Pricing European options with dividends under a stochastic...
Transcript of Pricing European options with dividends under a stochastic...
저 시 2.0 한민
는 아래 조건 르는 경 에 한하여 게
l 저 물 복제, 포, 전송, 전시, 공연 송할 수 습니다.
l 차적 저 물 성할 수 습니다.
l 저 물 리 목적 할 수 습니다.
다 과 같 조건 라야 합니다:
l 하는, 저 물 나 포 경 , 저 물에 적 된 허락조건 명확하게 나타내어야 합니다.
l 저 터 허가를 면 러한 조건들 적 되지 않습니다.
저 에 른 리는 내 에 하여 향 지 않습니다.
것 허락규약(Legal Code) 해하 쉽게 약한 것 니다.
Disclaimer
저 시. 하는 원저 를 시하여야 합니다.
저 시 2.0 한민
는 아래 조건 르는 경 에 한하여 게
l 저 물 복제, 포, 전송, 전시, 공연 송할 수 습니다.
l 차적 저 물 성할 수 습니다.
l 저 물 리 목적 할 수 습니다.
다 과 같 조건 라야 합니다:
l 하는, 저 물 나 포 경 , 저 물에 적 된 허락조건 명확하게 나타내어야 합니다.
l 저 터 허가를 면 러한 조건들 적 되지 않습니다.
저 에 른 리는 내 에 하여 향 지 않습니다.
것 허락규약(Legal Code) 해하 쉽게 약한 것 니다.
Disclaimer
저 시. 하는 원저 를 시하여야 합니다.
이학석사 학위논문
Pricing European options withdividends under a stochastic
volatility and stochasticinterest rate
(배당을 고려한 확률론적 변동성과 이자율 하에서의유럽형 옵션의 가격 결정)
2015년 2월
서울대학교 대학원
수리과학부
김 동 준
Pricing European options withdividends under a stochastic
volatility and stochasticinterest rate
(배당을 고려한 확률론적 변동성과 이자율 하에서의유럽형 옵션의 가격 결정)
지도교수 강 명 주
이 논문을 이학석사 학위논문으로 제출함
2014년 12월
서울대학교 대학원
수리과학부
김 동 준
김 동 준의 이학석사 학위논문을 인준함
2015년 1월
위 원 장 (인)
부 위 원 장 (인)
위 원 (인)
Pricing European options withdividends under a stochastic
volatility and stochasticinterest rate
A dissertation
submitted in partial fulfillment
of the requirements for the degree of
Master of Science
to the faculty of the Graduate School ofSeoul National University
by
Dongjun Kim
Dissertation Director : Professor Myungjoo Kang
Department of Mathematical SciencesSeoul National University
Feburary 2015
c© 2014 Dongjun Kim
All rights reserved.
Abstract
In this paper, we can derive the first-order price approximation of European
options with dividends under a stochastic volatility and stochastic interest
rate by using Feynman-Kac Theorem and Poisson Equations. Under a risk-
neutral measure, we choose the Ornstein-Uhlenbeck process for a stochastic
interest rate. Lastly, we can obtain the result that corrected price will not
depend on the present volatility and compare a difference with Monte-Carlo
simulations.
Key words: Stochastic volatility, Stochastic interest rate, Feynman-Kac for-
mula, European options with dividend, Ornstein-Uhlenbeck Process, Poisson
equations
Student Number: 2012-20243
i
Contents
Abstract i
1 Introduction 1
1.1 Risk-Neutral Pricing . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Multidimensional Stochastic Calculus . . . . . . . . . . . . . . 3
2 Model 7
2.1 Asset price model . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . 11
3 Asymptotics for Pricing 13
3.1 The Formal Expansion . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Poisson Equations . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3 Computation of Aε . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Accuracy of the Approxmation . . . . . . . . . . . . . . . . . 20
4 Numerical Results 22
5 Conclusion 29
Abstract (in Korean) 32
Acknowledgement (in Korean) 33
ii
List of Figures
1.1 Stock paths with S0 = 100, µ = 0.1, σ = 0.2 . . . . . . . . . . . 2
2.1 Interest rate for long run mean is 0.02. . . . . . . . . . . . . . 11
4.1 Monte-Carlo Simulation . . . . . . . . . . . . . . . . . . . . . 23
4.2 Explicit method . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Implicit method . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4 Crank-Nicolson method . . . . . . . . . . . . . . . . . . . . . . 26
4.5 Explicit, Implicit method with Various maturity time t . . . . 27
4.6 Crank-Nicolson method with Various maturity time t . . . . . 27
iii
Chapter 1
Introduction
The object of this first section is to review the basic ideas, objects, results of
the classical pricing theory.
1.1 Risk-Neutral Pricing
The main result we want to make in this section is that there is a unique
probability measure P∗ equivalent to P such that, under this probability,
the discounted price e−∫ t0 rsdsSt is a martingale. By fundamental theorem
of asset pricing, the expected value under P∗ of the discounted payoff of a
derivative gives its no-arbitrage price. Such a probability measure describing
a risk− neutral world is called an equivalent martingale measure.[1]
The Black-Scholes basic model for the risky stock price model corresponds
to a continuous process St such that with a constant rate of return µ, and
centered random fluctuations independent of the past up to time t. These
fluctuations are modeled by σdWt, where σ is a positive constant volatility.
The corresponding process for the infinitesimal return is
dStSt
= µdt+ σdWt (1.1.1)
The right-hand side of (1.1.1) has the financial interpretation of a return
term(µ) plus a risk term(σ).
