STOCK PRICE MODELING AND INSIDER TRADING THEORY

A Thesis
Submitted to the Graduate Faculty of the
Louisiana State University and
Agricultural and Mechanical College
in partial fulllment of the
requirements for the degree of
Master of Science
in
The Department of Mathematics
by
Jessica J. Guillory
B.S. McNeese State University, 2001
December 2003
Acknowledgments
This thesis would not be possible without the help of several people. It is a
pleasure to thank Dr. Padmanabhan Sundar for his guidance, time, and patience.
This work was motivated by course work with Dr. Hui-Hsiung Kuo in stochas-
tic analysis and discussions with Dr. Padmanabhan Sundar. Integrating business
applications with probability theory piqued my interested.
This thesis is dedicated to my mother Patricia V. Guillory, father Gilbert Guil-
lory, brother Randell Dejean, and sister Kayla Dejean for their encouragement and
support.
ii
Table of Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Chapter 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter 2. Stochastic Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1 Geometric Brownian Motion . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Motivation of the Stochastic Derivative . . . . . . . . . . . . . . . . 5
2.4 Market Eciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Chapter 3. Portfolio of an Investor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1 Stock Price Process . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.1 Multivariate Case . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Portfolio Analysis of a Small Investor . . . . . . . . . . . . . . . . . 10
Chapter 4. Insider Trading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.1 Enlargement of a Filtration: An Example of G = B(1) . . . . . . . 13
4.2 Change of Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.3 Detecting Insider Trading . . . . . . . . . . . . . . . . . . . . . . . 19
4.4 Fair Pricing of an Option with Additional Information . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Appendix: Black-Scholes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
iii
Abstract
The mathematical study of stock price modeling using Brownian motion and
stochastic calculus is a relatively new eld. The randomness of nancial markets,
geometric brownian motions, martingale theory, Itos lemma, enlarged ltrations,
and Girsanovs theorem provided the motivation for a simple characterization of
the concepts of stock price modeling.
This work presents the theory of stochastic calculus and its use in the nancial
market. The problems on which we focus are the models of an investors portfolio
of stocks with and without the possibility of insider trading, opportunities for fair
pricing of an option, enlarged ltrations, consumptions, and admissibility.
This survey has two parts. The rst part explores the theoretical aspects of
stochastic calculus, and the second part shows its application in predicting stock
prices and the wealth of an investors portfolio.
iv
Chapter 1. Introduction
The subject of money is fascinating to many people. Whether one has a little or
a lot, a good portion of ones time is spent trying to gure out how to make more,
spend less, or stretch it as far as possible.
The world we live in has never been very predictable. Randomness has always
been a part of daily activity. Time, change, and uncertainty encompass some of the
most important business, economical, and nancial decisions. Stochastic models
become the tools required to deal with such problems. Stochastic calculus grew out
of the need to assign meaning from ordinary dierential equations to continuous-
time stochastic processes.
Discussions of stochastic models in the nancial market presented in this paper
focus on several topics:
Motivation of stochastic calculus and its use in stock price modeling.
Portfolio analysis of an investor.
Information of an insider and its eects on the market.
The fair price of an option.
1
Chapter 2. Stochastic Calculus
A range of problems came under the eld of the theory of functions of a real
variable when Newton and Leibniz invented calculus. The main focus of this in-
vention was the use of dierentiation to describe rates of change and limits in
approximating sums. Similar concepts are needed when dealing with stochastic
environments.
In nancial markets, pricing assets deal with stochastic variables. Therefore, the
notion of risk becomes an important factor. Can the same dierential rules apply
when dealing with stochastic variables?
Consider a one-dimensional continuous stochastic process S(t) where, t [0, T].
We dene the stochastic dierential equation
dS(t) = a(S(t), t)dt +b(S(t), t)dB(t)
where B(t) represents unpredictable events occurring in the interval dt. Since dS(t),
and dB(t) are random increments, they have to be justied by means other than
deterministic calculus.
2.1 Geometric Brownian Motion
In 1828, botanist Robert Brown observed the irregular movements of pollen
suspended in water, a phenomenon which is now known as Brownian movements.
The range of applications of Brownian motion goes further than the study of
particles suspended in water. The use of Brownian motion is the basis for modeling
certain aspects of the nancial market [14].
2
Denition 2.1.1. A Brownian Motion, B(t, ), is a process dened on some prob-
ability space (, F, P) satisfying the following properties:
1. P(; B(0, ) = 0) = 1.
2. For any 0 s < t, the random variable B(t) B(s) has distribution
N(0, t s).
3. For any 0 t
1
< t
2
< < t
n
, the random variables
B(t
1
), B(t
2
) B(t
1
), , B(t
n
) B(t
n1
) are independent.
4. P(; B( , )is a continuous function) = 1.
The rst quantitative work on Brownian motion was introduced in 1900 by the
French mathematician Bachelier who used it in his dissertation to model the price
movements of stocks and commodities [20]. However, Brownian motion has two
major aws that does not allow this model to function properly in an nancial
market setting.
1. Stock prices are always positive, and since the price of a stock is a normal
random variable, it can theoretically become negative.
2. Fluctuations in the price are proportional to the price of the stock.
Instead we introduce a nonnegative functional of Brownian motion called Geo-
metric Brownian motion which is dened as
S(t) = S
0
exp X(t),
where X(t) = B
t
+ (
1
2
)t is a Brownian motion.
3
2.2 Martingales
One of the most fundamental concepts of the theory of nance are martingales.
Consider a real-valued process S = S
t
: 0 t on a probability space (, F, P).
Denition 2.2.1. A ltration, F, is an increasing family of -elds, i.e:
F = (F
t
: t [0, T]), s t, F
s
F
t
.
For 0 t
1
< t
2
< < t
n
this family of information sets will satisfy
F
t
0
F
t
1
F
t
2
F
t
n

