Journal of Mathematical Finance, 2011, 1, 90-97 doi:10.4236/jmf.2011.13012 Published Online November 2011 (http://www.SciRP.org/journal/jmf)
Stochastic Volatility Jump-Diffusion Model for Option Pricing
Nonthiya Makate, Pairote Sattayatham School of Mathematics, Suranaree University of Technology, Nakhon Ratchasima, Thailand E-mail: [email protected], [email protected] Received July 26, 2011; revised September 2, 2011; accepted September 15, 2011
Abstract
An alternative option pricing model is proposed, in which the asset prices follow the jump-diffusion model with square root stochastic volatility. The stochastic volatility follows the jump-diffusion with square root and mean reverting. We find a formulation for the European-style option in terms of characteristic functions of tail probabilities.
Keywords: Jump-Diffusion Model, Stochastic Volatility, Characteristic Function, Option Pricing
1. Introduction work by incorporating jumps in volatility and their model is given by Let ,, P be a probability space with filtration SS dddSStt tvWSYN tt tt dt, (3) t 0tT. All processes that we shall consider in this dddvvtvWZ vv dN. section will be defined in this space. An asset price ttttv t model with stochastic volatility has been defined by Eraker et al. [3] developed a likelihood-based estima- Heston [1] which has the following dynamics: tion strategy and provided estimates of parameters, spot dddSS tvWS , (1) volatility, jump times, and jump sizes using S&P 500 and tt tt Nasdaq 100 index returns. Moreover, they examined the v volatility structure of the S&P and Nasdaq indices and ddvvtv dW, tttt indicated that models with jumps in volatility are pre- where S is the asset price, is the rate of return of t ferred over those without jumps in volatility. But they the asset, vt is the volatility of asset returns, 0 is did not provide a closed-form formula for the price of a a mean-reverting rate, is the long term variance, S v European call option. 0 is the volatility of volatility, Wt and Wt are In this paper, we would like to consider the problem of standard Brownian motions corresponding to the proc- finding a closed-form formula for a European call option esses St and vt , respectively, with constant correlation where the underlying asset and volatility follow the Model . In 1996, Bate [2] introduced the jump-diffusion sto- (3). This formula will be useful for option pricing rather chastic volatility model by adding log normal jump Yt than an estimation of it as appeared in Eraker’s work. to the Heston stochastic volatility model. In the original The rest of the paper is organized as follows. In Sec- formulation of Bate, the model has the following form: tion 2, we briefly discuss the model descriptions for the ddSS tvdWSYNS dS, (2) option pricing. The relationship between stochastic dif- tt tt ttt ferential equations and partial differential equations for ddvvtv dWv , the jump-diffusion process with jump stochastic volatil- tttt ity is presented in Section 3. Finally, a closed-form for- S where Nt is the Poisson process which corresponds to the mula for a European call option in terms of characteristic underlying asset St , Yt is the jump size of asset price functions is presented. return with log normal distribution and St means that there is a jump the value of the process before the jump 2. Model Descriptions is used on the left-hand side of the formula. Moreover, in 2003, Eraker Johannes and Polson [3] extended Bate’s It is assumed that a risk-neutral probability measure
Copyright © 2011 SciRes. JMF N. MAKATE ET AL. 91
2 exists, the asset price St under this measure follows a jump- Nt Poisson random measure of the process R Z with diffusion process, and the volatility vt follows a pure mean t n n1 reverting and square root diffusion process with jump, i.e. 2 our models are governed by the following dynamics: intensity measure Z ddrt.
