Maskininlärning & Random Forest: Överträffar traditionella kreditmodeller

Sammanfattning: The Altman Z-Score model is one of the most famous models för predicting bankruptcy and measuring financial distress for companies. It uses multivariate discriminant analysis to classify companies in three different groups based on their calculated Z-Score. The purpose of this thesis is to analyze how well the Altman Z-Score model for emerging markets performs, to then attempt to create a new binary classification model using machine learning and random forest-algorithms in order to get a more precise model that better predicts bankruptcy for companies within 2 years. This was done using financial statements from 9520 Polish firms along with their respective bankruptcy status after 2 years. The results show that regardless of how you interpret the Altman model, it was possible to create a random forest-model that outperformed it by all measurements. The random forest models managed to beat the Altman Z-Score models by predicting a slightly higher percentage of both bankruptcies as well as non-bankruptcies, resulting in an overall higher accuracy. This was done by using the same four financial ratios that are used in Altman's model, i.e. no further information was added.