Stochastic Calculus for Finance II

46-945
Spring 2024

Solutions to Final Exam

10:00am – 1:00pm, Thursday, February 29

1. Stochastic differential equation:

dX(u) = β u, X(u) du + γ u, X(u) dW
f (u).

Kolmogorov Backward Equation:

1
pt (t, x; T, y) + β(t, x)px (t, x; T, y) + γ 2 (t, x)pxx (t, x; T, y) = 0.
2
Kolmogorov Forward Equation:

∂
1 ∂2 2

pT (t, x; T, y) + β(T, y)p(t, x; T, y) − γ (T, y)p(t, x; T, y) = 0.
∂y 2 ∂y 2

2. Dupire’s formula:


dS(t) = rS(t)dt + σ t, S(t) S(t)dW
f (t),
 −rT
+ 
c 0, S(0); T, K) = E
e e S(T ) − K ,
v 
u


u 2 cT 0, S(0); T, K + rKcK 0, S(0); T, K
σ(T, K) = .
t
K 2 cKK 0, S(0); T, K)

3. HJM no-arbitrage condition:

df (t, T ) = α(t, T )dt + σ(t, T )dW (t),

Z T
∗
σ (t, T ) = σ(t, u)du.
t

α(t, T ) − σ ∗ (t, T )σ(t, T )

θ(t) =
σ(t, T )
is independent of T . In terms of the risk-neutral Brownian motion,

df (t, T ) = σ(t, T )σ ∗ (t, T )dt + σ(t, T )dW

f (t).

4. SOFR set at time Tj+1 for the accrual period [Tj , Tj+1 ] is
"Z # !
Tj+1
1 D(Tj ) − D(Tj+1 )
S(Tj+1 ; Tj , Tj+1 ) = exp R(u)du − 1 = ,
τ Tj τ D(Tj+1 )

1
 where τ = Tj+1 − Tj . Forward SOFR is

B(t, Tj ) 1
− , 0 ≤ t ≤ Tj ,


τ B(t, Tj+1 ) τ

ForS (t; Tj , Tj+1 ) =
D(Tj ) 1
− , Tj ≤ t ≤ Tj+1 .


τ D(t)B(t, Tj+1 ) τ


5. Black’s formula with nonrandom, time-dependent volatility. Let S be an asset that

does not pay a dividend and has zero cost of carry. Let C(0) be the price at time
zero of a European call on S with strike price K expiring at time T . Assume that the
forward price For(t, T ) of S for delivery at time T has nonrandom, time-dependent
volatility σ(u), 0 ≤ u ≤ T . We do not require σ(u) to be positive; we require only that
s
1 T 2
Z
σ= σ (u)du
T 0

is not zero and hence positive so that we can divide by it in the formula for d± below.
Then


C(0) = B(0, T ) For(0, T )N (d+ ) − KN (d− ) = S(0)N (d+ ) − KB(0, T )N (d− ),

where  
1 For(0, T ) 1 2
d± = √ log ± σ T .
σ T K 2

You may not communicate with anyone during the exam, and you may not access the
internet.
In order to receive credit, you must show full derivations of your solutions.

2
Problem 1 (20 points). In the Hull-White interest rate model, the differential of the
interest rate R(u) (also called the spot rate or the short rate) is


dR(u) = κ θ(u) − R(u) du + σdW f (u), u ≥ 0.

Here θ is a nonrandom function of time, κ is a positive constant, σ is a positive constant,

and W
f is a Brownian motion under a risk-neutral measure P.e
The Hull-White model is Markov, which implies that for each maturity T there exists
of function g(t, r; T ) of t and r such that the price of a default-free zero-coupon bond paying
1 at maturity T can be written as
h R i
B(t, T ) = Ee e− tT R(u)du) F(t) = g t, R(t); T ).

(i) (10 pts.) Find the partial differential equation satisfied by g(t, r; T ). Justify all the
steps you used to find this equation. Remember that there are no random variables in
a partial differential equation.
(ii) (10 pts.) Assume g(t, r; T ) has the form
 
g(t, r; T ) = exp − C(t, T )r − A(t, T ) ,
where for each T , C(t, T ) and A(t, T ) are nonrandom functions of t. Find ordinary
differential equations for C(t, T ) and A(t, T ). You do not need to solve these equations.

