Measuring business cycles with a dynamic Markov switching factor model: an assessment using Bayesian simulation methods

Abstract

A Markov switching common factor is used to drive a dynamic factor model for important macroeconomic variables in eight countries. Bayesian estimation of the model is based on Markov chain Monte Carlo simulation methods which yield inferences about the unobservable path of the common factor, the latent variable of the state process and all model parameters. Additionally, simulation based filtering provides us with samples from the prediction density that can be used for model diagnostics and specification tests. The mean posterior state probabilities are used to date business cycle turning points that follow quite closely previous datings reported in the literature. Moreover, we test the Markov switching against a no-switching specification by means of a Bayes factor. The evidence proves to be quite favorable for the Markov switching model.