1
CHAPTER 1. INTRODUCTION
In integral form, (1.1.1) is
St = S0 + µ
∫ t
0
Stdt+ σ
∫ t
0
StdWt, (1.1.2)
where the last integral is a stochastic integral and S0 is the initial value.
Figure 1.1: Stock paths with S0 = 100, µ = 0.1, σ = 0.2
A function of the Brownian motion Wt defines a new stochastic process
h(Wt). By the simplest version of Ito’s formula,
dh(t, St) =
(∂h
∂t+ µ
∂h
∂s+
1
2σ2∂
2h
∂s2
)ds+ σ
∂h
∂sdWt. (1.1.3)
Let r is constant interest rate, we can compute the differential of the
discounted stock price g(t, St) = e−rtSt :
d(e−rtSt) = −re−rtStdt+ e−rtdSt
= e−rt(−rSt + µ)dt+ e−rtσdSt
= (µ− r)(e−rtSt)dt+ σ(e−rtSt)dWt.
(1.1.4)
2
CHAPTER 1. INTRODUCTION
The discounted stock price e−rtSt satisfies the same equation as St of
(1.2.1) where the expected return µ has been replaced by µ− r.In order to find a probability measure under which the discounted stock
price St = e−rtSt is a martingale,
dSt = σSt
(dWt +
(µ− rσ
)dt
). (1.1.5)
We set θ =(µ−rσ
), called the market price of stock risk [5]
or risk premium of stock price, and define
W ∗t = Wt +
∫ t
0
θds = Wt + θt, (1.1.6)
so that
dSt = σStdW∗t . (1.1.7)
We denote by ξθT = e−θWT− 12θ2T . By the property of log-normal distribu-
tion, ξθT has an expected value with respect to P equal to 1.
We now introduce the probability measure P∗. It is the equivalent measure
to P , meaning that both have the same null sets(they agree on which events
have probability 0), which has the density ξθT with respect to P .
Denoting by E∗· the expectation with respect to P∗, for any integrable
random variable X we have
E∗[X] = E[ξθTX].
The process (ξθt )0≤t≤T is called Radon−Nikodym process. [2]
1.2 Multidimensional Stochastic Calculus
This section defines multidimensional Ito formula, Girsanov’s theorem , Feynman-
Kac formula and develops their properties. These are are used to model the
value of a portfolio that results from trading assets in continuous time.
3
CHAPTER 1. INTRODUCTION
We consider the generalization of the stochastic differential equations
dSt = µ(t, St)dt+ σ(t, St)dWt to the case of systems of such equations:
dSit = µi(t, St)dt+d∑j=1
σi,jdWjt , i = 1, · · · , d,
where W jt are d−independent standard Brownian motions, and
St = (S1t , · · · , Sdt ) is a d−dimensional process. The d− dimensional Ito formula
can then be written:
df(t, St) =∂f
∂t(t, St)dt+
d∑i=1
∂f
∂si(t, St)dS
it
+1
2
d∑i,j=1
∂2f
∂si∂sj(t, St)d < Si, Sj >t,
(1.2.1)
where
d < Si, Sj >t=d∑
k=1
σi,k(t, St)σj,k(t, St)dt = (σσT )i,j(t, St)dt (1.2.2)
follows from the cross-variation rules d < t,W jt >= d < W j
t , t >= 0,
d < W it ,W
jt >= d < W j
t ,Wit >= 0 for i 6= j, and d < W i
t ,Wit >= dt.
(1.2.1) can then be rewritten:
df(t, St) =
(∂f
∂t+
d∑i=1
µi∂f
∂si+
1
2
d∑i,j=1
(σσT )i,j∂2f
∂si∂sj
)dt
+d∑i=1
∂f
∂si
(d∑i=1
σi,jdWjt
),
(1.2.3)
where the partial derivatives of f and the coefficients µ and σ are evaluated
at (t, St).
4
CHAPTER 1. INTRODUCTION
In (1.1.6) we have used a change of probability measure so that the
1−dimensional process W ∗t = Wt + θt becomes a standard Brownian mo-
tion under the new probability P∗. We assume that the d−dimensional pro-
cess (θt) is of the form (θj(St), j = 1, · · · , d),. We define the real process
(ξθt )0≤t≤T by:
ξθt = exp
(−
d∑j=i
(∫ t
0
θj(Ss)dWjs +
1
2
∫ t
0
θ2j (Ss)ds
)), (1.2.4)
which satisfies
dξθt = ξθt
d∑j=1
θj(St)dWjt .
and, therefore, is a martingale. Assume that E∫ T0‖θj‖2 ξθt
2ds < ∞, then
Girsanov′s− theorem states that the processes (W j∗t )0≤t≤T , j = 1, · · · , d,
defined by
W j∗t = W j
t +
∫ t
0
θj(Ss)ds, j = 1, · · · , d, (1.2.5)
are independent standard Brownian motions under P∗.Consider the stochastic differential equation
dSt = β(t, St)dt+ γ(t, St)dWt. (1.2.6)
Let h(s) be a Borel-measurable function and let r be constant. Fix T > 0,
and let t ∈ [0, T ] be given. Define the function
f(t, s) = Et,s[e−r(T−t)h(ST )],
assume that Et,s |h(ST )| <∞. By Feynman−Kac Theorem, then f(t, s)
satisfies the partial differential equation
ft(t, s) + β(t, s)fs(t, s) +1
2γ2(t, s)fss(t, s) = rf(t, s) (1.2.7)
and the terminal condition
f(T, s) = h(s).