Solution. Comment: The Hull-White model belongs to the class of affine yield models, and
for these models there is a standard method of deriving the ordinary differential equations for
the functions C(t, T ) and A(t, T ). We used this method in class. Lecture 7, pages 1 and 2 for
the Ho-Lee model. Jonghwa applied the method in Recitation 3 for the Cox-Ingersoll-Ross
model. You used the method in Exercise 2 of Homework 3 for the Hull-White model. Here
you are asked to repeat Exercise 2, parts (ii) and (iii), of Homework 3.

(i) Define Rt
D(t) = e− 0 R(u)du

so that h R i
e e− 0T R(u)du F(t) .
D(t)B(t, T ) = E
Let 0 ≤ s ≤ t ≤ T be given. From the tower property (iterated conditioning), we have
  h   i h i
e D(t)B(t, T ) F(s) = E
E e Ee D(T ) F(t) F(s) = E e D(T ) F(s) = D(s)B(s, T ).

This is the martingale property for the discounted bond price.

Because D(t)g(t, R(t); T ) = D(t)B(t, T ), D(t)g(t, R(t); T ) is a martingale. We com-
pute the differential of D(t)g(t, R(t); T ). To take the differential
R t of this product, we
−y
first note that D(t) = f (Y (t)), where f (y) = e and Y (t) = 0 R(u)du. We have
f 0 (y) = −f (y), f 00 (y) = f (y), dY (t) = R(t)dt, dY (t)dY (t) = 0.

3
 Therefore,
1
dD(t) = f 0 Y (t) dY (t) + f 00 Y (t) dY (t)dY (t) = −f Y (t) R(t)dt = −D(t)R(t)dt.




2

Itô’s product rule implies



d D(t)g t, R(t); T
= g dD + D dg + dD dg
 
1
= −DRgdt + D gt dt + gr dR + grr dRdR + 0
2
 
1 2
= D −Rgdt + gt dt + gr κ(θ − R)dt + gr σdW + grr σ dt
f
2






= D(t) −R(t)g t, R(t); T + gt t, R(t); T + κ θ(t) − R(t) gr t, R(t); T

1 2


+ σ grr t, R(t); T dt + σD(t)gr t, R(t); T dWf (t).
2

Because D(t)g(t, R(t); T ) is a martingale, the dt term in this last expression must
vanish along every path of the short rate process R. But R(t) can take any real value,
which implies that
1
−rg(t, r; T )+gt (t, r; T )+κ θ(t)−r gr (t, r; T )+ σ 2 grr (t, r; T ) = 0,


0 ≤ t ≤ T, r ∈ R.
2

(ii) We assume  
g(t, r; T ) = exp − C(t, T )r − A(t, T )
and compute

gt (t, r; T ) = − C 0 (t, T )r + A0 (t, T ) g(t, r; T ),

gr (t, r; T ) = −C(t, T )g(t, r; T ),

grr (t, r; T ) = C 2 (t, T )g(t, r; T ),

where C 0 (t, T ) = ∂t
∂
C(t, T ) and A0 (t, T ) = ∂t
∂
A(t, T ). Substitution into (??) results in
 
0 0

1 2 2
g(t, r; T ) −r − C (t, T )r − A (t, T ) − κ θ(t) − r C(t, T ) + σ C (t, T ) = 0.
2

But g(t, r; T ) > 0 so the term in square brackets must be zero. We write this as
 
 0
 0 1 2 2
− 1 − C (t, T ) + κC(t, T ) r + −A (t, T ) − κθ(t)C(t, T ) + σ C (t, T ) = 0. (7.1)
2

4
 This equation must hold for all r ∈ R. Differentiating with respect to r, we obtain
−1 − C 0 (t, T ) + κC(t, T ) = 0, 0 ≤ t ≤ T.
Substituting this back into (7.1), we obtain
1
−A0 (t, T ) − κθ(t)C(t, T ) + σ 2 C 2 (t, T ) = 0, 0 ≤ t ≤ T.
2
We write these equations in the more standard ordinary differential equation form
C 0 (t, T ) = κC(t, T ) − 1, 0 ≤ t ≤ T,
1
A0 (t, T ) = −κθ(t)C(t, T ) + σ 2 C 2 (t, T ), 0 ≤ t ≤ T.
2

