Insider Trading Patterns That Actually Predict Stock Moves
Insider Trading Patterns That Actually Predict Stock Moves
Three patterns separate informative insider trading from noise: extended trade sequences, after-hours disclosure timing, and cluster buys among multiple insiders in the same window. When you control for role and transaction type, these three account for most of the abnormal-return signal that academic research has isolated in Form 4 data. Isolated, single-day trades from beneficial owners carry far less information than a CEO’s third purchase in a nine-day stretch, filed after the closing bell.
If you’re building a screen, start here:
-
Cluster-buy filter: Flag tickers with two or more open-market “P” code purchases from different insiders within a 7 to 14 day window.
-
Disclosure-timing filter: Isolate trades reported after market close, which the literature associates with larger average abnormal returns.
-
Role segmentation: Separate CEO and CFO transactions from director and beneficial-owner transactions before scoring anything.
-
Sequence-duration tag: Distinguish one-off trades from multi-trade sequences spanning several weeks.
Statistic to anchor your prior: Academic work covering insider trading patterns finds that both purchases and sales predict abnormal returns once you control for how long a trading sequence runs. One caveat before you build anything: Form 4 filings lag actual trade execution by up to two business days, and no legitimate research use of this data should cross into acting on material nonpublic information yourself.
Key Takeaways
Insider trading patterns become reliable predictors of abnormal returns only when researchers separate trade sequences by duration, disclosure timing, and insider role rather than treating Form 4 filings as a single undifferentiated dataset.
| Point | Details |
|---|---|
| Sequence duration matters most | Extended trading sequences carry more signal than isolated single trades. |
| After-hours filings deserve attention | Disclosures filed post market close correlate with larger average abnormal returns. |
| Cluster buys need parameters | Define clusters as 2+ insiders buying within a 7 to 14 day window before testing. |
| Role segmentation sharpens signals | CEO and CFO cluster buys often outperform pooled or director-only signals. |
| 10b5-1 flags prevent false positives | Scheduled trading plans should never score as discretionary conviction buys. |
Table of Contents
-
What the Academic Research Says About Insider Trading Patterns
-
An Editorial Take on Bridging Academic Signals and Practitioner Screens
What the Academic Research Says About Insider Trading Patterns
Two research streams dominate this field, and they largely agree while approaching the problem from different angles.
-
Biggerstaff, Cicero, and Wintoki (2020) examined trade-sequence duration directly. Their central finding: insiders trade in short bursts when their informational edge is fleeting, and in longer, extended sequences when it persists. Sales and purchases both predicted abnormal returns once the researchers controlled for sequence duration, and after-hours disclosure correlated with both longer sequences and larger effect sizes.
-
Tamersoy and colleagues (2013) took a different route, mining roughly 12 million Form 4 transactions filed between 1986 and 2012. Their large-scale exploratory analysis surfaced temporal clustering, role-based behavioral differences, and evidence of short-swing profit patterns among specific insider cohorts. A companion write-up on the same dataset, hosted through CMU’s network analysis group, pushed further into network structure, showing that insiders who trade similarly tend to cluster into small, persistent groups.
Insiders who share a company, a sector, or even an advisory relationship often exhibit correlated trading timing that isolated event-study designs miss entirely.
Where these two streams diverge is methodological. Biggerstaff and colleagues built an event-study framework around abnormal returns and sequence duration as the primary variable of interest. Tamersoy’s team ran a broader, more exploratory pass across the full historical Form 4 corpus, prioritizing pattern discovery over hypothesis testing. Neither paper treated 10b5-1 scheduled trading plans identically. This matters: a trade executed under a pre-arranged 10b5-1 plan carries essentially zero discretionary signal, and any screen that fails to separate scheduled sales from opportunistic ones will dilute its own results. Event-window choice also varies across studies, meaning replicated tests should not expect identical effect sizes even when the underlying pattern definitions look similar on paper.
The Four Empirical Patterns Worth Operationalizing
Academic findings translate into four measurable, testable pattern categories.
Sequence length versus isolated trades. A sequence is a run of transactions by the same insider within a bounded window, typically defined as no more than 30 days between consecutive trades. Isolated trades, by contrast, sit alone with no follow-up activity for months. Biggerstaff and colleagues found that trading-duration patterns sharpen the predictive power of both buy and sell signals substantially compared to treating every trade as independent.
After-hours disclosure timing. Form 4 filings submitted after regular trading hours are disproportionately associated with extended sequences and, on average, larger abnormal returns than filings submitted during market hours. This is a subtle signal because most retail-facing trackers ignore filing timestamp entirely and only report trade date.
