How We Turn a Cluster Buy Into a 0–100 Score

The first two posts in this series covered which insider buys carry information and where that information is loudest. This one is about ranking. Once we know a cluster is real, opportunistic, and sitting in the micro-cap sweet spot, we still have a problem: ten clusters can clear the same filters in a single week. Which one do you read first?

That is what the Cluster Signal Score is for. It is a number from 0 to 100 that collapses a messy pile of Form 4 filings into one figure you can sort on. It does not predict price. It ranks conviction.

What the Score Is, and What It Isn't

The score is a heuristic. We are honest about that. There is no regression that outputs "this cluster will return 4.2% next month," and anyone who tells you their insider screener does that is selling you a backtest, not a forecast.

What the score does is encode the findings from the academic literature into weights. Bigger clusters, more senior buyers, larger dollar commitments relative to company size, tighter timing, and a clean opportunistic signal all push the number up, because every one of those traits is associated with stronger predictive power in the research. The score is the literature, compressed into points.

The Six Components

A cluster earns points across six dimensions. The maximum from each is capped, and the total is capped at 100. Here is the full breakdown.

Component	Max Points	What it measures
Cluster size	25	How many distinct insiders bought
Dollar value vs. market cap	25	Total purchased as a share of company size
Role seniority	20	How senior the buyers are
Recency tightness	10	How close together the buys landed
Largest single trade	10	The biggest individual purchase
Signal quality	10	First-buy and opportunistic vs. routine

The two heaviest components, cluster size and dollar value, are worth 25 points each. That is deliberate. They map most directly to the two effects we trust most: internal agreement and skin in the game.

Cluster Size (max 25)

Each insider in the cluster is worth 10 points, capped at 25. So a two-person cluster earns 20, and any cluster of three or more maxes out the component.

This is the Alldredge and Blank (2019) effect, made into points. Their work found that purchases made within days of a colleague's purchase generate roughly double the abnormal return of a solo trade. One executive can be wrong, or buying for liquidity reasons we cannot see. When two or three independently put money in within the same window, the odds that they are all reacting to the same private read on the business go up sharply. The cap at three sits where the marginal information from a fourth or fifth buyer starts to flatten.

Dollar Value vs. Market Cap (max 25)

This is the component that makes the score size-aware. We take the total dollars purchased across the cluster, divide by the company's market cap, and scale that ratio. A cluster that buys an amount equal to a quarter of one percent of the company's market cap maxes the component out.

The ratio matters more than the raw dollar figure, and this is the entire reason it carries 25 points. A $1M buy at a $200 million micro-cap is a real commitment relative to the size of the business. The same $1M at a mega-cap is a rounding error. As covered in the previous post, the insider's informational edge lives in small, under-covered companies, so we score the purchase against the size of the company, not in absolute terms.

Role Seniority (max 20)

Not all insiders sit equally close to the numbers. We weight the roles in the cluster and cap the total at 20:

Role	Points
CEO	15
CFO	12
President	12
COO	10
10% Owner	8
Director	5
Other / unmapped	3

The CFO sits at 12, just below the CEO, on purpose. The finance chief sees the cash position, the receivables, and the forecast before almost anyone. A CFO opening her checkbook is one of the cleaner signals on this list. Directors are weighted lowest among named officers because they are a step removed from daily operations, and a board full of directors buying is weaker than a C-suite doing the same. A cluster of two or three senior officers will hit the 20-point cap easily, which is the point: past a certain level of seniority, more titles do not add much information.

Recency Tightness (max 10)

We measure the span in days between the first and last buy in the cluster and award 10 − (span × 2) points, floored at zero. Buys on the same day earn the full 10. A five-day span earns nothing.

Timing is information. The tighter the window, the more the purchases look like a coordinated response to the same event or the same private read, rather than a handful of unrelated decisions that happened to land in the same month. A cluster that forms over a single day is a louder signal than the same names spread across two weeks, even if the dollars are identical. This component is capped low, at 10, because tightness sharpens a signal but cannot manufacture one.

Largest Single Trade (max 10)

We take the single biggest purchase in the cluster and award one point per $50,000, capped at 10. A $500,000 individual buy maxes it out.

This is a conviction tiebreaker. Cluster size already rewards how many people bought. This component rewards how hard the most committed one of them leaned in. A cluster where the CEO alone puts in half a million dollars carries a different weight than one where five people each bought $20,000, even though both might look similar on a count. The single large check is often the person who knows the most acting on it.

Signal Quality (max 10)

This is where the filtering from the first post becomes a number. Each insider in the cluster is classified, and we score them:

Signal type	Points per insider
First buy	5
Opportunistic	3
Routine	0

The total is capped at 10. A first-time buyer is worth more than a known opportunistic trader, who is worth more than someone on a routine, calendar-driven pattern.

This traces straight back to Cohen, Malloy, and Pomorski (2012), who found that routine insider trades generate essentially zero excess returns while opportunistic ones drive nearly all of the alpha. A routine buyer adds nothing to the score, by design. A first buy, a purchase from someone with no prior history at all, is the strongest version of opportunistic: there is no pattern to regress to, so the trade can only be a fresh decision. That is why it earns the most.

Why the Caps Matter

Every component is capped, and the total is capped at 100. This is not cosmetic. Caps stop any single dimension from running away with the score. Without them, a billionaire 10% owner making one enormous buy could swamp everything else and produce a high score off a signal we trust less than a tight, multi-officer cluster. The caps force a high score to be broad: a 90 cannot come from one impressive number. It has to come from several components agreeing at once, which is exactly the kind of cluster the research says is worth your attention.

The Takeaway

The Cluster Signal Score is not a crystal ball. It is a ranking tool built on top of the same evidence the first two posts laid out: clusters beat solo trades, ratios beat raw dollars, senior officers beat the board, tight timing beats scattered, and opportunistic beats routine. Each of those findings is worth points, the points are capped so no one trait dominates, and the result is a single number you can sort a week's worth of filings by.

When you see an 88 at the top of the dashboard, it is not a prediction. It is the cluster where the most of those things lined up at once. That is the one to read first.

---

References:

• Cohen, L., Malloy, C. & Pomorski, L. (2012). "Decoding Inside Information." Journal of Finance, 67(3), 1009–1043.
• Alldredge, D. M. & Blank, B. (2019). "Do Insiders Cluster Trades With Colleagues?" Journal of Financial Research, 42(2), 331–360.
• Lakonishok, J. & Lee, I. (2001). "Are Insider Trades Informative?" Review of Financial Studies, 14(1), 79–111.