Bayesian Change Point Analysis of Time Series

Sinsup Cho* and Seungmin Nam

Seoul National University

In this paper we consider a Bayesian change point analysis in autoregressive conditional heteroscedastic model. We assume that all or part of the parameters in the ARCH equation change over time. We model the occurrence of the change points as a discrete time Markov process with the unknown transition probabilities and estimate parameters by Markov chain Monte Carlo methods following Chib (1998). Model selection is performed using the marginal likelihood.