Вы находитесь на странице: 1из 26

Option Pricing in a Black-Scholes Model with Markov Switching

Fuh, Cheng-Der1 and Wang, Ren-Her


Institute of Statistical Science, Academia Sinica, Taipei, Taiwan, R.O.C.
Cheng, Jui-Chi
Graduate Institute of Finance, National Taiwan University, Taipei, Taiwan, R.O.C.

ABSTRACT
The theory of option pricing in Markov volatility models has been developed in recent
years. However, an efficient method to compute option price in this setting remains
lacking. In this article, we present a way of modeling time-varying volatility; to generalize
the classical Black-Scholes model to encompass regime-switching properties. Specifically,
the unobserved state variables for stock fluctuations are governed by a first-order Markov
process, and the drift and volatility parameters take different values depending on the
state of this hidden Markov model. Standard option pricing procedure under this model
becomes problematic. We first provide a closed-form formula for the arbitrage-free price
of the European call option. In the case of the underlying Markov process has two
states, we have an explicit analytic formula for option price. When the number of states
for the underlying Markov process is bigger than two, we propose an approximation
formula for option price. The idea of the approximation formula is that we replace
the distribution of occupation time for a given state by its corresponding stationary
distribution multiplies the whole time period. Numerical methods, such as Monte-Carlo
simulation and Markovian tree, to compute European call option price are presented
for comparison. Our numerical results show that the approximation formula provide
an efficient and reliable implementation tool for option pricing. Implied volatility and
implied volatility surface are also given.

Keywords: Arbitrage, Black-Scholes model, European call option, hidden Markov model,
implied volatility, Laplace transform, Monte Carlo, tree.
1. Research partially supported by NSC 91-2118-M-001-016.

1.

Introduction

It is well-known that the volatility of financial data series tends to change over time,
and changes in return volatility of stock returns tend to be persistent. The statistical
properties of return volatility have been deeply studied and uncovered in the financial
economic literatures, for instance, the extensive work on modeling stock fluctuations
with stochastic volatility (cf. Anderson, 1996; Hull and White, 1987; Stein, 1991; Wiggins, 1987; Heston, 1993), and uncertain volatility (Avellaneda, Levy, and Par
as, 1995).
Many efforts have been made to study financial markets with different information levels
among investors (cf. Duffie and Huang, 1986; Ross, 1989; Anderson, 1996; Karatzas and
Pikovsky, 1996; Guilaume et. al., 1997; Grorud and Pontier, 1998; Imkeller and Weisz,
1999). Like the Markov switching model of Hamilton (1988, 1989) may provide more
appropriate modeling of volatility. In this regard, Hamilton and Susmel (1994) proposed
the Markov switching ARCH model. In a partial equilibrium model, Turner, Startz and
Nelson (1989) formulate a switching model of excess returns in which returns switch exogenously between a Gaussian low variance regime and a Gaussian high variance regime.
So, Lam and Li (1998) generalizes the stochastic volatility model to incorporate Markov
regime switching properties. Diebold and Inoue (2001) considers the properties of long
memory and regime switching.
In deed, recent research (cf. Bittlingmayer, 1998) has shown that investors uncertainty over some important factors affecting the economy may greatly impact the
volatility of stock returns. More generally, there is evidence that investors tend to be
more uncertain about the future growth rate of the economy during recessions, thereby
partly justifying a higher volatility of stock returns. In this paper, to encompass the
empirical phenomena of the stock fluctuations relate to business cycle, we introduce a
model of an incomplete market by adjoining the Black-Scholes exponential Brownian
motion model for stock fluctuations with a hidden Markov process. Specifically, we
assume that stock of asset pricing are generated by realization of a Gaussian diffusion
process, and the drift and volatility parameters take different values depending on the
state of a hidden Markov process. That is, we assume that identical investors cannot
observe the drift rate, as well as the volatility of course, of the dividend process, but
they have to infer it from the observation of past dividends. We call this model a BlackScholes model with Markov switching, or a hidden Markov model in brief. Related works
in this type have been studied and documented in the literatures, for instance, Detemple
(1991) and David (1997) encompass the regime switching properties of the drift to the
classical Cox-Ingersoll-Ross model, while Veronesi (1999) studies the Gaussian diffusion
model. These papers focus on the issue of modeling stock returns and their empirical
phenomena. They also study investors optimal portfolio allocation and show that equilibrium marketwise excess returns display changing volatility, negative skewness, and
negative correlation with future volatility.
2

For the concern of option pricing in Markov volatility models. Di Masi, Kabanov
and Runggaldier (1994) considers the problem of hedging an European call option for
a diffusion model where drift and volatility are function of a two state Markov jump
process. Guo (2001) also considers the same model and gives a closed-form formula for
the European call option. However, her formula is not appropriate stated and has a
mistake. The contribution of this paper is to give a closed-form formula in the case of
a two state hidden Markov model, and to propose an approximation formula for the
arbitrage-free price of the European call option in general case. The idea of the approximation formula is that we replace the distribution of occupation time for a given state by
its corresponding stationary distribution multiplies the whole time period. Numerical
methods, such as Monte-Carlo simulation and Markovian tree, to compute European
call option price are presented for comparison. Our numerical results show that the
approximation formula provide an efficient and reliable implementation tool for option
pricing.
This article is organized as follows. In Section 2, we first define the Black-Scholes
model with Markov switching that capture the phenomenon of business cycle in stock
fluctuations. Then, we provide a closed-form formula for the price of the European call
option, and in particular, we give an explicit analytic formula for option price when the
underlying Markov process has two states. In Section 3, we propose an approximation
formula for option price in a finite state hidden Markov model. To illustrate the performance of the approximation formula, in Section 4, we presents numerical and simulation
results for comparison. These outcomes are reported in terms of Monte-Carlo simulation
and Markovian tree, respectively. Implied volatility and implied volatility surface are
also given to support our results. Section 5 is conclusions. The derivation of the option
price formula is in the Appendix.

2.

The hidden Markov model and option price

We incorporate the existence of business cycle by modeling the fluctuations of a single


stock price Xt , using an equation of the form
dXt = Xt (t) dt + Xt (t) dWt ,

(1)

where (t) is a stochastic process representing the state of business cycle, Wt is the
standard Wiener process which is independent of (t). For each state of (t), the drift
parameter (t) and volatility parameter (t) are known, and take different values when
(t) is in different state.
We also assume that the supply of the risky asset is fixed and normalized to 1. The
risky free asset is supplied and has an instantaneous rate of return equal to r. Note that
3

this assumption not only simplify the analysis, in particular for option pricing, but also
matches the empirical finding that the volatility of the risk-free rate is much lower than
the volatility of market returns.
Assume that (t) is a Markov process with a few states (maybe two or three states).
In financial economic, we know that business cycle can divide into two different states
called expansion and contraction. A growing economy is described as being in expansion.
In this state, let (t) = 0, (t) = 0 and (t) = 0 . On the other hand, we can take
the value (t) = 1, (t) = 1 and (t) = 1 to represent the state in contraction. More
generally, one can use the state space = {0, 1, , N } for (t) to model more complex
business cycle structures. In this section, for simplicity, we consider a two state hidden
Markov model for a single stock price Xt by using an equation of (1), where (t) is a
Markov process representing the state. Let
(t) =

0, when the business cycle in expansion,


1, when the business cycle in contraction.

