1: \begin{abstract}

2: This paper considers the maximum likelihood estimation of factor models

3: of high dimension, where the number of variables ($N$) is comparable

4: with or even greater than the number of observations~($T$). An

5: inferential theory is developed. We establish not only consistency but

6: also the rate of convergence and the limiting distributions. Five

7: different sets of identification conditions are considered. We show

8: that the distributions of the MLE estimators depend on the

9: identification restrictions. Unlike the principal components approach,

10: the maximum likelihood estimator explicitly allows

11: heteroskedasticities, which are jointly estimated with other

12: parameters. Efficiency of MLE relative to the principal components

13: method is also considered.

14: \end{abstract}