## tarea2025055-print-pdf

## Source details

**Canonical URL:** [tarea2025055-print-pdf](https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025055-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/tar/2025/english/tarea2025055-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/tar/2025/english/tarea2025055-print-pdf.pdf.json)

---

### Preface: mission, scope, and data
- Mission period: December 16, 2024 – January 3, 2025.
- Mission team: Mr. Soren Pedersen (FAD) and Mr. Tobias Gabel Christiansen (FAD short-term expert).
- Financing: Global Public Finance Partnership (GPFP).
- Requestor: Armenia State Revenue Committee (SRC).
- Objective: Estimate Personal Income Tax (PIT) and Social Security Contribution (SSC) gaps based on operational audits for income years 2020-2023.
- Data: PIT/SSC returns from employers for income years 2020-2023 provided by SRC; findings based on SRC-provided data.
- Acknowledgements: SRC support highlighted, with specific recognition of Mr. Arsen Sarikyan, Head of Development and Administration Strategy Programs Department.
- Report status: Final version of draft submitted to Mr. Ashot H. Muradyan, Deputy Chairman, and Mr. Arsen Sarikyan on January 3, 2025.
- Report structure: Executive Summary; (I) Introduction; (II) Key Findings; (III) Next Steps.

### Executive summary: headline estimates and scope
- Methodology: Bottom-up approach using operational audits (2020-2023). Compliance gap only (policy gap and payment gap excluded).
- Average yearly PIT gap (2020-2023): 14.5 billion AMD or 3.8 percent of potential PIT liability.
- Average yearly SSC gap (2020-2023): 0.6 billion AMD or 1.6 percent of potential SSC liability.
- Scope limitation: Estimates cover compliance gaps associated to registered corporations and dependent workers included in their payrolls. Compliance gaps associated to non-registered businesses and/or informal workers are out of scope.
- Year-by-year PIT gap (percent of potential PIT liability):
  - 2020 = 2.6 percent
  - 2021 = 3.1 percent
  - 2022 = 3.8 percent
  - 2023 = 5.5 percent
- Year-by-year SSC gap (percent of potential SSC liability):
  - 2020 = 2.8 percent
  - 2021 = 1.5 percent
  - 2022 = 0.7 percent
  - 2023 = 1.8 percent
- PIT gap as percent of average GDP (2020–2023): 0.2 percent.
- PIT gap as percent of GDP by year:
  - 2020 = 0.1 percent
  - 2021 = 0.2 percent
  - 2022 = 0.2 percent
  - 2023 = 0.3 percent
- SSC gap as percent of average GDP (2020–2023): 0.003 percent; SSC gap estimated to be 0.002–0.003 percent of GDP in all years covered.
- Absolute PIT gap by year (2020-prices):
  - 2020 = 8.9 billion AMD
  - 2021 = 10.8 billion AMD
  - 2022 = 14.4 billion AMD
  - 2023 = 24.1 billion AMD
- Absolute SSC gap by year:
  - 2020 = 0.5 billion AMD
  - 2021 = 0.5 billion AMD
  - 2022 = 0.3 billion AMD
  - 2023 = 1 billion AMD
- Sensitivity to large firms: PIT gap as percent of potential PIT excluding largest corporations:
  - 2020 = 0.6 percent
  - 2021 = 1.4 percent
  - 2022 = 2.8 percent
  - 2023 = 4.1 percent
- Next-step recommendation summary:
  - SRC adopts the methodology and continues PIT/SSC gap estimation in 2025 and subsequent years.
  - SRC data analysts equipped with R code and detailed instructions.
  - SRC may seek IMF capacity development (CD) for assisted self-assessment in winter 2025/2026.

### Recommendations (summary and timing)
- Primary recommendation: SRC should adopt the methodology and continue to carry out the PIT/SSC gap estimation in the coming years.
- Implementation options:
  - SRC data analysts can perform the analysis using provided R code and detailed instructions.
  - Alternatively, SRC can seek IMF capacity development (CD) for an assisted self-assessment based on operational audits in winter 2025/2026.
- Suggested target date for action: January 2026.

