Sei sulla pagina 1di 59

U.U.D.M.

Project Report 2009:6


Examensarbete i matematik, 30 hp
Handledare och examinator: Johan Tysk
Maj 2009
Pricing Some American Multi-Asset Options
Jun Han
Department of Mathematics
Uppsala University
Contents
Abstract iv
Acknowledgement v
1 Introduction 1
1.1 Some Basic Option Theory . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 The Denition of an Option . . . . . . . . . . . . . . . . . 1
1.1.2 The Value of an Option . . . . . . . . . . . . . . . . . . . 2
1.2 The Black-Scholes Equation . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 The Black-Scholes Equation . . . . . . . . . . . . . . . . . 2
1.2.2 Boundary and Final Conditions of European options . . . 4
1.2.3 The Explicit Solution of the Black-Scholes Equation . . . 5
1.3 American Options . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 American Options on Multiple Assets 10
2.1 Stochastic Models of Multi-Assets Pricing . . . . . . . . . . . . . 10
2.2 The Mathematical Model of American Multi-Asset Option . . . . 12
2.3 American Options with Two Assets . . . . . . . . . . . . . . . . . 13
2.3.1 American Better-of Option on Two Assets . . . . . . . . . 13
2.3.2 American Call-max Option on Two Risky Assets . . . . . 17
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3 Finite Dierence Methods 24
3.1 Fundamentals of Numerical Dierentiation . . . . . . . . . . . . . 24
3.2 Finite Dierence Method . . . . . . . . . . . . . . . . . . . . . . 26
3.2.1 Types of Finite Dierence Methods . . . . . . . . . . . . . 26
3.2.2 Finite Dierence Methods for One-factor American Options 27
i
3.2.3 Finite Dierence Methods for Two-factor American Options 31
3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 Conclusion 36
A MATLAB Code 38
A.1 Direct Discretization - Explicit Method . . . . . . . . . . . . . . 38
A.2 Direct Discretization - Implicit Method . . . . . . . . . . . . . . 40
A.3 Brennan & Schwartz Model - Explicit Method . . . . . . . . . . . 41
A.4 Brennan & Schwartz Model - Implicit Method . . . . . . . . . . . 43
A.5 American Better-of Options on 2 assets at time t<T . . . . . . . 44
Bibliography 51
ii
List of Figures
1.1 European Call Option . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 European Put Option . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 At t = T,
1
and
2
degenerate to a ray: S
1
= S
2
. . . . . . . . . 17
2.2 At t < T, positions of
1
and
2
. . . . . . . . . . . . . . . . . . . 17
3.1 Direct Discretization - Explicit Method - Global Map . . . . . . 30
3.2 Direct Discretization - Explicit Method - Part Map . . . . . . . . 30
3.3 Direct Discretization - Implicit Method . . . . . . . . . . . . . . 31
3.4 Direct Discretization - Implicit Method Compared with Brennan
& Schwartz Model - Implicit Method . . . . . . . . . . . . . . . . 32
3.5 Brennan & Schwartz Model - Implicit Method Compared with
Brennan & Schwartz Model - Explicit Method . . . . . . . . . . . 33
3.6 Option Values for an American Better-of Option at time t = T =
1/12, r = 0.05, q
1
= 0.02, q
2
= 0.03,
1
=
2
= 0.4 . . . . . . . . 34
3.7 Option Values for an American Better-of Option at time t = 0,
T = 1/12, r = 0.05, q
1
= 0.02, q
2
= 0.03,
1
=
2
= 0.4 . . . . . . 34
iii
Abstract
In the rst two chapters of this thesis, we consider some basic facts about
options such as the dierence between American and European options. We
study in more detail the American better-of and call-max options. In order to
numerically solve the free boundary PDEs corresponding to these options, we
use a nite dierence method developed in Chapter 3. All the programming is
done in MATLAB, and the corresponding code can be found in the Appendix.
We solve these PDEs in the original variables, without transforming them to
equations on the whole space.
iv
Acknowledgement
I would like to express my gratitude to my supervisor professor Johan Tysk
for his patience, support and guidance. He has always provided the balance
of assistance, pressure and encouragement - helping me when I needed it, and
leaving me on my own when I needed to gure out something by myself. More-
over, in completing this thesis, he spent a lot of time helping me to correct it,
even very small spelling and font mistakes. His knowledge and insight have been
invaluable resources to me in my current and future career.
I am very grateful to my schoolmates, Yang Zeng, Alan, and Xuemao Zhang
for the assistances about the nite dierence method, MATLAB, and LATEX.
I also thank the department of mathematics of Uppsala University for giving
me the opportunity to study here.
Last but not least, I thank my parents. Though they could not be with
me for the last a few years, they have given me constant love, encouragement
and support, and they have always been eager to help me in any way that they
could.
v
vi
Chapter 1
Introduction
1.1 Some Basic Option Theory
1.1.1 The Denition of an Option
Before we value an option, we have to know what an option is. From [1] we can
nd the denition of a European call option: a European call option is
a contract with the following conditions:
At a prescribed time in the future, known as the expiry date, the owner
of the option may
Purchase a prescribed asset, known as the underlying asset or, briey,
the underlying, for a
Prescribed amount, known as the exercise priceor strike price.
In this denition, the author uses the word may to imply that the holder
of the option has a right and not an obligation, but the writer does have a
potential obligation, since if the holder chooses to exercise this call option, he
has to sell the asset. From [2] we get another direct deniton: A call option
gives its holder the right but not the obligation to purchase a share of stock in
the underlying company at a xed price for a xed length of time.
A European put option is exactly the opposite of a European call option:
the holder of a put option has the right but not the obligation to sell a prescribed
asset at a prescribed time in the future.
From what has been discussed above, we may safely draw the denition of
an option from [3] : an option is an agreement that the holder can buy from,
1
or sell to, the seller of the option at a specied future time a certain amount of
an underlying asset at a specied price. But the holder is under no obligation
to exercise the contract.
1.1.2 The Value of an Option
Before talking about the value of an option, we have to introduce some notation,
which are used consistently throughout my thesis.
We denote the value of an option by V , and V is a function with two
variables: the current value of the underlying asset, S, and time, t : V =
(S, t), such that V
t
= V (S
t
, t);
The volatility of the underlying asset, ;
The exercise price, E;
The expiry, T;
The interest rate, r.
From these assumptions, we know that V only depends on S and t. An
options value at expiration date, V
T
, is already set, which is just the options
payo:
V
T
=
_
(S
T
E)
+
, (call option)
(E S
T
)
+
. (put option)
The value of an option, which we just want to nd out, is a functionV =
V (S, t),(0 S < , 0 t T), such that
V (S, T) =
_
(S E)
+
, (call option)
(E S)
+
. (put option)
1.2 The Black-Scholes Equation
1.2.1 The Black-Scholes Equation
Before describing the Black-Scholes analysis (Black & Scholes 1973) which leads
to the value of anoption, I list some assumptions, which I make for most of my
thesis.
2
0 2 4 6 8 10
0
1
2
3
4
5
6
7
8
S
T
C
European Call Option with E=2 at time T
Call Option
Figure 1.1: European Call Option
The asset price follows the lognormal random walk.
The risk-free interest rate r and the asset volatility are known functions
of time over the life of the option.
The market is ideal, which means: continuous trading, innitely divisible
assets, no transaction costs, taxes, no restrictions on short sales, and so
on.
The underlying asset pays no dividends during the life of the option.
There are no arbitrage possibilities.
I assume that the market consists of two assets with dynamics given by
dB(t) = rB(t)dt, (1.2.1)
dS(t) = S(t)dt + S(t)dW(t), (1.2.2)
where r, , and are deterministic constants for convenience. Consider a simple
contingent claim of the form
= (S(T)), (1.2.3)
and assume that the claim can be traded on a ideal market and its process has
the form
(t) = V (t, S(t)), (1.2.4)
3
0 1 2 3 4 5 6
0
0.5
1
1.5
2
2.5
3
3.5
4
S
T
P
European Put Option with E=2 at time T
Put Option
Figure 1.2: European Put Option
for some smooth function V .
Then follow the process which is given in [4], we obtain a very important
theorem.
Theorem 1.2.1 (Black-Scholes Equation). Assume that the market is specied
by 1.2.11.2.2, and we want to price a contingent claim of the form 1.2.3. Then
the only pricing function of the form 1.2.4 which is consistent with the absence
of arbitrage is when V is the solution of the following boundary value problem
in the domain [0, T] R
+
.
V
t
(t, s) + rsV
s
(t, s) +
1
2
s
2

2
(t, s)V
ss
(t, s) rV (t, s) = 0, (1.2.5)
V (T, s) = (s). (1.2.6)
From Theorem1.2.1, we nd that if we let (S(t)) be a payo function of an
option at time T, then the function V (t, s) will be price of the option.
1.2.2 Boundary and Final Conditions of European options
After deriving the Black-Scholes equation for the price of an option, we need to
nd out the boundary and nal conditions, otherwise, the PDE does not have
a unique solution.
4
First, let focus on a European call option, whose value is denoted by c(S, t),
with expiry date T, and exercise price E. At t = T, the value of a call is the
payo function:
c(S, T) = max(S E, 0), (1.2.7)
and this is our nal condition. Then we need to nd out the boundary condi-
tions, which are applied at S = 0, and as S . From [1] we know that if S
is ever zero, then it must remain zero, we can get the boundary conditions:
c(0, t) = 0, (1.2.8)
c(S, t) S as S . (1.2.9)
For a European call option, with nal condition 1.2.7 and boundary conditions
1.2.81.2.9, we can get the Black-Scholes value of a European call option.
For a European put option, whose value is p(S, t), the nal condition is also
the payo function
p(S, t) = max(E S, 0). (1.2.10)
The boundary condition when S = 0 is
p(0, t) = Ee
r(Tt)
, (1.2.11)
when the interest rate r is constant. As S , holders of put options are
unlikely to exercise the contract, so
p(S, t) 0 as S . (1.2.12)
1.2.3 The Explicit Solution of the Black-Scholes Equation
In the last section, we have established the PDEs and their nal and boundary
conditions, which are satised by the prices of European call and put options.
In this section, we will nd the explicit solution of the Black-Scholes equation.
Let us look at a European call with value c(S, t). The Black-Scholes equation
and boundary conditions for it are
c
t
+
1
2

