Do insider signals actually work?

Every signal we compute is tested against what the stock did next, across every company and every month we cover. The whole registry is published here, including the signals that do not work and the ones with too little data to judge. Nothing is cherry-picked.

The big number is the hit rate: how often the trade beat the market over that horizon (50% is a coin flip). Below it, the median market-adjusted return. Cells with too few events or too short a history are marked and greyed out.

AMF (FR)

Signal	Events	21 days	63 days	126 days	252 days
High-conviction buy (score 70+)	892	54%1%	53%2%	43%-3%	37%-11%
All open-market buys (baseline)	14,593	50%0%	49%0%	45%-2%	43%-5%
Cluster buy	5,088	53%0%	55%2%	49%-1%	44%-3%
First buy after 12 quiet months	2,900	53%0%	51%0%	47%-2%	43%-5%
Unusually large buy	1,071	53%0%	50%0%	46%-2%	45%-4%
Cluster sell	1,881	43%-1%	49%-1%	52%1%	50%0%
Routine buy (control)	759	54%1%	59%2%	56%2%	52%1%

SEC (US)

Signal	Events	21 days	63 days	126 days	252 days
High-conviction buy (score 70+)	20,152	55%1%	51%0%	47%-2%	44%-7%
All open-market buys (baseline)	447,212	53%1%	49%-1%	45%-3%	40%-10%
Cluster buy	230,960	55%1%	50%0%

Snapshot of 2026-07-01, methodology v1.1-2026-07. Market-adjusted returns (BHAR) measured from the filing date; the mean is winsorised so vendor price artifacts cannot distort it.