Research Overview
Investigating whether Federal Reserve speech sentiment correlates with macroeconomic conditions using 2,421 US Fed speeches (1996-2025) and six FRED indicators.
Key Finding
Near-zero correlation (0.005) between CB speech sentiment and macro indices - challenging assumptions about central bank communication effectiveness.
Key Results
0.005
Macro-Hawkish Corr
87%
Variance (3 PCs)
2,421
US Fed Speeches
-0.43
Hawkish Autocorr
Quick Start
git clone https://github.com/Digital-AI-Finance/CB-Speeches-Analysis.git
cd CB-Speeches-Analysis && pip install -r requirements.txt
python -c "from analysis.run_all import run_pipeline; run_pipeline()"
cd charts && python run_all_charts.pyMethodology
Data Sources
- FRED: Fed Funds Rate, CPI, PPI, GDP, Unemployment, Nonfarm Payrolls
- Speeches: BIS/Gigando dataset with NER and sentiment labels
Pipeline
- PCA: 6 macro vars to PC1 (Macro Index) + PC2 (Inflation Index)
- PELT: Structural breaks with RBF kernel (penalty=4)
- Regression: 36-month rolling betas and R-squared
Implications
- Policy Independence: CB communication may be forward-looking, not reactive
- Narrative Disconnect: Speech sentiment does not mirror macro reality
- Mean Reversion: Strong negative autocorrelation (-0.43) suggests rapid narrative shifts