Stochastic analysis in discrete and continuous settings. With normal martingales. (Q730785)

The author presents several aspects of stochastic analysis for discrete and continuous-time normal martingales. Normal martingales constitute a class of stochastic processes which contains both continuous and pure jump processes including for example the standard Brownian motion, the compensated Poisson process and some ``unusual'' processes like the so-called Azéma martingales. In the literature, several authors have described a stochastic analysis and especially Malliavin's calculus either for continuous diffusion processes defined on Gaussian and Wiener spaces, or for jump processes. The particular feature of this volume of the Lecture Notes in Mathematics is to consider a framework (normal martingales) for which a stochastic analysis can be derived simultaneously for continuous and for jump processes. Another achievement of this book is to deal with both discrete and continuous-time processes. Especially a Malliavin calculus for some discrete time processes is presented which constitutes a material interesting in itself. The book is decomposed into eight chapters whose contents is described in detail below. In the first chapter, the author presents a stochastic analysis for discrete time normal martingales. This case is simpler than the continuous one but it still captures the basic and important properties of the continuous-time setting. First a stochastic integral in the spirit of Itô (meaning in an \(L^2\)-sense) is defined for discrete-time normal martingales. This construction is immediately followed by the introduction of multiple stochastic integrals which will play an important role for the chaotic representation a little bit later in the book. Then, writing the discrete-time structure equation satisfied by a sequence of independent Bernoulli random variables, a chaotic representation property for discrete-time random walks with independent increments is given. Once the chaotic decomposition is obtained, one can define a discrete-time gradient and a discrete-time divergence operator related to each other by the \textit{integration by parts formula}. Other material like the Clark formula, the predictable representation property or the Ornstein-Uhlenbeck semi-group follows from the previous notions. The last sections of this first chapter are composed of several applications including applications to covariance identities and deviation inequalities under Bernoulli measures, logarithmic-Sobolev inequalities for random variables and finally to hedging options with respect to a Cox-Ross-Rubinstein discrete-time financial market. We now turn to the description of Chapter 2 which deals with the continuous-time case. More precisely, the definition of continuous-time normal martingales is given and immediately followed by the most significant examples that are the Brownian motion and the compensated Poisson process. The other sections of this chapter are devoted to extend the main concepts encountered in Chapter 1 to continuous-time normal martingales like the stochastic integral, the multiple integrals, the predictable representation property, the chaotic decomposition, the product formula, Itô's formula and the structure equations satisfied by the quadratic variations of these continuous-time normal martingales. The Malliavin calculus with respect to a normal martingale is introduced in Chapter 3. More precisely, the author presents how the existence of a gradient operator and of its dual operator (the divergence) is related to the predictable representation property and to the Clark formula. The former entails that every square integrable random variable \(F\) can be represented as \[ F=\mathbb{E}\left[ F \right]+\int_0^\infty u_t dM_t \] where \(u\) is a predictable stochastic process whereas the latter allows one to be more precise about the representing process \(u\) since it holds that \(u_t=\mathbb{E}[D_t F | \mathcal{F}_{t-}]\) where \(D\) denotes the Malliavin derivative. Note that the main advantage of the construction in Chapter 3 is that the gradient operator \(D\) has not been specified (it is just assumed to exist), exhibiting general relations between a gradient operator and other properties or notions of stochastic analysis like {e.g.} the stochastic integral, covariance inequalities and logarithmic Sobolev inequalities. Obviously, this abstract construction calls for an example of such a pair (gradient, divergence) which is presented in Chapter 4. In Chapter 4, a gradient operator is defined for a general normal martingale using the chaotic representation and the multiple integrals defined in Chapter 2. More precisely, the gradient operator is defined as an \textit{annihilation operator} meaning that the gradient of a multiple integral of order \(n\) is defined as a multiple integral of order \(n-1\) extending the classical notion of real analysis where the gradient is the inverse of the integration operator. The definition of the gradient is then extended to a dense subset of the space of square-integrable random variables. The divergence operator which is defined as the dual operator of the gradient \textit{via} the integration by parts formula is then shown to be a \textit{creation operator} meaning that the divergence of a process defined as a multiple integral of order \(n\) as \((I_n(f_{n+1}(\ast,t)))_{t \geq 0}\) coincides, up to a constant, with the multiple stochastic integral \(I_{n+1}(f_{n+1})\) of order \(n+1\). This definition is extended to more general processes and, for instance, it is proved that the divergence operator \(\delta\) generalizes Itô's stochastic integral since the divergence of every square-integrable adapted stochastic process coincides with its Itô integral. The chapter is completed by additional material not described in this review. Chapters 5 and 6 complete the specification and the analysis of the gradient operator and of the divergence operator respectively for the Brownian motion in Chapter 5 and for a general Poisson process in Chapter 6. In Chapter 5, the gradient operator is related to a directional derivative on the Wiener space whose directions are elements of the Cameron-Martin space. The chaotic decomposition is obtained using the canonical decomposition of the Gaussian measure with respect to the Hermite polynoms. In Chapter 6, a general Poisson point process is presented and its chaotic decomposition property is shown using Charlier polynomials. A first gradient operator, that is the \textit{finite difference operator}, is defined on the Poisson space according to the construction of Chapter 4. Many other properties on these Malliavin type operators are given in Section 5 and 6. For instance one can notice that the gradient operator in the Wiener setting satisfies the usual chain rule property (i.e., that \(D(FG)=F DG+G DF\)) whereas in the Poisson case the chain rule property of the difference gradient operator contains an additional term \(DF DG\) and thus does not satisfy the usual \textit{derivation property} (general chain rule type formulae are obtained for general normal martingales as in Proposition 4.5.2). The standard chain rule property (also called the derivation property) \(D(FG)=F DG+G DF\) is very useful in applications, so it is interesting to investigate whether or not one can construct other gradients on the Poisson space which satisfy the derivation property. This question constitutes the subject of Chapter 7 where several gradients (and their adjoints) satisfying the derivation property are derived. Here, we should stress that this material is rarely presented in the literature and this variety of operators on the Poisson space is an highlight of this book. Finally, Chapter~8 presents an application of the tools developed in the previous chapters to pricing and hedging in continuous time. To sum up this review, in this monograph the author proposes a framework (normal martingales) for which a general stochastic analysis and a general Malliavin calculus can be obtained in both the discrete-time and the continuous-time setting. This setup is valid for a quite general class of processes including both continuous processes like the Brownian motion and jump processes like general Poisson point processes or Azéma martingales. Another main highlight of this monograph is the description of several Malliavin derivatives on the Poisson space including the usual difference operator which does not enjoy the standard derivation property and other gradients (which extend the gradient of Carlen and Pardoux) which satisfy the derivation property. It is finally worth mentioning that this volume of the Lecture Notes in Mathematics includes many interesting applications and that the various notions, properties and proofs are clear and detailed.