Would You Trade This System?
- Thread starter jayjay121
- Start date
+1 to that.
Based on your walk forward results I think it's safe to test the system live with a small position size, unless the back test makes postdictive errors or unrealistic fills. Since that part of the strategy has not been disclosed (I may have missed it), I guess you're on your own.
Then again, you never know just whose opinions you're actually getting in here.
Quote from Wide Tailz:
+1 to that.
Based on your walk forward results I think it's safe to test the system live with a small position size, unless the back test makes postdictive errors or unrealistic fills. Since that part of the strategy has not been disclosed (I may have missed it), I guess you're on your own.
Then again, you never know just whose opinions you're actually getting in here.
thanks......
yes thats right, this is only one symbol and if i can find a way to go basket trading with this then should in theory improve, but hey nothing surprises me with trading. Things you think should work normally don't work out the way you think it.
I think walk forward only means something if you are optimizing your parameters to a past subset of data and then fixing them for future trades. To me it's almost identical to the out-of-sample testing that you had already done. Of course, a possible wrinkle is that if you believe your parameters will last only for say 5-10 years, then the walk forward might be more relevant and more reflective of your actual live trading than out-of-sample from the distant past.
should i do another wfo test with slightly different settings?, like change exit to more, not entry. Say for first 3 years to optimise then use out of sample after that to test on?
this way it would a completley new test and may even show some more robustness as im altering the exit.
I could even slightly change the timeframes, what do you think?
Im just trying to rule out ive coincedently fitted some data or i actually have discovered a real pattern
That is a good point: a sensitivity analysis of your PL with small changes to parameters can at least show if you have a curve-fit system. If you can make small changes to each of your parameters and compare the PL on those and if the PL changes drastically then you have most certainly fitted against noise. On the other hand, if your PL is robust to small changes in all your parameters, then you may have a real signal (but I think you can't prove the latter unfortunately)...