Bootstrapping Through Discrete Convolutional Methods

TL;DR

This paper introduces a convolutional approach to bootstrap methods that offers exact solutions and faster computation compared to traditional Monte Carlo resampling, especially for discrete distributions.

Contribution

It presents a novel convolutional technique for bootstrapping that improves accuracy and efficiency over Monte Carlo methods for discrete data.

Findings

Convolutional methods provide exact bootstrap quantities.

Faster computation than Monte Carlo for certain problems.

Comparable or improved accuracy over saddlepoint methods.

Abstract

Bootstrapping was designed to randomly resample data from a fixed sample using Monte Carlo techniques. However, the original sample itself defines a discrete distribution. Convolutional methods are well suited for discrete distributions, and we show the advantages of utilizing these techniques for bootstrapping. The discrete convolutional approach can provide exact numerical solutions for bootstrap quantities, or at least mathematical error bounds. In contrast, Monte Carlo bootstrap methods can only provide confidence intervals which converge slowly. Additionally, for some problems the computation time of the convolutional approach can be dramatically less than that of Monte Carlo resampling. This article provides several examples of bootstrapping using the proposed convolutional technique and compares the results to those of the Monte Carlo bootstrap, and to those of the competing saddlepoint method.