This action might not be possible to undo. Are you sure you want to continue?

Frédéric Abergel, Rémi Tachet November 17, 2009 Ecole Centrale Paris, 92295 Châtenay-Malabry, France Electronic addresses: frederic.abergel@ecp.fr, remi.tachet@ecp.fr

Introduction

Financial modelling has been an area of extremely rapid growth in the past 30 years, and some extremely interesting mathematical challenges have emerged. One of the utmost importance for real-life applications to derivatives trading is that of calibration. Similar to common situations in many areas of physics and engineering, once a model has been suggested, its parameters have to be estimated using external data. In the case of derivative modelling, those data are the liquid (tradable) options, generally known as the “vanilla” products. It is well known since the pioneering work of Litzenberger and Breeden [1] and its celebrated extension by Bruno Dupire [3] that the knowledge of market data such as the prices of vanilla options across all strikes and maturities is equivalent to the knowledge of the risk-neutral marginals of the underlying stock distribution, and moreover, that there is a unique one-dimensional driftless diffusion which recovers exactly such marginals. However, it has also been well-known for almost as many years that the evolution in time of the so-called “local volatility” is not stable, thereby leading researchers and ﬁnancial engineers to look for a more robust, stochastic volatility type of modelling. In this paper, we consider the calibration problem for a generic stochastic volatility model: more precisely, we address the issue of calibrating to market data a generic model with a stochastic component and a local component for the volatility process. Such models are very useful in practice, since they offer both the ﬂexibility and realistic dynamics of stochastic volatility models, and the exact calibration properties of local volatility models. In mathematical terms, the problem we consider is a non linear partial integrodifferential equation for which we are able to prove short-time existence of classical solutions under suitable assumptions. The paper is organized as follows: Section 1 is devoted to the mathematical formulation of the problem. Section 2, to notations and statement of the main result. In Section 3, we recall some important technical results stemming from the general theory of parabolic PDE’s. Section 4 contains the proof of the main result. Finally, Section 5 is a short conclusion.

1

The Local and Stochastic Volatility model and its calibration

The LSV model is an extension of the Dupire local volatility model. In the simplest situation - the two-dimensional case - the dynamics of the model are given by the following system of SDE’s dSt = a(t , St )b(Yt )dBt1 + µt dt St dYt = α(t , Yt )dBt2 + ξt dt

1

Electronic copy available at: http://ssrn.com/abstract=1508490

T [ × ∂Ω the different parts of the boundary. T [ × Ω the domain of deﬁnition and by B = {0} × Ω. ΩS y = {y ∈ R n ∂p ∂2 ∂2 2 − 2 (ρ11 α1 I ( p) p) − ∑ (ρ1i α1 αi O( p) := ∂t ∂S i=2 ∂S∂xi n I ( p) p) (1) − n ∂2 ∂ (ρi j αi α j p) + ∑ (βi p) + γ p = 0 on DT ∪ BT i.com/abstract=1508490 . The function b simply transforms that factor into a proper volatility. This gives us the value of a2 (t . xn ) ∈ Ω ⊂ Rn the n-dimensional space variable where Ω is an open subset with a sufﬁciently smooth boundary (we will precise this notion later). Eventually.. In order to ﬁt the vanillas of this model. Hence.. Given the particular part played by the spot. S.Here.. j=2 ∂xi ∂x j i=1 ∂xi ∑ Electronic copy available at: http://ssrn. y0 ) the initial conditions. We also denote by DT =]0. y. we denote by 0 < t ≤ T the time-variable and by x = (x1 . we know that q has to solve the following equation in order to ﬁt perfectly the vanillas of the market ∂2 1 ∂ ∂q 2 − 2 ( σ2 (rSq) + rq = 0 D S q) + ∂t ∂S 2 ∂S where σD is Dupire’s local volatility and contains the information about the vanillas we want to reproduce. α is the volatility of the volatility factor and µ and ξ are drift terms that may depend on the state variables and on time. (St . 2 Generalized equation and notations Throughout this article. y = y0 ) with (S0 . 0) = δ(S = S0 . S) = pdy q 2 σ2 D (t . y) of the couple (St . Taking q = pdy the marginal of S. S) b2 pdy = σD (t . We are interested in the following equation: ∀S ∈ ΩS . t ≥ 0) is the stock price process and (Yt . We identify the terms in this last formula. the ﬁrst-variable x1 stands for the spot and the last n − 1 for the volatility. we consider ΩS = {S ∈ R/∃y ∈ Rn−1 . y) ∈ Ω}. we get the equation ∂q ∂2 1 ∂ − 2 ( a2 S2 ( b2 pdy)) + (rSq) + rq = 0 ∂t ∂S 2 ∂S Using Dupire’s results from [3].. When we consider the equation from a ﬁnancial point of view. BT = {T } × Ω and CT =]0. we write the Kolmogorov forward equation on the joint density p(t . (S. x2 . . the joint density that calibrates the smile of our model is solution of the nonlinear partial integro-differential equation ∂2 1 2 ∂2 1 2 2 2 pdy ∂2 ∂p pdy 1 2 − ( σ b S p) − (ρσD bαS( 2 ) p) − 2 ( α p)) ∂t ∂S2 2 D b2 pdy ∂S ∂y b pdy ∂y 2 ∂ ∂ + (β p) + (rSp) + rp = 0 ∂y ∂S The rest of this paper is devoted to the study of a more general n-dimensional version of this equation. t ≥ 0) the stochastic component of the volatility. choosing its value properly will enable us to calibrate the vanillas of the model. a is the local volatility part of the model.. B1 and B2 are standard brownian motions with correlation ρ. xn ). y = (x2 .. y) ∈ Ω} and n−1 /(S. we write S = x1 . S) b2 pdy . . Yt ) ∂2 1 2 2 2 ∂2 ∂2 1 2 ∂ ∂ ∂p − 2 ( a b S p) − (ρabαSp) − 2 ( α p)) + (β p) + (rSp) + rp = 0 ∂t ∂S 2 ∂S ∂y ∂y 2 ∂y ∂S p(S.

