Insider Buying Clusters (Legal Filings) — a Signal With Decaying but Nonzero Numbers

Legally disclosed insider stock purchases (Form 4 filings) carry a measurable, positive abnormal-return signal in the academic literature — but the signal is concentrated in the first days after the trade, mostly consumed by a mandatory 2-day disclosure lag, weaker outside the US, and largely absent for large-caps in the post-2000s period. Clustering (multiple insiders buying together) roughly doubles the effect versus solitary purchases.

Academic evidence on abnormal returns following disclosed insider purchases

- Seyhun (1975-1981, 60,000 transactions, 769 NYSE/AMEX firms): net insider-buying firms earned 4.3% abnormal returns over 300 days; net-selling firms underperformed by 2.2%. — https://www.insidermonkey.com/insider-trading/education-center/academic-studies-on-insider-trading/3 - Lakonishok & Lee (1975-1995, NYSE/AMEX/Nasdaq): insider buys in small-cap firms earned 7.4% abnormal returns over the following 12 months; insider sales showed no meaningful underperformance (buy/sell asymmetry). — https://quantdecoded.com/en/insider-trading-signals-informative-trades - Jeng, Metrick, Zeckhauser (1975-1996, 563,863 transactions: 214,897 purchases + 348,966 sales): purchase portfolio earned abnormal returns of about 40 basis points/month; about one-sixth of the return accrued within the first 5 days, about one-third by month's end — this is the key timing fact for anyone trying to trade the signal. — https://www.nber.org/papers/w6913 - Cohen, Malloy, Pomorski (2012, *Journal of Finance*, "Decoding Inside Information"): "opportunistic" insider trades yielded 82 bps/month abnormal return; "routine" (scheduled/repeated) trades earned essentially zero — the trade's *type*, not just its direction, carries the signal. — https://www.nber.org/system/files/working_papers/w16454/w16454.pdf

Cluster buying vs single-insider signal strength

- Alldredge & Blank (2017, US insider trading 1986-2016): clustered purchases generated 2.1% monthly abnormal returns vs. 1.2% for solitary purchases; over 21 trading days, 3.8% cluster vs. 2% non-cluster; the gap widened to 2.5% over a 90-day horizon. — https://www.2iqresearch.com/blog/profiting-from-insider-transactions-a-review-of-the-academic-research - Kang, Kim, Wang (2018) replicate the same magnitude: clusters (3+ insiders buying within a 30-day window) achieve 3.8% over 21 trading days vs. 2% for non-cluster — roughly a 2x multiplier confirmed independently. — https://www.2iqresearch.com/blog/what-is-cluster-buying-and-why-is-it-such-a-powerful-insider-signal - Dardas (2011, European insider transactions): "high conviction" purchases (the cluster-like subset) generated an average 12-month excess return of 20.94% — the largest figure in these notes, from a European rather than US sample. — https://www.2iqresearch.com/blog/profiting-from-insider-transactions-a-review-of-the-academic-research - Operational definition (academic consensus, synthesized): cluster buying = 3+ insiders buying on the open market within a 30-day window, ideally including at least one independent director (director-only clusters are read as a stronger, less self-interested signal). Some sources use a 2-insider minimum instead. — https://www.heygotrade.com/en/blog/insider-buying-signals-should-you-follow-the-c-suite/

When it applies

the cluster premium is a *relative* multiplier on top of the base single-insider effect, not a separate strategy — apply it as a filter (require 3+ open-market buyers in 30 days) on top of the base signal in Post-Earnings-Announcement Drift (PEAD): A Real, Decaying Anomaly With Numbers-style event studies, not as a standalone screen.

Sample period, number of events, significance

- Seyhun 1975-1981: 60,000 transactions, 769 firms, significant at both firm and aggregate level. — https://www.insidermonkey.com/insider-trading/education-center/academic-studies-on-insider-trading/3 - Jeng et al. 1975-1996: 563,863 transactions; t-tests on the purchase portfolio's ~40 bps/month abnormal return are significant. — https://www.nber.org/papers/w6913 - Alldredge & Blank, 1986-2016 (31-year window): statistically significant cluster-vs-non-cluster gap across thousands of events. — https://onlinelibrary.wiley.com/doi/10.1111/jfir.12172 (cited via secondary source) - Recent microcap study (arXiv 2602.06198, 2018-2024): 17,237 open-market purchases across 1,343 microcap firms ($30M-$500M market cap). A gradient-boosting model identifies clusters preceding a 30-day CAR > 10% at 69% recall, 38% precision; the appreciation-vs-decline bucket difference is significant at t = -5.13, p << 0.001. This is the most recent and most granular sample in these notes, and confines the strongest confirmed effect to microcaps. — https://arxiv.org/html/2602.06198 - FTSE-350 study (UK, 2005-2015): 38,924 insider purchase transactions, 7,394 sales. Found much *weaker* results than the US literature: -0.84% CAR for purchases over a ±20-day window, significant at the 1% level but economically weak, and it does not survive realistic round-trip costs (see below). This is the one study in these notes that finds the signal essentially fails outside the US large/mid-cap context. — https://pmc.ncbi.nlm.nih.gov/articles/PMC8886886/

