Stochastic Calculus for Finance

Umesh Kumar

Lecture Notes for the Course MMATH18-305(ii) oered in Semester III of

M.Sc.(Mathematics) of University of Delhi.
2021-22

December 08, 2021

Contents
1 Introduction to Probability Theory 3
1.1 Probability Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
PROBLEMS - 1.1 : Probability Space . . . . . . . . . . . . . . . . . . . . 8
1.2 Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
PROBLEMS - 1.2 : Random Variables . . . . . . . . . . . . . . . . . . . . 11
1.3 Expectation of a Random Variable . . . . . . . . . . . . . . . . . . . 13
PROBLEMS - 1.3 : Expectation . . . . . . . . . . . . . . . . . . . . . . . 15
1.4 Independence and Conditional Expectation . . . . . . . . . . . . . . 17
1.4.1 Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.2 Joint Distribution of a Random Vector . . . . . . . . . . . . . 18
1.4.3 Conditional Expectation . . . . . . . . . . . . . . . . . . . . . 20
PROBLEMS - 1.4: Independence and Conditional Expectation . . . . . . 22
2 Brownian Motion and Martingales 24
2.1 Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
PROBLEMS-2.1: Brownian Motion . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Quadratic Variation of a Brownian Motion . . . . . . . . . . . . . . . 27
2.3 Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
PROBLEMS-2.2 : Martingales . . . . . . . . . . . . . . . . . . . . . . . . 30
3 Stochastic Calculus 31
3.1 Itô Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.1 Itô Integral for simple process . . . . . . . . . . . . . . . . . . 32
3.1.2 Itô Integral for general process . . . . . . . . . . . . . . . . . 33
3.1.3 Itô Integral Process . . . . . . . . . . . . . . . . . . . . . . . . 34
PROBLEMS 3.1 - Itô Integral . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Itô Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.1 Itô Formula for Itô Process . . . . . . . . . . . . . . . . . . . 38
PROBLEMS - 3.2 : Itô Formula . . . . . . . . . . . . . . . . . . . . . . . 39
4 Financial Mathematics 40
4.1 Black-Scholes Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Black Scholes PDE: Derivation . . . . . . . . . . . . . . . . . . . . . 42
4.3 Black-Scholes PDE: Solution . . . . . . . . . . . . . . . . . . . . . . . 44

1
Stochastic Calculus Umesh Kumar

4.4 Black-Scholes Option Pricing Formulae . . . . . . . . . . . . . . . . . 46

4.4.1 Put-Call Parity . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.5 Option Greeks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.6 Girsanov's Theorem and Risk-Neutral Valuation . . . . . . . . . . . 50

2
Chapter 1

Introduction to Probability
Theory
1.1 Probability Space

Denition 1.1.1 (Random Experiment ). A random experiment is an experiment

whose outcome can not be predicted without performing the experiment. A random
experiment can be described by the following:
(1.) Sample Space : the set of all possible outcomes of a random experiment. We
will denote it by Ω.
(2.) Event Space: an event is a set of outcomes of the random experiment, which
can be observed to hold or not to hold after the experiment. An event is a subset
of the Sample Space, i.e., A ⊂ Ω. The family of all events will be denoted by F .
(3.) Probability. the probability is a number between 0 and 1 which is assigned to
an event.
Example 1.1.2 (Tossing a coin ). Suppose a coin is tossed once. The outcomes are
head (H) and tail (T). The sample space is the set Ω = {H, T }. An event is a subset
of Ω. For example, A = {H} is the event of getting head as the outcome. The
probability of this event can be assigned as P (A) = 1/2, if the coin is a fair coin.
Example 1.1.3 (Tossing a coin twice ). Suppose a (fair) coin is tossed twice. The
sample space is the set Ω = {HH, HT, T H, T T }. Some examples of events are
A = {HH, HT, T H} (atleast one toss results in head);
B = {T T } (both tosses result in tails);
C = {HT, T H} (exactly one of the tosses results in head).

The corresponding probabilities are P (A) = 3/4, P (B) = 1/4, P (C) = 1/2.

3
Stochastic Calculus Umesh Kumar

Example 1.1.4 (Throwing two dice ). The sample space for rolling two dice is Ω =
{(i, j) : i, j ∈ {1, . . . , 6}}. An example of an event in this experiment is that the sum
of numbers on two dice is 5, that is, A = {(1, 4), (2, 3), (3, 2), (4, 1)}. One can assign
the probability as P (A) = 4/36 = 1/9.
In each of the examples above, the sample space is nite. Every subset of the
sample space is an event and probability is assigned by
|A|
A ⊂ Ω, P (A) = ,
|Ω|

where |A| denotes the cardinality of A, i.e., the number of elements in A. We have
also assumed here that all the outcomes are equally likely to happen. The latter
assumption of equally likely outcomes can be easily removed as we only want that
the probability is assigned in a "consistent" way to all events and will be discussed
in a later example. In an innite set not every subset can be assigned a probability
in a consistent way. In other words, not every subset of an innite sample space is
an event.
Example 1.1.5. Suppose we draw randomly a number from [0, 1]. The sample space
in this case is given by Ω = [0, 1]. An event A is identied with the event that the
number that we draw happens to be in the set A. For A = [a, b], 0 < a < b < 1, one
can assign the probability as P ([a, b]) = b − a, that is, the length of the interval [a, b].
The question now is whether every subset of [0, 1] has a "length"? You may be aware
of non-measurable subsets of [0, 1], so not every subset of [0, 1] can be regarded as
an event.
Remark 1.1.6. Do you know about Vitali sets? Do you know about Banach-Tarski
paradox? If not, may be you can try to nd out.
Example 1.1.7 (Share Price tomorrow ). Suppose we want to predict the price of
a share tomorrow. The sample space is Ω = [0, ∞). An example of an event is
that the share price will be larger than or equal to 100 rupees tomorrow, that is,
A = [100, ∞). Another event can be that the share price will lie between 100 rupees
and 400 rupees, that is B = [100, 400]. To assign probability, one needs to know how
share prices are modelled.
In probability theory, we model a random experiment using a probability space.
Kolmogorov dened a probability space by a triple (Ω, F, P ), where the event space
F is a sigma algebra and P is a probability measure. We now discuss these in detail.

Denition 1.1.8 (Sigma Algebra). Let Ω be a non-empty set and F be a family

of subsets of Ω. Then F is called a σ -algebra of Ω if the following are satised:
(1) ϕ ∈ F ;
(2) if A ∈ F , then Ac ∈ F ;

4
Stochastic Calculus Umesh Kumar

∞
(3) if A1 , A2 , · · · ∈ F , then An ∈ F .
[

n=1

Remark 1.1.9. By (1) and (2), it is obvious that Ω ∈ F . Some authors take Ω ∈ F
instead of (1) in the denition of a σ -algebra. If A, B ∈ F , then it easily follws that
A ∪ B ∈ F.

Example 1.1.10. Let Ω be a non-empty set.

(1) F = {ϕ, Ω} is trivially a σ -algebra on Ω;
(2) F = P(Ω), where P(Ω) denotes the power set of Ω is a σ -algebra on Ω;
(3) For any proper subset A ⊂ Ω, the collection F = {ϕ, A, Ac , Ω} is a σ -algebra of
Ω.

Remark 1.1.11. A σ-algebra is closed under countable set operations. For example,
∞
if A1 , A2 , · · · ∈ F , then An ∈ F because by De Morgan's laws we have
\

n=1

∞ ∞
!c
\ [
An = Acn .
n=1 n=1

Theorem 1.1.12. Intersection of an arbitrary family of σ-algebras of Ω is also a

σ -algebra.

Proof. Exercise
Denition 1.1.13 (generated σ-algebra). Let C be a class of subsets of Ω. The
σ -algebra generated by C , denoted by σ(C), is the smallest σ -algebra on Ω containing
C.

It is the intersection of all σ -algebras that contain C as a subclass.

Remark 1.1.14. Note that σ(C) is the unique σ-algebra satisfying:
(a) C ⊂ σ(C)
(b) if A is a σ -algebra of Ω with C ⊂ A, then σ(C) ⊂ A
Example 1.1.15. If A ⊂ Ω and C = {A}, then σ(C) = {ϕ, A, Ac , Ω}.
Denition 1.1.16 (Borel σ-algebra). The σ-algebra generated by all open sets in
R is called the Borel σ-algebra on R and it is denoted by B(R).
We shall use the shorthand B := B(R).
Theorem 1.1.17. Each of the following collection of sets generates the Borel σ-
algebra B of R (that is, B = σ(C) where):

5
Stochastic Calculus Umesh Kumar

(a) C := {(a, b) : a, b ∈ R};

(b) C := {(a, b] : a, b ∈ R};
(c) C := {[a, b] : a, b ∈ R};
(d) C := {[a, ∞) : a ∈ R};
(e) C := {(−∞, a] : a ∈ R};
(f) C := {(a, ∞) : a ∈ Q};
(g) C := {A : A is compact in R};
Proof. Exercise
Denition 1.1.18 (Probability Measure). A probability measure on a σ-algebra
F is a function P : F → [0, 1] which satises:
(1) P (Ω) = 1;
(2) whenever A1 , A2 , . . . is a sequence of pairwise disjoint sets in F , then
∞
X
P (∪∞
n=1 An ) = P (An ).
n=1

We are now in a position to understand probability space as dened by Kol-

mogorov.
Denition 1.1.19. (a) A measurable space (Ω, F) is a non-empty set Ω and a σ -
algebra F of Ω.
(b) A probability space is a triplet (Ω, F, P ) where Ω is a non-empty set, F is a
σ -algebra of Ω and P is a probability measure on F .
Example 1.1.20 (Discrete Probability space: equally likely case). Let Ω =
{ω1 , . . . , ωn } be a nite set. We take F = P(Ω), the power set of Ω. We assign a
constant probability, say c, to each outcome, that is we assume that each outcome
is equally likely to happen. Since the total probability is 1, we must have
n
X
P ({ωi }) = 1,
i=1

which gives c = 1/n. Now, we can compute probability of any event A ⊂ Ω by

|A|
A ⊂ Ω, P (A) = ,
|Ω|
where |A| denotes the cardinality of A, i.e., number of elements in A. This probability
is also called uniform probability and it describes the uniform distribution on the
nite set Ω.

6
Stochastic Calculus Umesh Kumar

The above example is a particular case of the following more general example.
Example 1.1.21 (Discrete Probability space). Let Ω be a nite or countable
set. We take F = P(Ω), the power set of Ω. We assign the probability to each
outcome by
P ({ω}) = pω ,
where
and
X
pω > 0, pω = 1.
ω∈Ω
We can dene a probability measure on F by
for all A ∈ F.
X
P (A) = pω ,
ω∈A

Verify yourself that P is indeed a probability measure. Conversely, every probability

measure on a countable set is of this form.
Example 1.1.22. Let Ω = [0, 1], F = B([0, 1]) and P = Leb where Leb denotes the
Lebesgue measure on B([0, 1]). Then (Ω, F, P ) is a probability space.
Example 1.1.23. Let Ω be a set with σ-algebra F . Let a ∈ Ω. Dene δa : F → [0, 1]
by
1, if a ∈ A;

δa (A) =
0, if a ∈
/ A.
Then δa denes a probability measure on F called Dirac measure and (Ω, F, δa ) is a
probability space.
Theorem 1.1.24. (Properties of Probability Measure) Let (Ω, F, P ) be a prob-
ability space and A, B, A1 , A2 , · · · ∈ F . Then we have
(a) P (ϕ) = 0;
(b) P (A ∪ B) = P (A) + P (B) − P (A ∩ B);
(c) P (A) = 1 − P (Ac );
(d) A ⊂ B, =⇒ P (A) 6 P (B); (monotone);
∞ ∞
!
(e) P P (Ak ); (σ-subadditive);
[ X
Ak 6
k=1 k=1
∞
!
(f) if A1 ⊆ A2 ⊆ · · · , then n→∞ ;
[
lim P (An ) = P Ak
k=1
∞
!
(g) if A1 ⊇ A2 ⊇ · · · , then n→∞ .
\
lim P (An ) = P Ak
k=1

Proof. Exercise.

