2
$\begingroup$

Suppose there is a random variable, $X$, with finite variance, and c.d.f. $F(x)$. Does this imply that the upper Matuszewska index of $\bar F(x)$ exists and is strictly smaller than $-2$?

The upper Matuszewska index of $f(x)$ is defined as the infimum of $\alpha$ such that there exists a $C$ such that for each $\Lambda > 1$,

$$ f(\lambda x)/f(x) \le C(1 + o(1))\lambda^\alpha \,\,\,(x \rightarrow \infty)\text{ uniformly in }\lambda \in [1, \Lambda]$$

I ask this because I was reading a paper where a result was proven for distributions with upper Matuszewska index less than $-2$, but in the conclusion the authors stated that the result held for finite variance distributions, seemingly without justification.

$\endgroup$

1 Answer 1

1
$\begingroup$

The answer is no: $\bar F$ does not have to have any negative Matuszewska index (any tail function $\bar F$ trivially has any nonnegative Matuszewska index).

Indeed, take any negative real $\alpha$. Suppose that $P(X=n^n)=c/n^{3n}$ for natural $n$, where $c:=1/\sum_1^\infty1/n^{3n}$. Then $EX^2<\infty$. However, $$\bar F(n^n/2)\sim c/n^{3n}\sim \bar F(n^n/(2\lambda))$$ for any real $\lambda>1$ as $n\to\infty$. Taking here any large enough $\lambda>1$ such that $C\lambda^\alpha<1$, we see that $\bar F$ does not have the upper Matuszewska index $\alpha$, for any negative real $\alpha$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.