Interesting idea ....
So this is one architecture, what others are there?
What ideas are there?
That is, I propose that we discuss TC architectural concepts here. For example the one above, it can be explained differently, simpler...
1) We take history, take a simple strategy, fit it, get a set of parameters
2) Compare the current market situation with the one we had at fitting, if it is too low then we exit the trade.
3) Gone 3 until similarity starts again at a given level.
The term "fitting" means that the TS works (brings profit) on history, but it does not work in the future on the real account (it fails). This is the essence of the term "fitting". Then what is the sense of testing it on real account if it fails?
Пачка должна быть солидной, 20 стратегий мало. Каждая стратегия в ней - со своим удельным весом. Солидная пачка - есть модель рынка.
I would rather put it this way: we don't need a pack, but rather a family of strategies, homogeneous in principle and differing in some parameters. Then we need to ask a question: what parameters should have been set, say, an hour ago, in order to obtain the maximum profit within this hour until the current moment (or the minimum drawdown, or maximum pips, or to lose deposits as quickly as possible - your choice). After that we consider that within (conditional) 10 minutes after this so called optimization the market properties will be preserved and - if there is a signal - we make an entry.
То есть я предлагаю тут по обсуждать архитектурные концепции ТС. Вот например одна выше, ее можно исзложить иначе, проще...
1) Берем историю берем простую стратегию, делаем подгонку, получаем набор параметров
2) Делаем сравнительный анализ текущего состяния рынка, с тем что был при подгонке, если подобие низкое выходим из торговли.
3) гото 3 пока не начнется снова подобие на заданном уровне.
***********
Собираем пачку таких примитивных стратегий, гоняем их с подгонкой на истории, таким образом получам пакет стратегий и им соответсвующих подобий "рынка" ( я говорю рынка, имею ввиду что это график цены ну буквально ). Собираем их в пачку и натравливаем на рынок, из 20 скажем таких пар, несколько будут торговать... Что и требовалось. Просто и сердито. Никаких высших математик. :)
***
Начало было тут - https://www.mql5.com/ru/forum/123412/page2
Secondly, everything gets stuck at the stage of identifying the physiognomy of the current market
Firstly, it cannot work in principle if it is based on fitting
2) Make a comparative analysis of the current market situation with the one at the time of adjustment, if the similarity is low, exit the trade.
Slippery point. We evaluate new HISTORICAL data for similarity of MA (training sample). But how does it follow that there is such similarity ahead of us?
Once upon a time I thought a lot in this direction. Here's another somewhat more advanced "architecture", we find some similarities on history to fresh historical data (the latest), and use the data following it as a synthetic BP for OB. And train our simple strategies on it. As soon as the current situation changes, we look for new similarities and so on. My result is at least we don't merge even on a simple crossing of wipes...
That is, I propose that we discuss TC architectural concepts here. For example the one above, it can be explained differently, simpler...
1) We take history, take a simple strategy, fit it, get a set of parameters
2) Compare the current state of the market with the one we had at fitting, if the similarity is low then exit the trade.
3) Goto3 until similarity starts again at a given level.
***********
We gather a pack of these primitive strategies, run them with fitting on the history, thus obtaining a package of strategies and their corresponding "market" images (I say market, I mean a literal price chart). We assemble them into a pack and apply them to the market, out of say 20 pairs, several will trade ... What we need. Simple and easy. No higher mathematics. :)
***
Started here - https://www.mql5.com/ru/forum/123412/page2
How does p.2 differ from: we pick a filter that improves results? The filter is the formal rule of comparing the current state of the market with what it was when it was adjusted or in good periods. It gives "True" if the market is as it should be and "False" otherwise.
...
Собираем пачку таких примитивных стратегий, гоняем их с подгонкой на истории, таким образом получам пакет стратегий и им соответсвующих подобий "рынка" ( я говорю рынка, имею ввиду что это график цены ну буквально ). Собираем их в пачку и натравливаем на рынок, из 20 скажем таких пар, несколько будут торговать... Что и требовалось. Просто и сердито. Никаких высших математик...
If the goal is to find "similar market conditions", then why should these conditions be analyzed using strategies that themselves may be very inadequate (at least because they are "primitive")? Just analyze the price itself and its movement (my thread was here. Has anyone tried to train their experts this way? My solution is also available there).
or am i missing something in the idea of using "primitive EAs"?
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
I like your idea, thank you for it!
In other words, it's a contrary approach - we make a primitive system that is strictly based on the fit, then we make some analyzer that detects deviation of the fact from the fit, and if there is deviation, we stop the work. Then we can make a detector of "adjusted" strategies, and when the "fact" is inside, we start working. The idea is good, but it's the same as in adaptive filtering. But just as another view of TC concept, thanks.
The topic should be marked as self-active, without relative to filters.
That is, I propose to discuss here architectural concepts of TC. Here is for example one above, it can be stated differently, more simply...
1) Take history, take a simple strategy, fit it, get a set of parameters
2) Compare the current market situation with the one we had at fitting, if it is too low then we exit the trade.
3) Goto3 until similarity starts again at a given level.
***********
We gather a pack of these primitive strategies, run them with fitting on the history, thus obtaining a package of strategies and their corresponding "market" images (I say market, I mean a literal price chart). We assemble them into a pack and apply them to the market, out of say 20 pairs, several will trade ... What we need. Simple and easy. No higher mathematics. :)
***
Started here - https://www.mql5.com/ru/forum/123412/page2