You are on page 1of 53

Contents

1 Introduction 1
1.1 Problem overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Uncertain Volatility Model . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Stochastic exit time control problem . . . . . . . . . . . . . . . . . . 7
1.2.3 Stochastic control and dominating strategy . . . . . . . . . . . . . . . 8
1.3 Probability Spaces and Filtrations . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Brownian Motion and Wiener Process . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Single-asset Barrier Option 15
2.1 Pricing and Hedging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Practical Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Numerical simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Multi-asset Barrier Options 23
3.1 Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Hedging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 The replicating strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Two-asset barrier option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.5 Optimization problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4 Numerical solutions 35
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Numerical Scheme and algorithm . . . . . . . . . . . . . . . . . . . . . . . . 36
3
CONTENTS CONTENTS
4.2.1 Single-asset case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.2.2 Two-asset case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
0
Chapter 1
Introduction
1.1 Problem overview
The year of 1973 was a special year for nancial mathematics. In this year, the most out-
standing paper on option pricing by Black and Scholes [14] appeared. The model is known
as the Black-Scholes formula. Since then, the model has been very popular among practi-
tioners and researchers, because it is easy to use and open for development. An important
characteristic of this model is the assumption that the volatility of the underlying asset is
constant. This assumption has been argued by researchers or practitioners, especially after
the identication of the so-called volatility smile by Rubinstein and Reiner [64] who claim
that the volatility is stochastic in the real market. The constant volatility can not explain
the observed market price for options. As a result, hedging strategy using a constant value
of volatility can result in a problem.
Realizing this problem, one may try to nd another model that employs volatility which
depends on time and stock prices. However, such a model may result in a misspecication
of the volatility. El-Karoui, Jeanblanc-Picque, and Shreve [33] provided conditions under
which the Black-Scholes formula is robust with respect to a misspecication of volatility.
Romagnoli and Vargiolu in [63], Gozzi and Vargiolu [39], proposed a new method for the
study of robustness of the Black-Scholes formulae for several assets. Avellaneda, Levy, and
Paras [2] and Avellaneda and Paras [3], considered the volatility following stochastic process
but lying on an interval band [
min
,
max
]. Then bounds on the option prices are obtained
by setting the volatility equal to
min
and
max
depending on the convexity or concavity of
the option price functions.
1
1.1 Problem overview Introduction
In the case of barrier option or other options where the payo function are non-convex,
the method proposed by El-Karoui, Jeanblanc-Picque, and Shreve [33], is not applicable.
This is due to the fact that the barrier option price may not increase monotonically with
volatility. Moreover, the value function of the option is neither convex nor concave. To
sell barrier options, one generally trades them above their theoretical Black-Scholes price.
Another method used to hedge and price barrier options is by static hedging. This strategy
does not involve continuous rebalancing as in dynamic hedging. Such static hedging normally
involves setting up a portfolio at the beginning of the contract that is guaranteed to match
the payout of the options to be hedged. This method of hedging is rstly discussed by
Derman, Ergener, and Kani [30]. This paper describes a numerical algorithm for single
barrier options in the context of a binomial tree representing the evolution of a stock with
time and level dependent volatility. In a related paper, Carr, Ellis and Gupta [19] discuss
the static replication of barrier option under the Black-Scholes model. Similarly, Brown,
Hobson, and Rogers [17] demonstrate how to set up model-free overhedges and underhedges
for certain simple classes of single barrier options using a probability approach.
In this thesis, we analysis the robustness of European multi-asset barrier option. Our
work is motivated by Gozzi and Vargiolu in [39], but we discuss hedging strategy of a multi-
asset barrier option, an option governed by a multidimensional diusion process. Consider
a riskless asset M whose price is assumed to be constantly 1 for t [0, t) and d risky asset
whose vector price S
t
= (S
1
, , S
d
t
) follows the dynamic
dS
t
=

S
t

t
dW
t
, (1.1)
where W
t
is an d-dimensional Brownian motion under risk-neutral measure Q. Our main
assumption in this model is that the volatilities are stochastic, presented in a matrix process
(
t
)
t
taking the values in a closed bounded set M(d, n, R). We dene here, M(d, n, R)
is a space of d n real matrices and

S
t
is a diagonal matrix who elements are (S
1
t
, , S
d
t
).
Consider a payo function for a barrier option h(S
T
)1
{>T}
, where is the rst moment of
time where the stock price hits a bounded domain O R
d
prespecied level of barrier. The
payo function h in this case is discontinous and the value function is
v(t, x) = E
Q
_
h(S
t,x,
Tt
)1
{>(Tt)}
|F
t

which is not a convex function. Since we work on the assumption of stochastic volatility,
then the market could be incomplete. As a result, the agent can not perfectly hedge the
2
Introduction 1.1 Problem overview
volatility. Then he uses the superprice to hedge the option. Following Avellaneda, Levy, and
Paras [2] and Avellaneda and Paras [3], Romagnoli and Vargiolu [63], Gozzi and Vargiolu
[39], we x the price of the option as v
t
= v(t, S
t
) and set a self-nancing portfolio consisting
of quantity
t
of the risky asset S
t
in the hedging portfolio X
t
as

i
t
=
v
S
i
t
(t, S
t
), 0 t T.
Here v is the solution of following nonlinear PDE,
_

t
v(t, x) + H(x, D
2
x
v(t, x)) = 0, t [0, T), x R
d
v(T, x) = h(x) x R
d
v(t, x) = 0, x O
(1.2)
where H is given by
H(x, D
2
x
v) =
1
2
sup

Tr(D
2
x
v(t, x)( x
t
)( x
t
)

).
This is known as the Black-Scholes-Barenblatt equation, a version of the Hamilton-Jacobi-
Bellman equation. From the nancial point of view, in order to obtain the superprice, the
payo function h
E
Q
_
h(S
t,x,
t
Tt
)1
{>(Tt)}
|F
t

.
is maximized with respect to . By this strategy, the agent is able to protect himself against
the worst possible case.
The results of this thesis are Theorem ?? in Chapter 3. We show, by a probability
approach, that the value function of the exit control problem v is continuous with respect
to time t and space of price x, and is regular enough to apply the Ito formula. Using this
regularity we show that the pair (v, ) is a superstrategy, see Chapter 4, Theorem 3.3. In
the case of a single-asset barrier option, the exit control problem is a bang-bang solution,
see Theorem 2.1 in Chapter 5. In Chapter 3, Theorem ??, we show that
i
t
=

S
i
v(t, S
t
)
is bounded. Therefore, choosing as a superstrategy is valid and makes sense. We also
demonstrate that the value function of the exit control problem can be approximated by a
sequence of functions (v

). We show that v

satises the dynamic programming principle


3
1.2 Literature review Introduction
(DPP), and v

v as 0. We also demonstrate that v is also a viscosity solution of the


HJB equation.
This kind of approach to stochastic volatility models was initiated by Avellaneda Levy,
and Paras [2], Avellaneda and Paras [3]. They considered the case of single-asset European
with dierent maturity time in the portfolio. The recent paper by El-Kouri, Jeanblanc-
Picque, and Shreve [33] discuss a similar problem, single-asset case, but they assume that
the payo is a convex function. Under this assumption, they succeed in showing that the
BS equation is robust. This is obtained by dominating the stochastic volatility with a
deterministic function of the stock price. Romagnoli and Vargiolu [63] extend this problem
into a multi-asset European derivative. A more detail and deep discussion in this area is
given by Gozzi and Vargiolu [39].
1.2 Literature review
Our interest is in hedging strategy using super-replication method, known also as a dominat-
ing strategy, in a multi-asset barrier option of which the volatilities are unspecied. We need
to solve the Hamilton-Jacobi-Bellman equation, which is a version stochastic exit control
problem. Therefore, we need to review some topics related to this thesis.
1.2.1 Uncertain Volatility Model
According to Arbitrage Pricing Theory, if the market presents no arbitrage opportunities,
there exists a probability measure on future scenario such that the price of any security
is the expectation of its discounted cash-ows, Due [28]. Such a probability is known
as a martingale measure, Harisson and Pliska [40]. It is true that pricing measure is often
dicult to calculate precisely and there may exist more than one measure which is consistent
with a given market
1
, Avellaneda, Levy, and Paras [2]. Based on this fact, it is useful to
view incomplete markets as they are reecting the many choices for derivatives asset prices
that can exist in an uncertainty market. The source of the uncertainty mainly comes from
unpredictable volatility. Avellaneda, Levy, and Paras [2] and Avellaneda and Paras [3],
1
Uniqueness of the martingale is equivalent to market completeness. A model for security market is said
to be complete if the volatility matrix is full rank in the sense that the number of underlying asset equals to
the number of source of randomness.
4
Introduction 1.2 Literature review
assume that the underlying asset S
t
follows a diusion process with non-constant interest
rate and volatility
dS
t
= r
t
S
t
dt +
t
S
t
dW
t
. (1.3)
The volatility process (
t
) uctuates within an interval
0 <
min

t

max
. (1.4)
The volatility process (
t
) that satises (1.4) induces a unique probability measure Q = Q

on the space of prices S


t
. Let denote the set of all measures that can be induced within
the constraint (1.4). Now consider a portfolio X of d options with expiration dates t
1
t
2

t
d
1
and payo functions h
1
(S
t
1
), h
2
(S
t
2
), , h
d
(S
t
d
). Avellaneda, Levy, and Paras [2]
show that the present day worst-case scenario estimate for the buyer side is the value v(t, S
t
)
of the function
v(t, S
t
) = sup

E
Q
_
d

i:t
i
t
e

t
i
t
r
s
ds
h
i
(S
t
i
)
_
(1.5)
where E
Q
is the expectation operator with respect to the measure Q and the dynamic
price process (1.3). Then, by Itos theorem, the problem can be converted into a nonlinear
partial dierential equation (DPP), which is a version of the Black-Scholes-Barenblatt (BSB)
equation.
Now we consider the multi-asset case. Consider a riskless asset M whose price is assumed
to be constantly 1 for t [0, t) and d risky assets whose vector price S
t
= (S
1
, , S
d
t
) follows
the dynamic
dS
t
=

S
t

t
dW
t
. (1.6)
Here W is an d-dimensional Brownian motion under risk-neutral measure Q. The matrix
process () takes the values in a closed bounded set M(d, n, R), and