Let Uxx 12, be a bounded real-valued function and dddSSr SS mtv W SYNdS, (4) tttt ttt twice continuously differentiable with respect to x1 and x2 and ddvvtWZN vddvv , ttt ttt 12 1 2 uxx12,, t EUXTT , X X t xX1 , t x 2(6) S v where St , vt , , , , Wt and Wt are defined as S in Bate’s model, r is the risk-free interest rate, Nt and By the two dimensional Dynkin formula [5], u is a v Nt are independent Poisson processes with constant in- solution of the partial integro-differential equation (PIDE) S v tensities and respectively. Yt is the jump size uxx 12,, t of the asset price return with density Y y and 0,,uxx t t 12 EY t :m and Zt is the jump size of the volatility with density z . Moreover, we assume that the jump (1) Z ux yxt,, uxx ,, t y d y S v 12 12Y processes Nt and Nt are independent of standard S v Brownian motions Wt and Wt . (2) uxx12,, zt uxx 12 ,,d tZ z z 3. Partial Integro-Differential Equations
subject to the final condition uxxT12,, Uxx 1 ,2 . Consider the process X XX12, where X 1 ttt t The notation is defined by 2 and X t are processes in and satisfy the following equations: uxx 12,, t uxx12 ,, t Au x12,, x t f 1 f 2 112 12 1 xx12 d,,,,X tttf11XXtd tg XXtW ttdt (5) 22 11 1 2 uxx12,, t uxx12,, t XYtt dNt gg112 g2 2 x1 xx12 21212 2 d,,,,Xttttt f22XXtd tg XXtWdt 2 1 2 uxx12,, t 2 g2 2 ZNttd 2 x2 where f112,,gf and g2 are all continuously differenti- (7) 1 2 able, Wt and Wt are standard Brownian motions 12 1 2 with Corrd, WttdW , Nt and Nt are 4. A Closed-Form Formula for the Price of a European Call Option independent Poisson processes with constant intensities 1 2 and respectively. Let C denote the price at time t of a European style call Since every compound Poisson process can be repre- option on the current price of the underlying asset St sented as an integral form of a Poisson random measure with strike price K and expiration time T . [4] then the last term on the right hand side of (5) can be The terminal payoff of a European call option on the written as follows: underlying stock St with strike price K is 1 ttNt 11 1 1 max SKT ,0 . X ssYNdd s XY nnX s qJ Q sqd, 00n1 This means that the holder will exercise his right only 2 ttNt 2 if SKT and then his gain is SKT . Otherwise, if ZssddNZrJs n Rdr, SKT , then the holder will buy the underlying asset 00n1 from the market and the value of the option is zero. where Yn are i.i.d. random variables with density Y y Assuming the risk-free interest rate r is constant 1 and NJtQ is a Poisson random measure of the process over the lifetime of the option, the price of the European 1 Qtn Ywith intensity measure Y ddqt, Zn are call at time t is equal to the discounted conditional n1 i.i.d. random variables with density Z z , and J R is a expected payoff
Copyright © 2011 SciRes. JMF 92 N. MAKATE ET AL.
function of the stock log-return L denoted by CSvtKTtt,,;, t rT t CLvtkT ,,;, CeLt ,, vteT;,k eE max SKSvTtt ,0 , tt t
ln St ln K rT t Ce,,; vtet , T e SPSSvST T tt,d T KPSSvS T tt ,d T KK CStt,,v tKT;,, 1 i.e., SSPSSvS,d tTTttTrT t rT t L e S ClvtkT ,,;, e E maxeT K,0 L lv , v t K tt rT() t and satisfies the following PIDE: Ke P (,)d STtt S v S T K C 0, ClvtkT ,;, 1 t SStTP (,)dSTStvtST S Cl yvtkT,,;, Cl ,,;, vtkT y d y ESSv[|,]TttK Y KerT() t PSSvS(,)d v Ttt T Clv,,;, ztkTCl ,,;, vtkTZ z d z K rT() t SPSvtKTttt12(,,;,) Ke PSvtKT (,,;,)tt (10) Here the operator as in (7) is defined by (8) S 1 CC ClvtkT,,;, r m v v where E is the expectation with respect to the risk- 2 lv 22 neutral probability measure, P SSvTtt, is the cor- 1 CC vv2 responding conditional density given Svtt, and 2 l lv 2 1 2 C vr2 C PSvtKT1 tt,,;, SPT S T Sv tt, d S T E S T Sv tt, 2 v K In the