MATH3075 3975 Course Notes 2016

MATH3075/3975
FINANCIAL MATHEMATICS
Valuation and Hedging of Financial Derivatives
Christian-Oliver Ewald and Marek Rutkowski

School of Mathematics and Statistics
University of Sydney
Semester 2, 2016
• Students enrolled in MATH3075 are expected to understand and learn all

the material, except for the material marked as (MATH3975).
• Students enrolled in MATH3975 are expected to know all the material.
Related courses:
• MATH2070/2970: Optimisation and Financial Mathematics.
• STAT3011/3911: Stochastic Processes and Time Series.
• MSH2: Probability Theory.
• MSH7: Introduction to Stochastic Calculus with Applications.
• AMH4: Advanced Option Pricing.
• AMH7: Backward Stochastic Differential Equations with Applications.
Suggested readings:
• J. C. Hull: Options, Futures and Other Derivatives. 9th ed. Prentice Hall,
2014.
• S. R. Pliska: Introduction to Mathematical Finance: Discrete Time Models.
Blackwell Publishing, 1997.
• S. E. Shreve: Stochastic Calculus for Finance. Volume 1: The Binomial
Asset Pricing Model. Springer, 2004.
• J. van der Hoek and R. J. Elliott: Binomial Models in Finance. Springer,
2006.
• R. U. Seydel: Tools for Computational Finance. 3rd ed. Springer, 2006.
Contents
1 Introduction 5
1.1 Financial Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Trading in Securities . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Perfect Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 European Call and Put Options . . . . . . . . . . . . . . . . . . . . 9
1.5 Interest Rates and Zero-Coupon Bonds . . . . . . . . . . . . . . . . 10
1.5.1 Discretely Compounded Interest . . . . . . . . . . . . . . . 10
1.5.2 Continuously Compounded Interest . . . . . . . . . . . . . 10
2 Single-Period Market Models 11

2.1 Two-State Single-Period Market Model . . . . . . . . . . . . . . . 12
2.1.1 Primary Assets . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.2 Wealth of a Trading Strategy . . . . . . . . . . . . . . . . . 13
2.1.3 Arbitrage Opportunities and Arbitrage-Free Model . . . . 14
2.1.4 Contingent Claims . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.5 Replication Principle . . . . . . . . . . . . . . . . . . . . . . 17
2.1.6 Risk-Neutral Valuation Formula . . . . . . . . . . . . . . . 19
2.2 General Single-Period Market Models . . . . . . . . . . . . . . . . 21
2.2.1 Fundamental Theorem of Asset Pricing . . . . . . . . . . . 25
2.2.2 Proof of the FTAP (MATH3975) . . . . . . . . . . . . . . . . 26
2.2.3 Arbitrage Pricing of Contingent Claims . . . . . . . . . . . 33
2.2.4 Completeness of a General Single-Period Model . . . . . . 40
3 Multi-Period Market Models 44

3.1 General Multi-Period Market Models . . . . . . . . . . . . . . . . . 44
3.1.1 Static Information: Partitions . . . . . . . . . . . . . . . . . 45
3.1.2 Dynamic Information: Filtrations . . . . . . . . . . . . . . . 47
3.1.3 Conditional Expectations . . . . . . . . . . . . . . . . . . . 49
3.1.4 Self-Financing Trading Strategies . . . . . . . . . . . . . . 52
3.1.5 Risk-Neutral Probability Measures . . . . . . . . . . . . . . 54
3.1.6 Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1.7 Fundamental Theorem of Asset Pricing . . . . . . . . . . . 57
3.1.8 Pricing of European Contingent Claims . . . . . . . . . . . 58
3.1.9 Risk-Neutral Valuation of European Claims . . . . . . . . 61
3
4 MATH3075/3975
4 The Cox-Ross-Rubinstein Model 64

4.1 Multiplicative Random Walk . . . . . . . . . . . . . . . . . . . . . . 64
4.2 The CRR Call Pricing Formula . . . . . . . . . . . . . . . . . . . . 68
4.2.1 Put-Call Parity . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2.2 Pricing Formula at Time t . . . . . . . . . . . . . . . . . . . 69
4.2.3 Replicating Strategy . . . . . . . . . . . . . . . . . . . . . . 70
4.3 Exotic Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.4 American Contingent Claims in the CRR Model . . . . . . . . . . 71
4.4.1 American Call Option . . . . . . . . . . . . . . . . . . . . . . 72
4.4.2 American Put Option . . . . . . . . . . . . . . . . . . . . . . 73
4.4.3 American Contingent Claims . . . . . . . . . . . . . . . . . 74
4.5 Implementation of the CRR Model . . . . . . . . . . . . . . . . . . 78
4.6 Game Contingent Claims (MATH3975) . . . . . . . . . . . . . . . 85
4.6.1 Dynkin Games . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.6.2 Arbitrage Pricing of Game Contingent Claims . . . . . . . 88
5 The Black-Scholes Model 90

5.1 The Wiener Process and its Properties . . . . . . . . . . . . . . . . 90
5.2 Markov Property (MATH3975) . . . . . . . . . . . . . . . . . . . . 92
5.3 Martingale Property (MATH3975) . . . . . . . . . . . . . . . . . . 93
5.4 The Black-Scholes Call Pricing Formula . . . . . . . . . . . . . . . 94
5.5 The Black-Scholes PDE . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.6 Random Walk Approximations . . . . . . . . . . . . . . . . . . . . 99
5.6.1 Approximation of the Wiener Process . . . . . . . . . . . . 99
5.6.2 Approximation of the Stock Price . . . . . . . . . . . . . . . 101
6 Appendix: Probability Review 102

6.1 Discrete and Continuous Random Variables . . . . . . . . . . . . . 102
6.2 Multivariate Random Variables . . . . . . . . . . . . . . . . . . . . 104
6.3 Limit Theorems for Independent Sequences . . . . . . . . . . . . . 105
6.4 Conditional Distributions and Expectations . . . . . . . . . . . . . 106
6.5 Exponential Distribution . . . . . . . . . . . . . . . . . . . . . . . . 107
Chapter 1
Introduction
The goal of this chapter is to give a brief introduction to financial markets.
1.1 Financial Markets

A financial asset (or a financial security) is a negotiable financial instru-
ment representing financial value. Securities are broadly categorised into:
debt securities (such as: government bonds, corporate bonds (debentures),
municipal bonds), equities (e.g., common stocks) and financial derivatives
(such as: forwards, futures, options, and swaps). We present below a tenta-
tive classification of existing financial markets and typical securities traded on
them:
1. Equity market - common stocks (ordinary shares) and preferred stocks
(preference shares).
2. Equity derivatives market - equity options and forwards.
3. Fixed income market - coupon-bearing bonds, zero-coupon bonds, sovereign
debt, corporate bonds, bond options.
4. Futures market - forwards and futures contracts, index futures, futures
options.
5. Interest rate market - caps, floors, swaps and swaptions, forward rate
agreements.
6. Foreign exchange market - foreign currencies, options and derivatives on
foreign currencies, cross-currency and hybrid derivatives.
7. Exotic options market - barrier options, lookback options, compound op-
tions and a large variety of tailor made exotic options.
8. Credit market - corporate bonds, credit default swaps (CDSs), collater-
alised debt obligations (CDOs), bespoke tranches of CDOs.
9. Commodity market - metals, oil, corn futures, etc.
5
6 MATH3075/3975
Stocks (shares), options of European or American style, forwards and futures,

annuities and bonds are all typical examples of modern financial securities
that are actively traded on financial markets. There is also a growing interest
in the so-called structured products, which are typically characterised by a
complex design (and sometimes blamed for the market debacles).
Organised exchanges versus OTC markets.
There are two types of financial markets: exchanges and over-the-counter

(OTC) markets. An exchange provides an organised forum for the buying and
selling of securities. Contract terms are standardised and the exchange also
acts as a clearing house for all transactions. As a consequence, buyers and
sellers do not interact with each other directly, since all trades are done via
the market maker. Prices of listed securities traded on exchanges are pub-
licly quoted and they are nowadays easily available through electronic media.
OTC markets are less strictly organised and may simply involve two institu-
tions such as, for instance, a bank and an investment company. This feature
of OTC transactions has important practical implications. In particular, pri-
vately negotiated prices of OTC traded securities are either not disclosed to
other investors or they are harder to obtain.
Long and short positions.
If you hold a particular asset, you take the so-called long position in that
asset. If, on the contrary, you owe that asset to someone, you take the so-called
short position. As an example, we may consider the holder (long position)
and the writer (short position) of an option. In some cases, e.g. for interest rate
swaps, the long and short positions need to be specified by market convention.
Short-selling of a stock.
One notable feature of modern financial markets is that it is not necessary to

actually own an asset in order to sell it. In a strategy called short-selling,
an investor borrows a number of stocks and sells them. This enables him to
use the proceeds to undertake other investments. At a predetermined time,
he has to buy the stocks back at the market and return them to the original
owner from whom the shares were temporarily borrowed. Short-selling prac-
tice is particularly attractive to those speculators, who make a bet that the
price of a certain stock will fall. Clearly, if a large number of traders do indeed
sell short a particular stock then its market price is very likely to fall. This
phenomenon has drawn some criticism in the last couple of years and restric-
tions on short-selling have been implemented in some countries. Short-selling
is also beneficial since it enhances the market liquidity and conveys additional
information about the investors’ outlook for listed companies.
F INANCIAL M ATHEMATICS 7
1.2 Trading in Securities

Financial markets are tailored to make trading in financial securities efficient.
But why would anyone want to trade financial securities in the first place?
There are various reasons for trading in financial securities:
1. Profits: A trader believes that the price of a security will go up, and
hence follows the ancient wisdom of buying at the low and selling at the
high price. Of course, this naive strategy does not always work. The
literal meaning of investment is “the sacrifice of certain present value
for possibly uncertain future value”. If the future price of the security
happens to be lower than expected then, obviously, a trader makes a loss.
Purely profit driven traders are sometimes referred to as speculators.
2. Protection: Let us consider, for instance, a protection against the uncer-
tainty of exchange rate. A company, which depends on imports, could wish
to fix in advance the exchange rate that will apply to a future trade (for
example, the exchange rate 1 AUD = 1.035 USD set on July 25, 2013 for
a trade taking place on November 10, 2013). This particular goal can be
achieved by entering a forward (or, sometimes, a futures) contract on the
foreign currency.
3. Hedging: In essence, to hedge means to reduce the risk exposure by
holding suitable positions in securities. A short position in one security
can often be hedged by entering a long position in another security. As we
will see in the next chapter, to hedge a short position in the call option, one
enters a short position in the money market account and a long position
in the underlying asset. Traders who focus on hedging are sometimes
termed hedgers.
4. Diversification: The idea that risk, which affects each particular secu-
rity in a different manner, can be reduced by holding a diversified port-
folio of many securities. For example, if one ‘unlucky’ stock in a portfolio
loses value, another one may appreciate so that the portfolio’s total value
remains more stable. (Don’t put all your eggs in one basket!)
The most fundamental forces that drive security prices in the marketplace are
supply (willingness to sell at a given price) and demand (willingness to buy
at a given price). In over-supply or under-demand, security prices will gener-
ally fall (bear market). In under-supply or over-demand, security prices will
generally rise (bull market).
Fluctuations in supply and demand are caused by many factors, including:

market information, results of fundamental analysis, the rumour mill and, last
but not least, the human psychology. Prices which satisfy supply and demand
are said to be in market equilibrium. Hence
Market equilibrium: supply = demand.

8 MATH3075/3975
1.3 Perfect Markets

One of central problems of modern finance is the determination of the ‘fair’
price of a traded security in an efficient securities market. To address this
issue, one proceeds as follows:
1. first, the market prices of the most liquidly traded securities (termed pri-
mary assets) are modelled using stochastic processes,
2. subsequently, the ‘fair’ prices of other securities (the so-called financial
derivatives) are computed in terms of prices of primary assets.
To examine the problem of valuation of derivative securities in some detail, we
need first to make several simplifying assumptions in order to obtain the so-
called perfect market. The first two assumptions are technical, meaning that
they can be relaxed, but the theory and computations would become more com-
plex. Much of modern financial theory hinges on assumption 3 below, which
postulates that financial markets should behave in such a way that the prover-
bial free lunch (that is, an investment yielding profits with no risk of a loss)
should not be available to investors.
We work throughout under the following standing assumptions:

1. The market is frictionless: there are no transaction costs, no taxes or
other costs (such as the cost of carry) and no penalties for short-selling of
assets. The assumption is essential for obtaining a simple dynamics of the
wealth process of a portfolio.
2. Security prices are infinitely divisible, that is, it is possible to buy or
sell any non-integer quantity of a security. This assumption allows us to
make all computations using real numbers, rather than integer values.
3. The market is arbitrage-free, meaning that there are no arbitrage oppor-
tunities available to investors. By an arbitrage opportunity (or simply,
an arbitrage) we mean here the guarantee of certain future profits for
current zero investment, that is, at null initial cost. A viable pricing of
derivatives in a market model that is not arbitrage-free is not feasible.
The existence of a contract that generates a positive cash flow today with no
liabilities in the future is inconsistent with the arbitrage-free property. An im-
mediate consequence of the no-arbitrage assumption is thus the so-called:
Law of One Price: If two securities have the same pattern of

future cash flows then they must have the same price today.
Assumptions 1. to 3. describe a general framework in which we will be able

to develop mathematical models for pricing and hedging of financial deriva-
tives, in particular, European call and put options written on a stock.
1.4 European Call and Put Options

We start by specifying the rules governing the European call option.
Definition 1.4.1. A European call option is a financial security, which gives
its buyer the right (but not the obligation) to buy an asset at a future time T for
a price K, known as the strike price. The underlying asset, the maturity time T
and the strike (or exercise) price K are specified in the contract.
We assume that an underlying asset is one share of the stock and the ma-
turity date is T > 0. The strike price K is an arbitrary positive number. It
is useful to think of options in monetary terms. Observe that, at time T , a
rational holder of a European call option should proceed as follows:
• If the stock price ST at time T is higher than K, he should buy the stock
at time T for the price K from the seller of the option (an option is then
exercised) and immediately sell it on the market for the market price ST ,
leading to a positive payoff of ST − K.
• If, however, the stock price at time T is lower than K, then it does not
make sense to buy the stock for the price K from the seller, since it can
be bought for a lower price on the market. In that case, the holder should
waive his right to buy (an option is then abandoned) and this leads to a
payoff of 0.
These arguments show that a European call option is formally equivalent to
the following random payoff CT at time T
CT := max (ST − K, 0) = (ST − K)+ ,
where we denote x+ = max (x, 0) for any real number x.
In contrast to the call option, the put option gives the right to sell an underlying
asset.
Definition 1.4.2. A European put option is a financial security, which gives
its buyer the right (but not the obligation) to sell an asset at a future time T for
a price K, known as the strike price. The underlying asset, the maturity time T
and the strike (or exercise) price K are specified in the contract.
One can show that a European put option is formally equivalent to the follow-
ing random payoff PT at time T
PT := max (K − ST , 0) = (K − ST )+ .
It is also easy to see that CT − PT = ST − K. This is left an as exercise.
European call and put options are actively traded on organised exchanges.
In this course, we will examine European options in various financial market
models. A pertinent theoretical question thus arises:
What should be the ‘fair’ prices of European call or put options?

10 MATH3075/3975
1.5 Interest Rates and Zero-Coupon Bonds

All participants of financial markets have access to riskless cash through bor-
rowing and lending from the money market (retail banks). An investor who
borrows cash must pay the loan back to the lender with interest at some time
in the future. Similarly, an investor who lends cash will receive from the bor-
rower the loan’s nominal value and the interest on the loan at some time in the
future. We assume throughout that the borrowing rate is equal to the lend-
ing rate. The simplest traded fixed income security is the zero-coupon bond.
Definition 1.5.1. The unit zero-coupon bond maturing at T is a financial
security returning to its holder one unit of cash at time T . We denote by B(t, T )
the price at time t ∈ [0, T ] of this bond; in particular, B(T, T ) = 1.
1.5.1 Discretely Compounded Interest

In the discrete-time framework, we consider the set of dates {0, 1, 2, . . . }. Let a
real number r > −1 represent the simple interest rate over each period [t, t + 1]
for t = 0, 1, 2, . . . . Then one unit of cash invested at time 0 in the money market
account yields the following amount at time t = 0, 1, 2, . . .
Bt = (1 + r)t .
Using the Law of One Price, one can show that the bond price satisfies
Bt
B(t, T ) = = (1 + r)−(T −t) .
BT
More generally, if r(t) > −1 is the deterministic simple interest rate over the
time period [t, t + 1] then we obtain, for every t = 0, 1, 2, . . . ,
∏
t−1 ∏
T −1
Bt = (1 + r(u)), B(t, T ) = (1 + r(u))−1 .
u=0 u=t
1.5.2 Continuously Compounded Interest

In the continuous-time setup, the instantaneous interest rate is modelled
either as a real number r or a deterministic function r(t), meaning that the
money market account satisfies dBt = r(t)Bt dt. Hence one unit of cash invested
at time 0 in the money market account yields the following amount at time t
Bt = ert
or, more generally, (∫ )
t
Bt = exp r(u) du .
0
Hence the price at time t of the unit zero-coupon bond maturing at T equals,
for all 0 ≤ t ≤ T , ( ∫ T )
Bt
B(t, T ) = = exp − r(u) du .
BT t
Chapter 2
Single-Period Market Models
Single period market models are elementary (but useful) examples of financial
market models. They are characterised by the following standing assumptions
that are enforced throughout this chapter:
• Only a single trading period is considered.
• The beginning of the period is usually set as time t = 0 and the end of the
period by time t = 1.
• At time t = 0, stock prices, bond prices, and prices of other financial as-
sets or specific financial values are recorded and an agent can choose his
investment, often a portfolio of stocks and bond.
• At time t = 1, the prices are recorded again and an agent obtains a payoff
corresponding to the value of his portfolio at time t = 1.
It is clear that single-period models are unrealistic, since in market practice

trading is not restricted to a single date, but takes place over many periods.
However, they will allow us to illustrate and appreciate several important eco-
nomic and mathematical principles of Financial Mathematics, without facing
technical problems that are mathematically too complex and challenging.
In what follows, we will see that more realistic multi-period models in discrete
time can indeed be obtained by the concatenation of many single-period models.
Single period models can thus be seen as convenient building blocks when con-
structing more sophisticated models. This statement can be reformulated as
follows:
Single period market models can be seen as ‘atoms’ of

Financial Mathematics in the discrete time setup.
11
12 MATH3075/3975
Throughout this chapter, we assume that we deal with a finite sample space
Ω = {ω1 , ω2 , . . . , ωk }
where ω1 , ω2 , . . . , ωk represent all possible states of the world at time t = 1.
The prices of the financial assets at time t = 1 are assumed to depend on the
state of the world at time t = 1 and thus the asset price may take a different
value for each particular ωi . The actual state of the world at time t = 1 is yet
unknown at time t = 0 as, obviously, we can not foresee the future. We only
assume that we are given some information (or beliefs) concerning the proba-
bilities of various states. More precisely, we postulate that Ω is endowed with
a probability measure P such that P(ωi ) > 0 for all ωi ∈ Ω.
The probability measure P may, for instance, represent the beliefs of an agent.
Different agents may have different beliefs and thus they may use different un-
derlying probability measures to build and implement a model. This explains
why P is frequently referred to as the subjective probability measure. How-
ever, it is also called the statistical (or the historical) probability measure
when it is obtained through some statistical procedure based on historical data
for asset prices. We will argue that the knowledge of an underlying probability
measure P is irrelevant for solving of some important pricing problems.
2.1 Two-State Single-Period Market Model

The most elementary (non-trivial) market model occurs when we assume that
Ω contains only two possible states of the world at the future date t = 1 (see
Rendleman and Bartter (1979)). We denote these states by ω1 = H and ω2 = T .
We may think of the state of the world at time t = 1 as being determined by the
toss of a (possibly asymmetric) coin, which can result in Head (H) or Tail (T),
so that the sample space is given as
Ω = {ω1 , ω2 } = {H, T }.
The result of the toss is not known at time t = 0 and is therefore considered as
a random event. Let us stress that we do not assume that the coin is fair, i.e.,
that H and T have the same probability of occurrence. We only postulate that
both outcomes are possible and thus there exists a number 0 < p < 1 such that
P(ω1 ) = p, P(ω2 ) = 1 − p.
We will sometimes write q := 1 − p.

2.1.1 Primary Assets

We consider a two-state single-period financial model with two primary assets:
the stock and the money market account.
The money market account pays a deterministic interest rate r > −1. We
denote by B0 = 1 and B1 = 1+r the values of the money market account at time
0 and 1. This means that one dollar invested into (or borrowed from) the money
market account at time t = 0 yields a return (or a liability) of 1 + r dollars at
time t = 1.
By the stock, we mean throughout a non-dividend paying common stock,

which is issued by a company listed on a stock exchange. The price of the stock
at time t = 0 is assumed to be a known positive number, denoted by S0 . The
stock price at time t = 1 depends on the state of the world and can therefore
take two different values S1 (ω1 ) and S1 (ω2 ), depending on whether the state of
the world at time t = 1 is represented by ω1 or ω2 . Formally, S1 is a random
variable, taking the value S1 (ω1 ) with probability p and the value S1 (ω2 ) with
probability 1 − p. We introduce the following notation:
S1 (ω1 ) S1 (ω2 )
u := , d := ,
S0 S0
where, without loss of generality, we assume that 0 < d < u. The evolution of
the stock price under P can thus be represented by the following diagram:
8 S0 u
pppppp
pp
ppppp
p
S0 NN
NNN
NN1−p
NNN
NNN
&
S0 d
2.1.2 Wealth of a Trading Strategy

To complete the specification of a single-period market model, we need to in-
troduce the concept of a trading strategy (also called a portfolio). An agent
is allowed to invest in the money market account and the stock. Her portfolio
is represented by a pair (x, φ), which is interpreted as follows:
• x ∈ R is the total initial endowment in dollars at time t = 0,
• φ ∈ R is the number of shares purchased (or sold short) at time t = 0,
• an agent invests the surplus of cash x−φS0 in the money market account if
x−φS0 > 0 (or borrows cash from the money market account if x−φS0 < 0).
14 MATH3075/3975
Let us emphasise that we postulate that φ can take any real value, i.e., φ ∈ R.
This assumption covers, for example, the short-selling of stock when φ < 0, as
well as taking an arbitrarily high loan (i.e., an unrestricted borrowing of cash).
The initial wealth (or initial value) of a trading strategy (x, φ) at time t = 0
is clearly x, the initial endowment. Within the period, i.e., between time t = 0
and time t = 1, an agent does not modify her portfolio. The terminal wealth
(or terminal value) V (x, φ) of a trading strategy at time t = 1 is given by the
total amount of cash collected when positions in shares and the money market
account are closed. In particular, the terminal wealth depends on the stock
price at time t = 1 and is therefore random. In the present setup, it can take
exactly two values:
V1 (x, φ)(ω1 ) = (x − φS0 )(1 + r) + φS1 (ω1 ),
V1 (x, φ)(ω2 ) = (x − φS0 )(1 + r) + φS1 (ω2 ).
Definition 2.1.1. The wealth process (or the value process) of a trading
strategy (x, φ) in the two-state single-period market model is given by the pair
(V0 (x, φ), V1 (x, φ)), where V0 (x, φ) = x and V1 (x, φ) is the random variable on Ω
given by
V1 (x, φ) = (x − φS0 )(1 + r) + φS1 .
2.1.3 Arbitrage Opportunities and Arbitrage-Free Model
An essential feature of an efficient market is that if a trading strategy can turn

nothing into something, then it must also run the risk of loss. In other words,
we postulate the absence of free lunches in the economy.
Definition 2.1.2. An arbitrage (or a free lunch) is a trading strategy that
begins with no money, has zero probability of losing money, and has a positive
probability of making money.
This definition is not mathematically precise, but it conveys the basic idea of
arbitrage. A more formal definition is the following:
Definition 2.1.3. A trading strategy (x, φ) in the two-state single-period market
model M = (B, S) is said to be an arbitrage opportunity (or arbitrage) if
1. x = V0 (x, φ) = 0 (that is, a trading strategy needs no initial investment),
2. V1 (x, φ) ≥ 0 (that is, there is no risk of loss),
3. V1 (x, φ)(ωi ) > 0 for some i (that is, strictly positive profits are possible).
Under condition 2., condition 3. is equivalent to
EP (V1 (x, φ)) = pV1 (x, φ)(ω1 ) + (1 − p)V1 (x, φ)(ω2 ) > 0.
Real markets do sometimes exhibit arbitrage, but this is necessarily temporary

and limited in scale; as soon as someone discovers it, the forces of supply and
demand take actions that remove it. A model that admits arbitrage cannot be
used for an analysis of either pricing or portfolio optimisation problems and
thus it is not viable (since a positive wealth can be generated out of nothing).
The following concept is crucial:
A financial market model is said to be arbitrage-free whenever

no arbitrage opportunity in the model exists.
To rule out arbitrage opportunities from the two-state model, we must assume
that d < 1 + r < u. Otherwise, we would have arbitrages in our model, as we
will show now.
Proposition 2.1.1. The two-state single-period market model M = (B, S) is
arbitrage-free if and only if d < 1 + r < u.
Proof. If d ≥ 1 + r then the following strategy is an arbitrage:
• begin with zero initial endowment and at time t = 0 borrow the amount
S0 from the money market in order to buy one share of the stock.
Even in the worst case of a tail in the coin toss (i.e., when S1 = S0 d) the stock at
time t = 1 will be worth S0 d ≥ S0 (1 + r), which is enough to pay off the money
market debt. Also, the stock has a positive probability of being worth strictly
more then the debt’s value, since u > d ≥ 1 + r, and thus S0 u > S0 (1 + r).
If u ≤ 1 + r then the following strategy is an arbitrage:

• begin with zero initial endowment and at time t = 0 sell short one share
of the stock and invest the proceeds S0 in the money market account.
Even in the worst case of a head in the coin toss (i.e., when S1 = S0 u) the cost
S1 of repurchasing the stock at time t = 1 will be less than or equal to the value
S0 (1 + r) of the money market investment at time t = 1. Since d < u ≤ 1 + r,
there is also a positive probability that the cost of buying back the stock will be
strictly less than the value of the money market investment. We have therefore
established the following implication:
No arbitrage ⇒ d < 1 + r < u.
The converse is also true, namely,
d < 1 + r < u ⇒ No arbitrage.
The proof of the latter implication is left as an exercise (it will also follow from
the discussion in Section 2.2).
16 MATH3075/3975
Obviously, the stock price fluctuations observed in practice are much more com-
plicated than the mouvements predicted by the two-state single-period model
of the stock price. We examine this model in detail for the following reasons:
• Within this model, the concept of arbitrage pricing and its relationship to
the risk-neutral pricing can be easily appreciated.
• A concatenation of several single-period market models yields a reason-
ably realistic model (see Chapter 4), which is commonly used by practi-
tioners. It provides a sufficiently precise and computationally tractable
approximation of a continuous-time market model.
2.1.4 Contingent Claims

Let us first recall the definition of the call option. It will be considered as a
standard example of a derivative security, although in several examples we will
also examine the digital call option (also known as the binary call option),
rather than the standard (that is, plain vanilla) European call option.
Definition 2.1.4. A European call option is a financial security, which gives
its buyer the right, but not the obligation, to buy one share of stock at a future
time T for a price K from the option writer. The maturity date T and the strike
price K are specified in the contract.
We already know that a European call option is formally equivalent to a con-
tingent claim X represented by the following payoff at time T = 1
X = C1 := max (S1 − K, 0) = (S1 − K)+
where we denote x+ = max (x, 0) for any real number x. More explicitly,
C1 (ω1 ) = (S1 (ω1 ) − K)+ = (uS0 − K)+ ,
C1 (ω2 ) = (S1 (ω2 ) − K)+ = (dS0 − K)+ .
Recall also that a European put option is equivalent to a contingent claim X,
which is represented by the following payoff at time T = 1
X = P1 := max (K − S1 , 0) = (K − S1 )+ .
The random payoff CT (or PT ) tells us what the call (or put) option is worth
at its maturity date T = 1. The pricing problem thus reduces to the following
question:
What is the ‘fair’ value of the call (or put) option at time t = 0?
In the remaining part of this section, we will provide an answer to this question
based on the idea of replication. Since we will deal with a general contingent
claim X, the valuation method will in fact cover virtually any financial contract
one might imagine in the present setup.
To solve the valuation problem for European options, we will consider a more
general contingent claim, which is of the type X = h(S1 ) where h : R+ →
R is an arbitrary function. Since the stock price S1 is a random variable on
Ω = {ω1 , ω2 }, the quantity X = h(S1 ) is also a random variable on Ω taking the
following two values
X(ω1 ) = h(S1 (ω1 )) = h(S0 u),
X(ω2 ) = h(S1 (ω2 )) = h(S0 d),
where the function h is called the payoff profile of the claim X = h(S1 ). It is
clear that a European call option is obtained by choosing the payoff profile h
given by: h(x) = max (x−K, 0) = (x−K)+ . There are many other possible choices
for a payoff profile h leading to different classes of traded (exotic) options.
2.1.5 Replication Principle

To price a contingent claim, we employ the following replication principle:
• Assume that it is possible to find a trading strategy replicating a con-

tingent claim, meaning that the trading strategy guarantees exactly the
same payoff as a contingent claim at its maturity date.
• Then the initial wealth of this trading strategy must coincide with the
price of a contingent claim at time t = 0.
• The replication principle can be seen as a consequence of the Law of One
Price, which in turn is known to follow from the no-arbitrage assumption.
• We thus claim that, under the postulate that no arbitrage opportunities
exist in the original and extended markets, the only possible price for the
contingent claim is the initial value of the replicating strategy (provided,
of course, that such a strategy exists).
Let us formalise these arguments within the present framework.

