## wpiea2020142-print-pdf

## Source details

**Canonical URL:** [wpiea2020142-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020142-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020142-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020142-print-pdf.pdf.json)

---

### Key findings and policy recommendations
- The operational strength of tax administration agencies—composed of compliance risk management (CRM) practices, use of third-party data, degree of digitalization of services, service orientation, public accountability, and autonomy—is tightly associated with tax collections.
- Among emerging market and low-income economies, countries at the top 25 percent (in terms of operational strength) collect more tax revenues (by 3.25 percent of GDP) than countries at the lowest 25 percent, assuming other conditions are equal.
- Priority reforms to enhance tax collection:
  - (i) Strengthen CRM by adopting automated risk profiling and electronic audits.
  - (ii) Utilize third-party data by adopting computer systems for processing the data and prefilling returns.
- Reform design insight:
  - Many factors are highly correlated; benefits derive from the strategic integration of multiple components rather than isolated measures (e.g., effective operation of an office/program for large taxpayers matters more than mere establishment).
- Staffing and taxpayer base effects:
  - Increased staffing improves revenue performance up to a threshold of 0.25 percent of the labor force.
  - The staff complement of many LIDC tax administrations is below this level.
  - Broadening the taxpayer base matters: the number of active taxpayers (in relation to the labor force) affects tax collection.

### Study scope, data, and methodology
- Dataset: International Survey on Revenue Administration (ISORA).
- ISORA coverage: responses from 135 administrations in 2016, and 159 administrations in 2018; the most recent data used in the paper covers 2017 with 37 Advanced Economies (AEs), 76 Emerging Markets (EMs), and 46 Low-Income Developing Countries (LIDCs).
- ISORA dataset composition: 982 data points, responses to numerical and categorical questions grouped into indices.
- Empirical approach:
  - Two-step approach (Hsiao (2003)) to focus on effects of tax administration while controlling for macroeconomic and tax policy factors.
  - First step: fixed-effect panel regression estimating time-varying control coefficients with model y_{i,t} = β0 X_{i,t} + u_i + γ_t + ε_{it}, where y = tax-to-GDP (excluding trade taxes and social security contributions).
  - Novel proxy for tax policy: “projected” tax revenues (WEO April vintage) and alternative proxies including Budgeted (t) minus Actual (t-1) tax revenue (as share of GDP) and Budgeted (t) minus Budgeted (t-1) tax revenue (as share of projected GDP).
  - Second step: regress estimated country fixed effects û on time-invariant ISORA variables: û = β1 I_i + h_i, using between-effect least squares and weighted least squares to correct heteroskedasticity.
- Caveat: ISORA data for vintages used are not publicly available; inaccuracies may remain despite checks.

### Constructed indices and descriptive metrics
- Operational strength index constructed from seven sub-indices:
  - (i) degree of autonomy; (ii) establishment of LTO or large taxpayer program; (iii) public accountability; (iv) service orientation; (v) CRM approach; (vi) use of third-party data; (vii) digitalization of services. Range 0–1.
- ISORA-derived descriptors:
  - Ratio of full-time-equivalents (FTEs) to labor force (scale of tax administration). Note: Full Time Equivalent (FTE) definition—An FTE of 1 means resources equal to one staff member available for one full year.
  - Ratio of active taxpayers to labor force (proxy for the administration’s workload).
- Stylized cross-country patterns:
  - AEs tend to have higher values across most indices relative to EMs and LIDCs.
  - LTO indicator is binary (0=no LTO, 1=LTO); almost all EMs and LIDCs have LTOs, while just under 80 percent of AEs indicate an LTO or program for large taxpayers.

