Incentive Compatibility in Risk Assessment Mechanisms

Abstract

This paper defines a risk assessment mechanism and compares its incentive properties with those of deterministic incentive mechanisms, particularly the Groves mechanism. Many risk assessments involve prediction for rare or unique events; in such cases there is limited opportunity for feedback and evaluation of the assessment process. To develop a feedback mechanism, the paper requires assessments to be made for indicator events, linked to the rare or unique events of ultimate interest. Assessments are made by several assessors, or assessment techniques, acting in competition. The feedback mechanism is a transfer function based on the probability assessments of all the assessors and the outcome of the indicator event. The incentive properties of risk assessment mechanisms are in some ways similar to those for deterministic mechanisms and in some ways quite different. The paper defines one risk assessment mechanism that looks like a Groves mechanism: it directly reveals probability and for risk neutral assessors has an unbiased or truthful dominant strategy which is discontinuous and which cannot solve the budget problem. The paper also defines a class of risk assessment mechanisms which do not look like a Groves mechanism; mechanisms in this class have unbiased dominant strategies which are continuous and which do solve the budget problem.