Consistent Non-Utilitarian Welfare Weights

Saez and Stantcheva (2016) proposed evaluating tax reforms using generalized welfare weights, which encode the social value of giving a dollar to each person, to incorporate non-utilitarian values without specifying a social welfare function. However, Sher (2024) showed that generalized welfare weights cannot both incorporate non-utilitarian values and evaluate tax reforms consistently (in the sense that they do not generate preference cycles). We show that by relaxing a restriction imposed by both Saez and Stantcheva (2016) and Sher (2024) on how weights vary with the tax schedule, the generalized welfare weight approach can achieve both of these objectives. We characterize when weights are consistent: a policymaker with consistent generalized welfare weights necessarily behaves as if they were determining desirability of tax reforms via a well-defined global objective function, even if they never wrote one down; we show consistency can be checked via a symmetry condition. We also show that consistency does not require the policymaker to specify the underlying objective: any set of mechanical weights (society's valuation of a dollar transferred to each person, holding behavior fixed), utilitarian or not, can be completed into consistent generalized welfare weights by adding a term that captures the implicit value of behavioral responses. Our results are robust to multidimensional tax schedules and heterogeneity, behavioral agents, general equilibrium effects, and non-smooth behavioral responses such as bunching.