Cluster buys and sells. A cluster occurs when multiple distinct insiders at the same company execute open-market transactions in the same direction within a short window, commonly 7 to 14 days. The network-based clustering evidence from the CMU study shows these groups aren’t random. Insiders who cluster once tend to cluster again, forming a repeatable behavioral fingerprint at the company level.
Role and sector variation. Signal strength is not uniform across insider types.
-
CEO and CFO cluster buys tend to carry more conditional alpha than director-only clusters.
-
Beneficial owners (often large institutional holders filing as insiders) trade on different motivations, frequently portfolio rebalancing, and dilute a pooled signal if not separated out.
-
Sector matters too: biotech and small-cap insider buying tends to precede sharper price moves than insider activity in large, heavily covered financials, partly a function of information asymmetry and analyst coverage density.
A trading-pattern screen that ignores these role and sector splits is effectively averaging a strong signal with a weak one and calling the blend “insider sentiment,” which understates how much a properly segmented CEO cluster-buy signal can actually tell you.
How to Turn Patterns Into Testable Signals
Academic findings are directional. Building a usable screen requires committing to specific parameter choices, and the choices you make here determine whether your backtest reflects reality or survivorship bias.
-
Set the cluster-buy rule. A common construction: flag a ticker when two or more distinct insiders file open-market “P” code purchases within a 7 to 14 day rolling window, with a combined transaction value above a minimum threshold (commonly $100,000 to $250,000, or a percentage of float for micro-caps).
-
Exclude non-discretionary transaction codes. Codes A (grant/award), M (option exercise), G (gift), and F (tax withholding) reflect compensation mechanics, not conviction. Strip these before scoring anything as a buy or sell signal.
-
Flag 10b5-1 language explicitly. Parse Form 4 footnotes for phrases referencing a Rule 10b5-1 trading plan. A scheduled sale under a pre-existing plan should never score the same as a discretionary open-market purchase.
-
Design the event window. Use symmetric abnormal-return windows, such as −1 to +5 days or −2 to +10 days, and report bootstrap confidence intervals rather than point estimates alone, since insider-return distributions carry heavy tails.
-
Account for transaction costs and slippage. Any backtest that ignores realistic execution costs on small-cap names will overstate net alpha meaningfully.
Watch for three pitfalls specifically: look-ahead bias from using amended filings that weren’t public at the signal date, multiple-testing inflation when you scan hundreds of parameter combinations without correction, and survivorship bias from testing only on tickers that still exist today. A pattern described well in smart-money research on event-driven timing applies here too: informed positioning ahead of a resolution tends to show up in timing clusters before it shows up in size.
Pro Tip: Run your cluster-buy screen twice, once including 10b5-1 flagged trades and once excluding them. The gap between the two result sets tells you how much of your “signal” was actually scheduled selling noise.
Building a Reproducible Data Pipeline From Form 4 and EDGAR
Every signal above depends on clean inputs. The raw EDGAR Form 4 feed contains fields you need to extract deliberately, not scrape incidentally.
-
Transaction code, share count, and price per share for every line item.
-
Filing timestamp (not just trade date), since after-hours disclosure is itself a signal.
-
Footnote text in full, to catch 10b5-1 plan references and gifting language.
-
Amendment status, since amended filings should replace prior entries rather than duplicate them in your dataset.
Merge filing data with a clean price series, matching trade date, and reserve out-of-sample periods for validation.
| Data Element | Why It Matters |
|---|---|
| Filing timestamp | Distinguishes after-hours disclosures from intraday ones |
| Footnote text | Reveals 10b5-1 plans and gift transactions |
| Transaction code | Separates discretionary buys from awards and exercises |
| Amendment flag | Prevents double-counting corrected filings |
Recommended controls include market beta, sector, and firm size, plus role-based segmentation in both in-sample and out-of-sample tests, since conditional alpha for CEO clusters often diverges sharply from pooled results.
What These Patterns Mean for Alpha and Compliance
Published effect sizes are real but modest once transaction costs enter the picture. A cluster-buy signal that shows several percentage points of abnormal return in an academic event window can shrink substantially after slippage on illiquid small-caps and after accounting for the fact that markets have grown more attentive to Form 4 filings over the past decade.
-
Surveillance and public tracking have compressed the historical edge; a pattern published in 2013 doesn’t carry the same alpha in 2026 simply because more capital now watches for it.
