1: \begin{abstract}

2: This paper studies the principal component (PC) method-based estimation of

3: weak factor models with sparse loadings. We uncover an intrinsic

4: near-sparsity preservation property for the PC estimators of loadings, which

5: comes from the approximately \textit{upper triangular (block)} structure of

6: the rotation matrix. It implies an \textit{asymmetric} relationship among

7: factors: the rotated loadings for a stronger factor can be contaminated by

8: those from a weaker one, but the loadings for a weaker factor is almost free

9: of the impact of those from a stronger one. More importantly, the finding

10: implies that there is no need to use complicated penalties to sparsify the

11: loading estimators. Instead, we adopt a simple screening method to recover

12: the sparsity and construct estimators for various factor strengths. In

13: addition, for sparse weak factor models, we provide a singular value

14: thresholding-based approach to determine the number of factors and establish

15: uniform convergence rates for PC estimators, which complement Bai and Ng

16: (2023). The accuracy and efficiency of the proposed estimators are

17: investigated via Monte Carlo simulations. The application to the FRED-QD

18: dataset reveals the underlying factor strengths and loading sparsity as well

19: as their dynamic features.

20: 

21: \noindent \textbf{JEL\ Classification: } C12, C53, C55

22: 

23: \noindent \textbf{Key Words: }Factor number, Factor strength, Principal

24: component, Rotation matrix, Sparse factor loadings, Sparsity preservation,

25: Weak factors

26: \end{abstract}

27: