Stochastic Calculus for Finance II

-
some Solutions to Chapter V
Matthias Thul∗
Last Update: June 19, 2015

Exercise 5.1

(i) Let f (t, x) = S(0)ex . We have

∂f ∂f ∂ 2f
= 0, = f (t, x) = f (t, x)
∂t ∂x ∂x2

and

 
1 2
dX(t) = α(t) − R(t) − σ (t) dt + σ(t)dW (t)
2
(dX(t))2 = σ 2 (t)dt

By the Itô formula, the differential of the discounted stock price D(t)S(t) is given
by

d (D(t)S(t)) = df (t, X(t))

1
= D(t)S(t)dX(t) + D(t)S(t) (dX(t))2
2
= (α(t) − R(t)) d(t)S(t)dt + σ(t)D(t)S(t)dW (t) (q.e.d.)

(ii) We first note that the cross variation dD(t)dS(t) = 0, since D(t) is a nonrandom
function of time and thus has zero quadratic variation. We thus obtain
∗
The author can be contacted via <<firstname>>.<<lastname>>@gmail.com and
http://www.matthiasthul.com.

1
 d (D(t)S(t)) = −R(t)D(t)S(t)dt + α(t)D(t)S(t)dt + σ(t)D(t)S(t)dW (t)

= (α(t) − R(t)) d(t)S(t)dt + σ(t)D(t)S(t)dW (t) (q.e.d.)

Exercise 5.2 (State Price Density Process)

This assertion follows from Equation (5.2.30) and Lemma 5.2.2.

1
D(t)V (t) = Ẽ [D(T )V (T )| F(t)] = E [D(T )Z(T )V (T )| F(t)]
Z(t)
and thus

D(t)Z(t)V (t) = E [D(T )Z(T )V (T )| F(t)] . (q.e.d.)

Exercise 5.3

(i) Differentiating inside the expected value and applying the chain rule yields

    
−rT 1 2
cx (0, x) = Ẽ e I{x exp{σW̃ (T )+(r− 1 σ2 )T }>K } exp σ W̃ (T ) + r − σ T
2 2
  
1
= Ẽ I{S(T )>K} exp σ W̃ (T ) − σ 2 T .
2

Here, we have defined S(t) by

   
1 2
S(t) = x exp σ W̃ (t) + r − σ t
2

(ii) Let P̂ be a probability measure equivalent to P̃ and let Z(t) be a Radon-Nikodým.

By Lemma 5.2.1, By Lemma 5.2.1, the expected value in of an F(t)-measureable
random variable Y satisfies

Ê [Y (t)] = Ẽ [Z(t)Y (t)]

We note that I{S(T )>K} is F(T )-measurable and can thus write

2
  
P̂ (S(T ) > K) = Ê I{S(T )>K}
 
= Ẽ Z(T )I{S(T )>K}
  
1 2
= Ẽ I{S(T )>K} exp σ W̃ (T ) − σ T
2
= cx (0, x) (q.e.d.) (1)

Here, we have defined Z(t) to be

 
1 2
Z(t) = exp σ W̃ (t) − σ t .
2

By Girsanov’s theorem (Theorem 5.2.3), this is just the Radon-Nikodým derivative

process that renders a P̃-Brownian motion into a P̂-Brownian motion if

Ŵ (t) = W̃ (t) − σt.

Thus, the Z(t) fulfills all required properties such that the change of measure in
Equation (1) is defined.

(iii) Substituing for S(T ) and using W̃ (T ) = Ŵ (T ) + σT yields

       
1 2
P̂ (S(T ) > K) = P̂ x exp σ Ŵ (T ) + σT + r − σ T > K
2
    
1 K
= P̂ σ Ŵ (T ) + r + σ 2 T > ln
2 x
K

1 2

!
Ŵ (T ) ln x − r + 2 σ T
= P̂ √ > √
T σ T
!
Ŵ (T )
= P̂ − √ < d+ (T, x)
T

Since − Ŵ√(T
T
)
∼ N (0, 1), we finally obtain

P̂ (S(T ) > K) = N (d+ (T, x)) (q.e.d.)

3
Exercise 5.4

(i) Let f (t, x) = ln x. We have

∂f ∂f 1 ∂ 2f 1
= 0, = , 2
=− 2
∂t ∂x x ∂x x
and

(dS(t))2 = σ 2 (t)S 2 (t)dt.

