Particle Filters for Markov-Switching Stochastic-Correlation Models

This work deals with multivariate stochastic volatility models that account for time-varying stochastic correlation between the observable variables. We focus on the bivariate models. A contribution of the work is to introduce Beta and Gamma autoregressive processes for modelling the correlation dynamics. Another contribution f our work is to allow the parameter of the correlation process to be governed by a Markov-switching process. Finally we propose a simulation-based Bayesian approach, called regularised sequential Monte Carlo. This framework is suitable for on-line estimation and the model selection.