Insider Signal Scorecard — September 2026
Don’t outsource your judgment · monthly scorecard
Did our insider signals beat the market? Every month we grade our own published pipeline in public — every bucket, winners and losers, with the statistics next to the returns, and against two benchmarks. This is the September 2026 report.
The honest headline
The unit is a company-event (one ticker + a 10-day window), so a cluster of insiders counts once — not five times. Entry is taken ~2 trading days after the transaction (when the Form 4 is public). On that basis, across 963 matured company-events, broad opportunistic insider buying returned +3.0% vs the S&P and -2.2% vs small-caps (IWM), beating the S&P just 42% of the time. In plain English: “insiders are buying” is not, by itself, an edge — and our own conviction score isn’t beating its control yet either. We publish that rather than hide it.
The pocket that leans best on both benchmarks: a meaningful stake increase bought at or below the insider’s own price — +8.7% vs S&P, +6.3% vs small-caps across 110 company-events — but it only beat the market 38% of the time, so that average leans on a few big winners. A hypothesis worth forward-testing, not a proven edge.
The contrarian setup — currently our strongest result
A different signal type: buying a ≥12% drop when insiders sold only via automated plans (a possible overreaction). Historically that returned +9.9% vs the S&P (t 3.18) and +6.3% vs small-caps, beating the S&P 57% of the time across 113 events. Caveats: modest sample, and it has a survivorship bias (names that dropped then delisted can’t complete the window, so this reads high). Promising hypothesis, forward-tracked — not proof.
Every bucket, no cherry-picking
| Signal bucket | Company-events | Raw +63d | ★ vs S&P | vs small-caps | Hit |
|---|---|---|---|---|---|
| All opportunistic buys (the baseline) | 963 | +8.3% | +3.0% | -2.2% | 54% |
| ★ Big stake increase + below insider cost | 110 | +14.0% | +8.7% | +6.3% | 55% |
| New position + below insider cost | 48 | +15.9% | +10.4% | +7.8% | 46% |
| High conviction (score 70+) | 64 | +7.8% | +2.2% | -2.5% | 50% |
| Low conviction (<45) — internal control | 330 | +7.3% | +2.4% | -1.5% | 55% |
| Brand-new position | 100 | +9.4% | +4.2% | +0.0% | 45% |
| Grew stake 30–99% | 97 | +18.2% | +12.6% | +10.0% | 60% |
| CFO / COO buys | 96 | +10.3% | +5.3% | -0.9% | 64% |
| Cluster buys (≥3 insiders) | 111 | +7.8% | +2.7% | -0.3% | 54% |
| $100K+ buys | 716 | +9.1% | +3.9% | -1.6% | 52% |
| Offensive (quiet accumulation) | 169 | +1.1% | -3.9% | -7.1% | 46% |
A signal is only real if it beats both the S&P and small-caps — beating the S&P alone usually just means the names were small and cheap, which you can buy in an index fund. Raw returns are greyed (mostly market beta). Rows under 25 company-events are faded: too few to mean anything. The “low conviction” row is our own control — if it ever beats our high-conviction picks on a real sample, our scoring isn’t working, and this report will say so.
Caveat: we test ~two dozen buckets. The best one can look good by chance (multiple comparisons), so the ★ combination is our strongest hypothesis, not proof — we track it forward, out-of-sample, in the public ledger. Forward results are the real test.
Check our work
The signal ledger lists every high-conviction signal ever logged — timestamped before the outcome, graded as it matures, never deleted. The live feed and full methodology (Cohen–Malloy–Pomorski, Journal of Finance) are on the Insider Signals page.
All figures are hypothetical event-study returns computed from public SEC filings and market prices — no trading costs, taxes, or slippage. Past performance, real or hypothetical, does not predict future results. Educational, not investment advice. See Disclosures.