S
t
is a diagonal
matrix who elements are (S
1
t
, , S
d
t
). Such a problem has been considered by Lyons [55],
Romagnoli and Vargiolu [63], Gozzi and Vargiolu [39]. The fair price v(, ) of the multi-asset
contingent claim at time t 0 can be found via the non-arbitrage principle, Due [28]:
v(t, x) = sup

E
Q
_
h(S
t,x,
t
Tt
)

. (1.7)
Then v must satisfy the associated Black-Scholes-Barenblatt (BSB) equation which is a
5
1.2 Literature review Introduction
version of the Hamilton-Jacobi-Bellman (HJB):
_

t
v(t, x) +
1
2
sup

Tr(D
2
x
v(t, x)( x
t
)( x
t
)

) = 0, t [0, T), x R
d
v(T, x) = h(x) x R
d
(1.8)
where D
2
x
v(t, x) =
_

2
x
i
x
j
v(t, x)
_
ij
.
The situation becomes more complex if the multi-asset European options and barrier
options are combined in a portfolio. Let be the rst moment of time when the stock price
hits the prespecied level of barrier. Then, the price of the barrier options is given by
v(t, x) = sup

E
Q
_
h(S
t,x,
Tt
)1
{>(Tt)}

. (1.9)
The standard approach to hedge this option is to model the behaviour of the underlying
asset and from the model, a fair price then derived. This price is expected to be perfectly
hedging the contingent claim. The approach is to maintain an ever-changing position in
the underlying assets. This method is known as dynamic hedging which needs continuous
rebalancing: maintaining the tracking error, that is the dierence between the actual value
and the theoretical value of a self-nancing portfolio, to be always positive. In works by
Avellaneda Levy, and Paras [2], Avellaneda and Paras [3], and El-Kouri, Jeanblanc-Picque,
and Shreve [33], the maximum volatility always overestimates the claim. This does not
happen if our claim is a barrier option. The maximum volatility sometimes underestimates
the contingent claim price. This is due to the fact that the option price function can not be
convex or concave for all prices x and all times t. Therefore, the option price is not increasing
monotonically as the volatility increases.
The problem now comes from the fact that the value function might not be smooth
enough to satisfy the boundary condition v = 0 in the classical sense. To solve this problem,
Crandall and Lions [21], Crandall, Ishii, and Lions [22], Lions [52],[53], [54], Fleming and
Soner [34] introduces the notion of a viscosity solution. This notion is not only giving a sense
of a continuous solution for the value function v, but also a sense of a discontinuous viscosity
solution. Equation (1.8) is a version of a stochastic exit control problem which is extensively
studied, see for example Barles and Burdeau [6], Barles and Rouy [7], Barles and Perthame
[8],[9], Barles and Souganidis [10]. For the non-degenerate case of (1.12), one may refer to
6
Introduction 1.2 Literature review
Bensoussan [11], Fleming and Rishel [35], Fleming and Soner [34], Krylov [49], where the
classical PDE approaches for stochastic control problem are discussed.
1.2.2 Stochastic exit time control problem
For the exit control problem the main diculty comes from the treatment of the boundary
condition. As suggested by Lions [52] or Barles and Rouy [7], it is rather simple to create
examples in which the value function is continuous in an open set O and can be extended
continuously to

O, but where its extension does not satisfy the boundary condition. Problems
may also occur when the diusion degenerates along the normal direction to the boundary.
Following Barles and Rouy [7], it is necessary to relax the boundary condition which has to
be read as
min
_

t
v(t, x) + H(x, D
2
x
v(t, x), v(t, x)
_
0 on O
and
max
_

t
v(t, x) + H(x, D
2
x
v(t, x), v(t, x)
_
0 on O.
These inequalities have to be understood in the viscosity sense. The viscosity notion has been
well presented in the Users Guide of Crandall, Ishii and Lions [22], including the presentation
of these boundary conditions. This type of boundary condition was rst considered by Lions
[52], Barles and Perthame [8],[9], Barles and Souganidis [10]. A more general denition
of viscosity solution, including equations with discontinuous Hamiltonian is discussed in
Fleming and Soner [34], Bardi and Capuzzo-Dolcetta [4]
There are two strategies to solve the this type of problem: the continuous approach
and the discontinuous one. Lions [52],[53], [54], uses the continuous approach to solve this
problem. He approximates the exit control problem by a continuous function. Then he
solves the problem analytically using PDE theory. The discontinuous approach is more
likely a control problem rather than a PDEs one, where one needs to apply the techniques
of variational inequalities, relaxed control and weak convergence. See for examples Barles
and Burdeau [6], Barles and Rouy [7], Barles and Perthame [8], or Flaming and Soner [34].
The continuous approach discussed by Lions [52],[53], [54] suggests that one rst prove
that the value function satises the Dynamic Programming Principle (DPP), [52]. Then
from DPP, he derives the Hamilton-Jacobi-Bellman equation to which the value function
is a viscosity solution. When the non-degenerate condition is assumed on the boundary,
7
1.2 Literature review Introduction
one may prove that the value function is the unique continuous solution to the exit control
problem by using a Strong Comparison Theorem ( Barles and Burdeau [6], Barles and Rouy
[7]), or one may refer to Users Guide of Crandall, Ishii and Lions [22]. Once the strong
comparison theorem is established, one can obtain the existence theorem by the Perrons
method known as Ishii Lemma, see Fleming and Soner [34].
The most recent method to solve this problem uses the notion of the L
p
-viscosity solution.
One reason for using L
p
-viscosity solutions is that H does not need to be continuous in x as
in [22]. It has been shown by Caarelli, Crandall, Kocan, and Swiech in [25] that when the
equations are degenerate, uniqueness for viscosity solutions fails even for the one dimensional
case, see example 2.4 in [25]. However, it has been noted by Caarelli et. al [25] that viscosity
solutions are L
p
-viscosity solutions whenever H is continuous, see Proposition 2.9 in [25].
Note that a comprehensive treatment of the L
p
-theory of fully nonlinear parabolic equations
is presented in a recent paper by Crandall, Kocan, and Swiech [24],[25].
1.2.3 Stochastic control and dominating strategy
The application of the stochastic control theory to nancial markets is relatively new in
applied mathematics. Intensive research in this area has been initiated by Merton [58].
In his paper, the optimal wealth and the optimal consumption-rate process are formulated
as a stochastic control model with the logarithmic value function. Later, Cvitanic and
Karatzas [20] published a paper dealing with the hedging contingent claim using the theory
of stochastic control. Hedging contingent claims based on the idea of dominating strategy in
the case of incomplete market, was discussed by Cvitanic and Karatzas [26], and El-Karoui
and Quenez [32]. The price of a contingent claim obtained by this method is called the
upper-hedging price or super price. Cvitanic and Karatzas [26] and El Karoui and Quenez
[32] demonstrate that the upper hedging price, which is a minimization problem, can be
transformed to a dual maximization problem. Similar transformation has also been done by
Schmock, Shreve, and Wystup [65], but in a dierent problem.
The solution of the BSB equation gives rise to an optimal dominating strategy for deriva-
tive securities. This result implies that the BSB equation is the dynamic programming
equation for the following control problems:
v

(t, x) = inf

t
A()
E
Q
_
h(S
t,x,
t
Tt
)

(1.10)
8
Introduction 1.3 Probability Spaces and Filtrations
and
v
+
(t, x) = sup

t
A()
E
Q
_
h(S
t,x,
t
Tt
)

. (1.11)
The BSB equation corresponding to the function v
+
in (1.11) can be written in the form
_

t
v
+
(t, x) +
1
2
sup

Tr(D
2
x
v
+
(t, x)( x
t
)( x
t
)

) = 0, t [0, T), x R
d
v
+
(T, x) = h(x) x R
d
. (1.12)
1.3 Probability Spaces and Filtrations
In this subsection, we will dene Brownian motion and the associated mathematical model
known as the Wiener process. Our discussion closely follows Durrett [29], and the book
written by Krylov [50], for the basic theory of diusion processes. We also refer to Karatzas
and Shreve [48], and Oksendal [60] for an introduction to stochastic dierential equations.
A probability space is a triple (, F, P) where is the space of elementary events or
outcomes, F is a -algebra of subsets of and P : F [0, 1] is a probability measure. A
collection of -algebras F = (F
t
)
t0
, satisfying
F
s
F
t
F
for all s t is called a ltration.
Denition 1.1. A ltration F = {F
t
} is said to satisfy the usual conditions if it is right-
continuous and F
0
contains all the P-negligible events in F.
A random variable X is a real-valued function dened on , such that for every Borel
set B B(R), we have X
1
(B) = { : X() B} F. Let us recall that the Borel
-algebra B(R) is the smallest -algebra containing the open sets of R. A stochastic process
{X
t
; 0 t T} is a family of R
d
-valued random variables dened on (, F, P). The
stochastic process {X
t
; 0 t T} is called measurable if, for every B B(R
d
), the set
{(t, ); X
t
() B} belongs to the product -algebra B([0, )) F, in other words, if the
mapping
(t, ) X
t
() : ([0, ) , B[0, ]) F) (R
d
, B(R
d
))
is measurable.
9
1.4 Brownian Motion and Wiener Process Introduction
We will denote by F
X
t
= (X
s
; s t) the smallest -algebra such that X
s
is F
X
t
-
measurable for all s t. We say that a stochastic process (X
t
, 0 t T) is adapted to the
ltration (F
t
) if X
t
is a F
t
-measurable random variable for each t 0. A stochastic process
(X
t
, 0 t T) is called progressively measurable with respect to the ltration (F
t
) if, for
each t 0 and B B(R
d
) , the set {(s, ); 0 s t, , X
s
() B} belongs to the
product -algebra B([0, t)) F
t
, in other words, if the mapping
(s, ) X
s
() ; ([0, t] , B[0, t]) F
t
) (R
d
, B(R
d
))
is measurable, for each t 0.
Denition 1.2. A random variable X is square integrable if EX
2
< . A process (X
t
, t 0)
is square integrable if sup
t0
EX
2
t
< . If (X
t
, t 0) is considered on a nite time interval
0 t T, then it is square integrable if sup
0tT
EX
2
t
< .
1.4 Brownian Motion and Wiener Process
Brownian motion is one of the most important objects in the theory of stochastic processes.
It has been applied in many branches of science and engineering. Recently, it has also been
widely applied in the eld of mathematical nance. One may refer to [50] or [48] for a
construction of Brownian motion. There are many ways to dene the Brownian motion. We
adopt the point of view that Brownian motion is dened on the space = C([0, ), R
d
)
equipped with a Wiener measure P, see Krylov [50] for details. This version is known as
canonical Brownian motion. The mathematical formulation of Brownian motion is known as
the Wiener process
2
. The mathematical denition of the Wiener process is given as follows:
Denition 1.3. An F-adapted process {W
t
: 0 t T} dened on a probability space
(, F, P) is called a Wiener process on [0, T] if it has the following four properties:
(i) W
0
= 0.
(ii) The increments of W
t
are independent; that is for any nite set of times, the random
variable
W
t
W
s
is independent of F
s
for any s, t such that 0 s t T.
2
In honor to N. Wiener who rstly constructed a mathematical model of Brownian motion
10
Introduction 1.5 Martingales
(iii) For any 0 s t T the increment W
t
W
s
has a Gaussian distribution with mean
0 and variance t s.
(iv) For all in a set of probability one, W
t
() is a continuous function of t.
1.5 Martingales
The aim of this subsection is to summarize the basic theory of martingales which is related
to our discussion. Our discussion in this section follows closely Steele [69]. We also refer to
Musiela M and M. Rutkowski, [59] for theorems which are relevant to Non-arbitrage pricing.
Denition 1.4. A stochastic process {X
t
, t 0} which is adapted to a ltration F is a
martingale with respect to F = (F
t
) if
1. E(|X
t
|) < for all 0 t T,
2. E(X
t
|F
s
) = X
s
, a.s. for all 0 s t T.
Denition 1.5. A stochastic process {X
t
, t 0}, adapted to a ltration F is a supermartin-
gale (submartingale) if it is integrable, and for any t and s, 0 s t T,
E(X
t
|F
s
) ()X
s
a.s.
Remark 1.6. If (X
t
) is a supermartingale, then (X
t
) is a submartingale.
Remark 1.7. The Wiener process {W
t
, 0 t T} is a square integrable martingale, since
EW
2
t
= t
Denition 1.8. A stochastic process {X
t
, t 0} is said to be uniformly integrable, if

