Forecasting gold price using artificial neural networks: a comparative study with linear regression and strategy performance analysis

This study investigates the application of Artificial Neural Networks (ANNs) and Linear Regression models for short-term gold price forecasting. Using technical indicators such as the 7-day and 30-day moving averages (MA7 and MA30), along with daily returns, both models were trained to predict next-day gold prices. The ANN architecture was developed through a trial-and-error process, while the regression model served as a benchmark. Model performance was evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE), along with a rule-based trading strategy that incorporated predicted returns. The ANN model had a higher RMSE (305.65) than the Linear Regression model (30.82) and underperformed in trading, resulting in a cumulative loss of –52.91%. In contrast, the regression-based strategy delivered a profit of 150.56%, based on the total net return from the same trading rules.