Optimisation and Out-of-Sample Testing. - page 10
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Optimisation 24/6/6/3 - number of months.
24 is the main part, identifying sets of parameters which give acceptable results.
2 to 6 is the OOS: running the resulting sets over these intervals.
In this thread, somewhere in the middle, Vita voiced a very competent idea, but nobody heard him - there is no point in doing monkey's work in the form of OOS runs.
In this thread, somewhere in the middle, Vita made a very good point, but no one listened to it - there's no point in doing monkey's work in the form of OOS runs.
"I'm not trying to be clever, I'm just trying to be diligent" ...... Have you still not understood that Forex is 90 percent routine?
I have only heard that after optimization for any lengthy period of time I do not have an instrument to check my selected set.
I've already burned myself on this, and more than once :((
It is much more harmless to immediately write experts who work by bars rather than by ticks :)Thanks for the insight, it's not about the entries but about the follow-up of open positions :)
I've heard it, except that after optimising for however long a period, I don't have a tool to check the selected set.
Burned myself on this already, and more than once :((
And you don't have it in the case of OOS runs either. Or do you count differently? If different, very interesting to hear why.
Thanks for the lib, only I wasn't talking about entrances, I was talking about escorting open positions :)
Well, with the escorting is really more complicated. It depends on the specific vehicle, but the performance on the test points can be an indication of its thickness.
And you don't have it in the case of OOS runs either. Or do you think differently? If not, it would be very interesting to hear why.
There is such a thing as "confidence in the choices you make", not progamming :)
It comes from this area. Take the Vita option. I do not optimize more than from the beginning of 2005 - further in the past the quotes are completely different (look at the minutes). Draconian conditions on tab "optimization" will give in the end (if the Expert Advisor is charged with something positive) about a couple hundred variants of parameters. The question is how to choose? There is only one thing left - to manually run and select the fairest graph - tiresome, don't you agree?
Taking my option. "A conscious choice of 24/6/6/3" ....... a post where you read the states, otherwise you will have to repeat it.
Moreover I've tried forwards too, after selection and optimization: the result is 90% positive.
I may throw the "grail child" overboard, but a sustainable result is worth more.
You are being too frivolous with the notion of a sustainable result. You don't have one :)
Suppose you have a TS with two parameters A and B. You, using your sophisticated multipass selection method, selected the best option and it has A=10, B=20. Now I take your TS and optimize it on the whole sample without splitting 24/6/6/3 and come to the conclusion that of all the results the best is A=10, B=20. So? Your result is now abruptly unsustainable because one pass gave it to me instead of your four passes?
You are being too frivolous with the notion of a sustainable result. You don't have one :)
Suppose you have a TS with two parameters A and B. You, using your sophisticated multipass selection method, selected the best option and it has A=10, B=20. Now I take your TS and optimize it on the whole sample without splitting 24/6/6/3 and come to the conclusion that of all the results the best is A=10, B=20. So? Your result is now abruptly unstable because one pass gave it to me instead of your four passes?
I mean, everyone is free to handle it any way they want..... is not a high-precision lab - we deal with finer matter :)
No. With my result it is the same: I saw expert's work on this set of parameters, at least, on 4 periods ("with features" as you say), and I optimize it on 24 on all variants - Balans and so on.
And, after all, optimising over the whole sample (by the way, will you answer the question of what it should be?) will not only give option A. How to choose?
And, after all, optimising over the whole sample (by the way, will you answer the question of what it should be?) will not only give option A. How to choose?
What can you see after running the system through four adjacent periods, that you won't be able to see after running it through all period at once? So we choose the same way, according to the criteria we need.
What can you see after running the system on four adjacent periods that you cannot see after running it on all periods at once? So we choose the same way, based on the criteria we need.
You've put the question wrong. Better: the rejection of what I do not want to see in any case and never :)
"Guided by the criteria we need" - more on this point, please, it's very interesting, even if we all have different opinions)
PS Do you think I like this multipass pain in the neck? No. I just haven't seen anything better yet.
And there is also one interesting thing, when 6/6/24/3 selection is completely inadequate, although logically it should be vice versa ;)
"Guided by the criteria we need" - more on that point, please, it's very interesting, even if we all have our own opinions.)
Well, here's a simple example. If we use maximal drawdown as one of selection criteria, results of one run are enough to choose a variant with optimal value of this criterion at all segments.
PS Do you think I like this multipass pain myself? No. I just haven't seen anything better yet.
And here's an interesting trick, when 6/6/24/3 selection is completely inadequate, although logically it should be vice versa ;)
Which only confirms my words :)
Well here's a simple example. If we as one of selection criteria, are guided by maximum relative drawdown, results of one run is enough to choose variant with optimal value of this criterion on all sites.
Still 'by hand' and 'by eye' on charts and reports on all sets..... Same thing, "which only reinforces my words :)"
Only my version also agrees on the evenness of profits over all periods.
Guarantees for the future, of course no, 100% only God gives and that's not always...... :)
The question is about "trust" - which expert to trust with your own money..... you are convinced by your version, I am convinced by mine.... it's a matter of taste, nothing more.
But you still haven't answered the question how big the sample should be :)