PARAMETER ESTIMATION OF THE

BLACK-SCHOLES-MERTON MODEL
by
Kubrom Hisho Teka
B.Sc., University of Asmara, 2002
M.S., Kansas State University, 2011
Ph.D., Kansas State University, 2012
A REPORT
submitted in partial fulllment of the
requirements for the degree
MASTER OF SCIENCE
Department of Statistics
College of Arts and Sciences
KANSAS STATE UNIVERSITY
Manhattan, Kansas
2013
Approved by:
Major Professor
Prof. James Neill
Abstract
In nancial mathematics, asset prices for European options are often modeled according to
the Black-Scholes-Merton (BSM) model, a stochastic dierential equation (SDE) depending
on unknown parameters. A derivation of the solution to this SDE is reviewed, resulting
in a stochastic process called geometric Brownian motion (GBM) which depends on two
unknown real parameters referred to as the drift and volatility. For additional insight, the
BSM equation is expressed as a heat equation, which is a partial dierential equation (PDE)
with well-known properties. For American options, it is established that asset value can be
characterized as the solution to an obstacle problem, which is an example of a free boundary
PDE problem. One approach for estimating the parameters in the GBM solution to the
BSM model can be based on the method of maximum likelihood. This approach is discussed
and applied to a dataset involving the weekly closing prices for the Dow Jones Industrial
Average between January 2012 and December 2012.
Table of Contents
Table of Contents iii
List of Figures iv
Acknowledgements v
1 Introduction 1
2 Option Theory 3
3 The Black-Scholes Model 5
3.1 Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 The Black- Scholes Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4 Basic Partial Dierential Equations 14
4.1 The Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2 Black-Scholes as a Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . 15
5 Free boundary Problems 17
5.1 The Obstacle Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 American Options as an Obstacle Problem . . . . . . . . . . . . . . . . . . . 20
6 Overview of estimation procedure 24
6.1 Maximum Likelihood Estimation . . . . . . . . . . . . . . . . . . . . . . . . 25
6.2 Condence Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
7 Simulations for Sample Mean and Variance of GBM 30
8 Summary 33
Bibliography 35
A Data 36
iii
List of Figures
3.1 Ten Geometric Brownian Motion Sample Path . . . . . . . . . . . . . . . . . 8
5.1 The Obstacle problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 American put - pay o function . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3 The Obstacle problem in R
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.1 Weekly Closing Prices for the DJIA between Jan 2012 - Dec 2012 . . . . . . 25
6.2 Autocorrelation Function for x(t
k
) . . . . . . . . . . . . . . . . . . . . . . . . 26
6.3 Partial Autocorrelation Function for x(t
k
) . . . . . . . . . . . . . . . . . . . 27
6.4 Probability plot for x(t
k
) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
7.1 Simulation of Sample Mean of GBM . . . . . . . . . . . . . . . . . . . . . . 30
7.2 Simulation of Sample Standard Deviations of GBM . . . . . . . . . . . . . . 31
iv
Acknowledgments
I would like to thank to all the faculty and sta of the department of Statistics, Kansas State
University. I am grateful to Prof. Louis Pigno, Chair of the department of Mathematics at
Kansas State University, and Prof. Ivan Blank for supporting me to puruse M.S in Statistics.
I am grateful for Drs. Baltus, Nanthakumar and Preston of SUNY Oswego for their support
and encouragement. I would also like to thank my M.S supervisory committee, Drs. Nelson
and Song, for their comment and feedback.
Most of all, my deepest gratitude to my advisor, Prof. James Neill, without whom none of
this would have been possible.
v
Chapter 1
Introduction
Oftentimes the mathematical modeling of numerous phenomena lead to solutions of dieren-
tial equations which involve unknown parameters. Consider, for instance, the Black-Scholes
model, or Black-Scholes-Merton model, which was rst presented in
SB
. Merton and Scholes
received the 1997 Noble prize in Economics for their work. Though ineligible for the prize
due to his death in 1995, Black was mentioned as a contributor by the Swedish academy. It
is a model for the asset prices given by the stochastic dierential equation :
dS(t) = S(t){rdt + dW(t)}, t 0, (1.1)
where S is the asset value, r 0 is the drift rate, is the volatility, W(t) is Brownian
motion, and t is time in years.
Equation (1.1) can also be written in integral form as:
S(t) = S(0) + r
_
t
0
S(s)ds +
_
t
0
S(s)dW(s), t R
+
.
In chapter 2 we introduce options. Options are rights to buy or sell underlying assets for
an exercise price (strike), which is xed by the terms of the option contract. That is, the
purchaser of the option is not obligated to buy or sell the asset. This decision will be based
on the payo, which is contingent on the underlying assets behavior. Both American and
1
European option contracts are discussed.
In chapter 3 the Black-Scholes model is derived, including a discussion of Brownian motion
and It o s integral. The Black-Scholes equation used to price European call and put op-
tions based on an asset price dynamic model. This model is a stochastic dynamic systems
model that may be written as a stochastic dierential equation. See
FM
. The solution to this
model takes the form of a stochastic process called geometric Brownian motion (GBM). The
model contains two real parameters: drift (r) and volatility ( ). In chapter 4, we express
the Black-Scholes model as a heat equation, a partial dierential equation whose properties
are well known.
We discuss free boundary problems in Chapter 5. We also establish that asset value can be
