Financial Uncertainty and Bank Performance in Pakistan’s Dual Banking System: Evidence from System GMM and Machine Learning Models

Abstract

Purpose: This article examines how financial uncertainty affects the profitability and stability of banks in Pakistan’s dual banking system. It focuses on oil price volatility, exchange-rate volatility and gold-price/market uncertainty, and compares Islamic and conventional banks through an integrated econometric and machine-learning design.

Design/methodology/approach: The study uses annual bank-level panel data for 21 Pakistani banks over 2010-2024. Bank performance is measured using Return on Assets (ROA), Return on Equity (ROE) and Z-Score. The empirical design combines one-step and two-step System GMM for dynamic explanatory inference with linear regression, Ridge, Lasso, Random Forest, XGBoost, SVR, ANN, LSTM and GRU models for predictive comparison and variable-ranking evidence.

Findings: The empirical results indicate that Islamic banks show significantly higher ROA, lower NPLs and stronger liquidity than conventional banks, while differences in ROE and Z-Score are not statistically decisive. System GMM confirms persistence in ROA and Z-Score and identifies credit risk as the most damaging internal mechanism. Linear, Ridge, Lasso, Random Forest and XGBoost results consistently show that NPL, CIR, CAR, liquidity and bank size dominate oil, exchange-rate and gold-price variables in explaining and predicting bank outcomes. Recurrent deep-learning models provide limited additional predictive value because of the small annual panel and weak sequential depth.

Originality/value: The study contributes an integrated explanatory-predictive framework for Pakistan’s dual banking system by combining dynamic panel estimation with machine-learning and deep-learning evidence. The main originality is not the use of algorithms alone, but the empirical triangulation showing that external financial uncertainty matters through banking resilience channels, while credit risk, operating efficiency, capital adequacy, liquidity and bank size dominate the predictive structure of bank performance.

Keywords: Financial Uncertainty; Oil Price Volatility; Islamic Banking; Conventional Banking; Bank Performance; System GMM; Machine Learning; Deep Learning; Pakistan.