Arlie O. Petters, Xiaoying Dong-An Introduction To Mathematical Finance With Applications - Understanding and Building Financial Intuition-S

Springer Undergraduate Texts
in Mathematics and Technology
ArlieO.Petters
XiaoyingDong
An Introduction
to Mathematical
Finance with
Applications
Understanding and Building Financial Intuition
Springer Undergraduate Texts in Mathematics
and Technology
Series Editors:
J.M. Borwein
H. Holden
V.H. Moll
More information about this series at http://www.springer.com/series/7438

Arlie O. Petters Xiaoying Dong
An Introduction to Mathematical
Finance with Applications
Understanding and Building Financial Intuition
123
Arlie O. Petters Xiaoying Dong
Department of Mathematics Department of Mathematics
Duke University Duke University
Durham, NC, USA Durham, NC, USA
ISSN 1867-5506 ISSN 1867-5514 (electronic)

Springer Undergraduate Texts in Mathematics and Technology
ISBN 978-1-4939-3781-3 ISBN 978-1-4939-3783-7 (eBook)
DOI 10.1007/978-1-4939-3783-7
Library of Congress Control Number: 2016939449
Mathematics Subject Classification (2010): 91Gxx
Arlie O. Petters and Xiaoying Dong 2016

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction
on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation,
computer software, or by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed
to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty,
express or implied, with respect to the material contained herein or for any errors or omissions that may have been
made.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG Switzerland
To my loving wife, Elizabeth Petters, for being at my
side unconditionally and my child, Preston Petters,
who inspires me with his intense curiosity.
A.O. Petters
To my dear husband and best friend, Xin Zhou. I could

not imagine to complete my part of the contribution to
this book without his love and support.
X. Dong
Preface
Rationale and Aim
Given the increasing intricacies and interconnectedness of financial firms ac-

tivities and the potential opportunities and risks to which they expose them-
selves and the worlds economy, the next generation of financial engineers
needs to master an extensive array of mathematical financial models. Indeed,
one of the current challenges in finance is that the complexity of modern se-
curities and markets has forced modelers to employ increasingly sophisticated
mathematical tools to address financial issues, creating a widening gap be-
tween the qualitative and quantitative approaches to finance.
Our book seeks to address this gap by introducing the quantitative aspects
of finance to students with either a qualitative background or no background
in the subject. At a firm the traders, risk managers, etc. employ proprietary an-
alytical and numerical models custom made to the needs of their firm. How-
ever, since open access to such models is prohibited, the book instead strives
to give students a fundamental understanding of key financial ideas and tools
that form the basis for building more realistic models, including those of a
proprietary nature.
Distinctive Features and Benefits
This book is distinct in how it emphasizes and pedagogically conveys in an

accessible manner the theoretical understanding and applications of the math-
ematical models forming key pillars of modern finance.
First, the book keeps a good balance between mathematical derivation and
description for the sake of providing an adequate level of rigor and depth in
mathematics and maintaining accessibility to the reader, which in turn adds
flexibility of material selection for the instructor (e.g., Chapter 7 may be taught
earlier). Specifically, this book addresses the gap between textbooks that of-
vii
viii Preface
fer a theoretical treatment without many applications and those that simply
present and apply formulas without appropriately deriving them. Indeed, the-
oretical understanding is incomplete without enough practice in applications,
and applications are risky without a rigorous theoretical understanding. To
accomplish this, the book contains numerous carefully chosen examples and
exercises that reinforce a students conceptual understanding and develop a
facility with applications. Indeed, the exercises are divided into conceptual,
application, and theoretical problems that probe the material deeper.
Second, beyond a few required undergraduate mathematics courses (see
Prerequisites below), this book is essentially self-contained. The large num-
ber of necessary financial terminologies and concepts can be overwhelming
to a student new to finance. For this reason, after introducing some central,
big-picture financial ideas in the first chapter, we present the financial minu-
tia along the way as needed. We have tried to make the book self-contained
in this regard through thoughtfully chosen illustrative applications starting at
the ground level with simple interest. We then gradually increase the difficulty
as the book develops, ranging across compound interest, annuities, portfolio theory,
capital market theory, portfolio risk measures, the role of linear factor models in portfo-
lio risk attribution, binomial tree models, stochastic calculus, derivatives, the martin-
gale approach to derivative pricing, the Black-Scholes-Merton model, and the Merton
jump-diffusion model.
Third, the book is also useful for students preparing either for higher level
study in mathematical finance or for a career in actuarial science. For example,
the syllabi for the actuarial Financial Mathematics Exam (Exam 2/FM) and
Models of Financial Economics Exam (Exam 3F/MFE) include many topics
covered in the book.
Prerequisites
The required mathematics consists of introductory courses on multivariable

calculus, probability, and linear algebra. Along the way, we introduce addi-
tional mathematical tools as needede.g., some measure theory is presented
from scratch.
No background in finance is assumed. As noted above, the necessary financial
concepts and tools are introduced in the text, with the first chapter giving an
overview of several common finance terminologies associated with securities
and securities markets.
Our book does not require computer programming. In our experience, fi-
nance courses based on computer programming are best taken after students
have developed a fundamental understanding of the theoretical architecture
of financial models.
Preface ix
Audience
The text is aimed at advanced undergraduates and masters degree students who are
either new to finance or want a more rigorous treatment of the mathematical
models used in finance. The students typically are from economics, mathemat-
ics, engineering, physics, and computer science.
We also believe that a faculty member who is teaching finance for the first
time will find this introduction readily manageable. Professionals working in
finance who would like a refresher or even clarification on some of the the-
oretical and conceptual aspects of mathematical finance will benefit from the
text.
Scope and Guide
The chapters are organized naturally into four parts and range over the fol-
lowing topics:
- Part I (Chapters 1 and 2):
introduction to securities markets and the time value of money
- Part II (Chapters 3 and 4):
Markowitz portfolio theory, capital market theory, and portfolio risk measures
- Part III (Chapters 5 and 6):
modeling underlying securities using binomial trees and stochastic calculus
- Part IV (Chapters 7 and 8):
derivative securities, BSM model, and Merton jump-diffusion model
The material was tested in courses offered to upper-level undergraduates and
masters degree students. Below are two examples of possible topics that may
serve as a guide for semester-long courses:
- Introduction to Mathematical Finance: securities markets (Chapter 1), the time
value of money (Chapter 2), Markowitz portfolio theory, capital market
theory, and portfolio risk measures (Chapters 34), binomial security pric-
ing (Chapter 5, omit most derivations), Itos formula and geometric Brow-
nian motion (Sections 6.8 and 6.9), forwards, futures, and options (Sec-
tions 7.2, 7.3, and 7.5), and call option pricing with applications (Sections 8.3,
8.2.2, 8.5, and 8.6.2).
- Introduction to Financial Derivatives: modeling underliers in discrete time
(Sections 5.15.3), stochastic calculus and modeling underliers in continuous
time (Section 5.4 and Chapter 6), general aspects of forwards, futures, swaps,
and options, including trading strategies (Chapters 7), the Black-Scholes-
Merton (BSM) model, BSM p.d.e. approach to pricing European-style op-
tions, risk-neutral approach to pricing European-style options, applications
to warrants, delta hedging, managing portfolio risk, and extension of the
BSM model to the Merton jump-diffusion model (Chapter 8).
x Preface
A year-long course on introductory mathematical finance can be based on the

entire book. The book can also be used as a reference for students enrolled in a
mathematical finance independent study course.
Acknowledgments
Specials thanks to the following individuals for their feedback and assistance:
Daniel Aarhus Lu Liu Chi Trinh
Amir Aazami Ruisi Ma Dan Turtel
Stanley Absher Tanya Mallavarapu Kari Vaughn
Vibhav Agarwal Xavier Mela Robert Vanderbei
Hengjie Ai Vadim Mokhnatkin Kevin Wan
Mitesh Amarthaluru Julia Ni Chenyu Wang
Vlad Bouchouev James Nolen David Williams
Michael Brandt Vivek Oberoi Chao Xu
Esteban Chavez Feng Pan Hangjun Xu
Rui Chen Chloe Peng Lu Xu
Kyuwon Choi Junkai Xue
Hal Press
Qian Deng Hui Qi Chao Yang
Christian Drappi Zhaozhen Qian Jiahui Yang
Zachary Freeman Hayagreev Ramesh Ashley Yeager
Tingran Gao Emma Rasiel Jeong Yoo
William Grisaitis Tianhua Ren Yanchi Yu
Xiaosheng Guo Chelsea Richwine Yunliang Yu
Zhonglin Han Irving Salvatierra Javier Zapata
John Hyde Andrew Schretter Xiaodong Zhai
Yuhang Si
Huseyin Kortmaz Biyuan Zhang
Baolei Li John Sias
Yang Zhang
Junchi Li Maxwell Stern
Lingran Sun Bowen Zhao
Li Li Ruiyang Zhao
Nan Li Alberto Teguia
Qiao Li Xiaoyang Zhuang
Nicholas Tenev
Li Liang Dominick Totino Zilong Zou
We are also thankful to Elizabeth Loew of Springer for her support and guid-
ance along the entire way and to Lisa Goldberg for her valuable comments and
constructive suggestions. AP is indebted to Duke University for providing
the financial support needed to hire many students who assisted with writ-
ing computer codes, checking calculations, etc. He is also extremely grateful to
his wife, Elizabeth Petters, for her patience, love, and steadfast encouragement
throughout the project. XD would like to express her gratitude to her husband
Xin Zhou who saw her through this book and offered great suggestions.
Durham, NC, USA A.O. Petters

Durham, NC, USA X. Dong
2016
Contents
1 Preliminaries on Financial Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 A Primer on Banks and Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Banks and the Federal Funds Rate . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Short-Term and Long-Term Rates and Yield Curves . . . . . . . 4
1.2 A Primer on Securities Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Securities Markets Organization . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Professional Participants in Securities Markets . . . . . . . . . . . 8
1.2.3 Bid-Ask Spreads and Market Liquidity . . . . . . . . . . . . . . . . . . 8
1.2.4 Trading Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Economic Indicators That May Affect Financial Markets . . . . . . . . 10
2 The Time Value of Money . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.1 Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Interest Rate and Return Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.1 Interest Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Required Return Rate and the Risk-Free Rate . . . . . . . . . . . . 16
2.2.3 Total Return Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Simple Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Compound Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.1 Compounding: Nonnegative Integer Number of Periods . . 22
2.4.2 Compounding: Nonnegative Real Number of Periods . . . . 24
2.4.3 Fractional Compounding Versus Simple Interest . . . . . . . . . 30
2.4.4 Continuous Compounding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5 Generalized Compound Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.1 Varying Interest and Varying Compounding Periods . . . . . 31
2.5.2 APR Versus APY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.5.3 Geometric Mean Return Versus Arithmetic Mean Return . . 35
2.6 The Net Present Value and Internal Rate of Return . . . . . . . . . . . . . 38
2.6.1 Present Value and NPV of a Sequence of Net Cash Flows . 38
2.6.2 The Internal Return Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
xi
xii Contents
2.6.3 NPV and IRR for General Net Cash Flows . . . . . . . . . . . . . . . 42

2.7 Annuity Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.7.1 Future and Present Values of Simple Ordinary Annuities . . 47
2.7.2 Amortization Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.7.3 Annuities with Varying Payments and Interest Rates . . . . . 56
2.8 Applications of Annuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.8.1 Saving, Borrowing, and Spending . . . . . . . . . . . . . . . . . . . . . . . 59
2.8.2 Equity in a House . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.8.3 Sinking Funds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.9 Applications to Stock Valuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.9.1 The Dividend Discount Model . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.9.2 Present Value of Preferred and Common Stocks . . . . . . . . . . 65
2.10 Applications to Bond Valuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.10.1 Bond Terminologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.10.2 Bond Prices Versus Interest Rates and Yield to Maturity . . . 69
2.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.11.1 Conceptual Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.11.2 Application Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.11.3 Theoretical Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3 Markowitz Portfolio Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.1 Markowitz Portfolio Model: The Setup . . . . . . . . . . . . . . . . . . . . . . . . 83
3.1.1 Security Return Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.1.2 What About Multivariate Normality of Security Return
Rates? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.1.3 Investors and the Efficient Frontier . . . . . . . . . . . . . . . . . . . . . . 87
3.1.4 The One-Period Assumption, Weights, and Short Selling . . 88
3.1.5 Expected Portfolio Return Rate . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.1.6 Portfolio Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.1.7 Risks and Covariances of the Portfolios Securities . . . . . . . . 96
3.1.8 Expectation and Volatility of Portfolio Log Return . . . . . . . . 100
3.2 Two-Security Portfolio Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.2.2 Efficient Frontier of a Two-Security Portfolio . . . . . . . . . . . . . 107
3.2.3 Reducing Risk Through Diversification . . . . . . . . . . . . . . . . . . 114
3.3 Efficient Frontier for N Securities with Short Selling . . . . . . . . . . . . 117
3.3.1 N-Security Portfolio Quantities in Matrix Notation . . . . . . . 118
3.3.2 Derivation of the N-Security Efficient Frontier . . . . . . . . . . . 120
3.4 N-Security Efficient Frontier Without Short Selling . . . . . . . . . . . . . 126
3.5 The Mutual Fund Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.5.1 The Global Minimum-Variance Portfolio . . . . . . . . . . . . . . . . . 128
3.5.2 The Diversified Portfolio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
Contents xiii
3.5.3 The Mutual Fund Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

3.6 Investor Utility Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.6.1 Utility Functions and Expected Utility Maximization . . . . . 131
3.6.2 Risk-Averse, Risk-Neutral, and Risk-Seeking Investors . . . . 133
3.7 Diversification and Randomly Selected Securities . . . . . . . . . . . . . . 138
3.7.1 Mean Portfolio Variance and the Uniform Dirichlet
Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.7.2 Mean Portfolio Variance using the NASDAQ . . . . . . . . . . . . . 142
3.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4 Capital Market Theory and Portfolio Risk Measures . . . . . . . . . . . . . . 151

4.1 The Capital Market Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.1.1 The Capital Market Line (CML) . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.1.2 Expected Return and Risk of the Market Portfolio . . . . . . . . 157
4.1.3 The Capital Asset Pricing Model (CAPM) . . . . . . . . . . . . . . . . 158
4.1.4 The Security Market Line (SML) . . . . . . . . . . . . . . . . . . . . . . . . 163
4.1.5 CAPM Security Risk Decomposition . . . . . . . . . . . . . . . . . . . . 164
4.2 Portfolio Risk Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.2.1 The Sharpe Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.2.2 The Sortino Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
4.2.3 The Maximum Drawdown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.2.4 Quantile Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4.2.5 Value-at-Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
4.2.6 Conditional Value-at-Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
4.2.7 Coherent Risk Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
4.3 Introduction to Linear Factor Models . . . . . . . . . . . . . . . . . . . . . . . . . . 185
4.3.1 Definition and Intuition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
4.3.2 Portfolio Variance Decomposition . . . . . . . . . . . . . . . . . . . . . . . 189
4.3.3 Factor Categorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
4.3.4 Alpha and Beta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.3.5 CAPM Beta Versus Linear Factor Beta . . . . . . . . . . . . . . . . . . . 195
4.3.6 Fama-French Three-Factor Model . . . . . . . . . . . . . . . . . . . . . . . 196
4.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
xiv Contents
5 Binomial Trees and Security Pricing Modeling . . . . . . . . . . . . . . . . . . . . 209

5.1 The General Binomial Tree Model of Security Prices . . . . . . . . . . . . 209
5.2 The Cox-Ross-Rubinstein Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
5.2.1 The Real-World CRR Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
5.2.2 The Risk-Neutral CRR Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.3 Continuous-Time Limit of the CRR Pricing Formula . . . . . . . . . . . . 237
5.3.1 The Lindeberg Central Limit Theorem . . . . . . . . . . . . . . . . . . . 237
5.3.2 The Continuous-Time Security Price Formula . . . . . . . . . . . . 241
5.4 Basic Properties of Continuous-Time Security Prices . . . . . . . . . . . . 246
5.4.1 Some Statistical Formulas for Continuous-Time Security
Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
5.4.2 Some Probability Formulas for Continuous-Time
Security Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
6 Stochastic Calculus and Geometric Brownian Motion Model . . . . . . 253

6.1 Stochastic Processes: The Evolution of Randomness . . . . . . . . . . . . 253
6.1.1 Notation for Probability Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.1.2 Basic Concepts of Random Variables . . . . . . . . . . . . . . . . . . . . 257
6.1.3 Basic Concepts of Stochastic Processes . . . . . . . . . . . . . . . . . . 260
6.1.4 Convergence of Random Variables . . . . . . . . . . . . . . . . . . . . . . 265
6.1.5 Skewness and Kurtosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
6.2 Filtrations and Adapted Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
6.2.1 Filtrations: The Evolution of Information . . . . . . . . . . . . . . . . 268
6.2.2 Conditional Expectations: Properties and Intuition . . . . . . . 270
6.2.3 Adapted Processes: Definition and Intuition . . . . . . . . . . . . . 273
6.3 Martingales: A Brief Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
6.3.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
6.3.2 Martingale as a Necessary Condition of an Efficient Market 277
6.4 Modeling Security Price Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
6.4.1 From Deterministic Model to Stochastic Model . . . . . . . . . . . 278
6.4.2 Innovation Processes: An Intuition . . . . . . . . . . . . . . . . . . . . . . 279
6.4.3 Securities Paying a Continuous Cash Dividend . . . . . . . . . . . 281
6.5 Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
6.5.1 Definition of Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . 282
6.5.2 Some Properties of Brownian Motion Paths . . . . . . . . . . . . . . 284
6.5.3 Visualization of Brownian Paths . . . . . . . . . . . . . . . . . . . . . . . . 285
6.5.4 Markov Property for Brownian Motion . . . . . . . . . . . . . . . . . . 288
6.6 Quadratic Variation and Covariation . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Contents xv
6.6.1 Motivation, Definition, and Notation . . . . . . . . . . . . . . . . . . . . 289

6.6.2 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
6.6.3 Quadratic Variation and Covariation Properties of BM . . . . 293
6.6.4 Significance of Quadratic Variation . . . . . . . . . . . . . . . . . . . . . . 297
6.7 Ito Integral: A Brief Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
6.7.1 Importance of Ito Integral with Respect to BM . . . . . . . . . . . . 299
6.7.2 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
6.7.3 A Famous Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
6.8 Itos Formula for Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
6.8.1 Ito Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
6.8.2 Itos Lemma for Brownian Motion . . . . . . . . . . . . . . . . . . . . . . 304
6.8.3 Risk-Neutral Probability Measure . . . . . . . . . . . . . . . . . . . . . . . 309
6.8.4 Girsanov Theorem for a Single Brownian Motion . . . . . . . . . 311
6.9 Geometric Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
6.9.1 GBM: Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
6.9.2 GBM: Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
6.9.3 Relation Between Binomial Tree Model and GBM Model . . 317
6.10 BM as a Limit of Simple Symmetric RW . . . . . . . . . . . . . . . . . . . . . . . 320
6.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
7 Derivatives: Forwards, Futures, Swaps, and Options . . . . . . . . . . . . . . 329

7.1 Derivative Securities: An Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
7.1.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
7.1.2 Basic Functions of Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 331
7.1.3 Characteristics of Derivative Valuation . . . . . . . . . . . . . . . . . . 332
7.1.4 No-Arbitrage Principle and Law of One Price . . . . . . . . . . . . 334
7.2 Forwards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
7.2.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
7.2.2 Forwards on Assets Paying a Continuous Cash Dividend . 340
7.2.3 Forward Price Formula and the Spot-Forward Parity . . . . . 341
7.2.4 Forward Value Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
7.3 Futures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
7.3.1 Evolution from Forwards to Futures . . . . . . . . . . . . . . . . . . . . . 345
7.3.2 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
7.3.3 Impact of Daily Settlement: A Brief Discussion . . . . . . . . . . . 347
7.4 Swaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
7.4.1 A Brief Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
xvi Contents
7.5 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353

7.5.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
7.5.2 How Options Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
7.5.3 Terminal Payoff and Profit Diagrams . . . . . . . . . . . . . . . . . . . . 359
7.5.4 Market Sentiment Terminologies and Option Moneyness . 363
7.5.5 Option Strategies: Straddle, Strangle, and Spread . . . . . . . . . 365
7.5.6 Put-Call Parity for European Options Revisited . . . . . . . . . . 368
7.5.7 Relation Among Put, Call, and Forward . . . . . . . . . . . . . . . . . 369
7.5.8 Intrinsic Value and Time Value . . . . . . . . . . . . . . . . . . . . . . . . . . 370
7.5.9 Some General Relations of Options . . . . . . . . . . . . . . . . . . . . . . 372
7.5.10 Put-Call Parity for American Options . . . . . . . . . . . . . . . . . . . 373
7.5.11 Boundary Conditions for European Options . . . . . . . . . . . . . 376
7.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
8 The BSM Model and European Option Pricing . . . . . . . . . . . . . . . . . . . . 383

8.1 The BSM Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
8.1.1 Marketplace Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
8.1.2 Money Market Account and the Underlier Model . . . . . . . . 386
8.1.3 Self-Financing, Replicating Portfolio . . . . . . . . . . . . . . . . . . . . . 388
8.1.4 Derivation of the BSM p.d.e. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
8.2 Applications of BSM Pricing to European Calls and Puts . . . . . . . . 392
8.2.1 Solving the BSM p.d.e. for European Calls . . . . . . . . . . . . . . . 392
8.2.2 BSM Pricing Formula for European Puts . . . . . . . . . . . . . . . . . 395
8.2.3 Delta and the Partial Derivative Relative to Strike Price . . . 396
8.2.4 European Call and Put Deltas at Expiration . . . . . . . . . . . . . . 398
8.3 Application to Pricing Warrants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
8.4 Risk-Neutral Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
8.4.1 Review of Conditional Expectation . . . . . . . . . . . . . . . . . . . . . . 400
8.4.2 From BSM Pricing to Risk-Neutral Pricing . . . . . . . . . . . . . . . 404
8.4.3 The Fundamental Theorems of Asset Pricing . . . . . . . . . . . . . 408
8.4.4 Risk-Neutral Pricing of European Calls and Puts . . . . . . . . . 410
8.5 Binomial Approach to Pricing European Options . . . . . . . . . . . . . . . 411
8.5.1 One-Period Binomial Pricing by Self-Financing Replication 412
8.5.2 One-Period Binomial Pricing by Risk Neutrality . . . . . . . . . . 415
8.5.3 Two-Period Binomial Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
8.5.4 n-Period Binomial Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
8.6 Delta Hedging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
8.6.1 Theoretical Delta Hedging for European Calls . . . . . . . . . . . . 422
8.6.2 Application of Delta Hedging to Selling European Calls . . 427
Contents xvii
8.7 Option Greeks and Managing Portfolio Risk . . . . . . . . . . . . . . . . . . . 433

8.7.1 Option Greeks for Portfolios and the BSM p.d.e. . . . . . . . . . . 433
8.7.2 Delta-Neutral Portfolios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
8.7.3 Delta-Gamma-Neutral Portfolios . . . . . . . . . . . . . . . . . . . . . . . . 438
8.8 The BSM Model Versus Market Data . . . . . . . . . . . . . . . . . . . . . . . . . . 440
8.8.1 Jumps in Security Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
8.8.2 Skewness and Kurtosis in Security Log Returns . . . . . . . . . . 441
8.8.3 Volatility Skews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
8.9 A Step Beyond the BSM Model: Merton Jump Diffusion . . . . . . . . 448
8.9.1 Poisson Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
8.9.2 The MJD Stochastic Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
8.9.3 Illustration of MJD Jump, Skewness, and Kurtosis . . . . . . . . 456
8.9.4 No-Arbitrage Condition and Market Incompleteness . . . . . 458
8.9.5 Pricing European Calls with an MJD Underlier . . . . . . . . . . . 461
8.9.6 MJD Volatility Smile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
8.10 A Glimpse Ahead . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
8.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
Chapter 1
Preliminaries on Financial Markets
Financial markets are a part of a financial system. In a modern economy, a fi-

nancial system consists of financial institutions (e.g., banks) that essentially are
intermediaries among policy makers, companies, and consumers who from
time to time alternate between their roles as savers and investors and chan-
nel savings into investments in financial markets through buying and selling
financial products.
Just as an industrial society needs physical infrastructure for transporting
goods, an information society needs financial infrastructure for innovations.
Just as a modern society needs mechanical and electrical engineers to apply the
principles of science and mathematics to build and maintain our mechanical
and electronic systems, a world increasingly built on finance needs financial
engineers to support financial systems. Financial engineers apply mathemati-
cal tools like modeling to support the details of investment decision-making,
such as risk management,1 and the machinery of market making, such as
derivatives pricing.2
The field of mathematics that is concerned with financial markets is called
mathematical finance, also known as quantitative finance. Although mathematical
models derived from mathematical finance need not be compatible with either
economic or financial theory, the student of mathematical finance will find it
important, and sometimes necessary, to have a basic understanding of interest
rates and financial markets, as well as of the financial system as a backdrop.
1 See Chapters 3 and 4.

2 See Chapters 6, 7 and 8.
Arlie O. Petters and Xiaoying Dong 2016 1

A.O. Petters, X. Dong, An Introduction to Mathematical Finance with Applications, Springer
Undergraduate Texts in Mathematics and Technology, DOI 10.1007/978-1-4939-3783-7 1
2 1 Preliminaries on Financial Markets
1.1 A Primer on Banks and Rates
Interest rates are a key concept in economics. The level of interest rates plays
an extremely important role in a market economy and therefore in financial
markets as well.
1.1.1 Banks and the Federal Funds Rate
Banks can be classified into central banks, investment banks, and commercial
banks.
1. A central bank is the monetary authority in a nation.

The central bank of the United States is the Federal Reserve System (also
known as the Federal Reserve or simply as the Fed) which is structurally com-
posed of the Federal Reserve Board (a central board of governors), the Fed-
eral Open Market Committee (FOMC), and twelve Federal Reserve Banks.
The FOMC meets eight times a year and is in charge of setting the reserve
ratio and targets for the federal funds rate and the federal discount rate:
Most banks operate under a form called fractional reserve banking, in

which banks maintain reserves of cash for only a fraction of the cus-
tomers deposits and lend the rest out. This fraction is referred to as the
reserve ratio.
The federal funds rate is the interest rate at which US banks lend money to
each other overnight.
Since sometimes banks try to stay as close to the reserve limit as possi-
ble (for instance, when they try to maximize their lending profit), they
may go under the reserve requirement. If at the end of a day, a bank has
a reserve deficit, it needs to borrow money (demand) to boost its reserve
from another bank that has excess reserve (supply). The fed funds rate is
determined by the market of such demand and supply. The FOMC influ-
ences the fed funds rate by buying or selling financial instruments (such
as bonds issued by the U.S. government) on the open market.
The federal discount rate is the interest rate at which the Federal Reserve
lends money to banks overnight. This lending window is designed as the
last resort for banks to borrow money to meet the reserve requirement.
For this reason, the federal discount rate is higher than the federal funds
rate.3
3 Due to the Feds quantitative easings (QEs) in the last few years, where QEs were programs for large-
scale purchases of assets from the banks that drove up the volume of excess reserves to unprecedented
levels, presently many US banks can meet their reserve requirements without borrowing.
1.1 A Primer on Banks and Rates 3
2. Investment banking provides services such as securities underwriting, merg-

ers and acquisitions advisory, and asset management and securities trading,
primarily for institutional clients.
3. Commercial banking provides services mainly in deposit and lending activi-
ties (such as mortgages (e.g., home loans), auto loans, credit card services,
and student loans) to individuals and small business.
While discussion of the separation of commercial and investment banking

is beyond the scope of this book, we can say that those banks whose pri-
mary business is investment banking are investment banks and that those banks
whose primary business is commercial banking are commercial banks. Commer-
cial banks often have many local retail branches.
Although they both have commercial banking divisions, Goldman Sachs
and Morgan Stanley are examples of investment banks. JPMorgan Chase, the
Bank of America, and Wells Fargo have significant business in investment
banking but are primarily examples of commercial banks.
Example 1.1. The following discussion is designed to help the reader under-
stand how the fractional reserve banking system works and establish an intu-
ition of the concepts of money, credit/debt, and leverage.
When JD1 (John Doe number 1) deposits $10,000 into his bank account, if
the banks reserve ratio is 10%, then the bank needs to reserve only $1,000 and
may lend the remaining $9,000 out. Lets say JD2 (John Doe number 2) gets the
$9,000 loan to buy a car, and the car dealer puts the $9,000 back into another
bank. The second bank reserves $900 and lends out $8,100 to JD3. Theoretically
speaking, this series of financial maneuvers can go on to JD4, JD5, and so on,
forever. As the result, the original $10,000 (cash) could generate $100,000 (cash
and credit/debt), for

10, 000
10, 000 (90%)n = = 100, 000.
n =0
1 0.9
This is why 10% reserve (ratio) is equivalent to 10:1 leverage (ratio).

Remark 1.1.
1. The last example shows that the money needs to travel to generate the credit.
Although there are times when cash is king, there are also times when
money doesnt count unless it is in motion.
While the discussion on developed world central banks starting to impose negative interest rates is
beyond the scope of this book, it is worth noting that on June 5, 2014, the European Central Bank cuts
its deposit rate for banks from 0 to -0.1% to encourage banks to lend rather than hold on to money; on
January 29, 2016, the Bank of Japan announced that it would cut the interest rate, set at -0.1%, further
into negative territory if necessary, that compared with -0.65% in Denmark, -1.1% in Sweden, and
-0.75% in Switzerland already (Source: WSJ, 01/29/2016).
2. High risk in financial investing is often caused by high leverage. In fact,

Lehman Brothers was leveraged 30.7:1 according to its last annual financial
statement before filing for bankruptcy in September 2008. And in October
2011, another Wall Street bank was taken down by bad bets financed by
excessive leverage: MF Global. Their leverage ratio? 40:1.4

1.1.2 Short-Term and Long-Term Rates and Yield Curves
Interest rates (and bond yields5) affect all financial markets. Among them,
there are short-term (fewer than 12 months) rates, there are long-term (usually
10 years or longer) rates, and there are some in between.
Since short-term interest rates and long-term interest rates are determined by
different mechanisms, we consider them separately.
Short-Term Rates
Short-term rates are administered by the FOMC in the USA (and by central
banks in other nations).
The fed funds rate not only is important to banks but also has trickle-down
effects that affect investors and consumers (or end users). This rate is also
used as the benchmark for other short-term interest rates.
When pricing a loan, a lender often uses the formula
{ index rate } + { spread (or margin)}

= { the interest rate which the borrower is to pay},
where the size of a spread (or margin) depends on how risky the lender feels
the loan is; the riskier the loan, the bigger the spread (or the higher the margin).
The often used index rates (or base rates) are the prime rate, LIBOR, or COFI,
depending on which of the following loans is under consideration:
variable-rate credit card credits,
variable-rate auto loans,
variable-rate student loans,
home equity lines of credit (HELOC),
adjustable-rate mortgages (ARM),
small business loans,
personal loans.
Here are the descriptions of prime rate, LIBOR, and COFI:
4 Source: http://azizonomics.com/2011/11/15/zombie-economics/
5 See Section 2.10.
1.1 A Primer on Banks and Rates 5
The prime rate, as a generic term, is the interest rate that the banks charge
their most credit-worthy customers; it primarily refers to the Wall Street
Journal Prime Rate which is the consensus prime rate published by the
WSJ after polling ten of Americas largest banks. The prime rate moves
up or down in lock step with changes by the FOMC. Normally, it runs
approximately 300 basis points (or 3 percentage points) above the fed funds
target rate, since the break-even lending rate of interest for banks making
loans is essentially the same as the prime lending rate.
The prime rate is the most popular index rate used in the USA.
The London Interbank Offered Rates (LIBOR) can be described as a daily refer-
ence of the wholesale cost of money in the London interbank money market
of 18 banks.6 Loosely speaking, LIBOR is like an international counterpart
of the fed funds rate. However, unlike the fed funds rate, there are many
different LIBOR rates with maturities ranging from overnight to 12 months.
The LIBOR is frequently the basis of investments including interest swap
agreements and forward contracts,7 and for many adjustable mortgage
loans as well. American banks use LIBOR because LIBOR is updated much
more frequently than the US prime rate as we mentioned earlier; this has
advantages particularly when global credit market conditions deteriorate
rapidly.
The 11th District Cost of Funds Index (COFI) is an index rate primarily used
in the western USA to set the cost of variable-rate loans and is computed
by using data from three western states Arizona, California, and Nevada
(which are covered by the 11th district).
Long-Term Rates
Long-term rates are determined by market forces.

This is to say that long-term interest rates are not an influencing factor
(which short-term rates often are). Instead, they are determined by the eco-
nomic fundamentals. In other words, they serve as a measure of how our econ-
omy is doing.
The fixed-rate mortgage (such as 15- or 30-year) rates are determined by the
trading of mortgage securities on Wall Street.
6 These 18 banks include Bank of America, Barclays, Credit Suisse, Deutsche Bank, HSBC, and JP Mor-
gan Chase. For a complete list, visit the website at https://en.wikipedia.org/wiki/Libor.
7 See Sections 7.2 and 7.4.1.
Yield Curve Basics
Since there are different interest rates based on various terms as we explained
earlier, it is desirable to represent all these interest rates simultaneously. A
graphical representation of where interest rates are today is called a yield curve.
More precisely, a yield curve is the graph of y = r (t) where t represents time to
maturity for bonds of the same asset class and credit quality (e.g., US Trea-
suries, LIBOR, zeros,8 or AA-rated corporate bonds, and so on), and y is the
corresponding yield to maturity. In short, a yield curve is a plot of bond yields
to maturity against times to maturity.
Example 1.2. In daily financial news, the Treasury yield curve is often shortened
to the yield curve. The yield curve is considered a leading indicator of economic
activity (see Section 1.3) and often used as a reference point for forecasting
interest rates by investors.
The zero-coupon yield curve or the spot-rate curve is created by plotting the
yields of zero-coupon treasury bills against their corresponding maturities.
The primary use of zero (or spot) interest rates is to discount cash flows.9
1.2 A Primer on Securities Markets
Market structures are defined by the trading rules and trading systems which
include what information (e.g., orders and quotations) traders can see. A con-
cise interpretation of the financial jargons includes (but is not limited to) the
following:
Securities are financial products10 that can be traded on securities markets.
Securities markets are the trade execution venue.
Clearing houses are responsible for settling trades.
Securities depositories are responsible for holding security certificates.
Brokers arrange trades for their clients (including clearing and settlement).
Dealers trade with their clients and are obligated liquidity providers.
1.2.1 Securities Markets Organization
Securities markets may be classified into two levels. One is the primary market
or new issue market,11 and another is the secondary market or after market.
8 Conceptually speaking, the most basic debt instrument is a zero-coupon bond or simply a zero, which
is a bond with a single cash flow equal to face value at maturity.
9 See Chapter 2 for the concept of discount cash flows.
10 See Section 7.1.1 for more detailed explanation. Examples of securities are stocks and bonds.
11 The market where new securities are issued.
1.2 A Primer on Securities Markets 7
Primary markets deal with the trading of newly issued securities, whereas
secondary markets deal with the trading of securities that have already been
issued in the primary market (i.e., existing securities). Thus, after market
may be interpreted as after new issue market.
Secondary markets are organized in two basic ways. One is the exchanges
and another is the over-the-counter (OTC) markets.
Exchanges are highly organized and centralized markets where securities
are traded. Exchanges started as physical places, trading floors, where trading
took place (although the advent of ECNs, or electronic trading, has eliminated
the need for such traditional floors), whereas OTC markets are less formal,
have never been a physical place, and connect broker-dealers only electroni-
cally.
Examples of exchanges are the New York Stock Exchange (NYSE) and the
NASDAQ Stock Exchange (NASDAQ), where the majority of larger US public
companies are traded.
Examples of OTC markets are over-the-counter bulletin board (OTCBB), an
electronic trading service, and pink sheets (a quotation service). In general,
stocks that are traded on OTC are considered to be more speculative than
stocks that are listed on exchanges.
NYSE is an order-driven market (it functions like an auction market), whereas
OTC, by contrast, is a quote-driven dealer market.
For a private company to go public, issue shares and be traded on an ex-
change thereafter, it needs12 to choose an exchange on which to be listed, which
means that it must be able to meet that exchanges listing requirements (among
other satisfactions13).
If a listed stock, at a later point in time, fails to comply with the exchanges
listing requirements, it will be delisted, i.e., removed from the exchange on
which the stock was issued. After a stock is officially delisted, normally it will
be traded on the OTC markets, mainly either on OTCBB or on pink sheets.14
Hence by the very organization of the secondary securities markets, an in-
ternal quality control mechanism has been put in place. Exchanges with list-
ing requirements are motivated to ensure that only high-quality securities are
traded on them and to uphold the exchanges reputation among investors.
12 Otherwise, a public company is traded on OTC and the stock is referred to as unlisted stock, whereas
those stocks traded on exchanges are referred to as listed stocks.

13 These satisfactions include paying both the exchanges entry and yearly listing fees. Listing re-
quirements usually include minimum stockholders equity, a minimum share price, and a minimum
number of shareholders. The standards vary by exchange. For example, listing on the NASDAQ is
considerably less expensive than listing on the NYSE, which partially explains why newer companies
often opt for the NASDAQ if they meet its requirements.
14 A stock traded on pink sheets is considered to be riskier than that on OTCBB. In general, a stock
traded on pink sheets is very speculative.

1.2.2 Professional Participants in Securities Markets
Both clearing houses and securities depositories are important professional

participants in securities markets. These central organizations provide the in-
frastructure to support trades.
Buyers and sellers do not deal directly with each other but with a clearing
house. Clearing is the procedure by which an organization acts as a centralized
(single) counterparty to all counterparties in trades.
Central securities depositories facilitate ownership transfer through a book
entry, which is a system of tracking ownership of securities where no certificate
is given to investors.
Other professional participants in securities markets include, but are not
limited to, brokers, dealers, mutual funds, and hedge funds. Both brokers and
dealers are intermediaries between investors and financial markets.
Brokerage houses or simply brokers are the first place to contact for almost any
one willing to participate in financial markets directly (rather than through
mutual funds, for instance) since setting up a brokerage account is required
for using services from the broker to implement trading. Electronic brokers are
automated order-driven execution systems, whereas others may have capaci-
ties for providing clients with investment information and advice. All brokers
are compensated by commissions for their services.
Securities dealers or simply dealers are just like merchants in that they make
money from the difference between bid prices and ask prices (bid-ask spreads),
where the bid/ask price is the price at which traders are willing to buy/sell. De-
pending on different exchanges, dealers may be known by different names
such as specialists, market makers15 or floor traders, and so on. Most of the time
dealers are passive traders as they have affirmative obligations to provide liq-
uidity (see Section 1.2.3) to stabilize the market. Generally speaking, dealers
face inventory and adverse-information risks. Therefore they are motivated to
hedge their portfolios.16
1.2.3 Bid-Ask Spreads and Market Liquidity
A limit buy/sell order is a trade instruction with a limit bid/ask price and a size
(i.e., a quantity). A liquidity pool usually consists of a large number of limit or-
ders which cannot be matched currently. These orders are referred to as current
nonmarketable orders. A new order is called nonmarketable with respect to such
15 The US Securities and Exchange Commission (SEC) defines a market maker as a firm that stands
ready to buy and sell stock on a regular and continuous basis at a publicly quoted price. (Source:
http://www.sec.gov/answers/mktmaker.htm)
16 See Chapter 8 for details.
1.2 A Primer on Securities Markets 9
a pool if this order cannot be matched in the pool. As a result, this new order
will be added to the pool and this is called adding liquidity to the pool. On the
other hand, if the new order can be matched in the pool immediately, the new
order is filled and the size of the pool is reduced and therefore this is called
taking liquidity from the pool. Usually, liquidity is said to be high if the bid-ask
spread is small, and the ask size and bid size are large.
Given time t during market hours, let b(t) and a(t) denote the best bid price
and the best ask price at time t, respectively; the bid-ask spread at time t is
a(t) b(t). In other words, the current bid-ask spread or simply the spread is
the amount by which the current lowest ask price exceeds the current highest
bid price.
Example 1.3. Suppose that dealer quotations for MSFT, the ticker symbol for
Microsoft, show that the best bid price is $50 and the best ask price is $50.02,
then the bid-ask spread is $0.02.17 If you are an impatient trader, then you
either have to buy at price $50.02 or have to sell at price $50 at best. If the
market dealer can buy and/or sell one million shares of Microsoft at the best
bid price and/or sell at the best ask price, then he or she will make $20,000 in
a short period of time.

Liquidity plays a central role in the functioning of securities markets. In fact,
market liquidity is the single most important characteristic of well-functioning
markets.
Although there is no specific liquidity formula, the size of the (bid-ask) spread
may be used as a rule of thumb for a trader to measure market liquidity, since
a maximum spread rule is the most common affirmative obligation of desig-
nated market makers. The smaller the spread and the larger the size, the more
liquid the market is.
If the definition of a security market risk is the standard deviation of the
security return, then liquid markets are less risky than illiquid ones as liquid
markets are less volatile than illiquid ones.
1.2.4 Trading Costs
A list of trading costs includes, but is not limited to, the following:
1. Brokerage cost: commissions and fees

2. Liquidity cost: bid-ask spreads and price impacts to the market that cause
further slippage18
17 Note that the bid-ask spread is the ask price minus the bid price.
18 Note: the smaller the spreads, the lower the trading costs resulting in a higher return for investors.
To a certain extent, tightening spreads is the single most important factor in improving security
returns.
3. Difference between the short-term capital gain rate and the long-term capital
gain rate (as the former is usually much higher than the latter)
In practice, trading costs cannot be neglected at all and an active investor must
scrutinize the per trade cost.
1.3 Economic Indicators That May Affect Financial Markets
An economic cycle (or business cycle) is the term used to describe certain pat-
terns of wide fluctuations in economic activities followed by economies: an
expansion, until a peak, followed by a contraction, and until a trough. It is
called a cycle because this pattern repeatsthe trough phase is then followed
by expansion phase, peak phase, contraction phase, and trough phase again,
to compose another cycle.
An economic variable is a random variable whose sample space consists
of economic-related events. Often used economic variables are population,
poverty rate, available resources, dividend yield, inflation rate, imports and
exports, etc. An economic variable that reveals the direction in which the econ-
omy is moving (i.e., signs of contraction or expansion) is an economic indicator.
Just as the tense of a verb group can be classified into future, present, or past,
economic indicators can be classified into leading, coincident, and lagging in-
dicators:
Those that change before the economy changes are called leading indicators.
For example, new factory orders for consumer durable goods and the dif-
ference between interest rates at two different maturities (e.g., term spread).
Those that occur at the same time as the related economic activity are called
coincident indicators, e.g., GDP (gross domestic product19 ), nonfarm payrolls,
and retail sales.
Those that only become apparent after the related economic activity are
called lagging indicators. For example, CPI (consumer price index20 ) and the
unemployment rate.
The US Census Bureau always releases economic indicators on schedule.
Online economic calendars provide convenient access to many types of in-
formation, including economic indicator announcements with forecasts; the
definition of each indicator, with prior, prior revised and actual numbers,
19 Gross domestic product is the monetary value of all the finished goods and services produced
within a country in a specific time period.
20 Consumer price index is a measure of average change over time in the prices paid by urban con-
sumers for a market basket of consumer goods and services, such as transportation, food, and medical
care.
1.3 Economic Indicators That May Affect Financial Markets 11
Table 1.1 Example of a simplified version of Bloombergs domestic economic calendar for the busi-
ness week beginning January 28, 2013. Source: based on the Bloomberg table at http://www.
bloomberg.com/markets/economic-calendar.
Mon Jan 28 Tue Jan 29 Wed Jan 30 Thu Jan 31 Fri Feb 1
Durable FOMC MBA Job-Cut Employment

goods Meeting Purchase Report Situation
Orders Begins Applica- 7:30am 8:30am
8:30am tions
7:00am
Pending Store Sales ADP Empl Jobless PMI Mfg

Home Sales 7:45am Rept 8:15am Claims Index
Index 10:am 8:30am 8:58am
Dallas Fed Redbook GDP 8:30am Personal Consumer

Mfg Survey 8:55am Income and Sentiment
10:30am Outlays 9:55am
8:30am
4 week Bill Home Price Petroleum Employment ISM Mfg

Anncmnt Index Status Rept Cost Index Index
11:00am 9:00am 10:30am 8:30am 10:00am
3-Month Bill Consumer 7-Yr Note Chicago Construction

Auction Confidence Auction PMI 9:45am Spending
11:30am 10:00am 1:00pm 10:00am
6-Month Bill Investor FOMC Consumer

Auction Confidence Meeting Comfort
11:30am Index 10:am Anncmnt Index
2:15pm 9:45am
2-Yr Note 4 week Bill Natural Gas

Auction Auction Rept
1:00pm 11:30am 10:30am
Farm Prices
3:00pm
Fed Balance
Sheet
4:30pm
Money
Supply
4:30pm
highlights and reflections on daily stock market focus; bond auction informa-
tion; and other relevant events. A simplified economic calendar example is
provided in Table 1.1.
The effects of the release of economic indicators on financial markets depend

on many factors such as (including, but not limited to) the indicator itself, the
business cycle phase that the economy is in, the markets expectations, market
momentum, and many other factors.
We end Chapter 1 on an important note: modern financial markets do need and
will continue to need mathematical tools, and the mathematical assumptions must be
verified carefully to ensure these tools are implemented properly.
Chapter 2
The Time Value of Money
You may have heard the expression, A dollar today is worth more than a dollar
tomorrow, which is because a dollar today has more time to accumulate inter-
est. The time value of money deals with this basic idea more broadly, whereby
an amount of money at the present time may be worth more than in the fu-
ture because of its earning potential. In this chapter, we discuss the valuing
of money over different time intervals, which includes a study of the present
value of future money and the future value of present money. The theory is
laid out in a rigorous, detailed, and general framework and accompanied by
numerous applications with direct relevance to personal finance.
To be self-contained for readers new to finance, Sections 2.1 to 2.5 intro-
duce our conventions and terminologies associated with time, interest rates,
required return rates, total return rates, simple interest, compound interest for
integral and nonintegral periods, and generalized compound interest, where
the interest rate and compounding period vary. Readers already familiar with
these topics should skim those sections for our notational usage. In Sec-
tion 2.6, we introduce the net present value and internal return rate, including
Descartess Rule of Signs. The theory of annuities is presented in Section 2.7
and includes amortization theory and annuities with varying payments and
varying interest rates. Applications of annuity theory to saving, borrowing,
equity in a house, sinking funds, the present value of preferred and common
stocks, and bond valuation are given in Sections 2.8 to 2.10.

14 2 The Time Value of Money
2.1 Time
Before delving into the value of money over time, it is important to be clear
about our conventions and notation for time.
Throughout the book, the default unit of time is a year. Unless stated to the con-
trary, assume that a year consists of 365 calendar days and 252 trading days.1
When designating time, assume that there is a fixed starting time relative to
which the other moments of time are defined. The explicit choice of starting
time will depend on the context of the application, but we shall always repre-
sent it by 0. Note that the starting time need not be the current time.
We employ the following notation:
t0 = fixed current time, t = general moment of time.
Note that a general moment of time t > 0 simultaneously designates the number
of years of elapsed time from 0 to the given moment. For example, writing
1 1
t0 = , t=
4 2
means that the current time is 3 months after the starting time and t is 3 months
from now. If October 1, November 1, and December 1 in 2015 mark the times
0, t1 , and t2 , respectively, then
1 1
t1 = , t2 = .
12 6
We shall distinguish between an interval of time, say, [t0 , t f ], and its time span
, which is the length of the interval:
= t f t0 = number of years from t0 to t f , ( t0 0 ) .
If a time interval is partitioned into equal-length subintervals, then the length

of a subinterval is called a period. For example, a year has 12 monthly periods
and 4 quarterly periods.
We shall employ the following abbreviations:
mth = month(s), yr = year(s), prd = period(s).
In particular, one year is written as 1 yr and two years as 2 yr.
1 Apart from being mindful of leap years, note that banks may use a 360-day year when computing
their charge on loans. Any deviation from a 365-day year will be stated explicitly.
2.2 Interest Rate and Return Rate 15
2.2 Interest Rate and Return Rate
2.2.1 Interest Rate
You are perhaps most familiar with interest as the rate a bank pays into your
savings account (where you lend the bank money) or the rate a bank charges
you for a loan (where the bank lends you money). Overall, interest is the cost
of money. It is the compensation received for lending or investing money. The
initial amount of money you lend or borrow is called the principal and will be
denoted by F0 . Henceforth, assume that money investedwhether in a sav-
ings account or in a start-up companyis money lent with the expectation of
receiving back more than the amount invested (principal plus interest).
The compensation for lending or the charge for borrowing a principal F0 is
typically expressed as a percent r of F0 per year:

compensation or charge per year = r F0 .
The percent r is called the annual interest rate or the quoted ratee.g., a 5% per
annum interest means r = 0.05. By default, all interest rates will be on or con-
verted to a per annum basis. For this reason, we sometimes refer to r simply
as the interest rate rather than the annual interest rate. Interest rates appear in
numerous settingssavings accounts, certificates of deposit, credit cards, auto
loans, mortgages, treasuries, bonds, etc.
Remark 2.1. Bear in mind that the interest rate used for lending need not equal
the interest rate employed for borrowing. However, in later modeling, we shall
assume that the two rates are equal (e.g., see page 84).

We shall also switch freely between expressing r as a percent and decimal.

It is possible to have r > 1 (interest rate of over 100% per year) or r < 0, which
can be interpreted as a bank charging you for holding your principal. For sim-
plicity, however, we abide by the following:
Unless stated to the contrary, assume that r is a positive constant.
Though r is constant by default, later in the chapter (e.g., Section 2.5), we shall
study models where r varies discretely and continuously with time. When the
interest rate r is a function of time, it is common practice to express this as
r (t)an abuse of notation that should not cause undue confusion.
Interest rates can, of course, be quoted for any time span (week, month,
etc.). For example, an interest rate of 12% per year is mathematically the same
as 1% per month. More generally, if we divide a year into k equal-size interest
periods, then
r
interest rate per interest period = .
k
Exact Interest, Ordinary Interest, and Bankers Rule
The exact time of a time interval measures the length of the interval in days,
but excludes the first day. Exact interest is interest computed using 365 days
in a year or 366 days for leap years. Credit card companies tend to use exact
time and exact interest. Ordinary interest is interest calculated using 360 days
in a year with 30 days in each month. Banks usually lend using exact time and
ordinary interest, which has come to be known as Bankers Rule.
2.2.2 Required Return Rate and the Risk-Free Rate
We always assume that when an investor commits her money for a specific
period of time, whether to a security, portfolio, or start-up, she expects to be
compensated. An investors required rate of return over an investment period is
then the interest rate the investor demands as compensation for the following:
Opportunity cost: Since lending prevents an investor from using that money
for other investment opportunities, the investor requires compensation for
her money being tied up.
Inflation: Since inflation erodes the value of money, the investor requires
compensation that covers the impact of inflation.
Risk: Since there is a nonzero probability that earnings promised to the in-
vestor will not materialize or that the investor can lose some or all of her
money, the investor requires compensation for the risks of the investment.
Unless stated to the contrary, we assume that no compensation to cover taxes and
transaction costs is part of a required return rate. It is messy to include these
items in an introduction to mathematical finance, not to mention that tax
laws and transaction costs change. Readers are referred to Reilly and Brown
[16, Chap. 1] for a detailed discussion of the required return rate.
In the absence of inflation and risk, the required return rate is called the real
risk-free rate and denoted rreal . It is a compensation purely for opportunity cost.
If there is no risk, but you have inflation and an opportunity cost, then the
required return rate is termed the nominal risk-free rate or, simply, the risk-free
rate. When the real risk-free rate is intended as opposed to the risk-free rate,
we shall indicate so explicitly.
Notation. Let r denote the risk-free rate.
There is a simple relationship among rreal , r, and the inflation rate i. Assume
that you invest F0 in a riskless asset over 1 year. Your required return rate is r,
which compensates you for opportunity cost and inflation. Specifically, your
compensation for opportunity cost a year from now is rreal F0 . However, a year
from now, the value of your compensation rreal F0 for opportunity cost will
reduce by i (rreal F0 ) due to inflation. Furthermore, your initial investment will
also reduce in value by i F0 due to inflation. Your required return rate amount
r F0 beyond your initial investment should then be
r F0 = i F0 + rreal F0 + i (rreal F0 ).
Consequently, we obtain a formula for the real risk-free rate:

ri
rreal = .
1+i
A common proxy (i.e., substitute or model) for the risk-free rate r is the
coupon rate of a US Treasury. The specific type of US Treasury chosen in ap-
plications depends on the time horizon over which an analysis is conducted.
In the modeling of derivatives, however, traders typically choose LIBOR as a
proxy for r (see Hull [9, p. 74] for more).
When inflation constitutes a major portion of the market risk-free rate r,
sometimes r is even called the inflation rate. It is also possible for the inflation
rate to be above the market risk-free rate, which, for instance, can be due to
the government lowering interest rates to increase liquidity. Hence, one can-
not always assume r i, but would expect it to hold under normal market
conditions.
2.2.3 Total Return Rate
Receiving an interest rate of 4% per year on a $20,000 investment means that

over 1 year, say, starting at time 0, you get
r F0 = 0.04 $20, 000 = $800.
In other words, your investment would grow from $20,000 to $20,800 over 1
year. The return rate R(0, 1) on your investment over 1 year is the fractional
percentage change
$20, 800 $20, 000

R(0, 1) = = 0.04 = r.
$20, 000
If you put F(t0 ) today in an investment that does not pay you any income and
the value of the investment at a future time t f = t0 + is F(t f ), then the total
return rate on your investment from time t0 to t f is defined to be
F ( t f ) F ( t0 )
R ( t0 , t f ) =
F ( t0 )
with the total return amount defined by R(t0 , t f ) F(t0 ).

The total return rate will not necessarily equal the interest rate. First, on
mathematical grounds the interest rate is always positive, while the total re-
turn rate can be negative. Second, on financial grounds, the return rate is con-
cerned only with the initial and final values of the investment and so is a per-
formance measure of the investment. However, it is possible to apply an inter-
est rate during each period into which a time span is divided, i.e., the interest
rate is involved with the evolution of F(t0 ) to the final value F(t f ). This will be
made explicit when we look at simple and compound interest.
In general, we formalize the total return rate on a per-unit basis and with
a cash dividend, i.e., an income.2 Suppose that your investment has a current
per-unit market value of V (t0 ), e.g., the price of a stock per share, and per-
unit market value V (t f ) at a final time t f > t0 . Assume that the value of an
investment at any point in time is nonnegative, i.e., there is no liability:
V ( t) 0 ( t 0). (2.1)
Assume that the investment pays a per-unit cash dividend of D (t0 , t f ) dur-
ing the interval [t0 , t f )e.g., a cash payout per share by a company to share-
holders.
Several clarifying remarks are needed about cash dividends:
For simplicity, we do not include any cash dividend at t f , but tally it as part
of the subsequent time interval starting at t f .3
It is also common practice to assume that D (t0 , t f ) excludes any income
such as interest from the cash dividend during [t0 , t f ). This is not a serious
concern for sufficiently short investment time intervals. We also exclude
complications like share splits and noncash payouts.
When an investment pays out a cash dividend, it has lost value by the
amount of dividend. The market value V (t f ) is then the ex-dividend (with-
out dividend) value and the cum-dividend (with dividend) value is
V c ( t f ) = V ( t f ) + D ( t0 , t f ) .
In the case of a cash dividend-paying stock, there is actually an ex-dividend

date, which is the cutoff date to be eligible for a declared cash dividend. It is
actually the close of trading on the trading day before the ex-dividend date.
The stock is said to be traded cum-dividend before the ex-dividend date and
ex-dividend after that date. For this reason, when modeling, the value of the
2 A dividend does not have to be in the form of cash. It can be a stock dividende.g., a company can
pay you additional (typically, fractional) shares for each share of company stock you own.
3 This bookkeeping for the cash dividend makes it convenient mathematically when considering rein-
vesting dividends to buy more units of the investment over consecutive time intervals.
stock is adjusted downward by the dividend amount on the ex-dividend

date, not on the payment date. Stock price data sets like Yahoo! Finance
have a column for adjusted prices, where the adjustments are for cash div-
idends and stock splits. The ex-dividend dates of stocks typically do not
coincide with an exact quarter and are not the same for all companies.
For securities like stocks and bonds, the cash dividends flow in discretely
e.g., quarterly, semiannually, and even annually in some cases. The divi-
dend stream for a sufficiently broad stock index is often modeled as con-
tinuous.
Expressing the per-unit total return amount on your investment from t0 to
t f as a percent R(t0 , t f ) of the initial value V (t0 ), we obtain
R ( t0 , t f ) V ( t0 ) = V ( t f ) V ( t0 ) + D ( t0 , t f ) .

return amount capital gain cash dividend
The spread V (t f ) V (t0 ) is called a capital gain. Note that a negative capital
gain is a capital loss. Equivalently,
V ( t f ) V ( t0 ) D ( t0 , t f ) V c ( t f ) V ( t0 )
R ( t0 , t f ) = + = . (2.2)
V ( t0 ) V ( t0 ) V ( t0 )

capital-gain return dividend yield
This is called the total rate of return or holding-period return of the investment
from t0 to t f . We shall often refer to R(t0 , t f ) simply as the return rate and
at times will even refer to R(t0 , t f ) as the return when it is clear from the con-
text that a rate is intended as opposed to the return amount R(t0 , t f ) V (t0 ).
Note that if your ownership in the investment consisted of n units (shares),
then the return rate is still given by (2.2) since the numerator and denominator
of each term would be multiplied by n and so n would drop out.
Notation. When the return rate depends on the length of [t0 , t f ] rather than
on the location of [t0 , t f ] on the positive time axis [0, ), we set
R ( t0 , t f ) = R ( ) .
D (t0 ,t )
The ratio V (t )f in (2.2) is called the dividend yield and represents the per-
0
unit cash dividend from the investment as a percent of the initially invested
V (t f )
capital V (t0 ). Additionally, we refer to the ratio V ( t0 )
as the gross return from t0
to t f .4 It expresses the final value V (t f ) as a percent of the initial value V (t0 ).
4 V (t f )
Some authors call V ( t0 )
the return rate, but we shall not abide by that usage.
Example 2.1. Suppose that after 1 year, the return rate on your investment is
50%. Then the gain to you, beyond your initial investment, is 50% of your
initial investment. If the return rate is 100%, then you have a complete loss.
If the return rate is 200%, then your gain is twice the initial investment, i.e.,
your initial investment tripled in value over the year.

Finally, observe that the return rate becomes random if the future value
V (t f ) and/or the cash dividend D (t0 , t f ) is random. Almost all the return
rates we encounter in this chapter are nonrandom, while all the return rates
in Chapter 3 are random.
2.3 Simple Interest
A principal of $1, 000 held for a year at a 12% interest rate has a simple interest
of $120 at the end of 1 year. This amount is the same as adding 12 monthly
interests of $10, each of which is obtained from a monthly interest rate of 1%.
For a time span of years, if we assume that the interest rate r is applied only
to the principal F0 , then
(simple interest amount earned or owed over years) = r F0 . (2.3)
If an annual simple interest rate is applied over multiple years (or periods)
to a principal, then at the end of each year (or period), interest is applied only to
the principal and the entire balance is reinvested back into the account. In other
words, all interest accrued at the end of each period or year is carried forward
without gaining interest. Under simple interest growth at rate r, a principal F0
increases to the following amount at years from the present:

F ( ) = F0 + r F0 = 1 + r F0 , (2.4)
where, by a slight abuse of notation, we write F( ) instead of F(t0 + ) since

the value depends on the length of the time interval.
Example 2.2. Suppose that an account has $700 and pays 4% per annum. Ap-
plying a 4% annual simple interest growth to the $700 for 1 year yields an
interest of 0.04 $700 = $28 and a total amount accrued of

$700 + 0.04 $700 = $728.
To obtain simple interest growth of $700 over 2 years, we add to the prin-
cipal a simple interest of 0.04 $700 at the end of the first year and simple
interest of 0.04 $700 at the end of the second year:

$700 + 0.04 $700 + 0.04 $700 = $756,
2.4 Compound Interest 21
or, equivalently,
$756 = 1 + 0.04 2 $700. (2.5)
Note that a 4% annual simple interest growth applied to $700 for 2 years is the
same as applying 8% per 2 years.

Investing $700 under simple interest growth of 4% per annum yields a future
value of $756 2 years from now. Conversely, the present value of $756 under 4%
annual simple interest discounting is $700. In general, if at the current time,
you invest (or borrow) a principal F0 under simple interest growth at an inter-
est rate r applied over years, then the amount of money you receive (or owe)
at the end of the time span is called the future value of F0 and given by

future value of F0
= F ( ) = ( 1 + r ) F0 . (2.6)
at years from now
The principal F0 is called the present value of the future amount F( ):

present value of the amount F( ), F( )
= F0 = . (2.7)
which occurs years in the future 1 + r
The quantity (1 + r )1 is called a discount factor since it reduces the amount

F( ) at the end of the time interval to the amount F0 at the start of the interval.
In the context of (2.6), we sometimes call the interest rate r the simple interest
growth rate of the principal F0 , while in the setting of (2.7), we call r the simple
interest discount rate on the future value F( ). The return rate when F0 grows
under simple interest r over years is then
F ( ) F0
R( ) = = r . (2.8)
F0
2.4 Compound Interest
We saw above that under simple interest for 2 years, an account with $700 at
4% per annum will grow to

$700 + 0.04 $700 = $728 (2.9)
at the end of the first year and to $756 at the end of second year, after the
interest of $28 for the second year is added. However, there is a way to accu-
mulate more money over the same 2 years using the same simple interest rate.
Assume that at the end of the first year, you withdrew the $728, closed the
account, and immediately used the $728 as principal to open another simple
interest account paying the same interest rate. Then a year later, i.e., at the end
of the second year, the total you would accrue is
$728 + (0.04 $728) = $757.12, (2.10)
which is greater than the original total of $756! This type of growth is called
compound interest. In fact, an annually compounded account earning 4% per
annum over 2 years would earn you the latter amount without you needing to
engage in the previous inconvenient strategy.
Using (2.9), we can rewrite (2.10) as

$700 + 0.04 $700 + 0.04 ($700 + 0.04 $700) = $757.12. (2.11)
Equation (2.11) summarizes exactly how the growth process works: annual
compounding of a principal of $700 over 2 years at an interest rate of 4% means
that one applies 4% simple interest to $700 at the end of the first year and then
applies 4% simple interest again at the end of the second year to the entire
balance (principal plus interest) carrying forward from the end of the first year.
Rewriting (2.11) as
$757.12 = (1 + 0.04)2 $700 (2.12)
yields the standard form for two annual compounds.
Let us extend (2.12) to a finite number of compoundings. In general, com-
pound interest occurs when the time span is divided into multiple periods, and
simple interest is applied over each period to the balance at the end of the pe-
riod. We assume that the entire balance at the end of each period is reinvested back into
the total being accrued, i.e., no money is withdrawn and no extra money is added. For
mathematical modeling purposes, we also treat the end of a period as equiva-
lent to the start of the next period.
2.4.1 Compounding: Nonnegative Integer Number of Periods
Assume that an account with an initial amount F0 (principal) pays an interest

rate of r. Divide a year into k interest periods, each of equal length:
1
1 prd = yr.
k
In a compound interest setting, the end of each period marks when interest is
applied to the balance from the start of the period. Consequently, we shall refer
to each such period as an interest period, compound interest period (when being
explicit), or a compounding period.
Unless stated to the contrary, assume that the date when the prin-
cipal is deposited coincides with the start of an interest period.
Following the structure of (2.12), we now compute the future value to which
the principal F0 will grow under k-periodic compounding at interest rate r over
n interest periods, where n is a nonnegative integer. Since n periods correspond
to n/k years, the future value at the end of the nth period is F(n/k). However,
in compound interest theory, the emphasis is on the number n of periods over
which compounding occurs, rather than the number of years. For this reason,
the future value is written as a function of the number of periods as follows:
n
F = Fn .
k
At the end of the first period, apply simple interest to F0 to obtain the future
value F1 to which F0 grows over the first period:
r r
F1 = F0 + F0 = 1 + F0 .
k k
Now, do not take out any of the money. Instead, reinvest the entire amount
F1 in the account at the end of the first period until the end of the second
period.
At the end of the second period, apply simple interest to F1 to get the future
value F2 to which F1 grows over the second period:
r r 2
F2 = F1 + F1 = 1 + F0 .
k k
Note that compound interest occurs since interest was added to the whole
F1 , yielding interest on the principal F0 and interest on the interest (r/k) F0 .
Next, reinvest the entire amount F2 in the account at the end of the second
period until the end of the third period.
Continuing the above process, at the end of the nth period, apply simple
interest growth to Fn1 to obtain the future value Fn to which Fn1 grows
over the nth period:
r r n
Fn = F n 1 + F n 1 = 1 + F0 , n = 0, 1, 2, . . . .
k k
We have established the following: Under k-periodic compounding over n inter-
est periods at an interest rate r, a principal F0 will increase to the value Fn at the end
of the nth interest period:
r n
Fn = 1 + F0 , n = 0, 1, 2, . . . , (2.13)
k
where kr is the periodic interest rate. Observe that Fn depends on the size of the
time interval over which the compounding occurs. This is because the interest
rate is constant for the n periods.
In (2.13), we call Fn the future value of F0 under k-periodic compounding over

n periods at interest rate r. The present value of Fn under the above compound-
ing is defined to be F0 .
Example 2.3. Borrow $1, 000 for a year at 12% interest rate. Applying this in-
terest with monthly compounding yields a balance due of
12
0.12
F12 = 1 + $1, 000 = $1, 126.83.
12
The interest owed is then $1, 126.83 $1, 000 = $126.83, which is more than the
$120 due when simple interest is applied.

Example 2.4. (Moneys Growth Under Different Compounding Periods) In-

vest $1, 000 at an interest rate of 7% and consider monthly, weekly, and daily
compounding. Determine the future values after 2 years.
Solution. We have F0 = $1, 000, r = 0.07, = 2, and k = 12 (monthly), 52

(weekly), and 365 (daily). The respective number of compounding periods is
then 24 (monthly), 104 (weekly), and 730 (daily). By (2.13) on page 23, the fu-
ture values at the end of 2 years are
F24 = $1, 000 1.14981 = $1, 149.81 (monthly compounding)

F104 = $1, 000 1.15017 = $1, 150.17 (weekly compounding)
F730 = $1, 000 1.15026 = $1, 150.26 (daily compounding).

2.4.2 Compounding: Nonnegative Real Number of Periods
Suppose a principal of $10, 000, 000 undergoes monthly compounding at 10%

per annum over a time span of 15.36 mth. What is the principals value at the
end of the time span? First, view 15.36 mth as 15 mth + 0.36 mth. By Equation
(2.13), the value at the end of the first 15 months is

0.1 15
F15 = 1 + $10, 000, 000 = $11, 325, 616.82.
12
To how much will F15 grow during the remaining 0.36 mth? For the partial
interest period, assume that a bank applies simple interest growth to F15 , which
yields a total of

0.1

F15.36 =
F15 + 0.36 = 1 + 0.36 F15 = $11, 359, 593.67. (2.14)
12
However, it may concern the reader that compounding occurs over the first
15 mth, but then stops during the remaining 0.36 mth, and is replaced by sim-
ple interest growth. We claim that the latter is actually an approximation of the
exact mathematical compounding that should be applied during the partial
month. We apply fractional compounding to F15 during the remaining 0.36 mth,
which gives
0.36
0.1
F15.36 = F15+0.36 = 1+ F15 = $11, 359, 503.48. (2.15)
12
In this example, we see that the accrued total in (2.14) is higher by $90.19 than
the total in (2.15) obtained from exact modeling. The bank would be paying
more interest if (2.14) is used.
We now present a theoretical basis for (2.15) and the approximation used
in (2.14). First, we shall introduce the key defining mathematical property of
compound interest as in the treatment by Kellison [10, Sec. 1.5]. For an integral
number of interest periods, Equation (2.13) shows that
r m+n r m r n
Fm + n = 1 + F0 = 1 + 1+ F0 ,
k k k
where m and n are nonnegative integers. We denote the compound interest
growth function over n interest periods by
r n
G (n) = 1 + ,
k
where
r
G (0) = 1, G (1) = 1 + , G (n) > 1 for n = 1, 2, . . . .
k
The inequality G (n) > 1 for positive integers n means that the principal will
increase for compounding over at least one interest period. The compound
interest growth function satisfies:
G ( m + n ) = G ( m ) G ( n ). (2.16)
In other words, compound interest is such that compounding a principal F0

over m + n interest periods is the same as compounding F0 over n interest pe-
riods and then compounding the balance at the end of the nth interest period
over the remaining m interest periods. Of course, one can interchange m and
n. Equation (2.16) embodies the core multiplication property of compound in-
terest.
We then extend (2.16) to a more general defining mathematical property of
compound interest, one applicable to a nonintegral number of interest periods.
Specifically, for any nonnegative real number x, a principal F0 is said to grow
to the value
F x = G ( x ) F0
by k-periodic compounding over x interest periods at interest rate r if the growth
function G ( x ) satisfies the following properties:
G ( x + y) = G ( x ) G (y) for all real numbers x 0 and y 0,

G (0) = 1,
r
G (1) = 1 + ,
k
G ( x ) > 1 for all real numbers x > 0.
(2.17)
The top equation in (2.17) generalizes (2.16) to a nonintegral number of pe-

riods. The same intuition carries over from the integral case: compounding a
principal F0 over x + y interest periods to the value
F x + y = G ( x + y ) F0
is identical to compounding F0 for y interest periods to the value Fy = G (y)F0

and then compounding Fy over the remaining x interest periods to the value
G ( x )Fy . The equation G (0) = 1 in (2.17) states that no growth occurs when
there is no interest period, while G (1) = 1 + kr means that the growth over
one interest period is given by simple interest (as we have done all along).
Finally, we require the condition G ( x ) > 1 for all x > 0 because we assume
that compound interest growth increases the principal over a nonzero interest
period, even if it is fractional.
Let us now solve for the growth function satisfying (2.17). For mathemati-
cal modeling reasons, we shall assume that G ( x ) is differentiable. Applying a
trick similar to the one used in deriving an exponential function, we first con-
sider the derivative of the growth function at x. Using the limit definition of a
derivative, we find (Exercise 2.30)
G ( x ) = G ( x ) G (0). (2.18)
Dividing by G ( x ), which is allowed since G ( x ) > 0 for all x 0, and recalling

that G ( x )/G ( x ) is the derivative of ln G ( x ), we obtain
d ln G ( x )
= G (0),
dx
or, equivalently,
d ln G ( x ) = G (0) dx.
Integrating the equation from 0 to x yields:
ln G ( x ) ln G (0) = G (0) x.
But ln G (0) = ln 1 = 0. Hence:
ln G ( x ) = G (0) x. (2.19)
Equation (2.19) implies:

r
G (0) = ln G (1) = ln 1 + .
k
Inserting G (0) back into (2.19), we find:
r r x
ln G ( x ) = x ln 1 + = ln 1 + .
k k
x
The binomial series 1 + kr with nonintegral x converges for 0 r/k < 1. Ex-
ponentiating both sides of the above equation, we obtain the growth function:
r x
G( x) = 1 + .
k
We summarize the result in the following theorem:
Theorem 2.1. Under k-periodic compound interest at r per annum over a time span
of x interest periods, where x is a nonnegative real number, a principal F0 will increase
to the following future value at the end of the time span:
r x r
Fx = 1 + F0 , 0 < 1, x 0 , (2.20)
k k
where k = 1, 2, . . . .
The periodic interest rate kr in (2.20) is constrained to 0 kr < 1 to assure con-

vergence of Fx when the nonnegative real x is not an integer. We do not need
this requirement when x is a nonnegative integer. In most applications, we
consider 0 < r < k. For example, under monthly compounding (k = 12), the
upper-bound condition expressed in percent means that the compounding in-
terest rate satisfies r < 1200%, which will surely be the case in most applica-
tions.
We also call Fx the future value of F0 at the end of x interest periods from the
present and refer to
Fx r
F0 = x , 0 < 1, x 0 , (2.21)
1 + kr k
as the present value of Fx . The interest rate r is applied as a growth rate in the
future valuing of (2.20) and as a discount rate in the context of (2.21). Since x
interest periods is x/k years, the future value Fx occurs x/k years from now,
i.e., x
Fx = F .
k
The number x of interest periods can always be expressed as the sum of an
integral number n of interest periods and a fraction of an interest period:
x = n + ,
where n is the greatest integer part of x and 0 < 1. For example, x = 15.36
interest periods splits into a sum of n = 15 and = 0.36 interest periods.
We can then rewrite (2.20) as
r n r r
Fx = 1 + F = 1 + Fn , 0 < 1, 0 < 1 . (2.22)
k k k
Here F is the amount to which F0 grows over the fraction of an interest
period, i.e., we have fractional compounding during mth:
r
F = 1 + F0 .
k
For a proper fractional period, i.e., for 0 < < 1, the leftmost equality in (2.22)
states that the fractionally compounded amount F is compounded over n in-
terest periods, and the rightmost equality captures that the accrued amount
Fn is compounded over the fraction of an interest period. The left equality
applies to settings where the start of the time span does not coincide with the
beginning or end of an interest period, while the right equality is for when the
end of the time span is not the beginning or end of an interest period.
The rightmost equality in (2.22) also shows that if the interest rate per inter-
est period r/k is sufficiently small, expanding the binomial series yields:
r r
1+ 1+ . (2.23)
k k
The amount accrued at the end of x interest periods can then be approximated
as follows:
r r
Fx = 1 + Fn 1 + Fn , (0 < 1, 0 r/k 1). (2.24)
k k
Example 2.5. Returning to the example from the start of this section (page 24),
Equation (2.20) shows that a principal of $10, 000, 000 compounded monthly at
10% per annum for 15.36 mth will grow to:

0.1 15.36
F15.36 = 1 + $10, 000, 000 = $11, 359, 503.48.
12
Equivalently,
0.36
0.1
F15.36 = $11, 359, 503.48 = 1 + Fn ,
12
which is the form in (2.22) and the origin of (2.15). Equation (2.14) uses simple
interest, rather than fractional compounding, during the remaining 0.36 mth
and is justified by (2.24):

r 0.1

F15.36 = 1 + Fn = 1 + 0.36 Fn = $11, 359, 593.67
k 12
and since r/k = 0.0083 1, we have

F15.36 F15.36 .

For k-periodic compounding at a constant interest rate r per year, Equation

(2.20) yields that a principal F0 will grow to the following future value over
years or k periods:
r k
Fk = 1 + F0 . (2.25)
k
Example 2.6. (Doubling Your Investment) Suppose that you invest F0 today
in an account with k-periodic compounding at r per year. Find a formula for
how long it will take you to increase your investment to x0 F0 , where x0 > 1.
Does the length of time depend on the initial amount F0 ? In particular, how
long will it take to double an investment of $1, 000 using 6% per annum with
daily compounding? What about $2, 000? Compare with the time it would take
using simple interest growth at the same interest rate.
Solution. We want to find how many years it will take to have F0 grow to
Fk = x0 F0 . By (2.25),
r k
x 0 F0 = 1 + F0 ,
k
which implies that
ln x0
= (r > 0).
k ln 1 + kr
The time does not depend on the initial F0 .
For F0 = $1, 000, x0 = 2, r = 0.06, and k = 365 (daily), we obtain
ln 2
= 11.55,
365 ln 1 + 0.06
365
so it will take 11.55 years. Since the time span will not depend on the initial
investment, we obtain the same answer for $2, 000. For simple interest growth,
x0 1
we have x0 F0 = (1 + r )F0 , which yields = 16.67. The doubling time
r
is 5.12 years longer.

2.4.3 Fractional Compounding Versus Simple Interest
Compound interest is constructed by applying simple interest over each inter-

est period to the balance at the start of the interest period. This may suggest
that simple interest should then be applied over each (proper) fraction of an
interest period to the balance at the start of the fractional interest period (since
simple interest adds over different time segments). However, if simple inter-
est is applied over a given fraction of a period, then it does not account for
the compounding that has to occur over every fraction of the given fractional
interest period. Indeed, a new insight from Theorem 2.1 is that compounding
occurs over every portion of an interest period. For example, start with a balance
F and have it compound for 13 prd. You should not apply simple interest over
the 13 prd because the balance F also has to compound over every fraction of
the 13 prd. For instance, compounding occurs over the first fourth of the 13 prd
and the remaining three-fourths of the period. By (2.20), the correct growth is:
r 13 r 34 ( 13 ) r 14 ( 13 )
F1/3 = 1 + F = 1 + 1+ F .
k k k
By (2.23), using simple interest over a fraction of an interest period would
only yield an approximation under the following condition:
r r
F = 1 + F 1 + F , (0 < < 1, 0 r/k 1).
k k
If we do not use simple interest over a fraction of an interest period, then
why can we apply simple interest over a whole interest period? The reason is
that simple interest over one interest period is equivalent to fractional com-
pounding over the interest period. In fact, decompose an interest period into
any two fractional periods, say, prd and (1 ) prd. Suppose that the bal-
ance at the start of the interest period is F . Then the balance at the end of the
period is:
r r 1 r
F1 = 1 + F = 1 + 1+ F .
k k k
In other words, the simple interest growth of F over 1 prd is the same as frac-
tional compounding of F over prd followed by fractional compounding of

the accrued amount 1 + kr F over the remaining (1 ) prd. We could di-
vide an interest period into an arbitrary finite number of subperiods and still
obtain that simple interest over one period is fractional compounding over the
subperiods:

r m
r j
F1 = 1 + F = 1 + F ,
k j =1
k
where 1 = 1 + 2 + + m and 0 < j < 1 with j = 1, . . . , m.

2.5 Generalized Compound Interest 31
2.4.4 Continuous Compounding
When the number k of compounding periods per year increases without

bound, we have continuous compounding. Applying (2.25), the future value
under continuous compounding is

r k r k
Fcts = lim Fk = lim 1 + F0 = F0 lim 1 + = F0 er , (2.26)
k k k k k
with return rate
R( ) = er 1. (2.27)
Under continuous compounding, $1 will grow to $ er t over the time interval
[0, t]. We can apply the same idea to a security (possibly risky) paying a contin-
uous cash dividend at a constant yield rate q. Suppose that at time 0, you have
1 unit of a security, and as the cash dividend flows in, you continuously buy
more units of the security, i.e., the number of units of the security is continu-
ously compounded at rate q. Then 1 unit of the security at time 0 will grow to
eq t at time t. Consequently, the cum-dividend value of the security at t is
Stc = eq t St , (2.28)
where St is the ex-dividend price at t of one unit of the security. We assume that
St models the market price at t since it discounts the cum-dividend price at the
dividend yield rate: St = eq t Stc . See the discussion on page 18.
2.5 Generalized Compound Interest
2.5.1 Varying Interest and Varying Compounding Periods
This section extends compound interest from a fixed interest rate over a non-
negative real number of compounding periods to discretely varying interest
rates across compounding intervals of different lengths.
We begin with some needed notation. Suppose that you put the amount F0
(principal) in an account for a time interval [t0 , t f ], where t0 0. Assume that
each compound interest period is 1k yr. Divide [t0 , t f ] into n subintervals (not
necessarily of the same length), say,
[ t0 , t1 ], [ t1 , t2 ], ..., [ t i 1 , t i ], ..., [ t n 1 , t n ],
where n is a positive integer, tn = t f . Denote the length of the ith subinterval

[ti1 , ti ] by i , which corresponds to the following number of periods:
i yr = ki prd, i = 1, . . . , n.
Suppose that k-periodic compounding at ri per annum applies during the ith
interval [ti1 , ti ] for i = 1, . . . , n.
We now determine a formula for the amount to which the principal F0 will
grow at the future time tn .
Over the time interval [t0 , t1 ], we have k-periodic compounding at interest

rate r1 of the principal F0 . Applying (2.20) on page 27 with x = k1 , the
value of F0 grows to the following at time t1 :
r k1
F ( t1 ) = 1 + 1 F0 .
k
Reinvest the entire amount F1 in the account.
Over the next time interval [t1 , t2 ], the balance F1 at time t1 is k-periodically
compounded at rate r2 . By (2.20) with x = k2 , the value of F(t1 ) grows to:
r2 k2 r2 k2 r k1
F ( t2 ) = 1 + F ( t1 ) = 1 + 1+ 1 F0 .
k k k
Reinvest F2 in the account.
Continuing this process, we find that over the final time interval [tn1 , tn ],
the balance F(tn1 ) at time tn1 is k-periodically compounded at rate rn .
Again (2.20) yields that F(tn1 ) grows to:
rn kn
F( t n ) = 1 + F ( t n 1 ) .
k
Explicitly:
rn kn r k2 r k1
F( t n ) = 1 + 1 + 2 1+ 1 F0 . (2.29)
k k k
Observe F(tn ) depends on the lengths of the subintervals over which the
various interest rates are constant.
We call F(tn ) the generalized compound interest future value of F0 at time tn .

It is given in product notation as follows:

n
ri ki r
F( t n ) = 1 + F0 , 0 i <1 . (2.30)
i =1
k k
Here F0 is termed the present value of F(tn ).

Special Case
Assume that each interval [ti1 , ti ], where i = 1, . . . , n, coincides with a com-

pound interest period. Then the generalized future value (2.30) becomes:
rn r r
F( t n ) = 1 + 1 + 2 1 + 1 F0 , (2.31)
k k k
where F0 is the principal at the initial time t0 . For a constant interest rate
amount ri = r, where i = 1, . . . , n, we recover the usual k-periodic compounding
formula over n periods or nk years:
n r n
Fn = F = 1+ F0 = F n .
k k
Example 2.7. How much will $1, 000 grow after 1.5 years if it is compounding
semiannually with annual interest rate 7% applied at the end of the first 6
months, 8% at the end of the first year, and 9% at the end of 1.5 years?
Solution. Use the generalized compound interest formula (2.31) with k = 2

(semiannual compounding), n = 3 (number of periods), t0 the current time,
t3 = t0 + 1.5 (future time), r1 = 0.07, r2 = 0.08, and r3 = 0.09. We obtain the
following future value:
r3 r2 r
F ( t3 ) = 1 + 1+ 1 + 1 F0 = 1.045 1.040 1.035 $1, 000
k k k
= $1, 124.84.

2.5.2 APR Versus APY
We begin by showing how the interest rate r relates to the return rate in the
context of compound interest.
At time t0 invest an amount F0 > 0 (principal) in an account that grows
under k-periodic compounding at interest rate r. Suppose that the account
pays no dividend. Let F(t f ) > 0 be the value of the principal at a future time
t f = t0 + . Since a time span of years has k periods, Equation (2.20) on
page 27 yields that the return rate on the principal F0 is:
F( t f ) r k
RC I ( ) = 1= 1+ 1, (2.32)
F0 k
where the subscript C I indicates that the return rate is in the context of com-
pound interest. Note the dependence on the length of the time interval [t0 , t f ].
For n periods, the return rate becomes:
n r n
RC I = 1+ 1. (2.33)
k k
In addition, the interest rate r can be expressed in terms of RC I ( ) as follows:
1
(1 + RC I ( )) k 1
r= . (2.34)
1/k
Equation (2.32) also shows that growing the initial amount V (t0 ) to the value
V (t f ) under compounding at interest rate r is the same as growing V (t0 ) to V (t f )
under simple interest using the return rate RC I ( ) over the time span :
r k
V ( t f ) = 1 + R C I ( ) V ( t0 ) = 1 + V ( t0 ) .
k
The return rate RC I (1) over a year is also commonly used. Equation (2.32)
yields:
r k
RC I (1) = 1 + 1, (2.35)
k
which is also called the annual percentage yield (APY) or effective interest rate and
denoted by RC I (1) = APY. The interest rate r corresponding to RC I (1) is called
the annual percentage rate (APR) or nominal interest rate and is given by:
1
(1 + APY) k 1
APR = .
1/k
The APR should not be confused with the APY, which involves compounding:
k
APR
APY = 1 + 1.
k
For instance, if you are quoted an APR of 12% per annum on a loan, then the
APR arises from a monthly interest rate of APR/12 = 1%. However, since in-
terest on debt typically involves compounding, the APY gives a true reflection
of the interest rate a borrower pays. In this case, the 1% per month interest
compounds to an annual percentage yield of
APY = (1 + 0.01)12 1 = 12.68%,
not 12%. The next example further illustrates the difference.
Example 2.8. If a credit card company quotes only its APR on the card, say,
10.99%, it can cause a consumer to think that after 1 year, the interest amount
on a balance of $2, 500 is
0.1099 $2, 500 = $274.75.

However, this is not correct because it assumes simple interest for the year.
Most credit cards compound daily or monthly (and may add fees). The true
interest rate for a 365-day year with daily compounding is given by the APY:
365
0.1099
APY = 1 + 1 = 11.6148%.
365
The actual interest amount for the year is then the (effective) return amount:
APY $2, 500 = 0.116148 $2, 500 = $290.37,
which is more than the amount $274.75 naively inferred from an APR of
10.99%.

2.5.3 Geometric Mean Return Versus Arithmetic Mean Return
An argument essentially the same as the one used to derive (2.35) shows that,
given any period return rate Rprd , the return rate over a year with compound-
ing at rate Rprd per period is given by:
k
Rann = 1 + Rprd 1, (2.36)
where (as usual) a year is assumed to have k periods. For example, a weekly
return rate of 1% annualizes as follows under weekly compounding:
Rann = (1 + 0.01)52 1 = 67.8%.
We can generalize (2.36) further. First, the return rate (2.33) extends naturally
to compound interest with varying interest rates over a time span of n com-
pounding periods, where each period is 1k yr. Assume that the annual interest
rates used for the various n consecutive compounding periods are r1 , . . . , rn ,
i.e., the interest over the ith period is rki . By (2.31) on page 33, the return rate
(2.33) generalizes to:
F( t n ) rn r r
R C I ( t0 , t n ) = 1= 1+ 1 + n1 1 + 1 1. (2.37)
F0 k k k
Now, assume that you invest F0 in a nondividend-paying investment that
has return rate Ri over the ith period, where i = 1, . . . , n. Explicitly, if Vi1 and
Vi are the respective values of the investment at the start and end of the ith
period, then return rate is
prd Vj Vj1
Rj = .
Vj1
Employing arguments similar to those used to derive (2.37), we can extend

(2.37) from an ith-period interest rate of rki , which is always positive, to the
prd
return rate of Ri , which is not necessarily positive. In other words, we are
generalizing (2.36) to the return rate over n periods by compounding at the
prd prd
respective return rates R1 , . . . , Rn :
n
prd prd prd
Rtot R t0 , t0 + = 1 + Rn 1 + R n 1 1 + R 1 1. (2.38)
k
Note that by (2.1), each factor in the product (2.38) is nonnegative since it is a
gross return:
prd Vj
1 + Rj = 0, ( j = 1, . . . , n).
Vj1
Equation (2.38) shows that the initial investment F0 will grow to the following
value: n
F = (1 + Rtot ) F0 . (2.39)
k
We remind the reader that in (2.39), we assume you do not withdraw or add
any funds to the investment during the n periods. Unless otherwise stated, this
is always our assumption when compounding; see Section 2.4.1.
Now, suppose that n periods ago, an investor put F0 into a nondividend-
paying fund and her investment grew by the process in (2.39) to the current
value F( nk ). She would now like to forecast the behavior of the fund over the
next period using a single mean return rate x. In other words, we seek a
single rate x such that when compounding F0 using x over each of the past
n periods, we obtain the same answer as compounding F0 using the n return
prd prd
rates R1 , . . . , Rn :
n
F = ( 1 + x ) n F0 . (2.40)
k
Comparing (2.39) and (2.40), we see that x must be the geometric mean return
prd prd prd
Rgeom of R1 , . . . , Rk , namely,
1/n
prd prd prd prd
Rgeom = 1 + Rn 1 + R n 1 1 + R 1 1.
We have: n
prd n
F = 1 + Rgeom F0 .
k
The geometric mean return relates as follows to the total return rate:

prd n
Rtot = 1 + Rgeom 1. (2.41)
In general, the geometric mean return does not equal the arithmetic mean
return,
1 n prd
n j
Rprd = Rj .
=1
In fact,
prd
Rgeom R prd .
prd
The two means coincide when the period return rates R j are identical for j =
1, . . . , n. The example below illustrates these two means; see Reilly and Brown
[16, Sec. 1.2.2] for more.
Example 2.9. (Geometric Mean Return Versus the Arithmetic Mean Return)
Suppose that you initially invest $3, 000 in a fund that pays no dividend. As-
sume that the investment decreases to $2, 000 at the end of 1 year, decreases
from $2, 000 to $1, 000 from the end of year 1 to the end of year 2, and increases
from $1, 000 to $3, 000 from the end of year 2 to the end of year 3. Then the total
return rate on your investment over the 3 years is zero.
Let us compare what the arithmetic and geometric mean returns forecast for
the total return rate. The year-to-year return rates over the 3 years are:
$2, 000 1 $1, 000 1

R ( t0 , t0 + 1)) = 1= , R (t0 + 1, t0 + 2)) = 1= ,
$3, 000 3 $2, 000 2
$3, 000
R ( t0 + 2, t0 + 3)) = 1 = 2.
$1, 000
The arithmetic mean return is:

1 1 1 7
Ryr = +2 = = 0.3889,
3 3 2 18
which when compounded annually over the 3 years yields a grossly incorrect
return rate:
(1 + Ryr )3 1 = (1 + 0.3889)3 1 = 1.6792 = 168%!
On the other hand, the geometric mean return is:

1/3
yr 1 1
Rgeom = (1 + 2) 1 1 1 = 0,
2 3
which gives the correct total return rate:
3
yr
R(t0 , t0 + 3) = 1 + Rgeom 1 = 0%.
This is, of course, an illustration of (2.41).

The arithmetic mean return deviated significantly from the geometric mean
because of the high volatility in the yearly return rates. The geometric mean
return was better able to capture this dispersion and, hence, produced the
correct total return rate. The two measures approximate each other when the
return rates do not change significantly from period to period. The geometric
mean return is usually employed for longer time horizons, where there is more
opportunity for higher volatility.

2.6 The Net Present Value and Internal Rate of Return
Compound interest can also be applied to develop the notion of a net present
value. This tool helps with deciding whether to partake in a particular in-
vestment opportunity. The opportunity can be a project, product line, start-up
company, etc. We assume that with an initial capital, the investment opportu-
nity produces net cash flows, i.e., cash inflows minus cash outflows, at different
future dates.
Unless stated to the contrary, assume that each net cash flow takes
taxes into account.
In addition, when the net cash flow on a particular future date is being esti-
mated, the estimate usually reflects activities over the year leading up to the
date. We shall then consider net cash flows on future dates separated by a year.
Furthermore, over the time span that an investment opportunity is analyzed
for its growth potential, we assume that all the annual net cash flows can be
modeled as arising from annual compounding at a constant interest rate. We
refer to this constant interest rate as the compounding growth (annual) rate from
investing in the opportunity.
2.6.1 Present Value and NPV of a Sequence of Net Cash Flows
For concreteness, we shall consider a credible, innovative start-up company.

Suppose that the start-up forecasts that, with an initial investment of $250, 000,
it will generate net cash flows of $155, 000 1 year from now, $215, 000 2 years
from now, and $350, 000 3 years from now. The entire initial capital is assumed
to be invested to produce these cash flows; e.g., none of the money is put aside
in an account unrelated to the companys activities.
An important mathematical function we shall employ is the present value
PV(r ) of the sequence of net cash flows at an annual discount rate r. Note
that in the current context, the present value is expressed as a function of the
discount rate r rather than the number of periods n since n will be fixed and r
will play a more key role. In our example, we have:
2.6 The Net Present Value and Internal Rate of Return 39
$155, 000 $215, 000 $350, 000

PV(r ) = + + .
1+r (1 + r ) 2 (1 + r )3
Equally important will be the spread between PV(r ) and the initial capital.
This function is called the net present value (NPV) at rate r of the sequence of
cash flows and is given by:
NPV(r ) = PV(r ) $250, 000.
For simplicity, we shall write expressions such as NPV(r ) > 0, where it is understood
that the 0 represents a zero amount of cash in the currency of the net cash flows.
Now, an important step in deciding whether to invest in the start-up is to
research the marketplace to find the mean compounding growth rate from
investing in an alternative opportunity with a similar business profile and
riske.g., research competitor companies comparable to the start-up in scale,
risk, business sector, etc. For illustration, assume that the mean compounding
growth rate from investing in an appropriate alternative opportunity is esti-
mated to be:
rRRR = 15%.
We then take rRRR as our required return rate for investing in the start-up.
The current market value of the start-ups projected stream of future net cash
flows is the present value of these net cash flows discounted at the required
return rate of 15%:
$155, 000 $215, 000 $350, 000
PV(rRRR ) = + +
1 + 0.15 (1 + 0.15) 2 (1 + 0.15)3
= $134, 782.61 + $162, 570.89 + $230, 130.68
= $527, 484.18. (2.42)
It is important to observe that when determining the present value in (2.42), no

assumption is being made about reinvesting the net cash flows $155, 000, $215, 000,
and $350, 000.
Equation (2.42) tells us that since the alternative opportunity grows your
investment at 15% per annum compounded annually, such an opportunity can
generate the start-ups forecasted net cash flows if you invest $527, 484.18 in the
opportunity today. To see this, separate the required investment of $527, 484.18
into three parts as follows:
PV(rRRR ) = $527, 484.18 = $134, 782.61 + $162, 570.89 + $230, 130.68 .

ARRR BRRR CRRR
We can then produce the net cash flows by thinking theoretically of the alter-
native opportunity as growing ARRR to a future value of FVARRR (1) at 1 year
out, BRRR to a future value of FVBRRR (2) at 2 years out, and CRRR to a future
value of FVCRRR (3) at 3 years out:
FVARRR (1) = (1 + 0.15) $134, 782.61 = $155, 000
FVBRRR (2) = (1 + 0.15)2 $162, 570.89 = $215, 000
FVCRRR (3) = (1 + 0.15)3 $230, 130.68 = $350, 000.
On the other hand, the credible start-up claims that it can generate the
above future net cash flows with an investment today of less than the amount
$527, 484.18 required by the alternative opportunity, namely, with an initial in-
vestment of only
C0 = $250, 000.
Naturally, investors will favor the start-up since the amount PV(rRRR ) required
by the alternative opportunity is more expensive than the amount C0 required
by the start-up. In other words, the start-up appears favorable when the net
present value at the market required return rate is positive:
NPV(rRRR ) = PV(rRRR ) C0 > 0.
The net present value at the required return rate, namely,
NPV(rRRR ) = $527, 484.18 $250, 000 = $277, 484.18,
then measures how much cheaper (or more expensive, if the difference were negative)
it is to invest in the start-up than in the alternative opportunity. Of course, any final
decision to invest in a start-up will not rely solely on the NPV, but will be com-
plemented with a detailed analysis of the start-ups business plan, innovative
products/services, market environment, management team, etc.
If we had NPV(rRRR ) = 0, i.e., the initial capital required by the start-up
to produce the given future net cash flows was the same as that required by
the alternative opportunity, then there would be no extra value received from
investing in the start-up. In this borderline situation, however, some investors
may still invest in the start-up if, for example, it has more long-term promise.
The start-up would not be attractive to investors if NPV(rRRR ) < 0, i.e., if
it costs more to receive the same future net cash flows from the start-up than
from the alternative opportunity.
2.6.2 The Internal Return Rate
Indeed, the start-up can achieve these future net cash flows with less initial
capital only if it grows the initial capital at a rate greater than the alternative
opportunitys compounding annual growth rate of 15%. The start-ups com-
pounding annual growth rate on the initial capital C0 is called the internal rate
of return (IRR) and denoted rIRR . To determine the start-ups IRR, we must find
the interest rate rIRR that generates the forecasted net cash flows starting from
C0 = $250, 000:
C1 = $155, 000 end of year 1

C2 = $215, 000 end of year 2
C3 = $350, 000 end of year 3.
(2.43)
First, separate $250, 000 into three amounts given by the present values of the
net cash flows $155, 000, $215, 000, and $350, 000 at the unknown discount rate
rIRR . The sum of these individual present values is the present value PV(rIRR )
of the sequence of net cash flows. Explicitly:
$155, 000 $215, 000 $350, 000

$250, 000 = + + = PV(rIRR ). (2.44)
(1 + rIRR ) (1 + rIRR ) 2 (1 + rIRR )3

AIRR BIRR CIRR
Then the future values at rate rIRR of the three portions of the $250,000 in (2.44)
yield the desired future net cash flows. Specifically, the future value of AIRR at
1 year out is $155, 000, of BIRR at 2 years out is $215, 000, and of CIRR at 3 years
out is $350, 000. It suffices then to find the IRR by solving (2.44) for rIRR . Note
that (2.44) is equivalent to the vanishing of the net present value at the rate
rIRR :
NPV(rIRR ) = PV(rIRR ) $250, 000 = 0. (2.45)
Employing a software, we find that an approximate solution of (2.44) or,
equivalently, (2.45) is:
rIRR = 0.652811.
Note that inserting this IRR into (2.44) actually produces $250, 000.04, which,
of course, is not the exact value $250, 000 due to the approximate value of rIRR .
In other words, decomposing the start-up capital approximately as
$250, 000 $93, 779.63 + $78, 703.14 + $77, 517.27 = PV(0.652811)

AIRR BIRR CIRR
and future valuing each term by compounding annually at the rate rIRR will
yield the desired stream of net cash flows.
The start-ups IRR of 65.2811% compounded annually exceeds the alterna-

tive opportunitys compounding annual growth rate of 15%, which makes the
start-up favorable. If it turned out that the IRR were 15%, then the start-ups
growth rate would be no better than that of the alternative opportunity in the
market (borderline case). If, on the other hand, the IRR were less than 15%, the
start-up would be unattractive to investors (start-up not favorable).
In our example, we have: rIRR > rRRR if and only if NPV(rRRR ) > 0. In other
words, the IRR basis for deciding whether to favor the start-up is equivalent,
in this example, to the choice being based on the NPV. We have to be careful to not
generalize this observation widely. The examples equivalence of the IRR and NPV
criteria is actually based on (2.44) or (2.45) having a unique positive solution
and on the net present value being a strictly decreasing function. These two
requirements need not hold in general. We address these issues next.
2.6.3 NPV and IRR for General Net Cash Flows
Extend the previous example to a general sequence of net cash flows. Suppose
that you are considering a new investment opportunity requiring an initial
capital of C0 > 0 to generate future net cash flows,
C1 , C2 , ..., Cn ,
at respective future years 1, 2, . . . , n.

Making no assumptions about reinvesting the net cash flows C1 , C2 , . . . , Cn ,
we see that the present value of this sequence of cash flows at the compound-
interest discount rate of r is:
C1 C2 Cn
PV(r ) = + + + , (r > 0). (2.46)
(1 + r ) (1 + r ) 2 (1 + r ) n
The net present value of the net cash flows is the cost of the alternative invest-
ment opportunity minus the cost of the new investment opportunity:
NPV(r ) = PV(r ) C0 , (r > 0, C0 > 0 ). (2.47)
As before, denote the required return rate of the new investment oppor-
tunity by rRRR . Recall that rRRR is the mean compounding (annual) growth
rate from investing in an alternative opportunity in the marketplace with busi-
ness profile and risk similar to the new investment opportunity. An NPV-based
decision-making rule about whether to invest in the new opportunity is as follows:
If NPV(rRRR ) > 0, then the new investment opportunity is cheaper than the
alternative investment and so is favorable.
If NPV(rRRR ) < 0, then the new opportunity is more expensive and not
favorable.
If NPV(rRRR ) = 0, then the cost of the new opportunity is the same as the
alternative investment and it is borderline whether to invest.
As noted in Section 2.6.1, even with a robust NPV estimate, a real-world busi-
ness decision about whether to invest in a new opportunity will not use the
NPV as the only measure. One has to factor in the business environment, ex-
perience of the management team, etc.
An IRR of the new investment is a positive solution, r = rIRR , of the following
equation:
C1 C2 Cn
0 = NPV(r ) = C0 + + + + . (2.48)
(1 + r ) (1 + r ) 2 (1 + r ) n
Equation (2.48) is equivalent to a real polynomial, so we are seeking the pos-
itive roots of such a real polynomial. Without loss of generality, suppose that
the real polynomial has degree k and is of the following form:5
ak r k + ak1 r k1 + + a1 r + a0 = 0, ( a k = 0). (2.49)
There is no general formula for the real solutions of (2.49) for all positive inte-
gers k.
Perhaps the most cited general result about the number of positive solutions
of (2.49) is Descartess Rule of Signs. Before stating this result, we gather some
notation. Let N+ denote the number of positive solutions of (2.49), where we
count the solutions with multiplicity. For example, the polynomial,
r2 10r + 25 = (r 5)2 = 0,
has N+ = 2, corresponding to two positive solutions r = 5 counted with multi-

plicity. Let Nsgn be the number of sign changes in the ordered sequence of the
coefficients in (2.49):
a k , a k 1 , . . . , a 1 , a 0 . (2.50)
Since the zero coefficients do not contribute to a sign change, it suffices to con-
sider the sign changes due to the ordered nonzero coefficients.
Theorem 2.2. (Descartess Rule of Signs) The number N+ of positive solutions

of (2.49) is the number Nsgn of sign changes of its ordered coefficients in (2.50) or is
Nsgn minus an even positive integer. Specifically, N+ equals either Nsgn , Nsgn 2,
Nsgn 4, . . . , Nsgn 2(n 1), or Nsgn 2n for some nonnegative integer n.6
5If ak = 0, then simply apply the same discussion to the lower degree polynomial.
6Using N+ Nsgn , the reason a nonnegative even integer is subtracted from Nsgn in the theorem is
because N+ and Nsgn have the same parity, i.e., N+ is even (odd) if and only if Nsgn is even (odd). This
Proof. See Meserve [14, p. 156] and Wang [17] for a proof.

For example, the polynomial equation
r5 r2 + r 1 = 0
has three sign changes in its ordered nonzero coefficients: +1, 1, +1, 1. By
Theorem 2.2, this polynomial equation has either 3 or 1 positive solutions.
The IRR Equation (2.48) is equivalent to a polynomial equation of the form
(2.49) with ordered coefficients (2.50). By Theorem 2.2, if these ordered coeffi-
cients have one sign change, then there is at most one positive solution. If, in
addition, you can prove that the polynomial equation has at least one positive
solution, then this solution is the unique positive solution and the desired IRR.
In the example of the start-up, Equation (2.45) is equivalent to a cubic equation:
p(r ) = 250, 000 r3 595, 000 r2 225, 000 r + 470, 000 = 0. (2.51)
There is one sign change, so there is at most one positive solution. Since p(r ) >
0 at r = 0 and p(r ) as r , its graph must cross the positive r-axis,
which means that p(r ) must have at least one positive solution. Hence, the
cubic equation has a unique, positive solution, which is the desired rIRR . Using
a software, we found the approximate positive solution to be rIRR = 0.652811.
We also observed that for the required return rate of rRRR = 15%, the IRR
criterion to favor the start-up, namely, rIRR > rRRR , is equivalent to the NPV
criterion of NPV(rRRR ) > 0. This is not true in general, but holds in the example
because the function NPV(r ) in (2.45) is strictly decreasing. The next result
shows when the situation of the example holds.
Theorem 2.3.
1) Suppose that all the future net cash flows are positive. Then NPV(r ) is a strictly
decreasing function of r and, if there is an r = rIRR , then rIRR is the only IRR.7
2) If there is an rIRR and NPV(r ) is strictly decreasing, then the IRR and NPV
decision-making criteria are equivalent:
a) rIRR > rRRR if and only if NPV(rRRR ) > 0.
b) rIRR = rRRR if and only if NPV(rRRR ) = 0.
c) rIRR < rRRR if and only if NPV(rRRR ) < 0.
Proof.
1) Since Ci > 0 for i = 1, . . . , n, the derivative of the NPV function satisfies:
implies Nsgn N+ is a nonnegative even number, i.e., N+ = Nsgn even. In particular, N+ is either
Nsgn , Nsgn 2, . . . , Nsgn 2( n 1), or Nsgn 2n for some nonnegative integer n.
IRR > 0.
7 By definition, we assume r
d C1 C2 Cn
NPV(r ) = 2 n < 0.
dr (1 + r ) 2 (1 + r ) 3 ( 1 + r ) n +1
Consequently, the function NPV(r ) is strictly decreasing and, hence, if the
graph of NPV(r ) crosses the positive r-axis, i.e., there is an IRR, the graph
will do so only once, namely, at a unique value rIRR .
2) We are given that an IRR exists: rIRR > 0. Since NPV(r ) is strictly decreasing,
we have r2 > r1 > 0 if and only if NPV(r1 ) > NPV(r2 ). For part a) choose
r2 = rIRR , r1 = rRRR , and observe that NPV(rIRR ) = 0. For part c) choose r2 =
rRRR , r1 = rIRR . Part b) holds since NPV(rIRR ) = 0.

It is important to note that in a real-world decision-making setting, the NPV

and IRR criteria have aspects not explicitly spelled out in Theorem 2.3. For
example, consider the start-up we have been exploring:
C0 C1 C2 C3 rRRR NPV(rRRR ) rIRR

$250,000 $155,000 $215,000 $350,000 15% $277,484.18 65.28%
The table shows that with an initial investment of C0 , the start-up is attractive
because it has:
Positive future net cash flows C1 , C2 , and C3 that are nontrivial as a percent
of the initial capital and are nontrivially increasing. In particular, C1 is more
than half the initial capital, C2 is about 86% of the initial capital, and C3 is
140% of the initial capital. Moreover, the net cash flow increases by about
39% from year 1 to 2 and about 63% from year 2 to 3.
A nontrivially positive NPV value of NPV(rRRR ) = $277, 484.18, which
makes the start-up much cheaper to invest in than a comparable alterna-
tive opportunity by more than the initial capital.
A quite large compounding growth rate of rIRR = 65.28% compared to the
required return rate of rRRR = 15%, i.e., the rIRR is more than four times
rRRR .
No IRR and Multiple IRRs
The discussion so far assumes a unique IRR. However, complications already

arise in the simple case of net cash flows over 2 years (n = 2), where the IRR
Equation (2.48) becomes a quadratic in r:
C r2 + (C1 2C0 ) r + (C0 + C1 + C2 ) = 0, (C0 > 0). (2.52)

0
a b c
For example, if the net cash flows are C1 = 2C0 and C2 = 2C0 , then the
quadratic reduces to one with no real solution: r2 + 1 = 0. In this case, there
is no IRR.
It is also possible to have multiple IRRs. The quadratic (2.52) has two positive
solutions, say, r1 and r2 , if and only if the following positivity conditions hold:
0 < b2 4ac = C12 + 4C0 C2

b C 2 C0
0 < r1 + r2 = = 1
a C0
c C0 C1 C2
0 < r1 r2 = = .
a C0
These positivity conditions are equivalent to
C12
2 C0 < C1 , < C2 < C0 C1 , (C0 > 0).
4 C0
Choosing C0 = $10, 000, C1 = $25, 000, and C2 = $15, 620, we obtain two IRRs:
r1IRR = 22.76%, r2IRR = 27.24%.
In the case of multiple IRRs, it is possible to construct a modified IRR. Nonethe-

less, in practice it is simplest to work with the NPV when there is no IRR or
multiple IRRs. Additionally, in cases where usage of the NPV and/or IRR are
unclear, it may be wise to hold off from making an investment decision.
Readers are referred to Bodie, Kane, and Marcus [1, Chaps. 5, 6] for a de-
tailed practical discussion of the uses of the NPV, IRR, and other tools in in-
vestment decision-making.
2.7 Annuity Theory
In this section, we continue our study of cash flow sequences by considering

annuities. An annuity is a series of payments made at equal time periods with
interest. Examples of annuities are the payment sequences of Social Security
funds, pensions, car loans, credit card debt, and mortgages. We shall study
annuities with identical payments and a constant interest rate and then gener-
alize them to payments and interest rates that vary discretely in time.
For simplicity, we shall explicitly indicate when the annuity payments vary
and, by default, abide by the following:
Unless stated to the contrary, assume that each annuity payment
is the same amount.
The term of an annuity is the time from the start of the first payment period to
the end of the last payment period. For an ordinary annuity, payments occur at
2.7 Annuity Theory 47
the end of each time period. When the payments occur at the start of each pe-
riod, we have an annuity due, which will not be treated in the text; see Guthrie
and Lemon [8] and Muksian [15] for an introduction.
An ordinary annuity is called simple if, at the end of each payment period,
both a payment and the simple interest on the balance from the beginning of
the payment period are applied. Note that the entire balance from the previous
period is reinvested. Hence, for a simple ordinary annuity, the total accrued at
the end of a payment period has the following form:
(total accrued) = (payment) + (previous balance)

+(simple interest on previous balance). (2.53)
Here previous balance refers to the balance from the end of the previous
payment period, which recall we treat mathematically the same as the start of
the current period. Since the simple interest applied to a previous balance will
yield interest on the principal and interest on the interest, we obtain compound
interest naturally.
Unless stated to the contrary, assume that all loans are simple
ordinary annuities.
2.7.1 Future and Present Values of Simple Ordinary Annuities
Future Value of a Simple Ordinary Annuity
The future value of a simple ordinary annuity is the amount to which the se-
quence of payments of the annuity will grow, taking into account appreciation
due to periodic compounding. We shall see that the annuitys future value is
the sum of the end-of-term future values of the individual payments of the
annuity.
Consider a simple ordinary annuity based on k-periodic compounding at
interest rate r. This divides each year into k equal-length payment periods.
Assume that each payment is the same amount P and the annuity has a term
of n periods, where n is a positive integer. The total accrued at the end of the
ith period will be denoted by Si . We shall apply (2.53) to obtain an expression
for the total amount Sn accrued over the n periods:
At the end of the first payment period, a payment P is made. Since there is
no balance from the beginning of this period, the total accrued at the end
of the first period is:
S1 = P .
Reinvest the entire amount S1 in the annuity.
At the end of the second period, the payment is P , the previous balance
is S1 , and the simple interest earned on the entire reinvested amount S1 is
(r/k)S1 . The total accrued at the end of the second period is then:
r r
S2 = P + S1 + S1 = P + 1 + P.
k k
Reinvest the entire amount S2 in the annuity.
At the end of the 3rd period, the payment is P , the previous balance is S2 ,
and the simple interest earned on S2 is (r/k)S2 . The total accrued is:
r r r 2
S3 = P + S2 + S2 = P + 1 + P + 1+ P.
k k k
Reinvest S3 .
Continuing the above process, at the end of the nth period, the payment is
P , the previous balance is Sn1 , and the simple interest earned on Sn1 is
(r/k)Sn1 . The total accrued at the end of the nth period is:
r
S n = P + S n 1 + S n 1
k
or
r r 2 r n 1
Sn = P + 1 + P + 1+ P + + 1 + P. (2.54)
k k k
Equation (2.54) shows that the future value of a simple ordinary annuity is the
sum of each of the payments future valued to the end of the annuity. To see this, in
(2.54) the future values of these payments are shown from right to left. Explic-
itly, the 1st payment P is at the end of the first period, so its future value at the
end of term (i.e., end of the nth period) is (1 + r/k)n1 P . The 2nd payment P
is at the end of the second period, which has a future value at the end of the
term of (1 + r/k)n2 P . The (n 2)nd payment P has an end-of-term future
value of (1 + r/k)2 P , and the (n 1)st payment P has (1 + r/k)P . The nth
payment P is at the end of the term so it equals its end-of-term future value.
By (2.54), the sum of these future values is Sn .
The right-hand side of (2.54) has a simpler expression. Applying the geo-
metric sum,

m 1 1 xm
a + ax + + ax = a, (m 1, x = 1), (2.55)
1x
with a = P , x = 1 + r/k = 1 (since r > 0), and m = n 1, we obtain:

(1 + kr )n 1
Sn = P (r > 0, n 1).
r/k
Remark 2.2. In actuarial science, the future value Sn is denoted by sn (pro-

nounced s angle n).

