Fixed-T estimation of the matrix-valued factor models

Due to the emergence of large-scale matrix-variate data, the matrix-valued factor model (MVFM) has received much attention of the literature in the past few years. As a natural extension of the high-dimensional factor model, most existing methods operate under the large $N$, large $T$'' context. However, many such datasets have either a short time span or a low frequency. This paper considers the estimation of the MVFM under thelarge $N$, fixed $T$’’ context. We show that the 2DSVD method continues to work. We derive the consistency and asymptotic normality of the estimator. The performance of our estimator is evaluated through simulations and applications with real data.