We obtain a decomposition of the call option price for a very general stochastic volatility diffusion model, extending a previous decomposition formula for the Heston model. We realize that a new term arises when the stock price does not follow an exponential model. The techniques used for this purpose are nonanticipative. In particular, we also see that equivalent results can be obtained by using Functional Itô Calculus. Using the same generalizing ideas, we also extend to nonexponential models the alternative call option price decomposition formula written in terms of the Malliavin derivative of the volatility process. Finally, we give a general expression for the derivative of the implied volatility under both the anticipative and the nonanticipative cases.
1. Introduction
Stochastic volatility models are a natural extension of the Black-Scholes model in order to manage the skew and the smile observed in real data. It is well known that in these models the average of future volatilities is a relevant quantity. See, for example, [1, Chapter 2]. Unfortunately, adding a stochastic volatility structure makes pricing and calibration more complicated, as closed formulas do not always exist. See, for example, [2], for a first reference on this topic. Moreover, even when these formulas exist, like for the Heston model (see [3]), in general, they do not allow a fast calibration of the parameters.
During last years, different developments for finding approximations to the closed-form option pricing formulas have been published. Malliavin techniques are naturally used to solve this problem in [4, 5] as the average future volatility is an anticipative quantity. Otherwise, a nonanticipative method to obtain an approximation of the pricing formula is developed for the Heston model in [6]. The method is based on the use of the adapted projection of the average future volatility. As a result, the model allows obtaining a decomposition of the call option price in terms of such future volatility.
In the present paper, we generalize [6] to general stochastic volatility diffusion models. Similarly, following the same kind of ideas, we extend the expansion based on Malliavin calculus obtained in [4, 5]. This is important because Heston model is not the unique stochastic volatility model currently used in practice, and some of them, like SABR model, are not of exponential type. For a general discussion about stochastic volatility models in practice, see [7].
The main ideas developed in this paper are the following:
A generic call option price decomposition is found without having to specify the volatility structure.
A new term emerges when the stock option prices do not follow an exponential model, as, for example, in the SABR case.
The Feynman-Kac formula is a key element in the decomposition. It allows expressing the new terms that emerge under the new framework (i.e., stochastic volatility) as corrections of the Black-Scholes formula.
The decomposition found using Functional Itô calculus appears to be the same as the decomposition obtained through our techniques.
A general expression of the derivative of the implied volatility, both for nonanticipative and anticipative cases, is given.
2. Notations
Let S={S(t),t∈[0,T]} be a strictly positive price process under a market chosen risk neutral probability that follows the model(1)dSt=μt,Stdt+θt,St,σtρdWt+1-ρ2dBt,where W and B are independent Brownian motions, ρ∈(-1,1), μ:[0,T]×R+→R, θ:[0,T]×R+2→R+, and σ(t) is a positive square-integrable process adapted to the filtration of W. We assume on μ and σ sufficient conditions to ensure the existence and uniqueness of the solution of (1). Notice that we do not assume any concrete volatility structure. Thus, our decompositions can be adapted to many different models. In particular, we cover the following models:
Black-Scholes model: μ(t,S(t))≔rS(t), θ(t,S(t),σ(t))≔σS(t), ρ=0, r>0, and σ>0.
CEV model: μ(t,S(t))≔rS(t), θ(t,S(t),σ(t))≔σS(t)β with β∈(0,1], ρ=0, r>0, and σ>0.
Heston model: μ(t,S(t))≔rS(t), θ(t,S(t),σ(t))≔σ(t)S(t), r>0, σ>0, and(2)dσ2t=kθ-σ2tdt+νσ2tdWt,where k, θ, and ν are positive constants satisfying the Feller condition 2kθ>ν2.
SABR model: μ(t,S(t))≔rS(t), θ(t,S(t),σ(t))≔σ(t)S(t)β with β∈(0,1], r>0, σ>0, and (3)dσt=ασtdWtwith α>0.
For existence and unicity of the solution in the Heston case, see, for example, [8, Section 2.2]. For the CEV and SABR models, see [9] and the references therein.
The following notation will be used in all the paper:
We will denote by BS(t,S,σ) the price of a plain vanilla European call option under the classical Black-Scholes model with constant volatility σ, current stock price S, time to maturity τ=T-t, strike price K, and interest rate r. In this case, (4)BSt,S,σ=SΦd+-Ke-rτΦd-,where Φ(·) denotes the cumulative probability function of the standard normal law and (5)d±=lnS/K+r±σ2/2τστ.