2
S
2

2
c
S
2
+ rS
c
S
rc = 0, (1.2.13)
with
c(0, t) = 0, c(S, t) S Ke
r(Tt)
as S ,
and
c(S, T) = max(S E, 0).
5
First we set
S = Ee
x
, t = T

1
2

2
, c = E(x, ).
Then we turn Equation 1.2.13 into a new equation

=

2

x
2
+ (k
1
1)

x
k
1
, (1.2.14)
where k
1
=
r
1
2

2
. The initial condition becomes
(x, 0) = max(e
x
1, 0).
We turn Equation 1.2.14 into a diusion equation by a simple change of variable.
Try putting
= e
x+
u(x, ),
for some constants and to be found, then dierentiation gives us
u +
u

=
2
u + 2
u
x
+

2
u
x
2
+ (k
1
1)(u +
u
x
) k
1
u.
By choosing
=
2
+ (k
1
1) k
1
,
we can get rid of all u terms, and with the choice
0 = 2 + (k
1
1)
we can eliminates the
u
x
term as well. From these equations, we get
=
1
2
(k
1
1), and =
1
4
(k
1
+ 1)
2
.
Then we can get
= e
(
1
2
(k
1
1)x
1
4
(k
1
+1)
2
)
u(x, t),
where
u

=

2
u
x
2
for < x < , > 0, (1.2.15)
with
u(x, 0) = u
0
(x) = max(e
1
2
(k
1
+1)x
e
1
2
(k
1
1)x
, 0). (1.2.16)
The solution of the diusion equation problem 1.2.15 is
u(x, ) =
1
2

u
0
(s)e
(xs)
2
4
ds (1.2.17)
here u
0
(x) is given by Equation 1.2.16.
6
Now, we need to evaluate the integral in Equation 1.2.17. Making the change
of variable x

=
(xs)

2
, we obtain
u(x, ) = I
1
I
2
,
where
I
1
= e
1
2
(k
1
+1)x+
1
4
(k
1
+1)
2

N(d
1
),
d
1
=
1

2
+
1
2
(k
1
+ 1)

2,
and
N(d
1
) =
1

2
_
d
1

1
2
s
2
ds
is the cumulative distribution function for the normal distribution.
The calculation of I
2
is similar to that of I
1
, just use (k
1
1) to replace
(k
1
+ 1) throughout.
Finally, we retrace the steps, using
(x, ) = e

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

u(x, )
and putting x = log(
S
E
), =
1
2

2
(T t), and c = E(x, ), then we can get
c(S, t),
c(S, t) = SN(d
1
) Ee
r(Tt)
N(d
2
),
with
d
1
=
log(
S
E
) + (r +
1
2

2
)(T t)

_
(T t)
,
d
2
=
log(
S
E
) + (r
1
2

2
)(T t)

_
(T t)
.
We can get the value function of a European put option just following the
similar steps, but after having that of a European call option, we can use put-call
parity formula
c p = S Ee
r(Tt)
to get the value p of a European put option.
1.3 American Options
From this section on, we focus on American options, which give their holders the
right to exercise the option at any time, thus oering more opportunity to make
prot than the corresponding European options. Intuitively, we conclude that
the price of American option cannot be less than that of an equivalent European
7
option. In mathematical theory, American option pricing corresponds to a free
boundary problem. This boundary divides the domain [0, T] R
+
into two
parts: the continuation region, within which it is better to hold the option than
to exercise, and the stopping region, within which it is better to exercise the
option than to hold it. The dividing price between exercise and non-exercise is
called the optimal exercise price S
f
(t).
We will take non-dividend-paying American put option as example to estab-
lish the mathematical models.
From the discussion above, we know that for an American put option with
expiration date t = T, there exist two regions: the continuation region
1
,
within which:
P(S, t) > max(E S, 0),
and the stopping region
2
, within which:
P(S, t) = max(E S, 0).
When S is very small, the American put option should be exercised at once,
on the other hand, when S > E, the holder should continue to keep the option,
because the payo is zero. Therefore we conclude: there must exists S
f
(t)
(0, E), such that

1
= {(S, t)|S
f
(t) S < , 0 t },

2
= {(S, t)|0 S S
f
(t), 0 t }.
And between these two regions is the optimal exercise boundary : S = S
f
(t).
In the continuation region
1
, with the -hedging principle and the Ito
formula, we nd that in the continuation region the option price P = P(S, t)
satises the Black-Scholes equation:
P
t
(S, t) + rsP
s
(S, t) +
1
2
s
2

2
(S, t)P
ss
(S, t) rP(S, t) = 0,
On the optimal exercise boundary ,
P(S
f
(t), t) = E S
f
(t),
P
S
(S
f
(t), t) = 1,
when S ,
P 0,
8
and at t = T,
P(S, T) = max(E S).
This is known as the smooth t principle. From a nancial point of view it is
natural, since this means that the hedging parameter is continuous across the
free boundary.
In order to price an American put option, we need to nd out {P(S, t), S
f
(t)}
in the continuation region
1
, such that they satisfy the Black-Scholes equation
with optimal exercise boundary conditions.
At the end of this section, we will talk a little about American call option.
Since we know that the price of American option cannot be less than that of an
equivalent European option, we will nd that for an American call option,we
always have:
C(S, t) > max(E S, 0),
which means that the price function of American call option C(S, t) always stays
in the continuation region
1
. So we can conclude that the holder of American
call options should always wait until time t = T, which means that an American
call option has the same price function as an equivalent European call option.
1.4 Summary
In this chapter, we consider the denition of nancial options, the Black-Scholes
equation, boundary and nal conditions of European options, and the free
boundary conditions of American options. However, in the real world, options
are not only derived from one underlying risky asset. In many cases, options are
derived from two or more underlying risky assets, and they are called multi-
asset options. In next chapter, we will focus on the mathematical model of
American options derived from two or more underlying risky assets.
9
Chapter 2
American Options on
Multiple Assets
2.1 Stochastic Models of Multi-Assets Pricing
In order to price options derived from multiple assets, we need to establish the
price movement model for the underlying multi-assets. From[3], let S
i
be the
price of the i-th risky asset (i = 1, . . . , n), and S
i
satises a stochastic dierential
equation with the following form:
dS
i
(t) = S
i
(t)
i
dt + S
i
(t)
i
dW
i
(t), (2.1.1)
where dW
i
(t)(i = 1, . . . , n) are standard Brownian motion that satisfy
E(dW
i
) = 0, (2.1.2)
V ar(dW
i
) = dt, (2.1.3)
and
Cov(dW
i
, dW
j
) =
ij
dt. (i = j) (2.1.4)
In this equation, Cov(., .) is the covariance, and can be expressed as
Cov(dW
i
, dW
j
) = E(dW
i
dW
j
), (i = j)
10
since
Cov(dW
i
, dW
j
) = E([dW
i
E(dW
i
)][dW
j
E(dW
j
)])
= E(dW
i
dW
j
dW
i
E(dW
j
) E(dW
i
)dW
j
+(E(dW
i
)E(dW
j
)))
= E(dW
i
dW
j
)
by the denition of covariance and Equation 2.1.2.
Equation 2.1.1 can be modeled in another form:
dS
i
= S
i

i
dt + S
i
m

j=1

ij
dW
j
, (i = 1, . . . , n) (2.1.5)
where dW
i
(i = 1, . . . , n) are one-dimensional Brownian motions, such that:
E(dW
i
) = 0, (2.1.6)
V ar(dW
i
) = dt, (2.1.7)
Cov(dW
i
, dW
j
) = 0, (i = j) (2.1.8)
In many cases, using Equation 2.1.5 will be easier for us.
In order to price options on multiple assets, we need to nd out the Black-
Scholes equation for multiple assets. Let S
1
, . . . , S
n
be n risky assets, which
satisfy Equation 2.1.5, and let V (S
1
, . . . , S
n
, t) be an option derived from these
risky assets.
With -hedging principle, we choose
i
shares of asset S
i
(i = 1, . . . , n) to
construct a portfolio :
= V (S
1
, . . . , S
n
, t)
n

i=1

i
S
i
,
such that this portfolio is risk-free in (t, t + dt).
With the Ito formula for the multivariate stochastic process, we will have
d = dV
n

i=1

i
dS
i

n

i=1

i
S
i
q
i
dt
=
_
V
t
+
1
2
n

i,j=1

ij
S
i
S
j

2
V
S
i
S
j
_
dt
+
n

i=1
V
S
i
dS
i

n

i=1

i
dS
i

n

i=1

i
S
i
q
i
dt,
11
where q
i
is the dividend rate of asset S
i
, and

ij
=
m

k=1

ik

jk
. (i, j = 1, . . . , n)
Since we need bo be risk-free in time (t, t + dt), we have
d = rdt = r(V
n

i=1

i
S
i
)dt, (2.1.9)
We use
i
=
V
S
i
to replace
i
in Equation 2.1.9, and eliminate dt, then we will
get the equation:
V
t
+
1
2
n

i,j=1

ij
S
i
S
j

2
V
S
i
S
j
+
n

i=1
(r q
i
)S
i
V
S
i
rV = 0. (2.1.10)
Equation 2.1.10 is the Black-Scholes equation for multi-asset options.
2.2 The Mathematical Model of American Multi-
Asset Option
From [3], we know that the mathematical model of American multi-asset option
can be seen as a variational inequality.
The domain of it is:
=
_
(S
1
, . . . , S
n
, t)|(S
1
, . . . , S
n
) R
n
+
, 0 t T
_
,
where
R
n
+
=
_
(S
1
, . . . , S
n
)|0 S
i
< , i = 1, . . . , n
_
.
Let V = V (S
1
, . . . , S
n
, t) be the American option price, then V is the solution
of the following problem in :
_
_
_
min
_