Assume 0 6= 1 .
For the concern of the transition rate, let i be the rate of leaving state i, and i the
time of leaving state i. We assume that
P (i > t) = ei t ,

i = 0, 1.

Then the memoryless property of this process is plausible in that, from a practical
standpoint, the information flow be identified more easily otherwise.
It is conceivable that sometimes investors will try to manipulate their buying and
selling in such a way that the existence of such information is not detectable from the
change of volatilities, namely are identical, see Detemple (1991), David (1997), and
Veronesi (1999). The problem of detecting the state change of (t) when remain
unchanged appears to be hard to solve mathematically. It is plausible that change in
business cycle distribution, hence predictability, manifests itself in the diffusion coefficient in the form of both stochastic volatility and drift. If we assume that the are
distinct then it is no loss of generality to assume that (t) is actually observable, since
the local quadratic variation of Xt in any small interval to the left of t will yield (t)
exactly. (For details, see McKean, 1969.) Hence, even if Xt is not Markovian, the joint
distribution (Xt , (t)) is so.
Despite the success of the classical Black-Scholes model, some empirical phenomena have received much attention recently. An important assumption in the BlackScholes model is that the underlying asset distribution is assumed to be Normal, and
the volatility is a fixed constant. However, empirical evidences suggest that it has cluster phenomenon, leptokurtic and unsymmetrical feature. Due to the adjoin of the stock
4

fluctuations with a hidden Markov process (the drift and volatility parameters take different values depending on the state of this hidden Markov process) in the Black-Scholes
exponential Brownian motion model, David (1997) shows that model (1) displays the
asymmetric leptokurtic features, negative skewness, and negative correlation with future
volatility. Veronesi (1999) shows that model (1) is better than the Black-Scholes model
to explain features of stock returns, including volatility clustering, leverage effects, excess volatility and time-varying expected returns. They both show that model (1) can
be embedded into a rational expectation equilibrium framework. For the concern of the
empirical phenomenon on volatility smile, we will present implied volatility and implied
volatility surface in the end of Section 4.
To get the option price formula, we know model (1) is arbitrage-free but incomplete
(cf. Harrison and Pliska (1981), Harrison and Kreps (1979)). One way to treat this
situation can be found in Follmer and Sondermann (1986), and Schweizer (1991), based
on the idea of hedging under a mean-variance criterion. Here, we follow the method by
D. Duffie to complete the market: at each time t, there is a market for a security that
pays one unit of account (say, a dollar) at the next time (t) = inf{u > t|(u) 6= (t)}
that the Markov chain (t) changes state. One can think of this as an insurance contract
that compensates its holder for any losses that occur when the next state change occurs.
Of course, if one wants to hedge a given deterministic loss C at the next state change,
one holds C of the current change-of-state (COS) contracts. For the detail, see Guo
(2001). It can be shown that the absence of arbitrage is effectively the same as the
existence of a probability Q, equivalent to P , under which the price of any derivative is
the expected discounted value of its future cash flow.
The same exercise applied to the underlying risky-asset implies that its price process
X must have the form
dXt
= (r d(t) )dt + (t) dWtQ ,
(2)
Xt
where WtQ is the standard Brownian motion under the risk-neutral probability Q, and
d(t) = r (t) . Note that d(t) is the cost of change of state at the time t. When the
state change at time t, the drift is different from the riskless interest rate r by d0 d1 ,
and the arbitrage opportunity emerges at this moment.
Denote Ti as the total time between 0 and T during which (t) = 0, starting from
state i for i = 0, 1. Let fi (t, T ) be the probability distribution function of Ti .
Theorem 1 Under hidden Markov model (1). Given Equation (2), COS, and riskless
interest rate r, the arbitrage free price of a European call option with expiration date T
and strike price K is
Vi (T, K, r) = E[erT (XT K)+ |(0) = i]
5

= e

rT

T
0

y
(ln(y + K), m(t), v(t))fi (t, T )dtdy,
y+K

(3)

where (x, m(t), v(t)) is the normal density function with expectation m(t) and variance
v(t),
1
1
m(t) = ln(X0 ) + (d1 d0 (02 12 ))t + (r d1 12 )T,
2
2
v(t) = (02 12 )t + 12 T,
and

(x, m(t), v(t)) =

1
q

2v(t)

exp

(x m(t))2

,
2v(t)


f0 (t, T ) = e0 T 0 (T t) + e1 (T t)0 t [0 I0 (2(0 1 t(T t))1/2 )


0 1 t 1/2
+(
) I1 (2(0 1 t(T t))1/2 )],
T t
f1 (t, T ) = e1 T 0 (T t) + e1 (T t)0 t [1 I0 (2(0 1 t(T t))1/2 )
0 1 (T t) 1/2
+(
) I1 (2(0 1 t(T t))1/2 )],
t

(4)

(5)

where I0 and I1 are modified Bessel functions such that

z X
(z/2)2k
Ia (z) = ( )a
.
2 k=0 k!(k + a + 1)!

Remarks: 1. When 0 = 1 , 0 = 1 , m(t) and v(t) are independent of t and


Equation (3) reduces to the classical Black-Scholes formula for European call options.
2. By using the properties of order statistics, Di Masi, Kabanov and Runggaldier
(1994) obtains the distribution of the Kac process, which is a key step for option pricing
in a two state hidden Markov model. In this paper, we apply the method of Laplace
transform to get the distribution. The proof of Theorem 1 will be given in the Appendix.

3.

Option pricing under a finite state hidden Markov


model

In this section, we extend our results to a finite state hidden Markov model. When
the number of states is two, Equation (3) gives a closed-form formula of the European
call option price, and Equations (4) and (5) provide the explicit form to compute the
distributions of the occupation times. Note that the idea of duality to compute fi (t, T )
for i = 0, 1 in a two state hidden Markov process can not be applied to general states,
and the computation of the probability distribution function fi (t, T ) with three or more
6

states becomes complicated and infeasible. However, detailed observation of Equation


(3) reveals that there are two important features in the option price formula. First, it
depends only the occupation time Ti , but not depend on the sample path of visiting each
particular state. Secondly, Equation (3) is valid in no respect to the number of states.
Therefore, instead of direct computation of fi (t, T ) as those in Equations (4) and (5),
we can approximate Ti as follows: Note that Ti /T i in probability as T for
any finite state ergodic Markov process, hence we can give an approximation formula by
simply substituting Ti in Equation (3) by i T . Since the expiration time T for a market
option is half to one year, the approximation should be accurate enough in such time
period.
In the case of a two state hidden Markov model, we also have the following empirical
evidence. Note that in Equations (3) to (5), there are two different states, 0 and 1, and
the option prices depend on the initial state. To the extremely case, we may consider the
situation to which the state is the same in whole time period. That is, the price given by
the classical Black-Scholes models at state 0 (or at state 1), respectively. For example, in
Figure 1, the dash lines represent the option prices given by the classical Black-Scholes
model for each given state, and the real lines represent the prices by Equation (3). Note
that the option prices by Equation (3) are within the two classical Black-Scholes models
and converge to a fixed option price when T is large enough. This empirical phenomenon
also provide the rational of the approximation formula for large enough time period.

