Gestão do risco de crédito com score dinâmico para previsão de inadimplência de Pequenas e Médias Empresas (PME)

Abstract

This study aimed to develop a dynamic credit risk score to predict default using binary logistic regression, serving as a monitoring tool for the active portfolio of a Direct Credit Society focused on financing Small and Medium Enterprises (SME) backed by card receivables. The approach enables continuous tracking of credit risk in already granted operations, supporting strategic management and default mitigation decisions. The analysis was conducted on a dataset of 2,400 active contracts, with explanatory variables categorized as static or behavioral characteristics of the companies. The dependent variable was client status (performing or defaulted), defined by arrears exceeding 90 days. The proposed model demonstrated high explanatory power, with a Nagelkerke pseudo R² of 0.962 and statistical significance (p < 0.001). The most influential predictors of default were the percentage of outstanding balance, deregistered status with the Federal Revenue Service, and a “Critical” rating in the payment alert indicator. Results indicate that combining behavioral and cadastral variables improves predictive accuracy, contributing to more effective credit portfolio management. The proposed methodology is suitable for continuous risk monitoring and dynamic score development, with potential to enhance credit policy frameworks