Forecasting Pakistan Stock Returns Using Economic Policy Uncertainty: Deep Learning Evidence from an Emerging Market

Abstract

Accurate forecasting of stock returns remains a persistent challenge in financial economics, particularly in emerging markets characterized by policy uncertainty, macroeconomic volatility, and evolving institutional structures. Pakistan offers a relevant setting due to recurring political transitions, inflationary pressures, exchange rate instability, external financing constraints, and periodic policy shifts that frequently influence investor sentiment and market behavior. This study examines whether Economic Policy Uncertainty (EPU) improves the forecasting of Pakistan stock returns and whether advanced deep learning models outperform conventional econometric approaches. Using monthly data from 2010 to 2024, traditional benchmark models, including ARIMA, GARCH, and ARDL, are compared with modern deep learning architectures comprising Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), CNN-LSTM, and Transformer models. Forecasting performance is evaluated using root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The empirical findings show that incorporating EPU significantly enhances predictive accuracy across competing models, confirming the informational relevance of policy uncertainty for financial markets. Deep learning models consistently outperform benchmark econometric models, indicating the presence of nonlinear, dynamic, and time-varying relationships in Pakistan’s stock market. Among all competing approaches, Transformer and CNN-LSTM models deliver the strongest forecasting performance, with gains becoming more pronounced during crisis periods, including COVID-19-related disruptions and episodes of macroeconomic stress. The study contributes new evidence from an underexplored emerging market and demonstrates the value of integrating uncertainty indicators with modern artificial intelligence techniques for financial forecasting. The findings provide practical implications for investors, regulators, and policymakers seeking to strengthen decision-making, risk management, and market monitoring under uncertain economic conditions.