We consider the optimal nonlinear income taxation problem in a dynamic, stochastic environment when the government cannot change the tax rule as uncertainty resolves. Due to such a stationarity constraint, our taxation problem is reduced to a static one over an expanded type space that incorporates type evolution. We strengthen the argument in the static model that the zero top marginal tax rate result is of little practical importance because it only applies to the top of the expanded type space. If the maximal type increases over time, the person with top ability in any period but the last has a positive marginal tax rate.