Fourier Space Time-stepping Method for Option Pricing with Lévy Processes

Size: px

Start display at page:

Download "Fourier Space Time-stepping Method for Option Pricing with Lévy Processes"

Adelia Gibbs
5 years ago
Views:

1 FST method Extensions Indifference pricing Fourier Space Time-stepping Method for Option Pricing with Lévy Processes Vladimir Surkov University of Toronto Computational Methods in Finance Conference University of Waterloo July 27, 2007 Joint work with Ken Jackson and Sebastian Jaimungal, University of Toronto 1 / 32

2 FST method Extensions Indifference pricing 1 Fourier Space Time-stepping method Infinitesimal generator and characteristic exponent Method derivation Numerical results 2 Extensions Multi-asset options Regime switching 3 Indifference pricing Optimal investment problems Application of FST to solution of HJB equations 2 / 32

3 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results The Option Pricing Problem Option payoff is given by ϕ(s) Stock price follows an exponential Lévy model: S(t) = S(0)e µt+x (t), X (t) is a Lévy process 3 / 32

4 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results The Option Pricing Problem Option payoff is given by ϕ(s) Stock price follows an exponential Lévy model: S(t) = S(0)e µt+x (t), X (t) is a Lévy process Generalizing PIDE for Lévy processes { t v + Lv = 0 v(t, x) = ϕ(s(0) e x ) where L is the infinitesimal generator of the Lévy process: Lf = γ x f + σ2 2 [ xxf + f (x + y) f (x) y1 { y <1} x f (x) ] ν(dy) R/{0} 3 / 32

5 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Finite Difference Methods for Option Pricing Alternating Direction Implicit-FFT - Andersen and Andreasen (2000) Implicit-Explicit (IMEX) - Cont and Tankov (2004) IMEX Runge-Kutta - Briani, Natalini, and Russo (2004) Fixed Point Iteration - d Halluin, Forsyth, and Vetzal (2005) 4 / 32

6 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Finite Difference Methods for Option Pricing Alternating Direction Implicit-FFT - Andersen and Andreasen (2000) Implicit-Explicit (IMEX) - Cont and Tankov (2004) IMEX Runge-Kutta - Briani, Natalini, and Russo (2004) Fixed Point Iteration - d Halluin, Forsyth, and Vetzal (2005) Common Features: Treat the integral term explicitly to avoid solving a dense system of linear equations. Use the Fast Fourier Transform (FFT) to speed up the computation of the integral term (which can be regarded as a convolution) 4 / 32

7 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Finite Difference Methods for Option Pricing Alternating Direction Implicit-FFT - Andersen and Andreasen (2000) Implicit-Explicit (IMEX) - Cont and Tankov (2004) IMEX Runge-Kutta - Briani, Natalini, and Russo (2004) Fixed Point Iteration - d Halluin, Forsyth, and Vetzal (2005) Common Features: Treat the integral term explicitly to avoid solving a dense system of linear equations. Use the Fast Fourier Transform (FFT) to speed up the computation of the integral term (which can be regarded as a convolution) Drawbacks: Diffusive and integral terms are treated asymmetrically Large jump are truncated and small jumps approximated by diffusion Difficult to extend to higher dimensions 4 / 32

8 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Infinitesimal Generator and Characteristic Exponent The characteristic exponent of a Lévy-Khinchin representation can be factored from a Fourier transform of the operator (Sato 1999) F[Lv](τ, ω) = {iγω σ2 ω 2 } + [e iωx 1 iωy1 { ω <1} ]ν(dy) F[v](τ, ω) 2 = ψ(ω)f[v](τ, ω) Model Characteristic Exponent ψ(ω) Black-Scholes-Merton iµω σ2 ω 2 2 Merton Jump-Diffusion iµω σ2 ω λ(e i µω σ2 ω 2 /2 1) Variance Gamma 1 κ log(1 iµκω + σ2 κω 2 2 ) CGMY CΓ( Y ) [ (M iω) Y M Y + (G +iω) Y G Y ] 5 / 32

9 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Numerical Method Derivation Apply the Fourier transform to the pricing PIDE { t F[v](t, ω) + Ψ(ω)F[v](t, ω) = 0, F[v](T, ω) = F[ϕ](ω) 6 / 32

