I assume you want to know about techniques that will allow you to establish if the performance of system A is significantly different from system B ?

If the distribution of gains and losses generated by your various systems are normally distributed you could use a simple students T test.

The problem is that the distributions in return from any system with a positive or negative edge will by definition exhibit a degree of kurtosis and skew, so something as simple as students T probably wont be appropriate

You need to be looking at non-parametric methods, probably stuff like the Mann Whitney U test or Wilcoxon Signed Rank Test

Did some stats at university but its a long time ago, but I'm not a statistician by any means