### Introduction: institutional and methodological context
- SRC request: Chairman of the Armenian State Revenue Committee requested IMF assistance to measure PIT and SSC gaps; confirmed at the 2024 IMF-World Bank Spring Meetings.
- System context: Armenia operates a well-established withholding tax system by employers; final PIT/SSC withheld by employers, reported electronically, and paid monthly to SRC.
- Employer reporting: Tax Agents report names and unique identifiers of employees and provide wide-ranging information in their reports.
- Audit data availability: SRC possesses extensive audit results related to withholding PIT and SSC and detailed taxpayer characteristics, enabling granular tax gap estimates by sector, company size, age, etc.
- Audit selection bias and correction: Non-random selection of audited corporations creates Sample Selection Bias; the Heckman two-step method (Heckman, 1979) is used to correct for this bias.
- Audit type used: Estimates based on comprehensive audits covering all aspects of taxpayers; thematic audits and thematic inspections are not used for gap estimation.
- Audit coverage and limitations:
  - Total comprehensive audits used (2020–2023): 4,098 audits.
  - Thematic audits impose penalties for unregistered workers (250,000 AMD per unregistered worker) but do not report PIT adjustments and lack available risk criteria, making them unsuitable for the Heckman method.
  - Compliance gaps for non-registered workers excluded; SRC used a labor input method separately and reported consistent findings (IMF team did not assess that method).
- Note on audit periods: Approximately 75 percent of audits conducted between 2020 and 2022 cover a period of 10 months or more; some audits cover shorter periods. The report does not adjust for these variations.

### Data on audits (summary)
- Number of audits by year:
  - 2020 = 2,656
  - 2021 = 2,373
  - 2022 = 2,044
  - 2023 = 1,262
- PIT assessed amounts by year:
  - 2020 = 1.3 billion AMD
  - 2021 = 900 million AMD
  - 2022 = 1.4 billion AMD
  - 2023 = 1 billion AMD
- SSC assessed amounts by year:
  - 2020 = 6 million AMD
  - 2021 = 21 million AMD
  - 2022 = 12 million AMD
  - 2023 = 125 million AMD
- Audit coverage shares:
  - Percentage of companies audited — 2020 = 3.5 percent; 2021 = 3.3 percent; 2022 = 3.0 percent; 2023 = 2.8 percent.
  - Percentage of revenues audited — 2020 = 55.8 percent; 2021 = 55.7 percent; 2022 = 53.6 percent; 2023 = 50.5 percent.
  - Example: In 2023, 1,262 companies audited reported approximately 210 billion AMD in PIT, amounting to 50.5% of total PIT of 413 trillion AMD in 2023.

### Methodological details and caveats
- Estimation approach: Heckman Sample Selection model (two-stage):
  - First-stage probit for audit selection using 30 summed risk scores and risk rankings used by SRC.
  - Second-stage models audit outcome (PIT/SSC uncovered) using company characteristics and a selection regressor.
- Risk scores provided: SRC provided access to five summed risk scores and risk rankings for PIT, CIT, and VAT for income years 2019–2023; underlying score details were not disclosed to protect audit selection integrity.
- Applicability conditions: Heckman estimator suitability depends on audit practices—comprehensive audits make Heckman appropriate; narrowly focused audits or targeted sector audits reduce reliability and may cause underestimation.
- Uncertainty notes:
  - Estimates for 2023 more uncertain due to lower number of audits and ongoing SRC auditing of 2023 returns.
  - SSC gap estimates show larger variation across years due to significantly fewer SSC corrections, increasing uncertainty.

