Criterion for automatic selection of optimisation results. - page 7
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
Vita, нормально в принципе. Предотсеивание - по числу сделок, м.о. и PF, а окончательное отсеивание - по вполне приличному интегральному критерию вместе с "облачностью". Т.е. фактически вся процедура предполагает фильтрацию примерно по пяти разным критериям.
Сами цифры предотсеивания (м.о., надеюсь, в старых полноценных пунктах, т.е. на четырехзнаке?) вполне логичны. Я бы, наверно, увеличил минимальное количество сделок (скажем, до 200), чтобы увеличить статдостоверность результатов.
Yes, the MO is 50 four digit points. I'd increase the minimum number of trades too, but 200 trades a year, i.e. 50 pips a day, is just my dream for now :). Although I must confess I practically do it this way - I check my chosen parameters on more trades, probably, to amuse my ego - the results are too amazing. :)
And I completely agree, that there is more than one criterion, no matter how I turn this problem with integral criterion. However, I understand the author of the theme - well, the optimizer got a lot of rainbow results. And who will tell me which one to choose? That's what I want, throw the results in the table, pull the formula and voila! That's what I do too, but I'd rather use pre-screening. It kills >99% of all the optimizer's results. Unfortunately, optimization based on profitability or expected payoff quickly turns into searching for one superprofitable deal, and optimization based on balance turns into one million deals with profitability close to one and expected payoff - zero. Setting the minimum balance cures the disease, but not to the end.
Would it be good if the genetic algorithm was hustling around looking for PF>2 & EP>50 & N>50? Or something like that, depending on your tastes?
Прибыль * Прибыльность / Просадка в %. это хорошо, но чтоб период тестирования не влиял на показатель прибыль
The effect of the period occurs only when the optimizer gives me results for different periods, i.e. when I do it myself with my own hands - first I optimize one period, then another, then I puzzle over the problem. No, I don't make life that difficult for myself.
As far as I understood the author of the theme, he would like what I would like (otherwise I would not understand it :) ) - To overlay an integral indicator on the optimization results obtained by strategy tester. So I do and I don't see the problem of testing period influence, because the results show the same period, but the index is absolute, i.e. it cannot be used for comparison either with other periods or other instruments etc.
I don't see any prospect in the whole idea of complex integral indicators. Maximum drawdown is the worst reality, and so it is a sufficient indicator of "worstness" for me. I also believe that averaging, normalisations and various delving into details is an illusion that the problem is under control, until you prove that your result is not an accident.
However, I once did an "integral indicator" indicating how far the results of the trades have strayed from the random distribution. I would like to ask Matemat for his opinion on this idea. Suppose we have a spherical MTS in a vacuum, namely, it has fixed stop levels. Suppose TP=75 and SL=25. And so the signals are such that TP is closed in 50% of cases, and stoploss, so also in 50%. Obviously, we have squeezed out 25% of the randomly trades in favour of a profitable one. Is it possible to squeeze some scientifically valid coefficient of reliability of this very 25% from the results of trades?
It is an excellent system by the way, even if it is spherical in a vacuum. A spherical grail.
Obviously, we have squeezed 25% from a random layout in favour of a profitable one.
If you calculate the PF, it is very good and equals 3. The higher the PF, the greater our margin compared to a random entry. For this system, with the same TP and SL, we will not get into serious drawdown of the system until the ratio of frequency of profitable to frequency of unprofitable becomes permanently worse than 1:3, i.e. 25%:75%. Well, that's probably our 25% chance squeeze (or is it triple?). But we have to remember that while we have squeezed 25% out of a profitable one, we have also squeezed the same 25% out of an unprofitable one. This is a very good thing.
Is it possible to squeeze some scientifically valid coefficient of confidence of that very 25% from the results of the trades?
If there are plenty of deals, why not?What is the probability that the system's PF on the interval of 500 trades, equal at the moment to 3.0, will become 1.0 or worse on other 500 trades? Everything actually comes down only to the ratio of profitable to unprofitable trades.
I researched this question superficially in my article on peaking sandwiches. There are no definite answers (probably the insidious pipsqueak let me down here), but there is a clear dependence of the degree of reliability of conclusions on the sample size: if you have PF=3 on 40 trades, the probability of losing the next 40 is not too low at all (I used to call it a "shock drawdown" in my article). But if at the same PF the number of trades is 500, then on the next 500 it is very unlikely that the system will sag.
If you can be sure that your spherical system is Bernoulli's, then experiment.
P.S. I guess I've already bored everyone here with my Bernoulli...
Не вижу перспективы в затее со сложными интегральными показателями. Максимальная просадка - это худшая реальность, и поэтому это достаточный для меня показатель "худшести". Также считаю, что усреднения, нормализации и различные углубления в детали - иллюзия того, что проблема под контролем, до тех пор, пока вы не докажете, что ваш результат не случаен.
I used to think the same way until I learned how to correct the maximum drawdown for extra profit. It sounds like science fiction, but strangely enough it is reality.
This will require a strategy within the main strategy. The following signals are developed:
- signal to determine the start of the deviation (signal for the beginning of the drawdown, i.e. closing the position for its subsequent reopening with the old TP)
- a signal to cancel the reopening (reset the expectation of reopening, i.e. the deviation turned out to be a reversal without a return)
- a signal to open with the old TP.
Раньше я также думал пока не научился исправлять максимальную просадку на дополнительную прибыль. Звучит конечно из разряда фантастики, но как ни странно это реальность.
Для этого потребуется стратегия внутри основной стратегии. Разрабатываются следующие сигналы:
- сигнал определения начала отклонения (сигнал начала просадки, т.е. закрытие позиции для ее последующего переоткрытия со старым TP)
- сигнал на отмену переоткрытия (сброс ожидания переоткрытия, т.е. отклонение оказалось - разворотом без возврата)
- сигнал на открытие со старым TP.
As I understand it, the first signal, the drawdown start signal, is meant to indicate that we should close the position if the drawdown increases more than desired?
The second signal is to make us forget that we opened at all. The third one is to enter the old position after "recovery from a drawdown". Or what?
Как я понимаю, первый сигнал, сигнал начала просадки, призван указать на закрытие позиции, если просадка усиливается больше желаемого?
Второй сигнал - заставляет забыть, что мы вообще открывались. Третий - войти в старую позицию после "восстановления из просадки". Или как?
That's right, in fact the first signal is to close the position and wait for the third or second, and the second signal is to cancel waiting for the third...
Все верно, фактически первый сигнал - это закрытие позиции и ожидание третьего или второго, а второй сигнал - это отмена ожидания третьего...
And then the question about the nature of these signals - are they simply levels that we can optimally pick up on history or is there a calculation in the nature of the signals, maybe even related to basic opening and closing signals?
These can be a variety of signal variations, e.g:
first - weakening of the main one, appearance of bad signs, deviation with discontinuities, etc...
Second - crossing of an important level, reversal of the main trend, break of the channel, or some strong signal that cancels or postpones possible returns...
Third - recovery of the main signal, a bounce from the trendline or horizontal line