Asset-neutral close-H H10 run 282 — concentration PnL exacte
Rejouer le ledger 282 sélectionné gelé (ridge_a1000, coverage 0.50, cap1, c30, H10, no whitelist) et calculer contributions PnL exactes par actif/mois, leave-one-asset, leave-best-month, rolling subperiods, worst-month et top contribution; aucun changement de modèle/seuil/cap.
Audit plutôt favorable mais non promu: la config 282 gelée conserve un score Validation positif après retrait du top actif/mois, avec concentration sous garde-fou, tandis que Test reste positif; Recent demeure trop mince.
Contrôle anti-leakage: config gelée de 282 choisie sur Validation seulement: ridge_a1000, coverage 0.50, cap1, coût 30 bps, H10. Les diagnostics Test/Recent sont affichés seulement.
Validation — leave-one actif
| Actif | Trades | Σ ret | Abs share | Leave score | Leave gain | Leave DD | Avg/trade |
|---|---|---|---|---|---|---|---|
| DOT | 4 | 48.02% | 15.0% | 127.63 | 156.9% | -19.5% | 12.005% |
| ADA | 13 | 37.10% | 11.6% | 138.79 | 168.1% | -19.5% | 2.854% |
| XRP | 8 | 35.49% | 11.1% | 151.14 | 180.4% | -19.5% | 4.436% |
| LINK | 4 | 19.42% | 6.1% | 192.01 | 221.3% | -19.5% | 4.854% |
| SOL | 16 | 19.28% | 6.0% | 198.12 | 227.5% | -19.6% | 1.205% |
| BCH | 1 | -8.96% | 2.8% | 289.08 | 318.4% | -19.5% | -8.959% |
| SUI | 7 | 3.82% | 1.2% | 236.71 | 268.2% | -21.0% | 0.545% |
| ETH | 1 | 2.55% | 0.8% | 242.14 | 271.4% | -19.5% | 2.547% |
Validation — leave-one mois
| Mois | Trades | Σ ret | Abs share | Leave score | Leave gain | Leave DD | Avg/trade |
|---|---|---|---|---|---|---|---|
| 2025-10 | 3 | 52.00% | 16.2% | 117.70 | 147.0% | -19.5% | 17.332% |
| 2024-12 | 10 | 37.39% | 11.7% | 137.01 | 169.9% | -21.9% | 3.739% |
| 2025-03 | 8 | 23.58% | 7.4% | 178.12 | 207.4% | -19.5% | 2.947% |
| 2025-04 | 3 | 19.90% | 6.2% | 185.29 | 214.6% | -19.5% | 6.635% |
| 2025-02 | 7 | 16.44% | 5.1% | 208.54 | 237.8% | -19.5% | 2.349% |
| 2026-02 | 2 | 5.64% | 1.8% | 231.47 | 260.8% | -19.5% | 2.819% |
| 2024-07 | 3 | 5.50% | 1.7% | 232.37 | 261.6% | -19.5% | 1.833% |
| 2024-08 | 5 | -4.85% | 1.5% | 283.52 | 312.8% | -19.5% | -0.971% |
| 2025-01 | 3 | 1.84% | 0.6% | 244.98 | 274.3% | -19.5% | 0.613% |
| 2024-11 | 4 | 1.00% | 0.3% | 248.27 | 277.6% | -19.5% | 0.250% |
| 2025-08 | 1 | -0.99% | 0.3% | 255.41 | 284.7% | -19.5% | -0.991% |
| 2025-11 | 1 | -0.82% | 0.3% | 254.76 | 284.1% | -19.5% | -0.823% |
Test — concentration reporting-only
| Actif | Trades | Σ ret | Abs share | Leave score | Leave gain | Leave DD | Avg/trade |
|---|---|---|---|---|---|---|---|
| DOT | 1 | 47.44% | 26.3% | 23.70 | 49.6% | -17.3% | 47.440% |
| ADA | 4 | 18.03% | 10.0% | 61.01 | 86.9% | -17.3% | 4.507% |
| LINK | 2 | 10.64% | 5.9% | 74.76 | 100.7% | -17.3% | 5.320% |
| SOL | 10 | 10.08% | 5.6% | 73.61 | 101.1% | -18.4% | 1.008% |
| SUI | 2 | 4.74% | 2.6% | 84.65 | 110.6% | -17.3% | 2.368% |
| XRP | 5 | 3.13% | 1.7% | 110.01 | 119.8% | -6.5% | 0.626% |
| Période | Trades | Score | Gain | DD | Avg/trade |
|---|---|---|---|---|---|
| 2025-Q1 | 16 | -4.10 | 21.8% | -17.3% | 1.560% |
| 2025-Q2 | 3 | 11.97 | 14.0% | -1.3% | 4.578% |
| 2025-Q4 | 3 | 49.96 | 50.0% | 0.0% | 16.380% |
| 2026-Q1 | 2 | 1.19 | 5.9% | -3.2% | 3.114% |
Recent — concentration reporting-only
| Actif | Trades | Σ ret | Abs share | Leave score | Leave gain | Leave DD | Avg/trade |
|---|---|---|---|---|---|---|---|
| XRP | 2 | 6.23% | 49.6% | 0.00 | 0.0% | 0.0% | 3.114% |
| Période | Trades | Score | Gain | DD | Avg/trade |
|---|---|---|---|---|---|
| 2026-Q1 | 2 | 1.19 | 5.9% | -3.2% | 3.114% |
Leçon et implication
Leçon causale: La performance asset-neutral 282 n'est robuste que si le PnL Validation ne dépend pas d'un actif/mois jackpot. Top actif Val=DOT share_abs=15.0% leave_score=127.63; top mois Val=2025-10 share_abs=16.2% leave_score=117.70.
Next implication: Si on poursuit, lancer un paper-shadow gelé forward-safe sans retuning ou un veto prédéclaré uniquement si le veto est choisi sur Validation dans une hypothèse séparée; ne pas utiliser Test/Recent pour choisir une whitelist.
Graphiques Val/Test
Sélection sur Validation uniquement. Le Test est affiché seulement pour juger le transfert.
