• $(\Omega,\mathcal{A},\operatorname{P})$ be a probability space
  • $I\subseteq\mathbb{R}$
  • $(\mathcal{F}_t,t\in I)$ be a filtration on $(\Omega,\mathcal{A})$
  • $Y\in\mathcal{L}^1(\operatorname{P})$ and $$X_t:=\operatorname{E}\left[Y\mid\mathcal{F}_t\right]\;\;\;\text{for }t\in I$$

I don't understand why it holds $$\operatorname{E}\left[\operatorname{E}\left[\left|Y\right|\mid\mathcal{F}_t\right]\right]=\operatorname{E}\left[|Y|\right]\tag{1}$$

$(1)$ would imply $$\operatorname{E}\left[\left|Y\right|\mid\mathcal{F}_t\right]=|Y|\;\;\;\color{blue}{\operatorname{P}\text{-almost surely}}$$ So, $Y$ would need to be $\mathcal{F}_t$-measurable. I really don't understand why this should hold.

Please note:

A random variable $Z$ is called conditional expectation of $X$ given $\mathcal{F}\subseteq\mathcal{A}$ $:\Leftrightarrow$

  • $Z$ is $\mathcal{F}$-measurable
  • $\operatorname{E}[1_AX]=\operatorname{E}[1_AZ]$ for all $A\in\mathcal{F}$

We write $\operatorname{E}[X\mid\mathcal{F}]=:Z$.

  • 1
    $\begingroup$ No (1) does not imply that. Likewise E(U)=E(V) does not imply U=V P-almost surely. $\endgroup$
    – Did
    Commented Apr 2, 2015 at 20:20
  • $\begingroup$ @Did You're right. But we've got $\operatorname{E}[X]=\operatorname{E}[Y]$ $\Rightarrow$ $X=Y$ almost surely, if $X\ge Y$ almost surely. I'd consufed that. $\endgroup$
    – 0xbadf00d
    Commented Apr 2, 2015 at 20:26
  • 1
    $\begingroup$ It is not true that $E[|Y||\mathcal{F}_t]\leq |Y|$ almost surely, nor is the reverse true. $\endgroup$
    – shalop
    Commented Apr 2, 2015 at 20:28
  • $\begingroup$ No, @Shalop's point is different and, should I admit, slightly subtler than mine (and I see nothing wrong with the notations in this comment). Say, 0/oxbadfo/00/od, what are you playing at? $\endgroup$
    – Did
    Commented Apr 2, 2015 at 20:32
  • $\begingroup$ @Did He has edited his comment. $\endgroup$
    – 0xbadf00d
    Commented Apr 2, 2015 at 20:46

1 Answer 1


If $X$ is any random variable, and $\mathcal{H} \subset \mathcal{A}$ is any sub-sigma-algebra, then by the definition of conditional expectation, it is always true that $$E[E[X|\mathcal{H}]] = \int_{\Omega} E[X|\mathcal{H}] dP = \int_{\Omega} X dP = E(X)$$ Now just apply this when $X=|Y|$ and $\mathcal{H}=\mathcal{F}_t$.

As Did has pointed out, the equality of the expectations does not imply a.s. equality of the random variables.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .