Generalized Additive Modelling of Credit Rating

• Activity Number: 156
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: International Chinese Statistical Association
• Author(s): Shuzhuan Zheng and Rong Liu*+ and Lijian Yang and Lifeng Wang
• Companies: Michigan State University and University of Toledo and Michigan State University and Michigan State University
• Keywords: Generalized Additive Model ; Confidence Band ; Credit Rating ; Accuracy Ratio ; Spline ; Kernel

Abstract

One central field of modern financial risk management is corporate credit rating in which default prediction plays a vital role. Parametric models of default prediction lack flexibility of model specification that leads to a low prediction power, while nonparametric models with higher prediction power are computationally demanding. We propose spline-backfitted kernel (SBK) estimator in the context of generalized additive model (GAM) with simultaneous confidence bands and BIC constructed for components testing and selection. First, GAM performs well in dimension reduction that allows to deal with a large set of covariates chosen from financial statements. Second, SBK estimator is much more computationally expedient than kernel smoothing, thus very practical for the fast prediction, and inference can be made on component functions with confidence. Third, we developed a BIC criterion to search significant covariates in the GAM modelling situation. Our method is applied to predict default probability of 3,472 listed companies of Japan, and the prediction displays nearly perfect cumulative accuracy profile (CAP) curves and very high accuracy ratio (AR) scores.