I was using a WWL system until near the end of 2012. I used the monthly historical results of 26 SIP screens from 2/1998. I could rank each screen with trailing 1,4,6,9,and 12 mo plus total Long term results/StdDev^1.5 and a couple of other algo’s.
I was then able to run a series of genetic optimization programs which selected the weighting of the individual parameters to decide the best selection criteria and how many screens should be invested in. Obviously, overfitting the available data. I do think this helped a little deciding what screens to invest in but there were no spectacular advantages. The primary factors that worked best when I abandoned the system was using the top 5 of 26 screens with Long term CAGR/Std ^ 1.6 and the last year’s CAGR/Std. One thing that stands out looking at the screens selected over time is that individual screens are not consistent winners but some seem to have one or two years of outperformance followed by 3 or 4 years of not being selected. Unfortunately, there are more instances where screens were too intermittent to make selection all that profitable.