We use in all the paper the notation Et[·]≔E[·∣Ft], where Ft,t≥0 is the natural filtration of S.
In our setting, the call option price is given by(6)Vt=e-rτEtST-K+.
Recalling that from the Feynman-Kac formula, the operator (7)Lθ≔∂t+12θt,St,σt2∂S2+μt,St∂S-rsatisfies LθBS(t,S(t),θ(t,S(t),σ(t)))=0.
We will also use the following definitions for y≥0:(8)Gt,St,y≔S2t∂S2BSt,St,y,Ht,St,y≔St∂SGt,St,y,Kt,St,y≔S2t∂S2Gt,St,y,Lt,St,y≔θt,St,ySt.
3. A Decomposition Formula Using Itô Calculus
In this section, following the ideas in [6], we extend the decomposition formula to a generic stochastic volatility diffusion process. We note that the new formula can be extended without having to specify the underlying volatility process, obtaining a more flexible decomposition formula. When the stock price does not follow an exponential process, a new term emerges. The formula proved in [6] is a particular case.
It is well known that if the stochastic volatility process is independent from the price process, then the pricing formula of a plain vanilla European call is given by(9)Vt=EtBSt,St,σ-t,where σ-2(t) is the so-called average future variance and it is defined by(10)σ-2t≔1T-t∫tTσ2sds.Naturally, σ-(t) is called the average future volatility. See [1, page 51].
The idea used in [6] consists in considering the adapted projection of the average future variance(11)v2t≔Etσ-2t=1T-t∫tTEtσ2sdsto obtain a decomposition of V(t) in terms of v(t). This idea switches an anticipative problem related with the anticipative process σ-(t) into a nonanticipative one with the adapted process v(t). We apply this technique to our generic stochastic differential equation (1).
Theorem 1 (decomposition formula).
For all t∈[0,T), we have (12)Vt=BSt,St,vt+12·Et∫tTe-ru-tGu,Su,vu·L2u,Su,σu-σ2udu+18·Et∫tTe-ru-tKu,Su,vudM,Mu+ρ2Et∫tTe-ru-tLu,Su,σu·Hu,Su,vudW,Mu,where M(t)≔∫0TEtσ2(s)ds=∫0tσ2(s)ds+(T-t)v(t)2.
Proof.
Notice that e-rTBS(T,S(T),v(T))=e-rTV(T). As e-rtV(t) is a martingale, we can write (13)e-rtVt=Ete-rTVT=Ete-rTBST,ST,vT.Our idea is to apply the Itô formula to the process e-rtBS(t,S(t),v(t)).
As the derivatives of BS are not bounded, we have to use an approximation to the identity changing BS(t,S,σ) by (14)BSnt,S,σ≔BSt,S,σψnS,where ψn(S)=ϕ(1/nS) for some ϕ∈Cb2 such that ϕ(S)=1 for all |S|<1 and ϕ(S)=0 for all |S|>2, and v(t) by vε(t)=1/(T-t)δ+∫tTEσ2(s)ds, where ε>0, and finally apply the dominated convergence theorem. For simplicity, we skip this mollifying argument across the paper.
So, applying Itô formula, using the fact that(15)∂σBSt,S,σ=S2στ∂S2BSt,S,σand the Feynman-Kac operator (7), we deduce (16)e-rTBST,ST,vT-e-rtBSt,St,vt=∫tTe-ruLvSBSu,Su,vudu+∫tTe-ru∂SBSu,Su,vuθu,Su,σu·ρdWu+1-ρ2dBu+12∫tTe-ruS2u·∂S2BSu,Su,vudMu+12∫tTe-ruS2u·∂S2BSu,Su,vuL2u,Su,σudu-σ2udu+18∫tTe-ruS2u·∂S2S2u∂S2BSu,Su,vudM,Mu+ρ2∫tTe-ruθu,Su,σu·∂SS2u∂S2BSu,Su,vudW,Mu.
Taking conditional expectation and multiplying by ert, we have(17)Ete-rT-tBST,ST,vT=BSt,St,vt+12Et∫tTe-ru-tS2u∂S2BSu,Su,vu·L2u,Su,σudu-σ2udu+18·Et∫tTe-ru-tS2u·∂S2S2u∂S2BSu,Su,vudM,Mu+ρ2Et∫tTe-ru-tθu,Su,σu∂SS2u·∂S2BSu,Su,vudW,Mu.
