InsidersTradesSigma

Understanding Swiss Anonymized Insider Disclosures and Signals, Sigma Journal

Analyzing Swiss Insider Disclosures Without Names

Can anonymized insider disclosures still provide actionable insights? Despite the absence of names, Swiss filings reveal enough structure to analyze insider behavior effectively.

By Sigma Research·May 17, 2026·14 min · 3,139 words

One missing name can hide a person, but it does not hide a pattern.

Anonymized insiders in Switzerland, and the stubborn survival of signal

Switzerland is awkward for anyone who likes neat insider datasets. The country offers a serious market, sophisticated issuers and a disclosure regime that is real enough to matter, yet often sparse enough to irritate. Names are frequently absent from public-facing insider transaction disclosures. For journalists, that is an aesthetic problem. For investors, it is a modelling problem.

The interesting question is not whether Swiss disclosures are inferior to the US model of naming names. They are, for many use cases. The better question is whether anonymized filings still contain enough information to extract a tradable or at least analytically useful signal. The answer is yes, with caveats sharp enough to keep one honest.

What survives anonymization is not identity, but structure. A filing can still tell you the role of the person discharging managerial responsibilities, or a closely associated party, the direction of the trade, the instrument, the date, the issuer and often the rough context. That is enough to build role-level clusters and intensity scores. It is not enough to write a good gossip column. Sigma Journal can live with that.

Act on this

SignalsToday's recommendations ScreenerInsider-signal screener UniverseAll covered companies

What the Swiss regime gives you, and what it withholds

Swiss insider disclosure sits in a different tradition from the SEC's Forms 3, 4 and 5, or the EU's MAR Article 19 notifications as commonly disseminated. The legal and market architecture matters because it determines the raw material available to researchers.

In Switzerland, issuers listed on SIX Swiss Exchange are subject to rules on the disclosure of management transactions. The public disclosure is designed to inform the market about dealings by members of the board of directors and executive management. The framework has long been administered through exchange regulation rather than a direct copy of US-style named insider reporting. The result is a public feed that can be timely and useful, but less person-specific than many quants would prefer.

Privacy by design, signal by residue

The privacy-preserving element is not an accident. Swiss financial regulation has historically tolerated a somewhat more discreet public posture around individual identities than the Anglo-American model. Public notices of management transactions often identify the issuer, the date, the transaction type, the instrument and the function of the person involved, while omitting the individual's name.

That omission removes a large class of features:

no direct insider fixed effects,
no clean separation of one habitual seller from another,
no easy mapping of one executive across issuers or over a career,
no straightforward linkage to compensation disclosures or prior enforcement history.

What remains is still richer than it first appears. Role labels such as chief executive officer, chief financial officer, board chair, non-executive director, executive committee member or closely associated person are not decorative. They encode hierarchy, information access and likely motive. A CFO sale two days after annual results is not the same species of event as a non-executive director buying into a drawdown six weeks before a capital markets day.

The minimum viable feature set

Even in anonymized regimes, a surprisingly robust feature set can usually be assembled from the filing itself and issuer metadata:

Role or function of the transacting person.
Trade direction, buy, sell, exercise, grant, conversion, pledge-related movement where disclosed.
Instrument type, common shares, options, restricted stock, derivatives.
Transaction date and disclosure date.
Issuer characteristics, market cap, sector, volatility, liquidity, ownership concentration.
Event proximity, earnings, guidance, M&A, capital raises, lock-up expiries.
Transaction size, where available directly or inferable from units and price.
Sequence information, whether the issuer has multiple filings in a short interval.

This is enough to score intensity and to cluster events into behavioural archetypes. It is not enough to know that "Mr X always sells before disappointing quarters". One must settle for "CFO-type actors at this issuer and peers exhibit a repeated pattern of post-result disposal with low subsequent alpha". Less cinematic, more useful.

Role-level data is enough to cluster insiders

The central claim of this article is simple. If names disappear but roles remain, clustering is still viable. In fact, role labels often carry more stable information than names for first-pass modelling.

Why role beats identity in early-stage models

A common mistake in insider analysis is to overfit to personalities. Named datasets tempt researchers into creating bespoke narratives around famous founders, serial sellers or unusually aggressive buyers. Those stories can be true. They can also be expensive forms of overfitting.

