Say I have two independent variables $X$ and $Y$ that are exponentially distributed with respective rates $\lambda_X$ and $\lambda_Y$. How do I compute $\mathbb{E}[X\mid \min\{X,Y\}]$?

  • $\begingroup$ Which approaches to compute conditional expectations do you know? $\endgroup$ – Did Apr 27 '14 at 21:45
  • $\begingroup$ I understand how to compute conditional expectations when the conditioned variable is independent of the other variables. For example $E[aX + bZ|Y]$. Also I can compute the answer if $Y$ takes a finite amount of values. $\endgroup$ – Laxy_economist Apr 27 '14 at 22:23
  • $\begingroup$ No general formula/definition? $\endgroup$ – Did Apr 27 '14 at 22:28
  • $\begingroup$ This is the only formula that come to mind:$$f_{X|Y}(x|y) = \frac{f_{X,Y}(x,y)}{f_Y(y)}$$ integrated over the correct bounds. $\endgroup$ – Laxy_economist Apr 27 '14 at 22:36
  • 1
    $\begingroup$ After a while, some serious definitions ARE useful... For a good introduction, quite readable, you might try Probability with martingales by David Williams. $\endgroup$ – Did Apr 27 '14 at 22:41

(Answering a query made in the comments.)

Recall that, for every integrable random variable $X$, $E(X\mid Z)$ is defined as the almost surely unique random variable $u(Z)$ such that $E(Xv(Z))=E(u(Z)v(Z))$ for every bounded measurable function $v$.

In words, $E(X\mid Z)$ is at the same time measurable with respect to $Z$ and similar to $X$ in that $E(X\mid Z)$ and $X$ give the same expectation when integrated against any function of $Z$.

In the present case, $Z=\min\{X,Y\}$ hence one needs the distribution of $Z$ to compute $E(u(Z)v(Z))$. Next, the task is to compute $E(Xv(Z))$. And finally one must equate these, that is, find the unique $u$ such that these coincide for every $v$...

The final result is $$ E(X\mid\min\{X,Y\})=\min\{X,Y\}+\frac{\mu}{\lambda+\mu}\,\frac1{\lambda}, $$ which has some nice interpretations in terms of marked Poisson processes or other similar notions.

  • $\begingroup$ I guess you mean "Marked Poisson processes". $\endgroup$ – Arash Apr 28 '14 at 22:06
  • $\begingroup$ @Arash I do. Thanks. $\endgroup$ – Did Apr 29 '14 at 5:52

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.