This paper combines policy response explanations of the uncovered interest parity puzzle with a time series approach that accounts for discrete central bank interventions. When monetary authorities manage the interest rate differential through an anti-inflationary policy rule, which allows for discrete shifts, then a stochastic segmented trends representation seems appropriate for the exchange rate and the interest rate differential series. In this setting, rational forecast errors are possible, and a test of the uncovered parity hypothesis, based on the cross-equation restrictions on a Markov switching process, suggests that the parity relationship cannot be rejected for three European currencies vis-" a avis the US dollar. Copyright # 2002 John Wiley & Sons, Ltd.JEL CODE: F31 KEY WORDS: switching regimes; policy rule; markov process; rational forecast errors 1. SUMMARY Empirical rejections of the relationship between expected exchange rate changes and the interest rate differential, also known as the uncovered interest parity (UIP) hypothesis, pose important questions concerning the rationality of economic agents as well as the usefulness of exchange rate structural theories and of many macro-econometric models for which UIP is a building block. Indeed, a rejection of UIP implies that there are systematically predictable excess returns in foreign exchange investments, a result that contradicts rational behaviour.Given that there are no riskless profit opportunities, i.e. the interest rate spread is equal to the forward premium, the validity of UIP implies that the forward premium is the best predictor of future exchange rate changes. Thus, most of the empirical research has relied on linear projections of ex post exchange rate changes on the forward premium and has produced regression coefficients significantly different from one (the value that corresponds to UIP) and more often negative. However, few economists are ready to accept the implications of UIP failure and, therefore, several research directions have been pursued for reconciling the parity relationship with empirical findings.One line of argument suggests that estimates of the above regression slope below 1 2 indicate the presence of a risk premium which is more variable than expected exchange rate changes, but neither static nor general equilibrium models can explain the observed variability of the risk premium. Another approach, based on survey data for exchange rate expectations, attributes the deviation of the regression coefficient from unity to the correlation between the forward premium and forecast errors, an outcome arising from irrational behaviour associated with heterogeneous agents. However, other researchers have argued that