8
$\begingroup$

Does $E[E(X|Y)|Z] =E[X|Y,Z]$? Also, what about $E[E(X|Y=y)|Z=z] =E[X|Y=y,Z=z]$ I'm confused by the relations. It seems intuitively to be the case. If it is correct, how do I mathematically prove it. I've searched on this site and elsewhere...

$\endgroup$
  • 3
    $\begingroup$ Is this a question from a course or textbook? If so, please add the [self-study] tag & read its wiki. $\endgroup$ – gung Dec 1 '16 at 16:41
  • 1
    $\begingroup$ No, this is not from a course or textbook. Just trying to understand for myself. $\endgroup$ – KUZ Dec 1 '16 at 17:26
10
$\begingroup$

Those two conditional expectations differ in general:$$\mathbb{E}[\mathbb{E}(X|Y)|Z] \ne\mathbb{E}[X|Y,Z]$$

As a matter of fact, strictly speaking, they do not even live in the same functional space as the first one is a function of $Z$, measurable wrt $\sigma(Z)$, the $\sigma$ algebra induced by $Z$, while the second one is a function of $(Y,Z)$, hence measurable wrt $\sigma(Y,Z)$, the $\sigma$ algebra induced by $(Y,Z)$,

As a counter-example, consider the setting when

  1. $X$ and $Y$ are independent
  2. $X$ and $Z$ are dependent, with $\mathbb{E}[X|Z]\ne \mathbb{E}[X]$

Then, because of the independence between $X$ and $Y$, $\mathbb{E}(X|Y)=\mathbb{E}[X]$ and therefore$$\mathbb{E}[\mathbb{E}(X|Y)|Z]= \mathbb{E}[X]\ne\mathbb{E}[X|Y,Z]$$

A valid equality is instead$$\mathbb{E}[\mathbb{E}(X|Y,Z)|Z]=\mathbb{E}[X|Z]$$that holds for all dependence relations between the three random variables.


Notations: The difference between the notations $\mathbb{E}[\mathbb{E}(X|Y)|Z]$ and $\mathbb{E}[\mathbb{E}(X|Y=y)|Z=z]$ is that

  1. $\mathbb{E}[\mathbb{E}(X|Y)|Z]$ is a random variable, transform of the random variable $Z$ (and not of the random variable $Y$ since $Y$Y is also conditioned on $Z$);
  2. $\mathbb{E}[\mathbb{E}(X|Y=y)|Z=z]$ is a function of apparently both $y$ and $z$, but in fact only of $y$ (as explained below) which has no clear meaning from a probabilistic point of view. Indeed, for a given value $y$, $\mathbb{E}(X|Y=y)$ is a constant for which taking a conditional expectation conditional on the realisation $Z=z$ makes little sense as it also returns $\mathbb{E}(X|Y=y)$. For instance, if $X$ depends on both $Y$ and $X$ as a random variable, for a given realisation $y$ of $Y$ and $Z$ of $z$, $\mathbb{E}(X|Y=y)$ is a constant that generally differs from $\mathbb{E}(X)$ and from $\mathbb{E}(X|Y=y,Z=z)$. But $\mathbb{E}[\mathbb{E}(X|Y=y)|Z=z]$ is not a realisation of the random variable $\mathbb{E}[\mathbb{E}(X|Y)|Z]$. The correct realisation is $\mathbb{E}[\mathbb{E}(X|Y)|Z=z]$
$\endgroup$
  • $\begingroup$ Thank you for this comment, Xi'an. I am unclear about your response on one thing: in your counterexample, if X and Y are independent (therefore $\mathbb{E}(X|Y)= \mathbb{E}[X]$) and if X and Z are dependent, then why is $\mathbb{E}[\mathbb{E}(X|Y)|Z]= \mathbb{E}[\mathbb{E}(X)|Z]= \mathbb{E}[X]$? $\endgroup$ – KUZ Dec 1 '16 at 22:07
  • $\begingroup$ ..because $\mathbb{E}[X]$ is a constant... $\endgroup$ – Xi'an Dec 1 '16 at 22:09

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.