<![CDATA[Micro and Small Industries (MSEs) serve as the backbone of the regional economy in Java, Indonesia, characterized by high volatility and potential spatial interdependencies. While advanced spatio-temporal models like GSTARIMA are theoretically superior in capturing inter-regional spillover effects, their empirical effectiveness on limited annual aggregate data remains questionable. This study evaluates the forecasting performance of the univariate ARIMA model against the multivariate GSTARIMA model for predicting the number of MSE business units across six provinces in Java from 2013 to 2024. Using a Queen Contiguity spatial weight matrix and Root Mean Square Error (RMSE) as the evaluation metric, the study rigorously tests the Principle of Parsimony. Preliminary analysis using Moran’s I indicates non-significant spatial autocorrelation, suggesting that business growth is predominantly driven by internal temporal inertia rather than immediate spatial propagation. Consequently, the results demonstrate that the simpler ARIMA model outperforms the complex GSTARIMA model in 10 out of 12 testing scenarios (83.3%), with accuracy improvements reaching up to 94% in specific provinces. The study concludes that adding spatial complexity to short-term annual time series (N=12) leads to over-fitting without proportional gains in accuracy. The ARIMA-based forecast for 2025–2029 identifies three distinct regional growth typologies: expansive growth, market saturation, and structural correction, providing critical insights for differentiated regional policy planning.]]>
