Out of Sample for Random Generation
6 replies
rafaeldelrey
7 years ago #115565
Hi,
Is there any reason for using “Out of Sample Period” in the Data tab, under Settings, when using Random Generation? I understand its use on Genetic Generation, where strategies are evolved. But if the Random generation doesnt make any kind of evolution, optimization, improvement during the generation process, is there any reason for an OOS period in this specific phasis?
Patrick
7 years ago #139421
It helps you to sort out the strategies. If you add some filter, as OOS test, but also PF, Return/DD ratio etc it will produce much smaller databank than without those filter.And you would use those filters later anyway. OOS test helps you to not save strategies which are curvefitted to data used for development. It is difference when there is 50k or 1k strategies in databank for next filtering.
_Cujo
7 years ago #139424
At an absolute bare minimum, you probably want OOS to be net profit greater than 0. I’m not saying that’s all you should do, but you probably want an out of sample test, on “unseen” data, to be at least positive…let alone good ret/DD, nice PF, etc..etc..
Patrick
7 years ago #139427
but when OOS is a few years, you definetly want some minimal PF, higher than 1.00000001 etc
rafaeldelrey
7 years ago #139434
Since it appears there is no optimization or evolution in a Random Generation, I cannot see how an OOS test would help determine curve-fitting. Curve-fitting happens during optimization and model fitting. There is no fitting, nor evolution, nor optimization. Such OOS filtering, the way it is done in Random Generation, would just mean another in-sample filtering in a diferent period, unless such generation involves some kind of internal optmization.
The same thing is observed with the third OOS test in the ebook. There was no adjustment, optimization or improvement after the robustness tests, soh de OOS3 has no meaning besides testing in another period. I am new to SQ, so I might be missing something.
Patrick
7 years ago #139435
my mistake i havent noticed RANDOM in the post, sorry. in that case it helps not.
Edinho
7 years ago #139469
I disagree with you guys. Curve fitting doesn`t happen just with optimization.If you don`t do OOS test how can you determine the strategy is suitable for real trading? You need to realize you looking for a strong statistical edge from the history which may and may not happen in the future. If you avoid OOS test you will get strategy which was profitable on a short set of data but it may crash soon in real market condition.
Viewing 6 replies - 1 through 6 (of 6 total)