In the nancial perspective, one can think of F as containing the information
available to the investor at time t. The investor keeps all past and present infor-
mation about the market and we assume there is no foreknowledge of events.
Denition 2.2.2. A process, S(t), is a martingale with respect to the ltration
F if for any t > 0, S(t) is F
t
-measurable and satises:
1. E|S(t)| < ;
2. E(S(u)|F
t
) = S(t) whenever t u.
Based on the denition of a martingale, the expected value of a stock price in the
future given the information known at the present time t, is equal to the present
stock price. That is, stock prices that are martingales are risk neutral.
Denition 2.2.3. A process, S(t), is a submartingale(respectively, a supermartin-
gale) with respect to the ltration F if for any t > 0, S(t) is F
t
-measurable and
satises:
1. E|S(t)| < ;
2. E[S(u)|F
t
] S(t) ( respectively, E[S(u)|F
t
] S(t)) whenever t u.
4
However, stock prices are not completely unpredictable. In general, the price of
a stock is expected to increase over time, therefore stock prices are expected to
follow submartingale properties.
2.3 Motivation of the Stochastic Derivative
From standard calculus we know the derivative of a function f can be dened
as
lim
h0
f(x +h) f(x)
h
= f
x
. (2.3.1)
Suppose f is a function of a stochastic process x. A Taylor series expansion of
f(x) around x
0
will give
f(x) = f(x
0
) +f
x
(x
0
)[x x
0
] +
1
2
f
xx
(x
0
)[x x
0
]
2
+
1
3!
f
xxx
(x
0
)[x x
0
]
3
+ +R.
(2.3.2)
Rewrite f(x) as f(x
0
+x) , where x = x x
0
. Hence, the Taylor series can be
written as
f(x
0
+ x) f(x
0
) = f
x
(x
0
)[x] +
1
2
f
xx
(x
0
)[x]
2
+
1
3!
f
xxx
(x
0
)[x]
3
+ +R.
(2.3.3)
If the variable x were deterministic, one would say that the term (x)
2
was small,
therefore negligible. However, we assumed x was a random variable, therefore,
changes in x are random.
What happens to (x)
3
,(x)
4
,...? We will show that the quadratic variations of
the random process x converge to a meaningful random variable, while the higher
ordered variations approximate to zero.
5
Denition 2.3.1. A continuous semimartingale X = {X
t
, F
t
; 0 t } is an
adapted stochastic process which has the decomposition
X
t
= M
t
+A
t
, 0 t < ,
where M
t
is a continuous local martingale and A
t
is a continuous stochastic process
with paths of nite variation a.s.
Lemma 2.3.2. For a continuous semimartingale, X = {X
t
, F
t
; 0 t } has
quadratic variation that converges to a random variable, while higher-order varia-
tions of X converges to 0.
Proof. Let [A]
t
= lim
n

n
i=1
|A
ti+1
A
ti
|
2
, where limit is in the sense of prob-
ability.
lim
n

n
i=1
|A
ti+1
A
ti
|
2
lim
n

n
i=1
|A
ti+1
A
ti
| max
i
|A
ti+1
A
ti
|.
Since A
t
has nite variations, there exist K > 0 such that lim
n

n
i=1
|A
ti+1

A
ti
| < K
By the continuity of A
t
,
max
i
|A
ti+1
A
ti
| < , when |t
i+1
t
i
| < , for some > 0. (2.3.4)
Hence, lim
n

n
i=1
|A
ti+1
A
ti
|
2
< K .
Allowing 0 , we have [A]
t
= lim
n

n
i=1
|A
ti+1
A
ti
|
2
0.
[X]
t
= [M]
t
+ 2[M, A]
t
+ [A]
t
.
By Cauchy-Schwartz inequality:
[M, A]
t
[M]
t
1
2
[A]
t
1
2
.
Hence, [M, A]
t
= 0. Therefore, [X]
t
= [M]
t
.
6
Consider,
lim
n
n

i=1
|X
ti+1
X
ti
|
3
[X]
t
max
i
|X
ti+1
X
ti
|
= [M]
t
max
i
|X
ti+1
X
ti
|
= [M]
t
max
i
|M
ti+1
M
ti
+A
ti+1
A
ti
|
[M]
t

