Income Distribution and the Limits to Policy Reform: Shock Therapy or Gradualism?

Abstract: The question of whether shock therapy or gradualism is the best choice for an economy engaged in trade policy reform is hotly debated. A central proposition in the literature is the observation by Mussa (1978) that, absent other distortions, the optimal reform policy is always shock therapy. The starting point of this paper is the simple observation that, in a context in which lump sum redistribution is unavailable, the fact that shock therapy is the best policy in terms of aggregate efficiency is insufficient to establish its optimality; rather, its income distributional effects will be crucial in determining its political feasibility and hence 'political optimality'. I model this in a simple way by analyzing a two-sector economy with adjustment costs, augmented by a two-tiered political economy of agenda-setting policymaker and legislature representing constituency interests. I show that the feasibility of shock therapy depends on the magnitude of the adjustment costs faced in retraining workers displaced by the reform. Whenever adjustment costs are smaller than a critical amount, shock therapy is politically feasible (as well as being optimal in terms of aggregate efficiency) and will therefore be proposed by the agenda-setter and ratified by the legislature. However, when adjustment costs are larger than this critical amount, shock therapy is not politically feasible. In this case, there always exists a gradualist reform agenda which is politically feasible. The possibility of time-inconsistency of the legislature's adoption of a gradualist agenda turns out to strengthen the case for gradualism, since if the relevant time-consistency constraint is binding, the reform program will need to be even more gradual than in the perfect precommitment case. Gradualism is also generally a more efficacious solution to the political feasibility problem than other alternatives such as pre-announced shock therapy or subsidizing worker retraining. The former is proneto serious time-inconsistency problems, whereas the latter involves higher efficiency losses in numerical simulation.