characterized as the solution to an obstacle problem. In particular, the American option is
expressed as an obstacle problem, which is an example of a free boundary problem.
In chapter 6, we discuss estimation procedures. Specically we use the method of Maxi-
mum Likelihood Estimation (MLE) to estimate the parameters r and . We also construct
a 95% condence interval for the parameters of the GBM. In chapter 7, we do a simulation
for sample mean variance of GBM. We conclude the report by giving a summary in chapter 8.
To summarize, in this report we establish relationships between Statistics, Finance and
Partial Dierential Equations.
2
Chapter 2
Option Theory
An option is the right (but not the obligation) to buy or sell a risky asset at a prespecied
xed price within a specied period. An option is a nancial instrument that allows, among
other things, one to make a bet on rising or falling values of an underlying asset. The
underlying asset typically is a stock, or a parcel of shares of a company. An option is a
contract between two parties about trading the asset at a certain future time. One party
is the writer, often a bank, who xes the terms of the option contract and sells the option.
The other party is the holder, who purchases the option, paying the market price, which is
called a premium
S
.
The holder of the option must decide what to do with the rights that the option contract
grants. The decision will depend on the market situation, and on the type of options. But
rst, a few denitions, as given in
S
, are collected that are relevant to the discussion.
The maturity date T xes the time horizon. At this date the rights of the holder expire,
and for later times (t > T) the option is worthless. There are two basic types of options.
The call option gives the holder the right to buy the underlying asset for an agreed price K,
known as the strike or exercise price , by the date T. The put option gives the holder
the right to sell the underlying asset for the price K by the date T. Thus, in summary, at
time t the holder of the option can choose to:
3
1. sell the option at its current market price on some options exchange at (t < T),
2. retain the option and do nothing,
3. exercise the option (t T), or
4. let the option expire worthless (t T).
If you are the writer of these options, you have received a premium, but may be forced to
either buy or sell the underlying asset in the future, according to the terms of the contract.
Not every option can be exercised at any time t T
S
. For European options, exercise is
only permitted at expiration time T. American options can be exercised at any time up
to and including the expiration date.
If we denote the current price of the underlying asset by S, then the payos at expiration,
T, for a given strike price, K, of European calls and puts are, respectively,
V (S, T) = max(S K, 0) = (S K)
+
and V (S, T) = max(K S, 0) = (K S)
+
.
The value of V (S, t) also depends on other factors. Dependence on the strike K and the
maturity T is evident. Market parameters aecting the price are the interest rate r, the
volatility of the price S
t
, and dividends in case of a dividend-paying asset.
4
Chapter 3
The Black-Scholes Model
The Black-Scholes option pricing formula is the most famous continuous-time derivative
pricing model which gives the price of a European put or call based on ve quantities:
the intial price of the underlying stock, which is known,
the strike price of the option, which is known,
the time to expiration, which is known,
the risk-free rate during the lifetime of the option, which is assumed to be constant
and can only be estimated,
the volatility of the stock price, a constant that provides a measure of the uctuation
in the stocks price, and thus is a measure of the risk involved in the stock. This
quantity can only be estimated as well.
In this chapter we will derive the Black-Scholes option pricing model. But rst Brownian
motion.
3.1 Brownian Motion
In 1827, just 35 years after the New York Stock Exchange was founded, an English botanist
named Robert Brown studied the motion of small pollen grains immersed in a liquid
5
medium
RO
. Brown wrote that pollen grains exhibited a continuous swarming motion when
viewed under the microscope. Brownian motion is the most important stochastic process,
being the archetype of Gaussian processes, of continuous time martingales, and of Markov
processes. It is fundamental to the study of stochastic dierential equations, nancial math-
ematics, and ltering, for example. A formal mathematical construction of Brownian motion
and its properties was rst given by the mathematician Norbert Wiener beginning in 1918,
based on Fourier series. Subsequently, martingale techniques have been employed to con-
struct Brownian motion as well.
In this section we describe Brownian motion and construct the associated It o stochastic
integral. First, some relevant denitions are given. See
P
3.1.1 Denition. A stochastic process is a family (X
t
)
tT
of random variables X
t
: R
indexed by a set T.
Real-life examples of stochastic processes include:
the time evolution of a risky asset. In this case X
t
represents the price of the asset at
time t T
the time evolution of a physical parameter. For instance, X
t
represents a temperature
observed at time t T.
Brownian motion is a fundamental example of a stochastic process. Here we work on
a probability space (, F, P), where = C
0
(R
+
) is the space of continuous real-valued
functions on R
+
started at 0.
3.1.2 Denition. Brownian motion is a stochastic process (W
t
)
tR
+
such that
1. W
0
= 0 almost surely
6
2. the sample trajectories t W
t
are continuous with probability 1
3. for any nite sequence of times t
0
< t
1
< ... < t
n
, the increments W
t
1
W
t
0
, W
t
2

