Bayesian Analysis of Dynamic Multivariate Models with Multiple Structural Breaks

This paper considers a vector autoregressive model or a vector error correction model with multiple structural breaks in any subset of parameters, using a Bayesian approach with Markov chain Monte Carlo simulation technique. The number of structural breaks is determined as a sort of model selection by the posterior odds. For a cointegrated model, cointegrating rank is also allowed to change with breaks. Bayesian approach by Strachan (Journal of Business and Economic Statistics 21 (2003) 185) and Strachan and Inder (Journal of Econometrics 123 (2004) 307) are applied to estimate the cointegrating vectors. As empirical examples, we investigate structural changes in the predictive power of the yield curve and the US term structure of interest rates. We find strong evidence of three structural changes in both applications.