To do it. we suppose that b does not vary too much and approximate the nonlocal term I ( p) with a suitable constant. one has to assume that the coefﬁcients of the equation belong to Hölder spaces H k. That kind of equation has been properly solved for quite some time now.h.. This gives us two results on p0 . 4. S. x .. our case doesn’t fall under the scope of that paper: the operator I (. since it may become very challenging to study a PDE on a domain with corners.h/2 (DT ) and second 0 < p0 = in f p0 ≤ sup p0 = p0 (H4) O(Ψ) = 0 on ∂B in a sense described in the preliminaries Under the previous assumptions.h/2 (we shall deﬁne them in the preliminaries). S) = ΩS y p(t .h/2 (DT ) and if γ belongs to H 0. we assume that b is non-negative and ΩS y that Ψ the initial condition is strictly positive. from a ﬁnancial viewpoint. By restricting ourselves to short times.h. x ) p(t . Let us now make a few remarks about our particular equation: 1. 5.. First. However. there exists 0 < T ∗ ≤ T and a solution of the equation (1) on DT ∪ BT . | ∂ ∂xi | ≤ b on Ωy where b is a constant we will choose later 2 (H3) Ψ is strictly positive and in H 2. To answer it. S. j≤n is a correlation matrix ie is symetric positive deﬁnite and veriﬁes: ρii = 1 2 for all 1 1 T i and − 2 < ρi j < 2 for i = j. First. xn )dx2 . we are sure that p is not too far from its initial condition and thus is strictly positive. However. y)dy ΩS y (2) (1) belongs to the class of nonlinear. for b∗ small ∗ ∗ enough. y). We also let p0 denote the function p0 (t . see [4] or [5]..h. y)dy ΩS y b2 (y) p(t . 3.h/2 . This observation is the key to our resolution method. we have to prove that b2 (y) p(t .) is not deﬁned on Cb (DT ).. in order to use this method and the results from [4]. y)dy is bounded away from 0. . we reduce our study to domains with a S-section depending on S.. (H1) b ∈ C1 (Rn−1 ). 0 < δ1 ≤ b ≤ δ2 on Rn−1 b ∗ ∗ (H2) ∀2 ≤ i ≤ n. The theorem we will prove requires the following assumptions on b and Ψ.. in the case of an equation with no I ( p) term.. are positive and bounded away from 0 by a stricly positive constant e.dxn b ΩS y 2 (x . x2 . 2. y) = Ψ(S.h. the problem is reduced to the previous remark. S. We add the boundary condition p = Ψ on B ∪ C with Ψ constant on CT . If the αi belong to H 2.. parabolic and nonlocal equations.. ∃(δ1 . it is natural to consider a domain Ω cylindrical with respect to the spot. δ2 ) ∈ R2 .dx n n n 2 2 2 = p(t .where (ρi j )1≤i. ... . we have the following result: Theorem 1. S.h. when b is constant. . then. the question whether I is properly deﬁned is natural.h. S. The complexity of this equation stems from the following integral term: I ( p)(t . if the βi are in H 1. it becomes a classic linear equation of parabolic type. x )dx . The rest of the paper will be devoted to the proof of Theorem 1 above..h/2 (DT ). An interesting reference concerning that kind of equations is [7].h/2 (DT ). p0 belongs to H 2. S. We then isolate the error made during this process in the second term and use a ﬁxed point method to solve the new equation.

