CHINA·77779193永利(集团)有限公司-Official website

题目：TARGET INFERENCE FOR HIGH-DIMENSIONAL QUANTILE REGRESSION

报告人：蒋建成 （北卡大学夏洛特分校）

摘要：Quantile regression is a powerful tool for uncovering relationships between predictors and responses, particularly in high-dimensional data where varied effects can be detected. Focusing on key predictors at a specific quantile, while treating others as nuisance variables, presents a unique hypothesis testing challenge. This research introduces an innovative inference framework that employs dimension-reduced convolution-smoothed quantile regression, while avoiding estimating the inverse of high-dimensional covariance matrix of the predictors. By calibrating the regularization parameter, we develop a data-driven test that can be shown to be an oracle test with probability tending to one. To mitigate the selective bias induced by dimension reduction and ensure valid inference, we implement a cross-fitting strategy by dividing the dataset into two parts: one for model selection and the other for parameter estimation. This process yields a fused estimator, derived from an informative weighting method that combines estimators from both dataset partitions.

The optimal fused estimator aids in constructing confidence intervals and performing Wald-type tests for targeted parameters. We establish the Bahadur’s representation of this estimator and obtain limiting distributions of the test statistics under both null and alternative hypotheses, with the number of parameters diverging to infinity. Advantages of our tests are further highlighted by theoretical power comparisons to some competitive tests. Empirical studies confirm effectiveness of the proposed tests across various linear parameter hypotheses. Additionally, we illustrate the use of the proposed methodology through two real-world data analyses.      

报告人简介：蒋建成博士目前是北卡大学夏洛特分校数学统计系和数据科学学院的双聘教授，也曾兼任南开大学讲座教授。他的研究领域包括非参数建模，金融时间序列分析，生存分析，统计和机器学习，大数据分析等。目前已在包括美国统计学会会刊，英国皇家统计学会会刊，统计年刊，计量经济学杂志等刊物上发表科研论文70余篇。自2018至2024年担任北卡大学数学统计系统计学项目主管。目前兼任该校可信人工智能中心联合研究员。

报告时间：2025年5月28日（周三）下午16：00-17：00

报告地点：教二楼727

联系人：胡涛