Zero Restricted GSTAR Modeling for Regional Inflation Forecasting in East Java

Abstract

The Generalized Space-Time Autoregressive (GSTAR) model is a spatial-temporal time series model that captures both temporal dynamics and spatial interactions among multiple locations while allowing heterogeneous parameters across regions. However, as the number of locations and lag orders increases, GSTAR often suffers from overparameterization, which may reduce estimation efficiency and model stability. To address this limitation, this study proposes the Zero Restricted Generalized Space-Time Autoregressive (R-GSTAR) model, which adapts the parameter restriction concept from Restricted Vector Autoregressive (VAR) models by constraining insignificant coefficients to zero. The proposed approach is applied to monthly inflation data from Surabaya, Malang, and Kediri using inverse distance and uniform spatial weighting schemes. Model performance is evaluated using the Akaike Information Criterion (AIC), residual diagnostics, and the Mean Squared Error (MSE). The results indicate that the R-GSTAR model with uniform weights provides the best performance, yielding the lowest AIC value of -273.1054 and MSE values of 0.1225, 0.1454, and 0.2025 for Surabaya, Malang, and Kediri, respectively. These findings demonstrate that the proposed model offers an efficient and accurate approach for regional inflation forecasting.