The out-of-sample forecasts of nonlinear long-memory models of the real exchange rate

We consider a new time series model that can describe long memory and nonlinearity simultaneously and can be used to assess an extensive evaluation of the out-of-sample forecasting performance of the nonlinear long-memory model. Upon fitting it to the real exchange rate, we find that a parsimonious version of the model captures the salient features of the data rather well. We then use this nonlinear long-memory model to forecast dynamically out-of-sample over the sample period for OECD countries. Overall, we find clear evidence that favours the nonlinear long-memory model over any other estimated models. Copyright © 2006 John Wiley & Sons, Ltd.