If $\mathbf {Z}$ is random vector and $A$ is a fixed matrix, could someone explain why $$\mathrm{cov}[A \mathbf {Z}]= A \mathrm{cov}[\mathbf {Z}]A^\top.$$


For a random (column) vector $\mathbf Z$ with mean vector $\mathbf{m} = E[\mathbf{Z}]$, the covariance matrix is defined as $\operatorname{cov}(\mathbf{Z}) = E[(\mathbf{Z}-\mathbf{m})(\mathbf{Z}-\mathbf{m})^T]$. Thus, the covariance matrix of $A\mathbf{Z}$, whose mean vector is $A\mathbf{m}$, is given by $$\begin{align}\operatorname{cov}(A\mathbf{Z}) &= E[(A\mathbf{Z}-A\mathbf{m})(A\mathbf{Z}-A\mathbf{m})^T]\\ &= E[A(\mathbf{Z}-\mathbf{m})(\mathbf{Z}-\mathbf{m})^TA^T]\\ &= AE[(\mathbf{Z}-\mathbf{m})(\mathbf{Z}-\mathbf{m})^T]A^T\\ &= A\operatorname{cov}(\mathbf{Z})A^T. \end{align}$$

  • 1
    $\begingroup$ I corrected my typo. Thanks for pointing out my error. $\endgroup$
    – user92612
    Aug 29 '14 at 14:54

I would add to the answer by Dilip Sarwate that the same result holds also for transformation of the form $\mathbf{Z}A^T$: $$\mathrm{cov}(\mathbf{Z}A^T) = A\mathrm{cov}(\mathbf{Z})A^T$$

Using the same approach: $$ \begin{align} \mathrm{cov}(\mathbf{Z}A^T)&=\mathbb{E}[(\mathbf{Z}A^T-\mathbf{m}A^T)(\mathbf{Z}A^T-\mathbf{m}A^T)^T] \\ &=\mathbb{E}[(\mathbf{Z}-\mathbf{m})A^TA(\mathbf{Z}-\mathbf{m})^T] \\ &=\mathbb{E}[A(\mathbf{Z}-\mathbf{m})(\mathbf{Z}-\mathbf{m})^TA^T] \\ &=A\mathbb{E}[(\mathbf{Z}-\mathbf{m})(\mathbf{Z}-\mathbf{m})^T]A^T \\ &=A\mathrm{cov}(\mathbf{Z})A^T \\ \end{align} $$

Using $AB^TBA^T=BAA^TB^T$ in step (3): $$\begin{align}AB^TBA^T &= \left(\left(AB^TBA^T\right)^T\right)^T \\ &= \left(BA^TAB^T\right)^T \\ &= B\left(BA^TA\right)^T \\ &= BAA^TB^T \end{align}$$

  • $\begingroup$ Why is $AB^TBA^T=BAA^TB^T$? $\endgroup$ Feb 21 '20 at 17:42
  • $\begingroup$ Looking at it now, this step seems incorrect to me... Thanks for pointing that out! $\endgroup$ Feb 25 '20 at 11:17

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.