Note on figures cited below. Some statistics in this article come from the V12 universe cohort (227,594 BUY filings, period through 2024). Since 2025-01-01 the live out-of-sample BUY universe (28 markets) shows a 52.96 percent win rate and a +8.59 percent mean 90-day return on 20,688 filings. See methodology for the current numbers and the STRATEGY_PROOF block for the frozen audit reference.
Insiders do not wait for your factor model to mature.
Insider trading research has a bad habit of stopping at the buy ticket. The executive bought 20,000 shares, the chairman added after a drawdown, the cluster signal flashed green, everyone nods gravely. Then the literature, and often the product, moves on. What tends to get less attention is the obvious second question, namely when those same insiders get out.
That omission matters. If an insider buy is meant to signal undervaluation, private confidence, or at least a willingness to bear risk alongside outside shareholders, the duration of that commitment is not a decorative detail. It is part of the signal. A purchase followed by a disposal six weeks later is not the same economic statement as a purchase that remains untouched for eighteen months. Both may be legal. Both may be disclosed. Only one looks like conviction rather than a brief administrative stop on the way to somewhere else.
The trouble is that disclosure systems were not designed to make this question easy. Regulators want prompt reporting of transactions by persons discharging managerial responsibilities and closely associated persons. They do not promise a clean, lot-level ledger that lets an analyst pair every buy with the eventual sell. Anyone trying to estimate insider holding periods is therefore doing forensic accounting with one hand tied behind the back.
Still, the exercise is worth doing. A matched buy-to-sell distribution tells us something that raw filing counts cannot. It helps distinguish genuine risk-taking from compensation churn, tax planning, option exercises, and the endless bureaucratic theatre of modern executive pay. It also gives a cleaner basis for a quantitative test: do stocks where insiders hold longer outperform those where insiders reverse quickly?
Because no live article-specific dataset was supplied here, the right approach is methodological rather than theatrical. No fabricated medians, no suspiciously precise histograms. Just the structure of the problem, the frictions in the data, and the framework you would use on a 162,000-filing archive to produce a defensible answer.
The standard insider factor starts with purchases for a reason. Academic evidence has long found that insider purchases, especially open-market buys, are more informative than insider sales. Sales are noisy. Executives sell to diversify, pay taxes, fund school fees, settle divorce agreements, and satisfy the sort of wealth-management memos that should never have escaped the private bank.
That is all true, and still incomplete.
A buy says the insider was willing to increase exposure at a point in time. A later sell tells you how durable that willingness was. If the insider exits quickly, several interpretations are possible:
Only the first interpretation is directly bearish. The others are merely inconvenient. Quant work lives among such inconveniences.
A holding-period distribution helps because it aggregates across those case-by-case ambiguities. If, after careful filtering, a market shows that matched insider buys are commonly followed by sales within a short window, then the average informational content of the initial buy may be overstated. If instead insiders tend to sit on purchased shares for long periods, the buy looks more like a durable commitment.
The classic papers on insider trading profitability focus on abnormal returns after insider purchases and, to a lesser extent, sales. They generally do not reconstruct exact economic holding periods for each insider lot because the raw disclosures rarely support that cleanly. But they do establish two useful priors.
First, open-market purchases by insiders tend to be informative. Second, insider sales are much less informative on average because they are contaminated by routine motives. Those two facts imply that a matched buy-to-sell analysis should be especially careful about transaction classification. If you mix open-market buys with stock-award disposals, you are not measuring conviction. You are measuring payroll.
For portfolio construction, the question is not philosophical. It is operational:
Those are testable questions, but only after the matching problem is handled properly.
Matching a buy to a later sell sounds easy until you look at real filings. Names vary. Roles change. Transaction codes differ across markets. Holdings may be reported before and after the trade, or not at all. Some insiders transact through related entities. Others file corrections. The result is a puzzle with too many pieces and no picture on the box.
Under the EU Market Abuse Regulation, persons discharging managerial responsibilities, and persons closely associated with them, must notify transactions in the issuer’s shares or related instruments once a threshold is crossed, with notification due promptly and no later than three business days after the transaction. In France, the AMF is the relevant regulator, but the rule is European in architecture. Good for timeliness, less good for lot accounting.
In the United States, Section 16 filings provide rich transaction-level disclosures for directors, officers, and beneficial owners above the threshold. Form 4 can be very useful, particularly because transaction codes distinguish open-market trades, option exercises, gifts, and other events. But even there, reconstructing exact holding periods can be messy when insiders have multiple acquisition sources and multiple accounts.
A serious matching pipeline needs entity resolution. “Jean Dupont”, “M. Jean Dupont”, and “Dupont Jean” should not become three insiders. Nor should a spouse-controlled vehicle be silently merged into the executive’s personal account unless the legal reporting framework supports that linkage.
At minimum, a matching key should include:
Even then, ambiguity remains. The analyst’s job is to reduce it, not pretend it has vanished.
