Let $(\Omega,\mathcal{F},P)$ be a probability space, and $A\in\mathcal{F}$ such that $P(A)>0$. Let $P_{A}$ denote the conditional probability measure given $A$ defined on $(\Omega,\mathcal{F})$ Now obviously $dP_A/dP=1_A/P(A)$, so then for any $X\in L^1$

$$E[X|A]:=E_{P_A}[X]=\int XdP_A=\int X\frac{dP_A}{dP}dP=E[X1_A]/P(A)$$ Now let $\sigma(A)=\{\emptyset,A,\Omega\backslash A,\Omega\}$ denote the sigma filed generated by $A$. So we must have for any $\omega\in A$, $E[X|\sigma(A)](\omega)=E[X|A]1_A(\omega)$. Is that true? But I have trouble showing that that for any $\omega\in A$ and r.v. $X\in L^1$

$$E[X|\sigma(A)](\omega)=\frac{1}{P(A)}\int_\Omega X1_A(\omega)dP=\frac{1}{P(A)}\int_AXdP=E[X1_A(\omega)]/P(A):=E[X|A]$$

So I canot prove the above equation, though it seems quite obvious but am quite stuck or is it wrong?. I started with the definition of conditional expectation but am getting nowhere, the only thing I can see is, $E[X|\sigma(A)]=E[X|\sigma(A)]1_A+E[X|\sigma(A)]1_{\Omega\backslash A}$, and can't see where I can pull out $1/P(A)$, so any help will be greatly appreciated. So the question is in fact how are $E[X|\sigma(A)]$ and $E[X|A]$ related? If there are any mistakes or if I have got hings wrong please correct me. Thanks and any help feedback is most needed and appreciated.

  • 1
    $\begingroup$ As in the very recent question that (you probably deleted and) this one is a duplicate of, "E[X|σ(A)](ω)=E[X|A]1A(ω)" is pure fantasy, as well as the displayed formula right after it. (No need to ask a third question, these inconvenient facts will remain the same.) Do you have a definition of E[X|G] at your disposal, when G is a sigma-algebra? $\endgroup$ – Did May 23 '14 at 11:03
  • $\begingroup$ First of all I may not of put things across correctly, so confusion on my part as well as yours might of arisen, my apologies. I was trying to verify a comment in Koralav and Sinai and what I am trying to say is that on $A$, we have $E[X|\sigma(A)]=E[X|A]$. I now see this is (trivially)true as on $A$ $$\int_A E[X|\sigma(A)]dP= \int_A XdP=E[X1_A]=(E[X1_A]/P(A))\int_A1dP=\int_AE[X|A]dP$$. In fact $E[X|\sigma(A)]=E[X|A]1_A+E[X|\Omega\backslash A]1_{\Omega\backslash A}$. $\endgroup$ – user152874 May 23 '14 at 14:53
  • $\begingroup$ Furthermore if $\mathcal{G}$ is a sub sigma field of $\mathcal{F}$ generated by a countable partition $\{G_i\}_{i=0}^{\infty}$ with $P(G_0)=0$ and $P(G_i)>0$ $\forall i=1,2,...$ then $E[X|\mathcal{G}]=\sum\limits_{i=1}^{\infty} E[X|G_i]1_{G_i}$ $\endgroup$ – user152874 May 23 '14 at 14:59
  • $\begingroup$ Which confusion "on my part"? Curious statement... // Does the answer below allow to close this discussion? $\endgroup$ – Did May 23 '14 at 18:13
  • $\begingroup$ Sorry, contrarily to what I suggested above, you did not delete the previous question, you probably logged under a new username to repost the same one. $\endgroup$ – Did May 23 '14 at 23:08

Indeed, $E[X\mid\sigma(A)]=E[X\mid A]\,\mathbf 1_A+E[X\mid \Omega\setminus A]\,\mathbf 1_{\Omega\setminus A}$.

| cite | improve this answer | |

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.