Role-level modelling starts with a cleaner prior. A board chair, a CFO and a non-executive director face different incentives, possess different information and transact for different reasons. Those distinctions are economically grounded.

A practical clustering framework might begin with three dimensions:

Information access, high for CEO/CFO, moderate for executive committee, lower for non-executive directors.
Discretionary capacity, high for open-market buys and sells, lower for automatic plan-related or compensation-linked events.
Motive ambiguity, high for option exercises and tax sales, lower for outright open-market purchases.

From there, one can derive behavioural clusters.

A workable clustering taxonomy

For Swiss anonymized filings, a useful taxonomy is less granular than in the US, but still meaningful:

1. Conviction buyers

Usually open-market purchases by board members or executives, especially after price weakness or around periods of elevated uncertainty. These tend to carry the strongest positive prior in the academic literature across markets. The reason is dull and persuasive: insiders have many reasons to sell, but fewer reasons to buy with their own cash.

2. Compensation disposers

Sales or conversions linked to vesting, option exercise or tax settlement. These are common and often low-signal. If the filing language or instrument path suggests mechanical disposal, the event should be down-weighted heavily.

3. Executive liquidity sellers

Open-market sales by senior executives outside obvious compensation mechanics. These can be informative, especially when clustered across multiple roles or when unusually large relative to estimated holdings or pay.

4. Board accumulators

Repeated purchases by non-executive directors or chairs. These often matter more in smaller issuers, where boards may be closer to financing, strategic review or operational inflection points.

5. Event-driven transactors

Transactions occurring near earnings, guidance changes, capital raises or M&A. These need separate treatment because the event context can dominate the role effect.

The taxonomy does not require names. It requires consistent role parsing and disciplined context enrichment.

Features that make clustering less naive

Role alone is not enough. Role plus context is.

Useful clustering features include:

role seniority score,
buy versus sell indicator,
open-market versus non-open-market classification,
instrument complexity score,
lag between trade date and disclosure date,
count of same-issuer insider filings in the prior 30 days,
abnormal return in the 20 trading days before the transaction,
issuer size bucket,
sector bucket,
earnings-window proximity,
whether the transaction follows a major price gap.

A simple unsupervised approach, k-means on standardized features or hierarchical clustering on mixed data types, can already separate routine compensation activity from potentially informative discretionary trades. More advanced methods, such as HDBSCAN or Gaussian mixture models, may uncover sparse but interesting clusters, especially in small-cap names where insider behaviour is less homogenized.

The point is not to discover metaphysical truths. It is to stop treating every anonymized Swiss filing as equivalent. They are not.

Intensity scoring, the practical substitute for named-insider tracking

If clustering tells you what kind of event you are looking at, intensity scoring tells you how much attention it deserves.

This is where anonymized data can still earn its keep. A role-level intensity model does not need to know who the insider is. It needs to know whether the event is unusual, discretionary and plausibly informed.

Building an intensity score

A sensible intensity score can be built as a weighted combination of five components.

1. Role weight

Assign higher baseline weights to roles with greater presumed access to operational and financial information.

Illustrative ordering:

CEO
CFO
COO or equivalent senior executive
Board chair
Other executive management
Non-executive director
Closely associated person, unless linked to a known managerial role

This ranking is not sacred. In some firms, the chair is deeply involved; in others, ceremonial. The model should allow issuer-specific overrides when governance data justify them.

2. Transaction discretion weight

Open-market purchases deserve the highest positive weight. Open-market sales receive moderate weight. Option exercises, grants and vesting-related transactions receive low weight unless accompanied by a clear discretionary hold-or-sell decision.

This principle is well supported by the broader insider-trading literature. Seyhun's work, and later studies across jurisdictions, repeatedly finds stronger predictive content in purchases than in sales, precisely because sales are contaminated by liquidity, diversification and tax motives.

3. Abnormal size weight

Size matters, but only relative to context. In a named regime one might scale by insider holdings. In Switzerland, anonymity blocks that route. The workaround is to scale by issuer-level variables:

transaction value relative to issuer free-float market cap,
transaction value relative to median disclosed management transaction size for the issuer,
transaction value relative to estimated annual executive cash compensation, when public remuneration reports allow rough benchmarking.

