Finding a Connection Between Exchange Rates and Fundamentals, How Should We Model Revisions to Forecasting Strategies?

In this paper I compare the performance of three approaches to modeling temporal instability of the relationship between the euro-dollar exchange rate and macroeconomic fundamentals. Each of the three approaches considered -- adaptive learning, Markov-switching and Imperfect Knowledge Economics (IKE) -- recognize that market participants revise forecasting strategies, at least intermittently, and, as a result, the relationship between the exchange rate and fundamentals is temporally unstable. The central question in the literature addressed by this paper is which of the three approaches to modeling revisions of market participants' forecasting strategies is most empirically relevant for understanding the connection between currency fluctuations and fundamentals? One of the objectives of comparing the out-of-sample forecasting of the three approaches to change is to test to what extent growth-of-knowledge considerations, as proposed by Frydman and Goldberg (2007, 2011), are empirically relevant for our understanding of currency fluctuations. I find that only the IKE model, developed from Sullivan (2013) is able to significantly outperform the random walk benchmark, suggesting that different sets of fundamentals matter during different time periods in ways that do not conform to an overarching probability law.