Detection of Multiple Changes of Variance Using Posterior Odds

Abstract

This article uses a Bayesian procedure based on obtaining posterior odds to assess the evidence about the existence of multiple changes of variance in a time series. The approach is developed for sequences of independent observations. An extension to consider autoregressive models is also discussed. The information on the data about the location of the change points and the magnitude of the variances at the different pieces of the series is summarized through posterior distributions. The procedure is illustrated with a well-known financial series.