Machine learning natural gas price predictions

Abstract

Many market participants have historically placed a great deal of importance on price projections for energy commodities. To tackle the problem, our analysis looks at Henry Hub natural gas prices on a monthly basis. The examined sample covers almost 30 years, from January 1997 to March 2024, and this price series has significant economic implications. In this case, price projections are generated through the use of Gaussian process regression techniques, which are obtained through the application of cross-validation processes and Bayesian optimization approaches. A relative root mean square error of 3.1744% indicates that our empirical prediction method produces reasonably accurate price estimates for the out-of-sample testing phase of November 2018–March 2024. Models for predicting prices give investors and governments the information they need to make wise choices about the natural gas market.

Topics & Concepts