Asset-neutral close-H H10 run 282 — contrôle leakage/alignement
Exécuter des contrôles négatifs rapides sur la forme 282: mêmes folds/features/scaler Train-only mais labels close-H permutés par blocs ou décalés hors causalité, petit menu Ridge uniquement, sélection Validation-only identique. Publier distribution Val/Test sans promotion.
Contrôle négatif OK: aucun pseudo-signal exploitable comparable au run 282 n'est apparu. Meilleur contrôle Validation-only: block_permute_train_y_64 ridge_a1000 coverage 0.50; Val pénalisé -19.40, gain 39.13% DD 0.00% sur 5 trades; Test reporting-only -3.27% DD -3.27% avg -3.272% sur 1 trades; Recent 0.00% sur 0 trades.
Anti-leakage: scaler et modèle fit uniquement sur Train; labels Train volontairement non causaux; Validation choisit seulement le moins mauvais contrôle; Test/Recent restent descriptifs. Ce run ne peut pas promouvoir un système.
Courbes du contrôle sélectionné
Leçon et suite
Leçon causale: Ce contrôle vérifie l'hygiène du pipeline asset-neutral: en remplaçant uniquement les labels Train par des séries non causales, le même code ne doit pas reconstruire une alpha stable sur les retours originaux. Un résultat faible/négatif renforce la confiance procédurale dans 282 sans le promouvoir; un résultat fort signalerait fuite, alignement ou sélection indirecte.
Implication suivante: Si le contrôle reste faible, continuer uniquement des audits gelés/paper-shadow de 282 et attendre du forward; si suspect, arrêter les audits promotionnels 282 et auditer l'alignement des labels, la construction yL et les splits avant toute nouvelle hypothèse alpha.
Distribution des contrôles
| Sel. | Contrôle | Modèle | Coverage | Split | Val pen. | Score | Gain | DD | Trades | Avg/trade | Actifs | Top PnL |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| block_permute_train_y_64 | ridge_a100 | 0.50 | val | -223.73 | -95.10 | -34.0% | -40.7% | 9 | -4.080% | 4 | 47.1% | |
| block_permute_train_y_64 | ridge_a100 | 0.50 | test | -223.73 | -71.79 | -28.7% | -28.7% | 5 | -6.436% | 3 | 43.3% | |
| block_permute_train_y_64 | ridge_a100 | 0.50 | recent | -223.73 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| block_permute_train_y_64 | ridge_a100 | 0.65 | val | -83.45 | -32.73 | -3.4% | -19.6% | 6 | -0.125% | 4 | 51.5% | |
| block_permute_train_y_64 | ridge_a100 | 0.65 | test | -83.45 | -8.18 | -3.3% | -3.3% | 1 | -3.272% | 1 | 100.0% | |
| block_permute_train_y_64 | ridge_a100 | 0.65 | recent | -83.45 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| ✓ | block_permute_train_y_64 | ridge_a1000 | 0.50 | val | -19.40 | 39.13 | 39.1% | 0.0% | 5 | 7.003% | 3 | 54.2% |
| ✓ | block_permute_train_y_64 | ridge_a1000 | 0.50 | test | -19.40 | -8.18 | -3.3% | -3.3% | 1 | -3.272% | 1 | 100.0% |
| ✓ | block_permute_train_y_64 | ridge_a1000 | 0.50 | recent | -19.40 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% |
| block_permute_train_y_64 | ridge_a1000 | 0.65 | val | -141.29 | 2.21 | 2.2% | 0.0% | 1 | 2.211% | 1 | 100.0% | |
| block_permute_train_y_64 | ridge_a1000 | 0.65 | test | -141.29 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| block_permute_train_y_64 | ridge_a1000 | 0.65 | recent | -141.29 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| roll_forward_train_y_240 | ridge_a100 | 0.50 | val | -405.57 | -247.73 | -98.9% | -99.2% | 1605 | -0.223% | 10 | 25.1% | |
| roll_forward_train_y_240 | ridge_a100 | 0.50 | test | -405.57 | -249.59 | -99.8% | -99.8% | 1206 | -0.475% | 10 | 28.8% | |
| roll_forward_train_y_240 | ridge_a100 | 0.50 | recent | -405.57 | -196.12 | -76.8% | -79.6% | 337 | -0.393% | 10 | 38.9% | |
| roll_forward_train_y_240 | ridge_a100 | 0.65 | val | -406.53 | -248.22 | -99.1% | -99.4% | 1604 | -0.237% | 10 | 19.5% | |
| roll_forward_train_y_240 | ridge_a100 | 0.65 | test | -406.53 | -249.18 | -99.6% | -99.7% | 1203 | -0.419% | 10 | 29.5% | |
| roll_forward_train_y_240 | ridge_a100 | 0.65 | recent | -406.53 | -169.52 | -64.4% | -70.1% | 335 | -0.269% | 10 | 34.9% | |
| roll_forward_train_y_240 | ridge_a1000 | 0.50 | val | -409.13 | -249.24 | -99.6% | -99.7% | 1604 | -0.293% | 10 | 18.5% | |
| roll_forward_train_y_240 | ridge_a1000 | 0.50 | test | -409.13 | -249.27 | -99.7% | -99.7% | 1204 | -0.423% | 10 | 20.3% | |
| roll_forward_train_y_240 | ridge_a1000 | 0.50 | recent | -409.13 | -166.33 | -63.6% | -68.5% | 336 | -0.262% | 10 | 30.3% | |
| roll_forward_train_y_240 | ridge_a1000 | 0.65 | val | -402.51 | -246.18 | -98.1% | -98.7% | 1602 | -0.193% | 10 | 24.6% | |
| roll_forward_train_y_240 | ridge_a1000 | 0.65 | test | -402.51 | -249.15 | -99.6% | -99.7% | 1202 | -0.407% | 10 | 23.0% | |
| roll_forward_train_y_240 | ridge_a1000 | 0.65 | recent | -402.51 | -151.44 | -54.3% | -64.7% | 334 | -0.199% | 10 | 19.0% | |
| random_sign_train_y | ridge_a100 | 0.50 | val | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a100 | 0.50 | test | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a100 | 0.50 | recent | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a100 | 0.65 | val | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a100 | 0.65 | test | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a100 | 0.65 | recent | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a1000 | 0.50 | val | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a1000 | 0.50 | test | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a1000 | 0.50 | recent | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a1000 | 0.65 | val | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a1000 | 0.65 | test | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% | |
| random_sign_train_y | ridge_a1000 | 0.65 | recent | -96.00 | 0.00 | 0.0% | 0.0% | 0 | 0.000% | 0 | 0.0% |
Graphiques Val/Test
Sélection sur Validation uniquement. Le Test est affiché seulement pour juger le transfert.
