CEO vs CFO Insider Purchases: No Universal CFO Premium

The appealing version of the CEO-versus-CFO hypothesis is easy to tell. CEOs see the strategic map; CFOs see the cash flows. If job title captures a repeatable information advantage, a model should be able to assign more weight to one role before observing the outcome.

The data does not support that shortcut.

In the July 29, 2026 release of the InsiderAlpha Form 4 Research Database, CEO and CFO purchase events produced similar descriptive outcomes. CEOs had a slightly higher median 180-day return in the broad sample. Within companies that had repeated purchases by both roles, CFO outcomes did not show a reliable advantage.

The defensible conclusion is not that CEOs are better market timers. It is that title alone is too weak a feature to justify a universal CFO premium.

Research question

We tested a narrow question: after a reported code P transaction, did the security's subsequent price path differ depending on whether the normalized reporting-owner role was CEO or CFO?

The SEC defines code P as an open-market or private purchase. It should not automatically be described as an exchange-only purchase. The official definition appears in the SEC Form 4 instructions.

This is an observational event study, not a trading backtest. Returns are measured from the transaction-date market-price baseline in the dataset. They are not benchmark-adjusted, do not begin after public filing acceptance, and do not include transaction costs.

Sample construction

The analysis uses the versioned July 29, 2026 release and applies the following rules:

1. Keep transactions classified as code P.
2. Keep normalized roles CEO and CFO.
3. Start on January 1, 2007, when historical coverage becomes materially denser.
4. Collapse multiple transaction lines to one ticker–reporting owner–transaction date event.
5. Use the median price outcome if multiple lines map to the same event.
6. Require a calculated 180-day market-price outcome.

This produces 1,465 CEO events across 265 issuers and 587 CFO events across 180 issuers. These are event counts, not raw filing-line counts.

Broad-sample results

Normalized role	Buyer events	Issuers	Median 30D	Win rate 30D	Median 90D	Win rate 90D	Median 180D	Win rate 180D
CEO	1,465	265	+1.98%	59.3%	+5.29%	62.5%	+8.22%	65.7%
CFO	587	180	+2.44%	60.0%	+4.79%	62.5%	+7.41%	64.2%

The distributions overlap substantially. At 180 days, the middle 50% of CEO outcomes ran from -4.39% to +25.94%; the CFO interquartile range ran from -5.87% to +22.37%. A difference of less than one percentage point in the sample median is small relative to that dispersion.

The within-company check

The broad comparison can be distorted by issuer composition. CFOs and CEOs do not buy the same companies at the same frequency, and a role can appear concentrated in a particular market regime or type of issuer.

To reduce that problem, we isolated 51 tickers with at least two mature CEO events and two mature CFO events. For each ticker, we calculated the median 180-day outcome for each role and then measured CFO median − CEO median.

Matched-issuer diagnostic	Result
Matched tickers	51
Median CFO-minus-CEO spread	-1.61 percentage points
Tickers where CFO median was higher	39.2%
Issuer bootstrap 95% interval	-3.71 to +2.68 pp

Because the bootstrap interval crosses zero, this release does not establish a stable difference between the two roles. The matched comparison also runs against the claim that CFO purchases should always receive more model weight.

What the result changes

A role field remains useful. It can help distinguish an operating executive from a director or a ten-percent owner, support governance research, and become informative in an interaction with other features. What fails is the one-dimensional rule:

CFO purchase = stronger signal than CEO purchase

A more defensible model would test role jointly with:

• the purchase as a share of the insider's reported post-transaction holdings;
• whether multiple owners bought within the same documented window;
• issuer sector and industry;
• whether the transaction is direct or indirect ownership;
• filing delay and the information actually available at a hypothetical entry time;
• size, liquidity and benchmark controls supplied by an external market dataset.

Those interactions must be validated out of sample. They should not be inferred from job-title narratives.

What this study does not claim

• It does not measure abnormal return or alpha.
• It does not establish that the reported purchase caused the later price change.
• It does not reconstruct an executable strategy from the EDGAR acceptance timestamp.
• It does not adjust for market capitalization, sector exposure or survivorship in the metadata-qualified universe.
• It does not treat the reporting owner's role as a proxy for personal wealth or purchase conviction.

The return-field definitions and missingness rules are documented in the research methodology. Reproducible work should retain those limitations alongside every result table.

Conclusion

The updated evidence is deliberately less dramatic than the conventional story: CEO and CFO purchases look broadly similar, and the matched-issuer test does not reveal a universal CFO advantage.

That is still useful information. It prevents a model from assigning confidence to a feature that has not earned it.

Researchers who want to test role interactions can inspect the 64-field schema in the free dataset sample or query targeted Form 4 records through the InsiderAlpha API.