IEOR E4707: Financial Engineering: Continuous-Time Models Fall 2013

c 2013 by Martin Haugh

Introduction to Stochastic Calculus for Diffusions

These notes provide an introduction to stochastic calculus, the branch of mathematics that is most identified
with financial engineering and mathematical finance. We will ignore most of the “technical” details and take an
“engineering” approach to the subject. We will cover more material than is strictly necessary for this course.
Any material that is not required, however, should be of value for other courses such as Term Structure Models.
We make the following assumptions throughout.
• There is a probability triple (Ω, F, P ) where

– P is the “true” or physical probability measure

– Ω is the universe of possible outcomes. We use ω ∈ Ω to represent a generic outcome, typically a
sample path(s) of a stochastic process(es).
– the set1 F represents the set of possible events where an event is a subset of Ω.

• There is also a filtration, {Ft }t≥0 , that models the evolution of information through time. So for example,
if it is known by time t whether or not an event, E, has occurred, then we have E ∈ Ft . If we are working
with a finite horizon, [0, T ], then we can take F = FT .
• We also say that a stochastic process, Xt , is Ft -adapted if the value of Xt is known at time t when the
information represented by Ft is known. All the processes we consider will be Ft -adapted so we will not
bother to state this in the sequel.
• In the continuous-time models that we will study, it will be understood that the filtration {Ft }t≥0 will be
the filtration generated by the stochastic processes (usually a Brownian motion, Wt ) that are specified in
the model description.

• Since these notes regularly refer to numeraires and equivalent martingale measures (EMMs), readers
should be familiar with these concepts in advance. My Martingale Pricing Theory in Discrete-Time and
Discrete-Space lecture notes, for example, covers these topics in a discrete-time, discrete-space framework.
In Section 13 of these notes we will discuss martingale pricing theory in the continuous-time setting and
state without proof2 the two Fundamental Theorems of Asset pricing.

1 Martingales and Brownian Motion

Definition 1 A stochastic process, {Wt : 0 ≤ t ≤ ∞}, is a standard Brownian motion if
1. W0 = 0

2. It has continuous sample paths

3. It has independent, normally-distributed increments.

1 Technically,
F is a σ-algebra.
2 These
Theorems, however, are stated and proven in the discrete-time, discrete-space setting of the Martingale Pricing
Theory in Discrete-Time and Discrete-Space lecture notes.
Introduction to Stochastic Calculus for Diffusions 2
(1) (n)
Definition 2 An n-dimensional process, Wt = (Wt , . . . , Wt ), is a standard n-dimensional Brownian
(i) (i)
motion if each Wt is a standard Brownian motion and the Wt ’s are independent of each other.

Definition 3 A stochastic process, {Xt : 0 ≤ t ≤ ∞}, is a martingale with respect to the filtration, Ft , and
probability measure, P , if
1. EP [|Xt |] < ∞ for all t ≥ 0
2. EP [Xt+s |Ft ] = Xt for all t, s ≥ 0.

Example 1 (Brownian martingales)

Let Wt be a Brownian motion. Then Wt , Wt2 − t and exp θWt − θ2 t/2 are all martingales.
The latter martingale is an example of an exponential martingale. Exponential martingales are of particular
significance since they are positive and may be used to define new probability measures.

Exercise 1 (Conditional expectations as martingales) Let Z be a random variable and set

Xt := E[Z|Ft ]. Show that Xt is a martingale.

2 Quadratic Variation
Consider a partition of the time interval, [0, T ] given by
0 = t0 < t1 < t2 < . . . < tn = T.
Let Xt be a Brownian motion and consider the sum of squared changes
n
X 2
Qn (T ) := [∆Xti ]
i=1

where ∆Xti := Xti − Xti−1 .

Definition 4 (Quadratic Variation) The quadratic variation of a stochastic process, Xt , is equal to the
limit of Qn (T ) as ∆t := maxi (ti − ti−1 ) → 0.

Theorem 1 The quadratic variation of a Brownian motion is equal to T with probability 1.

The functions with which you are normally familiar, e.g. continuous differentiable functions, have quadratic
variation equal to zero. Note that any continuous stochastic process or function3 that has non-zero quadratic
variation must have infinite total variation where the total variation of a process, Xt , on [0, T ] is defined as
n
X
Total Variation := lim |Xtk − Xtk−1 |.
∆t→0
i=1

This follows by observing that

n n
!
X X
(Xtk − Xtk−1 )2 ≤ |Xtk − Xtk−1 | max |Xtk − Xtk−1 |. (1)
1≤k≤n
i=1 i=1

If we now let n → ∞ in (1) then the continuity of Xt implies the impossibility of the process having finite total
variation and non-zero quadratic variation. Theorem 1 therefore implies that the total variation of a Brownian
motion is infinite. We have the following important result which will prove very useful when we price options
when there are multiple underlying Brownian motions, as is the case with quanto options for example.
3A sample path of a stochastic process can be viewed as a function.
Introduction to Stochastic Calculus for Diffusions 3

Theorem 2 (Levy’s Theorem) A continuous martingale is a Brownian motion if and only if its quadratic
variation over each interval [0, t] is equal to t.

Another interesting result is the following:

Theorem 3 Any non-constant continuous martingale must have infinite total variation.

We know from our discrete-time models that any arbitrage-free model must have an equivalent martingale
measure. The same is true in continuous-time models. Theorem 3 then implies that if we work in
continuous-time with continuous (deflated) price processes, then these processes must have infinite total
variation.

3 Stochastic Integrals
We now discuss the concept of a stochastic integral, ignoring the various technical conditions that are required
to make our definitions rigorous. In this section, we write Xt (ω) instead of the usual Xt to emphasize that the
quantities in question are stochastic.
Definition 5 A stopping time of the filtration Ft is a random time, τ , such that the event {τ ≤ t} ∈ Ft for all
t > 0.
In non-mathematical terms, we see that a stopping time is a random time whose value is part of the information
accumulated by that time.
Definition 6 We say a process, ht (ω), is elementary if it is piece-wise constant so that there exists a sequence
of stopping times 0 = t0 < t1 < . . . < tn = T and a set of Fti -measurable4 functions, ei (ω), such that
X
ht (ω) = ei (ω)I[ti ,ti+1 ) (t)
i

where I[ti ,ti+1 ) (t) = 1 if t ∈ [ti , ti+1 ) and 0 otherwise.

Definition 7 The stochastic integral of an elementary function, ht (ω), with respect to a Brownian motion,
Wt , is defined as
Z T n−1
X

ht (ω) dWt (ω) := ei (ω) Wti+1 (ω) − Wti (ω) . (2)
0 i=0

Note that our definition of an elementary function assumes that the function, ht (ω), is evaluated at the
left-hand point of the interval in which t falls. This is a key component in the definition of the stochastic
integral: without it the results below would no longer hold. Moreover, defining the stochastic integral in this
way makes the resulting theory suitable for financial applications. In particular, if we interpret ht (ω) as a trading
strategy and the stochastic integral as the gains or losses from this trading strategy, then evaluating ht (ω) at
the left-hand point is equivalent to imposing the non-anticipativity of the trading strategy, a property that we
always wish to impose.
For a more general process, Xt (ω), we have
Z T Z T
(n)
Xt (ω) dWt (ω) := lim Xt (ω) dWt (ω)
0 n→∞ 0

(n)
where Xt is a sequence of elementary processes that converges (in an appropriate manner) to Xt .

