Cash vs. Stock: What I Found in 45 Acquisitions
Research · About 4 min read
My Pfizer study raised a question: would a difference between cash- and stock-funded acquisitions appear in a broader sample? The saved dataset contains 45 candidate events. Its corrected figures show a gap, but an audit also exposed limits in the classifications and event dates.
What can be reproduced
The table below can be reproduced from the saved cumulative abnormal returns (CARs) after changing Coeur Mining's financing label from cash to stock. The original saved file still had the old label. Both versions are available below, so the correction is visible.
Each window uses fewer than 45 observations because some return values are missing. The cash and stock counts shown here are the observations used in each comparison.
| Window | Cash n | Stock n | Cash CAR | Stock CAR | Gap | Gap p-value | 95% CI for gap |
|---|---|---|---|---|---|---|---|
| [-1, +1] | 17 | 20 | +0.77% | -3.75% | +4.52pp | 0.0006 | +2.14 to +6.90pp |
| [-3, +3] | 17 | 18 | +1.08% | -3.64% | +4.72pp | 0.016 | +0.97 to +8.47pp |
| [-5, +5] | 16 | 17 | +0.95% | -4.26% | +5.20pp | 0.040 | +0.27 to +10.14pp |
Cash-minus-stock gaps are positive in all three windows. The p-values come from a Welch two-sample test that reproduces the published figures; the original testing script was not recovered. The 95% intervals were calculated for this update using the Welch standard error and degrees of freedom. They are unadjusted for multiple comparisons. These overlapping windows are not independent replications, and the candidate events have not all been checked for duplication or correct financing labels.
What the verification record says
The handoff records twelve selected entries: ten financing labels supported by extracted filing quotes, one correction for Coeur / SilverCrest, and one Cheniere entry examined for possible duplication rather than financing type. That is eleven reported financing checks, not twelve. The remaining 34 financing labels, including that Cheniere entry, do not have a completed check in this record.
The verification record includes extracted filing quotes and a review of the two Cheniere filings. It was reconstructed from earlier project records rather than re-verified during this update. Whether the two Cheniere entries represent separate transactions remains unresolved.
How the saved code works
- Classification: the harvest assigned cash or stock from its search-phrase bucket. The later text filter checked announcement language, but did not use its cash/stock term flags to validate financing labels.
- Event timing: the return code uses the first trading day on or after the filing date. It does not establish the original announcement time or adjust for after-hours news.
- Return model: adjusted closing prices from yfinance, simple daily returns, and an OLS market model against ^GSPC. The estimation slice is 200 aligned observations at positions [-220, -20) relative to the event.
- Windows: summed abnormal returns over [-1,+1], [-3,+3], and [-5,+5], expressed as percentages.
- Earnings screening: the surviving script reads recent SEC Item 2.02 filings and approximates trading-day gaps using calendar days multiplied by 5/7. Older earnings history and other company news are not comprehensively screened.
The written method describes a $10 billion current-market-cap filter. Its code and intermediate data were not recovered. The saved earnings script marks affected windows, but the step that removed their returns from the final file is also missing.
What I would not conclude yet
This is an exploratory association in saved data, not evidence that payment type causes announcement returns. Deal size, timing, sector, repeated acquirers, and classification errors could affect the comparison. A stock-payment signal is one possible interpretation, not a mechanism this study establishes.
The earlier write-up's 54% filtering claim is unconfirmed: its notes report both 54.2% and 60 of 107, which do not agree. The intermediate file needed to resolve that discrepancy is missing, so I have removed the percentage as a finding.
Data and reproduction
- Original saved data (CSV)
- Derived data with the Coeur correction (CSV)
- Statistics reproduction script and scope and instructions
These materials reproduce the table from saved returns. They do not reproduce the full route from SEC filings and market prices. Fully checking the candidate events, rebuilding missing stages, and documenting uncertainty are the next steps before a broader claim would be justified.