Insider buying and selling refers to transactions in a company's own stock by its officers, directors, and large (>10%) beneficial owners — the people legally defined as "insiders" under Section 16 of the Securities Exchange Act of 1934. Because these people are presumed to know more about the firm than outsiders, their open-market trades are watched as a sentiment and information signal. The core tension is asymmetry: buying is generally informative because there is essentially only one reason an insider spends personal cash on more shares (they expect the price to rise), whereas selling is noisy because insiders sell for many reasons unrelated to their view — diversification, taxes, divorce, a new house, or scheduled compensation. Insider activity is not the same thing as illegal insider trading; these are legal, disclosed trades, and the signal lives entirely in publicly filed forms.

How it's formed and disclosed

The data comes from mandatory SEC filings, not estimation:

• Form 3 — initial statement of ownership, filed when a person first becomes an insider.
• Form 4 — the workhorse. Reports each change in beneficial ownership (purchases, sales, option exercises, grants, gifts). Since the Sarbanes-Oxley Act of 2002, Form 4 must be filed within two business days of the transaction (previously by the 10th of the following month), which dramatically shortened the lag and made the signal more usable (SEC; NASPP).
• Form 5 — annual catch-up for small or exempt transactions not reported earlier.

Each Form 4 line carries a transaction code. The most important to distinguish are P (open-market purchase) and S (open-market sale) — the discretionary trades that carry signal — versus A (grant/award), M (option exercise), F (shares withheld to pay taxes), and G (gift), which are largely mechanical compensation events and carry little to no directional information. A Form 4 also now contains a checkbox indicating whether the trade was made under a Rule 10b5-1 plan — a pre-scheduled, formula-driven trading arrangement adopted while the insider had no material non-public information. 10b5-1 sales are the least informative kind, because their timing was set in advance.

How to read it

The structure of the activity matters far more than any single trade:

• Open-market purchases (code P) are the strongest single tell, especially when made with personal cash near or below the current market price.
• Cluster buying — multiple distinct insiders buying within a short window — is the most-cited bullish configuration; academic work finds the predictive content rises when buying is broad rather than a lone trade (Lakonishok & Lee).
• Size relative to wealth/holdings matters more than dollar size. A director adding 50% to a previously small stake is louder than a billionaire founder's routine top-up.
• Selling is read with heavy skepticism. Routine, scheduled, or 10b5-1 sales are usually treated as noise. Selling becomes interesting only when it is unusual — a long-time holder dumping a large fraction of a position, a cluster of insiders all selling at once, or sales clustered just ahead of bad news.

How it's used in practice

Most practitioners use insider data as a confirming overlay, not a standalone trigger. Common deployments:

• Conviction confirmation. A value or contrarian investor who already likes a thesis treats fresh cluster buying as evidence that management agrees the stock is cheap.
• Screening for "skin in the game." Quant and discretionary screens filter for opportunistic open-market purchases, cluster buys, or unusually large purchases as one factor among several.
• Catalyst-window awareness. Insiders are barred from trading on material non-public information and typically observe self-imposed blackout windows around earnings. A purchase after a blackout opens, or unusual selling before a known catalyst, draws scrutiny.
• Avoiding false alarms on selling. Sophisticated users strip out option-exercise-and-sell, tax-withholding, and 10b5-1 sales before drawing any conclusion, because raw "insider selling" dollar totals are dominated by these non-signal events.

Adoption, debate & evidence

The signal is widely tracked (OpenInsider, Form 4 aggregators, Bloomberg, and most institutional research desks) and is one of the better-documented anomalies in the academic literature — but with important boundaries:

• The buy-side signal is real but modest. The landmark study, Lakonishok & Lee (Review of Financial Studies 14(1), 2001, "Are Insider Trades Informative?"), examined NYSE/AMEX/Nasdaq firms over 1975–1995 and found that the predictive power comes almost entirely from purchases, not sales, that it is concentrated in smaller firms, and that the effect strengthens with cluster activity. The commonly cited figure is roughly a few percent of outperformance over the following 12 months for the heaviest-buying stocks — meaningful but not dramatic, and partly entangled with the size and value factors.
• Most insider trades are uninformative. Cohen, Malloy & Pomorski (Journal of Finance, 2012, "Decoding Inside Information") showed that "routine" insiders who trade on a predictable calendar carry essentially zero predictive power; stripping them out leaves "opportunistic" traders whose trades earned value-weighted abnormal returns of about 82 basis points per month (~10% annualized) and predicted future firm news. This is the key refinement: which insider and why matters more than the headline buy/sell count.
• Selling is largely noise. The same literature repeatedly finds insider selling has little to no reliable predictive value, consistent with its many benign motivations.
• Regulatory drift has weakened parts of the signal. The SEC's December 2022 amendments to Rule 10b5-1 imposed a mandatory cooling-off period (generally 90 days for officers/directors, 30 days for others) before plan trades can begin, after research showed insiders trading shortly after adopting plans earned abnormal returns — i.e., gaming the safe harbor. Tighter rules reduce that particular edge.

Strengths & limitations

Strengths: It is hard data from a legally compelled disclosure, available free and quickly (two-day filing), and the buy-side signal survives decades of out-of-sample scrutiny. It works best in small- and mid-caps, on open-market cluster purchases by opportunistic insiders.

Limitations: The edge is small, slow (a multi-month horizon, not a day-trade signal), and decays as it gets crowded. The single most common misuse is reading insider selling as bearish — most selling is liquidity- or tax-driven and predicts nothing. A close second is failing to separate transaction codes: treating option exercises, grants, and 10b5-1 sales as discretionary signals. Insiders can also be early or simply wrong about their own stock, and a CEO buying on the way down ("catching a falling knife") is common.

Sources

• Lakonishok, J. & Lee, I., "Are Insider Trades Informative?" Review of Financial Studies — SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=253079
• Cohen, L., Malloy, C. & Pomorski, L., "Decoding Inside Information," Journal of Finance 67(3), 2012 — NBER: https://www.nber.org/papers/w16454 ; Wiley: https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1540-6261.2012.01740.x
• SEC, "Insider Transactions and Forms 3, 4, and 5": https://www.sec.gov/files/forms-3-4-5.pdf
• NASPP, "Section 16: The Basics of Forms 3, 4, and 5": https://www.naspp.com/blog/section-16-basics-of-forms-3-4-and-5
• SEC, Final Amendments to Rule 10b5-1 (Dec 2022) — Fact Sheet: https://www.sec.gov/files/33-11138-fact-sheet.pdf ; Gensler statement: https://www.sec.gov/newsroom/speeches-statements/gensler-insider-trading-20221214
• Perkins Coie, Public Company Handbook Ch. 6 (insider reporting & 10b5-1): https://perkinscoie.com/public-company-handbook-chapter-6-insider-reporting-obligations-and-insider-trading-restrictions

Dispute flags: the magnitude of the buy-side outperformance varies by study, sample period, and factor controls (commonly cited around a few percent/12 months per Lakonishok–Lee; ~82 bps/month for the opportunistic subset per Cohen–Malloy–Pomorski) — treat all such figures as period- and method-dependent, not guaranteed.