### Key findings — PIT gap (highlights)
- Average yearly PIT gap (2020–2023): 14.5 billion AMD (measured in 2020-prices) or 3.8 percent of potential PIT liability; corresponds to 0.2 percent of average GDP across 2020–2023.
- PIT gap by year (amount and percent of potential PIT liability, measured in 2020-prices):
  - 2020: 8.9 billion AMD; 2.6 percent.
  - 2021: 10.8 billion AMD; 3.1 percent.
  - 2022: 14.4 billion AMD; 3.8 percent.
  - 2023: 24.1 billion AMD; 5.5 percent (corresponds to 0.3 percent of GDP in 2023).
- PIT gap trend: Increased from 2020 to 2023 but remains relatively low compared to the CIT gap.
- Drivers of increase:
  - Growth in newly established firms with higher relative PIT gaps.
  - Number of PIT-liable firms grew from 65,193 (2020) to 87,854 (2023).
  - PIT gap for firms less than five years old:
    - 2020 = 1.6 billion AMD (3.3 percent of potential PIT)
    - 2023 = 12.6 billion AMD (13.2 percent of potential PIT)
  - Newly established firms accounted for 11 billion AMD of the 15.2 billion AMD increase in the PIT gap from 2020 to 2023 (72 percent of total increase).
- Sectoral concentration: PIT gap as percent of total PIT highest in “Wholesale”; second-highest in “Accommodation”; lowest among top-10 sectors in “Professional, Scientific and Technical Activities.”

### PIT gap by firm size and age
- Relative PIT gap (average across 2020-2023) by number of employees:
  - 0–5 employees: 31.98 percent of potential tax
  - 6–10 employees: 9.86 percent of potential tax
  - 10–50 employees: 3.13 percent of potential tax
  - 50–100 employees: 1.09 percent of potential tax
  - >100 employees: 0.25 percent of potential tax
- Interpretation:
  - Collusion between employers and employees is more prevalent in smaller corporations, while it is less problematic in larger corporations.
  - Similar patterns are observed in Norway.
  - Kleven et al (2016) argue that collusive tax evasion is very hard in large firms.
- Relative PIT gap (average across 2020-2023) by corporation age:
  - 0–2 years: 13.96 percent of potential tax
  - 2–5 years: 4.86 percent of potential tax
  - 5–9 years: 3.27 percent of potential tax
  - 10–19 years: 3.46 percent of potential tax
  - 20+ years: 2.34 percent of potential tax
- Aggregate PIT estimates (Heckman model, excluding largest corporations):
  - PIT gap in 2023 (excluding largest corporations): 12.8 billion AMD
  - PIT gap in 2023 (excluding largest corporations): 4.1 percent of potential PIT
  - PIT gap in 2023 corresponds to: 0.2 percent of GDP
  - Note on exclusion: Top 0.1 percent largest companies (measured by average number of employees) were excluded before estimating the Heckman model and making predictions. Excluding largest corporations can reduce noise from outliers while still accounting for most of the population; the PIT gap remains similar after excluding the largest corporations.

### SSC gap results
- Average yearly SSC gap across 2020-2023: 0.6 billion AMD
- Average yearly SSC gap across 2020-2023: 1.6 percent of potential SSC liability
- SSC gap corresponds to: 0.0 percent of average GDP across 2020-2023
- SSC gap in 2023: 1 billion AMD
- SSC gap in 2023: 1.8 percent of potential PIT liability
- SSC gap in 2023 corresponds to: 0.0 percent of GDP
- Year-by-year SSC gap (amount and percent of potential SSC liability / potential PIT as presented):
  - 2020: 0.5 billion AMD or 2.8 percent of potential SSC liability
  - 2021: 0.5 billion AMD or 1.5 percent of potential PIT
  - 2022: 0.3 billion AMD or 0.7 percent of potential SSC
- Using Heckman model excluding largest corporations:
  - SSC gap in 2023: 0.3 billion AMD
  - SSC gap in 2023: 0.5 percent of potential PIT
  - SSC gap in 2023 corresponds to: 0.0 percent of GDP
- Interpretation:
  - SSC gap indicates social security contributions are not a significant compliance problem.
  - SSC gap estimates are associated with greater uncertainty because significantly fewer companies have been found with errors in SSC controls compared to PIT.

