American Option Pricing Using A Markov Chain Approximation
Jin-Chuan Duan
Hong Kong University of Science and Technology
Correspondence to:
Prof. Jin-Chuan Duan Department of Finance
Hong Kong University of Science & Technology Clear Water Bay, Kowloon, Hong Kong
Tel: (852) 2358 7671; Fax: (852) 2358 1749 E-mail: jcduan@ust.hk
Markov Chain OPM JC Duan (3/2000)
2
A two-way classification of option valuation problems
• Path-independent vs. path-dependent payoff functions
Path-independent: European options, American options, Bermudan options, digital options Path-dependent: lookback options, Asian options,
knock-out (in) options
• Path-independent vs. path-dependent underlying stochastic processes
Path-independent: Black-Scholes model, CEV model, jump-diffusion
Path-dependent: GARCH model
• In a numerical sense, it is typically easier to deal with path- dependence (a non-Markovian feature) arising from the payoff function than path-dependence inherent in the underlying dynamic.
• The traditional numerical methods are poor in either handling early exercise or the cases involving path-
dependent payoffs (and worse for the cases involving path- dependent underlying dynamic).
Different numerical methods for option valuation
• Monte Carlo (quasi and pseudo) methods
Exceedingly flexible in dealing with path-dependent payoffs and path-dependent underlying dynamic. Poor in handling options with early exercise possibilities.
Computing time intensive even with some variance reduction technique.
• Finite difference/element methods
Good at solving the option pricing problem that can be cast as a partial differential equation. Cannot deal with the
discrete time valuation models.
• Lattice methods
Hard to ensure recombination and harder still to
accommodate path-dependent payoffs and/or underlying dynamics.
• Analytical approximation methods Highly valuation problem specific.
• Neural network methods
Don’t stand alone and are intended to be a complementary speed accelerator for real-time runs (see Hanke, 1997).
• Markov chain method (Duan & Simonato, 1999)
Markov Chain OPM JC Duan (3/2000)
4
Black-Scholes Model
• Asset return under the risk neutral probability measure :
t
t r dt dW
S
d σ +σ
−
= 2
2 ln 1
where
σ : volatility rate r : risk free rate
W : standard Brownian motiont
• American option price with discrete exercise points :
[ ]
{
f St K e rEQ V St t t}
t t S
V( , ) = max ( , ), − ( +1, +1)|ℑ
where
V S t( t, ) : American option’s price K : Strike price
f S K( t, ) : European option’s payoff
V S( T , )T = f S( T , )K
• Discretize the underlying asset price
Let 2) ln( )
2
(r 1 t St
pt ≡ − − σ +
Then,
dWt
St d dt t r
dp
σ
σ
=
+
−
−
= 2) ln( )
2 ( 1
• Jarrow and Rudd’s (1983, eq (13-18)) n-step binomial tree approximation
Approximation target: tp
Up and down moves: and
n d T
n
u =σ T = −σ
Probability:
2
= 1 q
An n-step binomial tree has m discrete prices where 1
2 +
= n
m .
Markov Chain OPM JC Duan (3/2000)
6
Vector for the logarithmic adjusted stock price:
[
p(1), p(2), , p(m)]
P = !
where
nu p
m p
u p n
p
p n
p
d n
p p
nd p
p
+
=
+
= +
= +
− +
=
+
=
0 0 0 0 0
) (
) 2 (
) 1 (
) 1 (
) 2 (
) 1 (
"
"
Example: A 2-step binomial tree (m = 5)
p(1) p(2) p(3) p(4) p(5)
p(1) p(2) p(3) p(4) p(5)
p(1) p(2) p(3) p(4) p(5)
A Markov chain interpretation with the following transition probability matrix
= Π
1 0
0 0
0
5 . 0 0 5 . 0 0 0
0 5 . 0 0 5 . 0 0
0 0
5 . 0 0 5 . 0
0 0
0 0
1
American option valuation numerically
{ }
V P t( , ) = max g P K t( , , ) , e−rΠV P t( , +1)
where
V P t( , ) : American option’s price K : strike price
g P K t( , , ) : European option’s payoff
) , , ( )
,
(P T g P K T
V =
Example: a put option
− − +
= ) 1 ],0
2 exp[( 1
1 max )
