| Abstract [eng] |
The growing attention to corporate sustainability and tightening regulatory requirements create a need to supplement financial risk assessment with non-financial indicators – environmental, social responsibility, and governance metrics. Although the scientific literature suggests that higher sustainability scores may be associated with better financial performance and lower risk, there is still a lack of studies that systematically evaluate the impact of these factors on financial risk prediction using machine learning methods. The master’s thesis is dedicated to investigating the impact of environmental, social responsibility, and governance indicators on predicting corporate financial risk. The literature review presents the definition of financial risk and its assessment methods, and reviews scientific studies examining the application of machine learning methods in financial risk prediction tasks. The scientific literature review identifies two main problems in this research field: the ambiguity of the definition of financial risk and the quality of sustainability indicator data. For this reason, the thesis examines five different definitions of financial risk and develops machine learning models for each of them. Since the data sample contains few companies experiencing financial difficulties (about 10–15% of all companies), an additional method was applied in this thesis to address the problem of data imbalance. The conducted research into financial risk prediction includes not only companies’ financial data, but also aggregated environmental, social responsibility, and governance scores. The study compares the results of machine learning models using two data samples: one including environmental, social, and governance scores and one excluding them. Regardless of the chosen definition of financial risk (target variable), there is no statistically significant difference between models that include these sustainability indicators and models that contain only company financial data. The predictive accuracy of the machine learning models differs depending on the definition of the target variable; however, among the best models, it always remains higher than 87%. The machine learning model interpretation methods applied in the study showed that market capitalization, net profit, and operating profit before taxes and interest are the main factors determining the model’s predictions. The results obtained from the interpretation methods are significant both at the individual company level and in the broader economic, social, and business management context, as they provide the opportunity to make more informed decisions when assessing corporate financial risk. |