4A function f (ω) is Ft measurable if its value is known by time t.

Introduction to Stochastic Calculus for Diffusions 4
RT
Example 2 We want to compute 0
Wt dWt . Towards this end, let 0 = tn0 < tn1 < tn2 < . . . < tnn = T be a
partition of [0, T ] and define
n−1
X
Xtn := Wtni I[tni ,tni+1 ) (t)
i=0

where I[tni ,tni+1 ) (t) = 1 if t ∈ [tni , tni+1 ) and is 0 otherwise. Then Xtn is an adapted elementary process and, by
continuity of Brownian motion, satisfies limn→∞ Xtn = Wt almost surely as maxi |tni+1 − tni | → 0. The Itô
integral of Xtn is given by
Z T n−1
X
Xtn dWt = Wtni (Wtni+1 − Wtni )
0 i=0
n−1
1 X 2 
= Wtni+1 − Wt2ni − (Wtni+1 − Wtni )2
2 i=0
n−1
1 2 1 2 1X
= WT − W0 − (Wtni+1 − Wtni )2 . (3)
2 2 2 i=0

By the definition of quadratic variation the sum on the right-hand-side of (3) converges in probability to T . And
since W0 = 0 we obtain Z T Z T
1 1
Wt dWt = lim Xtn dWt = WT2 − T.
0 n→∞ 0 2 2

Note that we will generally evaluate stochastic integrals using Itô’s Lemma (to be discussed later) without
having to take limits of elementary processes as we did in Example 2.

Definition 8 We define the space L2 [0, T ] to be the space of processes, Xt (ω), such that
"Z #
T
E Xt (ω)2 dt < ∞.
0

Theorem 4 (Itô’s Isometry) For any Xt (ω) ∈ L2 [0, T ] we have

 !2  "Z #
Z T T
2
E  Xt (ω) dWt (ω)  = E Xt (ω) dt .
0 0

Proof: (For the case where Xt is an elementary process)

Introduction to Stochastic Calculus for Diffusions 5
P
Let Xt = i ei (ω)I[ti ,ti+1 ) (t) be an elementary process
where the ei (ω)’s and ti ’s are as defined in Definition 6.
RT Pn−1

We therefore have 0 Xt (ω) dWt (ω) := Wti+1 (ω) − Wti (ω) We then have
i=0 ei (ω)
 !2   !2 
Z T n−1
X

E Xt (ω) dWt (ω)  = E ei (ω) Wti+1 (ω) − Wti (ω) 
0 i=0
n−1 h
X
2 i
= E e2i (ω) Wti+1 (ω) − Wti (ω)
i=0
n−1
X 


+2 E ei ej (ω) Wti+1 (ω) − Wti (ω) Wtj+1 (ω) − Wtj (ω)
0≤i<j≤n−1
 
n−1 h
X  2
2 i
= E
e i (ω) E ti W ti+1 (ω) − W t i (ω) 

i=0 | {z }
= ti+1 −ti
 
n−1
X


+2 E ei ej (ω) Wti+1 (ω) − Wti (ω) Etj Wtj+1 (ω) − Wtj (ω) 

0≤i<j≤n−1
| {z }
=0
"n−1 #
X
= E e2i (ω)(ti+1 − ti )
i=0
"Z #
T
2
= E Xt (ω) dt
0

which is what we had to show.

Theorem 5 (Martingale Property of Stochastic Integrals) The stochastic integral,

Rt
Yt := 0 Xs (ω) dWs (ω), is a martingale for any Xt (ω) ∈ L2 [0, T ].

Rt
Exercise 2 Check that 0 Xs (ω) dWt (ω) is indeed a martingale when Xt is an elementary process. (Hint:
Follow the steps we took in our proof of Theorem 4.)

4 Stochastic Differential Equations

Definition 9 An n-dimensional Itô process, Xt , is a process that can be represented as
Z t Z t
Xt = X0 + as ds + bs dWs (4)
0 0

where W is an m-dimensional standard Brownian motion, and a and b are n-dimensional and n × m-dimensional
Ft -adapted5 processes, respectively6 .

We often use the notation

dXt = at dt + bt dWt
5 a and b are F -‘adapted’ if a and bt are Ft -measurable for all t. We always assume that our processes are Ft -adapted.
t t t t
6 Additional technical conditions on at and bt are also necessary.
Introduction to Stochastic Calculus for Diffusions 6

as shorthand for (4). An n-dimensional stochastic differential equation (SDE) has the form

dXt = a(Xt , t) dt + b(Xt , t) dWt ; X0 = x (5)

where as before, Wt is an m-dimensional standard Brownian motion, and a and b are n-dimensional and
n × m-dimensional adapted processes, respectively. Once again, (5) is shorthand for
Z t Z t
Xt = x + a(Xs , s) dt + b(Xs , t) dWs . (6)
0 0

While we do not discuss the issue here, various conditions exist to guarantee existence and uniqueness of
solutions to (6). A useful tool for solving SDE’s is Itô’s Lemma which we now discuss.
Introduction to Stochastic Calculus for Diffusions 7

5 Itô’s Lemma
Itô’s Lemma is the most important result in stochastic calculus, the “sine qua non” of the field. We first state
and give an outline proof of a basic form of the result.

Theorem 6 (Itô’s Lemma for 1-dimensional Brownian Motion)

Let Wt be a Brownian motion on [0, T ] and suppose f (x) is a twice continuously differentiable function on R.
Then for any t ≤ T we have
Z t Z t
1
f (Wt ) = f (0) + f 00 (Ws ) ds + f 0 (Ws ) dWs . (7)
2 0 0

Proof: (Sketch) Let 0 = t0 < t1 < t2 < . . . < tn = t be a partition of [0, t]. Clearly
n−1
X

f (Wt ) = f (0) + f (Wti+1 ) − f (Wti ) . (8)
i=0

Taylor’s Theorem implies

1 00
f (Wti+1 ) − f (Wti ) = f 0 (Wti )(Wti+1 − Wti ) + f (θi )(Wti+1 − Wti )2 (9)
2
for some θi ∈ (Wti , Wti+1 ). Substituting (9) into (8) we obtain
n−1 n−1
X 1 X 00
f (Wt ) = f (0) + f 0 (Wti )(Wti+1 − Wti ) + f (θi )(Wti+1 − Wti )2 . (10)
i=0
2 i=0

If we let δ := maxi |ti+1 − ti | → 0 then it can be shown that the terms on the right-hand-side of (10) converge
to the corresponding terms on the right-hand-side of (7) as desired. (This should not be surprising as we know
the quadratic variation of Brownian motion on [0, t] is equal to t.)
A more general version of Itô’s Lemma can be stated for Itô processes.

Theorem 7 (Itô’s Lemma for 1-dimensional Itô process)

Let Xt be a 1-dimensional Itô process satisfying the SDE

dXt = µt dt + σt dWt .

