Abstract:We establish central limit theorems for principal eigenvalues and eigenvectors under a large factor model setting, and develop two-sample tests of both principal eigenvalues and principal eigenvectors. One important application is to detect structural breaks in large factor models. Compared with existing methods for detecting structural breaks, our tests provide unique insights into the source of structural breaks because they can distinguish between individual principal eigenvalues and/or eigenvectors. We demonstrate the application by comparing the principal eigenvalues and principal eigenvectors of S\&P500 Index constituents' daily returns over different years.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is to develop a method to detect structural breaks in large - scale factor models and be able to distinguish whether these structural breaks are caused by changes in the principal eigenvalues, changes in the principal eigenvectors, or changes in both. Specifically, the authors established the central limit theorems for the principal eigenvalues and eigenvectors and proposed two sample - testing methods to test whether the principal eigenvalues and eigenvectors in two different time periods are equal. This method is of great significance for understanding the structural changes in economic or financial data, because different sources of change may have different economic or financial implications.
### Background and Problem Description of the Paper
In economics and finance, factor models are widely used to explain the dynamic relationships among a large number of variables. Time - invariance is a standard assumption in the study of factor models, that is, the factor loadings are fixed or at least approximately fixed over time. However, in actual time - series data, parameter instability is a common phenomenon, and this instability may be caused by changes in policy regimes, changes in economic or financial fundamentals, etc. Therefore, potential structural changes in actual data need to be treated with caution.
Existing methods for detecting structural breaks usually compare whether the factor loading matrix \(B\) is the same in two time periods, but these methods cannot distinguish the specific causes of structural changes, that is, whether they are caused by changes in the principal eigenvalues, changes in the principal eigenvectors, or changes in both.
### Objectives of the Paper
The objective of this paper is to precisely locate the sources of structural changes by detecting changes in individual principal eigenvalues and eigenvectors. Specifically, when a structural break occurs, it can be determined whether the break is caused by changes in the principal eigenvalues, changes in the principal eigenvectors, or changes in both. This helps to gain a deeper understanding of the economic or financial significance behind the structural changes.
### Main Contributions
1. **Establishing Central Limit Theorems**: The authors established central limit theorems for the principal eigenvalues and eigenvectors in large - scale factor models, which is the basis for statistical testing.
2. **Proposing Two Sample - Testing Methods**: Based on the above central limit theorems, the authors developed two sample - testing methods to detect whether the principal eigenvalues and eigenvectors in two different time periods are equal.
3. **Distinguishing Sources of Changes**: By detecting changes in individual principal eigenvalues and eigenvectors, the causes of structural changes can be identified more precisely, thus providing more targeted economic or financial analysis.
### Application Examples
The authors demonstrated the application of this method by comparing the principal eigenvalues and principal eigenvectors of the daily returns of the S&P 500 index components in different years. This not only verified the effectiveness of the method but also provided specific case support for practical applications.
### Conclusions
This paper provides a new tool for detecting structural changes in large - scale factor models by establishing central limit theorems for the principal eigenvalues and eigenvectors and proposing two corresponding sample - testing methods. This method can not only detect the existence of structural changes but also precisely locate the sources of changes, thus providing deeper insights for economic and financial data analysis.