讲座题目： Periodicity Learning for Non-stationary Functional Time Series
主 讲 人： 尤进红
时 间： 2023年2023年12月7日14:30-15:30
地 点： 创新港涵英楼经济金融研究院8121会议室
报告人简介：
尤进红教授， 加拿大女皇大学（University of Regina）统计学博士，美国北卡罗纳教堂山分校博士后， 上海财经大学统计与管理学院tenured常任轨教授、博士生导师和统计与数据科学研究院副院长，全国工业统计学教学研究会第九届理事会副会长，全国统计学交叉学科分会副会长，曾任Quality Technology and Quantitative Management (QTQM), special issue: Mathematical and Statistical Finance的客座编委（Guest Editor）。 尤进红教授有十余年的北美学习、工作经历。长期从事计量经济学、数理统计以及生物统计的科学研究； 在面板数据、函数型数据数据分析及其在经济学， 金融学和生物医学方面的应用开展了许多有价值的研究工作；在国际和国内著名的统计学、经济学和数据科学杂志（包括Journal of the American Statistical Association，Journal of Econometric，Journal of Machine Learning Research等）上发表学术论文八十余篇，其中三大检索论文六十余篇，被SCI他引几百余次；主持和参与了多个国家自科基金项目。为国际著名统计和计量经济学杂志Annals of Statistics, Journal of the American Statistical Association , Biometrika和Journal of Econometric等的论文评审人。曾获上海市教委教学成果一等奖，上海财经大学先进工作者称号等奖项。
Abstract:
Existing literature on non-stationary functional time series primarily focuses on testing for stationarity and the existence of periodicity or trends. In this article, we present the first attempt to investigate the non-stationary periodic functional time series that has an unknown period and a functional trend. To achieve this, we propose a nonparametric profile estimation method that accurately estimates the unknown period, periodic component, and trend function. We establish the asymptotic properties of the estimators, including period estimate consistency and the asymptotic behaviours of the estimated periodic component and trend function. The finite sample performance of our method is investigated through simulation studies. Additionally, we apply our method to three real data applications, which include the functional time series of global temperature anomaly curves, daily CO and NOX concentrations in Hong Kong, and annual sunspot numbers. We report several intriguing discoveries. For instance, the estimated period of the functional time series of global temperature anomaly curves is 46 years, which is much shorter than the previously estimated period of 60 years based on conventional time series analysis.
经济与金融学院
2023年11月24日