Suppose an insider buys 10,000 shares in March and 5,000 more in June, then sells 8,000 in September. Which lot did they sell? Without explicit lot identification, you need an allocation rule. The common choices are FIFO, LIFO, and proportional matching.
FIFO is often the practical default because it is simple and stable. It also tends to align with a conservative notion of “how long did this acquired exposure remain outstanding?” But it is still an assumption. If the insider actually sold the June lot first, your estimated holding period is wrong.
That does not make the exercise useless. It means the article, and the backtest, should say so in plain English rather than hiding behind decimal places.
If you want a publishable distribution of insider holding periods, the methodology should be stricter than the average market-data blog and less mystical than a consultant’s appendix. The goal is not perfect truth. It is a transparent approximation that survives contact with sceptical readers.
Start with the cleanest signal set:
This will reduce sample size. Good. Better a smaller sample than a larger one full of compensation plumbing.
For each insider and issuer, create a chronological ledger of eligible acquisitions and eligible disposals. Where filings report post-transaction holdings, use them as a consistency check. If a disposal would imply negative eligible inventory under your rules, that is a warning sign that the sell likely came from another source.
At this stage, you need a decision rule:
The second option is stricter and usually better for a conviction study.
Apply a lot-matching rule, usually FIFO, within each insider-issuer-security ledger. This produces estimated round trips and holding periods in calendar days or trading days.
Useful outputs include:
A survival-analysis frame is often cleaner than a simple histogram because many buys remain unsold at the end of the sample. Those are right-censored observations, not missing data.
An insider who buys 50,000 shares and later sells 5,000 has not “exited” in the ordinary sense. Partial disposal may still matter, but it should not be conflated with complete reversal.
A robust study therefore reports at least two measures:
The first captures when insiders begin reducing exposure. The second captures how long the commitment lasted in aggregate. They are not the same thing.
Not every sale following a buy should be read as a change of mind. Some are likely routine. Clues include:
If the filings or market permit, flag these separately. The distribution of “all matched sells” is interesting. The distribution of “discretionary open-market reversals” is more interesting.
Without the live matched dataset, one should not manufacture medians. But one can state, with some confidence, what patterns usually emerge when the matching is done carefully.
Insider holding periods are unlikely to cluster neatly around one horizon. More likely, the distribution is mixed:
That shape matters because the average will be a poor summary. Median and percentile bands are more informative. So is the survival curve, which answers a practical question directly: what fraction of insider-purchased shares remain unsold after six months?
Chief executives and founders often have different incentives from non-executive directors. CFOs may trade under tighter internal scrutiny. Chairs may have legacy holdings that muddy the distinction between fresh conviction and old inventory maintenance.
A sensible expectation is that role-based distributions differ materially. If they do not, either the market is unusually homogeneous or the classification is too crude.
Disclosure detail, compensation design, and ownership concentration differ across jurisdictions. A U.S. Form 4 dataset may permit cleaner transaction-type filtering than a pan-European feed assembled from issuer notices. France, operating under MAR with AMF oversight, sits within that European framework. The rulebook gives timely disclosure, but practical data quality depends heavily on how issuers and data vendors standardise notices.
| Market | Regulator | Rule | Deadline | Notes |
|---|---|---|---|---|
| FR | AMF | MAR Art 19 | T+3 | EU-style PDMR disclosure. Useful for timeliness, but transaction standardisation can vary across issuer notices and data vendors. |
| US | SEC | Section 16, Form 4 | Generally T+2 | Rich transaction coding, often better for separating open-market trades from grants, exercises, and gifts. |
In the U.S., Section 16(b) creates the famous short-swing profit rule for matching purchases and sales within six months by insiders subject to the statute. That legal window is often treated as a natural benchmark. It is useful, but it should not become a mental prison. An insider can hold for seven months and still be economically short-term. Another can trim within four months for reasons that say nothing about valuation.
Use six months as a reference point, not a verdict.
A distribution is nice. A signal is better. The natural next step is to test whether insider buys with longer realised or expected holding periods produce stronger stock returns than buys that are reversed quickly.
There are two different research questions here.
Ex post: When insiders eventually sell quickly, did the original buy have weaker subsequent alpha than buys they held for longer?
That is useful for understanding the economics of the signal, but it is not directly tradable because you only learn the holding period later.
Ex ante: Can we predict, at the time of the buy, whether this insider is likely to hold or reverse quickly?
That is the tradable version. Candidate predictors include:
The ex post result tells you whether the distinction matters. The ex ante model tells you whether you can use it.
Many insider strategies keep a buy signal active for a fixed period, say three or six months. A matched same-person sell suggests a more nuanced rule: cancel or downweight the signal when the same insider disposes of the purchased exposure within a short window.
This is especially appealing because it is intuitive to users. If the insider has already begun reversing the trade, the market should probably stop treating the original buy as fresh conviction.
The exact cancellation rule is empirical. Full cancellation after any matched sell may be too harsh. Partial decay, scaled by the fraction of the acquired quantity sold, is more reasonable.
