Cluster Buy Signals — Why Multiple Insiders Buying Together Matters

What counts as a cluster buy?

The academic definition of a cluster buy varies across studies but converges on a common structure: multiple independent insiders at the same issuer trading in the same direction within a short time window. Form4API's implementation uses the following definition, which reflects the most widely cited academic parameters:

• 3 or more distinct insiders — each identified by their individual CIK, not transaction count. One insider buying three times in a week is one participant, not three.
• Same company — identified by issuer CIK. Insiders at a parent and subsidiary filing separately are treated as the same company if the beneficial ownership relationship maps correctly.
• Same direction — all buys (code P on Table I). Mixed buy/sell windows do not constitute a cluster buy.
• 5-day rolling window — calendar days from first transaction to last. The window slides forward daily, so a new cluster can form each day as older transactions age out.
• Open-market transactions only — transaction code P. Grants (A), option exercises (M), and tax-withholding transactions (F) are excluded automatically.
• 10b5-1 plan trades excluded — Form4API excludes transactions flagged as is10b5Plan: true by default. Pre-scheduled plan trades carry no information signal. See the 10b5-1 guide for context.

Alternative academic definitions use a 30-day window, a 2-insider minimum, or an excess-return threshold to confirm the signal post-hoc. Form4API uses the tighter 5-day, 3-insider definition because it produces higher precision at the cost of lower recall — fewer false positives, even if some genuine consensus events are missed at the edges.

What we found when we tested it against our own data

This is the theoretical case for why cluster buys should be a stronger signal than single-insider buys — and it's the reasoning behind the academic result cited below. We tested it against Form4API's own data and it didn't hold up past the one-month mark. Here's the intuition first, then what actually happened.

Insiders buy for one reason

An insider may sell shares for many reasons that have nothing to do with their views on the company: tax obligations, diversification, children's college tuition, divorce settlements, or pre-arranged selling plans. But when an insider spends their own money on the open market to buy additional shares of their employer, the most parsimonious explanation is that they believe the shares will appreciate. This makes open-market purchases (transaction code P) the highest-signal subset of all Form 4 transactions.

Independent agreement amplifies the signal

A single insider buying could be idiosyncratic — a CFO who personally believes the market misunderstands a complex accounting item, or a director who wants to signal public confidence after a negative news cycle. But when a CEO, CFO, and two independent directors all independently purchase shares within the same week — without coordinating (which would constitute improper insider trading) — their shared action reflects independent convergence on a shared view. The probability that each of them is wrong independently, in the same direction, in the same week, is lower than the probability that any single one is wrong.

Then we measured it, and it reversed

Across 75,436 open-market buys in Form4API's database (2023-04 through 2026-09), cluster buys (3+ insiders, 5-day window) underperformed non-cluster buys (1-2 insiders) at every horizon on mean return, and at 3 and 6 months on median return as well:

• 1 week: cluster mean +2.89% vs non-cluster +4.09% (median: +0.78% vs +0.59% — clusters were fractionally ahead here)
• 1 month: cluster mean +3.52% vs non-cluster +4.65% (median +1.43% vs +1.15% — still fractionally ahead)
• 3 months: cluster mean +5.60% vs non-cluster +8.43% (median +2.23% vs +2.64% — the median flips negative here too)
• 6 months: cluster mean +8.09% vs non-cluster +12.00% (median +3.33% vs +5.11%)

The gap widens with horizon and every reported difference at 3 and 6 months has a 95% bootstrap confidence interval that excludes zero. This is a genuine reversal of what this page said until now, not a refinement of it — we checked our own signal against our own data and the result wasn't the one we expected. Full numbers, the by-insider-count breakdown, and the caveats (there's an unresolved sector confound) are in our companion study: We Tested Our Own Cluster-Buy Signal — Here's What We Found →

What the literature says vs. what we measured

The most-cited foundational study is Lakonishok and Lee (2001), "Are Insider Trades Informative?" — published in the Review of Financial Studies. This is what the literature found on market-adjusted returns from 1975-1995 data. It's the reason the intuition above exists, and it's the claim we set out to check against our own, much more recent, unadjusted data:

