<![CDATA[People commonly invest in gold to protect their assets because of its role as a stable, safe-haven asset against economic volatility. However, the uncertainty of future gold price movements creates challenges, making the ability to predict gold prices beneficial for analysis. This study proposes a comparative approach to forecasting PT Antam's gold prices, evaluating the performance of three architectures: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Hybrid LSTM-GRU. The study uses historical daily price data from 2010 to early 2025, chronologically split into 80% training and 20% testing sets. Data is processed through a 60-day sliding window framework, and recursive multi-step forecasting is applied for a 30-day horizon. Two optimization techniques, Adam and Stochastic Gradient Descent (SGD), are evaluated. The experimental results show that the Hybrid LSTM-GRU model with Adam optimization achieved the best performance among the evaluated configurations, evidenced by an R-Squared (R²) value of 0.9972, a Mean Square Error (MSE) of 9.553542e+07, and a Root Mean Square Error (RMSE) of 9,774.22. An RMSE of 9,774.22 indicates that the standard deviation of the prediction errors is approximately IDR 9,774. When contextualized with the latest actual gold price, this yields an error ratio of 0.64%. The 30-day projection indicates a downward trend in prices until the end of January 2025. However, due to the univariate nature of the model, it should be viewed as a supplementary analytical tool rather than a sole basis for financial decision-making.]]>
