线性回归M估计量的Wild Bootstrap方法研究

Wild Bootstrap Method for M Estimators of Linear Regression

Wild Bootstrap是一种适用于回归方程中存在异方差时的再取样方法。本文通过线性回归Huber估计量的模拟研究,比较了不同的bootstrap方法,并验证了wild bootstrap方法在有限样本下的有效性。通过运用一种简单有限样本统计量对wild bootstrap加以修正,对于存在异方差性且基于固定设计的回归模型而言,wild bootstrap成为首选的重复抽样法。

The wild bootstrap is capable of accounting for heteroscedasticity in a regression model. In this paper,a simulation study on Huber estimator of linear regression is carried out to compare various bootstrap methods and to demonstrate the relevance of our work in finite-sample problems. With a simple finite-sample correction,the wild bootstrap is a preferred resampling method when there exists heteroscedasticity in a regression model with fixed design points.