Chapter 4

The Black-Scholes Model

4.1 Continuous Time Models
So far all of our theoretical developments have been for discrete values of both
time and equity values. This has been very helpful for gaining understanding
and insight, however it is more realistic to deal with continuous time and real-
valued processes. From now on time will be modelled by using the interval
[0, T], where T is the terminal date for all relevant economic activity. For
simplicity we will work with a model which only involves two nancial assets.
These will be
A risk-free security (a bond or bank account) whose value at time t is
A(t).
1
We assume that the principal A(0) is invested at a xed interest
rate r > 0 and so the formula for continuous compound interest yields
A(t) = A(0)e
rt
.
Of course
dA(t)
dt
= rA(0)e
rt
= rA(t) and we will often write this Ito-
style as dA(t) = rA(t)dt.
For simplicity, well assume that A(0) = 1.
A stock whose value at time t is S(t).
We must model S = (S(t), t 0) as a stochastic process and so we need
a probability space (, F, P) and a ltration (F
t
, 0 t T). Intuitively the
-algebra F
t
should contain all information about the stock price behaviour
up to and including time t. We always take F
0
to be the trivial -algebra.
1
We are using A(t) here in order to reserve B(t) for Brownian motion.
25
The question of how to model S is a fundamental one. Financial analysts
like to study the return on their investment and over a small time period (t)
this is
S(t)
S(t)
where S(t) = S(t + t) S(t). If S behaves like A, we would
write
S(t)
S(t)
= t, where R. Of course no stock price behaves in this
way. We need to include a factor which describes the random behaviour of
stock prices and so we introduce an adapted process (Y (t), 0 t T) and
write
S(t)
S(t)
= t +Y (t),
so that
S(t) = S(t)t +S(t)Y (t). (4.1.1)
There is still debate as to which is the best choice for the process
Y . In the Black-Scholes model we choose Y (t) = B(t). Here > 0 and
B = (B(t), 0 t T) is a Brownian motion which is adapted to the
ltration (F
t
, t 0) and is such that B(t) B(s) is independent of F
s
whenever 0 s < t T.
2
In this model we take the formal limit of (4.1.1)
as t 0 and interpret the result as a stochastic dierential equation (SDE)
in the Ito sense. We then obtain
dS(t) = S(t)dt +S(t)dB(t), (4.1.2)
with initial condition S(0) = S
0
which is the stock price when the investor
enters the market. S = (S(t), 0 t T) is then an adapted process. In fact
(as was shown in Chapter 7 of PAS 401/6052) the unique solution to (4.1.2)
is geometric Brownian motion:
S(t) = S
0
exp
_
B(t) +
_

1
2

2
_
t
_
, (4.1.3)
for all 0 t T.
We also saw in PAS367/6051 that this process is obtained in the binomial
asset pricing model when the interval [0, T] is split into n subintervals of equal
length and we take the limit as n .
The parameter is called the stock drift. It measures the logarithmic rate
of return of the stock in the absence of noise. The parameter is called the
volatility. It measures the strength of the random uctuations in the stock
price.
2
These requirements are automatically satised when (F
t
, t 0) is the natural ltration
of B.
26
For the rest of this chapter, it will be convenient to take (F
t
, 0 t T)
to be the natural ltration generated by the Brownian motion.
4.2 Pricing and Hedging in the Black-Scholes
Model
4.2.1 Basic Concepts
We will generalise key denitions from the nite market model in an obvious
way.
Denition. A contingent claim X is a F
T
-measurable random variable.
If X = f(S(T)) for some measurable function f then X is called a European
contingent claim or ECC e.g. a European put option X = (kS(T))
+
where
k is the exercise price.
Denition. A portfolio is a pair (, ) of (real-valued) adapted processes
= ((t), 0 t T) and = ((t), 0 t T). We interpret (t) as the
number of stocks held at time t and (t) as the number of bonds held at
time t. The value of the portfolio at time t is V (t) where
V (t) = (t)A(t) +(t)S(t),
and it is easily veried that V = (V (t), 0 t T) is an adapted process.
Given a contingent claim X, the portfolio (, ) is said to be replicating if
V (T) = X. A portfolio is self-nancing if
dV (t) = (t)dA(t) +(t)dS(t). (4.2.4)
Of course the dierentials in (4.2.4) are to be interpreted in the Ito sense.
The economic interpretation of (4.2.4) is (as in the nite market case) that all
changes in the wealth of the portfolio are due entirely to changes in the values
of the underlying assets and there are no external injections (or removals) of
wealth. (4.2.4) is quite general and applied to any model of the market in
which A and S are processes having stochastic dierentials. In our case we
can combine (4.2.4) with (4.1.1) and the compound interest formula to get
dV (t) = (r(t)A(t) +(t)S(t))dt +(t)S(t)dB(t). (4.2.5)
This formula (4.2.5) turns out to be of limited value. In fact it is more
important to nd the relationship between the discounted wealth process