We summarize the result of the above analysis in the following theorem.
Theorem 2.4. At the end of n periods, the future value of the simple ordinary annuity
with payments P and k-periodic compounding at r per annum is:

(1 + kr )n 1
Sn = P (r > 0, n = 1, 2, . . . ). (2.56)
r/k
Let us consider how Sn behaves as a function of r 0. Intuitively, we expect

that as the interest rate r increases, the total Sn accumulated after n periods
should increase. Of course, this is not true for n = 1.8 However, for n = 2, 3, . . . ,
Equation (2.54) readily yields:
dSn P r P r n 2 P
= + 2 1+ + + ( n 1) 1 + > 0.
dr k k k k k
It follows that for n 2, the total amount Sn accrued over n periods increases as r
increases. Additionally, for n = 2 we have
d2 S2
= 0.
dr2
However, if n = 3, 4, . . . , then
d2 S n P r n 3 P
= 2 + + ( n 1 )( n 2 ) 1 + > 0.
dr2 k2 k k2
Hence, for n 3, the total amount Sn accumulated over n periods accelerates9 in
value as the interest rate r increases.
Present Value of a Simple Ordinary Annuity
The present value, denoted by An , of a simple ordinary annuity is the amount

needed today, taking interest into account, in order to be able to pay the
amount P at the end of each period for a total of n periods. In particular, an in-
terest per period of r/k is applied at the end of each period to the balance from
the start of that period. The funds are, of course, exhausted by the end of the
last period. For example, if An is a loan, then it would be paid off completely
at the end of the nth period. The total payout would be nP .
8 If there is only one period, then S1 = P (constant) for all r since the principal is added only at the
end of the first period, but the first interest payment occurs at the end of the second period.
9 That is, S is concave up as a function of r (it has an increasing slope).
n
Let us determine a formula for An . The first payment P will be made at

the end of the first period. The present value of that future payment P is then
P
. This is the amount needed at the start of the annuitys term in order
(1 + r/k)
for the amount to grow to P after one period. The present value of the second
P
payment P is since after two compoundings, i.e., at the end of the
(1 + r/k)2
second period, it grows to P . Consequently, at the start of the annuity, the
individual present values of the n payments are given in sequential order as
follows:
P P P

r , 2 , , n .
1+ k 1+ r 1 + kr
k
Then An is the sum of the present values of all payments:
P P P
An =
r + + + n . (2.57)
1+ k 1 + kr
2
1 + kr
Equation (2.57) can be expressed as:
r 1 r 2 r n
An = 1 + P + 1+ P + + 1 + P
k k k
r 1 r 1 r 2
= 1+ 1+ 1+ + 1+ +
k k k

r (n1)
+ 1+ P
k
1 n
1 + kr 1 1 + kr P
= .
1 (1 + r/k)1
The last equality above follows from the geometric series (2.55) with a = P and
x = (1 + r/k)1 and m = n. Further simplification yields:
Theorem 2.5. The present value of a simple ordinary annuity over n periods and with
payments P and k-periodic compounding at r per annum is:
n
1 1 + kr
An = r P (r > 0, n = 1, 2, . . . ). (2.58)
k
Theorem 2.5 gives a formula for the amount needed today at interest rate r
in order to be able to pay out the amount P each period for n periods.
Remark 2.3. The present value An is usually denoted by an and the discount
factor (1 + r/k)1 by in actuarial science.

The Total Number of Periods as a Function of the Payment per Period
We know intuitively that increasing the per-period payments of a loan will

shorten the time it takes to pay off the loan. In fact, we can solve (2.58) for an
exact formula relating the number n of periods to payoff in terms of the inputs
P , An (loan amount), r, and k:

(r/k ) A n
ln 1 P
n= . (2.59)
ln 1 + kr
To understand how n varies with P , we can fix An , r, and k and treat n for-
mally as a function of P given by (2.59). This treatment, of course, will lead
to noninteger values of n, which we round off to find the approximate integer
value. For general values of An > 0, r > 0, and k (nonnegative integer), as P in-
creases, the total number of periods n strictly decreases and the rate of decrease slows
down. In other words, the quantity n as a function of P is convex, i.e., n(P )
is everywhere concave up. Explicitly, though the function n(P ) is a strictly
decreasing function, it has an increasing slope:
dn (r/k) An
= <0 (r > 0)
dP 1
(r/k ) A n
P 2 ln 1 + kr
P

(r/k ) A n
2
d n 2 P (r/k) An
= > 0.
dP 2 (r/k )A n 2 3
1 P P ln 1 + k r

Here we used ln 1 + kr > 0 (since r > 0) and employed (2.58) to conclude that
(r/k) An r n
1 = 1+ > 0.
P k
Note that the quantity (r/k)An is the (simple) interest on the loan at the end
of the first period. An example of n as a function of the per-period payment P
is shown in Figure 2.1.
Present Value of a Perpetuity
A perpetuity is a sequence of cash flows that continues indefinitely. Though

a perpetuity has no future value, a simple-ordinary-annuity perpetuity has a
present value given by the following geometric series:
n
1 1 + kr P
A = lim An = lim r P= r (r > 0), (2.60)
n n
k k
Number of Periods
360
180
106
Monthly Payment
1000 1500 2000
Fig. 2.1 The graph shows the total number of periods n as a function of monthly payments P for a
loan of A n = $162, 412 at 6.25% per annum compounded monthly (k = 12). The loan is paid off in
about 30 years (or 360 months) if the monthly payment is $1, 000. Doubling the payments yields a
payoff time of 8 years and 10 months (or 106 months), which is much less than half of the time for a
$1, 000 monthly payment.
which is an immediate consequence of (2.58). See Section 2.9.1 on page 64 for

a growing perpetuity, i.e., one where the payments P increase at a certain rate.
Relating Future and Present Values of a Simple Ordinary Annuity
If you put aside the amount An today and have it grow by k-periodic com-
pounding with interest rate r, then after n periods, the initial amount will grow
to Sn . In other words, the initial amount An is the present value of the future
amount Sn under periodic compounding. To see this, note that by (2.56) and
(2.58), we obtain:

Sn (1 + kr )n 1 P r n
n = r 1 +
1 + kr k k

n
1 1 + kr
= r P
k
= An , (2.61)
where r > 0. Equation (2.61) shows that an equivalent way of determining the
future value of a simple ordinary annuity is to take the present value of the
sequence of payments and then take the future value of that present value.
2.7.2 Amortization Theory
Amortization is the reducing of a given loan amount (the principal) through a

series of payments over a fixed time span whereby one accounts explicitly for
the portion of each payment that goes toward the principal and the portion to-
ward the interest owed on the loan. The most common amortization is through
a mortgage, which is a loan where the borrower (mortgagor) gives the lender
(mortgagee) a lien on property as security for the repayment of the loan. The
mortgagor has use of the property and the lien is removed when the obligation
is fully paid. A mortgage usually involves real estate.10
What happens if you are amortizing a debt with equal periodic payments
and at some point decide to pay off the remainder of the debt in one lump-
sum payment? This occurs each time a house with an outstanding mortgage is
sold. How much of each periodic payment is used for interest and how much is
used to reduce the unpaid balance of the principal? This issue is also important
because the interest part of the payment may be tax deductible (as it is in the
USA). In order to answer these questions, we must take a close look at the
mathematical structure of amortization.
A loan is paid off with interest through the full sequence of its stipulated
minimal payments. We then model the amount of a loan by the present value
of the entire sequence of its required future payments. Specifically, we model
the loan amount by the present value of a simple ordinary annuity. The initial
amount of the loan is An (principal balance), the payment at the end of each
period is P , and the loan is for n periods. Each period is (1/k)th of a year and
the annual interest rate is r, i.e., the interest applied at the end of each period
is r/k.
Unpaid Principal Balances
We determine the unpaid balance on the principal at the end of each period of
the loan.
For notational simplicity, define
r
y1+ , (r > 0).
k
Then by (2.58), each end-of-period payment can be expressed as:
(y 1) yn (y 1)
P= A n = An . (2.62)
1 y n yn 1
10While a typical mortgage is a loan used to buy a fixed asset like a house or land, which also secures
the loan, a mortgage used to buy movable property such as a mobile home or operational equipment
that acts as security for the loan is called a chattel mortgage or secured transaction.
Denote the initial amount of the loan by:

B0 = A n .
Then the unpaid principal balances at the end of the different periods are given
as follows:
At the end of the first period, an interest (r/k)An is added to the starting
balance An and a payout/withdrawal of P is made. The unpaid principal
balance at the end of the first period is:
r
B 1 = A n + A n P = yA n P .
k
At the end of the second period, an interest (r/k)B1 is added to the balance
B1 from the start of the second period and then a payout/withdrawal of P
is made. The unpaid principal balance at the end of the second period is:
r
B2 = B1 + B1 P = yB1 P = y(yAn P ) P = y2 An (1 + y)P .
k
At the end of the 3rd period, an interest (r/k)B2 is added to the balance
B2 from the start of the 3rd period and then a payout/withdrawal of P is
made. The unpaid principal balance at the end of the 3rd period is
r
B 3 = B 2 + B 2 P = yB 2 P
2 k
= y y An (1 + y)P P
= y3 An (1 + y + y2 )P .
Continuing the above process, at the end of the th period, an interest
(r/k)B1 is added to the balance B1 from the start of the th period and
then a payout/withdrawal of P is made. The unpaid principal balance at
the end of the th period is:
r
B = B1 + B1 P
k
= y An (1 + y + y2 + + y1 )P
1 y
= y A n P
1y
(1 y ) yn (y 1)
= y A n + An , [by (2.62)]
(y 1) (yn 1)
y (yn 1) + yn (1 y )
= An
yn 1
yn y
= n An ( = 1, 2, . . . , n),
y 1
where B0 = An . Hence:
Theorem 2.6. The unpaid principal balance at the end of the th period is given in
terms of An , r > 0, and k by:
n
1 + kr 1 + kr
B = n An , (2.63)
1 + kr 1
where n = 1, 2, . . . and = 0, 1, 2, . . . , n.
The unpaid principal balance at the end of the loans term is Bn = 0.
Amount of Per-Period Payment Toward Interest and Unpaid Balance
At the end of each period, a portion of the payment P is used toward interest
on the loan, the other portion toward reduction of the loans unpaid principal
balance.
Notation. Let:
I = the portion of the payment P at the end of th period that is applied
toward interest on the loan (i.e., the interest payment at the end of
period ).
P = the portion of payment P at the end of the th period that is applied
toward the unpaid principal balance of the loan.
We now express I and P in terms of An and r. The interest payment at the
end of period is

r r 1 + kr n 1 + kr 1
I = B1 = n An , (2.64)
k k 1 + kr 1
where r > 0, n = 1, 2, . . . and = 1, 2, . . . , n.

For the payment P toward the principal, Equations (2.62) and (2.64) yield:
P = P I

(y 1) yn (y 1) yn y1
= An An
yn 1 yn 1
(y 1)y1
= An
yn 1
r 1 + r 1
= k
n An . (2.65)
k 1 + kr 1
As a check, we show that the payments, P1 , P2 , . . . , Pn , toward the principal

add up to the total loan amount An :
n n
(y 1)y1 y1 n
P = y n1
An = An n
y 1 y1
=1 =1 =1
n 1
y1 y1 1
yn
= An
yn 1 y = A n yn 1 y1
=0
= An .
Therefore:
Theorem 2.7. The total interest paid during a loan with n periods and k-periodic
compounding at interest rate r is then:
n n
I = (P P ) = nP An , (2.66)
=1 =1
where n = 1, 2, . . . .
Note that nP is the total amount paid into the loan over the life of the loan and
nP An is the total cost of the loan.
Remark 2.4. If you receive a loan today for the amount An at fixed interest rate
r, fixed payment P per period, and a term of n periods, then the sum n P of
all your future payments adds money at different future times without present
or future valuing them. In fact, the present value of all the future payments is
the loan amount An and the future value is Sn , neither of which is n P . The
meaning of n P is the amount you would, in principle, have to pay the lender
today if immediately after receiving the loan you want to pay the loan off, but
the lender penalizes you by requiring you to pay the principal An plus the
total interest for the full term of the loan. Of course, this is merely theoretical
since the majority of loans would not have such a drastic penalty.

2.7.3 Annuities with Varying Payments and Interest Rates
Applying essentially the same arguments used to establish the future value
annuity Equation (2.54), we can generalize to a simple ordinary annuity with a
sequence of varying payments, P1 , P2 , . . . , Pn , and respective varying interest
rates, r1 , r2 , . . . , rn , over n interest periods that coincide with the payment
periods. We assume k-periodic compounding. The payment P occurs at the end
of the th period, and the interest r is applied at the end of the th period to the balance
from the start of the th interest period, where = 1, . . . , n. Assume that there is no
balance at the start of the first period.
Future Value of a Generalized Simple Ordinary Annuity
The pattern for the future value of a simple ordinary annuity generalized to
varying payments and varying interest rates emerges as follows:
At the end of the first payment period, a payment P1 is made. Because
there is no balance from the beginning of this period, the total accrued at
the end of the first period is:
S 1 = P1 .
Reinvest S1 in the annuity.
At the end of the second period, the payment is P2 , the previous balance is
S1 , and the simple interest earned on S1 is (r2 /k)S1 . The total accrued is:
r r
S 2 = P2 + S 1 + 2 S 1 = P2 + 1 + 2 P1 .
k k
At the end of the 3rd period, the payment is P3 , the previous balance is S2 ,
and the simple interest earned on S2 is (r3 /k)S2 . The total accrued is:
r r r r2
S 3 = P3 + S 2 + 3 S 2 = P3 + 1 + 3 P2 + 1 + 3 1+ P1 .
k k k k
Continuing the above process, at the end of the nth period, the payment is
Pn , the previous balance is Sn1 , and the simple interest earned on Sn1 is
(rn /k)Sn1 . The total accrued is
rn
S n = P n + S n 1 + S n 1
k
or
rn rn r
S n = Pn + 1 + P n 1 + 1 + 1 + n 1 P n 2 +
k r k
r n 1
k
r2
n
+ 1 + 1+ 1 + P1 .
k k k
(2.67)
Observe that, by letting rn+1 = 0 and rewriting (2.67) as

r r rn
S n = 1 + n +1 P n + 1 + n +1 1+ P n 1
k r rn k
r
k

+ 1 + n +1 1+ 1 + n 1 P n 2
k k
r n +1 rn
k
r r2
+ 1 + 1+ 1 + n 1 1 + P1 ,
k k k k
we see that (2.67) can be expressed more compactly as follows:
Theorem 2.8. The future value at the end of n payment periods, which coincide with
the interest periods, of the simple ordinary annuity with payments P1 , . . . , Pn and
k-periodic compounding at respective interest rates r2 , . . . , rn during the consecutive
interest periods is:

n 1 r n +1 j
Sn = 1 + Pn , (n = 1, 2, . . . ), (2.68)
=0 j=0
k
where ri > 0 for i = 2, . . . , n and rn+1 = 0.
When ri = r for i = 2, . . . , n, and Pi = P for i = 1, . . . , n, Equation (2.68) recovers

(2.54) on page 48.
An application of (2.68) to sinking funds is given in Section 2.8.3.
Present Value of a Generalized Simple Ordinary Annuity
Similarly, the present value Equation (2.57) generalizes naturally to the case
of a sequence of payments, P1 , . . . , Pn , and interest rates r1 , . . . , rn . Here the
amount Pi is paid at the end of the ith period, and the interest ri is applied
at the end of the ith period to the balance from the end of the (i 1)st period.
When simple interest at rate r1 is applied at the end of the first period to
the initial amount P1 (1 + r1 /k)1 , we obtain the first payment P1 . Apply-
ing compound interest with rates r1 and r2 at the end of the first and second
periods, respectively, to the initial amount P2 (1 + r1 /k)1 (1 + r2 /k)1 yields
the second payment P2 . Continuing this process gives the initial amount that
will grow to the nth payment Pn . These initial amounts are the present values
of the sequence of payments under compound interest at different rates. Sum-
ming all the present values gives the following present value for the generalized
annuity:
P1 P2
An = + r1
1 + rk1 1+ k 1 + rk2
Pn
+ + r1 ,
1+ k 1 + rk2 1 + rkn
(2.69)
or, more compactly,

n
P
An = r , (2.70)
j=1 1 + k
j
=1
where n = 1, 2, . . . and ri > 0 for i = 1, . . . , n. In the special case where ri = r and

Pi = P for i = 1, . . . , n, Equation (2.70) yields (2.57) on page 50.
2.8 Applications of Annuities 59
Applications of (2.69) to the dividend discount model and bond pricing are
given, respectively, in Section 2.9.1 and 2.10 (see page 68).
Relating Future and Present Values of a Generalized Simple Ordinary

Annuity
We now show that Sn in (2.67) is the generalized future value of An in (2.69)

under the generalized periodic compounding in (2.31) (see page 33). Using
Equation (2.67), direct calculation shows that
Sn
r1 r2 r n 1
1+ k 1+ k 1 + k 1+ rn
k
Pn P2 P1
= r1 r2 rn + + r1 +
1+ k 1 + k 1 + k 1+ k 1 + rk2 1 + rk1
= An .
It immediately follows that the relationship between the future and present
values in Equation (2.61) generalizes to
Sn
An = n , (2.71)
rj
1+
k
j =1
where n = 1, 2, . . . and ri > 0 for i = 1, . . . , n.
2.8 Applications of Annuities
2.8.1 Saving, Borrowing, and Spending
Example 2.10. (Saving During College) A prospective college student plans

to deposit $25 every month in an untouchable savings account, starting the
first of July of the year she enters college until the last deposit on the thirtieth
of June of her graduating year. Assume that she secured a fixed interest rate of
2.25% annually. Assume that the account compounds monthly.
a) Using this average interest rate, estimate how much she would have on July
1st of her graduating year.
Solution. Use the future value Sn in (2.56). We have k = 12 for monthly com-
pounding and since the period is 4 years, we have n = 4 12 = 48 periods,
r = 0.0225, and P = $25. By (2.56),

0.0225 48
1+ 12 1
S48 = 0.0225
$25 = $1, 254.43.
12
b) If her target is to have at least $1, 300 on July 1st of her graduating year,
determine the minimum required interest rate.
Solution. Solving the equation,

r 48
1+ 12 1
$1, 300 = S48 = r $25,
12
implicitly for r (use a software package), we obtain the smallest interest rate
to be r = 4.04%. Note that this is the smallest value of r that works since Sn
is a strictly increasing function of r for natural numbers n 2 (see page 49).

Example 2.11. (Saving for Retirement) Suppose that you open a retirement
fund at the start of a month and you deposit $200 at the end of each month.
If the fund pays 4% per annum compounded monthly, how much would you
accumulate at the end of 25 years?
Solution. This problem deals with the future value Sn in (2.56). For monthly
compounding (k = 12), we have n = 25 12 = 300 periods, r = 0.04, and P =
$200. Equation (2.56) then yields the following future value:
300
1 + 12
0.04
1
S300 = 0.04
$200 = 514.13 $200 = $102, 826.
12

Example 2.12. (Total Paid on Loan) A relative is considering a 20-year loan
of $150, 000 with an interest rate of 8% compounded monthly. Assuming you
hold the loan the entire term and make the minimum payment at the end of
each month, what is the total amount you pay into the loan?
Solution. We have An = $150, 000, k = 12, r = 0.08, and n = 20 12 = 240 peri-

ods, so by the present value annuity formula, we obtain the minimum monthly
payment:
(r/k) An
P= n = $1, 254.66.
1 1 + kr
Since there are 240 months, the total paid is: 240 $1, 254.66 = $301, 118.40.

2.8 Applications of Annuities 61
Example 2.13. (Paying Off Debt) Suppose that you borrow $100, 000 at an an-
nual interest rate of 6% with monthly compounding. For an ordinary annuity
based on this compounding, what is your minimum payment per month to
pay off the loan in 10 years?
Solution. The problem requires the present value An . Since k = 12, there are
10 12 = 120 periods. Using A120 = $100, 000 and kr = 0.06
12 = 0.005, we get:
(0.005) $100, 000

P= = $1, 110.21.
1 (1.005)120

Example 2.14. (How Much Loan Can You Afford) Suppose that you can pay
$1, 495 per month for the next 15 years. What is the largest loan you can afford
at 6.25% per annum with monthly compounding?
Solution. Assume that the first payment is made 1 month from now. We have
n = 15 12 = 180 periods (months), P = $1, 495, and r = 0.0625. The maximum
loan you can afford is:
180
1 1 + 0.0625
12
An = 0.0625
$1, 495 = $174, 359.71.
12

Example 2.15. (Living Off a Lump Sum) Suppose that you inherited $300, 000
and invested it in an account with an annual interest rate of 7% compounded
monthly. For an ordinary annuity based on this compounding, if you want
your inheritance to last 20 years, what is the maximum fixed amount you can
spend from the account per month?
Solution. Using the present value annuity formula with An = $300, 000, k = 12,
r = 0.07, n = 20 12 = 240 periods, we obtain:
(r/k) An (0.07/12) $300, 000

P= =
r n 240
= $2, 325.90.
1 1+ k 1 (1 + 0.07
12 )

2.8.2 Equity in a House
Example 2.16. (House Equity) A couple bought their house 11 years ago for
$225, 000 and put down 10% on the house. On the balance, they took out a
15-year mortgage at 5.75% per annum with monthly compounding. The cur-
rent net market value of the house is its current market value minus all costs
in selling the house today. Suppose that the current net market value is now
$350, 000 and the couple wants to sell their house.
a) How much equity (to the nearest dollar) is in the house today? Equity in a
house is defined as:
equity = (current net market value) (unpaid loan balance).
Solution. The couple puts down 10% or $22, 500 at the start, so the mortgage
is for An = $225, 000 $22, 500 = $202, 500. Since n = 15 12 = 180, r =
0.0575, k = 12, and = 132, Equation (2.63) yields the unpaid balance at the
end of the 132nd month:
n
1 + kr 1 + kr
B132 = n An = $71, 952.87.
1 + kr 1
Hence, the equity is: $350, 000 B132 = $278, 047.13.
b) What are the 1st and 132nd interest payments?
Solution. We have I1 = kr B0 = $970.31 and I132 = kr B131 = $351.15.

2.8.3 Sinking Funds
A sinking fund is an account into which one (individual or company) regularly

deposits money in order to cover an obligation or debt that will come due at a
known future date.
Example 2.17. (Saving for College Tuition) When a child was born in 2011,
her parents decided to invest in her college education. This was motivated by
a forecast that 4 years of in-state tuition at an average public college will be
about $96, 000 when she will attend college. Suppose that the parents want to
accumulate that amount by their childs 17th birthday. They open a sinking
fund into which they make a deposit on each birthday of the child up to the
17th birthday. Assume that the first deposit is for the amount P and thereafter
the parents increase the deposited amount by 4% annually. Suppose that the
bank where they have the sinking fund pays a fixed 5.5% per annum com-
pounded annually. What should the minimum annual deposits be in order for
the amount in the fund to reach at least $96, 000 after her 17th deposit?
2.9 Applications to Stock Valuation 63
Solution. This example applies generalized future value annuity formula in

(2.68), i.e.,

n 1 r n +1 j
Sn = 1 + Pn , n = 1, 2, 3 . . . ,
=0 j=0
k
where rn+1 = 0. Note that r1 does not appear in the formula since no interest is
paid at the end of the first period (because the first deposit is not made at the
start of the first period, but at the end of the first period).
We have n = 17, S17 = $96, 000, r2 = = r17 = 0.055, and k = 1. The product
in the sum above then becomes:

r n +1 j rn r n 1 rn(1)
1+ k = 1+ 1+ 1 + = (1.055) ,
j =0
k k k
where = 0, 1, 2, . . . , n 1.
Let us determine the deposits P1 , . . . , Pn . The deposit P1 = P is made on
the first birthday. On the second birthday, it is increased by 4% to P2 = P1 +
0.04 P1 = (1.04) P . On the 3rd birthday, the deposit is P3 = P2 + 0.04 P2 =
(1.04)2 P . For j = 1, . . . , n, the deposit on the jth birthday is then P j = (1.04) j1 P .
It follows: Pn = (1.04)16 P .
The target amount for the sinking fund can then be expressed as:
16 16
1.055
$96, 000 = P (1.055) (1.04)16 = P (1.04)16 1.04
=0 =0
= P 1.87298 19.1104 = 35.7934 P .
This yields P = $2, 682.06, which is the first deposit. Hence, for j = 1, . . . , 17,
the minimum deposit on the jth birthday must be: P j = (1.04) j1 $2, 682.06,
which has values
P1 = $2, 682.06, P2 = $2, 789.34, ..., P16 = $4, 830.24, P17 = $5, 023.45.

2.9 Applications to Stock Valuation
This section applies the theory of annuities to determining the present val-
ues of preferred and common stocks. The main tool is the dividend discount
model. A stochastic model for the future value of a stock will be taken up in a
later chapter.
2.9.1 The Dividend Discount Model
The dividend discount model (DDM) was pioneered by Williams [18] (1938) and
Gordon [7] (1959). The fundamental hypothesis of the DDM is that, if a stock is held
for n years, then its current value is the present value of the sequence of its expected
future cash dividends through n years plus the present value of the stocks expected
price in n years.
A stock has no maturity date and so is a security in perpetuity. Suppose
that the stock pays a dividend and the (annual) required return rate of the
stock is k.11 Assume that you will hold the stock for n years. Let D0 be the
current cash dividend, i.e., the total cash dividend per share over the previous
year. Suppose that all future cash dividends are expected to grow at a constant
annual rate g, which we assume is less than the required return rate (k > g).
Let D(i) denote the expected cash dividend per share for the interval from the
present time to i years out, where i = 1, . . . , n. Then the expected sequence of
future cash dividends per share for years 1 through n is:
D(1) = (1 + g) D0 , D(2) = (1 + g)2 D0 , ..., D(n) = (1 + g)n D0 .

(n)
The share value S0 of the stock today is the present value of the expected
dividend cash flows and the expected price of the stock n years from now:
(n) D(1) D(2) D(n) Tn

S0 = + + + + , (k > 0), (2.72)
1+k (1 + k ) 2 (1 + k ) n ( 1 + k) n
where Tn is the terminal price, i.e., the expected price of the stock in n years.
Note that (2.72) is a special case of the generalized present value equation (2.69)
(page 58) with k = 1, ri = k, Pi = D(i) for i = 1, . . . , n 1, and Pn = D(n) + Tn .
Now, if you hold the stock in perpetuity (indefinitely) rather than for n
years, then there is no terminal price, and the stocks present share price be-
comes:

(n) 1+g
S0 = lim S0 = D0 , (k > g > 0).
n
=1
1+k
1+ g
Since 1 +k
< 1 due to k > g, the geometric series yields

1 1+g 1+g

1+ g
= 1+k
=1+ 1+k
.
1 1 +k =0 =1
Hence, the present share price of the stock becomes:
11 Recall that the marketplace is assumed to be in equilibrium, which allows for the required return
rate of the stock to be estimated using the CAPM model; see Chapter 4 for an introduction.
2.9 Applications to Stock Valuation 65
( 1 + g ) D0 D(1)
S0 = = , (k > g > 0). (2.73)
kg kg
Equation (2.73) is called the Gordon growth model. This is an example of a grow-
ing perpetuity, i.e., a perpetuity with payments that increase each period.
The Gordon growth model generalizes naturally to allow for k compound-
ings per year through the replacements g g/k and k k/k:
g
1 + k D0 D(1)
S0 = = k g (k > g > 0),
kg
k k k k
g j
where D( j) = 1 + k D0 .
2.9.2 Present Value of Preferred and Common Stocks
A preferred stock grants its holder ownership in a corporation, but no vot-

ing rights, and a claim on assets in the event of bankruptcy that comes before
any claim of the common stock holders. Preferred stocks are considered fixed-
income securities because they promise to pay a fixed cash dividend, which
has priority over any cash dividends paid to common stock holders. Since the
future cash dividends are fixed and expected to be paid indefinitely, the value
of a preferred stock is then obtained from Equation (2.73) using g = 0:
D0
S0 = (k > 0) (2.74)
k
Example 2.18. (Preferred Stocks) Suppose that a preferred stock has a fixed to-
tal annual cash dividend per share of $2.50. Assume an annual required return
rate of 13% for the stock. How much should you pay for the preferred stock?
Solution. We apply Equation (2.74) with D0 = $2.50 and k = 0.13. The current
share price of the preferred stock is: S0 = Dk0 = $2.50
0.13 = $19.23.

Common stocks do not have a promise to pay cash dividends. Nonetheless,

if a common stock currently pays no cash dividends, there is still investor ex-
pectation that earnings are being reinvested in the company to create growth
which will lead to cash dividends in the future. Due to the uncertainty of fu-
ture cash dividends for common stocks, we shall model their valuation under
certain assumptions about the expected cash dividends.
Example 2.19. (Common Stocks) Suppose that the total cash dividend of a
stock last year was $2.75 per share and dividends are expected to increase at
3% per annum. If the annual required return rate is 10%, find the share price
of the stock today.
(1+ g )D0 (1+0.03)$2.75

Solution. By (2.73), the price is: S0 = k g = (0.10.03) = $40.46.

2.10 Applications to Bond Valuation
The US bond market is vastmuch bigger than its stock market. As measured
at the end of 2012 in terms of capitalizations, the US bond market was twice
as big as the US stock market for domestic companies.12 As with other fixed-
income financial investments, the price of a bond is the present value of its
cash flow. We shall explore how to value bonds.
2.10.1 Bond Terminologies
A bond is a contract between an issuer (bond seller) and a lender (bondholder)

legally binding the issuer to repay the lender a specified fixed amount at ma-
turity and a series of interest payments during the life of the bond. In essence,
a bond is an IOU.13 The specific terms for a bonds duration, interest payment,
etc. are described fully in the contract (indenture). The funds raised by bond is-
sues are used for capital expenditures, operations, corporate takeovers, public
projects, etc.
Bonds are usually redeemed on the maturity date. However, some bonds
have the option to be callable,14 i.e., such bonds give the issuer the right, but not
the obligation, to redeem (call) the bond prior to the maturity date. There are
also bonds with the option to be convertible, i.e., the bondholder has the right
to exchange the bond for a different security (e.g., shares of common stock),
and have a prescribed variable interest rate or even deferred interest. To avoid
confusion about which types of bonds are intended, we assume
Unless stated to the contrary, all bonds are without options, i.e.,
they are noncallable, nonconvertible, etc., and have a fixed interest
paid every 6 months.
12 http://www.learnbonds.com/how-big-is-the-bond-market/
13 IOU is an abbreviation for I owe you.
14 Most corporate bonds are callable. Also, the US Treasury has not issued callable bonds since 1985.
2.10 Applications to Bond Valuation 67
Although both bonds and stocks are securities of a company, they are
different in the sense that bondholders are creditors of the company, whereas
stockholders are owners of the company. The cash flows from a companys
bonds are more reliable than those from its stocks since the company has a le-
gal obligation to repay its bondholders. Sometimes even when a company be-
comes insolvent, its bondholders may still get back some compensation, while
compensation is not guaranteed for its common stockholders.
We now list and discuss some basic terminologies and features of bonds:
The issue date of a bond is the date on which the bond issuer receives the
loan from the lender and from which the lender is entitled to receive inter-
est from the issuer.
The maturity value M (also known as the par value, face value, principal) of
a bond is the unit of the amount borrowed at the time it was issued. It is
traditionally in units of $1, 000, but municipal bonds are usually sold in
units of $5,000.
There are two main markets for bonds: primary market, where bonds are
sold for the first time to institutional investors, and secondary market, where
the resale of bonds taking place after their initial offering is open to the
public, though individual investors will need to have a brokerage account
to transact trades. Bonds selling at their maturity value are called par bonds.
In the secondary market, bonds are traded at prices that are typically dif-
ferent from the maturity value. If a bond sells at a market price above
(respectively, below) its maturity value, then it is called a premium bond
(respectively, discount bond).
Remark 2.5. The primary bond market is essentially an institutional mar-

ket. In practice, the US Treasury uses an auction process to sell treasury
bills, notes, and bonds in the primary market, whereas the pricing of newly
issued corporate bonds is negotiated between the corporation (or its repre-
sentative) on the one hand and investment bankers and large institutional
investors on the other.

The maturity date is the date on which the bond issuer must repay the lender
the bonds maturity value. Note that callable bonds have features which
allow for the principal to be repaid before the maturity date.
The term to maturity, or simply maturity, of a bond is the length of the time
interval between the issue date and the maturity date.
Bonds can be classified into three groups: short term, intermediate term and
long term according to maturities of, respectively, 15 years, 512 years, and
greater than 12 years.
A coupon payment C , or simply coupon, is an interest payment of a bond.

Most bonds have a fixed coupon that does not change during the life of the
bond and is paid at regular time intervals, usually semi-annually.
If a bond does not pay a coupon during its life, it is called a zero-coupon
bond. To compensate for no coupon payments, such bonds are issued at a
deep discount from their value at maturity.15 For this reason, they are also
called discount bonds or deep discount bonds. Zero-coupon bonds are similar
to US government savings bonds16 in concept and have significant theoret-
ical value.
The current yield indicates the yield of a security based on its current market
value. The current yield of a bond, denoted by r, is determined by the formula
below:
annual coupon amount

r= . (2.75)
current bond price
The coupon rate or interest rate, denoted by rC , is defined by the current yield
when the bond price is equal to its maturity value. That is:
annual coupon amount

rC = .
maturity value
A bonds yield to maturity (YTM), denoted by rY , is the marketplaces an-

nual required return rate of the bond held to maturity and whose future
coupon payments are reinvested at the same rate. Equivalently, a bonds
YTM equates the present value of the bonds future cash flows to the bonds
current market price:
n 1
C C+M
current bond price = +
(1 + rkY ) (1 + rkY )n
, (2.76)
=1
where n is the number of coupon payments remaining on the bond and

k is the number of coupon payments per annum (typically, k = 2). Equa-
tion (2.76) is obtained by applying Equation (2.69) on page 58 with Pi = C
for i = 1, 2, . . . , n 1, Pn = C + M, and ri = rY for i = 1, 2, . . . , n.
Denote the current bond price and rkY in (2.76) by B(n) and rY , respectively.
Then:
M n
C
B(n) =
(1 + rY )
+
. (2.77)
=1 (1 + rY )
n
15For example, such a bond might be issued at a 50% discount from its maturity value.
16A savings bond offers a fixed rate of interest over a fixed period of time, but cannot be traded after
being purchased.
By (2.58) on page 50, Equation (2.77) is equivalent to

M (1 + rY )n 1 C
B(n) = + (rY > 0). (2.78)
(1 + rY )n (1 + rY )n rY
Finally, the following relationships among yield to maturity (rY ), current

yield (r), and coupon rate (rC ) hold (Exercise 2.35):
1. A bond trades at par iff rY = r = rC (see Proposition 2.1).

2. A bond trades as a discount bond iff rY > r > rC .
3. A bond trades as a premium bond iff rY < r < rC .
Remark 2.6. Generally, evaluating financial investment performance can be a

complicated task as there are different measures to be applied to serve different
purposes. Yield is a measure of an investment income that an investor receives
annually. As a bond investor, if you just want to hold on to your bond until
its maturity, the coupon rate is the only measure that matters. However, if you
need to sell your bond before maturity, you have to adopt the current yield as a
measure. Yield to maturity is a measure that enables you to compare different
bonds by taking the effect of compound interest into consideration under the
assumption that all the coupon payments are reinvested at the same rate and
you hold the bond to maturity.17

2.10.2 Bond Prices Versus Interest Rates and Yield to Maturity
Bond Price with YTM at the Coupon Rate
For a bond being traded after it was originally issued, we expect intuitively
that when the YTM is at the coupon rate, then the market value of the bond
should be its maturity value. The following proposition confirms that intuitive
result and its converse:
Proposition 2.1. Suppose that a bond has n coupon payments remaining. The market
price of the bond equals its maturity value exactly when its coupon rate is the YTM:
rY = r C if and only if B ( n ) = M. (2.79)
17 It is worth noting that comparing different bonds by their percentage change in price is often mis-
leading since the significance is not the same for an identical percentage price change of bonds with
different interest rates. Also, it is important to realize that reinvesting all the coupon payments at the
same rate is rather difficult if not impossible in practice.
Proof. If rC = rY , where rC = rC /2 and rY = rY /2, then C = rC M = rY M and

the bond valuation Equation (2.78) becomes:

M (1 + rY )n 1 M M
B(n) = + = 1 + (1 + rY ) n 1
(1 + rY ) n (1 + rY ) n
(1 + rY ) n
= M.
Conversely, if B(n) = M, then (since C = rC M) Equation (2.78) reduces to:

1 (1 + rY )n 1 rC
1= + .
(1 + rY )n (1 + rY )n rY
rC ((1+rY )n 1)
Multiplying through by (1 + rY )n yields: (1 + rY )n 1 = rY . Hence,
rC = rY .

Bond Prices Move in Opposite Direction to Interest Rates
The interest rate probably has the single largest impact on the prices of all
bonds. The following three examples are related to each other and illustrate
the relationship between bond prices on the one hand and interest rates and
YTM on the other.
Example 2.20. Suppose that a 30-year bond with an annual 3% coupon rate
payable semiannually was issued by the US Treasury on the first trading day of
2013. If the maturity value is $1, 000, what is the semiannual coupon amount?
Solution. Solve for the semiannual coupon amount C from the equation
2C
3% =
$1, 000
to obtain C = $15. Assume, for simplicity, that the bond was sold in the primary
market at its maturity value. Then by Proposition 2.1, the YTM equals 3%.

In the next two examples, the bond in Example 2.20 will be referred to as
the first bond.
Example 2.21. Since the Feds kept interest rates artificially low in 2013, dou-
bling the interest rate in 10 years from 2013 is not an unreasonable speculation.
Suppose that another 30-year bond with an annual 6% coupon rate payable
semiannually will be issued by the Treasury on the first trading day of 2023.
What will be the price of the first bond at the time of the second bond initial
offering?
Solution. For the simplicity of our argument, we assume almost no intraday

bond price fluctuations on the issue date of the second bond. Let
t0 = the issue date of the second bond,

r1 = the (current) yield of the first bond at t0 ,
B1 = the price of the first bond at t0 .
Since no investors will buy a bond with 3% annual yield when they have the
choice to purchase a bond of the same type with 6% annual yield, we have
r1 = 6%. In other words, the current yield of the first bond will be forced to
approach 6% on the issue date of the second bond under the law of supply
and demand. To speculate on the price of the first bond, we apply (2.75) and
solve for B1 from the equation, r1 = 6% = 2B$151
, to obtain B1 = $500. Observe
that when the interest rate rises from 3% to 6%, the first bonds price will fall from
$1, 000 to $500.

Example 2.22. Suppose that you will purchase the first bond on the first trad-
ing day of 2023 at the price $500 and hold it to the maturity date of the first
trading day of 2043. What will be the yield to maturity?
Solution. We need to solve the bond Equation (2.78) for rY , which in our set-
ting is18
((1 + rkY )n 1) C M
B1 = rY rY n + .
k (1 + k ) (1 + rkY )n
Using B1 = $500, C = $15, M = $1, 000, k = 2 (semiannual compounding), and

n = 40 (number of coupon payments remaining), we obtain rY = 8.084%. In-
deed, when the price of a bond drops from $1, 000 to $500, the yield to maturity rises
from 3% to 8.084%. Note that we assumed the bond was sold in the primary
market at its maturity value.

We now establish, in general, the observation at the end of the solution of

Example 2.22. Take the first and second derivatives of the bonds present value
(2.77):
dB(n) M n

= n
(1 + rY ) +
C +1
< 0
=1 (1 + rY )
drY n 1
and
18 As before, there is no general analytical solution rY for every n. In most applications, we can only
estimate rY numerically using a software.
Bond Price
1600
1400
1200
1000
800
600
YTM
0.00 0.02 0.04 0.06 0.08 0.10
Fig. 2.2 The price of a bond is a strictly decreasing, concave-up function of the bonds YTM. The
graph illustrates this for a bond with $1, 000 maturity value and 6% coupon rate. Note that when the
YTM is 6%, the bonds price is its maturity value.
d2 B ( n ) M n
( + 1)
= n ( n + 1 )
(1 + rY ) +
+ C +2
> 0.
=1 (1 + rY )
2
drY n 2
In other words, the bonds price is not only strictly decreasing as the yield
increases, but has a convex graph, i.e., the graph is everywhere concave up
(increasing slope). Figure 2.2 depicts this property for a bond with $1, 000 ma-
turity value and 6% annual coupon rate.
2.11 Exercises
2.11.1 Conceptual Exercises
2.1. A physicist summed up the growth rate of an initial sum of money held
over a fixed time span as follows: If simple interest is applied during the
time span, then the initial sum will grow with uniform (constant) velocity as
the interest rate increases. If periodic compound interest is applied, then the
growth of the initial sum will accelerate as interest increases. Do you agree
with this interpretation? Justify your answer.
2.2. Theorem 2.1 on page 27 yields that k-periodic compounding of a principal

F0 at r per annum over a time span of years consisting of x interest periods
gives a future value,
r x
Fx = 1 + F0 ,
k
2.11 Exercises 73
where x is a nonnegative real number and 0 kr < 1 for k = 1, 2, . . . . Does Fx

increase or decrease as k increases indefinitely? Justify your answer.
2.3. Suppose you purchase a lottery ticket for $2. What is your return rate if
you lose? What if you win $200 million? Express your answer as a percentage.
2.4. Consider an investment that promises a fixed sequence of future cash div-
idends. Briefly explain why an increase in the required return rate on the in-
vestment would decrease the current value of the investment.
2.5. Explain what the following is stating financially about the start-up: A
start-ups NPV at 30% is $35, 000.
2.6. A friend borrows $1, 000 from a lender that gives him the loan as a simple
ordinary annuity at a fixed interest rate over 2 years with a payment of $100
per month. If your friend carries the loan to its full term, then he will have
to pay more than the amount of the loan in just interest. True or false? Justify
your answer.
2.7. A loan with a fixed payment of $1, 000 per month for 5 years has the stipu-
lation that you will have to pay all the interest due on the loan even if you pay
the loan off early. If immediately after you receive the loan, you want to pay it
off, how much do you have to pay the lender?
2.8. How would you modify the interpretation of the noncallable bond pricing
formula (2.78) on page 69 to obtain the current price of a callable bond, i.e.,
a bond where the issuer has the right, but not the obligation, to redeem (in
practice, cancel) the bond before maturity? Use a single call date, i.e., a date
when the issuer can redeem the bond before maturity. Compare the price of
a callable bond with a noncallable one. Corporations issue callable bonds be-
cause if interest rates go down, they can call their bonds and refinance their
debt at a lower interest rate.
2.9. How would you modify the interpretation of the noncallable bond pricing
formula (2.78) on page 69 to obtain the current price of a puttable bond, i.e.,
a bond where the investor has the right, but not the obligation, to redeem the
bond before maturity? Use a single put date, i.e., a date on which the investor
can redeem the bond before maturity. Compare the price of a puttable bond
with a noncallable one. Investors buy puttable bonds because if interest rates
increase, they can sell back their original bonds at the put value and invest the
proceeds in a higher interest rate bond.
2.11.2 Application Exercises
2.10. Consider a principal F0 that is held for nexact days during a non-leap
year at the simple interest rate r. By what percent is the simple interest amount
using Bankers Rule greater than the simple interest amount employing exact
time and exact interest?
2.11. (Selling or Buying a Loan) On November 12, 2007, a borrower closes on

a loan for $176, 000 at 6.25% per annum compounded daily. Repayment of the
loans maturity value (principal plus interest) is due in full on April 15, 2008.
Suppose that the fine print of the original loan stipulated that the lender can
sell the loan on the condition that the interest rate and maturity date remain
the same. The lender sells the loan to another lender on January 5, 2008. The
new lender agrees to purchase the debt for the present value of the maturity
value at 10% per annum compounded daily. Assume that interest compounds
daily and the borrower does not default on the loan. Use Bankers Rule when
solving the following:
a) What is the maturity value of the loan?
b) What will the first lender receive for selling the loan? Is any profit made by
the first lender?
c) What profit will the second lender make on the loans maturity date if the
conditions of the original loan are unchanged?
d) Though the original interest rate and maturity date are unchanged, the sec-
ond lender is not prevented from reissuing the loan with a new start date
set as the loans purchase date and with the new loans principal set as the
value of the loan on the purchase date. Does the second lender make more
profit by resetting the loan in this way? Explain.
2.12. For an interest rate of 4% per year, compare the future value 2 years from
now to which $10,000 increases under daily compounding versus continu-
ous compounding. Assume 365 days per year and express your answer as a
fractional-difference percentage of the daily compounding case.
2.13. Suppose that at the start of college, you have $1,000 to invest and would
like for it to grow to $1,250 at the end of your senior year through monthly
compounding. Determine the general formula for the interest rate required for
the growth and then compute the interest rate.
2.14. Assume that college tuition is currently 30 times its cost 15 years ago.
Assuming annual compounding, what is the interest rate r that gives the rate
of increase in tuition?
2.15. How much should you have today in an account with monthly com-
pounding and annual interest rate of 4% to receive $1,000 per month forever?
2.16. (Equity in a House) A couple purchased a house 7 years ago for $375, 000.
The house was financed by paying 20% down and signing a 30-year mort-
gage at 6.5% on the unpaid balance. The net market value of the house is now
$400, 000. Assume that the couple wishes to sell the house.
2.11 Exercises 75
a) How much equity (to the nearest dollar) does the family have in the house
now, after making 84 monthly payments?
b) Find the first interest payment I1 and the 84th interest payment I84 .
2.17. (Social Security Benefits) We present a simplified problem to illustrate
Social Security benefits. A college graduate begins work at age 22. She has an
annual income of $70, 000 until retirement (a simplification), pays 12.4% of this
income into Social Security each year, and retires at age 65 with Social Security
benefits of $20, 000 annually. How long must she live before the present value
of these benefits equals the present value of her annual contributions? In other
words, how long must she live after retirement to get back the full value of
her contributions to Social Security? Will she get the entire value? Assume a
discount rate of 4% per year, no change in her salary, and that all payments
and benefits occur at the end of each year.
2.18. (Workers Compensation) The usual legal settlement for an industrial
accident is the present value of the employees lifetime earnings. If you expect
to work for 10 more years, make $70, 000 a year in the next 2 years, and get
a raise of $5, 000 every 2 years, what would be your settlement? Assume an
annual discount rate of 4% in the first 5 years and 6% in the second 5 years,
and that your paycheck is received at the end of each year.
2.19. (Bonds) Suppose that you bought a 30-year bond with 4% annual coupon
rate. You wish to sell that bond at a later date when the remaining life of the
bond is 2.5 years and the current YTM of your bond has declined to 2%.
a) What is the fair value, as determined by the present value method, of the
bond at the time of your sale?
b) How much would you earn if you purchased the bond for $1, 000, sold it at
the fair value, and did not reinvest the coupon payments?
2.20. (Bonds) Bonds are generally quoted as a percentage of their face value.
A bond selling at 99.2% of its face value is quoted as 99.2. The following in-
formation for a treasury bond was provided by the WSJ market data center on
December 4, 2013:
Maturity Coupon Current price Previous price Change Yield

11/30/20 2.000 99.20 99.00 0.203 2.123
The coupon column refers to the annual coupon rate. Verify that the last col-
umn indicates YTM.
Purchasing a House
The remaining Application Exercises deal with purchasing a house. Assume

that you are currently renting an apartment for $1,040 per month and you have
been considering buying a house. You have saved $10,000 toward a down pay-
ment for the house.
A salesperson informs you that he has a new house for sale, where the house
and land were independently appraised at $200, 000, but are being sold by the
builder at a discount price of $185, 000. The builder wants to get rid of the prop-
erty quickly because the house is the last one to be sold in the development and
the builder is moving on to construction of a new development.
The salesperson connects you with his in-house lender, to whom you give
details about your income and grant permission to review your credit and
eligibility for a loan. You inform her that you are prepared to make a down
payment of $10,000 toward the house if necessary. She gets back to you with
good news that, if you put $8, 100 toward the house, then they can give you
a 30-year loan for the balance of $176, 900 at 6.25% per annum (compounded
monthly). Note that lenders require the house to appraise at or above the pur-
chase price; otherwise, they may reject the loan or require more down pay-
ment. The lender computes the monthly mortgage payment at $1,089.20. She
informs you that the remaining $1,900 of your $10,000 can be used toward costs
associated with the final evaluation of the physical property and the closing of
the purchase (property inspector fee, termite inspector fee, official survey, at-
torney fees, etc.). The builder agrees to pay for costs beyond your $1,900 and
make necessary repairs you identify during the period you have to inspect the
property (the due diligence period).
Hearing the news about your qualification for the loan, the salesperson asks
you how much rent you are now paying. When you inform him that you pay
$1,040 per month, he quickly points out that it would be a mere extra $50 per
month for you to meet the mortgage payments. He emphasizes that it is better
to own than to rent, especially if the mortgage is just a bit more than your
current rent.
You are thrilled! After the excitement subsides, however, you decide to run
the numbers yourself to make sure you get a clear understanding of what
you are getting into financially.19 The problems in this project help guide you
through some of this analysis.
2.21. Show that the monthly loan payment on the unpaid principal balance of
$176, 900 is $1, 089.20.
2.22. In addition to closing fees paid to settle the loan, there are expenses be-
yond the monthly mortgage payments.
First, since your deposit was less than 20% of the purchase price, you are
required to take out a private mortgage insurance (PMI) to protect the lender
19 Mortgages on a house are generally modeled as simple ordinary annuities by lenders.

2.11 Exercises 77
if you default on the loan. The PMI typically lasts until the unpaid principal
balance of the mortgage is paid down to 80% of the original value of the house,
where the houses original value is the lesser of the purchase price and the
official appraised value of the house used in closing the sale. Note that the
bank may also require your payment history to be in good standing (e.g., no
late payments in the past year or two) before removing PMI. Of course, if the
value of the house increases nontrivially, you may be able to remove the PMI
earlier. Suppose that the PMI is $141.52 per month.
Second, along with PMI, you have to pay for hazard insurance to cover un-
planned damages to the house due to fire, smoke, wind, etc. Assume that the
hazard insurance is $36.50 per month.
Third, you have to pay property taxes to the tax district (e.g., county and
city) where the house is located. The property (i.e., house and land) will be
valued within your tax district, which is a valuation that is separate from the
appraisal done when purchasing the house. The resulting tax districts valua-
tion is the taxable value of the house and is the amount to which the property
tax rate will be applied. Suppose that the annual property tax rate is 1.3% and
the taxable value of the property is $189, 986. For this project, the taxable prop-
erty value is less than the appraised value (i.e., $200, 000) used for the purchase.
Sometimes, however, the taxable value can be higher which was not uncom-
mon in the aftermath of the 2008 mortgage crisis.
The PMI, hazard insurance, and property tax payments are in addition to
the monthly loan payment, and all together they form a single payment you
make to the lender. The lender or a company hired by the lender manages
these payments by taking out the portion for the loan payment (principal plus
interest) and depositing the rest into an escrow account, which is used to pay
the annual insurance premiums and property taxes on behalf of the borrower.
Finally, assume that the property is in a housing development that comes
with a mandatory Homeowners Association (HOA) fee. The HOA fee is used
to maintain the grounds, roads, etc. in the development. If you do not pay the
fee, the HOA can foreclose on your property. Assume an HOA fee of $100 per
month.
a) What is the estimated total monthly PITI, i.e., the minimum monthly pay-
ment covering the principal, interest, taxes, and (hazard) insurance?
b) Identify two other mandatory house expenses that are outside of the PITI
payment and other basic house costs like utilities and repairs. Do exclude
costs like groceries, tuition, medical expenses, etc., which are more associ-
ated with running a home. What is the minimum monthly cost of the house
during the first year if you now include these two mandatory house ex-
penses and PITI? Which of these housing costs will likely increase in the
future?
c) What is your opinion about the salespersons pitch about the cost of renting
versus buying a house?
2.23. Fill out the amortization schedule below, which is for the first 5 months
of the loan.
Payment # Payment (P ) Principal (P ) Interest (I ) Bal. (B )
1 1, 089.20 167.85 921.35 176, 732.15
2 1, 089.20
3 1, 089.20
4 1, 089.20
5 1, 089.20
2.24. Are there discrepancies in the above amortization table? If so, explain
how to remove them mathematically.
For the remaining problems, note that only the payments toward principal
and interest (PI) are relevant to the loans balance. Costs associated with prop-
erty taxes, hazard insurance, PMI, HOA, etc. are separate expenses and do not
impact the balance of the loan. Such costs are typically not included in the
loans cost.
2.25. Using a software, compute the numbered payment at which the unpaid
balance on the loan will first dip below 80% of the original value of the house.
Roughly how many years and months does it take to reach that balance? If the
value of the house has not decreased below its original value at that point in
time, you would stop paying PMI henceforth.
2.26. Determine the total amount you would pay into the mortgage, excluding
escrow payments, if you make only the minimum payment over the full 30
years. What is the total cost of the mortgage? Is it more than the mortgage?
2.27. Estimate the number of years and months it would take to pay off the
mortgage if you double your monthly payments.
2.28. Estimate the total you would pay into the mortgage if you double your
monthly payments. What is the total cost of the mortgage for doubled pay-
ments? Is it more than the mortgage?
2.11.3 Theoretical Exercises
2.29. Suppose that an initial capital F0 grows to an amount F ( ) over a time

span . A mathematician modeling the growth observes that for all time spans
2.11 Exercises 79
x and y, the accumulated amount F ( x ) is a differentiable function satisfying

the following:
dF
F ( x + h ) = F ( x ) + F ( h ) F0 , F ( 0 ) = F0 , ( 0 ) = r F0 ,
dx
where r 0. Determine the type of growth model, i.e., find F ( x ).
2.30. Derive Equation (2.18) on page 26: G ( x ) = G ( x ) G (0).
2.31. (Capital After Spending, Inflation, and Interest) Consider the following
setup:
- Begin with an initial capital C (0) in an interest-bearing account and let C (n)
be the remaining capital at the end of the nth year.
- Assume an interest rate r is applied at the end of each year to the capital
remaining on that date.
- At the end of the first year, assume that an amount S was spent from C (0) on
goods and services, and money will be spent on similar goods and services
in each of the subsequent years.
- Suppose that the amount spent at the end of any specific year is the total
amount spent by the end of the first year increased in subsequent years at
the annual inflation rate i compounding annually until the end of the spec-
ified year. Assume that r > i since investors are not interested in a market
interest rate that is below the inflation rate.
a) Show that the total capital at the end of the (n + 1)st year can be expressed
recursively as follows in terms of the capital at the end of the previous year,
taking into account spending, inflation, and interest growth:
C ( n + 1) = (1 + r ) [ C ( n ) (1 + i) n S ] . (2.80)
b) Use induction to show that

1+r 1+r
C (n) = (1 + r )n C0 S + (1 + i) n S.
ri ri
2.32. Suppose that after this year, your grandmother will receive regular pay-
ments from a retirement fund, but she has to choose between two options for
how to receive the payments during n + 1 years. She does not plan to spend
any of the money until after the n + 1 years. Assume that she will save all the
disbursements in an account that accrues the payments as a simple ordinary
annuity with k-periodic compounding at interest rate r (e.g., each payment
date coincides with an interest date).
The payment start date will differ for the two plans, but both payment op-
tions will have the last payment at the start of the last interest period during
the (n + 1)st year. Your job is to help her choose between the two options.
a) (A General Future Value Formula) The current problem determines a gen-

eral formula that incorporates the future value of the payments into your
grandmothers account. Suppose that regular payments of P into an interest-
bearing account form a simple ordinary annuity with k-periodic compound-
ing at interest rate r. Assume that the account receives the first payment at
the end of the first interest period of a certain year and the last payment at
the end of the Nth interest period going forward, with no payment at the
end of the ( N + 1)st interest period. Show that the amount accrued in the
account at the end of the ( N + 1)st period is:

(1 + r/k) N +1 (1 + r/k)
FV P, (2.81)
r/k
where N is the total number of payments into the account.
b) We now explore the future values associated with the following two plans
for receiving payment.
i. Plan A. Assume that Plan A begins officially at the start of next year with
payments of A starting at the end of the first interest period of next year.
Show that the total amount she would accrue at the end of the (n + 1)st
year is:
(1 + r/k)(n+1)k (1 + r/k)
FV A A.
r/k
ii. Plan B. Under Plan B, your grandmother receives payments of B with the
choice of officially starting at the beginning of the (q + 1)st year after Plan
A starts and the first payment disbursing at the end of the first interest
period of the official starting year. Show that the total amount she would
accrue by the end of the (n + 1)st year is

(1 + r/k)[(n+1)q]k (1 + r/k)
FVB B,
r/k
where q = 1, 2, . . . . Note that for q = 0, the two options coincide.

c) (Choosing Between Plans A and B) Naturally, since Plan B starts out later
than Plan A and both have the same last-payment date, the payment amount
of Plan B has to be higher than that of Plan A, i.e., B > A. Suppose that the
accounts interest rate exceeds a threshold as follows:

r > k ( B/A)1/q 1 .
i. Show that there is no n such that the amounts accrued under both options
are equal by the end of the (n + 1)st year.
ii. Show that Plan A is superior to Plan B, i.e., prove FV A > FVB .
References 81
2.33. (Relating Present and Future Values of a Generalized Annuity) Using

n 1 r n +1 j
Sn = 1 + Pn ,
=0 j=0
k
verify the formula

S
An = n r ,
nj=1 1 + kj
where n = 1, 2, . . . , r j > 0 for j = 1, . . . , n, and rn+1 = 0.
2.34. (Bonds) Given a coupon bond described by Equation (2.76) on page 68,
find the future value at maturity of the bonds cash flow.
2.35. (Bonds) Show that for a coupon bond, its yield to maturity (rY ), current
yield (r), and coupon rate (rC ) have the following relationships:
a) A bond trades at a discount if and only if rY > r > rC .

b) A bond trades at a premium if and only if rY < r < rC .
References
[1] Bodie, Z., Kane, A., Marcus, A.: Investments, 9th edn. McGraw-Hill Irwin,
New York (2011)
[2] Brealey, R., Myers, S., and Allen, F.: Principles of Corporate Finance.
McGraw-Hill Irwin, New York (2011)
[3] Brown, S., Kritzman, M.: Quantitative Methods for Financial Analysis.
Dow Jones-Irwin, Homewood (1990)
[4] Chaplinsky, S., Doherty, P., Schill, M.: Methods of evaluating mergers and
acquisitions. Note Number UVA-F-1274. University of Virginia Darden
Business Publishing (2000)
[5] Choudhry, M.: Fixed-Income Securities and Derivatives Handbook.
Bloomberg Press, New York (2005)
[6] Davis, M.: The Math of Money. Copernicus Books, New York (2001)
[7] Gordon, M.J.: Dividends, earnings and stock prices. Rev. Econ. Stat. 41, 99
(1959)
[8] Guthrie, G, Lemon, L.: Mathematics of Interest Rates and Finance. Pren-
tice Hall, Upper Saddle River (2004)
[9] Hull, J.: Options, Futures, and Other Derivatives, 7th edn. Pearson Pren-
[10] Kellison, S.: The Theory of Interest, 2nd edn. Irwin McGraw-Hill, Boston
(1991)
[11] Koller, T., Goedhart, M., Wessels, D.: Valuation: Measuring and Managing
the Value of Companies. Wiley, Hoboken (2010)
[12] L.E.K. Consulting, LLC: Discounted Cash Flow Valuation Primer. L.E.K.
Consulting, Chicago (2003)
[13] Lovelock, D., Mendel, M., Wright, A.: An Introduction to the Mathematics
of Money. Springer, New York (2007)
[14] Meserve, B.: Fundamental Concepts of Algebra. Dover, New York (1981)
[15] Muksian, R.: Mathematics of Interest Rates, Insurance, Social Security, and
Pensions. Prentice Hall, Upper Saddle River (2003)
[16] Reilly, F., Brown, K.: Investment Analysis and Portfolio Management.
South-Western Cengage Learning, Mason (2009)
[17] Wang, X.: A simple proof of Descartess Rule of Signs. Am. Math. Mon.
111, 525 (2004)
[18] Williams, J.B.: The Theory of Investment Value. Harvard University Press,
Cambridge (1938). Reprinted in 1997 by Fraser Publishing
Chapter 3
Markowitz Portfolio Theory
In 1952 Harry Markowitz [17] pioneered a Nobel Prize-winning1 mathematical

model that showed how to distribute an initial capital across a collection of
risky securities to create an efficient portfolio, namely, one with the least risk
given an expected return and largest expected return given a level of portfolio
risk.
We shall first present the assumptions of the Markowitz portfolio model
and the models formulas for the expected return rate and risk of a port-
folio (Section 3.1). These formulas are then applied in detail over a single
period to two-security portfolios (Section 3.2) and extended to N securities
to show how to obtain efficient portfolios (Sections 3.33.4). This is followed
by an application of the theory to determining the global minimum-variance
portfolio and the diversified portfolio, resulting in the Mutual Fund Theorem
(Section 3.5). In Section 3.6 we introduce an investors utility function. Given
an infinite collection of efficient portfolios, we illustrate how optimizing an in-
vestors expected utility function leads to the selection of an efficient portfolio
that maximizes the investors satisfaction relative to the investors risk toler-
ance. The chapter then concludes with portfolio diversification and the issue
of systematic and unsystematic risk (Section 3.7). We shall see that the major
contributor to the risk of a portfolio with a sufficiently large number of sec-
urities from across the marketplace is not the risks of the securities, but the
movements of the securities returns relative to each other (their covariances).
3.1 Markowitz Portfolio Model: The Setup
In this section, we overview the key concepts, assumptions, and quantities of

the Markowitz portfolio model and introduce some needed notation.
1 Harry Markowitz, Merton Miller, and William F. Sharpe shared the 1990 Nobel Prize in Economic
Sciences. Markowitz won for his work on portfolio selection (see Press Release at Novelprize.org).

84 3 Markowitz Portfolio Theory
We begin with the market assumptions assumed throughout the book:

Unless stated to the contrary, assume a US market environment
and the following ideal market conditions hold:
Investors: all investors are rational, i.e., they make financial decisions that
maximize their expected satisfaction with possible wealth gains in the face
of the risks these possible gains require.
Equilibrium: supply equals demand.
No arbitrage: no-arbitrage opportunities exist, which means intuitively that
there is no opportunity to make a costless, riskless profit. For an arbi-
trage, none of your funds is required, loans would be settled with interest,
and a profit is still guaranteed. However, for mathematical modeling pur-
poses, a broader and more precise definition of arbitrage is used later; see
Definition 7.1 (page 334).
Access to information: rapid availability of accurate information on securities
exists.
Efficiency: a securitys price adjusts quickly to new information, so its
current price reflects all known information impacting the security, which
includes information about the past and expected future behavior of the
security.
Liquidity: any number of units of a security can be bought and sold quickly.
No transaction costs: transaction costs are assumed to be negligible com-

pared to the value of the trades and so are ignored.
No taxes: transactions occur without taxation.
Borrowing/lending: borrowing and lending are at the risk-free rate r.2
We now turn to the setup of the Markowitz model. First, a portfolio is a col-
lection of different securities, each of whose value is a specific percentage of
the entire portfolios value.
Throughout the chapter, all portfolios consist only of risky securi-
ties, i.e., securities whose future return rates cannot be predicted
with certainty, and each portfolio has at least two securities.
Chapter 4 will explore the consequences of adding a risk-free asset to a portfo-
lio with a sufficiently large number of risky securities. Furthermore, given the
vast array of different risky securities, it may be helpful to the reader to keep
stocks or ETFs in mind specifically.
2No lending or borrowing of money will be done in the current chapter, but it will be part of the
modeling in Chapter 4, which generalizes the Markowitz model.
3.1 Markowitz Portfolio Model: The Setup 85
3.1.1 Security Return Rates
Let N 2 be an integer and consider a portfolio of N securities with unit prices

S1 (t), . . . , S N (t) at time t. Assume that the liability of each security is limited to
its value, i.e., for i = 1, . . . , N, we have Si (t) 0 with Si (t) = 0 corresponding
to bankruptcy. We shall exclude the latter situation:
Unless stated to the contrary, assume that each security has a pos-
itive price at every moment of time: Si (t) > 0 for i = 1, . . . , N.
We shall fix an investment time interval [t0 , t f ], where t0 0 is the present
time, t f is a future time, and = t f t0 is the length of the investment interval.
Intuitively, the return rate of a security (or portfolio) for the given investment
interval measures what comes back to you beyond your initial investment. By
(2.2) on page 19, the return rate from t0 to t f of the ith risky security is the
following random percentage:
S i ( t f ) S i ( t0 ) D i ( t0 , t f )
R i ( t0 , t f ) = + , (3.1)
S i ( t0 ) S i ( t0 )

capital-gain return dividend yield
where i = 1, . . . , N. Here Di (t0 , t f ) 0 is the per-unit total cash dividend from

the ith security during the time interval [t0 , t f ). Furthermore, the return rate is
unchanged when the total number of units of the ith security is included; see
Section 2.2.3 for more. The amount that comes back to you for the investment
interval is then a percentage Ri (t0 , t f ) of the initial investment Si (t0 ):
R i ( t0 , t f ) S i ( t0 ) , (i = 1, . . . , N ).
We shall see that the return rates of the securities are the core quantities from
which all the other Markowitz portfolio inputs are calculated.
Notation. In most of this chapter, we shall consider security return rates over
a fixed investment time interval [t0 , t f ] and so the following simpler notation
will be used:
R i = R i ( t0 , t f ) , i = 1, . . . , N.
In the formula (3.1) for Ri , the futures price Si (t f ) is a discrete random vari-
able in models like binomial trees (Chapter 5), while in some continuous-time
models, it is lognormal (Chapter 6). Additionally, the future dividend Di (t0 , t f )
is also random since it is typically unknown. However, in most applications in
the book, we shall model the dividend as a known percentage of the securitys
unit price at t0 :
D i ( t0 , t f ) = q i S i ( t0 ) , i = 1, . . . , N,
where qi is the (assumed known) annual dividend yield rate of the ith security.
A basic assumption about the securities returns is that the covariance matrix V of
the return rates R1 , R2 , . . . , R N is invertible: the financial implication is that there
is no redundant security in the portfolio, i.e., no security with a return rate that is
a linear combination of the others. To see this, suppose for illustration that
R1 = a 2 R2 + a N R N .
Write the covariance matrix as

11 12 1N
21 22 2N

V = .. .. .. .. ,
. . . .
N1 N2 N N
where ij = Cov( Ri , R j ) = ji , and consider the first column of V. The top entry
in the first column expands to
11 = Cov( R1 , R1 ) = Cov( R1 , a2 R2 + + a N R N )
= a2 Cov( R1 , R2 ) + + a N Cov( R1 , R N )
= a2 12 + + a N 1N .
Similarly, for the other entries 21 , . . . , N1 , we obtain

11 a2 12 + + a N 1N
21 a2 22 + + a N 2N

.. = .. .
. .
N1 a2 N2 + + a N N N
In other words, the first column of V is a linear combination of the other

columns:
c1 = a 2 c 2 + + a N c N ,
where ci is the ith column of V . This contradicts V being invertible.
Now, a covariance matrix V is always semi-positive definite since for all
x = ( x1 , . . . , x N ) in R N , we have
xT V x = Var( x1 R1 + + x N R N ) 0.
Additionally, the eigenvalues of a semi-positive definite matrix are nonnega-

tive. Since the determinant of V is the product of its eigenvalues and because
V is invertible (so det V > 0), the eigenvalues of V are all positive. Hence, the
covariance matrix V of security return rates is positive definite.3
3 The theoretical importance of the positive definite property will be seen in Section 3.3.
3.1.2 What About Multivariate Normality of Security Return Rates?
An assumption about the Markowitz model that is often encountered in text-

books (in part, maybe because students are very familiar with the normal dis-
tribution) is the multivariate normal condition on the securities return rates:
Multivariate normality condition: The N-tuple of return
rates ( R1 , . . . , R N ) of the risky securities in a portfolio has
a joint multivariate normal density with a positive definite
covariance matrix. In other words, any linear combination
iN=1 ai Ri of the return rates is a normal random variable;
hence, each Ri is normal.
The validity of this assumption has been widely debated. One immediate crit-
icism is as follows: since Si (t f ) + Di (t0 , t f ) > 0, the return rate satisfies
S i ( t f ) + D i ( t0 , t f )
Ri = 1 > 1.
S i ( t0 )
But, if Ri is normal, then it has a nonzero probability of satisfying Ri 1,
which is inconsistent with a positive security price and nonnegative dividend.
However, the issue depends on the length of the time interval for which
one is considering the return rate. It is typically assumed that the normality of
security and portfolio return rates holds for a sufficiently short time span .
See Bodie, Kane, and Marcus [1, pp. 139153] for an elementary introduction
as well as the research paper [15] by Levy and Duchin.
The multivariate normality condition is actually not necessary for Markowitz
mean-variance analysissee Markowitz [18] and Markowitz and Blay [19]and
so will not be enforced. Readers are also referred to the insightful reviews by
Goldberg [6] and Levy [14], which provide excellent guides to the book [19].
3.1.3 Investors and the Efficient Frontier
Assume the following about investors:
Investors assess a portfolio only through its expected return rate and risk and
agree on the joint distribution of the securities return rates from t0 to t f .
Investors are risk averse, i.e., for a portfolio with a given risk, investors
demand the largest possible expected return rate and for a portfolio with
a given expected return rate, they demand the least possible risk.
By construction, risk-averse investors are interested only in portfolios that

are efficient. An efficient portfolio is a portfolio having simultaneously the
smallest possible risk for its given level of expected return rate and the largest
possible expected return rate for its given level of risk. The collection of all
efficient portfolios is called the efficient frontier. The efficient frontier contains
infinitely many efficient portfolios and each represents a different risk-return
tradeoff. We shall show how the Markowitz model is applied to determine the
efficient frontier of a two-security portfolio (Section 3.2.2) and then the general
case of a portfolio with N securities. We shall show how the Markowitz model
yields that the more one spreads a portfolios capital across different risky sec-
urities, the more the portfolios risk is reduced (Sections 3.2.3 and 3.7), lending
theoretical support to dont put all your eggs in one basket.
Finally, the place where an investor positions her portfolio on the efficient
frontier will have to do with her utility function, which indicates her satisfaction
with the risk-return tradeoff. In other words, she will rank potential returns in
the face of the potential risks it takes to realize those returns in such a way as to
maximize her expected satisfaction or happiness (utility). In other words, we
assume that an investor will seek an optimal portfolio, i.e., an efficient portfolio
that maximizes her expected utility functionsee Section 3.6 for more.
3.1.4 The One-Period Assumption, Weights, and Short Selling
We now turn to the setup of a portfolio in the Markowitz model.
One-Period Assumption
Assume that today, denoted by t0 , we use an initial investment capital
V P ( t0 ) > 0
to create a portfolio by distributing the entire capital among N different prese-

lected risky securities. We shall show how the Markowitz model addresses the
issue of how to allocate the initial capital among different securities. However,
the details of how to screen the marketplace to preselect the N different risky
securities for the portfolio are beyond the scope of this book. For the practi-
calities of portfolio management, readers are referred to, for example, Grinold
and Kahn [9] and Reiley and Brown [22]. Furthermore, it is possible that not all
the preselected securities will be used. Some securities can have a zero percent
assigned to them, which we refer to as having no position or being flat in the
security.
Now suppose that we will hold the portfolio until a future end date t f . Dur-
ing the investment period [t0 , t f ], we abide by
The one-period assumption: from the current time t0 to the final time t f ,
we make no change to the total number of securities, the type of securities,
or the number of units of any security in the portfolio.
Weights and Number of Units of the Security
Assume that the percentage of the initial capital V P (t0 ) invested in the ith
security is wi , which is called the weight of the ith security. Since the entire
initial capital V P (t0 ) is distributed across the N securities, the weights of all
the securities add up to 100%:
w1 + + w N = 1. (3.2)
The column vector

w = [ w1 w2 . . . w N ] T
is called the portfolio weight vector at t0 . Note that w = 0. The amount V P (t0 )
invested in the ith security is then wi V P (t0 ) and the total investment spreads
among the securities are follows:
V P ( t0 ) = w 1 V P ( t0 ) + + w N V P ( t0 ) .
Because the amount of the initial capital V P (t0 ) to be used to purchase units
of the ith security at t0 is wi V P (t0 ), the number of units the money buys is
w i V P ( t0 )
ni = , i = 1, . . . , N, (3.3)
S i ( t0 )
where Si (t0 ) is the price of the ith security at t0 . Equivalently, the cost of ni
units of the ith security is ni Si (t0 ). Note that a non-integer number of units of a
security is allowed.
The initial value of the portfolio can then be expressed as the sum of the
costs of the various securities, where a securitys cost is a product of cost per
unit and the number of units:
V P ( t 0 ) = n 1 S1 ( t 0 ) + + n N S N ( t 0 ) . (3.4)
Since the portfolio is constructed at t0 by obtaining the following specific num-

bers of units of the N securities,
n ( t0 ) = ( n 1 , . . . , n N ) , (3.5)
this vector (3.5) is called the trading strategy of the portfolio at t0 . The value of
the trading strategy at t0 is defined to be the initial capital V P (t0 ) as expressed
in (3.4). By the one-period assumption, the number of units of each security is
held fixed to the end date t f , i.e., the trading strategy is held constant during
the period.
Example 3.1. Assume that today an initial capital of $5, 000 is used to create a
three-security portfolio with 20% of the money in stock 1, 30% in stock 2, and
50% in stock 3. Suppose that the current share prices of the stocks are, respec-
tively, $40, $70, and $10. Then the current trading strategy of the portfolio is to
buy the following numbers of shares of stocks 1, 2, and 3, respectively:
0.2 $5, 000 0.3 $5, 000 0.5 $5, 000

n1 = = 25, n2 = = 21.43, n3 = = 250.
$40 $70 $10

Short Selling
At this stage, you may be implicitly assuming that the number of units and
weight of a security are nonnegative real numbers. However, we apply no such
restriction. This is because we assume that each security in the portfolio is obtained
by either buying, short selling, or taking no trading position (being flat). When you
buy ni units of a security, you are adding redundant securities to your portfolio
and so we represent this position by ni > 0. When you sell ni units of a security,
your portfolio has ni fewer units of the security and we express this position by
ni < 0. When you do not hold a security, we represent that position by ni = 0.
In general, to close or liquidate a buy (respectively, sell) position in ni units of
a security, you must sell (respectively, buy) ni units of the security. The weight
wi corresponding to a position of ni units in a security will have the same sign
as ni .
Short selling securities is a selling of securities that varies in its details dep-
ending on the type of security. The most common example is short selling a
stock, where you sell a certain number of shares of a stock borrowed from a
broker. You will almost definitely need to have a margin (a certain amount of
required funds) in your account in case you are unable to return the borrowed
shares. You close the stock short sale by buying back the shares of the given
stock. The rationale behind short selling a stock is that you hope to make a
profit from a nontrivial decrease in the stocks price. If you sell the borrowed
shares of stock for $50 per share and the share price drops to $45 a month later,
then you can use your proceeds to buy back the shares and still have a payoff
of $5 per share (excluding transaction fees). When you buy back and return the
borrowed shares, you are said to have closed the short position.
Next, consider an option, i.e., a legal contract between two parties whereby
one party (the issuer/writer) sells to the other (the holder) the right, but not the
obligation, to buy from or sell to the issuer a fixed amount of a security (e.g.,
stock) at a preagreed price (called the strike price or exercise price) on or by a
preagreed date (called the expiration date). In particular, a call option is a legal
contract between a buyer (holder) and seller (issuer) granting the holder the
right, but not the obligation, to buy a stipulated amount of the asset from the
issuer at the strike price on or by the expiration date.
Short selling an option contract is issuing an option contract. For example,
short selling an equity call option, i.e., a call option contract on a stock means
that you are obligated to sell 100 shares of the stock at the strike price if the
option is exercised and you are assigned the exercise. Specifically, when an
equity option is exercised by a holder, the exercise is randomly assigned (in
the USA, by the Options Clearing Corporation) to a market participant who
short sold the same call option (i.e., same underlying stock, strike price, and
expiration). If you are assigned, then you close the short position by obtaining
100 shares of the stock (if you do not already have them) and sell each share at
the strike price. If you were not assigned yet, then you can close the position by
buying back the exact call contract (same stock, strike price, and expiration). In
general, you close a buy position in an option by selling the exact option and
close a short-sell position by buying back the exact option.
In our Markowitz context, for any proper subset of securities in a portfolio
that are short sold, we always use the proceeds along with the initial capital to
purchase the units of the remaining securities. We shall not consider a portfolio
where all its securities are short sold.
Now, at t0 our portfolio has n1 , . . . , n N units, respectively, of securities 1
through N. When ni > 0, it means that ni units of the ith security are bought
at time t0 , while for ni < 0, the interpretation is that ni units of the security are
short sold at t0 . When no action is taken on the ith security, we write ni = 0.
Furthermore, by (3.3) the ith weight can be written as
n i S i ( t0 )
wi = , i = 1, . . . , N. (3.6)
V P ( t0 )
Since Si (t0 ) > 0 and V P (t0 ) > 0, we see from (3.6) that the weight wi has the
same sign as ni . We then interpret the sign of the weights as:
wi > 0 means buy ni units of the ith security at time t0 (long position).
wi = 0 means neither buy nor sell the ith security at time t0 (flat position).
wi < 0 means short sell ni units of the ith security at time t0 (short position).
Equation (3.4) also shows that the sum of the weights is still unity as in (3.2),
even if a proper subset of weights is negative:
w1 + + w N = 1.
When no short selling is used to construct the portfolio, the weights satisfy
wi 0 (no short selling), i = 1, . . . , N.
Consequently, in the absence of short selling, since the weights are nonnegative
and sum to unity, we always have
0 wi 1, i = 1, . . . , N.
Example 3.2. Suppose that you identified a stock and a call option contract on
the stock to create a portfolio at t0 . Assume that the positions you take in these
securities are to long (buy) (t0 ) shares of the stock priced at S(t0 ) per share
and to short (sell) the call option contract on the same stock, where the call is
sold at price C (t0 ) per share of the stock. In practice, call option contracts are
typically based on 100 units of the underlier, but for modeling purposes, it is
simpler to quote the call price per unit of the underlier.
Since long positions are an inflow of securities into the portfolio and short
positions are an outflow, they are represented by positive and negative signs,
respectively, when tallying the total value of a portfolio. The portfolios value
at t0 is then
V P ( t0 ) = S ( t0 ) ( t0 ) C ( t0 ) . (3.7)
Equivalently, you can obtain (3.7) if you unwind the two positions. Specifically,
liquidate (sell) the (t0 ) shares of the stock at t0 , which yields a cash inflow
of S(t0 ) (t0 ), and close your short position on the call (assuming it was not
assigned at t0 ). The latter means that you buy back a call contract on the same
stock and with the exact strike price and expiration date. The latter yields a
cash outflow of C (t0 ), which is represented mathematically as C (t0 ). The
total value of the portfolio at t0 is then the net sum of these cash flows, which
gives (3.7).
The trading strategy that created this portfolio is
n ( t0 ) = ( ( t0 ) , 1 ) .
In other words, short selling the call contract brings in proceeds of C (t0 ), which
when added to the initial capital V P (t0 ) in (3.7) yields the funds S(t0 ) (t0 ) to
buy (t0 ) shares of the stock.

Example 3.3. Suppose that we selected stocks 1, 2, and 3 to create a portfolio.

Assume that the current prices of the stocks are:
$40 per share for stock 1

Starting with an initial capital of $5, 000, we create a portfolio by distributing

the money across the three stocks using the following weights:
w1 = 20%, w2 = 50%, w3 = 70%.
Note that w1 + w2 + w3 = 1.
Let us interpret the meaning of the above weight assignment. First, the
weights state that we use the following trading strategy to form the portfo-
lio:
0.2 $5, 000
n1 = = 25 (short sell 25 shares of stock 1)
$40
0.5 $5, 000
n2 = = 35.7143 (buy 35.7143 shares of stock 2)
$70
0.7 $5, 000
n3 = = 350 (buy 350 shares of stock 3).
$10
Specifically, we create the portfolio by first short selling 25 shares of stock 1 to
obtain $1,000, which is 20% of the initial capital (since the proceeds are from
a short position). Adding these proceeds to the initial capital, we then have
$6,000 to invest in stocks 2 and 3.
Weight w2 tells us that we take 50% of the initial capital $5,000 to buy 35.7143
shares of stock 2. This reduces the initial capital to $2,500. Weight w3 indicates
that we use 70% of the initial capital, i.e., $3,500, to purchase 350 shares of
stock 3. Though the cost of the purchase exceeds the $2,500 remaining from
the initial capital, we have an extra $1,000 from the short sale to cover the
purchase. Finally, observe that (3.4) holds (to two decimal places)
$5, 000.00 = 25 $40 + 35.7143 $70 + 350 $10.
Note that using four decimal places in 35.7143 gives the desired accuracy,
while 35.71 would yield 4, 999.70.

To avoid confusion about whether short selling is or is not allowed in a

portfolio, we shall abide by the following convention:
Unless stated to the contrary, assume that short selling is allowed

in constructing a portfolio.
3.1.5 Expected Portfolio Return Rate
Consider the investment interval [t0 , t f ]. As with a security, Equation (2.2) on

page 19 yields that the return rate (the percentage you get back) of the portfo-
lio for the given investment interval comes from the percentage change in its
market value and from the total cash dividend it pays:
V P ( t f ) V P ( t0 ) D P ( t0 , t f )
R P ( t0 , t f ) = + , (3.8)
V P ( t0 ) V P ( t0 )
where V P (t0 ) and V P (t f ) are the values of the portfolio at the start and end of
the investment period, and DP (t0 , t f ) is the total cash dividend during [t0 , t f )
from all the securities in the portfolio. For the investment interval, the amount
you get back beyond the initial investment is the percentage R P (t0 , t f ) of the
initial investment V P (t0 ):
R i ( t0 , t f ) S i ( t0 ) , (i = 1, . . . , N ).
Since the investment interval [t0 , t f ] is fixed, we use the following simpler
notation:
R P = R P ( t0 , t f ) .
By (3.4), the initial portfolio market value is
V P ( t 0 ) = n 1 S1 ( t 0 ) + + n N S N ( t 0 )
and the final portfolio market value is
V P ( t f ) = n 1 S1 ( t f ) + + n N S N ( t f ) . (3.9)
The total cash dividend received from the securities during [t0 , t f ) is
DP (t0 , t f ) = n1 D1 (t0 , t f ) + + n N D N (t0 , t f ). (3.10)
Using (3.9), the portfolio return rate (3.8) from t0 to t f becomes
N
ni
RP = V (t ) i f S ( t ) S (
i 0t ) + D ( t
i 0 f, t )
i =1 P 0

N
n i S i ( t0 ) S i ( t f ) S i ( t0 ) + D i ( t0 , t f )
= V P ( t0 ) S i ( t0 )
.
i =1
By (3.1) and (3.6), the portfolio return rate is the weighted sum of the securities
return rates:
N
RP = wi Ri . (3.11)
i =1
The expected (or mean) portfolio return rate for the period [t0 , t f ] is then
N
P = E( RP ) = wi i , (3.12)
i =1
where
i = E ( Ri ) (3.13)
is the expected return rate of the ith security. In (3.12), every weight wi is assumed
nonrandom, unless stated otherwise, and each expected return rate i is assumed finite
and known. The weights are to be determined in the search for an efficient port-
folio, while the expected return rates 1 , . . . , N are typically estimated using
samples of historical return rates.
Example 3.4. Consider a portfolio with two stocks over a time interval [t0 , t f ]
corresponding to the next month. Denote the expected return rate over the next
month of a stock in the portfolio by

monthly = E R(t0 , t f ) .
Suppose that t0 = t0 > t1 > > tn denotes a sample of end-of-month to end-
of-month trading dates for m consecutive months from the present date t0 into
the past. Denote the corresponding historical return rates as follows:4
! (tn , tn1 ),
R ! (tn1 , tn2 ),
R ..., ! (t1 , t0 ).
R
The stocks theoretical ensemble expected monthly return rate monthly is esti-
mated using the time average of the monthly return data (see Exercise 3.12):
1 n !
n j
Rmonthly = R ( t j , t j 1 ) .
=1

Remark 3.1. The question of whether a data-sample time average Rmonthly

is an accurate approximation of the theoretical expected value monthly is a
rather thorny issue. Should we have used weekly or daily data rather than
monthly? Or, how many return rates should we have used? In general, the
sampling frequency (daily, monthly, quarterly, etc.) and sample size used to
estimate statistical financial quantities will depend on the context and typi-
cally become a debatable matter. See Graham, Smart, and Megginson [8, p.
212] for a discussion.

4Note that the returns are notationally the reverse of the case for future times: for past times tj < tj1 ,
! ( t , t ) instead of R( t j1 , t j ), which is for future times t j1 < t j .
we use R j j 1
3.1.6 Portfolio Risk
We saw that since the securities in the portfolio have random futures prices
and (in general) random future dividends, the portfolio return rate R P is also
random. This uncertainty in the return rate gives rise to the portfolios risk. In
other words, portfolio risk is determined by how much the possible values of
the random portfolio return rate R P can spread away from the expected return
rate P . More precisely, the risk of a portfolio is modeled in Markowitz theory
by the standard deviation of its return rate:
" #
P = Var( R P ) = E ( R P P )2 . (3.14)
Sometimes the standard deviation of a random variable X will be called the

volatility of X. Note that some authors refer to the variance P2 , rather than
the volatility P , as the risk of the portfolio, but we shall not abide by that
usage. The larger the portfolio risk P , the more the portfolio return rate R P can
spread away from the expected return rate P , while the smaller P becomes,
the closer R P concentrates to P .
The portfolio variance P2 is given as follows explicitly in terms of the volatil-
ities of the individual securities return rates and the covariances of these
return rates:
N
P2 (w1 , . . . , w N ) = wi 2 i2 + 2 wi w j ij , (3.15)
i =1 1 i < j N
where " #
i = Var( Ri ) = E ( Ri i )2 (3.16)
is the volatility of Ri and

ij = Cov( Ri , R j ) = E ( Ri i ) R j j = ji
is the covariance of Ri and R j , with the variance of Ri denoted by
ii = i2 .
In (3.15), the volatilities i s and covariances ij s are assumed to be finite and known.
3.1.7 Risks and Covariances of the Portfolios Securities
The Markowitz model (3.15) of portfolio risk implies that the risk of a portfolio
comes from two sources: the weighted contributions of the variances i2 , where
i = 1, . . . , N, of the individual securitiess returns and the covariances ij , where

1 i < j N, between the returns of all pairs of the securities. Putting this
informally in terms of the portfolio variance, we have

$ % weighted sum of weighted sum of
portfolio variance = the + the .

securities variances securities covariances
In the current section, we review some basic results about the volatility i and
covariances ij of the securities in a portfolio.
Risk of a Security
At the start of the Section 3.1 we informally defined a risky security as one
whose return rates cannot be predicted with certainty. In Markowitz setting,
risk is modeled more narrowly using volatility. Specifically, the risk of the ith
security is modeled by i for i = 1, . . . , N. In other words, the ith securitys risk
is a measure of how much the random return rate Ri spreads about the secu-
ritys expected return rate i . The risks 1 , . . . , N , are usually estimated from
historical data (see Exercise 3.12).
Remark 3.2. The portfolio risk (3.14) and security risk (3.16) measure how the
random return rates R P and Ri disperse above and below the expected return
rates P and i , respectively. Some have argued that risk should instead be
modeled by how much the return rates spread below the mean (downside risk)
or the probability of the return rates being below some threshold (shortfall
probability). Later we shall explore three portfolio risk measures: the Sortino
ratio (Section 4.2.2), the maximum drawdown (Section 4.2.3), and the value-
at-risk (Section 4.2.5). See Grinold and Kahn [9, pp. 4146] for a critique of
these risk measures relative to the standard deviation. Throughout our text,
however, the primary measure of risk shall be the standard deviation.

Example 3.5. (We continue with Example 3.4 on page 95.) Let us consider the
variance of a stock in the portfolio using n historical monthly consecutive
return rates over times t0 = t0 > t1 > t2 > > tn1 > tn . The data runs from
the past time tn to the present time t0 : R! (tn , t ), . . . , R! ( t , t ), R
! ( t , t ).
n 1 2 1 1 0
The theoretical variance month of the stock for the next month is estimated
2
using the sample monthly variance ! monthly

2 , which can be expressed as

n
1
R!2 (tj , tj1 )
2
! 2
monthly = n Rmonthly . (3.17)
n1 j =1
Exercise 3.12 involves a computation of !

monthly
2 .

Mathematically, the risk of a security is nonnegative. What happens when

the risk vanishes? Intuitively, the more tightly concentrated the return rates are
about their mean, the smaller the risk and vice versa. The following proposi-
tion expresses this insight for discrete random variables, which is the context
when dealing with data:
Proposition 3.1. Let X be a discrete random variable with finite mean = E ( X ) and
finite variance 2 = Var( X ). Then the volatility of X vanishes if and only if X =
almost surely, i.e., with probability 1.
Motivated by Proposition 3.1, we define a risky security to be one whose return
rate has a positive volatility.
Unless stated to the contrary, assume i > 0 for i = 1, . . . , N.
Covariance Between Two Securities
The other contributor to portfolio risk is the weighted collection of the covari-
ances of the random return rates of the securities in the portfolio. We shall see
in Section 3.7 that, for a portfolio with a sufficiently large number of different
securities, the weighted sum of the covariances of the securities dominates the
weighted sum of the securities volatilities. In this section, we instead review
some basic insights into the covariance of a pair of return rates of risky secu-
rities using the associated correlation coefficient. The correlation coefficient of
the return rates Ri and R j of the ith and jth securities will be written as follows:
ij
( Ri , R j ) = = ij ,
i j
The respective risks of the two securities are i and j , while the covariance is
ij = Cov( Ri , R j ).
In general, the (Pearson) correlation coefficient of random variables X and Y
with nonzero volatilities X and Y , respectively, is defined by
Cov( X, Y )
( X, Y ) = .
X Y
A basic property is
1 ( X, Y ) 1.
The correlation coefficient is a unit-independent measure of how X and Y
vary relative to each other, which is not the case for the covariance Cov( X, Y ),
where we assume the units carry a positive sign. This is a special case of the
following general property showing how the covariance and correlation coef-
ficient behave under affine transformations of X and Y (see Exercise 3.16 on
page 146):
Cov( aX + b, cY + d) = ac Cov( X, Y ),
and for a c = 0,
( aX + b, cY + d) = ( X, Y ) with +1 if a c > 0 and 1 if a c < 0.
The securities correlation coefficients ij , where i, j = 1, . . . , N, are assumed

known and are estimated using historical data. In the discrete context of est-
imates from data, we can think intuitively of ij as measuring the degree to
which the random return rates of the ith and jth securities move together along
a straight line. The following proposition captures this insight more generally:
Proposition 3.2. Let X and Y be discrete random variables. Then ( X, Y ) = 1 if

and only if Y = aX + b with probability 1, where a = 0 and b are real numbers with
a > 0 corresponding to ( X, Y ) = +1 and a < 0 to ( X, Y ) = 1.
In other words, for any two discrete random variables X and Y, which is the
setting when working with data, the closer the random variables are to being
perfectly positively correlated, i.e., ( X, Y ) = 1, the more likely the values of X and
Y are close to a positively sloped line. Similarly, the closer the random variables
are to having a perfectly negative correlation, ( X, Y ) = 1, the more the values
of X and Y concentrate near a negatively sloped line. For 1 < ( X, Y ) < 1, we
then have varying degrees of how much the values of X and Y spread away
from a straight line.
A pair of random variables X and Y is called uncorrelated if ( X, Y ) = 0. This
means that a scatter plot of possible values of the two random variables has
no linear relationship and so may appear as a cluster of independent points
or points showing an overall nonlinear relationship. Indeed, if X and Y are
independent, then they are uncorrelated. The converse is not true since it is
possible for two uncorrelated random variables to be dependent, though their
dependence will be nonlinear.
As noted earlier, the covariances ij and correlation coefficients ij are esti-
mated using historical data. There are several Web resources that compute the
correlation coefficients of pairs of stocks.5
Example 3.6. (We continue with Example 3.4 on page 95.) Let Rmonth A and
B
Rmonth be the random return rates over the next month of the two stocks in
the portfolio. Write their covariance and correlation coefficient as follows:
month
AB
AB
month = Cov Rmonth
A B
, Rmonth , AB
month = .
month
A month
B
5 For example, see the Correlation Tracker at: http://www.sectorspdr.com/correlation.

The theoretical covariance monthAB and correlation coefficient month AB of the

stocks for the next month are estimated using the sample data as follows:
n
1 ! A A ! B B
n 1 j
! AB
monthly = R ( t , t
j j 1 ) R monthly R ( t , t
j j 1 ) R monthly
=1
and
!
monthly
AB
AB
!monthly = ,
!
monthly
A !
monthly
B
respectively. Here ! monthly

A and ! monthly
B are the estimated volatilities of the
stocks for the next month. Note that for historical data, we have tj < tj1 .
Exercise 3.12 illustrates these estimates.
3.1.8 Expectation and Volatility of Portfolio Log Return
Throughout this section assume that the portfolio pays no dividend.
Portfolio Log Return for a Single Time Horizon
Suppose that a nondividend-paying asset has a current value of $500,000 and

its value 1 month later is $505,000. At what interest rate6 was $500,000 continu-
ously compounded to reach the value $505,000 after 1 month? In other words,
we want an interest rate r such that
$505, 000 = er ( 12 ) $500, 000,

1
which yields an annual interest rate of

$505, 000
r = 12 ln .
$500, 000
Expressing the interest as a monthly rate, we obtain

1 $505, 000
r = ln = 0.995% (per month).
12 $500, 000
The continuously compounded interest rate,

$505, 000
ln = 0.995%,
$500, 000
is called the log return of the asset over a month.
6 Recall from Chapter 2 that interest rate is per annum, unless otherwise stated.
Using the logarithmic return is advantageous not only for continuous com-
pounding but also for discrete compounding problems. In general, for a
nondividend-paying portfolio with initial value V P (t0 ) and end-of-period value
V P (t f ), we define the portfolio log return from t0 to t f to be

V P (t f )
span = ln
L
r P, .
V P ( t0 )
Intuitively, the log return r P, L

span is the continuously compounded rate across
[t0 , t f ] that transforms VP (t0 ) to VP (t f ), namely,
VP (t f ) = VP (t0 ) er P, span .
L
In the example above, t0 is the current time, t f is 1 month from now, and the
log return over [t0 , t f ] is
span = 0.995%.
L
r P,
For the typical situations we shall consider, the initial value V P (t0 ) is known
and V P (t f ) is a random future value, so r P,
L
span is random. In this case, the
span over [t0 , t f ] are denoted by

L
expectation and volatility of r P,
#
L
P, span =E L
r P, span , L
P, span = L
Var r P, span .
Portfolio Log Return for Different Time Horizons
Now, let us consider how the portfolio log returns, as well as their expectation
and volatility, behave under different time horizons. Divide [t0 , t f ] into n equal-
length subintervals:

[ t0 , t1 ], [ t1 , t2 ], , [ t n 1 , t n ], t n = t f , t j t j 1 = h n .
n
Let V P (t j ) be the value of the portfolio at time t j , where j = 1, . . . , n. The log
return from t j1 to t j is given by

V P (t j )
prd (t j ) = ln
L
r P, .
V P ( t j 1 )
These log returns over the subintervals relate to the log return over the entire
interval as follows:

VP (tn ) V P ( t1 ) V P ( t2 ) V (tn ) n VP (t j )
L
r P, span = ln
V P ( t0 )
= ln
V P ( t0 ) V P ( t1 )
P
V P ( t n 1 )
= ln V P ( t j 1 )
j =1
n
= r P,L prd (t j ). (3.18)
j =1
Before considering the expectation and volatility over different time hori-
zons, we need two assumptions:
For the first assumption, recall that the present value V P (t0 ) of the port-
folio is known and reflects all available information about the asset at the
current time t0 .7 Suppose that the value of the portfolio V P (t1 ) at the future
date t1 is dependent on information that is not known today. Similarly, the
value V P (t2 ) at date t2 is assumed to be based on information not known
on date t1 . For this reason, we assume that the random log returns
prd (t1 ), prd (tn )

L L
r P, ..., r P,
are uncorrelated random variables across the n successive subintervals:

Cov r P,L
(
prd k t ) , r L
(
P, prd jt ) = 0,
where k = j and k, j = 1, . . . , n.
For the second assumption, recall that by the one-period assumption, we
do not make changes to the portfolio during the time interval [t0 , t f ]. Conse-
prd (t j )
L
quently, we assume that the probability distributions of the log returns r P,
across the future subintervals are identically distributed. Write the expectation
and volatility of the log returns for each subinterval as
#
prd = E r P, prd (t j ) , prd = prd (t j ) ,
L L L
P, P, L
Var r P,
where j = 1, . . . , n.
We can now relate the expectation of the log return for the full time interval
[t0 , t f ], where t f = tn , to the expected log return over the n subintervals. Using
(3.18), we obtain

V ( t ) n
r P,L prd (tj ) = n P,L prd.
P n
L
P, = E ln = E (3.19)
span
V P ( t0 ) j =1
Hence, by (3.19) the expected portfolio log return over the time span is n
times the expected portfolio log return over a period. For instance, an annual
7 This issue pertains to the Weak Efficient Market Hypothesis.

expected portfolio log return is 12 times the monthly expected portfolio log
return. In other words, longer time horizons have a higher expected log return com-
pared to shorter ones.
For the variance of the log return over [t0 , t f ] with uncorrelated and identi-
cally distributed log returns across consecutive subintervals, we obtain
2 2
L
P, span = n P,
L
prd ,
which yields the volatility:

span =
L L
P, n P, prd . (3.20)
The volatility of the portfolio log return increases as the square root of the
number of periods in the time horizon increases. Longer time horizons will have
higher volatility than shorter
time horizons. In particular, the volatility of an an-
nual portfolio log return is 12 times the volatility of a monthly portfolio log
return.
Relating the Portfolio Log Return and Portfolio Return Rate
Lastly, the portfolio log return r P, span relates to the portfolio return rate R P as
follows:

V P ( tn )
r P, span = ln = ln (1 + R P ) ,
V P ( t0 )
where tn = t f . Taylor expanding the log return yields
R2P R3 R4
r P, span = R P + P P + for | R P | < 1. (3.21)
2 3 4
Consequently, if | R P | is sufficiently small, then the log return and the return
rate are approximately equal:
r P, span R P , | R P | 1.
However, over a sufficiently long time span, there is no guarantee that the
risk of a portfolio will not increase or the magnitude of its return rate will
be sufficiently small. In this case, we cannot treat the portfolio log return and
portfolio return rate as approximately equal.
Looking Ahead
The remainder of the chapter will show how to apply the infrastructure of
the Markowitz model to selecting the weights that produce efficient portfo-
lios as well as optimal portfolios, which is an efficient portfolio determined by

maximizing an investors expected utility function (Section 3.6). It will also be
shown how diversification reduces portfolio risk.
We shall first study portfolios with two risky securities and then extend to N
risky securities. The original Markowitz treatment (e.g., [18]) employs a geo-
metric approach using iso-mean and iso-variance curves and surfaces. This
approach, though, gets extremely intricate in higher dimensions. Instead, we
shall carry out the efficient portfolio selection using the optimization approach
by Merton [20].
3.2 Two-Security Portfolio Theory
Suppose that you have an amount of money V P (t0 ) to create a portfolio with
two risky securities. After an appropriate amount of due diligence, you have
identified two risky securities to buy today t0 and hold in the portfolio until
future date t f . The fundamental portfolio question we shall address is
What percentage of the money V P (t0 ) should you allocate today to each
security to create an efficient portfolio?
In other words, find the weights that give rise to an efficient portfolio.
Let us first collect some quantities needed for Markowitz model in a setting
of two securities, say, securities 1 and 2. The respective random return rates
are R1 and R2 . The securities expected return rates 1 and 2 , risks 1 > 0 and
2 > 0, and correlation coefficient 12 = are assumed to have been estimated
(either by you or by a company offering such a service). For real data the two
securities will typically not have identical expected return rates (1 = 2 ), iden-
tical risks (1 = 2 ), or return rates with either a perfectly positive correlation
( = 1) or perfectly negative correlation ( = 1). These mathematical ideal-
izations will not be assumed by default and will explicitly be identified when
considered:
Unless stated to the contrary, assume that 1 = 2 , 1 = 2 , and
1 < < 1. Without loss of generality, we assume 2 > 1 > 0.
Though it may not be obvious at this stage, the assumption on is actu-

ally already included in the Markowitz setup. We show this and more in Sec-
tion 3.2.1.
3.2 Two-Security Portfolio Theory 105
3.2.1 Preliminaries
Before addressing how to determine the weights that produce an efficient port-
folio, we shall present the needed quantities conveniently in matrix form.
For two securities, the weights are given by a vector

w1
w= .
w2
Since w1 + w2 = 1, we have w = 0. Furthermore, letting

1
e= ,
1
the sum-of-weight condition becomes
1 = w1 + w2 = w T e.
The random return rates R1 and R2 , the expected return rates 1 and 2 ,
and the covariances of R1 and R2 can also be compiled conveniently in matrix
form:

R1 1 12 1 2
R= , = E ( R) = , V= ,
R2 2 1 2 22
where 1 2 = 12 = Cov( R1 , R2 ). Here w, R, , and V are the weight vector, re-

turn rate vector, expected return rate vector, and covariance matrix, respectively, of
the portfolio. The portfolio
, expected return (i.e., P (w) = E ( R P )) and portfolio
risk (i.e., P (w) = Var ( R P )) are
P ( w ) = w 1 1 + w 2 2 = T w (3.22)
"
P (w ) = w21 12 + w22 22 + 2 w1 w2 1 2 = w T V w. (3.23)
The assumption of distinct expected return rates is captured by
1 = 2 = c e for any constant c.
In particular, we have = 0 (i.e., for c = 0) and = e. In general, and e are

linearly independent vectors.
We saw that the covariance matrix V of security returns is positive definite
(page 86). Recall that an n n real symmetric matrix is defined to be positive
definite if
xT A x > 0 for all x = 0 in R n . (3.24)
This is equivalent to each leading principal submatrix having a positive det-

erminant (Sylvesters criterion). In particular, the positive definiteness of V is
equivalent to
12 > 0 and det V = (1 2 ) (1 2 )2 > 0.
Since 2 > 1 > 0 and V is positive definite within our setup of the Markowitz
model, we also have 1 < < 1.
We now collect some basic consequences of the positive definiteness of the
covariance matrix V and introduce additional notation:
Since w = 0, by (3.24) the positive definiteness of V implies that the portfolio

always has risk, i.e.,
P (w) = w T V w > 0 (3.25)
for all portfolio weight vectors w.
The positive definiteness of V implies that V is invertible and its symmetric
inverse
1 1 22 1 2
V = , (3.26)
det V 1 2 12
is positive definite.
We introduce the following quantities which are needed later:
12 + 22 2 1 2
A e T V 1 e = (3.27)
det V
2
T 1 2 1 2 1 + 12 1 2 2
B V e= (3.28)
det V
12 22 + 22 21 2 1 2 1 2
C T V 1 = (3.29)
det V
( 1 2 ) 2
A C B2 = . (3.30)
det V
Let us consider their signs. Since e = 0 and = 0, the positive definiteness
of V 1 gives
A>0 (3.31)
C > 0. (3.32)
For the cross term B, we cannot draw any conclusion about its sign at this
stage. However, Equation (3.32) and the linear independence of and e
yield8
8If B C e = 0, then the linear independence of and e implies B = C = 0. This contradicts C > 0.
Hence, B C e = 0.
B C e = 0.
Consequently, the positive definiteness of V 1 gives
( B C e) T V 1 ( B C e) > 0.
This implies (Exercise 3.19)
A C B2 > 0. (3.33)
Note that we could have obtained (3.33) directly from (3.30) since 1 = 2
and det V > 0. However, our previous arguments for the signs of A, C, and
A C B2 were independent of the detailed expression for these quantities.
This will allow us to carry these results over to the N-security case.
3.2.2 Efficient Frontier of a Two-Security Portfolio
To determine an efficient portfolio for two securities, we must find a weight

vector for the two securities such that the resulting portfolio has the mini-
mum risk for a given portfolio expected return rate and the maximum expected
return rate for a given portfolio risk. The set of all such portfolio weight vectors
will determine the two-security efficient frontier.
We shall approach this problem by first determining a weight that gives the
smallest portfolio risk given an expected portfolio return rate . More precisely,
we seek a portfolio weight vector w = w that solves the following optimiza-
tion problem:

minimize P (w) = w T V w (3.34)
subject to wT e = 1 and w T = , (3.35)
where < < . There is no constraint wi 0 on the weights, i.e., unlimited

short selling is allowed. The constraints (3.35) are two equations in the two
unknowns (w1 , w2 ) and have a unique solution w = w . In fact, Equation (3.35)
is given explicitly by
w1 + w2 = 1, w1 1 + w2 2 = ,
or, equivalently, in matrix form by

1 1 w1 1
= . (3.36)
1 2 w2

K
Because = e, which is equivalent to 1 = 2 , we see that
det K = 2 1 = 0.
Consequently, Equation (3.36) has a unique solution:

w1, 1 1 1 2
w =K = .
w2, 2 1 1
Now, since w1 + w2 = 1, we can write the weights in terms of a single param-

eter w1 = w:
w1 w
w= = .
w2 1w
The portfolio weight vector w corresponding to is then written simply as

w 1 2
w = = . (3.37)
1 w 1 2 1
Note that since unlimited short selling is permitted, we have < w <
and so < < . If short selling is forbidden (i.e., 0 w 1) and if we
assume 1 < 2 , then 1 2 .
Though the expression of w in (3.37) is specific to the two-security case, we
can actually cast w in the following form, which carries over to N-securities:

C B 1 A B
w = V e + V 1 . (3.38)
AC B2 AC B2
Establishing (3.38) is a rather lengthy computation (Exercise 3.20). Moreover,
the form (3.38) is more complicated than (3.37) and its origin seems mysterious
at this stage. However, it will appear naturally during the N-security efficient
frontier analysis in Section 3.3.2 and allow us to link back readily to the two-
security efficient frontier (see page 123).
Given that the constraint equations in (3.35) have a unique solution w , this
is the only portfolio weight vector available to solve (3.34). In other words,
the solution w yields a unique portfolio risk P (w ) given a portfolio exp-
ected return rate P (w ) = . The quantity P (w ) is then the minimum pos-
sible portfolio risk, being the only risk associated with . However, this is not
enough to decide whether w determines an efficient portfolio because we do
not know whether P (w ) = is the maximum possible expected return rate
given P (w ).
We shall show that will have to lie in a restricted range in order for w
to give an efficient portfolio. Allowing the expected portfolio return to vary
over R, we shall see that the corresponding portfolio risk P (w ) traces out
a branch of a hyperbola. The turning point on this branch will be the global
minimum portfolio risk. The efficient frontier will be the curve segment from
the turning point to the upper part of the given branch of the hyperbola. We
now detail the computation of the efficient frontier.
First, to emphasize that the portfolio risk is a function of , we also write
P (w ) = P ().
Next, we express the portfolio variance as a quadratic in . Since

P2 () = wT V w = 12 + 22 21 2 w2 + 2 1 2 22 w + 22 ,
substituting (using (3.37))

2
w =
1 2
yields
P2 () = A 2 + B + C, (3.39)
where
12 + 22 2 1 2
A=
( 1 2 ) 2
2
2 1 2 1 + 12 1 2 2
B = 2
( 1 2 ) 2
12 22 + 22 21 2 1 2 1 2
C= .[1pt]
( 1 2 ) 2
Using Equations (3.27), (3.28), (3.29), and (3.30), we obtain
A 2B C
A= , B= , C= .
A C B2 A C B2 A C B2
Hence, the portfolio variance (3.39) becomes
A 2 2 B + C
P2 () = . (3.40)
A C B2
To identify the graph of (3.40), complete the square of the numerator to get
2
A B 1
P2 () = + , (3.41)
A C B2 A A
and introduce the following (upright) variables:
P = P (), P = .
Then (3.41) is equivalent to the equation of a hyperbola in the (P , P )-plane:

2
2P P B
A
= 1, (3.42)
1 AC B2
A A2
Because P > 0, the graph is the branch of a hyperbola opening to the

right along the P -axis. Each point on the graph is a portfolio risk-mean pair
(P (), ). Figure 3.1 illustrates a portion of the graph, which consists of solid
and dotted curves. The turning point or vertex of the graph is called the global
minimum-variance portfolio, denoted (G , G ), and given by

1 B
turning point = (G , G ) = , . (3.43)
A A
Fig. 3.1 The solid curve (including ) shows the Markowitz efficient frontier M E,2 for a two-security
portfolio. The turning point of the curve is the global minimum-variance portfolio ( G , G ). The
efficient frontier curve and the dotted curve extend to infinity when unlimited short selling is allowed
(since < < ). If there is no short selling and 1 < G < 2 , then 1 2 and the upper
portion of the efficient frontier ends at security 2, while the dotted curve ends at security 1. Here
1 < < 1, 1 = 2 , and 1 = 2
Since the turning point is the furthest point to the left on the graph, the port-
folio risk G is indeed the global minimum value of the portfolio risk function
P () as varies over R. Furthermore
Unless stated to the contrary, assume that G > 0, so B > 0.
This is a financially reasonable assumption because it requires the global

minimum-variance portfolio to have a positive expected return rate; other-
wise, no investor will be interested in the portfolio. Note that by (3.27) and
(3.28), we obtain
-
1 (1 2 )(1 2 )2
G = = (3.44)
A 1 + 22 2 1 2
2
2
B 2 1 2 1 + 12 1 2 2
G = = . (3.45)
A 12 + 22 2 1 2
The portfolio weight vector producing (G , G ) is given by (3.38)

C G B 1 G A B 1 V 1 e
w G w G = V e + V = . (3.46)
AC B2 AC B2 A
Explicitly

wG
wG = ,
1 wG
where
22 1 2 12 1 2
wG = , 1 wG = . (3.47)
12 + 22 21 2 12 + 22 21 2
Let us now determine the efficient frontier, which is the set of all efficient
portfolios. Recall that an efficient portfolio is determined by a risk-mean pair
(P , P ), where P is the smallest possible portfolio risk for the expected return
rate P and P is the largest possible expected return rate for the portfolio risk
P . Inspection of Figure 3.1 shows that the efficient frontier for the given plot
consists of the solid curve, including the turning point.
More generally, given a portfolio expected return rate P = , there is a
unique portfolio risk P = P () determined by (3.40) with P the minimum
possible portfolio risk associated with P . However, given the risk P , for
the pair (P , P ) to be efficient, we also need the expected return rate P to
be the maximum possible. Solving (3.40) for the expected return rate = P
yields either no solution, one solution, or two solutions depending on whether
P < G , P = G , or P > G , respectively. This is captured by Figure 3.1. In
fact, the expected return rate solutions correspond in Figure 3.1 to the inter-
section points of the vertical lines, P = constant, with the right branch of the
hyperbola. For P = G , the unique intersection point determines the portfolio
expected return P = G . Consequently, the turning point is an efficient port-
folio. For P = constant > G , there are two intersection points with the largest
possible portfolio expected return rate determined by the upper intersection
point. Indeed, the two-security efficient portfolios are given by the turning
point and the upper part of the hyperbola.
Equations (3.40) and (3.43) and the discussion above show that the Markowitz
efficient frontier ME,2 , i.e., the collection of all efficient two-security portfolios,
is given by
-
A2P 2 BP + C B
ME,2 = (P , P ) : P = , P . (3.48)
AC B2 A
The set of portfolio weight vectors that give rise to ME,2 is

C P B 1 P A B 1 B
WE,2 = w : w = V e+ V , P .
AC B2 AC B2 A
(3.49)
Finally, observe from Figure 3.1 that the efficient frontier indicates theoreti-
cally that to obtain a higher expected return rate the portfolio has to take on more risk.
Note, however, in a real-world setting, portfolio management involves more
complexity due to transaction costs, taxes, the trading platform, etc.
Example 3.7. Suppose that you have $2,000 to invest in two stocks, say, stocks 1
and 2, which have current share prices of $40.25 and $35.10, respectively. From
an analysis of historical data of the two stocks, suppose that
1 = 8%, 2 = 12%, 1 = 9%, 2 = 15%, = 0.5,
where these are annualized percentages. Using these two stocks, create an ef-
ficient portfolio that has an expected annual return rate of 20%.
Solution. The goal is to find a trading strategy9 (n1 , n2 ) such that n1 and n2 are
the respective number of shares of stocks 1 and 2 needed to build an efficient
portfolio with P = 0.2.
The initial capital is V P (t0 ) = $2, 000. Let us now collect the quantities from
Section 3.2.1 that are used in determining the portfolio weight vector. We shall
employ the expressions of these quantities that involve vectors and matrices
since that form carries over to the N-security analysis. First,

1 0.08 0.0081 0.00675
e= , = , V= ,
1 0.12 0.00675 0.0225
and

1 164.609 49.3827
V = .
49.3827 59.2593
Second
A = e T V 1 e = 322.6337449
B = T V 1 e = 30.1563786
C = T V 1 = 2.8549794
A C B2 = 11.7055327.
9 See (3.5) on page 89.
Third
C P B P A B
= 0.27135, = 2.93625.
AC B 2 AC B2
Note that the global minimum-variance portfolio,
B 1
G = = 9.3%, G = = 5.6%,
A A
does not meet the requirement of the portfolio we are trying to create since
P = 20% > G .
The portfolio weight vector is actually given by

C P B 1 P A B 1 2
w= V e+ V = .
AC B2 AC B2 3
Consequently, we create the desired efficient portfolio using the following

trading strategy:
w1 V P ( t0 ) (2) $2, 000
n1 = = = 99.379 (short sell 99.379 shares of stock 1)
S1 (t0 ) $40.25
w2 V P ( t0 ) 3 $2, 000
n2 = = = 170.9402 (buy 170.9402 shares of stock 2).
S2 (t0 ) $35.10
In other words, to create the portfolio, short selling n1 shares of stock 1

brings in
n1 $40.25 = 99.379 $40.25 = $4, 000.
Adding this amount to the initial $2,000 then allows one to buy n2 shares of
stock 2, which costs
n2 $35.10 = 170.9402 $35.10 = $6, 000.
The decimal places are maintained simply for mathematical consistency with
the amounts received from shorting and needed for purchasing. In an actual
trading setting, an integer number of shares is traded. Also, note that, to ob-
tain the required high expected portfolio return rate of 20%, the constructed
efficient portfolio ends up with a risk much higher than that of the individual
stocks:

P = w T V w = 56.2% max{1 , 2 } = 15%.

