ClusterSignal

About

Open methodology. Public data. Rigorous filtering.

ClusterSignal is powered by SEC EDGAR Form 4 filings and enriched with StockFit sector data and Yahoo Finance market context. We document exactly how clusters are detected, scored, and filtered — so you can evaluate the signal quality yourself.

Why clusters, not just trades

Insiders sell for a hundred reasons that have nothing to do with the stock — paying for a house, taxes, diversifying. A single sale tells you almost nothing.

Buys are different. Insiders only buy because they think the price is going up. And when three or more execs at the same company write personal checks for shares within the same short window, that's not noise — that's a coordinated bet from the people who actually know what's happening inside the business.

ClusterSignal filters for that pattern specifically, instead of dumping every Form 4 filing into a table and leaving you to find the signal yourself.

Data sources

All insider transaction data is extracted from public SEC EDGAR Form 4 filings. These mandatory disclosures cover open-market purchases, sales, option exercises, and other transactions by company insiders (executives, directors, and 10%+ holders).

Sector and industry classification is sourced from StockFit. Price context — current price, 52-week range, market cap, P/E ratio, earnings dates, and news headlines — is fetched from Yahoo Finance and cached per ticker. We do not use any non-public, proprietary, or insider-only sources.

Detection methodology

A cluster is flagged when 3 or more distinct insiders file Form 4 open-market transactions (purchases or sales) for the same company within a rolling 5-day window.

Transaction floors eliminate noise before cluster detection runs. Trades under $5,000 total value are ignored for all insiders. VP and Director-level trades under $10,000 are excluded. CEO and CFO trades under $25,000 are excluded. The aggregate cluster value must exceed $50,000 before a cluster is created.

The poller scans SEC EDGAR every 30 minutes using the EFTS feed. New Form 4 filings are parsed for non-derivative transactions, stored, and immediately evaluated for cluster patterns. Existing clusters are not duplicated — each unique combination of ticker, direction, and overlapping window is stored once.

Scoring methodology

Each cluster receives a 0–100 composite score from five factors. Every factor is measured beyond the bar a cluster must clear to be detected at all (3+ insiders, $50K+, a 5-day window), so the score reflects how far a cluster stands out from its peers rather than the fact that it qualified:

  • Capital Deployed (up to 30 pts): Log-scaled and ranked against other clusters in the same direction. Insiders sell far more than they buy — the median sell cluster is $2.6M against $535K for buys — so buys and sells are scored on separate dollar ladders and a large buy is not penalised for being small next to a large sale.
  • Seniority Depth (up to 26 pts): CEO 30, CFO 27, President 23, VP 10, Director 7 — summed at a halving weight across every participant, so a deep bench of senior officers outscores a lone signatory.
  • Cluster Breadth (up to 24 pts): Insiders above the 3-insider detection minimum. A 3-insider cluster scores 0 here: three is the price of admission, not a distinguishing feature.
  • Time Velocity (up to 12 pts): Same-day trades score 12, a 5-day spread scores 2. Tight compression is rare and statistically meaningful.
  • Filing Speed (up to 8 pts): Deliberately low-weight. Form 4 is legally due within two business days, so filing promptly is compliance rather than conviction — only the unusual late tail moves the score.

Grade thresholds: A (≥70), B (≥45), C (<45). Across the current population that is roughly 11% A, 55% B, 34% C.

Short Radar

SOLD clusters and SHORT RADAR clusters are distinct. Short Radar is reserved for sell clusters that pass additional filters: $500,000+ aggregate value with 3+ distinct sellers — or $250,000+ if C-suite insiders (CEO, CFO, President) are among the sellers. These filters are designed to separate large coordinated liquidations from routine diversification or tax-driven sales.

ClusterSignal also parses Form 4 footnotes to detect 10b5-1 plan language. Routine pre-scheduled plan sales are noted as such. Amended or modified 10b5-1 plans are treated as higher-conviction, since the insider voluntarily changed a pre-committed program.

What the scores are not

Scores are research inputs, not investment recommendations. A Grade A cluster does not mean a stock will go up. A Short Radar flag does not mean you should sell or short. Insider activity is one signal among many. Evaluate each cluster alongside earnings context, valuation, sector trends, and your own research.

Past cluster patterns do not guarantee future results. ClusterSignal is not a registered investment advisor. Nothing on this site is financial advice.