Abstract:Estimates of the approximate factor model are increasingly used in empirical work. Their theoretical properties, studied some twenty years ago, also laid the ground work for analysis on large dimensional panel data models with cross-section dependence. This paper presents simplified proofs for the estimates by using alternative rotation matrices, exploiting properties of low rank matrices, as well as the singular value decomposition of the data in addition to its covariance structure. These simplifications facilitate interpretation of results and provide a more friendly introduction to researchers new to the field. New results are provided to allow linear restrictions to be imposed on factor models.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to simplify the theoretical proofs in the estimation of large - dimensional approximate factor models. Specifically, the authors simplify the estimation proofs of these models by using alternative rotation matrices, exploiting the properties of low - rank matrices, and performing singular value decomposition of the data. This not only helps in the interpretation of the results but also provides a more user - friendly introductory guide for new researchers entering this field. In addition, the paper also presents new results that allow linear restrictions to be imposed in factor models.
### Background of the Paper
In the past two decades, there has been an active research area in panel data analysis, namely, the analysis of panel data with cross - sectional correlation, where the dimension of the panel is \(T\times N\), and both \(T\) (the time dimension) and \(N\) (the cross - sectional dimension) are very large. Classical factor models (such as those studied by Anderson and Rubin (1956) and Lawley and Maxwell (1974)) capture cross - sectional correlation when \(T\) or \(N\) is fixed and assume that the error terms are independently and identically distributed over time and units. The approximate factor model proposed by Chamberlain and Rothschild (1983) relaxes these assumptions, allowing the theory to be applied to real - world data.
### Contributions of the Paper
1. **Simplified Proofs**: The authors simplify the theoretical proofs of approximate factor model estimation by using alternative rotation matrices, the properties of low - rank matrices, and the singular value decomposition of the data. These simplifications not only make the results easier to understand but also provide a more user - friendly introductory guide for new researchers.
2. **New Results**: The paper presents new results that allow linear restrictions to be imposed in factor models, which are very useful in economic applications.
3. **Asymptotic Properties**: The paper re - derives the asymptotic properties of factor estimation, including the consistent estimation of the factor space, the consistent estimation of the number of factors, and the asymptotic normality of factors, loadings, and common components.
### Main Results
1. **Consistent Estimation of the Factor Space**: The factor space can be consistently estimated at the rate of \(\min(N, T)\).
2. **Consistent Estimation of the Number of Factors**: The number of factors can be consistently estimated.
3. **Asymptotic Normality**: The estimates of factors, loadings, and common components are asymptotically normal under \(\sqrt{N}\), \(\sqrt{T}\), and \(\min(\sqrt{N}, \sqrt{T})\) respectively.
### Methods
1. **Rotation Matrices**: The authors introduce four asymptotically equivalent rotation matrices, which simplify the proof of asymptotic normality.
2. **Eigenvalue Decomposition**: Using the eigenvalue decomposition of the data, rather than relying solely on its covariance structure, provides different representations of factor estimates, thereby simplifying the analysis.
3. **Linear Restrictions**: The paper provides an estimation method for imposing linear restrictions in factor models, which is very useful for economic analysis.
### Conclusions
By simplifying the theoretical proofs of approximate factor model estimation, the paper provides a more user - friendly introductory guide for new researchers and expands the application range of factor models, especially in the case of imposing linear restrictions in economic analysis. These results not only improve the understandability of the theory but also provide more flexibility for practical applications.