You are on page 1of 84

Stochastic Calculus for Finance, Volume I and II

by Yan Zeng
Last updated: August 20, 2007
This is a solution manual for the two-volume textbook Stochastic calculus for nance, by Steven Shreve.
If you have any comments or nd any typos/errors, please email me at yz44@cornell.edu.
The current version omits the following problems. Volume I: 1.5, 3.3, 3.4, 5.7; Volume II: 3.9, 7.1, 7.2,
7.57.9, 10.8, 10.9, 10.10.
Acknowledgment I thank Hua Li (a graduate student at Brown University) for reading through this
solution manual and communicating to me several mistakes/typos.
1 Stochastic Calculus for Finance I: The Binomial Asset Pricing Model
1. The Binomial No-Arbitrage Pricing Model
1.1.
Proof. If we get the up sate, then X
1
= X
1
(H) =
0
uS
0
+ (1 + r)(X
0

0
S
0
); if we get the down state,
then X
1
= X
1
(T) =
0
dS
0
+(1 +r)(X
0

0
S
0
). If X
1
has a positive probability of being strictly positive,
then we must either have X
1
(H) > 0 or X
1
(T) > 0.
(i) If X
1
(H) > 0, then
0
uS
0
+ (1 + r)(X
0

0
S
0
) > 0. Plug in X
0
= 0, we get u
0
> (1 + r)
0
.
By condition d < 1 + r < u, we conclude
0
> 0. In this case, X
1
(T) =
0
dS
0
+ (1 + r)(X
0

0
S
0
) =

0
S
0
[d (1 +r)] < 0.
(ii) If X
1
(T) > 0, then we can similarly deduce
0
< 0 and hence X
1
(H) < 0.
So we cannot have X
1
strictly positive with positive probability unless X
1
is strictly negative with positive
probability as well, regardless the choice of the number
0
.
Remark: Here the condition X
0
= 0 is not essential, as far as a property denition of arbitrage for
arbitrary X
0
can be given. Indeed, for the one-period binomial model, we can dene arbitrage as a trading
strategy such that P(X
1
X
0
(1 +r)) = 1 and P(X
1
> X
0
(1 +r)) > 0. First, this is a generalization of the
case X
0
= 0; second, it is proper because it is comparing the result of an arbitrary investment involving
money and stock markets with that of a safe investment involving only money market. This can also be seen
by regarding X
0
as borrowed from money market account. Then at time 1, we have to pay back X
0
(1 + r)
to the money market account. In summary, arbitrage is a trading strategy that beats safe investment.
Accordingly, we revise the proof of Exercise 1.1. as follows. If X
1
has a positive probability of being
strictly larger than X
0
(1 + r), the either X
1
(H) > X
0
(1 + r) or X
1
(T) > X
0
(1 + r). The rst case yields

0
S
0
(u1r) > 0, i.e.
0
> 0. So X
1
(T) = (1+r)X
0
+
0
S
0
(d1r) < (1+r)X
0
. The second case can
be similarly analyzed. Hence we cannot have X
1
strictly greater than X
0
(1 + r) with positive probability
unless X
1
is strictly smaller than X
0
(1 +r) with positive probability as well.
Finally, we comment that the above formulation of arbitrage is equivalent to the one in the textbook.
For details, see Shreve [7], Exercise 5.7.
1.2.
1
Proof. X
1
(u) =
0
8 +
0
3
5
4
(4
0
+1.20
0
) = 3
0
+1.5
0
, and X
1
(d) =
0
2
5
4
(4
0
+1.20
0
) =
3
0
1.5
0
. That is, X
1
(u) = X
1
(d). So if there is a positive probability that X
1
is positive, then there
is a positive probability that X
1
is negative.
Remark: Note the above relation X
1
(u) = X
1
(d) is not a coincidence. In general, let V
1
denote the
payo of the derivative security at time 1. Suppose

X
0
and

0
are chosen in such a way that V
1
can be
replicated: (1 + r)(

X
0

0
S
0
) +

0
S
1
= V
1
. Using the notation of the problem, suppose an agent begins
with 0 wealth and at time zero buys
0
shares of stock and
0
options. He then puts his cash position

0
S
0

0

X
0
in a money market account. At time one, the value of the agents portfolio of stock, option
and money market assets is
X
1
=
0
S
1
+
0
V
1
(1 +r)(
0
S
0
+
0

X
0
).
Plug in the expression of V
1
and sort out terms, we have
X
1
= S
0
(
0
+

0
)(
S
1
S
0
(1 +r)).
Since d < (1 +r) < u, X
1
(u) and X
1
(d) have opposite signs. So if the price of the option at time zero is

X
0
,
then there will no arbitrage.
1.3.
Proof. V
0
=
1
1+r

1+rd
ud
S
1
(H) +
u1r
ud
S
1
(T)

=
S0
1+r

1+rd
ud
u +
u1r
ud
d

= S
0
. This is not surprising, since
this is exactly the cost of replicating S
1
.
Remark: This illustrates an important point. The fair price of a stock cannot be determined by the
risk-neutral pricing, as seen below. Suppose S
1
(H) and S
1
(T) are given, we could have two current prices, S
0
and S

0
. Correspondingly, we can get u, d and u

, d

. Because they are determined by S


0
and S

0
, respectively,
its not surprising that risk-neutral pricing formula always holds, in both cases. That is,
S
0
=
1+rd
ud
S
1
(H) +
u1r
ud
S
1
(T)
1 +r
, S

0
=
1+rd

S
1
(H) +
u

1r
u

S
1
(T)
1 +r
.
Essentially, this is because risk-neutral pricing relies on fair price=replication cost. Stock as a replicating
component cannot determine its own fair price via the risk-neutral pricing formula.
1.4.
Proof.
X
n+1
(T) =
n
dS
n
+ (1 +r)(X
n

n
S
n
)
=
n
S
n
(d 1 r) + (1 +r)V
n
=
V
n+1
(H) V
n+1
(T)
u d
(d 1 r) + (1 +r)
pV
n+1
(H) + qV
n+1
(T)
1 +r
= p(V
n+1
(T) V
n+1
(H)) + pV
n+1
(H) + qV
n+1
(T)
= pV
n+1
(T) + qV
n+1
(T)
= V
n+1
(T).
1.6.
2
Proof. The banks trader should set up a replicating portfolio whose payo is the opposite of the options
payo. More precisely, we solve the equation
(1 +r)(X
0

0
S
0
) +
0
S
1
= (S
1
K)
+
.
Then X
0
= 1.20 and
0
=
1
2
. This means the trader should sell short 0.5 share of stock, put the income
2 into a money market account, and then transfer 1.20 into a separate money market account. At time one,
the portfolio consisting of a short position in stock and 0.8(1 +r) in money market account will cancel out
with the options payo. Therefore we end up with 1.20(1 +r) in the separate money market account.
Remark: This problem illustrates why we are interested in hedging a long position. In case the stock
price goes down at time one, the option will expire without any payo. The initial money 1.20 we paid at
time zero will be wasted. By hedging, we convert the option back into liquid assets (cash and stock) which
guarantees a sure payo at time one. Also, cf. page 7, paragraph 2. As to why we hedge a short position
(as a writer), see Wilmott [8], page 11-13.
1.7.
Proof. The idea is the same as Problem 1.6. The banks trader only needs to set up the reverse of the
replicating trading strategy described in Example 1.2.4. More precisely, he should short sell 0.1733 share of
stock, invest the income 0.6933 into money market account, and transfer 1.376 into a separate money market
account. The portfolio consisting a short position in stock and 0.6933-1.376 in money market account will
replicate the opposite of the options payo. After they cancel out, we end up with 1.376(1 + r)
3
in the
separate money market account.
1.8. (i)
Proof. v
n
(s, y) =
2
5
(v
n+1
(2s, y + 2s) +v
n+1
(
s
2
, y +
s
2
)).
(ii)
Proof. 1.696.
(iii)
Proof.

n
(s, y) =
v
n+1
(us, y +us) v
n+1
(ds, y +ds)
(u d)s
.
1.9. (i)
Proof. Similar to Theorem 1.2.2, but replace r, u and d everywhere with r
n
, u
n
and d
n
. More precisely, set
p
n
=
1+rndn
undn
and q
n
= 1 p
n
. Then
V
n
=
p
n
V
n+1
(H) + q
n
V
n+1
(T)
1 +r
n
.
(ii)
Proof.
n
=
Vn+1(H)Vn+1(T)
Sn+1(H)Sn+1(T)
=
Vn+1(H)Vn+1(T)
(undn)Sn
.
(iii)
3
Proof. u
n
=
Sn+1(H)
Sn
=
Sn+10
Sn
= 1+
10
Sn
and d
n
=
Sn+1(T)
Sn
=
Sn10
Sn
= 1
10
Sn
. So the risk-neutral probabilities
at time n are p
n
=
1dn
undn
=
1
2
and q
n
=
1
2
. Risk-neutral pricing implies the price of this call at time zero is
9.375.
2. Probability Theory on Coin Toss Space
2.1. (i)
Proof. P(A
c
) +P(A) =

A
c P() +

A
P() =

P() = 1.
(ii)
Proof. By induction, it suces to work on the case N = 2. When A
1
and A
2
are disjoint, P(A
1
A
2
) =

A1A2
P() =

A1
P() +

A2
P() = P(A
1
) + P(A
2
). When A
1
and A
2
are arbitrary, using
the result when they are disjoint, we have P(A
1
A
2
) = P((A
1
A
2
) A
2
) = P(A
1
A
2
) + P(A
2
)
P(A
1
) +P(A
2
).
2.2. (i)
Proof.

P(S
3
= 32) = p
3
=
1
8
,

P(S
3
= 8) = 3 p
2
q =
3
8
,

P(S
3
= 2) = 3 p q
2
=
3
8
, and

P(S
3
= 0.5) = q
3
=
1
8
.
(ii)
Proof.

E[S
1
] = 8

P(S
1
= 8) + 2

P(S
1
= 2) = 8 p + 2 q = 5,

E[S
2
] = 16 p
2
+ 4 2 p q + 1 q
2
= 6.25, and

E[S
3
] = 32
1
8
+ 8
3
8
+ 2
3
8
+ 0.5
1
8
= 7.8125. So the average rates of growth of the stock price under

P
are, respectively: r
0
=
5
4
1 = 0.25, r
1
=
6.25
5
1 = 0.25 and r
2
=
7.8125
6.25
1 = 0.25.
(iii)
Proof. P(S
3
= 32) = (
2
3
)
3
=
8
27
, P(S
3
= 8) = 3 (
2
3
)
2

1
3
=
4
9
, P(S
3
= 2) = 2
1
9
=
2
9
, and P(S
3
= 0.5) =
1
27
.
Accordingly, E[S
1
] = 6, E[S
2
] = 9 and E[S
3
] = 13.5. So the average rates of growth of the stock price
under P are, respectively: r
0
=
6
4
1 = 0.5, r
1
=
9
6
1 = 0.5, and r
2
=
13.5
9
1 = 0.5.
2.3.
Proof. Apply conditional Jensens inequality.
2.4. (i)
Proof. E
n
[M
n+1
] = M
n
+E
n
[X
n+1
] = M
n
+E[X
n+1
] = M
n
.
(ii)
Proof. E
n
[
Sn+1
Sn
] = E
n
[e
Xn+1
2
e

+e

] =
2
e

+e

E[e
Xn+1
] = 1.
2.5. (i)
Proof. 2I
n
= 2

n1
j=0
M
j
(M
j+1
M
j
) = 2

n1
j=0
M
j
M
j+1

n1
j=1
M
2
j

n1
j=1
M
2
j
= 2

n1
j=0
M
j
M
j+1
+
M
2
n

n1
j=0
M
2
j+1

n1
j=0
M
2
j
= M
2
n

n1
j=0
(M
j+1
M
j
)
2
= M
2
n

n1
j=0
X
2
j+1
= M
2
n
n.
(ii)
Proof. E
n
[f(I
n+1
)] = E
n
[f(I
n
+M
n
(M
n+1
M
n
))] = E
n
[f(I
n
+M
n
X
n+1
)] =
1
2
[f(I
n
+M
n
)+f(I
n
M
n
)] =
g(I
n
), where g(x) =
1
2
[f(x +

2x +n) +f(x

2x +n)], since

2I
n
+n = [M
n
[.
2.6.
4
Proof. E
n
[I
n+1
I
n
] = E
n
[
n
(M
n+1
M
n
)] =
n
E
n
[M
n+1
M
n
] = 0.
2.7.
Proof. We denote by X
n
the result of n-th coin toss, where Head is represented by X = 1 and Tail is
represented by X = 1. We also suppose P(X = 1) = P(X = 1) =
1
2
. Dene S
1
= X
1
and S
n+1
=
S
n
+b
n
(X
1
, , X
n
)X
n+1
, where b
n
() is a bounded function on 1, 1
n
, to be determined later on. Clearly
(S
n
)
n1
is an adapted stochastic process, and we can show it is a martingale. Indeed, E
n
[S
n+1
S
n
] =
b
n
(X
1
, , X
n
)E
n
[X
n+1
] = 0.
For any arbitrary function f, E
n
[f(S
n+1
)] =
1
2
[f(S
n
+b
n
(X
1
, , X
n
)) +f(S
n
b
n
(X
1
, , X
n
))]. Then
intuitively, E
n
[f(S
n+1
] cannot be solely dependent upon S
n
when b
n
s are properly chosen. Therefore in
general, (S
n
)
n1
cannot be a Markov process.
Remark: If X
n
is regarded as the gain/loss of n-th bet in a gambling game, then S
n
would be the wealth
at time n. b
n
is therefore the wager for the (n+1)-th bet and is devised according to past gambling results.
2.8. (i)
Proof. Note M
n
= E
n
[M
N
] and M

n
= E
n
[M

N
].
(ii)
Proof. In the proof of Theorem 1.2.2, we proved by induction that X
n
= V
n
where X
n
is dened by (1.2.14)
of Chapter 1. In other words, the sequence (V
n
)
0nN
can be realized as the value process of a portfolio,
which consists of stock and money market accounts. Since (
Xn
(1+r)
n
)
0nN
is a martingale under

P (Theorem
2.4.5), (
Vn
(1+r)
n
)
0nN
is a martingale under

P.
(iii)
Proof.
V

n
(1+r)
n
= E
n

V
N
(1+r)
N

, so V

0
,
V

1
1+r
, ,
V

N1
(1+r)
N1
,
V
N
(1+r)
N
is a martingale under

P.
(iv)
Proof. Combine (ii) and (iii), then use (i).
2.9. (i)
Proof. u
0
=
S1(H)
S0
= 2, d
0
=
S1(H)
S0
=
1
2
, u
1
(H) =
S2(HH)
S1(H)
= 1.5, d
1
(H) =
S2(HT)
S1(H)
= 1, u
1
(T) =
S2(TH)
S1(T)
= 4
and d
1
(T) =
S2(TT)
S1(T)
= 1.
So p
0
=
1+r0d0
u0d0
=
1
2
, q
0
=
1
2
, p
1
(H) =
1+r1(H)d1(H)
u1(H)d1(H)
=
1
2
, q
1
(H) =
1
2
, p
1
(T) =
1+r1(T)d1(T)
u1(T)d1(T)
=
1
6
, and
q
1
(T) =
5
6
.
Therefore

P(HH) = p
0
p
1
(H) =
1
4
,

P(HT) = p
0
q
1
(H) =
1
4
,

P(TH) = q
0
p
1
(T) =
1
12
and

P(TT) =
q
0
q
1
(T) =
5
12
.
The proofs of Theorem 2.4.4, Theorem 2.4.5 and Theorem 2.4.7 still work for the random interest
rate model, with proper modications (i.e.

P would be constructed according to conditional probabili-
ties

P(
n+1
= H[
1
, ,
n
) := p
n
and

P(
n+1
= T[
1
, ,
n
) := q
n
. Cf. notes on page 39.). So
the time-zero value of an option that pays o V
2
at time two is given by the risk-neutral pricing formula
V
0
=

E

V2
(1+r0)(1+r1)

.
(ii)
Proof. V
2
(HH) = 5, V
2
(HT) = 1, V
2
(TH) = 1 and V
2
(TT) = 0. So V
1
(H) =
p1(H)V2(HH)+ q1(H)V2(HT)
1+r1(H)
=
2.4, V
1
(T) =
p1(T)V2(TH)+ q1(T)V2(TT)
1+r1(T)
=
1
9
, and V
0
=
p0V1(H)+ q0V1(T)
1+r0
1.
5
(iii)
Proof.
0
=
V1(H)V1(T)
S1(H)S1(T)
=
2.4
1
9
82
= 0.4
1
54
0.3815.
(iv)
Proof.
1
(H) =
V2(HH)V2(HT)
S2(HH)S2(HT)
=
51
128
= 1.
2.10. (i)
Proof.

E
n
[
Xn+1
(1+r)
n+1
] =

E
n
[
nYn+1Sn
(1+r)
n+1
+
(1+r)(XnnSn)
(1+r)
n+1
] =
nSn
(1+r)
n+1

E
n
[Y
n+1
] +
XnnSn
(1+r)
n
=
nSn
(1+r)
n+1
(u p +
d q) +
XnnSn
(1+r)
n
=
nSn+XnnSn
(1+r)
n
=
Xn
(1+r)
n
.
(ii)
Proof. From (2.8.2), we have

n
uS
n
+ (1 +r)(X
n

n
S
n
) = X
n+1
(H)

n
dS
n
+ (1 +r)(X
n

n
S
n
) = X
n+1
(T).
So
n
=
Xn+1(H)Xn+1(T)
uSndSn
and X
n
=

E
n
[
Xn+1
1+r
]. To make the portfolio replicate the payo at time N, we
must have X
N
= V
N
. So X
n
=

E
n
[
X
N
(1+r)
Nn
] =

E
n
[
V
N
(1+r)
Nn
]. Since (X
n
)
0nN
is the value process of the
unique replicating portfolio (uniqueness is guaranteed by the uniqueness of the solution to the above linear
equations), the no-arbitrage price of V
N
at time n is V
n
= X
n
=

E
n
[
V
N
(1+r)
Nn
].
(iii)
Proof.

E
n
[
S
n+1
(1 +r)
n+1
] =
1
(1 +r)
n+1

E
n
[(1 A
n+1
)Y
n+1
S
n
]
=
S
n
(1 +r)
n+1
[ p(1 A
n+1
(H))u + q(1 A
n+1
(T))d]
<
S
n
(1 +r)
n+1
[ pu + qd]
=
S
n
(1 +r)
n
.
If A
n+1
is a constant a, then

E
n
[
Sn+1
(1+r)
n+1
] =
Sn
(1+r)
n+1
(1a)( pu+ qd) =
Sn
(1+r)
n
(1a). So

E
n
[
Sn+1
(1+r)
n+1
(1a)
n+1
] =
Sn
(1+r)
n
(1a)
n
.
2.11. (i)
Proof. F
N
+P
N
= S
N
K + (K S
N
)
+
= (S
N
K)
+
= C
N
.
(ii)
Proof. C
n
=

E
n
[
C
N
(1+r)
Nn
] =

E
n
[
F
N
(1+r)
Nn
] +

E
n
[
P
N
(1+r)
Nn
] = F
n
+P
n
.
(iii)
Proof. F
0
=

E[
F
N
(1+r)
N
] =
1
(1+r)
N

E[S
N
K] = S
0

K
(1+r)
N
.
(iv)
6
Proof. At time zero, the trader has F
0
= S
0
in money market account and one share of stock. At time N,
the trader has a wealth of (F
0
S
0
)(1 +r)
N
+S
N
= K +S
N
= F
N
.
(v)
Proof. By (ii), C
0
= F
0
+P
0
. Since F
0
= S
0

(1+r)
N
S0
(1+r)
N
= 0, C
0
= P
0
.
(vi)
Proof. By (ii), C
n
= P
n
if and only if F
n
= 0. Note F
n
=

E
n
[
S
N
K
(1+r)
Nn
] = S
n

(1+r)
N
S0
(1+r)
Nn
= S
n
S
0
(1 +r)
n
.
So F
n
is not necessarily zero and C
n
= P
n
is not necessarily true for n 1.
2.12.
Proof. First, the no-arbitrage price of the chooser option at time m must be max(C, P), where
C =

E

(S
N
K)
+
(1 +r)
Nm

, and P =

E

(K S
N
)
+
(1 +r)
Nm

.
That is, C is the no-arbitrage price of a call option at time m and P is the no-arbitrage price of a put option
at time m. Both of them have maturity date N and strike price K. Suppose the market is liquid, then the
chooser option is equivalent to receiving a payo of max(C, P) at time m. Therefore, its current no-arbitrage
price should be

E[
max(C,P)
(1+r)
m
].
By the put-call parity, C = S
m

K
(1+r)
Nm
+P. So max(C, P) = P +(S
m

K
(1+r)
Nm
)
+
. Therefore, the
time-zero price of a chooser option is

P
(1 +r)
m

+

E

(S
m

K
(1+r)
Nm
)
+
(1 +r)
m

=

E

(K S
N
)
+
(1 +r)
N

+

E

(S
m

K
(1+r)
Nm
)
+
(1 +r)
m

.
The rst term stands for the time-zero price of a put, expiring at time N and having strike price K, and the
second term stands for the time-zero price of a call, expiring at time m and having strike price
K
(1+r)
Nm
.
If we feel unconvinced by the above argument that the chooser options no-arbitrage price is

E[
max(C,P)
(1+r)
m
],
due to the economical argument involved (like the chooser option is equivalent to receiving a payo of
max(C, P) at time m), then we have the following mathematically rigorous argument. First, we can
construct a portfolio
0
, ,
m1
, whose payo at time m is max(C, P). Fix , if C() > P(), we
can construct a portfolio

m
, ,

N1
whose payo at time N is (S
N
K)
+
; if C() < P(), we can
construct a portfolio

m
, ,

N1
whose payo at time N is (K S
N
)
+
. By dening (m k N 1)

k
() =

k
() if C() > P()

k
() if C() < P(),
we get a portfolio (
n
)
0nN1
whose payo is the same as that of the chooser option. So the no-arbitrage
price process of the chooser option must be equal to the value process of the replicating portfolio. In
particular, V
0
= X
0
=

E[
Xm
(1+r)
m
] =

E[
max(C,P)
(1+r)
m
].
2.13. (i)
Proof. Note under both actual probability P and risk-neutral probability

P, coin tosses
n
s are i.i.d.. So
without loss of generality, we work on P. For any function g, E
n
[g(S
n+1
, Y
n+1
)] = E
n
[g(
Sn+1
Sn
S
n
, Y
n
+
Sn+1
Sn
S
n
)] = pg(uS
n
, Y
n
+ uS
n
) + qg(dS
n
, Y
n
+ dS
n
), which is a function of (S
n
, Y
n
). So (S
n
, Y
n
)
0nN
is
Markov under P.
(ii)
7
Proof. Set v
N
(s, y) = f(
y
N+1
). Then v
N
(S
N
, Y
N
) = f(

N
n=0
Sn
N+1
) = V
N
. Suppose v
n+1
is given, then
V
n
=

E
n
[
Vn+1
1+r
] =

E
n
[
vn+1(Sn+1,Yn+1)
1+r
] =
1
1+r
[ pv
n+1
(uS
n
, Y
n
+ uS
n
) + qv
n+1
(dS
n
, Y
n
+ dS
n
)] = v
n
(S
n
, Y
n
),
where
v
n
(s, y) =
v
n+1
(us, y +us) + v
n+1
(ds, y +ds)
1 +r
.
2.14. (i)
Proof. For n M, (S
n
, Y
n
) = (S
n
, 0). Since coin tosses
n
s are i.i.d. under

P, (S
n
, Y
n
)
0nM
is Markov
under

P. More precisely, for any function h,

E
n
[h(S
n+1
)] = ph(uS
n
) +

h(dS
n
), for n = 0, 1, , M 1.
For any function g of two variables, we have

E
M
[g(S
M+1
, Y
M+1
)] =

E
M
[g(S
M+1
, S
M+1
)] = pg(uS
M
, uS
M
)+
qg(dS
M
, dS
M
). And for n M+1,

E
n
[g(S
n+1
, Y
n+1
)] =

E
n
[g(
Sn+1
Sn
S
n
, Y
n
+
Sn+1
Sn
S
n
)] = pg(uS
n
, Y
n
+uS
n
)+
qg(dS
n
, Y
n
+dS
n
), so (S
n
, Y
n
)
0nN
is Markov under

P.
(ii)
Proof. Set v
N
(s, y) = f(
y
NM
). Then v
N
(S
N
, Y
N
) = f(

N
K=M+1
S
k
NM
) = V
N
. Suppose v
n+1
is already given.
a) If n > M, then

E
n
[v
n+1
(S
n+1
, Y
n+1
)] = pv
n+1
(uS
n
, Y
n
+uS
n
) + qv
n+1
(dS
n
, Y
n
+dS
n
). So v
n
(s, y) =
pv
n+1
(us, y +us) + qv
n+1
(ds, y +ds).
b) If n = M, then

E
M
[v
M+1
(S
M+1
, Y
M+1
)] = pv
M+1
(uS
M
, uS
M
) + v
n+1
(dS
M
, dS
M
). So v
M
(s) =
pv
M+1
(us, us) + qv
M+1
(ds, ds).
c) If n < M, then

E
n
[v
n+1
(S
n+1
)] = pv
n+1
(uS
n
) + qv
n+1
(dS
n
). So v
n
(s) = pv
n+1
(us) + qv
n+1
(ds).
3. State Prices
3.1.
Proof. Note

Z() :=
P()

P()
=
1
Z()
. Apply Theorem 3.1.1 with P,

P, Z replaced by

P, P,

Z, we get the
analogous of properties (i)-(iii) of Theorem 3.1.1.
3.2. (i)
Proof.

P() =

P() =

Z()P() = E[Z] = 1.
(ii)
Proof.

E[Y ] =

Y ()

P() =

Y ()Z()P() = E[Y Z].


(iii)
Proof.

P(A) =

A
Z()P(). Since P(A) = 0, P() = 0 for any A. So

P(A) = 0.
(iv)
Proof. If

P(A) =

A
Z()P() = 0, by P(Z > 0) = 1, we conclude P() = 0 for any A. So
P(A) =

A
P() = 0.
(v)
Proof. P(A) = 1 P(A
c
) = 0

P(A
c
) = 0

P(A) = 1.
(vi)
8
Proof. Pick
0
such that P(
0
) > 0, dene Z() =

0, if =
0
1
P(0)
, if =
0
.
Then P(Z 0) = 1 and E[Z] =
1
P(0)
P(
0
) = 1.
Clearly

P( `
0
) = E[Z1
\{0}
] =

=0
Z()P() = 0. But P( `
0
) = 1 P(
0
) > 0 if
P(
0
) < 1. Hence in the case 0 < P(
0
) < 1, P and

P are not equivalent. If P(
0
) = 1, then E[Z] = 1 if
and only if Z(
0
) = 1. In this case

P(
0
) = Z(
0
)P(
0
) = 1. And

P and P have to be equivalent.
In summary, if we can nd
0
such that 0 < P(
0
) < 1, then Z as constructed above would induce a
probability

P that is not equivalent to P.
3.5. (i)
Proof. Z(HH) =
9
16
, Z(HT) =
9
8
, Z(TH) =
3
8
and Z(TT) =
15
4
.
(ii)
Proof. Z
1
(H) = E
1
[Z
2
](H) = Z
2
(HH)P(
2
= H[
1
= H) + Z
2
(HT)P(
2
= T[
1
= H) =
3
4
. Z
1
(T) =
E
1
[Z
2
](T) = Z
2
(TH)P(
2
= H[
1
= T) +Z
2
(TT)P(
2
= T[
1
= T) =
3
2
.
(iii)
Proof.
V
1
(H) =
[Z
2
(HH)V
2
(HH)P(
2
= H[
1
= H) +Z
2
(HT)V
2
(HT)P(
2
= T[
1
= T)]
Z
1
(H)(1 +r
1
(H))
= 2.4,
V
1
(T) =
[Z
2
(TH)V
2
(TH)P(
2
= H[
1
= T) +Z
2
(TT)V
2
(TT)P(
2
= T[
1
= T)]
Z
1
(T)(1 +r
1
(T))
=
1
9
,
and
V
0
=
Z
2
(HH)V
2
(HH)
(1 +
1
4
)(1 +
1
4
)
P(HH) +
Z
2
(HT)V
2
(HT)
(1 +
1
4
)(1 +
1
4
)
P(HT) +
Z
2
(TH)V
2
(TH)
(1 +
1
4
)(1 +
1
2
)
P(TH) + 0 1.
3.6.
Proof. U

(x) =
1
x
, so I(x) =
1
x
. (3.3.26) gives E[
Z
(1+r)
N
(1+r)
N
Z
] = X
0
. So =
1
X0
. By (3.3.25), we
have X
N
=
(1+r)
N
Z
=
X0
Z
(1 + r)
N
. Hence X
n
=

E
n
[
X
N
(1+r)
Nn
] =

E
n
[
X0(1+r)
n
Z
] = X
0
(1 + r)
n

E
n
[
1
Z
] =
X
0
(1 +r)
n 1
Zn
E
n
[Z
1
Z
] =
X0
n
, where the second to last = comes from Lemma 3.2.6.
3.7.
Proof. U

(x) = x
p1
and so I(x) = x
1
p1
. By (3.3.26), we have E[
Z
(1+r)
N
(
Z
(1+r)
N
)
1
p1
] = X
0
. Solve it for ,
we get
=

X
0
E

Z
p
p1
(1+r)
Np
p1

p1
=
X
p1
0
(1 +r)
Np
(E[Z
p
p1
])
p1
.
So by (3.3.25), X
N
= (
Z
(1+r)
N
)
1
p1
=

1
p1
Z
1
p1
(1+r)
N
p1
=
X0(1+r)
Np
p1
E[Z
p
p1
]
Z
1
p1
(1+r)
N
p1
=
(1+r)
N
X0Z
1
p1
E[Z
p
p1
]
.
3.8. (i)
9
Proof.
d
dx
(U(x) yx) = U

(x) y. So x = I(y) is an extreme point of U(x) yx. Because


d
2
dx
2
(U(x) yx) =
U

(x) 0 (U is concave), x = I(y) is a maximum point. Therefore U(x) y(x) U(I(y)) yI(y) for every
x.
(ii)
Proof. Following the hint of the problem, we have
E[U(X
N
)] E[X
N
Z
(1 +r)
N
] E[U(I(
Z
(1 +r)
N
))] E[
Z
(1 +r)
N
I(
Z
(1 +r)
N
)],
i.e. E[U(X
N
)] X
0
E[U(X

N
)]

E[

(1+r)
N
X

N
] = E[U(X

N
)] X
0
. So E[U(X
N
)] E[U(X

N
)].
3.9. (i)
Proof. X
n
=

E
n
[
X
N
(1+r)
Nn
]. So if X
N
0, then X
n
0 for all n.
(ii)
Proof. a) If 0 x < and 0 < y
1

, then U(x) yx = yx 0 and U(I(y)) yI(y) = U() y =


1 y 0. So U(x) yx U(I(y)) yI(y).
b) If 0 x < and y >
1

, then U(x) yx = yx 0 and U(I(y)) yI(y) = U(0) y 0 = 0. So


U(x) yx U(I(y)) yI(y).
c) If x and 0 < y
1

, then U(x) yx = 1 yx and U(I(y)) yI(y) = U() y = 1 y 1 yx.


