Geopolitical risks and oil production in the Middle East and Africa: connectedness and predictability using machine learning ©

Abstract

Given the importance of oil production from the Middle East and African countries to the global oil market and the incessant tensions in the region in recent years, this study empirically examines the connectedness and the predictability of the Middle East and African geopolitical risks (GPR) on oil production. By utilizing the Quantile Connectedness framework, the machine learning XGBoost model, and models with natural inspired optimization algorithms, we show that oil production in the Middle East and Africa is significantly sensitive to geopolitical risks, especially in extreme conditions. From the Quantile Connectedness analysis, oil production always exhibits a net shock receiver, suggesting the potential predictability of GPR indices for oil production. We find that the XGBoost and the XGBoost with natural inspired optimization algorithms are capable of predicting both total oil production and country-level oil production based on the geopolitical risks, while PSO-XGBoost is the most optimal prediction model. Our study reveals that Saudi Arabia generates the optimal results for total oil production, demonstrating the impact of Saudi Arabia on oil production in the region. However, Saudi Arabia has the lowest predicting performance among the six countries when examining the predictability of geopolitical risks in country-level oil production.