I am trying to solve the following exercise from my professor's notes on conditional expectation:

Let $x: \Omega \rightarrow \mathbb{R}^n$, $x \in G(0, Q_x)$, $Q_x = Q_x^T>0$, $y: \Omega \rightarrow \mathbb{R}^k$, $y \in G(0, Q_y)$. Assume that $x$, $y$ are jointly Gaussian random variables and that \begin{equation*} E[\exp(iu^Ty)|F^x] = \exp\left(iu^TCx - \frac{1}{2} u^T \tilde{Q}u\right), \end{equation*} for some $C \in \mathbb{R}^{k \times n}$, $\tilde{Q} = \tilde{Q}^T \geq 0$. Prove that there exists a random variable $w: \Omega \rightarrow \mathbb{R}^k$, such that $x$, $w$ are independent, $w \in G(0, Q_w)$ and $y = Cx + w$

My problem is that I cannot find a way to prove the existence (the inverse is easy to prove). I was thinking to use the property $E[E[\exp(iu^Ty)|F^x]] = E[\exp(iu^Ty)]$ (because the characteristic function would give an expression for $y$) but following this path it gets overly complex.


Define $w=y-Cx$, then $$ E[\mathrm e^{\mathrm iu^Tw}|\mathcal F^x] =E[\mathrm e^{\mathrm iu^Ty}|\mathcal F^x]\,\mathrm e^{-\mathrm iu^TCx} = \mathrm e^{-u^T \tilde{Q}u/2} $$ is independent of $x$. Thus $w$ is independent of $x$ and centered normal with covariance $\tilde{Q}$.

| cite | improve this answer | |
  • $\begingroup$ Thank you! Quite simple in the end and similar concept to the proof of the gaussian conditional expectation, but my mind didn't go there because I was thinking of existence theorems. $\endgroup$ – CuriousNik Feb 20 '14 at 10:13

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.