Algorithmic-Statistical Model Under a Computational Scheme for Forecasting Insolvency Under Financial Risk

This research focuses on the analysis of insolvency risk in credit unions in Ecuador using a predictive approach based on automatic learning. Based on international and regional precedents on the determinants of financial solvency, the study aims to develop a model capable of classifying credit unions as solvent or not solvent, according to a return on eq-uity (ROE) threshold of 5 %. The methodology adopted was quantitative, explanatory and predictive, using the Random Forest algorithm on a structured database with coded finan-cial variables. The slope variable, called ROE - LOGIC, classifies as “not solvent” (1) those observations with ROE lower than 5 %, and as “solvent” (0) the rest. The model was trained with 80 % of the data and validated with the remaining 20 %. The results show excellent performance, with precision, recall and F1-score metrics above 0.88 in the test set, and an AUC of 0.95, indicating a high discriminative power. The most influential variables were the net interest margin on promised assets, operating expenses relative to net interest margin and the proportion of performing assets. These were used to construct an interpretive formula that estimates the probability of insolvency