So U(x) yx U(I(y)) yI(y).
d) If x and y >
1

, then U(x) yx = 1 yx < 0 and U(I(y)) yI(y) = U(0) y 0 = 0. So


U(x) yx U(I(y)) yI(y).
(iii)
Proof. Using (ii) and set x = X
N
, y =
Z
(1+r)
N
, where X
N
is a random variable satisfying

E[
X
N
(1+r)
N
] = X
0
,
we have
E[U(X
N
)] E[
Z
(1 +r)
N
X
N
] E[U(X

N
)] E[
Z
(1 +r)
N
X

N
].
That is, E[U(X
N
)] X
0
E[U(X

N
)] X
0
. So E[U(X
N
)] E[U(X

N
)].
(iv)
Proof. Plug p
m
and
m
into (3.6.4), we have
X
0
=
2
N

m=1
p
m

m
I(
m
) =
2
N

m=1
p
m

m
1
{m
1

}
.
So
X0

2
N
m=1
p
m

m
1
{m
1

}
. Suppose there is a solution to (3.6.4), note
X0

> 0, we then can conclude


m :
m

1

= . Let K = maxm :
m

1

, then
K

1

<
K+1
. So
K
<
K+1
and
X0

K
m=1
p
m

m
(Note, however, that K could be 2
N
. In this case,
K+1
is interpreted as . Also, note
we are looking for positive solution > 0). Conversely, suppose there exists some K so that
K
<
K+1
and

K
m=1

m
p
m
=
X0

. Then we can nd > 0, such that


K
<
1

<
K+1
. For such , we have
E[
Z
(1 +r)
N
I(
Z
(1 +r)
N
)] =
2
N

m=1
p
m

m
1
{m
1

}
=
K

m=1
p
m

m
= X
0
.
Hence (3.6.4) has a solution.
10
(v)
Proof. X

N
(
m
) = I(
m
) = 1
{m
1

}
=

, if m K
0, if m K + 1
.
4. American Derivative Securities
Before proceeding to the exercise problems, we rst give a brief summary of pricing American derivative
securities as presented in the textbook. We shall use the notation of the book.
From the buyers perspective: At time n, if the derivative security has not been exercised, then the buyer
can choose a policy with o
n
. The valuation formula for cash ow (Theorem 2.4.8) gives a fair price
for the derivative security exercised according to :
V
n
() =
N

k=n

E
n

1
{=k}
1
(1 +r)
kn
G
k

=

E
n

1
{N}
1
(1 +r)
n
G

.
The buyer wants to consider all the possible s, so that he can nd the least upper bound of security value,
which will be the maximum price of the derivative security acceptable to him. This is the price given by
Denition 4.4.1: V
n
= max
Sn

E
n
[1
{N}
1
(1+r)
n
G

].
From the sellers perspective: A price process (V
n
)
0nN
is acceptable to him if and only if at time n,
he can construct a portfolio at cost V
n
so that (i) V
n
G
n
and (ii) he needs no further investing into the
portfolio as time goes by. Formally, the seller can nd (
n
)
0nN
and (C
n
)
0nN
so that C
n
0 and
V
n+1
=
n
S
n+1
+ (1 + r)(V
n
C
n

n
S
n
). Since (
Sn
(1+r)
n
)
0nN
is a martingale under the risk-neutral
measure

P, we conclude

E
n

V
n+1
(1 +r)
n+1

V
n
(1 +r)
n
=
C
n
(1 +r)
n
0,
i.e. (
Vn
(1+r)
n
)
0nN
is a supermartingale. This inspired us to check if the converse is also true. This is exactly
the content of Theorem 4.4.4. So (V
n
)
0nN
is the value process of a portfolio that needs no further investing
if and only if

Vn
(1+r)
n

0nN
is a supermartingale under

P (note this is independent of the requirement
V
n
G
n
). In summary, a price process (V
n
)
0nN
is acceptable to the seller if and only if (i) V
n
G
n
; (ii)

Vn
(1+r)
n

0nN
is a supermartingale under

P.
Theorem 4.4.2 shows the buyers upper bound is the sellers lower bound. So it gives the price acceptable
to both. Theorem 4.4.3 gives a specic algorithm for calculating the price, Theorem 4.4.4 establishes the
one-to-one correspondence between super-replication and supermartingale property, and nally, Theorem
4.4.5 shows how to decide on the optimal exercise policy.
4.1. (i)
Proof. V
P
2
(HH) = 0, V
P
2
(HT) = V
P
2
(TH) = 0.8, V
P
2
(TT) = 3, V
P
1
(H) = 0.32, V
P
1
(T) = 2, V
P
0
= 9.28.
(ii)
Proof. V
C
0
= 5.
(iii)
Proof. g
S
(s) = [4 s[. We apply Theorem 4.4.3 and have V
S
2
(HH) = 12.8, V
S
2
(HT) = V
S
2
(TH) = 2.4,
V
S
2
(TT) = 3, V
S
1
(H) = 6.08, V
S
1
(T) = 2.16 and V
S
0
= 3.296.
(iv)
11
Proof. First, we note the simple inequality
max(a
1
, b
1
) + max(a
2
, b
2
) max(a
1
+a
2
, b
1
+b
2
).
> holds if and only if b
1
> a
1
, b
2
< a
2
or b
1
< a
1
, b
2
> a
2
. By induction, we can show
V
S
n
= max

g
S
(S
n
),
pV
S
n+1
+

V
S
n+1
1 +r

max

g
P
(S
n
) +g
C
(S
n
),
pV
P
n+1
+

V
P
n+1
1 +r
+
pV
C
n+1
+

V
C
n+1
1 +r

max

g
P
(S
n
),
pV
P
n+1
+

V
P
n+1
1 +r

+ max

g
C
(S
n
),
pV
C
n+1
+

V
C
n+1
1 +r

= V
P
n
+V
C
n
.
As to when < holds, suppose m = maxn : V
S
n
< V
P
n
+ V
C
n
. Then clearly m N 1 and it is possible
that n : V
S
n
< V
P
n
+ V
C
n
= . When this set is not empty, m is characterized as m = maxn : g
P
(S
n
) <
pV
P
n+1
+ qV
P
n+1
1+r
and g
C
(S
n
) >
pV
C
n+1
+ qV
C
n+1
1+r
or g
P
(S
n
) >
pV
P
n+1
+ qV
P
n+1
1+r
and g
C
(S
n
) <
pV
C
n+1
+ qV
C
n+1
1+r
.
4.2.
Proof. For this problem, we need Figure 4.2.1, Figure 4.4.1 and Figure 4.4.2. Then

1
(H) =
V
2
(HH) V
2
(HT)
S
2
(HH) S
2
(HT)
=
1
12
,
1
(T) =
V
2
(TH) V
2
(TT)
S
2
(TH) S
2
(TT)
= 1,
and

0
=
V
1
(H) V
1
(T)
S
1
(H) S
1
(T)
0.433.
The optimal exercise time is = infn : V
n
= G
n
. So
(HH) = , (HT) = 2, (TH) = (TT) = 1.
Therefore, the agent borrows 1.36 at time zero and buys the put. At the same time, to hedge the long
position, he needs to borrow again and buy 0.433 shares of stock at time zero.
At time one, if the result of coin toss is tail and the stock price goes down to 2, the value of the portfolio
is X
1
(T) = (1 +r)(1.36 0.433S
0
) +0.433S
1
(T) = (1 +
1
4
)(1.36 0.433 4) +0.433 2 = 3. The agent
should exercise the put at time one and get 3 to pay o his debt.
At time one, if the result of coin toss is head and the stock price goes up to 8, the value of the portfolio
is X
1
(H) = (1 + r)(1.36 0.433S
0
) + 0.433S
1
(H) = 0.4. The agent should borrow to buy
1
12
shares of
stock. At time two, if the result of coin toss is head and the stock price goes up to 16, the value of the
portfolio is X
2
(HH) = (1+r)(X
1
(H)
1
12
S
1
(H)) +
1
12
S
2
(HH) = 0, and the agent should let the put expire.
If at time two, the result of coin toss is tail and the stock price goes down to 4, the value of the portfolio is
X
2
(HT) = (1 +r)(X
1
(H)
1
12
S
1
(H)) +
1
12
S
2
(HT) = 1. The agent should exercise the put to get 1. This
will pay o his debt.
4.3.
Proof. We need Figure 1.2.2 for this problem, and calculate the intrinsic value process and price process of
the put as follows.
For the intrinsic value process, G
0
= 0, G
1
(T) = 1, G
2
(TH) =
2
3
, G
2
(TT) =
5
3
, G
3
(THT) = 1,
G
3
(TTH) = 1.75, G
3
(TTT) = 2.125. All the other outcomes of G is negative.
12
For the price process, V
0
= 0.4, V
1
(T) = 1, V
1
(TH) =
2
3
, V
1
(TT) =
5
3
, V
3
(THT) = 1, V
3
(TTH) = 1.75,
V
3
(TTT) = 2.125. All the other outcomes of V is zero.
Therefore the time-zero price of the derivative security is 0.4 and the optimal exercise time satises
() =

if
1
= H,
1 if
1
= T.
4.4.
Proof. 1.36 is the cost of super-replicating the American derivative security. It enables us to construct a
portfolio sucient to pay o the derivative security, no matter when the derivative security is exercised. So
to hedge our short position after selling the put, there is no need to charge the insider more than 1.36.
4.5.
Proof. The stopping times in o
0
are
(1) 0;
(2) 1;
(3) (HT) = (HH) = 1, (TH), (TT) 2, (4 dierent ones);
(4) (HT), (HH) 2, , (TH) = (TT) = 1 (4 dierent ones);
(5) (HT), (HH), (TH), (TT) 2, (16 dierent ones).
When the option is out of money, the following stopping times do not exercise
(i) 0;
(ii) (HT) 2, , (HH) = , (TH), (TT) 2, (8 dierent ones);
(iii) (HT) 2, , (HH) = , (TH) = (TT) = 1 (2 dierent ones).
For (i),

E[1
{2}
(
4
5
)

] = G
0
= 1. For (ii),

E[1
{2}
(
4
5
)

]

E[1
{

2}
(
4
5
)

], where

(HT) =
2,

(HH) = ,

(TH) =

(TT) = 2. So

E[1
{

2}
(
4
5
)

] =
1
4
[(
4
5
)
2
1 + (
4
5
)
2
(1 + 4)] = 0.96. For
(iii),

E[1
{2}
(
4
5
)

] has the biggest value when satises (HT) = 2, (HH) = , (TH) = (TT) = 1.
This value is 1.36.
4.6. (i)
Proof. The value of the put at time N, if it is not exercised at previous times, is K S
N
. Hence V
N1
=
maxKS
N1
,

E
N1
[
V
N
1+r
] = maxKS
N1
,
K
1+r
S
N1
= KS
N1
. The second equality comes from
the fact that discounted stock price process is a martingale under risk-neutral probability. By induction, we
can show V
n
= K S
n
(0 n N). So by Theorem 4.4.5, the optimal exercise policy is to sell the stock
at time zero and the value of this derivative security is K S
0
.
Remark: We cheated a little bit by using American algorithm and Theorem 4.4.5, since they are developed
for the case where is allowed to be . But intuitively, results in this chapter should still hold for the case
N, provided we replace maxG
n
, 0 with G
n
.
(ii)
Proof. This is because at time N, if we have to exercise the put and K S
N
< 0, we can exercise the
European call to set o the negative payo. In eect, throughout the portfolios lifetime, the portfolio has
intrinsic values greater than that of an American put stuck at K with expiration time N. So, we must have
V
AP
0
V
0
+V
EC
0
K S
0
+V
EC
0
.
(iii)
13
Proof. Let V
EP
0
denote the time-zero value of a European put with strike K and expiration time N. Then
V
AP
0
V
EP
0
= V
EC
0


E[
S
N
K
(1 +r)
N
] = V
EC
0
S
0
+
K
(1 +r)
N
.
4.7.
Proof. V
N
= S
N
K, V
N1
= maxS
N1
K,

E
N1
[
V
N
1+r
] = maxS
N1
K, S
N1

K
1+r
= S
N1

K
1+r
.
By induction, we can prove V
n
= S
n

K
(1+r)
Nn
(0 n N) and V
n
> G
n
for 0 n N 1. So the
time-zero value is S
0

K
(1+r)
N
and the optimal exercise time is N.
5. Random Walk
5.1. (i)
Proof. E[
2
] = E[
(21)+1
] = E[
(21)
]E[
1
] = E[
1
]
2
.
(ii)
Proof. If we dene M
(m)
n
= M
n+m
M
m
(m = 1, 2, ), then (M
(m)

)
m
as random functions are i.i.d. with
distributions the same as that of M. So
m+1

m
= infn : M
(m)
n
= 1 are i.i.d. with distributions the
same as that of
1
. Therefore
E[
m
] = E[
(mm1)+(m1m2)++1
] = E[
1
]
m
.
(iii)
Proof. Yes, since the argument of (ii) still works for asymmetric random walk.
5.2. (i)
Proof. f

() = pe

qe

, so f

() > 0 if and only if >


1
2
(lnq lnp). Since
1
2
(lnq lnp) < 0,
f() > f(0) = 1 for all > 0.
(ii)
Proof. E
n
[
Sn+1
Sn
] = E
n
[e
Xn+1
1
f()
] = pe
1
f()
+qe
1
f()
= 1.
(iii)
Proof. By optional stopping theorem, E[S
n1
] = E[S
0
] = 1. Note S
n1
= e
Mn
1
(
1
f()
)
n1
e
1
,
by bounded convergence theorem, E[1
{1<}
S
1
] = E[lim
n
S
n1
] = lim
n
E[S
n1
] = 1, that is,
E[1
{1<}
e

(
1
f()
)
1
] = 1. So e

= E[1
{1<}
(
1
f()
)
1
]. Let 0, again by bounded convergence theorem,
1 = E[1
{1<}
(
1
f(0)
)
1
] = P(
1
< ).
(iv)
Proof. Set =
1
f()
=
1
pe

+qe

, then as varies from 0 to , can take all the values in (0, 1). Write
in terms of , we have e

=
1

14pq
2
2p
(note 4pq
2
< 4(
p+q
2
)
2
1
2
= 1). We want to choose > 0, so we
should take = ln(
1+

14pq
2
2p
). Therefore E[
1
] =
2p
1+

14pq
2
=
1

14pq
2
2q
.
14
(v)
Proof.

E[
1
] = E[

1
] = E[
1

11
], and

1 4pq
2
2q

=
1
2q

(1

1 4pq
2
)
1

=
1
2q
[
1
2
(1 4pq
2
)

1
2
(4pq2)
1
+ (1

1 4pq
2
)(1)
2
].
So E[
1
] = lim
1

E[
1
] =
1
2q
[
1
2
(1 4pq)

1
2
(8pq) (1

1 4pq)] =
1
2p1
.
5.3. (i)
Proof. Solve the equation pe

+ qe

= 1 and a positive solution is ln


1+

14pq
2p
= ln
1p
p
= lnq lnp. Set

0
= lnq lnp, then f(
0
) = 1 and f

() > 0 for >


0
. So f() > 1 for all >
0
.
(ii)
Proof. As in Exercise 5.2, S
n
= e
Mn
(
1
f()
)
n
is a martingale, and 1 = E[S
0
] = E[S
n1
] = E[e
Mn
1
(
1
f()
)
1n
].
Suppose >
0
, then by bounded convergence theorem,
1 = E[ lim
n
e
Mn
1
(
1
f()
)
n1
] = E[1
{1<}
e

(
1
f()
)
1
].
Let
0
, we get P(
1
< ) = e
0
=
p
q
< 1.
(iii)
Proof. From (ii), we can see E[1
{1<}
(
1
f()
)
1
] = e

, for >
0
. Set =
1
f()
, then e

=
1

14pq
2
2p
. We
need to choose the root so that e

> e
0
=
q
p
, so = ln(
1+

14pq
2
2p
), then E[
1
1
{1<}
] =
1

14pq
2
2q
.
(iv)
Proof. E[
1
1
{1<}
] =

E[
1
1
{1<}
][
=1
=
1
2q
[
4pq

14pq
(1

1 4pq)] =
1
2q
[
4pq
2q1
1 + 2q 1] =
p
q
1
2q1
.
5.4. (i)
Proof. E[
2
] =

k=1
P(
2
= 2k)
2k
=

k=1
(

2
)
2k
P(
2
= 2k)4
k
. So P(
2
= 2k) =
(2k)!
4
k
(k+1)!k!
.
(ii)
Proof. P(
2
= 2) =
1
4
. For k 2, P(
2
= 2k) = P(
2
2k) P(
2
2k 2).
P(
2
2k) = P(M
2k
= 2) +P(M
2k
4) +P(
2
2k, M
2k
0)
= P(M
2k
= 2) + 2P(M
2k
4)
= P(M
2k
= 2) +P(M
2k
4) +P(M
2k
4)
= 1 P(M
2k
= 2) P(M
2k
= 0).
15
Similarly, P(
2
2k 2) = 1 P(M
2k2
= 2) P(M
2k2
= 0). So
P(
2
= 2k) = P(M
2k2
= 2) +P(M
2k2
= 0) P(M
2k
= 2) P(M
2k
= 0)
= (
1
2
)
2k2

(2k 2)!
k!(k 2)!
+
(2k 2)!
(k 1)!(k 1)!

(
1
2
)
2k

(2k)!
(k + 1)!(k 1)!
+
(2k)!
k!k!

=
(2k)!
4
k
(k + 1)!k!

4
2k(2k 1)
(k + 1)k(k 1) +
4
2k(2k 1)
(k + 1)k
2
k (k + 1)

=
(2k)!
4
k
(k + 1)!k!

2(k
2
1)
2k 1
+
2(k
2
+k)
2k 1

4k
2
1
2k 1

=
(2k)!
4
k
(k + 1)!k!
.
5.5. (i)
Proof. For every path that crosses level m by time n and resides at b at time n, there corresponds a reected
path that resides at time 2mb. So
P(M

n
m, M
n
= b) = P(M
n
= 2mb) = (
1
2
)
n
n!
(m+
nb
2
)!(
n+b
2
m)!
.
(ii)
Proof.
P(M

n
m, M
n
= b) =
n!
(m+
nb
2
)!(
n+b
2
m)!
p
m+
nb
2
q
n+b
2
m
.
5.6.
Proof. On the innite coin-toss space, we dene M
n
= stopping times that takes values 0, 1, , n,
and M

= stopping times that takes values 0, 1, 2, . Then the time-zero value V

of the perpetual
American put as in Section 5.4 can be dened as sup
M

E[1
{<}
(KS )
+
(1+r)

]. For an American put with


the same strike price K that expires at time n, its time-zero value V
(n)
is max
Mn

E[1
{<}
(KS )
+
(1+r)

].
Clearly (V
(n)
)
n0
is nondecreasing and V
(n)
V

for every n. So lim


n
V
(n)
exists and lim
n
V
(n)
V

.
For any given M

, we dene
(n)
=

, if =
n, if <
, then
(n)
is also a stopping time,
(n)
M
n
and lim
n

(n)
= . By bounded convergence theorem,

1
{<}
(K S

)
+
(1 +r)

= lim
n

1
{
(n)
<}
(K S

(n) )
+
(1 +r)

(n)

lim
n
V
(n)
.
Take sup at the left hand side of the inequality, we get V

lim
n
V
(n)
. Therefore V

= lim
n
V
(n)
.
Remark: In the above proof, rigorously speaking, we should use (KS

) in the places of (KS

)
+
. So
this needs some justication.
5.8. (i)
16
Proof. v(S
n
) = S
n
S
n
K = g(S
n
). Under risk-neutral probabilities,
1
(1+r)
n
v(S
n
) =
Sn
(1+r)
n
is a martingale
by Theorem 2.4.4.
(ii)
Proof. If the purchaser chooses to exercises the call at time n, then the discounted risk-neutral expectation
of her payo is

E

SnK
(1+r)
n

= S
0

K
(1+r)
n
. Since lim
n

S
0

K
(1+r)
n

= S
0
, the value of the call at time
zero is at least sup
n

S
0

K
(1+r)
n

= S
0
.
(iii)
Proof. max

g(s),
pv(us)+ qv(ds)
1+r

= maxs K,
pu+ qv
1+r
s = maxs K, s = s = v(s), so equation (5.4.16) is
satised. Clearly v(s) = s also satises the boundary condition (5.4.18).
(iv)
Proof. Suppose is an optimal exercise time, then

E

S K
(1+r)

1
{<}

S
0
. Then P( < ) = 0 and

K
(1+r)

1
{<}

> 0. So

E

S K
(1+r)

1
{<}

<

E

S
(1+r)

1
{<}

. Since

Sn
(1+r)
n

n0
is a martingale
under risk-neutral measure, by Fatous lemma,

E

S
(1+r)

1
{<}

liminf
n

E

Sn
(1+r)
n
1
{<}

=
liminf
n

E

Sn
(1+r)
n

= liminf
n

E[S
0
] = S
0
. Combined, we have S
0


E

S K
(1+r)

1
{<}

< S
0
.
Contradiction. So there is no optimal time to exercise the perpetual American call. Simultaneously, we have
shown

E

S K
(1+r)

1
{<}

< S
0
for any stopping time . Combined with (ii), we conclude S
0
is the least
upper bound for all the prices acceptable to the buyer.
5.9. (i)
Proof. Suppose v(s) = s
p
, then we have s
p
=
2
5
2
p
s
p
+
2
5
s
p
2
p
. So 1 =
2
p+1
5
+
2
1p
5
. Solve it for p, we get p = 1
or p = 1.
(ii)
Proof. Since lim
s
v(s) = lim
s
(As +
B
s
) = 0, we must have A = 0.
(iii)
Proof. f
B
(s) = 0 if and only if B+s
2
4s = 0. The discriminant = (4)
2
4B = 4(4B). So for B 4,
the equation has roots and for B > 4, this equation does not have roots.
(iv)
Proof. Suppose B 4, then the equation s
2
4s + B = 0 has solution 2

4 B. By drawing graphs of
4 s and
B
s
, we should choose B = 4 and s
B
= 2 +

4 B = 2.
(v)
Proof. To have continuous derivative, we must have 1 =
B
s
2
B
. Plug B = s
2
B
back into s
2
B
4s
B
+B = 0,
we get s
B
= 2. This gives B = 4.
6. Interest-Rate-Dependent Assets
6.2.
17
Proof. X
k
= S
k
E
k
[D
m
(S
m
K)]D
1
k

Sn
Bn,m
B
k,m
for n k m. Then
E
k1
[D
k
X
k
] = E
k1
[D
k
S
k
E
k
[D
m
(S
m
K)]
S
n
B
n,m
B
k,m
D
k
]
= D
k1
S
k1
E
k1
[D
m
(S
m
K)]
S
n
B
n,m
E
k1
[E
k
[D
m
]]
= D
k1
[S
k1
E
k1
[D
m
(S
m
K)]D
1
k1

S
n
B
n,m
B
k1,m
]
= D
k1
X
k1
.
6.3.
Proof.
1
D
n

E
n
[D
m+1
R
m
] =
1
D
n

E
n
[D
m
(1 +R
m
)
1
R
m
] =

E
n
[
D
m
D
m+1
D
n
] = B
n,m
B
n,m+1
.
6.4.(i)
Proof. D
1
V
1
= E
1
[D
3
V
3
] = E
1
[D
2
V
2
] = D
2
E
1
[V
2
]. So V
1
=
D2
D1
E
1
[V
2
] =
1
1+R1
E
1
[V
2
]. In particular,
V
1
(H) =
1
1+R1(H)
V
2
(HH)P(w
2
= H[w
1
= H) =
4
21
, V
1
(T) = 0.
(ii)
Proof. Let X
0
=
2
21
. Suppose we buy
0
shares of the maturity two bond, then at time one, the value of
our portfolio is X
1
= (1 +R
0
)(X
0
B
0,2
) +
0
B
1,2
. To replicate the value V
1
, we must have

V
1
(H) = (1 +R
0
)(X
0

0
B
0,2
) +
0
B
1,2
(H)
V
1
(T) = (1 +R
0
)(X
0

0
B
0,2
) +
0
B
1,2
(T).
So
0
=
V1(H)V1(T)
B1,2(H)B1,2(T)
=
4
3
. The hedging strategy is therefore to borrow
4
3
B
0,2

2
21
=
20
21
and buy
4
3
share of the maturity two bond. The reason why we do not invest in the maturity three bond is that
B
1,3
(H) = B
1,3
(T)(=
4
7
) and the portfolio will therefore have the same value at time one regardless the
outcome of rst coin toss. This makes impossible the replication of V
1
, since V
1
(H) = V
1
(T).
(iii)
Proof. Suppose we buy
1
share of the maturity three bond at time one, then to replicate V
2
at time
two, we must have V
2
= (1 + R
1
)(X
1

1
B
1,3
) +
1
B
2,3
. So
1
(H) =
V2(HH)V2(HT)
B2,3(HH)B2,3(HT)
=
2
3
, and

1
(T) =
V2(TH)V2(TT)
B2,3(TH)B2,3(TT)
= 0. So the hedging strategy is as follows. If the outcome of rst coin toss is
T, then we do nothing. If the outcome of rst coin toss is H, then short
2
3
shares of the maturity three
bond and invest the income into the money market account. We do not invest in the maturity two bond,
because at time two, the value of the bond is its face value and our portfolio will therefore have the same
value regardless outcomes of coin tosses. This makes impossible the replication of V
2
.
6.5. (i)
18
Proof. Suppose 1 n m, then

E
m+1
n1
[F
n,m
] =

E
n1
[B
1
n,m+1
(B
n,m
B
n,m+1
)Z
n,m+1
Z
1
n1,m+1
]
=

E
n1

B
n,m
B
n,m+1
1

B
n,m+1
D
n
B
n1,m+1
D
n1

=
D
n
B
n1,m+1
D
n1

E
n1
[D
1
n

E
n
[D
m
] D
1
n

E
n
[D
m+1
]]
=

E
n1
[D
m
D
m+1
]
B
n1,m1
D
n1
=
B
n1,m
B
n1,m+1
B
n1,m+1
= F
n1,m
.
6.6. (i)
Proof. The agent enters the forward contract at no cost. He is obliged to buy certain asset at time m at
the strike price K = For
n,m
=
Sn
Bn,m
. At time n + 1, the contract has the value

E
n+1
[D
m
(S
m
K)] =
S
n+1
KB
n+1,m
= S
n+1

SnBn+1,m
Bn,m
. So if the agent sells this contract at time n +1, he will receive a cash
ow of S
n+1

SnBn+1,m
Bn,m
(ii)
Proof. By (i), the cash ow generated at time n + 1 is
(1 +r)
mn1

S
n+1

S
n
B
n+1,m
B
n,m

= (1 +r)
mn1

S
n+1

Sn
(1+r)
mn1
1
(1+r)
mn

= (1 +r)
mn1
S
n+1
(1 +r)
mn
S
n
= (1 +r)
m

E
n1
[
S
m
(1 +r)
m
] + (1 +r)
m

E
n
[
S
m
(1 +r)
m
]
= Fut
n+1,m
Fut
n,m
.
6.7.
Proof.