Applying Itô’s lemma, the differential of the log stock price d ln S(t) becomes

d ln S(t) = df (t, S(t))

1 1 1
= dS(t) − (dS(t))2
S(t) 2 S 2 (t)
1
= r(t)dt + σ(t)dW̃ (t) − σ 2 (t)dt
  2
1 2
= r(t) − σ (t) dt + σ(t)dW̃ (t).
2

In integral form, we get

Z t  Z t
1 2
ln S(t) = ln S(0) + r(s) − σ (s) ds + σ(s)dW̃ (s). (2)
0 2 0

Taking the exponential yields

S(t) = S(0)eX(t)

where

Z t  Z t
1 2
X(t) = r(s) − σ (s) ds + σ(s)dW̃ (s).
0 2 0

By Theorem 4.4.9, Itô integrals of a deterministic integrand are normally distributed

with zero mean. Thus, X(t) is normally distributed with

Z t   Z t 
1 2 2
Xt ∼ N r(s) − σ (s) ds, σ (s)ds (3)
0 2 0

4
(ii) Since the payoff of a European call option only depends on the asset price ST at
maturity, two different diffusion processes for the underlying that have imply a
common risk-neutral density for ST at time T will yield the same option prices. By
setting r(t) = R and σ(t) = Σ in Equation (3), we see that the distribution of Xt
under constant interest rate and volatility is given by

Z t   Z t 
1 2 2
Xt ∼ N R − Σ ds, Σ ds
0 2 0
  
1 2 2
∼ N R − Σ T, Σ T . (4)
2

We now equate the mean and variance in Equations (3) and (4) to obtain

sZ
Z t t
1
Σ2 T = σ 2 (s)ds ⇔ Σ= σ 2 (s)ds
0 T 0

and

  Z t  Z t
1 2 1 2
R− Σ T = r(s) − σ (s) ds ⇔ R= r(s)ds
2 0 2 0

Rt qR
t
It follows that we can replace R by 0
r(s)ds and Σ by 0
σ 2 (s)ds in the Black-
Scholes formula for European call options to obtain the value under deterministic
interest rates and volatilities.

Exercise 5.5

(i) Let f (t, x) = x1 . We have

∂f ∂f 1 ∂ 2f 2
= 0, = − 2, 2
= 3
∂t ∂x x ∂x x

and thus

 
1 1 1 2
d = − dZ(t) + (dZ(t))2
Z(t) Z 2 (t) 2 Z 3 (t)
Θ(t) Θ2
= dW (t) + dt.
Z(t) Z(t)

5
 Here, we have used that

dZ(t) = −Θ(t)Z(t)dW (t),

as shown in the proof of Theorem 5.2.3.

(ii) As shown in the proof of Theorem 5.2.3, Z(t) is a Radon-Nikodým derivative process.
Thus, Lemma 5.2.2 applies and we get

h i h i
E M̃ (t)Z(t) F(s) = Z(s)Ẽ M̃ (t) F(s) = Z(s)M (s) (q.e.d.).

Here, the first equality follows by multiplying both sides of the equality in Lemma
5.2.2 by Z(s) and in the second step we use that M̃ (t) is a martingale under P̃.

(iii) By the Itô product rule (Corollary 4.6.3), we get

 
M (t)
dM̃ (t) = d
Z(t)
   
1 1 1
= M (t)d + dM (t) + dM (t)d
Z(t) Z(t) Z(t)
M (t) 2
  Γ(t) Γ(t)Θ(t)
= Θ (t)dt + Θ(t)dW (t) + dW (t) + dt
Z(t) Z(t) Z(t)
1 
M (t)Θ2 (t) + Γ(t)Θ(t) dt + (M (t)Θ(t) + Γ(t)) dW (t) .


=
Z(t)

(iv) Corollary 5.3.2 defines

dW̃ (t) = dW (t) + Θ(t)dt

and substituting for W (t) in the differential of M̃ (t) yields

1 h 2

 i
dM̃ (t) = M (t)Θ (t) + Γ(t)Θ(t) dt + (M (t)Θ(t) + Γ(t)) dW̃ (t) − Θ(t)dt
Z(t)
1
= (M (t)Θ(t) + Γ(t)) dW̃ (t)
Z(t)
 
Γ(t)
= M̃ (t)Θ(t) + dW̃ (t).
Z(t)

6
 We now define

Γ(t)
Γ̃(t) = M̃ (t)Θ(t) +
Z(t)

and integrate to obtain

Z t
M̃ (t) = M̃ (0) + Γ̃(u)dW̃ (u) (q.e.d.).
0

Note, that Γ(t) is adapted to the filtration F(t) as required by Corollary 5.3.2, since
M̃ (t), Θ(t), Γ(t) and Z(t) are adapted processes.

