Discussion of article "Calculating the Hurst exponent" - page 2
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
...
For some reason, the author believes that this coefficient can be estimated by ISC, and that there are no other methods of estimation in this case.
...
Where in the article did you see this?
3.4 Actual values of the Hurst index for currency pairs
We have finished outlining the basic principles of the fractal analysis theory. Before proceeding to the direct implementation of RS-analysis with the help of MQL5 programming language, I thought it necessary to give some more illustrations.
The table below shows the values of the Hurst coefficient for 11 currency pairs of the FOREX market at different timeframes and number of bars. The coefficients are calculated by solving the regression using the least squares method (LSM). As we can see, formally, most currency pairs support the persistent process, although there are some antipersistent ones. But how significant is this result? Can we trust these figures?
But the point is not to justify the method of estimation, but something else entirely.
In view of your previous post.
One should be concerned with position decision blocks, not with demonstrating knowledge of the implementation of certain algorithms.
3.4 Actual values of Hurst index for currency pairs
We have finished outlining the basic principles of the fractal analysis theory. Before proceeding to the direct implementation of RS-analysis with the help of MQL5 programming language, I thought it necessary to give some more illustrations.
The table below shows the values of the Hurst coefficient for 11 currency pairs of the FOREX market at different timeframes and number of bars. The coefficients are calculated by solving the regression using the least squares method (LSM). As we can see, formally, most currency pairs support the persistent process, although there are some antipersistent ones. But how significant is this result? Can we trust these figures?
But the point is not to justify the method of estimation, but something else entirely.
In view of your previous post.
One should be concerned with position decision blocks, not with demonstrating knowledge of the implementation of certain algorithms.
It is not enough what is written. In general, the article contains a lot of surprising suggestions and even strange ones. It is not even worth discussing. You should never trust what is written. Free interpretation and word creation are in vogue nowadays. But if the code is attached to the article, we can talk about it.
Applying a ready-made function and not understanding how it counts is not interesting. I am not interested in references to r. I am interested in understanding, not in mindless use a la monkey.
Personally, the value of this article is higher for me because the code is attached, unlike your r.
It is not enough what is written. In general, the article has a lot of amazing suggestions and even marvellous ones. It is not even worth discussing. You should never trust what is written. But there is a code attached to the article - we can talk about it.
Applying a ready-made function and not understanding how it counts is not interesting. References to r I am not interested in. I am interested in understanding, not in mindless use a la monkey.
Personally, the value of this article is higher for me because the code is attached, unlike your r.
The ISC function RegCulc1000 is used to calculate the Hurst coefficient. It has a comment: //---Calculation of the Beta regression coefficient or the Hirst coefficient you are looking for
If you cannot use ISC to calculate the Hurst coefficient, and the possibility of using ISC is not proven, and analogues use non-parametric estimation methods, then all the code that "we can talk about" has no value. It is the use of questionable code that you call "I am interested in understanding, not in mindless use a la monkey".
So which one of us is a monkey?
The RegCulc1000 ISC function is used to calculate the Hurst coefficient. It has a comment: //---Calculation of the Beta regression coefficient or the required Hurst coefficient
If you cannot use ISC to calculate the Hurst coefficient, and the possibility of using ISC is not proven, and analogues use non-parametric estimation methods, then all the code that "we can talk about" has no value. It is the use of questionable code that you call "I am interested in understanding, not in mindless use a la monkey".
So which one of us is a monkey?
You have to understand the code, understand how it is counted, draw conclusions. Feel the physical meaning of these calculations. And why don't you trust this code, but you trust your black box "R"? I'm not impressed with the R at all and have a lot of doubts. It's a hell of a dump - one library pulls 100 others behind it, and you can't figure out how it's calculated and the R code itself looks ugly. And R users...
In all seriousness, I don't trust the Hurst index at all, but I would like to understand how it is calculated.