Remark 2.
In [6], the following operators are defined for X(t)=logS(t):
G~(t,X(t),σ(t)):=∂x2-∂xBS(t,X(t),σ(t)).
H~(t,X(t),σ(t)):=∂x3-∂x2BS(t,X(t),σ(t)).
K~(t,X(t),σ(t)):=∂x4-2∂x3+∂x2BS(t,X(t),σ(t)).
We observe the following:
G~(t,X(t),σ(t))=G(t,S(t),σ(t)).
K~(t,X(t),σ(t))=K(t,S(t),σ(t)).
H~(t,X(t),σ(t))=H(t,S(t),σ(t)).
Remark 3.
We have extended the decomposition formula in [6] to the generic SDE (1). When we apply Itô Calculus, we realize that Feynman-Kac formula absorbs some of the terms that emerge. It is important to note that this technique works for any payoff or any diffusion model satisfying Feynman-Kac formula.
Remark 4.
Note that when θ(t,S(T),σ(t))=σ(t)S(t) (i.e., the stock price follows an exponential process), then (18)Vt=BSt,St,vt+18·Et∫tTe-ru-tKu,Su,vudM,Mu+ρ2Et∫tTe-ru-tσuHu,Su,vudW,M·u,and the term(19)12Et∫tTe-ru-tS2u∂S2BSu,Su,vu·L2u,Su,σu-σ2uduvanishes.
Indeed, we show that, due to the use of Feynman-Kac formula, (20)Movement of the asset+Movement of the volatility=12θu,Su2·∂S2BSu,Su,vudu+∂σBSu,Su,vudvu=12σ2uS2u·∂S2BSu,Su,vudu+12S2u·∂S2BSu,Su,vudM+v2du-σ2du=12·S2u∂S2BSu,Su,vudM+v2du,where 1/2S2(u)∂S2BS(u,S(u),v(u))v2 is used into the Feynman-Kac formula and (21)Et∫tT12S2∂S2BSu,Su,vudM=0.
4. Basic Elements of Functional Itô Calculus
In this section, we give the insights of the Functional Itô Calculus developed in [10–13].
Let X:[0,T]×Ω↦R be an Itô process, that is, a continuous semimartingale defined on a filtered probability space (Ω,F,(Ft)t∈[0,T],P) which admits the stochastic integral representation (22)Xt=x0+∫0tμudu+∫0tσudWu,where W is a Brownian motion and μ(t) and σ(t) are continuous processes, respectively, in L1(Ω×[0,T]) and L2(Ω×[0,T]).
We define D([0,T],R) the space of cadlag functions. Given a path x∈D([0,T],R), we will denote by xt its restriction to [0,t]. For h≥0, the horizontal extension xt,h is defined as (23)xt,hu=xtu=xu,u∈0,t;xt,hu=xt,u∈t,t+hand the vertical extension is defined as (24)xthu=xtu=xu,u∈0,t;xtht=xt+h,that is xthu=xu+h1t=u.A process Y:[0,T]×R→R, progressively measurable with respect to the natural filtration of X, may be represented as (25)Yt=Ft,Xs,0≤s≤t=Ft,Xtfor a certain measurable functional F:[0,t]×D([0,t],R)→R. Let F∞ be the space of local Lipschitz functionals with respect to the norm of the supremum on D([0,t+h],R); that is, there exists a constant C>0 such that for any compact K and for any x∈D([0,t],K) and y∈D([0,t+h],K) we have(26)Ft,xt-Ft+h,yt+h≤Cxt,h-yt+h∞.
Under this framework, we have the next definitions of derivative.
Definition 5 (horizontal derivative).
The horizontal derivative of a functional F∈F∞ at t is defined as (27)DtFt,xt=limh→0+Ft+h,xt,h-Ft,xth.
Definition 6 (vertical derivative).
The vertical derivative of a functional F∈F∞ at t is defined as (28)∇xFt,xt=limh→0+Ft,xth-Ft,xth.Of course we can consider iterated derivatives as ∇xx.
We also have the following Itô formula that works for nonanticipative functionals:
Theorem 7 (Functional Itô Formula).
For any nonanticipative functional F∈F∞ and any t∈[0,T], we have(29)Ft,Xt-F0,X0=∫0tDuFu,Xudu+∫0t∇xFu,XudXu+12∫0t∇xxFu,XudX,Xu,provided that DtF, ∇xF, and ∇xxF belong to F∞.