### Empirical relationships (tax-to-GDP and controls) — First-stage summary (selected coefficients)
- Log (PerCapitaGDP_PPP), lagged:
  - (1) 0.1810*** (0.0402)
  - (2) 0.1374*** (0.0406)
  - (3) 0.1417*** (0.0419)
- Square log (PerCapitaGDP_PPP), lagged:
  - (1) -0.0110*** (0.0024)
  - (2) -0.0084*** (0.0024)
  - (3) -0.0088*** (0.0025)
- Trade openness (out of GDP), lagged:
  - (1) 0.0228*** (0.0058)
  - (2) 0.0236*** (0.0066)
  - (3) 0.0251*** (0.0067)
- Oil exports (as a % of GDP), lagged:
  - (1) 0.1414*** (0.0254)
  - (2) 0.1570*** (0.0224)
  - (3) 0.1452*** (0.0225)
- Standard VAT Rate (%), lagged:
  - (1) 0.0030*** (0.0006)
- Budgeted (t) minus Actual (t-1) tax rev.:
  - (2) 0.1273*** (0.0407)
- Budgeted (t) minus Budgeted (t-1) rev.:
  - (3) 0.1953*** (0.0323)
- Model diagnostics:
  - Observations: (1) 675; (2) 894; (3) 830.
  - R-squared: (1) 0.2301; (2) 0.1984; (3) 0.2441.
  - Number of countries: (1) 105; (2) 127; (3) 125.
  - Country FE: YES; Year FE: YES.

### Second-stage findings — Operational Strength and staffing (selected)
- Operational Strength index (0–1) association with estimated country fixed effects:
  - All Countries examples: 0.2226*** (0.0349); 0.1767*** (0.0358); 0.1228*** (0.0417).
  - EMs and LIDCs examples: 0.1546*** (0.0357); 0.1485*** (0.0377); 0.1277*** (0.0438).
- Tax administration size metrics:
  - #Tax Staff/LaborForce: 0.0802*** (0.0235); 0.0535** (0.0250); 0.0472* (0.0278).
  - Sq(#Tax Staff/LaborForce): -0.0157* (0.0083); additional specs show negative coefficients.
  - ActiveTaxpayer/LaborForce: 0.0443*** (0.0122); 0.0346* (0.0196).
- R-squared examples: 0.2911; 0.4937; 0.5753 across specifications.
- Number of countries examples: 101; 85; 70.
- Interpretation:
  - Operational strength is positively and statistically significantly associated with higher tax collection after controlling for time-varying determinants.
  - Relative size measures are positively associated with tax collection, though squared staff ratio suggests diminishing returns.

### Sub-index results — single-variate and multivariate highlights
- Single-variate coefficients (Table 4 / Appendix VI; all with Number of countries = 101 unless noted):
  - Compliance Risk Management: 0.1154*** (Std. Err. 0.0184); R-squared 0.2835.
  - Third Party Data: 0.1136*** (Std. Err. 0.0181); R-squared 0.2843.
  - Digitalization: 0.0830*** (Std. Err. 0.0197); R-squared 0.1515.
  - Service Orientation: 0.1175*** (Std. Err. 0.0348); R-squared 0.1034.
  - Public Accountability: 0.0874*** (Std. Err. 0.0286); R-squared 0.0860.
  - Autonomy: 0.0809** (Std. Err. 0.0322); R-squared 0.0600.
  - Est. LTO: -0.0245 (Std. Err. 0.0325); R-squared 0.0057 (not significant).
- Multivariate second-stage highlights (Table 5 / Appendix VII):
  - Compliance Risk Management remains significant: e.g., 0.0830*** (Std. Err. 0.0212) across multivariate specifications.
  - Third Party Data remains significant: e.g., 0.0763*** (Std. Err. 0.0199).
  - Digitalization coefficients near zero and often not statistically significant in multivariate specifications.
  - Service Orientation and Public Accountability: small and not consistently significant in multivariate specs.
  - Staffing variables when included:
    - #Tax Staff/LaborForce: positive and significant (e.g., 0.0725***; 0.0641**).
    - Sq(#Tax Staff/LaborForce): negative and significant (e.g., -0.0153*; -0.0177**).
    - ActiveTaxpayer/LaborForce: positive and sometimes significant (e.g., 0.0301** (Std. Err. 0.0132)).
  - R-squared by column examples: 0.4210; 0.5755; 0.6492.
  - Number of countries by column examples: 101; 85; 70.

### Staff-size threshold and quantitative implications
- Marginal benefit of additional tax administration staff decreases with staff size; critical threshold point appears to be 0.25 percent of labor force, beyond which marginal benefit may turn negative.
- Appendix Table 2: Tax gains from increasing staff by 0.01 percent of labor force (Additional Tax as % of GDP):
  - Initial Level of Staff (% of labor force) -> Additional Tax (% of GDP)
    - 0.05  -> 0.67
    - 0.10  -> 0.49
    - 0.15  -> 0.31
    - 0.20  -> 0.13
    - 0.25  -> -0.06
- EMs and LIDCs are generally below the 0.25 percent threshold; additional staffing in these countries is likely associated with higher tax collection.

