SciELO - Scientific Electronic Library Online

 
vol.16 issue3Forecasting a call center demand using a Multiple Regression modelStudies on non-random patterns in Shewhart control charts author indexsubject indexarticles search
Home Pagealphabetic serial listing  

Services on Demand

Journal

Article

Indicators

Related links

Share


Gestão & Produção

Print version ISSN 0104-530XOn-line version ISSN 1806-9649

Abstract

SELAU, Lisiane Priscila Roldão  and  RIBEIRO, José Luis Duarte. Methodology for the construction and choice of credit risk prediction models. Gest. Prod. [online]. 2009, vol.16, n.3, pp.398-413. ISSN 1806-9649.  https://doi.org/10.1590/S0104-530X2009000300007.

Due to the growing consumer credit market and, therefore, insolvency indices, companies are seeking to improve their credit analysis by incorporating objective judgments. Multivariate techniques have been used to construct credit models. These models, based on consumer registration information, allow the identification of behavior standards concerning insolvency. The objective of this work is to propose a methodology for the construction of credit risk models and to evaluate prediction performance using three specific models: discriminant analysis, logistic regression, and neural networks. The proposed method (entitled PRC Model) embraces six steps: (i) population definition, (ii) sampling, (iii) preliminary analysis, (iv) model development, (v) model selection, and (vi) implementation steps. The PRC Model was applied to a sample of 17,005 customers of an organization which manages its own credit system and controls a pool of drugstores. The results for this specific database show slight superiority of neural networks over the other two techniques, which can be attributed to its non-linear approach when dealing with the combined effect of explanatory variables

Keywords : Credit analysis; Discriminat analysis; Logistic regression; Neural networks.

        · abstract in Portuguese     · text in Portuguese     · Portuguese ( pdf )

 

Creative Commons License All the contents of this journal, except where otherwise noted, is licensed under a Creative Commons Attribution License