Proof.
See [11, 12].
5. A General Decomposition Using Functional Itô Calculus
In this section, we apply Functional Itô Calculus to the problem of finding a decomposition for the call option price. The decomposition problem is an anticipative path-dependent problem. Using a smart choice of the volatility process into the Black-Scholes formula, we can convert it into a nonanticipative one. It is natural to wonder whether the Functional Itô Calculus brings some new insides into the problem.
We consider the functional (30)Ft,St,σt2=e-rtBSt,St,ft,σt2,where σ2 is the path-dependent process and f∈F∞ is a nonanticipative functional.
Under this framework, we calculate the derivatives using Functional Itô Calculus with respect to the variance. Then we write them in terms of the classical Black-Scholes derivatives. We must realize that, for simplicity, the new derivatives are calculated with respect to the variance instead of the volatility of the process.
Remark 8.
If ∂ denotes the classical derivative, we have
Alternative Vega: (31)∇σ2F=e-rt∂fBSt,St,ft,σt2∇σ2ft,σt2.
Alternative Vanna: (32)∇σ2,SF=∂f,SBSt,St,ft,σt2∇σ2ft,σt2.
Alternative Vomma: (33)∇σ2,σ2F=e-rt∂f,fBSt,St,ft,σt2∇σ2ft,σt22-e-rt∂fBSt,St,ft,σt2∇σ22ft,σt2.
Alternative Theta: (34)DtF=-re-rtBSt,St,ft,σt2+e-rt∂tBSt,St,ft,σt2+e-rt∂fBSt,St,ft,σt2Dtft,σt2.
Theorem 9 (decomposition formula).
For all t∈[0,T), S(t), and f(t,σt2)>0, we have (35)Vt=BSt,St,fu,σt2+Et∫tTe-ru-tfu,σu2τGu,Su,fu,σu2·Dufu,σu2du+12·Et∫tTe-ru-tGu,Su,fu,σu2·L2u,Su,σu-f2u,σu2du+12·Et∫tTe-ru-tfu,σu22·τ2Ku,Su,fu,σu2dfu,σu2,fu,σu2+ρEt∫tTe-ru-tL2u,Su,σu·Hu,Su,fu,σu2fu,σu2·τdWu,ft,σu2.
Proof.
Notice that F(T,X(T),σT2)=e-rTBS(T,S(T),f(T,σT2))=e-rTVT. As e-rtV(t) is a martingale, we can write (36)e-rtVt=Ete-rTVT=Ete-rTBST,ST,fT,σT2=EtFT,ST,σT2.Our idea is to apply an approximation to the identity argument as in Theorem 1 and then use the Functional Itô formula to (37)Ft,St,σt2=e-rtBSt,S,ft,σt2.
We deduce that (38)FT,ST,σT2-Ft,St,σt2=∫tTDuFu,Su,σu2du+∫tT∇SFu,Su,σu2dSu+∫tT∇σ2Fu,Su,σu2du+12∫tT∇S2Ft,St,σt2dS,Su+12∫tT∇σ22Fu,Su,σu2dσu2,σu2+12∫tTDu2Fu,Su,σu2du+∫tT∇S,σ22Fu,Su,σu2dSu,σu2+∫tT∂f∇SFu,Su,σu2dSu,fu,σu2+∫tT∂f∇σ2Fu,Su,σu2dσ2,fu,σu2.
Note that
as S(t) is not path-dependent, we have that ∇S(·)=∂S(·);
as u>t and f is a nonanticipative functional, then ∇σ2(u)f(t,σt2)=0.
So, we have (39)FT,ST,σT2-Ft,St,σt2=∫tTDuFu,Su,σu2du+∫tT∂SFu,Su,σu2dSu+12∫tT∂S2Fu,Su,σu2dS,Su+12∫tTDu2Fu,Su,σu2du+∫tT∂f,S2Fu,Su,σu2dSu,fu,σu2.We deduce that (40)FT,ST,σT2-Ft,St,σt2=∫tTLfu,σu2BSdu+∫tTe-ru∂fBSu,Su,fu,σu2Dufu,σu2du+12∫tTe-ru∂S2BSu,Su,fu,σu2·θ2u,Su,σu-S2f2u,σu2du+∫tT∂SBSu,Su,fu,σu2θu,Su,σu·ρdWu+1-ρ2dBu+12·∫tTe-ru∂f2BSu,Su,fu,σu2dfu,σu2,fu,σu2+ρ∫tTe-ru∂f,S2BSu,Su,fu,σu2·θu,Su,σudWu,fu,σu2.Taking now conditional expectations, using (15), and multiplying by ert, we obtain that (41)e-rT-tEtFT,ST,σT2=BSt,St,fu,σt2+Et∫tTe-rufu,σu2T-t·S2∂S2BSu,Su,fu,σu2Dufu,σu2du+12·Et∫tTe-ruGu,Su,fu,σu2·L2u,Su,σu-f2u,σu2du+12·Et∫tTe-rufu,σu22·τ2Ku,Su,fu,σu2dfu,σu2,fu,σu2+ρEt∫tTe-ruHu,Su,fu,σu2fu,σu2·τθu,Su,σudWu,fu,σu2.
