Forecasting with dynamic factor models

The validity of previous findings that dynamic factor models are useful for macroeconomic forecasting is of great importance for subsequent studies which use these models not only as a starting point for further developments but also as a benchmark for the evaluation of the forecasting performance of these further developments. Reanalyzing a standard macroeconomic dataset, we do not find any evidence corroborating the usefulness of dynamic factor models. We therefore explore two possible ways for improvement. First, we try to find those factors which have the greatest predictive power and then we try to utilize frequency-domain information. Our empirical results indicate that only the latter attempt is promising. Focusing on the low-frequency components of the macroeconomic time series and disregarding the high-frequency components can actually improve the forecasting performance.