MODELING OF INCREASE IN COVID-19 CASES IN BANDUNG CITY USING GSTAR MODEL WITH HETEROSCEDASTIC EFFECTS

• Main Author: Rizky Kurniawan, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55601
• Institution: Institut Teknologi Bandung

The generalized space-time autoregressive (GSTAR) model is one of several models that can be used to model a process whose observed values depend not only on the time of the observation, but also the location of the observation. The GSTAR model is derived from the generalized space-time autoregressive moving average (GSTARMA) model, and those models are analogous to the univariate autoregressive (AR) and autoregressive moving average (ARMA) models respectively. Univariate time series models can be modified to model processes that feature non-constant residual variances by using the generalized autoregressive conditional heteroskedasticity (GARCH). A similar modification can be done on space-time series. In this final project, the daily increase of COVID-19 cases in 30 districts of Bandung City during a time period from November 26th, 2020 to March 31st, 2021, is modeled using a GSTAR model by taking heteroscedastic effects into account. To identify the GSTAR model, the uniform weight matrix is used, which assumes that the increase of COVID-19 cases in each of the districts are affected equally by neighbouring districts. The uniform weight matrices are based on the first- and second-order neighbours of each district, and is limited to two matrices to simplify the modeling process. Based on the weight matrices, one possible model is GSTARI(12) (since the data underwent differencing). From the residuals of this model, the GARCH(2,1) model is chosen to model heteroscedasticity, with a note that the GARCH parameters associated with spatial lags 1 and 2 are very small. The data generated by this model is then compared to the actual dataset. The results showed that this model is not a good fit for the data during the above mentioned time period, as the MASE value for the majority of the districts is greater than one.