Forecasting German GDP Using Alternative Factor Models Based on Large Datasets

Abstract

This paper discusses the forecasting performance of alternative factor models based on a large panel of quarterly time series for the german economy. One model extracts factors by static principals components analysis, the other is based on dynamic principal components obtained using frequency domain methods. The third model is based on subspace algorithm for state space models. Out-of-sample forecasts show that the prediction errors of the factor models are generally smaller than the errors of simple autoregressive benchmark models. Among the factors models, either the dynamic principal component model or the subspace factor model rank highest in terms of forecast accuracy in most cases. However, neither of the dynamic factor models can provide better forecasts than the static model over all forecast horizons and different specifications of the simulation design. Therefore, the application of the dynamic factor models seems to provide only small forecasting improvements over the static factor model for forecasting German GDP.