Results: what are the published signals worth?
In short. As of September 11, 2026, 1038 verdicts are dated and none has reached 90 days yet: the first possible reading is given below, and the method is published before the figures. A verdict counts as a hit when the stock beats its exchange index over the horizon, from the first close strictly after the verdict.
How are verdicts evaluated?
- look_ahead : entrée à la première clôture strictement postérieure à la date du verdict / de rebalancement ; macro et calendriers filtrés par ts_connu ; fondamentaux : date de connaissance = fetched (sinon +90 j après clôture d'exercice), donc exclus du walk-forward avant 2026-08-30
- survivance : univers = instruments présents en base au jour du calcul (3995 actions) : les valeurs retirées de la cote avant l'entrée en base manquent (biais de survivance PARTIEL, non corrigé ; les relevés datés de univers_indices permettront de reconstituer les compositions passées)
- narratif : aucune sélection d'exemples : toutes les fenêtres, tous les verdicts, avec n
- objectif : métrique fixée avant le calcul : IC de Spearman hors échantillon à 90 j et excès contre l'indice de place en EUR ; aucune métrique ajoutée après lecture des résultats
- couts : rendements BRUTS : ni courtage, ni TTF, ni spread, ni fiscalité — un IC positif brut ne prouve pas un gain net (haircut 50 % recommandé, McLean-Pontiff 2016)
Hit rate by signal and horizon
No verdict resolved to date; first resolution expected on {'30': '2026-09-29', '90': '2026-11-28', '180': '2027-02-26'}.
| Horizon | Signal | n | Beats index [95% CI] | Universe base rate | Mean excess [CI] |
|---|---|---|---|---|---|
| 30 j | strong | 0 | – | – | – |
| 30 j | moderate | 0 | – | – | – |
| 30 j | weak | 0 | – | – | – |
| 30 j | no signal | 0 | – | – | – |
| 90 j | strong | 0 | – | – | – |
| 90 j | moderate | 0 | – | – | – |
| 90 j | weak | 0 | – | – | – |
| 90 j | no signal | 0 | – | – | – |
| 180 j | strong | 0 | – | – | – |
| 180 j | moderate | 0 | – | – | – |
| 180 j | weak | 0 | – | – | – |
| 180 j | no signal | 0 | – | – | – |
Do market factors hold out of sample?
30 factor tests on 3995 stocks, temporal split fenêtres mensuelles triées ; les premières 2/3 = entraînement, le dernier tiers = test (hors échantillon). Best 90-day test: secteur_mom_6. The cumulative number of trials (47) enters the deflated Sharpe ratio.
Benchmark caveats
- Personal analysis, decision support: not investment advice.
- Gross returns (no fees, no taxes); past performance is no guide to future results.
- Only 0 verdict(s) resolved at 90 d: no conclusion on verdict quality is possible before 2026-11-28 at the earliest.
- Walk-forward: 5 test(s) out of 30 significant at FDR 10% out of sample; 47 cumulative trial(s) enter the deflated Sharpe.
- Joint bootstrap at 90 d (69 common test windows, blocks of 3): 2 feature(s) significant under Romano-Wolf (FWER 10%); Hansen SPA p = 0.000 for the best one (secteur_mom_6).
Public methodology · All analysed stocks
- This content is an automated analysis produced using artificial intelligence. Personal analysis for information purposes — no personalised investment advice. MMR content consists of general research. It does not take account of your financial circumstances, objectives or risk tolerance. Data, analyses and results, including those produced using artificial intelligence, may contain errors, be incomplete or become outdated. No return or outcome is guaranteed. All investments carry a risk of partial or total loss of capital. Past performance and simulations do not predict future results. Before making any decision, check information against original sources and, where necessary, consult an authorised professional. You remain responsible for your investment decisions. This notice does not exclude or limit any liability that cannot legally be excluded or limited.