A0704
Title: Estimation and inference for CP tensor factor models
Authors: Bin Chen - University of Rochester (United States) [presenting]
Yuefeng Han - University of Notre Dame (United States)
Qiyang Yu - University of Rochester (United States)
Abstract: High-dimensional tensor-valued data have recently gained attention from researchers in economics and finance. The estimation and inference of high-dimensional tensor factor models are considered, where each dimension of the tensor diverges. The focus is on a factor model that admits CP-type tensor decomposition, which allows for non-orthogonal loading vectors. Based on the contemporary covariance matrix, an iterative simultaneous projection estimation method is proposed. The estimator is robust to weak dependence among factors and weak correlation across different dimensions in the idiosyncratic shocks. An inferential theory is established, demonstrating both consistency and asymptotic normality under relaxed assumptions. Within a unified framework, two eigenvalue ratio-based estimators are considered for the number of factors in a tensor factor model, and their consistency is justified. Through a simulation study and two empirical applications featuring sorted portfolios and international trade flows, the advantages of the proposed estimator are illustrated over existing methodologies in the literature.
