Forecasting the Real Exchange Rates Behavior: An Investigation of Nonlinear Competing Models

Ruxandra Prodan
Yu Liu, University of Texas at El Paso


A large amount of literature finds that real exchange rates appear to be characterized by several non-linear specifications. While each of these nonlinear models fits some particular real exchange rates especially well, leading to good in-sample properties, the recent studies have not come to any consensus whether the nonlinear models could provide a better specification than the linear model and/or the random walk model according to their out-of-sample forecasting performances. Our goal is to examine two important nonlinear models (Band-TAR and ESTAR) concerning their abilities to generate out-of-sample forecasts, when estimating real exchange rates for 20 OECD countries. We find strong evidence that the ESTAR model, but not the linear or the Band-TAR model, outperforms the random walk model when forecasting out-of-sample. On the other hand, a comparison between the nonlinear models and the linear model is inconclusive due to the low power of tests for predictive ability when bootstrapping critical values.