Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 50

 
VNG:


Yeah, well, yeah, except I haven't had to wait all my life. I know plenty of people who look at such posts and only smile quietly and chop cabbage.

By the way, moskitman, I read your posts in a branch about graality, ring, in the Neuter and others. You seem to talk reasonably. No one's forcing you to take it on faith. Check it out.

Nikolai, check what? Maybe I did not search well and googled "Tactics of Adverts" along with Agni Yoga missed among the search results something profoundly important?

Then give me at least a normal link, or what, and after getting acquainted with which you can already think about TC.
So far I see only guessing at two rays of extrema. Of course, for any instrument price there will be a pair of local extrema in the history around which the current price will be.

 

Lots of material on Spider, there's a Trafaretto programme describing Tactics and Skilful on the authors' website for free.

I don't keep links, I use the search engine. Dragging it to my mink, then savoring it.

I think one of the authors will also not refuse advice, although all materials are publicly available. It is better not to ask questions at first, awareness comes slowly but surely.

 
VNG:

Lots of material on Spider, there's a Trafaretto programme describing Tactics and Skilful on the authors' website for free.

I don't keep links, I use the search engine. Dragging it to my mink, then savoring it.

I think one of the authors will also not refuse advice, although all materials are publicly available. It is better not to ask questions at first, awareness comes slowly but surely.

The site no longer exists. Closed the other day. I won't give TA advice for sure. Busy with my own affairs, and everything is laid out. The skilful builds automatically, connects different sources and generates statistics. The project is open source.

But there are some bugs in public version. So, you need to be more careful. As for the rest, there are a lot of programmers here, they will figure it out for those who need it.

 
Avals:

It's a bit complicated)) It's better to make money :)


Yeah, yeah, yeah. Give it to me and my kids.

And when they give you a fishing rod, it's complicated, better than money.

 
...:

The website no longer exists. Closed the other day.


A pity, this loss is akin to Marat's departure and the closure of the Tactics Adverse Forum site. There are few worthy and disinterested sites on the web.
 
...:

What does this have to do with the issues raised in this thread?

A medal on your chest, an orchestra to help you and go to the platform to meet the trains.

It has direct relevance, because EAs work on the basis of one of the options for analyzing statistics (time dependence) of quotes in the real market. You're right though, they talk about theory here and pay less attention to the practice of their application.
 
VNG: I read both the article and the topstarter's first post carefully. Only in the article I did not see how information theory is applied to the question under study. This is something, but not the methodology of application of TI.

It's not an application methodology, just a study that directly demonstrates the inefficiency of the marketplace. How much more straightforward I can't even imagine.

Application is a separate issue.

You have taken the quantiles from the ceiling. Answer what criterion of usefulness they are selected by, why are there 40 and not 476, what is the expediency of this choice? Probably the word "perhaps".

OK, let it be from the ceiling. That's better for me, as all the letters are now equally likely. It's a perfectly natural split, but you can choose another one. The results won't change anyway.

But I can say for sure that if you take 476 quantiles, the results are insignificant. And 40 quantiles is almost the limit for a clockwork. For H4 you can no longer - insignificant results.

It's not you who should set the breakdown into ranges, it's the market. It doesn't care what you have imagined. Therefore the price and time ranges will be different. The market doesn't care about this chart, it just draws its own chart. This is why the opening and closing prices are meaningful only in the timeframes that are higher than the days; everything that is lower is just values in the quote flow having no statistical advantage over others.

Yes I know, how many times have I seen those words...

It's just that the subject was chosen, which is quite specific, and it was shown that with such input data (returns on the selected TF - say, H4) there is a clear relationship between the data - or, in other words, market inefficiency. Varying the number of quantiles in a fairly wide range, as well as the TF, does not affect the result. Dependencies remain, and in huge numbers.
 
Mathemat:

This is not an application methodology, but simply a study that directly demonstrates the inefficiency of the market. How much more straightforward is beyond me.

The application is a separate issue.

OK, let it be from the ceiling. That's better for me, as all the letters are now equally likely. It's a perfectly natural split, but you can choose another one. The results won't change anyway.

But I can say for sure that if you take 476 quantiles, the results will be insignificant. And 40 quantiles is almost the limit for watchtips. For H4 you cannot - the results are insignificant.

Yes I know, how many times have I seen these words...

It's just that the subject was chosen, which is quite specific, and it was shown that with such input data (returns on the selected TF - say, H4) there is a clear relationship between the data - or in other words, market inefficiency. Varying the number of quantiles in a fairly wide range, as well as the TF, does not affect the result. Dependencies remain, and in huge numbers.
Alexei, and what do econometricians say?
 
...:
Alexei, and what do econometricians say?

We have only one econometrician here - faa1947. He thinks there is some novelty, but lacks rigour in the research. Better to ask him himself, though.

Some think that the applicability of the TI itself to this object is not justified. I don't see any barriers here, as there is a clear source and receiver of the signal (e.g. 238 bar is the source, 0 is the receiver). There is an alphabet that is the same for source and receiver.

The communication channel... yes, it's more complicated here. But it is there as the 0th bar is dependent on 238th!

Avals and most others think it's just daily volatility (the dependencies are clearly greater on bars divisible by 24, on H1), i.e. GARCH and the like. I disagreed with that, but no counter-arguments yet.

Then where do the dependencies on bars remote from the current one by thousands, i.e. at least a year back in the data come from?

And another thing: if the phenomenon is detected on returns, then it will also, imho, be detected on the quotes (prices) themselves. But I'm afraid it will be a false correlation. I will try to post the results for the price itself. I need to dig in the code, something does not come out the first time.

 
Mathemat:

It is not a methodology of application, but just a research that directly demonstrates the ineffectiveness of the market.

Application is a separate issue.

OK, let it be from the ceiling. That's better for me, as all the letters are now equally likely. This is a perfectly natural partitioning, but you can choose another one. The results won't change anyway.

But I can say for sure that if you take 476 quantiles, the results are insignificant. And 40 quantiles is almost the limit for a watchdog. For H4 you can't - insignificant results.

I know, how many times have I seen these words...

It was simply that the subject of the study was chosen, which is quite specific, and it was shown that with this kind of input data (returns on a selected TF - say, H4) there is a clear correlation between the data - or in other words, market inefficiency.


Yep, what's there to prove market inefficiency, how can you apply it to trading? Much less for forecasting. If I told you that ALL previous price movement has an effect on its future state, would you believe me? No, but it does. And Adverse Tactics proves it. There are also Vadimchi's Channels and Swings, for sure he knows "every pimple on the price's body" and has proven it many times on the air. The question is something else, the practical value of the research. The patterns must be studied and applied in practical trading. But these patterns must be repeatable, universal and cover the whole market for an instrument. Such bright minds have gathered here, but they have been repeating the same thing for many years.

The price movement is like this - up and down, up and down. That's the repetitive pattern. Two primitives, two related movements. Why not look at it from a statistical point of view?

Here's another pattern, Vadimchi's channel. I assure you, it works.

Here's another one of Vadimchi's swing patterns. And it works, too. Sweet as can be.

I wish we could describe these patterns statistically with formulas.