Forecasting gold prices: a comparative analysis of traditional and deep learning methods

Gold prices significantly influence financial activities at both global and microeconomic levels. Forecasting gold prices is particularly challenging due to the diverse and unpredictable influence of political, economic and financial factors. This complexity has made the development of reliable prediction models for future gold prices a persistent challenge over decades. In this study, we develop three models for gold price forecasting, utilizing a dataset spanning the period from January 2014 to October 2024. The selected methods represent the main categories of forecasting algorithms: Linear Regression (LR) from classical statistics, Multilayer Perceptron (MLP) from machine learning and Long Short-Term Memory (LSTM) networks from deep learning techniques. Through comparative analysis, we evaluate and assess the constructed models based on their ability to forecast gold price movements using a variety of performance measures. The results reveal the strengths and limitations of each method in capturing the dynamics of gold price behaviour within current uncertain environment. These findings can assist investors and financial experts in making informed decisions and establish a foundation for future research focused on improving forecasting accuracy in financial time series analysis.