Foreign Exchange Rate Predictability: Seek and Ye Shall Find It

There is ample literature on foreign exchange rate predictability and the results point out toward entirely different directions, some of it claiming that foreign exchanges are unpredictable random walks, some others that they are mean-reverting and cointegrating and yet some others insists that they can be best modeled and forecasted by nonlinear methods. In this chapter, we take a broader view for the methodology and a narrow on the application: first, we explore the predictability of the EUR/USD exchange rate, the leading currency in terms of liquidity and market attention and, second, we bring forth a number of incremental contributions. Specifically, we propose and implement two types of models that address the dual nature of exchange rates, the linear and nonlinear one, and we illustrate that on our dataset they provide possibly the best performance overall. Then, we provide strong supportive evidence on the time-varying nature of foreign exchange rate predictability and the need for forecast evaluation based on statistical and financial performance criteria. Finally, we provide several illustrations that in-sample fit and out-of-sample predictability are not necessarily separable and that increased in-sample fit can in many instances lead to improved such predictability. Practitioners that are interested in foreign exchange forecasting should consider the results of our chapter for they offer coherent suggestions to understand and exploit predictability when it decides to manifest in any currency pair.