Exchange rate forecasting based on an artificial neural network and the random walk model

A financial time series (FTS) is related to the economic and business environments, and selecting an appropriate financial data model requires correct recognition of the inherent relationship between a financial market and the whole economy. The random nature of financial data makes modeling and forecasting extremely difficult. This study proposes a new combination forecasting method, which assumes that any financial time series data consist of linear and nonlinear parts. The linear part of the data is simulated using the random walk (RW) model, whereas the remaining residual nonlinear part is coprocessed using the feedforward artificial neural network (FANN) and the Elman artificial neural network (EANN). Empirical results show that the combination method exhibits higher prediction accuracy compared with the individual RW, FANN, and EANN models.