### High-risk sectors and firm characteristics: implications for compliance strategy
- Sectors with notable PIT non-compliance risk include: human health, construction, and transporting sectors.
- Corporations with few employees and younger corporations show higher PIT gaps.
- Understanding whether non-compliance stems from:
  - deliberate evasion,
  - lack of knowledge of the tax law, or
  - complicated legislation,
  will inform a compliance improvement plan.

### Next steps and operational recommendations
- Institutional capacity:
  - SRC should appoint 2-3 data analysts responsible for compiling future tax gap estimations.
  - At least 2-3 individuals should be trained in data analytics to sustain the CIT/PIT/SSC gap models.
  - The appointed team should invest around 80 percent of their time on data analytics to perform good data analytics and learn necessary programming tools.
- Data quality:
  - SRC addresses PIT/SSC gaps through implementation of a personal income tax declaration; high-quality data from employers—especially if preprinted on personal tax declaration forms—is crucial.
  - Tackling identified tax gaps will enhance the quality of data used in personal income declaration forms.
- Contingency:
  - If SRC cannot carry out the estimation in 2025 when relevant data are available with its own resources, SRC could seek IMF assistance for an assisted self-assessment of the CIT/PIT/SSC Gaps to develop required data analytical skills and become self-sufficient.

### Methodology: Heckman 2-step estimator and application
- Rationale:
  - Tax gap estimates from non-random risk-based audits are prone to sample selection bias because audited companies are not representative of the general population.
  - The Heckman 2-step estimator corrects for sample selection bias by estimating both the selection process and the level of non-compliance in the same model.
- Model specification (as presented):
  - Outcome equation: Y_i = X_i β + u_i
    - Y_i measures tax uncovered from audit; X_i are company characteristics (lines from the CIT return, sector, number of employees, etc.).
  - Selection equation: S_i = 1[Z_i δ + v_i > 0]
    - S_i indicates audit selection; Z_i are factors determining audit selection (includes 30 summed risk scores and risk rankings used by SRC to target audits).
  - Assumptions: v_i ~ N(0,1); E[u_i | v_i] = γ v_i
  - Inverse Mills ratio γ(·) = φ(·)/Φ(·) enters the outcome equation to correct for selection.
- Implementation notes:
  - Estimate δ via probit on selection equation; compute inverse Mills ratio; regress Y on X and the estimated inverse Mills ratio to obtain β̂.
  - Predicted unconditional expectation of non-compliance for the population: E[Ȳ | X̄]̂ = X̄ β̂.
  - Standard errors: manual 2-step estimation requires bootstrap for correct standard errors.
  - Identification: avoid using identical variables in selection and outcome models; include exclusion restrictions (variables that affect audit selection but not non-compliance). Here, 10 summed risk scores and 10 risk rankings for VAT and CIT are used as exclusion restrictions.

### Empirical model performance and regression results (summary)
- Data used:
  - All data on CIT returns and operational audits from 2020, 2021, 2022, and 2023.
  - Number of observations:
    - Selection model: 306,666
    - Outcome model: 9,532
- Model fit:
  - Selection model McFadden’s Pseudo R2: 0.77
  - Outcome model R2: 0.03
  - Coefficient on inverse Mills ratio (IMR) in outcome model: insignificant, indicating limited sample selection bias.
- Variables:
  - Selection model includes PIT/CIT/VAT scores and rankings (2019–2023), among others; includes sector dummies.
  - Outcome model includes variables such as UnjustPayTotalAmountIncomeSum, RentalTotalAmountIncomeSum, PropertyTotalAmountIncomeSum, UnjustPayIncomeTaxAmountSum, PercentIncomeTaxAmountSum, EmployeesAvg, Age, InsuranceTotalAmountIncomeSum, OthertaxbaseSum, RentalIncomeTaxAmountSum, UnjustPayNoRecipientsIncomeSum, HoursworkedSum, and PIT score/ranking variables; includes sector dummies.
- Note on statistical inference:
  - Standard errors computed using bootstrap with resampling at the TIN-level.
  - ***, **, * denotes p < 0.01, p < 0.05, p < 0.10.

