Machine learning in trading: theory, models, practice and algo-trading - page 463
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
Hello to all the invisible warriors!
No one has made a neural network robot for forex yet?
I already did, I think. But in the form of a multicurrency robot.
According to its calculations I have to buy dollar and sell euro.
I think that everyone already has a neural network trading robot, while those who don't have one still cherish the hope that they will.
Can you explain his calculations?
Or you have more money than the central bank?I don't understand why all of you are so excited about GARCH?
Does anyone have any results on GARCH?
At this stage, I can re-post links (I posted them in this thread) to hundreds of publications on the application of GARCH in financial markets, including forex.
And what can you, well, except empty words...
PS.
I have a working Expert Advisor, in which part of decision-making is a random forest.
A year ago I invited the PAMM with this advisor - silence.
So we stock rags and in a rag, quietly, without a forum.
SEE
GARCH is like everything else - it's a hobby for me. I've mastered it, as if I've achieved something else in my life. I've got ME, then articles, then advisers, then book.... And so I live and give advice to everybody. Otherwise, all the dough, dough.... are miserable.
I don't understand why all of you are so excited about GARCH?
Does anyone have any results on GARCH?
At this stage, I can re-post links (I posted them in this thread) to hundreds of publications on the application of GARCH in financial markets, including forex.
And what can you do, well, except empty words...
PS.
I have a working Expert Advisor, in which part of decision-making is a random forest.
A year ago I invited the PAMM with this advisor - silence.
So we stock rags and in a rag, quietly, without a forum.
SEE
GARCH is like everything else - it's a hobby for me. I've mastered it, as if I've achieved something else in my life. I've got ME, then articles, then advisers, then book.... And so I live and give advice to everybody. Otherwise, all the dough, dough.... are miserable.
Here are the formulas
https://basegroup.ru/community/glossary/garch-mod
Start a branch and work on it. There's a lot to work on. Trust me. But you have to work with your head, not with idle mindless runs in your favorite R.
Get to work, and then, you know, some knowledgeable people will come along.
Here are the formulas
https://basegroup.ru/community/glossary/garch-mod
Start a branch and work on it. There's a lot to work on. Trust me. But you have to work with your head, not with idle mindless runs in your favorite R.
Start working, and then, you'll see, competent people will come along.
This is child's play. There are much more serious publications with results on real currency pairs.
We are talking about dumb work on fitting parameters. There are three groups of them:
Each of these has some variety. By fitting one group well, you can't fit another.
You just have to work, which is what I do.
Understanding overfitting: an inaccurate meme in supervised learning
Good article, I also struggle with model overfitting in a similar way.
But for forex I divide data to trace/test not by random, but by consecutive chunks.
For example if a table has 100 rows, then with 10 fouls I train ten models -
1) training on rows 11-100; test on rows 1-10
2) training 1-10 and 31-100; test 11-20
3) training 1-20 and 41-100; test 21-30
etc.
As a result, test results are combined into a vector, which I compare with the required values
Here's an example in atache.
The more cross-validation fouls the better, but the more time you waste on model creation.
This is a good article, I also struggle with model overfits in a similar way.
But for forex I divide data into trains/tests not randomly, but by consecutive chunks.
For example if a table contains 100 lines, then I train ten models with 10 fouls -
1) training on lines 11-100; test on lines 1-10
2) training 1-10 and 31-100; test 11-20
3) training 1-20 and 41-100; test 21-30
etc. Finally test results are combined into one vector, which I then compare with the required values
Here is an example in atache.
The more crossvalidation fouls the better, but the more time will be spent to create models too.
What I don't like about MO is that it doesn't take into account the nuances of the input quotient. The biggest thing that has been achieved is to separate the noise predictors from the predictors relevant to the target. This completely ignores the statistical characteristics of the input data. I have written about it many times and you know it as well as anyone else on the forum. I write about it not so much for you as for other public.
I'm very impressed by the idea of GARCH: non-stationary time series are not predictable. So we start to identify non-stationaries in the original time series and model them.
We differentiate - much better than the original series.
Remains:
Let us start modeling. In rugarch there are three independent pieces: the trend, GARCH (volatility) and the distribution of this volatility.
I think there is something in this approach. And it is amazing that the number of publications on GARCH for financial series is just huge in comparison with MO. Why?
This is child's play. There are much more serious publications with results on real currency pairs.
We are talking about dumb work on fitting parameters. There are three groups of them:
Each of these has some variety. By fitting one group well, you can't fit another.
You just have to work at it, which is what I do.
You see... you dismiss it as "child's play...". But this is just a fixation of formulas, nothing more, without any "babbling".
And then you mention some "much more serious publications with results on real currency pairs". But you didn't get any use out of them. That said, you've already realized that these "much more serious" publications are just a dumb fit. This understanding is already a good result, and a sign of progress on the path of knowledge.
I'm telling you to turn your head on and work with your head to get out of these "much more serious" frameworks, instead of running around in circles, fixated on "much more serious".
You see... you immediately dismiss it as "childish babble"... But this is only a fixation of formulas, nothing more, without any "babbling".
And then you mention some "much more serious publications with results on real currency pairs". But you didn't get any use out of them. That said, you've already realized that these "much more serious" publications are just a dumb fit. This understanding is already a good result, and a sign of progress on the path of knowledge.
I'm telling you to turn your head on and work with your head to get out of these "much more serious" frames, instead of running in circles, fixated on the "much more serious" ones.
Oleg!
You used to write serious stuff, but now it's like this, well, just nothing.
Sorry.
Oleg!
You used to write serious stuff, but now it's like this, well, just nothing.
I'm sorry.
Apparently, you have not played enough, not run around in circles...
It brings me to tears...
You don't say :) High percentage of culling. I have requirements for many people, I work by Larry Williams, Larry has no losing trades, in ML it cannot be achieved, but it can be achieved by understanding the motivation of different groups of investors with different trading horizons, for this I need ML, there are many students, but not all want to work, I tried to force the winner of the contest, but he still resists, this is only temporary.