The thesis
Officers, directors, and 10% owners of a US-listed company must report their open-market trades to the SEC within two business days on Form 4. The data is public, the latency is short, and decades of research have shown that aggregate insider buying predicts positive abnormal returns. The trick is filtering the signal from the noise: most insider activity is automated compensation or pre-planned 10b5-1 selling, not informed buying.
Academic basis
Lakonishok and Lee (2001) found that aggregate insider buying predicts positive returns and that cluster patterns — multiple insiders buying within a tight window — are particularly informative. Cohen, Malloy, and Pomorski (2012) showed that the predictive power is concentrated in “opportunistic” trades by insiders without consistent trading patterns, not in the routine trades. Both findings inform how Alpha Suite weights and filters Form 4 filings.
How Alpha Suite implements it
The scanner pulls every Form 4 from the SEC EDGAR feed, parses both purchases (P) and sales (S), and computes a net flow per ticker rather than treating buys and sells separately. Each transaction is then weighted along several axes:
- Dollar conviction — transactions are tiered by absolute dollar size, with a market-cap floor so a $100K buy in a $500M small-cap counts more than a $100K buy in a $500B mega-cap.
- Role weighting — CEO and CFO purchases get more weight than director or 10% owner purchases, consistent with the empirical finding that C-suite trades carry the most information.
- Cluster window — three or more distinct insiders buying within a 10-day window triggers a cluster bonus.
- Exponential time decay — a 35-day half-life so signals fade smoothly rather than vanishing on a hard cutoff.
- Net-flow penalty — concurrent insider selling discounts the buying signal, so a "mixed" filing pattern doesn't get scored as if only the buys exist.
The Form 4 score is then layered with a technical overlay (volume-confirmed breakout, momentum, SPY relative strength, earnings-date penalty) and a conviction-scaled boost. Final signals are ranked cross-sectionally by z-score, sized via fractional Kelly (f=0.25), and shipped with a take-profit, stop-loss, and time-stop derived from a volatility-anchored barrier model.
When it fires
The strongest signals come from clusters of three or more C-suite insiders buying significant dollar amounts in a small or mid-cap with no concurrent insider selling. The weakest "signals" are routine 10b5-1 sells by directors and small option-exercise-and-hold transactions, both of which the model down-weights or filters out.
What it does not catch: Form 4 only covers US-listed equities. Non-US insider equivalents (UK PDMR notifications, French AMF disclosures) are out of scope. The signal is also vulnerable to insider buying that follows public bad news, where the buying is more about supporting the share price than expressing private conviction.
See live signals on the dashboard
Recon (free) gives you read-only access to today's ranked signals across all 10 strategies. No card, no commitment — sign up to see the model output for yourself.
Open the Dashboard