Bankruptcy Prediction of Listed Corporations in Tehran Stock Exchange Using Data Mining Techniques

Aims: This study aimed at predicting bankruptcy based on two data mining techniques, i.e. logistic regression and classification and regression tree (CART). Study design: This was an applied, descriptiveanalytical, cross-sectional study. Place and Duration of Study: This research was carried out in Iran. Annual financial statements of companies in Tehran stock market (Iran) during 1999-2010 were evaluated. Methodology: No sampling was performed. In total, 98 successful companies and 71 bankrupt companies were included. In designing the models, financial ratios were considered as independent variables and successful and bankrupt companies were regarded as dependent variables. Results: According to the CART, the most important variables in predicting bankruptcy were return on assets, debt ratio, operating profit to total assets ratio, working capital to total assets ratio, and operating profit margin. The logistic regression model suggested return on assets, debt ratio, and operating profit to total assets ratio as predicting variables. The overall accuracy of prediction using the CART was 100% on training data and 92.8% on test data. The accuracy of regression model was 95.9%. Conclusion: Both models revealed return on assets, debt ratio, and operating profit to total assets ratio to be the most important variables in predicting bankruptcy. However, considering the area under the receiver operating characteristic (ROC) curve, the logistic regression model had better performance in predicting bankruptcy. Keywords— Prediction of bankruptcy, Data mining, Decision tree, Classification and regression tree, Logistic regression.