current state variable, the last line of (8) becomes l kr Tt Note that P1 is the risk-neutral probability that C lvtkT,,;, eP12 lvtkT ,,;, e P lvtkT,,;, SKT (since the integrand is nonnegative and the inte- (11) gral over 0, is one), and finally that lk where PlvtkTjj ,,;, : PevteT ,,; , , j 1,2. PSvtKT,,;, P SSv, d S The following lemma shows the relationship between P1 2 tt T tt T K and P2 in the option value of (11). Pr ob(SKSvTt | ,t ) Lemma 1 The functions P and P in the option is the risk-neutral in-the-money probability. Moreover, 1 2 value of (11) satisfy the following PIDEs ESSv,e rT t S for t 0 . Ttt t PP 11 0,PlvtkTv1 ,;, Assume that the asset price St and the volatility vt tl satisfy (4), we would like to compute the price of a P European call option with strike price K and maturity vrmP1 S v 1 T . To do this, we make a change of variable from St to L ln S, i.e. where S satisfies (4) and its inverse Sy tt t ePlyvtkTy1,,:,1 Y dy Lt St e. Denote k ln K the logarithm of the strike price. By the jump-diffusion chain rule, ln S satisfies t and subject to the boundary condition at expiration time the SDE tT ; v SSt S dSln ttt r m d t vW d ln 1 YNt d t(9) PlvTkT1 ,, ;, 1lk (12) 2 moreover, P satisfies the equation Applying the two-dimensional Dynkin formula [5] for 2 the price dynamics (9) and volatility v in system (4), P2 t 0, PlvtkTrP,;,, we obtain the value of a European-style option, as a t 22
Copyright © 2011 SciRes. JMF N. MAKATE ET AL. 93
and subject to the boundary condition at expiration time 2) The characteristic function f2 is given by tT ; , f22lvtxt,,;, exp g vh2 ixl r PlvTkT2 ,, ;, 1lk , (13) 22 The operator is defined by e 2 1 22 where h2 , flvtkT,,;, 2 e 2 22 22 2 S 11fff :rmv v v2 S 22lvl g2 rmix r
2 2 f 1 2 f vvrf (14) 1 e 2 2 22 lv 2 v 2ln 1 2 2 22 S 2 f lyvtkTflvtkT,,;, ,,;,Y yy d v Sixyeyye1d vzh2 1 zdz f lv,,;, ztkT f lvTkT ,,;,Z z d z. YZ ix Note that 1lk 1 if lk and otherwise 10lk . 2
The following lemma shows how to calculate the 22 and 22ix ix 1 . functions P1 and P2 as they appeared in Lemma 1. In summary, we have just proved the following main Lemma 2 The functions P1 and P2 can be calculated theorem. by the inverse Fourier transforms of the characteristic Theorem 3 The value of a European call option of (4) is function, i.e. kr Tt ixk l eflvtkT,,;, C lvtkT,,;, eP12 lvtkT ,,;, e P lvtkT,,;, 11 j PlvtkTj ,,;, Re d x, 2 π 0 ix where P1 and P2 are given in Lemma 2. for j 1, 2 with Re[ ] denoting the real component of a complex number. By letting Tt. 5. Conclusions 1) The characteristic function f is given by 1 This paper has proposed asset price dynamics to accom- modate both jump-diffusion and jump stochastic vola- f11 lvtxt,,;, exp g vh1 ixl, tility. Under this proposed model, an analytical solution where is derived for a European call option via the characteris- 221 tic function. 11e 1 h1 2 1 11 11 e 6. Acknowledgements
SS This research is (partially) supported by The Centre of g1 rmixm Excellence in Mathematics, the Commission on Higher Education (CHE). 1 e 1 11 2ln 1 Address: 272 Rama VI Road, Ratchathewi District, 2 11 21 Bangkok, Thailand. 7. References S ix1 y eyy1d Y , [1] S. L. Heston, “A Close Form Solution for Options with Stochastic Volatility with Applications to Bond and Cur- v ezzzh1 1d Z rency Options,” The Review of Financial Studies, Vol. 6, No. 2, 1993, pp. 327-343. doi:10.1093/rfs/6.2.327 ix 1 [2] D. Bates, “Jump and Stochastic Volatility: Exchange Rate 1 Processes Implicit in Deutsche Mark in Options,” Review
22 of Financial Studies, Vol. 9, No. 1, 1996, pp. 69-107. and 11 ix ix 1 . tp://dx.doi.org/10.1093/rfs/9.1.69
Copyright © 2011 SciRes. JMF 94 N. MAKATE ET AL.