If f (t, x) : [0, ∞) × R → R is a C 1,2 function and Zt := f (t, Xt ) then

∂f ∂f 1 ∂2f
dZt = (t, Xt ) dt + (t, Xt ) dXt + (t, Xt ) (dXt )2
∂t ∂x 2 ∂x2

1 ∂2f
 
∂f ∂f 2 ∂f
= (t, Xt ) + (t, Xt ) µt + (t, X t ) σt dt + (t, Xt ) σt dWt
∂t ∂x 2 ∂x2 ∂x

The “Box” Calculus

In the statement of Itô’s Lemma, we implicitly assumed that (dXt )2 = σts dt. The “box calculus” is a series of
simple rules for calculating such quantities. In particular, we use the rules

dt × dt = dt × dWt = 0 and
dWt × dWt = dt
Introduction to Stochastic Calculus for Diffusions 8

when determining quantities such as (dXt )2 in the statement of Itô’s Lemma above. Note that these rules are
(1) (2)
consistent with Theorem 1. When we have two correlated Brownian motions, Wt and Wt , with correlation
(1) (2)
coefficient, ρt , then we easily obtain that dWt × dWt = ρt dt. We use the box calculus for computing the
quadratic variation of Itô processes.
(1) (2)
Exercise 3 Let Wt and Wt be two independent Brownian motions. Use Levy’s Theorem to show that
(1) (2)
p
Wt := ρ Wt + 1 − ρ2 Wt
is also a Brownian motion for a given constant, ρ.

Example 3
Suppose a stock price, St , satisfies the SDE
dSt = µt St dt + σt St dWt .
Then we can use the substitution, Yt = log(St ) and Itô’s Lemma applied to the function7 f (x) := log(x) to
obtain Z t Z t 
St = S0 exp (µs − σs2 /2) ds + σs dWs . (11)
0 0
Note that St does not appear on the right-hand-side of (11) so that we have indeed solved the SDE. When
µs = µ and σs = σ are constants we obtain
St = S0 exp (µ − σ 2 /2) t + σ dWt


(12)


so that log(St ) ∼ N (µ − σ 2 /2)t, σ 2 t .

Example 4 (Ornstein-Uhlenbeck Process)

Let St be a security price and suppose Xt = log(St ) satisfies the SDE
dXt = [−γ(Xt − µt) + µ] dt + σdWt .
Then we can apply Itô’s Lemma to Yt := exp(γt)Xt to obtain
dYt = exp(γt) dXt + Xt d (exp(γt))
= exp(γt) ([−γ(Xt − µt) + µ] dt + σdWt ) + Xt γ exp(γt) dt
= exp(γt) ([γµt + µ] dt + σdWt )
so that Z t Z t
Yt = Y0 + µ eγs (γs + 1) ds + σ eγs dWs (13)
0 0
or alternatively (after simplifying the Riemann integral in (13))
Z t
Xt = X0 e−γt + µt + σe−γt eγs dWs . (14)
0

Once again, note that Xt does not appear on the right-hand-side of (14) so that we have indeed solved the
SDE. We also obtain E[Xt ] = X0 e−γt + µt and
 Z t  "Z
t 2 #
−γt γs 2 −2γt γs
Var(Xt ) = Var σe e dWs = σ e E e dWs
0 0
Z t
= σ 2 e−2γt e2γs ds (by Itô’s Isometry) (15)
0
σ2
1 − e−2γt .


= (16)
2γ
7 Note that f (·) is not a function of t.
Introduction to Stochastic Calculus for Diffusions 9

These moments should be compared with the corresponding moments for log(St ) in the previous example.

Theorem 8 (Itô’s Lemma for n-dimensional Itô process) Let Xt be an n-dimensional Itô process
satisfying the SDE
dXt = µt dt + σt dWt .

where Xt ∈ Rn , µt ∈ Rn , σt ∈ Rn×m and Wt is a standard m-dimensional Brownian motion. If

f (t, x) : [0, ∞) × Rn → R is a C 1,2 function and Zt := f (t, Xt ) then

∂f X ∂f (i) 1 X ∂2f (i) (j)

dZt = (t, Xt ) dt + (t, Xt ) dXt + (t, Xt ) dXt dXt
∂t i
∂xi 2 i,j
∂xi ∂xj

(i) (j) (i) (i) (i)

where dWt dWt = dt dWt = 0 for i 6= j and dWt dWt = dt.

Exercise 4 Let Xt and Yt satisfy

(1) (1,1) (1)
dXt = µt dt + σt dWt
(2) (2,1) (1) (2,2) (2)
dYt = µt dt + σt dWt + σt dWt

and define Zt := Xt Yt . Apply the multi-dimensional version of Itô’s Lemma to find the SDE satisfied by Zt .

6 The Martingale Representation Theorem

The martingale representation theorem is an important result that is particularly useful for constructing
replicating portfolios in complete financial models.

Theorem 9 Suppose Mt is an Ft -martingale where {Ft }t≥0 is the filtration generated by the n-dimensional
(1) (n)
standard Brownian motion, Wt = (Wt , . . . , Wt ). If E[Mt2 ] < ∞ for all t then there exists a unique8
n-dimensional adapted stochastic process, φt , such that
Z t
Mt = M0 + φTs dWt for all t ≥ 0
0

where φTs denotes the transpose of the vector, φs .

Example 5 Let F = WT3 and define Mt = Et [F ]. We will show that

Z t
Mt = 3 (T − s + Ws2 ) dWs (17)
0

which is consistent with the Martingale Representation theorem. First we must calculate Mt . We do this using
the independent increments property of Brownian motion and obtain

Mt = Et [WT3 ] = Et [(WT − Wt + Wt )3 ]
= Et [(WT − Wt )3 ] + Et [Wt3 ] + 3Et [Wt (WT − Wt )2 ] + 3 Et [Wt2 (WT − Wt )]
| {z } | {z }
= 0 = 0
= Wt3 + 3Wt (T − t). (18)
8 To be precise, additional integrability conditions are required of φs in order to claim that it is unique.
Introduction to Stochastic Calculus for Diffusions 10

We can now apply Itô’s Lemma to (18) to obtain

1
dMt = 3Wt2 dWt + 6Wt dt + 3(T − t) dWt − 3Wt dt
2
2
= 3(Wt + T − t) dWt

which, noting M0 = 0, is (17).

7 Gaussian Processes
Definition 10 A process Xt , t ≥ 0, is a Gaussian process if (Xt1 , . . . , Xtn ) is jointly normally distributed for
every n and every set of times 0 ≤ t1 ≤ t2 ≤ . . . ≤ tn .

If Xt is a Gaussian process, then it is determined by its mean function, m(t), and its covariance function,
ρ(s, t), where

m(t) = E [Xt ]
ρ(s, t) = E [(Xs − m(s))(Xt − m(t))] .

In particular, the joint moment generating function (MGF) of (Xt1 , . . . , Xtn ) is given by
 
T 1 T
Mt1 ,...,tn (θ1 , . . . , θn ) = exp θ m(t) + θ Σθ (19)
2
T
where m(t) = (m(t1 ) . . . m(tn )) and Σi,j = ρ(ti , tj ).

Example 6 (Brownian motion)

Brownian motion is a Gaussian process with m(t) = 0 and ρ(s, t) = min(s, t) for all s, t ≥ 0.

