Last updated: August 8, 2026

Methodology

Every number on this site should be reproducible from public data. This page documents the definitions and the arithmetic so you can check our work — or disagree with it on the merits.

Where the data comes from

What	Source
Insider transactions (Form 3/4/5)	SEC EDGAR full-text filings
Prices, charts, volume, index levels	Yahoo Finance
Company news headlines	Yahoo Finance
Social sentiment	StockTwits public message streams

All of it is public. None of it is proprietary, and none of it comes from a private or paid feed. We do not have information that you could not obtain yourself with enough time.

What counts as an insider purchase

We keep only open-market purchases — Form 4 transaction code P. That is an insider buying shares with their own money.

We deliberately exclude the transaction types that look like buying but aren't a decision to invest:

• Code A — grants and awards. The company gave them the shares.
• Code M — option exercises. Converting existing compensation.
• Code F/S — shares withheld for taxes, and sales.
• Code G — gifts.

This is the single most important filter on the site. A screener that counts grants as "insider buying" will show you a flood of meaningless activity.

Each purchase is attributed to a distinct reporting owner as named in the filing, so one person filing several times in a week counts once toward the buyer count.

What counts as a cluster buy

A cluster buy is two or more distinct insiders making open-market purchases in the same company within a rolling 90-day window.

• Distinct buyers, not transactions. Ten purchases by one executive is not a cluster. Two purchases by two different people is.
• 90 days, rolling. The window moves with the calendar, so a cluster stops being current once the buying stops.
• Two is the floor, not the target. The screener ranks by buyer count first and total dollars second, because a five-insider cluster is a materially stronger signal than a two-insider one. You can raise the threshold yourself on the screener.

Our guides use "three or more" as the textbook definition, which reflects the convention in the research literature. The site's threshold is deliberately set lower so that forming clusters are visible early — the ranking, not the cutoff, is what separates strong from weak.

Cluster events and formation dates

A live screener answers "what's happening now." To measure a track record we need a fixed, non-moving record of the past, so each qualifying cluster is also written down once as a cluster event:

• The event's formation date is the date of the last qualifying purchase — the moment the cluster became visible to the public.
• Events are keyed by symbol and calendar month, so a company with continuous insider buying produces one event per month rather than a new one every day.
• Once written, an event is immutable. We do not retroactively delete clusters that went on to perform badly, and we do not backfill ones we wish we'd caught. That is the entire point of recording them.

How we measure outcomes

For each cluster event we compute a simple, unweighted return comparison:

1. Baseline — the first closing price at or after the formation date, for both the stock and the S&P 500 (^GSPC).
2. Current — the most recent available close for both.
3. Return = (current − baseline) ÷ baseline, for each.
4. Excess return = the stock's return minus the S&P 500's return over the identical window.

A positive excess return means the stock beat the index since the cluster formed. That's it — there is no risk adjustment, no beta, no sector-relative comparison, and no attempt to model what a portfolio would actually have earned.

Be careful about what this does and does not measure. It is not a backtest and it is not a strategy return. There is no position sizing, no entry or exit rule, no transaction costs, no dividends, and no survivorship correction. It answers one narrow question: since this cluster became public, has the stock outperformed the index?

How often the data refreshes

• Filings are pulled from EDGAR and clusters recomputed on a daily automated run. New Form 4s appear once EDGAR has published them, so there is an inherent lag of up to a couple of days from the trade itself.
• Prices, quotes and charts are fetched live and cached briefly.
• Cluster events and outcome scoring update on the same daily run.
• Daily market briefs are generated once per completed trading session. Weekends and market holidays produce no brief.

Which pages we let search engines index

A data site can generate near-infinite thin pages. We try not to.

• Company pages are indexable only once a ticker has enough insider-purchase history to say something. Sparse tickers stay out of the sitemap.
• Weekly leaderboards and reports are published only for weeks that actually had qualifying cluster activity.
• Daily market briefs carry noindex. They are useful to a regular reader and useless as search results.

Known limitations

We would rather state these plainly than have you discover them.

• Insiders are early. Cluster buys frequently precede months of further decline before any turn. Nothing here is a timing signal.
• Headcount isn't conviction. Five directors each buying $5,000 to satisfy an ownership guideline is much weaker than one executive committing $2 million. Read the dollar column, not just the buyer count.
• 10b5-1 and routine purchases dilute the signal. Pre-arranged plan purchases are less informative than discretionary ones, and Form 4 does not always make the distinction easy to parse.
• Parsing is imperfect. Form 4 is filed as XML by thousands of different filing agents, with inconsistent naming of both people and companies. We normalise what we can, and errors get through.
• Upstream data can be wrong or revised. Filings get amended. Price feeds have bad ticks. We inherit both.
• Small samples. Outcome statistics over a handful of clusters, or over a short window, are noise. Treat a track record with few events as unproven.
• Not risk-adjusted. A cluster in a volatile small-cap that beat the index is not evidence of skill — it may just be evidence of beta.

Corrections

If a filing is misparsed, an owner misattributed, or a number wrong, please report it with the ticker and, where possible, the EDGAR accession number. Confirmed errors are corrected in the underlying data, which flows through to every page.