The relevance of traditional and non-traditional borrower data in predicting default in financial co-operatives

In this paper, we examine the relevance of both traditional and non-traditional data in predicting default in two financial co-operatives (co-ops) in Kenya. Using micro-level secondary data representing 1753 borrower data extracted from the co-op systems of the two sample financial co-ops from June 2018 to July 2019, random panel logistic regressions are performed. The results, which are performed at both disaggregated and aggregated levels for both traditional and non-traditional features, reveal that both sets of features are useful in predicting default in financial co-ops. More specifically, we find that traditional features such as a longer member duration, higher value of deposits, and higher outstanding loan amounts are associated with lower default. In the case of nontraditional features, we find that borrowers drawn from the top 5 centres exhibit higher default rates. The results further show that borrowers who visit co-op offices more often are less likely to default. We further establish that the predictive power of the models improves when both traditional and non-traditional features are incorporated. The results in this study provide useful insights to managers and leaders when seeking operational and loan management systems for co-ops.