Currently generating strategies using a lot of MC test and the following:
I then go ahead and run 2 other tests.
- Out of sample 2: 2017 – Current date. Profit Factor > 1.2
- Retest on correlated market. GBPUSD on EURUSD as an example. Profit Factor > 1.1
However I see some issues with this, so would like to get more opinions.
- Strategies are bound to have a bad year, this might just be it. Or is 1 bad year not acceptable?
- Currency pairs don’t always have GREAT correlation.
Please let me know what you think?
1. There are very few strategies I know without any bad years in backtested history, in the end your total DD should be managed on portfolio level. If you will delete strategies only because they had 1 bad year, you might be also deleting good candidates.
2. I think the GBPUSD test is only meant to see how your systems will perform on an another (but similar) data set. In this case PF>1.1 check is a common/standard value. Basically you are checking, if your system is not totally curve-fitted to the ‘noise’ on EURUSD, this because the ‘noise’ by definition can not be reproduced, so it should be very different on a different currency pair. But this is just my opinion ;)
Do you want to become a profitable trader? Join our free community to learn more about algo-trading:forum.coensio.com
I agree with Coensio on both points.
To elaborate on point 2, if you have a strategy which has great RET/DD, PF etc on EURUSD, but miserably fails on GBPUSD, that’s a red flag and you might want to consider being cautious, since the strategy might be curve fitted. But if it’s slightly winning or at least not badly loosing, that’s good enough for me to go on further with the strategy. If it is significantly profitable on the other market -> bonus! but not a necessity.
You must be logged in to reply to this topic.