 











 



Figure 1: Option pricing with Black-Scholes model and two state hidden Markov model.
The parameter values: X0 = 100, K = 110, 0 = 1 = 10, d0 = d1 = 0, r = 0.1, 0 =
0.2, 1 = 0.3.
Next, we will explain how to compute the stationary distribution as follows: Let
{(t), t 0} be the underlying continuous time Markov process in (1). Denote the
7

transition probability from state i to state j in the time period t as


pt (i, j) = P ((t) = j|(0) = i).
Let

ph (i, j)
h0
h
be the jump rates for i 6= j. We have the stationary distribution by solving the balance
equation Q = 0, where Q is the matrix of transition rates
q(i, j) = lim

Q(i, j) =
where i =

j6=i

q(i, j) j 6= i,
i
j = i,

q(i, j) is the total rate of transitions out of state i.

As the expiration date T tends to infinity, Equations (4) and (5) become
f0 (t, T ) = f1 (t, T ) =

0, when t = T,
1, when t 6= T.

(6)

Therefore the approximation formula of Equation (3) in Theorem 1 can be written as


Vi

(7)
y
= erT
((ln(y + K), m(0 T ), v(0 T ))(1 e0 T 0 (i 0) e1 T 0 (1 i))
y+K
0
+(ln(y + K), m(T ), v(T ))e0 T 0 (i 0)) + (ln(y + K), m(0), v(0))e1 T 0 (1 i))dy.
Z

Figure 2 displays option prices under the same parameters. The dash lines represent
the prices given by the approximation formula (7), and the real lines represent the prices
given by Equation (3) under different initial state. We note that two dash lines are very
close to the two real lines, and they are within the two real lines. Hence they will
converge to the fixed option price faster than the real lines. The reason to describle this
phenomenon is simply that we use the stationary distribution in (7).
By using the same idea as that in (7), we have the approximation formula for European call option in hidden Markov model (t) with finite state space = {0, 1, , N }
as
y
((ln(y + K), m(t0 , t1 , , tN ), v(t0 , t1 , , tN ))
y+K
0
(1 e0 T 0 (0 i) eN T 0 (N i))

Vi = erT

+(ln(y + K), m(T, 0, , 0), v(T, 0, , 0))e0 T 0 (0 i) +

+(ln(y + K), m(0, 0, , T ), v(0, 0, , T ))eN T 0 (N i))dy,

(8)

&'()&+* '-,)./
$
#
!

 



!

" #

Figure 2: Option pricing for a two states hidden Markov model. The parameter values:
X0 = 100, K = 110, 0 = 1 = 1, d0 = d1 = 0, r = 0.1, 0 = 0.2, 1 = 0.3.
where ti = i T , i = 0, , N and
m(t0 , t1 , , tN ) = ln(X0 ) +
v(t0 , t1 , , tN ) =

N
X

N
X

1
(r di i2 )ti ,
2
i=0

i2 ti .

i=0

In practice, a three state hidden Markov chain is good enough to capture the empirical phenomena of most financial data series, and the computational method is the same
for each finite state Markov process. Hence, we only consider the case of three state
hidden Markov model in our simulation study. Figure 3 displays option prices under the
same parameters. Here, the dash lines represent the prices given by the approximation
formula, and the real lines represent the prices given by Black-Scholes formula under different initial state. In principle, the number of states exceed three can be implemented
in the same matter, but to discriminate and find out the matrix of transition rates Q in
some time interval might be computational demanding.

9:;=<=9>

:@?-<=ACB

73
7-0
6
5
4
3
0213

0214

0215

0216

Figure 3: Option pricing for a three states hidden Markov model. The parameter values:
X0 = 100, K = 110, 0 = 1 = 2 = 10, d0 = d1 = d2 = 0, r = 0.1, 0 = 0.1, 1 =
0.2, 2 = 0.3.

4.

Numerical performance

4.1 Design of the simulation


Numerical performance of the approximation formula (8) for option pricing is presented
in this section. We compare European call option prices obtained by Monte-Carlo simulation, Markovian tree method and the approximation formula. Two state and three
state hidden Markov models are considered in the simulation study, respectively. The
performance of this comparison is based on the difference of the option prices. The
definition of difference will be given later. We use the exact analytic formula as the
benchmark in the case of a two state hidden Markov model, and the Monte-Carlo simulation result as the benchmark in the case of a three state hidden Markov model. The
results are shown in Tables 1 to 10, in which four factors are used:
(1) Low and high volatilities.
(2) Different transition rates in the case of a three state hidden Markov model.
(3) Different strike prices. Specifically, we consider the cases of in the money, at the
money and out of the money.
(4) Different expiration date; T = 0.1, 0.2, 0.5, 1, 2, 3, with unit of year.
Since model (1) is a continuous time model, we discretize this continuous time model
in the vein of Cox, Ross, and Rubinstein (1979). A corresponding two state hidden
Markov model based on Cox, Ross and Rubinstein (1979) can be found in David (1997).
It is worth pointing out that this methodology applies also to the general case where
the hidden Markov process (t) takes more than three states, i.e., the state space can

10

be = {0, 1, , N }, when, for example, more complex information patterns can be


imposed.
We can apply Equation (2) for Monte-Carlo simulation of the option price. First,
rewrite Equation (2) in the discrete time form
Xn+h = Xn e

(r

1 2
n dn )h + n h
2
,

(9)

where is simulated from the standard normal random variable N (0, 1). Second, consider the random variable n with transition probability of changing state by (i,j +
(1)i,j eq(i,i)h )(|q(i, j)/i |). Repeat the steps for M times, where M is large enough to
guarantee the convergence of the simulation.
To illustrate the idea of discretization in Markovian tree method, we first construct
the structure of a two state hidden Markov model as follows: Suppose the time interval
[0, t] is divided into n sub-intervals such that t = nh. Let X = (Xk , k 0) and Xk is a
price at time kh. Define
Xkk = (Xk , k ) = (X(kh), (kh)).
Let ni,j be independent and identically distributed (i.i.d.) random variables, taking
values uj with probability pj (i,1j + (1)i,1j ei h ), and 1/uj with probability (1
pj )(i,1j + (1)i,1j ei h ), i, j = 0, 1, respectively, where

i h + i h 0.5i2 h
i h

ui = e
.
, pi =
2i h
Then, we have the following recurrence relation,
(Xn , n ) = n(n),(n1) (Xn1 , n1 ).

