报告题目【Inferences for Partially Conditional Quantile Treatment Effect Models】
时间：2019年9月29日 (星期日) 上午11:00
地点：福建师范大学行政楼附二楼宏达厅
主讲：美国堪萨斯大学教授，蔡宗武
主办：数学与信息学院
参加对象：相关教师和研究生
报告人简介：美国堪萨斯大学经济系经济学教授和计量经济学Charles Oswald 讲席教授。中央“千人计划”特聘专家，教育部“长江学者”讲座教授，厦门大学王亚南经济研究院特聘教授并兼任现代统计研究中心主任，“闽江学者” 讲座教授。蔡宗武教授曾为“中国留美经济学会”会长（2018.9-2019.8），是Journal of business and economic statistics、Econometric reviews和Big data and cloud innovation等国际一流学术期刊副主编，美国统计协会Fellow (资深会员)、国际数理统计协会会员、国际计量经济学会会员和国际泛华统计协会会员。蔡宗武教授是中国自然科学基金管理学部和数学学部、中央组织部“千人计划”和教育部“长江学者”评审专家和评审团成员。蔡宗武教授在国际计量经济学、统计学以及数据科学领域有很高的影响力，在国际顶尖级的经济学与统计学以及金融学等期刊上发表了论文100 多篇。主要研究领域为理论和应用计量经济学、宏观计量经济学、微观计量经济学、经济分析和政策评估、金融计量学、金融大数据、风险管理、非线性和非平稳时间序列建模和检验、非参数函数估计和检验，以及大数据分析与建模等多个领域。
报告摘要：Motivated by studying how the first-time mother's smoking during pregnancy has an effect on the baby's birth weight as a function of the mother's age across different baby's birth weights (particularly for low baby's birth weights), due to the asymmetry of the distribution of birth weight, this paper proposes a new model called the partially conditional quantile treatment effect (PCQTE) model, designed to capture the heterogeneity of a treatment effect across sub-populations, say mother’s age or other variables. First, we show that the PCQTE parameter is nonparametrically identified under selection on observable variables, which leads to an estimation procedure with two steps: nonparametric or parametric estimation of the propensity score and computation of the difference between the solutions of two separate minimization problems. Under some regularity conditions, we then show consistency and asymptotic normality of a fully nonparametric and a semiparametric estimator. The Monte Carlo study shows that, for a moderate sample size, the method produces good estimates. Moreover, the method developed here is applied to estimating the effect of the first-time mother's smoking during pregnancy on the baby's birth weight as a function of the mother's age across different quantiles for both whites and blacks. As result, the most interesting findings are that the effect changes over the mother’s age only for whites but not for blacks for a certain quantile. This leads us to further consider the testing issue that the effect has a particular form or there is no effect at all over age or a constant effect. In the other words, our interest is to test if the partially conditional quantile treatment effect model is correctly specified. To this end, we propose a consistent test based on the well-known Cramér–von Mises criterion and derive the asymptotic properties, including consistency and asymptotic normality, which is the novel in the econometrics/statistics literature. Finally, we apply the proposed procedure to testing if the partially conditional quantile treatment effect for both whites and blacks changes over age. It turns out that the partially conditional quantile treatment effect for whites indeed changes over age but not for blacks for some quantiles.
