Insider Clusters: When Form 4 Filings Actually Matter
Insider selling is noise by construction and a single buy is an opinion. The signal in Form 4 data lives in clusters — several insiders independently paying to disagree with the market price.
The common sophistication in reading insider filings is symmetry: insiders buying is bullish, insiders selling is bearish, weight both. The symmetry is the error. Buying and selling by insiders are not mirror images of one signal — they are two different kinds of event, generated by two different processes, and only one of them is priced in the economic sense of the word. Treating them symmetrically means averaging a genuine signal with structural noise, which is why so many investors conclude that insider data is useless: they have diluted the part that works with the part that cannot.
Start with the asymmetry, because everything else follows from it. An insider sale has many innocent explanations and needs none of them to be sinister: diversification out of a concentrated position, tax obligations, an exercise-and-sell on expiring options, a house, a divorce, a pre-scheduled trading plan executing mechanically. Compensation at most public companies arrives substantially in stock, which means insiders are structural sellers the way farmers are structural sellers of grain — the selling is the harvest, not the forecast. An open-market purchase has essentially one explanation: the insider believes the stock is worth meaningfully more than the price, strongly enough to add to a concentration that every adviser they employ is telling them to reduce. This is the framing we use: a buy is a purchased opinion; a sell is a liquidity event until proven otherwise. One side of the ledger costs the filer nothing but the position they were paid in. The other side costs them cash, diversification, and — because filings are public — reputation if they are wrong. Markets should, and in our reading of the evidence do, treat paid-for opinions differently from free ones.
Why the buy signal exists at all
It is worth being precise about why insider buying should carry information, because the mechanism determines the filters. It is not that executives possess the forecast — insiders are famously mediocre at predicting their own stock's near-term path. It is that they hold a different kind of information than the market does, and a buy reveals when that information diverges from the price.
The cluster test
A single buy, however sincere, is one person's opinion — and executives are as capable of value-trap romanticism about their own company as any outside holder, arguably more. The signal compounds along identifiable dimensions, and the compounding is the point: one insider buying is an opinion; a cluster is a consensus of the best-informed. The dimensions we weight:
Breadth with independence. Several insiders buying within a short window, at different times and sizes, are approximately independent draws from the same private information set — the disclosure-data equivalent of replication. Discount the cluster sharply if it looks coordinated: identical amounts on the same day reads as a morale gesture organized in a boardroom, not a set of independent judgments. Role. Operators outrank observers. A chief financial officer or a divisional president sees the order book; an outside director sees a quarterly deck. Buying by the people closest to the cash register carries the most operating information. Size against wealth. The unit that matters is not dollars but proportion — a purchase that is trivial against the insider's disclosed holdings and compensation certifies little, whatever its absolute size. Structure. Open-market purchases only. Option exercises, plan-driven transactions, and grants are compensation mechanics wearing a filing's clothes. Context. Buys against a falling price after bad news are worth more than buys into strength, because they express disagreement with a market that has just voted the other way — the maximum-information configuration is a broad, independent, operator-heavy cluster of meaningful open-market buys into weakness.
The illustration: clusters at the point of maximum disagreement
The pattern is old enough to state as history. Around the major dislocations of the modern era — the depths of the 2008-2009 financial crisis, the pandemic break of early 2020 — aggregate insider buying broadened dramatically, with clusters appearing across sectors at precisely the moments public commentary was most apocalyptic. Insiders were not forecasting the macro turn; almost nobody did. They were reporting something narrower and more reliable: that the prices being printed on their own companies implied operating outcomes far worse than what they could see from inside the building. That is the signal working exactly as the mechanism predicts — ground-level operating knowledge disagreeing, expensively and in public, with a panicked discount rate.
The same history teaches the signal's principal limitation: insiders were early, in some cases painfully so, and an investor who treated the first cluster as a bottom-tick indicator was often carried lower before being carried higher. The cluster certifies the divergence between price and operating reality; it says nothing about when the market will close that divergence, or how much wider it will get first. It is a valuation instrument, not a timing instrument, and it should be sized and horizoned accordingly — which for a portfolio owner means it pairs naturally with capital that can wait, and pairs badly with leverage that cannot.
The strongest case against this framework
The serious objection is reflexive, and it mirrors the decay argument that applies to every public filing: insiders know Form 4s are watched. Once buying becomes a known bullish signal, it becomes cheap public relations — a chief executive spending a visible but personally trivial sum to steady a falling stock is buying the signal, not the stock. On this view the filing stream is contaminated by performances of conviction, and the cluster test can be gamed by any management team willing to coordinate a show of confidence.
We accept the mechanism — signaling games are real — and note that the framework's filters are already the counter to it, because a performance and a conviction trade differ in exactly the dimensions the filters measure. Performative buying is small relative to wealth, tightly synchronized, concentrated in the most visible officers, and timed to press coverage; conviction buying is proportionate, staggered, spread across operators who gain nothing reputationally, and indifferent to attention. The gaming objection, followed to its end, does not defeat the framework; it explains why the framework has to be multidimensional rather than a headline count of buyers. A second objection deserves standing: even honest clusters have a value bias — insiders systematically buy their own stock too early in declines, and a cluster in a structurally challenged business may certify only that management underestimates the disruption they are inside of. This is why the cluster belongs at the certification layer of a research process, weighing on theses about businesses you have independently judged durable, rather than at the idea-generation layer where it would import that bias wholesale.
How to apply this framework
- Split the ledger before reading it. Route sales to the noise column by default — compensation structure makes insiders structural sellers — and promote a sale to signal only when it is anomalous in size, breadth, and timing against that insider's own history. Route open-market buys to the analysis column always.
- Score clusters on the five dimensions, not the headline count. Breadth with independence, operator-heavy roles, size relative to the filer's wealth, open-market structure, and direction against the tape. A cluster strong on all five is the rarest and most informative configuration in insider data.
- Read absence as evidence. When a holding's narrative is 'the market is irrational about this business' and the people who see the order book are declining to buy at these prices, the narrative is missing its most natural sponsor — treat that as data against the thesis, not as neutral.
- Horizon-match the signal. Clusters certify mispricing, not turning points, and history shows insiders run early. Pair the signal with patient capital and independent business-quality work; never with timing-sensitive or leveraged structures that cannot survive the divergence widening first.