2
$\begingroup$

I saw two versions of the law of iterated expectations, this one: \begin{align} E(E(Y\vert X)) = E(Y) \end{align} and this one: \begin{align} E(E(Y\vert X_1, X_2)\vert X_1) = E(Y \vert X_1) \end{align}

I was wondering if this one is also correct or not: \begin{align} E(E(Y\vert Z) \vert X) = E(Y\vert X) \end{align}

Here $X,X_1,X_2,Y,Z$ are random variables.

$\endgroup$
2
$\begingroup$

The conjecture is false. If $X=Y$, $Y$ and $Z$ are independent, $E(Y)=0$ then $E(Y|X)=Y$, $E(Y|Z)=0$, and $E(E(Y|Z)|X)=0$.

$\endgroup$
0
$\begingroup$

The reason the second version you state works but the third one does not is that "the smaller information set dominates", see this elaborate answer: A generalization of the Law of Iterated Expectations

Conditioning on $X_1,X_2$ means conditioning on more information than conditioning on $X_1$ alone, which implies that first conditioning on $X_1,X_2$ and then conditioning on $X_1$ is as good as the latter only.

In the third statement, we have no such relationship between $X$ and $Z$.

$\endgroup$

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.