Hilbert transform approach for pricing Bermudan options in … · Introduction Bermudan vanilla...
Transcript of Hilbert transform approach for pricing Bermudan options in … · Introduction Bermudan vanilla...
IntroductionBermudan vanilla
Bermudan barrier/look
Hilbert transform approach for pricing Bermudanoptions in Levy models
Liming Feng1
1Dept. of Industrial & Enterprise Systems EngineeringUniversity of Illinois at Urbana-Champaign
Joint with Xiong Lin
Spectral and cubature methods in finance and econometricsUniversity of Leicester, 6/19/2009
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Levy modelsHilbert transform methods
Levy models
Better fit to empirical data, explain volatility smiles
Finite activity jump diffusion modelsMerton 76 (normal jump diffusion), Kou 02 (double exponential
jump diffusion)
Infinite activity pure jump Levy modelsMadan et al 90, 91, 98 (variance gamma), Barndorff-Nielsen 98(normal inverse Gaussian), Carr, Geman, Madan & Yor 02(CGMY), Eberlein, Keller & Prause 98 (Generalized hyperbolic)
Books on Levy processesBoyarchenko & Levendorskii 02, Cont & Tankov 04, Schoutens03, Bertoin 96, Sato 99, Applebaum 04, Kyprianou 06
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Levy modelsHilbert transform methods
Levy-Khintchine theorem
Levy process: independent and stationary increments
Analytical tractability: characteristic functions available viaLevy-Khintchine Theorem
φt(ξ) = E[e iξXt ] = e−tψ(ξ)
where ψ is the characteristic exponent
ψ(ξ) =1
2σ2ξ2 − iµξ +
∫R(1− e iξx + iξx1|x |≤1)Π(dx)
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Levy modelsHilbert transform methods
Fourier transform method
Fourier transform method for European options: Carr &Madan 99
c(k) =1
2πe−αk
∫R
e−iξkΦ(ξ)dξ
where
Φ(ξ) =e−rTφT (ξ − i(α+ 1))
(iξ + α)(iξ + α+ 1)
Extended in Lee 04: more general payoffs, exponentiallydecaying errors
Bermudan options: vanilla, discrete barrier/lookback options,Hilbert transform method
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Levy modelsHilbert transform methods
Hilbert transform
Hilbert transform of f ∈ Lp(R), 1 ≤ p <∞
Hf (x) =1
πp.v .
∫ ∞
−∞
f (y)
x − ydy
For any f ∈ L1(R) with f ∈ L1(R)
F(sgn · f )(ξ) = iHf (ξ)
F(1(l ,∞) · f )(ξ) =1
2f (ξ) +
i
2e iξlH(e−iηl f (η))(ξ)
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Levy modelsHilbert transform methods
Option pricing
European vanilla options:
E[(K − ST )+] = KE[1ST<K]− E[ST1ST<K]
P(X ≤ x) =
∫R
1(−∞,x)(y)p(y)dy
European style discrete barrier options: Feng & Linetsky 08
f j(x) = 1(l ,u) · Etj ,x [fj+1(Xtj+1)]
European style discrete lookbacks: Feng & Linetsky 09
Mj−Xj = max(Mj−1,Xj)−Xj = max(0,Mj−1−Xj−1−(Xj−Xj−1))
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Levy modelsHilbert transform methods
Discrete Hilbert transform
Discrete Hilbert transform with step size h > 0
Hhf (x) =∞∑
m=−∞f (mh)
1− cos[π(x −mh)/h]
π(x −mh)/h, x ∈ R
For f analytic in a horizontal strip z ∈ C : |=(z)| < d
||Hf −Hhf ||L∞(R) ≤Ce−πd/h
πd(1− e−πd/h)
Related to Whittaker cardinal series (sinc) expansion
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Levy modelsHilbert transform methods
Sinc expansion of analytic functions
Whittaker cardinal series (sinc expansion)
c(f , h)(x) =∞∑−∞
f (kh)sin(π(x − kh)/h)
π(x − kh)/h
For entire functions of exponential type π/h, sinc expansion isexact: c(f , h) = f
For functions analytic in a strip z ∈ C : |=(z)| < d, Stenger93
||f − c(f , h)||L∞ ≤ Ce−πd/h
πd(1− e−2πd/h)
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Levy modelsHilbert transform methods
Trapezoidal rule
Take Hilbert transform on c(f , h), obtain discrete Hilberttransform
Trapezoidal rule very accurate for f analytic in a stripz ∈ C : |=(z)| < d∣∣∣∣∣
∫R
f (x)dx −∞∑
m=−∞f (kh)h
∣∣∣∣∣ ≤ Ce−2πd/h
1− e−2πd/h
