Enhancing Rice Price Forecasts with Generalized Space-Time Autoregressive (GSTAR) Models and Spatial Weighting Variations

Rising rice demand in Indonesia, driven by population growth, causes price fluctuations that impact household spending. Accurate forecasting is crucial for price stability and government planning. This study employs the Generalized Space-Time Autoregressive (GSTAR) model with spatial weight variations to forecast rice prices across six provinces in Java. The results indicate that the GSTAR (71)I(1) model, utilizing radial distance weights (RDW), was identified as the optimal model. It satisfies the white noise assumption and achieves superior performance metrics, with a mean squared error (MSE) that is considerably lower than those obtained from other spatial weight models tested in this study. The mean absolute error (MAE) also demonstrates a strong accuracy, and the mean absolute percentage error (MAPE) is exceptionally small, suggesting minimal deviation from actual values when compared to other models. These values are notably lower compared to those of other spatial weight models tested in this study.