Exchange rate is the currency value of a country that is expressed by the value of another countrys currency. Changes in exchange rates indicate risks or uncertainties that would return obtained by investors. With the predicted value of return, investors can make informed decisions when to sell or buy foreign currency to gain an advantage. Forecasting of return values can be using artificial neural network with backpropagation. In backpropagation procedure, data is divided into two pairs, namely training data for training process and testing data for testing process. In the training process, the network is trained to minimize the MSE. One of optimization method that can minimize the MSE is one step secant backpropagation. In this research, the data used is the return of the exchange rate of rupiah against US dollar in the period of January 1st, 2015 until December 31st, 2015. The results were obtained architecture best model neural network that was built from 8 neurons in the hidden layer, 1 unit of input layer with input xt-1 and 1 unit of output layer. The activation function used in the hidden layer and output layer are bipolar sigmoid and linear, respectively. The architecture chosen based on the smallest MSE of testing data is 0.0014. After obtaining the best model, data is foreseen in the period of November 2016 produce MAPE=153.23%.Keyword : Artificial Neural Network, Backpropagation, One Step Secant, Time Series, Exchange Rate.
Copyrights © 2017