The brief asks for a direct comparison on three metrics. The honest scorecard, given the missing database extract, looks like this:
| Metric |
CEO buys |
CFO buys |
Provisional read |
| Hit-rate at T+90 |
n/a |
n/a |
Cannot be estimated from the provided material |
| Mean excess return at T+90 |
n/a |
n/a |
Cannot be estimated from the provided material |
| Sharpe at T+90 |
n/a |
n/a |
Cannot be estimated from the provided material |
That may look unsatisfying. It is also the difference between research and decorative certainty.
Still, the evidence and market logic support a testable prior:
- Base case: both CEO and CFO open-market buys should be positive on average.
- Conditional edge: CFO buys are more likely to outperform when the market is misreading financial quality, liquidity, or the durability of weak reported numbers.
- Attention edge: CEO buys may react faster and more visibly, particularly in larger or more followed names.
- Best practical signal: clustered buying, meaningful trade size, and post-drawdown context probably matter more than role alone.
A likely ranking, with proper caveats
If forced to state a hypothesis before seeing the backtest, I would phrase it this way:
- Hit-rate: slight edge to CFO buys.
- Mean excess return: moderate edge to CFO buys, but with fatter tails and more sample sensitivity.
- Sharpe: ambiguous, depends heavily on sample size, concentration, and whether CFO signals are sparse but cleaner.
That is not a forecast dressed as a fact. It is the most defensible prior given the institutional roles and the literature on informed insider purchases.
The article-specific test should also include interaction terms
If this were being turned into a production note rather than a magazine article, I would not stop at CEO versus CFO. I would test:
- CFO buy × negative prior 6-month return,
- CFO buy × small-cap bucket,
- CFO buy × clustered insider purchases,
- CEO buy × founder-led indicator,
- CEO buy × high media-attention names.
That is where the useful signal probably lives. Simple role dummies are tidy for headlines and mediocre for portfolios.
A blueprint for the next pass on the 162k-filings database
The missing data is irritating, but not fatal. It simply means the next step is obvious.
Minimum spec for the rerun
Run the role-split backtest on the 162k-filings universe with the following constraints:
- Universe: only open-market purchases.
- Role map: CEO, CFO, and optionally “other executive” as a control bucket.
- Timing: entry at next tradable close after public dissemination.
- Horizon: T+90, plus T+20 and T+180 as sensitivity checks.
- Benchmark: sector- and size-adjusted excess return.
- Portfolio: equal-weighted calendar-time portfolios, issuer cap at 5 percent.
- Filters: exclude events within blackout ambiguities where disclosure timing is unclear.
- Robustness: report medians, winsorised means, and bootstrap confidence intervals.
The outputs that would actually settle the argument
To make the article’s thesis stand or fall, I would want these tables:
- Event counts by role and market.
- Average trade size and percentage increase in holdings by role.
- T+90 hit-rate, mean excess return, median excess return, volatility, Sharpe.
- Results split by market cap tercile.
- Results split by prior 6-month return bucket.
- Results for solo versus clustered buys.
- A regression table showing whether the CFO indicator survives controls.
Then, and only then, can we stop speaking in priors and start speaking in evidence.
If the CFO bucket wins on raw return but loses after controls, the story becomes “CFOs buy in better setups”. If it still wins after controls, then “higher information density” graduates from plausible narrative to empirical result.
The distinction matters. Finance has enough stories already.
The payoff is simple. If you are screening insider buys today, do not rank a CFO purchase above a CEO purchase merely because the title sounds more numerate. Rank both above most forms of executive optimism, then demand confirmation from trade type, size, clustering, and context. The concrete next step is to rerun the 162k-filings backtest with a strict role map and dissemination-time entry rules. The open question is not whether insiders know more than the market, they plainly do at times. It is whether the CFO’s extra information survives contact with sample design, benchmark choice, and the market’s habit of noticing the chief executive first.
