OptF factor selection bias

Posted By: nanotir

OptF factor selection bias - 01/07/16 11:58

Hi

I read in a post about z12 http://www.opserver.de/ubb7/ubbthreads.php?ubb=showflat&Number=449794&page=19. There, jcl is referring to a possible bias due to the calculation of the OptF factor

Quote:

Second, it is too optimistic, due to the OptF factor calculation over the whole simulation period. This introduces selection bias. So it is possible and has in fact happened that you made loss in a period when the simulation made profit.

So I wonder if there is a way to eliminate or evaluate such a bias. Something similar to the detrend function which is used for different kind of bias.

Posted By: jcl

Re: OptF factor selection bias - 01/07/16 14:52

The bias could be eliminated by calculating the OptF factors not from the whole period, but using new factors for every WFO period. This would indeed work for factors calculated in a different way, but not for Optimal-F, since they rely on the extrema over the whole backtest.

So the answer is no, the factor bias can not be eliminated in an easy way. It is thus recommended to develop the system with no factors for verifying if it is profitable or not. Use the factors only in the last step.

Posted By: nanotir

Re: OptF factor selection bias - 01/07/16 19:41

Good to know that there is a way eventhough the strategy comes with this bias.

Posted By: Jeff1228

Re: OptF factor selection bias - 01/12/17 02:18

Interesting stuff