Zero sample correlation does not necessarily mean there is no linear relationship - page 16
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
The use of a sliding window is a basic principle of tehanalysis, nobody counts anything on all the data because it is not realistic in principle. It is the same with DSP.
Your autocorrelation function shows several correlation values with different parameters (window offset, window length (or something like that, I did not get into details)). It also uses a sliding window. The function draws autocorrelation values for one data point, no one prevents it from moving the window and calculating values for each bar, but a three-dimensional plot must be drawn.
The definition of autocorrelation can be obtained from Yandex. Everything is much simpler than it seems.
I will not prove and argue, as it is useless, just take note.
IS THAT CLEAR?
is that clear?
There is a category of people who do not push the idea, but themselves, hotly beloved. The more delirious the concept, the more successful is the process of self-promotion. hrenfx has thought up something (quite possibly valuable) and this thought up has called widely known and established terms. It is impossible to explain to him that this is not allowed, because he is advertising himself, and the whole branch has nothing to do with correlations. Hrenfx writes about himself and the rest of us write about correlation, why did we get caught up in this? Only because he wasn't banned in time with the admonition: "Read some books".
1. you take a piece of 100 bars and compare it to a piece of 100 bars. there is no other way. QC cannot be calculated on arrays of different lengths.
don't blink. for all 100,000 bars. Spell it out ACF is a comparison of BP to itself, not to a piece of itself. It's with HIMSELF.
What 100 bars or 100 thousand is still a sample, not the whole BP. And here, it is up to someone to decide which length of sampling to use. The result is the same - autocorrelation on a sample, though numbers can be very different.
About the header - the correlation on the sample does not tell us much. Correlation is zero only for independent infinite stationary series, i.e. it is an abstraction which cannot be seen in real life, but one still needs to know.
The function draws autocorrelation values for one data point, no one is stopping anyone from moving the window and calculating values for each bar, only a 3D graph will have to be drawn.
I wrote a script that prepares the data for Mathcad for 3D visualization. Script and Mathcad file attached.
This is the appearance of QC changes for EURUSD and GBPUSD since the beginning of October:
Your autocorrelation function shows several correlation values with different parameters {...}
I would only agree that it shows something incomprehensible :-) /from lack of practice in DSP
The use of a sliding window is a basic principle of tehanalysis, nobody counts anything on all the data because it is not realistic in principle. It is the same with DSP.
Your autocorrelation function shows several correlation values with different parameters (window offset, window length (or something like that, I did not get into details)). It also uses a sliding window. The function draws autocorrelation values for one data point, no one prevents it from moving the window and calculating values for each bar, but a three-dimensional plot must be drawn.
The definition of autocorrelation can be obtained from Yandex. Everything is much simpler than it seems.
I will not prove and argue, as it is useless, just take note.
Roger that.
Here is more detail, it may not be three dimensional, but it is exactly what you say
https://www.mql5.com/ru/forum/105740/page5#50590
We were building this ACF.
composter, Candid, the mathematician kept an eye on us there further down the branch and checked it through FFT . Of course it depends on the size of the window (sample) and the shift
https://www.mql5.com/ru/forum/105740/page16
If anyone is interested they have the indicators there.
Dmitry, I'm not an idea killer. I'm against the use of terms that are well known and yet have a different meaning (other mathematics). That way...
we don't understand each other. After all, more often than not, it's this misunderstanding that makes everything happen.
Judge for yourself, he wrote the indicator and added it to codebase and showed that there is a correlation. Many thanks to him. I also did something similar https://www.mql5.com/ru/forum/107695 the correlation at 24 hour lag. It has been two years and this correlation still exists. I have noticed in the morning flat breakdown that many people use this idea.
Is it bad? No, it's great, it's perfect. But you can't put everyone on the forum in passing (including Pirson) he's the only one who gets it right and understands it, while we're all clumsy.
He has accused everyone, hence you, of never having calculated the CC, that you have coded it incorrectly, and if you have coded something, you have applied it incorrectly ... I am against it, you cannot do it that way.
Z.U. and I'm pissed as hell. Just to put the matcad (and show me what's wrong), I tore down Windows 7 yesterday, but forgot that MT5 stores everything on disc C by default (although it's on D)... 4 months of work, wasted, unformat didn't help, and no copies... ((
That's the thing, it's not its autocorrelation function :-). There is a clear definition for ACF.
Except that it shows something incomprehensible :-) /from lack of practice in DSP