经典非参数回归模型和贝叶斯非参数分位数回归模型的比较  

Comparison Between Classical Nonparametric Regression Model and Bayesian Nonparametric Quantile Regression Model

作　　者：孔航 Kong Hang(Institute of Marxism,Nanjing University of Science ＆ Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学马克思主义学院

出　　处：《统计与决策》2018年第17期34-39,共6页Statistics & Decision

基　　金：国家社会科学基金青年项目(17CJL035)

摘　　要：文章基于贝叶斯法对非参数函数进行分位数处理,研究函数在每个分位点的基本特征,构建了一种新的基于贝叶斯法的非参数分位数回归模型,并与传统非参数回归模型进行算例比较。新模型具有以下优点：第一,分位点差异性。该模型有别于传统的非参数模型,可以对每个分位点的差异进行分析。第二,高效性。基于贝叶斯的基本方法对非参数函数进行分位数拓展研究,可以大大提高运算效率。第三,可靠性。Gibbs抽样校准结果较为理想、蒙特卡洛模拟的精度较高。Based on the Bayesian method,this paper deals with the quantile processing of nonparametric functions,analyzes the basic features of the function at each quantile, constructs a new nonparametric quantile regression model based on Bayesian method,and compares the estimation results with the traditional nonparametric regression model. The new model has the following advantages： Firstly, quantile difference. This model, different from the traditional non-parametric model, can analyze the differences in each quantile. Second, high efficiency. Based on the Bayesian basic method, the quantile expansion of non-parametric functions can be studied, with efficiency greatly improved. The third is its reliability. The Gibbs sample calibration results are ideal and Monte Carlo simulations have higher accuracy.

关 键 词：非参数回归模型 分位数 Gibbs抽样算法 联合密度函数 

分 类 号：O21[理学—概率论与数理统计]