Exercise 5.6

We will proof the more general case of the multidimensional Girsanov theorem, i.e. The-
orem 5.4.2 for any d ∈ {1, 2, . . .}. We note, that by the two dimensional Lévy theorem
(Theorem 4.6.5) two continuous martingales M1 (t) and M2 (t) with respect to some filtra-
tion F(t) that start at zero and have unit quadratic variation and zero cross variation are
independent Brownian motions. If we thus consider the multidimensional martingale

M (t) = (M1 (t), M2 (t), . . . , Md (t)) ,

then we require that the conditions of the two dimensional Lévy theorem are satisfied
for any two Mi (t) and Mj (t) where i, j ∈ {1, 2, . . . , d}. I.e. we want to show that

dt if i = j

dMi (t)dMj (t) =
0

otherwise
or equivalently in matrix notation

dM (t)dM 0 (t) = Id dt,

where 0 is used to denote the vector/matrix transpose and Id is the identity matrix of
dimension d.

(i) Continuity:

7
 Clearly,

Z t
W̃ (t) = W (t) + Θ(u)du
0

is continuous since by Definition 3.3.1 the Brownian motion W (t) has continuous
sample paths. Similarly, the ordinary Lebesgue integral is a continuous function of
the upper limit of integration.

(ii) Starting at zero:

It is obvious that W̃ (0) = W (0) = 0.

(iii) Unit quadratic and zero cross variation:

We have

dW̃ (t)dW̃ 0 (t) = (dW (t) + Θ(t)dt) (dW (t) + Θ(t)dt)0

= (dW (t) + Θ(t)dt) (dW 0 (t) + Θ0 (t)dt)

= dW (t)dW 0 (t) = Id dt (q.e.d.).

(iv) Martingale property:

We first define

Z t Z t
1
X(t) = − Θ(u) · dW (u) − ||Θ(u)||2 du
0 2 0

such that

1
dX(t) = −Θ(t) · dW (t) − ||Θ(t)||2 dt
2

and

8
 d d
!2
X 1X 2
dX(t)dX(t) = − Θj (t)dWj (t) − Θ (t)dt
j=1
2 j=1 j
d X
X d
= Θj (t)Θk (t)dWj (t)dWk (t)
j=1 k=1
d
X
= Θ2j (t)dt
j=1

= ||Θ(t)||2 dt.

In analogy to the proof of Theorem 5.2.3 (Girsanov, one dimension), we now define
f (t, x) = ex and apply the Itô formula to obtain the differential of Z(t) as

1
dZ(t) = Z(t)dX(t) + Z(t)dX(t)dX(t)
2
1 1
= −Z(t)Θ(t) · dW (t) − Z(t)||Θ(t)||2 dt + Z(t)||Θ(t)||2 dt
2 2
= −Z(t)Θ(t) · dW (t).

Since the differential of Z(t) contains no dt term, it follows that Z(t) is a P martingale
with E [Z(T )] = Z(0) = 1 and thus qualifies as a Radon-Nikoým derivative process.
The differential of W̃ (t)Z(t) can be computed using the Itô product rule (Corollary
4.6.3) as

 
d W̃ (t)Z(t) = W̃ (t)dZ(t) + Z(t)dW̃ (t) + dW̃ (t)dZ(t)

= −W̃ (t)Z(t)Θ(t) · dW (t) + Z(t) (dW (t) + Θ(t)dt)

+ (dW (t) + Θ(t)dt) (−Z(t)Θ(t) · dW (t))

= −W̃ (t)Z(t)Θ(t) · dW (t) + Z(t)dW (t) + Z(t)Θ(t)dt

−Z(t)Θ(t)dt

= −W̃ (t)Z(t)Θ(t) · dW (t) + Z(t)dW (t)

 
= Z(t) −W̃ (t) · Θ(t) + 1 dW (t).

By the same argument at before, it follows that W̃ (t)Z(t) is a P martingale. Using

Lemma 5.2.2, we obtain

9
 h i 1 h i
Ẽ W̃ (t) F(s) = E W̃ (t)Z(t) F(s)
Z(s)
1
= W̃ (s)Z(s) = W̃ (s) (q.e.d.).
Z(s)

Thus, W̃ (t) is a d-dimensional P̃ martingale.

Since all conditions of the multidimensional Lévy theorem are satisfied, we conclude
that W̃ (t) is a d-dimensional Brownian motion under P̃.

Exercise 5.7

(i) We can obtain this portfolio process by investing the amount X2 (0) > 0 in the
money market such that the value at time T is

Z T 
X2 (0)
X2 (0) exp R(t)dt = .
0 D(T )
In addition, we invest in the portfolio process X1 (t). The initial cost of setting up
this portfolio are X2 (0) + X1 (0) = X2 (0). At time T , we have

   
X2 (0) X2 0 X2 (0)
P X2 (T ) ≥ =P + X1 (T ) ≥ = P {X1 (T ) ≥ 0} = 1
D(T ) D(T ) D(T )

and

   
X2 (0) X2 0 X2 (0)
P X2 (T ) > =P + X1 (T ) > = P {X1 (T ) > 0} > 0.
D(T ) D(T ) D(T )

