Abstract:
"Non-Performing Loans (NPLs) represent financial assets where borrowers fail to meet principal or interest payments for an extended period. Prolonged neglect of NPLs can erode a bank’s profitability due to increased loan loss reserves and write-offs, undermine risk management credibility, and alienate stakeholders. On a broader scale, rising NPLs may tighten credit supply, reduce spending, and lead to economic contraction.
This study explores the predictive dynamics of NPLs in licensed commercial banks in Sri Lanka by analyzing their relationship with macroeconomic conditions and bank-specific factors. It pursues three objectives: identifying statistical links between NPLs and relevant variables, comparing findings with global and local studies to identify unique Sri Lankan factors, and developing a predictive model using advanced machine learning techniques.
The research utilizes quarterly data spanning 2013–2023 from 12 licensed commercial banks listed on the Colombo Stock Exchange, supplemented by macroeconomic data from the Central Bank of Sri Lanka and the Department of Census and Statistics. Twelve independent variables were analyzed using 13 regression-based machine learning models, adopting two approaches: one with all variables and another with statistically significant ones. Model performance was evaluated against six key metrics, with linear regression emerging as the most suitable, yielding low error rates and accurate predictions.
The study underscores the importance of predictive insights for bankers, regulators, and policymakers to proactively manage risks and adapt strategies. However, limitations include a relatively small dataset and the lack of domain-specific expertise for optimal model tuning, addressed as future work. This research highlights the potential of predictive modeling to enhance financial risk management, fostering a stable banking system and a resilient economy.
"
