Insider Transaction Signals

Legal insider trading — a company's officers, directors, and large shareholders buying or selling their own company's stock, publicly reported (in the US, via SEC Form 4) within days of the trade — has been studied since the 1970s as a signal, and the finding holds up better on the buy side than the sell side. An executive selling stock can mean almost anything: diversifying a concentrated position, funding a home purchase, covering a tax bill on vested options. An executive buying stock on the open market, with their own cash, has a much narrower set of plausible motives, and "I think this is undervalued" is by far the most common one.

Why buys are more informative than sells

Lakonishok and Lee's landmark study separated insider purchases from insider sales and tracked subsequent stock performance for each. Insider purchase clusters (multiple insiders buying in the same window) predicted meaningfully higher forward returns; insider sales, especially routine, small, or plan-based sales, predicted little to nothing, and even large discretionary sales were weaker predictors than the equivalent-sized buy signal was in the opposite direction. This asymmetry is the whole design principle of a tradeable insider-signal strategy: weight buys much more heavily than sells, and especially weight cluster buying, where several insiders act independently within a short window.

Worked example. A mid-cap company's CEO, CFO, and two independent directors each buy shares on the open market within a two-week span, a total purchase of $3.2m against a company with a $1.5 billion market cap — small relative to the company, but notable because four separate insiders acted independently rather than one person making a single large trade. Studies of cluster buying like this find average abnormal returns in the following six months on the order of 4–7%, meaningfully higher than the return following an isolated single-insider purchase of similar dollar size. A strategy flagging this cluster would go long, sized modestly given the signal is one input among several, and hold through the typical 3–6 month window over which the historical edge has concentrated.

What this means in practice, and what erodes it

Insider data is public and cheap (Form 4 filings are free from EDGAR, and vendors normalize them within a day), so the raw signal is heavily used and its easiest form is largely arbitraged into standard quant models already. What remains valuable is filtering out noise: routine 10b5-1 plan sales (scheduled in advance, uninformative), option-exercise-and-hold transactions (not open-market purchases with cash), and small transactions relative to the insider's net worth. Funds that build cleaner, filtered insider datasets — distinguishing discretionary open-market buys from everything else — retain more edge than funds using raw filing counts.