Dynamic Tensor Factor Model Based on CP Decomposition

• Activity Number: 392 - Recent Advances in Tensor Learning
• Type: Invited
• Date/Time: Thursday, August 12, 2021 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Yuefeng Han* and Rong Chen and Cun-Hui Zhang
• Companies: Rutgers University and Rutgers University and Rutgers University
• Keywords: Tensor data analysis; Factor model; CP decomposition; Time series; High dimension; Autocovariance

Abstract

Tensor time series, which is a time series consisting of tensorial observations, has become ubiquitous. Existing work adopts Tucker tensor decomposition, which produces very complicated core tensor factor process. In this talk, we consider CP decomposition of the tensor factor model. The factor process is univariate and thus can be efficiently modelled with linear and nonlinear models. We develop and study a class of higher-order-efficiency (HOHE) methods based on a new and exciting idea. We investigate efficient estimation procedures for the loading vectors based on the HOHE and related iterative algorithm.