Understanding the Impact of Large Insider Trades on Scoring Models

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.

One filing can swamp a model for reasons that have nothing to do with information.

Insider models love size until size starts lying.

A simple scoring system usually treats a larger purchase or sale as a stronger signal. That works reasonably well in the middle of the distribution. A chief financial officer buying €200,000 of stock after a drawdown is not the same as buying €20,000. The problem arrives when the right tail turns grotesque. Founder placements, secondary offerings, stake monetisations, and cross-holding reorganisations can produce reported trades in the hundreds of millions or billions. Feed those notionals straight into a ranking model and the result is less “signal extraction” than “tribute to arithmetic”.

This is why we cap trades above €1 billion in a liquidity-adjusted score. Not because those trades do not matter, and not because the number is sacred, but because beyond that point the economics of execution, the motives of the seller, and the market’s capacity to absorb stock become qualitatively different. A model that pretends otherwise is not sophisticated. It is merely obedient.

The rest of this piece sets out the rationale. First, why very large trades are a special case. Second, why liquidity matters more than notional at the top end. Third, how the cap can be implemented without throwing away useful information. Fourth, what the cap does not solve, because every respectable methodology note needs one mildly disappointing section.

When size stops meaning conviction

The naive model and its bad habits

Most insider scoring systems begin with an intuitive premise: bigger insider buys are better, bigger insider sells are worse. Sometimes the variable is raw notional, sometimes it is transformed with a logarithm, sometimes it is scaled by salary or market capitalisation. The common instinct is sensible. Insiders have finite wealth, finite risk tolerance, and finite willingness to transact in their own company’s stock. Size should tell us something.

But size is only informative if the transaction is discretionary enough to reflect a view. That assumption weakens as trades get larger.

At modest scales, an executive deciding to buy shares in the open market is usually making a choice that can plausibly encode information or at least confidence. At very large scales, especially on the sell side, the transaction often reflects a different set of forces:

• portfolio diversification after years of concentrated wealth,
• tax liabilities tied to vesting or estate planning,
• pledged-share management,
• lock-up expiry,
• secondary offerings underwritten and marketed to institutions,
• family holding-company restructurings,
• negotiated block placements,
• merger-related exchanges or internal transfers.

These are not obscure edge cases. They are routine features of founder-heavy listed markets.

Founder trades are not just bigger, they are different

A founder holding 20 percent of a company can sell €1.5 billion of stock and still remain economically and politically dominant. A non-founder executive cannot. Treating both events as points on the same linear scale is an invitation to category error.

The issue is not merely that founders transact in larger size. It is that their trades often occur under constraints and motives that differ from those of hired managers. Founders may be monetising decades of accumulated equity, funding philanthropy, settling family ownership matters, or broadening the shareholder base. None of these motives implies bullishness or bearishness about next quarter’s earnings. Yet a raw-score model can easily rank such a sale as the most “negative” event in the entire universe for months.

That is how a model ends up shouting about a block sale that every investment bank on the street already knew was coming.

Regulation captures the filing, not the motive

Insider disclosure regimes are designed to get transactions into the public domain, not to classify their economic meaning with editorial finesse. In Europe, the Market Abuse Regulation requires persons discharging managerial responsibilities and closely associated persons to notify transactions, subject to thresholds and deadlines. In the United States, Section 16 reporting captures many officer, director, and beneficial-owner transactions. These regimes are indispensable, but they are not motive detectors.

A filing can tell you that stock changed hands, in what amount, on what date, and under what broad transaction code. It often cannot tell you whether the sale was opportunistic, forced, pre-arranged, tax-driven, collateral-related, or simply the least dramatic way to move a very large ownership block from one pocket to another.

That ambiguity becomes more dangerous as notional rises. The larger the trade, the more likely it is to involve a process, not a moment of conviction.

