WAVEMAN
Corrections
> Ask across 15+ datasets ⌘K
LIVE
CORRECTIONS
We publish every error that changed a number you could see, what caused it and what now guards against it.

Corrections

26 September 2026The insider-buying backtest used unadjusted prices
The study read raw closing prices, so every stock split looked like a crash or a jump and dividends were missing. Re-run on split- and dividend-adjusted prices, all open-market purchases against the Russell 2000 over 90 days moved from -0.62% to -0.11% (median -1.76%, 45.4% of purchases beat the index). The page also called the old figure statistically significant; that sentence is now computed from the numbers, and at t = -0.8 it is not. The conclusion did not change: bought blindly, insider purchases are not an edge.
25 September 2026Amended Form 4 filings were counted twice
When an insider files a Form 4/A it restates the whole transaction. We stored both the original and the amendment, so 12,401 purchases and sales were counted twice. The largest visible case: Jefferies (JEF) showed $1.3B of insider buying in 180 days instead of $677M. Restated originals are now removed every day before any figure is built, and the 20 largest figures on the site are re-checked each morning for this pattern.