Bankruptcy Prediction Models: Artificial Neural Networks versus Discriminant Analysis and Logit Model

Abstract

The purpose of this paper is to develop the validation of bankruptcy prediction models for Mongolian companies by using most useful current techniques. To do this, we reviewed 510 bankruptcy prediction models that had been published in 296 academic journals from 1966 to 2015. We also focused on the methodology of the models predictability, application, and relating factors. There are more than 600 different variables in the models and 86 percent out of them are applied as financial variables. Therefore, using financial ratios is an essential method to analyze the financial reports, the prediction of financial distress and bankruptcy. Moreover, some variables, such as corporate governance, macroeconomic and industry effect reflected variables have been used to develop bankruptcy prediction models in modern studies. We selected a sample of 16 bankrupt and 426 non-bankrupt companies for the years 2010 to 2015. Based on research results, ANN model with variables of EBIT to total asset, equity to total asset, liabilities to equity and logarithm of total asset has shown more capability to predict corporate bankruptcy in Mongolia.