This is imperfect. It is still better than treating CHF 200,000 and CHF 20 million as equal expressions of sentiment.

4. Temporal concentration weight

A single filing can be noise. A burst of filings is harder to ignore. If multiple role categories transact in the same direction within a short window, the score should rise. This is especially true when the cluster includes both executives and directors.

Temporal concentration can be measured as:

number of insider filings in same direction over 7, 14 and 30 days,
number of distinct role categories involved,
net buy-sell imbalance by value.

Anonymity weakens person counting, but role diversity remains observable and useful.

5. Event-context weight

The same trade can mean different things depending on timing. A director purchase after a 25 percent drawdown and before a strategy update is not the same as the same-sized purchase after a routine quarter. Likewise, executive sales immediately after lock-up expiry or compensation vesting should be discounted.

A practical event-context layer includes:

days to or from earnings,
days to or from guidance changes,
days to or from financing announcements,
recent abnormal volatility,
recent analyst estimate revisions.

This turns a static filing into a contextualized event.

Why this works despite anonymity

Named-insider models often derive alpha from repeated actor effects. Some insiders are simply better than others. Switzerland makes that harder to capture. But a large share of insider signal in the literature does not depend on personality. It depends on asymmetry of information and costly action. Role-level intensity scoring preserves both concepts.

A CFO buying stock in the open market after a sell-off is a costly action by a highly informed actor. The model does not need the CFO's surname to understand the economic content.

What Swiss anonymity costs you, quantitatively and conceptually

There is no point pretending the Swiss setup is merely a charming local variant. It imposes real losses.

You lose longitudinal actor history

In the US, one can ask whether a specific executive has a strong record of buying before positive surprises. In Switzerland, public anonymity often prevents that. The inability to estimate insider fixed effects is a material handicap.

This matters because repeated behaviour is informative. Some insiders are chronic sellers with no signal. Others buy rarely and well. Without names, one cannot easily separate them.

You lose network analysis

Named regimes allow linkage to outside board seats, family offices, venture affiliations and prior issuer relationships. Those networks can explain why a transaction occurred and whether it should be interpreted as informed or ceremonial.

Swiss anonymized data largely closes that door. One can still use public board biographies and remuneration reports to enrich issuer-level governance context, but not with the same precision.

You lose cleaner legal-event linkage

In named datasets, enforcement actions, blackout-policy breaches or amended filings can be tied to individuals. That is useful for reliability scoring. In anonymized public disclosures, reliability assessment must lean more heavily on issuer reporting quality and exchange notices.

But you do not lose the first-order economics

The first-order economics of insider trading signal remain:

informed actors know more than the market,
purchases are generally more informative than sales,
discretionary trades matter more than mechanical ones,
clusters matter more than isolated events,
context determines interpretation.

Those principles survive the removal of names. The signal degrades, but it does not vanish.

How Switzerland compares with more transparent regimes

The contrast with France and the US is instructive because it clarifies what is missing, and what is still salvageable.

Market	Regulator	Rule	Deadline	Notes
CH	SIX Exchange Regulation	Management transactions disclosure	prompt public disclosure under exchange rules	Public disclosures often identify role and transaction details, but not the individual's name.
FR	AMF	MAR Art 19	T+3 business days	Persons discharging managerial responsibilities and closely associated persons must notify transactions; public dissemination is more standardized under EU MAR.
US	SEC	Exchange Act Section 16, Form 4	T+2 business days	Named insider reporting with granular ownership and transaction coding, highly suitable for person-level modelling.

Public insider-disclosure regimes differ less in principle than in granularity. Switzerland is the clear outlier on public identity transparency.

The US advantage is obvious

SEC Form 4 is a quant's comfort blanket. It names the insider, specifies the relationship, codes the transaction, reports holdings before and after in many cases, and arrives on a tight deadline. One can construct person-level histories, issuer-level aggregates and cross-sectional factors with relative ease.

Swiss data does not permit that degree of precision. It is a different sport.

The EU model sits in the middle

Under MAR Article 19, PDMRs and closely associated persons must notify transactions, and regulators such as the AMF disseminate disclosures in a fairly structured format. The data quality varies by market and vendor, but identity transparency is generally stronger than in Switzerland.

