Bridging Joint and Grouped Sufficient Dimension Reduction: Application in Forecasting the Equity Risk Premium

• Activity Number: 683
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Haileab Hilafu*
• Companies: University of Tennessee
• Keywords: Central Subspace ; Sufficient Dimension Reduction ; Equity Risk Premium

Abstract

Sufficient dimension dimension methods are widely used in variety of applications to mitigate the effect of dimensionality and facilitate analysis of high-dimensional data. When there is prior predictor domain knowledge regarding grouping structure of the variables, it is useful to incorporate such domain information into dimension reduction and subsequent model formulation. Two popular approaches in this scenario are: grouped dimension reduction that carries dimension reduction of the variables in each group separately by ignoring the inter-dependence among the groups, and joint dimension reduction that ignores the grouping structure and reduces the variables jointly. In this article we propose a method that bridges the two approaches in the sense that it utilizes both the prior domain knowledge of grouping structure and potential inter-dependence among the groups, simultaneously. The usefulness of the method is studied via simulations. The method is also applied to a real data with the goal of forecasting the equity risk premium using 14 popular macroeconomic variables and 14 technical variables.