n+1
(0) =

E[D
n+1
V
n+1
(0)]
=

E[
D
n
1 +r
n
(0)
1
{#H(1n+1)=0}
]
=

E[
D
n
1 +r
n
(0)
1
{#H(1n)=0}
1
{n+1=T}
]
=
1
2

E[
D
n
1 +r
n
(0)
]
=

n
(0)
2(1 +r
n
(0))
.
19
For k = 1, 2, , n,

n+1
(k) =

E

D
n
1 +r
n
(#H(
1

n
))
1
{#H(1n+1)=k}

=

E

D
n
1 +r
n
(k)
1
{#H(1n)=k}
1
{n+1=T}

+

E

D
n
1 +r
n
(k 1)
1
{#H(1n)=k}
1
{n+1=H}

=
1
2

E[D
n
V
n
(k)]
1 +r
n
(k)
+
1
2

E[D
n
V
n
(k 1)]
1 +r
n
(k 1)
=

n
(k)
2(1 +r
n
(k))
+

n
(k 1)
2(1 +r
n
(k 1))
.
Finally,

n+1
(n + 1) =

E[D
n+1
V
n+1
(n + 1)] =

E

D
n
1 +r
n
(n)
1
{#H(1n)=n}
1
{n+1=H}

=

n
(n)
2(1 +r
n
(n))
.
Remark: In the above proof, we have used the independence of
n+1
and (
1
, ,
n
). This is guaranteed
by the assumption that p = q =
1
2
(note if and only if E[[] = constant). In case the binomial model
has stochastic up- and down-factor u
n
and d
n
, we can use the fact that

P(
n+1
= H[
1
, ,
n
) = p
n
and

P(
n+1
= T[
1
, ,
n
) = q
n
, where p
n
=
1+rndn
undn
and q
n
=
u1rn
undn
(cf. solution of Exercise 2.9 and
notes on page 39). Then for any X T
n
= (
1
, ,
n
), we have

E[Xf(
n+1
)] =

E[X

E[f(
n+1
)[T
n
]] =

E[X(p
n
f(H) +q
n
f(T))].
20
2 Stochastic Calculus for Finance II: Continuous-Time Models
1. General Probability Theory
1.1. (i)
Proof. P(B) = P((B A) A) = P(B A) +P(A) P(A).
(ii)
Proof. P(A) P(A
n
) implies P(A) lim
n
P(A
n
) = 0. So 0 P(A) 0, which means P(A) = 0.
1.2. (i)
Proof. We dene a mapping from A to as follows: (
1

2
) =
1

5
. Then is one-to-one and
onto. So the cardinality of A is the same as that of , which means in particular that A is uncountably
innite.
(ii)
Proof. Let A
n
= =
1

2
:
1
=
2
, ,
2n1
=
2n
. Then A
n
A as n . So
P(A) = lim
n
P(A
n
) = lim
n
[P(
1
=
2
) P(
2n1
=
2n
)] = lim
n
(p
2
+ (1 p)
2
)
n
.
Since p
2
+ (1 p)
2
< 1 for 0 < p < 1, we have lim
n
(p
2
+ (1 p)
2
)
n
= 0. This implies P(A) = 0.
1.3.
Proof. Clearly P() = 0. For any A and B, if both of them are nite, then A B is also nite. So
P(AB) = 0 = P(A) +P(B). If at least one of them is innite, then AB is also innite. So P(AB) =
= P(A) +P(B). Similarly, we can prove P(
N
n=1
A
n
) =

N
n=1
P(A
n
), even if A
n
s are not disjoint.
To see countable additivity property doesnt hold for P, let A
n
=
1
n
. Then A =

n=1
A
n
is an innite
set and therefore P(A) = . However, P(A
n
) = 0 for each n. So P(A) =

n=1
P(A
n
).
1.4. (i)
Proof. By Example 1.2.5, we can construct a random variable X on the coin-toss space, which is uniformly
distributed on [0, 1]. For the strictly increasing and continuous function N(x) =

x

2
e

2
2
d, we let
Z = N
1
(X). Then P(Z a) = P(X N(a)) = N(a) for any real number a, i.e. Z is a standard normal
random variable on the coin-toss space (

, T

, P).
(ii)
Proof. Dene X
n
=

n
i=1
Yi
2
i
, where
Y
i
() =

1, if
i
= H
0, if
i
= T.
Then X
n
() X() for every

where X is dened as in (i). So Z


n
= N
1
(X
n
) Z = N
1
(X) for
every . Clearly Z
n
depends only on the rst n coin tosses and Z
n

n1
is the desired sequence.
1.5.
21
Proof. First, by the information given by the problem, we have


0
1
[0,X())
(x)dxdP() =

1
[0,X())
(x)dP()dx.
The left side of this equation equals to

X()
0
dxdP() =

X()dP() = EX.
The right side of the equation equals to

1
{x<X()}
dP()dx =


0
P(x < X)dx =


0
(1 F(x))dx.
So EX =

0
(1 F(x))dx.
1.6. (i)
Proof.
Ee
uX
=

e
ux
1

2
e

(x)
2
2
2
dx
=

2
e

(x)
2
2
2
ux
2
2
dx
=

2
e

[x(+
2
u)]
2
(2
2
u+
4
u
2
)
2
2
dx
= e
u+

2
u
2
2

2
e

[x(+
2
u)]
2
2
2
dx
= e
u+

2
u
2
2
(ii)
Proof. E(X) = Ee
uX
= e
u+
u
2

2
2
e
u
= (EX).
1.7. (i)
Proof. Since [f
n
(x)[
1

2n
, f(x) = lim
n
f
n
(x) = 0.
(ii)
Proof. By the change of variable formula,

f
n
(x)dx =

2
e

x
2
2
dx = 1. So we must have
lim
n

f
n
(x)dx = 1.
(iii)
Proof. This is not contradictory with the Monotone Convergence Theorem, since f
n

n1
doesnt increase
to 0.
22
1.8. (i)
Proof. By (1.9.1), [Y
n
[ =

e
tX
e
snX
tsn

= [Xe
X
[ = Xe
X
Xe
2tX
. The last inequality is by X 0 and the
fact that is between t and s
n
, and hence smaller than 2t for n suciently large. So by the Dominated
Convergence Theorem,

(t) = lim
n
EY
n
= Elim
n
Y
n
= EXe
tX
.
(ii)
Proof. Since E[e
tX
+
1
{X0}
] +E[e
tX

1
{X<0}
] = E[e
tX
] < for every t R, E[e
t|X|
] = E[e
tX
+
1
{X0}
] +
E[e
(t)X

1
{X<0}
] < for every t R. Similarly, we have E[[X[e
t|X|
] < for every t R. So, similar to
(i), we have [Y
n
[ = [Xe
X
[ [X[e
2t|X|
for n suciently large, So by the Dominated Convergence Theorem,

(t) = lim
n
EY
n
= Elim
n
Y
n
= EXe
tX
.
1.9.
Proof. If g(x) is of the form 1
B
(x), where B is a Borel subset of R, then the desired equality is just (1.9.3).
By the linearity of Lebesgue integral, the desired equality also holds for simple functions, i.e. g of the
form g(x) =

n
i=1
1
Bi
(x), where each B
i
is a Borel subset of R. Since any nonnegative, Borel-measurable
function g is the limit of an increasing sequence of simple functions, the desired equality can be proved by
the Monotone Convergence Theorem.
1.10. (i)
Proof. If A
i

i=1
is a sequence of disjoint Borel subsets of [0, 1], then by the Monotone Convergence Theorem,

P(

i=1
A
i
) equals to

i=1
Ai
ZdP =

lim
n
1

n
i=1
Ai
ZdP = lim
n

n
i=1
Ai
ZdP = lim
n
n

i=1

Ai
ZdP =

i=1

P(A
i
).
Meanwhile,

P() = 2P([
1
2
, 1]) = 1. So

P is a probability measure.
(ii)
Proof. If P(A) = 0, then

P(A) =

A
ZdP = 2

A[
1
2
,1]
dP = 2P(A [
1
2
, 1]) = 0.
(iii)
Proof. Let A = [0,
1
2
).
1.11.
Proof.

Ee
uY
= Ee
uY
Z = Ee
uX+u
e
X

2
2
= e
u

2
2
Ee
(u)X
= e
u

2
2
e
(u)
2
2
= e
u
2
2
.
1.12.
Proof. First,

Z = e
Y

2
2
= e
(X+)

2
2
= e

2
2
+X
= Z
1
. Second, for any A T,

P(A) =

A

Zd

P =

(1
A

Z)ZdP =

1
A
dP = P(A). So P =

P. In particular, X is standard normal under

P, since its standard
normal under P.
1.13. (i)
23
Proof.
1

P(X B(x, )) =
1

x+

2
x

2
1

2
e

u
2
2
du is approximately
1

2
e

x
2
2
=
1

2
e

X
2
( )
2
.
(ii)
Proof. Similar to (i).
(iii)
Proof. X B(x, ) = X B(y , ) = X + B(y, ) = Y B(y, ).
(iv)
Proof. By (i)-(iii),

P(A)
P(A)
is approximately

2
e

Y
2
( )
2

2
e

X
2
( )
2
= e

Y
2
( )X
2
( )
2
= e

(X( )+)
2
X
2
( )
2
= e
X( )

2
2
.
1.14. (i)
Proof.

P() =

e
(

)X
dP =


0
e
(

)x
e
x
dx =

x
dx = 1.
(ii)
Proof.

P(X a) =

{Xa}

e
(

)X
dP =

a
0

e
(

)x
e
x
dx =

a
0

x
dx = 1 e

a
.
1.15. (i)
Proof. Clearly Z 0. Furthermore, we have
EZ = E

h(g(X))g

(X)
f(X)

h(g(x))g

(x)
f(x)
f(x)dx =

h(g(x))dg(x) =

h(u)du = 1.
(ii)
Proof.

P(Y a) =

{g(X)a}
h(g(X))g

(X)
f(X)
dP =

g
1
(a)

h(g(x))g

(x)
f(x)
f(x)dx =

g
1
(a)

h(g(x))dg(x).
By the change of variable formula, the last equation above equals to

a

h(u)du. So Y has density h under

P.
24
2. Information and Conditioning
2.1.
Proof. For any real number a, we have X a T
0
= , . So P(X a) is either 0 or 1. Since
lim
a
P(X a) = 1 and lim
a
P(X a) = 0, we can nd a number x
0
such that P(X x
0
) = 1 and
P(X x) = 0 for any x < x
0
. So
P(X = x
0
) = lim
n
P(x
0

1
n
< X x
0
) = lim
n
(P(X x
0
) P(X x
0

1
n
)) = 1.
2.2. (i)
Proof. (X) = , , HT, TH, TT, HH.
(ii)
Proof. (S
1
) = , , HH, HT, TH, TT.
(iii)
Proof.

P(HT, TH HH, HT) =

P(HT) =
1
4
,

P(HT, TH) =

P(HT) +

P(TH) =
1
4
+
1
4
=
1
2
,
and

P(HH, HT) =

P(HH) +

P(HT) =
1
4
+
1
4
=
1
2
. So we have

P(HT, TH HH, HT) =



P(HT, TH)

P(HH, HT).
Similarly, we can work on other elements of (X) and (S
1
) and show that

P(A B) =

P(A)

P(B) for any


A (X) and B (S
1
). So (X) and (S
1
) are independent under

P.
(iv)
Proof. P(HT, TH HH, HT) = P(HT) =
2
9
, P(HT, TH) =
2
9
+
2
9
=
4
9
and P(HH, HT) =
4
9
+
2
9
=
6
9
. So
P(HT, TH HH, HT) = P(HT, TH)P(HH, HT).
Hence (X) and (S
1
) are not independent under P.
(v)
Proof. Because S
1
and X are not independent under the probability measure P, knowing the value of X
will aect our opinion on the distribution of S
1
.
2.3.
Proof. We note (V, W) are jointly Gaussian, so to prove their independence it suces to show they are
uncorrelated. Indeed, EV W = EX
2
sin cos +XY cos
2
XY sin
2
+Y
2
sin cos = sin cos +
0 + 0 + sin cos = 0.
2.4. (i)
25
Proof.
Ee
uX+vY
= Ee
uX+vXZ

= Ee
uX+vXZ
[Z = 1P(Z = 1) +Ee
uX+vXZ
[Z = 1P(Z = 1)
=
1
2
Ee
uX+vX
+
1
2
Ee
uXvX

=
1
2
[e
(u+v)
2
2
+e
(uv)
2
2
]
= e
u
2
+v
2
2
e
uv
+e
uv
2
.
(ii)
Proof. Let u = 0.
(iii)
Proof. Ee
uX
= e
u
2
2
and Ee
vY
= e
v
2
2
. So Ee
uX+vY
= Ee
uX
Ee
vY
. Therefore X and Y cannot
be independent.
2.5.
Proof. The density f
X
(x) of X can be obtained by
f
X
(x) =

f
X,Y
(x, y)dy =

{y|x|}
2[x[ +y

2
e

(2|x|+y)
2
2
dy =

{|x|}

2
e

2
2
d =
1

2
e

x
2
2
.
The density f
Y
(y) of Y can be obtained by
f
Y
(y) =

f
XY
(x, y)dx
=

1
{|x|y}
2[x[ +y

2
e

(2|x|+y)
2
2
dx
=


0(y)
2x +y

2
e

(2x+y)
2
2
dx +

0y

2x +y

2
e

(2x+y)
2
2
dx
=


0(y)
2x +y

2
e

(2x+y)
2
2
dx +

0(y)

2x +y

2
e

(2x+y)
2
2
d(x)
= 2


|y|

2
e

2
2
d(

2
)
=
1

2
e

y
2
2
.
So both X and Y are standard normal random variables. Since f
X,Y
(x, y) = f
X
(x)f
Y
(y), X and Y are not
26
independent. However, if we set F(t) =

t
u
2

2
e

u
2
2
du, we have
EXY =

xyf
X,Y
(x, y)dxdy
=

xy1
{y|x|}
2[x[ +y

2
e

(2|x|+y)
2
2
dxdy
=

xdx


|x|
y
2[x[ +y

2
e

(2|x|+y)
2
2
dy
=

xdx


|x|
( 2[x[)

2
e

2
2
d
=

xdx(


|x|

2
e

2
2
d 2[x[
e

x
2
2

2
)
=


0
x

2
e

2
2
ddx +

2
e

2
2
ddx
=


0
xF(x)dx +

xF(x)dx.
So EXY =

0
xF(x)dx

0
uF(u)du = 0.
2.6. (i)
Proof. (X) = , , a, b, c, d.
(ii)
Proof.
EY [X =

{a,b,c,d}
EY 1
{X=}

P(X = )
1
{X=}
.
(iii)
Proof.
EZ[X = X +EY [X = X +

{a,b,c,d}
EY 1
{X=}

P(X = )
1
{X=}
.
(iv)
Proof. EZ[X EY [X = EZ Y [X = EX[X = X.
2.7.
27
Proof. Let = EY X and = EY X [(. Note is (-measurable, we have
V ar(Y X) = E(Y X )
2

= E[(Y EY [() + (EY [( X )]


2

= V ar(Err) + 2E(Y EY [() +E


2

= V ar(Err) + 2EY EY [( +E
2

= V ar(Err) +E
2

V ar(Err).
2.8.
Proof. It suces to prove the more general case. For any (X)-measurable random variable , EY
2
=
E(Y EY [X) = EY EY [X = EY EY = 0.
2.9. (i)
Proof. Consider the dice-toss space similar to the coin-toss space. Then a typical element in this space
is an innite sequence
1

3
, with
i
1, 2, , 6 (i N). We dene X() =
1
and f(x) =
1
{odd integers}
(x). Then its easy to see
(X) = , , :
1
= 1, , :
1
= 6
and (f(X)) equals to
, , :
1
= 1 :
1
= 3 :
1
= 5, :
1
= 2 :
1
= 4 :
1
= 6.
So , (f(X)) (X), and each of these containment is strict.
(ii)
Proof. No. (f(X)) (X) is always true.
2.10.
Proof.

A
g(X)dP = Eg(X)1
B
(X)
=

g(x)1
B
(x)f
X
(x)dx
=

yf
X,Y
(x, y)
f
X
(x)
dy1
B
(x)f
X
(x)dx
=

y1
B
(x)f
X,Y
(x, y)dxdy
= EY 1
B
(X)
= EY I
A

A
Y dP.
28
2.11. (i)
Proof. We can nd a sequence W
n

n1
of (X)-measurable simple functions such that W
n
W. Each W
n
can be written in the form

Kn
i=1
a
n
i
1
A
n
i
, where A
n
i
s belong to (X) and are disjoint. So each A
n
i
can be
written as X B
n
i
for some Borel subset B
n
i
of R, i.e. W
n
=

Kn
i=1
a
n
i
1
{XB
n
i
}
=

Kn
i=1
a
n
i
1
B
n
i
(X) = g
n
(X),
where g
n
(x) =

Kn
i=1
a
n
i
1
B
n
i
(x). Dene g = limsupg
n
, then g is a Borel function. By taking upper limits on
both sides of W
n
= g
n
(X), we get W = g(X).
(ii)
Proof. Note EY [X is (X)-measurable. By (i), we can nd a Borel function g such that EY [X =
g(X).
3. Brownian Motion
3.1.
Proof. We have T
t
T
u1
and W
u2
W
u1
is independent of T
u1
. So in particular, W
u2
W
u1
is independent
of T
t
.
3.2.
Proof. E[W
2
t
W
2
s
[T
s
] = E[(W
t
W
s
)
2
+ 2W
t
W
s
2W
2
s
[T
s
] = t s + 2W
s
E[W
t
W
s
[T
s
] = t s.
3.3.
Proof.
(3)
(u) = 2
4
ue
1
2

2
u
2
+(
2
+
4
u
2
)
2
ue
1
2

2
u
2
= e
1
2

2
u
2
(3
4
u+
4
u
2
), and
(4)
(u) =
2
ue
1
2

2
u
2
(3
4
u+

4
u
2
) +e
1
2

2
u
2
(3
4
+ 2
4
u). So E[(X )
4
] =
(4)
(0) = 3
4
.
3.4. (i)
Proof. Assume there exists A T, such that P(A) > 0 and for every A, lim
n

n1
j=0
[W
tj+1
W
tj
[() <
. Then for every A,

n1
j=0
(W
tj+1
W
tj
)
2
() max
0kn1
[W
t
k+1
W
t
k
[()

n1
j=0
[W
tj+1
W
tj
[()
0, since lim
n
max
0kn1
[W
t
k+1
W
t
k
[() = 0. This is a contradiction with lim
n

n1
j=0
(W
tj+1

W
tj
)
2
= T a.s..
(ii)
Proof. Note

n1
j=0
(W
tj+1
W
tj
)
3
max
0kn1
[W
t
k+1
W
t
k
[

n1
j=0
(W
tj+1
W
tj
)
2
0 as n , by an
argument similar to (i).
3.5.
29
Proof.
E[e
rT
(S
T
K)
+
]
= e
rT

(ln
K
S
0
(r
1
2

2
)T)
(S
0
e
(r
1
2

2
)T+x
K)
e

x
2
2T

2T
dx
= e
rT

T
(ln
K
S
0
(r
1
2

2
)T)
(S
0
e
(r
1
2

2
)T+

Ty
K)
e

y
2
2

2
dy
= S
0
e

1
2

2
T

T
(ln
K
S
0
(r
1
2

2
)T)
1

2
e

y
2
2
+

Ty
dy Ke
rT

T
(ln
K
S
0
(r
1
2

2
)T)
1

2
e

y
2
2
dy
= S
0

T
(ln
K
S
0
(r
1
2

2
)T)

T
1

2
e

2
2
d Ke
rT
N

T
(ln
S
0
K
+ (r
1
2

2
)T)

= Ke
rT
N(d
+
(T, S
0
)) Ke
rT
N(d

(T, S
0
)).
3.6. (i)
Proof.
E[f(X
t
)[T
t
] = E[f(W
t
W
s
+a)[T
s
][
a=Ws+t
= E[f(W
ts
+a)][
a=Ws+t
=

f(x +W
s
+t)
e

x
2
2(ts)

2(t s)
dx
=

f(y)
e

(yWss(ts))
2
2(ts)

2(t s)
dy
= g(X
s
).
So E[f(X
t
)[T
s
] =

f(y)p(t s, X
s
, y)dy with p(, x, y) =
1

2
e

(yx)
2
2
.
(ii)
Proof. E[f(S
t
)[T
s
] = E[f(S
0
e
Xt
)[T
s
] with =
v

. So by (i),
E[f(S
t
)[T
s
] =

f(S
0
e
y
)
1

2(t s)
e

(yXs(ts))
2
2(ts)
dy
S0e
y
=z
=


0
f(z)
1

2(t s)
e

(
1

ln
z
S
0

ln
Ss
S
0
(ts))
2
2
dz
z
=


0
f(z)
e

(ln
z
Ss
v(ts))
2
2
2
(ts)
z

2(t s)
dz
=


0
f(z)p(t s, S
s
, z)dz
= g(S
s
).
3.7. (i)
30
Proof. E

Zt
Zs
[T
s

= E[exp(W
t
W
s
) +(t s) ( +

2
2
)(t s)] = 1.
(ii)
Proof. By optional stopping theorem, E[Z
tm
] = E[Z
0
] = 1, that is, E[expX
tm
( +

2
2
)t
m
] =
1.
(iii)
Proof. If 0 and > 0, Z
tm
e
m
. By bounded convergence theorem,
E[1
{m<}
Z
m
] = E[ lim
t
Z
tm
] = lim
t
E[Z
tm
] = 1,
since on the event
m
= , Z
tm
e
m
1
2

2
t
0 as t . Therefore, E[e
m(+

2
2
)m
] = 1. Let
0, by bounded convergence theorem, we have P(
m
< ) = 1. Let +

2
2
= , we get
E[e
m
] = e
m
= e
mm

2+
2
.
(iv)
Proof. We note for > 0, E[
m
e
m
] < since xe
x
is bounded on [0, ). So by an argument similar
to Exercise 1.8, E[e
m
] is dierentiable and

E[e
m
] = E[
m
e
m
] = e
mm

2+
2 m

2 +
2
.
Let 0, by monotone increasing theorem, E[
m
] =
m

< for > 0.


(v)
Proof. By > 2 > 0, we get +

2
2
> 0. Then Z
tm
e
m
and on the event
m
= , Z
tm

e
m(

2
2
+)t
0 as t . Therefore,
E[e
m(+

2
2
)m
1
{m<}
] = E[ lim
t
Z
tm
] = lim
t
E[Z
tm
] = 1.
Let 2, then we get P(
m
< ) = e
2m
= e
2||m
< 1. Set = +

2
2
. So we get
E[e
m
] = E[e
m
1
{m<}
] = e
m
= e
mm

2+
2
.
3.8. (i)
Proof.

n
(u) =

E[e
u
1

n
Mnt,n
] = (

E[e
u

n
X1,n
])
nt
= (e
u

n
p
n
+e

n
q
n
)
nt
=

e
u

r
n
+ 1 e

n
e

n
e

+e

r
n
1 +e

n
e

n
e

nt
.
(ii)
31
Proof.
1
x
2
(u) =

e
ux

rx
2
+ 1 e
x
e
x
e
x

e
ux

rx
2
+ 1 e
x
e
x
e
x

t
x
2
.
So,
ln 1
x
2
(u) =
t
x
2
ln

(rx
2
+ 1)(e
ux
e
ux
) +e
(u)x
e
(u)x
e
x
e
x

=
t
x
2
ln

(rx
2
+ 1) sinhux + sinh( u)x
sinhx

=
t
x
2
ln

(rx
2
+ 1) sinhux + sinhxcoshux coshxsinhux
sinhx

=
t
x
2
ln

coshux +
(rx
2
+ 1 coshx) sinhux
sinhx

.
(iii)
Proof.
coshux +
(rx
2
+ 1 coshx) sinhux
sinhx
= 1 +
u
2
x
2
2
+O(x
4
) +
(rx
2
+ 1 1

2
x
2
2
+O(x
4
))(ux +O(x
3
))
x +O(x
3
)
= 1 +
u
2
x
2
2
+
(r

2
2
)ux
3
+O(x
5
)
x +O(x
3
)
+O(x
4
)
= 1 +
u
2
x
2
2
+
(r

2
2
)ux
3
(1 +O(x
2
))
x(1 +O(x
2
))
+O(x
4
)
= 1 +
u
2
x
2
2
+
rux
2


1
2
ux
2
+O(x
4
).
(iv)
Proof.
ln 1
x
2
=
t
x
2
ln(1 +
u
2
x
2
2
+
ru

x
2

ux
2
2
+O(x
4
)) =
t
x
2
(
u
2
x
2
2
+
ru

x
2

ux
2
2
+O(x
4
)).
So lim
x0
ln 1
x
2
(u) = t(
u
2
2
+
ru


u
2
), and

E[e
u
1

n
Mnt,n
] =
n
(u)
1
2
tu
2
+ t(
r



2
)u. By the one-to-one
correspondence between distribution and moment generating function, (
1

n
M
nt,n
)
n
converges to a Gaussian
random variable with mean t(
r



2
) and variance t. Hence (

n
M
nt,n
)
n
converges to a Gaussian random
variable with mean t(r

2
2
) and variance
2
t.
4. Stochastic Calculus
4.1.
32
Proof. Fix t and for any s < t, we assume s [t
m
, t
m+1
) for some m.
Case 1. m = k. Then I(t)I(s) =
t
k
(M
t
M
t
k
)
t
k
(M
s
M
t
k
) =
t
k
(M
t
M
s
). So E[I(t)I(s)[T
t
] =

t
k
E[M
t
M
s
[T
s
] = 0.
Case 2. m < k. Then t
m
s < t
m+1
t
k
t < t
k+1
. So
I(t) I(s) =
k1

j=m

tj
(M
tj+1
M
tj
) +
t
k
(M
s
M
t
k
)
tm
(M
s
M
tm
)
=
k1

j=m+1

tj
(M
tj+1
M
tj
) +
t
k
(M
t
M
t
k
) +
tm
(M
tm+1
M
s
).
Hence
E[I(t) I(s)[T
s
]
=
k1

j=m+1
E[
tj
E[M
tj+1
M
tj
[T
tj
][T
s
] +E[
t
k
E[M
t
M
t
k
[T
t
k
][T
s
] +
tm
E[M
tm+1
M
s
[T
s
]
= 0.
Combined, we conclude I(t) is a martingale.
4.2. (i)
Proof. We follow the simplication in the hint and consider I(t
k
) I(t
l
) with t
l
< t
k
. Then I(t
k
) I(t
l
) =

k1
j=l

tj
(W
tj+1
W
tj
). Since
t
is a non-random process and W
tj+1
W
tj
T
tj
T
t
l
for j l, we must
have I(t
k
) I(t
l
) T
t
l
.
(ii)
Proof. We use the notation in (i) and it is clear I(t
k
) I(t
l
) is normal since it is a linear combination of
independent normal random variables. Furthermore, E[I(t
k
) I(t
l
)] =

k1
j=l

tj
E[W
tj+1
W
tj
] = 0 and
V ar(I(t
k
) I(t
l
)) =

k1
j=l

2
tj
V ar(W
tj+1
W
tj
) =

k1
j=l

2
tj
(t
j+1
t
j
) =

t
k
t
l

2
u
du.
(iii)
Proof. E[I(t) I(s)[T
s
] = E[I(t) I(s)] = 0, for s < t.
(iv)
Proof. For s < t,
E[I
2
(t)

t
0

2
u
du (I
2
(s)

s
0

2
u
du)[T
s
]
= E[I
2
(t) I
2
(s)

t
s

2
u
du[T
s
]
= E[(I(t) I(s))
2
+ 2I(t)I(s) 2I
2
(s)[T
s
]

t
s

2
u
du
= E[(I(t) I(s))
2
] + 2I(s)E[I(t) I(s)[T
s
]

t
s

2
u
du
=

t
s

2
u
du + 0

t
s

2
u
du
= 0.
33
4.3.
Proof. I(t) I(s) =
0
(W
t1
W
0
) +
t1
(W
t2
W
t1
)
0
(W
t1
W
0
) =
t1
(W
t2
W
t1
) = W
s
(W
t
W
s
).
(i) I(t) I(s) is not independent of T
s
, since W
s
T
s
.
(ii) E[(I(t) I(s))
4
] = E[W
4
s
]E[(W
t
W
s
)
4
] = 3s 3(t s) = 9s(t s) and 3E[(I(t) I(s))
2
] =
3E[W
2
s
]E[(W
t
W
s
)
2
] = 3s(t s). Since E[(I(t) I(s))
4
] = 3E[(I(t) I(s))
2
], I(t) I(s) is not normally
distributed.
(iii) E[I(t) I(s)[T
s
] = W
s
E[W
t
W
s
[T
s
] = 0.
(iv)
E[I
2
(t)

t
0

2
u
du (I
2
(s)

s
0

2
u
du)[T
s
]
= E[(I(t) I(s))
2
+ 2I(t)I(s) 2I
2
(s)

t
s
W
2
u
du[T
s
]
= E[W
2
s
(W
t
W
s
)
2
+ 2W
s
(W
t
W
s
) W
2
s
(t s)[T
s
]
= W
2
s
E[(W
t
W
s
)
2
] + 2W
s
E[W
t
W
s
[T
s
] W
2
s
(t s)
= W
2
s
(t s) W
2
s
(t s)
= 0.
4.4.
Proof. (Cf. ksendal [3], Exercise 3.9.) We rst note that

j
Bt
j
+t
j+1
2
(B
tj+1
B
tj
)
=

Bt
j
+t
j+1
2
(B
tj+1
Bt
j
+t
j+1
2
) +B
tj
(Bt
j
+t
j+1
2
B
tj
)

j
(Bt
j
+t
j+1
2
B
tj
)
2
.
The rst term converges in L
2
(P) to

T
0
B
t
dB
t
. For the second term, we note
E

Bt
j
+t
j+1
2
B
tj

t
2

= E

Bt
j
+t
j+1
2
B
tj

j
t
j+1
t
j
2

j, k
E

Bt
j
+t
j+1
2
B
tj

t
j+1
t
j
2

Bt
k
+t
k+1
2
B
t
k

t
k+1
t
k
2

j
E

B
2
t
j+1
t
j
2

t
j+1
t
j
2

j
2

t
j+1
t
j
2

T
2
max
1jn
[t
j+1
t
j
[ 0,
34
since E[(B
2
t
t)
2
] = E[B
4
t
2tB
2
t
+t
2
] = 3E[B
2
t
]
2
2t
2
+t
2
= 2t
2
. So

j
Bt
j
+t
j+1
2
(B
tj+1
B
tj
)