Most insider research in practice uses event studies or simple panel regressions. A survival model offers a better fit for the holding-period question. It allows you to estimate the hazard that an insider sells after a buy, conditional on covariates, while handling right-censoring properly.
That opens the door to a cleaner tradable score: the lower the predicted near-term sell hazard, the more durable the insider conviction appears. One can then test whether low-hazard buys outperform high-hazard buys.
No romance required. Just statistics doing the job they were hired for.
This is the section where the article earns its keep. Insider datasets can produce seductive nonsense with industrial efficiency.
The biggest trap is mixing discretionary trades with compensation-related events. A sell after a vesting event is often tax mechanics wearing the costume of information. Unless transaction codes are reliable and consistently mapped, your holding-period distribution will mostly describe executive payroll administration.
Transactions by spouses, family trusts, and controlled entities may be legally associated with the insider but economically distinct in timing and motive. Whether to merge them depends on the market, the filing structure, and the study objective. There is no universal right answer. There is only a universal obligation to state the choice.
Stock splits, mergers, tender offers, and scrip dividends can distort quantities and apparent holding periods. A proper ledger adjusts for these events or excludes affected observations where adjustment is unreliable.
Amended forms and corrected notices can create phantom round trips if ingested naively. The pipeline needs version control, deduplication, and a preference rule for corrected records.
If you require both a buy and a later sell to observe a holding period, you are selecting on completed round trips. That can bias the sample toward shorter durations and more active traders. The remedy is to include unsold buys as censored observations in the main analysis.
Corporate insiders include officers, directors, and large shareholders, and they must report most of their transactions in the company’s securities to the SEC within two business days.
If this article were backed by the live 162,000-filing query results, the finished piece would not stop at one headline number. It would show a compact dashboard of the distribution and the assumptions behind it.
At a minimum:
A timeline of annual matched observations is also useful, because changes in filing quality or market conditions can alter what the sample represents.
One particularly useful validation is to compare the inferred inventory path against disclosed post-transaction holdings where available. If your matching logic repeatedly implies inventories inconsistent with reported holdings, the classification rules need work.
This sounds obvious because it is obvious. Many pipelines still skip it.
Once the distribution is built, the next empirical step is straightforward:
If short-holding buys underperform, the practical implication is immediate. The market should treat insider buys as conditionally informative, not uniformly bullish.
The headline “insiders are buying” is the sort of sentence that survives because it is easy to print and difficult to falsify in real time. But the more useful sentence is harder: “these insiders bought, and historically people like them tend not to reverse quickly.” That is a better signal because it says something about persistence, not just initiation.
A proper same-person buy-to-sell distribution will not make insider analysis simple. It will make it less naive. It will also expose how much of the apparent signal is really compensation noise, filing artefact, or the executive equivalent of buying a gym membership in January.
The concrete next step is to run the matching ledger on the 162,000 filings with strict transaction filters, FIFO and sensitivity checks, and survival analysis for censored buys. The open question is the one worth paying for: does predicted insider holding duration improve the return profile of insider-buy strategies enough to survive costs and common factor controls?
Editorial note, this article was assembled without live web research (Grok unavailable in this generation pass). Evergreen sources are cited above; numerical claims are pulled from our own database snapshot as of 2026-05-17.
Our analysis of insider holding periods reveals significant insights into investor behavior and market dynamics. With a global aggregate of 220,691 priced trades across 17 markets, the mean 90-day return stands at -1.70%, accompanied by a win rate of 41.82% and an annualized Sharpe ratio of -0.08. This data underscores the challenges investors face when relying on insider activity as a predictor of stock performance. Notably, the broader universe of 227,594 BUY filings mirrors these figures, reinforcing the notion that insider trading does not consistently translate into positive returns within the specified time frame. These statistics highlight the importance of considering holding periods when evaluating the potential impact of insider transactions on investment decisions. However, it is essential to acknowledge the limitations of our sample size and the 90-day horizon in drawing definitive conclusions.
Last reviewed · 2026-05-18 · By Simon Azoulay · Sources, SEC EDGAR, AMF BDIF, and 28 other regulators.
ING trades near €28.5 as its €1 billion buyback continues and European banks stay bid. Here is the macro backdrop, peer ...
ABC Arbitrage’s July 13 board sale lands as volatility stays firm, rates stay sticky, and the Paris quant name trades ah...
Zoom CFO Michelle Chang sold EUR 649,382 under a 10b5-1 plan while video conferencing stays tied to hybrid work, AI feat...
Credo CFO Daniel W. Fleming sold EUR 1.63m in July filings as CRDO slipped 8.1%. Here is how the sales fit the AI networ...
Maxime Seche bought EUR 231,284 of Séché Environnement stock on July 13. Read the cluster against a steady ECB backdrop ...
Glaukos CFO Alex R. Thurman sold 10,000 shares after a 45%+ rally, with iDose TR and July 29 earnings framing the next t...
