I have come across a needed condition on a continuous probability distribution defined over $[0, \infty]$ and wonder whether it has a name. For a distribution with CDF $F$ and pdf $f$ I need that the quantity: $$\phi(x) \equiv \frac{f(x)}{F(x)+xf(x)}$$ be monotonically non-increasing. Putting the condition in another form (by taking derivatives), the requirement is that for all $x \in [0,\infty]$ such that $f'(x) > 0$: $$f(x) \geq \sqrt{\frac{F(x) f'(x)}{2}}$$

This seems to be a commonly satisfied property, so does it have a name? It is related to, but distinct from a monotone hazard rate condition.

  • 1
    $\begingroup$ Have you looked at monotonicity of the mean residual life? (Maybe that's how you derived the conditions in the first place?) $\endgroup$ – guest Dec 19 '11 at 17:52
  • 2
    $\begingroup$ Can you tell us a bit about how the quantity ϕ comes about? $\endgroup$ – r.e.s. Dec 19 '11 at 18:38

This is almost the condition for the cumulative distribution function to be log-concave , which is a very useful property with many applications. But almost.

A function $F(x)$ is log-concave if

$$\frac {\partial^2 \ln F(x)}{\partial x^2} \le 0 \Rightarrow F''(x)F(x) - \left[F'(x)\right]^2 \le 0$$

Write $\phi(x)$ in terms of $F(x)$

$$\phi(x) \equiv \frac{F'(x)}{F(x)+xF'(x)}$$

and we want

$$\frac {\partial \phi(x)}{\partial x} \le 0 \Rightarrow F''(x)\Big(F(x)+xF'(x)\Big)-F'(x)\Big(F'(x)+F'(x) +xF''(x)\Big) \le 0$$

$$\Rightarrow F''(x)F(x)-2\left[F'(x)\right]^2 \le 0 $$

...which is not enough for log-concavity, due to the existence of the factor $2$.

Assume that the condition is satisfied. If we divide by $[F(x)]^2$ and rearrange we obtain

$$\frac {\partial \phi(x)}{\partial x} \le 0 \Rightarrow \frac {\partial^2 \ln F(x)}{\partial x^2} \le \left( \frac{F'(x)}{F(x)}\right)^2 = \left(\frac {\partial \ln F(x)}{\partial x}\right)^2$$

| cite | improve this answer | |

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.