### Cross-index correlations and reform implications
- High correlations among ISORA sub-indices (2015 and 2017 correlations):
  - Compliance risk management: 0.8677
  - Use of third party data: 0.8439
  - Degree of digitalization: 0.7290
  - Service orientation: 0.8559
  - Public Accountability: 0.8825
  - Autonomy: 0.8319
  - Tax admin staff out of labor force: 0.9561
  - Active taxpayer out of labor force: 0.9267
- Implication:
  - Countries stronger on one sub-index tend to be stronger on others; strategic integration of multiple components is likely to yield larger benefits than isolated reforms.
  - Use of third-party data strongly correlates with CRM and digitalization; Service orientation correlates strongly with most other indices except LTO establishment.

### Large Taxpayer Office (LTO) findings
- Prevalence:
  - 92 percent of tax administrations had an LTO in ISORA dataset; all LIDCs indicate existence of an LTO.
- Statistical relationship:
  - Existence of LTOs did not yield a statistically significant relationship with tax revenue performance in cross-section analysis.
  - Variation in LTO resourcing and operations is large (staff allocation ranges below 1 percent to close to 20 percent; ratio of corporate taxpayers per LTO staff ranges from <1 to almost 200).
- Interpretation:
  - Because LTOs are widespread, insufficient cross-sectional variation exists to identify establishment effects; effective operationalization (e.g., CRM and third-party data use) likely necessary for LTOs to improve revenue performance.

### COVID-19 impacts on tax administration and operational priorities
- Pandemic effects likely to adversely impact tax administration:
  - Slower economic activity reduces taxable base and erodes tax collection.
  - Easing of taxpayer obligations (extensions to filing and payment deadlines) and administration of crisis relief complicate compliance monitoring and non-compliance response.
  - Filing, declaration, and payment compliance may deteriorate due to extended deadlines, limited availability of staff, and weakened taxpayer finances.
  - Face-to-face interactions limited; many administrations shifted from face-to-face audits to desk audits.
  - Administrations with limited digital services and/or remote data access face more severe challenges; staff illness can further undermine collection.
- Post-pandemic priorities and recommendations:
  - Strengthen tax administration to generate fiscal resources for development needs and to safeguard debt sustainability.
  - Prioritize staffing for “critical” areas using a risk management approach if staff shortages occur.
  - Expand digital services where possible.
  - LIDCs face greater challenges due to lower digitalization, weaker CRM, and smaller tax administration staff.

### Robustness checks and additional notes
- Results robust to alternative control variable selection to reduce multicollinearity (dropping variables with correlation > 0.4).
- Findings broadly unchanged when limiting sample to EMs and LIDCs except the quadratic staffing term loses significance in some specifications.
- Standard errors recalculated with alternative methods (Bootstrapping, Huber-White) yield robust outcomes.
- Operational strength remains significant after adding presumed static structural controls (e.g., land-locked, inequality, informality, control of corruption) in second-stage regressions.
- Normality tests (Skewness/Kurtosis) on sub-indices do not reject normality in most cases (selected Prob>chi2 values: Comp. risk mgmt 0.6612; Third party data 0.6934; Digitalization 0.3964; Service orientation 0.1866; Public accountability 0.0888; Autonomy 0.3431).

*Source: wpiea2020142-print-pdf*

### REFERENCES ___________________________________________________________________________________ 27

### REFERENCES

### Key findings and policy recommendations
- The practices and characteristics of tax administration agencies matter significantly for tax performance. The operational strength of the agency—comprising compliance risk management (CRM) practices, the use of third-party data, degree of digitalization of services, service orientation, public accountability, and autonomy—is tightly associated with tax collections.
- Among emerging market and low-income economies, countries at the top 25 percent (in terms of operational strength) collect more tax revenues (by 3.25 percent of GDP) than countries at the lowest 25 percent, assuming other conditions are equal.
- Priority reforms to enhance tax collection:
  - (i) Strengthen CRM by adopting automated risk profiling and electronic audits.
  - (ii) Utilize third-party data by adopting computer systems for processing the data and prefilling returns.
- Reform design insight: Many factors are highly correlated; benefits derive from the strategic integration of multiple components rather than isolated measures (e.g., effective operation of an office/program for large taxpayers matters more than mere establishment).
- Staffing and taxpayer base effects:
  - Increased staffing of a tax administration agency improves revenue performance up to a threshold of 0.25 percent of the labor force.
  - The staff complement of many LIDC tax administrations is below this level.
  - Broadening the taxpayer base matters: the number of active taxpayers (in relation to the labor force) affects tax collection.