V
t
LV, V (S
1
, . . . , S
n
)
_
= 0, () (2.2.1)
V (S
1
, . . . , S
n
) = (S
1
, . . . , S
n
), (S
1
, . . . , S
n
R
n
+
). (2.2.2)
where L is the multivariate Black-Scholes dierential operator, which is given
as:
LV =
1
2
n

i,j=1

ij
S
i
S
j

2
V
S
i
S
j
+
n

i=1
(r q
i
)S
i
V
S
i
rV,
and (S
1
, . . . , S
n
) is the payo function.
From Equation 2.2.1, we conclude that: in the continuation region
1
:
V
t
+LV = 0,
12
V > (S
1
, . . . , S
n
);
and in the stopping region
2
:
V
t
+LV 0,
V = (S
1
, . . . , S
n
).
Between
1
and
2
is the optimal exercise boundary .
From [3], we have the following theorem:
Theorem 2.2.1. Suppose the payo function satises:
(1) is a Lipschitz continuous function;
(2) There exists

(R
n
+
), such that:
lim
0

= ,
and is uniformly convergent in any nite bounded domain within R
n
+
;
(3)

S
i
> 0 (or

S
i
< 0), (2.2.3)
and in any nite bounded domain D within (R
n
+
), there exists a constant
C
D
which depends on D only and is independent of , such that
L

C
D
, (2.2.4)
then the solution of the American multi-asset option pricing problem 2.2.12.2.2,
V (S
1
, . . . , S
n
, t), has the following properties:
(1) V (S
1
, . . . , S
n
, t) is a nondecreasing (monotone increasing) function of S
i
(i =
1, . . . , n),
(2) V (S
1
, . . . , S
n
, t) is a nonincreasing function of t.
2.3 American Options with Two Assets
2.3.1 American Better-of Option on Two Assets
From[3],the holder of better-of option receives exercise payo associated with the
better performer of the underlying assets.
13
There are two kinds of better-of option: one is based on the price, the payoff
function of which is:
payoff = max(
1
S
1
(T), . . . ,
n
S
n
(T)),
where S
i
(T) is the price of the i-th risky asset at time t = T, and
i
is a
coefficient so that all risky asset prices are at the same level, and the other one
is based on the growth rate, the payoff function of which is:
payoff(rate) = max(

S
1
(T), . . . ,

S
n
(T)),
where

S
i
is the growth rate of the i-th risky asset at time t = T.
Under the transformation:

S
i
(T) =
S
i
(T) S
i
(0)
S
i
(0)
=
S
i
(T)
S
i
(0)
1,
we will nd these two kinds of better-of options have an relationship, since their
payoff functions have an relationship, which is:
max(

S
1
(T), . . . ,

S
n
(T)) = max(
1
S
1
(T), . . . ,
n
S
n
(T)) 1,
where

i
=
1
S
i
(0)
, (i = 1, . . . , n).
From now on, we will talk about American better-of option on two assets,
the payoff function of which is max(S
1
, S
2
), in this section.
The mathematical model is: (from[3])
_
_
_
min
_

V
t
LV, V max(S
1
, S
2
)
_
= 0, () (2.3.1)
V |
t=T
= max(S
1
, S
2
), (S
1
, S
2
R
2
+
). (2.3.2)
where
LV =
1
2
_

2
1
S
2
1

2
V
S
2
1
+ 2
1

2
S
1
S
2

2
V
S
1
S
2
+
2
2
S
2
2

2
V
S
2
2
_
+(r q
1
)S
1
V
S
1
+ (r q
2
)S
2
V
S
2
rV, (2.3.3)
and this is the variational inequality model for American better-of option
on two assets.
We introduce the transformation:
=
S
1
S
2
, u(, t) =
V (S
1
, S
2
, t)
S
2
, (2.3.4)
14
then we have:
V
S
1
= S
2
u

S
1
=
u

,
V
S
2
= u + S
2
u

S
2
= u
u

2
V
S
2
1
=
1
S
2

2
u

2
,

2
V
S
1
S
2
=

2
u

S
2
=

S
2

2
u

2
,

2
V
S
2
2
=
u

S
2

S
2

2
u

S
2
=

2

2
u
S
2

2
.
Substituting these into Equation 2.2.12.2.2,we will have a problem in one di-
mensional:
_

_
min
_

u
t

1
2

2

2
u

2
(q
2
q
1
)
u

+q
2
u, u max(, 1)
_
= 0, (2.3.5)
u|
t=T
= max(, 1). (2.3.6)
where the domain is

= {(, t)|0 , 0 t T} and
2
=
2
1
+2
1

2
+

2
2
.
Since q
1
, q
2
are dividend rate, in the real world, the dividend rates are always
nonnegative, so we can assume that the dividend rate are positive (q
1
, q
2
> 0)
for simple. Then we can write the domain

to be

2
, where

1
is the
continuation region of the option:
u > max(, 1),
u
t
+
1
2

2

2
u

2
+ (q
2
q
1
)
u

q
2
u = 0;

2
is the stopping region of the option:
u = max(, 1),
u
t
+
1
2

2

2
u

2
+ (q
2
q
1
)
u

q
2
u 0;

is the interface of

1
and

2
.
Theorem 2.3.1. If q
1
, q
2
> 0, then for the problem 2.3.52.3.6, the continua-
tion region is:

1
=
_
(, t)|
1
(t)
2
(t), 0 t T
_
,
15
where

1
(T) =
2
(T) = 1,

1
(t) nondecreasing,

2
(t) nonincreasing.
In

1
, u satises the free boundary problem:
_

_
u
t
+
1
2

2

2
u

2
+ (q
2
q
1
)
u

q
2
u = 0, (

1
) (2.3.7)
u(
1
(t), t) = 1, (0 t T) (2.3.8)
u

(
1
(t), t) = 0, (0 t T) (2.3.9)
u(
2
(t), t) = , (0 t T) (2.3.10)
u

(
2
(t), t) = 1, (0 t T) (2.3.11)
We omit the proof of the theorem. The complete proof is given in [3].
Now, we need to go back to the original function V (S
1
, S
2
, t), and we will
get the price of the American better-of option:
V (S
1
, S
2
, t) = S
2
u(
S
1
S
2
) =
_

_
S
2
, 0
S
1
S
2

1
(t),
S
2
u(
S
1
S
2
, t),
1
(t)
S
1
S
2

2
(t),
S
1
,
2
(t)
S
1
S
2
< ,
(2.3.12)
and the optimal boundary consists of surfaces
1
and
2
:

1
: S
1
=
1
(t)S
2
, (2.3.13)
and

2
: S
2
=
2
(t)S
2
, (2.3.14)
where
1
(t) =
2
(t) = 1,
1
(t) ,
2
(t) .
Then we can get the gures at t = T and t < T.
From the Figure 2.1 and 2.2, we conclude that at time t = T, the optimal
boundary lines
1
and
2
coincide to the line S
1
= S
2
; at time 0 < t < T,
the optimal boundary lines separate, and the optimal consists of two lines:
S
1
=
1
(t)S
2
and S
1
=
2
(t)S
2
, between these two lines lies the continuation
region, and beyond them lies the stopping region; and at time t = 0, the optimal
boundary lines are S
1
=
1
(0)S
2
and S
1
=
2
(0)S
2
.
16
0 1 2 3 4 5
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
S
1
S
2

2
V=S
2
V=S
1
S
1
=S
2
Figure 2.1: At t = T,
1
and
2
degenerate to a ray: S
1
= S
2
.
0 1 2 3 4 5
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
S
1
S
2
V=S
2
V=S
1

2
V=S
2
u(S
1
/S
2
,t)
Figure 2.2: At t < T, positions of
1
and
2
.
2.3.2 American Call-max Option on Two Risky Assets
In this section, we will talk about American call-max option on two risky assets,
the exercise payo function is:
(max(S
1
, S
2
) E)
+
, (2.3.15)
where the notationx
+
is short for max(x, 0). This option can be regarded as a
generalization of the American better-of option, since the American better-of
option is the E = 0 case of American call-max option on two risky assets.
We should set up the mathematical model for American call-max options on
two risky assets. From [3], we get the mathematical model: In domain
: {(S
1
, S
2
) R
2
+
, 0 t T}
17
solve the variational inequality
_
_
_
min
_

V
t
LV, V (max(S
1
, S
2
) E)
+
_
= 0, () (2.3.16)
V |
t=T
= (max(S
1
, S
2
) E)
+
, (S
1
, S
2
R
2
+
). (2.3.17)
where
LV =
1
2
_

2
1
S
2
1

2
V
S
2
1
+ 2
1

2
S
1
S
2

2
V
S
1
S
2
+
2
2
S
2
2

2
V
S
2
2
_
+(r q
1
)S
1
V
S
1
+ (r q
2
)S
2
V
S
2
rV. (2.3.18)
From the Theorem 2.2.1, we infer some properties of the price function of
an American call-max option, V (S
1
(t), S
2
(t), t).
(1) If

S
i
S
i
, for i = 1, 2, then
V (

S
1
(t),

S
2
(t), t) V (S
1
(t), S
2
(t), t); (2.3.19)
(2) If

E E, then
0 V (S
1
(t), S
2
(t), t, E) V (S
1
(t), S
2
(t), t,

E)

