Forecasting GDP at the regional level with many predictors

In this paper, we assess the accuracy of macroeconomic forecasts at the regional level using a unique data set at quarterly frequency. We forecast gross domestic product (GDP) for two German states (Free State of Saxony and Baden-Württemberg) and Eastern Germany. We overcome the problem of a "data-poor environment" at the sub-national level by including more than 300 international, national and regional indicators. We calculate single–indicator, multi–indicator and pooled forecasts. Our results show that we can significantly increase forecast accuracy compared to an autoregressive benchmark model, both for short- and long-term predictions. Furthermore, our best leading indicators describe the specific regional economic structure better than other indicators.