_
max
i
|M
ti+1
M
ti
| + max
i
|A
ti+1
A
ti
|
_
.
Continuity of M
t
implies that for any , there exist an > 0 such that
|M
u
() M
v
()| < when |u v| < for > 0
From 2.3.4 we know max
i
|A
ti+1
A
ti
| 0, hence
[M]
t

_
max
i
|M
ti+1
M
ti
| + max
i
|A
ti+1
A
ti
|
_
0 as 0.
Therefore,
lim
n
n

i=1
|X
ti+1
X
ti
|
3
0.
Notation 2.3.3. Similar arguments hold for higher-ordered variations.
As a result of 2.3.2, the higher-ordered terms of x can be approximated by zero.
The Taylor approximation can be written as
f(x
0
+ x) f(x
0
)

= f
x
(x
0
)[x] +
1
2
f
xx
(x
0
)[x]
2
. (2.3.5)
Therefore, the total change in f(x
0
) is given by (2.3.5).
Theorem 2.3.4 (The Ito Formula). Let F(S(t), t) be a twice dierentiable func-
tion of t and of the random process S(t) where
dS(t) = a(t)dt +(t)dB(t), t 0,
with drift and diusion parameters, a(t), (t). Then we have
dF(S(t), t) =
F
S(t)
dS(t) +
F
t
dt +
1
2

2
F
S(t)
2

2
(t)dt,
7
2.4 Market Eciency
A fundamental assumption about the market known as the Ecient Market
hypothesis, claims that asset prices quickly reect all available information. This
hypothesis states that a market digests new information in such an ecient way
that all the current information about the market development, including past
prices, is at all times contained in the present price. The basic idea behind any test
for market eciency is that if investors can use the information to earn abnormal
prots, the market is not ecient.
Denition 2.4.1. The weak-form of the ecient market hypothesis states that
all information contained in past prices are reected in current stock prices.
If the weak-form of market eciency were not true, investors would make above-
average returns by interpreting the past history of stock prices. Unfortunately, the
competition in the marketplace ensures that the weak-form of the ecient market
hypothesis holds by responding to the increasing demand and price.
Denition 2.4.2. Let d be a positive integer and P be a probability measure on
(
d
, B(
d
)). A d-dimensional process S = (S
t
, F
t
; t 0) on some probability space
(, F, P) is said to be a Markov process if:
The stochastic process S
n
is adapted to the ltration F
n
and
P(S
k+1
A|F
k
) = P(S
k+1
A|S
k
) for all A B(
d
), P- a.s.
According to this denition, future movements in S
k
, given what we observe
until time k, are likely to be the same as starting the process at time k. That is,
no advantage is gained by taking into account any of the previous price evolutions.
Therefore, the Markov property of stock prices is consistent with the weak-form of
market eciency.
8
Chapter 3. Portfolio of an Investor
3.1 Stock Price Process
Consider a non-dividend-paying risky asset, S
t
, with the following parameters:
, the expected rate of return.
> 0, the volatility.
S
0
> 0, the initial stock price.
The stock price follows a continuous-time model:
S
t
= S
0
e
B
t
+(
1
2

2
)t
t [0, ) (3.1.1)
Example 3.1.1. Deriving the Stock Price model:
The stock price model is given by the linear stochastic dierential equation:
dS(t) = Sdt +SdB(t).
Integrating both sides we get,
_
t
0
dS(u)
S(u)
= t +B(t). (3.1.2)
Since S(t) is a random variable, Itos formula is used. Let F(S) = ln(S), by Itos
formula (2.3.4):
dF(S) =
F
S
dS +
1
2

2
F
S
2
dS
2
=
1
S
dS +
1
2
(
1
S
2
)(S
2

2
dt)
=
dS
S

1
2

2
dt
Therefore,
_
t
0
dF(S) =
_
t
0
dS(u)
S(u)

