[Soundwave-]

Originally Posted by xXOpkillerXx

Why not do machine learning with "predictions" as output and the trades you do manually as dataset. That way you get to learn even more about the params needed for a good trade, and you might even see surprising results in the weights in the end. Just an idea...

I mean this is basically the point of backtesting, although I'm not sure if traditional machine learning has been applied to it before.

I've wanted to develop some open source procedural trading software, but the problem is to test it well you either need money or market data, neither of which are amazingly easy to obtain.