A Bayesian Network Model to Evaluate the Credit Risk of Mexican Microfinance Institutions in 2023

Abstract

Assessing credit risk is crucial for financial institutions to make timely and accurate decisions. This research proposes a model for microfinance institutions to estimate credit risk in simplified single-client scenarios. The model is based on a Bayesian Network algorithm and uses a historical database to demonstrate the forecast risk potential. The database includes variables such as age, income, credit history, home ownership, and the final state of the loan, which can be pay, default, or breach. By establishing a relationship between variables through an inferential network and joint probability tables, the research explores three scenarios to obtain risk probability distributions based on different age and income ranges. The inference probability is obtained via a Bayesian network, where the interrelation between variables is structured in a specific topology. We use the causality assumption to estimate the probability of default or credit risk, which is closer to the reality of credit institutions. Therefore, it is a powerful tool for risk analysis agencies to make informed decisions in credit evaluation.