Context: What Makes a Trade "Scorable"?
The raw Sigma dataset contains 595,000 insider declarations. A trade is "scorable" when all conditions are met: market within the EU_strict universe (XPAR, XAMS, XWBO, XBRU, XHEL, XOSL, XSTO, XETR), transaction type BUY (or SELL with sign flip), market cap available, reference price available, V14e score > 0.
Across this population, we mined patterns with the best edge/robustness ratio: positive lift, N >= 30 in OOS, cohort stable across both validation periods.
Pattern 1: CEO Bid on European Mid-Cap, Direct Purchase
Defined on XPAR and XETR, market cap EUR 300M -- 2B, CEO direct buyer (not via holding), pctMcap >= 0.01.
| Metric | Value |
|---|---|
| N OOS | 312 |
| Mean r90 | +8.4% |
| Win% | 58.3% |
| Isolated Sharpe | 1.47 |
This is the simplest and most robust pattern. A European mid-cap CEO who buys directly, with at least 0.01% of market cap at stake, systematically shows above-average EU_strict returns. This is consistent with the literature: mid-size CEOs have fewer liquidity obligations than mega-cap executives and their purchases are more discretionary.
Pattern 2: CFO + BOARD Cluster Same Week
Defined when a CFO and at least one BOARD member both buy the same ticker in the same week, same market.
| Metric | Value |
|---|---|
| N OOS | 87 |
| Mean r90 | +12.1% |
| Win% | 62.1% |
| Isolated Sharpe | 1.89 |
Clusters are potentially the model's strongest signal, and also the rarest. The CFO + BOARD combination (rather than CEO + someone) is particularly powerful: the CFO knows the exact numbers, the BOARD knows the strategy. When both buy together, it is rarely noise.
Limit: N=87 in OOS over 14 months. Too few for a definitive conclusion, but cohort stable across the two years.
Pattern 3: Controlled Holding Buying Above pctMcap Threshold 0.05
Defined for declarations where related-insider kind is "controlled" and pctMcap >= 0.05 (significant threshold).
| Metric | Value |
|---|---|
| N OOS | 143 |
| Mean r90 | +9.7% |
| Win% | 56.6% |
| Isolated Sharpe | 1.52 |
Significant purchases via a controlled entity (holding company, portfolio company) above the 0.05% market cap threshold signal deliberate intent, not mere RSU management. Legal structure plus significant relative amount equals a clean signal.
Pattern 4: Concurrent Sector Momentum
Defined when the stock's sector (per ICB or SIC classification mapped to the corresponding European sector ETF: EXV1.DE, EXV4.DE, etc.) is in positive momentum regime (63-day return > 0) at the time of the declaration.
| Metric | Value |
|---|---|
| N OOS | 4,218 |
| Lift vs baseline | +1.8pp on mean r90 |
| Win% increment | +2.1pp |
| DSR impact | +0.06 |
This is not an isolated pattern but a contextual multiplier. Insider buys in a sector with positive momentum outperform identical buys in a contracting sector. The effect is modest (+1.8pp) but stable and present across both validation periods.
This is the foundation of the sector multiplier in V13.3s_sector_only, which served as the baseline in all audits 100-103.
Pattern 5: Q4 Seasonality on Nordic Small and Mid-Caps
Observed on XHEL, XOSL, XSTO in Q4 (October -- December), market cap < EUR 500M.
| Metric | Value |
|---|---|
| N OOS (Q4 only) | 289 |
| Mean r90 Q4 | +13.2% |
| Mean r90 non-Q4 | +6.1% |
| Delta | +7.1pp |
Nordic small and mid-caps show significant Q4 seasonality: insiders who buy before fiscal year-end signal strong conviction about upcoming annual results. The effect is specific to Nordic markets (less present on XPAR or XETR at equivalent capitalization).
This pattern is not yet integrated in V14e. It will be robustness-tested in a Q3 2026 audit once the Nordic OOS Q4 2026 cohort is available.
Conclusion
These five patterns are not magic recipes. They represent regularities observed on a specific dataset, over a 14-month OOS window, with the inherent biases of any retrospective analysis. They guide model evolution (V15 in research) but do not constitute investment recommendations.
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).