The paper addresses the Kydland and Prescott (1977) argument that the optimal policy in models with rational expectations is time-inconsistent. This, it is argued, undermines the credibility of the optimal policy in the eyes of the private sector, who will expect the policy-maker to reoptimize. Therefore policy, if it is to be credible, must be constrained to be time-consistent. For many models, this is a serious constraint. Barro and Gordon offer a different approach to the time inconsistency problem. They assume that policy-makers suffer a loss of reputation if they renege on earlier commitments. With this "punishment" mechanism in place, Barro and Gordon show that there exist policies superior to the time-consistent policy which are credible and sustainable. The Barro-Gordon analysis is, however, model-specific and, in particular, applies only to static models. The main purpose of this paper is to show how their analysis can be generalized to any rational expectations model with structural dynamics and stochastic exogenous shocks.
