面板数据分位回归中适应性LASSO调节参数的选择

发布时间：2018-03-31 02:28

本文选题：调节参数选择　切入点：分位回归　出处：《数理统计与管理》2017年03期

【摘要】：在带有罚函数的变量选择中,调节参数的选择是一个关键性问题,但遗憾的是,在大多数文献中,调节参数选择的方法较为模糊,多凭经验,缺乏系统的理论方法。本文基于含随机效应的面板数据模型,提出分位回归中适应性LASSO调节参数的选择标准惩罚交叉验证准则(PCV),并讨论比较了该准则与其他选择调节参数的准则的效果。通过对不同分位点进行模拟,我们发现当残差ε来自尖峰分布和厚尾分布时,该准则能更好地估计模型参数,尤其对于高分位点和低分位点而言.选取其他分位点时,PCV的效果虽稍逊色于Schwarz信息准则,但明显优于Akaike信息准则和交叉验证准则。且在选择变量的准确性方面,该准则比Schwarz信息准则、Akaike信息准则等更加有效。文章最后对我国各地区多个宏观经济指标的面板数据进行建模分析,展示了惩罚交叉验证准则的性能,得到了在不同分位点处宏观经济指标之间的回归关系。
[Abstract]:In the variable selection with penalty function, the choice of adjusting parameters is a key problem. Unfortunately, in most literatures, the method of adjusting parameter selection is rather fuzzy, mostly based on experience and lacking of systematic theory.Based on the panel data model with random effects, this paper presents a standard penalty cross validation criterion for the selection of adaptive LASSO adjustment parameters in quantile regression, and discusses and compares the effectiveness of this criterion with other criteria for selecting adjustment parameters.By simulating the different loci, we find that when the residual 蔚 comes from the peak distribution and the thick tail distribution, the criterion can better estimate the model parameters, especially for the high and low fraction sites.The effect of Schwarz information criterion is slightly inferior to that of Akaike information criterion and cross validation criterion when other loci are selected.Moreover, this criterion is more effective than Schwarz information criterion and Akaike information criterion in the accuracy of selecting variables.Finally, the panel data of several macroeconomic indicators in various regions of China are modeled and analyzed, and the performance of the penalty cross-validation criterion is demonstrated, and the regression relationship between macroeconomic indicators at different sub-sites is obtained.
【作者单位】：中国人民大学应用统计科学研究中心;中国人民大学统计学院;兰州商学院统计系;
【基金】：教育部哲学社会科学研究重大课题攻关项目(15JZD015),教育部高等学校博士学科点专项科研基金(20130004110007),教育部人文社会科学重点研究基地重大项目(15JJD910001) 国家自然科学基金(11271368) 北京市社会科学基金重大项目(15ZDA17) 国家社会科学基金重点项目(13AZD064) 中国人民大学科学研究基金(中央高校基本科研业务费专项资金资助)项目成果(15XNL008)
【分类号】：O212.1

【相似文献】