讲座题目：Specification analysis of conditional distribution functions
内容摘要：This paper proposes different tests for the correct specification of conditional distribution functions. We first transform our hypothesis of interest by means of the integrated moment approach, focusing on the linear indicator and exponential weight functions. Empirical processes based on the two characterizations are constructed and an extra projection onto the tangent space of parameter is introduced to eliminate the parameter estimation uncertainty. Eight Cram′er-von Mises-type test statistics using both integrable and nonintegrable weights are proposed. The asymptotic null distributions of our test statistics are derived. Local alternatives converging to the null at a parametric rate can be detected. Besides, our tests are fully data-driven, neither rely on any tuning parameters such as bandwidth nor require estimators to admit an asymptotic linear representation. With the assistance of an easy-to-implement multiplier bootstrap procedure, the critical values can be approximated as accurately as desired. Finite sample performance in simulation studies illustrates the usefulness of our proposed tests.
主讲人简介：Dr.Xiaojun Song is an Assistant Professor in Guanghua School of Management, Peking University. He has published several papers in top-field econometric journals, such as Journal of Econometrics, Journal of Business & Economic Statistics, Oxford Bulletin of Economics and Statistics, among others.