Predicting bankruptcy based on the full population of Croatian companies

This paper analyses the bankruptcy prediction based on the population of companies representative of the total business sector in Croatia. The representativity of the sample is achieved through the propensity score matching of the full population of bankrupt and similar non-bankrupt companies. The robust estimation of bankruptcy prediction is carried out through the multiple discriminant analysis (MDA) and logistic regression (logit). The results indicate high classification accuracy of both models, but more favourable performance of the logit estimation. Overall accuracy of the MDA model was 73.7%, while the overall accuracy of the logit model was 76.3%. The results serve as a bankruptcy estimation benchmark for the business sector in Croatia.