面板数据模型的惩罚复合分位回归方法  

A Penalized Composite Quantile Regression Method for Panel Data Model

作　　者：朱利荣 胡超竹 罗幼喜[1] Zhu Lirong;Hu Chaozhu;Luo Youxi(School of Science,Hubei University of Technology,Wuhan 430068,China)

机构地区：[1]湖北工业大学理学院,武汉430068

出　　处：《统计与决策》2022年第13期40-45,共6页Statistics & Decision

基　　金：国家社会科学基金资助项目(17BJY210);国家自然科学基金资助项目(11701161)。

摘　　要：针对含个体效应的面板数据模型,文章提出了一种带Adaptive Lasso惩罚的复合分位回归方法来估计回归系数。通过对模型两边左乘一个合适的幂等矩阵有效地消除了个体效应的影响,并使用MM算法迭代求解未知参数,用SIC准则对惩罚参数进行选取。同时,利用蒙特卡洛方法模拟了在不同误差和不同稀疏模型下回归系数的估计和选择情况,并与最小二乘回归、中位回归、复合分位回归估计结果进行对比,最后用实例数据进行验证。结果表明:带Adaptive Lasso惩罚的复合分位回归方法能够对回归系数进行精确估计,且其在稀疏模型上相比稠密模型具有更好的表现。在变量选择问题上,带Adaptive Lasso惩罚的复合分位回归方法能够很好地排除无关解释变量的影响。For panel data models with individual effects, this paper proposes a composite quantile regression method with Adaptive Lasso penalty to estimate the regression coefficients. By multiplying both sides of the model by an appropriate idempotent matrix, the influence of individual effects is effectively eliminated;the MM algorithm is adopted to iteratively solve the unknown parameters, and the SIC criterion is used to select the penalty parameters. At the same time, Monte Carlo method is used to simulate the estimation and selection of regression coefficients under different errors and different sparse models, and the results are compared with those of least square regression, median regression and composite quantile regression. Finally, the example data is used to make verification. The results show that the composite quantile regression method with Adaptive Lasso penalty can estimate the regression coefficients accurately, and has better performance in sparse model than in dense model, and that in variable selection, the composite quantile regression method with Adaptive Lasso penalty can eliminate the influence of irrelevant explanatory variables well.

关 键 词：面板数据 Adaptive Lasso惩罚 复合分位回归 

分 类 号：C81[社会学—统计学]