The bootstrap of the mean with arbitrary bootstrap sample size (Q749074): Difference between revisions

Let \(X,X_ 1,X_ 2,..\). be independent, identically distributed random variables defined on a probability space (\(\Omega\),\({\mathcal A},P)\). For each \(n\in N\) and \(\omega\in \Omega\) let \(\{X^{\omega}_{nj}(\omega ')\), \(j=1,...,m_ n\}\) be independent, identically distributed random variables with law \(P_ n(\omega)=n^{- 1}\sum^{n}_{i=1}\delta_{X_ i}(\omega),\) defined on another probability space \((\Omega ',{\mathcal A}',P')\). Further, let \(\hat L\) be the conditional distribution function given \(\{X_ i\}\). We say that the bootstrap CLT holds in probability (or a.s.) if \[ \hat L[\sum^{m_ n}_{j=1}X^{\omega}_{nj}/a_ n-c_ n(\omega)]\to N(0,\sigma^ 2)\text{ in } probability\quad (or\quad a.s.), \] where d denotes any distance metrizing weak convergence in R. It was already shown that if E \(X^ 2<\infty\), then the bootstrap CLT holds a.s. as long as \(m_ n\to \infty\) [see \textit{D. Freedman} and \textit{P. J Bickel}, Ann. Stat. 9, 1196- 1217 (1981; Zbl 0449.62034)]. The present authors prove the following: (i) If X is in the domain of attraction of the normal law then the bootstrap CLT holds in probability for all \(m_ n\to \infty;\) (ii) If \(m_ n(\log \log n)/n\to 0\) and X is in the domain of attraction of a stable law, then the bootstrap CLT holds a.s., but if F \(X^ 2=\infty\) and \(\inf m_ n(\log \log n)/n>0,\) then it does not hold a.s.