A Statistical Study of High-Frequency Crypto Reactions to FOMC NLP Tone

Author: Thibault Charbonnier
Date: November 2025
Contact: thibault.charbonnier@ensae.fr

🔎 The full empirical study, methods, figures and discussion are in the Jupyter notebook: demo.ipynb (see repository root).

Abstract

This empirical study examines various cryptocurrencies' intraday reactions to U.S. monetary-policy communication and reaches a conclusion consistent with recent literature: crypto returns appear orthogonal to monetary policy at high frequency. Using an NLP-based tone score on FOMC statements and press-conference transcripts to proxy policy stance, we find no statistically significant predictive power for the sign nor magnitude of returns (10–60m horizons). However, distributions differ in the tails: most dovish vs most hawkish events display significant asymmetry (heavier right vs left tails). Results should be read with caution given structural data limits (few FOMC events; limited minute-level crypto history).

Project Architecture

Data & Reproducibility

FOMC documents are fetched from the Federal Reserve websites; structure varies by period (fallbacks implemented).

Crypto minute data are sourced from exchange archives; coverage varies around listing dates.

See demo.ipynb for the full pipeline and analysis steps (bucketing, skew/Δ-quantiles, OLS/logit, plots).

Citation

If you use this project, please site : Charbonnier, T. (2025). A Statistical Study of High-Frequency Crypto Reactions to FOMC NLP Tone. GitHub repository.

Contact & Issues

Questions / bugs: open an issue on GitHub or email thibault.charbonnier@ensae.fr or linkedin https://www.linkedin.com/in/thibault-charbonnier