Form 4 Cluster Buys vs. Single Buys: What 5,947 Open-Market Purchases Show (2024–2026)
A Form 4 cluster buy is three or more company insiders making open-market purchases of the same stock within 30 days. We tested whether cluster buys beat single insider buys across 5,947 purchases at activist-target companies. Over 90 days, clusters delivered a higher median return (+3.07% vs +0.35%) and hit rate, but not a higher mean. The cluster edge is consistency, not magnitude.
The claim everyone repeats, and what the data says now
Cluster buying has a long academic pedigree. Lakonishok and Lee (2001) found the predictive content of insider purchases rises when multiple insiders buy, with the strongest results where at least three insiders bought sizable amounts of small value stocks. Jeng, Metrick, and Zeckhauser (2003) showed insider purchase portfolios earned abnormal returns above 50 basis points per month between 1975 and 1996. Cohen, Malloy, and Pomorski (2012) sharpened the picture: only "opportunistic" (non-routine) insider trades carry information, worth roughly 82 basis points per month value-weighted.
From this literature, a practitioner shorthand emerged and now circulates on every retail insider-tracking blog: cluster buys generate roughly double the excess return of single buys. The data behind that shorthand is decades old. Nobody, as far as we can find, publishes a current out-of-sample cut.
We ran one. The result complicates the shorthand: in our 2024–2026 window, cluster buys did not beat single buys on mean 90-day return. They beat them on median return, hit rate, and dispersion. If you follow clusters expecting bigger wins, the recent record says no. If you follow them expecting fewer disasters, the recent record says yes — modestly, and without statistical significance.
Methodology
The sample is every Form 4 transaction coded "P" — open-market or private purchase, per the SEC Form 4 transaction codes — in the 13dwatch insider_transactions table as of June 4, 2026: 5,947 purchase rows across 51,059 total Form 4 transactions, collapsing to 5,584 unique insider-day purchase records. Form 4s must be filed with the Securities and Exchange Commission (SEC) within two business days of the transaction under Section 16 of the Exchange Act.
One scope note that cuts both ways: the 13dwatch pipeline ingests Form 4s at activist-target issuers — companies with at least one Schedule 13D or 13G filing since 2018 in our activist filings database. This skews the universe toward small- and mid-cap, event-driven names. It is not the whole market, and we don't claim it is. It is, however, exactly the universe where insider signals and activist pressure interact — the intersection this site exists to measure.
Definitions:
- Cluster buy: three or more unique insiders (distinct SEC filer CIKs) making code-P purchases of the same issuer within a trailing 30-day window. The trigger date is the day the third unique insider buys. Repeat triggers within 90 days of a prior trigger at the same issuer are folded into one episode.
- Single buy: a code-P purchase with no other unique insider buying the same issuer within 30 days either side, with the same 90-day per-issuer cooldown.
- Return measurement: 90-calendar-day forward return, close-to-close on adjusted closes, entry at the first trading close on or after the trigger. Benchmark-adjusted returns subtract the SPDR S&P 500 ETF (SPY) return over the identical dates. Risk-free rate from FRED series DGS3MO (average 4.40% annualized over the window, about 1.08% per 90 days).
That yields 325 cluster episodes and 1,100 single-buy episodes since 2018. Return statistics use the subset with trigger dates between May 1, 2024 and March 5, 2026 — the span where a full 90-day forward price window exists in our price history (13dwatch daily prices plus Yahoo Finance for tickers outside our price table): 67 measurable cluster episodes and 250 measurable singles.
The numbers
| 90-day forward | Cluster buys (n=67) | Single buys (n=250) |
|---|---|---|
| Mean raw return | +3.89% | +5.77% |
| Median raw return | +3.07% | +0.35% |
| Hit rate, raw | 52.2% | 50.8% |
| Mean SPY-adjusted | +0.23% | +1.57% |
| Median SPY-adjusted | −1.25% | −3.38% |
| Hit rate, SPY-adjusted | 49.3% | 45.2% |
| Standard deviation, raw | 35.96% | 38.33% |
| 90-day Sharpe (vs 3-month Treasury) | 0.078 | 0.122 |
90-calendar-day forward returns after insider purchase events at activist-target issuers, triggers May 2024 – March 2026. Source: 13dwatch insider_transactions (SEC EDGAR Form 4), 13dwatch daily_prices + Yahoo Finance, FRED DGS3MO. Event-level CSV available free on request: contact@13dwatch.com.