T
0
B
t
dB
t
+
T
2
=
1
2
B
2
T
in L
2
(P).
4.5. (i)
Proof.
d lnS
t
=
dS
t
S
t

1
2
d'S`
t
S
2
t
=
2S
t
dS
t
d'S`
t
2S
2
t
=
2S
t
(
t
S
t
dt +
t
S
t
dW
t
)
2
t
S
2
t
dt
2S
2
t
=
t
dW
t
+ (
t

1
2

2
t
)dt.
(ii)
Proof.
lnS
t
= lnS
0
+

t
0

s
dW
s
+

t
0
(
s

1
2

2
s
)ds.
So S
t
= S
0
exp

t
0

s
dW
s
+

t
0
(
s

1
2

2
s
)ds.
4.6.
Proof. Without loss of generality, we assume p = 1. Since (x
p
)

= px
p1
, (x
p
)

= p(p 1)x
p2
, we have
d(S
p
t
) = pS
p1
t
dS
t
+
1
2
p(p 1)S
p2
t
d'S`
t
= pS
p1
t
(S
t
dt +S
t
dW
t
) +
1
2
p(p 1)S
p2
t

2
S
2
t
dt
= S
p
t
[pdt +pdW
t
+
1
2
p(p 1)
2
dt]
= S
p
t
p[dW
t
+ ( +
p 1
2

2
)dt].
4.7. (i)
Proof. dW
4
t
= 4W
3
t
dW
t
+
1
2
4 3W
2
t
d'W`
t
= 4W
3
t
dW
t
+ 6W
2
t
dt. So W
4
T
= 4

T
0
W
3
t
dW
t
+ 6

T
0
W
2
t
dt.
(ii)
Proof. E[W
4
T
] = 6

T
0
tdt = 3T
2
.
(iii)
Proof. dW
6
t
= 6W
5
t
dW
t
+
1
2
6 5W
4
t
dt. So W
6
T
= 6

T
0
W
5
t
dW
t
+15

T
0
W
4
t
dt. Hence E[W
6
T
] = 15

T
0
3t
2
dt =
15T
3
.
4.8.
35
Proof. d(e
t
R
t
) = e
t
R
t
dt +e
t
dR
t
= e
t(dt+dWt)
. Hence
e
t
R
t
= R
0
+

t
0
e
s
(ds +dW
s
) = R
0
+

(e
t
1) +

t
0
e
s
dW
s
,
and R
t
= R
0
e
t
+

(1 e
t
) +

t
0
e
(ts)
dW
s
.
4.9. (i)
Proof.
Ke
r(Tt)
N

(d

) = Ke
r(Tt)
e

d
2

2
= Ke
r(Tt)
e

(d
+

Tt)
2
2

2
= Ke
r(Tt)
e

Ttd+
e

2
(Tt)
2
N

(d
+
)
= Ke
r(Tt)
x
K
e
(r+

2
2
)(Tt)
e

2
(Tt)
2
N

(d
+
)
= xN

(d
+
).
(ii)
Proof.
c
x
= N(d
+
) +xN

(d
+
)

x
d
+
(T t, x) Ke
r(Tt)
N

(d

)

x
d

(T t, x)
= N(d
+
) +xN

(d
+
)

x
d

+
(T t, x) xN

(d
+
)

x
d
+
(T t, x)
= N(d
+
).
(iii)
Proof.
c
t
= xN

(d
+
)

x
d
+
(T t, x) rKe
r(Tt)
N(d

) Ke
r(Tt)
N

(d

)

t
d

(T t, x)
= xN

(d
+
)

t
d
+
(T t, x) rKe
r(Tt)
N(d

) xN

(d
+
)


t
d
+
(T t, x) +

2

T t

= rKe
r(Tt)
N(d

)
x
2

T t
N

(d
+
).
(iv)
36
Proof.
c
t
+rxc
x
+
1
2

2
x
2
c
xx
= rKe
r(Tt)
N(d

)
x
2

T t
N

(d
+
) +rxN(d
+
) +
1
2

2
x
2
N

(d
+
)

x
d
+
(T t, x)
= rc
x
2

T t
N

(d
+
) +
1
2

2
x
2
N

(d
+
)
1

T tx
= rc.
(v)
Proof. For x > K, d
+
(T t, x) > 0 and lim
tT
d
+
(T t, x) = lim
0
d
+
(, x) = . lim
tT
d

(T t, x) =
lim
0
d

(, x) = lim
0

ln
x
K
+
1

(r +
1
2

2
)

= . Similarly, lim
tT
d

= for x
(0, K). Also its clear that lim
tT
d

= 0 for x = K. So
lim
tT
c(t, x) = xN(lim
tT
d
+
) KN(lim
tT
d

) =

x K, if x > K
0, if x K
= (x K)
+
.
(vi)
Proof. It is easy to see lim
x0
d

= . So for t [0, T], lim


x0
c(t, x) = lim
x0
xN(lim
x
d
+
(T t, x))
Ke
r(Tt)
N(lim
x0
d

(T t, x)) = 0.
(vii)
Proof. For t [0, T], it is clear lim
x
d

= . Note
lim
x
x(N(d
+
) 1) = lim
x
N

(d
+
)

x
d
+
x
2
= lim
x
N

(d
+
)
1

Tt
x
1
.
By the expression of d
+
, we get x = K exp

T td
+
(T t)(r +
1
2

2
). So we have
lim
x
x(N(d
+
) 1) = lim
x
N

(d
+
)
x

T t
= lim
d+
e

d
2
+
2

2
Ke

Ttd+(Tt)(r+
1
2

2
)

T t
= 0.
Therefore
lim
x
[c(t, x) (x e
r(Tt)
K)]
= lim
x
[xN(d
+
) Ke
r(Tt)N(d)
x +Ke
r(Tt)
]
= lim
x
[x(N(d
+
) 1) +Ke
r(Tt)
(1 N(d

))]
= lim
x
x(N(d
+
) 1) +Ke
r(Tt)
(1 N( lim
x
d

))
= 0.
4.10. (i)
37
Proof. We show (4.10.16) + (4.10.9) (4.10.16) + (4.10.15), i.e. assuming X has the representation
X
t
=
t
S
t
+
t
M
t
, continuous-time self-nancing condition has two equivalent formulations, (4.10.9) or
(4.10.15). Indeed, dX
t
=
t
dS
t
+
t
dM
t
+(S
t
d
t
+dS
t
d
t
+M
t
d
t
+dM
t
d
t
). So dX
t
=
t
dS
t
+
t
dM
t

S
t
d
t
+dS
t
d
t
+M
t
d
t
+dM
t
d
t
= 0, i.e. (4.10.9) (4.10.15).
(ii)
Proof. First, we clarify the problems by stating explicitly the given conditions and the result to be proved.
We assume we have a portfolio X
t
=
t
S
t
+
t
M
t
. We let c(t, S
t
) denote the price of call option at time t
and set
t
= c
x
(t, S
t
). Finally, we assume the portfolio is self-nancing. The problem is to show
rN
t
dt =

c
t
(t, S
t
) +
1
2

2
S
2
t
c
xx
(t, S
t
)

dt,
where N
t
= c(t, S
t
)
t
S
t
.
Indeed, by the self-nancing property and
t
= c
x
(t, S
t
), we have c(t, S
t
) = X
t
(by the calculations in
Subsection 4.5.1-4.5.3). This uniquely determines
t
as

t
=
X
t

t
S
t
M
t
=
c(t, S
t
) c
x
(t, S
t
)S
t
M
t
=
N
t
M
t
.
Moreover,
dN
t
=

c
t
(t, S
t
)dt +c
x
(t, S
t
)dS
t
+
1
2
c
xx
(t, S
t
)d'S
t
`
t

d(
t
S
t
)
=

c
t
(t, S
t
) +
1
2
c
xx
(t, S
t
)
2
S
2
t

dt + [c
x
(t, S
t
)dS
t
d(X
t

t
M
t
)]
=

c
t
(t, S
t
) +
1
2
c
xx
(t, S
t
)
2
S
2
t

dt +M
t
d
t
+dM
t
d
t
+ [c
x
(t, S
t
)dS
t
+
t
dM
t
dX
t
].
By self-nancing property, c
x
(t, S
t
)dt +
t
dM
t
=
t
dS
t
+
t
dM
t
= dX
t
, so

c
t
(t, S
t
) +
1
2
c
xx
(t, S
t
)
2
S
2
t

dt = dN
t
M
t
d
t
dM
t
d
t
=
t
dM
t
=
t
rM
t
dt = rN
t
dt.
4.11.
Proof. First, we note c(t, x) solves the Black-Scholes-Merton PDE with volatility
1
:


t
+rx

x
+
1
2
x
2

2
1

2
x
2
r

c(t, x) = 0.
So
c
t
(t, S
t
) +rS
t
c
x
(t, S
t
) +
1
2

2
1
S
2
t
c
xx
(t, S
t
) rc(t, S
t
) = 0,
and
dc(t, S
t
) = c
t
(t, S
t
)dt +c
x
(t, S
t
)(S
t
dt +
2
S
t
dW
t
) +
1
2
c
xx
(t, S
t
)
2
2
S
2
t
dt
=

c
t
(t, S
t
) +c
x
(t, S
t
)S
t
+
1
2

2
2
S
2
t
c
xx
(t, S
t
)

dt +
2
S
t
c
x
(t, S
t
)dW
t
=

rc(t, S
t
) + ( r)c
x
(t, S
t
)S
t
+
1
2
S
2
t
(
2
2

2
1
)c
xx
(t, S
t
)

dt +
2
S
t
c
x
(t, S
t
)dW
t
.
38
Therefore
dX
t
=

rc(t, S
t
) + ( r)c
x
(t, S
t
)S
t
+
1
2
S
2
t
(
2
2

2
1
)
xx
(t, S
t
) +rX
t
rc(t, S
t
) +rS
t
c
x
(t, S
t
)

1
2
(
2
2

2
1
)S
2
t
c
xx
(t, S
t
) c
x
(t, S
t
)S
t

dt + [
2
S
t
c
x
(t, S
t
) c
x
(t, S
t
)
2
S
t
]dW
t
= rX
t
dt.
This implies X
t
= X
0
e
rt
. By X
0
, we conclude X
t
= 0 for all t [0, T].
4.12. (i)
Proof. By (4.5.29), c(t, x) p(t, x) = x e
r(Tt)
K. So p
x
(t, x) = c
x
(t, x) 1 = N(d
+
(T t, x)) 1,
p
xx
(t, x) = c
xx
(t, x) =
1
x

Tt
N

(d
+
(T t, x)) and
p
t
(t, x) = c
t
(t, x) +re
r(Tt)
K
= rKe
r(Tt)
N(d

(T t, x))
x
2

T t
N

(d
+
(T t, x)) +rKe
r(Tt)
= rKe
r(Tt)
N(d

(T t, x))
x
2

T t
N

(d
+
(T t, x)).
(ii)
Proof. For an agent hedging a short position in the put, since
t
= p
x
(t, x) < 0, he should short the
underlying stock and put p(t, S
t
) p
x
(t, S
t
)S
t
(> 0) cash in the money market account.
(iii)
Proof. By the put-call parity, it suces to show f(t, x) = x Ke
r(Tt)
satises the Black-Scholes-Merton
partial dierential equation. Indeed,


t
+
1
2

2
x
2

2
x
2
+rx

x
r

f(t, x) = rKe
r(Tt)
+
1
2

2
x
2
0 +rx 1 r(x Ke
r(Tt)
) = 0.
Remark: The Black-Scholes-Merton PDE has many solutions. Proper boundary conditions are the key
to uniqueness. For more details, see Wilmott [8].
4.13.
Proof. We suppose (W
1
, W
2
) is a pair of local martingale dened by SDE

dW
1
(t) = dB
1
(t)
dW
2
(t) = (t)dB
1
(t) +(t)dB
2
(t).
(1)
We want to nd (t) and (t) such that

(dW
2
(t))
2
= [
2
(t) +
2
(t) + 2(t)(t)(t)]dt = dt
dW
1
(t)dW
2
(t) = [(t) +(t)(t)]dt = 0.
(2)
Solve the equation for (t) and (t), we have (t) =
1

1
2
(t)
and (t) =
(t)

1
2
(t)
. So

W
1
(t) = B
1
(t)
W
2
(t) =

t
0
(s)

1
2
(s)
dB
1
(s) +

t
0
1

1
2
(s)
dB
2
(s)
(3)
39
is a pair of independent BMs. Equivalently, we have

B
1
(t) = W
1
(t)
B
2
(t) =

t
0
(s)dW
1
(s) +

t
0

1
2
(s)dW
2
(s).
(4)
4.14. (i)
Proof. Clearly Z
j
T
tj+1
. Moreover
E[Z
j
[T
tj
] = f

(W
tj
)E[(W
tj+1
W
tj
)
2
(t
j+1
t
j
)[T
tj
] = f

(W
tj
)(E[W
2
tj+1tj
] (t
j+1
t
j
)) = 0,
since W
tj+1
W
tj
is independent of T
tj
and W
t
N(0, t). Finally, we have
E[Z
2
j
[T
tj
] = [f

(W
tj
)]
2
E[(W
tj+1
W
tj
)
4
2(t
j+1
t
j
)(W
tj+1
W
tj
)
2
+ (t
j+1
t
j
)
2
[T
tj
]
= [f

(W
tj
)]
2
(E[W
4
tj+1tj
] 2(t
j+1
t
j
)E[W
2
tj+1tj
] + (t
j+1
t
j
)
2
)
= [f

(W
tj
)]
2
[3(t
j+1
t
j
)
2
2(t
j+1
t
j
)
2
+ (t
j+1
t
j
)
2
]
= 2[f

(W
tj
)]
2
(t
j+1
t
j
)
2
,
where we used the independence of Browian motion increment and the fact that E[X
4
] = 3E[X
2
]
2
if X is
Gaussian with mean 0.
(ii)
Proof. E[

n1
j=0
Z
j
] = E[

n1
j=0
E[Z
j
[T
tj
]] = 0 by part (i).
(iii)
Proof.
V ar[
n1

j=0
Z
j
] = E[(
n1

j=0
Z
j
)
2
]
= E[
n1

j=0
Z
2
j
+ 2

0i<jn1
Z
i
Z
j
]
=
n1

j=0
E[E[Z
2
j
[T
tj
]] + 2

0i<jn1
E[Z
i
E[Z
j
[T
tj
]]
=
n1

j=0
E[2[f

(W
tj
)]
2
(t
j+1
t
j
)
2
]
=
n1

j=0
2E[(f

(W
tj
))
2
](t
j+1
t
j
)
2
2 max
0jn1
[t
j+1
t
j
[
n1

j=0
E[(f

(W
tj
))
2
](t
j+1
t
j
)
0,
since

n1
j=0
E[(f

(W
tj
))
2
](t
j+1
t
j
)

T
0
E[(f

(W
t
))
2
]dt < .
4.15. (i)
40
Proof. B
i
is a local martingale with
(dB
i
(t))
2
=

j=1

ij
(t)

i
(t)
dW
j
(t)

2
=
d

j=1

2
ij
(t)

2
i
(t)
dt = dt.
So B
i
is a Brownian motion.
(ii)
Proof.
dB
i
(t)dB
k
(t) =

j=1

ij
(t)

i
(t)
dW
j
(t)

l=1

kl
(t)

k
(t)
dW
l
(t)

1j, ld

ij
(t)
kl
(t)

i
(t)
k
(t)
dW
j
(t)dW
l
(t)
=
d

j=1

ij
(t)
kj
(t)

i
(t)
k
(t)
dt
=
ik
(t)dt.
4.16.
Proof. To nd the m independent Brownion motion W
1
(t), , W
m
(t), we need to nd A(t) = (a
ij
(t)) so
that
(dB
1
(t), , dB
m
(t))
tr
= A(t)(dW
1
(t), , dW
m
(t))
tr
,
or equivalently
(dW
1
(t), , dW
m
(t))
tr
= A(t)
1
(dB
1
(t), , dB
m
(t))
tr
,
and
(dW
1
(t), , dW
m
(t))
tr
(dW
1
(t), , dW
m
(t))
= A(t)
1
(dB
1
(t), , dB
m
(t))
tr
(dB
1
(t), , dB
m
(t))(A(t)
1
)
tr
dt
= I
mm
dt,
where I
mm
is the mm unit matrix. By the condition dB
i
(t)dB
k
(t) =
ik
(t)dt, we get
(dB
1
(t), , dB
m
(t))
tr
(dB
1
(t), , dB
m
(t)) = C(t).
So A(t)
1
C(t)(A(t)
1
)
tr
= I
mm
, which gives C(t) = A(t)A(t)
tr
. This motivates us to dene A as the
square root of C. Reverse the above analysis, we obtain a formal proof.
4.17.
Proof. We will try to solve all the sub-problems in a single, long solution. We start with the general X
i
:
X
i
(t) = X
i
(0) +

t
0

i
(u)du +

t
0

i
(u)dB
i
(u), i = 1, 2.
41
The goal is to show
lim
0
C()

V
1
()V
2
()
= (t
0
).
First, for i = 1, 2, we have
M
i
() = E[X
i
(t
0
+) X
i
(t
0
)[T
t0
]
= E

t0+
t0

i
(u)du +

t0+
t0

i
(u)dB
i
(u)[T
t0

=
i
(t
0
) +E

t0+
t0
(
i
(u)
i
(t
0
))du[T
t0

.
By Conditional Jensens Inequality,

t0+
t0
(
i
(u)
i
(t
0
))du[T
t0

t0+
t0
[
i
(u)
i
(t
0
)[du[T
t0

Since
1

t0+
t0
[
i
(u)
i
(t
0
)[du 2M and lim
0
1

t0+
t0
[
i
(u)
i
(t
0
)[du = 0 by the continuity of
i
,
the Dominated Convergence Theorem under Conditional Expectation implies
lim
0
1

t0+
t0
[
i
(u)
i
(t
0
)[du[T
t0

= E

lim
0
1

t0+
t0
[
i
(u)
i
(t
0
)[du[T
t0

= 0.
So M
i
() =
i
(t
0
) +o(). This proves (iii).
To calculate the variance and covariance, we note Y
i
(t) =

t
0

i
(u)dB
i
(u) is a martingale and by Itos
formula Y
i
(t)Y
j
(t)

t
0

i
(u)
j
(u)du is a martingale (i = 1, 2). So
E[(X
i
(t
0
+) X
i
(t
0
))(X
j
(t
0
+) X
j
(t
0
))[T
t0
]
= E

Y
i
(t
0
+) Y
i
(t
0
) +

t0+
t0

i
(u)du

Y
j
(t
0
+) Y
j
(t
0
) +

t0+
t0

j
(u)du

[T
t0

= E [(Y
i
(t
0
+) Y
i
(t
0
)) (Y
j
(t
0
+) Y
j
(t
0
)) [T
t0
] +E

t0+
t0

i
(u)du

t0+
t0

j
(u)du[T
t0

+E

(Y
i
(t
0
+) Y
i
(t
0
))

t0+
t0

j
(u)du[T
t0

+E

(Y
j
(t
0
+) Y
j
(t
0
))

t0+
t0

i
(u)du[T
t0

= I +II +III +IV.


I = E[Y
i
(t
0
+)Y
j
(t
0
+) Y
i
(t
0
)Y
j
(t
0
)[T
t0
] = E

t0+
t0

i
(u)
j
(u)
ij
(t)dt[T
t0

.
By an argument similar to that involved in the proof of part (iii), we conclude I =
i
(t
0
)
j
(t
0
)
ij
(t
0
) +o()
and
II = E

t0+
t0
(
i
(u)
i
(t
0
))du

t0+
t0

j
(u)du[T
t0

+
i
(t
0
)E

t0+
t0

j
(u)du[T
t0

= o() + (M
i
() o())M
j
()
= M
i
()M
j
() +o().
42
By Cauchys inequality under conditional expectation (note E[XY [T] denes an inner product on L
2
()),
III E

[Y
i
(t
0
+) Y
i
(t
0
)[

t0+
t0
[
j
(u)[du[T
t0

E[(Y
i
(t
0
+) Y
i
(t
0
))
2
[T
t0
]
M

E[Y
i
(t
0
+)
2
Y
i
(t
0
)
2
[T
t0
]
M

E[

t0+
t0

i
(u)
2
du[T
t0
]
M M

= o()
Similarly, IV = o(). In summary, we have
E[(X
i
(t
0
+) X
i
(t
0
))(X
j
(t
0
+) X
j
(t
0
))[T
t0
] = M
i
()M
j
() +
i
(t
0
)
j
(t
0
)
ij
(t
0
) +o() +o().
This proves part (iv) and (v). Finally,
lim
0
C()

V
1
()V
2
()
= lim
0
(t
0
)
1
(t
0
)
2
(t
0
) +o()

(
2
1
(t
0
) +o())(
2
2
(t
0
) +o())
= (t
0
).
This proves part (vi). Part (i) and (ii) are consequences of general cases.
4.18. (i)
Proof.
d(e
rt

t
) = (de
Wt
1
2

2
t
) = e
Wt
1
2

2
t
dW
t
= (e
rt

t
)dW
t
,
where for the second =, we used the fact that e
Wt
1
2

2
t
solves dX
t
= X
t
dW
t
. Since d(e
rt

t
) =
re
rt

t
dt +e
rt
d
t
, we get d
t
=
t
dW
t
r
t
dt.
(ii)
Proof.
d(
t
X
t
) =
t
dX
t
+X
t
d
t
+dX
t
d
t
=
t
(rX
t
dt +
t
( r)S
t
dt +
t
S
t
dW
t
) +X
t
(
t
dW
t
r
t
dt)
+(rX
t
dt +
t
( r)S
t
dt +
t
S
t
dW
t
)(
t
dW
t
r
t
dt)
=
t
(
t
( r)S
t
dt +
t
S
t
dW
t
) X
t

t
dW
t

t
S
t

t
dt
=
t

t
S
t
dW
t
X
t

t
dW
t
.
So
t
X
t
is a martingale.
(iii)
Proof. By part (ii), X
0
=
0
X
0
= E[
T
X
t
] = E[
T
V
T
]. (This can be seen as a version of risk-neutral pricing,
only that the pricing is carried out under the actual probability measure.)
4.19. (i)
Proof. B
t
is a local martingale with [B]
t
=

t
0
sign(W
s
)
2
ds = t. So by Levys theorem, B
t
is a Brownian
motion.
43
(ii)
Proof. d(B
t
W
t
) = B
t
dW
t
+ sign(W
t
)W
t
dW
t
+ sign(W
t
)dt. Integrate both sides of the resulting equation
and the expectation, we get
E[B
t
W
t
] =

t
0
E[sign(W
s
)]ds =

t
0
E[1
{Ws0}
1
{Ws<0}
]ds =
1
2
t
1
2
t = 0.
(iii)
Proof. By It os formula, dW
2
t
= 2W
t
dW
t
+dt.
(iv)
Proof. By It os formula,
d(B
t
W
2
t
) = B
t
dW
2
t
+W
2
t
dB
t
+dB
t
dW
2
t
= B
t
(2W
t
dW
t
+dt) +W
2
t
sign(W
t
)dW
t
+sign(W
t
)dW
t
(2W
t
dW
t
+dt)
= 2B
t
W
t
dW
t
+B
t
dt +sign(W
t
)W
2
t
dW
t
+ 2sign(W
t
)W
t
dt.
So
E[B
t
W
2
t
] = E[

t
0
B
s
ds] + 2E[

t
0
sign(W
s
)W
s
ds]
=

t
0
E[B
s
]ds + 2

t
0
E[sign(W
s
)W
s
]ds
= 2

t
0
(E[W
s
1
{Ws0}
] E[W
s
1
{Ws<0}
])ds
= 4

t
0


0
x
e

x
2
2s

2s
dxds
= 4

t
0

s
2
ds
= 0 = E[B
t
] E[W
2
t
].
Since E[B
t
W
2
t
] = E[B
t
] E[W
2
t
], B
t
and W
t
are not independent.
4.20. (i)
Proof. f(x) =

x K, if x K
0, if x < K.
So f

(x) =

1, if x > K
undened, if x = K
0, if x < K
and f

(x) =

0, if x = K
undened, if x = K.
(ii)
Proof. E[f(W
T
)] =

K
(x K)
e

x
2
2T

2T
dx =

T
2
e

K
2
2T
K(
K

T
) where is the distribution function of
standard normal random variable. If we suppose

T
0
f

(W
t
)dt = 0, the expectation of RHS of (4.10.42) is
equal to 0. So (4.10.42) cannot hold.
(iii)
44
Proof. This is trivial to check.
(iv)
Proof. If x = K, lim
n
f
n
(x) =
1
8n
= 0; if x > K, for n large enough, x K +
1
2n
, so lim
n
f
n
(x) =
lim
n
(x K) = x K; if x < K, for n large enough, x K
1
2n
, so lim
n
f
n
(x) = lim
n
0 = 0. In
summary, lim
n
f
n
(x) = (x K)
+
. Similarly, we can show
lim
n
f

n
(x) =

0, if x < K
1
2
, if x = K
1, if x > K.
(5)
(v)
Proof. Fix , so that W
t
() < K for any t [0, T]. Since W
t
() can obtain its maximum on [0, T], there
exists n
0
, so that for any n n
0
, max
0tT
W
t
() < K
1
2n
. So
L
K
(T)() = lim
n
n

T
0
1
(K
1
2n
,K+
1
2n
)
(W
t
())dt = 0.
(vi)
Proof. Take expectation on both sides of the formula (4.10.45), we have
E[L
K
(T)] = E[(W
T
K)
+
] > 0.
So we cannot have L
K
(T) = 0 a.s..
4.21. (i)
Proof. There are two problems. First, the transaction cost could be big due to active trading; second, the
purchases and sales cannot be made at exactly the same price K. For more details, see Hull [2].
(ii)
Proof. No. The RHS of (4.10.26) is a martingale, so its expectation is 0. But E[(S
T
K)
+
] > 0. So
X
T
= (S
T
K)
+
.
5. Risk-Neutral Pricing
5.1. (i)
Proof.
df(X
t
) = f

(X
t
)dt +
1
2
f

(X
t
)d'X`
t
= f(X
t
)(dX
t
+
1
2
d'X`
t
)
= f(X
t
)

t
dW
t
+ (
t
R
t

1
2

2
t
)dt +
1
2

2
t
dt

= f(X
t
)(
t
R
t
)dt +f(X
t
)
t
dW
t
.
This is formula (5.2.20).
45
(ii)
Proof. d(D
t
S
t
) = S
t
dD
t
+ D
t
dS
t
+ dD
t
dS
t
= S
t
R
t
D
t
dt + D
t

t
S
t
dt + D
t

t
S
t
dW
t
= D
t
S
t
(
t
R
t
)dt +
D
t
S
t

t
dW
t
. This is formula (5.2.20).
5.2.
Proof. By Lemma 5.2.2.,

E[D
T
V
T
[T
t
] = E

D
T
V
T
Z
T
Zt
[T
t

. Therefore (5.2.30) is equivalent to D


t
V
t
Z
t
=
E[D
T
V
T
Z
T
[T
t
].
5.3. (i)
Proof.
c
x
(0, x) =
d
dx

E[e
rT
(xe

W
T
+(r
1
2

2
)T
K)
+
]
=

E

e
rT
d
dx
h(xe

W
T
+(r
1
2

2
)T
)

=

E

e
rT
e

W
T
+(r
1
2

2
)T
1
{xe

W
T
+(r
1
2

2
)T
>K}

= e

1
2

2
T

E

W
T
1
{

W
T
>
1

(ln
K
x
(r
1
2

2
)T)}

= e

1
2

2
T

E

W
T

T
1
{

W
T

T>
1

T
(ln
K
x
(r
1
2

2
)T)