, ,
(P K t k r 2 t P
g σ
Shortcoming of the lattice approach
1. Rigidity of geometry, i.e., the number of states is tied to the number of steps.
2. Recombined lattices are hard to construct.
Markov Chain OPM JC Duan (3/2000)
8
• A general Markov chain method The structure
p(1) p(2) p(3) p(4) p(5)
p(1) p(2) p(3) p(4) p(5)
p(1) p(2) p(3) p(4) p(5)
Transition probability matrix
= Π
)
; 5 , 5 ( )
; 4 , 5 ( )
; 3 , 5 ( )
; 2 , 5 ( )
; 1 , 5 (
)
; 5 , 4 ( )
; 4 , 4 ( )
; 3 , 4 ( )
; 2 , 4 ( )
; 1 , 4 (
)
; 5 , 3 ( )
; 4 , 3 ( )
; 3 , 3 ( )
; 2 , 3 ( )
; 1 , 3 (
)
; 5 , 2 ( )
; 4 , 2 ( )
; 3 , 2 ( )
; 2 , 2 ( )
; 1 , 2 (
)
; 5 , 1 ( )
; 4 , 1 ( )
; 3 , 1 ( )
; 2 , 1 ( )
; 1 , 1 (
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
τ π
where
{ }
{ }
π τ τ
σ τε
σ τ ε
σ τ
σ τ σ τ
( , ; ) Pr ( ) ( ) | ( )
Pr ( ) ( )| ( )
Pr ( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
i j Q
c j pt c j pt p i Q
t t t
Q
t
c j p c j p p i
c j p i c j p i
c j p i c j p i
= ≤ + < + =
≤ + < + =
− ≤ < + −
+ −
− −
1
1 1 1
=
=
= Φ Φ
and Φ(.) is the standard normal distribution function
Density of the transition probability matrix
Fact: For a fixed partition, the density of the transition probability matrix depends on the length of one operation time interval
Transition probability matrix using one step T=90, m=101, σ=0.2, S0=50, τ=90/365
Transition probability matrix using 90 steps T=90, m=101, σ=0.2, S0=50, τ=1/365
Markov Chain OPM JC Duan (3/2000)
10
• A numerical example
European put options in the Black and Scholes framework
Parameters:
1
Strike Price 48
Stock Price 50
Interest Rate 0.05 d1 0.55411
Standard Deviation 0.2 d2 0.35411
Price computed by Black and Scholes Formula: 2.024003 Binomial Tree Approach # of steps 2
Step size 0.141421
3.62918 3.62918 3.62918
3.77060 3.77060 3.77060
3.91202 3.91202 3.91202
4.05344 4.05344 4.05344
4.19487 4.19487 4.19487
Backstep 2 Backstep 1 Payoff K - S
8.72325 8.94408 9.17050 9.17050
5.13971 4.47204 3.27191 3.27191
2.18081 1.59556 0.00000 -3.52273
0.77808 0.00000 0.00000 -11.34954
0.00000 0.00000 0.00000 -20.36532
Transition probability matrix
1 0 0 0 0
0.5 0 0.5 0 0
0 0.5 0 0.5 0
0 0 0.5 0 0.5
0 0 0 0 1
Maturity
Markov Chain OPM JC Duan (3/2000)
12
Markov Chain Approach # of prices 5
3.62918 3.62918 3.62918
3.77060 3.77060 3.77060
3.91202 3.91202 3.91202
4.05344 4.05344 4.05344
4.19487 4.19487 4.19487
Backstep 2 Backstep 1 Payoff K - S
5.71204 6.95589 9.17050 9.17050
3.91773 3.98155 3.27191 3.27191
1.96324 1.36892 0.00000 -3.52273
0.69951 0.24891 0.00000 -11.34954
0.17864 0.02115 0.00000 -20.36532
Transition Probability Matrix
0.69146 0.24173 0.06060 0.00598 0.00023 0.30854 0.38292 0.24173 0.06060 0.00621 0.06681 0.24173 0.38292 0.24173 0.06681 0.00621 0.06060 0.24173 0.38292 0.30854 0.00023 0.00598 0.06060 0.24173 0.69146
• Numerical results for the Black-Scholes model (Duan &
Simonato, 1999)
⇒ Table 1 : European puts
⇒ Table 2 : American puts
GARCH Option Pricing Model
• The reasons for GARCH option pricing A. Option prices
1. Volatility smile/smirk
2. Term structure of volatilities
3. Black-Scholes implied volatilities are higher than historical (or realized) volatilities
