Hearst index - page 21

 
Vita писал(а) >>

Here's a variant in which the error is counted. Unfortunately, I can't find where I stole the C source of this wonder, but it claims to count by Feder E. Fractals. For him Test H=0.6807 for the same file. Seems like it's not bad.

For 78 values it's the hardest. A lot of work is devoted to how to estimate Hurst on half a hundred observations. Even without understanding the calculations, you get very different results from one author to another. There's nothing surprising about that. As many algorithms as there are indicators :). Oh, and another problem - in attached version on 1000 observations with error taken into account we can't say anything about the price - is it consistent or not at the moment, because 0.5 lies just between error channel (red lines at cRSGraphic=false).

The input is either the price difference or the logarithm of the price ratio.

Which time series model are we talking about? In its classical form(Box and Jenkins) BP consists of regular and noise components. Taking the differences has quite definite goals, not "just chance" to get something.
What goals do you have by taking the first differences? In general, can Hirst be calculated for all Box models?
 
Yurixx >>:

Вы приложили файл brown72.txt. Однако, Ваш индикатор тестирует на файле brown72.csv. За неимением других инструкций я просто переименовал его и положил в папку \experts\files. Вот результат:
На Н1:


На тиках:


Ваш файл содержит 1024 значения. Вот первые 4 из них:
45.47422
42.55601
46.5188
41.61502

Quite right. That's right.
That's the problem. I have to think about why and how to solve it.

 
faa1947 >>:

О какой модели временного ряда идет речь? В ее классическом виде (Бокс и Дженкинс) ВР состоит из регулярной и шумовой компонент. Взятие разностей преследует вполне определенные цели, а не "авось" получим что-либо.
Какие цели Вы преследуете, беря первые разности? Вообще, для всех ли моделей Бокса может быть посчитан Херст?

Suppose we want to know if there is a long-term memory in price movements. I think it's appropriate to feed the algorithm's input with the actual movements - the first differences. What do you want to feed to the input of the enclosed implementation of a particular algorithm?

 
Vita писал(а) >>

Suppose we want to know if there is a long-term memory in price movements. I think it's appropriate to feed the algorithm's input with the actual movements - the first differences. What would you like to feed to the input of the enclosed implementation of a particular algorithm?

>> I don't know. For a start, you need to identify the BP model. There may be trends, cycles, noise, with parameters ranging from operational to those in which the model is not workable. Box considers several models with a different set of parameters. If you take at least what Box has identified and calculate Hurst for them - will it be the same Hurst or will it be different, with different values or algorithms.
I've seen attempts to calculate Hearst on several forums and in the literature. None workable. This has led to the above thoughts.

 
Good afternoon, I read this thread with great attention, as I am interested in this topic, although I am still a novice in these matters. During my research of Hurst indicator I am interested in realization of the following task: it is necessary to determine the effectiveness of iVAR indicator, _hurst_classik for a financial series.
My vision of the realization of this task is as follows: I need to make an indicator that calculates the distance d1(number of bars) between two adjacent points (minimum and maximum) based on ZigZag indicator data and get the distance d2 (number of points within the corresponding interval) that gives iVAR indicator (set of values less than 0.5 in case of variation index) and _hurst_classik indicator (set of values greater than 0.5 in case of Hurst index). The final result is an array of d2/d1 ratios. The final result is presented as a histogram.
I hope there are some MQL4 programmers on this site who will help this girl in her research, I will be glad for any help! Thanks a lot earlier!
PS: although the ZigZag indicator is a trending one, maybe some kind of software solution with flat. If there are any ready-made tools to solve this problem, please specify them. In addition I am re-posting the codes of iVAR and _hurst_classik indicators.
Files:
 
Index of variation
Files:
ivar.mq4  4 kb
 
_Forex19_ >>:
Надеюсь на этом сайте есть джентльмен - программисты на MQL4,которые помогут девушке

Madam, there are gentlemen here, and even more than one, and a girl-researcher of unexplored Forex will surely help her in the best way.

But Madame must show some interest in the final result and not only in words. Isn't it so? :)

 
Dear Gentlemen, I'm not very good at MQL4 programming, I can put my vision of the algorithm for this task into words and I would be very grateful for help in realizing this task :)
 

Visions and programming are incompatible. :)

In the next branch there are programs for drawing block diagrams, you can first draw a complete and detailed block diagram of your algorithm - the one you want to program. And then the code will be quite close.

 

Please tell me how the length of the VR to be analysed is selected in RS analysis.