报告题目：Inference for High Dimensional Quantile Regression

报告时间：2017年12月27日（周三） 8:30

报告地点：统计学院办公楼1层104会议室

主讲人：王会霞, 乔治华盛顿大学统计系教授。复旦大学统计系获得本科（1999年）及硕士（2002年）学位，从美国伊利诺伊大学获得博士学位（2006年），曾于2012年获得美国科学研究基金会Career Award及国际数理统计协会Tweedie Award，担任美国统计协会杂志和统计年鉴等顶尖统计杂志的副主编， 主持了多项美国国家自然科学基金的项目。

报告摘要: In this talk I will discuss inference for high dimensional quantile regression, and present two new tests based on maximum-type statistics to detect the presence of significant predictors associated with the quantiles of a scalar response. The first test is based on the t-statistic associated with the chosen most informative predictor at the quantiles of interest. A resampling method is devised to calibrate this test statistic, which has non-regular limiting behavior due to a weak identifiability issue. The second test is based on the maximum of score-type statistics. We show that for diverging dimensions, the test statistic converges to the extreme value distribution of Type I under the null hypothesis under some regularity conditions. For finite samples, we also introduce a convenient multiplier bootstrap method to construct critical values. The proposed tests are more flexible than existing methods based on mean regression, and have the added advantage of being robust against outliers in the response.