Theorem 10 (Integration of a deterministic function w.r.t. a Brownian motion) Let Wt be a

Brownian motion and suppose Z t
Xt = δs dWs
0
R min(s,t)
where δs is a deterministic function. Then Xt is a Gaussian process with m(t) = 0 and ρ(s, t) = 0
δs2 ds.

Proof: (Sketch)
(i) First use Itô’s Lemma to show that
Z t
1
E euXt = 1 + u2 δs2 E euXs ds.
   
(20)
2 0
 
If we set yt := E euXt then we can differentiate across (20) to obtain the ODE

dy 1
= u2 δt2 y.
dt 2
This is easily solved to obtain the MGF for Xt ,
 Z t 
1 2
E euXt = exp δs2
 
u ds (21)
2 0
Introduction to Stochastic Calculus for Diffusions 11
Rt
which, as expected, is the MGF of a normal random variable with mean 0 and variance 0
δs2 ds.
(ii) We now use (21) and similar computations to show that the joint MGF of (Xt1 , . . . , Xtn ) has the form
R min(s,t) 2
given in (19) with m(t) = 0 and ρ(s, t) = 0 δs ds. (See Shreve’s Stochastic Calculus for Finance II for
further details.)

The next theorem again concerns Gaussian processes and is often of interest9 when studying short-rate term
structure models.

Theorem 11 Let Wt be a Brownian motion and suppose δt and φt are deterministic functions. If
Z t Z t
Xt := δu dWu and Yt := φu Xu du
0 0

then Yt is a Gaussian process with m(t) = 0 and

Z min(s,t) Z s  Z t 
ρ(s, t) = δv2 φy dy φy dy dv.
0 v v

Proof: The proof is tedious but straightforward. (Again, see Shreve’s Stochastic Calculus for Finance II for
further details.)

Note for example that Brownian motion with drift and the Ornstein-Uhlenbeck process are both Gaussian
processes. Nonetheless, we saw in Example 4 that these two processes behave very differently: the
Ornstein-Uhlenbeck process mean-reverts and its variance tends to a finite limit as T → ∞. This is not true of
Brownain motion with or without drift.

8 The Feynman-Kac Formula

Suppose Xt is a scalar stochastic process satisfying the SDE
dXt = µ(t, Xt ) dt + σ(t, Xt ) dWt (22)
where as usual µ(t, Xt ) and σ(t, Xt ) are Ft -adapted and satisfy sufficient conditions to guarantee existence of a
unique solution to (22). Now consider the function, f (x, t), given by
"Z #
T
x (t) (t)
f (t, x) = Et φs h(Xs , s) ds + φT g(XT )
t

where  Z s 
φ(t)
s = exp − r(Xu , u) du
t
and the notation Ext [·] implies that the expectation should be taken conditional on time t information with
Xt = x. Note that f (x, t) may be interpreted as the time t price of a security that pays dividends at a
continuous rate, h(Xs , s) for s ≥ t, and with a terminal payoff g(XT ) at time T . Of course E[·] should then be
interpreted as an expectation under an equivalent martingale measure with the cash account as the
corresponding numeraire and r(·, ·) as the instantaneous risk-free rate.
The Feynman-Kac Theorem10 states that f (·, ·) satisfies the following PDE

∂f ∂f 1 ∂2f
(t, Xt ) + (t, Xt ) µt (t, Xt ) + (t, Xt ) σt2 (t, Xt ) − r(x, t)f (x, t) + h(x, t) = 0, (x, t) ∈ R × [0, T )
∂t ∂x 2 ∂x2
f (x, T ) = g(x), x∈R
9 See, for example, Hull and White’s one-factor model.
10 Additional technical conditions on µ, σ, r, h, g and f are required.
Introduction to Stochastic Calculus for Diffusions 12

Proof: (Sketch proof )

We will use the martingale property of conditional expectations which states that if Mt := Et [F ] where F is a
given fixed random variable then Mt is a martingale. First let
"Z #
T
(t)
f (t, Xt ) := Et φ(t)
s h(Xs , s) ds + φT g(XT ) (23)
t

and note that the random variable inside the expectation on the right-hand-side of (23) is a function of t. This
means we cannot yet apply the martingale property of conditional expectations. However by adding
R t (t) (0)
φ h(Xs , s) ds to both sides of (23) and then multiplying both sides by φt we obtain
0 s
"Z #
 Z t  T
(0) (0)
φt f (t, Xt ) + φ(t)
s h(Xs , s) ds = Et φ(0)
s h(Xs , s) ds + φT g(XT ) (24)
0 0

= Et [Z] , say.

Note that Z does not depend on t and so we can now apply the martingale property of conditional expectations
to conclude that the left-hand-side of (24) is a martingale. We therefore apply Itô’s Lemma to the left-hand-side
and set the coefficient of dt to zero. This results in the Feynman-Kac PDE given above.

Remark 1 The Feynman-Kac result generalizes easily to the case where Xt is an n-dimensional Itô process
driven by an m-dimensional standard Brownian motion.

The Feynman-Kac Theorem plays an important role in financial engineering as it enables us to compute the
prices of derivative securities (which can be expressed as expectations according to martingale pricing theory) by
solving a PDE instead. It is worth mentioning, however, that in the development of derivatives pricing theory,
the PDE approach preceded the martingale approach. Indeed by constructing a replicating11 strategy for a given
derivative security, it can be shown directly that the price of the derivative security satisfies the Feynman-Kac
PDE.

9 The Kolmogorov Equations

We assume in this section that Xt again satisfies the SDE of (22). Let

u(x, t) := E [Φ(Xs ) | Xt = x]

for some function Φ(·) and with s > t. Then following the same argument that we used in Section 8 we see that
u satisfies
1
ut + µ(t, x)ux + σ 2 (t, x)uxx = 0 for t < s with u(x, s) = Φ(x). (25)
2
Suppose we now take 
1, for x ≤ y
Φ(x) =
0, otherwise.
so that u(x, t) = P (Xs ≤ y | Xt = x) =: P (t, x; s, y), say. Let p(t, x; s, y) be the corresponding PDF so that
p(t, x; s, y) = ∂P (t, x; s, y)/∂y. By (formally) differentiating across (25) with respect to y we obtain
Kolmogorov’s backward equation for the transition density, p(t, x; s, y). In particular we obtain
1
pt + µ(t, x)px + σ 2 (t, x)pxx = 0 for t < s. (Kolmogorov’s Backward Equation) (26)
2
It is worth emphasizing that (26) does not always admit a unique solution. In particular boundary conditions
need to be specified. For example absorbing and reflecting Brownian motions both satisfy (26) but they are very
11 We are implicitly assuming that the security can be replicated.
Introduction to Stochastic Calculus for Diffusions 13

different processes. It is also worth mentioning that (25) itself is what people sometimes have in mind when
they refer to Kolmogorov’s backward equation.
In order to state Kolmogorov’s forward equation we must now view p(t, x; s, y) as a function of (s, y) and
therefore keep (t, x) fixed. It can be shown that
1 2
ps + (µ(s, y)p)y − (σ (t, y)p)yy = 0 for s > t. (Kolmogorov’s Forward Equation) (27)
2
with the initial condition p = δx (y) at s = t. The forward equation can be derived quickly in a formal manner
from the backward equation using integration by parts but doing so rigorously can be very challenging. In fact,
depending on technical and boundary conditions the forward equation (27) may not actually be satisfied. The
forward equation plays a key role in local volatility models of security prices where it is used to derive the Dupire
formula.

