Taxation under Learning-by-Doing∗

We study optimal labor-income taxation in a dynamic economy in which the agents’ productivity is their own private information, is stochastic, and evolves endogenously over the lifecycle due to learning-by-doing. First, we identify novel distortions that originate in the endogeneity of the agents’ private information and that contribute to higher labor wedges under the optimal tax code. Next, we calibrate the model to US earnings data and find that reforming the current US tax code while accounting for learning-by-doing calls for a smaller increase in marginal tax rates and for more progressivity, compared to the optimal reform that ignores learning-by-doing. We also find that ignoring learning-by-doing overestimates the benefits of the reform. At the same time, reforming the code while accounting for learning-by-doing brings significant welfare gains. Lastly, we show that simple age-dependent taxes that are invariant in past incomes are approximately optimal and generate most of the welfare gains from the optimal reform. ∗For comments and useful suggestions, we thank Emmanuel Farhi, Mikhail Golosov, Marek Kapicka, Paul Klein, Alex Mennuni, Ludovic Renou, Stephanie Stantcheva, Arnau Valladares-Esteban, and seminar participants at the European Meetings of the Econometric Society, the Latin America and Caribbean Meetings of the Econometric Society, the World Congress of the Econometric Society, the second Shanghai Macroeconomics Workshop, CERGE-EI, Cergy-Pontoise, the Chicago FED Research Department, Northwestern University, the University of Western Ontario, and the 2016 ESSET Conference. Abdoulaye Ndiaye and Laura Doval provided excellent research assistance. The usual disclaimer applies. Email addresses: [email protected] (Makris); [email protected] (Pavan).