### Study scope, data, and methodology
- Dataset: International Survey on Revenue Administration (ISORA).
- ISORA coverage: responses from 135 administrations in 2016, and 159 administrations in 2018; the most recent data used in the paper covers 2017 with 37 Advanced Economies (AEs), 76 Emerging Markets (EMs), and 46 Low-Income Developing Countries (LIDCs).
- ISORA dataset composition: 982 data points, responses to numerical and categorical questions grouped into indices.
- Empirical approach: two-step approach (as proposed by Hsiao (2003)) to focus on effects of tax administration while controlling for other factors (e.g., macroeconomic developments and tax policy changes).
- Caveat: ISORA data are not in the public domain for these vintages; inaccuracies may remain despite checks. (From ISORA 2020 onwards, all data will be publicly available.)

### Constructed indices and key descriptive statistics
- Operational strength index constructed from seven sub-indices:
  - (i) degree of autonomy;
  - (ii) establishment of LTO or large taxpayer program;
  - (iii) public accountability;
  - (iv) service orientation;
  - (v) CRM approach;
  - (vi) use of third-party data;
  - (vii) digitalization of services.
- ISORA-derived descriptors:
  - Ratio of full-time-equivalents (FTEs) to labor force (scale of tax administration). Note: Full Time Equivalent (FTE) definition—An FTE of 1 means resources equal to one staff member available for one full year.
  - Ratio of active taxpayers to labor force (proxy for the administration’s workload).
- Stylized cross-country patterns:
  - AEs tend to have higher values across most indices relative to EMs and LIDCs (radar diagram and box-and-whisker plots).
  - LTO indicator is binary (0=no LTO, 1=LTO); almost all EMs and LIDCs have LTOs, while just under 80 percent of AEs indicate an LTO or program for large taxpayers.
  - The spread in the operational strength index shows mean and median values for AEs are greater than for EMs and LIDCs.

### Literature context and evidence gaps
- Existing literature links per capita GDP positively with revenue; agriculture’s share often negatively correlated with central government revenue; trade openness correlates with revenue. (References: Gupta (2007); Yohou (2017); Boukbech (2018); IMF (2018).)
- Socio-political factors matter: Gini coefficient appears negatively correlated with revenue; spending on education correlates positively; among countries with similar incomes, those with the lowest levels of corruption collect four more percentage points of GDP in tax revenues than those with the highest corruption levels (IMF, 2019).
- Empirical literature specifically on tax administration practices and characteristics and their effect on revenue is limited due to lack of comparable cross-country data; Crivelli (2018) and selected case studies (e.g., Pomeranz (2015); Almunia and Lopez-Rodriquez (2018)) examine aspects such as third-party data and LTOs but often lack controls for macroeconomic and policy confounders.
- This paper addresses the gap using ISORA to control for macroeconomic environments, tax policy, and socio-political factors.

### Operational implications and recommended focus areas for reformers
- Strengthen CRM capabilities, including automated risk profiling and electronic audits.
- Invest in systems and processes to use third-party data effectively and prefill returns.
- Advance digitalization of services to support compliance and reduce corruption potential.
- Consider integrated reform packages rather than isolated technical fixes; focus on effective operation of programs/offices (e.g., LTOs) rather than merely creating them.
- Assess staffing relative to the labor force and consider increasing staff up to a threshold of 0.25 percent of the labor force where needed.
- Broaden the active taxpayer base through policy and administrative measures that bring more taxpayers into regular interaction with the tax administration.

