Prognose von Insolvenzwahrscheinlichkeiten mit Hilfe logistischer neuronaler Netzwerke

Summary In this article we forecast the probability of small and medium sized companies to go bankrupt. Although such companies face a particular high insolvency risk, very little research has been done in this area. The difficulty of forecasting the insolvency risk of small and medium sized companies mainly consists of a lack of balance sheet data so that one has to rely on qualitative Information as well. We show that despite of that constraint good forecasts of insolvency risk are possible. The methods used are logistic regression and neural networks. In order to select appropriate modeis we apply Statistical inference techniques. For logistic regression this is a Standard. However, for neural networks this is new. By help of Statistical inference techniques we create parsimonious modeis and avoid overfitting of the data. We then analyze the mutual dependencies in the resulting modeis. As opposed to the logistic regression neural network modeis are able to map nonlinear dependencies between the variables, which allows for a more precise deduction as to which and how variables contribute to the insolvency risk.