Definition 2.1.5. A replicating strategy (or a hedge) for a contingent claim
X = h(S1 ) in the two-state single period market model is a trading strategy
(x, φ) which satisfies V1 (x, φ) = h(S1 ) or, more explicitly,
(x − φS0 )(1 + r) + φS1 (ω1 ) = h(S1 (ω1 )), (2.1)
(x − φS0 )(1 + r) + φS1 (ω2 ) = h(S1 (ω2 )). (2.2)
From the above considerations, we obtain the following result.
Proposition 2.1.2. Let X = h(S1 ) be a contingent claim in the two-state single-
period market model M = (B, S) and let (x, φ) be a replicating strategy for
X. Then x is the only price for the claim at time t = 0, which does not allow
arbitrage in the extended market in which X is a traded asset.
We write x = π0 (X) and we say that π0 (X) is the arbitrage price of X at
time 0. This definition will be later extended to more general market models.
18 MATH3075/3975
In the next proposition, we address the following question:
How to find a replicating strategy for a given claim?
Proposition 2.1.3. Let X = h(S1 ) be an arbitrary contingent claim. The pair

(x, φ) given by
h(S1 (ω1 )) − h(S1 (ω2 )) h(uS0 ) − h(dS0 )
φ= = (2.3)
S1 (ω1 ) − S1 (ω2 ) S0 (u − d)
and
1 ( )
π0 (X) = pe h(S1 (ω1 )) + qe h(S1 (ω2 )) (2.4)
1+r
is the unique solution to the hedging and pricing problem for X.
Proof. We note that equations (2.1) and (2.2) represent a system of two linear
equations with two unknowns x and φ. We will show that it has a unique solu-
tion for any choice of the function h. By subtracting (2.2) from (2.1), we compute
the hedge ratio φ, as given by equality (2.3). This equality is often called the
delta hedging formula since the hedge ratio φ is frequently denoted as δ.
One could now substitute this value for φ in equation (2.1) (or equation (2.2))
and solve for the initial value x. We will proceed in a different way, however.
First, we rewrite equations (2.1) and (2.2) as follows:
( )
1 1
x+φ S1 (ω1 ) − S0 = h(S1 (ω1 )), (2.5)
1+r 1+r
( )
1 1
x+φ S1 (ω2 ) − S0 = h(S1 (ω2 )). (2.6)
1+r 1+r
Let us denote
1+r−d u − (1 + r)
pe := , qe := 1 − pe = . (2.7)
u−d u−d
Since we assumed that d < 1 + r < u, we obtain 0 < pe < 1 and 0 < qe < 1. The
following relationship is worth noting
( )
1 ( ) 1 1+r−d u − (1 + r)
peS1 (ω1 ) + qeS1 (ω2 ) = S0 u + S0 d
1+r 1+r u−d u−d
(1 + r − d)u + (u − (1 + r))d)
= S0 = S0 .
(u − d)(1 + r)
Upon multiplying equation (2.5) with pe and equation (2.6) with qe and adding
them, we obtain
( )
1 ( ) 1 ( )
x+φ peS1 (ω1 ) + qeS1 (ω2 ) − S0 = pe h(S1 (ω1 )) + qe h(S1 (ω2 )) .
1+r 1+r
By the choice of pe and qe, the equality above reduces to (2.4)
From Proposition 2.1.3, we conclude that we can always find a replicating strat-
egy for a contingent claim in the two-state single-period market model. Models
that enjoy this property are called complete. We will see that some models
are incomplete and thus the technique of pricing by the replication principle
fails to work for some contingent claims.
2.1.6 Risk-Neutral Valuation Formula

We define the probability measure P e on Ω = {ω1 , ω2 } by setting P(ω
e 1 ) = pe and
e 2 ) = 1 − pe. Then equation (2.4) yields the following result.
P(ω
Proposition 2.1.4. The arbitrage price π0 (X) admits the following probabilis-
tic representation
( ) ( )
π0 (X) = EPe X
1+r
= EPe h(ST )
1+r
. (2.8)
We refer to equation (2.8) as the risk-neutral valuation formula. In par-

ticular, (as we already
) observed in the proof of Proposition 2.1.3, the equality
S0 = EPe 1+r ST holds.
1
• The probability measure P e is called a risk-neutral probability mea-

sure, since the price of a contingent claim X only depends on the expecta-
tion of the payoff under this probability measure, and not on its riskiness.
• As we will see, the risk-neutral probability measure will enable us to com-
pute viable prices for contingent claims also in incomplete markets, where
pricing by replication is not always feasible.
Remark 2.1.1. It is clear that the price π0 (X) of a contingent claim X does not
depend on subjective probabilities p and 1 − p. In particular, the arbitrage price
of X usually does not coincide with the expected value of the discounted payoff
of a claim under the subjective probability measure P, that is, its actuarial
value. Indeed, in general, we have that
( ) ( )
h(ST ) h(ST ) 1 ( )
π0 (X) = EPe ̸= EP = ph(S1 (ω1 )) + qh(S1 (ω2 )) .
1+r 1+r 1+r
Note that the inequality above becomes equality only if either:
(i) the equality p = pe holds (and thus also q = qe) or
(ii) h(S1 (ω1 )) = h(S1 (ω2 )) so that the claim X is non-random.
Using (2.8) and the equality CT − PT = ST − K, we obtain the put-call parity
relationship
Price of a call − Price of a put = C0 − P0 = S0 − 1

1+r
K. (2.9)
20 MATH3075/3975
Remark 2.1.2. Let us stress that the put-call parity is not specific to a single-
period model and it can be extended to any arbitrage-free multi-period model.
It suffices to rewrite (2.9) as follows: C0 − P0 = S0 − KB(0, T ). More generally,
in any multi-period model we have
Ct − Pt = St − KB(t, T )
for t = 0, 1, . . . , T where B(t, T ) is the price at time t of the unit zero-coupon

bond maturing at T .
Example 2.1.1. Assume the parameters in the two-state market model are
given by: r = 13 , S0 = 1, u = 2, d = .5 and p = .75. Let us first find the price of
the European call option with strike price K = 1 and maturity T = 1. We
start by computing the risk-neutral probability pe
1+r−d 1 + 13 − 1
5
pe = = 2
= .
u−d 2 − 12 9
Next, we compute the arbitrage price C0 = π0 (C1 ) of the call option, which is
formally represented by the contingent claim C1 = (S1 − K)+
( ) ( ) ( )
C1 (S1 − K)+ 1 5 4 15
C0 = EPe = EPe = 1 · (2 − 1) + · 0 = .
1+r 1+r 1+ 3 9 9 36
This example makes it obvious once again that the value of the subjective prob-
ability p is completely irrelevant for the computation of the option’s price. Only
the risk-neutral probability pe matters. The value of pe depends in turn on the
choice of model parameters u, d and r.
Example 2.1.2. Using the same parameters as in the previous example, we
will now compute the price of the European put option (i.e., an option to sell
a stock) with strike price K = 1 and maturity T = 1. A simple argument shows
that a put option is represented by the following payoff P1 at time t = 1
P1 := max (K − S1 , 0) = (K − S1 )+ .
Hence the price P0 = π0 (P1 ) of the European put at time t = 0 is given by
( ) ( ) ( )
P1 (K − S1 )+ 1 5 4 ( 1) 1
P0 = EPe = EPe = 1 ·0+ · 1− = .
1+r 1+r 1+ 3
9 9 2 6
Note that the prices at time t = 0 of European call and put options with the
same strike price K and maturity T satisfy
15 1 1 1 1
C0 − P0 = − = =1− 1 = S0 − K.
36 6 4 1+ 3
1+r
This result is a special case of the put-call parity relationship.
2.2 General Single-Period Market Models

In a general single-period model M = (B, S 1 , . . . , S n ):
• The money market account is modeled in the same way as in Section 2.1,
that is, by setting B0 = 1 and B1 = 1 + r.
• The price of the ith stock at time t = 0 (resp., at time t = 1) is denoted by
S0j (resp., S1j ). The stock prices at time t = 0 are known, but the prices the
stocks will have at time t = 1 are not known at time t = 0, and thus they
are considered to be random variables.
We assume that the observed state of the world at time t = 1 can be any of the
k states ω1 , . . . , ωk , which are collected in the set Ω, so that
Ω = {ω1 , . . . , ωk }. (2.10)
We assume that a subjective probability measure P on Ω is given. It tells us

about the likelihood P(ωi ) of the world being in the the ith state at time t = 1,
as seen at time t = 0. For each j = 1, . . . , n, the value of the stock price S1j at
time t = 1 can thus be considered as a random variable on the state space Ω,
that is,
S1j : Ω → R.
Then the real number S1j (ω) represents the price of the jth stock at time t = 1
if the world happens to be in the state ω ∈ Ω at time t = 1. We assume that
each state at time t = 1 is possible, that is, P(ω) > 0 for all ω ∈ Ω.
Let us now formally define the trading strategies (or portfolios) that are
available to all agents.
Definition 2.2.1. A trading strategy in a single-period market model M is a
pair (x, φ) ∈ R × Rn where x represents the initial endowment at time t = 0
and φ = (φ1 , . . . , φn ) ∈ Rn is an n-dimensional vector, where φj specifies the
number of shares of the jth stock purchased (or sold) at time t = 0.
Given a trading strategy (x, φ), it is always assumed that the amount
∑
n
φ := x −
0
φj S0j
j=1
is invested at time t = 0 in the money market account if it is a positive number

(or borrowed if it is a negative number). Note that this investment yields the
(positive or negative) cash amount φ0 B1 at time t = 1.
22 MATH3075/3975
Definition 2.2.2. The wealth process of a trading strategy (x, φ) in a single-

period market model M is given by the pair (V0 (x, φ), V1 (x, φ)) where
∑n
V0 (x, φ) := φ0 B0 + j=1 φj S0j = x (2.11)
and V1 (x, φ) is the random variable given by
∑n ( ∑n ) ∑n
V1 (x, φ) := φ0 B1 + j=1 φ j j
S 1 = x − j=1 φ j j
S 0 B1 +
j j
j=1 φ S1 . (2.12)
Remark 2.2.1. Note that equation (2.12) involves random variables, meaning
that the equalities hold in any possible state the world might attend at time
t = 1, that is, for all ω ∈ Ω. Formally, we may say that equalities in (2.12) hold
P-almost surely, that is, with probability 1.
The gains process G(x, φ) of a trading strategy (x, φ) is given by
G0 (x, φ) := 0, G1 (x, φ) := V1 (x, φ) − V0 (x, φ) (2.13)
or, equivalently,
∑n ( ∑n ) ∑n
0
G1 (x, φ) := φ ∆B1 + j=1
j
φ ∆S1j = x− j=1 φ j
S0j ∆B1 + j=1 φj ∆S1j
where we denote
∆B1 := B1 − B0 , ∆S1j := S1j − S0j . (2.14)
As suggested by its name, the random variable G1 (x, φ) represents the gains
(or losses) the agent obtains from his trading strategy (x, φ).
It is often convenient to study the stock price in relation to the money market
account. The discounted stock price Sbj is defined as follows
bj S0j
S0 := = S0j ,
B0
bj S1j 1
S1 := = S1j .
B1 1+r
Similarly, we define the discounted wealth process Vb (x, φ) of a trading

strategy (x, φ) by setting, for t = 0, 1,
Vbt (x, φ) := Vt (x,φ)

Bt
. (2.15)
It is easy to check that
Vb0 (x, φ) = x,
( ∑
n ) ∑
n ∑
n
b
V1 (x, φ) = x− j j
φ S0 + j bj
φ S1 = x + φj ∆Sb1j ,
j=1 j=1 j=1
where we denote
∆Sb1j := Sb1j − Sb0j .
b1 (x, φ) is defined by
Finally, the discounted gains process G
b0 (x, φ) := 0,
G b1 (x, φ) := Vb1 (x, φ) − Vb0 (x, φ) = ∑n φj ∆Sb1j
G (2.16)
j=1
where we also used the property of Vb1 (x, φ). It follows from (2.16) that G
b1 (x, φ)
does not depend on x, so that, in particular, the equalities
b1 (x, φ) = G
G b1 (0, φ) = Vb1 (0, φ)
hold for any x ∈ R and any φ ∈ Rn .

Example 2.2.1. We consider the single-period market model M = (B, S 1 , S 2 )
and we assume that the state space Ω = {ω1 , ω2 , ω3 }. Let the interest rate be
equal to r = 91 . Stock prices at time t = 0 are given by S01 = 5 and S02 = 10,
respectively. Random stock prices at time t = 1 are given by the following table
ω1 ω2 ω3
60 60 40
S11 9 9 9
40 80 80
S12 3 9 9
Let us consider a trading strategy (x, φ) with x ∈ R and φ = (φ1 , φ2 ) ∈ R2 . Then

(2.12) gives
( )
1
V1 (x, φ) = (x − 5φ − 10φ ) 1 +
1 2
+ φ1 S11 + φ2 S12 .
9
24 MATH3075/3975
More explicitly, the random variable V1 (x, φ) : Ω → R is given by

( )
1 60 40
V1 (x, φ)(ω1 ) = (x − 5φ − 10φ ) 1 +
1 2
+ φ1 + φ2 ,
9 9 3
( )
1 60 80
V1 (x, φ)(ω2 ) = (x − 5φ1 − 10φ2 ) 1 + + φ1 + φ2 ,
9 9 9
( )
1 40 80
V1 (x, φ)(ω3 ) = (x − 5φ1 − 10φ2 ) 1 + + φ1 + φ2 .
9 9 9
The increments ∆S1j for j = 1, 2 are given by the following table

ω1 ω2 ω3
∆S11 5
3
5
3
− 59
∆S12 10
3
− 10
9
− 10
9
and the gains process G(x, φ) satisfies: G0 (x, φ) = 0 and

1 15 1 30 2
G1 (x, φ)(ω1 ) = (x − 5φ1 − 10φ2 ) + φ + φ,
9 9 9
1 15 1 10 2
G1 (x, φ)(ω2 ) = (x − 5φ1 − 10φ2 ) + φ − φ,
9 9 9
1 5 1 10 2
G1 (x, φ)(ω3 ) = (x − 5φ1 − 10φ2 ) − φ − φ.
9 9 9
Let us now consider the discounted processes. The discounted stock prices at
time t = 1 are:
ω1 ω2 ω3
Sb11 6 6 4
Sb12 12 8 8
and the discounted wealth process at time t = 1 equals
Vb1 (x, φ)(ω1 ) = (x − 5φ1 − 10φ2 ) + 6φ1 + 2φ2 ,

Vb1 (x, φ)(ω2 ) = (x − 5φ1 − 10φ2 ) + 6φ1 + 8φ2 ,
Vb1 (x, φ)(ω3 ) = (x − 5φ1 − 10φ2 ) + 4φ1 + 8φ2 .
The increments of the discounted stock prices ∆Sb1j are given by

ω1 ω2 ω3
∆Sb11 1 1 −1
∆Sb21 2 −2 −2
b1 (x, φ) = G
The discounted gains process G b1 (0, φ) = Vb1 (0, φ) satisfies: G
b0 (x, φ) =
0 and
b1 (x, φ)(ω1 ) = φ1 + 2φ2 ,
G
b1 (x, φ)(ω2 ) = φ1 − 2φ2 ,
G
b1 (x, φ)(ω3 ) = −φ1 − 2φ2 .
G
2.2.1 Fundamental Theorem of Asset Pricing

Let us return to the study of a general single-period model. Given the definition
of the wealth process in a general single-period market model, the definition of
an arbitrage in this model is very similar to Definition 2.1.3.
Definition 2.2.3. A trading strategy (x, φ), where x denotes the total initial
endowment and φ = (φ1 , . . . , φn ) with φj denoting the number of shares of stock
S j , is called an arbitrage opportunity (or simply an arbitrage) whenever
1. x = V0 (x, φ) = 0,
2. V1 (x, φ) ≥ 0,
3. there exists ωi ∈ Ω such that V1 (x, φ)(ωi ) > 0.
Recall that the wealth V1 (x, φ) at time t = 1 is given by equation (2.12). The
following remark is useful:
Remark 2.2.2. Given that a trading strategy (x, φ) satisfies condition 2. in
Definition 2.2.3, condition 3. in this definition is equivalent to the following
condition: ∑
3.′ EP (V1 (x, φ)) = ki=1 V1 (x, φ)(ωi )P(ωi ) > 0.
The definition of an arbitrage can also be formulated in terms of the discounted
wealth process or the discounted gains process. This is sometimes very useful,
when one has to check whether a model admits an arbitrage or not. The fol-
lowing proposition gives us such a statement. The proof of Proposition 2.2.1 is
left as an exercise.
Proposition 2.2.1. A trading strategy (x, φ) in a single-period market model
M is an arbitrage opportunity if and only if one of the following equivalent
conditions hold:
1. Conditions 1.–3. in Definition 2.2.3 are satisfied with Vbt (x, φ) instead of
Vt (x, φ) for t = 0, 1.
2. x = V0 (x, φ) = 0 and conditions 2.–3. in Definition 2.2.3 are satisfied with
b1 (x, φ) instead of V1 (x, φ).
G
Furthermore, condition 3. can be replaced by condition 3.′
We will now return to the analysis of risk-neutral probability measures and
their use in arbitrage pricing. Recall that this was already indicated in Section
2.1 (see equation (2.8)).
Definition 2.2.4. A probability measure Q on Ω is called a risk-neutral prob-
ability measure for a single-period market model M = (B, S 1 , . . . , S n ) if
1. Q(ωi ) > 0 for every i = 1, . . . , k,
2. EQ (∆Sb1j ) = 0 for every j = 1, . . . , n.
We denote by M the set of all risk-neutral probability measures for the model M.
26 MATH3075/3975
Condition 2. of Definition 2.2.4 can be represented as follows: for every j =

1, . . . , n ( )
EQ S1j = (1 + r)S0j .
We thus see that when Ω consists of two elements only and there is a single
traded stock in the model, we obtain exactly what was called a risk-neutral
probability measure in Section 2.1.
We say that a model is arbitrage-free is no arbitrage opportunities exist.

Risk-neutral probability measures are closely related to the question whether
a model is arbitrage-free. The following result, which clarifies this connection,
is one of the main pillars of Financial Mathematics. It was first established by
Harrison and Pliska (1981) and later extended to continuous-time models.
Theorem 2.2.1. Fundamental Theorem of Asset Pricing. A single-period

market model M = (B, S 1 , . . . , S n ) is arbitrage-free if and only if there exists a
risk-neutral probability measure for the model.
In other words, the FTAP states that the following equivalence is valid:
A single-period model M is arbitrage-free ⇔ M ̸= ∅.
2.2.2 Proof of the FTAP (MATH3975)
Proof of the implication ⇐ in Theorem 2.2.1.
Proof. (⇐) We assume that M ̸= ∅, so that there exists a risk-neutral proba-

bility measure, denoted by Q. Let (0, φ) be an arbitrary trading strategy with
x = 0 and let Vb1 (0, φ) be the associated discounted wealth at time t = 1. Then
( ) (∑n ) ∑ n
( )
EQ Vb1 (0, φ) = EQ φj ∆Sb1j = φj EQ ∆Sb1j = 0.
j=1 j=1
| {z }
=0
If we assume that Vb1 (0, φ) ≥ 0, then the last equation clearly implies that the
equality Vb1 (0, φ)(ω) = 0 must hold for all ω ∈ Ω. Hence, by part 1. in Proposition
2.2.1, no trading strategy satisfying all conditions of an arbitrage opportunity
may exist.
The proof of the second implication in Theorem 2.2.1 is essentially geometric

and thus some preparation is needed. To start with, it will be very handy
to interpret random variables on Ω as vectors in the k-dimensional Euclidean
space Rk . This can be achieved through the following formal identification
X = (X(ω1 ), X(ω2 ), . . . , X(ωk )) ∈ Rk .

This one-to-one correspondence means that every random variable X on Ω can

be interpreted as a vector X = (x1 , . . . , xk ) in Rk and, conversely, any vector
X ∈ Rk uniquely specifies a random variable X on Ω. We can therefore identify
the set of random variables on Ω with the vector space Rk .
Similarly, any probability measure Q on Ω can be interpreted as a vector in Rk .
The latter identification is simply given by
Q = (Q(ω1 ), Q(ω2 ), . . . , Q(ωk )) ∈ Rk .
It is clear that there is a one-to-one correspondence between the set of all prob-
ability measures on Ω and the set P ⊂ Rk of vectors Q = (q1 , . . . , qk ) with the
following two properties:
1. qi ≥ 0 for every i = 1, . . . , k,
∑k
2. i=1 qi = 1.
Let X be the vector representing the random variable X and Q be the vector
representing the probability measure Q. Then the expected value of a ran-
dom variable X with respect to a probability measure Q on Ω can be identified
with the inner product ⟨·, ·⟩ in the Euclidean space Rk , specifically,
∑k ∑k
EQ (X) = i=1 X(ωi )Q(ωi ) = i=1 xi qi = ⟨X, Q⟩.
Let us define the following subset of Rk

{ }
b1 (x, φ) for some (x, φ) ∈ Rn+1 .
W = X ∈ Rk | X = G
Recall that G b1 (x, φ) = G

b1 (0, φ) = Vb1 (0, φ) for any x ∈ R and φ ∈ Rn . Hence W is
the set of all possible discounted values at time t = 1 of trading strategies that
start with an initial endowment x = 0, that is,
{ }
W = X ∈ Rk | X = Vb1 (0, φ) for some φ ∈ Rn . (2.17)
From equality (2.16), we deduce that W is a vector subspace of Rk generated by

the vectors ∆Sb11 , . . . , ∆Sb1n . Of course, the dimension of the subspace W cannot
be greater than k. We observe that in any arbitrage-free model the dimension
of W is less or equal to k −1 (this remark is an immediate consequence of equiv-
alence (2.19)).
Next, we define the following set

{ }
A = X ∈ Rk | X ̸= 0, xi ≥ 0, i = 1, . . . , k . (2.18)
28 MATH3075/3975
The set A is simply the closed nonnegative orthant in Rk (the first quadrant
when k = 2, the first octant when k = 3, etc.) but with the origin excluded. In-
deed, the conditions in (2.18) imply that at least one component of any vector X
from A is a strictly positive number and all other components are non-negative.
Remark 2.2.3. In view of Proposition 2.2.1, we obtain the following useful

equivalence
M = (B, S 1 , . . . , S n ) is arbitrage-free ⇔ W ∩ A = ∅. (2.19)
To establish (2.19), it suffices to note that any vector belonging to W ∩ A can be

interpreted as the discounted value at time t = 1 of an arbitrage opportunity.
The FTAP can now be restated as follows: W ∩ A = ∅ ⇔ M ̸= ∅.
We will also need the orthogonal complement W⊥ of W, which is given by
{ }
W⊥ = Z ∈ Rk | ⟨X, Z⟩ = 0 for all X ∈ W . (2.20)
Finally, we define the following subset P + of the set P of all probability mea-
sures on Ω
{ ∑ }
P + = Q ∈ Rk | ki=1 qi = 1, qi > 0 . (2.21)
The set P + can be identified with the set of all probability measures on Ω that
satisfy property 1. from Definition 2.2.4.
Lemma 2.2.1. A probability measure Q is a risk-neutral probability measure

for a single-period market model M = (B, S 1 , . . . , S n ) if and only if Q ∈ W⊥ ∩P + .
Hence the set M of all risk-neutral probability measures satisfies M = W⊥ ∩ P + .
Proof. (⇒) Let us first assume that Q is a risk-neutral probability measure.

Then, by property 1. in Definition 2.2.4, it is obvious that Q belongs to P + . Fur-
thermore, using property 2. in Definition 2.2.4 and equality (2.16), we obtain
for an arbitrary vector X = Vb1 (0, φ) ∈ W
( ) (∑n ) ∑ n
( )
⟨X, Q⟩ = EQ Vb1 (0, φ) = EQ φj ∆Sb1j = φj EQ ∆Sb1j = 0.
j=1 j=1
| {z }
=0
This means that Q ∈ W⊥ and thus we conclude that Q ∈ W⊥ ∩ P + .