V
and the discounted stock process

S where each

V (t) = A(t)
1
V (t) = e
rt
V (t)
and

S(t) = e
rt
S(t). You can explore this for yourself in exercise 21.
27
In the nite market model (discrete time setting), the key to pricing and
hedging options were the fundamental theorems of asset pricing. The exis-
tence of self-nancing replicating portfolios was intimately related to nding
an equivalent measure Q under which the discounted stock process

S was a
Q-martingale. (In our continuous time setting you can check for yourself in
Exercise 21 that

S is quite far from being a P-martingale, except in a very
special case.) There are generalisations of the fundamental theorems to con-
tinuous time but these are more dicult to set-up and prove and they require
additional assumptions such us the NFLVR hypothesis - no free lunch with
vanishing risk. Details can be found in Chapter 6 of Bingham and Keisel.
We will follow the philosophy of the fundamental theorems and work in a
concrete setting. Our rst job is to nd candidates for the measure Q under
which

S will be a martingale.
4.2.2 Change of Measure
Recall from Chapter 6 of PAS401/6052 that if F P
2
(T) then its stochastic
exponential is the process E
F
= (E
F
(t), 0 t T) where
E
F
(t) = exp
__
t
0
F(s)dB(s)
1
2
_
t
0
F(s)
2
ds
_
, (4.2.6)
and this process satises the SDE
dE
F
(t) = F(t)E
F
(t)dB(t).
Furthermore, we may dene a new probability measure Q on (, F) which
is equivalent to P and which has Radon-Nikod ym derivative:
dQ
dP
= E
F
(T).
Girsanovs theorem then tells us that (W(t), 0 t T) is a Brownian motion
on (, F, Q) where for each 0 t T,
W(t) = B(t)
_
t
0
F(s)ds.
The key result is the following one:
Theorem 4.2.1 There exists an equivalent probability measure Q under which
(

S(t), 0 t T) is a martingale. It is obtained by taking F(t) =

r

in
(4.2.6) for all 0 t T. We then have
d

S(t) =

S(t)dW(t). (4.2.7)
28
Proof. First note that if F(t) =
r

then dW(t) = dB(t)

_
r

_
dt.
By exercise 21,
d

S(t) = ( r)

S(t)dt +

S(t)dB(t)
= ( r)

S(t)dt +

S(t)dW(t) + (r )

S(t)dt
=

S(t)dW(t).
So we have derived (4.2.7). Since W is a Q-Brownian motion, it follows that

S is a Q-martingale, indeed

S is a Q-geometric Brownian motion

S(t) = S(0) exp

_
W(t)
1
2

2
t
_
.
4.2.3 The Claim Martingale
The next step in the Black-Scholes theory is to obtain a martingale from the
contingent claim X. In fact we need a constraint at this stage - we assume
that X is square-integrable with respect to the measure Q, i.e. E
Q
(X
2
) < .
We will turn the random variable X into a Q-martingale in a two stage
process. First discount so X becomes e
rT
X and then condition to dene

X(t) = e
rT
E
Q
(X|F
t
).
Lemma 4.2.1

X = (

X(t), 0 t T) is a square-integrable Q-martingale.

The process (

X(t), 0 t T) is clearly adapted. It satises the martingale

property by (CE4) indeed if 0 s t T

X(s) = E
Q
(e
rT
X|F
s
) = E
Q
(E
Q
(e
rT
X|F
t
)|F
s
) = E
Q
(

X(t)|F
s
).
Square-integrability follows from (CE8) (the conditional Jensen inequality).
Indeed when we apply this and then use (CE5) we get for all 0 t T,
E
Q
(

X(t)
2
) = e
2rT
E
Q
(E
Q
(X|F
t
)
2
)
e
2rT
E
Q
(E
Q
(X
2
|F
t
))
= e
2rT
E
Q
(X
2
) < .
As the process

X is square-integrable, it is automatically integrable (see
Exercise 4 of PAS 401/6052).
We call the process