E E; (2.3.20)
(3) If

t t, then
V (S
1
(t), S
2
(t),

t) V (S
1
(t), S
2
(t), t) 0. (2.3.21)
Standard American Options
Before we proceed further, we review some basic results for a standard American
call option derived from a single underlying risky asset. Let be the optimal
exercise boundary, that is = inf{S(t) : (S(t), t)
2
}.
From [9], we know that Van Moerbeke(1976) and Jacka(1991) show that
is continuous; Kim(1990) and Jacka(1991) show that is decreasing in t;
Kim(1990) shows that
T
lim
tT

t
= max((
r
q
E, E)); Merton(1973) shows
that is bounded above and derives a formula for

lim
t

t
; Jacka(1991)
shows that the option value C
t
(S
t
)(the theoretical value of an American call op-
tion) is continuous and the stopping region
2
is closed.
Proposition 2.3.1. For a standard American call option, (S(t), t)
2
implies
that (S(t), t)
2
for all 1.
Proof. (S(t), t)
2
implies that C(S(t), t) = S(t) E > 0. Since 1, we
have S(t) E > 0. Then we have C(S(t), t) = S(t) E > 0, which means
that (S(t), t)
2
.
18
Stopping Region of American Call-max Option on Two Risky Assets
From Proposition 2.3.1, we will have similar conjectures for American call-max
option on two risky assets:
Conjecture 2.3.1. (S
1
(t), S
2
(t), t)
2
implies (
1
S
1
(t),
2
S
2
(t), t)
2
) for
all
1
1 and
2
1.
Conjecture 2.3.2. If q
1
> 0 and q
2
> 0 then there exist constants M
1
and M
2
such that (S
1
(t), S
2
(t), t)
2
for all S 1(t) M
1
and S
2
(t) M
2
.
Conjecture 2.3.3. (S
1
(t), S
2
(t), t)
2
and (

S
1
(t),

S
2
(t), t)
2
implies
(S
1
(t), S
2
(t), t) + (1 )(

S
1
(t),

S
2
(t), t)
2
for all 0 1.
From [9], we know that all these three conjectures are false. However, also
from [9], we know that properties similar to those on one risky asset do hold on
certain subregions of
2
(stopping region of American call-max option on two
risky assets).
Dene the subregion
i
2
of the stopping region
2
by
i
2
=
2
G
i
where
G
i
{(S
1
(t), S
2
(t), t) : S
i
(t) = max(S
1
(t), S
2
(t)} for i = 1, 2
Proposition 2.3.2. If S
1
(t) = S
2
(t) > 0 and t < T then (S
1
(t), S
2
(t), t) /

2
. That is, prior to maturity exercise is not optimal when the prices of the
underlying assets are equal.
This proposition states that: prior to maturity, stopping is suboptimal when
the prices of the underlying assets are equal, no matter how large the prices are
and no matter how large the dividend rates are. We can prove this proposition
by intuition: we can exercise this option at some xed time t
1
where t < t
1
< T,
then at least we will have
PV (t
1
t) = S
1
(t)e
q
1
(t
1
t)
Ee
r(t
1
t)
plus a European option to exchange asset 2 for asset 1 with a maturity date t
1
,
and the value of this European option is
E

t
[e
r(t
1
t)
(S
2
(t
1
) S
1
(t
1
))
+
].
As t
1
converges to t, PV (t
1
t) converges to S
1
(t) E at a nite rate, but at
the same time, the value of that European exchange option converges to 0 at
an increasing rate that approaches innity in the limit. So, there must be some
time t
1
> t such that waiting until t
1
gives a strictly positive return relative to
stopping at time t. you can nd the strict proof in [9].
19
Proposition 2.3.3 (Subregion Convexity). Let S = (S
1
, S
2
) and

S = (

S
1
,

S
2
).
Suppose (S, t)
i
2
and (

S, t)
i
2
for a xed i = 1 or 2. Given , with
0 1, dene S() = S + (1 )

S. Then (S(), t)
i
2
. That means, if
stopping is optimal at S and

S and if (S, t) G
i
and (

S, t) G
i
then stopping
is optimal at S().
Since the payo function is convex with respect to (S
1
, S
2
) and the mul-
tiplicative structure of the uncertainty in Equation 2.1.1 when i = 1, 2, the
stopping region is convex.
Proposition 2.3.4. Let
2
represent the stopping region for a max-option.
Then
2
satises the following properties.
(1) (S
1
(t), S
2
(t), t)
2
implies (S
1
(t), S
2
(t), s)
2
for all t s T.
(2) (S
1
(t), S
2
(t), t, E
1
)
2
implies (S
1
(t), S
2
(t), t, E
2
)
2
for all E
2
> E
1
.
(3) (S
1
(t), S
2
(t), t)
1
2
implies (S
1
(t), S
2
(t), t)
1
2
for all 1.
(4) (S
1
(t), S
2
(t), t)
1
2
implies (S
1
(t), S
2
(t), t)
1
2
for all 0 1.
(5) (S
1
(t), 0, t)
1
2
implies S
1
(t)
1
(t).
where
1
(t) is the optimal exercise boundary for a standard American option
derived from a single underlying risky asset. In (3), (4), (5), analogous results
hold for the subregion
2
2
.
From property (1), we can conclude that the stopping region expands as time
moves forward, since a short maturity option cannot be more valuable than the
longer maturity option,which gives holder more opportunity to make prot.
Property (2) implies that the stopping region shrinks as the exercise price E
increases. Property (3) implies that the stopping subregion is connected in the
direction of increasing S
1
, and property (4) implies that the stopping subregion
is connceted in the direction of decreasing S
2
.
Now, I will give the proof of the property (1), you can nd these two proof
in [3]. Since the stopping region expands as time moves forward, we conclude
that if t
2
> t
1
, then

2
(t
2
, E)
2
(t
1
, E) (2.3.22)
Proof. Relation 2.3.22 is a corollary of Equation 2.3.21. Suppose otherwise:
there exist (S
0
1
(t), S
0
2
(t)) and t = t
0
1
, t
0
2
with t
0
2
> t
0
1
, such that:
(S
0
1
(t), S
0
2
(t), t
0
1
)
2
(t
0
1
, E),
20
but
(S
0
1
(t), S
0
2
(t), t
0
2
)
1
(t
0
2
, E).
Then we will have
V (S
0
1
(t), S
0
2
(t), t
0
2
) > (S
0
1
(t), S
0
2
(t)) = V (S
0
1
(t), S
0
2
(t), t
0
1
),
which contradicts to Equation 2.3.21.
Next, I will prove the property (2). Since the stopping region shrinks as the
exercise price E increases, we have: if E
2
> E
1
, then

2
(t, E
2
)
2
(t, E
1
). (2.3.23)
Proof. Assume otherwise, then we will have (S
0
1
(t), S
0
2
(t)) and t = t
0
, such that
when E
2
> E
1
,
(S
0
1
(t), S
0
2
(t), t
0
)
2
(t, E
2
),
but
(S
0
1
(t), S
0
2
(t), t
0
)
1
(t, E
1
).
Due to Equation 2.3.20, we have
(S
0
1
(t), S
0
2
(t); E
2
) = V (S
0
1
(t), S
0
2
(t), t
0
; E
2
)
V (S
0
1
(t), S
0
2
(t), t
0
; E
1
) E
2
+ E
1
> (S
0
1
(t), S
0
2
(t); E
1
) E
2
+ E
1
,
i.e.
(max(S
0
1
(t), S
0
2
(t)) E
2
)
+
+ E
2
> (max(S
0
1
(t), S
0
2
(t)) E
1
)
+
+ E
1
. (2.3.24)
Thus there must be
max(S
0
1
(t), S
0
2
(t)) < E
2
, (2.3.25)
otherwise, if
max(S
0
1
(t), S
0
2
(t)) E
2
> E
1
,
then we will have
(max(S
0
1
(t), S
0
2
(t))E
2
)
+
+E
2
= max(S
0
1
(t), S
0
2
(t)) = (max(S
0
1
(t), S
0
2
(t))E
1
)
+
+E
1
.
This contradicts to Equation 2.3.24.
But Equation 2.3.25 and (S
0
1
(t), S
0
2
(t), t
0
)
2
(t, E
2
) are contradictory,
since Equation 2.3.23, therefore (S
0
1
(t), S
0
2
(t), t
0
) must belong to
2
(t, E
1
).
21
Proposition 2.3.5 (Divergence of the stopping region). For a xed time t,
satisfying t < T. There exists
1
and
2
with
2
< 1 <
1
such that

2
(t) R(
1
,
2
) = .
where
R(
1
,
2
)
_
(S
1
, S
2
) R
2
+
:
2
S
1
< S
2
<
1
S
1
_
and for
2
<
1
, R(
1
,
2
) denotes the open cone dened by the price rations

1
and
2
.
With this result, we can draw the shape of a typical stopping region
2
.
Proposition 2.3.6. Let V (S
1
(t), S
2
(t), t) be the value of the American call-max
option, we have:
(1) V (S
1
(t), S
2
(t), t) is continuous on R
+
R
+
[0, T];
(2) V (, S
2
(t), t) and V (S
1
(t), , t) are nondecreasing on R
+
for all S
1
(t), S
2
(t)
in R
+
and all t in [0, T];
(3) V (S
1
(t), S
2
(t), ) is nonincreasing on [0, T] for all S
1
(t) and S
2
(t) in R;
(4) V (, , t) is convex on R
+
R
+
for all t in [0, T].
The continuity of the function can be proved by the continuity of the payo
function and the continuity of Equation 2.1.1. The monotonicity of the function
follows the monotonicity of the payo function. The property (3) is right since
a shorter maturity option cannot be more valuable. The convexity is implied
by the convexity of the payo function.
Remark 1 (American call-max option with more than two underlying assets).
We will talk about some properties of American call-max option with more than
two underlying assets. Since they are in more general situation, I think they are
correct in case of two underlying assets.
First, we introduce some notations:
The stopping region of the American call-max option derived from n assets,

n
2
.
The corresponding price, V (S
1
(t), . . . , S
2
(t), t).
The vector of underlying asset prices, S (S
1
(t), . . . , S
n
(t)).
22
G
n
i