Problem 2 (20 points). Consider a model in which for each T ∈ [0, T ] the price at time
t ∈ [0, T ] of a default-free zero-oupon bond paying 1 at maturity T is
 Z T 
1 2 2 2
B(t, T ) = exp − f (0, u)du − γ t(T − t ) − γ(T − t)W (t) .
t 2
Here f (0, u), 0 ≤ u ≤ T , is the time-zero forward rate curve, γ is a positive constant, and
W is a Brownian motion under the physical measure P. A trader can trade these bonds
and finance the trading using the money market account with random intrest rate R(t),
0 ≤ t ≤ T . The interest rate process R(t), which is not given explicitly in the problem
statement, can be determined from the bond prices that are given.

(i) (10 pts.) Does there exist an arbitrage in this model? If there is, specify a trading
strategy that produces arbitrage. In particular, specify which maturity bonds the
trading strategy holds at each time t and how many of these bonds it holds. If there
is no arbitrage, give a formula for the Brownian motion W f under the risk-neutral
measure.
(ii) (10 pts.) Provide a formula for the interest rate process in terms of the Brownian
motion W under the physical measure.

Solution.

(i) To determine if there is an arbitrage, we compute the differential of the forward interest
rates and check condition 3 on page 1 of the exam. We have
∂
f (t, T ) = − log B(t, T )
∂T
Z T 
∂ 1 2 2 2
= f (0, u)du + γ t(T − t ) + γ(T − t)W (t)
∂T t 2
2
= f (0, T ) + γ tT + γW (t).

5
 Therefore,
df (t, T ) = γ 2 T dt + γdW (t).
The general from for the differential of the forward rate, given by item 3 on page 1 of
the exam, is
df (t, T ) = α(t, T )dt + σ(t, T )dW (t).
We see that in this case

α(t, T ) = γ 2 T and σ(t, T ) = γ.

Therefore, Z T
∗
σ (t, T ) = σ(t, u)du = γ(T − t).
t
We have
α(t, T ) − σ ∗ (t, T )σ(t, T ) γ 2 T − γ 2 (T − t) γ 2t
= = = γt, (7.2)
σ(t, T ) γ γ
which is independent of T . This model is free of arbitrage.
The Brownian motion under the risk-neutral measure satisfies

dW
f (t) = dW (t) + θ(t)dt

for some process θ(t). You may remember that θ(t) is given by (7.2). If not, you can
proceed as follows. We have from the final formula in condition 3 on page 1 of the
exam and from the formulas just derived that

df (t, T ) = σ(t, T )σ ∗ (t, T )dt + σ(t, T )dW

f (t)
= σ(t, T )σ ∗ (t, T )dt + σ(t, T ) dW (t) + θ(t) dt

= γ 2 (T − t)dt + γdW (t) + γθ(t)dt

= γ 2 T dt + γdW (t),

from which we see that

θ(t) = γt.
Therefore,
Z t Z t
1
W
f (t) = W (t) + θ(u)du = W (t) + γudu = W (t) + γt2 .
0 0 2

(ii) We have
R(t) = f (t, t) = f (0, t) + γt2 + γW (t).
Instead of providing the formula for R(t), some students gave the formula for dR(t),
which we can compute from the formula above for R(t):
∂
dR(t) = f (0, T ) dt + 2γtdt + γdW (t),
∂T T =t

6
 ∂
where the notation ∂T f (0, T ) means to first take the derivative of f (0, T ) with
T =t
respect to its second argument T and then replace T by t.
Note that to compute dR(t), you must first compute R(t) = f (t, t) and then take the
differential with respect t. This way the differential with respect to t applies to both
the first and second arguments of f (t, t). If you instead replace T by t in the formula
for df (t, T ), you will get the wrong answer because the differential df (t, T ) is respect
to the first argument of f (t, T ) only.