So you have to understand the code, understand how it is calculated, draw conclusions. Feel the physical meaning of these calculations. And why don't you trust this code, but trust your black box "R"? I'm not impressed with the R at all and have a lot of doubts. It's a hell of a dump - one library pulls 100 others behind it, and you can't figure out how it's calculated and the R code itself looks ugly. And R users...
In all seriousness, I don't trust the Hurst index at all, but I would like to understand how it is calculated.
You and a lot of people on this site are conflating two completely different activities:
The latter is of almost no value - it's a chore.
But algorithm development is often of great value. This is an independent problem and has nothing to do with coding. It is very often a scientific problem. People who have developed algorithms get world names.
Therefore, when we talk about any code, in our case Hirst's calculation (I'll note, the most pre-passable algorithm), under no circumstances should we mix these two different actions.
This is well understood by all R package developers without exception. Any function in R always has a reference to the algorithm it implements. You see, ALWAYS. People like you who want to get into the algorithm, and often it is very useful, can refer to the algorithm description, see its justification, criticism..... in general, to approach closely and consider from all sides, to see the application in practice ... It is the wide discussion of the algorithm as such that gives at least some guarantees that the algorithm is workable. And if you want to see the source code, R always has it.
It is a standard.
Deviation from such a standard is always fraught. If we take the article under discussion, we cannot be sure that the author calculates exactly the Hurst coefficient.
You and so many on this site are mixing two completely different actions in a bunch:
The latter is practically of no value - it is a chore.
But algorithm development is often of great value. This is an independent problem and has nothing to do with coding. It is very often a scientific problem. People who have developed algorithms get world names.
Therefore, when we talk about any code, in our case Hirst's calculation (I'll note, the most pre-passable algorithm), under no circumstances should we mix these two different actions.
This is well understood by all R package developers without exception. Any function in R always has a reference to the algorithm it implements. You see, ALWAYS. People like you who want to get into the algorithm, and often it is very useful, can refer to the algorithm description, see its justification, criticism..... in general, to approach closely and consider from all sides, to see the application in practice ... It is the broad discussion of the algorithm as such that gives at least some assurance that the algorithm is workable.
It is a standard.
Deviation from such a standard is always fraught. If we take the article under discussion, we cannot be sure that the author calculates exactly Hirst's algorithm.
I'm not confusing anything. The best description of an algorithm is code. A description of an algorithm without code is empty blablabla. If there is a code, you can understand everything by the code.
What use is a book about the Hirst index if it is poorly written and nobody has ever coded anything good with it?
I'm not confused. The best description of an algorithm is the code. Algorithm description without code is empty blablabla. If there is a code, you can understand everything by the code.
What use is a book about the Hirst index if it is poorly written and no one has ever coded anything good with it?
You know better.
Good luck
You know best.
Good luck
An extremely weak article that could have been a term paper 30 years ago.
If you read the article, the current state of affairs related to Hurst is completely left out of the picture.
For some reason, the author believes that this coefficient can be estimated by ANC, and that there are no other methods of estimation in this case.
For example, the FGN package with the HurstK(z) function, in which non-parametric estimation of the Hurst coefficient is performed, which gives a much more accurate value.
If the author had bothered to do a literature review in this area, he would not have passed by the classic paper that in particular introduces the concept of fractional ARIMA, which allows us to consider the Hurst coefficient not only as such, but within the framework of appropriate models, moreover the author would have seen that there are packages in R that have generalised the Hurst coefficient.
The Hurst coefficient outside the framework of models is of little interest and Hurst's ideas were developed within the framework of fractionally differentiated models - Fractionally differentiated ARIMA aka ARFIMA(p,d,q) models
The fracdiff package provides a fairly complete set of tools in this area.
And this is not all in the field related to the Hurst coefficient.
Once again I state that any articles in the field of time series processing without a proper review of the tools available within R look extremely ignorant with a lag of several decades
May I remind you that this is an MQL site, not an R site? And maybe you should stop shoving this into every topic wherever necessary and not necessary.
After such clever commentators a person may have no desire to write articles at all, but I am personally very interested in it.