1: \begin{abstract}

2:   Tensor time series, which is a time series consisting of tensorial observations, has become ubiquitous. It typically exhibits high dimensionality. One

3:   approach for dimension reduction is to use a factor model structure, in

4:   a form similar to Tucker tensor decomposition, except that the time

5:   dimension is treated as a dynamic process with a time dependent structure.

6:   In this paper we introduce two approaches to estimate such a tensor factor

7:   model by using iterative orthogonal projections of the original tensor

8:   time series. The approaches extend the existing estimation procedures and

9:   our theoretical investigation shows that they

10:   improve the estimation accuracy and convergence rate significantly.

11:   The developed approaches are similar to higher order orthogonal projection

12:   methods for tensor decomposition, but with significant differences and

13:   theoretical properties.

14:   Simulation study is conducted to further illustrate the statistical properties of these estimators.

15: 

16: \end{abstract}

17: