Statistical Inference for High-dimensional Matrix-variate Factor Models with Missing Observations

This paper develops an inferential theory for high-dimensional matrix-variate factor models with missing observations. We propose an easy-to-use all-purpose method that involves two straightforward steps. First, we perform principal component analysis on two re-weighted covariance matrices to obtain the row and column loadings. Second, we utilize these loadings along with the matrix-variate data to derive the factors. We develop an inferential theory that establishes the consistency and the rate of convergence under general conditions and missing patterns. The simulation results demonstrate the adequacy of the asymptotic results in approximating the properties of a finite sample. Finally, we illustrate the application of our method using a real numerical dataset.