Title: Stationary and nonstationary time series models for categorical data
Authors: Lionel Truquet - ENSAI (France) [presenting]
Konstantinos Fokianos - University of Cyprus (Cyprus)
Abstract: Finite-state Markov chains are of limited use in the modeling of dependent data because the number of parameters grows exponentially with the order of the chain. To get more parsimonious models, some logistic type autoregressions that involve a latent process are considered in the literature. We will present optimal conditions for stationarity and ergodicity of such processes. In the nonstationary case, we will also discuss locally stationary versions of these models as well as some asymptotic results for statistical inference. Our results are highly based on coupling techniques for a general class of finite-state processes, the chains with complete connections.