W
t
1
, ..., W
tn
W
t
n1
are independent
4. for any times 0 s < t, W
t
W
s
is normally distributed with mean zero and variance
t s.
The existence of Brownian motion as a stochastic process (W
t
) is given in
P
.
Continuing the treatment as in
P
, we regard Brownian motion as a random walk over in-
nitesimal time intervals of length t, with increments W
t
over the time interval
[t, t + t] given by
W
t
=

t (3.1)
with equal probabilities.
By splitting the interval [0, T] into N intervals
(
k 1
N
T,
k
N
T], k = 1, ..., N,
of length t =
T
N
, with N large and letting
X
k
=

T =

t =

N W
t
with V ar(X
k
) = T,
W
t
=
X
k

N
=

t
is the increment of W
t
over ((k 1)t, kt] and we get
W
T

0<t<T
W
t

X
1
+ ... + X
N

N
Hence, by the central limit theorem we recover the fact that W
t
has a centered Gaussian
distribution with variance T, see point (4) of denition 3.1.2.
7
Next, to illustrate, we generate GBM sample paths with S(0) = 1, r = 1 and = 0.2. See
gure 3.1.
Figure 3.1: Ten Geometric Brownian Motion Sample Path
3.1.3 Denition. A stochastic process of the form {rt + W
t
|t 0} where r is a constant
and W
t
is Brownian motion with volatility is called Brownian motion with drift r and
volatility . If we let S
t
= e
Wt
, then the process S
t
, t 0 is said to be a Geometric
Brownian Motion.
3.1.4 Denition. A Brownian motion process {Z
t
|t 0} with drift r = 0 and volatility
= 1 is called Standard Brownian motion. In this case Z
t
has mean 0 and variance t.
If {W
t
|t 0} is Brownian motion with drift r and variance
2
, then we can write
W
t
= rt + Z
t
where {Z
t
|t 0} is Standard Brownian motion.
8
We next describe stochastic integrals, also known as It o integrals, with respect to Brownian
motion. See
K
.
Consider integrals of a non-random simple process X
t
, which is a function of t and does not
depend on W
t
. By denition a simple non-random process X
t
is a process for which there
exist times 0 t
0
< t
1
< ... < t
n
= T and constants c
1
, ..., c
n
, such that
X
t
=
n

i=1
c
i

(t
i1
,t
i
]
(t), t R
+
.
3.1.5 Denition. The Ito integral
_

0
X
t
dW
t
is dened as the sum
_

0
X
t
dW
t
=
n

i=1
c
i
(W
t
i
W
t
i1
). (3.2)
As given in
P
, the probability distribution of
_

0
X
t
dW
t
is independent of the representation
of X
t
. A relevant denition and results follow next.
3.1.6 Denition. A measurable function f L
2
(R
+
) if and only if
_

0
|f(x)|
2
dx < .
3.1.7 Proposition. For X
t
L
2
(R
+
), the integral
_

0
X
t
dW
t
has a centered Gaussian distribution
_

0
X
t
dW
t
N(0,
_

0
|X
t
|
2
dt)
and we have the Ito isometry
E{(
_

0
X
t
dW
t
)
2
} =
_

0
|X
t
|
2
dt.
Proof. See Proposition 4.1 of
P
.
By a Taylor expansion,
df(x) = f

(x)dx +
1
2
f

(x)(dx)
2
+
1
3!
f

(x)(dx)
3
+ ...
9
and by applying Taylors formula to Brownian motion
dW
t
= W
t+dt
W
t
,
and letting
df(W
t
) = f(W
t+dt
) f(W
t
),
we have
df(W
t
) = f

(W
t
)dW
t
+
1
2
f

(W
t
)(dW
t
)
2
+
1
3!
f

(W
t
)(dW
t
)
3
+ ...
By (3.1) and dropping higher order terms gives
df(W
t
) = f

(W
t
)dW
t
+
1
2
f

(W
t
)dt, for a small dt.
By integrating and applying the fundamental theorem of calculus gives the integral form of
It os formula for Brownian motion
f(W
t
) = f(W
0
) +
_
t
0
f

(W
s
)dW
s
+
1
2
_
t
0
f

(W
s
)ds.
Next a general expression of It os formula is given which applies to an It o process of the
form
X
t
= X
0
+
_
t
0
u
s
dW
s
+
_
t
0
v
s
ds, t R
+
. (3.3)
3.1.8 Theorem (Itos formula for Ito process). For any Ito process (X
t
)
tR
+
of the form
(3.3) and any f C
1,2
(R
+
, R), we have
f(t, X) = f(0, X
0
) +
_
t
0
V
s
f
x
(s, X
s
)ds +
_
t
0
U
s
f
x
(s, X
s
)dW
s
+
_
t
0
f
s
(s, X
s
)ds
+
1
2
_
t
0
|U
s
|
2

2
f
x
2
(s, X
s
)ds
Proof. See
P
10
3.1.9 Theorem. The solution to (1.1) is given by
S(t) = S(0)e
(r
1
2

