贝叶斯复合分位回归的Gibbs抽样算法

摘要

大多数基于传统均值回归的建模方法都对非正态误差表现出不稳健的估计结果.和传统均值回归相比,复合分位回归(CQR)可以产生稳健的估计.基于一个复合反对称Laplace分布(CALD),我们建立了加权复合分位回归(WCQR)的贝叶斯分层模型.Gibbs抽样算法被发展用于WCQR的后验推断.最后,我们提供了一些模拟研究和一个实际数据分析来验证所提方法.%Most regression modeling is based on traditional mean regression which results in non-robust estimation results for non-normal errors.Compared to conventional mean regression,composite quantile regression (CQR) may produce more robust parameters estimation.Based on a composite asymmetric Laplace distribution (CALD),we build a Bayesian hierarchical model for the weighted CQR (WCQR).The Gibbs sampler algorithm of Bayesian WCQR is developed to implement posterior inference.Finally,the proposed method are illustrated by some simulation studies and a real data analysis.