1
$\begingroup$

I'm reading this paper, which contains the following covariance matrix:

Covariance Matrix

In the example there are six forecasters who estimate some quantity, and then we look at the covariances of those estimates.

I understand from this question that the rank of a covariance matrix is at most $n-1$, and I also understand that the rank of a matrix is also equal to its number of nonzero eigenvalues.

I entered this covariance matrix into MATLAB and it says the rank of the matrix is 6. MATLAB says this matrix also has 6 nonzero eigenvalues. What's going on?

$\endgroup$
2
$\begingroup$

Don't confuse the sample size with the number of vectors for which you calculate the covariance.

You have $k=6$ forecasters, each of which provides some number $n$ of forecasts (which you don't give in your question). So you get a $k\times k$ covariance matrix, which can have a rank of at most $n-1$. But if $n>k$, then the matrix can certainly have full rank $k$, as it does here (it's block diagonal, and both blocks have nonzero determinant, so the matrix is nonsingular and has full rank).

$\endgroup$
  • 1
    $\begingroup$ There might not be any $n$ involved as this might not be a sample covariance matrix $\endgroup$ – Juho Kokkala Nov 5 '15 at 11:19
  • $\begingroup$ It is a sample covariance matrix, though. Actually the example in the paper is hypothetical but they're referring to a situation which would in the regular course of events involve a sample covariance matrix. $\endgroup$ – user1205901 Nov 5 '15 at 11:52
  • 1
    $\begingroup$ Why is the rank tied to the number of forecasts $n$ rather than the number of forecasters $k$? I realise that it is true, but I don't understand why the number of forecasts should be relevant, since the size of the covariance matrix is completely determined by the number of forecasters. $\endgroup$ – user1205901 Nov 5 '15 at 12:17
  • 2
    $\begingroup$ @user1205901: that's exactly the question you link to. See also this question, which the other one links to. $\endgroup$ – Stephan Kolassa Nov 5 '15 at 12:21
  • 2
    $\begingroup$ @user1205901: I have now provided an answer to the question you linked to; you might be interested to take a look. +1 to Stephan's answer, by the way. $\endgroup$ – amoeba Nov 5 '15 at 17:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.