6
$\begingroup$

A question for the statisticians and other math lovers: Are all symmetric matrices with diagonal elements 1 and other values between $-1$ and 1 correlation matrices?

$\endgroup$
6
$\begingroup$

I thought this must be asked & answered before, but cannot find it, so here it goes ... Let $S$ be a covariance matrix (for the algebra that follows it does not matter if it is theoretical or empirical). Let $D$ be a diagonal matrix with the diagonal of $S$. Then the correlation matrix $R$ is given by $$ R= D^{-1/2} S D^{-1/2} $$ (how you can see this directly is explained here.)

To see that $S$ must be positive (semi)-definite (abbreviated posdef), let $X$ be a random variable with covariance amtrix $S$, and $c$ a vector. Then $$ \DeclareMathOperator{\var}{\mathbb{V}ar} \var(c^T X)= c^T S c \ge 0 $$ since variance is always nonnegative. Then this transfers to the correlation matrix: $$ c^T R c = c^T D^{-1/2} S D^{-1/2} c = (D^{-1/2} c)^T S (D^{-1/2} c) \ge 0 $$ Armed with this it is easy to make an counterexample, the following is not a correlation matrix: $$ \begin{pmatrix} 1 & -0.9 & -0.9 \\ -0.9& 1 & -0.9 \\ -0.9 & -0.9 & 1 \end{pmatrix} $$

$\endgroup$

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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