Investigating the accuracy of credit bureau data in predicting borrowers' repayment of consumer loans in Nigeria

Consumer lending, especially microloans, is a critical component of financial inclusion and economic growth. However, the increasing reliance on credit bureau data to assess borrowers' repayment capabilities raises questions about its predictive accuracy. In this study, we evaluated the effectiveness of credit bureau data in predicting loan repayment behavior using a dataset from First Central, a Nigerian credit bureau, and Irorun, a digital lending institution in Nigeria. A total of 3,741 loan applicants, aged 21 to 60, were analyzed using traditional statistical tools and machine learning (ML) models, including logistic regression, random forests, and gradient boosting. Our results indicated that while credit bureau data provides some predictive insights, its standalone accuracy is limited due to inconsistencies in borrower credit histories and incomplete data reporting. Correlation analyses show weak associations between borrower-reported and bureau-reported overdue loan records and repayment outcomes, with Cramer's V values below 0.05. ML models, particularly gradient boosting, outperformed traditional statistical approaches, achieving an AUC-ROC of 0.77, highlighting the potential of advanced algorithms in credit risk assessment. Our findings suggest that integrating alternative borrower data, such as utility bill payments and digital transaction records, could enhance credit risk modeling. The study emphasizes the need for improved credit reporting accuracy and regulatory measures to ensure comprehensive borrower profiles. Enhancing predictive models with supplementary data sources can mitigate default risks and promote responsible lending, leading to greater financial inclusion in Nigeria.