10 Change of Probability Measure

The majority of financial engineering models price securities using the EMM, Q, that corresponds to taking the
cash account, Bt , as numeraire. Sometimes, however, it is particularly useful to work with another numeraire,
Nt , and its corresponding EMM, PN say. We now describe how to create new probability measures and how to
switch back and forth between these measures.
Let Q be a given probability measure and Mt a strictly positive Q-martingale such that EQ
0 [Mt ] = 1 for all
t ∈ [0, T ]. We may then define a new equivalent probability measure, P M , by defining

PM (A) = EQ [MT 1A ] .

Note that

(i) PM (Ω) = 1
(ii) PM (A) ≥ 0 for every event A and
(iii) It can easily be shown that !
[ X
PM Ai = PM (Ai )
i i

whenever the Ai ’s form a sequence of disjoint sets.

Points (i), (ii) and (iii) above imply that PM is indeed a probability measure. Because the null-sets of Q and
PM coincide we can conclude that PM is indeed an equivalent probability measure (to Q). Expectations with
respect to PM then satisfy
EP
0
M
[X] = EQ 0 [MT X] . (28)

Pn
Exercise 5 Verify (28) in the case where X(ω) = i=1 ci I{ω∈Ai } , MT is constant on each Ai , and where
A1 , . . . , An form a partition12 of Ω.
When we define a measure change this way, we use the notation dPM /dQ to refer to MT so that we often write
 
PM Q dPM
E0 [X] = E0 X .
dQ
12 The
S
Ai ’s form a partition of Ω if Ai ∩ Aj = ∅ whenever i 6= j and i Ai = Ω.
Introduction to Stochastic Calculus for Diffusions 14

The following result explains how to switch between Q and PM when we are taking conditional expectations. In
particular, we have
h i
EQ t
dPM
dQ X
EPt
M
[X] = h i (29)
EQ t
dPM
dQ
h i
EQ
t
dPM
dQ X
= . (30)
Mt
Since Mt is a Q-martingale (30) follows easily from (29). We defer a proof of (29) to Appendix A as it requires
the measure-theoretic definition of a conditional expectation.

Q
Exercise 6 Show that if X is Ft -measurable, i.e. X is known by time t, then EP
0 [X] = E0 [Mt X].
M

Remark 2 Since MT is strictly positive we can set X = IA /MT in (28) where IA is the indicator function of
Q
the event A. We then obtain EP
0 [IA /MT ] = E0 [IA ] = Q(A). In particular, we see that dQ/dPM is given by
M

1/MT .

Remark 3 In the context of security pricing, we can take Mt to be the deflated time t price of a security with
strictly positive payoff, normalized so that its expectation under Q is equal to 1. For example, let ZtT be the
time t price of a zero-coupon bond maturing at time T , and let13 Bt denote the time t value of the cash
account. We could then set14 MT := 1/(BT Z0T ) so that
 
1
Mt = EQ t
BT Z0T

is indeed a Q-martingale. The resulting measure, denoted by PT , is sometimes called the T -forward measure.
Note that we have implicitly assumed (why?!) that in this context, Q refers to the EMM when we take the cash
account as numeraire. We discuss PT in further detail in Section 11.

11 A Useful EMM: The Forward Measure

We now discuss15 the τ -forward measure, P τ , which is the EMM that corresponds to taking the zero-coupon
bond with maturity τ as the numeraire. We therefore let Ztτ denote the time t price of a zero-coupon bond
maturing at time τ ≥ t and with face value $1. As usual, we let Q denote the EMM corresponding to taking the
cash account, Bt , as numeraire. We assume without loss of generality that B0 = $1 and now use Ztτ to define
our new measure. To do this, set
dP τ 1
= . (31)
dQ Bτ Z0τ

Exercise 7 Check that (31) does indeed define an equivalent probability measure.
13 We assume the zero-coupon bond has face value $1 and B0 = $1.
14 Note that it is not the case that Mt = 1/(Bt Z0t ) which is not a Q-martingale.
15 This section may be skipped if you are not previously familiar with the concepts of an equivalent martingale measure

(EMM), numeraire security, cash account etc. The purpose of this section is simply to provide a concrete example of a change
of probability measure using the techniques of Section 10.
Introduction to Stochastic Calculus for Diffusions 15

Now let Ct denote the time t price of a contingent claim that expires at time τ . We then have
 
Q Cτ
Ct = Bt Et (by martingale pricing with EMM Q) (32)
Bτ
τ
h i
Cτ
Bt EP
t Bτ Bτ Z0
τ

= τ (going from Q to P τ using (29)) (33)

EPt [B Z
τ 0
τ]

τ
Bt Z0τ EP
t [Cτ ]
= (going from P τ to Q in the denominator of (33) using (29))
EQ Q τ
t [1] /Et [1/(Bτ Z0 )]
τ
= Ztτ EP
t [Cτ ] (by martingale pricing with EMM Q). (34)

We can now find Ct , either through equation (34) or through equation (32) where we use the cash account as
numeraire. Computing Ct through (32) is our “usual method” and is often very convenient. When pricing
equity derivatives, for example, we usually take interest rates, and hence the cash account, to be deterministic.
This means that the factor 1/Bτ in (32) can be taken outside the expectation so only the Q-distribution of Cτ
is needed to compute Ct .
When interest rates are stochastic we cannot take the factor 1/Bτ outside the expectation in (32) and we
therefore need to find the joint Q-distribution of (Bτ , Cτ ) in order to compute Ct . On the other hand, if we
use equation (34) to compute Ct , then we only need the P τ -distribution of Cτ , regardless of whether or not
interest rates are stochastic. Working with a univariate-distribution is generally much easier than working with a
bivariate-distribution so if we can easily find the P τ -distribution of Cτ , then it can often be very advantageous
to work with this distribution. The forward measure is therefore particularly useful when studying term-structure
models.
Switching to a different numeraire can also be advantageous in other circumstances, even when interest rates
are deterministic. For example, we will find it convenient to do so when pricing quanto-options and options on
multiple underlying securities. Moreover, we will see how we can use Girsanov’s Theorem16 to move back and
forth between Q and other EMMs corresponding to other numeraires.

12 Girsanov’s Theorem
Girsanov’s Theorem is one of the most important results for financial engineering applications. When working
with models driven by Brownian motions17 it enables us to (i) identify the equivalent martingale measure(s)
corresponding to a given numeraire and (ii) to move back and forth between different EMM-numeraire pairs.
Consider then the process  Z t
1 t 2
Z 
Lt := exp − ηs dWs − ηs ds (35)
0 2 0
where ηs is an adapted process and Ws is a P -Brownian motion. Using Itô’s Lemma we can check that
dLt = −Lt ηt dWt so Lt is a positive martingale18 with EP0 [Lt ] = 1 for all t.

