Turning Points Detection of Business Cycles: A Model Comparison

Abstract

This paper compares three models, namely Markov-Switching Model (MS), Self-Exciting-Threshold-Autoregressive Model (SETAR) and Hidden Markov Model (HMM), for detecting turning points of business cycles. The aim is to find out which of the three models produces the most reliable signals of turning points. For this purpose we apply these models to the US Industrial Production Index (IPI) and make a comparison of the results of the different models. We observe that the HMM performs best in our testing environment. The turning points detected by the HMM correspond well to the different phases of the business cycle. In comparison, the MS and SETAR produce rather instable signals.

Keywords: Turning point detection, business cycle, Markov switching, hidden markov model, US industrial production, SETAR