Probability theory question... - page 6

 
Yurixx писал (а) >>

Can you give me a quick primer?

Is the ACF you are talking about a correlation of neighbouring values ? How is it calculated ? I remember that ACF and correlogram are "two big differences", but I don't remember which is which. :-(

I also remember that in a big thread they talked about all this, but I don't remember where. Maybe it's already sclerosis ? :-)

If ACF is a function of a series point and, consequently, depends on the reference number, you can't reproduce it, it's not even worth trying.

But if it is a non-local statistical function of series CB, i.e. depends on other parameters, then maybe just the opposite and I just need it to get rid of that arbitrariness, which is still present, if we deal with the generation of synthetics.

Yes, arbitrariness is present. Therefore, we will call (as far as I remember now) correlogram, autocorrelation coefficients between samples in the series of the first BP difference lagging behind each other by n samples. Thus, we get a function defined on the interval 1,2...n, where n is any integer non-negative number and taking values from -1 to 1 inclusive.

This is exactly the function I meant in my previous post.

 

So ACF is a set of autocorrelation coefficient values for different lag values ? And the same thing you call a correlogram. So I was wrong and it is the same thing: ACF=correlogram. But that makes life a lot easier !

In that case, perhaps the ACF could be used to determine a measure of the correlation between the return and the trend. Just the kind of arbitrariness I wanted to get rid of.

 
Yurixx писал (а) >>

And what I wanted to share with you concerns the behaviour of H-volatility to the right and left of 2. If you remember, the right side is a trend market, the left side is a return market. So H-volatility behaves very differently in these areas. This difference makes it completely unsuitable for determination of the market trend condition.

The notion of "trendiness" and "flatness" is very relative. Indeed, if one plots a Cagi-Split (the same Zig-Zag) with an arbitrary step H on the historical data, then, regardless of the value of H, the market will always be trending... It is quite another matter when we have to work in casual mode, i.e., at the right border of BP, then a top of WP is formed with an inevitable lag in its amplitude (I intentionally did not tie it with a time scale). In this case we have to decide whether to open in the direction of movement or against it. In the first case, the market is considered as a trend one, in the second - as a pullback; there is no third one, and thank God! You and I understand this very well. In such a formulation there is no controversy in interpreting the H-volatility as a "trend-flat" indicator, or you put some sacred meaning into this concept?

So the ACF is a set of autocorrelation coefficient values for different lag values ? And the same thing you call a correlogram.

That is how it is found in the literature.
 
Mathemat писал (а) >>

Hehe, a rash statement... Given chart Close[n] = 2 + sin(n/20) (Close[n] - consecutive counts). It has the following signals: if Close < 1.2, buy with Take 1.79 and Stop 0.21; if Close >= 2.8, sell with the same parameters. Good signals that are never wrong.

I am not saying I have a solution to the problem of predicting the future. The above is just a vague hint at the direction of research that I haven't even started.

In this unstarted research are you going to build a model or are you going to crush the price series with almighty transformations?

 
Vita писал (а) >>

In these unstarted studies are you going to build a model or are you going to crush the price range with almighty transformations?

Bingo, Vita. There are a countless number of transformations we can think of. That's another degree of freedom, roughly speaking. And the model may well come out of what comes out (if it comes out well). Why, have you already tried, judging by the accuracy of the shot?

P.S. We started off sort of "on a first-name basis".

 
Totally agree with you, except maybe that "no matter how big H is, the market will always be trending". We should look at those pictures we posted in the big thread. But those are details. I don't put any sacral meaning into it. But Pastukhov claimed that H-volatility may be used as a measure of trend-flat condition. But according to my results it is not. In the left neighborhood of H=2 it still may be, as H-volatility behaves linearly there, but definitely not in the right one. Firstly, its behavior there is highly non-linear, and secondly, at the H=2 point H-volatility has a minimum, i.e. tangent to the chart is horizontal there. As a result, H-volatility in the nearest vicinity is almost insensitive to changes in the market condition. And this is the most important point - the moment of transition to the trend state.
 
Neutron писал (а) >>
I, you remember, was messing around with tick flow modelling at one time. All that I managed to realize was either coincidence of probability function density function (PDF) of synthetic and real VR, but then there is a strong divergence of their autocorrelation functions (ACF). Or, on the contrary, I have achieved coincidence of the ACF and then there is a divergence of their PDFs. It is clear that these two things are somehow related, but I could not solve this problem - not in my way. I don't think I could be of any help, although I'll go in and read about it.

I have the same ACF and FR, where to go to read plz. I might be good for something.

 
vizit писал (а) >>

I will try to be very brief but clear:

There is a set of indicators (number = N). All indicators at the same time give a signal to either buy or sell. The probability of choosing the right direction for each indicator = Dn (where index n=1...N). The order opening direction, following the signals of the indicators, is selected in the direction that most of the indicators give signals to.

Question: What is the probability of selecting the correct order opening direction?

-

I have been searching for an answer for a week. I do not have energy anymore. Maybe someone has more knowledge in this area. Thanks in advance for the answer!

or if it's different.... don't get hung up on statistics (science I mean) etc. - just calculate that these 2 out of 3 or 3 out of 4 give the most correct choice???

 

have been trying to voice this question for a long time, no one is responding..... too dumb probably.... last time I try....

Let's say:

- there is some statistical preference (2-3-4 eyes seen)

- not fully defined: there are many options, but the perspective is intuitively guessed.....

Which is easier, and more importantly, more expedient, financially and temporally:

- to study it to the end, and then already TC to sculpt?

- to make 2-3-4 versions of TC.... test them, etc.?

In general, of course, the question is much wider: Are there such statistical-time patterns in forex, which can be the basis for a trading system, or is there some symbiosis needed? )

And I'm not shouting, CL suddenly turned on :)))

 
Prival писал (а) >>

I have the same ACF and FR, where to go to read plz. Maybe I'll be good for something.

Hi Sergey.

If you really managed to build a tick flow model satisfying two requirements:

1. Coincidence of probability density function for series of first difference of initial BP and model BP;

2. coincidence of correlation coefficients between samples separated by a positive integer in the first difference series and the model return series.

My respect to you!

Can you post the proper graphs, confirming what you say and tell us a little bit about the method used?