Predicting the Bankruptcy of Construction Companies: A CART-Based Model

Abstract

Company bankruptcy is a frequent research topic, and with regard to current economic developments, its importance and relevance is undisputable. Models created in different conditions do not achieve the accuracy claimed by their authors. The aim of this article is to present the results of research associated with the development of a bankruptcy model designed for construction companies. Due to the properties of financial data, the quality of the model, i.e. its discrimination ability, is strongly influenced by the choice of the method used to derive the model. The model was developed for the years 2011–2014 based on 29 financial indicators of companies operating in the construction industry, which were calculated on the basis of accounting data of the companies contained in the AMADEUS database. A non-parametric method of Classification and Regression Trees (CART) was used to derive this model. The discrimination ability of the model was evaluated based on the percentage of correctly differentiated companies and the percentage of Type I and Type II errors. To test the discrimination accuracy of models, the Receiver Operating Characteristic curve (ROC curve) and the Area Under Curve (AUC) were also used. Based on these tests, it is possible to graphically and numerically measure the discrimination ability of the models. The discrimination accuracy of the model created on the basis of the data of construction companies was compared with other selected models, which were not created using the data of construction companies. The comparison clearly showed that the model created especially for the construction companies achieves the highest discrimination ability.Company bankruptcy is a frequent research topic, and with regard to current economic developments, its importance and relevance is undisputable. Models created in different conditions do not achieve the accuracy claimed by their authors. The aim of this article is to present the results of research associated with the development of a bankruptcy model designed for construction companies. Due to the properties of financial data, the quality of the model, i.e. its discrimination ability, is strongly influenced by the choice of the method used to derive the model. The model was developed for the years 2011–2014 based on 29 financial indicators of companies operating in the construction industry, which were calculated on the basis of accounting data of the companies contained in the AMADEUS database. A non-parametric method of Classification and Regression Trees (CART) was used to derive this model. The discrimination ability of the model was evaluated based on the percentage of correctly differentiated companies and the percentage of Type I and Type II errors. To test the discrimination accuracy of models, the Receiver Operating Characteristic curve (ROC curve) and the Area Under Curve (AUC) were also used. Based on these tests, it is possible to graphically and numerically measure the discrimination ability of the models. The discrimination accuracy of the model created on the basis of the data of construction companies was compared with other selected models, which were not created using the data of construction companies. The comparison clearly showed that the model created especially for the construction companies achieves the highest discrimination ability.