Bibliographic citations
Valdivia, M., (2019). Comparación del pronóstico de riesgo de crédito utilizando regresión binaria asimétrica cloglog y perceptrón multicapa [Tesis, Universidad Nacional Agraria La Molina]. https://hdl.handle.net/20.500.12996/4206
Valdivia, M., Comparación del pronóstico de riesgo de crédito utilizando regresión binaria asimétrica cloglog y perceptrón multicapa [Tesis]. : Universidad Nacional Agraria La Molina; 2019. https://hdl.handle.net/20.500.12996/4206
@mastersthesis{renati/247164,
title = "Comparación del pronóstico de riesgo de crédito utilizando regresión binaria asimétrica cloglog y perceptrón multicapa",
author = "Valdivia Carbajal, Manuel",
publisher = "Universidad Nacional Agraria La Molina",
year = "2019"
}
This thesis takes as a case of study a recognized cosmetics company from the City of Lima in Peru. To predict the credit risk, two models will be analyzed: The Cloglog Asymmetric Binary Regression and the Perceptron Multilayer Artificial Neural Networks. The selection of these models arises from recent studies that reveal the advantages of artificial intelligence techniques over statistical models in terms of prediction due to their high ability to discern patterns. The company has a business model called Red Binaria, which means that they hire sellers and they offer products to their clients through catalogs. Due to the lack of information from the final customers, the probability of nonpayment was measured through the vendors. The population studied was made up of the company's salespeople, who handled a client portfolio of 51,183 people as of July 2017. The data were previously treated considering the analysis of atypical values at the univariate and multivariate level, the latter using the K-means segmentation algorithm. Once this was done to classify sellers into good and bad payers, a Perceptron Multilayer Artificial Neural Networks model was used with a single intermediate layer and a Binary regression model on which the asymmetric link Cloglog was chosen due to the nature of the data. The results showed a 0.846 and 0.809 ROC index in the training samples, and a 0.762 and 0.733 ROC index in the test samples respectively for each model. Finally, it is concluded that the application of the Percptron Multilayer Neural Networks technique defines a better discrimination rule than the Cloglog Asymmetric Binary Regression in the probability of default study. In addition, Neural Networks present better prognostic indicators. For future research it is recommended to build new variables, because these could have a better predictive capacity.
This item is licensed under a Creative Commons License