Sentiment Analysis with Large Language Models Applied to the Federal Reserve Beige Book

Abstract

We present the application of Large Language Models (LLMs) to perform sentiment analysis on the Beige Book of the United States Federal Reserve. These reports are a critical qualitative resource for understanding the economic conditions in the United States and are instrumental in the decision-making of the Federal Reserve. We use different LLM models over a dataset of more than ten years to evaluate their effectiveness in capturing sentiment in the reports. Our findings show that certain sections of the Beige Books are a more accurate representation of the overall sentiment than others. We compare the measured sentiment with the macroeconomic time series. Our work highlights a potential application of LLMs for economic forecasting and is a novel approach to studying qualitative data critical to monetary policy in the United States.

Keywords: large language model, llm, federal reserve, sentiment analysis, beige book

JEL Classification: C61, E58