Federal Reserve · NLP · Machine Learning

FedSignal

Predicts US Treasury yield movements following FOMC meetings by combining NLP sentiment analysis of Federal Reserve statements with macroeconomic regime classification and gradient boosting.

GitHub

Latest Signal

May 2025

→ Flat

Predicted 1Y yield change: -0.04%

Transition

Dec 2024

✓ +0.04

Jan 2025

✓ +0.07

Mar 2025

✓ +0.03

NLP Sentiment

FOMC Hawkishness Over Time

Hawkish Dovish

Walk-Forward CV

Predicted vs Actual Yield Change

GoldilocksReflationOverheatingStagflationTransitionSecular StagnationDeflationary Recession

Model Performance

GBR · Walk-Forward CV · 67 OOS Predictions

OOS R²

+0.314

Dir. Acc.

67.7%

MAE

0.183%

Feature Importance (Gini)

2Y − FF Spread

27.1%

NLP Hawkishness

19.8%

NLP vs Regime

17.1%

Core PCE YoY

14.2%

Regime (Ordinal)

9.8%

NLP Momentum

7.9%

Sent. Dispersion

4.1%

Live Demo

Statement Scorer

Select an FOMC statement excerpt:

“The Committee decided to raise the target range for the federal funds rate by 75 basis points.”

Hawkishness Score

Very Hawkish

Dovish (0.0)0.95Hawkish (1.0)

Scores generated by a DistilBERT model fine-tuned on 88 labeled FOMC sentences.