基于M-H抽样算法的贝叶斯Probit分位回归模型研究  

Bayesian Analysis of Probit Quantile Regression Models Based on Metropolis-hastings Algorithm

作　　者：朱慧明[1] 李荣[1] 曾昭法[2] 虞克明[3] 

机构地区：[1]湖南大学工商管理学院,湖南长沙410082 [2]湖南大学金融与统计学院,湖南长沙410079 [3]Brunel大学数学系

出　　处：《湖南大学学报（自然科学版）》2013年第2期98-102,共5页Journal of Hunan University:Natural Sciences

基　　金：国家自然科学基金资助项目(71031004;71171075);国家自然科学基金创新研究群体项目(71221001);教育部创新团队项目(IRT0916);教育部博士点基金项目(20110161110025);湖南省自然科学基金资助项目(11JJ3090;09JJ7002)

摘　　要：针对Probit分位回归在参数随机化条件下的建模问题,提出基于Metropolis-Hastings算法的贝叶斯Probit分位回归模型.通过分析Probit分位回归模型结构,选择模型的先验分布,运用M-H算法进行参数估计.利用Monte Carlo仿真技术,得到不同分位点模型参数后验分布,同时用贝叶斯probit分位回归与分位回归方法和光滑分位回归方法对模型参数估计进行比较分析.研究结果表明:贝叶斯Probit分位回归模型可以更全面描述离散选择变量的影响,能够得到更加准确有效的参数估计.To address the problem of modelling under the condition of random parameters using Probit quantile regression, this paper proposed Bayesian Probit quantile regression model based on the Metropolis-Hastings algorithrn. According to the Probit quantile structure, the M-H algorithm was ulitized to simulate the posterior marginal distribution by choosing the prior distribution of the parameters. The model posterior distribution in the different quantile points was obtained by using Monte Carlo simulation, and at the same time, Probit quantile regression, smoothed Probit quantiel regression and Bayesian probit quantile regression method were used to estimate the parame- ters of the model and to compare the differences from the estimation of parameters, respectively. The research shows that Bayesina Probit quantile regression comprehensively describes the influencing factors of discrete variables, and the estimation of parameters is more accurate and efficient.

关 键 词：仿真 PROBIT模型 贝叶斯方法 分位回归 离散选择 

分 类 号：O212.8[理学—概率论与数理统计]