0
$\begingroup$

If $(X,Y)$, $(X,Z)$, and $(Y,Z)$ are all Gaussian, does it follow that $(X,Y,Z)$ is also Gaussian? I'm having trouble coming up with a counterexample...

$\endgroup$
6
  • 3
    $\begingroup$ Possible duplicate of Is it possible to have a pair of Gaussian random variables for which the joint distribution is not Gaussian? A straightforward counterexample is to extend the example in the upper right figure in @whuber's answer to three dimensions. $\endgroup$ – Jarle Tufto Nov 19 '19 at 20:13
  • 2
    $\begingroup$ Why do three normal random variable necessarily have to have a Gaussian joint distribution? They're not necessarily independent. @Konstantin $\endgroup$ – Bindiya12 Nov 19 '19 at 20:18
  • 1
    $\begingroup$ @Konstantin - Just being able to construct a covariance matrix does not mean the three variables are jointly Gaussian. $\endgroup$ – jbowman Nov 19 '19 at 21:14
  • $\begingroup$ @jbowman Thank you, you are so right, I've learned from the post Jarle Tufto ponted at. $\endgroup$ – Konstantin Nov 19 '19 at 21:20
  • 1
    $\begingroup$ The example in the duplicate was posted by @Cardinal, not by me. However, at stats.stackexchange.com/a/434151/919 I recently posted an example of a triviariate distribution that has uniform 2D marginals but is not uniform. By treating this as a copula one obtains an (intriguing) example involving pairwise Gaussian variables that are not jointly Gaussian, but in a subtle way. This approach extends to any univariate distribution whatsoever. $\endgroup$ – whuber Nov 19 '19 at 21:33