Modelo de scoring para crédito de consumo en una entidad del sector solidario

Abstract

In Colombia, the Superintendence of the Solidarity Economy through the Basic Accounting Circular, Title IV, Chapter II regulates the SARC as the Credit Risk Management System that must be implemented and/or complemented by the supervised solidarity organizations, with the purpose of, identify, measure, control and monitor the credit risk to which they are exposed in the development of their granting process. Having said the above, it is necessary for the entities of the sector that place credits, to have their own models that obtain the most relevant variables and that allow the calculation of the probability of default based on the information of the credits that the entity has and to qualify accordingly. periodically each of the current credits. The main objective of this article is to explain how the most relevant variables can be obtained to calculate the probability of default of a client and create an internal score for new loans in order to support the granting decision through a machine learning model. This article was carried out using a database of a solidarity sector entity that contains a total of 5974 consumer loans and through automatic learning different models were trained to determine the probability of default of a client. The light gradient boosting machine model had the best performance with an AUC of 0.7550, recall of 0.7111, precision 0.1587. In addition, among the available variables, the ones that are most important to infer the probability of default for a consumer loan are age, total assets, total liabilities, monthly income, monthly expenses, credit value, term, level of study, status civil, form of payment and age.