| Abstract [eng] |
In the fields of finance and management, it is crucial to analyse the financial risks that companies face. Various econometric methodologies are applied to assess and mitigate these risks, but advancements in natural language processing and artificial intelligence technologies have opened new opportunities for evaluating financial risks using big data. Furthermore, non-financial information, such as textual or comparative data, is gaining increasing importance alongside financial data due to technological advancements, as it helps improve the accuracy of financial forecasts and consider current market trends. The aim of this study is to evaluate whether non-financial data, together with indicators reflecting reputation, constructed by assessing customer feedback and articles in the press, contribute to assessing the financial position of a company. In the literature analysis section, it is determined that non-financial information is a valuable part of financial analysis, improving the accuracy of forecasting models. Additionally, the sentiment analysis process is reviewed, along with the challenges encountered during its implementation, and the benefits of analysis are identified. In the methodology section, the stages of sentiment analysis based on lexicology and neural networks are outlined: text pre-processing, training process, and sentiment determination. The Zmijewski bankruptcy prediction model is chosen, according to which indicators the financial condition of companies will be evaluated. For financial risk forecasting, the following categorizations methodologies are selected and described: logistic regression, random forests, neural network models and support vector classifier. After conducting sentiment analyses of various types in Lithuanian texts, it was found that lexical analysis yielded the best results when analysing informal text, while sentiment analysis based on neural networks more accurately determined the sentiment of formal text. During the study, bankruptcy probabilities for the real estate companies were evaluated using a slightly modified Zmijewski method. Based on the values of this indicator, it was determined that the bankruptcy probability of companies, using non-financial data, was best classified by a random forest-based classifier, achieving an accuracy of 86,67 %. Considering the obtained results, it can be concluded that the financial risk of real estate companies can be forecasted using non-financial indicators such as: customer and press sentiment towards the company, number of employees, the company's ability to meet its obligations, and others. |