3.2.3 Reducing Risk Through Diversification
The discussion so far makes no mention about the diversification of the two
securities in the portfolio. We first discuss how diversification relates to the
correlation coefficient between the return rates of two stocks. Using two sepa-
rate online correlation coefficient Calculators,10 we see that for the time span
from August 10, 2010, to August 10, 2013, using adjusted closing prices, the for-
profit stocks Apollo Group, which owns the University of Phoenix, and Strayer
Education, Inc., which owns Strayer University, have a positive correlation of
0.77. Indeed, we would expect intuitively that on average two stocks from
the same business sector have a nontrivial positive correlation. On the other
hand, the correlation calculators output a negative correlation of 0.5 for
American Airlines11 and Exxon-Mobil. We also expect this intuitively since in-
creases in oil prices benefit oil companies, but hurt airlines due to the resulting
higher fuel cost. Finally, the correlation between the Apollo Group and Amer-
ican airlines was quite weak, namely, 0.15, which is expected intuitively
since the online education sector and the airline industry do not compete with
each other.
Remark 3.3. Bear in mind that there are always exceptions to the above sim-
plistic intuition. Historical-data estimates of correlation coefficients (and other
quantities) are part art and part science. The degree to which two stocks
vary relative to each other is influenced by overall market movements during
the time span of the datae.g., a period of an overall market rise lifts the
return rates of most stocks, creating positive correlations between them. Add-
itionally, correlation estimates are affected by the datas sample size, sample
frequency, etc.

Overall, as the correlation coefficient decreases from 1 down to 1, we are

considering pairs of securities whose return rates covary less and less in the
same direction, yielding more and more diversification in their return rates.
The two-security efficient frontier in Figure 3.1 was plotted for varying weights
w and fixed values of 1 , 2 , 1 , 2 , and . Suppose, instead, we consider vari-
ous pairs of securities with the same 1 , 2 , 1 , and 2 , but different correlation
coefficients, i.e., we allow to vary. To identify the associated efficient frontier,
each in (1, 1), we shall vary the weight w assuming no short selling, i.e.,
0 w 1. The portfolio risk is given by
10 We used the free online Correlation Tracker tool at (www.sectorspdr.com) and Stock Correlation
Calculator at Buyupside (www.buyupside.com/calculators).
11 Note that the ticker symbol of American Airlines at the time was AAMRQ, which changed after the
merger with US Airways.

"
P (w) = w2 12 + (1 w)2 22 + 2w(1 w)1 2 .
Since w(1 w) 0, we see that the portfolio risk decreases as decreases from
1 to 1 (the securities covary less and less in the same direction). In other
words, portfolio risk decreases as the diversification increases.
Figure 3.2 illustrates the above situation. The curves from right to left depict
the different efficient frontier curves for increasing diversification, i.e., as dec-
reases from positive to negative values over the interval (1, 1). Figure 3.2
then captures the benefit of diversification in the two-security setting, namely,
increasing diversification creates efficient-frontier curves that push to the left,
reducing the overall portfolio risk. Diversification in the N-security setting will
be explored in Section 3.7.
Is There a Riskless Portfolio with Two Risky Securities?
The answer is affirmative if we drop the requirement that 2 < 1 and consider
a portfolio with two risky securities having perfect negative correlation:
= 1,
which is an idealization. The portfolio variance becomes
P2 (w) = w2 12 + (1 w)2 22 2w(1 w) 1 2 = (w1 (1 w)2 )2 ,
Fig. 3.2 From left to right, the solid curves show typical efficient frontiers for a two-security portfolio
as the correlation coefficient varies over (1, 1) from negative to positive values. Each associated
set of feasible portfolios for a given has risk and expected return rates determined by the union of
the solid and dashed curves. The turning point of each curve identifies the efficient portfolio with the
lowest risk for the given curve
which yields the portfolio risk for w in R: P (w) = |w 1 (1 w) 2 | 0.

Hence, the smallest value of the portfolio risk is zero:
2
P (w ) = 0 at w = .
1 + 2
We now resume our assumption that 1 < < 1.
Diversification: Two Securities Versus One Security
Is it better to put all your money in two risky securities versus one risky secu-
rity? Intuitively, it would seem better to spread your money between the two
risky securities to lower your risk, i.e., not to put all your eggs in one basket.
We shall construct a portfolio with two uncorrelated risky securities having less risk
than a portfolio consisting of either one of the securities. In particular, we consider
securities 1 and 2 with risks 1 > 0 and 2 > 0 and correlation coefficient
= 0.
Then the two-security portfolio has variance given by
P2 (w) = (12 + 22 )w2 2 22 w + 22 .
Moreover, Equation (3.44) yields the global minimum-variance portfolio:
1 2 22 1 + 12 2
G = " , G = .
12 + 22 12 + 22
Furthermore, by (3.47) the global minimum portfolio has weight
22 12
wG = , 1 wG = . (3.50)
12 + 22 12 + 22
Note that since 1 > 0 and 2 > 0, we have

0 < w G < 1.
Consequently, the weight w G is not a short selling position.
Since
- -
1 12 2 22
" = < 1 and " = < 1,
2 + 2 12 + 22 2 + 2 12 + 22
1 2 1 2
we see that the smallest value of the portfolio risk satisfies
0 < G < 1 and 0 < G < 2 ,
or, more compactly,

0 < G < min{1 , 2 }.
3.3 Efficient Frontier for N Securities with Short Selling 117
Therefore
A portfolio of uncorrelated risky securities with its fraction w G of the ini-
tial capital invested in one security and 1 w G in the other will have less
risk than either of the two securities. Moreover, this portfolio involves no
short selling since 0 < w G < 1.
In other words, spreading the investment capital strategically as above be-
tween the two uncorrelated securities yields a portfolio with less risk than a
portfolio consisting of just one of the two securities.
3.3 Efficient Frontier for N Securities with Short Selling
Many of the ideas and quantities introduced for two securities (Section 3.2)
carry over naturally to N securities in determining the efficient frontier. In
this section, we shall compute the efficient frontier for an N-security portfo-
lio without putting any restrictions on short selling. The only restriction on the
weights will then be that they sum to unity. When there is no short selling, the
efficient frontier is analytically more complex and usually presented by using
numerical plots; see Section 3.4.
Suppose that, after researching a collection of different risky securities, you
have identified N of them for which you are confident in your estimates of
their expected return rates i , of their risks i , and of the correlations ij of the
return rates of each pair of the securities. It would also be unrealistic for all of
the securities to have the same expected return and risk and to have a perfect
correlation between any pair of return rates. For these reasons
Unless stated to the contrary, for any N-security portfolio, as-

sume:
The securities are risky: i > 0 for i = 1, . . . , N.
The expected returns 1 , . . . , N , risks 1 , . . . , N , and correla-
tion coefficients ij , where 1 i, j N and i = j, are known
and fixed.
None of the following occurs: identical expected returns
1 = 2 = = N ,
identical risks
1 = 2 = = N ,
or perfect correlation ij = 1 for any distinct pair i, j.
Unlimited short sales are allowed, i.e., < wi < .
At an initial time t0 , you want to use an amount of money V P (t0 ) to pur-

chase the N risky securities. Our goal is to form an efficient portfolio out of these
securities, i.e., a portfolio whose risk is the least for a given expected return
and whose expected return is the most for a given level of risk. In other words,
we shall determine the percentage of V P (t0 ) that should be allocated today to each
security to obtain an efficient portfolio.
3.3.1 N-Security Portfolio Quantities in Matrix Notation
The portfolio weight vector and column of 1s are

w1 1
w2 1

w = . , e = . .
.
. ..
wN 1
The summing of the weights to unity is expressed by
1 = w1 + + w N = w T e.
The weight space of an N-security portfolio that allows for unlimited short sell-
ing is $ %
WN = w R N : w T e = 1 . (3.51)
Note that WN is a line for N = 2 and plane for N = 3. In general, the space WN
is an ( N 1)-dimensional plane in R N = {(w1 , . . . , w N )} passing through the
N standard unit basis vectors:
e1 = [ 1 0 0 . . . 0 0 ] T , e 2 = [0 1 0 . . . 0 0 ] T , ..., e N = [0 0 0 . . . 0 1 ] T .
Of course, because a realistic weight space will not extend to infinity, it will be
a proper subset of the mathematical space WN .
The random return rates and expected return rates of the securities are

R1 1
R2 2

R = . , = E ( R) = . .
.
. .
.
RN N
The portfolio return rate and expected portfolio return rate can then be ex-
pressed as
N N
R P (w) = w T R = wi Ri , P (w) = E (w T R) = w T = wi i .
i =1 i =1
The assumption that the securities do not all have the same expected return
means that = ce for some constant c.
The covariance matrix of the return rates R1 , . . . , R N is
2
1 12 1N
22 2N

V = . .. ,
. . .
N2
where the entries below the diagonal are not shown since the covariance ma-
trix is symmetric, ij = Cov( Ri , R j ) = ji . The matrix V is invertible and so
positive definite (see page 86). It follows that V 1 is symmetric and positive
definite. Estimating the covariance matrix using historical data then requires
N ( N 1)
N+ estimates,
2
N ( N 1 )
which correspond to N variances and 2 correlation coefficients. For ex-
ample, one hundred stocks require 5, 050 estimates to determine V.
The portfolio risk is given by

P (w) = w T V w > 0.
Note that the positivity follows since V is positive definite and w = 0.

The quantities A, B, and C that were introduced in the two-security case
generalize naturally to an N-security portfolio. Essentially the exact arguments
used for the two-security setting give the following (see (3.31), (3.32), and
(3.33); page 106):
A = e T V 1 e > 0 (3.52)
1
B= V T
e (3.53)
C = T V 1 > 0 (3.54)
A C B > 0.2
(3.55)
It will be argued in the next section that it is financially reasonable to assume

B > 0 (as we did in the two-security case).
Finally, we shall employ the following notation for the gradient and Hessian
of a twice continuously differentiable function g(w ):

2 g 2 g 2 g

g w21 w1 w2 w1 w N
2 g 2 g
g . w1
2 g
w2 w N
= . , = w22
,
w . ww T .. ..
g . .
w N 2 g

w2N
where since the Hessian matrix is symmetric, we do not show the entries below
its diagonal. Two useful properties (Exercise 3.25) to keep in mind are that for
an n n real matrix A, the gradient and Hessian of x T Ax are given by
( x T Ax) 2 ( x T Ax)
= ( A + AT ) x, = A + AT , x Rn . (3.56)
x xx T
3.3.2 Derivation of the N-Security Efficient Frontier
We first find the portfolio with the smallest risk given an expected portfolio
return rate. We then vary through all possible expected portfolio return rates
to obtain the corresponding set of minimum-risk portfolios. The set of all such
pairs of portfolio risks and expected return rates forms the right branch of a
horizontal hyperbola. We shall argue that the portion of the branch from the
turning point to the upper part of the hyperbola forms the desired efficient
frontier.
Let us now detail the above. We wish to find a portfolio weight vector w
that solves the following:

Problem I: minimize P (w ) = w T V w (3.57)
subject to wT e = 1 and w T = . (3.58)
Note that unrestricted short selling is allowed. A major difference between the
above optimization problem and the two-security problem on page 107 is that
the constraints (3.58) are now two equations in N unknowns w1 , w2 , . . . , w N .
In other words, for N 3, there is not a unique portfolio weight vector w
satisfying (3.58). In fact, generically there are infinitely many solutions w of
(3.58) and, hence, infinitely many portfolio risks corresponding to for N 3.
However, we shall show that there is a unique portfolio weight vector w = w
yielding the smallest portfolio risk associated with the given . That is, we find
a unique solution of (3.57) and (3.58) together.
It is analytically simpler to solve instead the following optimization
problem:
wT V w
Problem II: minimize f c (w ) = , where c > 0, (3.59)
c
subject to wT e = 1 and w T = . (3.60)
Problems I and II have the same solution set (Exercise 3.6).12
Remark 3.4. The equivalence of Problems I and II is a simple example of obt-

aining a desired solution by first identifying the mathematical equivalence of
the two optimization problems and then solving the simpler of the two prob-
lems, which in this case is Problem II (since it does not involve the differenti-
ation of square roots). For a deeper study of how to exploit the equivalence of
portfolio optimization problems to develop an iterative scheme to find a solu-
tion, see Korn [12, p. 8] (compare with Korn and Korn [13, pp. 35]). Students
interested in the programming aspects of optimization problems are referred
to Vanderbeis text on linear programming [25].
It suffices to consider Problem II with c = 2, which is a convenient choice

since f c (w ) is a quadratic:
wT V w
minimize f (w ) =
2
subject to w T e = 1 and w T = .
Since the constraints are equalities, the tool for solving this problem is the
method of Lagrange multipliers.13 The Lagrange function for Problem II is
L(w, ) = f (w ) + 1 (1 w T e) + 2 ( w T ) = f (w ) + T h (w),
with multipliers and constraints given, respectively, by

1 h1 1 wT e
= , h(w ) = = .
2 h2 wT
By (3.56) and the fact that V T = V, we have the following:
L
(w, ) = V w 1 e 2 (3.61)
w
12 Do not attempt to show the equivalence via the Lagrange conditions (i.e., (3.64) to (3.66)). Simply
use the statement of the problems and logically imply one from the other.
13 For optimization problems with inequality constraints, the Karush-Kuhn-Tucker Theorem is em-
ployed.
L
(w, ) = h (w) (3.62)

2 L
(w, ) = V. (3.63)
w w T
The Lagrange Multiplier Theorem yields that w is a solution of Problem II
if and only if there is a pair (w , ), where is unique to w , satisfying
L
(w , ) = 0 (3.64)
w
L
(w , ) = 0 (3.65)

2
L
x T
(w , ) x 0 for every x = 0 such that (3.66)
ww T
both of the following hold
T
h1
0= (w ) x = e T x = ( x1 + + x N )
w
T
h2
0= (w ) x = T x = (1 x1 + + N x N ).
w
In other words, the set of solutions of Problem II is in 1-1 correspondence with
the set of solutions of (3.64)(3.66). We shall show that the latter set of equa-
tions has a unique solution (w , ), which, by the Lagrange Multiplier The-
orem, gives a unique solution w to Problem II, which in turn is a unique
solution to Problem I. Without loss of generality, we set c = 2 for convenience.
To determine solutions (w , ) of (3.64)(3.66), first observe that condition
(3.66) holds automatically because by (3.63) the Hessian matrix,
2 L
(w , ) = V ,
ww T
is positive definite. We next search for the pairs (w , ) satisfying (3.64). Let

1,
= .
2,
Equation (3.61) shows that (3.64) is equivalent to
V w = 1, e + 2, .
Multiplying through by V 1 , we get

w = 1, V 1 e + 2, V 1 (3.67)
or T
w = 1, e T V 1 + 2, T V 1 .
Though (w , ) solves (3.64), we also need the pair to satisfy (3.65). But Equa-
tion (3.65) is equivalent to the two constraint equations, which fortunately are
linear in the weight vector and so, by (3.67), linear in the multipliers:
1 = wT e = 1, (e T V 1 e) + 2, ( T V 1 e)
= wT = 1, (e T V 1 ) + 2, ( T V 1 ).
Employing the quantities A, B, and C, the above equations simplify to
1 = 1, A + 2, B (3.68)
= 1, B + 2, C. (3.69)
Note that B = T V 1 e = e T V 1 (since B is a scalar). The constraints in Equa-

tions (3.68) and (3.69) can be expressed more compactly in the matrix form:

A B 1, 1
= . (3.70)
B C 2,

A B
Setting K = , Equation (3.55) yields det K = AC B2 > 0. Because K
B C
is invertible, we obtain a unique solution for the Lagrange multipliers:

1, 1 1 C B 1
= = K 1 = ,
2, AC B2 B A
where
C B A B
1, = , 2, = . (3.71)
AC B 2 AC B2
Then the uniqueness of yields by (3.67) a unique portfolio weight vector:

C B 1 A B
w = V e+ V 1 . (3.72)
AC B2 AC B2
We call w the minimum-variance portfolio weight vector with expected portfolio
return rate . Observe that, for a two-security portfolio, Equation (3.72) coin-
cides with (3.38) (page 108).
To summarize, we found a unique pair (w , ) satisfying (3.64) and (3.65),
and automatically satisfying (3.66) by the positive definiteness of V. The
Lagrange Multiplier Theorem then implies that the weight vector w is a
unique solution of Problem II or, equivalently, Problem I. Note that our deriva-
tion of (w , ) drew fundamentally upon the linearity of the constraint equa-
tions and the positive definiteness of V .
With w available, let us determine the efficient frontier. The portfolio vari-
ance associated with the expected portfolio return is
P2 (w ) = wT V w = wT V (1, V 1 e + 2, V 1 )
= 1, (w T e) + 2, (w T )
= 1, + 2,
since wT e = 1 and wT = . Using (3.71), we get
A 2 2 B + C
P2 () = . (3.73)
A C B2
Equation (3.73) has the exact form as the portfolio variance (3.40) for two se-
curities (page 109). Indeed, the methodology for the two-security case in Sec-
tion 3.2.2 carries over almost verbatim to determining the N-security efficient
frontier. Equation (3.73) is equivalent to the right branch of the following hy-
perbola in the (P , P )-plane:
2
2P P B
A
= 1. (3.74)
1 AC B2
A A2
For our portfolios, we have P = P () and P = and consider the right

branch since P () > 0. The turning point of the right branch of the hyperbola
yields the global minimum-variance portfolio, which has risk G , expected return
rate G , and weight vector w G :

1 B V 1 e
(G , G ) = , , wG = . (3.75)
A A A
Since investors are not interested in a portfolio with negative expected return
rate, we assume G > 0, which yields B > 0.
The portion of the branch of the hyperbola from the turning point to along
the upper portion of the branch then forms the efficient frontier, which we call
the Markowitz efficient frontier for N securities
-
A2P 2BP + C B
ME,N = (P , P ) : P = , P
AC B2 A
. /
2P (P ( B/A))2 B
= ( P , P ) : = 1, P > 0, P .
(1/A) (( AC B2 )/A2 ) A
(3.76)
The set of portfolio weight vectors that produce the efficient frontier ME,N is:

C P B 1 P A B 1 B
WE,N = w : w = V e+ V , P .
AC B2 AC B2 A
(3.77)
Equations (3.76) and (3.77) coincide, respectively, with (3.48) and (3.49) (see
page 111) when N = 2.
Finally, since not all portfolios are efficient, we consider how the efficient
frontier sits in the set of feasible portfolios, i.e., the collection of all possible pairs
of portfolio risk and expected return rate. The discussion will highlight a dif-
ference between the N-security portfolio case with N 3 and the two-security
case. Introduce a mapping fP,N from the weight space WN into the (P , P )-
plane by
fP,N (w) = (P (w ), P (w)).
The set of feasible portfolios with no short selling is defined as the range of fP,N :
$ %
FP,N = fP,N [WN ] = (P (w ), P (w )) : w T e = 1 .
For two securities, the weight space W2 is a line and the transformation fP,2
maps it into the right branch of a hyperbola. In fact, by (3.37) and (3.40) (see
pages 108 and 109), we can express w and P (w) as functions of the expected
portfolio return rate :
-
1 2 A2 2B + C
P (w ) = , w= , P (w) = .
1 2 1 AC B2
Employing the upright variables P and P defined by
P = P (w ) = , P = P (w)
and using Equation (3.42) on page 109, we see that the feasible set FP,N is the
right branch of a hyperbola:

"
FP,2 = (P , P ) : P = ( AP 2BP + C )/( AC B ), < P <
2 2
. /
2P (P ( B/A))2
= ( P , P ) : = 1, P > 0, < P < .
(1/A) (( AC B2 )/A2 )
The efficient frontier MP,2 is the top half, including the turning point, of the
curve FP,2 , i.e., the solid curve in Figure 3.1 on page 110.
For N 3, the set of feasible portfolios is different. For each expected port-
folio return rate P (w) = P , an efficient portfolio has the least risk, which by
(3.76) is -
A2P 2BP + C
P (w ) = , ( N 3).
AC B2
In other words, a portfolio with P (w) = P is inefficient if and only if it has
risk greater than the above amount. The set of feasible portfolios, i.e., the locus
of all efficient and inefficient portfolios, is then given as follows for N 3:
-
2BP + C
A2P
FP,N 3 = (P , P ) : P , < P <
AC B2
. /
2P (P ( B/A))2
= (P , P ) : 1, P > 0, < P < .
(1/A) (( AC B2 ) /A2 )
(3.78)
Consequently, the feasible set FP,N , where N 3, forms an infinite (unb-

ounded) region whose boundary is the right branch of a hyperbola. In this
case, we call FP,N a Markowitz bullet. The efficient frontier ME,N is then a proper
subset of FP,N and forms the upper half of the feasible sets boundary curve,
including the turning point. Consequently, any portfolio in the feasible region that
is not in ME,N is inefficient.
Figure 3.3 illustrates the feasible region FP,3 and its efficient frontier.
3.4 N-Security Efficient Frontier Without Short Selling
By default, we assume that short selling is allowed with restriction. In this

section, we briefly treat the case of no short selling. Unfortunately, in the latter
case the study of the efficient frontier is much more complex mathematically
and so is usually treated numerically using quadratic programming software.
For our purpose, since short selling will be part of most models considered,
we shall simply illustrate the no-short-selling case graphically. Interestingly,
when short selling is forbidden, many of the efficient portfolios have several
vanishing weights (e.g., Luenberger [16, p. 161]).
When there is no short selling, the space of weights for N securities is
$ %

WN = w : w T e = 1, wi 0 . (3.79)
The space WN is a standard ( N 1)-simplex in R N . Figure 3.4 depicts the
weight spaces for portfolios with two and three securities. The two-security
case is a line segment, while the three-security case is an equilateral triangle.
3.4 N-Security Efficient Frontier Without Short Selling 127
0.30
0.25
0.20
0.15
0.10
0.05
0.16 0.18 0.20 0.22 0.24 0.26 0.28 0.30
Fig. 3.3 The feasible region FP,3 for three securities with unlimited short selling. The outer boundary
curve is a branch of a hyperbola. The efficient frontier M E,3 is the portion of the hyperbola starting
from the turning point and going along the upper segment of the hyperbola
Note that WN is the intersection of the weight space W , where short sell-
N
ing is allowed, with the positive orthant {w : wi 0, i = 1, . . . , N } in R N . For
example, in Figure 3.4 the line segment and equilateral-triangle weight spaces
arise, respectively, from the intersection of the straight line W1 with the first
quadrant and from the plane with the first octant.
Fig. 3.4 The weight spaces for two-security (left) and three-security (right) portfolios

Define a mapping fP,N into the ( , )-plane
from the weight space WN P P
R by
2

fP,N (w ) = (P (w), P (w)), w WN .

The range of fP,N
is defined to be the set FP,N of feasible portfolios with no
short selling:

fP,N [WN ] = FP,N

.
For a two-security portfolio, the no-short-selling feasible set FP,2 is a segment
of the right branch of a hyperbola. The end points of the segment correspond
to the risk-mean points (1 , 1 ) and (2 , 2 ) due to the individual securities.
Figure 3.1 on page 110 gives an illustration.
For a three-security portfolio, the feasible set FP,3 forms a region. The left
panel in Figure 3.5 depicts an example. The three cusp points in the figure
correspond to the pair of risk and expected return rate due to the securities,
namely, (1 , 1 ), (2 , 2 ), and (3 , 3 ). As in the short selling case, the turning
point (shown as ) designates the portfolio with the global minimum risk and
expected return rate. We can then identify the efficient frontier as the outer
boundary curve segment from the turning point to the uppermost cusp point.
Each bold curve joins a pair of cusp points corresponding to two securities, so
each point on such a curve is a two-security portfolio.
The right panel in Figure 3.5 shows how the three-security feasible set with
no short selling fits inside the feasible set in Figure 3.3 with short selling. The
two figures are generated using the same three-security inputs, except that the
weights are nonnegative in the case of no short selling. The efficient frontier
for short selling in Figure 3.3 extends higher up than its analog for no short
selling in the left panel of Figure 3.5.
Remark 3.5. Portfolio theory with no short selling and with bounds on the
portfolio expectation and portfolio variance is usually treated with optimiza-
tion software. For a mathematical treatment, see, for example, Korn and Korn
[13, Chap. 1] and references therein.

3.5 The Mutual Fund Theorem
3.5.1 The Global Minimum-Variance Portfolio
Let us recall how to characterize an efficient portfolio when unlimited short

selling is allowed. By (3.77) on page 125, an efficient portfolio has a weight
vector w and expected return satisfying

C B 1 A B B
w = V e+ V 1 , G = . (3.80)
AC B 2 AC B 2 A

1, 2,
3.5 The Mutual Fund Theorem 129
0.30
0.25
0.20
0.15
0.10
0.05
0.16 0.18 0.20 0.22 0.24 0.26 0.28 0.30
of feasible portfolios for three securities without short selling (shaded region).
Fig. 3.5 Left: set FP,3
We use the same input parameters as in Figure 3.3, except the weights are restricted to being nonneg-
ative. Each cusp point corresponds to a pair of risk and expected return rate due to one of the three
securities. The efficient frontier is the outer curve segment from the turning point to the uppermost
cusp point. The three solid curves joining pairs of cusp points consist of two-security portfolios con-
taining only the two joined securities (cusp points). Right: superposition of the feasible sets for no
short selling (left) and short selling (Figure 3.3). The efficient frontier for the short selling example
extends higher up than the efficient frontier with no short selling
The global minimum-variance portfolio in (3.75) on page 124 arises when the
second Lagrange multiplier 2, vanishes:
V 1 e
2, = 0 = G w = w G = .
A
To see
V 1 e
w = = = G ,
A
note that since

V 1 e 1
w = 1, V 1 e + 2, V 1 = 0
A A
and because V 1 e and V 1 are linear independent (due e and being linearly
independent), we get
1
1, = , 2, = 0,
A
which yields = G .
3.5.2 The Diversified Portfolio
When the first Lagrange multiplier 1, vanishes, the corresponding portfolio

is called the diversified portfolio, and its weight vector, expected return rate, and
risk are denoted by w D , D , and D , respectively. Similar to the above, we find
C V 1
1, = 0 = D = w = wD = .
B B
However, we must check one more property, i.e., to conclude that the diversi-
fied portfolio is efficient, Equation (3.80) shows that we still need to establish
= D G .
Since
C B AC B2
D G = = ,
B A AB
we have
AC B2
D = G + .
AB
But AB > 0 and AC B2 > 0. Consequently
D > G .
Hence, the diversified portfolio is at a point (D , D ) on the efficient frontier
MP,N that is higher up than the global minimum-variance portfolio (G , G ).
The variance of the diversified portfolio is
C
P2 (w D ) = .
B2
Because we assume that G > 0 and so B > 0, the risk of the diversified port-
folio is

C
P (w D ) = D .
B
Note that because G is the smallest possible variance for the N securities, we
have
D > G .
3.5.3 The Mutual Fund Theorem
It will now readily follow that every minimum-variance portfolio, i.e., a port-
folio with weight vector that solves Problem II, can be expressed as an app-
ropriate combination of the global minimum-variance portfolio w G and the
diversified portfolio w D . In fact, for a minimum-variance portfolio, we have
3.6 Investor Utility Function 131
w = 1, (V 1 e) + 2, (V 1 ) = (1, A)w G + (2, B)w D ,
where by (3.68) on page 123, the coefficients sum to unity:
1, A + 2, B = 1.
Equation (3.71) (page 123) yields
A (C B) B (A B)
a 1, A = , 1 a = 2, B = .
AC B2 AC B2
Hence:
w = a w G + (1 a ) w D . (3.81)
The result (3.81) is called a Mutual Fund Theorem or Separation Theorem. It
states that an N-security minimum-variance portfolio with expected return has the
same risk-mean as the two-security portfolio with the percentage a invested in port-
folio w G and percentage 1 a invested in portfolio w D . Proxies for the portfolios
w G and w D could be mutual funds. The idea would then be to purchase two
such mutual funds in the proportions a and 1 a to create a two-security
portfolio that replicates the risk-reward profile of the N-security portfolio.
In general, there are many minimum-variance portfolio separations
available:
Theorem 3.1. (Mutual Fund Theorem) Any minimum-variance portfolio w can

be expressed in terms of any two distinct minimum-variance portfolios:
w = s1 w a + s2 w b ,
where w a = wb (distinct portfolios) and the scalars si obey s1 + s2 = 1.
Readers are referred to Ingersoll [10] for more on the Mutual Fund Theorem.
3.6 Investor Utility Function
3.6.1 Utility Functions and Expected Utility Maximization
A utility function is sometimes informally called a happiness function. Intu-

itively speaking, it is a one-to-one function that represents the level of happi-
ness of the investor as a function of return (or wealth). The investor applies her
utility function to all investments and chooses the investment that maximizes
her expected utility. We shall also see in Section 3.6.2 that the shape of the
graph of an investors utility function carries information about the investors
risk tolerance.
An investors utility function, denoted u( x ), is assumed to be deterministic.

It becomes a random variable u( X ) after substituting in a random variable
X representing a random future return rate (or random future wealth). The
possible values of X are denoted by x. In particular, if an investment, such as
a portfolio, has an array of possible return rates x1 , x2 , . . . xn , we assume that
an investor will assign numerical values u( x1 ), u( x2 ), . . . u( xn ) that indicate her
preferences for those opportunities based on her satisfaction with the possible
returns. The actual numerical values u( xi ), where i = 1, . . . , n, are inessential in
the sense that it is how two investment choices are ranked relative to each other
that will matter. This brings us to the optimization of an investors expected
utility.
We assume that investors maximize their expected utility when choosing portfo-
lios. In other words, when faced with a choice between two investments, an
investor will choose the one with higher expected utility. For instance, if the
investor owns security A and discovers that A has less expected utility than
that of security B, then the investor will sell A and buy B. The investor will
favor the security that is expected to bring higher satisfaction. In general, since
we assume that an investor always maximizes her expected utility when dis-
tributing her wealth across different investment choices, there is no trade that
can further increase her expected utility. Hence, if an investor chooses invest-
ments with returns X and Y, then the investors expected utilities for these
choices are the same:

E u ( X ) = E u (Y ) . (3.82)
An optimal portfolio for an investor is the one for which the investors expected
utility is maximized.
The derivative u ( x ) is called the investors marginal utility function. We shall
assume that utility functions are analytic (i.e., they can be Taylor expanded) and have
positive marginal utility, i.e., u is strictly increasing:
x2 > x1 u( x2 ) > u( x1 ). (3.83)
In other words, a positive incremental change x2 x1 > 0 in returns implies a

positive change u( x2 ) u( x1 ) > 0 in utility and vice versa. This underscores
that a higher return (x2 > x1 ) leads to higher satisfaction (u( x2 ) > u( x1 )) and,
hence, requires a higher degree of risk tolerance for the potential higher return.
There is no free ride: a higher return requires higher risk.
Taylor expanding the utility function about the expectation E ( X ), we get
1 2
u( X ) = u(E ( X )) + u (E ( X )) X E ( X ) + u (E ( X )) X E ( X )
2!
1 (3 ) 3
+ u (E ( X )) X E ( X ) + ,
3!
where u(k) ( x ) is the kth derivative of u. Since u(E ( X )) and u(k) (E ( X )) are
constants and because E ( X E ( X )) = 0, it follows
1
E u( X ) = u(E ( X )) + u (E ( X )) Var( X )
2!
1 3
+ u(3) (E ( X )) E X E ( X ) + .
3!
A key assumption of Markowitz portfolio theory is that investors assess an
investment only through the expectation and variance of its return. Unless
otherwise stated, let us assume that investors evaluate an investment exclusively in
terms of its expected return E ( X ) and variance Var( X ) and, for simplicity, treat all
higher-order derivatives of the utility function as zero, even if those terms in the Taylor
expansion are a function of E ( X ) and Var( X ):
u(k) (E ( X )) = 0 for k = 3, 4, . . . .
Our simplifying assumption implies

1
E u( X ) = u(E ( X )) + u (E ( X )) Var( X ). (3.84)
2
Examples of utility functions satisfying (3.84) are (Exercise 3.32)

u( x ) = ax + b a>0
b a
u( x ) = ax x2 b = 0, x < .
2 b
3.6.2 Risk-Averse, Risk-Neutral, and Risk-Seeking Investors
Investors can be divided into three broad categories. To illustrate, let 0 be the
current time and suppose that an investor is considering a choice of one of the
following portfolios:
Portfolio A: riskless with a 100% guaranteed return rate of RtA = 10% from
the present time 0 to a time t in the future. This sure investment has an
expected utility given by
E (u( RtA )) = E (u(10%)) = u(10%).
Portfolio B: risky with a random return rate from 0 to a future time t

given by

5% with probability of 50%

Rt =
B
15% with probability of 50%,
which has an expected return E ( RtB ) = 10% identical to that of Portfolio A.
Fig. 3.6 Depiction of three different utility functions, where x represents possible return rates on an
investment and u( x ) assigns a number designating the investors utility for the given possible return
rate. Lower, middle, and upper curves show graphs of the utility functions of a risk-averse, risk-
neutral, and risk-seeking investor, respectively
We now consider choosing between these two portfolios through the eyes of
risk-averse, risk-seeking, and risk-neutral investors.
Risk-Averse Investors
A risk-averse investor is one with a utility function satisfying u ( x ) > 0 and

u ( x ) < 0. In other words, the utility function is strictly increasing and strictly
concave, i.e., it has a decreasing slope; see the lower curve in Figure 3.6.
How would a risk-averse investor evaluate Portfolio B? The increasing,
concave shape of the graph shows that the magnitude of the decrease of the
investors utility (happiness) due to a 5% drop in return from 10% to 5% is
greater than the increase in utility from a 5% increase in return from 10% to
15%. Such an investor is affected more by a 5% drop in wealth than a 5% inc-
rease in wealth and so she assigns a stiffer penalty to losses. This implies that
a risk-averse investor assigns a lower expected utility to Portfolio B than to
portfolio A:14
1 1
E (u( RtA )) = u( RtA ) = u E ( RtB ) > u(0.05) + u(0.15) = E (u( RtB )).
2 2
In other words, a risk-averse investor will prefer Portfolio A over B. Intuitively,
for the investor to be just as happy with risky Portfolio B as she is with riskless
14 A strictly concave function f on an interval I satisfies f ( x a + (1 x ) b ) > x f ( a) + (1 x ) f (b) for all

0 < x < 1, and a, b in I. In the example, the strict inequality follows using f = u with x a + (1 x ) b =

E ( R B ), where x = P R B = 0.05 , 1 x = P R B = 0.15 , a = 0.05, and b = 0.15.
Portfolio A, the expected return of Portfolio B must exceed that of Portfolio A,

which will be shown below. The extra return beyond the 10% of Portfolio A is
the risk premium required by a risk-averse investor for bearing risk.
Risk premiums show up more generally for risk-averse investors. Let R f be
the return rate across [0, t], where 0 is the current time, of a risk-free security
that continuously compounds at the risk-free rate r. Then
R f = er t 1. (3.85)
Note that since R f is a constant, the value u( R f ) is not random. Let RS be

the random return rate on [0, t] of a risky security that pays a continuous cash
dividend at yield rate q, which is continuously reinvested to buy more units of
the security. Then the cum-dividend price of the risky security at t is given by
Stc = eq t St .15 The return rate becomes
Stc
RS = 1 Stc = eq t St , (3.86)
S0
where S0 is the (known) current price and St is the random market price at the
future time t. Since u( R f ) is a constant, Equation (3.82) implies

u( R f ) = E u( R f ) = E u( RS ) .
Employing the assumption that investors evaluate their expected utility of

a security only through its expectation and variance and assuming that they
treat all higher moments as zero, Equation (3.84) yields
1
E u( RS ) = u E ( RS ) + u E ( RS ) Var( RS ). (3.87)
2
Because u ( x ) < 0 for a risk-averse investor and Var( RS ) > 0 (since the secu-
rity is risky), it follows
1
u( R f ) = u E ( RS ) + u E ( RS ) Var( RS ) < u E ( RS ) .
2
By (3.83), we see that on [0, t] the expected return rate of the risky secu-
rity is higher than the risk-free return rate: R f < E ( RS ). The positive quan-
tity E ( RS ) R f is the risk premium required for a risk-averse investor to be
equally happy or indifferent between the risky investment and the risk-free
E(Sc )
one. Using (3.85) and (3.86), we also get S0t > er t . In other words, a risk-
averse investor chooses risky securities with
E ( S t ) > S0 e ( r q ) t . (3.88)
Most investors are risk averse.

15 See (2.28) on page 31.
Finally, the reader may wonder how our current definition of a risk-averse
investor, i.e., one with utility function where u ( x ) > 0 and u ( x ) < 0, relates to
our original definition given in the Markowitz theory. Recall that an investor
is termed risk averse in the Markowitz model if, for a portfolio with a given
level of risk, the investor requires the maximum expected return and if, for a
portfolio with a given expected return, the investor requires the minimum risk.
To see the link with utility functions, for a portfolio return rate R P , Equation
(3.84) yields
1
E u( R P ) = u E ( R P ) + u E ( R P ) Var( R P ).
2

Since u E ( R P ) < 0, we see that maximizing the expected utility E u( R P )
given a fixed expected return E ( R P ) implies that the variance Var( R P ) of the
portfolio is minimized. Conversely, if we are given a fixed variance Var( R P ),

then maximizing E u( R P ) implies u E ( R P ) is maximized, which yields
that the expected portfolio return E ( R P ) is also maximized (since u is strictly
increasing).
Risk-Seeking Investor
A risk-seeking investor is one whose utility function satisfies u ( x ) > 0 and

u ( x ) > 0. The graph is an increasing function with an increasing slope; see
the upper curve in Figure 3.6. The strict convexity of the function shows that
the increase in the investors utility due to a 5% rise in return from 10% to 15%
is greater than the decrease in utility from a 5% drop in return from 10% to
5%. Such investors are less impacted from a 5% drop in wealth than from a 5%
increase in wealth. They gravitate toward risk. A risk-seeking investor then
assigns a higher expected utility to Portfolio B than to A:16
1 1
E (u( RtA )) = u( RtA ) = u E ( RtB ) < u(0.05) + u(0.15) = E (u( R B )).
2 2
A risk-seeking investor will select Portfolio B.
Analogous to the risk-averse case, given a risk-free security and risky secu-
rity with returns R f and RS , respectively, and since u ( x ) > 0 for a risk-seeking
investor, we obtain
1
u( R f ) = u E ( RS ) + u E ( RS ) Var( RS ) > u E ( RS ) .
2
16 A strictly convex function f on an interval I satisfies f ( x a + (1 x ) b ) < x f ( a) + (1 x ) f ( b ) for all

0 < x < 1 and a, b in I. In the example, the strict inequality follows using f = u with x a + (1 x ) b =

E ( R B ), where x = P R B = 0.05 , 1 x = P R B = 0.15 , a = 0.05, and b = 0.15. Also, see Jensens
inequality.
Equation (3.83) yields R f > E ( RS ), i.e., a risk-seeking investor has a negative

E (Stc )
risk premium E ( RS ) R f . By (3.85) and (3.86), we obtain er t > S0 . Hence, a
risk-seeking investor prefers risky securities with
E ( S t ) < S0 e ( r q ) t . (3.89)
Risk-Neutral Investors
A risk-neutral investor is one with a utility function satisfying u ( x ) > 0 and

u ( x ) = 0. This means u( x ) = a x + b, where a > 0. The graph is a straight
line with positive slope, which is shown in Figure 3.6 for the case u( x ) = a x,
where a > 0. The magnitude of the decrease of the investors utility due to a
5% drop in return from 10% to 5% is the same as the increase in utility from
a 5% increase in return from 10% to 15%. Such an investor is equally affected
by a 5% drop in wealth compared with a 5% increase in wealth. A risk-neutral
investor is then neutral toward or indifferent between Portfolios A and B:

E u( RtA ) = E u( RtB ) .
Considering the risk-free security with return R f and risky one with return
RS , we have
1
u( R f ) = u E ( RS ) + u E ( RS ) Var( RS ) = u E ( RS ) .
2
E (Stc )
Since u strictly increases, we must have er t = S0 . Thus, a risk-neutral investor
prefers risky securities with
E ( S t ) = S0 e ( r q ) t . (3.90)
Remark 3.6.
1. Though investors tend to be risk averse versus risk seeking, they are un-
likely to be risk neutral. Despite this, risk-neutral investors will play an im-
portant role in the pricing of derivatives (Chapter 8).
2. Portfolio theory can also be presented starting from the theory of utility
functions; see Pennacchi [21, Chap. 2] for more. This approach, though,
would take us far deeper into utility functions than is appropriate for this
introductory text.

We now return to our original assumption that all investors are risk averse.
3.7 Diversification and Randomly Selected Securities
Diversification is a risk management technique that mixes a variety of assets in

a portfolio. Its effect is to reduce portfolio risk by distributing a given invest-
ment capital across more and more securities to spread the risk around. We
first explored diversification in Section 3.2.3 (on page 116) for a two-security
portfolio. We showed how a portfolios risk can be reduced by spreading an
initial capital in a certain way across two uncorrelated securities as opposed
to investing all of the capital in one of the securities. This section explores
how the mean portfolio risk is impacted when we create diversification by
randomly and uniformly selecting more and more securities and weights. We
shall see that the Markowitz portfolio model provides a mathematical basis for
the statement, Dont put all your eggs in one basket.
Consider a market of all possible risky securities. By default we focus on sec-
urities in the USA and one may choose say, the NASDAQ, as a proxy. To create
an N-security portfolio, we pick a random pair of N risky securities and N
weights. Specifically, we simultaneously pick N random securities uniformly
from across the market and uniformly choose N nonnegative random weights
for these securities. For each such random pair, there is a portfolio variance,
and it is the mean of this variance for many random pairs that we shall explore
as N increases.
It is important to emphasize that since we shall randomly draw securities from the
marketplace and randomly assign weights, the resulting portfolios will, in general, not
be efficient. Also, as N we assume that all the securities in the marketplace are
included in the portfolio.
We now model these ideas mathematically under the assumption of no short
selling.
3.7.1 Mean Portfolio Variance and the Uniform Dirichlet Distribution
We determine a formula for the expectation of the portfolio variance of N

random securities.
First, consider an N N random covariance matrix V resulting from N ran-
domly and uniformly selected securities, where N 2. Since V is symmetric,
fix an ordering of the entries of the diagonal and upper triangular portion of V
as follows:
(11 , 12 , . . . , 1N , 22 , 23 , . . . , 2N , 33 , 34 , . . . , 3N , . . . , N N ),
N ( N 1 )
where ii = i2 . Denote the joint p.d.f. of these N + 2 entries by fV and the
marginal p.d.f.s of the entries ij by f ij , respectively. Expectations with respect
3.7 Diversification and Randomly Selected Securities 139
to fV and f ij are denoted by EV and E ij , respectively. We shall set E ii = E i . The

expectation of the sample mean of the random variances i2 is

1 N 2 1 N
Var ( N ) E w
N i =1
i =
N i
E i (i2 ), (3.91)
=1
which is the average of the expected individual security variances. The expec-
tation of the sample mean of the covariances ij , where i = j, is

1 1
Cov ( N ) EV N ( N 1) ij = N ( N 1) Eij [ij ] ,
2 1 i < j N 2 1 i < j N
(3.92)
which is the average of the expected individual security covariances. Assume
that as N , the quantities Var ( N ) and Cov ( N ) converge to finite values, which
we denote simply as Var and Cov , respectively.
Second, uniformly and randomly choose weight vectors w from the weight
space WN , which excludes short selling:
$ %

WN = ( x1 , . . . , x N 1, x N ) R N : xi 0, x1 + + x N 1 + x N = 1 .
Since WN is a standard ( N 1)-simplex, the uniform Dirichlet density is used to
choose randomly the elements of WN . The joint Dirichlet p.d.f of
w = [ w 1 . . . w N 1 w N ] T ,
where wi 0 and w N = 1 (w1 + + w N 1 ), is then given relative to the first

N 1 weights:

( N 1) ! ( x1 , . . . , x N ) WN
f w ( x1 , . . . , x N ) =

0 ( x1 , . . . , x N ) WN ,
where xi is the variable representing the possible values of wi for i = 1, . . . , N.

See, for example, Fine [4, Secs 7.4.3, 7.7] and Frigyik, Kapila, and Gupta [5] for
introductions to the Dirichlet distribution.
Figure 3.7 illustrates the uniform random selection from the weight spaces
for two and three securities. The associated marginal density of each weight is
identical and given by
f w ( xi ) = ( N 1)(1 xi ) N 2 , i = 1, . . . , N ( N 2).
Expectations, variances, and covariances relative to the Dirichlet density (joint

or marginal) will be indicated with w as a subscript. The following basic prop-
erties hold, where i, j = 1, . . . , N and N 2:
0.8
1
0.6
0.8
w3
0.4
0.6
0.2
w2
0.4 0
0 0.2
0 0.4
0.2 w1
0.6
0.5
0.8
w2
0 1 1
0 0.2 0.4 0.6 0.8 1
w1
Fig. 3.7 Randomly and uniformly chosen weights from the weight spaces with no short selling for
two securities (left) and three securities (right). The weights were drawn using a uniform Dirichlet
distribution
1
E w (wi ) =
N
N1
Varw (wi ) =
N2 ( N
+ 1)
1
Covw (wi , w j ) = 2 , i = j.
N ( N + 1)
Moreover:
2
E w (w2i ) = Varw (wi ) + (E w (wi ))2 = (3.93)
N ( N + 1)
and
1
E w (wi w j ) = Covw (wi , w j ) + E w (wi ) E w (w j ) = (i = j), (3.94)
N ( N + 1)
where i, j = 1, . . . , N and N 2.
Third, let the pair (w, V ) represent the following string of random variables:
(w1 , . . . , w N , 11 , 12 , . . . , 1N , 22 , 23 , . . . , 2N , 33 , 34 , . . . , 3N , . . . , N N ).
Denote the joint p.d.f. of (w, V ) by f (w,V ) and expectations relative to f w,V
by E (w,V ) . With respect to the degree of independence between w and V , we
assume:
E w (wi w j ij ) = E w (wi w j ) E ij (ij ). (3.95)
Now, turning to the portfolio variance, we have the random variable:
N
2
P,N = wT V w = w2i i2 + 2 wi w j ij .
i =1 1 i < j N
3.7 Diversification and Randomly Selected Securities 141
Using (3.91)(3.95), we compute the expectation of the variance:

N
2
E (w,V ) P,N = E (w,V ) w2i i2 + 2 E (w,V ) wi w j ij
i =1 1 i < j N
N
= E w (w2i ) E i (i2 ) + 2 E w (wi w j ) E ij (ij )
i =1 1 i < j N
2 N E i (i2 ) N1 E ij (ij )
=
N+1 N
+
N+1 ( N ( N 1)/2)
.
i =1 1 i < j N
Consequently:
1 1
2
E (w,V ) P,N = 2 Var ( N ) Cov ( N ) + Cov ( N )
N+1 1+ 1
N
(3.96)
Since Cov ( N ) Cov and Var ( N ) Var as N , Equation (3.96) yields

2
E (w,V ) P,N Cov as N . (3.97)
Hence, the limiting value of the mean portfolio variance as N increases shows that
the mean sample variance of the individual securities return rates is dominated by the
mean sample covariance of these returns. That is, for a sufficiently large number
of securities, the covariances between securities have a greater impact on a
typical portfolios variance than the variances of the individual securities. The
next section illustrates these ideas using data from the NASDAQ.
Remark 3.7.
1. The above is consistent with our findings in Section 3.2.3 for two securities
with no short selling. By decreasing the correlation coefficient between
the two securities from 1 toward 1, we increased the diversification and
reduced the portfolios risk. See Figure 3.2 on page 115.
2. For the case when the weights are nonrandom and equal, namely,
1
wi = , i = 1, . . . , N,
N
but the securities are randomly chosen (i.e., the covariance matrix is ran-
dom), the expected portfolio variance becomes
1
2
E (w,V ) P,N = (Var ( N ) Cov ( N )) + Cov ( N ).
N
By increasing the number of securities indefinitely, we obtain the same result

as in (3.97):

2
E (w,V ) P,N Cov as N .

3.7.2 Mean Portfolio Variance using the NASDAQ
Example 3.8. We estimated the mean portfolio variance for 2, 385 NASDAQ
stocks using 503 adjusted closing daily prices for each stock over the time
span from March 15, 2011, to March 15, 2013. The data was obtained from fi-
nance.yahoo.com. Figure 3.8 depicts the results generated by 100, 000 pairs of
random stock picksand random weights for each N. The portfolio variance for
60
real data
theoretical
50
Mean Portfolio Variance
40
30
20
10
0
0 10 20 30 40 50
Number of Stocks in the Portfolio
Fig. 3.8 Mean portfolio variance as a function of the number of randomly chosen stocks on the NAS-
DAQ (see text). The theoretical line is the mean portfolio variance based on Equation (3.96), which
employs a uniform Dirichlet distribution of weights with no short selling. The real data line is based
on random draws from the NASDAQ. Courtesy of Li Li
all 100, 000 pairs was computed and averaged to estimate the mean portfolio
variance for the given N. The figure also depicts the theoretical mean portfolio
variance determined using the uniform Dirichlet distribution.

Equation (3.97) and Figure 3.8 illustrate that even after diversification by ran-
domly and uniformly choosing a large number of securities across the marketplace
and across weights, portfolio risk arising from the mean sample covariance remains.
3.8 Exercises 143
This type of risk, i.e., risk that cannot be eliminated by diversification, is called
undiversifiable risk or systematic risk. It will play an important role in Chapter 4.
The portion of portfolio risk that can be removed by diversification, namely,
the risk that contributes to the mean portfolio variance values above the con-
vergence value in Figure 3.8, is called unsystematic risk or diversifiable risk.
Unless stated to the contrary, assume that a portfolios risk is sys-

tematic, i.e., the portfolio is sufficiently diversified that all other
risks are negligible.
Remark 3.8. Readers interested in learning more about Markowitz portfolio

theory should consult the book Risk-Return Analysis: The Theory and Practice of
Rational Investing by Markowitz and Blay[19].
3.8 Exercises
3.1. An investor plans to create a portfolio of ten stocks by shorting all of them.
Can he use the Markowitz theory presented in this chapter? Explain your
answer.
3.2. Can you find five examples of pairs of stocks in the USA that are negatively
correlated? Are such occurrences common?
3.3. The volatility of the log-return rate of a portfolio over 120 days is roughly
11 times the volatility over a day. Agree or disagree? Explain your answer.
3.4. Decide whether you agree or disagree with the statements below. Justify
your answer.
a) Investors who are not risk averse are irrational.
b) When the risk of a portfolio vanishes, the risk of each security has to
vanish.
3.5. A clever wealth manager constructed a portfolio of stocks such that the
portfolio has no risk and has an expected return of 25%. What is the probability
that the portfolio return rate will actually be 25%?
3.6. Show that Problem I on page 120 is equivalent to Problem II on page 121,
i.e., show that these two optimization problems have the same set of solutions
for all c > 0.
P2 (w)
3.7. Explain the financial meaning of minimizing the function f P (w ) = P ( w)
,
where P (w) is the portfolio expected return.
3.8. It can be shown that the covariance of the return of the global minimum-
variance N-security portfolio with the return of any other efficient N-security
portfolio is always 1/A (Exercise 3.28). Interpret this result.
3.9. If your initial capital increases by $100, then we would expect an increase
in your utility. To which scenario would you assign a higher utility, assuming
the same risk for both? Scenario A: initial capital of $1, 000. Scenario B: initial
capital of $10, 000.
3.10. Is u( x ) = a a utility function for a a constant? Justify your answer.
3.11. Is u( x ) = 1 eb x , where b > 0, a risk-averse, risk-neutral, or risk-seeking

utility function? Justify your answer.
3.12. (Two Securities) The table below gives a sample of artificial historical
data for an energy company (Stock A) and a phone company (Stock B). Each
indicated return rate is end of month to end of month and is based on adjusted
A
closing prices. For example, RJan 2012 is the return rate from the last trading
day in December 2011 to the last trading day in January 2012. Using the table,
express your answers in percentages where appropriate.
3.8 Exercises 145
Date A
Rmonthly B
Rmonthly
Jan-2012 2.45% 2.69%
Feb-2012 3.35% 1.81%
Mar-2012 3.24% 4.94%
Apr-2012 2.93% 5.88%
May-2012 6.13% 2.51%
Jun-2012 6.19% -0.35%
Jul-2012 0.78% 1.59%
Aug-2012 -0.19% -3.83%
Sep-2012 4.65% 5.24%
Oct-2012 3.53% 4.85%
Nov-2012 5.03% 2.48%
Dec-2012 -1.71% 4.03%
a) Sketch the graph of the monthly total return rates of each stock as a function
of time during 2012. Briefly discuss the movement of the stocks during two
equal-length-time periods in 2012.
b) Estimate the expected monthly total return rate of each stock in the table.
What is your answer if you use data only from Dec-2011 to Jun-2012? Com-
pute the monthly volatility of each stock for the year 2012.
c) Estimate the monthly variance of each stock and determine the covariance
and correlation coefficient between the monthly total return rates for the two
stocks during 2012. Is the result what you expected? Briefly discuss.
d) For a portfolio consisting of stocks A and B, use the data in the table to det-
ermine the portfolios expected monthly total return rate and the portfolios
monthly risk under selection I, where funds are split evenly between the
two stocks, and selection II, where two-thirds of your funds are in stock A
and one-third in stock B. Annualize the portfolio expected returns and risks.
e) Would you recommend portfolio selection I or II to an investor? Briefly dis-
cuss your answer.
f) Critique how the data in the table is being applied to the theoretical frame-
work used in these exercises. Include two drawbacks with using historical
data to estimate the expected returns and risks.
3.13. (Three Securities) Suppose that you have $5,000 to invest in stocks 1, 2,

S1 ( t 0 ) $10.20
and 3 with current prices S2 (t0 ) = $53.75 , covariance matrix
S3 ( t 0 ) $30.45

0.03 0.04 0.02
V = 0.04 0.08 0.04 ,
0.02 0.04 0.04
and expected return vector

0.10
= 0.15 .
0.075
For example, stock 3 has a volatility of 3 = 20% and expected return rate of
3 = 7.5%. The values in V and are pure numbers (not percentages). Answer
the following using an appropriate software.
a) Determine the weights needed to create the global minimum-variance port-
folio of these three stocks.
b) Create an efficient portfolio with an expected return rate of 18%. Explicitly
state the number of shares one must hold for each stock and how you fund
each position. State the portfolio risk and compare it with maximum risk
among the stocks.
3.14. (N Securities) Consider managing a portfolio with 500 risky securities,

and assume that the variances of the securities are robustly estimated from
reliable historical data. If 1% of the remaining independent covariances of the
securities are poorly estimated due to inaccuracies in the data, then determine
the number of poorly estimated covariances.
3.15. Suppose a client just inherited $1, 000, 000 and has come to you seeking
advice on how to split the money between two of his favorite securities so as to
maximize return. Security A has expected rate of return r A = 0.13 and standard
deviation of A = 0.15. Security B has expected rate of return r B = 0.14 and
standard deviation B = 0.20. The correlation coefficient between their rates of

return is = 0.3. If the investor has a utility function U ( x ) = 3 x, how should
he invest in each stock to maximize his overall rate of return?
3.16. Let X and Y be two random variables. Let ai , i = 1, 2, 3, 4 be real numbers

with a2 a4 = 0. Let be the correlation coefficient. Prove that
.
( X, Y ) if a2 a4 > 0
( a1 + a2 X, a3 + a4 Y ) =
( X, Y ) if a2 a4 < 0.
The significance of the result is to allow a change of scale to one convenient for
computation.
3.8 Exercises 147
3.17. Prove (3.17) on page 97.
3.18. Verify Equation (3.20) on page 103, where the portfolio log-return rates
are assumed uncorrelated and identically distributed.
3.19. (Two Securities) Let A = e T V 1 e, B = T V 1 e, and C = T V 1 . Show

that if
( B C e) T V 1 ( B C e) > 0,
then A C B2 > 0. See (3.33) on page 107.
3.20. (Two Securities) Verify Equation (3.38) on page 108, i.e., show that

C B 1 A B
w = V e + V 1 ,
AC B2 AC B2
where
w 1 2
w = = .
1 w 1 2 1
3.21. Let f ( x ) = ax2 + bx + c, where a, b, and c are real numbers with a = 0.
a) Show that if b2 4 a c < 0 and f ( x ) 0 for all x R, then f ( x ) > 0 for all
x R.
, that if a > 0 and f ( x ) > 0 for all x R, then the global minimum point
b) Show
of f is
b
x =
2a
and the corresponding global minimum value is
" #
(b2 4ac)
f ( x ) = .
4a
c) Use the above results to give an alternative proof that the two-security port-
folio variance,
P2 (w) = w2 12 + (1 w)2 22 + 2w(1 w)1 2 ,
is strictly positive. Compute the global minimum point w of the portfolio

risk P (w) and find P (w ). Compare with (3.44) and (3.47); see page 110.
3.22. (Two Securities) Given two securities Si , i = 1, 2. Let Ri , i = 1, 2, be their

return rates, respectively. Assuming that Ri , i = 1, 2, are independent and iden-
tically distributed continuous random variables, determine the portfolio that a
risk-averse investor would select.
3.23. (Two Securities) Consider a portfolio with two securities having returns
1 and 2 , risks 1 and 2 , and a correlation coefficient that vanishes. To
minimize this portfolios risk-to-reward ratio, a natural quantity to minimize is
P2 (w)
f (w) = ,
P w)
where w is the fraction of the total investment in the security with expected
return 1 .
a) Determine an equation that any critical point of f must satisfy. What type of
equation is it?
b) Show that if 1 = 2 , then we obtain a linear equation for w with solution
22
w= .
12 + 22
This critical point coincides with the global minimum w we found for the
two-security portfolio when minimizing the variance P2 with = 0. Why
are the two critical points identical?
3.24. (Three Securities) Consider a portfolio with three securities having risks
1 , 2 , and 3 and correlation coefficients 12 , 13 , and 23 . Let V be the covari-
ance matrix of the security returns. Show that V is positive definite if and only
if the following hold:
a) 1 > 0, 2 > 0, and 3 > 0,
b) |12 | < 1 and 212 + 213 + 223 212 13 23 < 1.
3.25. Let A be an n n matrix. Show that the gradient and Hessian of the
quadratic xT Ax are
( x T Ax) 2 ( x T Ax)
= ( A + A T ) x, = A + AT , x Rn ,
x xx T
2
f f f T 2 ( x T Ax) f
where x = x . . . xn and xxT = x x .
1 i j n n
3.26. (N Securities) For an N security portfolio, show that the portfolio vector
w which minimizes the variance P2 (w ) = w T V w, subject to w T e = 1, is the
global minimum-variance portfolio vector. Explain why this is expected.
3.27. (N Securities) Determine the equations for the lines asymptotic to the set
of all minimum-variance N-security portfolios.
3.28. Show that the covariance of the return of the global minimum-variance
N-security portfolio with the return of any other efficient N-security portfolio
is always 1/A.
References 149
3.29. Where does the tangent line at the diversified portfolio on the Markowitz
N-security efficient frontier intersect the P -axis?
3.30. Determine the equation of a line asymptotic to the Markowitz N-security

efficient frontier in the (P , P )-plane.
3.31. Let w a and wb be any two distinct minimum-variance portfolio vectors.

For suitable constants a and b, these vectors can be expressed in the following
form:
w a = (1 a)w G + aw D , wb = (1 b)w G + bw D ,
where w G and w D are the global minimum-variance and diversified portfolio
vectors, respectively.
a) Show that any minimum-variance portfolio vector w can be expressed as

1 A + b 1 1 a 1 A
w= wa + wb .
ba ba
b) Show that the covariance is given as follows:
1 AC B2
Cov( R P (w a ), R P (w b )) = + ab ,
A AB2
where A = e T V 1 e, B = T V 1 e, and C = T V 1 .
3.32. Let a > 0 and b = 0. Show that the utility functions u( x ) = ax + b and
u( x ) = ax 2b x2 , where x < ba , obey
1
E u( X ) = u(E ( X )) + u (E ( X )) Var( X ).
2
3.33. A power utility function refers to one of the form u( x ) = x a . When is u a
risk-averse utility function?
References
[1] Bodie, Z., Kane, A., Marcus, A.: Investments, 9th edn. McGraw-Hill,
New York (2011)
[2] Capinski, M., Zastawniak, T.: Mathematics forFinance. Springer,
New York (2003)
[3] Durrett, R.: Probability: Theory and Examples, 4th edn. University Press,
Cambridge (2010)
[4] Fine, T.: Probability and Probabilistic Reasoning for Electrical Engineer-
ing. Pearson Prentice Hall, Upper Saddle River (2006)
[5] Frigyik, B., Kapila, A., Gupta, M.: Introduction to the Dirichlet distribu-
tion and related processes. University of Washington Electrical Engineer-
ing Technical Report, Number UWEETR-2010-0006 (2010)
[6] Goldberg, L.: A review of Risk-Return Analysis: The Theory and
Practice of Rational Investing (Volume I) by Harry Markowitz and Ken-
neth Blay. http://www.cfapubs.org/doi/full/10.2469/br.v9.
n1.9%40faj.2014.70.issue-3 (2014)
[7] Goodman, J.: Statistical Optics. Wiley Classics Library Edition. Wiley,
New York (2000)
[8] Graham, J., Smart, S., Megginson, W.: Corporate Finance. South-Western
Cengage Learning, Mason (2010)
[9] Grinold, R., Kahn, R.: Active Portfolio Management. McGraw-Hill,
New York (2000)
[10] Ingersoll, J.: Theory of Financial Decision Making. Rowman and Little-
field, Savage (1987)
[11] Jorion, P.: Value at Risk, 3rd edn. McGraw-Hill, New York (2007)
[12] Korn, R.: Optimal Portfolios. World Scientific, River Edge (1997)
[13] Korn, R., Korn, E.: Option Pricing and Portfolio Optimization. American
Mathematical Society, Providence (2001)
[14] Levy, H.: A review of Risk-Return Analysis: The Theory and Practice of
Rational Investing (Volume I) by Harry Markowitz and Kenneth Blay
(2014). Quant. Finance 14(7), 1141 (2014)
[15] Levy, H., Duchin, R.: Asset return distributions and the investment hori-
zon. J. Portf. Manag. 30(3), 47 (2004)
[16] Luenberger, D.: Investment Science. Oxford University Press, New York
(1998)
[17] Markowitz, H.: Portfolio selection. J. Financ. Res. 7(1), 77 (1952)
[18] Markowitz, H.: Portfolio Selection. Blackwell, Cambridge (1959)
[19] Markowitz, H., Blay, K.: Risk-Return Analysis: The Theory and Practice
of Rational Investing, vol. I. McGraw-Hill, New York (2014)
[20] Merton, R.: An analytic derivation of the efficient portfolio frontier.
J. Financ. Quant. Anal. 7(3), 1851 (1972)
[21] Pennacchi, G.: Theory of Asset Pricing. Pearson Addison Wesley, Boston
(2008)
[22] Reiley, F., Brown, K.: Investment Analysis and Portfolio Management.
Dryden Press, Fort Worth (1997)
[23] Roman, S.: Introduction to the Mathematics of Finance. Springer,
New York (2004)
[24] Ross, S.: An Elementary Introduction to Mathematical Finance.
Cambridge University Press, Cambridge (2011)
[25] Vanderbei, R.: Linear Programming: Foundations and Extensions, 4th
edn. Springer, New York (2014)
Chapter 4
Capital Market Theory and Portfolio Risk Measures
The process of dividing a portfolio among major asset categories such as

stocks, bonds, real estate1, and cash is generally referred to as asset allocation.
A key aspect of asset allocation is portfolio risk management.
The risk that the entire financial system bears is called the systematic risk,
whereas the risk that a portfolio bears is called portfolio risk.
Systematic risk can be characterized by the potential of financial system dis-
ruption with substantial and adverse effects on the economy. The most re-
cent example of such risks was exhibited by 20072008 financial crisis.
Portfolio risk can be classified into two broad categories: market risk and
idiosyncratic risk.
Market risk is the risk that is correlated with price fluctuations of the gen-
eral market. Since an essential feature of the market is cross-sectional, it is
also called undiversifiable risk or aggregated risk. Sources of macroeconomic
factors such as inflation and changes in exchange rates are often considered
to be such a risk since the likelihood they will cause adverse market price
fluctuations is high.
Idiosyncratic risk is company-specific risk that is uncorrelated with price
fluctuations of the general market. As the Markowitz portfolio theory
showed in the last chapter, it can be greatly reduced through diversifica-
tion. Thus, it is also called diversifiable risk. Sources such as strikes, slump-
ing sales, and unexpected poor earning reports or forecasts are examples
of such risks.
Though in the current chapter we shall extend Markowitzs portfolio the-
ory, risk will still be measured by employing the variance and covariance of
returns. We shall see that the sensitivity to the market return plays an impor-
tant role. But later, we will introduce other measures of risk and tools used to
determine the risk-reward profile of a portfolio.
1 For example, REITs, which is an acronym for Real Estate Investment Trusts.

152 4 Capital Market Theory and Portfolio Risk Measures
Remark 4.1.
1. Unless stated otherwise, in this chapter, R P and r P denote the return2 and
logarithmic return of a portfolio P on a general time interval [t0 , t f ], respec-
tively. On the other hand, by default the risk-free rate r is a percent quoted
on an annual basis. To keep the mathematical expressions simple, in the
current chapter, both R, P and r are for the same period; consequently, so are
P = E ( R P ) and P = Var( R P ). For example, if one of them is monthly,
so are the rest.
2. Although we will provide all the concepts in this chapter using only total
returns, some concepts such as the Sharpe ratios and linear factor models
can be defined using logarithmic returns as well. The latter has advantages
in studying the properties of individual securities since logarithmic returns
are more tractable than ordinary returns. Furthermore, statistical tools can
be more conveniently applied with logarithmic returns, particularly under
the lognormal assumption of security prices.

4.1 The Capital Market Theory
Capital market theory, which includes the Capital Asset Pricing Model
(CAPM3 ), was developed by William Sharpe,4 John Lintner, and Jan Mossin.
This theory naturally generalizes the Markowitz mean-variance portfolio
model by introducing both a new efficient frontier that extends beyond the
Markowitz efficient frontier and a model for pricing individual securities. The
new efficient frontier is formed by adding a risk-free borrowing or lending
consideration, which turns the old efficient frontier into a half line.
Recall that the one-period Markowitz model is interested in risk-averse in-
vestors selecting portfolios at time t0 that produce stochastic returns at time t f .
Besides all the assumptions we made about the Markowitz model in the last
chapter,5 we also assume that:
all investors have equal access to borrowing and lending, which occur at
the same risk-free rate r, and lenders bear no risk of not being repaid;
the inflation rate is no more than the risk-free rate r.6
2 The return can be applied either in a simple or compounded context.

3 Pronounced CAP-M.
4 Harry Markowitz, Merton Miller, and William F. Sharpe shared the 1990 Nobel Prize in Economic
Sciences. Sharpe won for his contributions to the Capital Asset Pricing Model. See the Press Release
at Novelprize.org.
5 See Section 3.1.
6 Under normal circumstances, inflation constitutes a major portion of the risk-free rate. The only
problem with this is when the inflation is far above the risk-free rate due to a central bank intervention.
4.1 The Capital Market Theory 153
4.1.1 The Capital Market Line (CML)
Let A be a risk-free security with risk-free rate r.7 Consider a set of N risky
securities,8 and assume that the initial investment at time t0 is the amount V0 .
We shall investigate the best risk-return trade-off portfolio design that is based
only on allocating the initial investment between A and a portfolio B consisting
of the given N risky securities.
As in Markowitz portfolio theory, we represent each portfolio P by a point
in the (P , P )-plane. To avoid possible confusion about the time basis for these
quantities and r, we clarify our usage in the remark below:
Remark 4.2. The definitions of R P , P , and P used in this section are the same
as that in (3.8) (page 94), (3.13) (page 95), and (3.14) (page 96), respectively,
which are all relative to a time interval [t0 , t f ]. As noted in Remark 4.1, we allow
in this chapter for r to be not necessarily on an annual basis, but to have the
same period as R P . The same applies to R M , Ri , etc., which will be introduced
later.

Now, if we let V0 be the amount of the initial investment at time t0 , then P

with coordinates (0, r) on the -axis means that the investor puts all the money
V0 into the risk-free security A. Otherwise, P is a point in the right half plane
P > 0 and with P > r, since no investor would want to take on any risk to get
an expected return below or equal to r.
Let A and B be two points in the (P , P )-plane described by
A = (0, r) and B = (B , B ) with B > 0 and B > r.
First, we present a more intuitive discussion. Any point ( x, y) on the line de-
termined by the points A and B can be represented by
( x, y) = w0 (0, r) + (1 w0 )(B , B )
for some real number w0 , and (see Figure 4.1)
if w0 [0, 1], then ( x, y) is between A and B;

if w0 < 0, then ( x, y) is on the right side of B;
if w0 > 1, then ( x, y) is on the left side of A.
7 A risk-free security is, of course, a theoretical concept. In reality, any investment carries a certain
amount of risk. In this context, by risk-free securities we mean US T-bonds or FDIC insured bank
accounts to which we can lend out our money and obtain a sufficient credit line from which we can
borrow money. Under our assumption in the last section, the interest rate for lending is equal to that
for borrowing and occurs at the risk-free rate.
8 At this stage, N is arbitrary. Eventually we need to consider only a sufficiently large N for which we
have a diversified portfolio.

Fig. 4.1 The points M, B, and A have coordinates ( M , M ), ( B , B ), and (0, r ), respectively. The curve
M E,N is the efficient frontier of FP,N . The line CAL is the graph of equation (4.1), while the CML is the
graph of equation (4.3).
In our context, obviously, the case w0 > 1 is not applicable as no investor would
want to get an expected return below r. That is to say, a possible portfolio
design in terms of the portfolios expected return rate P and variance P2 is
given by
(P , P ) = w0 (0, r) + (1 w0 )(B , B ) where w0 1. (4.1)
The fact that
risk with lending < B < risk with borrowing
implies that
if w0 [0, 1], then the investor lends amount w0 V0 at time t0 ;

if w0 < 0, then the investor borrows amount w0 V0 at time t0 .
The graph of (4.1) in the (P , P )-plane as w0 varies is shown in Figure 4.1 and
is called a Capital Allocation Line (CAL).
Second, we take a more theoretical approach. Let FP,N denote the set of fea-
sible portfolios (the Markowitz bullet) that contains only the given N risky
securities, and let FP,N +1 denote the set of all feasible portfolios that contain
the given risk-free security and the N risky securities. We have
N
FP,N WN = {w = [w1 , w2 , , w N ] : w i = 1},
i =1
N
FP,N +1 WN +1 = {w = [w0 , w1 , w2 , , w N ] : w i = 1},
i =0
where w0 is the weight of the risk-free security A, and wi is the weight of the
ith risky security for i = 1, 2, . . . , N, and WN and WN +1 are defined as indicated.
Since we shall need N sufficiently large, it suffices to assume at this stage that N 3.9
We are interested in finding the efficient frontier of FP,N +1 .10 Recall that the
Markowitz efficient frontier of FP,N is denoted by ME,N ; see Figure 4.1. For ease
of presentation, we shall slightly abuse our notation and denote the efficient frontier of
FP,N +1 by ME,N +1 (though all the securities are not risky). For each P FP,N +1 ,
there exists vector w P WN +1 such that P can be expressed by
N
w P = [ w0 , w1 , w2 , , w N ] with wi = 1. (4.2)
i =0
Let
w N = [ w1 , w2 , , w N ] ,
where wi , i = 1, 2, . . . , N remain the same as in (4.2). Notice that w N
/ WN unless
w0 = 0 for iN=1 wi = 1 w0 . We have
N
P = w0 r + w i i , P2 = w
N V wN ,
i =1
where V is the covariance matrix of the random vector [ R1 , R2 , , R N ] and

here Ri is the return from the ith risky security over the time period [t0 , t f ] for
i = 1, 2, . . . , N.
Let wi = 1wwi , and w = [w1 , w2 , , w N ] . Then w WN as iN=1 wi = 1.
0
Since w N = (1 w0 )w, we have
N
P = w0 r + (1 w0 ) wi i = w0 r + (1 w0 ),
i =1
P2 = w 2
N V w N = (1 w0 ) w V w = (1 w0 ) ,
2 2
where (, ) represents a portfolio containing no risk-free securities. Thus, for

w0 < 1,
P = (P , P ) ME,N +1 iff (, ) ME,N .
9Recall that the feasible set for N = 2 is a curve, while for N 3, it is a region (see page 126).
10The proof to be provided here is for a one-period portfolio. The proof for n-period self-rebalancing
portfolios is similar. The details are left as an exercise for the reader.
Obviously, the graph of ME,N +1 in the (P , P )-plane is the (half) tangent line
to the graph of ME,N at the point (, ) and with the left end point (0, r). Other-
wise, either the graph of ME,N +1 is a secant line to that of ME,N or is above that
of ME,N . In either case, a contradiction can be produced. The graph of ME,N +1
in the (P , P )-plane is called a Capital Market Line (CML).
Finally, combining our above intuitive and theoretical discussions, we have
the equation for the CML:
(P , P ) = w0 (0, r) + (1 w0 )(, ),
where (, ) ME,N .
It follows from the Markowitz portfolio theory of Chapter 3 that the efficient
portfolio (, ) offers the best risk-return trade-off. Everyone will then want to
invest in the portfolio (, ) and will want a risk-free asset to be on a CML. If
a security is not part of this portfolio, then there is no interest in the security,
which will cause it to drop out the marketplace. Consequently, the tangent point
portfolio (, ) then consists of all securities in the marketplace and so N is now the
total number of securities in the market. It also follows that the weight of each security
in (, ) is the percent of the marketplace the security occupies, i.e., the weight is the
securitys market capitalization. For this reason, (, ) is called the market portfolio
and denoted by (M , M ). Therefore, our desired portfolio should be designed
by putting the percentage w0 in the risk-free security and 1 w0 in the market
portfolio:
(P , P ) = w0 (0, r) + (1 w0 )(M , M ). (4.3)
Again, the graph of (4.3) in the (P , P )-plane is called a capital market line
(CML). It is depicted in Figure 4.1.
Using the parametric equations of the CML(obtained from the equation (4.3)),
.
P = w0 r + (1 w0 ) M
,
P = (1 w0 )M
to express P in terms of P , we obtain the slope-intercept form of the CML:

M r
P (P ) = P + r . (4.4)
M
Example 4.1. (No Leverage Versus Leverage) Suppose that you have $2,000 to
invest. Assume a risk-free rate of 6% and market expected return of 12%.
a) If you invest $1,500 in a risk-free security at the rate of 6% and put the rest
in the market portfolio, then what are the expected portfolio return rate and
portfolio risk?
Solution. Since w0 = $1,500

$2,000 = 0.75 and r = 0.06, we have
P = w0 r + (1 w0 ) M = 0.75 0.06 + 0.25 0.12 = 0.075 = 7.5%

P = (1 w0 )M = (1 0.75)M = 0.25M .
The portfolio has less expected return than the markets, but the portfolio
risk is only 25% of the market risk.
b) If you add leverage to your portfolio by borrowing $1, 500 at the risk-free
rate, then what are the expected portfolio return rate and portfolio risk?
Compare with the previous case.
Solution. In this case, w0 = 0.75. Consequently,
P = w0 r + (1 w0 ) M = 0.75 0.06 + 1.75 0.12 = 0.165 = 16.5%

P = (1 w0 )M = (1 (0.75))M = 1.75M .
This means that after paying back the loan, the expected return is 16.5%,
which is the sum of the loss in returns due to paying the loan and the gain in
returns from investing the loan and the initial capital. The expected return is
more than double that for the previous case. However, such higher expected
returns expose you to risks that are much more volatile than the market,
about 1.75 times the market volatility. The leveraged portfolio is seven times
more risky than the portfolio in the previous case.

4.1.2 Expected Return and Risk of the Market Portfolio
The CML shows that the market portfolio is the best efficient frontier portfo-
lio to combine with a risk-free security to exceed the expected returns of the
Markowitz model.
Though in practical applications we can employ a proxy for the market
portfolio to position a portfolio on the CML, the proxy does not give us any
quantitative understanding of where the market portfolio is on the Markowitz
efficient frontier. Moreover, one may be interested in investing in a limited
number of stocks, for which there may be no obvious proxy, and wish to con-
struct a CML-type tangent to the Markowitz efficient frontier of those stocks.
For these reasons, we compute, for a general N-security risky portfolio, the
point of tangency of the CML to the Markowitz efficient frontier of the N secu-
rities. We shall obtain explicit expressions for the expected return and risk of
the point of tangency, which we call the market portfolio for the N securities, even
if the securities form a subset of the true market portfolio. The mathematical
framework will, therefore, be general enough to incorporate the cases ranging
from two securities to any finite number.
Given a risk-free rate r and N risky securities, we apply the definition of
and formula for in the Markowitz portfolio theory:
-
" A2M 2B M + C
M = w
M , M ( M ) = w
MV wM = ,
AC B2
where w
M e = 1, to obtain the expressions of M , M and w M as follows:
C Br Ar2 2Br + C V 1 ( M re )
M = , 2
M = , wM = . (4.5)
B Ar ( B Ar)2 B Ar
The computational details are left as an exercise for the reader (Exercise 4.23).
The reader may recall the diversified portfolio that was introduced in the
Markowitz model as part of establishing the Mutual Fund Theorem (see (3.81)
on page 131). The market portfolio is a generalization of the diversified port-
folio (D , D ). In fact, if r = 0, then we obtain the diversified portfolio from the
market portfolio:

C C V 1
M = = D , M = = D , wM = = wD ,
B B B
where we use B > 0. In this case, the CML runs from the origin to the diversi-
fied portfolio (D , D ).
4.1.3 The Capital Asset Pricing Model (CAPM)
Consider a security with return Ri . The securitys beta, denoted i , measures

the degree to which the securitys return moves in step with the markets re-
turn:
Cov( Ri , R M )
i = iM i ,
M2 M
where R M is the return of the market portfolio and M is the market portfolios
risk.
The risk of the security is bounded below in terms of beta and the market
risk as follows:
i | i | M .
We shall see that the result can be improved to an equality relating the risk
premium of the security in terms of beta and the market risk premium.
The risk premium of a security is defined to be
i r,
which is how much the securitys expected return is above/below the risk-free
rate r. The risk premium of the market portfolio is
M r.
This is how much the markets expected return is expected to differ from the
risk-free rate. The Capital Asset Pricing Theorem, which is due to Sharpe, Lint-
ner, and Mossin, relates the risk premiums of the security and market via beta.
Theorem 4.1. (Capital Asset Pricing Theorem) Assume that the covariance ma-
trix V of all securities is positive definite and and e are linearly independent. Let
i = e
i , where e i = [0 0 1 0 0] with 1 in the ith slot, be the expected return
on the ith security. Then
i r = i ( M r ). (4.6)
Proof. The idea is to show that
Cov( Ri , R M ) e
i V wM r
i = =
= i .
M2 wMV wM M r
Let wM be the weight of the th security in the marketplace. Then
N N
Cov( Ri , R M ) = Cov( Ri , w
M R M ) = Cov( R i , w R ) =
M
wM Cov( Ri , R )
=1 =1
= e
i V w M.
Furthermore, since

11 1N
.. .. w M
. .
.1 N

ei V w M = 0 1 0 i1 iN . = wM i ,
.
.. .. M =1
. . wN
N1 N N
it follows
e
i V wM
i = .
w
MV wM
Now, we saw that the market portfolio vector is given by
V 1 ( M re )
wM = ,
B Ar
or equivalently,
M re
V wM = .
B Ar
Because e
i M = i , ei e = 1, w M M = M , and w M e = 1, we get

e
i V wM e
i M re i e B Ar
i = =
w
MV wM
B Ar w
MM rw
Me
i r
= .
M r

Although beta is a relatively stable measure of a stocks relative volatility

(a stocks risk relative to the market), a stocks beta may change over time.
Additionally, (4.6) carries the following messages:
1. Since as the slope of the regression line (see (4.6)) relates the excess returns
of the stock to that of the market, measures the sensitivity of the stocks
return to the fluctuations in the market.
2. ( M r) is a risk premium, where M r is the premium in units of .
3. The sign of the beta of a stock indicates the direction of the movement of
the stock price with respect to that of the market portfolio, and moreover,
statistically speaking,

<0 indicates that the stock is losing money while the market

as a whole is gaining and vice versa;

= 0 indicates that the stock fluctuates uncorrelatedly

with the market;

( 0, 1 ) indicates that the stock fluctuates less than the market;

=1 indicates that the stock fluctuates the same as the market

(i.e., same direction and magnitude as the market);

>1 indicates that the stock price fluctuates more than the market.
Example 4.2. (Security Pricing via CAPM) Let D (t0 , t f ) be the stock cash div-
idend issued in the time period [t0 , t f ). Let S(t) be the stocks price at time t.
Recall the one-period return (see (3.1) on page 85):
S ( t f ) + D ( t0 , t f ) S ( t0 )
R ( t0 , t f ) = ,
S ( t0 )
where S(t f ) and D (t0 , t f ) are random variables, whereas S(t0 ) is deterministic.
Taking expectations on both sides of the last equation yields

E S ( t f ) + E D ( t0 , t f ) S ( t0 )
E R ( t0 , t f ) = .
S ( t0 )
Solving for S(t0 ) from the last equation produces


E S ( t f ) + E D ( t0 , t f )
S ( t0 ) = . (4.7)
1 + E R ( t0 , t f )
Applying the CAPM to valuation of the stock price at time t0 , we obtain the
asset pricing formula based on the CAPM
S(t f ) + D (t0 ,t f )
S ( t0 ) = ,
1 + r + ( M r)

where S(t f ) = E S(t f ) and D (t0 ,t f ) = E D (t0 , t f ) .

Remark 4.3.
1. Given a time interval [t0 , t f ], (4.7) provides a relation between an asset price
at time t0 and the expectation of its return over the interval. This is to say
that every asset return model corresponds to an asset pricing model. This is
why authors use terminologies like modeling asset prices and modeling
asset returns interchangeably.
2. It is worth noting that the CAPM shows how the market must price an indi-
vidual security in relation to its asset class index, which we call beta, a risk
measure. Thus, the CAPM as an asset pricing model also shows how the set
of all securities can be classified by one risk measure, which is the beta in the
case of the CAPM. Looking ahead from these two points, the latter sections
of the chapter will branch into risk measures and linear factor models which
are a class of much more practically useful asset pricing models.

Example 4.3. The hurdle rate refers to the minimum acceptable (rate of an)
investment return. This concept plays an important role in decision-making
when a project is under consideration.
Here is a table of project specifications:
Project A B
Project beta 1.5 1.4
Initial Investment $10,000 $10,000
Expected Payoffs $8,000 in 2 years $9,000 in 2 years
$16,000 in 5 years $9,000 in 5 years
$9,000 in 8 years
Assuming that the hurdle rate is calculated by the formula,
hurdle rate = E ( R project ) = r + project (E ( R M ) r),

where r = 2% and E ( R M ) = 20%, determine if any of the projects is worth

pursuing.
Solution. We have
E ( R project A ) = 0.02 + 1.5 (0.2 0.02) = 0.290,

E ( R project B ) = 0.02 + 1.4 (0.2 0.02) = 0.272.
Since
$8, 000 $16, 000
NPV A (0.290) = $10, 000 + + = $713.70,
1.292 1.29
5

1 1 1
NPVB (0.272) = $10, 000 + $9, 000 + +
1.2722 1.2725 1.2728
= $421.52,
neither project is worth pursuing.

Finally, the CAPM Theorem also applies to portfolios. First, the concept of beta
extends naturally to portfolios. If P represents a portfolio of n risky securities
with portfolio weight vector w given by

w1 n
..
w = . , wi = 1, (4.8)
i =1
wn
the portfolio beta is defined to be the weighted average of the individual risky
security betas:
P = w1 1 + w2 2 + + wn n , (4.9)
where i is the beta of the ith risky security in P for i = 1, 2, . . . n. Second, for
portfolio P we denote by P and i the expected portfolio return and expected
return of the ith security in P , respectively. Then by applying (4.6) on page 159
to the ith risky security in P and employing (4.8) and (4.9), we obtain
n n
P r = wi (i r) = wi i ( M r).
i =1 i =1
Hence, we obtain the CAPM for portfolios:
P r = P ( M r ). (4.10)
Remark 4.4.
1. The CAPM is a theoretically significant equilibrium pricing model. Without
a model of market equilibrium, the efficient market hypothesis cannot be
tested (see Fama [16]).
2. Although the CAPM has a beautiful simplicity in theory, the empirical ev-
idence shows its weaknesses in practice (see Fama and French [19] for a
detailed and comprehensive discussion).

4.1.4 The Security Market Line (SML)
The Capital Asset Pricing Theorem can be viewed as expressing the expected
return of a security as an affine function of the securitys beta, i.e., it defines a
straight line:
i ( i ) = ( M r) i + r.
This line is called the security market line (SML). That is, the SML is a graphical
representation of the CAPM on the ( , )-plane.
An illustration of the SML is shown in Figure 4.2. A security with i = 0 has
an expected return at the risk-free rate rthe security has no risk premium. For
i = 1, we have a security with i = M , i.e., the risk premium of the security
coincides with the market risk premium. If i = 1.5, then the securitys risk
premium is 1.5 times the market risk premium or 50% larger than the market
risk premium. For i = 1, the securitys risk premium is minus the market
risk premium, which means that the expected security return is less than the
expected market return by twice the market premium. Indeed, if i = b < 0,
then
i = M (1 + b)( M r).
In theory, the SML provides an equilibrium in the sense that each stock on
the line is fairly valued. Otherwise, it is located off the line:
1. A stock is overvalued if it is below the SML because such a stock offers
too low a risk premium. In other words, the level of expected return is not
adequate for the given level of risk measured by .
2. A similar argument can be made to show that a stock is undervalued if it is
above the SML.
mi
0.20
e
L in
0.15
et
k
ar
1, mM
M
rity
cu
Se
0.10
0.05
rf
1.0 0.5 0.5 1.0 1.5 2.0

bi
0.05
Fig. 4.2 An illustration of the security market line (SML) for risky securities identified by ( , ). A
security with its risk measured by i has expected return i . In particular, for i = 0, the security earns
the risk-free rate (i = r), so the security has no risk premium. A security with i = 1 has an expected
return at the market expected return, i = M 13%. For i = 1, the security has i 5%, i.e., the
expected security return is less than the expected market return by twice the market risk premium
(i = M 2( M r ) 5%)
4.1.5 CAPM Security Risk Decomposition
By the Capital Asset Pricing Model, we have
E ( Ri r) = E ( i ( R M r)) ,
which yields the following model for the return of a risky security:11
Ri = r + i ( R M r) + i ,
where i is a random variable with mean zero. Assume that i is normal with
variance denoted by 2i and suppose that i is independent of R M and j for
j = i.
The securitys risk can be found from
i2 = Var( Ri ) = Var( i R M ) + Var( i ) = 2i M

2
+ 2i . (4.11)
11If E ( X ) = E (Y ), then X = Y + , where is a random variable with E ( ) = 0, called an error term.

Typically, an error term is assumed to be normal (with mean zero).
4.2 Portfolio Risk Measures 165
The term
Var( i R M ) = 2i M
2
is the market risk (also known as systematic risk), while
Var( i ) = 2i
is the idiosyncratic risk (also known as unsystematic risk). Expression (4.11)

asserts that the total risk of a security can be decomposed into two orthogonal
components (under the inner product of covariance). In this sense we write
Total Risk = Systematic Risk + Unsystematic Risk.
Under the CAPM for security returns, we see that the covariance between
the returns of any two securities is determined by the betas of the securities
and market risk:

Cov( Ri , R j ) = Cov r + i ( R M r) + i , r + j ( R M r) + j
= i j Cov( R M , R M )
= i j M
2
.
4.2 Portfolio Risk Measures
Risk measures are a challenging topic as the notion of risk itself is hard to con-
ceptualize. The most popular measure of risk is volatility, which by definition
measures the dispersion of the investment return from its mean, regardless of
the direction of an investment prices movement. However, for investors in the
real world, risks are often associated with the adverse movement of the market
only. This is to suggest we also take a different perspective in the further de-
velopment of risk measures: if a risk measure is about the sustainability of los-
ing money, then maximum drawdown is employed (Section 4.2.3). If the risk
measure is about the odds of losing money, then VaR and CVaR are used (Sec-
tions 4.2.5 and 4.2.6). Especially in a leveraged investment, the performance is
measured by the return on unit risk, such as the Sharpe ratio (Section 4.2.1), the
Sortino ratio (Section 4.2.2), or the ratio of return and maximum drawdown.
It is worth pointing out, nevertheless, that in spite of the intuitiveness of the
latter two ratios, the system developer often prefers to optimize the in sam-
ple Sharpe ratio as it is intimately connected to the statistical t-test, which is a
measurement of the reliability of the system in out sample period.
In short, this section addresses several approaches to risk measures, which
provide a variety of portfolio evaluation techniques.
Remark 4.5.
1. Before formally introducing any mathematical terminologies about risk, we
encourage the readers to think about how they themselves as (individual
or institutional) investors interpret risk. For insight into how some of the
worlds greatest minds have viewed risk, we refer the reader to Peter Bern-
steins book [5].
2. The lack of statistics in the prerequisite presents us with a great challenge in
this section and the next. To circumvent this difficulty, one of our pedagog-
ical approaches is to focus on basic understanding of concepts and avoid
statistical tests.

4.2.1 The Sharpe Ratio
Consider a portfolio P over a time period [t0 , t f ]. Let R P be the portfolio return
over [t0 , t f ] and let r be the best available risk-free rate corresponding to the
same period (e.g., T-bills).12 The portfolios Sharpe ratio,13 denoted by S ( P), is
defined by
E ( R P r)
S ( P) = .
P ( R P r)
The Sharpe ratio was originally developed as a forecasting tool with the
expected return to calculate the forward-looking ratio (see Sharpe [36]). But
with the historical returns, which can be of any frequency, e.g., hourly, daily,
monthly, and so on, it is used to evaluate the risk-reward trade-off of an invest-
ment over that period.
Example 4.4. Suppose that we have the data set of monthly returns of a portfo-
lio P and that of monthly rates of 90-day T-bills over the past 54 months. This
is to say that each data set consists of 54 data points. Using the sample mean
and sample standard deviation, we can approximate the Sharpe ratio of the
portfolio over the past 4.5 years (54 months). A Sharpe ratio of
S ( P) = 0.0401
is interpreted as the portfolio P earned an average excess return of 4% per unit

risk over the time period under consideration.
12 Recall that US Treasuries can be classified into bills, notes, and bonds according to their initial ma-
turities (in years) in terms of time intervals: (0, 1], (1, 10], and (10, ), respectively. We consider only
T-bills here since, the longer the maturities, the bigger the risk of inflation, and consequently, the less
reliable.
13 Named after William Sharpe.
Fig. 4.3 The slope of the CML equals the Sharpe ratio
Statistically speaking, data sets need to be sufficiently large. The more data
points we use (the shorter the sub-sample period), the more accurate our ap-
proximation of S ( P) becomes.

For a constant risk-free rate, we have
E ( R P r) = E ( R P ) r, Var( R P r) = Var( R P ),
so
E( RP ) r
S ( P) = .
P ( R P )
The following explains the significance of the Sharpe ratio:
1. Since E ( R P r) and P represent the expected excess return and risk, re-
spectively, the Sharpe ratio is a measure of the excess return per unit of risk
of the portfolio. In other words, the ratio describes how much risk premium
you are receiving for the extra volatility that you endure for holding a riskier
portfolio. In short, the Sharpe ratio measures risk-adjusted performance of
the portfolio.
2. Recall that if you invest part of your money in a risk-free security (e.g.,
T-bills) and the remainder in an efficient portfolio, the capital market line
(CML) can help you find the portfolio P that offers the most favorable
risk-return trade-off. In fact, assuming that R P and r are the same as we
used in the discussion of the CAPM, the slope of this CML is equal to the
Sharpe ratio of P (see Figure 4.3). This observation provides a method for
finding the best possible portfolio from the given collection of securities.
3. The Sharpe ratio is a leverage-environment14 measure of performance in the
sense that if r is omitted, then approximately
14 Two basic ways of achieving leverage are (a) to borrow money for investment and (b) to use finan-
cial instruments such as futures and options (see Chapter 7).
E( RP ) bE ( R P )
SP = = ,
P ( R P ) b P ( R P )
where b denotes the leverage factor. For instance, if we double every invest-
ment, then the return is doubled, but the risk (standard deviation ) is also
doubled. If a hedge fund predetermines the risk level, then the Sharpe ratio
determines the return from which the leverage level can be determined for
products allowing high leverage (e.g., futures, commodities, currencies, and
options). In this sense, the Sharpe ratio provides a method of optimizing a
portfolio. Otherwise, the Sharpe ratio does not provide a portfolio of highest
return possible.
4. A negative correlation can considerably reduce the standard deviation (com-
pare page 115). Even in the case that the portfolio return is reduced, the
Sharpe ratio may still increase.
Remark 4.6.
1. Sharpe ratios can be defined by using logarithmic returns as well. Keeping
this in mind, we replace R P by r P in the definition of the Sharpe ratio to
obtain
E (r P r )
S ( P) = .
P (r P r)
For a constant risk-free rate, we have
E (r P ) r
S ( P) = .
P (r P )
2. For the reader who has had an introductory statistical background and
wishes to delve further into the significance of the Sharpe ratio, the
t-statistic is useful. In testing the null hypothesis = 0 , where denotes
the population mean, one uses
x 0
t= ,
( x )/ n
!
where x is the sample mean of the data, ! ( x ) is the sample standard devia-
tion of the x-data, and n is the sample size. The ! and are used to indicate
that a sample-data estimate is being carried out.
In connection to our interest, we rewrite for constant r,
RP r
t= = n!
SP ,
!
P / n
where
P = !
! ( R P r) = !
( RP )
and n is the number of returns used in the calculation of ! S P . This relation

shows us how reliable the portfolio strategy is in obtaining excess returns.

Example 4.5. Consider a portfolio with the following annual returns:
Year Year 1 Year 2 Year 3 Year 4 Year 5 Year 6
Return Rate 21% 7.8% -13% 59.4% 0.2% -1.2%
Suppose that the average T-bill return for those 6 years is 4%. Find the Sharpe
ratio for the portfolio in the period of these 6 years.
Solution. In this case, model the risk-free rate r over the total time span as a
constant given by 4% per year. The Sharpe ratio is then
! R r
SP = P .
!
P
Applying the formula for sample mean
1 n
n i
X= Xi ,
=1
we obtain
1 21 + 7.8 13 + 59.4 + 0.2 1.2 74.2 12.3667
RP = = = .
6 100 6 100 100
Note that we converted the percentages to pure fractions. The sample variance
formula,
2
ni=1 Xi2 nX
P =
! 2
,
n1
yields
1 212 + 7.82 + 132 + 59.42 + 0.22 + 1.22 6 12.36672 656.6137

P2 =
! = .
5 1002 1002
Thus,
! R r 100 R P 100 r 12.3667 4 8.3667

SP = P = = = = 0.3265.
!
P 100 !
P 656.6137 25.6245

Example 4.6. Suppose that the risk-free rate is 4% and that portfolio i is oper-
ated under strategy i, where i = 1, 2. Consider the following information:
E ( R) S P
Portfolio 1 17% 9% 1.44
Portfolio 2 15% 5% 2.2
Which portfolio is more attractive to investors?
Solution. Assuming that S P1 and S P2 are equally statistically significant, port-

folio 2 is preferable. Although strategy 1 generates a higher expected return,
strategy 2 is able to generate a higher return on a risk-adjusted basis. Other-

wise, since t = n S P , without information on n, one cannot tell the reliability
of S Pi for i = 1, 2.

Remark 4.7.
1. For long-term investments, a Sharpe ratio of S P > 1 is typically considered
desirable. However, some short-term traders may consider only S P 3 good
enough, while other fund managers may consider S P > 2 to be an appropri-
ate target level.
2. Under the lognormal assumption, the unit conversions of volatility can eas-
ily be made by applying the property that the variance of its increments is
linear in the observation interval.
3. One should always remember that the Sharpe ratio is calculated based on
the historical returns and that past returns might be an indicator of future
performance, but they are certainly not a guarantee.
4. For a complex trading or investing system, the Sharpe ratio may provide
false information.

4.2.2 The Sortino Ratio
The standard deviation of the return of an investment does not distinguish

between up (good) and down (bad) volatility. A variation of the Sharpe ratio
is the Sortino ratio, which uses semivariance to differentiate between good
and bad volatility. Thus, defining the concept of semivariance is a preparation
for understanding the Sortino ratio. We begin with a brief review of variance.
Let X be a random variable with the density function f ( x ). The variance of
X is a measure of the dispersion of X from its mean u = E ( X ) and can be
computed by
1
2 = Var( X ) = E (( X u)2 ) = ( x u)2 f ( x ) dx.

Using the same idea, we can compute the dispersion of X from a given num-
ber a by
1
a2 = ( x a)2 f ( x ) dx.

In a similar fashion, if we are interested only in the dispersion of X from one
side of number a, say, from the downside (, a], then the formula below
serves that purpose: 1 a
( x a)2 f ( x ) dx.

This leads to a natural way of defining sample semivariance. Let X1 , . . . , Xn be
a random sample of size n drawn from a population X. Let a be a number. Let
.
a if Xi a
Yi =
Xi if Xi < a,
where i = 1, . . . , n. The downside sample semivariance a2 of X is defined by
1 n
n i
a2 = (Yi a)2 . (4.12)
=1
Note that (4.12) is equivalent to
1 n
n i
a2 = (min{0, Xi a})2 . (4.13)
=1
We are now ready for the definition of the Sortino ratio. Let
P represent a portfolio over a time period [t0 , t f ],
R P be the portfolio return over [t0 , t f ],
r0 be the target or required rate of return for the investment strategy under
consideration.15
The Sortino ratio, denoted by SD ( P), is defined by
E ( R P ) r0
SD ( P) = ,
r0
"
where r0 = r20 and is the downside deviation of the portfolio equity.
Example 4.7. Let us use the same information given in Example 4.5 and find
the Sortino ratio in the period of those 6 years.
To apply (4.13), we first compute
min{0, 21 4} = 0, min{0, 7.8 4} = 0, min{0, 13 4} = 17,

min{0, 59.4 4} = 0, min{0, 0.2 4} = 3.8, min{0, 1.2 4} = 5.2.
15The quantity r0 was originally known as the minimum acceptable return (or hurdle rate) and is often
taken to be r.
Thus,
1 172 + 3.82 + 5.22 330.48 55.08
r20 = = = .
6 100 2 6 100 2 1002
We obtain
RP r 8.3667
SD ( P) = = = 1.1273.
r0 55.08

Since the Sortino ratio captures the downside risk only, one can use the
Sortino ratio as a measure to rank the performance of a portfolio or an in-
vestment strategy.
4.2.3 The Maximum Drawdown
A drawdown is a portfolio equity retracement. The maximum drawdown is the

maximum equity retracement of a portfolio over a period of time. In other
words, it is the peak-to-trough portfolio equity decline during a specific time
period. The maximum drawdown measures how sustained ones losses can
be and is usually quoted as the percentage between the peak and the trough.
(For example, let the peak be 1, and the trough be 0.8. Then the maximum
drawdown is 1 0.8 = 0.2 = 20%.) To express this concept in a rigorous fashion
we introduce the following notation:
V (t) = the value of the portfolio at time t,

M (t) = max V (u), the maximum value of the portfolio over time period [0, t].
u [0,t]
Hence, our description of the maximum drawdown given above can be ex-
pressed below:
Definition 4.1. Given a time period [0, T ] and a portfolio, the maximum draw-
down of portfolio equity over [0, T ], denoted by MDD( T ), or simply MDD, is
defined by
MDD( T ) = max (V (u) V (v)). (4.14)
0 uv T
However, in practice, the formula
MDD( T ) = max ( M (t) V (t)) (4.15)

0 t T
may be easier to use in computations as there is only one parameter involved.

For the definition to be meaningful, we must show that (4.14) and (4.15) are
equivalent. Let
M(1 ) = max (V (u) V (v)),

0 uv T
M(2) = max ( M (t) V (t)).

0 t T
Since M (v) = maxu[0,v] V (u) given v [0, T ], we have
max (V (u) V (v)) M (v) V (v) for each fixed v.

0 uv T
This in turn implies that
max (V (u) V (v)) max ( M (v) V (v)).

0 uv T 0 v T
That is, M(1) M(2) .

On the other hand, since
V (u) M (u) M (v) for u v,
V (u) V (v) M (u) V (v) M (v) V (v) for u v,

we have
max (V (u) V (v)) max ( M (v) V (v)) .
0 uv T 0 v T
That is, M(1) M(2) . Therefore, M(1) = M(2) .

We conclude that the definition of MDD( T ) can be given by either of the
following equivalent statements:
Statement 1: MDD( T ) = max (V (u) V (v)).

0 uv T
Statement 2: MDD( T ) = max ( M (t) V (t)).

0 t T
Statement 1 is more intuitive in understanding the concept. Statement 2 is eas-

ier to use in computation.
Example 4.8. Consider a portfolio with these given daily values:

Date Day 1 Day 2 Day 3 Day 4 Day 5 Day 6
$ (Vi ) 1141.95 1143.73 1147.73 1142.24 1140.81 1149.88
By the definition of M (v),
M (1) = V1 , M (2) = max{V1 , V2 } = V2 ,
M (3) = max{V1 , V2 , V3 } = max{V2 , V3 } = V3 ,

and the rest of values of M are
M (4) = V3 , M (5) = V3 , M (6) = V6 .

Since
max { M (t) V (t)} = max{0, 0, 0, V3 V4 , V3 V5 , 0} = V3 V5 ,

0 t 6
the maximum drawdown of the portfolio equity over the time period of these
6 days is
MDD = 1, 147.73 1, 140.81 = 6.92.
Since a drawdown is usually quoted as the percentage between the peak and
trough, in terms of percentage, MDD is 6.92/1147.73 0.00603 0.6%.

4.2.4 Quantile Functions
The quantile function is one of the basic statistical concepts. We are primarily
interested in its fundamental role in defining distortion risk measures.16
The materials in quantiles covered in this section mainly serve as a prepara-
tion for introduction to the concepts of value-at-risk and conditional value-at-
risk in the next two sections.
Definition 4.2. Let X be a random variable with c.d.f. FX . Given p (0, 1), a
p-quantile of X (or its distribution) is a number a satisfying the properties
P(X < a) p and P ( X a) p. (4.16)
The number a is also referred to as a quantile of order p for the distribution of X.

Let FX ( a ) denote P ( X < a), (4.16) is equivalent to
FX ( a ) p and FX ( a) p.
Keep in mind that a c.d.f. is a right-continuous and monotonically increas-

ing (i.e., nondecreasing) function whose domain is the entire real line.
Example 4.9. Consider a random variable X taking on the values 1, 2, 3, 4, 5, 6,
7, 8, 9, and 10 with p.d.f. given by P ( X = i) = 11
1
for i = 7 and P ( X = 7) = 11
2
.
The fact that
6 8
P ( X < 7) = and P ( X 7) = ,
11 11
8 9
P ( X < 8) = and P ( X 8) =
11 11
shows that both 7 and 8 are 8
11 -quantile of X, and so is each a [7, 8].

16 A distortion risk measure is a type of risk measure related to the cumulative distribution func-
tion of a financial portfolio return. CVaR (to be introduced shortly) is an example of a distortion risk
measures.
Definition 4.3. The quantile function of a random variable X is denoted by FX1

and defined by
FX1 ( p) = min{ x R | FX ( x ) p}, p (0, 1). (4.17)
In words, the value of a quantile function of X at a point p is defined by the

lowest p-quantile of X.
In a situation where there is no ambiguity of the random variable, to ease
the notation, the quantile function is also denoted by Q( p).
Example 4.10. Let X be an exponential random variable with parameter .
Find the quantile function of X.
Solution. The probability density function of X is

.
0 if x < 0
f (x) = x
e if x 0.
Let F be the cumulative distribution function of X. Then

1 x
0 if x < 0
F( x) = P(X x) = f (t) dt = 2 x t
e dt if x 0
. 0
0 if x < 0
=
1 ex if x 0.
By definition, Q( p) = F 1 ( p), where p (0, 1). Thus
F(x) = p if and only if 1 ex = p.

ln(1 p )
Solving for x, we obtain Q( p) = .
Observe that Q( p) is a continuously increasing function on (0, 1).

Example 4.11. We revisit random variable X taking on the values 1, 2, 3, 4, 5, 6,

7, 8, 9, and 10 with p.d.f. defined by P ( X = i) = 11
1
for i = 7 and P ( X = 7) = 11
2
.
A straightforward verification shows that

i 1 i
i if p ( 11 , 11 ], i = 1, 2, 3, 4, 5, 6

Q( p) = 7 if p ( 11
6 8
, 11 ]

i if p ( 11i , i+
11 ], i = 8, 9, 10.
1
Observe that Q( p) is nondecreasing and left-continuous on (0, 1). Shortly we

will show that these properties of Q are not accidental.

The next property helps the reader to visualize graphs of quantile functions.
Property 4.1.
1. A quantile function is monotonically increasing on (0, 1).
2. A quantile function is left-continuous on (0, 1).
Proof. Let X be a random variable. As usual, F and Q represent the c.d.f. and
quantile function of X, respectively.
To show item 1, given p1 , p2 (0, 1) with p1 < p2 , we let
A i = { x R | F ( x ) p i }, i = 1, 2.
Clearly, x A2 implies x A1 . Thus A2 A1 . Consequently, Q( p2 ) Q( p1 ).

