Training

Reading insider buying

Insider filings are one of the few datasets where decades of academic work agree on the broad shape: purchases have historically carried information; sales mostly haven’t. This page is a field guide to which patterns the published research actually supports — and a warning label, because none of it makes any individual filing predictive. Nothing here is investment advice; it is a reading list applied to a dataset.

What the research supports

• Purchases over sales. The asymmetry is the oldest result in the field (Jaffe 1974; Seyhun 1986; Lakonishok & Lee 2001). Insiders sell for a hundred reasons — taxes, houses, diversification — and buy for approximately one. Jeng, Metrick and Zeckhauser (2003) found insider purchases earned abnormal returns while sales showed essentially none.
• Clusters over solo trades. Several distinct insiders buying inside a short window has more consistent support than any single-filer pattern — one insider buying is a datapoint, three is a pattern. That is exactly what the cluster-buys page computes, with the window and minimum-buyer knobs exposed.
• Opportunistic over routine. Cohen, Malloy and Pomorski (2012) split insiders into routine traders (same month, every year) and opportunistic ones; the signal concentrated almost entirely in the opportunistic group. A purchase that breaks an insider’s own pattern is worth more than one that continues it.
• Size relative to stake. A $200,000 buy by an insider whose position was $400,000 is a different statement than the same dollars atop a $50 million position. The Δ-stake column in every feed here does this division, from the filing’s own owned-after figure.
• Smaller and less-covered companies. Most studies find the effect stronger where fewer analysts look — which is also where data quality and liquidity are worst. The two facts travel together.

What to discount

• Nearly everything that isn’t a purchase. Grants, vestings, option exercises and tax withholdings dominate filing volume and are compensation mechanics. The transaction codes separate them in one letter.
• Scheduled sales. A sale under a 10b5-1 plan was decided months before it executed.
• Headline dollar amounts without context. Big numbers from 10% owners are often funds rebalancing; the role filter on the screener exists for this reason.

The honest caveats

The abnormal returns in these studies are averages over thousands of trades and long horizons, before costs, and mostly from decades past; effects documented in academic papers have a way of shrinking once everyone can screen for them — and everyone can: the two-business-day deadline made this data effectively real-time for all. Insiders are also frequently early, sometimes by years, and sometimes simply wrong. Treat insider buying as one input that earns a closer look, never as a conclusion.