5
CHAPTER 1. INTRODUCTION
Let h(S(T )) be the payoff at time T of a derivative security whose un-
derlying asset is the geometric Brownian motion
dS(t) = αS(t)dt+ σS(t)dW (t).
We may rewrite this as
dS(t) = rS(t)dt+ σS(t)dW (t),
where W (t)is a Brownian motion under the risk-neutral probability measure
P∗. The price of the derivative security at time t is
V (t) = E[e−r(T−t)h(S(T ))|F(t)].
Because the stock price is Markov and the payoff is a function of the stock
alone, there is a function p(t, s) such that V (t) = v(t, S(t)). Using Feynman-
Kac Theorem, function p(t, s) must satisfy the following discounted partial
differential equation. This is the Black− Shocles−Merton equation
vt(t, s) + rsvs(t, s) +1
2σ2s2vss(t, s) = rv(t, s).
One should never offer prices derived from a model that admit free-arbitrage.
So we define a Complete market model as every derivative security can
be hedged.
Followings are the Fundamental Theorem of Asset Pricing(FTAP).
If a market model has a risk-neutral probability measure, then it does not
admit arbitrage(FTAP1). Consider a market model that has a risk-neutral
probability measure. The model is complete if and only if the risk-neutral
probability measure is unique(FTAP2). All section provides conditions that
guarantee that such such a model does not admit arbitrage and that every
derivative security in the model can be hedged.[3]
6
Chapter 2
Model
2.1 Asset price model
In this section, we consider an underlying asset price model given by the
stochastic differential equations (SDEs)
dSt = (µt −D)Stdt+ f(Yt)StdWst , (2.1.1)
dYt =1
ε(m− Yt)dt+
ν√
2√εdW y
t , (2.1.2)
where St is the underlying risky asset price and f is a smooth positive
function that is bounded away from zero and f is the function of fast-mean
reverting process Yt. Also, W st and W y
t are standard Brownian motions mu-
tually correlated such that d〈W s,W y〉t = ρxydt. We note that the solution
of the Ornstein–Uhlenbeck SDE (2.1.2) is a Gaussian process with a distri-
bution given by Yt ∼ N(m + (Y0 − m)e−1εt, ν2(1 − e−
2εt)). So, its invariant
distribution is given by N(m, ν2). Later, we will use notation 〈 · 〉 for the
average with respect to this invariant distribution.
Since the model (2.1.1) and (2.1.2) describes an incomplete market, there
is a possibility of more than one equivalent martingale measure. Using nota-
tion γ (the market prices of volatility risk) for the non-uniqueness and rt for
7
CHAPTER 2. MODEL
an interest rate, the processes defined by
W(s)∗t = W s
t +
∫ t
0
(µt − rt)/f(Ys)ds,
W(y)∗t = W y
t +
∫ t
0
γ(Ys)ds
are Brownian motions under a risk-neutral measure, say P ∗γ , by the Gir-
sanov theorem, where we assume that (µt−rtf(Yt)
, γ(Yt)) satisfies the Novikov
condition. Here, γ is assumed to be smooth bounded functions of y only.
We can find a P (t, s, r, y) by trying to construct a hedged portfolio of
assets which can be priced by the no-arbitrage principle. Unlike in the Black-
Scholes case(Geometric Brownian Motion), it is not sufficient to hedge solely
with the underlying asset St, since the dW(s)∗t terms can be balanced, but
the dW(y)∗t term cannot. So we try to hedge with the underlying asset(P (1))
and another option which has a different expiration date(P (2)), and which
we also trade continuously.
Let P (1)(t, s, r, y) be the price of a European derivative with expiration
date T 1, and P (2)(t, s, r, y) be the price of another European derivative with
expiration date T 2 > T 1. We try to find processes M1t ,M
2t , Bt such that the
self − financing portfolio (No further funds or assets are required after
the initial investment so that if, for example, more of the asset is bought,
then money would have to be obtained by selling bonds to pay for it)
Πt = M1t P
(1)(t, St, r, Yt)−BtSt −M2t P
(2)(t, St, r, Yt)
instantaneously riskless (or hedged) at any time t < T .
By 2-dimensional Ito’s formula,
dg =
(∂g
∂t+ L(S,Y )g
)dt+
(f(Yt)St
∂g
∂s+ ρsy
ν
ε√
2
∂g
∂y
)dW s∗
t
+√
1− ρ2syν
ε√
2
∂g
∂ydW y∗
t
where LS,Y is the infinitesimal generator of (S, Y ) given by
8
CHAPTER 2. MODEL
LS,Y =1
2f 2(y)s2
∂2
∂s2+ ρsy
ν
ε√
2f(y)s
∂2
∂s∂y+ν2
4ε2∂2
∂y2
+ µts∂
∂s+
1
ε(m− y)
∂
∂y
for smooth function g.
By the self-financing property,
dΠt = M1t dP
(1)(t, St, r, Yt)−BtdSt −M2t dP
(2)(t, St, r, Yt),
and applying 2-dimensional Ito’s formula to P (1) and P (2) yields
dΠt = M1t dP
(1)(t, St, r, Yt)−BtdSt −M2t dP
(2)(t, St, r, Yt)
=
(M1
t
[∂
∂t+ LS,Y
]P (1) −BtµtSt −M2
t
[∂
∂t+ LS,Y
]P (2)
)dt
+ Stf(Yt)
[M1
t
∂P (1)
∂s−M2
t
∂P (2)
∂s−Bt
]dW s∗
t
+ ρsyν
ε√
2
[M1
t
∂P (1)
∂y−M2
t
∂P (2)
∂y
]dW s∗
t
+√
1− ρ2syν
ε√
2
[M1
t
∂P (1)
∂y−M2
t
∂P (2)
∂y
]dW y∗
t
.
