A Neural NARX Approach for Exchange Rate Forecasting

This paper presents a Nonlinear Autoregressive Exogenous (NARX) model using feed-forward neural network learning to forecast the exchange rate. We have evaluated the model performances by considering the exchange rate of the Thai baht per US dollar using the historical data from the Bank of Thailand for 10 years, from 2009 to 2018. We have used the following forecasting evaluation indices: Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE) and correlation coefficient (R). We have considered the following financial inputs for the neural system that predicts the current exchange rate: gross domestic product rate (GDP), interest rate, inflation rate, balance account, trade balance and a finite set of previous exchange rates. The best result showed that the NARX neural network technique leads to a MAPE of 3.001% and a MSE of 0.006. Therefore, we can conclude that the considered predictive model using the NARX neural network technique can be used to accurately forecast the exchange rate.