Backtesting Strategies

Sebastian Arsjo · 2020-05-13T15:41:09+00:00

I just got my PRT Premium and of course when i now run a backtest on my strategies optimized at 100k for a 200k period they mostly break down , in other words the where more curve fitted then i thought they would be. What is a good way to do a good proper backtest when testing strategies Should i optimize at 100k and then see if it holds up at 200k? ( walk backwards?) or optimize directly at 200k and then do a normal WF test ?

Viewing 2 posts - 1 through 2 (of 2 total)

Author

Posts
05/13/2020 at 4:41 PM #131528 quote

Sebastian Arsjo
Participant

Junior

I just got my PRT Premium and of course when i now run a backtest on my strategies optimized at 100k for a 200k period they mostly break down , in other words the where more curve fitted then i thought they would be.

What is a good way to do a good proper backtest when testing strategies

Should i optimize at 100k and then see if it holds up at 200k? ( walk backwards?)

or optimize directly at 200k and then do a normal WF test ?

05/14/2020 at 10:25 AM #131629 quote

nonetheless
Participant

Master

Sebastian Arsjo wrote:

Should i optimize at 100k and then see if it holds up at 200k? ( walk backwards?) or optimize directly at 200k and then do a normal WF test ?

Hi Seb, I do a combination of both those things. First I optimise all the data I’ve got (200k), then do a WF varying the weakest values, ie the ones I’m least sure of, usually SL, TP, trailing stop, some MA values. If that looks good, then I take the values for the last 70/30 period and run that over all the data. This gives a few months of forward OOS + a much longer historical OOS. If that still looks good then I usually run a VRT randomisation, sometimes several and go for the one that performs best. I expect everyone has a different way of doing these things, but that seems to work for me.

Sebastian Arsjo thanked this post
Author

Posts