Dynamic Variable Selection for Corporate Bankruptcy Prediction

• Activity Number: 631
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Business and Economic Statistics Section
• Author(s): Yan Yu and Shaonan Tian*+ and Hui Guo
• Companies: University of Cincinnati and University of Cincinnati and University of Cincinnati
• Keywords: Default Prediction ; Hazard Model ; LASSO ; Model Selection ; Panel Data

Abstract

Corporate bankruptcy forecasting has received paramount interest both in academic research, business practice and government regulation, especially during today's challenging economy. To facilitate accurate default probability estimation, we introduce the "least absolute shrinkage and selection operator" (LASSO) method, a fully data driven variable selection process to select important financial ratios and/or market predictors with time-varying panel data. We investigate such dynamic variable selection on a comprehensive US bankruptcy database spanning 1980 to 2009, by merging Compustat and CRSP. The simple hazard model using the selected predictor variable yields significantly improved out-of-sample predictive power. In addition, we find six out of eight predictor variables proposed in Campbell et al (2008)'s seminal work also enters the variable set selected by LASSO method, except for the equity size and the short-term-asset variable. We also note the significance of the variable Market-to-Book ratio however appears to decrease when the prediction horizon extends from one month to one year.