h/2 and | f |D h <∞ T i.Q∈DT D D |u|D h = |u|0 + Hh (u) T T T D (u) < ∞ if and only if u is uniformly hölder (exponent h) in DT . in the form xi = r(t . Indeed.h/2 (DT ). ∂2 xx r . we have: D D |uv|D k+h ≤ |u|k+h |v|k+h T T T (4) We can now make the assumptions about DT more precise: for every point Q of CT . xi+1 . ∂t Ψ is uniquely deﬁned (by continuity) on the boundary ∂B of B. ∂t r Hölder continuous (exponent h) and 2 r simply continous.h/2 .. We denote by H 0. + Σ|∂x u|h + |∂t u|h T T T T T (3) where the sums are taken over all the partial derivatives of the indicated order. t ) + ∑ bi (x.h.h.h/2 (DT ) such that Ψ = ψ on B ∪ CT . for all u.h. t )ξiξ j ≥ K2 | ξ |2 (K2 > 0) In addition. ∂ xt tt We also have to consider functions ψ deﬁned on B ∪ CT .. x) and Q = (t . for some i (1 ≤ i ≤ n).h/2 (DT ).v in H k. For a function u..h.. x1 . Such a function ψ is said to belong to H k. we write: |u|D 0 = sup|u| DT T T D Hh (u) = sup T |u(P) − u(Q)| d (P. . .h/2 (DT ) Hh T the set of all functions u for which |u|D h < ∞.h/2 (DT ) which coincide with ψ on |Ψ|D k+h where the inf is taken with respect to all the Ψ’s in H B ∪ CT .h. if we consider a function ψ ∈ H 2. We denote by T H k.h/2 . x ) belong to DT and |x| is the norm of the n-dimensionnal vector x. • ψ ∈ H 2. Thus. and the deﬁnition is independent of Ψ. xi−1 .. It is a Banach space and an algebra with the norm given by deﬁnition 3.h. let K1 be a bound on their norm n • for all (x. given the assumption about DT . They concern the PDE: T Lu := n n ∂u ∂2 u ∂u − ∑ ai j (x.. ∂2 r . We denote this function (on ∂B) by ∂t ψ. .h. if all the derivatives used in the equation exist. The following results will be useful in the proof of our result.h.h/2 (DT ) the set of all functions u for which |u|D k+h < ∞. there exists an (n+1)-dimensional neighborhood V such that V ∩ CT can be represented.h. Q)h P. j=1 ∑ ai j (x. Q) = [|x − x |2 + |t − t |]1/2 where P = (t . the quantity Lψ is well-deﬁned on ∂B. We then deﬁne |ψ|k+h = inf k... t ) + c(x. This process deﬁnes a norm on H k. we can deﬁne the concept of Hölder continuity.h/2 if there exists a Ψ in H k. t )u = f (x. t ) on DT ∪ BT ∂t i. xn ) with r. for any extension Ψ of ψ. j=1 ∂xi ∂x j i=1 ∂xi (5) u = ψ on B ∪ CT We shall need the assumptions: • the coefﬁcients of the operator L belong to H 0. we write ∂xi and ∂∂ xi without distinction.h.3 Preliminaries In this section. Given such a metric d. The other terms of Lψ are also uniquely deﬁned (by continuity) on ∂B. Let us start as in [4] and [5] with the following notion of distance d (P. we write for k ≤ 2: D D k D D |u|D k+h = |u|h + Σ|∂x u|h + . ∂x r. Now. t ) in DT and for all ξ ∈ Rn .

