This paper introduces a groundbreaking study on the influence of Federal Open Market Committee (FOMC) communications on financial markets by constructing the largest tokenized and annotated dataset of FOMC documents, including speeches, meeting minutes, and press conference transcripts. The research develops a novel hawkish-dovish classification task to analyze monetary policy stances, benchmarking various pre-trained language models, with RoBERTa-large identified as the top performer. Using this model, the authors create a measure of monetary policy stance for FOMC document release days, enabling quantitative assessment of policy impacts.

The study evaluates this measure by examining its effects on treasury markets, stock markets, and macroeconomic indicators, demonstrating its utility in understanding market dynamics driven by policy communications. All resources, including the dataset, models, and code, are publicly released under a CC BY-NC 4.0 license, promoting reproducibility and advancing research in financial natural language processing. This work bridges gaps in quantitative finance by providing tools for systematic analysis of central bank communications, with implications for investors, policymakers, and academics.