[|X
t
|a]
|X
t
| 0 uniformly in t, as a
Teorema 1.1. (Doobs martingale)
Let Y be an integrable random variable, that is E|Y | < , and dene
M
t
= E(Y |F
t
).
Then M
t
is uniformly integrable.
The next result is an obvious consequence of the Burkholder-Davis-Gundy theorem.
11
1.5 Martingales Introduction
Teorema 1.2. Burkholder-Davis-Gundy
For every p 0, there exist two constants c
p
and C
p
such that, for all continuous local
martingales M vanishing at zero,
c
p
E
_
M, M
p/2

E[(M

)
p
] C
p
E
_
M, M
p/2

where M

t
= sup
st
|M
s
|.
Corollary 1.9. Let
M
t
=

tT
0
F
s
dW
s
.
Then
c
p
E
_
T
0
|F
t
|
2
dt
_
p/2
Esup
tT
|M
t
|
p
C
p
E
_
T
0
|F
t
|
2
dt
_
p/2
.
Denition 1.10. A random variable 0 is a stopping time , with respect to the ltration
F, if the event { t} belongs to the -algebra F
t
for every t 0. A stopping time is an
optional time of the ltration if { < t} F
t
for every t 0.
Proposition 1.11. Let X = {X
t
, 0 t < } be a progressively measurable process, and
let be a stopping time of the ltration F. Then the random variable X

dened on the
set { < } F

, is F

-measurable, and the stopped process {X


t
, F
t
, 0 t < } is
progressively measurable.
The proof of Proposition 1.11 is given in Karatzas and Shreve [48] page 9.
Teorema 1.3. Optional Sampling Theorem
Let (X
t
) be a martingale and a stopping time. Then the stopped process (X
t
) is a
martingale. In particular, for any t, EX
t
= EX
0
.
Denition 1.12. An adapted process (X
t
) is called a local martingale if there exists a
sequence of stopping times
n
, such that
n
as n and for each n, (X
t
n
) is a
uniformly integrable martingale.
Denition 1.13. Quadratic Variation of Martingales
Let
n
= max
i
(t
n
i+1
t
n
i
) as n . The Quadratic variation of a process (X
t
) is
dened as a limit in probability
X
t
= lim

i=1
(X
t
n
i
X
t
n
i1
)
2
. (1.13)
12
Introduction 1.5 Martingales
If (X
t
) is a martingale, then (X
2
t
) is a submartingale. By compensating X
2
t
by an
increasing process, it is possible to make it into a martingale. The process which compensates
X
2
t
to form a martingale turns out to be the quadratic variation of process X
t
.
Teorema 1.4. If (X
t
) is a local martingale, then X, X
t
exists. Moreover X
2
t
X, X
t
is
a local martingale.
13
1.5 Martingales Introduction
14
Chapter 2
Single-asset Barrier Option
In this chapter we discuss the single asset case, which is another important result of this
thesis. My result here is original and it might be one of some contributions of this thesis.
2.1 Pricing and Hedging
Let us consider a very common example of barrier option, that is the knock out and up call
of the European type. If 0 t T, S
t
= x and the call has not knocked out prior to time
t, then the price process for this option is given by an adapted process, {v
t
; 0 t T},
satisfying
v
T
= (S
T
K)1
{>T}
.
Here K is the strike price of the option and is the rst moment of time when the process
S
t
hits the barrier H, dened by
= inf{t 0; S
t
H}. (2.1)
Assuming that P is already the risk neutral measure and 0 < K < H, the value of the
knock-out barrier option at time t with initial stock price x is given by
J(t, x; ) = E
_
(S
t,x,
Tt
K)
+
1
{>(Tt)}

, 0 t T. (2.2)
For a constant volatility,
t
= , the explicit solution of (2.2) can be derived by the method
of reection principle of Brownian motion, Rich [62]. One may refer to Rich [62] for the
closed form solution of (2.2).
15
2.1 Pricing and Hedging Single-asset Barrier Option
As is shown in the previous chapter, Theorem 3.1, (2.2) is also the solution of partial
dierential equation

t
v(t, x) +
1
2

2
x
2

2
x
2
v(t, x) = 0, 0 t < T, 0 x < H (2.3)
with terminal and boundary conditions
v(T, x) = (x K)
+
, 0 x < H (2.4)
v(t, x) = 0, x H, 0 t T. (2.5)
Now assume that the true volatility is limited to move in a certain interval, i.e.

t
I = [
min
,
max
].
The assumption that (
t
) is adapted to F makes it functional of the Brownian paths {W
t
, 0
t T}, so that it is dependent on the past of the Brownian motion or stock price. This
volatility can be interpreted as a control to nd the worst and the best case price of the
barrier option. Since the seller does not know the true volatility, he will estimate the fair
price of the claim within an interval of prices, which is known as the interval of admissible
prices. Therefore, we expect that the price of the claim lies in the interval
v

(t, S
t,x,
t
) v

t
v
+
(t, S
t,x,
t
), (2.6)
where
v

(t, S
t,x,
t
) = inf
I
E
_
(S
t,x,
Tt
K)
+
1
{>(Tt)}

(2.7)
and
v
+
(t, S
t,x,
t
) = sup
I
E
_
(S
t,x,
Tt
K)
+
1
{>(Tt)}

. (2.8)
As already discussed in the previous chapter, in order to have superstrategy, we x the price
v
t
of the option and the quantities
t
of the risky asset S
t
in the hedging portfolio X
x,,v
t
as
v
t
= v(t, S
t,x,
t
),
t
=

x
v(t, S
t,x,
t
), 0 t T (2.9)
where v is the solution of the HJB equation

t
v(t, x) +
1
2
sup
I

2
t
x
2

2
x
2
v(t, x) = 0, 0 t < T, 0 x < H (2.10)
with terminal and boundary conditions
v(T, x) = (x K)
+
, 0 x < H (2.11)
v(t, x) = 0, x H, 0 t T. (2.12)
16
Single-asset Barrier Option 2.1 Pricing and Hedging
Therefore, the portfolio process satises
d(X
x,,v
t
) =
t
S
t,x,
t
dW
t
.
Initially, at t = 0, take
X
x,,v
0
= v(0, S
0
).
Then
X
x,,v
T
= v( T, S
t,x,
T
)
with terminal and boundary conditions
v(T, x) = (x K)
+
, if > T (2.13)
v(t, H) = 0, if T. (2.14)
Remark 2.1. In Theorem ??, we have shown that in order to have delta hedging admissible,
we have to impose the condition
E

T
0

2
t
dt < .
Moreover, in the case when

t
=

x
v(t, S
t,x,
t
), 0 t T,
then X
x,,v
t
is a supermartingale.
Proposition 2.2. Suppose that v is a solution of (2.10)-(2.12) for any convex payo function.
Then v is not convex or concave in x for any t > 0.
Proof. We prove by contradiction. Suppose that v is convex or concave for all t and x. Note
that v is positive for x < H and v approaches zero when x H and x 0 for every xed
time t. Therefore, it must be concave. However, if t approaches T, v(T, x) = h(x) which is
a convex function. This produces a contradiction.
Teorema 2.1. Let v be a solution of the HJB equation (2.10) with terminal condition (2.11)
and boundary condition (2.12), and dene

t
(x) =
_

max
if

2
x
2
v(t, x) > 0

min
if

2
x
2
v(t, x) < 0.
(2.15)
Then
t
is an optimal bang-bang control, v is the superprice and is the superstrategy.
17
2.2 Practical Issues Single-asset Barrier Option
Proof. Since v is a unique solution of the HJB equation (2.10)-(2.12), then v is the optimal
price; see Chapter 4. Then, clearly
t
is an optimal control, because HJB is computed with
supremum.
2.2 Practical Issues
The contingent claim h(x) = (x K)
+
1
{>T}
is discontinuous at the barrier H. This results
in an unbounded delta hedging at the maturity of the barrier option. The large delta hedging
may cause instability in the hedging strategy (See gure 2.1). The delta hedging becomes
14 15 16 17 18 19 20 21 22 23 24 25 26
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Stock Price (S)
O
p
t
i
o
n

