Zero sample correlation does not necessarily mean there is no linear relationship - page 5
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
This is the right thing to do. you have to get to the bottom of it. You can't criticise them straight away. What, for example, the same Pearson you have failed to apply.
I don't apply Pearson anywhere. It simply did not exist in MQL4 in a correct form. Now we have it.
Doesn't mean Pearson is lying at all. A formula can't lie, it's just a formula... maybe you're just trying to misapply it. Or you have too high an expectation for it. Pearson's got nothing to do with it. He's good. He wrote the formula. A lot of people use it... thank you.
When one considers correlation - that's one thing. But when they start talking about linear correlation - that's another thing. There are plenty of places where people write that supposedly zero sample correlation is the absence of linear correlation. Not only is that not true. It's also that people don't understand what a linear relationship is. The example thread warns you not to take our word for it.
Z.I. about matkad. look for it there for sure is (AKF). unfortunately, on this Windows 7-ku can not put matkad. will soon be demolished. will put. can send a personal file. where I did all the checks.
Sergey, thank you for paying attention to my remark :-),
to clarify: I wrote "how I would interpret the word "autocorrelation" :-).
Such, you know, naive approach - when you immediately understood what was meant by
and you don't care what is actually meant.
:-)
look again at the formula https://ru.wikipedia.org/wiki/Автокорреляционная_функция ACF depends only on tau, on bias, no window there.
If you enter an additional variable N, it means that for the same data set, say 1 2 3 4 5 6 7 8 9, different ACFs can occur, depending on the selected N. This is wrong. One dataset - one ACF, another dataset - another ACF, etc.
Fundamental error. This ACF of a random variable with known variance and expectation is a theoretical definition.
In practice, it always refers to a sample. Sample autocorrelation is defined by sample size (window). There is no single sigma, but sigma(t) and sigma(t + Shift). And the sample autocovariance is divided by their product.
This is very important to understand:
A little literacy.
Another common misconception is to confuse the concepts of "correlation coefficient" (i.e. a characteristic of the stochastic relationship between s.v.) and "sample correlation coefficient"(an estimate - one of many possible - of the true SC). Actually these are completely different things, and substituting one for the other is fundamentally wrong.
The fundamental error. This ACF of a random variable with a known variance and expectation is the theoretical definition.
In practice, it always refers to a sample. Sample autocorrelation is defined by sample size (window). There is no single sigma, but sigma(t) and sigma(t + Shift). And the sample autocovariance is divided by their product.
So you want to prove that there can be different ACFs for the same dataset. ACF by the way can also be calculated through the Fourier transform. I will soon install Matcad and prepare all methods of calculating ACF (built into Matcad, through the Fourier transform and through the formula given in the indicator).
I.e. you want to prove that for the same set of data. can be different ACF. this is not true. ACF by the way can also be calculated through the Fourier transform. I will soon install Matcad and prepare all methods of calculating ACF (built into Matcad, through the Fourier transform and through the formula given in the indicator).
You have a fundamental misunderstanding of the concept of estimation in a sample.
No one knows the true variance and matrix expectation of EURUSD. And you make the calculation as if you know these values. And you also calculate through a linear regression model.
Apparently, autocorrelation as well as correlation can be implemented in the form of an indicator. This is a resource-intensive task that requires serious optimization. Not 20 lines of code.
And another huge fundamental error in calculations of the correlation (auto or overall) - is the use of absolute values of prices of financial instruments, instead of relative ones. You have to do logarithm before calculating the price series correlation.
If you dissect a row, you can draw an abyss of ACF.
One dabbles in smoothing with a Parzen window or something...
Another one subtracts a linear regression.
Or did you mean something else?
Фундаментальная ошибка. Эта АКФ случайной величины, у которой известна дисперисия и мат.ожидание - теоретическое определение.
In practice, it always refers to sampling. Sampling autocorrelation is determined by sample size (window). There is no single sigma, but sigma(t) and sigma(t + Shift). And the sample autocovariance is divided by their product.
It is very important to understand:
By window, what do you mean? The sample size...? :о)
So there's a lot of time series on minutes.
;)
By window, what do you mean? Sample size...? :о)
You have a fundamental misunderstanding of the concept of estimation in a sample.
No one knows the true variance and matrix expectation of EURUSD. And you make the calculation as if you know these quantities. And you also calculate through a linear regression model.
Apparently, autocorrelation as well as correlation can be implemented in the form of an indicator. This is a resource-intensive task that requires serious optimization. And not 20 lines of code.
And another fundamental error in calculations of the correlation (autocorrelation or overall) - is the use of absolute values of prices of financial instruments, instead of relative ones. It is necessary to do logarithm before calculating the price series correlation.
I'm afraid you don't fully understand this. There is a built-in function in the matrix package. I don't care what it is, you can logarithm it or not. The output is ACF. I'll do what I promised. I'll show you that they all match. You can double-check everything. Then we'll talk. Right. Wrong. Now it's just words. There's a code on my end in code byes. I double-checked it. But it's very important to me. I'll do all the checks and I'll post it. Not to prove anything to you. What's important to me is that I really got it right from the mathcad. If you find a mistake, I'll be glad. I really am. Because all my adaptive algorithms are based on ACF, they don't have any input data, everything is taken from ACF. That's why it's so important to me...
You have a fundamental misunderstanding of the concept of valuation in a sample.
And another huge fundamental error in calculating correlation (auto or overall) is to use absolute values of prices of financial instruments instead of relative ones. One should do logarithm before calculating price series correlation.
A lot of fundamental mistakes...
Have you forgotten that forex cannot have zero or infinity in "values of financial instruments"? DDD
here prices are almost always relative.
It's not commodity prices or stock curves.
;)
Have you forgotten that forex cannot have zero or infinity in "finance tool values"? DDD
prices here are almost always relative.
it's not commodity prices or stock market curves.
You are being told about the proper preparation of the price BP for correlation estimates. And it doesn't matter what market the financial instrument belongs to. It is, indeed, fundamental.
It should be understood that it is a global mistake to assume correlation for EURUSD and USDJPY without logarithms.