Identifying key features affecting bank liquidity risk: A machine learning regression approach

Abstract

 This study investigates the determinants of liquidity risk by applying advanced machine learning techniques to overcome the limitations of traditional econometric methods. Using data collected from the financial statements of 31 Vietnamese commercial banks over the period 2009-2024, the study implements ten regression algorithms classified into three main groups: regularized linear models (Lasso, Ridge, Elastic Net) compared with standard linear regression; tree-based regression models (Decision Tree, Random Forest); and gradient boosting models (Gradient Boosting, XGBoost, LightGBM, and CatBoost). The empirical results quantify and rank the importance of key financial features affecting liquidity risk, thereby providing further evidence of the nonlinear nature of banking data in Vietnam. These findings contribute to the development of quantitative tools that support bank managers and regulators in designing risk monitoring frameworks and early warning systems for liquidity instability.