摘要
在生产实践和计量经济领域中,控制产品质量的方差就能保证产品的合格品数相对稳定,所以当前学者对联合均值与方差模型的研究倍感兴趣.基于解释变量经常是具有相关关系的实际情况,提出了一种由SCAD惩罚和岭回归混合在一起的组合惩罚,该惩罚充分利用了岭回归能克服解释变量相关性过高对估计效果的影响,同时也证明了这样的惩罚具有相合性和Oracle性质.使用该组合惩罚对联合均值与方差模型进行了变量选择.最后的随机模拟结果表明该模型和方法是有效的.
In the production and econometric area , controlling the variance of the quality of the product can guarantee the stable quality of products .So scholars are very interested in joint mean and variance models nowadays .In general ,it is uncommon for explanatory variables to be uncorrelated .A combined penalization ,w hich is mixed by the smoothly clipped absolute deviation （SCAD ） penalty and ridge ,is proposed .It can outperform the SCAD penalty technique w hen the correlation among predictors is high . At the same time , the consistency and the Oracle properties of the combined penalization are proved .Then ,the combined penalization is used to select variables in joint mean and variance models .The results of stochastic simulation show that this model and method are effective .
出处
《大连理工大学学报》
EI
CAS
CSCD
北大核心
2014年第1期147-151,共5页
Journal of Dalian University of Technology
基金
国家自然科学基金资助项目(61175041
11371077)
国家自然科学基金青年基金资助项目(11101062)
中央高校基本科研业务费专项资金资助项目(DUT12LK29)
大连民族学院自主科研基金资助项目(DC120101115)
关键词
组合惩罚
联合均值与方差模型
变量选择
惩罚极大似然估计
combined penalization
joint mean and Variance model
variable selection
penalizedmaximum likelihood
