Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 74

New comment
 
what can this tell us? patterns, namely to draw conclusions from lags, is there a cyclical pattern?
 

We (Alexei and Alexei) now hold the view that memory can be traced back hundreds or even thousands of hour bars into the past. If you cut out the false "correlations", you can see how far the memory actually extends. Now the calculation is not quite correct, to be honest. For example, zero bar depends on the first bar, and the first respectively on the second bar, and we can see that zero bar depends on the second bar. But, the second bar affects the zero bar indirectly, through the first one. If we cut off the influence of the first (intermediate) bar, we see that the second bar has a much weaker effect on the zero bar.

But, this does not yet lead us to the question of exploiting the remaining dependencies. It may not work there at all, or it may have a very weak return. We shall see.

 
alexeymosc:

We (Alexei and Alexei) now hold the view that memory can be traced back hundreds or even thousands of hour bars into the past. If you cut out the false "correlations", you can see how far the memory actually extends. Now the calculation is not quite correct, to be honest. For example, the zero bar depends on the first bar, and the first, respectively, on the second bar, and we can see that the zero bar depends on the second bar. But, the second bar affects the zero bar indirectly, through the first one. If we cut off the influence of the first (intermediate) bar, we see that the second bar has a much weaker effect on the zero bar.

But, this does not yet lead us to the question of exploiting the remaining dependencies. It may not work there at all, or it may have a very weak return. We shall see.

I'll keep an eye on it, I take it you looked at the usual acf and chakf estimates, then saw the gains, saw that there was nothing useful there, and started to transform the original series on the sly. are there no such ways to get non-linear analogues of acf and chakf?
 
orb:
and there are no such ways to get non-linear analogues of ACF and CHAFC?

The whole topic here from the beginning is about non-linear analogues of ACF, or more precisely, a function showing dependencies of an arbitrary kind. I have calculated the ACF, see the graphs from the very beginning of the thread (page 1). There's a cyclic chart there, oddly enough, it's dependencies on a series of returns of EURUSD almost without any transformation of that series, except quantizing its values into 5 or 7 characters.

Now I am so to speak aware of the need to calculate the CHAFC. It is more difficult, because I don't have a ready algorithm yet, I need to think. Do you understand now?

 
Has anyone on the forum used the classification trees?
 
FAGOTT:
Has anyone on the forum used classification trees?

Here it was promised to end with trees

 
Viteek:

This is where they promised to finish with trees

Yes, thank you, I know that, but they haven't got to the trees yet
1...6768697071727374
New comment