Suppose that $X$ and $Y$ are r.v.s such that $F_X$ (the cdf of $X$) is continuous and $$ \sup_{r\in\mathbb{R}}|F_X(r)-F_Y(r)|\le \epsilon. $$ Is it true that $\mathsf{P}(X\le q_Y(\alpha))\le \alpha+\epsilon$, where $q_Y(\alpha)$ is the $\alpha$-quantile of $Y$?

If $q_Y(\alpha)$ is a continuity point of $F_Y$, then it is true. Also by the uniform bound the jumps of $F_Y$ cannot exceed $2\epsilon$ so that the bound holds when $F_Y$ is discontinuous at $q_Y(\alpha)$ ($\because$ $F_Y$ is cadlag).


You have for all $r\in\Bbb R$ $$F_X(r) - F_Y(r) \le \sup_{r\in\mathbb{R}}|F_X(r)-F_Y(r)|\le \epsilon$$

Hence for all $r\in\Bbb R$ it holds $$F_X(r) \le F_Y(r) + \varepsilon$$ $q_Y(\alpha)$ is the $\alpha$-quantile of Y hence we have for each $n\in\Bbb N$ $$F_Y\left(q_Y(\alpha) - \frac{1}{n}\right) = P\left( Y \le q_Y(\alpha) - \frac{1}{n}\right) \le P\left( Y < q_Y(\alpha) \right) \le \alpha$$

And because $F_X$ is continuous we get: $$\begin{align*}P\left(X \le q_Y(\alpha)\right) &= F_X\left(q_Y(\alpha)\right) \\ &= \lim_{n\to\infty} F_X\left(q_Y(\alpha) - \frac{1}{n}\right) \\&\le \lim_{n\to\infty} F_Y\left(q_Y(\alpha) - \frac{1}{n}\right) + \varepsilon \\ &\le \alpha + \varepsilon\end{align*}$$

  • 1
    $\begingroup$ There is nothing to be known… it's the definition of the cdf. $F_Y$ is cdf of $Y$ iff $F_Y(y) = P(Y \le y)$. Independently of $F_Y$ being continuous or not. $\endgroup$ – Gono Jun 20 '18 at 19:17

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.