T}

= e

1
2

2
T

2
e

z
2
2
e

Tz
1
{z

T>d+(T,x)}
dz
=

2
e

(z

T)
2
2
1
{z

T>d+(T,x)}
dz
= N(d
+
(T, x)).
(ii)
Proof. If we set

Z
T
= e

W
T

1
2

2
T
and

Z
t
=

E[

Z
T
[T
t
], then

Z is a

P-martingale,

Z
t
> 0 and E[

Z
T
] =

E[e

W
T

1
2

2
T
] = 1. So if we dene

P by d

P = Z
T
d

P on T
T
, then

P is a probability measure equivalent to

P, and
c
x
(0, x) =

E[

Z
T
1
{S
T
>K}
] =

P(S
T
> K).
Moreover, by Girsanovs Theorem,

W
t
=

W
t
+

t
0
()du =

W
t
t is a

P-Brownian motion (set =
in Theorem 5.4.1.)
(iii)
Proof. S
T
= xe

W
T
+(r
1
2

2
)T
= xe

W
T
+(r+
1
2

2
)T
. So

P(S
T
> K) =

P(xe

W
T
+(r+
1
2

2
)T
> K) =

P

W
T

T
> d
+
(T, x)

= N(d
+
(T, x)).
46
5.4. First, a few typos. In the SDE for S, (t)d

W(t) (t)S(t)d

W(t). In the rst equation for


c(0, S(0)), E

E. In the second equation for c(0, S(0)), the variable for BSM should be
BSM

T, S(0); K,
1
T

T
0
r(t)dt,

1
T

T
0

2
(t)dt

.
(i)
Proof. d lnS
t
=
dSt
St

1
2S
2
t
d'S`
t
= r
t
dt +
t
d

W
t

1
2

2
t
dt. So S
T
= S
0
exp

T
0
(r
t

1
2

2
t
)dt +

T
0

t
d

W
t
. Let
X =

T
0
(r
t

1
2

2
t
)dt +

T
0

t
d

W
t
. The rst term in the expression of X is a number and the second term
is a Gaussian random variable N(0,

T
0

2
t
dt), since both r and ar deterministic. Therefore, S
T
= S
0
e
X
,
with X N(

T
0
(r
t

1
2

2
t
)dt,

T
0

2
t
dt),.
(ii)
Proof. For the standard BSM model with constant volatility and interest rate R, under the risk-neutral
measure, we have S
T
= S
0
e
Y
, where Y = (R
1
2

2
)T +

W
T
N((R
1
2

2
)T,
2
T), and

E[(S
0
e
Y
K)
+
] =
e
RT
BSM(T, S
0
; K, R, ). Note R =
1
T
(E[Y ] +
1
2
V ar(Y )) and =

1
T
V ar(Y ), we can get

E[(S
0
e
Y
K)
+
] = e
E[Y ]+
1
2
V ar(Y )
BSM

T, S
0
; K,
1
T

E[Y ] +
1
2
V ar(Y )

1
T
V ar(Y )

.
So for the model in this problem,
c(0, S
0
) = e

T
0
rtdt

E[(S
0
e
X
K)
+
]
= e

T
0
rtdt
e
E[X]+
1
2
V ar(X)
BSM

T, S
0
; K,
1
T

E[X] +
1
2
V ar(X)

1
T
V ar(X)

= BSM

T, S
0
; K,
1
T

T
0
r
t
dt,

1
T

T
0

2
t
dt

.
5.5. (i)
Proof. Let f(x) =
1
x
, then f

(x) =
1
x
2
and f

(x) =
2
x
3
. Note dZ
t
= Z
t

t
dW
t
, so
d

1
Z
t

= f

(Z
t
)dZ
t
+
1
2
f

(Z
t
)dZ
t
dZ
t
=
1
Z
2
t
(Z
t
)
t
dW
t
+
1
2
2
Z
3
t
Z
2
t

2
t
dt =

t
Z
t
dW
t
+

2
t
Z
t
dt.
(ii)
Proof. By Lemma 5.2.2., for s, t 0 with s < t,

M
s
=

E[

M
t
[T
s
] = E

Zt

Mt
Zs
[T
s

. That is, E[Z


t

M
t
[T
s
] =
Z
s

M
s
. So M = Z

M is a P-martingale.
(iii)
47
Proof.
d

M
t
= d

M
t

1
Z
t

=
1
Z
t
dM
t
+M
t
d
1
Z
t
+dM
t
d
1
Z
t
=

t
Z
t
dW
t
+
M
t

t
Z
t
dW
t
+
M
t

2
t
Z
t
dt +

t

t
Z
t
dt.
(iv)
Proof. In part (iii), we have
d

M
t
=

t
Z
t
dW
t
+
M
t

t
Z
t
dW
t
+
M
t

2
t
Z
t
dt +

t

t
Z
t
dt =

t
Z
t
(dW
t
+
t
dt) +
M
t

t
Z
t
(dW
t
+
t
dt).
Let

t
=
t+Mtt
Zt
, then d

M
t
=

t
d

W
t
. This proves Corollary 5.3.2.
5.6.
Proof. By Theorem 4.6.5, it suces to show

W
i
(t) is an T
t
-martingale under

P and [

W
i
,

W
j
](t) = t
ij
(i, j = 1, 2). Indeed, for i = 1, 2,

W
i
(t) is an T
t
-martingale under

P if and only if

W
i
(t)Z
t
is an T
t
-martingale
under P, since

E[

W
i
(t)[T
s
] = E

W
i
(t)Z
t
Z
s
[T
s

.
By Itos product formula, we have
d(

W
i
(t)Z
t
) =

W
i
(t)dZ
t
+Z
t
d

W
i
(t) +dZ
t
d

W
i
(t)
=

W
i
(t)(Z
t
)(t) dW
t
+Z
t
(dW
i
(t) +
i
(t)dt) + (Z
t

t
dW
t
)(dW
i
(t) +
i
(t)dt)
=

W
i
(t)(Z
t
)
d

j=1

j
(t)dW
j
(t) +Z
t
(dW
i
(t) +
i
(t)dt) Z
t

i
(t)dt
=

W
i
(t)(Z
t
)
d

j=1

j
(t)dW
j
(t) +Z
t
dW
i
(t)
This shows

W
i
(t)Z
t
is an T
t
-martingale under P. So

W
i
(t) is an T
t
-martingale under

P. Moreover,
[

W
i
,

W
j
](t) =

W
i
+

i
(s)ds, W
j
+

j
(s)ds

(t) = [W
i
, W
j
](t) = t
ij
.
Combined, this proves the two-dimensional Girsanovs Theorem.
5.7. (i)
Proof. Let a be any strictly positive number. We dene X
2
(t) = (a +X
1
(t))D(t)
1
. Then
P

X
2
(T)
X
2
(0)
D(T)

= P(a +X
1
(T) a) = P(X
1
(T) 0) = 1,
and P

X
2
(T) >
X2(0)
D(T)

= P(X
1
(T) > 0) > 0, since a is arbitrary, we have proved the claim of this problem.
Remark: The intuition is that we invest the positive starting fund a into the money market account,
and construct portfolio X
1
from zero cost. Their sum should be able to beat the return of money market
account.
(ii)
48
Proof. We dene X
1
(t) = X
2
(t)D(t) X
2
(0). Then X
1
(0) = 0,
P(X
1
(T) 0) = P

X
2
(T)
X
2
(0)
D(T)

= 1, P(X
1
(T) > 0) = P

X
2
(T) >
X
2
(0)
D(T)

> 0.
5.8. The basic idea is that for any positive

P-martingale M, dM
t
= M
t

1
Mt
dM
t
. By Martingale Repre-
sentation Theorem, dM
t
=

t
d

W
t
for some adapted process

t
. So dM
t
= M
t
(

t
Mt
)d

W
t
, i.e. any positive
martingale must be the exponential of an integral w.r.t. Brownian motion. Taking into account discounting
factor and apply Itos product rule, we can show every strictly positive asset is a generalized geometric
Brownian motion.
(i)
Proof. V
t
D
t
=

E[e

T
0
Rudu
V
T
[T
t
] =

E[D
T
V
T
[T
t
]. So (D
t
V
t
)
t0
is a

P-martingale. By Martingale Represen-
tation Theorem, there exists an adapted process

t
, 0 t T, such that D
t
V
t
=

t
0

s
d

W
s
, or equivalently,
V
t
= D
1
t

t
0

s
d

W
s
. Dierentiate both sides of the equation, we get dV
t
= R
t
D
1
t

t
0

s
d

W
s
dt +D
1
t

t
d

W
t
,
i.e. dV
t
= R
t
V
t
dt +

t
Dt
dW
t
.
(ii)
Proof. We prove the following more general lemma.
Lemma 1. Let X be an almost surely positive random variable (i.e. X > 0 a.s.) dened on the probability
space (, (, P). Let T be a sub -algebra of (, then Y = E[X[T] > 0 a.s.
Proof. By the property of conditional expectation Y
t
0 a.s. Let A = Y = 0, we shall show P(A) = 0. In-
deed, note A T, 0 = E[Y I
A
] = E[E[X[T]I
A
] = E[XI
A
] = E[X1
A{X1}
] +

n=1
E[X1
A{
1
n
>X
1
n+1
}
]
P(AX 1)+

n=1
1
n+1
P(A
1
n
> X
1
n+1
). So P(AX 1) = 0 and P(A
1
n
> X
1
n+1
) = 0,
n 1. This in turn implies P(A) = P(AX > 0) = P(AX 1) +

n=1
P(A
1
n
> X
1
n+1
) =
0.
By the above lemma, it is clear that for each t [0, T], V
t
=

E[e

T
t
Rudu
V
T
[T
t
] > 0 a.s.. Moreover,
by a classical result of martingale theory (Revuz and Yor [4], Chapter II, Proposition (3.4)), we have the
following stronger result: for a.s. , V
t
() > 0 for any t [0, T].
(iii)
Proof. By (ii), V > 0 a.s., so dV
t
= V
t
1
Vt
dV
t
= V
t
1
Vt

R
t
V
t
dt +

t
Dt
d

W
t

= V
t
R
t
dt + V
t

t
VtDt
d

W
t
= R
t
V
t
dt +

t
V
t
d

W
t
, where
t
=

t
VtDt
. This shows V follows a generalized geometric Brownian motion.
5.9.
Proof. c(0, T, x, K) = xN(d
+
) Ke
rT
N(d

) with d

=
1

T
(ln
x
K
+ (r
1
2

2
)T). Let f(y) =
1

2
e

y
2
2
,
then f

(y) = yf(y),
c
K
(0, T, x, K) = xf(d
+
)
d
+
y
e
rT
N(d

) Ke
rT
f(d

)
d

y
= xf(d
+
)
1

TK
e
rT
N(d

) +e
rT
f(d

)
1

T
,
49
and
c
KK
(0, T, x, K)
= xf(d
+
)
1

TK
2

TK
f(d
+
)(d
+
)
d
+
y
e
rT
f(d

)
d

y
+
e
rT

T
(d

)f(d

)
d

y
=
x

TK
2
f(d
+
) +
xd
+

TK
f(d
+
)
1
K

T
e
rT
f(d

)
1
K

e
rT
d

T
f(d

)
1
K

T
= f(d
+
)
x
K
2

T
[1
d
+

T
] +
e
rT
f(d

)
K

T
[1 +
d

T
]
=
e
rT
K
2
T
f(d

)d
+

x
K
2

2
T
f(d
+
)d

.
5.10. (i)
Proof. At time t
0
, the value of the chooser option is V (t
0
) = maxC(t
0
), P(t
0
) = maxC(t
0
), C(t
0
)
F(t
0
) = C(t
0
) + max0, F(t
0
) = C(t
0
) + (e
r(Tt0)
K S(t
0
))
+
.
(ii)
Proof. By the risk-neutral pricing formula, V (0) =

E[e
rt0
V (t
0
)] =

E[e
rt0
C(t
0
)+(e
rT
Ke
rt0
S(t
0
)
+
] =
C(0) +

E[e
rt0
(e
r(Tt0)
K S(t
0
))
+
]. The rst term is the value of a call expiring at time T with strike
price K and the second term is the value of a put expiring at time t
0
with strike price e
r(Tt0)
K.
5.11.
Proof. We rst make an analysis which leads to the hint, then we give a formal proof.
(Analysis) If we want to construct a portfolio X that exactly replicates the cash ow, we must nd a
solution to the backward SDE

dX
t
=
t
dS
t
+R
t
(X
t

t
S
t
)dt C
t
dt
X
T
= 0.
Multiply D
t
on both sides of the rst equation and apply Itos product rule, we get d(D
t
X
t
) =
t
d(D
t
S
t
)
C
t
D
t
dt. Integrate from 0 to T, we have D
T
X
T
D
0
X
0
=

T
0

t
d(D
t
S
t
)

T
0
C
t
D
t
dt. By the terminal
condition, we get X
0
= D
1
0
(

T
0
C
t
D
t
dt

T
0

t
d(D
t
S
t
)). X
0
is the theoretical, no-arbitrage price of
the cash ow, provided we can nd a trading strategy that solves the BSDE. Note the SDE for S
gives d(D
t
S
t
) = (D
t
S
t
)
t
(
t
dt + dW
t
), where
t
=
tRt
t
. Take the proper change of measure so that

W
t
=

t
0

s
ds +W
t
is a Brownian motion under the new measure

P, we get

T
0
C
t
D
t
dt = D
0
X
0
+

T
0

t
d(D
t
S
t
) = D
0
X
0
+

T
0

t
(D
t
S
t
)
t
d

W
t
.
This says the random variable

T
0
C
t
D
t
dt has a stochastic integral representation D
0
X
0
+

T
0

t
D
t
S
t

t
d

W
t
.
This inspires us to consider the martingale generated by

T
0
C
t
D
t
dt, so that we can apply Martingale Rep-
resentation Theorem and get a formula for by comparison of the integrands.
50
(Formal proof) Let M
T
=

T
0
C
t
D
t
dt, and M
t
=

E[M
T
[T
t
]. Then by Martingale Representation Theo-
rem, we can nd an adapted process

t
, so that M
t
= M
0
+

t
0

t
d

W
t
. If we set
t
=

t
DtStt
, we can check
X
t
= D
1
t
(D
0
X
0
+

t
0

u
d(D
u
S
u
)

t
0
C
u
D
u
du), with X
0
= M
0
=

E[

T
0
C
t
D
t
dt] solves the SDE

dX
t
=
t
dS
t
+R
t
(X
t

t
S
t
)dt C
t
dt
X
T
= 0.
Indeed, it is easy to see that X satises the rst equation. To check the terminal condition, we note
X
T
D
T
= D
0
X
0
+

T
0

t
D
t
S
t

t
d

W
t

T
0
C
t
D
t
dt = M
0
+

T
0

t
d

W
t
M
T
= 0. So X
T
= 0. Thus, we have
found a trading strategy , so that the corresponding portfolio X replicates the cash ow and has zero
terminal value. So X
0
=

E[

T
0
C
t
D
t
dt] is the no-arbitrage price of the cash ow at time zero.
Remark: As shown in the analysis, d(D
t
X
t
) =
t
d(D
t
S
t
) C
t
D
t
dt. Integrate from t to T, we get
0 D
t
X
t
=

T
t

u
d(D
u
S
u
)

T
t
C
u
D
u
du. Take conditional expectation w.r.t. T
t
on both sides, we get
D
t
X
t
=

E[

T
t
C
u
D
u
du[T
t
]. So X
t
= D
1
t

E[

T
t
C
u
D
u
du[T
t
]. This is the no-arbitrage price of the cash
ow at time t, and we have justied formula (5.6.10) in the textbook.
5.12. (i)
Proof. d

B
i
(t) = dB
i
(t) +
i
(t)dt =

d
j=1
ij(t)
i(t)
dW
j
(t) +

d
j=1
ij(t)
i(t)

j
(t)dt =

d
j=1
ij(t)
i(t)
d

W
j
(t). So B
i
is a
martingale. Since d

B
i
(t)d

B
i
(t) =

d
j=1
ij(t)
2
i(t)
2
dt = dt, by Levys Theorem,

B
i
is a Brownian motion under

P.
(ii)
Proof.
dS
i
(t) = R(t)S
i
(t)dt +
i
(t)S
i
(t)d

B
i
(t) + (
i
(t) R(t))S
i
(t)dt
i
(t)S
i
(t)
i
(t)dt
= R(t)S
i
(t)dt +
i
(t)S
i
(t)d

B
i
(t) +
d

j=1

ij
(t)
j
(t)S
i
(t)dt S
i
(t)
d

j=1

ij
(t)
j
(t)dt
= R(t)S
i
(t)dt +
i
(t)S
i
(t)d

B
i
(t).
(iii)
Proof. d

B
i
(t)d

B
k
(t) = (dB
i
(t) +
i
(t)dt)(dB
j
(t) +
j
(t)dt) = dB
i
(t)dB
j
(t) =
ik
(t)dt.
(iv)
Proof. By It os product rule and martingale property,
E[B
i
(t)B
k
(t)] = E[

t
0
B
i
(s)dB
k
(s)] +E[

t
0
B
k
(s)dB
i
(s)] +E[

t
0
dB
i
(s)dB
k
(s)]
= E[

t
0

ik
(s)ds] =

t
0

ik
(s)ds.
Similarly, by part (iii), we can show

E[

B
i
(t)

B
k
(t)] =

t
0

ik
(s)ds.
(v)
51
Proof. By It os product formula,
E[B
1
(t)B
2
(t)] = E[

t
0
sign(W
1
(u))du] =

t
0
[P(W
1
(u) 0) P(W
1
(u) < 0)]du = 0.
Meanwhile,

E[

B
1
(t)

B
2
(t)] =

E[

t
0
sign(W
1
(u))du
=

t
0
[

P(W
1
(u) 0)

P(W
1
(u) < 0)]du
=

t
0
[

P(

W
1
(u) u)

P(

W
1
(u) < u)]du
=

t
0
2

1
2


P(

W
1
(u) < u)

du
< 0,
for any t > 0. So E[B
1
(t)B
2
(t)] =

E[

B
1
(t)

B
2
(t)] for all t > 0.
5.13. (i)
Proof.

E[W
1
(t)] =

E[

W
1
(t)] = 0 and

E[W
2
(t)] =

E[

W
2
(t)

t
0

W
1
(u)du] = 0, for all t [0, T].
(ii)
Proof.

Cov[W
1
(T), W
2
(T)] =

E[W
1
(T)W
2
(T)]
=

E

T
0
W
1
(t)dW
2
(t) +

T
0
W
2
(t)dW
1
(t)

=

E

T
0

W
1
(t)(d

W
2
(t)

W
1
(t)dt)

+

E

T
0
W
2
(t)d

W
1
(t)

T
0

W
1
(t)
2
dt

T
0
tdt
=
1
2
T
2
.
5.14. Equation (5.9.6) can be transformed into d(e
rt
X
t
) =
t
[d(e
rt
S
t
) ae
rt
dt] =
t
e
rt
[dS
t
rS
t
dt
adt]. So, to make the discounted portfolio value e
rt
X
t
a martingale, we are motivated to change the measure
in such a way that S
t
r

t
0
S
u
duat is a martingale under the new measure. To do this, we note the SDE for S
is dS
t
=
t
S
t
dt+S
t
dW
t
. Hence dS
t
rS
t
dtadt = [(
t
r)S
t
a]dt+S
t
dW
t
= S
t

(tr)Sta
St
dt +dW
t

.
Set
t
=
(tr)Sta
St
and

W
t
=

t
0

s
ds +W
t
, we can nd an equivalent probability measure

P, under which
S satises the SDE dS
t
= rS
t
dt +S
t
d

W
t
+adt and

W
t
is a BM. This is the rational for formula (5.9.7).
This is a good place to pause and think about the meaning of martingale measure. What is to be
a martingale? The new measure

P should be such that the discounted value process of the replicating
52
portfolio is a martingale, not the discounted price process of the underlying. First, we want D
t
X
t
to be a
martingale under

P because we suppose that X is able to replicate the derivative payo at terminal time,
X
T
= V
T
. In order to avoid arbitrage, we must have X
t
= V
t
for any t [0, T]. The diculty is how
to calculate X
t
and the magic is brought by the martingale measure in the following line of reasoning:
V
t
= X
t
= D
1
t

E[D
T
X
T
[T
t
] = D
1
t

E[D
T
V
T
[T
t
]. You can think of martingale measure as a calculational
convenience. That is all about martingale measure! Risk neutral is a just perception, referring to the
actual eect of constructing a hedging portfolio! Second, we note when the portfolio is self-nancing, the
discounted price process of the underlying is a martingale under

P, as in the classical Black-Scholes-Merton
model without dividends or cost of carry. This is not a coincidence. Indeed, we have in this case the
relation d(D
t
X
t
) =
t
d(D
t
S
t
). So D
t
X
t
being a martingale under

P is more or less equivalent to D
t
S
t
being a martingale under

P. However, when the underlying pays dividends, or there is cost of carry,
d(D
t
X
t
) =
t
d(D
t
S
t
) no longer holds, as shown in formula (5.9.6). The portfolio is no longer self-nancing,
but self-nancing with consumption. What we still want to retain is the martingale property of D
t
X
t
, not
that of D
t
S
t
. This is how we choose martingale measure in the above paragraph.
Let V
T
be a payo at time T, then for the martingale M
t
=

E[e
rT
V
T
[T
t
], by Martingale Representation
Theorem, we can nd an adapted process

t
, so that M
t
= M
0
+

t
0

s
d

W
s
. If we let
t
=

te
rt
St
, then the
value of the corresponding portfolio X satises d(e
rt
X
t
) =

t
d

W
t
. So by setting X
0
= M
0
=

E[e
rT
V
T
],
we must have e
rt
X
t
= M
t
, for all t [0, T]. In particular, X
T
= V
T
. Thus the portfolio perfectly hedges
V
T
. This justies the risk-neutral pricing of European-type contingent claims in the model where cost of
carry exists. Also note the risk-neutral measure is dierent from the one in case of no cost of carry.
Another perspective for perfect replication is the following. We need to solve the backward SDE

dX
t
=
t
dS
t
a
t
dt +r(X
t

t
S
t
)dt
X
T
= V
T
for two unknowns, X and . To do so, we nd a probability measure

P, under which e
rt
X
t
is a martingale,
then e
rt
X
t
=

E[e
rT
V
T
[T
t
] := M
t
. Martingale Representation Theorem gives M
t
= M
0
+

t
0

u
d

W
u
for
some adapted process

. This would give us a theoretical representation of by comparison of integrands,
hence a perfect replication of V
T
.
(i)
Proof. As indicated in the above analysis, if we have (5.9.7) under

P, then d(e
rt
X
t
) =
t
[d(e
rt
S
t
)
ae
rt
dt] =
t
e
rt
S
t
d

W
t
. So (e
rt
X
t
)
t0
, where X is given by (5.9.6), is a

P-martingale.
(ii)
Proof. By Itos formula, dY
t
= Y
t
[d

W
t
+ (r
1
2

2
)dt] +
1
2
Y
t

2
dt = Y
t
(d

W
t
+ rdt). So d(e
rt
Y
t
) =
e
rt
Y
t
d

W
t
and e
rt
Y
t
is a

P-martingale. Moreover, if S
t
= S
0
Y
t
+Y
t

t
0
a
Ys
ds, then
dS
t
= S
0
dY
t
+

t
0
a
Y
s
dsdY
t
+adt =

S
0
+

t
0
a
Y
s
ds

Y
t
(d

W
t
+rdt) +adt = S
t
(d

W
t
+rdt) +adt.
This shows S satises (5.9.7).
Remark: To obtain this formula for S, we rst set U
t
= e
rt
S
t
to remove the rS
t
dt term. The SDE for
U is dU
t
= U
t
d

W
t
+ ae
rt
dt. Just like solving linear ODE, to remove U in the d

W
t
term, we consider
V
t
= U
t
e

Wt
. Itos product formula yields
dV
t
= e

Wt
dU
t
+U
t
e

Wt

()d

W
t
+
1
2

2
dt

+dU
t
e

Wt

()d

W
t
+
1
2

2
dt

= e

Wt
ae
rt
dt
1
2

2
V
t
dt.
53
Note V appears only in the dt term, so multiply the integration factor e
1
2

2
t
on both sides of the equation,
we get
d(e
1
2

2
t
V
t
) = ae
rt

Wt+
1
2

2
t
dt.
Set Y
t
= e

Wt+(r
1
2

2
)t
, we have d(S
t
/Y
t
) = adt/Y
t
. So S
t
= Y
t
(S
0
+

t
0
ads
Ys
).
(iii)
Proof.

E[S
T
[T
t
] = S
0

E[Y
T
[T
t
] +

E

Y
T

t
0
a
Y
s
ds +Y
T

T
t
a
Y
s
ds[T
t

= S
0

E[Y
T
[T
t
] +

t
0
a
Y
s
ds

E[Y
T
[T
t
] +a

T
t

Y
T
Y
s
[T
t

ds
= S
0
Y
t

E[Y
Tt
] +

t
0
a
Y
s
dsY
t

E[Y
Tt
] +a

T
t

E[Y
Ts
]ds
=

S
0
+

t
0
a
Y
s
ds

Y
t
e
r(Tt)
+a

T
t
e
r(Ts)
ds
=

S
0
+

t
0
ads
Y
s

Y
t
e
r(Tt)

a
r
(1 e
r(Tt)
).
In particular,

E[S
T
] = S
0
e
rT

a
r
(1 e
rT
).
(iv)
Proof.
d

E[S
T
[T
t
] = ae
r(Tt)
dt +

S
0
+

t
0
ads
Y
s

(e
r(Tt)
dY
t
rY
t
e
r(Tt)
dt) +
a
r
e
r(Tt)
(r)dt
=

S
0
+

t
0
ads
Y
s

e
r(Tt)
Y
t
d

W
t
.
So

E[S
T
[T
t
] is a

P-martingale. As we have argued at the beginning of the solution, risk-neutral pricing is
valid even in the presence of cost of carry. So by an argument similar to that of '5.6.2, the process

E[S
T
[T
t
]
is the futures price process for the commodity.
(v)
Proof. We solve the equation

E[e
r(Tt)
(S
T
K)[T
t
] = 0 for K, and get K =

E[S
T
[T
t
]. So For
S
(t, T) =
Fut
S
(t, T).
(vi)
Proof. We follow the hint. First, we solve the SDE

dX
t
= dS
t
adt +r(X
t
S
t
)dt
X
0
= 0.
By our analysis in part (i), d(e
rt
X
t
) = d(e
rt
S
t
) ae
rt
dt. Integrate from 0 to t on both sides, we get
X
t
= S
t
S
0
e
rt
+
a
r
(1 e
rt
) = S
t
S
0
e
rt

a
r
(e
rt
1). In particular, X
T
= S
T
S
0
e
rT

a
r
(e
rT
1).
Meanwhile, For
S
(t, T) = Fut
s
(t, T) =

E[S
T
[T
t
] =

S
0
+

t
0
ads
Ys

Y
t
e
r(Tt)

a
r
(1e
r(Tt)
). So For
S
(0, T) =
S
0
e
rT

a
r
(1 e
rT
) and hence X
T
= S
T
For
S
(0, T). After the agent delivers the commodity, whose value
is S
T
, and receives the forward price For
S
(0, T), the portfolio has exactly zero value.
54
6. Connections with Partial Dierential Equations
6.1. (i)
Proof. Z
t
= 1 is obvious. Note the form of Z is similar to that of a geometric Brownian motion. So by Itos
formula, it is easy to obtain dZ
u
= b
u
Z
u
du +
u
Z
u
dW
u
, u t.
(ii)
Proof. If X
u
= Y
u
Z
u
(u t), then X
t
= Y
t
Z
t
= x 1 = x and
dX
u
= Y
u
dZ
u
+Z
u
dY
u
+dY
u
Z
u
= Y
u
(b
u
Z
u
du +
u
Z
u
dW
u
) +Z
u

a
u

u

u
Z
u
du +

u
Z
u
dW
u

+
u
Z
u

u
Z
u
du
= [Y
u
b
u
Z
u
+ (a
u

u

u
) +
u

u
]du + (
u
Z
u
Y
u
+
u
)dW
u
= (b
u
X
u
+a
u
)du + (
u
X
u
+
u
)dW
u
.
Remark: To see how to nd the above solution, we manipulate the equation (6.2.4) as follows. First, to
remove the term b
u
X
u
du, we multiply on both sides of (6.2.4) the integrating factor e

u
t
bvdv
. Then
d(X
u
e

u
t
bvdv
) = e

u
t
bvdv
(a
u
du + (
u
+
u
X
u
)dW
u
).
Let

X
u
= e

u
t
bvdv
X
u
, a
u
= e

u
t
bvdv
a
u
and
u
= e

u
t
bvdv

u
, then

X satises the SDE
d

X
u
= a
u
du + (
u
+
u

X
u
)dW
u
= ( a
u
du +
u
dW
u
) +
u

X
u
dW
u
.
To deal with the term
u

X
u
dW
u
, we consider

X
u
=

X
u
e

u
t
vdWv
. Then
d

X
u
= e

u
t
vdWv
[( a
u
du +
u
dW
u
) +
u

X
u
dW
u
] +

X
u

u
t
vdWv
(
u
)dW
u
+
1
2
e

u
t
vdWv

2
u
du

+(
u
+
u

X
u
)(
u
)e

u
t
vdWv
du
= a
u
du +
u
dW
u
+
u

X
u
dW
u

u

X
u
dW
u
+
1
2

X
u

2
u
du
u
(
u
+
u

X
u
)du
= ( a
u

u

u

1
2

X
u

2
u
)du +
u
dW
u
,
where a
u
= a
u
e

u
t
vdWv
and
u
=
u
e

u
t
vdWv
. Finally, use the integrating factor e

u
t
1
2

2
v
dv
, we have
d

X
u
e
1
2

u
t

2
v
dv

= e
1
2

u
t

2
v
dv
(d

X
u
+

X
u

1
2

2
u
du) = e
1
2

u
t

2
v
dv
[( a
u

u

u
)du +
u
dW
u
].
Write everything back into the original X, a and , we get
d

X
u
e

u
t
bvdv

u
t
vdWv+
1
2

u
t

2
v
dv

= e
1
2

u
t

2
v
dv

u
t
vdWv

u
t
bvdv
[(a
u

u

u
)du +
u
dW
u
],
i.e.
d

X
u
Z
u

=
1
Z
u
[(a
u

u

u
)du +
u
dW
u
] = dY
u
.
This inspired us to try X
u
= Y
u
Z
u
.
6.2. (i)
55
Proof. The portfolio is self-nancing, so for any t T
1
, we have
dX
t
=
1
(t)df(t, R
t
, T
1
) +
2
(t)df(t, R
t
, T
2
) +R
t
(X
t

