Computable infinite-dimensional filters with applications to discretized diffusion processes (Q855688)

The authors provide non-trivial examples of nonlinear models where the filtering steps can be explicitly described. The signal may or may not be stationary. Let one consider a pair signal-observation \((x_n, y_n) (n\geq 0)\), where the unobserved signal \((x_n)\) is a Markov chain and the observed component is such that, given the whole sequence \((x_n)\), the random variables \((y_n)\) are independent and the conditional distribution of \(y_n\) only depends on the corresponding state variable \(x_n\). This is a classical setting in the field of nonlinear filtering and the process \((y_n)\) is often called a hidden Markov model. In this context, a central problem that has been the subject of a huge number of contributions is the study of the exact filter, i.e. the sequence of conditional distributions of \(x_n\) given \(y_n, y_1, y_0, n\geq 0\). The filter system is not finite-dimensional. Nevertheless, optimal and prediction filters are specified by a finite number of parameters so they are computable. Their main example (Section 3) is interesting for modeling purposes in the field of finance and especially for stochastic volatility models. Other fields of applications of scale perturbation models remain to be investigated. Numerical simulation results has been done for the model corresponding to a one-dimensional radial Ornstein-Uhlenbeck process. The above results may be extended to the case of a non-time-homogeneous signal.