For a random variable $T \geq 0$ with distribution function $F$, a real number $t > 0$ and $\mathcal{G}= \{ \emptyset, \{ T > t \}, \, \{ T \leq t \}, \, \Omega \}$, I need to evaluate the conditional expectation of $E|T-t||\mathcal{G}$, where $\mathcal{G}$ is my sub-$\sigma$-algebra.

Now, I asked somebody how to do this, and they told me that because the $\sigma$-algebra is so small, if I want to find this conditional expectation, all I have to do is find $P(T\leq x| T>t)$, which seems very strange to me. What is the reasoning behind this.

This person then said to apply this formula and then express everything in terms of $F$, but I don't know what the distribution is in this case, do I? So then, how would I go about doing this?

I am extremely confused...

I thank you in advance for your time and patience in helping me to become un-confused!

  • $\begingroup$ Can you give what $F$ is? Or the answer should be given in terms of $F$ without specifying what $F$ is? $\endgroup$ – Shashi Dec 21 '17 at 18:19
  • $\begingroup$ @Shashi the answer should be given in terms of $F$ wihtout specifying what $F$ is. .Can you do that? $\endgroup$ – ALannister Dec 21 '17 at 18:24
  • $\begingroup$ One more question is $T$ a continuous random variable? I don't think it affects the answer much, but it is nice to know $\endgroup$ – Shashi Dec 21 '17 at 18:29
  • $\begingroup$ @Shashi again, that is unknown. I gave all the information I had in my question above. $\endgroup$ – ALannister Dec 21 '17 at 18:31

$\newcommand{\PM}{\mathbb{P}}\newcommand{\E}{\mathbb{E}}$From your comments I think you want to keep it as general as possible. Let $(\Omega, \mathcal F ,\PM)$ be the probability space where everything is going on. Since you are talking about $T\geq 0$ then $T$ is a random variable from $(\Omega,\mathcal F)$ to $([0,\infty),\mathcal B_{[0,\infty)})$.

It is given that $\mathcal G=\{\emptyset, A,A^c,\Omega\}$, where $A=\{T>t\}$ and to make it a little bit interesting let us assume $\PM(A)\in(0,1)$ otherwise this exercise is boring $(\star)$. One knows that $Y:=\E[ |T-t|\mid \mathcal G]$ is the only (up to a null set) $\mathcal G$-measurable function that satisfies: \begin{align}\tag{1} \int_B |T-t|\,d\PM = \int_B Y\,d\PM ,\hspace{15pt} \forall_{ B\in\mathcal G} \end{align} I claim the following: \begin{align} Y(\omega)= \begin{cases} \E[\mathbf{1}_A|T-t|](\PM(A))^{-1} & \text{ if } & \omega\in A\\ \E[\mathbf{1}_{A^c}|T-t|](\PM(A^c))^{-1} & \text{ if } & \omega\in A^c\\ \end{cases} \end{align} You may verify that this indeed satisfies $(1)$.

We have a problem now, because we did not put everything in terms of $F$. Let's do it now. We have: \begin{align} \E[\mathbf{1}_A |T-t|](\PM(A))^{-1}&\stackrel{?}{=}\E[\mathbf{1}_AT](\PM(A))^{-1}-\E[\mathbf{1}_At](\PM(A))^{-1}\\ &=\E[\mathbf{1}_AT](\PM(A))^{-1}-t\\ &=\frac{\E[\mathbf{1}_AT]}{1-F(t)}-t \end{align} And you think about the question mark. Similarly we get: \begin{align} \E[\mathbf{1}_{A^c}| T-t|](\PM(A^c))^{-1}=t-\frac{\E[\mathbf{1}_{A^c}T]}{F(t)} \end{align} Putting these things together yields: \begin{align} \E[|T-t|\mid \mathcal G] = \mathbf{1}_{A^c}\left(t-\frac{\E[\mathbf{1}_{A^c}T]}{F(t)}\right)+\mathbf{1}_{A}\left(\frac{\E[\mathbf{1}_AT]}{1-F(t)}-t \right) \end{align} Do you now understand why we assumed $\PM(A)\in(0,1)$ at ($\star$)? What would change?


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.