• Cluster purchases outperform single-insider purchases by roughly 2x in excess return over 21-trading-day windows. The directional finding has been replicated in multiple subsequent studies across different time periods and international markets.
• Small-cap stocks show stronger effects. The information asymmetry between insiders and the public market is larger for small and mid-cap companies where analyst coverage is sparse and institutional ownership is lower. Cluster buy signals in the Russell 2000 have historically shown larger excess returns than in the S&P 500.
• The effect is strongest for open-market purchases. Studies that filter to transaction code P (open market buy) and exclude option exercises, compensation grants, and 10b5-1 plan trades consistently show stronger results than studies using unfiltered Form 4 data.
• Whether the alpha still holds is now an open question at Form4API. Some post-2015 replications report a compressed but still-positive effect. Our own test, on 2023-2026 open-market Form4API data, found cluster buys underperforming non-cluster buys rather than a smaller version of the same outperformance — see below.

Important caveats: all academic studies on insider trading signals are subject to survivorship bias (companies that experienced insider buying and then failed are underrepresented in long-term datasets), selection bias (academic samples often exclude illiquid securities where execution is impractical), and data-snooping risk. The cluster buy signal is best interpreted as a probabilistic screen, not a deterministic trading rule.

That caveat about survivorship and selection bias applies to the academic literature. It also turns out to matter for testing the claim on fresh data: when we ran the same 5-day, 3-insider definition against Form4API's own 75,436-buy population, the result was the opposite sign from Lakonishok & Lee. See "The full study" below for the numbers, the by-insider-count breakdown, and an honest accounting of an unresolved sector confound (cluster buys skew toward REITs and community banks) that keeps this correlational rather than proven causal.

The full study

We ran a full breakdown — cluster vs. non-cluster returns at four horizons, a by-insider-count cut that shows the underperformance appears as a step right at the 3-insider threshold rather than a gradual decline, robustness checks across 1,696 companies and four years, and the caveats (an unresolved sector confound, since top cluster tickers skew toward REITs and community banks) — in a companion article: We Tested Our Own Cluster-Buy Signal — Here's What We Found →

How to detect cluster buy signals programmatically

There are two approaches: build your own cluster-detection logic on raw Form 4 data, or use Form4API's pre-computed signals endpoint.

Manual approach on raw Form 4 data

To detect clusters yourself against raw Form 4 data from EDGAR or the /v1/transactions endpoint:

1. Filter to transaction_code = "P" (open-market purchase)
2. Exclude transactions where is10b5Plan = true
3. Group remaining transactions by issuer CIK and transaction date
4. For each transaction, check whether 2+ other distinct insiders (by filerCik) filed P transactions at the same issuer within the preceding 5 calendar days
5. If yes, emit a cluster event with the participant list, dates, and aggregate share value

This approach requires a rolling window computation and deduplication by insider CIK. The main failure mode is double-counting insiders who file multiple transactions in the same window — always deduplicate on filerCik before counting participants.

Pre-computed signals via Form4API

Form4API pre-computes cluster signals and exposes them via /v1/signals:

# Get recent cluster buy signals across all tickers
curl "https://api.form4api.com/v1/signals?cluster_buy=true&per_page=20" \
  -H "X-Api-Key: $FORM4API_KEY"

# Get cluster signals for a specific ticker
curl "https://api.form4api.com/v1/signals?cluster_buy=true&ticker=NVDA" \
  -H "X-Api-Key: $FORM4API_KEY"

Each signal in the response includes the ticker, company name, signal date, the buy/sell ratio, cluster-buy and cluster-sell flags, and the number of participating insiders. Cluster signals are computed from open-market transactions with 10b5-1 plan trades already excluded. The /v1/signals endpoint is available on the Business plan and above. See /docs for the full schema.

Combining cluster signals with quality filters

Raw cluster signal detection produces a noisy output. The following filters, applied in priority order, remove the most common false positives:

These filters assume cluster buys carry a positive signal worth refining. Our own data (above) complicates that assumption at 3+ month horizons — treat the filters below as ways to reduce noise in the cluster population you detect, not as a guarantee that a filtered cluster buy will outperform.

