Bayesian Model Selection for Structural Break Models

Abstract:     
We take a Bayesian approach to model selection in regression models with structural breaks in conditional mean and residual variance parameters. A novel feature of our approach is that it does not assume knowledge of the parameter subset that undergoes structural breaks, but instead conducts model selection jointly over the number of structural breaks and the subset of the parameter vector that changes at each break date. Simulation experiments demonstrate that conducting this joint model selection can be quite important in practice for the detection of structural breaks. We apply the proposed model selection procedure to characterize structural breaks in the parameters of an autoregressive model for post-war U.S. inflation. We find important changes in both residual variance and conditional mean parameters, the latter of which is revealed only upon conducting the joint model selection procedure developed here.

Keywords: Posterior Model Probability, Markov Chain Monte Carlo, Inflation Persistence

JEL Classifications: C11, C22, C52, E31

Working Paper Series