7
Stochastic Calculus Umesh Kumar

PROBLEMS - 1.1 : Probability Space

1. Do all the exercises of Section 1.1 and prove all the theorems left as exercises.
2. Let F be a σ -algebra on Ω and let A ⊆ Ω. Dene
H := {B ∩ A : B ∈ F}.

Show that H denes a σ -algebra on A.

3. Find the smallest σ -algebra σ(C) on Ω generated by C , if
(a) Ω = {1, 2, 3, 4} and C = {ϕ, {2, 3}};
(b) Ω = {1, 2, 3, 4} and C = {{3}, {2, 3, 4}};
(c) Ω = {a, b, c, d, e, f } and C = {{a, b}, {c, d, e}}.
(d) Ω = R and C = {(1, 2], {2}}.
4. Show by only using the Denition of the Borel-σ -algebra that the following
sets are in the Borel σ -algebra B(R):
(a) {x} for x ∈ R;
(b) [a, b], [a, b), (a, b), [a, ∞) for a < b.
(c) N;
(d) Q.
5. Let f : Ω → R be a function on a set Ω. Show that
A := {f −1 (B) : B ∈ B},

denes a σ -algebra of Ω, where f 1 (B) := {ω ∈ Ω : f (ω) ∈ B}.

6. Let A and B be subsets of a non-empty set Ω. Determine σ(C) where
(a) C := {A, B} with A ∩ B = ϕ;
(b) C := {A, B} with A ∩ B 6= ϕ.
7. Let Ω be an uncountable set, e.g. Ω = R. Dene
F := {A ⊆ Ω : A is countable or Ac is countable}.

(a) Show that F denes a σ -algebra on Ω.

(b) Show that
0, if A is countable;

P (A) :=
1, if Ac is countable.
denes a probability measure on (Ω, F)
(c) Take Ω = R. Is A := {A ⊆ R : A is nite or Ac is nite} a σ -algebra on
R?

8
Stochastic Calculus Umesh Kumar

1.2 Random Variables

Denition 1.2.1 (Random Variable). Let (Ω, F, P ) be a probability space. A

function X : Ω → R is called a random variable if for each Borel set B ∈ B, we have
X −1 (B) := {ω ∈ Ω : X(ω) ∈ B} ∈ F.

In other words, a random variable is just a F -B measurable function.

It is a standard notation to denote the set X −1 (B) by {X ∈ B}.
Example 1.2.2. Let A ⊂ R. Dene the indicator function 1A : R → R by
1 if x ∈ A;

1A (x) =
0 if x ∈
/ A.

Then the function 1A is a random variable if and only if A is a measurable subset

of R.
Proof. For each B ∈ B, it follows that
if 0 ∈/ B, 1 ∈/ B;

 ϕ
if 0 ∈/ B, 1 ∈ B;

A

1−1
A (B) =  Ac if 0 ∈ B, 1 ∈/ B;
if 0 ∈ B, 1 ∈ B.

Ω


Then, 1−1
A (B) ∈ B if and only if A ∈ B .

Example 1.2.3 (Discrete Probability space ). Let Ω be a nite or countable set and
σ -algebra F be the power set on Ω. Then every function X : Ω → R is a random
variable. (Prove it yourself.)
Denition 1.2.4. Let X be a random variable on a probability space (Ω, F, P ).
The distribution measure or law of X is the probability measure µX : B → R dened
by
µX (B) := P (X ∈ B) = P (X −1 (B)).

Exercise 1.2.5. Show that µX dened above is indeed a probability measure on

(R, B).

Note that two random variables can have the same distribution.
Denition 1.2.6 (CDF). Let X be a random variable on a probability space
(Ω, F, P ). The cumulative distribution function (CDF) of X is the function FX : R →
[0, 1] dened by
FX (x) := P (X 6 x).

9
Stochastic Calculus Umesh Kumar

Knowing the CDF of a random variable is same as knowing its distribution mea-
sure. This is because, if X has distribution measure µX , then the CDF of X is
FX (x) = µX (−∞, x]. Conversely, if we know the CDF FX of X , the distribution
can be computed as:
µX (x, y] = FX (y) − FX (x) for x < y.

For a 6 b, we have
   
1 1
µX [a, b] = lim µX a − , b = FX (b) − lim FX a − .
n→∞ n n→∞ n

Using this, µX can be obtained for any Borel set B .

Theorem 1.2.7. A function F is the cumulative distribution function of a (unique)
probability measure on (R, B) if and only if
(a) F is non-decreasing;
(b) x→−∞
lim F (x) = 0 and lim F (x) = 1;
x→∞

(c) F is right continuous.

Denition 1.2.8 (PDF). Let X be a random variable on a probability space
(Ω, F, P ). The probability density function (PDF) of X is a function f : R → R
which satises Z b
µX [a, b] = P (a 6 X 6 b) = f (x) dx.
a

From the denition, it is obvious that if FX is the CDF of a random variable X

with density f , then
Z x
FX (x) = µX (−∞, x] = f (y) dy.
−∞

Remark 1.2.9. PDF may not exist for every random variable. Even if it exists, it
may not be unique.
It follows that if FX is continuously dierentiable (except possibly at nitely
many points), then the density f exists and we have
dFX (x)
= f (x),
dx
at all points where FX is continuously dierentiable and f may be dened arbitrary
at other points.

10
Stochastic Calculus Umesh Kumar

Theorem 1.2.10. If f : R → [0, ∞) is a continuous function satisfying

Z ∞
f (x) dx = 1,
−∞

then there exists a unique probability (distribution) measure Q on B such that

Z
Q(B) = f (x) dx.
B

Example 1.2.11. Let X be a random variable with density ϕ given by

1 2
ϕ(x) := √ e−x /2 .
2π
Then X is said to follow standard normal distribution. The CDF is given by
Z x
1 2 /2
N (x) = √ e−y dy.
2π −∞

Some authors also use the notation Φ(x) = N (x).

Exercise 1.2.12. Draw the graph of the function ϕ. Also, prove that
Z ∞
ϕ(x) dx = 1.
−∞

Exercise 1.2.13. Learn more about the normal distribution. In particular, recall
about the probability distribution N (µ, σ 2 ) and its properties that you may have
learnt in your undergraduate studies.

PROBLEMS - 1.2 : Random Variables

1. Assume that in the measurable space (Ω, F) the σ -algebra is given by F =

{ϕ, Ω}. Show that if X : Ω → R is a random variable then X is constant.

2. Consider a measurable space (Ω, F), and a family of sets C such that σ(C) =
B(R). Prove that if X is any real valued function on Ω such that

X −1 (A) ∈ F for all A ∈ C,

then X is a random variable.
3. Consider a measurable space (Ω, F). Prove that a function X : Ω → R is a
random variable if and only if
{X 6 a} := {ω ∈ Ω : X(ω) 6 a}

belongs to F for each a ∈ R.

11
Stochastic Calculus Umesh Kumar

4. Suppose that X, Y are random variables dened on a measurable space (Ω, F),
and λ is a real number. Prove that X +Y , XY and λX are all random variables.
5. A function f : R → R is called Borel-measurable if
f −1 (B) ∈ B for all B ∈ B,
where B denotes the Borel-σ -algebra on R. Prove that if f : R → R is contin-
uous, then f is Borel-measurable.
6. Let (Ω, F, P ) be a probability space, X : Ω → R be a random variable and
f : R → R be Borel-measurable. Prove that f (X) (that is, f ◦ X ) is a random
variable.
7. In your undergraduate studies, you may have learnt about discrete and con-
tinuous random variables. Recall about probability mass function (PMF) of a
discrete random variable. Learn about the following distributions, write their
PMF or PDF, and nd their CDF:
(a) Discrete uniform on {x1 , . . . , xn };
(b) Bernoulli distribution;
(c) Binomial ditribution;
(d) Poisson distribution;
(e) Continuous uniform distribution on [a, b];
(f) Normal Distribution;
(g) Exponential distribution;
(h) Gamma distribution.
8. Do all the exercises of Section 1.2 and prove all the theorems left as exercises.

12
Stochastic Calculus Umesh Kumar

1.3 Expectation of a Random Variable

Let X : Ω → R be a random variable. We now dene the integral of X with respect

to the probability measure P , (sometimes called P -integral), that is
Z Z
X dP or X(ω) dP (ω).
Ω Ω

The P -integral can be dened using the following three steps as in Lesebgue integral:
1. (First Step) We rst dene the P -integral for a simple random variable. A
simple random variable is a function X : Ω → R+ of the form
n
αk 1Ak (ω),
X
X(ω) =
k=1

where αk ∈ R+ (note that R+ := [0, ∞]) and Ak ∈ F are pairwise disjoint for
k = 1, . . . , n. For such X , we dene the P -integral as
Z n
X
X(ω) dP (ω) := αk P (Ak ).
Ω k=0

2. (Second Step) We now dene P -integral of a positive random variable, that

is, a measurable X : Ω → R+ . For such X , there exists an increasing sequence
of simple random variables Xn such that
Xn (ω) 6 Xn+1 (ω) and X(ω) = lim Xn (ω) for all ω ∈ Ω.
n→∞

For such X , we dene P -integral by

Z Z
X(ω) dP (ω) := lim Xn (ω) dP (ω).
Ω n→∞ Ω

3. (Third Step) In the nal step, we dene P -integral for any random variable
X : Ω → R. We rst dene:

X + (ω) := max(X(ω), 0) and X − (ω) := max(−X(ω), 0) for all ω ∈ Ω.

It follows that X + and X − are positive random variables and X = X + − X − .

We say that X is P -integrable if
Z Z
X (ω) dP (ω) < ∞ and
+
X − (ω) dP (ω) < ∞.
Ω Ω

In this case, we dene P -integral of X by

Z Z Z
X(ω) dP (ω) = +
X (ω) dP (ω) − X − (ω) dP (ω) < ∞.
Ω Ω Ω

13
Stochastic Calculus Umesh Kumar

Denition 1.3.1 (Expectation). A random variable has nite expectation if X is

P -integrable and we dene expectation of X as
Z
E[X] := X dP.
Ω

It follows that X has nite expectation if and only if E[|X|] < ∞.

Theorem 1.3.2. (Properties of Expectation.) Let X , Y be random variables
with nite expectation and α, β ∈ R. Then
(a) E[αX + βY ] = αE[X] + βE[Y ] (Expectation is linear);
(b) if X 6 Y , then E[X] 6 E[Y ];
(c) |E[X]| 6 E[|X|];
(d) (Jensen's Inequality) Let g : R → R be a convex function. If E[|X|] < ∞, then
g(E[X]) 6 E[g(X)].

Proof. Exercise.
Theorem 1.3.3. Let X be a random variable on a probability space (Ω, F, P ) and
let g be a Borel-measurable function on R. Then,
Z
E[|g(X)|] = |g(x)| dµX (x),
R
and if this quantity is nite, then
Z
E[g(X)] = g(x) dµX (x).
R
Theorem 1.3.4. Let X be a random variable on a probability space (Ω, F, P ) and
let g be a Borel-measurable function on R. Suppose that X has a density f . Then
Z ∞
E[|g(X)|] = |g(x)|f (x) dx,
−∞

and if this quantity is nite, then

Z ∞
E[g(X)] = g(x)f (x) dx.
−∞

In particular, we have Z ∞
E[X] = xf (x) dx,
−∞
provided that Z ∞
E[|X|] = |x|f (x) dx < ∞.
−∞

14
Stochastic Calculus Umesh Kumar

Remark 1.3.5. By L1 (Ω, P ) we denote the set of all P -integrable random variables,
that is,
L1 (Ω, P ) := {X : Ω → R : E[|X|] < ∞}.
In general, we dene for 1 6 p < ∞,
Lp (Ω, P ) := {X : Ω → R : E[|X|p ] < ∞}.
From Jensen's inequality, it follows that
L2 (Ω, P ) ⊂ L1 (Ω, P ).
Denition 1.3.6 (Variance of a random variable.). Let X : Ω → R be a random
variable and E[|X|2 ] < ∞. We dene variance of X , denoted as V ar[X], by
Z
2
V ar[X] := E[(X − µ) ] = (X(ω) − µ)2 dP (ω),
Ω

where µ := E[X].
Example 1.3.7. If X follows standard normal distribution of Example 1.2.11, then
E[X] = 0 and V ar[X] = 1.

