报告题目：Robust Matrix FactorModels withQn-basedAutocovariancesfor Outlier Detection

报告人：郝程程，上海对外经贸大学

报告时间：2026年10月20日(星期二) 10:15-11:00

报告地点：伍卓群楼第一报告厅

校内联系方式：邹婷婷[email protected]

报告摘要：

  Matrix factor models provide two-way dimension reduction for matrix-valued time series, but loading-spaceestimatorsbasedon samplelaggedautocovariancescanbehighlysensitivetoextremeobservations.Weproposetworobustspectralestimators,rTOPUP andrTIPUP,by replacing the entrywise lagged autocovariances underlying TOPUP and TIPUP with scale-difference estimatorsconstructedfromtherobustQnstatistic.UnderGaussian short-range dependence, a uniform root-TconditionfortherelevantQnscales,and mode-specific iden-tification conditions, we establish root-Tconvergenceof therobustlagged-autocovariance operators and the resultingloading-spaceestimators. Wealsodevelopa residual-based procedure that first detects outlying time pointsthroughrobustlycalibratedFrobenius norms and then localizes outlying cells usingQn-standardized residuals.Simulations under cellwise additive contamination show thatrTIPUP substantially improves loading-space recovery relative to TIPUP whileincurringonlyamodestlossundercleansampling; rTOPUP is less affected by contamination but remains inaccurate in the simulationdesign.AnapplicationtoChina’sexportsacross97commoditycategoriesand30tradingpartners over 2014–2024 identifies2024astheonlygloballyanomalousyearandlocalizes146 commodity–partner cells contributing to that departure.

报告人简介：

  郝程程，上海对外经贸大学副教授，瑞典斯德哥尔摩大学统计学博士。主要研究方向为张量数据分析、异常点诊断，在《Scandinavian Journal of Statistics》等期刊上发表一作论文10余篇；主持完成国自科、上海市科委、瑞典皇家科娱乐城等课题，指导学生获全国研究生统计建模大赛一等奖、“三下乡”全国优秀调研报告、“挑战杯”上海市一等奖等，获上海高校青年东方学者、上海市白玉兰人才、上海市教学成果特等奖。