P
r
i
c
e

v
(
t
,
S
)
Call and Up Barrier Option
3 weeks to maturity
2 weeks to maturity
1 week to maturity
Figure 2.1: The barrier option price given by (2.10)- (2.12) with K=20, H=23, = 0.20
very negative near the barrier,
t
=

x
v(t, x) as t T.
If our portfolio consists of a non-risky asset invested in a money market and risky assets
in a stock, then in the case where the stock price does not cross the barrier, the seller covers
this short position with funds shares in the stock. If the stock price hits the barrier and the
option is knocked out, the hedging strategy is in the region where
t
is large and negative.
In this case, the seller covers his short position with the money market.
To avoid the large delta being taken, one can put a constraint on the hedging portfolio
and then use this constraint to bound the super-replication strategy. This approach has been
18
Single-asset Barrier Option 2.3 Numerical simulation
suggested by Schmock, Shreve, and Wystup [65]. They impose constraints on the delta and
show that the cheapest super-replicating claim that satises this constraint can be found as
the solution of a dual problem of a stochastic control problem. Another method to avoid
instability in the hedging strategy is proposed by Shreve [68], Chap.20, p.218. He imposes
the boundary condition
v(t, x) + H

x
v(t, x) = 0, x H, 0 t T,
instead of
v(t, H) = 0 x H, 0 t T
where is a tolerance parameter. This approach guarantees that the H
t
remains bounded
and the value of the portfolio is always sucient to cover a hedging error within H
t
of
the short position.
2.3 Numerical simulation
In this subsection, we consider a numerical example which illustrates the previous discussion.
In particular, we generate a Call and Up barrier option of European type with strike price
K =$20 and barrier H =$23. Since the true volatility is not known, we expect the volatility
to be moving within interval [
min
,
max
] = [0.10, 0.20] and option expiration at T = 0.25
year. Then we use the HJB equation (2.10)- (2.12) to calculate the superprice. We also
assume that we initially can buy or sell the option at the mid volatility, (
max
+
min
)/2.
Here we report in Figure 2.2 the subprice and superprice barrier option computed using
explicit schemes (the algorithm is given in the next chapter).
Figure 2.2 illustrates a comparison between the extreme prices that are obtained by
pricing with a constant volatility, linear PDEs and those obtained from the BSB equation.
Since the extreme prices for options are obtained by using the two extreme volatilities, one
might believe that the extreme price for the portfolios would be given by the Black-Scholes
prices with some constant volatility in the range
min

max
. As shown in Figure
2.2, the theoretical price calculated by the Black-Scholes formula is too low to enter into
a delta-hedging strategy that protects against the worst case situation. The superhedging
strategy obtained from the BSB equation would protect the hedger against the movement
of the volatilities within the band.
19
2.3 Numerical simulation Single-asset Barrier Option
14 16 18 20 22 24 26
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Stock Price (S)
O
p
t
i
o
n

P
r
i
c
e

v
(
t
,
S
)
Call and Up Barrier Option
20%
18%
12%
10%
Superprice
Subprice
Figure 2.2: The dotted lines represent the superprice and subprice of the barrier option computed by
(2.10)- (2.12) and the solid lines represent the extreme value of the option computed by the linear equation
(2.3)-(2.5).
The following gure shows superstrategy with 1-3 months to maturity, computed using
equation (2.9). It shows how delta of the portfolio superstrategy varies as the option gets
closer to maturity.
20
Single-asset Barrier Option 2.3 Numerical simulation
15 16 17 18 19 20 21 22 23
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Stock Price (S)
S
u
p
e
r
s
t
r
a
t
e
g
y
,

D
e
l
t
a
Call and Up Barrier Option
1 month to maturity
3 months to maturity
2 months to maturity
Figure 2.3: Delta superstrategy computed from the superprice given in gure 2.2
21
2.3 Numerical simulation Single-asset Barrier Option
22
Chapter 3
Multi-asset Barrier Options
3.1 Pricing
Based on the position of the barrier, we categorize the multi-asset barrier options into three
dierent types. The rst one is the external barrier option. The value of the option depends
on the value of another asset. If the tradeable stock hits a certain level of barrier then the
value of such an option is zero. The second one is the basket barrier option. The value
of this option depends on whether the underlying assets in the basket hit a certain level of
barrier or not. The third one termed the max/min barrier option is a barrier option where
the value of the option depends on whether the maximum of the underlying assets hits a
certain level of barrier or not. To begin with let us dene a price process for the multi-asset
barrier option.
Denition 3.1. A price process for a barrier option is any adapted process {v
t
; 0 t T}
satisfying
v
T
= h(S
t,x,
t
T
)1
{>T}
, a.s.
where h : R
d
+
[0, ) is a given function and is the rst moment of time when S
t
hits
the barrier, dened as
= inf{t > 0; S
t
O}. (3.1)
Here S
t
is the solution of (??), O R
d+1
and O is the boundary of O
As an illustration, we rstly, discuss the class of multi-asset barrier options whose prices
depend on an external barrier variable.
23
3.1 Pricing Multi-asset Barrier Options
The external barrier determines whether the option is knocked out when the stock price
breaches the prespecied level, or stays alive until the expiry time T. The valuation of
multi-asset barrier options with a single-sided external barrier has been discussed in several
papers, for example in Heynen and Kat [42], who have presented analytic valuation formulae
for European-style barrier options with a single barrier. A similar problem, but with the
external barrier following an exponential function, has also been discussed by Kwok, Wu,
and Yu [51]. They employ the method of images to nd the Green function of the governing
dierential equation. In this subsection, we follow closely the discussion by Wong and Kwok
[73], but we are not interested in the analytic valuation of the option as discussed in this
working paper.
We propose here a payo function for a multi-asset barrier option with an external barrier
in which the terminal payo is characterized by
h(S
T
) = (max(S
2
T
, , S
d
T
) K)
+
1
{>T}
.
We adopt the usual Black-Scholes assumptions on the capital market and we assume that
the volatilities are xed. In the risk-neutral assumption, the stock price S
i
t
, i = 1, , d
follow the lognormal diusion processes. Let
ij
denote the correlation coecients between
dW
i
and dW
j
which are constant. We dene
x
i
=
1

i
ln
S
i
t
S
i
and
i
=
i
/2, i = 1, 2, , d.
Let H denote the upper barrier. The call option will be knocked out when S
1
t
H at any
time before expiry time T. We dene

H =
1

1
ln
H
S
1
.
The value of the multi-asset barrier option with barrier level H is given by
v(t, x) = E
_
h(S
x,
Tt
)1
{>(Tt)}


H
0

D
d1
(x, T, )(max(S
2
T
e

2
x
2
, , S
d
T
e

d
x
d
) K)
+
dx
d
, , dx
1
0 t T. (3.2)
Here D
d1
is the domain in the (d 1)-dimensional (x
1
, , x
d
) - plane in which
max(S
2
T
e

2
x
2
, , S
d
T
e

d
x
d
) > K is attained. Let D
d1
i
denote the domain in which S
i
T
e

i
x
i
24
Multi-asset Barrier Options 3.1 Pricing
is the maximum of the (d 1) -quantities S
2
e

2
x
2
, , S
d
e

d
x
d
. Then D
d1
i
is given by
D
d1
i
=
_
(x
2
, , x
d
) : x
i

i
ln
K
S
i
x
i

i
x
j

i
ln
S
i
S
j
, j = 2, , d, j = i
_
, i = 2, , d.
Following Wong and Kwok [73], (x, T, ) is the fundamental solution to the d-dimensional
Fokker-Plank equation

t
=
1
2
d

i=1
d

j=1

ij

x
i
x
j

j=1

x
j
,
x
1
< H, < x
j
< , j = 2, , d, t > 0. (3.3)
This formulation gives rise to the analytical evaluation of the expectation integral in many
dimensions. This is beyond our discussion. Instead we convert the problem into the partial
dierential equation given by the following theorem.
Teorema 3.1. Suppose that v is a solution of the partial dierential equation

t
v(t, S
t
) +
1
2
tr(D
2
x
v(t, S
t
)(

S
t
)(

S
t
)

) = 0, 0 t T, S
1
t
H, (3.4)
with terminal and boundary conditions
v(T, S
1
T
) = h(S
1
T
), S
1
t
< H, 0 t T (3.5)
v(t, S
1
t
) = 0, S
1
t
H, 0 t T. (3.6)
Then v is given by (3.2).
Proof. The proof is standard in terms of its application to the exit control problems. How-
ever, we here is dealing with its application to the pricing of barrier option. So, in view of its
application to the nancial problem, the proof is original. To simplify notation, we denote
the operator
L =
1
2
tr[(

)D
2
x
v(t, S
t
)].
By the Markov property, the stochastic process e
r(T)
v( T, S
T
) is a martingale under
Q. Without loss of generality, take r = 0, by applying Itos formula to (3.2) then integrating
from t to T, we have
v( T, S
T
) = v(t, S
t
) +

T
t
_

t
v(r, S
r
) +Lv(r, S
r
)
_
dr
+

T
t

x
v(r, S
r
)(

S
r
)

dW
r
(3.7)
25
3.2 Hedging Multi-asset Barrier Options
Since v( T, S
T
) and the last term of (3.7) is also a martingale, then the Reimann integral

T
t
_

t
v(r, S
r
)) +Lv(r, S
r
)(

S
t
)

dW
r
_
dr (3.8)
is also a martingale. Therefore

t
v(r, S
r
) +Lv(r, S
r
) = 0, 0 r T, S
1
r
< H, (3.9)
which also satisfy the Black-Scholes equation with nal and boundary conditions (3.5)-
(3.6).
3.2 Hedging
The barrier option sellers objective is to nd a strategy, i.e. an amount , which enables
him to make a good commitment to hedge the contingent claim h(x) at time t = T. He is
expecting that his starting wealth, x will increase such that he can cover his obligation
X
x,
T
h(S
x,
T
)1
{>T}
a.s.
where X
x,
T
is the solution of the linear stochastic dierential equation
dX
x,
t
=
n

i=0

i
t
dS
i
t
, t T. (3.10)
which is
X
x,
t
= x +

t
0
n

i=1

i
u
dS
i
u
, 0 t T. (3.11)
In other words, given a contingent claim h(S
x,
T
)1
{>T}
, we consider the smallest price that
the seller can accept from the buyer at the initial contract which will enable the seller to
cover his obligation at the nal contract T without any risk, in the sense of (3.11). This is
denoted by
v
+
(t, x) = inf
_
x 0 | such that X
x,
T
h(S
x,
Tt
)1
{>(Tt)}
; P a.e.
_
; (3.12)
This price v is the smallest amount such that X
x,
T
super-replicates h(S
x,
T
)1
{T}
. Any such
satisfying the above condition (3.12) is also called a superstrategy or a superprice.
The buyers objective is to nd a portfolio strategy so that the payment that he receives
at time T makes it possible for him to cover the debt he incurred at time t = 0. The largest
26
Multi-asset Barrier Options 3.3 The replicating strategy
amount x 0 that enables the buyer to achieve this is called the subhedging price for the
contingent claim h(S
x,
T
)1
{>T}
, and denoted by
v

