Multivariate Forecasting of GAU/IDR Gold Prices Using a Hybrid Prophet-LSTM Model Based on Macroeconomic Indicators

Abstract

Forecasting gold prices in the Indonesian domestic market (GAU/IDR) presents its own unique challenges, as its volatility is simultaneously driven by global gold price fluctuations and the dynamics of the Rupiah exchange rate against the US dollar. This study proposes a multivariate Hybrid Prophet-LSTM model that integrates seven macroeconomic indicators (GAU/USD, USD/IDR, DXY, Federal Funds Rate, BI Rate, Indonesia's domestic inflation, and crude oil prices) as exogenous variables to forecast daily GAU/IDR gold prices for the 2021-2026 period. The hybrid architecture uses a residual learning approach: Prophet captures the long-term trend and seasonal decomposition, while LSTM corrects the remaining nonlinear residuals. A forecast continuity mechanism is also implemented to prevent unrealistic vertical jumps during the transition from historical data to future projections. A 90-day out-of-sample evaluation (January 2026-April 2026) shows that the Hybrid Prophet-LSTM achieves a MAPE of 0.8525%, RMSE of IDR 28,605/gram, MAE of IDR 22,398/gram, and an  of 0.9429, outperforming Prophet (MAPE 1.0655%) and a univariate LSTM (MAPE 22.7450%). A 30-day projection for May 2026 estimates the GAU/IDR price to be in the range of IDR 2,431,453 to IDR 2,651,542 per gram