Predicting financial distress in Norway : using logistic regression and random forest models

Abstract

Financial distress can be a highly costly and disruptive event, both on the level of the firm as well as for the society. Models to predict financial distress for this reason have been beneficial. In this thesis, we aim to develop a similar model which is applicable to Norwegian companies. Rather than solely focusing on bankruptcy predictions as previous research has done, we use financial ratios and other related company information, to predict whether firms are likely to enter financial distress within the next two years. Furthermore, we seek to identify early warning signs of financial distress in order for the management to start financial reconstruction in time. A traditional and a more recent algorithm – logistic regression and random forest – were utilized in our analysis for their complementary properties. The models were created based on data provided by the Norwegian School of Economics where we selected a sample of 30 000 companies in the period from 2013 - 2016 after thorough cleaning of data. We find very similar performance for both models where random forest shows slight superiority to logistic regression. Both models yield an AUC of ~ 0.65, and from the results obtained, it indicates that they are able to correctly predict ~ 60% of both healthy and financially distressed companies ahead of time. Moreover, the results indicate that our models assign high importance to some commonly used ratios in the past, such as Size (Log of total assets), ROA, Retained earnings/Total assets, Total debt/Total assets and Debt/Equity. We also find Cash ratio and Net profit margin as important variables, which have been neglected previously. All these variables may contribute as warnings signs of financial distress when making predictions.