*Source: IMF calculations based on data from SRC.*

### Preface ____________________________________________________________________________ 6

### Preface ____________________________________________________________________________ 6

### Preface: mission, scope, and data
- Mission period: December 16, 2024 – January 3, 2025.  
- Mission team: Mr. Soren Pedersen (FAD) and Mr. Tobias Gabel Christiansen (FAD short-term expert).  
- Financing: Global Public Finance Partnership (GPFP).  
- Requestor: Armenia State Revenue Committee (SRC).  
- Objective: Estimate Personal Income Tax (PIT) and Social Security Contribution (SSC) gaps based on operational audits for income years 2020-2023.  
- Data: PIT/SSC returns from employers for income years 2020-2023 provided by SRC; findings based on SRC-provided data.  
- Acknowledgements: SRC support highlighted, with specific recognition of Mr. Arsen Sarikyan, Head of Development and Administration Strategy Programs Department.  
- Report status: Final version of draft submitted to Mr. Ashot H. Muradyan, Deputy Chairman, and Mr. Arsen Sarikyan on January 3, 2025.  
- Report structure: Executive Summary; (I) Introduction; (II) Key Findings; (III) Next Steps.

### Executive Summary: headline estimates and scope
- Methodology: Bottom-up approach using operational audits (2020-2023). Compliance gap only (policy gap and payment gap excluded).  
- Average yearly PIT gap (2020-2023): 14.5 billion AMD or 3.8 percent of potential PIT liability.  
- Average yearly SSC gap (2020-2023): 0.6 billion AMD or 1.6 percent of potential SSC liability.  
- Scope limitation: Estimates cover compliance gaps associated to registered corporations and dependent workers included in their payrolls. Compliance gaps associated to non-registered businesses and/or informal workers are out of scope.  
- Year-by-year PIT gap (percent of potential PIT liability): 2020 = 2.6 percent; 2021 = 3.1 percent; 2022 = 3.8 percent; 2023 = 5.5 percent.  
- Year-by-year SSC gap (percent of potential SSC liability): 2020 = 2.8 percent; 2021 = 1.5 percent; 2022 = 0.7 percent; 2023 = 1.8 percent.  
- PIT gap as percent of average GDP (2020–2023): 0.2 percent.  
- PIT gap as percent of GDP by year: 2020 = 0.1 percent; 2021 = 0.2 percent; 2022 = 0.2 percent; 2023 = 0.3 percent.  
- SSC gap as percent of average GDP (2020–2023): 0.003 percent; SSC gap estimated to be 0.002–0.003 percent of GDP in all years covered.  
- Absolute PIT gap by year (2020-prices): 2020 = 8.9 billion AMD; 2021 = 10.8 billion AMD; 2022 = 14.4 billion AMD; 2023 = 24.1 billion AMD.  
- Absolute SSC gap by year: 2020 = 0.5 billion AMD; 2021 = 0.5 billion AMD; 2022 = 0.3 billion AMD; 2023 = 1 billion AMD.  
- Sensitivity to large firms: PIT gap as percent of potential PIT excluding largest corporations—2020 = 0.6 percent; 2021 = 1.4 percent; 2022 = 2.8 percent; 2023 = 4.1 percent. Excluding largest corporations likely provides a more accurate estimate due to reduced noise from a limited number of large companies.  
- Next-step recommendation summary: SRC adopts the methodology and continues PIT/SSC gap estimation in 2025 and subsequent years; SRC data analysts equipped with R code and detailed instructions; SRC may seek IMF capacity development (CD) for assisted self-assessment in winter 2025/2026.

