Score-based mechanisms

Abstract: We propose a mechanism design framework that incorporates both soft information, which can be freely manipulated, and semi-hard information, which entails a cost for falsification. The framework captures various contexts such as school choice, public housing, organ transplant and manipulations of classification algorithms. We first provide a canonical class of mechanisms for these settings. The key idea is to treat the submission of hard information as an observable and payoff-relevant action and the contractible part of the mechanism as a mapping from submitted scores to a distribution over decisions (a score-based decision rule). Each type report triggers a distribution over score submission requests and a distribution over decision rules. We provide conditions under which score-based mechanisms are without loss of generality. In other words, situations under which the agent does not make any type reports and decides without a mediator what score to submit in a score-based decision rule. We proceed to characterize optimal approval mechanisms in the presence of manipulable hard information. In several leading settings optimal mechanisms are score-based (and thus do not rely on soft information) and involve costly screening. The solution methodology we employ is suitable both for concave cost functions and quadratic costs and is applicable to a wide range of contexts in economics and in computer science.