A Primer on Machine Learning Methods for Credit Rating Modeling

Using machine learning methods, this chapter studies features that are important to predict corporate bond ratings. There is a growing literature of predicting credit ratings via machine learning methods. However, there have been less empirical studies using ensemble methods, which refer to the technique of combining the prediction of multiple classifiers. This chapter compares six machine learning models: ordered logit model (OL), neural network (NN), support vector machine (SVM), bagged decision trees (BDT), random forest (RF), and gradient boosted machines (GBMs). By providing an intuitive description for each employed method, this chapter may also serve as a primer for empirical researchers who want to learn machine learning methods. Moody's ratings were employed, with data collected from 2001 to 2017. Three broad categories of features, including financial ratios, equity risk, and bond issuer's cross-ownership relation with the credit rating agencies, were explored in the modeling phase, performed with the data prior to 2016. These models were tested on an evaluation phase, using the most recent data after 2016.