Simplex Techniques for Quantile Regression Model Selection

• Activity Number: 36
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Yonggang Yao*+
• Companies: SAS Institute
• Keywords: Quantile Regression ; Model Selection ; Simplex Algorithm ; Cross Validation ; Greedy Algorithm ; Statistical Regularization

Abstract

It is well known that a regular quantile regression problem can be cast as a linear programming problem and be solved by using a Simplex algorithm. However, fast Simplex algorithms for quantile model selection have not been sufficiently explored in literature. This work discusses the Simplex-based techniques that achieve fast computation for quantile model selection. The focus is on how to manipulate a Simplex algorithm for a variety of model selection processes such as greedy algorithms (forward, backward, and stepwise selection methods), leave-one-out cross validation, and penalty methods (e.g., LASSO penalty).