Theorem 12 (Girsanov’s Theorem) Define an equivalent probability measure, Qη , by setting

Qη (A) := EP
0 [LT 1A ]. (36)
16 We will introduce Girsanov’s Theorem in Section 12.
17 Versions of Girsanov’s Theorem are also available for jump-diffusion and other processes.
18 In fact we need η to have some additional properties before we can claim L is a martingale. A sufficient condition is
s h  R i t
1 T 2
Novikov’s Condition which requires EP0 exp 2 0 ηs ds < ∞.
Introduction to Stochastic Calculus for Diffusions 16

ct := Wt + t ηs ds is a standard Qη -Brownian motion. Moreover, W

R
Then W 0
ct has the martingale representation
property under Qη .

Remark 4 Suppose ηs = η, a constant. Then since W ct := Wt + ηt is a standard Qη -Brownian motion, it

η
ct − ηt is a Q -Brownian motion with drift −η.
implies that Wt = W

Example 7 Let dXt = µt dt + σt dWt and suppose we wish to find a process, ηs , such that Xt is a
Qη -martingale. This is easily achieved as follows:

dXt = µt dt + σt dWt
= ct − ηt dt)
µt dt + σt (dW
= σt dW
ct

if we set ηt = µt /σt in which case Xt is a Qη -martingale. We can obtain some intuition for this result: suppose
for example that µt and σt are both positive so that ηt is also positive. Then we can see from the definition of
Lt in (35) that Lt places less weight on paths where Wt drifts upwards than it does on paths where it drifts
downwards. This relative weighting adjusts Xt for the positive drift induced by µt with the result that under
Qη , Xt is a martingale.

Remark 5 Note that Girsanov’s Theorem enables us to compute Qη -expectations directly without having to
switch back to the original measure, P .

We can get some additional intuition for the Girsanov Theorem by considering a random walk,
X = {X0 = 0, X1 , . . . , Xn } with the interpretation that Xi is the value of the walk at time iT /n. In particular,
Xn corresponds to the value of the random walk at time T . We assume that ∆Xi := Xi − Xi−1 ∼ N (0, T /n)
under P and is independent of X0 , . . . , Xi−1 for i = 1, . . . n.
Suppose now that we want to compute θ := EQ 0 [h(X)] where Q denotes the probability measure under which
∆Xi ∼ N (µ, T /n), again independently of X0 , . . . , Xi−1 . In particular, if we set µ := −T η/n then Xt
approximates standard Brownian motion under P and Xt approximates Brownian motion with drift −η under Q.
If we let f (·) and g(·) denote the PDF’s of N (µ, T /n) and N (0, T /n) random variables, respectively, then we
obtain
Z n
!
Q
Y
θ = E0 [h(X)] = h(x1 , . . . , xn ) f (∆xi ) d∆x1 . . . d∆xn
Rn i=1
Z Y  f (∆xi ) 
= h(x1 , . . . , xn ) g(∆xi ) d∆x1 . . . d∆xn
Rn i
g(∆xi )
!
Z Y  f (∆xi )  Y
= h(x1 , . . . , xn ) g(∆xi ) d∆x1 . . . d∆xn
Rn g(∆xi )
i i
" #
Y  f (∆Xi ) 
P
= E0 h(X1 , . . . , Xn )
i
g(∆Xi )
" !#
2
X η T
= EP
0 h(X1 , . . . , Xn ) exp −η ∆Xi −
i
2

η2 T
  
= EP
0 h(X1 , . . . , X n ) exp −ηX n −
2

which is consistent with our statement of Girsanov’s Theorem in (35) and (36) above. (See Remark 4.)
Introduction to Stochastic Calculus for Diffusions 17

Remark 6 (i) As in the statement of the Girsanov Theorem itself, we could have chosen µ (and therefore η)
to be adapted, i.e. to depend on prior events, in the random walk.
(ii) Note that Girsanov’s Theorem allows the drift, but not the volatility of the Brownian motion, to change
under the new measure, Qη . It is interesting to see that we are not so constrained in the case of the random
walk. Have you any intuition for why this is so?

Multidimensional Girsanov’s Theorem

The multidimensional version of Girsanov’s Theorem is a straightforward generalization of the one-dimensional
version. In particular let Wt be an n-dimensional standard P -Brownian motion and define
 Z t
1 t T
Z 
Lt := exp − ηsT dWs − ηs ηs ds
0 2 0
Rt
for t ∈ [0, T ]. Then19 W
ct := Wt +
0
ηs ds is a standard Qη -Brownian motion where dQη /dP = LT .

13 An Application: Martingale Pricing Theory

We now briefly outline how the martingale pricing theory that we have seen in a discrete-time, discrete-space
framework translates to a continuous-time setting. We use St to denote the time t price of a risky asset and Bt
to denote the time t value of the cash account. For ease of exposition we will initially assume that the risky
(s) (b)
asset does not20 pay dividends. Let φt and φt denote the number of units of the security and cash account,
respectively, that is held in a portfolio at time t. Then the value of the portfolio at time t is given by
(s) (b)
Vt = φt St + φt Bt .

(s) (b)
Definition 11 We say φt := (φt , φt ) is self-financing if
(s) (b)
dVt = φt dSt + φt dBt .

Note that this definition is consistent with our definition for discrete-time models. It is also worth emphasizing
the mathematical content of this definition. Note that Itô’s Lemma implies
(s) (b) (s) (b) (s) (b)
dVt = φt dSt + φt dBt + St dφt + Bt dφt + dSt dφt + dBt dφt
| {z }
A

so that the self-financing assumption amounts to assuming that the sum of the terms in A are identically zero.
Our definitions of arbitrage, numeraire securities, equivalent martingale measures and complete markets is
unchanged from the discrete-time setup. We now state21 without proof the two fundamental theorems of asset
pricing. These results mirror those from the discrete-time theory.

Theorem 13 (The First Fundamental Theorem of Asset Pricing) There is no arbitrage if and only if
there exists an EMM, Q.

A consequence of Theorem 13 is that in the absence of arbitrage, the deflated value process, Vt /Nt , of any
self-financing trading strategy is a Q-martingale. This implies that the deflated price of any attainable security
can be computed as the Q-expectation of the terminal deflated value of the security.
19 Again it is necessary to make some further assumptions in order to guarantee that L is a martingale. Novikov’s condition
t
is sufficient.
20 We can easily adapt our definition of a self-financing trading strategy to accommodate securities that pay dividends.
21 Additional technical conditions are generally required to actually prove these results. See Duffie’s Asset Pricing Theory or

Shreve’s Stochastic Calculus for Finance II: Continuous-Time Models for further details.
Introduction to Stochastic Calculus for Diffusions 18

Theorem 14 (The Second Fundamental Theorem of Asset Pricing) Assume there exists a security
with strictly positive price process and that there are no arbitrage opportunities. Then the market is complete if
and only if there exists exactly one risk-neutral martingale measure, Q.

Beyond their theoretical significance, these theorems are also important in practice. If our model is complete
then we can price securities with the unique EMM that accompanies it, assuming of course the absence of
arbitrage. This is the case for the Black-Scholes model as well as local volatility models. If our model is
incomplete then we generally work directly under an EMM, Q, which is then calibrated to market data. In
particular, the true data-generating probability measure, P , is often completely ignored when working with
incomplete models. In some sense the issue of completeness then only arises when we discuss replicating or
hedging strategies.