(ii) The same idea as in (i) applies here. We can obtain the portfolio process X1 (t) by
borrowing the amount X2 (0) in the money market and investing the proceeds in the
portfolio process X2 (t). At time T , we have to repay X2 (0)/D(T ) and obtain

   
X2 (0) X2 (0)
P {X1 (T ) ≥ 0} = P X2 (T ) − ≥ 0 = P X2 (T ) ≥ =1
D(T ) D(T )
and

   
X2 (0) X2 (0)
P {X1 (T ) > 0} = P X2 (T ) − > 0 = P X2 (T ) > > 0.
D(T ) D(T )

10
Exercise 5.8 (Every strictly positive Asset is a generalized geo-
metric Brownian Motion)

(i) We have that

D(t)V (t) = Ẽ [D(T )V (T )| F(t)] .

is a P̃-martingale since for s < t

h i
Ẽ [D(t)V (t)| F(s)] = Ẽ Ẽ [ D(T )V (T )| F(t)] F(s)

= Ẽ [D(T )V (T )| F(s)]

= D(s)V (s).

By Corollary 5.3.2, there exists a an adapted process Γ̃(t) such that

Z t
D(t)V (t) = V (0) + Γ̃(u)dW̃ (u)
0

or in differential form

d(D(t)V (t)) = Γ̃(t)dW̃ (t).

By the Itô product rule,

dV (t) = D−1 (t)d(D(t)V (t)) + D(t)V (t)dD−1 (t) + d(D(t)V (t))dD−1 (t)
Γ̃(t)
= R(t)V (t)dt + dW̃ (t).
D(t)

Here we used that

dD−1 (t) = R(t)D−1 (t)dt

is a process of finite variation and thus the cross variation term in the Itô differential
drops out.

11
 (ii) We define a new random variable X = 0 with E[X] = 0. Note that the constant
X satisfies Definition 1.2.1 for random variables. Since both D(T ) and V (T ) are
P̃-almost surely positive, it follows that X < D(T )V (T ) P̃-almost surely. Conse-
quently, 0 = E[X] < E[D(T )V (T )] = D(t)V (t). This does not follow directly from
Theorem 1.3.1 but is a straight forward modification of it. Now, since D(t) is positive
P̃-almost surely, it follows that V (t) is too.

(iii) Since V (t) is P̃-almost surely positive, we can write its differential as

Γ̃(t)
dV (t) = R(t)V (t)dt + V (t)dW̃ (t)
D(t)V (t)
= R(t)V (t)dt + σ(t)V (t)dW̃ (t),

where we defined σ(t) = Γ̃(t)/(D(t)V (t)). σ(t) is adapted to the filtration F(t) since
all three processes that define it are.d

Exercise 5.9 (Implying the risk-neutral distribution)

We apply the Leibniz integral rule to obtain

 Z ∞ 
∂ −rT
cK = e (y − K)p̃(0, T, x, y)dy
∂K K
Z ∞
−rT ∂
= e (y − K)p̃(0, T, x, y)dy
K ∂K
Z ∞
−rT
= −e p̃(0, T, x, y)dy.
K

Applying the Leibniz integral rule for a second time yields

cKK = e−rT p̃(0, T, x, K).

Thus, the implied risk-neutral density can be recovered from a continuum of call prices
by

p̃(0, T, x, K) = erT cKK .

12
Exercise 5.10 (Chooser Option)

i) At time t0 (choice date) the holder of a long position in a chooser option has the right
to choose whether his option is a call or a put. If he acts rationally, he will always
choose the higher priced option. Thus, the price V (t0 ) of a chooser option at t = t0
has to be the greater of the call option price C (t0 ) and the put option price P (t0 ).
Using put-call parity we obtain

V (t0 ) = max {C (t0 ) , P (t0 )}

= max {C (t0 ) , C (t0 ) − F (t0 )}

= C (t0 ) + max {0, −F (t0 )}

+
= C (t0 ) + e−r(T −t0 ) K − S (t0 ) . (q.e.d.)

ii) By the risk-neutral pricing formula we have

V (0) = EQ e−rt0 V (t0 )

 
h 
+ i
= EQ e−rt0 C (t0 ) + e−r(T −t0 ) K − S (t0 )
 −rt0 Q  −r(T −t0 ) h
+ i
−rt0 −r(T −t0 )

Q
= E e E e C(T ) F (t0 ) + E e
Q
e K − S (t0 )

= C(0) + EQ e−rt0 max 0, e−r(T −t0 ) K − S (t0 )

  

= C(0, T, K) + P (0, t0 , e−r(T −t0 ) K). (q.e.d)

While the first term is the current value of a call option with strike price K and
maturity in T , the second term is the current value of a put option with strike price
e−r(T −t0 ) K and maturity in t0 . Thus the value of a chooser option is equal to a
portfolio consisting of a long call and a long put with different strike prices and
expiry dates.

