Dating and predicting phase changes in the U.S. business cycle

Abstract

Hamilton's quasi-Bayesian, Markovian, regime-switching model is applied to monthly growth rates of leading, long-leading and coincident indexes of the US economy. A simple rule applied to regime probabilities for each data point of the coincident index produces a phase chronology which exactly reproduces the turning points of the index produced by the Bry-Boschan method. The chronology is also therefore almost identical to the officially recognised US business cycle chronology. To gain some insight into how quickly the last eight business cycle turning points could have been identified, the dating algorithm is applied to the data sequentially, augmenting the sample period one monthly observation at a time. The regime-switching model is also applied to the leading and long-leading indexes. The application of a simple rule to the regime probabilities is found to result in a very reliable advance signalling system for the US business cycle.