10 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Numerical Method Derivation Apply the Fourier transform to the pricing PIDE { t F[v](t, ω) + Ψ(ω)F[v](t, ω) = 0, F[v](T, ω) = F[ϕ](ω) Resulting ODE has explicit solution F[v](t 1, ω) = F[v](t 2, ω) e (t2 t1)ψ(ω) 6 / 32

11 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Numerical Method Derivation Apply the Fourier transform to the pricing PIDE { t F[v](t, ω) + Ψ(ω)F[v](t, ω) = 0, F[v](T, ω) = F[ϕ](ω) Resulting ODE has explicit solution F[v](t 1, ω) = F[v](t 2, ω) e (t2 t1)ψ(ω) Apply the inverse Fourier transform v(t 1, x) = F 1 { F[v](t 2, ω) e (t2 t1)ψ(ω)} (x) 6 / 32

12 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Numerical Method Derivation Apply the Fourier transform to the pricing PIDE { t F[v](t, ω) + Ψ(ω)F[v](t, ω) = 0, F[v](T, ω) = F[ϕ](ω) Resulting ODE has explicit solution F[v](t 1, ω) = F[v](t 2, ω) e (t2 t1)ψ(ω) Apply the inverse Fourier transform v(t 1, x) = F 1 { F[v](t 2, ω) e (t2 t1)ψ(ω)} (x) Fourier Space Time-stepping (FST) method v n 1 = FFT 1 [FFT[v n ] e Ψ t ] 6 / 32

13 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results European Call Option N Value Change log 2 Ratio CPU-Time Option: European call S = 1.0, K = 1.0, T = 0.2 Model: Kou jump-diffusion σ = 0.2, λ = 0.2, p = 0.5, η = 3, η + = 2, r = 0.0 Quoted Price: Almendral and Oosterlee (2005) 7 / 32

14 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results American Put Option N M Value Change log 2 Ratio CPU-Time Option: American put S = 90.0, K = 98.0, T = 0.25 Model: CGMY C = 0.42, G = 4.37, M = 191.2, Y = , r = 0.06 Quoted Price: Forsyth, Wan, and Wang (2006) 8 / 32

15 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results American Put Option N M Value Change log 2 Ratio CPU-Time Option: American put S = 10.0, K = 10.0, T = 0.25 Model: CGMY C = 1.0, G = 8.8, M = 9.2, Y = 1.8, r = / 32

16 FST method Extensions Indifference pricing Characteristic exponent Method derivation Numerical results Up-and-Out Barrier Call Option N M Value Change log 2 Ratio CPU-Time Option: Up-and-Out Barrier Call S = 100.0, K = 100.0, B = 110, T = 1.0 Model: Black-Scholes-Merton σ = 0.15, r = 0.05, q = 0.02 Closed-Form Price: Hull (2005) 10 / 32

17 FST method Extensions Indifference pricing Multi-asset options Regime switching Multi-asset options Pricing PIDE { ( t + L) v = 0, v(t, x) = ϕ(s(0) e x ) 11 / 32

18 FST method Extensions Indifference pricing Multi-asset options Regime switching Multi-asset options Pricing PIDE { ( t + L) v = 0, v(t, x) = ϕ(s(0) e x ) Multi-dimensional FST method v n 1 = FFT 1 [FFT[v n ] e Ψ t ] 11 / 32

19 FST method Extensions Indifference pricing Multi-asset options Regime switching Multi-asset options Pricing PIDE { ( t + L) v = 0, v(t, x) = ϕ(s(0) e x ) Multi-dimensional FST method v n 1 = FFT 1 [FFT[v n ] e Ψ t ] Jackson, Jaimungal and Surkov (2007) discuss application of FST to pricing of spread, American spread and catastrophe equity put options 11 / 32

20 FST method Extensions Indifference pricing Multi-asset options Regime switching Catastrophe Equity Put (CatEPut) In the event of large (catastrophic) losses U, the insurer receives a put option on its own stock ϕ(s(t ), L T ) = 1 LT >U (K S(T )) + 12 / 32

21 FST method Extensions Indifference pricing Multi-asset options Regime switching Catastrophe Equity Put (CatEPut) Presence of losses drives the share value down, not an independent jump process S(t) = S(0) exp { α L(t) + γ t + σ W t } L(t) = N(t) l i n=1 13 / 32

22 FST method Extensions Indifference pricing Multi-asset options Regime switching Catastrophe Equity Put (CatEPut) Presence of losses drives the share value down, not an independent jump process S(t) = S(0) exp { α L(t) + γ t + σ W t } L(t) = N(t) l i n=1 When losses are modeled as a Gamma r.v., characteristic exponent is [ ] ( Ψ(ω 1, ω 2 ) = i γ ω σ2 ω1 2 + λ 1 i( αω 1 + ω 2 ) v ) m 2 m ) v 1 13 / 32

