A Comparison of Forecasting Performance of Seasonal Arimax and Hybrid Seasonal Arimax-Ann of Surabaya’s Currency Circulation Data
This study concern on forecasting monthly currency outflow data of Bank Indonesia in Surabaya Region. The aim
of this paper is to compare the forecasting models based on Seasonal ARIMAX and hybrid Seasonal ARIMAX-ANN model.
The ARIMAX model is autoregressive integrated moving average (ARIMA) model with exogenous input. The Eid al-Fitr
effect is the exogenous input in Seasonal ARIMAX model. The hybrid model combine linear and nonlinear model. The
hybrid model in this study use Seasonal ARIMAX to model the linier components and using artificial neural network (ANN)
to model the non-linier components. Finally, the performance of the methods were evaluated based on the out-of-sample root
mean squared error (RMSE) for all series. The results showed that the hybrid Seasonal ARIMAX-ANN method perform best
for currency outflow data of Rp 50.000, Rp 20.000, and 5.000 series. However, for Rp 100.000 and Rp 10.000 series,
Seasonal ARIMAX method are much better than hybrid model.
Keywords- currency circulation, hybrid model, ARIMAX, Eid al-Fitr, calendar variation.