The Optimal Number of Tax Audits: Evidence from Italy

Abstract

Tax audits are the main tool adopted by tax administrations to collect taxes. Their optimal number depends on two parameters, i.e. the enforcement elasticity of tax revenue with respect to the audit effort and the sum of private compliance costs and public administrative costs entailed by audits. In turn, the enforcement elasticity critically depends on audit selection criteria actually chosen by tax authorities. In this paper, we apply a machine learning approach to Italian data and we provide evidence that, in 2010 and 2011, audited taxpayers are those whose reporting behaviour in between the report year and the audit year has deviated from the business cycle. We use these audit criteria to match audited taxpayers to non-audited ones and we obtain an estimate of the enforcement elasticity that allows us to characterize the optimal number of tax audits as a function of the ratio between private compliance and public administrative costs.

Keywords: Optimal Tax Administration, Enforcement Elasticity of Tax Revenue, Machine Learning