(10)

By the memoryless property of i , (Xnn , n 0) is a Markov chain. The Markov chain Xnn
with initial
state X0 = x is a random walk on the set Ex = {xur |r = 0 n0 +1 n1 , n0 , n1

Z, u = e h }.
In general, we can apply the same idea to the hidden Markov model with state
space = {0, 1, , N }. Denote Q as the matrix of transition rates. Note that the
ni,j in Equation (9) are independent and identically distribute random variables, taking
values uj with probability pj (i,j +(1)i,j eq(i,i)h )(|q(i, j)/i |), and 1/uj with probability
(1 pj )(i,j + (1)i,j eq(i,i)h )(|q(i, j)/i |), where i represents the current state and j
represents the oncoming state in one sub-interval. Similarly, the Markov chain Xnn with
initial state X0 = x is a random
walk on the set Ex = {xur |r = 0 n0 + 1 n1 + +

N nN , n0 , n1 , , nN Z, u = e h }.
11

Here we consider n, the number of sub-interval, to be 30, since our simulation


shows that this number is large enough to provide accurate results. The Monte-Carlo
replication size for both Monte-Carlo simulation and Markovian tree method is B =
50, 000, 000. Computations were performed using Visual Basic programs on the personal
computer system with a Pentium 4 CPU, 1.6G and 256MB of RAM, of the Institute
of Statistical Science, Academia Sinica, Taipei, Taiwan, R.O.C.. The pseudo-random
numbers were generated by using IMSL routines. All tests were compared on the basis of
the same random numbers, samples of different size were nested. The reported running
time is given the CPU time in seconds.
4.2 Simulation results
We will use the following abbreviations in Tables 1 to 10. B-S i refers to the classical Black-Scholes formula with state i = 0, 1; Vi refers to the exact price of a two
state hidden Markov model; Vi refers to the approximation price of a two states hidden
Markov model; sim i and tree i refer to Monte-Carlo simulation for Equation (2) and
Markovian tree method, respectively.
We first consider the case of a two state hidden Markov model, with the same
transition rates. Tables 1 and 2 report the numerical and simulation results according
to low and high volatilities. The other parameters are the same. Note from Tables 1 and
2 that the numerical values given by the Monte-Carlo simulation and the Markov tree
method are very close to the same values for various time period. In the case of short
time period, the performance of the approximation formula is better than Monte-Carlo
simulation and Markovian tree method, in the sense to compare the results obtained
by the exact formula. And the computational time is reduced significantly. Therefore,
the approximation formula (7) gives a rather promising result in the case of a two state
hidden Markov model.
Table 1: The parameters are: X0 = 100, K = 90, 0 = 1 = 1, d0 = d1 = 0, r =
0.1, 0 = 0.2, 1 = 0.3, n = 30.
T (year)
0.1
0.2
0.5
1.0
2.0
3.0
running
time(sec)

B-S 0
10.975
12.088
15.288
19.988
28.037
34.996

B-S 1
11.381
12.968
17.034
22.510
31.281
38.504

V0
10.993
12.164
15.614
20.721
29.287
36.476

V1
11.360
12.889
16.718
21.812
30.085
37.061

< 105

< 105

1.51

1.45

12

tree 1
11.314
12.698
16.960
21.747
30.060
36.877

V0
10.993
12.166
15.639
20.811
29.478
36.689

V1
11.360
12.886
16.698
21.739
29.917
36.863

240.55 251.32 140.24 144.23

0.061

0.063

sim 0
10.986
12.151
15.592
20.681
29.256
36.426

sim 1
11.346
12.883
16.719
21.796
30.051
37.057

tree 0
10.910
12.257
15.497
20.740
29.166
36.219

Table 2: The parameters are: X0 = 100, K = 90, 0 = 1 = 1, d0 = d1 = 0, r =


0.1, 0 = 0.1, 1 = 1.0, n = 30.
T (year)
0.1
0.2
0.5
1.0
2.0
3.0
running
time(sec)

B-S 0
17.007
21.612
31.027
41.547
55.470
64.952

B-S 1
18.100
23.180
33.429
44.688
59.236
68.853

V0
17.059
21.752
31.478
42.460
56.934
66.638

V1
18.050
23.045
32.996
43.818
57.858
67.286

< 105

< 105

1.44

1.52

tree 1
18.050
23.135
32.995
44.106
57.616
66.550

V0
17.060
21.757
31.513
42.570
57.158
66.884

V1
18.050
23.042
32.969
43.726
57.668
67.078

249.46 252.55 100.23 101.17

0.060

0.063

sim 0
17.048
21.731
31.468
42.455
56.945
66.671

sim 1
18.045
23.050
32.994
43.760
57.871
67.415

tree 0
16.842
21.860
31.765
42.662
56.817
66.080

Let Q1 be the matrix of transition rate

1 1
0

Q1 = 1/2 1 1/2 ,
0
1
1

and let Q2 be the matrix of transition rate

1 1
0

Q2 = 5/2 5 5/2 .
0
10 10
The results of a three state hidden Markov model are recorded in Tables 3 to 6. Note
that the matrix of transition rates Q is the same for a two state hidden Markov model;
while the matrix of transition rates Q may be different in a three or more state hidden
Markov models. Since the difference of the matrix of transition rates in a three or more
states hidden Markov model may also play an important role for option pricing, we will
present our simulation results accordingly. Tables 3 and 4 displays the option prices for
a three states hidden Markov model, with matrix of transition rate Q1 (with the same
transition rate), the same parameters, but different volatilities. Tables 5 and 6 displays
the option prices for a three states hidden Markov model, with matrix of transition rate
Q2 (with different transition rate), the same parameters, but different volatilities.
Note from Tables 3 to 6 that we have three numerical consequences. First, the
option prices obtained by the approximation formula (8) are less sensitive to the initial
state than those obtained by Monte-Carlo simulation and Markovian tree method. The
rational behind this phenomenon is that we use the stationary distribution , instead of
exactly probability distribution, in the approximation formula (8). Secondly, the option
prices in Tables 5 and 6 with bigger i values, i.e. 1 and 2 , are closer to the values given
by the Monte-Carlo simulations than the price with 0 . This result is reasonable, since
13

Table 3: The parameters are: X0 = 100, K = 90, 0 = 1 = 2 = 1, d0 = d1 = d2 =


0, r = 0.1, 0 = 0.1, 1 = 0.2, 2 = 0.3, n = 30.
T (year)
0.1
0.2
0.5
1.0
2.0
3.0
running
time(sec)

tree 2
11.309
12.694
17.050
21.744
29.776
36.386

V0
10.906
11.854
14.833
19.656
28.103
35.250

V1
10.978
12.103
15.358
20.155
28.327
35.331

V2
11.345
12.823
16.417
21.083
28.766
35.506

269.25 272.84 276.63 171.25 175.22 178.32

0.061

0.062

0.061

sim 0
10.898
11.806
14.572
19.061
27.201
34.333

sim 1
10.973
12.078
15.282
19.978
28.028
34.953

sim 2
11.346
12.886
16.721
21.761
29.736
36.426

tree 0
10.904
11.840
14.639
18.987
27.067
34.057

tree 1
10.988
12.199
15.084
20.167
27.409
34.667

Table 4: The parameters are: X0 = 100, K = 90, 0 = 1 = 2 = 1, d0 = d1 = d2 =


0, r = 0.1, 0 = 0.8, 1 = 0.9, 2 = 1.0, n = 30.
T (year)
0.1
0.2
0.5
1.0
2.0
3.0
running
time(sec)