We use trapezoidal rule to compute Fourier inverse integral
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Bermudan style vanilla options
Bermudan put: payoff G (S) = (K −S)+, discrete monitoring
T = t0, t1, · · · , tN = 0,∆, 2∆, · · · ,N∆ = T
Exponential Levy model: Xt a Levy process in (Ω,F ,F,P),start from ln(S0/K ), equivalent martingale measure P is given
St = KeXt
Optimal stopping
V 0(S0) = supτ
E0[e−rτG (Sτ )]
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Backward induction in state space
Variable change x = ln(S/K ),
g(x) = G (Kex) = K (1− ex)+, f 0(x) = V 0(Kex)
Backward induction
f N(x) = g(x)
f j(x) = max(g(x), e−r∆Ej∆,x [f
j+1(X(j+1)∆)]), 0 ≤ j < N
V 0(S0) = f 0(ln(S0/K ))
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Direct implementation
Knowing the transition density p∆(·)
Ej∆,x [fj+1(X(j+1)∆)] =
∫R
f j+1(y)p∆(y − x)dy
When discretized, the convolution becomes a Toeplitz matrixvector multiplication, can be implemented using FFT(Eydeland 94 )
Density may not be available, polynomial convergence
Double exponential fast Gauss transform (Broadie &Yamamoto 05 ), limited to the Black-Schoels-Merton modeland Merton’s model
For general Levy models: COS method Fang & Oosterlee 08
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Dampening for integrability
For α > 0 (for puts), define
f jα(x) = eαx f j(x), gα(x) = eαxg(x)
Backward induction
f Nα (x) = gα(x)
f jα(x) = max
(gα(x), eαx−r∆Ej∆,x [e
−αX(j+1)∆f j+1α (X(j+1)∆)]
),
V 0(S0) = e−α ln(S0/K)f 0α (ln(S0/K ))
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Esscher transform
Define new measure Pα via
dPα
dP|Ft =
e−αXt
φt(iα)
Esscher transformed Levy process is still a Levy process
φαt (ξ) =φt(ξ + iα)
φt(iα)
Applying property Eαs [Yt ] = Es [YtZt/Zs ]
Ej∆,x [e−αX(j+1)∆f j+1
α (X(j+1)∆)] = e−αx−∆ψ(iα)Eαj∆,x [f j+1α (X(j+1)∆)]
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Dampened backward induction
Dampened backward induction
f Nα (x) = gα(x)
For 0 ≤ j < N
f jα(x) = max
(gα(x), e−(r+ψ(iα))∆Eαj∆,x [f j+1
α (X(j+1)∆)])
= gα(x)1(−∞,x∗j )(x)
+e−(r+ψ(iα))∆Eαj∆,x[f j+1α (X(j+1)∆)
]1[x∗j ,∞)(x)
where x∗j is the early exercise boundary at time j∆
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Backward induction in Fourier space
By convolution theorem, Fourier transform of
Eαj∆,x[f j+1α (X(j+1)∆)
]=
∫R
f j+1α (y)pα∆(y − x)dy
is a product f j+1α (ξ)φα∆(−ξ)
Backward induction in Fourier space
f Nα (ξ) = gα(ξ)
f jα(ξ) = F(gα1(−∞,x∗j ))(ξ) + e−(r+ψ(iα))∆
(1
2f j+1α (ξ)φα∆(−ξ)
+i
2e iξx∗j H
(e−iηx∗j f j+1
α (η)φα∆(−η))
(ξ)
)Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Early exercise boundary
Early exercise boundary solves
gα(x) = e−(r+ψ(iα))∆Eαj∆,x [f j+1α (X(j+1)∆)]
Using Fourier inverse representation
gα(x) =1
2πe−(r+ψ(iα))∆
∫R
e−iξx f j+1α (ξ)φα∆(ξ)dξ
x∗N = K . To solve for x∗j , use Newton-Raphson, with startingpoint x∗j+1
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Algorithm summarized
Start with Fourier transform of dampened payoff f Nα = gα
At time j∆, with f j+1α , compute early exercise boundary x∗j
using Newton Raphson
Compute f jα from f j+1
α and x∗j (Hilbert transform)
With f 1α , option value at time 0
f 0α (x) = max
(gα(x),
1
2πe−(r+ψ(iα))∆
∫R
e−iξx f 1α (ξ)φα∆(ξ)dξ
)
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Discrete approximation
Need to repeatedly evaluate a Fourier inverse integral andHϕ(ξ)
Trapezoidal rule for Fourier inverse integral, truncate infiniteseries with truncation level M, computational cost O(M)
Replace Hϕ by discrete Hilbert transform, truncate resultinginfinite series
Hϕ(ξ) ⇐M∑
m=−M
ϕ(mh)1− cos[π(ξ −mh)/h]
π(ξ −mh)/h
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Error estimate
Discretization error ∼ O(exp(−πd/h)
With φt(ξ) ∼ exp(−ct|ξ|ν), truncation error is essentially