Problem 3 (35 points). Consider a risky asset whose initial price S(0) is positive and
whose price differential is

dS(t) = α(t)S(t)dt + σS(t)dW (t),

where W is a Brownian motion under the physical measure P, α(t) is a possibly random
process adapted to this Brownian motion, and σ is a positive constant. In addition to this
asset, our model has a money market account with a constant interest rate r.
The asset in this case is something physical, like oil or gold. If an agent holds the asset,
she must pay a storage cost at rate c per unit asset per unit time, where c is a positive
constant. The storage cost paid depends only on the number of units of the asset held, not
on the price of the asset. For example, if the agent holds ∆(t) units of the asset at each time
t between 0 and T , then the total storage cost she pays is
Z T
c ∆(t)dt.
0

Now consider a derivative security that pays S(T ) − K at time T . (Note: The final
payment is S(T ) − K, not (S(T ) − K)+ . This is not an option.) At each time t, the price
of this derivative security is g(t, S(t)), where g(t, x), 0 ≤ t ≤ T , x > 0 is a function to be
determined. Of course, g(T, x) = x − K for all x > 0.

(i) (10 pts.) Consider a portfolio that holds ∆(t) units of the asset S at each time t between
0 and T and finances this trading using the money market with constant interest rate
r. Write the formula for dX(t), the differential of the value of this portfolio.

(ii) (5 pts.) It is possible to choose X(0) and ∆(t), 0 ≤ t ≤ T , so that X(t) = g(t, S(t) for
all t ∈ [0, T ] provided that g(t, x) satisfies a certain partial differential equation. What
is this partial differential equation?

(iii) (5 pts.) Find a measure P e under which e−rt g(t, S(t)) is a martingale. In particular,
what is the Radon-Nikodym derivative process that changes from the physical measure
P to P
e and what is the Brownian motion Wf under P. e

(iv) (5 pts.) Use the measure P

e to compute g(t, x) explicitly.

7
(v) (5 pt.) The time-t forward price For(t, T ) of S(T ) delivered at time T is the value of
K that causes the derivative security paying S(T ) − K at time T to have price zero at
time t. What is the forward price of S(T ) at time t?

Solution. Comment: In this problem, the portfolio value consists of the shares of S held
and cash, and the cash position may be negative. This means that

X(t) = ∆(t)S(t) + Cash(t),

the cash position is X(t) − ∆(t)S(t). The cost of carry reduces the value of the portfolio at
rate c∆(t), but it is not a debt incurred by the agent holding the portfolio. In particular, the
cash position is not X(t) − ∆(t)S(t) − ∆(t)c. If you used this cash position and wrote


dX(t) = ∆(t)dS(t) − c∆(t)dt + r X(t) − ∆(t)S(t) − c∆(t) dt,

you lost full credit for part (i), but could still get credit for the remaining parts if you worked
correctly from your wrong answer to part (i). If you did this, everywhere that c appears in
the correct answer, you would have (r + 1)c.

(i)

dX(t) = ∆(t)dS(t) − c∆(t)dt + r X(t) − ∆(t)S(t) dt.

(ii) From (i) we have

d e−rt X(t) = e−rt − rX(t)dt + dX(t)

= e−rt ∆(t)dS(t) − c∆(t)dt − r∆(t)S(t)dt

 
= e−rt ∆(t)S(t) α(t) − r) − c∆(t) dt + e−rt ∆(t)σS(t)dW (t).

On the other hand,



−rt
d e g t, S(t)


= e−rt − rg t, S(t) dt + dg t, S(t)


 
−rt



1

=e −rg t, S(t) dt + gt t, S(t) dt + gx t, S(t) dS(t) + gxx t, S(t) dS(t)dS(t)
2
 
−rt



1 2 2

=e −rg t, S(t) + gt t, S(t) + α(t)gx t, S(t) + σ S (t)gxx t, S(t) dt
2
−rt


+ e gx t, S(t) σS(t)dW (t).

Equating the dW (t) terms in these two expressions, we see that we should take

∆(t) = gx t, S(t)), 0 ≤ t ≤ T.

8
 Equating the dt terms and substituting in this value for ∆(t), we obtain




α(t) − r S(t)gx t, S(t) − cgx t, S(t)

1
= −rg t, S(t) + gt t, S(t) + α(t)gx t, S(t) + σ 2 S 2 (t)gxx t, S(t) .




