Adjusted MSE Sequential Method

• Activity Number: 500
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): Limin Yang*+ and Li Yang
• Companies: Fiserv and Renmin University of China
• Keywords: Variable selection ; SAS phreg ; Stepwise ; Collinearity ; Modeling

Abstract

Variable selection is a common problem in modeling, especially for financial data. Having several hundred of higher correlated or collinear variables is not unusual. This makes it difficult to select more important variables into a model. Although many works have been done in the area, the problem is far from being solved in practice. This paper suggests an adjusted MSE sequential method, which sequentially excludes and includes variables based on their contributions to the MSE adjusted with collinear factor. By considering both their collinearity and contribution to the model, we can keep more important and less collinear variables. There are four steps: stepwise selection; variable excluding loop; variable including loop; model finalization. An example is shown and it confirmed that this method did select better variables and produced an improved model than just using proc phreg stepwi