To show item 2, we need to show that > 0, > 0, such that
0 < p0 p < implies 0 Q( p0 ) Q( p) < .
In fact, for a fixed p0 , let x = Q( p0 ). Given > 0, let x = x and p = F ( x ).

Clearly, p < p0 by the definition of Q and property of F. It follows from item 1
that for = p0 p > 0,
x < Q( p) Q( p0 ) whenever p ( p0 , p0 ).
This implies that 0 Q( p0 ) Q( p) < x x = .

Now we summarize the above intuition of FX and FX1 as follows:

A distribution function is nondecreasing and continuous from the right.
A quantile function is nondecreasing and continuous from the left.
Example 4.12. In statistics, if x = ( x1 , x2 , . . . , xn ) represents the observed values

of a sample of size n corresponding to a random variable X, the order statistic
of rank k, denoted by x(k) , is the k-th smallest value in the data set x. That is,
x (1 ) x (2 ) x ( n ) .
The (sample) quantile of order p = 12 is known as the (sample) median of X.

The (sample) quantile of order p = 14 is known as the first (sample) quartile of
X. Similarly, the (sample) quantile of order p = 34 is known as the third (sample)
quartile of X.
The quantile function is a useful tool in dividing ordered (from lowest to
highest) data (i.e., order statistics) into finitely many essentially probability-
wise equally sized data subsets (see Exercise 4.7 on page 200). In this sense, the
term quantile is synonymous with percentile.
As the data size n increases, a natural question to ask is what are asymp-
totic behaviors of sample quantile functions. We refer the reader with sufficient
statistic backgrounds to the literature (e.g., Ma, Genton, and Parzen [29]).
Example 4.13. In modern financial theory, the concept of quantile can be used
as a measure of the downside portfolio risk. If X represents the possible loss on
a portfolio, this measure is determined by a prescribed p-quantile (e.g., p = 1%)
of X such that the likelihood of X (i.e., loss in dollar amount) to take on a value
larger than that p-quantile is less than probability p (i.e., 1% chance).
Such a measure of the downside risk is called VaR, an abbreviation for value-
at-risk, which will be introduced shortly.

The next remark is for the reader interested in statistical importance and
applications of quantile functions.
Remark 4.8.
1. The inverse relation between the quantile function and the cumulative dis-
tribution function makes the quantile function one of basic concepts used to
describe the probability distribution of a random variable. Since the quan-
tile function also plays an essential role in the concept of mid-distribution,
which is important for discrete distributions, the quantile function is espe-
cially important for sample distribution functions (therefore for statistical
data modeling).
2. Regression analysis is a way to determine whether or not there is a correla-
tion between two or more variables and how strong any correlation may
be. Quantile regression is a type of regression analysis used in statistics
and econometrics (e.g., detection of heteroscedasticity). Both statistics and
econometrics are employed in mathematical finance. Just as the method of
least squares enables one to estimate models for conditional means, methods
of quantile regression enable one to estimate models for conditional quantile
functions (e.g., conditional median function).
For extensive discussions on the utility of quantile functions in statistical
applications, we refer the reader to the literature (e.g., Gilchrist [21]).

4.2.5 Value-at-Risk
Value-at-risk (VaR) and conditional value-at-risk (CVaR) are two risk measures
widely used by financial institutions and financial regulators. The latter may
be viewed as an extension of, or complement to, the former.
We focus on a basic understanding of the concepts of these two measures
in Sections 4.2.5 and 4.2.6 and explore a deeper issue behind these concepts in
Section 4.2.7.
For the sake of convenience and clarity of notation, we begin with the fol-
lowing notational remarks:
Let X represent data of (random) returns of a portfolio. By saying that X

is in the form of profit/loss (P/L), we mean that the data assign positive
values to profits and negative values to losses. In a similar fashion, by say-
ing that X is in the form of loss/profit (L/P), we mean that the data assign
positive values to losses and negative values to profits.
In studying VaR and CVaR in relation to financial risk management, it is
more convenient to define X in L/P form because VaR and CVaR as mea-
sures of portfolio risk are denominated in loss terms.
Definition 4.4. Given p (0, 1), the value-at-risk of a random variable X for the
level of probability p is denoted by VaR p ( X ) and defined by
VaR p ( X ) = FX1 ( p). (4.18)
That is, VaR p ( X ) = min{ x | FX ( x ) p} = Q( p), the lowest p-quantile of X.

Depending on the interpretation of X, there are different versions17 of the
notion VaR of X. In the above definition we interpret X as the possible loss in
dollar amount of a portfolio of securities (i.e., X in L/P form).
Assuming normal financial market conditions, under mark-to-market ac-
counting18 and single-period framework (see Chapter 3), for a given portfolio,
a given holding period (time horizon) and a given probability p (0, 1), the
VaR (often referred to the p-VaR) is a threshold loss value such that the proba-
bility that the loss on the portfolio over the given holding period exceeds this
value is p.
Example 4.14. A stock portfolio with a one-day p-VaR = 20, 000, where p = 1%,
is interpreted as that there is a 0.01 probability that the portfolio will lose more
than $20, 000 on a day if there is no trading during the day.

Example 4.15. Consider a portfolio of a single asset. Suppose that the return
of the asset is normally distributed with mean return of 14% per annum and
annual standard deviation of 35%. The value of the portfolio today is $100,000.
With 1% probability (equivalently, a 99% level of confidence), what is the max-
imum loss at the end of the year (equivalently, what is the annual VaR)?
Solution. Let X be the annual return of the portfolio (in P/L form). Then X is
a normal random variable with
17 One of those versions is given as follows:

Let X be (random) returns of a portfolio in P/L form, the VaR of X at confidence level (1 p)100%
is defined by
VaR p ( X ) = min{ x | P( X x ) p}.
That is, the VaR at tail probability p is the negative of the lower p-quantile of the return distribution.
18 Mark-to-market accounting is an accounting process by which the price of an asset held in an
account is valued each day to reflect the daily closing price of the asset.
Fig. 4.4 The probability distributions for X1% and Z1% are given by the left corners in the top and
bottom plots, respectively
E ( X ) = 0.14 $100, 000 = $14, 000,

"
Var( X ) = 0.35 $100, 000 = $35, 000,
and
X $14, 000
Z= , ( Z1% ) = 1%.
$35, 000
We illustrate X and Z in Figure 4.4.
A software can be used to obtain Z1% . However, lets use a linear interpola-
tion as in Figure 4.5, which is a plot of the line segment,
y2 y1
y y1 = ( Z Z1 ).
Z2 Z1
We need to find the value Z1% satisfying ( Z1% ) = 0.01. Since
(2.33) = 0.0099, (2.32) = 0.0102,
Fig. 4.5 Linear interpolation

we have
Z1 = 2.33, Z2 = 2.32,
y1 = 0.0099, y2 = 0.0102.
Solving for Z when y = 0.01 yields the desired value:

Z2 Z1
Z1% = Z = Z1 + ( y y1 )
y2 y1
0.01
= 2.33 + 0.0001
0.0003
= 2.32667.
From
X = $35, 000 Z + $14, 000,
we then obtain
VaR1% ( X ) = $35, 000 (2.32667) + $14, 000 = $67, 433.45.

In the next example, the historical performance of a portfolio return is de-
scribed by a frequency distribution.
Example 4.16. Suppose that the historical data of a portfolio shows the weekly
returns over the past 750 weeks in the table below.
Percentage Gain/Loss Number of weeks Other information
(Frequency)
< 5 2
[5, 4.5) 2 Suppose the
[4.5, 4) 0 seventh-highest
[4, 3.5) 3 weekly loss is 3.6%
[3.5, 3) 1
.. ..
. .
Find the weekly loss that will not be exceeded in 99% of cases. Assume that
the initial investment is $100,000.
Solution. One percent of observations is 750 1% = 7.5. Not exceeded in 99%

of cases implies that we need to consider only the seven lowest observations.
From the table, the seventh-highest weekly loss is 3.6% of the initial invest-
ment. Hence, the weekly VaR of the portfolio is
0.036 $100, 000 = $3, 600
Equivalently, $3,600 is the weekly loss that will not be exceeded in 99% of cases.

The next property provides a relation between VaRs under two different in-
terpretations of random variable X (e.g., P/L vs L/P). The proof of it provides
strategy and tactics in proving (4.20).
Property 4.2.
VaR p ( X ) = VaR1 p ( X ), p (0, 1). (4.19)
Proof. Notice that (4.19) is equivalent to
FX1 ( p) = F
1
X (1 p ), p (0, 1),
and (4.16) is equivalent to
P ( X < a ) p P ( X a ).
Let Y = X, x R and y = x. It is sufficient to show that for each fixed

p (0, 1), x is a p-quantile of X if and only if y is a (1 p)-quantile of Y. In
fact,
X<x iff X > x
implies
P ( X < x ) = P ( X > x ) = P(Y > y) = 1 P (Y y).
Similarly,
Xx iff X x
implies
P ( X x ) = P ( X x ) = P (Y y) = 1 P (Y < y).
Therefore,
P(X < x) p P(X x) iff 1 P (Y y ) p 1 P (Y < y ) ,
which is equivalent to
P(X < x) p P(X x) iff P (Y < y ) 1 p P (Y y ) ,
which implies the desired result.

By applying arguments similar to that in the proof of Property 4.2, we can

show that for any constant c = 0 and b,
.
c VaR p ( X ) + b if c > 0
VaR p (cX + b) = (4.20)
c VaR1 p ( X ) + b if c < 0.
Verification of (4.20) is left as an exercise to the reader. (Hint: Let Y = cX + b

and y = cx + b, and consider case 1, c > 0, and case 2, c < 0 separately.)
Remark 4.9.
1. There are three basic methods for calculating VaR:
a. The historical method.

b. The variance-covariance method.
c. The Monte Carlo simulation method.
Without getting into technical details, we note that the historical method in-
volves historical simulation and is a nonparametric approach, the variance-
covariance method involves estimation of the standard deviation and is a
parametric approach, and the Monte Carlo simulation involves generation
of time series such as (sample) paths of security prices, etc. For sophisti-
cated examples and detailed and comprehensive treatments of VaR, readers
are referred to the literature (e.g., Dowd [15], Hull [25], and Jorion [27]).
2. While VaR is conceptually simpler to understand and operationally easier
to implement than most other risk measures, it can provide a false sense of
security if it is misused due to its lack of subadditivity and other limitations
(see Section 4.2.7 and the literature, e.g., Artzner, Delbaen, Eber, and Heath
[4]; Follmer and Schied [20]).

4.2.6 Conditional Value-at-Risk
We begin this section with a natural question:

If a VaR of a portfolio is $10, 000, when the portfolio loses more than
$10, 000, what is the expected loss (over the corresponding time horizon)?
The answer to this question is provided by the CVaR of the portfolio.
Definition 4.5. Given p (0, 1), the conditional value-at-risk of a random vari-
able X for the level of probability p is denoted by C VaR p ( X ) and defined by
C VaR p ( X ) = E ( X |X VaR p ( X )). (4.21)
The conditional value-at-risk (C VaR) is also called the average value-at-risk

(A VaR) for the following obvious reason:
C VaR p ( X ) = E ( X |X VaR p ( X ))
1 1 1
1 1
= x dFX ( x ) = FX1 (y) dy
1p
FX ( p )
1 1 p p
1 1
1
= VaRy ( X ) dy.
1p p
Indeed, if X represents the possible loss of a portfolio, then the last expression
represents the average of the VaRs on the losses in the tail, which are larger
than VaR p ( X ). This explains why notations C VaR p ( X ) and A VaR p ( X ) are of-
ten interchangeable and why C VaR is also called expected tail loss, tail VaR , or
expected shortfall.
Example 4.17. The possible return information of a portfolio is given in the

table below:
X 70 30 0 30 70
p 10% 20% 40% 20% 10%
where X represents the annual return of the portfolio (in P/L form). The table
indicates that, for instance, if the unit on X is one thousand dollars, the chance
of the portfolios losing $70 thousand per year is 10%.
Determine C Var p ( X ) for the cases p = 10% and p = 20%.
Solution.
70 10
C Var10% ( X ) = 10
100
= 70,
100
70 100 30 100
10 10
C Var20% ( X ) = 20
= 50.
100

Remark 4.10. CVaR satisfies the subadditivity property, which is not valid for
ordinary VaR. For further discussions of the CVaR and its applications, we re-
fer the reader to the literature (e.g., Follmer and Schied [20]; Goldberg, Hayes,
Menchero, and Mitra [23]; Goldberg and Hayes [24]; Rockfellar and Uryasev
[32, 33]).

4.2.7 Coherent Risk Measures
From the standpoint of practical application (especially from an institutional

perspective), risk has many different dimensions that not only should have
exposure and uncertainty components but also should include sources and
decision-making inputs.
Just as a coherent and integrated air quality measure needs monitoring sta-
tions to measure the presence of contaminants in the air such as carbon monox-
ide, ozone and particulate matter, and so on, a coherent financial risk measure
needs to satisfy a set of properties that covers a number of different dimen-
sions of risk as briefly mentioned above. A proposal of such a set of properties
with mathematical clarity is postulated by Artzner et alia [3, 4] and described
in the definition below.
Definition 4.6. Let random variables X and Y represent two portfolio returns.
A coherent risk measure for portfolio return is a function that satisfies each of
the following properties with probability 1:
1. Monotonicity: If X Y, then ( X ) (Y ).
2. Subadditivity: ( X + Y ) ( X ) + (Y ).
3. Positive homogeneity: (c X ) = c ( X ), for any c (0, ).
4. Translational invariance: ( X + b) = ( X ) b, for any b (, ).
Intuition and interpretation.

1. Monotonicity: The higher the future return is, the smaller the risk is.
2. Subadditivity: The sum of sub-portfolio risks is an upper bound of the total
portfolio risk. A practical application could be for decentralized decision-
making in a financial institution based on an upper bound on risk prescribed
by regulation.
Notice that subadditivity provides an incentive to diversify a portfolio.
3. Positive homogeneity: The risk of a position is proportional to the size of the
position.
4. Translational invariance: Cash needed is considered a risk measure. Thus,
margin account requirements and margin call concerns have been factored
into a coherent risk measure.
Note that the gist of the definition of coherent risk measure is subadditivity.
The rest of the properties are designed to ensure that portfolio returns under
consideration are essentially well behaved.
Rockafellar and Uryasev [32] showed that the CVaR is a coherent risk measure.
On the other hand, the VaR has been criticized for violating subadditivity by
many, notably by Artzner et alia [4]. Thus, the VaR is a not a coherent risk measure.
The next example elaborates this point (for more examples, see the literature
Acerbi and Tasche [1], Acerbi et alia [2], Artzner et alia [4], and Dowd [15] for
original ideas).
Example 4.18. Suppose that we bought two securities which are issued by two
companies and have identical price movements. Let X and Y represent returns
from each of these two securities. Assuming that each company goes bankrupt
4.3 Introduction to Linear Factor Models 185
independently with probability 8%, and we lose $10, 000 if a company goes
bankrupt, and we lose $0 if no bankruptcy occurs. Therefore VaR90% ( X ) =
VaR90% (Y ) = 0, and we obtain
VaR90% ( X ) + VaR90% (Y ) = 0.
On the other hand, let Ai be the event that we lose $10, 000 i, i = 0, 1, 2, a
straightforward elementary probability calculation establishes that
P ( A0 ) = 0.92 = 0.81 = 81%,

P ( A2 ) = 0.12 = 0.01,
P ( A1 ) = 1 0.81 0.01 = 0.18,
which implies Var90% ( X + Y ) = 10, 000. The fact that
Var90% ( X + Y ) = 10, 000 > 0 = VaR90% ( X ) + VaR90% (Y )
indicates that the VaR does not hold subadditivity for all possible random vari-
ables. Again, we note that subadditivity provides an incentive to diversify a
portfolio, which the VaR clearly discourages in this case.

Remark 4.11. For conditions under which the VaR becomes subadditive, we re-
fer the reader to the literature (e.g., Danelsson et alia [12] for subadditivity for
VaR in the tail and Dhaene et alia [14] for risk measures and comonotonicity).

4.3 Introduction to Linear Factor Models
Pioneered by Charles Spearman [38], an English psychologist, factor analy-

sis has more than 100 years of history and has become a principal statistical
method of investigation in almost all scientific, economic, and financial fields
today. In this section, we briefly introduce linear factor models in a context
where only financial investment returns are under consideration.
Applying linear factor models to finance in a multifactor setting was in-
troduced in 1976 by Stephen Ross [34] through his work on arbitrage pricing
theory (APT). While two major applications of linear factor models to portfolio
management are asset allocation and risk management, we briefly discuss the
latter only (despite the interconnect between these two areas).
For a systematic study of factor models for their role in portfolio risk analy-
sis, we refer the reader to the book by Connor, Goldberg, and Korajczyk [11].
4.3.1 Definition and Intuition
Let P = {S1 , S2 , . . . , Sn } be a portfolio of n assets (which can be stocks, bonds,

mutual funds, and so on). Let f1 , . . . , f m represent m fundamental risk factors
associated with portfolio P.
Definition 4.7. A (linear) factor model relates the return19 of an asset of the port-
folio to the values of a limited number of factors, say f j , j = 1, 2, . . . , m, and is
represented by
Ri = i1 f1 + i2 f2 + + im f m + ei , (4.22)
where:
Ri is the return on asset i, where i = 1, 2, . . . , n.
ij is the change in the return on asset i per unit change in factor j, where
i = 1, 2, . . . , n and j = 1, 2, . . . , m. (In other words, ij indicates the sensitivity
of asset i to factor f j .)
ei is the portion of the return on asset i not related to the m factors. The
quantity ei is called the idiosyncratic return of asset i.
Note that f j (j = 1, 2, . . . , m) and ei (i = 1, 2, . . . , n) are random variables and that

ij (i = 1, 2, . . . , n, j = 1, 2, . . . , m) are constants. For simplicity, we shall refer to
linear factor models simply as factor models.
Often we write
e i = i + i ,
where i = E (ei ) and i is the error term.20 It follows that a factor model can
be expressed by
m
Ri = i + ik fk + i , i = 1, 2, . . . , n, (4.23)
k =1
where i is called the alpha value (or simply, alpha) of asset i, ik s are called
the beta values (or simply, betas) of asset i, and we assume that Cov( i , i ) = 0
whenever i = i , and that for each i, Cov( i , f j ) = 0 for all j.
Example 4.19. To make the concept of factor models easier to understand, lets
consider a single stock portfolio. Let f j , j = 1, 2, 3, 4, 5 represent house price-
to-income ratio, new home sales, housing market index, biotech ETF perfor-
mance, and FDA fast track development program, respectively. The following
is a five-factor model
19 Annualizing returns with compounding would make these factor models almost useless because
of the linearity of the model. As we pointed out in Remark 4.1, using the logarithmic return in fac-
tor models can avoid this shortcoming and take advantage of time-additivity and statistical tools in
studying properties of individual securities.
20 Error terms are usually assumed to be normal with mean zero.
R = + 1 f1 + 2 f2 + 3 f3 + 4 f4 + 5 f5 + .
A least squares linear regression fit will determine whether or not the model
is meaningful in practice.

Because we imposed few conditions on the factors (for instance, no probabil-

ity spaces specified or distributions required), the definition above is simple;
however, in practice the redundancy in the factors may make our already
cloudy data points cloudier and create an unnecessarily more complicated and
prolonged computational process. The Principal Component Analysis (PCA) is a
powerful factor redundancy reduction procedure that compresses the data set
by keeping only important information that is represented by a likely smaller
number of orthogonal and centralized new factors.
In general, a simpler model is less prone to overfitting. While the topic is
beyond the scope of this book, the next example provides an intuition of it.
Example 4.20. Revisit our example above. Recall the model with five factors
R = + 1 f1 + 2 f2 + 3 f3 + 4 f4 + 5 f5 + ,
where f j , j = 1, 2, 3, 4, 5 correspond to the following factors: house price-to-

income ratio, new home sales, housing market index, biotech ETF perfor-
mance, and FDA fast track development program, respectively.
Now, suppose that the (sample) correlation matrix (with artificial entries) of
the random vector f T = [ f1 f2 f3 f4 f5 ] is below:

1.00 0.82 0.93 0.01 0.02
0.82 1.00 0.91 0.02 0.00

0.93 0.91 1.00 0.00 0.01 .

0.01 0.02 0.00 1.00 0.89
0.02 0.00 0.01 0.89 1.00
The matrix suggests two underlying constructs: housing (related to f1 , f2 , and

f3 , and so to their linear combinations) and the biotech sector (related to f4 and
f5 , and again, to their linear combinations). In other words, the above correla-
tion matrix displays the redundancy of the five factors described above. There
is an obvious computational advantage if f1 , f2 , and f3 are collapsed into a
single new factor f1 corresponding to housing, and f4 and f5 are collapsed
into another single new factor f2 corresponding to biotech sector. We can then
form a factor model with only two factors
R = + 1 f1 + 2 f2 + .
Intuitively, what we illustrated in this example is in essence what the PCA

accomplishes.

To make it more precise, PCA converts the original factors into uncorrelated
linear combinations of them. These linear combinations form an orthonormal
set of eigenvectors of the (sample) correlation matrix of factors. Very often in
practice, a few leading eigenvalues of the correlation matrix explain, in terms
of percentage, close to the total variation in the entire data set. Therefore, the
original model may be reduced to a new model with fewer uncorrelated fac-
tors which are the eigenvectors belonging to those leading eigenvalues. Such
reduction, in practice, not only reduces computation but also often results in
more robust models with higher sustainability for the future.
This discussion leads to another statement of the definition of linear factor
model:
Definition 4.8. A (linear) factor model relates the return of an asset of the port-
folio to the values of a limited number of factors, say fj , j = 1, 2, . . . , m, and is
represented by
m
Ri = i + ij fj + i , i = 1, 2, . . . , n, (4.24)
j =1
where:
Ri is the return on asset i, i = 1, 2, . . . , n.
E ( fj ) = 0, j = 1, 2, . . . , m (i.e., fj , j = 1, 2, . . . , m are centered).
fj , j = 1, 2, . . . , m are orthonormal factors under covariance (i.e., Cov( fj , fj ) =
0 whenever j = j , and Var( fj ) = 1 for all j).
ij indicates the sensitivity of asset i to factor fj and is called the factor loading
of return Ri .
Cov( i , i ) = 0 whenever i = i and that for each i, Cov( i , fj ) = 0 for all j.
i + i is the portion of the return on asset i not related to the m factors (thus
it is called the idiosyncratic return of asset i), where E ( i ) = 0.
The conditions imposed on the quantities in Definition 4.8 ensure that (4.24)
can be estimated by the method of least squares (see Exercises 4.30 and 4.31 on
page 206). Note that the alphas and betas are both measures of risks and tools
used to determine the risk-reward profile of a portfolio.
Observe that (4.23) shows that a factor model decomposes an assets return
into factors common to all assets in the portfolio and an asset-specific factor.
Even though this is a considerable simplification of reality, it is computation-
ally reasonable. Factor models are practically useful in many domains in the
field of investments, particularly in analyzing historic results, because they
provide a tool to allow analysts to separate components of the overall return
of the asset.
Indeed, factor models provide a versatile tool for an elegant framework of

trading strategy analysis and investment portfolio design. However, finding
good factors is always a challenging task. Almost none of the published mod-
els remain profitable (as of the writing of this text), as the market is becoming
much more efficient than 10 years ago. The idea of studying these models is
to help readers to develop their own models in the future. In practice, none of
these models will be applied alone, and almost certainly not in the exclusive
form provided here.
4.3.2 Portfolio Variance Decomposition
Under the single time period framework, we rewrite (4.23)

m
Ri = i + ik fk + i , i = 1, 2, . . . , n,
k =1
in a matrix form
X = + f + ,
where
X = [ R1 , R2 , . . . , R n ] , = [ 1 , 2 , . . . , n ] , i = [ i1 , i2 , . . . , im ],

f = [ f1 , f2 , . . . , f m ] , = [ 1 , 2 , . . . , n ] , = [ ik ]nm .
Under the assumptions that
E ( f k ) = 0, k = 1, 2, . . . , m, E ( i ) = 0, i = 1, 2, . . . , n,
Cov( i , j ) = 0 whenever i = j, Cov( i , f k ) = 0 for all i and k,
we compute
Cov( Ri , R j ) = E (( Ri E ( Ri ))( R j E ( R j ))) = E (( Ri i )( R j j ))

m m
=E ik fk + i jl fl + j
k =1 l =1

m m m m
=E ik fk jl fl + ik E( fk j ) + jl E( i fl ) + E( i j )
k =1 l =1 k =1 l =1

= E(i ( f f ) j ) + E ( i j ) = i E( f f )
j + E ( i j ).
This result suggests a relation between covariance matrices. To see this, we let
= [ij ]nn be the covariance matrix of a portfolio return X,

= [ f ij ]mm be the covariance matrix of factors in the model,
= [ ij ]nn be the covariance matrix of the error terms,
where
ij = Cov( Ri , R j ), f ij = Cov( f i , f j ), ij = Cov( i , j ).
Then we have
.
0 if i = j
= E ( f f ), Cov( i , j ) = E ( i j ) =
Var( i ) if i = j,
and the covariance matrix of the returns can be expressed by
= + . (4.25)
It is readily understood that this relation of covariance matrices implies a de-

composition of portfolio variance into common factor variance and idiosyn-
cratic variance (see Exercise 4.24 on page 205).
Note that given underlying factors, a portfolio return, and its volatility are
completely determined by its factor loadings.
Remark 4.12.
1. Different portfolios may be exposed to different types of risk when different
scenarios occur in the market. Intuitively speaking, portfolio risk manage-
ment is about what to anticipate in dealing with different market scenarios.
The additive decomposition in (4.25) is a significant result (or rather a key
objective of factor models) for portfolio risk management. It suggests that
factor models allow portfolio managers to perform risk management by ap-
proaching these very underlying factors and examining how they impact
covariance matrix of returns in a direct way.
2. Achieving optimal asset allocation requires a robust understanding of port-
folio risk. To pass the test for robust understanding, one needs to quantify
portfolio risk. Quantitative portfolio risk management requires quantifying
the overall portfolio risk (e.g., VaR and CVaR), and slicing and dicing into
sources of portfolio risk (e.g., multifactor models). In this sense, factor anal-
ysis refines the risk profile of the portfolio and allows portfolio managers to
perform allocation and hedging from factor risk perspective.
For portfolio optimization, we refer the reader to the literature (e.g., Chan,
Karceski, and Lakonishokk [6]; Connor and Korajczyk [8, 9]).

4.3.3 Factor Categorization
Generally speaking, financial markets will reflect the economic conditions of

an economy. Empirical evidence for this can be found in the literature (e.g.,
Chen, Roll, and Ross [7]).
The field of economics can be divided into many subfields depending on
how finely one wishes to make the division. The terms macroeconomics and
microeconomics were coined in 1933. As the names suggest, the former studies
how the economy as a whole works, whereas the latter studies how specific in-
dividual units function. Examples of major factors affecting macroeconomics
are the rate of change of the gross domestic product (GDP), unemployment
rate, and inflation (see Section 1.3), and examples of factors affecting microe-
conomics are companies size, dividend yield, and price-to-earnings ratio (see
Section 4.3.6).
Factors can be divided into observable21 and unobservable (or latent22 ).
Example 4.21. A publicly traded companys market capitalization and indus-

try classification are observable factors.
Factors that are determined by PCA may be unobservable or latent (see f1
and f2 in Example 4.20 on page 187).

Depending on the different factor types and factor construction methods,

multifactor models of security returns may be categorized by macroeconomic,
fundamental, and statistical models:
A macroeconomic factor model is a factor model whose common risk factors
are determined by observable macroeconomic factors.
A fundamental factor model is a factor model whose common risk factors
are created from stock-specific fundamentals (e.g., company size and price-
to-market ratio) that affect the corresponding returns.
A statistical factor model is a factor model whose common risk factors are
extracted from historical returns by using analytical methods (e.g., PCA).
Note that such a division of factor models has blurred boundaries. Fama-
French three-factor model will be introduced in Section (4.3.6) and can be
viewed as a fundamental-based factor model or a combination of fundamental
and macroeconomic factor model.
For an overview of the empirical procedures for the three types of fac-
tor models including inputs, outputs, and estimation technique, we refer the
reader to Table 1 in [10] by Connor.
21
Observable variables, as a statistical term, are those that can be directly measured.
22
Latent variables, as opposed to observable variables in statistics, are those inferred through mathe-
matical models and cannot be directly observed.
4.3.4 Alpha and Beta
In order to understand the significance of the factor model in relation to in-

vestments, lets consider its simplest form (with n = m = 1).
R = + R M + , (4.26)
where R M is the stock market return, say, the return of the S&P 500 over time
period [0, T ], and R is the return of a stock over the same time period. Here
and are referred as the the stocks and . (If R is the return of a portfolio,
then and are referred as the the portfolios and .) Taking the expectation
of both sides of (4.26) yields
E ( R) = + E ( R M ).
Accordingly, we have
R = + R M , (4.27)
where R and R M are sample means of R and R M , respectively.
In general, and are more stable than the equity price itself. Consequently,
they provide a certain predictive value. In other words, they are not only mea-
sures of risk but also tools used to determine the risk-reward profile of a port-
folio to form investment strategies.
Example 4.22. The mathematical expression of the single-factor model seen in
(4.27) suggests the following possible investment strategies:
a) Getting stock returns from both and . Perhaps a number of investors do
so without knowing the formal meanings of and .
b) Making stock investment profits based on a better judgment of the direction
of major stock market (say, the S&P 500). If one can do it well, then with a
certain level of leverage, one can make good returns on high stocks.
c) Making stock investment profits based on selecting high stocks. In fact,
many quantitative funds try to use index futures positions to hedge away
the beta part to obtain so-called market-neutral portfolios.

Before we illustrate how to determine and given the historical data on
R and R M in (4.27), let us understand the general idea of the method of least
squares, which is a classical technique in finding the approximate solution
of an overdetermined23 system of linear equations. Such systems often occur
based on raw field data and usually have no solutions.
Let f i , where i = 1, 2, , n, be n functions of m variables x1 , x2 , , xm ,
where n > m. Suppose that we are interested in solving for m unknowns
23 A system of linear equations is called overdetermined if there are more equations than unknowns.
x1 , x2 , , xm from the system of n equations:
f1 ( x1 , x2 , , xm ) = 0,
f2 ( x1 , x2 , , xm ) = 0,

f n ( x1 , x2 , , xm ) = 0.
Note that the system holds if and only if the equation
n
( fi ( x1 , x2, , xm ))2 = 0
i =1
holds. Unfortunately, in general, the system does not have a solution. What we
intend to do is to find points ( x1 , x2 , , xm ) that minimize
n
( fi ( x1 , x2, , xm ))2 .
i =1
Example 4.23. To simplify the notations in (4.27), we let y = R and x = R M .

Rewrite (4.27) into y = + x. Given n sets of data ( xi , yi ), i = 1, 2, , n, where
n > 2, determine and .
Solution. Consider the following overdetermined system with and as un-

knowns and n > 2:
y1 = + x1 ,
y2 = + x2 ,

yn = + xn .
By the above discussion, we need to minimize
n
L= (yi xi )2 .
i =1
We compute
n n
L
= 2 (yi xi )(1) = 2 (yi xi ),
i =1 i =1
n n
L
= 2 (yi xi )( xi ) = 2 ( xi yi xi x2i ).
i =1 i =1
Setting24
24 The solution to this system can only be minimum point(s) since L does not have maximum point.
L L
= 0, = 0,

we obtain
n n n
0= (yi xi ) = yi n xi ,
i =1 i =1 i =1
n n n n
0= ( xi yi xi x2i ) = xi yi xi x2i .
i =1 i =1 i =1 i =1
Equivalently,
1 n 1 n
= yi xi ,
n i =1 n i =1

n n n
xi yi = xi + x2i .
i =1 i =1 i =1
Straightforward algebraic manipulations yield
ni=1 yi (ni=1 xi ) n ni=1 xi yi ( ni=1 xi ) (ni=1 yi )

= , = . (4.28)
n n ni=1 x2i (ni=1 xi )2

As a matter of fact, what we have just accomplished is a procedure to de-

termine a line that best fits the data {( xi , yi ), i = 1, 2, , n}. This procedure is
called the method of least squares. The best25 line to fit the data is the line with
the slope and y-intercept as we derived above. More rigorously speaking,
the least squares approximate solution of an overdetermined system of linear
equations should be referred to as the best fit graph26 to data.
Remark 4.13.
1. Although a stocks and are more stable than the stock price in general,
they change over time as the risk profile of the company changes. While the
discussion of their (random) variability is beyond the scope of this book, we
would like to point out that sometimes a stocks and may become quite
unstable when the volatility of the market is high.
2. The calculation of the actual values of a stocks and depends on several
factors including the length of the time period in the data sampling and the
frequency of the data sampling (see Exercise 4.22 on page 204). For instance,
consider a stock performance over a period of 3 years. If wk and mth de-
note the stocks calculated based on weekly returns and monthly returns,
25The best line to fit the data is in the sense of the least Euclidean distance.
26 For example, if the data space is R2 , and the unknown space is R3 , say y = + x + x2 , then
the best fit graph is a curve in the data space. If both the data space and unknown space are R3 , say
y = + x2 + z2 , then the best fit graph is a surface in the data space.
respectively, then wk may be different from mth . Also, a stocks calcu-

lated based on a 3-year period is likely to be different from that based on a
5-year period. Similar results apply to a stocks .
3. Most academics use log returns to calculate and since log returns are
additive. In practice, the difference between the log return and (total) return
is small unless the return rates are large.

4.3.5 CAPM Beta Versus Linear Factor Beta
We have the quantity beta in the CAPM:
E ( R r) = E ( R M r), (4.29)
where R M is the return of a market portfolio. Expression (4.29) is equivalent to
R r = ( R M r) + 1 , (4.30)
where the error term 1 satisfies E ( 1 ) = 0.

On the other hand, there is also a beta in the linear factor model in the form of
excess returns:
R r = + ( R M r) + 2 , (4.31)
where the error term27 2 satisfies E ( 2 ) = 0. Since a portfolio or security beta
plays an important role in investment strategy, one should be clear about the
beta value under the consideration. We would like to point out that these two
betas are different unless R r and R M r have zero means.
To verify the previous claim, consider a data set {( Ri , R M,i ), i = 1, 2, n}
of n observations. Plug it into (4.30) and let
yi = Ri r f ,i , xi = R M,i r f ,i ,
where R M,i and r f ,i are, respectively, the market return and risk-free rate for
the ith sample period. Also, let
x = [ x1 x2 . . . x n ] , y = [y1 y2 . . . y n ] , e = [1 1 . . . 1 ] .
We then obtain the system

y = x, (4.32)
which has the least squares solution
27 We assume that error terms are normal with mean zero.

xy n 1 xi yi
= = i= CAPM . (4.33)
xx ni=1 x2i
Similarly, plug the data set into (4.31). After eliminating from
y = e + x,
we obtain the system

y ye = ( x xe), (4.34)
which has the following least squares solution:
( x xe) (y ye) ni=1 ( xi x )(yi y)

= = LF . (4.35)
( x xe) ( x xe) ni=1 ( xi x )2
It follows that if x = 0 (i.e., E ( R M r) = 0) and y = 0 (i.e., E ( R r) = 0), then

by (4.33) and (4.35), we obtain CAPM = LF .
4.3.6 Fama-French Three-Factor Model
A well-known example of factor models is the Fama-French three-factor model.

It requires some basic understanding of the following financial terminologies:
The book value of a company is the total asset value of the company carried
on its balance sheet.28
Given a publicly traded company, suppose that it has a total number of N
shares outstanding. Let t0 be the current time, p the current stock price, and B
its current book value. Here p and B carry the same units, say, in dollars:
p is called the market value or market price at time t0 .
Np is called the market capitalization of the company at time t0 , which is the
product of the total number of its outstanding shares and its market value.
According to a companys market capitalization, stocks can be classified into
small-cap, mid-cap, and big-cap. Although the definitions of these categories
can vary across financial institutions, the classification ranges are generally
from $300 million to $2 billion, from $2 billion to $10 billion, and more than
$10 billion, respectively. In a similar way, stocks can also be further classified
into mega-cap, large-cap, mid-cap, small-cap, micro-cap and nano-cap.
28 A balance sheet gives a snapshot of the financial position of a company at a given time. Most account-
ing balance sheets consist of two sides: the left side indicates ASSETS (things that have value), and the
right indicates LIABILITIES (things that are owed to third parties) and STOCKHOLDERS EQUITY
(which is the value of the remaining assets if the company were to go out of business immediately).
For an oversimplified example, consider equity in your home = what you paid - what you owe (loan
remaining). It is called balance sheet because it has to balance between both sides: Assets = Liabilities
+ Stockholders Equity.
B/N
p is called the book-to-price ratio of the company at time t0 . It is also
known as book-to-market ratio.
P/B ratio or the price-to-book ratio is the reciprocal of the book-to-price ratio.
In practice, the book-to-price ratio is often calculated by dividing the latest

quarters book value per share by the current day closing price of the stock.
The same time frame/units apply to the calculations of market capitalization.
The book-to-price ratio is also used to identify undervalued (the ratio > 1) or
overvalued (the ratio < 1) stocks (although the practical definitions of under-
valued and overvalued can be in other relative terms).
Example 4.24. The key statistics of companies can be easily accessed online.
For instance, valuation measures for IBM provided by Yahoo! Finance on
November 8, 2012, indicated
Market Cap : 214.28B and Price/Book (mrq) : 10.3,
where mrq stands for most recent quarter. Given that the closing price of
the stock on that day was $190.10, determine the capitalization of IBM and the
book-to-price ratio of IBM on that day.
Solution. The market capitalization was 214.28 billion dollars and

1
book-to-price ratio = = 0.097.
10.3
IBMs P/B ratio = 10.3 simply means that the stock costs 10.3 times as much as
its asset could be sold for if the company were liquidated.

By definition, the P/B ratio provides a measure of the markets valuation of

a company in relation to the value of that company indicated on its financial
statements. In other words, a book-to-price ratio tells investors how the com-
panys stock value measures up to its book value. In this sense, the higher the
book-to-price ratio is, the higher the value investors get (assuming that the
fundamentals of a company are accurately and completely reflected on its bal-
ance sheet). By this measure, a stock at $1 per share may not be cheap, whereas
a stock at $100 per share may be a bargain.
The following terminologies are useful in describing additional returns:
Size premium is the additional return that investors receive by investing in
stocks of small-market-capitalization companies rather than in big-market-
capitalization companies:
size premium = RS R B .
Value premium is the additional return that investors receive by investing

in stocks of high book-to-price ratio companies to that of low book-to-price
ratio companies:
value premium = R H R L .
Both size premium and value premium are called risk premiums in the theory
of financial investments.
Intuitively, the stock price is expected to depend on market capitalization
(size), book-to-price ratio (value), and the systematic risk in stock investing
that is directly associated with the general stock market (e.g., S&P 500). Thus,
one natural approach to stock investments is to consider the general market,
size, and value as three risk factors. As a matter of fact, the factor models we
defined before are linear approximations of this dependence.
Although natural ideas do not always lead to successful investment strate-
gies, once they are quantified, statistical testing provides a mechanism for ac-
cepting or rejecting the ideas. It is in such frameworks that factor models can
come in handy.
Motivated by the desire to better explain differences in the returns of diver-
sified equity portfolios by asset pricing models, Fama and French created a
three-factor model [18] in the form of excess returns29 :
R r = + 1 ( R M r) + 2 ( RS R B ) + 3 ( R H R L ) + , (4.36)
which implies
E ( R r ) = + 1 E ( R M r ) + 2 E ( R S R B ) + 3 E ( R H R L ).
Denote E ( RS R B ) and E ( R H R L ) by SMB and HML, respectively. We ob-

tain
E ( R) = r + + 1 E ( R M r) + 2 SMB + 3 HML, (4.37)
where r is the risk-free rate, E ( R M r) is the expected excess return of the
general market, and SMB and HML represent the expected size premium and
expected value premium, respectively.
The regression test (4.36) has been widely applied to various equity markets
of both developed and emerging economies. The following example is to help
the reader to understand that the Fama-French three-factor model provides an
effective asset pricing mechanism in the US stock markets.
Example 4.25. Let S, MS , and B denote the sets of stocks of companies with
small, medium, and big capitalizations, respectively. Let L, M, and H denote
the sets of stocks of companies with low, medium, and high book-to-market
29Equation (4.36) is the form that most researchers currently use for the Fama-French three-factor
model. We refer the reader to [18] and [19] for the original form of this model.
4.4 Exercises 199
ratios, respectively. By taking the intersections among them as indicated in the

table below, we obtain nine portfolios.
Low b/m ratio Medium b/m ratio High b/m ratio

3 3 3
Small size S L S M S H
3 3 3
Medium size MS L MS M MS H
3 3 3
Big size B L B M B H
In the 1990s, Davis, Fama, and French performed empirical tests on the nine
portfolios above. Confirmatory factor analysis of the information in [13], which
includes data source, sample size, data organization, regression results, fac-
tor loadings, and R-square values, strongly supports that value and small-cap
stock portfolios outperformed markets on a regular basis in the USA from 1929
to 1996.

As a contrast to the CAPM which explains, on average, 70% of the variabil-

ity in well-diversified portfolios returns, the Fama-French three-factor model
explains over 90% of the variability (see Fama and French [17, 18]).
There are many more nice discussions in the literature related to the topics
presented in this chapter, e.g., [22, 26, 28, 30, 31, 35, 37, 39].
4.4 Exercises
4.1. State the definition of the market portfolio.
4.2. Explain how the sign of the beta of a stock indicates the direction of the
movement of the stock price with respect to that of the market portfolio.
4.3. Let A be a class of stocks with beta between 0.5 and 2. What is the main
property that all stocks in A have in terms of the market returns?
M r f
4.4. Give a financial interpretation of the mathematical expression: M .
4.5. Can you find an example of a company in the USA with a negative beta? If
so, do you think that they are as abundant as companies with a positive beta?
Explain.
4.6. Consider random variable X = 1, 2, 3 satisfying P ( X = i) = 13 , i = 1, 2, 3.

Find the quantile function of X.
4.7. Consider a random variable X = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 with p.d.f. given by

P ( X = i) = 11
1
for i = 7 and P (i = 7) = 11
2
. Find the quantile function of X.
4.8. Use the table given in Example 4.17 on page 183 to determine Var p ( X ) for
arbitrary p.
4.9. Use the table given in Example 4.17 on page 183 to determine C Var p ( X )
for p = 30% and p = 50%.
4.10. What does a negative VaR imply?
4.11. Returns and risks are two aspects involved in every investment. Identify
each statement below as true or false or identify scenarios when it is true and
when it is false. Justify your answer.
a) The quantity relates to factors affecting the performance of an individual
stock or the fund managers skill in selecting the stocks.
b) The factor relates an individual stock-to-market risks.
c) A higher stock and a lower stock would be preferred choices.
d) The quantity = 0 if a stock market is efficient.
4.12. Briefly justify your answer to each of the following:

a) Suppose that empirical evidence were sufficient to confirm the CAPM. What
would be an investment implication?
b) Suppose that extraordinary premiums from the (Fama-French) three-factor
investing during the period of 19261996 were to repeat today, what would
be an investment implication?
4.13. All investments carry some form of risk. Major risks include, but are not
limited to, the following:
a) Systematic risk
b) Interest rate risk
c) Liquidity risk30
d) Regulatory/political risk31
e) Leverage risk32
f) Credit risk
g) Currency risk
h) Counterparty risk33
Find an example for each type of risk listed above.
30 Liquidity risk is the risk that an investor cannot execute a buy/sell order in the market due to the
lack of anticipated/reasonable bid/ask spread or sufficient volume.
31 Regulatory changes or governmental policy changes may have significant impact on asset values.
Such risk can be either systematic risk or market risk.

32 Such risk is often associated with unexpected and unfavorable volatility.
33 Counterparty here means the other party in a financial transaction.
4.4 Exercises 201
4.14. Use the data in Table 4.1 below to compute the Sharpe ratio of the S&P
500 for the periods 1986 to 1999. Note that the risk-free rate is not constant.
Table 4.1 Annual return rate data from 1986 to 1999 for the S&P 500 and 1-year treasury bills. Data
Source: istockanalyst.com
Year S&P500 annual return 1-year T-bill rate
1986 18.82% 7.21%

1987 5.40% 5.46%
1988 15.99% 6.52%

1989 31.56% 8.37%
1990 -2.97% 7.38%
1991 30.51% 6.25%
1992 7.45% 3.95%
1993 10.09% 3.35%
1994 1.33% 3.39%
1995 37.28% 6.59%

1996 22.69% 4.82%
1997 33.60% 5.30%
1998 30.73% 4.98%
1999 21.10% 4.31%
4.15. Assume a risk-free rate of 1.5%. Answer the questions below using the
information in the following table:
Portfolio A B C D E F
Expected Return 3.2% 8.1% 9.8% 5.1% 10.7% 4.8%
Standard Deviation 2.7% 9.9% 13.7% 6.2% 17% 6.1%
a) Among the portfolios in the table, which one is closest to the market portfo-
lio? Justify your answer.
b) Plot the capital market line (CML) based on your answer in part (a).
c) For portfolio C, what is the portfolio risk premium per unit of portfolio risk?
d) Suppose we are willing to make an investment only with = 6.2%. Is a
return of 6.5% a realistic expectation for us?
4.16.
a) Given the information in the table below,
Stock Beta Expected Return

A 1.25
B 0.7
C -0.4
and assuming that r f is 3% and that the market return is 5%, find the ex-
pected returns for each stock listed in the table and plot them on an SML
graph.
b) Suppose the table below provides further information about the stocks in
part a):
Stock Current Price Expected Price Expected Dividend

A 29.5 21.5 0.71
B 47 49.75 1.85
C 35.4 38.7 1.05
Indicate your estimated returns on each stock on the graph from part a),
and decide your buy/sell/hold rating on each stock based on your graph.
Justify your decisions.
4.17. Suppose that we borrow an amount equal to 25% of our original wealth
at the risk-free rate 4.125%. Use the CML to find P and P .
4.18. Assume the risk-free rate is 1.5% and consider the information in the table
below:
Portfolio Expected Return Standard Deviation

A 3.2% 2.7%
B 8.1% 9.9%
C 9.8% 13.7%
D 5.1% 6.2%
E 10.7% 17%
F 4.8% 6.1%
a) Which of these six portfolios offers investors the best combination of risk
and return? Justify your answer from a capital market perspective.
4.4 Exercises 203
b) Use the formula

P = w0 r f + (1 w0 ) M
to determine your investment asset allocation.
c) If you plan to invest $100, 000, what is your investment strategy based on
the information given in this exercise?
4.19. Given the following information, find the securitys beta and expected
return:
a) The risk-free rate is 1.72%. The market portfolio has its standard deviation
as 15.92% and its expected return as 5%. The covariance of the security with
the market is 0.04.
b) The risk-free rate is 2%. The market portfolio has its standard deviation as
14% and its expected return as 11%. The security is uncorrelated with the
market and has a standard deviation of 39.7%.
c) The risk-free rate is 1.72%. The market portfolio has its standard deviation
as 15.92% and its expected return as 5%. The covariance of the security with
the market is 0.04.
4.20. The ticker symbol for the Goldman Sachs Group is GS. The table below
provides the daily closing prices for GS and S&P 500 index on the trading
days during the period between June 30 and July 14 of 2011. Let r and r M be
the daily log returns of GS and S&P 500 index, respectively. Using the model
r = + r M , determine GSs and for the period.
Date GS S&P 500

June 30 $133.09 $1320.64
July 1 $136.65 $1339.67
July 5 $134.5 $1337.88
July 6 $133.89 $1339.22
July 7 $135.01 $1353.22
July 8 $134.08 $1343.8
July 11 $132.02 $1319.49
July 12 $130.31 $1313.64
July 13 $129.7 $1317.72
July 14 $129.89 $1308.87
4.21. The ticker symbols (stock symbols) for the Goldman Sachs Group and
SPDR S&P 500 ETF are GS and SPY34 , respectively. The table below provides
34SPDR (Spiders) is a short form of Standard & Poors depositary receipt, an exchange-traded fund
(ETF) that tracks the Standard & Poors 500 Index (S&P 500). Each share of SPY contains one-tenth
the daily closing prices for GS and SPY on the trading days during the period
between June 30 and July 14 of 2011. Let P be a portfolio consisting of longing
200 shares of SPY and shorting 100 shares of GS. Suppose that r f = 0. Find the
maximum drawdown and the Sharpe ratio for the portfolio in the time period
indicated in the table.
Date GS SPY
June 30 $133.09 $131.97
July 1 $136.65 $133.92
July 5 $134.5 $133.81
July 6 $133.89 $133.97
July 7 $135.01 $135.36
July 8 $134.08 $134.4
July 11 $132.02 $131.97
July 12 $130.31 $131.4
July 13 $129.7 $131.84
July 14 $129.89 $130.93
4.22. Although most stocks and can be found online, the actual values of
and for the same stock may be different at different sites. Besides, how the
actual values are calculated might be considered as proprietary information.
Thus, it is critical to understand the factors that affect the calculations.
Use Yahoo Finance as a data source to complete each of the following prob-
lems:
a) The stock symbol of Apple Inc. is AAPL. Estimate and for AAPL by
using the weekly adjusted closing prices over the last 2 years and the S&P
500 index as the market portfolio.
b) Estimate and for AAPL by using the weekly adjusted closing prices over
the last 4 years and the S&P 500 index as the market portfolio.
c) Estimate and for AAPL by using the daily adjusted closing prices over
the last 2 years and the S&P 500 index as the market portfolio.
d) Estimate and for AAPL by using the weekly adjusted closing prices over
the last 2 years and the NASDAQ-100 index as the market portfolio.
e) Observe the results above and give the factors that the actual calculated
value of depends on.
4.23. Establish (4.5) on page 158, i.e., show
of the S&P index and trades at approximately one-tenth of the dollar value of the S&P 500. Thus, the
rate of daily returns of SPY and S&P 500 index are basically the same.
4.4 Exercises 205
C Br f Ar2f 2Br f + C V 1 ( M r f e )
M = , 2
M = , wM = .
B Ar f ( B Ar f )2 B Ar f
4.24. Show that under linear factor model framework, portfolio variance can
be decomposed into common factor variance and idiosyncratic variance. (Hint:
Apply relation (4.25) on page 190.)
4.25. Use the single-factor model (4.26) on page 192, namely,
R = + R M + ,
to express in terms of the variance of the total market return and the covari-
ance of the market return with an individual securitys return.
4.26. Use the single-factor model (4.26) on page 192, namely,
R = + R M + ,
to obtain a corresponding asset pricing formula.

4.27. A self-financing or dollar-neutral portfolio is established by using the
proceeds of the short sales to finance the long purchases. In other words, under
the assumption of Frictionless Trading, a self-financing portfolio is a zero-cost
portfolio. For example, the excess return R M r f can be viewed as the return
of a portfolio that is formed by using the borrowed amount at interest rate r f
to purchase shares of SPY.
A traditionally common representation of many asset pricing models is in
the linear factor form:
m
E( Ri r f ) = + k k ,
k =1
where k with k = 1, 2, . . . , m are the values of the corresponding risk factors.

Given two risk factorsone is the excess return on the market portfolio and
the other is an economic recession factoruse a self-financing portfolio to es-
tablish a two-factor linear model for the excess return of a security.
4.28. Let
n
1
n Xi Yi XY
i =1
A= - ,
n n
1
n2 ( Xi X )2 (Yi Y )2
i =1 i =1

n n n
n Xi Yi Xi Yi
i =1 i =1 i =1
B= - 2 - 2 .
n n n n
n Xi2 Xi n Yi2 Yi
i =1 i =1 i =1 i =1
Prove that A = B.
4.29. Given a portfolio P, show that its Sortino ratio is no less than its Sharpe
ratio.
4.30. Let A be an n m matrix and A T be the transpose of A. Prove the follow-
ing property:
rank( A T A) = rank( A).
4.31. We continue from the last exercise. Let

1 x1 y1
1 x2 y2

A = . . with x1 = x2 and y= .
.. .. ..
1 xn yn
be given. Find the best fit to the system

A =y

4 4
4 4
4
such that the norm 4 A y44 is minimized.

Hint: Let = and L = | A y|2 , then

L = ( A y) T ( A y) = T A T A T A T y y T A + y T y.
References
[1] Acerbi, C., Tasche, D.: Expected shortfall: a natural coherent alternative to
value at risk. Econ. Notes 31(2), 379388 (2002)
[2] Acerbi, C., Nordio, C., Sirtori, C.: Expected Shortfall as a Tool for Financial
Risk Management. Abaxbank Working Paper. arXiv:condmat/0102304.
http://arxiv.org/pdf/cond-mat/0102304.pdf
[3] Artzner, P., Delbaen, F., Eber, J.-M., Heath, D.: Thinking coherently. Risk
10, 6871 (1997)
[4] Artzner, P., Delbaen, F., Eber, J.-M., Heath, D.: Coherent measures of risk.
Math. Finance 9(3), 203228 (1999)
[5] Bernstein, P.L.: Against the Gods: The Remarkable Story of Risk. Wiley,
New York (1996)
[6] Chan, L.K.C., Karceski, J., Lakonishokk, J.: On portfolio optimization:
forecasting covariances and choosing the risk model. Rev. Financ. Stud.
12(5), 937974 (1999)
[7] Chen, N., Roll, R., Ross, S.A.: Economic forces and the stock market. J.
Bus. 59(3) 383(1986)
References 207
[8] Connor G., Korajczyk, R.A.: Risk and return in an equilibrium APT: ap-
plication of a new test methodology. J. Financ. Econ. 21(2), 255290 (1988)
[9] Connor G., Korajczyk, R.A.: A test for the number of factors in an approx-
imate factor model. J. Finance 48, 12631291 (1993)
[10] Connor G.: The three types of factor models: a comparison of their ex-
planatory power. Financ. Anal. J. 51(1), 4246 (1995)
[11] Connor G., Goldberg, L.R., Korajczyk, R.A.: Portfolio Risk Analysis.
Princeton University Press, Princeton (2010)
[12] Danelsson, J., Jorgensen, B.N., Samorodnitsky, G., Mandira, M.: Subaddi-
tivity Re-Examined: the Case for Value-at-Risk. Financial Markets Group,
Discussion paper, 549. London School of Economics and Political Science,
London (2005)
[13] Davis, J., Fama, E., French, K.: Characteristics, covariances, and average
returns: 1929-1997. J. Finance 55(1), 396 (2000)
[14] Dhaene, J., Vanduffel, S., Goovaerts, M.J., Kaas, R., Tang, Q., Vyncke, D.:
Risk Measures and Comonotonicity: A Review. Stochastic Models. Taylor
& Francis Group, LLC (2006)
[15] Dowd, K.: Measuring Market Risk. Wiley, New York (2005)
[16] Fama, E.: Efficient capital markets: a review of theory and empirical work.
J. Finance 25, 383(1970)
[17] Fama, E., French, K.: The cross-section of expected stock returns. J. Fi-
nance 47(2), 427(1992)
[18] Fama, E., French, K.: Common risk factors in the returns on stocks and
bonds. J. Financ. Econ. 33(1), 3(1993)
[19] Fama, E., French, K.: The capital asset pricing model: theory and evidence.
J. Econ. Perspect. 18(3), 25(2004)
[20] Follmer, H., Schied, A.: Stochastic Finance. Walter de Gruyter, Berlin
(2004)
[21] Gilchrist, W.G.: Statistical Modelling with Quantile Functions. Chapman
and Hall/CRC, Boca Raton (2000)
[22] Glasserman, P., Heidelberger, P., Shahabuddin, P.: Portfolio value-at-risk
with heavy-tailed risk factors. Math. Finance 12(3), 239(2002)
[23] Goldberg, L.R., Hayes, M.Y., Menchero, J., Mitra, I.: Extreme risk analysis.
J. Perform. Meas. 14, 3 (2010)
[24] Goldberg, L.R., Hayes, M.Y.: The long view of financial risk. J. Investment
Manag. 8, 3948 (2010)
[25] Hull, J.C.: Risk Management and Financial Institutions. Prentice Hall, Up-
per Saddle River (2007)
[26] Ingersoll, J.: Theory of Financial Decision Making. Rowman and Little-
field, Savage (1987)
[27] Jorion, P.: Value at Risk, 3rd edn. McGraw-Hill, New York (2007)
[28] Luenberger, D.: Investment Science. Oxford University Press, New York
(1998)
[29] Ma, Y., Genton, M., Parzen, E.: Asymptotic properties of sample quantiles
of discrete distributions. Ann. Inst. Stat. Math. 63, 227(2011)
[30] Markowitz, H.: Portfolio Selection. Blackwell, Cambridge (1959)
[31] Reiley, F., Brown, K.: Investment Analysis and Portfolio Management.
[32] Rockafellar, R.T., Uryasev, S.: Optimization of Conditional Value-at-Risk.
J. Risk 2, 2141 (2000)
[33] Rockafellar, R.T., Uryasev, S.: Conditional value-at-risk for general loss
distributions. J. Bank. Finance 26, 1443(2002)
[34] Ross, S.A.: The arbitrage theory of capital asset pricing. J. Econ. Theory
13, 341(1976)
[35] Sharpe, W.: Capital asset prices: a theory of market equilibrium under
conditions of risk. J. Finance 19(3), 425(1964)
[36] Sharpe, W.: The Sharpe Ratio objectively determined and measured. J.
Portf. Manag. 21, 1 (1994)
[37] Sharpe, P., Alexander, G.J., Bailey, J.V.: Investments. Prentice-Hall, Upper
Saddle Rive (1999)
[38] Spearman, C.: General Intelligence objectively determined and mea-
sured. Am. J. Psychol. 15, 201(1904)
[39] Wilmott, P.: Paul Wilmott on Quantitative Finance. Wiley, New York
(2006)
Chapter 5
Binomial Trees and Security Pricing Modeling
We introduce a discrete-time model of a risky securitys futures price using a

binomial tree. By increasing the number of time steps in the tree, the assump-
tion is that one obtains a more and more accurate model of the random futures
price of a security. The chapter starts with a general n-period binomial tree
model and then restricts to the case of a Cox-Ross-Rubinstein tree. The latter
allows us to determine a continuous-time model, i.e., one in the limit n , of
the random futures price of a risky security. The chapter concludes with some
basic properties of the continuous-time security price model.
5.1 The General Binomial Tree Model of Security Prices
We shall present a general binomial tree model of the futures price of a security
given its current price, which is assumed known to all market participants.
Fix a time interval [t0 , t f ] over which to model a securitys price. The time
span of the interval [t0 , t f ] is denoted by
= t f t0 .
Let n be a positive integer. Divide [t0 , t f ] into n subintervals [t0 , t1 ], . . . , [tn1 , tn ]

of the same period (i.e., length) hn :
0 t0 , t1 = t0 + h n , t2 = t0 + 2 h n , ..., t n = t0 + n h n = t f ,
where

hn = .
n
Here t0 is the current time with t1 , . . . , tn future times. The subintervals [t j1 , t j ],
j = 1, . . . , n, will be called time steps. Since the overall time span is divided into
n periods, the tree will also be called an n-period binomial tree. Note that as the
number n of time steps changes, the label tn for the fixed final time t f changes.

210 5 Binomial Trees and Security Pricing Modeling
Denote the price of the security at time t j by S(t j ), where j = 0, 1, . . . , n. As-

sume that at the current time t0 , the price of the security is known from the market,
say, S(t0 ) = S0 . Then the futures prices S(t1 ), . . . , S(tn ) are random.
Remark 5.1. To avoid our notation becoming too cumbersome, we denote the
price of a security at time t by S(t) without additional notation to indicate the
type of model used for the security. In particular, whether S(t j ) represents a
securitys price at t j using a discrete or continuous-time model will be made
clear from the context. For example, almost all of this chapter (except at the
end) will employ a discrete model, while the next chapter and beyond typically
will use continuous-time modeling.

We now list some key assumptions and properties of an n-period binomial

tree. It is helpful to reference Figure 5.1 during the presentation.
Fig. 5.1 An n-period binomial tree over the interval [t0 , t f ], where n is a positive integer and each
[t j1 , t j ] has the same length /n, where = t f t0 . The tree satisfies the recombining property and
has independent paths. There are n + 1 possible values of S( tn ) and 2n possible price paths from t0 to
tn . The random gross returns S( t j ) /S( t j1 ), where j = 1, . . ., n are assumed independent.
5.1 The General Binomial Tree Model of Security Prices 211
Recombining property. If the price Snode at any node increases over the
next time step and is followed by a decrease in the subsequent time step,
then we get the same value if, instead, we had a price decrease followed
by an increase: (Snode un ) dn = (Snode dn ) un . By the recombining property,
there are then n + 1 possible prices at time tn :
S0 dnn , S0 un dnn1 , ..., S0 uin dnni , ..., S0 unn1 dn , S0 unn .
The prices increase as one moves through the list from left to right, equivalently,
from the end of the bottommost branch of the tree to the topmost one. In particular,
for the ith price, S0 uin dnni , where i = 0, 1, . . . , n, the power i of un is the
number of times the price had to increase during the n time steps to arrive
at S0 uin dnni , while n i is the number of times the price had to decrease.
In other words, we can express the price of the security at time t j as
NU, j j NU, j
S ( t j ) = S ( t0 ) u n dn , (5.1)
where S(t0 ) = S0 and NU, j is the random number of upticks in the secu-
ritys price from time t0 to t j .
Gross returns, capital-gain returns, and log returns. For each time step
[t j1 , t j ], where j = 1, . . . , n, the random up or down price movement of the
security is given by its gross return S(t j )/S(t j1 ) over the interval. We as-
sume that the gross returns of the n time steps, namely,
S ( t1 ) S ( t2 ) S ( tn )
, , ..., , (5.2)
S ( t0 ) S ( t1 ) S ( t n 1 )
are independent and identically distributed (i.i.d.). Explicitly, the gross returns
are assumed to be independent Bernoulli random variables with each hav-
ing the same probability distribution determined by

S(t j ) un with probability pn
= (5.3)
S ( t j 1 )
dn with probability 1 pn ,
where j = 1, . . . , n and
un > 1, 0 < dn < 1, 0 < pn < 1. (5.4)
In other words, from time t j1 to t j , each possible value of the price S(t j1 )
at t j1 either goes up by the factor un with probability pn > 0 or down by
the factor dn with probability 1 pn > 0. The situations pn = 0 and pn = 1
are excluded since there is little interest in binomial trees with such proba-
bilities.
The expected gross return over [t0 , t f ] is

S ( tn ) S ( t1 ) S ( t2 ) S ( tn )
E =E ...
S ( t0 ) S ( t0 ) S ( t1 ) S ( t n 1 )
n
S ( t1 )
= E (gross returns are i.i.d.)
S ( t0 )
= ( p n un + (1 p n ) dn ) n . (5.5)
Since the gross returns (5.2) are i.i.d., the capital-gain returns
S ( t1 ) S ( t2 ) S ( tn )
R1 = 1, R2 = 1, ..., Rn = 1,
S ( t0 ) S ( t1 ) S ( t n 1 )
are i.i.d. as well as the log returns

S ( t1 ) S ( t2 ) S ( tn )
ln , ln , ,..., ln ,
S ( t0 ) S ( t1 ) S ( t n 1 )
where

S(t j ) ln(un ) with probability pn
ln = (5.6)
S ( t j 1 )
ln(dn ) with probability 1 pn ,
for j = 1, . . . , n. We are using the fact that if X and Y are independent ran-
dom variables and if f and g are continuous1 functions, then f ( X ) and g(Y )
are independent.2
The triple pn , un , dn . These quantities depend on the size hn of the time

step. This is reasonable because under normal market conditions, we would
expect the price of a stock to be more likely to vary less during, say, a time-
step size of an hour as opposed to a day. Notationally, the subscript n in
pn , un , dn is there as a reminder of the dependence of these quantities on hn .
Additionally, since each time step in an n-period binomial tree has the same size,
the quantities pn , un , anddn are assumed to have the same value over every time
step in the tree. However, though their values are the same across an n-
period binomial tree, they are not necessarily the same for different trees.
For instance, the triplet pn , un , anddn does not carry over to 1-period, 2-
period, . . . , and (n 1)-period binomial trees since those trees have larger
time steps and different probability spaces. In fact, each element of the
1 For readers familiar with measure theory, it suffices for the functions f and g to be measurable,
which includes the continuous functions. In fact, all the functions you will encounter in our financial
applications are measurable. Some measure theory will be introduced in Chapter 6.
2 Proof. P ( f ( X ) A, g (Y ) B ) = P ( X f 1 ( A ), Y g 1 ( B )) = P ( X f 1 ( A )) P (Y g 1 ( B )) =
P ( f ( X ) A) P ( g(Y ) B), where the independence of X and Y was used in the second to the last
equality.
Fig. 5.2 One- and two-period binomial trees (left to right). Since the time interval is [t0 , t f ] for both
trees, each time step in the tree on the right is actually half the size of the one on the left.
sample space n of an n-period binomial tree is a sequence of n up (U)

and down (D) movements of the securitys price. Consequently, the sam-
ple spaces are differente.g., the 1- and 2-period trees have the following
sample spaces:
1 = {U, D } , 2 = {UU, UD, DU, DD } .
Sample price paths. Geometrically, each element n of the sample space

can be viewed as a discrete sample price path from t0 to tn , namely,
t St ( n ),
where
St = S ( t ), t t0 , t1 , . . . , t n ,
and St0 (n ) = S0 for every n in n . Moreover, the realizations of the ran-
dom price S(t j ) at a fixed time t j , where 1 j n, are given by the values
St j (n ) as n varies over the n . In addition, for a fixed t j and n , the pos-
sible price St j (n ) depends only on the portion of the sequence of Us and
Ds in n up to time t j , j = 1, . . . , n.
Example 5.1. Let us illustrate the above observations using the 2-period
binomial tree in Figure 5.2. The associated sample space of possible out-
comes is

2 = UU, UD, DU, DD .
Consider the following specific possible outcome:
2 = UD .
Table 5.1 Possible outcomes 2 and security price values at times t0 , t1 , and t2 for a 2-period binomial

tree. The sample space is 2 = UU, UD, DU, DD . For example, if 2 = DU, then St1 ( 2 ) = S0 d2 .
The random variable St1 depends only on the first slot of 2 (e.g., St1 (UD ) depends only on U), while
the random variable St2 depends on both slots. Note that St0 is nonrandom, i.e., it is a constant random
variable.
2 St 0 ( 2 ) St 1 ( 2 ) St 2 ( 2 )
UU S0 S0 u2 S0 u22
UD S0 S0 u2 S0 u2 d2
DU S0 S0 d2 S0 d2 u2 = S0 u2 d2
DD S0 S0 d2 S0 d22
Then 2 determines the following prices at times t0 , t1 , and t2 :
S t 0 ( 2 ) = S0 , St1 (2 ) = S0 u2 , St2 (2 ) = S0 u2 d2 .
In other words, we can view 2 as a discrete curve given by
t St ( 2 ), t = t0 , t1 , t2 .
We also see that the possible price St1 (2 ) depends only on the first entry
U of 2 = UD. Table 5.1 presents all the possible outcomes in 2 along
with the prices across time that each possible outcome determines. Com-
pare with Figure 5.2.

Since every path to one of the n + 1 possible prices,
S0 dnn , S0 un dnn1 , ..., S0 uin dnni , ..., S0 unn1 dn , S0 unn ,
consists of a sequence of n up or down price movements (i.e., gross returns)

and because these movements are independent, all the paths are indepen-
dent.
The number of price paths in the tree is 2n , which quickly becomes ex-
tremely large. For example, a 50-period binomial tree has
250 1015 (one quadrillion!) price paths.
Probability measure. The probability of a particular price path occurring

is obtained by multiplying the probabilities along the path. Explicitly, let
NU be the random number of upticks in the price of the security from t0 to
tn = t f . Then NU is a binomial random variable. The probability measure
on the sample space n of an n-period binomial tree is then defined at a
possible outcome n by
N (n )
P (n ) = pn U (1 pn )n NU (n ) .
Thinking of n as a discrete price path, we see that P (n ) is indeed the

product of the probabilities along the path.
Additionally, since the price paths are independent, the probability of a

specific security price occurring at time t f is the sum of the probabilities of
the individual paths to the given price. For instance, the probability of the
price S0 uin dnni , conditioned on the current price being S(t0 ) = S0 , is
n
P S(tn ) = S0 uin dnni = pin (1 pn )ni
i
= probability of i price increases during [t0 , t f ],
(5.7)
where i = 0, 1, 2, . . . , n, and
k
n
P S(tn ) S0 ukn dnnk = i pin (1 pn )ni
i =0
= probability of at most k price increases,
(5.8)
where k = 0, 1, 2, . . . , n.
The expected security price at time t f , conditioned on the current price

being S(t0 ) = S0 , is
n
i n i n
E S ( t f ) = S0 u n d n pin (1 pn )ni
i =0
i
n n i
n
= S0 ( p n un ) i (1 p n ) dn
i =0
i
= S0 ( p n u n + ( 1 p n ) d n ) n , (5.9)
where the binomial formula was used for the last equality. Note that Equa-
tion (5.9) agrees with (5.5).
Cash dividends and the return rate. From time t0 to t f , assume that the
security pays a constant, continuous, proportional annual dividend yield rate q
giving a cash dividend of
D ( t j 1 , t j ) = q S ( t j 1 ) h n , ( j = 1, 2, . . . , n).
Unless stated to the contrary, assume that in the continuous-time limit (i.e., n
or hn 0), the cash dividends are continuously reinvested in the security to
buy more units of the security. The (total) return rate from t j1 to t j is then
S ( t j ) S ( t j 1 ) + D ( t j 1 , t j )
R ( t j 1 , t j ) = = R j + q hn . (5.10)
S ( t j 1 )
Note that the return rates R(t0 , t1 ), . . . , R(tn1 , tn ) are i.i.d.
Example 5.2. Consider a nondividend-paying stock with a current price of

$155. Model the futures price behavior of the stock over the next 6 months
using a 100-period binomial tree with
u100 = 1.01424, d100 = 0.985957, p100 = 0.52299.
a) What is the 57th possible price 6 months out, where the 101 possible prices
are counted from bottom to top in the tree?
Solution. The (i + 1)st possible price is
i n i
S0 u100 d100 ,
100 d100 = $183.64.

so the 57th one is $155 u56 44
b) What is the probability that the 57th possible price will occur?
Solution. By (5.7), do not even need to know the 57th possible price in order
to compute the desired probability:
100
P S(0.5) = $155 u100 d100 =
56 44
100 (1 p100 ) = 6.1%.
p56 44
56
c) What is the probability that the stocks price increases 56 times?
Solution. Equation (5.7) yields that this probability is the same as that of the
57th possible price occurring, which is 6.1%.
d) What is the probability that the stocks price is at most $183.64?
Solution. Since
10056
$183.64 = $155 u56
100 d100 ,
Equation (5.8) gives
56
100 i
P S(0.5) $155 u56
100 d44
100 = p100 (1 p100 )100i = 80%.
i =0
i
e) What is the expected price of the stock 6 months from now?

100
Solution. E (S(0.5)) = $155 p100 u100 + (1 p100 ) d100 = $167.05.

Log Returns and Security Price Modeling
We now conclude the section with a formal expression for the futures price of
the security in terms of the log returns over the various time steps. Writing

S ( tn ) ln
S(tn )
S ( t n ) = S0 = S0 e S0
S0
and using

S ( tn ) S ( t1 ) S ( t2 ) S ( t3 ) S ( tn ) n S(t j )
ln
S0
= ln
S0 S ( t 1 ) S ( t 2 )

S ( t n 1 )
= ln S ( t j 1 )
, (5.11)
j =1
we can express the price S(tn ) in terms of the time-step log returns as

n S(t j )
S(tn ) = S0 exp ln , (5.12)
j =1
S ( t j 1 )
where tn = t f . Equation (5.12) highlights that, to determine the securitys price, it

suffices to determine the log returns. Log returns are then important.
We can further breakdown Equation (5.12) into individual parts that can be
determined. Let Xn,j be the standardization of ln S(t j )/S(t j1 ) :
S(t j )

S(t )
ln E ln S(t j )
S ( t j 1 ) j 1
Xn,j = # S(t ) .
Var ln S(t j )
j 1
Equivalently,
5
6
S(t j ) S(t j ) 6 S(t j )
ln = E ln + 7 Var ln Xn,j .
S ( t j 1 ) S ( t j 1 ) S ( t j 1 )
Since the log returns are i.i.d., we have

S(t j ) S ( t1 )
E ln = E ln ( j = 1, . . . , n)
S ( t j 1 ) S ( t0 )
and

S(t j ) S ( t1 )
Var ln = Var ln ( j = 1, . . . , n).
S ( t j 1 ) S ( t0 )
It follows
-
n S(t j ) n
S ( t1 ) n
S ( t1 )
ln S ( t j 1 )
= E ln S ( t0 )
+ Var ln
S ( t0 )
Xn,j
j =1 j =1 j =1
-
S ( t1 ) S ( t1 ) n
= n E ln
S ( t0 )
+ Var ln
S ( t0 ) Xn,j .
j =1
Let
n
1
Zn =
n
Xn,j .
j =1
Then the discrete pricing formula (5.12) can be expressed as

-
S ( t1 ) S ( t1 )
S(tn ) = S0 exp n E ln + n Var ln Zn .
S ( t0 ) S ( t0 )
(5.13)
Equation (5.13) shows that, to characterize the futures price of the security in

the continuous-time limit, it suffices to determine nE ln SS((tt10 )) , n Var ln SS((tt10 )) ,
and Z n as n .
The goal of Sections 5.2 and 5.3 is to characterize these quantities for n suf-
ficiently large and then in the full limit n .
5.2 The Cox-Ross-Rubinstein Tree
The Cox-Ross-Rubinstein (CRR) tree is a special case of a binomial tree and,

consequently, satisfies all the assumptions and properties of a general binomial
tree. For a sufficiently large number n of time steps, this tree will allow us to ex-
press un , dn , and pn approximately in terms of the time-step period hn and two
quantities, denoted RW and , which measure the respective instantaneous ex-
pectation and volatility of the securitys log returns, respectively.
# In fact, the

S(t ) S(t )
CRR tree will allow us to estimate n E ln S(t1 ) and n Var ln S(t1 ) in
0 0
the limit n . Additionally, in Section 5.3 we shall characterize Z n as n
by using the CRR tree and the Lindeberg Central Limit Theorem.
We present the n-period CRR tree by using a 1/n perturbative approach.
The approximations are presented informally to preserve the intuitive ideas
behind the analysis. Our discussion will utilize and parallel some aspects of
the development in Cox, Ross, and Rubinstein [4], Cox and Rubinstein [5],
Roman [15], and Wilmott, Dewynne, and Howison [16]. See Roman [15, Chap.
9] for a very detailed mathematical treatment that keeps track of the probabil-
ity spaces.
5.2 The Cox-Ross-Rubinstein Tree 219
5.2.1 The Real-World CRR Tree
It is assumed that, as the number n of time steps increases, the CRR tree be-
comes a more and more accurate model of a securitys price. For this reason,
the model is also called a real-world CRR tree. Of course, such terminology
should not be taken literally. It is another matter whether the real-world CRR
tree actually becomes an accurate fit to security prices in the marketplace as n
increases. These issues are beyond the scope of this introductory text, but we
do note that other types of trees have been studied, e.g., trinomial trees. Inter-
estingly, we shall see in Chapter 8 that the risk-neutral CRR tree is a natural
choice when pricing derivatives.
Assumptions and the Quantities m, RW , and 2
We now list some additional assumptions and define certain key quantities
associated with the real-world CRR tree:
Computations for large n. For n sufficiently large, we shall compute only

to first order in 1/n, i.e., we ignore all terms of the form (1/n) a with a > 1.
Almost surely continuous sample price paths. As n , assume that the
CRR binomial trees probability space transforms to a continuous-time ana-
log with sample space denoted . Similar to the discrete case, we iden-
tify each element in with the sample price path t (S(t))( ), where
0 t t f . The sample price paths are assumed to be continuous almost surely in
the limit n . Since we shall explore probabilities of price paths, we do
not need to require that literally every price path must be continuous in the
limit n ; only that they are continuous with probability 1 in the limit.
Expected returns and the constant m. Assume that the expected time-step re-
turn rate per unit time period converges to a constant m in the continuous-time
limit:
E ( R(t0 , t1 ))
m as n . (5.14)
hn
In other words, since the time period shrinks to a single moment of time as
n , we think of m as the securitys instantaneous expected return rate
per unit time and will call it the instantaneous expected return rate or, simply,
the instantaneous expected return. The rate m is quoted per annum.
We now show explicitly that m applies at each instant of time in the sense
that the expected price of the security is obtained by continuously com-
pounding the current price at the rate m q. First, since the (total) return
rate R(t0 , t1 ) arises from the capital-gain return R1 plus the dividend yield
contribution q hn , i.e.,
R ( t 0 , t 1 ) = R1 + q h n ,
employing (5.14) gives

E ( R1 ) E R ( t0 , t1 ) q h n
= mq as n . (5.15)
hn hn
We then interpret m q as the instantaneous capital-gain return rate or sim-
ply the instantaneous capital-gain return. Note that m q is also quoted per
annum. Now, by (5.15), we have
( m q ) h n E ( R1 ) (n sufficiently large) (5.16)
and since
e( m q ) h n 1 + ( m q ) h n (n sufficiently large),
it follows3
E (S(t1 )) S0 e(mq) hn (n sufficiently large). (5.17)
Equation (5.17) shows that as n increases, the gross return over the interval
[t0 , t f ] becomes independent of the number of time periods n in a partition
of the interval. In fact, recalling (5.5), we see that (5.17) yields
n
S(t f ) S ( t1 ) n
E = E e( m q ) h n = e( m q )
S ( t0 ) S ( t0 )
(n sufficiently large)
or
E (S(t f )) S0 e(mq) (n sufficiently large). (5.18)
The expectation on the left hand side of (5.18) is for discrete time.
We assume that m and q are known. Additionally, Equation (5.18) shows that
the instantaneous expected capital-gain return m q can be interpreted as
the continuously compounded rate at which the securitys expected price increases
(m q > 0) or decreases (m q < 0). In most cases, m q > 0. Naturally, a
security must have nontrivial promise for investors to tolerate an expected
return of m q < 0 for an extended period.
3 Strictly speaking, we are considering the conditional expectation E ( S( t1 )|S( t0 ) = S0 ) S0 e( m q ) h n

for n sufficiently large.
Example 5.3. Suppose that the time span = t f t0 is 2 years and the time
period hn is a trading day. Assuming 252 trading days in a year, we then
consider a 504-period CRR tree. Assume that the annual dividend yield
rate is 2% and the instantaneous annual expected return is 10%. If the cur-
rent security price is $75, then (5.18) yields that the expected price of the
security 1 year from now is obtained by continuously compounding $75 at
the annual rate m q = 8%. Explicitly, taking the current time to be 0, we
get
E (S(2)) $75 e0.082 = $88.01.

Remark 5.2. Estimating m by using historical security prices is problematic

since its value is affected by the sampling frequency (e.g., daily, weekly,
monthly, etc.) and size of the data (e.g., 6 months of prices versus 1.5 years,
versus 10 years, etc.) See Leunberger [11, pp. 214216] for more. However,
in the study of derivatives, this issue is avoided all together since the price
of a derivative will be independent of m! See pages 404 and 415.
First constraint. Assume the following recombining condition:
un dn = 1. (5.19)
This condition makes the CRR tree recombine along the horizontal line
through the initial price S0 since S0 un dn = S0 .
Second constraint and the constant RW . Define n to be the expected time-
step log return per unit time period:

1 S ( t1 )
n = E ln .
hn S ( t0 )
Explicitly
n hn = pn ln un + (1 pn ) ln dn . (5.20)
Additionally, since the time-step log returns are identically distributed,
Equation (5.11) yields that the expected log return over the interval [t0 , t f ]
is n times the expected log return over the time step [t0 , t1 ]:

S ( tn ) S ( t1 )
E ln = n E ln = n . (5.21)
S ( t0 ) S ( t0 )
A second constraint assumed for a CRR tree is that n converges to a constant RW
as n increases without bound:
n RW as n . (5.22)
The constant RW is called the real-world4 instantaneous drift or, simply, the
real-world drift of the securitys price. Also, some authors refer to RW as a
natural drift or physical drift. We quote RW per annum.
Finally, assume that RW is known. As in the case of m q, the quantity RW is

estimated by using historical security prices over an appropriate choice of
time steps (e.g., daily) and

n E ln S(t1 )
1 S ( t1 ) S ( t0 )
RW E ln = (n sufficiently large).
hn S ( t0 )
(5.23)
Example 5.4. Suppose that the time interval [t0 , t f ] is a year and partition
the year into 252 trading days. Then
1
= 1, n = 252, hn = .
252
Taking the current time to be t0 = 0, one trading day later is t1 = 252
1
, and 1
year away is t f = 1. Historical adjusted closing prices of the security can
5
be used to estimate the expected daily log return:

S(1/252)
E ln .
S (0)
Employing (5.21), the expected log return over a year is then

S (1) S(1/252)
E ln = 252 E ln . (5.24)
S (0) S (0)
In other words, the annual expected log return on the left hand side of (5.24)
is obtained by annualizing the expected daily log return, i.e., by multiply-
ing the expected daily log return by 252. Equations (5.23) and (5.24) then
show that the real-world drift RW is estimated by the annualized expected
daily log return:

S(1/252)
RW 252 E ln .
S (0)
Note that the daily log returns are assumed to be i.i.d. Also, we shall see
later (Equation (5.30) on page 226) that RW is a function of m, which as
mentioned in Remark 5.2 is problematic to estimate.

4 We remind readers to be mindful of the usage of the terminology real world; see the introductory
paragraph to this section on page 219.
5 See page 19.
Third constraint and the constant . Define n2 to be the variance of the

time-step log return per unit time step:

1 S ( t1 )
n =
2
Var ln .
hn S ( t0 )
We have (Exercise 5.19)
2
un
n2 hn = pn (1 pn ) ln . (5.25)
dn
Observe that by the i.i.d. property of the time-step log returns, it follows

S ( tn )
n = Var ln
2
.
S ( t0 )
Equivalently, the variance of the log return over [t0 , t f ] is n times the vari-
ance of the log return over a time step:

S ( tn ) S ( t1 )
Var ln = n Var ln . (5.26)
S ( t0 ) S ( t0 )
A third assumed constraint for a CRR tree is that n2 converges to a positive con-
stant 2 as n increases without bound:
n2 2 > 0 as n . (5.27)
The quantity > 0 is called the continuous-time volatility or, simply, the
volatility of the securitys price. We assume that is known.
Example 5.5. Employing the same inputs as in Example 5.4, Equation (5.26)
shows that the annual variance of the log return is obtained by annualizing
the variance of the daily log returns:

S (1) S(1/252)
Var ln = 252 Var ln .
S (0) S (0)
Equation (5.27) then yields that if the standard deviation of the daily log
returns is estimated by using historical security price data, then the annual
volatility of the security is estimated as follows:
1/2
S(1/252)
252 Var ln .
S (0)
In other words, the standard deviation of the daily log returns is annual-
ized through multiplying by 252. As noted in Example 5.4, the daily log
returns are assumed to be i.i.d. in order to apply the CRR tree.

Remark 5.3. The appearance of the constraints (5.19), (5.22), and (5.27) may
seem mysterious. We shall see that they make it possible to determine
approximate expressions of pn , un , and dn for n sufficiently large; see
page 226. Perhaps a more practical explanation for the constraints is that
they allow in the limit n for the discrete CRR pricing formula to con-
verge to the continuous-time security pricing formula utilized in the Black-
Scholes-Merton model.

Approximating the variances of the time-step log return and gross return.
A useful approximation of the variance of the log return is

S ( t1 )
Var ln E R21 (n sufficiently large). (5.28)
S ( t0 )
To obtain this result, first observe that
2
S ( t1 ) S ( t1 ) 2 S ( t1 )
Var ln = E ln E ln
S ( t0 ) S ( t0 ) S ( t0 )

S ( t1 ) 2 2
n
= E ln
S ( t0 ) n

S ( t1 ) 2
E ln (n sufficiently large).
S ( t0 )
Now, by the continuity assumption of the sample price paths, we have

|R1 | 0 almost surely as n . Using this result and the leading terms
in the Taylor expansion of the natural logarithm, namely,
1
ln x = ( x 1) ( x 1)2 + (| x 1| 1),
2
with
S ( t1 )
x= , x 1 = R1 ,
S ( t0 )
it follows

S ( t1 ) 2 R21 2
E ln E R1 (n sufficiently large)
S ( t0 ) 2

R41
E R21 R31 + (n sufficiently large).
4
For n sufficiently large, we have |R1 | 1 almost surely and so ignore any
3
contributions
from skewness (i.e., the term E R1 ) and kurtosis (i.e., the
term E R41 ). We then enforce the following approximation:
2
S ( t1 )
E ln E R21 (n sufficiently large).
S ( t0 )
Our heuristic arguments then imply the desired result:

S ( t1 )
Var ln E R21 (n sufficiently large).
S ( t0 )
Under our assumptions, the variances of the log return and gross return
are also approximately equal:

S ( t1 ) S ( t1 )
Var Var ln (n sufficiently large), (5.29)
S ( t0 ) S ( t0 )
which is a consequence of

S ( t1 ) S ( t1 )
Var = Var 1 = E R21 (E (R1 ))2
S ( t0 ) S ( t0 )
(m q) 2
= E R21 (by (5.16))
n

E R21 (n sufficiently large)
and Equation (5.28).

Expressing RW in terms of m q and . Using the Taylor expansion of the
natural logarithm, we get

S ( t1 ) R2
n = E ln E R1 1 (n sufficiently large)
n S ( t0 ) 2
or
2 n
E R21 2 E (R1 ) (n sufficiently large).
n
On the other hand, employing (5.28) yields

n2 E R21 (n sufficiently large).
n
Consequently
n2 2 n E (R1 ) 2 n .
Taking the limit n and making use of the convergences (5.15), (5.22),
and (5.27), namely,
n E (R1 ) (m q), n RW , n2 2 ,
we obtain the following expression for real-world drift RW in terms of the

instantaneous capital-gain return m q and volatility :
2
RW = m q . (5.30)
2
The Real-World CRR Equations
To obtain formulas for the unknowns un , dn , pn of a real-world CRR tree in

terms of the inputs m q, , hn , solve the three constraints (5.19), (5.22), and
(5.27) for the three unknowns. For the readers convenience, we restate these
equations below:
un d n = 1 (5.31)
n RW (n sufficiently large) (5.32)
n2 2 (n sufficiently large), (5.33)
where
n hn = pn ln un + (1 pn ) ln dn (5.34)
2
un
n2 hn = pn (1 pn ) ln . (5.35)
dn
We claim that for n sufficiently large, Equations (5.31)-(5.35) are solved by
1 RW ,
un e hn
, dn e hn
, pn 1+ hn , (5.36)
2
which are called the real-world CRR equations. They are the governing formulas
for the real-world CRR tree. Note that
1
pn as n (5.37)
2
and by (5.17),
p n un + (1 p n ) dn e ( m q ) h n (n sufficiently large),
which yields another expression for the uptick probability:
e( m q ) h n dn
pn (n sufficiently large). (5.38)
un dn
Equation (5.38) and the expression for pn in (5.36) are equivalent to first order
in 1/n (see Exercise 5.20).
To establish (5.36), first observe that by (5.31), we have lnun = ln dn , which

when inserted in (5.34) yields
n hn = pn ln dn + (1 pn ) lndn = 2 pn ln dn + lndn .
We then solve for the desired uptick probability:

n hn ln dn n hn lndn 1 n h n
pn = = = 1 + .
2 ( ln dn ) 2 lnun 2 ln un
Substituting the above expression for pn and ln(un /dn ) = 2 ln un into (5.35)
gives

1 n h n n h n
n2 hn = 1 + 1 4 (ln un )2
4 ln un lnun

n h n 2
= 1 (ln un )2 .
ln un
Employing (5.33) and computing to first order in hn , it follows
2 hn n2 hn = (ln un )2 (n hn )2 (ln un )2 (n sufficiently large).
Since un > 1, we have

,
hn ln un = ln dn (n sufficiently large). (5.39)
Applying (5.32) then gives

1 RW ,
pn 1 + hn (n sufficiently large) (5.40)
2
and
un e hn
, dn e hn
(n sufficiently large).
Example 5.6. Suppose that a nondividend-paying stock with a current price of
$75 has an instantaneous annual expected return of 12% and annual volatility
of 10%. Model the behavior of the stocks price using a 100-period CRR tree.
a) What is the expected price (i.e., the forecasted price) of the stock 3 months
from now?
Solution. Given the large number of periods, we employ the CRR tree to
determine u100 , d100 , and p100 in terms of h100 , , and m:
e(mq) h100 d100
u100 e h100
, d100 e h100
, p100 .
u100 d100

We have n = 100, h100 = 0.25100 = 0.0025, h100 = 0.05, m = 0.12, q = 0,
= 0.10, and S(t0 ) = $75. Moreover
u100 = 1.00501, d100 = 0.995012, p100 = 0.528754.
Therefore, the expected price is

100
E (S(0.25)) = $75 p100 u100 + (1 p100 ) d100 = $77.28.
b) What is the probability that 3 months from now, the stocks price is less than
or equal to its expected price?
Solution. First, we check whether the expected price $77.28 is one of the 101
possible prices 3 months from now. To do this, we must find the number k
of price upticks such that
100k
$77.28 = $75 u100
k
d100 .
In general, the number of upticks associated with a price S is the nonnega-

tive integer k that solves
S = S0 ukn dnnk .
It follows
S
ln S0 dnn
k= . (5.41)
un
ln dn
Since S = $77.28, we get k = 53, i.e., the expected price $77.28 is the 54th
possible price, counting from bottom to top.
With k available, we can apply (5.8) that yields

P ( S(0.25) $77.28) = P S(0.25) $75 u53 47
100 d100
53
100 i
= p100 (1 p100 )100i
i =0
i
= 55%.
For the 100-step tree, the probability that the price 3 months from now is less
than or equal to its mean is more than 50%. We emphasize that this property is
not an accident for n sufficiently large. It carries over to the continuous-time
limit; see Equation (5.78) on page 248.
Remark 5.4. Note that if k were not an integer, say, k = 42.6785, then the
dollar amount $75 ukn d100 k would not be one of the possible price values of
n
the stock. However, we can still compute the desired probability by omitting
that dollar amount (which has probability zero) and adding the probabilities
up to k = 42.

c) What is the probability that 3 months from now, the stocks price is less than
or equal to its current price? What would be the probability if the volatility
of the stock were much higher, say, 40%?
Solution. By (5.41), the number of upticks associated with the price S = $75
is k = 50, which means
10050
$75 u50
100 d100 = $75.
The current price is then the 51st possible price, counting from bottom to
top. The probability of the price 3 months out being less than or equal to $75
is actually less that 50%. In fact, by (5.8) the probability is
50
100 i
P ( S(0.25) $75) = p100 (1 p100 )100i = 32%.
i =0
i
Now, if we increase the volatility from 10% to 40%, the CRR tree yields:
u100 = 1.0202, d100 = 0.980199, p100 = 0.502501.
In addition, the number of upticks for S = $75 is also k = 50. However, the
probability that the stock price 3 months from now is less than or equal to its
current price increases to 52%. In other words, for a sufficiently large volatility,
we see that there is more than a 50% probability that the price 3 months away is less
than or equal to the current price. This property is also not coincidental for n
and sufficiently large. We shall encounter it again in our continuous-time
study of security prices; see Equation (5.79) on page 248.

Security Price Formula for a Real-World CRR Tree
We now express the general binomial tree security price formula (5.13) on
page 218, namely,
-
S ( t1 ) S ( t1 )
S(tn ) = S0 exp n E ln + n Var ln Zn ,
S ( t0 ) S ( t0 )
in the framework of a real-world CRR tree. Recall that

n
1
Zn =
n
Xn,j .
j =1
S(t j )

Here Xn,j is the standardization of ln S ( t j 1 )
and, since the log returns being
identically distributed, it follows
S(t j )

S(t ) S(t )
S(t )
ln E ln S(t j )
S ( t j 1 )
ln S(t j ) E ln S(t1 )
j 1 j 1 0
Xn,j = # S(t ) = # .
j S ( t1 )
Var ln S(t ) Var ln S(t )
j 1 0
Because

S ( t1 ) S ( t1 )
n E ln = n , n Var ln = n2 ,
S ( t0 ) S ( t0 )
we have
S(t j )
ln S ( t j 1 )
n h n
Xn,j = . (5.42)
n hn
Also, it can be shown (Exercise 5.21) using the formulas (5.31)(5.35) that in a
real-world CRR setting, the standardization Xn,j becomes
1 p n

with probability pn
p n (1 p n )
Xn,j = (5.43)

pn
with probability 1 pn .
p n (1 p n )
The security pricing formula now takes the following more compact form for
a real-world CRR tree:

S(tn ) = S0 en + n Zn
. (5.44)
The goal is to determine the quantity to which (5.44) converges in the limit
n .
5.2.2 The Risk-Neutral CRR Tree
The current section explores the CRR tree in the context of a risk-neutral world,
that is, in a world of only risk-neutral investors (see page 137).
In a risk-neutral world, meaning a world of risk-neutral investors (see page 137),
the expected futures price of a security is assumed to be given by its current price
continuously compounded at the risk-free rate r minus any cash dividend yield rate.
For such a world, there is no compensation required for the securitys risk
since the rate r compensates only for opportunity cost and inflation. Explicitly,
if a security pays a continuous, proportional cash dividend at constant annual
yield rate q, then in a risk-neutral world, the securitys expected price at the
future time t1 , given the current price S(t0 ) = S0 , is

E S ( t 1 ) = S0 e ( r q ) h n . (5.45)
On the other hand, a real marketplace consists primarily of risk-averse

rather than risk-neutral investors, so it may seem like a mathematical indul-
gence to explore risk neutrality. We shall show in Chapter 8 that the price of
a derivative in the BSM model is actually independent of the instantaneous
expected return m; see Remark 5.2 on page 221. In other words, if we assume
m = r for the security, which is the case in a risk-neutral world, then though our
pricing model for the underlying security would be unrealistic, the resulting
price of a derivative based on the security would be the same as if we had used
a real-world value for m.6 For this reason, we shall investigate a risk-neutral
CRR tree and will do so to first order in 1/n.
We assume that in a risk-neutral world, the security prices and dividend yield
rate are the same as those in the real marketplace. Explicitly, the real-world prices
S(t0 ), . . . , S(tn ) and q will also be used for the risk-neutral CRR analysis. In
particular, for a risk-neutral CRR tree, we shall employ the securitys current market
price S(t0 ) and the same up and down factors un and dn , where un > 1, 0 < dn < 1,
and also assume that the relationship between un and dn is unchanged: un = d1n . Be-
cause the up and down factors un and dn and the period hn are the same for
the real world and risk-neutral world, we can make use of

un e hn
, dn e hn
(n sufficiently large), (5.46)
where is the real-world, continuous-time volatility of the security.
Risk-Neutral Uptick Probability pn
What actually changes in the switch from the real world to a risk-neutral world is
the probability. In particular, the uptick probability will no longer be pn . This is
because, in a risk-neutral world, Equation (5.45) is true. Can we find an uptick
probability pn that makes the risk-neutral condition (5.45) hold? Writing out
(5.45) formally using the unknown quantity pn , we obtain
pn S0 un + (1 pn ) S0 dn = S0 e(rq) hn . (5.47)
Equation (5.47) is readily solved:
e (r q ) h n dn
pn = . (5.48)
un dn
Strictly speaking, we do not yet know if the quantity pn given by (5.48) is ac-
tually a probability. It will be a probability if we can prove that pn is between
zero and one. As pointed out on page 211, we shall exclude binomial trees with
pn = 0 or pn = 1. By (5.47), this means that the constraints
6 Recall that it is essentially impossible to determine a reliable value for m in the marketplace.
e (r q ) h n = dn , e (r q ) h n = un , (5.49)
are enforced.
We claim that, for an n-period risk-neutral binomial tree, the no-arbitrage condition
implies
0 < pn < 1.
To establish the claim, it suffices to prove that if there is no arbitrage, then an
n-period risk-neutral binomial tree satisfies
dn < e (r q ) h n .
un > e (r q ) h n .
To show these two statements, it suffices to consider the time step [t0 , t1 ]
since the uptick probability is assumed the same across every time step in a
binomial tree. Assume dn > e(rq) hn . At the current time t0 , short (borrow) at
the risk-free rate r an amount equal to the cost S0 eq hn of eq hn units of the
security. Use these funds to long eq hn units of the security. At time t1 , the
number of units of the security grows to 1 due to continuous cash dividend
reinvesting to buy more shares. Sell the one unit of the security to receive S(t1 ).
The amount owed on the loan at t1 is S0 e(rq) hn . The net profit/loss is
S ( t 1 ) S0 e ( r q ) h n .
Since un > 1 and 0 < dn < 1, we see

S0 un with probability pn
S ( t1 ) = S0 d n .

S0 dn with probability 1 pn
Consequently,
S ( t 1 ) S0 e ( r q ) h n S0 d n S0 e ( r q ) h n .
If dn > e(rq) hn , then
S(t1 ) S0 e(rq) hn > 0,
which is an arbitrage. Hence
dn e (r q ) h n .
However, we exclude the case dn = e(rq) hn since (5.47) shows that it leads to a
binomial tree with pn = 0. The argument for un > e(rq) hn is left as an exercise
(Exercise 5.7).
The no-arbitrage condition (along with the constraints (5.49)) then yields
dn < e (r q ) h n < un .
Therefore, by (5.48) we obtain 0 < pn < 1. The quantity pn is called a risk-neutral

uptick probability.
Notation. For the risk-neutral CRR tree, we shall designate expectations, vari-
ances, etc. with respect to pn using an .
The Risk-Neutral CRR Equations
Equations (5.46) and (5.48) show that the following quantities govern the risk-
neutral CRR tree for n sufficiently large:
e (r q ) h n dn
un e hn
, dn e hn
, pn = . (5.50)
un dn
Let us now express pn in a form analogous to the expression for pn in (5.40).
The quantities m, n , andn change to the following m , n , andn in the risk-
neutral setting:
m hn E ( R(t0 , t1 )) (n sufficiently large)

S ( t1 )
n hn
= E ln
S ( t0 )

2 S ( t1 )
(n ) hn = Var ln .
S ( t0 )
We assume there are constants and > 0 such that
n , n as n .
Since in a risk-neutral world, the expected futures price comes from com-
pounding S0 at the rate r q, the instantaneous expected return rate m is the
risk-free rate. In fact, analogous to (5.17) we have

S0 e ( m q ) h n E S ( t 1 ) (n sufficiently large),
which with (5.45) gives

q) h
S0 e ( m n
S0 e ( r q ) h n (n sufficiently large),
i.e., in the limit n we get

m = r.
Arguing analogously to the real-world CRR tree case (see page 225), we ob-
tain
( )2
= r q ,
2
which is called the risk-neutral drift of the security. Similarly, the analogs of
(5.20) and (5.25) are
2
2 un
n hn = pn ln un + (1 pn ) ln dn , (n ) hn = pn (1 pn ) ln ,
dn
which with un = 1
dn yield the following for n sufficiently large:

1 ,
un e hn
, dn e hn
, pn 1 + hn .
2
Because the up and down factors are the same for the risk-neutral world and
the real world, we have

e hn
un e hn
(n sufficiently large).
In other words, the continuous-time volatility is the same in the real world and the
risk-neutral world:
= .
Consequently, the risk-neutral CRR equations become the following for n sufficiently
large:
1 ,
un e hn
, dn e hn
, pn 1+ hn , (5.51)
2
where
2
= r q . (5.52)
2
Relationship Between pn and pn
The risk-neutral uptick probability pn yields

E S ( t 1 ) = S0 e ( r q ) h n ,
while the real-world uptick probability pn gives

E S ( t 1 ) S0 e ( m q ) h n (n sufficiently large).
We can express pn in terms of pn through the following transformation:

"
pn pn n pn (1 pn ) (n sufficiently large), (5.53)
where
E ( R(t0 , t1 )) r hn
n = , . (5.54)
Var ( R(t0 , t1 ))
Here n is the Sharpe ratio of the security given as the ratio of the spread
between the expected total return rate and the risk-free rate7 r hn across the
time step [t0 , t1 ], to the security risk across the same time step. The relationship
(5.53) is a discrete-time example of Girsanov theorem (see Neftci [13, Chap. 14]
and references therein).
We now give a heuristic proof of (5.53). Our approach is to compute the
Sharpe ratio n first. By (5.14) on page 219, the expectation in the numerator of
(5.54) is
E ( R(t0 , t1 )) m hn (n sufficiently large).

,
To determine the denominator Var ( R(t0 , t1 ) of (5.54), employ (5.29):

S ( t1 ) S ( t1 )
Var Var ln (n sufficiently large).
S ( t0 ) S ( t0 )
Since
S ( t1 ) S ( t1 )
Var ( R(t0 , t1 )) = Var 1 + q hn = Var ,
S ( t0 ) S ( t0 )
we obtain

S ( t1 )
Var ( R(t0 , t1 )) Var ln = n2 hn (n sufficiently large).
S ( t0 )
Equations (5.25) on page 223 and (5.39) on page 227 then yield
2
un
Var ( R(t0 , t1 )) pn (1 pn ) ln
dn
p n ( 1 p n ) ( 4 2 h n ) (n sufficiently large),
where
1 RW ,
pn 1+ hn (n sufficiently large)
2
with
2
RW = m q .
2
Consequently, the Sharpe ratio becomes
(m r) hn
n , (n sufficiently large).
p n (1 p n ) 2 h n
Now, for n sufficiently large, note that

"
((m q) (r q)) hn ((m q) (r q) ) hn
n p n (1 p n ) =
2 hn 2
7 Recall that r is the risk-free rate per annum (by default).

and
1 (m q) , 2 ,
pn 1 + hn hn .
2 2
Hence, for n sufficiently large, it follows
"
1 (r q ) , 2 ,
p n n p n (1 p n ) 1 + hn hn
2 2
1 ,
= 1 + hn
2
pn ,
where
2
= r q .
2
Security Price Formula for a Risk-Neutral CRR Tree
Though the price of a security has the same value in the real world and the
risk-neutral world, its mathematical expression can also be given in terms of
quantities in a risk-neutral CRR tree. In parallel to the discussion for the secu-
rity price formula of a real-world CRR tree (see page 229), we obtain

Z n
S(tn ) S0 en + n (n sufficiently large), (5.55)
where we used n n for sufficiently large n and employed the quantity

n
1
Z n =
Xn,j
n j =1
with

1 pn with probability pn
p n (1 p n )

Xn,j =

p n

p n (1 p n )

We shall obtain the function to which (5.55) converges in the limit n .

Observe that since the risk-neutral security price expression (5.55) equals the
real-world one in (5.44), we have a transformation between Zn and Zn :
(n n )
Z n Z n + (n sufficiently large). (5.56)
n
We shall also show this transformation explicitly in the continuous-time limit
(see (5.70) on page 243).
5.3 Continuous-Time Limit of the CRR Pricing Formula 237
5.3 Continuous-Time Limit of the CRR Pricing Formula
The goal is to determine an explicit expression for the random price of a se-
curity in the continuous-time limit n of the discrete CRR pricing formu-
las for the real-world and risk-neutral world CRR trees. We shall carry out the
analysis in detail for the real-world CRR tree. The risk-neutral case is the same,
except for a minor relabeling of the notation.
5.3.1 The Lindeberg Central Limit Theorem
For a real-world CRR tree, Equation (5.44) on page 230 expresses the price of a
security as follows:
S(tn ) = S0 en + n Zn .
Here
n
1
Zn =
n
Xn,j ,
j =1
S(t j )

with Xn,j the standardization of the log return ln S ( t j 1 )
(see (5.42)):
S(t j )

ln S ( t j 1 )
n h n
Xn,j = ,
n hn
where j = 1, . . . , n. Note that
1 n
n j
Var(Z n ) = E X n,j = 1.
2
(5.57)
=1
In fact, each Z n is the standardization of nj=1 Xn,j . In terms of the uptick prob-
ability pn , we have explicitly (see (5.43))
1 p n

with probability pn
p n (1 p n )
Xn,j = (5.58)

pn with probability 1 pn .
p n (1 p n )
Table 5.2 lists some of the random variables Xn,j and Z n . Note that the row
sequences of Xn,j s form a triangular-array pattern. For each n 1, the random
variables Xn,1 , Xn,2 , . . . , Xn,n in the nth row are independent (since the time-step log
returns are independent) and identically distributed, i.e., the probability measures
P n,1 , P n,2 , . . . , P n,n of Xn,1 , Xn,2 , . . . , Xn,n , respectively, are the same:
Table 5.2 A triangular array of the random variables Xn,j , where n = 1, 2, 3, . . . and j = 1, . . ., n. Each
row sequence Xn,1 , Xn,2 , . . ., Xn,n , where n = 1, 2, . . ., consists of i.i.d. standardized random variables,
whereas the sequence Z1 , Z2 , . . ., is not i.i.d.
n Sequence Zn
1 X1,1 X1,1
1
2 X2,1 , X2,2 ( X2,1 + X2,2 )
2
1
3 X3,1 , X3,2 , X3,3 ( X3,1 + X3,2 + X3,3 )
3
.. .. ..
. . .
1
n Xn,1 , Xn,2 , Xn,3 , . . ., Xn,n ( Xn,1 + Xn,2 + Xn,3 + + Xn,n )
n
.. .. ..
. . .

1 pn pn
P n,j Xn,j = , = pn , P n,j Xn,j = , = 1 pn
p n (1 p n ) p n (1 p n )
for j = 1, . . . , n. It may also be tempting to apply the classical Central Limit The-
orem (CLT)8 to conclude that Z n = 1n nj=1 Xn,j converges in distribution to a
standard normal random variable as n . However, this is not allowed because
X1,1 , X2,1 , X2,2 , X3,1 , X3,2 , X3,3 , X4,1 , . . . are not identically distributed. In fact,
if n = n , then Xn,j and Xn ,j have different probability distributions since
pn = pn .
To determine the convergence of Z n as n , we employ a generalization
of the classical CLT due to Lindeberg. Consider a triangular array of random
variables, namely, the following collection of rows of random variables:
X1,1
X2,1 , X2,2
X3,1 , X3,2 , X3,3
..
.
Xn,1 , Xn,2 , Xn,3 , ..., Xn,n
..
. (5.59)
8 Theorem. (Classical CLT) Assume that X1 , X2 , . . . are i.i.d. random variables with each having a
finite mean E ( Xi ) = 0 and finite variance Var( Xi ) = 02 > 0. Let Zn be the standardization of the
sample mean Xn = n1 ( X1 + + Xn ), namely,

Xn 0 1 n X j 0
Zn = =
( 0 / n) n
0
.
j =1
Then Zn converges in distribution to a standard normal Z as n .

