How reliable should auditors be?

Abstract This paper considers a principal–agent model with auditing and collusion, in which the audit costs are a convex function of the audit reliability. The focus of interest is the question of the audit reliability chosen by the principal. It turns out that the optimal audit reliability strongly depends on the given maximum punishment and on whether collusion is possible or not. The principal chooses a higher level of reliability if collusion is possible. Moreover, in the presence of a high maximum punishment, the optimal corruptible auditor is very reliable and very expensive, whereas the optimal honest auditor is unreliable and cheap.