(t, x) = sup
_
x 0 | such that X
x,
T
h(S
x,
Tt
)1
{>(Tt)}
; P a.e.
_
. (3.13)
The price v(t, x) is the largest amount that the buyer can aord to pay at initial contract,
which guarantees that this amount can cover the debt at nal contract without any risk.
Any such satisfying the above condition (3.13) is also called a substrategy or a subprice.
In order to ensure that the portfolio is self-nanced, we should set X
x,
t
= v(t, S
x,
t
) for
all t [0, T]. Then by Itos formula, we have
dX
t
= X
t
, dW
r
+

t
v(t, S
x,
t
) +Lv(t, S
x,
t
). (3.14)
We also know, from the denition of the self-nancing value of the portfolio , that
dX
x,
t
=
d

i=1

i
t
dS
i
t
.
This gives

i
r
=

x
i
v(r, S
t,x,
), 0 r T, i = 0, , d;
then
X
x,
T
= v( T, S
x,
T
) (3.15)
subject to
_

_
v(T, S) = h(S
x,
T
) if x O
v(t, x) = 0 if x / O.
(3.16)
3.3 The replicating strategy
We assume that the volatilities are stochastic, but restricted to move within an admissible
set A(). In the real situation the agent does not know the true volatilities, instead he uses
another model, that is,
dS
i
t
= S
i
t
n

j=1

ij
t
dW
j
t
, i = 1, , d (3.17)
to hedge the contingent claim, where A() is a certain admissible volatility, S
i
t
= x
i
is
the initial condition and x = (x
i
) is a vector in R
d
. These volatilities can be interpreted as a
control to nd the worst or best case price of the multi-asset barrier option. In this model,
there are two sources of uncertainty, that is W
t
and the volatility . Since the agent does
27
3.3 The replicating strategy Multi-asset Barrier Options
not know these two objects, he will estimate the fair price of the claim within the interval
price, which is known as the interval of admissible prices. The arbitrage free price of the
barrier option is given by
v

t
= E
Q
_
h(S
t,x,
Tt
)1
{>(Tt)}

. (3.18)
Since we do not know yet whether our contingent claim is attainable or not. Therefore, we
expect that the arbitrage free price of the claim lies in the interval
v

(t, x) v

t
v
+
(t, x), 0 t T, (3.19)
where
v
+
(t, x) = sup
A()
E
_
h(S
t,x,
Tt
)1
{>(Tt)}

(3.20)
and
v

(t, x) = inf
A()
E
_
h(S
t,x,
Tt
)1
{>(Tt)}

. (3.21)
Now we adopt the denition of a replicating strategy for unspecied volatilities given in
Touzi [70], Karatzas [47], or Frey [37].
Denition 3.2. A super-replicating price for a contingent claim h at time t is given by
v(t, x) = inf {x 0 | A() admissible, such that
X
t,x,
T
h(S
t,x,
Tt
)1
{>(Tt)}
; P a.e.,
_
. (3.22)
This is the minimum price that the agent can accept in order to super-replicate the claim.
If the set is empty, then is zero. Any such process , which may depend on , is called a
super-replicating strategy or superstrategy.
Denition 3.3. A sub-replicating price for the contingent claim h at time t is given by
v(t, x) = sup {x 0 | A() admissible, such that
X
t,x,
T
h(S
t,x,
Tt
)1
{>(Tt)}
; P a.e.,
_
. (3.23)
Any such process , which may depend on , is called a sub-replicating strategy or sub-
strategy.
Remark 3.4. Another version of super-replicating and sub-replicating strategy is also given
by El-Kouri et al. [33] or Romagnoli and Vargiolu [63]. If v = v = v
t
, then v
t
is the arbitrage
free price of the contingent claim h.
28
Multi-asset Barrier Options 3.3 The replicating strategy
Teorema 3.2. The process
X
t,x,
t
X
t
= sup
A()
E
_
h(S
t,x,
Tt
)1
{>(Tt)}

(3.24)
is a supermartingale
Proof. The portfolio is self-nancing, and X
t
is bounded from below, hence by Theorem 3.5
in Krylov [49], p.149, X
t
is a supermartingale.
As we noticed in Denition 3.2 the super/substrategy depends on the choice of volatility
process (
t
). This choice can create arbitrage opportunities. Therefore, v
+
t
(respectively v

t
)
may be considered as a stochastic control problem where the lower bound and the upper
bound of its solution can be interpreted as a sub arbitrage price and super arbitrage
price, respectively. Before we convert our problem into a stochastic control problem in the
HJB equation, the following theorem gives an idea that with a superstrategy, one can have
X
t,x,
T
h(S
t,x,
Tt
)1
{>(Tt)}
. This means the portfolio overhedges the contingent claim.
Teorema 3.3. Let v be a price process for a contingent claim and let be a portfolio process.
If v is the super-hedging price as dened by Denition 3.2, then there exists a pair (v, )
such that
v(t, x) = v(t, x) = sup
A()
E
_
h(S
t,x,
Tt
)1
{>(Tt)}

.
In particular,
v(T, x) h(S
t,x,
Tt
) a.s. (3.25)
Proof. The following proof is original. This might be a signicant contribution of the thesis
in hedging barrier option. Take a superstrategy associated with an upper hedging price
as dened in Denition 3.2. Then
X
x,,v
T
E(S
t,x,
Tt
)1
{>(Tt)}
for every admissible control , but by Proposition 4.4.2 in Krylov [49], X
x,
T
is a super-
martingale. This implies that
v(t, x) = X
x,
t
E
_
X
t,x,
T
|F
t
_
E
_
h(S
t,x,
Tt
)1
{>(Tt)}

admissible . (3.26)
Hence, v(t, x) v(t, x).
29
3.3 The replicating strategy Multi-asset Barrier Options
To prove that v(t, x) v(t, x), we apply Itos formula to the process S
x,
T
, giving
v( T, S
x,
T
) = v(t, x) +

T
t
_

t
v(r, S
t,x,
r
) +Lv(t, r, S
x,
r
)
_
dr
+

T
t
x

x
v(r, S
t,x,
r
)dW
r
. (3.27)
Taking expectation of both sides, we have
Eh(S
t,x,
T
)1
{>T}
= v(t, x) +E

T
t
_

r
v(r, S
x,
r
) +Lv(r, S
x,
r
)
_
dr. (3.28)
Now we take the supremum of both sides, giving
v(t, x) v(t, x) + sup
A()
E

T
t
_

r
v(r, S
r
) +Lv(r, S
x,
r
)
_
dr. (3.29)
Since the expectation in (3.29) is zero, we have
v(t, x) v(t, x). (3.30)
Therefore,
v(t, x) = v(t, x) = sup
A()
E
_
(S
t,x,
Tt
)1
{>(Tt)}

.
Before we discuss the more specic example of the multi-asset barrier option, let us write
the HJB equation (??)-(??) in sense of multi-asset barrier option version
Teorema 3.4.
_

t
v(t, S
t
) +
1
2
sup

Tr(D
2
x
v(t, S
x,
t
)( x)( x)

) = 0, t [0, T), x R
d
v(T, x) = h(x) x R
d
v(t, x) = 0, x O
(3.31)
This equation is also known as the Black-Scholes-Barenblatt (BSB) equation. To give a
better idea about this problem, in the following section we discuss two-asset barrier option.
30
Multi-asset Barrier Options 3.4 Two-asset barrier option
3.4 Two-asset barrier option
In standard European options, there are several types of payo functions of two-asset barrier
options. For examples:
1. Max/min of two-asset barrier option:
h(T, S
1
T
, S
2
T
) = (max(S
1
T
, S
2
T
) K)
+
1
{>T}
,
where
= {t 0, max(S
1
t
, S
2
t
) H}.
2. Basket barrier option :
h(T, S
1
T
, S
2
T
) = (aS
1
T
+ (1 a)S
2
T
K)
+
1
{>T}
,
where a is a portion of asset S
1
and
= {t 0, max(S
1
t
, S
2
t
) H}.
3. External barrier option :
h(T, S
1
T
, S
2
T
) = (S
1
T
K)
+
1
{>T}
,
where
= {t 0, S
2
t
H}.
We choose the max/min of two-asset barrier option as an example. Let the prices of the
stocks at time t be S
1
t
and S
2
t
. The risk neutral price processes for the two assets S
1
t
and S
2
t
follow the stochastic dierential equations
dS
1
t
=
1
t
S
1
t
dW
1
t
(3.32)
dS
2
t
=
2
t
S
2
t
dW
2
t
(3.33)
S
1
0
= x
1
, S
2
0
= x
2
. (3.34)
Let denote the correlation coecient of the Brownian motion dW
1
t
and dW
2
t
. Assume that
the interest rate is zero. We can write the dynamics of the two assets in a more compact
vectorial notation:
d
_
_
_
_
S
1
t
S
2
t
_
_
_
_
=
_
_
_
_
S
1
t
0
0 S
2
t
_
_
_
_
_
_
_
_
(
1
t
)
2

1
t

2
t

1
t

2
t
(
2
t
)
2
_
_
_
_
d
_
_
_
_
W
1
t
W
2
t
_
_
_
_
(3.35)
31
3.4 Two-asset barrier option Multi-asset Barrier Options
or
dS
t
=

S
t

t
dW
t
. (3.36)
Here

S
t
= diag(S
t
) =
_
_
_
_
S
1
t
0
0 S
2
t
_
_
_
_
and (
t
) is a two-dimensional process such that
t
A(), where A() is a set of admissible
volatilities which is progressively measurable with respect to (F).
Consider a price process {v
t
; 0 t T} for a two-asset barrier option satisfying
v
T
= (max(S
1
T
, S
2
T
) K)
+
1
{>T}
, 0 t T.
Let S

t
= max(S
1
t
, S
2
t
), 0 t T. Then the rst moment of time when the process S

t
hits
the barrier H is given by
= inf{t 0; S

t
H}, (3.37)
where 0 < K < H. The value of the option at time t is given by
v(0, x
1
, x
2
) = E
_
min((S
1
T
, S
2
T
) K)
+
1
{>T}
|S
1
t
= x
1
, S
2
t
= x
2

. (3.38)
For constant volatilities
t
= , the option price v can be computed by solving the partial
dierential equation
v
t
+
1
2
(
1
)
2
x
2
1

