The Efficiency of Neural Networks, Logistic Regression & Discriminant Analysis in Defaults Prediction

There are different statistical models for prediction and classifications in science. Statistical and econometric models like regression, discriminant analysis, time series, ratings and the others are used for forecasting and classifying, based on variables and data about certain subject. However, statistical models have many presumptions and limitations, so in spite of different existing statistical models, recently, neural networks as modern method of forecasting have received considerable attention for its high efficiency and no needs to the presumption and limitation in data distributions. This study is aimed to compare the ability of artificial neural networks model to that of logistic regression and discriminant models in forecasting default events. To do this, logistic regression, discriminant analysis and neural network models were evaluated using the data of 23801 contracts and taking the contract time, contract amount, industry type, contract type, guarantee type and credit policy as forecasting variables. The ability of models in predicting default risks was compared using ROC analysis and the classification accuracy reviews. The results of this study indicate that the variables mentioned above are significant in forecasting defaults and artificial neural networks model is more efficient as compared to the two other models.