VARX and GSTARX Models for Forecasting Currency Inflow and Outflow with Multiple Calendar Variations Effect

Authors

  • Suhartono Suhartono Department of Statistics, Faculty of Mathematics, Computing, and Data Science Institut Teknologi Sepuluh Nopember (ITS), 60111 Surabaya, Indonesia
  • Muhammad Munawir Gazali Department of Statistics, Faculty of Mathematics, Computing, and Data Science Institut Teknologi Sepuluh Nopember (ITS), 60111 Surabaya, Indonesia
  • Dedy Dwi Prastyo Department of Statistics, Faculty of Mathematics, Computing, and Data Science Institut Teknologi Sepuluh Nopember (ITS), 60111 Surabaya, Indonesia

DOI:

https://doi.org/10.11113/matematika.v34.n3.1139

Abstract

VARX and GSTARX models are an extension of Vector Autoregressive (VAR) and Generalized Space-Time Autoregressive (GSTAR) models. These models include exogenous variable to increase the forecast accuracy. The objective of this research is to develop and compare the forecast accuracy of VARX and GSTARX models in predicting currency inflow and outflow in Bali, West Nusa Tenggara, and East Nusa Tenggara that contain multiple calendar variations effects. The exogenous variables that are used in this research are holidays in those three locations, i.e. EidFitr, Galungan, and Nyepi. The proposed VARX and GSTARX models are evaluated through simulation studies on the data that contain trend, seasonality, and multiple calendar variations representing the occurrence of EidFitr, Galungan, and Nyepi. The criteria for selecting the best forecasting model is Root Mean Square Error (RMSE). The results of a simulation study show that VARX and GSTARX models provide similar forecast accuracy. Furthermore, the results of currency inflow and outflow data in Bali,West Nusa Tenggara, and East Nusa Tenggara show that the best model for forecasting inflow and outflow in these three locations are VARX and GSTARX (with uniform weight) model, respectively. Both models show that currency inflow and outflow in Bali, West Nusa Tenggara, and East Nusa Tenggara have a relationship in space and time, and contain trends, seasonality and multiple calendar variations.

Downloads

Published

2018-12-31