Comparison of Classical Time Series Method and Artificial Neural Network Method to Forecast Rupiah against US Dollar Currency Exchange Rate
M. Fuad Qamarul Alam (a*), Brady Rikumahu (b)

a) Business and Economic Faculty, Telkom University, Banduing, Indonesia
*fuadalam[at]student.telkomuniversity.ac.id
b) Business and Economic Faculty, Telkom University, Banduing, Indonesia

Abstract

The currency exchange rate is one of the macroeconomic components that have a distinctive characteristic of fluctuation and heteroskedasticity pattern. When referring to the historical data from 2008 to 2017, Indonesia rupiah has been depreciated toward US dollar as much as 44,59% were during that time it was consists of high fluctuation periods especially between 2008-2009 and 2014-2016 but also interspersed by a relatively stable period during 2010-2013. Consequently, research to find the forecasting model that can fit with exchange rate distinctive characteristic is critical. This research will focus on the projection performance comparison of ARIMA-GARCH classical time series method and Backpropagation Artificial Neural Network (BP-ANN) method for rupiah to the US dollar exchange rate. From Mean Squared Error (MSE) and accuracy level measurement, BP-ANN shows a better performance compared to ARIMA GARCH, while it also can be concluded if the forecasting performance of both models is decreasing along with the projection time duration.

Keywords: Time series; Forecasting; ARIMA; GARCH; Artificial Neural Network

Topic: Financial Technology

Link: https://ifory.id/abstract-plain/yWm63MwbU78E

Web Format | Corresponding Author (Muhammad Fuad Qamarul Alam)