Predictive Analytics for Non-performing Loans and Bank Vulnerability During Crises in Philippines

An economic downturn like the COVID-19 pandemic and the Russia-Ukraine war impact economies worldwide and threaten the stability of every country’s financial system, bringing challenges to governments and policymakers. During crises, a country’s financial system experiences slow credit growth, asset deterioration, and increasing defaults, threatening individual banks’ survival, performance, and development. This study examined the stability and vulnerability of the banking sector of the Philippines, a developing country in Asia using statistical and machine learning models. The first result of this study is a Vector Autoregression (VAR) to forecast the Non-Performing Loan (NPL) ratio, considering the multiple time-series data of bank-specific and macroeconomic variables. The VAR model result is useful for individual banks to conduct peer comparisons, regular credit risk management, and stress testing exercises. The second result is a predictive model which was developed as an early warning system to identify vulnerable banks during an economic downturn and determine the indicators of vulnerability. Among the three classification models that were fitted namely logistic regression, decision trees, and support vector machines (SVM), decision trees emerged as the strongest model in predicting vulnerability. The resulting model sheds light on the dynamics of vulnerability indicators during a crisis and provides the Central Bank insights into their policymaking and bank supervision. Likewise, individual banks can draw supplementary information from the results of this study to help them formulate strategies during a crisis.