Zero sample correlation does not necessarily mean there is no linear relationship - page 11
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
It is nonsense to get a correlation value without an estimate of significance.
What kind of confidence assessment can we talk about when the calculations before analysing the sliding window results are incorrect...
It is nonsense to apply a correlation (which describes a linear relationship) so that it shows different absolute values for {EURUSD; USDJPY} and {EURUSD; JPYUSD} pairs.
I wonder who gets such different (delusional) results?
What confidence assessment can we talk about when the calculations before analysing the sliding window results are incorrect...
What else is in the pocket?
The dynamics are given by the sliding sampling window.
Really?! :)))
but for a sample of 500 samples? and we need to identify the relationship (or lack thereof) for the last 100 samples for
EURUSD and GBPUSD, for example? To see how this pair relationship changes, how far the quotes of
of one pair advance or lag behind the other? :)
I argue that when using Pearson this approach gives birth to this saying "There are lies, there are grandiose lies, but there are also statistics".
:)
It is nonsense to apply a correlation (which characterizes a linear relationship) so that it shows different absolute values for {EURUSD; USDJPY} and {EURUSD; JPYUSD} pairs.
OK, let y=ax+b. Prove that there is also a linear relationship between y and 1/x.
I wonder who gets such different (nonsense) results?
For example, you have...you use a formula for calculating the correlation coefficient which is not obvious to me (below is the corr2 function).
Below I show the correlation calculation without first logarithming the original BPs:
You can see that 1 / X already gives a different absolute QC value.
Now with logarithm:
You can see that 1 / X gives an identical result.
You can also see that Mathcad calculates the correlation as I wrote above: covariance divided by the product of RMS - function corr3.
OK, let y=ax+b. Prove that there is also a linear relationship between y and 1/x.
What kind of linear relationship is there?! It's hard to understand what we're talking about.
You're not going to present EURUSD = a * USDJPY + b. Or is linear regression applied here without logarithmization of price VRs?
If so, it would be: log(EURUSD) = a * log(USDJPY) + b. And in fact this b should be discarded as a zero value.
It's not clear to you that log(USDJPY) == -log(JPYUSD). And that the linear relationship by definition cannot change in absolute value when the BP price is inverted, but only changes its sign?
The above has clearly demonstrated this.
You wouldn't think EURUSD = a * USDJPY + b. Or are they also using linear regression without logarithm of price VPs?!
If so, it would be: log(EURUSD) = a * log(USDJPY) + b. And in fact this b must be discarded as a zero value.
You're not convincing, hrenfx! I understand that logarithmic returns are more appropriate to describe the quoting process, but we have two processes here instead of one.
And the second: what are we fighting for? One thousandth in correlation coefficients? And what will it give you, such precision?
Logarithmetic makes sense when the values change over wide ranges, not by single percentages.
That's not convincing, hrenfx! I understand that logarithmic returns are more appropriate to describe the quoting process, but here we have not one but two processes.
And the second: what are we fighting for? One thousandth in correlation coefficients? And what does it give you, such precision?
Logarithmetic makes sense when the values change over wide ranges, not by single percentages.
I won't show examples on real price BPs where the differences are substantial, not in the "thousandths". And you just have to understand that studying absolute prices of financial instruments is nonsense. I am surprised that almost no one sees this. You should see the wording of Markowitz's portfolio problem. Or better yet, Recycle, where one doesn't care at all about the nature of the original BPs: price, equity of the TS, etc. The linear relationship is perfectly clear and unambiguous between these BPs.