Exercise 5.11 (Hedging a Cash Flow)

Following the hint, we start by defining the market price of risk by

α(t) − R(t)
Θ(t) = .
σ(t)

13
 The process Θ(t) is adapted to the filtration F(t) since all three processes that define
it are. It is well defined since σ(t) is strictly positive. Next, we define

 Z t
1 t 2
Z 
Z(t) = exp − Θ(u)dW (u) − Θ (u)du
0 2 0
and

Z t
W̃ (t) = W (t) + Θ(u)du.
0

By Girsanov’s theorem (Theorem 5.2.3), W̃ (t) is a Brownian motion under the prob-
ability measure P̃ defined by

Z
P̃(A) = Z(ω)dP (ω)
A

for all A ∈ F. Further following the hint, we define

Z T 
M̃ (t) = Ẽ D(u)C(u)du F(t) .
0

Then, for 0 ≤ s ≤ t we get

h i  Z T  
Ẽ M̃ F(s) = Ẽ Ẽ D(u)C(u)du F(t) F(s)
0
Z T 
= Ẽ D(u)C(u)du F(s)
0
= M̃ (s)

and thus M̃ (t) is a P̃-martingale. Since the filtration F(t) is the one generated by the
Brownian motion W (t), we can apply Corollary 5.3.2. There exists an adapted process
Γ̃(t) such that

Z t
M̃ (t) = M̃ (0) + Γ̃(u)dW̃ (u).
0

Using Itô’s product rule, we now compute the differential of the discounted portfolio
value under P as

14
d(D(t)X(t)) = D(t)dX(t) + X(t)dD(t) + dD(t)dX(t)

= D(t)∆(t)dS(t) + R(t)D(t)(X(t) − Delta(t)S(t))dt − D(t)C(t)dt − R(t)D(t)X(t)

= ∆(t)D(t)(dS(t) − R(t)S(t)dt) − D(t)C(t)dt

= ∆(t)D(t)S(t)((α(t) − R(t))dt + σ(t)dW (t)) − D(t)C(t)dt

  
= ∆(t)D(t)S(t) (α(t) − R(t))dt + σ(t) W̃ (t) − Θ(t)dt − D(t)C(t)dt

= ∆(t)σ(t)D(t)S(t)dW̃ (t) − D(t)C(t)dt.

In integral form, we have

Z T Z T
D(T )X(T ) = X(0) + ∆(u)σ(u)D(u)S(u)dW̃ (u) − D(u)C(u)du.
0 0

where we used that D(0) = 1. Choose X(0) = M̃ (0) and

Γ̃(t)
∆(t) = ,
σ(t)D(t)S(t)
then

Z T Z T
D(T )X(T ) = M̃ (0) + Γ̃(u)dW̃ (0) − D(u)C(u)du
0 0
Z T
= M̃ (0) + M̃ (T ) − M̃ (0) − D(u)C(u)du
0
Z T  Z T
= Ẽ D(u)C(u)du F(T ) − D(u)C(u)du
0 0
= 0.

Here, we used that the integral in the second last equality is F(T )-measurable. Since
D(T ) is P̃-almost surely positive, it follows that X(T ) = 0 P̃-almost surely. It follows
that if we start with an initial wealth of

Z T 
X(0) = Ẽ D(u)C(u)
0

and choose the portfolio process given by ∆(t) above, then we can hedge the random
cash-flow P̃-almost surely.

15
Exercise 5.12 (Correlation under Change of Measure)

(i) Remember that Bi (t) is defined by

d Z t
X σij (u)
Bi (t) = dWj (u).
j=1 0 σi (u)

We have

d Z t 
X σij (u)
B̃i (t) = dWj (u) + γi (u)du
i=1 0
σi (u)
d Z t
X σij (u)
= (dWj (u) + Θ(u)du)
i=1 0
σi (u)
d Z t
X σij (u)
= dW̃j (u).
i=1 0
σi (u)

It has been shown using Lévy’s theorem (Theorem 4.6.4) that Bi (t) is a Brownian
motion under P. Repeating the same proof but replacing Wj (t) by W̃j (t) it can be
shown that B̃i (t) is a Brownian motion under P. Note that by Theorem 4.3.1, each
of the Itô integrals in the definition of B̃i (t) is a continuous martingale starting at
zero and thus B̃i (t) is too. Its quadratic variation is

d
X σij2 (t)
dB̃i (t)dB̃i (t) = dt
j=1
σi2 (t)
= dt,

where we used that by definition

v
u d
uX
σi (t) = t σ 2 (t). ij
j=1

As anticipated, it follows by Theorem 4.6.4 that B̃i (t) is a P̃-martingale.

