From theory to practice - page 14

 
СанСаныч Фоменко:

For some reason there is no response to my link, which has everything on this thread chewed up and much more.

Why not, I downloaded it. Then I compared the title, title page and length of the pdf file with the one in the article recommended two days earlier(by Yury Kirillov https://www.mql5.com/ru/forum/221552/page3#comment_6145613). I found a complete match and did not reread it. There was no reaction to Yury Kirillov's post either. Except that I've found and downloaded about 30 other sources on related topics, including the PhD dissertation of one of Orlov's students, where the information should be more than in this preprint. I can't say that I can recommend these sources to forum members even with an engineering background, sometimes it is the mathematical one that is required. Often quite naturally the authors of these articles go to spaces with a probabilistic measure, which is not part of higher mathematics for engineers.

От теории к практике
От теории к практике
  • 2017.12.02
  • www.mql5.com
Добрый вечер, уважаемые трейдеры! Решил было на какое-то время покинуть форум, и сразу как-то скучно стало:)))) А просто читать, увы - неинтересно...
 

I got a little excited about the conclusions about the ticking returns. It's not a series there - the timeline is a curve. Alexander said it right: you need to take prices with some periodicity. I will think about it.

I see 2 ways:

  1. Averaging and creating a second timeframe;
  2. to take price by timer.

 
Dennis Kirichenko:

SanSanych, of course! I'll try to submit the details in the next few days. I will be happy to discuss and listen to comments and opinions.

What revolution and why the surprise? Returns tend to show stationarity.


Where did you get this information from?

ARMA models have been refined to ARIMA, in which I just means "incremental", in the terminology of the model - differentiation. The use of ARIMA in practice is extremely rare precisely because of non-stationarity and that is why GARCH models have appeared in an uncanny variety. The reasons for the emergence of these models for increments, usually log(p1/p0), are:

  • remains a trend in the increments
  • variable variance with various additional effects
  • The law of the incrementdistribution is NOT normal; it varies greatly and changes as the window moves.


Perhaps the most advanced package in this area rugarch explicitly requires specifying modelling of all these three components, with AFRIMA instead of ARIMA, where the differentiation can be fractional to simulate Hearst-adjusted series. My attempts to use this package have so far failed precisely because of the non-stationarity of the increments - all the time there are parameters that are NOT significant.

 
Renat Akhtyamov:

San Sanych.

Honestly - I did.

//At least there is no error in the material from the beginning as the topikstarter. And I said to the topik - not 0.05, and 0.5...., see if you're writing bullshit.

But the thing is, you have to think 100 times about the fact that the market is chaotic and you can't predict it mathematically.

The simplest example - when will there be a reversal?

It is impossible to make a forecast because the market has no average because it is either flat or trending and the average is strongly descending from its average state.

//I just have this opinion for some reason and that's all

This is when trading trends.

When trading increments this problem does not exist, but there is the problem of gaps, or more precisely structural changes, when the price changes not only unpredictably by leaps and bounds, but it may well change all the statistical characteristics of these increments or may not.

 
Alexander_K:

...Reading all the ticks in a row is a road into the abyss, from which you can only escape with an empty wallet. But an empty wallet is half the trouble, an empty soul and head is the trouble. I was philosophizing again - what is it? :))))))))))).

...But, I repeat - reading all ticks in a row, with unknown period between data - it's a completely unmodelable situation.

And why should it be modelled? An arbitrage situation has appeared - let's withdraw the profit. Of course, by reading any tick as soon as possible, without waiting for the end of a round second. Is the goal to simulate? No, the goal is to take profits.

 
Alexander_K:

Yes, and if you suddenly started to doubt the quasi-stationarity of the returns, you shouldn't. This thing is one of the keys to solving the problem. To admit that returns are completely non-stationary is to sign oneself helpless before Forex and just catch luck.


And here the man uses the same Fokker-Planck and considers it completely different: the merit of this method is its work on non-stationary data.

 
       - Comrade scientists! Associate professors and candidates! 
	You're tired of the X's, you're confused by the zeros! 
	You're sitting here, decomposing molecules into atoms, forgetting that potatoes decompose in the fields. 

	You try to extract the balsam from the rot and from the mould and you extract the roots ten times a day. 
	Oh, you'll add to it! Oh, you'll get the potatoes rotten and mouldy on the roots!
... (V. Vysotsky)
 
Alexander_K:

By the way, this work is VERY close to what I do. And they all do it right. They are only wrong about one thing - they think that returns are independent and non-stationary. AND THEY ARE NOT!!! They work with parametric statistics and this misleads them. They are definitely strong in analytical calculations, but in numerical modelling they are obviously just trusting this work to students, while they should do it themselves. That's how it is!

The authors of this paper, if they read this forum, I would advise, as an elective, to attend lectures on theoretical physics, specifically - quantum mechanics. They will be quickly taught to think a bit outside the box there :))))

Apart from bragging of nobody's diplomas I have not seen any proof of stationarity of increments and I have not seen any definition of this very stationarity.

 
СанСаныч Фоменко:

Apart from bragging about the stationarity of the increments, I have not seen any evidence of stationarity, nor have I seen any definition of stationarity.

I could write an article on the materials of this thread - Non-stationarity and non-normality of the distribution as a bugbear).

Signals and noise have never been stationary or normal; however, radio engineering and signal theory coped very well with their processing in the era of analogue technology. Even in the past they were successfully taken out and restored on old BESMs and ECs.

Literature on these issues is abundant - on methods in radio engineering, image processing, geophysics, astrophysics, etc. I can point out the literature - the books are on the shelf).

 
Alexander_K:

By the way, I was thinking about this... If the hypothesis about quasi-stationarity of returns is true - and it is true, it cannot be otherwise - then the way you described, Vladimir, is certainly the simplest way of making profit.

But we don't look for easy ways, do we? :))))))))))

Alexander, arbitrage was described a long time ago and not by me at all. In 2008-2014 was the heyday of this method. Since then many things have changed, there are companies that make additives for MT4 and MT5 servers to oppose arbitrage and scalping. The leading DCs themselves have learned to counter arbitrage, and ahead of these companies. So the method has already faded. Particularly because of the active move by the brokerage companies from providing accounts with Instant Execution to providing accounts with Market Execution, where the client is in no way able to contest arbitrary slippage in the execution of a trade order.