PROBLEMS - 1.3 : Expectation

1. Do all the exercises of Section 1.3 and prove all the theorems left as exercises.
2. Let P be the probability measure with density
f: R → [0, ∞), f (x) = cxα 1[0,3] (x)
for some parameter α > 1 and a constant c > 0.
(a) Determine the constant c such that f is the density of a probability mea-
sure P for each α > 1.
(b) Dene a function g : R → R by
if x ∈ [0, 1),

 7,
if x ∈ [1, 2),

3,

g(x) =
 4, if x ∈ [2, 3],
else.

0,


Give reason why g is a simple function and calculate R g(u) dP .

R

3. Let f : R → R be dened by
1 1 x−µ 2
f (x) = √ e− 2 ( σ ) ,
2π σ
where µ ∈ R and σ > 0. Let X be a random variable with probability density
function f . We say that X is normally distributed with parameters µ (called
mean) and σ 2 (called variance) and write X ∼ N (µ, σ 2 ).

15
Stochastic Calculus Umesh Kumar

(a) Prove that f is indeed a probability density function.

(b) Prove that
E[X] = µ, V ar[X] = σ 2 .

(c) Compute E[eαX ] and V ar[eαX ] where α ∈ R.

(d) Show that the moment generating function (m.g.f.) of X is given by

t ∈ R.
1 2 t2
MX (t) = eµt+ 2 σ ,

Hence or otherwise compute E[X n ] for n = 1, 2, 3, 4.

(e) Let Y := eX . We say that the random variable Y follows log-normal
distribution (that means, log Y = X is normal). Find the probability
density of Y . (Hint : First nd CDF FY of Y . Then the PDF of Y is
given by fY (y) = FY0 (y).)
(f) Find the mean and variance of Y . (You can use part (c)).
4. Find expected value and variance of a random variable X which follows respec-
tively each of the distributions mentioned in Question 7 of Problems-1.2.

16
Stochastic Calculus Umesh Kumar

1.4 Independence and Conditional Expectation

In your undergraduate studies you may have learnt about independence of events
and of random variables. In this section we study these concepts with more formal
measure theoretic approach.

1.4.1 Independence

Denition 1.4.1. Let X: Ω → R be a random variable dened on a probability

space (Ω, F, P ). The σ -algebra generated by X is dened as
σ(X) := {X −1 (B) : B ∈ B}.

Remark 1.4.2. σ(X) denes a σ -algebra on Ω (see Question 5 of Problems-1.1). It

is also obvious that X is a random variable if and only if σ(X) ⊆ F . Indeed σ(X)
is the smallest σ -algebra with respect to which X is measurable.
Example 1.4.3. Let A ∈ F and X := 1A . Then X is a random variable and
σ(X) = {ϕ, A, Ac , Ω}.

We can dene measurability of X with respect to any σ -algebra on Ω.

Denition 1.4.4. Let X :Ω→ R be a random variable and G be a σ-algebra on
Ω. We say that X is G -measurable if σ(X) ⊆ G .

A random variable X is G -measurable if and only if the information in G is

sucient to determine the value of X .
Recall that two events A and B are called independent if P (A ∩B) = P (A)P (B).
We now dene independence of two random variables X and Y .
Denition 1.4.5. Let (Ω, F, P ) be a probability space, and let G and H be sub-σ-
algebras of F . We say that
(a) G and H are independent if
P (A ∩ B) = P (A) · P (B) for all A ∈ G, B ∈ H;

(b) two random variables X and Y are independent if σ(X) and σ(Y ) are indepen-
dent;
(c) X is independent of the σ -algebra G if σ(X) and G are independent.
From the denition of independence of two random variables, it easily follows
that X and Y are independent if and only if
P (X ∈ B1 , Y ∈ B2 ) = P (X ∈ B1 )P (X ∈ B2 ) for all B1 , B2 ∈ B.

Exercise 1.4.6. Prove this statement.

17
Stochastic Calculus Umesh Kumar

We can extend this denition to more than two σ -algebras or more than two
random variables also. Further, a sequence G1 , G2 , . . . of sub-σ -algebras of F is inde-
pendent if whenever Ai ∈ Gi (i ∈ N) and i1 , . . . , in are distinct, then
n
Y
P (Ai1 ∩ · · · ∩ Ain ) = P (Aik ).
k=1

Similarly, we can dene the independence of events as follows:

Denition 1.4.7. Events A1 , A2 , . . . are independent if the generated σ -algebras
σ(A1 ), σ(A2 ), . . . are independent where

σ(Ai ) := σ({Ai }) = {ϕ, Ai , Aci , Ω} for i = 1, 2, . . . .

It is clear from this denition that events A1 , A2 , . . . are independent if and only
if the random variables 1A1 , 1A2 , . . . are independent. (Prove it!)
It also follows easily that two events A and B are independent if and only if
P (A ∩ B) = P (A) · P (B).
Exercise 1.4.8. What does independence of three events A, B, C of Ω mean in terms
of probability?
Theorem 1.4.9. Let X and Y be independent random variables, and let f and g
be Borel-measurable functions on R. Then f (X) and g(Y ) are independent random
variables.
Proof. Exercise.

1.4.2 Joint Distribution of a Random Vector

Let X and Y be random variables. The pair of random variables (X, Y ) takes values
in the plane R2 , that is it is a measurable map
(X, Y ) : Ω → R2 .

The joint distribution measure of (X, Y ) is given by

µX,Y (C) = P ({(X, Y ) ∈ C}) for all Borel sets C ∈ R2 .

This is indeed a probability measure on (R2 , B(R2 )).

The joint cumulative distribution function of (X, Y ) is
FX,Y (x, y) = µX,Y ((−∞, x] × (−∞, y]) = P (X 6 x, Y 6 y), (x, y) ∈ R2 .

We say that a non-negative, Borel-measurable function fX,Y (x, y) is a joint den-

sity for the pair of random variables (X, Y ) if
Z
µX,Y (C) = fX,Y (x, y) dy dx for all Borel sets C ∈ R2 .
C

18
Stochastic Calculus Umesh Kumar

Thus, we have
Z a Z b
FX,Y (a, b) = fX,Y (x, y) dy dx, a, b ∈ R .
−∞ −∞

If the joint density fX,Y exists, then the marginal densities exist and are given by
Z ∞ Z ∞
fX (x) := fX,Y (x, y) dy and fY (y) := fX,Y (x, y) dx.
−∞ −∞

Similarly marginal distribution measures and marginal CDF's can be dened.

Denition 1.4.10. The moment generating function MX of a random variable X
is dened by
MX (t) = E[etX ] for t ∈ R,
if the expectation is nite.
Theorem 1.4.11. Let X and Y be random variables. The following conditions are
equivalent:
(a) X and Y are independent.
(b) The joint distribution measure factors:
µX,Y (A × B) = µX (A) · µY (B) for all Borel subsets A, B ∈ B.

(c) The joint cumulative distribution function factors:

FX,Y (a, b) = FX (a) · FY (b) for all a ∈ R, b ∈ R .

(d) The joint moment-generating function factors:

E[euX+vY ] = E[euX ]·E[evY ] for all u ∈ R, v ∈ R for which the expectations are nite.

If there is a joint density, each of the conditions above is equivalent to the

following.
(e) The joint density factors:
fX,Y (x, y) = fX (x) · fY (y) for almost every x ∈ R, y ∈ R .
The conditions above imply but are not equivalent to the following:
(f) The expectation factors:
E[XY ] = E[X] · E[Y ],

provided E[|XY |] < ∞.

19
Stochastic Calculus Umesh Kumar

Denition 1.4.12. If X and Y are two random variables, we dene covariance of

X and Y as
Cov(X, Y ) := E[(X − E[X])(Y − E[Y ])].

By linearity of expectation, it follows that

Cov(X, Y ) := E[XY ] − E[X] E[Y ].

From the Statement (f) of Theorem 1.4.11, it is clear that if X and Y are indepen-
dent, then Cov(X, Y ) = 0 but the converse may not be true.
Exercise 1.4.13. Give an example of random variables X and Y which satisfy
(f) of Theorem 1.4.11 holds) but X and Y are not
Cov(X, Y ) = 0 (equivalently,
independent.

1.4.3 Conditional Expectation

Before dening conditional expectation, we note the following very important equiv-
alent idea of almost everywhere (a.e.) concept of measure theory for probability
theory.
Denition 1.4.14. We say that a property holds P -a.s. (that is, P almost surely)
if it is true for all ω in a set of probability 1. For example, we say that X = Y ,
P -a.s., if
P (X = Y ) = P {ω : X(ω) = Y (ω)} = 1.

Denition 1.4.15 (Conditional Expectation). Let (Ω, F, P ) be a probability

space and X : Ω → R be a random variable. Let G be a sub-σ -algebra of F . A
random variable Z is called conditional expectation of X given G if
(a) Z is G -measurable and E[|Z|] < ∞;
(b) dP for all A ∈ G .
R R
AZ dP = AX

We use the notation E[X|G] = Z.

Remark 1.4.16. Note that such a random variable Z always exists by Radon-
Nikodym Theorem. Further, if is any other random variable satisfying the condi-
Z0
tions of above denition, then Z 0 = Z , P -a.s.
The conditional expectation E[X|G] is the estimate of X based on the information
in G .
Denition 1.4.17. For any two random variables X and Y , we dene E[X|Y ] :=
E[X|σ(Y )].

Theorem 1.4.18 (Properties of Conditional Expectation). Let (Ω, F, P ) be a

probability space, X, Y be random variables and G be a sub-σ-algebra of F .

20
Stochastic Calculus Umesh Kumar

(i) E[E[X|G]] = E[X].

(ii) If X is G -measurable, then E[X|G] = X .
(iii) (Linearity) E[c1 X + c2 Y |G] = c1 E[X|G] + c2 E[X|G], for all c1 , c2 ∈ R.
(iv) (Tower property) If H is a sub-σ-algebra of G , then
E [E[X|G]|H] = E[X|H].

(v) (Taking out what is known) If XY is integrable and Y G -measurable, then

E[Y X|G] = Y E[X|G].

(vi) (Independence) If X is independent of G , then

E[X|G] = E[X].

In particular, if X and Y are independent random variables, then E[X|Y ] =

E[X].

(vii) (Conditional Jensen's inequality) If g : R → R is a convex function, then

g(E[X|G]) 6 E[g(X)|G].

Proof. Exercise
Remark 1.4.19. Note that if G = {ϕ, Ω}, then E[X|G] = E[X].
Lemma 1.4.20 (Independence Lemma). If X is G -measurable and Y is inde-
pendent of G , then
E[f (X, Y )|G] = g(X),
where g(x) := E[f (x, Y )].
If (X, Y ) has a joint density and X is P -integrable, then E[X|Y ] can be computed
using the following theorem.
Theorem 1.4.21. Let X and Y be random variables with joint density function
f : R2 → R+ .
Assume that h : R → R is a measurable function such that E[|h(X)|] <
∞. Dene the
function g : R → R by
if fY (y) > 0,
 1
R∞
g(y) := fY (y) −∞ h(x)f (x, y) dx,
0, else.
where fY denotes the marginal density of Y . Then it follows that
E[h(X)|Y ] = g(Y ).

21
Stochastic Calculus Umesh Kumar

PROBLEMS - 1.4: Independence and Conditional Expec-

tation

Let (Ω, F, P ) be a probability space.