Assume that the random variables Xn,1 , Xn,2 , . . . , Xn,n in the nth row are independent
for every n (i.e., every row) and satisfy
n
E Xn,j = 0 for j = 1, . . . , n, and E X2n,j = 1. (5.60)
j =1
For the array (5.59), we do not require that the random variables in any row be identi-
cally distributed and nor do we require that those in different rows are independent.
The Lindeberg Condition
Before we can get the desired generalization of the classical CLT, we need a
constraint, called the Lindeberg condition, on the random variables in the rows of
(5.59). Our discussion will be in parallel with the treatment in Pitmans lecture
[14]; see Section 27 and Theorem 27.2 of Billingsley [1] for more.
Our assumptions about the array in Equation (5.59) yield that the summa-
tion in (5.60) is actually the total variance of the nth row:

n n
Var Xn,j = E X2n,j = 1, (5.61)
j =1 j =1

where Var Xn,j = E X2n,j . Note that the largest value of the variance terms
in the summation (5.61) is less than or equal to 1, and, if the largest value
equals 1, then it is attained by only one variance term and the other terms
must vanish. Intuitively speaking, the Lindeberg condition is the requirement
that as one moves further and further down the array in (5.59) (i.e., as n ),
the variances of the random variables in each row become smaller and smaller.
To state the Lindeberg condition precisely, recall the definition of the indica-
tor function on { X > c}, where X is a random variable and c a real number:

1 if X > c
1{ X > c } =

0 if X c.
Then the Lindeberg condition is defined as follows:

n
For every fixed > 0, lim
n
E X2n,j 1{|X |> } = 0.
n,j
(5.62)
j =1
Note that as n in (5.62), the value of remains unchanged since it is fixed

during the limit.
To see how the Lindeberg condition implies that the variance of each ele-
ment in a row of the array gets smaller as n , first observe that for every
random variable Xn, i in the nth row, we have
X2n, i 2 + X2n, i 1{|X |> } (i = 1, . . . , n).

n,i
Since the inequality is between nonnegative quantities, taking the expectation

yields an upper bound on the variance of Xn,i :

Var (Xn, i ) 2 + E X2n, i 1{|X |> } (i = 1, . . . , n). (5.63)
n, i
Adding to the upper bound in Equation (5.63) the contributions (each of which
is also nonnegative)
from the remaining terms in the nth row, namely, the sum
E Xn,j 1{|Xn,j |> } , we find
2
1 j = i n
n
Var (Xn, i ) 2 + E X2n,j 1{|X |> }
n,j
(i = 1, . . . , n).
j =1
Because this upper bound holds for the variance of each random variable in
the nth row, it holds in particular for the largest value of those variances:
n
max Var (Xn, i ) 2 +
1 i n
E X2n,j 1{|X |> }
n,j
(i = 1, . . . , n). (5.64)
j =1
By Equation (5.64), if the array (5.59) satisfies the Lindeberg condition, then
as one moves down the array, the maximum of the variances of the random
variables in each row starts to approach zero:
lim max Var (Xn, i ) 2 for all >0 (no matter how small), (5.65)
n 1 i n
i.e.,
lim max Var (Xn, i ) = 0.
n 1 i n

Note: since we always have Var Xn,j 1, Equation (5.65) gives no new con-
straint on the maximum variance for 1, but it does for 0 < < 1.
The Lindeberg CLT
With the Lindeberg condition available, we can now state the following gener-
alization of the classical CLT:
Theorem 5.1. (The Lindeberg Central Limit Theorem) Assume that the trian-
gular array of random variables,
X1,1
X2,1 , X2,2
X3,1 , X3,2 , X3,3
..
.
Xn,1 , Xn,2 , Xn,3 , ..., Xn,n
..
.
is such that the random variables Xn,1 , Xn,2 , . . . , Xn,n in the nth row are independent
for every n 1 (i.e., every row) and satisfy
n
E Xn,j = 0 for j = 1, . . . , n, and E X2n,j = 1.
j =1
If the triangular array obeys the Lindeberg condition, then
n
d
Zn = Xn,j Z as n ,
j =1
where Z is the standard normal random variable.
Readers are referred to Billingsley [1, p. 360], Durrett [6, p. 129], and Pitman
[14] for proofs of Theorem 5.1.9
The Lindeberg CLT tells us that, for n sufficiently large, all probabilities
about Z n can be approximated by those of a standard normal random vari-
able Z. Explicitly, for n sufficiently large and for every real number x, we have
P n (Z n x ) P ( Z x ) = N( x ) ( < x < , n sufficiently large),

2x y2
where N( x ) is the c.d.f. of Z at x, i.e., N( x ) = 1
e 2 dy.
2
5.3.2 The Continuous-Time Security Price Formula
We now show that the Lindeberg CLT applies to both the real-world and risk-
neutral CRR security price formulas.
Continuous-Time Real-World Security Price Formula
Starting with the real-world CRR tree, consider the triangular array of random
variables Xn,j in Table 5.2 associated with the security price formula (5.44) on
page 230. Let
9 The classical CLT follows from the Lindeberg CLT through the Dominated Convergence Theorem
(expectation form) from measure theory; see Durrett [6, pp. 29,129].
Xn,j
Xn,j = .
n
To check that these random variables obey the hypotheses of the Lindeberg
CLT, recall that for each n 1, the random variables Xn,1 , Xn,2 , . . . , Xn,n are
independent and standardized, and satisfy (5.57) on page 237. Then for ev-
ery n 1, the random variables Xn,1 , Xn,2 , . . . , Xn,n are also independent and
satisfy (5.60). Next, to see that the Lindeberg condition holds, Equation (5.58)
yields
1 p n

with probability pn
1 p n (1 p n )
Xn,j =
n
p n (1 p n )
Since pn 1/2 as n (see (5.37) on page 226), we get
Xn,j 0 as n .
This implies that, for any fixed > 0, the following condition does not hold for
n sufficiently large: 4 4
4Xn,j 4 > > 0.
Consequently, 1{|X |> } = 0 for n sufficiently large and so

n,j

E X2n,j 1{|X |> } = 0 (n sufficiently large).
n,j
Hence, as n , the Lindeberg condition (5.62) holds. By the Lindeberg CLT,

we obtain
n
d
Zn = Xn,j = Z n Z as n .
j =1
Thus, the discrete-time real-world CRR security price at the future time t f = t0 + ,
which is labeled by tn in an n-period real-world CRR tree, converges in distribution to
the following continuous-time security price formula at t f :

S0 en + n S0 eRW +
Zn d Z
as n . (5.66)
Here = t f t0 , RW = m q 2 (i.e., the real-world drift), and the standard

2
normal Z is randomly drawn at a time span from the present.
Notation. In security price modeling, we typically choose the current time to be

t0 = 0 and allow the final time t f to be a variable t. By (5.66), the continuous-time
security price at time t is then written as

S(t) = S0 eRW t +
d t Zt
(t 0), (5.67)
where S0 = S(0) and Zt N (0, 1). Interpret the standard normal Zt to mean
that its realizations are randomly drawn at time t. We insert the subscript t
rather than use the notation Z to avoid some potential notational and concep-
tual confusion later; see page 246. Notice the dependency of S(t) on the length
t of the interval [0, t]. We shall also switch freely between S(t) and St as nota-
tion for the securitys price at time t and, more generally, between X (t) and Xt
for a random variable dependent on t:
S ( t ) = St , X ( t ) = Xt .
Continuous-Time Risk-Neutral Security Price Formula
The security price is the same in the real world and the risk-neutral world, but
has different expressions and probabilities. Explicitly, we have

Xn,j n
Xn,j = , Zn = Xn,j ,
n j =1
where

1 pn with probability pn
p n (1 p n )

Xn,j =

p n

p n (1 p n )

Essentially the same arguments as in the real-world case carry over to the risk-
neutral setting to allow application of the Lindeberg CLT. For example, Equa-
tion (5.51) on page 234 shows that pn 1/2 as n , which in turn yields
the following condition used to verify the Lindeberg condition: Xn,j 0 as
n . Since n = as n , the Lindeberg CLT yields
Z n

Z
S0 en + n S0 e +
d
as n , (5.68)
where = r q 2 (i.e., the real-neutral drift) and Z is a standard normal

2
random variable drawn at a time span, = t f t0 , from the present relative to

a risk-neutral probability. By our continuous-time convention that t0 = 0 and
t f = t, where t is a variable, the resulting continuous-time risk-neutral security
price is

t Zt
S ( t ) = S0 e t +
d
(t 0). (5.69)
Equations (5.67) and (5.69) yield the relationship between the real world and
risk-neutral standard normal random variables:

d (RW ) t mr
Zt = Zt + = Zt + t, (5.70)
t
which is the continuous-time analog of Equation (5.56) on page 236. Note that
m r
is the securitys Sharpe ratio.
A Remark About the Notation for Security Prices
As mentioned in Remark 5.1 on page 210, to avoid notation that is too cumber-
some, we do not introduce any new notation to distinguish the security price
formula in discrete time versus continuous time. For example, given a parti-
tion, 0 = t0 < t1 < < tn = t, of the interval [0, t], it will be made clear from
the context whether the real-world security price S(t j ) at time t j is given by

t + n t j Z n t + t j Zt j
S0 e n j (discrete time) or S0 e RW j (continuous time).
Similarly, for a risk-neutral world, we shall clarify whether we intend

t j + n t j Z n t j + t j Zt
S0 e (discrete time) or S0 e j (continuous time).
In fact, for most of our study going forward, we shall use the continuous-time
formula and abide by the following:
Unless stated to the contrary, assume that the price of a security is

expressed by the continuous-time formula (5.67) for the real-world
setting or formula (5.69) for a risk-neutral world.
Some Assumed Properties of the Continuous-Time Security Price
We now motivate and present several properties we shall assume about secu-
rity prices for continuous time.
Continuous sample paths. For an n-period CRR tree, a possible outcome
0 is a sequence of n up and down movements in a securitys price. There
are 2n such possible outcomes giving rise to 2n possible security price
paths. In the continuous-time limit n , we have assumed that the result-
ing uncountably many sample paths of the security, whether in a real world or
risk-neutral world, are continuous functions of time t with probability 1. Later,
we shall explore the discontinuous case, where the price paths can have
jumps; see the Merton jump-diffusion model in Section 8.9 (page 448).
Markov property and the Weak Efficient Market Hypothesis. The futures
price of a security depends on the current price S(0) and does not depend ex-
plicitly on the securitys past prices; see Equations (5.67) and (5.69). That is,
probabilities about futures prices of a security are independent of the secu-
ritys past price path (Markov property). Nonetheless, information about the
past is not totally excluded in the sense that such information is included
in the current price. Indeed, one of our key assumptions is the Weak Efficient
Market Hypothesis, which states that the current price of a security reflects
all market information concerning the security. In particular, the securitys
past price path and present market expectations about the securitys future
behavior are already taken into account in the current price of the security.
Stationarity of log returns. By Equations (5.67) and (5.69), the securitys
continuous-time log return from time 0 to t does not depend on the price
of the security at the start of the interval [0, t]. It depends on the length of
the interval:
S ( t) d
ln = RW t + t Zt ,
S (0)

where t 0. Similarly, in a risk-neutral world ln SS((0t)) = t + t Zt . Mo-
d
tivated by this, assume that in the continuous-time setting, the log return of a
security is stationary, which means

S ( t) d S ( t + u)
ln = ln
S( x ) S ( x + u)
for all 0 x < t and every u such that x + u 0 and t + u 0. In other
words, rigidly shifting the interval does not change the log return probabilities.
Independence of log returns over nonoverlapping10 intervals. For the n-
period CRR tree, the discrete-time log returns over different time steps are
independent. Motivated by the latter, we assume in the continuous-time set-
ting that for any finite sequence of times, 0 t1 < t2 < < tk1 < tk , which is
not required to coincide
with the time steps of the CRR tree we studied earlier, the
S(t j )
log returns ln S(t ) are independent for j = 2, . . . , k.
j 1
Geometric Brownian motion model of security prices. The price of a secu-

rity is an example of a stochastic process, which, intuitively speaking, is a
random variable that is a function of time. In Chapter 6, we shall model the
continuous-time price of a security more precisely as a stochastic process
called geometric Brownian motion, namely,
S(t) = S(0) eRW t + B(t) ( t 0).
Here B(t), where t 0, is also a stochastic process, called standard Brownian

motion, and equals in distribution the following:
d
B(t) = t Zt .
An important property (stationarity) of standard Brownian motion is
10 Recall that two intervals are nonoverlapping if their interiors are disjoint.
d
B(t) B() = B(t ) N (0, t ) (0 t ).
d
Note that if we used the notation B(t) = t Z, then it is tempting to write
d
B(t) B() = tZ
Z = ( t ) Z,

which is incorrect since it states B(t) B() N 0, t + 2 t .
The risk-neutral expression of the securitys price is

(t)
S ( t ) = S (0) e t + B ( t 0),
where B (t) is also standard Brownian motion, but in a risk-neutral world.

d
We have B (t) = t Zt , where the asterisk indicates that the probability
distribution is risk neutral. Chapter 6 and Section 8.4 will have more on
risk-neutral probabilities. Note that Equation (5.70) gives the link between
Zt and Zt .
5.4 Basic Properties of Continuous-Time Security Prices
For each future moment of time t > 0, the continuous-time security price,

S(t) = S(0) eRW t +
d t Zt
,
is a lognormal random variable with p.d.f.

2
ln x
t
1 S (0 ) RW

f S(t) ( x ) = exp ( t > 0).
2t x 2 2 t
5.4.1 Some Statistical Formulas for Continuous-Time Security Prices
The statistics of a lognormal random variable are well known and readily yield
formulas for the mean, median,11 variance, and covariance of continuous-time
security prices:
11The median of an absolutely continuous random variable X is a number, denoted Med( X ), such
that P ( X > Med( X )) = 12 = P ( X < Med( X )) . Additionally, if X N ( a, b2 ), where a and b > 0 are

constants, then the mean and median of X satisfy E e X = e a +b /2 and Med A e a X = A e a Med( X ) .
2
5.4 Basic Properties of Continuous-Time Security Prices 247
E (S(t)) = S(0) e(mq) t ( t 0) (5.71)

2
2 t
Med(S(t)) = e E (S(t)) ( t 0) (5.72)
2 2
Var(S(t)) = e t 1 E (S(t)) ( t 0) (5.73)
2
Cov(S(t), S(u)) = e u 1 E (S(u)) E (S(t)) (t > u 0). (5.74)
Equation (5.71) shows that the expected price of the continuous-time security at
a future time t is obtained by continuously compounding the current price S(0) at
the instantaneous capital-gain rate m q over the time span t. Also, since > 0,
Equation (5.72) readily shows that the median of a continuous-time security price
at t > 0 is always below its mean:
0 < Med(S(t)) < E (S(t)) ( t > 0). (5.75)
This is due to the skewness of the lognormal p.d.f. Equations (5.73) and (5.74)
yield, since and the mean price E (S(t)) are positive, that the variance and
covariance of the securitys prices are positive for all future times t > u > 0.
We can also obtain formulas for the conditional expectations of a continuous-
time security price (Exercise 5.23):12
N(d+,RW (t))
E (S(t) | S(t) > K ) = E (S(t)) (5.76)
N(d,RW (t))
N(d+,RW (t))
E (S(t) | S(t) < K ) = E (S(t)) , (5.77)
N(d,RW (t))

ln
S ( 0)
+RW t
where t > 0, K > 0, and d,RW (t) = with d+,RW (t) = d,RW (t) +
K
t.
t
5.4.2 Some Probability Formulas for Continuous-Time Security Prices
We present some probability results for certain basic continuous-time security

price behavior.
Since the median of a securitys price is below the mean, there is more than a 50%
probability that the futures price of a security will be below its mean price. To see
this, observe that

P S(t) < E (S(t)) = P S(t) < Med(S(t))

+ P Med(S(t)) S(t) E (S(t)) .
12 Recall that the conditional expectation of an absolutely continuous random variable V given an
E (V 1 ) 2
event A with probability P ( A) > 0 is defined by E (V | A) = P ( AA) = P (1A ) A v f V ( v) dv, where f V is
the p.d.f. of V.

By definition of the median, we have P S(t) < Med(S(t)) = 1
2 and by
(5.75) we see that P Med(S(t)) S(t) E (S(t)) > 0. Hence
1
P S(t) < E (S(t)) > . (5.78)
2
We made a similar observation in Example 5.6 (page 227) using a 100-
period CRR tree.
If the securitys volatility is sufficiently large, then there is also more than a 50%
probability that the futures price of the security will be below its current price. In
fact, note that (5.72) implies
2
2 + m q t 2 t
Med(S(t)) = S(0) e S (0) e 2 < S (0)
for sufficiently large. Then an argument similar to the one above yields
1
P S ( t ) < S (0) > ( sufficiently large). (5.79)
2
Example 5.6 (page 227) also pointed out this property using a CRR tree.
For any K > 0, the probability that S(t) is less than K is (Exercise 5.24)

P S(t) < K = N(d,RW (t)) ( t > 0). (5.80)
Here N( x ) is the standard normal c.d.f. at x. Since N( x ) = 1 N( x ), the

probability that the securitys price at t > 0 exceeds K is

P S(t) > K = 1 P S(t) < K = 1 N(d,RW (t)) = N(d,RW (t)).
(5.81)
The probability of the securitys price lying between two values is (Exercise 5.24)
K K
P K1 < S(t) < K2 = N(d1,RW (t)) N(d2,RW (t)), (5.82)

S ( 0)
ln +RW t
where 0 < t, 0 < K1 < K2 , and ,RW (t) =
x x
d .
t
We add that the risk-neutral analogs13 of the conditional expectation in

Equation (5.76) and probability in Equation (5.81) appear in the pricing for-
mula for a European call. Similarly, risk-neutral analogs of Equations (5.77)
and (5.80) are part of the pricing formula for a European put. Readers are re-
ferred to McDonald [12, Sec. 18.4] for a detailed treatment.
13 Replace m by r and (hence) RW by in the formulas.

5.5 Exercises 249
5.5 Exercises
5.1. Suppose that a binomial tree has an initial price of $80. If the tree has 21
periods, then is $80 one of the possible prices at time t21 ? If the tree has 300
periods, then is $80 one of the possible prices at time t300 ? Justify your answers.
5.2. For an n-period binomial tree, give a financial interpretation of each of the
following: un dn = 1, un 1, and dn 1.
5.3. Explain why the condition, dn < e(mq)hn < un , holds for n-period CRR
trees with n sufficiently large.
5.4. How many 1-period subtrees are in an n-period binomial tree?
5.5. For an n-period binomial tree, let NU be the number of security price
upticks from time t0 to tn . Explain why NU is a binomial random variable.
What are its expected value and variance if the tree has 40 steps and the uptick
probability is 60%?
5.6. An n-period CRR tree has reflection symmetry about the horizontal line
S(t) = S0 since the tree recombines. Agree or disagree? Justify your answer.
5.7. For an n-period risk-neutral binomial tree, show that if un < e(rq) hn , then
there is an arbitrage.
5.8. The risk-neutral uptick probability pn is related to the real-world probabil-
, E ( R(t ,t ))r h n
ity pn by pn = pn n pn (1 pn ), where n = 0 1 . Interpret n .
Var( R(t0 ,t1 ))
5.9. If in an n-period real-world CRR tree, the real-world probability pn is re-

placed by the risk-neutral uptick probability pn , then the expected annualized
return rate m is unchanged, but the annualized variance 2 changes. Agree or
disagree? Justify your answer.
n
5.10. For an n-period binomial tree, the collection of all paths has 22 subcollec-
tions of paths, where the empty subcollection is included. Agree or disagree?
Justify your answer.
5.11. A trader believes that a certain stock currently at $51.25 per share has by
the end of the trading day a 70% chance of increasing by 50 and a 30% chance
of decreasing by 25. Using a 1-step binomial tree with this information, what
is the expected price of the stock at the end of the day?
5.12. Assume that the current share price of a stock is $100 with a volatility
of 10%. Using a CRR tree model over a year with each being one trading day,
predict the maximum spread in the stocks possible prices a trading day from
now. Is your prediction impacted if you employ a Taylor approximation to
e h252 using the CRR assumptions?
5.13. Suppose that a nondividend-paying stock with current price of $45 has
an instantaneous annual expected return of 8% and annual volatility of 15%.
a) Using a 100-period CRR tree, forecast the price of the stock 3 months from
now, i.e., find the expected price of the stock 3 months from now.
b) What is your forecast if you use an 80-period CRR tree?
c) Using an 80-period CRR tree, determine the probability of your forecasted
price occurring.
5.14. Consider a nondividend-paying stock with a current price of $45, an in-

stantaneous annual expected return of 8%, and annual volatility of 15%. As-
sume an 80-period CRR tree.
a) Can the stock price be at the same value 3 months from now? If so, how
many times would the price have to increase and decrease for this to hap-
pen?
b) What is the probability that the stock price will be at the same value 3
months?
c) What is the probability that in 3 months the stock price is greater than its
current price?
5.15. Suppose that the current date is January 21, 2016. Estimate the volatility
and instantaneous drift RW for Google (ticker symbol GOOGL) using its ad-
justed closing prices from Yahoo! Finance for the period from January 20, 2016
to September 15, 2015. The data will consist of 90 daily log returns over the
past 91 trading days. Carry out similar estimates with the past 60 daily log re-
turns and then the past 30 daily log returns. Annualize your results using 252
trading days in a year. Discuss your findings.
5.16. Assume that a nondividend-paying security has a current price of $50,

instantaneous expected return of 8%, and volatility of 15%. Estimate the prob-
ability, as a fraction (not percentage) to two decimal places, that 3 months from
now, the price of the real-world, continuous-time price of the security is greater
than $50. What fractional probability to two decimal places does a 100-period
real-world CRR tree predict? How about a 1,000-period real-world CRR tree?
Find a value of n for which an n-period real-world CRR tree gives the same
fractional probability to two, three, and four decimal places as that obtained
from the continuous-time security price model. Use a software for this prob-
lem.
5.5 Exercises 251
5.17. For an n-period binomial tree, show that

n
i n i
P S ( t n ) = S0 u n d n | S ( t 0 ) = S0 = pin (1 pn )ni , i = 0, 1, 2, . . . , n.
i
5.18. Determine the number of elements in the sample space n of price paths
of an n-period binomial tree.
2
5.19. Show that n2 hn = pn (1 pn ) ln udnn .
(m q) hn
5.20. Show that for a CRR tree, the uptick probability pn e un d dn
satisfies
RW n
pn 2 1 +
1
hn for n sufficiently large.
1 p n

with probability pn
p n (1 p n )
5.21. For a CRR tree, show Xn,j =

p n (1 p n )
5.22. A Jarrow-Rudd (JR) tree for the price of a security paying a continuous
dividend yield rate q is a binomial tree where pn = 12 and the per-period ex-
pectation n and variance n2 defined by

1 S ( t j ) 1 S(t j )
n = E ln , (n )2 = Var ln ,
hn S ( t j 1 ) hn S ( t j 1 )
satisfy n and n as n , where = r q 2 . Then for n suffi-

2
ciently large, the constraint equations for un , dn , pn in a JR tree are

2
1 un
pn = , hn pn ln un + (1 pn ) ln dn , hn pn (1 pn ) ln
2
.
2 dn

a) Show that un e hn + hn and dn e hn hn for n sufficiently large.
b) Show that the security price formula for a JR tree has the following form for

n sufficiently large: SJR (tn ) S0 e + ZJR where ZJR JR
n = 1 nj=1 Xn,j
n ,
n
JR
and Xn,j 1 with probability 1/2 for each possibility.
XJR
c) Verify that the triangular array XJR
n,j = n , where n 1 and j = 1, . . . , n, sat-
n,j
isfies the hypotheses of the Lindeberg CLT.

d) Using the Lindeberg CLT, determine the continuous-time security price for-
mula to which the JR tree security price converges in distribution. Is this the
same formula obtained by using a risk-neutral CRR tree?
e) (Risk-Neutral JR Tree) Is the JR tree risk neutral? If not, then how would
you make it risk neutral? For a sufficiently large n, what would be the ap-
proximate governing equations of a risk-neutral JR tree, i.e., the equations
expressing un , dn , pn in terms of the inputs r q, , hn ?
N(d (t))
5.23. For t > 0 and K > 0, show that E (S(t) | S(t) > K ) = E (S(t)) N(d+,RW (t))
,RW
N(d (t))
and E (S(t) | S(t) < K ) = E (S(t)) N(d+,RW (t)) .
,RW

5.24. For t > 0, K > 0, and K2 > K1 > 0, show that P S(t) < K = N(d,RW (t))
K
and P K1 < S(t) < K2 = N(d1,RW (t)) N(dK ,RW (t)).
2
References
[1] Billingsley, P.: Probability and Measure, 3rd edn. Wiley, New York (1995)
[2] Chance, D.: Proofs and derivations of binomial models. Technical Finance
Notes, Louisiana State University (2007)
[3] Chance, D.: Risk neutral pricing in discrete time. Teaching Note 96-02,
Louisiana State University (2008)
[4] Cox, J., Ross, S., Rubinstein, M.: Option pricing: a simplified approach. J.
Financ. Econ. 7, 229 (1979)
[5] Cox, J., Rubinstein, M.: Options Markets. Prentice Hall, Upper Saddle
River (1985)
[6] Durrett, R.: Probability: Theory and Examples, 4th edn. University Press,
Cambridge (2010)
[7] Epps, T.: Pricing Derivative Securities. World Scientific, Singapore (2007)
[8] Ghahramani, S.: Fundamentals of Probability with Stochastic Processes.
Pearson Prentice Hall, Upper Saddle River (2005)
[9] Jarrow, R., Rudd, A.: Option Pricing. Richard Irwin, Homewood (1983)
[10] Jarrow, R., Turnbull, S.: Derivative Securities. South-Western College
Publishing, Cincinnati (2000)
[11] Leunberger, D.: Investment Science. Oxford University Press, Oxford
(1998)
[12] McDonald, R.: Derivative Markets. Addison-Wesley, Boston (2006)
[13] Neftci, S.: An Introduction to the Mathematics of Financial Derivatives.
Academic, San Diego (2000)
[14] Pitman, J.: Setup for the Central Limit Theorem. STAT 205 Lecture Note
10, scribe D. Rosenberg. University of California, Berkeley (2003)
[15] Roman, S.: Introduction to the Mathematics of Finance. Springer, New
York (2004)
[16] Wilmott, P., Dewynne, N., Howison, S.: Mathematics of Financial Deriva-
tives: A Student Introduction. Cambridge University Press, Cambridge
(1995)
Chapter 6
Stochastic Calculus and Geometric Brownian
Motion Model
Stochastic calculus is the methodology of choice in modern finance: it provides

intuitive constructions of financial objects and is the most identified mathemat-
ical tool with financial engineers.
The materials in stochastic calculus covered in this chapter mainly serve as
a preparation for pricing derivatives in a later chapter.
6.1 Stochastic Processes: The Evolution of Randomness
6.1.1 Notation for Probability Spaces
A probability space is denoted by the triple (, F, P ), where:

1. denotes the sample space, that is, a nonempty set of all the possible outcomes
of a random experiment;
2. F denotes a -algebra on . For our purpose, it is sufficient to know that F
represents a collection of subsets of , called events;
3. P denotes a probability measure on F that assigns probabilities to events. (Note
that P : F [0, 1] is a function whose domain is F.)
The next remark provides intuitive explanations of 2 and 3 and motivates
the concept of -algebra.
Remark 6.1. Ideally, one would like to assign a probability for every subset
of . Unfortunately, because of some obstacles in measure theory, this cannot
always be done. It is often the case that one can only assign a probability P on
a collection F of subsets of . An element in F is called an event in probability
theory. For further development of the probability theory, we need additional
algebraic structures for F as intuitively explained below:

254 6 Stochastic Calculus and Geometric Brownian Motion Model
A probability P should be additive in the sense that if A and B are disjoint

events, then
P ( A B ) = P ( A ) + P ( B ),
and therefore F should be closed under the union operation. For the appli-
cation of modern integration theory, one actually wants P to have a slightly
deeper nature that for a countable pairwise disjoint collection of events Ai , i =
1, 2, . . . ,
P ( A1 A2 ) = P ( A1 ) + P ( A2 ) + ,
and therefore F should be closed under countable union. Furthermore, the
empty set should be an event with probability zero, and if A is an event,
then the complement of Ac should also be an event such that
P ( A) + P ( Ac ) = 1.
By Morgans law, finite or countable intersections of events should also be

events.

Although intuitive explanations of the -algebra will be provided in great

detail shortly, the definitions scope best captures the intuitive ideas.
Let be a nonempty set. The power set of , denote by 2 , is a set consisting
of all subsets of .
Definition 6.1. Let be a nonempty set. F 2 is called an algebra over if

it satisfies the following conditions:
1. F ;
2. A F Ac F ;
3. A, B F A B F .
Note that the last condition is equivalent to closure under finite union, i.e.,
Ai F , i = 1, 2, . . . , n ni=1 Ai F .
If the last condition is replaced with closure under countable union, i.e.,
Ai F , i = 1, 2, . . . ,
i =1 A i F ,
then F is called a -algebra over .
Note that the definition of -algebra does not rely on the existence of a proba-
bility measure.
The next definition is a natural extension.
Definition 6.2. Let Fi , i = 1, 2, be two -algebras on a nonempty set . F2 is

said to be a sub--algebra of F1 if F2 F1 .
6.1 Stochastic Processes: The Evolution of Randomness 255
Definition 6.3. For C 2 , the smallest -algebra that includes C is called the
-algebra generated by C and denoted by (C ).
In fact, (C ) is the intersection of all -algebras including C.
Example 6.1. Let A . The collection {, A, Ac , } is the smallest -algebra
on containing A, i.e., ( A), the -algebra generated by A.

Example 6.2. {, } is the smallest -algebra on , whereas 2 , the power set
of , is the largest -algebra on .
Clearly, for any C 2 , {, } is a sub--algebra of (C ), and (C ) is a
sub--algebra of 2 .

In modern probability theory, a probability is defined on a -algebra as a
measure with whole space measure equal to one. Precisely, we define the prob-
ability measure as follows:
Definition 6.4. Let be a probability space and F a -algebra over . A real-
valued function P from F to [0, 1] is said to be a probability measure on F if it
satisfies the following conditions:
1. P () = 1;
2. For each countable collection { Ai F, i I } of pairwise disjoint sets,
P (i I Ai ) = P ( Ai ).
i I
This property is referred to as the countable additivity.

Note that items 1 and 2 in the definition imply P () = 0.
A proper sub--algebra F may be considered as a coarsification of F, in the
sense that fewer events are observable and this in turn, as we will see in the
next section, implies that fewer random variables are measurable. In this
regard, F gives rise to lower resolution when one tries to observe a random
variable.
Example 6.3. (One-Period Binomial Tree) Consider a random experiment of
tossing a biased coin. Assume that the chance of landing heads up is 30% and
that of landing tails up is 70%. Then, the corresponding probability space is
represented by (, F, P ), where
= { H, T },
F = {, H, T, },
P : F [0, 1] is defined by
P () = 0, P ( H ) = 0.3, P ( T ) = 0.7, P () = 1.
In a parallel fashion, if we consider an oversimplified stock price behavior

in which the stock price either goes up by a factor u with probability p or goes
down by a factor d with probability 1 p over the time period [t0 , t1 ], then the
corresponding probability space is represented by (, F, P ), where
= {U, D },
F = {, U, D, },
P () = 0, P (U ) = p, P ( D ) = 1 p, P () = 1.

The random experiment associated with the next example is parallel to re-
peatedly flipping a biased coin.
Example 6.4. (Two-Period Binomial Tree) Consider an oversimplified stock

price behavior as described by a two-period ([t0 , t1 ] and [t1 , t2 ] with t0 < t1 < t2 )
binomial tree. In each period the stock price either goes up by a factor u with
probability p or goes down by a factor d with probability 1 p. Then, the
corresponding probability space is represented by (, F, P ), where
= {1 , 2 , 3 , 4 }, where
1 = UU, 2 = UD, 3 = DU, 4 = DD,
F = {, {1 }, {2 }, {3 }, {4 },
{ 1 , 2 } , { 1 , 3 } , { 1 , 4 } , { 2 , 3 } , { 2 , 4 } , { 3 , 4 } ,
{ 1 , 2 , 3 } , { 1 , 2 , 4 } , { 1 , 3 , 4 } , { 2 , 3 , 4 } , } ,
P ({1 }) = p2 , P ({2 }) = P ({3 }) = p(1 p), P ({4 }) = (1 p)2 ,
which along with additive property of P (see Remark 6.1) imply
P ({1 , 2 , 3 }) = 1 (1 p)2 , P ({2 , 3 , 4 }) = 1 p2 ,
P ({1 , 3 , 4 }) = P ({1 , 2 , 4 }) = 1 p(1 p),
P ({1 , 2 }) = P ({1 , 3 }) = p, P ({2 , 3 }) = 2p(1 p),
P ({2 , 4 }) = P ({3 , 4 }) = 1 p, P ({1 , 4 }) = p2 + (1 p)2 ,
P () = 0, P () = 1.
Note that each simple event in is a representation of (or corresponds to)

a path (i.e., (S(t0 ), S(t1 ), S(t2 )), where S(ti ), i = 0, 1, 2 denote the stock price at
time ti ).
Also, note that F is closed under countable union and intersection, and
F = {, {1 }, {2 , 3 , 4 }, }
is a sub--algebra.

6.1.2 Basic Concepts of Random Variables
A random variable on a probability space (, F, P ) is a measurable real-valued

function from to R. That is
X : R,
X ( ),
where measurability refers to the -algebra F and means that for each real
number a, the set { X a} is an event in the probability space. That is,
X 1 ((, a]) = { X a} = { | X ( ) a} F, (6.1)
where X 1 ((, a]) is the pre-image of (, a] under X.

In words, (6.1) says that X is F-measurable. Thus,
a random variable X on (, F, P ) is a F-measurable function from to R;
the notion of X being F-measurable assures that the distribution function
of X: FX ( a) P ( X a) = P ( X 1 ((, a])) is well defined. In fact, a sig-
nificant attribute of random variables is that they allow us to work on dis-
tributions without recognizing sample spaces.
For our purpose, if a random variable X is known given a sub--algebra

F F, then we say that X is measurable with respect to F and write X F .
Example 6.5. (One-Period Binomial Tree Again) Let us revisit Example 6.3,
the one-period binomial tree model of an oversimplified representation of
stock price behavior in which the stock price either goes up by a factor u
with probability p or goes down by a factor d with probability 1 p over the
time period [t0 , t1 ]. Then the corresponding probability space is represented by
(, F, P ), where
= {U, D },
F = {, U, D, },
P () = 0, P (U ) = p, P ( D ) = 1 p, P () = 1.
Given S(t0 ) = S0 and factors u > 0 and d > 0, we define X : R by

S0 u if = U
X ( ) = (6.2)
S0 d if = D.
Suppose that S0 = 100, u = 6

5 and d = 56 . Then

120 if = U
X ( ) =
83.33 if = D.
To show that X is a random variable on the probability space (, F, P ), we

need to verify (6.1):
X 1 ((, a]) = { X a} = { | X ( ) a}

if a < 83.33
= D if 83.33 a < 120

if 120 a.
Indeed, X 1 ((, a]) F for each real number a:

Using mathematical language, we say that X is F-measurable. Translat-
ing it into ordinary English, we say that F contains sufficient information
about X. This intuition of measurability is important for gaining a correct
intuition of the concept of conditional expectation with respect to -algebra
in a later section.
It follows from similar arguments that X defined by (6.2) is a random variable
on (, F, P ).
Note that the smallest -algebra containing { X a, a R } = {, D, } is F.

In general, ( X ) is the notation for the smallest -algebra containing the

events { X a, a R } and called the -algebra generated by X. In the last
example, ( X ) happens to be F, which, in fact, is also the largest -algebra
containing { X a, a R }.
Let X be an arbitrary random variable. By definition, X is ( X )-measurable.
Example 6.6. Let X be a random variable. Without using mathematical jargon

for the notation ( X ), how one can intuitively describe ( X )?
Intuitive Explanation.
The power set 2X is the largest -algebra. On this -algebra, every random
variable, no matter how complex, is measurable. A particular random variable
may be simpler and therefore measurable on a smaller -algebra. By simpler we
mean that the random variable X is constant on more events. For example,
the simplest random variable X c, where c is a real constant and is measur-
able even on the smallest -algebra {, }. Most random variables require
-algebras in between so that they are measurable. For an economic reason,
we sometimes may want to have a smallest -algebra which is sufficient for
a random variable X to be measurable. Such a smallest -algebra, denoted by

( X ), always exists, as the intersection of all -algebras sufficient for X to be
measurable is itself a -algebra.
In the language of information, this is tantamount to saying that we use the
least amount of information to determine X.

Sometimes random variable X and a -algebra F are both given, but X is not
measurable on F . Intuitively, we say that F does not have enough resolution
to read off all information contained in X. The meaning of E ( X |F ) is the least
coarsification of X so that F has enough resolution for E ( X |F ) (i.e., E ( X |F )
is measurable on F ). More discussion on conditional expectation with respect
to a -algebra will be given shortly.
Remark 6.2. Now we summarize the above intuition of X being F -measurable

(X F ) in the following:
X is F -measurable F contains sufficient information of X

X is determined or known given F .
1. The intuition of the measurability of a random variable as we have provided

in the equivalent statements above has fundamental importance in study-
ing modern finance theory. For instance, it is important for understanding
adapted processes, which in turn is important for understanding martin-
gales. Readers should always have such an intuition in their mental model.
2. If X is not F -measurable, then probabilities of events described by X may

not be computed based on the information contained in F . In this sense, X
is unknown.

Recall that two events A and B are independent if and only if their joint
probability equals the product of their probabilities: P ( A B) = P ( A)P ( B).
Since -algebras are collections of events, the next definition is natural.
Definition 6.5. Two -algebras F and G are said to be independent if
P ( A B ) = P ( A )P ( B ), A F, B G.
In words, two -algebras are independent, if any two events, one from each
-algebra, are independent.
Using the language of independent -algebras, we have an equivalent state-
ment for the definition of independent random variables:
Definition 6.6. Two random variables X and Y are independent if and only if
two corresponding -algebras ( X ) and (Y ) are independent.
Definition 6.7. A random variable X and a -algebra F are said to be indepen-

dent, denoted by X F , if two -algebras ( X ) and F are independent.
6.1.3 Basic Concepts of Stochastic Processes
We begin with a continuation of Example 6.5.
Example 6.7. Recall from Example 6.5 that the probability space is given by
(, F, P ), where
= {U, D },
F = {, U, D, },
P () = 0, P (U ) = p, P ( D ) = 1 p, P () = 1,
and a random variable X is defined by

S0 u if = U
X ( ) =
S0 d if = D,
where S(t0 ) = S0 and factors u > 0 and d > 0.

To extend a one-period tree to a multi-period tree, we rewrite random vari-
able X into

u if = U
X ( ) = S0 Y ( ) where Y ( ) =
d if = D.
If we use a three-period binomial tree (with [ti , ti+1 ], ti < ti+1 and i = 0, 1, 2, 3,
as time periods), to model an oversimplified stock price behavior (with S(ti )
as the price at time ti ) so that in each period the stock price either goes up
by a factor u with probability p or goes down by a factor d with probability
1 p, then we can introduce three random variables on (, F, P ), Yi : R
defined by

u if = U
Yi = Yi ( ) = where i = 1, 2, 3.
d if = D,
Then
X1 = S0 Y1 = S(t1 ), X2 = S0 Y1 Y2 = S(t2 ), X3 = S0 Y1 Y2 Y3 = S(t3 ).
In a similar fashion, we can extend the binomial tree over infinitely many time
periods: [t0 , t1 ], [t1 , t2 ], [t2 , t3 ], . . . , [ti1 , ti ], . . . by introducing infinitely many ran-
dom variables on (, F, P ), Yi : R defined by

u if = U
Yi = Yi ( ) = where i = 1, 2, 3, . . . , (6.3)
d if = D,
which form a sequence of random variables denoted by
Y = {Y1 , Y2 , Y3 , . . . }, or Y = {Y1 ( ), Y2 ( ), Y3 ( ), . . . }, .
A deterministic counterpart of Y is a sequence of functions denoted by y:
y = { f 1 , f 2 , f 3 , . . . }, or y = { f1 ( x ), f2 ( x ), f3 ( x ), . . . }, x R,
where we assume that f i : R R are deterministic functions. In vector calcu-

lus, y can be viewed as a function of two variables with domain I R and
range S R, where I = {1, 2, . . . }, and expressed as follows:
y : I R S, defined by (i, x ) f i ( x ) (i.e., y(i, x ) = f i ( x )),
where I is naturally called an index set, and range S =
i=1 Si if we denote the
range of function f i by Si .

Mimicking the familiar notation from vector calculus, we give the following
definition.
Definition 6.8. A stochastic process (or a random process, or simply a process) on

the probability space (, F, P ), denoted by X = { Xt : t J }, is a function of
two variables with domain J and range S R and is expressed by
X :J S,
(t, ) Xt (i.e., X (t, ) = Xt ( ) = Xt ( is dropped to ease notation)),
where J R is nonempty and called the index set1 of the process X, and the
range S is called the state space2 of the process X.
In short, a stochastic process is a collection of random variables on the same

probability space, representing the evolution of randomness over time.
Example 6.8. Let J = {1, 2, . . . }. The process X = { Xt : t J } is called a

stochastic sequence and sometimes written in the form { Xn }
n =1 or
X = { X1 , X2 , X3 , . . . } .
If X is defined by Xi = S0 Y1 Yi , where S0 and Yi are defined in Example 6.7

(see (6.3)), then clearly, the random variable Xi will depend on earlier value
X i 1 .

1 For our purpose, the index set is always a time index set although it is not necessarily so by defini-
tion.
2 In mathematics, the set of objects that we are considering is often referred to as a space.
Example 6.9. Let J = [0, ). We simply write the process X = { Xt : t J } into

X = { X ( t) : t 0}.
Suppose that X (0) = 0, i.e., P ( |X (0, ) = 0) = 1. We say that process
a.s.
X starts at 0 almost surely.

By Definition 6.8, the process X + 7 = { X (t) + 7 : t 0} starts at 7 almost
surely.

Since the index set J usually represents time indeed, and the random vari-
able describes the state of the process at time t, the index t admits a natural
interpretation: if Xt = s, we say that the process is in state s at time t.
Classifications of Stochastic Processes
Stochastic processes can be classified according to the index set J into discrete-
time processes and continuous-time processes.
Example 6.10. { Xt : t J } is a discrete-time process if Xt represents the clos-

ing price of a stock on the t-th trading day, and J = {1, 2, . . . }.

Example 6.11. { Xt : t J } is a continuous-time process if Xt represents the

intraday price of a stock at time t, and J = [9 : 30am, 4 : 00pm] on July 7, 2015.

Stochastic processes can also be classified according to the state space S into
discrete-state processes and continuous-state processes. The state space is discrete if
it consists of a finite number of points or a countably infinite number of points;
otherwise, it is continuous.
Example 6.12. { Xt : t J } is a discrete-state process if Xt represents the total

number of heads in the first t flips of a coin and J = {1, 2, . . . }.

Example 6.13. { Xt : t J } is a continuous-state process if Xt represents the log

return from investing in a stock and is normally distributed with parameters
(t, 2t2 ) and J = {t 0}:

In this chapter, we focus on continuous-time and continuous-state pro-

cesses { Xt : t 0}, although we may use discrete-time processes as ex-
amples to ease unnecessary technicalities.
Sample Paths of Stochastic Processes
Since sample paths of a process may have a direction almost nowhere (noise),
one can visualize them only through imagination; however, an intuition of
the notion of sample paths may come from some idea in analytic geometry.
Example 6.14. Recall from analytic geometry and vector calculus that the graph
of the function of two variables is a surface in R3 . For example, the graph of
y = f (t, x ) = t2 + x2 , (t, x ) R2
is an elliptic paraboloid.
However, either the t-cross section or the x-cross section of y = f (t, x ) is a
continuous path (curve) in R3 . For example, let x = 1. The graph of
y = f (t, 1) = t2 + 1, t R,
which is a function of t alone, is a parabola. In fact, for each fixed x, the graph
of y = f (t, x ) = t2 + x2 , t R, a function of t alone, is a parabola, a continuous
path in R3 .

Let X = { Xt : t 0} be a process on (, F, P ). Observe that for each fixed

t 0, X defines a random variable
Xt : R,
X (t, ) (i.e., Xt ( )),
and for each fixed , X defines a deterministic function of time
X ( ) : J R,
t X (t, ) (i.e., t Xt ( )). (6.4)

The latter part of the observation leads to the next definition.
Definition 6.9. For each fixed , the sample path (or realization, or trajectory,
or sample function) of a stochastic process X = { Xt : t J } on (, F, P ) is the
(graph of) function defined in (6.4) and denoted by either Xt ( ) or X (t, )
(where J may be discrete or continuous).
Note that for a fixed , a continuous sample path of the process { Xt : t 0} is
defined in the ordinary calculus sense. That is,
lim X (s, ) = X (t, ) for each t > 0,

st
lim X (s, ) = X (0, ) for t = 0.
s0, s>0
It is important for the reader to keep the following notational remarks in

mind throughout the rest of this book.
Notational Remark
There may be several notations for the same mathematical concept for good
reasons. Consider a familiar environment as in one variable calculus, where
f ( x ), dx , dx , and D f ( x ) may all represent the derivative of function y = f ( x )
dy d f
with respect to x.
In our discussion, for the sake of convenience or clarity relative to different
contexts, different notation for the same mathematical object may be used. The
reader should keep the following in mind in the rest of this chapter:
1. Notation for sample paths. Xt ( ), X (t, ), X ( ) (where t is dropped to ease
the notation and to emphasize the (relation) rule of function rather than the
value of the function) and X (t) (where is dropped to ease the notation and
to emphasize the fact that a path is defined by a function of t alone) may all
represent a sample path of the stochastic process X.
2. Notation for random variables in a process. Xt and X (t) may be used for a ran-
dom variable in process X.
3. Background probability space. Unless stated otherwise (, F, P ) represents
the background probability space in the rest of our discussions involving
stochastic processes.
Continuous stochastic process
Definition 6.10. A stochastic process X = { Xt : t 0} is said to be sample-

continuous (or almost surely continuous, or simply continuous) if almost surely
all sample paths are continuous. That is,
X ( ) : [0, ) R,
t X (t, )
is a continuous sample path for almost surely (a.s.) all , which means that
P ( |X ( ) is not a continuous sample path) = 0.
The sample continuity is a nice property for a stochastic process to possess as

it implies that sample paths of the process are well behaved in some sense and
therefore easier to analyze than general processes.
Example 6.15. Recall Example 6.13:
X = { Xt : t 0} where Xt N (t, 2t2 ),
is sample-continuous. Explanations will be given in the section on Brownian

motion.

Note that continuity is a convergence property.3
6.1.4 Convergence of Random Variables
We will make use of the following definitions and relationships among differ-
ent notions of convergence of random variables.
Definition 6.11. A sequence { Xn } of random variables is said to converge almost
surely (or converge almost everywhere or converge with probability 1) to a random
a.s.
variable X, written Xn X, if
P ( lim Xn = X ) = 1. (6.5)
n
Definition 6.12. A sequence { Xn } of random variables is said to converge in

P
probability to a random variable X, written Xn X, if for each > 0,
lim P (|Xn X | ) = 0. (6.6)

n
P
Equivalently, Xn X if and only if for each > 0,
lim P (|Xn X | < ) = 1. (6.7)

n
Definition 6.13. A sequence { Xn } of random variables is said to converge in

mean square (or converge in the L2 -norm, or simply converge in L2 ) to a random
variable X, written Xn X, if
m.s.
lim E (|Xn X |2 ) = 0. (6.8)

n
Definition 6.14. A sequence { Xn } of random variables is said to converge in dis-

tribution (or converge in law or converge weakly) to a random variable X, written
d
Xn X, if
lim Fn ( x ) = F ( x ), (6.9)
n
for each x at which F is continuous, where Fn and F are the cumulative distri-
bution functions of random variables Xn and X, respectively.
The next property is often applied. We provide the statement without proofs.
Property 6.1. Let Xn , n = 1, 2, . . . and X be random variables. The following re-
lationships hold:
3 There are different kinds of continuity for stochastic processes although we have only introduced
one (because we will only use one). Other kinds such as continuity in mean, continuity in probability
(or stochastic continuity), and cadlag continuity are also important concepts in the study of mathe-
matical finance. We recommend that the reader who has a serious interest in mathematical finance
begin with studying measure theory.
a.s. P d
1. Xn X = Xn X = Xn X.
P
2. Xn X = Xn X.
m.s.
d
Consequently, Xn X = Xn X.
m.s.
6.1.5 Skewness and Kurtosis
Although both the Markowitz portfolio mean-variance analysis (see Chapter3)

and the Black-Scholes option pricing model (see Chapter 8), which are among
the most important models in quantitative finance, use only the first and sec-
ond central moments of a distribution, the empirical evidence shows that
higher moments are also important in financial modeling.
Given a random variable X, the first moment is a measure of the center of the
distribution; the second moment is a measure of spread. In the following well
see that the third moment measures the amount and direction of the skewness,
and the fourth moment measures the height and sharpness of the central peak.
Thus, the third and fourth moments are measures of shape.
Skewness of Random Variables
Let X be a random variable with expectation and variance 2 . The concept of

skewness of a distribution is defined by a normalized form of the third central
moment of a distribution as follows:
Definition 6.15. The skewness of X, denoted by skew( X ), is defined by

( X )3
skew( X ) = E .
3
X is called positively skewed or skewed right if skew( X ) > 0. X is called negatively

skewed or skewed left if skew( X ) < 0.
The skewness of a random variable is a measure of the symmetry of its dis-
tribution. Clearly, X is symmetric if skew( X ) = 0.
The following intuition provides an easy visualization of the direction of
skewness: let f X be the p.d.f. of X. Suppose that a distribution of a data set X
is unimodal. Skew( X ) > 0 implies that more data are located on the right side
of the center (i.e., E [X ], see Figure 6.4 on page 320 and the rightmost graph in
Figure 8.5 on page 443). Since the area under the graph of f X is 1, f X peaks on
the left side of the center. This is to say that a skewed right distribution has the
right tail longer than the left tail as the p.d.f. is tilted to the left.
Similarly, one can see that a skewed left distribution has the left tail longer
than the right tail, and the p.d.f. is tilted to the right (see the leftmost graph in
Figure 8.5 on page 443).
Kurtosis of Random Variables
The concept of kurtosis is a measure of the degree of peakedness of a distri-

bution and is defined by a normalized form of the fourth central moment of a
distribution as follows:
Definition 6.16. The kurtosis of X, denoted by kurt( X ), is defined by

( X )4
kurt( X ) = E .
4
The following intuition provides an easy visualization of a distribution with
large kurtosis: again, suppose that a distribution of a data set X is unimodal.
Let a be a real number. Note that a4 is much larger than |a| if |a| > 1, and much
X
smaller than |a| if |a| < 1, and that normalizes X. Thus, a large kurt( X )
means that |X | can take large values with relatively high probability. This is to
say that the peak of f X is tall and sharp as the area under the graph of f X is 1
(see Figure 8.7 on page 444).
Remark 6.3. Although both first and second moments of a random variable X
have units, skew( X ) and kurt( X ) do not.

Example 6.16. If X is a normal random variable with parameters and 2 , then
skew( X ) = 0 and kurt( X ) = 3. The detailed computation is left as an exercise
for the reader.

The following terminologies appear frequently in the literature of financial as-
set modeling:
Definition 6.17. The kurtosis excess of a random variable X is defined by
KurtExcess( X ) = kurt( X ) 3.
Definition 6.18. A random variable X is called leptokurtic if kurt( X ) > 3.

Equivalently, a random variable X is leptokurtic if KurtExcess( X ) > 0.
Note that relative to normal distributions, a unimodal distribution with kur-
tosis larger than 3 has a taller and sharper peak and longer and fatter tails.
Financial data are often collected from observations of the same variable
(e.g., a stock price) in a form of discrete-time series { Xt : t = 0, 1, 2, . . . n}.
Although these data are always discretely sampled, we often establish
continuous-time series to model them as mathematical tools can be more easily
applied.
6.2 Filtrations and Adapted Processes
6.2.1 Filtrations: The Evolution of Information
Stochastic processes are often viewed as useful tools to model randomness

evolving over a period of time. For example, we let X = { Xt : t 0} represent
a process of a stock price at a future time. In normal circumstances, the market
price of a stock is the price at which a buyer and a seller agree to trade (with-
out any official arbiter of stock prices). Thus, the change of the market price
of a stock is caused by the supply and demand, which reflect expectations of
the companys profitability. To figure out such expectations, one has to figure
out how any news about the company will be interpreted by traders and in-
vestors. Therefore, we need a time-evolving information structure to study a
random process of a stock price. Another part of the motivation of establish-
ing such a structure comes from the mathematical object itself: if the incre-
ment Xt+h Xt , where h > 0, is assumed to be independent of t (to make a
full description4 of the process X mathematically manageable), then what we
really say is that Xt+h Xt is independent of information up to time t. This
observation suggests we describe the evolution of information as information
propagation over time. The mathematical concept that serves this purpose is
called filtrations, which give a rigorous definition for the past at a given time
and captures the desired intuition in the above discussion.
Definition 6.19. A filtered probability space is a quadruple (, F, {Ft }, P ), where

(, F, P ) is a probability space and {Ft } is a filtration. A filtration is a nonde-
creasing collection of -algebras {Ft F : t 0} such that
Fs Ft , for s, t 0 with s t. (6.10)
One intuitive way to describe a filtration is that it functions like a filter of in-
formation flow to control information propagation. For our purpose, it is suf-
ficient to know the following:
1. Ft represents the information available at time t.
2. The information structure designated by (6.10) assures that the amount of
information grows as time evolves and that no information is lost with in-
creasing time (e.g., no computer crashes) in the sense that whatever infor-
mation available at time s is still available at time t as long as t s.
In short, we translate the mathematical language the -algebra Ft to the

ordinary English as information set Ft under the structure (6.10).
4 A legitimate full description of the stochastic process is based on the notion of finite-dimensional
distributions.
6.2 Filtrations and Adapted Processes 269
Note that filtrations can also be defined on a discrete-time index set by the
same idea. An example of such filtrations is provided below.
Example 6.17. Recall Example 6.4, where we considered an oversimplified

stock price behavior described by a binomial tree with two periods [t0 , t1 ] and
[t1 , t2 ] such that in each period, the stock price either goes up by a factor u with
probability p or goes down by a factor d with probability 1 p. The corre-
sponding probability space is represented by (, F, P ), where
= {1 , 2 , 3 , 4 }, where
1 = UU, 2 = UD, 3 = DU, 4 = DD,
F = {, {1 }, {2 }, {3 }, {4 },
{ 1 , 2 } , { 1 , 3 } , { 1 , 4 } , { 2 , 3 } , { 2 , 4 } , { 3 , 4 } ,
{ 1 , 2 , 3 } , { 1 , 2 , 4 } , { 1 , 3 , 4 } , { 2 , 3 , 4 } , } .
Now, if we let
Ft0 = {, },
Ft1 = {, {1 , 2 }, {3 , 4 }, },
Ft2 = F,
then {Ft : t = t0 , t1 , t2 } is a filtration. Note that Ft0 Ft1 Ft2 F indeed.

Also, note that as t increases, information set Ft becomes finer and reveals
more information about the evolution of stock price in the following sense:
S ( t0 ) S ( t1 ) S(t2 ), outcomes
S0 u 2 , 1
S0 u, {1 , 2 }
S0 ud, 2
S0
S0 du, 3
S0 d, {3 , 4 }
S0 dd, 4
at t = t0 we have no information available about outcomes,

at t = t1 we know whether we will have {1 , 2 } or {3 , 4 },
at t = t2 we know which i we have.
We say that Ft2 is finer than Ft1 , and Ft1 is finer than Ft0 , or equivalently, Ft0
is coarser than Ft1 , and Ft1 is coarser than Ft2 .

Definition 6.20. The filtration {Ft F : t 0} is called a natural (or standard)

filtration of process X if
F t = ( Xs , 0 s t ), t 0, (6.11)
where ( Xs , 0 s t) is called the -algebra generated by random variables

Xs , 0 s t.
We also say that the filtration {Ft : t 0} defined by Ft = ( Xs : s t) is the

filtration induced by { Xt : t 0}. Note that
( Xu , 0 u s ) ( Xu , 0 u t ), 0 s t.
Example 6.18. The -algebra generated by two random variables X1 and X2 , writ-
ten ( X1 , X2 ), is the smallest -algebra which is sufficient for both random
variables X1 and X2 to be measurable (see page 257 and Example 6.6).
In the language of information, this is tantamount to saying that we use the
least amount of information to determine both X1 and X2 . For this reason, we
say that:
( X1 , X2 ) represents the information set that contains the least amount
of information to determine both X1 and X2 .
Clearly, ( X1 ) ( X1 , X2 ) and we see that given a discrete-time stochastic
process { Xt , t = 1, 2, 3 . . . }, we have ( X1 ) ( X1 , X2 ) ( X1 , X2 , X3 ) .

6.2.2 Conditional Expectations: Properties and Intuition
The significance of conditional expectation with respect to a -algebra will be

discussed at an intuitive level while suppressing the technicalities.
The notion of information set described in Section 6.2.1 is used in the concept
of conditional expectation of random variable X on (, F, P ) with respect to a
-algebra (an information set) F F, denoted by E ( X |F ).
The conditional expectation E ( X |F ) is itself a random variable.5
We will make use of the following properties of conditional expectation:
Property 6.2. (Measurability) If X is F -measurable, then
E ( X |F ) = X.
a.s.
(6.12)
5 A rigorous definition of conditional expectation can be found in standard graduate-level textbook

of probability theory.
Intuition and interpretation. As we explained in Example 6.6 and on page 257,

X being F -measurable simply means that F contains sufficient information of
X (sufficient to determine X). That is to say that X is known given F . Since the
best guess (or estimate or predication) of the known is itself, (6.12) holds. In
mathematical language the best guess (or estimate or predication) of a random
variable Y is expressed by E (Y ) (in the context of our discussion).
Example 6.19. Given two -algebras F and G with F G and a F -measurable

random variable X, compute E ( X |G).
Solution. Since X being F -measurable implies that X is G -measurable for

F G (stated in common language: if the smaller information set contains
sufficient information of X, so does the larger one), it follows by the measura-
bility property of conditional expectation that E ( X |G) = X.
a.s.

Example 6.20. Given two -algebras F and G with F G and a G -measurable
random variable X, what can you say about E ( X |F ) or the prediction of X?
Solution. Unknown (because a smaller information set may not contain suffi-
cient information of X even the larger one does).

Example 6.21. Since ( X ) contains sufficient information of X, (6.12) implies
E ( X |( X )) = X.
a.s.
Similarly,
E ( X1 |( X1 , X2 )) = X1 , and E ( X2 |( X1 , X2 )) = X2 .
a.s. a.s.
More generally, let X = { Xt , t 0} be a stochastic process, then the natural

filtration of X is defined by Ft = ( Xs , 0 s t), t 0. We have
E ( Xs |Ft ) = Xs , for each s [0, t].

a.s.
But E ( Xu |Ft ) is unknown for each u > t.

Property 6.3. (Taking Out What Is Known) If X and Y are random variables
and X is F -measurable, then
E ( XY |F ) = XE (Y |F ).
a.s.
(6.13)
Intuition and interpretation. Again, X being F -measurable means that X is

known given F . That is to say that X is being treated as a constant given infor-
mation set F . It follows that (6.13) holds.
Example 6.22. We continue from Example 6.21 and have
E ( Xt Xt+0.2 |Ft ) = Xt E ( Xt+0.2 |Ft ).

a.s.

Property 6.4. (Computing Expectations by Conditioning) If X is a random

variable and F is a -algebra, then
E (E ( X |F )) = E ( X ). (6.14)
Property 6.5. (Tower Property) If F G , i.e., F is a sub--algebra of G , then
E (E ( X |G)|F ) = E ( X |F ).
a.s.
(6.15)
Intuition and interpretation. Calculation of conditional expectation on an in-

formation set may be taken by two steps. First, calculate conditional expecta-
tion on a less coarse (or more detailed) information set than the original one.
Then, finish the calculation by conditioning on the original information set,
which is coarser (or in less detail).
Property 6.6. (Linearity) Let ai be constants and Xi be random variables, then
E ( a1 X1 + a2 X2 |F ) = a1 E ( X1 |F ) + a2 E ( X2 |F ).
a.s.
(6.16)
More intuition of E ( X |F ).
Sometimes random variable X and a -algebra F are both given, but X is not
measurable on F . Intuitively, we say that F does not have enough (contrast)
resolution to read off all information contained in X. In this sense, X being
F -measurable is at one end of a yardstick, where F has highest resolution to
read off all information contained in X, whereas X being independent of F
(see Definition 6.7 on page 260) is at the other end of the yardstick, where F
has no resolution (thus read off no information of X). In other words, we say
that X being independent of F means that F contains no information of X.
The meaning of E ( X |F ) is the least coarsification of X so that F has enough
resolution for E ( X |F ) (i.e., E ( X |F ) is measurable on F ).
Now, we are ready for the next property.
Property 6.7. (Independence) If X is independent of F , then
E ( X |F ) = E ( X ). (6.17)
Finally, to avoid notational confusion, we have the following definition:

Definition 6.21. Let X and Y be two random variables with finite expectations.
we define the conditional expectation of X given Y by
E ( X |Y ) = E ( X |(Y )). (6.18)
Interpretation. In a special case of E ( X |F ), when F = (Y ), we have a more

concise notation: E ( X |Y ). Thus, E ( X |Y ) = E ( X |Y = y) since (6.18) says that
the random variable E ( X |Y ) depends only on the events that Y defines (e.g.,
{0.2 < sin(Y 2 + eY ) 0.7}) instead of just on specific value of Y.
6.2.3 Adapted Processes: Definition and Intuition
Definition 6.22. A stochastic process { Xt : t 0} defined on a filtered prob-

ability space (, F , {Ft }, P ) is said to be adapted (or non-anticipating) if Xt is
Ft -measurable for each t 0.
Intuition and interpretation.

1. On the one hand, by this definition, Property 6.2 and Example 6.19,
Xs = E ( Xs |Ft ), for each t s,

a.s.
particularly,
Xs = E ( Xs |Fs ).
a.s.
(6.19)
Using the language in Remark 6.2 on page 259, we say that Xs is considered
known at any time t, t s if { Xt : t 0} is adapted to the filtration {Ft }.
In other words:
A stochastic process { Xt : t 0} being adapted to the filtration implies
that the value of Xt is (almost surely) completely determined by the fil-
tration Ft in the sense of (6.19).
2. On the other hand, since for each t > s, Xt may not be Fs -measurable, at
time s (see Example 6.20 or 6.21 on page 271), Xt is considered unknown
because probabilities of events described by Xt may not be computed based
on the information available at any time earlier than the moment t. In this
sense:
The notion of adaptedness can be interpreted as inability to have knowl-
edge about future events. For this reason an adapted process is also called
non-anticipating because the propagation or progressive revelation of in-
formation under adaptedness allows no anticipation of future informa-
tion. An illustration of this concept is provided in the next example.
3. By definition of natural filtrations (see (6.11) on page 270):

A stochastic process is always adapted to its natural filtration.
Example 6.23. Let us revisit Example 6.17 (see page 269) where we have
= { 1 , 2 , 3 , 4 } ,
and {Ft F : t = t0 , t1 , t2 } with t0 < t1 < t2 , a filtration defined by

{, }
if t = t0
Ft = ({1 , 2 }) if t = t1

F if t = t2 ,
where
({1 , 2 }) = {, {1 , 2 }, {3 , 4 }, }.
If we define
Xt0 ( i ) = 1 if i = 1, 2, 3, 4,
.
2 if i = 1, 2
Xt1 ( i ) =
3 if i = 3, 4,

4 if i = 1

5 if i = 2
Xt2 ( i ) =

6 if i = 3

7 if i = 4,
then it can be easily verified that Xt is Ft -measurable where t = t0 , t1 , t2 respec-

tively. Thus X = { Xt : t = t0 , t1 , t2 } is adapted.
Observe that events represented by { Xt1 a}, a < 2 or a 3 (which are
either or ) are information that was propagated or revealed at time t = t0 ,
and none of them is {1 , 2 } or {3 , 4 }, both of which are information
revealed at time t = t1 . In a similar fashion, the reader may observe that events
represented by { Xt1 a}, 2 a < 3 (which are {1 , 2 }) are information that
was revealed at time t = t1 , and none of them is {i }, i = 1, 2, 3, 4, which are
information revealed at time t = t2 .
In this sense, we say that the propagation or progressive revelation of infor-
mation under adaptedness ensures no anticipation of future information:
It is also worth noting that the concept of information in financial theory
is not the same as that in the ordinary sense. Under the circumstances sur-
rounding our discussion, by propagation of information, we mean the pro-
gressive revelation of the events represented in the form of filtrations.
In a similar context, by observable realities (by time t), we mean those
events that are observable (precisely, Ft -measurable or in Ft ) by time t
6.3 Martingales: A Brief Introduction 275
(e.g., events represented by { Xt a} assuming that { Xt } is adapted to the

filtration).6
Example 6.24. In financial theory, the mathematical concept of a filtered prob-
ability space (, F, {Ft }, P ) can be used to represent an economy, where each
simple event represents an economic state, and the mathematical con-
cept of adapted stochastic processes { Xt : t 0} to the filtration can be used
to represent the evolution of a security price and to ensure that events such as
{ Xt a} = { : Xt ( ) a} are not anticipated at any time s < t. Also, note
that the concept of filtration suggests that the economic states are not
instantaneous states but represent entire possible histories of the economy.
Remark 6.4. The concept of an adapted stochastic process is essential in study-

ing modern finance theory. For instance, it is essential in the definition of mar-
tingales.

6.3 Martingales: A Brief Introduction
6.3.1 Basic Concepts
Note that in our next definition, t may be an integer for a discrete-time process
or a real number for a continuous-time process.
Definition 6.23. A stochastic process { Xt : t 0} on a filtered probability space
(, F, {Ft }, P ) is said to be a martingale with respect to the filtration {Ft } if { Xt }
is adapted to {Ft } and satisfies the condition E (|Xt |) < for t 0 and the
property
E ( Xt |Fs ) = Xs for s < t, t 0. (6.20)
Stating the defining characteristic of the martingale property given by (6.20) in
common language:
The best prediction for a future realization is the current value of the pro-
cess.
Taking expectation on both sides of (6.20) yields
E ( Xt ) = E ( Xs ),
which is equivalent to E ( Xt Xs ) = 0. That is, the expected future gain (or

loss) is zero. In this sense, martingales model a fair game.7 For this reason the
6We also say that F t consists of all events that are observable (i.e., F t -measurable) by time t.
7For this reason, a risk-neutral probability measure is referred to as an equivalent martingale measure,
which is a key concept in derivative pricing (see Remark 7.3, 2 on page 336).
importance of martingales in modern finance is self-explanatory since security

valuation is the determination of the fair price of a security.
Intuition. If Xt represents the log return from investing in a stock at time t, and
{Ft F : t 0} is a natural filtration of { Xt : t 0}, where Ft = ( Xs : s t)
for every t 0 is the filtration induced by { Xt } (see (6.11) on page 270), expres-
sion (6.20) implies that different styles of investing will not produce different
expected returns regardless of the amount of financial research that investors
have access to. In other words, active management or passive management,
buy-and-hold investing, or market-timing trading based on technical analysis
will all have the same expected returns.
Remark 6.5. Since all Ito integrals with respect to Brownian motion are mar-
tingales, martingales form an important class of stochastic processes.

Martingales are defined for discrete-time stochastic processes by Defini-

tion 6.23. However, for the special case when t takes nonnegative integer val-
ues, we have a more intuitive statement:
Definition 6.24. A stochastic sequence { Xn : n = 0, 1, 2, . . . } on a filtered prob-

ability space (, F, {Fn }, P ), where Fn = ( Xk , 0 k n), is said to be a
martingale with respect to its natural filtration if { Xn } satisfies the condition
E (|Xn |) < for all n 0 and the property
E ( Xn+1 |Fn ) = Xn for all n 0. (6.21)
Using a more concise notation of conditional expectation (in the spirit of Defi-
nition 6.21), (6.21) can be written into
E ( X n + 1 | X0 , X1 , . . . , X n ) = X n for all n 0. (6.22)
Expression (6.22) in ordinary English says that:

Knowing the past history of the process, the best prediction for one step
ahead is the current observation.
Example 6.25. (Random Walk) Let { Rn : n 1} be a sequence of independent

identically distributed random variables with mean = 0 and finite variance
2 . Let
Sn = R1 + R2 + + R n . (6.23)
The sequence {Sn : n 1} is a random walk. For example, if
.
1 with p = 12
R1 = and set S0 = 0,
1 with p = 12
6.3 Martingales: A Brief Introduction 277
the stochastic sequence {Sn : n 0} is called the simple random walk on Z.

Clearly, the walk starts at 0 and at each step moves either +1 (to the right)
or 1 (to the left) with equal probability.
We are interested in showing that {Sn : n 1} defined by (6.23) is a martin-
gale.
Proof. We verify two conditions required by Definition 6.24:

Step 1.
E (Sn ) = nE ( R1 ) = 0.
n n n
E (S2n ) = E ( Ri R j ) = 2 = n2
j =1 i =1 j =1
implies8
E (|Sn |) < .
Step 2.
E ( S n + 1 | S1 , S2 , . . . , S n ) = E ( R 1 + R 2 + + R n + R n + 1 | S1 , S2 , . . . , S n )
= R 1 + R 2 + + R n + E ( R n +1 ) = S n + 0 = S n .

6.3.2 Martingale as a Necessary Condition of an Efficient Market
According to Fama (1970), a market in which prices always fully reflect avail-
able information is called efficient (see Fama [11]).
Recall (6.20)
E ( Xt |Fs ) = Xs for s < t.
Let s be the current time and Fs = ( Xu , u s). If Xt represents a security
price at time t, (6.20) indicates that the information contained in the past prices
is instantly and fully reflected in the security current price. For this reason,
the martingale is considered to be a necessary condition for efficient security
market.
8 A proof can be done either by applying Cauchy-Schwarz inequality or by using the result at the link
http://mathworld.wolfram.com/RandomWalk1-Dimensional.html to show that

#
E (|Sn |) 2
lim = .
n n
Since the efficient market hypothesis (EMH) is viewed as a fundamental as-

sumption in theoretical finance, showing that a security price process is a mar-
tingale becomes highly important as otherwise establishing any model based
on the study of the process behavior could be in a framework that violates the
EMH. A further in-depth discussion on this subject will lead to the fundamen-
tal theorem of finance.
6.4 Modeling Security Price Behavior
6.4.1 From Deterministic Model to Stochastic Model
We will illustrate deterministic versus stochastic models by an example.

Deterministic model
The continuously compounding interest model can be described by an ordi-
nary differential equation (o.d.e.)
dS
= rS with S ( 0 ) = S0 ,
dt
dS = rS dt with S ( 0 ) = S0 , (6.24)
where S = S(t) is the amount at time t and r is a constant interest rate.

S(t)
Letting X (t) = ln S and substituting in (6.24), we obtain a simplified
0
model:
dX = rdt with X (0) = 0, (6.25)
where X = X (t) is the log return over the time period [0, t].
Stochastic model
We consider the scenario of having a perturbation to the constant interest rate.
In other words, we decompose r into a sum of a nominal value and its per-
turbation (which is unpredictable and causes r to change infinitely fast):
r=+ or rt = + t , (6.26)
where the subscript t is to emphasize the change with respect to time.

Corresponding to the perturbation term, the (deterministic) o.d.e. in (6.25)
becomes
dX = dt + noise,
where the noise term is often considered as a Gaussian white noise stochastic
process (see Definition 6.25 on page 280), which is related to Brownian motion
6.4 Modeling Security Price Behavior 279
(see Section 6.5) since it has been shown that, although Brownian motion paths
are nowhere differentiable, their (formal) time derivatives form the white
noise process (see Definition 6.25). Thus, we can write
dXt = dt + dBt , (6.27)
where is a positive constant and B = {Bt } is the standard Brownian motion

which will be introduced shortly.9
Notice that X = X (t) in (6.25) is a deterministic function, whereas X = Xt in
(6.27) becomes a stochastic process.
Using a mathematical jargon, we say that the process X in (6.27) is driven by
the standard Brownian motion B.
6.4.2 Innovation Processes: An Intuition
An innovation (process) is obtained by taking the difference between the ob-

served value of a random variable at time t (e.g., xt where t is in discrete-time)
and the optimal forecast of that value based on information available prior to
time t (e.g., E ( xt |Ft1 )).
In other words, the quantity modeled by a process can be decomposed into
two components; one is predictable and another is unpredictable (e.g., (6.26)).
The unpredictable component is called an innovation (process). If a stochastic model
successfully captures the predictable component structure in the data, then
the innovation process constitutes a white noise process (see Definition 6.25
below) or an martingale difference sequence (not to be discussed in this book).
Before the next example, we need the definition of autocovariance of a pro-
cess.
Let X = { Xt : t J } be a stochastic process with discrete or continuous
time. The autocovariance (function) of process X is defined by
(t, s) = Cov( Xt , Xs ) for s, t J , (6.28)
which is a function of two variables and satisfies
(s, t) = (t, s).
For each t, we let be a time lag such that t + J and denote by (t) the
autocovariance of X (where (t, s) = (t, t + ) with = s t). That is,
(t) = (t, t + ) = Cov( Xt , Xt+ ), t, t + J .
9 White noise thought of as the derivative of Brownian motion, dB t

dt , does not exist in the ordinary
dB
sense. It is related to the notion of generalized stochastic process, since dt is well defined as a generalized
function on an infinite dimensional space, which is a topic beyond the scope of this book.
Note that a process has no linear forecasting value if its autocovariance func-
tion (t) is identically equal to zero.
Since a noise is presumably unpredictable, we expect the covariance func-
tion of a white noise process at all nontrivial time lags to show a value of zero.
We define the mean value function of a process { Xt } by m(t) = E ( Xt ) and
give the following definitions.
Definition 6.25.
1. The process { t } is said to be a white noise (process) if its mean value function
and autocovariance function respectively are
.
2 if = 0
m(t) = E ( t ) = 0, and (t) = for t,
0 if = 0,
where 2 > 0 is a constant. We write
t WN (0, 2 ).
2. An independent white noise (process) { t } is a white noise process consisting

of mutually independent random variables. We write
t i.WN (0, 2 ).
3. A strict white noise (process) { t } is a white noise process consisting of inde-

pendent and identically distributed (i.i.d.) random variables. We denote it
by
t i.i.d.WN (0, 2 ).
4. A white noise process { t } is said to be a Gaussian white noise process if
t N (0, 2 ).
A white noise process has no linear prediction value (it is unpredictable)

because it is serially uncorrelated.
Example 6.26. { i } in (4.23) on page 186, is a white noise.

Example 6.27. A stochastic process {Wi } defined by

i
Wi = W0 + Rk = Wi1 + Ri , i = 1, 2, . . .
k =1
is said to be a random walk on a probability space (, F, P ) if { Ri , i = 0, 1, 2, . . . }

is a sequence of i.i.d. random variables and W0 a random variable that is inde-
pendent of each Ri :
6.4 Modeling Security Price Behavior 281
1. A random walk is not a white noise. (Hint: compute covariance.)

2. A random walk is a martingale with respect to its natural filtration if
E ( Ri ) = 0 for all i (see Example 6.25 on page 276).
The proof is left as an exercise for the reader.

6.4.3 Securities Paying a Continuous Cash Dividend
Let us first define the dividend yield with continuously compounding. To do

so intuitively, recall that the (annualized constant) interest rate r under contin-
uously compounding is a constant satisfying the equation
I (t) = A(t) r dt,
where I (t) is the simple interest on the time interval [t, t + dt], and A(t) is the
initial principal on the same time period. Note that the infinitesimal change on
the principal is
d A(t) = I (t) = A(t) r dt.
Similarly, if S(t) is the price of the underlier, without loss of generality, e.g.,
a stock at time t, and D(t) is the dividend paid by the stock over the time
interval [t, t + dt], we define the (annualized) dividend yield, denoted by q, to
be the constant satisfying
D(t) = S(t) q dt. (6.29)
If we reinvest the dividends immediately in the stock, (6.29) says that the divi-
dend paid on [t, t + dt] buys us qdt shares of the stock. It follows that over the
time period [t, t + dt], if we own N (t) shares of the stock initially (i.e., at time
t), the dividends from these N (t) shares of the stock on the small interval buy
us N (t) qdt more shares than that we initially owned. Restated in mathematical
language,
d N (t) = q N (t) dt, t [0, T ]. (6.30)
In words, the infinitesimal change in the shares held on [t, t + dt] is q N (t) dt.
Solving the initial value problem of o.d.e.
d N (t) = q N (t) dt, N (0) = N0 , t [0, T ], (6.31)
we obtain N ( T ) = N0 eqT . In particular,
N0 = 1 N ( T ) = eqT , and N0 = eqT N ( T ) = 1. (6.32)
In words:
The dividend reinvestment yields eqT 1 more shares at time T from one share of
the stock at time 0;
The dividend reinvestment yields 1 eqT more shares at time T from eqT shares
of the stock at time 0.
This result will be used again and again, particularly in Chapter 7.
Example 6.28. There are two ways to profit from a dividend-paying security:
capital gains and dividends. Let {S(t)} be the price process of such a secu-
rity with dividend yield q. Assuming that the risk-neutrality hypothesis holds,
the expected return from the security is the risk-free interest rate r. There-
fore, the expected return of the capital gains (i.e., based solely on the appreci-
ation of the stock price) must be r q. This is to say that if S(t0 ) is the security
price at the beginning of time interval [t0 , t0 + t], then the expected value of
the security price at the end of the interval becomes S(t0 )e(r q)t . If {S(t)} is
modeled by a binomial tree with parameters p, u, and d, then the expectation
of capital gains over this time period can be expressed by
S(t0 )(e(rq)t 1) = S(t0 )((p u + (1 p) d) 1).

6.5 Brownian Motion
6.5.1 Definition of Brownian Motion
The following familiar properties of normal random variables and notation

will be frequently used:
Property 6.8. (Normal Distributions) We denote a normally distributed ran-

dom variable X with mean and variance 2 by X N (, 2 ). Let a and b be
constants. The following properties hold:
1. X N (0, 1) a + bX N ( a, b2 ).
2. X N (, 2 ) a + bX N ( a + b, b2 2 ).
3. X N (, 2 ) MX (t) = et+ 2 t ,
1 2 2
where MX (t) E (etX ) is the moment generating function of X.

6.5 Brownian Motion 283
Definition 6.26. A standard (one-dimensional) Brownian motion10 or Wiener pro-

cess,11 denoted by B = {B(t) : t 0}, is a stochastic process on (, F, P )12
satisfying the following properties:
1. B(0) = 0 (i.e., P ( |B(0) = 0) = 0).
a.s.
2. With probability 1, sample paths of B are continuous, i.e.,
P ( | B(t, ) is not continuous at each t) = 0.
3. For every choice of nonnegative real numbers
0 t1 < t2 < t3 < tn1 < tn < ,
the increments
B(t2 ) B(t1 ), B(t3 ) B(t2 ), . . . , B(tn ) B(tn1 )
are mutually independent random variables.

4. For each 0 s < t < , the increment B(t) B(s) is a normal random vari-
able with mean 0 and variance t s, written B(t) B(s) N (0, t s).
It is easier to remember the properties in Definition 6.26 if we can concisely

describe them in words:
1. B starts at 0 almost surely (i.e., the initial state of the process is 0 a.s.).
2. B is a sample-continuous process (see Definition 6.10).
3. B has independent increments.
4. B has stationary increments (i.e., time homogeneity), which are normally dis-
tributed.
The reason for the names (i.e., stationary increments or time homogeneity)
is due to the fact that Definition 6.26, 4 implies that for all s, t, h 0,
d
B(t) B(s) = B(t + h) B(s + h).
That is, the increments over the time intervals with same length have the
same probability distributions.
10 Robert Brown (1773 - 1858).

11 Nobert Wiener (1894 - 1964). Wiener process is a more popular name among mathematicians than
among physicists whereas Brownian motion is vice versa.
12 B = {B( t) : t 0} on ( , F, P ) is often referred to as a P-Brownian motion where P represents
the probability measure in the real world (or physical world) in contrast to Q-Brownian motion. Q-
Brownian motion means that B = {B( t) : t 0} is a process on ( , F, Q ) where Q represents the
probability measure in the risk-neutral world. More detailed explanation is given in Section 6.8.3.
Definition 6.27. A Brownian motion with starting point b is a stochastic process

that can be expressed by b + B where b R is a constant and B is a standard
Brownian motion.
In words, a Brownian motion X = { X (t)} with initial value X (0) = b is ob-

a.s.
tained by adding b to a standard Brownian motion B = {B(t)}, i.e., X = B + b

or equivalently X (t) = b + B(t), t 0.
Definition 6.28. A Brownian motion with drift and scaling is a stochastic process
that can be expressed by b + t + B where b, R and > 0 are constant
and B is a standard Brownian motion.
Remark 6.6. Although Section 6.10 provides a brief discussion on a Brownian

motion as a limit of a random walk, it is worth noting that the existence of
Brownian motion is not a trivial fact.13 For a constructive proof of this non-
trivial fact, we refer the reader to the literature (e.g., Durrett [9]; Morters, and
Peres [22]; and Paley, Wiener, and Zygmund [25]).
6.5.2 Some Properties of Brownian Motion Paths
For the sake of convenience of conversation, we refer to a sample path of a

Brownian motion as a Brownian path.
Three of the most important Brownian path properties are given below,
which will help us to visualize Brownian motion.
Property 6.9.
1. With probability 1, Brownian paths are continuous.
2. With probability 1, Brownian paths are nowhere differentiable.
3. With probability 1, Brownian paths do not have bounded total variation14
on [0, t] (nor on any interval by the time-homogenous property of Brownian
motion).
Proof. The first property follows from the fact that a Brownian motion B + b is
sample-continuous if and only if the corresponding standard Brownian motion
B is sample-continuous (and this is straightforward from Definition 6.26).
13 It is not obvious that all four properties in Definition 6.26 are compatible with each other. For in-
stance, it is not obvious that stationary independent increments and sample continuity are compatible
properties.
14 The total variation is a way to measure the variation of a (deterministic) real-valued function (see
Section (6.6)).
Since a proof of the second property requires deeper understanding of anal-

ysis that is beyond the scope of this book, we refer the reader to the literature
(e.g., Billingsley [3]).
For a proof of the third property, see page 295.

Scaling properties of Brownian motion are invariant properties under trans-
formations.
Property 6.10. (Scaling Invariance) For a > 0, process { X (t) : t 0} with

1
X (t) = B( a2 t)
a
is a standard Brownian motion.
Proof. It is sufficient to show that increments are normally distributed. In fact,
1 1 1
X ( t) X ( s ) = B( a2 t) B( a2 s) = (B( a2 t) B( a2 s))
a a a
1 d a
N (0, a2 (t s)) = N (0, t s) = N (0, t s),
a a
where the third equal sign holds due to Property 6.8, 2.

Intuition. Scaling invariance simply means that the geometric structure of

Brownian paths has a fractal15 nature or self-similarity property. In loose terms,
if we zoom in or zoom out on a Brownian path, we always see a Brownian
path.
Example 6.29. Show that if B is a standard Brownian motion, so is B.
What is your visualization of B? Is it different from that of B?
Straightforward verifications of continuity of sample paths, independence,
normal distribution, and proper means and variances of all the increments are
left as an exercise to the reader.
Hint: Notice that F X ( x ) = P ( X x ) = 1 P ( X < x ) = 1 FX ( x ) implies
that f X ( x ) = f X ( x ) and that f X ( x ) = f X ( x ) if f X is an even function (e.g.,
X N (0, 2 )).

6.5.3 Visualization of Brownian Paths
Although any graphs can only depict a Brownian motion traveling in a manner
far from desirable due to a host of microscopic random effects, a mental visu-
alization of them may be achieved. The following explanation may be helpful.
15In short, a fractal is a natural phenomenon or a mathematical set that exhibits a repeating pattern
that displays at every scale. For more explanation, we refer the reader to http://en.wikipedia.
org/wiki/Fractal.
Three Keys to Reading Brownian Motion Paths

Mimicking the notion of graphs of cross-sections of a function, say
x = f (t, w)
in analytic geometry, we try to visualize a stochastic process
X = { X (t, )}
in terms of its sample paths (cross-sections when is fixed) and the p.d.f. of
random variables (cross-sections when t is fixed):
1. Visualization of Brownian paths (cross-sections when is fixed)
Each Brownian path is continuous but at almost nowhere directional,

i.e., makes sharp turns (or forms sharp corners) everywhere due to
Property 6.9, 1 and 2. In fact, it can be shown that with probability 1, a
Brownian path is not monotonic on any interval. This is a manifestation of
Property 6.9, 1 and 2 combined.
To elaborate this, if a path is drawn by a physical pen, then at most points
the path has directions as driven by the force exerted on the pen. This ex-
plains why a Brownian path cannot be drawn by a pen and therefore quite
nonintuitive.
Each Brownian path has an infinitely large variation over any time interval
due to Property 6.9, 3.
To elaborate this, we might imagine a piece of Brownian path over a time
interval merely the diameter of a thread of hair yet with infinite length.
All sample paths of the same Brownian motion diffuse (or radiate) from the
(same) initial state (at t = 0) due to Definition 6.26, 1.
If you zoom in or zoom out on a Brownian path, you always see a Brownian
path.
2. Visualization of distributions of random variables (cross-sections when t is fixed)
The graph of probability density function of each random variable B(t) is

a bell-shaped curve due to Definition 6.26, 4. That is, the distributions of
the dots, which are t-cross-sections of Brownian paths, are governed by
bell-shaped curves.
In other words, since B(t) N (0, t) , for each fixed t, we know that
16
68% of Brownian paths are within t units from the time-axis;

95% of Brownian paths are within 2 t units from the time-axis;
99.7% of Brownian paths are within 3 t units from the time-axis.
16

Note that 1 standard deviation from 0 is t.
3. Combining sample paths and graphs of p.d.f. of (state) random variables

Just as we can achieve 3-D visualization of the graph of z = f ( x, y) by men-
tally combining the graphs of cross-sections in analytic geometry, we can ob-
tain some 2-D intuition of a Brownian motion by mentally combining its sam-
ple paths and those bell-shaped curves (see Figure 6.1).
Fig. 6.1 Standard Brownian motion is shown using 5,000 randomly selected sample paths over the
interval [0, 1], where X ( t) = B( t) is plotted on the vertical axis. The current time is 0 and for each
future t > 0, the random variable B( t) is normal with mean 0 and variance t, which can be seen to
increase with time in the figure. At time t = 1, the frequency distribution of the sample paths is shown
as a histogram, which indeed has the shape of a normal distribution with mean 0 and variance 1. The
horizontal line shows the mean value of standard Brownian motion, namely, E (B( t)) = 0 for all t 0
Example 6.30. Because of a host of microscopic random effects (e.g., see scaling
invariance Property 6.10), graphs can depict a Brownian motion traveling only
in a manner far from desirable; however, to visualize the Brownian motion B +
b, one may vertically translate the graph in Figure 6.1 by b units, and imagine
that Brownian paths are diffusing from its initial state B(0) = b, and travel in
a.s.
unpredictable directions all the time (therefore impossible to draw).

6.5.4 Markov Property for Brownian Motion
One may view the Markov property as a somewhat extended independent

increment requirement in the definition of Brownian motion.
Observation.
Let {Ft } be the natural filtration of the standard Brownian motion {Bt }. That
is, Ft = (Bs : 0 s t), for t 0.
Given s 0, it follows from the definition of Brownian motion that:
The pre-s process {Bt : 0 t s} is independent of the post-s process
{Bs+t Bs : t 0}.
The post-s process {Bs+t Bs : t 0} has the same distribution as the orig-
inal process {Bt : t 0} (i.e., the post-s process is also a standard Brownian
motion).
Thus, by Definition 6.6, (Bs+t Bs : t 0) is independent of Ft , 0 t s.
Consequently, it follows from Definition 6.7 that
the post-s process {Bs+t Bs : t 0} is independent of the filtration Fs .
Now, we formally state the last result from our observation:
Property 6.11. (Markov Property for Brownian Motion) Let B = {Bt : t 0}
be a standard Brownian motion on (, F, P, {Ft }), where {Ft } is the natural
filtration of process B. Then:
1. For all s 0, the post-s process Bpost-s = {Bs+t Bs : t 0} is independent
of Fs .
2. The post-s process Bpost-s and the original process B have equivalent distri-
butions.
Remark 6.7. We emphasize that the last property is crucial in the definition of
the Ito integral with respect to Brownian motion.

Intuitively, a process { Xt : t 0} is said to be a Markov process if, given t,

all information in the truncated process { Xs : s t} relevant to the probability
distribution of a future Xu , u > t is all contained in Xt . In mathematical terms,
P ( Xu a|( Xs , s t)) = P ( Xu a|Xt ) for a (, ).
As an example, a Brownian motion is a Markov process.
Example 6.31. Show that the Brownian motion {Bt } is a martingale with re-
spect to its natural filtration (i.e., Ft = (Bs : s t),17 the filtration induced
by {Bt }).
17 It is worth noting that both filtrations below are used frequently in the literature:
6.6 Quadratic Variation and Covariation 289
Proof.
E (Bt |Fs ) = E (Bs + Bt Bs |Fs )

= E (Bs |Fs ) + E (Bt Bs |Fs ) = E (Bs |Fs ) = Bs ,
where the third equal sign holds due to the fact that (Bt Bs )Fs by Proper-
ties 6.11, 1 and 6.7, and E (Bt Bs ) = 0.

Remark 6.8. Since all Ito integrals with respect to Brownian motion are mar-
tingales, martingales form an important class of stochastic processes.

Finally, we note that Brownian motion is a basic building block for the con-
struction of a diffusion process (or a diffusion for short), which is a continuous-
time Markov process with (almost surely) continuous sample paths.
The simplest and most fundamental diffusion process is Brownian motion.
A more general example of diffusion processes is a Brownian motion with drift
(e.g., Xt = t + Bt , also see Figure 6.2).
6.6 Quadratic Variation and Covariation
6.6.1 Motivation, Definition, and Notation
There are different ways to measure the variation of a function. Total varia-
tion is a tool for us to measure the total, therefore the absolute value (as you
will see in the definition below), up-and-down movement of a function. The
total variation is used for deterministic real-valued functions (not as sample
paths of random processes), whereas quadratic variation is used for stochastic
processes (we will show that processes such as Brownian motions do not have
finite total variations, but can be dealt with if one uses quadratic variation).
The total variation of a real-valued function on an interval is denoted by
Va ( f ) and defined as
b
n 1
Vab ( f ) = lim | f ( xk+1 ) f ( xk )|,
|P|0 k =0
(6.33)
(1) Fto = (Bs : 0 s t), (2) Ft+ = st Fto , for t 0.
The first is the smallest filtration that makes {Bt } adapted. The second is an extension of the first by
including some zero-probability subsets and has advantages of being complete and right-continuous,
which are convenient and important properties to have. The second filtration is referred to as the
Brownian filtration. We use the first filtration in this example to suppress some conceptual and technical
details involved in the second.
In loose terms and for our purpose, both filtrations are denoted by {F t }, where F t is interpreted
as the set of information generated by the standard Brownian motion on the time interval [0, t].
Fig. 6.2 Brownian motion with drift parameter = 0.2 and volatility parameter = 0.1 is illustrated
using 5,000 randomly selected sample paths over the interval [0, 1]. Note how the over structure drifts
upward about the mean line t. At time t = 1, the frequency distribution of X (1) is shown as a
histogram, which approximates a normal distribution with mean 0.2 and variance 0.1. The current
time is 0 and the solid line is a plot of the expected value E ( X ( t)) = 0.2 t, where t 0
where P : a = x0 < x1 < x2 < < xn = b is a partition on [a, b].

The quadratic variation process is one of the central concepts in classical
continuous-time martingale theory. Martingale processes will be introduced
in a later section.
Definition 6.29. Let X = { Xt , t 0} and Y = {Yt , t 0} be two real-valued

stochastic processes defined on a probability space (, F, P ).
The quadratic variation of X is the process denoted by {[X ]t } and defined by
n
[X ]t = lim ( X t k X t k 1 ) 2 ,
|P|0 k =1
t 0, (6.34)
where the limit is taken in probability18 (see Definition 6.12), and P repre-
sents partitions over the interval [0, t] and |P| is the length of the longest
subintervals associated to P.
18Notice that, among different concepts of convergence of a sequence of random variables (i.e., con-
vergence in probability, almost sure convergence, and convergence in mean square), convergence in
probability is the weakest.
More generally, the quadratic covariation or cross-variation (or just covariation)

of two processes X and Y is the process denoted by {[X, Y ]t } and defined by
n
[X, Y ]t = lim (Xtk Xtk1 )(Ytk Ytk1 ),
|P|0 k =1
t 0, (6.35)
where the limit is taken in probability, and P and |P| are the same as stated
above.
Clearly, [X ]t = [X, X ]t . If there is no ambiguity, to ease notation, [X ]t and [X, Y ]t

may be used to represent quadratic variation process (i.e., {[X ]t }) and covariation
process (i.e., {[X, Y ]t }), respectively.
If X and Y have finite quadratic variations, we rewrite (6.34) and (6.35) with
substitutions
k X = X t k X t k 1 and k Y = Ytk Ytk1
and obtain
n
[X ]t = lim
t0
( k X )2 ,
k =1
n
[X, Y ]t = lim
t0
(k X )(k Y ),
k =1
which motivate the following notation, respectively:
d [ X ] t = ( d Xt )2 , (6.36)
d [X, Y ]t = (d Xt )(d Yt ). (6.37)
6.6.2 Basic Properties
The next two properties of the covariation process can be easily verified.
(i)
Property 6.12. (Symmetry and Bilinearity) Let X (i) = { Xt , t 0}, i = 1, 2, 3 be
three processes with finite quadratic variations. The following properties hold:
1. The covariation process is symmetric. That is,
[ X (1 ) , X (2 ) ] t = [ X (2 ) , X (1 ) ] t , t 0.
2. The covariation process is bilinear. That is, for any constants a, b (, ),
[X (1) , aX (2) + bX (3) ]t = a[X (1) , X (2) ]t + b[X (1) , X (3) ]t , t 0,

(1 ) (2 ) (3 ) (1 ) (3 ) (2 ) (3 )
[aX + bX ,X ]t = a [ X ,X ]t + b[ X ,X ]t , t 0.
Property 6.13 below is motivated by the following:

n
lim
t0
( k X + k Y )2
k =1
n n n
= lim
t0
(k X )2 + 2 lim t0
(k X )(k Y ) + lim
t0
( k Y )2 ,
k =1 k =1 k =1
which is obtained by applying the binomial formula and shows that the co-
variation can be defined by quadratic variations.
Property 6.13. (Covariation Expressed in Terms of Quadratic Variations) Let
X = { Xt } and Y = {Yt } be two processes with finite quadratic variations. Then
1
[X, Y ]t = ([X + Y ]t [X ]t [Y ]t ).
2
Again, let X = { Xt } and Y = {Yt } be two processes. As a motivation for the
next property, let us verify the following identity:
( Xt Yt ) = Xt Yt + Yt Xt + Xt Yt , (6.38)
where Xt , Yt , and ( Xt Yt ) are defined as the corresponding increments over

time interval [t, t + t]. In fact,
LHS = Xt+t Yt+t Xt Yt ,

RHS = Xt (Yt+t Yt ) + Yt ( Xt+t Xt ) + ( Xt+t Xt )(Yt+t Yt ).
A verification of LHS = RHS is straightforward.

The infinitesimal version of (6.38) is a useful product rule:
d ( Xt Yt ) = Xt dYt + Yt dXt + dXt dYt (6.39)
holds under certain condition (e.g., both X and Y are Ito processes, which will
be introduced later) and has an equivalent form:
d ( Xt Yt ) = Xt dYt + Yt dXt + d[X , Y ]t . (6.40)
A proof of the product rule can be done by applying the two-dimensional Itos
lemma (see Exercises 6.32 and 6.33 on page 326). For this the product rule is
also referred to as the Ito product rule.
If f is a deterministic function, we denote [ f ]ba the quadratic variation of f
over interval [a, b]. That is, [ f ]ba [ f ]t , t [a, b].
We will make use of the next two properties.
Property 6.14. Let f be a continuous (deterministic) function on [0, T ]. If f has

finite total variation, then [ f ]ba , the quadratic variation of f , is identically equal
to zero on [a, b]. That is,
[ f ]ba 0. (6.41)
In words, the quadratic variation of a continuous f is identically equal to zero.
Proof. A proof is provided in Remark 6.9 on page 297.

Property 6.15. Let f be a continuous (deterministic) function and X = { Xt } be
a sample-continuous process on [0, T ] (see Definition 6.10 on page 264). Then
[X, f ]t 0, t [0, T ].
Proof. A proof is provided in the Remark 6.10 on page 297.

Example 6.32. Let X = B, a standard Brownian motion, and f (t) = t. It follows
from Property 6.15 that
[B, t ]t 0, t 0.
Using the notation d [X, Y ]t = (d Xt )(d Yt ) defined by (6.37), we write
d[B, t ]t = dB(t) dt
(keep in mind: B(t) Bt by the notational remark on page 264) and obtain
dB(t) dt = 0. (6.42)

As a good exercise, the reader is encouraged to derive identity (6.42) by directly
using the definition of covariation.
Example 6.33. Compute d(et
2 +t
S(t)) given dS(t) = 0.2 dt + 0.095 dB(t).
Solution. Let f (t) = et

2 +t
.
Applying the product rule provided by (6.40), we have
d (e t
2 +t
S(t)) = et
2 +t
dS(t) + d(et
2 +t
)S(t) + d[S, f ]t
t2 + t
dS(t) + (2t + 1)et
2 +t
=e S(t)dt + 0
t2 + t
(0.2 dt + 0.095 dB(t)) + (2t + 1)et
2 +t
=e S(t)dt
t2 + t t2 + t
= (0.2 + (2t + 1)S(t))e dt + 0.095e dB(t).

6.6.3 Quadratic Variation and Covariation Properties of BM
Property 6.16. (Quadratic Variation Property of Brownian Motion) Let B be a

standard Brownian motion on a probability space (, F, P ) (i.e., B = {Bt , t
0} is a P-standard Brownian motion). Then
[B]t = t. (6.43)
Note that the equality (6.43) can be expressed by
(dBt )2 = dt or dB2t = dt. (6.44)
Proof. It is sufficient to show that for partitions Pn : 0 = t0 < t1 < t2 < < tn = t
on the interval [0, t] with tk = kt
n , k = 0, 1, . . . , n 1, n > 0,
n n 1
lim
|Pn |0
(Btk Btk1 )2 = nlim

(Btk Btk1 )2 = t,
k =1 k =0
where the limit is taken in the mean square sense (see Definition 6.13 and Prop-
erty 6.1, 2 on page 265).
We let Bk = Btk+1 Btk , and t = tk+1 tk , k = 0, 1, . . . , n 1, and
n 1 n 1
Sn = (Btk Btk1 )2 = B2k .
k =0 k =0
Since

n 1 n 1 n 1 n 1
E (Sn ) = E B2k = E(B2k ) = Var(Bk ) = t = t,
k =0 k =0 k =0 k =0
we have
n 1 n 1
E ((Sn t)2 ) = Var(Sn ) = Var(B2k ) = [E(B4k ) (E(B2k ))2 ]
k =0 k =0
n 1
= [( Var(Bk ))2 kurt(Bk ) ( Var(Bk ))2 ]
k =0
n 1 n 1
t2
= [( Var(Bk ))2 (3 1)] = 2 t2 = 2 n 0 as n .
k =0 k =0
Since the convergence in mean square implies the convergence in probability,

we have proved [B]t = t indeed.

It follows from [B]t = t and Property 6.14 that with probability 1, sample
paths of Brownian motion do not have bounded variation on [0, t] (nor on
any interval by the time-homogenous property of Brownian motion).
In fact, we have just proved Property 6.9, 3 on page 284.

Now, consider two standard Brownian motions
(i)
B(i) = {Bt , t 0}, i = 1, 2,
which are defined on the same filtered probability space (, F, {Ft }, P ) and
adapted to the filtration {Ft }. Let
(i) (i)
B(i) = Bt Bs , where t > s and i = 1, 2. (6.45)
We say that B(1) and B(2) have correlation if, for t > s 0,
1. Cov (B(1) , B(2) ) = t,

2. B(1) and B(2) are independent of Fs , and
3. B(1) and B(2) have a bivariate normal distribution.19
Property 6.17. (Quadratic Covariation Property of Brownian Motion) If B(1)

and B(2) have correlation , then
[B(1) , B(2) ]t = t. (6.46)
Note that (6.46) can be expressed by

(dB(1) )(dB(2) ) = dt or dB(1) dB(2) = dt. (6.47)
Proof. Let
1
X = a(B(1) + B(2) ) with a2 = .
2 + 2
Thus, 1
2a2
= 1 + .
Step 1. We verify that X is a standard Brownian motion as follows:
Let X = a(B(1) + B(2) ), where B(i) are defined in (6.45). We establish
Var(X ) = E (X 2 ) = E ( a2 (B(1) + B(2) )2 )
= a2 E ((B(1) )2 + 2(B(1) )(B(2) ) + (B(2) )2 )
= a2 [E ((B(1) )2 ) + 2E ((B(1) )(B(2) )) + E ((B(2) )2 )]
= a2 [ Var(B(1) ) + 2 Cov(B(1) , B(2) ) + Var(B(2) )]
= a2 [t s + 2(t s) + t s]
1
= (2 + 2)(t s) = t s,
2 + 2
E (X ) = a(E (B(1) ) + E (B(2) )) = 0,

(1 ) (2 )
X0 = a(B0 + B0 ) = 0.
a.s.
By Property 6.16, [X ]t = t.
Step 2. Applying Property 6.12, the symmetry and bilinearity of covariation
process yield
19The random n-vector X = [ X1 X2 . . . Xn ] is (or the random variables X1 , X2 , . . . , Xn are) said to have
a multivariate normal distribution if and only if all linear combinations of X1 , X2 , . . ., Xn are normally
distributed. When n = 2, X is said to have a bivariate normal distribution.
t = [X ]t = [X, X ]t = [a(B(1) + B(2) ), a(B(1) + B(2) )]t

= a[B(1) , a(B(1) + B(2) )]t + a[B(2) , a(B(1) + B(2) )]t
= a2 ([B(1) , B(1) ]t + 2[B(1) , B(2) ]t + [B(2) , B(2) ]t )
= a2 ([B(1) ]t + 2[B(1) , B(2) ]t + [B(2) ]t )
= a2 (t + 2[B(1) , B(2) ]t + t) = 2a2 (t + [B(1) , B(2) ]t ).
Solving for [B(1) , B(2) ]t , we obtain

(1 ) (2 ) t 1
[B ,B ]t = 2 t = t 1 = t(1 + 1) = t.
2a 2a2

Clearly, if B(1) and B(2) are uncorrelated, then [B(1) , B(2) ]t = 0.

For future convenience, we summarize the results represented by the iden-
tities (6.41), (6.42), (6.44), and (6.47) in the following multiplication table for
Brownian motion variation and covariation:
dt dB(1) dB(2)
dt 0 0 0
dB(1) 0 dt dt
dB(2) 0 dt dt
In fact, the multiplication table can be enlarged to any size we wish. More
specifically, given that the correlation of Brownian motions B(i) and B( j) is
i,j , i, j = 1, 2, 3, we can establish a table of size 5 5 as below:
dt dB(1) dB(2) dB(3)

dt 0 0 0 0
dB(1) 0 dt 1,2 dt 1,3 dt
dB(2) 0 1,2 dt dt 2,3 dt
dB(3) 0 1,3 dt 2,3 dt dt
Such tables function like a handy computational tool in studying stochastic
models involving more than one random source.
Example 6.34. (Interpretation of the Multiplication Product Table) In the

(2, 2) entry of the table,
(dt)2 = d[ t ]t 0
by Property 6.14, which states that [ f ]t 0 if f is a continuous (deterministic)

function (in our case, f (t) = t).

6.6.4 Significance of Quadratic Variation
The following three remarks provide some insights into the concept of quadratic
variation and are for the interested reader.
Remark 6.9. Recall from (6.33) on page 289 that the total variation of a real-
valued function on interval is denoted Vab ( f ) and defined as
n 1
Vab ( f ) = lim
|P|0 k =0
| f ( xk+1 ) f ( xk )|,
where P : a = x0 < x1 < x2 < < xn = b is a partition on [a, b].
Let f be a continuous function on [a, b]. We claim that if f has finite total
variation, then [ f ]ba , the quadratic variation of f , is identically equal to zero on
[a, b] (this is why quadratic variation is not defined for deterministic functions).
In fact, Vab ( f ) being finite implies that M > 0 such that
n 1
lim | f ( xk+1 ) f ( xk )| M.
|P|0 k =0
Since f being continuous on a closed interval [a, b] implies the uniform conti-
nuity of f on [a, b]: > 0, N > 0 such that | f ( xk+1 f ( xk )| < whenever
| xk+1 xk | < N1 . Consequently,
n 1 n 1
( f ( xk+1 ) f ( xk ))2 < | f ( xk+1 ) f ( xk )| M,
k =0 k =0
which implies that

n 1
0 < [ f ]xa [ f ]ba lim sup ( f ( xk+1 ) f ( xk ))2 M 0.
k =0
Thus, [ f ]xa 0 where a x b.

Remark 6.10. A proof of Property 6.15 on page 293 is provided below for the
interested reader.
Since for each fixed , sample path X (t, ) = X (t) of X being continuous
on a closed interval [0, T ] implies X (t) being absolutely continuous on [0, T ],
> 0, N > 0, such that
1
|X (tk+1 ) X (tk )| < whenever |tk+1 tk | < .
N
Consequently,
n 1 n 1
| (X (tk+1 ) X (tk ))( f (tk+1 ) f (tk ))| < | f (tk+1 ) f (tk )| V0T ( f ),
k =0 k =0
where V0T ( f ) is the total variation of f on [0, T ], which is finite due to the con-
tinuity of f . It follows that
n 1
0 |[X, f ]t | lim | f ( xk+1 ) f ( xk )| V0T ( f ) 0.
k =0
Thus, [X, f ]t = 0.

Remark 6.11. Since a stochastic differential equation (s.d.e.) is defined by a

stochastic integration equation (s.i.e.) which we have not yet defined, our dis-
cussion here can only be formal. However, it may help to immediately satisfy
the curiosity of those who are wondering about the significance of quadratic
variation.
Loosely speaking, the local dynamics of a stochastic process driven by Brow-
nian motion over time interval [t, t + t] can be represented by approximations
in the form below
( X (t)) = (t) t + (t) B(t), (6.48)
where {(t)} is an adapted drift process, {(t)} is an adapted volatility pro-
cess, and B(t) causes fluctuations. The adaptedness refers to Brownian fil-
tration (the set of information of the past history of the Brownian motion B).
The infinitesimal version of (6.48) is
dX (t) = (t) dt + (t) dB(t). (6.49)
When both (t) and (t) are functions of X (t) only, we replace (t) and (t)
by ( X (t)) and ( X (t)) respectively and have
dX (t) = ( X (t)) dt + ( X (t)) dB(t), (6.50)
which defines, in loose terms, a diffusion process.

Note that (6.27) on page 279 is a special case of (6.50) when both ( X (t))
and ( X (t)) are constant, and (6.50) is a special case of (6.49).
The corresponding s.i.e. to s.d.e. (6.49) can be expressed by
1 t 1 t
X (t) = X (t, ) = (s, ) ds + (s, ) dB(s)
0 0
= f.t.v. term + i.t.v. term,
where f.t.v. and i.t.v. stand for finite total variation and infinite total variation
respectively. The second term on the right labeled by the infinite total variation
term is because, with probability 1, the total variation of a Brownian path on
any interval no matter how small is infinite. It is the infinite total variation term
that complicates our interpretation of the limiting procedure in the ordinary
sense.
6.7 Ito Integral: A Brief Introduction 299
A straightforward computation shows that for a diffusion process X (i.e., for

X satisfying diffusion equation (6.50))
1 t
[ X ]t = 2 (s) ds. (6.51)
0
This is why we impose the condition of 2 (t) = 2 ( X (t)) in the diffusion equa-
tion being square integrable.

6.7 Ito Integral: A Brief Introduction
6.7.1 Importance of Ito Integral with Respect to BM
Stochastic integrations may be taken with respect to different stochastic pro-

cesses. For both theoretical and practical considerations, we will introduce
only the one with respect to the standard Brownian motion:
Brownian motion is theoretically important because its sample paths are
(almost surely) continuous but nowhere differentiable (see Billingsley [3];
Durrett [9]; and Paley, Wiener, and Zygmund [25]).
The integration with respect to Brownian motion is practically impor-
tant because {dB(t)} is a white noise process (see Definition 6.25 on
page 280).20
To avoid involving too many technicalities, we will approach the concept of

the Ito integral with respect to standard Brownian motion by emphasizing the
key idea of the definition under which certain computations become easier and
the integral of a stochastic process will produce a martingale.
The mathematical rigor in the definition of a stochastic integral can be per-
fected when we display a similar drill to the familiar one in Riemann integrals
over interval [a, b].21
20 Brownian noise (also known as brown noise or red noise) is the kind of signal noise produced by
Brownian motion. Naturally, it is also called random walk noise as a Brownian motion can be viewed
as a limit of random walks. Note that a random walk noise is not a white noise (see Example 6.27).
21 That is, following an approximation procedure: Step 1. Divide the interval into finitely many subin-
tervals (the partition). Step 2. Construct a simple function (use step functions for intuition) that has a
constant value on each of the subintervals of the partition (the upper and lower sums). Step 3. Define
integrals of simple functions (simple processes are random step functions). Step 4. Take the limit of
these simple functions as more and more dividing points are added to the partition. If the limit exists,
Just as the indefinite integral defines a function in the deterministic calculus,

the Ito integral defines a stochastic process.
Let B = {Bt , t 0} be the standard Brownian motion adapted to its natural
filtration {Ft } and keep in mind that independent increments mean that for
s, t with s < t, the random variable Bt Bs is independent of Fs .
Let f = { f t , t 0} be a stochastic process to integrate and adapted to the
same filtration {Ft }, where the idea of the adaptedness requirement is that f
is allowed to depend on the history of B but not on the future development
of B. That is, f s , s < t do not contain any information of future increments
Bt Bs , t > s. Consequently,
E ( f s (Bt Bs )|Fs ) = f s E (Bt Bs |Fs ) = f s E (Bt Bs ) = 0,
where the first equal sign holds due to a property of conditional expectation
(taking out f s , which is known), the second equal sign holds due to the Markov
property of Brownian motion, and the last equal sign holds due to the defini-
tion of Brownian motion. Thus, if we let dBs = Bs+ds Bs (where ds > 0 is
considered to be infinitesimal), then
E ( f s dBs |Fs ) = 0.
To introduce the Ito integral with respect to Brownian motion B, we begin

with dividing interval [0, t] into n subintervals 0 = t0 < t1 < t2 < tn = t and
forming a sum associated with the partition
n
Sn = fti1 (Bti Bti1 ).
i =1
The Ito integral of the process { f t } with respect to Brownian process {Bt } is denoted
by the process {Yt } and defined by
1 t
Yt = f s dBs = lim Sn , (6.52)
0 n
where the limit is taken in terms of the mean squared errors:
lim E ((Yt Sn )2 ) = 0.
n
{Yt } is called a process driven by Brownian motion {Bt } and transformed by the
integrand process { f t }.
Before considering computational details, let us understand (6.52) concep-
tually:
it is called the Riemann integral and the function is called the Riemann integrable (Ito integrability
requires convergence in mean square).
Note that a simple function is a finite linear combination of indicator functions of measurable sets.
Thus all step functions are simple functions.
6.7 Ito Integral: A Brief Introduction 301
1. Sn by definition is a random variable for each n, so is Yt for each t. Just as an

indefinite integral transforms an integrand function to another function in
the deterministic calculus, an Ito integral transforms an integrand process
to another process in the stochastic calculus. One application of Ito integrals
is to construct a new process.
2. An important property of Ito integrals with respect to Brownian motion is
that
{Yt } is a martingale.
In fact, for a, b [0, t] with b > a, if a and b are not among the dividing
points ti in the partition, then we can insert them into the sequence of divid-
ing points to obtain that
1 b
E (Yb Ya |Fa ) = E f s dBs |Fa
a
1 b 1 b
= E ( f s dBs |Fa ) = E (E ( f s dBs |Fs )|Fa )
a a
1 b
= E (0|Fa ) = 0.
a
3. In a more general version of stochastic integrals, the driven process does
not have to be Brownian motion. In that case, from the argument in item 2,
we see that {Yt } is always a martingale as long as the driven process is a
martingale.
6.7.3 A Famous Example
The example below is well known.

2t
Example 6.35. To calculate 0 Bs dBs , we divide interval [0, t] into n subinter-
vals with equal length t = nt ,
1 2 1 1 n
Bt = (B2t B20 ) = (B2it B2(i1)t )
2 2 2 i =1
1 n
2 i
= (Bit B(i1)t )(Bit + B(i1)t )
=1
1 n
2 i
= (Bit B(i1)t )(Bit B(i1)t + 2B(i1)t )
=1
1 n n
=
2 i =1
(Bit B(i1)t )2 + (Bit B(i1)t )B(i1)t
i =1
= I1 + I2
1 t
1
[B]t + Bs dBs
2 0
1 t
t
= + Bs dBs ,
2 0
where the convergence as n is in probability,
1 n
2 i
I1 = (Bit B(i1)t )2 ,
=1
n
I2 = (Bit B(i1)t )B(i1)t.
i =1
We obtain 1 t
1 t
Bs dBs = B2t .
0 2 2
Equivalently,
d(B2t ) = 2Bt dBt + dt.

