Playbook
How to find insider buying that actually predicts returns
“Insiders are buying” is one of the most over-quoted phrases in markets, and one of the least useful as stated. Most Form 4 purchases are noise. The edge that survives in the academic record is specific and narrow: opportunistic, open-market buys, by operating insiders, clustered together, in small-caps. Here is how to separate that signal from the artifacts — and how to ask an agent to do it for you.
The short version
- Insider buying predicts returns; insider selling mostly doesn’t (people sell for many reasons).
- The strong edge is narrower still: opportunistic (non-routine) buys, by operating insiders, in clusters, in small-caps.
- The biggest trap is mistaking primary issuance (IPO subscriptions, PIPEs, VC fund buy-ins) for open-market conviction.
- Validate empirically before acting: report the sample size, hit rate and event-window return — a single name is descriptive, not predictive.
What the research actually says
Insider trading data is one of the few signals with a deep, peer-reviewed literature behind it. Quote these as base rates, never as guarantees.
Buying beats selling. Lakonishok and Lee (2001, Review of Financial Studies) studied two decades of insider transactions and found that companies with heavy insider buying outperformed those with heavy selling by about 7.8% over the next year, and that the effect was concentrated in small-cap stocks. Crucially, the signal came from the buy side. Insiders sell for diversification, tax, divorce, a new house — a thousand reasons that have nothing to do with the stock being expensive. So selling is mostly noise.
Opportunistic beats routine. Cohen, Malloy and Pomorski (2012, Journal of Finance) added the sharpest refinement. They split insiders into routine traders — those who buy or sell at the same time every year — and opportunistic traders who break their own pattern. Almost the entire abnormal return, around 82 basis points per month, came from the opportunistic group. A CFO who suddenly buys after years of doing nothing is saying something. The one who buys every January is on autopilot.
Clusters beat singles. One insider buying can be cosmetic — a board member topping up to meet ownership guidelines, a routine grant exercise. Several distinct insiders buying on the open market within the same short window is harder to explain away. Independent decisions converging is the texture of real information.
The four filters that separate signal from noise
Put together, the literature implies a checklist. A raw “insiders are buying” headline that fails these is usually an artifact.
1. Open-market buys only (and beware primary issuance)
This is the single most common trap. Filings routinely dress up primary issuance as insider buying:
- IPO subscriptions — venture funds buying into the offering on day one.
- PIPE deals — private investment in public equity, often by the same funds.
- Reverse-merger and fund-subscription rows dressed up as Form 4 buys.
None of these is open-market conviction; they are participation in new share issuance. The tell: an S-1, IPO, or PIPE 8-K in the same window as the “buys”. Always cross-check the issuer’s recent filings before believing a cluster.
2. Operating insiders, not funds
Real signal comes from operating-company officers and directors — the CEO, CFO, founders, board members with skin in the game. A name like “X Capital LLC / GP / Fund / Advisors” buying its own vehicle is a different animal: it can be genuine, but it is a fund position, not an operator betting on the business they run. Weight the operator buys.
3. Weight small-caps
The Lakonishok-Lee edge lives in small-caps, where insiders genuinely know more than the market and analyst coverage is thin. A founder buying a $400M company off the radar is more informative than a director adding to a mega-cap that fifty analysts already cover.
4. Opportunistic, not calendar
Favor buys that break a pattern. A first purchase in years, a buy into weakness, a buy outside the usual grant cycle — these are the opportunistic trades that carry the documented edge.
The strongest read combines all four: opportunistic + operating insider + small-cap + no concurrent offering. That intersection is rare, which is the point — rare is where the signal is.
How to screen for it with an agent
Once your assistant is connected to a markets MCP server, you don’t run SQL — you describe the screen and let it call the tools. The workflow ClawTerminal’s tools support maps directly onto the filters above:
| Step | Tool | What it does |
|---|---|---|
| Find clusters | cluster_insider_buys | Tight groupings of independent open-market buys; floors on distinct insiders and total value drop noise; PIPEs excluded by default |
| Rank buyers | top_insider_buyers | Market-wide ranking of recent open-market buying |
| Add context | insider_features | Counts, aggregate value, CEO/CFO flags, price context for one name |
| Confirm no offering | list_filings_universal | Check for a concurrent S-1 / IPO / PIPE 8-K — discard if present |
| Validate the edge | get_insider_price_reaction | Historical post-buy returns for that issuer |
A plain-English version you can paste to a connected agent:
“Find insider cluster buys in the last two weeks: at least three distinct insiders, at least $500k total, open-market only. Drop any name with a concurrent IPO, S-1 or PIPE filing. For the survivors, tell me the insider roles, whether the buyers are operators or funds, the market cap, and the historical post-buy price reaction. Flag the small-caps.”
That single prompt encodes the four filters and the validation step. The agent resolves entities, runs the cluster screen, cross-checks filings, and reports the survivors with context.
Validate before you act
The cardinal rule of smart-money following: never assert the literature edge — confirm it, then report the numbers. For any candidate, you want the sample size, the hit rate, the mean and median event-window return, and a t-stat. A single issuer’s post-buy move is descriptive, not causal — n of one proves nothing. If the sample is tiny or the t-stat is insignificant, the honest conclusion is “unconfirmed,” and you stop there rather than forcing a story.
Insider buying is an origination signal, not a thesis. It tells you where to look, not what to conclude. The next step is a fundamental and valuation check on the name — see the factor screen for the quant overlay, and pair it with the 13F view to see whether institutions agree.
What not to conclude
- Don’t call an IPO, PIPE or VC subscription “insider conviction.” It is the number-one cluster artifact.
- Don’t read insider selling as a bearish signal — its predictive content is close to zero.
- Don’t treat a single name’s post-buy move as proof of an edge.
- Don’t trust valuation context on foreign issuers and ADRs — share-count quirks make their market caps unreliable, even though the insider signal itself is unaffected.
Insider buying done right is one of the cleanest event-driven signals available to a retail researcher, because the data is public, timely and legally disclosed. The discipline is in the filters. Apply them, validate the edge, and “insiders are buying” stops being a slogan and starts being a screen.