_
t
0
1
2

2
du.
9
We can rewrite this as
_
t
0
dS(u)
S(u)
=
_
t
0
dF(S) +
_
t
0
1
2

2
du
= ln(S(t) ln(S(0)) +
1
2

2
t.
Combining 3.1.2 and 3.1.3:
t +B(t) = ln(
S(t)
S(0)
) +
1
2

2
t.
ln(
S(t)
S(0)
) = t +B(t)
1
2

2
t.
Hence,
S(t) = S(0)e
B(t)+(
1
2

2
)t
.
3.1.1 Multivariate Case
In general, most investors will have more than one risky asset. Consider a
set of n-risky assets. We can use the linear stochastic dierential equation (3.1.1),
where
S(t) =
_

_
S
1
(t)
S
2
(t)
.
.
.
S
n
(t)
_

_
, (t) =
_

1
(t)

2
(t)
.
.
.

n
(t)
_

_
, =
_

11

1n
.
.
.

n1

nn
_

_
.
Hence the price S
i
(t) for a share of the ith stock at time t is modeled by the
linear stochastic dierential equation:
dS
i
(t) =
i
(t)S
i
(t)dt +S
i
(t)
j

n=1

ij
dB
i
(t).
3.2 Portfolio Analysis of a Small Investor
The structure of a portfolio of an investor relies on the theory of stochastic
calculus and dierential equations to model the risky assets in continuous time.
10
Denition 3.2.1. A market is d+1-valued stochastic process S(t) = S
0
(t), , S
d
(t).
Denition 3.2.2. Let
i
(t) denote the number of shares of asset i owned by the
investor at time t. The value, (wealth), (t) of a portfolio (t) for the market S(t)
is given by
X(t) =
d

i=0

i
(t)S
i
(t) (3.2.1)
Denition 3.2.3. A portfolio process = (t) = (
1
(t), ,
d
(t)), F
t
is a stochas-
tic process where
j
(t)s are non anticipating and
d

i=1
_
0
T

2
i
(t)dt < , a.s. (3.2.2)
We denote
i
(t) =
i
(t)S
i
(t) as the amount invested in the i-th stock, 1 i d,
at time t.
Denition 3.2.4. A portfolio is said to be self-nancing if
dX(t) = d =
d

i=0

i
(t)dS
i
(t)
Denition 3.2.5. A consumption process C = C
t
, F
t
is a measurable , adapted
process such that, C [0, ) and
_
0
T
C
t
< , a.s. (3.2.3)
Example 3.2.6 (Wealth of a small investor). Consider an investor who has
d + 1-asset from a given market where these assets are traded continuously. One
of these assets is riskless, and has a price S
0
(t) which evolves according to the
dierential equation
dS
0
(t) = r(t)S
0
(t)dt, 0 t T. (3.2.4)
The remaining risky assets have a price modeled by the linear stochastic dierential
equation
dS
i
(t) =
i
(t)S
i
(t)dt +S
i
(t)
n

j=1

ij
dB
i
(t) i = 1 d. (3.2.5)
11
If the investor choses at time t + to consume an amount C
t+
, then the rate of
change of the reduced wealth would be
dX(t) =
d

i=0

i
(t)dS
i
(t) Cdt (3.2.6)
The wealth of the portfolio can be written as
dX(t) = [r(t)X(t) C
t
]dt +
d

i=1
[
i
(t) r(t)]
i
(t)d(t) +
d

j=1
d

i=1

i
(t)
ij
dB
j
(t).
Which has the solution:
X(t) = e

t
0
r(s)ds
_
X
0
+
_
t
0
e

s
0
r(u)du
_
(s)
T
((s) r(s)1) C
s

ds +
_
t
0
e

s
0
r(u)du
(s)
T
(s)dB(s)
_
.
(3.2.7)
Denition 3.2.7. A pair (, C) is said to be admissible for the initial endowment
X(0) if the wealth process 3.2.2 satises
X(t) 0; 0 t T, a.s.
Denition 3.2.8. An arbitrage is a portfolio that starts with X(0) = 0 and ends
with
P[X(t) 0] = 1, P[X(t) > 0] > 0 a.s.
12
Chapter 4. Insider Trading
Suppose we have a natural ltration, F
t
on a trading interval [0, 1]. We let the
nancial market model consist of a mean rate of return process and a volatility
process which determines the stock price process given by
dS(t) = (t)dt +dB(t).
There are two type of traders that can act on the market:
The regular trader whose information levels correspond to the natural infor-
mation ow of the market F
t
The inside trader, who has access to information beforehand.
This knowledge consists of a random variable G that is F
1
-measurable. Hence
the insiders ltration is given by (G
t
), t [0, 1] where G
t
= F
t
(G).
4.1 Enlargement of a Filtration: An Example of
G = B(1)
Let G = B(1). The process {B(t)} is an F
t
-adapted Brownian motion and
does not remain as a Brownian motion under the enlarged ltration G
t
. In order
to nd the semimartingale decomposition of {B(t)} under the new ltration, we
need the following lemmas:
Lemma 4.1.1. For any T > t, the conditional distribution of B(t) given B(T) is
the N(
B(T)t
T
;
t(Tt)
T
) distribution.
13
Proof.
f
B(t)|B(T)
(x, y) =
f(x, y)
_

f(x, y)dx
=
f
B(t)|B(T)B(t)
(x, y x)
f
B(T)
(y)
.
By iid of increments: =
f
B(t)
(x) f
B(T)B(t)
(y x)
f
B(T)
(y)
=
1