6.8 Itos Formula for Brownian Motion
6.8.1 Ito Processes
Definition 6.30. A one-dimensional Ito process is a stochastic processX = { X (t)}

defined on a probability space (, F, P ) and has an expression
1 t 1 t
X ( t ) = X (0) + ( X (s), s) ds + ( X (s), s) dB(s), 0 t T, (6.53)
0 0
where {( X (t), t)} is an adapted drift process, and {( X (t), t)} is an adapted
volatility process and square integrable. The adaptedness refers to the Brownian
filtration (the set of information of the past history of the Brownian motion B).
An Ito process X = { X (t)} is said to be an Ito diffusion if both drift process
and volatility process are functions of X (t) only. That is, X has the expression
1 t 1 t
X ( t ) = X (0) + ( X (s)) ds + ( X (s)) dB(s), 0 t T, (6.54)
0 0
where ( X (t)) and ( X (t)) are also known as the drift coefficient and diffusion
coefficient of X, respectively.
As a shorthand notation, we write stochastic integral equation (6.53) in the

form of the stochastic differential equation:
dX (t) = ( X (t), t) dt + ( X (t), t) dB(t), 0 t T. (6.55)

6.8 Itos Formula for Brownian Motion 303
Remark 6.12.
1. An Ito process is a process of a sum of the initial state of the process and two
integrals:
X (t, ) = X (0, ) + I1 (t, ) + I2 (t, ), 0 t T, where
1 t
I1 (t, ) = ( X (s, ), s) ds, 0 t T,
0
1 t
I2 (t, ) = ( X (s, ), s) dB(s), 0 t T.
0
For each fixed , I1 is an ordinary integral and I2 an Ito integral. For (6.53)
to be well defined, (t, X (t)) must be integrable in the ordinary sense, and
( X (t), t) must be integrable in the stochastic sense as we defined in Sec-
tion 6.7.2. The square integrability of 2 ( X (t), t) ensures that the quadratic
variation of the process X is finite. In fact, a straightforward calculation
shows that 1 t
[ X ]t = 2 ( X (u), u) du.
0
The detailed verification is left as an exercise for the reader.
2. Although the form of stochastic differential equations has the advantage of
being intuitive in modern financial theory, stochastic differential equations
acquire mathematical meanings only through their corresponding stochastic
integral equations. Equation (6.55) alone is not well defined, and dividing
both sides of the equation by dt is forbidden because the Brownian path is
non-differentiable.22

Example 6.36. Both familiar processes X and S with constants and on
page 278 defined by
dX (t) = dt + dB(t),
dS(t) = S(t) dt + S(t) dB(t),
respectively, are Ito diffusions.
We emphasize that both s.d.e.s should be understood as defined by their
corresponding s.i.e.s:
1 t 1 t
X ( t ) = X (0) + dt + dB(s),
0 0
1 t 1 t
S ( t ) = S (0) + S(u) du + S(u) dB(u).
0 0

22 With probability 1 the Brownian path is non-differentiable in the ordinary sense, but in the context
of generalized stochastic process, dB
dt is well defined as a generalized function on an infinite dimensional
space, which is a topic beyond the scope of this book.
6.8.2 Itos Lemma for Brownian Motion
Itos lemma is the tool of the trade in continuous-time stochastic process mod-
eling.
Theorem 6.1. (Itos Lemma) Let f ( x, t) be a function that is continuously differen-

tiable in t and twice continuously differentiable in x.
Let X = { X (t)} be an Ito process represented by the s.d.e.
dX (t) = ( X (t), t) dt + ( X (t), t) dB(t), 0 t T. (6.56)
Define a process Y = {Y (t)} by Y (t) = f ( X (t), t), 0 t T.

Then Y = {Y (t)} is an Ito process that has a s.d.e. representation with

f f 1 2 f
dY (t) = ( X (t), t) + ( X (t), t) ( X (t), t) + 2 ( X (t), t) 2 ( X (t), t) dt
t x 2 x
f
+ (t) ( X (t), t) dB(t). (6.57)
x
In less detail, formula (6.57), which is referred to as Itos formula, can be written
into
1 2
dY = f t + f x + f xx dt + f x dB. (6.58)
2
An informal proof of the case when X = B (i.e., = 0 and = 1) will be
given shortly (see Example 6.39 on page 307). The idea of the proof for the
general case is similar but with many more tedious computational details.
Remark 6.13.
1. Note that f ( x, t) in Theorem 6.1 is a smooth function, and that
1
Y = f t + f x + 2 f xx
2
Y = f x
are the drift and volatility processes, respectively, for the new process Y.
2. In words, Itos lemma says that Ito processes are stable under smooth maps
in the sense that any smooth function maps (sends) an Ito process X in terms
of its driven process B, drift X , and volatility coefficient X to another Ito
process in terms of its driven process B, drift Y , and volatility coefficient
Y . In short, a smooth function of an Ito process is an Ito process.
3. There are different versions of Itos lemma (e.g., Itos lemma for jump-
diffusion processes, whereas Theorem 6.1 for Brownian motions), which are
widely employed in modern financial theory. The best known application
of Itos lemma is in the derivation of the Black-Scholes-Merton equation for
option values, which will be introduced in a later chapter.

Example 6.37. Suppose that a stock price is modeled by a process S = {S(t)}

where
S(t) = et+B(t)
with = 0 and > 0 being constant. What is the expected growth rate of the
stock at any given time t T?
Solution 1.
Step 1. Identify the given process X and smooth function f in Itos formula:
X (t) = B(t) (thus, X = 0, X = 1) and f ( x, t) = et+x ,
where X = ( X (t), t) and X = ( X (t), t) as defined in (6.56). Clearly, the con-

ditions of Itos lemma are satisfied.
Step 2. Compute partial derivatives of f in (6.58):
f t = f (B(t), t) = S(t),
f x = f (B(t), t) = S(t),
f xx = 2 f (B(t), t) = 2 S(t).
Step 3. Apply Itos lemma to obtain dY = dS(t):

1 2
dS(t) = dS = f t + X f x + X f xx dt + X f x dB
2

1 2
= S(t) + 0 + S(t) dt + S(t) dB(t)
2

1
= + 2 S(t) dt + S(t) dB(t).
2
Step 4. The answer to the question is + 12 2 .

As a postscript, we point out that in a risk-neutral world, if the stock pays a
known dividend yield q, then

dS(t) 1 2
= + dt + dB(t)
S ( t) 2
implies + 12 2 = r q. We obtain = r q 12 2 .
Solution 2.
Step 1. Identify the given process X and smooth function f in Itos lemma:
Keep s.d.e. (6.56) in mind.
dX (t) = X ( X (t), t) dt + X ( X (t), t) dB(t), 0tT
is a shorthand of s.i.e. (6.53):

1 t 1 t
X ( t ) = X (0) + X ( X (s), s) ds + X ( X (s), s) dB(s), 0 t T.
0 0
It is natural to attempt X (t) = t + B(t), which is equivalent to

1 t 1 t
X ( t) = 0 + ds + dB(s), 0 t T.
0 0
Thus, we identify X = and X = and summarize all we need for applying

Itos lemma below:
X (t) = t + B(t), X = , X = , and f ( x, t) = f ( x ) = ex .
Notice that the conditions of Itos lemma are satisfied.

Step 2. Compute partial derivatives of f in (6.58):
f t = 0, f x = f xx = f ( X (t)) = S(t).
Step 3. Apply Itos lemma to obtain dY = dS(t):

1 2
dS(t) = dS = f t + X f x + X f xx dt + X f x dB
2

1 2
= 0 + S(t) + S(t) dt + S(t) dB(t)
2

1 2
= + S(t) dt + S(t) dB(t).
2
Step 4. The answer to the question is + 12 2 , which is the same as the one we
obtained earlier.

Using (6.58)

1 2
dY = f t + f x + f xx dt + ( f x ) dB,
2
Y
Y
we have a corollary of Itos lemma:

Corollary 6.1. Let {Ft } be the Brownian filtration.
Let Y be an Ito process expressed by (6.58). Then

1
E (dY |Ft ) = f t + f x + 2 f xx dt,
2
Var(dY |Ft ) = ( f x )2 dt.
Proof. It follows from:

1. Itos lemma23 (conditions are satisfied indeed),

2. Properties 6.2, 6.3, 6.6, and 6.11,
3. dB N (0, dt),
that both formulas in Corollary 6.1 hold.

Example 6.38. Suppose that a stock price is modeled by a process S = {S(t)}

where
S(t) = e 2 t +2B(t) .
1 2
What is the (conditional) expected growth rate of the stock at any given time
t T?
Solution 1.
Since X (t) = 12 t2 + 2B(t) is an Ito process with X = t and X = 2,
1 2 +2B( t )
Y = f ( X (t)) = e 2 t = S ( t ), where f ( x ) = ex
is also an Ito process. Note that the conditions of Itos lemma must be verified
before the lemma can be applied.
Applying Corollary 6.1, we obtain

1 2
E (dS(t)|Ft ) = E (dY |Ft ) = f t + f x + f xx dt
2
= (0 + tS(t) + 2S(t)) dt
= (2 + t)S(t) dt.
The answer to the question is 2 + t.

Solution 2.
Take the same steps in Example 6.37. The detailed work is left as an exercise
to the reader.

Example 6.39. (Informal Proof of Itos Lemma) Consider the case = 0, = 1

and X = B.
Let f ( x ) be a function that is twice continuously differentiable. Then
n
1
f ( x + x ) f ( x ) = f ( x )x + f ( x )(x )2 + o ((x )2 ),
2 i =1
where the little o is a popular notation in analysis: o ( ) refers to a quantity such

that lim 0 o ( )/ = 0. This notation is very convenient as in many cases such
a quantity needs not be specified.
We recall the multiplication table:
23Itos lemma ensures that Y in (6.58) is an Ito process. Thus, both drift process Y and volatility
process Y are adapted to the Brownian filtration by definition.
dt dBt
dt 0 0
dBt 0 dt
which implies that (dBt )m 0 as dt 0 for m 2.
For an arbitrary partition on interval [0, t]: 0 = t0 < t1 < t2 < tn = t,
n
f (Bt ) f (Bs ) = ( f (Bti ) f (Bti1 ))
i =1
n
1 n
f (Bti )(Bti Bti1 ) +
2 i
= f (Bti1 )(Bti Bti1 )2
i =1 =1
n
+ o ((Bti Bti1 )2 )
i =1
= I1 + I2 + I3 ,
where
n 1 t

I1 = f (Bti )(Bti Bti1 ) f (Bu ) dBu ,
i =1 0
n 1 t
1 1
I2 = f (Bti1 )(Bti Bti1 )2 f (Bu ) du,
2 i =1 2 0
n
I3 = o((Bti Bti1 )2 ) 0,
i =1
where, by applying the multiplication table, I3 0 as the partition becomes

finer and finer so that max(ti+1 ti ) 0.
Consequently, for twice differentiable function f , the basic form of Itos
lemma can be understood as a Taylor expansion to second order
1
d f ( X ) = f ( X )dX + f ( X )(dX )2 ,
2
where X = B = {Bt }, the standard Brownian motion, and the quadratic term
(dX )2 is the quadratic variation of the process X, i.e., (dB)2 = dt. That is,
1
d f (B) = f (B)dB + f (B)dt. (6.59)
2
Assume that all partial derivatives of g( x, t) exist and are continuous at
( x, t). Define Yt = g(B(t), t), then
dYt = g(B(t + dt), t + dt) g(B(t), t)

= ( g(B(t + dt), t + dt) g(B(t + dt), t)) + [ g(B(t + dt), t) g(B(t), t)]

g 1 2 g g 1 2 g
= dt + (dt) + higher order terms +
2
dB + dt ,
t 2 t2 x 2 x2
where we apply (6.59) to the inside of brackets with
d f (B) = g(B(t + dt), t) g(B(t), t),

g(B(t), t)
f (B) = ,
x
2 g(B(t), t)
f (B) = .
x2
We discard all terms involving dt to a power higher then 1 and obtain
g(B(t + dt), t) 1 2 g(B(t), t) g(B(t), t)

dYt = dt + dt + dB
t 2 x
2 x
g(B(t + dt), t) g(B(t), t) g(B(t), t)
= dt + dt
t t t
1 2 g(B(t), t) g(B(t), t)
+ dt + dB
2 x 2 x

2 g(B(t), t) 1 3 g(B(t), t) g(B(t), t)
= dB + dt dt + dt
xt 2 x t
2 t
1 2 g(B(t), t) g(B(t), t)
+ dt + dB,
2 x 2 x
where we apply (6.59) to the inside of the last brackets with
g(B(t + dt), t) g(B(t), t)

d f (B) = ,
t t
g2 (B(t), t)
f (B) = ,
xt
3 g(B(t), t)
f (B) = .
x2 t
Once again, we discard all terms involving dt to a power higher than 1 and
obtain the form of Itos lemma in the case where X = B (i.e., = 0 and = 1)
given in (6.58):

g 1 2 g g
dYt = + dt + dB.
t 2 x 2 x

6.8.3 Risk-Neutral Probability Measure
Those who on the sell side of the security industry (e.g., market makers) and
policy makers (e.g., the Federal Reserve) usually work with the risk-neutral
probability measure.
In the risk-neutral world, if we ignore dividends, the (conditional) expecta-
tion of the stock returns must be equal to the risk-free rate. To interpret this
statement in mathematical language, let us recall (6.55) with (t, X (t)) = (t)
and (t, X (t)) = (t):
dX (t) = (t) dt + (t) dB(t), (6.60)
where {B(t)} is a P-Brownian motion. To emphasize this fact, we rewrite B(t)

into BP (t). Accordingly, (6.60) becomes
dX (t) = (t) dt + (t) dBP (t). (6.61)
Thus, in the risk-neutral world (i.e., under the risk-neutral probability mea-
sure Q),
dX (t) = r(t) dt + (t) dBQ (t), (6.62)
where BQ is a Q-Brownian motion. This Q is called the risk-neutral measure
because the expected appreciation rate of the log return on the stock is identical
to the risk-free rate despite the presence of risk in the form of (t) dBQ (t).
Note that under our consideration, drift parameter (t), scale parameter
(t), and the risk-free rate r(t) are all constants. Thus, we write
(t) = , (t) = , r(t) = r.
Recall the definition of the Sharpe ratio (i.e., the market price of risk from
Section 4.2.1 on page 166), written S, and we have
r
S= r = S.

Taking the difference of (6.61) and (6.62) yields
r
dBQ (t) = S dt + dBP (t) = dt + dBP (t), (6.63)

where BQ and BP are Brownian motions under probability measure Q and
P, respectively. Notice that (6.63) connects P and Q only implicitly (an explicit
connection will be given shortly). Nevertheless, since (6.63) is equivalent to
r
BQ (t) = t + BP (t), (6.64)

BP is adapted if and only if BQ is adapted to the same filtration.
It follows from the martingale property that E (dB(t)|Ft ) = 0 if B is adapted
to the filtration {Ft } that
EP (dX (t)|Ft ) EQ (dX (t)|Ft ) = EP (dX (t)|Ft ) rdt,

dS(t) 44 dS(t) 44 dS(t) 44
EP 4 Ft EQ 4 Ft = EP 4 Ft rdt. (6.65)
S ( t) S ( t) S ( t)
In words, (6.65) says that the difference between the (conditional) expected
log returns over time period [t, t + dt] (of a stock) under real-world and risk-
neutral probability measures is the risk premium.
Example 6.40. Equality (6.65) assumes that the stock pays no dividend. How
should it be modified if a dividend-paying stock is under consideration?
Solution. For a dividend-paying stock, (6.62) needs to be modified to
dX (t) = (r q) dt + (t) dBQ (t),
where q is the annual dividend yield of the stock. Consequently, (6.65) is

changed accordingly to

dS(t) 44 dS(t) 44 dS(t) 44
EP F
4 t E Q F
4 t = E P 4 t (r q)dt.
F
S ( t) S ( t) S ( t)

6.8.4 Girsanov Theorem for a Single Brownian Motion
Relation (6.64)
r
BQ (t) = t + BP (t)

changes Brownian motion with no drift to Brownian motion with drift, which
can be described by the simplest version of Girsanov theorem for a single Brow-
nian motion.
The Girsanov theorem, also referred to as the Cameron-Martin-Girsanov theo-
rem, is a tool of changing probability measures. Changes of probability mea-
sures can be used for changes of the expectation of a random variable, which
in turn can be used for security pricing in finance, particularly for derivative
pricing. After all, establishing a probability measure in practice may not al-
ways be done in an objective way. It is desirable to look at the effect of different
probability measures on expectations.
Theorem 6.2. (Girsanov Theorem for a Single Brownian Motion with Drift)
Let B = {B(t)} be a standard Brownian motion on (, F, {Ft }, P ), where {Ft }
is the natural filtration of B. Let { (t)} be an adapted process to {Ft } satisfying
2T
1
e2 0 ( (s))2 ds < (Novikovs condition). For each t [0, T ], define
2t 2t
1. D (t) = e (s) dB(s) 12 0 ( ( s ))
2 ds
0 ,
2t
2. W (t) = B(t) + 0 (s) ds,
dQ
3. a measure Q with dP = D ( T ).
Then { D (t)} is a martingale under P and {W (t)} is a standard Brownian motion
under Q.24
Novikovs condition is a sufficient condition for { D (t)} to be a martingale.

For our purpose, it is sufficient to know a special case of Theorem 6.2:
If {B(t)} is a standard Brownian motion under P and (t) = is constant,
then {t + B(t)} is a standard Brownian motion under Q.
The Girsanov theorem (the version in Theorem 6.2, change of measure) de-
scribes how the dynamics of stochastic processes change when the original
probability measure P is changed to an equivalent probability measure Q. In
other words, it describes the distribution of the process {W (t)} under the new
probability measure Q. As a by-product, Girsanov theorem also provides a tool
for the martingale approach to pricing derivatives.
The following note is for the interested reader.
Interpretation and explanation.
1. To ease the notation, we drop T and let D = D ( T ). Note that
dQ dQ
=D ( ) = D ( ),
dP dP 1
Q ( A) = D ( ) dP ( ) for each A F.
A
At this point of our formal observation, D functions like a probability den-

sity (in fact, it is, as we will explain shortly). For this { D (t)} is also called
the density process of Q relative to P.
2. There are two basic concepts involved in the hypotheses of more rigorous
versions of Girsanov theorem. One is called the absolute continuity of mea-
sures.25 This relation between P and Q assures the existence of a nonnega-
tive random variable D such that
1
Q ( A) = D ( ) dP ( ) for all A F.
A
2t
( s) dB( s). A straightforward computation leads to D ( t) = e Xt 2 [ X ] t and W ( t) =
1
24 (a) Let Xt = 0
B( t) [B, X ]t . (b) If ( t) = is constant, then D ( t) = e B( t ) 12 2 t
and W ( t) = t + B( t).
25 Given a filtered probability space ( , F, {F }, P ) and a probability measure Q defined on measur-
t
able space ( , F), Q is said to be absolutely continuous with respect to probability measure P if any
A F with P ( A) = 0 implies Q ( A) = 0 (i.e., every P-null event is a Q-null event). This is one of the
reasons we prefer another filtration (infinitesimally larger than the natural filtration) for the Brownian
motion (see the footnote on page 289) and work with complete probability space.
D is called the Radon-Nikodym derivative (or density) of Q with respect to P,

denoted by
dQ
( ) = D ( ).
dP
The Radon-Nikodym derivative connects probabilities in one measure P to
probabilities in an equivalent measure26 Q, which is another basic concept
involved in changing measures. This relation between P and Q assures that
P ( | D ( ) > 0) = 1. Indeed, Radon-Nikodym derivative gives a probability
density.
The Radon-Nikodym derivative is a topic beyond the scope of this book. We
refer the reader to any standard graduate-level textbook on measure theory
so that it can be learned properly.
S(t)
3. Keep in mind that our model assumes that X (t) = ln S , the log return on
0
a stock over time period [0, t], is an Ito process satisfying s.d.e. (6.27) on
page 279:
dX (t) = dt + dB(t).
Equivalently, the stock process S(t) is a geometric Brownian motion (to be

defined shortly) governed by the s.d.e.
dS(t) = S(t) dt + S(t) dB(t).
The conditions of Itos lemma are satisfied if we let f ( x, t) = ert x and S(t)
satisfy the last equation. We apply Itos lemma to the deflated (discounted)
stock price, Y (t) = ert S(t), and obtain
dY (t) = (rert S(t) + S(t)ert + 0) dt + S(t)ert dBP (t)

= ert S(t)[(r + )dt + dBP (t)]

r
= Y (t) dt + dBP (t)

= Y (t) dBQ (t),
r
where the last equal sign holds due to Girsanov theorem with = ,
which assures that {BQ (t)} defined by
r
BQ (t) = t + BP (t)

26 Two measures P and Q are equivalent if for any A F, P ( A) = 0 Q ( A) = 0.

is a standard Brownian motion under Q, therefore a martingale process with

respect to its natural filtration under Q. Consequently, so is the deflated
price process {Y (t)} under Q for27
dY (t) = Y (t) dBQ (t). (6.66)

For these reasons we note that Q is also referred to as an equivalent martingale
measure.
It is worth noting that:
s.d.e. (6.66) says that the risk-neutral probability measure Q renders the
price process {S(t)} to be a martingale {Y (t)}, and when probability
measures are changed in the way described by Girsanov theorem, the
volatility is not altered.
Remark 6.14. In practice, an equivalent martingale measure Q is derived based

on the market data of various financial instruments rather than assumed by the
derivative pricing model. From the stand point of the construction of Q, it is
an empirical measure. For the technique of estimating risk-neutral distribu-
tions, we refer the reader to the literature (e.g., Hunt and Kennedy [15], and
Malz [20]).

A formal statement of Girsanov theorem must operate at a higher level

of mathematical abstraction. We refer the reader to standard graduate-level
stochastic calculus textbooks. For the proof of Theorem 6.2, we refer the reader
to the literature (e.g., Karatzas and Shreve [17]).
6.9 Geometric Brownian Motion
6.9.1 GBM: Definition
In terms of stochastic differential equations, the definition of Brownian motion

with drift and scaling can be restated in the following:
A stochastic process { X (t)} is said to be a Brownian motion with drift and
scaling if it is a solution to the stochastic differential equation
dX (t) = dt + dB(t), (6.67)
where and > 0 are constant.

Since taking the integral on both sides of (6.67) with initial condition X (0) = x0 :
a.s.
27 This is significant because martingales model a fair game, and security valuation is the determina-
tion of the fair price of a security.
6.9 Geometric Brownian Motion 315
1 t 1 t 1 t
dX (s) = ds + dB(s)
0 0 0
yields
X (t) = x0 + t + B(t),
{ X (t)} is a Brownian motion with drift and scaling if and only if
X (t) = x0 + t + B(t) for some x0 R. (6.68)
Note that is also referred to as a scale parameter or volatility parameter or dif-

fusion coefficient for obvious reasons. Also, note that (6.68) is equivalent to the
expression in Definition 6.28 on page 284.
A class of stochastic processes that are closely related to Brownian motion
with drift and scaling are called the geometric Brownian motion.
Definition 6.31. A stochastic process { X (t)} is said to be a geometric Brownian
motion (GBM) if it is a solution to the stochastic differential equation
dX (t) = X (t) dt + X (t) dB(t), (6.69)
where and > 0 are constant.

Since solving (6.69) with initial condition X (0) = x0 , where x0 > 0, by ap-
plying Itos lemma with function f ( x ) = ln( x ) (see Exercise 6.20 on page 324)
yields
X ( t) = x0 e( 2 )t+B(t)
1 2
, (6.70)
{ X (t)} with X (t) = eY(t) is a geometric Brownian motion if and only if
{Y (t)} is a Brownian motion with drift and scaling.
6.9.2 GBM: Basic Properties
Let us recall (6.27) on page 279

S ( t)
dX (t) = dt + dB(t), where X (t) = ln ,
S (0)
X (t) = t + B(t), (6.71)
for X (0) = ln 1 = 0. Clearly, { X (t) : t 0} is a Brownian motion with drift
parameter and volatility parameter .
For each fixed t, since B(t) N (0, t), applying Property 6.8 on page 282,
we have
X (t) N (t, 2 t).
Fig. 6.3 Geometric Brownian motion with drift parameter = 0.15 and volatility parameter = 0.3
is illustrated using 5,000 randomly selected sample paths over the interval [0, 1]. The current time
is 0 and S(0) = 1. At time t = 1, the frequency distribution of S(1) is shown as a histogram,
which approximates a lognormal distribution. The solid curve shows the plot of the expected value
E ( S( t)) = e(0.15+ 2 (0.3) ) t , where t 0
1 2

S(t)
That is, X (t) = ln S(0) is a normal random variable. Let S(0) = S0 be a posi-
tive number. Note that the nonnegative random variables
S(t) = S0 eX (t) = S0 et+B(t) , t0 (6.72)
define a stochastic process {S(t) : t 0}, which is a geometric Brownian mo-

tion with parameters and (see Figure 6.3). We say that random variable
S(t) has a lognormal distribution, written
S(t) log-normal (t, 2 t). (6.73)
Notice that E (S(t)) and Var(S(t)) are determined by (6.74) and (6.75), re-
spectively (which are not t and 2 t).
Property 6.8, 3 on page 282 indicates
X (t) N (t, 2 t) MX (t) (s) E (esX (t) ) = ets+ 2

1 2 ts2
,
and therefore with s = 1,

S ( t)
= E (eX (t) ) = MX (t) (1) = et+ 2 t ,
1 2
E (6.74)
S0
and with s = 2,
2
S ( t)
= E (e2X (t) ) = MX (t) (2) = e2t+2 t .
2
E (6.75)
S0
That is,
2
S ( t) (+ 12 2 )t S ( t)
= e2 ( + ) t .
2
E =e and E (6.76)
S0 S0
Remark 6.15. Since stock prices are never negative, it is more reasonable to
use geometric Brownian motions to model stock price dynamics than to use
Brownian motions as the latter may take on negative values.
Similar arguments apply to a comparison between the binomial tree model
and random walk model (the latter, in fact, converges to a Brownian motion;
see Theorem 6.3 on page 321).

6.9.3 Relation Between Binomial Tree Model and GBM Model
Recall the Brownian motion with drift and scaling represented by (6.71)
X (t) = t + B(t) (note that X (0) = 0),
which implies that increments have normal distributions:
X (t) X (s) = (t s) + (B(t) B(s)) N ((t s), 2 (t s)).
Recall (6.72) and consider geometric Brownian motion S(t) = S0 eX (t) . We ob-
tain
S ( t) eX (t)
= X (s) = eX (t)X (s) log-normal((t s), 2 (t s)). (6.77)
S(s) e
Given t > 0 and S(0) = S0 , let P be a partition on [0, t] with equal length:
it
P : 0 = t0 < t1 < t2 < < t n = t where ti = , i = 0, 1, . . . n.
n
To establish a binomial tree model, we let

Si1 u with probability p

Si = where i = 1, 2, . . . , n.
Si1 d with probability 1 p,
Given the stock price at current time, S(0) = S0 , the evolution of a stock price
process from time 0 to time t governed by each of the two different models is
illustrated in the table below.
t1 t2 t3 t n 1 t
Binomial tree model forecast S1 S2 S3 S n 1 Sn
GBM model forecast S ( t1 ) S ( t2 ) S ( t3 ) S ( t n 1 ) S ( t )
Again, {S(t), t 0}, is a geometric Brownian motion with parameters (, )
given in (6.72). Our goal is to find a relation between the binomial tree model
and the geometric Brownian motion model when the partition P becomes finer
and finer.
Recall that the parameters u, d, and p in the binomial tree model satisfy the
relations
0 < d < 1 + r < u, 0 < p < 1, ud = 1. (6.78)
To achieve our goal, we are going to look for specific values of these parameters
u, d, and p such that they can be expressed in terms of and as well as satisfy
relations in (6.78). To do so, we take four steps:
Step 1. Let us recall (6.3) on page 261 and write

u if = U with probability p
Yi = Yi ( ) = where i = 1, 2, . . . , n.
d if = D with probability 1 p,
(6.79)
Thus
Si = S0 Y1 Yi , i = 1, 2, . . . , n. (6.80)
Note that Yi s are independent and identically distributed, and straightforward
computation yields
St i
Yi = , i = 1, 2, . . . , n, (6.81)
S t i 1
E (Yi ) = pu + (1 p)d, (6.82)
E ((Yi ) ) = pu + (1 p)d .
2 2 2
(6.83)
Step 2. Notice that (6.77) implies

S ( ti ) t 2t
log-normal , , i = 1, 2, . . . , n,
S ( t i 1 ) n n
and that the formulas in (6.76) imply

S ( ti )
= e(+ 2 ) n , i = 1, 2, . . . , n,
1 2 t
E (6.84)
S ( t i 1 )

S ( ti ) 2
= e2(+ ) n , i = 1, 2, . . . , n.
2 t
E (6.85)
S ( t i 1 )
Step 3. Setting the expectations in (6.82) and (6.84) equal, with corresponding
i, yields28
( + 12 2 ) nt
pu + (1 p)d = e ,
( + 12 2 ) nt
d e
. p=
ud
Setting the expectations in (6.83) and (6.85) equal, with corresponding i,
yields
pu2 + (1 p)d2 = e2(+ ) n .
2 t
Step 4. By solving for parameters p, u, and d from the system of equations

( + 12 2 ) nt
pu + (1 p)d = e
2( +2 ) nt (6.86)
pu2 + (1 p)d2 = e

ud = 1
we can express p, u, and d in terms of , , and n.

It can be shown that for large n,
( + 12 2 ) nt t t
e d n n
p= , u=e , d=e (6.87)
ud
provide an approximation to the solution of the system.
Recall the comment in a solution provided in Example 6.37 on page 305 that
in a risk-neutral world, if the stock pays a known dividend yield q, then
1
+ 2 = r q.
2
This explains why we choose the binomial model parameters as
(r q) t
e d t t
p= , u=e , d=e . (6.88)
ud
By making such a choice of parameters, the geometric Brownian motion can
be simulated by the binomial tree model (see Figure 6.4, which depicts at time
t = 1 a security prices lognormal density, mean, and median along with a
histogram approximation to the density and illustrates that S(1) is skewed
right (see Definition 6.15 on page 266), thus mode < median < mean).
Verification of p, u, and d in (6.87) is left as an exercise for the reader.
Finally, we point out that it is readily understood from (6.80) that the bino-
mial tree approximates a lognormal distribution.
28 The subscript n is dropped from pn , un , and dn to ease the notation.

Fig. 6.4 Lognormal density

of S(1) with initial price
fs(1)(x)
S(0) = $1, drift parameter
RW = 0.1, and volatility 1.0
parameter = 0.4. This
right-skewed density has 0.8
a median (dashed verti-
cal line) and mean (solid
0.6
vertical line) at 1.1 and
1.2, respectively. The his-
togram approximation is 0.4
for 100,000 values ran-
domly drawn from a log- 0.2
normal distribution with
the same drift and volatil- x
ity parameters 0.5 1.0 1.5 2.0 2.5 3.0
6.10 BM as a Limit of Simple Symmetric RW
A random walk consists of a succession of random steps, which means that ei-
ther the direction or size of each step (or both) is chosen at random. To make
a mathematical formalization of this notion, we associate each step with a ran-
dom variable, say Ri . Thus, the dynamics of the walk are governed by the
distribution of these random variables.
Construction of Brownian motion from simple symmetric random walk.
Recall Example 6.27 on page 280. In mathematical language, a stochastic
process {Wi } on (, F, P ) is said to be a one-dimensional random walk (or ran-
dom walk on Z) if
i
Wi = W0 + Rk = Wi1 + Ri , i = 1, 2, . . .
k =1
where { Ri , i = 1, 2, . . . } is a sequence of independent, identically distributed

random variables and W0 a random variable that is independent of each Ri .
A (one-dimensional) random walk is simple if

1 with probability p
W0 = 0 and Ri = i = 1, 2, . . . ,
1 with probability 1 p,
i.e., a walk starting from its mean 0 with unit step size only. A (one-
dimensional) simple random walk is symmetric if p = 12 , i.e., a walk taking each
step equally likely to be in all possible directions (for a one-dimensional sim-
ple symmetric walk, it means that moving to the left is as likely to occur as
moving to the right at each step).
6.10 BM as a Limit of Simple Symmetric RW 321
Example 6.41. For each positive integer n, let

(n) 1 nm 1 with probability 12

Sm =
n i =1
Ri where Ri =
1 with probability 12 ,
i = 1, 2, . . . .
with RW E ( Ri ) = 0 and RW
2 Var( R ) = 1.
i
Apply the central limit theorem to obtain

nm
1 (n) 1 d
Sm =
m

RW nm i=1
Ri nm RW N (0, 1) as n ,
and observe

(n) 1 nm
d
Sm = m
RW

nm
Ri nmRW N (0, m) as n .
i =1
Notice that different scalings lead to different limits and B(m) N (0, m).
This observation makes the next theorem plausible.

For each
$ positive integer
% n, we define a continuous-time stochastic process
(n)
W (n) = Wt : t 0 by
n t
(n) 1
Wt = Ri , (6.89)
n i =1
whereby convention nt is the greatest integer less than or equal to nt, and
random variables Ri are defined in Example 6.41.
In words, (6.89) defines the rescaled random walk with steps of size 1n
taken every 1
n time units (i.e., steps are taken at time i/n, i = 1, 2, . . . , n t).
Theorem 6.3. Standard Brownian motion can be approximated by the rescaled ran-
dom walk:
n t
d 1
B(t) = lim Ri , t 0.
n n i =1
Intuitively, the theorem implies that Brownian motion has the microscopic
structure that may emerge from a random walk. Although Example 6.41 might
seem intuitively reasonable in suggesting a proof of this theorem by applying
the (ordinary version of) central limit theorem, a proper proof of the theorem
involves sophisticated technicality at the mathematical level beyond the scope
of this book (it requires functional central limit theorem). We refer the reader
to the literature (e.g., Billingsley [2]).
presented in this chapter, e.g., [1, 5, 6, 7, 8, 10, 12, 13, 14, 16, 18, 19, 21, 23, 24,
26, 27, 28].
6.11 Exercises
6.1. Consider an oversimplified stock price behavior as described by a two-

period binomial tree. In each period the stock price either goes up by a factor u
with probability p or goes down by a factor d with probability 1 p. Identify
the corresponding probability space (, F, P ).
6.2. Continue from Exercise 6.1.

Let 1 = UU (i.e., the event that the stock price goes up in both periods).
Construct a sub--algebra F such that {1 } F 2 .

Construct a filtration {Ft F, t = t0 , t1 , t2 } such that as time t increases, Ft
reveals more information about the evolution of the stock price.
6.4. Consider an oversimplified stock price behavior as described by a three-

period binomial tree. In each period the stock price either goes up by a factor u
with probability p or goes down by a factor d with probability 1 p. Identify
the corresponding sample space (in the notation for the probability space
(, F, P )). How many elements are there in the -algebra F = 2 ? (Hint: Each
simple event corresponds to a path.)
6.5. Continue from Exercise 6.4 on page 322.

Given the corresponding sample space
= {1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 }, where
1 = UUU, 2 = UUD, 3 = UDU, 4 = UDD,
5 = DUU, 6 = DUD, 7 = DDU, 8 = DDD,
Construct a sub--algebra F such that
{ 1 , 2 , 3 , 4 } F 2 .

Construct a filtration {Ft F, t = t0 , t1 , t2 , t3 } such that as time t increases,
Ft reveals more information about the evolution of the stock price.
6.7. Let B = {B(t)} be standard Brownian motion. What is the probability that
B(1) lies between 1 and 1?
6.8. Let X = { Xt } and Y = {Yt } be two processes. Verify the following identity:
( Xt Yt ) = Xt Yt + Yt Xt + Xt Yt
6.11 Exercises 323
where Xt , Yt , and ( Xt Yt ) are defined as the corresponding increments over

time interval [t, t + t]. (Hint: see (6.38) on page 292.)
6.9. Let B = {B(t)} be standard Brownian motion. Describe your visualization

of B. Is it different from that of B?
6.10. Use the definition of covariation to derive
dB(t) dt = 0.
6.11. Suppose that a stock price is modeled by a process S = {S(t)} where

2 +2.3B( t )
S(t) = e0.7t .
What is the expected growth rate of the stock at any given time t?
6.12. Let {S(t)} be governed by the s.d.e.
dS(t) = S(t)dt + S(t)dB(t) ,
where and > 0 are constants. Is {S(t)} an Ito diffusion process?
6.13. Recall s.d.e. (6.58) on page 304:

1 2
dY = f t + f x + f xx dt + f x dB.
2
Compute E (dY |Ft ) and Var(dY |Ft ), where {Ft } is the Brownian filtration
described in Definition 6.30. Indicate properties of conditional expectation that
you applied.
6.14. Let {S(t)} be governed by the s.d.e.
dS(t) = S(t)dt + S(t)dB(t) ,
where and > 0 are constants. Let Y (t) = ert S(t). Use the Ito product rule
(see (6.40) on page 292) to compute dY (t).
6.15. Continue from the last exercise. Is {Y (t)} defined in Exercise 6.14 a mar-
tingale under a risk-neutral probability measure?
2t
6.16. Compute 0 B(s) dB(s) first, then express the s.i.e. in the form of s.d.e.
6.17. Is it true that X = { X (t), t 0} is a Brownian motion process if and only if

Y = {Y (t), t 0}, where Y (t) = eX (t) , is a geometric Brownian motion process?
6.18. In continuous-time financial mathematics, in what situations is geometric

Brownian motion useful?
6.19. Rewrite s.d.e. (6.27) on page 279 into
dS(t) = S(t) dt + S(t) dBt .
Consider a time period of length t. Compute

the ratio of the per-period stan-
Var(S (t)
dard deviation to the per-period drift, i.e., E (S (t))
, and interpret your re-
sult.
6.20. Find a solution to the s.d.e. dX (t) = X (t)dt + X (t)dB(t) by applying
Itos lemma with f ( x ) = ln x.
6.21. Verify that p, u, and d given by (6.88) on page 319
(r q) nt t t
e d n n
p= , u=e d=e
ud
are an approximation to the solution of system (6.86):

( + 12 2 ) nt
pu + (1 p)d = e
2( +2 ) nt
pu2 + (1 p)d2 = e

ud = 1
6.22. Given a standard Brownian motion B(t), show that each of following
stochastic processes is also a standard Brownian motion:
a) X (t) = 1
c
B(c t) for all constants c > 0.
b) Y (t) = B(t + c) B(c) for all constants c > 0.
6.23. Let {B(t)} be a standard Brownian motion. Let t1 , t2 , . . . , tn (0, ) with
0 < t1 < t2 < < tn . Show that the random vector (or multivariate random
variable) (B(t1 ), B(t2 ), . . . , B(tn )) has a multivariate normal distribution for
any fixed choice of n time points 0 < t1 < t2 < < tn , n 1.
6.24. Continue from the last exercise. Show that the joint probability density
function of (B(t1 ), B(t2 ), . . . , B(tn )) is

n 1 ( x j +1 x j )
2
x12
exp 2 t + j=1 t t
1
1 j +1 j
f ( x1 , . . . , x n ) = , ,
( 2 ) t1 ( t2 t1 ) ( t n t n 1 )
k
where < x j < for j = 1, . . . , n.

6.11 Exercises 325
6.25. Let {B(t)} be a standard Brownian motion. Compute E (B(t)) for t 0

and Cov(B(s), B(t)) for s, t 0. Is Brownian motion a white noise?
6.26. Let { Zn : n 1} be a sequence of independent identically distributed ran-

dom variables with mean = 0 and finite variance 2 . Let
Sn = Z1 + Z2 + + Zn and Xn = S2n n2 .
Show that { Xn } is a martingale with respect to the natural filtration of the

sequence (Sn : n 1).
6.27. Let W = {Wi } be a random walk defined in Example 6.27. Show that W
is not a white noise.
6.28. Prove (6.51) on page 299:

1 t
[ X ]t = 2 (s) ds,
0
where X satisfies (6.50) on page 298 and (t) is continuous on [0, ).
6.29. Show that B2t [B]t is a martingale with respect to the filtration gener-
ated by the Brownian motion itself. That is, B is adapted to its natural filtration
{Ft }, where Ft = ({Bs , s t}).
6.30. Show that as positive integer n , the sequence of random variables

(n)
{Wt }, defined by (6.89) on page 321, converges in distribution to a normal
random variable in N (0, t), where t > 0 is an integer.
6.31. Let X = { X (t)} be an Ito process represented by s.d.e. (6.56)
dX (t) = ( X (t), t) dt + ( X (t), t) dB(t), 0 t T.
Show that if we define a process Y = {Y (t)} by Y (t) = f ( X (t)), 0 t T, then

Itos formula (see (6.57) on page 304) has a convenient form:
1
dY = f ( X )dX + f ( X )(dX )2 ,
2
where X = X (t). Clearly, this form is easier to remember than the form in (6.57)
on page 304 because it bears greater similarity to the Taylor expansion.
6.32. Let X = ( X1 , X2 ), where X1 and X2 are Ito processes governed respec-

tively by the s.d.e.s
dXi = i dt + i dBi , i = 1, 2.
Two-dimensional Itos lemma states that if B1 and B2 are standard Brownian
motion processes with correlation 12 , then Y = f ( X ), where f : R2 R is twice
continuously differentiable, is an Ito process satisfying the s.d.e.
d f ( X ) = f1 ( X )dX1 + f2 ( X )dX2
1
+ f11 ( X )dX12 + 2 f12 ( X )dX1 dX2 + f22 ( X )dX22 , (6.90)
2
f 2 f
where f i = Xi and f ij = Xi X j , i, j = 1, 2.29
Show that (6.90) is equivalent to (6.57) if X2 = t (i.e., 2 = 1 and 2 = 0).
6.33. Prove the Ito product rule. (Hint: prove (6.39) on page 292 by applying
two-dimensional Itos lemma given in Exercise 6.32).
6.34. An investment in a foreign asset carries exchange risk. The model under
our consideration is introduced by Briys and Solnik [4] in study of hedging
such risk.
Let V (t) be the local (domestic) currency value of a foreign asset at time t.
Let S(t) be the exchange rate at time t expressed as the local currency value
of one unit of foreign currency (e.g., 1.11 USD/Euro). The model assumes that
both {V (t)} and {S(t)} are geometric Brownian motion processes:
dV
= V dt + V dBV
V
dS
= S dt + S dBS ,
S
where two standard Brownian motion processes BV and BS have correla-
tion VS .
Let V = VS, the value of the foreign investment expressed in domestic cur-

rency. Compute dV V and interpret your answer. (Hint: apply Ito product rule.)
References
[1] Baxter, M., Rennie, A.: Financial Calculus: AnIntroduction to Derivative

Pricing. Cambridge University Press, Cambridge (1996)
[2] Billingsley, P.: Weak convergence of measures. In: CBMS-NSF Regional
Conference Series in Applied Mathematics (1971)
[3] Billingsley, P.: Probability and Measures. Wiley, New York (1995) or (2012)
[4] Briys, E., Solnik, B.: Optimal currency hedge ratios and interest rate risk.
J. Int. Money Finance 11, 431446 (1992)
[5] Chung, K.L.: A Course in Probability. Academic, New York (1974)
[6] Clark, J.M.C.: The representation of functionals of Brownian motion by
stochastic integrals. Ann. Math. Stat. 41(4), 1282(1970)
[7] Dellacherie, C., Meyer, P.-A.: Probabilities and Potential C. North-
Holland, Amsterdam (1988)
29More precisely speaking, we assume that B1 and B2 are defined on the same filtered probability
space { , F, {F t }, P } and adapted to the filtration {F t }.
References 327
[8] Durrett, R.: Probability: Theory and Examples, 4th edn. Cambridge Uni-
versity Press, Cambridge (2010)
[9] Durrett, R.: Brownian Motion and Martingales in Analysis. Wadsworth
Advanced Books and Software, Belmont (1984)
[10] Epps, T.: Pricing Derivative Securities. World Scientific, Singapore (2007)
[11] Fama, E.: Efficient capital markets: a review of theory and empirical work.
J. Finance 25(2), 383 (1969)
[12] Girsanov, I.V.: On transforming a certain class of stochastic processes by
absolutely continuous substitution of measures. Theory Probab. Appl. 5,
285(1960)
[13] Harrison, J.M., Pliska, S.R.: Martingales and stochastic integrals in the the-
ory of continuous trading. Stoch. Process. Appl. 11(3), 215(1981)
[14] Hull, J.: Options, Futures, and Other Derivatives, 7th edn. Pearson Pren-
[15] Hunt, P., Kennedy, J.: Financial Derivatives in Theory and Practice. Wiley
Series in Probability and Statistics. Wiley, New York (2004)
[16] Ito, K.: Multiple wiener integral. J. Math. Soc. Japan, 3, 157169 (1951)
[17] Karatzas, I., Shreve, S.: Brownian Motion and Stochastic Calculus.
Springer, New York (1991)
[18] Korn, R., Korn E.: Option Pricing and Portfolio Optimization. American
[19] Mackean, H.: Stochastic Integrals. Academic, New York/London (1969)
[20] Malz, A.: A Simple and Reliable Way to Compute Option-Based Risk-
Neutral Distributions. Federal Reserve Board of New York Staff Reports
(2014)
[21] Mikosch, T.: Elementary Stochastic Calculus with Finance in View. World-
Scientific, Singapore (1998)
[22] Morters, P., Peres, Y.: Brownian Motion. Cambridge University Press,
Cambridge (2010)
[23] Musiela, M., Rutkowsk, M.: Martingale Methods in Financial Modelling.
[25] Paley, R., Wiener, N., Zygmund, A. Note on some random functions.
Math. Z. 37, 647668 (1993)
[26] Roman, S.: Introduction to the Mathematics of Finance, 1st edn. Springer,
New York (2004)
[27] Shreve, S.: Stochastic Calculus for Finance II: Continuous-Time Models.
tives: a Student Introduction. Cambridge University Press, Cambridge
(1995)
Chapter 7
Derivatives: Forwards, Futures, Swaps, and Options
To introduce major concepts and ideas about derivatives in a simple and

concise fashion, we make some assumptions1 throughout this chapter so that
we can concentrate on the task at hand and make the main ideas easier to
understand.
Unless otherwise stated, we impose the following assumptions:
There are no transaction costs.
There are no taxes.
The risk-free interest rate is the same for borrowing and lending.
There are no restrictions (e.g., margin requirements) to the short seller.
There is sufficient liquidity in all markets (e.g., stocks, bonds, derivatives,
foreign exchange, and so on).
7.1 Derivative Securities: An Overview
In financial accounting, an asset represents value of ownership that can be con-

verted into cash.
A financial asset is an (intangible) asset whose value is derived from a con-
tractual claim, such as bank deposits, bonds, and stocks. Unlike real-estate
properties and commodities (which are tangible, physical assets), financial
assets do not necessarily have physical worth.
A security is a tradable financial asset. Thus, the set of all securities is a subset
of the set of all assets as any security is an asset by definition.
1 These assumptions are often either unreasonable or unrealistic from a practical point of view (e.g., a
time t is assumed to be continuous, but in reality it is discrete).

330 7 Derivatives: Forwards, Futures, Swaps, and Options
Securities are represented by certificates (this explains the name securi-

ties) which may be either in paper form, in electronic form, or in book-entry
form.
Example 7.1. Cash is an asset but not a security.

Securities may be classified into three categories:
1. debt securities or debts (e.g., bonds)

2. equity securities or equities (e.g., common stocks)
3. derivative securities or derivatives (e.g., forwards, futures, options, and
swaps).
Derivatives are securities in the form of contracts between two parties, the
buyer and the seller. Since these contracts are either for contingent claims or for
forward commitments, derivatives can be classified into contingent claims and
forward commitments (or noncontingent claims) according to the type of contracts.
A contingent claim is a contract which gives the buyer the right, but not the
obligation, to buy or sell a security, called underlying security or underlier, at a
specified price, called strike price, on or before a specified date, called expiration
date. Examples of contingent claims are options.2
A forward commitment is a contract by which the buyer and seller have the
obligation to buy or to sell (to deliver) an underlying security at a predeter-
mined price in the future, called the delivery date. Examples of forward com-
mitments are forwards and futures.
Derivative contracts can be created on and traded in some exchanges such
as the Chicago Board of Trade (CBOT), which is the worlds oldest futures
and options exchange, or on OTC markets, which are less transparent than
exchanges. For example, a futures contract is considered to be a standardized
version of a forward commitment and therefore traded on CBOT, whereas a
forward contract can be customized to any commodity, amount, and delivery
date and, therefore, is nonstandardized and traded on OTC markets.
Remark 7.1. One might attempt to classify securities in terms of underlying

securities and derivative securities. If you do so, think again, for underlying
securities are often real assets such as gold, oil, and metals or financial assets
such as bonds, stocks, and currencies. But underlying securities can also be
derivatives on other underliers; for example, options on futures which are on
metals are derivatives on derivatives.
2 Later we will show that the definition of options coincides with the definition of contingent claim in
the financial dictionary: a claim that can be made only if one or more specified outcomes occur.
7.1 Derivative Securities: An Overview 331
Keep in mind that derivative contracts can be written on real assets such
as commodities3 as well as on financial assets such as bonds, stocks, currencies,
and other derivatives. Derivative contracts written on precious metals or agri-
cultural products are called commodity derivatives, whereas those written on
securities are called financial derivatives.

Although we will primarily focus on stock options and have not yet pro-
vided all the details in the financial jargon used in this section, the next exam-
ple will help the reader to understand Remark 7.1.
Example 7.2. The premium of an option on futures tracks the price of its under-
lying futures contract which, in turn, tracks the price of the underlying cash. In
fact, regardless of the underlying security, it is a fairly safe bet that the futures
price will generally converge to the spot price of the underlying security as
the delivery month of a futures contract approaches because otherwise there
would be arbitrage opportunities.
For instance, the July copper option tracks the July copper futures contract.
The March S&P 500 index option follows the March S&P 500 index futures.
Furthermore, the prices of these futures converge to the spot prices of copper
and S&P 500 index with high probability at least in their delivery months.
7.1.2 Basic Functions of Derivatives
Derivatives may serve three basic functions, which are price discovery, specu-
lative activity, and hedging activity:
Price discovery is a process involving buyers and sellers arriving at a transac-
tion price for a given product with given quality and quantity at a given time
in a given location. Although they are interrelated, price discovery and price
valuation are different concepts. The former is a mechanism, whereas the lat-
ter is a determination. Price discovery plays an important role in economic
decision-making that involves either entrepreneurs or policy makers.
Speculators are those who take calculated risks in the hope of making large
short-term profits. By definition, speculators are usually not interested in hold-
ing possession of the underlying securities. They are typically sophisticated
investors with expertise in the markets in which they are trading who usually
use highly leveraged investments such as futures and options.
Hedgers are those who take steps to reduce the risk of an investment by mak-
ing an offsetting investment. By definition, hedgers do not usually seek a profit
but rather seek to stabilize the performance of their portfolios or the revenues
3A commodity is a raw material used in commerce or primary agricultural product that can be bought
and sold such as copper, silver, crude oil, natural gas, wheat, beef cattle, and coffee.
or costs of their business operations. Their gains or losses are usually offset to
some degree by a corresponding loss or gain in the market for the underlying
securities.
Example 7.3. Price discovery begins with market price information. Imagine
how difficult business decision-making would be in the extreme scenario in
which no one knows at what price competitors have sold a given product.
When more varieties of a product are sold through more venues, more mar-
ket price information becomes available. Derivatives serve this purpose by
providing a wider variety of assets through many venues, including cash
settlements.

Example 7.4. Speculators provide very important market information. Arbi-

trage makes prices converge to the fair price as a riskless way of making a
profit would only be a transient opportunity. In other words, free-lunch opp-
ortunities will not last long as other people may seek and grab them quickly.

Example 7.5. If you are in a manufacturing business and sell products to

foreign countries, you may have specific business ideas in mind. But foreign
exchange markets may be in turmoil and highly volatile. With the help from
derivatives, you can minimize or even eliminate the unwanted volatility.

7.1.3 Characteristics of Derivative Valuation
Security valuation is the determination of the fair price of a security.

If we focus only on a straight bond (a bond with no embedded options4 ), the
valuation is based on discounting its expected cash flows at the appropriate
discount rate.5
If we focus only on a single (dividend-paying) stock, valuation models
include the dividend discount and discounted cash flow model.6
To summarize in loose terms, the valuation methods above are based on
forecasting future cash flows, growth rates, and risks.
A derivative valuation, by contrast, involves no forecasting and is directly
linked to the price of the underlier in the sense that, in relative terms, the higher
the volatility of the price movement of the underlier, the higher the value of the
derivative.
Derivative valuation models are derived mostly by using replication or
hedging arguments under the Law of One Price (see Section 7.1.4).
4 A callable bond is an example of a bond with an embedded call option.

5 Section 2.10
6 http://www.investopedia.com/terms/d/ddm.asp
Example 7.6. (Price by Hedging) Consider a derivative7 with expiration T on

a stock whose price is modeled by a one-step binomial tree (where the prob-
ability the stock price will go up is p and the probability it will go down is
1 p). Let C (t) and S(t) denote the prices of the derivative and stock at time
t [0, T ], respectively. What is the fair value of C (0) if it is priced by hedging
based on the following assumptions:
(a) S(0) = $50,

(b) Only two possible states at time T: either S( T ) = $60 or S( T ) = $40,
(c) C ( T ) max(S( T ) 50, 0),
(d) The risk-free interest rate r = 0.
Solution. Note that an equivalence between the no-arbitrage condition and the
existence of the risk-neutral probability measure implies that p = 12 (which is
the solution to the equation 50 = 60p + 40(1 p)).
Suppose that we sold one derivative contract and need to buy shares of
the underlying stock to hedge away any portfolio risk, where the portfolio
consists of a long position in shares of the stock and a short position in one
derivative contract. Let ( T ) denote the value of the portfolio at time T. Since
.
$10 if S( T ) = $60
C ( T ) max{S( T ) 50, 0} =
0 if S( T ) = $40,
we obtain
.
60 10 if S( T ) = 60
( T ) =
40 if S( T ) = 40,
which should be a constant if the risk is completely hedged away. It follows

from the equation 60 10 = 40 that = 12 .
Thus, 50 C (0) = (0) = E (( T )|(0)) = ( T ) = 20 with = 12 implies
that C (0) = $5.

7 One can think of a call option (see Definition 7.8). It is defined in a later section, but this example
should be readily understandable.
7.1.4 No-Arbitrage Principle and Law of One Price
The key principle of mathematical finance is the principle of no arbitrage.

We consider a portfolio8 on time interval [0, T ] and denote by (t) the
value of portfolio at time t. There are different versions9 of the notion of
arbitrage. We will use the one below.
Definition 7.1. A portfolio is said to be an arbitrage portfolio, or simply an
arbitrage, if it satisfies either of the two sets of conditions:
1. (0) = 0, P (( T ) 0) = 1, and P (( T ) > 0) > 0;
2. (0) < 0 and P (( T ) 0) = 1.
In words, the first set of conditions describes a portfolio that is at no cost, never
loses money, and sometimes makes money. The second set of conditions des-
cribes a portfolio that starts from a debt and guarantees to finish with a non-
negative amount of money.
Example 7.7. A no-arbitrage portfolio must satisfy the property that
( T ) 0 implies (t) 0 for t < T,
for otherwise, t0 < T such that (t0 ) < 0 and ( T ) 0 with probability 1,
contradicting to the no-arbitrage assumption!

Definition 7.2. An arbitrage opportunity in a market is an opportunity to con-

struct an arbitrage portfolio consisting of securities in the market.
An arbitrage-free market is a market providing no arbitrage opportunities.
Let A and B be two portfolios on the same time interval.

The law of one price is said to hold if A ( T ) = B ( T ) implies A (t) = B (t)
for t < T.
Equivalently, the law of one price (LOP) is said to hold if there do not exist
two portfolios, say A and B , such that A ( T ) = B ( T ) but A (t) = B (t)
for some t < T.
In other words, the law of one price simply says that two portfolios that will pro-
duce exactly the same cash flows in the future must have the same value to begin with.
8 It is understood that by a portfolio we mean a self-financing portfolio (i.e., no money is added to or
withdrawn from the portfolio after time 0).

9 One of those versions is given as follows:
A portfolio is said to be a statistical arbitrage portfolio, or simply a statistical arbitrage if it satisfies

either of the two sets of conditions:
( a) (0) = 0 and E ( ( T ) > 0) > 0, ( b ) (0) < 0 and E ( ( T ) 0) > 0.

The next example shows that no arbitrage is a sufficient condition of the law
of one price. Equivalently, the law of one price is a necessary condition of no
arbitrage.
Example 7.8. Show that the law of one price holds if there are no arbitrage
portfolios.
Proof. Suppose otherwise, i.e., there exist two portfolios A and B and some
t0 < T such that A ( T ) = B ( T ) and A (t0 ) = B (t0 ).
Without loss of generality, say A (t0 ) < B (t0 ). Then we construct a port-
folio by long portfolio A and short portfolio B, written = A B , which
leads to
(t0 ) = A (t0 ) B (t0 ) < 0, and

( T ) = A ( T ) B ( T ) = 0 with probability 1.
Thus, by Definition 7.1, 2, is an arbitrage! This contradicts the no-arbitrage

assumption.

Example 7.9. Let A and B be two portfolios on [0, T ] with
A (0) = B (0) and A ( T ) < B ( T ).
Construct an arbitrage portfolio.
Solution. Consider the portfolio defined by B A . We have
(0) = B (0) A (0) = 0,

( T ) = B ( T ) A ( T ) > 0 with probability 1.
Thus, is an arbitrage by Definition 7.1, 1.

If the given information becomes
A (0) = B (0) and A ( T ) > B ( T ),
is defined by A B an arbitrage?

Remark 7.2. (Law of One Price) The approaches to constructing an arbitrage
demonstrated in Examples 7.8 and 7.9 provide a general guideline which will
be applied throughout this chapter. Combining these two examples, we con-
clude the following:
If each of two investment portfolios produces a deterministic stream of cash
flows as indicated below,
t=0 t=T
cash flows from A A (0) A (T )
cash flows from B B (0) B (T )
then
A ( T ) = B ( T ) A (0) = B (0) by the law of one price,

A ( T ) > B ( T ) and A (0) = B (0) = A B is an arbitrage,
A ( T ) < B ( T ) and A (0) = B (0) = B A is an arbitrage.

Remark 7.3.
1. Recall that under the continuous-time framework, the investors are allowed
to trade up to time T < . Let S(t) = ( S0 (t), S1 (t), S2 (t), . . . , S N (t) ) be a
market, where price process {Si (t)} is defined on filtered probability space
(, F, P, {Ft F : 0 t T }) with FT = F for each i = 0, 1, 2, . . . , N.
Suppose that positions of a portfolio P respectively on securities
S0 ( t ) , S1 ( t ) , S2 ( t ) , . . . , S N ( t )
at time t are expressed by n0 (t), n1 (t), n2 (t), . . . , n N (t).

In order to understand the law of one price in a more precise fashion, let us
define a self-financing portfolio trading strategy (or simply a trading strategy) to
be a vector-valued function of t denoted by n(t) and expressed by
n(t) = (n0 (t), n1 (t), n2 (t), . . . , n N (t)).
Also, let us denote by Pn (t) the process of the value of a portfolio P con-
structed by trading strategy n. Then
N
Pn (t) = n(t) S(t) = n i ( t ) Si ( t ).
i =0
The law of one price is said to hold if there do not exist two trading strategies,

say n and n such that Pn ( T ) = Pn ( T ) but Pn (t) = Pn (t) for some t < T.
An interpretation of this more precise version of the law of one price along
with Example 7.8 is that arbitrage opportunities exist when the prices of similar
assets are set at different levels.
2. A relationship between no arbitrage and the equivalent martingale measure
is given by the First Fundamental Theorem of Asset Pricing, which states: The
market is arbitrage free if and only if there exists an equivalent martingale measure.
(See page 314.)
3. The non-arbitrage assumption is a reasonable one for financial theory:
In the real world, arbitrage opportunities do exist, but they are only transient
because, once more investors jump in to share the free lunch, soon the free
lunch will be over. The market price will be adjusted and move from an old
equilibrium to a new one.
7.2 Forwards 337
4. Although the no arbitrage assumption is reasonable theoretically, there are

many different arbitrage strategies which are practically efficient and are
often applied by quantitative models.

7.2 Forwards
If today, denoted by time 0, you want to secure a transaction on a future date,

denoted by time T, rather than waiting with no certainty of getting the market
price, you may consider locking in the terms of the transaction by entering into
a forward contract today.
Forward contracts offer users the ability to lock in a purchase or sale price
without incurring any direct cost. In fact, forward contracts are the founda-
tion of all derivatives as futures contracts are standardized forward contracts,
swaps are series of forward contracts, and options are a variation of forward
contracts.
Definition 7.3. A forward contract is an OTC-traded agreement between two

parties to buy or sell a specified quantity of a specified asset at a specified
future time at a price agreed today.
The forward price is the agreed unit price of an asset in a forward contract.
The long position is the position that the buyer enters in a forward contract.
The short position is the position that the seller enters in a forward contract.
The delivery date or expiry is the specified future time in a forward contract.
The underlier is the asset to be delivered on the expiry in a forward contract.
The contract size is the specified quantity in a forward contract.
There is no payment by either party when the contract is first entered into.10
Thus the value of a forward contract at the time the contract is entered into
is zero. The delivery date or expiry is also called the exercise date or maturity
or expiration date, the time at which the asset changes hands. The seller is also
called the writer of the contract. The forward price is also called the exercise
price.
The spot market or cash market or physical market is a financial market where
assets are traded for cash and immediately delivered on spot (e.g., the stock
10 In old times, people would most often shake hands when agreeing on deals.
market, commodity market, and foreign exchange market), whereas a deriva-

tive market (e.g., forward market) is a market in which the delivery of the un-
derlier asset is due at a later date(s). The spot market price is also called the
spot price.
Since forward contracts are not standardized, the payments can be cus-
tomized to fit a given situation. The delivery price is a price negotiated at the
time the contract is entered into, which may or may not coincide with the for-
ward price (see Example 7.13).
Forward contracts offer users the ability to lock in a purchase or sale price
without incurring any direct cost.
Example 7.10. It is a well-known fact that grain prices may swing substantially
between highs and lows.
In order to secure a smooth wheat supply, a flour mill A enters into a for-
ward contract with a farmer B on June 1 to buy 100 (metric) tons of wheat at
$222 per ton on September 30.
To become familiar with the terminologies in Definition 7.3, we identify that
in this contract, the buyer is A, the seller is B, the underlying asset is wheat,
the forward price is $222, the expiration date is September 30, the contract size
is 100, and the delivery price is $222 as well (which happens to be the same as
the forward price).
On June 1, A and B sign the contract and shake hands. No money changes
hands. However, on September 30, A will pay $22,200, irrespective of the price
of wheat in the spot market, and B will deliver 100 tons of wheat to the flour
mill.
Both parties A and B are bound by the contract and have to honor their
commitments.

Let
t = 0 be the time the forward contract is entered into (e.g., June 1),
t = T be the expiration of the forward contract (e.g., September 30),
S(t) be the spot price of the underlier in the forward contract at time t.
We denote by FT (0) the forward price specified in a forward contract initi-

ated at time 0 and expiring on time T (e.g., FT (0) = $222).
Example 7.11. We continue from the last example.

The wheat spot price fluctuations either naturally or artificially become an
interplay between supply and demand. Keep in mind that the spot wheat price
is a stochastic process. Thus the forward price specified in the forward contract
on wheat is a stochastic process as well.
Suppose that starting from July 10, massive droughts (or floods) negatively
affect the supply of wheat, which in turn triggers the spot price of wheat to
surge (say, to $322). Imagine that another flour mill X enters into a forward
7.2 Forwards 339
contract on July 10 to buy 100 tons of wheat on September 30. Would X be able
to negotiate with any counterparty to get the same forward price, $222? Of
course not!
If we denote July 10 by time t1 , in this new contract, denoted by FT (t1 ), the
forward price FT (t1 ) would be much higher than FT (0).

Notice that two forward contracts from the last two examples have the same
terms except the forward price and the initial time of the contract.
To emphasize the fact that forward prices are functions of time, we denote
by FT (t) the forward price.
Example 7.12. We continue from Example 7.10.

In order to understand the profit and loss (P&L) from the deal in terms of a
market value, we argue that if, on September 30, the spot price of wheat were
$232, A could sell 100 tons of wheat immediately and profit $1000 (i.e., ($232 -
$222)100). However, if on September 30, the spot price of wheat were $202,
A could sell 100 tons of wheat immediately and lose $2000. That is,
the payoff for the buyer = S( T ) FT (0),

the payoff for the seller = FT (0) S( T ).

For the convenience of conversation,
we define the terminal payoff of a forward contract to be the payoff from a
long forward contract, that is,
forward payoff = long forward payoff = S( T ) FT (0). (7.1)
The payoff from a short forward contract is the negative value given by (7.1):
short forward payoff = FT (0) S( T ) = forward payoff.
Definition 7.4. A forward payoff diagram is a graph of the terminal payoff from
long position of a forward contract as a function of the underlier price at T.
In short, a forward payoff diagram is a graph of long forward payoff

against S( T ).
If we simplify notation by letting K = FT (0), x = S( T ), and y = x K, then
a forward payoff diagram is a graph on xy-plane, which is a straight line that
can be obtained by translating the graph of y = x to the right by K units.
In a parallel way, one can define and visualize a payoff diagram from the
short position of a forward contract. It is the graph of the straight line y =
x + K, where x = S( T ) and K = FT (0) on xy-plane and can be obtained by
translating the line y = x to the right by K units (see diagrams below).
Example 7.13. (Delivery Price of a Forward) Once again, we continue from

Example 7.10.
Now, keep in mind that forward contracts are not standardized. This means
that payments can be made in many different ways. Suppose that there were
an upfront payment of $2000 (e.g., a down payment) from party A to party B
at the time when the contract is entered into; the delivery price would become
$202 (different from the forward price).

7.2.2 Forwards on Assets Paying a Continuous Cash Dividend
The spot-forward parity (see Section 7.2.3) for underliers with continuously
paid cash dividends is useful when the underliers are stock indexes containing
many stocks,11 since such an index can be modeled as the dividend being paid
continuously at a rate that is proportional to the level of the index.
To obtain such a spot-forward parity in the next section, let us now recall
(6.29) on page 281 from Section 6.4.3: D(t) = S(t) q dt, where S(t) is the spot
price of the underlier of a forward contract and q is the (annualized) dividend
yield of the underlier. In the rest of this chapter, we will use the familiar results
on page 281:
The dividend reinvestment yields eqT 1 more units at time T from 1 unit
of the underlier at time 0,
The dividend reinvestment yields 1 eqT more units at time T from eqT
units of the underlier at time 0,
and the following assumptions: Unless stated otherwise,
1. By stating that an asset with annual dividend yield q, we mean that an asset
pays a constant, continuous proportional annual dividend yield rate q that is con-
tinuously reinvested to buy more units of the asset and that holding the asset neither
incurs cost of carry nor provides any other convenience yield.
2. r is the risk-free interest rate compounded continuously.
11
We leave out of the theoretical arguments such as whether a no-arbitrage condition can be verified
when the underlier is the S&P 500 Index.
7.2 Forwards 341
7.2.3 Forward Price Formula and the Spot-Forward Parity
We note that the assumption of cost of carry12 and convenience yield13 other
than the dividend yield in the next theorem is not a reasonable one.
Theorem 7.1. Suppose that a forward contract entered into at time 0 with expiration
T is on an asset with annual dividend yield q (see page 341) and that r is the risk-free
interest rate, the fair forward price is given by
FT (0) = S(0) e(rq) T , (7.2)
where S(0) is the price of the asset at the time 0.
In words, (7.2) says that under an arbitrage-free assumption, buying the for-
ward contract and taking delivery is equivalent to buying the underlying asset
from its spot market today and holding in the sense that the cost of both strate-
gies must have the same present value.14
Proof.
Case 1. If FT (0) < S(0) e(rq) T , we construct two portfolios A and B, denoted
by A and B , respectively, with positions established at time 0 indicated
below:
A : short 1 forward.
B : short eqT units of the asset to long bond15 with S(0)eqT at rate r.
We denote by A (t) the value of portfolio A and B (t) the value of portfolio
B at time t. Then the initial and terminal values of two portfolios are
A (0) = 0 (no initial cost for a forward contract),

B (0) = 0 (the proceed from shorting is used to buy the bond
immediately),
A ( T ) = FT (0) (FT (0) is the forward price, the cash flow of A at T),
(r q ) T
B ( T ) = S (0) e (the principle S(0)eqT earns interest at r),
12 The cost of carry is the cost incurred by holding the underlying asset such as storage costs or insur-
ance as well as other incidental costs.

13 For example, being able to take advantage of shortages of the underlying asset.
14 Note that in the equality F (0)erT = S (0) eqT , the LHS is the present value of F (0), and the RHS
T T
is the present value of S( T ) because eqT shares at time 0 grow to 1 share at time T and, during the
same time period, the stock value changes from S(0) to S( T ).
15 The bond here is a zero-coupon bond. For our purpose, the position of long 1 par zero maturing
at time T can be interpreted as lend $S(0)eqT at interest rate r. Similarly, a position of short zero
here would be interpreted as borrow $S(0)eqT at interest rate r. Using the bond terminologies will
provide us convenience (e.g., for letter expression of long or short position) later.
which are summarized in the table below.

At time t At time T
Cash flows from A 0 FT (0)
Cash flows from B 0 S (0 ) e (r q ) T
An arbitrage portfolio = B A (note: A has position: long 1 for-
ward) has been established (see Remark 7.2). This contradicts the no-arbitrage
assumption.
Case 2. If FT (0) > S(0) e(rq) T , using the table in case 1, let = A B , then
is an arbitrage portfolio by Definition 7.1, 1 (for , at time 0, consisting of
the positions: short 1 forward, short bond with S(0)eqT at rate r to long eqT
units of the asset).
Thus, under the assumption of no arbitrage, (7.2) holds.

For a given t [0, T ], if we replace 0 by t in (7.2), then we establish a relation
between the underlier spot price S(t) and the forward price FT (t). This rela-
tion is referred to as the spot-forward parity and shows how a forward can be
replicated.
Corollary 7.1. The spot-forward parity is given by
FT (t) = S(t) e(rq)( T t) , t [0, T ]. (7.3)
Example 7.14. Suppose that the current spot price of a continually paying div-
idend asset is $222, the interest rate is r = 3%, and the dividend yield is q = 2%.
1. What are the one-month and seven-month forward prices for the asset in an
arbitrage-free market?
2. Let be a portfolio on time interval [0, T ] consisting of three positions start-
ing from time 0: borrow $222 at the rate 3%, long 1 unit of the asset, and
short the three-month forward. Is an arbitrage portfolio?
Solution. Denote by K1 and K7 the one-month and seven-month forward

prices respectively. Applying (7.2), we obtain
K1 = 222 e(0.030.02) 12 = $222.19,

1
K7 = 222 e(0.030.02) 12 = $223.30.

7
In order to answer the second part of questions, one needs to fill out the table
below first.
At time 0 At time T
Cash flows from
The detailed work is left as an exercise for the reader.

7.2 Forwards 343
Example 7.15. Let F = F (S, t) = FT (t) = S(t) e(rq)( T t) . Suppose that the un-
derliers price follows a geometric Brownian motion:
dS = S dt + S dB.
We claim that so does the forward price. In fact,
dF = ( r + q) F dt + F dB.
The detailed computation is straightforward and is left for the reader as an

exercise (see Exercise 7.26 on page 381).

Remark 7.4.
1. Using the language of stochastic calculus that we introduced in the last
chapter and only considering the forward contracts that can be replicated
by self-financing trading strategies, the forward price can be expressed by
FT (t) = S(t) /dt,T , (7.4)

2t
where dt,T = EQ ( B(t)/B( T ) | Ft ) with B(0) = 1 and B(t) = e 0 r (u)du, in
which {r (t)} represents a stochastic interest rate process adapted to the fil-
tration {Ft }, and Q is an equivalent martingale measure. It is worth noting
that formula (7.4) can be verified, or proved, by establishing the equality
EQ ((S(t ) FT (t)) B(t)/B(t ) | Ft ) = 0, where t < t T. (Hint: The time t
value of the forward contract entered into at time t is S(t ) FT (t) and FT (t)
is Ft -measurable.)
2. The treatment of forwards and futures in the current and next sections em-
phasizes the intuitive nature of these financial products. For a systematically
derivational approach to them, we refer the reader to the literature (e.g.,
Anderson and Kercheval [1]).

7.2.4 Forward Value Formula
We begin this section with a natural question:

How much is a particular forward contract worth today that was entered
into in the past but has not yet expired?
Let t [0, T ] be the current time; a natural way to evaluate a forward contract
initiated at 0 and expiring at T is to consider the difference between two for-
ward terminal payoffs: one is entered into at time 0 with forward price FT (0)
and the other is entered into currently, at time t, with forward price FT (t).
Both forwards must be written on the same asset and have the same expira-
tion and contract size. Recalling the definition of a (long) forward payoff from
(7.1) on page 339, we obtain
Difference between two forward payoffs = S( T ) FT (0) (S( T ) FT (t))
= FT (t) FT (0).
To discount this difference to get its value at the current time, we obtain the
current (market) value of the (long) forward contract that was entered into at
time 0 in an arbitrage-free market:
Theorem 7.2. The value of a (long) forward contract at time t entered into at time 0
with expiration at time T, denoted by FT (t), is given by
FT (t) = ( FT (t) FT (0)) er( T t) , t [0, T ]. (7.5)
The time-t value of a short forward contract is F T (t).

Proof. An idea of proof is to establish two portfolios at time t such that each
of them produces a deterministic stream of cash flows as indicated in the table
below (where FT (t) is to be determined).
At time t At time T
Cash flows from A FT ( t) FT (t) FT (0)
Cash flows from B ( FT (t) FT (0)) er( T t) FT (t) FT (0)
With this idea, the proof is straightforward and left as an exercise to the reader.

From (7.5), as what we expected indeed, the initial and terminal values of a
forward are
FT (0) = 0,
F T ( T ) = FT ( T ) FT (0) = S( T ) FT (0) = terminal payoff of forward.
Formula (7.5) is also expressed by
FT (t) = ( FT (t) K0 ) er( T t) , t [0, T ], (7.6)

where K0 = FT (0).
Example 7.16. Given K0 = $252, FT (t) = $222, T t =6 months, and r = 3%,
determine the value of the forward contract in an arbitrage-free market.
Solution.
FT (t) = (222 252)e0.030.5 = 29.55,
which (a negative value) means that if the forward contract were to be closed
out currently, the buyer would compensate the writer $29.55.

7.3 Futures 345
As a final note of this section, we emphasize that

The forward price (or contract price) and forward value (or contract value)
are conceptually different. One should not confuse the two.
7.3 Futures
Futures contracts are standardized forward contracts.
7.3.1 Evolution from Forwards to Futures
Forward trading has survived for a few hundred years. The forward contract
was created to stabilize grain prices by farmers on top of the already central-
ized grain trade.
In 1848, the Chicago Board of Trade (CBOT) was formed, and trading was
originally in forward contracts.
Although the forward contract is customized to meet the users special
needs, it has illiquidity due to the lack of market exposure to potential buy-
ers or sellers,16 and it has counterparty risk, the risk that their counterparties
fail to meet their obligations (e.g., default in the payment or even bankrupt).
Thus, counterparties must check each others creditworthiness before a for-
ward contract is entered into. This is why the end users of forward contracts
are mainly big institutions.
Illiquidity and counterparty risk are the inherent limitations of forward
contracts. A way to overcome these shortcomings is to standardize forwards.
The standardized forward contract is called the futures contract. In 1972, the
Chicago Mercantile Exchange (CME) started to offer futures contracts.
To mitigate or even remove credit risks, regulations made in accordance
with laws impose mark-to-market and daily settlement on futures traders
margin accounts (to be explained shortly). The daily balance on such an
account is calculated based on the settlement price defined by an exchange.
To increase liquidity, regulations made in accordance with laws impose stan-
dardizations on terms of futures contracts including what can be delivered,
when it can be delivered, how it can be delivered, where it can be delivered,
16To compare to a familiar environment, simply consider, in a housing market, the difference between
the market exposure of for sale by owner and that of for sale by real-estate agency. Generally
speaking, the bigger the market exposure, the higher the level of liquidity.
and so on. In otherwords, standardize underlying assets, contract size, expira-

tion days, and method of settlement (cash or physical at what location17), and
so forth.
In this way, futures contractors deal with an exchange rather than each other.
That is, in futures trading, your counterparty is an exchange.
Example 7.17. A forward contract allows a farmer to sell 1234 bushels of wheat
next February, whereas a futures contract does not because the size of one
futures contract is 5000 bushels and the expiration months are March, May,
July, September, and December. However, the farmer does not expose himself
to any credit risk by entering into a futures contract.

Definition 7.5. A futures contract is an exchange-traded standardized agree-

ment between two parties to buy or sell an asset at a specified future time
at a price agreed today. The contract is mark to market daily and guaranteed
by the clearinghouse.
The definitions of futures price, long position, short position, expiration date,
etc. are parallel to their counterparts for forwards. For example, the futures price
is the agreed price of an asset in a futures contract.
A futures contract is similar to a forward contract except that contractors
deal with a third party, i.e., an exchange, rather than each other. By doing so,
the inherent credit risk of a forward contract is mitigated because both parties
must meet margin account requirements under the mark-to-market accounting
rule, which makes a futures contract like a sequence of daily forward contracts
until maturity.
Although by definition futures and forward contracts are similar in terms
of the final results, the mark-to-market accounting rule requires futures con-
tractors to settle up daily (or make daily settlement) through the exercise day,
whereas by contrast there is no settlement for forward contracts until the exer-
cise day.
Since margin account requirements and the mark-to-market accounting rule
differentiate the futures contract from the forward contract, to understand fu-
tures we begin with explaining these two concepts. To do so intuitively, we
illustrate margin account requirements and compute the daily margin balance
in the example below.
Before trading a futures contract, the prospective trader must deposit funds
with a broker. This deposit serves as a performance bond and is referred to as
the initial margin, the level of which is based on a function of the price volatility
of the underlier (e.g., a commodity).
17 Futures contracts allow fewer delivery options than forward contracts.
7.3 Futures 347
Example 7.18. (Computing Daily Margin Balance Under Mark-to-Market

Rule) Suppose an investor takes a long position in two October gold futures
contracts with contract size of 100 (i.e., the forward price is for 100 oz of gold).
The initial margin requirement for a futures trader by the clearinghouse is
$5,000 per contract with a maintenance margin level of $4,000.
Suppose that on day one the futures price moves up from $1200 to $1220.
Then the margin account balance will change from $10,000 (2 $5000) to
$14,000 (change = 2 100 $20 = 4, 000). More daily margin account balance
fluctuations are shown in the table18 below:
Day Futures price Daily price change Contract No. Gain/loss Margin Bal.
0 1, 200 10, 000
1 1, 220 +20 2 4, 000 14, 000
2 1, 190 30 2 6, 000 8, 000
3 1, 180 10 2 2, 000 6, 000
The margin account owner receives a margin call for $6, 000 < $8, 000 and has
lost 40% of the account value in only 3 days. Suppose that the exercise date
is still a month away; the buyer of the futures contract still needs to send at
least $4, 000 to the broker before the beginning of the next business day to
bring the margin balance to $10, 000, the initial margin level, or close positions.
Otherwise, the broker can and will cancel the contract.
Although margin required is used as collateral to cover losses, margin calls
can only mitigate, not eliminate, the risk in futures markets, particularly in
FX (foreign exchange) markets. For example, the Swiss Franc fiasco in January
2015 brought many traders margin balance to negative levels over night.
Note that the table shows that the margin balance column fully reflects all
the daily losses from the long position of 2 futures contracts, and observe that
when the initial margin is cut off 40%, the futures price has lost only 1.7% of
its original value.

7.3.3 Impact of Daily Settlement: A Brief Discussion
We let FT (t) be the futures price at time t and FT (t) be the futures value at time
t and close the futures section by the following remark.
18 Precisely speaking, the futures price in the second column should be the daily settlement price or
simply settlement price, which is defined by the exchange. There are different types of settlement pro-
cedures. Each derivative exchange has a set of procedures used to calculate the settlement price. Mar-
gin requirements are based on the daily settlement price, not the daily closing price. For our purpose,
we consider the settlement price to be essentially the closing price on that day.
Remark 7.5.
1. Since the daily settlement makes a futures contract like a sequence of daily
forward contracts until maturity, and the value of a forward is zero at the
initial time, FT (t) = 0 for t < T (although the terminal payoff of a futures
contract is the same as that of its forward counterpart), the futures value is
different from the value of its forward counterpart (as the latter needs not
equal zero except at the time the contract is entered into).
2. Because of the daily settlement and the exchange-treated nature of a futures
contract, FT (t), a futures price for delivery of an asset at time t < T, is an
agreed price between the trader and the exchange (i.e., a price determined
by the exchange rather) and, therefore, behaves more like a market price. It
has been proved that if futures interest rates are deterministic, then under
the no-arbitrage assumption, FT (t) = FT (t). Otherwise the equality may not
hold. More description about the behavior of FT (t) can be found in the liter-
ature (e.g., Hull [11]).
For more impact of daily mark to market on futures contracts such as the
correlation between the futures price movements and interest rate move-
ments, we refer the reader to the literature (e.g., Cox, Ingersoll, and Ross [4]
and Duffie and Stanton [8]).
3. For the following reasons:
Futures have lower transaction cost, since the evolution from OTC-traded
forwards to exchange-traded futures encourages liquidity;
With futures, it is equally easy to go short as to go long (a convenience
inherited from forwards);
The value of forward kept at zero (i.e., FT (t) = 0) allows a futures con-
tractor (either buyer or seller) to close out his position at any time;
it is easier to use futures to hedge, particularly easier to go short with futures
market than with the spot market because of the uptick rule19 imposed on
the spot market.

7.4 Swaps
Swap contracts are basically a series of cash-settled forward contracts that re-
quire action to be taken by investors periodically over time.
The tailor-made structures of swaps create a wide variety of financial instru-
ments that financial institutions trade in order to hedge against risk.
19 The SEC requires that every short sale transaction be entered at a price that is higher than the price
of the last trade.
7.4 Swaps 349
7.4.1 A Brief Introduction
Definition 7.6. A swap contract, or simply a swap, is an OTC agreement between

two parties to exchange (or swap) two streams of cash flows at specified future
times according to certain specified rules.
These cash flows are most commonly the interest payments associated with
debt service. Financial institutions and companies dominate the swaps market
with almost no individual participation.
By definition, swaps are basically sequential cash-settled forward contracts
that require action to be taken by the counterparties on periodic dates. Conse-
quently, the initial value of a swap should be zero.
Example 7.19. Consider a forward contract on 100 ounces of gold at expiration

in 6 months at the price $1281 per ounce. If the gold is viewed as a currency,
this forward can be viewed as a simple example of a swap: a party entering
into the forward contract to buy 100 ounces of gold for $1281 in 6 months is
equivalent to a party entering into a swap to use $128, 100 to exchange for 100
ounces of gold in 6 months.

Popular swaps are classified into different categories.

An interest rate swap is a transfer of interest rate cash flows without transfer-
ring underlying debt and requires both cash flow streams to be in the same cur-
rency. The predetermined amounts on which the exchange interest payments
are based are called the notional principal.
A currency swap requires the principal to be specified in each of the two
currencies and two parties to exchange cash flows as well as the principals
at the beginning and end of the life of the swap.
An interest rate swap is called a plain vanilla swap20 if one stream of cash flow
consists of floating interest rate payments and another consists of fixed interest
rate payments. The former stream is called the floating leg and the latter the fixed
leg. By tradition, the floating rate payer is called a buyer of the swap and the
fixed rate payer is a seller of the swap. During the life of a plain vanilla swap,
every period there is an exchange of payments. The payments under the fixed
leg are always known, since the fixed rate is set on the pricing date of the swap,
but the payments under the floating leg are known only one period prior to
the exchange of payments. For example, if the payments are made at the end
of each period, then the floating rate payment will be based on the floating
rate that prevailed at the beginning of the period, since the floating rate is reset
periodically based on market levels after the rate for the first payment is set at
market level on the pricing date (see Example 7.20).
20 The name plain vanilla swap reflects that those swaps do not possess any special or unusual features.
Other than the above categories of swaps, there are also credit default swaps
(CDS), which can be used as a protection against credit loss, commodity swaps,
which can be used by commodity producers to manage their exposure to price
fluctuations, and other variations of swaps, since tailor-made swap structures
create a wide variety of financial instruments that financial institutions trade
in order to hedge against risk.
Among all different types and variations of swaps, the most widely used are
interest rate swaps, which are used mostly to reduce borrowing costs:
1. Based on changes in its long-term or short-term assets and its credit rating,
an institution with an existing debt service obligation faces higher than ex-
pected borrowing costs because of a change in its initial interest rate outlook.
To avoid these costs, the institution wants to swap to a different exposure
(see Example 7.20).
2. To receive lower borrowing costs than those available by directly accessing
the fixed-rate or floating-rate markets, two institutions may work together
by exploiting their comparative advantages as borrowers in different mar-
kets and swapping the proceeds (see Example 7.21).
The interest rate swaps market is the largest and fastest growing financial
derivative market in the world.
It is worth noting that in the next two examples, we will display only
the mechanics of interest rate swaps and ignore credit risk differences, even
though counterparty credit risks are very important for swap traders to
evaluate.
Example 7.20. (Plain Vanilla Swap) Consider the scenario of two companies A
and B. Company A has taken a loan at a six-month LIBOR21 plus 1% floating,22
but now would like to have the loan at a fixed rate of 3% since the company
expects interest rates to rise above 3% soon, whereas company B has a loan at
a fixed annual rate of 3% and expects interest rates to drop below 3% soon.
Since for a company to pay off an old loan and apply for a new loan or to
refinance a loan can be not only costly but also legal document-intensive due
to regulations, companies A and B decide to enter into a swap contract, which
is a much simpler way of exchange of interest rates between fixed and floating.
Assume that the terms of the swap contract include the following:
The notional principal is two million dollars,
The life of the contract is 4 years,
A pays B six-month LIBOR + 1%,
21 See Section 1.1.2 on page 4.

22 Such a loan is called a floating rate loan or a variable or adjustable rate loan, which is a debt, such as a
bond, mortgage, or credit, that does not have a fixed interest rate over the life of the debt. The interest
rate on a floating rate loan is referred to as a floating interest rate, or variable or adjustable rate.
7.4 Swaps 351
B pays A 3% fixed and

There is an exchange of payments every 6 months from the initialization.
Given the LIBOR rates in the table below, since the floating leg payment is
always known 6 months before the exchange of interest payments, company
B receives 2, 000, 000 (1% + 1%)/2 = $20, 000 from company A as the first
period interest payment (out of a total of eight variable payments), and the
rest of the floating cash flow and fixed cash flow of the swap is illustrated in
the table:
Time (months) Six-month LIBOR Payment from A to B Payment from B to A
0 1%
6 2% $20,000 $30,000
12 3% $30,000 $30,000
18 2% $40,000 $30,000
24 3% $30,000 $30,000
30 5% $40,000 $30,000
36 4% $60,000 $30,000
42 2% $50,000 $30,000
48 $30,000 $30,000
A straightforward verification of the interest payments in the above table is
left as an exercise for the reader.

Example 7.21. (Mechanics of Interest Rate Swaps) Suppose that both com-
panies X and Y need to borrow US dollars and that company X would like to
borrow at fixed rate, whereas company Y would like to borrow at floating rate.
The cost to each company of accessing either the fixed rate or the floating rate
market for a new debt issue is given below23:
Borrower Fixed rate Floating rate

X 5.50% LIBOR + 0.40%
Y 4.30% LIBOR + 0.20%
Difference (margin) 1.2% (120bp) 0.2% (20bp)
Given the differences in rates indicated in the table above, companies X and
Y realize that they could achieve a combined 100 basis point (i.e., 1%) savings
and decide to enter a swap with a swap bank24 B.
We claim that X, Y, and B can all benefit financially if X borrows at the
floating rate, Y borrows at the fixed rate, X and Y swap interest payments,
and B as the intermediary charges 0.1% of the notional principal. To prove our
23 Company Y enjoys a lower borrowing cost in both markets because we assume that company Y
has a better credit rating than company X.
24 A swap bank is a generic term for a financial institution that facilitates swaps between counterparties
and serves as a broker or a dealer for the trading.

claim to be true, we demonstrate the simple mechanics of the swap with the
calculations below:
1. X pays 4.9% fixed to B and B pays 4.8% fixed to Y, and
2. Y pays LIBOR floating to B and B pays the same LIBOR floating to X.
3. The borrowing cost for X after swapping proceeds becomes at the fixed rate
4.9% LIBOR + ( LIBOR + 0.4%) = 5.3%,
which is less than 5.5%. Thus, the savings for X is 0.2%.

4. The borrowing cost for Y after swapping proceeds becomes at the floating
rate
LIBOR 4.8% + 4.3% = LIBOR 0.5%,
which is less than LIBOR + 0.2%. Thus, the savings for Y is 0.7%.

Although payments paid in a foreign currency may entail currency risk, a

currency swap can be used to transform loans or cash flows from one currency
to another. The next example briefly explains certain purposes of doing so.
Example 7.22. (Purpose of Currency Swaps) Consider the scenario of two

companies A and B in two different countries. Company A is an US-based
company, whereas company B is a European company. Each company has
plan to enter global markets and needs to take loans to fund capital expen-
ditures. Moreover, company A needs to have Eurodollars to carry out its plan
and company B needs to have US dollars.
Because of a variety of reasons such as different regulations in different
countries and geopolitical concerns, it is often advantageous for a company
to borrow from its domestic banks and disadvantageous to borrow from for-
eign banks. To get around the financial difficulties of borrowing from foreign
banks, company A could take a loan from an US bank and company B from a
European bank, and then the two companies could enter a swap contract.
Remark 7.6.
1. A variation of swap is called a variance swap which is an OTC instrument
that allows one to speculate on or hedge risks associated with volatility. Vari-
ance swaps on major stock indexes such as S&P 500 are actively traded.
2. Since an interest rate swap contract is basically a series of forward contracts,
intuitively, its value is zero at the time the swap is entered into (although
some of these forward contracts may have nonnegative values and others
may have negative values, the sum of all of them is zero) and may change
over time due to the change of interest rates. For the buyer of the swap,
the party who receives fixed and pays floating, the value of the swap is
positive if the fixed rate is greater than the floating rate; the value of the
7.5 Options 353
swap is zero if the fixed rate equals the floating rate; and the value of the
swap is negative if the fixed rate is less than the floating rate.
Since the floating rates cannot be observed beforehand, the valuation of an
interest rate swap may involve calculations of forward rates based on a yield
curve in addition to present value techniques.
Generally speaking, the pricing and valuation of swaps can be complex and
requires vigorous financial analysis to derive fair values. We refer the reader
to the literature on this subject.

7.5 Options
Options are contingent claims which may be viewed as a variation of for-

wards25 that provide an investor with an option, but not an obligation, to com-
plete a transaction on or by a future date.
The most practical elements covered in this section are terminal payoff and
profit diagrams as they provide the simplest way to analyze option strategies.
A theoretical discussion on the risk-neutral valuation of options will be pro-
vided in the next chapter.
Definition 7.7. An option contract, or simply an option, is either an individually

negotiated OTC or exchange-traded agreement between two parties, which
grants the buyer (or holder or owner) the right, but not the obligation, to buy
or sell a specified quantity of a specified asset on or by a specified date at a
specified price.
The premium of an option contract is the amount that the buyer has to pay
and the seller (or writer) receives at the time when both parties enter into the
contract.
The specified asset is called the underlier or underlying asset of the option.
The specified date is called the expiration or maturity (date) of the option.
The specified price is called the strike price or exercise price of the option.
Naturally, an expiration date is also called an exercise date.
The specified quantity is called the contract size, which denotes how much of
the underlying asset will change hands if the option is exercised, where exercise
an option means to put into effect the right in an option contract.
25We mean the European-style option (to be defined shortly), which is the basic option style from
which other styles of options derive.
By definition, an option contract is structured for one to pay money to have

a choice in the future. This explains the name option.
Through the standardization of the underliers, contract sizes, and expiration
dates, organized option exchanges such as CBOE26 and NYSE provide the
advantages of liquidity, low transaction costs, and safety for option trading.
Option Styles
American-style options, or simply American options, can be exercised at any

time prior to or on expiration day, whereas European-style options, or sim-
ply European options, can only be exercised on expiration day.27 Therefore,
American options are more flexible than European options.
Since the ability to exercise an American option at any time prior to or on
expiration makes American options more flexible than European options,
American options are more valuable than European options.28
Although the flexibility and ease of exercise makes American options more
valuable, it also makes the valuations or pricing of the American options more
complex and difficult than European options.29
Option Types
The basic bread-and-butter options fall into one of the following two types,
designated according to buying and selling rights:
Definition 7.8.
1. A call contract, or a call option or simply a call, is an option contract which
grants the buyer the right to buy a specified quantity of an underlying asset
on or by an expiration date T at a strike price K. The payoff is max{S( T )
K, 0} when exercised at expiration T.
2. A put contract, or a put option or simply a put, is an option contract which
grants the buyer the right to sell a specified quantity of an underlying asset
26 The Chicago Board Options Exchange (CBOE), a spin off from the Chicago Board of Trades, first
traded standardized options in 1973 and NYSE in 1982.

27 There are other option styles such as Asian or Bermuda. We only consider American and European
options because they are the most actively traded options.

28 Indeed, American-style stock options tend to cost more than equivalent European-style options
for the same stock in practice. Almost all exchange-traded stock options are American-style options,
whereas stock index options can be issued as either American or European options (e.g., S&P 100
index options are American options, and Nasdaq 100 index options are European options).
29 Option pricing done by the Black-Scholes-Merton model applies to European options, not to Amer-
ican options, and reflects the risk associated with having to wait to exercise the option, which is not
appropriate for American options because of the possibility of early exercise.
7.5 Options 355
on or by an expiration date T at a strike price K. The payoff is max{K

S( T ), 0} when exercised at expiration T.
A call option is called a call because the buyer (owner) has the right to call
the underlying asset away from the seller (writer). A put option is called a
put because the buyer (owner) has the right to put the underlying asset to
the seller (writer). The payoff may be considered as the option value at exercise.
Clearly, an option is a financial instrument with nonnegative value at any time
(for it involves no obligation prior to expiration).
Calls and puts are also referred to as vanilla options.30
Although in reality stock options are American options and American op-
tions are more useful, all the examples in this section will be confined to Eu-
ropean options hypothetically on stocks. They are easier to understand and
give the background needed for studying the option pricing models in Chap-
ter 8 and for following related literature beyond. The next example provides
an intuition about what a call option actually means.
Example 7.23. (European Call) To make a call option easier to understand, in

the example we assume the contract size to be 1, that is, a call on 1 unit of the
underlier.
Consider a six-month (i.e., T = 0.5) European call written on stock XYZ with
strike price K = $7.
By entering this contract as a buyer, you have the right (not the obligation)
to buy 100 shares of stock XYZ at the price $7 per share at time T (in 6 months).
On expiration day T (a Friday), whether you exercise the call (i.e., exercise
the right to buy the stock) depends on the stock price in the spot market on
that Friday:
If the stock price is above $7, say $10, you exercise the call to buy 1 share of
the stock at $7 and sell it immediately for $10 to make a profit of $3 (assuming
no commissions). We say that the terminal payoff of the call is $3 if the stock
price is $10 at expiration.
If the stock price is below $7, you do not want to exercise the call because you
can buy the stock at a lower price in the spot market. Therefore, you let the call
expire worthless by doing nothing. We say that the terminal payoff of the call
is $0 if the stock price is below the strike price at expiration.
More generally speaking, if we denote by C (S, t) the value of European call
option at time t, then at expiration T, the terminal payoff of an option position
of long a call (or simply, terminal payoff of a call), i.e., C (S, T ), is found to be
.
0 if S( T ) K
C (S, T ) = max{S( T ) K, 0} = (7.7)
S( T ) K if S( T ) > K.
30The name vanilla reflects that calls and puts do not possess any special or unusual features. In
contrast, exotic options have more complex features.
In words,
The terminal payoff of a call, C (S( T ), T ), is the value of the call C (S(t), t)
at expiration T, which is its market value at expiration.

In a similar fashion, if we denote by P(S, t) the value of European put option
at time t, then at expiration T ,
.
K S( T ) if S( T ) < K
P(S, T ) = max{K S( T ), 0} = (7.8)
0 if S( T ) K.
Option contracts are defined by their terms, which are standardized by the
exchange on which the option is listed. In the next two examples, we explain
standardized contract size and expiration dates for practical purposes.
Example 7.24. (Option Contract Size) For equity options (underliers are
stocks), the contract size (also called the option trading unit or multiplier) is
100. In other words, one contract controls 100 shares of the underlying stock.
Suppose that you want to purchase a call on XYZ stock with strike price
$50 and premium $1.50. Then you will have to pay $150 for the right to buy
100 shares of XYZ stock in the contract. Note that, in practice, you would also
have to pay commissions to your broker.
For standard index options (underliers are stock indexes), the contract size
is also 100. In other words, the notional value underlying each contract equals
$100 multiplied by the index value.
However, for mini options, the contract size is 10 (representing 10 shares of
an underlier). For example, 10 Mini-SPX options equal 1 SPX full value con-
tract. That is, the notional value underlying each mini SPX contract equals $10
multiplied by the S&P 500 index value.31

Example 7.25. (Option Contract Expiration Date) Equity and index options
expire at 4 pm EDT on the third Friday32 of the expiration month in the sense
that they no longer trade; however, the official expiration day is the Saturday
immediately following that Friday.
Knowing the month in which the option you want to purchase will expire
is very important, since it is a natural part of your trading strategy design and
will have a significant impact on the outcome of your trade.
Traditionally, there is an option expiration cycle for each equity on which
options are written. Each cycle contains 4 months. For example, suppose that
today is May 11, 2015 and a stock XYZ has a February option expiration
31 For standard S&P 500 index futures, the multiplier is 250 (index level 250 = price); for E-mini SPX
futures (smaller contract), the multiplier is 50. Multiplier varies for indices.
32 If the third Friday is a market holiday, then those options expire on the third Thursday.
7.5 Options 357
cycle. Then tradable option contracts written on stock XYZ have expiration
months (at least) in May (the current or front month), August (the near month),
November, and February 2016.
General information on exchange-traded option expiration dates is available
on the Options Expiration Calendar at the CBOE website.

7.5.2 How Options Work
Equity options can be used for a variety of purposes such as hedging exist-
ing positions and speculating or buying or selling stocks. We illustrate some
simple applications in the following examples.
Example 7.26. (How Call Buying Work) If the price of stock XYZ is $7 per
share today (say, May 11) and you speculate that it could rocket above $15
within 30 days, then you could buy the June (expiration) 15 (strike) call option.
Suppose that the premium is $1. Ignoring commissions, in order for you to
break even, the stock price would need to rise to $16. In order for you to make
a profit, the stock price would need to exceed $16.
As a call holder, your maximum possible loss is the premium you paid ($100
per contract for 100 shares).
If you exercise the call when the stock price is $20 you immediately have
$300 profit (per contract) on paper. You may continue to hold the shares if you
think the stock price will continue to rise. Otherwise, you could sell your call
contract (i.e., close your position) and nail down the profit which is a return of
300% (much higher than 2077 = 185.7% return from buying the stock).
Further discussion. You may also buy call options in the situation when you
wait for cash coming in, say, from selling stocks other than the underlier.

Example 7.27. (How Call Selling Work) Suppose you have bought 100 shares
of stock XYZ at price $7 per share. Today (say, May 11), you think that, alt-
hough the stock has a good potential to continue to move much higher in the
long run, in the short run, the price may move down before breaking a resis-
tance level at about $10 (e.g., historical stock price movements showed that
the price moved down from a level at about $10 a number of times). To hedge
your position, you could write 1 June 10 call option.
Suppose that the premium is $1.50. The premium that you receive produces
income ($150) on the stock that is already in your portfolio. Ignoring commis-
sions, you will not lose money as long as the price is not below $5.50.
Further discussion. In fact, the way of using call options as we just explained in
this example is referred to as a covered call strategy, which means that you write
calls when you have enough shares of the underlying stock in your portfolio.
In contrast, a call contract is referred to as a naked or uncovered call if it is not

backed by an offsetting position of underlying stock. In other words, you write
a call without any shares of the underlying stock in your portfolio.
Generally speaking, the covered call strategy is applied for the purpose of
holding a stock for long term, possibly for tax or dividend advantages. This
strategy allows you to earn the premium from the option writing but caps
your upside potential gains.
A covered call creates a portfolio consisting of longing 1 share of underlier
and shorting a 1 unit call. The initial value of the portfolio is S(0) c where c
is the premium received by the writer at initial time (so S(0) c 0 by Prop-
erty 7.1 on page 372). Assume the underlier to be nondividend paying and
recall (7.7), the portfolio value at expiration is S( T ) max{S( T ) K, 0}, and
therefore, the terminal profit generated by a covered call is

S( T ) + c S(0) if S( T ) K
S( T ) max{S( T ) K, 0} (S(0) c) = (7.9)
K + c S(0) if S( T ) > K.
Indeed, the profit is capped by the constant value K + c S(0) if S( T ) > K as

we expected.

Example 7.28. (How Put Buying Work) Again, suppose that you bought 100
shares of stock XYZ at price $7 per share. Today (say, May 11), you think that,
although the stock has a good potential to continue to move much higher in
the long run, in the short run, the price may move down before breaking a
resistance level at about $10.
Aside from engaging in the covered call strategy in the last example, you
can also consider buying puts if you think the stock may go down much lower
than $5.50 in the short term. If you are wrong, you lose the premium that you
paid when you entered into the contract. In a way, put options are parallel to
insurance policies.
Further discussion. One can make a real profit in a big downward movement of
stock price by either buying puts or shorting the stock directly. Ignoring mar-
gin requirements and commissions, in order to make a profit from buying puts,
one needs to be right about both direction and timing of the price movement;
when shorting stocks, one needs to be right only about the direction. However,
the advantage of buying puts over shorting stock is that buying puts allows
you to determine and prepare for a worst-case scenario as you know that your
loss cannot exceed the premium paid when you entered into the contract.
Example 7.29. (How Put Selling Work) One strategy of buying stocks is by
selling puts.
Suppose you would like to buy 100 shares of stock XYZ at about $6 and the
price of the stock is $7 today (May 11). You could sell a June $6 put option on
7.5 Options 359
the stock and earn premium income $25 (i.e., premium is $0.25) immediately.
If the price of the stock drops below $6, the put buyer will exercise the put,
and you will have to honor your commitment to buy the stock at $6 (even if
the stock price plunges to $2 unexpectedly). If the price of the stock stays above
or at $6, then the put buyer will not exercise the put. Although you will not get
a chance to buy the stock, you still keep the premium income.
In fact, one way for speculative traders to earn premium income is selling
puts by thinking (being so confident) that the stock will not reach the strike
price minus the premium.
Further discussion. By definition, a put contract grants the buyer a right, not an
obligation, to sell the underlying stock. However, by selling this put, the put
writer assumes an obligation, not a right, to buy 100 shares of the stock at the
strike price if the buyer of the put wants to sell (i.e., if the buyer exercises
the put), regardless of the price of the stock in the spot market. Because of
this obligation, when writing a naked put, the put writer should not rely on a
wishful thinking but be prepared to take a loss or to own the stock for a while.
The same, if not higher, level of prudence should apply to consideration of
writing a naked call (see rationale in Exercise 7.11).

7.5.3 Terminal Payoff and Profit Diagrams
For a portfolio of options on the same underlier and with the same expiration,
an option terminal payoff diagram provides an extremely useful visualization
that traders rely on to analyze a portfolio strategy. This diagram illustrates how
a right combination of option positions can form a portfolio that has a risk exp-
osure to almost any chosen kind of market volatility scenarios. In addition, a
terminal profit diagram, a chart obtained by a simple translation of the cor-
responding terminal payoff diagram, provides a clear visual presentation of
the range of profit and loss and the break-even point as outcomes of a chosen
trading strategy.
We emphasize that one should not confuse the two.
Just to make these tools of the trade easier to underhand, our discussion
in this section will focus on European options only, since, prior to expiration, the
payoff diagrams of American options may be more complicated.
Also, without loss of generality, we assume that all options in a given portfolio
are written on the same stock and have the same expiration T.
Terminal Payoff Diagrams
Again, let C (S, t) and P(S, t) respectively denote the value of European call
and European put options at time t, t [0, T ], where T is the expiration.
Definition 7.9. An option terminal payoff diagram is a graph of the value of the
option position (e.g., long a call or short a put) at expiration T as a function of
the underlier price at T.
Recall (7.7) and (7.8) on page 355; the terminal payoffs of an option position of
a long call and that of a long put are represented respectively by
.
0 if S( T ) K
C (S, T ) = max{S( T ) K, 0} =
S( T ) K if S( T ) > K,
and
.
K S( T ) if S( T ) < K
P(S, T ) = max{K S( T ), 0} =
0 if S( T ) K.
Example 7.30.
1. If we let x = S( T ) and y = f (S( T )) C (S( T ), T ), then the graph of y = f ( x )
on xy-plan is the graph of C (S( T ), T ) against S( T ) given in Fig. (3). That is,
Fig. (3) is the terminal payoff diagram of an option position of long a call by
Definition 7.9.
2. By Definition 7.9, the graph of P(S( T ), T ) against S( T ) is the terminal payoff
diagram of long-a-put. Can you sketch this graph? Does your graph coin-
cide with the one in Fig. (6)?
3. Since the value of a short position on a call at expiration T is C (S( T ), T ),
Fig. (4) gives the terminal payoff diagram of short-a-call.
4. Similarly, we see that the terminal payoff diagram of short a put is given by
Fig. (7).

Fig. (3) long call (C) Fig. (4) short call (C) Fig. (5) long stock (S )
O O O

payoff payoff payoff

K K

/ ?
? / /
0 K S(T ) 0 K ??? S(T ) 0 S(T )
? ??
K K
7.5 Options 361
Fig. (6) long put ( P) Fig. (7) short put ( P) Fig. (8) long bond
O O O
?
K ?? K K
??
??
? / / /
S(T ) K S(T ) S(T )

0 K 0 0

K
Fig. (9) P+C Fig. (10) P1 +C2 Fig. (11) short bond
O O O
? K1 ?? /
K ?? ??
?? ??
0 S(T )
?? ??
? / / K
0 K S(T ) 0 K1 K2 S(T )
Notice that the result of superimposing Fig. (3) and Fig. (7) is the graph of
translating Fig. (5) to the right by K units, i.e., C P = S K at time T.
Also notice that a bond generates only vertical translations of payoff dia-
grams. Therefore, taking a position in a bond does not play any hedging
role.
Recall that a portfolio is a linear combination of securities. If we let rep-

resent a portfolio consisting of longing a call, shorting two puts, and longing
three shares of the underlier, then we write = C 2P + 3S. Also, as in the
past, (t) = C (S(t), K ) 2P(S(t), K ) + 3S(t) represents the value of the port-
folio at time t.
Let be a portfolio consisting of long or short positions of a stock (S) calls
(Cs) and puts (Ps) on the same underlier (S) with the same expiration and zero-
coupon bonds (Bs); we introduce the terminology of portfolio terminal payoff
diagram:
Definition 7.10. A portfolio terminal payoff diagram is a graph of ( T ) as
a function of S( T ).
Example 7.31.
1. If consists of long one share of stock (S), then the terminal payoff diagram
of portfolio is the payoff diagram of long a stock. Can you sketch this
graph? Does your graph coincide with Fig. (5) on page 360?
2. If consists of simultaneous long a call (C) and a put (P), which are on
the same stock and with same expiration, then = C + P, and the terminal
payoff diagram of portfolio is given by Fig. (9).
Portfolio = C + P is called a straddle, which will be revisited in Section 7.5.5.

We provide an intuitive way to derive the put-call parity in the next example.
Example 7.32. (Put-Call Parity of European Options) Verify that superimpos-

ing Fig. (4), Fig. (5), and Fig. (6) on the previous page yields Fig. (8). It follows
that S + P C = K at time T. To discount K back to time t [0, T ), we obtain
the put-call parity:
S(t) + P(t) C (t) = Ker( T t) , (7.10)
where we assume that the underlier is nondividend paying and r is the risk-
free interest rate.
If the underlier is an asset with annual dividend yield q (see page 341),
the put-call parity takes the form
S ( t ) e q ( T t ) + P ( t ) C ( t ) = K e r ( T t ) . (7.11)
Further discussion about the put-call parity for European options will be given
in Section 7.5.6 and Section 7.5.7.

Remark 7.7. Four Keys to Reading Option Payoff Diagrams:

1. Each sharp-corner point corresponds to a strike price (on S(T)-axis).
2. Each vertical parallel shift can be done by using a bond.
3. Each call-put can be converted to a put-call under the put-call parity.
4. Each of the eight most basic diagrams (long call, short call, long put, short
put, long stock, short stock, long bond, short bond) must be kept in mind.

Terminal Profit Diagrams
Since the profit generated by a portfolio is the difference between the termi-
nal payoff and the initial price you pay (e.g., the option premium), ignoring
commissions, the terminal profit diagram of a portfolio is the graph that is the
vertical translation of the corresponding terminal payoff diagram by the initial
price you pay in the same coordinate system.
Example 7.33. A portfolio consisting of simultaneously buying a call and a put

is called a straddle if the call and put are on the same underlier with the same
expiration (T) and strike (K). The terminal payoff of a straddle is provided by
Fig. (9) on page 360, and the terminal profit diagram is provided below:
7.5 Options 363
where a is the net premium of the straddle. Then the range of profit is given by
S( T ) [0, K a) (K + a, ) and the range of loss by S( T ) (K a, K + a),
and the break-even point is S( T ) = K a or K + a.

Remark 7.8. More precisely speaking, the terminal profit of a portfolio is defined as
the difference between the terminal payoff and the future value at time T of the initial
cost. For example, by this definition, the profit for a call buyer at the expiration
T is C (S( T ), K ) c erT = max{S( T ) K, 0} c erT , where we assume that the
buyer purchases the call at time t = 0 by paying premium $ c.
However, taking into consideration the two facts:
1. 1 erT is relatively too small and
2. The terminal profit diagram is only used for visualization of option strategy
outcomes,
we may ignore the compounding factor erT in the future value of the initial
cost.

7.5.4 Market Sentiment Terminologies and Option Moneyness
In this section, we introduce some often used jargon in option trading.

Market sentiment is a terminology used to describe the overall attitude of
investors toward a particular security or a financial market.
The words bullish and bearish are often used to describe a market sentiment.
The bullish sentiment on a security (or a market) means that the security
(or the market) price is believed to be going up.
The bearish sentiment on a security (or a market) means that the security
(or the market) price is believed to be going down.
Example 7.34. The statement that investors who buy calls are bullish on the
underlying stock is equivalent to the statement that investors who buy calls
think that the underlying stock price will go up.
In short, we can say that buying a call is a bullish bet on the underlier.