tree 2
18.025
23.145
32.939
43.887
57.049
65.646

V0
16.036
20.344
29.590
40.427
55.033
64.851

V1
17.010
21.620
31.053
41.602
55.563
65.058

V2
17.999
22.904
32.510
42.758
56.072
65.252

267.11 282.29 288.13 164.54 170.91 178.30

0.062

0.063

0.062

sim 0
15.984
20.170
29.118
39.464
53.760
63.782

sim 1
16.993
21.584
31.025
41.555
55.456
64.822

sim 2
18.021
23.062
32.906
43.674
57.202
66.494

tree 0
15.683
20.105
29.027
39.726
53.621
63.304

tree 1
16.797
21.725
31.345
42.048
56.074
64.395

the one with bigger i means that the Markov process will leave state i in a short period
of time, and hence the effect of such initial state is smaller. Thirdly, the option prices
in a longer time period, for instance a 3-year contract in Tables 3 to 6, have different
outcome compared to the results in a smaller time period. Note that the option price
converges to the true value by the i s order, and the prices with two larger values will
emerge faster than the third one. Therefore, the simulation region for option prices can
not cover the whole approximation prices. In other words, the option prices with many
initial states will merge to a fewer one in a reasonable time period. Hence, one does
not need too many states for modeling stock returns from the perspective of computing
option prices, and it is reasonable to replace a finite state hidden Markov model by a
two or three state hidden Markov model.
For the concern of the effect of the time to maturity T , in Tables 7 to 10, we provide

14

Table 5: The parameters are: X0 = 100, K = 90, 0 = 1, 1 = 5, 2 = 10, d0 = d1 =


d2 = 0, r = 0.1, 0 = 0.1, 1 = 0.2, 2 = 0.3, n = 30.
T (year)
0.1
0.2
0.5
1.0
2.0
3.0
running
time(sec)

tree 2
11.191
12.473
15.632
19.873
27.542
34.245

V0
10.896
11.799
14.541
18.949
26.891
33.884

V1
10.949
11.947
14.771
19.139
26.971
33.910

V2
11.082
12.015
14.741
19.134
26.971
33.910

263.01 264.90 263.45 172.78 171.44 174.20

0.059

0.062

0.062

sim 0
10.901
10.768
14.740
18.862
26.840
33.745

sim 1
10.937
12.083
14.798
19.171
27.131
34.059

sim 2
11.175
12.669
15.261
19.389
27.281
34.168

tree 0
10.906
11.845
14.636
18.904
26.777
33.692

tree 1
10.925
12.178
15.187
19.508
27.260
34.047

Table 6: The parameters are: X0 = 100, K = 90, 0 = 1, 1 = 5, 2 = 10, d0 = d1 =


d2 = 0, r = 0.1, 0 = 0.8, 1 = 0.9, 2 = 1.0, n = 30.
T (year)
0.1
0.2
0.5
1.0
2.0
3.0
running
time(sec)

tree 2
17.653
22.366
31.353
41.218
54.170
63.142

V0
15.974
20.172
29.016
39.209
53.000
62.498

V1
16.761
21.042
29.748
39.719
53.226
62.586

V2
17.014
21.044
29.659
39.707
53.226
62.586

261.34 265.69 267.45 171.34 182.22 183.20

0.061

0.061

0.062

sim 0
15.979
20.186
29.109
39.518
53.729
63.525

sim 1
16.993
21.564
31.008
41.553
55.569
64.918

sim 2
17.671
22.231
31.278
41.434
55.113
64.695

tree 0
15.689
20.111
28.953
39.541
52.986
62.244

tree 1
16.790
21.599
30.714
40.607
53.893
62.905

approximate European call option prices for various T in a hidden Markov model with
three states. Tables 7 and 9 reports the case of low volatilities, and Tables 8 and 10
reports the case of high volatilities. The matrix of transition rates in Tables 7 and 8 is
Q2 , and the matrix of transition rate in Tables 9 and 10 is Q3 , which is defined as

1 1
0

5 5
Q3 = 0
.
10 0
10
For each strike price K, we consider three different strike-to-stock price ratios K/X 0 .
They are 1.1, 1.0 and 0.9. Note that in Tables 7 to 10, the first row in each panel is
the approximate analytical prices with different initial state, whereas the second and
forth rows report Monte-Carlo tree prices and their standard deviation, respectively.
The number in the third row is error estimate between the approximation value and
15

Monte-Carlo simulation. Specifically, in Tables 7 to 10, Vi refers the approximate price


for a three state hidden Markov model with state i = 0, 1, 2.; tree i refers to Monte-Carlo
method for tree with Q2 ; difference is the ratio between approximate price and tree price,
i.e. difference:= (Vi tree i)/tree i; and Std is the standard deviation of Monte-Carlo
tree prices.
Note that the option prices Vi for i = 0, 1, 2 in Table 7 is higher than the corresponding option prices in Table 9. The reason can be described as follows: Although
the parameters in both tables are the same, the matrix of transition rates are different.
We use Q2 in Table 7; while Q3 in Table 9. By the definition of Q2 and Q3 , it is easy to
see that the occupation time of state 0 in Q3 is longer than that of Q2 , and state 0 is
the worst state. By using the same argument, we have the option prices Vi for i = 0, 1, 2
in Table 8 is higher than the corresponding option prices in Table 10. Note from Tables
1 to 10, the convergence rate of option price in approximation formula depends on the
value of and difference of prices from initial state.
Furthermore, we observe that the option price obtained by the approximation formula (8) depends on the strike price K. It is more accurate in the cases of in the money
and at the money; while it is less accurate in the case of out of the money for each time
period. A heuristic argument based on Equation (3) can be described as follows:
We consider the case of a two state hidden Markov model for simplicity. Let K(1)
and K(2) be two given strike prices with K(1) < K(2). Without loss of generality, we
may assume 1 > 0 ; that is the volatility at state 1 is bigger than the volatility at state
0. Then from Equation (3), we have V1 (T, K, r) > V0 (T, K, r). It is also known from
Equation (3) again that Vi (T, K(1), r) > Vi (T, K(2), r), for i = 0, 1. Under the condition
V1 (T, K(1), r) V0 (T, K(1), r)
V1 (T, K(2), r) V0 (T, K(2), r)
<
,
V1 (T, K(1), r)
V1 (T, K(2), r)

(11)

which is correct for T is not too large. Note also that |Vi (T, K(1), r) Vi (T, K(1), r)|
V1 (T, K(1), r) V0 (T, K(1), r), and |Vi (T, K(2), r) Vi (T, K(2), r)| V1 (T, K(2), r)
V0 (T, K(2), r) for i = 0, 1. Therefore,
|V1 (T, K(1), r) V1 (T, K(1), r)|
|V1 (T, K(2), r) V1 (T, K(2), r)|
<
.
V1 (T, K(1), r)
V1 (T, K(2), r)

(12)