B. Historical returns
1. Volatility clustering
2. Negative return skewness 3. Excess return kurtosis
• Asset return under the data generating probability measure P
1 2 1
2 1
1 0
2
1 1 1
1 1
) (
2 ln 1
θ ξ
β β
β
ξ λ
− +
+
=
+
− +
=
+ +
+ +
+ + +
+ +
t t
t t
t t t
t t
t
h h
h
h h
h S r
S
where
r : risk-free rate
λ : risk premium parameter θ : leverage parameter
ξt+1 ~ N(0,1) with respect to P
Markov Chain OPM JC Duan (3/2000)
14
• Asset return under the risk neutral probability measure Q (Duan 1995)
2 1
1 2 1
1 0
2
1 1 1 1
) (
2 ln 1
λ θ ε
β β
β
ε
−
− +
+
=
+
−
=
+ +
+ +
+ + + +
t t
t t
t t t
t t
h h
h
h h
S r S
where
εt+1 ~ N(0,1) with respect to Q
• American option valuation in the GARCH framework
[ ]
{ }
V S h( ,t t+1, )t = max f S K( , ) ,t e−rEQ V S( t+1,ht+2,t + ℑ1)| t
where
V S h( t, t+1, )t : American option’s price
K : Strike price
f S K( t, ) : European option’s payoff
V S( T,hT+1, )T = f S( T, )K
• Discretize the underlying asset prices
1) 1 ln(
) ln(
*) 2 ( 1
= + +
+
−
−
=
ht qt
St t
h t r
p
where
] ) (
1 [ 1
*
2 2
1
0
λ θ β
β
β
+ +
−
= − h
Example: m=3, n=3
(p(1), q(1)) (p(2), q(1)) (p(3), q(1))
(p(1), q(1)) (p(2), q(1)) (p(3), q(1))
(p(1), q(2)) (p(2), q(2)) (p(3), q(2))
(p(1), q(2)) (p(2), q(2)) (p(3), q(2))
(p(1), q(3)) (p(2), q(3)) (p(3), q(3))
(p(1), q(3)) (p(2), q(3)) (p(3), q(3))
Markov Chain OPM JC Duan (3/2000)
16
• Transition probability matrix
Π =
π π π π π
π π π π π
π π π π
( , ; , ) ( , ; , ) ( , ; , ) ( , ; , ) ( , ; , ) ( , ; , ) ( , ; , ) ( , ; , ) ( , ; , ) ( , ; , )
( , ; , ) ( , ; , ) ( , ; , ) ( , ; ,
1 11 1 1 1 2 1 1 1 1 1 11 2 1 1
2 11 1 2 1 2 1 2 1 1 2 11 2 2 1
11 1 1 2 1 1 1 11 2
! !
! !
" " # " " # "
!
m m n
m m n
m m m m m ) ! ( , ; , )
" " # " " # "
π 1 2 m n
Since
2 1
1 2 1
1 0
2
1 1 1 1
) (
2 ln 1
λ θ ε
β β
β
ε
−
− +
+
=
+
−
=
+ +
+ +
+ + + +
t t
t t
t t t
t t
h h
h
h h
S r S
qt+2 is a deterministic function of qt+1, pt+1, pt; that is, qt+2 = Φ( qt+1, pt+1, pt )
{ } ( )
∈ = = Φ ∈
=
+ +
otherwise 0
) ( )
( ), ( ), ( if
) ( ),
(
| ) ( Pr
) ,
; , (
1
1 C k p p i q q j q j p k p i D l
p l k j i
t t
t Q
π
• American option prices in the GARCH framework
Price vector
[
(1) (2) ( ) (1) (2) ( )]
' p p p m p p p m
P = ! ! !
American option price computation
{
( , , ), ( , 1)}
max )
,
(P t = g P K t e− ΠV P t +
V r
where
V P t( , ) : American option’s price K : Strike price
g P K t( , , ) : European option’s payoff function V P T( , ) = g P K T( , , )
Markov Chain OPM JC Duan (3/2000)
18
• Density of the transition probability matrix (GARCH)
m=101, n=11,
S0 = 50,r = 0 05. ,β0 = 0 00001. ,β1 = 0 8. ,β2 = 0 1. ,λ = 0 2.
• Numerical results for the GARCH model (Duan &
Simonato, 1999)
⇒ Table 3 : European put
⇒ Table 4 : American put
References
Duan, J.-C. and J.-G. Simonato, 1999, “American Option Pricing under
GARCH by a Markov Chain Approximation,” Journal of Economic Dynamics and Control, forthcoming.
Hanke, M., 1997, “Neural Network Approximation of Option Pricing Formulas for Analytically Intractable Option Pricing Models,” Journal of Computational Intelligence in Finance 5, 20-27.
Jarrow, R. and A. Rudd, 1983, Option Pricing, Richard D. Irwin, Inc.