_
(S, t) : S
i
(t) = max(S
1
(t), . . . , S
n
(t))
_
for i = 1, . . . , n.
Proposition 2.3.7. If max(S
1
(t), . . . , S
n
(t)) = S
i
(t) = S
j
(t) for i = j, i
{1, . . . , n}, j {1, . . . , n} and if t < T then (S, t) /
n
2
. Which means prior
to maturity stopping is suboptimal if the maximum is achieved by two or more
asset prices.
This property parallels Proposition 2.3.2, and implies that stopping is subop-
timal on all regions where the maximum asset price is achieved by two or more
asset prices.
I will give some properties that parallel the properties in Section 2.3.2, and
I omit the proofs of these properties, which can be found in [9].
Proposition 2.3.8 (Subregion Convexity). Consider two vectors S R
n
+
and

S R
n
+
. Suppose that (S, t)
n,i
2
and (

S, t)
n,i
2
for the same i {1, . . . , n},
where
n,i
2

2
G
n
i
. Given with 0 1 denote S() = S + (1 )

S.
Then (S(), t)
n,i
2
. That is, if stopping is optimal at S and

S and if (S, t)
G
n
i
and (

S, t) G
n
i
then stopping is optimal at S().
Proposition 2.3.9.
n
2
satises the following properties.
(1) (S, t)
n
2
implies (S, s)
n
2
for all t s T;
(2) (S, t)
n,i
2
implies (S
1
(t), . . . , S
i
(t), . . . , S
n
(t), t)
n,i
2
for all 1;
(3) (S, t)
n,i
2
implies (
1
S
1
(t),
2
S
2
(t), . . . , S
i
(t),
i+1
S
i+1
(t), . . . ,
n
S
n
(t))

n,i
2
for all 0
j
1, j = 1, . . . , i 1, i + 1, . . . , n;
(4) S
i
(t) = 0 and (S, t)
n,i
2
implies (S
1
(t), . . . , S
i1
(t), S
i+1
(t), . . . , S
n
(t), t)

n1,i
2
.
2.4 Summary
In this chapter, we describe mathematical models of American multi-asset op-
tions, especially American better-of options on two assets and American call-
max option on two assets. In Remark 1, we discuss some properties of American
call-max options with n > 2 underlying assets. Since American option pricing is
a free boundary problem to the Black-Scholes equation, there is no explicit so-
lution of the option price in general, except for the perpetual American option.
Therefore we have to use numerical methods to nd the solution of American
option price, and we will discuss the numerical method in the next chapter.
23
Chapter 3
Finite Dierence Methods
When we attempt to price an option, we use the knowledge of nance and
the partial dierential equation to formulate the PDEs, which the option price
satises. It is very dicult, sometime even impossible, to nd a closed-form
solution to these PDEs, so we need numerical methods to solve them. From
[10], we know that there are several most common methods:
Monte Carlo methods, which involve generating large numbers of numeri-
cally simulated realizations of the random walk followed by the underlying
assets and averaging the derivative prices obtained from these realizations.
Lattice methods, which include the binomial and trinomial models, as-
sume that the underlying stochastic process is discrete i.e. the underlying
asset jumps to a nite number of values(each associated with a certain
probability) with a small advancement in time.
Finite Dierence methods, which consist of discretizing the PDE and the
given boundary conditions to form a set of dierence equations, which can
be solved either directly or iteratively.
In this chapter, we will use nite dierence methods to price American op-
tions.
3.1 Fundamentals of Numerical Dierentiation
When we have a real-valued function of a real variable, for example:
y = f(x), (3.1.1)
24
we are interested in the rst and second derivatives of the function. When the
form of the function f is unknown, it is impossible to calculate its derivatives
analytically, then we have to use numerical approximations. Assume that we
want to approximate the rst derivative of y at some point a, and assume that
h is a very small positive number, then we have several ways to approximate its
rst derivative:
the centred dierence formula:
f

(a)
f(a + h) f(a h)
2h
. (3.1.2)
the forward dierence formula:
f

(a)
f(a + h) f(a)
h
. (3.1.3)
the backward dierence formula:
f

(a)
f(a) f(a h)
h
. (3.1.4)
For convenience, we use following notations:
D
0
f(a)
f(a + h) f(a h)
2h
,
D
+
f(a)
f(a + h) f(a)
h
,
D

f(a)
f(a) f(a h)
h
.
Now, let us check how good these approximations to the derivative of f at
a are. Take the centred dierence as an example, use Taylors expansion, then
we have
f(a h) = f(a) hf

(a) +
h
2
2!
f

(a)
h
3
3!
f

), (3.1.5)
where

(a h, a),
+
(a, a + h).
Then we conclude that in this particular case
D
0
f(a) = f

(a) +
h
2
6
_
f

(
+
) + f

)
2
_
. (3.1.6)
From the equation above, we know that the centred dierences give a second-
order approximation to the rst derivative if h is small enough and if f has
continuous derivatives up to order 3. With similar arithmetic, we know that the
25
forward dierences and the backward dierences give rst-order approximation
to the rst derivative of f at the point a:
D
+
f(a) = f

(a) +
h
2
f

(
+
),
+
(a, a + h); (3.1.7)
D

f(a) = f

(a)
h
2
f

),

(a h, a). (3.1.8)
One-side schemes are rst-order accurate, but they ask for low continuity
constraints on the function f, which means that we only need to assume that
the second derivative of f is continuous.
Next, let us discuss divided dierences for the second derivative of f at some
point a. We propose the three-point formula, which is popular and much used:
D
+
D

f(a)
f(a h) 2f(a) + f(a + h)
h
2
. (3.1.9)
The discretisation error is given by
D
+
D

f(a) = f

(a) +
h
4
4!
_
f
(iv)
(
+
) + f
(iv)
(

)
_
(3.1.10)
3.2 Finite Dierence Method
From this section, we will focus on the numerical methodsnite dierence
method. When we use nite dierence method, we have to consider the consis-
tency, stability, and convergence of the scheme, which we use to discretize time
and space. I omit the statement and the proof of these theories, and they can
be found in [6] and [8], since in this thesis, we are interested in the application
of the nite dierence method.
3.2.1 Types of Finite Dierence Methods
Depending on how we approximate the partial derivative with respect to time,
we have three dierent nite dierence schemes:
Explicit nite dierence scheme, when we use the forward dierence for-
mula, Equation 3.1.3;
Implicit nite dierence scheme, when we use the backward dierence
formula, Equation 3.1.4;
Crank-Nicholson nite dierence scheme, when we use the centred dier-
ence formula, Equation 3.1.2.
26
The Crank-Nicholson scheme can be seen as a special case of the -method,
which can be thought of as a -weighted average of the explicit and implicit
nite dierence methods. When =
1
2
, the -method is the Crank-Nicholson
method.
When we calculate an option price given the payo function and boundary
conditions with nite dierence method, we transform the Black-Scholes equa-
tion into a system of equations, which can be solved by using matrix notation.
In general, the explicit method can be transformed into:
V
n+1
= AV
n
+ C
n
,
while the implicit method can be transformed into:
AV
n+1
= V
n
+ C
n
,
and the Crank-Nicholson method can be transformed into:
AV
n+1
= BV
n
+ C
n
.
We nd that the explicit method can be solved directly with the matrix A, while
the implicit and Crank-Nicholson can be solved indirectly with the inversion of
the matrix A.
3.2.2 Finite Dierence Methods for One-factor American
Options
In this section, we will use nite dierence methods to price one-factor American
options, since for Chapter 1, we know that a one-factor American call option
has the same price function as an equivalent one-factor European call option,
we will pay our attention to one-factor American put option, which satises the
Black-Scholes equation:
P
t
(S, t) + rsP
s
(S, t) +
1
2
s
2

2
(S, t)P
ss
(S, t) rP(S, t) = 0,
On the optimal exercise boundary ,
P(S
f
(t), t) = E S
f
(t),
P
S
(S
f
(t), t) = 1,
when S ,
P 0,
27
and at t = T,
P(S, T) = max(E S).
For convenience, we assume (S, t) is a constant.
Direct Discretization of PDE
We have:
P
t
(S, t) + rsP
s
(S, t) +
1
2
s
2

2
P
ss
(S, t) rP(S, t) = 0, (3.2.1)
we use centred dierence formula to approximate P
s
and P
ss
, and when we use
the forward dierence formula to approximate P
t
, we use the explicit method,
and when we use the backward dierence formula, we use the implicit method.
First, let us use the explicit method. Discretize the space variable S into N
equally spaced units of S, and the time variable t into M equally spaced units
of t:
S
n
= nS, forn = 0, 1, , N;
t
m
= mt, form = 0, 1, , M.
When n = N, we have S
N
= S
max
, where S
max
is a realistic approximation
to innity, and is subjectively chosen, depending on the maturity T, and type
of the option. From [10], we know that S
max
does not have to be too large in
practice, typically it should be three or four times the value of the exercise price
or more generally, three to four times the value of the asset at which there is
some important behaviour.
Substituting approximations into Equation 3.2.1, we have:
0 =
P(S, t + t) P(S, t)
t
+ rS
P(S + S, t) P(S S, t)
2S
+
1
2

2
S
2
P(S + S, t) 2P(S, t) + P(S S, t)
(S)
2
rP(S, t).(3.2.2)
P(S, t) can be approximated by P(S
n
, T
m
) P
m
n
. From Equation 3.2.2, we
have:
P
m
n
= a
n
P
m1
n1
+b
n
P
m1
n
+c
n
P
m1
n+1
, for n = 1, , N 1, and m = 1, , M,
(3.2.3)
where
a
n
=
1
1 + rt
(
1
2
rnt +
1
2

