PEMODELAN GENERALIZED SPACE TIME AUTOREGRESSIVE (GSTAR) PADA DATA INDEKS HARGA KONSUMEN (IHK) 5 IBUKOTA PROVINSI DI PULAU KALIMANTAN

Abstract

Generalized Space Time Autoregressive (GSTAR) model is a development model from the generalized STAR (Space Time Autoregressive) model. GSTAR model have autoregressive order to see the effect of the time element and location weighting matrix to see the effect of the location element. Unlike the STAR model, it can assume that each location research has different characteristics. The purpose of this research is to apply the Generalized Space Time Autoregressive (GSTAR) model to the Consumer Price Index (CPI) data in Kalimantan Island, especially in the capital city of each province in Kalimantan Island to find out the best estimation model with the best location weight. The location weights used the distance inverse location weights and the normalized cross-correlation location weights by estimating the parameters of the GSTAR model using the Ordinary Least Square (OLS) method. The best estimated model can be seen from the smallest Akaikae’s Information Criterion (AIC) and Root Mean Square Error (RMSE) value. From the research results, it was found that the best GSTAR prediction model for CPI data for 5 cities in Kalimantan Island was the GSTAR(1,1)-I(1). These results are based on the GSTAR prediction model with the smallest AIC value and the data is differencing 1 time. The best location weight based on the smallest RMSE value for the GSTAR(1,1)-I(1) model is the normalized cross-correlation location weight.