Geopolitical risks and crude oil futures volatility: Evidence from machine learning

Abstract

This paper conducts a dynamic analysis of the forecasting impact of categorical geopolitical risks on crude oil futures volatility, employing the Transformer-based neural network. Empirical results indicate geopolitical risk linked to war and terrorism consistently exerts the most significant impact across all forecast horizons. Our investigation further reveals that the impact of different subcategories of geopolitical risk on crude oil futures volatility exhibits noteworthy time-varying characteristics. Furthermore, the predictive impact of geopolitical risk on crude oil futures volatility exhibits asymmetry across distinct economic states. In short-term forecasts, the incremental predictive information derived from geopolitical risks primarily concentrated in the economic expansion, gradually transitioning towards economic recession as the forecast horizon extends. More importantly, our research emphasizes that the predictive information derived from geopolitical risks enhances the precision of crude oil futures volatility forecasts and delivers significant economic benefits to investors by integrating valuable information into their portfolio strategies.