Predicting cyclical turning points with leading index in a markov switching model

We have evaluated the Commerce Department's Composite Index of Leading Indicators as a predictor of business cycle turning points using the two‐state Markov switching model as the filter. Contrary to some recent studies, we found that the predictive performance of CLI is quite good and, with an exception of the 1973:11 peak, it made very little difference to the prediction of turning points whether real‐time data are used instead of the revised series. We found, however, that imposing any degree of autoregression in the errors on the simple regime‐shift model caused the filter to signal turning points inappropriately. Also, we found no evidence of duration dependence in post‐war U.S. business cycles.