Economic Forecasting With German Newspaper Articles

Abstract

We introduce a new leading indicator for the German business cycle based on the content of newspaper articles from the Süddeutsche Zeitung. We use the rapidly evolving technique of Natural Language Processing (NLP) to transform the content of daily newspaper articles between 1992 and 2021 into topic time series using an LDA model. These topic time series reflect broad areas of the German economy since 1992, in particular the recession phases of the High-Tech Crisis, the Great Financial Crisis and the Covid-19 pandemic. We use the Newspaper Indicator in a Probit model to demonstrate that our data can be considered as a new leading indicator for predicting recession periods in Germany. Moreover, we show in an out-of-sample forecast experiment that our newspaper data have a predictive power for the German business cycle across 12 target variables that is as strong as established survey indicators. Industrial Production, the Stock Market Index DAX, and the Consumer Price Index for Germany can even be predicted out-of-sample more accurately with our newspaper data than with survey indices of the Ifo Institute and the OECD.