To provide actionable forecasting insights for gold prices in Indonesia’s public sentiment-driven market, this study developed a machine learning framework using the Google Trends Index (GTI) as a sentiment proxy. We employed an Optuna-optimized Support Vector Regression (SVR) model to comparatively evaluate three feature sets (GTI, historical Lag, and a Mix) across seven forecasting horizons (t+1 to t+30). A key advantage of our approach was the identification of horizon-dependent predictor dynamics: results revealed that while historical data excelled for short-term forecasts (MAPE 0.50% at t+5), the contribution of GTI became vital for long-term accuracy, where the hybrid model achieved its peak performance (MAPE 1.92% at t+30). Notably, the GTI-only model showed solid standalone potential (MAPE < 20%). We conclude that a hybrid approach is most effective, validating GTI as a relevant predictor for Indonesia. Furthermore, the proposed SVR-Optuna framework offers a generalizable methodology for forecasting other sentiment-driven assets, providing a clear, actionable guide for model selection based on forecasting horizons.
Ilahi et al. (Mon,) studied this question.