5
$\begingroup$

I've been working on this problem for 3 hours now, and my complete lack of progress is getting disheartening. I've looked up definitions, proofs, and have even seen a solution for this particular problem. However, I did not understand the steps the solution used, which means I have no business reading it. I'm hoping that if I show my attempt, y'all can give me some pointers. Thanks in advance!

Here is my work:

$$\mathrm{Cov}(X,Y)=E[(X-E[X])(Y-E[Y])] =E[XY-X*E[Y]-Y*E[X]+E[X]*E[Y]]$$

By linearity of expected value:

=E[XY]-E[X*E[Y]]-E[Y*E[X]]+E[E[X]*E[Y]], as E[Y] and E[X] are constants, E[X*E[Y]]=E[Y]*E[X]

=E[XY]-E[X]E[Y]-E[X]E[Y]+E[X]E[Y]

=E[XY]-E[X]E[Y]

Cov(X,E[Y|X])

=E[(X-E[X])(E[Y|X]-E[E[Y|X]])]

=E[(X-E[X])(E[Y|X]-E[Y])]

=E[X*E[Y|X]-X*E[Y]-E[X]E[Y|X]+E[X]E[Y]]

=E[X*E[Y|X]]-E[X*E[Y]]-E[E[X]E[Y|X]]+E[E[X]E[Y]]

=E[X]E[Y|X]-E[X]E[Y]-E[X]E[E[Y|X]]+E[X]E[Y]

=0

I really have no idea where to go with this problem. If E[X*E[Y]]=E[X]E[Y], I cannot figure out why E[XE[Y|X]]$\neq$E[X]E[Y|X]. I believe I am making a mistake when I go from the bold line to the next, but I can't figure out why.

$\endgroup$
  • $\begingroup$ First observe that it suffices to prove $Cov(X,Y)=Cov(X,E[Y|X])$ for $X,Y$ with $E(X)=E(Y)=0$. Now, can you prove this? $\endgroup$ – Miheer Nov 24 '13 at 6:14
  • $\begingroup$ Also observe that $E(XE[X|Y])$ is a constant while $E(X)E[Y|X]$ is a random variable, so they may not always be equal. $\endgroup$ – Miheer Nov 24 '13 at 6:24
3
$\begingroup$

If E[X*E[Y]]=E[X]E[Y], I cannot figure out why E[XE[Y|X]]≠E[X]E[Y|X].

Indeed, for every integrable random variables $X$ and $Y$, $E[XE[Y]]=E[X]E[Y]$ since $E[Y]$ is simply a constant and one knows that $E[\lambda X]=\lambda E[X]$ for every constant $\lambda$.

But $Z=E[Y\mid X]$ is a random variable and, for some general random variable $Z$, $E[XZ]\ne E[X]Z$ simply because the LHS is a number and the RHS a random variable.

What happens here is that $Z$ is a specific random variable, namely, the conditional expectation of some random variable $Y$ with respect to $X$. And it is a fact (actually, almost part of the definition) that $E[u(X)E[Y\mid X]]=E[u(X)Y]$ for every measurable function $u$ (technically, such that $u(X)Y$ is integrable). You might want to apply this to your problem (if you try and still fail to solve it, I might add some further hints).

$\endgroup$

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.