Consistency of the generalized bootstrap for degenerate \(U\)-statistics (Q1317252): Difference between revisions

Let \(X_ 1,\dots,X_ n\) be i.i.d. random variables defined on a probability space \((\Omega, {\mathcal F}, P)\) and \(h\) be a symmetric kernel of degree 2. Consider the degenerate \(U\)-statistic \[ U_ n (h) = {n \choose 2}^{-1} \cdot \sum_{1 \leq i<j \leq n} h(X_ i,X_ j) \] \(Eh(X_ 1,x) = 0\) for almost all \(x\). Let \(W_ n = (W_{n1}, \dots, W_{nn})\) be a vector of random weights defined on a probability space \((\widetilde \Omega, \widetilde {\mathcal F}, \widetilde P)\) and independent of the data \(X_ 1,\dots,X_ n\). For a degenerate \(U\)-statistic the authors define the corresponding bootstrapped \(U\)-statistic as \[ U_{w_ n} (h) = {\underset i \neq {j} {\sum \sum}} (W_{n_ i} - n^{-1}) (W_{n_ j} - n^{-1}) h(X_ i,X_ j). \] The main result states that given \(X_ n = (X_ 1, \dots, X_ n)\) the distribution function of \(n \cdot U_{w_ n} (h)\) is a uniformly consistent estimator for the unknown distribution of \(n \cdot U_ n(h)\). Some examples are considered.