• $(\Omega,\mathcal A,\operatorname P)$ be a probability space
  • $\mathcal F$, $\mathcal G$ and $\mathcal H$ be $\sigma$-algebras on $\Omega$ with $\mathcal F\subseteq\mathcal G\subseteq\mathcal H\subseteq\mathcal A$
  • $(E,\mathcal E)$ be a measurable space
  • $X,Y,Z:\Omega\to E$ be $(\mathcal F,\mathcal E)$-, $(\mathcal G,\mathcal E)$- and $(\mathcal H,\mathcal E)$-measurable, respectively

Assume $$\operatorname P\left[Y\in B\mid\mathcal F\right]=\operatorname P\left[Y\in B\mid X\right]\;\;\;\text{for all }B\in\mathcal E\tag1.$$ How can we conclude that $$\operatorname E\left[1_{\left\{\:Y\:\in\:B\:\right\}}\operatorname E\left[1_{\left\{\:Z\:\in\:C\:\right\}}\mid Y\right]\mid\mathcal F\right]=\operatorname E\left[1_{\left\{\:Y\:\in\:B\:\right\}}\operatorname E\left[1_{\left\{\:Z\:\in\:C\right\}}\mid Y\right]\mid X\right]\tag2$$ for all $B,C\in\mathcal E$?

Seems to be easy, but I can't figure out how to do it.


If we can prove that

$$\mathbb{E}(U \mid \mathcal{F}) = \mathbb{E}(U \mid X) \tag{3}$$

for any bounded function $U:\Omega \to \mathbb{R}$ which is measurable with respect to $\sigma(Y)$, then this gives $(2)$; simply choose $U := 1_{\{Y \in B\}} \mathbb{E}(1_{\{Z \in C\}} \mid Y)$ which is clearly bounded and $\sigma(Y)$-measurable.

The proof of $(3)$ is a standard monotone class argument:

  • It follows from $(1)$ that $(3)$ holds for functions $U$ of the form $U=1_{\{Y \in B\}}$.
  • Because of the linearity of the conditional expectation, this implies that $(3)$ holds for bounded step functions which are $\sigma(Y)$-measurable.

  • If $U$ is bounded and $\sigma(Y)$-measurable, there exists a sequence of $\sigma(Y)$-measurable step functions $(U_j)_{j \in \mathbb{N}}$ such that $U_j \to U$ and $|U_j| \leq |U|$. Since we already know that $(3)$ holds for each $j$, we can use the dominated convergence theorem to conclude that $$\mathbb{E}(U \mid \mathcal{F}) = \lim_{j \to \infty} \mathbb{E}(U_j \mid \mathcal{F}) = \lim_{j \to \infty} \mathbb{E}(U_j \mid X) = \mathbb{E}(U \mid X).$$

| cite | improve this answer | |
  • $\begingroup$ You're right. We should even be able to generalize it as follows: Let $\mathcal G\subseteq\mathcal F\subseteq\mathcal A$ be $\sigma$-algebras on $\Omega$ and $X:\Omega\to E$ be $(\mathcal A,\mathcal E)$-measurable with $\operatorname P\left[X\in B\mid\mathcal F\right]=\operatorname P\left[X\in B\mid\mathcal G\right]$ for all $B\in\mathcal E$, then $\operatorname E\left[Y\mid\mathcal F\right]=\operatorname E\left[Y\mid\mathcal G\right]$ for all bounded $\sigma(X)$-measurable $Y:\Omega\to\mathbb R$; if I'm not missing something. $\endgroup$ – 0xbadf00d Oct 14 '18 at 15:09
  • 1
    $\begingroup$ @0xbadf00d Yeah sure. $\endgroup$ – saz Oct 14 '18 at 15:10
  • $\begingroup$ My question was motivated by a different problem. I've asked a separate question for that one. Maybe you can help there too. $\endgroup$ – 0xbadf00d Oct 14 '18 at 16:05

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.