Note that (12) is the definition of difference in Tables 7 to 10, and it is an increasing
function of K.
It is easy to see that under the condition (11), we also have
|V0 (T, K(1), r) V0 (T, K(1), r)|
|V0 (T, K(2), r) V0 (T, K(2), r)|
<
.
V0 (T, K(1), r)
V0 (T, K(2), r)
16

(13)

17

0.1948 2.8094 10.949 0.6870 4.0995 11.947 2.5753 7.1804 14.771 6.3232 11.763 19.139
0.2884 3.0140 10.925 1.0173 4.5460 12.178 3.3714 7.9582 15.187 7.1656 12.484 19.508
-0.325 -0.068 0.0022 -0.325 -0.098 -0.019 -0.236 -0.098 -0.027 -0.118 -0.058 -0.019
0.0006 0.0018 0.0028 0.0013 0.0027 0.0036 0.0031 0.0044 0.0054 0.0053 0.0065 0.0074
0.4155 3.1278 11.082 0.7567 4.1465 12.015 2.4981 7.1080 14.741 6.3108 11.753 19.134
0.7362 3.8411 11.191 1.7075 5.4005 12.473 4.2281 8.7693 15.632 8.0111 13.153 19.873
-0.436 -0.186 -0.010 -0.557 -0.232 -0.037 -0.409 -0.189 -0.057 -0.212 -0.106 -0.037
0.0011 0.0025 0.0035 0.0020 0.0034 0.0045 0.0038 0.0051 0.0061 0.0059 0.0071 0.0081

V1
tree 1
difference
Std

V2
tree 2
difference
Std

1.1
1.6640
1.6462
0.0108
0.0018

K/X0
V0
tree 0
difference
Std

T=0.5
T=1.0
1.0
0.9
1.1
1.0
0.9
6.3350 14.541 5.5403 11.221 18.949
6.3186 14.636 5.4421 11.159 18.904
0.0026 -0.006 0.0180 0.0056 0.0024
0.0031 0.0037 0.0040 0.0051 0.0058

T=0.1
T=0.2
1.1
1.0
0.9
1.1
1.0
0.9
0.0106 1.8759 10.896 0.1600 3.0882 11.799
0.0128 1.8829 10.906 0.1641 3.0783 11.845
-0.172 -0.004 -0.001 -0.025 0.0032 -0.004
0.0001 0.0010 0.0016 0.0004 0.0016 0.0022

Table 7: The parameters are: X0 = 100, 0 = 1, 1 = 5, 2 = 10, d0 = d1 = d2 = 0, r = 0.1, 0 = 0.1, 1 =


0.2, 2 = 0.3, n = 30.

18

7.6075 11.481 16.761 12.295 16.168 21.042 21.834 25.481 29.748 32.804 36.057 39.719
7.7310 11.678 16.790 12.898 16.617 21.599 22.944 26.279 30.714 33.856 36.866 40.607
-0.016 -0.016 -0.001 -0.046 -0.027 -0.025 -0.048 -0.030 -0.031 -0.031 -0.021 -0.021
0.0075 0.0089 0.0103 0.0122 0.0136 0.0148 0.0233 0.0246 0.0258 0.0397 0.0408 0.0418
7.8960 11.766 17.014 12.297 16.169 21.044 21.733 25.385 29.659 32.790 36.044 39.707
8.6757 12.494 17.653 13.824 17.506 22.366 23.687 27.069 31.353 34.632 37.508 41.218
-0.089 -0.058 -0.036 -0.110 -0.076 -0.059 -0.082 -0.062 -0.054 -0.053 -0.039 -0.036
0.0083 0.0097 0.0110 0.0131 0.0145 0.0157 0.0244 0.0257 0.0269 0.0411 0.0421 0.0432

V2
tree 2
difference
Std

T=1.0
0.9
1.1
1.0
0.9
29.016 32.222 35.506 39.209
28.953 32.467 35.419 39.541
0.0022 -0.007 0.0025 -0.008
0.0235 0.0366 0.0377 0.0388

V1
tree 1
difference
Std

T=0.5
0.9
1.1
1.0
20.172 20.998 24.684
20.111 21.188 24.403
0.0030 -0.009 0.0115
0.0134 0.0209 0.0222

1.1
6.6994
6.6165
0.0125
0.0064

T=0.2
0.9
1.1
1.0
15.974 11.300 15.201
15.689 11.408 15.064
0.0182 -0.009 0.0091
0.0094 0.0106 0.0121

K/X0
V0
tree 0
difference
Std

T=0.1
1.0
10.570
10.486
0.0080
0.0079

Table 8: The parameters are: X0 = 100, 0 = 1, 1 = 5, 2 = 10, d0 = d1 = d2 = 0, r = 0.1, 0 = 0.8, 1 =


0.9, 2 = 1.0, n = 30.

19

T=0.1
T=0.2
T=0.5
T=1.0
1.1
1.0
0.9
1.1
1.0
0.9
1.1
1.0
0.9
1.1
1.0
0.9
0.0085 1.8658 10.896 0.1448 3.0618 11.795 1.5748 6.2516 14.511 5.3237 11.056 18.873
0.0148 1.8894 10.904 0.1805 3.1023 11.849 1.7180 6.3803 14.663 5.5045 11.196 18.909
-0.426 -0.012 -0.001 -0.198 -0.013 -0.005 -0.083 -0.020 -0.010 -0.033 -0.013 -0.002
0.0001 0.0010 0.0016 0.0004 0.0016 0.0022 0.0018 0.0031 0.0038 0.0040 0.0051 0.0058
0.1863 2.7678 10.947 0.6339 4.0074 11.932 2.3671 6.9859 14.702 5.9828 11.504 19.021
0.3497 3.1817 10.959 1.1985 4.7978 12.249 3.5432 8.1102 15.230 7.1038 12.423 19.418
-0.467 -0.130 -0.001 -0.471 -0.165 -0.026 -0.332 -0.139 -0.035 -0.158 -0.074 -0.020
0.0006 0.0020 0.0029 0.0015 0.0029 0.0039 0.0032 0.0045 0.0055 0.0051 0.0063 0.0073
0.4017 3.0609 11.079 0.6841 4.0205 11.994 2.2729 6.8976 14.666 5.9682 11.492 19.015
0.5988 3.5314 11.127 1.2190 4.7581 12.223 3.0319 7.6072 15.118 6.6177 11.989 19.248
-0.329 -0.133 -0.004 -0.439 -0.155 -0.019 -0.250 -0.093 -0.030 -0.098 -0.041 -0.012
0.0010 0.0023 0.0033 0.0016 0.0030 0.0039 0.0029 0.0042 0.0051 0.0048 0.0060 0.0068

K/X0
V0
tree 0
difference
Std

V1
tree 1
difference
Std

V2
tree 2
difference
Std

Table 9: The parameters are: X0 = 100, 0 = 1, 1 = 5, 2 = 10, d0 = d1 = d2 = 0, r = 0.1, 0 = 0.1, 1 =


0.2, 2 = 0.3, n = 30.

