PEMODELAN NEURO-GARCH PADA RETURN NILAI TUKAR RUPIAH TERHADAP DOLLAR AMERIKA

Abstract

Exchange rate can be defined as the value of a currency against other currencies. Exchange rates always fluctuate all the time. Very high fluctuations and unconstant becoming problem in forecasting where the data changed extremely. Most of economic data have heteroskedasticity characteristic analyzed using (Generalized Autoregressive Conditional Heteroskedasticity) GARCH models. Another model that commonly used as an alternative is Artificial Neural Network (ANN). However, both models have weaknesses. ARIMA models are linear, but the residual probably still contains non-linear relationship, while the ANN model used to non-linear relationship there is difficulty in determining the input. In this research combination of two models is Neuro-GARCH model, with GARCH model used as input of ANN model. The purpose of this study was determined the best variance model Neuro-GARCH of return exchange rates rupiah against US dollar. The data used is daily return value of the rupiah (IDR) against the US dollar (USD) from August 27th, 2012 to March 31st, 2016. In this research, the mean model obtained is MA (1) and varian model is GARCH (1,1). The best model is Neuro-GARCH (2-10-1) with MSE smaller than the GARCH (1,1). Keywords: exchange rate, return, GARCH, Neuro–GARCH.