Suppose we have two random variables that follow a multivariate normal distribution,
$[x,y]\sim MVN([a,b], \begin{bmatrix} \rho_1 & \rho_3 \\ \rho_3 & \rho_2 \end{bmatrix})$
Then what is the distribution for the sum of x and y?
That is,
$z=x+y \sim N(?,?)$
Surely, $N(a+b,\rho_1+\rho_2)$ does not seem right, since we need to consider the correlation between x and y. But what is the correct way to incorporate this correlation aspect into the final distribution?