Many financial problems invovle classifying items into classes wich have a natural ordering. Examples include bond rating, credit scoring and equity risk rating. This paper discusses the special features of financial classification problems where the classes are ordered. The use of ordinal problbility models reslts in simpler, more interpretable classifiers. We propose a flexible ordinal model based on a feed-forward neural network: non-linear ordinal logistic regression (NOLR). NOLR is demonstrated on a synthetic example and is applied to the problem of risk rating UK equities. The model is used for prediction and as a tool for understanding the effects of fundamental factors on risk.
展开▼
