The backtest
should look very good because you've optimized the parameters on that data. It goes through a huge number of iterations to see what would have worked best in your testing period.
Now, that's not necessarily a bad thing, but that's why it looks so good. The best way to test this kind of system is a walk-forward backtest. For example, "train" it using Jan-Jun. 2007 data then test it in July 2007. Keep doing that 1 month at a time. Unfortunately it's time-consuming and not fully automated, but that's the best way to test a system that relies so much on optimization.
Quote:
Originally Posted by Harre
Yup, did this and that's how the optimisation actually works. Backtests look very good and I'll try posting them here tonight. I used data from October to the 19 December to avoid holiday trades.
|