(ii) We have

16
 dSi (t) = αi (t)Si (t)dt + σi (t)Si (t)dBi (t)
d
X σij (t)
= αi (t)Si (t)dt + σi (t)Si (t) dWj (t)
j=1
σi (t)
d
X σij (t)  
= αi (t)Si (t)dt + σi (t)Si (t) dW̃j (t) − Θj (t)dt
j=1
σi (t)
d
!
X σij (t)Θj (t)
= αi (t)Si (t)dt + σi (t)Si (t) dB̃i (t) − dt
j=1
σi (t)
d
!
X
= αi (t) − σij (t)Θj (t) Si (t)dt + σi (t)Si (t)dB̃t
j=1

= R(t)Si (t)dt + σi (t)Si (t)dB̃i (t) (q.e.d.).

Here, we used the expression previously obtained for B̃i (t) in equality four. The last
equality follows from the market price of risk equations

d
X
αi (t) − R(t) = σij (t)Θj (t).
j=1

(iii) We have for i 6= k

dB̃i (t)dB̃k (t) = (dBi (t) + γi (t)dt) (dBk (t) + γk (t)dt)

= dBi (t)dBk (t)

= ρik (t)dt (q.e.d.).

Here, we used that the covariation between a process of bounded variation and any
other process is zero in the second step.

(iv) Using the Itô product rule, we get for the first expectation

17
  Z t 
E [Bi (t)Bk (t)] = Ẽ Bi (0)Bk (0) + d (Bi (u)Bk (u))
0
Z t 
= Ẽ (Bi (u)dBk (u) + Bk (u)dBi (u) + dBi (u)dBk (u))
0
Z t 
= Ẽ dBi (u)dBk (u)
0
Z t 
= Ẽ ρik (u)du
0
Z t
= ρik (u)du.
0

Here, we used that Bi (0) = Bk (0) = 0 in the second equality. By Theorem 4.3.1,
the Itô integrals in the third equality are martingales starting at zero and have thus
zero expected value. In the last step, we can drop the expectation operator since
the function ρik (t) is deterministic. The same steps can be repeated to show that

h i Z t
Ẽ B̃i (t)B̃k (t) = ρik (u)du
0

and the proof is complete.

(v) Again using the Itô product rule, we get for the first expectation

 Z t 
E [B1 (t)B2 (t)] = Ẽ B2 (0)B2 (0) + d (B1 (u)B2 (u))
0
Z t 
= Ẽ (B1 (u)dB2 (u) + B2 (u)dB1 (u) + dB1 (u)dB2 (u))
0
Z t 
= Ẽ dB1 (u)dB2 (u)
0
Z t 
= Ẽ sign (W1 (u)) du
0
Z t
= E [sign (W1 (u))] du
0
Z t
= (P {W1 (u) ≥ 0} − P {W1 (u) < 0}) du
0
= 0.

Skipping some intermediate steps, we get for the second expectation

18
 h i Z t
Ẽ B̃1 (t)B̃2 (t) = Ẽ [sign (W1 (u))] du
0
Z t   Z u 
= Ẽ sign W̃1 (u) − Θ1 (v)dv du
0 0
Z t h  i
= Ẽ sign W̃1 (u) − u du
Z0 t  n o n o
= P̃ W̃1 (u) ≥ u − P̃ W̃1 (u) < u du
0
≤ 0.

Here, strict inequality holds for all t > 0. Consequently, for t > 0,

h i
E [B1 (t)B2 (t)] 6= Ẽ B̃1 (t)B̃2 (t) (q.e.d.).

Exercise 5.13

(i) We have

 Z t 
Ẽ [W1 (t)] = Ẽ W̃1 (t) − Θ1 (u)du
0
= 0,

where the last equality follows from W̃1 being a P̃-martingale with zero initial value
and Θ1 (t) = 0 for all t ≥ 0. Similarly,

 Z t 
Ẽ [W2 (t)] = Ẽ W̃2 (t) − Θ2 (u)du
0
 Z t 
= Ẽ − W1 (u)du
0
 Z t  
= Ẽ − W̃1 (u) − Θ1 (u) du
0
 Z t 
= Ẽ − W̃1 (u)du .
0

To evaluate the remaining expectation, we define a function f (t, x) = tx where

∂f ∂f ∂ 2f
= x, = t, = 0.
∂t ∂x ∂x2

19
 Applying Itô’s formula yields the differential

   
d tW̃1 (t) = df t, W̃1 (t)

= W̃1 (t)dt + tdW̃1 (t).

Thus, in integral form and after rearranging, we get

Z t Z t
− W̃1 (u)du = udW̃1 (u) − tW̃1 (t).
0 0

Substituting back yields

Z t 
Ẽ [W2 (t)] = Ẽ udW̃1 (u) − tW̃1 (t)
0
= 0 (q.e.d.).

Here, we used that by Theorem 4.3.1, both the Itô integral w.r.t. W̃ (t) and W̃ (t)
have zero expectation under P̃.

