We use a Markov switching approach in which we account for the presence of two potential regimes: ordinary and turbulent. We also recognize the fact that, even within each regime, the volatility of exchange rate returns is not constant, and we therefore include a GARCH specification. The probabilities of switching between the two regimes are time-varying. The attractiveness of this approach is that we do not need to distinguish ex-ante between ordinary and turbulent times, but we let the estimation results supply us with this information. We differ from Mariano et al. (2002) in that we recognize the importance of volatility dynamics, and we enlarge the set of potential explanatory variables to include the M2-reserve ratio, real domestic credit, real effective exchange rate, stock market returns and volatility, and banking sector returns and volatility.