Improved prediction of global gold prices: An innovative Hurst-reconfiguration-based machine learning approach

Abstract

With gold being one of the most important precious metals that play irreplaceable roles in the global market, understanding the future movement of the gold price is of significant importance for investment and risk management worldwide. However, gold prices are subject to volatility and can experience significant fluctuations over time due to economic uncertainty shocks such as the China-US Trade War, Russia-Ukraine war, and COVID-19, which make the forecasting of gold price a challenging task. In this paper, we propose a hybrid forecasting model for gold prices based on the Hurst-oriented reconfiguration and machine learning approach and illustrate its usefulness by analyzing the gold prices of three major markets. We conduct a multifractal analysis of the decomposed series and scrutinize the predictability of each sub-series and its relationship with the Hurst exponent. Empirical results show that there are negative relationships between forecasting error and the Hurst exponent and between the number of embedding dimensions and the Hurst exponent. Our Hurst-based hybrid model outperforms other conventional prediction models in terms of prediction errors and accuracy of direction prediction. The findings of this study shed light on a better understanding of the temporal features of the gold market and provide references for improving investment and hedging strategies.