Yuji Yamada

Control and Dynamical Systems

California Institute of Technology

First Term
Fall 2001
 Mathematics
(Basic stochastic
calculus)

This course
Engineering Finance
(Numerical (Derivative pricing
technique) And hedging)
What are we going to learn?

• Basics of stochastic calculus and its application to finance

• Underlying theory and computational tools for stochastic

processes and simulations in derivative pricing/hedging
problems

• Topics: basic probability theory (measure theory), martingale

theory, Markov processes, and stochastic control

Lecture notes: “Steven Shreve, Stochastic Calculus and

Finance,” downloadable at
http://www.cs.cmu.edu/~chal/shreve.html
Before starting…

• From the preface of “Thomas Mikosh, Elementary

Stochastic Calculus with Finance in View, 1998”
Ten years ago I would not have dared to write a book like
this: a non-rigorous treatment of a mathematical theory. I
admit that I would have been ashamed, and I am afraid
that most of my colleagues in mathematics still think like
this. However, my experience with students and
practitioners convinced me that there is a strong demand
for popular mathematics…….
• From the preface of “Bernt Oksendal,
Stochastic Differential Equations, 1985”
There are several reasons why one should learn more about
stochastic differential equations: They have a wide range
of
applications out side mathematics…
…. Unfortunately most of the literature about stochastic
differential equations seems to place so much emphasis on
rigor and completeness that it scares many nonexperts
away. These notes are an attempt to approach the subject
from the nonexpert point of view….
• From the preface of “David Williams,
Probability with Martingales, 1990”
Preface – please read!
……….
You cannot avoid measure theory: an event in probability
is a measurable set, a random variable is a measurable
function on the sample space, the expectation of a random
variable is its integral with respect to the probability
measure….
 Course material
“Steven Shreve, Stochastic Calculus and Finance”
• T. Mikosch, Elementary Stochastic Calculus with Finance in
View, World Scientific, 1998
• B. Oksendal, Stochastic Differenctial Equations: An Introduction
with Applications, 5th ed., Springer Berlin Heidelberg, 1998
• D. Williams, Probability with Martingales, Cambridge Univ.
Press, 1991
- D. Duffie, Dynamic Asset Pricing Theory, 2nd ed., Princeton
Univ. Press, 1996
- D.T. Gllespie, Markov Process: An Intoroduction for Physical
Scientists, Academic Press, 1992
- J. Hull, Options, Futures, and Other Derivative Securities, 4th
ed., Prentice-Hall, 1999
Course outline

1. Basics of arbitrage pricing and probability theory

2. The Markov property and American options
3. Properties of continuous models
4. Numerical techniques

Theory Application
Discrete
Continuous
Course outline

Application Theory
Pricing and Hedging on Conditional Expectation
- European option Martingale theory
- American option Markov processes
- Exotic option Ito formula

Recommend “BEM 105: Options”

 Theory Application
Discrete
Continuous

• Using a discrete random walk model, we are going to

learn the basic theory of “pricing and hedging for
derivative securities”

Derivative = an instrument whose price depends on, or is

derived from, the price of another asset [Hull]
European call option
:= the right to buy an asset at a strike price K
under specified terms T

St : price of stock at time t  [0, T ]

Ct : price of stock at time t  [0, T ]

CT  max(ST  K , 0)

• What is the fair price of the call option between t  [0, T ) ?

Arbitrage pricing theory
Arbitrage = a strategy that is guaranteed to make money with
no initial cost, or no future payment.

Arbitrage pricing theory = the theory of asset pricing which

permits no arbitrage opportunity

Comparison principle: if you know for sure that two securities

will have the same price, then the initial
prices have to be the same, too.

A(1)  B (1)  A(0)  B (0)

A(1)  B(1), A(0)  B (0)  A(0)  B(0)  0, A(1)  B(1)  0

Replicating portfolio

t  t  [0, T ] : price of a risk-free asset (or a bond)

Xt  t  [0, T ] : value of a portfolio

X T   t St   t t

• If there exists a self-financing trading strategy s.t.

X T  CT
 C0  X 0
In fact,
Ct  X t , t  [0, T ]
Perfect replication

• Is perfect replication XT = CT possible?

Yes, if the market is complete.

– Which market (or model) allows us perfect replication?

– What kind of hedging strategy do we need?

Binomial lattice model: an example of complete market

 Single period binomial model

t=0 t=1
uS (1+r)
p
Stock S Bond 
1-p d<1+r<u
dS (1+r)

Portfolio X1(uS)=uS+r)

X0=S+

X1(dS)=dS+r)
Single period binomial model

C1 (uS )  max(uS  K , 0)

C0

C1 (dS )  max(dS  K , 0)
• Compare with portfolio process

C1 (uS )  X 1 (uS )  uS   (1  r ) Two equations for

C1 (dS )  X 1 (dS )  dS   (1  r ) two unknowns

Solve these equations for  and 

 C1 (uS )  C1 (dS )
C1 (uS )  uS   (1  r ) 
uS  dS
C1 (dS )  dS   (1  r ) uC (uS )  dC1 (dS )
  1
(1  r )(u  d )

C1  X 1 for each state

Comparison principle
C0  X 0  S  

1 1  r  d u  (1  r ) 
C0  C (uS )  C ( dS )
1  r  u  d 
1 1
ud
 1 1  r  d u  (1  r ) 
C0  C (uS )  C ( dS )
1  r  u  d 
1 1
ud 

~
p q~

1 ~
C0   pC1 (uS )  q~C1 (dS )
1 r
~
p  q~  1, ~ p  0, q~  0
~ ~
It is (notationally) convenient to regard p and q as probabilities
~
p , q~ : Risk neutral probability (real probability is irrelevant)
 Multi-period binomial lattice model
Stock price Call price
u4S u 4S  K
u 3S
u2S
u 3 dS u 3 dS  K
uS u 2 dS
Finite number
udS
S u 2d 2 S u 2d 2S  K of one step
ud 2 S models
dS
ud 3 S 0
2
d S
d 3S
d 4S 0
 Stock price Call price
uS3 ~ C4 (uS3 )  max(uS3  K , 0)
p
S3 C3 ( S 3 )

dS3 q~ C4 (dS3 )  max(dS3  K , 0)

C3 ( S3 ) 
1 ~
 pC4 (uS3 )  q~C4 (dS3 ), ~p  1  r  d , q~  1  ~p
1 r ud
Apply one step pricing formula at each step, and solve
backward until initial price is obtained.

