Is $Cov(A,BA)$ equal to $Var(A)E(B)$ when $A$ is a vector and $B$ is a matrix?

105 Views Asked by At

I have a square matrix $B_{nxn}$ and a vector $A_{nx1}$ where $B$ and $A$ are independent with each other. I solved $Cov(A,BA)$ and obtained $Var(A)E(B)$ as follows,

$Cov(A,BA)=E(A^2)E(B)-E(A)(E(A)E(R))$

$=[E(A^2)-(E(A))^2]E(B)$

$=Var(A)E(B)$

$Var(A)$ will result in a scalar value. But, $B$ is a matrix. So, the result just not justify enough for $Cov(A,BA)$ which is a scalar value.

Please be kind enough to provide any instruction on the mistake I have made.

If $Var(A)E(B)$ is not correct, how can $Cov(A,BA)$ be solved and what is the answer for that?

1

There are 1 best solutions below

9
On BEST ANSWER

Since we are dealing with vectors and matrices, we must be careful with operations like squares and so on. The calculation would go like this I think:

\begin{eqnarray*}Cov(A,BA)&=&E[(A-E[A])(BA-E[BA])^T]\\ &=&E[(A-E[A])(B(A-E[A]))^T]\\ &=&E[(A-E[A])(A-E[A])^TB^T]\\ &=&Cov(A)E[B]^T\end{eqnarray*}

Here $Cov(A)$ is the covariance matrix of the vector $A$, which corresponds to $Var(A)$. So the result is basically what you expected, but we need to respect matrix rules.