Machine learning in trading: theory, models, practice and algo-trading - page 1108
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
Do you know Gorchakov? Again he writes some thoughts on smradlab
https://smart-lab.ru/blog/499678.php
No, not familiar, perhaps I have seen him, but I have not communicated with him. I'll read it later.
I read it. Yes, I have been to his seminars at Finam a couple of times. About 5 or 6 years ago. He used to talk all about distributions and tails.
No, I don't know him, maybe I've seen him, but I haven't talked to him. I'll read it later.
ZS I read it. Yes, I've been to his seminars at Finam a couple of times. About 5 or 6 years ago. He was going on about distributions and tails.
He says that he hasn't been able to conquer them for 70 years ) or something like that. The topic is already beaten up on this forum.
writes that he can't beat them for 70 years) or something like that. Well the topic is already beaten up on this forum
And this, in fact, is correct:
And therefore there are no similarities between different timeframes and there can't be. And what does it mean: with "one measure" for five-minute periods and daily periods, we should be very careful and distrustful of those who formulate it as an axiom.
Well, that too, by the way:
I'll add to what I wrote in my commentary: an important gnoseological conclusion: if we build trading methods on past daily or even hourly rates (if the hours have many ticks) then it makes no sense to "dig deeper" the functions of the second degree nonlinearity of price increments or increments of price logarithms (prices in the text easily change to price logarithms without losing the essence).
Well, that too, by the way:
I'll add to what I wrote in the commentary: an important gnoseological conclusion: if we build trading methods on past daily or even hourly rates (if there are many ticks in the hours), then it makes no sense to "dig deeper" the second degree non-linearity functions from price increments or price logarithms increments (prices in the text are easily changed to price logarithms without losing the essence).
I did not understand the second conclusion :) from what it is deduced
ah, like from the fact that 2 distributions sit in them?One of the very few signals I watched was yours.
In general I do not see any point in watching someone else's signals. I have no doubts that there are traders who make money. I know some of them from sitting together at auto-trading conferences and seminars. Well, what is the use of it? The speakers did not offer their own strategies. By the way, they are the most valuable. By the way, they are the most valuable.
I thought that in science, everything is the same. Scientific achievements are open - take it, use it. But the technology of their realization-application is a big secret. They are protected by patents.
So it is, who will disclose the technology, which is financially beneficial. But initially AI was not developed for stock exchanges, that's why in principle there is a lot of information on the subject and it is open, but it is all mostly theoretical in nature, which is why I personally wonder if any of those who are on the subject managed to achieve any sustainable results. I already know of one in this thread, succeeded.
So it is, who will disclose technology that is financially beneficial. But initially AI was not developed for exchanges, that is why in principle there is a lot of information on this topic and it is open, but it is all mostly theoretical in nature, which is why I personally wonder if any of those who are in the subject managed to achieve any sustainable results. I already know for sure one from this thread has succeeded.
Forget the AI. There is no AI here. And MoD (NS, RF, SVM, etc.) has nothing to do with AI. It's just a mathematical model.
And this, in fact, is correct:
And therefore there are no similarities between different timeframes. And what does it mean? And the fact that with "one measure" to the five-minute and daily periods should be approached with great caution and with distrust to those who formulate it as an axiom.
Well, that too, by the way:
I'll add to what I wrote in my commentary: an important gnoseological conclusion: if we build trading methods on past daily or even hourly rates (if there are many ticks in the hours), then it makes no sense to "dig deeper" the second degree non-linearity functions of price increments or price logarithms (prices in the text are easily changed to logarithms of prices without losing the essence).
Gentlemen, what do timeframes have to do with it? Operate the time, it will help.
About increments of price logarithms without losing any of their essence, the idea is also very interesting and thus amusing.
I do not understand the second conclusion :) from what it is deduced
ah, like from the fact that 2 distributions sit in them?Where he got that from, I don't know - I read it diagonally. But I agree with it from my considerations. In my opinion, excessive complication of the model leads to an increase in degrees of freedom and, very possibly, to a loss of stability.
In my own systems I tried to go up to 3rd order, but ended up going back to 2nd order. This one is about three years old.
Do you know Gorchakov? Again he writes some thoughts on smradlab
https://smart-lab.ru/blog/499678.php
In order to meaningfully judge the characteristics of a process with nonstationary increments one needs either an ensemble of its realizations or one realization together with a parametric model. We cannot have several realizations in principle, so only the second variant remains.
I think it is impossible to prove (or even show) the Gaussianity of the price in any way. We can only take some parametric family of Gaussian processes and study how well it fits to describe the price. For example, for simplicity, we can take Gaussian processes with independent (but different!) increments.
In order to meaningfully judge the characteristics of a process with nonstationary increments, one needs either an ensemble of its realizations or one realization together with a parametric model. We cannot have several realizations in principle, so only the second variant remains.
I think it is impossible to prove (or even show) the Gaussianity of the price in any way. We can only take some parametric family of Gaussian processes and study how well it fits to describe the price. For example, for simplicity, one can take Gaussian processes with independent (but different!) increments.
That's why he compared it with Fermat's theorem, which couldn't be proved for 300 years.)