I don't think this is going to work, it's way too easy to overoptimize...

I've been going in circles trying to find a way to do it that tests well in and out of sample, and anything that works for a few years, doesn't work for the next few. It will break about even out of sample and perform amazingly on the data it's tested on, no matter what I do with it. I may need to filter the trades more, etc., but it's becoming quite time-consuming...

__________________
Banned from FF! A badge of honor...