Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model Numerical option pricing for the Barndorff-Nielsen Shephard stochastic volatility model A utility indifference pricing approach Martin Groth martijg@math.uio.no Bachelier World Congress Tokyo, Japan, August 2006 1(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model Outline I The model I The problem I The equations I The numerics I The results I General risk aversion case The model 2(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The model The Barndorff-Nielsen - Shephard model Stochastic volatility model proposed by Barndorff-Nielsen Shephard [BNS01] dSt = α(Yt )St dt + σ(Yt )St dBt , S0 = s > 0 dYt = −λYt dt + dLλt , Y0 = y > 0 dRt = 0, α(y ) = µ + βy , σ(y ) = √ R0 = 1 y on the complete filtered probability space (Ω, F, Ft , P) where {Ft }t≥0 is the completion of the filtration σ(Bs , Lλs ; s ≤ t). 3(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The problem The indifference pricing problem Investor trying to maximise her utility, having exponential utility function U(x) = 1 − exp(−γx). Two different strategies in the market I Buy the underlying asset with the initial wealth x I Issue a claim f (ST ) with price Λ(γ) (t, y , s) and invest the incremental wealth x + Λ(γ) in the market The indifference price of the claim is when the investor is indifferent between the investment alternatives. In the zero risk aversion limit the indifference price corresponds to pricing under the minimal entropy martingale measure (MEMM). 4(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The problem Value functions If the value functions for the two choices are V 0 (t, x, y ) = sup E[1 − exp(−γXT )|Xt = x, Yt = y ] π∈At V (t, x, y , s) = sup E[1 − exp(−γ(XT − f (ST )))|Xt = x, Yt = y , St = s] π∈At the utility indifference price of the claim is the unique solution Λ(γ) s.t. V 0 (t, x, y ) = V (t, x + Λ(γ) (t, y , s), s, y ) The problem is studied in Benth and Meyer-Brandis [BMB05] who use a dynamic programming approach to derive the associated Hamilton-Jacobi-Bellman equations for the value functions of the investor. 5(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The equations Integro-PDE for value function without a claim issued Assuming V 0 (t, x, y ) = 1 − exp(−γx)H(t, y ) we get the integro-PDE Ht (t, y ) − α2 (y ) H(t, y ) − λyHy (t, y ) 2σ 2 (y ) Z ∞ +λ {H(t, y + z) − H(t, y )} ν(dz) = 0 0 Together with the terminal condition H(T , y ) = 1 the integro-PDE has a solution H(t, y ) ∈ C 1,1 ([0, T ] × R+ ) and allow the Feynman-Kac representation " 1 H(t, y ) = E exp − 2 Z t T ! # α2 (Yu ) du Yt = y σ 2 (Yu ) 6(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The equations Integro-PDE for the option price In the zero risk aversion limit the option price Λ(t, y , s), for (t, y , s) ∈ [0, T ) × R+ × R+ , is governed by the integro-PDE: 1 Λt + σ 2 (y )s 2 Λss − λy Λy 2 Z ∞ (Λ(t, y + z, s) − Λ(t, y , s)) +λ 0 H(t, y + z) ν(dz) = 0 H(t, y ) The terminal condition Λ(T , y , s) = f (s) yields the Feynman-Kac representation eT )|Y et = y , S et = s] Λ(t, y , s) = E[f (S 7(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The equations New stochastic processes The stochastic processes under the Minimal entropy martingale measure are now et dS et dY et )S et dB et , = σ(Y ft dt + de = −λY Lλt where e Lt is a pure jump Markov process with jump measure νe(ω, dz, dt) = et (ω) + z) H(t, Y ν(dz) dt et (ω)) H(t, Y 8(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model Numerical difficulties Two spatial dimensions The numerics 9(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model Numerical difficulties Numerical approximation of the integral The numerics 9(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model Numerical difficulties Need to calculate H(t, y ) first before the option price The numerics 9(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model Numerical difficulties Infinite mass of the Lévy measure around zero The numerics 9(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model Boundary conditions The numerics 10(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The results Preparing an example Use parameters from Nicolato and Venardos [NV03] and assume that the stationary distribution of the Yt process is inverse Gaussian. The corresponding Lévy measure of Lt is then 1 2 δ −3/2 2 (1 + γ z) exp − γ z dz ν(dz) = √ z 2 2 2π The log-marginal stock returns are then approximately Normal inverse Gaussian (NIG) distributed. 