Quantitative funds do not read insider filings one company at a time. They ingest them. Every Form 4 filed on EDGAR is machine readable, and a systematic fund can parse the reporting person, transaction codes, prices, share counts, and remaining balances for thousands of insiders within minutes of filing. That data becomes one signal among many in models that rank stocks, size positions, and manage risk. The approach is less about spotting a single dramatic purchase and more about measuring whether insider behavior, in aggregate, tilts a probability.
The engineering work comes first. Filings arrive in structured form but with messy human variation: footnotes that explain plans and tax withholding, indirect holdings through trusts and funds, amended filings that correct earlier ones, and late filings that arrive after the two business day deadline. A usable system has to classify transactions correctly, separating open market purchases, code P, and open market sales, code S, from grants, option exercises, gifts, and code F tax withholding. Getting that classification wrong poisons everything downstream, so much of the effort goes into cleaning rather than cleverness.
Funds also normalize by size. A purchase of ten thousand shares means one thing at a low priced small company and another at a high priced large one, so models typically scale trades by the insider existing holding, by salary proxies, by company market value, and by usual trading volume. They track who is trading as well, because a chief executive or chief financial officer purchase carries different information from a small director grant, and a first purchase after years of only selling differs from a routine addition.
Insider behavior has been studied as a factor for decades, in academic papers and inside funds. The broad finding, stated carefully, is that open market purchases by insiders have, on average and over long samples, been followed by somewhat better than average returns, with the effect stronger in smaller companies and when several insiders buy together. Selling shows a much weaker and less consistent relationship with later returns, for reasons that match common sense: insiders sell for many reasons unrelated to outlook, including tax, diversification, retirement, and preset plans.
These are averages with wide variation, not rules for individual stocks. Effects found in older samples can shrink once they are widely known and traded on, transaction costs eat into small edges, and a signal that works across thousands of stocks can fail badly in any single case. Reputable funds treat insider data as a modest tilt, combined with valuation, earnings, price, and quality signals, and they test whether the tilt still earns its place after costs. Some conclude it does in limited form. Others drop it. Both conclusions come from the same public data.
The weighting follows from the motives. Buying shares in the open market with personal money is a voluntary commitment with a clear alternative: do nothing. Selling has many routine explanations that carry no view on the company. A model that treated a code F tax withholding line or a scheduled plan sale as equal and opposite to a discretionary purchase would systematically misread the data, so systematic users filter routine trades out and weight what remains by discretion and size.
Clusters weigh more for a statistical reason as well as an intuitive one. One insider can be wrong, idiosyncratic, or buying for personal reasons. Three officers and a director buying in the same few weeks, at a company where insider buying is normally rare, is harder to explain by chance or by individual circumstance. Funds therefore build features such as the number of distinct buyers in a window, the net buying across all insiders, and whether buying is unusual relative to that company own history. Rarity matters. A purchase at a firm where insiders buy every quarter carries less new information than the first purchase in five years.
Latency and discipline complete the picture. Filings must be made within two business days, but some arrive late, and a fund edge, if any, depends on processing the bulk quickly and trading without moving the price against itself, which limits the strategy mostly to liquid names or small position sizes. The result is a signal used with restraint, which is a fair summary of what the evidence supports.
This description of systematic practice is educational information only and is not investment advice or a promise about any strategy returns.