讲座题目：Slicing-free supervised dimension reduction for multivariate time series

主讲人：王国长

时间：2024年12月22日（星期日）15:00-17:00

地点：西安交通大学创新港涵英楼8121会议室

报告人简介：

王国长，暨南大学经济学院统计与数据科学系，教授、博士生导师。主要研究方向为函数型数据，时间序列和机器学习，至今在JoE、JBES、Sinica和Scandinavian Journal of Statistics等重要学术期刊发表论文30余篇。主持国家级项目4项，省部级项目4项。任中国现场统计研究会资源与环境统计分会常务理事；中国旅游大数据协会，理事，副秘书长；广东省现场统计协会常务理事，秘书长。

Abstract:Sufficient dimension reduction (SDR) methods have been widely studied for the regression model with independent data. However, there are not many sufficient dimension reduction methods for the time series mode, and most of them focus on multivariate predictors and scalar response such as time series sliced inverse regression(TSIR), time series sliced average variance estimation(TSAVE), and the TSSH, which is a hybrid version of TSIR and TSAVE. These above SDR methods for time series are very useful, but all of them are designed for scalar response and \sethlcolor{yellow}\hl{are} based on the slice approach. We can naturally extend these methods to multivariate responsees by marginally slicing each response, but this will create too many hyper-rectangular slices. Furthermore, the slice approach raises two main questions: how many slices should be chosen and how to divide all samples into different slices. To solve these problems, we will propose a time series version of slicing-free SDR method and call it time series martingale difference divergence matrix (TMDDM).

This method applies approximate joint diagonalization of several supervised lagged martingale difference divergence matrices(MDDM) to consider the temporal nature of the data. Based on TMDDM, we also discuss the strategies for selecting the lags of time series and the dimensionality of dimension reduction space. Simulations and real data analysis demonstrate the favorable finite sample performance of the proposed method.

经济与金融学院

2024年12月18日