摘要：This paper sets up a common unobserved factor model with smooth transition autoregressive dynamics. This model is compared to the already classical common factor model with regime-switching. Both models' in-sample and out-of-sample performance in terms of capturing and predicting the business cycle turning points is evaluated. The comparison of the model-derived probabilities to the NBER business cycle dating shows statistically equivalent in-sample forecasting accuracy of these techniques. The common factor model with exponential STAR outperforms the model with logistic STAR and that with Markov switching in terms of out-of-sample prediction with up to 3 month horizon.