2t
e

x
2
2t
1

2(Tt)
e

(yx)
2
2(Tt)
1

2T
e

y
2
2T
=

T
2t(T t)
e

(xy)
2
T
2t(Tt)
Hence, B(t)|B(T) N(
B(T)t
T
;
t(Tt)
T
).
Lemma 4.1.2. For any 0 < s < r < t < 1,
E [B(t) B(s)|G
s
] =
t s
1 s
(B(1) B(s).
Proof. Let s < t be xed in the interval [0, 1].
E [B(t) B(s)|G
s
] = E [B(t) B(s)|B(u) : u s, B(1)]
= E [B(t) B(s)|B(u) : 0 u s, B(1) B(s)]
= E [B(t) B(s)|B(1) B(s)] by independence of increments.
Dene a processes

B as follows:

B(r) = B(s + r) B(s) for all r 0, and a
xed s. {

B}
r0
is a standard Brownian motion. By lemma 4.1.1, the conditional
distribution of

B(t s) given

B(1 s) is the normal distribution
N
_

B(1 s)(t s)
1 s
;
(t s)(1 t)
(1 s)
_
.
Thus,
E
_

B(t s)|

B(1 s)
_
=

B(1 s)(t s)
(1 s)
.
14
That is,
E [B(t) B(s)|B(1) B(s)] =
B(1) B(s)(t s)
(1 s)
.
Hence, the expected value of B(t) under the enlarged ltration is
E(B(t)|G
s
) = B(s) +
t s
1 s
(B(1) B(s)). (4.1.1)
Theorem 4.1.3. The process {B(t)
_
t
0
1
1u
(B(1) B(u))du}
0t1
is a {G}
0t1
-
adapted Brownian motion.
Proof. Suppose G
s
= F
s
(B(1)). Let s < r < t < 1, then G
s
G
r
.
E (B(t)|G
s
) = E [E(B(t)|G
r
)|G
s
]
By (4.1.1), E (B(t)|G
s
) = E[B(r) +
t r
1 r
(B(1) B(r))|G
s
]
= E(B(r)|G
s
) +
t r
1 r
E(B(1) B(r)|G
s
).
Suppose we partition the interval s < t < 1 such that
s = r
n+1
< r
n
< r
n1
< r
1
< r
0
= t < 1.
15
Using equation (4.1.2) with r = r
1
, r
2
, ...
E (B(t)|G
s
) = E(B(r
1
)|G
s
) +
t r
1
1 r
1
E(B(1) B(r
1
)|G
s
)
= E(B(r
2
)|G
s
) +
r
1
r
2
1 r
2
E(B(1) B(r
2
)|G
s
) +
t r
1
1 r
1
E(B(1) B(r
1
)|G
s
)
.
.
.
= E(B(r
n
)|G
s
) +
n

j=1
r
j1
r
j
1 r
j
E(B(1) B(r
j
)|G
s
)
= B(s) +
n+1

j=1
r
j1
r
j
1 r
j
E(B(1) B(r
j
)|G
s
)
= B(s) +E
_
n+1

j=1
r
j1
r
j
1 r
j
(B(1) B(r
j
)|G
s
_
= B(s) +E
__
t
s
1
1 u
(B(1) B(u))du|G
s
_
, as n with lim
j
|r
j1
r
j
|
= B(s) +E
__
t
0
1
1 u
(B(1) B(u))du|G
s
_

_
s
0
1
1 u
(B(1) B(u))du,
since B(1) B(u) is G-measurable if u s.
E(B(t)|G
s
= B(s)+E
__
t
0
1
1 u
(B(1) B(u))du|G
s
_