*Source: wpiea2020142-print-pdf*

### 14.      The tax-to-GDP   ratio is positively correlated with the operational strength index, as

### 14.      The tax-to-GDP   ratio is positively correlated with the operational strength index, as

### Key empirical relationships between tax administration features and tax-to-GDP
- The tax-to-GDP ratio (excluding trade taxes and social security contributions) is positively correlated with:
  - The operational strength index (an average of seven ISORA sub-indices, range 0–1).
  - The ratio of tax administration staff (FTEs) to the labor force.
  - The ratio of active taxpayers to the labor force.
- In countries with a higher active taxpayer-to-labor-force ratio, tax administrations are likely to have higher compliance risk management and third-party data indices.
- Ratios of staff to labor force and active taxpayers to labor force vary by one and two orders of magnitude respectively, with far smaller values in general for LIDCs than for AE or EM countries.

### Covid-19: impacts on tax administration and operational priorities (Box 2)
- Pandemic effects likely to adversely impact tax administration:
  - Slower economic activity reduces taxable base and erodes tax collection.
  - Easing of taxpayer obligations (extensions to filing and payment deadlines) and administration of crisis relief complicate compliance monitoring and non-compliance response.
  - Filing, declaration, and payment compliance may deteriorate due to extended deadlines, limited availability of staff, and weakened taxpayer finances.
  - Face-to-face interactions limited; most administrations shifted from face-to-face audits to desk audits.
  - Administrations with limited digital services and/or remote data access face more severe challenges.
  - Staff illness in already understaffed agencies may further undermine tax collection.
- Post-pandemic priorities and recommendations:
  - Strengthen tax administration to generate fiscal resources for development needs and to safeguard debt sustainability.
  - Prioritize staffing for “critical” areas using a risk management approach if staff shortages occur.
  - Expand digital services where possible.
  - LIDCs face greater challenges due to lower digitalization, weaker CRM, and smaller tax administration staff.

### Empirical methodology (two-step approach)
- Rationale:
  - ISORA data available only for 2014–17 with little within-period variation; fixed-effects panel estimation is not appropriate.
  - Use Hsiao (2003) two-step approach to handle time-invariant tax administration variables.
- First step:
  - Fixed-effect panel regression estimating time-varying control coefficients:
    - Model: y_{i,t} = β0 X_{i,t} + u_i + γ_t + ε_{it}
    - y = tax-to-GDP (excluding trade taxes and social security contributions); X includes time-varying macro controls (GDP, CPI, trade openness, external debt), plus oil export share and agriculture share.
  - Novel proxy for tax policy: “projected” tax revenues (WEO April vintage) and alternative proxies:
    - Budgeted (t) minus Actual (t-1) tax revenue (as share of GDP).
    - Budgeted (t) minus Budgeted (t-1) tax revenue (as share of projected GDP).
    - Top tax rates (from IMF Tax Rate Database, DART) and Standard VAT Rate (%).
- Second step:
  - Regress estimated country fixed effects u_i (from first stage) on time-invariant ISORA variables:
    - Model: û = β1 I_i + h_i
    - I_i = set of time-invariant tax administration practices and characteristics (ISORA sub-indices).
  - Estimation uses panel between-effect with least squares; weighted least square used to correct heteroskedasticity.

