How We Measure Whether Insider Cluster Buys Beat the S&P 500

A signal is only worth anything if you can check whether it works. Plenty of sites will tell you insiders are buying; far fewer will tell you, honestly, how those buys turned out. This page explains exactly how we measure the outcome of every insider cluster buy against the S&P 500 — the definitions, the math, and the limitations — so you can judge the evidence for yourself.

The principle: an immutable, dated starting line

The core integrity problem in any "insider buying works" claim is hindsight. It is easy to look at winners after the fact and forget the losers. We avoid that by fixing a starting line the moment a cluster forms and never moving it.

When several distinct insiders have bought the same stock on the open market within a short window, we record a cluster event with:

• the symbol,
• the formation date (formed_on) — the date the cluster qualified,
• the number of distinct buyers at formation,
• the total dollar value of the purchases.

That snapshot is immutable. We do not retroactively add buyers, shift the date, or delete clusters that went on to lose. Each cluster gets its own permanent page — for example, /cluster/[symbol]/[period] — so the record is fixed and auditable. The starting line is set on day one and never edited to flatter the result.

The benchmark: return since formation, vs. the S&P 500

Once a cluster has a fixed formation date, measuring it is straightforward. For each cluster we compute:

• Stock return — the percentage change in the stock's price from the formation date to "as of" today.
• Benchmark return — the percentage change in the S&P 500 over the exact same window.
• Excess return — the stock's return minus the S&P 500's return. This is the number that actually matters.

Excess return is the honest scorecard. A cluster stock that rose 8% while the S&P 500 rose 10% did not beat the market — it lagged by 2 points, even though the raw return looks positive. By always comparing to the index over the identical period, we strip out "a rising tide lifted everything" and isolate whether the insider signal added anything.

The prices and the S&P 500 series are pulled from market data at render time, so the figures you see are current rather than frozen at some flattering past date.

Why the S&P 500 is the right yardstick

The S&P 500 is the default benchmark for U.S. equity performance: broad, liquid, and the index most investors are implicitly choosing against when they pick an individual stock. The relevant question for any stock pick is never "did it go up?" but "did it beat what I could have earned by doing nothing and buying the index?" Measuring excess return against the S&P 500 answers that directly.

(For one stock you might argue a sector benchmark is fairer. Across a large, diversified set of clusters spanning many sectors, the broad index is the cleaner, less arbitrary comparison — and it is the one investors care about most.)

What we deliberately do not do

Transparency means being clear about the guardrails:

• No survivorship editing. Clusters that underperformed stay in the record. We do not quietly drop the losers.
• No moving the goalposts. The formation date is fixed at formation. We never re-baseline to a lower entry price after the fact.
• No leverage, options, or timing overlays in the core measure. It is a simple buy-and-hold-from-formation comparison, which is the most honest test of the raw signal.
• No claim that every cluster wins. Insiders are sometimes early and sometimes wrong. The honest claim is about the tendency of clustered open-market buying, measured one event at a time.

Honest limitations

No methodology is perfect, and pretending otherwise would defeat the purpose of a trust page:

• Survivorship in the source data. If a company is delisted or acquired, its price history can become messy; such cases are inherently harder to score cleanly.
• Dividends. A pure price-return comparison does not capture dividends on either side; for most growth-oriented cluster names the effect is small, but it exists.
• Window sensitivity. "Return since formation" depends on how much time has elapsed. A cluster that formed last week and one that formed two years ago are not directly comparable; always read the elapsed window alongside the number.
• Past behavior is not a promise. A historical tendency to outperform is evidence, not a guarantee. Size positions accordingly.

See it for yourself

The whole point of this approach is that you do not have to take our word for it. Every cluster's outcome is shown on its own page, scored against the S&P 500 over the same window.

Browse current clusters on the cluster-buy screener, see the largest fresh formations on the biggest cluster buys this week, and open any individual cluster page to read its formation snapshot and its return-vs-S&P-500 outcome. For the broader picture of which insiders' buys have actually paid off, see the insider track records.

Want to follow new clusters and their outcomes as they form? Start with the live screener.

Frequently asked questions

How do you define a cluster buy? Three or more distinct corporate insiders each buying the same stock on the open market (Form 4 code P) within a short window. See insider cluster buys, explained for the full definition.

What benchmark do you compare against? The S&P 500, measured over the exact same window as the stock — from the cluster's formation date to today. We report excess return (stock return minus index return), not just raw return.

Do you remove clusters that lost money? No. Clusters are recorded immutably at formation and stay in the record regardless of outcome. Removing losers would make the data meaningless.

Where does the price data come from? Stock prices and the S&P 500 series are pulled from market data at render time, so each cluster's outcome reflects current prices rather than a frozen historical snapshot.

Does beating the S&P 500 historically mean a cluster will beat it in the future? No. A historical tendency is evidence, not a guarantee. Insiders are sometimes early and sometimes wrong — treat each cluster as a research prompt, not a recommendation.