_
s
0
1
1 u
(B(1)B(u))du.
Therefore,
E(B(t)
_
t
0
1
1 u
(B(1) B(u))du|G
s
) = B(s) +
_
s
0
1
1 u
(B(1) B(u))du.
Hence B(t)
_
t
0
1
1u
(B(1)B(u))du is a G-martingale. Since the quadratic variation
of {B(t)
_
t
0
1
1u
(B(1) B(u))du}
t0
is t, we get that {B(t)
_
t
0
1
1u
(B(1)
B(u))du} is a G-Brownian motion.
We dene the new stock price process in terms of the enlarged ltration as
dS(t) = S(t)dt +S(t)dM(t) +S(t)
B(1) B(t)
1 t
dt. (4.1.2)
where M(t) = B(t)
_
t
0
1
1u
(B(1) B(u))du is a G-martingale.
16
Remark 4.1.4. As an example the ltration G was an enlargement of F by B(1).
In general, an enlarged ltration has the form
_
1
0
f(u)dB(u), wheref L
2
[0, 1].
Let
_
1
0
f
2
(u)du > 0, for all t < 1, then under enlargement of ltration,
B(t)
_
t
0
_
1
u
f(v)dB(v)
_
1
u
f
2
(v)dv
f(u)du
is a G -Brownian Motion [15].
4.2 Change of Measure
Theorem 4.2.1 (Girsanov Theorem). Let {B(t)}
0tT
be a Brownian motion
on a probability space (, F, P). Let {F
t
}
0tT
be the accompanying ltration, and
let {X(t)} be an adapted process such that E
_
e
1
2

T
0
|X(u)|
2
du
_
< .
For 0 t T let M(t) be the martingale dened by
M(t) = e

t
0
X(u)dB(u)
1
2

t
0
X(u)
2
du
.
Let Q
t
be a new probability measure on (, F
t
) given by
Q
t
(A) =
_
A
M(t)dP, A F
t
.
If Q
T
is denoted by Q, then Q|
F
t
= Q
t
, and the stochastic process

B(t) dened by

B(t) = B(t) +
_
t
0
X(u)du, t [0, T] is a Brownian motion on (, F, {F
t
}, Q).
In general, bonds and stocks exhibit submartingale (2.2.3) characteristics. To
nd a fair market value of an asset,
S(t)
S
0
(t)
should be converted to a martingale.
Using the Girsanov Theorem, one could easily change the measure P so that the
drift of
S(t)
S
0
(t)
is zero.
17
Recall the following equation for the wealth of an investor:
dX(t) = [r(t)X(t)C
t
]dt+
d

i=1
[
i
(t)r(t)]
i
(t)dt+
d

j=1
d

i=1

i
(t)
ij
dB
j
(t) (4.2.1)
Girsanov theorem allows us to remove the drift

d
i=1
[
i
(t)r(t)] by considering a
new Brownian motion,

B, under a dierent probability measure Q. Let us introduce
the process
(t) = (t)
1
[(t) r(t)1] .
Let
Z(t) = e

d
j=1

t
0
(u)dB(u)
1
2

t
0
(u)
2
du
.
We dene the probability measure
Q
T
(A) = E
P
[Z(t)1
A
] ; A F
t
and the process

B(t) = B(t) +
_
t
0
(u)du
is a Brownian motion under Q
T
.
Hence we can write the wealth process as
X(t)e

t
0
r(s)ds
+
_
t
0
e

s
0
r(u)du
C
s
ds = X
0
+
_
t
0
e

s
0
r(u)du
(s)
T
(s)d

B(s).
Denition 4.2.2. Let (, F, P) be a probability space equipped with a ltra-
tion {F
t
}
0t
. A random time T is a stopping time of the ltration if the event
{, T() t} belongs to the ltration {F
t
}, for every t 0.
Denition 4.2.3. Let X = {X
t
: F
t
, 0 t } be a stochastic process. If there
exist a nondecreasing sequence {T
n
}

n=1
of stopping times of {F
t
} such that X
tT
n
is a martingale and P[lim
n
T
n
= ] = 1, then X is a local martingale.
18
Lemma 4.2.4. If for an admissible pair (, C), and X
0
+
_
t
0
e

