Economic Sentiment from News: Predicting Regional Unemployment in Germany

Abstract

Economic sentiment influences economic outcomes by shaping public perceptions and behaviors. Traditional survey-based indices, such as the ZEW Economic Sentiment Index, GfK Consumer Climate Indicator, and IFO Business Climate Index, have limitations like limited participants and infrequent updates. This thesis constructs a continuously updated economic sentiment index for Germany's regions using natural language processing (NLP) and machine learning (ML) on the RegNeS-database of regional news.

The study investigates the relationship between this news-based sentiment index and regional unemployment rates, finding a significant negative correlation. This suggests that changes in economic sentiment can predict changes in unemployment, offering timely and region-specific economic insights.

Despite its contributions, the study acknowledges limitations, including short temporal coverage and potential NLP inaccuracies. Future research could expand data sources, refine NLP techniques, and integrate traditional and news-based indices.

In conclusion, this thesis provides a novel approach to measuring economic sentiment at the regional level, offering valuable insights for policymakers and economists to facilitate informed decision-making.