X the claim martingale. As it is square-integrable we
can apply the martingale representation theorem (see Chapter 6 of PAS401/6052)
29
in the probability space (, F, Q) to deduce that there exists a process
= ((t), 0 t T) in H
2
(T) such that for each 0 t T,

X(t) =

X(0) +
_
t
0
(u)dW(u)
=

X(0) +
_
t
0
(u)d

S(u), (4.2.8)
where (t) =
(t)

S(t)
and we have used (4.2.7). We can rewrite (4.2.8) in
stochastic dierential form to obtain
d

X(t) = (t)d

S(t). (4.2.9)
As we will see the process plays a vital role in constructing the self-
nancing replicating portfolio which perfectly hedges the option.
4.2.4 The Black-Scholes Portfolio
In this section, we will draw together the ideas of the previous two sections
to show how to hedge and price an arbitrary contingent claim X. We dene
the Black-Scholes portfolio by
(t) = (t) ; (t) =

X(t) (t)

S(t), (4.2.10)
for all 0 s t. Its value at time t is
V (t) = (t)S(t) + (

X(t) (t)

S(t))A(t)
= (t)A(t)

S(t) + (

X(t) (t)

S(t))A(t)
=

X(t)A(t). (4.2.11)
Theorem 4.2.2 The Black-Scholes portfolio is self-nancing and replicat-
ing.
Proof. The portfolio is replicating since by (4.2.11)
V (T) = A(T)

X(T)
= e
rT
.e
rT
E
Q
(X|F
T
)
= X, (4.2.12)
where weve used the fact that X is F
T
-measurable.
30
To see that the portfolio is self-nancing, we nd the stochastic dierential
of (4.2.11) using Itos product formula and apply (4.2.9) and (4.2.10) to
obtain
dV (t) =

X(t)dA(t) +A(t)d

X(t)
= ((t) +(t)

S(t))dA(t) +(t)A(t)d

S(t)
= (t)dA(t) +(t)[

S(t)dA(t) +A(t)d

S(t)]
= (t)dA(t) +(t)d[A(t)

S(t)]
= (t)dA(t) +(t)dS(t).
Applying the usual arguments (c.f. Exercise 14) we deduce that the
arbitrage price of the option at time t is
V (t) = A(t)

X(t)
= e
rt
E
Q
(e
rT
X|F
t
)
= e
r(Tt)
E
Q
(X|F
t
), (4.2.13)
i.e. to attempt to sell the option at this time at a greater or lower price
creates arbitrage opportunities. In particular the price of the option at time
zero is
V (0) = e
rT
E
Q
(X). (4.2.14)
You should compare (4.2.13) with the discrete time formula (2.2.1).
4.2.5 Pricing a European Call Option
We begin by considering the case of a general ECC - so X = f(S(T)) where
f is a Borel measurable function from [0, ) to R. Then (4.2.14) takes the
form
V (0) = e
rT
E
Q
(f(S(T))).
Since

S is a geometric Brownian motion (see (4.2.7) and the formula at the
end of the proof of theorem 4.2.1) we have
S(T) = A(T)

S(T)
= e
rT
S(0) exp
_
W(T)
1
2

2
T
_
= S(0)e
U+rT
,
31
where U = W(T)
1
2

2
T. Since (under the measure Q) W(T) N(0, T), it
follows that U N
_

1
2

2
T,
2
T
_
. Using the convenient notation S(0) = s,
we thus obtain
V (0) =
e
rT

2T
_

f(se
x+rT
) exp
_

_
x +
1
2

2
T
_
2
2
2
T
_
dx. (4.2.15)
This is the Black-Scholes pricing formula for a general European contin-
gent claim.
Now consider the case of a European call option. In this case f(x) =
(x k)
+
. In particular, f(se
x+rT
) = max{se
x+rT
k, 0} = 0 if and only if
x > log
_
k
s
_
rT. Pulling the factor of e
rT
through the integral in (4.2.15)
we then obtain
V (0) =
1

2T
_

log
(
k
s
)
rT
(se
x
ke
rT
) exp
_

_
x +
1
2

2
T
_
2
2
2
T
_
dx. (4.2.16)
Let be the cdf of a standard normal, i.e. (z) =
1

2
_
z

1
2
u
2
du, then
in Exercise 24, you can check that (4.2.16) can be rearranged to yield
V (0) = s
_
_
log
_
s
k
_
+
_
r +

2
2
_
T

T
_
_
ke
rT

_
_
log
_
s
k
_
+
_
r

2
2
_
T

T
_
_
.
(4.2.17)
Equation (4.2.17) is the celebrated Black-Scholes pricing formula for a
European call option. It is often written more succinctly as
V (0) = s(d
1
) ke
rT
(d
2
), (4.2.18)
where d
1
=
log
_
s
k
_
+
_
r +

