We study the forecasting performance of three alternative large scale approaches using a dataset for Germany that consists of 123 variables in quarterly frequency. These three approaches handle the dimensionality problem evoked by such a large dataset by aggregating information, yet on different...