2
v
x
2
1
+
1
2
(
2
)
2
x
2
2

2
v
x
2
2
+
1

2
x
1
x
2

2
v
x
1
x
2
= 0 (3.39)
with terminal and boundary conditions
v(T, x
1
, x
2
) = (min(x
1
, x
2
) K)
+
, 0 max(x
1
, x
2
) < H (3.40)
v(t, x
1
, x
2
) = 0, max(x
1
, x
2
) H, 0 t T. (3.41)
Equation (3.39) can be written as
v
t
+
1
2
tr(D
2
x
v( x)( x)

= 0 (3.42)
where the vector x = (x
1
, x
2
) and the matrix
D
2
x
v =
_
_
_
_
_
_
_
_
_

2
v
x
2
1

2
v
x
1
x
2

2
v
x
1
x
2

2
v
x
2
2
_
_
_
_
_
_
_
_
_
.
32
Multi-asset Barrier Options 3.5 Optimization problem
Hence we have
A
t,x
= x D
2
x
v x =
_
_
_
_
_
_
_
_
_
x
2
1

2
v
x
2
1
x
1
x
2

2
v
x
1
x
2
x
1
x
2

2
v
x
1
x
2
x
2
2

2
v
x
2
2
_
_
_
_
_
_
_
_
_
.
Now we assume that the true volatilities are not known and limited to move in a certain
interval. We write the set of admissible volatilities as follows:
=
_

_
R
22

=
_
_
_
_

1
t

1
t

2
t

2
t
_
_
_
_
,

1

1
t

+
1
,

2

2
t

+
2
,


+
_

_
.
Then we have

=
_
_
_
_
_
_
_
_
(
1
t
)
2

1
t

2
t

1
t

2
t
(
2
t
)
2
_
_
_
_
_
_
_
_
.
Let A
t,x
= A be
A =
_
_
_
_
a b
b c
_
_
_
_
.
Then the corresponding Hamilton-Jacobi-Bellman equation can be written as:
v
t
+
1
2
max
A()
tr(A

) = 0, (3.43)
with terminal and boundary condition
v(T, x
1
, x
2
) = (min(x
1
, x
2
) K)
+
, 0 max(x
1
, x
2
) < H (3.44)
v(t, S
1
T
, S
2
T
) = 0, , max(x
1
, x
2
) H, 0 t T. (3.45)
3.5 Optimization problem
In this subsection we discuss the optimization problem appearing in the HJB equation (3.43).
First, we write the function to be maximized as follows:
f(
1
,
2
, ) = tr(A

) = a
2
1
+ 2b
1

2
+ c
2
2
, (3.46)
33
3.5 Optimization problem Multi-asset Barrier Options
where

1

1
t

+
1
,

2

2
t

+
2
,


+
. The standard form is given by
QP1
_

_
minimize f(
1
,
2
, )
subject to

1

1
t

+
1
,

2

2
t

+
2
,

+
.
(3.47)
We present the variables as a vector x and its lower bound and upper bound as l and u,
respectively:
l =
_
_
_
_
_
_
_
_

_
_
_
_
_
_
_
_
and u =
_
_
_
_
_
_
_
_

+
1

+
2

+
_
_
_
_
_
_
_
_
. (3.48)
The problem QP1 is a special form of bound constraint of the optimization problem. This
form leads to considerable simplication of the optimality conditions. The Lagrange mul-
tiplier for an active bound on x
i
, = 1, , 3 is given by g
i
(x

) if x

i
= l
i
, and by g
i
(x

) if
x

i
= u
i
( since the constraint x
i
u
i
can be written as x
i
u
i
). Then the sucient
condition for x

to be a local minimizer of f subject to bound (3.48) is given by the following


theorem.
Teorema 3.5. First order sucient conditions for optimality(Gill, Murray, and
Wright [38])
If x is feasible for QP1 and there exists x

such that x

satises the following conditions


1. l x

u
2. x

= l if f

(x

) > 0
3. g

L
> 0 and g

U
< 0
then x

is a local minimizer of f.
34
Chapter 4
Numerical solutions
The following algorithm for solving the HJB equation is original. Our result here is a
signicant contribution of the thesis.
4.1 Introduction
In Chapter 4, we have shown that the stochastic exit control problem for pricing barrier
options with unspecied volatilities can be converted into the nonlinear PDE (HJB equation).
We have also shown that the nonlinear PDE of the stochastic exit control problems converges
to the viscosity solutions (see Crandall et al. [22]). Barles [5] gives a detailed discussion about
stability, consistence, and convergence to the viscosity solution.
One may refer to [75] for the implicit method or the Crank-Nicolson method, used to
solve linear PDEs for barrier options. Boyle [16], also developed an explicit nite dierence
method for pricing barrier options based on a lattice or trinomial approach. The linear PDE
for pricing barrier options has been discussed in Zvan et al. [75], [76]. For the nonlinear PDE,
one may refer to Pooley, Forsyth and Vetzal [61]. The authors use the implicit discretization
method to nd the best and the worst price of the uncertain volatility. They also demonstrate
the numerical convergence properties of the nonlinear PDE in the case of a single asset.
In this chapter, we focus on computational and implementation issues of the nite dif-
ference methods for solving the HJB equation. We not only discuss the single asset case but
also the two-asset-case. We also present here the algorithm of how to solve the HJB equation
in two-asset case. In the algorithm, we demonstrate how we can make use MATLAB func-
tion, FMINCON, to nd the optimal price. In this discussion we are not interested in the
35
4.2 Numerical Scheme and algorithm Numerical solutions
eciency or the accuracy of the approximation. Such a problem is beyond our discussion.
One may refer to Pooley, Forsyth and Vetzal [61] or [75] for the eciency or the accuracy
for the implicit method or the Crank-Nicolson method
4.2 Numerical Scheme and algorithm
4.2.1 Single-asset case
In the case of a single-asset, we use explicit schemes to approximate the HJB equation.
The reason for choosing these schemes is that we want to demonstrate the condition under
which the approximation converges to the closed form solution. Other schemes, such as the
implicit and Crank-Nicolson schemes, are unconditionally stable. Now we rewrite again the
HJB equation:

r
v(r, S
r
) + sup

min

max
1
2

2
S
2
r

2
S
2
v(r, S
r
) = 0, r < T, S
min
S
r
S
max
(4.1)
with terminal condition
v(T, S
T
) = (S
T
K)
+
S
r
< H, 0 r T (4.2)
and boundary condition
v(r, S
r
) = 0 S
r
H, 0 r T. (4.3)
To discuss the discretization of (4.1), we rstly introduce the new variables x and t dened
by
x = log(S/K), t = T /
1
2

2
, S = Ke
x+
u(, x).
Then (4.1) can be written as
u

=

2
u
x
2
, x
min
< x < x
max
, 0 < <
1
2

2
T (4.4)
with the initial condition
u(0, x) = max(e
1
2
(k+1)x
e
1
2
(k1)x
, 0), x
min
x x
max
(4.5)
36
Numerical solutions 4.2 Numerical Scheme and algorithm
T
v=0
v = (S
T
K)
+
v=0
H K
0
S
s
Figure 4.1: The initial and boundary conditions of the schemes
and the boundary condition
u(, x
max
) =
1
K
e
1
2
(k1)x
max
+
1
4
(k+1)
2

(S
max
Ke
r(T)
) (4.6)
u(, x
max
) = 0, S
max
> H. (4.7)
Here
u(, x) =
1
K
e
1
2
(k1)x+
1
4
(k+1)
2

v(, S

), 0 < <
1
2

2
T
and k = r/0.5
2
, x
min
= log(S
min
/K), x
max
= log(S
max
/K). The choice of volatility
depends on the sign of the

2
v
x
2
. To discuss the discretization, we divide the x axis into
equally spaced nodes with interval length x and the t time axis into equally spaced nodes
with interval t. Now we choose positive integers N and M as the number of intervals in
the x axis and = t axis, respectively. We dene grid points along the space and time axes,
x
i
= x
min
+ ix for i = 1, N,
t
j
= jt for j = 1, , M,
37
4.2 Numerical Scheme and algorithm Numerical solutions
where x = (x
max
x
min
)/N, t =
1
2

2
T/M. We denote the midpoint of the j-th time
interval by
t
j
1
2
=
1
2
(t
j1
+ t
j
) = (j
1
2
)t
and we write
U
1
2
i
=
1
2
(U
j
i
+ U
j1
i
) = u(x
i
, t
j
1
2
).
Now we approximate
u
t
by
u
t

u(t
j
, x
i
) u(t
j
t, x
i
)
t
=
U
j
i
U
j1
i
t
(4.8)
and approximate

2
u
x
2
by

2
u
x
2

u(t
j
1
2
, x
i
+ x) 2u(t
j
1
2
, x
i
) + u(t
1
1
2
, x
i
x)
t

U
j
1
2
i+1
2U
j
1
2
i
+ U
j
1
2
i1
(x)
2
.
We can write (4.4) in general form as follows
v
j
i
v
j1
i
t
=
v
j
i+1
2v
j
i
+ v
j
i1
(x)
2
, (4.9)
where [0, 1] is a temporal weighting factor determining the type of scheme being used:
fully implicit when = 1, Crank-Nicolson when =
1
2
and fully explicit when = 0. Now
consider the case when = 0, that is explicit scheme.
Because the scheme is explicit, it is stable under some condition. The stability depends
on the step lengths t and x. From (4.9), we have
v
j
i
=
_
1
t
(x)
2
_
v
j
i
+
1
2
_
t
(x)
2
_
_
v
j+1
i+1
+ v
j+1
i1
_
. (4.10)
In order to prevent any oscillation, the following condition must be satised
(1
t
(x)
2
) 0 (4.11)
The PDE is solved using an explicit scheme which is conditionally convergent. Based on
this scheme, we propose the following algorithm for a numerical solution of (4.1) and (4.2):
(
j+1
)
2
=
_