(⇐) Assume now that Q is an arbitrary vector in W⊥ ∩ P + . We first note that
Q defines a probability measure satisfying condition 1. in Definition 2.2.4. It
remains to show that it also satisfies condition 2. in Definition 2.2.4. To this
end, for a fixed (but arbitrary) j = 1, . . . , n, we consider the trading strategy
(0, φ) with φ = (0, . . . , 0, 1, 0, . . . , 0) = ej . The discounted wealth of this strategy

clearly satisfies Vb1 (0, φ) = ∆Sb1j . Since Vb1 (0, φ) ∈ W and Q ∈ W⊥ , we obtain
( )
0 = ⟨Vb1 (0, φ), Q⟩ = ⟨∆Sb1j , Q⟩ = EQ ∆Sb1j .
Since j is arbitrary, we see that Q satisfies condition 2. in Definition 2.2.4.
Remark 2.2.4. Using Lemma 2.2.1, we obtain a purely geometric reformula-
tion of the FTAP: W ∩ A = ∅ ⇔ W⊥ ∩ P + ̸= ∅.
Separating hyperplane theorem.
In the proof of the implication ⇒ in Theorem 2.2.1, we will employ an auxil-

iary result from the convex analysis, known as the separating hyperplane
theorem. We state below the most convenient for us version of this theorem.
Let us first recall the definition of convexity.
Definition 2.2.5. A subset D of Rk is said to be convex if for all d1 , d2 ∈ D and
every α ∈ [0, 1], we have that αd1 + (1 − α)d2 ∈ D.
It is important to notice that the set M is convex, that is, if Q1 and Q2 belong to
M then the probability measure αQ1 +(1−α)Q2 belongs to M for every α ∈ [0, 1].
A verification of this property is left as an exercise.
Proposition 2.2.2. Let B, C ⊂ Rk be nonempty, closed, convex sets such that
B ∩ C = ∅. Assume, in addition, that at least one of these sets is compact (i.e.,
bounded and closed). Then there exist vectors a, y ∈ Rk such that
⟨b − a, y⟩ < 0 for all b∈B
and
⟨c − a, y⟩ > 0 for all c ∈ C.
Proof. The proof of this classic result can be found in any textbook on the
convex analysis. Hence we will merely sketch the main steps.
Step 1. In the first step, one shows that if D is a (nonempty) closed, convex set
such that the origin 0 is not in D then there exists a non-zero vector v ∈ D such
that for every d ∈ D we have ⟨d, v⟩ ≥ ⟨v, v⟩ (hence ⟨d, v⟩ > 0 for all d ∈ D). To
this end, one checks that the vector v that realises the minimum in the problem
min d∈D ∥d∥ has the desired properties (note that since D is closed and 0 ∈/ D we
have that v ̸= 0).
Step 2. In the second step, we define the set D as the algebraic difference of B
and C, that is, D = C − B. More explicitly,
{ }
D = x ∈ Rk | x = c − b for some b ∈ B, c ∈ C .
It is clear that 0 ∈
/ D. One can also check that D is convex (this always holds
if B and C and convex) and closed (for closedness, we need to postulate that at
least one of the sets B and C is compact).
30 MATH3075/3975
From the first step in the proof, there exists a non-zero vector y ∈ Rk such that
for all b ∈ B and c ∈ C we have that ⟨c − b, y⟩ ≥ ⟨y, y⟩. This in turn implies that
for all b ∈ B and c ∈ C
⟨c, y⟩ ≥ ⟨b, y⟩ + α
where α = ⟨y, y⟩ is a strictly positive number. Hence there exists a vector a ∈ Rk
such that
inf ⟨c, y⟩ > ⟨a, y⟩ > sup⟨b, y⟩.
c∈C b∈B
It is now easy to check that the desired inequalities are satisfied for this choice
of vectors a and y.
Let a, y ∈ Rk be as in Proposition 2.2.2. Observe that y is never a zero vector.
We define the (k − 1)-dimensional hyperplane H ⊂ Rk by setting
{ }
H = a + x ∈ Rk | ⟨x, y⟩ = 0 = a + {y}⊥ .
Then we say that the hyperplane H strictly separates the convex sets B and
C. Intuitively, the sets B and C lie on different sides of the hyperplane H
and thus they can be seen as geometrically separated by H. Note that the
compactness of at least one of the sets is a necessary condition for the strict
separation of B and C.
Corollary 2.2.1. Assume that B ⊂ Rk is a vector subspace and set C is a com-
pact convex set such that B ∩ C = ∅. Then there exists a vector y ∈ Rk such
that
⟨b, y⟩ = 0 for all b ∈ B
and
⟨c, y⟩ > 0 for all c ∈ C.
Proof. Note that any vector subspace of Rk is a closed, convex set. From
Proposition 2.2.2, there exist vectors a, y ∈ Rk such that the inequality
⟨b, y⟩ < ⟨a, y⟩
is satisfied for all vectors b ∈ B. Since B is a vector space, the vector λb belongs
to B for any λ ∈ R. Hence for any b ∈ B and λ ∈ R we have that
⟨λb, y⟩ = λ⟨b, y⟩ < ⟨a, y⟩.
This in turn implies that ⟨b, y⟩ = 0 for any vector b ∈ B, meaning that y ∈ B ⊥ .
To establish the second inequality, we observe that from Proposition 2.2.2, we
obtain
⟨c, y⟩ > ⟨a, y⟩ for all c ∈ C.
Consequently, for any c ∈ C
⟨c, y⟩ > ⟨a, y⟩ > ⟨b, y⟩ = 0.
We conclude that ⟨c, y⟩ > 0 for all c ∈ C.
Corollary 2.2.1 will be used in the proof of the implication ⇒ in Theorem 2.2.1.
Proof of the implication ⇒ in Theorem 2.2.1.
Proof. (⇒) We now assume that the model is arbitrage-free. We know that
this is equivalent to the condition W ∩ A = ∅. Our goal is to show that the class
M of risk-neutral probabilities is non-empty. In view of Lemma 2.2.1 (see also
Remark 2.2.4), it suffices to show that the following implication is valid
W∩A=∅ ⇒ W⊥ ∩ P + ̸= ∅.
We define the following auxiliary set A+ = {X ∈ A | ⟨X, P⟩ = 1}. Observe that
A+ is a closed, bounded (hence compact) and convex subset of Rk . Recall also
that P is the subjective probability measure (although any other probability
measure from P + could have been used to define A+ ). Since A+ ⊂ A, it is clear
that
W ∩ A = ∅ ⇒ W ∩ A+ = ∅.
By applying Corollary 2.2.1 to the sets B = W and C = A+ , we see that there
exists a vector Y ∈ W⊥ such that
⟨X, Y ⟩ > 0 for all X ∈ A+ . (2.22)
Our goal is to show that Y can be used to define a risk-neutral probability Q
after a suitable normalisation. We need first to show that yi > 0 for every i. For
this purpose, for any fixed i = 1, . . . , k, we define the auxiliary vector Xi as the
vector in Rk whose ith component equals 1/P(ωi ) and all other components are
zero, that is,
1 1
Xi = (0, . . . , 0, 1, 0 . . . , 0) = ei .
P(ωi ) P(ωi )
Then clearly
1
EP (Xi ) = ⟨Xi , P⟩ = P(ωi ) = 1
P(ωi )
and thus Xi ∈ A+ . Let us denote by yi the ith component of Y . It then follows
from (2.22) that
1
0 < ⟨Xi , Y ⟩ = yi ,
P(ωi )
which means that the inequality yi > 0 holds for all i = 1, . . . , k. We will now
define a vector Q = (q1 , . . . , qk ) ∈ Rk through the normalisation of the vector Y .
To this end, we define
yi
qi = = cyi
y1 + · · · + yk
and we set Q(ωi ) = qi for i = 1, . . . , k. In this way, we defined a probability
measure Q such that Q ∈ P + . Furthermore, since Q is merely a scalar multiple
of Y (i.e. Q = cY for some scalar c) and W⊥ is a vector space, we have that
Q ∈ W⊥ (recall that Y ∈ W⊥ ). We conclude that Q ∈ W⊥ ∩P + so that W⊥ ∩ P + ̸=
∅. By virtue of by Lemma 2.2.1, the probability measure Q is a risk-neutral
probability measure on Ω, so that M ̸= ∅.
32 MATH3075/3975
Example 2.2.2. We continue the study of the market model M = (B, S 1 , S 2 )

introduced in Example 2.2.1. Our aim is to illustrate the fact that the exis-
tence of a risk-neutral probability is a necessary condition for the no-arbitrage
property of a market model, that is, the ‘only if ’ implication in Theorem 2.2.1.
Recall that the increments of the discounted prices in this example were given
by the following table:
ω1 ω2 ω3
∆Sb11 1 1 −1
∆Sb2 1 2 −2 −2
From the definition of the set W (see (2.17)) and the equality
Vb1 (0, φ) = φ1 ∆Sb11 + φ2 ∆Sb12 ,
it follows that   
 φ1 + 2φ2 

W =  φ1 − 2φ2  (φ1 , φ2 ) ∈ R2 .
 
−φ1 − 2φ2
We note that for any vector X ∈ W we have x1 + x3 = 0, where xi is the ith
component of the vector X.
Conversely, if a vector X ∈ R3 is such that x1 + x3 = 0 than we may choose

φ1 = 21 (x1 + x2 ) and φ2 = 14 (x1 − x2 ) and obtain
   1 
x1 φ + 2φ2
X =  x2  =  φ1 − 2φ2  .
x3 −φ1 − 2φ2
We conclude that W is the plane in R3 given by
W = {X ∈ R3 | x1 + x3 = 0} = {X ∈ R3 | X = (γ, x2 , −γ)⊤ for some γ ∈ R}.
Hence the orthogonal complement of W is the line given by
W⊥ = {Y ∈ R3 | Y = (λ, 0, λ)⊤ for some λ ∈ R}.
It is now easily seen that W⊥ ∩ P + = ∅, so that there is no risk-neutral prob-
ability measure in this model, that is, M = ∅. One can also check directly
that the sub-models (B, S 1 ) and (B, S 2 ) are arbitrage-free, so that the corre-
sponding classes of risk-neutral probabilities M1 and M2 are non-empty, but
M = M1 ∩ M2 = ∅.
In view of Theorem 2.2.1, we already know at this point that there must be
an arbitrage opportunity in the model. To find explicitly an arbitrage opportu-
nity, we use (2.19). If we compare W and A, we see that
W ∩ A = {X ∈ R3 | x1 = x3 = 0, x2 > 0}
so that the set W ∩ A is manifestly non-empty. We deduce once again, but this
time using equivalence (2.19), that the considered model is not arbitrage-free.
We will now describe all arbitrage opportunities in the model. We start with
any positive number x2 > 0. Since
 
0
 x2  ∈ W ∩ A,
0
we know that there must exist a trading strategy (0, φ) such that
 
0
Vb1 (0, φ) =  x2  .
0
To identify φ = (φ1 , φ2 ) ∈ R2 , we solve the following system of linear equations:
Vb1 (0, φ)(ω1 ) = φ1 + 2φ2 = 0,
Vb1 (0, φ)(ω2 ) = φ1 − 2φ2 = x2 ,
Vb1 (0, φ)(ω3 ) = −φ1 − 2φ2 = 0,
where the last equation is manifestly redundant. The unique solution reads
x2 x2
φ 1 = , φ2 = − .
2 4
As we already know, these numbers give us the number of shares of each stock
we need to buy in order to obtain an arbitrage. It thus remains to compute
how much money we have to invest in the money market account. Since the
strategy (0, φ) starts with zero initial endowment, the amount φ0 invested in
the money market account satisfies
x2 ( x )
2
φ0 = 0 − φ1 S01 − φ2 S02 = − 5 − − 10 = 0.
2 4
This means that any arbitrage opportunity in this model is a trading strategy
that only invests in risky assets, that is, stocks S 1 and S 2 . One can observe that
the return on the first stock dominates the return on the second.
2.2.3 Arbitrage Pricing of Contingent Claims

In Sections 2.2.3 and 2.2.4, we work under the standing assumption that a gen-
eral single-period model M = (B, S 1 , . . . , S n ) is arbitrage-free or, equivalently,
that the class M of risk-neutral probability measures is non-empty. We will
address the following question:
How to price contingent claims in a multi-period model?
In Section 2.1, we studied claims of the type h(S1 ) where h is the payoff’s profile,
which is an arbitrary function of the single stock price S1 at time t = 1. In our
general model, we now have more than one stock and the payoff profiles may be
complicated functions of underlying assets. Specifically, any contingent claim
can now be described as h(S11 , . . . , S1n ) where h : Rn → R is an arbitrary function.
34 MATH3075/3975
It is thus natural to introduce the following definition of a contingent claim.

Definition 2.2.6. A contingent claim in a single-period market model M is
a random variable X on Ω representing a payoff at time t = 1.
Let us state the arbitrage pricing principle for a general contingent claim.
Definition 2.2.7. We say that a price x for the contingent claim X complies
with the principle of no-arbitrage provided that the extended market model
f = (B, S 1 , . . . , S n , S n+1 ) consisting of the savings account B, the original stocks
M
S 1 , . . . , S n , and an additional asset S n+1 , with the price process satisfying S0n+1 =
x and S1n+1 = X, is arbitrage-free.
The additional asset S n+1 introduced in Definition 2.2.7 may not be interpreted
as a stock, in general, since it can take negative values if the contingent claim
takes negative values. For the general arbitrage pricing theory developed so
far, positiveness of stock prices was not essential, however. It was only essen-
tial to assume that the price process of the money market account is strictly
positive.
Step A. Pricing of an attainable claim.
To price a contingent claim for which a replicating strategy exists, we apply the
replication principle.
Proposition 2.2.3. Let X be a contingent claim in a general single-period mar-
ket model and let (x, φ) be a replicating strategy for X, i.e. a trading strategy
which satisfies V1 (x, φ) = X. Then the unique price of X which complies with
the no arbitrage principle is x = V0 (x, φ). It is called the arbitrage price of X
and denoted as π0 (X).
Proof. The proof of this proposition hinges on the same arguments as those
used in Section 2.1 and thus it is left as an exercise.
Definition 2.2.8. A contingent claim X is called attainable if there exists a

trading strategy (x, φ) which replicates X, that is, satisfies V1 (x, φ) = X.
For attainable contingent claims the replication principle applies and it is clear
how to price them, namely, the arbitrage price π0 (X) is necessarily equal to the
initial endowment x needed for a replicating strategy. There might be more
than one replicating strategy, in general. However, it can be easily deduced
from the no arbitrage principle that the initial endowment x for all strategies
replicating a given contingent claim is unique.
In equation (2.8), we established a way to use a risk-neutral probability mea-

sure to compute the price of an option in the two-state single-period market
model. The next result shows that this probabilistic approach works fine in
the general single-period market model as well, at least when we restrict our
attention to attainable contingent claims, for which the price is defined as the
initial endowment of a replicating strategy.
Proposition 2.2.4. Let X be an attainable contingent claim and Q be an arbi-

trary risk-neutral probability measure, that is, Q ∈ M. Then the arbitrage price
π0 (X) of X at time t = 0 satisfies
π0 (X) = EQ ((1 + r)−1 X) . (2.23)
Proof. Let (x, φ) be any replicating strategy for an attainable claim X, so that
the equality X = V1 (x, φ) is valid. Then we also have that
(1 + r)−1 X = Vb1 (x, φ).
Using Definition 2.2.4 of a risk-neutral probability measure, we obtain
( ) ( ∑ )
( ) n
−1 b
EQ (1 + r) X = EQ V1 (x, φ) = EQ x + j bj
φ ∆ S1
j=1
∑
n
( )
= x+ φj EQ ∆Sb1j = x
j=1
| {z }
=0
and thus formula (2.23) holds. Note that (2.23) is valid for any choice of a
risk-neutral probability measure Q ∈ M.
Step B. Example of an incomplete model.
A crucial difference between the two-state single-period model and a general

single-period market model is that in the latter model a replicating strategy
might not exist for some contingent claims. This may happen when there are
more sources of randomness than there are stocks to invest in. We will now
examine an explicit example of an arbitrage-free single-period model in which
some contingent claims are not attainable, that is, an incomplete model.
Example 2.2.3. We consider the market model consisting of two traded assets,
the money market account B and the stock S so that M = (B, S). We also
introduce an auxiliary quantity, which we call the volatility and denote by v.
The volatility determines whether the stock price can make big jumps or small
jumps. In this model, the volatility is assumed to be random or, in other words,
stochastic. Such models are called stochastic volatility models. To be more
specific, we postulate that the state space consists of four states
Ω := {ω1 , ω2 , ω3 , ω4 }
and the volatility v is the random variable on Ω given by
{
h if ω = ω1 , ω4 ,
v(ω) :=
l if ω = ω2 , ω3 .
We assume here 0 < l < h < 1, so that l stands for a lower level of the volatility
whereas h represents its higher level.
36 MATH3075/3975
The stock price S1 is then modeled by the following formula

{
(1 + v(ω))S0 if ω = ω1 , ω2 ,
S1 (ω) :=
(1 − v(ω))S0 if ω = ω3 , ω4 ,
where, as usual, a positive number S0 represents the initial stock price. The
stock price can therefore move up or down, as in the two-state single-period
market model from Section 2.1. In contrast to the two-state single-period
model, the amount by which it jumps is itself random and it is determined
by the realised level of the volatility. As usual, the money market account is
given by: B0 = 1 and B1 = 1 + r.
Let us consider a digital call in this model, i.e., an option with the payoff
{
1 if S1 > K,
X=
0 otherwise.
Let us assume that the strike price K satisfies
(1 + l)S0 < K < (1 + h)S0 .
Then a nonzero payoff is only possible if the volatility is high and the stock
jumps up, that is, when the state of the world at time t = 1 is given by ω = ω1 .
Therefore, the contingent claim X can alternatively be represented as follows
{
1 if ω = ω1 ,
X(ω) =
0 if ω = ω2 , ω3 , ω4 .
Our goal is to check whether there exists a replicating strategy for this contin-
gent claim, i.e., a trading strategy (x, φ) ∈ R × R satisfying
V1 (x, φ) = X.
Using the definition of V1 (x, φ) and our vector notation for random variables,
the last equation is equivalent to
     
1+r (1 + h)S0 1
 1+r   (1 + l)S0   0 
(x − φS0 )      
 1 + r  + φ  (1 − l)S0  =  0  .
1+r (1 − h)S0 0
The existence of a solution (x, φ) to this system is equivalent to the existence
of a solution (α, β) to the system
     
1 h 1
 1   l   0 
α 
 1 +β
  
 −l  =  0
.

1 −h 0
It is easy to see that this system of equations has no solution and thus a digital
call is not an attainable contingent claim within the framework of the stochas-
tic volatility model.
The heuristic explanation of this example is that there is a source of random-

ness in the volatility, which is not hedgeable, since the volatility is not a directly
traded asset. To summarise Example 2.2.3, the stochastic volatility model in-
troduced in this example is incomplete, as for some contingent claims a repli-
cating strategy does not exist.
Step C. Non-uniqueness of a risk-neutral value.
Let us now address the issue of non-attainability of a contingent claim from

the perspective of the risk-neutral valuation formula.
• From Propositions 2.2.3 and 2.2.4 we deduce that if a claim X is attain-
able, then we obtain the same number for the expected value in right-hand
side of equation (2.23) for any risk-neutral probability measure Q for the
model M.
• The following example shows that the situation changes dramatically if
we consider a contingent claim that is not attainable, specifically, a risk-
neutral expected value in equation (2.23) is no longer unique.
Example 2.2.4. We start by computing the class of all risk-neutral probability
measures for the stochastic volatility model M introduced in Example 2.2.3. To
simplify computations, we will now assume that r = 0, so that the discounted
processes coincide with the original processes. We have
{
b v(ω)S0 if ω = ω1 , ω2 ,
∆S1 (ω) :=
−v(ω)S0 if ω = ω3 , ω4 ,
or, using the vector notation,
 
h
 l 
∆Sb1 = S0  
 −l  .
−h
Recall that for any trading strategy (x, φ), the discounted gains satisfies
b1 (x, φ) = G
G b1 (0, φ) = Vb1 (0, φ) = φ∆Sb1 .
Hence the vector space W is a one-dimensional vector subspace of R4 spanned

by the vector ∆Sb1 , that is,
     

 h 
   h 

    l  
 l   
W = span  = λ , λ ∈ R .

 −l 
 

 −l  

 −l   −h 
38 MATH3075/3975
The orthogonal complement of W is thus the three-dimensional subspace of R4

given by       
 z z h 

 1 1 

z   z   l 
⊥
W =   2 
∈ R4  2  
⟨ , ⟩ = 0 .
 z3   z3   −l  

 z 

4 z 4−h
Recall also that a vector (q1 , q2 , q3 , q4 )⊤ belongs to P + if and only if the equality
q1 + q2 + q3 + q4 = 1 holds and qi > 0 for i = 1, 2, 3, 4. Since the set of risk-neutral
probability measures is given by M = W⊥ ∩ P + , we find that
 
q1
 q2 
 
 q3  ∈ M ⇔ q1 > 0, q2 > 0, q3 > 0, q4 > 0, q1 + q2 + q3 + q4 = 1
q4
and h(q1 − q4 ) + l(q2 − q3 ) = 0.
The class of all risk-neutral probability measures in our stochastic volatility

model is therefore given by
  
 q 

 1 

q  q > 0, q > 0, q > 0, q + q + q < 1
M=  
2 

1 2 3 1 2 3
.

 q3 l(q2 − q3 ) = h(1 − (2q1 + q2 + q3 ))  
 1 − (q + q + q ) 
1 2 3
Clearly, this set is non-empty and thus we conclude that our stochastic volatil-
ity model is arbitrage-free.
In addition, it is not difficult to check that for every 0 < q1 < 21 there exists
a probability measure Q ∈ M such that Q(ω1 ) = q1 . Indeed, it suffices to take
q1 ∈ (0, 12 ) and to set
1
q4 = q1 , q 2 = q3 = − q1 .
2
Let us now compute the (discounted) expected value of the digital call X =
(1, 0, 0, 0)⊤ under a probability measure Q = (q1 , q2 , q3 , q4 )⊤ ∈ M. We have
EQ (X) = ⟨X, Q⟩ = q1 · 1 + q2 · 0 + q3 · 0 + q4 · 0 = q1 .
Since q1 is here any number from the open interval (0, 12 ), we obtain many differ-
ent values as discounted expected values under risk-neutral probability mea-
sures. In fact, every value from the open interval (0, 12 ) can be achieved.
Note that the situation in Example 2.2.4 is completely different than in Propo-
sitions 2.2.3 and 2.2.4. The reason is that, as we already have shown in Exam-
ple 2.2.3, the digital call is not an attainable contingent claim in the stochastic
volatility model and thus it is not covered by Propositions 2.2.3 and 2.2.4.
Step D. Arbitrage pricing of an arbitrary claim.
In view of Example 2.2.4, the next result might be surprising, since it says
that for any choice of a risk-neutral probability measure Q, formula (2.23),
yields a price which complies with the principle of no-arbitrage. Let us stress
once again that we obtain different prices, in general, when we use different
risk-neutral probability measures. Hence the number π0 (X) appearing in the
right-hand side of (2.24) represents a plausible price (rather than the unique
price) for a non-attainable claim X. We will show in what follows that:
• an attainable claim is characterised by the uniqueness of the arbitrage
price, but
• a non-attainable claim always admits a whole spectrum of prices that com-
ply with the principle of no-arbitrage.
Proposition 2.2.5. Let X be a possibly non-attainable contingent claim and
Q ∈ M be any risk-neutral probability measure for an arbitrage-free single-
period market model M. Then the real number π0 (X) given by
( )
π0 (X) := EQ (1 + r)−1 X (2.24)
defines a price for the contingent claim X at time t = 0, which complies with the
principle of no-arbitrage. Moreover, the probability measure Q belongs to the
e of risk-neutral probability measures for the extended market model M.
class M f
Proof. In view of the FTAP (see Theorem 2.2.1), it is enough to show that
there exists a risk-neutral probability measure for the corresponding model M, f
which is extended by S n+1
as in Definition 2.2.7. By assumption, Q is a risk-
neutral probability measure for the original model M, consisting of consisting
of the savings account B and stocks S 1 , . . . , S n . In other words, the probability
measure Q is assumed to satisfy conditions 1. and 2. of Definition 2.2.4 for
every j = 1, . . . , n. For j = n + 1, the second condition translates into
( ) ( )
EQ (∆Sb1n+1 ) = EQ (1 + r)−1 X − π0 (X) = EQ (1 + r)−1 X − π0 (X)
= π0 (X) − π0 (X) = 0.
Hence, by Definition 2.2.4, the probability measure Q is a risk-neutral prob-

ability measure for the extended market model, that is, Q ∈ M. e In view of
Theorem 2.2.1, the extended market model M f is arbitrage-free, so that the
price π0 (X) complies with the principle of no arbitrage.
By applying Proposition 2.2.5 to the digital call within the setup of Example
2.2.4, we conclude that any price belonging to (0, 12 ) does not allow arbitrage
and can therefore be considered as ‘fair’ (or ‘viable’). The non-uniqueness of
prices for non-attainable claims is a challenging problem, which was not yet
completely resolved.
40 MATH3075/3975
2.2.4 Completeness of a General Single-Period Model

Let us first characterise the models in which the problem of non-uniqueness of
prices does not occur.
Definition 2.2.9. A financial market model is called complete if for any con-
tingent claim X there exists a replicating strategy (x, φ) ∈ Rn+1 . A model is
incomplete when there exists a claim X for which a replicating strategy does
not exist.
By Proposition 2.2.4, the issue of computing prices of contingent claims in a
complete market model is completely solved. But how can we tell whether an
arbitrage-free model is complete or not?
Step A: Algebraic criterion for market completeness.
The following result gives an algebraic criterion for the market completeness.
Note that the vectors A0 , A1 , . . . , An ∈ Rk represent the columns of the matrix A.
Proposition 2.2.6. Let us assume that a single-period market model M =
(B, S 1 , . . . , S n ) defined on the state space Ω = {ω1 , . . . , ωk } is arbitrage-free. Then
this model is complete if and only if the k × (n + 1) matrix A given by
 
1 + r S11 (ω1 ) · · · S1n (ω1 )
 1 + r S11 (ω2 ) · · · S1n (ω2 ) 
 
 · · · · 
A=  = (A0 , A1 , . . . , An )
 · · · · 
 · · · · 
1 + r S1 (ωk ) · · · S1 (ωk )
1 n
has a full row rank, that is, rank (A) = k. Equivalently, a single-period mar-
ket model M is complete whenever the linear subspace spanned by the vectors
A0 , A1 , . . . , An coincides with the full space Rk .
Proof. On the one hand, by the linear algebra, the matrix A has a full row rank
if and only if for every X ∈ Rk the equation AZ = X has a solution Z ∈ Rn+1 .
∑n
On the other hand, if we set φ0 =x− j=1 φj S0j then we have
   0   
1 + r S11 (ω1 ) · · · S1n (ω1 ) φ V1 (x, φ)(ω1 )
 1 + r S11 (ω2 ) · · · S1n (ω2 )   φ1   V1 (x, φ)(ω2 ) 
     
 · · · ·   ·   · 
   = .
 · · · ·   ·   · 
 · · · ·   ·   · 
1 + r S11 (ωk ) · · · S1n (ωk ) φn V1 (x, φ)(ωk )
This shows that computing a replicating strategy for a contingent claim X is
equivalent to solving the equation AZ = X. Hence the statement of the propo-
sition follows immediately.
Example 2.2.5. We have seen already that the stochastic volatility model dis-
cussed in Examples 2.2.3 and 2.2.4 is not complete. Another way to establish
this property is to use Proposition 2.2.6. The matrix A in this case has the form
 
1 + r (1 + h)S0
 1 + r (1 − h)S0 
A= 
 1 + r (1 + l)S0  .
1 + r (1 − h)S0
The rank of this matrix is 2 and thus, of course, it is not equal to k = 4. We
therefore conclude that the stochastic volatility model is incomplete.
Step B: Probabilistic criterion for attainability of a claim.
Proposition 2.2.6 offers a simple method of determining whether a given mar-

ket model is complete, without the need to make explicit computations of repli-
cating strategies. Now, the following question arises: in an incomplete market
model, how to check whether a given contingent claim is attainable, without
trying to compute the replicating strategy? The following results yields an an-
swer to this question.
Proposition 2.2.7. A contingent claim X is attainable in an arbitrage-free
single-period market model M = (B, S 1 , . . . , S n ) if and only if the expected value
( )
EQ (1 + r)−1 X
has the same value for all risk-neutral probability measures Q ∈ M.
Proof. (⇒) It is immediate from Proposition 2.2.4 that if a contingent claim X

is attainable then the expected value
( )
EQ (1 + r)−1 X
has the same value for all Q ∈ M.
(⇐) (MATH3975) We will prove this implication by contrapositive. Let us

thus assume that the contingent claim X is not attainable. Our goal is to find
b for which
two risk-neutral probabilities, say Q and Q,
( ) ( )
EQ (1 + r)−1 X ̸= EQb (1 + r)−1 X . (2.25)
Consider the (k × (n + 1))-matrix A introduced in Proposition 2.2.6. Since X is
not attainable, there is no solution Z ∈ Rn+1 to the system
AZ = X.
We define the following subsets of Rk :
{ }
B = image (A) = AZ | Z ∈ Rn+1 ⊂ Rk
and C = {X}. Then B is a subspace of Rk and, obviously, the set C is convex
and compact. Moreover, B ∩ C = ∅.
42 MATH3075/3975
From Corollary 2.2.1, there exists a vector Y = (y1 , . . . , yk ) ∈ Rk such that

⟨b, Y ⟩ = 0 for all b ∈ B,
⟨c, Y ⟩ > 0 for all c ∈ C.
In view of the definition of B and C, this means that for j = 0, . . . , n
⟨Aj , Y ⟩ = 0 and ⟨X, Y ⟩ > 0 (2.26)
where Aj is the jth column of the matrix A.
Let Q ∈ M be an arbitrary risk-neutral probability measure. We may choose

a real number λ > 0 to be small enough in order to ensure that for every
i = 1, . . . , k
b i ) := Q(ωi ) + λ(1 + r)yi > 0.
Q(ω (2.27)
We will check that Qb is a risk-neutral probability measure. From the definition
of A in Proposition 2.2.6 and the first equality in (2.26) with j = 0, we obtain
∑
k
λ(1 + r)yi = λ⟨A0 , Y ⟩ = 0.
i=1
It then follows from (2.27) that

∑
k ∑
k ∑
k
b i) =
Q(ω Q(ωi ) + λ(1 + r)yi = 1
i=1 i=1 i=1
and thus Q b is a probability measure on the space Ω. Moreover, in view of (2.27),

b satisfies condition 1. in Definition 2.2.4.
it is clear that Q
It remains to check that Qb satisfies also condition 2. in Definition 2.2.4. To

this end, we examine the behaviour under Q b of the discounted stock prices Sbj .
1
We have that, for every j = 1, . . . , n,
( ) ∑
k
EQb Sb1j = b i )Sbj (ωi )
Q(ω 1
i=1
∑k ∑
k
= Q(ωi )Sb1j (ωi ) + λ Sb1j (ωi )(1 + r)yi
i=1 i=1
( )
= EQ Sb1j + λ ⟨Aj , Y ⟩ (in view of (2.26))
| {z }
=0
= Sb0j (since Q ∈ M)
( j)
We conclude that EQb ∆Sb1 = 0 and thus Q b ∈ M, that is, Q
b is a risk-neutral
probability measure for the market model M.
b We have thus proven that if M is arbitrage-

From (2.27), it is clear that Q ̸= Q.
free and incomplete then there exists more than one risk-neutral probability
measure.
To complete the proof, it remains to show that inequality (2.25) is satisfied. We

observe that
( ) ∑ k ∑k ∑k
X b i ) X(ωi ) = X(ωi )
EQb = Q(ω Q(ωi ) +λ yi X(ωi )
1+r 1 + r 1 + r
i=1 i=1
| i=1 {z }
>0
∑
k ( )
X(ωi ) X
> Q(ωi ) = EQ
i=1
1+r 1+r
where we used the second part of (2.26) and the inequality λ > 0 in order to
conclude that the braced expression is a strictly positive real number.
Step C: Probabilistic criterion for market completeness.
The following important result complements the Fundamental Theorem of As-

set Pricing (see Theorem 2.2.1). It states that for an arbitrage-free model:
Completeness ⇔ Uniqueness of a risk-neutral probability measure.
Theorem 2.2.2. Assume that a single-period market model M = (B, S 1 , . . . , S n )

is arbitrage-free. Then M is complete if and only if the class M consists of a
single element, that is, there exists a unique risk-neutral probability measure.
Proof. Since M is assumed to be arbitrage-free, it follows from Theorem 2.2.1

that there exists at least one risk-neutral probability measure, that is, M ̸= ∅.
(⇐) Assume first that a risk-neutral probability measure for M is unique.
Then the condition in Proposition 2.2.7 is trivially satisfied for any claim X.
Hence any claim X is attainable and thus the market model is complete.
(⇒) Assume M is complete and consider any two risk-neutral probability mea-
sures Q and Q b from M. For a fixed, but arbitrary, i = 1, . . . , k, let the contingent
claim X i be given by
{
i 1 + r if ω = ωi ,
X (ω) =
0 otherwise.
Since M is now assumed to be complete, the contingent claim X i is attainable.

