Predicting Bankruptcy with Support Vector Machines

The purpose of this work is to introduce one of the most promising among recentlydeveloped statistical techniques – the support vector machine (SVM) –to corporate bankruptcy analysis. An SVM is implemented for analysing suchpredictors as financial ratios. A method of adapting it to default probabilityestimation is proposed. A survey of practically applied methods is given. Thiswork shows that support vector machines are capable of extracting useful information from financial data, although extensive data sets are required in orderto fully utilize their classification power. The support vector machine is a classification method that is based on statistical learning theory. It has already been successfully applied to optical character recognition, early medical diagnostics, and text classification. One application where SVMs outperformed other methods is electric load prediction (EUNITE, 2001), another one is optical character recognition (Vapnik, 1995). SVMs produce better classification results than parametric methods and such a popular and widely used nonparametric technique as neural networks, which is deemed to be one of the most accurate. In contrast to the latter they havevery attractive properties. They give a single solution characterized by theglobal minimum of the optimized functional and not multiple solutions associatedwith the local minima as in the case of neural networks. Moreover, SVMsdo not rely so heavily on heuristics, i.e. an arbitrary choice of the model andhave a more flexible structure.