2
max
if
j+1
0

2
min
if
j+1
0
(4.12)
38
Numerical solutions 4.2 Numerical Scheme and algorithm
where
=
_
v
j+1
i+1
2v
j+1
i
+ v
j+1
i1
(x)
2
_
for i = 1, , M and j = 1, , N.
Algorithm :
Initialize {v
j
0
}
N1
j=0
using the initial and boundary conditions (4.5).
For i = 0 to M 1 do
v
0
i
= 0
For j = 1 to N 2 do
grad
j+1
i
= v
j+2
i1
2v
j+1
i1
+ v
j
i1
If grad
j+1
i
0
=
min
= 0.5
2
t/x
2
Else
=
max
= 0.5
2
t/x
2
endif
v
j+1
i
= (1 2)v
j+1
i1
+ v
j+2
i1
+ v
j
i1
For k = h to N 1 do
v
k
i
= 0
end do j
end do i
Here h is the point at which the scheme hits the barrier.
In the algorithm above, the barrier is assumed to be observed continuously. In other
words, the barrier is checked at every time during the life of the option. In practice, the
barrier is often observed at discrete-time intervals such as daily or weekly. The algorithm
above can easily be adjusted to compute prices and to hedge the discrete-time barrier option.
Moreover, it also can be used to handle double discrete barrier options.
39
4.2 Numerical Scheme and algorithm Numerical solutions
4.2.2 Two-asset case
Consider the HJB equation:
v
s
+ max
A()
_
1
2

2
1
S
2
1

2
v
S
2
1
+
1
2

2
2
S
2
2

2
v
S
2
2
+
1

2
S
1
S
2

2
v
S
1
S
2
_
= 0,
S
min
S
1
, S
2
S
max
, (4.13)
with initial conditions
v(s, S
min
, S
2
) = 0 (4.14)
v(s, S
1
, S
min
) = 0; (4.15)
terminal condition
v(T, S
1
, S
2
) = (max(S
T
1
, S
T
2
) K)
+
, (4.16)
and the boundary condition
v(s, S
t
1
, S
t
2
) = 0., max(S
t
1
, S
t
2
) H. (4.17)
Here () is a two-dimensional matrix such that A(). To simplify computation we
convert the HJB equation (4.13) into a dimensionless form. We consider the following change
of variables
S
1
= Ke
x
,
S
2
= Ke
y
,
s = T
t
1
2

2
,
v = Ku(x, y, s).
Then the rst derivative of v with respect to time t and price S is given by
v
s
=
1
2

2
K
u
t
(4.18)

2
v
S
2
1
=

S
1
_
K
S
1
u
S
1
_
=
K
S
2
1
u
x
+
K
S
2
1

2
u
x
2
(4.19)

2
v
S
2
2
=

S
2
_
K
S
2
u
S
2
_
=
K
S
2
2
u
y
+
K
S
2
2

2
u
y
2
(4.20)

2
v
S
1
S
2
=

S
1
_
K
S
2
u
S
2
_
=
K
S
1
S
2

2
u
xy
(4.21)
By substituting these into (4.13), then the problem can be formulated as follows;
v(s, S
1
, S
2
) = Ku(t, x, y) (4.22)
40
Numerical solutions 4.2 Numerical Scheme and algorithm
where u is the solution of following dynamic programming equation
u
t
= max
A()
_

2
u
x
2
+

2

2
u
y
2
+ 2

2
u
xy
_
(4.23)
with initial conditions
u(t, x
min
, y) = 0 (4.24)
u(t, x, y
min
) = 0, (4.25)
terminal condition
u(T, x, y) = (max(e
x
, e
y
) 1)
+
, (4.26)
and the boundary condition
u(t, x, y) = 0, max(S
t
1
, S
t
2
) H. (4.27)
Here x
max
= y
max
= log(S
max
/K) andx
min
= y
min
= log(S
min
/K).
In nite dierence mesh, we divide the x and y axes into J equally spaced nodes with
interval x and y, and the t axis into N equally spaced nodes with interval t. We dene
grid points along the space and time axis as follows;
t =
1
2

2
T/J,
t
j
= jt for 0 j J,
x
m
= mx for N

m N
+
,
y
n
= nx for N

n N
+
,
where x = (x
max
x
min
)/N = (y
max
y
min
)/N, and N

and N
+
are positive integer with
N
+
N

= N. Now we introduce grid function


U
j
m,n
u(x
m
, y
n
, t
j
)
u
0
m,n
u
0
(x
m
, y
n
)
t
j
= (t
j1
+ t
j
) = (j )t
Now we approximate
u
t
by
u
t

u(x
m
, y
n
, t
j
) u(x
m
, y
m
, t
j
t)
t

U
j
mn
U
j1
mn
t
, (4.28)
41
4.2 Numerical Scheme and algorithm Numerical solutions
and approximate

2
u
x
2
,

2
u
y
2
and

2
u
xy
by second central dierence, we have

2
u
x
2

u(x
m
+ x, y
n
, t
j
) 2u(x
m
, y
n
, t
j
) + u(x
m
x, y
n
, t
1
)
(x)
2

U
i
m+1,n
2U
j
m,n
+ U
j
m1,n
(x)
2
.

2
u
y
2

u(x
m
, y
n
+ y, t
j
) 2u(x
m
, y
n
, t
j
) + u(x
m
, y
n
y, t
1
)
(y)
2

U
i
m,n+1
2U
j
m,n
+ U
j
m,n1
(y)
2
.

2
u
xy

1
2x
_
u(x
m
+ x, y
n
+ y, t
j
) u(x
m
x, y
n
+ y, t
j
)
2x

u(x
m
+ x, y
n
y, t
j
) u(x
m
x, y
n
y, t
j
)
2x
_

U
i
m+1,n+1
U
j
m1,n+1
U
j
m+1,n1
+ U
j
m1,n1
(2x)
2
.
Substituting these into (4.23), we obtain
U
j
mn
U
j1
mn
t
= max
A()
_

2
U
i
m+1,n
2U
j
m,n
+ U
j
m1,n
(x)
2

1
U
i
m+1,n
2U
j
m,n
+ U
j
m1,n
(x)
2
2
U
i
m+1,n+1
U
j
m1,n+1
U
j
m+1,n1
+ U
j
m1,n1
(2x)
2
_
for N

+ 1 m, n N
+
1 and 1 j J,
with U
0
m,n
= u
0
m,n
for N

m, n N
+
.
We will use the implicit method solve the problem, so we choose the temporal weighting
= 1. The algorithm is given as follows
42
Numerical solutions 4.2 Numerical Scheme and algorithm
Algorithm :
input N, J, T, r
Calculate k = 0.5
1

2
T/J, h = (x
max
x
min
)/N, a = 2r/(
1

2
), b =
1
/
2
Initialize v
0
(1 : N + 1, 1 : N + 1) using the initial value v
0
= (max(e
x
, e
y
) 1)
+
Search for the barrier
If found H = k then
v
0
(k : N + 1, :) = 0
v
0
(:, k : N + 1) = 0
EndIf
x0 = [x
1
0
, x
2
0
, x
3
0
] (starting point)
lb = [

1
,

2
,

] (lower bound)
ub = [
+
1
,
+
2
,
+
] (upper bound)
For i = 2 to N do
For j = 2 to N do
a
1
= v
i+1,j
2v
i,j
+ v
i1,j
a
2
= v
i+1,j+1
v
i,j+1
u
i+1,j1
+ v
i1,j1
a
2
= v
i,j+1
2v
i,j
+ v
i,j1
f(sg
1
, sg
2
, rh) = a
1
sg
2
1
+ a
3
sg
2
2
+ 2a
2
sg
1
sg
2
rh
CALL MATLAB function fmincon(f,x0,[],[],[],[],lb,ub)

1
= sg
1
,
2
= sg
2
, = rh
aa = 2 r/(
1

2
), bb =
1
/
2
cc = k b/(2 h
2
), dd = k/(2 b h
2
), ee = k/(4 h
2
), ss = 1 dd cc
u
i,j
= ss v
i,j
+ cc(v
i+1,j
+ v
i1,j
) + dd(v
i,j+1
+ v
i,j1
) + ee(v
i+1,j+1