From Proposition 2.2.4, it thus follows that
( i ) ( i )
X X b i ).
Q(ωi ) = EQ = π0 (X ) = EQb
i
= Q(ω
1+r 1+r
Since i was arbitrary, we see that the equality Q = Q b holds. We have thus
shown that the class M consists of a single risk-neutral probability measure.
Chapter 3
Multi-Period Market Models
3.1 General Multi-Period Market Models

The two most important new features of a multi-period market model when
compared to a single-period market model are:
• Agents can buy and sell assets not only at the beginning of the trad-
ing period, but at any time t out of a discrete set of trading times t ∈
{0, 1, 2, . . . , T } where t = 0 is the beginning of the trading period and t = T
is the end.
• Agents can gather information over time, since they can observe prices.
For example, they can make their investment decisions at time t depen-
dent on the prices of the asset observed at time t. These are unknown at
time t − 1 and could not be used in order to choose the investment at time
t − 1 for t = 1, . . . , T .
These two aspects need special attention. We have to model trading as a dy-
namic process, as opposed to the static trading approach in single-period mar-
ket models. We need also to take care about the evolution over time of the
information available to investors. The second aspect leads to the probabilistic
concepts of a σ-field and a filtration.
Definition 3.1.1. A collection F of subsets of the state space Ω is called a σ-field
(or a σ-algebra) whenever the following conditions hold:
1. Ω ∈ F ,
2. if A ∈ F then Ac = Ω \ A ∈ F ,
∪
3. if Ai ∈ F for i ∈ N then ∞
i=1 Ai ∈ F .
When the state space is finite, the third condition is manifestly equivalent to
the following condition:
3’. if A, B ∈ F then A ∪ B ∈ F ,
which in turn, in view of condition 2., is equivalent to
3”. if A, B ∈ F then A ∩ B ∈ F .
44
3.1.1 Static Information: Partitions

A σ-field is supposed to model a certain quantity of information. If F is chosen
to model the level of information of an agent, it is understood that she can
distinguish between two events A and B which belong to F, but not necessarily
between the particular elements of A or the particular elements of B.
• Intuitively, if an agent looks at the state space Ω then her resolution is not
high enough to recognise the actual states ω, but only to see the particular
subsets belonging to the σ-field F.
• One can think of particular states ω as atoms, whereas the sets contained
in the σ-field can be interpreted as molecules built from these atoms. An
agent can only see the molecules, but usually she does not observe partic-
ular atoms. The larger the σ-field, the higher the resolution is and thus
the information conveyed by the σ-field is richer.
Let I be some index set. By a partition of Ω we mean any collection (Ai )i∈I
of non-empty subsets of Ω such ∪ that the sets Ai are pairwise disjoint, that is,
Ai ∩ Aj = ∅ whenever i ̸= j and i∈I Ai = Ω. It is known that any σ-field on a
finite state space Ω can represented by some partition of Ω.
Definition 3.1.2. We say that a partition (Ai )i∈I generates a σ-field∪F if every
set A ∈ F can be written as a union of some of the Ai s, that is, A = j∈J Aj for
some subset J ⊂ I.
• The sets Ai in a partition satisfy a certain minimality condition. If, for
example, A ∈ F and A ⊂ Aj for some j then A = Aj , since otherwise it
could not be written as a union of some of the Ai s. It is not hard to show
that, given a σ-field F on a finite state space Ω, a partition generating this
σ-field always exists and is in fact unique.
Let us consider an arbitrary mapping X : Ω → R. Recall that the preimage
(i.e., the inverse image) of a set U ⊂ R under X, denoted as X −1 (U ), is defined
by X −1 (U ) := {ω ∈ Ω | X(ω) ∈ U }. The following notion of F-measurability
formalise the statement that the values of X only depend on the information
conveyed by a σ-field F.
Definition 3.1.3. We say that a mapping X : Ω → R is F-measurable, if
for every closed interval [a, b] ⊂ R the preimage under X belongs to F, that is,
X −1 ([a, b]) ∈ F . Then X is called a random variable on (Ω, F).
The extreme case where a = b, and thus the interval [a, b] reduces to a single
point, is also allowed. If we know the partition corresponding to a σ-field F,
then we may establish a handy criterion for the F-measurability. In Proposi-
tion 3.1.1, it is not assumed that the cj s all have different values.
Proposition 3.1.1. Let (Ai )i∈I be a partition generating the σ-field F. Then a
map X : Ω → R is F-measurable if and only if X is constant on each of the sets
of the partition, that is, for every j ∈ I there exists cj ∈ R such that
X(ω) = cj for all ω ∈ Aj .
46 MATH3075/3975
Proof. (MATH3975) (⇒) Assume that (Ai )i∈I is a partition of the σ-field F
and that X is F-measurable. Let j ∈ I be an index and let an element ω ∈ Aj
be arbitrary. Define cj := X(ω). We wish to show that X(ω) = cj for all ω ∈ Aj .
Since X is F-measurable, from Definition 3.1.3 we have that X −1 (cj ) ∈ F .
Therefore, by properties 2. and 3. in the definition of a σ-field, we have
∅ ̸= X −1 (cj ) ∩ Aj ∈ F .
Since, obviously, the inclusion X −1 (cj )∩Aj ⊂ Aj holds, from the aforementioned
minimality property of the sets contained in the partition we obtain that
X −1 (cj ) ∩ Aj = Aj .
But this means that X(ω) = cj for all ω ∈ Aj and hence X is constant on Aj . By
varying j, we obtain such cj s for all j ∈ I.
(⇐) Assume now that X : Ω → R is a function, which is constant on all sets Aj
belonging to the partition and that the cj are given as in the statement of the
proposition. Let [a, b] be a closed interval in R. We define
{ }
C := cj | j ∈ I and cj ∈ [a, b] .
∪
Since i∈I Ai = Ω no other elements then the cj occur as values of X. Therefore,
∪ ∪
X −1 ([a, b]) = X −1 (C) = X −1 (cj ) = Aj ∈ F,
j | cj ∈C j | cj ∈C
where the last equality holds by property 3. of a σ-field.
Example 3.1.1. 1. If Ω is a finite state space, then the power set 2Ω is a

σ-field. This is the largest possible σ-field on Ω, but beware: if Ω is not
finite then the power set of Ω is quite awkward to work with since it is
then hard to define a probability measure P on the power set.
2. The trivial σ-field corresponding to a state space Ω is the σ-field F =
{∅, Ω}. This set clearly satisfies the conditions in Definition 3.1.1. The
trivial σ-field is the smallest σ-field on Ω. Every random variable, which
is measurable with respect to the trivial σ-field is necessarily constant,
that is, deterministic.
3. On a state space Ω = {ω1 , ω2 , ω3 , ω4 } consisting of four elements, the fol-
lowing set is a σ-field:
F = {∅, {ω1 , ω2 }, {ω3 , ω4 }, Ω}.
This can be easily verified. A partition of this σ-field, is given by the two
sets A1 = {ω1 , ω2 } and A2 = {ω3 , ω4 }.
4. If one has a σ-field F and a collection Ai ∈ F for i ∈ I, one can show
that there is always a smallest σ-field which contains the sets Ai . This
σ-field is denoted by σ (Ai | i ∈ I) and is called the σ-field generated by
the collection of sets (Ai )i∈I .
3.1.2 Dynamic Information: Filtrations

We now focus on modelling of the development of information over time.
Definition 3.1.4. A family (Ft )0≤t≤T of σ-fields on (Ω, F) is called a filtration
if Fs ⊂ Ft ⊂ F for all s ≤ t. For brevity, we denote F = (Ft )0≤t≤T .
• We interpret the σ-field Ft as the information available to the agent or
observer at time t. In particular, F0 represents the information available
at time 0, that is, the initial information.
• We assume that the information accumulated over time can only grow, so
that we never forget anything!
To model the dynamic random behaviour, we use the concept of a stochastic
process (or simply a process), that is, a sequence of random variables indexed
by the time parameter t = 0, 1, . . . , T . Recall that we say that X : Ω → R is a
random variable on (Ω, F) whenever X is F-measurable.
Definition 3.1.5. A family X = (Xt )0≤t≤T of random variables on (Ω, F, P) is
called a stochastic process. A stochastic process X is said to be F-adapted if
for every t = 0, 1, . . . , T the random variable Xt is Ft -measurable.
If a filtration F is already known from the context, we also sometimes just write
adapted instead of F-adapted.
We assume from now on that the sample space Ω is equipped with a fixed σ-
field F. Including a probability measure P, which is defined for all sets from F,
we denote this dataset as a triple (Ω, F, P). Given a stochastic process X, we
can define the unique filtration, denoted by FX , such that:
1. X is adapted to the filtration FX ,
2. FX represents the information flow induced by observations of X.
Definition 3.1.6. Let X = (Xt )0≤t≤T be a stochastic process on (Ω, F, P). Define
for any t = 0, 1, . . . , T
( )
FtX := σ Xu−1 ([a, b]) | 0 ≤ u ≤ t, a ≤ b
so that FtX is the smallest σ-field containing all the sets Xu−1 ([a, b]) where 0 ≤
u ≤ t and a ≤ b. Clearly FsX ⊂ FtX for s ≤ t and thus FX := (FtX )0≤t≤T is a
filtration. It follows immediately that a process (Xt )0≤t≤T is FX -adapted. The
filtration FX is said to be generated by the process X.
Example 3.1.2. Let St be the level of the stock price at time 0 ≤ t ≤ T . Since
for t ≥ 1 the price St is not yet known at time 0, on the basis of the initial
information at that time, the price St is modelled by means of a random vari-
able St : Ω → R+ . At time t, however, we observe the price of St and we thus
assume that St is Ft -measurable. We thus model the stock price evolution as
an F-adapted process S, where the filtration F is otherwise specified. In most
cases, we will set F = FS .
48 MATH3075/3975
Example 3.1.3. The following diagram describes the evolution of a single stock
over two time steps:
S: 2 = 9 ω1 P(ω1 ) = 6
tt
25
3
tt
tt
5
ttt
t
S1C = 8J
JJ 2
JJ 5
JJ
JJ
J$
2
5
S2 = 6 ω2 P(ω2 ) = 4
25

S0 =7 5
77
77
77
77 3
775 S: 2 = 6 ω3 P(ω3 ) = 6
77
tt
25
77 5 tt
2
77 ttt
7 t
tt
S1 = 4J
JJ 3
JJ 5
JJ
JJ
J$
S2 = 3 ω4 P(ω4 ) = 9
25
The underlying probability space is given by Ω = {ω1 , ω2 , ω3 , ω4 } and the corre-

sponding σ-field is assumed to be the power set of Ω, i.e. F = 2Ω . The number
next to an arrow indicates the probability with which a given move occurs.
• At time t = 0, the stock price is known and the only one possible value is
S0 = 5. Hence the σ-field F0S is the trivial σ-field.
• At time t = 1, the stock can take two possible values. The following is easy
to verify:


 Ω if a ≤ 4 and b ≥ 8, so that {4, 8} ⊂ [a, b]

{ω , ω } if 4 < a ≤ 8 and b ≥ 8, so that [a, b] ∩ {4, 8} = {8}
S1−1 ([a, b]) = 1 2

 {ω3 , ω4 } if a ≤ 4 and 4 ≤ b < 8, so that [a, b] ∩ {4, 8} = {4}
 ∅ otherwise, that is, when [a, b] ∩ {4, 8} = ∅.
We conclude that F1S = {∅, {ω1 , ω2 }, {ω3 , ω4 }, Ω}. An agent is only able to
tell at time t = 1, whether the true state of the world belongs to the set
{ω1 , ω2 } or to the set {ω3 , ω4 }.
• At time t = 2, an agent is able to pinpoint the true state of the world. The
reason for this is as follows: F2S must contain the sets A1 := S2−1 ([9, 10]) =
{ω1 }, A2 := S2−1 ([6, 9]) = {ω1 , ω2 , ω3 }, A3 := S2−1 ([5, 6]) = {ω2 , ω3 } and A4 :=
S1−1 ([6, 9]) = {ω1 , ω2 } and, since F2S is a σ-field, it must also contain: {ω1 } =
A1 , {ω2 } = A3 ∩ A4 , {ω3 } = A2 \ A4 and {ω4 } = Ac2 . Hence F2S = 2Ω .
3.1.3 Conditional Expectations

Let (Ω, F, P) be a finite probability space and let X be an arbitrary F-measurable
random variable. Assume that G is a σ-field which is contained in F. Let (Ai )i∈I
be the unique partition associated with G.
Definition 3.1.7. The conditional expectation EP (X| G) of X with respect to
G is defined as the random variable which satisfies, for every ω ∈ Ai ,
∑ ∑
EP (X| G)(ω) = 1
P(Ai ) ω∈Ai X(ω)P(ω) = xl xl P(X = xl |Ai )
where the summation is over all possible values of X and

P({X = xl } ∩ Ai )
P(X = xl | Ai ) =
P(Ai )
stands for the conditional probability of the event {ω ∈ Ω | X(ω) = xl } given the
event Ai . This means that
∑ 1
EP (X| G) = EP (X1Ai ) 1Ai . (3.1)
i∈I
P(Ai )
• Note that EP (X| G) is well defined by equation (3.1) and, by Proposition

3.1.1, the conditional expectation EP (X| G) is in fact a G-measurable ran-
dom variable, meaning here that EP (X| G) is constant on each event Ai
from the partition (Ai )i∈I .
• Intuitively, the conditional expectation EP (X| G) is the best estimate of X
given the information represented by the σ-field G.
• One can prove that the following averaging property uniquely charac-
terises the conditional expectation (in addition to G-measurability):
∑ ∑
ω∈G X(ω)P(ω) = ω∈G EP (X| G)(ω)P(ω), ∀ G ∈ G. (3.2)
• One can represent (3.2) in terms of (discrete) integrals. Then it becomes

∫ ∫
X dP = EP (X| G) dP, ∀ G ∈ G.
G G
Using this representation, the definition of conditional expectations can

be extended to arbitrary random variables and σ-fields. This extension is
beyond the scope of this course, however.
Example 3.1.4. We will check that in Example 3.1.3 we have
{ 39
if ω ∈ {ω1 , ω2 },
EP (S2 | F1 ) =
S 5
21
5
if ω ∈ {ω3 , ω4 }.
We have shown already that F1S = {∅, {ω1 , ω2 }, {ω3 , ω4 }, Ω} and that the sets
A1 = {ω1 , ω2 } and A2 = {ω3 , ω4 } form a partition of F1S .
50 MATH3075/3975
We have
P({ω ∈ A1 } ∩ {S2 (ω) = 9}) P(ω1 ) 6
3
P(S2 = 9 | A1 ) = = = 25
= ,
P(A1 ) P({ω1 , ω2 }) 2
5
5
P({ω ∈ A1 } ∩ {S2 (ω) = 6}) P(ω2 ) 4
2
P(S2 = 6 | A1 ) = = = 25
= ,
P(A1 ) P({ω1 , ω2 }) 2
5
5
P({ω ∈ A1 } ∩ {S2 (ω) = 3}) P(∅)
P(S2 = 3 | A1 ) = = = 0,
P(A1 ) P({ω1 , ω2 })
P({ω ∈ A2 } ∩ {S2 (ω) = 9}) P(∅)
P(S2 = 9 | A2 ) = = = 0,
P(A2 ) P({ω3 , ω4 })
P({ω ∈ A2 } ∩ {S2 (ω) = 6}) P(ω3 ) 6
2
P(S2 = 6 | A2 ) = = = 25
= ,
P(A2 ) P({ω3 , ω4 }) 3
5
5
P({ω ∈ A2 } ∩ {S2 (ω) = 3}) P(ω4 ) 9
3
P(S2 = 3 | A2 ) = = = 25
= .
P(A2 ) P({ω3 , ω4 }) 3
5
5
It then follows from Definition 3.1.7 that
3 2 39
EP (S2 | F1S )(ω) = 9 · + 6 · + 3 · 0 = , ∀ ω ∈ A1 ,
5 5 5
2 3 21
EP (S2 | F1S )(ω) = 9 · 0 + 6 · + 3 · = , ∀ ω ∈ A2 .
5 5 5
We also note that
( ) 39 2 21 3 141 6 4 6 9
EP EP (S2 | F1S ) = · + · = = 9· +6· +6· +3· = EP (S2 ).
5 5 5 5 25 25 25 25 25
Proposition 3.1.2. Consider a probability space (Ω, F, P) endowed with sub-σ-
fields G, G1 and G2 of F. Assume furthermore that G1 ⊂ G2 . Then
1. Tower property: If X : Ω → R is an F-measurable random variable then
( ) ( )
EP (X| G1 ) = EP EP (X| G2 ) | G1 = EP EP (X| G1 ) | G2 . (3.3)
2. Pull out property: If X : Ω → R is a G-measurable random variable and

Y : Ω → R is an F-measurable random variable then
EP (XY | G) = X EP (Y | G). (3.4)
3. Trivial conditioning: If X : Ω → R is an F-measurable random variable

independent of G then
EP (X| G) = EP (X). (3.5)
In particular, if G = {∅, Ω} is the trivial σ-field then any random variable

X is independent of G and thus (3.5) is valid. By taking G1 = {∅, Ω} in
(3.3), we obtain EP (EP (X| G)) = EP (X) for any sub-σ-field G.
Abstract Bayes formula. (MATH3975) Assume that two equivalent proba-

bility measures, P and Q say, are given on a space (Ω, F). Suppose that the
Radon-Nikodým density of Q with respect to P equals
dQ
(ω) = L(ω), P-a.s.
dP
meaning that, for every A ∈ F ,
∫ ∫
X dQ = XL dP.
A A
If Ω is finite then this equality takes the following form

∑ ∑
X(ω) Q(ω) = X(ω)L(ω) P(ω).
ω∈A ω∈A
Note that the random variable L is strictly positive P-a.s. Moreover, L is P-

integrable with EP (L) = 1. Finally, it is clear that the equality EQ (X) = EP (XL)
holds for any Q-integrable random variable X (it suffices to take A = Ω). Recall
the following property of the conditional expectation EP (X| G)
∫ ∫
G
X dP = G
EP (X| G) dP, ∀ G ∈ G.
We take for granted that this property uniquely determines the conditional ex-
pectation EP (X| G). Let us also mention that when Ω is finite then any random
variable is P-integrable with respect to any probability measure P on Ω.
Lemma 3.1.1. Let G be a sub-σ-field of F and let X be a random variable in-
tegrable with respect to Q. Then the following abstract version of the Bayes
formula holds
EP (XL | G)
EQ (X | G) = .
EP (L | G)
Proof. It is easy to check that EP (L | G) is strictly positive so that the right-
hand side of the asserted formula is well defined. By our assumption, the
random variable XL is P-integrable. Therefore, it suffices to show that
EP (XL | G) = EQ (X | G)EP (L | G).
Since the right-hand side of the last formula defines a G-measurable random
variable, we need to verify that for any set G ∈ G, we have
∫ ∫
XL dP = EQ (X | G)EP (L | G) dP.
G G
But for every G ∈ G, we obtain

∫ ∫ ∫ ∫
XL dP = X dQ = EQ (X | G) dQ = EQ (X | G)L dP
G ∫ G G ∫ G
( )
=
EP EQ (X | G)L G dP = EQ (X | G) EP (L | G) dP
G G
and was required to demonstrate.

52 MATH3075/3975
3.1.4 Self-Financing Trading Strategies

In order to specify a multi-period market model M = (B, S 1 , . . . , S n ), we need
to describe traded assets and the class of all trading strategies that the agents
in our model are allowed to use. As in Chapter 2, we assume that the market
consists of:
• the money market (i.e. savings) account B with deterministic evolution
given by Bt = (1 + r)t where r denotes the interest rate,
• n risky assets (stocks) S 1 , . . . , S n with prices assumed to follow F-adapted
stochastic processes on the underlying probability space (Ω, F, P) endowed
with a filtration F = (Ft )0≤t≤T .
Definition 3.1.8. A trading strategy is an Rn+1 -valued, F-adapted stochastic
process (φt ) = (φ0t , φ1t , . . . , φnt ) where φjt , j = 1, . . . , n denotes the number of shares
of the jth stock held (or shorted) at time t and φ0t Bt represents the amount of
money in the savings account at time t. The wealth process (or value process)
of a trading strategy φ = (φt )0≤t≤T is the stochastic process V (φ) = (Vt (φ))0≤t≤T
where
∑
n
Vt (φ) = φ0t Bt + φjt Stj .
j=1
In general, it is not reasonable to allow all trading strategies. A useful class of

trading strategies is the class of self-financing trading strategies.
Definition 3.1.9. A trading strategy φ = (φt )0≤t≤T is called self-financing if
for all t = 0, . . . , T − 1
∑n ∑n
φ0t Bt+1 + j=1 φjt St+1
j
= φ0t+1 Bt+1 + j=1
j
φjt+1 St+1 . (3.6)
The financial interpretation of equation (3.6) can be summarised as follows:

• An agent invests money at the beginning of the period, i.e. at time t and
he only can adjust his portfolio at time t + 1. Hence he is not allowed to do
any trading during the time interval (t, t + 1).
• The left-hand side of equation (3.6) represents the value of the agent’s
portfolio at time t + 1 before rebalancement, whereas the right-hand side
of equation (3.6) represents the value of his portfolio at time t + 1 after the
portfolio was revised.
• The self-financing condition says that these two values must be equal and
this means that money was neither withdrawn nor added.
• If we set t = T − 1 then both sides of formula (3.6) represent the agent’s
wealth at time T , that is, VT (φ).
It is often useful to represent the self-financing condition in terms of the gains

process. Given a trading strategy φ, the corresponding gains process G(φ) =
(Gt (φ))0≤t≤T is given by
Gt (φ) := Vt (φ) − V0 (φ). (3.7)
Definition 3.1.10. A multi-period market model M = (B, S 1 , . . . , S n ) is

given by the following data:
1. A probability space (Ω, F, P) endowed with a filtration F = (Ft )0≤t≤T .
2. The money market account B given by Bt = (1 + r)t .
3. A number of financial assets with prices S 1 , . . . , S n , which are assumed to

be F-adapted stochastic processes.
4. The class Φ of all self-financing trading strategies.
Assume that we are given a multi-period market model, as described in Defi-

nition 3.1.10.
Definition 3.1.11. The increment process ∆S j corresponding to the jth stock

is defined by
j
∆St+1 j
:= St+1 − Stj for t = 0, . . . , T − 1. (3.8)
The increment process ∆B of the money market account are given by
∆Bt+1 := Bt+1 − Bt = (1 + r)t r = Bt r
for all t = 0, . . . , T − 1.
As in the case of a single-period market model, it is will be convenient to con-

sider the discounted price processes as well.
Definition 3.1.12. The discounted stock prices are given by
Sbtj :=
Stj
Bt
, (3.9)
so that the increments of discounted prices are
j
∆Sbt+1 := Sbt+1 − Sbtj =
St+1 Stj
j j
Bt+1
− Bt
. (3.10)
54 MATH3075/3975
The discounted wealth process Vb (φ) of a trading strategy φ = (φt )0≤t≤T is

given by
Vbt (φ) := Vt (φ)

Bt
. (3.11)
b
The discounted gains process G(φ) of a trading strategy φ = (φt )0≤t≤T equals
bt (φ) := Vbt (φ) − Vb0 (φ).

