Appendix ¡ — Cohort Research Projects
Fifteen empirical-finance papers — ASSIP 2026
Over the summer, the ASSIP 2026 cohort builds a portfolio of real, data-backed empirical finance papers — each with a research question, data from WRDS or public sources, an identification strategy, regression tables, figures, and a written draft. Results are reported exactly as they came out: findings and honest nulls alike. Below is the portfolio, by project.
The fifteen papers
| # | Paper | Headline result |
|---|---|---|
| 1 | Cognitive vs. Physical Automation: AI and Robotics Patents | AI patents track employment growth; robotics tracks firm value |
| 2 | The Diffusion of AI into Medicine | Health incumbents hold about 47% of medical-AI patents; the market values them |
| 3 | Clawbacks and CEO Risk-Taking after the 2023 SEC Rule | Null — the one significant result is a size-and-COVID artifact that fails pre-trends |
| 4 | Do Anomalies Die When They Are Published? | Anomaly long-short returns fall about half after publication |
| 5 | Import Exposure and the April-2025 Tariff Shock | Import-exposed firms fell on the shock and rebounded on the pause |
| 6 | Does Housing Re-Price Disaster Risk Only Where People Believe? | Null — no robust belief-conditional re-pricing |
| 7 | BNPL Enters the Credit File | Null — no furnisher-specific complaint spike; the surge is sector-wide |
| 8 | Is the LLM Reading the 10-K or Remembering the Stock? | Null at scale — no look-ahead contamination detected |
| 9 | Did ChatGPT Reprice AI-Exposed Labor? | AI-exposed firms earned abnormal returns; the sign flips by economic role |
| 10 | What Cracked a Bank in March 2023? | Uninsured deposits, not hidden bond losses, drove the SVB-week crash |
| 11 | Do CHIPS Act Awards Move the Awardee’s Stock? | Null — no average effect; the apparent size-scaling is a single outlier |
| 12 | Did the 2025 Half-Cent Tick Tighten Spreads? | Null — the difference-in-differences fails its placebo and parallel-trends |
| 13 | Where Does Price Discovery Happen: Kalshi vs. Polymarket? | Polymarket leads price discovery on identical contracts |
| 14 | What Kind of Layoff Does the Market Punish? | Layoff announcements cost about 1.6%, driven by firm materiality |
| 15 | Cohort capstone: A Policy-Shock Event-Study Atlas of 2022–2025 | Firm differentiation scales with shock size; the SVB shock uniquely propagates |
What the tally says
Nine significant findings, five diagnosed nulls, and one that scaled to a null — with zero fabricated numbers. That mix is what honest research looks like. A project that reports fifteen “significant” results out of fifteen is the one to distrust; a diagnosed null (a failed placebo, a single leverage point, a non-parallel pre-trend) is a real result, and it is what makes the significant findings credible.
How the work is divided
The heavy computational scaffolding — building the data pipeline, running the regressions, drafting the tables — is done with AI assistance. Each student then owns the part a machine cannot do reliably: hand-collecting and hand-coding documents, pinning exact event dates, resolving messy entity crosswalks, and making the judgment calls that determine whether a result holds up. That hand-work is what makes each paper trustworthy.