(ii) We have

g [W1 (T ), W2 (T )] = Ẽ [W1 (T )W2 (T )] − Ẽ [W1 (T )] Ẽ [W2 (T )]

Cov

= Ẽ [W1 (T )W2 (T )] .

Here, the second equality follows from the result in (i). Now using the hint, we have
in integral form

Z T Z T
W1 (T )W2 (T ) = W1 (u)dW2 (u) + W2 (u)dW1 (u)
0 0
Z T   Z T
= W1 (u) dW̃2 (u) − W1 (u)du + W2 (u)dW̃1 (u)
0 0
Z T Z T Z T
2
= W1 (u)dW̃2 (u) − W̃1 (u)du + W2 (u)dW̃1 (u).
0 0 0

Thus,

20
 Z T Z T Z T 
Cov
g [W1 (T ), W2 (T )] = Ẽ W1 (u)dW̃2 (u) − W̃12 (u)du + W2 (u)dW̃1 (u)
0 0 0
 Z T 
2
= Ẽ − W̃1 (u)du
0
Z T h i
2
= − Ẽ W̃1 (u) du
0
Z T
= − udu
0
1
= − T2 q.e.d..
2

Here, we again used the martingale property of the two Itô integrals w.r.t. W̃1 (t)
and W̃2 (t) under P̃ in the second equality.

Exercise 5.14 (Cost of Carry)

(i) The differential of the discounted portfolio value is given by

d(D(t)X(t)) = D(t)dX(t) + X(t)dD(t) + dD(t)dX(t)

= ∆(t)D(t)dS(t) − a∆(t)D(t)dt + rD(t)(X(t) − ∆(t)S(t))dt − rD(t)X(t)

= ∆(t)D(t) (dS(t) − rS(t)dt) − a∆(t)D(t)dt

= ∆(t)σD(t)S(t)dW̃ (t).

Since W̃ (t) is a P̃-martingale, the claim follows.

(ii) Let f (t, x) = ex where

∂f ∂f ∂ 2f
= 0, = ex , = ex
∂t ∂x ∂x2

and define

 
1 2
X(t) = r − σ t + σ W̃ (t).
2

Then,

21
 dY (t) = df (t, X(t))
1
= Y (t)dX(t) + Y (t)dX(t)dX(t)
  2
1 2 1
= r − σ Y (t)dt + σY (t)dW̃ (t) + σ 2 dt
2 2
= rY (t)dt + σY (t)dW̃ (t) (q.e.d.).

Furthermore,

d(D(t)Y (t)) = D(t)dY (t) + Y (t)dD(t) + dD(t)dY (t)

= rD(t)Y (t)dt + σD(t)Y (t)dW̃ (t) − rD(t)Y (t)dt

= σD(t)Y (t)dW̃ (t) (q.e.d.).

Since the differential contains no dt-term, it follows that D(t)Y (t) is a P̃-martingale.
Finally, from the second definition of S(t), we obtain the differential form by applying
the Itô product rule

Z t
 Z t  Z t 
a a a
dS(t) = S(0)dY (t) + Y (t)d ds + ds dY (t) + dY (t)d ds
0 Y (s) 0 Y (s) 0 Y (s)
 Z t 
a a
= S(0) + ds dY (t) + Y (t) dt
0 Y (s) Y (t)
 Z t   Z t 
a a
= r S(0) + ds Y (t)dt + σ S(0) + ds Y (t)dW̃ (t) + adt
0 Y (s) 0 Y (s)
= rS(t)dt + σS(t)dW̃ (t) + adt (q.e.d.).

We could have directly obtained the solution to the linear stochastic differential
equation for S(t) using the (yet to be obtained) result from Exercise 6.1. Using the
notation of that exercise, we have

dS(t) = (a(t) + b(t)X(t))du + (γ(t) + σ(t)S(t))dW̃ (t)

where

a(t) = a, b(t) = r, γ(t) = 0, σ(t) = σ.

22
 Let

Z t Z t 
1
A(t) = exp σ(v)dW̃ (v) + b(v) − σ 2 (v)dv
2
 0  0
 
1
= exp σ W̃ (t) + r − σ 2 t
2

and

t t
a(v) − σ(v)γ(v)
Z Z
γ(v)
B(t) = S(0) + dv + dW̃ (v)
0 A(v) 0 Z(v)
Z t
a
= S(0) + dv.
0 A(v)

Then the solution to the stochastic differential equation is

S(t) = A(t)B(t)
Z t
a
= S(0)A(t) + A(t) dv.
0 A(v)

This coincides with the result that we just proved.

