🎯La Conviction des Initiés comme Stratégie : Pourquoi la Qualité l'Emporte sur la Quantité

Key Lesson: A Signal Without Quality Filter is a Trap

The most important insight about insider strategies is counterintuitive: the mere existence of an insider purchase guarantees no advantage. It depends on WHICH purchases you include and WHICH you exclude. The difference between a naive and a quality approach is vast.

In a full market cycle (2021–2026, a fair window across both bull and bear phases), the naive approach of "buy every insider buy across all capitalizations" resulted in a substantial loss against SPY. This is a disaster — and the main culprit is the micro-cap trap: in the smallest companies, one purchase distorts the picture, liquidity is fragmented, and survivorship and noise dominate.

On the other hand, a quality conviction filter — market capitalization of at least $5 billion OR a genuine cluster with a CEO purchase — ended the same period above SPY. The same raw data source, opposite result. A quality filter is the whole game.

Anatomy of a Quality Filter

A quality filter excludes everything that generates noise:

• Only P-code — no grants, options, or gifts disguised as purchases.
• Size threshold — exclusion of micro-caps where one purchase skews statistics, or a requirement for a genuine cluster + CEO purchase as a size substitute.
• Hierarchy — CEO/CFO weigh more.
• Purity — no concurrent insider sales (Sell Contamination).
• Price proximity — purchase near or below the current price, not after a meteoric rise.

The filter does not promise that every position will profit. It promises that the distribution of returns shifts towards where conviction historically has an edge over the base of pure sales.

Construction of HIGH and ELITE

QMA works with layered levels of conviction, built on the ICS score:

HIGH conviction (strong ICS) did better in backtesting than a base of companies where insiders were net sellers. That is the crux: the value lies in the gap between the two cohorts, not in an absolute return. The 2024–2026 window was strongly bullish, snapshots overlapped and market capitalisation was approximated — any one of those three caveats is enough to make the absolute figure untrustworthy.

ELITE is a narrower construction: HIGH conviction and an above-average overall platform rating — insider conviction independently confirmed by a strong fundamental-technical view.

🚨 CORRECTION NOTE (31 July 2026) — the performance figures that used to stand here are WITHDRAWN.

>

They rested on a historical scores table whose older part was shown on 15 July 2026 to have been

derived from later prices — that is, from the future. On 31 July 2026 the significance

threshold was also tightened: it has to account for every test we have ever run, not for the

variants of a single run.

>

The figures are not being replaced by smaller ones. The inputs were never valid as of the day

in question, so an honest replacement does not exist — the claim goes away entirely. Only

forward tracking since 18 February 2026 is trustworthy.

>

Added 31 July 2026: by the same logic, the last remaining figures in this article go too —

the gap between the naive and the quality approach against SPY, and the contribution of

congressional overlap. They had been kept because they came from a different source than the

withdrawn scores table. But the article announced above that it was withdrawing performance

figures and then quoted three of them two paragraphs below; that contradiction is worse than a

missing number. The direction stays as a description, the number does not.

ELITE in particular had two defects at once: the cohort was defined by a composite score from the
withdrawn table, and the sample ran to a few dozen trades. What remains is a hypothesis worth
monitoring, not a result. That is precisely why QMA runs forward tracking of ELITE and HIGH
conviction selections — to test it in real time, outside the backtest.

Confluence as a Confirmation Layer

Above conviction, additional smart money traces can be layered: congress (STOCK Act), dark pool (FINRA), unusual options activity, and 13F institutional positions. In backtesting, congressional overlap was on average mildly additive; the specific figure is not published (see the correction note above).

The key is to correctly understand the role: confluence is a confirmation tilt, not a standalone advantage. Alpha still stems from the conviction and quality filter. Confluence helps decide between two similarly strong signals, but it does not replace conviction on its own. Layering multiple weak signals on a poor foundation does not create a strong strategy.

Thinking About Risk and Position Size

When constructing a framework around insider conviction, at the level of thinking (not as a directive), these principles are often relevant:

• Asymmetry of quantity and quality — fewer positions with high ICS historically outperform many positions with low ICS. Concentration on conviction is itself a risk filter.
• Diversification across clusters and sectors — even quality signals fail individually; dispersion dampens the idiosyncratic risk of a single name.
• Horizon matching the signal — backtests work with a 90-day window; insider conviction is a medium-term, not intraday signal.
• Company size as a risk threshold — capitalization threshold is not just a yield filter, but also liquidity and robustness.

These are considerations for method construction, not directives for action. How large a position someone holds is an individual decision beyond the scope of this text.

The Limits Within Which the Methodology Was Measured

The specific results are withdrawn (see the note above). What remains are the limits of the measurement itself, and it is honest to state them:

• The backtest 2024–2026 took place in a bull market — results may differ in another regime.
• Snapshots overlap, which overstates consistency.
• Market capitalization is a proxy, not an exact historical reconstruction.
• ELITE rests on a few dozen trades — exploratory, not robust.
• The full cycle 2021–2026 showed how easily a naive approach loses money — the advantage exists only with a quality filter.

Therefore, QMA displays a live panel Win Rate — Backtesting ICS Signals and forward tracking: so users can see how the historical hypothesis behaves further.

Summary

• Quality beats quantity: the naive all-cap approach lost against SPY, the quality filter ended above the index — specific figures withdrawn.
• Quality filter = only P-code + size threshold (or genuine cluster + CEO) + hierarchy + purity + price proximity.
• HIGH conviction: the value is the gap against the net-selling base, not an absolute return.
• ELITE (HIGH + an above-average overall rating): figure withdrawn, exploratory construction.
• Confluence (congress, dark pool, options, 13F) is confirmation, not standalone alpha.
• ICS + live forward tracking serve for ongoing verification, not blind trust in backtesting.

---

Past performance is not indicative of future results. This text is educational and analytical — it is not an investment recommendation.