2
2
_
T

T
and d
2
= d
1

T.
It is interesting to observe that the price as given by (4.2.17) has no
dependence on the stock drift . It does however depend critically on the
value of the volatility .
4.3 The Black-Scholes PDE
4.3.1 Deriving the Black-Scholes PDE
In their celebrated paper which began the option pricing revolution
3
Black
and Scholes played down the use of Ito calculus (which they thought econo-
3
F. Black, M. Scholes, The pricing of options and corporate liabilities, Journal of
Political Economy 81, 637-59 (1973)
32
mists wouldnt understand) and expressed their main results in the language
of ordinary calculus, especially partial dierential equations (pdes). This ap-
proach is still important for hedging purposes as well see below. The deriva-
tion of the Black-Scholes pde is similar to the method we used in Chapter 7
of PAS401/6052 to associate a PDE to an SDE, but the argument is a little
bit more subtle as we deal with a terminal boundary value problem rather
than an initial value problem. Our claim X is an ECC of the form f(S(T)).
Before we start the analysis we should be clear about notation. If g is
a dierentiable function of both space and time variables we will use the
notation
g
x
(t, y) for the partial derivative
g
x
(t, x) evaluated at x = y, e.g. if
g(t, x) = e
t+x
2
,
g
x
(t, y) = 2ye
t+y
2
.
We begin with the SDE (4.2.7) for the discounted stock price. Since
S(t) = e
rt

S(t), we easily deduce

dS(t) = rS(t)dt +S(t)dW(t).
Now apply Itos formula for a function h C
1,2
to get
dh(t, S(t)) =
h
t
(t, S(t))dt +
h
x
(t, S(t))dS(t)
+
1
2

2
h
x
2
(t, S(t))(dS(t))
2
=
h
t
(t, S(t))dt +rS(t)
h
x
(t, S(t))dt
+ S(t)
h
x
(t, S(t))dW(t) +
1
2

2
S(t)
2

2
h
x
2
(t, S(t))dt
= (Lh)(t, S(t))dt +S(t)
h
x
(t, S(t))dW(t). (4.3.19)
Here weve introduced the useful notation
4
(Lh)(t, S(t)) =
h
t
(t, S(t)) +rS(t)
h
x
(t, S(t)) +
1
2

2
S(t)
2

2
h
x
2
(t, S(t)).
We integrate (4.3.19) from t to T, we obtain
h(T, S(T))h(t, S(t)) =
_
T
t
S(u)
h
x
(u, S(u))dW(u)+
_
T
t
(Lh)(u, S(u))du.
(4.3.20)
Now take conditional expectations E
Q
(|S(t) = s) of both sides of this
equation. The martingale property of stochastic integrals tells us that
E
Q
__
T
t
S(u)
h
x
(u, S(u))dW(u)

S(t) = s
_
= 0,
4
In fact L is a linear operator on the vector space C
1,2
.
33
and so we obtain
E
Q
(h(T, S(T))|S(t) = s) h(t, s) = E
Q
__
T
t
(Lh)(u, S(u))du|S(t) = s
_
.
(4.3.21)
Now suppose (a) that h has the property that (Lh)(t, x) = 0 for all
0 t T, x R, i.e.
h
t
(t, x) +rx
h
x
(t, x) +
1
2

2
x
2

2
h
x
2
(t, x) = 0, (4.3.22)
then (4.3.21) tells us that
h(t, s) = E
Q
(h(T, S(T))|S(t) = s).
Now suppose (b) that we have the terminal boundary condition h(T, S(T)) =
f(S(T)), then we get
h(t, s) = E
Q
(f(S(T))|S(t) = s),
which is quite close to the price of the option. To get the price as the solution
of a PDE, we need to introduce discounting. We do this in the following way.
Dene
F(t, s) = e
r(Tt)
h(t, s),
then F(t, s) = e
r(Tt)
E
Q
(f(S(T))|S(t) = s) is by (4.2.13) the price at time
t conditioned on the event (S(t) = s) F
t
. Moreover
F
t
(t, s) = rF(t, s) +e
r(Tt)
h
t
(t, s),
and substituting in (4.3.22) we deduce that
F
t
(t, s) rF(t, s) +rs
F
x
(t, s) +
1
2