z. Under the previous assumptions and if Lψ = f on ∂B. j=2 ∂xi ∂x j i=1 ∂xi and want to prove the Theorem. y.h/2 (DT ).Theorem 2. If the αi belong to H 2. Proof: the ﬁrst part of the result is classic. one needs a result from [5] about volume potentials and representation of solutions of parabolic equations.h. there exists 0 < T ∗ ≤ T and a solution of the equation (1) on DT ∪ BT . One reads that the solution of the equation 5 with ψ = 0 can be written as t u(x. We use the assumption on b to approximate the integral term I ( p) with 1/b2 . on h and on Ω. are positive and bounded away from 0 by a stricly positive constant e. As to the result with ψ = 0.16 we shall use.h/2 (DT ) and we have the Schauder inequality (with KH 2 depending only on K1 .h.h/2 (DT ). Using both these results. τ) f (z. t . τ)dz where G is the Green’s function for the operator L and veriﬁes |G(x. if the βi are in H 1.h. this solution belongs to H 2. we can write a bound containing the time on the supremum of the solution |u|D 0 t ≤ tK0 | f |D 0 t (7) where K0 only depends on K1 . its proof can be found in [4]. Let us denote by b = b(y0 ) a strictly positive value taken by b (with y0 ∈ ΩS y for some arbitrary S ∈ ΩS ). on K2 . for all t ≤ t and x ∈ Ω |u(x. t ) = 0 dτ Ω G(x. there exists a unique solution of the equation 5. It is the theorem (16. τ)| ≤ K (t − τ)− 2 exp(−K n |x − y|2 ) t −τ (8) with K and K two constants depending on the data of the problem. which is more original. t )| ≤ t | f |D 0 t t dτ 0 Ω K (t − τ)− 2 −1 exp(−K n t |x − y|2 )dz ≤ t | f |D 0 K0 t −τ where K0 depends on K1 . t . Proof: The assumption (H2) gives us some control over the variations of b. if ψ = 0. the gap between those two quantities is quantiﬁed with the . 4 Proof of Theorem 1 n ∂2 ∂2 ∂p 2 − (ρ11 α1 I ( p) p) − ∑ (ρ1i α1 αi ∂t ∂S2 i=2 ∂S∂xi n We are interested in the equation I ( p) p) n ∂2 ∂ −∑ (ρi j αi α j p) + ∑ (βi p) + γ p = 0 on DT ∪ BT i. then.2) of section IV.h. on h and on DT . we get. for b∗ small ∗ ∗ enough. on K2 . on h and on DT ) D |u|D 2+h ≤ KH 2 (|ψ|2+h + | f |h ) t t (6) Furthermore. on K2 .h/2 (DT ) and if γ belongs to H 0.