1
(t)f(t, R
t
, T
1
)
2
(t)f(t, R
t
, T
2
))dt,
and
d(D
t
X
t
)
= R
t
D
t
X
t
dt +D
t
dX
t
= D
t
[
1
(t)df(t, R
t
, T
1
) +
2
(t)df(t, R
t
, T
2
) R
t
(
1
(t)f(t, R
t
, T
1
) +
2
(t)f(t, R
t
, T
2
))dt]
= D
t
[
1
(t)

f
t
(t, R
t
, T
1
)dt +f
r
(t, R
t
, T
1
)dR
t
+
1
2
f
rr
(t, R
t
, T
1
)
2
(t, R
t
)dt

+
2
(t)

f
t
(t, R
t
, T
2
)dt +f
r
(t, R
t
, T
2
)dR
t
+
1
2
f
rr
(t, R
t
, T
2
)
2
(t, R
t
)dt

R
t
(
1
(t)f(t, R
t
, T
1
) +
2
(t)f(t, R
t
, T
2
))dt]
=
1
(t)D
t
[R
t
f(t, R
t
, T
1
) +f
t
(t, R
t
, T
1
) +(t, R
t
)f
r
(t, R
t
, T
1
) +
1
2

2
(t, R
t
)f
rr
(t, R
t
, T
1
)]dt
+
2
(t)D
t
[R
t
f(t, R
t
, T
2
) +f
t
(t, R
t
, T
2
) +(t, R
t
)f
r
(t, R
t
, T
2
) +
1
2

2
(t, R
t
)f
rr
(t, R
t
, T
2
)]dt
+D
t
(t, R
t
)[D
t
(t, R
t
)[
1
(t)f
r
(t, R
t
, T
1
) +
2
(t)f
r
(t, R
t
, T
2
)]]dW
t
=
1
(t)D
t
[(t, R
t
) (t, R
t
, T
1
)]f
r
(t, R
t
, T
1
)dt +
2
(t)D
t
[(t, R
t
) (t, R
t
, T
2
)]f
r
(t, R
t
, T
2
)dt
+D
t
(t, R
t
)[
1
(t)f
r
(t, R
t
, T
1
) +
2
(t)f
r
(t, R
t
, T
2
)]dW
t
.
(ii)
Proof. Let
1
(t) = S
t
f
r
(t, R
t
, T
2
) and
2
(t) = S
t
f
r
(t, R
t
, T
1
), then
d(D
t
X
t
) = D
t
S
t
[(t, R
t
, T
2
) (t, R
t
, T
1
)]f
r
(t, R
t
, T
1
)f
r
(t, R
t
, T
2
)dt
= D
t
[[(t, R
t
, T
1
) (t, R
t
, T
2
)]f
r
(t, R
t
, T
1
)f
r
(t, R
t
, T
2
)[dt.
Integrate from 0 to T on both sides of the above equation, we get
D
T
X
T
D
0
X
0
=

T
0
D
t
[[(t, R
t
, T
1
) (t, R
t
, T
2
)]f
r
(t, R
t
, T
1
)f
r
(t, R
t
, T
2
)[dt.
If (t, R
t
, T
1
) = (t, R
t
, T
2
) for some t [0, T], under the assumption that f
r
(t, r, T) = 0 for all values of r
and 0 t T, D
T
X
T
D
0
X
0
> 0. To avoid arbitrage (see, for example, Exercise 5.7), we must have for
a.s. , (t, R
t
, T
1
) = (t, R
t
, T
2
), t [0, T]. This implies (t, r, T) does not depend on T.
(iii)
Proof. In (6.9.4), let
1
(t) = (t), T
1
= T and
2
(t) = 0, we get
d(D
t
X
t
) = (t)D
t

R
t
f(t, R
t
, T) +f
t
(t, R
t
, T) +(t, R
t
)f
r
(t, R
t
, T) +
1
2

2
(t, R
t
)f
rr
(t, R
t
, T)

dt
+D
t
(t, R
t
)(t)f
r
(t, R
t
, T)dW
t
.
This is formula (6.9.5).
If f
r
(t, r, T) = 0, then d(D
t
X
t
) = (t)D
t

R
t
f(t, R
t
, T) +f
t
(t, R
t
, T) +
1
2

2
(t, R
t
)f
rr
(t, R
t
, T)

dt. We
choose (t) = sign

R
t
f(t, R
t
, T) +f
t
(t, R
t
, T) +
1
2

2
(t, R
t
)f
rr
(t, R
t
, T)

. To avoid arbitrage in this


case, we must have f
t
(t, R
t
, T) +
1
2

2
(t, R
t
)f
rr
(t, R
t
, T) = R
t
f(t, R
t
, T), or equivalently, for any r in the
range of R
t
, f
t
(t, r, T) +
1
2

2
(t, r)f
rr
(t, r, T) = rf(t, r, T).
56
6.3.
Proof. We note
d
ds

s
0
bvdv
C(s, T)

= e

s
0
bvdv
[C(s, T)(b
s
) +b
s
C(s, T) 1] = e

s
0
bvdv
.
So integrate on both sides of the equation from t to T, we obtain
e

T
0
bvdv
C(T, T) e

t
0
bvdv
C(t, T) =

T
t
e

s
0
bvdv
ds.
Since C(T, T) = 0, we have C(t, T) = e

t
0
bvdv

T
t
e

s
0
bvdv
ds =

T
t
e

t
s
bvdv
ds. Finally, by A

(s, T) =
a(s)C(s, T) +
1
2

2
(s)C
2
(s, T), we get
A(T, T) A(t, T) =

T
t
a(s)C(s, T)ds +
1
2

T
t

2
(s)C
2
(s, T)ds.
Since A(T, T) = 0, we have A(t, T) =

T
t
(a(s)C(s, T)
1
2

2
(s)C
2
(s, T))ds.
6.4. (i)
Proof. By the denition of , we have

(t) = e
1
2

2

T
t
C(u,T)du
1
2

2
(1)C(t, T) =
1
2
(t)
2
C(t, T).
So C(t, T) =
2

(t)
(t)
2
. Dierentiate both sides of the equation

(t) =
1
2
(t)
2
C(t, T), we get

(t) =
1
2

2
[

(t)C(t, T) +(t)C

(t, T)]
=
1
2

2
[
1
2
(t)
2
C
2
(t, T) +(t)C

(t, T)]
=
1
4

4
(t)C
2
(t, T)
1
2

2
(t)C

(t, T).
So C

(t, T) =

1
4

4
(t)C
2
(t, T)

(t)

/
1
2
(t)
2
=
1
2

2
C
2
(t, T)
2

(t)

2
(t)
.
(ii)
Proof. Plug formulas (6.9.8) and (6.9.9) into (6.5.14), we get

(t)

2
(t)
+
1
2

2
C
2
(t, T) = b(1)
2

(t)

2
(t)
+
1
2

2
C
2
(t, T) 1,
i.e.

(t) b

(t)
1
2

2
(t) = 0.
(iii)
Proof. The characteristic equation of

(t) b

(t)
1
2

2
(t) = 0 is
2
b
1
2

2
= 0, which gives two
roots
1
2
(b

b
2
+ 2
2
) =
1
2
b with =
1
2

b
2
+ 2
2
. Therefore by standard theory of ordinary dierential
equations, a general solution of is (t) = e
1
2
bt
(a
1
e
t
+ a
2
e
t
) for some constants a
1
and a
2
. It is then
easy to see that we can choose appropriate constants c
1
and c
2
so that
(t) =
c
1
1
2
b +
e
(
1
2
b+)(Tt)

c
2
1
2
b
e
(
1
2
b)(Tt)
.
57
(iv)
Proof. From part (iii), it is easy to see

(t) = c
1
e
(
1
2
b+)(Tt)
c
2
e
(
1
2
b)(Tt)
. In particular,
0 = C(T, T) =
2

(T)

2
(T)
=
2(c
1
c
2
)

2
(T)
.
So c
1
= c
2
.
(v)
Proof. We rst recall the denitions and properties of sinh and cosh:
sinhz =
e
z
e
z
2
, coshz =
e
z
+e
z
2
, (sinhz)

= coshz, and (coshz)

= sinhz.
Therefore
(t) = c
1
e

1
2
b(Tt)

e
(Tt)
1
2
b +

e
(Tt)
1
2
b

= c
1
e

1
2
b(Tt)

1
2
b
1
4
b
2

2
e
(Tt)

1
2
b +
1
4
b
2

2
e
(Tt)

=
2c
1

2
e

1
2
b(Tt)

(
1
2
b )e
(Tt)
+ (
1
2
b +)e
(Tt)

=
2c
1

2
e

1
2
b(Tt)
[b sinh((T t)) + 2 cosh((T t))].
and

(t) =
1
2
b
2c
1

2
e

1
2
b(Tt)
[b sinh((T t)) + 2 cosh((T t))]
+
2c
1

2
e

1
2
b(Tt)
[b cosh((T t)) 2
2
sinh((T t))]
= 2c
1
e

1
2
b(Tt)

b
2
2
2
sinh((T t)) +
b

2
cosh((T t))
b

2
cosh((T t))
2
2

2
sinh((T t))

= 2c
1
e

1
2
b(Tt)
b
2
4
2
2
2
sinh((T t))
= 2c
1
e

1
2
b(Tt)
sinh((T t)).
This implies
C(t, T) =
2

(t)

2
(t)
=
sinh((T t))
cosh((T t)) +
1
2
b sinh((T t))
.
(vi)
Proof. By (6.5.15) and (6.9.8), A

(t, T) =
2a

(t)

2
(t)
. Hence
A(T, T) A(t, T) =

T
t
2a

(s)

2
(s)
ds =
2a

2
ln
(T)
(t)
,
and
A(t, T) =
2a

2
ln
(T)
(t)
=
2a

2
ln

e
1
2
b(Tt)
cosh((T t)) +
1
2
b sinh((T t))

.
58
6.5. (i)
Proof. Since g(t, X
1
(t), X
2
(t)) = E[h(X
1
(T), X
2
(T))[T
t
] and e
rt
f(t, X
1
(t), X
2
(t)) = E[e
rT
h(X
1
(T), X
2
(T))[T
t
],
iterated conditioning argument shows g(t, X
1
(t), X
2
(t)) and e
rt
f(t, X
1
(t), X
2
(t)) ar both martingales.
(ii) and (iii)
Proof. We note
dg(t, X
1
(t), X
2
(t))
= g
t
dt +g
x1
dX
1
(t) +g
x2
dX
2
(t) +
1
2
g
x1x2
dX
1
(t)dX
1
(t) +
1
2
g
x2x2
dX
2
(t)dX
2
(t) +g
x1x2
dX
1
(t)dX
2
(t)
=

g
t
+g
x1

1
+g
x2

2
+
1
2
g
x1x1
(
2
11
+
2
12
+ 2
11

12
) +g
x1x2
(
11

21
+
11

22
+
12

21
+
12

22
)
+
1
2
g
x2x2
(
2
21
+
2
22
+ 2
21

22
)

dt + martingale part.
So we must have
g
t
+g
x1

1
+g
x2

2
+
1
2
g
x1x1
(
2
11
+
2
12
+ 2
11

12
) +g
x1x2
(
11

21
+
11

22
+
12

21
+
12

22
)
+
1
2
g
x2x2
(
2
21
+
2
22
+ 2
21

22
) = 0.
Taking = 0 will give part (ii) as a special case. The PDE for f can be similarly obtained.
6.6. (i)
Proof. Multiply e
1
2
bt
on both sides of (6.9.15), we get
d(e
1
2
bt
X
j
(t)) = e
1
2
bt

X
j
(t)
1
2
bdt + (
b
2
X
j
(t)dt +
1
2
dW
j
(t)

= e
1
2
bt
1
2
dW
j
(t).
So e
1
2
bt
X
j
(t) X
j
(0) =
1
2

t
0
e
1
2
bu
dW
j
(u) and X
j
(t) = e

1
2
bt

X
j
(0) +
1
2

t
0
e
1
2
bu
dW
j
(u)

. By Theorem
4.4.9, X
j
(t) is normally distributed with mean X
j
(0)e

1
2
bt
and variance
e
bt
4

2

t
0
e
bu
du =

2
4b
(1 e
bt
).
(ii)
Proof. Suppose R(t) =

d
j=1
X
2
j
(t), then
dR(t) =
d

j=1
(2X
j
(t)dX
j
(t) +dX
j
(t)dX
j
(t))
=
d

j=1

2X
j
(t)dX
j
(t) +
1
4

2
dt

=
d

j=1

bX
2
j
(t)dt +X
j
(t)dW
j
(t) +
1
4

2
dt

d
4

2
bR(t)

dt +

R(t)
d

j=1
X
j
(t)

R(t)
dW
j
(t).
Let B(t) =

d
j=1

t
0
Xj(s)

R(s)
dW
j
(s), then B is a local martingale with dB(t)dB(t) =

d
j=1
X
2
j
(t)
R(t)
dt = dt. So
by Levys Theorem, B is a Brownian motion. Therefore dR(t) = (a bR(t))dt +

R(t)dB(t) (a :=
d
4

2
)
and R is a CIR interest rate process.
59
(iii)
Proof. By (6.9.16), X
j
(t) is dependent on W
j
only and is normally distributed with mean e

1
2
bt
X
j
(0) and
variance

2
4b
[1 e
bt
]. So X
1
(t), , X
d
(t) are i.i.d. normal with the same mean (t) and variance v(t).
(iv)
Proof.
E

e
uX
2
j
(t)

e
ux
2 e

(x(t))
2
2v(t)
dx

2v(t)
=

(12uv(t))x
2
2(t)x+
2
(t)
2v(t)

2v(t)
dx
=

2v(t)
e

(
x
(t)
12uv(t)
)
2
+

2
(t)
12uv(t)


2
(t)
(12uv(t))
2
2v(t)/(12uv(t))
dx
=

1 2uv(t)

2v(t)
e

(
x
(t)
12uv(t)
)
2
2v(t)/(12uv(t))
dx
e

2
(t)(12uv(t))
2
(t)
2v(t)(12uv(t))

1 2uv(t)
=
e

u
2
(t)
12uv(t)

1 2uv(t)
.
(v)
Proof. By R(t) =

d
j=1
X
2
j
(t) and the fact X
1
(t), , X
d
(t) are i.i.d.,
E[e
uR(t)
] = (E[e
uX
2
1
(t)
])
d
= (1 2uv(t))

d
2
e
ud
2
(t)
12uv(t)
= (1 2uv(t))

2a

2
e

e
bt
uR(0)
12uv(t)
.
6.7. (i)
Proof. e
rt
c(t, S
t
, V
t
) =

E[e
rT
(S
T
K)
+
[T
t
] is a martingale by iterated conditioning argument. Since
d(e
rt
c(t, S
t
, V
t
))
= e
rt

c(t, S
t
, V
t
)(r) +c
t
(t, S
t
, V
t
) +c
s
(t, S
t
, V
t
)rS
t
+c
v
(t, S
t
, V
t
)(a bV
t
) +
1
2
c
ss
(t, S
t
, V
t
)V
t
S
2
t
+
1
2
c
vv
(t, S
t
, V
t
)
2
V
t
+c
sv
(t, S
t
, V
t
)V
t
S
t

dt + martingale part,
we conclude rc = c
t
+rsc
s
+c
v
(a bv) +
1
2
c
ss
vs
2
+
1
2
c
vv

2
v +c
sv
sv. This is equation (6.9.26).
(ii)
60
Proof. Suppose c(t, s, v) = sf(t, log s, v) e
r(Tt)
Kg(t, log s, v), then
c
t
= sf
t
(t, log s, v) re
r(Tt)
Kg(t, log s, v) e
r(Tt)
Kg
t
(t, log s, v),
c
s
= f(t, log s, v) +sf
s
(t, log s, v)
1
s
e
r(Tt)
Kg
s
(t, log s, v)
1
s
,
c
v
= sf
v
(t, log s, v) e
r(Tt)
Kg
v
(t, log s, v),
c
ss
= f
s
(t, log s, v)
1
s
+f
ss
(t, log s, v)
1
s
e
r(Tt)
Kg
ss
(t, log s, v)
1
s
2
+e
r(Tt)
Kg
s
(t, log s, v)
1
s
2
,
c
sv
= f
v
(t, log s, v) +f
sv
(t, log s, v) e
r(Tt)
K
s
g
sv
(t, log s, v),
c
vv
= sf
vv
(t, log s, v) e
r(Tt)
Kg
vv
(t, log s, v).
So
c
t
+rsc
s
+ (a bv)c
v
+
1
2
s
2
vc
ss
+svc
sv
+
1
2

2
vc
vv
= sf
t
re
r(Tt)
Kg e
r(Tt)
Kg
t
+rsf +rsf
s
rKe
r(Tt)
g
s
+ (a bv)(sf
v
e
r(Tt)
Kg
v
)
+
1
2
s
2
v

1
s
f
s
+
1
s
f
ss
e
r(Tt)
K
s
2
g
ss
+e
r(Tt)
K
g
s
s
2

+sv

f
v
+f
sv
e
r(Tt)
K
s
g
sv

+
1
2

2
v(sf
vv
e
r(Tt)
Kg
vv
)
= s

f
t
+ (r +
1
2
v)f
s
+ (a bv +v)f
v
+
1
2
vf
ss
+vf
sv
+
1
2

2
vf
vv

Ke
r(Tt)

g
t
+ (r
1
2
v)g
s
+(a bv)g
v
+
1
2
vg
ss
+vg
sv
+
1
2

2
vg
vv

+rsf re
r(Tt)
Kg
= rc.
That is, c satises the PDE (6.9.26).
(iii)
Proof. First, by Markov property, f(t, X
t
, V
t
) = E[1
{X
T
log K}
[T
t
]. So f(T, X
t
, V
t
) = 1
{X
T
log K}
, which
implies f(T, x, v) = 1
{xlog K}
for all x R, v 0. Second, f(t, X
t
, V
t
) is a martingale, so by dierentiating
f and setting the dt term as zero, we have the PDE (6.9.32) for f. Indeed,
df(t, X
t
, V
t
) =

f
t
(t, X
t
, V
t
) +f
x
(t, X
t
, V
t
)(r +
1
2
V
t
) +f
v
(t, X
t
, V
t
)(a bv
t
+V
t
) +
1
2
f
xx
(t, X
t
, V
t
)V
t
+
1
2
f
vv
(t, X
t
, V
t
)
2
V
t
+f
xv
(t, X
t
, V
t
)V
t

dt + martingale part.
So we must have f
t
+ (r +
1
2
v)f
x
+ (a bv +v)f
v
+
1
2
f
xx
v +
1
2
f
vv

2
v +vf
xv
= 0. This is (6.9.32).
(iv)
Proof. Similar to (iii).
(v)
Proof. c(T, s, v) = sf(T, log s, v) e
r(Tt)
Kg(T, log s, v) = s1
{log slog K}
K1
{log slog K}
= 1
{sK}
(s
K) = (s K)
+
.
6.8.
61
Proof. We follow the hint. Suppose h is smooth and compactly supported, then it is legitimate to exchange
integration and dierentiation:
g
t
(t, x) =

t


0
h(y)p(t, T, x, y)dy =


0
h(y)p
t
(t, T, x, y)dy,
g
x
(t, x) =


0
h(y)p
x
(t, T, x, y)dy,
g
xx
(t, x) =


0
h(y)p
xx
(t, T, x, y)dy.
So (6.9.45) implies

0
h(y)

p
t
(t, T, x, y) +(t, x)p
x
(t, T, x, y) +
1
2

2
(t, x)p
xx
(t, T, x, y)

dy = 0. By the ar-
bitrariness of h and assuming , p
t
, p
x
, v, p
xx
are all continuous, we have
p
t
(t, T, x, y) +(t, x)p
x
(t, T, x, y) +
1
2

2
(t, x)p
xx
(t, T, x, y) = 0.
This is (6.9.43).
6.9.
Proof. We rst note dh
b
(X
u
) = h

b
(X
u
)dX
u
+
1
2
h

b
(X
u
)dX
u
dX
u
=

b
(X
u
)(u, X
u
) +
1
2

2
(u, X
u
)h

b
(X
u
)

du+
h

b
(X
u
)(u, X
u
)dW
u
. Integrate on both sides of the equation, we have
h
b
(X
T
) h
b
(X
t
) =

T
t

b
(X
u
)(u, X
u
) +
1
2

2
(u, X
u
)h

b
(X
u
)

du + martingale part.
Take expectation on both sides, we get
E
t,x
[h
b
(X
T
) h
b
(X
t
)] =

h
b
(y)p(t, T, x, y)dy h(x)
=

T
t
E
t,x
[h

b
(X
u
)(u, X
u
) +
1
2

2
(u, X
u
)h

b
(X
u
)]du
=

T
t

b
(y)(u, y) +
1
2

2
(u, y)h

b
(y)

p(t, u, x, y)dydu.
Since h
b
vanishes outside (0, b), the integration range can be changed from (, ) to (0, b), which gives
(6.9.48).
By integration-by-parts formula, we have

b
0
(u, y)p(t, u, x, y)h

b
(y)dy = h
b
(y)(u, y)p(t, u, x, y)[
b
0

b
0
h
b
(y)

y
((u, y)p(t, u, x, y))dy
=

b
0
h
b
(y)

y
((u, y)p(t, u, x, y))dy,
and

b
0

2
(u, y)p(t, u, x, y)h

b
(y)dy =

b
0

y
(
2
(u, y)p(t, u, x, y))h

b
(y)dy =

b
0

2
y
(
2
(u, y)p(t, u, x, y))h
b
(y)dy.
Plug these formulas into (6.9.48), we get (6.9.49).
Dierentiate w.r.t. T on both sides of (6.9.49), we have

b
0
h
b
(y)

T
p(t, T, x, y)dy =

b
0

y
[(T, y)p(t, T, x, y)]h
b
(y)dy +
1
2

b
0

2
y
2
[
2
(T, y)p(t, T, x, y)]h
b
(y)dy,
62
that is,

b
0
h
b
(y)


T
p(t, T, x, y) +

y
((T, y)p(t, T, x, y))
1
2

2
y
2
(
2
(T, y)p(t, T, x, y))

dy = 0.
This is (6.9.50).
By (6.9.50) and the arbitrariness of h
b
, we conclude for any y (0, ),

T
p(t, T, x, y) +

y
((T, y)p(t, T, x, y))
1
2

2
y
2
(
2
(T, y)p(t, T, x, y)) = 0.
6.10.
Proof. Under the assumption that lim
y
(y K)ry p(0, T, x, y) = 0, we have


K
(yK)

y
(ry p(0, T, x, y))dy = (yK)ry p(0, T, x, y)[

K
+


K
ry p(0, T, x, y)dy =


K
ry p(0, T, x, y)dy.
If we further assume (6.9.57) and (6.9.58), then use integration-by-parts formula twice, we have
1
2


K
(y K)

2
y
2
(
2
(T, y)y
2
p(0, T, x, y))dy
=
1
2

(y K)

y
(
2
(T, y)y
2
p(0, T, x, y))[

y
(
2
(T, y)y
2
p(0, T, x, y))dy

=
1
2
(
2
(T, y)y
2
p(0, T, x, y)[

K
)
=
1
2

2
(T, K)K
2
p(0, T, x, K).
Therefore,
c
T
(0, T, x, K) = rc(0, T, x, K) +e
rT


K
(y K) p
T
(0, T, x, y)dy
= re
rT


K
(y K) p(0, T, x, y)dy +e
rT


K
(y K) p
T
(0, T, x, y)dy
= re
rT


K
(y K) p(0, T, x, y)dy e
rT


K
(y K)

y
(ry p(t, T, x, y))dy
+e
rT


K
(y K)
1
2

2
y
2
(
2
(T, y)y
2
p(t, T, x, y))dy
= re
rT


K
(y K) p(0, T, x, y)dy +e
rT


K
ry p(0, T, x, y)dy
+e
rT
1
2

2
(T, K)K
2
p(0, T, x, K)
= re
rT
K


K
p(0, T, x, y)dy +
1
2
e
rT

2
(T, K)K
2
p(0, T, x, K)
= rKc
K
(0, T, x, K) +
1
2

2
(T, K)K
2
c
KK
(0, T, x, K).
7. Exotic Options
7.1. (i)
63
Proof. Since

(, s) =
1

[log s + (r
1
2

2
)] =
log s

1
2
+
r
1
2

(, s) =
log s

(
1
2
)

3
2

t
+
r
1
2

1
2

1
2

t
=
1
2

log s

(1)
r
1
2

(1)

=
1
2

1

log ss + (r
1
2

2
))

=
1
2

(,
1
s
).
(ii)
Proof.

(,
x
c
) =

x

log
x
c
+ (r
1
2

2
)

=
1
x

(,
c
x
) =

x

log
c
x
+ (r
1
2

2
)

=
1
x

.
(iii)
Proof.
N

(, s)) =
1

2
e

(,s)
2
=
1

2
e

(log s+r)
2

2
(log s+r)+
1
4

2
2
2

.
Therefore
N

(
+
(, s))
N

(, s))
= e

2
2
(log s+r)
2
2

=
e
r
s
and e
r
N

(, s)) = sN

(
+
(, s)).
(iv)
Proof.
N

(, s))
N

(, s
1
))
= e

[
(log s+r)
2
(log
1
s
+r)
2
]

2
(log slog
1
s
)
2
2

= e

4r log s2
2
log s
2
2

= e
(
2r

2
1) log s
= s
(
2r

2
1)
.
So N

(, s
1
)) = s
(
2r

2
1)
N

(, s)).
(v)
Proof.
+
(, s)

(, s) =
1

log s + (r +
1
2

2
)

log s + (r
1
2

2
)

=
1

2
=

.
(vi)
Proof.

