A Comparative Study with Quantile Regression and Back Propagation Neural Network for Credit Rating

Abstract

In this study, we use the quantile regression and the back propagation neural network to construct a credit rating model for companies listed in Taiwan Stock Exchange and Over-The-Counter. The data we use is from 1997 to 2013 in Taiwan. The data in the period from 1997 to 2005 is in sample and the data in the period from 2006 to 2013 is out of sample. TCRI established by TEJ is used as a dependent variable to analyze the relationship between 12 financial ratios and credit rating. Our results show that the average forecasting correction rate based on the propagation neural network, which is about 70%, is higher than that based on the quantile regression, which is about 60%. However, investors and financial institution are mainly concerned about the companies facing bankruptcy so they are more interested in which companies bear higher risk. In this case, the quantile regression can provide higher forecasting correction rate for low-credit-ranking companies, which is about 80%, than that provided by the back propagation neural network, which is about 55%.