1 ~ Ck (uS k 1 )  Ck (dS k 1 )
Ck 1 ( S k 1 )   pCk (uSk 1 )  q~Ck (dSk 1 )  k 1 
uS k 1  dS k 1
1 r
Multi-period binomial lattice model

• Perfect replication is possible

Market is complete

• Real probability is irrelevant

• Risk neutral probability dominates the pricing formula

 Theory Application
Discrete
Continuous

• Using the binomial lattice model as a guide, we are

going to introduce “Probability spaces”
Basics of
- Measure spaces (probability spaces)
- Measurable functions (random variables)
- Filtration
 Finite probability spaces
• Random experiment of 3 coin tosses
t=0 t=1 u 2 S  uS1 ( H )  S 2 ( HH )
S1 ( H )  uS 0
p uS
duS  dS1 ( H )  S 2 ( HT )
S0 S
udS  uS1 (T )  S 2 (TH )
1-p
dS
S1 (T )  dS 0
d 2 S  dS1 (T )  S 2 (TT )

After 3 tosses, the set  of all possible outcomes are given as

   HHH , HHT , HTH , HTT , THH , THT , TTH , TTT 
: sample space
   : sample point
S k ( ) : stock price at time k
S 0 ( ) S1 ( ) S 2 ( ) S ( )
3

u 3S
u 2S

uS u 2 dS
udS
S
ud 2 S
dS

d 2S
d 3S
 • Given a random experiment (three coin tosses), you are
only told if    or   
• This does not tell us anything except S 0 ( )
• We want S k ( ) to be “measurable”

Definition1.1: A -algebra is a collection  of subsets of  with

the following three properties:
1.  
2. If A  , then its complement A  
c

 
3. 
 Ak   ,  Ak  
 k 1 

 A pair (, ) is called a measurable space.

 An element of  is called a  measurable subset of 
 • A -algebra contains , so does 

0   ,  : trivial -algebra

   HHH , HHT , HTH , HTT , THH , THT , TTH , TTT 

1   , ,  HHH , HHT , HTH , HTT  , THH , THT , TTH , TTT 

H on the first toss T on the first toss
 AH  AT
• The additional information,   AH or   AT , gives us if
the first toss is H or T
2   , ,  HHH , HHT  ,  HTH , HTT  , THH , THT  , TTH , TTT  ,
HH on the first  AHT  ATH  ATT
two toss  AH
and all sets which can be build by taking unions of these }

• AH  AHT  AHH , AT  ATH  ATT , so 1  2

• The additional information,

  AHH ,   AHT ,   ATH , or   ATT ,
gives us if the first two tosses are HH, HT, TH, or TT
Filtration

0  1  2
• 0 contains no information
• 1 contains the information up to time 1 (the first toss)
• 2 contains the information up to time 2 (the first two tosses)

Definition 1.2: A filtration is an increasing sequence

of -algebras w.r.t time s.t.
0  1  2    k  
 Probability measure

Definition 1.3: A probability measure is a function mapping

into [0, 1] with the following properties:

1. ()  1
2. If A1 , A2 ,  is a sequence of disjoint sets in , then
  
 Ak    ( Ak )
 k 1  k 1

• If the coin has probability p for H and 1-p for T,

 AH    HHH , HHT , HTH , HTT 
  HHH    HHT    HTH    HTT 
 p 3  p 2 (1  p )  p(1  p ) p  p (1  p 2 )  p
Probability space and filtered space

 (, , P) is called a probability triple

 (, , {k }, P) is called a filtered space, where
0  1    

• What else?

We have not formally introduced random variables…

 • Random experiment of 3 coin tosses
S k ( ) : stock price at time k
S 0 ( ) S1 ( ) S 2 ( ) S ( )
3
Sk :   
u 3S
u2S Consider
uS u 2 dS S k1 ( B ) :    S k  B
udS B   (open interval)
S
ud 2 S
dS k  measurable?

d 2S
d 3S
 There are four sets for S11 ( B)

   S 1  uS   AH ,    S
1  uS   AH
   S 1  uS  dS   ,    S
1  uS , S1  dS   

On the other hand,

1   , ,  HHH , HHT , HTH , HTT  , THH , THT , TTH , TTT 
 AH  AT

S11 ( B) is k  measurable for all B (open interval on )

 In fact, we defined  k  s.t. S k1 ( B) is k  measurable
for all B (open interval on  )

Definition 1.4: Given (, , P), a function f :    is called

  measurable if
f 1 ( B)     f  B  
for all B (open interval on  )

 A random variable X :    is an   measurable function

 Adapted processes
Definition 1.5:
Given (, , {k },P), a process X k is called adapted
to the filtration {k } if for each k, X k is k  measurable.

Definition 1.6:
algebra (X) generated by a random variable X is the
smallest -algebra on containing all the sets
X 1 ( B); B (open interval on )

In the coin toss example, the information content of  ( S k )

is exactly the information learned by observing S k only.
Usually, k may be described as
k   ( S 0 , S1 ,  , S k )