Looking for patterns - page 26
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
Bad solution, it's more complicated and more importantly much worse than a normal equivolume chart. Switching to the Range Bars explain, how are they equivalent to volumetric charts?
I have thought about it and changed my mind, if we operate exactly with ticks, then everything is correct. A tick is cheaper at night in a thin market than during the day, all true. But it does not cancel the problem of the average volume profile, while distributions are very different from day to day.
There are some ideas on how to improve and multiply the probability of correct inputs. Your views, Alexei, do not support much of what I see as ideal. But if you look at it from our general perspective, you could tentatively do the following:
Take a few valid and effective TS, analyse them, perhaps add some details. and try to combine them. You expressed the idea that the more data, the more they will conflict. I'm definitely of a different opinion and suggest different uses for the data from yours. But in the case of multiple TCs, you could, for example, do the following:
1) Each of the TS has certain conditions for opening positions. So the idea is not to combine them into one, but to observe these conditions in parallel. That is, each algorithm of the system works on its own.
2) Add an additional algorithm, which receives from each system the estimation of the situation, which is expressed in one of three conclusions: favorable conditions for entering, unfavorable, uncertain.
3) For example, the user sees with his/her eye suitable conditions to open a position. Then he/she refers to that algorithm, which gives conclusions from a parallel working group of systems. The probability of a correct entering a position increases by orders of magnitude based on their conclusions.
Many people work in this way. But it is problematic to continuously keep several TS and conditions of each of them in mind. The presence of an indicator, which would signal several final parallel conclusions of the TS, would simplify the process of analysis and decision making.
Moving on from the search for flavours to the trivial, already baked by someone else cutlet....
If you mean my ideas, I don't insist. Suggest yours.
At the moment the topicstarter encourages sharing observations, but everyone is basically observing for themselves. That's why I suggested specific options for implementation.
Good day to you all!
Actually, I once wrote an article, or rather a post, on price fluctuations (see appendix). If you are interested you can dig deeper ... :)
The idea is that if the oscillations have the amplitude close to the average value (for example for EURUSD it is ~200 pips), then the forecast of the next "knee" is quite real. Another thing is to determine if the zigzag break has occurred. But we have certain tools for that as well, including those based on the popular neural networks.
Here are certain regularities ... :)
Regards, RomFil.
You could also mention the search for pAtterns: one-bar, two-bar, three-bar, etc. It takes a long time to find more than three. For example, for one-bar - we take OHLC prices, enter a delta of changes, for example +-1 pips from the price and divide all the bars on the history (for example, 1000000 bars) by the following principle: the first bar is a pAttern, if the next bar differs from the previous one by the delta value, it will be a new pAttern, etc. If one of the pATterns shows that, for example, there were 10 such bars on the history and 9 of them are going in one direction - this is an important pAttern and it works in 70-80% of cases in a forward test.
It's a bit complicated, but it seems to be understandable... :) You may go deeper into the subject here as well.
Regards, RomFil.
You could also mention the search for pAtterns: one-bar, two-bar, three-bar, etc. It takes a long time to find more than three. For example, for one-bar - we take OHLC prices, enter a delta of changes, for example +-1 pips from the price and divide all the bars on the history (for example, 1000000 bars) by the following principle: the first bar is a pAttern, if the next bar differs from the previous one, it is a new pAttern and so on. If one of the pATterns shows that, for example, there were 10 such bars on the history and 9 of them are going in one direction - this is an important pAttern and it works in 70-80% of cases in a forward test.
It's a bit complicated, but it seems to be understandable... :) You may go deeper into the subject here as well.
Regards, RomFil.
I'm far from technical notation, but if I understand correctly, you suggest to take into account certain patterns and define them in parts, predicting continuation, right?
I'm also far from neural networks, but I imagine information processing in approximately the same way: mass of data with the most interesting patterns, and identifying them by individual manifestations. Your ideas are very interesting. I've read the document, but the calculations don't fit in my head. Could you give me a couple of examples of your algorithms, but expressed not mathematically, but in user language?
Thanks for the useful ideas!
I'm far from technical notation, but if I understand correctly, you propose to take into account certain patterns and identify them piecemeal, predicting continuation, right?
I'm also far from neural networks, but I see information processing in a similar way: mass of data with the most interesting patterns, and identifying them by individual manifestations. Your ideas are very interesting. I've read the document, but the calculations don't fit in my head. Could you give me a couple of examples of your algorithms, but expressed not mathematically, but in user language?
Thanks for the useful ideas!
Ok! But it's hard for me to explain it to you without formulas ... :(
The point is to find such a combination of OHLC on history that the next bar has a sufficiently accurate direction, for example downwards - this is for pATterns.
About the document - it is a different topic.
Right! But it's hard for me to explain it to you without formulas ... :(
The idea is to find such a combination of OHLC on the history that the next bar has a quite accurate direction, for example downwards.
Well, I generally understood the idea. I will look forward to your further comments. So far no one has offered such variants of approach, it is the closest to me.
Right! But it's hard for me to explain it to you without formulas ... :(
The point is to find such a combination of OHLC on the history that the next bar has a quite accurate direction, for example downwards.
OK! But it is difficult to explain it to you without formulas... :(
The point is to find a combination of OHLC on history with the next bar having a quite accurate direction, for example downwards.
And also the next bar has a big size, from H to L. If you know the direction of the next bar and the price changes within this bar only by 10 pips, you will not earn much.