Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I want to use a rule for conditional expectation I found in (German) wikipedia, not in my script/textbook of probability theory, I guess it should be simple and follow more or less straight from the general definition (I want have a proof to be sure that I don't build up on a wikipedia mistake)

Let X be independent of Z and of Y (XY integrable and X,Y,Z random variables) $$ E(XY|Z) = E(X) E(Y|Z) $$

My idea: Showing that the rhs meets the conditions of the general definition of E(XY|Z), that is (i) it should be $\sigma(Z)$-measureable, check. (ii) $E\left( E(X) E(Y|Z) 1_A \right) \stackrel{!}{=} E(XY 1_A) \forall A \in \sigma(Z)$ Now the lhs$=E(X) E(E(Y|Z)1_A) = E(X)E(Y 1_A)$ (according to (ii) of the definition of $E(Y|Z)$, for all $A\in \sigma(Z)$)

$= E(XY1_A)$ as wished (X, Y are independent).

But, in this proof I did not use that X,Z are independent, so it would follow as well $E(XY|Z)=E(Y)E(X|Z)=E(Y)E(X)=E(XY)$ which shouldn't be this way.

Maybe it's all much simpler and I have just the wrong point of view on it. Q: Does anybody see the flaw in my proof? Can anybody hint me to a proof or a reference to a proof?

(@Didier, i try to get Williams book, I have to see if my library can get it for me).

share|improve this question
    
The faulty step is when you assert that $E(X)E(Y1_A)=E(XY1_A)$. Here you must not only assume that $X$ is independent of $Y$ (which is not enough to conclude) but that $X$ is independent of $(Y,Z)$. (+1 for showing the steps you tried.) –  Did Aug 4 '11 at 8:56
    
The usual counterexample works: take X and Y i.i.d. centered Bernoulli random variables and Z=XY. Then (X,Y) is independent (by definition), as are (Y,Z) and (Z,X) (easy to check), but (X,Y,Z) is not (for example P(X=Y=1,Z=-1)=0 instead of $\frac18$). And E(XY|Z)=Z although E(X)E(Y|Z)=0. –  Did Aug 4 '11 at 12:05
    
@did I think your first comment serves as a good answer for me already. By (Y,Z) you mean $\sigma(\sigma(Y), \sigma(Z))$? And independence of Y, Z might not be enough to have independence from (Y,Z) (see my new question ) –  Johannes L Aug 4 '11 at 12:34
    
Yes, [X is independent of (Y,Z)] means that the sigma-algebras sigma(X) and sigma(Y,Z) are independent. And sigma(Y,Z) coincides with sigma(sigma(Y),sigma(Z)). –  Did Aug 4 '11 at 13:37
    
So I would have to have (i) X independent from (Y,Z).. I wonder if that follows if I assume Y and Z independent (which would be more natural to my application than to have to introduce the assumption (i)) - i make a new question out of this. –  Johannes L Aug 5 '11 at 11:47

1 Answer 1

up vote 1 down vote accepted

The faulty step is when you assert that $E(X)E(Y1_A)=E(XY1_A)$. Here one must not only assume that $X$ is independent on $Y$ but that $X$ is independent on $(Y,Z)$.

To see that there is a difference, the usual example works here: take $X$ and $Y$ i.i.d. centered Bernoulli random variables and $Z=XY$. Then $X$ and $Y$ are independent (by definition), as are $Y$ and $Z$, as are $Z$ and $X$ (easy to check), but $X$, $Y$ and $Z$ are not*. And $E(XY|Z)=Z$ although $E(X)E(Y|Z)=0$.

*For example, $P(X=1,Y=1,Z=-1)=0$ but $P(X=1)P(Y=1)P(Z=-1)=1/8$.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.