Volumes, volatility and Hearst index - page 11

New comment
 
joo:
Can't you just make up formulas and declare them Hearst? You can call it Stool, as long as it makes sense. That would be Yurixx's Criterion.
Somebody can call a cart painted red Ferrari to attribute properties of the Ferrari to his cart, and somebody can call his home-brewed calculation for the derivative series Hearst to attribute properties of Hearst to his calculation.

You see, "Yurixx's criterion" has no significance and meaning, which Hirst's criterion has, the properties of the former are not known and require proof. It is another matter to call yourself Hearst.
 

I'm going to get in a bit... :o) I analyzed Hearst's behavior with 5 or 7 (I don't remember) calculation variants (I wrote briefly here https://www.mql5.com/ru/forum/123519/page387 and a bit farther). The result is similar, apart from that I would like to add:

  • Huge dependence on sample size (but this is a classic). "Outliers"/"spreads"/"deviations" (as you like) are not large or small by themselves, everything is learned by comparison. And the condition for comparison lies in the sample itself, or rather in what it contains.
  • A very tricky manifestation of "memory". It "slips away" all the time, in other words it jumps on different scales, like a monkey on a palm tree. It may not exceed a month (e.g. at 15m). And when approaching significant extremums (especially near global levels or levels of price "concentration", i.e. high frequencies), memory length can significantly (or rather jumps) increase. You should use wavelet method of Hearst calculation, it is more accurate.
  • The longer the memory length, the worse it is - the number of equal variants of trajectories increases catastrophically. In other words "long tail", the rightmost/leftmost parts, is more often manifested in this period. (But that's already my conclusion.)
  • I do not take ticks on forex. On ticks any brokerage company will show any dependence (remember wine-wine). But this is my personal conviction.
 
Vita:

1. In the first place not only the argument of the function, but also the multiplier of that function. For the numerical experiment which is carried out here in Table 2b, the result of this function is constant, but we are already too deep in search of the truth.

2. Yes, and can you yourself explicitly say that High - Low = k * sqrt(N) is wrong?

1. I don't have time now to parse the work in detail, but it seems as if the root of T arises only for the case m = n/2 . Which can roughly be taken to be true for large n, but in no way for small n.

2. Well I made the above assertion, after all:

Yurixx's graphs clearly show that there is no relation between the RMS and the spread. Of course if its calculation is correct.

And since it is proven that RMS = A * sqrt(N), then High - Low = k * sqrt(N) is generally incorrect. Again, provided Yurixx's calculation is correct .

 

Farnsworth:

The only difference is that I have come to the humble conclusion that TA does not work at all.

Maybe it is because the market does not obey the theory of probability as it is not random?
 
joo:
Can't you just make up formulas and declare them Hearst? You can call it Stool, as long as it makes sense. That would be Yurixx's Criterion.
Here we have a "victim" of propaganda :). It was Vita, not Yurixx, who invented the formula, while Yurixx just did the correct procedure for calculating Hurst.
 
Candid:
So much for the "victim" of propaganda :). It wasn't Yurixx who invented the formula, it was Vita, and Yurixx was just doing the correct procedure for calculating Hurst.
Why a victim? I was trying to get at Vita.
 
joo:
Why the victim? I was going to poke fun at Vita.
So I put inverted commas. :)
 
Candid:
So much for the "victim" of propaganda :). It wasn't Yurixx who invented the formula, it was Vita, and Yurixx was just doing the correct procedure for calculating Hurst.

I didn't make anything up, here is what Yurixx writes in his post (I underlined it for you):

So, the final form for Hearst's indicator is: h = Log(High-Low)/Log(N). Here N is the number of single ticks on the time interval, High and Low are maximum and minimum price values, reached on this interval. Their difference is expressed in 4-digit points.

Firstly, the author of the formula h = Log(High-Low)/Log(N) is Yurixx. I don't need his laurels.

Secondly, please note that there is no standard deviation in this formula, which you refer to in your previous post to question my calculation.

The average mileage formula is a textbook formula, it's not mine either, and again it doesn't have any standard deviation

Why are you bringing up, now not Hirst, but the standard deviations? Where do you see them in my calculation? In the formula Үurixxa or the formula for average mileage for SB? There are no standard deviations there, nor is there any Hearst.

I merely argue that the average run is directly proportional to the average spread, and therefore true, let it be, "my" formula High - Low = k * sqrt(N), after substituting which into the formula Үurixxa, we get the result tending to 1/2 from above. Converges with table 2b, the result of the experiment. But you still like to call the formula Үurichxa Hurst, despite all the inconsistencies with the experiment and restrictions on the original series.

Has anyone calculated Hurst for N in a cube by Үurichx yet? Does anyone see this log? Or are we going to chase the specks in my eye?

 
Candid:


his calculation has both the average mileage and the RMS mileage and the spread. - Give me his formula which has this RMS mileage in it. I see a completely different formula on his front page. See my post above. Your formula with the root is only proven for RMS, that is also not relevant to Open - Close.
Show in Yurixx 's post the formula in which you did the substitution. - I did, see the post above.
Where is the table or graph? At least the value to the led, at which the agreement comes.
In your original reasoning you introduce a variable h and call it the Hearst exponent. This is incorrect, it is not the Hearst exponent.
The answer is 1/2 - Oopsie. Can I have the calculation? but it won't be the Hearst figure, the Hearst figure is calculated by the spread.



 
to Andrei01:
Could it be because the market does not obey the theory of probability as it is not random?

The properties of the market (as a whole) are very close to random. Consistently came to the following conclusion (I'll even highlight :o):

You can't treat a quotation process as a whole. Moreover, the quoting process as a whole does not exist in nature - it is an illusion. It makes no sense to take any quote statistics and study it, even the reduction to a stationary series will not give anything. It is senseless to take any lengths and it is impossible to take the whole history.

But these are my conclusions and they are confirmed (unfortunately, it makes the development of TS extremely difficult). I wanted to make a thread on this subject, but probably much later. Not much free time at all.

PS: TV always works, you should not confuse it with the conclusions of TV about "something" as not working.

1...456789101112131415161718...37
New comment