We use natural language processing methods to quantify the sentiment expressed in the Federal Reserve's anecdotal summaries of current economic conditions in the national and 12 Federal Reserve District-level economies as published eight times per year in the Beige Book since 1970. We document that both national and District-level economic sentiment tend to rise and fall with the US business cycle. But economic sentiment is extremely heterogeneous across Districts, and we find that national economic sentiment is not always the simple aggregation of District-level sentiment. We show that the heterogeneity in District-level economic sentiment can be used, over and above the information contained in national economic sentiment, to better forecast US recessions.
Read full abstract