A MGSTAR: An Extension of The Generalized Space-Time Autoregressive Model

Abstract Up to now, Generalized Space-Time Autoregressive (GSTAR) models are focused only for univariate spatial-temporal data. This research proposes an extension of GSTAR multivariate data, known as Multivariate or MGSTAR. Three studies were conducted in this research, i.e., theoretical, simulation, and applied studies. These initially developed based on bivariate A theoretical study was done by developing MGSTAR the framework Vector (VAR) models. In proposed model, parameter estimation obtained implementing Ordinary Least Square (OLS) method. The simulation showed that OLS method yielded unbiased estimator. Furthermore, have forecasting CO PM10 at three stations Surabaya City. results model could explain well dynamic relationship between variables locations. However, Root Mean Error Prediction (RMSEP), less accurate forecast than ARIMA due employed simpler order Autoregressive. Further is needed expand with a higher Autoregressive, particularly handle trend seasonal order.