Abstract
The observations at a certain location may be linearly influenced by the previous times of observations at that location and neighbor locations, which could be analyzed by Generalized STAR(1,1). In this paper, the weekly red-chili prices secondary-data of five main traditional markets in Bandung are used as case study. The purpose of GSTAR(1,1) model is to forecast the next time red-chili prices at those markets. The model is identified by sample space-time ACF and space-time PACF, and model parameters are estimated by least square estimation method. Theoretically, the errors’ independency assumption could simplify the parameter estimation’s problem. However, practically that assumption is hard to satisfy since the errors may be correlated each other’s. In red-chili prices modeling, it is considered that the process has time-correlated errors, i.e. martingale difference process, instead of follow normal distribution. Here, we do some simulations to investigate the behavior of errors’ assumptions. Although some of results show that the behavior of the errors’ model are not always followed the martingale difference process, it does not corrupt the goodness of GSTAR(1,1) model to forecast the red-chili prices at those five markets.