1: \begin{abstract}

2: 

3: This paper considers the estimation and inference of the low-rank components in high-dimensional matrix-variate factor models, where each dimension of the matrix-variates ($p \times q$) is comparable to or greater than the number of observations ($T$).

4: We propose an estimation method called $\alpha$-PCA that preserves the matrix structure and aggregates mean and contemporary covariance through a hyper-parameter $\alpha$.

5: We develop an inferential theory, establishing consistency, the rate of convergence, and the limiting distributions, under general conditions that allow for correlations across time, rows or columns of the noise.

6: We show both theoretical and empirical methods of choosing the best $\alpha$, depending on the use-case criteria.

7: Simulation results demonstrate the adequacy of the asymptotic results in approximating the finite sample properties.

8: The $\alpha$-PCA compares favorably with the existing ones.

9: Finally, we illustrate its applications with a real numeric data set and two real image data sets.

10: In all applications, the proposed estimation procedure outperforms previous methods in the power of variance explanation using out-of-sample 10-fold cross-validation.

11: 

12: \vspace{1em}

13: 

14: \noindent {\it Key words}: Matrix-variate; Latent low rank; Factor models; Asymptotic normality; High-dimension.

15: \end{abstract}

16: