报告题目：Statistical Diagnostics for Time Series Models

报告时间：2025年12月29日中午12:30

报告地点：北湖东校区数统新楼201室

主办单位：激情视频

报告人：刘双喆

报告人简介：刘双喆，教授，现任澳大利亚堪培拉大学科技学院数据科学组组长。他在荷兰阿姆斯特丹大学廷伯根研究所获得计量经济学博士学位，专门研究矩阵微分学、多元分析和统计学习。他的主要工作发表于数学、统计学和相关领域的著名期刊。此外，参与编著了SAS Enterprise Guide进行时间序列分析的综合书籍。担任多家统计期刊的副主编，积极为该领域做出贡献，并在《Statistical Papers》担任编辑工作。

摘要：The reliability of a statistical model hinges on validating its assumptions and identifying influential observations. While regression diagnostics are well-established for linear models, their application to modern time series frameworks—such as those for count data or skewed financial returns—presents significant challenges. This talk introduces powerful diagnostic tools, with a focus on local influence analysis. We develop and apply novel influence diagnostics for Integer-Valued GARCH (INGARCH) models under Poisson-type distributions, which are essential for analyzing discrete data like daily stock transactions or infectious disease case counts. Using real-world applications, we demonstrate how these diagnostics contribute to more robust, reliable, and trustworthy time series analysis in practice.