Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Suppose the "mean residual lifetime," $\mathbb{E}[X-x|X≥x]$ is approximately constant for large $x$. Then, I believe that the conditional tail distribution is approximately exponential, in the sense of being stochastically dominated by an exponential and dominating a similar exponential. Formally:

Conjecture Given any random variable $X$ with support on $[0,∞)$. If, for some $\lambda \in(0,∞)$, $$\lim_{x→∞}\mathbb{E}[X-x|X≥x]= \lambda ,$$

then, for all $ε>0$ and for all $\Delta>0$, there is some $c$ such that $x≥c$ implies $$e^{-\frac{t}{λ-ε}}\leq \mathbb{P}[X≥x+t|X≥x] \leq e^{-\frac{t}{λ+ε}} \qquad ∀t≥\Delta.$$

I posted this question on StackExchange. Robert Israel provided a counterexample to an earlier conjecture, which was wrong.

Update The approximation result is stronger than weak convergence. Let $Y$ be distributed exponentially with parameter $\lambda$. The conclusion of the conjecture implies that

$$\lim_{x→∞}\mathbb{E}[f(X-x)|X≥x]=\mathbb{E}[f(Y)]$$

for all nondecreasing functions for which $\mathbb{E}[f(Y)]$ exists. In particular, $f$ is allowed to be unbounded. The first great response by Ori Gurel-Gurevich implies a slightly weaker approximation result.

share|improve this question

2 Answers 2

up vote 2 down vote accepted

$\newcommand{\eps}{\varepsilon} \newcommand{\E}{\mathbb{E}} \renewcommand{\P}{\mathbb{P}}$ Fix $\eps>0$ and let $a=\E[X]$ and $b=\E[X-\eps \mid X>\eps]$. Let $p=\P(X > \eps)$. Then $$ (1-p)(b+\eps)\le a \le (1-p)(b+\eps) + p \eps \ .$$

Solving, we get $$ 1-\frac{a+\eps}{b+\eps}+\frac{\eps}{b+\eps} < p < 1-\frac{a}{b}+\frac{\eps}{b} $$

In other words, $$ \frac{\eps}{\lambda+\eps} -\delta < p < \frac{\eps}{\lambda} +\delta $$

Where $\delta=\delta(a,b)$ goes to 0 as $a$ and $b$ approach $\lambda$.

Compounding this we get that when $a < \E[X-s \mid X > s] < b$ for all $s>0$ and we take $a$ and $b$ to $\lambda$, we have for any $t>0$

$$(1-\frac{\eps}{\lambda+\eps})^{\frac{t}{\eps}} - \delta_1 < \P(X > t) < (1-\frac{\eps}{\lambda})^{t/\eps} + \delta_1$$

where $\delta_1 \to 0$ when $a,b \to \lambda$.

Taking now $\eps$ small enough yields the desired result.

share|improve this answer
    
The idea to use compounding seems right to me. I don't see how the result is implied, however. Taken literally, the displayed conclusion implies that for all t, the absolute value of P(X>t) is bounded from above and below by the corresponding values of the exponential distribution +/- d. The approximation of the exponential distribution is weaker than the approximation from the conclusion of the conjecture. I might not understand the argument fully. Ps: I might really simply overlook how the structure of your argument extends. I very much appreciate your work! Thanks a lot! –  Stephan Apr 26 '12 at 15:41

Here is a work which gives general conditions under which your conjecture is true.

"Limiting Properties of the Mean Residual Lifetime Function" by Isaac Meilijson in Ann. Math. Statist. Volume 43, Number 1 (1972), 354-357.

http://projecteuclid.org/euclid.aoms/1177692731

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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