Dating U.S. Business Cycles with Macro Factors

I propose a framework for the assessment of current business conditions using a factor-augmented autoregressive probit model where the dependent variable is the state of the economy as defined by the NBER. Results show that latent common factors estimated by principal components analysis from a large number of macroeconomic series have important predictive power for NBER recession dates. The models generate in-sample recession probabilities that almost perfectly reproduce NBER dates. A pseudo out-of-sample forecasting exercise, designed to approximate real time conditions, shows that predicted recession probabilities consistently rise during subsequently declared NBER recession dates. With the appropriate classification rule, the models exhibit good performance as real time dating algorithms. The latent variable in the probit model can be interpreted as an index of business conditions which is used to assess the strength of an expansion or the depth of a recession.