2
The terms involving α(t) cancel out of this equation. We replace the random variable
S(t) by the real variable x and rearrange terms in this equation to obtain the partial
differential equation
1
−rg(t, x) + gt (t, x) + rxgx (t, x) + cgx (t, x) + σ 2 x2 gxx (t, x) = 0, 0 ≤ t ≤ T, x > 0.
2

(iii) Let g(t, x) be a function satisfying the partial differential equation found in part (ii)
and the terminal condition g(t, x) = x − K. We need to find a measure P e under which
e−rt g(t, S(t)) is a martingale. If W
f is a Brownian motion under this measure and

dS(t) = rS(t)dt + cdt + σS(t)dW

f (t), (7.3)

then


−rt
d e g t, S(t)


= e−rt − rg t, S(t) dt + dg t, S(t)


 
−rt



1

=e −rg t, S(t) dt + gt t, S(t) dt + gx t, S(t) dS(t) + gxx t, S(t) dS(t)dS(t)
2

= e−rt −rg t, S(t) + gt t, S(t) + rS(t)gx t, S(t) + cgx t, S(t)


1 2 2


+ σ S (t)gxx t, S(t) dt + σS(t)gx t, S(t) dW f (t)
2


= σS(t)gx t, S(t) dWf (t)

because g(t, x) satisfies the partial differential equation found in part (ii). Thus, if dS(t)
is given by (7.3), then e−rt g(t, S(t)) is a martingale under P.e We derive a formula for
dWf (t) using (7.3) and the formula for dS(t) given in the problem statement:

dS(t) = α(t)S(t)dt + σS(t)dW (t)

   
α(t) − r c
= rS(t)dt + cdt + σS(t) dW (t) + − dt
σ σS(t)
= rS(t)dt + cdt + σS(t)dW
f (t)

when  
α(t) − r c
dW
f (t) = dW (t) + − dt.
σ σS(t)

9
 Equivalently, we have
Z t 
α(u) − r c
W
f (t) = W (t) + − du.
0 σ σS(u)

The Radon-Nikodym derivative process that changes from the physical measure P to
a measure P
e under which W
f is a Brownian motion is
" Z  2 #
t
1 t α(u) − r
 Z 
α(u) − r c c
exp − − dW (u) − − du .
0 σ σS(u) 2 0 σ σS(u)

(iv) Because e−rt g(t, S(t)) is a martingale under P,

e we have

e−rt g t, S(t) = E
e e−rT g T, S(T ) F(t) = E
e e−rT S(T ) − K F(t) ,


 

0 ≤ t ≤ T.

In other words,
e e−rT S(T ) − K F(t) = ert E
g t, S(t) = ert E e e−rT S(T ) F(t) − e−r(T −t) K.


  

e e−rT S(T ) F(t) . Equation (7.3) implies

 
It remains to compute E

d e−ru S(u) = e−ru − rS(u) + dS(u) = e−ru cdt + e−rt σS(u)dW

f (u).

Integrating from u = t to u = T , we obtain

Z T Z T
−rT −rt −ru
e S(T ) = e S(t) + c e du + σ e−ru S(u)dWf (u)
t t
Z T
−rt c −rt −rT
e−ru S(u)dW


= e S(t) + e − e +σ f (u).
r t

It follows that
e e−rT S(T ) F(t) = e−rt S(t) + c e−rt − e−rT .
 

E
r
Substituting this into the formula for g(t, S(t)) derived above we have
c
g t, S(t) = S(t) + 1 − e−r(T −t) − e−r(T −t) K.



r
Replacing the random variable S(t) by the real variable x, we conclude that
c
1 − e−r(T −t) − e−r(T −t) K,


g(t, x) = x + 0 ≤ t ≤ T, x > 0.
r

You were not asked to check that the function g(t, x) you found satisfies the terminal
condition g(T, x) = x − K and the partial differential equation derived in part (ii), but

10
 we do that here. It is easily verified that g(T, x) = x − K. For the partial differential
equation, we first note that
gt (t, x) = −ce−r(T −t) − re−r(T −t) K, gx (t, x) = 1, gxx (t, x) = 0.
We have
1
−rg(t, x) + gt (t, x) + rxgx (t, x) + cgx (t, x) + σ 2 x2 gxx (t, x)
2
−r(T −t) −r(T −t) −r(T −t)
− re−r(T −t) K + rx + c


= −rx − c 1 − e + re K − ce
= 0.

