Comparing linear and non-linear benchmarks of exchange rate forecasting

M.Com. (Financial Economics)Exchange rate forecasting has been an important and complex field of study originating mainly from the introduction of floating exchange rates in the 1970s. Since then, various models have been developed to explain exchange rate behaviour, all contributing in their own way to the understanding of what economic and financial information reveal about the future price of exchange rates. To measure the performance of a variety of exchange rate models, researchers in exchange rate forecasting almost always use the random walk model as benchmark to evaluate the forecasting performance of exchange rate models. An exchange rate model is regarded as superior if it can outperform a random process. The random walk model, a special case of the unit root process, helps us to identify the kinds of disturbances that drive the exchange rate to follow an independent successive process. If the exchange rate follows a random walk process, it has no mean reversion tendency and a directional shock in the exchange rate will cause it to deviate from its long-run equilibrium. Conversely, if the exchange rate does not follow a random walk, it has mean reverting tendencies, and will follow a stationary process which allows us to accurately forecast the exchange rate based on historic observations (Lam, Wong and Wong, 2005:1). However, it seems unrealistic that exchange rates will follow either a random walk process or a stationary process. If we assume that the exchange rate follows a random walk, we also assume that the order flow information from exchange rate trades follows a random walk, and by implication that macroeconomic exchange rate information follows a random walk [see Lyons (2001) for the link between order flow and macroeconomic fundamentals]. It seems unrealistic that exchange rates will follow an identifiable mean reverting (stationary) process, as daily exchange rates are exposed to risk, news and speculation which functions independent from long-run exchange rate fundamentals. Ironically, Meese and Rogoff (who laid the foundation for the use of random walk models as benchmark in exchange rate forecasting) emphasize that exchange rates do not follow an exact random walk (Meese and Rogoff, 1983:14). However, if it is known that exchange rates do not follow a random process explicitly, alternative exchange rate benchmark models should be considered. Yet, judging by the universal..