AI Disclosure & S&P 500 Returns — Empirical Study
Links the semantic AI-disclosure factors (EDGAR Research Database) to realized stock
performance (Yahoo Finance) and tests whether AI language in 10-K filings explains
the cross-section of subsequent returns. Full 40-page paper in report/.
Pipeline (run in order)
| Script | Purpose | Key output |
|---|---|---|
01_download_prices.py |
Monthly prices (500 tickers), SPY, market caps | data/prices_monthly.parquet |
02_build_panel.py |
Look-ahead-free forward returns (July formation) + controls; merge | data/merged_panel.parquet |
03_analysis.py |
Sorts, FF6 alphas, two-way FE panel, Fama-MacBeth, post-ChatGPT break | tables/results.json |
06_extended.py |
Placebo, sector-neutral/double sorts, Mag-7 exclusion, alt signals/horizons, PCA, DiD, fwd-vol | tables/results2.json |
07_event_study.py |
ChatGPT (Nov 2022) market-model event study, daily CARs | tables/results_event.json |
04_figures.py, 08_figures_ext.py |
13 publication figures | figures/fig1..13_*.png |
05_make_tables.py, 09_make_tables_ext.py |
12 LaTeX table fragments | report/tab_*.tex |
FF5 + momentum factors fetched into data/ff5_monthly.parquet, data/mom_monthly.parquet.
Report
report/ai_disclosure_returns_full.tex → ai_disclosure_returns_full.pdf (40 pp).
Compile: cd report && tectonic ai_disclosure_returns_full.tex.
(ai_disclosure_returns.tex is the earlier short 6-page version.)
Headline finding
AI-disclosure intensity barely predicted returns for a decade, then was sharply rewarded after ChatGPT (Nov 2022). High−low (Q5−Q1) forward-return spread: +2.4 pp/yr pre-2022 → +14.6 pp/yr post-2022; long-short Sharpe 0.15 → 1.33. Identified by: (1) a placebo — the jump is AI-specific (ESG/climate go negative); (2) an event study — high-AI firms earned +10.1% higher abnormal return in the year after ChatGPT (t=1.97); (3) it survives Mag-7 exclusion (+14.1pp). It is a cross-sectional, regime-dependent re-rating (absorbed by firm FE, halved by sector FE), not within-firm news timing.
Design notes
- Signals: section-weighted cosine similarity to AI anchors (
*_wmean) for adoption/infrastructure/disruption-risk; standardized cross-sectionally each year. - Timing: 10-K(FY=Y) → form June Y+1 → hold July Y+1..June Y+2 (no look-ahead).
- Inference: Newey-West (sorts, FM, monthly LS), double-clustered SE (panel/DiD), Welch t (event study).
- Caveats: 3 post-period cohorts (concentrated in 2024); current-membership survivorship; shares-held-constant size proxy; AI signals collinear (ρ=0.75–0.86); effect operates substantially through a sector (tech) channel.