Dynamic periods for indicators - page 3

 
Svinozavr писал(а) >>

What do you call 'continuous adaptation'?

Perhaps what you mean by continuous is that only the prevalue is involved in the EMA. But this is a private case.

Here are, for example, two MAs: the red one is the SMA and the blue one is the EMA. The principle of adaptation is the same and is described above. The periods (or the reduced period for EMA) vary from 3 to 111 (lower sub-window). Naturally with SMA the whole sample is recalculated, with EMA only the weighting factor of the new k-value and the feedback (1-k) for the previous EMA bar is recalculated from the other coefficient. Of course, I am aware of the fact that if you use iMA for EMA, it will be recalculated over the entire history if you change the period dynamically. In order to avoid this, a built-in calculation is used.


What I meant was that changing the SMA (i.e. its period) can only happen in steps = 1. And the change in the EMA parameter (weighting factor k) can be any double. Therefore you cannot make the two SMAs as close as possible, but for EMA it is possible.

 
Yurixx >>:

Я имел в виду, что изменение переметра SMA (т.е. его периода) может происходить только с шагом = 1. А изменение параметра EMA (весовой коэффициент k) может быть любым double. Поэтому вы не можете сделать так, чтобы две SMA были сколь угодно близки, а для EMA это возможно.

А... Yeah, I see. Although the point of this contrast... the series is discrete anyway. Okay, well... It doesn't matter in the end.

 
Svinozavr писал(а) >>

))) That's what we're talking about.

Another thing is that indicators are developed for a very specific purpose - to display the desired state of the market, which is further used in the TS. In this case, the goal was to find an adequate channel for a breakout TS.

I have created so many indicators of that type in so many years that I can't even remember them. I can tell you one thing - there is no "grail" here. Better, yes, but not fundamental. Within the framework of general hopelessness.)) Anyway, it's been two years since I practically did it. So... Transferred some of my old ones to MT from Metas and Omegas.

In short, as was put by the topicstarter sub, and tried to answer. Although with ZZ gone a little sideways - there is no din.period (there is a threshold). And auto-optimization of EAs is, you know, quite a different topic.

If you answer the question, and if it even makes sense, then, I repeat, - yes, but we should not count on a dramatic improvement. IMHO, within the time-frame price series paradigm it's unrealistic.


All I've seen on the marketplace is adaptation within the sub-engine. We make a nice indicator. But the idea of adaptive systems exists outside the marketplace and is put as I put it: adaptation should be evaluated by the final indicator of the whole system, in our case by profit (others may be used, which is not crucial).
For some reason it seems to me that we evaluate everything at once. In particular, we fix the profitability of the TS for some time ahead. The market has changed, we adjusted parameters of TS, the past says that there will be profit, etc. Of course it is not clear how to evaluate all the profit components at each next step, but nevertheless
 
faa1947 >>:


Все, что я видел по адаптации на рынкете - это адаптация в рамках сабжа. Делаем красивый индикатор. Но идея адаптивных систем существует вне рынкета и ставится так как ставлю ее я: оценивать адаптацию нужно по конечному показателю всей системы, в нашем случае по прибыли (можно по другим, что не принципиально).
Мне почему-то кажется, что оцениваем все сразу. В частности, мы фиксируем прибыльность ТС на некоторое время вперед. Рынок поменялся, мы подогнали параметры ТС, прошлое говорит, что будет прибыль и т.д. Конечно не ясно как оцнивать все слагаемые прибыли на каждом очередном шаге, но тем не менее

I see your point - I have nothing against it and, in fact, it seems to me just as logical an approach as it does to you.

I was just talking about something else. That's all. And the topic you touched on is also not new - there are threads on the forum.

 
Svinozavr писал(а) >>
Or here's another. Two classic threshold ZZs. The blue one is with a hard threshold of 10pts, the purple one with a dynamic threshold calculated as a fast 3*ATR(5), smoothed by the attenuation of EMA(444). The bottom threshold is also limited to 10 pps. The threshold controlling indicator is in the lower sub-window (upper blue dashed line).