We want to hedge the risk from W y∗t , we can choose
Σt = M1t
(∂P (2)
∂y
)−1(∂P (1)
∂y
).
To hedge the remaining risk from W s∗t , we choose
At = M(1)t
∂P (1)
∂s− Σt
∂P (2)
∂s.
For there to be no arbitrage, the now riskless portfolio Πt must grow at
the risk-free rate: dΠt = rΠtdt, which gives
9
CHAPTER 2. MODEL
(∂P (1)
∂y
)−1LP (1) =
(∂P (2)
∂y
)−1LP (2),
with the notation
L =∂
∂t+ LS,Y − (µt − r)s
∂
∂s− r · .
The left-hand side contains terms depending on T 1 but not T 2 and vice
versa for the right-hand side. Thus both sides must be equal to a function
that does not depend on expiration date T 1, T 2. We denote this function
Λ(y) =ρsy(µt − rt)
f(y)+ γ(y)
√1− ρ2sy
and γ(t, s, y) is an arbitrary function.
The Vasicek model is chosen as an interest model for rt. Then under the
risk-neutral measure P ∗γ the evolution dynamics of (2.1.1) and (2.1.2) are
transformed into
dSt = (rt −D)Stdt+ f(Yt)StdW(s)∗t , (2.1.3)
drt = (b− art) dt+ σdW(r)t = a
(b
a− rt
)dt+ σdW
(r)t , (2.1.4)
dYt =
(1
ε(m− Yt)−
ν√
2√ε
Λ(Yt)
)dt+
ν√
2√εdW
(y)∗t , (2.1.5)
where a (mean reversion rate of interest), b (average direction of interest
rate movement) and σ (volatility of interest rate) are constants and we assume
that W (r) is correlated with W (s)∗ such that d〈W (s)∗,W (r)〉t = ρsrdt but it
is independent of W (y)∗. Figure (2.1) shows the paths of interest rate from
(2.1.4). As one can notice from figure, if we assume more large volatility,
then interest rate tends to the long run mean more sensitive. Note that the
market price of interest rate risk is absorbed by the term b(t) based on an
argument in Pelsser (2000).
As opposed to stock prices for instance, interest rates cannot rise in-
definitely. This is because at very high levels they would hamper economic
10
CHAPTER 2. MODEL
activity, prompting a decrease in interest rates. Similarly, interest rates can
not decrease below 0. As a result, interest rates move in a limited range,
showing a tendency to revert to a long run value. So we choose OU-process
to model long-run interest rates.
Figure 2.1: Interest rate for long run mean is 0.02.
2.2 Problem Formulation
Considering the Markovian property of the joint process (St, rt, Yt) and by
using notation E∗{·} for the expectation with respect to the risk-neutral
11
CHAPTER 2. MODEL
measure, the no-arbitrage price of a European option with a payoff function
h(s) is given by
P ε(t, s, r, y) = E∗{e−∫ Tt rsdsh(ST )|St = s, rt = r, Yt = y}, (2.2.1)
where the dependence on the small parameter ε is expressed explicitly. As
usual, by the application of the Feynman–Kac formula (cf. Oksendal, 2002
for instance) to (2.1.3 - 2.1.5), we obtain a characterization of P ε(t, x, r, y)
as a solution of the terminal value problem
LεP ε(t, s, r, y) = 0, t < T,
P ε(T, s, r, y) = h(s),(2.2.2)
where the partial differential operator Lε is given by
Lε =1
εL0 +
1√εL1 + L2, (2.2.3)
by using the notation
L0 = (m− y)∂
∂y+ ν2
∂2
∂y2, (2.2.4)
L1 = ν√
2
(ρsyf(y)s
∂2
∂s∂y− Λ(y)
∂
∂y
), (2.2.5)
L2 =∂
∂t+
1
2f 2(y)s2
∂2
∂s2+ (r −D)(s
∂
∂s)− r·
+ ρsrσf(y)s∂2
∂s∂r+ (b− ar) ∂
∂r+
1
2σ2 ∂
2
∂r2
(2.2.6)
with the terminal condition P (T, s, r, y) = h(s).
12
Chapter 3
Asymptotics for Pricing
In this part, a multi scale asymptotic method(Fouque, Papanicolaou, Sir-
cac, and Solna(2011)) is used to derive an approximated European option
price for the model given by (2.1.3)-(2.1.5). Also, the error of the corrected
price is obtained.
3.1 The Formal Expansion
The method is to expand the solution
P ε = P0 +√εP1 + εP2 + ε
√εP3 + · · · , (3.1.1)
where P0, P1, · · · are functions of (t, s, r, y) to be determined such that
P0(T, s, r, y) = h(s). We are primarily interested in the first two terms
P0 +√εP1. The terminal condition for the second term is P1(T, s, r, y) = 0.
Also, we called P0 +√εP1 as corrected price.
Plugging (3.1.1) into (2.2.2), the leading order term P ε0 is given by the
unique solution to the problem(1
εL0 +
1√εL1 + L2
)P ε0(t, s, r, y) = 0 (3.1.2)
13
CHAPTER 3. ASYMPTOTICS FOR PRICING
with the terminal condition P ε0(T, s, r, y) = h(s).