Example 7.35. The statement that investors who buy puts are bearish on the
underlying stock is equivalent to the statement that investors who buy puts
think that the underlying stock price will go down.
In short, we can say that buying a put is a bearish bet on the underlier.

Moneyness is a term describing the condition of an option in terms of its

strike price in relation to the underliers spot price. Again, let S(t) be the spot
price of the underlier at time t and K be the strike of the option.
An option is said to be in-the-money at time t if the option has a positive
payoff if it is exercised at time t. More specifically,
A call is in-the-money at time t if S(t) > K,
A put is in-the-money at time t if S(t) < K.
An option is said to be at-the-money at time t if the payoff of the option is

zero if it is exercised at time t. More specifically,
A call is at-the-money at time t if S(t) = K,
A put is at-the-money at time t if S(t) = K.
An option is said to be out of the money at time t if the option has a negative
payoff if it is exercised at time t. More specifically,
A call is out of the money at time t if S(t) < K,
A put is out of the money at time t if S(t) > K.
Example 7.36. In option trading, the statement that a call is deep in the money at
time t means that the underlier spot price S(t) is well above the strike price of
the call.
In practice, the deep in-the-money condition is a condition in which a call
value changes dollar for dollar with the spot price movement of the underlier;
for an at-the-money call, its value changes only about 50% of the spot price
change.
A parallel statement can be made for a deep in-the-money put.

Example 7.37. In option trading, the statement that a call is deep out of the money
at time t means that the underlier spot price S(t) is well below the strike price
of the call.
A parallel statement can be made for a deep out-of-the-money put.
In practice, deep out-of-the-money options are always worth something bec-
ause there is always a probability that the condition may change. They may be
used for two trading strategieshedging and speculation. In some ways, deep
out-of-the-money options are almost like purchasing lottery tickets, i.e., they
present an opportunity for profits but with a low probability of success.

Remark 7.9. From a pure financial theory point of view, one would define the
condition for a call being in the money at time t to be S(t) > Ker( T t) . However,
this definition would not make any real difference in terms of the purpose
which the option moneyness terminologies serve, since the difference between
K and Ker( T t) is very small. An expression like S(t) > K is much simpler,
and therefore more convenient to use, than S(t) > Ker( T t) .
7.5 Options 365
Similar argument applies to the definitions of out of the money and at the
money.

7.5.5 Option Strategies: Straddle, Strangle, and Spread
A straddle is a combination of two positionsone is long on a call and the other

is long on a put, where the call and put have the same underlier (S), strike (K),
and expiration (T).
The long straddle, or buy straddle or simply straddle, is a (underlier spot)
market-neutral33 option strategy that involves simultaneously buying a call
and a put. It is a bullish bet on the underliers volatility and one of the least
sophisticated option strategies.
Example 7.38. (Straddle) Let us consider a portfolio = C + P where the und-

erlier is modeled by one period binomial tree on [0, T ], with
.
$13 with probability 35
S(0) = $7, S( T ) =
$1 with probability 25 ,
and
c = $1, p = $1, K = $7.
What is the expected return of ? What is the terminal profit diagram of the
straddle?
Solution. Note that

3 2
E (( T )) = E (C (S( T ), K )) + E ( P(S( T ), K )) = 6 + 6 = 6.
5 5
6 2
Therefore, the expected return is 2 = 200%.
The terminal profit diagram of long a straddle (C+P) can be found in

Example 7.33 on page 362 with K = 7 and a = 2. The break-even spot price
of the underlier is either $5 or $9. The range of loss is [2, 0) corresponding to
S( T ) (5, 9).
We let the reader observe the potential profits when either the call or the put
in the straddle is deep in the money.
Further discussion. If you are confident that a stock price will change dramat-
ically, but not sure about the direction of the price move, you may consider
the long straddle. However, in practice, it is often the situation that both call
33 A neutral option strategy is a strategy that is designed to profit from either a rise or fall (non-
directional) in the underlier price.
and put prices become much higher than expected when there is a major news
pending.

A strangle is a combination of two positions, one is long on a put with strike
K1 and another is long on a call with strike K2 (K2 > K1 ), where the call and
put have the same underlier (S) and expiration (T).
The long strangle, or buy strangle or simply strangle, is a (underlier spot)
market-neutral option strategy that involves simultaneously buying an out-
of-the-money put and an out-of-the-money call. It is also a bullish bet on the
underliers volatility.
Example 7.39. (Strangle) Typically, when you long a strangle, the two strike
prices K2 > K1 are near the money and out of the money (thus, close to in the
money). Note that you are betting on volatility rather than on the underlying
stock alpha when you long either a straddle or a strangle and the one with
lower net premium is preferred. The terminal payoff of a strangle is provided
on page 360, Fig. (10), and the terminal profit diagram is provided below:
where a is the net premium of the strangle. Identifying the range of profit, loss
and break- even point as outcomes of the corresponding portfolio strategy is

A spread strategy corresponds to a portfolio that consists of two or more
options of the same type to achieve a certain level of hedging effect. Spreads
are the basic building blocks of many option strategies although we will only
briefly introduce three of them in the following.
Example 7.40. (Price Spread or Vertical Spread) One way to construct a price
spread, or vertical spread, is to use two calls (or two puts) on the same und-
erlying asset (S) and the same expiration date (T) but with different strike
prices34 (Ki , i = 1, 2). One is bought and another is sold in order to achieve a
level of hedging effect.
A spread is called a bull spread if it is designed to profit from an upward
movement of the underliers price (see payoff diagram below).
A spread is called a bear spread if it is designed to profit from a downward
movement of the underliers price (see payoff diagram below).
34This explains the name price spread (more precisely, strike price spread). Since the strike prices
are listed vertically by the news media, a price spread is also referred to as a vertical spread.
7.5 Options 367
Using a formula to express the terminal payoff of each spread shown above is

Example 7.41. (Calendar Spread or Time Spread or Horizontal Spread) One
way to construct a calendar spread, also called time spread or horizontal spread,
is to use two calls (or two puts) on the same underlier (S) and with the same
strike price (K) but on different expiration dates35 (Ti , i = 1, 2).
If you predict (as a pure speculation) that the underlying stock price will rise
above $45 in a few months, you may want to consider the following bullish
calendar call spread.
Suppose that XYZ 45 (American) calls are priced below.
Expiration month May June July August
Premium of XYZ 45 call $0.25 $2 $3 $5
By buying 1 XYZ Aug 45 call only, you will lose $500 if the stock XYZ drops
to, say, $35. In contrast, by making the transaction of buying 1 XYZ Aug 45 call
and writing 1 XYZ July 45 call, you will loss only $200 if the stock drops to $35.
The hedging effect created by the spread cuts the loss by more than half.

Example 7.42. (Butterfly Spread) A butterfly spread is a portfolio consisting of
four options of the same type on the same underlier (S) with the same exp-
iration date (T) but three different strike prices (K1 < K2 < K3 ). Generally, K2 =
2 (K1 + K3 ) and is close to the current underlier price. The graph below is the
1
terminal payoff diagram of a butterfly spread consisting of long 1 call with

strike K1 , write (i.e., short) 2 calls with strike K2 , and long 1 call with strike K3 .
The butterfly spread is designed to profit when the underlier price movement
stays close to the current price. Using a formula to express the terminal payoff
of the butterfly spread indicated above is left as an exercise for the reader.

35This explains the name calendar spread or time spread. Since the expiration months are listed
across the top of the newspaper page horizontally, a calendar spread is also referred to as a horizontal
spread.
7.5.6 Put-Call Parity for European Options Revisited
The basic forms of put-call parity given in Example 7.32 on page 362 provide
not only a conversion between a European put and a European call but also a
perfect hedged portfolio36 = S + P C for
S ( t ) e q ( T t ) + P ( t ) C ( t ) = K e r ( T t ) ,
from (7.11) on page 362, with dividend yield rate q 0, of which a proof was
given by using option terminal payoff diagrams.
In this section, we provide a different proof of the put-call parity by using
the law of one price.
We denote by C E (t) and P E (t) the price of a European call and the price of a
European put options at time t [0, T ], where time 0 is current date and both
the call and put are on the same underlying asset and have the same expiration
date T and strike price K. Let S(t) be the price of the asset at time t [0, T ] and
let r be the risk-free interest rate compounded continuously.
Theorem 7.3. (Put-Call Parity for European Options) The current price relation
between European put and call options on the same asset with annual dividend yield
q(see page 341) and with the same expiration T and strike price K is given by
C E (0) + K erT = P E (0) + S(0) eqT , (7.12)
where r is the risk-free interest rate.
Proof. We consider two portfolios A and B, denoted by A and B , respec-

tively, having positions established at time 0 as indicated below:
A : long 1 European call, and long zero-coupon bond with KerT .
B : long 1 European put, and long eqT units of the asset.
We denote by A (t) the time-t value of portfolio A and B (t) the time-t value
of portfolio B and compute A ( T ) and B ( T ):
A ( T ) = C E ( T ) + Ke0 = max{S( T ) K, 0} + K = max{S( T ), K }

B ( T ) = P E ( T ) + e0 S( T ) = max{K S( T ), 0} + S( T ) = max{K, S( T )}.
Since the two portfolios have identical values at a future time, it follows from
the definition of arbitrage and the law of one price that they must have identi-
cal values today:
A (0) = B (0),
36 A perfect hedged portfolio is a portfolio with complete risk elimination.

7.5 Options 369
C E (0) + KerT = P E (0) + S(0) eqT .

Replacing 0 by t < T in (7.12), we obtain (7.11) again
C E (t) P E (t) = S(t) eq( T t) Ker( T t) .
7.5.7 Relation Among Put, Call, and Forward
Observe that a superposition of Fig. (3) and Fig. (7) on page 360 yields Fig. (1)
on page 340. It follows that at time T, C P = F T ( T ), where we assume that
the put, call, and forward are on the same underlier and expiration T and with
the same strike K (thus FT (0) = K; also see page 344). Under the law of one
price, we obtain an equivalent form of the put-call parity in (7.12):
C E ( t ) P E ( t ) = F T ( t ), 0 t T, (7.13)
where the forward value FT (t) = ( FT (t) K ) er( T t) by (7.6) on page 344. In
words, (7.13) says that the difference between the current values of the Euro-
pean call and put is the current value of the forward.
Example 7.43. Using the forward price formula FT (t) = S(t)e(rq)( T t) , the
put-call parity (7.13) can also be derived by straightforward computation as
follows:
( FT (t) K )er( T t) = (S(t)e(rq)( T t) K )er( T t)

= S(t) eq( T t) Ker( T t) = C E (t) P E (t),
where the last equal sign holds due to (7.12).

The new version of put-call parity expressed by (7.13) says that
The payoff from long a forward equals the payoff from simultaneously
long a call and short a put provided the forward, call, and put are on the
same underlier and with the same strike and expiration.
Equivalently,
A forward can be replicated by simultaneously long a call and short a put
on the same underlier and with the same strike and expiration as the for-
ward.
Example 7.44. A call can be replicated by a forward and a put, and a put can
be replicated by a forward and a call.

7.5.8 Intrinsic Value and Time Value
An option valuation is a procedure for assigning a market value to an option.

Different models generate different procedures, which, in turn, produce
different valuations for different purposes. The procedure generated by the
Black-Scholes-Merton model in the next chapter is a theory-based approach
for calculating the fair value of an option. The procedure to be introduced in
the current section is based on a simple decomposition of the market value of
an option into two components, one of which is considered to be the value of
that option that most professional trading is based upon.
We begin with the following observations.
Observation 1.
max{( FT (t) K )er( T t) , 0} = er( T t) max{ FT (t) K, 0}

.
r( T t) FT (t) K if FT (t) > K
= e
0 if FT (t) K
.
er( T t) ( FT (t) K ) if FT (t) > K
=
0 if FT (t) K
.
S(t) er( T t) K if S(t) > er( T t) K
=
0 if S(t) er( T t) K
.
(S(t) K ) + K (1 er( T t) ) if S(t) > er( T t) K
=
0 if S(t) er( T t) K.
Keeping Remark 7.9 (see page 364) in mind, the expression right after the last
equal sign represents the degree to which the call option is in the money at
time t, as does the expression er( T t) max{ FT (t) K, 0}.
Observation 2.
max{( FT (t) K )er( T t) , 0} = er( T t) max{K FT (t), 0}

.
(S(t) er( T t) K ) if S(t) < er( T t) K
=
0 if S(t) er( T t) K
represents the degree to which the put option is in the money at time t.
Observation 3.
er( T t) max{ FT (t) K, 0} er( T t) max{K FT (t), 0}

= ( FT (t) K )er( T t) = C E (t) P E (t).
Now, we define the intrinsic value of an option at time t to be the degree to

which it is in the money at t and define the time value of a call option at time t,
7.5 Options 371
written TVc (t), by
TVc (t) = C E (t) er( T t) max{ FT (t) K, 0}. (7.14)
Then, a call option price can be expressed by a sum of its intrinsic value and
its time value:
C E (t) = er( T t) max{ FT (t) K, 0} + TVc (t).
Similarly, we define the time value of a put option at time t, written TV p (t), by
TV p (t) = P E (t) er( T t) max{K FT (t), 0}. (7.15)
Then, a put option price can be expressed by a sum of its intrinsic value and
its time value:
P E (t) = er( T t) max{K FT (t), 0} + TV p (t).
Since definitions (7.14) and (7.15) and Observation 3 yield
TVc (t) TV p (t) = C E (t) P E (t) (C E (t) P E (t)) 0, t [0, T ],
we obtain TVc (t) = TV p (t) during the life of the call and put. This result allows
us to simply denote by TV(t) the time value of a European option and write
C E (t) = er( T t) max{ FT (t) K, 0} + TV(t),

P E (t) = er( T t) max{K FT (t), 0} + TV(t).
In words,
Option Value = Intrinsic Value + Time Value.
Notice that an out-of-the-money option has time value only. As a result, the
value of an out-of-the-money option erodes quickly with time as it gets closer
to its expiration.
Because out-of-the-money options have time values only, they are signifi-
cantly cheaper and offer great leverage and, therefore, have better liquidity
(are more actively traded). For these reasons, most professional option traders
trade the time value only with confidence in turning time value decay into
potential profits; for trading purpose, the time value of an option is where the
professional traders see the value of an option.
Remark 7.10.
1. Time value is subject to several factors, primarily time to expiration and
implied volatility. The latter concept will be discussed in Chapter 8.
2. The rate at which time value decays is represented by , one of the Greeks to
be introduced along with Black-Scholes-Merton model, again in Chapter 8.

Example 7.45. SPY is trading at $213 on May 18, 2015. Call options with strike
prices below $213 are in-the-money calls. Call options with strike prices above
$213 are out-of-the-money calls. Call options with strike prices equal $213 are
at-the-money calls. Given the following option information,
Expiration month May June July
Premium of SPY 212.5 call $1.28 $2.92 $3.67
Time value of SPY 212.5 call $0.75 $2.42 $3.17
observe the time value decay in action: the time value drops from June to May
much faster than that from July to June.

Example 7.46. (Trading on Time Value) Option writers attempt to benefit from
the time value decay. They collect time value premiums paid by option buyers.
Such premiums can become steady cash flows if the underlying security is
stationary.

7.5.9 Some General Relations of Options
We assume that the underlying asset pays a constant, continuous proportional

annual dividend yield rate q that is continuously reinvested to buy more units
of the asset and that holding the asset neither incurs cost of carry nor provides
other convenience yields in the interval [0, T ] where time 0 is current date. We
also assume that r is the risk-free interest rate compounded continuously.
We denote by C A (t) and C E (t) the American call price and the European
call price at time t, where both the American call and the European call are on
the same asset and with the same terms.
Similarly, we denote by P A (t) and P E (t) the American put price and Euro-
pean put price at time t, where both the American put and the European put
are on the same asset and with the same terms.
Assuming that notation S(t), T, and K are defined as usual, we have the
following property and provide a partial proof.
Proposition 7.1. The following relations of options hold.
1. S(0) C A (0) C E (0) max{S(0) eqT K erT , 0}.

2. K P A (0) P E (0) max{K erT S(0) eqT , 0}.
7.5 Options 373
Proof. We provide a proof for the first set of inequalities. A proof of the second
set is left as an exercise for the reader.
Step 1. If S(0) < C A (0), then construct a portfolio at time 0, by shorting 1
American call and using the proceeds to immediately long 1 unit of the un-
derlier. We obtain (0) = C A (0) S(0) > 0, an immediate profit with the cash
amount a = C A (0) S(0) at time 0. Thus, ( T ) > 0 is guaranteed regardless
of the fluctuations of the asset price, even if the cash amount $a is kept under
mattress for taking a short position on a covered call means that the pos-
session of the asset can always cover the exercise made by the buyer in case it
happens. Therefore, is an arbitrage!
Step 2. The ability to exercise an American option at any time prior to or at
expiration makes American options more flexible than European options; thus,
C A (0) C E (0).
Step 3. Applying the put-call parity (7.12) and P E (0) 0, we obtain
C E (0) = P E (0) + S(0) eqT K erT

S(0) eqT K erT .
It follows from C E (0) 0 that C E (0) max{S(0) eqT K erT , 0}.

Proposition 7.2. One should not prematurely exercise an American call on a

nondividend-paying asset if the risk-free interest rate r > 0. That is,
C A (0) = C E (0).
Proof. It follows from Proposition 7.1, 1 with q = 0 and r > 0, that
C A (0) max{S(0) K erT , 0} > S(0) K.
Note that S(0) K is the payoff if the call is exercised currently.

7.5.10 Put-Call Parity for American Options
We denote by C A (t) and P A (t) the price of the American call and the price
of the American put options at time t [0, T ], where time 0 is current date
and both the call and put are on the same underlying asset and have the same
expiration date T and strike price K. Let S(t) be the price of the asset at time
t [0, T ] and let r be the risk-free interest rate compounded continuously.
Theorem 7.4. (Put-Call Parity for American Options) The current price relation
between American put and call options on the same asset with annual dividend yield
q(see page 341) and with the same expiration T and strike price K is given by
S(0)eqT K C A (0) P A (0) S(0) KerT , (7.16)
where r is the risk-free interest rate.
Proof.
Step 1. To prove the inequality C A (0) P A (0) S(0) KerT , we construct
two portfolios A and B, denoted by A and B , respectively:
A : long 1 American call and long zero-coupon bond with KerT at rate r.
B : long 1 American put and long 1 unit of the underlying asset.
We denote by A (t) the value of portfolio A and B (t) the value of portfolio
B at time t and by DPV the time-0 value, or present value, of the dividends.
Notice that the American call cannot be exercised early because there is not
enough cash to buy 1 unit of the asset until time T and
A ( T ) = max{S( T ) K, 0} + K = max{S( T ), K }.
With the possibility that the American put could be exercised at t [0, T ],
.
K + DPV ert if S(t) < K
B ( t) = rt
= max{S(t), K } + DPV ert ,
S(t) + DPV e if S(t) K
which implies that
B ( T ) = max{S( T ), K } + DPV erT A ( T ).
It follows from the definition of arbitrage and the law of one price that
A (0) B (0).
Therefore, the desired inequality C A (0) + KerT P A (0) + S(0) is obtained.

Step 2. To show the other inequality
S(0)eqT K C A (0) P A (0), (7.17)
we write DPV = S(0)(1 eqT ) for DPV eqT = S(0)(eqT 1) being the total
dividends of 1 unit of the underlier over the interval [0, T ]. Then
S(0)eqT = S(0) DPV
implies that (7.17) is equivalent to
C A (0) + K + DPV P A (0) + S(0). (7.18)
We construct two portfolios X and Y, denoted by X and Y , respectively:

X : long 1 European call and long zero-coupon bond with K + DPV at r.
Y : long 1 American put and 1 unit of the asset simultaneously.
7.5 Options 375
Since the European call cannot be exercised early, we only need to consider
whether the American put is exercised early.
If the American put is not exercised early, then
Y ( T ) = max{K S( T ), 0} + S( T ) + DPV erT

= max{K, S( T )} + DPV erT ,
X ( T ) = max{S(t) K, 0} + DPV erT + KerT
= max{S(t) K, 0} + K + DPV erT + KerT K
= max{S( T ), K } + DPV erT + K (erT 1)
> Y ( T ).
If the put is exercised early at time t [0, T ) (i.e., sell stock to receive K), then
Y (t) K + DPV ert

DPV ert + Kert X (t).
Thus, regardless of the early exercise, X (0) Y (0), which is equivalent to
C E (0) + K + DPV P A (0) + S(0).
Since C A (0) C E (0), we obtain
C A (0) + K + DPV C E (0) + K + DPV P A (0) + S(0),
which implies (7.18).

Replacing 0 by t < T in (7.16), we obtain
S(t)eq( T t) K C A (t) P A (t) S(t) Ker( T t) .
Example 7.47. (Intuition of the Put-Call Parity for American Options) To in-
terpret the put-call parity (bounds) for American options geometrically, we let
x = S (0) and y = C A (0) P A (0)
and visualize the region in the xy-plane, which is bounded by two straight
lines,
L1 : y = eqT x K and L2 : y = x KerT .
Notice that for nondividend-paying underliers, these two lines are parallel.
The relation between S(0) and C A (0) PA (0) has a geometric interpretation
in terms of the points in the region bounded by L1 and L2 as shown below:
put-call parity for American options when q = 0

O
C (0) P (0)
A A
L2

L1

/
Ke rT
0

K S(0)

The geometric interpretation of the put-call parity for American option in the
case of q > 0 is left as an exercise for the reader. (Hint: sketch a graph!)

Example 7.48. Given the current price of a stock and the current price of a put
option on the stock with expiration T and strike K along with the stocks divi-
dend yield q and risk-free rate r, find the upper and the lower price bounds for
the American call option with the same expiration and strike on the stock.
Solution. The lower bound is P A (0) + S(0)eqT K.

The upper bound is P A (0) + S(0) KerT .

7.5.11 Boundary Conditions for European Options
The next proposition gives the boundary conditions on calls and puts; they are
needed later in our study of the Black-Scholes-Merton p.d.e.
Again, we let C E (S, t) = C E (S(t), t) and P E (S, t) = P E (S(t), t) for t T.
Boundary conditions for C E (S, t) and P E (S, t) are applied for S 0 and S .
The next proposition can be established by applying Proposition 7.1 (see
page 372) and the put-call parity (see (7.12) on page 368) along with following
observations:
1. C E (S, t) 0 as S 0 since the call is unlikely to be exercised when the
underliers price is sufficiently small.
2. C E (S, t) Seq( T t) Ker ( T t) as S since the call is likely to be exer-
cised when the underliers price is sufficiently large.
Proposition 7.3. (European Call and Put Boundary Conditions)
1. limS0 C E (S, t) = 0 and limS C E (S, t) = limS S eq ( T t) .
2. limS0 P E (S, t) = Ker ( T t) and limS P E (S, t) = 0.
A proof of Proposition 7.3, 2, can also be done directly (without applying the
put-call parity). A geometric interpretation of the properties can be done by
sketching the graphs of C E (S, t) and P E (S, t) against S.
presented in this chapter, e.g., [2, 3, 5, 6, 7, 9, 10, 12, 13, 14, 15, 16, 17, 18, 19].
7.6 Exercises 377
7.6 Exercises
Unless stated otherwise, an option is either American or European.
7.1. Is a forward a contingent claim? Is an option a contingent claim?

7.2. There are mainly three types of derivative traders: hedgers, speculators,
and arbitrageurs. What are their definitions?
7.3. Sketch the terminal payoff diagram of a forward with expiration T and for-
ward price K. If you short this forward, what is the terminal payoff diagram?
7.4. What are the key features of futures that differ from forwards?
7.5. If you believe that the market price of a stock will stay at approximately
the same price for a period of time, can you still make money from the stock if
your hunch is correct? Explain your answer.
7.6. What is the possible maximum gain or loss if you sell a call?
7.7. What is the possible maximum gain or loss if you buy a call?
7.8. Consider the call options given in Example 7.45 (see page 372). Are they
all in the money? Identify their intrinsic value(s).
7.9. If a call is in the money sufficiently close to the expiration date, then the
call price will rise dollar for dollar with the stock price. Agree or disagree?
Explain.
7.10. An at-the-money American call with a strike price of $80 is being sold for
$200. Assume that the stock goes up to $84 per share on the day of expiration.
a) If you bought the option, what is your return from exercising the call and liq-
uidating your stock position? If you did not buy the option, but had bought
100 shares of the stock in the market at $80 per share and then sold them on
the options expiration date at $84 per share, what would be your return?
Do the two scenarios have equivalent gain/loss?
b) If you do not exercise the option, what is your approximate return from
selling the call right before expiration?
c) Which would you then prefer? Exercise the call or sell the call?
7.11. You sell an American call on 1 round lot of a stock at $40 per share.
A month later, the market value of that stock is $46 per share. If the buyer
exercises the option, you will be obligated to deliver 100 shares at $6 below
current market value.
a) If you own those shares, what is your gain/loss from settling the position?
b) If you had naked short sold the American call, what is your gain/loss from
settling the position?
7.12. You paid $300 for an American call on a stock several months ago. It will
expire next month and is now worth only $100. What are the feasible actions
that you can take? What are the consequences of your actions?
7.13. Argue how American put buying/selling works.
a) How American put buying works.

Buyers of a put expect the underlying stock to fall in value. In each of the
following cases, what are the feasible actions of a buyer and their outcomes
in terms of monetary gain or loss?
Case 1: The price of the stock increases after the buyer purchased the put.
Case 2: The price of the stock almost does not change.
Case 3: The price of the stock decreases and the exercise price of the put is
higher than the price of the stock at the expiration date.
b) How American put selling works.
Under the plan of selling puts, you grant someone else the right to sell 100
shares to you at the exercise price. At the time you sell, you receive a pre-
mium. Like the call seller, you do not have much control over the outcome
of your investment since the buyer will decide whether to exercise the put
you sold him. In each of the following cases, what are the feasible actions of
a put seller and their monetary results?
Case 1: The price of the stock increases.
Case 2: The price of the stock remains stable till the expiration date.
Case 3: The price of the stock decreases and the put is in the money at the
expiration.
7.14. Hedge/hedging is a strategy used to offset investment risk. A perfect

hedge is one eliminating the possibility of future gain or loss.
A stockholder worried about declining stock prices, for instance, can hedge
his or her holdings by buying a put on the stock or by selling a call.
a) How does each case work?
b) Which type of hedging is preferable?
7.15. (Call Time Spread Bearish) Recall Example 1. Given XYZ 40 call price
table:
Expiration Nov Dec Jan
Premium 2 3 5
If one expected XYZ stock to decline, one might establish a bear spread by
taking a position opposite of a bullish one.
7.6 Exercises 379
Make two transactions to establish a spread in the hope of making a profit if

XYZ stocks price declines and of limiting the loss if the expectation turns out
to be wrong.
7.16. (Price Put Spread Bearish) Open a bear spread by using the following
puts:
XYZ Dec 40 Put at 3

XYZ Dec. 45 Put at 7
in the hope of making a profit if XYZ stock declines in price. What is the pos-
sible maximum gain or loss? Justify your answers.
7.17. (Forward Price and Arbitrage) Suppose that the current spot price of a
continually paying dividend asset is $222, the interest rate is r = 3% and the
dividend yield is q = 2%.
a) What are the one-month and eight-month forward prices for the asset in an
arbitrage-free market?
b) Let be a portfolio on time interval [0, T ] consisting of three positions start-
ing from time 0: borrow $222 at the rate 3%, long 1 unit of the asset, and
short the three-month forward at FT (0) = $222.56. Is an arbitrage portfo-
lio? If your answer is no, show a proof. If your answer is yes, explain how
you can make a profit by taking the arbitrage opportunity.
7.18. (Forward Value) Given K0 = $222, FT (t) = $252, T t = 6 months, and

r = 3%, determine the value of the forward in an arbitrage-free market and
interpret the result.
7.19. (Swaps) Assume that the terms of the swap contract include the follow-
ing:
a) The notional principal is one million dollars,
b) The life of the contract is 2 years,
c) A pays B three-month LIBOR + 0.2%,
d) B pays A 1.5% fixed,
e) There is an exchange of payments every 3 months from the initialization.
Given the LIBOR rates in the table below, calculate both the floating cash flow
and fixed cash flow of the swap.
Period LIBOR Payment from A to B Payment from B to A

0 1%
1 0.8%
2 1%
3 1.2%
4 1.06%
5 1.1%
6 1.2%
7 1.4%
8
7.20. (Swaps) Suppose that both companies X and Y need to borrow US dollars
and that company X would like to borrow at a fixed rate, whereas company
Y would like to borrow at a floating rate. If X can borrow at 6.00% fixed and
LIBOR + 0.60% floating, and Y can borrow at 5.00% fixed and LIBOR + 0.20%
floating, what is the range of possible cost savings that company X can real-
ize through an interest rate swap with company Y? Use an example of swap
mechanics to demonstrate how a cost saving to be done for either company
(ignoring credit risk differences).
7.21. Identify the range of profit, loss, and break-even point as outcomes of the
corresponding strangle strategy given in Example 7.39 (page 366).
7.22. Use a formula to express the terminal payoff of each spread strategy given
in Example 7.40 (page 366).
7.23. Use a formula to express the terminal payoff of the butterfly spread given
in Example 7.42 (page 367).
7.24. (Arbitrage)
a) Suppose that the price of a stock at time t, denoted by S(t), is modeled by a
one-step binomial tree over the time period [0, T ] with
.
Sb with probability p
S( T ) =
Sa with probability 1 p,
where Sb > Sa .
Show that Sb > S(0) > Sa is a necessary condition for a non-arbitrage oppor-
tunity for any investor (assuming r f = 0).
References 381
b) Show that there exists a (risk-neutral) probability p > 0 holding the equation
S(0) = pSb + (1 p)Sa .
7.25. (Forward Value) Prove (7.5) (page 344).
7.26. Show that if the price of the underlier of a forward contract follows a geo-
metric Brownian motion, so does the forward price process (see Example 7.15).
7.27. (General Property of Options) Let C A (0, K, T1 ) and C A (0, K, T2 ) be two

American call options on the same terms except that they have different expi-
rations with T1 < T2 . Show that C A (0, K, T1 ) C A (0, K, T2 ).
7.28. Establish the following bounds for American puts on nondividend-

paying underliers:
K P A (0) max{K S(0), 0}.
7.29. Establish the following relation between American and European puts on
the same nondividend-paying underlier and with the same expiration T and
strike K:
P A (0) P E (0).
7.30. Is it never advantageous to exercise early an American put on a

nondividend-paying stock? Justify your answer.
7.31. Establish the following put-call parity bounds for American options:
S(t0 ) eq K C A (t0 ) P A (t0 ) S(t0 ) K er ,
where = T t0 .
References
[1] Anderson, G., Kercheval, A.: Lectures on Financial Mathematics: Discrete

Asset Pricing. Morgan and Claypool, San Rafael (2010)
[2] Bingham, N.H., Kiesel, R.: Risk-NeutralValuation: Pricing and Hedging
of Financial Derivatives. Springer Science and Business Media, New York
(2004)
[3] Bjork, T.: Arbitrage Theory in Continuous Time. Oxford University Press,
Oxford (2009)
[4] Cox, J.C., Ingersoll, J.E., Ross, S.A.: The relation between forward prices
and futures prices. J. Financ. Econ. 9. (1981) [Columbia Business School,
New York]
[5] Delbaen, F., Schachermayer, W.: A general version of the fundamental
theorem of asset pricing. Math. Ann. 300, 463(1994)
[6] Delbaen, F., Schachermayer, W.: The fundamental theorem of asset pricing
for unbounded stochastic processes. Math. Ann. 312, 215(1998)
[7] Delbaen, F., Schachermayer, W.: The Mathematics of Arbitrage. Springer,
Berlin/Heidelberg (2006)
[8] Duffie, D., Stanton, R.: Pricing continuously resettled contingent claims.
J. Econ. Dyn. Control 16, 561573 (1992)
[9] Epps, T.W.: Pricing Derivative Securities. World Scientific, River Edge
(2007)
[10] Harrison, J., Pliska, S.: Martingales and stochastic integrals in the theory
of continuous trading. Stoch. Process. Appl. 11, 215260 (1981)
[11] Hull, J.C.: Options, Futures, and Other Derivatives. Pearson Princeton
Hall, Upper Saddle River (2015)
[12] Jacod, J., Protter, P.: Probability Essentials. Springer, Berlin/Heidelberg
(2004)
[13] Kolb, R.W.: Financial Derivatives. New York Institute of Finance, New
York (1993)
[14] Korn, R., Korn, E.: Option Pricing and Portfolio Optimization. American
[15] Kreps, D.M.: Arbitrage and equilibrium in economies with infinitely
many commodities. J. Math. Econ. 8(1), 15(1981)
[16] Musiela, M., Rutkowsk, M.: Martingale Methods in Financial Modelling.
[17] Reilly, F.K., Brown, K.C.: Investment Analysis and Portfolio Management.
[18] Whaley, R.: Derivative: Markets, Valuation, and Risk Management. Wiley,
Hoboken (2006)
(1995)
Chapter 8
The BSM Model and European Option Pricing
The Black-Scholes-Merton (BSM) model, also known as the Black-Scholes model, is

one of the pillars of finance, providing a powerful theoretical framework that
is widely applicable in financial engineering and corporate finance.
The BSM model is a partial differential equation (p.d.e.) approach to pricing
derivatives, not just call and put options. The model is presented in Section 8.1,
which includes the marketplace assumptions, self-financing, replicating port-
folios, and a derivation of the BSM p.d.e. using such portfolios. Applications
of the BSM model to the special case of European calls and puts are given
in Section 8.2. An outline is presented for how to solve the BSM p.d.e. for a
European call price. The associated put price is obtained via put-call parity,
and the deltas and rate of change relative to strike price are explored for Eu-
ropean calls and puts. A corporate finance application of the BSM model is
given in Section 8.3, where warrants are priced. Section 8.4 then takes up an
alternative approach to pricing derivatives. It prices derivatives using a risk-
neutral probability measure, i.e., by discounting the conditional expectation of
future prices at the risk-free rate minus the dividend yield rate. The alterna-
tive approach provides a deep link between the BSM p.d.e. and existence of
risk-neutral measures. Exploration of the latter gives rise to the fundamental
theorems of asset pricing. These results show how the no-arbitrage condition
and market completeness are related to the existence and uniqueness of risk-
neutral probability measures.
The pricing of derivatives can be done not only in a continuous-time set-
ting with a continuum of possible prices of the underlying security but also in
discrete time with discrete possible underlier prices. The latter approach is il-
lustrated in Section 8.5 using binomial trees, which provide excellent intuition
into the pricing process. Using a European call, we give a discrete-time ver-
sion of the self-financing, replicating portfolio of the BSM model to price the
call as well as a discrete-time version of the risk-neutral probability measure
approach to pricing.

384 8 The BSM Model and European Option Pricing
A core challenge is managing risk in a portfolio when the diversification

approach of the Markowitz model does not apply. One of the significant appli-
cations of the BSM model is managing portfolio risk through what is known as
delta and gamma hedging. Section 8.6 details the theoretical framework for dy-
namical delta hedging, while Section 8.6.2 illustrates how to use the framework
to manage the risk from selling European call options. Section 8.7 extends these
ideas to a portfolio of options, explaining how to make the portfolio delta-
and gamma-neutral, i.e., stable against movements in the underlying security
price.
In Section 8.8, we investigate how the assumptions of the BSM model stand
up against data using the S&P 500 and an IBM European call option for il-
lustration. Contrary to the specific BSM assumption of a geometric Brownian
motion underlier, the underlier price can exhibit a jump discontinuity, and its
log return can show skewness and kurtosis. In addition, the implied volatility
of a European call when plotted as a function of strike price (and time to expi-
ration) is not constant as predicted by the BSM model, but is curved and can
even reveal shapes that look like a volatility smile.
The issue then is how to amend the BSM model. Section 8.9 gives a detailed
introduction to the Merton jump diffusion (MJD) model as an example ad-
dressing the above concerns with the BSM model. The MJD model extends the
underlier price process to one that is a mixture of geometric Brownian motion
and jump discontinuities. We show that adding jump discontinuities to geo-
metric Brownian motion produces log returns with skewness and kurtosis. In
this sense, the current price of a European call with a jump-diffusion underlier
is viewed as a better approximation to the volatility smiles in market data. We
shall also see that the MJD model is in an arbitrage-free, incomplete market.
This means that there is not a unique price for a derivative.
The chapter ends in Section 8.10 with a glimpse beyond the theory of
derivatives presented in the book. We highlight that though the Merton jump-
diffusion model incorporates several features not present in the BSM model, it
does assume that the volatility of the underliers price is deterministic, while
data supports a stochastic volatility. This brings us to issues currently of inter-
est in the research on derivatives: pricing derivatives in incomplete markets
with underliers having stochastic volatility with jumps.
8.1 The BSM Model
The Black-Scholes-Merton (BSM) model gives a p.d.e. (i.e., partial differential

equation) approach to pricing a derivative. We shall overview the basic as-
sumptions of the model and derive the BSM p.d.e. It is important to empha-
size that the BSM p.d.e. applies not only to European call options but also to
8.1 The BSM Model 385
a rather general class of derivative contracts. Our approach is in parallel with

the treatments by Bjork [5], Epps [15], and Privault [34].
Notation (Current time). The current time is typically set at time 0. In this case,
since we are considering risky securities, their prices at a future time t > 0 are
random. However, in certain settings, we shall need to treat the current time
as advancing and shall denote it by the variable t. For such situations, assume
that time t + dt is in the future relative to t. Overall, the context in which the
current time is intended should be clear to the reader.
8.1.1 Marketplace Assumptions
The marketplace is assumed to be idealized, i.e., the following holds:

Equilibrium: supply equals demand.
No arbitrage: no arbitrage opportunity exists, i.e., there is no costless, risk-
less profit. See Section 7.1.4 (page 334) for a precise treatment.
Access to information: there is immediate availability of accurate information
on all securities.
Efficiency: a securitys price adjusts instantly to new information, so its cur-
rent price reflects all known information concerning the security, which
includes information about the past and expected future behavior of the
security.
Liquidity: any number of unitseven a fractional amountof a security
can be bought and sold instantly.
No transaction costs and no bid-ask spreads: transaction costs, which include
fees and margin account requirements, are ignored.
No taxes: this includes no taxes on capital gains, interest, or dividends.
Borrowing/lending: borrowing and lending are at the risk-free rate r, and
there is no limit to how much one can borrow or lend.
Short selling: short selling is allowed without restriction; in particular, the
funds from a short sale can be used immediately to trade.
Remark 8.1. (Number of Days in a Year) For simplicity, a 365-day year is used
in this chapter (e.g., McDonald [27]). However, in Chapter 5 (see page 223),
when estimating a securitys historical volatility, the non-trading days were ig-
nored, and a year was treated as a trading year, which consists of 252 trading
days; see Hull [22, p. 328] for more. On the other hand, a daily-compounded
loan will compound on the non-trading days during the term of the loan, so
such days are not omitted in this case.

8.1.2 Money Market Account and the Underlier Model
The BSM model prices a derivative using two securities that act as fundamen-
tal drivers: a money market account and the security serving as underlier of
the derivative.
Define a money market account to be a riskless security that has a value B0
at the current time 0 and grows by continuous compounding at the risk-free
rate r. Its value at a general time t is then
Bt = B0 er t ,
where B0 is an arbitrary constant. By default, we choose B0 = 1 (one dollar).

The instantaneous change is
dBt = r Bt dt ( B0 = 1). (8.1)
Remark 8.2. The value Bt is often used as a numeraire for an asset, which means
that the value of the asset at t can be expressed as a multiple of Bt , i.e., in units
of Bt .
The underlying security of a derivative is assumed to pay a continuous divi-

dend at yield rate q and to be a geometric Brownian motion with drift parameter
RW and volatility parameter :
St = S0 eRW t + Bt (0 t T),
where S0 is the known current price of the underlying security and
2
RW = m q .
2
The instantaneous change is given by the following s.d.e.:
dSt = (m q) St dt + St dBt . (8.2)
The underliers cash dividend will be continuously reinvested either in the

underlying security or the money market account.1 Our default assumption is
to reinvest the cash dividend in the underlier. In this case, we must distinguish
between the underliers ex-dividend (without dividend) unit price St , which is
the market price of the security at t, and its cum-dividend (with dividend) unit
price Stc ; see page 18. Starting with 1 unit of the underlier at 0, the 1 unit will
increase to eq t units at time t. The cum-dividend unit price of the security will
then be (see (2.28) on page 31)
1 Reinvesting the cash dividend in the security is then simply the investor acquiring more units of the
security.
Stc = St eq t ( t 0).
In other words, as time advances the market price St = eq t Stc is a continu-

ous downward adjustment of the cum-dividend price. This is because when a
security pays out a cash dividend, value is flowing out of the security.
Finally, we restate Itos formula2 for easy reference. Suppose that { Xt }t0 is
an Ito process, i.e.,
dXt = u( Xt , t) dt + v( Xt , t) dB(t), 0 t T, (8.3)
where u( x, t) and v( x, t) are deterministic functions and { Xt }t0 is an adapted

process (e.g., a stochastic process that is a deterministic function of standard
Brownian motion). Assume that Yt = g( Xt , t), where g( x, t) is a deterministic
function that is twice continuously differentiable in x and once continuously
differentiable in t. Then:

1 2 2 g g g
dYt = v ( Xt , t ) 2 ( Xt , t ) + u ( Xt , t ) ( Xt , t ) + ( Xt , t ) dt
2 x x t
g
+ v ( Xt , t ) ( Xt , t) dB(t). (8.4)
x
For example, to obtain the instantaneous change in the cum-dividend price,
apply (8.4) to Stc = St eq t with g( x, t) = x eq t , which gives the s.d.e. for cum-
dividend security prices (Exercise 8.14):
dStc = m Stc dt + Stc dBt . (8.5)
Finally, in keeping with our intuitive approach, we abide by the following:
Unless stated to the contrary, we shall freely apply Itos formula

and stochastic integration when necessary and so assume that all
processes considered satisfy the properties that allow for such ac-
tions.
For instance, a stochastic process { Xt }t0 is always assumed to be adapted to
standard Brownian motion, meaning 2Xt is a deterministic function of B for
t
all 0 t. When integrating, say, 0 Xv dBv , we also always assume that
2t
the process is square integrable, i.e., 0 E ( Xv2 ) dv < . See Bjork [5, Chap 4],
Mikosch [32, Chap. 2], and Privault [34, Chaps. 46] for an introduction and
Elliot and Kopp [14, Chaps. 6, 7] and Korn and Korn [24, Chaps. 2, 3] for an
advanced treatment.
2 See (6.57) on page 304.

8.1.3 Self-Financing, Replicating Portfolio
In this section, we shall employ a certain trading strategy (nt , bt )i.e., take
a position with nt units of the underlying security and bt units of the money
market accountto construct a portfolio whose value replicates the price of a
derivative in a self-financing manner.
Before carrying out this strategy, we fix a derivative and assume that its price
based on 1 unit of its underlying security is a stochastic process { f t }0t T that
is a deterministic function of the underliers market price (i.e., ex-dividend
price) St and time t,
f t = f ( St , t ), (8.6)
where f ( x, t) is assumed to be at least twice continuously differentiable in x
and once continuously differentiable in t for x > 0 and 0 < t < T. In particular,
the current price of the derivative is f (S0 , 0). We assume that the derivative does
not pay a cash dividend.
Now, assume that we have an initial capital Vt at t. With this money, create a
portfolio using a trading strategy (nt , bt ), i.e., hold nt units of the cum-dividend
underlying security and bt units of the money market account. Think of (nt , bt )
as a stochastic process with values evolving in R2 . The value of the portfolio
at t is
Vt = nt Stc + bt Bt .
As t advances and (nt , bt ) evolves, a key issue will be how to pay for these
changes in the number of units of the underlying security and the money mar-
ket account.
At time t + dt, suppose that (nt+dt , bt+dt ) replicates the price of the deriva-
tive:
Vt+dt = nt+dt Stc+dt + bt+dt Bt+dt = f (St+dt , t + dt).
After the initial capital at t, the portfolio is self-financing if the trading strategy
(nt+dt , bt+dt ) at t + dt is funded without withdrawing or adding any external
funds to the portfolio. In other words, during the transition from time t to
t + dt, the value of the portfolio at t + dt arises only from an increase, decrease,
or neither in the values of the underlying security and/or the money market
account. The original strategy (nt , bt ), along with the possibly new values of
the underlier and money market account at t + dt, must then fund the portfo-
lios replication of the derivative:
Vt+dt = nt Stc+dt + bt Bt+dt .
Explicitly, the self-financing condition is then
nt Stc+dt + bt Bt+dt = nt+dt Stc+dt + bt+dt Bt+dt ,

which can be written more compactly as3

c
dnt St+dt + dbt Bt+dt = 0. (8.7)
Equation (8.7) is equivalent to (Exercise 8.15)
dVt = nt dStc + bt dBt . (8.8)
If the self-financing portfolio replicates the price of the derivative at t + dt,

then by the law of one price, the portfolios value at t also replicates the price
of the derivative at t:
Vt = nt Stc + bt Bt = f (St , t). (8.9)
Let us then determine the trading strategy (nt , bt ) that will make (8.9) possible.
The self-financing condition (8.8) and replicating condition (8.9) will allow us
to determine the desired (nt , bt ). By (8.9),
f (St , t) nt Stc
bt = .
Bt
Substituting into (8.8) and employing (8.1) and (8.5) give
dBt
dVt = nt dStc + f (St , t) nt Stc
B
t
= r f (St , t) + nt (m r) Stc dt + nt Stc dBt .
(8.10)
On the other hand, Itos formula (8.4) yields

1 2 2 2 f f f
dVt = d f (St , t) = St 2 ( St , t ) + ( m q ) St ( St , t ) + (St , t) dt
2 x x t
f
+ St (St , t) dBt .
x
(8.11)
Now, an Ito process has a unique representation; see, for example, Korn and
Korn [24, p. 77]. This means that if
dXt = at dt + bt dBt = at dt + bt dBt ,
where at , bt , at , and bt are functions of standard Brownian motion, then
at = at , bt = bt .
3 Recall: dxt = xt +dt xt .

Equating the dBt coefficients in (8.10) and (8.11), we obtain:

St f
nt = ( St , t ). (8.12)
Stc x
The partial derivative of f in (8.12) is our first encounter with a financial

derivatives Greek, which we define loosely as a partial derivative of f with
respect to x, t, or any of the parameters used in modeling the financial deriva-
tive. Specifically, the partial of f relative to x is called the delta of the financial
derivative and denoted by
f
f ( St , t) = ( St , t ).
x
Delta is perhaps the most popular of the Greeks and will appear many times
in this chapter.
Thus, the price f (St , t) of the derivative at any time t can be replicated by a
self-financing strategy (nt , bt ), where
f (St , t) = nt Stc + bt Bt (8.13)
with

St f ( St , t ) St f ( St , t )
nt = f ( St , t ), bt = . (8.14)
Stc Bt
8.1.4 Derivation of the BSM p.d.e.
Interestingly, the price of a derivative in the BSM model will arise from solving
a partial differential equation (p.d.e.). A second order p.d.e. in two indepen-
dent variables ( x, t) is an equation of the form
2 y 2 y 2 y y
A( x, t) ( x, t ) + B ( x, t ) ( x, t ) + C ( x, t ) ( x, t) + D ( x, t) ( x, t)
x 2 xt t 2 x
y
+ E( x, t) ( x, t) + F ( x, t)y( x, t) = 0, (8.15)
t
where the coefficients and y( x, t) are deterministic functions. The p.d.e. (8.15)
is called
hyperbolic if B2 4AC > 0
parabolic if B2 4AC = 0
elliptic if B2 4AC < 0.
Example 8.1. (Heat Equation) The following p.d.e. is well known in physics:
y 2 y
( x, t) = c 2 ( x, t), (8.16)
t x
where c is a positive constant and y( x, t) designates temperature at position

x and time t. It describes how heat diffuses (spreads) along a rod over time,
where there is no heat source along the rod. The heat Equation (8.16) is also
called a diffusion equation. Since B = C = 0, we have B2 4AC = 0, and so the
heat equation is a parabolic p.d.e.

Returning to the self-financing, replicating portfolio of Section 8.1.3, we al-

ready did all the work needed to obtain the p.d.e. of the BSM model. By the
unique representation of Ito processes (page 389), we can equate the dt coeffi-
cients in (8.10) and (8.11), which along with (8.12) gives
1 2 2 2 f f f
St 2 ( St , t ) + (r q ) St ( St , t ) + (St , t) r f (St , t) = 0. (8.17)
2 x x t
In general, Equation (8.17) is not a deterministic p.d.e. because St is random for
t > 0. However, for each t in (0, T ), the possible values of the lognormal ran-
dom variable St range over (0, ) independent of t. In other words, Equation
(8.17) holds at all points ( x, t) in (0, ) (0, T ). The associated deterministic
p.d.e. is then
1 2 2 2 f f f
x ( x, t) + (r q) x ( x, t) + ( x, t) r f ( x, t) = 0, (8.18)
2 x 2 x t
where 0 < x < and 0 < t < T. Equation (8.18) is called the BSM p.d.e. Because
B = C = 0, we get B2 4AC = 0, i.e., the BSM p.d.e. is parabolic.
Assuming a solution f ( x, t) of the BSM p.d.e. exists, the derivatives price is
given by f (St , t). The issue is that if f ( x, t) is a solution, then we can construct
infinitely many other solutions, which creates an infinity of derivative prices.
For example, if f ( x, t) is a solution, then for every positive real number c > 0
the function f c ( x, t) = f (c x, t) is also a solution (Exercise 8.17). The existence
and uniqueness of a derivative price require additional constraints on f ( x, t):
Final condition: f ( x, T ) is given.
Boundary conditions: f (0, t) is given and the growth behavior of

f ( x, t) as x is given.
Note that the explicit nature of these conditions cannot be stated a priori be-
cause they depend on the contractual structure of the derivative.
For sufficiently well-behaved final and boundary conditions, the theory of parabolic
p.d.e.s yields that the BSM p.d.e. will have a unique solution f ( x, t) and, hence, the
derivative will have a unique price f (St , t). See Korn and Korn [24, Sec. 3.3] and
Miersemann [31, Chap. 6] for more.
8.2 Applications of BSM Pricing to European Calls and Puts
Below we state the final and boundary conditions for European calls and puts
as well as present the associated unique solution of the BSM p.d.e. and the
derivatives price.
Notation. For European calls and puts, write the solutions of the BSM p.d.e.
as C E ( x, t) and P E ( x, t), respectively, rather than f ( x, t).
8.2.1 Solving the BSM p.d.e. for European Calls
For a European call with strike K and expiration at T, the BSM p.d.e. is
1 2 2 2 C E C E C E
x ( x, t) + (r q) x ( x, t) + ( x, t) r C E ( x, t) = 0, (8.19)
2 x 2 x t
the final condition is
C E ( x, T ) = max{ x K, 0}, (8.20)
and boundary conditions are
C E (0, t) = 0 and C E ( x, t) eq ( T t) x as x . (8.21)
Note that (8.20) and (8.21) follow from properties of European calls using no
arbitrage and put-call parity.
We outline how to solve the p.d.e. along the lines of Wilmott, Dewynne, and
Howison [42, Sec. 5.4], leaving the computational details as exercises:
Transform the p.d.e. (8.19) to a form without dividend. It can be shown

(Exercise 8.20) that with the new underlier price process
St = St eq( T t) ,
Equation (8.19) transforms to a form without dividend:
1 2 2 2 C E C E C E
x ( x, t ) + r x ( x, t ) + ( x, t) r C E ( x, t) = 0, (8.22)
2 x2 x t
where x = x eq( T t) and the call price is now viewed as a function of ( x, t).
The associated terminal and boundary conditions are
C E ( x, T ) = max{x K, 0}, C E (0, t) = 0, C E ( x, t) x as x . (8.23)
Note that x = x at t = T.
8.2 Applications of BSM Pricing to European Calls and Puts 393
Convert to the following convenient variables:

x 2 C E ( x, t) r
x = ln , = ( T t), v( x, ) = , k = .
K 2 K (2 /2)
Note that < x < and > 0, where the initial time = 0 corresponds
to the expiration date T (i.e., the final value problem becomes an initial
value one). Equations (8.22) and (8.23) then become (Exercise 8.21):
v 2 v v
( x, ) = 2 ( x, ) + (k 1) ( x, ) k v( x, ) (8.24)
x x
and
v( x, 0) = max{ex 1, 0}, lim v( x, ) = 0, v( x, ) ex as x . (8.25)
x
Using a trial solution
v( x, ) = u( x, ) ea x+b , (8.26)
it can be shown (Exercise 8.22) that the choices

1 1
a = (k 1), b = (k + 1)2 , (8.27)
2 4
transform (8.24) into the heat equation
u 2 u
( x, ) = 2 ( x, ) (8.28)
x
and (8.25) to
$ 1 %
u( x, 0) = max e 2 (k+1) x e 2 (k1) x , 0 , lim u( x, ) ec x = 0, (8.29)
1 2
| x|
where c > 0.
The heat equation has been extensively studied in physics and mathemat-
ics. The key result is that Equations (8.28) and (8.29) have a unique solution
given by
1 ( xs)2
1
u( x, ) = u(s, 0) e 4 ds, (8.30)
2
where u(s, 0) is given in (8.29). After some change of variables and com-
pleting the square, the solution (8.30) can be transformed (see Wilmott,
Dewynne, and Howison [42, Sec. 5.4]) to
u( x, ) = e 2 (k+1) x + (k +1)2
N(d+ ) e 2 (k1) x + (k 1)2
1 1 1 1
4 4 N( d ), (8.31)
where N() is the standard normal c.d.f. and

x + (k + 1)
d+ = , d = d+ 2 .
2
Inserting (8.31) and (8.27) in (8.26), the call price becomes (Exercise 8.23)
C E ( x, t) = K v( x, ) = x N(d+ ) K er ( T t) N(d ). (8.32)
Transforming back to the original variables ( x, t), Equation (8.32) shows that
the unique solution of the BSM p.d.e. (8.19) subject to (8.20) and (8.21) is

C E ( x, t) = x eq ( T t) N d+ ( x, T t) K er ( T t) N d ( x, T t) , (8.33)
where4

1 x 1 2
d ( x, T t) = ln ( T t) . (8.34)
Tt K e(rq) ( T t) 2
Note that Equation (8.34) can be written as follows:

ln Kx + r q + 12 2
d+ ( x, ) = (8.35)

ln Kx + r q 12 2 ln Kx +
d ( x, ) = = (8.36)

d+ ( x, ) = d ( x, ) + ,
where = T t and = r q 12 2 (which appeared in our study of the CRR

tree (see Equation (5.52) on page 234)).
Equation (8.33) gives the BSM formula for pricing European call options. Ex-
plicitly, the price of a European call on 1 unit of the underlying security is then
a stochastic process given by the value of the function C E ( x, t) at ( x, t) = (St , t)
for all t 0:

C E (St , t) = St eq ( T t) N(d+ (St , T t)) K er ( T t) N d (St , T t) , (8.37)
or, more compactly,

CtE = St eq N(d+ ) K er N d ( = T t ). (8.38)
The current price of the European call is then
C0E = C E (S0 , 0).
4 Some authors use the alternative notation d1 = d+ and d2 = d .

It is important to emphasize that European call prices in the real world are determined
by market forces, not by the BSM formula.
The BSM pricing formula (8.37) is also applied when the underlying security
is replaced by a risky portfolio of securities. In this situation, the total value of
the portfolio at a general time t is St , and the portfolio is assumed to follow
geometric Brownian motion for t 0. Like the underlying security of a deriva-
tive, the portfolio is also assumed to be tradable.
Remark 8.3. The original 1973 formula by Black and Scholes for a European
calls price assumed q = 0, while Mertons 1973 paper extended the result to
q > 0.

Notation. The price of a European call will be written in several different

ways, depending on the degree of notational simplicity needed and the de-
pendence we wish to highlight:
CtE = C E (St , t) = C E (St , K, , r, , q) ( = T t ).
The rightmost expression emphasizes the calls dependence on all the inputs
St , K, , r, , and q.
Example 8.2. What is the fair current price of 500 European calls on an index
with current dollar value $1,100, strike price $1,100, volatility 15%, two months
to expiration, and dividend yield of 2.5%? Assume a risk-free rate of 2%.
Solution. The fair price of a derivative is its no-arbitrage price, which in the
case of a call is the BSM price. At the current time t = 0, the inputs are:
S0 = $1, 100, K = $1, 100, = 0.15, r = 0.02, = 0.166667, q = 0.025.
Then the European call formula (8.37) yields C0E = $26.31437.5 Each call is
based on 100 indexes, so the total cost of the 500 calls is
500 100 $26.31437 = $1, 315, 718.50.

8.2.2 BSM Pricing Formula for European Puts
Inserting the call price (8.37) into the put-call parity formula, namely,
P E ( S t , t ) = K e r S t e q + C E ( S t , t ) ,
immediately yields the price process of a European put:

P E (St , t) = K er N d (St , ) St eq N(d+ (St , )). (8.39)
5 Many free online calculators are available for computing the prices of calls and puts.
The current price is then P0E = P E (S0 , 0). As noted earlier for calls, the actual prices
of European puts are dictated by the marketplace and not by (8.39).
The pricing formula (8.39) can also be obtained by solving the BSM p.d.e.
with final condition
P E ( x, T ) = max{K x, 0}
and boundary conditions P E (0, t) = K er ( T t) and P E ( x, t) 0 as x . The
final and boundary conditions also follow from no arbitrage and put-call par-
ity. The corresponding unique solution is
P E ( x, t) = K er ( T t) N(d ) x eq ( T t) N(d+ ),
which yields the put price (8.39) at t via ( x, t) = (St , t).
Example 8.3. What is the fair current price of 1,000 European puts on a stock
with current price $82, strike price $82, volatility 10%, and six months to ex-
piration? Assume a risk-free rate of 3% and that the stock pays no dividend.
Solution. The inputs at the current time t = 0 are:
S0 = $82, K = $82, = 0.10, r = 0.03, = T = 0.5, q = 0.
The BSM pricing formula (8.39) gives P0E = $1.7365. Hence, the fair cost of the
1,000 puts is: 1, 000 100 $1.7365 = $173, 650.

8.2.3 Delta and the Partial Derivative Relative to Strike Price
The European call price naturally involves partial derivatives relative to the
underlier price (delta) and the strike price. In fact, Equation (8.37) can be ex-
pressed more compactly in terms of partial derivatives:
C E
C E ( St , t ) = St C ( t ) + K ( t ), (8.40)
K
where the delta of the call is
4
C E 44
C ( t) = = eq N(d+ (St , )) > 0 (8.41)
x 4( x,t)=(St ,t)
and
C E
( t ) = e r N d ( S t , ) < 0, (8.42)
K
where the strike price K is treated as a variable.
Notation. When confusion is less likely, we shall also write
C E
C ( t) = ( t ).
S
The partial derivative C

E
S (t) should be understood as the partial of C ( x, t)
E
with respect to x with evaluation at ( x, t) = (St , t):

4
C E C E 44
( t) = .
S x 4( x, t)=(St , t)
A similar notation will be used for European puts.
An important consequence of (8.40), (8.42), and Itos formula is that the price
of a European call is more volatile than that of its underlying security (Exercise 8.25).
In addition, Equation (8.41) yields that the delta of a European call is always
positive, so the price of a European call increases as the price of the underlying secu-
rity increases. Furthermore, Equation (8.42) shows that C
E
K (t) < 0, i.e., the price
of a European call decreases as the strike price increases. In other words, an out-of-
the-money European call is cheaper than an at-the-money or in-the-money call with
the same inputs:
C E (St , K , , r, , q) < C E (St , K , , r, , q) for K St < K .

out-of-the-money at-the-money or in-the-money
(8.43)
For European puts, Equation (8.39) becomes
P E
P E ( St , t ) = St P ( t ) + K ( t ),
K
where
P E
P ( t) = (t) = eq N(d+ (St , )) < 0 (8.44)
S
and
P E
( t ) = e r N d ( S t , ) > 0. (8.45)
K
The puts delta is negative by (8.44), i.e., the price of a European put decreases as the
price of the underlying security increases. Moreover, since (8.45) gives P
E
K (t) > 0,
the price of a European put increases as the strike price increases. It follows that an
out-of-the-money European put is cheaper than an at-the-money or in-the-money put
with the same inputs:
P E (St , K , , r, , q) < P E (St , K , , r, , q) for K < St K .

out-of-the-money at-the-money or in-the-money
(8.46)
8.2.4 European Call and Put Deltas at Expiration
We conclude with the behavior of the deltas of European calls and puts as time
approaches expiration, i.e., as = T t 0. First, recall that (see page 394)

St e q
ln 1 2
St
K e r 2 ln K (r q 12 2 )
d ( St , ) = = +

d (St , ) = d+ (St , ) .
Then the following holds as t approaches expiration T:

If S( T ) = K, then d (St , ) 0 as 0.
If S( T ) > K, then d (St , ) + as 0.
If S( T ) < K, then d (St , ) as 0.
Using (8.41), (8.44), and the above, we find
If S( T ) = K, then C (t) 1/2 and P (t) 1/2 as 0. (8.47)

If S( T ) > K, then C (t) 1 and P (t) 0 as 0. (8.48)
If S( T ) < K, then C (t) 0 and P (t) 1 as 0. (8.49)
If t < T, then C (t) < 1 and P (t) > 1. (8.50)
8.3 Application to Pricing Warrants
Warrants are call options issued (i.e., sold) by a company on its own stock. They
provide a way for companies to raise money. When the warrants are exercised,
the company issues new shares of its stock and sells them to the holders at the
strike price. Note that issuing these new shares dilutes the share price of the
stock.
Suppose that at the current time, t = 0, a company has Nout outstanding shares
with each of price S(0) and issues Nw warrants, where each warrant is a European call
on 1 share of the companys stock with strike price K and expiration T. The number of
outstanding shares will not change until the warrants are exercised. The equity
value of the company, i.e., the value of the companys asset minus the value
of its debt, at 0 is denoted by V (0). It consists of the current value Nout S(0)
of the Nout outstanding shares and the proceeds Nw W (0) from selling the Nw
warrants, where W (0) is the current value of each warrant at 0:
V (0) = Nout S(0) + Nw W (0). (8.51)

8.3 Application to Pricing Warrants 399
Assume that all Nw warrants are exercised at expiration T. Denote by Va ( T ) the

equity value of the company immediately after the warrants are exercised. It
is the sum of the companys equity value V ( T ) just before the warrants are
exercised and the proceeds from selling Nw new shares at unit price K to settle
the warrants being exercised:
Va ( T ) = V ( T ) + Nw K.
The price of a share of the companys equity instantly after the warrants are
exercised is
Va ( T )
,
Nout + Nw
where Nout + Nw is the number of outstanding shares after exercise.
To value the warrant today, construct two portfolios as follows:
Portfolio A: long 1 warrant on 1 share of the company stock. The current
payoff of portfolio A is then the current value W (0) of the warrant.
Nout V (0 )
Portfolio B: long Nout + Nw European calls with current underlier price Nout ,
strike price K, expiration T, and no dividend. The price of the underlier at
V (t)
a general time t is the value Nout of a share of company equity before ex-
V (t)
ercise. We assume that Nout follows a geometric Brownian motion with volatility
parameter V . The current payoff of Portfolio B is then determined using the
BSM European call pricing formula:

Nout E V (0)
C , K, V , r, T, q .
Nout + Nw Nout
The payoffs of portfolios A and B at T are:

Va ( T ) Nout V (T)
(Payoff of A at T) = max K, 0 = max K, 0 ,
Nout + Nw Nout + Nw Nout

Nout V (T)
(Payoff of B at T) = max K, 0 .
Nout + Nw Nout
Because both portfolios have the same payoff at expiration, the law of one price
yields that they have the same payoff today, which gives the current value of
each warrant to be

Nout V (0)
W (0) = CE , K, V , r, T, q . (8.52)
Nout + Nw Nout
V (0 ) V (0 )
Equation (8.52) assumes that Nout and V are known. If Nout is unknown, then
employing (8.51) we find:

Nout Nw
W (0) = C S (0) +
E
W (0), K, V , r, T, q . (8.53)
Nout + Nw Nout
This is a BSM-type pricing formula for the current price of the warrant in terms
of the current price of the warrant. In other words, we can numerically solve
(8.53) for W (0) implicitly.
Example 8.4. Suppose that a company has 3 million outstanding shares, the
current value of a share of the companys equity is $110, and the companys
equity per share has a volatility parameter of 20% per annum. The company
plans to issue 500,000 European warrants. Each warrant is based on 1 share of
the companys stock, the strike price is $125, and the expiration date is 3 years
away. Determine the fair price of the issuance, i.e., the BSM price. Assume the
company pays no dividend and the risk-free rate is 2.5%.
Solution. Applying Equation (8.52) with Nout = 3 106 , Nw = 0.5 106 ,

V (0 )
Nout = $110, V = 0.2, K = $125, r = 0.025, T = 3, and q = 0, the price per war-
rant is:
3 E
W (0) = C ($110, $125, 0.025, 3, 0) = 0.85714286 $12.73093 = $10.91.
3.5
The fair total price of the issuance is:
500, 000 $10.91 = $5, 455, 000.
This is how much money the company would raise if it sold all the warrants
at the BSM price.

8.4 Risk-Neutral Pricing
In this section, we shall restrict our presentation to European style deriva-

tives and give an intuitive, informal presentation. The case of American-
style derivatives requires a more extended, complex discussion and is be-
yond the scope of this text; see, for example, Bjork [5, Chaps. 7, 21] and Epps
[15, Chap. 7].
8.4.1 Review of Conditional Expectation
Conditional expectations E ( Xt |Fs ) will play an important role in our discus-

sion of risk-neutral pricing, so we shall review the concept intuitively starting
with the -algebra Fs . See Section 6.2 (page 268) for more.
8.4 Risk-Neutral Pricing 401
First, the in the name -algebra is not the volatility parameter of a

security price modeled as geometric Brownian motion. The term -algebra
was developed in mathematics independent of finance, and its usage is in-
grained in the literature, so we abide by its customary usage.
Second, unless otherwise stated, the -algebra Fs refers to the collection of all
events generated by standard Brownian motion B = {Bt }t0 on the interval [0, s]:
Fs = FsB = (B , 0 s).
In other words, the events in Fs are typically in terms of ranges of values of B.

Some examples are:
A = { : B ( ) 10, 0 s} , B = { : Bs ( ) > 15} ,

C = { : 1 B ( ) 1, 0 s} .
The collection Fs also includes the sample space , which consists of all pos-
sible paths of standard Brownian motion, the empty set , complements of
events in Fs , and countable unions of events in Fs . Each event in Fs then car-
ries a piece of information about standard Brownian motion B on [0, s]. For example,
the sample space is the event that the actual path standard Brownian motion
will follow will be one of the possible paths in , i.e., one of the possible paths
of will occur (superficial information), while C is the event that the values
of B are between 1 and 1 from time 0 to time s. The -algebra Fs is then an
information set about the history of B up to and including time s. Bear in
mind that the word history should not be interpreted literally to imply that
s is the current time or a past time; the time s can be in the future.
If standard Brownian motion is observed on [0, s], then we know the ac-
tual path it took on [0, s], i.e., we can confirm which events in Fs occurred or
not. Since the current time is at 0 and s > 0 is in the future, we may not yet
know which events in Fs have occurred. However, all events in Fs are still
confirmable in the sense that when the current time reaches s, we shall know
for each event in Fs whether or not it occurred. On the other hand, the event
= { : B ( ) 10, 0 s + 1}
A
is not in Fs because it involves a portion of the possible Brownian motion paths

beyond time s, namely, on (s, s + 1], and so its occurrence cannot be tested
based on observations of B on [0, s]. Event A is not confirmable relative to the
information set Fs , but confirmable relative to Fs+1 . Note that some events in
Fs+1 may still be in Fs , though not all.
In summary, we think of the information set Fs as the set of all confirmable events
about standard Brownian motion on [0, s]. As time increases, we assume that no
information is lost, i.e., Fs Ft for s t. The collection {Ft }t0 , where Fs
Ft for every s t, is called a filtration and is viewed as capturing the entire

history of standard Brownian motion.
Third, a stochastic process X = { Xt }t0 is said to be Fs -measurable if all
events generated by X on [0, s] are also events in Fs :
FsX = ( X , 0 s) Fs .
Analogous to Fs , the elements of FsX include , , the events

X1 (, b] = { : X ( ) b}
for all b in R and 0 s, and complements and countable unions of events

in FsX . If X is Fs -measurable, then all the information in FsX is contained in
Fs , i.e., each event in FsX is confirmable. This implies that when the path of
standard Brownian motion is observed on [0, s], we shall know the actual path
the stochastic process X took on [0, s] and can confirm the occurrence of each
event in FsX . In a nutshell, X is Fs -measurable if all confirmable events about
X on [0, s] are determined by the confirmable events about B on [0, s]. The
stochastic process X is called adapted to standard Brownian motion or, simply,
an adapted process, if X is Fs -measurable for all s 0. Intuitively, to say that a
stochastic process X = { Xt }t0 is an adapted process means that for any t, the
values of the random variable Xt are determined (almost surely) by the con-
firmable events about B up to and including time t. A deterministic function
of standard Brownian motion is then an adapted process. On the other hand,
the stochastic process {Yt }t0 , where Yt = Bt+1 , is not adapted since at a given
t the random variable Yt requires information about B on (t, t + 1] and so is
not completely determined by the information about B on [0, t]. The process
{Yt } looks into the future of B at each t. We shall be interested primarily in
adapted stochastic processes.
Our geometric Brownian motion model {St }t0 of an underlying security is
an adapted process since
g( x, t) = S0 e(mq 2
1 2 ) t+ x
St = g(Bt , t), .
Moreover, given that the constants S0 , m, q, and are assumed known, we see
that for every fixed t the quantity Bt uniquely determines St through g(Bt , t)
and St uniquely determines Bt via

1 x 1 1
Bt = g(St , t), g( x, t) = ln (m q 2 ) t.
S0 2
For this reason, the -algebra Fs generated by {Bt }t0 is the same as the one
generated by the security price process {St }t0 : 6
6 Note: ( X ) (Y ) if and only if X = f (Y ); see Mikosch [32, p. 66] for more.

Fs = (B , 0 s) = (S , 0 s). (8.54)
Finally, we now turn to the conditional expectation E ( Xt | Fs ). Intuitively, it

computes the expectation of Xt relative to some probability measure from the
vantage point of time s, taking into account the information set Fs and looking
toward time t. In other words, the conditional expectation E ( Xt | Fs ) takes the
expectation of Xt given the history of B up to and including time s. Note (as
pointed out earlier) that if t > s, then knowing the history of B up to time s,
i.e., given the information set Fs , is not enough to determine Xt .
The conditional expectation E ( Xt | F0 ) computes the expectation of Xt given
the information set F0 = {, }, which merely tells you that one of the paths in
the sample space will occur, while a path not in will not occur. With such
insignificant information, the expectation of Xt has to be computed by taking
into account all the possible sample paths of X. This yields the unconditional
expectation:
E ( X t | F0 ) = E ( X t ) .
When s > 0, the quantity E ( Xt | Fs ) is typically random since for each event
in Fs , we use the events information when computing the expectation of Xt .
This usually yields different possible values of E ( Xt | Fs ) as we range through
the events in Fs . Overall, the conditional expectation E ( Xt | Fs ) is a random vari-
able that can be thought of, in a certain sense, as the best approximation of Xt given the
history of B up to time s. With our best-approximation interpretation, if { Xt }t0
is an adapted process, then E ( X | Fs ) = X for all 0 s. Moreover, for
a.s.
adapted processes { Xt }t0 and {Yt }t0 , we have:
E ( X Ys | F ) = X E (Ys | F ) (0 s ).
a.s.
We also see that if Xt is independent of the information contained in Fs , then

we are back to the unconditional expectation:
E ( Xt | F s ) = E ( Xt ) ( Xt is independent of Fs ). (8.55)
Another useful result is the tower property:
E ( E ( Xt |Fs ) |F ) = E ( Xt |F ) (0 s t ).
a.s.
(8.56)
For = 0, the tower property (8.56) implies that
E ( E ( Xt |Fs )) = E ( Xt ) . (8.57)
8.4.2 From BSM Pricing to Risk-Neutral Pricing
First, a basic assumption is that there is a market probability measure P that

determines the true probabilities of all traded securities. Of course, no one
knows the exact nature of P, and all market participants try to approximate P
as closely as possible. Each investor then has his or her own subjective proba-
bility measure to forecast the expected security prices. To illustrate this explic-
itly, recall from Section 3.6.2 that investors always maximize their expected
utility and can be divided into three categories according to risk preference:
risk averse, risk seeking, and risk neutral. Let PRA , PRS , and PRN be their re-
spective subjective probabilities. The lognormal nature of the underlying secu-
rity yields that investors maximizing their expected utility will hold the under-
lier if their subjective probabilities determine the same expected utility for the
underlier and a risk-free security. Equations (3.88),7 (3.89), and (3.90) imply:
Risk-averse investor: EPRA (St ) = S0 e(mq) t > S0 e(rq) t m>r

(mq) t (r q ) t
Risk-seeking investor: EPRS (St ) = S0 e < S0 e m<r
Risk-neutral investor: EPRN (St ) = S0 e(mq) t = S0 e(rq) t m = r,
where m is the instantaneous mean return of the underlier and r is the risk-free
rate.
Turning to the BSM model, it determines the price of a derivative as the so-
lution of the BSM p.d.e. under appropriate final and boundary conditions. A
simple but profound property that is easy to gloss over is that the coefficients
of the BSM p.d.e. are independent of the instantaneous expected return rate m.
In other words, given a derivative with well-behaved final and boundary con-
ditions, the solution of the BSM p.d.e. gives a unique derivative price process
that is independent of m. This means that the investors risk preference is irrelevant!
Denote the value of the instantaneous expected return rate m more explicitly
by mP , where P is the probability measure used to compute m. The above ob-
servation suggests that a derivative can be priced in a world of risk-neutral in-
vestors, where each investor has a probability measure PRN such that mPRN = r.
Remark 8.4. The insight for pricing a derivative in a risk-neutral world is due
to Cox and Ross [9].

Two critical points need to be emphasized:

We have assumed the mathematical existence of a risk-neutral world, i.e.,
the existence of a probability measure PRN for which mPRN = r.
We have assumed that there is a single probability measure PRN for the
risk-neutral world.
7 See page 135.
How do we know that a risk-neutral probability measure exists and there is

only one? It may be possible for different such probability measures to produce
different prices of the same derivative. We shall address below the existence of
a risk-neutral measure in the context of the BSM model, which assumes that
all underlying securities follow a geometric Brownian motion. In general, the
existence and uniqueness of these probability measures is a deep issue that
culminates with the fundamental asset pricing theorems (Sections 8.4.3). These
theorems will bring into sharper focus some of the core market assumptions
we are making in the BSM model.
Given the real market probability measure P, we shall now discuss the issue
of whether there exists a risk-neutral probability measure Q relative to which
derivatives can be priced.
Risk-Neutral Pricing of Underliers
A key assumption of the BSM model is that any underlying security of a

derivative follows geometric Brownian motion:
dSt = (m q) St dt + St dBt . (8.58)
Assume that Bt is standard Brownian motion with respect to P. In particular,

the instantaneous expected return is mP = m and the expected capital-gain re-
turn (m q) dt over the instant dt is not due to the subjective probability of
some risk-averse investor, but is relative to the true market probability mea-
sure P.
Now, recall Girsanov theorem:8 there is a probability measure Q that is equivalent
to P and given by

( mr ) BT ( mr )
1 2
T
dQ = e 2
dP (0 t T ) (8.59)
such that
mr
BQ
t = Bt + t

is a standard Brownian motion on (, FT , Q ).9 Note that mr is a Sharpe ratio.
Under the probability measure Q, underlying security price, which by assump-
tion is a geometric Brownian motion, has instantaneous expected return given
by the risk-free rate r. To see this, rewrite (8.58) as
dSt = (r q) St dt + St dBQ
t , (8.60)
where
8 See Privault [34, Thm. 6.1].
2 2
( mr ) BT ( ) 12 ( mr ) T
9 Q ( A) = A DT ( ) dP ( ), where A, A F T , and DT ( ) = e .

mr
dBQ = dBt + dt.
t

Since BQ t is a standard Brownian motion with respect to Q, the underlier price
St in (8.60) is also a geometric Brownian motion relative to Q, and its instan-
taneous expected return is the risk-free rate, i.e., mQ = r. The latter reveals the
risk neutrality of Q.
Let us express the risk neutrality of Q more explicitly by showing the link to
martingales. Equation (8.60) is solved by
+ BQ
S t = S0 e ( r q 2 )t
1 2
t (0 t T),
where S0 is the known initial price of the underlying security. Consider the
discounted price process
St = er t Stc = e(rq) t St ,
where Stc is the cum-dividend price of the underlier. Taking the conditional
expectation under Q yields (Exercise 8.28)

EQ St |F = S (0 t T ),
where the conditioning is relative to F since BQ is a function of B. It follows

that the discounted price process {St }t0 is a martingale. In other words, the value
of S is obtained by continuously discounting EQ (St |F ) at the net risk-free
rate r q:
S = e(rq) (t) EQ (St |F ) (0 t T ). (8.61)
For this reason, the probability measure Q is called risk-neutral probability mea-
sure, i.e., for every underlying security price process {St }t0 , the discounted
process {e(rq) t St }t0 is a martingale relative to Q. Equation (8.61) is the
risk-neutral price of the underlying security at time . Note that the market
probability measure P is risk neutral only when m = r, in which case P = Q.
Risk-Neutral Pricing of Derivatives
We now show how to price a derivative relative to Q. In the BSM model, we

obtained the price of a derivative using a self-financing, replicating portfolio:10
f (St , t) = Vt = nt Stc + bt Bt , (8.62)
where Vt is the value of the portfolio at time t and
10 See (8.13) on page 390.


St f ( St , t ) St f ( St , t )
nt = f ( St , t ), bt = . (8.63)
Stc Bt
In fact, we saw that the self-financing, replicating nature of the portfolio deter-
mines f as a solution of the BSM p.d.e.
Consider the discounted portfolio value process
t = er t Vt ,
V
where there is no downward adjustment by q since the derivative is assumed

to pay no dividend. As noted earlier, we always assume that nt and bt satisfy
the properties needed to apply stochastic integration and Itos formula.
Since (Exercise 8.24)
dSt = St dBQ

t , dVt = r Vt dt + nt (m r) Stc dt + nt Stc dBt ,
we find
t = nt St dBQ .
dV t
Integrating yields
1 t
t = V
V 0 + nv Sv dBQ
v. (8.64)
0
An important result from stochastic calculus is that stochastic processes like
those in (8.64) are martingales under the given measure and, conversely, a
(square integrable) martingale can be expressed in such a form (martingale rep-
t is a martingale under Q, we have
resentation theorem).11 Since V

EQ V t |F = V (0 t T ),
V = er (t) EQ (Vt |F ) (0 t T ).
Because the portfolio value Vs replicates the price of the derivative at s for all
0 s T, we obtain
f (S , ) = er (t) EQ ( f (St , t)|F ) (0 t T ).
In particular, the derivatives price at time t can be expressed relative to the

expiration date T:
f (St , t) = er ( T t) EQ ( f (ST , T )|Ft ) = er ( T t) EQ ( f (ST , T )) (0 t T ),

(8.65)
where
11 See, for example, Bjork [5, Sec. 4.4] and Elliot and Kopp [14, p. 176] for details.
Q Q Q
ST = St e ( T t) + (BT Bt ) = St e ( T t) + BTt (0 t T )
d
and
1
= r q 2 .
2
To obtain the rightmost equality in (8.65), recall that the independence of Brow-
nian motion increments implies that BQ Q Q
T Bt is independent of B for all
0 t. Consequently, the increment BQ Q
T Bt is independent of Ft . Since
the f is a deterministic function of St with St an invertible deterministic func-
tion of standard Brownian motion, Equations (8.54) (page 403) and (8.55) yield
the rightmost equality in (8.65).
Equation (8.65) is a stochastic process giving the risk-neutral price process of a
European-style derivative under the probability measure Q during the time interval
[0, T ]. Since at the current time, t = 0, we know the entire history of the under-
lying securitys prices up to and including time t0 , Equation (8.65) implies that
the derivatives current risk-neutral price is the following constant:
f (S0 , 0) = er T EQ ( f (ST , T )) , (8.66)
where T is the expiration date. Because of the conditional expectation in (8.66),

risk-neutral pricing lends itself naturally to numerical simulations.
In summary, the function f on the left-hand side of (8.66) is determined by
solving the BSM p.d.e. under appropriate final and boundary conditions, and
the value f (St0 , t0 ) of f at the current security price St0 and current time t0 is
exactly the value obtained from discounting the risk-neutral conditional expec-
tation in accordance with the right-hand side of (8.66). The important lesson is
that we can price a derivative either by solving the BSM p.d.e. under appropriate final
and boundary conditions or by computing a discounted risk-neutral conditional expec-
tation. This connection between partial differential equations and probability
theory is not coincidental and is far reaching.
Remark 8.5. Readers interested in more about the link between p.d.e.s and
stochastic differential equations should explore the Feynman-Kac formula.

8.4.3 The Fundamental Theorems of Asset Pricing
Two issues were raised in Section 8.4.2 (see page 404): does there exist a risk-
neutral probability measure relative to which a European-style derivative can
be priced and, if it exists, is there a unique such probability measure? In Sec-
tion 8.4.2, we employed Girsanov theorem to invoke the existence of such a
measure Q and derived a formula (8.66) for a derivatives risk-neutral price

under Q. In this section, we shall address which assumption of the BSM model
gives rise to the existence of a risk-neutral measure.
Consider a more general market setting where we do not enforce a priori
the no-arbitrage condition. The theorem below gives a link between the no-
arbitrage condition and the existence of a risk-neutral probability measure.
Theorem 8.1 (First Fundamental Theorem of Asset Pricing). 12 A market has no

arbitrage if and only if there is a risk-neutral probability measure Q that is equivalent13
to the market probability measure P.
The theorem reveals that fundamentally it is actually the no-arbitrage assump-

tion of the BSM model which grants us the existence of a risk-neutral proba-
bility. The remaining outstanding item is the uniqueness of the risk-neutral
probability measure.
A derivative is called attainable if there is a self-financing portfolio that repli-
cates its price. A securities market is called complete if all its derivatives are
attainable.14 The next theorem takes up the issue of market completeness and
the uniqueness of risk-neutral measures.
Theorem 8.2 (Second Fundamental Theorem of Asset Pricing). Suppose that

a market has no arbitrage. Then the market is complete if and only if there is a unique
risk-neutral probability Q that is equivalent to the market probability measure P.
Notation. For simplicity, we shall often use an asterisk to indicate when the
risk-neutral probability is being employed. For example, if fYQ is the p.d.f. of
Y relative to Q, we shall also write fY instead of fYQ . In some cases, we shall
also write E ( X ) instead of EQ ( X ) for efficiency. This carries over from the
study of binomial trees. It should be clear from the context whether E ( X ) is
an expectation relative to the risk-neutral uptick probability pn of a binomial
tree or the risk-neutral measure Q of a continuous-time setting.
We saw in Section 8.1.3 (page 388) that all the European-style derivatives are
attainable in the BSM model. Since the marketplace for the BSM model has no
arbitrage and is complete, the BSM model has a unique risk-neutral probability
measure and so only one price for a derivative. Indeed, the unique risk-neutral
probability measure is the one given via Girsanov theorem by (8.59). This is
consistent with the BSM p.d.e. having a unique solution under appropriate
final and boundary conditions.
12 Some authors call Theorem 8.1 the Fundamental Theorem of Asset Pricing.
13 Probability measures P and Q are called equivalent when P ( A) = 0 if and only if Q ( A) = 0.
14 See Chapter 12 (e.g., Appendices A and B) by Staum in Birge and Linetskys handbook [4] for a
further discussion of the financial meaning and mathematical modeling of a complete/incomplete

market.
Later, we shall study the Merton jump diffusion model (Section 8.9) and en-
counter a no-arbitrage market that is incomplete. Moreover, since the Merton
model reduces to the BSM one, it also yields a unique risk-neutral probability
measure for the BSM model. See Section 8.9.4 on page 458.
Remark 8.6. The papers by Ross [37], Harrison and Kreps [19], Harrison and
Pliska [20], Kreps [25], and Delbaen and Schachermayer [11] laid the math-
ematical foundation for the two fundamental theorems of asset pricing. See
Schachermayer [39] for a history and discussion as well as Epps [15] for an
insightful summary. The lecture notes by Privault [34, Chaps. 2, 5, 15] give an
excellent accessible introduction to martingale pricing and the fundamental
asset pricing theorems in discrete and continuous time.
8.4.4 Risk-Neutral Pricing of European Calls and Puts
In Section 8.4, we saw that it suffices to price European-style derivatives in the

BSM model using its unique risk-neutral probability measure. This will pro-
vide a relatively fast way of obtaining the price of a European call and, via
put-call parity, the price of a European put. In general, however, risk-neutral
pricing does not lead to a simple analytical derivation of the price of a deriva-
tive. Oftentimes it has to be computed by numerical simulation.
By (8.65), the continuous-time risk-neutral current price of a European
call is:
C E (0) = er T E (C E ( T )).
We compute the risk-neutral expectation as follows:
1
E (C E ( T )) = E (max{ST K, 0}) = (v K ) f S( T ) (v) dv
K
1 1
= v f ST (v) dv K f ST (v) dv,
K K
where f ST is the risk-neutral lognormal density of ST .

The integrals are in terms of the risk-neutral probability density of the un-
derlying securitys price at expiration T, where in a risk-neutral world, we
have
2
S T = S0 e T + BT , = r q . (8.67)
2
The following theorem yields the desired evaluation of the integrals:
Theorem 8.3. Suppose YT = Y0 eX , where Y0 is a known constant and X N (0 , 02 )

is the price of a security. Let fYT be the risk-neutral lognormal density of YT . Then:
1
fYT (v) dv = N(d2 ) (8.68)
K
8.5 Binomial Approach to Pricing European Options 411
and 1
v fYT (v) dv = Y0 e0 +0 /2 N(d1 ),
2
(8.69)
K
where N() is the standard normal cumulative distribution function and
ln(Y0 /K ) + 0 + 02 ln(Y0 /K ) + 0
d+ = , d = .
0 0
Apply Theorem 8.3 with YT = ST , Y0 = S0 , and
X = T + BT N (0 , 02 ).
Here 0 = T, 02 = 2 T, and

ln SK0 + (r q + 12 2 ) T ln S0
K + (r q 12 2 ) T
d+ = , d = .
T T
That is, d = d (S0 , T ); see Equations (8.35) and (8.37) (page 394). It follows:
1 1
E (C E ( T )) = v f ST (v) dv K f ST (v) dv
K K

= Y0 e0 +0 /2 N d+ (S0 , T ) KN d (S0 , T )
2

= St e(rq) T N d+ (S0 , T ) KN d (S0 , T ) .
Hence, we obtain the BSM formula for the current price of a European call:

C (0) = S0 eq T N(d+ (S0 , T )) K er T N d (S0 , T ) . (8.70)
8.5 Binomial Approach to Pricing European Options
The binomial trees of Chapter 5 provide a useful and pedagogically insightful

way to explore the pricing of options. Basically, this approach approximates
the standard BSM formula in discrete time using binomial trees with a suffi-
ciently large number n of time steps and recovers the exact BSM formula in the
continuous-time limit n .
We shall first summarize the binomial-tree strategy for pricing European call
options and then determine explicitly the option pricing formula for one- and
two-period trees and present the n-period case in Section 8.5.4. Note that, in
applications with n sufficiently large, a specific binomial tree is chosene.g.,
a CRR treeso we can determine the values of the uptick factor un , downtick
factor dn , and uptick probability pn from the inputs r q, , and hn (time-

period size).
For simplicity, suppress notationally the dependence of un , dn ,

and pn on the number n of time steps:
un = u, dn = d, pn = p.
Additionally, write the n-period binomial tree price of a European

call at time ti as
C E ( t i ; n ),
where ti = ti1 + h for i = 1, . . . , n. The current time is t0 = 0.
8.5.1 One-Period Binomial Pricing by Self-Financing Replication
For a one-period binomial tree from time t0 to t1 , we shall determine the cur-
rent price C E (t0 ; 1) of a European call option with strike K and expiration T in
terms of a replicating portfolio that finances itself after an initial capital.
Assume an initial capital V0 at the current time t0 , and create a portfolio as
follows:
Take a position consisting of n0 units of a risky security, where the unit

price is S(t0 ) and the security pays a continuous cash dividend at annual
yield rate q. Recall that, unless otherwise stated, we assume that the cash
dividend is continuously invested back in the security by purchasing more
units.
Take a position consisting of b0 units of the money market fund, which is
B(t0 ) per unit at time t0 .
We do not yet know the explicit nature of the trading strategy (n0 , b0 )e.g.,
whether we have a long position in the security (n0 > 0) or short position in
the money market account (b0 < 0). However, since only the initial capital is
used to form these positions, we must have:
V (t0 ) = n0 S(t0 ) + b0 B(t0 ),
where the choice of B(t0 ) is an arbitrary positive number that we have simply
set to be $1. The value of the portfolio at t1 arises from changes in the values
of the security and the money market fund:
V (t1 ) = n0 Sc (t1 ) + b0 B(t1 ), (8.71)

Fig. 8.1 One-period binomial tree models of the prices of the security (left tree) and European call
option (right tree). Here p is the probability of an upward movement in the tree
where Sc (t1 ) is the value of security cum dividend (with dividend). In other
words, due to dividend reinvesting, the original one unit of the security has
grown to eq h . Consequently, the cum-dividend unit price of the security at t1 is
S c ( t1 ) = S ( t1 ) e q h , (8.72)
where S(t1 ) is the securitys ex-dividend (without dividend) unit price. Note
that the prices shown in Figure 8.1 are ex-dividend prices. The value of the
money market account at t1 is
B ( t 1 ) = B ( t 0 ) er h . (8.73)
Now, suppose that at time t1 we have a trading strategy (n1 , b1 ). Then the
value of the trading strategy is
n1 Sc (t1 ) + b1 B(t1 ).
However, for the portfolio to be self-financing, any trading strategy (n1 , b1 )

at t1 must be financed from the portfolio value (8.71) at t1 resulting from the
original trading strategy (n0 , b0 ):
V (t1 ) = n0 Sc (t1 ) + b0 B(t1 ) = n1 Sc (t1 ) + b1 B(t1 ). (8.74)
That is, after using the initial capital, no outside funds can be added, and no
funds can be taken out the portfolio.
We want to find a trading strategy (n1 , b1 ) at expiration t1 such that the port-
folios value replicates the calls price
V (t1 ) = n1 Sc (t1 ) + b1 B(t1 ) = C E (t1 ; 1). (8.75)
If this is the case, then by the law of one price, the current price of the European
call will equal the current value of the portfolio. Basically, the goal is to show
that the European call is attainable, i.e., there is a self-financing portfolio that
replicates its price.
Let us first solve for the trading strategy (n1 , b1 ) using the self-financing
condition (8.74). The two possible values of the S(t1 ) are (Figure 8.1):

Su (t1 ) = S(t0 ) u with probability p
S ( t1 ) = (8.76)

Sd (t1 ) = S(t0 ) d with probability 1 p,
where
u > 1, 0 < d < 1, 0 < p < 1.
By (8.72) and (8.76), Equation (8.74) becomes
n0 eq h Su (t1 ) + b0 B(t1 ) = n1 eq h Su (t1 ) + b1 B(t1 )

(8.77)
n0 eq h Sd (t1 ) + b0 B(t1 ) = n1 eq h Sd (t1 ) + b1 B(t1 ).
This is solved by
n1 = n0 , b1 = b0 . (8.78)
Turning to the replicating condition (8.75), the possible prices of the call at
t1 are (Figure 8.1):
E
Cu (t1 ) = max{Su (t1 ) K, 0} with probability p
C ( t1 ; 1 ) =
E
(8.79)
E
Cd (t1 ) = max{Sd (t1 ) K, 0} with probability 1 p.
Note that in (8.79), we made use of the fact that the terminal value of a Euro-
pean call is its payoff:
C E ( T ) = max{S( T ) K, 0}, (8.80)
where K is the strike price. Employing (8.78), we see that (8.75) becomes
n0 eq h Su (t1 ) + b0 er h B(t0 ) = CuE (t1 )

(8.81)
rh
n0 e qh
Sd (t1 ) + b0 e B(t0 ) = CdE (t1 ).
The coefficient matrix of this linear system of two equations in two unknowns
is invertible:
qh
e Su ( t 1 ) er h B ( t 0 )
det = e(r+q) h S(t0 ) B(t0 ) (u d) > 0.
e q h Sd ( t 1 ) er h B ( t 0 )
Hence, the unique solution of (8.81) is


CuE (t1 ) CdE (t1 ) e r h CuE (t1 ) d CdE (t1 ) u
n0 = e q h , b0 = .
Su ( t 1 ) Sd ( t 1 ) B ( t0 ) ud
(8.82)
Remark 8.7. The expression for n0 in (8.82) includes a discrete version of a Eu-
ropean calls delta, namely, the partial difference at t1 of the call price with
respect to the underlier price:
C E C E (t1 ) CdE (t1 )

( t1 ) = u .
S Su ( t 1 ) Sd ( t 1 )

In summary, since the dividends are reinvested in the underlying security,

there is a unique trading strategy (n0 , b0 ) given by (8.82) that can replicate the
call value at t1 in a self-financing way: (n1 , b1 ) = (n0 , b0 ). In other words, the
call is attainable. By the law of one price, the portfolio and the call have the
same price today:
C E (t0 , 1) = V (t0 ) = n0 S(t0 ) + b0 B(t0 ),
where n0 and b0 are given (8.82). Explicitly, the current price of the European
call can be expressed as

e (r q ) h d u e (r q ) h
C E ( t0 ; 1 ) = e r h CuE (t1 ) + CdE (t1 ) , (8.83)
ud ud
where (8.76) and (8.80) with T = t1 give
CuE (t1 ) = max{S(t0 ) u K, 0}, CdE (t1 ) = max{S(t0 ) d K, 0}.
Equation (8.83) yields that the options current price (8.83) is independent of the
underliers instantaneous expected return rate m! Hence, the one-period call price is
independent of investors view on m. This is consistent with the earlier observation
(page 404) from the BSM p.d.e. that the price of a derivative is independent of
m.
8.5.2 One-Period Binomial Pricing by Risk Neutrality
As noted above, the current price of the European call is independent of

the instantaneous expected return m of the underlier. In other words, even
if the marketplace were a risk-neutral world, i.e., one where the value of m is r,
the current price of the call would be unchanged. Since r is known, while es-
timating m from market data is problematic, it is simpler to price the call in a

risk-neutral world, i.e., price the call using a risk-neutral uptick probability.
How do we know that there is a risk-neutral uptick probability? First, let us
assume that there is one, i.e., there is a probability p ,
0 < p < 1, (8.84)
such that, relative to p , the expected price at t1 of the underlying security is

given by
E (S(t1 )) = S(t0 ) er h , (8.85)
where (as always) the current price S(t0 ) is assumed known. Equation (8.85) is
independent of S(t0 ) and equivalent to
p u + (1 p ) d = e (r q ) h . (8.86)
The risk-neutral uptick probability p is then characterized by (8.84) and (8.86).

Recall that for a risk-neutral binomial tree, we always assume15
e(rq) h = d, e(rq) h = u.
Otherwise, we get p = 0 or p = 1, which are excluded in our binomial-tree

analysis. Now, since u d > 0, we then obtain the following unique solution
of (8.86):
e (r q ) h d
p = . (8.87)
ud
Furthermore, recall that if there is no arbitrage, then (see page 232):
d < e(rq) h < u. (8.88)
Since u d > 0, Equations (8.87) and (8.88) yield
0 < p < 1.
Hence, p given by (8.87) is a unique risk-neutral uptick probability.

The probability p allows us to compute a unique current price for the call.
In fact, the expected future value of the European call at t1 is:

E C E ( t 1 ; 1 ) = C E ( t 0 ; 1 ) er h , (8.89)
where the call does not pay a dividend (by assumption). By (8.89), the current
call price is:

C E (t0 ; 1) = er h p CuE (t1 ) + (1 p ) CdE (t1 ) , (8.90)
15 See the constraints (5.49) on page 232.

where
CuE (t1 ) = max{S(t0 ) u K, 0}, CdE (t1 ) = max{S(t0 ) d K, 0}.
Comparing the risk-neutral price (8.90) and the replicating-portfolio price

(8.83) immediately shows that they are identical. In other words, it suffices to
price a European call using the risk-neutral approach, which we shall do going for-
ward for binomial trees.
Example 8.5. Suppose that the risk-free rate is 2% per annum and a stock with
current price of $50. Assume that the stock pays no dividend and its price 3
months from now is either $53.8900 or $46.3850, where four decimal places are
used to minimize rounding-off errors. Using a binomial tree, compute the one-
period current price of a 3-month European call on this stock given a strike
price of $50.
Solution. The formula for the one-period European call price at the current
time t0 is

C E (t0 ; 1) = er h p max{S(t0 ) u K, 0} + (1 p ) max{S(t0 ) d K, 0} ,
where t0 = 0 and
e (r q ) h d
p = .
ud
The needed inputs are:
h = 0.25, r = 0.02, q = 0, S(t0 ) = $50, K = $50
$53.8900 $46.3850
u= = 1.0778, d= = 0.9277, p = 0.5151.
$50 $50
Direct calculation then yields the current call price: C E (t0 ; 1) = $1.99.

8.5.3 Two-Period Binomial Pricing
Consider the two-period risk-neutral binomial tree in Figure 8.2, where the
time remaining on the European call is 2 h and runs from t0 to t2 . We shall
employ risk-neutral pricing to determine the current price of the call:

C E (t0 ; 2) = e r (2h) E C E (t2 ; 2) .
Fig. 8.2 Possible European call prices for a two-period binomial tree

Direct approach. To compute the expectation E C E (t2 ; 2) directly, Figure 8.2
shows that there are four paths leading to the three possible prices of the call.
The probability of a given price is the sum of the probabilities of each path
leading to the price, where the probability along a path is the product of the
probabilities along each section. It follows:

C E (t0 ; 2) = e r (2h) p2 CuE2 (t2 ) + 2p (1 p ) Cud
E
(t2 ) + (1 p )2 CdE2 (t2 ) .
(8.91)
Using (8.20) and (8.76), we get the explicit forms:
CuE2 (t2 ) = max{Su (t1 ) u K, 0} = max{S(t0 ) u2 K, 0}

E
Cud (t2 ) = max{Su (t1 ) d K, 0} = max{S(t0 ) ud K, 0}
CdE2 (t2 ) = max{Sd (t1 ) d K, 0} = max{S(t0 ) d2 K, 0}.
(8.92)
Note that when a binomial tree has many paths, our direct method to com-

pute E C E (t2 ; 2) becomes nontrivial quicklye.g., a 20-step binomial tree
already has over 1 million paths.
Algorithmic approach. We can also obtain (8.91) using a pedagogically simple

algorithm. The strategy is to divide [t0 , t2 ] into two equal-length subintervals,
namely, [t1 , t2 ] and [t0 , t1 ], and then apply the one-period risk-neutral result
(8.90) to [t1 , t2 ] and [t0 , t1 ]. That is, we work backward through the tree. Let
us now begin with the one-period subinterval [t1 , t2 ]. To apply the one-period
analysis to [t1 , t2 ], think of t1 and t2 as now playing the respective roles of the
present time t0 and the expiration time t1 in (8.90). Since the call price at time
t1 has two possible values CuE (t1 ) and CdE (t1 ), apply (8.90) to each of them with
t0 and t1 in (8.90) replaced by t1 and t2 , respectively:

CuE (t1 ) = er h E CuE (t2 ) | S(t1 ) = Su (t1 )

CdE (t1 ) = er h E CdE (t2 ) | S(t1 ) = Sd (t1 ) ,
where E
Cu2 (t2 ) with probability p
CuE (t2 ) =
E (t )
Cud 2 with probability 1 p
and E
Cud (t2 ) with probability p
CdE (t2 ) =

CdE2 (t2 ) with probability 1 p .
The possible call prices at time t1 are then:

C E ( t ) = e r h p C E ( t ) + ( 1 p ) C E ( t )
u 1 u2 2
ud 2
C ( t1 ; 2 ) =
E
(8.93)

C E ( t1 ) = e r h p C E ( t2 ) + ( 1 p ) C E ( t2 ) .
d ud d 2
Moving backward in the tree, consider the interval [t0 , t1 ]. Application of

(8.90) yields the current call price as:

C E (t0 ; 2) = er h E C E (t1 ; 2) = er h p CuE (t1 ) + (1 p ) CdE (t1 ) ,
where CuE (t1 ) and CdE (t1 ) are now given by (8.93). The European call option
price over the two-period interval [t0 , t2 ] is then given as follows:

C E (t0 ; 2) = e r (2h) p2 CuE2 (t2 ) + 2p (1 p ) Cud
E
(t2 ) + (1 p )2 CdE2 (t2 ) ,
(8.94)
where Cu2 (t2 ), Cud (t2 ), and Cd2 (t2 ) are given by (8.92). Equation (8.94) agrees
E E E
with (8.91).
Filtration approach. We can also obtain Equation (8.94) via risk-neutral pricing
using the -algebra Ft for discrete time t = t j ; see Example 6.17 (page 269) for
more on the filtration. In our binomial tree setting, the -algebra Ft j F j is
generated by the underlying securitys prices up to time t j . For example, F1 is
generated by S0 , S1 and given by
F1 = {, AU , A D , },
where
AU = {UU, UD }, A D = { DU, DD }, = {UU, UD, DU, DD }.

Note that each element in the events AU , A D , and is an entire price path
from t0 from t2 (not just to t1 ). In particular, the path = UU has an uptick at
t1 followed by another uptick at t2 and AU consists of all price paths with an
uptick at t1 , while A D are those with a downtick at t1 . The risk-neutral price of
the European call is

C E (t0 ; 2) = e r (2h) E C E (t2 ; 2) , (8.95)
where the tower property yields (see (8.57) on page 403):

4
E C E ( t2 ; 2 ) = E E C E ( t2 ; 2 ) 4 F1 . (8.96)
4
The conditional expectation E C E (t2 ; 2) 4 F1 is relative to each event in F1 .
Since the sample space merely tells us 4that one of the four price paths will

occur, the possible values of E C E (t2 ; 2) 4 F1 will come from the information
carried by AU and A D . Figure 8.2 shows that
4
E Cu (t2 ) 4 S(t1 ) = Su (t1 )
E with probability p
4
E C E ( t 2 ; 2 ) 4 F1 =
E C E (t ; 2) 44 S(t ) = S (t ) with probability 1 p

u 2 1 d 1

p Cu2 (t2 ) + (1 p ) Cud (t2 ) with probability p
E E
=

p C E (t ) + (1 p ) C E (t ) with probability 1 p .
ud 2 d2 2
Consequently,

E C E (t2 ; 2) = p2 CuE2 (t2 ) + 2p (1 p ) Cud
E
(t2 ) + (1 p )2 CdE2 (t2 ).
Inserting the above into (8.95) yields the same call price as in (8.94).
Example 8.6. Suppose that a stock with current price of $50 pays not dividend.
Assume that at 1.5 months from now, the price of the stock is either $52.7250
or $47.4150, where four decimal places are used for pedagogical reasons to
minimize rounding errors. Employing a binomial tree and risk-free rate of 2%
per annum, compute the two-period current price of a 3-month European call
option on this stock given a strike price of $50.
Solution. The necessary inputs are as follows: h = 0.25

2 = 0.125, r = 0.02,
q = 0, S(t0 ) = $50, K = $50, and
$52.7250 $47.4150 e (r q ) h d
u= = 1.0545, d= = 0.9483, p = = 0.5104.
$50 $50 ud
Hence, the two-period pricing formula (8.94) yields C E (t0 ; 2) = $1.45.

8.5.4 n-Period Binomial Pricing
For the n-period case 0 = t0 < t1 < < tn , an induction argument yields the
current price of a European call as given by (Exercise 8.32):
n
r (nh ) r (nh ) n
C ( t0 ; n ) = e
E
E (C (tn , n)) = e
E
i pi (1 p )ni CuEidni (tn ),
i =0
(8.97)
where
CuEi dni (tn ) = max{S(t0 )ui dni K, 0}
for i = 0, 1, . . . , n. The formula (8.97) can be expressed more simply as (Exer-
cise 8.33):
C E (t0 ; n) = S(t0 ) eq N(n, k , p ) K er N(n, k , p ), (8.98)
where k is the smallest value of i for which S(t0 )ui dni K > 0 and
n
n i p u
= tn t0 , N(n, k , p ) = p (1 p )ni , p = (rq)h .
i=k
i e
The reader may have noticed that the pricing formula (8.98) looks similar
to the BSM pricing formula (8.70) on page 411. In the continuous-time limit
n , it can be shown (see Hsia [21]) that (8.98) converges to (8.70):
lim C E (t0 ; n) = C0E = S0 eq T N(d+ (S0 , T )) K er T N(d (S0 , T )).

n
Example 8.7. Suppose that a stock paying no dividend has a current price of
$50 and annual volatility of 15%. For a risk-free rate of 2% per annum, compute
the 100-period current price of a 3-month European call on this stock with a
strike price of $50.
Solution. Given the large number

of periods,

we employ the CRR formulas for
(r q ) h
u, d, and p , namely, u e h , d e h , and p e udd . Applying (8.97)

with the values n = 100, h = 0.25
100 = 0.0025, h = 0.05, r = 0.02, q = 0, = 0.15,
S(t0 ) = $50, K = $50, and
u = 1.007530, d = 0.992528, p = 0.501458,
we obtain a current call price of C E (t0 ; 100) = $1.62. This coincides with the re-
sult of the continuous-time BSM pricing formula (8.70). In fact, for an 80-period
tree, the two prices already agree to the given decimal places; see Figure 8.3.

2.0
1.9
CallPrice $ 1.8
1.7
1.6
1.5
0 20 40 60 80 100
n
Fig. 8.3 The per share prices of a 3-month European call option, where the strike price is K = $50
per share, risk-free rate is 0.02 per annum, and underlying stock has current share price of $50 and
volatility of 15%. The bullets show the call prices given by the n-period binomial option-pricing for-
mula using a CRR tree for n = 1, . . . , 100. The horizontal dashed line is the per share call price of $1.62
obtained using the BSM formula. Even values of n give call prices below the BSM line, while odd
values are above. The binomial per share call price is $1.99 for n = 1 and $1.45 for n = 2. The price to
two decimal places is already $1.62 for n = 80
Remark 8.8. The original binomial-tree approach to deriving the continuous-

time BSM price for European calls used the CRR formulas for u, d, and p .
However, it was shown by Hsia [21] that such specificity is unnecessary to
take the continuous-time limit. He gave a simpler proof for general u, d, and
p using the de Moivre-Laplace Theorem. The lecture notes by Chance [8] also
give an accessible treatment of the proof in [21].

8.6 Delta Hedging
In this section, we develop the theoretical infrastructure for delta hedging of

European calls (Section 8.6.1) and apply the framework to an example (Sec-
tion 8.6.2).
8.6.1 Theoretical Delta Hedging for European Calls
When selling European calls, the risk to the seller is to be able to meet the
obligation should the calls be exercised. For instance, if you sell 500 European
calls on a stock and they are exercised at expiration, then you need to have
8.6 Delta Hedging 423
50,000 shares of the stock available at expiration. The risk lies, for example, in
not having them available at expiration. Or, if you buy the 50,000 shares right
after selling the calls to be assured of covering your obligations in the event of
exercise, and the calls expire out of the money, you are left with a lot of shares
that may have even lost value.
Our core strategy for managing the risk from selling a European call is to
create simultaneously an offsetting position in a synthetic European call. The
synthetic call will be constructed using the funds from the short sale and bor-
rowing a certain amount. The core of the strategy is the process of delta hedging
and will result in the loan being paid off at expiration. Note that the cash divi-
dend from the underlier can be used either to buy more units of the underlier,
which is assumed by default, or to pay toward the loan. To make the presen-
tation different, instead of employing our default assumption, assume that the
cash dividend is used continuously as payment toward the loan.
We now detail the theoretical framework for hedging.
Time t
At time t < T, construct a costless portfolio as follows:
a) (Short Call) Short sell 1 European call on 1 unit of an underlying security.

This brings in proceeds of CtE .
b) (Long Synthetic Call) Using the proceeds CtE as start-up capital and bor-
rowing the amount
L t = K e r N d ( S t , ) ,
we can buy
C (t) = eq N(d+ (St , ))
units of the underlying security. This is because (8.38) (page 394) yields
CtE + Lt = C (t) St .
The portfolio with C (t) units of the underlier and loan Lt is called a syn-
E (t) replicates the call price:
thetic call because its value Csyn
E
Csyn (t) = C (t) St Lt = CtE .
Remark 8.9.

1. E ( t) = n S c
Since Csyn t t + bt Bt , where nt = SSct C (t) and bt = BLtt , the syn-
t
thetic call can be viewed as a portfolio with a position in nt units of the
cum-dividend underlier and bt units of the money market account. Equa-

tion (8.14) then yields that the strategy (nt , bt ) is self-financing and repli-
cates the call price for all times t 0.
2. By (8.50), the number C (t) of units we purchased of the underlying se-
curity is less than 1 since t < T. We did not purchase 1 unit of the underlier
to cover the call and neither did we take a naked position on the call.
The portfolio with positions (a) and (b) has a value:
0 = VC (t) = CtE + Csyn

E
(t) = 0. (8.99)
Since
E
Csyn ( t) = C ( t) St L t with Lt = C (t) St CtE , (8.100)
the value of the portfolio at t can then be expressed as
0 = VC (t) = CtE + It Lt + ( t ) St ,
C
short call risk-free investment loan long C ( t) units
(8.101)
where
It = CtE , L t = C ( t ) St .
Equation (8.101) has a natural financial interpretation: the portfolio with posi-
tions (a) and (b) at t is equivalent in value to a costless portfolio where one short sells
1 European call for CtE , invests the proceeds CtE in a risk-free investment, borrows
the amount L t = C (t) St , and uses the loan to buy C (t) units of the underlying
security.
Time t + dt
As time t advances, the positions on the right-hand side of (8.101) will change.
We shall have to update continuously our earlier positions and will do so in a
costless, self-financing way to maintain the equation at zero, i.e., maintain the
replication by the synthetic call. Recall that self-financing means the change in
the value of the portfolio comes strictly from the change in the value of the
securities in the portfolio, which includes a loan. For instance, any purchases
of additional units of the security will be funded by increasing the loan, i.e.,
more borrowing, and any proceeds from selling units of the security will not
be withdrawn, but paid toward the loan. Note that borrowing is at the risk-free
rate.
Let us look closely at this process. At time t + dt, the portfolios value has
the form:
0 = VC (t + dt) = CtE+dt + It+dt L t+dt + C (t + dt) St+dt .