we see that ∀ p ∈ H 2. y) f j (t .h/2 (DT ).h/2 (Dt ).Lemma 1. S.h.h. It remains to prove that this operator is elliptic: let (ξi )1≤i≤n be n real numbers 1 and (t . t p0 p0 ≤ p ≤ p0 + . We then consider the application M which takes a The set Xx 0 function u ∈ X and sends it on v ∈ H 2. function P strictly positive and increasing on R∗ 0 + such that ∀ p ∈ H we have |I ( p) − 1 Dt 1 Dt ∗ Dt | + | I ( p ) − | ≤ b K P ( | p | b 2 + h 2 + h 2+h ). j=2 ∂xi ∂x j i=1 ∂xi i=2 ∂S∂xi = n ∂2 1 ∂2 1 2 ( ρ α ( I ( u ) − ( ρ α α ( ) u ) + I ( u ) − )u) (9) i 11 1 i 1 ∑ 1 ∂S 2 b b2 i=2 ∂S∂xi with the boundary condition v = Ψ on B ∪ CT . y)ξ2 i ≥ Kρ e 2 i=1 i=1 ∑ ξ2 i where the existence of Kρ is a consequence of ρ being a positive deﬁnite matrix. We write f1 = α b and f i = αi for i ≥ 2. I ( p) belongs to H 2. As a consequence of this lemma.h/2 (DT ) solution of the equation O v := n n n ∂v ∂2 1 1 ∂2 ∂2 ∂ − 2 (ρ11 α2 ( ρ α α v ) − ( ρ α α v ) + (βi v) + γv v ) − i i j i j 1 i 1 ∑ ∑ ∑ 1 2 ∂t ∂S b b i. v exists and belongs to H 2.h/2 (DT ) veriﬁying p0 ≤ p.h. We then write the equation as n n n 1 ∂2 ∂2 ∂ ∂p ∂2 1 − 2 (ρ11 α2 p ) − ( ρ α α p ) − ( ρ α α p ) + (βi p) + γ p ij i j 1i 1 i ∑ ∑ ∑ 1 2 ∂t ∂S b b i=2 ∂S∂xi i. Indeed. the coefﬁcients of this equation belong to the appropriate spaces and because of (H4) the necessary condition Oψ= n ∂2 1 ∂2 1 2 ( ρ α ( I ( Ψ ) − ) Ψ ) + (ρ1i α1 αi ( I (Ψ) − )Ψ) 11 1 ∑ 2 2 ∂S b b i=2 ∂S∂xi on ∂B is veriﬁed.h. | p|D 2+h ≤ x. T ∗ t We take a real number x ≥ | p0 |D 2+h and t ∈ R+ and let Xx denote the set t Xx = { p ∈ H 2. b b2 Remark. S.h. v belongs to Xx |v|D 2+h ≤ x t p0 p0 ≤ v ≤ p0 + 2 2 . y) ∈ DT .h/2 (DT ) veriﬁying p ≤ p. The existence of v is given by Theorem 2. p = Ψ on B ∪ CT } 2 2 t clearly contains the function p . j=2 ∂xi ∂x j i=1 ∂xi = n ∂2 1 ∂2 1 2 ( ρ α ( I ( p ) − (ρ1i α1 αi ( I ( p) − ) p) ) p ) + 11 ∑ 1 2 2 ∂S b b i=2 ∂S∂xi To solve this equation. we have n i. we apply a ﬁxed point method and use the lemma 1 to get an upper bound on the second term. n. There exists a constant Kb (depending only on h.h/2 (DT ). This proves the ellipticity of the operator. j=1 n n ∑ ρi j fi(t .h. S. y)ξiξ j ≥ Kρ ∑ fi2 (t . S. We now want to show that for t ie that suitable x and t . δ2 . p0 and Ω) and a polynomial 2. δ1 .

