Comparison of Forecasting Methods for Gold Prices under the Russia-Ukraine War

Abstract

In the context of the Russia-Ukraine war that erupted in February 2022, global gold prices have shown an upward trend, driven by gold accumulation policies adopted by certain countries to ensure economic security and increased investor demand for safe-haven assets. This study aims to evaluate the forecasting performance of daily gold prices (XAU/USD) by comparing three forecasting approaches, namely Holt's Exponential Smoothing, the ARIMA method following the Box-Jenkins framework, and the XGBoost machine learning algorithm applied with a Direct multi-step-ahead forecasting strategy to construct gold price forecasting models. The study utilizes 773 daily gold price observations, with the first 80% (618 days) used for model development and training, and the remaining 20% (155 days) reserved for performance evaluation and comparison. The results indicate that XGBoost achieved the highest performance across all three evaluation metrics, with a MAD of 132.55, RMSE of 156.49, and MAPE of 4.96%. The findings suggest that while time series methods are capable of capturing overall trend patterns, they are limited in detecting short-term volatility due to inherent model constraints. In contrast, XGBoost combined with the Direct multi-step-ahead forecasting strategy demonstrates superior ability to handle nonlinear relationships within the data, making it more suitable for forecasting the prices of highly volatile assets during geopolitical crises.