Simplifying Conditional Expectation with two Random Variables

185 Views Asked by At

In my introductory probability class I ran across these two expressions in a solution to a homework problem.

X and Y are two random variables, and f(Y) is any function.

$\mathbb{E}[\mathbb{E}[X|Y]f(Y)]$ = $\mathbb{E}[\mathbb{E}[f(Y)X|Y]]$

or without the outside expectiaton:

$\mathbb{E}[X|Y]f(Y)$ = $\mathbb{E}[f(Y)X|Y]$

Can someone help clarify, more intuitively, why these two expressions are equivalent? The only idea I have is that since we are "fixing" Y in the expectation, f(Y) acts almost as a constant. However, the "fixing" of Y happens in the expectation, yet the function of Y is multiplied afterwards. I know this is probably supposed to be very basic, but it simply isn't clicking for me.

1

There are 1 best solutions below

0
On

It reminds me of one of the basic properties of conditional expectations, namely "Taking out what's known": if $Y$ is $\mathcal{G}$- measurable and $XY$ - integrable, then $\mathbb{E}[XY|\mathcal{G}] = Y\mathbb{E}[X|\mathcal{G}]$.

The proof of this is quite standard - start by showing the result when Y is an indicator function, then apply linearity to prove it for $Y$ - simple, Monotone Convergence for any non-negative measurable function and finally use $Y = Y^+ -Y^-$ to complete the argument.

In your case $\mathcal{G}$ is simply the sigma-algebra generated by $Y$. However, I'm not entirely sure if $f$ can be any function - $f(Y)$ needs to remain $\sigma(Y)$-measurable.