20

7.5624 11.436 16.722 12.191 16.067 20.951 21.600 25.259 29.543 32.469 35.740 39.425
7.8958 11.853 16.950 13.157 16.858 21.784 22.882 26.405 30.697 33.650 36.707 40.399
-0.042 -0.035 -0.013 -0.073 -0.046 -0.038 -0.056 -0.043 -0.037 -0.035 -0.026 -0.024
0.0077 0.0091 0.0104 0.0124 0.0138 0.0150 0.0233 0.0247 0.0258 0.0392 0.0402 0.0412
7.8236 11.693 16.952 12.154 16.030 20.920 21.480 25.144 29.438 32.453 35.725 39.411
8.2894 12.141 17.328 12.918 16.743 21.595 22.257 25.819 30.159 33.105 36.277 40.032
-0.056 -0.036 -0.021 -0.059 -0.042 -0.031 -0.034 -0.026 -0.023 -0.019 -0.015 -0.015
0.0079 0.0093 0.0107 0.0123 0.0137 0.0149 0.0225 0.0237 0.0249 0.0380 0.0392 0.0402

V1
tree 1
difference
Std

V2
tree 2
difference
Std

T=0.5
T=1.0
0.9
1.1
1.0
0.9
1.1
1.0
0.9
20.146 20.897 24.589 28.928 32.009 35.304 39.022
20.127 21.189 24.493 29.009 32.466 35.392 39.459
0.0009 -0.013 0.0039 -0.002 -0.014 -0.002 -0.011
0.0134 0.0210 0.0223 0.0236 0.0365 0.0377 0.0387

1.1
6.6885
6.6217
0.0101
0.0064

T=0.2
0.9
1.1
1.0
15.964 11.270 15.172
15.683 11.428 15.091
0.0179 -0.013 0.0054
0.0094 0.0106 0.0121

K/X0
V0
tree 0
difference
Std

T=0.1
1.0
10.559
10.487
0.0069
0.0079

Table 10: The parameters are: X0 = 100, 0 = 1, 1 = 5, 2 = 10, d0 = d1 = d2 = 0, r = 0.1, 0 = 0.8, 1 =


0.9, 2 = 1.0, n = 30.

From the results appeared in Tables 3 to 10, we note that the performance of analytic approximation is very promising, compare to the examined numerically outcomes
obtained by Monte-Carlo simulation and Markovian tree method. The advantage of the
analytical approximation is that it accelerates the computation time of the option prices
in hidden Markov models.
4.3 Volatility smiles and surfaces
If the Black-Scholes model is correct, then the implied volatility should be constant.
But it is widely recognized that the volatility has a smile feature. For instance, Figure
4 illustrates our notion of volatility smile that is presented in the Microsoft call option
with the Black-Scholes model and the hidden Markov model. Model (1) can reveal the
phenomena of volatility smile as shown in Figure 4.
We can show implied volatility against both maturity and strike in a three-dimensional
plot. That is we consider (X, t) as a function of X and t. One is shown in Figure 5 that
IBM call option with Black-Scholes model and hidden Markov model from 2002/3/15
with five expiry of nine months. There are call option traded with an expiry of five
months and strikes of 100, 105, 110, 115, 120, 125, 130, 135 and 140. This implied
surface represents the constant value of volatility that gives each traded option a theoretical value equal to the market value. We can see how the time dependence in implied
volatility could be turned into a volatility of the underlying that was time dependent.
In the case of (X, t) can be deduced from volatility surface at a specific time t , we
might call it the local volatility surface. This local volatility surface can be thought of
as the markets view of the future value of volatility when the asset price is X at time t.
We should emphasize that the examples pressed in Figures 4 and 5 are not an
empirical test of the model (1), it is only an illustration to show that the model can
produce a close fit to the empirical phenomenon.
D=E FGF HI@JLKNM OGPGQ@RSP FGHTE RVUT E HFVQ PGT KNW E T E W X

no pSqSr sGtuv qVwu v tsVx rGu yNz v u v z {

YZ \
fef ll m
e kj
hi
ef gd
bc

YZ ]\

YZ ]

YZ [G\
YZ [

]\

\G\

^G\
a

_\

` \

|} 
|} G
|} 
|}
|} ~
|} ~G
|} ~
|} ~ 
|} ~

Figure 4: Implied volatility of Black-Scholes model and two-state hidden Markov model

21


 C+
" =  
 =
-
 = =
" ==

Figure 5: Implied volatilities against expiry and strike price. The parameter values:
X0 = 107, 0 = 1 = 1, d0 = d1 = 0, r = 0.0261.

5.

Conclusions

A closed form formula for the Black-Scholes with Markov switching option pricing model
has been developed in this paper. In the case of a two state hidden Markov model, we
have an exact analytic formula; while in the case of a three or more state hidden Markov
model, we have an approximation formula. Numerical evaluation of the formula is
studied for both two state and three state cases. The performance of the approximation
was examined numerically using option prices obtained from Monte-Carlo simulation
and Markovian tree method. The pricing error, in terms of ratio, by the analytical
approximation is shown to be small except in extreme cases (out of the money).
An exact closed formula provides useful insight for the European option pricing in
the Black-Scholes model with Markov switching. It not only explains the effect of regime
switching for option pricing, but also gives the idea of an analytical approximation, in
which it accelerates the computation of the European option pricing in the Black-Scholes
model with Markov switching. It has a number of further applications, for instance, it
can be used to compute hedge ratios and implied hidden Markov models parameters,
i.e. calibrate the parameters by using implied volatility surface. The approximation can
facilitate empirical studies of the index options that are, in many cases, of European
style. This feature is worthy of further exploration and may have many implications in
other applications.

22

References
Anderson, T. G. (1996). Return volatility and trading volume: an information
flow interpretation of stochastic volatility. Journal of Finance, 51, 169-204.
Avellaneda, M., Levy, P., and Par
as, A. (1995). Pricing and hedging derivative
securities in markets with uncertain volatilities. Applied Mathematical Finance, 2,
73-88.
Bittlingmayer, G. (1998). Output, stock volatility and political uncertainty in a
natural experiment: Germany, 1880-1940. Journal of Finance, 53, 2243-2257.
Black, F. and Scholes, M. (1973). The pricing of options and corporate liabilities.
Journal of Political Economy, 81, 637-654.
Cox, J., Ross, S., and Rubinstein, M. (1979). Option pricing, a simplified approach.
Journal of Financial Economics, 7, 229-263.
David, A. (1997). Fluctuating confidence in stock markets: Implications for returns and volatility. Journal of Financial and Quantitative Analysis, 32, 427-462.
Detemple, J. B. (1991). Further results on asset pricing with incomplete information. Journal of Economic Dynamics and Control, 15, 425-453.
Di Masi, G. B., Kabanov, Yu, M., and Runggaldier, W. J. (1994). Mean-variance
hedging of options on stocks with Markov volatility. Theory of Probability and Its
Applications, 39, 173-181.
Diebold, F. X., and Inoue, A. (2001). Long memory and regime switching. Journal
of Econometrics, 105, 131-159.
Duffie, D., and Huang, C. F. (1986). Multiperiod security markets with differential
information. Journal of Mathematical Economics, 15, 283-303.
Follmer, H. and Sondermann, D. (1986). Hedging of nonredundant contingent
claims. In: Hildenbrand and Mas-Colell, eds., Contributions to Mathematical
Economics, 205-223.
Grorud, A., and Pontier, M., (1998). Insider trading in a continuous time market
model. International Journal of Theoretical and Applied Finance, 1, 331-347.
Guilaume, D. M., Dacorogna, M., Dav
e, R., Muller, U., Olsen, R., and Pictet, P.
(1997). From the birds eye to the microscope, a survey of stylized facts of the
intra-daily foreign exchange market. Finance and Stochastics, 1, 95-129.
Guo, X. (2001). Information and option pricing. Journal of Quantitative Finance,
1, 38-44.
23

