Evaluating the Accuracy of ARIMA vs. Holt’s Double Exponential Smoothing in Predicting Indonesia’s Gold Price Trends

Abstract

This research is focused on evaluating how accurately the Autoregressive Integrated Moving Average (ARIMA) technique and Holt’s Double Exponential Smoothing predict gold prices in Indonesia. The dataset comprises monthly gold price records from PT Aneka Tambang Tbk for the period January 2021 to December 2025, totaling 60 data points. The analytical approach utilizes a time series method that incorporates both ARIMA and Holt’s Double Exponential Smoothing model phases, followed by an assessment of accuracy through the Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). Findings reveal that the ARIMA (1,1,6) model performed best regarding historical accuracy assessment, showing reduced error metrics in comparison to Holt’s approach. Nonetheless, verification using real data from January to April 2026 demonstrates that Holt’s Double Exponential Smoothing yields forecasts that align more closely with actual outcomes. This discrepancy suggests that while the ARIMA model excels in capturing long-term historical trends, the Holt method adapts better to recent shifts in trends, thus proving more precise for short-term predictions.  Consequently, the selection of a forecasting technique should be customized according to the objectives of the analysis and the specific features of the data utilized.