-
Liquidity constraints matter more for small-cap cluster signals than for large-cap ones, where trade sizes barely move the market.
-
Rule 10b5-1 plans and Form 4 reporting deadlines remain the SEC’s primary framework for separating scheduled trades from potentially informed discretionary ones, and enforcement actions continue to target trades executed outside those plans shortly before material news.
-
Promising extensions for further research include cross-venue signal propagation and network-based insider similarity clustering, both of which the Tamersoy dataset supports but few practitioners have fully mined.
How Filingsiq Turns This Research Into a Working Screen
Manually parsing thousands of Form 4 footnotes for 10b5-1 language isn’t a realistic weekly workflow for most research teams. Filingsiq automates that extraction directly.
-
Pulls transaction codes, filing timestamps, and footnote text automatically for every ticker in your workspace.
-
Flags 10b5-1 plan language so scheduled trades don’t contaminate your discretionary-buy screen.
-
Builds a summarized insider timeline per ticker, making sequence-length and cluster-buy patterns visible without manual spreadsheet work.
The gap between raw Form 4 volume and an actionable insider signal is almost entirely a filtering problem, and filtering well is what separates a usable screen from a noisy one.
A typical workflow: build a cluster-buy watchlist inside a dedicated ticker workspace, validate each flagged disclosure against its footnote text, then export a research memo summarizing the pattern for your investment committee. For teams formalizing this into a repeatable process, Filingsiq’s SEC filing analysis practices guide covers the validation step in more depth, and the red-flag detection guide extends the same filtering logic to broader filing analysis beyond insider trades.
Pro Tip: Export your cluster-buy watchlist as a research memo before earnings season starts. Insider filing volume spikes in the weeks after earnings, and a pre-built filtering workflow saves real time when the volume hits.
Ready to stop parsing footnotes by hand? FilingsIQ’s platform extracts these fields automatically across your entire ticker coverage list, and the pricing page outlines which plan tier fits a solo researcher versus a full team running this screen at scale.
An Editorial Take on Bridging Academic Signals and Practitioner Screens
The academic literature on insider trading patterns is more actionable than most practitioners give it credit for. The problem isn’t a lack of evidence, it’s that the evidence sits in event-study tables built for peer review, not in a format anyone can plug into a screen on a Tuesday morning.
Conventional retail insider-trading commentary treats every Form 4 purchase as equally meaningful, which the research flatly contradicts. A director’s routine annual purchase and a CEO’s third open-market buy in nine days, filed after the bell, are not the same event, and lumping them together is why so much “insider sentiment” content produces mediocre results.
If you take one thing from this research, prioritize sequence and timing over raw transaction count. A single well-timed, after-hours-disclosed cluster buy from senior management tells you more than a dozen scattered director purchases. Build your filtering logic around that hierarchy first, then layer in sector and role controls before you ever look at a backtest’s headline return.
Sources
-
Insider trading patterns — Journal of Corporate Finance (Biggerstaff, Cicero & Wintoki, 2020)
-
Investopedia: Insider trading — definition and regulatory context
FAQ
What Is the 7% Rule in Insider Trading Analysis?
There is no established academic “7% rule” tied to insider trading patterns in the research covered here; the term is more commonly associated with retail trading stop-loss strategies unrelated to Form 4 signal analysis.
What Is the 84% Rule in Trading?
What Are Some Recent Examples of Insider Trading Patterns Worth Studying?
Researchers commonly study cluster buys around earnings surprises and executive purchases ahead of merger announcements, both of which the SEC’s enforcement record documents as recurring discretionary-trade fact patterns.
How Do I Know Which Stocks Insiders Are Buying Right Now?
Screening current Form 4 filings for open-market “P” code purchases, then applying a cluster-buy and role-based filter, is the standard approach; platforms like FilingsIQ automate the extraction step so you can focus on validating the pattern rather than parsing raw filings.
Why Do After-Hours Form 4 Filings Matter for Signal Strength?
Trades disclosed after market close are more often part of extended sequences and carry larger average abnormal returns, based on Biggerstaff, Cicero & Wintoki’s findings.
Can 10b5-1 Trading Plans Distort Insider Trading Signals?
Yes. Scheduled sales under a Rule 10b5-1 plan reflect pre-arranged timing rather than discretionary conviction, so any screen that fails to flag this footnote language will understate the strength of genuinely informed trades.
Recommended
Related insights
Ready to analyze filings faster?
Try FilingsIQ free and turn SEC filings into actionable research in minutes.