v
i,j+1
v
i+1,j1
+ v
i1,j1
)
end do j
end do i
Solve the system equation of size (N 1)
2
(N 1)
2
43
4.2 Numerical Scheme and algorithm Numerical solutions
44
Bibliography
[1] Andersen L., J. Andreasen, and David Elizer. 2000. Static replication of barrier option.
Working Paper.
[2] Avellaneda, M., Levy, A., and Paras, A. 1995. Pricing and hedging derivatives in markets
with stochastic volatilities. Applied Mathematical Finance. 2, p.73-88.
[3] Avellaneda, M. and A. Paras. 1996. Managing the volatility risk of portfolios of deriva-
tives securities: the Lagrangian uncertain volatility model. Applied Mathematical Fi-
nance 3, p.21-52.
[4] Bardi, M. and I. Capuzo Dolcetta. 1997. Optimal control and viscosity solutions to
Hamilton-Jacobi-Bellman equations. Birkhauser Boston.
[5] Barles, G. 1997. Convergence of numerical schemes for degenerate parabolic equations
arising in nance. In L.C.Rogers and D. Talay (Eds.), Numerical Methods in Finance,
p.1-21. Cambridge University Press, Cambridge.
[6] Barles G. and J. Burdeau. 1995. The Dirichlet problem for semilinear second-order de-
generate elliptic equation and application to stochastic exit time control problem. Com-
munication in Partial Dierential Equations, 20(1-2), p.129-178.
[7] Barles G. and E. Rouy. 1998. A strong comparison result for the Bellman equation arsing
in stochastic exit time control problem and its application. Communication in Partial
Dierential Equations, 23(11-122), p.1995-2033.
[8] Barles G. and Perthame. 1988. Exit time problems in optimal control and vanishing
viscosity method. SIAM journal in Control and Optimisation. 26. p.1133-1148.
45
BIBLIOGRAPHY BIBLIOGRAPHY
[9] Barles G. and Perthame. 1990. Comparison principle for Dirichlet type Hamilto-Jacobi
equations and singular pertubations of degenerated elliptic equations. Applied Math. and
Optimization, 21.
[10] Barles G. and P.E. Souganidis. 1991. Convergence of approximation schemes for fully
nonlinear second order equations. Asymptotic Analysis. 4. p.271-283.
[11] Bensoussan, A. 1982. Stochastic control by functional analysis methods. North-Holland,
Amsterdam.
[12] Bergman Y.Z., B.D. Grundy, and Z. Wiener. 1996. General properties of option prices.
The Journal of Finance, 51(5). p.1573-1610.
[13] Bingham, N.H. and R. Kiesel. 1998. Risk-Neutral Valuation: Pricing and Hedging Fi-
nancial Derivatives. Spinger, New York.
[14] Black, F. and M. Schole. 1973. Pricing of Options and Corporate Liabilities. Journal of
Political Economy, 81,p637-659.
[15] Borkar, V.S. 1989. Optimal Control of Diusion Processes, Pitman Research Notes, No.
203, Longman Sci. and Tech. Harlow, UK.
[16] Boyle, P.B. and Yisong Tian. 1998. An explicite nite dierence approach to the pricing
of barrier options. Applied Mathematics Finance, 5, p.17-43.
[17] Brown, H., D. Hobson, and L. Rogers. 2001. Robust hedging of barrier options. Mathe-
matical Finance. 11(3), p.285-314.
[18] Caarelli, L.A. 1989. Interior a priori estimates for solutions of fully non-linear equa-
tions. Annals of Mathematics 130. p.189-213.
[19] Carr, P. K. Ellis, and V. Gupta. 1998. Static hedging of exotic options. Journal of
Finance, 53, p.1165-1191.
[20] Cvitanic, J. and I. Karatzas. 1993. Hedging contingent claim with constrained portfolio.
Annal Applied Probability. 3. p.652-681.
[21] Crandall M. and P.L. Lions. 1983. Viscosity solutions of Hamilton -Jacobi equations.
Transaction American Mathematic Society. 277(1). p.1-42.
46
BIBLIOGRAPHY BIBLIOGRAPHY
[22] Crandall, M.G., H. Ishii, and P.L. Lions. 1992. Users guide to viscosity solution of
second order partial dierential equation. Bull. Amer. Math. Soc., 27, p.1-67.
[23] Crandall, M. G. 1997. Viscosity Solutions : A Primer. Viscosity solution and Applica-
tions. Eds. I.C. Dolcetta and P.L. Lions. p.134-185.
[24] Crandall, M. G., M. Kocan, and A. Swiech. 2000.L
P
-Theory for fully nolinear uni-
formly parabolic equations. Communication in Partial Dierential Equations, 25(11&12),
p.1997-2053.
[25] Crandall, M. G., M. Kocan, and A. Swiech. 1999.Existence results for boundary prob-
lems for uniformly elliptic and parabolic fully nonlinear equations. Electronic Journal of
Dierential Equations. 24, p.1-20.
[26] Cvitanic, J and Karatzas. 1993. Hedging contingent claim with constrained portfolion.
Ann. Applied Probability 3, p.523-545.
[27] Davis M.H.A. and J.M.C. Clark. 1994. A note on super-replicating strategies. Phil.
Trans. R. Soc. London A 347, p.485-494.
[28] Due D. 1996. Dynamic Asset Pricing Theory. Standford University, Academic Press,
California
[29] Durrett, R. 1996. Stochastic Calculus: A practical Introduction. CCR Press, Florida.
[30] Derman, E., D Ergener, and I. Kani. 1995. Static options replication. Journal of Deriva-
tives, 2(4). p.78-95
[31] Elliott, R.J. and P E. Kopp. 1999. Mathematics of Financial Markets Springer Verlag.
[32] El-Karoui N. and M.C. Quenez. 1995. Dynamic programming and the pricing of contin-
gent claim in an incomplete market. SIAM, Journal of Control and Optimisation. 33.
p.29-66.
[33] El-Karoui N., M. Jeanblanc-Pique, S.E. Shreve. 1998. Robustness of the Black-Schole
Formula. Math. Finance 8(2), p.93-126.
[34] Fleming, W.H. and H.M. Soner. 1993. Controlled Markov Processes and Viscosity So-
lution. Springer-Verlag, New-York.
47
BIBLIOGRAPHY BIBLIOGRAPHY
[35] Fleming, W.H and R.W. Rishel. 1975. Deterministic and Stochastic Optimal Control.
Springer-Verlag, New-York.
[36] Frey, R. and C.A. Sin. 1999. Bounds on European option prices under stochastic volatil-
ity, Mathematical Finance. 9(2), p.97-116.
[37] Frey R. 2000. Superreplication in stochastic volatility models and optimal stopping. Fi-
nance and Stochastics. 4. p.161-187.
[38] Gill P.E., W. Murray, and M.H. Wright.1981. Practical Optimization. Academic Press,
London.
[39] Gozzi, F and T. Vargiolu. 2002. Supperreplication of European Multiasset Derivatves
with Bounded Stochastic Volatility. Math. Methods of Oper. Research.
[40] Harrison, J.M. and S.R Pliska. 1981. Matingales and stochastic integrals in the theory
of continuous trading. Stochastic processes and Applications. 11. p.215-260.
[41] Harrison, J.M. and S.R Pliska. 1983. A stochastic calculus model of continuous trading:
Complete markets. Stochastic processes and Applications. 15. p.313-316.
[42] Heynen R. and Harry Kat. 1994. Crossing Barrier. Risk, 7(6). p.46-49
[43] Hobson, David G. 1998. Volatility misspecication, option pricing and superreplication
via coupling. The Annals of Applied Probability 8(1), p.193-205
[44] Hull, J.C. and White, A. 1987.The pricing of Option on Assets with Stochastic Volatil-
ities. Journal of Finance 42, p.281-300.
[45] Hull J.C. 1997. Options, Futures, and Other Derivatives 3rd Edition, Prentice-Hall, New
Jersey.
[46] Jensen, R. 1988. The maximum principle fro viscosity solutions of fully nonlinear second
order. Arch. Rational Mech. Anal. 101. p.1-27. PDE
[47] Karatzas, I. 1996. Lectures on mathematics of nance. CRM monograph series, ISSN
1065-8599; v.8
[48] Karatzas, I. and S.E. Shreve. 1991. Brownian Motion and Stochastic Calculus. Springer-
Verlag, New York.
48
BIBLIOGRAPHY BIBLIOGRAPHY
[49] Krylov, N.V. 1990. Control Diusion Processes. Springer-Verlag, New York.
[50] Krylov, N.V. 1995. Introduction to the Theory of Diusion Processes. American Math-
ematical Society.
[51] Kwok, Y.K., L. Wu, and H. Yu. 1998. Pricing multi-asset options with an external
barrier. International Journal of Theoretical and Applied Finance. 1(4), p.523-541
[52] Lions, P.L. 1983. Optimal Control of Diusion Process and Hamilton-Jacobi-Bellman
Equation: Part I: The Dynamic Programming Principle and Applications. Comm. in
Partial Dierential Equations, 8(10), p.1101-1174.
[53] Lions, P.L. 1983. Optimal Control of Diusion Process and Hamilton-Jacobi-Bellman
Equation: Part II: Viscosity Solutions and Uniqueness . Comm. in Partial Dierential
Equations, 8(11), p.1229-1276.
[54] Lions, P.L. 1983. Optimal Control of Diusion Process and Hamilton-Jacobi-Bellman
Equation: Part III: Regularity of the optimal cost function . Pitman, College de France
Seminar.
[55] Lyons, T. 1993. Uncertain volatility and the risk-free synthesis of derivatives. Applied
Mathematical Finance 6, p.1969-1984.
[56] Ma, Jin and Jiongmin Yong. 1999. Dynamic programming for multidimensional stochas-
tic control problems. Acta Mathematica Sinica, English Series 15 (4), p.485-506.
[57] Margrabe, W. 1978. The price to exchange an asset for another. The Journal of Finance
1, p.177-186.
[58] Merton R.C. 1973. The theory of rational option pricing. Bell Journal of Economics and
Management Scince. 4. p.141-183.
[59] Musiela, M. and M. Rutkowski. 1997. Martingale Methods in Financial Modelling, vol.
36 of Applications of mathematics: Stochastic Modelling and Applied Probability. Sp-
inger, New York.
[60] Oksendal, B. 1995. Stochastic Dierential Equations. Fourth Edition, Springer Verlag,
New York.
49
BIBLIOGRAPHY BIBLIOGRAPHY
[61] Pooley, D.M., P.A. Forsyth, and K.R.Vetzal. 2001. Numerical convergence properties of
option pricing PDEs with uncertain volatility. Working Paper. University of Waterloo,
Waterloo, Ontario, Canada.
[62] Rich D. R. 1994. The mathematical foundations of barrier option-pricing theory. Ad-
vances in Futures and Operation Research, 7,p.267-311.
[63] Romagnoli S. and T. Vargiolu. 2000. Robustness of the Black-Scholes approach in the
case of option on several assets. Finance and Stochastics, 4, p.325-341.
[64] Rubinstein M. and E.S. Reiner. 1991. Breaking down the barrier. Risk 4(8), p.28-35.
[65] Schmock U., S. E. Shreve and U. Wystup. 2001. Valuation of exotic under shortselling
constraints. Working Paper. http://www.math.ethz.ch/ schomock
[66] Soner, Halil M. 1986. Optimal control problems with state-space constraints. SIAM Jour-
nal on Control and Optimisation. 24. p.552-562.
[67] Soner, Halil M. 1997. Controlled Markov Processes, Viscosity Solutions and Applications
to mathematical Finance. in Viscosity solution and Applications. Eds. I.C. Dolcetta and
P.L. Lions. p.134-185.
[68] Shreve, E.S.Lecture on Stochastic Calculus and Finance .
http://www-2.cs.cmu.edu/ chal/shreve.html
[69] Steele, J.M. 2001. Stochastic Calculus and Financial Applications. Springer-Verlag, New
York.
[70] Touzi Nizar. 1999. Super-replication under proportional transaction costs: From discrete
to continuous-time models. Mathematical Methods of Operations Reserach. 50, p.297-
320.
[71] Wang, L. 1992. On the regularity of fully nonlinear parabolic equation : I, Communica-
tions in Pure and Applied Mathematics. 45.p.27-76
[72] Wang, L. 1992. On the regularity of fully nonlinear parabolic equation : II, Communi-
cations in Pure and Applied Mathematics. 45.p.141-178
50
BIBLIOGRAPHY BIBLIOGRAPHY
[73] Wong H.Y. and Y.K. Kwok. 2001. Multi-asset barrier option and occupation time de-
rivetives. Working paper. Dept. of Mathematics, Hongkong University of Science and
Technology.
[74] Yong J. and X.Y, Zhou. 1999. Stochastic Control : Hamiltonian Systems and HJB
Equations Springer-Verlag, New York.
[75] Zvan R., K.R. Vetzal, and P.A. Forsyth. 2000. PDE method for pricing barrier option.
Journal of Economic Dynamics and Control. 24. p.1563-1590.
[76] Zvan R., K.R. Vetzal, and P.A. Forsyth. 1998. A general nite element for PDE option
pricing models. Working paper. University of Waterloo, Canada.
51

You might also like