Gold Price Modeling and Prediction in Indonesia Using Gated Recurrent Units (GRU)

Abstract

Abstract

Gold has long been recognized as a safe-haven asset against inflation, economic uncertainty, and geopolitical turmoil. This study aims to predict daily gold prices in Indonesia using the Gated Recurrent Unit (GRU) method, incorporating exchange rate and Bank Indonesia interest rate as exogenous variables. The dataset covers April 2016-December 2024, including PT Antam gold prices, exchange rates, and interest rates. Preprocessing involved handling missing values through interpolation, Min-Max normalization, and seasonal decomposition. The dataset was split into 80% training and 20% testing using a time-series windowing approach. The GRU architecture consisted of two layers with dropout, trained with the Adam optimizer and MSE loss function. Evaluation results show accuracy with RMSE 19,563.49, MAE 9,377.12, MAPE 0.73%, and R² 0.9868, indicating the model explains more than 98% of data variability. GRU effectively captured long-term trends and short-term fluctuations, though less responsive to extreme market shocks. Future work includes adding macroeconomic variables, comparing with other deep learning models, and implementing interactive dashboards for practical decision-making.

Keywords: Exchange Rate, Gold Price Prediction, Gated Recurreent Unit (GRU), Interest Rate, Time Series