2
)t+Wt
, t R
+
. (3.4)
Proof. First rewrite (1.1) in integral form as
S(t) = S(0) + r
_
t
0
S(s)ds +
_
t
0
S(s) dW(s), t R
+
.
By applying It os formula to f(S(t)) = log S(t) with f(x) = log x, we have
dlogS(t) = rS(t) f

(S(t))dt + S(t)f

(S(t))dW(t) +
1
2

2
S
2
(t) f

(S(t))dt
= rdt + dW(t)
1
2

2
dt
=
_
r
1
2

2
_
dt + dW(t).
Hence,
logS
t
logS(0) =
_
t
0
d logS(x)
=
_
t
0
(r
1
2

2
) dx +
_
t
0
dW(x)
= (r
1
2

2
)t + W(t), t R
+
,
and
S(t) = S(0)e
(r
1
2

2
)t+W(t)
, t R
+
.
Asset prices can be modeled by the following stochastic dierential equation (SDE)
R
:
dS(t) = S(t){rdt + dW(t)}, t 0,
where S is the asset value, r 0 is the drift rate, is the volatility, W(t) is Brownian
motion (as described in the preceding), and t is time in years. Because of the dW term, S
11
itself is also a random variable.
3.1.10 Lemma. If S satises the above SDE, then the value function V (S, t) satises the
following
dV = S
V
S
dW +
_
rS
V
S
+
1
2

2
S
2

2
V
S
2
+
V
t
_
dt
Proof. See
Hu
3.2 The Black- Scholes Equation
For the value function V (S, t), the Black-Scholes equation is given by
V
t
+
1
2

2
S
2

2
V
S
2
+ (r )S
V
S
rV = 0 (3.5)
where = 0 for the European option, as no dividend is paid.
The assumptions that led to (3.5) are
S
:
1. There are no arbitrage opportunities. (Arbitrage means the existence of a portfolio
which requires no investment initially, and which with guarantee makes no loss but
very likely a gain at maturity.
2. The market is frictionless. This means that there are no transaction costs (fees or
taxes), the interest rates for borrowing and lending money are equal, all parties have
immediate access to any information, and all securities and credits are available at
any time and in any size.
3. The asset price follows a geometric Brownian motion.
4. r and are constant for 0 t T. No dividends are paid in that time period. The
option is European.
12
Under the above assumptions, there exists a closed form solution to some types of option.
For some of the more advanced option contracts, the closed form solution may not be pos-
sible to determine, and thus a numerical solution is usually obtained.
A closed form solution for the European put option given by
Hu
is
P(S, t) = Ke
r(Tt)
N(d
2
) SN(d
1
)
where
N(x) =
1

2
_
x

1
2
s
2
ds, d
1
=
log
S
K
+ (r +
1
2

2
)(T t)

T t
, and d
2
=
log
S
K
+ (r
1
2

2
)(T t)

T t
.
13
Chapter 4
Basic Partial Dierential Equations
When considering a partial dierential equation, there are few things one needs to ask. The
rst question is: Do we have a well-posed problem? In other words, does a solution to the
problem exist and is it unique. And the other question: Is the solution well-behaved? In
other words, does the solution depend continuously on the initial and boundary conditions,
so that a small perturbation in the conditions does not bring a big change in the solution?
One also wants to know the regularity of the solution, that is, which conditions give the
best regularity etc. In this chapter we describe the techniques of deriving the solution of
the Black-Scholes by converting the Black-Scholes equation to a heat equation, a partial
dierential equation whose properties are well known.
4.1 The Heat Equation
The heat equation, also know as the diusion equation, in R is given by:
_
_
_
u

= u
xx
for x R, > 0
u(x, 0) = g(x).
(4.1)
By using the Fourier transform, it is known that the solution of equation (4.1) is:
14
u(x, ) =
1
2

_
R
g(y)e

(xy)
2
4
dy for < x < , > 0. (4.2)
4.2 Black-Scholes as a Heat Equation
The main purpose of this section is to express the Black-Scholes equation as a heat equation.
As described in the above section, the solution of a heat equation is known, and consequently
this will allow us to nd the solution of the Black-Scholes for a European call.
The Black-Scholes equation and boundary conditions for a European call with values V (S, t),
as described in (3.5) , is given by
V
t
+
1
2

2
S
2

2
V
S
2
+ rS
V
S
rV = 0
with
V (0, t) = 0, V (S, t) S as S ,
and
V (S, t) = (S K)
+
.
To write the Black-Scholes equation as a heat equation requires a series of change vari-
ables
DHW
. The rst change of variable is to let
S = Ke
x
, t = T 2/
2
, V = Kv(x, ).
This results in
v

=

2
v
x
2
+ (b 1)
v
x
bv and v(x, 0) = (e
x
1)
+
where b = 2r/
2
.
15
If we do a change of variable one more time by setting v = e
x+
u(x, ), for some constants
and to be specied later, we get
u +
u

