I study the optimal audit mechanism when the principal cannot commit to an audit strategy. Invoking a revelation principle, the agent reports her type to a mediator who assigns contracts and recommends the principal whether to audit. For each reported type the mediator randomizes over a base-contract and the audit contract, which is accompanied by a recommendation to audit. I characterize the unique threshold value for penalties, that renders auditing profitable. For large penalties the optimal mechanism uses strictly more contracts than types and cannot be implemented via simple menu offers. These results provide a proper benchmark for studying audits under limited commitment, and provide insights on the institutional design of separating auditing from contracting. The analysis sheds new light on the usefulness of mediation in contracting and on the structure of optimal mechanisms in general. Creation Date: 2015-03-05