For comparative research, this suggests a useful benchmark. If a role-level model performs reasonably in France or other MAR markets after stripping names, then one can estimate how much predictive power identity adds. That is the right experiment. Not pining for a dataset one does not have.

A research agenda for anonymized insider data

The proper response to Swiss anonymity is not despair. It is method.

Start with event studies, not grand claims

The cleanest first step is a set of event studies around clustered role-level signals:

open-market buys by senior executives,
board-only buy clusters,
mixed executive-and-board buy clusters,
large discretionary sales by finance executives,
compensation-linked disposals as a low-signal control group.

Measure abnormal returns over 1, 5, 20 and 60 trading days, adjusted for market, sector and size. If no signal survives there, more ornate modelling will not rescue the exercise.

Then test issuer-relative abnormality

Because anonymity blocks person-level baselines, issuer-relative abnormality becomes more important. Ask whether the filing is unusual for that issuer's own history:

first insider purchase in 18 months,
largest management transaction in two years,
first multi-role cluster since listing,
first board purchase after a prolonged drawdown.

These issuer-relative features are often potent because they capture local rarity, which the market may underreact to.

Use text and taxonomy carefully

Swiss issuer disclosures may include enough wording to separate routine events from discretionary ones. A restrained natural-language layer can classify descriptions into categories such as grant, exercise, sale, conversion or transfer. This is not glamorous machine learning. It is useful housekeeping.

The rule here is simple: taxonomy before transformers. Most insider datasets need better labels, not a larger neural network.

Cross-market transfer learning is plausible

A practical idea is to train a role-level signal model on more transparent markets, then apply it to Swiss data after removing identity-dependent features. For example, one could train on French MAR data or US data collapsed to role-event level, then test whether the learned weights on role, discretion, size and timing generalize to Switzerland.

If they do, that would strengthen the case that first-order insider economics travel across disclosure regimes. If they do not, the failure would itself be informative. Swiss governance and ownership structures may produce genuinely different insider behaviour.

The likely edge is in small and mid caps

Large-cap Swiss issuers are heavily covered. Their insider disclosures matter, but less dramatically. The more interesting hunting ground is likely small and mid caps, where analyst coverage is thinner, ownership can be concentrated and management actions may carry more incremental information.

That does not mean every small-cap board purchase is gold. It means the market's prior is weaker, and role-level insider events may move the posterior more.

Practical guidance for investors and data teams

This is the part where one either becomes useful or decorative.

For discretionary investors

Treat Swiss anonymized insider disclosures as a ranking input, not a verdict. A sensible checklist:

Is it a buy or a sell?
Is it clearly discretionary?
Which role is involved?
Is the size unusual for the issuer?
Are there multiple filings in the same direction?
What corporate event sits nearby?
Is the stock in a neglected part of the market?

If four or five of those line up, the event deserves work. If not, it belongs in the archive.

For systematic teams

Build the pipeline around robust parsing and context enrichment. Specifically:

normalize role labels,
classify transaction mechanics conservatively,
attach issuer fundamentals and event calendars,
compute issuer-relative rarity features,
score temporal clustering,
maintain a "mechanical transaction" suppression layer.

Do not waste the first six months trying to infer names. That is a poor use of time and a good way to create legal and methodological trouble. The market does not pay extra for detective fiction.

For compliance and governance researchers

Swiss anonymized data is also useful beyond alpha. It can support studies of governance responsiveness, board confidence during stress and the interaction between disclosure design and market efficiency. A regime that protects individual privacy while preserving some market transparency is not analytically barren. It simply asks different questions.

The honest verdict

Can we still extract signal from anonymized Swiss insider data? Yes. Not the full signal, and certainly not the easiest version of it, but enough to justify disciplined analysis.

The key is to stop mourning the missing names and start exploiting the surviving structure. Role-level data, transaction mechanics, issuer context and temporal concentration are sufficient to cluster events into meaningful behavioural buckets and to build intensity scores that rank likely informational value. The resulting models will be blunter than their US equivalents. They will also be less prone to personality worship, which is not always a loss.

The next concrete step is straightforward: run a Swiss event study on role-weighted, issuer-relative abnormal insider buys versus mechanically filtered disposals, split by market-cap bucket. If the spread survives transaction costs and basic risk controls, anonymity has not killed the signal. It has merely forced it to wear a different suit.