Short-Term and Long-Term Forecasting of Global Gold Prices Using LSTM and GRU Models

Abstract

Global gold prices exhibit high volatility and complex temporal patterns, making accurate forecasting a challenging task. This study aims to compare the performance of deep learning models for short-term and long-term gold price prediction using daily historical data. Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) were selected because both models can capture temporal dependencies in financial time-series data, while having different architectural complexities and learning characteristics. Comparing these models is important to identify the most suitable approach for different forecasting horizons. The dataset consists of daily global gold prices denominated in USD obtained from an open financial data source covering the period from 2010 to 2024. The models were evaluated under two forecasting horizons, namely short-term prediction (1 day ahead) and long-term prediction (30 days ahead). Model performance was assessed using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Experimental results indicate that the GRU model outperforms LSTM in short-term forecasting by producing lower prediction errors, while LSTM demonstrates slightly better stability in long-term forecasting. These findings suggest that the effectiveness of deep learning models for gold price prediction is highly dependent on the forecasting horizon.