<![CDATA[Cloves are a strategic plantation commodity in Indonesia with important economic and cultural value, and their price volatility directly affects farmers’ welfare, supply chain stability, and regional economic planning. Although previous studies have shown that the generalized space time autoregressive (GSTAR) model is more flexible than the space time autoregressive (STAR) model for heterogeneous locations, empirical studies comparing GSTAR and vector autoregressive (VAR) models for clove price forecasting across geographically interconnected provinces remain limited. This study addresses that gap by comparing the forecasting performance of GSTAR and VAR for monthly clove prices in North Sulawesi, Central Sulawesi, and South Sulawesi. The novelty of this study lies in the application of GSTAR with three spatial weighting schemes uniform, inverse distance, and cross-correlation normalization and its comparison with VAR in the context of clove price forecasting. Monthly data from January 2015 to December 2024 obtained from the Central Statistics Agency were analyzed using an 80:20 training-testing split. Stationarity testing showed that all series became stationary after first differencing, and lag selection based on the Akaike information criterion identified lag 1 as optimal for both models. The results indicate that the GSTAR(1)I(1) model with cross-correlation normalization weights provides the best forecasting performance, with an average MAPE of 3.18% and RMSE of 5,729.84, outperforming the VARI(1,1) model, which produced an average MAPE of 10.57% and RMSE of 15,214.11. These findings confirm that incorporating spatial dependence significantly improves forecasting accuracy and demonstrates that GSTAR is a more effective model for geographically interconnected commodity markets.Keywords: Love price, forecast, GSTAR, SDGs 8, decent work and economic growth, VAR.]]>
