A multivariate generalization of Hall's theorem for Edgeworth expansions of bootstrap distributions

TL;DR

This paper extends Hall's 1992 theorem to multivariate cases, providing a rigorous justification for Edgeworth expansions of bootstrap distributions in higher dimensions, which was previously unaddressed.

Contribution

It introduces a multivariate version of Hall's theorem and outlines the necessary modifications to the original proof for higher-dimensional applications.

Findings

Provides a multivariate generalization of Hall's theorem

Sketches proof modifications for multivariate Edgeworth expansions

Fills a gap in bootstrap distribution theory for multivariate statistics

Abstract

Theorem 5.1 in the monograph by Hall (1992) provides rigorous in-probability justification of Edgeworth expansions of bootstrap distributions. Proving this result was rather challenging because bootstrap distributions do not satisfy the classical Cram\'er condition and therefore classical methods for justifying Edgeworth expansions, e.g. Bhattacharya and Rao (1976) and Bhattacharya and Ghosh (1978), are not available. Hall's (1992) theorem is for a univariate statistic which can be expressed as a smooth function of means, though the underlying population can be multivariate. However, there are a number of applications where a multivariate version of Hall's theorem is needed, and generalizing the proof from the univariate case to the multivariate case is not immediate. Our primary purpose in this article is to fill this gap by stating a multivariate version of the theorem and sketching the modifications to the proof of Hall's (1992) Theorem 5.1 that are needed.