We test whether the relationship between the nominal exchange rate and the news in its underlying fundamentals has non-linear features. In order to do so, we develop a Markov switching model and apply it to a sample of low and high inflation countries. The empirical analysis shows that for the high inflation countries the relationship between news in the fundamentals and the exchange rate changes is stable and significant. This is not the case, however, for the low inflation countries, where frequent regime switches occur. We develop two non-linear models that are capable of explaining our empirical findings. A first model is based on the existence of transaction costs; a second one assumes the existence of agents using different information to forecast the future exchange rate. In both cases we find that these simple non-linear models are capable of replicating the empirical evidence uncovered in this paper.