A Test for Structural Change in Markov-Switching Models : Has the U.S. Economy Become More Stable?

We develop a Bayesian test for structural change at an unknown changepoint in Markov-switching models. Unlike the usual Bayesian treatment of the unknown changepoint problem in the literature, we cast the problem into a `model selection' framework. This is done by adopting a prior for the shift parameters that is a mixture of a truncated Normal and point mass at 0. That is, instead of making an inference on structural change indirectly based on the posterior distribution of the changepoint, we make an inference directly based on the posterior probabilities that the shift parameters are 0. When the proposed framework and procedure are applied to the Markov-switching models of postwar real GNP growth and the fixed real investment growth, we find sample evidence in favor of a structural shift toward narrowing gap between the growth rates during booms and recessions. We interpret this as a structural change in the postwar U.S. economy toward more stability