2
n
2
t),
b
n
=
1
1 + rt
(1
2
n
2
t),
28
c
n
=
1
1 + rt
(
1
2
rnt +
1
2

2
n
2
t).
Then we need to nd the payo function and the boundary conditions:
The nal condition at t = T, which means that m = M:
P
M
n
= max(0, E n(S)), for n = 0, , N. (3.2.4)
The payo function at t = 0, which means that m = 0:
P
0
n
= max(0, E n(S)), for n = 0, , N. (3.2.5)
The boundary condition at S = 0:
P
m
0
= E, for m = 1, , M. (3.2.6)
The boundary condition t S = S
max
:
P
m
N
= 0, for m = 1, , M. (3.2.7)
Next, let us use the implicit method to solve the same PDE 3.2.1. When we
use the implicit method, we use the backward dierence formula to approximate
P
t
, then the equivalent equation of Equation 3.2.2 is:
0 =
P(S, t) P(S, t t)
t
+ rS
P(S + S, t) P(S S, t)
2S
+
1
2

2
S
2
P(S + S, t) 2P(S, t) + P(S S, t)
(S)
2
rP(S, t).(3.2.8)
P(S, t) can be approximated by P(S
n
, T
m
) P
m
n
. From Equation 3.2.8, we
have:
P
m1
n
= a
n
P
m
n1
+ b
n
P
m
n
+ c
n
P
m
n+1
, for n = 1, , N 1, and m = 1, , M,
(3.2.9)
where
a
n
=
1
2
rnt
1
2

2
n
2
t,
b
n
= 1 +
2
n
2
t + rt,
c
n
=
1
2
rnt
1
2

2
n
2
t.
The initial condition(since we use the backward dierence formula, the nal
condition becomes the initial condition) and the boundary conditions are exactly
the same as those of the explicit method.
29
0 20 40 60 80 100
1
0
1
2
3
4
5
6
7
x 10
133
Option Values for an American Put, r=0.1, vol=0.3, T=5/12, E=50
Stock Price
O
p
t
i
o
n

P
r
i
c
e
Figure 3.1: Direct Discretization - Explicit Method - Global Map
0 20 40 60 80 100
0
5
10
15
20
25
30
35
40
45
50
Option Values for an American Put, r=0.1, vol=0.3, T=5/12, E=50
Stock Price
O
p
t
i
o
n

P
r
i
c
e
Figure 3.2: Direct Discretization - Explicit Method - Part Map
When we use the explicit method, we have to consider the stability of the
scheme. From [6], we know that the scheme for the explicit method is only
conditionally stable for the value of time-step, t, meanwhile the implicit method
and the Crank-Nicolson method are unconditionally stable for the value of time-
step, t. Figure 3.1 and 3.2 show what can happen when the scheme of the
explicit method is not stable. Figure 3.3 shows what we get when we use the
implicit method, and we conclude that the implicit method is superior over the
explicit method in this respect.
Brennan & Schwartz Models
When we use Brennan & Schwartz models, we use a log transform of the stock
price S, and this transform can reduce Equation 3.2.1 to a new PDE, which has
30
0 20 40 60 80 100
0
5
10
15
20
25
30
35
40
45
50
Option Values for an American Put, r=0.1,vol=0.4, T=5/12, E=50
Stock Price
O
p
t
i
o
n

P
r
i
c
e
Figure 3.3: Direct Discretization - Implicit Method
constant coecients.
Dene:
Z = log S, (3.2.10)
W(Z, t) = P(S, t), (3.2.11)
then we will have:
P
S
= e
Z
W
Z
, (3.2.12)

2
P
S
2
= e
2Z
(

2
W
Z
2

W
Z
), (3.2.13)
P
t
=
W
t
. (3.2.14)
Substitute these equations into Equation 3.2.1, then we will get the following
equation:
W
t
+ (r
1
2

2
)
W
Z
+
1
2

2
W
Z
2
rW = 0. (3.2.15)
We use centred dierence formula to approximate
W
Z
and

2
W
Z
2
, when we use
the forward dierence formula to approximate
W
t
, we use the explicit method,
and when we use the backward one, we use the implicit method. Then we can
follow the same steps that we used previously to approximate W(Z, t), then we
will have P(S, t).
3.2.3 Finite Dierence Methods for Two-factor American
Options
Let us look at an American better-of option on two risky assets. For conve-
nience, we assume that the underlying assets are independent and thus the
31
0 50 100 150
0
5
10
15
20
25
30
35
40
45
50
Option Values for an American Put, r=0.1, vol=0.3, T=5/12, E=50
Stock Price
O
p
t
i
o
n

P
r
i
c
e
B & S Model Implicit
Direct Discretization Implicit
Figure 3.4: Direct Discretization - Implicit Method Compared with Brennan &
Schwartz Model - Implicit Method
cross-derivative term is zero. The Black- Scholes PDE is:
V
t
+ (r q
1
)S
1
V
S
1
+
1
2

2
1
S
2
1

2
V
S
2
1
+ (r q
2
)S
2
V
S
2
+
1
2

2
2
S
2
2

2
V
S
2
2
rV = 0
(3.2.16)
with the nal, and boundary conditions:
V (S
1
, S
2
, T) = max(S
1
, S
2
), for 0 S
1
, S
2
< , (3.2.17)
V (0, S
2
, t) = S
2
, for 0 t T, 0 S
2
< , (3.2.18)
V (, S
2
, t) = , for 0 t T, 0 S
2
< , (3.2.19)
V (S
1
, 0, t) = S
1
, for 0 t T, 0 S
1
< , (3.2.20)
V (S
1
, , t) = , for 0 t T, 0 S
1
< . (3.2.21)
First, we use the centred dierence formula to approximate
V
S
1
,

2
V
S
2
1
,
V
S
2
,
and

2
V
S
2
2
.
At each time level, we have:
V
t
= r
1
S
1
V (S
1
+ S
1
, S
2
) V (S
1
S
1
, S
2
)
2S
1

1
2

2
1
S
2
1
V (S
1
+ S
1
, S
2
) 2V (S
1
, S
2
) + V (S
1
S
1
, S
2
)
(S
1
)
2
r
2
S
2
V (S
1
, S
2
+ S
2
) V (S
1
, S
2
S
2
)
2S
2

1
2

2
2
S
2
2
V (S
1
, S
2
+ S
2
) 2V (S
1
, S
2
) + V (S
1
, S
2
S
2
)
(S
2
)
2
+rV (S
1
, S
2
). (3.2.22)
32
0 50 100 150
0
5
10
15
20
25
30
35
40
45
50
Stock Price
O
p
t
i
o
n

P
r
i
c
e
Option Values for an American Put, r=0.1, vol=0.3, T=5/12, E=50
B & S Model Explicit
B & S Model Implicit
Figure 3.5: Brennan & Schwartz Model - Implicit Method Compared with Bren-
nan & Schwartz Model - Explicit Method
where r
1
= r q
1
and r
2
= r q
2
. V (S
1
, S
2
) can be approximated by
C(S
1i
, S
2j
) C
i,j
, then Equation 3.2.22 becomes:
V
t
= d
i,j
V
i1,j
+ b
i,j
V
i,j1
+ a
i,j
V
i,j
+ c
i,j
V
i,j+1
+ e
i,j
V
i+1,j
, (3.2.23)
where:
a
i,j
= r +
2
1
i
2
+
2
2
j
2
,
b
i,j
=
1
2
r
2
j
1
2

