Comparação do desempenho de Modelos Lineares Generalizados (MLG) e Modelos Aditivos Generalizados (MAG) na predição de dados financeiros em credit score
This study aimed to present and compare the performance of two different methodologies for statistical modeling of financial data with dichotomous response, specifically exemplified by models of credit score as well as methodologies for validation and performance analysis of these models. One of the me...
| Autor: | |
|---|---|
| Tipo de documento: | dissertação |
| Estado: | Versão publicada |
| Data de publicação: | 2010 |
| País: | Brasil |
| Recursos: | Universidade Federal de São Carlos (UFSCAR) |
| Repositório: | Repositório Institucional da UFSCAR |
| Idioma: | português |
| OAI Identifier: | oai:repositorio.ufscar.br:20.500.14289/10158 |
| Acesso em linha: | https://repositorio.ufscar.br/handle/20.500.14289/10158 |
| Access Level: | Acceso aberto |
| Palavra-chave: | Dados binários Modelos lineares generalizados Modelos aditivos generalizados Lift Credit Score CIENCIAS EXATAS E DA TERRA::PROBABILIDADE E ESTATISTICA::ESTATISTICA::REGRESSAO E CORRELACAO |
| Resumo: | This study aimed to present and compare the performance of two different methodologies for statistical modeling of financial data with dichotomous response, specifically exemplified by models of credit score as well as methodologies for validation and performance analysis of these models. One of the measures used in this analysis is the lift, often used in marketing, but little used in the financial area, this measure is also used as a descriptive technique for categorizing variables. The techniques presented here are the Generalized Linear Models (GLM), the most usual method, and Generalized Additive Models (GAM), unusual in finance because it is a semi-parametric or nonparametric model, generating even some difficulty in interpretation because it does not present parameters. The predictive capabilities of the two techniques are compared in an application on real data and in a simulation study. |
|---|