S&P 500 Firm-Day Panel with LLM Exposure Index, 2019–2024

742,444 firm-day observations for 498 S&P 500 constituents from January 2019 to December 2024. This dataset is the complete replication archive for a paper analyzing the impact of large language model adoption on market efficiency and systemic fragility. It includes raw data, pre-computed analysis files, and Python scripts for full reproduction of the study's results.

Use Cases

• Replicate causal difference-in-differences designs based on the described ChatGPT outage event study.
• Analyze the decay of post-earnings-announcement drift based on the LLM Exposure Index.
• Study correlated-error and crash-risk tests using the provided firm-day panel data.
• Construct firm-level textual features for financial analysis using the included SEC EDGAR data and sentiment dictionaries.

Strengths

• Includes 742,444 firm-day observations covering a 6-year period that spans the pre- and post-ChatGPT launch era.
• Provides a complete replication pipeline with 17 pre-computed CSV files and numbered Python scripts for full result reproduction.
• Integrates multiple data sources, including Yahoo Finance price data, Fama–French factors, and SEC EDGAR textual features.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for individual component files is unknown, which may limit suitability assessment.

Provenance

Source

Harvard Dataverse

Collection Method

Assembled from Yahoo Finance (yfinance API), Kenneth French Data Library, SEC EDGAR APIs, and public historical constituent records.

Time Range

January 2019 to December 2024

Freshness

Last updated 2026-06-16 16:51:15; freshness should be verified.

Geography

Firms listed in the S&P 500 index (United States focus).