1. Fix an event B ∈ F such that P (B) > 0. Prove that the function P (·|B) : F →
R dened by
P (A ∩ B)
P (A|B) = , A ∈ F,
P (B)
is a probability measure on (Ω, F).
2. Assume that A ∈ F satises P (A) = 0 or P (A) = 1. Show that A is indepen-
dent of A.
3. Prove that if A∩B = ϕ, then A and B cannot be independent unless P (A) = 0
or P (B) = 0.
4. For sets A, B, C ∈ F , show the following:
(a) A and B are independent if and only if A and B c are independent.
(b) If P (B) = 0 or P (B) = 1, then A and B are independent.
(c) If A, B and C are independent, then A ∪ B and C are independent.
5. Show that if X and Y are independent random variables in L2 (Ω, P ), then
V ar[X + Y ] = V ar[X] + V ar[Y ].

6. Let X : Ω → R and Y : Ω → R be random variables with the joint probability

density
c(x + y), if x, y ∈ [0, 1],

f: R → R , f (x, y) =
2 +
0, else,
for a constant c ∈ R.
(a) Determine the constant c.
(b) Calculate the probability densities of X and Y .
(c) Calculate the joint distribution function.
(d) Calculate the covariance of X and Y .
7. Let X and Y be independent random variables which are exponentially dis-
tributed with parameter α > 0. Calculate the density of X + Y .
8. If X is an integrable random variable and G is a sub-σ -algebra of F . Show
that E[X|X] = X .
9. If Y = α P -a.s., with α a constant, show that E[Y |G] = α.

22
Stochastic Calculus Umesh Kumar

10. Let (X, Y ) : Ω → R2 be a random variable with joint density

f: R2 → R+ , f (x, y) = xe−x(y+1) 1[0,∞) (x)1[0,∞) (y).
(a) Find the densities of X and Y .
(b) Determine a version of the conditional expectation E[Y |X].
11. Assume that X : Ω → R is a random variable with P (X > 0) = 1 and E[X] =
1. Dene a mapping by

Q : F → R, Q(A) = E[1A X].

(a) Show that Q is a probability measure on (Ω, F).

(b) Show that for each A ∈ F we have the equivalence:
P (A) = 0 ⇐⇒ Q(A) = 0.

(c) Show that a random variable Y is Q-integrable if and only if Y X is

P -integrable.
In this case, it follows E Q [Y ] = E[Y X], where E Q [Y ] denotes the expec-
tation w.r.t. Q.
(d) Show that the random variable X 1 is Q-integrable.
(e) Dene a mapping by
Z
R : F → R, R(A) = 1A (ω)X −1 (ω) dQ(ω).
Ω

Conclude that P = R.

23
Chapter 2

Brownian Motion and Martingales

2.1 Brownian Motion

Denition 2.1.1 (Stochastic Process). Let I be a subset of [0, ∞). A stochastic

process is a family (X(t) : t ∈ I) of random variables X(t) : Ω → R for each t ∈ I .
Remark 2.1.2. It should be noted that a stochastc process (X(t))t∈I depends on
two quantities: t ∈ I and ω ∈ Ω. In other words, we can consider a stochastic process
as a function of two variables X(t, ω) := X(t)(ω). For each t ∈ I and ω ∈ Ω, we have
X(t, ω) ∈ R. If we x t, the map ω 7→ X(t, ω) is a random variable, that is,for xed
t ∈ I , X(t) : Ω → R is a random variable. For a xed ω ∈ Ω, the map t 7→ X(t, ω)
is called a trajectory or path of the stochastic process.
If I ⊂ N, we say the stochastic process is discrete-time. In this course, we will
usually take, I = [0, T ] for some T > 0 or I = [0, ∞) and in these cases, we say
that X is continuous-time stochastic process. If the trajectories t → X(t)(ω) are
continuous for almost all ω ∈ Ω, we say that the stochastic process is a continuous
stochastic process.
Denition 2.1.3 (Brownian Motion). A stochastic process (W (t) : t > 0) is
called a (standard) Brownian motion or Wiener Process if it satises the following:
(a) W (0) = 0;
(b) the increments
W (t2 ) − W (t1 ), W (t3 ) − W (t2 ), . . . , W (tn ) − W (tn−1 ),

are independent for 0 6 t1 < t2 < t3 · · · < tn and all n ∈ N.

(c) the increments are normally distributed:
W (t) − W (s) ∼ N (0, t − s) for all 0 6 s < t;

(d) for each ω ∈ Ω, the map t 7→ X(t)(ω) is continuous.

24
Stochastic Calculus Umesh Kumar

Some properties of Brownian Motion.

Let (W (t) : t > 0) be a Brownian motion. The following properties easily follow
from the denition:
(1) E[W (t) − W (s)] = 0 and E[|W (t) − W (s)|2 ] = t − s.
(2) W (t) ∼ N (0, t) for each t > 0.
(3) W (t) − W (s) ≡ W (t − s) in Law.
x2
(4) P (a < W (t) < b) = e− 2t dx.
Rb
√1
2π σ a
√
(5) P (W (t) < b) = N (b/ t) where N denotes CDF of standard normal.
(6) E[eαW (t) ] = e 2 α t .
1 2

(7) E[(W (b) − W (a))(W (d) − W (c))] = 0 for any 0 6 a < b < c < d.
(8) E[W (s)W (t)] = min{s, t} for any s, t > 0
(9) The sample paths of the standard Brownian motion are nowhere dierentiable
functions, P -a.s. (Proof not required).
Exercise 2.1.4. Prove properties (1)-(8) given above.
Example 2.1.5. Let W = (W (t) : t > 0) be a Brownian motion. Then the process
W = (W (t) : t > 0) is also Brownian motion in each of the following cases:
f f

(a) W
f (t) := −W (t);

f (t) := 1 W (α2 t) where α > 0 is a constant;

(b) W α

(c) W
f (t) := W (t + a) − W (a) where a > 0 is a xed constant.

Proof. Exercise.

PROBLEMS-2.1: Brownian Motion

(Note: In the following, by a Brownian motion we always mean a one-dimensional

standard Brownian Motion.)
1. Prove that a Brownian Motion (W (t) : t > 0) has stationary increments, that
is, for any 0 6 s < t, we have
W (t + h) − W (s + h) ≡ W (t) − W (s) in Law (i.e. distribution),

for all h > 0.

25
Stochastic Calculus Umesh Kumar

2. (a) If X ∼ N (0, σ 2 ), prove that

if k is odd;
(
k 0
E[X ] = k!
2k/2 (k/2)!
σk if k is even.

(Hint : Use integration by parts and induction.)

(b) Using part (a) or otherwise, prove that a Brownian motion (W (t) : t > 0)
satises
E[(W (t) − W (s))3 ] = 0 and E[(W (t) − W (s))4 ] = 3(t − s)2 .

(c) Prove that for each α ∈ N, we have

E[|W (t) − W (s)|2α ] 6 c|t − s|α ,

where c is a constant depending on α.

(This property is used to show that a Brownian motion has Holder con-
tinuous paths of every order β < 1/2.)
3. Let (W (t) : t > 0) be a Brownian motion. Consider any times 0 6 r < s < t.
Calculate
(a) E[(W (t) − W (r))(W (s) − W (r))2 ]
(b) E[W (r)W (s)W (t)]
4. Let (W (t) : t ∈ [0, T ]) be a Brownian motion. Prove that W
f dened by

f (t) := W (T ) − W (T − t),
W t ∈ [0, T ],

is a Brownian motion.
5. Let (W (t) : t > 0) be a Brownian motion. Prove that the following processes
f (t) : t > 0) are Brownian motion
(W

(a) W
f (t) = √1 W (ct);
c

(b) W
f (t) = tW (1/t) for t > 0 and W
f (t) = 0 for t = 0.

26
Stochastic Calculus Umesh Kumar

2.2 Quadratic Variation of a Brownian Motion

We recall the rst variation of a function. Let f be a real-valued function dened on

[0, T ]. Let Π = {t0 , t1 , . . . , tn } be a partition of [0, T ], that is,

0 = t0 < t1 < · · · < tn = T.

The mesh of partition Π is dened by

kΠk = max (ti+1 − ti ).
i=0,...,n

We dene the rst variation of f up to time T by

n−1
X
F VT (f ) := lim |f (ti+1 ) − f (ti )|.
kΠk→0
i=0

If f ia an increasing function, then F VT (f ) = f (T ) − f (0). If the derivative f 0 (t)

of f (t) is dened everywhere, then it follows by a simple application of mean-value
theorem that Z T
F VT (f ) = |f 0 (t)| dt.
0
The quadratic variation of f up to time T is dened by
n−1
X
[f, f ](T ) := lim |f (ti+1 ) − f (ti )|2 .
kΠk→0
i=0

If f has continuous derivative, then [f, f ](T ) = 0.

We can similarly dene the rst variation and quadratic variation (process) for
stochastic processes. In particular, we dene the quadratic variation of a Brownian
motion W as
n−1
X
[W, W ](T ) := lim |W (ti+1 ) − W (ti )|2 .
kΠk→0
i=0

Theorem 2.2.1. Let W be a Brownian motion. Then [W, W ](T ) = T for all T >0
almost surely.
Proof. Let T > 0 and Π == {t0 , t1 , . . . , tn } be a partition of [0, T ]. Dene
n−1
X
QΠ = |W (ti+1 ) − W (ti )|2 .
i=0

We show that E[QΠ ] = T and V ar[QΠ ] → 0 which will imply that QΠ → T in the
mean-square sense as kΠk → 0. More specically, it will follow that
 !2 
n−1
X
lim E[|QΠ − T |2 ] = lim E  |W (ti+1 ) − W (ti )|2 − T  = 0.
kΠk→0 kΠk→0
i=0

27
Stochastic Calculus Umesh Kumar

We have
"n−1 #
X
E[QΠ ] = E |W (ti+1 ) − W (ti )|2
i=0
n−1
X h i
= E |W (ti+1 ) − W (ti )|2
i=0
n−1
X
= (ti+1 − ti ) = T.
i=0

Using independent increment property of Brownian motion, it follows that

"n−1 #
X 2
V ar[QΠ ] = V ar |W (ti+1 ) − W (ti )|
i=0
n−1
X
= V ar[|W (ti+1 ) − W (ti )|2 ]
i=0
n−1 h
X
2 i
= E |W (ti+1 ) − W (ti )|2 − (ti+1 − ti )
i=0
n−1
X
E |W (ti+1 ) − W (ti )|4 − 2(ti+1 − ti )|W (ti+1 ) − W (ti )|2 + (ti+1 − ti )2
 
=
i=0
n−1
X
E |W (ti+1 ) − W (ti )|4 − 2(ti+1 − ti )E[|W (ti+1 ) − W (ti )|2 ] + (ti+1 − ti )2
 

=
i=0
n−1
X
3(ti+1 − ti )2 − 2(ti+1 − ti )2 + (ti+1 − ti )2


=
i=0
n−1
X
=2 (ti+1 − ti )2
i=0
n−1
X
6 2kΠk (ti+1 − ti )
i=0
= 2kΠkT → 0 as kΠk → 0.

Remark 2.2.2. We proved that QΠ → T in mean-square sense (that is L2 -convergence)

but mean square convergence implies almost sure convergence along a subsequence.
Theorem 2.2.3. With probability one, a Brownian motion is of innite variation
on every interval [0, T ], that is
P (F VT (W ) = ∞ for every T > 0 ) = 1.

28
Stochastic Calculus Umesh Kumar

Remark 2.2.4. Following Itô Calculus rules follow:

dW (t) dW (t) = dt, dW (t) dt = 0, dt dt = 0.

2.3 Martingales

Denition 2.3.1 (Filtration). A family {F(t)}t∈I of σ-algebras F(t) for all t ∈ I

with
F(s) ⊆ F(t) for all s 6 t and s, t ∈ I
is called a ltration.
It can be interpreted as information owing in the market.
Denition 2.3.2 (Adapted Process). A stochastic process (X(t) : t ∈ I) is called
adapted with respect to a ltration {F(t)}t∈I if X(t) is F(t)-measurable for every
t ∈ I.