Hamilton, J. D. (1988). Rational-Expectations econometric analysis of changes in


regime: an investigation of term structure of interest rates. Journal of Economic
Dynamics and Control, 12, 385-423.
Hamilton, J. D. (1989). A new approach to the economic analysis of nonstationary
time series and the business cycle. Econometrica, 57, 357-384.
Hamilton, J. D., and Susmel, R. (1994). Autoregressive conditional heteroscedasticity and changes in regime. Journal of Econometrics, 64, 307-333.
Heston, S. L. (1993). A closed-form solution for options with stochastic volatility
with applications to bond and currency options. Review of Financial Studies, 6,
327-343.
Hull, J. and White, A. (1987). The pricing of options on assets with stochastic
volatility. Journal of Finance, 2, 281-300.
Karatzas, I., and Pikovsky, I. (1996). Anticipative portfolio optimization. Advances in Applied Probability, 28, 1095-1122.
McKean, H. P. (1969). Stochastic Integrals. Academic Press, New York.
Ross, S. A. (1989). Information and volatility, the no-arbitrage martingale approach to timing and resolution irrelevancy. Journal of Finance, 44, 1-8.
Schweizer, M. (1991). Option hedging for semimartingales. Stochastic Processes
and their Applications, 37, 339-363.
So, M. K. P., Lam, K., and Li, W. K. (1998). A stochastic volatility model with
Markov switching. Journal of Business & Economic Statistics, 16, 244-253.
Stein, E. M., and Stein, C. J. (1991). Stock prices distribution with stochastic
volatility, an analytic approach. Review of Financial Studies, 4, 727-752.
Turner, C., Startz, R., and Nelson, C. (1989). A Markov model of heteroscendasticity, risk, and learning in the stock market. Journal of Financial Economics, 25,
3-22.
Veronesi, P. (1999). Stock market overreaction to bad news in good times: a
rational expectations equilibrium model. Review of Financial Studies, 12, 5, 9751007.
Wiggins, J. B. (1987). Option values under stochastic volatility. Theory and
empirical evidence. Journal of Financial Economics, 19, 351-372.

Appendix
24

Proof of Theorem 1.
Since the arbitrage price of the European option is the discounted expected value of
Xt under the equivalent martingale measure Q, we have
Vi (T, K, r) = EQ [erT (XT K)+ |(0) = i].
Recalling that
Xt = X0 exp(

t
0

1 2
(r d(s) (s)
)ds +
2

t
0

(s) dWsQ ),

the key point is to calculate the instantaneous distribution of Xt . Let Yt = ln Xt , then


Yt = Y 0 +

t
0

1 2
(r d(s) (s)
)ds +
2

t
0

(s) dWsQ .

Denote Ti as the total time between 0 and T during which (t) = 0, starting from
state i, and we consider the probability distribution function fi (t, T ), which is defined
as the probability distribution function of Ti .
Vi (T, K, r) = E[erT (XT K)+ |(0) = i]
= erT E[E[(XT K)+ |Ti ]|(0) = i]
= erT Ei [E[(XT K)+ |Ti ]|F (0) = i]
Z Z T
y
= erT
(ln(y + K), m(t), v(t))fi (t, T )dtdy,
0
0 y+K
where (x, m(t), v(t)) is the normal density function with expectation m(t) and variance
v(t),
1
1
m(t) = ln(X(0)) + (d1 d0 (02 12 ))t + (r d1 12 )T,
2
2
v(t) = (02 12 )t + 12 T,
and

Then

(x, m(t), v(t)) = q

1
2v(t)

fi (t, T )dt = P (

exp(

(x m(t))2
).
2v(t)

0 (s )ds dt),

where 0 is the indicate function at state 0.


Let

RT

r
0 (s )ds
0
i (r, T ) = E[e
|(0) = i]
Z
=
ert fi (t, T )dt
0

:= Lr (fi (, T )),

then we have

i (r, T ) = erT ei T 0 (i 0) + ei T 0 (1 i)
+

ei u i 1i (T u)eru0 (i0) du.


25

That is,
0 (r, T ) = erT e0 T +
1 (r, T ) = e

1 T

T
0

e0 u 0 1 (T u)eru du,

0
1 u

1 0 (T u)du.

Taking Laplace transforms on both sides, and writing


Ls (i (r, )) = L
[Lr (fi (, T ))(r, )]
Z s
esT i (r, T )dT
=
0
:= i (r, s),
then

1
0
1 (r, s),
+
r + s + 0 r + s + 0
1
1
0 (r, s).
1 (r, s) =
+
s + 1 s + 1
Solving these equations, we get
0 (r, s) =

s + 0 + 1
,
+ s1 + s0 + rs + r1
r + s + 0 + 1
.
1 (r, s) = 2
s + s1 + s0 + rs + r1
0 (r, s) =

s2

Taking r the inverse Laplace transform with represent w


s + 0 + 1

s + 1 s + 0 + 1 ,
L1
r (0 (r, s))(w, ) = e
s + 1
s + 0 + 1
sw

s + 1 1 (s + 0 + 1 ) + 0 (w) .
L1
r (1 (r, s))(w, ) = e
(s + 1 )2
s + 1
sw

Again, tacking s the inverse Laplace transform with represent v


1
0 v
L1
0 (v w) + e1 (vw)0 w [0 I0 (2(0 1 w(v w))1/2 )
s [Lr (0 (r, s))(w, )](, v) = e
0 1 w 1/2
+(
) I1 (2(0 1 w(v w))1/2 )],
vw
1
1 v
L1
0 (v w) + e1 (vw)0 w [1 I0 (2(0 1 w(v w))1/2 )
s [Lr (1 (r, s))(w, )](, v) = e
1 0 (v w) 1/2
+(
) I1 (2(0 1 w(v w))1/2 )].
w

Thus, we get f0 (w, v) and f1 (w, v), the distribution function of T0 and T1 , such that
f0 (w, v) = e0 v 0 (v w) + e1 (vw)0 w [0 I0 (2(0 1 w(v w))1/2 )
0 1 w 1/2
+(
) I1 (2(0 1 w(v w))1/2 )],
vw
f1 (w, v) = e1 v 0 (v w) + e1 (vw)0 w [1 I0 (2(0 1 w(v w))1/2 )
1 0 (v w) 1/2
) I1 (2(0 1 w(v w))1/2 )].
+(
w
26

Вам также может понравиться