Another possibility is to analyze the equity curve itself so as to go live with the system when good performance is expected and to either reduce risk or just paper trade when performance is expected to be negative. Are a series of positive returns followed by negative returns (i.e. is there mean reversion)? Does a trend-following "meta-system" and/or a trailing stop loss (on the total account equity) reduce risk or at least make it more tolerable? A couple other ideas might be to try [supervised learning][1] for the drawdown periods or incorporate concepts from [control charts][2].

There is some danger that you might just end up multiplying your [model risk][3] by overlaying a meta-strategy on your original system. Statistically significant changes may be very hard to come by and in the end you might just have to trade smaller size. Good luck. 


  [1]: http://en.wikipedia.org/wiki/Supervised_learning
  [2]: http://en.wikipedia.org/wiki/Control_chart
  [3]: http://en.wikipedia.org/wiki/Model_risk