Forecasting industrial production in Germany: The predictive power of leading indicators

We investigate the predictive power of several leading indicators in order to forecast industrial production in Germany. In addition, we compare their predictive performance with variables from two competing categories, namely macroeconomic and financial variables. The predictive power within and between these three categories is evaluated by applying Dynamic Model Averaging (DMA) which allows for timevarying coefficients and model change. We find that leading indicators have the largest predictive power. Macroeconomic variables, in contrast, are weak predictors as they are even not able to outperform a benchmark AR model, while financial variables are clearly inferior in terms of their predictive power compared to leading indicators. We show that the best set of predictors, within and between categories, changes over time and depends on the forecast horizon. Furthermore, allowing for time-varying model size is especially crucial after the Great Recession.