### First-stage regression empirical findings (Table 2, dependent variable: Tax revenue/GDP)
- Significant associations (coefficients and significance levels preserved):
  - Log (PerCapitaGDP_PPP), lagged:
    - (1) 0.1810*** (0.0402)
    - (2) 0.1374*** (0.0406)
    - (3) 0.1417*** (0.0419)
  - Square log (PerCapitaGDP_PPP), lagged:
    - (1) -0.0110*** (0.0024)
    - (2) -0.0084*** (0.0024)
    - (3) -0.0088*** (0.0025)
  - Trade openness (out of GDP), lagged:
    - (1) 0.0228*** (0.0058)
    - (2) 0.0236*** (0.0066)
    - (3) 0.0251*** (0.0067)
  - External debt (out of GDP), lagged:
    - (1) -0.0005 (0.0033)
    - (2) -0.0038 (0.0046)
    - (3) -0.0063 (0.0051)
  - CPI, lagged:
    - (1) -0.0094 (0.0119)
    - (2) -0.0348*** (0.0131)
    - (3) -0.0308** (0.0128)
  - Terms of Trade (2000=1), lagged:
    - (1) 0.0020 (0.0038)
    - (2) 0.0021 (0.0039)
    - (3) 0.0038 (0.0039)
  - Oil exports (as a % of GDP), lagged:
    - (1) 0.1414*** (0.0254)
    - (2) 0.1570*** (0.0224)
    - (3) 0.1452*** (0.0225)
  - Log (Agri, %GDP), lagged:
    - (1) 0.0029 (0.0046)
    - (2) -0.0032 (0.0054)
    - (3) -0.0006 (0.0056)
  - Control Corruption, lagged:
    - (1) 0.0056 (0.0044)
    - (2) 0.0027 (0.0047)
    - (3) 0.0047 (0.0049)
  - Top Combined CIT Rate (%), lagged:
    - (3) 0.0002 (0.0003)
  - Top Combined PIT Rate (%), lagged:
    - (3) 0.0002 (0.0002)
  - Standard VAT Rate (%), lagged:
    - (1) 0.0030*** (0.0006)
  - Budgeted (t) minus Actual (t-1) tax rev.:
    - (2) 0.1273*** (0.0407)
  - Budgeted (t) minus Budgeted (t-1) rev.:
    - (3) 0.1953*** (0.0323)
- Model diagnostics:
  - Observations:
    - (1) 675
    - (2) 894
    - (3) 830
  - R-squared:
    - (1) 0.2301
    - (2) 0.1984
    - (3) 0.2441
  - Number of countries:
    - (1) 105
    - (2) 127
    - (3) 125
  - Country FE: YES; Year FE: YES

### Second-stage regression empirical findings (Table 3, dependent variable: estimated country fixed effects)
- Operational Strength index (average of seven ISORA sub-indices, 0–1):
  - Coefficients across specifications (statistical significance preserved):
    - All Countries (columns shown): 0.2226*** (0.0349), 0.1767*** (0.0358), 0.1228*** (0.0417)
    - EMs and LIDCs (columns shown): 0.1546*** (0.0357), 0.1485*** (0.0377), 0.1277*** (0.0438)
- Tax administration size metrics:
  - #Tax Staff/LaborForce:
    - 0.0802*** (0.0235)
    - 0.0535** (0.0250)
    - 0.0472* (0.0278)
    - Additional spec: 0.0487 (0.0366)
  - Sq(#Tax Staff/LaborForce):
    - -0.0157* (0.0083)
    - -0.0130 (0.0087)
    - -0.0058 (0.0112)
    - -0.0110 (0.0171)
  - ActiveTaxpayer/LaborForce:
    - 0.0443*** (0.0122)
    - 0.0346* (0.0196)
- Constants (examples):
  - -0.1187*** (0.0222)
  - -0.1444*** (0.0235)
  - -0.1122*** (0.0273)
  - -0.0729*** (0.0215)
  - -0.1027*** (0.0245)
  - -0.0965*** (0.0297)
- Model diagnostics:
  - R-squared values across specifications: 0.2911, 0.4937, 0.5753, 0.1935, 0.3672, 0.3941
  - Number of countries across specifications: 101, 85, 70, 80, 64, 54
- Interpretation:
  - The operational strength index is positively and statistically significantly associated with higher tax collection (tax-to-GDP) after controlling for time-varying determinants.
  - Relative size measures (#Tax Staff/LaborForce and ActiveTaxpayer/LaborForce) are positively associated with tax collection, though squared staff ratio suggests diminishing returns at higher staffing levels.
  - Results are robust when sample restricted to emerging markets and low-income developing countries (EMs and LIDCs).

### Cross-index correlations and implications for reform
- High correlations exist among ISORA sub-indices:
  - Countries stronger on one sub-index tend to be stronger on others.
  - Service orientation index correlates strongly with all other indices (except LTO establishment).
  - Use of third-party data strongly correlates with compliance risk management and degree of digitalization.
- Implication for reform:
  - Major benefits of administrative reform arise from strategic integration of multiple components rather than isolated measures.
  - Correlations complicate isolating which specific practices most directly enhance revenue performance; multicollinearity necessitates complementary approaches (aggregate index, single-index regressions, and all-index regressions).