2
s
2

2
F
x
2
(t, s) = 0. (4.3.23)
(4.3.23) is the Black-Scholes pde. We have established the important fact
that
The price of a ECC X = f(S(T)) at time t (given that S(t) = s) is
the (in fact, unique) solution F to the pde (4.3.23) subject to the terminal
boundary condition F(T, S(T)) = f(S(T)).
Our rst application of the pde (4.3.23) is to hedging. Recall that we
have constructed a portfolio V (t) = (t)A(t) +(t)S(t) where the process
has so far been found indirectly using the martingale representation theorem.
We would like to know a little bit more about . The following result gives
us some useful information
34
Theorem 4.3.1 For each 0 t T,
(t) =
F
x
(t, S(t))
Proof. Dene

F(t, x) = e
rt
F(t, xe
rt
). In exercise 27, you can apply
(4.3.23) to deduce that

F
t
(t, x) =
1
2

2
x
2

F
x
2
(t, x) . . . (i)
Applying Itos formula and using (4.2.7) we obtain,

F(T,

S(T)) F(0, S(0)) =
_
T
0

F
u
(u,

S(u))du
+
_
T
0

S(u)

F
x
(u,

S(u))dW(u) +
_
T
0
1
2

S(u)
2

F
x
2
(u,

S(u))du
=
_
T
0

F
x
(u,

S(u))

S(u)dW(u)
=
_
T
0

F
x
(u,

S(u))d

S(u),
where we have used (i) and (4.2.7) again. Since

V (t) =

F(t,

S(t)), weve
shown that

V (T) = V (0) +
_
T
0

F
x
(u,

S(u))d

S(u).
However by exercise 22,

V (T) = V (0) +
_
T
0
(u)d

S(u) and so we deduce that

(t) =

F
x
(t,

S(t)) =
F
x
(t, S(t)),
as was required.
Example Hedging a European Call Option
Write = T t, Then by exercise 26(b) ,
F(t, x) = e
r
E
Q
((xe
U+r
k)
+
),
where U N
_

1
2

2
,
2

_
. We can then write this in the same form as
(4.2.16) with T replaced throughout by . Notice that this equation is of the
form
F(t, x) =
_
(x)
(x)
g(t, x, y)dy, (4.3.24)
35
where (x) = log
_
k
x
_
r and (x) = . Lets take (x) = N for now. A
well-known result in calculus tells us that:
F(t, x)
x
=
_
(x)
(x)
g(t, x, y)
x
dy +
(x)
x
g(t, x, (x))
(x)
x
g(t, x, (x)).
When we apply this carefully and then let N we obtain
F
x
(t, s) =
1

2
_

log
(
k
s
)
r
e
y
exp
_

_
y +
1
2

_
2
2
2

_
dy
and by similar computations to those used to obtain (4.2.18) we get
F
x
(t, S(t)) = (d
1
()),
where d
1
() is given by the same formula as d
1
in (4.2.18) except that T is
replaced by and s is replaced by S(t).
In general, nancial analysts call
F
s
the delta of the option and denote
it by the Greek letter . Using to nd the process in the portfolio is
called delta hedging. There are a whole range of derivatives (in the sense of
calculus, not nance) which measure the sensitivity of the option price to
various parameters. These are called the Greeks. We have already met delta
which is denoted by =
F
s
. We also have
vega: =
F

,
theta: =
F
T
.
rho : =
F
r
.
gamma: =

2
F
s
2
.
Note that these are all dened at t = 0. Even for the simplest example of
the European call option, these can be dicult to calculate by hand. Using
e.g. Maple or Mathematica, you can check the following in this case:
= s

Tn(d
1
).
=
s
2

T
n(d
1
) +Kre
rT
(d
2
)
36
= Tke
rT
(d
2
)
=
n(d
1
)
s

T
.
Here n is the pdf of a standard normal distribution. Note that the Greeks
are all strictly positive for a European call option and this indicates that the
stock price increases as a function of the given parameter, e.g. > 0 i.e.
F

