Nudges and Tax Compliance

Firms receive official letters from the Secretaría de Hacienda de Medellín reminding them of their obligation to file their 2024 Industry and Commerce Tax (ICA) declaration. Letters vary in content across four groups: (1) neutral reminder (control), (2) deterrence message highlighting penalties and enforcement, (3) social norm message emphasizing peer compliance rates, and (4) combined deterrence and social norm message.

Due to administrative and operational considerations in identifying eligible non-filers, the study is implemented in two phases:

Phase 1 (May 2026): n≈1,000 firms. Messages sent May 5, 2026 via email and certified mail, with a 15-day deadline to file voluntarily.

Phase 2 (July-August 2026): Approximately 6,000 additional firms receive messages using identical protocols.

All messages are official communications from the Tax Authority. Outcomes measured at 90 days post-intervention from administrative records.

2026-05-05

2026-08-31

Primary Outcome 1: Tax Filing (Binary)

Definition: Whether the firm filed its 2024 Industry and Commerce Tax (ICA) declaration within 90 days of receiving the Phase 1 treatment letter (April 16 - June 7, 2026)
Measurement: Binary indicator (1 = filed, 0 = not filed) extracted from Tax Authority administrative records
Timeline: Measured on June 1, 2026

Primary Outcome 2: Days to Filing (Continuous)

Definition: Number of calendar days between April 16, 2026 (treatment letter sent) and the date the firm filed its ICA declaration
Measurement: Continuous variable (0-90 days) for firms that filed; censored at 90 days for non-filers
Source: Tax Authority filing timestamp data

Primary Outcome 3: Tax Payment (Binary)

Definition: Whether the firm paid its self-declared 2024 ICA tax liability within 90 days of receiving the treatment letter
Measurement: Binary indicator (1 = paid, 0 = not paid) from Tax Authority payment records
Timeline: Measured on June 1, 2026

Primary Outcome 4: Amount Paid (Continuous)

Definition: Total amount (Colombian Pesos) of 2024 ICA tax paid by the firm within the 90-day window
Measurement: Continuous variable (COP) from Tax Authority payment records
Note: Zero for firms that did not pay
All primary outcomes are measured at the firm level using administrative data from the Secretaría de Hacienda de Medellín. The 90-day window (April 16 - June 7, 2026) corresponds to Phase 1 voluntary compliance period before enforcement actions begin.

All primary outcomes are directly observable from administrative records and require no construction or index creation:
Tax Filing (Binary): Extracted directly from Tax Authority database field indicating declaration submission status. Coded as 1 if filing timestamp falls between April 16, 2026 00:00:01 and June 7, 2026 23:59:59; coded as 0 otherwise.
Days to Filing (Continuous): Calculated as: (Filing Date - April 16, 2026). For firms that filed, this equals the number of calendar days elapsed. For firms that did not file within 90 days, the variable is censored at 90 days for survival analysis, or coded as missing for linear regression specifications.
Tax Payment (Binary): Extracted from Tax Authority payment records. Coded as 1 if any payment toward 2024 ICA liability is recorded between April 16, 2026 and June 7, 2026; coded as 0 if no payment recorded in this window.
Amount Paid (Continuous): Sum of all payments (in Colombian Pesos) applied to 2024 ICA tax liability during the 90-day window, extracted directly from payment transaction records. No winsorizing or transformation applied for primary analysis; robustness checks will use log(Amount + 1) and inverse hyperbolic sine transformations to address skewness.
No composite indices or constructed variables are used for primary outcomes. All outcomes represent actual administrative events (filing, payment) or direct calculations from timestamps and transaction amounts recorded by the Tax Authority in the normal course of tax administration.

Phase 1 Secondary Outcomes (Voluntary Compliance Period):
1. Filing Within 15 Days (Binary)

Definition: Whether firm filed within the initial 15-day deadline stated in the letter (April 16- May 1, 2026)
Measurement: Binary indicator from administrative records
Purpose: Test responsiveness to explicit deadline

2. Tax Base Declared (Continuous)

Definition: Self-reported tax base (gross revenue subject to ICA) in Colombian Pesos
Measurement: Extracted from filed declaration form
Note: Only observable for firms that filed

3. Tax Liability Declared (Continuous)

Definition: Self-calculated ICA tax owed based on declared tax base and applicable rate
Measurement: From filed declaration (COP)
Purpose: Assess whether messages affect reporting accuracy

4. Partial vs. Full Payment (Categorical)

Definition: Payment status categorized as: (0) No payment, (1) Partial payment (<100% of declared liability), (2) Full payment (≥100% of declared liability)
Measurement: Ratio of amount paid to tax liability declared

Phase 2 Secondary Outcomes (Enforcement Response - Exploratory):
5. Filing After Enforcement Notice (Binary)