to this function v ˜ t |v ˜|D 0 n t ∂2 1 1 ∂2 2 ≤ tK0 |O p0 + 2 (ρ11 α1 (I (u) − 2 )u) + ∑ (ρ1i α1 αi ( I (u) − )u)|D 0 ∂S b b i=2 ∂S∂xi ≤ tK0 (|O T p0 |D 0 T n t ∂2 ∂2 1 1 2 + | 2 (ρ11 α1 (I (u) − 2 )u) + ∑ (ρ1i α1 αi ( I (u) − )u)|D h ) ∂S b b i=2 ∂S∂xi ≤ tK0 (|O p0 |D 0 + 1) T (11) t 0 0 Taking T ∗ K0 (|O p0 |D ˜|D ˜. T∗ Using this statement. we construct a bounded sequence ( pn )n∈N of functions belonging to Xx ∗ p p • p0 has been previously deﬁned • by induction. | pn |D 2+h ≤ x . the inequality 7. t ) − ∂xi ∂x j ∂xi ∂x j (x . thus |u|D ≤ x and then We remember that u belongs to Xx 2+h ∗ D |v|D 2+h ≤ KH 2 (|ψ|2+h + b K P(x)x) D ∗ Taking x∗ = max(KH 2 (|ψ|D 2+h + 1). (x. we apply 6 D |v|D 2+h ≤ KH 2 (|ψ|2+h + | T t t n t ∂2 1 ∂2 1 2 ( ρ α ( I ( u ) − ) u ) + (ρ1i α1 αi ( I (u) − )u)|D 11 ∑ 1 h ) 2 2 ∂S b b i=2 ∂S∂xi n t 1 Dt 1 ) u | + |ρ1i α1 αi ( I (u) − )u|D ∑ 2+h 2+h ) 2 b b i=2 n T 2 ≤ KH 2 (|ψ|D 2+h + |ρ11 α1 (I (u) − T T D 2 D ≤ KH 2 (|ψ|D 2+h + (|ρ11 α1 |2+h + ∑ |ρ1i α1 αi |2+h )(|I (u) − ≤ where K T ∗ KH 2 (|ψ|D 2+h + b K i=2 t Dt P(|u|2+h )|u|D 2+h ) T 1 Dt 1 t Dt | + | I (u) − |D 2+h )|u|2+h ) 2 2+h b b (10) DT = (|ρ11 α2 1 |2+h + n i=2 ∑ |ρ1iα1αi|D 2+h )Kb (we apply lemma 1 for the last line). It is clear that v ˜ veriﬁes n ∂2 1 1 ∂2 2 ( ρ α ( I ( u ) − ) u ) + ( ρ α α ( I ( u ) − )u) i 11 1 i 1 ∑ 1 ∂S 2 ∂ S ∂ x b b2 i i=2 on DT ∪ BT with v ˜ = 0 on B ∪ CT (here we use the fact that Ψ is constant on CT ). Repeated applications of the Ascoli-Arzelà theorem ∗ ∗ give us a function p ∈ C2 (DT ) limit in C2 (DT ) of a subsequence of pn . ∗ 10 gives us |v|D 2+h ≤ x . T Let us write v ˜ = v − p0 . (x (| x − x |2 + | t − t |)α/2 . t ). Since T | sup{ ∂2 pn ∂2 pn ( x . We now apply the second part of Theorem 2. 1 K P(x∗ )x∗ . t t . | p0 |2+h ) and b ≤ T T t T It remains to prove that Ov ˜ = O p0 + p0 2 ≤ v ≤ p0 + p0 2. Eventually. The application M maps Xx∗ into itself.For the ﬁrst inequality. we get |v 0 ≤ 2 . since v = p0 + v ∗ ∗ T T is proved and v belongs to Xx∗ . we have ∀n ∈ N. t ) ∈ DT } ≤ x ∗ . the last inequality 0 + 1) = 2 . we write pn+1 = M ( pn ) ∗ By construction. t ) | .

.. S... Let us now prove lemma 1.. xi−1. xn)| ∂(b2 ) Dt | |x − xi | ≤ (n − 2)b∗ |y − y| ∂xi 0 i ∑| .h/2 . xi+1. the denominator is bounded away from 0. . n. . t ) | . (x. xn ) − b2 (x2 . The only result needed is I ( pn ) → I ( p). .. we ﬁnd that p ∈ H 2. xn) − b2(x2. We get the last line from the following computation where y = (x2 .. K stands for some constant depending only on the data of the problem (Ω. Thus. p is solution of our equation. xn ) and y0 = (x2 .h. δ1 . y)|b2 − b2 (y)|dy ≤ | p|D 0 T T ΩS y |b2 − b2 (y)|dy |y0 − y|dy| p|D 0 T ≤ | p|D 0 ΩS y T |b2 (y0 ) − b2 (y)|dy ≤ b∗ (n − 2) ΩS y ≤ b∗ K | p|D 0 here and in the rest of the proof. . We have | ΩS y p(t . y)(b2 − b2 (y))dy| ≤ ΩS y p(t . S. S) belong to ]0. xn )| ≤ ≤ i=3 n i=3 ∑ |b2(x2.We have | sup{ ∂2 p ∂2 p ∂xi ∂x j (x. we have: |I ( p) − 1 DT 1 DT pdy p(b2 − b2 )dy DT − = | = | | | |2+h 2+h b2 pdy b2 2+h b2 b2 b2 pdy T T 1 1 p(b2 − b2 )dy|D |D ≤ 2| 2 2+h | 2+h b pdy b Let us compute one after another the terms appearing in this norm (we remember that those functions only depend on t and S). t ) ∈ DT } ≤ x ∗ And this computation being true for all the derivatives appearing in the norm H 2. j=2 ∂xi ∂x j = n ∂2 1 ∂2 1 2 ( I ( p ) − ( ρ α (ρ1i α1 αi ( I ( p) − ) p) ) p ) + 11 ∑ 1 2 2 ∂S b b i=2 ∂S∂xi That concludes the proof of the theorem.) but not on p nor on b∗ . The last step of the proof is to take the limit in 9. t ).. xi. . xn ): n |b2 (y0 ) − b2 (y)| = |b2 (x2 . Let (t .h. Two applications of the dominated convergence Xx ∗ theorem give us the convergence we need.. T [ × ΩS . By deﬁnition. Since pn ∈ T ∗ . xi. (x (| x − x |2 + | t − t |)α/2 . . . .h/2 . t ) − ∂xi ∂x j (x .. it is possible to take the limit in 9 which gives us n n n ∂p ∂2 ∂2 1 ∂ ∂2 2 1 − 2 (ρ11 α1 2 p) − ∑ (ρ1i α1 αi p) − ∑ (ρi j αi α j p) + ∑ (βi p) + γ p ∂t ∂S b b i=2 ∂S∂xi i=1 ∂xi i.

