A flexible prior distribution for Markov switching autoregressions with Student-t errors

Abstract

This paper proposes an empirical Bayes approach for Markov switching autoregressions that can constrain some of the state-dependent parameters (regression coefficients and error variances) to be approximately equal across regimes. By flexibly reducing the dimension of the parameter space, this can help to ensure regime separation and to detect the Markov switching nature of the data. The permutation sampler with a hierarchical prior is used for choosing the prior moments, the identification constraint, and the parameters governing prior state dependence. The empirical relevance of the methodology is illustrated with an application to quarterly and monthly real interest rate data.