Abstract

Noise can be efficiently introduced by a decision-maker into data to protect identity (differential privacy) or to reduce gaming between a decision-maker and an agent who can the manipulate data. We present a new benefit of noise: to efficiently reduce distortions in a second-best setting. We derive a condition---which quickly converges to standard DARA preferences in the number of agents---where the introduction of noise in the private provision of public goods is Pareto improving. Despite producing a risk cost, noise reduces free-riding, which is more valuable under our condition. The effect is large: total Nash giving, while still less than first best, now diverges in the number of donors instead of converges (the standard result). A second application relates to tax.