High weak order discretization schemes for stochastic differential equation.

Summary The development of technology and computer science in the last decades, has led the emergence of numerical methods for the approximation of Stochastic Differential Equations (SDE) and for the estimation of their parameters. This thesis treats both of these two aspects. In particular, we study the effectiveness of those methods. The first part will be devoted to SDE's approximation by numerical schemes while the second part will deal with the estimation of the parameters of the Wishart process. First, we focus on approximation schemes for SDE's. We will treat schemes which are defined on a time grid with size $n$. We say that the scheme $ X^n $ converges weakly to the diffusion $ X $, with order $ h in mathbb{N} $, if for every $ T> 0 $, $ vert mathbb{E} [f (X_T) -f (X_T^n)]vert leqslant C_f / h^n $. Until now, except in some particular cases (Euler and Victoir Ninomiya schemes), researches on this topic require that $ C_f$ depends on the supremum norm of $ f $ as well as its derivatives. In other words $C_f =C sum_{vert alpha vert leqslant q} Vert partial_{alpha} f Vert_{ infty}$. Our goal is to show that, if the scheme converges weakly with order $ h $ for such $C_f$, then, under non degeneracy and regularity assumptions, we can obtain the same result with $ C_f=C Vert f Vert_{infty}$. We are thus able to estimate $mathbb{E} [f (X_T)]$ for a bounded and measurable function $f$. We will say that the scheme converges for the total variation distance, with rate $h$. We will also prove that the density of $X^n_T$ and its derivatives converge toward the ones of $X_T$. The proof of those results relies on a variant of the Malliavin calculus based on the noise of the random variable involved in the scheme. The great benefit of our approach is that it does not treat the case of a particular scheme and it can be used for many schemes. For instance, our result applies to both Euler $(h = 1)$ and Ninomiya Victoir $(h = 2)$ schemes. Furthermore, the random variables used in this set of schemes do not have a particular distribution law but belong to a set of laws. This leads to consider our result as an invariance principle as well. Finally, we will also illustrate this result for a third weak order scheme for one dimensional SDE's. The second part of this thesis deals with the topic of SDE's parameter estimation. More particularly, we will study the Maximum Likelihood Estimator (MLE) of the parameters that appear in the matrix model of Wishart. This process is the multi-dimensional version of the Cox Ingersoll Ross (CIR) process. Its specificity relies on the square root term which appears in the diffusion coefficient. Using those processes, it is possible to generalize the Heston model for the case of a local covariance. This thesis provides the calculation of the EMV of the parameters of the Wishart process. It also gives the speed of convergence and the limit laws for the ergodic cases and for some non-ergodic case. In order to obtain those results, we will use various methods, namely: the ergodic theorems, time change methods or the study of the joint Laplace transform of the Wishart process together with its average process.