Group detection in energy commodity markets through manifold-informed Wasserstein barycenter

A novel approach based on unsupervised Machine Learning techniques is proposed to explore the complex interconnections between the dynamics of energy commodity prices, such as oil, gas and electricity prices in the USA, and the dynamics of certain macroeconomic variables that reflect the behavior of the US economy, such as interest rates and the Standard and Poor’s index. This methodology combines the Wasserstein barycenter with Graph Machine Learning and Manifold Learning techniques to identify common stochastic factors that drive the dynamics of energy commodity prices. Our analysis reveals the presence of a well-defined group of energy commodity markets that share similar characteristics. To study common stochastic factors, a Gaussian Mixture Model is fitted to the Wasserstein barycenter of the discovered cluster. The fitting is performed by maximum likelihood using the Expectation–Maximization algorithm with an initialization strategy based on Graph Machine Learning techniques. A fine-tuning of specific factors affecting the dynamics of energy commodity prices is also discussed.