1: \begin{abstract}

2: This paper deals with the factor modeling for high-dimensional time series

3: based on a dimension-reduction viewpoint.

4: % Following a brief survey on the factor models for multiple time

5: % series in econometrics, we introduce a statistical approach from the

6: % viewpoint of dimension reduction. Our method can handle

7: % nonstationary factors. However

8: Under stationary settings, the inference is simple in the sense that

9: both the number of factors and the factor loadings are estimated in

10: terms of an eigenanalysis for a~nonnegative definite matrix, and is

11: therefore applicable when the dimension of time series is on the order

12: of a few thousands. Asymptotic properties of the proposed method are

13: investigated under two settings: (i) the sample size goes to infinity

14: while the dimension of time series is fixed; and (ii) both the sample

15: size and the dimension of time series go to infinity together. In

16: particular, our estimators for zero-eigenvalues enjoy faster

17: convergence (or slower divergence) rates, hence making the estimation

18: for the number of factors easier. In particular, when the sample size

19: and the dimension of time series go to infinity together, the

20: estimators for the eigenvalues are no longer consistent. However, our

21: estimator for the number of the factors, which is based on the ratios

22: of the estimated eigenvalues, still works fine. Furthermore, this

23: estimation shows the so-called ``blessing of dimensionality'' property

24: in the sense that the performance of the estimation may improve when

25: the dimension of time series increases. A two-step procedure is

26: investigated when the factors are of different degrees of strength.

27: Numerical illustration with both simulated and real data is also

28: reported.

29: \end{abstract}