异方差下条件处理效应的非参数估计及应用

处理效应模型作为分析政策效应的量化工具,在社会与经济的各个领域有着广泛的应用.现有文献为了得到处理效应的一致估计量,通常需要加一些较强的限制条件(如条件均值独立)或采用工具变量法,在较弱的与实际更吻合的条件下给出处理效应的一致估计量并不多见.本文在误差项对称的假定下,讨论了条件处理效应模型的非参数识别和估计,并进一步估计了平均处理效应.我们通过放松模型中函数形式的假定,同时考虑了较为普遍的广义异方差形式,大大减少了模型误设的可能性,拓展了现有模型的适用性.本文对估计量的大样本性质进行了分析,表明了估计量的一致性和渐近正态性,蒙特卡罗模拟显示了估计量良好的有限样本性质.最后,本文将估计量应用于研究大学教育回报及其性别差异,进一步解释了估计量的实用价值.

Nonparametric estimation and application of conditional treatment effect under heteroscedasticity

Treatment effect is widely used in microeconomics, especially in the literature of program evaluation. In order to estimate the treatment effect consistently, there usually imposed strong restrictions (for example, conditional mean independence) or adopt instrumental variable method. There are few researchers on treatment effect model under less restrictive assumptions. In this paper, we consider nonparametric identification and estimation of conditional treatment effect model under joint symmetry restriction, and further estimate the average treatment effect. Our model relaxes restriction of the functional form, and also considers a more general form of heteroscedasticity, which greatly reduces the possibility of model misplacement and expands the applicability of the existing model. The proposed estimator is shown to be consistent and asymptotically normally distributed, and it also performs well in the simulation. Finally, we apply our approach to estimate the return of college education and gender difference to explain the practical value.