I approached it in a slightly different way - I wrote a gating that has no parameters at all. As a result it defines its own configuration, which generally corresponds to the idea of dynamic adaptation. However, in this case the result is achieved not by changing parameters, but by the algorithm. Here's what happens:

For staking:


Mine behaves in the same way as your blue one, only it doesn't twitch. But that doesn't mean it misses small variations. Where it suits the nature of the market it displays them as well.
 
Yurixx >>:


А я подошел к этому немного иначе - написал ЗЗ который вообще не имеет параметров. В результате он сам определяет свою конфигурацию, что в общем соответствует идее динамической адаптации. Однако, в этолм случае результат достигается не за счет изменения параметров, а за счет алгоритма. Вот что получилось:

Для ставнения:
https://c.mql4.com/forum/2010/04/tdybzy3.JPG
Мой ведет себя также как и твой синий, только не дергается. Но это не значит, что он пропускает мелкие колебания. Там, где это соответствует характеру рынка он отображает и их.

There are different ways of doing this. I just cited the first one that came up. There's an algorithm similar to yours. And many, many others. And then it depends on the task at hand - what we're catching.

No, you can, if you are interested, make a compilation of approaches/algorithms in general or to a zone in particular. Only, you know, I personally it is not so interesting any more - I will hardly take an active part in it. My opinion on prospectivity I have already formed (and even not yesterday) and has stated here. All, of course, IMHO.

 
faa1947 >>:


Адаптировать нужно ТС а не параметры настройки индикаторов. У меня давно такая идея: с приходом нового бара оптимизируем ТС, а затем решаем вопрос с позицией. И так на каждом баре (или n-барах). Будет ли это адаптивной ТС?

Of course this was engaged, including on this forum. However, I don't remember the nicknames, so I can't help you with the search.

Last I remember, Neutron was retraining his NS at every step. He seemed to be happy with the results at first, I wonder how he is doing now.

By the way, I copied my approach from the topic "In the follow-up" completely into MQL and now it's also adapting at every step (more precisely after every position closing, it makes no sense more often). So far no miracles, but spread seems to be compensated :) . Here is a fairly typical picture (balance charts) of a minute history from 1999, that is for more than 11 years.

The horizontal axis is seconds, but the grid lines are approximately years. The lower curve is the basic set of entries; strictly according to the logic, it has a slope of about minus spread per trade (by the way, we can see that the slope, i.e. the density of deals in time, has large-scale fluctuations). The upper curve is the result of that very adaptive filtering of this base set. One can even see that OR for this set of efforts has a positive slope but it is quite negligible from the practical point of view :). By the way, had the testing period been shorter, I could have got a "grail", two plots about a year old would have looked pretty good by themselves :) .

Now I'm thinking, what and in what order (of FP parameters) to feed this beast. Although after these first experiments, some subconscious skepticism seems to be brewing.

P.S. By the way, I have this in the form of an indicator, so not quite an off-topic here :)

 
And my ZZ gives the third option a mark-up :)
 
Guys... Why are we bragging?! :)
There are as many coders as there are variants of rendering, and everyone has about a dozen of his own variants...

The question is different: Are there any adequate algorithms of dynamic adjustment (in a good sense) of period length of one or another TA indicator and the related question: when one can consider that the previous parameter's period is over and since the current bar you have to choose a new value? If they exist - I would like to discuss the methods of their usage.
 
ForexTools >>:
Ребяты... ну чего хвастаемся?! :)
Сколько кодеров столько и вариантов отрисовок может быть да плюс у каждого в загашнике по десяточку своих вариантов наберется...
That's for sure.
The question is whether there are "in nature" adequate algorithms of dynamic adjustment (in a good sense) of period length of this or that TA indicator and, consequently, the related question:when one can consider that the validity zone of the previous parameter is over, and since the current bar one should choose a new value. If they exist - I would like to discuss the methods of their usage.

Yes. That would be in essence. Only, you know...)) it's a fundamental thing for the whole trading theme. Not just for the "dynamic period". By the way, why are you focusing so much on the period? There are some indicators, where the parameter is, for example, a threshold.

I am afraid, that because of globality of a problem (it is allocated in a quote) all again will slide to original disputes from which already sometimes it would not be desirable to look on a forum. I would be glad to be wrong. (Maybe there is Father Christmas after all?))