7
$\begingroup$

Let $X_1, \ldots, X_n$ be i.i.d. random variables with expectation $a$ and variance $\sigma^2$, taking only positive values. Let $m < n$. Find the expectiation of $\displaystyle\frac{X_1 + \cdots + X_m}{X_1 + \cdots + X_n}$.

My attemps to solve this probles are rather straightforward. Denote $X = X_1 + \cdots + X_m$ and $Y = X_{m+1} + \dots + X_n$. So, $X$ has the expectation $ma$ and the variance $m\sigma^2$. And $Y$ has the expectation $(n-m)a$ and variance $(n-m)\sigma^2$. And also $X$ and $Y$ are independent. So we can compute the expectation by the definition $\mathbb{E}\displaystyle\frac{X}{X+Y} = \int\limits_{\Omega^2}\frac{X(\omega_1)}{X(\omega_1) + Y(\omega_2)}\mathbb{P}(d\omega_1)\mathbb{P}(d\omega_2)$. But we do not know the distribution, so we do not have chance to calculate it.

I would be glad to any help or ideas!

$\endgroup$
  • 6
    $\begingroup$ Hint. You have $$\mathbb{E}\left[ \frac{X_1}{X_1+\cdots+X_n} \right] = \cdots = \mathbb{E}\left[ \frac{X_n}{X_1+\cdots+X_n} \right] $$ and $$ \frac{X_1+\cdots+X_n}{X_1+\cdots+X_n} = 1. $$ $\endgroup$ – Sangchul Lee Sep 7 '17 at 10:29
  • $\begingroup$ @SangchulLee how do you prove $\mathbb{E}\left[ \frac{X_1}{X_1+\cdots+X_n} \right] = \mathbb{E}\left[ \frac{X_n}{X_1+\cdots+X_n} \right]$ ? It doesn't look obvious to me. $\endgroup$ – Gabriel Romon Sep 7 '17 at 12:30
  • 4
    $\begingroup$ @LeGrandDODOM, Since $X_1, \cdots, X_n$ are i.i.d., they are exchangable: for any permutation $\sigma$ on $\{1,\cdots,n\}$ we have the following equality in distribution: $$ (X_1, \cdots, X_n) \stackrel{d}{=} (X_{\sigma(1)}, \cdots, X_{\sigma(n)} ). $$ And since the expectation depends only on the distribution, we have $$ \mathbb{E}\left[\frac{X_1}{X_1+\cdots+X_n}\right] = \mathbb{E}\left[\frac{X_{\sigma(n)}}{X_{\sigma(1)}+\cdots+X_{\sigma(n)}}\right] = \mathbb{E}\left[\frac{X_{\sigma(n)}}{X_1+\cdots+X_n}\right]. $$ $\endgroup$ – Sangchul Lee Sep 7 '17 at 12:33
  • $\begingroup$ @SangchulLee if say $s_{i} = \frac{X_{i}}{\sum_{j=1}^{n}}X_{j}$, then while calculating expectation (in continuous case) we will have $s_{i}$ under the integrsl sign too,that is while calculating $E(s_{i})$ the term inside integral that is $s_{i}$ will be varing ,so how can we show that expectation of each $s_{i}$ are the same? or if $X_{i}$ are identically distributed then does that mean $s_{i}$ are identically distributed? I too get theintuition that they are same but how do i prove it? $\endgroup$ – BAYMAX Sep 16 '17 at 1:06
  • $\begingroup$ @BAYMAX, If $X_i$ has common p.d.f. $f$, then $$\mathbb{E}\left[\frac{X_i}{X_1+\cdots+X_n}\right]=\int_{(0,\infty)^n}\frac{x_i}{x_1+\cdots+x_n}f(x_1)\cdots f(x_n)\,dx_1\cdots dx_n. $$ Now you can interchange the role of $x_1$ and $x_i$ to find that this expectation does not depend on $i$. This line of reasoning can be extended to arbitrary distribution on $(0,\infty)$. $\endgroup$ – Sangchul Lee Sep 16 '17 at 2:03
11
$\begingroup$

Suppose $S_m=\sum\limits_{i=1}^{m} X_i$ and $S_n=\sum\limits_{i=1}^n X_i$.

Now, $$\frac{X_1+X_2+\cdots+X_n}{S_n}=1\,, \text{ a.e. }$$

Therefore,

$$ \mathbb E\left(\frac{X_1+X_2+\cdots+X_n}{S_n}\right)=1$$

Since $X_1,\ldots,X_n$ are i.i.d (see @SangchulLee's comments on main), we have for each $i$,

$$\mathbb E\left(\frac{X_i}{S_n}\right)=\frac{1}{n}$$

So for $m\le n$, $$\mathbb E\left(\frac{S_m}{S_n}\right)=\sum_{i=1}^m \mathbb E\left(\frac{X_i}{S_n}\right)=\frac{m}{n}$$

$\endgroup$
  • $\begingroup$ How to prove that $\frac{X_i}{S_n}$ and $\frac{X_j}{S_n}$ are independent? $\endgroup$ – Alex Grey Sep 7 '17 at 11:10
  • 1
    $\begingroup$ @AlexGrey That isn't really used anywhere. As the $X_i$'s are identically distributed, $E\left(\frac{X_1}{S_n}\right)=E\left(\frac{X_2}{S_n}\right)=...=E\left(\frac{X_n}{S_n}\right)$ and we simultaneously have $\sum_{i=1}^nE\left(\frac{X_i}{S_n}\right)=1$. $\endgroup$ – StubbornAtom Sep 7 '17 at 11:18
  • 5
    $\begingroup$ @AlexGrey, And in general they are not independent. $\endgroup$ – Sangchul Lee Sep 7 '17 at 11:21
  • $\begingroup$ @StubbornAtom How do you prove $E\left(\frac{X_1}{S_n}\right)=E\left(\frac{X_2}{S_n}\right)$ ? It doesn't look obvious to me. $\endgroup$ – Gabriel Romon Sep 7 '17 at 12:30
  • 1
    $\begingroup$ @LeGrandDODOM I feel we don't need anything more after Sangchul's comment. $\endgroup$ – StubbornAtom Sep 7 '17 at 13:10

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.