Geopolitical conflict, global uncertainty and equity market resilience in the MENA region: evidence from a quantile machine-learning framework

Purpose This study aims to examines how global uncertainty and domestic geopolitical conflict affect equity market resilience in six Middle East and North Africa (MENA) economies across stress, normal and expansion regimes. Design/methodology/approach Using monthly MSCI country-index series with country-specific starting dates between 2010 and 2016 and observations through 2025, the study constructs a Geopolitical Fatal Conflict Index (GFCI) from ACLED violent events and fatalities. The GFCI is combined with the Geopolitical Risk Index (GPR), the Financial Stress Index (FSI) and the Volatility Index (VIX) in a Quantile Gradient Boosting (QGB) framework. SHAP values are used to compare the relative contribution and direction of these predictors at Q10, Q50 and Q90. Findings The results are heterogeneous across countries and quantiles rather than following a single regional pattern. At Q10, global indicators lead the SHAP ranking in Egypt, Morocco and Tunisia, whereas GFCI is the leading predictor in Jordan, Palestine and Saudi Arabia. At Q50 and Q90, GFCI remains or becomes prominent in Egypt, Palestine and Saudi Arabia, while VIX or FSI remain important in Jordan, Morocco and Tunisia. In the reported Q50 robustness comparison, including GFCI reduces MAE and RMSE and increases R² in all six markets. Practical implications The findings support country- and quantile-specific monitoring that combines global uncertainty indicators with country-level violent-conflict data. The results are intended for equity-market risk assessment and scenario monitoring rather than causal policy evaluation. Originality/value This study introduces the Geopolitical Fatal Conflict Index (GFCI) as a country-level measure of violent conflict intensity within a quantile machine-learning framework. It distinguishes domestic conflict dynamics from global geopolitical and financial uncertainty and shows that their relative predictive importance varies across countries and conditional quantiles.