y)||b2 (y0 ) − b2 (y)|dy + + S ΩS y \Ωy S ΩS y \Ωy T D ≤ Hh ( p)D((t . S )) T S ΩS y ∩Ωy |b2 (y0 ) − b2 (y)|dy + | p|D 0 +| p|D 0 S ΩS y \Ωy T ΩS y \ΩS y ≤ b∗ K | p|D h (D((t . S )) + T S ΩS y \Ωy |y0 − y|dy + S ΩS y \Ωy By assumption on the boundary of our domain. y)dy − t b ΩS y 2 (y) p(S. S . t . (t . S. ∂S 2 ∂p 2 2 DT ∂t (b − b )dy|h +| ∂p 2 2 DT ∂S (b − b )dy|h + we get from the previous computation | ∗ D p(b2 − b2 )dy|D 2+h ≤ b K | p|2+h T 1 |D . y)dy| ΩS y b2 (y) p(S. S ) belong to ]0. ΩS \ΩS |y0 − y|dy ≤ KD(S. S). S )) ≤ KHh We used the same kind of arguments than earlier. As for derivatives of b2 pdy . S ) belong to ]0. p0 and Ω. t . T [ × ΩS . y) − p(t . This gives us y y | T ∗ D p(b2 − b2 )dy|D h ≤ b K | p|h T T t 2 2 D And since | p(b2 − b2 )dy|D 2+h = | p(b − b )dy|h + | | T ∂2 p 2 (b − b2 )dy|D h . S ). it is possible to ﬁnd a constant K depending only on Ω such as ∀S. let (t . y)dy ΩS ΩS y y b2 (y) p(S . δ2 1 p0V (Ω) T . T [ × ΩS . let (t . y)(b2 (y0 ) − b2 (y))dy − ΩS y p(t . S). K denotes here another constant depending on δ1 . This gives us | b21pdy |D h ≤ K (1 + | p|h ). (t . S ∈ ΩS . (t . t . y)|b2 (y0 ) − b2 (y)|dy p(t . t . We write | 1 | ≤ b2 (y) p(S . S) and (t . y)dy D ( p)D((t . b2 pdy 2+h T t We now have to ﬁnd a bound on | 2 . S). we have | Since p belongs to Xx ∗ ∗ T 1 |D b2 pdy 0 ≤ Now. S . t . S) and (t . we get | T 1 |D 2+h 2 b pdy D 2 D 3 D 4 D 5 ≤ K (1 + | p|D 2+h + (| p|2+h ) + (| p|2+h ) + (| p|2+h ) + (| p|2+h ) ) T T T T T . we have p b2 ∂ ∂ 1 DT 2 DT DT ∂S dy DT | ( 2 )|h = | − | ≤ K ( 1 + | p | ) | p | h h 1+h ∂S b pdy ( b2 pdy)2 The same kind of computation is true for the derivative of second order p p p 2 dy DT b2 ∂ b2 ∂ 2( b2 ∂ T 3 ∂ DT DT 2 DT 2 DT ∂S dy ∂S dy) ∂S2 | (− )|h = | − |h ≤ K [(1 + | p|D h ) (| p|1+h ) + (1 + | p|h ) | p|2+h ] 2 2 2 3 2 2 ∂S ( b pdy) ( b pdy) ( b pdy) 2 Eventually. We compute | ≤ ΩS y p(t . T 1 DT δ2 . S. y)(b2 (y0 ) − b2 (y))dy| p(t . for instance with respect to S. y)dy ΩS y b2 (y) p(S . S. S. t . y)dy ΩS y 1 − | 2 b (y) p(S.Now. y)|b2 (y0 ) − b2 (y)|dy |b2 (y0 ) − b2 (y)|dy |b2 (y0 ) − b2 (y)|dy |y0 − y|dy) S ΩS y ∩Ωy | p(t .