*Source: wpiea2020142-print-pdf*

### 25.      The square term of tax administration staff controls for the non-linear relationship

### 25.      The square term of tax administration staff controls for the non-linear relationship

### Key empirical findings on operational strength and tax collections
- Among EMs and LIDCs, a country at the top 25 percentile threshold in terms of operational strength collects larger tax revenues by 3.25 percent of GDP than a country at the lowest 25 percent threshold, assuming other conditions are equal.
- The operational strength index is closely associated with tax collections; stronger operational capacity is associated with significantly more tax revenues in EMs and LIDCs.
- Six out of seven ISORA sub-indices are positive and statistically significant when regressed separately on estimated country fixed effects; the establishment of the LTO was not found to be significant.
- Compliance Risk Management (CRM) and Third Party Data usage show stronger correlation with tax collections than other sub-indices.
- The share of tax staff out of labor force is positively associated with tax collection while its quadratic term is negatively associated, indicating decreasing marginal returns to additional staff and an optimal staffing threshold.

### Quantitative results from sub-index single-variate regressions (Table 4 highlights)
- Comp. Risk Management coefficient: 0.1154*** (Std. Err. 0.0184); R-squared 0.2835; Number of countries 101.
- Third Party Data coefficient: 0.1136*** (Std. Err. 0.0181); R-squared 0.2843; Number of countries 101.
- Digitalization coefficient: 0.0830*** (Std. Err. 0.0197); R-squared 0.1515; Number of countries 101.
- Service Orientation coefficient: 0.1175*** (Std. Err. 0.0348); R-squared 0.1034; Number of countries 101.
- Public Accountability coefficient: 0.0874*** (Std. Err. 0.0286); R-squared 0.0860; Number of countries 101.
- Autonomy coefficient: 0.0809** (Std. Err. 0.0322); R-squared 0.0600; Number of countries 101.
- Est. LTO coefficient: -0.0245 (Std. Err. 0.0325); R-squared 0.0057; Number of countries 101.
- Note: *, **, *** denote statistical significance at the 1, 5, and 10 percent levels, respectively.

### Multivariate second-stage results (Table 5 highlights)
- Compliance Risk Management coefficients: 0.0830***; 0.0772***; 0.0873*** (Std. Err. 0.0212; 0.0226; 0.0247 across columns).
- Third Party Data coefficients: 0.0763***; 0.0607***; 0.0612** (Std. Err. 0.0199; 0.0206; 0.0237).
- Digitalization coefficients: 0.0228; 0.0098; 0.0016 (Std. Err. 0.0215; 0.0220; 0.0225).
- Service Orientation coefficients: -0.0297; -0.0295; -0.0425 (Std. Err. 0.0438; 0.0432; 0.0472).
- Est. LTO coefficients: -0.0087; -0.0162; 0.0123 (Std. Err. 0.0271; 0.0271; 0.0295).
- Public Accountability coefficients: -0.0033; 0.0112; -0.0180 (Std. Err. 0.0314; 0.0318; 0.0340).
- Autonomy coefficients: 0.0160; 0.0140; 0.0103 (Std. Err. 0.0315; 0.0319; 0.0331).
- #Tax Staff/LaborForce coefficients: 0.0725***; 0.0641** (Std. Err. 0.0228; 0.0242).
- Sq (#Tax Staff/LaborForce) coefficients: -0.0153*; -0.0177** (Std. Err. 0.0081; 0.0086).
- ActiveTaxpayer/LaborForce coefficient (column 3): 0.0301** (Std. Err. 0.0132).
- R-squared by column: 0.4210; 0.5755; 0.6492.
- Number of countries by column: 101; 85; 70.
- Interpretation: CRM and Third Party Data remain significant when sub-indices and quantitative variables are included together; quadratic staffing term negative implies optimal staffing level.

### Staff-size threshold and implications
- The marginal benefit of additional tax administration staff decreases with staff size; among all countries the critical threshold point seems to be 0.25 percent of labor force, beyond which the marginal benefit of additional staffing may turn negative.
- EMs and LIDCs have not passed this threshold; additional staffing in these countries is likely associated with higher tax collection.

### Policy recommendations and priorities
- Prioritize enhancement of compliance risk management (CRM): adopt automated risk profiling and electronic audits.
- Prioritize utilization of third-party data: adopt computer systems for processing third-party data and prefilling returns.
- Strengthening CRM and third-party data systems should be prioritized given limited tax administration resources; these actions deliver stronger tax collection and facilitate effective LTO functioning.
- Broaden the taxpayer base where active taxpayers/labor force is low; identify whether low active taxpayer ratios stem from tax administration issues or tax policy (e.g., exemptions).
- During pandemic-related revenue risks, preserve taxpayer compliance through third-party data and digital technology and, where agencies are understaffed, preserve existing staff to safeguard revenue collections.

