Change-points in US business cycle durations

This paper develops a change-point model that can endogenously detect a structural shift in a time series of durations. The model is applied to NBER data on U.S. business cycle durations for expansions and contractions. There are two primary results. First, the change-point model endogenously detects a shift in the distribution for the phases of the U.S. business cycle around WWII. The pattern of duration dependence for both contractions and expansions correspond to earlier work, such as Diebold and Rudebusch (1990), Sichel (1991) and Zuehlke(2003), that exogenously split the sample at WWII. The second result is that the change-points for expansions and contractions generally occur earlier than WWII when controlling for various factors, such as the duration of the preceding half-cycle, wars and a trend variable. For expansions, the only significant explanatory variable is a trend, resulting in each successive expansion's hazard rate uniformly shifting down. For contractions, both a trend and the lagged duration of the preceding expansion are found, when estimated separately, to be significant. Controlling for a trend, contractions no longer exhibit positive duration dependence following the estimated change-point.