Paper: arXiv 2305.07972

Authors: Agam Shah, Suvan Paturi, Sudheer Chava

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 markets. 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 measure of monetary policy stance for the FOMC document release days. To evaluate the constructed measure, we study its impact on the treasury market, stock market, and macroeconomic indicators. Our dataset, models, and code are publicly available on Huggingface and GitHub under CC BY-NC 4.0 license.

Complexity vs Empirical Score

  • Math Complexity: 3.5/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 8.0)

Why this score: The paper focuses on building and benchmarking NLP models (RoBERTa-large, FinBERT) on a novel annotated dataset, with empirical validation on treasury and stock market impact, but lacks complex mathematical derivations or theoretical modeling.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Monetary Policy & Market Impact"] --> B{"Key Methodology"}
  
  B --> C["Data Collection<br>FOMC Speeches & Transcripts"]
  C --> D["Hawkish-Dovish<br>Classification Task"]
  D --> E["Model Training &<br>Benchmarking (RoBERTa-large)"]
  E --> F["Policy Stance<br>Measurement"]
  
  F --> G{"Evaluation / Market Impact"}
  G --> H["Treasury Market<br>Analysis"]
  G --> I["Stock Market<br>Analysis"]
  G --> J["Macroeconomic<br>Indicators"]
  
  H --> K["Key Findings:<br>Policy Measure Predicts Asset Returns"]
  I --> K
  J --> K

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper introduces a large tokenized, annotated FOMC text dataset, a hawkish‑dovish classification task, an aggregate monetary policy stance measure, and a simple trading strategy that outperforms the QQQ benchmark.

Source sentence
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 markets.

Assumes 2

  • Trading results are reported without transaction fees, assuming low‑frequency trading incurs negligible costs. execution
    Source sentence
    In our trading strategy construction, we do not include transaction fees as it involves low-frequency trading.
  • Sentence splitting is performed using a simple rule‑based approach, assuming this is sufficient for handling opposite tones. data
    Source sentence
    In dataset construction, while splitting sentences, we use a simple rule-based approach.

Data

  • FOMC Meeting Minutes · Federal Reserve text-news public
    1996-2022
    Source sentence
    Meeting Minutes 1996 - 2022 214 44,923 1,346,674 29.98
  • FOMC Press Conferences · Federal Reserve text-news public
    2011-2022
    Source sentence
    Meeting Press Conferences 2011-2022 63 19,068 468,941 24.59
  • FOMC Speeches · Federal Reserve text-news public
    1996-2022
    Source sentence
    Speeches 1996-2022 1,026 108,463 3,222,285 29.71

Recognised sources: Yahoo Finance

Costs

Not modeled A trading or execution result is reported without costs.

Source sentence
In our trading strategy construction, we do not include transaction fees as it involves low-frequency trading.

Finds 3

  • Finetuned PLMs, especially RoBERTa‑large, outperform rule‑based and LSTM models by a significant margin across all datasets. performance
    Source sentence
    Finetuned PLMs outperform rule-based model and LSTM models by a significant margin.
  • The constructed monetary policy stance measure shows statistically significant correlations with CPI and PPI changes. statistical
    Source sentence
    Table 6: Correlation of immediate next CPI and PPI data with our measure. All values are statistically significant.
  • A simple trading strategy based on the stance measure outperforms the QQQ index benchmark over the last decade. performance
    Source sentence
    We also propose a simple trading strategy that outperforms the high benchmark set by the QQQ index over the last decade.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 13 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 9 papers in the archive (8 by independent authors): Open-FinLLMs: Open Multimodal Large Language Models for … (uses its method or data); SusGen-GPT: A Data-Centric LLM for Financial NLP and … (uses its method or data); FinBloom: Knowledge Grounding Large Language Model with … (uses its method or data)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.