Strategic Auditing for Fraud.

Abstract

Abstract Two features of current audit practice are striking. First, auditors often use sampling methods to audit probabilistically. Second, the probability of auditing is usually contingent on information about that item. This essay studies an audit-reporting game in which a manager reports on the privately observed value of the firm. The report may be understated if some of this value is appropriated to the manager's own use. Only a costly audit can directly verify the report and impose penalties on the manager when fraud is discovered. The optimal audit policy for the owner of the firm is completely characterized and is shown to be both probabilistic and contingent: a report is never audited if it is sufficiently large, and is audited with a probability that is strictly less than one, otherwise. The optimal audit policy is contingent on the manager's report because the report depends on, and so provides some signal of, the firm's value. In contrast to optimal strategies in variance investigation models, the audit policy is also probabilistic. Variance investigation models are based on moral hazard problems, and efficient monitoring is deterministic because the agent has imperfect control over the outcome. In the adverse selection model studied here, the agent has perfect control over the report and would never choose a fraudulent report that will certainly be discovered. This makes probabilistic monitoring efficient. Because the model yields an exact solution, it can generate some useful comparative statics. As the audit cost decreases, the audit region expands and the amount of misreporting declines. The model also shows how the auditor's information affects the audit strategy and suggests how additional information can be efficiently incorporated into the audit policy.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 46 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.