Out-of-sample forecasting of unemployment rates with pooled STVECM forecasts
In this paper we use smooth transition vector error-correction models (STVECMs) in a simulated out-of-sample forecasting experiment for the unemployment rates of the four non-Euro G-7 countries, the U.S., the U.K., Canada, and Japan. For the U.S., pooled forecasts constructed by taking the median value across the point forecasts generated by the linear and STVECM forecasts perform better than the linear AR(p) benchmark, and more so during business cycle expansions. Such pooling leads to statistically significant forecast improvement for the U.K. across the business cycle. ''Reality checks'' of these results suggest that they do not stem from data snooping.