0
$\begingroup$

Consider two random variables $X_1=\min (W_1, W_2)$ and $ X_2=\min (W_3, W_4),$ where $W_1$, $W_2$,$W_3$ and $W_4$ are positive, identically distributed random variables. While $W_1$, $W_2$ are independent $W_3$, $W_4$ are correlated. We assume that the range of both $X_1$ and $X_2$ is from $0$ to $1$. Is it true that

$$\Pr \{X_2 \leq x \} \leq \Pr \{X_1 \leq x \},~~~ \forall x?$$

If yes, how to prove it? I would be grateful if any pointers to existing literature are given. Thanks!

$\endgroup$
3
  • 1
    $\begingroup$ "Correlated", like in positively correlated? $\endgroup$
    – Did
    Commented Jan 26, 2015 at 9:01
  • $\begingroup$ Yes we can assume as in positively correlated.. $\endgroup$
    – Oliver
    Commented Jan 26, 2015 at 11:47
  • 1
    $\begingroup$ We "can"? It seems we must definitely assume that... $\endgroup$
    – Did
    Commented Jan 26, 2015 at 19:54

1 Answer 1

2
$\begingroup$

$$ \begin{eqnarray*} P\left(X_2> x\right)&{}={}&P\left(W_3> x,\ W_4> x\right)\newline &{}={}&\mathbb{E}\left[{\bf{1}}_{\left\{W_3> x,\ W_4> x\right\}}\right]\newline &{}={}&\mathbb{E}\left[{\bf{1}}_{\left\{W_3> x\right\}}{\bf{1}}_{\left\{W_4> x\right\}}\right]\newline &{}={}&\mathbb{E}\left[{\bf{1}}_{\left\{W_3> x\right\}}\right]\mathbb{E}\left[{\bf{1}}_{\left\{W_4> x\right\}}\right]{}+{}\mathbb{C}ov\left({\bf{1}}_{\left\{W_3> x\right\}},\ {\bf{1}}_{\left\{W_4> x\right\}}\right)\newline &{}={}&P\left(W_1> x\right)P\left(W_2> x\right){}+{}\mathbb{C}ov\left({\bf{1}}_{\left\{W_3> x\right\}},\ {\bf{1}}_{\left\{W_4> x\right\}}\right)\newline &{}={}&P\left(X_1> x\right){}+{}\mathbb{C}ov\left({\bf{1}}_{\left\{W_3> x\right\}},\ {\bf{1}}_{\left\{W_4\ge x\right\}}\right)\newline \end{eqnarray*} $$

Therefore,

$$ \begin{eqnarray*} \mathbb{C}ov\left({\bf{1}}_{\left\{W_3> x\right\}},\ {\bf{1}}_{\left\{W_4> x\right\}}\right){}\ge0&\iff& P\left(X_2> x\right)\ge P\left(X_1> x\right)\newline &&\newline &\iff& P\left(X_2\le x\right)\le P\left(X_1\le x\right)\,. \end{eqnarray*} $$

$\endgroup$
10
  • $\begingroup$ Thanks for your answer ki3i!. but, I cannot follow the proof from the second line. What does it mean the event $\textbf{1}_{A,B}?$ $\endgroup$
    – Oliver
    Commented Jan 27, 2015 at 0:42
  • $\begingroup$ @Oliver, These are indicator functions en.wikipedia.org/wiki/Indicator_function. That is, ${\bf{1}}_{A}(\omega){}:={}\left\{\begin{array}{cc}1;&\,\,\,if\,\,\,\omega\in A\\0;&\,\,otherwise\end{array}\right.\,.$ $\endgroup$
    – ki3i
    Commented Jan 27, 2015 at 0:53
  • $\begingroup$ @Oliver, A useful property of indicator functions is that $\mathbb{E}\left[{\bf{1}}_{A}\right]{}={}P\left(A\right)\,.$ This is also discussed at the Wikipedia link above. $\endgroup$
    – ki3i
    Commented Jan 27, 2015 at 1:09
  • $\begingroup$ Thanks a lot ki3i!! I am looking at your proof and the wiki link. Thanks once again. $\endgroup$
    – Oliver
    Commented Jan 27, 2015 at 2:10
  • $\begingroup$ @ ki3i, how come the line 3 to line 4? I agree that must be true, but, could not find out quickly; may be some axioms of indicator functions? $\endgroup$
    – Oliver
    Commented Jan 27, 2015 at 2:47

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .