Hearst index - page 12

 
Mathemat >> :

And that's where Rosh hit the mark. It takes a lot of historical data to calculate the Hearst figure. It is not a muwing whose memory is limited to a period, but a global characteristic of BP as a whole - or a large chunk of it.

A possible solution is to build the Hurst(N) dependence, analyzing it and deciding on a sufficient amount of data. That is, starting from a certain point, k.X. will change little.

Here the question is different, how to use XH? The first thing that comes to mind is the construction of a 2x and 3x sigma channel on both sides of the average. The period of the average is based on the first maximum of the V-statistics. This is where the creative process begins :)

 
surfer писал(а) >> The period of the average is based on the first maximum of the V-statistics.

Can you go into more detail on this point, surfer?

 

A window of 400 counts or somewhere near there is usually used.


The PC shows the transition conditions to the Levy statistic.

 
Mathemat >> :

Can you go into more detail here, surfer?

the idea is to find an instrument with one maximum, and the period is rather short

entry in the area between 2nd and 3rd sigma

the problem is that V-statistics need to be evaluated visually

 

Figure to the question of choosing the interval for estimating kX based on 30 years for the DJIA (closing prices of days)

Since we are interested in interval estimation of KX, the choice of the number of R/S is not particularly difficult

 
Rosh писал(а) >>

How are you going to get the Hearst figure for the current situation? It means to consider a limited number of N bars at the moment in order to calculate Hearst on this particular sample. So you need another criterion for finding a moment in the past, from which the calculations for the current moment are made.

That's correct, but I'm sticking with it for the following reason. It varies from 0 to 1. That is its value. Just to determine the size of the window.

Look at the two indicators and the ideas behind them, both of which you programmed.

  1. Spearman's Rank Correlation Coefficient
  2. and the Perry Kaufman AMA optimised.

The AMA is based on the idea of adapting the N window size. So I want to see if Spearman can be improved by changing its N (window size). Using for this purpose Hearst or algorithm, which is included in AMA. I just haven't got around to it. I'd be interested to see an adaptive Spearman.

 

This is what I've got so far, I haven't double-checked it yet. the tilt angle tangent is somehow not calculated correctly

Files:
ckkfn.rar  31 kb
 

I don't like this indicator. Whoever wants to see it can look it up and try it out.

In order to apply different signals to the input described above in the line Y=Y0 just change to the corresponding Y1 Y2 ...

file attached. Matcad version 14.

If you suddenly see an error. I will correct it. I may have made a wrong calculation.

Files:
whknt.rar  35 kb
 
Prival >> :

If you happen to see an error. Be sure to let me know and I'll correct it. Cause I might have miscounted something.

I'm not sure, but X[N] should have a large N above the summation sign, not n.

 
Prival писал(а) >>

I don't like this indicator.

It's a good indicator!

You shouldn't do that. It's probably a mistake in your formulas. I've been trying to figure it out, but I don't have the stomach for it. At first glance, the R parameter should be calculated like this:

In general, I quickly sketched my version of RF (the algorithm described above) for your BP:

Y0 is trend + noise, Y1 is integrated noise (analog of kotier), Y2 is noise with zero MO (analog of the first difference of kotier), Y3 is sin + noise.

Here are the results of plotting the VC for different TFs:

Seems to be all according to science.

The range of variation of PC is from 0 (series of the first difference) up to 1 (linear trend on large TF). The special place is occupied by a random Brownian one-dimensional motion (integrated SV with zero MO), for it PC=1/2, and a noisy sine, at this comrade, PC smoothly oscillates that should be, as on small TF the noise plays a large role, on large TF the trend is already visible, etc.

Files:
hearst.zip  10 kb