Asymptotic properties of the bootstrap for heavy-tailed distributions (Q751110)

Let \((X_ n)\) be a sequence of i.i.d. random variables with common distribution function F and let \({\mathcal X}=\{X_ 1,X_ 2,...,X_ n\}\) be the sample of the first n \(X_ i's\). Let \({\mathcal X}^*=\{X^*_ 1,X^*_ 2,...,X^*_ n\}\) be a resample drawn randomly, with replacement, from \({\mathcal X}\). Define \(\bar X=n^{- 1}\sum^{n}_{i=1}X_ i\) and the bootstrapped mean \(\bar X^*=n^{- 1}\sum^{n}_{i=1}X^*_ i.\) Then it is shown that (Theorem 2.1) the distribution of \(\bar X^*\), suitably normalized, converges in probability to some nondegenerate distribution if and only if either F belongs to the domain of attraction of a normal law or F has a slowly varying tail at \(\infty\) (-\(\infty)\). The limit law is shown to be normal in the first case and displaced Poisson in the second case. In particular when F belongs to the domain of normal attraction of a normal law (E \(X^ 2_ 1<\infty)\) the above convergence is shown to be almost sure. As convergence in probability fails when F belongs to the domain of attraction of a nonnormal stable law, it is shown that the distribution of \(\bar X^*\), suitably normalized, converges weakly to some nondegenerate distribution. Results of \textit{K. B. Athreya} [see Ann. Stat. 15, 724-731 (1987; Zbl 0628.62042), and Mathematical statistics theory and applications, Proc. World Congr. Bernoulli Soc., Tashkent/USSR 1986, Vol. 2, 95-98 (1987; Zbl 0669.62023)] have been established as a special case.