Forecasting the Chinese energy security index price: A Gaussian process regression-based machine learning framework enhanced by Bayesian optimisation and cross-validation

Abstract

Accurate forecasting of energy security index pricing represents a significant challenge for policymakers and financial stakeholders. This study evaluates the applicability of a Gaussian process regression framework, optimised through Bayesian techniques and cross-validation, to address this forecasting task. The analysis employs historical data comprising daily closing values of the energy security index listed on the Shanghai Stock Exchange, spanning 4 January 2016 to 31 December 2020. A methodical exploration of kernel configurations and basis functions was conducted to establish a robust predictive framework. The optimised model demonstrated strong performance, achieving a relative root mean square error of 1.4884% during the out-of-sample evaluation period (2 January 2020 to 31 December 2020). Empirical evidence suggests that machine learning technologies demonstrate substantial potential in enhancing the precision of energy market forecasts. The results can complement existing forecasting methods to inform analyses of market trends and policy development, or serve independently as advanced technical projections.