Machine learning in trading: theory, models, practice and algo-trading - page 1640
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 don't want to go back to Renko, I've already wasted time on it, not only does it completely lose OHLC information, but in addition you get a lag of two heights of Renko bricks - it lags very much
The same will probably be true for ZigZag, but I have not dealt with it specifically.
Of OHLC, only O is unambiguously identified immediately when the corresponding tick arrives. In reality, the opening can be missed when there is a delay in processing previous ticks.
There is no perfection in the world)
From OHLC, only O is unambiguously identified immediately at the moment of arrival of the corresponding tick. In reality, the opening may be missed if there is a delay in the processing of previous ticks.
There is no perfection in the world)
Everything is very complicated here
Many participants work with bar opening and closing prices, there are both trivial indicators and complicated modeling and manipulations of those who quote prices.
The High and Low prices make sense during the creation of channels, ZigZags and breakdowns of historical maximums (min), the graphical analysis, as well as settaps and patterns - this also makes sense
and these grandfatherly methods are still used in trading, i know for sure that americans actively use graphical analysis - i communicated with them several years ago.... But they also know it doesn't work)))
The problem is generally with the market participants themselves - they constantly create a disturbance on an ideal price series!
It is possible to get rid of session volatility fluctuations by switching to a zigzag or renko, isn't it? Of course, the natural time structure will suffer, but it is possible to introduce the normal time as an indicator set for each knee/brick.
I've turned to zigzag... But since the beginning of March they are just not comparable to what they were before March. If before it could take half an hour or an hour to build a knee, it may take 5 minutes now because of the high volatility with the same parameters. So it makes no sense to study on the data before March. Everything is different now.
We should still think of something universal for high and low volatility.
Maybe something wave like that. The waves have remained, they have just become wider.
I googled it, but I'm still going to ask here
what types of neural networks can be used as a control system?
At least for this example: NS should give a response that will open an order in the desired direction and set the value of take and stoploss, i.e. 3 control parameters (buy/sell + tp + sl)
EN: reinforcement learning
https://github.com/EliteQuant/EliteQuant#quantitative-model
RU: reinforcement learning
https://www.mql5.com/ru/search#!keyword=%D0%BF%D0%BE%D0%B4%D0%BA%D1%80%D0%B5%D0%BF%D0%BB%D0%B5%D0%BD%D0%B8%D0%B5%D0%BC
although if the parameter set is fixed, as well as their values, it looks like the usual classificationI googled it, but I'm still going to ask here
what types of neural networks can be used as a control system?
At least for the following example: NS should give a response that will open an order in the necessary direction and set the value of take and stoploss, i.e. 3 parameters of control (buy/sell + tp + sl)
What a problem you're trying to solve!
Here's the plan:
1. make a network which guesses the direction of price movement after a fixed time.
2. if p.1 is successful, make a network which guesses the direction dynamically, i.e. gives out the direction and time where it is most confident.
3. if you have item 2, make a network which guesses the direction and strength of movement.
4. if you have item 3, create a network which guesses the direction and strength + time of movement.
After that go to your 3 parameters.
EN: reinforcement learning
https://github.com/EliteQuant/EliteQuant#quantitative-model
RU: reinforcement learning
https://www.mql5.com/ru/search#!keyword=%D0%BF%D0%BE%D0%B4%D0%BA%D1%80%D0%B5%D0%BF%D0%BB%D0%B5%D0%BD%D0%B8%D0%B5%D0%BC
although if the parameter set is fixed, as well as their values, then it looks like a normal classificationtraining with reinforcement I don't want any more, I was getting a fit, more effective random GA tester to use, at least the forward passes the TS found by genetics better than training with reinforcement
Thanks for the git, I will study a large collection of ready-made material
What a task you are trying to solve!
Here is the plan:
1. Make a network that guesses the direction of price movement after a fixed time.
If p. 1, then make a network that guesses the direction dynamically, i.e. gives the direction and time where it is most confident.
3. if you have item 2, make a network which guesses the direction and strength of movement.
4. if you have item 3, create a network which guesses the direction and strength + time of movement.
Then go to your 3 parameters.
it's an RNN network or a standard multi-layer perceptron does it fine
the problem is just in the control of the object - google neurocontrol, you can read the terminology on the same wiki and google it afterwards
it's an RNN network or a regular multilayer perceptron that does it fine
the problem is just in the management of the object - google neurocontrol, in the same wiki you can read the terminology fluently and then google them
Not an expert, but at first glance this approach for markets is unlikely to give anything beyond NARX. And this model, it seems, can always be implemented through RNN. And the completeness of RNN by Turing also contributes to its sufficiency.
Article on the equivalence of NARX and RNN.
Not an expert, but at first glance this approach for the markets is unlikely to give anything beyond NARX. And this model, it seems, can always be implemented by RNN. And the completeness of RNN by Turing also contributes to its sufficiency.
I googled it, according to my observations I should use RBF-net.
OK, I'll ask more specifically: there is a portfolio of primitive TS, which are forward tested, each TS is time bound within a day and the TS may overlap in time - you need"something that will try to go through the portfolio" depending on the input data - OHLC
just do the portfolio optimization by enumerating in the tester's genetics can.... But I want some intelligence)))
I googled, according to my observations I should use RBF-network
Ok, I'll ask more specifically: there is a portfolio of primitive TS, which are forward tested, each TS is bound by the time of operation within the day and TS may overlap in time - you need"something that will try to enumerate the portfolio" depending on the input data - OHLC
just do the portfolio optimization by enumerating in the tester's genetics can.... but I want some intelligence)))
The task looks not quite formalized - the set of parameters is not clear. Is the complete set of systems finite, countable or continuous? Is the portfolio a fixed size? Is the system included in the portfolio with certain weights or just yes/no?