Estimating the Business Cycle in Korea: A model averaging approach

This paper analyzes how business cycle estimates showing deviations from trends have changed over the past 60 years using quarterly GDP data and 52-time series models. The SIC statistic of the empirical analysis indicates that non-linear models are superior to linear models, especially BBD_AR(0), UC-FP-0, and BBU_AR(0), which have relatively high posterior model probabilities. Therefore, the cyclical estimates weighted by the posterior probabilities of the 52 models show asymmetrically larger declines during recessions. These multi-model-based weighted average cyclical estimates are not significantly different from DFM’s common estimates or principal component analysis estimates using the same set of models. In addition, since 2000, the unemployment rate or the manufacturing average utilization rate has been more closely correlated with the multi-model-based weighted average cyclical estimate than with the cyclical component of the coincident index or HP and bandpass filter estimates. This study also shows that dating a reference cycle has been a challenging task in recent years.