Seven findings
- Singles won on mean, +5.77% vs +3.89% raw. The gap is 1.88 points with a standard error near 5 points — statistically indistinguishable from zero. Neither side "beats" the other on averages at conventional significance levels.
- Clusters won on median, +3.07% vs +0.35% raw. The 2.7-point median gap, and a 2.1-point gap after SPY adjustment, is the cleanest version of a cluster edge in this sample: the typical cluster outcome beats the typical single-buy outcome.
- The single-buy mean is a lottery-ticket artifact. Singles combine a worse median with a better mean — the signature of a fat right tail. A handful of large winners drag the average up while the typical single buy goes roughly nowhere.
- Clusters miss disaster slightly more often. SPY-adjusted hit rate: 49.3% for clusters vs 45.2% for singles. Lower dispersion too (35.96% vs 38.33% stdev). Consistency is the product clusters actually deliver.
- Neither signal printed money in 2024–2026. Mean SPY-adjusted returns were +0.23% (clusters) and +1.57% (singles) per 90 days. Compare the academic era: 50+ bps per month in Jeng, Metrick, and Zeckhauser's 1975–1996 sample. Either the signal has decayed with disclosure speed and screener ubiquity, or activist-target names are a harder tape, or both.
- Cluster activity is steady, not rare: 78 episodes totaling $718.9M in 2025–2026 alone. Most clusters are minimal — 222 of 325 episodes since 2018 had exactly 3 insiders. Nine-plus-insider pileups happened 3 times.
- Dollar-weighted leaderboards mislead without footnotes. The largest 2025–2026 cluster by dollars is a $515M set of purchases at Blackstone Private Real Estate Credit & Income Fund — a non-traded vehicle. Real Form 4s, no public ticker, no measurable return. Screens that don't annotate this will quietly mix non-traded entities into "top insider buying" lists.
The 2025–2026 cluster-buy leaderboard
Ten largest cluster episodes by code-P dollars, January 2025 through June 2026:
| Trigger | Ticker | Issuer | Insiders | Cluster $ |
|---|---|---|---|---|
| 2025-10-21 | — | Blackstone Private Real Estate Credit & Income Fund† | 3 | $515.0M |
| 2026-05-20 | BRCB | Black Rock Coffee Bar‡ | 3 | $64.5M |
| 2025-09-08 | ACDC | ProFrac Holding | 3 | $20.1M |
| 2025-02-27 | RCUS | Arcus Biosciences | 3 | $15.4M |
| 2025-10-07 | ZBIO | Zenas BioPharma | 3 | $11.8M |
| 2026-05-15 | EROK | EagleRock Land† | 5 | $10.5M |
| 2026-05-15 | BXDC | Blackstone Digital Infrastructure Trust† | 7 | $8.9M |
| 2026-03-02 | PMN | ProMIS Neurosciences | 3 | $8.5M |
| 2025-11-20 | GLOO | Gloo Holdings | 5 | $8.3M |
| 2026-04-07 | LW | Lamb Weston | 3 | $5.9M |
† Non-traded or thinly traded vehicle; excluded from return statistics. ‡ Dollar total dominated by one 10%-owner purchase (Cynosure Group, $64.4M on 2026-05-15); the officer cluster around it totaled ~$363K. Source: SEC EDGAR Form 4 filings via 13dwatch.
The ProFrac row is instructive for anyone counting "unique insiders." The August–September 2025 episode comprises THRC Holdings LP, $10.0M, Farris Wilks, $10.0M, and Executive Chairman Matt Wilks, $94.5K. THRC is a Wilks-family vehicle. Three filer CIKs, arguably one family's conviction. Our counts treat distinct CIKs as distinct insiders; any screen that does the same — including OpenInsider's cluster-buy screen — inherits the same inflation.
Why the cluster signal is weaker than the textbooks say
The canonical studies measured a slower world. Until 2002, insiders could report trades up to 10 days after month-end; the Sarbanes-Oxley Act's Section 403 compressed that to two business days, and the SEC's 2003 mandate for electronic Form 4 filing put each one on EDGAR essentially in real time. A signal that took weeks to reach the market in the Lakonishok and Lee sample now reaches every screener in seconds — OpenInsider publishes a free cluster-buy list daily, and Bloomberg, Verity (formerly InsiderScore), and 2iQ Research sell the same events to institutions.
