How to Sell Information Optimally: an Algorithmic Study
summary
Provides an algorithmic characterization: when type distribution and payoff matrix are explicit the revenue-optimal menu is computable via a polynomial-size linear program; when actions are implicit but a best-response oracle exists and the number of states is constant there is an additive FPT algorithm that merges close signals and reduces the action set; without the oracle even two-state/two-action instances make it NP-hard to approximate revenue within any constant factor; extends model to multi-agent winner-selection with a reduced-form LP.
AI-written summary · about