Optimisation of Hermes

My optimisation of my "Hermes" low-volatility breakout system is proving to be very messy.

 

The trouble with optimisation is that you can take it in any direction. At the moment I added an RSI filter to check how the sytem performs when going with and against the trend.

 

There is a noticable improvement in profit factor when trading with the trend. I have also noticed that the system performs best against the trend if I use a Reward:Risk ratio greater than 2, while trading with the trend performs best with a Reward:risk ratio of less than 2.

 

I suppose this makes sense. If you are trading with the trend, the likelihood of running into the end of the trend is relatively high. On the otherhand, if you trade against the trend and the trend-reversal is strong, it can turn into a new trend and you succeeded in catching the beginning of this trend. This would explain the bigger pay-off if trading with a larger reward:risk ratio.

 

With that in mind, I have also noticed other behaviour that can be used to "optimise" my system further. For example, I have noticed that trades perform better when you have a multitude of entry signals on the same day. I guess the entry signals on different pairs reinforce each other. Trading only on days with multiple signals may improve my profit factor, but at what point does it end being an exercise in optimisation, and the start of curve-fitting?

 

To be safe, I may stick with one level of filtering only. I also feel that I'm getting sidetracked. Mechanical systems are not intended to be perfect. They should be "rough but ready", and ideally deployed as part of a portfolio of independent trading systems. To win mechanically, you should trade with a low risk per trade, and let high frequency and exponential growth do the heavy lifting. To do this, though, you'll need more systems to trade with. I'll return my attention to system development.