稳健双自适应惩罚权重expectile方法及其在GDP数据中的应用  

Robust Double-Adaptive Regularized Weight Expectile Method and Its Application in GDP Data

作　　者：严笑 文诗涵 邹航 YAN Xiao;WEN Shi-han;ZOU Hang(School of International Studies,Jinan University,Guangzhou 510632,China;School of Economics,Jinan University,Guangzhou 510632,China)

机构地区：[1]暨南大学国际关系学院,广东广州510632 [2]暨南大学经济学院,广东广州510632

出　　处：《数理统计与管理》2024年第6期962-972,共11页Journal of Applied Statistics and Management

基　　金：国家自然科学基金项目(12171203);广东省自然科学基金项目(2022A1515010045);中央高校基本科研业务费专项资金资助(23JNQMX21)。

摘　　要：为了解决杠杆点存在时,惩罚expectile回归和惩罚分位数回归失效问题,基于expectile回归和稳健双自适应惩罚权重回归估计方法,本文提出了一种稳健双自适应惩罚权重expectile回归估计方法。该方法可以在自变量和因变量都含有异常值时,实现稳健变量选择和异方差检测。对于提出的模型,本文首先利用MM算法构建替代惩罚函数的优控函数,随后用迭代加权最小二乘算法估计参数,惩罚参数通过最小化BIC准则获得。模拟和实证表明,当数据中存在杠杆点时,所提方法在变量选择和异方差检测效果上优于惩罚最小二乘方法和惩罚分位数回归方法。In order to solve the problem of penalty expectile regression failure when leverage points exist,based on expectile regression and robust double adaptive penalty weight regression estimation method,this paper proposes a robust double adaptive penalty weight expectile regression estimation method.This method can realize robust variable selection and heteroscedasticity detection when both response variables and covariates contain outliers.For the proposed model,this paper first uses MM algorithm to construct the optimal control function instead of the penalty function,and then uses the iterative weighted least squares estimation algorithm to estimate the parameters.The penalty parameters are obtained by minimizing the BIC criterion.Simulation and empirical results show that the proposed method outperforms the penalized least squares method and the penalized quantile regression method in terms of variable selection and heteroscedasticity detection when there are leverage points in the data.

关 键 词：稳健双自适应惩罚权重expectile回归 变量选择 异方差 稳健性 

分 类 号：O212[理学—概率论与数理统计]