23 FST method Extensions Indifference pricing Multi-asset options Regime switching European CatEPut 14 / 32

24 FST method Extensions Indifference pricing Multi-asset options Regime switching American CatEPut 15 / 32

25 FST method Extensions Indifference pricing Multi-asset options Regime switching Regime Switching Let K := {1,..., K} denote the possible hidden states of the world, driven by a continuous time Markov chain Z t. The transition probability from state k at time t 1 to state l at time t 2 is given by P t1t2 kl = Q(Z t2 = l Z t1 = k) = (exp{(t 2 t 1 )A}) kl where A is the Markov chain generator Within state k, log-stock follows Lévy model k 16 / 32

26 FST method Extensions Indifference pricing Multi-asset options Regime switching Regime Switching Pricing PIDE { [ t + ( A kk + L (k))] v(x, k, t) + j k A jk v(x, j, t) = 0 v(x, k, T ) = ϕ(s(0)e x ) 17 / 32

27 FST method Extensions Indifference pricing Multi-asset options Regime switching Regime Switching Pricing PIDE { [ t + ( A kk + L (k))] v(x, k, t) + j k A jk v(x, j, t) = 0 v(x, k, T ) = ϕ(s(0)e x ) FST Method where v n 1 = FFT 1 [FFT[v n ] e Ψ t ] { Akk + Ψ Ψ(ω) kl = (k) (ω), k = l A kl, k l No time-stepping required for European options 17 / 32

28 FST method Extensions Indifference pricing Multi-asset options Regime switching American put option with 3 regimes N M Value Change log 2 Ratio CPU-Time Option: American put S = 100.0, K = 100.0, T = 1.0 Model: Merton jump-diffusion σ = 0.15, µ = 0.5, σ = 0.45, r = 0.05, q = 0.02, λ [0.3, 0.5, 0.7], p = [0.2, 0.3, 0.5], A = [ 0.8, 0.6, 0.2; 0.2, 1, 0.8; 0.1, 0.3, 0.4] 18 / 32

29 FST method Extensions Indifference pricing Multi-asset options Regime switching Exercise boundary for American put option with 3 regimes 19 / 32

30 FST method Extensions Indifference pricing Multi-asset options Regime switching FST Method Strengths Stable and robust, even for options with discontinuous payoffs Easily extendable to various stochastic processes and no loss of performance for infinite activity processes Can be applied to multi-dimensional and regime-switching problems in a natural manner Computationally efficient Computational cost is O(MNlogN) while the error is O( x 2 + t 2 ) European options priced in a single time-step Bermudan style options do not require time-stepping between monitoring dates 20 / 32

31 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Indifference Pricing In incomplete markets (non-traded assets, transaction costs, portfolio constraints, etc.) perfect replication is impossible Investor can still maximize the expected utility of wealth through dynamic trading The price of a claim is the initial wealth forgone so that the investor is no worse off in expected utility terms at maturity The framework incorporates wealth dependence, non-linear pricing and risk-aversion 21 / 32

32 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB The Modelling Framework Utility function is a strictly increasing and concave function ranking investor s preferences of wealth Popular choices include CRRA model U(x) = x 1 γ 1 γ or CARA model U(x) = 1 1 γ e γx 22 / 32

33 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB The Modelling Framework Utility function is a strictly increasing and concave function ranking investor s preferences of wealth Popular choices include CRRA model U(x) = x 1 γ 1 γ or CARA model U(x) = 1 1 γ e γx An economic agent over a fixed-time horizon attempts to optimally allocate his investment between risky (S t ) and risk-free (B t ) assets ds t = µs t dt + σs t dw t dx t = π t X t ds t S t + (1 π t )X t db t B t = (π t (µ r) + r)x t dt + π t σx t dw t where π t is the share of wealth allocated in stocks 22 / 32

34 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Two Optimal Control Problems Investor maximizes the expected utility of wealth at time horizon T, given initial wealth x at t: V 0 (t, x) = sup E [U(XT π X t = x)] π t 23 / 32