In Finance, a portfolio ∆(t) is assumed to be adapted process with respect to a

ltration {F(t)}t∈I of information in market. That is, the value ∆(t) must depend
on the information available at time t only.
Example 2.3.3. Let (W (t) : t > 0) be a Brownian motion. For each t ∈ I , dene
F W (t) := σ(W (s), s 6 t).

Then {F W (t)}t∈I is a ltration called the (natural ) ltration generated by Brownian

motion W .
In the same way, one can dene ltration generated by any stochastic process.
Denition 2.3.4 (Martingale). Let {F(t)}t∈I be a ltration. An adapted stochas-
tic process (M (t) : t ∈ I) is called a martingale with respect to {F(t)}t∈I if
(i) E[|M (t)|] < ∞ for all t ∈ I ;
(ii) E[M (t)|F(s)] = M (s) for all 0 6 s 6 t and s, t ∈ I .
Analogously, one denes submartingales and supermartingales by replacing (ii)
in the above denition respectively by
(ii)* E[M (t)|F(s)] > M (s) for all 0 6 s 6 t and s, t ∈ I .
(ii)** E[M (t)|F(s)] 6 M (s) for all 0 6 s 6 t and s, t ∈ I .
Remark 2.3.5. A martingale has constant expectation.
Example 2.3.6. Let (W (t) : t > 0) be a Brownian motion. The following processes
(M (t) : t > 0) are martingales with respect to the ltration {F W (t)}t>0

29
Stochastic Calculus Umesh Kumar

(a) M (t) = W (t);

(b) M (t) = eαW (t)− 2 α t .
1 2

(c) M (t) = cos(βW (t))e 2 β t .

1 2

Exercise 2.3.7. Does M (t) = W 2 (t) dene a martingale w.r.t. {F W (t)}t>0 ? What
about M (t) = W 2 (t) − t?

Remark 2.3.8. From the above example, it is clear that a Brownian Motion is a
martingale with respect to its natural (i.e. generated) ltration.
Example 2.3.9. From Exercise 5 of Problems 2.1, the following processes (M (t) :
t > 0) where

(i) M (t) = √1 W (ct);

c

(ii) M (t) = tW (1/t) for t > 0 and M (t) = 0 for t = 0;

are Brownian motions and hence are martingales with respect to their generated
ltrations {F M (t)}t>0 .

PROBLEMS-2.2 : Martingales

1. Let (F(t))t∈I be a ltration and let X be a random variable satisfying E[|X|] <
∞. If we dene
M (t) = E[X|F(t)], t ∈ I,
then prove that M = (M (t) : t ∈ I) is a martingale.
2. Let (M (t) : t ∈ I) be an (F(t))t∈I -martingale, such that E[|M (t)|2 ] < ∞.
Show that, for all s < t,
(a) E[(M (t) − M (s))2 |F(s)] = E[(M (t))2 |F(s)] − M 2 (s);
(b) E[(M (t) − M (s))2 ] = E[(M (t))2 ] − E[(M (s))2 ]; (Hint: use the tower
property)
(c) The function t 7→ E[(M (t))2 ] is increasing.
3. Check whether the following stochastic processes (M (t) : t > 0) are martingales
with respect to the generated ltration {F W (t)}t>0 where
(a) M (t) = W 3 (t) − 3tW (t);
(b) M (t) = t2 W (t) − 2 0t sW (s) ds;
R

(c) M (t) = W 4 (t) − 4t2 W (t);

c2
(d) M (t) = e− 2 t cosh(cW (t)).

30
Chapter 3

Stochastic Calculus
3.1 Itô Integral

By an Itô stochastic integral we mean an expression of the form 0T ∆(t) dW (t),

R

where (W (t) : t ∈ [0, T ]) is a Brownian motion and (∆(t) : t ∈ [0, T ]) is a stochastic

process adapted to the ltration of the Brownian motion. In this section, we give
meaning to such an expression. Recall that the Riemann integral of a function is
dened by
Z T n−1
X
f (t) dt := lim f (si )(ti+1 − ti ),
0 kΠk→0
i=0

where si ∈ [ti , ti+1 ]. The limit is independent of the choice of partitions and is also
independent of the values si . The Riemann-Stieltjes integral of f with respect to a
function g is dened by
Z T n−1
X
f (t) dg(t) := lim f (si )(g(ti+1 ) − g(ti )),
0 kΠk→0
i=0

where si ∈ [ti , ti+1 ]. The limit is independent of the choice of partitions and is
also independent of the values si . It is well-known that if the Riemann-Stieltjes
integral 0T f (t) dg(t) exists for all continuous functions f , then g must be of bounded
R

variation. Since the trajectories of a Brownian motion are of innite variation almost
surely, we cannot dene the stochastic integral 0T ∆(t) dW (t) as Riemann-Stieltjes
R

integral pathwise.
Example 3.1.1. Consider W (t) dW (t). For a partition Π = {t0 , t1 , . . . , tn } of
RT
0
[0, T ], we can dene the upper Riemann-Stieltjes sum
n−1
X
U (W, Π) := W (ti+1 )(W (ti+1 ) − W (ti )),
i=0

31
Stochastic Calculus Umesh Kumar

and the lower Riemann-Stieltjes sum

n−1
X
L(W, Π) := W (ti )(W (ti+1 ) − W (ti )).
i=0

Then,
n−1
X
U (W, Π) − L(W, Π) = (W (ti+1 ) − W (ti ))2 ,
i=0

and using Theorem 2.2.1, we obtain

n−1
X
lim |U (W, Π) − L(W, Π)| = lim (W (ti+1 ) − W (ti ))2 = [W, W ](T ) = T.
kΠk→0 kΠk→0
i=0

This shows that we cannot dene 0 W (t) dW (t) as Riemann-Stieltjes integral path-
RT

wise as the limit depends on the choice of values si ∈ [ti , ti+1 ].

Itô integral is dened in the following way
Z T n−1
X
∆(t) dW (t) = lim ∆(ti )(W (ti+1 ) − W (ti )).
0 kΠk→0
i=0

In the following we make it more precise.

3.1.1 Itô Integral for simple process

Let (W (t) : t ∈ [0, T ]) be a Brownian Motion with a ltration (F(t))t∈[0,T ] . Let

(∆(t) : t ∈ [0, T ]) be a stochastic process which satises

∆(t) = ∆(tk ) for t ∈ [tk , tk+1 ), k = 0, 1, . . . , n − 1,

and ∆(T ) = ∆(tn−1 ), where 0 = t0 < t1 < · · · < tn = T , such that ∆(tk ) are
F(tk )-measurable and satisfy E[|∆(tk )|2 ] < ∞. Such a process ∆ will be called a
simple process. We can write it as
n−1
1[tk ,tk+1 ) (t)∆(tk ) + 1{T } (t)∆(tn−1 ).
X
∆(t) =
k=0

We dene the Itô Integral for the simple process ∆ by

Z T n−1
X
∆(t) dW (t) := ∆(tk )(W (tk+1 ) − W (tk )).
0 k=0

For a xed T, we will write I(∆) = ∆(t) dW (t). Clearly, I(∆) is a random
RT
0
variable.

32
Stochastic Calculus Umesh Kumar

Theorem 3.1.2 (Properties of Itô Integral for simple processes). For a simple pro-
cess ∆, we have
(a) E[I(∆)] = E ∆(t) dW (t) = 0;
hR i
T
0
 2 
(b) (Itô Isometry).
hR i
T T
E[|I(∆)|2 ] = E 0 ∆2 (t) dt
R
=E 0 ∆(t) dW (t)

Proof. In Class.
Remark 3.1.3. If ∆(t) is deterministic (i.e. non-random), then it is easy to see that
 Z T 
2
I(∆) ∼ N 0, ∆ (t) dt .
0

3.1.2 Itô Integral for general process

We now assume that (∆(t) : t ∈ [0, T ]) is an adapted stochastic process and satises
Z T 
k∆k := E 2
∆ (t) dt < ∞. (3.1.1)
0

For such ∆, we can nd a sequence (∆n ) of simple stochastic processes such that
∆n → ∆ as n → ∞. Indeed, we have

lim k∆n − ∆k → 0.
n→∞

We dene Itô integral of ∆ by

Z T Z T
I(∆) = ∆(t) dW (t) := lim ∆n (t) dW (t),
0 n→∞ 0

where the limit is in the mean square sense. More precisely, there exists a square
integrable random variable I(∆) such that
lim E |I(∆n ) − I(∆)|2 = 0.
 
n→∞

It can be proved that the denition does not depend on the approximating sequence
∆n .
Theorem 3.1.4 (Properties of Itô Integral). For any adapted process ∆ which sat-
ises (3.1.1), we have
(a) E[I(∆)] = E ∆(t) dW (t) = 0;
hR i
T
0

(b) Itô Isometry:

"Z
T 2 # Z T 
2 2
E[|I(∆)| ] = E ∆(t) dW (t) =E ∆ (t) dt .
0 0

33
Stochastic Calculus Umesh Kumar

3.1.3 Itô Integral Process

For any t ∈ [0, T ], we dene the Itô integral ∆(s) dW (s) by

Rt
0
Z t Z T
∆(s) dW (s) := 1[0,t] (s)∆(s) dW (s).
0 0

Thus, we obtain the Itô Integral as a stochastic process

Z t 
∆(s) dW (s) : t ∈ [0, T ] .
0

In the following results, given a general process ∆, we will use the notation
Z t
I(t) := ∆(s) dW (s).
0

Theorem 3.1.5 (Properties of Itô Integral Process). The following properties are
satised:
(a) The Itô integral process (I(t) : t ∈ [0, T ]) has continuous trajectories, that is, the
map t 7→ I(t)(ω) is continuous pathwise;
(b) I(t) is F(t)-measurable for each t ∈ [0, T ], that is (I(t) : t ∈ [0, T ]) is an adapted
process;
(c) The Itô integral process (I(t) : t ∈ [0, T ]) is a martingale;
(d) The quadratic variation accumulated upto time t of the Itô integral process (I(t) :
t ∈ [0, T ]) is
Z t
[I, I](t) = ∆2 (s) ds;
0

(e) E[I(t)] = 0;
(f) Itô Isometry: Z t 
2 2
E[I(t) ] = E ∆ (t) dt .
0

Remark 3.1.6. Since I(t) = ∆(s) dW (s), we can write it as

Rt
0

dI(t) = ∆(t) dW (t).

Then (by Itô multiplication rules),

dI(t) dI(t) = ∆2 (t) dt,

which is another way of saying that

Z t
[I, I](t) = ∆2 (s) ds.
0

34
Stochastic Calculus Umesh Kumar

R t Note that dI(t) = ∆(t) dW (t) is a notation in dierential form to mean I(t) =
0 ∆(s) dW (s). Indeed, dI(t) = ∆(t) dW (t) actually means
Z t
I(t) = I(0) + ∆(s) dW (s),
0

where I(0) is an initial condition. For, I(t) = ∆(s) dW (s), we have I(0) = 0.
Rt
0

Example 3.1.7. Let us now compute the Itô integral 0T W (t) dW (t). Let Π =
R

{t0 , · · · , tn } be a partition of [0, T ]. Using

1 1
W (ti ) = (W (ti+1 ) + W (ti )) − (W (ti+1 ) − W (ti )),
2 2
we get
n−1 n−1 n−1
X 1X 2 1X
W (ti )(W (ti+1 ) − W (ti )) = (W (ti+1 ) − W 2 (ti )) − (W (ti+1 ) − W (ti ))2
2 2
i=0 i=0 i=0
n−1
1 1 X
= (W 2 (tn ) − W 2 (0)) − (W (ti+1 ) − W (ti ))2
2 2
i=0
n−1
1 1 X
= W 2 (T ) − (W (ti+1 ) − W (ti ))2 .
2 2
i=0

Hence, using Theorem 2.2.1, we obtain

Z T n−1
X
W (t) dW (t) = lim W (ti )(W (ti+1 ) − W (ti ))
0 kΠk→0
i=0
n−1
1 1 X
= W 2 (T ) − lim (W (ti+1 ) − W (ti ))2
2 2 kΠk→0
i=0
1 1
= W 2 (T ) − T.
2 2
Remark 3.1.8. Note that if f is a dierentiable function with f (0) = 0, then we
can evaluate the Riemann-Stieltjes integral
Z T Z T Z T  
0 d 1 2 1
f (t) df (t) = f (t)f (t) dt = f (t) dt = f 2 (T ).
0 0 0 dt 2 2

Note the occurrence of an extra term in the Itô integral in above example.