To derive an explicit expression for P0 and P1, respectively, we substitute the
expansion (3.1.1) into (3.1.2) and obtain
1
εL0P0
+1√ε(L0P1 + L1P0)
+ (L0P2 + L1P1 + L2P0)
+√ε(L0P3 + L1P2 + L2P1)
+ · · · = 0.
(3.1.3)
Equating terms of order 1ε, we must have
L0P0 = 0. (3.1.4)
The operator L0 acts only on the y variable; hence P0 must be a constant
with respect to that variable, which implies that
P0 = P0(t, s, r), (3.1.5)
a function of (t, s, r) only.
Moreover, in order to eliminate the term in 1√ε
in (3.1.3), we must have
L0P1 + L1P0 = 0. (3.1.6)
The operator L1 given by (2.2.5) takes derivatives with respect to y,
and we therefore deduce from (3.1.5) that L1P0 = 0 and consequently that
L0P1 = 0. P1 also does not depend on y so that
P1 = P1(t, s, r), (3.1.7)
a function of (t, s, r) only. This implies in particular that the corrected
price P0 +√εP1 will not depend on the present volatility y.
14
CHAPTER 3. ASYMPTOTICS FOR PRICING
3.2 Poisson Equations
From the O(1) terms in (3.1.3) and the y- independence of P1, we already
know L1P1 = 0 and have
L0P2 + L2P0 = 0 (3.2.1)
The variable x being fixed, L2P0 is a function of y since L2 involves f(y).
Focusing on the y dependence only, equation (3.2.1) is of the form
L0χ+ g = 0, (3.2.2)
which is known as a Poisson equation for χ(y) with respect to the
operator L0 in the variable y. Let us now consider Yt an Ornstein-Uhlenbeck
process as the solution of the stochastic differential equation (2.1.2)
dYt =1
ε(m− Yt)dt+
ν√
2√εdW y
t ,
where dW yt is a standard Brownian motion. We wrote the explicit solution
for Yt, which showed that Yt is a Gaussian process. Define the infinitesimal
generator of the Markov process Yt is the differential operator
L =1
ε(m− y)
∂
∂y+ν√
2√ε
∂2
∂2y.
We look for an invariant distribution, which will be a distribution for
Y ∗ such that ELg(Y ∗) = 0 for any smooth and bounded g. We denote the
density function of Y ∗ as Φ(y), which satisfies∫∞−∞Φ(y)Lg(y)dy = 0 for any
g. Using integration by parts yields∫∞−∞ g(y)L∗Φ(y)dy = 0, where L∗, the
adjoint of L, is L∗ = −1ε∂∂y
((m − y)·) + ν√2√ε∂2
∂2yand there are no boundary
terms because Φ,Φ′ → 0 as y → ±∞. If equation holds any smooth test
function g, then Φ should satisfy
L∗Φ =2ν2
εΦ′′ − 1
ε((m− y)Φ)′ = 0,
where differentiations are with respect to the variable y. Solving this
second-order ordinary differential equation with the constraint∫∞−∞Φ = 1
15
CHAPTER 3. ASYMPTOTICS FOR PRICING
gives
Φ(y) =1√
2πν2exp
(−(y −m)2
2ν2
).
The centering condition of Poisson equation
< g >=
∫g(y)Φ(y)dy = 0 (3.2.3)
is necessary for (3.2.2) to admit a solution. We average (3.2.2) with respect
to the invariant distribution of Yt, integrate by parts, and use the adjoint
operator L∗ and its property L∗Φ = 0 :
< g > = − < L0χ >
= −∫
(L0χ(y))Φ(y)dy
=
∫χ(y)(L∗0Φ(y))dy
= 0,
(3.2.4)
where there are no boundary terms because Φ,Φ′ → 0 as y → ±∞.Using the bracket notation < · > denoting integration with respect to the
Gaussian N (m, ν2) density, the centering condition of (3.2.1) is expressed as
< L2P0 >= 0.
Since P0 does not depend on y, this is < L2 > P0 = 0. Therefore, we
obtain a terminal value problem for the first term of (3.1.1) as follows.
< L2 > P0(t, s, r) = 0, t < T,
P0(T, s, r) = h(s),(3.2.5)
where
< L2 > =∂
∂t+
1
2< f 2(y) > s2
∂2
∂s2+ (r −D)(s
∂
∂s)− r·
+ ρsrσ < f(y) > s∂2
∂s∂r+ (b− ar) ∂
∂r+
1
2σ2 ∂
2
∂r2:= Lsvsi
(3.2.6)
16
CHAPTER 3. ASYMPTOTICS FOR PRICING
So, < L2 > exactly corresponds to a pricing operator for an asset pricing
model with a constant volatility replaced by the effective volatility√< f 2 > = σ. Then, from the already known formula in Brigo and Mercu-
rio(2007), P0 is given by the explicit solution
P0(t, s, r) = se−D(T−t)N(d1)−KP (t, T )N(d2),
d1 :=
12∆1 +
(1 + b
aβ
)∆2 − (r −D)
(e−a(T−t)−1
a
)+ ln( s
K)
12∆1 + 2∆2 + ∆3
,
d2 := d1 −√
∆1 + 2∆2 + ∆3
P (t, T ) := exp
(1
2∆3 −
b
β∆2 + (r −D)
(e−a(T−t) − 1
a
)),
∆1 := σ2 =< f 2 >=
∫ ∞−∞
f 2(y)Φ(y)dy,
∆2 :=β
a
(T − t+
e−a(T−t) − 1
a
),
∆3 :=σ2
a2
((T − t)− 3 + e−a(T−t)(e−a(T−t) − 4)
2a
)β :=
√< f 2 >σρsr
(3.2.7)
where N is the usual standard normal distribution function. The de-
tailed derivation of (3.2.7) uses Mellin−Transform. Definition of Mellin-
transform is given by Appendix. Also we note that P (t, T ) represents the cur-
rent price of pure discount bond from time t to maturity T under stochastic
interest rate.
