Journal of
Economics and International Finance

  • Abbreviation: J. Econ. Int. Finance
  • Language: English
  • ISSN: 2006-9812
  • DOI: 10.5897/JEIF
  • Start Year: 2009
  • Published Articles: 323

Article in Press

Investigation of Trade-off between Binary logistic regression and linear discriminant Models on Credit risk: Evidence from real life data

Suleman Issa and Adebayo Alabi Mudasiru

  •  Received: 09 August 2018
  •  Accepted: 18 October 2018
Credit scoring is used to determine the risk factors relevant to undertake an obligatory task and wiliness to pay. The purpose of this study is to investigate trade-off between linear discriminant and logistic regression models for credit scoring and risk evaluation in banking system. Various performances indices such as percentage correctly classified (PCC), Sensitivity, Specificity, and misclassification cost (type I error and type II error) were used to measure the accuracy and reliability of the models. The result shows that logistic regression model performs slightly better than linear discriminant model because it has higher percentage correctly classified (PCC), Sensitivity, Specificity and misclassification cost (type I error and Type II error) were used to measure the accuracy and reliability of the models. The result shows that logistic regression model performs slightly better than discriminant model because it has higher percentage correctly classified, high chance of increasing profit (sensitivity) and reducing risk (low type I error). However, it performed badly in terms of Negative prediction value (specificity) and type II error. Furthermore, the empirical evidence indicates that the age of applicant, marital status, length of service and amount request are significantly helpful in scoring credit applicants and evaluation credit risk. Finally, the models can help financial institution most especially FCMB to evaluate credit applicants and risk default as well as serve an early warning system.

Keywords: Credit scoring, Credit risk, Indicators, Logistic regression model and linear discriminant model