Financial credit risk measurement prediction using innovative soft-computing techniques

Correct default risk classification of an issuer is a critical factor. Practitioners and academics alike agree on this. Thus, under the supervision of financial experts, significant resources of investment advisory companies are used for this task. Researchers, both theoretical and empirical ones, are not the exception either. Nowadays, many methodological and technical advances allow support for the work of classification of issuers. Learning algorithms based on Kernel Machines, particularly Support Vector Machines (SVM), have provided good results in classification problems when data are not linearly separable or noise patterns are employed for training. Moreover, on using kernel structures it is possible to deal with any data space, metric space not being necessary. The study presented in this paper is oriented towards credit risk modelling and measurement through qualitative kernel techniques. In particular, the process followed by the agencies that analyze and value the credit risk of companies, assigning a rating to them, is replicated. Results are expounded for the credit risk forecast of a group of companies that supply the market with public information. Companies’ economic and financial variables and their risk classification, issued by a well-known assessor of the financial market, are used for this purpose.