Using neural networks in trading - page 36
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
Well, it is hard to say whether the NS or the regression is a better predictor
even linear regression can give better results than NS
It does. Absolutely. One way or another. My personal opinion - to find the highs and lows you need other approaches... If the market is calm, you can follow the market, I traded it for a long time. But when it throws feverishly, you will only catch tails ))))
It is. Absolutely. One way or another. My personal opinion - to find the highs and lows we need other approaches... If the market is calm, you can follow the market, I've traded it myself for a long time. But when it throws feverishly, you will only catch tails ))))
It's about something else entirely.
This is a matter of principle.
Take regression for example. Having built this model it is possible, without running the model, to answer the question "can this model be trusted". This model is parametric and the accuracy of parameter determination (sko) will be specified in the reference information. If we are talking about linear regression it will very quickly become clear that the error of parameter determination often exceeds the value of that parameter. That is, the figure we see does not actually exist at all! Hence the conclusion that linear regression, that particular one, cannot be applied regardless of what the tester shows.
TA and NS have no such tool, and for econometrics it is the standard, and extremely developed.
It's about something else entirely.
This is a matter of principle.
Take regression for example. Having built this model it is possible, without running the model, to answer the question "can this model be trusted". This model is parametric and the accuracy of parameter determination (sko) will be specified in the reference information. If we are talking about linear regression it will very quickly become clear that the error of parameter determination often exceeds the value of that parameter. That is, the figure we see does not actually exist at all! Hence the conclusion that linear regression, that particular one, cannot be applied regardless of what the tester shows.
TA and NS have no such tool, and for econometrics it is the standard, and extremely advanced.
Minimal quadratic error, maximum correlation percentage, everything can be tracked there. And in general, to be honest, use whatever you like.
p.s. In econometrics, I somehow remember a little bit about non-stationarity (and how much attention is paid to it). Well we are not going to make four-storey formulas to prove what is already obvious, are we?
And so you use econometrics? Show us something about how you use it?
Well, there's difficult to say which is more adequate forecasting - NS or regression
even linear regression can show better results than NS
I'd say it's not. difficult but impossible. At least because no one has seen any of the results. I'm reminded of Zhvanetsky: "... let's talk about the rise and fall of Hollywood without seeing a single film. let's talk about the rise and fall of Hollywood without seeing a single film".
Well, "even linear regression..." is simply unsubstantiated. Linear regression of what? Better than which one? What kind of result are we talking about?
Another thing I noticed. Some econometricians like to point out that they "... by numbers, not by faith". It seems that networkers, apparently, on the contrary - all of them are believers.
A little earlier, in a nod to Solar's Demi noted that "I construct a multiple regression and by partial correlation coefficients, the same thing will be determined" I personally envy. Honestly. I'll never do that in my life. But the question, "Why build a network?" is easily answered. I'll build a network and get the same thing without the slightest notion of partial coefficients.
And anyway, the topic of this thread seems to be dead. Society has steadily split into physicists and lyricists. Physicists press the lyricists with their powerful I-Q, lyricists sluggishly fight back. I want to advise econometricians, just to save their own time. You cannot prove that econometrics is better than NS. Networkers will not give up NS and will not go all at once to econometrics. So they will dig in their NS-companies to achieve the same goal as you do.
I would say, not difficult but impossible. At least because no one has seen any of the results. Involuntarily Zhvanetsky comes to mind: "... let's talk about the rise and fall of Hollywood without seeing a single film".
Well, "even a linear regression..." is simply unsubstantiated. A linear regression of what? Better than what kind of NS? What kind of result are we talking about?
Another thing I noticed. Some econometricians like to point out that they "... by numbers, not by faith". It seems that networkers, apparently, on the contrary - all of them are believers.
A little earlier, in a kick to Solar's Demi noted that "I construct a multiple regression and by partial correlation coefficients, I determine the same" I personally envy. Honestly. I'll never do that in my life. But the question, "Why build a network?" is easily answered. I'll build a network and get the same thing without the slightest notion of partial coefficients.
And anyway, the topic of this thread seems to be dead. Society has steadily split into physicists and lyricists. Physicists press the lyricists with their powerful I-Q, lyricists sluggishly fight back. I want to advise econometricians, just to save their own time. You cannot prove that econometrics is better than NS. Networkers will not give up NS and will not go all at once to econometrics. They will dig in their NS-companies to achieve the same goal as you do.
You're going in the wrong direction - you haven't seen the result, so get it. Go to EURUSD and build, for example, GBPUSD regression and VS and compare the results - what's so hard?
And what's more unclear - if you can't build the regression with the use of a statistical model (any), what's there to talk about? You can build a network, but how do you compare the contribution of each variable to the forecast? That's what private correlation coefficients were about.
You're going in the wrong direction - you haven't seen the result, so get it. For example, build EURUSD on GBPUSD regression and NS and compare the results - what's so hard?
And what's more unclear - if you can't build the regression with the use of a statistical model (any), what's there to talk about? You can build a network, but how do you compare the contribution of each variable to the forecast? This is exactly what we were talking about with partial correlation coefficients
Dear Demi, try re-reading my post a few times. Then maybe you will see the general sense. And you won't pull out the parts you like the most and build a "rebuttal" on it.
And in general, the topic of this thread seems to have died. Society has steadily split into physicists and lyricists. Physicists press the lyricists with their powerful I-Q, lyricists sluggishly fight back. I want to advise econometricians, just to save their own time. You cannot prove that econometrics is better than NS. Networkers will not give up NS and will not go all at once to econometrics. They will still poke around in their NS-companies to achieve the same goal as you do.
Anyway, I apologise if I have inadvertently offended anyone. I had no intention of insulting and offending. On the contrary, I thought there would be a powerful constructive discussion, as two sensible groups with different tools came together. No way. The human factor is more powerful. If you meet an interlocutor who is doing the same thing but in a different way, you need to urgently prove to him how wrong he is. That he's a bum, that he's floundering in shit, that he's wasted his life for nothing. To make it shorter, I am running out of patience with this stream of econometric teachings.
I wish the econometricians the best of luck and pleasure in making comparative analyses of NS in favour of econometrics.
Dear Yuri, I'd like to ask you not to use such statements in a general sense (I mean about all neural networkers). You see, neural networkers (in the general sense) quite often worry about the number of hidden layers, and also periodically the number of neurons in those hidden layers. And also sometimes there are difficulties with the choice of activation function. And sometimes you have to choose a gradient descent method as well. I'm not offended at all, not at all. But still, you've oversimplified the situation.
It's not a big deal. Usually we take one hidden layer and number of neurons in it equal to the number of grid inputs, sigmoid - hypertangent, RPROP learning method. And most often such architecture is the most optimal. Then you can adjust all this stuff by observing the reaction on forwards.
If you don't know anything about the real trading robot, you may use it as an example, for example: take a look at the market, or try to detect some strange programmatic errors, or you may be wrong. And in general, to be honest, use whatever you like.
p.s. In econometrics the discussion of non-stationarity, I somehow remember a bit (and how much attention is paid to it). Well, we are not going to make four-storey formulas to prove what is already obvious, are we?
Unfortunately, you didn't understand what I was writing about. Let me try again. In TA if there is a graph we believe there is one. In econometrics the graph drawn does not necessarily exist in reality. It just has a set of tools to distinguish phantoms from reality. And non-stationarity has nothing to do with it yet. The regression mentioned above is usually applied to stationary series.
And so you use econometrics ? Show me at least something about how you use it ?
No one will post real trading ideas on this forum, including me. I have tried to show you the tools, but to no avail. I've even written two articles. See my profile