There is statistical evidence that Marchus (@Whycliffes) has a good model:Two Pre-Registered Designer Models on OSF - #28 by Jrinne. Specifically he has one of the models in a set of 3 models with a FDR of 0.2 measuring the excess returns over the last 2 years.
To be clear the BH model to determine this is limited to the last 2 years for excess returns. I think other models would also do well if the period of excess returns could be extended. Also there were some assumptions made regarding the volatilities of the model's excess returns as weekly excess returns or the volatility of the excess returns is not provided P123's download of Designer Models. This weakness would be eliminated if P123 provided the weekly excess returns for designer models as a download or even if they provided the standard deviation of the weekly excess returns in the Designer Model Download. Maybe with options of greater than 2 years for excess returns if P123 wishes to facilitate statistical methods for evaluating Designer Models and newer methods.
It bears repeating that only the last 2 years could be tested due to the limitations of the data.
That having been said, he had the top model: https://www.portfolio123.com/app/r2g/summary?id=1764562. And advanced Machine Learning methods have a relatively short history at P123 anyway.
Marchus, in the past you have said you use evolutionary algorithms including genetic algorithms and CMA-ES:ML Script Performance vs. Genetic Algorithms & CMA-ES**
I would be interested in anything you wish to share about your method for this model including whether you are still using evolutionary algorithms.
Thanks in advance for sharing anything you are comfortable sharing.
Jim