Indonesian rupiah exchange rate prediction using a hybrid ARIMA and neural network model

Abstract

Indonesian Rupiah (IDR) exchange rate is an indicator to measure the economic stability in Indonesia. An effort to maintain the stability of the IDR exchange rate is very important because it would directly impact Indonesia’s national monetary conditions such as debt settlement and export-import. One way to measure government policy in reducing the exchange rate is by making a prediction. The accurate prediction is determined by the model which is suitable to the data characteristics. Generally, exchange rate data is nonlinear imply the linear model is less effective to be applied. This study aims to model and predict the IDR exchange rate using a hybrid ARIMA and Neural Network model (ARIMA-NN), where ARIMA is for modeling linear components while NN is for modeling nonlinear components. This study uses daily data on US Dollar (USD) to IDR exchange rate from January 2015 - June 2020, which is categorized into 80% for training and 20% for testing. The results show that the best hybrid ARIMA-NN model is a combined model of ARIMA (1,1,1) and the NN model with 1 input, 1 hidden layer, and 5 neurons. The accurate prediction of this model is quite good with the smallest MAPE value.