A Check on the Robustness of Hamilton's Markov Switching Model Approach to the Economic Analysis of the Business Cycle

This note explores the robustness of Hamilton's (Econometrica, 1989) two-regime Markov switching model framework for capturing business-cycle patterns. Applying his exact specification to a revised version of real GNP, I find parameter estimates that are similar to those he reported only when I use the same sample period (1952-1984) and a particular set of starting values for the maximum likelihood procedure. Two other local maxima exist that have higher likelihood values, and neither correspond to the conventional recession-expansion dichotomy. In fact, when the sample period is extended, there is no longer a local maximum near the parameter set reported by Hamilton. Exploring the model and data further, I reject cross-regime restrictions of Hamilton specification, but also find that relaxing these restrictions increases the number of local maxima. However, a parsimonious three-regime model for GNP growth is more robust and plausible, especially when each regime is required to last more than one quarter.