The usual index of leading indicators has constant weights on its components and is therefore implicitly premised on the assumption that the dynamical properties of the economy remain the same over time and across phases of the business cycle. We explore the possibility that the business cycle has phases, for example, recessions, recoveries and normal growth, each with its unique dynamics. Based on this possibility we develop a nonlinear model of the business cycle that combines a number of previous approaches. We model the state of the economy as a latent variable with a threshold autoregression structure. In addition to dependence on its own lags the latent variable is also determined by observed economic and financial variables. In turn these variables are modeled as following a nonlinear vector autoregression with regimes defined by the latent business cycle variable. A Markov Chain Monte Carlo algorithm is developed to estimate the model. Special attention is paid to specification of prior distributions given the large dimension of the model. We also investigate using the business cycle chronology of the NBER to aid in the classification of the latent variable. The two main empirical objectives of the model are to provide more accurate predictions of economic variables particularly at turning points and to describe how the dynamics differ across business cycle phases
