Replication Package

Policy Shocks and Algorithmic Infrastructure Dependence

Leippold, Colesanti Senni, and Vaghefi

Overview

This folder contains all code necessary to replicate the results in the paper. Raw and intermediate data files are not included because they contain proprietary WRDS/CRSP data and SEC-API extracts subject to licensing restrictions. The instructions below describe how to obtain the data and reproduce every table and figure.

Prerequisites

Software

• Python >= 3.10
• LaTeX distribution (TeX Live or MiKTeX) for compiling the manuscript
• Google Colab accounts (optional; notebooks run locally with minor path edits)

Data access (not included)

Source	What	How to obtain
WRDS/CRSP	Daily equity returns, market cap, share codes	Academic subscription at https://wrds-www.wharton.upenn.edu/
Kenneth French Data Library	Fama-French 5 factors, industry classifications, NYSE breakpoints	Free download at https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html
SEC EDGAR (via SEC-API)	10-K filing text (Items 1, 1A, 7)	API key from https://sec-api.io/
OptionMetrics (via WRDS)	Implied volatility surface	Included in WRDS subscription
Compustat (via WRDS)	Accounting fundamentals	Included in WRDS subscription

Copy .env.example to .env and fill in your WRDS and SEC-API credentials.

Python setup

pip install -e ".[full]"
# or
pip install -r requirements.txt

Reproduction steps

Step 1: Construct ADI scores

Notebook	Purpose	Inputs	Outputs
ADI_Score_Generator.ipynb	Build ADI from 10-K filings	SEC-API credentials	data/adi_scores.csv
ADI_Sentence_Extractor.ipynb	Extract AI-relevant sentences	10-K text	Sentence-level data

Step 2: Run main event study

Notebook	Purpose	Outputs
ADI_Event_Study_Colab.ipynb	Main event study (Tables 1-4, Figures 1-3)	tables/table1_summary_stats.tex through table4_alphas.tex, figures

Step 3: Robustness and extensions

Notebook	Purpose	Key outputs
ADI_Options_Event_Study.ipynb	IV channel, CAR/IV joint table	table_car_iv_combined.tex
ADI_Compustat_Validation.ipynb	Accounting validation	table_adi_vs_fundamentals.tex, table_decile_monotonicity.tex
ADI_Mechanical_Events.ipynb	Stacked regression, meta-regression, placebo	table_stacked_regression.tex, table_meta_regression.tex, table_placebo.tex
ADI_Component_Shock_Matching.ipynb	Component-event interaction matrix	table_component_event_matrix.tex
ADI_Diversification_Response.ipynb	Real-effects (diversification language)	table_diversification_response.tex
ADI_News_Index_Colab.ipynb	News-based event validation	table7_news_index.tex

Step 4: Transformer comparison (Internet Appendix)

Notebook	Purpose
ADI_Transformer_v2_Extract.ipynb	Extract training data for ModernBERT
ADI_Transformer_v2_Classify.ipynb	Fine-tune ModernBERT classifier
ADI_Transformer_v2_ModernBERT.ipynb	Score filings with transformer
ADI_Transformer_Event_Study.ipynb	Quadrant analysis, horse-race regressions

Step 5: Compile manuscript

pdflatex 00_master_paper.tex
bibtex 00_master_paper
pdflatex 00_master_paper.tex
pdflatex 00_master_paper.tex

Step 6: Consistency check

Notebook	Purpose
ADI_Consistency_Check.ipynb	Verify all tables match manuscript numbers

File inventory

replication/
  README.md              <- This file
  .env.example           <- Template for credentials (no secrets)
  requirements.txt       <- Python dependencies
  pyproject.toml         <- Package configuration
  src/adi/               <- Python package (scoring, event study, extraction)
  notebooks/             <- Jupyter notebooks (analysis pipeline)
  tests/                 <- Unit tests
  tables/                <- Generated LaTeX tables
  figures/               <- Generated figures
  sections/              <- LaTeX manuscript sections
  00_master_paper.tex    <- Master LaTeX file
  biblio.bib             <- Bibliography
  LICENSE                <- MIT license

What is NOT included

• Raw data files (data/): CRSP returns, Compustat fundamentals, OptionMetrics IV, SEC filing text. These are proprietary and must be obtained with valid WRDS/SEC-API credentials.
• Credentials (.env): API keys and passwords.
• Presentation materials: Slides, keynote scripts, transcripts.
• Working notes: REVISION_top_tier.md, CLAUDE.md, results_unified_sample.md.