1: \begin{abstract}

2: This paper introduces a Projected Principal Component Analysis 

3: (Projected-PCA), which employs principal component analysis to the

4: projected (smoothed) data matrix onto a given linear space spanned by

5: covariates. When it applies to high-dimensional factor analysis, the

6: projection removes noise components. We show that the unobserved latent

7: factors can be more accurately estimated than the conventional PCA if

8: the projection is genuine, or more precisely, when the factor loading

9: matrices are related to the projected linear space. When the

10: dimensionality is large, the factors can be estimated accurately even

11: when the sample size is finite. We propose a flexible semiparametric

12: factor model, which decomposes the factor loading matrix into the

13: component that can be explained by subject-specific covariates and the

14: orthogonal residual component. The covariates' effects on the factor

15: loadings are further modeled by the additive model via sieve

16: approximations. By using the newly proposed Projected-PCA, the rates of

17: convergence of the smooth factor loading matrices are obtained, which

18: are much faster than those of the conventional factor analysis. The

19: convergence is achieved even when the sample size is finite and is

20: particularly appealing in the high-dimension-low-sample-size situation.

21: This leads us to developing nonparametric tests on whether observed

22: covariates have explaining powers on the loadings and whether they

23: fully explain the loadings. The proposed method is illustrated by both

24: simulated data and the returns of the components of the S\&P 500 index.

25: \end{abstract}