We generalize the optimal tax models of Mirrlees (1976) and
Saez (2002) by allowing for externalities and a more
general social welfare function, and show that second best allocations can
be implemented through a non-linear, separable tax schedule if
ulterior motives of the social planner do not interfere with redistributive
preferences. This result solves a puzzle posed by Mirrlees in 1976, in the
sense that we can now show that tax-ratios are locally independent of type
because of incentive compatibility constraints. This result resonates very
strongly with the double deviation analysis done by Kocherlakota and others
in New Dynamic Public finance and our model can be used to relate many of
their outcomes to the more classical Mirrleesian optimal tax literature.
Furthermore this result holds strong technical implications for the optimal
tax literature on externalities, and merit and demerit goods.