Family and Holding Edge: What 595,000 Filings Teach Us

The Declarant Distribution

Across 595,000 declarations in the full dataset, the breakdown by related-insider kind:

Controlled entities are by far the largest category at 44,481 records. Trusts follow at 1,701. Direct family members (spouse, child, parent) remain very thin in the current dataset (79 rows total).

Why Related Entities Can Signal Differently

When an executive buys shares directly, their name is exposed. When they use a controlled holding or family trust, the mechanism differs:

• The decision is often more deliberate: creating or using a legal structure implies explicit intent.
• Traceability is maintained (AMF, ESMA, Autoriteit FM, and equivalent disclosure regimes require it), but immediate public visibility is reduced.
• Historically, declarations via holdings have been less explored in academic literature, potentially creating an information inefficiency.

V14e Numbers: Calibrated Multipliers

V14e applies differentiated multipliers to each declaration's score based on the declarant kind:

• Controlled entity: multiplier +0.10
• Trust: multiplier +0.10
• Family (spouse, child, parent): multiplier +0.20

These values were calibrated on the BUY-only 60k cohort (Sharpe 1.31 on that cohort), then validated on the EU_strict 38k universe (Sharpe 0.53). Stability of Sharpe across two independent cohorts is a non-trivial OOS passage.

Important: most of V14e's lift on EU_strict comes from the entity multiplier (controlled 44k rows, trust 1,701 rows), not the family multiplier. The 79 family declarations are too sparse to isolate their contribution statistically. The family multiplier is kept for future robustness but its real weight will be recalibrated when the family cohort exceeds 193 declarations.

What the Kind Breakdown Shows

On the EU_strict universe (8 European markets), declarations via controlled entities show T+90 returns systematically 1 to 3 percentage points higher than direct insider declarations, all else equal (same role, same mcap bucket, same week).

The signal is particularly strong on:

• European mid-cap names (EUR 300M -- 2B market cap).
• CFO and BOARD insiders acting through a holding company.
• Clusters combining direct purchase + controlled entity purchase the same week.

Limits and Honesty

Two important caveats:

1. Family cohort still thin. With 79 family declarations (spouse + child + parent) across 595k, any conclusion about "family" edge sensu stricto is anecdotal. We publish the multiplier to be transparent about the method, not to assert a statistically valid edge.

2. Completeness bias. Not all markets have the same related-entity disclosure requirements. In France (AMF), transparency is strong. In some Nordic markets, family trust declarations are less exhaustive. The dataset likely under-represents family structures in those jurisdictions.

Practical Impact on Scoring

In the Sigma interface, a signal displays "via controlled holding" or "via trust" when the kind is available. These declarations receive a slightly higher score than an equivalent direct declaration. The effect is marginal on the absolute score (0.10 pts on a 0-10 scale) but can shift a declaration's position around the top-14 selection threshold.

Note: the results above are validated on the EU_strict universe (8 European markets), after exclusion of XNAS following audit 101 (signal inversion on the US tape).