Property Tax Compliance in the Digital Transition: Evidence from Argentina

We partner with the municipal tax administration to test the causal effect of the physical bill itself. Accounts in the eligible universe are randomly assigned to either have their printed bill suppressed (treatment) or to continue receiving it through the usual hand-delivery channel (control). The treatment is the absence of the printed bill; all accounts retain access to electronic/online payment channels, due dates, and amounts owed, which are unchanged by the experiment. We measure how suppression affects timely payment and revenue, overall and by prior-behavior group, with particular attention to heterogeneity within the mass-emission group.

2026-07-01

2026-07-10

The primary outcome is tax compliance, measured from administrative records at extensive and intensive margins. The main indicator is timely payment, a binary variable equal to one if the account pays on or close to the due date in each month. The primary outcomes are: (i) timely payment of installment 7 (the installment nearest the due date), (ii) timely payment of installment 8 and 9 (which barcodes and payments code are also included in the physical bill), and (iii) revenue collected over the evaluation window, analyzed both as a level and decomposed into the extensive margin (any payment) and the intensive margin (amount conditional on paying). We will also look at longer-run compliance.

The primary estimate is the intention-to-treat (ITT) effect of suppression, estimated separately within each behavior group (1, 2, 3). Because the treatment share differs across strata, group-specific effects are the headline. For Groups 1 and 2, the substantive claim is that suppression has no meaningful effect; this is tested via a cost-benefit analysis comparing the effects or lack of effects against the total cost (per-bill print + delivery cost). Effects are reported in percentage points (binary outcomes) and in ARS or log-points (revenue).

Secondary Outcomes (end points) Any-time payment of installments 7–9 (including late); number/share of installments paid in the window; share and amount of pre-existing debt settled; enrollment in a debt-repayment plan; payment-channel adjustments or behavior (web / home banking vs. physical).
Within Group 1 we separately estimate effects for direct-debit accounts versus non-direct-debit excellent digital payers, since direct debit should be mechanically unaffected by the printed bill.
Additionally, there is a group excluded from the experiment for administrative reasons (referred to as "Periphery Accounts", 4,731 accounts, and “Broches”, 3,024 accounts), for which we will assess the impact using a Difference-in-Differences or synthetic Difference-in-Differences methodology.

Experimental design
The main analytical interest is Group 3 (rest of mass emission), whose internal heterogeneity we exploit via subgroup analysis to identify which segments are most dependent on the physical bill (largest revenue loss under suppression).
We will look for heterogeneity through historic compliance, both measured as timely and non-timely compliance, historical payment-channel mix (digital vs. physical), assessed-value of the property, region or neighborhood, previous exemptions, prior payment-plan use, type of property (lot, apartment, house, etc) and if it’s charged the public-lighting charge (TAP).
We expect larger effects among accounts with room to adjust (occasional/late payers) and reduced or near-zero effects among never-payers (Group 2) and always-on-time/digital payers (Group 1).
Behavior groups are constructed from compliance over a 23-month window (2024 m7–12, 2025, 2026 m1–5).
Group 1: direct-debit accounts (entered directly) OR non-direct-debit accounts that paid on time in every evaluated period with ≥90% of payments made online.
Group 2: zero payments (timely or late) over 2024-H2 + 2025 + 2026, no annual payment, and no payment plan.
Group 3: all remaining accounts that receive a physical bill.
All the relevant data (i.e. properties characteristics and payments records) will be produced by the Municipality of General Pueyrredón and shared with the researchers in anonymized form, both prior to the experiment (baseline) and in periodic batches following its conclusion.
We will use OLS or ANCOVA to estimate mean differences between treatment and control, clustering standard errors at the account level. We may also estimate a DiD specification when using the panel, in order to benchmark post-intervention mean differences relative to the pre-intervention baseline compliance. As standard in the literature, we will control for past compliance behavior when it is feasible.

Not available

Randomization is done in office by computer, at the account level, stratified by group, with a fixed treatment probability within each stratum (70% in Group 1, 90% in Group 2, 25% in Group 3).

Tax Account

No

No clustering.

Sample size: planned number of observations 196,454 property tax accounts (eligible universe).

Sample size (or number of clusters) by treatment arms
Group Control Treatment (no bill) Total
1 — Always payers (digital + direct debit) 7,384 17,232 24,616
2 — Never payers (uncollectible) 5,858 52,721 58,579
3 — Rest of mass emission 84,945 28,314 113,259
Total 98,187 98,267 196,454

We focus the minimum detectable effect (MDE) calculation on Group 3 because it is the only group with a non-degenerate outcome in the baseline values. The estimated values for Group 3 are: N = 113,259; mean (ȳ) = 0.549 and a resulting MDE of 0.81 percentage points (power = 0.80, α = 0.05).