• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.96.2019.tde-09082019-110750
Document
Author
Full name
Luís Felipe Barbosa Fernandes
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
Ribeirão Preto, 2019
Supervisor
Committee
Ribeiro, Evandro Marcos Saidel (President)
Abraham, Kuruvilla Joseph
Albuquerque, Andrei Aparecido de
Prataviera, Gilberto Aparecido
Title in Portuguese
Aplicação de Redes Bayesianas em modelos de classificação de risco de crédito
Keywords in Portuguese
Modelagem de crédito
Pontuação de crédito
Redes Bayesianas
Risco de crédito
Abstract in Portuguese
A demanda pelo estudo e aprimoramento de modelos de crédito que auxiliem na tomada de decisões, relativas a concessão creditícia, cresce de forma acelerada. Frente às dificuldades de ordem financeira que atingem os mais diversos países, incluindo o Brasil, verifica-se uma crescente preocupação dos órgãos reguladores do mercado financeiro, bem como, das próprias instituições credoras que atuam no mercado por modelos de crédito. A dificuldade para a obtenção de informações que reflitam a saúde financeira das empresas - assimetria informacional - aliada à carência de informações no mercado financeiro contribui para o aumento dos casos de default e empresas que decretam concordata. Em face dos problemas e dificuldades apresentados, a pesquisa empregou o método probabilístico de Redes Bayesianas com o objetivo de desenvolver um modelo de crédito que calcule o risco de crédito de uma empresa baseado apenas em um conjunto de indicadores financeiros, obtidos a partir das demonstrações financeiras dessas empresas. Para isso, foi usado um conjunto de demonstrações financeiras, referentes a um total de 852 empresas com faturamento superior à 200 milhões, cedidas pela instituição Serasa Experian. A partir dessas demonstrações foram implementadas as fórmulas usadas pela Serasa Experian para o cálculo de indicadores financeiros, a partir dos quais a Rede Bayesiana inicial foi formada. A técnica de Redes foi implementada através do algoritmo denominado Algoritmo Pc, que combina elementos de grafos probabilísticos e definições de probabilidades condicionais, para a selecionar as variáveis, representadas pelos indicadores financeiros, mais significantes para o cálculo do risco de crédito. Além disso, foi realizada uma comparação da técnica probabilística de Redes Bayesianas com a técnica de Regressão Logística, para verificar qual dos modelos melhor se adequava ao conjunto de dados. Após implementar a técnica, foi desenvolvido também um aplicativo, que calcula o risco de crédito de uma empresa, a partir de um conjunto de 17 indicadores financeiros e exibe ao usuário final, a classe de risco, dentre cada uma das 13 classes possíveis, a que uma empresa possui maior probabilidade de pertencer. Para validar a técnica de Redes Bayesianas foram empregadas duas métricas: a RMSE(Raiz Quadrada do Erro Médio) e o MAE(Erro Absoluto Médio). As métricas mostraram que o modelo de Redes Bayesianas foi pouco preditivo, com resultados aquém do esperado. Os resultados da técnica de Regressão Logística porém, mostram um percentual de acertos muito superior, classificando um percentual de 82% das empresas classificadas como de risco de crédito baixo, de forma correta
Title in English
Application of Bayesian Networks in models of classification of credit risk
Keywords in English
Bayesian networks
Credit modeling
Credit risk
Credit score
Abstract in English
The demand for studies and enhancement of credit models that helps at the decision making, associated with the granting credit, grows in a high speed. In the face of the recent troubles of financial order that accomplish innumerous countries nowadays, including Brasil, financial authorities have shown an increasing concern, as well as, the financial institutions that plays at the market for credit models. The challenge of search for informations that shows the financial health's companies - information asymmetry - together with the lack of data at the financial market contribute to increase the number of default cases and number of companies that fails. Due to the issues and difficulties described, this research used the probabilistic approach of Bayesian Networks to develop a credit model capable of calculate the credit risk of a company based on a set of financial indexes, obtained by the financial statements of these companies. For that, it was used a set of financial statements, regarding a set of 852 companies with revenues higher than 200 hundred billion reais, obtained through an agreement with the institution Serasa Experian. These financial statements were used to calculate the financial indexes through the formulae adopted by Serasa Experian, which gave the inicial set of the Bayesian Network. The Network technique was used through an algorithm called Pc Algorithm, that mix elements of probabilistic graphs with conditional's probability definitions, to select variables, represented by financials indexes, that are more significant to the calculation of credit's risk. Besides that, it was made a comparison between Bayesian Network and Logistic Regression technique, with the purpose of verify which one was the best to this set of variables. After the technique was implemented, it was also developed an application, capable of calculate the credit risk of a corporation, using a dataset of seventeen financial indexes. As a result, the app shows to the final user which of the thirteen risk's classes, has the biggest chance of being associated with the enterprise. To validate the technique it were employed two measurements, the RMSE(root mean square error) and the MAE( mean absolute error). The measurements showed that the Bayesian Networks model was not very predictive to the sample of companies which it was trained, since the outcomes fell short of expectations. On the other hand, the Logistic Regression technique showed better results when compared with the Bayesian Network technique. The percentage of right risk's class classifications were much higher, resulting at a percentage of 82% of companies classified as "low risk" , in the right way
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Publishing Date
2019-10-22
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.