[3] B. Eraker, M. Johannes and N. Polson, “The Impact of Processes,” CRC Press, Boca Raton, 2004. Jumps in Volatility and Returns,” The Journal of Finance, [5] F. B. Hanson, “Applied Stochastic Process and Control for Vol. 58, No. 3, 2003, pp. 1269-1300. Jump Diffusions: Modeling, Analysis and Computation,” doi:10.1111/1540-6261.00566 Society for Industrial and Applied Mathematics, Phila- [4] R. Cont and P. Tankov, “Financial Modeling with Jump delphia, 2007
Copyright © 2011 SciRes. JMF N. MAKATE ET AL. 95
Appendix ePl lvtkT , , ; , ekrTt P lvtkT,,;, 12 Proof of Lemma 1. We plan to substitute (11) into (10). ly e eP11 l yvtkT,,;, P l yvtkT ,,;, Firstly, we compute ePll l yvtkT,,;, eP lvtkT ,,;, C PP 11 eel 12krTt rekrTtP 2 krTt tt t e P11 l yvtkT,,;, P lvtkT ,,;, ly C llP12krTtP ee1,,;, PlyvtkT eePe 1 ll1 l l eP1 l yvtkT,,;, PlvtkT1 ,,;, C l PP12krTt ee ekrTt P l yvtkT ,,;, P lvtkT ,,;, vvv 21 and 2C 22PP P eeePelll112 krTt 2 Clv , ztkT ,;, ClvtkT ,,;, ll22l 1 l2 l krTt eP12 lv, ztkT ,;, e P lv, ztkT ,;, 2 22 C llPP11krTt P2 eee ePl lvtkT,,;, ekrTt P lvtkT,,;, lv lv v lv 12 l 2 22 e P lv, ztkT ,;, P lvtkT ,,;, C l PP12krTt 11 22ee 2, vv v krTt e P22 lv, ztkT ,;, P lvtkT ,,;, Cl yvtkT,,;, ClvtkT ,,;, We substitute all terms above into (10) and separate it by assumed independent terms of P and P . This ePlyvtkTly ,,;, 1 2 1 gives two PIDEs for the risk-neutralized probability for ePlyvtkTkrTt ,,;, 1 PlvtkTj (, ,; , ), j 1,2: (Equation (15))
2 PPPP11111S 11P 0()2rmv P11 v v2 P tlvl22 l
22PP1 P vvr11 2 1PeSy1Ply ,v ,;tk ,TP ly ,v ,;tk ,TP (,,;lvtk ,T ) y dy (15) 2 11 1 1Y lv v 2 v v Plv1 , ztkT,; , P1 lvtkT , ,; ,Z z d z subject to the boundary condition at the expiration time 22 2 11PP222 P 2 tT according to (12). vv22 vrP 2 22lvlv By using the notation in (14), PIDE (15) becomes S P22 l yvtkT,,;, P lvtkT ,,;,Y y d y PP11 0, PlvtkTv ,;, tl1 v PlvztkT, ,;, PlvtkT ,,;, z d z P1 S 22Z vrmP 1 v (16) SyePlyvtkTy1,,;, dy 1 Y subject to the boundary condition at the expiration time tT according to (13). Again, by using the notation in P :,1 PlvtkT ,;, (14), PIDE (16) becomes t 1 P 2 For PlvtkT,,;, : 0[](,,;,) PlvtkTrP22 2 t PPS 1 P 22 2 P2 0 rP2 r m v v :[](,,; PlvtkT ,). tl2 v t 22
Copyright © 2011 SciRes. JMF 96 N. MAKATE ET AL.
