A System of Insolvency Prediction for industrial companies using a financial alternative model with neural networks

We find in the accounting literature the use of neural networks (NN) for the prediction of insolvency data from the last financial year before the bankruptcy, with a success rate below 85%. The objective of this work is to increase the predictive power of the NN models to discriminate between solvent and insolvent companies incorporating for this purpose a new set of financial ratios. A sample of about 500 European industrial companies that have filed for bankruptcy between 2007 and 2009 was confronted with about 500 solvent companies, matched by year, country and asset size. To do this, we have used five sets of different input data for training the NN. For each input set, 20 NN have been trained for each number of neurons in hidden layer, from 1 to 50 neurons, giving a total of 5 000 trained NN. The proposed model predicts correctly the 92.5 and 92.1 percent of the estimates of the training set and testing set (accuracy), respectively, using financial information for two years prior to bankruptcy.