Clarida and Taylor (1993) and Clarida et al. (2001) offer a new alternative for nominal spot exchange-rate forecasting. They show that they can improve on forecasting the spot exchange rate over the random walk by using the combination of the full term structure of forward premia and a non-linear model. This study proposes an extension of the Clarida and Taylor framework by developing a Markov–switching model where the dynamics of the spot and forward exchange rate is tied to the term structure of interest rates. We motivate theoretically how an endogenous Markov switching model will emerge from a small open economy monetary model of exchange rate determination augmented with a monetary policy rule. And we estimate the resulting closed–form solution of the model with a Markov–switching vector error–correction model with a time–varying transition probability matrix. We show that our model can improve substantially over Clarida et al. (2001) model for certain currencies. Particularly on currencies for which the interest rate term structure has a strong relationship to the realized inflation regime.