This article analyzes structural changes in Japanese business cycles using the coincident index of composite indexes (CI) from January 1980 to February 2009. A Markov switching model where the average growth rates of CI differ between expansion and recession periods is extended so that the average growth rates for the both periods and the variance of error term may be subject to structural changes. Models with different numbers of change points are estimated using Bayesian method via Markov chain Monte Carlo techniques and the number of change points is selected by the marginal likelihood. Main results are: (1) the number of change points is two; (2) one is around July 1991 and the other is around July 2008; (3) the both structural changes decreased the average growth rates for the following periods.
