Unobserved Leading and Coincident Common Factors in the Post-War U.S. Business Cycle

The paper introduces a two-factor model of the common leading and coincident economic indicators. Both factors are unobserved and each of them captures the dynamics of a corresponding group of the observed time series. The common leading factor is assumed to Granger-cause the common coincident factor. This property is used to estimate these two factors simultaneously and hence more efficiently. Two models of the latent leading and coincident factors are studied : a model with linear dynamics and a model with Markov-switching dynamics introduced through the leading factor intercept term. Moreover, a possibility of the individual leading variables having different leads over the common coincident indicator is considered. These models - both with linear and with regime-switching dynamics - were applied to the US monthly macroeconomic time series. The business cycle dating resulting from the nonlinear model closely corresponds to the NBER chronology and leads its turning points by 3-5 months.