(iii) Following the hint, we get

 Z T
a
Ẽ [S(T )| F(t)] = Ẽ S(0)Y (T ) + Y (T ) ds F(t)
0 Y (s)
Z t
a
= S(0)Ẽ [Y (T )| F(t)] + Ẽ [Y (T )| F(t)] ds
0 Y (s)
 Z T 
a
+Ẽ Y (T ) ds F(t)
t Y (s)
Z t
a
= S(0)Ẽ [Y (T )| F(t)] + Ẽ [Y (T )| F(t)] ds
0 Y (s)
Z T  
Y (T )
+a Ẽ F(t) ds.
t Y (s)

We further have

23
    
1 2
Ẽ [Y (T )| F(t)] = Ẽ exp r − σ T + σ W̃ (t)
2
    
1 2 
= Ẽ Y (t) exp r − σ (T − t) + σ W̃ (T ) − W̃ (t) F(t)
2
  
r(T −t) 1 2 
= Y (t)e Ẽ exp − σ (T − t) + σ W̃ (T ) − W̃ (t)
2
  
r(T −t) 1 2
= Y (t)e Ẽ exp − σ (T − t) + σ W̃ (T − t)
2
= Y (t)er(T −t) .

The third equality uses that Y (t) is F(t)-measurable and that the increment W̃ (T )−
W̃ (t) is independent of the σ-algebra F(t) to drop the conditioning. In the fourth
equality we exploit the time homogeniety of the Brownian motion, i.e. that the law
of W̃ (T ) − W̃ (t) under P̃ is the same as that of W̃ (T − t). Finally, we use that

 
−rt 1
e Y (t) = exp − σ 2 t − σ W̃ (t)
2
is a P̃-martingale with an initial value of one as shown in (ii). Next, for t ≤ s ≤ T
and using similar steps we have

     
Y (T ) 1 2 
Ẽ F(t) = Ẽ exp r − σ (T − s) + σ W̃ (T ) − W̃ (s)
Y (s) 2
  
r(T −s) 1 2
= e Ẽ exp − σ (T − s) + σ W̃ (T − s)
2
= er(T −s) .

Thus,

Z t Z T
r(T −t) r(T −t) a
Ẽ [ S(T )| F(t)] = S(0)Y (t)e + Y (t)e ds + a er(T −s) ds
0 Y (s) t
a s=T
= S(t)er(T −t) − er(T −s)
r s=t
a
= S(t)er(T −t) − 1 − er(T −t) .


r

(iv) Let f (t, x) = xer(T −t) − ar 1 − er(T −t) where

∂f ∂f ∂ 2f
= −rxer(T −t) − aer(T −t) , = er(T −t) , = 0.
∂t ∂x ∂x2

24
 An application of Itô’s lemma yields the differential

 a

d S(t)er(T −t) − 1 − er(T −t)
r
r(T −t)
− aer(T −t dt + rS(t)er(T −t) dt + σS(t)er(T −t) dW̃ (t) + aer(T −t) dt


= −rS(t)e

= σS(t)er(T −t) dW̃ (t).

Since the differential contains no dt-term it follows that Ẽ [S(T )| F(t)] is a martingale
under P̃. This is also obvious by letting s ≤ t ≤ T and directly computing

h
r(T −t) a r(T −t)

i h i
Ẽ S(t)e − 1−e F(s) = Ẽ Ẽ [S(T )| F(t)] F(s)
r
= Ẽ [ S(T )| F(s)]
a
= S(s)er(T −s) − 1 − er(T −s)


(q.e.d.).
r

(v) This result immediately follows from the definition of the value of the forward con-
tract. We have

Ẽ e−r(T −t) (S(T ) − ForS (t, T )) F(t)

 
⇔ ForS (t, T ) = Ẽ [S(T )| F(t)] (q.e.d.).

Here, we used that ForS (t, T ) is F(t)-measurable to take it outside the conditional
expectation.

(vi) Following the hint, we first compute the differential of the discounted portfolio value

d(D(t)X(t)) = D(t)dX(t) + X(t)dD(t) + dD(t)dX(t)

= ∆(t)D(t)dS(t) − a∆(t)D(t)dt + rD(t)(X(t) − ∆(t)S(t))dt − rD(t)X(t)dt

= ∆(t)D(t)(dS(t) − rS(t)dt) − a∆(t)D(t)dt

= ∆(t)σd(D(t)S(t)) − a∆(t)D(t)dt.

Integrating and using that ∆(t) = 1 for all 0 ≤ t ≤ T yields

25
 Z T Z T
D(T )X(T ) = σ d(D(t)S(t)) − a D(t)dt
0 0
Z T
t=T
= σD(t)S(t)|t=0 − a e−rt dt
0
a −rT

= D(T )S(T ) − S(0) + e −1 .
r

Solving or X(T ) yields

rT
a −rT

X(T ) = S(T ) − e S(0) − e −1
r
 a 
1 − erT


= S(T ) − S(0) −
r
= S(T ) − ForS (0, T ) (q.e.d.).

26