G (3.12)
The self-financing condition (3.6) can now be reformulated as follows:

Proposition 3.1.3. Consider a multi-period market model M. An F-adapted
trading strategy φ = (φt )0≤t≤T is self-financing if and only if any of the two
equivalent statements hold:
∑ ∑t−1 ∑n
1. Gt (φ) = t−1 j=1 φu ∆Su+1 for all 1 ≤ t ≤ T ,
0 j j
u=0 φu ∆Bu+1 + u=0
bt (φ) = ∑t−1 ∑n φju ∆Sbu+1

2. G j
for all 1 ≤ t ≤ T .
u=0 j=1
Proof. The proof is elementary and thus it is left as an exercise. Note that
b
the process G(φ) given by condition 2. does not depend on the component φ0 of
φ ∈ Φ.
3.1.5 Risk-Neutral Probability Measures

We will now redefine the general concepts of financial mathematics, such as ar-
bitrage opportunity, arbitrage-free model, replicating strategy, arbitrage price,
etc., in the framework of a multi-period market model. This is also good revi-
sion of the basic ideas from Section 2.2.
Definition 3.1.13. A trading strategy φ = (φt )0≤t≤T ∈ Φ is an arbitrage op-
portunity if
1. V0 (φ) = 0,
2. VT (φ)(ω) ≥ 0 for all ω ∈ Ω,
3. VT (φ)(ω) > 0 for some ω ∈ Ω or, equivalently, EP (VT (φ)) > 0.
We say that a multi-period market model M is arbitrage-free if no arbitrage
opportunities exist.
As in Chapter 2, one can use either the discounted wealth process or the dis-
counted gains process in order to express the arbitrage conditions. It is also
important to note that the conditions 1.–3. in Definition 3.1.13 hold under P
whenever they are satisfied under some probability measure Q equivalent to P.
Let us now come to the point of risk-neutral probability measures. As in Sec-

tion 2.2, we will find that risk-neutral probability measures are very closely
related to the question of arbitrage-free property and completeness of a model.
In the dynamic setup, the concept of a risk-neutral probability measure has

to be extended. Using the notion of the conditional expectation, we can now
define a risk-neutral probability measure in the multi-period framework.
Definition 3.1.14. A probability measure Q on Ω is called a risk-neutral

probability measure for a multi-period market model M = (B, S 1 , . . . , S n )
whenever
1. Q(ω) > 0 for all ω ∈ Ω,
2. EQ (∆Sbt+1
j
| Ft ) = 0 for all j = 1, . . . , n and t = 0, . . . , T − 1.
We denote by M the class of all risk-neutral probability measures for M.
3.1.6 Martingales
Observe that condition 2. in Definition 3.1.14 is equivalent to the equality, for
all t = 0, . . . , T − 1,
( j )
EQ Sbt+1 | Ft = Sbtj .
The last equation leads us into the world of martingales, that is, stochastic
processes representing fair games.
Definition 3.1.15. An F-adapted process X = (Xt )0≤t≤T on a finite probability

space (Ω, F, P) is called a martingale whenever for all s < t
EP (Xt | Fs ) = Xs .
To establish the equality in Definition 3.1.5, it suffices to check that for every
t = 0, 1, . . . , T − 1
EP (Xt+1 | Ft ) = Xt .
Whether a given process X is a martingale obviously depends on the choice of
the filtration F as well as the probability measure P under which the condi-
tional expectation is evaluated.
Lemma 3.1.2. Let Q ∈ M be any risk-neutral probability measure for a multi-

period market model M = (B, S 1 , . . . , S n ). Then the discounted stock price Sbj is
martingale under Q for every j = 1, . . . , n.
Proof. (MATH3975) For s < t and arbitrary i we have that
∑
t
Sbtj = Sbsj + ∆Sbuj
u=s+1
56 MATH3075/3975
where ∆Sbuj = Sbuj − Sbu−1

j
. Hence, using the additivity as well as properties (3.4)
and (3.5) of the conditional expectation under Q, we obtain
( ∑t )
b b bj
EQ (St | Fs ) = EQ Ss +
j j
∆Su Fs
u=s+1
( ) ∑
t
( )
= EQ Sbsj | Fs + EQ ∆Sbuj | Fs
u=s+1
( )
∑
t
( )
= Sbsj + EQ EQ ∆Sbuj Fu−1 Fs
u=s+1
| {z }
=0
= Sbsj .
Note that to conclude that the braced expression is equal to 0, we used condition
2. in Definition 3.1.14.
The previous lemma explains why risk-neutral probability measures are also
referred to as martingale measures. As the next result shows, one can in fact
go one step further.
Proposition 3.1.4. Let φ ∈ Φ be a trading strategy. Then the discounted wealth
process Vb (φ) and the discounted gains process G(φ)
b are martingales under any
risk-neutral probability measure Q ∈ M.
Proof. (MATH3975) From equation (3.12), it follows that for all t = 0, . . . , T
Vbt (φ) = Vb0 (φ) + G
bt (φ).
Since Vb0 (φ) (the initial endowment) is a constant, it suffices to show that the
b
process G(φ) is a martingale under any Q ∈ M. From Proposition 3.1.3, we
obtain
∑
n ∑
t−1
b b
Gt (φ) = Gs (φ) + φju ∆Sbu+1
j
.
j=1 u=s
Therefore, arguing as in the proof of Lemma 3.1.2, we get

∑n ∑t−1 ( )
b b
EQ (Gt (φ)|Fs ) = Gs (φ) + EQ φju ∆Sbu+1
j
Fs
j=1 u=s
∑
n ∑
t−1 ( ( ) )
bs (φ) +
= G EQ EQ φju ∆Sbu+1
j
Fu Fs
j=1 u=s
( )
∑
n ∑
t−1
( j )
bs (φ) +
= G b
EQ φu EQ ∆Su+1 Fu Fs
j
j=1 u=s
| {z }
=0
bs (φ),
= G
where we used the fact that φju is Fu -measurable, as well as property (3.5) of
the conditional expectation.
3.1.7 Fundamental Theorem of Asset Pricing

The Fundamental Theorem of Asset Pricing is still valid in the framework of
a multi-period market model. The set of allowed trading strategies Φ in a
multi-period market model is assumed to be the full set of all self-financing
and F-adapted trading strategies. In that case, the relationship between the
existence of a risk-neutral probability measure and no arbitrage is “if and only
if”. We will only prove here the following implication:
Existence of a risk-neutral probability measure ⇒ No arbitrage.
As in the single-period case, the proof of the inverse implication relies on the
separating hyperplane theorem, but is more difficult. Hence the proof of
part (ii) in Theorem 3.1.1 is omitted.
Theorem 3.1.1. Fundamental Theorem of Asset Pricing. Given a multi-
period market model M = (B, S 1 , . . . , S n ), the following statements hold:
(i) if the class M of risk-neutral probability measures is non-empty then there
are no arbitrage opportunities in Φ and thus the model M is arbitrage-free,
(ii) if there are no arbitrage opportunities in the class Φ of all self-financing
trading strategies then there exists a risk-neutral probability measure, so that
the class M is non-empty.
Proof. (MATH3975) We will only prove part (i). We postulate that Q is a risk-
neutral probability measure for a general multi-period market model. Our goal
it to show that the model is arbitrage-free. To this end, we argue by contradic-
tion. Let us thus assume that there exists an arbitrage opportunity φ ∈ Φ.
Such a strategy would satisfy the following conditions (see Proposition 2.2.1):
1. the initial endowment Vb0 (φ) = 0,
2. the discounted gains process G bT (φ) ≥ 0,
3. there exists at least one ω ∈ Ω such that GbT (φ)(ω) > 0.
On the one hand, from conditions 2. and 3. above, we deduce easily that
( )
EQ G bT (φ) > 0.
On the other hand, using the definition of the discounted gains process, prop-
erty 2. of a risk-neutral probability measure (see Definition 3.1.14), and prop-
erties (3.4) and (3.5) of the conditional expectation, we obtain
(∑ n ∑
T −1 ) ∑ n ∑T −1
( ) ( )
b
EQ GT (φ) = EQ j bj
φu ∆Su+1 = EQ φju ∆Sbu+1
j
j=1 u=0 j=1 u=0
∑
n ∑
T −1
( ) ∑
n ∑
T −1
( )
= EQ EQ (φju ∆Sbu+1
j
| Fu ) = EQ φju EQ (∆Sbu+1
j
| Fu ) = 0.
j=1 u=0 j=1 u=0
| {z }
=0
This clearly contradicts the inequality obtained in the first step. Hence there
cannot be an arbitrage in the market model M = (B, S 1 , . . . , S n ).
58 MATH3075/3975
Remark 3.1.1. (MATH3975) We already know from Proposition 3.1.4 that

b
G(φ) is a martingale under Q. Hence we can conclude from (3.5) that
bT (φ)) = EQ (G
EQ (G bT (φ) | F0 ) = G
b0 (φ) = 0.
This observation simplifies the argument used in the proof of Theorem 3.1.1
significantly.
3.1.8 Pricing of European Contingent Claims

Let us extend the concepts of a European contingent claim and its replication.
Note that a contingent claim of European style can only be exercised at its
maturity date T (as opposed to contingent claims of American style studied in
Section 4.4). A European contingent claim in a multi-period market model
is an FT -measurable random variable X on Ω to be interpreted as the payoff
at the terminal date T . For brevity, European contingent claims will also be
referred to as contingent claims.
Definition 3.1.16. A replicating strategy (or hedging strategy) for a con-
tingent claim X is a trading strategy φ ∈ Φ such that VT (φ) = X, that is, the
terminal wealth of the trading strategy matches the claim’s payoff.
The standard method to price a contingent claim is to employ the replication
principle. The price will now depend on time t and thus one has to specify a
whole price process π(X), rather than just an initial price, as in the single-
period market model. Therefore, we need first to extend Definition 2.2.7 to a
multi-period setup. Obviously, πT (X) = X for any claim X.
Definition 3.1.17. We say that an F-adapted stochastic process (πt (X))0≤t≤T is
a price process for the contingent claim X that complies with the principle
of no-arbitrage if there is no F-adapted and self-financing arbitrage strategy
in the extended model M f = (B, S 1 , . . . , S n , S n+1 ) with an additional asset S n+1
given by St = πt (X) for 0 ≤ t ≤ T − 1 and STn+1 = X.
n+1
The following result generalises Proposition 2.2.3. We deal here with an

attainable claim, meaning that we assume a priori that a replicating strategy
for X exists. The proof of Proposition 3.1.5 is left as an exercise.
Proposition 3.1.5. Let X be a contingent claim in an arbitrage-free multi-
period market model M and let φ ∈ Φ be any replicating strategy for X. Then
the only price process of X that complies with the principle of no-arbitrage is the
wealth process V (φ).
• The arbitrage price of X at time t is unique and it is also denoted as πt (X).
Hence the equality πt (X) = Vt (φ) holds for any replicating strategy φ ∈ Φ
for X.
• In particular, the price at time t = 0 is the initial endowment of any repli-
cating strategy for X, that is, π0 (X) = V0 (φ) for any strategy φ ∈ Φ such
that VT (φ) = X.
Example 3.1.5. Let us consider the two-period model M = (B, S) with the
stock price S specified as in Example 3.1.3. We assume that the interest rate is
equal to zero, so that the equality Bt = 1 holds for t = 0, 1, 2.
We will examine the digital call paying one unit of cash at maturity date T = 2
whenever the stock price at this date is greater or equal than 8, that is,
{
1 if S2 ≥ 8,
X=
0 otherwise.
In our model, the contingent claim X pays off one unit of cash if and only if the
state of the world is ω1 and zero else.
How can we find a self-financing replicating strategy for X, that is, a self-
financing strategy φ satisfying
V2 (φ)(ω) = φ02 (ω) + φ12 (ω)S2 (ω) = X(ω)?
The computational trick now is to decompose the model into single-period mar-
ket models. In our case, we need to examine three sub-models: one starting at
time t = 0 going until time t = 1 with price S0 = 5, one starting at time t = 1
with price S1 = 8 and one starting at time t = 1 with price S1 = 4. To solve our
problem, we proceed by the backward induction with respect to t.
Step 1. We start by solving the hedging problem at time t = 1 assuming that

S1 (ω) = 8, that is, on the set {ω1 , ω2 }. This problem is in fact the same as in
the two-state single-period market model and thus the delta hedging formula
applies. Denoting the still unknown hedging strategy with φ = (φt )0≤t≤2 , we
find
1−0 1
φ11 (ω) = =
9−6 3
for ω ∈ {ω1 , ω2 }. As usual, the net amount of money available at time t = 1,
that is, V1 (φ)(ω) − 13 · 8 is invested in the money market account. In order to
hedge in this situation, we must have
( )
1 1
V1 (φ)(ω1 ) − · 8 + · 9 = 1,
3 3
( )
1 1
V1 (φ)(ω2 ) − · 8 + · 6 = 0.
3 3
It is easily seen that these equations are satisfied for V1 (φ)(ω1 ) = 32 = V1 (φ)(ω2 )
and thus π1 (X)(ω) = 32 for ω ∈ {ω1 , ω2 }. The equality of the wealth process for
the two states ω1 and ω2 is by no means a coincidence; it must hold, since V1 (φ)
has to be F1S -measurable (see Proposition 3.1.1). We then have, for ω ∈ {ω1 , ω2 },
2 1
φ01 (ω) = − · 8 = −2
3 3
so that two units of cash are borrowed.
60 MATH3075/3975
Let us now examine the hedging problem at time t = 1 when the stock price
is S1 (ω) = 4, that is, on the set {ω3 , ω4 }. There is no chance that the digital
call will payoff anything else than zero. The hedging strategy for this payoff is
trivial, invest zero in the money market account and invest zero in the stock.
We therefore have, for ω ∈ {ω3 , ω4 },
φ01 (ω) = φ11 (ω) = 0
and thus V1 (φ)(ω3 ) = 0 = V1 (φ)(ω4 ). Hence π1 (X)(ω) = 0 for ω ∈ {ω3 , ω4 }.
Step 2. In the second step, we consider the hedging problem at time t = 0

with price S0 = 5. In order to find a hedge of the digital call, it suffices to
replicate the contingent claim V1 (φ) in the single-period model, where
{ 2
3
if ω ∈ {ω1 , ω2 },
V1 (φ) = π1 (X) =
0 if ω ∈ {ω3 , ω4 }.
The rationale is the following: if we hedge this contingent claim in the period
from time t = 0 to time t = 1 and next we follow the strategy we computed at
time t = 1, then we are done. Solving the hedging problem at time t = 0 is easy.
By applying once again the delta hedging formula, we obtain
−0 2
1
φ10 = 3
= .
8−4 6
In order to replicate V1 (φ), we must have, for all ω ∈ {ω1 , ω2 , ω3 , ω4 },
( )
1 1 2
V0 (φ)(ω) − · 5 + · 8 = ,
6 6 3
( )
1 1
V0 (φ)(ω) − · 5 + · 4 = 0.
6 6
It appears that both equations are satisfied for V0 (φ) = 61 and thus this number
is the price of the digital call in this model. It is also clear that φ00 (ω) = − 54 .
One can now check that the digital call can be dynamically hedged using the
self-financing trading strategy computed above.
As in a single-period market model, there may be contingent claims for which
no replicating strategy in Φ exists.
Definition 3.1.18. A contingent claim X is called attainable in M if there
exists a trading strategy φ ∈ Φ which replicates X, i.e. satisfies VT (φ) = X.
For an attainable contingent claim, the replication principle applies and it is
clear how to price such a claim, namely, by the initial endowment needed for a
replicating strategy.
Definition 3.1.19. A multi-period market model M is called complete if and
only if for any contingent claim X there exists a replicating strategy φ. A model
which is not complete is called incomplete.
Example 3.1.6. One can show that a multi-period version of the stochastic
volatility model introduced in Example 2.2.3 is an incomplete multi-period
market model. This is left as an exercise.
3.1.9 Risk-Neutral Valuation of European Claims

Let us now consider a general European contingent claim X (attainable or not)
and assume that Φ consists of all self-financing and F-adapted trading strate-
gies. We assume that the market model M = (B, S 1 , . . . , S n ) is arbitrage-free.
Proposition 3.1.6. Let X be a contingent claim (possibly non-attainable) and
Q any risk-neutral probability measure for the multi-period market model M.
Then the risk-neutral valuation formula
( )

πt (X) = Bt EQ X
BT Ft (3.13)
defines a price process π(X) = (πt (X))0≤t≤T for the contingent claim X that
complies with the principle of no-arbitrage.
Proof. (MATH3975) The proof of this result is essentially the same as the
proof of Proposition 2.2.4 and thus it is left as an exercise.
• If a claim X is attainable then the conditional expectation in the right-
hand side of (3.13) does not depend on the choice of a risk-neutral proba-
bility measure Q.
• If a claim X is not attainable then, as in the case of single-period market
models, one faces the problem of non-uniqueness of a price complying with
the principle of no-arbitrage. Indeed, unless a claim X is attainable, using
Proposition 3.1.6, we obtain an interval of prices π0 (X) for the claim X.
As a criterion for completeness, one has the following result, which should be
seen as a multi-period counterpart of Theorem 2.2.2.
Theorem 3.1.2. Assume that a multi-period market model M = (B, S 1 , . . . , S n )
is arbitrage-free. Then M is complete if and only if there is only one risk-neutral
probability measure.
Example 3.1.7. Let us return to Examples 3.1.3 and 3.1.5. As before, we as-
sume that the interest rate is equal to zero, i.e., r = 0. We will re-examine
valuation of the digital call that pays one unit of cash at maturity date T = 2
whenever the stock price at this date is greater or equal than 8, that is,
{
1 if S2 ≥ 8,
X=
0 otherwise.
We already know that the arbitrage price equals π0 (X) = 61 and that this price
is in fact unique, since the claim X can be replicated. We will now compute the
arbitrage price of the digital call using the risk-neutral pricing formula (3.13).
For this purpose, we will first compute a risk-neutral probability measure Q =
(q1 , q2 , q3 , q4 ). The first two conditions for the qi s are obtained by the fact that Q
is a probability measure and condition 1. in Definition 3.1.14, that is:
q1 + q2 + q3 + q4 = 1, qi > 0, i = 1, 2, 3, 4.
62 MATH3075/3975
Additional conditions for the qi s can be obtained from property 2. in Definition

3.1.14. To this end, we note that since the interest rate is equal to zero, the
non-discounted prices agree with the discounted prices. The first condition is
obtained by taking t = 0
5 = EQ (Sb1 ) = q1 · 8 + q2 · 8 + q3 · 4 + q4 · 4.
Upon substituting q4 = 1 − q1 − q2 − q3 , we obtain
5 = 4 + 4(q1 + q2 ),
which is equivalent to
1
q1 + q2 = , (3.14)
4
so that necessarily
3
q3 + q4 = . (3.15)
4
We now examine the conditional expectation for t = 1. Using the definition of
the conditional expectation, we see that for ω ∈ {ω1 , ω2 } = A1 , we have
q1 q2
8 = EQ (Sb2 | F1S )(ω) = 9 +6 = 36q1 + 24q2 ,
q1 + q2 q1 + q2
which in turn yields
9q1 + 6q2 = 2. (3.16)
Furthermore, for ω ∈ {ω3 , ω4 } = A2 we must have
q3 q4
4 = EQ (Sb2 | F1S )(ω) = 6 +3 = 8q3 + 4q4
q3 + q4 q3 + q4
so that
2q3 + q4 = 1. (3.17)
Now, we have four equations (3.14), (3.15), (3.16) and (3.17) with four un-
knowns q1 , q2 , q3 and q4 . The unique solution of this system is
( )
1 1 1 1
Q = (q1 , q2 , q3 , q4 ) = , , , . (3.18)
6 12 4 2
We conclude that the market model M = (B, S) is arbitrage-free and complete.
In order to compute the price of the digital call option, which is represented
by the claim X = 1{S2 ≥8} , we observe that X(ω) equals 1 for ω = ω1 and it is
equal to 0 for all other ωs. Since r = 0, the discounting is irrelevant, and thus
1
π0 (X) = EQ (X) = q1 · 1 + q2 · 0 + q3 · 0 + q4 · 0 = .
6
One can also compute the price π1 (X) using formula (3.13).
Note that the computation of the unique risk-neutral probability Q can sim-
plified if we focus on the conditional probabilities, that is, the risk-neutral
probabilities of upward and downward mouvements of the stock price over one
period.
The following diagram describes the evolution of the stock price under Q
S2 = 9 ω1
2
sss9
s
sss
3
sss
S1B = 8K
KKK 1
KK3K
KKK
1 %
4
S 2 = 6 ω2

S0 =8 5
88
88
88
88 43
88 S2 = 6 ω3
88 1
sss9
88 s
sss
3
88
sss
S1 = 4K
KKK 2
KK3K
KKK
%
S2 = 3 ω4
where probabilities are the conditional risk-neutral probabilities, for instance,
1
Q(S2 = 6 | S0 = 5, S1 = 4) = Q(S2 = 6 | S1 = 4) = .
3
Let us now consider the claim Y = (Y (ω1 ), Y (ω2 ), Y (ω3 ), Y (ω4 ) = (1, 1, 3, −6)
maturing at T = 2. Using formula (3.13), we obtain the price process π(Y ):
Y (ω1 ) = 1 ω1
n6
nnn
2
nn
nnn
3
nnn
π1 (Y ) = 1
}> PPP 1
PPP 3
}}} PPP
}} PPP
1
4 }
}} (
}} Y (ω2 ) = 1 ω2
}}}
}
}}}
}}
π0 (Y ) = −2
AA
AA
AA
AA 3
AA 4
AA Y (ω3 ) = 3 ω3
AA n6
AA nnn
1
AA nn
nnn
3
AA
nnn
π1 (Y ) = −3
PPP 2
PPP 3
PPP
PPP
(
Y (ω4 ) = −6 ω4
Chapter 4
The Cox-Ross-Rubinstein Model
The most widely used example of a multi-period market model is the binomial
options pricing model, which we are going to discuss in this section. This
model is also commonly known as the Cox-Ross-Rubinstein model, since it
was proposed in the seminal paper by Cox et al. (1979). For brevity, it is also
frequently referred to as the CRR model.
As we will see in what follows, the binomial market model is the concatenation
of several elementary single-period market models, as discussed in Section 1.1.
We assume here that we have one stock S and the money market account B,
but a generalization to the case of more than one stock is also possible.
4.1 Multiplicative Random Walk

At each point in time, the stock price is assumed to either go ‘up’ by a fixed
factor u or go ‘down’ by a fixed factor d. We only assume that 0 < d < u, but
we do not postulate that d < 1 < u. The probability of an ‘up’ mouvement
is assumed to be the same 0 < p < 1 for each period, and is assumed to be
independent of all previous stock price mouvements.
Definition 4.1.1. A stochastic process X = (Xt )1≤t≤T on a probability space

(Ω, F, P) is called the Bernoulli process with parameter 0 < p < 1 if the ran-
dom variables X1 , X2 , . . . , XT are independent and have the following common
probability distribution
P(Xt = 1) = 1 − P(Xt = 0) = p.
The Bernoulli counting process N = (Nt )0≤t≤T is defined by setting N0 = 0

and by postulating that, for every t = 1, . . . , T and ω ∈ Ω,
Nt (ω) := X1 (ω) + X2 (ω) + · · · + Xt (ω).
64
The Bernoulli counting process is a very special case of an additive random

walk. The stock price process in the CRR model is defined via a deterministic
initial value S0 > 0 and for 1 ≤ t ≤ T and all ω ∈ Ω
St (ω) := S0 uNt (ω) d t−Nt (ω) . (4.1)
• The idea behind this construction is that the underlying Bernoulli process
X governs the ‘up’ and ‘down’ mouvements of the stock. The stock price
moves up at time t if Xt (ω) = 1 and moves down if Xt (ω) = 0. The dynamics
of the stock price can thus be seen as an example of a multiplicative
random walk.
• The Bernoulli counting process N counts the up mouvements. Before and
including time t, the stock price moves up Nt times and down t − Nt times.
Assuming that the stock price can only move up (resp. down) by the fac-
tors u (resp. d) we obtain equation (4.1) governing the dynamics of the
stock price over time.
Why is this model called the binomial model? The reason is that for each t, the
random variable Nt has the binomial distribution with parameters p and t.
To be more specific, for every t = 1, . . . , T and k = 0, . . . , t we have that
(t)
P(Nt = k) = k
pk (1 − p)t−k
and thus the probability distribution of the stock price St at time t is given by
(t)
P(St = S0 uk dt−k ) = k
pk (1 − p)t−k (4.2)
for k = 0, 1, . . . , t.
It it easy to show that the filtration FS = (FtS )0≤t≤T generated by the stock
price S = (St )1≤t≤T coincides with the filtration FX = (FtX )0≤t≤T generated by
the Bernoulli process X = (Xt )1≤t≤T , where F0X is set to be the trivial σ-field,
by definition. As usual, the money market account B is assumed to be defined
via B0 = 1 and
Bt = (1 + r)t . (4.3)
Definition 4.1.2. The binomial market model (or the CRR model) with
parameters p, u, d, r and time horizon T on a probability space (Ω, F, P) is the
multi-period market model consisting of the stock and the money market ac-
count, where the stock price evolution is governed by equation (4.1) and the
money market account satisfies (4.3). The underlying filtration F is assumed to
be the filtration FS = (FtS )0≤t≤T generated by S and the set of allowed trading
strategies Φ is given by all self-financing and FS -adapted trading strategies.
66 MATH3075/3975
Sample paths of the CRR model can be represented through the following lat-
tice, for T = 4,
4
S0 u
tt:
t
tt
ttt
t
3
;S0 u JJ
vvv JJ
vv JJ
vv JJ
vv J$
2
S0 u H S: 0 u3 d
yy< HH
HH tt
yy HH tt
yy HH ttt
yy # tt
S0 u E S; 0 u2 dJ
||> EE
EE vv JJ
|| vv JJ
| EE vv JJ
| EE v JJ
|| " vv $
S0 B S0 udH S0 u2 d2
BB yy< HH tt:
BB yy HH t
BB y HH tt
B yyy HH ttt
y # t
S0 d E 2
S; 0 ud J
EE vv JJ
EE v JJ
EE vvv JJ
E" vv JJ
v $
S0 d2 H S: 0 ud 3
HH tt
HH t
HH tt
HH tt
# tt
S0 d 3 J
JJ
JJ
JJ
JJ
$
S0 d4
Given a binomial market model, the underlying Bernoulli process X can be

recovered from the stock prices. Clearly, Xt = Nt − Nt−1 and thus
St S0 uNt dt−Nt
=
St−1 S0 uNt−1 dt−1−Nt−1
= uNt −Nt−1 d1−(Nt −Nt−1 )
Xt 1−Xt
= u
{ d
u if Xt (ω) = 1,
=
d if Xt (ω) = 0.
We will now address the crucial question:
Is the CRR market model arbitrage-free?

