| Abstract [eng] |
The main purpose of the thesis is to model and forecast the resilience of European Union regions to economic shocks, as this topic became very important after global financial crisis in 2008. Researchers and scientists have not agreed which specific microeconomic, macroeconomic indicators or indexes can define or evaluate the resilience of regions so far, as a result eight macroeconomic indicators will be compared to each other to evaluate economic resilience. There are calculations of resilience of macroeconomic indicators in each region, which are used in further analysis, also deeper analysis whether the year of becoming a member to the European Union and population affect the probability of non-recovery after the economic shock. Random forest regression were used for modeling and forecasting, where the dependent variable is the area of regional resilience, also five detection and one survival model, where the dependent variable is a two-class event, whether the region was able to recover from economic shock and reach pre-shock status or not. The results of the models are evaluated by prediction errors (RMSE, MAE, MDAE, MAAPE, SMAPE), and detection and survival analysis were evaluated by AUC area and error rate estimate (EER). In addition, confusion matrices and ROC, DET curves are placed in result section. Analysis showed that the strongest and most vulnerable economic indicators, which should be strengthened to avoid or recover easily from the economic shock, are: the most resistant region is Poland, the least resistant – Ireland of gross domestic product, house index: the most resistant – Luxembourg, least – Denmark, consumer price index: the most resistant – Portugal, the least resistant – Lithuania, government debt: the most resistant – Italy, the least – Spain, unemployment: the most resistant – Germany, the least – Spain, employment: the most resistant – Poland, the least resistant – Spain, export of goods and services: the most resistant – Croatia, the least – Luxembourg, import of goods and services: the most resistant – Netherlands, the least resistant – Lithuania. Modeling and forecasting results showed that two indicators were the most valuable for modeling with random forest regression: the consumer price index, which had the highest result of coeficient of regression evaluation (0.94) compared to other indicators, and the unemployment rate due to the lowest percentage errors. The results of the detection and survival models showed that the detection models are more accurate for all indicators, except housing price index, this indicator achieved the best results with random survival forest. |
