A Theory of Income Taxation under Multidimensional Skill Heterogeneity⇤

We develop a unifying framework for optimal income taxation in multi-activity economies with general production technologies. Agents are characterized by an N-dimensional skill vector that captures intrinsic abilities in N activities. The private return to each activity depends on individual skill and an aggregate activity-specific return, which is a general function of the economy-wide distribution of efforts across activities. The optimal tax schedule features a multiplicative income-specific correction to an otherwise standard tax formula. Because taxes affect the relative returns to different activities, this correction diverges, in general, from the weighted average of the Pigouvian taxes that would align private and social returns in each activity. We characterize this divergence as a function of relative return elasticities, and its implications for the shape of the income tax both generally and in a number of applications, including externality-free economies with general equilibrium effects, economies with increasing or decreasing returns to scale, zero-sum activities such as bargaining or rent extraction, and positive or negative spillovers. ⇤Email addresses: [email protected], [email protected]. We thank Spencer Bastani, Lizi Chen, Gordon Dahl, Alex Diesl, Roger Gordon, Laurence Jacquet, Guido Menzio, Torsten Persson, Edouard Schaal, Chris Sleet, Johannes Spinnewijn, Aleh Tsyvinski, Venky Venkateswaran, Gianluca Violante and seminar participants at Bocconi, Georgia State, Harvard, IIES Stockholm, LSE/IFS, NYU, University of Pennsylvania, Toulouse, UC San Diego, Wellesley, Zurich, the 2014 Taxation Theory Conference (Cologne), Barcelona GSE Summer Forum on Macro and Micro Perspectives on Taxation and National Tax Association Annual Meeting (Santa Fe) for helpful comments. Scheuer and Rothschild respectively thank Harvard University and the Radcliffe Institute for Advanced Study at Harvard University for funding and hospitality. All errors are our own.