Has the Business Cycle Changed in Japan? A Bayesian Analysis Based on a Markov-Switching Model with Multiple Change-Points.

Using a Markov-switching model and Bayesian inference, the turning points of Japanese business cycles are identified from a monthly coincident composite index series, taken over the last thirty years. Ordinarily, in taking such a long-range estimation approach, we would face the following questions: (1) Have there been any structural changes? (2) If so, does the existence of these structural changes prevent the detection of the turning points? (3) How many changes have occurred? (4) When did these changes occur? The Bayesian analysis approach easily provides answers. The estimation results suggest that the Markov-switching model with no changes is unable to identify turning points appropriately, whereas the model with changes selected via the Bayes factor robustly estimates these points for the long period of time considered, and also successfully facilitates estimation of the changes in amplitude that occurred between booms and recessions, as well as in the volatility of business cycles.