=
2
u + 2
u
x
+

2
v
x
2
+ (b 1)
_
u +
u
x
_
bu.
Now set = 1/2(b 1) and = 1/4(b + 1)
2
, to obtain
v = e
1/2(b1)x1/4(b+1)
2

u(x, )
where
u

=

2
u
x
2
for < x < , > 0,
with
u(x, 0) = g(x) :=
_
e
1/2(b+1)x
e
1/2(b1)x
_
+
.
Now that we have transformed the Black-Scholes equation into a heat equation, the solution
of the Black-Scholes equation will take the form of (4.2) .
In the next chapter, the American option will be expressed as a free boundary, specically
as an obstacle problem.
16
Chapter 5
Free boundary Problems
Free boundary problems deal with solving partial dierential equations (PDEs) in a domain,
a part of whose boundary is unknown in advance. That portion of the boundary is called
a free boundary. In addition to the standard boundary conditions that are needed in or-
der to solve the PDEs, an additional condition must be imposed at the free boundary. One
then seeks to determine both the free boundary and the solution of the dierential equations.
Free boundary problems have applications in Finance, and we will express the American
option as an obstacle problem, which is an example of a free boundary problem. A brief
description of the obstacle problem is rst given.
5.1 The Obstacle Problem
What happens when we pull an elastic membrane down over an obstacle?
To formulate what is happening mathematically:
assume the membrane is given by the graph
u : B
1
R
n
R, u 0 on B
1
, and : B
1
R
n
R, < 0 on B
1
.
17
Obstacle
Membrane
Contact Set
Figure 5.1: The Obstacle problem
We want to nd a function u (the membrane) which minimizes the Area integral:
I
1
(u) :=
_
B
1
_
1 +|u|
2
among u satisfying :
u = 0 on B
1
(i.e. the membrane is pinned down) and
u in B
1
(i.e. the membrane is above the obstacle).
From the calculus of variations, it follows that functions which minimize I
1
in a neighborhood
among functions with xed boundary data satisfy the minimal surface equation. Observe
that for a small deection of the membrane, |u|
2
is the rst important term in the Taylor
expansion of
_
1 +|u|
2
. (i.e.

1 + x 1 +
1
2
x, for x small.) Thus, we want to nd a
function u which minimizes
I
2
(u) :=
_
B
1
|u|
2
i.e. Energy - The Dirichlet Integral
among u satisfying:
u = 0 on B
1
(i.e. the membrane is pinned down) and
u in B
1
(i.e. the membrane is above the obstacle).
18
It is known that functions which locally minimize I
2
satisfy Laplaces equation. Linearizing
the Area integral is standard in the study of the obstacle problem, mainly because it adds
technical simplication in that it changes the operator from nonlinear to linear, without
altering the real diculties of the problem.
Therefore, the obstacle problem involves nding a function u which solves the problem
C
:
minimize
_
B
1
|u|
2
dx among all functions u K

,
where we dene K

to be the closed convex set:

:= {u W
1,2
0
(B
1
), u }
We also have
u = 0, for u > and u 0, everywhere. (5.1)
If we dene the height function w := u , and we let f := , then w satises:
w =
{w>0}
f .
Since the problem above is variational, existence and uniqueness of solutions follows.
Regularity of the solution has been studied by many authors, and in the case where is
smooth, Frehse showed in 1972 that the solutions belong to C
1,1
. Finally, in Caarellis
famous Acta paper in 1977, the regularity of the free boundary was addressed in the case
where f was Holder continuous and positive.
Since the obstacle problem was formulated, but especially in the last 15 years, there has been
interest in extending some of these results to related problems. Ki-Ahm Lee studied the
case where the Laplacian is replaced with a fully nonlinear (but smooth) operator. Blank
studied the case where the function f was not assumed to be H older continuous. Blank and
19
Teka
BT
studied the the case where the Laplacian is replaced with a general second order
elliptic operator in nondivergence form. Many people (e.g. Blanchet, Caarelli, Dolbeault,
Monneau, Petrosyan, Shahgholian, Weiss) have recently studied the case where the Lapla-
cian is replaced with the Heat Operator.
5.2 American Options as an Obstacle Problem
The explicit formula discussed in chapter 4 is valid for European options where early ex-
ercise is not allowed. It does not necessarily give the values for American options which
has the additional feature that exercise is permitted at any time during the life of the option.
A European option can have a value that is smaller than the payo. This cannot happen with
American options
S
. If, for instance, an American put would have a value V
Am
p
< (KS)
+
,
one would simultaneously purchase the asset and the put, and exercise immediately. Thus,
there is an obvious arbitrage opportunity, even though this opportunity would not last long
before the value of the option is pushed up by the demand of the arbitragers. The same
is true for an American call if V
Am
c
< (S K)
+
. Therefore, the following constraints are
imposed:
V
Am
p
(K S)
+
and V
Am
c
(S K)
+
for all (S, t) (5.2)
Thus, there must be a value of S for which it is optimal from the holders point of view
to exercise the American option. The valuation of the American options is therefore more
complicated, since at each time we have to determine not only the option value, but also,
for each value of S, whether or not it should be exercised. This is what is known as a free
boundary problem
DHW
.
If we focus on an American put, then for 0 t T and for a contact point S
f
(t) (0, K)
20
we have
V
Am
p
(S, t) > (K S)
+
for S > S
f
(t)
and
V
Am
p
(S, t) = (K S) for S S
f
(t).
The curve S
f
is the boundary separating the area with V > payo and the area with V =
payo. The curve S
f
of a put is illustrated in the diagram below.
Figure 5.2: American put - pay o function
A priori, just as in the obstacle problem, the location of the boundary S
f
is unknown, the
curve is free. This explains why the problem of calculating V
Am
p
(S, t) for S > S
f
(t) is called
free boundary problem, and specically the obstacle problem.
Let us introduce an operator

