Auditing policies and information

We first point out that, using any of the current criteria for comparing information systems in principal-agent models with moral hazard (such as Kim (1994)'s criterion), it is often impossible to contrast the value of information obtained from different policies of contingent audits that bear the same cost. Given two such policies A and B where, say, the lower cumulated frequencies of audits are always larger under B than under A, we show, however, that the likelihood ratio distribution associated with A dominates the one associated with B in the third order. A new, strictly finer, ranking of information systems then implies that the value of information is greater under A than under B when the agent's negative inverse utility function exhibits some prudence. The practical upshot is that the design of auditing policies involves somewhat more than the classical tradeoff between risk and incentives; it also requires to balance incentives and downside risk.