Prediction games

Abstract

We introduce a new class of mechanism design problems called prediction games. There are n self interested agents. Each agent i has a private input xi and a cost of accessing it. Given are a function f(x1, . . . , xn) which predicts whether a certain event will occur and an independent distribution on the agents' inputs. Agents can be rewarded in order to motivate them to access their inputs. The goal is to design a mechanism (protocol) which in equilibrium predicts f(.) and pays in expectation as little as possible. We investigate both, exact and approximate versions of the problem and provide several upper and lower bounds. In particular, we construct an optimal mechanism for every anonymous function and show that any function can be approximated with a relatively small expected payment. © Springer-Verlag Berlin Heidelberg 2005.