ABSTRACT
Since Meese and Rogoff showed in 1983 that a simple driftless random walk model is the best out-of-sample predictor of exchange rates in a group of various structural and purely time series models, the random walk process has become a benchmark in predicting foreign exchange rates. In the following decades, there have been numerous attempts to improve the random walk model; however, a reliable and accurate prediction model that is superior to the random walk model has yet to be discovered. In this chapter, I have attempted to compare the out-of-sample prediction results of eight exchange rate prediction models with the prediction result of the random walk model. Unfortunately, the random walk model has beaten all eight of the candidates. Among the eight models, nonlinear specifications tend to return better prediction results than linear specifications, supporting the idea that nonlinear models should be employed by future researchers.