(, s)

(, s
1
) =
1

log s + (r
1
2

2
)

log s
1
+ (r
1
2

2
)

=
2 log s

.
(vii)
Proof. N

(y) =
1

2
e

y
2
2
, so N

(y) =
1

2
e

y
2
2
(
y
2
2
)

= yN

(y).
64
To be continued ...
7.3.
Proof. We note S
T
= S
0
e

W
T
= S
t
e
(

W
T

Wt)
,

W
T


W
t
= (

W
T


W
t
) + (T t) is independent of T
t
,
sup
tuT
(

W
u

W
t
) is independent of T
t
, and
Y
T
= S
0
e

M
T
= S
0
e
sup
tuT

Wu
1
{

Mtsup
tuT

Wt}
+S
0
e

Mt
1
{

Mt>sup
tuT

Wu}
= S
t
e
sup
tuT
(

Wu

Wt)
1
{
Y
t
S
t
e
sup
tuT
(

Wu

W
t
)
}
+Y
t
1
{
Y
t
S
t
e
sup
tuT
(

Wu

W
t
)
}
.
So E[f(S
T
, Y
T
)[T
t
] = E[f(x
S
Tt
S0
, x
Y
Tt
S0
1
{
y
x

Y
Tt
S
0
}
+ y1
{
y
x

Y
Tt
S
0
}
)], where x = S
t
, y = Y
t
. Therefore
E[f(S
T
, Y
T
)[T
t
] is a Borel function of (S
t
, Y
t
).
7.4.
Proof. By Cauchys inequality and the monotonicity of Y , we have
[
m

j=1
(Y
tj
Y
tj1
)(S
tj
S
tj1
)[
m

j=1
[Y
tj
Y
tj1
[[S
tj
S
tj1
[

j=1
(Y
tj
Y
tj1
)
2

j=1
(S
tj
S
tj1
)
2

max
1jm
[Y
tj
Y
tj1
[(Y
T
Y
0
)

j=1
(S
tj
S
tj1
)
2
.
If we increase the number of partition points to innity and let the length of the longest subinterval
max
1jm
[t
j
t
j1
[ approach zero, then

m
j=1
(S
tj
S
tj1
)
2

[S]
T
[S]
0
< and max
1jm
[Y
tj

Y
tj1
[ 0 a.s. by the continuity of Y . This implies

m
j=1
(Y
tj
Y
tj1
)(S
tj
S
tj1
) 0.
8. American Derivative Securities
8.1.
Proof. v

L
(L+) = (K L)(
2r

2
)(
x
L
)

2r

2
1 1
L

x=L
=
2r

2
L
(K L). So v

L
(L+) = v

L
(L) if and only if

2r

2
L
(K L) = 1. Solve for L, we get L =
2rK
2r+
2
.
8.2.
Proof. By the calculation in Section 8.3.3, we can see v
2
(x) (K
2
x)
+
(K
1
x)
+
, rv
2
(x) rxv

2
(x)
1
2

2
x
2
v

2
(x) 0 for all x 0, and for 0 x < L
1
< L
2
,
rv
2
(x) rxv

2
(x)
1
2

2
x
2
v

2
(x) = rK
2
> rK
1
> 0.
So the linear complementarity conditions for v
2
imply v
2
(x) = (K
2
x)
+
= K
2
x > K
1
x = (K
1
x)
+
on [0, L
1
]. Hence v
2
(x) does not satisfy the third linear complementarity condition for v
1
: for each x 0,
equality holds in either (8.8.1) or (8.8.2) or both.
8.3. (i)
65
Proof. Suppose x takes its values in a domain bounded away from 0. By the general theory of linear
dierential equations, if we can nd two linearly independent solutions v
1
(x), v
2
(x) of (8.8.4), then any
solution of (8.8.4) can be represented in the form of C
1
v
1
+C
2
v
2
where C
1
and C
2
are constants. So it suces
to nd two linearly independent special solutions of (8.8.4). Assume v(x) = x
p
for some constant p to be
determined, (8.8.4) yields x
p
(rpr
1
2

2
p(p1)) = 0. Solve the quadratic equation 0 = rpr
1
2

2
p(p1) =
(
1
2

2
p r)(p 1), we get p = 1 or
2r

2
. So a general solution of (8.8.4) has the form C
1
x +C
2
x

2r

2
.
(ii)
Proof. Assume there is an interval [x
1
, x
2
] where 0 < x
1
< x
2
< , such that v(x) 0 satises (8.3.19)
with equality on [x
1
, x
2
] and satises (8.3.18) with equality for x at and immediately to the left of x
1
and
for x at and immediately to the right of x
2
, then we can nd some C
1
and C
2
, so that v(x) = C
1
x+C
2
x

2r

2
on [x
1
, x
2
]. If for some x
0
[x
1
, x
2
], v(x
0
) = v

(x
0
) = 0, by the uniqueness of the solution of (8.8.4), we
would conclude v 0. This is a contradiction. So such an x
0
cannot exist. This implies 0 < x
1
< x
2
< K
(if K x
2
, v(x
2
) = (K x
2
)
+
= 0 and v

(x
2
)=the right derivative of (K x)
+
at x
2
, which is 0).
1
Thus
we have four equations for C
1
and C
2
:

C
1
x
1
+C
2
x

2r

2
1
= K x
1
C
1
x
2
+C
2
x

2r

2
2
= K x
2
C
1

2r

2
C
2
x

2r

2
1
1
= 1
C
1

2r

2
C
2
x

2r

2
1
2
= 1.
Since x
1
= x
2
, the last two equations imply C
2
= 0. Plug C
2
= 0 into the rst two equations, we have
C
1
=
Kx1
x1
=
Kx2
x2
; plug C
2
= 0 into the last two equations, we have C
1
= 1. Combined, we would have
x
1
= x
2
. Contradiction. Therefore our initial assumption is incorrect, and the only solution v that satises
the specied conditions in the problem is the zero solution.
(iii)
Proof. If in a right neighborhood of 0, v satises (8.3.19) with equality, then part (i) implies v(x) = C
1
x +
C
2
x

2r

2
for some constants C
1
and C
2
. Then v(0) = lim
x0
v(x) = 0 < (K 0)
+
, i.e. (8.3.18) will be
violated. So we must have rv rxv


1
2

2
x
2
v

> 0 in a right neighborhood of 0. According to (8.3.20),


v(x) = (Kx)
+
near o. So v(0) = K. We have thus concluded simultaneously that v cannot satisfy (8.3.19)
with equality near 0 and v(0) = K, starting from rst principles (8.3.18)-(8.3.20).
(iv)
Proof. This is already shown in our solution of part (iii): near 0, v cannot satisfy (8.3.19) with equality.
(v)
Proof. If v satisfy (Kx)
+
with equality for all x 0, then v cannot have a continuous derivative as stated
in the problem. This is a contradiction.
(vi)
1
Note we have interpreted the condition v(x) satises (8.3.18) with equality for x at and immediately to the right of x
2

as v(x
2
) = (K x
2
)
+
and v

(x
2
) =the right derivative of (K x)
+
at x
2
. This is weaker than v(x) = (K x) in a right
neighborhood of x
2
.
66
Proof. By the result of part (i), we can start with v(x) = (K x)
+
on [0, x
1
] and v(x) = C
1
x +C
2
x

2r

2
on
[x
1
, ). By the assumption of the problem, both v and v

are continuous. Since (Kx)


+
is not dierentiable
at K, we must have x
1
K.This gives us the equations

K x
1
= (K x
1
)
+
= C
1
x
1
+C
2
x

2r

2
1
1 = C
1

2r

2
C
2
x

2r

2
1
1
.
Because v is assumed to be bounded, we must have C
1
= 0 and the above equations only have two unknowns:
C
2
and x
1
. Solve them for C
2
and x
1
, we are done.
8.4. (i)
Proof. This is already shown in part (i) of Exercise 8.3.
(ii)
Proof. We solve for A, B the equations

AL

2r

2
+BL = K L

2r

2
AL

2r

2
1
+B = 1,
and we obtain A =

2
KL
2r

2
+2r
, B =
2rK
L(
2
+2r)
1.
(iii)
Proof. By (8.8.5), B > 0. So for x K, f(x) BK > 0 = (K x)
+
. If L x < K,
f(x) (K x)
+
=

2
KL
2r

2
+ 2r
x

2r

2
+
2rKx
L(
2
+ 2r)
K = x

2r

2
KL
2r

2
+ 2r(
x
L
)
2r

2
+1
(
2
+ 2r)(
x
L
)
2r

(
2
+ 2r)L
.
Let g() =
2
+ 2r
2r

2
+1
(
2
+ 2r)
2r

2
with 1. Then g(1) = 0 and g

() = 2r(
2r

2
+ 1)
2r

2
(
2
+
2r)
2r

2r

2
1
=
2r

2
(
2
+ 2r)
2r

2
1
( 1) 0. So g() 0 for any 1. This shows f(x) (K x)
+
for
L x < K. Combined, we get f(x) (K x)
+
for all x L.
(iv)
Proof. Since lim
x
v(x) = lim
x
f(x) = and lim
x
v
L
(x) = lim
x
(K L

)(
x
L
)

2r

2
= 0, v(x)
and v
L
(x) are dierent. By part (iii), v(x) (K x)
+
. So v satises (8.3.18). For x L, rv rxv


1
2

2
x
2
v

= rf rxf
1
2

2
x
2
f

= 0. For 0 x L, rv rxv

1
2

2
x
2
v

= r(Kx) +rx = rK. Combined,


rv rxv

1
2

2
x
2
v

0 for x 0. So v satises (8.3.19). Along the way, we also showed v satises (8.3.20).
In summary, v satises the linear complementarity condition (8.3.18)-(8.3.20), but v is not the function v
L
given by (8.3.13).
(v)
Proof. By part (ii), B = 0 if and only if
2rK
L(
2
+2r)
1 = 0, i.e. L =
2rK
2r+
2
. In this case, v(x) = Ax

2r

2
=

2
K

2
+2r
(
x
L
)

2r

2
= (K L)(
x
L
)

2r

2
= v
L
(x), on the interval [L, ).
8.5. The diculty of the dividend-paying case is that from Lemma 8.3.4, we can only obtain

E[e
(ra)
L
],
not

E[e
r
L
]. So we have to start from Theorem 8.3.2.
(i)
67
Proof. By (8.8.9), S
t
= S
0
e

Wt+(ra
1
2

2
)t
. Assume S
0
= x, then S
t
= L if and only if

W
t

1

(r a
1
2

2
)t =
1

log
x
L
. By Theorem 8.3.2,

E[e
r
L
] = e

log
x
L

(ra
1
2

2
)+

2
(ra
1
2

2
)
2
+2r

.
If we set =
1

2
(r a
1
2

2
) +
1

2
(r a
1

2
)
2
+ 2r, we can write

E[e
r
L
] as e
log
x
L
= (
x
L
)

. So
the risk-neutral expected discounted pay o of this strategy is
v
L
(x) =

K x, 0 x L
(K L)(
x
L
)

, x > L.
(ii)
Proof.

L
v
L
(x) = (
x
L
)

(1
(KL)
L
). Set

L
v
L
(x) = 0 and solve for L

, we have L

=
K
+1
.
(iii)
Proof. By It os formula, we have
d

e
rt
v
L
(S
t
)

= e
rt

rv
L
(S
t
) +v

L
(S
t
)(r a)S
t
+
1
2
v

L
(S
t
)
2
S
2
t

dt +e
rt
v

L
(S
t
)S
t
d

W
t
.
If x > L

,
rv
L
(x) +v

L
(x)(r a)x +
1
2
v

L
(x)
2
x
2
= r(K L

x
L

+ (r a)x(K L

)()
x
1
L

+
1
2

2
x
2
()( 1)(K L

)
x
2
L

= (K L

x
L

r (r a) +
1
2

2
( + 1)

.
By the denition of , if we dene u = r a
1
2

2
, we have
r + (r a)
1
2

2
( + 1)
= r
1
2

2
+(r a
1
2

2
)
= r
1
2

2
+
1

u
2

2
+ 2r

2
+

2
+
1

u
2

2
+ 2r

u
= r
1
2

u
2

4
+
2u

u
2

2
+ 2r +
1

u
2

2
+ 2r

+
u
2

2
+
u

u
2

2
+ 2r
= r
u
2
2
2

u

u
2

2
+ 2r
1
2

u
2

2
+ 2r

+
u
2

2
+
u

u
2

2
+ 2r
= 0.
If x < L

, rv
L
(x) +v

L
(x)(r a)x +
1
2
v

L
(x)
2
x
2
= r(K x) +(1)(r a)x = rK +ax. Combined,
we get
d

e
rt
v
L
(S
t
)

= e
rt
1
{St<L}
(rK aS
t
)dt +e
rt
v

L
(S
t
)S
t
d

W
t
.
68
Following the reasoning in the proof of Theorem 8.3.5, we only need to show 1
{x<L}
(rKax) 0 to nish
the solution. This is further equivalent to proving rK aL

0. Plug L

=
K
+1
into the expression and
note
1

2
(r a
1
2

2
)
2
+
1

2
(r a
1
2

2
) 0, the inequality is further reduced to r( +1) a 0.
We prove this inequality as follows.
Assume for some K, r, a and (K and are assumed to be strictly positive, r and a are assumed to be
non-negative), rK aL

< 0, then necessarily r < a, since L

=
K
+1
K. As shown before, this means
r( + 1) a < 0. Dene =
ra

, then < 0 and =


1

2
(r a
1
2

2
) +
1

2
(r a
1
2

2
)
2
+ 2r =
1

(
1
2
) +
1

(
1
2
)
2
+ 2r. We have
r( + 1) a < 0 (r a) +r < 0
(r a)

(
1
2
) +
1

(
1
2
)
2
+ 2r

+r < 0
(
1
2
) +

(
1
2
)
2
+ 2r +r < 0

(
1
2
)
2
+ 2r < r (
1
2
)(< 0)

2
[(
1
2
)
2
+ 2r] > r
2
+
2
(
1
2
)
2
+ 2r(
1
2

2
)
0 > r
2
r
2
0 > r
2
.
Since
2
< 0, we have obtained a contradiction. So our initial assumption is incorrect, and rK aL

0
must be true.
(iv)
Proof. The proof is similar to that of Corollary 8.3.6. Note the only properties used in the proof of Corollary
8.3.6 are that e
rt
v
L
(S
t
) is a supermartingale, e
rt
L
v
L
(S
t

L
) is a martingale, and v
L
(x) (Kx)
+
.
Part (iii) already proved the supermartingale-martingale property, so it suces to show v
L
(x) (K x)
+
in our problem. Indeed, by 0, L

=
K
+1
< K. For x K > L

, v
L
(x) > 0 = (Kx)
+
; for 0 x < L

,
v
L
(x) = K x = (K x)
+
; nally, for L

x K,
d
dx
(v
L
(x) (K x)) = (K L

)
x
1
L

+ 1 (K L

)
L
1

+ 1 = (K
K
+ 1
)
1
K
+1
+ 1 = 0.
and (v
L
(x) (K x))[
x=L
= 0. So for L

x K, v
L
(x) (K x)
+
0. Combined, we have
v
L
(x) (K x)
+
0 for all x 0.
8.6.
Proof. By Lemma 8.5.1, X
t
= e
rt
(S
t
K)
+
is a submartingale. For any
0,T
, Theorem 8.8.1 implies

E[e
rT
(S
T
K)
+
]

E[e
rT
(S
T
K)
+
] E[e
r
(S

K)
+
1
{<}
] = E[e
r
(S

K)
+
],
where we take the convention that e
r
(S

K)
+
= 0 when = . Since is arbitrarily chosen,

E[e
rT
(S
T

K)
+
] max

0,T

E[e
r
(S

K)
+
]. The other direction is trivial since T
0,T
.
8.7.
69
Proof. Suppose [0, 1] and 0 x
1
x
2
, we have f((1 )x
1
+ x
2
) (1 )f(x
1
) + f(x
2
)
(1 )h(x
1
) +h(x
2
). Similarly, g((1 )x
1
+x
2
) (1 )h(x
1
) +h(x
2
). So
h((1 )x
1
+x
2
) = maxf((1 )x
1
+x
2
), g((1 )x
1
+x
2
) (1 )h(x
1
) +h(x
2
).
That is, h is also convex.
9. Change of Numeraire
To provide an intuition for change of numeraire, we give a summary of results for change of
numeraire in discrete case. This summary is based on Shiryaev [5].
Consider a model of nancial market (

B,

B, S) as in [1] Denition 2.1.1 or [5] page 383. Here

B and

B are both one-dimensional while S could be a vector price process. Suppose



B and

B are both strictly
positive, then both of them can be chosen as numeaire.
Several results hold under this model. First, no-arbitrage and completeness properties of market are
independent of the choice of numeraire (see, for example, Shiryaev [5] page 413 Remark and page 481).
Second, if the market is arbitrage-free, then corresponding to

B (resp.

B), there is an equivalent probability

P (resp.

P), such that

B
,
S

(resp.

B
,
S

) is a martingale under

P (resp.

P). Third, if the market is
both arbitrage-free and complete, we have the relation
d

P =

B
T

B
T
1
E


B0

B0
d

P.
Finally, if f
T
is a European contingent claim with maturity N and the market is both arbitrage-free and
complete, then

B
t

E

f
T

B
T
[T
t

=

B
t

f
T

B
T
[T
t

.
That is, the price of f
T
is independent of the choice of numeraire.
The above theoretical results can be applied to market involving foreign money market account. We con-
sider the following market: a domestic money market account M (M
0
= 1), a foreign money market account
M
f
(M
f
0
= 1), a (vector) asset price process S called stock. Suppose the domestic vs. foreign currency
exchange rate is Q. Note Q is not a traded asset. Denominated by domestic currency, the traded assets
are (M, M
f
Q, S), where M
f
Q can be seen as the price process of one unit foreign currency. Domestic risk-
neutral measure

P is such that

M
f
Q
M
,
S
M

is a

P-martingale. Denominated by foreign currency, the traded
assets are

M
f
,
M
Q
,
S
Q

. Foreign risk-neutral measure



P
f
is such that

M
QM
f
,
S
QM
f

is a

P
f
-martingale.
This is a change of numeraire in the market denominated by domestic currency, from M to M
f
Q. If we
assume the market is arbitrage-free and complete, the foreign risk-neutral measure is
d

P
f
=
Q
T
M
f
T
M
T
E

Q0M
f
0
M0
d

P =
Q
T
D
T
M
f
T
Q
0
d

P
on T
T
. Under the above set-up, for a European contingent claim f
T
, denominated in domestic currency, its
payo in foreign currency is f
T
/Q
T
. Therefore its foreign price is

E
f

D
f
T
f
T
D
f
t
Q
T
[T
t

. Convert this price into


domestic currency, we have Q
t

E
f

D
f
T
f
T
D
f
t
Q
T
[T
t

. Use the relation between



P
f
and

P on T
T
and the Bayes
formula, we get
Q
t

E
f

D
f
T
f
T
D
f
t
Q
T
[T
t

=

E

D
T
f
T
D
t
[T
t

.
The RHS is exactly the price of f
T
in domestic market if we apply risk-neutral pricing.
9.1. (i)
70
Proof. For any 0 t T, by Lemma 5.5.2,
E
(M2)

M
1
(T)
M
2
(T)

T
t

= E

M
2
(T)
M
2
(t)
M
1
(T)
M
2
(T)

T
t

=
E[M
1
(T)[T
t
]
M
2
(t)
=
M
1
(t)
M
2
(t)
.
So
M1(t)
M2(t)
is a martingale under P
M2
.
(ii)
Proof. Let M
1
(t) = D
t
S
t
and M
2
(t) = D
t
N
t
/N
0
. Then

P
(N)
as dened in (9.2.6) is P
(M2)
as dened in
Remark 9.2.5. Hence
M1(t)
M2(t)
=
St
Nt
N
0
is a martingale under

P
(N)
, which implies S
(N)
t
=
St
Nt
is a martingale
under

P
(N)
.
9.2. (i)
Proof. Since N
1
t
= N
1
0
e

Wt(r
1
2

2
)t
, we have
d(N
1
t
) = N
1
0
e

Wt(r
1
2

2
)t
[d

W
t
(r
1
2

2
)dt +
1
2

2
dt] = N
1
t
(d

W
t
rdt).
(ii)
Proof.
d

M
t
= M
t
d

1
N
t

+
1
N
t
dM
t
+d

1
N
t

dM
t
=

M
t
(d

W
t
rdt) +r

M
t
dt =

M
t
d

W
t
.
Remark: This can also be obtained directly from Theorem 9.2.2.
(iii)
Proof.
d

X
t
= d

X
t
N
t

= X
t
d

1
N
t

+
1
N
t
dX
t
+d

1
N
t

dX
t
= (
t
S
t
+
t
M
t
)d

1
N
t

+
1
N
t
(
t
dS
t
+
t
dM
t
) +d

1
N
t

(
t
dS
t
+
t
dM
t
)
=
t

S
t
d

1
N
t

+
1
N
t
dS
t
+d

1
N
t

dS
t

+
t

M
t
d

1
N
t

+
1
N
t
dM
t
+d

1
N
t

dM
t

=
t
d

S
t
+
t
d

M
t
.
9.3. To avoid singular cases, we need to assume 1 < < 1.
(i)
Proof. N
t
= N
0
e

W3(t)+(r
1
2

2
)t
. So
dN
1
t
= d(N
1
0
e

W3(t)(r
1
2

2
)t
)
= N
1
0
e

W3(t)(r
1
2

2
)t

W
3
(t) (r
1
2

2
)dt +
1
2

2
dt

= N
1
t
[d

W
3
(t) (r
2
)dt],
71
and
dS
(N)
t
= N
1
t
dS
t
+S
t
dN
1
t
+dS
t
dN
1
t
= N
1
t
(rS
t
dt +S
t
d

W
1
(t)) +S
t
N
1
t
[d

W
3
(t) (r
2
)dt]
= S
(N)
t
(rdt +d

W
1
(t)) +S
(N)
t
[d

W
3
(t) (r
2
)dt] S
(N)
t
dt
= S
(N)
t
(
2
)dt +S
(N)
t
(d

W
1
(t) d

W
3
(t)).
Dene =

2
2 +
2
and

W
4
(t) =

W
1
(t)

W
3
(t), then

W
4
is a martingale with quadratic
variation
[

W
4
]
t
=

2

2
t 2

2
t +

2
r
2
t = t.
By Levys Theorem,

W
4
is a BM and therefore, S
(N)
t
has volatility =

2
2 +
2
.
(ii)
Proof. This problem is the same as Exercise 4.13, we dene

W
2
(t) =

1
2

W
1
(t) +
1

1
2

W
3
(t), then

W
2
is a martingale, with
(d

W
2
(t))
2
=

1
2
d

W
1
(t) +
1

1
2
d

W
3
(t)

2
=


2
1
2
+
1
1
2

2
2
1
2

dt = dt,
and d

W
2
(t)d

W
1
(t) =

1
2
dt +

1
2
dt = 0. So

W
2
is a BM independent of

W
1
, and dN
t
= rN
t
dt +
N
t
d

W
3
(t) = rN
t
dt +N
t
[d

W
1
(t) +

1
2
d

W
2
(t)].
(iii)
Proof. Under

P, (

W
1
,

W
2
) is a two-dimensional BM, and

dS
t
= rS
t
dt +S
t
d

W
1
(t) = rS
t
dt +S
t
(, 0)

W
1
(t)
d

W
2
(t)

dN
t
= rN
t
dt +N
t
d

W
3
(t) = rN
t
dt +N
t
(,

1
2
)

W
1
(t)
d

W
2
(t)

.
So under

P, the volatility vector for S is (, 0), and the volatility vector for N is (,

1
2
). By Theorem
9.2.2, under the measure

P
(N)
, the volatility vector for S
(N)
is (v
1
, v
2
) = ( ,

1
2
. In particular,
the volatility of S
(N)
is

v
2
1
+v
2
2
=

( )
2
+ (

1
2
)
2
=

2
2 +
2
,
consistent with the result of part (i).
9.4.
Proof. From (9.3.15), we have M
f
t
Q
t
= M
f
0
Q
0
e

t
0
2(s)d

W3(s)+

t
0
(Rs
1
2

2
2
(s))ds
. So
D
f
t
Q
t
= D
f
0
Q
1
0
e

t
0
2(s)d

W3(s)

t
0
(Rs
1
2

2
2
(s))ds
72
and
d

D
f
t
Q
t

=
D
f
t
Q
t
[
2
(t)d

W
3
(t) (R
t

1
2

2
2
(t))dt +
1
2

2
2
(t)dt] =
D
f
t
Q
t
[
2
(t)d

W
3
(t) (R
t

2
2
(t))dt].
To get (9.3.22), we note
d

M
t
D
f
t
Q
t

= M
t
d

D
f
t
Q
t

+
D
f
t
Q
t
dM
t
+dM
t
d

D
f
t
Q
t

=
M
t
D
f
t
Q
t
[
2
(t)d

W
3
(t) (R
t

2
2
(t))dt] +
R
t
M
t
D
f
t
Q
t
dt
=
M
t
D
f
t
Q
t
(
2
(t)d

W
3
(t)
2
2
(t)dt)
=
M
t
D
f
t
Q
t

2
(t)d

W
f
3
(t).
To get (9.3.23), we note
d

D
f
t
S
t
Q
t

=
D
f
t
Q
t
dS
t
+S
t
d

D
f
t
Q
t

+dS
t
d

D
f
t
Q
t

=
D
f
t
Q
t
S
t
(R
t
dt +
1
(t)d

W
1
(t)) +
S
t
D
f
t
Q
t
[
2
(t)d

W
3
(t) (R
t

2
2
(t))dt]
+S
t

1
(t)d

W
1
(t)
D
f
t
Q
t
(
2
(t))d

W
3
(t)
=
D
f
t
S
t
Q
t
[
1
(t)d

W
1
(t)
2
(t)d

W
3
(t) +
2
2
(t)dt
1
(t)
2
(t)
t
dt]
=
D
f
t
S
t
Q
t
[
1
(t)d

W
f
1
(t)
2
d

W
f
3
(t)].
9.5.
Proof. We combine the solutions of all the sub-problems into a single solution as follows. The payo of a
quanto call is (
S
T
Q
T
K)
+
units of domestic currency at time T. By risk-neutral pricing formula, its price at
time t is

E[e
r(Tt)
(
S
T
Q
T
K)
+
[T
t
]. So we need to nd the SDE for
St
Qt
under risk-neutral measure

P. By
formula (9.3.14) and (9.3.16), we have S
t
= S
0
e
1

W1(t)+(r
1
2

2
1
)t
and
Q
t
= Q
0
e
2

W3(t)+(rr
f

1
2

2
2
)t
= Q
0
e
2

W1(t)+2

1
2
W2(t)+(rr
f

1
2

2
2
)t
.
So
St
Qt
=
S0
Q0
e
(12)

W1(t)2

1
2
W2(t)+(r
f
+
1
2

2
2

1
2

2
1
)t
. Dene

4
=

(
1

2
)
2
+
2
2
(1
2
) =

2
1
2
1

2
+
2
2
and

W
4
(t) =

1

2

W
1
(t)

2

1
2

W
2
(t).
Then

W
4
is a martingale with [

W
4
]
t
=
(12)
2

2
4
t +
2(1
2
)

2
4
t +t. So

W
4
is a Brownian motion under

P. So
if we set a = r r
f
+
1

2

2
2
, we have
S
t
Q
t
=
S
0
Q
0
e
4

W4(t)+(ra
1
2

2
4
)t
and d

S
t
Q
t

=
S
t
Q
t
[
4
d

W
4
(t) + (r a)dt].
73
Therefore, under

P,
St
Qt
behaves like dividend-paying stock and the price of the quanto call option is like the
price of a call option on a dividend-paying stock. Thus formula (5.5.12) gives us the desired price formula
for quanto call option.
9.6. (i)
Proof. d
+
(t) d

(t) =
1

Tt

2
(T t) =

T t. So d

(t) = d
+
(t)

T t.
(ii)
Proof. d
+
(t)+d

(t) =
2

Tt
log
For
S
(t,T)
K
. So d
2
+
(t)d
2

(t) = (d
+
(t)+d

(t))(d
+
(t)d

(t)) = 2 log
For
S
(t,T)
K
.
(iii)
Proof.
For
S
(t, T)e
d
2
+
(t)/2
Ke
d
2

(t)
= e
d
2
+
(t)/2
[For
S
(t, T) Ke
d
2
+
(t)/2d
2

(t)/2
]
= e
d
2
+
(t)/2
[For
S
(t, T) Ke
log
For
S
(t,T)
K
]
= 0.
(iv)
Proof.
dd
+
(t)
=
1
2

(T t)
3
[log
For
S
(t, T)
K
+
1
2

2
(T t)]dt +
1

T t

dFor
S
(t, T)
For
S
(t, T)

(dFor
S
(t, T))
2
2For
S
(t, T)
2

1
2
dt

=
1
2

(T t)
3
log
For
S
(t, T)
K
dt +

4

T t
dt +
1

T t
(d

W
T
(t)
1
2

2
dt
1
2

2
dt)
=
1
2(T t)
3/2
log
For
S
(t, T)
K
dt
3
4

T t
dt +
d

W
T
(t)

T t
.
(v)
Proof. dd

(t) = dd
+
(t) d(

T t) = dd
+
(t) +
dt
2

Tt
.
(vi)
Proof. By (iv) and (v), (dd

(t))
2
= (dd
+
(t))
2
=
dt
Tt
.
(vii)
Proof. dN(d
+
(t)) = N

(d
+
(t))dd
+
(t)+
1
2
N

(d
+
(t))(dd
+
(t))
2
=
1

2
e

d
2
+
(t)
2
dd
+
(t)+
1
2
1

2
e

d
2
+
(t)
2
(d
+
(t))
dt
Tt
.
(viii)
74
Proof.
dN(d

(t)) = N

(d

(t))dd

(t) +
1
2
N

(d

(t))(dd

(t))
2
=
1

2
e

d
2

(t)
2

dd
+
(t) +
dt
2

T t

+
1
2
e

d
2

(t)
2

2
(d

(t))
dt
T t
=
1

2
e
d
2

(t)/2
dd
+
(t) +
e
d
2

(t)/2
2

2(T t)
dt +
e

d
2

(t)(

Ttd
+
(t))
2
2(T t)

2
dt
=
1

2
e
d
2

(t)/2
dd
+
(t) +
e
d
2

(t)/2

2(T t)
dt
d
+
(t)e

d
2

(t)
2
2(T t)

2
dt.
(ix)
Proof.
dFor
S
(t, T)dN(d
+
(t)) = For
S
(t, T)d

W
T
(t)
e
d
2
+
(t)/2

2
1

T t
d

W
T
(t) =
For
S
(t, T)e
d
2
+
(t)/2

2(T t)
dt.
(x)
Proof.
For
S
(t, T)dN(d
+
(t)) +dFor
S
(t, T)dN(d
+
(t)) KdN(d

(t))
= For
S
(t, T)

2
e
d
2
+
(t)/2
dd
+
(t)
d
+
(t)
2(T t)

2
e
d
2
+
(t)/2
dt

+
For
S
(t, T)e
d
2
+
(t)/2

2(T t)
dt
K

e
d
2

(t)/2

2
dd
+
(t) +

2(T t)
e
d
2

(t)/2
dt
d
+
(t)
2(T t)

2
e
d
2

(t)/2
dt

For
S
(t, T)d
+
(t)
2(T t)

2
e
d
2
+
(t)/2
+
For
S
(t, T)e
d
2
+
(t)/2

2(T t)

Ke
d
2

(t)/2

2(T t)

