AN ENSEMBLE MODEL FOR PREDICTION OF CRISIS IN SLOVAK COMPANIES

Background: Globalization has increased not only the speed and importance of cooperation, but thanks to international relations between countries as well as between businesses, competition has also increased. Increased competition leads to increased probability of company failure. Early recognition of the threat of bankruptcy is therefore even more important. In 2016 in Slovakia entered into force the new provisions of Law No. 513/1991 Coll. Commercial Code on companies in crisis. The company is in crisis when it is in default or at risk of imminent default. Objectives: The goal is to create and evaluate an ensemble model that predict whether a Slovak company will find itself in a crisis in the following year. Paper also provide an overview of bagging, boosting, and stacking, arguably the most used ensemble methods. Methods and data: The ensemble model is created and tested on a sample of Slovak companies. The data used to create the model are from financial statements 2014 (47 414 companies) and model is tested on data from financial statements 2015 (64 757 companies) and 2016 (56 743 companies). The data was obtained from the register of the financial statements (www.registeruz.sk). To evaluate of the model, we use standard performance metrics (AUC, confusion matrix, RMSE, logloss, ...). Results: Using ensemble learning, we have created a model to predict whether a company will be in a crisis. Some of the model performance metrics: AUC = 0.89, RMSE = 0.28, Mean per class error = 0.20.