Remark 10.
Note that Functional Itô formula proved in [11] holds for semimartingales but in [12] is also proved for Dirichlet process. In both cases, the hypothesis hold by definition of f and differentiability of the derivatives of Black-Scholes function when τ,S,σ>0. Therefore, this technique can be applied to these models.
Remark 11.
Note that Theorem 9 coincides with Theorem 1 when we choose the volatility function as f(t,σt2)=v(t). We found an equivalence of the ideas developed by [6, 10–13] in the decomposition problem. Both formulas come from different points of view; the ideas under [10–13] are based on an extension to functionals of the work [14], while the main idea of [6] is to change a process by its expectation. Realize that standard Itô Calculus also can be applied to Dirichlet processes (for more information see [14]).
Remark 12.
Realize that Theorem 9 holds for any nonanticipative f(t,σt2). It is not trivial to find a different nonanticipative process f(t,σt2) different from the one chosen in [6].
6. Basic Elements of Malliavin Calculus
In the next section, we present a brief introduction to the basic facts of Malliavin calculus. For more information, see [15].
Let us consider a Brownian motion W=W(t),t∈0,T defined on a complete probability space Ω,F,P. Set H=L2(0,T), and denote by W(h) the Wiener integral of a function h∈H. Let S be the set of random variables of the form F=f(W(h1),…,W(hn)), where n≥1, f∈Cb∞, and h1,…,hn∈H. Given a random variable F of this form, we define its derivative as the stochastic process DtWF,t∈[0,T] given by(42)DtWF=∑i=1n∂xifWh1,…,Whnhix,t∈0,T.The operator DW and the iterated operators DW,n are closable and unbounded from L2(Ω) into L2([0,T]n×Ω), for all n≥1. We denote the closure of S with respect to the norm (43)Fn,22≔FL2Ω2+∑k=1nDW,kFL20,Tk×Ω2.We denote by δW the adjoint of the derivative operator DW. Note that δW is an extension of the Itô integral in the sense that the set La2([0,T]×Ω) of square integrable and adapted processes is included in Domδ and the operator δ restricted to La2([0,T]×Ω) coincides with the Itô stochastic integral. We use the notation δ(u)=∫0Tu(t)dW(t). We recall that LWn,2≔L2[0,T];DWn,2 is contained in the domain of δ for all n≥1.
We will use the next Itô formula for anticipative processes.
Proposition 13.
Let us consider the processes X(t)=x(0)+∫0tu(s)dW(s)+∫0tv(s)ds, where u,v∈La2([0,T]×Ω). Furthermore, consider also a process Y(t)=∫tTθ(s)ds, for some θ∈L1,2. Let F:R3→R be a twice continuously differentiable function such that there exists a positive constant C such that, for all t∈[0,T], F and its derivatives evaluated in (t,X(t),Y(t)) are bounded by C. Then it follows that (44)Ft,Xt,Yt=F0,X0,Y0+∫0t∂sFs,Xs,Ysds+∫0t∂xFs,Xs,YsdXs+∫0t∂yFs,Xs,YsdYs+∫0t∂x,y2Fs,Xs,YsD-Ysusds+12·∫0t∂x2Fs,Xs,Ysu2sds,where (D-Y)(s)≔∫sTDsWY(r)dr.
Proof.
See [4].
The next proposition is useful when we want to calculate the Malliavin derivative.
Proposition 14.
Let σ and b be continuously differential functions on R with bounded derivatives. Consider the solution X=Xt,t∈[0,T] of the stochastic differential equation: (45)Xt=x0+∫0tσXsdWs+∫0tbXsds.Then, we have (46)DsXt=σXs·exp∫stσ′XsdWs+∫stλsds10,ts,where λ(s)=[b′-1/2σ′2](X(s)).
