New econometric and statistical techniques have been used in recent years to provide with exchange rates forecasting models that can statistically outperform a random walk. In particular, a model that uses the term structure of forward premia into a regime-switching vector error correction model has proven to be successful at such a task. In this paper, we propose that the exchange rate fluctuations are not solely influenced by the economic fundamentals of those countries involved in the exchange. Therefore, the accuracy of the aforementioned model can be improved by separately forecasting the average change in value of each of the currencies involved in the exchange rate, instead of forecasting the exchange rate itself. This is achieved by using a low volatility currency basket to transform the data before and after the modeling.