Proposition 4.1.1. Assume that d < 1 + r < u. Then a probability measure P e

on (Ω, FT ) is a risk-neutral probability measure for the CRR model M = (B, S)
with parameters p, u, d, r and time horizon T if and only if:
e
1. X1 , X2 , X3 , . . . , XT are independent under the probability measure P,
e (Xt = 1) < 1 for all t = 1, . . . , T ,
2. 0 < pe := P
3. peu + (1 − pe)d = (1 + r),
where X is the Bernoulli process governing the stock price process S.
The proof of Proposition 4.1.1 is not hard and thus we leave it as an exercise.
Note that condition 3. here is the same as in Section 2.1 and it is equivalent to
pe = 1+r−d
u−d
. (4.4)
We thus see that the binomial model is arbitrage-free whenever d < 1 + r < u.
This is exactly the same condition as in a single-period model of Section 2.1. As
the value for pe in Proposition 4.1.1 is also unique, the risk-neutral probability
e is unique and thus, from Theorem 2.2.2, the binomial model is complete.
P
If d < 1 + r < u then the CRR market model M = (B, S) is

arbitrage-free and complete.
• We assume from now on that d < 1 + r < u. Otherwise, the CRR model is
not arbitrage-free.
• The sample paths of the CRR process are represented by the recombin-
ing tree (i.e. the lattice). Recall that since Nt can only take t + 1 values,
so does the stock price St for and date t = 0, 1, . . . , T . The corresponding
number of different sample paths of the stock price between times 0 and
t equals 2t , so that the number of sample paths grows very quickly when
the number of periods rises.
• For the sake of computational simplicity, it is sometimes postulated that
d = u−1 . (4.5)
In that case, formula (4.1) simplifies as follows
St = S0 u2Nt −t . (4.6)
68 MATH3075/3975
4.2 The CRR Call Pricing Formula

Since the CRR model is complete, the unique arbitrage price for the call option
can be computed using the risk-neutral valuation formula (3.13).
Proposition 4.2.1. The arbitrage price at time t = 0 of the European call option
CT = (ST − K)+ in the binomial market model M = (B, S) with parameters u, d
and r is given by the Cox-Ross-Rubinstein call pricing formula
∑T ( ) T ( )
∑
T k T −k K T k
C0 = S0 pb (1 − pb) − pe (1 − pe)T −k
k (1 + r)T k
k=b
k k=b
k
where
1+r−d peu
pe = , pb =
u−d 1+r
and b
k is the smallest integer k such that
(u) ( )
K
k log > log .
d S0 d T
Proof. The price at time t = 0 of the claim X = CT = (ST − K)+ can be
computed using the risk-neutral valuation under P e
( )
CT
C0 = EPe .
(1 + r)T
In view of Proposition 4.1.1, this can be represented more explicitly as follows
∑T ( )
1 T k T −k
( k T −k
)
C0 = pe (1 − e
p ) max 0, S 0 u d − K .
(1 + r)T k=0 k
We note that
( )
( u )k K log K
S0 dT
S0 uk dT −k − K > 0 ⇔ > ⇔ k> (u) .
d S0 dT log d
We define b k = bk(S0 , T ) as the smallest integer k such that this inequality is

satisfied. If there are less than bk upward mouvements, there is no chance that
the option will pay off anything (i.e. it will expire worthless). Therefore, we
obtain
k−1 ( )
b
∑
1 T k
C0 = T
pe (1 − pe)T −k 0
(1 + r) k=0 k
∑T ( )
1 T k ( )
+ T
pe (1 − pe)T −k S0 uk dT −k − K
(1 + r) k
k=b
k
∑
T ( ) T ( )
∑
S0 T k T −k k T −k K T k
= pe (1 − pe) u d − pe (1 − pe)T −k
(1 + r)T k (1 + r)T k
k=b
k k=b
k
so that
T ( )(
∑ )k ( )T −k T ( )
∑
T peu (1 − pe)d K T k
C0 = S0 − T
pe (1 − pe)T −k
k 1+r 1+r (1 + r) k
k=bk k=b
k
∑(
T )
T k K
T ( )
∑ T k
T −k
= S0 pb (1 − pb) − pe (1 − pe)T −k
k (1 + r)T k
k=b
k k=b
k
peu
where we write pb = 1+r
.
peu
Example 4.2.1. (MATH3975) Check that 0 < pb = 1+r < 1 whenever 0 < pe < 1.
b
Let P be the probability measure obtained by setting p = pb in Definition 4.1.1.
Show that the process BS is an F-martingale under P b and the price of the call
satisfies
b
C0 = S0 P(D) e
− KB(0, T ) P(D)
where D = {ω ∈ Ω : ST (ω) > K}. Using the abstract Bayes formula of Lemma
3.1.1, establish the second equality in the following formula
( )
b | Ft ) − KB(t, T ) P(D
Ct := Bt EPe BT−1 (ST − K)+ | Ft = St P(D e | Ft ).
4.2.1 Put-Call Parity

Since CT − PT = ST − K, we see that the following put-call parity holds at any
date t = 0, 1, . . . , T
Ct − Pt = St − K(1 + r)−(T −t) = St − KB(t, T )
where B(t, T ) = (1 + r)−(T −t) represents the price at time t of the zero-coupon
bond paying one unit of cash at time T . Using Proposition 4.2.1 and the put-call
parity, one can derive an explicit pricing formula for the European put option
with the payoff PT = (K − ST )+ . This is left as an exercise.
4.2.2 Pricing Formula at Time t

The pricing formula for the European call option, established in Proposition
4.2.1, can be extended to the case of any date t = 0, 1, . . . , T − 1, specifically,
∑
T −t ( ) ∑
T −t ( )
T −t k T −t−k K T −t k
Ct = St pb (1−b
p) − p)T −t−k
pe (1−e
k (1 + r)T −t k
k=b
k(St ,T −t) k=b
k(St ,T −t)
where b
k(St , T − t) is the smallest integer k such that
(u) ( )
K
k log > log .
d St dT −t
Note that Ct = C(St , T − t) meaning that the call option price depends on the
time to maturity T − t and the level St of the stock price observed at time t, but
it is independent of the evolution of the stock price prior to t.
70 MATH3075/3975
4.2.3 Replicating Strategy

To compute the replicating strategy for the call option, we note that the CRR
model can be seen as a concatenation of single-period models. For instance,
if we wish to find the replicating portfolio at time t = 0, we observe that the
replicating portfolio (φ00 , φ10 ) satisfies
φ00 + φ10 S0 = V0 (φ) = C0
and
φ00 (1 + r) + φ10 S1u = C1u ,

φ00 (1 + r) + φ10 S1d = C1d ,
where in turn C1u = C(uS0 , T − 1) and C1d = C(dS0 , T − 1). Consequently,
C1u − C1d C(uS0 , T − 1) − C(dS0 , T − 1)

φ00 = C0 − φ10 S0 , φ10 = = .
S1 − S1
u d S0 (u − d)
4.3 Exotic Options

So far, we considered contingent claims of the type X = h(ST ) where h : R → R
is the payoff function. In that case, the value of the claim X at its maturity
date T depends on the terminal stock price ST , but not on the values of stock
prices at times strictly before T . Contingent claims of this form are said to be
path-independent. However, several traded derivatives correspond to the so-
called path-dependent contingent claims, meaning that they are of the form
X = h(S0 , S1 , . . . , ST ) for some function h : RT +1 → R. We provide below some
examples of path-dependent claims:
1. Asian option: An Asian option is simply a European call (or put) op-
tion on the average stock price during the option’s lifetime. The following
payoffs are examples of traded Asian options.
(a) Arithmetic average option:
( )+
1 ∑
T
X= St − K . (4.7)
T + 1 t=0
(b) Geometric average option:

(( ∏
T ) T +1
1 )+
X= St −K . (4.8)
t=0
2. Barrier options: Barrier options are options that are activated or deac-
tivated if the stock price hits a certain barrier during the option’s lifetime.
As an example, we consider two examples of the the so-called “knock-out”
barrier options.
(a) Down-and-out call: This option has the same payoff as a European
call with strike K, but only if the stock price always remain over the
barrier level of H < K. Otherwise, it expires worthless. Hence
X = (ST − K)+ 1{min 0≤t≤T St >H} . (4.9)
(b) Down-and-in call: This option has the same payoff as a European
call with strike K, but only if the stock price once attains a price below
the barrier level of H < K. Otherwise, it expires worthless. Thus
X = (ST − K)+ 1{min 0≤t≤T St ≤H} . (4.10)
3. Lookback options: A lookback option is an option on a maximum or

minimum of a stock price during the option’s lifetime.
(a) Call option on a maximum:
( )+
X = max St − K .
0≤t≤T
(b) Call option on a minimum:

( )+
X = min St − K .
0≤t≤T
The crucial difference between the options presented here and the standard
(that is, plain vanilla) European call or put is that their payoffs depend not only
on the terminal stock price ST , but on the whole path taken by the stock price
over the time interval [0, T ]. Pricing these options, however, works in the same
way as for the European call. It suffices to compute the discounted expectation
under the risk-neutral probability measure described by Proposition 3.3.1.
4.4 American Contingent Claims in the CRR Model

In contrast to a contingent claim of a European style, a contingent claim of an
American style can by exercised by its holder at any date before its expiration
date T . We denote by T the class of all stopping times defined on the filtered
probability space (Ω, F, P), where Ft = FtS for every t = 0, 1, . . . , T .
Definition 4.4.1. A stopping time with respect to a filtration F is a map τ :
Ω → {0, 1, . . . , T } such that for any t = 0, 1, . . . , T the event {ω ∈ Ω | τ (ω) = t}
belongs to the σ-field Ft .
Intuitively, this property means that the decision whether to stop a given pro-
cess at time t (for instance, whether to exercise an option at time t or not)
depends on the stock price fluctuations up to time t only.
Definition 4.4.2. Let T [t,T ] be the subclass of stopping times τ with respect to F
satisfying the inequalities t ≤ τ ≤ T.
72 MATH3075/3975
4.4.1 American Call Option

Let us first consider the case of the American call option – that is, the option
to buy a specified number of shares, which may be exercised at any time before
the option expiry date T, or on that date. The exercise policy of the option
holder is necessarily based on the information accumulated to date and not on
the future prices of the stock. We denote by Cta the arbitrage price at time t of
an American call option written on one share of the stock S.
Definition 4.4.3. By an arbitrage price of the American call we mean a price
process Cta , t ≤ T, such that the extended financial market model – that is, a
market with trading in riskless bonds, stocks and an American call option –
remains arbitrage-free.
Lemma 4.4.1. The price of an American call option in the CRR arbitrage-free
market model with r ≥ 0 coincides with the arbitrage price of a European call
option with the same expiry date and strike price.
Proof. It is sufficient to show that the American call option should never be
exercised before maturity, since otherwise the issuer of the option would be
able to make riskless profit.
The argument hinges on the following simple inequality
Ct ≥ (St − K)+ , ∀ t ≤ T. (4.11)
which can be justified in several ways. An intuitive way of deriving (4.11) is

based on no-arbitrage arguments. Notice that since the option’s price Ct is al-
ways non-negative, it is sufficient to consider the case when the current stock
price is greater than the exercise price – that is, when St − K > 0.
Suppose, on the contrary, that Ct < St − K for some t, i.e., St − Ct > K. Then
it would be possible, with zero net initial investment, to buy at time t a call
option, short a stock, and invest the sum St − Ct in the savings account. By
holding this portfolio unchanged up to the maturity date T, we would be able
to lock in a riskless profit. Indeed, the value of our portfolio at time T would
satisfy (recall that r ≥ 0)
CT − ST + (1 + r)T −t (St − Ct ) > (ST − K)+ − ST + (1 + r)T −t K ≥ 0.
We conclude that inequality (4.11) is necessary for the absence of arbitrage op-
portunities.
Taking (4.11) for granted, we may now deduce the property Cta = Ct using
simple no-arbitrage arguments. Suppose, on the contrary, that the issuer of
an American call is able to sell the option at time 0 at the price C0a > C0 (it is
evident that, at any time, an American option is worth at least as much as a
European option with the same contractual features; in particular, C0a ≥ C0 ).
In order to profit from this transaction, the option writer establishes a dynamic
portfolio that replicates the value process of the European call, and invests the
remaining funds in the savings account. Suppose that the holder of the option
decides to exercise it at instant t before the expiry date T. Then the issuer of
the option locks in a riskless profit, since the value of portfolio satisfies
Ct − (St − K)+ + (1 + r)t (C0a − C0 ) > 0, ∀ t ≤ T.
The above reasoning implies that the European and American call options are
equivalent from the point of view of arbitrage pricing theory; that is, both op-
tions have the same price and an American call should never be exercised by
its holder before expiry.
4.4.2 American Put Option

An American put is an option to sell a specified number of shares, which may
be exercised at any time before or at the expiry date T . Let us consider an
American put with strike K and expiry date T . We then have the following
valuation result.
Proposition 4.4.1. The arbitrage price Pta of an American put option equals
( )
Pta = max τ ∈T [t,T ] EPe (1 + r)−(τ −t) (K − Sτ )+ | Ft , ∀ t ≤ T. (4.12)
For any t ≤ T , the stopping time τt∗ which realizes the maximum in (4.12) is
given by the expression
τt∗ = min {u ≥ t | Pua = (K − Su )+ }. (4.13)
The stopping time τt∗ will be referred to as the rational exercise time of an
American put option that is assumed to be still alive at time t. It should be
pointed out that τt∗ does not solve the optimal stopping problem for any indi-
vidual, but only for those investors who are risk-neutral.
An application of the classic Bellman principle reduces the optimal stopping

problem (4.12) to an explicit recursive procedure for the value process. We
state the following corollary to Proposition 4.4.1, in which the dynamic pro-
gramming recursion for the value of an American put option is given.
Corollary 4.4.1. Let the non-negative adapted process Ut , t ≤ T, be defined
recursively by setting UT = (K − ST )+ and
{ }
Ut = max (K − St )+ , (1 + r)−1 EPe (Ut+1 | Ft ) , ∀ t ≤ T − 1. (4.14)
Then the arbitrage price Pta of the American put option at time t equals Ut and
the rational exercise time after time t admits the following representation
τt∗ = min {u ≥ t | Uu = (K − Su )+ }. (4.15)
Therefore, τT∗ = T and for every t = 0, 1, . . . , T − 1
∗
τt∗ = t1{Ut =(K−St )+ } + τt+1 1{Ut >(K−St )+ } .
74 MATH3075/3975
It is also possible to go the other way around – that is, to first show directly
that the price Pta satisfies the recursive relationship, for t ≤ T − 1,
{ ( a )}
Pta = max (K − St )+ , (1 + r)−1 EPe Pt+1 | Ft (4.16)
subject to the terminal condition PTa = (K − ST )+ , and subsequently derive the

equivalent representation (4.12). In the case of the CRR model, formula (4.16)
is the simplest way of valuing American options. The main reason for this is
that an apparently difficult valuation problem is thus reduced to the simple
single-period case.
To summarise:
• In the case of the CRR model, the arbitrage pricing of an American put
reduces to the following simple recursive recipe, for t ≤ T − 1,
{ ( au )}
Pta = max (K − St )+ , (1 + r)−1 pePt+1 + (1 − pe)Pt+1
ad
(4.17)
with the terminal condition PTa = (K − ST )+ .

• The quantities Pt+1
au ad
and Pt+1 represent the values of the American put in
the next step corresponding to the upward and downward mouvements of
the stock price starting from a given node on the CRR lattice.
4.4.3 American Contingent Claims

We assume that a contingent claim of an American style (or, briefly, an Ameri-
can claim) does not produce any payoff unless it is exercised.
Definition 4.4.4. An American contingent claim X a = (X, T [0,T ] ) expiring
at T consists of a sequence of payoffs (Xt )0≤t≤T where the random variable Xt is
Ft -measurable for t = 0, 1, . . . , T and the set T [0,T ] of admissible exercise policies.
We interpret Xt as the payoff received by the holder of the claim X a upon ex-
ercising it at time t. The set of admissible exercise policies is restricted to the
class T [0,T ] of all stopping times of the filtration FS with values in {0, 1, . . . , T }.
Let g : R × {0, 1, . . . , T } → R be an arbitrary function. We say that X a is a
path-independent American contingent claim with the payoff function g if
the equality Xt = g(St , t) holds for every t = 0, 1, . . . , T . The arbitrage valuation
of an American claim in a discrete-time model is based on a simple recursive
procedure. In particular, to price a path-independent American claim in the
CRR model, it suffices to move backwards in time along the binomial lattice.
Proposition 4.4.2. For every t ≤ T , the arbitrage price π(X a ) of an American
claim X a in the CRR model equals
( )
πt (X a ) = max τ ∈T [t,T ] EPe (1 + r)−(τ −t) Xτ | Ft .
The price process π(X a ) satisfies the following recurrence relation, for t ≤ T − 1,
{ ( )}
πt (X a ) = max Xt , EPe (1 + r)−1 πt+1 (X a ) | Ft (4.18)
with πT (X a ) = XT and the rational exercise time τt∗ equals

{ ( )}
τt∗ = min u ≥ t | Xu ≥ EPe (1 + r)−1 πu+1 (X a ) | Fu .
For a path-independent American claim X a with the payoff process Xt = g(St , t)

we obtain, for every t ≤ T − 1,
{ ( u )}
πt (X a ) = max g(St , t), (1 + r)−1 pe πt+1 (X a ) + (1 − pe) πt+1
d
(X a ) (4.19)
u
where, for a generic stock price St at time t, we denote by πt+1 (X a ) and πt+1
d
(X a )
the values of the price πt+1 (X a ) at the nodes corresponding to the upward and
downward mouvements of the stock price during the period [t, t + 1].
Let us consider a path-independent American claim:
• By a slight abuse of notation, we write Xta to denote the arbitrage price at
time t of a path-independent American claim X a .
• Then the pricing formula (4.19) becomes (see (4.17))
{ ( )}
Xta = max g(St , t), (1 + r)−1 pe Xt+1
au
+ (1 − pe) Xt+1
ad
(4.20)
with the terminal condition XTa = g(ST , T ).

• The risk-neutral valuation formula (4.20) is valid for an arbitrary path-
independent American claim with a payoff function g in the CRR binomial
model.
Example 4.4.1. We consider here the CRR binomial model with the horizon
date T = 2, the risk-free rate r = 0.2, and the following values of the stock price
S at times t = 0 and t = 1:
S0 = 10, S1u = 13.2, S1d = 10.8.
Let X a be the American call option with maturity date T = 2 and the following
reward process g(St , t) = (St − Kt )+ , where the variable strike Kt satisfies K0 =
9, K1 = 9.9 and K2 = 12. We will first compute the arbitrage price πt (X a ) of this
option at times t = 0, 1, 2 and the rational exercise time τ0∗ . Subsequently, we
will compute the replicating strategy for X a up to the rational exercise time τ0∗ .
We start by noting that the unique risk-neutral probability measure P e satisfies
1+r−d (1 + r)S0 − S1d 12 − 10.8 1.2 1

pe = = = = = .
u−d S1 − S1
u d 13.2 − 10.8 2.4 2
76 MATH3075/3975
e are thus given by (note that S ud = S du )

The dynamics of the stock price under P 2 2
S2uu6 = 17.429 ω1
0.5nnnn
nn
nnn
nnn
S1u =
A
13.2
P
PPP
PP0.5
PPP
PPP
'
0.5
S2ud = 14.256 ω2

S0 = ;10
; ;;
;;
;;
;;0.5
;; S2du7 = 14.256 ω3
;; oo
;; 0.5oooo
;; ooo
ooo
S1d = 10.8O
OOO
OO0.5
OOO
OOO
'
S2dd = 11.664 ω4
Consequently, the price process πt (X a ) of the American option X a is given by
(17.429 − 12)+ = 5.429

hhh4
hhhhhhh
hhhh
hhhh
max (3.3, 3.202) = 3.3
q8 VVVV
qqqqq VVVV
VVVV
q VVVV
qqqqq VV*
qq (14.256 − 12)+ = 2.256
qqqqq
qq
qqqqq
q
max (1, 1.7667) = 1.7667
MMM
MMM
MMM
MMM
MMM
MMM (14.256 − 12)+ = 2.256
MMM hhh4
MMM hhhhhhh
MMM hhhh
& hhhh
max (0.9, 0.94) = 0.94
VVVV
VVVV
VVVV
VVVV
VV*
(11.664 − 12)+ = 0
Holder. The rational holder should exercise the American option at time t = 1
whenever he observes that the stock price has risen during the first period.
Otherwise, he should not exercise the option till time 2. Hence the rational
exercise time τ0∗ is a stopping time τ0∗ : Ω → {0, 1, 2} given by
τ0∗ (ω) = 1 for ω ∈ {ω1 , ω2 },
τ0∗ (ω) = 2 for ω ∈ {ω3 , ω4 }.
Issuer. We now take the position of the issuer of the option:
• At t = 0, we need to solve
1.2 φ00 + 13.2 φ10 = 3.3,
1.2 φ00 + 10.8 φ10 = 0.94.
Hence (φ00 , φ10 ) = (−8.067, 0.983) for all ωs.
• If the stock price has risen during the first period, the option is exercised
by its holder. Hence we do not need to compute the strategy at time 1 for
ω ∈ {ω1 , ω2 }.
• If the stock price has fallen during the first period, we need to solve
e01 + 14.256 φ11 = 2.256,
1.2 φ
e01 + 11.664 φ11 = 0.
1.2 φ
Hence (φ e01 , φ11 ) = (−8.46, 0.8704) if the stock price has fallen during the
first period, that is, for ω ∈ {ω3 , ω4 }. Note that φ
e01 = −8.46 is the amount of
cash borrowed at time 1, rather than the number of units of the savings
account B.
We conclude that the replicating strategy φ = (φ0 , φ1 ) is defined at time 0 for
all ωs and it is defined at time 1 on the event {ω3 , ω4 } only:
4−−−
hhhhhhh
hhh
hhhh
hhhh
V1u (φ) = 3.3
7 VVVV
ooooo VVVV
VVVV
ooo VVVV
ooooo VV*
oo −−−
ooooo
ooo
ooooo
o
(φ00 , φ10 ) = (−8.067, 0.983)
NNN
NNN
NNN
NNN
NNN
NNN V du (φ) = 2.256
NNN
iii4
2
NNN iii
NNN ii
NN' iiii
iiii
e01 , φ11 ) = (−8.46, 0.8704)
(φ
UUUU
UUUU
UUUU
UUUU
*
V2dd (φ) = 0
78 MATH3075/3975
4.5 Implementation of the CRR Model

To implement the CRR model, we proceed as follows:
• We fix maturity T and we assume that the continuously compounded
interest rate r is such that B(0, T ) = e−rT . Note that T is now expressed
in years. For instance, T = 2 months means that T = 1/6. In general, for
different maturities we may write B(0, T ) = e−Y (0,T )T where the function
Y (0, T ) represents the yield curve computed from the bond market data.
• From the stock market data, we take the current stock price S0 and we
estimate the stock price volatility σ per one time unit (i.e., one year).
• Note that it was assumed so far that t = 0, 1, 2, . . . , T , meaning that ∆t = 1.
In general, the length of each period can be any positive number smaller
than 1. We set n = T /∆t and we assume that n is an integer.
• Two widely used conventions for obtaining u and d from σ, r and ∆t are:
– The Cox-Ross-Rubinstein (CRR) parametrisation:
√ 1
u = eσ ∆t
and d= .
u
– The Jarrow-Rudd (JR) parameterisation:
( 2
) √ ( 2
) √
r− σ2 ∆t + σ ∆t r− σ2 ∆t − σ ∆t
u=e and d=e .
We first examine the Cox-Ross-Rubinstein parametrisation.
Proposition√
4.5.1. Assume that Bk∆t = (1 + r∆t)k for every k = 0, 1, . . . , n and
u = d1 = eσ ∆t in the CRR model. Then the risk-neutral probability measure P e
satisfies
1 r − σ2 √ √
2
e
P (St+∆t = St u |St ) = + ∆t + o( ∆t)
2 2σ
provided that ∆t is sufficiently small.
Proof. The risk-neutral probability measure for the CRR model is given by
e (St+∆t = St u |St ) = 1 + r∆t − d .

pe = P
u−d
Under the CRR parametrisation, we obtain
√
1 + r∆t − d 1 + r∆t − e−σ ∆t
pe = = √ √ .
u−d eσ ∆t − e−σ ∆t
The Taylor expansions up to the second order term are
√ √ σ2
eσ ∆t
= 1 + σ ∆t + ∆t + o(∆t),
2
√ √ σ2
e−σ ∆t = 1 − σ ∆t + ∆t + o(∆t).
2
By substituting the Taylor expansions into the risk-neutral probability mea-

sure, we obtain
( √ 2
)
1 + r∆t − 1 − σ ∆t + σ2 ∆t + o(∆t)
pe = ( √ ) ( √ )
1 + σ ∆t + σ2 ∆t − 1 − σ ∆t + σ2 ∆t + o(∆t)
2 2
√ ( 2
)
σ ∆t + r − σ2 ∆t + o(∆t)
= √
2σ ∆t + o(∆t)
1 r − σ2 √ √
2
= + ∆t + o( ∆t)
2 2σ
as was required to show.
To summarise:
• For ∆t sufficiently small, we get
√
1 r − σ2 √
2
1 + r∆t − e−σ ∆t
pe = √ √ ≈ + ∆t.
eσ ∆t − e−σ ∆t 2 2σ
• Note that 1 + r∆t ≈ er∆t when ∆t is sufficiently small.

• Hence we may also represent the risk-neutral probability measure as fol-
lows √
er∆t − e−σ ∆t
pe ≈ √ √ .
eσ ∆t − e−σ ∆t
Let us finally note that if we define rb such that (1 + rb)n = erT for a fixed T and
n = T /∆t then rb ≈ r∆t since ln (1 + rb) = r∆t and ln (1 + rb) ≈ rb when rb is close to
zero.
Example 4.5.1. We consider here the Cox-Ross-Rubinstein parameterisation.
• Let the annualized variance of logarithmic returns be σ 2 = 0.1.
• The interest rate is set to r = 0.1 per annum.
• Suppose that the current stock price is S0 = 50.
• We examine European and American put options with strike price K = 53
and maturity T = 4 months (that is, T = 13 ).
• The length of each period is chosen to be ∆t = 1
12
, that is, one month.
• Hence n = T
∆t
= 4 time steps.
• We adopt the CRR parameterisation to derive the stock price.
• Then u = 1.0956 and d = 1/u = 0.9128.
• We compute 1 + r∆t = 1.00833 ≈ er∆t and pe = 0.5228.
80 MATH3075/3975
Stock Price Process
75
72.0364
70
65 65.7516
60 60.0152 60.0152
Stock Price
55 54.7792 54.7792
50 50 50 50
45 45.6378 45.6378
41.6561 41.6561
40
38.0219
35 34.7047
30
25
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
t/∆ t
Figure 4.1: Stock Price Process
Price Process of the European Put Option
75
0
70
65 0
60 0.67199 0
Stock Price
55 2.2478 1.4198
50 4.4957 4.0132 3
45 7.0366 6.9242
10.4714 11.3439
40
14.5401
35 18.2953
30
25
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
t/∆ t
Figure 4.2: European Put Option Price

Price Process of the American Put Option
75
0
70
65 0
60 0.67199 0
Stock Price
55 2.3459 1.4198
50 4.7928 4.2205 3
45 7.557 7.3622
11.3439 11.3439
40
14.9781
35 18.2953
30
25
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
t/∆ t
Figure 4.3: American Put Option Price
1: Exercise the American Put; 0: Hold the American Option

80
1
70
1
60 0 1
Stock Price
0 0
50 0 0 1
0 1
1 1
40
1
1
30
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

t/∆ t
Figure 4.4: Rational Exercise Policy

82 MATH3075/3975
The next result deals with the Jarrow-Rudd parametrisation.

Proposition 4.5.2. Let Bk∆t = (1 + r∆t)k for k = 0, 1, . . . , n. We assume that
( 2
) √
r− σ2 ∆t + σ ∆t
u=e
and ( 2
) √
r− σ2 ∆t − σ ∆t
d=e .
e satisfies
Then the risk-neutral probability measure P
e (St+∆t = St u |St ) = 1 + o(∆t)

P
2
provided that ∆t is sufficiently small.
Proof. Under the JR parametrisation, we have
( 2
) √
r− σ2 ∆t−σ ∆t
1 + r∆t − d 1 + r∆t − e
pe = = ( ) √ ( ) √ .
u−d 2
r− σ2 ∆t+σ ∆t
2
r− σ2 ∆t−σ ∆t
e −e
The Taylor expansions up to the second order term are
( )
2 √
r− σ2 ∆t+σ ∆t
√
e = 1 + r∆t + σ ∆t + o(∆t)
( )
2 √
r− σ2 ∆t−σ ∆t
√
e = 1 + r∆t − σ ∆t + o(∆t)
and thus
1
pe = + o(∆t)
2
as was required to show.
Example 4.5.2. We now assume the Jarrow-Rudd parameterisation.
• We consider the same problem as in Example 7.2, but with parameters u
and d computed using the JR parameterisation. We obtain u = 1.1002 and
d = 0.9166 (note that u ̸= 1/d).
• As before, 1 + r∆t = 1.00833 ≈ er∆t , but pe = 0.5.
• We compute the price processes for the stock, the European put option,
the American put option and we find the rational exercise time.
• When we compare with Example 7.2, we see that the results are slightly
different than before, although it appears that the rational exercise policy
is the same.
• The CRR and JR parameterisations are both set to approach the Black-
Scholes model.
• For ∆t sufficiently small, the prices computed under the two parametrisa-
tions will be very close to one another.
Stock Price Process

80
73.2471
70
66.5787
60 60.5174 61.0238
Stock Price
55.0079 55.4682
50 50 50.4184 50.8403
45.8283 46.2118
42.0047 42.3562
40
38.5001
35.2879
30
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

t/∆ t
Figure 4.5: Stock Price Process
Price Process of the European Put Option

80
0
70
0
60 0.53103 0
Stock Price
2.0877 1.0709
50 4.4284 3.6792 2.1597
6.8428 6.3488
10.1204 10.6438
40
14.0607
17.7121
30
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

t/∆ t
Figure 4.6: European Put Option Price

84 MATH3075/3975
Price Process of the American Put Option

80
0
70
0
60 0.53103 0
Stock Price
2.1958 1.0709
50 4.7506 3.8971 2.1597
7.3846 6.7882
10.9953 10.6438
40
14.4999
17.7121
30
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

t/∆ t
Figure 4.7: American Put Option Price
1: Exercise the American Put; 0: Hold the American Option

80
1
70
1
60 0 1
Stock Price
0 0
50 0 0 1
0 1
1 1
40
1
1
30
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

t/∆ t
Figure 4.8: Rational Exercise Policy

4.6 Game Contingent Claims (MATH3975)

The concept of a game contingent claim is an extension of an American contin-
gent claim to a situation where both parties to a contract may exercise it prior
to its maturity date. The payoff of a game contingent claim depends, in general,
not only on the moment when it is exercised, but also on which party takes the
decision to exercise. In order to make a clear distinction between the exercise
policies of the two parties, in what follows, we will refer to the decision of the
seller (also termed the issuer) as the cancellation policy, whereas the decision
of the buyer (also referred to as the holder) is called the exercise policy.
Definition 4.6.1. A game contingent claim X g = (L, H, T e , T c ) expiring at time

T consists of F-adapted payoff processes L and H, a set T e of admissible exercise
times, and a set T c of admissible cancellation times.
It is assumed throughout that L ≤ H, meaning that the Ft -measurable random

variables Lt and Ht satisfy the inequality Lt ≤ Ht for every t = 0, 1, . . . , T . Un-
less explicitly otherwise stated, the sets of admissible exercise and cancellation
times are restricted to the class T [0,T ] of all stopping times of the filtration FS
with values in {0, 1, . . . , T }, that is, it is postulated that T e = T c = T [0,T ] .
We interpret Lt as the payoff received by the holder upon exercising a game

contingent claim at time t. The random variable Ht represents the payoff re-
ceived by the holder if a claim is cancelled (i.e., exercised by its issuer) at time
t. More formally, if a game contingent claim is exercised by either of the two
parties at some date t ≤ T , that is, on the event {τ = t} ∪ {σ = t}, the random
payoff equals
Xt = 1{τ =t≤σ} Lt + 1{σ=t<τ } Ht , (4.21)
where τ and σ are the exercise and cancellation times, respectively. Note that
both the holder and the issuer may choose freely their exercise and cancellation
times. If they decide to exercise their right at the same moment, we adopt the
convention that a game contingent claim is exercised, rather than cancelled, so
that the payoff is given by the process L. This feature is already reflected in
the payoff formula (4.21).
If the class of all admissible cancellation times is assumed to be given as T c =

{T } then a game contingent claim X g becomes an American claim X a with the
payoff process X = L. Therefore, the definition of a game contingent claim
covers as a special case the notion of an American contingent claim. If we
postulate, in addition, that T e = {T } then a game contingent claim reduces to
a European claim X = LT maturing at time T .
From the previous section, we know that valuation and hedging of American
claims is related to optimal stopping problems. Game contingent claims are as-
sociated with the so-called Dynkin games, which in turn can be seen as natural
extensions of optimal stopping problems.
86 MATH3075/3975
4.6.1 Dynkin Games

Before analyzing in some detail the game contingent claims, we first present a
brief survey of basic results concerning Dynkin games.
We assume that t = 0, 1, . . . , T and we investigate the Dynkin game (also known

as the optimal stopping game) associated with the payoff
Z(σ, τ ) = 1{τ ≤σ} Lτ + 1{σ<τ } Hσ ,
where L ≤ H are F-adapted stochastic processes defined on a finite probability

space (Ω, F, P) endowed with a filtration F.
Definition 4.6.2. For any fixed date t = 0, 1, . . . , T , by the Dynkin game
started at time t and associated with the payoff Z(σ, τ ), we mean a stochas-
tic game in which the min-player, who controls a stopping time σ ∈ T[t,T ] ,
wishes to minimize the conditional expectation
EP (Z(σ, τ ) | Ft ), (4.22)
while the max-player, who controls a stopping time τ ∈ T[t,T ] , wishes to maxi-
mize the conditional expectation (4.22).
Let us fix t. Since the stopping times σ and τ are assumed to belong to the class
T[t,T ] , formula (4.21) yields
(∑
T
( ) )
EP (Z(σ, τ ) | Ft ) = EP 1{τ =u≤σ} Lu + 1{σ=u<τ } Hu Ft .
u=t
We are interested in finding the value process of a Dynkin game and the corre-
sponding optimal stopping times. We start by stating the following definition
of the upper and lower value processes.
Definition 4.6.3. The F-adapted process Ȳ u given by the formula
Ȳtu = min max EP (Z(σ, τ ) | Ft )

σ∈T[t,T ] τ ∈T[t,T ]
is called the upper value process. The lower value process Ȳ l is an F-

adapted process given by the formula
Ȳtl = max min EP (Z(σ, τ ) | Ft ).

