Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 21

New comment
 

Alexei kind of let it slip that he didn't elaborate on the returns.

Whatever...
;)

 
Which Alexei? There are at least two.
 
If you're talking about me, I haven't developed it yet. I will now post the EURUSD M5 data. For EURUSD H1, I will do that too.
 

Calculation of mutual information for M5:

Result for randomly mixed returns:

Comparison:

Also a comparison of the sums of mutual information. For the original series, the sum over 500 lags: 1.08 Bits. For random series: 0.29 bits. Huge difference!

One more thought: for different timeframes, the sum of mutual information turns out different (I remember it was 0.6 bits for days). We can try to find the maximum variant, using conversion of minutes to arbitrary timeframes (1,2,3 ... 1000 minutes). A lot of calculations, maybe a day, but the result will be interesting.

 
alexeymosc:

One more thought: for different timeframes the total amount of mutual information is different (I remember that for days it was 0.6 Bits). We can try to find the maximum variant, using conversion of minutes to arbitrary timeframes (1,2,3 ... 1000 minutes). It may take up to 24 hours, but the result will be interesting.

So we need to find the TF with the highest mutual information? - yes, very interesting!

 
I suspect it's the M1.
 
Mathemat:
I suspect it's the M1.
I very much doubt that it is M1. I have not been able to achieve acceptable net learnability on M1.
 
Mathemat:
I suspect it's the M1.
It could be Alexey. At least the standard TFs can be checked in 3-4 hours. Is there even a way to unload custom timeframes from MT for analysis?
 
joo:
I highly doubt it's the M1. I have not been able to achieve acceptable teachability of the grids on M1.
Do you teach grids on your own technical indicators? Have you tried selecting inputs using mutual information?
 
alexeymosc:

1. Do you teach grids on your own technical indicators?

2. Have you tried selecting inputs using mutual information?

I do not use indicators, not at all lately (if you understand the term "indicator" as a procedure for converting a quotient into a form which gives fewer signals than the number of bars contained in the reporting period).

2. No, but I would love to try if I can figure out how to do it from reading this thread.

1...141516171819202122232425262728...74
New comment