L dened by
21

LV :=
1
2

2
S
2

2
V
S
2
+ (r )S
V
S
rV.
Thus, (3.5) can be written as
V
t
+

LV = 0.
For the case of a put, S S
f
and we have
V = K S,
V
t
= 0,
V
S
= 1,

2
V
S
2
= 0.
Thus,
V
t
+

LV = (r )S r(K S) = S rK < 0.
But since S < rK, we have
V
t
+

LV < 0.
The same holds for the call.
Thus, American options satisfy
V
t
+

LV 0.
Now let us consider the obstacle problem in R
1
. Graphically this can be described as in g.
5.2.
We express equation (5.1) as
u

= 0, for u > and u

0, u > .
It turns out we can express American options as
1. If V > payo, then Black-Scholes equation
V
t
+

LV = 0
22
Classical Obstacle problem Obstacle problem for the fractional laplacian Thin Obstacle problem main tools
Classical Obstacle problem
4u = 0 where u > , since there u is free to move
4u 0 everywhere, since the surface pushes down
u
u

Figure 5.3: The Obstacle problem in R

2
)t+Wt
, t R
+
.
Thus, for S(t) > 0,
ln(S(t)) = ln(S(0)) + (r
1
2

2
)t + W
t
, t R
+
,
and for k = 1, 2, ....,
ln(S(t
k
)) ln(S(t
k1
)) = (r
1
2

2
)(t
k
t
k1
) + (W
t
k
W
t
k1
) , t
k
R
+
(6.1)
where W
t
k
W
t
k1
is normally distributed with mean zero and variance t
k
t
k1
so that
ln(S(t
k
)) ln(S(t
k1
)) N((r
1
2

2
)(t
k
t
k1
),
2
(t
k
t
k1
))
for discrete time points t
k
.
Further, as given in
B
, the mean and variance of S(t) are, respectively,
E[S(t)] = S(0) e
rt
and Var[S(t)] = e
2rt
S
2
(0)
_
e

2
t
1
_
.
Also, as noted by
B
, to simulate this process, the continuous equation between discrete
instants t
0
< t
1
< ... < t
n
needs to be solved as follows:
24
S(t
k
) = S(t
k1
)e
[(r
1
2

2
)(t
k
t
k1
)+

t
k
t
k1
Z
k
]
(6.2)
where Z
1
, Z
2
, ...Z
n
are independent random draws from the standard normal distribution.
The parameters to be estimated are = (r, ) in the GBM process. Since GBM is a con-
tinuous time process, an approximate MLE based on discrete observations will be obtained
(see
AP
and
HN
).
6.1 Maximum Likelihood Estimation
The maximum likelihood estimation method is illustrated on data comprised of the weekly
closing prices for the Dow Jones Industrial Average for the year 2012. This data is given in
the Appendix.
Figure 6.1: Weekly Closing Prices for the DJIA between Jan 2012 - Dec 2012
25
First set x(t
k
) := ln(S(t
k
)) ln(S(t
k1
)). As indicated in gures 6.2 and 6.3, the auto-
correlation and partial autocorrelation graphs for x(t
k
) do not exhibit any signicant lags.
Thus, an assumption of independence for the x(t
k
) is tenable. Moreover, x(t
k
), satises the
assumption of normality. See gure 6.4.
Figure 6.2: Autocorrelation Function for x(t
k
)
With independence and normality tenable, the log likelihood function is given by
L() =
n

k=1
ln (f

(x(t
k
))) (6.3)
where the probability density function f

is given by
f

(x(t
k
)) =
1
S(t
k
)
_
2(t
k
t
k1
)
exp
_

_
x(t
k
) (r
1
2

2
)(t
k
t
k1
)

2
2
2
(t
k
t
k1
)
_
.
To determine

, let t := t
k
t
k1
in (6.1) . The mean and variance parameters are
26
Figure 6.3: Partial Autocorrelation Function for x(t
k
)
identied and computed as
m :=
_
r
1
2

