Looking for patterns - page 37
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
I notice that you are keen to determine what the next bar will be. And indeed such goals are mostly the goals of AI developers. It's not so much about predicting the trend as it is about the detailed development in steps. But in trading it is not so much important how the movement will be (i.e. location and type of bars), as how to catch the movement in a certain direction by a certain amount of points without errors. And speaking of neuronets and big data, what if we undertake not to OHLC of one or two bars but to determine individual reversal patterns that in principle do not also consist of hundreds of bars? Plus if we use interactive fundamental events as data (including parsing and recognizing Trump's tweets).
Guys, you can't guess anything. No one. Any masses of information, all the combined knowledge of people is incapable of predicting the future. Trying to guess is a waste of time. Many people delve heavily into theory, even though everything lies on the surface.
There are a few patterns (which we look for) that can be used to some advantage. That is, we will close more profitable trades than unprofitable ones. But at the moment of order opening, there is absolutely no confidence in its fate.
Further we have only one question: when to close it. If the price goes in our direction, what will we do? And if the price goes against us, what will we do? That's it.
The rest doesn't work, believe me, nets, counts, genetics. Anything that isn't simple is very unstable. The market's going to turn it all around all the time.
Guys, you can't guess anything. No one. Any masses of information, all the combined knowledge of people is incapable of predicting the future. Trying to guess is a waste of time. Many people delve heavily into theory, even though everything lies on the surface.
There are a few patterns (which we look for) that can be used to some advantage. That is, we will close more profitable trades than unprofitable ones. But at the moment of order opening, there is absolutely no confidence in its fate.
Further we have only one question: when to close it. If the price goes in our direction, what will we do? And if the price goes against us, what will we do? That's it.
The rest doesn't work, believe me, nets, counts, genetics. Anything that isn't simple is very unstable. The market's going to turn it all around all the time.
It's like that song. Your words
Guys, you can't guess anything. No one. Any masses of information, all the combined knowledge of people is incapable of predicting the future.
Not to reproach, but you, Alexey, have a habit of a priori dismissing what you have not tried to understand. Today, AI is not only guessing the future, it is shaping it. This can be clearly seen in various developments. I have even given you examples. But in principle, my aim is not to impose or prove anything. I am simply interested in the point of view of a person who knows about it. And this person really understands how these areas work.
I will try to limit my comments in the future, but there are key things that I think are important to state publicly. I can understand your reluctance to discuss at all what you yourself have discarded, but I ask you and everyone else to be loyal to some of the key points. Perhaps later on this seeming theory, or even someone might see the philosophy, will change the views of some developers.
I notice that you are keen to determine what the next bar will be. And indeed such goals are mostly the goals of AI developers. It's not so much about predicting the trend as it is about the detailed development in steps. But in trading it is not so much important how the movement will be (i.e. location and type of bars), as how to catch the movement in a certain direction by a certain amount of points without errors. And speaking of neuronets and big data, what if we undertake not to OHLC of one or two bars but to determine individual reversal patterns that in principle do not also consist of hundreds of bars? Plus if you use interactive fundamental events as data (including parsing and recognising Trump's tweeter posts).
You are wrong in your opinion about my aspirations. About OHLC you started it yourself ... :) I only said that there is such a possibility.
A detailed development might just suggest a trend! Here is an example: let your system predicts it is correct 80% of the time, on D1 it says that the next day will be down, on the current H4 it is also down, on the current H1 - where will you open? Who prevents "detailed development" from doing three, five or ten bars?
Now, the reversal pATterns: how to determine them? If we use the methodology of neural networks and AI (don't kick only those who know and understand... :)). Here the length of this pATtern is important. If it will be equal in the same number of bars, then everything has been thought out long ago. But nothing is the same ... :)
Who prevents "detailed development" from doing three, five or ten bars?
The thing is, I don't understand the technology itself. From my discussion with Alexey, I deduced that big data causes a heavy load. Even one extra bar can stretch calculations to infinity. So I understood your point. I won't litter the forum. We will find a moment sometime and discuss it in person.
Vitaly, don't be offended. I have been researching this issue. First with handwritten number recognition, there was such a question on the job, this MNIST database. The recognition was very successful. And after some time I tried to use the nets in trading. I've been doing it for a long time, and I've learned a few things from this experience, which I'm now saying. I understand that someone else's experience is always unimpressive because the necessary amount of sensations has not been felt and experienced.
You don't have to limit yourself, say what you want. My experience may not have been very good either, maybe somewhere along the line...
Guys, you can't guess anything. No one. Any masses of information, all the combined knowledge of people is incapable of predicting the future. Trying to guess is a waste of time. Many people delve heavily into theory, even though everything lies on the surface.
There are a few patterns (which we look for) that can be used to some advantage. That is, we will close more profitable trades than unprofitable ones. But at the moment of order opening, there is absolutely no confidence in its fate.
Further we have only one question: when to close it. If the price goes in our direction, what will we do? And if the price goes against us, what will we do? That's it.
The rest doesn't work, believe me, nets, counts, genetics. Anything that isn't simple is very unstable. The market's going to turn it all around all the time.
Your reasoning is flawed... And I can even guess why. You tried it and saw the result, after that you are not interested in this subject. I can understand you perfectly well, but there are others, eager like me, who are only spurred on by failures to understand the subject in more detail. And I will say that by using neural networks a trader has a definite statistical advantage over others like him.
And now I want to give you an idea, but not the one that I described today (I will describe the idea of "paired trading" a little later), but more of a "strategic" nature. You say "reversal" and what do you mean by it? Just on history. There's a thread somewhere on the forum about it - what counts as a U-turn. For example, if we consider an ordinary zigzag, its tops are reversals, but not all of them are important, i.e. there are some reversals after which the price goes not far, and then it turns and goes many...
The whole problem in determining the pATterns that we're all looking for is to create a target function that gives a clear signal of a reversal. If you and I can define such a function, then let the computer do the rest. I.e. if we take a neuronet (any neuronet, but something modern is better) and input any OHLC or 100500 indicators and reduce the values to this target function - we will be able to receive many of such necessary patterns having statistical advantage or, by name of the branch theme, regularities.
Sincerely, RomFil
P.S. I wrote a post, then I saw about neural network experience and I was right ... :) Sometimes I am afraid of myself.
Having researched a lot of approaches and done 100500 experiments, I want to say that the target function has long been invented! And as you probably already guessed - it's an ordinary lwma!!! About this later ... :):):)