Bayesian Analysis of Markov Switching Vector Error Correction Model

This paper introduces a Bayesian approach to a Markov switching vector error correction model that allows for regime shifts in the intercept terms, the lag terms, the adjustment terms and the variance-covariance matrix. The proposed Bayesian method allows for estimation of the cointegrating vector within a nonlinear framework through Gibbs sampling so that it generates more efficient estimation than classical approaches that require a multi-stage maximum likelihood procedure. The Bayes factors are applied to test for Markov switching and model specifications. We apply the proposed model to U.S. term structure of interest rates allowing the risk premium and other parameters in the model to change with regime.