Determining Replicating Strategies

The following example is particularly useful in many financial engineering applications. While we focus on the
case of a single risky security the result generalizes easily to multiple risky securities. We assume the market is
complete of course, since only in that case can we guarantee that a replicating strategy actually exists.
Example 8 (Wealth Dynamics and Hedging)
We know the value of the cash-account, Bt , satisfies dBt = rt Bt dt and suppose in addition that St satisfies
dSt = µt St dt + σt St dWt . (37)
(s) (b) (s) (b)
Then for a portfolio (φt , φt ), the portfolio value, Vt , at time t satisfies Vt := φt St + φt Bt . If
(s) (b)
φt := (φt , φt ) is self-financing, then we have
(s) (b)
dVt = φt dSt + φt dBt

(s) (s) (b)

= φt µt St dt + φt σt St dWt + φt rt Bt dt
" #
(s) (b) (s)
φt St φ Bt φt St
= Vt µt + t rt dt + σt Vt dWt
Vt Vt Vt

= Vt [rt + θt (µt − rt )] dt + θt σt Vt dWt (38)

where θt and (1 − θt ) are the fractions of time t wealth, Vt , invested in the risky asset and cash account,
respectively, at time t.
Suppose now that Vt is the time t value of some derivative security and that we want to determine the
(s) (b)
self-financing trading strategy, (φt , φt ), that replicates it. By using Itô’s Lemma to write the dynamics of Vt
as in (38), we can immediately determine θt , the fraction of time t wealth that is invested in the risky asset. In
particular, θt is the coefficient of σt Vt dWt in the dynamics of Vt . Note also that we can work under any
probability measure we choose since different probability measures will only change the drift of Vt and not the
(s) (b)
volatility. Once we have θt we immediately have 1 − θt and then also (φt , φt ).
Remark 7 If a security pays dividends or coupons, then in the above statements we should replace the security
price with the total gain process from holding the security. For example, if a stock pays a continuous dividend
yield of q, then Yt := eqt St is the total gain process. It is what your portfolio would be worth at time t if you
purchased one unit of the security at time t = 0 and reinvested all dividends immediately back into the stock.
(y) (b)
The portfolio (φt , φt ) is then said to be self-financing if
(y) (b)
dVt = φt dYt + φt dBt
(y) qt (b)
= φt d(e St ) + φt dBt
(y) qt

(b)
= φt e dSt + qYt dt + φt dBt (39)
(s) (b)
= φt (dSt + qSt dt) + φt dBt (40)
Introduction to Stochastic Calculus for Diffusions 19
(s)
where φt = eqt φt (y) is the number of units of the risky asset held at time t. We shall make use of (40) when
we price options on dividend-paying securities.

The Role of Girsanov’s Theorem

Suppose we again adopt the setting of Example 8 and that the risky security does not pay dividends. If we
deflate as usual by the cash account then we can let Zt := St /Bt denote the deflated time t value of the risky
asset. Assume also that Wt is a P -Brownian motion where P is the true data-generating probability measure.
Then it is easy to check using Itô’s Lemma that

dZt = Zt (µt − r) dt + Zt σt dWt . (41)

But we know from the first fundamental theorem that Zt must Rbe a Q-martingale. Girsanov’s Theorem,
ct := Wt + t ηs ds is a Q-Brownian motion with
however, implies that Zt is a Q-martingale only if W 0
ηs := (µs − r)/σs . Indeed, because there is only one such process, ηs , this Q is unique22 and the market is
therefore complete by the second fundamental theorem. We then obtain that the Q-dynamics for St satisfy

dSt = rSt dt + σt St dW
ct .

Exercise 8 Suppose now that the risky asset pays a continuous dividend yield, q. Show that the Q-dynamics
of St now satisfy
dSt = (r − q)St dt + σt St dW ct .

(Hint: Remember that it is now the deflated total gains process that is now a Q-martingale.)

We can use these observations now to derive the Black-Scholes formula. Suppose the stock-price, St , has
P -dynamics
dSt = µt St dt + σSt dWt
where Wt is a P -Brownian motion. Note that we have now assumed a constant volatility, σ, as Black and
Scholes originally assumed. If the stock pays a continuous dividend yield of q then martingale pricing implies
that the price of a call option on the stock with maturity T and strike K is given by

C 0 = EQ
 −rT
(ST − K)+

0 e (42)


where log(ST ) ∼ N (r − q − σ 2 /2)T, σT . We can therefore compute the right-hand-side of (42) analytically
to obtain the Black-Scholes formula.

Exercise 9 Consider an equity model with two non dividend-paying securities, A and B, whose price processes,
(a) (b)
St and St respectively, satisfy the following SDE’s
(a) (a) (a) (1)
dSt = rSt dt + σ1 St dWt
 
(b) (b) (b) (1) (2)
p
dSt = rSt dt + σ2 St ρ dWt + 1 − ρ2 dWt

(1) (2)
where (Wt , Wt ) is a 2-dimensional Q-standard Brownian motion. We assume the cash account is the
(a) (b)
numeraire security corresponding to Q (which is consistent with the Q-dynamics of St and St ) and that the
continuously compounded interest rate, r, is constant. Use the change of numeraire technique to compute the
(a) (b)
time 0 price, C0 of a European option that expires at time T with payoff max (0, ST − ST ). (We will have
plenty of practice with Girsanov’s Theorem and changing probability measures as the course progresses.)

22 If, for example, there were two Brownian motions and only one risky security then there would be infinitely many processes,

ηs , that we could use to make Zt a martingale. In this case we could therefore conclude that the market was incomplete.
Introduction to Stochastic Calculus for Diffusions 20

Appendix A
We need to show that (29) is true. That is, we must show
h i
EQ
t
dPM
dQ X
EtPM [X] = h i . (43)
EQ
t
dPM
dQ

To do this we first recall the definition of a conditional expectation. This definition states that Y is the
PM -expected value of X conditional on Ft if Y is an Ft -measurable random variable satisfying

EPM [ξX] = EPM [ξY ] (44)

for all bounded Ft -measurable random variables, ξ. Y is then typically written as E PM [X | Ft ] ≡ EtPM [X]. We
are now in a position to prove (43). First note that the right-hand-side of (43) is indeed Ft -measurable, as
required. Now let ξ be Ft -measurable. We then have
 Qh i   Qh i 
Et X dP M
dQ Et X dP dQ
M
dPM 
EPM  h i ξ  = EQ  h i ξ
Q dPM
Et dPQ Q dPM
Et dPQ dQ
 Qh i 
Et X dP dQ
M 
dP

M
= EQ  h i ξ EtQ  (45)
EtQ dP M dQ
dPQ
  
Q Q dPM
= E Et X ξ
dQ
 
dPM
= EQ X ξ = EPM [Xξ]
dQ

and so (43) holds.

Exercises
Unless otherwise stated, you may assume that Wt is a one-dimensional Brownian motion on the probability
space, (Ω, FT , P ), and that F = {Ft : t ∈ [0, T ]} is the filtration generated by Wt . Expectations should be
taken with respect to P unless otherwise indicated.

1. (From Chapter 2 in Back )

(a) Consider a discrete partition 0 = t0 < t1 < . . . < tN = T of the time interval [0, T ] with
ti − ti−1 = T /n for each i. Consider the function

X(t) = et .
PN
Write a function23 that takes T and N as inputs, and then computes and prints 2
i=1 [∆X(ti )] , where

∆X(ti ) = X(ti ) − X(ti−1 ) = eti − eti−1 .