2
_
t and v :=

2
t. (6.4)
The estimates for the GBM parameters are then deduced from the estimates of m and v
B
.
To get a closed form expression for the parameters m and v, the derivative of the above
density function is taken with respect to these parameters and setting the resulting deriva-
tives equal to zero, yielding
m =
n

k=1
x(t
k
)
n
and v =
n

k=1
(x(t
k
) m)
2
n
. (6.5)
Thus, for the dataset under consideration,
m =
52

k=1
x(t
k
)
52
= 0.00112
27
Figure 6.4: Probability plot for x(t
k
)
and
v =
n

k=1
(y (x(t
k
)) m)
2
n
=
52

k=1
(x(t
k
) 0.00112)
2
52
= 0.00023.
6.2 Condence Level
The 95% condence intervals for the parameters m and v, respectively, are given by
B
m1.96

n
m m + 1.96

n
(6.6)
and
n

2
n,0.025
v v
n

2
n,0.975
v (6.7)
where
2
n,0.025
and
2
n,0.975
are the quantiles of the chi-square distribution with n degrees of
freedom corresponding to a 95% condence level.
28
The computed 95% condence intervals for the mean and variance as given in equations
(6.6) and (6.7) , respectively are
0.00112 1.96

0.00023

52
m 0.00112 + 1.96

0.00023

52
0.003 m 0.005
and
52
594.62423519
0.00023 v
52
467.16334532
0.00023
0.00002 v 0.00003.
For t =
1
52
, and from (6.4) ,

2
=
v
t
= (0.00023)(52) = 0.01196 and hence = 0.1094
and
r =
1
2
0.01196 + (52)(0.00112) = 0.06422.
Thus,
r = 0.06422 and = 0.1094. (6.8)
In the next chapter, simulated parameter distributions are considered based on the so-called
rst parameters being taken as the estimates of r and . Such can be considered in the
absence of explicit relationships as discussed in
B
.
29
Chapter 7
Simulations for Sample Mean and
Variance of GBM
We generated 15,000 data sets each containing 52 normal random values with mean m =
0.00112 and standard deviation of v = 0.01525. This can be done in Matlab by using:
normrnd(0.00112, 0.01525, 15000, 52)
Figure 7.1: Simulation of Sample Mean of GBM
30
The mean and variance are computed for of each of the 15,000 data sets. These values are
used in (6.4) , to compute the sample estimates for the GBM parameters corresponding
to each of the 15,000 data sets. The t in this case is
1
52
as we are considering the weekly
closing prices for the DJIA.
Figure 7.2: Simulation of Sample Standard Deviations of GBM
The frequency histograms for the sample estimates are given in gure 7.1 and 7.2, respec-
tively. The sample mean, 0.06603, and sample standard deviation, 0.1086, are observed to
be very close to the closed form theoretical estimates found in (6.8) . These computations
were done in Matlab. The Matlab code is given below.
Code: For Simulation of Sample mean and Standard deviations of GBM
Z = normrnd(0.00112, 0.01525, 15000, 52);
31
R = Z

;
m = sum(R)/52;
p = m

;
for k = 1 : 15000;
for j = 1 : 52;
D(k, j) = (Z(k, j) p(k, 1)).
2
;
end;
end;
V = sum(D

)/52;
V = V

;
t = 1/52;
S = sqrt(V/t)
A = 0.5 S.
2
+ (1/t) p
32
Chapter 8
Summary
Stochastic dierential equations are utilized in nancial applications, as well as science and
engineering applications involving chemical engineering and neurobiology, for example. Such
equations generally involve unknown parameters, requiring statistical estimation techniques.
The Black-Scholes-Merton model used in nancial mathematics for European options has
been considered in this report, and the method of maximum likelihood for parameter esti-
mation used for estimating the unknown drift and volatility parameters. This approach is
discussed and applied to a dataset involving the weekly closing prices for the Dow Jones In-
dustrial Average between January 2012 and December 2012. In addition, it was established
that asset values for American options can be characterized as the solution to an obstacle
problem, which is an example of a free boundary partial dierential equation problem. For
future work, comparison of maximum likelihood estimation techniques with recent methods
involving data smoothing along with a generalization of proled estimation would be of
interest
CHR
.
33
Bibliography
[B] D. Brigo, A. Dalessandro, M. Neugebauer and F. Triki, A stochastic process toolkit for
risk management: mean reverting processes and jumps, Journal of risk management
in nancial institutions Vol. 3 (2009), no. 4-5, 6583.
[BT] I. Blank and K. Teka, The Caarelli Alternative in Measure for the Nondivergence
Form Elliptic Obstacle Problem with Principal Coecients in VMO. Submitted.
[C] L.A. Caarelli, The obstacle problem revisited, J. Fourier Anal. Appl., 4(1998), no.
4-5, 383402.
[CHR] J. Cao, D. Campbell, G. Hooker, and J.O.Ramsey, Parameter estimation for dier-
ential equations: a generalized smoothing approach, J.R. Statist. Soc., B(2007),no. 69,
Part 5, 741796.
[DHW] J. Dewynne, S. Howison and P. Wilmott, The Mathematics of Financial Derivatives,
University of Cambridge, 1995.
[FM] L. Fatone, F. Mariani, M. C. Recchioni and F. Zirilli. The use of statistical tests
to calibrate the Black-Scholes asset dynamics model applied to pricing options with
uncertain volatility. Journal of Prob and Stat. Vol. 2012, Article ID 931609.
[Hu] John C. Hull, Options, Futures, and Other Derivatives, Prentice-Hall, New Jersey,
International ed, 1997.
[HN] Hutton, J. E and Nelson, P. I, Estimation for Semi-martingales, Stoch. Proc. Appl.,
(1986),no. 22, 245257.
[K] Fima. C. Klebaner , Introduction to Stochastic Calculus with Applications, 2nd ed.,
Imperial College Press, 2005.
[AP] A. S. Pedersen, Consistency and Asymptotic Normality of an Approximate Maximum
Likelihood Estimator for Discretely Observed Diusion Processes, Bernoulli, Vol. 1.
(1995),no. 3, 257279.
34
[P] N. Privault , Notes on Stochastic Finance. Unpublished notes.
[R] M. Richardson, Numerical methods for Option pricing. Unpublished notes.
[RO] Steven Roman , Introduction to the Mathematics of Finance, 2nd ed., Springer, 2012.
[S] Rudiger U. Seydel , Tools for Computational Finance, 4th ed., Springer, 2002.
[SB] M. Scholes, F. Black. The pricing options and corporate liabilitiess. J. of Political
Economy Vol. 81(1973), 637654.
35
1

