Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 64

 
Mathemat:

It apparently claims to be the only truly scientific way of knowing the market.

P.S. A little later I will try to post the results of a similar study of prices rather than returns. Only it will be there with chi-squared, not in TI language.

If interested, let's discuss it all in private.
 
Mathemat:

Don't care about the structure of movement within an hour. It is a minimal atom, an indivisible part of the model.

Discretisation... OK, which one are you suggesting?

These are scale invariant networks. Our result shows that there is no scale invariance.

No fractality as a consequence either. This is an EWA myth that is not supported by anything.


getch 01.04.2010 11:03

Repeat:

EA counts the number of ZigZag knees (at least Pips) and writes to file:

EA text

Plot as a function of knees count against their min. size (Pips)

http://www.cognitivist.ru/er/kernel/prologi_02_signature_scalefree.xml

Perhaps the most essential signature of the cognitive order is scale invariance, that is, the fractal structure of the phenomenon.If the object or phenomenon we study has scale invariance, then this is a good reason to look for the effect of cognitive order regularities in it. On the contrary, if the phenomenon under study has no scale invariance, this suggests that it is governed mainly by laws of the physical order.
It is convenient to say that the signature of the cognitive order is the absence of a characteristic scale in the phenomenon, and this distinguishes it from manifestations of the physical order, in which there is almost always a characteristic scale setting a typical spatial size or other magnitude. This difference is easily captured in the statistical parameters of observed objects or phenomena.

The absence of a characteristic scale in the parameters of the phenomenon is a signature of the cognitive order controlling this phenomenon.

 
Mathemat: P.S. A little later I will try to post the results of a similar study of prices, not returns. Only there will be with chi-squared, not in TI language.

Bullshit, if you'll pardon the expression. Frankly, that's what I expected.

The zero bar turned out to be extremely severely dependent on all previous ones - not only on price, but also on a pure Wiener process (as pure as it can be based on the PRNG in MT4), in which the dependences should in principle be no more than a significance level (0.01). The chi-square criterion turned out to be hundreds of times higher than the boundary criterion. This is if you process the values themselves, not their returns.

There should be no dependencies, they are false on a Wiener process. Roughly speaking, you can't examine I(1), you have to examine its differences I(0). Of course, cotier returns are not I(0), but they are not too far from it either.

By the way, the returns of the usual Brownian motion show dependencies just at the level of significance 0.01, which is quite natural.

In short, everything is as it should be. The object of the study is correctly chosen - quotes returns. We did not miss the detrending.

 
Mathemat, what do you mean by quotes returns? Delta between quote 1 and quote 2, or what?
 
x4x:
Mathemat, what do you mean by quotes returns? Is it a delta between quote 1 and quote 2, or what ?

Return[i] = return[i] = close[i] - open[i] (mine).

If correct, it should be return[i] = close[i] - close[i+1].

In the first formula, I simply removed the gaps. But it's not too important really.

The tool is the same everywhere.

 
Ahhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh, the returns are cheatily. That's right returnees are more profitable to analyse. Does it make sense to analyse these returns return[i] = close[i] - close[i+1], relative to high[i] - low[i+1] Potential return on the instrument-https://forum.mql4.com/ru/22251
 
Also, it is probably more advantageous to take returns of equal-volume bars (by the number of ticks, for example).
 
More accurately, it's not. The analysis is not of returns, but of median line differences (((high+low)/2)-((open-close)/2))
 

Hai da lo... Nah, it's going to be hard to work on a zero bar. Don't confuse simple with complicated.

And here it's simple - opened at 14:00 and closed at 15:00, no problem. Or waited and continued, if forecast minus the first bar in the same direction.

And what highs and lows there will be - bullshit.

 

Vladimir? Have I offended you in any way? )))