Assessment of Credit Scoring Models Performance in Cameroon: A Contextual Empirical Analysis

Abstract

This paper analyses the performance of default risk prediction models in Cameroon. To do this, we first identify the predictive variables specific to the context. We then test commonly used scoring models on a sample of 448,364 credit files granted between 2003 and 2024. The results show that contextual variables such as job stability have a negative influence on the probability of default for borrowers working in the public sector, and a positive influence on that of employees in the formal and informal private sectors, as well as self-employed workers. In addition, all the models tested perform better in the presence of contextual variables. However, the logit model has better predictive power on the total number of correctly predictedcases. Indeed, it has an average correct classification rate of 90.80%, which is 4.83, 5.68 and 5.85percentage points above the probit, the NN and the SVM models respectively. On the other hand, the probit model performs better than the logit model in terms of ROC-AUC and F-score metrics. Finally, the NN and SVM models exhibit better ROC-AUC, sensitivity, and type II error rates if compared to those of the logit and probit models; yet, their performances are significantly lower than those of both the logit and probit models as shown by the terms of specificity, type I error, and F-score metrics.