Covariance of two normal distributions where one is squared.

55 Views Asked by At

I'm facing this question while looking for the variance of the product of two normal distributions. Let say we have $x$ and $y$ which are normally distributed with mean = 0 and sd = 1 and correlated by $\rho$. Is their a way to compute the expected covariance of $x$ and $y^2$? My simulations show that it should always be 0 in these conditions, but i'm looking for some examples, proof or references.

Thank you,

1

There are 1 best solutions below

0
On BEST ANSWER

Define $Z:=X-\rho Y$ so $Z\sim N(0,\,1-\rho^2)$ is uncorrelated with $Y$. If $X,\,Y$ are joint Normally distributed (which is a stronger condition than their each being Normally distributed), so are $Y,\,Z$, and $Z$ is also independent of $Y$ so$$\begin{align}\operatorname{Cov}(X,\,Y^2)&=\Bbb E(XY^2)-\Bbb EX\Bbb EY^2\\&=\Bbb E(ZY^2+\rho Y^3)\\&=\Bbb EZ\Bbb EY^2\\&=0.\end{align}$$In particular, the penultimate $=$ works because $Z$ will, on the above assumptions, be uncorrelated with $Y^2$.