τ ∈T[t,T ] σ∈T[t,T ]
Lemma 4.6.1. Let L0 (Ω, F, P) stand for the class of all random variables defined
on a probability space (Ω, F, P). Let A and B be two finite sets and let g : A×B →
L0 (Ω, F, P) be an arbitrary map. Then
min max g(a, b) ≥ max min g(a, b).

a∈A b∈B b∈B a∈A
It follows immediately from Definition 4.6.3 and Lemma 4.6.1 that the upper
value process Ȳ u always dominates the lower value process Ȳ l , that is, the
inequality Ȳtu ≥ Ȳtl holds for every t = 0, 1, . . . , T .
Definition 4.6.4. If the equality Ȳ u = Ȳ l is satisfied, we say that the Stackel-

berg equilibrium holds for a Dynkin game. Then the process Ȳ = Ȳ u = Ȳ l is
called the value process.
Definition 4.6.5. We say that the Nash equilibrium holds for a Dynkin game
if for any t there exist stopping times σt∗ , τt∗ ∈ T[t,T ] , such that the inequalities
( ) ( )
EP Z(σt∗ , τ ) | Ft ≤ EP (Z(σt∗ , τt∗ ) | Ft ) ≤ EP Z(σ, τt∗ ) | Ft (4.23)
are satisfied for arbitrary stopping times τ, σ ∈ T[t,T ] , that is, the pair (σt∗ , τt∗ ) is
a saddle point of a Dynkin game.
The next result shows that the existence of a Nash equilibrium for a Dynkin
game implies the Stackelberg equilibrium.
Lemma 4.6.2. Assume that a Nash equilibrium for a Dynkin game exists. Then
the Stackelberg equilibrium holds and
Ȳt = EP (Z(σt∗ , τt∗ ) | Ft ),
so that σt∗ and τt∗ are optimal stopping times as of time t.
Proof. From (4.23), we obtain

( ) ( )
max EP Z(σt∗ , τ ) | Ft ≤ EP (Z(σt∗ , τt∗ ) | Ft ) ≤ min EP Z(σ, τt∗ ) | Ft .
Consequently,
Ȳtu = min max EP (Z(σ, τ ) | Ft ) ≤ EP (Z(σt∗ , τt∗ ) | Ft )

≤ max min EP (Z(σ, τ ) | Ft ) = Ȳtl .

Since the inequality Ȳtu ≥ Ȳtl is known to be always satisfied, we conclude that
the value process Ȳ is well defined and satisfies Ȳt = EP (Z(σt∗ , τt∗ ) | Ft ) for all
t = 0, 1, . . . , T .
The following definition introduces a plausible candidate for the value process
of a Dynkin game.
Definition 4.6.6. The process Y is defined by setting YT = LT and, for any

t = 0, 1, . . . , T − 1,
{ { }}
Yt = min Ht , max Lt , EP (Yt+1 | Ft ) . (4.24)
88 MATH3075/3975
The assumption that L ≤ H ensures that, for any t = 0, 1, . . . , T − 1,

{ { }} { { }}
Yt = min Ht , max Lt , EP (Yt+1 | Ft ) = max Lt , min Ht , EP (Yt+1 | Ft ) .
It is also clear from (4.24) that Lt ≤ Yt ≤ Ht for t = 0, 1, . . . , T . In particular, if

the equality Lt = Ht holds then necessarily Yt = Lt = Ht .
Theorem 4.6.1. (i) Let the stopping times σt∗ , τt∗ be given by
{ }
σt∗ = min u ∈ {t, t + 1, . . . , T } | Yu = Hu
and { }
τt∗ = min u ∈ {t, t + 1, . . . , T } | Yu = Lu ∧ T.
Then we have, for arbitrary stopping times τ, σ ∈ T[t,T ] ,
( ) ( )
EP Z(σt∗ , τ ) | Ft ≤ Yt ≤ EP Z(σ, τt∗ ) | Ft
and thus also
( ) ( ) ( )
EP Z(σt∗ , τ ) | Ft ≤ EP Z(σt∗ , τt∗ ) | Ft ≤ EP Z(σ, τt∗ ) | Ft
so that the Nash equilibrium holds.
(ii) The process Y is the value process of a Dynkin game, that is, for every t =
0, 1, . . . , T ,
( ) ( )
Yt = min max EP Z(σ, τ ) | Ft = EP Z(σt∗ , τt∗ ) | Ft
( )
= max min EP Z(σ, τ ) | Ft = Ȳt
and thus the stopping times σt∗ and τt∗ are optimal as of time t.
4.6.2 Arbitrage Pricing of Game Contingent Claims

We place ourselves within the framework of the CRR model with the unique
e We define the discounted payoffs L
martingale measure denoted by P. b = B −1 L
b = B −1 H of a game contingent claim X g and we consider the Dynkin
and H
e with the payoff given by the expression
game under P
b τ ) = 1{τ ≤σ} L
Z(σ, bτ + 1{σ<τ } H
bσ .
In view of the financial interpretation, the max-player and the min-player will
now be referred to as the seller (issuer) and the buyer (holder), respectively.
b , where U
Let us write U = B U b is the value process of the Dynkin game, that is,
bt = min max Ee (Z(σ,

U b τ ) | Ft ).
b τ ) | Ft ) = max min Ee (Z(σ,
P
σ∈T[t,T ] τ ∈T[t,T ] P τ ∈T[t,T ] σ∈T[t,T ]
In view of Theorem 4.6.1, this also means that for t = 0, 1, . . . , T − 1

{ { ( )}}
b b b b
Ut = min Ht , max Lt , EPe Ut+1 | Ft
or, equivalently,
{ { ( −1 )}}
Ut = min Ht , max Lt , Bt EPe Bt+1 Ut+1 | Ft
with the terminal condition UT = LT . By convention, we will refer to U (rather

than Ub ) as the value process of the Dynkin game associated with a game con-
tingent claim X g . For a fixed t = 0, 1, . . . , T , we define the following stopping
times { }
σt∗ = min u ∈ {t, t + 1, . . . , T } | Uu = Hu
and { }
τt∗ = min u ∈ {t, t + 1, . . . , T } | Uu = Lu ∧ T.
One can show that they are rational exercise times for the holder and issuer,
respectively. Proofs of Propositions 4.6.1 and 4.6.2 are omitted.
Proposition 4.6.1. The seller’s price πts (X g ) at time t of a game contingent claim
∗ ∗
is equal to the seller’s price π s (X a,σt ) of an American claim X a,σt with the payoff
∗ ∗
process X σt given by the formula Xtσ = Z(σt∗ , t) for every t = 0, 1, . . . , T . Conse-
quently, πts (X g ) = Ut for every t = 0, 1, . . . , T , and thus πts (X g ) is the solution to
the problem
πts (X g ) = b ∗ , τ ) | Ft ) = max min Bt Ee (Z(σ,

max Bt EPe (Z(σ b τ ) | Ft ).
τ ∈T[t,T ]
t P
This also means that

( )
b t∗ , τt∗ ) | Ft .
πts (X g ) = Bt EPe Z(σ
Finally, we examine the buyer’s price of a game contingent claim.

Proposition 4.6.2. The buyer’s price π b (X g ) of a game contingent claim X g is
equal to the seller’s price π s (X g ), that is, πtb (X g ) = Ut for every t = 0, 1, . . . , T .
Hence the arbitrage price π(X g ) of a game contingent claim is unique and it is
equal to the value process U of the associated Dynkin game. This means that it
is given by the recursive formula
{ { ( −1 )}}
πt (X g ) = min Ht , max Lt , Bt EPe Bt+1 πt+1 (X g ) | Ft (4.25)
with πT (X g ) = LT .
Under the assumption that the payoffs Ht = h(St , t) and Lt = ℓ(St , t) are given
in terms of the current value of the stock price at time t, it is convenient to
denote Xtg = πt (X g ) and to rewrite the last formula as follows
{ { ( gu )}}
Xtg = min h(St , t), max ℓ(St , t), (1 + r)−1 peXt+1 + (1 − pe)Xt+1
gd
(4.26)
with the terminal condition XTg = πT (X g ) = ℓ(ST , T ).

Chapter 5
The Black-Scholes Model
5.1 The Wiener Process and its Properties

We will now introduce an important example of a continuous-time Markov pro-
cess, called the Wiener process in honour of American mathematician Nobert
Wiener (1894–1964). It is also known as the Brownian motion, after Scot-
tish botanist Robert Brown (1773–1858). The Wiener process can be seen as
a continuous-time counterpart of the Bernoulli counting process, which was
introduced in Section 3.3.
Definition 5.1.1. A stochastic process (Wt ) with time parameter t ∈ R+ is called
the Wiener process (or the Brownian motion) if:
1. W0 = 0,
2. the sample paths of the process W , that is, the maps t → Wt (ω) are continuous
functions,
3. the process W has the Gaussian (i.e. normal) distribution with the expected
value EP (Wt ) = 0 for all t ≥ 0 and the covariance
Cov (Ws , Wt ) = min (s, t), s, t ≥ 0.
As opposed to a Markov chain, that is, a Markov process taking values in a
countable state space, the state space of the Wiener process is the (uncount-
able) set of all real numbers. This important property can be easily deduced
from Definition 5.1.1 since, by assumption, the distribution of the Wiener pro-
cess at any date t is Gaussian (i.e., normal). We take for granted without proof
the following important result, first established by Wiener (1923).
Theorem 5.1.1. The Wiener process exists, that is, there exists a probability
space (Ω, F, P) and a process W defined on this space, such that conditions (1)-
(3) of Definition 5.1.1 are met.
Let N (µ, σ 2 ) denote the Gaussian distribution with the expected value µ and
the variance σ 2 . From Definition
√ 5.1.1, it follows that for every fixed t > 0 we
have Wt ∼ N (0, t), so that ( t)−1 Wt ∼ N (0, 1) for every t > 0. More explicitly,
the random variable Wt has the probability density function p(x, t) given by
1
e−x /2t .
2
p(t, x) = √
2πt
90
This means that, for any real numbers a ≤ b,

∫ b ∫ √b ( ) ( )
1 −x2 /2t
t 1 −x2 /2 b a
P(Wt ∈ [a, b]) = √ e dx = √ e dx = N √ − N √ ,
a 2πt √a
t
2π t t
where N is the standard Gaussian (i.e. normal) cumulative distribution func-

tion, that is, for every z ∈ R,
∫ z ∫ z
1 −x2 /2
N (z) = √ e dx = n(x) dx
−∞ 2π −∞
where n is the standard Gaussian probability density function. Since the Wiener
process W is Gaussian, the random vector
( )
Ws
, s, t > 0,
Wt
has the two-dimensional Gaussian distribution N (0, C), where 0 denotes the
null vector of expected values and C is the covariance matrix
( )
s min(s, t)
C= .
min(s, t) t
Consequently, the increment Wt −Ws has again Gaussian distribution and man-
ifestly
EP (Wt − Ws ) = EP (Wt ) − EP (Ws ) = 0.
Moreover, we obtain, for s < t,
Var (Wt − Ws ) = EP (Wt − Ws )2 = EP (Wt2 ) − 2EP (Ws Wt ) + EP (Ws2 )
= t − 2 min(s, t) + s = t − 2s + s = t − s.
We conclude that
Wt − Ws ∼ N (0, t − s), s < t. (5.1)
It appears that the Wiener process has a common feature with the Poisson
process, specifically, both are processes of independent increments. In fact, they
belong to the class of Lévy processes, that is, the time-homogeneous processes
with independent increments (in other words, the processes with stationary
and independent increments).
Proposition 5.1.1. The Wiener process has stationary and independent incre-
ments. Specifically, for every n = 2, 3, . . . and any choice of dates 0 ≤ t1 < t2 <
· · · < tn , the random variables
Wt2 − Wt1 , Wt3 − Wt2 , . . . , Wtn − Wtn−1
are independent and they have the same probability distribution as the incre-
ments
Wt2 +h − Wt1 +h , Wt3 +h − Wt2 +h , . . . , Wtn +h − Wtn−1 +h
where h is an arbitrary non-negative number.
92 MATH3075/3975
Proof. (MATH3975) For simplicity, we only consider the increments Wt2 − Wt1
and Wt4 − Wt3 . Note first that a random vector (Wt1 , Wt2 , Wt3 , Wt4 ) is Gaussian.
Moreover,
 
( ) ( ) W t1
Wt2 − Wt1 −1 1 0 0   Wt2  ,
Y = =
Wt4 − Wt3 0 0 −1 1  Wt3 
Wt4
and thus Y is a Gaussian vector as well. We compute
Cov (Wt2 − Wt1 , Wt4 − Wt3 ) = EP (Wt2 − Wt1 ) (Wt4 − Wt3 )
= EP (Wt2 Wt4 ) − EP (Wt2 Wt3 ) − EP (Wt1 Wt4 ) + EP (Wt1 Wt3 )
= t2 − t2 − t1 + t1 = 0.
Hence, by the normal correlation theorem, the random variables Wt2 − Wt1
and Wt4 − Wt3 are independent. Note also that it follows immediately from (5.1)
that the increments of W are stationary.
5.2 Markov Property (MATH3975)

Recall that the state space for the Wiener process is the real line R. In that
case, the Markov property of W is defined through equality (5.2).
Theorem 5.2.1. The Wiener process W is a Markov process in the following
sense: for every n ≥ 1, any sequence of times 0 < t1 < . . . < tn < t and any
collection x1 , . . . , xn of real numbers, the following holds:
P ( Wt ≤ x| Wt1 = x1 , . . . , Wtn = xn ) = P ( Wt ≤ x| Wtn = xn ) , (5.2)
for all x ∈ R. Moreover,
∫ y
P ( Wt ≤ y| Ws = x) = p(t − s, z − x)dz
−∞
where ( )
1 (z − x)2
p(t − s, z − x) = √ exp −
2π(t − s) 2(t − s)
is the transition probability density function of the Wiener process.
Proof. By the independence of increments, we obtain

P ( Wt ≤ x| Wt1 = x1 , . . . , Wtn = xn ) = P ( Wt − Wtn ≤ x − xn | Wt1 = x1 , . . . , Wtn = xn )
= P (Wt − Wtn ≤ xn ) .
Similarly,
P ( Wt ≤ x| Wtn = xn ) = P ( Wt − Wtn ≤ x − xn | Wtn = xn )
= P (Wt − Wtn ≤ x − xn ) (5.3)
and the Markov property follows. The last part of the theorem is an immediate
consequence of (5.1) and (5.3).
Given the transition density function, we can compute the joint probability
distribution of the random vector (Wt1 , . . . , Wtn ) for t1 < t2 < . . . < tn , using
the transition densities only. Let us consider the case n = 2. Invoking the
independent increments property again, we obtain
P (Wt1 ≤ x1 , Wt2 ≤ x2 )
∫ x1
= P ( Wt2 ≤ x2 | Wt1 = y) p (t1 , y) dy
−∞
∫ x1
= P (Wt2 − Wt1 ≤ x2 − y) p (t1 , y) dy
−∞
∫ x1 ∫ x2 −y
= p (t2 − t1 , u) p (t1 , y) dudy
−∞ −∞
∫ x1 ∫ x2
= p (t2 − t1 , v − y) p (t1 , y) dvdy.
−∞ −∞
Finally,
∫ x1 ∫ x2
P (Wt1 ≤ x1 , Wt2 ≤ x2 ) = p (t2 − t1 , v − y) p (t1 , y) dvdy.
−∞ −∞
In the same way, one can compute P (Wt1 ≤ x1 , Wt2 ≤ x2 , Wt3 ≤ x3 ) and so on.
5.3 Martingale Property (MATH3975)

An important class of stochastic processes are continuous-time martingales.
Definition 5.3.1. We say that stochastic process (Xt ) is a continuous-time
martingale with respect to its natural filtration FX whenever
EP (Xt | Xt1 , . . . , Xtn ) = Xtn
for any n ≥ 1 and 0 ≤ t1 < t2 < . . . < tn ≤ t.
It appears that the Wiener process enjoys the martingale property.
Proposition 5.3.1. The Wiener process is a continuous-time martingale with
respect to its natural filtration FW , that is,
EP (Wt | Wt1 , . . . , Wtn ) = Wtn
for any n ≥ 1 and 0 ≤ t1 < t2 < . . . < tn ≤ t.
Proof. By the independence of increments, we have

EP (Wt | Wt1 , . . . , Wtn ) = EP ( Wt − Wtn + Wtn | Wt1 , . . . , Wtn )
= EP ( Wt − Wtn | Wt1 , . . . , Wtn ) + EP (Wtn | Wt1 , . . . , Wtn )
= EP (Wt − Wtn ) + Wtn = Wtn ,
as required.
Another example of a continuous-time martingale associated with the Wiener
process is given in Proposition 3.4.3.
94 MATH3075/3975
5.4 The Black-Scholes Call Pricing Formula

In this section, we apply the Wiener process to model the stock price S in
the financial market with continuous trading. If you wish to learn more about
continuous-time financial models and pertinent results from Stochastic Analy-
sis based on the Itō stochastic integral (see, for instance, Kuo (2006)), you may
enrol in the fourth-year course Advanced Option Pricing.
Following Black and Scholes (1973), we postulate that the stock price process S
can be described under the risk-neutral probability measure Pe by the following
stochastic differential equation
dSt = rSt dt + σSt dWt , t ≥ 0, (5.4)
with a constant initial value S0 > 0. The term σ dWt in dynamics (5.4) of the
stock price is aimed to give a reasonable description of the uncertainty in the
market. In particular, the volatility parameter σ measures the size of random
fluctuations of the stock price. It turns out that stochastic differential equation
(5.4) can be solved explicitly yielding
( ( )
1 2)
St = S0 exp σWt + r − σ t . (5.5)
2
This means, in particular, that St has the lognormal probability distribution
for every t > 0. It can also be shown that S is a Markov process. Note, however,
that S is not a process of independent increments. We assume that the inter-
est rate r is deterministic, constant and continuously compounded. Hence the
money market account is given by the formula
Bt = B0 ert , t ≥ 0,
where, by the usual convention, we set B0 = 1. It is easy to check that
dBt = rBt dt, t ≥ 0.
The next result is for the advanced level only.
Proposition 5.4.1. The discounted stock price, that is, the process Sb given by
the formula
St
Sbt = = e−rt St , t ≥ 0, (5.6)
Bt
b e that
is a continuous-time martingale with respect to the filtration FS under P,
is, for every 0 ≤ s ≤ t,
EPe (Sbt | Sbu , u ≤ s) = Sbs .
Proof. It follows immediately from (5.5) and (5.6) that
Sbt = S0 eσWt − 2 σ t = Sbs eσ(Wt −Ws )− 2 σ (t−s) .
1 2 1 2
(5.7)
Hence if we know the value of Sbt then we also know the value of Wt and vice
b
versa. This immediately implies that FS = FW . Therefore, for any integrable
random variable X the following conditional expectations coincide
Ee (X | Sbu , u ≤ s) = Ee (X | Wu , u ≤ s).
P P (5.8)
Since the Wiener process W has independent increments, using also the well-
known properties of the conditional expectation, we obtain the following chain
of equalities
( )
EPe Sbt Sbu , u ≤ s
( )
= EPe Sbs eσ(Wt −Ws − 2 σ (t−s)) Sbu , u ≤ s
1 2
(from (5.7))
( )
= Sbs e− 2 σ (t−s) EPe eσ(Wt −Ws ) Sbu , u ≤ s
1 2
(property of conditioning)
( )
= Sbs e− 2 σ (t−s) EPe eσ(Wt −Ws ) Wu , u ≤ s
1 2
(from (5.8))
( )
= Sbs e− 2 σ (t−s) EPe eσ(Wt −Ws ) .
1 2
(independence of increments)
√
Recall also that Wt − Ws = t − s Z where Z ∼ N (0, 1), and thus
( ) ( √ )
EPe Sbt Sbu , u ≤ s = Sbs e− 2 σ (t−s) EPe eσ t−sZ .
1 2
Let us finally observe that if Z ∼ N (0, 1) then for any real a

( ) 2
EPe eaZ = ea /2 .
√
By setting a = σ t − s, we finally obtain
( )
Ee Sbt Sbu , u ≤ s = Sbs e− 2 σ (t−s) e 2 σ (t−s) = Sbs ,
1 2 1 2
P
e
which shows that Sb is indeed a martingale with respect to FS under P.
b
Recall that the European call option written on the stock is a traded security,
which pays at its maturity T the random amount
CT = (ST − K)+ ,
where x+ = max (x, 0) and K > 0 is a fixed strike (or exercise price). We take
for granted that for t ≤ T the price Ct (x) of the call option when St = x is given
by the risk-neutral pricing formula
( )
Ct (x) = e−r(T −t) EPe (ST − K)+ St = x . (5.9)
In fact, this formula can be supported by the replication principle. However,
this argument requires the knowledge of the stochastic integration theory with
respect to the Brownian motion, as developed by Itō (1944, 1946). The following
call option pricing result was established in the seminal paper by Black and
Scholes (1973). Recall that ln = log e .
Theorem 5.4.1. Black-Scholes Call Pricing Formula. The arbitrage price
of the call option at time t ≤ T equals
( ) ( )
Ct (St ) = St N d+ (St , T − t) − Ke−r(T −t) N d− (St , T − t) (5.10)
where ( )
ln SKt + r ± 12 σ 2 (T − t)
d± (St , T − t) = √
σ T −t
and N is the standard Gaussian (i.e. normal) cumulative distribution function.
96 MATH3075/3975
Proof. (MATH3975) Using (5.5), we can represent the stock price ST as fol-
lows
ST = St e(r− 2 σ )(T −t)+σ(WT −Wt ) .
1 2
√
As in the proof of Proposition 5.4.1, we write WT − Wt = T − tZ where Z has
the standard Gaussian probability distribution, that is, Z ∼ N (0, 1).
Using (5.9) and the independence of increments of the Wiener process W , we

obtain, for a generic value x > 0 of the stock price St at time t,
(( )+ )
−r(T −t) (r− 21 σ 2 )(T −t)+σ(WT −Wt )
Ct (x) = e EPe St e −K St = x
( √ )+
−r(T −t) (r− 12 σ 2 )(T −t)+σ T −tZ
=e EPe x e −K
∫ ∞( √ )+
−r(T −t) (r− 12 σ 2 )(T −t)+σ T −tz
=e xe − K n(z) dz
−∞
where n is the standard Gaussian probability density function.
It is clear that the function under the integral sign is non-zero if and only if
the inequality √
x e(r− 2 σ )(T −t)+σ T −tz − K > 0
1 2
holds. This in turn is equivalent to the following inequality

( )
ln Kx − r − 12 σ 2 (T − t)
z≥ √ = −d− (x, T − t).
σ T −t
Hence, denoting d+ = d+ (x, T − t) and d− = d− (x, T − t), we obtain
∫ ∞ ( √ )
−r(T −t) (r− 12 σ 2 )(T −t)+σ T −tz
Ct (x) = e xe − K n(z) dz
−d−
∫ ∞ √ ∫ ∞
− 21 σ 2 (T −t) σ T −tz −r(T −t)
= xe e n(z) dz − Ke n(z) dz
−d− −d−
∫ ∞ √
− 21 σ 2 (T −t) 1
eσ T −tz √ e− 2 z dz − Ke−r(T −t) (1 − N (−d− ))
1 2
= xe
−d− 2π
∫ ∞ √
1 2
√ e− 2 (z−σ T −t) dz − Ke−r(T −t) N (d− )
1
=x
−d 2π
∫ ∞−
1
√ e−u /2 du − Ke−r(T −t) N (d− )
2
=x √
−d −σ T −t 2π
∫ ∞−
1
√ e−u /2 du − Ke−r(T −t) N (d− )
2
=x
−d+ 2π
= xN (d+ ) − Ke−r(T −t) N (d− ) .
Upon substitution x = St , this establishes formula (5.10).