### Recommendations (summary and timing)
- Primary recommendation: SRC should adopt the methodology and continue to carry out the PIT/SSC gap estimation in the coming years.  
- Implementation options: SRC data analysts can perform the analysis using provided R code and detailed instructions; alternatively, SRC can seek IMF capacity development (CD) for an assisted self-assessment based on operational audits in winter 2025/2026.  
- Suggested target date for action: January 2026.

### Introduction: institutional and methodological context
- SRC request: Chairman of the Armenian State Revenue Committee requested IMF assistance to measure PIT and SSC gaps; confirmed at the 2024 IMF-World Bank Spring Meetings.  
- System context: Armenia operates a well-established withholding tax system by employers; final PIT/SSC withheld by employers, reported electronically, and paid monthly to SRC.  
- Employer reporting: Tax Agents report names and unique identifiers of employees and provide wide-ranging information in their reports.  
- Audit data availability: SRC possesses extensive audit results related to withholding PIT and SSC and detailed taxpayer characteristics, enabling granular tax gap estimates by sector, company size, age, etc.  
- Audit selection bias and correction: Non-random selection of audited corporations creates Sample Selection Bias; the Heckman two-step method (Heckman, 1979) is used to correct for this bias.  
- Audit type used: Estimates based on comprehensive audits covering all aspects of taxpayers; thematic audits and thematic inspections are not used for gap estimation.  
- Audit coverage and limitations:  
  - Total comprehensive audits used (2020–2023): 4,098 audits.  
  - Thematic audits impose penalties for unregistered workers (250,000 AMD per unregistered worker) but do not report PIT adjustments and lack available risk criteria, making them unsuitable for the Heckman method.  
  - Compliance gaps for non-registered workers excluded; SRC used a labor input method separately and reported consistent findings (IMF team did not assess that method).  
- Note on audit periods: Approximately 75 percent of audits conducted between 2020 and 2022 cover a period of 10 months or more; some audits cover shorter periods. The report does not adjust for these variations.

### Data on audits (Table 1 summary)
- Number of audits by year: 2020 = 2,656; 2021 = 2,373; 2022 = 2,044; 2023 = 1,262.  
- PIT assessed amounts by year: 2020 = 1.3 billion AMD; 2021 = 900 million AMD; 2022 = 1.4 billion AMD; 2023 = 1 billion AMD.  
- SSC assessed amounts by year: 2020 = 6 million AMD; 2021 = 21 million AMD; 2022 = 12 million AMD; 2023 = 125 million AMD.  
- Audit coverage shares: Percentage of companies audited — 2020 = 3.5 percent; 2021 = 3.3 percent; 2022 = 3.0 percent; 2023 = 2.8 percent. Percentage of revenues audited — 2020 = 55.8 percent; 2021 = 55.7 percent; 2022 = 53.6 percent; 2023 = 50.5 percent. Example: In 2023, 1,262 companies audited reported approximately 210 billion AMD in PIT, amounting to 50.5% of total PIT of 413 trillion AMD in 2023.

### Key findings — PIT gap (highlights)
- Average yearly PIT gap (2020–2023): 14.5 billion AMD (measured in 2020-prices) or 3.8 percent of potential PIT liability; corresponds to 0.2 percent of average GDP across 2020–2023.  
- PIT gap by year (amount and percent of potential PIT liability, measured in 2020-prices):  
  - 2020: 8.9 billion AMD; 2.6 percent.  
  - 2021: 10.8 billion AMD; 3.1 percent.  
  - 2022: 14.4 billion AMD; 3.8 percent.  
  - 2023: 24.1 billion AMD; 5.5 percent (corresponds to 0.3 percent of GDP in 2023).  
- PIT gap trend: Increased from 2020 to 2023 but remains relatively low compared to the CIT gap.  
- Drivers of increase: Growth in newly established firms with higher relative PIT gaps. Number of PIT-liable firms grew from 65,193 (2020) to 87,854 (2023). PIT gap for firms less than five years old: 2020 = 1.6 billion AMD (3.3 percent of potential PIT); 2023 = 12.6 billion AMD (13.2 percent of potential PIT). Newly established firms accounted for 11 billion AMD of the 15.2 billion AMD increase in the PIT gap from 2020 to 2023 (72 percent of total increase).  
- Sectoral concentration: PIT gap as percent of total PIT highest in “Wholesale”; second-highest in “Accommodation”; lowest among top-10 sectors in “Professional, Scientific and Technical Activities.”