> 0 so the price increases when there is more volatility in the market
(as you would expect !)
4.4 Foreign Exchange
We would like to be able to price and hedge claims in dierent currencies.
This introduces a new complication - the exchange rate between currencies.
Well consider US and UK investors seeking to hold assets in either currency.
For simplicity, well just deal with risk-free assets in each currency. So
(A(t), 0 t T) is a dollar cash bond.
(D(t), 0 t T) is a Sterling cash bond.
(E(t), 0 t T) is the exchange rate, i.e E(t) is the value of one pound in
dollars at time t.
We assume the following dynamical behaviour of these quantities
A(t) = e
rt
, D(t) = e
ut
,
E(t) = E
0
exp{t +B(t)},
so the risk-free assets A and D are continuously compounded at rates r
and u (respectively), while E evolves as a geometric Brownian motion (E
0
is
constant).
We take the point of view of a dollar investor who wants to hold dollar-
valued options that speculate on the future value of Sterling. From his point
of view neither the exchange rate E or the Sterling bond D are tradeable
assets, but their product is so we dene S(t) = E(t)D(t).
We can now apply the Black-Scholes methodology within the dollar mar-
ket to the two assets A and S. The rst step is to nd a measure Q
d
under
which the discounted product

S(t) = A(t)
1
S(t) becomes a martingale. Its
an easy exercise to check that
d

S(t) =

S(t)(dB(t) dt),
37
where =
r u
1
2

. We apply Girsanovs theorem exactly as in

theorem 4.2.1. We then nd that (check this!)
dQ
d
dP
= exp
_
B(T)
1
2

2
T
_
and (W(t), 0 t T) is a Q
d
-Brownian motion where W(t) = B(t) t so
that d

S(t) =

S(t)dW(t).
It then follows by the usual arguments that the arbitrage price at time t
of a contingent claim X (priced in dollars) is
V (t) = e
r(Tt)
E
Q
d
(X|F
t
). (4.4.25)
We will study this further in the case where X is a forward contract to buy
one unit of sterling at time T for the price k. So the value of the claim at
the terminal date is X = E(T) k and so
V (t) = e
r(Tt)
E
Q
d
(E(T) k|F
t
).
A forward contract costs nothing at time zero and so we must have
0 = V (0) = e
rT
E
Q
d
(E(T) k) = e
rT
E
Q
d
(E(T)) e
rT
k
and hence the (arbitrage-free) strike price must be k = E
Q
d
(E(T)).
We can simplify this even further. Writing E(T) in terms of the Q
d
-
Brownian motion W we nd that
E(T) = E
0
exp
_
W(T) +
_
r u
1
2

2
_
T
_
,
and so the calculation of E
Q
d
(E(T)) boils down to that of the moment gen-
erating function for a normal distribution (see Exercise 7 in PAS401/6052).
You can check that we then obtain
k = e
(ru)T
E
0
and so
V (t) = e
r(Tt)
E
Q
d
(E(T) e
(ru)T
E
0
|F
t
). (4.4.26)
Our last exercise in the dollar market is to nd the precise hedging port-
folio (, ) whose value at time t is V (t) = (t)S(t)+(t)A(t). To do this we
use the fact that for each 0 t T, E(t) = D(t)
1
S(t) = D(t)
1
A(t)

S(t).
Now D(T)
1
= e
u(Tt)
D(t)
1
and A(T) = e
r(Tt)
A(t). Since

S is a Q
d
-
martingale we have
E
Q
d
(E(T)|F
t
) = e
(ru)(Tt)
D(t)
1
A(t)E
Q
d
(

S(T)|F
t
)
= e
(ru)(Tt)
D(t)
1
A(t)

S(t)
= e
(ru)(Tt)
E(t).
38
Substituting into (4.4.26) we obtain
V (t) = e
u(Tt)
E(t) e
rtuT
E
0
= e
uT
(e
ut
E(t) e
rt
E
0
).
If we discount in dollars we have

V (t) = e
rt
V (t) and so

V (t) = e
uT
(e
(ru)t
E(t) E
0
)
= e
uT

S(t) e
uT
E
0
.
From this we see that the portfolio is constant, i.e. for all 0 t
T, (t) = e
uT
and (t) = e
uT
E
0
.
So far we have investigated this transaction from the point of view of the
dollar investor. Now lets take the perspective of the Sterling investor. From
their point of view, there are two tradeable assets, these being the Sterling
bond D = (D(t), 0 t T) and the Sterling value of the dollar bond which
is Z = (Z(t), 0 t T) where each Z(t) = E(t)
1
A(t). The discounted
Sterling value at time t is

Z(t) = D(t)
1
Z(t). In exercise 30, you can check
that the measure Q
p
under which

