Suppose $X$, $Y$, and $Z$ are normal random variables with means and variances of $\mu_X, \mu_Y, \mu_Z$ and $\sigma^2_X, \sigma^2_Y, \sigma^2_Z$ respectively. $X$ and $Y$ have correlation $\rho$. $X$ and $Z$ are independent, and $Y$ and $Z$ are independent. Then what is the correlation of $X$ and $Y-Z$? I know that $Y-Z$ has distribution $\mathcal{N}(\mu_Y - \mu_Z, \sigma^2_Y + \sigma^2_Z)$.

$corr(X, Y-Z) = \frac{cov(X,Y-Z)}{\sigma_X\sigma_{Y-Z}}$. $cov(X,Y-Z) = cov(X,Y) - cov(X,Z)$.

$corr(X,Y) = \rho = \frac{cov(X,Y)}{\sigma_X\sigma_Y}$. So $cov(X,Y) = \rho \sigma_X \sigma_Y$. $cov(X,Z) = 0$ since they are independent. So $cov(X,Y-Z) = cov(X,Y)$.

So $corr(X, Y-Z) = \frac{cov(X,Y)}{\sigma_X\sqrt{\sigma^2_Y + \sigma^2_Z}} = \frac{\rho \sigma_X \sigma_Y}{\sigma_X\sqrt{\sigma^2_Y + \sigma^2_Z}} = \frac{\rho \sigma_Y}{\sqrt{\sigma^2_Y + \sigma^2_Z}}$.

Is this right?

  • 2
    $\begingroup$ What is the covariance of $X$ and $Y$? What is the covariance of $X$ and $Y-Z$? What is the correlation of $X$ and $Y-Z$? $\endgroup$
    – Henry
    Dec 12, 2021 at 4:34

1 Answer 1


You have a joint (normal) distribution of $(X,Y,Z)'$, can you think of a transformation to get the joint distribution of $(X,Y-Z)$?

Hint: If you take A = $[1,-1,0]$ you get the distribution of $(X-Y)'$

  • $\begingroup$ The question doesn't say that (X,Y,Z) are jointly normal. Unless I missed something, we don't have that (X,Y) are bivariate normal, only that they're marginally normal with a specified correlation. $\endgroup$
    – Glen_b
    Dec 12, 2021 at 22:56
  • 1
    $\begingroup$ As @Glen points out, this approach requires an additional assumption. However, the idea is correct, because the question supplies full information about the variance-covariance matrix of $(X,Y,Z).$ Whether that has a multivariate Normal distribution is irrelevant. $\endgroup$
    – whuber
    Nov 19 at 17:10

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.