Replication package: reaction, not anticipation — macroeconomic and monetary-policy predictors of seven U.S. stock indices

# Replication package—"Reaction, not anticipation"Data, code and results for the study of macroeconomic and monetary-policy predictors ofseven U.S. stock indices (2000–2023), under a leakage-audited chronological protocolwith a distribution-matched placebo control.## Contents```macro_pipeline_notebook.ipynb executed pipeline (data build, models, probes)macro_pipeline/├── dataset_2000_2023.parquet full daily panel├── sub_2015_2023.parquet sub-sample used for the recent-period checks├── train.parquet / val.parquet / test.parquet chronological splits└── leakage_audit.csv random vs chronological validation AUCmacro_results/├── RECAP_canonical.json headline configuration and results├── RECAP_extended.json extended robustness results├── baseline_predictability.csv AUC by index, horizon and model├── diagnostic_blocs.csv lift by predictor block├── probe2_regime.csv regime dependence├── probe3_lag.csv surprise–return association by lag├── probe4_placebo.json placebo comparison├── placebo_lifts_*.npy raw placebo lift draws├── extended_*.csv lag, sign, specification and sub-period robustness└── overlap_diagnostic.json overlapping-window diagnosticpaper3/├── placebo_significance.csv permutation test: 21 index-horizon configurations,│ 500 draws, empirical p-values and 95% intervals└── significance_test.py the cell that produces the file above```## Headline resultTwelve of the 21 index-horizon configurations are statistically indistinguishable from adistribution-matched placebo (two-sided permutation test, 500 draws, +1 correction ofDavison and Hinkley, 1997). The nine exceedances occur only at the ten- and twenty-dayhorizons and none at one day, which is the signature of overlapping-window inflationrather than genuine forecastability.## Data sources (not redistributed)* Monetary-policy surprises: Bauer and Swanson (2023) replication data* Predictor variables: Goyal and Welch, `PredictorData` (updated annually)* Macroeconomic indicators: FRED* Index prices: Yahoo FinancePlace the two source spreadsheets in the working directory before running the notebook.## Reproduce```bashpip install -r requirements.txt# run macro_pipeline_notebook.ipynb, then:python paper3/significance_test.py```Note: re-running the full pipeline retrains the models and may not reproduce thepublished numbers exactly; the archived `macro_results/` files are the source of truth.## LicenseCode under MIT. Third-party data remain under their providers' licenses.