Z is a Q
p
-martingale is given by
dQ
p
dP
= exp
_
B(T)
1
2

2
T
_
,
where =
r +
1
2

2
u

and so W

is a Q
p
-Brownian motion where for
each 0 t T,
W

(t) = B(t) t.
So the arbitrage-price at time t in Stirling of the claim X (which we recall
is priced in dollars) is
U(t) = D(t)E
Q
p
(D(T)
1
E(T)
1
X|F
t
)
= e
u(Tt)
E
Q
p
(E(T)
1
X|F
t
). (4.4.27)
We have expressions for the price at time t of the claim in dollars (4.4.26)
and its price in pounds (4.4.27). For these to be compatible we must have
V (t) = E(t)U(t) for each 0 t T - otherwise arbitrage opportunities are
created. To see that this holds, we rst need some preliminaries. First check
that W

(t) = W(t) t, hence

(t) =
dQ
p
dQ
d

F
t
= exp
_
W(t)
1
2

2
t
_
.
39
Its an easy exercise in algebra to check that for all 0 t T,
(t) = A(t)
1
D(t)E
1
0
E(t). (4.4.28)
Now its a fact that you can derive from the denition of conditional
expectation that if Y is any F
T
measurable random variable then
E
Q
p
(Y |F
t
) = (t)
1
E
Q
d
((T)Y |F
t
). (4.4.29)
Now we are ready to check our compatibility condition. Starting from
(4.4.27) and then applying (4.4.29), (4.4.28) and (4.4.25) we get
E(t)U(t) = E(t)D(t)E
Q
p
(D(T)
1
E(T)
1
X|F
t
)
= E(t)D(t)(t)
1
E
Q
d
((T)D(T)
1
E(T)
1
X|F
t
)
= E(t)D(t)A(t)D(t)
1
E
0
E(t)
1
E
Q
d
(A(T)
1
D(T)E
1
0
E(T)D(T)
1
E(T)
1
X|F
t
)
= A(t)E
Q
d
(A(T)
1
X|F
t
) = V (t),
as was required.
4.5 Dividends
4.5.1 Continuous Dividend Payments
Our nal variation on the standard Black-Scholes theory will deal with divi-
dends. Weve already seen how to include these in the binomial asset pricing
model in Problem 17. Now we will work with continuous time and real-valued
asset prices. We will begin by making the simple (but somewhat articial
assumption) that a dividend is paid continuously at a xed rate c where
0 < c < 1. (Well modify this later.) So in the time period [t, t + t], the
owner of the stock receives a dividend payment of cS(t)(t).
The diculty now is that S(t) no longer represents the true worth of the
asset at time t as it does not take account of the accumulated dividends up
to that time. In other words S is not a tradeable asset. The solution to
this diculty is to translate the problem into one which involves tradeable
assets. We make the further assumption that whenever a dividend is paid
then it is immediately reinvested in the stock. So the dividend cS(t)(t) is
used to buy ct units of stock. We thus construct a new tradeable asset
Z = (Z(t), 0 t T) whose return at time t is
Z(t)
Z(t)
= ct +
S(t)
S(t)
= ct +t +B(t),
40
so that
Z(t) = ( +c)Z(t)t +Z(t)B(t), (4.5.30)
which by the usual arguments yields the SDE:
dZ(t) = ( +c)Z(t)dt +Z(t)dB(t).
We thus see that at time t rather than holding one unit of stock we have
e
ct
units with total value
Z(t) = e
ct
S(t) (4.5.31)
= e
ct
S
0
exp
_
B(t) +
_

1
2

2
_
t
_
= S
0
exp
_
B(t) +
_
+c
1
2

2
_
t
_
, (4.5.32)
and you can easily check that
dZ(t) = e
ct
dS(t) +cZ(t)dt. (4.5.33)
Our strategy will be to regard Z as a single asset which pays no dividends.
We now apply Black-Scholes theory but with the asset Z taking the place
of S. Any portfolio which (at time t) contains (t)e
ct
units of the original
stock S(t) and (t) units of the risk-free asset A(t) can be thought of as a
portfolio which contains (t) units of the new asset Z(t) and (t) units of
A(t). Note that in this context, for (, ) to be self-nancing, we require
dV (t) = (t)dZ(t) +(t)dA(t)
= (t)e
ct
dS(t) +c(t)Z(t)dt +(t)dA(t),
by (4.5.33).
We will now price a European call option with strike price k and maturity
date T which is written on the dividend paying stock. First dene

Z =
A(t)
1
Z(t) = e
rt
Z(t). You can then check that
d

Z(t) =

Z(t)dB(t) + ( +c r)

Z(t)dt.
We now apply Girsanovs theorem and theorem 4.2.1. Indeed if we take
F(t) =
r c

in (4.2.6) to obtain the equivalent probability measure

41
Q then W = (W(t), 0 t T) is a Q-Brownian motion where W(t) =
B(t) +
+c r

t and
d

Z(t) =

Z(t)dW(t).
The value of the replicating portfolio at time t (and hence the arbitrage price
of the option) is
V (t) = e
r(Tt)
E
Q
((S(T) k)
+
|F
t
).
Now remember that it is

