Geopolitical Shocks and Commodity Markets

This project investigates whether geopolitical events have a measurable and predictable effect on WTI crude oil prices, and whether machine learning models can detect and quantify that effect. Oil is the world's most traded commodity and its price affects everything from airline tickets to heating bills, making accurate price signal detection genuinely valuable for investors, energy companies, and policymakers alike.

Three key findings emerge from the analysis. First, machine learning models can predict WTI price levels with very high accuracy, achieving an R2 of 0.974 on unseen data. However this performance is largely explained by price momentum rather than geopolitical insight: the model is essentially observing that today's oil price tends to be close to yesterday's. Second, when momentum is removed from the analysis by predicting daily price deviations from trend instead of the price level itself, geopolitical variables including conflict event counts across the Middle East and North Africa and country-level risk scores for Russia, Israel, and Saudi Arabia emerge as genuine contributors, explaining approximately 10 percent of daily price surprises. Third, this geopolitical signal is not constant: it appears most strongly during acute supply-side disruptions such as the COVID crash and the Ukraine invasion, and fades during demand-driven or low-volatility market periods.

The central conclusion is that geopolitical risk has a real but conditional effect on short-term oil prices. The 10 percent explained by Part 2 is the floor of what a better-specified model could achieve. With more granular conflict data, satellite monitoring of energy infrastructure, and options market volatility as a real-time signal, this approach could be developed into a practical commodity risk tool. This project is a proof of concept for a longer research programme.