Predicting the financial failures of firms traded on the stock markets of developed and developing countries

The purpose of this thesis is developing early warning system models that can predict the financial failure of a business 1 or 2 years in advance and revealing which methods are more accurate in predicting the failure. 16 financial ratios of 570 companies traded in the benchmark indices of G20 countries were adopted. Aforesaid firms were then analyzed with logistic regression, artificial neural networks and decision trees over the period 2010 – 2019. Results showed that logistic regression achieved 94,7% accuracy, decision trees produce 96,1% accuracy rate and artificial neural networks had the highest prediction accuracy rate of 98,1%, one year prior to the failure. In developed countries, the obtained classification accuracy of logistic regression, artificial neural networks and decision trees were equal to 94,3%, 98,6% and 95,9%, respectively. Firms in developing countries were classified with an accuracy rate of 97,2% using logistic regression, 97,9% using artificial neural networks and 94,4% using decision trees. Models created to forecast financial failure two years in advance revealed that logistic regression achieved 84,4% classification accuracy while artificial neural networks and decision trees had an accuracy of 92,5% and 91,1%, respectively. Logistic regression provided a correct classification accuracy of 89,3%, artificial neural networks had 96,6% accuracy and decision trees reached 92,1% predictive accuracy for firms in developed countries. Logistic regression, on the other hand, correctly classified 80,6% of firms operate in developing countries while artificial neural networks and decision trees achieved accuracy rate of 94,4% and 91,3%, respectively.