Title: Markov chains models with time-varying parameters
Authors: Lionel Truquet - ENSAI (France) [presenting]
Abstract: The focus is on time-inhomogeneous Markov chains on general state spaces and for which the finite-dimensional distributions can be approximated locally by ergodic Markov chains via an infill asymptotic. Our approach, which is based on contraction properties of Markov kernels, can be seen as a Markov version of the notion of local stationarity. In particular, one can consider time-inhomogeneous autoregressive processes, finite-state Markov chains or Markov-switching processes for which statistical inference is still possible. We will explain how to construct such models and what type of statistical results it is possible to get.