Definition: Whether firm filed within 30 days of receiving emplazamiento (enforcement notice)
Measurement: Binary indicator for firms that received enforcement
Note: Conditional on non-random enforcement assignment; analyzed with propensity score matching

6. Payment After Enforcement (Binary)

Definition: Whether firm paid within 30 days of receiving emplazamiento
Measurement: Binary indicator from payment records

7. Long-term Compliance Persistence (Binary)

Definition: Whether firm filed AND paid during September-December 2026 follow-up period (delayed compliance)
Measurement: Binary indicator for firms that did not comply in Phase 1 or Phase 2 enforcement window
Purpose: Assess treatment effect persistence beyond immediate response

8. Email Engagement (Binary)

Definition: Whether firm opened the treatment email (if tracking data available)
Measurement: Email open rate from delivery system logs
Purpose: Mechanism check for message delivery

Design: 2×2 factorial randomized controlled trial testing two behavioral mechanisms in tax compliance messaging

Sample: Firms in Medellín, Colombia that failed to file their 2024 Industry and Commerce Tax (ICA) declaration. Implementation in two phases due to administrative process of identifying eligible non-filers on a rolling basis:
- Phase 1 (May 2026): n≈1,000 firms
- Phase 2 (July-August 2026): n≈6,000 firms
- Total: ≈7,000 firms

Randomization: Stratified random assignment by taxpayer type (2 categories) × sector (4 categories) × firm size (4 quartiles) × firm age (4 groups). Each phase uses independent randomization. Firms are randomly assigned with equal probability to one of four groups within strata.

Treatment Arms:
- Group 1: Control message (neutral reminder)
- Group 2: Treatment A
- Group 3: Treatment B
- Group 4: Treatment A + Treatment B (combined)

Intervention Timing:
- Phase 1: Messages sent May 5, 2026
- Phase 2: Messages sent July-August 2026
All messages sent via email and certified mail from Secretaría de Hacienda de Medellín. Firms receive a 15-day deadline to file voluntarily.

Primary Outcome Measurement: Tax filing and payment behavior measured at 90 days post-intervention using Tax Authority administrative records

Follow-up Period: Observational follow-up (through December 2026) tracks firms that did not file during the initial 90-day window. This phase analyzes responses to the Tax Authority's standard enforcement procedures using propensity score matching (non-experimental).

Analysis: Primary analysis pools both phases with phase fixed effects, comparing filing and payment rates across the four groups using ordinary least squares regression with stratification controls. The factorial design enables estimation of main effects for Treatment A and Treatment B, as well as their interaction effect to test for complementarity or substitution.

Statistical Power:
- Phase 1 only (n≈1,000): Main effects MDE ≈2.5pp, Interaction MDE ≈3.5pp
- Both phases combined (n≈7,000): Main effects MDE ≈1.1pp, Interaction MDE ≈1.5pp
Calculations assume baseline filing rate of 3-5% and α=0.05 (two-tailed).

Not available

Method: Computerized stratified block randomization conducted by the Secretaría
de Hacienda de Medellín using R statistical software

Procedure:

The Tax Authority uses a random number generator to assign firms to treatment
groups. Randomization is conducted on the message deployment date using the
following steps:

1. Stratification: Firms are sorted into strata based on:
- Taxpayer type (2 categories: natural person vs. legal entity)
- Economic sector (4 categories: construction, restaurants, tourism/RNT,
financial services)
- Firm size (4 quartiles based on log of taxable base)
- Firm age (4 groups: pre-2000, 2000-2009, 2010-2019, 2020 onwards)

2. Block randomization within strata: Within each stratum, firms are randomly
assigned to one of four groups (1=Control, 2=Social Norm, 3=Deterrence,
4=Combined) with equal probability using computer-generated random numbers

3. Implementation: The Tax Authority executes the randomization algorithm and
generates the treatment assignment list. Each firm receives a unique treatment
code determining which message variant they receive.

4. Documentation: The randomization seed and complete assignment list are stored
securely by the Tax Authority. Researchers receive only anonymized data with
treatment codes (no firm identifiers).

Fixed Seeds (for reproducibility):
- Phase 1 (May 2026): seed = 12345
- Phase 2 (July-August 2026): seed = 123452

Timing: Randomization occurs on the same day as message dispatch to minimize
time between assignment and treatment delivery.

Unit of randomization: Individual firm (identified by NIT - Número de
Identificación Tributaria)

Level: Firm-level randomization. Each firm is independently assigned to one of
the four treatment groups within its stratum. There is no group-level or
cluster-level randomization. Each firm is randomized individually, even if
multiple firms share the same address, ownership, or accountant.

