Economic analyses of farm policies generally focus on the long run, steady state impacts while the transition dynamics are often overlooked. In this paper we develop a determinist dynamic computable general equilibrium analysis allowing agents to form imperfect versus pefect expectations. Using an illustrative CAP reform scenario, we simulate an abrupt versus a gradual implementation of this reform. Our results show that if economic agents are able to perfectly anticipate the impacts of the reform, then delaying its implementation is never optimal. On the other hand, if agents gradually learn from market developments, then we find some cases where a gradual implementation of this reform is welfare improving. Such gradual implementation allows minimizing adjustment costs.