this seems like it is extremely prone to overfitting to noise...especially with only 4000 observations in your sample it is very easy to find seemingly profita

this seems like it is extremely prone to overfitting to noise...especially with only 4000 observations in your sample it is very easy to find seemingly profitable strategies in backtesting (even outsample) by just tinkering with a few parameters, in this paper many many parameters are tuned