2
2
j
2
,
c
i,j
=
1
2
r
2
j
1
2

2
2
j
2
,
d
i,j
=
1
2
r
1
i
1
2

2
1
i
2
,
e
i,j
=
1
2
r
1
i
1
2

2
1
i
2
.
Then we need to nd the boundary conditions,
The boundary condition at S
1
= 0:
V
0,j
= jS
2
, for 0 j J, (3.2.24)
The boundary condition at S
1
= S
1 max
:
V
I,j
= S
max
, for 0 j J, (3.2.25)
The boundary condition at S
2
= 0:
V
i,0
= iS
1
, for 0 i I, (3.2.26)
33
The boundary condition at S
2
= S
2 max
:
V
i,J
= S
max
, for 0 i I, (3.2.27)
for convenience, we assume that S
1 max
= S
2 max
= S
max
.
Then at each time level, we have:
AU =
U
t
,
where A is a matrix, U is a vector,which consists of all elements V
i,j
, which are
not on the boundary, at each time level.
0
5
10
0
2
4
6
8
10
0
2
4
6
8
10
S
1
Opton Values for an American Betterof Option at time t=T
S
2
V
(
S
1
,
S
2
,
T
)
Figure 3.6: Option Values for an American Better-of Option at time t = T =
1/12, r = 0.05, q
1
= 0.02, q
2
= 0.03,
1
=
2
= 0.4
0
5
10
0
2
4
6
8
10
0
2
4
6
8
10
12
S
1
Option Values for an American Betterof Option at time t=0
S
2
V
(
S
1
,
S
2
,
t
)
Figure 3.7: Option Values for an American Better-of Option at time t = 0,
T = 1/12, r = 0.05, q
1
= 0.02, q
2
= 0.03,
1
=
2
= 0.4
34
Now we begin to use the -Method, which is also called the -weighted
method. Follow the way in [8], and [10], we have:
U(t + k) U(t)
k
((1 ) + ) = (1 )AU(t) + AU(t + k), (3.2.28)
where k is time-step, A is just the same matrix above, and U is the vector at
each time level. Equation 3.2.28 can be rewritten in the equivalent form:
_
I kA
_
U(t + k) + kB(t + k) =
_
I + (1 )kA
_
U(t), (3.2.29)
where I is indentity matrix, B(t + k) is a vector, which consists of all elements
V
i,j
, which are on the boundary, at time level t = t + k. Since we have U(T)
and B(T) from the nal condition, we can solve U(t) at each time level with:
U(t k) =
_
I + (1 )kA
_
1
__
I kA
_
U(t) + kB(t)
_
. (3.2.30)
3.3 Summary
In this chapter, we describe the basic concepts of nite dierence methods, and
we use direct discretization model and Brennan & Schwartz model to numeri-
cally solve American put option derived from one asset. With each model we use
both explicit and implicit methods, and we nd that the explicit method is con-
ditionally stable for the time-step, meanwhile the implicit and Crank-Nicolsom
methods are unconditionally stable for the time-step. We also use the -method
to solve an American better-of option derived from two assets. When we use the
-method, we use centred dierence formula to approximate the space, and at
each time level, we solve a system of ODE equations, then we use the formular
U(t k) =
_
I +(1 )kA
_
1
__
I kA
_
U(t) +kB(t)
_
to calculate U(t) at each
time level, since we get U(T) with the nal condition.
35
Chapter 4
Conclusion
In the rst chapter of this thesis, we introduce the basic concepts of options, the
Black-Scholes equation, boundary and nal conditions of European options, and
the free boundary conditions of American options. We take American options
to compare with European options in Section 1.3, give the mathematics model
of American put options, and conclude that the holder of American call options
should always wait until time t = T, which means that an American call option
has the same price function as an equivalent European call option.
In the second chapter, we focus on American options derived from multiple
assets. In order to price American multi-asset options, we introduce stochastic
models of multi-assets pricing, the Black-Scholes equation for multi-asset op-
tions, and a very important theorem, Theorem 2.2.1, which we use to prove a
few propositions in Section 2.3. We also give the mathematics models of Ameri-
can better-of option and American call-max option derived from two assets, and
discuss some propositions of these two options.
In order to nd the solution of American option price, we introduce nite
dierence methods in Chapter 3. We use direct discretization model and Bren-
nan & Schwartz model to solve American put option derived from one asset,
and with each model we use both explicit and implicit methods. We use the
-method to solve American better-of options derived from two assets in Section
3.2. This method is a weighted average between explicit and implicit methods.
When people use the -method to solve Black-Scholes equations, in most
cases, they rst make changes of variables to transform the Black-Scholes PDE
to the diusion PDE (in one dimension) or the Heat PDE (in two dimension).
But I use the -method to solve the Black-Scholes equation for two assets options
36
directly, and it works and gives good approximations.
Further research can include the study of American better-of and similar
options on several underlying assets. I believe that the approach taken in this
paper, using the original variables, should be of advantage in higher dimensions.
37
Appendix A
MATLAB Code
A.1 Direct Discretization - Explicit Method
% Parameter I n i t i a l i z a t i o n
Smax=100; K=50; r =0. 10; vol =0. 30; T=5/12;
% Grid I n i t i a l i z a t i o n
N=149; % Number of Time I nt e r va l s ;
M=300; % Number of Stock Pr i ce I nt e r va l s
t i mes t ep=T/N;
s t oc ks t e p=Smax/M;
% Boundary Condi t i ons
% Payof f at Maturi ty t=T
f o r j =1:M+1
f ( j , N+1)=max(K(j 1) s t ocks t ep , 0 ) ;
end ;
% Stock Pr i ce=0 S=0
f o r i =1:N+1
f ( 1 , i )=K;
end ;
38
% S=Smax
f o r i =1:N+1
f (M+1, i )=0;
end ;
% Cal cul at e a , b , c ( parameters ) f o r Model
f o r j =1:M1
a ( j )=1/(1+r t i mes t ep )( 0. 5 r j t i mes t ep +0.5 vol 2 j 2 t i mes t ep ) ;
b( j )=1/(1+r t i mes t ep )(1vol 2 j 2 t i mes t ep ) ;
c ( j )=1/(1+r t i mes t ep ) ( 0. 5 r j t i mes t ep +0.5 vol 2 j 2 t i mes t ep ) ;
end ;
% Cal c ul at i ons ( Sol ve System Ax=d)
% Set up matri x A
f o r j =2:M
A( j , j 1)=a ( j 1);
A( j , j )=b( j 1);
A( j , j +1)=c ( j 1);
end ;
A( 1 , 1) =1;
A(M+1,M+1)=1;
% Sol ve System
f o r i=N: 1: 1
f ( : , i )=A f ( : , i +1); % Cal cul at e Val ues of Option
% American Condi ti on
f o r j =1:M+1
i f f ( j , i )<(K(j 1) s t oc ks t e p )
f ( j , i )=(K(j 1) s t oc ks t e p ) ;
end ;
end ;
end ;
j =1: 1:M1;
39
pl ot ( j , f ( j , 1 ) ) ;
A.2 Direct Discretization - Implicit Method
% Parameter I n i t i a l i z t i o n
Smax=100; K=50; r =0. 03; vol =0. 30; T=5/12;
% Grid I n i t i a l i z a t i o n
N=149; % Number of Time I nt e r va l s ;
M=300; % Number of Stock Pr i ce I nt e r va l s
t i mes t ep=T/N;
s t oc ks t e p=Smax/M;
% Boundary Condi t i ons
% Payof f at Maturi ty t=T
f o r j =1:M+1
f ( j , N+1)=max(K(j 1) s t ocks t ep , 0 ) ;
end ;
% Stock Pr i ce=0 S=0
f o r i =1:N+1
f ( 1 , i )=K;
end ;
% S=Smax
f o r i =1:N+1
f (M+1, i )=0;
end ;
% Cal cul at e a , b , c ( parameters ) f o r Model
f o r j =1:M1
a ( j )=0. 5 r j ti mestep 0.5 vol 2 j 2 t i mes t ep ;
b( j )=1+vol 2 j 2 t i mes t ep+r t i mes t ep ;
c ( j )=0.5 r j ti mestep 0.5 vol 2 j 2 t i mes t ep ;
end ;
40
% Cal c ul at i ons ( Sol ve System Ax=d)
% Set up matri x A
f o r j =2:M
A( j , j 1)=a ( j 1);
A( j , j )=b( j 1);
A( j , j +1)=c ( j 1);
end ;
A( 1 , 1) =1;
A(M+1,M+1)=1;
% I nver t Matri x A
A=i nv (A) ;
% Sol ve System
f o r i=N: 1: 1
f ( : , i )=A f ( : , i +1); % Cal cul at e Value
% American Condi ti on
f o r j =1:M+1
i f f ( j , i )<(K(j 1) s t oc ks t e p )
f ( j , i )=(K(j 1) s t oc ks t e p ) ;
end ;
end ;
end ;
j =1: 1:M1;
pl ot ( j , f ( j , 1 ) ) ;
A.3 Brennan &Schwartz Model - Explicit Method
% Parameter I n i t i a l i z a t i o n
Smax=100; Zmax=round ( l og (Smax ) ) ; K=50; r =0. 10; vol =0. 30; T=5/12;
% Grid I n i t i a l i z a t i o n
41
m=149; % Number of Time I nt e r va l s ;
n=300; % Number of Stock Pr i ce I nt e r va l s ;
k=T/m; % Time Step
h=vol s qr t (3k ) ; % Numeri cal l y most e f f i c i e n t to s e t changes to
% t hi s val ue
Zmin=Zmaxhn ;
P=z e r os ( n+1,m+1);
% Boundary Condi t i ons
f o r i =1: n+1
s t oc kpr i c e ( i )=exp ( Zmin+( i 1)h ) ;
P( i ,m+1)=max(Ks t oc kpr i c e ( i ) , 0 ) ; %Payof f Functi on
end ;
f o r j =1:m
P( 1 , j )=K; % S=0 i mpl i e s Put pr i c e = K
P( n : n+1, j )=0; % S=Smax i mpl i e s Put pr i c e = 0
% and P( n1, j )=P( n , j ) f o r a l l j
end ;
% Cal cul at e a , b , c ( parameters ) f o r Model
a =(0. 5( vol /h)2 0. 5( r 0.5 vol 2)/h)k/(1+r k ) ;
b=1( vol /h)2k/(1+r k ) ;
c =(0. 5( vol /h)2+0. 5( r 0.5 vol 2)/h)k/(1+r k ) ;
% Set up System Ax=b ;
f o r row=2: n
A( row , row1)=a ;
A( row , row)=b ;
A( row , row+1)=c ;
end ; A( 1 , 1) =1; A( n+1,n+1)=1;
% Cal c ul at i ons ( Sol ve System Ax=d)
f o r j=m: 1: 1
P( : , j )=AP( : , j +1);
42
% American Condi ti on
f o r i =1: n+1
i f P( i , j )<(Ks t oc kpr i c e ( i ) )
P( i , j )=(Ks t oc kpr i c e ( i ) ) ;
end ;
end ;
end ;
i =1: 1: n1; pl ot ( exp ( Zmin+( i 1)h) , P( i , 1 0 ) , : r ) ;
A.4 Brennan &Schwartz Model - Implicit Method
