We know that if a variable $X$ is iid from a $N(\mu,\sigma^2)$,
the distribution of $X+b$ is $N(\mu+b,\sigma^2)$
If we scale the $X$ by a scaling factor $k$, the new distribution will be $N(k\mu+b,k^2\sigma^2)$.
Does the same principle applies for multivariate normal distributions?
What happens if the scaling factor is a matrix?
It's ok if you don't give a full answer but a nice reference would be nice
Thanks
Best Answer
Almost.
Suppose the variance of $X\in\mathbb R^{n\times 1}$ is the $n\times n$ matrix $\Sigma$.
Suppose $A$ is a $k\times n$ matrix.
Then the variance of $AX$ must obviously be a $k\times k$ matrix.
It is $A\Sigma A^T$.
The distribution is still multivariate normal.