Hint: The sum can be computed as follows

23 In VBA, Matlab, Excel or some other language / package.
Introduction to Stochastic Calculus for Diffusions 21

sum = 0
For i = 1 To N
DeltaX = Exp(iT/N) - Exp((i-1)T/N)
sum = sum + DeltaX * DeltaX
Next i
PN
(b) Repeat part (a) for the function X(t) = t3 . In both cases what happens to i=1 [∆X(ti )]2 as
N → ∞, for a given T ?
PN
(c) Repeat part (b) to compute i=1 [∆W (ti )]2 where W is a simulated Brownian motion. For a given
T , what happens to the sum as N → ∞?
PN
(d) Repeat part (c), computing instead i=1 |∆W (ti )| where | · | denotes the absolute value. What
happens to this sum as N → ∞, for a given T ?

2. Let X be a process
 R satisfying
 dXt = Xt (µt dt + σt dWht ), X0 >0. RShow that
iYt is a martingale where
t T
Yt := Xt exp − 0 µs ds . Hence show that Xt = Et XT exp − t µs ds .

Rt Wt3 Rt
3. Use Itô’s Lemma to prove that 0
Ws2 dWs = 3 − 0
Ws ds.

4. (Hedging Strategies) Let G : Ω → R be a random variable. Ignoring technical restrictions, we can

write G as Z T
G = E[G] + θs dWs .
0

By computing Et [G] and then using Itô’s Lemma, explicitly calculate the process, θt , for:

(a) G = 1A where A = {exp(WT ) > K}.

(b) G = WT2 .

(c) G = exp(a2 T + aWT ) for a > 0.

5. (Oksendal Exercise 5.1)

(a) Show that Xt = Wt /(1 + t) solves
1 1
dXt = − Xt dt + dWt ; X0 = 0.
1+t 1+t
(b) Show that (X1,t , X2,t ) = (t, et Wt ) solves
     
dX1 1 0
= dt + dWt .
dX2 X2 eX1

6. (Oksendal Exercise 5.3)

Let (W1 , . . . , Wn ) be Brownian motion in Rn , α1 , . . . , αn constants. Solve the SDE
n
!
X
dXt = rXt dt + Xt αi dWi,t ; X0 > 0.
i=1

7. (Oksendal Exercise 5.5)

(a) Solve the Ornstein-Uhlenbeck equation dXt = µXt dt + σ dWt where µ and σ are real constants.
(Hint: Use the integrating factor e−µt and consider d(e−µt Xt ).)

(b) Find: (i) E[Xt ] (ii) Var(Xt ) (iii) Cov(Xt , Xt+s ).

Introduction to Stochastic Calculus for Diffusions 22

8. Consider the bivariate diffusion process below that describes the evolution of the security price, St , and
where (W (1) , W (2) ) is a 2-dimensional Brownian motion with correlation coefficient, ρ, i.e.,
(1) (2)
E[Wt Wt ] = ρt for all t > 0.
(1)
dSt = µt St dt + σSt dWt
(2)
dµt = k(θ − µt ) dt + σµ dWt
By direct manipulations, solve the above system of SDE’s for St .
Rt
9. Consider the stochastic process, Xt , satisfying Xt = 0 γs dWs where Ws is a standard Brownian
motion and γs is stochastic. Find an expression for Cov(Xs , Xt ) for s < t.

10. (Solving a System of Linear SDE’s (Appendix E of Duffie (1996))

Consider the stochastic differential equation
dXt = [a(t)Xt + b(t)] dt + c(t) dWt (46)
where a : [0, ∞) → RN ×N , b : [0, ∞) → RN and c : [0, ∞) → RN ×d are continuous, deterministic
functions. This linear SDE can be solved explicitly. In particular, let Φ denote the solution to the ODE
dΦ(t)
= a(t)Φ(t), Φ(0) = IN ×N .
dt
Then it may be shown that the matrix Φ(t) is non singular and the solution of (46) is
 Z t Z t 
Xt = Φ(t) X0 + Φ−1 (s)b(s) ds + Φ−1 (s)c(s) dWs , t ≥ 0. (47)
0 0

(a) What type of process is Xt ?

(b) Let m(t) := E[Xt ]. Show that m(t) satisfies the ODE
dm(t)
= a(t)m(t) + b(t), m(0) = X0 .
dt
(c) Let V (t) := Cov(Xt ). Show that V (t) satisfies the ODE
dV (t)
= a(t)V (t) + V (t)a(t)0 + c(t)c(t)0 , V (0) = 0.
dt
(d) Confirm that the solution in (47) is consistent with your solution to Exercise 8.
11. Consider the bivariate diffusion process of Exercise 8 again, and assume also that there exists a cash
account that earns interest at a constant continuously compounded rate, r.

(a) Describe the set of all possible equivalent martingale measures, Q. Is this a complete or incomplete
market?
(b) Can you compute unique prices for European options? Explain your answer.

12. Consider the following stochastic volatility model for the dynamics of a stock price
(1)
dSt = rSt dt + St Vt dWt
(1) (2)
dVt = α(Vt , t) dt + γ1 (Vt , t) dWt + γ2 (Vt , t) dWt
(1) (2)
where (Wt , Wt ) is a standard Q-Brownian motion. A particular derivative security pays dividends at
the rate h(St , t) for s ∈ [0, T ] and upon expiration at T has a final payout of F (ST ). The time t price of
this security is therefore given by
"Z #
T
Q
Ct = Et φt (u)h(Su , u) du + φt (T )F (ST )
t
Introduction to Stochastic Calculus for Diffusions 23

where  Z u 
φt (u) = exp − rs ds
t

and rs is the short-rate value at time s. (Note that this follows from martingale pricing with cash account
as numeraire.) Use the Feynman-Kac approach to write a PDE that is satisfied by Ct assuming that the
short-rate is a deterministic process.

13. (a) Without using Novikov’s condition, show that exp (Wt − t/2) is a P -martingale.

(b) Define Q by
Q(A) := E P [exp (WT − T /2) 1A ]
for all A ∈ FT . Verify that Q is a probability measure, and compute the Q- and P -probabilities that
exp (WT ) ≥ K for some K > 0.

(c) Find a Q-Brownian motion, W̃ , and a process At , such that Wt = W̃t + At .

14. Assume the price of a European call option is given by

 
1
C 0 = EQ 0 (ST − K)+
BT

where (BT , Q) is the usual cash-account, risk-neutral pair. Find a pair of measures, Q(1) and Q(2) , that
are equivalent to Q and such that

C0 = S0 Q(1) (ST > K) − KZ0T Q(2) (ST > K)

where Z0T is the time 0 value of the zero-coupon bond maturing at time T .

15. Write a simple Monte-Carlo simulation to verify

η2 T
  
Q P
θ := E0 [h(X)] = E0 h(X1 , . . . , Xn ) exp −ηXn − .
2

Your code should take n, T , µ and m as inputs where m is the number of Monte-Carlo samples. Ideally
the function h(·) should also be an input to your function. Test that your code is correct by running it
with different test functions, h(·).

16. Referring to Section 11, carefully explain and justify every step in going from equation (32) to equation
(34). Your explanations should refer to the results from Section 10.