Kd
+
(t)
2(T t)

2
e
d
2

(t)/2

dt
+
1

For
S
(t, T)e
d
2
+
(t)/2
Ke
d
2

(t)/2

dd
+
(t)
= 0.
The last = comes from (iii), which implies e
d
2

(t)/2
=
For
S
(t,T)
K
e
d
2
+
(t)/2
.
10. Term-Structure Models
10.1. (i)
Proof. Using the notation I
1
(t), I
2
(t), I
3
(t) and I
4
(t) introduced in the problem, we can write Y
1
(t) and
Y
2
(t) as Y
1
(t) = e
1t
Y
1
(0) +e
1t
I
1
(t) and
Y
2
(t) =

21
12
(e
1t
e
2t
)Y
1
(0) +e
2t
Y
2
(0) +
21
12

e
1t
I
1
(t) e
2t
I
2
(t)

e
2t
I
3
(t), if
1
=
2
;

21
te
1t
Y
1
(0) +e
1t
Y
2
(0)
21

te
1t
I
1
(t) e
1t
I
4
(t)

+e
1t
I
3
(t), if
1
=
2
.
75
Since all the I
k
(t)s (k = 1, , 4) are normally distributed with zero mean, we can conclude

E[Y
1
(t)] =
e
1t
Y
1
(0) and

E[Y
2
(t)] =

21
12
(e
1t
e
2t
)Y
1
(0) +e
2t
Y
2
(0), if
1
=
2
;

21
te
1t
Y
1
(0) +e
1t
Y
2
(0), if
1
=
2
.
(ii)
Proof. The calculation relies on the following fact: if X
t
and Y
t
are both martingales, then X
t
Y
t
[X, Y ]
t
is
also a martingale. In particular,

E[X
t
Y
t
] =

E[X, Y ]
t
. Thus

E[I
2
1
(t)] =

t
0
e
21u
du =
e
21t
1
2
1
,

E[I
1
(t)I
2
(t)] =

t
0
e
(1+2)u
du =
e
(1+2)t
1

1
+
2
,

E[I
1
(t)I
3
(t)] = 0,

E[I
1
(t)I
4
(t)] =

t
0
ue
21u
du =
1
2
1

te
21t

e
21t
1
2
1

and

E[I
2
4
(t)] =

t
0
u
2
e
21u
du =
t
2
e
21t
2
1

te
21t
2
2
1
+
e
21t
1
4
3
1
.
(iii)
Proof. Following the hint, we have

E[I
1
(s)I
2
(t)] =

E[J
1
(t)I
2
(t)] =

t
0
e
(1+2)u
1
{us}
du =
e
(1+2)s
1

1
+
2
.
10.2. (i)
Proof. Assume B(t, T) = E[e

T
t
Rsds
[T
t
] = f(t, Y
1
(t), Y
2
(t)). Then d(D
t
B(t, T)) = D
t
[R
t
f(t, Y
1
(t), Y
2
(t))dt+
df(t, Y
1
(t), Y
2
(t))]. By Itos formula,
df(t, Y
1
(t), Y
2
(t)) = [f
t
(t, Y
1
(t), Y
2
(t)) +f
y1
(t, Y
1
(t), Y
2
(t))(
1
Y
1
(t)) +f
y2
(t, Y
1
(t), Y
2
(t))(
2
)Y
2
(t)]
+f
y1y2
(t, Y
1
(t), Y
2
(t))
21
Y
1
(t) +
1
2
f
y1y1
(t, Y
1
(t), Y
2
(t))Y
1
(t)
+
1
2
f
y2y2
(t, Y
1
(t), Y
2
(t))(
2
21
Y
1
(t) + +Y
1
(t))]dt + martingale part.
Since D
t
B(t, T) is a martingale, we must have

(
0
+
1
y
1
+
2
y
2
) +

t
+ (
1
y
1
)

y
1

2
y
2

y
2
+
1
2

2
21
y
1

2
y
1
y
2
+y
1

2
y
2
1
+ (
2
21
y
1
+ +y
1
)

2
y
2
2

f = 0.
(ii)
76
Proof. If we suppose f(t, y
1
, y
2
) = e
y1C1(Tt)y2C2(Tt)A(Tt)
, then

t
f = [y
1
C

1
(T t) + y
2
C

2
(T t) +
A

(T t)]f,

y1
f = C
1
(T t)f,
f
y2
= C
2
(T t)f,

2
f
y1y2
= C
1
(T t)C
2
(T t)f,

2
f
y
2
1
= C
2
1
(T t)f,
and

2
f
y
2
2
= C
2
2
(T t)f. So the PDE in part (i) becomes
(
0
+
1
y
1
+
2
y
2
)+y
1
C

1
+y
2
C

2
+A

(
1
y
1
)C
1
+
2
y
2
C
2
+
1
2

2
21
y
1
C
1
C
2
+y
1
C
2
1
+ (
2
21
y
1
+ +y
1
)C
2
2

= 0.
Sorting out the LHS according to the independent variables y
1
and y
2
, we get

1
+C

1
+
1
C
1
+
21
C
1
C
2
+
1
2
C
2
1
+
1
2
(
2
21
+)C
2
2
= 0

2
+C

2
+
2
C
2
= 0

0
+A

C
1
+
1
2
C
2
2
= 0.
In other words, we can obtain the ODEs for C
1
, C
2
and A as follows

1
=
1
C
1

21
C
1
C
2

1
2
C
2
1

1
2
(
2
21
+)C
2
2
+
1
dierent from (10.7.4), check!
C

2
=
2
C
2
+
2
A

= C
1

1
2
C
2
2
+
0
.
10.3. (i)
Proof. d(D
t
B(t, T)) = D
t
[R
t
f(t, T, Y
1
(t), Y
2
(t))dt +df(t, T, Y
1
(t), Y
2
(t))] and
df(t, T, Y
1
(t), Y
2
(t))
= [f
t
(t, T, Y
1
(t), Y
2
(t)) +f
y1
(t, T, Y
1
(t), Y
2
(t))(
1
Y
1
(t)) +f
y2
(t, T, Y
1
(t), Y
2
(t))(
21
Y
1
(t)
2
Y
2
(t))
+
1
2
f
y1y1
(t, T, Y
1
(t), Y
2
(t)) +
1
2
f
y2y2
(t, T, Y
1
(t), Y
2
(t))]dt + martingale part.
Since D
t
B(t, T) is a martingale under risk-neutral measure, we have the following PDE:

(
0
(t) +
1
y
1
+
2
y
2
) +

t

1
y
1

y
1
(
21
y
1
+
2
y
2
)

y
2
+
1
2

2
y
2
1
+
1
2

y
2
2

f(t, T, y
1
, y
2
) = 0.
Suppose f(t, T, y
1
, y
2
) = e
y1C1(t,T)y2C2(t,T)A(t,T)
, then

f
t
(t, T, y
1
, y
2
) =

y
1
d
dt
C
1
(t, T) y
2
d
dt
C
2
(t, T)
d
dt
A(t, T)

f(t, T, y
1
, y
2
),
f
y1
(t, T, y
1
, y
2
) = C
1
(t, T)f(t, T, y
1
, y
2
),
f
y2
(t, T, y
1
, y
2
) = C
2
(t, T)f(t, T, y
1
, y
2
),
f
y1y2
(t, T, y
1
, y
2
) = C
1
(t, T)C
2
(t, T)f(t, T, y
1
, y
2
),
f
y1y1
(t, T, y
1
, y
2
) = C
2
1
(t, T)f(t, T, y
1
, y
2
),
f
y2y2
(t, T, y
1
, y
2
) = C
2
2
(t, T)f(t, T, y
1
, y
2
).
So the PDE becomes
(
0
(t) +
1
y
1
+
2
y
2
) +

y
1
d
dt
C
1
(t, T) y
2
d
dt
C
2
(t, T)
d
dt
A(t, T)

+
1
y
1
C
1
(t, T)
+(
21
y
1
+
2
y
2
)C
2
(t, T) +
1
2
C
2
1
(t, T) +
1
2
C
2
2
(t, T) = 0.
77
Sorting out the terms according to independent variables y
1
and y
2
, we get

0
(t)
d
dt
A(t, T) +
1
2
C
2
1
(t, T) +
1
2
C
2
2
(t, T) = 0

1

d
dt
C
1
(t, T) +
1
C
1
(t, T) +
21
C
2
(t, T) = 0

2

d
dt
C
2
(t, T) +
2
C
2
(t, T) = 0.
That is

d
dt
C
1
(t, T) =
1
C
1
(t, T) +
21
C
2
(t, T)
1
d
dt
C
2
(t, T) =
2
C
2
(t, T)
2
d
dt
A(t, T) =
1
2
C
2
1
(t, T) +
1
2
C
2
2
(t, T)
0
(t).
(ii)
Proof. For C
2
, we note
d
dt
[e
2t
C
2
(t, T)] = e
2t

2
from the ODE in (i). Integrate from t to T, we have
0 e
2t
C
2
(t, T) =
2

T
t
e
2s
ds =
2
2
(e
2T
e
2t
). So C
2
(t, T) =
2
2
(1 e
2(Tt)
). For C
1
, we note
d
dt
(e
1t
C
1
(t, T)) = (
21
C
2
(t, T)
1
)e
1t
=

21

2
(e
1t
e
2T+(21)t
)
1
e
1t
.
Integrate from t to T, we get
e
1t
C
1
(t, T)
=

212
21
(e
1T
e
1t
)
212
2
e
2T e
(
2

1
)T
e
(
2

1
)t
21
+
1
1
(e
1T
e
1T
) if
1
=
2

212
21
(e
1T
e
1t
)
212
2
e
2T
(T t) +
1
1
(e
1T
e
1T
) if
1
=
2
.
So
C
1
(t, T) =

212
21
(e
1(Tt)
1) +
212
2
e

1
(Tt)
e

2
(Tt)
21

1
1
(e
1(Tt)
1) if
1
=
2
212
21
(e
1(Tt)
1) +
212
2
e
2T+1t
(T t)
1
1
(e
1(Tt)
1) if
1
=
2
.
.
(iii)
Proof. From the ODE
d
dt
A(t, T) =
1
2
(C
2
1
(t, T) +C
2
2
(t, T))
0
(t), we get
A(t, T) =

T
t

0
(s)
1
2
(C
2
1
(s, T) +C
2
2
(s, T))

ds.
(iv)
Proof. We want to nd
0
so that f(0, T, Y
1
(0), Y
2
(0)) = e
Y1(0)C1(0,T)Y2(0)C2(0,T)A(0,T)
= B(0, T) for all
T > 0. Take logarithm on both sides and plug in the expression of A(t, T), we get
log B(0, T) = Y
1
(0)C
1
(0, T) Y
2
(0)C
2
(0, T) +

T
0

1
2
(C
2
1
(s, T) +C
2
2
(s, T))
0
(s)

ds.
Taking derivative w.r.t. T, we have

T
log B(0, T) = Y
1
(0)

T
C
1
(0, T) Y
2
(0)

T
C
2
(0, T) +
1
2
C
2
1
(T, T) +
1
2
C
2
2
(T, T)
0
(T).
78
So

0
(T) = Y
1
(0)

T
C
1
(0, T) Y
2
(0)

T
C
2
(0, T)

T
log B(0, T)
=

Y
1
(0)

1
e
1T

212
2
e
2T

Y
2
(0)
2
e
2T


T
log B(0, T) if
1
=
2
Y
1
(0)

1
e
1T

21

2
e
2T
T

Y
2
(0)
2
e
2T


T
log B(0, T) if
1
=
2
.
10.4. (i)
Proof.
d

X
t
= dX
t
+Ke
Kt

t
0
e
Ku
(u)dudt (t)dt
= KX
t
dt + d

B
t
+Ke
Kt

t
0
e
Ku
(u)dudt
= K

X
t
dt + d

B
t
.
(ii)
Proof.

W
t
= C

B
t
=

1
1
0

1
2
1
2

1
2

1
0
0
2


B
t
=

1 0

1
2
1

1
2


B
t
.
So

W is a martingale with '

W
1
`
t
= '

B
1
`
t
= t, '

W
2
`
t
= '

1
2

B
1
+
1

1
2

B
2
`
t
=

2
t
1
2
+
t
1
2
2

1
2
t =

2
+12
2
1
2
t = t, and '

W
1
,

W
2
`
t
= '

B
1
,

1
2

B
1
+
1

1
2

B
2
`
t
=
t

1
2
+
t

1
2
= 0. Therefore

W is a
two-dimensional BM. Moreover, dY
t
= Cd

X
t
= CK

X
t
dt+Cd

B
t
= CKC
1
Y
t
dt+d

W
t
= Y
t
dt+d

W
t
,
where
= CKC
1
=

1
1
0

1
2
1
2

1
2

1
0
1
2

1
[C[

1
2

1
2
0

1
2
1
1

1
1
0

1
1

1
2

1
2

1
2
2
2

1
2

1
0

2

2

1
2


1
0
2(21)1
2

1
2

2

.
(iii)
Proof.
X
t
=

X
t
+e
Kt

t
0
e
Ku
(u)du = C
1
Y
t
+e
Kt

t
0
e
Ku
(u)du
=

1
0

2

2

1
2

Y
1
(t)
Y
2
(t)

+e
Kt

t
0
e
Ku
(u)du
=


1
Y
1
(t)

2
Y
1
(t) +
2

1
2
Y
2
(t)

+e
Kt

t
0
e
Ku
(u)du.
79
So R
t
= X
2
(t) =
2
Y
1
(t)+
2

1
2
Y
2
(t)+
0
(t), where
0
(t) is the second coordinate of e
Kt

t
0
e
Ku
(u)du
and can be derived explicitly by Lemma 10.2.3. Then
1
=
2
and
2
=
2

1
2
.
10.5.
Proof. We note C(t, T) and A(t, T) are dependent only on T t. So C(t, t + ) and A(t, t + ) aare constants
when is xed. So
d
dt
L
t
=
B(t, t + )[C(t, t + )R

(t) A(t, t + )]
B(t, t + )
=
1

[C(t, t + )R

(t) +A(t, t + )]
=
1

[C(0, )R

(t) +A(0, )].


Hence L(t
2
)L(t
1
) =
1

C(0, )[R(t
2
)R(t
1
)]+
1

A(0, )(t
2
t
1
). Since L(t
2
)L(t
1
) is a linear transformation,
it is easy to verify that their correlation is 1.
10.6. (i)
Proof. If
2
= 0, then dR
t
=
1
dY
1
(t) =
1
(
1
Y
1
(t)dt +d

W
1
(t)) =
1

(
0
1

Rt
1
)
1
dt +d

W
1
(t)

= (
0

1
R
t
)dt +
1
d

W
1
(t). So a =
0

1
and b =
1
.
(ii)
Proof.
dR
t
=
1
dY
1
(t) +
2
dY
2
(t)
=
1

1
Y
1
(t)dt +
1
d

W
1
(t)
2

21
Y
1
(t)dt
2

2
Y
2
(t)dt +
2
d

W
2
(t)
= Y
1
(t)(
1

1
+
2

21
)dt
2

2
Y
2
(t)dt +
1
d

W
1
(t) +
2
d

W
2
(t)
= Y
1
(t)
2

1
dt
2

2
Y
2
(t)dt +
1
d

W
1
(t) +
2
d

W
2
(t)
=
2
(Y
1
(t)
1
+Y
2
(t)
2
)dt +
1
d

W
1
(t) +
2
d

W
2
(t)
=
2
(R
t

0
)dt +

2
1
+
2
2

2
1
+
2
2
d

W
1
(t) +

2

2
1
+
2
2
d

W
2
(t)

.
So a =
2

0
, b =
2
, =

2
1
+
2
2
and

B
t
=
1

2
1
+
2
2

W
1
(t) +
2

2
1
+
2
2

W
2
(t).
10.7. (i)
Proof. We use the canonical form of the model as in formulas (10.2.4)-(10.2.6). By (10.2.20),
dB(t, T) = df(t, Y
1
(t), Y
2
(t))
= de
Y1(t)C1(Tt)Y2(t)C2(Tt)A(Tt)
= dt term +B(t, T)[C
1
(T t)d

W
1
(t) C
2
(T t)d

W
2
(t)]
= dt term +B(t, T)(C
1
(T t), C
2
(T t))

W
1
(t)
d

W
2
(t)

.
So the volatility vector of B(t, T) under

P is (C
1
(T t), C
2
(T t)). By (9.2.5),

W
T
j
(t) =

t
0
C
j
(T
u)du +

W
j
(t) (j = 1, 2) form a two-dimensional

P
T
BM.
(ii)
80
Proof. Under the T-forward measure, the numeraire is B(t, T). By risk-neutral pricing, at time zero the
risk-neutral price V
0
of the option satises
V
0
B(0, T)
=

E
T

1
B(T, T)
(e
C1(

TT)Y1(T)C2(

TT)Y2(T)A(

TT)
K)
+

.
Note B(T, T) = 1, we get (10.7.19).
(iii)
Proof. We can rewrite (10.2.4) and (10.2.5) as

dY
1
(t) =
1
Y
1
(t)dt +d

W
T
1
(t) C
1
(T t)dt
dY
2
(t) =
21
Y
1
(t)dt
2
Y
2
(t)dt +d

W
T
2
(t) C
2
(T t)dt.
Then

Y
1
(t) = Y
1
(0)e
1t
+

t
0
e
1(st)
d

W
T
1
(s)

t
0
C
1
(T s)e
1(st)
ds
Y
2
(t) = Y
0
e
2t

21

t
0
Y
1
(s)e
2(st)
ds +

t
0
e
2(st)
d

W
2
(s)

t
0
C
2
(T s)e
2(st)
ds.
So (Y
1
, Y
2
) is jointly Gaussian and X is therefore Gaussian.
(iv)
Proof. First, we recall the Black-Scholes formula for call options: if dS
t
= S
t
dt +S
t
d

W
t
, then

E[e
T
(S
0
e
W
T
+(
1
2

2
)T
K)
+
] = S
0
N(d
+
) Ke
T
N(d

)
with d

=
1

T
(log
S0
K
+(
1
2

2
)T). Let T = 1, S
0
= 1 and = W
1
+(
1
2

2
), then
d
= N(
1
2

2
,
2
)
and

E[(e

K)
+
] = e

N(d
+
) KN(d

),
where d

=
1

(log K + (
1
2

2
)) (dierent from the problem. Check!). Since under

P
T
, X
d
=
N(
1
2

2
,
2
), we have
B(0, T)

E
T
[(e
X
K)
+
] = B(0, T)(e

N(d
+
) KN(d

)).
10.11.
Proof. On each payment date T
j
, the payo of this swap contract is (K L(T
j1
, T
j1
)). Its no-arbitrage
price at time 0 is (KB(0, T
j
) B(0, T
j
)L(0, T
j1
)) by Theorem 10.4. So the value of the swap is
n+1

j=1
[KB(0, T
j
) B(0, T
j
)L(0, T
j1
)] = K
n+1

j=1
B(0, T
j
)
n+1

j=1
B(0, T
j
)L(0, T
j1
).
10.12.
81
Proof. Since L(T, T) =
1B(T,T+)
B(T,T+)
T
T
, we have

E[D(T +)L(T, T)] =



E[

E[D(T +)L(T, T)[T


T
]]
=

E

1 B(T, T +)
B(T, T +)

E[D(T +)[T
T
]

=

E

1 B(T, T +)
B(T, T +)
D(T)B(T, T +)

=

E

D(T) D(T)B(T, T +)

=
B(0, T) B(0, T +)

= B(0, T +)L(0, T).


11. Introduction to Jump Processes
11.1. (i)
Proof. First, M
2
t
= N
2
t
2tN
t
+
2
t
2
. So E[M
2
t
] < . f(x) = x
2
is a convex function. So by conditional
Jensens inequality,
E[f(M
t
)[T
s
] f(E[M
t
[T
s
]) = f(M
s
), s t.
So M
2
t
is a submartingale.
(ii)
Proof. We note M
t
has independent and stationary increment. So s t, E[M
2
t
M
2
s
[T
s
] = E[(M
t

M
s
)
2
[T
s
] + E[(M
t
M
s
) 2M
s
[T
s
] = E[M
2
ts
] + 2M
s
E[M
ts
] = V ar(N
ts
) + 0 = (t s). That is,
E[M
2
t
t[T
s
] = M
2
s
s.
11.2.
Proof. P(N
s+t
= k[N
s
= k) = P(N
s+t
N
s
= 0[N
s
= k) = P(N
t
= 0) = e
t
= 1 t + O(t
2
). Similarly,
we have P(N
s+t
= k + 1[N
s
= k) = P(N
t
= 1) =
(t)
1
1!
e
t
= t(1 t + O(t
2
)) = t + O(t
2
), and
P(N
s+t
k + 2[N
2
= k) = P(N
t
2) =

k=2
(t)
k
k!
e
t
= O(t
2
).
11.3.
Proof. For any t u, we have
E

S
u
S
t

T
t

= E[( + 1)
NtNu
e
(tu)
[T
t
]
= e
(tu)
E[( + 1)
Ntu
]
= e
(tu)
E[e
Ntu log(+1)
]
= e
(tu)
e
(tu)(e
log(+1)
1)
(by (11.3.4))
= e
(tu)
e
(tu)
= 1.
So S
t
= E[S
u
[T
t
] and S is a martingale.
11.4.
82
Proof. The problem is ambiguous in that the relation between N
1
and N
2
is not clearly stated. According
to page 524, paragraph 2, we would guess the condition should be that N
1
and N
2
are independent.
Suppose N
1
and N
2
are independent. Dene M
1
(t) = N
1
(t)
1
t and M
2
(t) = N
2
(t)
2
t. Then by
independence E[M
1
(t)M
2
(t)] = E[M
1
(t)]E[M
2
(t)] = 0. Meanwhile, by Itos product formula, M
1
(t)M
2
(t) =

t
0
M
1
(s)dM
2
(s) +

t
0
M
2
(s)dM
1
(s) +[M
1
, M
2
]
t
. Both

t
0
M
1
(s)dM
2
(s) and

t
0
M
2
(s)dM
1
(s) are mar-
tingales. So taking expectation on both sides, we get 0 = 0 + E[M
1
, M
2
]
t
= E[

0<st
N
1
(s)N
2
(s)].
Since

0<st
N
1
(s)N
2
(s) 0 a.s., we conclude

0<st
N
1
(s)N
2
(s) = 0 a.s. By letting t = 1, 2, ,
we can nd a set
0
of probability 1, so that
0
,

0<st
N
1
(s)N
2
(s) = 0 for all t > 0. Therefore
N
1
and N
2
can have no simultaneous jump.
11.5.
Proof. We shall prove the whole path of N
1
is independent of the whole path of N
2
, following the scheme
suggested by page 489, paragraph 1.
Fix s 0, we consider X
t
= u
1
(N
1
(t)N
1
(s))+u
2
(N
2
(t)N
2
(s))
1
(e
u1
1)(ts)
2
(e
u2
1)(ts),
t > s. Then by Itos formula for jump process, we have
e
Xt
e
Xs
=

t
s
e
Xu
dX
c
u
+
1
2

t
s
e
Xu
dX
c
u
dX
c
u
+

s<ut
(e
Xu
e
Xu
)
=

t
s
e
Xu
[
1
(e
u1
1)
2
(e
u2
1)]du +

0<ut
(e
Xu
e
Xu
).
Since X
t
= u
1
N
1
(t)+u
2
N
2
(t) and N
1
, N
2
have no simultaneous jump, e
Xu
e
Xu
= e
Xu
(e
Xu
1) =
e
Xu
[(e
u1
1)N
1
(u) + (e
u2
1)N
2
(u)]. So
e
Xt
1
=

t
s
e
Xu
[
1
(e
u1
1)
2
(e
u2
1)]du +

s<ut
e
Xu
[(e
u1
1)N
1
(u) + (e
u2
1)N
2
(u)]
=

t
s
e
Xu
[(e
u1
1)d(N
1
(u)
1
u) (e
u2
1)d(N
2
(u)
2
u)].
This shows (e
Xt
)
ts
is a martingale w.r.t. (T
t
)
ts
. So E[e
Xt
] 1, i.e.
E[e
u1(N1(t)N1(s))+u2(N2(t)N2(s))
] = e
1(e
u
1
1)(ts)
e
2(e
u
2
1)(ts)
= E[e
u1(N1(t)N1(s))
]E[e
u2(N2(t)N2(s))
].
This shows N
1
(t) N
1
(s) is independent of N
2
(t) N
2
(s).
Now, suppose we have 0 t
1
< t
2
< t
3
< < t
n
, then the vector (N
1
(t
1
), , N
1
(t
n
)) is independent
of (N
2
(t
1
), , N
2
(t
n
)) if and only if (N
1
(t
1
), N
1
(t
2
) N
1
(t
1
), , N
1
(t
n
) N
1
(t
n1
)) is independent of
(N
2
(t
1
), N
2
(t
2
) N
2
(t
1
), , N
2
(t
n
) N
2
(t
n1
)). Let t
0
= 0, then
E[e

n
i=1
ui(N1(ti)N1(ti1))+

n
j=1
vj(N2(tj)N2(tj1))
]
= E[e

n1
i=1
ui(N1(ti)N1(ti1))+

n1
j=1
vj(N2(tj)N2(tj1))
E[e
un(N1(tn)N1(tn1))+vn(N2(tn)N2(tn1))
[T
tn1
]]
= E[e

n1
i=1
ui(N1(ti)N1(ti1))+

n1
j=1
vj(N2(tj)N2(tj1))
]E[e
un(N1(tn)N1(tn1))+vn(N2(tn)N2(tn1))
]
= E[e

n1
i=1
ui(N1(ti)N1(ti1))+

n1
j=1
vj(N2(tj)N2(tj1))
]E[e
un(N1(tn)N1(tn1))
]E[e
vn(N2(tn)N2(tn1))
],
where the second equality comes from the independence of N
i
(t
n
) N
i
(t
n1
) (i = 1, 2) relative to T
tn1
and
the third equality comes from the result obtained in the above paragraph. Working by induction, we have
E[e

n
i=1
ui(N1(ti)N1(ti1))+

n
j=1
vj(N2(tj)N2(tj1))
]
=
n

i=1
E[e
ui(N1(ti)N1(ti1))
]
n

j=1
E[e
vj(N2(tj)N2(tj1))
]
= E[e

n
i=1
ui(N1(ti)N1(ti1))
]E[e

n
j=1
vj(N2(tj)N2(tj1))
].
83
This shows the whole path of N
1
is independent of the whole path of N
2
.
11.6.
Proof. Let X
t
= u
1
W
t

1
2
u
2
1
t +u
2
Q
t
t((u
2
) 1) where is the moment generating function of the jump
size Y . Itos formula for jump process yields
e
Xt
1 =

t
0
e
Xs
(u
1
dW
s

1
2
u
2
1
ds ((u
2
) 1)ds) +
1
2

t
0
e
Xs
u
2
1
ds +

0<st
(e
Xs
e
Xs
).
Note X
t
= u
2
Q
t
= u
2
Y
Nt
N
t
, where N
t
is the Poisson process associated with Q
t
. So e
Xt
e
Xt
=
e
Xt
(e
Xt
1) = e
Xt
(e
u2Y
N
t
1)N
t
. Consider the compound Poisson process H
t
=

Nt
i=1
(e
u2Yi
1),
then H
t
E[e
u2Y
N
t
1]t = H
t
((u
2
) 1)t is a martingale, e
Xt
e
Xt
= e
Xt
H
t
and
e
Xt
1 =

t
0
e
Xs
(u
1
dW
s

1
2
u
2
1
ds ((u
2
) 1)ds) +
1
2

t
0
e
Xs
u
2
1
ds +

t
0
e
Xs
dH
s
=

t
0
e
Xs
u
1
dW
s
+

t
0
e
Xs
d(H
s
((u
2
) 1)s).
This shows e
Xt
is a martingale and E[e
Xt
] 1. So E[e
u1Wt+u2Qt
] = e
1
2
u1t
e
t((u2)1)t
= E[e
u1Wt
]E[e
u2Qt
].
This shows W
t
and Q
t
are independent.
11.7.
Proof. E[h(Q
T
)[T
t
] = E[h(Q
T
Q
t
+Q
t
)[T
t
] = E[h(Q
Tt
+x)][
x=Qt
= g(t, Q
t
), where g(t, x) = E[h(Q
Tt
+
x)].
References
[1] F. Delbaen and W. Schachermayer. The mathematics of arbitrage. Springer, 2006.
[2] J. Hull. Options, futures, and other derivatives. Fourth Edition. Prentice-Hall International Inc., New
Jersey, 2000.
[3] B. ksendal. Stochastic dierential equations: An introduction with applications. Sixth edition. Springer-
Verlag, Berlin, 2003.
[4] D. Revuz and M. Yor. Continous martingales and Brownian motion. Third edition. Springer-Verlag,
Berline, 1998.
[5] A. N. Shiryaev. Essentials of stochastic nance: facts, models, theory. World Scientic, Singapore, 1999.
[6] S. Shreve. Stochastic calculus for nance I. The binomial asset pricing model. Springer-Verlag, New
York, 2004.
[7] S. Shreve. Stochastic calculus for nance II. Continuous-time models. Springer-Verlag, New York, 2004.
[8] P. Wilmott. The mathematics of nancial derivatives: A student introduction. Cambridge University
Press, 1995.
84