11(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The results H(t, y ) and the measure change fraction The H(t, y ) function is essential in the measure change for the MEMM and needs to be simulated before the option prices. The fraction in the measure change says something about how we weight jump sizes under the new measure. 1.02 1.01 H(t,y+z)/H(t,y) 1 0.99 0.98 0.97 0.96 0 0.5 1 z 1.5 12(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The results The option prices and the relation to Black & Scholes 120 100 MEMM price 80 60 40 20 0 300 250 0.5 200 0.4 0.3 150 0.2 100 0.1 50 S 0 y Price difference Black−Scholes price with variance=0.007333 minus MEMM price with y=0.00733 0.3 0.25 0.2 Price difference The option prices conform to earlier results for exponential Lévy models with NIG distributed marginal returns. The difference between the BNS-prices and Black & Scholes prices display the characteristic W-shape reported by Eberlein et.al. 0.15 0.1 0.05 0 −0.05 −0.1 0.4 0.6 0.8 1 1.2 stockprice−strike ratio 1.4 1.6 1.8 13(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model The results 14(20) Volatility smile for the MEMM-prices 0.16 0.15 0.14 Implied Black−Scholes volatility As expected, with approximately NIG-distributed marginal returns, the model gives a skewed implied volatility smile. 0.13 0.12 0.11 0.1 0.09 0.08 0.5 0.6 0.7 0.8 0.9 1 1.1 spotprice−strike ratio 1.2 1.3 1.4 1.5 Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model General risk aversion Taking it further Is the market pricing under MEMM, or if not, what is the risk aversion of the investors in the market? 15(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model General risk aversion 16(20) PDE for the general risk aversion If we let Λ(γ) (t, y , s) = γ1 ln h(γ) (t, y , s) we can find the option price by solving the following PDE 1 (∂s h(γ) )2 1 + LY h(γ) = 0 ∂t h(γ) + ys 2 ∂ss h(γ) − ys 2 2 2 h(γ) where Z LY h(t, y ) = −λy ∂y h+λ ∞ {h(t, y +z)−h(t, y )} 0 and initial condition h(γ) (T , y , s) = exp(γf (s)). H(t, y + z) ν(dz) H(t, y ) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model General risk aversion 16(20) PDE for the general risk aversion If we let Λ(γ) (t, y , s) = γ1 ln h(γ) (t, y , s) we can find the option price by solving the following PDE 1 (∂s h(γ) )2 1 + LY h(γ) = 0 ∂t h(γ) + ys 2 ∂ss h(γ) − ys 2 2 2 h(γ) where Z LY h(t, y ) = −λy ∂y h+λ ∞ {h(t, y +z)−h(t, y )} 0 and initial condition h(γ) (T , y , s) = exp(γf (s)). H(t, y + z) ν(dz) H(t, y ) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model General risk aversion What is the market risk aversion? We look at Microsoft options, bid/ask prices, and historical stock prices. Using the number of trades as a measure of the volatility, the method by Lindberg [Lin05] estimates the parameters for the NIG-distribution and BNS-model. I How do the market data compare to MEMM prices simulated with these parameters? I Can we see if the market has any preference with regards to the risk aversion? 17(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model General risk aversion Comparison between market and MEMM prices Comparing implied Black & Scholes volatility for market and MEMM prices to see if the market is consistently in favour of one of the actors. 18(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model General risk aversion Risk aversion in the market We implement a search method and solve the PDE iteratively to find the risk aversion parameter γ that results in the market price. Aversion, Microsoft options, May 5, 2006 2.5 May 19, 2006 June 16, 2006 July 21, 2006 Sept 20, 2006 Jan 19, 2007 Jan 18, 2008 Aversion 2 1.5 1 0.5 5 10 15 20 25 Strike 30 35 40 45 19(20) Numerical option pricing for the Barndorff-Nielsen - Shephard stochastic volatility model General risk aversion Fred Espen Benth and Thilo Meyer-Brandis. The density process of the minimal entropy martingale measure in a stochastic volatility model with jumps. Finance and Stochastics, 9(4), 2005. Ole E. Barndorff-Nielsen and Neil Shepard. Non-Gaussian Ornstein-Uhlenbeck-based models and some of their uses in financial economics. J. the Royal Statistical Society, 63:167–241, 2001. Carl Lindberg. The estimation of a stochastic volatility model based on the number of trades. Submitted, 2005. Elisa Nicolato and Emmanouil Venardos. Option pricing in stochastic volatility models of the Ornstein-Uhlenbeck type. Math. Finance, 13(4):445–466, 2003. 20(20)