Data Sources
FRED Macroeconomic Series
| Series | Name | Description |
|---|---|---|
| FEDFUNDS | Fed Funds Rate | Federal funds effective rate (%) |
| CPIAUCNS | CPI | Consumer Price Index |
| PPIACO | PPI | Producer Price Index |
| GDP | GDP | Gross Domestic Product (quarterly, forward-filled) |
| UNRATE | Unemployment | Unemployment Rate (%) |
| PAYEMS | Nonfarm Payrolls | Total Nonfarm Employment |
Speech Data
2,421 US Federal Reserve speeches (1996-2025) from BIS/Gigando dataset with pre-computed sentiment labels.
Analysis Pipeline
Step 1: Rolling Standardization
12-month rolling window z-score standardization for comparability and stationarity.
Step 2: PCA Dimensionality Reduction
| Component | Variance | Interpretation |
|---|---|---|
| PC1 | 48.6% | Macro Strength Index (Fed Rate, Payrolls +; Unemployment -) |
| PC2 | 23.2% | Inflation Index (CPI, PPI +; Fed Rate -) |
Step 3: Structural Break Detection
PELT algorithm with RBF kernel identifies structural breaks. Penalty = 4.
Step 4: Sentiment Aggregation
Monthly hawkish/dovish counts with 12-month rolling standardization.
Step 5: Rolling Regression
36-month rolling window OLS regression of sentiment on macro indices.
Key Parameters
| Parameter | Value | Purpose |
|---|---|---|
ROLLING_WINDOW | 12 | Months for standardization |
REGRESSION_WINDOW | 36 | Months for rolling regression |
PELT_PENALTY | 4 | Breakpoint sensitivity |
RANDOM_STATE | 42 | Reproducibility seed |