A0242
Title: Bayesian neural networks applied to credit risk
Authors: Maria Rosa Nieto Delfin - Investigaciones y Estudios Superiores, S.C (Mexico) [presenting]
Luis Javier Espinosa Rios - Universidad Anahuac Mexico (Mexico)
Abstract: The improvement of information technology has driven the development of the financial system. Financial institutions implement high-performance methodologies that contribute to risk analysis and management. One such method is the use of machine learning algorithms. The aim is to explore credit risk supervised learning algorithms in three databases: 1) credit cards issued by a commercial bank, 2) mortgage loans, and 3) LGD of financial assets. For this, the Bayesian neural networks (BNN) algorithm evokes to train and estimate the probability of default. The Markov Chain Monte Carlo (MCMC) method evaluates posterior probabilities to strengthen the algorithm's operation. In addition, the performance of the BNNs is studied under different prior distributions. Given the importance of the activation functions, different functions are used for the hidden and final layers. The results obtained allow the conclusion that, for classifying and estimating the probability of default, the BNNs give a robust confidence interval and allow credits to be classified correctly, making their configuration ideal and replicable. Machine learning is a strong field of research, where its application to complex tasks is plausible given computational progress and the mathematical assumptions of the model.
