| Summary: | This paper develops a new Bayesian approach to clustering analysis of multiple time series with structural breaks. The number of breaks is treated as a random variable, and group membership and group-specific parameters can change upon the breaks. The group-specific parameters in each regime can be integrated analytically, so we only have a small number of parameters to be handled by posterior simulation. We discuss model identification, clustering, prediction, and model selection based on marginal likelihood. We use a simulation study to document the performance of the proposed approach in statistical efficiency, forecasting, and detection of the structural breaks. Applications to quarterly industrial production growth rates of 21 countries and monthly asset returns link regimes to specific historical periods and provide predictions and trading strategies based on the proposed method.
|
|---|
