{
"generatedAt": "2026-05-14T20:47:47.482Z",
"method": {
"reference": "Cohen, Malloy, Pomorski (2012, JoF) — Decoding Inside Information",
"filterApplied": "Exclude Insider.tradingPattern == 'routine'",
"universe": "BUY · cluster≥2 · mid-cap 200M-1B€ · role∈{PDG,CFO,Dir} · pubDelay≤7d · signalScore≥40",
"returnHorizon": "T+90 days from pubDate+1 close (retail-view)"
},
"patternBreakdown": {
"opportunistic": 73,
"routine": 85,
"new": 15,
"unclassified": 0
},
"perPatternStats": {
"opportunistic": {
"n": 73,
"avgReturn90d": 13.706543262616549,
"winRate": 0.8082191780821918,
"sharpe": 0.7610317131643077,
"medianReturn90d": 10.20909607725264
},
"routine": {
"n": 85,
"avgReturn90d": 48.62052736260003,
"winRate": 0.8352941176470589,
"sharpe": 1.6590870981167811,
"medianReturn90d": 57.85627256917885
},
"new": {
"n": 15,
"avgReturn90d": -5.623884556283721,
"winRate": 0.26666666666666666,
"sharpe": -0.3478048012922834,
"medianReturn90d": -8.627452263690953
}
},
"withoutFilter": {
"n": 173,
"avgReturn90d": 29.184764252299146,
"winRate": 0.7745664739884393,
"sharpe": 0.9354978004622497,
"medianReturn90d": 25.54166581895616
},
"withFilter": {
"n": 88,
"avgReturn90d": 10.411583975304001,
"winRate": 0.7159090909090909,
"sharpe": 0.5438593480983861,
"medianReturn90d": 6.864032872501762
},
"delta": {
"signalCountDelta": -85,
"avgReturn90dDelta": -18.773180276995145,
"winRateDelta": -0.05865738307934831,
"sharpeDelta": -0.39163845236386363
},
"interpretation": {
"signalsDropped": 85,
"pctSignalsDropped": 0.4913294797687861,
"avgReturnUpliftPctPts": -18.773180276995145,
"winRateUpliftPctPts": -5.865738307934832,
"sharpeUplift": -0.39163845236386363
},
"note": [
"HONEST FINDING (AMF/FR universe ≠ US CMP-2012 universe):",
"On this sample, the 'routine' cohort actually outperforms 'opportunistic' on T+90,",
"likely because the AMF 3-day mandatory disclosure deadline forces predictable cadences",
"even for legitimate private-info trades, and the small-N FR universe is dominated by a",
"few high-conviction repeat buyers. The CMP filter as published thus REDUCES alpha on",
"this dataset rather than amplifying it.",
"",
"RECOMMENDATION: keep the classifier (it is a valuable feature for the score) but",
"consider re-tuning the patternBonus signs (currently -8 routine / 0 new / +12 opp.)",
"via grid-search on the live retail-return data before relying on the filter in",
"production. The `excludeRoutineInsiders` flag is shipped opt-in for safety review."
]
}InsidersTradesSigma