35 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Two Optimal Control Problems Investor maximizes the expected utility of wealth at time horizon T, given initial wealth x at t: V 0 (t, x) = sup E [U(XT π X t = x)] π t In addition to initial endowment x, investor receives k derivative contracts with payoff C(S T ): V k (t, x, s) = sup E [U(XT π + k C(S T ) X t = x, S t = s)] π t 23 / 32

36 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Two Optimal Control Problems Investor maximizes the expected utility of wealth at time horizon T, given initial wealth x at t: V 0 (t, x) = sup E [U(XT π X t = x)] π t In addition to initial endowment x, investor receives k derivative contracts with payoff C(S T ): V k (t, x, s) = sup E [U(XT π + k C(S T ) X t = x, S t = s)] π t Indifference Pricing Principle Indifference buy price pbuy k (s) and sell price pk sell (s) satisfy V 0 (t, x) = V k (t, x p k buy (s), s) V 0 (t, x) = V k (t, x + p k sell(s), s) 23 / 32

37 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Optimal Investment Problem Optimal value HJB equation { t V 0 (t, x) + sup π t {A π V 0 (t, x)} = 0 where V 0 (T, x) = U(x) A π = (π t (µ r) + r)x x π2 t σ 2 x 2 xx 24 / 32

38 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Optimal Investment Problem Optimal value HJB equation { t V 0 (t, x) + sup π t {A π V 0 (t, x)} = 0 where V 0 (T, x) = U(x) A π = (π t (µ r) + r)x x π2 t σ 2 x 2 xx Example: CARA utility γ = 0.8; GBM µ = 0.1, σ = 0.2, r = / 32

39 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Optimal Expected Utility 25 / 32

40 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Optimal Investment Problem with Option Optimal value 2D HJB equation { t V k (t, x, s) + sup π t {A π V k (t, x, s)} = 0 where V k (T, x, s) = U(x + k C(s)) A π = (π t (µ r) + r)x x π2 t σ 2 x 2 xx + µs s σ2 ss ss + πσ 2 xs xs 26 / 32

41 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Optimal Investment Problem with Option Optimal value 2D HJB equation { t V k (t, x, s) + sup π t {A π V k (t, x, s)} = 0 where V k (T, x, s) = U(x + k C(s)) A π = (π t (µ r) + r)x x π2 t σ 2 x 2 xx + µs s σ2 ss ss + πσ 2 xs xs Example continued: European put K = 2, T = 5; k = 2 26 / 32

42 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Optimal Expected Utility (t=5) 27 / 32

43 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Optimal Expected Utility (t=0) 28 / 32

44 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Indifference Buy Price 29 / 32

45 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Application of FST to solution of HJB equations General Approach: Fix π t (x) over the time-step [t 1, t 2 ] and solve the resulting PIDE Iterate to converge to the optimal policy 30 / 32

46 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Application of FST to solution of HJB equations General Approach: Fix π t (x) over the time-step [t 1, t 2 ] and solve the resulting PIDE Iterate to converge to the optimal policy Optimization Algorithm 1: Fix π t (x) = π t over the entire space F[A π ] has an analytic form and the resulting ODE can be solved explicitly (just like in option pricing) Inefficient if optimal policy is uniformly distributed in space 30 / 32

47 FST method Extensions Indifference pricing Optimal investment problems Application of FST to HJB Application of FST to solution of HJB equations General Approach: Fix π t (x) over the time-step [t 1, t 2 ] and solve the resulting PIDE Iterate to converge to the optimal policy Optimization Algorithm 1: Fix π t (x) = π t over the entire space F[A π ] has an analytic form and the resulting ODE can be solved explicitly (just like in option pricing) Inefficient if optimal policy is uniformly distributed in space Optimization Algorithm 2: Let π t (x) vary in space F[A π ] involves convolutions of F[A π ] and F[π(x)], F[π 2 (x)] Can apply a policy iteration approach of Wang and Forsyth (2006) Working in Fourier space we can solve an Integral HJB 30 / 32

48 FST method Extensions Indifference pricing Future Work Exotic, multi-asset options Mean reverting processes in energy markets HJB equations arising from optimal control problems Efficient policy iteration algorithm Optimal control with jump-diffusions 31 / 32

49 FST method Extensions Indifference pricing Thank You! 32 / 32

Pricing and Hedging of Commodity Derivatives using the Fast Fourier Transform

Pricing and Hedging of Commodity Derivatives using the Fast Fourier Transform Vladimir Surkov vladimir.surkov@utoronto.ca Department of Statistical and Actuarial Sciences, University of Western Ontario