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)
Well you have to choose "parameters" for Walk Forward e.g. length of in-sample vs length of out-sample. Which measure of goodness to use. It's easily possible to start "optimizing" the walk forward process.
Kevin: Do you have more information about your WAlk-Forward process and the practicalities related to it. Maybe a webinar or something?
Trading: Primarily Energy but also a little Equities, Fixed Income, Metals, U308 and Crypto.
Frequency: Many times daily
Duration: Never
Posts: 5,059 since Dec 2013
Thanks Given: 4,410
Thanks Received: 10,226
Ahh okay. Yes I agree you can optimize the walkforward itself, different is/os periods, different start dates etc etc and that would obviously could lead to over fitting, which is why you shouldn't do it.
Thank's Kevin that's interesting. As you know I've looked at Cluster Analysis, and would say in that 4 out of 5 cases my cluster analysis produces better/more stable results. My clustering is just 'average gross profit of nearest neighbors' but maybe that is the exact type of overfitting you mean. I've had so few systems pass either walkforward test though it's difficult to analyze the effect of clustering on my results.
Off topic - for many years the Texans were called Denver South. After last night I'm not sure whether we should be Cleveland South or whether you should be Houston North!
Pretty soon, the Browns will have all the Texans picks in future drafts. But it does not matter, the Browns would screw up a draft even if they had every single pick!
I think there are ways to do Cluster analysis correctly, and ways to just curve fit with it. In any analysis/testing, I always like to see as much performance as possible with unseen data. That to me is the key.
Kevin, which system would you say is more robust - the one that uses no variables (or in other words the variable is fixed and is not wf) or a system where you wf a single variable after a certain time frame? Assume here that systems are built in the same manner where IS period is all data except for the past 2 years. I am asking this question as you come across both systems but not sure which one you should strive for as logic behind both of them is usually different.
Really, either one could be bad, or either one could be good. It all depends HOW it was developed. For example, using no variables (where you fix the variable value) can be nice, but not if it was derived after running an optimization, then realizing you could "fix" it to a set value. Similarly, you can screw up the wf approach it multiple ways (cluster analysis when done incorrectly is just one way to mess it up).
Ultimately, no matter how you do it, live (future) data is the best judge of which is better.
Not the answer you hoped for, but I hope it helps...
Do you think there could be an edge from counting consecutive Loser's and Winner's in a strategies history and from that analyze and implement increased or decreased position sizing . I think the key point is to cut the extrems from a sequence
when the probability of another losing trade or another winning trade is very very weak
I have three strategies working on ES, they did pretty good the last two years since I put them live but this year I'm really having hard times keeping them working. Overall they are still in profit YTD but numbers are very small and defenitely this is due to the low volatility. I'm trying to think which could be the events that could increase volatility, French election for example were a non event.
So I was wondering if anyother who have strategies on ES is experiencing my same difficulties...
That approach might help ONLY if your trades exhibit serial correlation.
Otherwise, it would work about as well as applying it to a random coin flip, which also has consecutive loser/winner runs. In other words, it would work only be sheer luck.