How can I tell the difference between a FOREX chart and a PRNG? - page 27
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
It may be in a couple of years at the earliest. It will be more useful for the forum. They will think more.
I'm sorry. I don't give a shit about your arrogant high-mathematical speculations at all.
Pearson's (linear) correlation in the context of quotes has probably been said all along. It's useless, what else is there to talk about.
Only non-linear correlations may be of real practical interest and no one here really talks about them. For they are too complicated and unexplored. This is information theory, chi-squares and other incomprehensibilities for the vast majority.
The topic was brought up before, but it was bluntly reduced to volatility clustering, i.e. to (G)ARCH. This is far from all, there is something more. Maybe semi-invariants will help here, and maybe something else.
P.S. Come back whenever you want, Alexey. Even though you talk a lot of crap, it stimulates the brain.
No offence meant, it's not about the tool at all, it's about how it's used.
I'm not offended. It's just that it's always scary, really, that there's a pro who'll dig something up and poke his nose in, by rights. :)
Why argue about who is cooler, the explanation is quite simple - the original signal is a segment of a sine wave in a rectangular window, its ACF is also a segment of a sine wave, but in a triangular window, i.e. exactly what we see in the second figure. This can be checked by elementary calculations. If we take a sinusoid unbounded in time, its ACF will be the same sinusoid. Conclusion 1: The mathdeck calculation is correct. Conclusion 2: if we calculate in this way the sampling ACF (and not the actual ACF, which we will never know) of the real signal, we must remember that the calculation is done in the window, and therefore the result is always distorted.
With all due respect, the ACF is defined as the dependence of the ACF on the distance between samples, so the difference is not so fundamental. And the classical formula itself (which, as rightly pointed out above, does imply at least stationarity of the process in the narrow sense, plus its ergodicity) confirms this.
This is nice and more correct. The difference is in what is compared to what. When calculating the ACF two different data sets are compared. When calculating ACF in the first step, the array is compared to itself (which is why in zero ACF = 1, the arrays are completely the same). At second step, the array is shifted along time axis and compared with initial one, and so on, until there is no sense to shift it anymore, the array has gone beyond the first ACF=0.
ACF is defined as the dependence of QC on the distance between samples, so the difference is not so fundamental.
I would say that ACF is a function of array shift (tau) relative to the first, not a function of distance between samples (distance between samples, usually a constant).
The point is different. I've given a formula, made an indicator and laid it out in codebase. But they say that it's not calculated correctly, sort of need to "tidy up", that there is a more correct calculation ... that has robustness properties, non-parametric ...
I'm asking you to tell me where it's wrong, what's the difference? show a better, correct.... it's just a formula, you take it and calculate it like MA. But how do you use these results and calculations ... You need to understand what you're calculating for.
In my personal message (and on the forum) they have written many things, like I am a moron, my indicator is dumb, I am dumb, I don't know mathematics, and I am not able to program, the indicator always shows one on the 0th bar, it's impossible to trade with it.... what should I say to them? I want to cry from illiteracy ... you don't even have an academic approach to analysis ... everyone is interested in when to press the button and which button...
he indicator always shows one on the 0 bar, it is impossible to trade with it..... what should I say to them? It makes me weep with illiteracy... you don't even have an academic approach to analysis... everyone wants to know when to push the button and which button to push...
People who understand ACF will not take it out of the kodobase, because ACF (a) must be accompanied by additional information, (b) in itself is of no special value, because you have to use it with other instruments, which are not in the kodobase. Therefore, by placing it in the kodobase, you have intended it precisely for citizens with open beaks, future billionaires.
As for the academic approach, you are wrong. There are such people on this forum and there are quite a few of them. And you cannot get out of a systematic error of the first kind: solving a wrong problem with the right methods. And you do not take criticism of these people who understand it.
Sorry to be blunt.
1. It's not my formula. Don't attribute it to me. I got it from textbooks and math packages. I didn't make it up. It's exactly the same on the wiki. The formula matches 100%. What's to clean up?