% Parameter I n i t i a l i z a t i o n
Smax=100; Zmax=round ( l og (Smax ) ) ; K=50; r =0. 10; vol =0. 30; T=5/12;
% Grid I n i t i z l i z a t i o n
m=149; % Number of Time I nt e r va l s
n=300; % Number of Stock Pr i ce I nt e r va l s
k=T/m; % Time Step
h=vol s qr t (3k ) ; % Numeri cal l y most e f f i c i e n t to s e t change to
% t hi s val ue
Zmin=Zmaxhn ; P=z e r os ( n+1,m+1);
% Boundary Condi t i ons
f o r i =1: n+1
s t oc kpr i c e ( i )=exp ( Zmin+( i 1)h ) ;
P( i ,m+1)=max(Ks t oc kpr i c e ( i ) , 0 ) ; % Payof f f unc t i on
end ;
f o r j =2:m+1
P( 1 , j )=K; % S=0 i mpl i e s Put pr i c e = K
P( n : n+1, j )=0; % S=Smax i mpl i e s Put pr i c e = 0
% and P( n1, j ) = P( n , j ) f o r a l l j
end ;
43
% Cal cul at e a , b , c ( parameters ) f o r Model
a=( 0.5( vol /h)2+0. 5( r 0.5 vol 2)/h)k ; b=1+(vol /h)2k+r k ;
c =( 0.5( vol /h)2 0. 5( r 0.5 vol 2)/h)k ;
% Set up System Ax=b
f o r row=2: n
A( row , row1)=a ;
A( row , row)=b ;
A( row , row+1)=c ;
end ; A( 1 , 1) =1; A( n+1,n+1)=1;
% I nver t Matri x A
A=i nv (A) ;
% Cal c ul at i ons ( Sol ve System Ax=d)
f o r j=m: 1: 1
P( : , j )=AP( : , j +1);
% American Condi ti on
f o r i =1: n+1
i f P( i , j )<(Ks t oc kpr i c e ( i ) )
p( i , j )=(Ks t oc kpr i c e ( i ) ) ;
end ;
end ;
end ;
i =1: 1: n1; pl ot ( exp ( Zmin+( i 1)h) , P( i , 10) , b ) ;
A.5 American Better-of Options on 2 assets at
time t<T
% Better of Option
% Parameter I n i t i a l i z a t i o n
44
Smax=15; T=5/12; r =0. 05; q1 =0. 02; q2 =0. 03; r1=rq1 ; r2=rq2 ;
vol 1 =0. 40; vol 2 =0. 40;
% Grid I n i t i a l i z a t i o n
t i mes t ep =0. 01; s t oc ks t e p =0. 5; h=0. 2;
N=8; % Number of Time
M=Smax/ s t oc ks t e p +1; % Number of Stock
s1=l i ns pa c e ( 0 , Smax,M) ; % a l l the nodes on the x coor di nat e
s2=l i ns pa c e ( 0 , Smax,M) ; % a l l the nodes on the y coor di nat e
[ S1 , S2]=meshgri d ( s1 , s2 ) ;
%t i mes t ep=T/N;
%s t oc ks t e p=Smax/M;
MM=M2; %the i n t e r i o r nodes on the space mesh
% Fi nal Condi ti on t=T, s e t up matri x V(T)
V T=z e r os (M,M) ;
f o r i =1:M
f o r j =1:M
V T( i , j )=max( S1 ( i , j ) , S2 ( i , j ) ) ;
end
end
% Set up the matri x A and the vect or b
A=z e r os (MM2 ,MM2) ; b=z e r os (MM2 , 1) ;
% us i ng the boundary c ondi t i ons : v ( 0 , s2 , t )=s2&v( s1 , 0 , t )=s1
i =1; j =1;
A( i , i )=( vol 1 2) s1 ( i +1)2/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( i , i +1)=(r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( i , i+MM)=(r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
45
b( i )=(( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh) ) s2 ( j +1 ) . . .
+(( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh) ) s1 ( i +1);
% us i ng the boundary c ondi t i on : v ( 0 , s2 , t )=s2
j =1;
f o r i =2: (MM1)
A( i , i )=( vol 1 2) s1 ( i +1)2/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( i , i +1)=(r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( i , i 1)=( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( i , i+MM)=(r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
b( i )=(( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh) ) s2 ( j +1);
end
% us i ng the boundary c ondi t i ons : v(INF, s2 , t )=Smax&v( s1 , 0 , t )=s1
i=MM; j =1;
A( i , i )=( vol 1 2) s1 ( i +1)/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( i , i 1)=( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( i , i+MM)=(r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
b( i )=(( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh) ) Smax . . .
+(( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh) ) s1 ( i +1);
% i n t e r i o r nodes
f o r j =2: (MM1)
f o r i =2: (MM1)
k=MM( j 1)+i ;
A( k , k)=( vol 1 2) s1 ( i +1)2/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( k , k+1)=(r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( k , k1)=( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( k , k+MM)=(r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
A( k , kMM)=( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
end
46
end
% us i ng the boundary c ondi t i on : v ( 0 , s2 , t )=s2
i =1;
f o r j =2: (MM1)
k=(MM( j 1)+1);
A( k , k)=( vol 1 2) s1 ( i +1)2/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( k , k+1)=(r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( k , k+MM)=(r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
A( k , kMM)=( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
b( k)=(( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh) ) s2 ( j +1);
end
% us i ng the boundary c ondi t i on : v(INF, s2 , t )=Smax
i=MM;
f o r j =2: (MM1)
k=MM( j 1)+MM;
A( k , k)=( vol 1 2) s1 ( i +1)2/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( k , k+1)=( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( k , k+MM)=(r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
A( k , kMM)=( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
b( k)=(( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh) ) s2 ( j +1);
end
% us i ng the boundary c ondi t i ons : v ( 0 , s2 , t )=s2&v( s1 , INF, t )=Smax
i =1; j=MM; k=MM( j 1)+1;
A( k , k)=( vol 1 2) s1 ( i +1)2/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( k , k+1)=(r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( k , kMM)=( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
b( k)=(( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh) ) s2 ( j +1 ) . . .
+((r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh) ) Smax;
47
% us i ng the boundary c ondi t i on : v( s1 , INF, t )=Smax
j=MM; f o r i =2: (MM1)
k=MM( j 1)+i ;
A( k , k)=( vol 1 2) s1 ( i +1)2/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( k , k+1)=(r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( k , kMM)=( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
A( k , k1)=( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
b( k)=(( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh) ) Smax;
end
% us i ng the boundayr c ondi t i ons : v( s1 , INF, t )=Smax&v( s2 , INF, t )=Smax
i=MM; j=MM; k=MM( j 1)+MM;
A( k , k)=( vol 1 2) s1 ( i +1)2/(hh)+( vol 2 2) ( s2 ( j +1)2)/( hh)+r ;
A( k , k1)=( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh ) ;
A( k , kMM)=( r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh ) ;
b( k)=(( r1 s1 ( i +1))/(2h)( vol 1 2) s1 ( i +1)2/(2hh) ) Smax . . .
+((r2 s2 ( j +1))/(2h)( vol 2 2) s2 ( j +1)2/(2hh) ) Smax;
% Sol ve U( t )
% Use the thetamethod
I=eye (MM2 ,MM2) ; t het a =1/2; U = z e r os (MM2 ,N) ;
% the i n i t i a l val ues on the i n t e r i o r nodes
f o r i =1:MM
f o r j =1:MM
kk=MM( i 1)+j ;
U( kk , N)=V T( i +1, j +1);
end
end
%c a l c ul a t e the U wi th ti me de c r e as i ng from T to 0
f o r n=(N1): 1:1
U( : , n)=i nv ( I +(1t het a ) t i mes t ep . A) . . .
( ( It het a t i mes t ep . A)U( : , n+1)+t i mes t ep . b ) ;
% American Condi ti on
f o r i =2: (M1)
48
f o r j =2: (M1)
i f U( ( i 2)(M2)+( j 1) , n)<max( ( i 1) s t ocks t ep , ( j 1) s t oc ks t e p )
U( ( i 2)(M2)+( j 1) , n)=max( ( i 1) s t ocks t ep , ( j 1) s t oc ks t e p ) ;
end
end
end
end
% Put val ues back to the matri x V
V=z e r os (M,M) ;
f o r i =2: (M1)
f o r j =2: (M1)
V( i , j )=U( ( i 2)(M2)+( j 1) , 4) ;
end
end
% Sur f
V( : , 1) = s2 ( : ) ; V( : ,M)=Smax; V( 1 , : ) = s1 ( : ) ; V(M, : ) =Smax;
s ur f ( S1 ( 1 : 2 1 , 1 : 2 1 ) , S2 ( 1 : 2 1 , 1 : 2 1 ) ,V( 1 : 2 1 , 1 : 2 1 ) )
49
Bibliography
[1] Paul Wilmott, Je Dewynne, Sam Howison. Option Pricing Mathematical
models and computation. Oxford Financial Press; Repr. with corrections
edition (May 1, 1994).
[2] Thomas E. Copeland, J. Fred Weston, Kuldeep Shastri. Financial Theory
and Corporate Policy. Addison Wesley; 4 edition (January 10, 2004).
[3] Lishang Jiang. Mathematical Modeling and Methods of Option Pricing.
World Scientic Publishing Company; illustrated edition edition (July 30,
2005).
[4] Tomas Bjork. Argitrage Theory in Continuous Time. Oxford University
Press, USA; 2 edition (May 6, 2004).
[5] Paolo Brandimarte. Numerical Methods in Finance and Economics: a
MATLAB-based Introduction. Wiley-Interscience; 2 edition (October 6,
2006).
[6] Daniel J. Duy. Finite Dierence Methods in Financial Engineering. Wiley
(May 23, 2006).
[7] Ravi Myneni. The Pricing of the American Option. The Annals of Applied
Probability, Vol. 2, No.1.(Feb., 1992), pp. 1-23.
[8] Owe Axelsson. The Finite Dierences Method.
[9] Mark Broadie, Jerome Detemple, Cirano. The Valuation of American Op-
tions on Multiple Assets.
[10] Shameer Sukha. Finite-Dierence Methods for Pricing the American Put
Option.
50
[11] F. AitSahlia, P. Carr. American Options: A Comparison of Numerical
methods.
[12] Samuli Ikonen, Jari Toivanen. Ecient Numerical Methods for Pricing
American Options Under Stochastic Volatility.
51

Potrebbero piacerti anche