s
0
r(u)du
(s)
T
(s)d

B(s)
a Q-local martingale then X(t)e

t
0
r(s)ds
+
_
t
0
e

s
0
r(u)du
C
s
ds is a nonnegative super-
martingale and E
Q
_
X(t)e

t
0
r(s)ds
+
_
t
0
e

s
0
r(u)du
C
s
ds
_
X
0
Proof. Let Y (t) = X(t)e

t
0
r(s)ds
+
_
t
0
e

s
0
r(u)du
C
s
ds be nonnegative. Since Y (t)
is a local martingale, it follows that Y (t T
n
)
t0
is a martingale. For s t,
E
Q
[Y (t T
n
)|F
s
] = Y (s T
n
). Allowing n , by Fatous Lemma
E
Q
_
liminf
n
Y (t T
n
)|F
s
_
lim
n
E
Q
[Y (t T
n
)|F
s
]
= lim
n
Y (s T
n
)
= Y (s)
Therefore E[Y (t)|F
s
] Y (s). Hence Y (t) is a nonnegative super-martingale.
Since E[Y (t)|F
s
] Y (s), it follows that
E [E(Y (t)|F
s
] E[Y (s)]
E[Y (t)] E[Y (s)]
Let s = 0, then
E[Y (t)] E[Y (0)]
E[Y (t)] X
0
4.3 Detecting Insider Trading
Denition 4.3.1. Recall from 4.1.3 we can dene a martingale in terms of an
enlarged ltration. Using Girsanov Theorem we dened a new probability measure.
Is it possible to have a risk-neutral processes with insider information?
19
Let (s) =
B(1)B(s)
1s
and (t) = (t)
1
[(t) r(t)1] .
By Girsanov Theorem, 4.2.1, we dene a new probability measure. Let M(t) be
a (F, P)-martingale dened as:
e

t
0
[(s)+(s)]dB(u)
1
2

t
0
[(s)+(s)]
2
du
.
Let Q be a probability measure on (, F) given by
Q
t
(A) = E
P
[M(t)1
A
] , A F
t
.
The stochastic process

B(t) dened by

B(t) = B(t) +
_
t
0
[(u) +(u)]du, t [0, T]
is a Brownian motion on (G, Q).
Proposition 4.3.2. Let (, C) be an admissible pair and the nal wealth associ-
ated with the pair is X
,C
. Then there exist a positive function h for which
E
Q
_
X
,C
T
e

T
0
r(s)ds
+
_
T
0
C(t)e

t
0
r(s)ds
dt|G
0
_
h(B(1)),
such that X
,C
0
= h(B(1)).
Theorem 4.3.3 (Martingale Representation Theorem). Suppose Z is a Q-
martingale adapted to G
t
then there exist an adapted process (t) such that
Z(t) = Z(0) +
_
t
0
(t)d

B(t).
Proposition 4.3.4. Given an initial wealth h(B(1)), a consumption C, and a
random variable Z(t) on the probability space (, G, Q) such that
E
Q
_
Z
T
e

T
0
r(s)ds
+
_
T
0
C(t)e

t
0
r(s)ds
dt|G
0
_
= h(B(1)),
there exist a portfolio
Z
that is G-adapted such (
Z
, C) is admissible and
X

Z
,C
T
= Z(T).
20
4.4 Fair Pricing of an Option with Additional
Information
Suppose at time t = 0 we sign an option which gives us the right to buy, at the
maturity time T, a share of the stock S
k
(t) at a strike price q. At maturity, if the
price P
k
(T) of the share is below the strike price q, the option is not exercised; on
the other hand, if P
k
(t) > q, we can exercise the option at time T to buy the stock
at price q, and then sell the share immediately in the market for P
k
(t).
This option is known as a European call option which has a payment of (P
k
(T)
q)
+
at T.
Let X(0) 0, and (, C) be a portfolio/consumption process which is admis-
sible, 3.2.7, for the initial endowment X(0). The pair (, C) is called a hedging
strategy against the contingent claim provided that
C
t
, the payo rate, and
X(T), the terminal payo at maturity
holds a.s., where X is the wealth process (3.2.2) associated with the pair (, C).
Suppose an investor knows privileged information about a stock that he wanted
to sell. To reduce the risk of loosing money on this stock transaction he needs to
nd an appropriate hedging strategy such that it is risk neutral.
Consider a contingent claim consisting of a payment Z at time T. By Proposition
4.3.4 there exist a hedging strategy against the contingent claim, (
Z
, C), whose
corresponding wealth process satises
X

Z
,C
T
= Z(T).
Therefore
X

Z
,C
0
= E
Q
_
Z
T
e

T
0
r(s)ds
+
_
T
0
C(t)e

t
0
r(s)ds
dt|G
0
_
.
21
If Z is a contingent claim with {C
t
} as the accompanying consumption process,
and if h(B(1)) is not given a priori, then
E
Q
_
Z
T
e

