Gold, a highly valued and significant investment asset, is subject to various influences including global economic conditions and geopolitical events. Recent advancements in machine learning have shown promising results in predicting financial time series, including gold prices. This study evaluates machine learning algorithms (Linear/Ridge/LASSO Regression, Decision Trees, Random Forest, XGBoost, SVM) for gold price forecasting. A comparative analysis of these algorithms reveals that tree-based machine learning techniques, specifically decision trees, random forest, and XGBoost, outperform other algorithms. Among them, random forest exhibits the highest R2 value (R2 = 0.99) and the lowest values for RMSE (1.38), MSE (1.89), and MAE (0.95). XGBoost and decision trees both achieve an R2 of 0.99 and obtain RMSE values of 1.51 and 1.76, MSE values of 2.28 and 3.09, and MAE values of 1.08 and 1.14, respectively. These findings suggest that tree-based machine learning models may be more suitable for predicting gold prices compared to traditional approaches.
TopIntroduction
For centuries, gold has held significant value as a precious metal. It serves as a crucial financial asset for nations and plays a vital role in the global monetary reserves used for trading and currency hedging (Baser et al., 2023; Wen et al., 2017). Additionally, gold assumes a significant position in investments, particularly as a safeguard against unfavorable financial circumstances. In fact, during periods of economic turmoil when major stock indices decline, the prices of precious metals often exhibit an inverse movement. Furthermore, accurately predicting fluctuations in gold prices holds vital importance for investors, associated companies, and any entity considering gold as an indicator of future performance in the global economy (Feldstein, 2023; Kilimci, 2022; Pattnaik et al., 2022; Plakandaras & Ji, 2022).
The prediction of gold prices is a complex task influenced by various factors that play a crucial role in determining its value. One of the primary factors is the prices of oil, minerals, and other precious metals. As gold is often considered a safe haven asset, its price can be impacted by the trends and fluctuations in these commodities (Hajek & Novotny, 2022; Madziwa et al., 2022). Additionally, inflation rates have a significant influence on the price of gold. When inflation rises, investors tend to seek the stability of gold as a hedge against the devaluation of fiat currencies. Furthermore, exchange rates also contribute to the gold price prediction. As the value of currencies fluctuates, it affects the purchasing power of individuals and institutions, consequently impacting the demand and price of gold. Considering these multifaceted factors, analysts and investors closely monitor and analyze economic indicators to make informed predictions about the future movements of gold prices (Boongasame, 2023; Pierdzioch et al., 2015).
Three primary paradigms are introduced for gold price prediction. The first paradigm employs traditional statistical methods, such as Autoregressive Integrated Moving Average (ARIMA) and multi-linear regression models, to analyze the linear relationships among various factors that influence gold price forecasting (Parisi et al., 2008; Pesaran & Smith, 2019). However, since gold price prediction is influenced by numerous variables with complex non-linear characteristics, the second paradigm incorporates intelligent artificial systems like neural networks and other machine learning models to tackle these challenges (Dhokane & Sharma, 2023; Kilimci, 2022; Vidal & Kristjanpoller, 2020). Finally, the third paradigm involves combining different hybrid models including mathematical, frequency decomposition, optimization algorithms with deep learning to predict gold price fluctuations efficiently and accurately (Alameer et al., 2019; Samee et al., 2022; Shah et al., 2022).