Financial Stability and Risk Resilience in Pakistan’s Islamic and Conventional Banks: Evidence from Dynamic Econometrics and AI Forecasting

Abstract

This article looks at the impact of credit, liquidity and solvency risk on profitability based financial stability in Pakistan's dual banking system and compares the findings of dynamic econometric models with AI forecasting models for complementary evidence for risk surveillance. Using a balanced annual panel of 21 Pakistani banks from 2010 to 2024, including 18 conventional and three fully fledged Islamic banks, the study uses return on assets (ROA) as the stability signal. The non-performing loans (NPL) and the ratio of liquid assets to deposits (LR) and that of liabilities to assets (SER) measure the credit, liquidity and solvency risks respectively, while GDP growth and inflation capture the macroeconomic environment. Analysis includes descriptive statistics, correlations, lag-order selection, Johansen cointegration, Panel VECM, Difference GMM and AI forecasting models (Random Forest, MLP, Hybrid RF-MLP, LSTM and GRU). The analysis reveals that the average ROA, NPLs and ROA volatility of Islamic banks are better than those of conventional banks, in addition to having a better solvency position. The most constant risk channel is credit risk. The findings of VECM support that the profitability indicators are in long run equilibrium, and the findings of GMM show the persistence and the solvency related effects with Islamic banks. The results indicate that GRU works best with conventional banks, whereas LSTM and Hybrid RF-MLP are more effective with Islamic banks using error measures. The results are relevant for risk surveillance of banks.

Keywords: Islamic banking; Conventional banking; Financial stability; Credit risk; Liquidity risk; Solvency risk; Panel VECM; AI forecasting.