• Open-market only (code P). Exclude grants (A), option exercises (M), and tax withholding (F). Only open-market purchases reflect discretionary capital commitment.
• Exclude 10b5-1 plan trades. Pre-scheduled trades have no information content. This is Form4API's default.
• Minimum dollar threshold. Filter out purchases below $10,000-$25,000 per participant to exclude symbolic director purchases that add no economic signal. For micro-cap companies, the absolute threshold may need to be lower.
• Officer and director participants only. Restrict the participant list to insiders with roles of CEO, CFO, COO, President, Director, or VP — exclude passive 10%+ ownership positions held by PE firms or institutional investors (who may have motivations unrelated to corporate prospects).
• Require at least one C-suite participant. A cluster formed entirely of outside directors who buy small symbolic amounts carries less signal than a cluster that includes at least one operating executive who has direct knowledge of business performance.
• Minimum market cap or trading volume. Penny stocks and micro-caps with thin trading volume can generate cluster signals that are not practically exploitable due to liquidity constraints. A minimum average daily volume filter (e.g., $500K) keeps the output actionable.

Applying all five filters typically reduces the raw cluster signal universe by 60-75%, but the remaining signals have materially better precision. Form4API exposes the insider role field on every transaction, making role-based filtering straightforward in both direct API queries and post-processing.

Cluster sell signals — are they symmetric?

The cluster sell equivalent — three or more insiders filing open-market sale transactions (code S) within the same 5-day window — is a natural counterpart to the cluster buy signal, but the academic literature treats it quite differently.

The asymmetry in insider sale signals stems from the diversity of motivations for selling. An insider may sell to: meet a tax obligation, diversify a concentrated position, fund a real estate purchase, satisfy a pre-arranged selling plan, exercise options near expiry, or respond to a personal financial emergency. None of these motivations are related to their view of the company's prospects, but all of them produce the same Form 4 sale record.

A cluster sell formed by three directors who all happen to have restricted stock vesting in the same month — and who all sell simultaneously — carries almost no information. The cluster pattern here is driven by a calendar artifact, not shared bearish conviction. This is why academic studies consistently find that cluster sell signals have lower precision and shorter shelf-life than cluster buy signals.

The exception worth monitoring: cluster immediate sells in the days following an earnings release, especially from C-suite insiders who are not on scheduled plans and who had no public pre-announced intention to sell. This pattern can indicate that insiders interpreted the company's own earnings results more negatively than the market did. Form4API does compute cluster sell signals and they are available in the /v1/signals output with type=cluster_sell.

Limitations of the cluster buy signal

• Small-cap bias. Cluster signals are more frequent and more impactful in small and mid-cap companies where insider ownership is proportionally larger and analyst coverage is thinner. Large-cap cluster buy signals exist but are rarer and typically smaller in relative size, making them harder to act on at scale.
• The horizon matters, and not the way we expected. Our own data (above) shows cluster buys running roughly even with non-cluster buys at 1 week and 1 month, then falling behind at 3 and 6 months — the opposite of a signal that decays from an early edge. Treat any claim about cluster-buy timing, including this one, as provisional until the sector confound is resolved.
• Not standalone — requires confirmation. A cluster buy signal is most useful as a screening tool that narrows a stock universe, not as a standalone buy trigger. Combining it with fundamental screens (positive earnings revisions, low short interest, improving free cash flow) and technical confirmation (uptrending price, volume support) produces more reliable outcomes than using the cluster signal alone.
• Insider knowledge is company-specific. A cluster buy at a semiconductor company may reflect visibility into an upcoming major design win. But insiders typically do not have superior macro-economic insight — they can be wrong about sector-wide headwinds or interest rate sensitivity even when they are right about their own company.
• Coordinated buying is legally prohibited. If cluster participants are actually coordinating their purchases (sharing non- public information about their intent to buy), that constitutes a violation of SEC rules. The predictive power of cluster signals assumes the participants are independent — and legally, they must be.

Frequently asked questions