T
0
r(s)ds
|G
0
_
(4.4.1)
gives the price of this contingent claim at time 0. Hence we can conclude there is
a fair price for the contingent claim to be paid given insider information.
For instance, the seller of a European call option with strike price q and maturity
date T will sell this claim at time t = 0 for the price give by (4.4.1) if C
s
is the
dividend paid at time s and Z = max[S
k
(T) q, 0] is the nal pay-o. The eect
of additional information known in advance is reected in the denition of the
measure Q.
22
References
[1] Arnold, Ludwig, Stochastic Dierential Equations: Theory and Applications,
John Wiley and Sons, New York, 1974.
[2] Baras, John, and Vincent Mirelli, Recent Advances in Stochastic Calculus,
Springer-Verlag, New York, 1990.
[3] Billingsley, Patrick, Probability and Measure, John Wiley and Sons, Inc., New
York, 1995.
[4] Brealey, Richard, Security Prices in a Competitive Market, The M.I.T. Press,
Cambridge, 1971.
[5] Breiman, Leo, Probability and Stochastic Processes: With A View Toward Ap-
plication, Houghton Miin Company, Boston, 1969.
[6] Cootner, Paul H., The random character of stock market prices, The MIT
press, Cambridge, 1964.
[7] Dudley, R.M.,Real Analysis and Probability, Wadsworth and Brooks, Pacic
Grove, 1989.
[8] Durrett, Richard, Stochastic Calculus: A practical introduction, CRC Press,
Boca Raton, 1996.
[9] Elliot, Robert J., P. Ekkehard Kopp,Mathematics of nancial Markets,
Springer, New York, 1999.
[10] Elton, Edwin J., Modern Portfolio Theory and Investment Analysis, John
Wiley and Sons, New York, 1981.
[11] Fernholz, E. Robert, Stochastic Portfolio Theory, Springer-Verlag, New York,
2002.
[12] Grorud, Axel, and Monique Pontier Comment detecter le delit dinities?,
Academy of Science in Paris, Vol 324 pg 1137-1142, 1997.
[13] Hull, John C., Options, futures and other derivatives, 4th Edition, Prentice
Hall, Upper Saddle River, 2000.
[14] Karatzas, Ioannis, and Steven E. Shreve, Brownian Motion and Stochastic
Calculus, Graduate Text in Mathematics, Springer-Verlag, New York, 1991.
[15] Koski, T., and P. Sundar, Two Applications of Reproducing Kernel Hilbert
Spaces and Stochastic Analysis, Stochastics in nite and innite dimensions.,
195-206, Birkhauser, 2000.
23
[16] Kuo, H.-H, Lecture notes in Stochastic Processes, Louisiana State University,
Department of Mathematics, 2002.
[17] Mehata, Srinivasan, Stochastic Processes, McGraw-Hill, New York, 1978.
[18] Neftci, Salih, Introduction to the Mathematics of Financial Derivatives, Aca-
demic Press, San Diego, 1996.
[19] Paul,Wolfgang, and Jorge Baschnagel, Stochastic Processes fro Physics to Fi-
nance, Springer-Verlag, Berlin, 1999.
[20] Ross, Sheldon, A rst course in Probability, 6th Edition,Prentice Hall, Upper
Saddle River, 20002.
[21] Tapiero, Charles,S., Applied Stochastic Models and control for Finance and
Insurance, Kluwer Academic Publisher, 1998.
24
Appendix: Black-Scholes
In the early 1970s a major breakthrough in the pricing of stock options was
developed known as the Black-Scholes model. This model has had such an inuence
on the way traders price and hedge options that Scholes later won the Nobel price
for economics [13]. The Black-Scholes formula gives the price of a call option,
F(t, S
1
(t)), when the following conditions apply:
The riskless interest rate, r, is constant over the life of the option.
The call options is European. Therefore it cannot be exercised before the
expiration date T.
There are no transaction costs, and the security pays no dividend before the
option matures.
The price S
1
(t) follows the stock price process of a geometric Brownian mo-
tion.
Example .0.1 (Black-Scholes option pricing). Let the price of a bond obey
the process
dS
0
(t) = rS
0
(t)dt,
and the price of a stock obeys
dS
1
(t) = S
1
dt +S
1
dB(t) = rS
1
dt +S
1
d

B(t).
The option to buy one share of the stock at time T at the price q follows the
wealth processes corresponding to the fair price of the contingent claim:
V
t
= E
Q
_
e
r(Tt)
(S
1
(t) q)
+
|F
t

; 0 t T.
25
Theorem .0.2 (Black-Scholes formula). The price of a European call option
with exercise price q and time of maturity T is given by :
F(S
1
(t), t) = S
1
(t)N(d
1
) qe
r(Tt)
N(d
2
), F(S
1
(T), T) = max[S
1
(T) q, 0]
where
d
1
=
ln
_
S
1
(t)
q
_
+ (r +

2
2
)(T t)

T t
, d
2
= d
1

T t,
and
N(d
i
) =
1

2
_
d
i

1
2
x
2
dx i = 1, 2.
The Black-Scholes formula explicitly interprets the valuation process V
t
into
F(S
1
(t), t) , making all security buying and selling bets fair under the previous
conditions.
26
Vita
Jessica Guillory was born on December 27, 1979, in Lake Charles, Louisiana.
She nished her undergraduate studies at McNeese State University May 2001. In
August 2001 she came to Louisiana State University to pursue graduate studies in
mathematics. She is currently a candidate for the degree of Master of Science in
mathematics, which will be awarded in December 2003.
27