In our situation, the centering condition of (3.2.1) gives
< L2P0 >= 0
Since P0 does not depend on y, this is < L2 > P0 = 0.
By (3.2.6), LsvsiP0 = 0 and we can write
L2P0 = L2P0− < L2P0 >=1
2
(f(y)2 − σ2
)s2∂2P0
∂s2+
1
2ρsrσ(f(y)− < f(y) >)s
∂2P0
∂s∂r.
17
CHAPTER 3. ASYMPTOTICS FOR PRICING
The second-order correction P2, solution of the L0P2 = −L2P0, is then given
by
P2(t, s, r, y)
= −1
2L−10
((f(y)2 − σ2)s2
∂2P0
∂s2+ ρsrσ(f(y)− < f(y) >)s
∂2P0
∂s∂r
)=
1
2
((φ(y) + c(t, s, r))s2
∂2P0
∂s2+ ρsrσ(ψ(y) + d(t, s, r))s
∂2P0
∂s∂r
) (3.2.8)
where φ(y), ψ(y) is a solution of the Poisson equation
L0φ = f(y)2 − σ2,L0ψ = f(y)− < f(y) >
and c(t, s, r), d(t, s, r) is a constant in y that may depend on (t, s, r).
Next, we derive an expression for P1. The order√ε terms in the expansion
(3.1.3) give the following Poisson equation for P3:
L0P3 + L1P2 + L2P1 = 0 (3.2.9)
The solvability condition for this equation is
< L1P2 + L2P1 >= 0. (3.2.10)
P1 does not depend on y, we can rewrite this centering condition as
< L1P2 > + < L2 > P1 = 0. (3.2.11)
Using the fact that < L2 > P0 = 0 and (3.2.1), we can write
P2 = −L−10 (L2− < L2 >)P0 + c(t, s, r). (3.2.12)
But c(t, s, r) does not depend on y. So L1c(t, s, r) = 0, and then
< L1P2 >= − < L1L−10 (L2− < L2 >)P0.
To write the problem defining P ε1 =
√εP1, it is convenient to intro-
duce the operator Aε =√ε < L1L−10 (L2− < L2 >). With this notation,
LsvsiP ε1(t, s, r) = AεP0(t, s, r) with a terminal condition P ε
1(T, s, r) = 0
18
CHAPTER 3. ASYMPTOTICS FOR PRICING
3.3 Computation of Aε
From (3.2.8) to (3.2.12), we deduce that
< L2 > P1 = LsvsiP1
= − < L1P2 >
=1
2< L1
(φ(y)s2
∂2P0
∂s2+ ρsrσψ(y)s
∂2P0
∂s∂r
)> .
(3.3.1)
First, operator Aε has the following explicit representation
AεP0(t, s, r) = LsvsiP ε1
=√εLsvsiP1
= −√ε < L1P2 >
= V ε3 s
3∂3P0
∂s3+ V ε
2 s2∂
2P0
∂s2+ U ε
3s3 ∂
3P0
∂s2∂r+ U ε
2s∂2P0
∂s∂r,
V ε3 :=
√εν√2ρsy < f
∂φ
∂y>
V ε2 := −
√εν√2< Λ
∂φ
∂y>
U ε3 :=
√εν√2ρsrσρsy < f
∂ψ
∂y>
U ε2 := −
√εν√2ρsrσ < Λ
∂ψ
∂y>
(3.3.2)
Second, the operator Lsvsi satisfies the following identities.
Lsvsi(−(T − t)f) = f − (T − t)Lsvsif
Lsvsi(sn∂nf
∂sn
)= sn
∂n
∂snLsvsif
Lsvsis∂
∂s
(sn∂nf
∂sn
)= s
∂
∂s
(sn∂n
∂sn
)Lsvsif
(3.3.3)
19
CHAPTER 3. ASYMPTOTICS FOR PRICING
Using above two results, P ε1 is given explicitly in terms of P0.
P ε1 = −(T − t)AεP0
+ w1(T − t)(U ε3s∂
∂s
(s2∂2
∂s2
)+ U ε
2s2 ∂
2
∂s2
)P0
+ w2(T − t)(U ε3s∂
∂s
(s∂2
∂s∂r
)+ U ε
2s∂2
∂s∂r
)P0
(3.3.4)
where P0 is given by (3.2.7). Also, w1, w2 are functions of (T − t) and
given by
w1(T − t) =1
a
(T − t+
1− ea(T−t)
a
),
w2(T − t) = T − t+1− ea(T−t)
a.
By (3.2.7), (3.3.4) the corrected price is given explicitly by
P0 + P ε1 = P0 − (T − t)AεP0
+ w1(T − t)(U ε3s∂
∂s
(s2∂2
∂s2
)+ U ε
2s2 ∂
2
∂s2
)P0
+ w2(T − t)(U ε3s∂
∂s
(s∂2
∂s∂r
)+ U ε
2s∂2
∂s∂r
)P0
(3.3.5)
3.4 Accuracy of the Approxmation
Our object in this section is to show that the approximation corrected price
to the price P is of order ε, in the sense that
|P − (P0 + P ε1)| ≤ constant · ε.