Proof.
See [15, Section 2.2].
7. Decomposition Formula Using Malliavin Calculus
In this section, we use the Malliavin calculus to extend the call option price decomposition in an anticipative framework. This time, the decomposition formula has one term less than in the Itô formula’s setup.
We recall the definition of the future average volatility as (47)σ-t≔1T-t∫tTσ2sds.
Theorem 15 (decomposition formula).
For all t∈[0,T), we have (48)Vt=EtBSt,St,σ-t+12·Et∫tTe-ruGu,Su,σ-u·L2u,Su,σu-σ2udu+ρ2·Et∫tTe-ru-tLu,Su,σuHu,Su,σ-u·∫uTDuWσ2rdrdu,where(49)Gt,St,σt≔S2t∂S2BSt,St,σt,Ht,St,σt≔St∂SGt,St,σt,Lt,St,σt≔θt,St,σtSt.
Proof.
Notice that e-rTBS(T,S(T),σ-(T))=e-rTVT. As e-rtV(t) is a martingale, we can write (50)e-rtVt=Ete-rTVT=Ete-rTBST,ST,σ-T.So, using the approximation to the identity argument and applying the Itô formula presented in Proposition 13 to(51)e-rtBSt,St,σ-t,we deduce by using (15) and (7) that (52)e-rTBST,ST,σ-T-e-rtBSt,St,σ-t=∫tTe-ruLσ-SBSu,Su,σ-udu+12·∫tTe-ruS2u∂S2BSu,Su,σ-u·θu,Su,σuSu2du-12∫tTe-ruS2u·∂S2BSu,Su,σ-uσ2udu+∫tTe-ru∂SBSu,Su,σ-uθu,Su,σu·ρdWu+1-ρ2dBu+ρ2·∫tTe-ruθu,Su,σu·∂SS2u∂S2BSu,Su,σ-u·∫uTDuWσ2rdrdu.
Taking conditional expectation and multiplying by ert, we have (53)Ete-rT-tBST,ST,σ-T=EtBSt,St,σ-t+12·Et∫tTe-ru-tGu,Su,σ-u·L2u,Su,σu-σ2udu+ρ2·Et∫tTe-ru-tLu,Su,σu·Hu,Su,σ-u∫uTDuWσ2rdrdu.
Remark 16.
As it is expected, a new term emerges when it is considered (1) like it happens in Theorem 1.
Remark 17.
In particular, when θ(t,S(T),σ(t))=σ(t)S(t), (54)Vt=BSt,St,σ-t+ρ2Et∫tTe-ru-tσu·Hu,Su,σ-u∫uTDuWσ2rdrdu.Additionally, the gamma effect is cancelled as we have seen in the Itô formula section.
Remark 18.
Note that when v(t) is a deterministic function, we have that all decomposition formulas are equal.
Remark 19.
When ρ=0, we have (55)EtBSt,St,σ-t-BSt,St,vt=12·Et∫tTe-ru-tGu,Su,σ-u-Gu,Su,vuL2u,Su,σudu-12·Et∫tTe-ru-tGu,Su,σ-u-Gu,Su,vuσ2udu-18·Et∫tTe-ru-tKu,Su,vudM,Mu.In particular, when θ(t,S(t),σ(t))=σ(t)S(t), (56)EtBSt,St,σ-t-BSt,St,vt=-18·Et∫tTe-ru-tKu,Su,vudM,Mu.The difference between the two approaches is given by the vol-vol of the option.
8. An Expression for the Derivative of the Implied Volatility
In this section, we give a general expression for the derivative of the implied volatility under the framework of Itô Calculus and Malliavin calculus. A previous calculation of this derivative in the case of exponential models by using Malliavin calculus is given in [5].
Let I(S(t)) denote the implied volatility process, which satisfies by definition V(t)=BS(t,S(t),I(S(t))). We calculate the derivative of the implied volatility in the standard Itô case.
Proposition 20.
