Title: An analysis of the key-variables of default risk using complex systems

Authors: Greta Falavigna

Addresses: Ceris-CNR, via Real Collegio 30, Moncaliere 10024, Turin, Italy; University of Eastern Piedmont 'Amedeo Avogadro', Novara, Italy

Abstract: This paper presents a method for the definition of the variables that determine the default. The choice of key-variables in default risk analysis is very important for banks and firms. Indeed, when a firm goes bankrupt, serious problems arise for all the stakeholders of the firm, as well as, for all subjects having relations with the failed subject. A feed-forward neural network with back propagation is used in this paper to classify firms and to divide them according to two groups: failed and not failed firms. The analysis of the model results shows that 13 variables, out of the 41 variables introduced in the network, are those that determine the default. This result is validated by using different techniques, such as the cluster analysis. Moreover, this study also finds that, by reducing the data introduced in the model, neural network models provide a very good performance. The conclusion is that Artificial Neural Network (ANN) models are able to identify key-variables for default risk.

Keywords: default risk; feed-forward neural networks; backpropagation; cluster analysis; regularisation; overfitting; key variables; risk assessment; artificial neural networks; ANNs; failed firms; firm classification.

DOI: 10.1504/IJBPM.2008.016640

International Journal of Business Performance Management, 2008 Vol.10 No.2/3, pp.202 - 230

Published online: 11 Jan 2008 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article