Doubly Constrained Factor Models: Estimation and Applications

• Activity Number: 173
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Henghsiu Tsai*+ and Ruey S. Tsay
• Companies: Academia Sinica and The University of Chicago
• Keywords: Approximate factor model ; Constrained principal component analysis ; Least squares estimation ; Likelihood ratio test ; Eigenvalues ; Noise-to-signal ratio

Abstract

Factor models have been widely used in recent years to improve the accuracy of forecasting when many explanatory variables are available. However, the models often encounter the difficulties of over-parameterization and factor interpretation. In this paper, we first consider constrained factor analysis to obtain a parsimonious factor model and propose likelihood ratio statistics to test the adequacy of factor constraints. We then extend the constrained models to the doubly constrained factor models by incorporating external information on both rows (e.g., subjects) and columns (e.g., variables) of a data matrix. Real and simulated examples are used to demonstrate the proposed analysis. In an application, we show that the doubly constrained factor analysis can provide a deeper understanding of variations in monthly financial asset returns.