Expert systems for bond rating: a comparative analysis of statistical, rule‐based and neural network systems

Abstract

An important problem in financial investment is the classification of bonds based on the likelihood that the issuing company may default on the promised payments. Much effort has been invested into simulating the bond rating process using statistical tools. A weakness of these tools is the requirement of statistical assumptions which may not be appropriate for the bond rating problem. In this paper we present results of a study comparing an artificial neural network system, a rule-based expert system and statistical techniques applied to the bond rating problem. The bond rating process is simulated by using published financial data.

Using Neural Networks to Model Sovereign Credit Ratings: Application to the European Union

“…In this regard, numerous empirical research has compared neural network methodology with econometric techniques for modeling ratings, and consistent with theoretical arguments, the results clearly demonstrate that neural networks represent a superior methodology for calibrating and predicting ratings relative to linear regression analysis, multivariate discriminant analysis and logistic regression (e.g [1,10]). Leshno and Spector [11] even point out that the methods traditionally used in credit risk analysis, which include regression analysis, multivariate discriminant analysis, and logistic regression among others, represent special cases of neural networks, and therefore it is not surprising that they are appropriate for credit rating modeling.…”

Using neural networks for forecasting volatility of S&P 500 Index futures prices

“…• Bond risk analysis: Dutta and Shekhar (1988), Moody and Utans (1991), Surkan and Singleton (1991), Kim, Weistroffer and Redmond (1993), Maher and Sen (1997).…”

Forecasting Bank Credit Ratings

“…In these studies different sets of variables were used and the prediction results were between 50% and 70%. Many studies on bond credit rating prediction with neural networks (Dutta et al (1988); Surkan et al (1990); Kim (1993)) show more promising results than statistical methods. Moody and Utans (1994) used neural networks to predict bond ratings of firms that had a rating from S&P. Using 10 input variables to predict 16 S&P subratings they managed to predict correctly just 36.2% of the ratings.…”