5
$\begingroup$

I came across this question in a review of an old exam I took. I didn't get the answer correctly then, and I'm struggling to figure the answer out now. Can anyone help me reason through this?

Prove or Disprove that if $F_X(z) > F_Y (z)$ for all $z\in \mathbb{R}$ then $P(X < Y ) > 0$. We may not assume independence.

Here is what I attempted:

I figured I might be able to approach this by proving this through contradiction. I started by assuming $P(X<Y)=0$. Then,

\begin{eqnarray*} F_{X}(z)=P(X\le z) & = & P(X\le z,X<Y)+P(X\le z,X\ge Y)\\ & = & 0+P(X\le z,X\ge Y) \end{eqnarray*}

Can anyone help from here?

$\endgroup$
  • 2
    $\begingroup$ This isn't true. Does the question ask you to assume $X$ and $Y$ are independent? $\endgroup$ – whuber Apr 15 at 20:12
  • 1
    $\begingroup$ @whuber, sorry, I wasn't clear about this earlier. The question asks to prove OR disprove that this is true. We may NOT assume independence. $\endgroup$ – StatCurious Apr 15 at 21:08
  • 2
    $\begingroup$ I found counterexamples by considering discrete variables supported on $\{0,1,2\}$ and constructing a table with given marginals (nearly satisfying your requirements, for of course the two distribution functions must agree on $(-\infty,0)$ and $(,\infty)$) and distributing the probability within the tables to make $\Pr(X\lt Y)$ very small. This provides the insight; such near-counterexamples are readily modified into genuine counterexamples. $\endgroup$ – whuber Apr 15 at 21:53
7
$\begingroup$

Firstly, it is worth noting that the antecedent condition in your conjecture is a slightly stronger version of the condition for strict first-order stochastic dominance (FSD) $X \ll Y$, so it implies this stochastic dominance relationship. This condition is much stronger than what you actually need to get the result in the conjecture, so I will give you a proof for a stronger result (same implication but with a weaker antecedent condition). Your chosen method of proof is a good one, and you are almost there - just one more step to go!


Theorem: If $F_X(z) > F_Y(z)$ for some $z \in \mathbb{R}$ then $\mathbb{P}(X<Y) > 0$.


Proof: We will proceed using a proof-by-contradiction. Contrary to the result in the theorem, suppose that $\mathbb{P}(X<Y)=0$. Then for all $z \in \mathbb{R}$ you have: $$\begin{equation} \begin{aligned} F_X(z) = \mathbb{P}(X \leqslant z) &= \mathbb{P}(X \leqslant z, X < Y) + \mathbb{P}(Y \leqslant X \leqslant z) \\[6pt] &= 0 + \mathbb{P}(Y \leqslant X \leqslant z) \\[6pt] &= \mathbb{P}(Y \leqslant X \leqslant z) \\[6pt] &\leqslant \mathbb{P}(Y \leqslant z) = F_Y(z), \\[6pt] \end{aligned} \end{equation}$$ which contradicts the antecedent condition for the theorem. This establishes the theorem by contradiction. $\blacksquare$

$\endgroup$
  • $\begingroup$ Ah, yes of course! I don't know how I didn't see those last few steps earlier! Thank you, @Ben. This is very helpful. I appreciate the FSD info too! $\endgroup$ – StatCurious Apr 16 at 0:55
3
$\begingroup$

Under the assumption that $X$ and $Y$ are independent and continuous, \begin{align*}\Bbb P(X<Y)&=\Bbb E^Y[\Bbb I_{X<Y}\mid Y]\\ &=\Bbb E^Y[F_X(Y)]\\&>\Bbb E^Y[F_Y(Y)]\\ &=\int_{\Bbb R} F_Y(y) \, \text{d}F_Y(y) \\&= \frac{1}{2} \int_{\Bbb R} \, \text{d}F_Y^2(y)\\&=\frac{1}{2}F_Y^2(\infty)-\frac{1}{2}F_Y^2(-\infty)\\&=1/2\end{align*} Further, $$\int_{\Bbb R} F_Y(y) \,\text{d}F_Y(y)=\int_{\Bbb R} \Bbb P(Y'<y) \,\text{d}F_Y(y)$$ when $Y'\sim F_Y(\cdot)$, or $$\int_{\Bbb R} F_Y(y) \,\text{d}F_Y(y)=\Bbb P(Y'<Y)$$ when $Y,Y'\stackrel{\text{iid}}{\sim} F_Y(\cdot)$, implying $$\int_{\Bbb R} F_Y(y) \,\text{d}F_Y(y)=1/2$$

$\endgroup$
  • 3
    $\begingroup$ You could have stopped at the first line by noting that $F_Y(y) \ge 0$ for all $y\in \mathbb R$ implies $\mathbb{E}^Y[F_Y(Y)] \gt 0.$ I don't fully believe the rest of the calculation because I can find discrete variables $Y$ where this expectation does not equal $1/2.$ A uniform distribution on $\{0,1,2\}$ will work as a counterexample. $\endgroup$ – whuber Apr 15 at 20:42
  • $\begingroup$ Sorry for any confusion. I cannot assume independence. I've updated the question to make it clear. $\endgroup$ – StatCurious Apr 15 at 21:12
  • 1
    $\begingroup$ @whuber: right and right. I assumed continuity in addition to independence. $\endgroup$ – Xi'an Apr 16 at 4:28

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.