Neural Networks Versus Logit Regression Model For Predicting Financial Distress Response Variables

Abstract

Neural networks are designed to detect complex relationships among variables better than traditional statistical methods. Our study examined whether the complexity of the response measure impacts whether logistic regression or a neural network produces the highest classification accuracy for financially distressed firms. We compared results obtained from the two methods for a four state response variable and a dichotomous response variable. Our results suggest that neural networks are not superior to logistic regression models for the traditional dichotomous response variable, but are superior for the more complex financial distress response variable.