Statistical arbitrage in commodity markets through PCA and OPTICS clustering

Abstract

This thesis explored the application of statistical arbitrage strategies on commodity related assets. The asset universe consisted of a diversified basket of 55 assets spanning three asset classes: commodity futures, commodity-linked equities, and commodity currencies. Two strategies were employed: a traditional PCA-based approach and a method that additionally involved clustering the assets using OPTICS. Over the period from 2014 to 2024, both strategies generated slight yet consistent returns. Notably, the strategy incorporating OPTICS clustering outperformed, both in absolute returns and also risk adjusted performance, suggesting that the inclusion of a clustering step may provide additional benefits in such strategies. Moreover, when tested on a post COVID-19 period, the PCA approach failed to generate returns, while the OPTICS strategy remained slightly profitable. Additional results are presented on the characteristics of the residual parametrisation, as well as a insights into which asset clusters and sectors performed the best. Any returns attributable to both strategies proved to be uncorrelated both with a broad based commodity index and also the S&P500.