Asian Journal of Advanced Research and Reports

The advent of digital credit has made much-needed funds available to a population traditionally regarded as “unbankable”. Both the use of, and demand for, these loans have increased over the past three years and continue to grow. Digital loans are characterised by distinctive features, including rapid disbursement upon request, proxy borrowing, the absence of collateral, inadequate customer identity verification in accordance with business requirements, and unclear purposes for the borrowed funds. These factors, together with the limitations of static conventional risk models, necessitate the adoption of dynamic approaches capable of capturing the temporal dimension of credit behaviour. This study aimed to model the credit risk of digital loans in commercial banks using the Cox proportional hazards regression model. The model permits the analysis of time to default while simultaneously accounting for the influence of multiple covariates. The study employed a simulated dataset comprising 6,000 digital borrowers to estimate the direction and magnitude of the effects of borrower characteristics and to develop a digital credit risk-scoring model. Synthetic data were used because access to real-world loan records was restricted. The Cox proportional hazards model identified two distinct borrower risk profiles. The high-risk profile comprised young, male, and unemployed borrowers with a low income-to-loan ratio and a high debt-to-income ratio, yielding a linear risk score of 3.0753 and a corresponding hazard ratio of 21.656 relative to the baseline profile. Conversely, the low-risk profile comprised older, female, and formally employed borrowers with a high income-to-loan ratio and a low debt-to-income ratio, yielding a linear risk score of -0.2402 and a hazard ratio of 0.7865. The model demonstrated excellent short-term discriminatory performance, with time-dependent AUC values of 0.867, 0.939, and 0.966 at 15, 30, and 45 days, respectively, whereas Harrell’s concordance index (C-index) of 0.665 indicated moderate overall discrimination between borrowers at high and low risk of default. These findings demonstrate the usefulness of survival analysis for dynamic digital credit-risk assessment and borrower-risk stratification. The findings are particularly useful because they enable banks to estimate the likelihood of default over time, including at the point of loan application, thereby strengthening credit-risk assessment. A limitation of the study was its reliance on simulated data, which may not fully capture the complexities of real-world digital-loan settings. Future research should validate the models using real-world borrower data from financial institutions to improve generalisability and assess operational performance under actual lending conditions.