35
Stochastic Calculus Umesh Kumar

PROBLEMS 3.1 - Itô Integral

1. Show that
n−1
1 1
(a) lim
X
W (ti+1 )(W (ti+1 ) − W (ti )) = W 2 (T ) + T ;
kΠk→0 2 2
i=0
n−1
X  W (ti+1 ) + W (ti )  1
(b) lim (W (ti+1 ) − W (ti )) = W 2 (T ).
kΠk→0 2 2
i=0

2. Dene a stochastic process (Φ(t) : t ∈ [0, 4]) by

if t ∈ [0, 1],

 2,
if t ∈ (1, 2],

W (1),

Φ(t) =
 W (1.5), if t ∈ (2, 3],
if t ∈ (3, 4].

0,


Give a reason why (Φ(t) : t ∈ [0, 4]) is an adapted stochastic process. Write
the stochastic integral Z 4
Φ(t) dW (t)
0
as the sum of three random variables. Calculate the mean and variance of the
stochastic integral Z 4
Φ(t) dW (t).
0

3. Verify the equality

Z T Z T
t dW (t) = T W (T ) − W (t) dt,
0 0

by computing the Itô integral from the denition. (The integral on the R.H.S.
can be understood as Riemann integral pathwise.)

36
Stochastic Calculus Umesh Kumar

3.2 Itô Formula

In standard calculus, we have the chain rule

d
f (g(t)) = f 0 (g(t))g 0 (t),
dt
that is,
df (g(t)) = f 0 (g(t)) dg(t).
But in Itô Calculus, we have the following chain rule.
Theorem 3.2.1. If f: R → R is twice continuously dierentiable function, then
for each t > 0,
Z t Z t
1
f (W (t)) = f (0) + 0
f (W (s)) dW (s) + f 00 (W (s)) ds. (3.2.1)
0 2 0

Example 3.2.2. For f (x) = x2 , we obtain

Z t Z t Z t
1
W 2 (t) = 2W (s) dW (s) + 2 ds = 2 W (s) dW (s) + t.
0 2 0 0

This also veries that

Z t
1 1
W (s) dW (s) = W 2 (t) − t.
0 2 2
Example 3.2.3. For f (x) = ex , we obtain
Z t Z t
W (t) W (s) 1
e =1+ e dW (s) + eW (s) ds.
0 2 0

Remark 3.2.4. The equation (3.2.1) in Theorem 3.2.1 can be written in dierential
form as follows:
1
df (W (t)) = f 0 (W (t)) dW (t) + f 00 (W (t)) dt. (3.2.2)
2
Theorem 3.2.5 (Itô Formula). Let f (t, x) be twice dierentiable in x and once
dierentiable in t and (W (t) : t > 0) be a Brownian Motion. Then for each t > 0,
Z t Z t
1 t
Z
f (t, W (t)) = f (0, 0)+ fs (s, W (s)) ds+ fx (s, W (s)) dW (s)+ fxx (s, W (s)) ds.
0 0 2 0
(3.2.3)
Remark 3.2.6. In dierential form, we can write (3.2.3) as
1
df (t, W (t)) = ft (t, W (t)) dt + fx (t, W (t)) dW (t) + fxx (t, W (t)) dt.
2
Example 3.2.7. In Black-Scholes theory, a stock price is1 assumed to follow Geo-
metric Brownian Motion (G.B.M.) given by S(t) = S0 e(µ− 2 σ . Then by Itô
2 )t+σW (t)

formula, we obtain
dS(t) = µS(t) dt + σS(t) dW (t).

37
Stochastic Calculus Umesh Kumar

3.2.1 Itô Formula for Itô Process

Denition 3.2.8. Let (W (t) : t > 0) be a Brownian motion and let (F(t))t>0 , be
an associated ltration. An Itô process (X(t) : t > 0) is a stochastic process of the
form Z t Z t
X(t) = X(0) + ∆(s) dW (s) + Θ(s) ds,
0 0
where X(0) is non-random and ∆ and Θ are adapted stochastic processes.
Remark 3.2.9. The appropriate conditions are assumed so that the integrals in the
denition of Itô process exist. In dierential notation, an Itô process is of the form
dX(t) = ∆(t) dW (t) + Θ(t) dt.

Remark 3.2.10. By Itô multiplication rules, we obtain

dX(t) dX(t) = ∆2 (t) dt.

In other words, the quadratic variation of Itô Process X is given by

Z t
[X, X](t) = ∆2 (s) ds.
0

Denition 3.2.11. Let (Γ(t) : t > 0) be an adapted process. We dene the integral
with respect to an Ito process (X(t) : t > 0) by
Z t Z t Z t
Γ(s) dX(s) = Γ(s)∆(s) dW (s) + Γ(s)Θ(s) ds.
0 0 0

Theorem 3.2.12. Let f (t, x) be twice dierentiable in x and once dierentiable in

t and (X(t) : t > 0) be an Itô process. Then for each t > 0, we have
Z t Z t
1 t
Z
f (t, X(t)) = f (0, X(0))+ ft (s, X(s)) ds+ fx (s, X(s)) dX(s)+ fxx (s, X(s)) d[X, X](s).
0 0 2 0

Remark 3.2.13. In dierential notation, we have

1
df (t, X(t)) = ft (t, X(t)) dt + fx (t, X(t)) dX(t) + fxx (t, X(t)) dX(t) dX(t).
2
Exercise 3.2.14. Suppose (S(t)) is G.B.M. given by the Itô process
dS(t) = µS(t) dt + σS(t) dW (t).

Write ln S(t), S 2 (t) and S(t) as Itô processes.

p

38
Stochastic Calculus Umesh Kumar

PROBLEMS - 3.2 : Itô Formula

1. Let S(t) follows G.B.M., that is,

dS(t) = µS(t) dt + σS(t) dW (t).

Let Y (t) = S(t) .

1
Using Itô formula write dY (t) as an SDE in terms of Y (t).
2. Show that the stochastic process (X(t) : t > 0) dened by
1
X(t) = exp( t) cos W (t)
2
satises the equation
Z t
1
X(t) = 1 − exp( s) sin W (s) dW (s) for all t > 0.
0 2

3. Let (Y (t) : t > 0) be dened by

1
Y (t) = (W (t))3 − tW (t).
3
Compute the dierential dY and Write Y as an Itô process.
4. Let dS(t) = (βS(t) + 1 − β) dW (t). Derive an SDE for X(t) = βS(t) + 1 − β
in terms of X(t).
5. Consider the Ornstein-Uhlenbeck process which satises
dY (t) = −αY (t) dt + σ dW (t),

for α > 0. Using Itô's Lemma, write down the SDE for Z(t) = eαt Y (t). Using
this, nd the solution of Ornstein Uhlenbeck process.
6. Consider the Bessel process which satises
2δ − 1
dR(t) = dt + dW (t),
R(t)

for δ > 0, R0 > 0. Write the SDE for Z(t) = (R(t))2 .

39
Chapter 4

Financial Mathematics
4.1 Black-Scholes Model

We assume the existence of a risk-free and risky asset in the nancial market. By
a risk-free asset we mean a Bond or Money Market account (M.M.A.). Let B(t)
denote the price of a Bond at time t (or the money invested in M.M.A. at time t).
Let r denote the continuously compounded risk-free interest rate in the market. The
dymanics of the Bond is given by the following dierential equation
dB(t) = rB(t) dt,
with initial condition B(0) = B0 , which denotes the price of Bond at time t = 0.
Solving this we get
B(t) = ert B0 .
The randomness in the market is modelled using a probability space (Ω, F, P ). Here,
P will be called real world probability measure. Let W be a Brownian motion dened
on (Ω, F, P ) . Let S denote a risky asset (or stock) in the market. By S(t), we denote
the price of a share of the stock at time t. The dynamics of the stock is given by the
following stochastic dierential equation (SDE):
dS(t) = µS(t) dt + σS(t) dW (t), (4.1.1)
where µ ∈ R and σ > 0. In other words, S(t) follows the Geometric Brownian
motion (G.B.M.). Here, µ will be called drift and σ will be called volatility of the
stock price.
Exercise 4.1.1. Explain why S(t) = W (t) and S(t) = µt + σW (t) are not good
models for the stock price. (It was discussed in class).
In Example 3.2.7, we saw that the process
(4.1.2)
1 2 )t+σW (t)
S(t) = S(0)e(µ− 2 σ
satises the SDE (4.1.1). Conversely, by using Itô lemma on ln S(t), we can prove
that the solution of SDE (4.1.1) is given by (4.1.2)

40
Stochastic Calculus Umesh Kumar

Exercise 4.1.2. Using Itô lemma on ln S(t) solve the SDE (4.1.1). (Discussed in
class)
It is clear from (4.1.2) that S(t) follows log-normal distribution.
Exercise 4.1.3. For any t > 0, nd the PDF of S(t). Also, nd E[S(t)] and
V ar[S(t)].

By a nancial derivative we mean a nancial instrument whose price depends on

some underlying asset, say, stock. Some examples of derivatives are forward, future,
call or put options. A forward is a contract to exchange(buy/sell) an asset (called
underlying) on a xed future date (called maturity or expiry) for a xed price (called
strike price). Let S(T ) denote the price of the underlying asset at expiry T and K
be the strike price. Then the payo of the holder (long position) i.e. the party who
will buy the asset is given by S(T ) − K . Similarly the payo of the seller (short
position) is given by K − S(T ). The holder of a forward contract is obligated to buy
the underlying asset and the seller is obligated to sell the underlying asset.
A call option is a nancial contract which gives its holder the right (but no
obligation) to buy an underlying asset by a certain future date for a xed price
(strike price). A put option is a nancial contract which gives its holder right (but
no obligation) to sell an underlying asset by a certain future date for a xed price.
The xed price in contract is called strike price or exercise price and xed date is
called maturity or expiry date of the contract. A European option is one that can be
exercised only on the maturity date but an american option can be exercised at any
time up to the maturity date. The payo of the holder (long position) of a European
call option is given by
(S(T ) − K)+ = max(S(T ) − K, 0).

The payo of the writer (short position) of a European call option is given by
−(S(T ) − K))+ = − max(S(T ) − K, 0) = min(K − S(T ), 0).

Similarly, one can write the payos of the long and short positions for a put option.
Exercise 4.1.4. Write the payos for the long position and short position of a put
option. Draw the payo diagrams for long and short positions of both call and put
options.
Since the payo of the holder of a call (or put) option is always non-negative, the
holder is always in benet position than the writer who faces potential liabilities in
the future. For this reason, the holder needs to pay the writer a premium (also called
option price) to enter into this contract. The natural question is: What should be
the fair option price so that the contract is fair to both the holder and the writer.
This is the basic question that we want to deal in this unit of this course. Before
that, we explain some basic terminologies and useful denitions:

41
Stochastic Calculus Umesh Kumar

1. Portfolio. By a portfolio we mean taking position in two or more assets.

Suppose at time t, we invest a(t) units in Bond and b(t) units in stock, then
the value X of our portfolio at time t is given by
X(t) = a(t)B(t) + b(t)S(t).

In this case, we sometimes say that the portfolio X is given by (a(t), b(t)).
2. No Arbitrage Principle. Arbitrage in a market means an opportunity to
make a positive prot with zero investment and without taking any risk. Math-
ematically, a portfolio V will produce an arbitrage opportunity if
V (0) = 0, P (V (T ) > 0) = 1 and P (V (T ) > 0) > 0.

