MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Dear all, I might just be blind, so forgive me if it is a trivial question. Given two normally distributed variables $x_1$ and $x_2$ (with zero mean), their correlation $c$ can be estimated from the samples $x_1 x_2$, $c = E[x_1 x_2]$ (where E denotes the expectation value). Now assume I want to estimate the square of the correlation, $c^2$. Unfortunately, the expectation value of $E[(x_1x_2)^2]$ for two zero-mean Gaussians is simply the product of their variances, so $E[x_1^2x_2^2] = \sigma_1^2\sigma_2^2$ where $\sigma_i^2$ is the variance of $x_i$. So how do I compute $c^2$ directly as an expectation value of independent samples?

Thanks so much! Wieland

share|cite|improve this question
$c$ is called the covariance, not correlation, between $x_1$ and $x_2$. – Lutz Mattner Mar 28 '13 at 14:48
up vote 1 down vote accepted

$E[x_1^2 x_2^2]=\sigma_1^2 \sigma_2^2 + 2 c^2$

(an application of Isserlis theorem)

share|cite|improve this answer
Thanks Carlo for pointing me to Isserlis theorem! Indeed I missed a 2$c^2$ here. This expression is still not completely what I wanted, but I should be able to use it for optimization purposes. Ideally I searched for an expectation value that is zero if the correlations are zero and positive otherwise. – Wieland Mar 28 '13 at 11:41

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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