Two non-exclusive hypotheses fit our numbers. First, crowding: when everyone sees the cluster on day one, the front half of the abnormal return gets competed away — consistent with Jeng, Metrick, and Zeckhauser's finding that half the abnormal return historically accrued within the first month. Second, composition: activist-target issuers are disproportionately stocks under pressure, where insider buying is sometimes a defensive signal to the market rather than a value judgment — management signaling resolve during a proxy contest is not the same trade as a CFO quietly buying a forgotten small-cap. We can't separate the two with 67 events. We can say the naive version of the signal, "3 insiders bought, go long," carried no mean edge in this window.
How to track cluster buys without a Bloomberg
Three workable tiers. Free and manual: SEC EDGAR search lists every Form 4 as filed; cross-referencing buyers by issuer is your spreadsheet problem. Free and screened: OpenInsider's cluster-buy screen does the grouping but not the verification — it inherits the affiliated-filer inflation shown above and mixes in non-traded vehicles. Joined to activist context: the 13dwatch feed cross-references each cluster against live Schedule 13D and 13G filings, short interest, and institutional consensus — the combination this study suggests is where the residual signal lives. The raw filing data underneath all three tiers is identical and public; the work is in the joins.
Limitations a skeptical analyst should press on
- Sample selection. Activist-target issuers only. Results are conditional on that universe.
- Survivorship. 10 of 77 cluster events (13%) and 65 of 316 single events (21%) lacked usable price history (mostly delistings) and were dropped. Since delistings skew toward bad outcomes and hit singles harder, the true single-buy numbers are plausibly worse than reported.
- Small N. 67 measurable clusters. Standard errors on means are roughly ±4.4 points. We present medians and hit rates for that reason, and claim no statistical significance anywhere.
- Affiliated filers. Distinct CIKs can be one economic actor (see ProFrac).
- Undercounting. Historical Form 4 rows ingested before June 2026 captured only the first transaction block per filing; a repair job is rebuilding the remainder. Unique-insider counts are unaffected; per-episode dollar totals may be understated.
- One outlier excluded. A single-buy event with a >500% absolute move was dropped as a data-error guard.
What this means for emerging managers
The recycled "clusters double your return" line is not supportable on 2024–2026 data in this universe — and any vendor still quoting it is quoting a pre-EDGAR tape. What survives scrutiny is narrower and more useful: a cluster of three or more genuinely distinct insiders buying in the open market shifts the distribution of 90-day outcomes toward the center-right — better median, better hit rate, less dispersion — without raising the average. That makes clusters a screening filter, not an alpha machine: a way to rank which activist situations have aligned management, not a standalone trade. The interesting trades in our data sit where the cluster coincides with an active 13D — the overlap we measured in our activist-insider cluster study and track live.
Get cluster-buy events cross-referenced against live 13D filings in the 13dwatch feed — $1,500/month, or read the free research index first.
Frequently Asked Questions
What is a Form 4 cluster buy?
A Form 4 cluster buy is three or more distinct company insiders — officers, directors, or 10% owners — reporting open-market purchases (transaction code P) of the same company's stock within a 30-day window on SEC Form 4 filings. The pattern is read as multiple informed people independently judging the stock undervalued.
Do cluster buys beat single insider buys?
Not on average returns in recent data. Across 67 cluster and 250 single-buy events at activist-target companies from May 2024 to March 2026, single buys had the higher mean 90-day return (+5.77% vs +3.89%), while clusters had the higher median (+3.07% vs +0.35%) and hit rate. No difference was statistically significant.
How many insiders make a cluster buy?
The standard threshold is three unique insiders within 30 days, following Lakonishok and Lee (2001). In the 13dwatch database, 223 of 325 cluster episodes since 2018 had exactly three insiders; only 3 episodes had nine or more. Verify the insiders are genuinely distinct — family investment vehicles file under separate CIKs.
How fast are insider purchases disclosed?
Form 4 must be filed within two business days of the transaction under Section 16, as amended by Sarbanes-Oxley Section 403, and appears on SEC EDGAR immediately upon acceptance. Before 2002, insiders had until the 10th day of the following month — one reason older academic return estimates likely overstate today's exploitable edge.
Where can I see current cluster buys for free?
OpenInsider publishes a free daily cluster-buy screen, and SEC EDGAR carries every underlying Form 4. For cluster buys cross-referenced against live activist 13D filings and short interest, 13dwatch publishes the joined feed and a free research index.
This is research, not investment advice. Long Street Consulting LLC.