Liquidity is the adult in the room

Notional is easy to count, hard to interpret

A €50 million sale in a mega-cap with deep daily turnover is one thing. A €50 million sale in a mid-cap with thin liquidity is another. The same is true, only more so, at €500 million or €1 billion.

What matters for market impact is not just the euro amount. It is the relationship between the trade and the stock’s liquidity. Average daily value traded, free float, order-book depth, index ownership, short interest, and the presence of block-trade intermediaries all shape how much informational content survives the mechanics of execution.

A billion-euro transaction almost never behaves like a scaled-up €10 million trade. It is arranged differently, placed differently, hedged differently, and interpreted differently by the market. Banks wall-cross investors. Discounts are negotiated. Allocation is curated. The event can be as much capital-markets logistics as insider sentiment.

This is why we describe the cap as liquidity-adjusted in spirit, even if the operational rule is a hard ceiling at €1 billion. The cap acknowledges that above a certain size, raw notional is no longer proportional to informational intensity.

Market impact is non-linear

Academic market microstructure has spent decades making one point with great persistence: trading costs and price impact are not linear in size. Large orders are split, intermediated, delayed, crossed, discounted, or warehoused precisely because markets are not infinitely deep. If they were, block desks would be much quieter places.

For insider scoring, the implication is straightforward. Once a trade enters the realm of exceptional block activity, the observed notional is partly a statement about ownership concentration and market plumbing, not just private information. A model that lets score contribution rise mechanically with every extra euro is therefore embedding a false linearity.

There is a practical side too. Cross-sectional ranking systems are fragile to outliers. One or two giant founder sales can compress the score distribution for everything else, making ordinary but genuinely informative transactions look trivial by comparison. You do not need a philosophical objection to cap these events. You only need a functioning ranking table.

Relative measures help, but they do not fully solve it

A common response is to scale trade size by market capitalisation, free float, or average daily value traded. That is good practice, but it does not eliminate the problem.

Suppose a founder sells 2 percent of a €100 billion company, worth €2 billion, in a carefully marketed block. Relative to market cap, the sale may not look catastrophic. Relative to daily liquidity, it may still be enormous. Relative to the founder’s remaining stake, it may be modest. Relative to the information content of the event, it may be close to nil. Which denominator should win?

The answer is not to worship one denominator. It is to accept that very large insider trades are multi-dimensional events. A hard cap is useful because it stops any single size-based transformation from pretending to know too much.

Why €1 billion, and not €500 million or €2 billion

Thresholds are judgement calls with paperwork

There is no regulator-issued stone tablet declaring that insider-trade signal quality expires at €1 billion. Any threshold is a methodological choice. The proper question is whether the choice is defensible, stable, and transparent.

A cap at €1 billion has three advantages.

First, it is high enough that ordinary executive trading is unaffected. This matters. A cap should target the pathological tail, not flatten the useful middle of the distribution.

Second, it is low enough to catch the class of transactions most likely to be block placements, founder monetisations, or ownership-structure events rather than straightforward open-market conviction trades.

Third, it is legible. Methodologies that require a whiteboard and a support ticket to explain are rarely robust in production.

One can imagine alternative designs. A percentile cap by market, a cap based on multiples of average daily value traded, or a piecewise function that reduces marginal score contribution as notional rises. All are respectable. We use the €1 billion cap because it is easy to audit, hard to game, and aimed squarely at the part of the distribution where raw notional is least trustworthy.

The cap is a winsorisation, not a deletion

This distinction matters. We do not discard trades above €1 billion. We winsorise their notional contribution to the score. In plain English, the event stays in the model, but its size-based component is treated as if it were €1 billion rather than whatever larger amount was reported.

That preserves several things at once:

• the fact that a transaction occurred,
• the direction of the trade,
• the identity and role of the insider,
• any timing signal around earnings, guidance, or corporate events,
• and any supplementary flags attached to the filing.

What it removes is the absurdity whereby a €4 billion founder sale is treated as four times more informative than a €1 billion founder sale merely because the arithmetic says so.