2. The picture you cited was mine, where I showedhrenfx what the differences were.
3. Yes, that is exactly how it turns out, and I would like to point out that it does not happen with me. And MathCAd, add MathLab here and it turns out exactly the same way, because lcorr(Y,Y) is a function built into matcad, I have not programmed it and did not invent ... (Anyone who knows Mathcad can go check it out) Do you honestly believe that both of these mathematical packages do not calculate ACF correctly?
4. give me the formula. I really want to see robust and also non-parametric...
1. Yeah. Yeah, you'll leave the forum here, how come......
As the State Duma deputy Maria Kozhevnikova says, "THIS IS BLEEP!"
Privalov, autocorrelation is a dimensionless quantity that shows the FUCKING FEATURES of a function to itself. The autocorrelation of a periodic function is also a periodic function.
The autocorrelation of a sine is COSINUS. The autocorrelation of cosine is COSINUS.
http://sfprime.net/lls/pcs.htm
The autocorrelation of a sine wave is a cosine waveshape [REF10].
10. Applied Fourier Analysis, Harcourt Brace College Outline Series, Hwei P. Hsu, Harcourt Brace College Publishers, New York (1984). ISBN 0-15-60169-5.
I can give you a dozen more references. Need one?
And according to you (and Mathworks) it turns out that a piece of sine at 0 is like a piece of sine at 200 a THOUSAND TIMES MORE than a piece of the same sine at 200,000, right?Privalov, this is 7th or 8th grade high school.
The formula on wikipedia is NOT the same, it's just normalized there (n-k) so that for different lags you get comparable numbers. Then there is ONE average mu-small in wiki, while you have many, many mu-small in your formula, and all some kind of indexed. What is it?
2. You got it wrong.
3. Yes, they are idiots. A bunch of underachieving physicists who didn't become physicists, so they decided to write mathWorks software in Fortran.
Here's a link where mathWorks staffer - when asked why their ACF is dumped, i.e. fading, says that it's produced IN WINDOW, and therefore the longer the period of ACF being tested, the less samples are left and therefore their ACF is always fading.
http://www.mathworks.co.uk/matlabcentral/answers/36882
-------------------------------------------------------------------
2 Comments
Thanks so much for your input, Sir.The result is what I expected.
I do have I last question,though: why does the autocorrelation function flatten out at the beginning and end of the period?
Because the summation necessarily involves fewer and fewer terms as the lag increases. Think about shifting one finite length vector in respect to itself, the greater the shift, the less the overlap and therefore the fewer products there will be in the sum.
-------------------------------------------------------------------
Privalov, just stop praying to MATLAB, it will take you out of some misconceptions. Why the hell should I blindly believe a bunch of unknown to me, unnamed physicists who wrote this clanking behemoth?
4. Why, are you getting lazy? Here's a link to the Russian section of Wikipedia:
https://ru.wikipedia.org/wiki/%D0%9A%D0%BE%D1%80%D1%80%D0%B5%D0%BB%D1%8F%D1%86%D0%B8%D1%8F
This, Privalov, is correlation. It is calculated here in non-parametric ways. You can calculate autocorrelation the same way if the SECOND function is the FIRST function, only with a shift. All known correlation calculation methods, of which there are dozens, apply to autocorrelation as well, since autocorrelation is just a special case of correlation.
Shit, colleagues, well let me leave this forum - to work for myself, please don't be stupid. This limelight on a level playing field bores me. All right, if it was something complicated, but it is elementary. Although, .... If MathWorks has been dumb for so many years, what's to ask of the rest of us.
It's strange to hear you say that. Do you really believe that ranking really does not take absolute values into account in any way?
The main requirement for non-parametric methods is robustness to "noise" and distributions (especially fat tails). This can be achieved at the expense of accuracy, which is often elusive and misleading.
What's the problem with Matcad? It counts what it is given as input. In this case, two samples of 1000 points each are shifted - of course, with a full shift, the data does not overlap, there will be nothing to compare.
Continue the first sample up to 2000 points, and there will be no fading.