A1684
Title: Estimating the probability of Federal Reserve policy rate changes using large language models
Authors: Takashi Matsuki - Ryukoku University (Japan) [presenting]
Abstract: While monetary policy decisions made by the Federal Reserve Board (FRB) at FOMC meetings exert a profound influence on both the U.S. and global economies, their occasional divergence from professional economic forecasts can intensify global uncertainty. The uncertainty surrounding FRB monetary policy changes is mitigated by leveraging Large Language Models (LLMs). First, key textual data reflecting the FRB's policy stance (including Statements, Minutes, Press Conference transcripts, the Beige Book, and speeches by the FRB Chair) are quantified using three LLMs (OpenAI-GPT 5.4, Google-Gemini 3.1, and Claude-Sonnet 4.6). From these quantifications, a policy stance indicator is constructed from Statements, Minutes, and Press Conference, a policy forecast indicator from speeches, and an economic assessment indicator from the Beige Book. Next, these indicators are employed to estimate a model of the probability of policy rate changes. Furthermore, by regressing the estimated probabilities of policy rate changes on key macroeconomic variables (such as economic growth rate, inflation rate, and unemployment rate), the macroeconomic factors that most significantly impact policy decisions are identified. The estimated probabilities are also compared with those derived from the FedWatch tool. Finally, the study examines which Language Model offers the most robust performance in forecasting monetary policy changes.
