5
$\begingroup$

Let

  • $(\Omega,\mathcal A,\operatorname P)$ be a probability space
  • $\mathcal F\subseteq\mathcal A$ be a $\sigma$-algebra on $\Omega$
  • $(E_i,\mathcal E_i)$ be a measurable space
  • $X_i$ be an $(E_i,\mathcal E_i)$-valued random variable on $(\Omega,\mathcal A,\operatorname P)$
  • $f:E_1\times E_2\to E_3$ be $(\mathcal E_1\otimes\mathcal E_2,\mathcal E_3)$-measurable
  • $X_3:=f(X_1,X_2)$

Assuming $X_2$ is independent of $\mathcal F$, are we able to show that $X_3$ is conditionally independent of $\mathcal F$ given $X_1$, i.e. $$\operatorname P\left[X_3\in B_3,F\mid X_1\right]=\operatorname P\left[X_3\in B_3\mid X_1\right]\operatorname P\left[F\mid X_1\right]\;\;\;\text{almost surely}\tag1$$ for all $B_3\in\mathcal E_3$ and $F\in\mathcal F$?

Let $B_3\in\mathcal E_3$ and $F\in\mathcal F$. We need to prove that $$\operatorname P\left[X_1\in B_1,X_3\in B_3,F\right]=\operatorname E\left[1_{\{\:X_1\:\in\:A\:\}}\operatorname P\left[X_3\in B_3\mid X_1\right]\operatorname P\left[F\mid X_1\right]\right]\tag2.$$ What's the easiest way to show $(2)$? Maybe we are able to reduce the problem to the case $f^{-1}(B_3)=A_1\times A_2$ for some $A_i\in\mathcal E_i$, but I'm missing the right argument for that.

EDIT: If necessary, feel free to impose a stronger notion of measurability on $f$.

$\endgroup$
2
+25
$\begingroup$

This is false. Indeed, consider Bernstein's example:

  • $\Omega = \{\omega = (\omega_1,\omega_2)\mid \omega_i \in \{0,1\}\}$,
  • $\mathrm P(\omega) = 1/4, \omega \in \Omega$,
  • $X_i(\omega) = \omega_i$, $i=1,2$;
  • $X_3 = (X_1+X_2) \mod 2$, $\mathcal F = \sigma(X_3)$.

Then, $X_1,X_2,X_3$ are pairwise independent, but not jointly independent. So for $B_3 = \{0\}$ and $F = \{X_3 = 0\}$ $$ \mathrm P \left[X_3=0,F\mid X_1\right] = \mathrm P [F] = 1/2\neq 1/4 = \mathrm P[F]^2 = \mathrm P \left[X_3=0\mid X_1\right] \cdot \mathrm P \left[F\mid X_1\right]. $$

Note that in this counterexample we even have that $X_1$ and $X_2$ are independent and $X_1$ and $\mathcal F$ are independent. So it is hard to imagine an additional assumption which would make this true (except that the pair $(X_1,X_2)$ is independent of $\mathcal F$, which makes this trivial).

| cite | improve this answer | |
$\endgroup$

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.