Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
I see, but if you don't care about their change, you are probably missing a point about the reason WHY your strategy reacts (positively or negatively, winning or losing) to the market. I say this because one of the purposes of WFO is to keep your algo correlated to the market you trade, with some kind of "smart" (not curve-fitted) adaptation. A not appropriate understanding of the way a WFO uses or suggests its values may result in a random exposition to the market. IMHO, to avoid this risk you must be able to make a difference between a correct "identity" of the WFO suggested parameters and a bunch of "magic" numbers a trader may want to use without even understanding why.
Can you help answer these questions from other members on NexusFi?
Yes I do see your point. I too wonder about the amount of importance being put on the WFO for ALL strategy development. After all is this in itself not a form of curve fitting? In sample and out of sample testing is important but with the WFO, the parameters are being constantly adjusted to "fit" the next period of the test. If the WFO then turns round and gives you the best average set of parameters well then what is this only a rake of curve fits just layered over each other. (Mind you, NT do say that the WFO process is their LEAST understood process(outside of NT I assume they mean..LOLLL))
I also think a lot will depend on the type of strategy. An intraday only strategy that works well over say a year of data should not blow up when introduced to another year of data with those same parameters. That is unless there is some seismic change to the behavior of the markets going forward, in which case the WFO would not have been any use anyway as everything is retrospective with its process.
ON THE OTHER HAND..... if your strategy is more of a position type or swing type or something that holds positions for longer periods (days/weeks) then this is a different ball game and I would imagine the WFO starts to become more important/valid....
BTW I think it would be great if people would post a screen shot of what they think are good and bad strategy analyzer results and why they think what they think. This way we could all get more of a feel for whats considered acceptable and not acceptable and indeed to discuss the metrics used a bit more.
As I said, I'm more into asking for your opinion than into demonstrating a thesis. Btw, here is some examples that support my impressions:
Crude Oil contract, excel chart of (daily) opening gap sizes since 2006.
You may notice that the distribution is more or less homogeneous on the long run, apart from a slightly evident bias to the down side. However, if you "normalize" these data by considering the progressive daily range, you may get a quite steady line of distribution.
Now, as Bosch777 correctly noticed, if you take care of the overall scenario, you may suppose that there isn't any real change/evolution in this parameter. Or maybe you might be persuaded that there's a slight bias to a specific direction (red dotted line), and this will widely make room for a WFO test.
However, you may get a very different impression if you give a look to the short-term perspective (the vertical positions of the contiguous blue dots): the change from one day to the other, or from one week to the following one, is totally erratic. This is immediately reflected by the WFO results, that usually take these changes too "seriously" by proposing an adaptation that has no logical or strategical grounding.
I was thinking about WF along similar lines. I was not sure if WF was the ultimate tester for robustness of the strategy because of limitations of in sample windows and market conditions that exit then and following change in the market conditions in out of sample results. And that is what i noticed that many models suffered during WF at the time then the market conditions have been changing - for example going into 2008-09. So, my concern where was that the data that i have been in sampling before 2008-09 did not have market conditions as in these years. As you roll over, you start to account for the these years and they take bigger part of your in sample optimized parameters when you come out from 2008-09 and thereafter when the market normalizes back again and you model gives still quite volatile out of sample results following such roll for the next couple of yours. And thereafter the results normalize.
So, i was not and am not sure whether to use WF or all in sample data and just leave 2-3 latest year as out of sample. Any suggestions are much appreciated.
Recently i have developed a strategy that does not have a single best optimum parameter for the whole data: I was testing on ES, NQ. It performs well most of the time but then has extended period of flat performance if you optimize for the all the data.
BUT, when I plugged in the model into WF, the OS results came nearly perfectly upward slopping. The model has only two variables and both of them did not fluctuate much at all, just a bit to accommodate for changing market conditions - just like as you mentioned above. I guess such models are better used with regular WF re-optimization as their robustness is prone to changing market conditions (not sure if that is not data fitting in itself)).