3
$\begingroup$

Suppose I have two correlated random variables, that were generated in the following way: \begin{align*} X_1 &\sim \mathcal{N}(0,1)\\ X_1' &\sim \mathcal{N}(0,1)\\ X_2 &= \rho X_1+\sqrt{1-\rho^2}\cdot X_1'\\ Y_1 &= \mu_1+\sigma_1 X_1\\ Y_2 &= \mu_2+\sigma_2 X_2. \end{align*}

Now, is it true that $Y_1+Y_2$ (or, more generally $\alpha_1 Y_1+\alpha_2Y_2$) normally distributed? (I can easily calculate the mean and the variance of $\alpha_1 Y_1+\alpha_2Y_2$, but I am not sure about the distribution...)

EDIT: just to clarify, $X_1$ and $X_1'$ are independent.

$\endgroup$
  • $\begingroup$ You mean $X_2 \sim \rho X_1 +\sqrt{1-\rho^2} X_1'$? $\endgroup$ – ntt Apr 14 '17 at 15:37
  • $\begingroup$ Sure, fixed, thanks. $\endgroup$ – Bach Apr 14 '17 at 15:37
  • $\begingroup$ Do you mean that $X_2 \mathbf{\,=\,} \rho X_1+\sqrt{1-\rho^2}\cdot X_1'$? $\endgroup$ – NCh Apr 14 '17 at 15:40
  • $\begingroup$ Sure, fixed as well, thanks :) $\endgroup$ – Bach Apr 14 '17 at 20:36
2
$\begingroup$

$\alpha_1 Y_1 + \alpha_2 Y_2$ is a linear combination of $X_1$ and $X_1^\prime$ - that is $\alpha_1 Y_1 + \alpha_2 Y_2 = \beta X_1 + \beta^\prime X_1^\prime$ for some $\beta, \beta^\prime$ that are a bit of a pain to calculate. Linear combinations of independent normal random variables are normal; there are several proofs of this (nontrivial, but well-known) fact. So the answer to your question is yes.

$\endgroup$
  • $\begingroup$ Right, I should have seen this. Thanks! $\endgroup$ – Bach Apr 14 '17 at 15:42
  • $\begingroup$ @Michael Lugo Here $X_1$ and $X_1'$ are correlated, how can you deduce $\alpha_1 Y_1 + \alpha_2 Y_2$ is normal? $\endgroup$ – ntt Apr 14 '17 at 15:50
  • 2
    $\begingroup$ Bach: as @ntt has pointed out, this only holds if $X_1$ and $X_1^\prime$ are independent. I'd assumed from your notation that you intended independence there, but if that's not the case then my answer is incorrect. $\endgroup$ – Michael Lugo Apr 14 '17 at 15:51
  • $\begingroup$ Yes, I meant that they are independent. Sorry for not being clear about that. $\endgroup$ – Bach Apr 14 '17 at 16:00
  • 1
    $\begingroup$ @ntt Uncorreletedness of two r.v. with given normal marginals does not imply independence. . $\endgroup$ – NCh Apr 14 '17 at 16:08
2
$\begingroup$

Take $X_1 \sim \mathcal N(0,1)$ and let $Z$ be a r.v. with $P(Z =-1)=P(Z=1)=1/2$ and $X_1, Z$ are independent. Denote $X_1'=Z\cdot X_1$, then $X_1' \sim \mathcal N(0,1)$. However $(X_1, X_1')$ is not Gaussian as $X_1+X_1'$ is not normal. Then $X_2$ should not Gaussian, perhaps $X_2$ will be Gaussian with some specific values of $\rho$. We have $Y_1$ is Gaussian, $Y_2$ is not Gaussian then $Y_1+Y_2$ is probably not Gaussian in general.

$\endgroup$

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.