In order to do this, one can use the expansion (3.1.1) up to order 3 in√ε
and write
20
CHAPTER 3. ASYMPTOTICS FOR PRICING
P ε = P0 +√εP1 + εP2 + ε
√εP3 − Zε.
At time T, P (T, s, r, y) = P0(T, s, r, y) = h(s, r) and P1(T, s, r) = 0,
Zε(T, s, r, y) = ε(P2(T, s, r, y) +√εP3(T, s, r, y))
Let Lε = 1εL0 + 1√
εL1 +L2, then P ε solves the original equation LεP ε = 0.
We can compute LεZε:
LεZε = Lε(P0 +√εP1 + εP2 + ε
√εP3 − P ε)
=
(1
εL0 +
1√εL1 + L2
)(P0 +
√εP1 + εP2 + ε
√εP3)− LεP ε
=1
εL0P0 +
1√ε(L0P1 + L1P0)
+ (L0P2 + L1P1 + L2P0) +√ε(L0P3 + L1P2 + L2P1)
+ ε(L1P3 + L2P2 +√εL2P3)
= ε(L1P3 + L2P2) + ε32L2P3 = O(ε).
(3.4.1)
We also known that Zε = O(ε), define
F ε(t, s, r, y) = L1P3(t, s, r, y) + L2P2(t, s, r, y) +√εL2P3(t, s, r, y)
Gε(s, r, y) = P2(T, s, r, y) +√εP3(T, s, r, y),
(3.4.2)
then LεZε = εF ε and Zε(T, s, r, y) = εGε. Under the smoothness and bound-
edness assumption on the payoff function h and the boundedness of f(y)
and Λ(y), F ε and Gε are bounded uniformly in s, r and are at most linearly
growing in |y|; Consequently,
P (t, s, r, y) = P0(t, s, r) + P ε1(t, s, r) +O(ε),
which shows that the error in our approximation is of order ε.
21
Chapter 4
Numerical Results
In this section, we’ll compare Call option price using the various meth-
ods such as Explicit Finite Difference method, Implicit Finite Difference
method, Crank-Nicolson Finite Difference method, Monte-Carlo simulation
with closed solution P0(3.2.7) and Black-Scholes Formula. Also, we assume
that S0 = 970, K = 1000, r = 0.03, D = 0.01, t = 0, T = 1, < f 2(y) >= 0.01.
For a simple and convenient calculation, we assume that interest rate r is
constant. In this situation, (3.2.6) is convert to
< L2 >=∂
∂t+
1
2< f 2(y) > s2
∂2
∂s2+ (r −D)
(s∂
∂s
)− r· = 0
Therefore, we obtain a terminal value problem for the first term of (3.1.1) as
follows.
< L2 > P0(t, s) = 0, t < T,
P0(T, s) = h(s),
h(s) =
{ST −K if ST > K
0 if ST ≤ K
(4.0.1)
22
CHAPTER 4. NUMERICAL RESULTS
1. Monte−Carlo Simulation
Monte−Carlo simulation is based on random number generation; ac-
tually, we must speak of pseudo-random numbers, since nothing is random
on a computer. If we feed random numbers into a simulation procedure, the
output will be a sequence of random numbers. Given this output, we use sta-
tistical techniques to build an estimate of a quantity of interest. Intuitively,
the more replications we run, the more reliable our estimates will be. [6]
Figure 4.1: Monte-Carlo Simulation
23
CHAPTER 4. NUMERICAL RESULTS
2. Explicit Finite Difference Method
Starting from the initial conditions(j = 0), we may solve the equation for
increasing values of j = 1, 2, · · · ,M . Since the unknown values are given by
explicit expression, this approach is called explicit. [4]
∂P
∂t=Pi,j − Pi,j−1
∆t+O(∆t)
∂P
∂s=Pi+1,j − Pi−1,j−1
2∆s+O(∆s2)
∂2P
∂s2=Pi+1,j − 2Pi,j + Pi−1,j
∆s2+O(∆s2)
Figure 4.2: Explicit method
24
CHAPTER 4. NUMERICAL RESULTS
3. Implicit Finite Difference Method
If we use a forward approximation for the derivative with respect to time,
we get an explicit method. We get a completely different scheme if we use
a backward approximation. Thus, the unknown values are given implicitly,
which is where the implicit method. [4]
∂P
∂t=Pi,j+1 − Pi,j
∆t+O(∆t)
∂P
∂s=Pi+1,j − Pi−1,j
2∆s+O(∆s2)
∂2P
∂s2=Pi+1,j − 2Pi,j + Pi−1,j
∆s2+O(∆s2)
Figure 4.3: Implicit method
25
CHAPTER 4. NUMERICAL RESULTS
4. Crank−Nicolson Finite Difference Method
We have seen methods involving three points on one time layer and one on
a neighboring layer. It is natural to wonder if a better scheme may be obtained
by considering three points on both layers. The Crank−Nicolson scheme
may be analyzed in a more general framework. [4] We may think of using a
convex combination of two approximations of the second-order derivative in
the finite difference scheme(Assume that λ = 0.5):
∂P
∂t=Pi,j+1 − Pi,j
∆t+O(∆t)
∂P
∂s= λ
Pi+1,j+1 − Pi−1,j+1
2∆s+ [1− λ]
Pi+1,j − Pi−1,j2∆s
+O(∆s2)
∂2P
∂s2= λ
Pi+1,j+1 − 2Pi,j+1 + Pi−1,j+1
∆s2
+ [1− λ]Pi+1,j − 2Pi,j + Pi−1,j
∆s2+O(∆s2)
Figure 4.4: Crank-Nicolson method
26
CHAPTER 4. NUMERICAL RESULTS
(a) Explicit method (b) Implicit method
Figure 4.5: Explicit, Implicit method with Various maturity time t
(c) Crank-Nicolson method
Figure 4.6: Crank-Nicolson method with Various maturity time t
27
CHAPTER 4. NUMERICAL RESULTS
S0 = 970, K = 1000, r = 0.03, D = 0.01, < f 2(y) >= 0.01, σ = 0.1
Black-Scholes P0 Explicit F.D.M. Implicit F.D.M. Crank-Nicolson F.D.M.