(v) We set g(t, S(t)) = 0 and solve for K to obtain

c r(T −t)
For(t, T ) = er(T −t) S(t) +


e −1 .
r
Comment: The formula for the forward price obtained in this problem can be justified
by a finance argument. For 0 ≤ t < T , an agent can begin with zero initial capital at time
t, borrow S(t), and buy one unit of S. Holding this until time T requires the agent to pay
the storage cost at rate c per unit time. At each time u ∈ [t, T ], the storage cost must be
borrowed, and the resulting loan grows at rate r between times u and T , so that the total debt
at time T from borrowing storage costs between times t and T is
Z T
c
er(T −u) cdu = er(T −t) − 1 .


t r
In addition, the debt incurred at time t to buy one unit of S grows to er(T −t) S(t) at time t.
Thus, beginning with zero capital at time t, one can create a portfolio whose value at time T
is  c

S(T ) − er(T −t) S(t) + er(T −t) − 1 .
r
By buying a forward contract at time t, an agent can begin with zero initial capital at time t
and create a portfolio whose value at time T is
S(T ) − For(t, T ).
In order to avoid arbitrage, we must have
c r(T −t)
For(t, T ) = er(T −t) S(t) +


e −1 .
r

Problem 4 (20 points). Consider two geometric motion risky asset price processes S1 and
S2 with differentials
dS1 (t) = rS1 (t)dt + σ1 S1 (t)dW
f1 (t),
p

f1 (t) + 1 − ρ2 dW
dS2 (t) = rS2 (t)dt + σ2 S2 (t) ρdW f2 (t),

11
where the interest rate r is constant, the volatilties σ1 and σ2 are positive constants, the
correlation ρ is a constant between −1 and 1, and W f1 and W
f2 are independent Brownian
motions under a risk-neutral measure P. e Compute the spread option price
h i
e e−r(T −t) S2 (T ) − S1 (T ) + F(t) , 0 ≤ t ≤ T.


E

Your answer should be in the form

Something × N d+ (t, Something) − Something × N d− (t, Something) .

You need to determine what should replace Something in four places and what are the
formulas for d+ (t, x) and d− (t, x).

Solution. Comment: The price of a spread option was computed in Exercise 4(iii) of
Homework 6. The change of measure needed here appeared in Homework 1, Exercise 4 and
in the second example of Recitation 3.
Many students attempted to use Black’s formula to solve this problem. We use Black’s
formula when the interest rate is random. In this problem, r is constant, so B(t, T ) = e−r(T −t)
and the Radon-Nikodym derivative that changes from P e to the forward measure PT is

D(t)B(t, T ) e−rt e−r(T −t)

= = 1.
B(0, T ) e−rT

Therefore, PT = P. e There is no reason to use a forward measure or a forward price in the

solution of this problem.
We use Bayes’ Rule with the P-martingale
e e−rt S1 (t)/S1 (0) as the Radon-Nikodym
derivative, which changes from the measure P e to a measure we call PS1 , in the following
computation:
" + #
−rT

h
−r(T −t)

+ i S 1 (0) e S1 (T ) S2 (T )
e e
E S2 (T ) − S1 (T ) F(t) = S1 (t) · −rt E
e −1 F(t)
e S1 (t) S1 (0) S1 (T )
" + #
S1 S2 (T )
= S1 (t)E −1 F(t) .
S1 (T )

Under PS1 , the quotient of P-martingales

e

e−rt S2 (t) S2 (t)

=
e−rt S1 (t) S1 (t)
p
is a martingale with volatility vector (σ2 ρ − σ1 , 1 − ρ2 σ2 ) and volatility scalar
q p q
σ = (σ2 ρ − σ1 )2 + ( 1 − ρ2 σ2 )2 = σ12 − 2ρσ1 σ2 + σ22 .