*IMF Technical Assistance Report | Preface; Executive Summary; Recommendations; Introduction; Key Findings (PIT) — based on materials provided by SRC and IMF mission analyses.*

### 18.        The PIT gap is large in corporations with few employees (Figure 3). It amounts to 31.98

### 18.        The PIT gap is large in corporations with few employees (Figure 3). It amounts to 31.98

### PIT gap by firm size
- Relative PIT gap (average across 2020-2023) by number of employees:
  - 0–5 employees: 31.98 percent of potential tax
  - 6–10 employees: 9.86 percent of potential tax
  - 10–50 employees: 3.13 percent of potential tax
  - 50–100 employees: 1.09 percent of potential tax
  - >100 employees: 0.25 percent of potential tax
- Interpretation:
  - Collusion between employers and employees is more prevalent in smaller corporations, while it is less problematic in larger corporations.
  - Similar patterns are observed in Norway.
  - Kleven et al (2016) argue that collusive tax evasion is very hard in large firms.

### PIT gap by firm age
- Relative PIT gap (average across 2020-2023) by corporation age:
  - 0–2 years: 13.96 percent of potential tax
  - 2–5 years: 4.86 percent of potential tax
  - 5–9 years: 3.27 percent of potential tax
  - 10–19 years: 3.46 percent of potential tax
  - 20+ years: 2.34 percent of potential tax

### Aggregate PIT estimates (Heckman model, excluding largest corporations)
- PIT gap in 2023 (excluding largest corporations): 12.8 billion AMD
- PIT gap in 2023 (excluding largest corporations): 4.1 percent of potential PIT
- PIT gap in 2023 corresponds to: 0.2 percent of GDP
- Note on exclusion:
  - Top 0.1 percent largest companies (measured by average number of employees) were excluded before estimating the Heckman model and making predictions.
  - Excluding largest corporations can reduce noise from outliers while still accounting for most of the population; the PIT gap remains similar after excluding the largest corporations.

### SSC gap results
- Average yearly SSC gap across 2020-2023: 0.6 billion AMD
- Average yearly SSC gap across 2020-2023: 1.6 percent of potential SSC liability
- SSC gap corresponds to: 0.0 percent of average GDP across 2020-2023
- SSC gap in 2023: 1 billion AMD
- SSC gap in 2023: 1.8 percent of potential PIT liability
- SSC gap in 2023 corresponds to: 0.0 percent of GDP
- Year-by-year SSC gap (amount and percent of potential SSC liability / potential PIT as presented):
  - 2020: 0.5 billion AMD or 2.8 percent of potential SSC liability
  - 2021: 0.5 billion AMD or 1.5 percent of potential PIT
  - 2022: 0.3 billion AMD or 0.7 percent of potential SSC
- Using Heckman model excluding largest corporations:
  - SSC gap in 2023: 0.3 billion AMD
  - SSC gap in 2023: 0.5 percent of potential PIT
  - SSC gap in 2023 corresponds to: 0.0 percent of GDP
- Interpretation:
  - SSC gap indicates social security contributions are not a significant compliance problem.
  - SSC gap estimates are associated with greater uncertainty because significantly fewer companies have been found with errors in SSC controls compared to PIT.

### High-risk sectors and firm characteristics: implications for compliance strategy
- Sectors with notable PIT non-compliance risk include: human health, construction, and transporting sectors.
- Corporations with few employees and younger corporations show higher PIT gaps.
- Understanding whether non-compliance stems from:
  - deliberate evasion,
  - lack of knowledge of the tax law, or
  - complicated legislation,
  will inform a compliance improvement plan.

