Hearst index - page 2

New comment
 
Yuri, I don't know how to calculate the RMS, to be honest... Alas, it's not as simple a question as it would seem. After all, we don't have ticks, we have bars. And what goes on inside a bar, Allah only knows... Anyway, I've tried different ways of doing it. Nothing good came out of it. And I just never got to the regression and additive term. I shut up on CKO :( Alas, unfortunately not all mathematics fits our specifics... Maybe with Hearst too, we need to think further about how to count it in our circumstances.
 
The ERR must be calculated according to the standard formula. But to take one price as Open, High, Low or Close. For some reason, it is assumed that Open should be taken, although from my point of view it makes no difference. Any of these prices is a regular sample of a random time series. Therefore it should retain all properties of the series itself. IMHO
 
eugenk писал (а):
Oasis, I myself was inspired by the discussion https://www.mql5.com/ru/forum/50458 and tried to write it. It's not very expressive or correct. The main problem is how to calculate S, or how it can be plausibly substituted. After all, we are working not with ticks, but with bars. And calculations strictly according to definitions are hardly good here... Anyway, I am sending you my preliminary and unsatisfactory (pardon the pun) result. Perhaps someone will find it useful.

The material near the end of the discussion is truly inspiring.
This is what inspired me to find out what the Hurst index is and how to calculate it.

Thanks for the programming code of the indicator.
I will look into it now

 
eugenk:
Yuri, I don't know how to calculate the RMS, to be honest...
What a bunch of philosophers you are! Take the turkey called Standart Deviation (included in the standard MT ammo), which not only counts the FCS, but also displays it in a separate window.
 

The ERR must be calculated according to the standard formula. But to take one price as Open, High, Low or Close. For some reason, it is assumed that Open should be taken, although from my point of view it makes no difference. Any of these prices is a regular sample of a random time series. Therefore it should retain all properties of the series itself. IMHO

High and Low - no, it is not a regular sample. On a random series (where H = 0.5) for a series of High and Low, H is significantly greater than 0.5 (~0.7). I.e. in fact High and Low are autocorrelated. But you won't earn a cent on it :)

 
kniff писал (а):

...... For a number of High and Low H is WAY over 0.5 (~0.7). I.e. in fact High and Low are autocorrelated. Only you won't make a dime out of it :)



Information from a spider, in my opinion.



We'll check everything... )))
 
Oasis:


I understand the formula to be

H = MathLog(R/S)/MathLog(N/2)


Not quite.
Look at page 30 https://www.mql5.com/ru/forum/50458
There is a code of correct script there, modify it for indicator. Enter external variable N and maybe Close[i]-Close[i+1] should be taken as basis.

for(n=0; n<=N-1; n++)
{
x[n]=Close[k]-Close[k+1];
k=k-1;
}

Wouldn't say no to the final version at gorillych@tut.by :)))
 
Gorillych писал (а):
Oasis wrote (a):


as I understand the formula is

H = MathLog(R/S)/MathLog(N/2)


Not really.
Look at page 30 https://www.mql5.com/ru/forum/50458
There is a code of correct script there, modify it for indicator. Enter external variable N and maybe Close[i]-Close[i+1] should be taken as basis.

for(n=0; n<=N-1; n++)
{
x[n]=Close[k]-Close[k+1];
k=k-1;
}

Wouldn't say no to the final version at gorillych@tut.by :)))

Yes thanks, you won't find what you need there at once )))) after all, more than 1000 posts in the subject...
 

2 kniff

High and Low - no, this is not a regular sample. On a random series (which has H = 0.5) For a series of High and Low, H is WAY over 0.5 (~0.7). In fact, High and Low are autocorrelated.

Yes, that's right. About High and Low, I was in a hurry. :-)
But Open and Close are supposed to be absolutely equal.

2 Reshetov

You philosophers! Take the turkey called Standart Deviation (included in the standard MT kit), which not only counts but also displays this same RMS in a separate window.

It's nice to see that knowing how to calculate the RMS is so encouraging. However, the way it is done in Standard Deviation is not appropriate here. In Standard Deviation the RMS is calculated relative to the moving average. But for Hearst it is necessary to calculate the usual RMS with reference to the mean value on the interval of N bars. But do not be embarrassed, go on giving advice. This is a great way to figure things out for yourself.

 
Oasis писал (а):
Gorillych wrote (a):
Oasis wrote (a):


as I understand the formula is

H = MathLog(R/S)/MathLog(N/2)


Not really.
Look at page 30 https://www.mql5.com/ru/forum/50458
There is a code of correct script, remake it for indicator. Enter external variable N and maybe Close[i]-Close[i+1] should be taken as base.

for(n=0; n<=N-1; n++)
{
x[n]=Close[k]-Close[k+1];
k=k-1;
}

Wouldn't say no to the final version at gorillych@tut.by :)))

Yes thanks, you won't find what you need there at once )))) after all, more than 1000 posts in the subject...
There's a question about its correctness:



although with x[n]=Close[k]-Close[k+1] - it turns out 0.4674, which already seems to be true
123456789...46
New comment
Reason: