Machine learning and statistical techniques. An application to the prediction of insolvency in spanish non-life insurance companies

Abstract

Prediction of insurance companies insolvency has arisen as an important problem in the field of financial research. Most methods applied in the past to tackle this issue are traditional statistical techniques which use financial ratios as explicative variables. However, these variables often do not satisfy statistical assumptions, which complicates the application of the mentioned methods. In this paper, a comparative study of the performance of two non-parametric machine learning techniques (See5 and Rough Set) is carried out. We have applied the two methods to the problem of the prediction of insolvency of Spanish non-life insurance companies, upon the basis of a set of financial ratios. We also compare these methods with three classical and well-known techniques: one of them belonging to the field of Machine Learning (Multilayer Perceptron) and two statistical ones (Linear Discriminant Analysis and Logistic Regression). Results indicate a higher performance of the machine learning techniques. Furthermore, See5 and Rough Set provide easily understandable and interpretable decision models, which shows that these methods can be a useful tool to evaluate insolvency of insurance firms.