样本选择模型截距项的半参数估计及应用———户籍工资差异分解研究

样本选择模型是解决样本选择问题的主要工具,广泛应用于工资差异分解、平均处理效应测算等实证研究。截距项的估计是样本选择模型半参数估计中相对独立且重要的一部分,现有的以无穷处识别为代表的半参数估计方法存在窗宽参数难以选取的问题。为此,本文把无穷处识别等价转化为边界处识别,并基于新的识别关系给出样本选择模型截距项的核估计方法。这种新方法的好处在于将样本选择模型截距项的估计纳入核估计框架中,从而可以采用经验法则解决现有方法的窗宽选取难题。数值模拟结果表明,本文所提出的估计方法在不同设定下均有良好的有限样本表现。把这种新的半参数估计方法应用于户籍工资差异分解后发现,我国劳动力市场目前不存在明显的户籍差别待遇。

Semiparametric Estimation of the Intercept and Application of the Sample Selection Model—Research on the Wage Difference Decomposition

Heckman sample selection model is a standard tool in sample selection correction, and the estimation of its intercept is important in applications such as wage difference decomposition and the estimation of average treatment effect. The identification at infinity is an important semiparametric estimation method but it is difficult to select appropriate bandwidths. This study rephrases the identification-at-infinity problem into identification-at-boundary, and accordingly proposes a kernel estimation method to semiparametrically estimate the intercept of sample selection model. A virtue of the kernel method is that the bandwidths can be chosen by the rule of thumb. A Monte Carlo simulation shows that the rule-of-thumb bandwidths are optimal or nearly optimal in various designs. Lastly, the new semiparametric estimation method is applied to wage difference decomposition by the registered population system, which shows that Chinese labor market has no significant differential treatment due to the registered population system at present.