DJIA
Date t_k S(t_k) ln(S(t_k)) x(t_k)=ln(S(t_k)) - ln(S(t_k-1))
3-Jan-12 0 12,359.92 9.422214258
9-Jan-12 1 12,422.06 9.427229203 0.005014945
17-Jan-12 2 12,720.48 9.450968572 0.023739369
23-Jan-12 3 12,660.46 9.44623903 -0.004729542
30-Jan-12 4 12,862.23 9.462050389 0.015811359
6-Feb-12 5 12,801.23 9.457296539 -0.00475385
13-Feb-12 6 12,949.87 9.468841028 0.011544489
21-Feb-12 7 12,982.95 9.471392237 0.002551209
27-Feb-12 8 12,977.57 9.470977762 -0.000414476
5-Mar-12 9 12,922.02 9.466688112 -0.00428965
12-Mar-12 10 13,232.62 9.490440272 0.02375216
19-Mar-12 11 13,080.73 9.478895434 -0.011544838
26-Mar-12 12 13,212.04 9.488883814 0.00998838
2-Apr-12 13 13,060.14 9.477320123 -0.011563692
9-Apr-12 14 12,849.59 9.461067183 -0.016252939
16-Apr-12 15 13,029.26 9.474952876 0.013885693
23-Apr-12 16 13,228.31 9.490114509 0.015161632
30-Apr-12 17 13,038.27 9.475644158 -0.014470351
7-May-12 18 12,820.60 9.458808531 -0.016835627
14-May-12 19 12,369.38 9.422979343 -0.035829188
21-May-12 20 12,454.83 9.429863778 0.006884436
29-May-12 21 12,118.57 9.402494266 -0.027369513
4-Jun-12 22 12,554.20 9.43781055 0.035316284
11-Jun-12 23 12,767.17 9.454632311 0.016821761
18-Jun-12 24 12,640.78 9.444683375 -0.009948937
25-Jun-12 25 12,880.09 9.463437987 0.018754613
2-Jul-12 26 12,772.47 9.455047352 -0.008390635
9-Jul-12 27 12,777.09 9.455409002 0.00036165
16-Jul-12 28 12,822.57 9.458962178 0.003553176
23-Jul-12 29 13,075.66 9.478507766 0.019545587
30-Jul-12 30 13,096.17 9.4800751 0.001567334
6-Aug-12 31 13,207.95 9.4885742 0.0084991
13-Aug-12 32 13,275.20 9.493652912 0.005078712
20-Aug-12 33 13,157.97 9.484782938 -0.008869974
27-Aug-12 34 13,090.84 9.479668028 -0.00511491
4-Sep-12 35 13,306.64 9.496018438 0.01635041
10-Sep-12 36 13,593.37 9.517337453 0.021319015
17-Sep-12 37 13,579.47 9.516314372 -0.00102308
24-Sep-12 38 13,437.13 9.50577705 -0.010537323
Appendix A
Data
Weekly closing price for the Dow Jones Industrial Average between Jan, 2012 and Dec,
2012.
36
2

1-Oct-12 39 13,610.15 9.518571117 0.012794067
8-Oct-12 40 13,328.85 9.497686138 -0.020884979
15-Oct-12 41 13,343.51 9.498785403 0.001099265
22-Oct-12 42 13,107.21 9.480917739 -0.017867664
31-Oct-12 43 13,093.16 9.479845235 -0.001072504
5-Nov-12 44 12,815.39 9.458402071 -0.021443164
12-Nov-12 45 12,588.31 9.440523885 -0.017878187
19-Nov-12 46 13,009.53 9.473437445 0.03291356
26-Nov-12 47 13,025.58 9.474670395 0.001232951
3-Dec-12 48 13,155.13 9.484567076 0.00989668
10-Dec-12 49 13,135.01 9.483036463 -0.001530612
17-Dec-12 50 13,190.84 9.487277928 0.004241465
24-Dec-12 51 12,938.11 9.467932499 -0.01934543
31-Dec-12 52 13,104.14 9.48068349 0.012750991

37