We make an assumption that there are no arbitrage opportunities in the mar-

ket. This is called no arbitrage principle. Sometimes it is expressed in simple
words as "there is no free lunch". This assumption is very important for de-
termining fair prices. If X and V are two portfolios such that X(T ) = V (T ),
then no arbitrage principle implies that we must have X(0) = V (0).
3. Short Selling. Short selling means selling an asset that one does not own.
Mathematically, short selling an asset means taking a negative position in the
asset. Short selling a risk-free asset means taking a loan from M.M.A. at risk
free interest rate. Since short selling a risky asset can incur huge losses, it is
usually prohibited in many markets.
4. Self Financing Portfolio. A portfolio is called self-nancing if any change
in the portfolio value is due to the change in the asset prices (e.g. bond and
stock prices) only and not due to any external cash ow. If X is a portfolio
with value X(t) = a(t)B(t) + b(t)S(t) at time t, then X is self-nancing means
that
dX(t) = a(t) dB(t) + b(t) dS(t).

5. Replication Portfolio. A portfolio X is said to replicate a derivative V

if at time T its value matches exactly the payo of the derivative, that is,
X(T ) = V (T ). A replicating portfolio is sometimes used to hedge a derivative
(i.e., to cover the risk due to holding the derivative).

4.2 Black Scholes PDE: Derivation

To obtain price of an option in Black-Scholes model, we make the following assump-

tions:
• The risk-free interest rate r is constant.

42
Stochastic Calculus Umesh Kumar

• The dynamics of stock price S = S(t) is given by

dS = µS dt + σS dW (t) (4.2.1)
where µ ∈ R and σ > 0 are constants. The stock pays no dividend.
• The no arbitrage principle holds.

• Short selling is allowed and assets are divisible.

• There are no transaction costs.

• The trading of stock can take place continuously in the market.

Let V (S, t) denote the value of an option. Using Itô's lemma we have
∂V ∂V 1 ∂2V
dV = dt + dS + ( dS)2 . (4.2.2)
∂t ∂S 2 ∂S 2
Using (4.2.1), we obtain
∂V ∂V 1 ∂2V 2 2
dV = dt + (µS dt + σS dW (t)) + σ S dt
∂t ∂S 2 ∂S 2
∂2V

∂V ∂V 1 ∂V
= + µS + σ 2 S 2 2 dt + σS dW (t). (4.2.3)
∂t ∂S 2 ∂S ∂S
Consider a portfolio consisting of one option and −∆ units of underlying stock. The
value of the portfolio is given by
Π = V − ∆ S.

Assuming that the portfolio is self-nancing we obtain

dΠ = dV − ∆ dS. (4.2.4)
Using (4.2.1) and (4.2.3), we obtain
1 2 2 ∂2V
 
∂V ∂V ∂V
dΠ = + µS + σ S 2
dt + σS dW (t) − ∆(µS dt + σS dW (t))
∂t ∂S 2 ∂S ∂S
1 2 2 ∂2V
   
∂V ∂V ∂V
= + µS + σ S − ∆µS dt + σS − ∆ dW (t).
∂t ∂S 2 ∂S 2 ∂S

We can eliminate the random term by choosing ∆ = ∂S .

∂V
It then follows that
∂2V
 
∂V 1
dΠ = + σ2S 2 2 dt. (4.2.5)
∂t 2 ∂S
Then Π is riskless and by the no arbitrage principle it should grow as the bond, that
is,
dΠ = rΠ dt

43
Stochastic Calculus Umesh Kumar

= r(V − ∆ S) dt
∂V
= r(V − S ) dt. (4.2.6)
∂S
Comparing (4.2.5) and (4.2.6) we have
∂2V
 
∂V ∂V 1
r(V − S ) dt = + σ2S 2 2 dt.
∂S ∂t 2 ∂S

We nally obtain the Black-Scholes PDE

∂V ∂V 1 ∂2V
+ rS + σ 2 S 2 2 − rV = 0. (4.2.7)
∂t ∂S 2 ∂S

4.3 Black-Scholes PDE: Solution

In the last section we obtained the Black-Scholes PDE as a backward parabolic

partial dierential equation. Note that it does not contain the drift µ of the stock
price process. The price V (t, S) of the option satises this PDE along with the
terminal condition given by its payo. Thus, if the payo of the option is given by
V (T, S) = Φ(S), then the price of the option can be obtained as solution of the
following PDE
∂V ∂V 1 ∂2V
+ rS + σ 2 S 2 2 − rV = 0 (4.3.1)
∂t ∂S 2 ∂S
V (T, S) = Φ(S). (4.3.2)
One can reduce the Black-Scholes PDE into Heat equation using a series of trans-
formations and thus the solution of Black-Scholes PDE can be obtained in terms
of the solution of the Heat equation. Instead of using these arguments in detail
we just quote the following more general and famous result to write the solution of
Black-Scholes PDE.
Theorem 4.3.1 (Feynman-Kac). The solution F (t, x) of the PDE
∂F ∂F 1 ∂2F
+ µ(t, x) + σ 2 (t, x) 2 − rF = 0 (4.3.3)
∂t ∂x 2 ∂x
F (T, x) = Ψ(x), (4.3.4)
is given by
F (t, x) = E[e−r(T −t) Ψ(X(T ))|X(t) = x],
where (X(s) : s > t) is a stochastic process which satises the SDE
dX(s) = µ(s, X(s)) ds + σ(s, X(s)) dW (s),

with X(t) = x.

44
Stochastic Calculus Umesh Kumar

As an application of Feynman-Kac theorem we note that the solution V (t, S) of

the Black-Scholes equation (4.3.3) with terminal condition (4.3.4) has a stochastic
representation given by Feynman-Kac formula:
V (t, S) = E Q [e−r(T −t) Φ(S(T ))|S(t) = S], (4.3.5)
and, in particular, the price V (0) = V (0, S0 ) of the option at time t = 0 is given by
V (0) = E Q [e−rT Φ(S(T ))], (4.3.6)
where (S(t) : t > 0) is a stochastic process which satises
dS(t) = rS(t) dt + σS(t) dW Q (t) (4.3.7)
and S(0) = S0 . It must be noted that the dynamics of S given by (4.3.7) is dierent
from (4.2.1). Indeed, instead of µ as drift we have r as drift. Further, the Brownian
motion in equation (4.3.7) is denoted by W Q deliberately to stress the fact that it
is a dierent Brownian motion than that appearing in (4.2.1). In other words, W Q
is a Brownian motion not under the real world measure P but a dierent measure
Q called the risk-neutral measure. Further, the expectation appearing in (4.3.6) is
expectation under the measure Q. Thus, the option pricing formula given by (4.3.6)
states that the option price is given by "expected value of the discounted payo
under risk neutral measure". The idea of risk-neutral measure will become more
clear in Section 4.6 where we will make precise how to obtain (4.3.7) from (4.2.1) by
a change of measure.
Exercise 4.3.2. Show that the dynamics of stock price under risk-neutral measure
is given by
1 2 )t+σW Q (t)
S(t) = S(0)e(r− 2 σ .
Prove that
E Q [S(t)] = ert S(0).
Interpret this statement.

Price of a Call option

Let C = C(t, S) denote the value of a European call option on an underlying S . The
payo of call option is Φ(S(T )) = (S(T ) − K)+ . Thus the Black-Scholes PDE for a
call option can be written as
∂C ∂C 1 ∂2C
+ rS + σ 2 S 2 2 − rC = 0 (4.3.8)
∂t ∂S 2 ∂S
C(T, S) = (S(T ) − K)+ . (4.3.9)
The price of the call option is given by
C(0) = E Q [e−rT (S(T ) − K)+ ]. (4.3.10)
Computing this expectation we obtain the famous Black-Scholes formulae for price
of a European call option.

45
Stochastic Calculus Umesh Kumar

Exercise 4.3.3. Write the Black-Scholes PDE for put option and mention its ter-
minal condition. Also, write the solution as an expectation.

4.4 Black-Scholes Option Pricing Formulae

In this section we derive the Black-Scholes formula for the price of a European call
option on a non-dividend paying stock. We noted in the last section that the price
of a call option is given by
C(0) = e−rT E Q [(S(T ) − K)+ ]
= e−rT E Q [(S(T ) − K)1{S(T )>K} ], (4.4.1)
where the expectation is taken under the risk-neutral measure Q. Under Q, the
dynamics of S is given by
1 2 )t+σW Q (t)
S(t) = S(0)e(r− 2 σ ,
√
for t ∈ [0, T ]. Since W Q (T ) ∼ N (0, T ) we can write W Q (T ) = T Z , where Z ∼
N (0, 1) under Q. Thus, we have
1 2 )T +σ
√
S(T ) = S(0)e(r− 2 σ TZ
.

Also,
1 2
√
{S(T ) > K} = {S(0)e(r− 2 σ )T +σ T Z > K}
√
  
1 2 K
= (r − σ )T + σ T Z > ln
2 S(0)
∗
= {Z > d }.

where  
K
ln S(0) − (r − 21 σ 2 )T
d∗ := √
σ T
Using this in (4.4.1) and using the properties of normal distribution, the price of a
call option can be computed as
√
C(0) = e−rT E Q [(S(0)e(r− 2 σ )T +σ T Z − K)1{Z>d∗ } ]
1 2

Z ∞ √
−rT 1 2 1 2
=e (S(0)e(r− 2 σ )T +σ T x − K) √ e−1/2x dx
∗ 2π
Z d∞ √ Z ∞
1 1 2 2 1 2
= S(0) √ e− 2 σ T +σ T x e−1/2x dx − e−rT K √ e−1/2x dx
∗ 2π ∗ 2π
Zd ∞ Z ∞ d
1 − 1 (σ2 T −2σ√T x+x2 ) 1 2
= S(0) √ e 2 dx − e−rT K √ e−1/2x dx
∗ 2π d∗ 2π
Zd ∞ Z ∞
1 − 1 (x−σ√T )2 1 2
= S(0) √ e 2 dx − e−rT K √ e−1/2x dx
d∗ 2π d∗ 2π

46
Stochastic Calculus Umesh Kumar

Z ∞ Z ∞
1 1 2 1 2
= S(0) √ √ e− 2 y dy − e−rT K √ e−1/2x dx
d∗ −σ T 2π d∗ 2π
√
Z−d∗ +σ T Z −d∗
1 1 2 1 2
= S(0) √ e− 2 y dy − e−rT K √ e−1/2x dx
−∞ 2π −∞ 2π
√
= S(0)N (−d + σ T ) − e−rT K N (−d∗ ).
∗

Recall that N (x) := dy denotes the CDF of a standard normal random

Rx 2
√1 e−1/2 y
0 2π
variable. Dene √
d1 := −d∗ + σ T and d2 := −d∗ .
Hence the price of the call option is given by
C(0) = S(0)N (d1 ) − Ke−rT N (d2 ) (4.4.2)
where
 
S(0)
ln + (r + 21 σ 2 )T
K
d1 = √ (4.4.3)
σ T
√
and d2 = d1 − σ T . (4.4.4)
Equation (4.4.2) gives us the Black-Scholes formula for the price of a call option
Exercise 4.4.1. Using similar computations as above, derive the following Black-
Scholes formula for the price of a put option
P (0) = −S(0)N (−d1 ) + Ke−rT N (−d2 ) (4.4.5)
where
 
S(0)
ln + (r + 21 σ 2 )T
K
d1 = √ (4.4.6)
σ T
√
and d2 = d1 − σ T . (4.4.7)
Exercise 4.4.2. Compute the price of a digital option whose payo is given by
1S(T )>K .

4.4.1 Put-Call Parity

Note that
C(T ) − P (T ) = (S(T ) − K)+ − (K − S(T ))+
= S(T ) − K
= F (T ).

47
Stochastic Calculus Umesh Kumar

This means that payo of call minus payo of put is same as the payo of a forward.
By no arbitrage argument, we must have
C(0) − P (0) = S(0) − Ke−rT .

Indeed, for all t ∈ [0, T ], we have

C(t) − P (t) = S(t) − Ke−r(T −t) .

This means that a forward derivative can be replicated by taking a long position in
a call option and a short position in a put option. This relation is called put-call
parity. We can use the put-call parity to nd the price of put option from that of
call option. Indeed, we have
P (0) = C(0) − S(0) + Ke−rT .

