Exchange Rates Forecasting Using Nonlinear Autoregressive

In this paper, an attempt has been made to forecast two major exchange rates USDAUD (US Dolar versus Australian Dolar currency) and USDJPY (US Dolar versus Japanese Yen currency) using Non-Linear Autoregressive (NAR) with monthly period data from January 1975 to April 2014. The data is collected from Bank of England. The ecchange rates forecasting use 3 layer models with input, hidden and output layer. The number of hidden nodes are 2, and 3 for USDAUD and USDJPY respectively. Levernberg-Marquardt (LM) learning algorithm is used to do training. The accuracy result for the experiments showed that Dstat parameter for USDAUD (60.56%) is a bit higher than industry standard accuracy but lower than industry standard accuracy for USDJPY (52.11%). The MSE result is 0.3239 for USDAUD and 7.518 for USDJPY.