Zero sample correlation does not necessarily mean there is no linear relationship - page 54

New comment
 
Demi:

Isn't there more than an ACF to put a QC anywhere else? What about a QC between instruments? No? Can't you think of it?

Well, take the prices of S&P500 and NASDAQ indices (^GSPC and ^IXIC on Yahoo.Finance, respectively). The correlation calculated on prices will again be positive. Can you build a profitable strategy?))

And spread trading? Also crossed out?

Because they do not use correlation, but cointegration.

Why these meaningless posts?

I'm giving you ideas on how to make grails with this approach. Don't you have any doubts that you are right, if there is correlation but you cannot make money on the boat? :D

 
anonymous:

Well then take the prices of the S&P500 and NASDAQ indices (^GSPC and ^IXIC on Yahoo.Finance, respectively). The correlation calculated on prices will again be positive. Can you build a profitable strategy?))

Scratch that, as it's not correlation but cointegration that's used.

I'm giving you ideas on how you can make grails with this approach. Don't you have any doubts that you are right, if there is a correlation, but you cannot make money on the boat? :D

Well, that's what I thought... It all makes sense now...

So:

1. market instruments are not cointegrated - remember that for life.

2. I can do anything - remember that too.

3. For all of the above, it is correlation that is used - see item 1.

4. do not spend time on the forum, and read, read, read. Googling, googling, googling.

P.S. stop playing the clown

 
Demi:

1. market instruments are not cointegrated - remember this for life.

Look at BRK-A, BRK-B share prices. Counter-example, once again.

2. i can do anything - remember that too.

Good for you.

3. it is correlation that is used for everything I listed - see point 1.

It has been discussed both in this forum and in MQL5 forum.

I'm tired of boring you here too.)

 
Demi: On a larger scale, QC can and should be calculated on the initial data (prices) of the forex market.
Well, well, do the math. You will find a lot of such correlations, which nobody here has dreamed of, because they are nonsense (false correlations).

QC for stationary and ergodic series is not needed at all - for them everything is clear and understandable.

Wrong in principle. Somewhere I saw a very illustrative example of calculation of the diffusion rate of paint in a liquid just using ACF. The process was stationary and very probably ergodic.

I shall post it when I find it.

 
Mathemat:
Do the math. You will find plenty of such correlations that no one here would dream of, because they would be nonsense (false correlations).

Financial instruments in forex are correlated. Once again, this is all inter-market analysis, pairs, etc. trading, spread trading with the exception of seasonal trading.

False correlation could be between the speed of hair growth on the head and the dynamics of continental plate movement.

 
Integer:

So the row is stationary... So you can't use it that way, but only the first differences. Let's imagine another row, exactly the same, and another one, only the line is pointing downwards.

So, the correlation is perfectly calculated, when both series are in the same direction - we get 1, when in different directions - -1. I.e. the result makes sense, the correlation is calculated and the value corresponds to reality.
However, the series are non-stationary, so you can't do it that way:) you have to count the correlation from the first difference. So we have series 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 and -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 - on such data the correlation cannot be calculated.

That's it! Gentlemen

* * *

I searched the internet a little on Granger, and there I met statements that Granger method should be applied only on the first differences... However in more competent textbooks there is no such thing, on the contrary it is written, that on stationary data another method is applied. But with what aplomb everyone proves their point... I don't know, it's obvious to me that I don't need any first difference.

* * *

All is clear with gentlemen econometricians and the like... Therefore, I take my leave and do not participate in conversations on the subject of correlation and the like.

In addition to manipulating formulas and terms, one must also understand the essence and meaning.


You're stressing the point, but meanwhile you've lost it yourself. A simple example, two stationary, random walks with zero MO:

It is obvious that both are directed in the same direction, it is also obvious that there is no correlation between these processes. Taking the QC for these two series as it is, we obtain the coefficient equal to 0.86, i.e. we have identified a strong correlation. But if it is reliably absent, then what have we got? Now we take the first differences of these two processes and calculate the correlation coefficient for them and now it is equal to 0.02, i.e. it has shown what it is supposed to show - there is no connection. Their movement in one direction is a simple coincidence.

By calculating QC on I(1) you are fitting statistical methods to what seems to you. And visually, the two series do appear to be similar, when in fact they are not.

 
C-4: It is obvious that both are pointing in the same direction, it is also obvious that there is no relationship between these processes. Taking the QC for the two series as it is, we get a coefficient of 0.86, i.e. we have identified a strong relationship. But if it is reliably absent, then what have we got? Now we take the first differences of these two processes and calculate the correlation coefficient for them and now it is equal to 0.02, i.e. it has shown what it should show - there is no connection. Their movement in one direction is simple coincidence.

Very good example, thank you. A pebble in the direction of lovers of false correlations, who think that they will never get them.

 
Integer:

Gentlemen, can you tell me if this data series is stationary or non-stationary?

It is a stationary process of the form I(1), if only because its mathematical expectation does not change with time. But it has no variation (dispersion) and hence it makes no sense to have at least Pearson QC because it uses the sum of products of variation of two vectors divided by their dispersion, which we don't have in this process.
 
C-4:


You are stressing the point, but in the meantime, you yourself have lost it. A simple example, two stationary, random walks with zero MO:

It is obvious that both are pointing in the same direction, it is also obvious that there is no relationship between these processes. Taking the QC for the two series as it is, we get a coefficient of 0.86, i.e. we have identified a strong relationship. But if it is reliably absent, then what have we got? Now we take the first differences of these two processes and calculate the correlation coefficient for them and now it is equal to 0.02, i.e. it has shown what it should show - there is no connection. Their movement in one direction is a simple coincidence.

By calculating QC on I(1) you are fitting statistical methods to what seems to you. And visually, the two series do appear to be similar, when in fact they are not.

1. MO=0? The MO of the series = 0? Or the increments of the rows?

2. both rows are stationary? Are you sure about that?

3. QC does not and never has established the presence or absence of any functional relationships. It is simply a numerical characteristic. The presence or absence of relationships is a matter of interpreting QC by other methods.

 
C-4: This is a stationary process of the form I(1)

No, there can be no such thing. Only series I(0) can be stationary.

Demi: 2 . Both series are stationary? Are you sure about that?

No, they are not stationary. They are just selected pieces of a Wiener process (i.e. a Brownian one), as far as I understand it. That is, the process I(1), if it is indeed a Wiener process.

1...4748495051525354555657585960
New comment