| Abstract [eng] |
The main objective of the work is to create a bankruptcy prediction model for the construction sector, which would include macroeconomic variables designed to predict the probability of bankruptcies during the economic downturn (Covid-19). Tasks of the work were focused on the analysis of scientific literature, the analysis of bankruptcy prediction models and macroeconomic variables, the development of a methodology and the development of a bankruptcy prediction model. The construction sector is one of the main branches of the global economy. It employs about 7 percent of the global workforce and accounts for about 13 percent of the global gross domestic product (GDP). The construction industry also faces challenges such as labor shortages and macroeconomic fluctuations, which make the sector particularly vulnerable. The study analyses the scientific literature on insolvency factors. It reviews traditional bankruptcy prediction methods, estimates macroeconomic variables, and develops an empirical model using logistic regression. The data set consists of financial statements and macroeconomic indicators of Lithuanian construction companies. Missing financial information was handled using the MissForest imputation method. The variables of the created model are statistically significant, since p < 0,05. Five statistically significant variables remained in the final created logistic regression equation: the ratio of liabilities to assets, the ratio of sales revenue to assets, the unemployment rate, the consumer price index and exports. The results of the created bankruptcy prediction model show that each variable in the equation has a different impact on the insolvency risk. An increase in the ratio of liabilities to assets increases the insolvency risk of the company, which means that more indebted companies in the construction sector are more likely to go bankrupt. The ratio of sales revenue to assets reduces the probability of bankruptcy, since more efficient use of company assets 76 helps them remain financially stable. After analyzing the resulting bankruptcy prediction equation, it was determined that the directions of the effects of some macroeconomic indicators do not correspond to economic theory. Both the unemployment rate and the consumer price index have a negative impact on insolvency risk. These results may be related to the specific conditions of the Covid-19 period. During this period, factors such as various support for businesses, restrictions on activities or fluctuations in demand may have distorted the relationship between macroeconomic indicators and the probability of a company's insolvency. The resulting export sign in the equation was consistent with economic theory. An increase in exports reduced the risk of insolvency, so exporting companies were more financially stable during the recession. The model has a high discriminatory power, its ROC-AUC value is 0.874. This confirms that in many cases it can correctly distinguish bankrupt and non-bankrupt companies. The results of the calibration and Hosmer-Lemeshaw test show that companies are classified with a small error into bankrupt and non-bankrupt. Using a classification threshold of 0.3, the model can achieve an overall accuracy of 78.7% and successfully identify approximately 85% of actual bankruptcies. The developed model can be used as one of the insolvency warning tools for shareholders and credit institutions to assess the bankruptcy risk of construction sector companies during periods of economic instability. The results obtained in the work can be used for further empirical research and provide assumptions for scientific publications in the field of bankruptcy forecasting. |