Under (1), for every fixed t∈[0,T) and assuming that (v(t))-1<∞ a.s., we have that (57)∂SIS∗t=Et∫tT∂SF2u,S∗u,vudu∂σBSt,S∗t,IS∗t-Et∫tTF1u,S∗u,vu+∂SF3u,S∗u,vudu2S∂σBSt,S∗t,IS∗t,where (58)Et∫tTF1u,Su,vudu=12·Et∫tTe-ru-tGu,Su,vu·L2u,Su,σu-σ2udu+18·Et∫tTe-ru-tKu,Su,vudM,Mu+ρ2Et∫tTe-ru-t·θu,Su,σuSuHu,Su,vudW,M·u,Et∫tTF2u,Su,vudu=ρ2Et∫tTe-ru-t·θu,Su,σuSuHu,Su,vudW,M·u,Et∫tTF3u,Su,vudu=12·Et∫tTe-ru-tGu,Su,vu·L2u,Su,σu-σ2udu+18·Et∫tTe-ru-tKu,Su,vudM,Mu.
Proof.
Taking partial derivatives with respect to S(t) on the expression V(t)=BS(t,S(t),I(S(T))), we obtain (59)∂SVt=∂SBSt,St,IST+∂σBSt,St,IST∂SISt.On the other hand, from Theorem 1, we deduce that (60)Vt=BSt,St,vt+Et∫tTF1u,Su,vudu,which implies that (61)∂SVt=∂SBSt,St,vt+Et∫tT∂SF1u,Su,vudu.Using the fact that (v(t))-1<∞, we can check that ∂SV(t) is well defined and finite a.s. Thus, using the fact that S∗t=Kexp(r(T-t)), (59), and (61), we obtain (62)∂SIS∗t=∂SBSt,S∗t,vt-∂SBSt,S∗t,ISt∂σBSt,S∗t,ISt+Et∫tT∂SF1u,S∗u,vudu∂σBSt,S∗t,ISt.From [16], we know that ∂SI0(t)=0, where I0(t) is the implied volatility in the case ρ=0, so (63)∂SBSt,S∗t,vt=∂SBSt,S∗t,I0St-Et∫tT∂SF3u,S∗u,vudu.So, we have that (64)∂SIS∗t=∂SBSt,S∗t,I0t-∂SBSt,S∗t,IS∗t∂σBSt,S∗t,IS∗t+Et∫tT∂SF2u,S∗u,vudu∂σBSt,S∗t,IS∗t.On the other hand, we have that (65)∂SBSt,S∗t,vt=ϕd,BSt,S∗t,vt=Sϕd-ϕ-d,where ϕ is the standard Gaussian density. Then (66)∂SBSt,S∗t,vt=BSt,S∗t,vt+S2S,∂SBSt,S∗t,I0t-∂SBSt,S∗t,IS∗t=12SBSt,S∗t,I0t-BSt,S∗t,IS∗t=-12SEt∫tTF1u,S∗u,vu+∂SF3u,S∗u,vudu.
Now, we derive the implied volatility using Malliavin calculus. This is done in [5] for the case θ(t,S(T),σ(t))=σ(t)S(t).
Proposition 21.
Under (1), for every fixed t∈[0,T) and assuming that (σ~(t))-1<∞ a.s., we have that (67)∂SIS∗t=Et∫tT∂SF2u,S∗u,σ-udu∂σBSt,S∗t,IS∗t-Et∫tTF1u,S∗u,σ-u+∂SF3u,S∗u,σ-udu2S∂σBSt,S∗t,IS∗t,where (68)Et∫tTF1u,Su,σ-udu=12·Et∫tTe-ru-tGu,Su,σ-u·L2u,Su,σu-σ2udu+ρ2·Et∫tTe-ru-tLu,Su,σu·Hu,Su,σ-u∫uTDuWσ2rdrdW,M·u,Et∫tTF2u,Su,σ-udu=ρ2·Et∫tTe-ru-tL2u,Su,σu·∫uTDuWσ2rdrdW,Mu,Et∫tTF3u,Su,σ-udu=12·Et∫tTe-ru-tGu,Su,σ-u·L2u,Su,σu-σ2udu.
Proof.
See [5] or the previous proof.
Remark 22.
Note that this is generalization of the formula proved in [5]. In that case, F1=F2 and F3=0.
9. Examples
In this section, we provide some applications of the decomposition formula to well-known models in Finance.
9.1. Heston Model
We consider that the stock price follows the Heston Model (1). Using Theorem 1 or Theorem 9, we have(69)Vt=BSt,Xt,vt+ρ2·Et∫tTe-ru-tHu,Xu,vu·∫uTe-kr-sdrσ2uνdu+18·Et∫tTe-ru-tKu,Xu,vu·∫uTe-kr-sdr2ν2σ2udu.
