Trillion Dollar Words: A New Financial Dataset, Task & Market Analysis

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6664–6679 July 9-14, 2023 ©2023 Association for Computational Linguistics Trillion Dollar Words: A New Financial Dataset, Task & Market Analysis Agam Shah , Suvan Paturi , Sudheer Chava Georgia Institute of Technology Abstract Monetary policy pronouncements by Federal Open Market Committee (FOMC) are a major driver of financial market returns. We construct the largest tokenized and annotated dataset of FOMC speeches, meeting minutes, and press conference transcripts in order to understand how monetary policy influences financial mar- kets. In this study, we develop a novel task of hawkish-dovish classification and benchmark various pre-trained language models on the proposed dataset. Using the best-performing model (RoBERTa-large), we construct a mea- sure of monetary policy stance for the FOMC document release days. To evaluate the con- structed measure, we study its impact on the treasury market, stock market, and macroeco- nomic indicators. Our dataset, models, and code are publicly available on Huggingface and GitHub under CC BY-NC 4.0 license1. 1 Introduction On August 26th, 2022, FOMC Chair Jerome H. Powell gave an 8-minute long speech at Jackson Hole which immediately resulted in an almost $3 Trillion USD decline in U.S. equity market value that day. The speech was followed by more than $6 Trillion USD loss in equity valuation over the next 3 days. Drastic market shifts to the Fed’s pronouncements indicate just how important the FOMC communications have become and high- light the need for a model which can capture the policy stance from Fed-related text. The Federal Open Market Committee (FOMC) is a federal organization responsible for controlling U.S.’s open market operations and setting interest rates. It tries to achieve its two main objectives of price stability and maximum employment by controlling the money supply in the market. Given Correspondence to Agam Shah {ashah482@gatech.edu} 1The fine-tuned model and data are available on the Hug- gingface. The code is available on FinTech Lab GitHub. the market condition (employment rate and infla- tion), the Fed either increases (dovish), decreases (hawkish), or maintains the money supply2 (neu- tral). To understand the influence the FOMC has on the different financial markets, we need to extract its monetary policy stance and the corresponding magnitude from official communications. Utilizing the traditional sentiment analysis model, which classifies text into positive vs neg- ative, one can’t extract policy stance. A sentence that has the word "increase" could either be dovish or hawkish without a clear negative connotation. For example, the word "increase" with the word "employment" means the economy is doing well, but the word "increase" with the word "inflation" is negative for the economy. Current SOTA fi- nance domain-specific language models (Araci, 2019; Shah et al., 2022) trained for sentiment anal- ysis find both cases to be positive, which is inac- curate. The performance analysis for FinBERT (Araci, 2019) model is provided in Appendix A. This problem creates a need to develop a new task for hawkish vs dovish classification accompanied …

Content truncated to 500 words