The arbitrage price of the put option in the Black-Scholes model can now be
computed without difficulties, as shown by the following result.
Corollary 5.4.1. From the put-call parity
Ct − Pt = St − Ke−r(T −t)
we deduce that the arbitrage price at time 0 ≤ t < T of the European put option
equals ( ) ( )
Pt = K e−r(T −t) N − d− (St , T − t) − St N − d+ (St , T − t) .
Example 5.4.1. Suppose that the current stock price equals $31, the stock
price volatility is σ = 10% per annum, and the risk-free interest rate is r = 5%
per annum with continuous compounding.
Call option. Let us consider a call option on the stock S, with strike price $30
and with 3 months to expiry. We may assume, without loss of generality, that
t = 0 and T = 0.25. We obtain (approximately) d+ (S0 , T ) = 0.93, and thus
√
d− (S0 , T ) = d+ (S0 , T ) − σ T = 0.88.
The Black-Scholes call option pricing formula now yields (approximately)
C0 = 31N (0.93) − 30e−0.05/4 N (0.88) == 25.42 − 23.9 = 1.52
since N (0.93) ≈ 0.82 and N (0.88) ≈ 0.81. Let Ct = φ0t Bt + φ1t St . The hedge ratio
for the call option is known to be given by the formula φ1t = N (d+ (St , T − t)).
Hence the replicating portfolio for the call option at time t = 0 is given by
φ00 = −23.9, φ10 = N (d+ (S0 , T )) = 0.82.
This means that to hedge the short position in the call option, which was sold
at the arbitrage price C0 = $1.52, the option’s writer needs to purchase at time
0 the number δ = 0.82 shares of stock. This transaction requires an additional
borrowing of 23.9 units of cash.
Note that the elasticity at time 0 of the call option price with respect to the
stock price equals
( )−1 ( )
c ∂C C0 N d+ (S0 , T ) S0
η0 := = = 16.72.
∂S S0 C0
Suppose that the stock price rises immediately from $31 to $31.2, yielding a
return rate of 0.65% flat. Then the option price will move by approximately
16.5 cents from $1.52 to $1.685, giving a return rate of 10.86% flat. The option
has nearly 17 times the return rate of the stock; of course, this also means
that it will drop 17 times as fast. If an investor’s portfolio involves 5 long call
options (each on a round lot of 100 shares of stock), the position delta equals
500 × 0.82 = 410, so that it is the same as for a portfolio involving 410 shares of
the underlying stock.
98 MATH3075/3975
Put option. Let us now assume that an option is a put. The price of a put
option at time 0 equals
P0 = 30 e−0.05/4 N (−0.88) − 31N (−0.93) = 5.73 − 5.58 = 0.15.
The hedge ratio corresponding to a short position in the put option equals ap-
proximately −0.18 (since N (−0.93) ≈ 0.18). Therefore, to hedge the exposure
an investor needs to short 0.18 shares of stock for one put option. The proceeds
from the option and share-selling transactions, which amount to $5.73, should
be invested in risk-free bonds.
Notice that the elasticity of the put option is several times larger than the
elasticity of the call option. In particular, if the stock price rises immediately
from $31 to $31.2 then the price of the put option will drop to less than 12
cents.
5.5 The Black-Scholes PDE

We introduce here an alternative method for valuing European claims within
the framework of the Black-Scholes model. Proposition 5.5.1 can be used to
value an arbitrary path-independent contingent claim of European style.
Proposition 5.5.1. Let g : R → R be a function such that the random variable
X = g(ST ) is integrable under P. e Then the arbitrage price in the Black-Scholes
model M of the claim X which settles at time T is given by the equality πt (X) =
v(St , t), where the function v : R+ × [0, T ] → R solves the Black-Scholes PDE
∂v 1 2 2 ∂ 2 v ∂v
+ σ s + rs − rv = 0, ∀ (s, t) ∈ (0, ∞) × (0, T ),
∂t 2 ∂s2 ∂s
with the terminal condition v(s, T ) = g(s).
Proof. The derivation of the Black-Scholes PDE hinges on results from Stochas-
tic Analysis (the Feynman-Kac formula and the Itō lemma) and thus it is be-
yond the scope of this course.
It can be checked that arbitrage prices of call and put options satisfy this PDE.
If we denote by c(s, τ ) the function c : R+ × [0, T ] → R such that Ct = c(St , T − t)
then it is not hard to verify by straightforward calculations that
cs = N (d+ ) = δ > 0,
n(d+ )
css = √ = γ > 0,
sσ τ
sσ
cτ = √ n(d+ ) + Kre−rτ N (d− ) = θ > 0,
2 τ
√
cσ = s τ n(d+ ) = λ > 0,
cr = τ Ke−rτ N (d− ) = ρ > 0,
cK = −e−rτ N (d− ) < 0,
where d+ = d+ (s, τ ), d− = d− (s, τ ) and n stands for the standard Gaussian
probability density function.
5.6 Random Walk Approximations

Our final goal is to examine a judicious approximation of the Black-Scholes
model by a sequence of CRR models:
• In the first step, we will first examine an approximation of the Wiener
process by a sequence of symmetric random walks.
• In the next step, we will use this result in order to show how to approxi-
mate the Black-Scholes stock price process by a sequence of the CRR stock
price models.
• We will also recognise that the proposed approximation of the stock price
leads to the Jarrow-Rudd parametrisation of the CRR model in terms of
the short term rate r and the stock price volatility σ.
5.6.1 Approximation of the Wiener Process

We first define the symmetric random walk on integers.
Definition 5.6.1. A process Y = (Yk , k = 0, 1, . . . ) on a probability space (Ω, F, P)
is
∑kcalled the symmetric random walk starting at zero if Y0 = 0 and Yk =
i=1 Xi where the random variables X1 , X2 , . . . are independent with the fol-
lowing common probability distribution
P(Xi = 1) = 0.5 = P(Xi = −1).
Next, we consider the scaled random walk Y h .
1.5
Y(3∆ t) = 3h
1
0.5
Y(2∆ t) = 2h
0.5
0.5 Y(∆ t) = h
0.5 ...
0.5 0.5 0.5
Y(3∆ t) = h
0
...
0.5 0.5
Y(0) = 0 Y(2∆ t) = 0
0.5
−0.5 Y(∆ t) = −h
0.5
Y(3∆ t) = −h
...
0.5
Y(2∆ t) = −2h
0.5
−1
Y(3∆ t) = −3h
−1.5
Figure 5.1: Representation of the scaled random walk Y h

100 MATH3075/3975
The scaled random walk Y h represented in Figure

√ 5.1 is obtained from the
symmetric random walk Y as follows: we fix h = ∆t and we set for every
k = 0, 1, . . .
√ ∑k
√
h
Yk∆t = ∆tYk = ∆tXi .
i=1
√ h
Of course, for h = ∆t = 1 we obtain = Yk1 = Yk . The following result is an
Yk∆t
easy consequence of the classic Central Limit Theorem (CLT) for sequences of
independent and identically distributed (i.i.d.) random variables.
h
Theorem 5.6.1. Let
√ Yt for t = 0, ∆t, . . . , be a random walk starting at 0 with
increments ±h = ± ∆t. If
( h ) ( h )
P Yt+∆t = y + h Yth = y = P Yt+∆t = y − h Yth = y = 0.5
then, for any fixed t ≥ 0, the limit limh→0 Yth exists in the sense of probability
distribution. Specifically, limh→0 Yth ∼ Wt where W is the Wiener process and
the symbol ∼ denotes the equality of probability distributions. In other words,
limh→0 Yth ∼ N (0, t).
Proof. We√fix t > 0 and we set k = t/∆t. Hence if ∆t → 0 then k → ∞. We recall
that h = ∆t and
∑k
√
h
Yk∆t = ∆tXi .
i=1
Since EP (Xi ) = 0 and Var (Xi ) = EP (Xi ) =
2
1, we obtain
√ ∑
k
EP (Yk∆t
h
) = ∆t EP (Xi ) = 0,
i=1
∑
k ∑
k
h
Var (Yk∆t ) = ∆t Var (Xi ) = ∆t = k∆t = t.
i=1 i=1
We conclude using a minor extension of the CLT (see Theorem 6.3.2).

h
The
√ sequence of random walks Y approximates the Wiener process W when
h = ∆t → 0 meaning that:
• For any fixed t ≥ 0, the convergence limh→0 Yth ∼ Wt holds, where ∼ de-
notes the equality of probability distributions on R. This follows from
Theorem 5.6.1.
• For any fixed n and any dates 0 ≤ t1 < t2 < · · · < tn , we have
lim (Yth1 , . . . , Ythn ) ∼ (Wt1 , . . . , Wtn )
h→0
where ∼ denotes the equality of probability distributions on the space Rn .

• The sequence of linear versions of the random walk processes Y h converge
to a continuous time process W in the sense of the weak convergence of
stochastic processes on the space of continuous functions. This result is
usually referred to as the Donsker theorem or the invariance principle
(see Donsker (1951)).
5.6.2 Approximation of the Stock Price

Recall that the JR parameterisation for the CRR binomial model postulates
that ( 2
) √ ( 2
) √
r− σ2 ∆t + σ ∆t r− σ2 ∆t − σ ∆t
u=e and d = e ,
√
whereas under the CRR convention we set u = eσ ∆t = 1/d. We will show that
the JR choice hinges on a particular approximation of the stock price process
S. To this end, we recall that for all 0 ≤ s < t
( 2
)
r− σ2 t+σWt
St = S0 e
( 2
) ( 2
)
r− σ2 s+σWs r− σ2 (t−s)+σ(Wt −Ws ) (5.11)
= S0 e e
( 2
)
r− σ2 (t−s)+σ(Wt −Ws )
= Ss e .
Let us set t − s = ∆t and let us replace the Wiener process W by the random
walk Y h in equation (5.11). Then
√
Wt+∆t − Wt ≈ Yt+∆t
h
− Yth = ±h = ± ∆t.
Consequently, we obtain the following approximation
 ( ) √
 2
r− σ2 ∆t + σ ∆t
St e( if the price increases,
St+∆t ≈ ) √
 Se r− σ2
2
∆t − σ ∆t
if the price decreases.
t
More explicitly, for k = 0, 1, . . .

( 2
)
h r− σ2 h
k∆t + σYk∆t
Sk∆t = S0 e .
If we denote t = k∆t, then
( 2
)
r− σ2 t+σYth
Sth = S0 e .
We observe that this approximation of the stock price process leads to the
Jarrow-Rudd parameterisation
( 2
) √ ( 2
) √
r− σ2 ∆t + σ ∆t r− σ2 ∆t − σ ∆t
u=e and d=e .
We conclude by making the following comments:
• The convergence of the sequence of random walks Y h to the Wiener pro-
cess W (that is, the Donsker theorem) implies that the sequence S h of CRR
stock price models converges to the Black-Scholes stock price model S.
• The convergence of S h to the stock price process S justifies the claim that
the JR parametrisation is more suitable than the CRR method.
• This is especially important when dealing with valuation and hedging of
path-dependent and American contingent claims.
Chapter 6
Appendix: Probability Review
We provide here a brief review of basic concepts of probability theory.
6.1 Discrete and Continuous Random Variables

For any random variable X, the cumulative distribution function (cdf) of
X is defined by the equality
FX (x) = P(X ≤ x), ∀ x ∈ R.
Definition 6.1.1. A random variable X is called discrete if there exists a

countable set of real numbers x1 , x2 , . . . such that
∑
∞
pi = P(X = xi ) > 0 for i = 1, 2, . . . and pi = 1.
i=1
Example 6.1.1. Discrete random variables: binomial, geometric, Poisson vari-

ables, etc.
The mth moment of a discrete random variable X is defined by
∑
∞ ∑
∞
EP (X ) :=
m
xm
i P(X = xi ) = xm
i pi
i=1 i=1
provided that the series converges absolutely, that is, if

( ) ∑
∞ ∑
∞
EP |X|m := |xi |m P(X = xi ) = |xi |m pi < ∞.
i=1 i=1
Definition 6.1.2. A random variable X is called continuous if there exists a

function fX : R → R+ such that
∫ x
FX (x) = fX (y) dy, ∀ x ∈ R.
−∞
The function fX is called the probability density function (pdf) of X.
102
We will sometimes write F (f resp.) instead of FX (fX resp.) if no confusion

may arise. If a random variable X is continuous then
∫
P(X ∈ B) = f (x) dx
B
for every Borel set B. Recall that intervals (a, b), (a, b], [a, b], etc, are examples
of Borel sets. Since ∫ x
F (x) = f (y) dy, ∀ x ∈ R,
−∞
d
it follows that f (x) = dx
F (x).
Example 6.1.2. Continuous random variables: uniform, normal (Gaussian),
exponential, Gamma, Beta, Cauchy variables, etc.
The mth moment of a continuous random variable X is defined by
∫ ∞
EP (X ) =
m
xm fX (x) dx,
−∞
provided that the integral converges absolutely, that is, if

∫ ∞
( m)
EP |X| = |x|m fX (x) dx < ∞.
−∞
Definition 6.1.3. The expectation or mean of a random variable X is de-

noted by EP (X) and is defined by (if it exists)
∫ ∞ { ∫∞
xfX (x) dx, if X is continuous,
EP (X) = x dFX (x) = ∑ −∞
∞ (6.1)
−∞ i=1 xi P(X = xi ), if X is discrete.
Equation (6.1) also defines the expectation of any function of X say g(X). Since
Y = g(X) is itself random variable, it follows from equation (6.1) that if EP [g(X)]
exists then it equals
∫ ∞
EP [g(X)] = EP (Y ) = y dFY (y),
−∞
where FY is the cumulative distribution function of the random variable Y =

g(X). However, you can easily show that
∫ ∞
EP [g(X)] = g(x) dFX (x),
−∞
that is,
{ ∫∞
g(x)fX (x) dx, if X is continuous,
EP [g(X)] = ∑ −∞
∞ (6.2)
i=1 g(xi ) P(X = xi ), if X is discrete.
Definition 6.1.4. The variance of the random variable X is defined by (if it

exists) ( )
Var(X) = EP [X − EP (X)]2 = EP X 2 − (EP (X))2 . (6.3)
104 MATH3075/3975
6.2 Multivariate Random Variables

Definition 6.2.1. The joint distribution FX,Y of two random variables X and
Y is defined by
FX,Y (x, y) = P(X ≤ x, Y ≤ y)
for all real numbers x and y.
The marginal distributions of X and Y can be obtained in the following way
FX (x) = lim FX,Y (x, y), FY (y) = lim FX,Y (x, y).
y→∞ x→∞
Definition 6.2.2. The random variables X and Y are said to be independent

if
FX,Y (x, y) = FX (x)FY (y)

for all real x and y.
It is also not hard to show that X and Y are independent if and only if
EP (g(X)h(Y )) = EP (g(X)) EP (h(Y ))

for all functions g and h for which the expectations exists.
Definition 6.2.3. Two random variables X and Y are said to be uncorrelated
if their covariance, which is defined by
Cov(X, Y ) = EP [(X − EP (X))(Y − EP (Y ))] = EP (XY ) − EP (X) EP (Y ),
is zero.
It is easy to see that independent random variables are uncorrelated, but the
converse is not true, as the following example shows.
Example 6.2.1. Let
P(X = ±1) = 1/4, P(X = ±2) = 1/4,
and Y = X 2 . Here Cov(X, Y ) = 0 by symmetry, but there is a functional depen-

dence of Y on X.
Example 6.2.2. Let U and V have the same probability distribution and let
X = U + V and Y = U − V . Show that Cov(X, Y ) = 0. Are they independent?
We say that two random variables X and Y with finite variances are corre-
lated if the correlation coefficient if
Cov(X, Y )
ρ = ρ(X, Y ) = √ √
Var(X) Var(Y )
exists and is non-zero. It is known that −1 ≤ ρ ≤ 1. Note that X and Y are
uncorrelated whenever ρ(X, Y ) = 0.
Note that the correlation coefficient ρ(X, Y ) is not a general measure of depen-
dence between X and Y . Indeed, if X and Y are uncorrelated then they may be
either dependent or independent (although if X and Y are independent then
they are always uncorrelated). The correlation coefficient can be seen as a mea-
sure of the linear dependency of X and Y in the context of linear regression.
The random variables X and Y are jointly continuous if there exists a func-
tion fX,Y (x, y), called the joint probability density function such that
∫ b∫ d
P(X ∈ [a, b], Y ∈ [c, d]) = fX,Y (x, y) dydx
a c
for all real numbers a < b and c < d.

Definition 6.2.4. The joint distribution FX1 ,...,Xn of a finite collection of ran-
dom variables X1 , X2 , . . . , Xn is defined by
FX1 ,...,Xn (x1 , x2 , . . . , xn ) = P(X1 ≤ x1 , X2 ≤ x2 , . . . , Xn ≤ xn )
for all real numbers x1 , x2 , . . . , xn .
The marginal distribution of Xi can be obtained in the following way
FXi (xi ) = P(Xi ≤ xi ) = lim FX1 ,...,Xn (x1 , x2 , . . . , xn ).

xj →∞, j̸=i
Definition 6.2.5. We say that the n random variables X1 , X2 , . . . , Xn are inde-

pendent if for all real numbers x1 , x2 , . . . , xn the following equality holds
P(X1 ≤ x1 , X2 ≤ x2 , . . . Xn ≤ xn ) = P(X1 ≤ x1 ) P(X2 ≤ x2 ) · · · P(Xn ≤ xn )
or, equivalently,
FX1 ,...,Xn (x1 , x2 , . . . , xn ) = FX1 (x1 )FX2 (x2 ) · · · FXn (xn ).
6.3 Limit Theorems for Independent Sequences

Some of the most important results in probability are limit theorems. We recall
here two important limit theorems: Law of Large Numbers (LLN) and Central
Limit Theorem (CLT).
Theorem 6.3.1. [Law of Large Numbers] If X1 , X2 , . . . are independent and
identically distributed random variables with mean µ then, with probability
one,
X1 + · · · + Xn
→ µ as n → ∞.
n
Theorem 6.3.2. [Central Limit Theorem] If X1 , X2 , . . . are independent and
identically distributed random variables with mean µ and variance σ 2 > 0 then
for all real x
{ } ∫ x
X1 + · · · + Xn − nµ 1
√ e−u /2 du.
2
lim P √ ≤x =
n→∞ σ n −∞ 2π
106 MATH3075/3975
6.4 Conditional Distributions and Expectations

Definition 6.4.1. For two random variables X1 and X2 and an arbitrary set A
such that P(X2 ∈ A) ̸= 0, we define the conditional probability
P(X1 ∈ A1 , X2 ∈ A2 )
P(X1 ∈ A1 | X2 ∈ A2 ) =
P(X2 ∈ A2 )
and the conditional expectation
EP (X1 1X2 ∈A )
EP (X1 | X2 ∈ A) =
P(X2 ∈ A)
where 1{X2 ∈A} : Ω → {0, 1} is the indicator function of {X2 ∈ A}, that is,
{
1, if X2 (ω) ∈ A,
1{X2 ∈A} (ω) =
0, otherwise.
The following result is useful
P(X1 ∈ A, X3 ∈ C | X2 ∈ B) = P(X1 ∈ A | X2 ∈ B, X3 ∈ C) P(X3 ∈ C | X2 ∈ B).
Discrete case. Assume that X and Y are discrete random variables
∑
∞
pi = P(X = xi ) > 0, i = 1, 2, . . . and pi = 1,
i=1
∑
∞
pbj = P(Y = yj ) > 0, j = 1, 2, . . . and pbj = 1.
j=1
Then
P(X = xi , Y = yj )
pX|Y (xi | yj ) = P(X = xi | Y = yj ) =
P(Y = yj )
and
∑
∞
P(X = xi , Y = yj ) ∑
∞
EP (X | Y = yj ) = xi = xi P(X = xi | Y = yj ).
i=1
P(Y = yj ) i=1
It is easy to check that

∑
∞
P(X = xi ) = P(X = xi | Y = yj ) P(Y = yj )
j=1
and
∑
∞
EP (X) = EP (X | Y = yj ) P(Y = yj ).
j=1
Definition 6.4.2. The conditional cdf FX|Y (· | yj ) of X given Y is defined for

all yj such that P(Y = yj ) > 0 by
∑
FX|Y (x | yj ) = P(X ≤ x | Y = yj ) = pX|Y (xi | yj )
xi ≤x
Note that the conditional expectation EP (X | Y = yj ) is the mean of the condi-

tional distribution FX|Y (· | yj ).
Continuous case. Assume that X and Y have a joint pdf fX,Y (x, y).
Definition 6.4.3. The conditional pdf of Y given X is defined for all x such
that fX (x) > 0 and equals
fX,Y (x, y)
fY |X (y | x) =
fX (x)
and the conditional cdf of Y given X equals
∫ y
fX,Y (x, u)
FY |X (y | x) = P(Y ≤ y | X = x) = du.
−∞ fX (x)
Definition 6.4.4. The conditional expectation of Y given X is defined for
all x such that fX (x) > 0 by
∫ ∞ ∫ ∞
fX,Y (x, y)
EP (Y | X = x) = y dFY |X (y | x) = y dy.
−∞ −∞ fX (x)
An important property of conditional expectation is that
∫ ∞
EP (Y ) = EP (EP (Y | X)) = EP (Y | X = x)fX (x) dx.
−∞
Hence the expectation of Y can be determined by first conditioning on X (to

give EP (Y |X)) and then integrating with respect to the pdf of X.
6.5 Exponential Distribution

The following functional equation occurs often in probability theory
f (s + t) = f (s)f (t) for all s, t ≥ 0. (6.4)
It is known that the only (measurable) solution to equation (6.4) is f (t) = eαt
for some constant α. Let us recall the definition of a memoryless random
variable.
Definition 6.5.1. A non-negative random variable X is memoryless if
P(X > s + t | X > t) = P(X > s) for all s, t ≥ 0. (6.5)
We will now use equation (6.4) to prove that the exponential distribution is
the unique memoryless distribution. Suppose that X is memoryless and let
F̄ (t) = P(X > t) for t ≥ 0. From equation (6.5), we have
P(X > s + t)
= P(X > s).
P(X > t)
or F̄ (s +t) = F̄ (s)F̄ (t) which in turn implies F̄ (t) = e−λt for some constant λ > 0.
This is the tail of the distribution of an exponential random variable. It is also
clear that the exponential distribution is memoryless.
108 MATH3075/3975
Table of Distributions
probability/density domain mean variance
Bernoulli P(X = 1) = p {0, 1} p p(1 − p)

P(X = 0) = 1 − p
Uniform (discrete) n−1 {1, 2, . . . , n} 1

2
(n + 1) 1
12
(n2 − 1)
(n)
Binomial k
pk (1 − p)n−k {0, 1, . . . , n} np np(1 − p)
Geometric p(1 − p)k−1 k = 1, 2, . . . p−1 (1 − p)p−2
Poisson λk
k!
e−λ k = 0, 1, . . . λ λ
Uniform (continuous) (b − a)−1 [a, b] 1

2
(a + b) 1
12
(b − a)2
Exponential λe−λ x [0, ∞) 1

λ
1
λ2
( )
(x−µ)2
Normal (Gaussian) √ 1
2πσ 2
exp − 2σ 2
R µ σ2
N (µ, σ 2 )
( )
x2
Standard Normal √1
2π
exp − 2
R 0 1
N (0, 1)
Gamma Γ(n, λ) 1
Γ(n)
λn xn−1 e−λx [0, ∞) n
λ
n
λ2
ab(a+b)2
Beta β(a, b) Γ(a+b)
Γ(a)Γ(b)
xa−1 (1 − x)b−1 [0, 1] a
a+b a+b+1
Cauchy 1
π(1+x2 )
R
Bibliography
[1] R. E. Bellman: On the theory of dynamic programming. Proceedings of the

National Academy of Sciences 38 (1952), 716–719.
[2] F. Black and M. Scholes: The pricing of options and corporate liabilities.
Journal of Political Economy 81 (1973), 637–654.
[3] J. C. Cox, S. A. Ross and M. Rubinstein: Option pricing: a simplified ap-
proach. Journal of Financial Economics 7 (1979), 229–263.
[4] M. D. Donsker: An invariance principle for certain probability limit theo-
rems. Memoirs of the American Mathematical Society 25 (1951), 1–12.
[5] J. M. Harrison and S. R. Pliska: Martingales and stochastic integrals in the
theory of continuous trading. Stochastic Processes and their Applications 11
(1981), 215–260.
[6] R. J. Elliott and P. E. Kopp: Mathematics of Financial Markets. Springer,
1999.
[7] K. Itō: Stochastic integral. Proceedings of the Imperial Academy (Tokyo) 20
(1944), 519–524.
[8] K. Itō: On stochastic integral equations. Proceedings of the Japan Academy
22 (1946), 32–35.
[9] R. A. Jarrow and A. Rudd: Approximate option valuation for arbitrary
stochastic processes. Journal of Financial Economics 10 (1982), 347–369.
[10] M. Kac: On distributions of certain Wiener functionals. Transactions of
the American Mathematical Society 65 (1949), 1–13.
[11] Y. Kifer: Game options. Finance and Stochastics 4 (2000), 443–463.
[12] H. Kuo: Introduction to Stochastic Integration. Springer, 2006.
[13] M. Musiela and M. Rutkowski: Martingale Methods in Financial Mod-
elling. 2nd edition, Springer, 2005.
[14] R. J. Rendleman and B. J. Bartter: Two-state option pricing. Journal of
Finance 34 (1979), 1093–1110.
[15] R. T. Rockafellar: Convex Analysis. Princeton University Press, 1970.
[16] N. Wiener: Differential space. Journal of Mathematics and Physics 58
(1923), 131–174.
109

MATH3075 3975 Course Notes 2016

Uploaded by

Copyright:

Available Formats

MATH3075 3975 Course Notes 2016

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

MATH3075 3975 Course Notes 2016

Uploaded by

Copyright:

Available Formats

MATH3075/3975

Christian-Oliver Ewald and Marek Rutkowski

• Students enrolled in MATH3075 are expected to understand and learn all

2 Single-Period Market Models 11

3 Multi-Period Market Models 44

4 The Cox-Ross-Rubinstein Model 64

5 The Black-Scholes Model 90

6 Appendix: Probability Review 102

The goal of this chapter is to give a brief introduction to financial markets.

1.1 Financial Markets

Stocks (shares), options of European or American style, forwards and futures,

Organised exchanges versus OTC markets.

There are two types of financial markets: exchanges and over-the-counter

Long and short positions.

One notable feature of modern financial markets is that it is not necessary to

1.2 Trading in Securities

Fluctuations in supply and demand are caused by many factors, including:

Market equilibrium: supply = demand.

1.3 Perfect Markets

We work throughout under the following standing assumptions:

Law of One Price: If two securities have the same pattern of

Assumptions 1. to 3. describe a general framework in which we will be able

1.4 European Call and Put Options

What should be the ‘fair’ prices of European call or put options?

1.5 Interest Rates and Zero-Coupon Bonds

1.5.1 Discretely Compounded Interest

1.5.2 Continuously Compounded Interest

Single-Period Market Models

It is clear that single-period models are unrealistic, since in market practice

Single period market models can be seen as ‘atoms’ of

where ω1 , ω2 , . . . , ωk represent all possible states of the world at time t = 1.

2.1 Two-State Single-Period Market Model

We will sometimes write q := 1 − p.

2.1.1 Primary Assets

By the stock, we mean throughout a non-dividend paying common stock,

2.1.2 Wealth of a Trading Strategy

V1 (x, φ) = (x − φS0 )(1 + r) + φS1 .

2.1.3 Arbitrage Opportunities and Arbitrage-Free Model

An essential feature of an efficient market is that if a trading strategy can turn

Real markets do sometimes exhibit arbitrage, but this is necessarily temporary

A financial market model is said to be arbitrage-free whenever

If u ≤ 1 + r then the following strategy is an arbitrage:

No arbitrage ⇒ d < 1 + r < u.

The converse is also true, namely,

d < 1 + r < u ⇒ No arbitrage.

2.1.4 Contingent Claims

2.1.5 Replication Principle

• Assume that it is possible to find a trading strategy replicating a con-

Let us formalise these arguments within the present framework.

In the next proposition, we address the following question:

How to find a replicating strategy for a given claim?

Proposition 2.1.3. Let X = h(S1 ) be an arbitrary contingent claim. The pair

2.1.6 Risk-Neutral Valuation Formula

We refer to equation (2.8) as the risk-neutral valuation formula. In par-

• The probability measure P e is called a risk-neutral probability mea-

Price of a call − Price of a put = C0 − P0 = S0 − 1

for t = 0, 1, . . . , T where B(t, T ) is the price at time t of the unit zero-coupon

Hence the price P0 = π0 (P1 ) of the European put at time t = 0 is given by

2.2 General Single-Period Market Models