$F(x)$ and $G(y)$ are distribution functions. Define the $\tau$th quantile for cdf $F(x)$, $G(y)$ as $$\xi_\tau\equiv F^{-1}(\tau)=\inf\{x:F(x)\ge \tau\}$$ and $$\eta_\tau\equiv G^{-1}(\tau)=\inf\{y:G(y)\ge \tau\}.$$ Does the linear combination of the quantile, say, $\alpha F^{-1}(\tau)+\beta G^{-1}(\tau)$ still a quantile of some cdf? That is, does there exist a cdf $H(z)$ s.t. $$H^{-1}(\tau)=\alpha F^{-1}(\tau)+\beta G^{-1}(\tau)?$$ Do we need some constraints about $\alpha$ and $\beta$, say, $\alpha,\beta>0$ and $\alpha+\beta=1$?

  • $\begingroup$ Those conditions are the sufficient (and morally the correct conditions). $\endgroup$ – Anthony Quas Jan 1 at 21:29
  • $\begingroup$ @AnthonyQuas I take your comment to mean that in "nontypical" cases they may not be necessary. Can you give an example? $\endgroup$ – kodlu Jan 1 at 21:55
  • 1
    $\begingroup$ No, we do not need $\alpha+\beta=1$. Eg: let $x$ be distributed uniformly on [0,1], let $y=2x^2$, then $\alpha=3,\ \beta =5$ gives the quantile function for $3x + 10x^2$. $\endgroup$ – Matt F. Jan 2 at 3:54
  • 2
    $\begingroup$ Indeed, doesn't any right-continuous nondecreasing function $g$ with domain $[0,1]$ define a quantile function? (Just check that $g(U)$, for $U$ a standard uniform random variable defines a random variable with cdf $g^{-1}$, the (right-) generalized inverse function of $g$)? $\endgroup$ – Stephan Sturm Jan 2 at 6:48
  • 1
    $\begingroup$ @StephanSturm, almost, but the domain should be $(0,1)$ to allow for unbounded distributions. $\endgroup$ – Matt F. Jan 2 at 13:16

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.