Z (and not

S) which is a martingale under the
probability measure Q and so we can write
V (t) = e
r(Tt)
E
Q
((e
cT
Z(T) k)
+
|F
t
)
= e
r(Tt)
e
cT
E
Q
((Z(T) ke
cT
)
+
|F
t
).
For each 0 t T dene G(t) = e
(rc)(Tt)
S(t). In Exercise 35 you can
deduce that, writing = T t as usual,
V (t) = e
r
_
_
_
G(t)
_
_
log
_
G(t)
k
_
+

2

_
_
k
_
_
log
_
G(t)
k
_

_
_
_
_
_
.
(4.5.34)
4.5.2 Periodic Dividend Payments
In the real world, dividend payments are paid at regular intervals - say at
times T
1
, .T
2
, . . .. At each time T
i
the holder of the stock receives a payout
of cS(T
i
) where 0 < c < 1. Between payouts the stock price will evolve
according to the usual geometric Brownian motion model. It can be shown
that the arbitrage-free model for the stock price in this context is
S(t) = S
0
(1 c)
N(t)
exp
_
B(t) +
_

1
2

2
_
t
_
,
where N(t) = max{i, T
i
t}. Again we have the problem that S is not a
tradeable asset and this can again be remedied by reinvesting the dividend
payment into stock. So at each time T
i
our total holding of stock will increase
by a factor of (1 c)
1
and we work with the process Z = (Z(t), 0 t T)
where
Z(t) = (1 c)
N(t)
S(t) = S
0
exp
_
B(t) +
_

1
2

2
_
t
_
.
42
We will thus apply the Black-Scholes model with the assets Z and A
instead of S and A. The analysis is fairly similar to that in the standard
Black-Scholes case case. So the equivalent martingale measure Q is again
obtained by taking F(t) =
r

in (4.2.6). To illustrate the use of the

theory in this case, well apply it to nd the fair price K of a forward contract
to buy one unit of periodic dividend paying stock at time T. So the claim is
X = S(T) K and as usual the arbitrage price at time t is
V (t) = e
r(Tt)
E
Q
((S(T) K)|F
t
)
= e
r(Tt)
E
Q
(((1 c)
N(T)
Z(T) K)|F
t
)
= (1 c)
N(T)
Z(t) Ke
r(Tt)
,
where we have used the fact that

Z is a Q-martingale where each

Z(t) =
e
rt
Z(t). Now since V (0) = 0, we immediately deduce that
K = e
rT
(1 c)
N(T)
Z(0) = e
rT
(1 c)
N(T)
S
0
.
4.6 Black-Scholes Theory - Concluding Re-
marks
In this chapter we have developed the Black-Scholes approach to pricing and
hedging of contingent claims and seen how to extend this to take account
of both exchange rate complications and dividend payments. There is much
more work that can be done beyond this within the Black-Scholes frame-
work. For example we can extend the analysis to exotic options such as
digital and multistage options. These are still of basic ECC type. There are
also important examples of options where the pay-os are not ECC claims
but which may depend on the whole history of the stock price (S(t), 0
t T). These include American options and Asian options where X =
_
1
T
_
T
0
S(t)dt K
_
+
. You can nd an introduction to the study of these
options in Chapter 6 of Etheridge.
The Black-Scholes paradigm is extremely powerful and has launched a
revolution in option pricing. However there are some problems with it. One
of these is that the main assumption - that the stock price is a geometric
Brownian motion - is a very crude assumption. In recent years there has
been a lot of work on replacing Brownian motion with more general sto-
chastic processes which more accurately model the behaviour of the market.
One particular direction that has been investigated by a number of workers
43
is to use a general Levy process - i.e. a process that has independent and
stationary increments - instead of Brownian motion. The paths of such a
process are no longer continuous and the discontinuous jumps in the cor-
responding stock price are a reection of supply and demand shocks to the
economy. A major diculty with moving away from Brownian motion is that
the market is incomplete, i.e. martingale measures are no longer unique and
so the second fundamental theorem of asset pricing is no longer valid. There
is some very interesting mathematics being developed in order to overcome
these diculties. For an introduction to these ideas see Chapter 5 of D. Ap-
plebaum Levy Processes and Stochastic Calculus, Cambridge University
Press (2004).
44