33.9703 34.1212 34.0740 33.8393 34.1517
99 % confidence interval [33.0603 , 35.0927]
Yes Yes Yes Yes Yes
95 % confidence interval [33.3045 , 34.8485]
Yes Yes Yes Yes Yes
Table 4.1: Call option price with S0 = 970 using various method
S0 = 1000, K = 1000, r = 0.03, D = 0.01, < f 2(y) >= 0.01, σ = 0.1
Black-Scholes P0 Explicit F.D.M. Implicit F.D.M. Crank-Nicolson F.D.M.
44.8482 44.7856 44.9703 44.6806 44.4856
99 % confidence interval [43.3825 , 45.7445]
Yes Yes Yes Yes Yes
95 % confidence interval [43.6663 , 45.4607]
Yes Yes Yes Yes Yes
Table 4.2: Call option price with S0 = 1000 using various method
28
Chapter 5
Conclusion
In Conclusion, we can obtain formal derivation of the first-order price
approximation of European options with dividends under a stochastic volatil-
ity and stochastic interest rate. We choose the Ornstein-Uhlenbeck process
and Hull-White process for a stochastic volatility model and interest rate
model, respectively. Based on this chosen model, we can conclude two main
results. First, we can obtain a closed formula for the first-order European
option price with dividend under stochastic volatility and stochastic interest
rate by applying the centering condition of Poisson Equation and method
of Mellin transform. This formula is simply given by the cumulative density
function of standard normal distribution. We can find not only the closed
formula to European put options by using Put-Call Parity, but also the price
that shaped similar to European options by using the Mellin transform. Sec-
ond, we can observe that the result brings out a minor difference by using
Black-Scholes formula and other numerical methods: finite difference method
with stochastic volatility and Monte-Carlo Simulation. However, in Black-
Scholes formula, drift and diffusion terms of stochastic differential equations
are constant. So, we can conclude that the stochastic model brings out better
results than pricing that using a constant volatility and interest rate. And
for Monte-Carlo Simulation, the more replications we run, the more reliable
our estimates we can derive.
29
Appendix
To derive the (3.2.7), we first need to define Mellin transform. For
a locally Lebesgue integrable function f(y), y ∈ R+, the Mellin transform
M(f(y), w), w ∈ C, is defined by
M(f(y), w) := f(w) =
∫ ∞0
f(y)yw−1dy,
and if a < Re(w) < b and c such that a < c < b exist, the inverse of the
Mellin transform is expressed by
f(y) =M−1(f(w)) =1
2πi
∫ c+i∞
c−i∞f(w)y−wdw.
If we define the sequence of the payoff function hn(s) such that limn→∞ hn(s) =
h(s) and the call option price with the payoff function hn(s) = pn(T, s, r) as
pn(t, s, r) = E∗[exp
(−∫ T
t
r∗t dt∗)hn(ST )|St = s, rt = r
],
pn(t, s, r) satisfies the PDE given by (3.2.6). If we find the solution pn(t, s, r)
by using the Mellin transform, we can obtain the formula of the option price
P (t, s, r) from P (t, s, r) = limn→∞ pn(t, s, r). If we define pn(t, w, r) as the
Mellin transform of Pn(t, s, r), then
pn(t, s, r) =1
2πi
∫ c+i∞
c−i∞pn(t, w, r)s−wds.
30
Bibliography
[1] Shreve, Steven E. stochastic calculus for finance II Continuous-time
models, (2004).
[2] Jean-Pierre Fouque, George Papanicolaou, K. Ronnie Sircar, Multiscale
Stochastic Volatility for Equity Interest Rate and Credit Derivatives.
(2011), pp. 157-175.
[3] Jean-Pierre Fouque, George Papanicolaou, K. Ronnie Sircar, Derivatives
in financial markets with stochastic volatility (2000), pp. 501-556.
[4] Paolo Brandimarte, Numerical methods in finance and economics
(2006), pp. 1779-1799.
[5] Robert L. McDonald, Derivatives markets (2006), pp. 161-171.
[6] Don L. McLeish, Monte Carlo simulation and finance (2005), pp. 57-
108.
31
국문초록
이 논문에서 우리는 파인만-칵 정리와 포아송 방정식을 이용하여 확률론
적 변동성과 확률론적 이자율 모형 하에서의 배당을 고려한 유럽형 옵션의
가격을 유도한다. 확률론적 변동성은 오스틴-유린벡 과정을 통해 모형화하고
이를위험중립측도하에서유도한다.마지막으로변동성과독립인옵션가격의
근사치를 구하고 그 차이를 몬테-카를로 시뮬레이션과 비한다.
주요어휘: 확률론적 변동성모형, 확률론적 이자율모형, 파인만-칵 정리, 배당
이 존재하는 유럽형 옵션, 오스틴-유린벡 과정, 포아송 방정식
학번: 2012-20243