Machine learning predictions of China commodity price indices

Source abstract

Regulators and investors have always placed a high premium on commodity price forecasting. This study examines the weekly price forecast issue for the China commodities price index for the period from June 2 2006 to 17 January 2020. This important commodity price indicator’s forecasting has not received enough attention in the literature. We use cross-validation and Bayesian optimizations during model training, and our analysis is supported by Gaussian process regressions. With an out-of-sample relative root mean square error of 0.1334%, the created models correctly forecasted the price index between 6 January 2017 and 17 January 2020. The generated models can be used by policymakers and investors for policy analysis and decision-making. The forecasting findings might be helpful in creating similar commodity price indices based on reference data on the price trends projected by the models.