Audit Rule Disclosure, Monitoring, and Tax Compliance: Experimental Evidence from Slovakia

The intervention combines audit-rule information disclosure and visible compliance monitoring. Randomly selected small enterprises receive official information about the Financial Administration’s risk-based approach to e-kasa/cash-register audits. One letter provides general information about risk-based auditing; a second letter provides more specific information about audit-selection criteria related to recorded sales and number of receipts relative to comparable businesses. In a crossed assignment, randomly selected firms are assigned to receive no visible e-kasa compliance check, one compliance check, or two compliance checks through audit/monitoring visits during the study period.

2025-03-10

2025-12-31

Primary outcomes are firm-level measures of recorded compliance based on administrative e-kasa/tax records: (1) total sales recorded through e-kasa/cash registers in post-intervention windows; (2) number of receipts recorded through e-kasa/cash registers in post-intervention windows; and (3) audit/inspection outcomes, such as whether a violation was found, if consistently available. Main windows will include periods after the letter and after assigned or realized audit rounds, such as 30-, 60-, and 90-day windows, subject to data availability and exact implementation dates.

Sales and receipt outcomes will be measured using administrative records and constructed as firm-level totals or logs over pre-specified post-intervention windows. When windows depend on an audit date, windows will be calculated relative to the firm-specific audit date. Pre-treatment controls will use analogous outcomes from the same calendar period in 2024, when available. Audit/inspection outcomes will be coded using administrative audit records if available and consistently measured.

Secondary outcomes may include VAT liabilities or payments, tax filing/payment behavior, changes in payment method or cash intensity if available, amendments/corrections, firm inactivity or exit, and heterogeneity by pre-treatment risk, sales, receipt counts, sector, region, municipality size, entity type, VAT status, establishment year, profitability, number of employees, number of cash registers, and prior audit history.

Secondary outcomes and heterogeneity variables will be constructed from pre-treatment and follow-up administrative records, if available and linkable to randomized firms. Heterogeneity variables will be measured before treatment assignment whenever possible.

The study is a randomized field experiment among small enterprises in Slovakia that use e-kasa online or virtual cash registers. Eligible units are selected from administrative records based on pre-treatment characteristics and risk criteria. The design crosses three information-disclosure conditions with three visible audit/monitoring conditions. Implementation records indicate that the letters and visible e-kasa compliance-check interventions were carried out during 2025, with some deviations from the originally assigned monitoring schedule. The public design description is intentionally high level while the study is ongoing, because exact operational criteria and timing may affect taxpayer behavior.

Not available

Computer-based randomization using administrative baseline variables and stratification/blocking. Randomization occurred after the final sampling frame was assembled in March 2025 and before implementation of the mailing/audit treatments. The randomization crossed letter assignment with audit/inspection assignment and used strata based on pre-treatment administrative characteristics.

Firm/taxpayer/cash-register user, aggregated to the taxpayer/firm level for analysis. Treatment was assigned at the firm/taxpayer/cash-register user level. If outcomes are observed at daily or monthly frequency, standard errors will be clustered at the randomized unit level.

No

Approximately 3,840 firms/taxpayers/cash-register users.

Approximately 3,840 firm-level randomized units. Administrative outcomes may also form a firm-period panel.

no letter: 1,280 firms; general risk-based audit-information letter: 1,280 firms; specific audit-selection criteria letter: 1,280 firms. No visible audit: 1,920 firms; one visible audit: 960 firms; two visible audits: 960 firms. 3-by-3 table: each letter arm has approximately 640 no-audit firms, 320 one-audit firms, and 320 two-audit firms.

Using the sample size design numbers, a two-sided 5 percent test, 80 percent power, and the conservative variance of a binary compliance outcome with a 50 percent baseline rate, the design can detect approximately: 4.8 percentage points for any letter versus no letter (2,560 versus 1,280 firms); 5.5 percentage points for the general versus specific letter comparison (1,280 versus 1,280 firms); 4.5 percentage points for any visible audit versus no visible audit (1,920 versus 1,920 firms); and 6.4 percentage points for one versus two visible audits (960 versus 960 firms). Comparisons within individual 3-by-3 cells have lower power; a 320-versus-320 comparison has an MDE of approximately 11.1 percentage points. These calculations do not include precision gains from stratification, lagged outcomes, or covariate adjustment. For log sales and log receipt outcomes, detectable proportional effects will depend on residual variance after controlling for pre-treatment outcomes; the planned lagged-outcome/log-linear specifications should improve precision. Evidence from tax-compliance mailing experiments suggests that enforcement/deterrence messages are among the more effective letter treatments, but average letter effects may be modest. The design is therefore primarily powered for economically meaningful letter effects of roughly 5-10 percentage points and for audit/inspection effects that may be larger than purely informational letters.