When Tax Audit Targeting Creates a New Route to Evasion

A business tax return contains more than a profit figure. It can also describe expenses, assets, payroll and activities. Those details may help a tax authority decide where audits are most useful. They can also create an opportunity: a firm might understate its profits and misclassify its activities to obtain more favorable treatment.

In a theoretical preprint, Saara Hämäläinen, Eduardo Ferraz and Sebastián Castillo Ramos study how an authority with a limited budget might respond. Their answer combines two forms of enforcement: audits of reported profits and checks of the information used to classify taxpayers.

The result is not a test of an existing tax agency’s practices. It is a model of how firms and an authority could behave under specified incentives and costs.

Two decisions on one return

The model’s firms know their own taxable profits and their relevant classification data. They choose what to report after seeing an audit policy to which the tax authority has committed. The authority, in turn, chooses that policy to minimize the expected tax gap: the difference between statutory tax on actual profits and tax on reported profits.

A profit audit checks the income declaration. A data audit checks the classification; if it discovers a mismatch, it triggers a profit audit. Both cost money, but the model assumes that checking classification data is cheaper. It also assumes that audits perfectly detect misreporting in the dimension examined and that penalties rise with the amount of tax evaded.

The enforcement budget is too small to deter all profit underreporting. That constraint makes targeting attractive: if taxpayer categories are informative about profits, the authority can allocate audits differently across them. But different treatment also gives firms a reason to claim membership in a more leniently treated category.

A shared verification frontier

The researchers find that, in their model, differentiated enforcement links decisions across categories. A firm facing a relatively high profit-audit cutoff could try to report both a lower profit and a category with a lower cutoff.

To deter that joint misreporting, the modeled policy covers reports below a common verification frontier with one of the two instruments. In some categories, profit audits cover reports up to that frontier. In others, profit audits stop at a lower category-specific cutoff, and classification checks cover reports from that cutoff up to the frontier.

Coverage does not mean every return is audited. In the model’s optimal policy, a covered report faces a probability of the relevant check sufficient to deter the targeted misreporting. A data check that finds a false classification can then lead to a profit audit.

The distinction between actual and reported profits also matters. Firms whose actual profits exceed their category’s profit-audit cutoff report at that cutoff in the model. When that cutoff falls below the common frontier, these firms’ reports are subject to classification checks—even if their actual profits are above the frontier. Consequently, the cost of checking classifications can be substantial.

Why the best mix changes

The authors characterize the benefit of extending profit-audit coverage using an enforcement index. It depends on the tax schedule and the distribution of profits within each category. If the marginal benefit of extending coverage decreases with profit, different categories can receive different profit-audit cutoffs, with classification checks filling the gaps. But it can also be cheaper to extend some categories’ profit audits all the way to the common frontier and avoid separate classification checks there.

If the index instead increases with profit, the model favors a different division: some categories receive profit audits up to the frontier, while others rely on classification checks. Being exempt from profit audits in this case does not mean being exempt from enforcement altogether.

The model predicts bunching of reported profits at audit cutoffs, but no strategic bunching of classification data under successful implementation. Observed classification manipulation could be consistent with insufficient verification, the authors suggest; it would not by itself prove the model or identify the cause.

The price of better information

In one numerical illustration, the authors compare policies that ignore classification data, use truthful data, or use strategically reported data with optimal verification. The modeled tax gap is 0.151 without data, 0.140 with truthful data and 0.148 with strategic data and verification. The authors estimate that roughly two-thirds of the potential targeting gain is lost to the constraints associated with strategic reporting. These are normalized model results, not estimates for a particular tax administration.

The illustration uses three equally likely categories and assumed profit distributions, costs and penalties. Its magnitudes change when those assumptions change. The framework also fixes the tax schedule and enforcement budget and does not provide a full welfare analysis of taxpayer burdens or incidence.

Within those limits, the study’s central warning is that useful information is not automatically useful enforcement data. An authority must weigh improved targeting against the resources needed to keep the classifications reliable—and, if applicable, the cost of setting up the information system.