### Next steps and operational recommendations
- Institutional capacity:
  - SRC should appoint 2-3 data analysts responsible for compiling future tax gap estimations.
  - At least 2-3 individuals should be trained in data analytics to sustain the CIT/PIT/SSC gap models.
  - The appointed team should invest around 80 percent of their time on data analytics to perform good data analytics and learn necessary programming tools.
- Data quality:
  - SRC addresses PIT/SSC gaps through implementation of a personal income tax declaration; high-quality data from employers—especially if preprinted on personal tax declaration forms—is crucial.
  - Tackling identified tax gaps will enhance the quality of data used in personal income declaration forms.
- Contingency:
  - If SRC cannot carry out the estimation in 2025 when relevant data are available with its own resources, SRC could seek IMF assistance for an assisted self-assessment of the CIT/PIT/SSC Gaps to develop required data analytical skills and become self-sufficient.

### Methodology: Heckman 2-step estimator and application
- Rationale:
  - Tax gap estimates from non-random risk-based audits are prone to sample selection bias because audited companies are not representative of the general population.
  - The Heckman 2-step estimator corrects for sample selection bias by estimating both the selection process and the level of non-compliance in the same model.
- Model specification (as presented):
  - Outcome equation: Y_i = X_i β + u_i
    - Y_i measures tax uncovered from audit; X_i are company characteristics (lines from the CIT return, sector, number of employees, etc.).
  - Selection equation: S_i = 1[Z_i δ + v_i > 0]
    - S_i indicates audit selection; Z_i are factors determining audit selection (includes 30 summed risk scores and risk rankings used by SRC to target audits).
  - Assumptions: v_i ~ N(0,1); E[u_i | v_i] = γ v_i
  - Inverse Mills ratio γ(·) = φ(·)/Φ(·) enters the outcome equation to correct for selection.
- Implementation notes:
  - Estimate δ via probit on selection equation; compute inverse Mills ratio; regress Y on X and the estimated inverse Mills ratio to obtain β̂.
  - Predicted unconditional expectation of non-compliance for the population: E[Ȳ | X̄]̂ = X̄ β̂.
  - Standard errors: manual 2-step estimation requires bootstrap for correct standard errors.
  - Identification: avoid using identical variables in selection and outcome models; include exclusion restrictions (variables that affect audit selection but not non-compliance). Here, 10 summed risk scores and 10 risk rankings for VAT and CIT are used as exclusion restrictions.

### Empirical model performance and regression results (summary)
- Data used:
  - All data on CIT returns and operational audits from 2020, 2021, 2022, and 2023.
  - Number of observations: Selection model: 306,666; Outcome model: 9,532.
- Model fit:
  - Selection model McFadden’s Pseudo R2: 0.77
  - Outcome model R2: 0.03
  - Coefficient on inverse Mills ratio (IMR) in outcome model: insignificant, indicating limited sample selection bias.
- Variables:
  - Selection model includes PIT/CIT/VAT scores and rankings (2019–2023), among others; includes sector dummies.
  - Outcome model includes variables such as UnjustPayTotalAmountIncomeSum, RentalTotalAmountIncomeSum, PropertyTotalAmountIncomeSum, UnjustPayIncomeTaxAmountSum, PercentIncomeTaxAmountSum, EmployeesAvg, Age, InsuranceTotalAmountIncomeSum, OthertaxbaseSum, RentalIncomeTaxAmountSum, UnjustPayNoRecipientsIncomeSum, HoursworkedSum, and PIT score/ranking variables; includes sector dummies.
- Note on statistical inference:
  - Standard errors computed using bootstrap with resampling at the TIN-level.
  - ***, **, * denotes p < 0.01, p < 0.05, p < 0.10.

*Source: IMF calculations based on data from SRC.*

---


_Source: https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025055-print-pdf.pdf_