Information and Incentives: The Agency Information Problem

This paper considers the use of imperfect information for risk sharing and incentive purposes when perfect observation of actions and outcomes is impossible, making complete contracting infeasible. The incentive-insurance problem is defined to consist of two parts: the choice of an information system and the design of a sharing rule based on the information system. A generalized agency model is formulated to analyse this problem. The agency models of Ross (1973a, b), Wilson (1968), Stiglitz (1974), Mirrlees (1976), Harris and Raviv (1979), Holmstrom (1979) a.o. appear as special cases of the generalized model. The analysis focuses on the value of information in the agency information problem. The set of information systems which are valuable—i.e. improve risk sharing and incentives in a Pareto sense—is characterized. A problem-independent ranking of information systems for the agency information problem is then characterized under the assumption that the agent's preferences are additive in money and actions. The ranking may be viewed as a generalization of Blackwell's ranking of information systems for decision problems, to this particular game. When the agent's risk preferences depend on his choice of action, on the other hand, it is shown that the Blackwell ranking may be invalid. Randomized incentive schemes are shown to be efficient when the incentive effect of risk is positive and sufficiently large relative to the absolute risk aversion of the partners.