This paper characterizes the optimal degree of jointness in the taxation of married couples. We develop a tax-perturbation framework to analyze reforms that vary the degree of jointness, decomposing their efficiency effects into a tagging gain from conditioning taxes on spousal earnings and a distortion arising because those earnings are endogenous. Our main theoretical result shows that, starting from a separable, revenue-maximizing (Rawlsian) tax schedule, introducing negative jointness—where marginal tax rates decrease with spousal earnings—increases tax revenue under a mild regularity condition satisfied by a broad class of empirically relevant joint income distributions. More generally, positive jointness is optimal only if the planner places sufficiently high welfare weight on rich couples with unequal earnings. Using U.S. data on married couples, we find that the globally optimal tax schedule raises the revenue-maximizing lump-sum transfer by about 3% relative to individual taxation, redirecting redistribution toward the poorest couples.

This paper studies the joint design of nonlinear income taxes and education subsidies when schooling raises productivity, signals ability, and improves employers' information, thereby enhancing worker--task assignment. I develop a model in which these roles operate jointly. Under utilitarian welfare, marginal education subsidies of roughly -60 to -70 percent are optimal across the spending distribution, primarily because of the externality generated by signaling. Signal precision reduces the magnitude of these negative subsidies by about 5 percentage points because better worker-specific information weakens employers' reliance on credentials. With stronger redistributive preferences, the education-subsidy schedule becomes progressive: subsidies remain strongly negative at low spending margins, become less negative at higher margins, and turn positive near the top. Precision-driven ability--wage sorting becomes more socially valuable when income taxation is redistributive because it improves task assignment, raises output, affects tax revenue, and makes income a more informative proxy for latent ability.

When should governments sacrifice production efficiency for redistribution? We generalize the celebrated result of Diamond and Mirrlees (1971a,b) by allowing for imperfect competition, suboptimal and nonlinear taxation. We demonstrate that production efficiency hinges on the flexibility of the tax system in compensating gains and losses from changes in factor prices. This requires the tax system to target each factor’s income. We show how to adjust tax systems or production policies for imperfect targeting and market failures, even when the tax system is not flexible enough. We then obtain new sufficient statistics formulas. Endogenous factor prices do not modify the test to identify Pareto-improving tax reforms.