FORECASTING THE NEW ENERGY INDEX TRADING AMOUNT IN CHINA MAINLAND: A MACHINE LEARNING FRAMEWORK EMPLOYING GAUSSIAN PROCESS REGRESSION TUNED WITH BAYESIAN OPTIMIZATION
Bingzi Jin, Xiaojie Xu
Abstract
The precise forecasting of trading amounts in energy indices persists as a significant concern for financial stakeholders and oversight institutions. This study fills a notable void in prior research by concentrating on predicting daily trading amounts for China’s new energy index from 2016 to 2020 — a crucial economic metric historically underexplored in the literature. The forecasting approach integrates Gaussian process regression (GPR) methodologies, with model optimization advanced through 10-fold cross-validation procedures and Bayesian parameter tuning. Experimental results validate the framework’s efficacy, attaining an out-of-sample relative root mean square error (RRMSE) value of 16.8197% during the 2020 evaluation phase, consistent with recognized precision standards in economic forecasting. These analytical instruments yield actionable insights for developing investment strategies and crafting regulatory measures, facilitating evidence-based decision-making frameworks. Additionally, the proposed analytical approach exhibits adaptability potential, offering transferable principles for the development and assessment of comparable energy benchmarks in global financial ecosystems.