I want to compute a conditionnal expectation, i know that

$Z=(Z_1,\ldots,Z_p)'$ where $ Z_j=\Phi ^{-1}(U_j)$ with $Z \sim N(0,R(\theta))$ and $R(\theta)$ the $p \times p$ positive definite correlation matrix, $\Phi ^{-1}(U_j) $ is the standard normal quantile function and $U\sim[0,1]$.

I want to show $E[Z_i|U_j=u_j]=R_{ij}Z_j$. I tried by using the correlation between $ Z_i $ et $Z_j $

$R_{ij}=cor(Z_i,Z_j)=E[Z_iZ_j]-0 $ so

$R_{ij}=E[Z_iZ_j]=E[\Phi ^{-1}(U_i)\Phi ^{-1}(U_j)]=E[E[\Phi ^{-1}(U_i)\Phi ^{-1}(U_j)|U_j=u_j]]=E[\Phi ^{-1}(U_j)E[\Phi ^{-1}(U_i)|U_j=u_j]] =E[Z_jE[Z_i|U_j=u_j]]. $

But i don't know if it's the right way. Any ideas ?



The task is to compute $$ \mathbb E\left[Z_i|U_j \right].$$ Observe that $\Phi$ is invertible hence the $\sigma$-algebra generated by $U_j$ is the same as the $\sigma$-algebra generated by $Z_j$. We are thus reduced to compute $$ \mathbb E\left[Z_i|Z_j \right].$$ Then decompose $Z_i$ as $Z_i-\frac{R_{i,j}}{R_{j,j}}Z_j+ \frac{R_{i,j}}{R_{j,j}}Z_j$ to get $$ \mathbb E\left[Z_i|Z_j \right]=\mathbb E\left[Z_i- R_{i,j} Z_j|Z_j \right]+\mathbb E\left[ R_{i,j} Z_j|Z_j \right]. $$ For the first terms, use and show the independence of $Z_i- R_{i,j} Z_j$ and $Z_j$; the second term is $R_{i,j}Z_j$.


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.