Abstract We analyze the optimal income tax problem when consumers work for many periods. Can information about abilities learned in one period be used later to attain more redistribution than is possible in a one-period world? When the government can commit to future policies and has the same discount rate as individuals, intertemporal nonstationarity of tax schedules can lead to Pareto improvements by relaxing lifetime self-selection constraints, analogous to randomization in a one-period world. More significant use of information is possible when the social discount rate is less than individuals' rates. The taxes converge over time to a nondistortionary schedule.