Zero sample correlation does not necessarily mean there is no linear relationship - page 36

 
chepikds:
You're right, it's not!!! a dead end, I checked it myself...

It's a dead end! It's the same way.
 
By the way, my textbook says that before you use the correlation coefficient, you should check the significance of the linear relationship of the variables using statistics, which are based on the correlation coefficient and the empirical coefficient of determination(the square of the correlation index). If the linear relationship is significant, then it makes sense to check the significance of the correlation coefficient (set limits), and if it is not, then the significance of the correlation index is checked, and if it is significant, then there is a relationship, but not linear, but of an unknown type, and if it is insignificant, then there is no relationship. If anything, please don't kick me, I am not well versed in this topic.
 
FAGOTT:

How do you apply it?

There is no way to apply it. Either instantaneous estimates of correlations would be of practical value, or conversely, on a long sample, but between shifted series. The former are unrealistic to estimate due to the finiteness of the sample, the latter are statistically indistinguishable from zero for all pairs in the market.
 
FAGOTT:

Well, it's a dead end! They're applying
They apply a lot of things, but what's the point? Maybe I'm overdoing it, come to the beginning))
 
Integer:

It's just so you understand a little bit about correlation coefficients before you talk about constructive things.
Amazing ignorance. Sharing research that is nowhere to be found. Show me a single correlation indicator. Anything where millions of QC values are statistically investigated. And you talk to me about not understanding QC?! Not even Spearman's poor QC calculation has been openly done here. Read the description of my work and then make your assumptions about the depth of understanding of correlation.
 
alsu:
There is no way to apply it. Either instantaneous estimates of correlations would be of practical value, or on the contrary, on a long sample, but between shifted series. The former are unrealistic to estimate due to the finiteness of the sample, the latter are statistically indistinguishable from zero for all pairs in the market.


is it between x(t) and x(t+1) for one instrument? Is it close to 0?

I was counting - I got quite a big one. Could it be a mistake?

But these models fall back to autoregressive models and they all say the same thing - if the price goes up, it is more likely to go up and less likely to go down.

 
chepikds:
There's a lot of application, but what's the point? Maybe I overstepped my bounds, back to square one))

If there is an alternative, there really is no point.
 
alsu:
Read the definition of stationarity and show how it is absent in the original series and present in the synthetic. Or by stationarity do you mean something of your own as well?

Finally, the terminological bullshit has set in again. It is impossible to prove stationarity from a finite sample. That's what different levels of hypothesis testing are for. I'm not going to engage in theoretical bullshit.

Quote:

alsu:

Nevertheless, my request to hrenfx stands - to give an example of a synthetic instrument that is supposed to have substantially different characteristics from individual FIs. Purely for research purposes, just to see if the performance improvement actually exists or is a sham.

Isn't a horizontal channel on any stretch a performance improvement?
 
hrenfx:
Isn't a horizontal channel in any area an improvement in performance?

I would like a clear numerical criterion: "The degree of horizontality of the channel was so-and-so for each BP, but we got so-and-so".

ps So on any stretch. =)

 
hrenfx, why are you trying to prove something not very clear to those present? just mow your own cabbage and that's it!!! or what's the problem? guys have eaten more than a dog here with these correlations)