--- In [email protected], "zozuzoza" <zoz...@...> wrote: > > Hi, > > I would like to open this topic to discuss and share WFO experience. There > are some interesting issues below to start with. Please let us know if you > come across with these and know the answer to them. > > - Dakota states that they step 1 bar ahead, so each bar is OOS result. There > is one question here. If the parameters change each bar how does it treat the > open trades? How does Amibroker treat open trades at the beginning/end of the > OOS period? > - Wathcing the videos in Biocomp web page, the WFO improves as the time > passes. You can see that the performance is poor in the beginning but it > improves a lot in terms of equity straightness later. How does it improve > itself as the time goes on? > - What is your experience what is the best time period of the IS and OOS data? > - How did you solve the problem that there are too few trades used in IS > optimisation because the optimisation engine pushes towards the parameters as > they give the best CAR/MDD for example. Is it somehow possible to tell the > optimiser to keep at least 50 trades in the IS period otherwise the > performance results are invalid? > > Cheers, > Zozu > I use an in sample period of 1/1/2000 - 12/31/2003, and of course, rolling every year after that.
Since I use the components of an index, I am especially mindful of survivorship bias (it really does exist). Too few trades in the iis period is not a problem for me - I make sure that I have at least 30-50 trades each year in the in sample period. Deciding on the objective function to optimize is a real issue. I have used several metrics, including a combined one. I am still convinced that there is no "best", but I tend to use kratio, and check the other key metrics.
