Credit risk management: comparison of logit and probit models under classical and Bayesian approaches

• Autor: Fernández Vásquez, Richard Fernando
• Formato: artículo
• Fecha de Publicación: 2026
• Institución: Universidad Nacional de Ingeniería
• Repositorio: Revistas - Universidad Nacional de Ingeniería
• Lenguaje: español; inglés
• OAI Identifier: oai:oai:revistas.uni.edu.pe:article/2738
• Enlace del recurso: https://revistas.uni.edu.pe/index.php/iecos/article/view/2738
• Nivel de acceso: acceso abierto
• Materia: incumplimiento de crédito; credit scoring; regresión logit; regresión probit; enfoque clásico; enfoque bayesiano; MCMC; credit default; logit regression; probit regression; classical approach; bayesian approach

In the field of credit risk management, credit scoring is a fundamental statistical tool that allows financial institutions to improve their decisions on whether to approve or reject loans, adjusting their policies to the risk profile of each applicant. In response to this context, the overall objective of this research was to compare the logit and probit statistical regression models under the classical and Bayesian approaches, with the aim of developing a credit scoring model applied to personal loan applicants, aimed at segmenting them and proposing actions according to their level of risk. The research took a quantitative approach; the method used was hypothetical-deductive, with a non-experimental, cross-sectional, descriptive, and correlational design. The population consisted of 5,584 personal loan applicants, both compliant and non-compliant, whose data were obtained from the DataCamp platform as part of an academic initiative aimed at credit risk analysis. For the development of the study, a training sample and a validation sample were established, corresponding to 70% and 30% of the data, respectively. The logit and probit statistical models estimated under the classical and Bayesian approaches were compared. The results showed that, when considering the performance indicators—accuracy, F1-score, area under the ROC curve, and Gini index—the logit regression model under the Bayesian approach performed best, with values of 0.6201, 0.6310, 65.6286, and 31.2573, respectively.