O(exp(−∆c(Mh)ν))
Select h = h(M) according to
h(M) =
(πd
∆c
) 11+ν
M− ν1+ν
Total error: O(exp(−CMν
1+ν ))
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Toeplitz matrix vector multiplication
Evaluate
Hψ(ξ) ⇐M∑
m=−M
ψ(mh)1− cos[π(ξ −mh)/h]
π(ξ −mh)/h
for ξ = −Mh, · · · ,Mh
Correspond to Toeplitz matrix vector multiplication
FFT based method for such multiplications: O(M log(M))
Total computational cost of the method: O(NM log(M))
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
NIG model
Pricing Bermudan put option in NIG
NIG process: time changed Brownian motion, with an inverseGaussian process as the stochastic time
Characteristic exponent
ψ(ξ) = −iµξ+ δNIG (√α2
NIG − (βNIG + iξ)2−√α2
NIG − β2NIG )
φt(ξ) has exponential tails with ν = 1, error estimate in M:
O(e−C√
M)
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Backward induction in state spaceBackward induction in Fourier spaceDiscrete approximation
Bermudan put in the NIG model
Figure: T = 1,N = 252,S0 = 100,K = 110, r = 0.1, q = 0, αNIG =25, βNIG = −5, δNIG = 0.5, implemented in Matlab using a Laptop withCPU 2GHz, RAM 1G, last node takes 2.9s
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Barrier optionsLookback options
Bermudan barrier options
Bermudan down-and-out put with lower barrier L, l = ln(L/K )
Backward induction
f N(x) = g(x)1(l ,∞)(x)
For 0 ≤ j < N,
f j(x) = 1(l ,∞)(x) max(g(x), e−r∆Ej∆,x [fj+1(X(j+1)∆)])
= g(x)1(l ,∞)(x)1(−∞,x∗j )(x)
+e−r∆Ej∆,x [fj+1(X(j+1)∆)]1(l ,∞)(x)1(x∗j ,∞,)(x)
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Barrier optionsLookback options
CGMY model
Pricing Bermudan down-and-out put in CGMY
Levy density
π(x) =CeGx
|x |1+Y1x<0 +
Ce−Mx
|x |1+Y1x>0
Characteristic exponent
ψ(ξ) = −iµξ−CΓ(−Y )((M − iξ)Y −MY +(G + iξ)Y −GY )
Characteristic function has exponential tails with ν = Y , errorestimate in M: O(exp(−CMY /(1+Y )))
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Barrier optionsLookback options
Bermudan DOP in the CGMY model
Figure: T = 1,N = 252,S0 = 100,K = 110, r = 0.1, q = 0,C = 3,G =22,M = 28,Y = 0.8, implemented in Matlab using a Laptop with CPU2GHz, RAM 1G, last node takes 2.0s
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Barrier optionsLookback options
Bermudan floating strike lookbacks
Standard backward induction involves two state variables:asset price, maximum asset price
Can be reduced to one state variable, maximum assetprice/asset price
f j(y) = max(ey − 1, e−q∆E∗j∆,y [f j+1(eY(j+1)∆)]
Double exponential fast Gauss transform method (Yamamoto05 ), limited to BSM and Merton’s models
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Barrier optionsLookback options
Kou’s model
Pricing Bermudan floating strike lookback put
Log asset price follows
Xt = µt+σBt+Nt∑
n=1
Zn, Zn ∼ pη1e−η1x1x>0+(1−p)η2e
η2x1x<0
Characteristic exponent
ψ(ξ) = −iµξ +1
2σ2ξ2 + λ
(1− pη1
η1 − iξ− (1− p)η2
η2 + iξ
)Characteristic function has exponential tails with ν = 2, errorestimate O(exp(−CM2/3))
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Barrier optionsLookback options
Bermudan floating strike lookback put in Kou’s model
Figure: T = 1,N = 252,S0 = 100, r = 0.1, q = 0, σ = 0.25, λ = 3, p =0.3, η1 = 17, η2 = 12, implemented in Matlab using a Laptop with CPU2GHz, RAM 1G, last node takes 1.1s
Liming Feng Pricing Bermudan options in Levy models
IntroductionBermudan vanilla
Bermudan barrier/look
Barrier optionsLookback options
Summary
Hilbert transform method for pricing Bermudan style optionsin Levy process models
Very accurate with exponentially decay errors
Fast with computational cost O(NM log(M))
European vanilla, barrier, lookback, defaultable bonds,Bermudan vanilla, barrier, floating strike lookback, montecarlo simulation etc.
Liming Feng Pricing Bermudan options in Levy models