π΅οΈββοΈ Insider Trading Intensity and 8-K Filings
π EPFL β Master in Financial Engineering, Year 2 (2025) π₯ Team: Matthias Wyss, William Jallot π¬ Supervisor: Prof. Pierre Collin-Dufresne π Final Report: Report π GitHub Repository: GitHub
This project investigates the dynamics of Informed Trading Intensity (ITI)βa machine-learning-based metricβaround corporate disclosures via SEC Form 8-K filings. The goal is to analyze the relationship between informed trading activity and asset prices during major corporate events.
π Key Project Highlights
- Large-Scale Event Study: Conducted a study on 99,384 US corporate 8-K filings (2001β2024) to analyze the relationship between ITI and asset prices.
- Abnormal Returns Calculation: Estimated via a Fama-French 5-factor + Momentum model, identifying significant volatility spikes around report dates.
- Informed Trading Validation: Discovered that ITI significantly increases prior to material events (e.g., M&A, financial results), validating information leakage and informed trading hypotheses.
- Statistical Rigor: Performed robust statistical validation using placebo tests (randomized event dates) to confirm the significance of volatility and ITI signals.
π Performance & Market Analysis
The results demonstrate that informed trading contains predictive power for prices, with volatility spikes closely tracking ITI movements.
| Metric | Observation |
|---|---|
| ITI Spike Window | Sharp rise at report date (π=0) followed by a rapid decline after public disclosure. |
| Volatility Correlation | Peak correlation of ~0.28 between Abnormal ITI and Absolute Returns at π=1. |
| Sentiment Impact | Using Mistral AI to summarize filings before FinBERT analysis provides a clearer separation of abnormal returns. |
Major Result: Filings including detailed exhibits (Item 9.01) are the primary drivers of informed trading activity.
π Tools & Libraries:
- SEC EDGAR & WRDS: Massive extraction of textual and financial data.
- Mistral-7B & FinBERT: NLP pipeline for automated summarization and financial sentiment analysis.
- Python & Polars: Efficient processing of high-volume datasets.
π§ Applied Techniques:
- Factor Modeling: Controlling for systematic risk factors including Size, Value, and Momentum.
- Volatility Analysis: Utilizing absolute abnormal returns as a proxy for market activity intensity.
- Placebo Testing: Temporal randomization to ensure the statistical significance of findings.