What does NOT work / where the edge dies

- Trading on the public filing alone captures only part of the return. The disclosure lag itself eats the edge: Ozlen & Batumoglu (Jan 2026) estimate 70-80% of insider-trading alpha lives in the unobservable window between the trade date and the public Form 4 filing — i.e. before a retail follower can act. — https://larryswedroe.substack.com/p/when-the-insiders-and-the-news-disagree - SEC's own 2-business-day Form 4 deadline (Sarbanes-Oxley) means a retail follower reacting at disclosure time captures only an estimated 20-30% of the abnormal return that accrued around the trade. — https://www.investor.gov/introduction-investing/general-resources/news-alerts/alerts-bulletins/investor-bulletins-69 - Rule 10b5-1 scheduled trading plans (SEC-mandated 90-day cooling-off period since 2023 between plan adoption and first trade) strip information content from what looks like a "buy" — Cohen-Malloy-Pomorski's "routine" vs "opportunistic" split above is the empirical version of this: routine/scheduled buys earn ~0% abnormal return. Filter these out before treating a cluster as a signal. — https://www.investor.gov/introduction-investing/general-resources/news-alerts/alerts-bulletins/investor-bulletins-69 - UK large/mid-cap purchases do not clear realistic transaction costs. FTSE-350 2005-2015: -0.84% CAR is "insufficient to overcome realistic round-trip costs of 2.9-11.3%" — net profitability disappeared in the most recent decade of that sample, even though the raw effect stayed statistically significant. This is a direct example of a signal for Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one: statistically real but not tradeable net of cost. — https://pmc.ncbi.nlm.nih.gov/articles/PMC8886886/ - Post-publication decay applies here too. McLean & Pontiff's general finding that documented anomalies see returns 35-50% smaller after academic publication is cited as applying to the insider-signal literature as institutional following increased. No insider-specific before/after split is given in these notes — flagged as a generic mechanism, not an insider-specific number. — https://www.heygotrade.com/en/blog/insider-buying-signals-should-you-follow-the-c-suite/ - Large-cap edge has largely evaporated since the 2000s; the small-cap/microcap advantage persists (see the arXiv 17,237-purchase study above) but with materially lower returns than the 1975-1995 baseline studies (Seyhun, Lakonishok-Lee, Jeng et al.), and with severe liquidity constraints on execution size. — https://pmc.ncbi.nlm.nih.gov/articles/PMC8886886/ - (unverified) A May-June 2026 pattern study found that *divergent* signals — insiders buying while news sentiment turns negative — yielded +2.46% next-day excess return, but on only 7 observations over 12 days across 536 stocks tested. The source itself flags this as not statistically robust; kept here only as a pointer, not as evidence. — https://larryswedroe.substack.com/p/when-the-insiders-and-the-news-disagree

Why this page matters for the wiki

This is a case where the evidence bar in What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique separates a real number from a tradeable strategy: the abnormal return is well-replicated across four independent samples spanning 1975-2024 (Seyhun, Lakonishok-Lee, Jeng et al., the 2026 microcap study), and clustering roughly doubles it — but the disclosure-lag mechanic means most of that return is earned before a filing-reader can act, and the FTSE-350 result shows the same nominal effect fails net of cost outside the microcap US context. Treat this as *partially verified*: real and replicated at the event-study level, unresolved at the "can you actually capture it" level. See Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one for the cost-threshold method and Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real for how the McLean-Pontiff decay mechanism generalizes here.

Related

- What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique — this page is a worked case of the four-requirement bar: effect size and proof regime are strong (multi-decade academic replication), but cost-survival and post-disclosure capture are the open questions that keep confidence at 0.6. - Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one — the FTSE-350 -0.84% CAR vs 2.9-11.3% round-trip cost figures here are a direct, sourced example for that page's "cost thresholds that kill strategies" section. - Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real — the McLean-Pontiff 35-50% post-publication shrinkage cited here is the general mechanism that page documents in depth. - Post-Earnings-Announcement Drift (PEAD): A Real, Decaying Anomaly With Numbers — another disclosure-driven, short-horizon drift signal with the same "does the retail follower capture it before it decays" structure.