The proof of Lemma 1 is now completed. and
For j 1, 2 the characteristic functions for yygvhix11 ly eflyvtxtee1,,;,1 1 PlvtkTj ,,;, , yixy eeflvtxt1,,;,1 with respect to the variable k are defined by Substituting all the above terms into (17) and after ixk canceling the common factor of f , we get a simplified f jjlvtxT,,;, : e d P lvtkT ,,;, , 1 form as follows: with a minus sign to account for the negativity of the 1 S 0 g11 vh r m v ix measure dPj . 2
Note that f j also satisfies similar PIDEs 1 2 vvhvx 1 2 f j flvtxT,,;, 0, (17) 1 22 S t jj vixh vh m 112 with the respective boundary conditions S eyyix1 y 1d Y ixk f jjlvtxT,,;, e d P lvtkT ,,;, v zh1 ezz1d.Z eklixk dk By separating the order v and ordering the remaining eixl terms, we can reduce it to two ordinary differential equa- tion (ODEs), since d,,;,d1dPlvTkTjl k Hlk k ld k. 1122 12 hh11 1 ixhix 1 x(19) Proof of Lemma 2 222 1) To solve for the characteristic function explicitly, and letting Tt be the time-to-go, we conjecture that SS g11 hrmix m the function f1 is given by ix 1 y S ey 1dy (20) f11lvtxt,,;, exp g vh1 ixl (18) Y and the boundary condition v zh1 ez1d.Z z gh1100 0.
This conjecture exploits the linearity of the coefficient Let 1 ix 1 and substitute it into (19). We get in PIDE (17). 11222 Note that the characteristic function f1 always exists. 1 hhhixix111221 In order to substitute (18) into (17), firstly we com- 2 pute 1 24422ix ix 1 2 11 f f h1 2 1 1 2 2 g11 vh f 1 ixf1 t l 2 24422ix ix 1 f1 f1 2 11 h f x f h1 11 2 1 2 2 v l 2 2 f1 f1 2 1 2 11 11 ixh1 f1 2 hf11 hh1122 lv v 2
f1 l yvtxt,,;, f1 lvtxt ,,;, 22 where 11ix ix 1 . ixy eflvtxt1,,;,1 By the method of variable separation, we have
2dh1 2 f11 lv, ztxt ,;, f lvtxt ,,;, d . 11 11 zh1 hh eflvtxt1,,;, 1122 1
Copyright © 2011 SciRes. JMF N. MAKATE ET AL. 97
ixk Using partial fractions, we get 11 eflvtkT,,;, PlvtkT ,,;, Rej d x(21) j 2 π ix 11 1 0 ddh . 1 for j 1, 2 . 1 hh11 11 1122 To verify (21), firstly we note that Integrating both sides, we obtain Eeixln St ln K ln SLvv , ttt 11 h 1 2 ix lk ln C . Eeix Lt k Llvv,d, e PlvtkT ,;, 1 tt j 11 h1 2 eePlvtkTeeixk ixld,,;, ixk ixk lkk d Using boundary condition h1 (0)0 , we get j
11 ixk C ln . eflvtkTj ,,;, . 11 Then Solving for h1 , we obtain ixk 11 eflvtkTj ,,;, 22e1 1 Re dx 11 2 π ix h1 . 0 2 1 11 11 e ixln St ln K Eeln SLvv , In order to solve g1 explicitly, we substitute h1 11 ttt Re dx into (20) and integrate with respect to on both sides. 2 π ix 0 Then we get
SS ixlk grmixm1 11 e ExLlv Re dtt , v 2 π ix 1 0 11 1 e 2ln 1 2 2 11 1 11 sin xl k ExLlvd,v tt S ix1 y 2 π x eyy1d 0 Y 11 sin x v zh1 Elksgn d,xL lv v ezz1d Z tt 2 π x 0 Proof of 2). The details of the proof are similar to case 1). Hence, we have 11 ElkLlvv sgn tt , 22 f22lvtxt,,;, exp g vh2 ixl r
ELlvv 1,lk t t where gh22 ,, 2 and 2 are as given in the Lemma. We can thus evaluate the characteristic functions in sin x where we have used the Dirichlet formula d1x explicit form. However, we are interested in the risk-neu- x 0 tral probabilities Pjj ,1,2 . These can be inverted from the characteristic functions by performing the following and the sgn function is defined as sgnx 1 if x 0 , integration 0 if x 0 and –1 if x 0 .
Copyright © 2011 SciRes. JMF