Using Theorem 15, we have that (70)Vt=BSt,St,σ-t+ρ2·Et∫tTe-ru-tHu,Su,σ-u·∫uTDuWσ2rdrσudu,where DuWσ2(r)=νσ(u)expν/2∫ur(1/σ(s))dW(s)+∫ur-k-ν2/8σ2(s)ds.
9.2. SABR Model
We consider that the stock price follows the SABR model (3). Using Theorem 1 or Theorem 9, we have(71)Vt=BSt,St,vt+12·Et∫tTe-ru-tGu,Su,vuσ2u·S2β-1u-1du+18·Et∫tTe-ru-tKu,Su,vudM,Mu+ρ2Et∫tTe-ru-tHu,Su,vuσudW,M·u,where(72)dM,M=4α2σ4t∫tTeα2s-tds2dt,dM,W=2ασ2t∫tTeα2s-tdsdt.
Using Theorem 15, we have that (73)Vt=EtBSt,St,σ-t+12·Et∫tTe-ru-tGu,Su,σ-uσ2u·S2β-1u-1du+ρ2·Et∫tTe-ru-tHu,Su,σ-u·∫uTDuWσ2rdrσudu,where DuWσ2(r)=2ασ2(u)1[0,r](u).
10. Conclusion
In this paper, we notice that the idea used in [6] can be used for a generic Stochastic Differential Equation (SDE). There is no need to specify the volatility process. Only existence and uniqueness of the solution of the SDE are needed, allowing much more flexibility in the decomposition formula. We observe what is the effect of assuming that the stock price follows an exponential process and how a new term arises in a general framework. Additionally, we have computed the decomposition using three different methods: Itô formula, Functional Itô Calculus, and Malliavin calculus. In the case of call options, the idea used in [6] is equivalent to the use of the Functional Itô formula developed in [10–13] but without the need of the theory behind the Functional Itô Calculus. Both formulas can be applied to Dirichlet process, in particular, to the fractional Brownian motion with Hurst parameter equal or bigger than 1/2. Furthermore, we realize that the Feynman-Kac formula has a key role in the decomposition process.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
FouqueJ.-P.PapanicolaouG.SircarK. R.2000Cambridge, UKCambridge University PressMR1768877HullJ.WhiteA.The pricing of options on assets with stochastic volatilities198742228130010.1111/j.1540-6261.1987.tb02568.xHestonS. L.A closed for solution for options with stochastic volatility with application to bond and currency options19936232734310.1093/rfs/6.2.327AlòsE.A generalization of the Hull and White formula with applications to option pricing approximation200610335336510.1007/s00780-006-0013-5MR22443502-s2.0-33747875057AlòsE.LeònJ. A.VivesJ.On the short-time behavior of the implied volatility for jump-diffusion models with stochastic volatility200711457158910.1007/s00780-007-0049-1MR23358342-s2.0-34548107274AlòsE.A decomposition formula for option prices in the Heston model and applications to option pricing approximation201216340342210.1007/s00780-012-0177-0MR29466132-s2.0-84862585631GatheralJ.2006New York, NY, USAJohn Wiley & SonsGulisashviliA.2012Berlin, GermanySpringerSpringer Finance10.1007/978-3-642-31214-4MR2976192ChenB.OosterleeC. W.van der WeideH.Efficient unbiased simulation scheme for the SABR stochastic volatility model2011152ContR.FourniéD.A functional extension of the Ito formula20103481-2576110.1016/j.crma.2009.11.013MR25867442-s2.0-74249122962ContR.FourniéD.-A.Change of variable formulas for non-anticipative functionals on path space201025941043107210.1016/j.jfa.2010.04.017MR26521812-s2.0-77952953212ContR.FourniiéD.-A.Functional Itô calculus and stochastic integral representation of martingales201341110913310.1214/11-aop721MR30591942-s2.0-84874965188DupireB.Functional itô calculus20092009-04-FRONTIERShttp://papers.ssrn.com/sol3/papers.cfm?abstract_id=14355511435551FöllmerH.Calcul d'ito sans probabilites1981850Berlin, GermanySpringer143150Lecture Notes in Mathematics10.1007/BFb0088364NualartD.20062ndSpringerMR2200233RenaultE.TouziN.Option hedging and implied volatilities in a stochastic volatility model19966327930210.1111/j.1467-9965.1996.tb00117.xZBL0915.900282-s2.0-0030557781