Now using the Black-Scholes formula for the price of call option, we can immediately
write the Black-Scholes formula for price of put option.

4.5 Option Greeks

The Black-Scholes option price depends on factors like S , K , σ , r and T (or T −

t). The sensitivity of the option price with respect to these underlying factors are
denoted by greek letters and are called Greeks of the option.

Delta

The sensitivity of the option price with respect to stock price is given by Delta of
the option. We dene the Delta, denoted by the symbol ∆, of an option with value
V , by
∂V
∆= .
∂S
In other words, Delta of an option measures the rate of change in the option value
with respect to the change in stock price. Let us compute the delta of a call option.
Recall that the price of a call option is given by
C = S N (d1 ) − Ke−rT N (d2 ) (4.5.1)
where
S
+ (r ± 12 σ 2 )T


ln
d1,2 = K √ . (4.5.2)
σ T
Delta of call option is given by
∂C
∆call =
∂S

48
Stochastic Calculus Umesh Kumar

∂N (d1 ) ∂N (d2 )
= N (d1 ) + S − Ke−rT .
∂S ∂S
We have
∂N (d1 ) ∂N (d1 ) ∂d1
=
∂S ∂d1 ∂S
1 − 1 d21 1
=√ e 2 √ .
2π σ TS

Similarly, we can compute ∂N (d2 )

∂S and show that
∂N (d1 ) ∂N (d2 )
S = Ke−rT .
∂S ∂S
This proves that
∆call = N (d1 ).

Exercise 4.5.1. Using Put-call parity or directly, prove that the Delta of European
put option is given by
∆put = N (d1 ) − 1.

Gamma

The rate of change in Delta of an option w.r.t. change is S is called Gamma of the
option, that is,
∂∆ ∂2V
Γ := = .
∂S ∂S 2
For call and put options, we have
1 1 2 1
Γcall = Γput = √ e− 2 d1 √ .
2π σ TS
We can similarly dene other Greeks Vega, Rho and Theta as the rate of change of
Option price V with respect to changes in other factors like volatility, interest rate
and maturity (or time to maturity) respectively. Formally, we have
∂V
Vega ν =
∂σ
∂V
Rho ρ =
∂r
∂V
Theta Θ = −
∂T

Exercise 4.5.2. Compute Vega, Rho and Theta for European call and put options.

49
Stochastic Calculus Umesh Kumar

4.6 Girsanov's Theorem and Risk-Neutral Valuation

Recall that we discussed the change of real world measure P to risk-neutral measure
Q for pricing of options in Black-Scholes model. In this section we discuss the formal
approach of change of measure. In particular, we will see that a Brownian motion
plus a drift term can be converted to a Brownian motion under a new probability
measure. This is the idea of Girsnov's theorem. We rst recall the idea of change of
measure and Radon-Nikodym density.
Let Z be a positive random variable dened on a probability space (Ω, F, P ) and
suppose that E[Z] = 1. Dene Q : F → [0, 1] by
Q(A) = E[Z 1A ] for all A ∈ F. (4.6.1)
Then Q satises the properties given in the following exercise: (already discussed in
Chapter 1, see Question 11 of Problems-1.4).
Exercise 4.6.1. Prove that
(a) Q dened by (4.6.1) is a probability measure on (Ω, F);
(b) Q is equivalent to P , that is, for A ∈ F, P (A) = 0 ⇐⇒ Q(A) = 0);
(c) if X : Ω → R is a random variable, then X is
E Q [X] = E[XZ].

The random variable Z is called the Radon Nikodym density of Q w.r.t. P and
we write
dQ
= Z.
dP
Theorem 4.6.2 (Girsanov's Theorem). Let W be a Brownian motion under P . Let
(Θ(t) : t ∈ [0, T ]) be an adapted stochastic process. Dene
 Z t
1 t 2
Z 
Z(t) = exp − Θ(s) dW (s) − Θ (s) ds .
0 2 0
If Novikov's condition   
1 2
E exp Θ (s) ds < ∞,
2
is satised, then (Z(t) : t ∈ [0, T ]) is a martingale. Dene
Q(A) = E[Z(T )1A ].
Then Q is a probability measure and (W Q (t) : t ∈ [0, T ]) is a Brownian motion under
Q where Z t
W Q (t) := W (t) + Θ(s) ds,
0
or equivalently
dW Q (t) = dW (t) + Θ(t) dt.

50
Stochastic Calculus Umesh Kumar

We now make the change of measure from P to Q in Black-Scholes model. Recall

that under Q, we noted that the drift of stock price in Black-Scholes model changes
from µ to r. To dene Q formally, note that
dS(t) = µS(t) dt + σS(t) dW (t)
= (µ − r + r)S(t) dt + σS(t) dW (t)
 
µ−r
= rS(t) dt + σS(t) dt + dW (t)
σ
= rS(t) dt + σS(t) dW Q (t).
where
µ−r
dW Q (t) := dt + dW (t).
σ
If we dene θ = σ ,
µ−r
then
1 2
Z(t) := e−θW (t)− 2 θ t

is a martingale. Here, θ is known as market price of risk. By Girsanov's theorem

Z
Q(A) := E[Z(T )1A ] = (4.6.2)
1 2
e−θW (T )− 2 θ T
dP
A

is a probability measure and

µ−r
W Q (t) := t + W (t)
σ
denes a Brownian motion under Q. Thus the dynamics of S under Q is given by
the equation
dS(t) = rS(t) dt + σS(t) dW Q (t). (4.6.3)
This justies the change of measure from P to Q discussed in Sections 4.3 and 4.4.
Recall that we called Q to be a risk-neutral measure. We now formally dene the
concept of a risk-neutral measure.
We assume the existence of n-assets S1 , . . . , Sn in the market. Let D(t) denote a
discount process. (In the Black-Scholes model D(t) = e−rt ).
Denition 4.6.3. A probability measure Q is called a risk-neutral measure or equiv-
alent martingale measure (E.M.M) if
(i) Q is equivalent to P (i.e., for A ∈ F, P (A) = 0 ⇐⇒ Q(A) = 0);
(ii) under Q, the discounted asset price D(t)Si (t) is a martingale for every i =
1, . . . , n.
In the Black-Scholes model we have a single asset S with dynamics given by
(4.6.3) under the measure Q dened by (4.6.2). Using Itô Lemma, it follows that
the discounted stock process S(t)
e := e−rt S(t) is a martingale under Q. Moreover, Q
and P are equivalent. Thus, Q is a risk-neutral measure.
Under a risk-neutral measure all discounted price processes become martingales.

51
Stochastic Calculus Umesh Kumar

Lemma 4.6.4. Let Q be a risk-neutral measure, and let X(t) be a portfolio. Under
Q, the discounted portfolio is a martingale.
This idea can be used to price options. For example, if V (t) is value of a derivative
at time t, and a risk-neutral measure Q exists in the market, then the disounted
derivative process (D(t)V (t) : t ∈ [0, T ] is a martingale under Q. So, we have
D(t)V (t) = E Q [D(T )V (T )|Ft ].
But the natural question is what happens if more than one risk-neutral measure
exists. This will make the things complicated as the price may not be determined
uniquely. But in a complete market, the risk neutral measure if exists is always
unique. By a complete market we mean the following:
Denition 4.6.5. A market model is complete if every derivative security can be
hedged, that is, it can be replicated.
We now state two very important theorems of nancial mathematics together
called the fundamental theorem of asset pricing.
Theorem 4.6.6 (First Fundamental Theorem of Asset Pricing). If a market model
has a risk-neutral probability measure, then it does not admit arbitrage.
Theorem 4.6.7 (Second Fundamental Theorem of Asset Pricing). Consider a mar-
ket model that has a risk neutral probability measure. The model is complete if and
only if the risk-neutral probability measure is unique.
Note that we have proved the existence of a risk-neutral measure in Black-Scholes
model. Therefore, the Black-Scholes model is arbitrage free. Further, as every deriva-
tive can be replicated, the Black-Scholes model is complete and hence there is a
unique risk neutral measure Q given by (4.6.2). This will guarantee the existence of
unique fair price for a derivative given by the risk-neutral pricing formula.
In the end we discuss the general risk neutral valuation approach. Suppose the
stock price process is given by the generalised geometric Brownian motion (G.G.B.M.)
dS(t) = µ(t)S(t) dt + σ(t)S(t) dW (t).
Here, we assume that µ(·) and σ(·) are adapted processes. By applying Itô lemma
on ln S(t), it follows that
Z t   Z t 
1 2
S(t) = S(0) exp µ(s) − σ (s) ds + σ(s) dW (s) .
0 2 0

We also assume that the money market account is given by

dB(t) = r(t)B(t) dt.
Here, r(t) is a deterministic function. Solving this dierential equation, we have
Z t 
B(t) = B(0) exp r(s) ds .
0

52
Stochastic Calculus Umesh Kumar

We take B(0) = 1 and dene the discount factor by D(t) = B(t) ,

1
that is
 Z t 
D(t) = exp − r(s) ds .
0

Obviously, D(t) satises

dD(t) = −r(t)D(t) dt.
Let us nd the risk-neutral measure Q. Under Q, the discounted stock price must
be a martingale. The discounted stock price is
Z t   Z t 
1 2
S(t)
e = D(t)S(t) = S(0) exp µ(s) − r(s) − σ (s) ds + σ(s) dW (s) ,
0 2 0

and its dierential is given by

e = (µ(t) − r(t)) S(t)
dS(t) e dt + σ(t)S(t)
e dW (t)
 
µ(t) − r(t)
= S(t)
e dt + dW (t) .
σ(t)

We dene the market price of risk by

µ(t) − r(t)
Θ(t) = .
σ(t)

Applying Girsanov's theorem, we get a probability measure Q such that W Q given

by
µ(t) − r(t)
dW Q (t) = dt + dW (t),
σ(t)
is a Brownian motion under Q. Hence, the discounted stock price is given by
dS(t) e dW Q (t),
e = S(t)

which being a driftless process is a martingale under Q.

Exercise 4.6.8. Prove that a driftless Itô process is a martingale, that is, if
dX(t) = b(t, X(t) dW (t),

then (X(t) : t > 0) is a martingale. (Note that X(t) can be written as an Itô integral)
Remark 4.6.9. The converse of this Exercise is also true, that is a martingale is
a driftless process. More formally, we have the martingale representation theorem,
which says that every martingale can be expressed as an Itô integral.

53
Stochastic Calculus Umesh Kumar

We have thus shown the existence of a risk-neutral measure Q under which the
discounted stock price is a martingale and the market model is arbitrage free. As-
suming that the market is complete, Q is unique by second fundamental theorem.
Consider an option V with payo V (T ) on underlying S . We want to compute its
price. Since the market is complete, we replicate the payo by a portfolio X , i.e.
X(T ) = V (T ). Then the discounted portfolio is a martingale under Q. We thus
have
D(t)X(t) = E Q [D(T )X(T )|Ft ] = E Q [D(T )V (T )|Ft ].
In particular at time t = 0, we have
X(0) = E Q [D(T )V (T )].

By no-arbitrage, X(0) = V (0). Thus we get the risk neutral pricing formula
V (0) = E Q [D(T )V (T )].

In general at any time t, we have

  Z T  
1 Q Q
V (t) = E [D(T )V (T )|Ft ] = E exp − r(s) ds V (T )|Ft
D(t) t

This gives us the general risk-neutral pricing formula which says that the price of
a derivative can be obtained by taking conditional expectation under risk neutral
measure of the discounted payo.
In general, we can have n stocks S1 , . . . , Sn in the market . We are not discussing
the risk neutral valuation approach in this case, but you can look in the Textbook by
Shreve for further details. Also, the interest rate r(t) is assumed deterministic but
can be stochastic also. We can model such interest rate by stochastic processes like
an Ornstein Uhlenbeck process etc. We have many interest rate models like Vasocek
model, CIR model etc. This itself is a nice area of study in nancial mathematics,
but is beyond the scope of current course.

54