5
$\begingroup$

If $X$ is an integrable random variable on probability space $(\Omega, \mathcal{F}, \mathbb{P})$ and $\mathcal{G_1}, \mathcal{G_2}$ sub sigma fields of $\mathcal{F}$ then how can we find an example where $\Omega = \{a, b, c\}$ in which $$E[E(X\mid\mathcal{G_1})\mid\mathcal{G_2}] \ne E[E(X\mid\mathcal{G_2})\mid\mathcal{G_1}].$$

I would really appreciate if you could analytically show the steps of conclusion. I am self learning.

$\endgroup$
3
$\begingroup$

$\def\Ω{{\mit Ω}}\def\F{\mathscr{F}}\def\G{\mathscr{G}}\def\emptyset{\varnothing}$Take $\F = 2^\Ω$ and denote $p(ω) = P(\{ω\})$ for $ω \in \Ω = \{a, b, c\}$. If$$ \G_1 = \{\Ω, \emptyset, \{a\}, \{b, c\}\},\ \G_2 = \{\Ω, \emptyset, \{b\}, \{a, c\}\},$$ then for any random variable $X$ on $(Ω, \F)$,\begin{gather*} E(X \mid \G_1)(ω) = \begin{cases} X(a); & ω \in \{a\}\\ \dfrac{p(b) X(b) + p(c) X(c)}{p(b) + p(c)}; & ω \in \{b, c\} \end{cases},\\ E(X \mid \G_2)(ω) = \begin{cases} X(b); & ω \in \{b\}\\ \dfrac{p(a) X(a) + p(c) X(c)}{p(a) + p(c)}; & ω \in \{a, c\} \end{cases}, \end{gather*} which implies\begin{gather*} E(E(X \mid \G_1) \mid \G_2)(ω) = \begin{cases} \dfrac{p(b) X(b) + p(c) X(c)}{p(b) + p(c)}; & ω \in \{b\}\\ \small\dfrac{p(a)(p(b) + p(c)) X(a) + p(b)p(c) X(b) + (p(c))^2 X(c)}{(p(a) + p(c))(p(b) + p(c))}; & ω \in \{a, c\} \end{cases},\\ E(E(X \mid \G_2) \mid \G_1)(ω) = \begin{cases} \dfrac{p(a) X(a) + p(c) X(c)}{p(a) + p(c)}; & ω \in \{a\}\\ \small\dfrac{p(a)p(c) X(a) + p(b)(p(a) + p(c)) X(b) + (p(c))^2 X(c)}{(p(a) + p(c))(p(b) + p(c))}; & ω \in \{a, c\} \end{cases}. \end{gather*} In order to have $E(E(X \mid \G_1) \mid \G_2) ≠ E(E(X \mid \G_2) \mid \G_1)$, it suffices to make$$ E(E(X \mid \G_1) \mid \G_2)(a) ≠ E(E(X \mid \G_2) \mid \G_1)(a),$$ i.e.$$ \frac{p(a)(p(b) + p(c)) X(a) + p(b)p(c) X(b) + (p(c))^2 X(c)}{(p(a) + p(c))(p(b) + p(c))} ≠ \frac{p(a) X(a) + p(c) X(c)}{p(a) + p(c)}, $$ which can be simplified as $X(b) ≠ X(c)$ assuming $p(ω) > 0$ for $ω \in \Ω$.

To summarize, it suffices to take $p(a) = p(b) = p(c) = \dfrac{1}{3}$ and $X = I_{\{c\}}$.

$\endgroup$
1
  • $\begingroup$ @Batominovski Thanks, I didn't edit this one when copy-pasting. $\endgroup$
    – Saad
    Apr 25 '20 at 14:07

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.