A simple folk theorem of repeated games states that Pareto-efficient outcomes can be achieved in a perfect equilibrium where deviations are punished by a reversion to Nash equilibrium of the associated stage game (Nash threats equilibrium), provided that players are sufficiently patient. In a dynamic version of a well-known monetary policy game we show that such asymptotic efficiency may not be possible, as the presence of a state variable introduces the possibility of state manipulation. Moreover, the lowest inflation rate in Nash threats equilibrium may be increasing as players become more patient