?????????????LASSO???????????????

删失回归模型是一种很重要的模型,它在计量经济学中有着广泛的应用. 然而,它的变量选择问题在现今的参考文献中研究的比较少.本文提出了一个LASSO型变量选择和估计方法,称之为多样化惩罚$L_1$限制方法, 简称为DPLC. 另外,我们给出了非0回归系数估计的大样本渐近性质. 最后,大量的模拟研究表明了DPLC方法和一般的最优子集选择方法在变量选择和估计方面有着相同的能力.

A LASSO-Type Approach to Variable Selection andEstimation for Censored Regression Model

Censored regression (``Tobit') modelis one of important regression models and has been widely used ineconometrics. However, studies for variable selection problem incensored regression model are rare at the present references. Inthis paper, for censored regression model we propose a LASSO-typeapproach, diverse penalty $L_1$ constraint method (DPLC), to selectvariables and estimate the corresponding coefficients. Furthermore,we obtain the asymptotic properties of nonzero elements' estimationof regression coefficient. Finally, extensive simulation studiesshow that DPLC method almost possesses the same performance ofselecting variables and estimation as generally best subsetselection method (GBSS).