(8.102)
We consider each term:

The short position CtE has changed to CtE+dt .
The original investment of It = CtE has grown at the risk-free rate r to
It+dt = CtE er dt .
Delta hedging. The previous two positions change on their own. The ac-
tual adjusting occurs in maintaining C (t) units of the underlier as time t
advances. This adjusting is called delta hedging and will be reflected in the
balance of the loan. In fact, at time t + dt, the new loan balance L t+dt is the
net result of using the securitys cash dividend to pay toward the loan and
adjusting, if necessary, the number of units of the underlier from C (t) to
C (t + dt). We then consider the difference
dC (t) = C (t + dt) C (t).
If dC (t) = 0, then C (t) suffices and the loans balance is
L t+dt = L t e(rq) dt .
If dC (t) > 0, then C (t) units are not enough, so we purchase dC (t) units

by borrowing dC (t) St+dt . The loans balance becomes

L t+dt = L t e(rq) dt + dC (t) St+dt .
If dC (t) < 0, then C ( t) units are too much. We then sell dC (t) units
and receive dC (t) St+dt . We pay this amount toward the loan, which
decreases the change in value of the loan to

L t+dt = L t e(rq) dt dC (t) St+dt .
All three cases are captured by the single equation,

L t+dt = L t e(rq) dt + dC (t) St+dt , (8.103)
where dC (t) can be zero, positive, or negative. The delta hedging process
is repeated during every instant until expiration.
Expiration Time T
At expiration T, the short position in the call will be CTE , the initial investment
CtE would have grown at the risk-free rate r to CtE er over the period = T t,
the loans balance will be L T , and the long position in the underlying asset will
consist of C ( T ) units and have value C ( T ) ST . The value of the portfolio at
expiration is then:
0 = VC ( T ) = CTE + IT L T + C ( T ) ST ,
where CTE = max{ST K, 0} by the final boundary condition, IT = CtE er , and

the loans balance is

L T = L T dt e(rq) dt + dC ( T dt) ST (8.104)
by (8.103). There are two possibilities to consider at expiration: either ST > K

or ST K.
In-the-money European call at expiration. Suppose ST > K, i.e., the call is
exercised.16 Then the obligation of the short sale of the call can be fulfilled. In
fact, Equation (8.48) shows that C ( T ) = 1 for ST > K, i.e., the delta hedging
process results in 1 unit of the underlying asset. It is this 1 unit that is sold
to the call holder at price K to fulfill the obligation of the exercised call. The
outstanding item now is whether the loan can be paid off. For the case ST > K,
we have CTE = ST K and C ( T ) = 1, which yield
0 = ST + K + CtE er L T + ST .
Consequently,
K + CtE er = L T .
Hence, if the call is exercised, then the loans balance L T can also be paid off by
using the proceeds K from selling the 1 unit of the underlying security to the
call holder and the cash CtE er from liquidating the risk-free investment.
At-the-money or out-of-the-money European call at expiration. Assume ST
K, i.e., the call is not exercised. Then even though the issuer has no obligation
to the call holder, the balance on the loan still has to be settled. For ST K, we
have CTE = 0 and so
0 = CtE er L T + C ( T ) ST ,
i.e.,
CtE er + C ( T ) ST = L T .
16We assume that a European call is exercised if and only if ST > K and not exercised if and only if
ST K.
In other words, if the call is not exercised, then the loans balance L T can still be paid
off by the cash inflow CtE er + C ( T ) ST from liquidating the risk-free invest-
ment and the long position of C ( T ) units of the underlying security. Note that
(8.47) yields C ( T ) = 12 for ST = K and (8.49) implies C ( T ) = 0 for ST < K.
The latter, along with (8.104) and
dC ( T dt) = C ( T ) C ( T dt),
implies that

(rq) dt + 1 ( T dt) S
L T dt e 2 C T at-the-money at T
LT =

L T dt e(rq) dt C ( T dt) ST out-of-the-money at T.
(8.105)
In summary, we see that employing delta hedging enables one to meet the
obligations of selling European calls using a costless, self-financing, replicating
process. Even though the process requires no initial capital, it involves borrow-
ing, but the loan is paid off at expiration.
8.6.2 Application of Delta Hedging to Selling European Calls
A major concern to a European call seller is fulfilling the obligations of the call
if it is exercised. To get a sense of this risk, we give a simple example:
Example 8.8. Suppose that a firm sells 500 European calls for $2.264248 per
share of a stock currently trading at $75 per share. Since each call involves 100
shares, the firm receives
500 100 $2.264248 = $113, 212.40.
Assume that there are 80 days to expiration, the strike price is $75, and the
firm invests the $113,212.40 proceeds in a risk-free investment growing at 2%
annually. Suppose that there are 365 days in a year; see Remark 8.1 (page 385).
Should the calls be exercised at expiration, the firm must sell 500 100 =
50, 000 shares of the stock to the call holder for $75 per share. But if the firm
does not own any share of the stock and plans to buy the 50,000 shares only
when the calls are exercised, then the firm is taking a naked position and ex-
posing itself to potential loss. For instance, if the share price at expiration is
$87.98, then the calls will be exercised and it will cost the firm $4,399,000 to buy
the shares to satisfy the obligations of the calls. Though firm would receive
$3,750,000 from selling the shares at the strike price and the firms proceeds
from selling the calls would have grown to e0.02(80/365) $113, 212.40 =
$113, 709.76 at expiration, the firm would still have a loss that more than
quadruples the gain from selling the calls:
$113, 709.76 $4, 399, 000 + $3, 750, 000 = $535, 290.24.
On the other hand, the firm can take a covered position by buying the 50,000
shares at the time it sells the calls. If the stock price falls to $65 at expiration,
i.e., drops in value to $500,000, then the call will not be exercised and, despite
the growth in the call-sale proceeds to $113,709.76, the firm will experience a
loss in value that more than triples this gain:
$113, 709.76 $500, 000 = $386, 290.24.

How many shares should the firm then hold to hedge against the risk from
selling a European call? The answer is actually not a fixed number of shares.
The number will have to change as time advances. It is called delta hedging.
This tool is actually contained in the construction of the costless, self-financing,
riskless portfolio employed in Section 8.1.4. We apply the ideas and results of
that section to illustrate, in principle, how delta hedging works. For simplicity,
the discussion will focus on the issuance of 1 call on 1 unit of an underlier.
The application in Section 8.6.2 and, in particular, Example 8.9 (page 428) will
illustrate delta hedging using the example above.
We now illustrate the theoretical framework for delta hedging European
calls by applying it to Example 8.8.
Example 8.9. Suppose that a firm sells 500 European calls for $113,212.40 on a
stock with share price of $75 and annual volatility of 15%. All of the calls then
involve 50,000 shares of the stock. Assume that there are 80 days to expira-
tion, the strike price is $75, and the firm invests the $113,212.40 in a risk-free
investment growing at 2% per annum. How can the firm manage the risk from
the call sale without taking a naked or covered position? We illustrate how to
manage the risk using daily delta hedging. Since the risk-free investment will
earn interest even on non-trading days, assume 365 days in a year and, for
simplicity, suppose that trading occurs on each of the 80 days remaining till
expiration.
Let us now delta hedge day by day for 80 days using Equations (8.101) and
(8.102) as a guide, which we rewrite for convenience:
0 = VC (t) = CtE + It Lt + ( t) S ,
C t
(8.106)
and
0 = VC (t + dt) = CtE+dt + It+dt L t+dt + C (t + dt) St+dt ,

short call risk-free investment loan long C ( t + dt) units
(8.107)
where

It+dt = CtE er dt , L t+dt = L t e(rq) dt + dC (t) St+dt (8.108)
with
dC (t) = C (t + dt) C (t).
Note that Equations (8.106) and (8.107) are based on one share of the stock.
The current time is Day 0, each instantaneous time change is approximated
by a day, and expiration is at T:
1
t = 0, dt h = , T = 80 h.
365
Because we cannot perfectly delta hedge, i.e., we cannot hedge every moment
of time (e.g., dt is replaced by a day) and cannot work to infinitely many deci-
mal places, our results will not yield VC (t) = 0 and VC (t + dt) = 0 as the current
time t advances day by day. However, though a perfect hedge would yield zero
loss at expiration, our approximate hedge will significantly reduce any losses
compared to the naked and covered positions in Example 8.8 (page 427). The
results are summarized in Table 8.1.
Remark 8.10. (Rounding Errors) Rounding errors are unavoidable whenever

results are expressed to a fixed number of decimal places. The delta hedging in
Table 8.1 is based on a MATLAB code that outputs results to much more than
six decimal places. We summarized the results in the table by rounding off
the MATLAB output at six decimal places to avoid making the table look too
dense. Naturally, the rounding off will produce errors. For instance, in the first
row of Table 8.1, we have C (0) S0 = $40.413671. However, using the initial
stock price S0 = $75 and the rounded value C (0) = 0.538849, the product
gives a different answer to six decimal places: C (0) S0 = $40.413675.
As noted, the actual MATLAB value of C (0) has far more than six deci-
mal places. To nine decimal places, it yields C (0) = 0.538848947 and so pro-
duces C (0) S0 = $40.413671025, which to six decimal places yields the entry
$40.413671 in the table. The latter entry is more accurate. Note that at the end
of the delta hedging, the total net value of the portfolio, i.e., the per-share value
V ( T ) times 50,000 shares, will be rounded off to cents.

We shall carry out the delta hedging on a per-share basis. Each per-share
position can then be multiplied by 50,000 to obtain the total size of the position.
Table 8.1 Delta hedging to mitigate against risk from selling 500 calls. The time to expiration is 80
days. Multiply the per-share values by 50,000 to get the total values. Though the European call expires
in the money, delta hedging enables the seller to have the required 50,000 shares and minimizes the
losses to only $2,326.10. Simulation is based on a MATLAB code, and we truncated the output at six
decimal places. The portfolios value at time t is VC ( t) = CtE + It L t + C ( t) St
Delta hedging values per share
t St C ( t) CtE It L t C ( t ) St V C ( t)
short call investment loan balance long C ( t) shares port. value
Day 0 $75.000000 0.538849 -$2.264248 $2.264248 -$40.413671 $40.413671 $0
Day 1 $75.806609 0.598789 -$2.707914 $2.264373 -$44.959756 $45.392182 -$0.011116

.. .. .. .. .. .. .. ..
. . . . . . . .
Day 79 $92.155473 1 -$17.159583 $2.274071 - $77.316481 $92.155473 -$0.046519
Day 80 $93.561474 1 -$18.561474 $2.274196 -$77.320717 $93.561474 -$0.046522
Day 0
The stock price is $75 and there are 80 days until expiration. We employ the
per-share Equation (8.106). The firm short sells each European call on one share
at the BSM price. With the starting time t = 0, the BSM price per share is

C0E = C E (S0 , K, , r, T, q) = C E $75, $75, 0.15, 0.02, 80 h, 0 = $2.264248445.
We shall employ at least six decimal places to minimize rounding errors. The
short sale creates a negative position with per-share value
C0E = $2.264248445.
The firm invests the proceeds from the sale in a risk-free account paying 2%
per annum:
I0 = C0E = $2.264248445 (per share).
Now, the current delta of the call to nine decimal places is
C (0) = 0.538848947,
which is rounded in Table 8.1 to 0.538849 merely for ease of presentation. In

other words, on a per-share basis, the firm needs to long 0.538848947 shares of
the stock, which results in longing 50, 000 C (0) shares of the stock when
considering the total. The firm takes out a loan at the risk-free rate to buy
the 0.538848947 shares. This creates a negative position, i.e., a per-share loan
amount of
L0 = 0.538848947 $75 = $40.413671.

Buying the shares also creates a positive position:
C (0) S0 = $40.413671 (per share).
The per-share value of the portfolio at the start is then zero:
VC (0) = $2.264248445 + $2.264248445 $40.413671

+ $40.413671 = 0.000000.
Day 1
We shall use (8.107). Suppose that the stock price is $75.80660915. There are 79
days until expiration and the per-share value of the short sale position is now

ChE = C E $75.80660915, $75, 0.15, 0.02, 79 h, 0 = $2.707914261.
and the investment has grown to a per-share value of
Ih = I0 er h = $2.264372517,
which follows from the more accurate value I0 = 2.264248445.

Since the stock price increased, the delta also increased:
C (h) = 0.598789243.
The extra number of shares needed relative to a per-share basis is
dC (0) = C (h) C (0) = 0.598789243 0.538848947 = 0.059940296.
To buy these shares, borrow

dC (0) Sh = $4.543870582.
The original loan with interest has grown to

0.02
L0 er h = $40.413671 e 365 = $40.415885508.
The per-share loan balance on Day 1 is then

L h = L0 er h + dC (0) Sh = $40.415885508 + $4.543870582 = $44.959756090.
The new value of the long position in the stock is
C (h) Sh = 0.598789243 $75.80660915 = $45.392182107 (per share).

The net value of all the positions is then
VC (h) = $2.707914261 + $2.264372516 $44.9597561 + $45.392182107

= $0.011116 (per share),
which rounds off at six decimal places to the value in the table.
The process from Day 0 to Day 1 is repeated across consecutive days. We
then skip ahead to the day before expiration.
Day 79
Suppose that on Day 79 we have the rounded values given in Table 8.1.
Day 80
The stock price is $93.56147415, which means that the call expires in the money
and will be exercised. The short sale position now has value
CTE = (ST K ) = ($93.56147415 $75) = $18.56147415 (per share).
The value of the investment is
IT = I79 h er h = $2.274195703 (per share).
Note that the above value, which is exposed to more rounding errors due to
the day-to-day delta hedging, still gives the same result to six decimal places
as the more accurate value
I0 er T = $2.264248445 e0.02 365 = $2.274195704.

80
Because the call is in the money, delta is exactly unity:
C ( T ) = 1.
In other words, at expiration the firm has the required 50,000 shares to meet the
obligations of the call being exercised. Since
dC ( T h) = dC (79 h) = C (80 h) C (79 h) = 1 1 = 0,
the balance on the loan to nine decimal places is

L T = L79 h er h + dC (79 h) ST = L79 h er h = $77.32071734 (per share).
The long position in the stock is $87.98 per share since C ( T ) = 1.

The net value to six decimal places of all the positions is
8.7 Option Greeks and Managing Portfolio Risk 433
$0.046522 = VC ( T ) = CTE + IT L T + C ( T ) ST
= ST + K + IT L T + ST

= K + IT L T (per share).
In other words, the proceeds K received from selling the call at strike plus the
amount IT from the risk-free investment are not enough to cover the balance
L T of the loan. The negative value corresponds to a total loss of
50, 000 VC ( T ) = $2, 326.10.
This loss is insignificant compared to the total monies involved in the sale.
In the idealized case of continuously delta hedging, the firm would net zero
when the calls are sold at the BSM price. Note that if the firm had sold the calls
sufficiently higher than the BSM price, then it can even make a profit.
Exercise 8.12 explores a case where the European call in this example expires
out of the money. In this case, we would make use of (8.105).

In practice, each transaction has a variety of costs, including bid-ask spreads,

brokerage commissions, and taxes. Delta hedging can be expensive and diffi-
cult to execute since it requires frequent rebalancing.
8.7 Option Greeks and Managing Portfolio Risk
The price of a European call is more volatile than that of its underlying secu-
rity (Exercise 8.25). A 1% price movement of the underlier can lead to a price
movement in the call that is significantly larger (Exercise 8.8). This section ex-
plores how to make the value of a portfolio of options with the same under-
lying security more stable against small price movements in the underlier. To
accomplish this, we first discuss option Greeks and then apply these ideas to
the construction of delta- and gamma-neutral portfolios.
Unless stated to the contrary, in this section assume for simplicity

that the underlying security pays no dividend (q = 0).
8.7.1 Option Greeks for Portfolios and the BSM p.d.e.
Before generalizing delta neutrality to portfolios, we introduce option Greeks

for a portfolio consisting of options and a security and show that the portfolios
value satisfies the BSM p.d.e.
Consider a portfolio consisting of k options (or derivatives) with the same
underlying security, where the per-unit value of the ith option at time t is
f i (St , t). Let Ni be the number, based on each unit of the underlier, of the ith
option in the portfolio. Here Ni > 0 indicates a long position, Ni = 0 no po-
sition, and Ni < 0 a short position in the ith option. For example, suppose
that the first position in the portfolio is long 300 calls on a stock. Since a call
involves 100 shares of the stock, the number of calls on a per-share basis is
N1 = 300 100 = 30, 000. The value of the portfolio at time t is
V (St , t) = N1 f1 (St , t) + + Nk f k (St , t).
The portfolio also has what are called option Greeks, which define the rate of
change of the portfolio relative to various parameters.
We shall first introduce these rates of change for the options in the portfolio.
Recall that the value of an option is a function of ( x, t) with x representing the
possible prices of the underlier at time t (see page 391). Using rates of change
in x and t, we then define the following three option Greeks of the ith option in
the portfolio :
4
fi f i 44
Delta : i (St , t) = ( St , t) =
S x 4( x, t)=(St , t)
4
2 f i 2 f i 44
Gamma : i (St , t) = ( St , t) =
S2 x2 4( x, t)=(St , t)
4
fi f i 44
Theta : i (St , t) = ( St , t) = .
t t 4( x, t)=(St , t)
The formulas for these option Greeks in the case of calls and puts are given in
Exercise 8.35 on page 473.
Now, viewing the portfolio value V as a function of ( x, t), the option Greeks
extend naturally to the portfolio:
4
V V 44 k
( St , t) =
S
( St , t) = = N ( St , t)
x 4( x, t)=(St , t) i=1 i i
4
2 V 2 V 44 k
( St , t) =
S2
( S t , t ) = = N ( St , t)
x2 4( x, t)=(St , t) i=1 i i
4
V V 44 k
( St , t) =
t
( St , t) = = N ( St , t ).
t 4( x, t)=(St , t) i=1 i i
Note that if the gamma of a portfolio stays sufficiently small, then delta
changes a little as the underlier price changes, which reduces the need for fre-
quent rebalancing in delta hedging.
Turning to the BSM p.d.e., we know that each option in the portfolio satisfies
this p.d.e. (q = 0):
1 2 2 2 f i f f
St 2 (St , t) + r St i (St , t) + i (St , t) r f i (St , t) = 0.
2 x x t
This can then be expressed in terms of option Greeks as follows:
1 2 2
St i (St , t) + r St i (St , t) + i (St , t) r f i (St , t) = 0.
2
Summing over the k options in the portfolio, it follows that V satisfies the BSM
p.d.e.:
1 2 2
St (St , t) + r St (St , t) + (St , t) r V (St , t) = 0. (8.109)
2
As mentioned earlier (page 391), this equation holds for all 0 t T and every
0 < St < .
Now, expand the portfolio to include NS units of the underlying security:
(St , t) = V (St , t) + NS St .
V
The expanded portfolio has option Greeks

2
(St , t) = V (St , t) = (St , t) + NS ,
( St , t ) = V ( St , t ) = ( St , t ),

S S2
and

( S t , t ) = V ( S t , t ) = ( S t , t ) .

t
Since the portfolio value V and the security price St satisfy the BSM p.d.e., the
value of the expanded portfolio does as well:
1 2 2 ( St , t) +
( St , t) r V
(St , t) = 0.
St ( St , t) + r St (8.110)
2
8.7.2 Delta-Neutral Portfolios
We saw that at the core of managing risk from short selling, an option (or
derivative) is delta hedging, i.e., maintaining a long position in a delta num-
ber of units of the underlying security. This process makes the portfolio with a
short position in the option and delta long position in the security delta neutral,
meaning the portfolio value is robust against sufficiently small price move-
ments in the underlier. For instance, suppose that at the current time t, a port-
folio is short 1 European call, and denote its value by
V ( St , t ) = C E ( St , t ).
To makethis portfolio insensitive to a small movement in the underlying se-

curity price over the next instant, long NS shares of the underlying stock. We
now determine the value of NS that will accomplish the desired insensitivity.
Denote the value of the expanded portfolio by
(St , t) = V P (St , t) + NS St = C E (St , t) + NS St .
V
Using Equation (8.33) on page 394 to express the call price as a function of two
variables ( x, t) with the security price as x, we have
P ( x, t) = C E ( x, t) + NS x.
V
Then
4 4

V 44
V C E 44
( St , t) = 4 = + NS (t) = C (t) + NS (t).
S x 4 x 4( x, t)=(St , t)
( x, t)=(St , t)
By choosing a delta hedged position, i.e., selecting NS = C (t), we obtain

V
(St , t) = 0.
S
In other words, a sufficiently small movement in the security price during the
instant from t to t + dt causes little change in the portfolio value. The portfolio
is said to be delta neutral at time t in the sense that it is neutral on whether
the stock has a small price movement up or down. Of course, the number of
shares NS = C (t) that was bought at time t to make the expanded portfolio
delta neutral at t will change as the current time t advances. This means that
the expanded portfolio will have to be continuously rebalanced.
Let us now turn to delta neutrality of the general portfolio of options con-
sidered in Section 8.7.1. We are interested in how an instantaneous change dS
in the price of a security impacts the instantaneous change dV in the value of
the expanded portfolio. Itos formula yields:

1 2 2 2 V
V V
( St , t) =
dV St (St , t) dt + (St , t) dSt + (St , t) dt.
2 x 2 x t
In terms of option Greeks, we obtain:
dV (St , t) dSt + 1
( St , t) = (St , t) dt,
(St , t) (dSt )2 + (8.111)
2
where

( St , t ) = ( St , t ), ( St , t ) = ( St , t ),
(dSt )2 = 2 S2t dt.
S, and t, Equation (8.111)
For sufficiently small incremental changes V,
yields
V (St , t) St + 1
( St , t) (St , t) t.
(St , t) (St )2 + (8.112)
2
Equation (8.112) shows that the impact of a price movement St on the
change V (St , t) in the value of the expanded portfolio can be reduced by mak-
ing the portfolio delta neutral, i.e., make delta vanish:
(St , t) = (St , t) + NS = 0.

This is accomplished by having a position of
NS = (St , t) (8.113)
units in the underlying security. In the special case of a short position in 1

option on 1 unit of the underlier, we have
V ( St , t ) = f 1 ( St , t ), ( St , t ) = 1 ( St , t ),
which gives
NS = (St , t) = 1 (St , t).
In other words, the portfolio is made delta neutral by expanding it to include
a long position with 1 (St , t) units of the underlier. In general, a portfolio of
options is made delta neutral by expanding it to include NS = (St , t) units of the
underlying security.
Example 8.10. (Delta Neutrality) A portfolio has a short position in 500 Eu-
ropean calls and long position in 300 puts, all on the same underlying stock,
which is assumed to pay no dividend. The per-share deltas of each call and put
are 0.5389 and 0.7584, respectively. How to make the portfolio delta neutral?
Solution. On a per-share basis, the number of calls is
N1 = 500 100 = 50, 000
and puts is
N2 = 300 100 = 30, 000.
By (8.113), the portfolio can be made delta neutral by expanding it to include
4,193 shares of the stock:
NS = (St , t) = ( N1 1 (St , t) + N2 2 (St , t))

= (50, 000 0.5389 + 30, 000 0.7584)
= 4, 193.

A delta-neutral portfolio also has an interesting link between its gamma and
theta. By the BSM p.d.e. (8.110), when the expanded portfolio is delta neutral,
we have
1 2 2 ( St , t) = r V
( St , t ).
St ( St , t) +
2
Since the right-hand side r V (St , t) is a fixed value at t, if theta (St , t) has a
sufficiently large positive (resp., negative) value, then it forces gamma ( St , t)
to have a sufficiently large negative (resp., positive) value to maintain the fixed
value on the right. For this reason, theta (St , t) is also interpreted intuitively as a
proxy for gamma (St , t) in a delta-neutral portfolio; see Hull [22, Sec. 19.7]. Note
that theta measures the impact of the change in time on the change in value of
the portfolio, while our discussion is focused on the impact of the underlying
securitys price change on the portfolios change in value. Readers are referred
to Hull [22, Sec. 19.5] and Kwok [26, Sec. 2.1.3] for more on theta.
8.7.3 Delta-Gamma-Neutral Portfolios
A delta-neutral portfolio is robust against a sufficiently small price change S

of the underlier over a sufficiently small time t. Nonetheless, Equation (8.112)
shows that even for a delta-neutral portfolio, a sufficiently nontrivial price
change S can still impact the change V in value of the portfolio through
gamma:
V ( St , t) 1 (St , t) t.
(St , t) (St )2 + (8.114)
2
However, a delta-neutral portfolio can also be made gamma neutral (i.e., gamma
vanishes) by adding an appropriate number of options.
Explicitly, consider a portfolio of options with the same underlying secu-
rity. Assume that the portfolio has value V, delta = 0, and gamma . Also,
identify an option with gamma o = 0 on the same underlier. Assume that the
option is tradable to allow us to take a long or short position in the option.
Create a new portfolio by expanding the delta-neutral portfolio to include No
units of the option. As before, we count the number of options on a per unit
of the underlier; e.g., 10 calls on a stock refer to No = 1, 000 calls counting on a
per-share basis. This gives a new portfolio value at t of

( St , t) = V
V (St , t) + No fo (St , t),
where fo (St , t) is the value of the option at t. The gamma of the new portfolio is

2 V

( St , t) = ( St , t) =
(St , t) + No o (St , t) = 0.
S2
The gamma

(St , t) vanishes exactly when

( St , t)
No = . (8.115)
o ( St , t)
Note that we cannot accomplish this gamma-neutrality by adding a certain
number of units of the underlying security since the gamma of the underlier is
zero.
The reader may have noticed from (8.114) that though No in (8.115) makes
the new portfolio gamma neutral, it causes the delta of the new portfolio to be
nonzero:

( St , t) =
(St , t) + No o (St , t) = No o (St , t),
where (St , t) = 0 and o (St , t) is the delta of the option. To make the new
portfolio delta neutral, we take a position
Nmod = No o (St , t) (8.116)
in the underlying security. The new portfolio then has the following value at t:

mod (St , t) = V
V (St , t) + No fo (St , t) + Nmod St , (8.117)
where No is given by (8.115) and Nmod is fixed. View the value of the modified
portfolio as a function of ( x, t), namely,

mod ( x, t) = V
V ( x, t) + No fo ( x, t) + Nmod x.
Then by (8.115), (8.116), and (8.117), the modified portfolio is both delta neutral
and gamma neutral at ( x, t) = (St , t):

mod (St , t) =
(St , t) + No o (St , t) + Nmod = 0
and

mod (St , t) =
+ No o = 0.
Example 8.11. (Gamma Neutrality) A delta neutral portfolio of options on the

same stock has a gamma = 2, 390 (per $), and a tradable call on the exact
stock has a delta C = 0.4681 and gamma C = 0.3467 (per $). How can we
construct a delta- and gamma-neutral portfolio using the original portfolio and
the call?
Solution. By (8.117), we expand the original delta-neutral portfolio to include

No units of the call per share and Nmod shares of the underlying stock. By
(8.115) and (8.116), we have:

( St , t) 2, 390
No = = = 6, 893.57
C ( St , t) 0.3467
and
Nmod = No o (St , t) = 6, 893.57 0.4681 = 3, 226.88.
In other words, buy 6,893.57 calls on a per-share basis (or 68.9357 calls in round
lots) and short sell 3,226.88 shares of the stock and include these positions
in the original portfolio. The resulting portfolio will be delta- and gamma-
neutral.

Readers are referred to Hull [22, Chap. 19] for more on option Greeks and
their applications.
8.8 The BSM Model Versus Market Data
We shall take a closer look at some of the assumptions of the BSM model to
see how they hold up against market data. A fundamental assumption of the
BSM model is that security prices follow geometric Brownian motion. Two
consequences are that security prices are continuous with probability 1, i.e.,
there are almost surely no jump discontinuities, and the log returns of security
prices are normal. We shall use the S&P 500 index as an example to illustrate
that real-world security prices can jump and exhibit log return behavior that
deviates from normality.
8.8.1 Jumps in Security Prices
Our first observation from the market data is that security prices can have
jumps. In particular, we consider the daily closing prices of the S&P 500 in-
dex from January 3, 1950, to January 2, 2015.17 These prices include the stock
market crash of October 19, 1987, which was marked by a negative jump (a
drop) in the price of the index that day. The opening price was 282.70, which
was also the closing price on the previous trading date of October 16, 1987.
The S&P 500 closed at 224.84 on October 19, 1987, creating a drastic fractional
percentage drop of -20.4669% in the price or a log return drop of -22.8997%.
The negative jump is shown in Figure 8.4.
The daily log returns for the data in the top left panel of Figure 8.4 are shown
in the bottom panel of the figure. The longest negative spike is due to the crash
of October 19, 1987. Note that the second longest spike and the accompany-
ing volatility reflect the 2008 financial crisis, which actually began in 2007 and
peaked in the latter part of 2008. Significant damaging effects spilled into early
2009, and an economic slowdown continued into 2012.
17 See Yahoos historical prices for GSPC.

8.8 The BSM Model Versus Market Data 441
2500 320
310
2000 300
S&P 500 Daily Closing Price
S&P 500 Daily Closing Price

290
1500 280
270
1000 260
250
500 240
230
0 220
1947 1957 1967 1977 1987 1997 2007 13-Oct 14-Oct 15-Oct 16-Oct 19-Oct 20-Oct 21-Oct 22-Oct 23-Oct
Time Time
0.15
0.1
0.05
S&P 500 Daily Log Return
-0.05
-0.1
-0.15
-0.2
-0.25
1947 1957 1967 1977 1987 1997 2007
Time
Fig. 8.4 Top panels: S&P 500 index closing prices from January 3, 1950, to January 2, 2015 (top left)
and from October 13, 1987, to October 23, 1987 (top right). The stock market crash of October 19, 1987,
is clearly seen in the middle panel. The S&P 500 index closed with a negative jump of about -20.5%
relative to the opening price that day, which is the same as the closing price on the previous trading
day of October 16. Note that there was no trading on October 17 and 18 (a weekend). Bottom panel:
daily log returns of the S&P 500 based on prices in the top left panel. The longest negative spike is
due to the stock market crash on October 19. The next longest negative spike is connected with the
2008 financial crisis that became highly pronounced in the latter part of 2008
The evidence for jumps is not restricted to the S&P 500 index, but shows
up in other securities and even in intraday trading. The study of jumps
has become a significant area of research; e.g., see the discussion by Taylor
[41, Sec. 13.6] and references therein.
8.8.2 Skewness and Kurtosis in Security Log Returns
Our second observation is that the log return of security prices is not necessar-
ily normal. Before showing this with S&P 500 data, let us review some of the
moments of a random variable. Let X be a random variable with a p.d.f. f ( x ).
Denote the first moment (i.e., mean) of X by
X = E ( X ),
which measures the probability-weighted center of the possible values of X,

and the second moment (i.e., variance) of X by
X2 = Var( X ),
which measures the dispersion or spread of the possible values of X to the left
and right of the mean E ( X ). In the BSM model, the log return of the price of
a security is normal and so is completely characterized by its first and second
moments. However, we shall see from data that all the higher moments cannot
be ignored.
Skewness
Given that the log return of a security in the BSM model is normal, the shape
of its p.d.f. is symmetric about the vertical line through its center (i.e., mean).
Deviations of the graph of f ( x ) from a symmetric shape about its center can be
measured through the standardized third moment of X. It is called the skewness
of X and defined by
1
X X 3 1
skew( X ) = E = 3 ( x X )3 f ( x ) dx.
X X
When the p.d.f. is symmetric about the center, we have skew( X ) = 0.18 Equiv-
alently, if skew( X ) = 0, then there is a break in the symmetry about the center.
Intuitively, if skew( X ) > 0, then a unimodal (single peak) p.d.f. will have a
more elongated right tail area like the dashed p.d.f. in Figure 8.5. In this case,
we say that f ( x ) is positively skewed. The opposite happens for skew( X ) < 0, in
which case we call f ( x ) negatively skewed; note the stretched left tail area of the
dotted p.d.f. in Figure 8.5.
Kurtosis
The standardized fourth moment of X, called the kurtosis of X, measures the

flatness or sharpness of the central peak of f ( x ) relative to that of a normal
p.d.f. determined by the mean X and standard deviation X of X. The kurtosis
of X is defined by
X 4 1
X x X 4
kurt( X ) = E = f ( x ) dx.
X X
18This is because the portion of skew( X ) from to 0 cancels the part from 0 to . In fact, all odd

power moments E ( X X )2n +1 vanish if the p.d.f. is symmetric about the line x = X .
f (x) f (x) f (x)
0.20
0.3 0.3
0.15
0.2 0.2
0.10
0.1 0.05 0.1
x x x
6 4 2 6 4 2 2 4 6 2 4 6
Fig. 8.5 The three p.d.f.s are due to a skew-normal distribution with location parameter 0, scale pa-
rameter 2, and shape parameters -6 (dotted graph), 0 (solid graph), and 6 (dashed graph). The dotted
graph has a negative skewness of -0.891159, the solid graph has skewness 0 (the skew-normal distri-
bution reduces to a normal in this case), and the dashed graph has a skewness of 0.891159
Since the kurtosis of all normal random variables is 3, it is convenient to use

that kurtosis as a benchmark and define an excess kurtosis by
ekurt( X ) = kurt( X ) 3.
For the cases ekurt( X ) < 0, ekurt( X ) = 0, and ekurt( X ) > 0, the p.d.f. f ( x )
is called platykurtic (platy means flat; think platypus), mesokurtic (meso
means middle), and leptokurtic (lepto means thin), respectively. In interpret-
ing kurtosis of a random variable X, we shall always compare the p.d.f. f ( x )
of X with the p.d.f. of the normal random variable determined by the mean X
and standard deviation X of X. Various kurtosises are illustrated in the left
graphs of Figure 8.6 using three symmetric, unimodal p.d.f.s with identical
mean 0 and standard deviation 1. The middle p.d.f. (solid curve) is mesokurtic
since it is a standard normal. The leptokurtic p.d.f. (dashed curve) has a thinner
peak and thicker (heavier) tails than the standard normal, while the platykur-
tic p.d.f. (dotted curve) has a flatter peak and thinner tail than the standard
normal. In addition, observe that the leptokurtic and platykurtic p.d.f.s inter-
sect the corresponding normal p.d.f. twice on the right of the mean and twice
on the left of the mean. The lower crossing on each side results in a tail thicker
than that of the associated normal; see the right graphs in Figure 8.6. These
properties are not uncommon for unimodal p.d.f.s. See DeCarlo [10] and Ba-
landa and MacGillivray [2] for more.
Remark 8.11. It is important not to confuse effects due to variance with that
due to kurtosis. For example, Figure 8.7 shows the two normal p.d.f.s in com-
parison with the standard normal. Though the dashed p.d.f. has a narrower
peak and the dotted one a flatter peak, all three p.d.f.s have the same excess
kurtosis of 0. Consult DeCarlo [10] for some of the pitfalls in the interpretation
of kurtosis.

f(x)
0.4
0.3
0.2
0.1
x x
3 2 1 1 2 3 1.8 2.0 2.2 2.4 2.6 2.8 3.0
Fig. 8.6 Left plots: the p.d.f.s all have the same mean 0 and standard deviation 1 and are symmetric
about the center (zero skewness). The solid p.d.f. is the standard normal, which has excess kurtosis
0 (mesokurtic). The dotted p.d.f. has excess kurtosis -1 (platykurtic) with a flatter peak and thinner
tail than the standard normal. It is given by a Wigner semicircle density with radius 2 centered at the
origin. The dashed p.d.f. has excess kurtosis 1.2 (leptokurtic) with a thinner peak and thicker tails than

the standard normal. It is given by the logistic density with mean 0 and scale parameter 3/. Right
plots: zoom-in of the right tails of the p.d.f.s in the left plots. Observe that the leptokurtic (dashed)
and platykurtic (dotted) p.d.f.s have tails that are thicker and thinner, respectively, than that of the
associated normal
Skewness and Kurtosis in the S&P 500 Index Daily Log Returns
Turning now to the daily log returns of the S&P 500 based on prices from Jan-
uary 3, 1950, to January 2, 2015, the frequency histogram of the log returns
reveals both asymmetry and leptokurtosis. Figure 8.8 shows the frequency his-
togram. The skewness is -1.0281, which is reflected in the left tail having a more
elongated area than the right tail.
The excess kurtosis is 27.6769, which is significantly above that of the as-
sociated normal. Indeed, the tails of the S&P 500 daily log returns are thicker
(heavier) than those of the corresponding normal. In other words, there is a
higher probability (than in the case of a normal) of having extreme values in
the daily log return. Figure 8.9 shows the thicker tails by zooming in on his-
f (x)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
x
3 2 1 1 2 3
Fig. 8.7 The peakedness and flatness of the dashed and dotted p.d.f.s relative to the standard nor-
mal (solid p.d.f.) should not be confused with leptokurtosis and platykurtosis, respectively. All three
graphs have the same excess kurtosis 0, but different standard deviations: 0.6 (dotted graph), 1 (solid
graph), and 1.4 (dashed graph)
Histogram of S&P 500 Daily Log Return
60
50
40
Frequency
30
20
10
0
-0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1
Daily Log Return
Fig. 8.8 The histogram shows the frequency of the log returns of the S&P 500 index based on prices
from January 3, 1950, to January 2, 2015. The mean and standard deviation are 0.0002945 and 0.0097,
respectively. The solid curve is the p.d.f. of a normal distribution with mean 0.0002945 and standard
deviation 0.0097. The skewness is -1.0281 and the excess kurtosis is 27.6769
togram and using a QQ-plot. A point ( x, y) on the QQ-plot is a pair of quan-

tiles, where x is a quantile of the standard normal and y is the corresponding
quantile in the standardized S&P 500 daily return data. For example, if x is the
30% quantile of the standard normal distribution, which means that there is at
most a 30% probability that X < x and 70% probability that X > x, then the
corresponding y-coordinate will be the 30% quantile of the standardized S&P
daily log returns.19 Indeed, if the standardized S&P daily log returns are stan-
dard normal, then its quantiles will be the same as for the standard normal,
i.e., the QQ-plot will be the 45 line x = y. The deviations of the S&P daily log
returns from normality are evident in Figure 8.9.
Overall, the data in Figures 8.4, 8.8, and 8.9 support that the the daily log
returns of the S&P 500 are not normal, i.e., the prices do not follow geometric
Brownian motion. Deviations of the log returns from normality show up not
only in the S&P 500 but also in common stocks, currencies, etc.; see the intro-
duction by Epps [15, Chaps. 8, 9] and references within.
8.8.3 Volatility Skews
A central assumption of the BSM model is that the inputs St , K, , r, , q are

known. However, the volatility is not market observable. Nonetheless, it can
be inferred from the BSM pricing formula. For example, if we assume that the
BSM call price (8.37), i.e.,
19 See Section 4.2.4 (page 174) for more on quantiles.
Histogram of S&P 500 Daily Log Return Histogram of S&P 500 Daily Log Return
8 8
7 7
Frequency 6 6
Frequency
5 5
4 4
3 3
2 2
1 1
0 0
-0.055 -0.05 -0.045 -0.04 -0.035 -0.03 -0.025 -0.02 -0.015 -0.01 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.055
Daily Log Return Daily Log Return
QQ Plot for Sample S&P 500 Log Returns

15
10
5
Quantiles of Input Sample
-5
-10
-15
-20
-25
-5 0 5
Standard Normal Quantiles
Fig. 8.9 Top: The histograms zoom-in on portion of the left and right tails of the histogram in Fig-
ure 8.8. The solid curve is the p.d.f. of a normal distribution with mean and standard deviation given
by the sample data of the S&P 500 daily log returns. Notice that the tails are thicker than the tails of of
corresponding normal. Bottom: QQ-plot of the standardization of the log returns of the S&P 500 log
returns in Figure 8.8. The left and right tails are heavier than those of the standard normal. The excess
kurtosis is 27.6208

C E (St , K, , r, , q) = St eq N(d+ (St , )) K er N d (St , ) ,
truly models the prices of European calls in the marketplace, then given the
E
current market price Cmarket (t) of a European call, we can solve the equation
E
Cmarket (t) = C E (St , K, , r, , q)
implicitly for . The resulting value of is called the implied volatility and de-
noted by im . In other words, the implied volatility is the volatility that makes
the theoretical BSM call price equal to the market price.
How do we know that one and only one implied volatility corresponds to
the market price of a European call? There is actually a 1-1 correspondence
between the possible prices of a European call and the possible volatilities of
its underlying security. To see this, observe that the vega of the call, which is
defined by
C E
C = ( St , t ),

is positive:
C = St e q N (d+ (St , t))) > 0.
The European call price is a strictly increasing function of . This implies that
for each possible European call price, there is a unique volatility and vice versa.
For this reason, one can freely switch between volatilities and European call
prices. A similar result holds for European puts, which have the same vega as
a European call. Hence, for each market price of a European call (or put), there
is a unique implied volatility, and for each implied volatility, only one market
price can correspond to it.
Unfortunately, implied volatility opens up several concerns with the BSM
model. The BSM model assumes that the volatility of the underlying security
is an inherent property of the underlier that is constant during the life of a call
or put. In other words, according to the BSM model, the value of is not only
unchanging; it is also independent of contractual elements like the options
strike price K, expiration date T, and type (call or put). Explicitly, if the BSM
formulas (8.37) and (8.39) truly model the respective prices of European calls
and puts in the marketplace, then:
The implied volatility im is the same for a European call (or put) with
inputs St , r, , q, but different strike prices Ki . In the BSM model, the graph
of implied volatility as a function of the Ki s is a horizontal line through the
y-value im .
The implied volatility im is the same for a European call (or put) with in-
puts St , K, r, q, but different times i to expiration. In the BSM model, the
graph of implied volatility as a function of the i s is a horizontal line
through the y-value im .
The implied volatility im is the same for a European call and put with the
same underlying security and identical inputs St , r, q, but different strike
prices Ki and times i to maturity.
Unfortunately, all three of the properties above are known to be inconsistent

with data. For example, data on equity options show that the implied volatility
im varies as the strike price and time to expiration change. The resulting plot
of im as a function of (Ki , i ) does not lie on a plane as predicted by the BSM
model, but on a curved discrete surface, called a volatility surface. Moreover, for
a fixed time to expiration, some of the plots of im as a function of Ki even show
shapes resembling smileshence, the name volatility smiles; see Figure 8.10.
In addition, the graph of the implied volatility as a function of strike price
changes as one switches from a European call to a European put with the same
underlier and inputs.
Volatility Smile - SPXPM European Call Option

90
80
70
60
Implied Volatility
50
40
30
20
10
0
1400 1600 1800 2000 2200 2400
Strike Price
Fig. 8.10 Implied volatility in percent versus strike price for the S&P 500 index European call option
with PM settlement (SPXPM). The data is from Yahoo! Finance and based on values at 4:49 p.m. EDT
on June 26, 2015, when the index was at 2,101.59. The expiration date is July 17, 2015. The data shows
that the implied volatility as a function of strike price is not constant as predicted by the BSM model.
The graph has a volatility smile
There is a vast literature on implied volatility. See, for example, the texts by
Hull [22, Chap. 20] and McDonald [27, Chap. 23] and the lecture by Rachev[35].
An extensive introduction to the volatility surface is given by Gatheral [16].
8.9 A Step Beyond the BSM Model: Merton Jump Diffusion
Several proposed modifications of the BSM model allow for more heavy tails,
peakedness, and volatility skews in security prices. By mixing geometric Brow-
nian motion (which is a diffusion) with jump discontinuities, Merton [29] intro-
duced in 1976 a model that naturally extends the BSM theoretical framework
and addresses some of the issues facing the BSM model. We give an introduc-
tion to this model.
For easy reference in the discussion to follow, we recap the BSM theoretical
prices for European calls and puts given in Equations (8.37) and (8.39):

C E (St , K, , r, , q) = St eq N(d+ (St , )) K er N d (St , ) ,
and

P E (St , K, , r, , q) = K er N d (St , ) St eq N(d+ (St , )),
where = T t.
8.9 A Step Beyond the BSM Model: Merton Jump Diffusion 449
8.9.1 Poisson Processes
The number of jumps in the future price of a security is random and assumed
to follow a Poisson process, which we now introducesee, for example, Pri-
vault [34, Chap. 14]. First, we can view the possible outcomes of a Poisson
process at a fixed moment of time in terms of successes and failures. In our
context, define a success at time t as the arrival of information that causes the
price of the security to jump in value. Such information can be important news
pertaining to earnings, sector outlook, serious macroeconomic concerns, etc.
We assume a 1-1 correspondence between successes and the securitys price
jumps and so freely identify a success with a price jump.
Let the current time be 0 and let Nt be the number of price jumps during
[0, t], i.e., over the next t years. The increment Nt Nx , where 0 x < t, is then
the number of price jumps during the time interval ( x, t]. Note that the times
when future price jumps occur are not known a priori. We assume that the
stochastic process {Nt }t0 is a Poisson process, which means that the following
properties hold:
The number of jumps at the starting time 0 is zero, i.e., N0 = 0, and the
mean number of price jumps per year20 is known and denoted by . The
parameter is called the intensity of the Poisson process.
The increments are stationary: for all 0 x < t and all u such that x + u 0
and t + u 0, we have
d
Nt N x = Nt +u N x +u . (8.118)
In other words, the probability of n price jumps during a time period21

T starting now is the same as during a time period of the same length T
starting at any other time. By choosing u = x in (8.118), we can always
shift to an interval starting out at 0:
d
Nt N x = Nt x . (8.119)
The increments are independent: for every sequence of times 0 = t0 < t1 <
< tk , the increments
Nt1 Nt0 , Nt2 Nt1 , ..., N t k N t k 1 ,
20 Recall that our assumed unit of time is a year.

21 Recall that in our usage a time period is the length of a time interval.
are independent. In other words, the number of price jumps during a given
time interval I is independent of the number of price jumps during a time
interval that does not overlap22 with I.
At most one jump can occur during [t, t + dt]:

1 if a jump occurs during [t, t + dt]
dNt = Nt+dt Nt = (8.120)

0 if no jump occurs during [t, t + dt].
In addition, during any instant dt, the probability of one price jump is dt
and the probability of more than one jump is zero:

1 with probability dt

dNt = 0 with probability 1 dt

k > 1 with probability 0.
Using the Ito multiplication rule (dt) a = 0 for a > 0, we see:
EP (dNt ) = dt = Var (dNt ) , (8.121)
where P is the probability measure of the Poisson process.

In general, for a fixed but arbitrary interval (u, t], where u 0, the probabil-
ity that the number Nt Nu of price jumps during (u, t] is n is given by a
Poisson distribution with parameter (t u) (e.g., Durrett [13, Sec. 3.6.3]).
In other words, the probability measure is
n
( t u) ( t u)
P N t Nu = n = e (n = 0, 1, 2, . . . ).
n!
(8.122)
Note that though a Poisson process has discrete values, it is a continuous-time

parameter stochastic process. Additionally, the inter-arrival times of the jumps
of a Poisson process must be independent exponential random variables with
parameter .
8.9.2 The MJD Stochastic Process
Merton models the instantaneous change in the price at a general time t as hav-
ing contributions from a no-jump component determined by geometric Brow-
nian motion and a jump component.
22 Two intervals are nonoverlapping if their interiors are disjoint.
No-Jump Case
When a securitys price has no possibility of jumping, denote its price at t by

St . This case corresponds to = 0 (i.e., the mean number of jumps per year is
zero) and has an instantaneous capital-gain return determined by geometric
Brownian motion:
dSt
= (m q) dt + dBt . (8.123)
St
This is solved by
St = S0 eRW t+Bt ( t 0), (8.124)
where the current time is at 0, S0 is the known current price, and
2
RW = m q .
2
Given that the history of the security price is known up to time t, the expected
instantaneous capital-gain return at t is then:

dSt 44
E 4 Ft = (m q) dt. (8.125)
St
The -algebra Ft in (8.125) is generated by standard Brownian motion B up

to time t, so in the conditional expectation we know St , but not St+dt , which
appears in dSt = St+dt St .
The Merton Jump-Diffusion s.d.e.
Assume it is possible for the security price to have jumps and, for this situa-
tion, denote its price at t by St . Let us explore the instantaneous capital-gain
return when jumps are possible, but not guaranteed. First, if there is no jump
during [t, t + dt], then the Merton model assumes that the capital-gain return
is determined by a geometric Brownian motion:
dSt
= (m
! q) dt + dBt (no jump during [t, t + dt]). (8.126)
St
The instantaneous total mean return m ! of the security should not be confused
with m, which is the instantaneous total mean return of a security with no
possibility of jumps. Second, suppose there is a jump (i.e., discontinuity) in the
price at t. By our assumptions, it is the only jump during [t, t + dt]. Then the
price has two values at t determined by the left and right limits at t. Let St be
the price of the security just before the jump, i.e., the left-hand limit price at t:
St = lim St .
t t
The price of the security at t, given a jump at t, is defined to be the right-hand

limit of the price at t and denoted by St :
St = St+ = lim St .
t t+
Express the price jump as a multiple Jt > 0 of St :
St = Jt St (a jump occurs at t, where Jt > 0). (8.127)
We shall call Jt the jump factor. The percentage change in the price at t due only
to the jump is then
St St
= Jt 1 (due only to the jump at t).
St
We can also think of Jt 1 as the fractional size of the jump at t, where a neg-
ative value is a downward jump. On the other hand, if we do not know that a
jump occurs at t, but only that it is possible, then the percentage change con-
tribution coming only from the jump is modeled by
St St
= ( Jt 1) dNt (contribution only from a possible jump at t).
St
(8.128)
We assume that Jt and dNt are independent. By this assumption and (8.121), the
expected capital gain only from the jump is then

E ( Jt 1) dNt = E ( Jt 1) E (dNt ) = dt, = E ( Jt 1) , (8.129)
where is the mean gain factor in the price due to the jump.
The Merton model assumes that the instantaneous capital-gain return at t is
given by the sum of a no-jump component (8.126) just before t and a possible
jump component (8.128) at t. The two equations can then be combined at a
general time t as follows:
dSt
= (m
! q) dt + dBt + ( Jt 1) dNt ( t > 0). (8.130)
St
Another key assumption of the Merton model is that the risk due to jumps is idiosyn-
cratic (see page 151). For example, it can be due to company-specific events like
the sudden finding of corruption among the senior management that threat-
ens to bring down the company. The model assumes there is no reward or risk
premium for jumps since, as we learned in the Markowitz theory, such risk can
be diversified away. Consequently, the expected instantaneous capital-gain return
at any time t, conditioned on the security price being known up to t , is given by the
case (8.125) when there is no jump possibility:

dSt 44 MJD
E 4F = (m q) dt ( t > 0). (8.131)
St t
In other words, whether the mean jump factor E ( Jt ) is positive or negative

will not impact the expected capital gain. The -algebra FtMJD is generated by
standard Brownian motion, the Poisson process, and the jump-factor process.
Computing the left-hand side of (8.131) using (8.130), it follows:
! = m .
m
Hence,
dSt
= (m q ) dt + dBt + ( Jt 1) dNt (t > 0). (8.132)
St
We call (8.132) a Merton jump-diffusion s.d.e. Note that if there is no jump at t,
then St = St and dNt = 0, while for a jump at t, we have dNt = 1. In other
words,

(m q ) St dt + St dBt given no jump at t
dSt =

(m q ) St dt + St dBt + ( Jt 1) St given a jump at t.
(8.133)
Note that if there is no jump at t, then (8.133) reduces to a geometric Brownian

motion with drift parameter RW and volatility parameter . If there is no
possibility for the security price to jump, then = 0 and dNt = 0, and so (8.133)
reduces to geometric Brownian motion with drift RW and volatility .
Solving the Merton Jump-Diffusion s.d.e.
We shall solve the Merton jump-diffusion s.d.e. by drawing on the fact that
the price between jumps is a geometric Brownian motion, while the price at a
jump time t is Jt times the price just before t.
Suppose that there are price jumps at times T with
0 < T1 < < T < T +1 < < TNt t.
Recall that Nt is the random number of price jumps in [0, t] and so TNt is the
time of the last jump in [0, t]. Note that for times 0 < T1 there are no jumps
in [0, ], while for T < T +1 the intervals [0, T ] and [0, ] have the same
number of jumps, namely, N . Denote the jump factor at T by J = JT for
= 1, . . . , Nt .
Assume that the jump factors J1 , . . . , JNt are i.i.d. and
= E ( J ) 1 ( = 1, . . . , Nt ).
Let us determine the securitys price as varies across [0, t]:
0 < T1
Since no jump occurs during this interval, the security price follows a geo-
metric Brownian motion
S = S0 e(RW ) +B (0 < T 1 ). (8.134)
= T1
The price is ST1 = J1 ST . But the price ST just before T1 has no jumps and
1 1
so is given by taking the left-hand limit of (8.134) at T1 :
ST = S0 e(RW ) T1 +BT1 .
1
Consequently:
ST1 = S0 e(RW ) T1 +BT1 J1 . (8.135)
T1 < < T2
Since there is no jump during this interval, the price at is a geometric
Brownian motion with initial price ST1 :23
S = ST1 e(RW ) (T1 )+ (B BT1 ) (T 1 < < T 2 ).
Equation (8.135) yields:
S = S0 e(RW ) +B J1 (T 1 < < T 2 ). (8.136)
= T2
We have ST2 = J2 ST , where ST is obtained by taking the left-hand limit
2 2
of (8.136) at T2 :
ST = S0 e(RW ) T2 +BT2 J1 .
2
Hence:
ST2 = S0 e(RW ) T2 +BT2 J1 J2 . (8.137)
T Nt 1 < < T Nt
Continuing across the remaining jump times, we have geometric Brownian
motion during the given interval:
(RW ) ( TNt 1 )+ (B BTN )
S = ST N e t 1 (T Nt 1 < < T Nt ),
t 1
where
23 Note: if S( t) = S(0) e t +Bt , then S( t) = S( ) e ( t )+(Bt B ) , where t 0 and 0.

(RW ) TNt 1 +BTN

ST N
t 1
= S0 e t 1 J1 J2 JNt 1 .
Equation (8.136) then extends to:
S = S0 e(RW ) +B J1 J2 JNt 1 (TNt 1 < < TNt ). (8.138)
= T Nt
This is the time of the last price jump in [0, t]. Taking the left limit of (8.138)
at TNt , we get
(RW ) TNt + BTN
ST = S0 e t J1 J2 JNt 1 .
Nt
Equation (8.137) generalizes to:

(RW ) TNt +BTN
STN = JTN ST = S0 e t J1 J2 JNt 1 JNt . (8.139)
t t Nt
T Nt < t
For this situation, a jump does not occur at since the last jump in [0, t] is
at TNt . The price at is then given by a geometric Brownian motion with
initial price STN :
t
(RW ) ( TNt )+ (B BTN )

S = ST N e t (T Nt < t ).
t
By (8.139), we obtain:
S = S0 e(RW ) +B J1 J2 JN 1 JN (T Nt < t ). (8.140)
Equation (8.140) shows that for an interval [0, t] containing a random num-
ber Nt of jumps at times 0 < T1 < < TNt t, the security price at t is
given by
Nt
St = S0 e(RW ) t+Bt J ( t 0), (8.141)
=1
where {Nt }t0 is a Poisson process with intensity and = E ( J ) 1. As
expected, the underliers price process is a mix of jumps and geometric Brow-
nian motion with drift RW and volatility . The stochastic process (8.141)
solves the s.d.e. (8.132) and is called a Merton jump diffusion (MJD).
d
For 0 u t, Equation (8.141) and Btu = Bt Bu imply:

Nt
St d
= exp (RW ) (t u) + Btu + X ,
Su =N +1 u
where X = ln( J ). Shifting the summation index to = Nu and em-

ploying (8.119) (page 449) yield:
N t N u
d
St = Su exp (RW ) (t u) + Btu + X
=1
Nt u
d
= Su exp (RW ) (t u) + Btu + X , (8.142)
=1
where 0 u t. Note that Su in (8.142) is random for u > 0.
Assumptions About the MJD Process
When studying options with underlier process (8.141), we add several as-
sumptions about the MJD process to make the analysis more tractable:
The jump factors J1 , . . . , JNt are i.i.d. lognormal random variables with
J + 12 J2
J = eX , X N ( J , J2 ), E ( J ) = e = + 1, (8.143)
where X is normal with mean J and variance J2 for = 1, . . . , Nt . We also

refer to J as the jump-factor drift parameter and J as the jump-factor volatility
parameter. The mean percentage change of the jump factor is the same for
each jump.
The jump factors are independent of the jump times.
The random variables Bt , Nt , and J are independent, where t > 0 and
= 1, . . . , Nt . In particular, Bt is independent of X for = 1, . . . , Nt .
Notation. Denote the probability measure for the MJD price process (8.141) by
P and, of course, assume that the above properties are enforced.

Remark 8.12. The natural log, ln N Nt
=1 J = =1 X , is an example of a com-
t
pound Poisson process. More generally, a compound Poisson process is a sum

N
=t 1 X , where Nt is a Poisson process and X1 , . . . , XNt are i.i.d. random vari-
ables that are independent of Nt . Furthermore,
the log return of the MJD price
process in Equation (8.141), i.e., ln St
S0 = (RW ) t + Bt + N
=1 X L t ,
t
is an example of a Levy process, while the MJD price process, i.e., St = S0 e Lt , is

an example of an exponential Levy process.

8.9.3 Illustration of MJD Jump, Skewness, and Kurtosis
Consider the MJD security-price process (8.141) over 1 trading year or 252 trad-
1
ing days. The current time is 0, time step is 252 (trading day), final time is t = 1,
and inputs are:
Simulated MJD Prices

1.2
1.1
MJD Security Price 1
0.9
0.8
0.7
0.6
0.5
0 0.2 0.4 0.6 0.8 1
Time
Fig. 8.11 Simulated daily MJD security prices over 1 trading year. A negative price jump occurs in the
transition from trading day 202 (0.8016 years) to trading day 203 (0.8056 years). The price fell from
$1.0207 to $0.6077, which is a 40.46% drop. The inputs of the simulation are given in Equation (8.144)
S0 = $1, RW = 0.1, = 0.15, = 0.5, J = 0.05, J = 0.08, (8.144)
+ 1 2
where = e J 2 J 1. Figure 8.11 depicts a sample path of {St }0t1 over 1
trading year, i.e., 252 trading days. A very pronounced negative jump occurs in
the transition from trading day 202 (0.8016 years), when the price was $1.0207,
to trading day 203 (or 0.8056 years), when the price dropped to $0.6077. It is a
percentage drop of:
St St 0.6077 1.0207
Jt 1 = = = 40.46%.
St 1.0207
The MJD security price model can also produce skewness and kurtosis. Fig-
ure 8.12 shows a histogram of the log returns of a simulated MJD price process
running over 65 years and with the same inputs as in (8.144). The skewness
and excess kurtosis of the simulated MJD prices are -1.3325 and 23.9169, re-
spectively. In other words, the MJD security price model has daily log return
behavior that deviates from normality. Notice the qualitative similarities be-
tween Figure 8.12 and Figure 8.8 (page 445), which shows log returns of the
S&P 500 for prices over a 65-year period.
A QQ-plot of the standardized MJD log returns is shown in Figure 8.13. The
figure gives a clear depiction of the MJD log returns deviating from normal-
ity. It is interesting to observe the qualitative features that Figure 8.13 shares
with the QQ-plot of the standardized S&P 500 log returns shown in Figure 8.9
(page 446). Notice the outlier in the bottom left in both figures. These qualita-
tive comparisons show that the MJD security model can address the difficul-
ties faced by the geometric Brownian motion model of underliers. Naturally, a
Histogram of Simulated MJD Log Returns

45
40
35
Frequency 30
25
20
15
10
0
-0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1
Daily Log Return
Fig. 8.12 Simulated daily log returns for an MJD price process over 65 years. The mean is 0.000424
(practically zero) and standard deviation is 0.0099. The skewness is -1.3325. In particular, the left tail
has a more elongated area (due to more histograms on that side) than the right tail; see the dark lines
along the x-axis. The excess kurtosis is 23.9169, which is much higher than that of the corresponding
normal with mean 0.000424 and standard deviation 0.0099. Compare with Figure 8.8 (page 445)
proper fitting of the MJD model to security prices will not depend on qualita-
tive comparisons, but will involve a detailed statistical investigation.
8.9.4 No-Arbitrage Condition and Market Incompleteness
We shall see in this section that a market with an MJD underlier has no ar-
bitrage, but is incomplete, i.e., not all its derivatives are attainable. In other
words, there is at least one derivative whose payoff cannot be replicated using
a self-financing trading strategy in other securities.
QQ Plot for Simulated MJD Log Returns

10
5
Quantiles of Input Sample
-5
-10
-15
-20
-25
-5 0 5
Standard Normal Quantiles
Fig. 8.13 QQ-plot of the standardization of the simulated MJD log returns in Figure 8.12. The left and
right tails are thicker than that of the standard normal. Compare with Figure 8.9 (page 446)
First, write the MJD security price process in (8.142) as

Nt u
St = Su e( m q 2 ) (tu )+Btu
d 1 2
J , (8.145)
=1
where the index was relabeled to for simplicity. This price is relative to a
real-world probability measure P , , which incorporates our assumptions
on page 456 and takes into account not only geometric Brownian motion but
also the intensity of the jumps and the probability measure of the jump
factors. In particular, Bs is a standard Brownian motion relative to P , . The
conditional expectation of St with respect to P , is (Exercise 8.37):
EP , (St | FuMJD ) = Su e(mq) (tu) (0 u t ). (8.146)
The -algebra FuMJD is generated by the standard Brownian motion {Bs }s0 ,
the Poisson process {Ns }s0 , and the jump-factor process { J }, where =
1, . . . , Ns .
If P , is a risk-neutral measure, then
EP , (St | FuMJD ) = Su e(rq) (tu) (0 u t ), (8.147)
and the price C E (St , t) of a European call on the security would satisfy

EP , C E (St , t) | FuMJD = C E (Su , u) er (tu) (0 u t). (8.148)
This allows us to obtain the current price of the call by setting u = t0 and t = T.
However, comparing (8.146) and (8.147), we see that this is only possible if
m = r.
Since the risk-free rate r is fixed independent of the security, the above con-
straint will not hold for all P , .
There is a transformation from P , to a risk-neutral probability measure
that allows us to price the European call. Though the details are beyond the
scope of this text, we shall sketch the basic idea and refer readers to Privault
[34, Chaps. 14, 15] for more. Choose c and > 0 as well as a probability
measure for the jump factors such that
c = m q (r q ), (8.149)
where
= E ( J 1).
We write the jump factor as J instead of J to indicate that the probability
measure being used is . Then the following can be shown using Girsanov the-
orem for jump processes (see Privault [34, Theorem 14.3, Chap. 15]): there exists a
risk-neutral probability measure, denoted Q c , , , associated with the choice (c , , )

in (8.149) that is equivalent to P , and such that is now the mean annual number
of jumps for the Poisson process, is the probability measure for the jump factors,
and
Bt = Bt + c t
is a standard Brownian motion.
The existence of Q c , , immediately implies via the First Fundamental
Theorem of Asset Pricing that a market with MJD underliers has no arbitrage. In
addition, using (8.149) and Bt , the MJD security price (8.145) becomes
Ntu
(r q 12 2 ) (tu ) + Btu
J .
d
St = Su e (8.150)
=1
An * on Nt and J is a reminder that their probabilities are relative to Q c , , ,

i.e., the Poisson process now has intensity and the jump factor distribution
is now .
As a consistency check, taking the expectation of (8.150) relative to Q c , ,
gives the risk-neutral condition:24
4
EQc , , St 4 FuMJD = Su e(rq) (tu) (0 u t ). (8.151)
Given the risk-neutral measure Q c , , , the price at t of a European call

with respect to it is then:
4
CQE
c , ,
( t ) = e r ( T t ) E Q c , , C E ( S T , T ) 4 F t (0 t T ). (8.152)

This price depends on Q c , , , which in turn is based on a choice (c , , )

satisfying (8.149). Unfortunately, there are infinitely many such choices, and so there
is no unique call price in the MJD model, though there is no arbitrage. The Second
Fundamental Theorem of Asset Pricing then implies that a market with an MJD-
underlier model is incomplete.
Remark 8.13. We saw that for each solution (c , , ) of (8.149), namely,

c = m q (r q), there exists a risk-neutral probability
measure Q c , , . If there is no possibility of a price jump, which is the setting
of the BSM model, then = = = 0 and the constraint equation reduces
to
c = m r.
Consequently, there is a unique solution given by the Sharpe ratio, i.e.,
mr
c = .

Nt u

For St = Su e a0 ( t u ) + Btu J , the desired expectation is e( a0 + 2
d 1 2 +
) ( t u );
24
see Exercise 8.37.
=1
Therefore, there is only one risk-neutral measure, namely, Q mr , 0, 0 , and so

only one call (or derivative) price. In other words, the BSM model has no arbi-
trage and its market is complete.

Further Reading on Pricing Derivatives in an Incomplete Market
For the MJD model, we saw above that though the market has no arbitrage,
there is not a unique risk-neutral probability to price a European call option
with MJD underlier. In fact, there are infinitely many such probability mea-
sures, and so the market is incomplete. How then can one price a derivative in
an arbitrage-free, incomplete market? A practical approach is to have the mar-
ket choose the risk-neutral probability measure, i.e., fit the model to mar-
ket data (market calibration), and employ the resulting measure to price the
derivative. Another approach is to take into account an investors utility func-
tion and then use an associated utility maximization to determine the risk-
neutral probability measure and, hence, price the derivative.
Properly addressing the above important and deep issue is surely beyond
the scope of our text. Readers are referred to the highly informative survey arti-
cle by Staum in Birge and Linetskys handbook [4]. In Chapter 12 of [4], Staum
discusses the meaning25 of market incompleteness and its causes, different ap-
proaches to derivative pricing in incomplete markets (including market cal-
ibration and expected utility maximization), hedging in incomplete markets,
and many other pertinent topics.
8.9.5 Pricing European Calls with an MJD Underlier
Analogous to Section 8.4.4, risk-neutral pricing gives a relatively quick method

of deriving the price of a European call with an MJD underlier.
Consider pricing a European call with expiration T, strike price K, and MJD
underlier. Assume that a risk-neutral probability measure Q c , , is chosen
(e.g., after market calibration). The price at time t of the call is computed using
Equation (8.152). For notational simplicity, we shall write the formula as
E
CMJD (t) = er E (C E ( T ) | Ft ) = er E (max{ST K, 0} | Ft ), (8.153)
where = T t. By (8.150), the underlier price at expiration is

N
ST = St exp ( ) + B + X
d
,
=1
25 Appendices A and B of Chapter 12 in reference [4] treat the meaning of market incompleteness.
where
2
= r q
2
and, analogous to (8.143),
2
E ( J ) = e J + 2 (J ) = + 1.
1
J = eX , X N (J , (J )2 ), (8.154)
We now evaluate (8.153) with the understanding that it is relative to Q c , , .

Assume N = n and let J0 = 1, so X0 = ln J0 = 0. Since X1 , . . . , Xn are i.i.d. with

X N J , (J )2 and because B N (0, ) is independent of each X , we
see that
n
( ) + B + X
=0
is a normal random variable. Its risk-neutral expectation and variance are

N 4
4
E ( ) + B + X 4 N = n = ( ) + n J

=0
and

N 4
4
Var ( ) + B + X 4 N = n = 2 + n (J )2 .
=0
Consequently, the security price at expiration T conditioned on N = n jumps

occurring during [t, T ] is26

ST | N = n
d
= St e m +s B ,

n

n

(8.155)
where m and s are defined by

n

n
m = + n ,

n

J s
( n )2 = 2 +
n 2
( ) .
J
(8.156)
m s
Hence, the conditional security price (8.155) is a lognormal random variable
with mean parameter n and variance parameter ( n )2 .
m
To cast the quantity n in (8.155) in a form analogous to r q 12 2 in the
risk-neutral lognormal security price (8.67) on page 410, define n such that r
m =r

n

n q
( n )2
.
s
2
By (8.143) and (8.156), we get
r
n = r +
n

ln(1 + ). (8.157)
d
26If X N ( a , b2 ), where * indicates a risk-neutral setting, then e X = e a +b Z . Compare with (5.69)
on page 243.
Equation (8.155) then becomes

ST | N
d
= n = St exp r
n
1
s
q ( n )2
2
+ s
n B ,
r s
where n and n play the roles of the risk-free rate and volatility, respectively.
Proceeding as in Section 8.4.4, we apply Theorem 8.3 to evaluate
e r E (max{ST K, 0} | N = n) .

n
s
The result is the price at t of a European call, conditioned on n jumps occurring
during the time remaining until expiration, with strike price K, volatility n ,
and risk-free rate n : r
e r E (max{ST K, 0} | N = n)

n = CBSM
E
(St , K, s , r , , q).

n

n (8.158)
Since the possible values of N are 0, 1, 2, . . . , the unconditional expectation

(8.153) can be expressed as

E
CMJD ( t ) = e r E (max{ST K, 0} | N = n) Q (N = n) .
n =0
Employing (8.158) and the Poisson distribution with intensity , it follows:

n
CMJD (t) =
E

n =0

n!

e( + n r) CBSM
E r
(St , K, n , n , , q). s r
By (8.157), we then get the following theorem:
Theorem 8.4 (MJD Price for a European Call). The price at time t of a European
call with an MJD underlier is given as follows relative to the risk-neutral measure
Q c , , :

CMJD (t) =
E
e n
n =0 n!
E
CBSM s r
(St , K, n , n , , q), (8.159)
where 0 t T and = T t. The current price relative to the choice Q c , , is

obtained at t = 0.
In Theorem 8.4, the quantities with an * are computed relative to Q c , , .

To summarize, the mean number of jumps per year is with
= (1 + ).

The jump factors J are i.i.d., and each is a lognormal random variable J = eX ,

where X N J , (J )2 , and
2
= E ( J ) 1 = e J + 2 (J ) 1,
1
s r
with
n 2 n
( n )2 = 2 +
(J ) ,
n = r + ln(1 + ).

We shall call (8.159) simply the MJD European call price. Put-call parity im-
mediately yields the associated MJD European put price:
E
PMJD (t) = K er St eq + CMJD
E
( t ).
In addition, the BSM European call price can be recovered from the MJD European
call price. Separate out the first-order term in (8.159):
E
CMJD

(t) = e CBSM
E
(St , K, s , r , , q)

0

0
n
s , r , , q).

e
+ n!
E
CBSM (St , K, n n
n =1
r s
Since 0 = r and 0 = , we see that when there is no possibility of a
price jump, i.e., = 0, the BSM formula follows:
E
CMJD (t) = CBSM
E
(St , K, , r, , q).
Let us illustrate the MJD European call price (8.159) in comparison with the
CRR and BSM prices in Example 8.7 on page 421.
Example 8.12. Consider a 3-month European call with strike price of $50 on a
stock whose price follows a Merton jump diffusion. Suppose that the stock is
a nondividend-paying stock with the current price $50 and annual volatility
15%. Let the risk-free rate be 2% per annum. We saw in Example 8.7 that the
BSM price and the 100-period CRR tree price of the European call on one share
is $1.62. What is the MJD European call price if the stock has a mean number
of jumps per year of 0.25, a jump-factor drift parameter value of -1.5%, and a
jump-factor volatility parameter value of 4%?
Solution. The inputs are
S0 = $50, K = $50, = 0.15, r = 0.02, = T = 0.25, q = 0,
with jump parameters
= 0.25, J = 1.5%, J = 4%.
Using a software, Equation (8.159) can be computed to give

E
CMJD (0) = $1.63 > CBSM
E
(0) = $1.62.
The MJD model gives a higher current price than that of the BSM model.

8.10 A Glimpse Ahead 465
8.9.6 MJD Volatility Smile
As noted earlier, the BSM model predicts that the implied volatility im (K ) is
constant as K changes, which does not agree with market data. We also saw
(see page 446) that the BSM call price formula produces a one-to-one corre-
spondence between call prices and implied volatilities. In other words, given
any current European call price C (0), whether it is the market price or even a
theoretical price arising in a model different from the BSM model, we can still
assign a unique implied volatility to the call price, i.e., we can assign the value
of obtained by solving the BSM call price formula for using C (0) as the
input price. In this sense, the unique BSM implied volatility assigned to a call price
can be used as a marker or proxy for the call price.
We now determine the BSM implied volatility associated with the European
call price due to the MJD model, i.e., given an MJD call price CMJDE (0), we solve
the following equation for the BSM implied volatility :

E
CMJD (0) = CBSM
E
(S0 , K, , r, T, q).
We illustrate this below using the call option in Example 8.12.

Example 8.13. (MJD Volatility Smile) A European call on a nondividend-
paying stock has strike price of $50 and three months until expiration. The
stock price is an MJD process with the current price $50, mean number of
jumps per year of 0.25, jump-factor drift parameter value of -1.5%, and jump-
factor volatility parameter value of 4%. Assume a risk-free rate of 2% per an-
num. In Figure 8.14, the implied volatility found using the MJD European call
price (8.159) is plotted as a function of strike price, showing a volatility smile.

Naturally, the accuracy of MJD volatility smiles has to be tested against the
associated market volatility smiles. See, for example, Epps [15, Sec. 9.3] and
references therein, for an introduction.
8.10 A Glimpse Ahead
A rigorous assessment of how well the MJD model for option pricing fits mar-
ket data will require an empirical analysis that is outside the scope of this
text. Nonetheless, the MJD model took an important first step beyond the BSM
model because it can incorporate jumps, skewness, and kurtosis. We also saw
that though there is no arbitrage in the MJD model, its market environment is
incomplete, unlike the BSM model.
As with any model, there are always aspects that need to be modified to
improve the MJDs fit with data. For instance, the BSM and MJD models
Simulated MJD Volatility Smile

0.154
MJD
0.1535 BSM
0.153
0.1525
Implied Volatility
0.152
0.1515
0.151
0.1505
0.15
0.1495
40 45 50 55 60
Stike Price
Fig. 8.14 Implied volatility of a European call as a function of strike price, which the underlying
security has price jumps. Each implied volatility was computed using the MJD European call price
(8.159). The shape shows a volatility smile. The inputs for the model are current underlier price St =
$50, risk-free rate r = 2%, time to expiration of = 0.25 years, dividend yield rate of q = 0, jump-
factor drift parameter value of J = 1.5%, and jump-factor volatility parameter value of J = 4%.
The horizontal line with constant implied volatility of 15% is due to the BSM model with a strike price
of K = $50
assume that the volatility of the underlying security is deterministic. How-

ever, data shows that volatility is stochastic. There are many sources that give
introductions to extensions of the BSM and MJD models that include the fol-
lowing (e.g., see Epps [15], Gathereal [16], and Taylor [41], and the references
therein):
Models with stochastic volatility, but no price jumps (e.g., Heston model)
Models with stochastic volatility and price jumps
Models with stochastic volatility and with jumps in the price and volatility
On the other hand, jumps and stochastic volatility in security prices may
cause market incompleteness.27 In fact, quite an extensive research literature
has been developed around derivative pricing, hedging, expected utility opti-
mization, etc., in incomplete markets. Readers are referred to Staums survey
chapter in the handbook by Birge and Linetsky [4, Chap. 12].
Without a doubt, a whole universe of adventures lies ahead.
27 See Staums article in [4, Chap. 12, Sec. 3]

8.11 Exercises 467
8.11 Exercises
8.1. A modeler who knows nothing about the BSM model is trying to find a
formula for the present value C (0) of a European call option, where the under-
lying security has current price S(0) and the strike price is K. She proposes the
following formula after considerable experimentation:
C (0) = w1 S n (0) + w2 K m ,
where the weights w1 and w2 are to be determined. Without using any infor-
mation about the BSM model, give a two-sentence argument that determines
the possible values of n and m.
8.2. Give a brief intuitive reason why a European call option is more risky than
its underlying security.
8.3. Express the return rate of a European call during an instant dt as a s.d.e.
8.4. If a stock satisfies the CAPM, then does a European call on the stock also
satisfy the CAPM? Justify your answer.
8.5. Traders often abide by simple intuitive rules concerning volatility. Here
are some examples you may have heard:
Sell a stock when its volatility is high.
Favor puts when volatility is high.
Buy a stock when volatility is low.
Are these rules of thumb captured by the BSM model for underliers? Justify
your answer.
8.6. Explain why a European call and put have the same implied volatility.
8.7. Briefly critique the MJD models assumption that the risk of price jumps is
diversifiable.
8.8. (Price Change in Options Versus Stocks) Traders use options for specula-
tion. To get an intuitive feel for why this is the case, we consider an example
of how the price of a European call option changes with variations in the un-
derlying security. A financial companys stock currently has a price of $40. The
risk-free interest rate is 7% per annum and the stock has volatility parameter
of 28%. Consider a European call option on the stock for a strike price of $41
with expiration in 6 months. Let t be the current time and t + h an hour later.
a) From time t to t + h, the price of the stock increases by 1%. What is the
percentage change in the value of the call? Would the price of the put move
by the same percentage?
b) From time t to t + h, the price of the stock decreases by 1%. What is the
percentage change in the value of the call? Would the price of the put move
by the same percentage?
8.9. (European Calls as Insurance) After careful research, a fund manager

would like to purchase 100,000 shares of a nondividend-paying stock currently
trading at $60. The fund manager estimates the stocks volatility at 15% and
believes the stock will rise over the coming months. The $6 million needed to
buy the shares now will not be available until a month away. He is concerned
that if the stock rises over the next month, it will become too expensive to buy.
How can the fund manager insure against the risk of the stock price increas-
ing? How much would this insurance cost? Suppose that the risk-free rate is
2%.
8.10. (European Puts as Insurance) An investor owns 10,000 shares of a stock

paying no dividend and currently trading at $160 per share. The stock has a
volatility of 20% and the current risk-free rate is 2.5%. Three months from now
she would like to liquidate the shares to purchase an investment property for
$1,500,000. She is concerned that if the stock price falls over the next three
months, she would not be able to buy the property. On the other hand, she
does not want to sell her shares now since there is also the possibility that the
stock price will increase over the next three months and so she would miss
out on such gains. How can she mitigate against this risk? Note that since
the portfolio has a single stock, Markowitz portfolio theory does not directly
apply.
8.11. (Warrants) Assume that the equity per share of a company satisfies the
BSM model and has volatility of 25%. Suppose that the current equity value of
the company is $50 million. Assume that its stock pays no dividend and equity
is presently $50 per share. The risk-free rate is 6%. The company plans to issue
300, 000 warrants with strike price of $70 and maturity in 3 years. Each warrant
is based on 1 share of the companys stock. Determine how much money the
company will raise if it sells all the warrants at a fair price.
8.12. (Delta Hedging European Calls That End Out-of-the-Money at Expi-

ration) Assume that a firm sells 1,000 European calls (in round lots) on a
nondividend-paying stock with current price $75, strike $75, annual volatil-
ity of 15%, and 80 days to expiration. Suppose that the risk-free rate is 2%
and assume 365 days in a year. In Table 8.2, some entries are shown for delta
8.11 Exercises 469
Table 8.2 Delta-hedging table for Exercise 8.12.
Delta Hedging Values Per Share
short call investment loan balance long C ( t) shares portf. value
Day 0 $75.000000 -$2.264248 $0.000000
Day 1 $75.472305 0.574123 $2.264373 $0.004812
Day 2 $75.646263
Day 3 $75.523208 $0.029374
Day 4 $ 76.652920
Day 5 $77.036379
.. .. .. .. .. .. .. ..
. . . . . . . .
Day 79 $69.932887 0.000000 $ 0.000000 $ 2.274071 -$2.146712 $0.000000 $ 0.127359
Day 80 $69.904710
hedging based on a MATLAB code that outputs values to at least nine decimal
places, but were rounded at the sixth decimal place so the entries appear less
congested. If you work to six decimal places only, then there naturally will be
rounding errors in the day-to-day delta hedging, and not all your numerical
values will exactly match those in the table. See Example 8.9 on page 428 and
Remark 8.10 in that example.
a) Complete the values for Days 15 and Day 80 in Table 8.2. Assume that the
firm sold the European calls at the BSM price. Did the firm experience a
profit or a loss? Determine how much.
b) If the firm sold the calls at $3.50 per share of the stock, did the firm have a
profit or loss? Determine the amount.
8.13. (Delta Hedging European Calls That End in-the-Money at Expiration)

Suppose that a firm sells 800 European calls (in round lots) on a nondividend-
paying stock with current price $110, strike price $110, annual volatility of 20%,
and 90 days to expiration. Suppose that the risk-free rate is 3% and assume 365
days in a year (see Remark 8.1). Table 8.3 shows a portion of delta hedging us-
ing a MATLAB code; see the comment in Exercise 8.12 about rounding errors.
a) Compute the values for Days 15 and Day 90 in Table 8.3 under the as-
sumption that the firm sold the European calls at the BSM price. Did the
firm experience a profit or a loss? Determine how much.
b) If the firm sold the calls at $5.50 per share of the stock, did the firm have a
profit or loss? Determine the amount.
Table 8.3 Delta-hedging table for Exercise 8.13.
Delta Hedging Values Per Share
short call investment loan balance long C ( t) shares port. value
Day 0 $ 110.000000 $ 4.757733
Day 1 $111.276638 -$0.004965
Day 2 $108.680788 0.499957 - $0.103196
Day 3 $110.757606
Day 4 $110.049030
Day 5 $111.589771 -$0.161611

.. .. .. .. .. .. .. ..
. . . . . . . .
Day 89 $125.092736 1.000000 -$15.101777 $4.792664 - $ 115.392349 $125.092736 -$ 0.608726
Day 90 $124.937041
8.14. For a cum-dividend security price Stc = eq t St , where St follows geometric

Brownian motion, show that dStc = m Stc dt + Stc dBt .

8.15. Show that the self-financing condition dnt Stc+dt + dbt Bt+dt = 0 is
equivalent to dVt = nt dStc + bt dBt .
8.16. Show that if the BSM p.d.e. does not hold, then there is an arbitrage.
8.17. If f ( x, t) is a solution of the BSM p.d.e., then show that for every positive
constant c > 0, the function f c ( x, t) = f (c x, t) is also a solution.
8.18. Given solutions f1 ( x, t), . . . , f n ( x, t) of the BSM p.d.e., show that all linear
combinations c1 f1 ( x, t) + + cn f n ( x, t) are also solutions.
8.19. If a solution f ( x, t) of the BSM p.d.e. has an nth partial derivative with
n f
respect to x, then show that x n ( x, t) is also a solution.
x n
8.20. Show that, for the price process S!t = St eq( T t) , the BSM p.d.e. (8.19) on
page 392 transforms to a form without dividend:
1 2 2 2 C E C E C E
x ( x, t ) + r x ( x, t ) + ( x, t) r C E ( x, t) = 0,
2 x2 x t
where x = x eq( T t) .
8.11 Exercises 471
8.21. Using the variables,

x 2 C E ( x, t) r
x = ln , = ( T t ), v( x, ) = , k = ,
K 2 K (2 /2)
show that (8.22) and (8.23) transform, respectively, to
v 2 v v
( x, ) = 2 ( x, ) + (k 1) ( x, ) k v( x, )
x x
and
v( x, 0) = max{ex 1, 0}, lim v( x, ) = 0, v( x, ) ex as x .
x
8.22. Using a trial solution v( x, ) = u( x, ) ea x+b , show that for the choices
a = 12 (k 1) and b = 14 (k + 1)2 , Equation (8.24) transforms into the heat
equation
u 2 u
( x, ) = 2 ( x, )
x
and (8.25) into
$ 1 %
u( x, 0) = max e 2 (k+1) x e 2 (k1) x , 0 , lim u( x, ) ec x = 0,
1 2
| x|
where c > 0.
8.23. Derive Equations (8.31) and (8.32) on page 394 and show that (8.32) equals

C E ( x, t) = x eq ( T t) N d+ ( x, T t) K er ( T t) N d ( x, T t) .
8.24. Consider the discounted underlier price process {St }t0 , where St =
er t Stc with Stc = eq t St the cum-dividend price process, and a discounted self-
financing, replication portfolio value process {V t }t0 , where V
= er t Vt . Show
m r
a) dSt = St dBQ Q
t , where dBt = dBt + dt.
b) dVt = r Vt dt + nt (m r) St dt + nt St dBt .
c c
c) dVt = nt dSt .
8.25. Consider a nonvanishing stochastic process { Xt }t0 such that
dXt
= a( Xt , t) dt + b( Xt , t) dBt ,
Xt
where a( x, t) and b( x, t) are deterministic functions. The coefficients a( Xt , t)
and b( Xt , t) are called the drift and volatility, respectively, of { Xt }t0 . For exam-
ple, a security price following geometric Brownian motion has constant volatil-
ity b( Xt , t) = . Show that the volatility of a European call is strictly greater
than the volatility of its underlying security.
8.26. Assume that a security satisfies the CAPM. Show that the beta of a Euro-
pean call on the security is strictly greater than the beta of the security.
8.27. Establish the following:
If S( T ) = K, then C (t) 1/2 and P (t) 1/2 as 0.

If S( T ) > K, then C (t) 1 and P (t) 0 as 0.
If S( T ) < K, then C (t) 0 and P (t) 1 as 0.
If t < T, then C (t) < 1 and P (t) > 1.
8.28. Show that the discounted underlier process Xt = e(rq) t St and dis-
counted derivative price process Yt = er t f (St , t) are martingales relative to
the risk-neutral measure Q of Girsanov theorem.
8.29. In a continuous-time approach, we saw that the BSM European option
pricing formula can be derived as the solution of the BSM p.d.e. On the other
hand, the BSM pricing formula can be determined as the continuum limit
of the discrete-time binomial tree model. Is there a discrete-time analog of
the BSM p.d.e. in the binomial tree framework? If so, then using appropriate
discrete-time interpretations, determine the partial difference equation analog
of the BSM p.d.e. directly from the binomial tree.
8.30. Consider the binomial tree model for option pricing.
a) Give a one-sentence mathematical reason why the constraint
d < e(rq) h < u holds. Do not use a specific binomial tree model such as a
CRR tree, JR tree, etc.
b) Give a financial reason why the condition d < e(rq) h < u holds. If this result
does not hold, then is any assumption of the BSM model violated? If so,
indicate which one.
8.31. Using a three-period binomial tree, show that a European call price is
given by
C (t0 , 3) = e(3h)r [ p3 Cu3 + 3p2 (1 p )Cu2 d + 3p (1 p )2 Cud2 + (1 p )3 Cd3 ].
8.32. For an n-period binomial tree, show that the price of a European call is
given by
n
n
C (t0 ) = er (nh) pi (1 p )ni Cui dni (tn ) ,
i =0
i
where Cui dni (tn ) = max{S(t0 )ui dni K, 0} for i = 0, 1, . . . , n.

8.33. Show that the n-period binomial formula for a European call can be ex-
pressed as
C (t0 ) = S(t0 ) eq N(n, k , p ) K er N(n, k , p ),

References 473
where k is the smallest value of i for which S(t0 )ui dni K > 0 and
n
n i p u
= tn t0 , N(n, k , p ) = p (1 p )ni , p = (rq)h .
i=k
i e
8.34. Find the deltas of a forward and futures on a nondividend-paying stock.
8.35. Show that delta, gamma, and theta of European calls and puts are:
C (St0 , t0 ) = eq N(d+ (St0 , )), P (St0 , t0 ) = eq N(d+ (St0 , ))
eq N (d+ (St0 , ))
C ( S t 0 , t0 ) = , P ( S t 0 , t0 ) = C ( S t 0 , t0 )
S t0
St0 eq N (d+ (St0 , ))

C ( S t 0 , t0 ) = + q St0 eq N(d+ (St0 , ))
2
r K er N(d (St0 , ))
St0 eq N (d+ (St0 , ))

P ( S t 0 , t0 ) = q St0 eq N( d+ (St0 , ))
2
+ r K er N( d (St0 , )).
8.36. Consider a portfolio of derivatives with the same underlying security that
pays no dividend. Prove that if the portfolio has zero gamma, then it is theta-
market neutral, meaning P
S = x (St0 , t0 ) = 0.
P
8.37. Consider an MJD security price process, i.e.,

Nt u
St = Su e0 (tu)+ Btu
d
J ,
=1
where 0 u t and 0 = m q 12 2 . Show:

Nt u
a) E J = e ( t u )
=1
b) EP , (St | FuMJD ) = Su e(0 + 2
1 2 + ) (tu )
= Su e( m q ) ( t u ) .
References
[1] Ai, H.: Lecture Notes on Derivatives. Fuqua School of Business, Duke
University, Durham (2008)
[2] Balanda, K., MacGillivray, H.: Kurtosis: a critical review. Am. Stat. 42(2),
111 (1988)
[3] Bemis, C.: The Black-Scholes PDE from Scracth (Lecture Notes). Financial
Mathematics Seminar, University of Minnesota (27 November 2006)
[4] Birge, J., Linetsky, V. (ed.): Handbooks in Operations Research and Man-
agement Science: Financial Engineering, vol. 15, 1st edn. North-Holland,
Amsterdam (2008)
[5] Bjork, T.: Arbitrage Theory in Continuous Time. Oxford University Press,
Oxford (2009)
[6] Black, F.: How we came up with the option formula. J. Portf. Manag. 15(2),
4 (1989)
[7] Black, F., Scholes, M.: The pricing of options and corporate liabilities. J.
Polit. Econ. 81, 637 (May/June 1973)
[8] Chance, D.: Lecture Notes on the Convergence of the Binomial to the
Black-Scholes Model. Louisiana State University, Baton Rouge (2008)
[9] Cox, J., Ross, S.: The valuation of options for alternative stochastic pro-
cesses. J. Financ. Econ. 3, 145 (1976)
[10] DeCarlo, L.: On the meaning and use of kurtosis. Psychol. Methods 2(3),
292 (1997)
[11] Delbaen, G., Schachermayer, W.: A general version of the fundamental
theorem of asset pricing. Math. Ann. 300, 463 (1994)
[12] Demeterfi, K., Derman, E., Kamal, M., Zou J.: More than you ever wanted
to know about volatility swaps. Goldman Sachs, Quantitative Strategies
Research Notes (March 1999)
[13] Durrett, R.: Probability: Theory and Examples, 4th edn. Cambridge Uni-
versity Press, Cambridge (2010)
[14] Elliot, R., Kopp, P.: Mathematics of Financial Markets, 2nd edn. Springer,
New York (2005)
[15] Epps, T.W.: Pricing Derivative Securities. World Scientific, Hackensack
(2007)
[16] Gatheral, J.: The Volatility Surface: A Practitioners Guide. Wiley, Hobo-
ken (2006)
[17] Gray, D., Malone, S.: Macrofinancial Risk Analysis. Wiley, West Sussex
(2008)
[18] Groebner, D., Shannon, P., Fry, P.: Business Statistics: A Decision-Making
Approach. Pearson, Boston (2014)
[19] Harrison, J., Kreps, D.: Martingales and arbitrage in multiperiod securities
markets. J. Econ. Theory 20, 381 (1979)
[20] Harrison, J., Pliska, S.: Martingales and stochastic integrals in the theory
of continuous trading. Stoch. Process. Appl. 11(3), 215 (1981)
[21] Hsia, C.-C.: On binomial option pricing. J. Financ. Res. 6, 41 (1983)
[22] Hull, J.C.: Options, Futures, and Other Derivatives. Pearson Princeton
Hall, Upper Saddle River (2015)
References 475
[23] Kolb, R.W.: Financial Derivatives. New York Institute of Finance,

New York (1993)
[24] Korn, R., Korn E.: Option Pricing and Portfolio Optimization. American
[25] Kreps, D.: Arbitrage and equilibrium economics with infinitely many
commodities. J. Math. Econ. 8, 15 (1981)
[26] Kwok, Y.: Mathematical Models of Financial Derivatives. Springer,
New York (1998)
[27] McDonald, R.: Derivative Markets. Addison-Wesley, Boston (2006)
[28] Merton, R.C.: Theory of rational option pricing. Bell J. Econ. Manag. Sci.
4, 141 (Spring 1973)
[29] Merton, R.C.: Option pricing when underlying stock returns are discon-
tinuous. J. Financ. Econ. 3, 125 (1976)
[30] Merton, R.C.: Applications of option-pricing theory: twenty-five years
later. Am. Econ. Rev. 88(4), 323 (1998) [Nobel Lecture]
[31] Miersemann, E.: Lecture Notes on Partial Differential Equations. Depart-
ment of Mathematics, Leipzig University (2012)
[32] Mikosch, R.: Elementary Stochastic Calculus with Finance in View. World-
Scientific, Singapore (1998)
[34] Privault, N.: Notes on Stochastic Finance. Nanyang Technological Univer-
sity, Singapore (2013)
[35] Rachev, S.: Lecture Notes on Option Pricing. Applied Mathematics and
Statistics, SUNY-Stony Brook, Stony Brook (2011)
[36] Reilly, F.K., Brown, K.C.: Investment Analysis and Portfolio Management.
[37] Ross, S.: A simple approach to the valuation of risky streams. J. Bus. 51,
453 (1978)
[38] Ross, S.: An Elementary Introduction to Mathematical Finance. Cam-
bridge University Press, Cambridge (2011)
[39] Schachermayer, W.: The Fundamental Theorem of Asset Pricing. Faculty
of Mathematics, University of Vienna (2009, Preprint). http://www.
mat.univie.ac.at/schachermayer/preprnts/prpr0141a.
pdf
[40] Scholes, M.: Derivatives in a dynamic environment. Am. Econ. Rev. 88(3),
350 (1998) [Nobel Lecture]
[41] Taylor, S.: Asset Price Dynamics, Volatility, and Prediction. Princeton Uni-
versity Press, Princeton (2005)
(1995)
Index
F -measurable function, 257 bid/ask price, 8

-algebra, 254 binomial pricing
generated by random variable, 258 comparison with BSM pricing, 422
generated by stochastic process, 270 of European calls, 411422
of underliers, 209246
adapted processes, 273, 402 binomial trees
almost surely continuous stochastic process, 264 Cox-Ross-Rubinstein tree, 218
alpha, 192 general, 209218
Amortization, 53 Jarrow-Rudd tree, 251
Annuity, 46 probability measure, 214
application to bond valuation, 66 recombining property, 211
application to equity in a house, 61 Black-Scholes-Merton model, see BSM model
application to sinking funds, 62 bonds, 66
application to stock valuation, 63 bond prices vs. interest rates, 70
applications to saving, borrowing, and bond prices vs. YTM, 72
spending, 59 bond valuation formula, 69
future value, 49 callable, 66
present value, 50 convertible, 66
with varying payments and interest rates, 56 coupon payment, 68
APR, 33 coupon rate, 68
APY, 33 current yield, 68
arbitrage, 84, 334 issue date, 67
asset, 329 maturity date, 67
financial, 329 maturity, par, or face value, 67
at-the-money, 364 par bonds, 67
average value-at-risk, 182 premium bonds, 67
Bankers Rule, 16 yield to maturity (YTM), 68
basis points, 5 zero-coupon bond, 68
bearish, 363 book value, 196
beta, 158, 160 book-to-market ratio, 197
for portfolios, 162, 192 boundary conditions, 391
linear factor beta, 195 for European calls, 392
bid-ask spread, 9 for European puts, 396

Undergraduate Texts in Mathematics and Technology, DOI 10.1007/978-1-4939-3783-7
478 Index
Brownian motion, 282289

with drift, 289 converge
with drift and scaling, 284, 314 almost surely, 265
with starting point, 284 in distribution, 265
Brownian path, 284 in mean square, 265
BSM model, 384391 in probability, 265
unique risk-neutral probability measure, 461 correlation of Brownian motion, 295
versus market data, 440448 covered call, see options
BSM p.d.e., 391 Cox-Ross-Rubinstein tree, see CRR tree
equivalent to heat equation, 393 credit default swaps, see swaps
existence and uniqueness of solutions, 391 CRR tree, 218246
solving for European calls, 392 continuous-time limit, 237246
bullish, 363 CRR equations, 226, 234
real world, 219230
real-world uptick probability, 226, 234
Cameron-Martin-Girsanov theorem, see
risk-neutral uptick probability, 233, 234
Girsanov theorem
risk-neutral world, 230236
Capital Allocation Line (CAL), 154
security price formula, 229, 236
Capital Asset Pricing Model, see CAPM
Capital Market Line (CML), 153157
dealers, 8
tangent pointmarket portfolio, 156
deep in-the-money, 364
CAPM, 152, 158165
deep out-the-money, 364
beta versus linear factor beta, 195
delivery market, 338
for portfolios, 162
delta, 390
formula, 159
discrete version, 415
risk premium of a security, 158
European call and put deltas at expiration, 398
risk premium of the market portfolio, 158
of a European call, 396
security price, 160
of a European put, 397
security risk decomposition, 164 delta hedging, 422433
cash market, 337 application, 427433
coherent risk measure, 184 theoretical framework, 422427
coincident indicators, 10 derivatives
commercial banking, 3 characteristics of its valuation, 332
commodity, 331 commodity, 331
commodity swap, see swaps defined, 330
compound interest, 21 financial, 331
continuous compounding, 31 purposes of, 331
formula, 27 Descartess Rule of Signs, 43
fractional compounding, 28 diffusion coefficient, 315
fractional vs. simple compounding, 30 diffusion equation, see heat equation
future value, 27 diffusion process, 289
generalized compounding, 31 Dirichlet density, 139
nonnegative integer number of periods, 22 discounted price process, 406
nonnegative real number of periods, 24 discrete-state processes, 262
present value, 27 discrete-time processes, 262
conditional expectation, 270273 diversifiable risk, 143, 151
conditional value-at-risk, 182 diversification, see Markowitz portfolio theory
contingent claim, 330 diversified portfolio, see Markowtiz portfolio
continuous stochastic process, 264 theory
continuous-state processes, 262 dividend
continuous-time processes, 262 continuously reinvested, 340
Index 479
cum-dividend, 18, 31, 387 short forward, 337

ex-dividend, 18, 31 spot-forward parity formula, 342
ex-dividend date, 18 underlier, 337
yield, 19, 215 writer, 337
dividend discount model, 64 fundamental factor model, 191
dividend yield, 281 futures, 345348
drift parameter, 386 evolution from forwards to futures, 345
drift process, 302 futures contract, 346
futures price, 346, 347
economic cycle, 10 futures value, 347
economic indicator, 10 impact of daily settlement, 347
efficient frontier, see Markowitz portfolio theory maintenance margin, 347
equity, 330 margin account, 346
equity in a house, 61 margin requirement, 347
equivalent martingale measure, 314 mark-to-market, 346
equivalent measures, 313
European call price, 394 geometric Brownian motion, 314319
behavior relative to security price, 397
Girsanov theorem, 235, 311, 405, 459
behavior relative to strike price, 397
global minimum-variance portfolio, see
European put price, 395
Markowitz portfolio theory
behavior relative to security price, 397
gradient, 120
behavior relative to strike price, 397
Greeks
exact time, 16
delta, 390, 396398, 434
exchanges, 7
for a portfolio, 434
expected short fall, 183
gamma, 434
expected tail loss, 183
theta, 434
vega, 446
Fama-French three-factor model, 196199
gross return, 19
feasible portfolios, 125
federal discount rate, 2
federal funds rate, 2 heat equation, 390, 393
Federal Reserve, 2 hedgers, 331
filtered probability space, 268 Hessian, 120
filtrations, 268270 hurdle rate, 161
final condition, 391
for European calls, 392 i.i.d., 211
for European puts, 396 idiosyncratic risk, 151
financial markets, 1 in-the-money, 364
First Fundamental Theorem of Asset Pricing, 408 incomplete market, 458
forward commitment, 330 index rates, 4
delivery date, 330 innovation process, 279
forwards, 337345 intensity, 449
contract size, 337 interest, 15
delivery price, 338 exact, 16
delivery, expiry, expiration, exercise, or interest rate per period, 15
maturity date, 337 negative interest rate, 15
forward or exercise price, 337 ordinary, 16
forward price formula, 341 quoted rate, 15
forward value formula, 344 total interest on a loan, 56
long forward, 337 intrinsic value, 370
relation to put-call parity, 369 investment banking, 3
480 Index
IRR, 41, 42 global minimum-variance portfolio, 124, 128

multiple, 45 model, 83
relation to NPV, 44 multivariate normality, 87
Ito diffusion, 302 Mutual Fund Theorem, 130
Ito integral, 299302, 314 one-period assumption, 88
Ito process, 302, 387 optimal portfolios, 132
unique representation, 389 portfolio log return, 100103
Ito product rule, 292 portfolio log return versus portfolio return
Itos formula, 302387 rate, 103
Itos lemma, 304 portfolio risk, 96, 119
return rates, 85
jumps in security prices, 440 securities variances and covariances, 96100
two-security portfolio analysis, 104117
kurtosis, 267, 442
utility function, 131
excess, 443
weight vector for minimum-variance
in S&P 500 log returns, 446
portfolio, 123
Levy process, 456 weights, 89
exponential, 456 martingale representation theorem, 407
lagging indicators, 10 martingales, 275278, 406
Lagrange Multiplier Theorem, 122 necessary condition of efficient market, 277
latent factor, 191 maximum drawdown, 172
Law of One Price, 334, 389, 399, 413 Merton jump-diffusion model, see MJD model
consequence of no-arbitrage condition, 335 mesokurtic, 443
leading indicators, 10 method of least squares, 192
leptokurtic, 443 MJD model, 448465
leverage ratio, 3 assumptions, 456
LIBOR, 5, 350 European call pricing, 461464
limit buy/sell order, 8 recovering BSM price, 464
Lindeberg Central Limit Theorem, 237 solving the MJD s.d.e., 453455
Lindeberg condition, 239 volatility smile, 465
linear factor models, 185 money market account, 386
long-term rates, 5 moneyness, 364
multivariate normality, 87
macroeconomic factor model, 191 mutual fund theorem, see Markowiz portfolio
maintenance margin, see futures theory
margin requirement, see futures
mark-to-market accounting rule, 346 naked call, see options
market capitalization, 196 NPV, 38, 42
market liquidity, 9 relation to IRR, 44
market portfolio, 156, 157 numeraire, 386
market risk, 151
market sentiment, 363 observable factor, 191
Markov process, 288 opportunity cost, 16
Markov property, 244, 288 optimization problem, 120
Markowitz bullet, 126 options, 353376, 383466
Markowitz portfolio theory American, 354, 372376
diversification, 138143 buyer, holder, or owner, 353
diversified portfolio, 130 call option, 354
efficient frontier for N securities, 117128 contract size, 353, 356
efficient frontier for two securities, 107117 covered call, 357
expected portfolio return rate, 94 European, 354, 359372, 383466
Index 481
exercise an option, 353 price-to-book ratio, 197

expiration, exercise, or maturity date, 353, 356 primary market, 6, 67
final or terminal payoff, 355 prime rate, 5
how options work, 357359 principal, 15
moneyness, 364 probability measure, 253, 255
naked or uncovered call, 358 probability space, 253
premium, 353 profit diagram
put option, 354 terminal, 359363
seller or writer, 353 put-call parity
strike or exercise price, 353 American options, 373
styles, 354 European options, 362, 368, 395
trading strategies, 365 relation to forward, 369
types, 354
QQ-plot, 445
underlier, 353
quadratic covariation, 291
vanilla, 355
quadratic variation, 289, 290
order statistic, 176
quantile function, 175
out-the-money, 364
quantiles, 445
over-the-counter market (OTC), 7
quoted interest rate, see interest
p-quantile, 174 Radon-Nikodym derivative, 313

p.d.e., see partial differential equations random variables, 257
partial differential equations, 390395 convergence of, 265
connection with probability, 408 independent, 259
parabolic p.d.e., 391 independent of -algebra, 260
payoff diagram random walk, 276, 280, 320
forward, 339 simple, 320
terminal, 359363 symmetric, 320
reserve ratio, 2
perpetuity, 51
return
physical market, 337
arithmetic mean return, 36
platykurtic, 443
capital-gain return, 212
Poisson process, 449
geometric mean return, 36
compound, 456
gross return, 211
portfolio
log return, 212
alpha, 192
required return rate, 16
beta, 162, 192
risk-averse investor, 87, 404
delta-gamma-neutral, 438440
risk-free rate, 16
delta-neutral, 435438
proxy, 17
log return, 101 real, 16
replicating, 388 risk-neutral investor, 137, 404
replicating condition, 389 risk-neutral pricing
risk, 96, 119 of European calls and puts, 410
risk measures, 151, 165180 of European-style derivatives, 404, 408
self-financing, 388, 412 with binomial trees, 415422
self-financing condition, 389 risk-neutral probability measure, 309, 406
trading strategy, 89, 388 and no-arbitrage, 409
weights, 89 for Merton jump-diffusion model, 458
positive definite matrix, 86, 105, 119 uniqueness for BSM model, 461
power set, 254 risk-seeking investor, 136, 404
price discovery, 331 rounding errors, 429
482 Index
s.d.e., see stochastic differential equation for Merton jump diffusion, 453
sample-continuous stochastic process, 264 stochastic processes
scale parameter, 315 basics, 260265
Second Fundamental Theorem of Asset Pricing, Merton jump diffusion, 450458
409 stock valuation, 63
secondary market, 6, 67 straddle, 365
securities strangle, 366
basic behavior, 278282 sub-sigma algebra, 255
cum-dividend price, 386 swap contract, 349
debt securities, 330 swaps, 348353
definition, 329 commodity swaps, 350
derivative securities, 330 credit default swap, 350
equity securities, 330 currency swap, 349
ex-dividend price, 386 fixed leg, 349
securities markets, 6 floating leg, 349
professional participants, 8 interest rate swap buyer, 349
Security Market Line (SML), 163 interest rate swap seller, 349
semivariance, 171 interest rate swaps, 349
Sharpe ratio, 166170, 244 mechanics of interest rate swaps, 351
as slope of CML, 167 notional principal, 349
in BSM model, 460 plain vanilla swap, 349, 350
short selling, 90 swap bank, 351
short-term rates, 4 variance swap, 352
simple interest, 20 systematic risk, 143, 151, 165
formula, 21
future value, 21 tail VaR, 183
present value, 21 time value, 370
return rate, 21 total variation, 289
versus fractional compounding, 30 tower property, 403, 420
sinking funds, 62 trading costs, 9
size premium, 197 trading strategies with options, see options
skewness, 266, 442
in S&P 500 log returns, 445 uncovered call, see options
Sortino ratio, 170, 174 unobservable factor, 191
speculators, 331 unsystematic risk, 143, 165
spot market, 337 utility function, 131137
spot price, 338 concave, 134
spread, 366 convex, 136
bear, 366 marginal utility, 132
bull, 366
butterfly, 367 value premium, 198
calendar, 367 value-at-risk, 178
horizontal, 367 VaR, 177180
price, 366 variance swap, see swaps
time, 367 volatility
vertical, 366 implied, 446
statistical factor model, 191 MJD volatility smile, 465
stochastic differential equation parameter, 386
for cum-dividend security price, 387 skews, 445
for geometric Brownian motion, 386, 387, smiles, 447
451 surface, 447
Index 483
volatility parameter, 315 white noise, 280

volatility process, 302 Gaussian, 280
independent, 280
warrants, 398400 strict, 280
Weak Efficient Market Hypothesis, 244
weights, 89, 118 yield curve, 6

Arlie O. Petters, Xiaoying Dong-An Introduction To Mathematical Finance With Applications - Understanding and Building Financial Intuition-S

Uploaded by

Copyright:

Available Formats

Arlie O. Petters, Xiaoying Dong-An Introduction To Mathematical Finance With Applications - Understanding and Building Financial Intuition-S

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Arlie O. Petters, Xiaoying Dong-An Introduction To Mathematical Finance With Applications - Understanding and Building Financial Intuition-S

Uploaded by

Copyright:

Available Formats

What is the topic of the document?

What is the topic of the document?

What are some of the mathematical concepts discussed?

What are some of the mathematical concepts discussed?

Springer Undergraduate Texts

in Mathematics and Technology

More information about this series at http://www.springer.com/series/7438

Understanding and Building Financial Intuition

ISSN 1867-5506 ISSN 1867-5514 (electronic)

Library of Congress Control Number: 2016939449

Mathematics Subject Classification (2010): 91Gxx

Arlie O. Petters and Xiaoying Dong 2016

Printed on acid-free paper

This Springer imprint is published by Springer Nature

To my dear husband and best friend, Xin Zhou. I could

Rationale and Aim

Given the increasing intricacies and interconnectedness of financial firms ac-

Distinctive Features and Benefits

This book is distinct in how it emphasizes and pedagogically conveys in an

The required mathematics consists of introductory courses on multivariable

Scope and Guide

A year-long course on introductory mathematical finance can be based on the

Durham, NC, USA A.O. Petters

1 Preliminaries on Financial Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 The Time Value of Money . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.6.3 NPV and IRR for General Net Cash Flows . . . . . . . . . . . . . . . 42

3 Markowitz Portfolio Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.5.3 The Mutual Fund Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

4 Capital Market Theory and Portfolio Risk Measures . . . . . . . . . . . . . . 151

5 Binomial Trees and Security Pricing Modeling . . . . . . . . . . . . . . . . . . . . 209

6 Stochastic Calculus and Geometric Brownian Motion Model . . . . . . 253

6.6.1 Motivation, Definition, and Notation . . . . . . . . . . . . . . . . . . . . 289

7 Derivatives: Forwards, Futures, Swaps, and Options . . . . . . . . . . . . . . 329

7.5 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353

8 The BSM Model and European Option Pricing . . . . . . . . . . . . . . . . . . . . 383

8.7 Option Greeks and Managing Portfolio Risk . . . . . . . . . . . . . . . . . . . 433

Financial markets are a part of a financial system. In a modern economy, a fi-

1 See Chapters 3 and 4.

Arlie O. Petters and Xiaoying Dong 2016 1

1.1 A Primer on Banks and Rates

1.1.1 Banks and the Federal Funds Rate

1. A central bank is the monetary authority in a nation.

Most banks operate under a form called fractional reserve banking, in

2. Investment banking provides services such as securities underwriting, merg-

While discussion of the separation of commercial and investment banking

This is why 10% reserve (ratio) is equivalent to 10:1 leverage (ratio).

2. High risk in financial investing is often caused by high leverage. In fact,

1.1.2 Short-Term and Long-Term Rates and Yield Curves

{ index rate } + { spread (or margin)}

Long-term rates are determined by market forces.

Yield Curve Basics

1.2 A Primer on Securities Markets

1.2.1 Securities Markets Organization

those stocks traded on exchanges are referred to as listed stocks.

traded on pink sheets is very speculative.

1.2.2 Professional Participants in Securities Markets

Both clearing houses and securities depositories are important professional

1.2.3 Bid-Ask Spreads and Market Liquidity

1.2.4 Trading Costs

and since r/k = 0.0083 1, we have