This is not statistical cowardice. It is recognition that the fourth billion often tells you much less than the first.

Why not just log-transform the notional

Log transforms are useful. They compress the range and reduce outlier dominance. But they do not fully address the conceptual problem.

A log-transformed €3 billion sale still outranks a €1 billion sale by a meaningful amount, even if both are likely to belong to the same family of structurally exceptional events. More importantly, a transform is a mathematical convenience, not an economic argument. If the issue is that trades above a certain size are generated by different mechanisms, then a threshold-based treatment has a cleaner interpretation.

The dry version is that the data-generating process changes in the tail. The less dry version is that once a founder is selling by the billion, you are no longer measuring enthusiasm with a ruler.

How to implement the cap without blinding yourself

Keep the event, cap the score, add context flags

The best implementation is modular. The transaction record remains complete. The score applies the cap only to the notional component. Then additional context variables are layered on top.

Useful flags include:

• founder or controlling-shareholder status, because motive sets differ materially;
• percentage change in beneficial ownership, because selling €1 billion while reducing ownership by 0.8 percentage points is different from exiting half a stake;
• trade type, such as open-market sale, private transfer, option exercise, tax withholding, or block placement where disclosed;
• pre-arranged plan indicators, for example Rule 10b5-1 in the US where relevant;
• relative liquidity metrics, such as trade size versus average daily value traded;
• cluster behaviour, whether multiple insiders are transacting in the same direction;
• timing windows, especially around earnings or major corporate announcements.

This structure solves a common false choice. You do not have to decide between “size matters” and “context matters”. Size matters until it starts impersonating context.

Ownership change is often more informative than cash amount

For founders in particular, percentage ownership change can be a better behavioural measure than euro value. A €1.2 billion sale sounds dramatic. A reduction from 24.7 percent to 24.1 percent sounds rather less like a vote of no confidence. Sometimes that calmer framing is closer to the truth.

This is not to say ownership change should replace notional. A tiny percentage sale in a very large company can still matter, especially if it breaks a pattern of previous holding behaviour. But ownership change is a useful antidote to the theatrical tendencies of large euro amounts.

If your model can only afford one extra variable alongside capped notional, ownership change is a strong candidate. It is simple, economically meaningful, and often available from the filing or related beneficial-ownership disclosures.

Distinguish sells from buys, because the asymmetry is real

The case for capping giant sales is stronger than for giant buys. Large insider buys are rarer, harder to explain away with liquidity management, and more likely to represent deliberate risk-taking. If a founder or executive is putting fresh capital to work in the open market at very large size, that is usually worth attention.

Even so, liquidity still matters. A huge buy can be connected to support operations, control contests, or structured transactions. The broader point is that scoring systems should not assume perfect symmetry between buys and sells. The motives differ, the constraints differ, and the base rates differ.

A sensible methodology can therefore cap both sides for consistency, while allowing the directional component and contextual modifiers to reflect the asymmetry.

What the cap does not fix

It cannot read around legal or tax structures

Some filings are maddeningly terse. They report a transaction that is economically linked to another leg elsewhere, or to a derivative, trust, or family vehicle not fully visible in the same document. A cap cannot solve disclosure opacity. It can only reduce the damage done by taking a giant headline number too literally.

This is especially relevant in cross-border cases, where filing formats, beneficial-ownership conventions, and transaction coding differ. Harmonising those records is hard enough without letting a single outlier notional dictate the score.

It does not replace event-level reporting

A quantitative score is a ranking device, not a substitute for reading. If a founder sells €2.5 billion in a block, that event may deserve coverage regardless of how the score is capped. The market may care about the discount, the investor mix, the lock-up, or the signalling effect on future supply. None of that disappears because the model has decided not to treat €2.5 billion as 2.5 times more informative than €1 billion.

This is the correct division of labour. The model keeps the ranking sane. The analyst keeps the story honest.

