0
$\begingroup$

I am trying to prove the above statement but not able to find an elegant way.... any thoughts would be appreciated.

$\endgroup$
2
  • $\begingroup$ yes 2x2 can be shown as sum of squared expressions with positive coefficients $\endgroup$
    – dasman
    Feb 14 '13 at 4:11
  • $\begingroup$ What if two variables are perfectly correlated? $\endgroup$
    – hardmath
    Feb 14 '13 at 4:26
2
$\begingroup$

That statement is not true in general.

For example,

$$\begin{pmatrix} 1 & 0.9 & 0.2 \\ 0.9 & 1 & 0.7 \\ 0.2 & 0.7 & 1 \end{pmatrix} $$

$\endgroup$
1
  • $\begingroup$ A correlation matrix must be positive semidefinite. Your example is not. $\endgroup$
    – user1551
    Feb 14 '13 at 9:09
0
$\begingroup$

You cannot. The statement is false. The most obvious example is $$ \begin{pmatrix}1&1\\1&1\end{pmatrix}, $$ where the random variables are perfectly correlated. For a less obvious example, take any $n\times k$ random matrix $X$ such that (a) $n\ge k$, (b) all row sums of $X$ are zero and (c) in each column, all entries have the same sign. Then the covariance matrix $XX^T$ has deficient rank but it almost always have all positive entries. And so does the correlation matrix. For a concrete example, consider $$ X=\begin{pmatrix} 1& 0&-1& 0\\ 1& 1& 0&-2\\ 0& 2&-1&-1\\ 1& 1&-2& 0\\ 1& 1&-1&-1 \end{pmatrix}, \ XX^T=\begin{pmatrix} 2&1&1&3&2\\ 1&6&4&2&4\\ 1&4&6&4&4\\ 3&2&4&6&4\\ 2&4&4&4&4 \end{pmatrix}. $$

$\endgroup$

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.