vol. writing √ I ( p)− 1 b2 I ( p)+ 1 b 1 I ( p) − b = . N.Combining this result with the previous computations. 5 Conclusion In this paper. V. N.. Berestycki. Pazy. Amer. This con- cludes the proof. Tachet.. R. Busca.. Proc AFFI Conf. 1968. we have shown that the equation driving the calibration problem for local and stochastic volatility models is well-posed in the case of suitably regularized initial conditions. Pricing and Hedging with Smiles. uniqueness and regularity for nonlinear parabolic equations with nonlocal terms equations with nonlocal terms.J.N. Now. Ural’ceva. Math. Math. Linear and quasilinear equations of parabolic type. R. 14 (3-4):259–289. Comm. It is however clear that the solution of the full Kolmogorov equation with Dirac initial condition does not obtain as a consequence of Theorem 1 : possible extensions of our results towards this direction are currently being explored. Prentice-Hall Inc. 23. Providence. we ﬁnd the same kind of results for the second term involved in the lemma. 1983 [7] N. State Contingent Prices Implicit in Option Prices. RI.. Breeden. I. Monographs. Springer-Verlag. 2007 [8] F. Math.A. Trans. [6] A. 621-651. Florent. Let us also mention that a generalization of Theorem 1 to multidimensional correlation calibration have already been investigated and will be presented in [8] References [1] D. Abergel. Pure Appl.A. 57(10):1352-1373. 1978 [2] H. Solonnikov. NoDEA Nonlinear Differential Equations Appl. La Baule. Journal of Business 51. 2004 [3] B. J. Friedman. Alibaud.. [5] O. Soc. Partial differential equations of parabolic type. 1993 [4] A. Litzenberger. strictly positive on R∗ + . 1964. New-York. Computing the implied volatility in stochastic volatility models. Ladyzhenskaya. we ﬁnd |I ( p) − t 1 DT | ≤ b∗ KP(| p|D 2+h ) 2 2+h b with P a polynomial function of degree 6. Dupire. Semigroups of linear operators and applications to partial differential equations. Existence. work in progress . Englewood Cliffs.

- 5 Love Languages Study Guide
- understanding_annuity_basics.pdf
- SEPTA _ Southeastern Pennsylvania Transportation Authority
- fw9
- Fact Sheet
- Crawfish Update 31 March
- BH12 Registration Form PDF
- 10.1.1.58.1537
- Damages
- About Downloads
- Ce Reg 200305
- 15993270 Mythical Swan
- 26116
- vic-recap
- nlReader.dll
- Jump Type Levy Processes
- Math Girls Sample
- SSRN-id2179188
- Postgres is Different From Better Than Your RDBMS
- 37453934-BurryTranscript
- sayc_card
- 32174243 JM I Want to Break Free
- Buffett Questions
- BHB1986_determinants of Portfolio Performance

Diff Eq

Diff Eq

- 0-içindekiler 6
- The Euler-Maclaurin Formula
- Mixed Model Assembly Line Balancing
- Solving Recurrences
- Appendix 1
- Dynamic Programming and Optimal Control Script
- 10-6 the Binomial Theorem
- AMTStudentNotes.pdf
- FEM Prof Notes
- 3-Equilibrium of Rigid Bodies_FE
- ALG1.2
- ch2
- 1998-GhirlandaGuerra
- Summability Methods r
- Binomial Theorem
- MIT Calculus 18-001 Chapter 10
- KroneckerDeltaAndLeviCivita
- Thesis 10
- PART2
- Tensor
- math_induction
- MODSIMv8.1UserManual_Appendices
- AppGeoph(b)
- Summation Notes
- Aakash - Binomial Theorem and Its Application
- Formula Sheet
- Fundamental Mathematics Explained
- 2126 Template
- Practice for Test 4 051
- Chapter 4
- Non Linear Pide

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd