二元选择分位数回归模型的贝叶斯估计方法及模拟研究  

Bayesian Estimation and Simulation of Binary Selective Quantile Regression Model

作　　者：邸俊鹏[1] 张晓峒[2] Di Junpeng;Zhang Xiaotong(Research Center of Econometrics,Shanghai Academy of Social Sciences,Shanghai 200020,China;Institute of Quantitative Economics,Nankai University,Tianjin 300071,China)

机构地区：[1]上海社会科学院数量经济研究中心,上海200020 [2]南开大学数量经济研究所,天津300071

出　　处：《统计与决策》2019年第5期11-16,共6页Statistics & Decision

基　　金：国家自然科学基金青年项目(71601120);上海社会科学院创新工程"数量经济学"学科创新团队资助项目

摘　　要：文章针对二元选择分位数回归模型的贝叶斯估计方法进行探索性研究。通过模拟实验比较分析了不同先验设定和不同抽样算法下贝叶斯二元选择分位数回归估计量的统计性质,以及贝叶斯方法与频率学派方法对二元选择分位数模型进行估计的优劣。结果表明,对二元选择分位数回归模型进行贝叶斯估计具有一定的优势,尤其是小样本下,估计效果更佳;而且采用Gibbs抽样得到的估计量精度更高,统计推断更准确,尤其是在高分位数下Gibbs抽样的优势更明显。This paper makes an exploratory study on Bayesian estimation of binary selection quantile regression model.By simulation experiments the paper comparatively analyzes the statistical properties of Bayesian binary selection quantile regression estimator under different priori settings and different sampling algorithms,as well as the merits and demerits of Bayes method and frequency school method to estimate the binary choice quantile model.The results show that the Bayesian estimation of binary selection quantile regression model has certain advantages,especially in the case of small samples,and the estimation effect is better.Moreover,the estimator precision obtained by Gibbs sampling is higher and the statistical inference more accurate;especially in the case of high score,Gibbs sampling has more obvious advantages.

关 键 词：分位数回归 贝叶斯估计 二元离散选择模型 

分 类 号：F064.1[经济管理—政治经济学] O212.8[理学—概率论与数理统计]