It will occasionally understate a genuinely important trade

Every cap creates false negatives at the margin. There will be cases where a transaction above €1 billion is genuinely extraordinary in a way the score should recognise more strongly. That is the price of robustness.

The alternative, however, is not perfection. It is a system in which a handful of founder block sales repeatedly hijack the top and bottom deciles, degrading the usefulness of the ranking for everyone else. In production, robust and slightly conservative usually beats theoretically pure and routinely derailed.

The regulatory backdrop, briefly and without romance

Disclosure rules give us the raw material

In Europe, the Market Abuse Regulation sets the framework for managers’ transactions, including the requirement to notify transactions conducted by persons discharging managerial responsibilities and persons closely associated with them. National regulators such as the AMF in France implement and publish within that framework. In the US, the SEC’s Section 16 regime, including Forms 3, 4, and 5, provides analogous raw material for officers, directors, and large beneficial owners.

These systems are invaluable because they create a public record. They are less helpful when one wants to distinguish a founder liquidity event from a discretionary informational trade. That is not a criticism. It is simply a reminder that a filing regime is a disclosure mechanism, not a scoring methodology.

The filing deadline is not the same as price discovery

Another reason to avoid over-weighting giant trades is that market awareness often precedes the filing. Large blocks are frequently intermediated by banks and placed with institutions through a process that leaks, by design or by gravity, into market expectations. By the time the formal insider filing appears, some or much of the information may already be in the price.

Again, this is not universal. But it is common enough that a model should be cautious about equating reported notional with fresh surprise, especially for very large founder sales.

A better way to think about extreme insider trades

They are events first, observations second

The cleanest mental model is to treat very large insider trades as event objects with several attributes, not as giant points on a single size axis. The attributes include who traded, how much ownership changed, how liquid the stock is, whether the trade was pre-arranged, whether it was marketed as a block, and what else was happening around the company.

Once you think this way, the €1 billion cap looks less like an arbitrary brake and more like a guardrail against category mistakes. It tells the score, politely but firmly, that there is a point beyond which “more euros” stops being the main story.

Models should be humble where market structure is loud

Quant finance has a recurring weakness for variables that are easy to measure and hard to interpret. Reported notional is one of them. It is clean, available, and temptingly numeric. It is also a magnet for spurious precision in the tail.

A liquidity-adjusted cap is an admission of humility. It says the model knows enough to use size, but not so much that it will extrapolate linearly through founder blocks, secondary placements, and ownership restructurings as if they were merely enthusiastic versions of ordinary trades.

That is not a glamorous insight. It is, however, the kind that keeps a production signal from becoming a museum of avoidable errors.

The practical payoff is simple: keep the giant trade in view, stop it from hijacking the score, and force the interesting questions back into context, where they belong. The next step is empirical rather than philosophical. Test whether capped-notional rankings, combined with ownership-change and liquidity flags, improve stability and forward signal quality versus raw-size models across founder-heavy markets. If they do, the cap stays. If not, the threshold, or the whole idea, deserves another audit. That is the pleasant thing about methodology. It can be argued with using data, which is more than can be said for most billion-euro founder sales.

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.

Insiders Trades backtest cohort

Our analysis of the backtest figures reveals critical insights into the implications of insider trade size on scoring models. 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 annualised Sharpe ratio of -0.08. This data mirrors the broader universe of 227,594 BUY filings, which also reflects a mean 90-day return of -1.70% and an identical Sharpe ratio of -0.08. These consistent figures underscore the potential pitfalls of relying solely on trade size as a predictive metric, as they highlight a pervasive trend of underperformance across various markets. It is essential to consider these statistics within the context of sample size and time horizon, as they may not fully capture the nuances of insider trading dynamics.

Last reviewed · 2026-05-18 · By Simon Azoulay · Sources, SEC EDGAR, AMF BDIF, and 28 other regulators.