Criterion for automatic selection of optimisation results. - page 5
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
Критерий = Сумма длин отрезков на которых ТС находилась в прибыли деленная на сумму длин отрезков на которых ТС находилась в просадке. Отрезки с прибылью и с просадкой могут пересекаться. Отрезки можно измерять в минутах или в количестве сделок.
Pardon me, I forgot - this criterion has to be considered together with the recovery factor.
And any others are fine :)
Azzx,
Интересно, сколько ордеров на самом деле будут закрыты с профитом на реале?
I have no idea - the EA was, of course, created solely for the tester.
And what's wrong - it clearly controls the beginning of the bar - i.e. the checks don't happen somewhere in the middle of the bar?
StatBars писал(а) >>
By the way, I see that people don't quite understand that the main difficulty is not to come up with evaluation parameters (there are plenty of them), but to come up with a way to bring them together, i.e. there are 1000 passes, we have several indicators for each pass, how to choose the best strategy using these several indicators? Or 5 best strategies?
I use my criterion ("quality of equity") exactly in the sense of justifying the choice of option in some approximation. I.e. the profit is there, the number of trades is normal, the profit factor is also real - you need to choose among several variants. Actually, imho, you should look through all variants, if they are not "close to each other" by parameters. The main criterion here, imho, will be the smoothness of equity. Imho, of course - that's how I see it.
Можно эти несколько показателей нормировать, и подать на вход НС или комитету НС. На выход –прибыль на OOS.
As an option, you could try to bolt on grids and teach them to choose the TCs most likely to be profitable on the CB.
1 - ("minus") skewed classes, because there are very few normal TCs.
2 - sample will be small for more or less adequate experiments, I think this point will not give the opportunity to work with the NS in this direction (do not forget that you also need to provide a test sample, and then the issue of fitting the grids will not be solved).
3 - input parameters will have to be normalized :)
Я свой критерий ("качество эквити") использую именно в том смысле, чтобы обосновать выбор варианта в некотором приближении. Т.е. профит есть, число сделок нормальное, профит-фактор тоже реальный - нужно выбрать из нескольких вариантов. А вообще-то, имхо, стоит просматривать все варианты, если они не "рядом сидящие" по параметрам. Тут главным критерием, имхо, и будет гладкость эквити. Имхо, конечно - я это так вижу.
Well, in principle I manually select strategies in the same way as you described, but I determine the quality of exvity by eye :), so we're OK with the indicators...
Here is some interesting information on the subject: http://www.marketprofit.ru/doc/wealth-lab-otsenka-rezultatov-testirovaniya-sistemy.htm
It may come in handy.
The more parameters in the EA that significantly change the behaviour, the more chance of getting a fit instead of optimisation. Even with a small drawdown, large profitability, and other good indicators, you can get a loss on OOS... Why? Yes because the fit. I tell you the cure for this terrible disease of the TS. The most important indicators are not those that we see, but those that we do not see. Let me explain: imagine that one passage is a track, profit is time of passing the track, deposit is a car. Every time we do not enter a turn, it is a losing trade. We only get to see the car at the start and at the finish. If we have good time (big deposit), there is not a scratch on the car (small drawdown), but how do we know how it has passed turns, may be it has flown them in millimetre from the post, and not one but 200 of available 300. The conclusion is this: we do not see those very dangerous deals that the TS passed in millimetre, and therefore on an unfamiliar track, we catch all these poles! Therefore, the strategy should be built so that all possible deals were opened. If not all of them, then those which are not necessary should be as far away from their triggering as possible. But it's easier to open the first of course, because it's easier to open everything possible and try to keep it than to move away what we don't need. If your strategy opens most of the bad trades, it is not at all certain that the filter will help. It is better to change your mind about the signal itself. It should be smoothly advancing and smoothly decaying, this will give smooth parameter obsolescence. As a result, the quality of trades will also fall gradually. And the better your TS is, the longer the parameters will be sufficient.
If you build your TS on this principle, you won't have to puzzle over the formula for calculating the criterion.
Although I've derived such a formula for myself. But it is only in order not to run a half an hour with eyes and calculator, looking for good indicators ...
Построив ТС на таком принципе, не придется ломать голову над формулой высчитывания критерия.
Хотя такую формулу я для себя вывел... Но она только для того, чтоб не бегать полчаса глазами и калькулятором, выискивая хорошие показатели...
Can you share the formula?
After this thread I decided to modify it a bit, but here is what I have now:
PF - Profit Factor.
SdDay - number of deals per day
ProcDay - profit percentage per day (complex formula with logarithms)
MD - Maximum drawdown
SrD - Average drawdown (drawdowns of each order divided by the number of orders)
if(PF>3) Vigoda=2*SdDay+(10*(ProcDay/((MD+SrD)/10));
else Vigoda=(PF-1)*SdDay+(10*(ProcDay/((MD+SrD)/10)));
I'll be glad to tweak it together or rewrite it in general...
The conclusion is that we do not see the very missed dangerous trades that the TS passed within millimetres, and consequently on an unfamiliar track we catch all those poles!
That is, if I understand correctly, how to evaluate not the result, but the quality of trades, how well do the trades meet what is built into the system?
I'll have to think it over in my head.