Cluster buying: the one insider signal that actually works

OpenStocks team·2026-04-10·6 min read
insidersForm 4signals

Why most insider data is noise

Form 4 catches every executive purchase or sale — but the vast majority of what shows up isn't a discretionary trade:

  • Option exercises (txCode = M) followed by an automatic same-day sale to cover taxes.
  • 10b5-1 plans — pre-scheduled sales that have nothing to do with the executive's view today.
  • Gifts to charity or family trusts (txCode = G).
  • Compensation grants vesting on schedule.

If you naively rank tickers by "insider sales last 30 days," you'll mostly find companies whose insiders are subject to a vesting calendar. That's not signal; that's payroll.

What does work: open-market purchases

The only insider transaction that consistently predicts forward returns is the open-market purchase (txCode = P). Why?

  • The insider is paying with their own after-tax cash.
  • They have no good reason to do this unless they think the stock is undervalued.
  • They're triggering a 6-month short-swing profit rule (Section 16(b)) — so they're committing to hold.

When three or more named officers/directors buy in the open market within ~30 days, the academic literature (Cohen, Malloy & Pomorski, Lakonishok & Lee) finds an outperformance of roughly 6–10% over the following 12 months.

We surface this on OpenStocks under Signals as "cluster buys."

What to filter out

Even within open-market buys, three filters dramatically improve the quality of the signal:

1. Skip 10%-owners unless they've been on the cap table for years. Activist funds buying a 5% block to push for change is its own thing — not a fundamental signal. 2. Skip transactions under $50K — directors topping up their stake by a token amount around board meetings is friendly noise. 3. Skip programmatic 10b5-1 buy plans — these now appear in Form 4 with a (2) footnote since the 2023 SEC amendment.

How we use it

The OpenStocks /insiders page sorts by net dollar flow per ticker (buys minus sells) — but the cluster signal lives in /signals, where we require:

  • ≥ 3 distinct insiders,
  • All open-market (P),
  • Within a 30-day window,
  • Net positive dollar flow.

That's typically 5–15 tickers per month. Most of them won't go up. But the hit rate is high enough — and the win-loss skew favorable enough — that it's the one piece of insider data we'd actually trade on.