Stratification note: While randomization occurs at the firm level, it is
stratified by firm characteristics to ensure balance across groups. Stratification
does not constitute a separate level of randomization—all randomization occurs
at the individual firm level within strata.

Unit of analysis: Matches unit of randomization (firm-level). All outcome
measures (filing, payment) are recorded at the firm level from administrative
records.

Phase 2 exclusions: Firms assigned in Phase 1 are explicitly excluded from
Phase 2 randomization to prevent double-treatment. This is implemented by
filtering the Phase 1 firm list (by NIT) from the Phase 2 eligible population
before randomization.

Quality checks: After randomization, balance checks verify that observable
characteristics (firm size, sector, taxpayer type, age, prior compliance history,
debt amounts) are statistically similar across the four treatment groups.

Unit of randomization: Individual firm (identified by NIT - Número de
Identificación Tributaria)

Level: Firm-level randomization. Each firm is independently assigned to one
of the four treatment groups within its stratum.

No cluster randomization: There is no group-level or cluster-level randomization.
Each firm is randomized individually, even if multiple firms share the same
address, ownership, or accountant.

Stratification: While randomization occurs at the firm level, it is stratified
by firm characteristics (taxpayer type × sector × firm size × firm age) to
ensure balance across groups. Stratification does not constitute a separate
level of randomization—all randomization occurs at the individual firm level
within strata.

Unit of analysis: Matches unit of randomization (firm-level). All outcome
measures (filing, payment) are recorded at the firm level from administrative
records.

No

Not applicable - no cluster randomization.

Randomization occurs at the individual firm level. Total sample: ≈7,000 firms
across two phases.

≈7,000 firms across two phases (≈1,750 firms per treatment group)

Total: ≈7,000 firms
- Phase 1 (May 2026): ≈1,000 firms
- Phase 2 (July-August 2026): ≈6,000 firms

By treatment group (expected):
- Group 1 (Control): ≈1,750 firms
- Group 2 (Treatment A): ≈1,750 firms
- Group 3 (Treatment B): ≈1,750 firms
- Group 4 (Combined): ≈1,750 firms

Assumptions: - Significance level: α = 0.05 (two-tailed) - Power: 80% - Baseline filing rate: 3-5% (assumed 3.5% for calculations) - Standard errors clustered at stratification level SCENARIO 1: Phase 1 Only (n≈1,000, ≈250 per group) Primary Outcome: Tax Filing (Binary) - Main effects (Treatment A alone, Treatment B alone): 2.5 percentage points - Interaction effect (complementarity/crowding-out): 3.5 percentage points - Unit: Percentage points change in filing rate - Baseline: 3.5% filing rate - Effect size as % of baseline: 71% increase for main effects Secondary Outcome: Tax Payment (Binary) - Main effects: 2.5 percentage points - Interaction effect: 3.5 percentage points - Unit: Percentage points change in payment rate - Baseline: 2-3% payment rate Secondary Outcome: Amount Paid (Continuous) - Main effects: 0.28 standard deviations - Interaction effect: 0.40 standard deviations - Unit: Standard deviations of amount paid (COP) SCENARIO 2: Both Phases Combined (n≈7,000, ≈1,750 per group) Primary Outcome: Tax Filing (Binary) - Main effects: 1.1 percentage points - Interaction effect: 1.5 percentage points - Unit: Percentage points change in filing rate - Baseline: 3.5% filing rate - Effect size as % of baseline: 31% increase for main effects Secondary Outcome: Tax Payment (Binary) - Main effects: 1.1 percentage points - Interaction effect: 1.5 percentage points - Unit: Percentage points change in payment rate - Baseline: 2-3% payment rate Secondary Outcome: Amount Paid (Continuous) - Main effects: 0.11 standard deviations - Interaction effect: 0.15 standard deviations - Unit: Standard deviations of amount paid (COP) Power Calculation Method: Following Muralidharan, Romero & Wüthrich (2025) methodology for factorial designs: For main effects: MDE = 2.8 × √[(p(1-p)/n)] For interaction effects: MDE = 2.8 × √[2×p(1-p)/n] Where p = baseline compliance rate, n = sample size per group Interpretation: The study is powered to detect effect sizes consistent with the tax compliance literature. For reference: - Hallsworth et al. (2017, UK): 2-5pp effects - Castro & Scartascini (2015, Argentina): 5pp effect - Del Carpio (2022, Peru): 10-20pp effects - Brockmeyer et al. (2019, Costa Rica): 3.4pp effect Scenario 1 MDEs (2.5pp main effects, 3.5pp interaction): Adequate power to detect effects typical in literature Scenario 2 MDEs (1.1pp main effects, 1.5pp interaction): Excellent power to detect policy-relevant impacts, smaller than most published effects