12
In particular, S2 (t)/S1 (t) is a geometric Brownian motion with zero mean rate of growth
and volatility σ. It follows from the Black-Scholes formula (with zero mean rate of return
and zero discounting and with strike price K = 1) that
" + #      
S1 S2 (T ) S2 (t) S2 (t) S2 (t)
E −1 F(t) = N d+ t, − N d− t, ,
S1 (T ) S1 (t) S1 (t) S1 (t)

where  
1 S2 (t) 1 2
d± (t, x) = √ log ± σ (T − t) .
σ T −t S1 (t) 2
Multiplying by S1 (t), we finally obtain
h i
e e−r(T −t) S2 (T ) − S1 (T ) + F(t)


E
     
S2 (t) S2 (t)
= S2 (t)N d+ t, − S1 (t)N d− t, , 0 ≤ t ≤ T.
S1 (t) S1 (t)

Problem 5 (10 points). Consider a model with two currencies in two countries, A and B.
Let X B|A (t) be the price in currency A of a unit of currency B at time t and let X A|B (t) be
the price in currency B of a unit of currency A at time t. Assume

dX B|A (t) = RA (t) − RB (t) X B|A (t)dt + σX B|A (t)dW A (t),

where RA (t) is a random interest rate in country A, RB (t) is a random interest rate in
country B, σ is a positive constant, and W A (t) is a Brownian motion under the risk-neutral
measure PA in country A. We define the time-t futures price denominated in currency A for
time-T delivery of a unit of currency B by

FutB|A (t, T ) = EA X B|A (T ) F(t) , 0 ≤ t ≤ T.

 

Let PB denote the risk-neutral measure in country B and define the time-t futures price
denominated in currency B for time-T delivery of a unit of currency A by

FutA|B (t, T ) = EB X A|B (T ) F(t) , 0 ≤ t ≤ T.

 

Find a condition on the model that so that

1
FutB|A (0, T ) = A|B
Fut (0, T )

if and only the condition holds. You may find it helpful to use the notation
 Z t   Z T 
A A B B
D (t) = exp − R (u)du , D (t) = exp − R (u)du .
0 0

13
Solution. Comment: There are many conditions that guarantee that
1
FutB|A (0, T ) = A|B
.
Fut (0, T )
One is that the interest rates are not random, but the problem statement emphasized that
they are random. One can obtain another condition by replacing the futures prices by their
definitions and writing
1
EA X B|A (T ) = B  A|B
 
.
E X (T )
There are other conditions one can write that involve expectations under one or both mea-
sures that are not much more insightful than that. A condition that involves both measures
is impossible to understand intuitively. Here we want a condition that has some intuitive
meaning.
There are two ideas in the solution. The first idea in the solution is to write both futures
prices using the same measure. Let us choose PA and rewrite FutA|B (0, T ) as
FutA|B (0, T ) = EB X A|B (T )
 
 B A A|B

A|B B D (T )M (T )X (T ) B A
= X (0)E M (T )D (T )
X A|B (0)
1
= B|A EA M B (T )DA (T ) .
 
X (0)
The second idea is to use the fact that M B DA X B|A is a PA martingale to write
X B|A (0) = EA M B (T )DA (T )X B|A (T ) .
 

Substituting this into the equation above, we have

A
 B A

E M (T )D (T )
FutA|B (0, T ) = A  B ,
E M (T )DA (T )X B|A (T )
and we seek a condition under which this is
1 1
B|A
= A  B|A .
Fut (0, T ) E X (T )
The condition is obviously that the expectation in the denominator factors:
EA M B (T )DA (T )X B|A (T ) = EA M B (T )DA (T ) · EA X B|A (T ) .
     

This is the case if and only if M B (T )DA (T ) and X B|A (T ) are uncorrelated under PA .
Comment: In FX models with random interest rates, this condition rarely holds. In such
models, it is nearly always the case that FutB|A (0, T ) is not the reciprocal of FutA|B (0, T ),
and more generally,
1
FutB|A (t, T ) 6= A|B
, 0 ≤ t ≤ T.
Fut (t, T )

14
 Comment: If one changes to the PB measure rather than the PA measure, the condition
obtained would be that M A (T )DB (T ) and X A|B (T ) are uncorrelated under PB .
Comment: Many students set out to find a condition under which
1
FutB|A (t, T ) = A|B
.
Fut (t, T )

That requires the use of Bayes’ Rule to change the measure. The resulting condition is that
M B (T )DA (T ) and X B|A (T ) are uncorrelated under PA conditioned on F(t), i.e.,

EA M B (T )DA (T )X B|A (T ) F(t) = EA M B (T )DA (T ) F(t) · EA X B|A (T ) F(t) .

     

15