Hearst index - page 34
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
maybe we should draw phase trajectories and look visually at the attractor? If there is indeed chaos, it should, as far as I remember, turn out ribbons... You can also calculate its (attractor's) dimensionality.
Yes, in principle, you can see it that way. Moreover, I'm not very good at phase space.
Here, I shuffled the last 1,000,000 bars of eurusd and added it to the chart: the result is excellent, the series is not different from random. I do not understand why it does not work with RTS. There must be some specific features that I do not understand:
I still don't understand why this doesn't work with the RTS. Apparently there are some peculiarities that I don't understand yet:
I was interested in the Hertz ratio and calculated it using the Matlab package for RTS minute and hour bars.
I got a paradoxical result: if I calculate the coefficient for open or for close it equals 0.5 and has a normal distribution for all chunks, while if I calculate it for low or high bar it equals 0.6 and the distribution is shifted. Who can comment on this fact? I found an article on the Internet where the author calculated for currency pairs and found the same pattern for low and high it is higher than for open and close bars. The result is the same for both one hour and one minute bars.
I was interested in the Hertz ratio and calculated it using the Matlab package for RTS minute and hour bars.
I got a paradoxical result: if I calculate the coefficient for open or for close it equals 0.5 and has a normal distribution for all chunks, while if I calculate it for low or high bar it equals 0.6 and the distribution is shifted. Who can comment on this fact? I found an article on the Internet where the author calculated for currency pairs and found the same pattern for low and high it is higher than for open and close bars. The result is the same for both one hour and one minute bars.
If you calculated the Hurst index using the Mandelbort-Peters formula, you would not be able to obtain the value 0.5 because it does not converge to 0.5 in principle. Taking high and low instead of close will certainly increase the index, but not significantly as the range will increase and the standard deviation (calculated by close) will remain unchanged. The increase in the range from 0.5 to 0.6 is excessive and only indicates a possible error in your algorithm, as well as the convergence of the index to 0.5.
Interested in the Hertz indicator, calculated it using the matlab package for RTS minute and hourly bars.
I've got a paradoxical result: if I calculate the coefficient for open or for close I get 0.5 and have normal distribution for all bars, while if I use low or high bar I get 0.6 and the distribution is skewed. Who can comment on this fact? I found an article on the Internet where the author calculated for currency pairs and found the same pattern for low and high it is higher than for open and close bars. The result is the same for both one hour and one minute bars.
I think there are errors in your calculations. Just in case, if you calculate by the traditional formula (given in the initial posts), you should use about 2000-3000 bars to get sane values.
Interested in the Hertz indicator, calculated it using the Matlab package for RTS minute and hourly bars.
I got a paradoxical result: if I calculate the coefficient for an open or close bar it is 0.5 and has a normal distribution for all chunks, while if I calculate the coefficient for a low or high bar it is equal to 0.6 and the distribution is shifted. Who can comment on this fact? I found an article on the Internet where the author calculated for currency pairs and found the same pattern for low and high it is higher than for open and close bars. The result is the same for both one hour and one minute bars.
I think it's because these bars are temporary; try doing it on equal-volume bars of 500 ticks or 1000 ticks. But probably there will also be a high-low bias (high-low bias))))) thick tails will show up. Already checked on the forums that the ticks are closer to a normal distribution. And maybe try to build equal-volume bars not by the number of ticks, but by the dimension of the diagonal line, that is, the TF by fixed lengths of line C, in fact, then there will be no hilo)))) and there will be only 2 points of open and close, But it is possible to build higher prices not also arbitrarily diagonally from the cathetuses of ticks and points as in figure 1, but from bars in figure 1, that is, instead of the number of ticks there will be the number of bars in figure 1, and the vertical line will also contain ticks.
Some kind of alignment of looseness between fractal structures.
I think it's because the bars are temporary, try to do it on equal volume bars of 500 ticks or 1000 ticks. But there will probably be a high-low bias as well (high-low bias))))) thick tails will show up. It has already been verified on the forums that the ticks are closer to a normal distribution...
I think there are errors in your calculations. Just in case, if you calculate by the traditional formula (given in the initial posts), you should use about 2000-3000 bar to get sane values.
The formula in the first posts is wrong and has nothing to do with the classic Mandelbort-Peters formula.
See page 22 of this thread for the classic calculation.
All Hearst exponent issues are solved within fractionally integrated models - FARIMA(p,d,q), where d<1. When d=0 corresponds to Hurst index = 1. In R, it is a fracdiff function that fits (estimates) model parameters. There is a corresponding instruction package that solves all these problems - in the attachment.
Once again: everything is stolen before us and for us - we can use it when building models, instead of arguing about correctness of formulas.