Every trade an executive makes in their own company's stock is public. It lands in an SEC filing called a Form 4 within two business days — who traded, what role they hold, how many shares, at what price. It's one of the most honest datasets in markets, because it's people betting their own money. And most of it is completely useless. The skill isn't watching insider activity — it's filtering it. Three filters do almost all of the work.
Filter 1: buys, not sells
The old market adage holds up: insiders sell for many reasons — diversification, taxes, a divorce, a house, options about to expire — but they buy on the open market for one. Selling is what a rational person does with concentrated wealth; buying more of the thing your salary, bonus, and career already depend on is a deliberate act of conviction. So the informative direction is purchases, and specifically open-market purchases — the Form 4 transaction code "P" — not option exercises, not grants, not conversions, which are compensation mechanics wearing a trade's clothing.
Filter 2: decisions, not schedules
Many insider trades are executed under a 10b5-1 plan: a program the insider committed to months earlier, trading on a fixed schedule regardless of what's happening at the company. These exist precisely so the trade cannot reflect current inside knowledge — which is exactly why they tell you nothing. A purchase only carries information if someone decided to make it recently. Form 4s flag plan trades; any honest screen throws them out before counting anything.
Filter 3: clusters, not lone wolves
One executive buying is a data point with a dozen innocent explanations — a new CFO expected to show skin in the game, a director topping up after joining the board, a personal portfolio decision. The pattern that research on insider trading has repeatedly found most predictive is the cluster: several different insiders buying within the same short window. People with independent views of the same business, separately deciding the same fortnight is the time to buy, is rarely a coincidence — each of them sees the gap between what they know and what the market believes.
A workable definition, and the one AnalystBook uses: three or more distinct insiders making open-market purchases within a 14-day window — plan trades excluded, multiple buys by the same person counted once. Strict on purpose: most companies never trip it, which is what makes it worth noticing when one does.
What a cluster does — and doesn't — tell you
A cluster is a prompt, not a verdict. Insiders know the company, not the market: they're often early, sometimes flat wrong, and they can't see the macro picture any better than you can. What a cluster earns is your next hour of research, pointed at three questions: What changed at the company in the last quarter? What's in the most recent filing — especially what changed in it? And is there a visible catalyst — new guidance, a strategic review, a product approval — that insiders might be positioning around? Buying into weakness after an ugly quarter is the classic tell; it says the people closest to the trouble think the market overreacted.
How to run this without reading Form 4s all day
The mechanics are tedious by hand: pull every Form 4, keep the code-P open-market buys, drop the 10b5-1 plan trades, group by person, slide a two-week window across the calendar. AnalystBook computes exactly this from the filings for every company we cover — when a cluster exists, it surfaces as a signal on the company page with the insider count, shares, and window, and the full transaction record sits one click away so you can see who bought, at what title, and when. No cluster, no card — the honest state for most companies most of the time is nothing to see here, and the product says so rather than manufacturing excitement. When you're researching a company end-to-end, the workflow guide shows where the insider record fits in the ten-minute read.
Insider data is the rare corner of markets where the disclosure is fast, complete, and about real money. Filtered down to deliberate, recent, collective buying, it's one of the best research prompts that exists. Unfiltered, it's noise with a good reputation. For research purposes only; not investment advice.