| Abstract [eng] |
95 pages, 8 tables, 9 figures, 7 appendices. The aim of the thesis is to evaluate the accuracy of classical bankruptcy prediction models and to develop extended models adapted for predicting bankruptcy risk in construction sector companies. The thesis consists of three main parts. The first part presents a systematic and comparative analysis of scientific literature, examining the development of bankruptcy prediction models, their limitations, and sector-specific characteristics. Classical statistical models (Altman, Springate, Zmijewski, Liss and others) are discussed. The literature review reveals that in the construction sector traditional models often lose accuracy due to high variability of financial indicators, project-based operations, and macroeconomic fluctuations. The importance of integrating additional financial, non-financial, and macroeconomic variables is also highlighted. The second part presents the research methodology, data sources, and sample selection principles. The empirical analysis is based on financial data of Lithuanian construction sector companies for the period 2020–2024, obtained from Okredo.lt. The accuracy of classical models is evaluated using accuracy, sensitivity, specificity, and AUC indicators. In order to develop extended models, logistic regression analysis is applied, integrating additional financial, non-financial, and macroeconomic variables. The third part presents the empirical results. It was found that the accuracy of classical models in the construction sector is limited, and their ability to distinguish between bankrupt and non-bankrupt companies varies depending on the model structure. In the group of non-bankrupt companies, the Altman and Zmijewski models demonstrated high accuracy; however, their ability to identify bankrupt companies was significantly weaker, revealing limited sensitivity of classical models to early signs of financial distress in a cyclical sector. After developing extended versions of the Altman and Zmijewski models by incorporating additional financial, non-financial, and macroeconomic variables, the accuracy of bankruptcy prediction improved significantly, and AUC values indicated better discriminatory power. However, the accuracy of extended models decreased in the group of non-bankrupt companies, suggesting that additional variables improve the identification of high-risk firms rather than the classification of stable companies. The conclusions confirm that classical bankruptcy prediction models exhibit limited accuracy in the construction sector, particularly in identifying bankrupt firms. It is also confirmed that the integration of additional financial, non-financial, and macroeconomic variables can improve prediction accuracy; however, this improvement is more related to the identification of risky firms than to the classification of stable companies. The limitations of the study are related to the sample size, variability of financial indicators in the construction sector, differences in accounting policies, and the limited significance of macroeconomic variables. Despite these limitations, the study reveals a clear tendency that extended models are more suitable for assessing bankruptcy risk in the construction sector, and that a broader data set and additional variables could further strengthen the findings. |