### Robustness checks and additional notes
- Results robust to alternative control variable selection to reduce multicollinearity (dropping variables with correlation > 0.4).
- Findings broadly unchanged when limiting sample to EMs and LIDCs except the quadratic staffing term loses significance.
- Keeping same sample size across first-stage regressions yields similar outcomes; tax base inclusion does not materially affect results.
- Standard errors recalculated with alternative methods (Bootstrapping, Huber-White) yield robust outcomes.
- Operational strength remains significant after adding presumed static structural controls (e.g., land-locked, inequality, informality, control of corruption) in second stage regressions.
- Some additional variables: Gini coefficient and landlocked dummy found insignificant; control of corruption shows significant association with tax collections but was excluded from baseline due to high correlation with other controls.

*Source: wpiea2020142-print-pdf, International Monetary Fund (excerpts as provided).*

### Appendix I. Sub-Indices Compiled to Reflect Facets of Revenue Administration

### Appendix I. Sub-Indices Compiled to Reflect Facets of Revenue Administration

### Sub-indices and component weights
- Compliance risk management (Weight: 3)
  - Formal approach for identifying; Assessing and prioritizing key compliance risks; Automated risk profiling; electronic audit
  - Component weights: 1/3, 1/3, 1/3
- Use of third-party data (Weight: 12)
  - Computer based information systems for processing various forms of third-party data:
    - Financial institutions, International exchange, Online trading, Wage and Salary, Insurance company, Property sale, Other government agencies, Asset leasing, Prescribed contractors with report of payment, VAT invoices, and Others
    - Component weights for listed items: each 1/22 except final item 1/2 (Use of third-party data in prefilling returns is weighted 1/2)
- Degree of digitalization (Weight: 10)
  - E-filing mandatory for some/all taxpayers (1/3)
  - E-payment mandatory for some/all taxpayers (1/3)
  - Web-based information and communication services; Tools and calculators on the webpage; Online application for taxpayer; Capture data from third parties; Digital mailbox for communication with taxpayers; Information on the webpage; Integrated taxpayer account; Electronic invoicing system; Others (each 1/24)
- Service orientation (Weight: 14)
  - Measures to facilitate taxpayer compliance and improve services to taxpayers:
    - Have a formal service strategy (1/14)
    - Have a formal set of service standards (1/14)
    - Conduct taxpayer satisfaction surveys (1/14)
    - Registration possible through other agencies (1/14)
    - Provision of rulings to taxpayers (1/14)
    - Availability of online application for taxpayers (1/14)
    - Make special provisions for taxpayers with disabilities (1/14)
    - Provision of services in unofficial language(s) (1/14)
    - Use of information on compliance burden (1/14)
    - End-user testing of new services (1/14)
    - End-user involvement in design of new services (1/14)
    - Simultaneous Registration for multiple tax types (1/14)
    - Formal document covering taxpayer rights (1/14)
    - Mechanism for managing taxpayer complaints (1/14)
- Public accountability (Weight: 12)
  - Measures enhancing tax agency’s accountability to the public (each component weight 1/12):
    - Publish strategic plan; Publish annual business/operation plan; Make public formal service delivery standards; Publish achievements vis-a-vis standards; Publish annual reports; Use of an external auditor; Make key compliance risk public; Make reports of outcomes in addressing compliance risk public; Publish results of taxpayer satisfaction surveys; Document that formally set out taxpayer rights; Have specific mechanism for managing taxpayer complaints; Publish periodic estimates of the tax gap
- Autonomy (Weight: 14)
  - Institutional form, degree of autonomy in managing expenditure and human resources (each component weight 1/14):
    - Autonomous vs. operating within Ministry; Discretion over designing internal structure; Discretion over operational budget; Discretion over capital budget; Authority to set performance standards; Determination of work requirements; Appointment of new staff; Promotion of existing staff; Decide on qualifications for appointment; Decision whether work is carried out by permanent or contractual staff; Placement of staff in salary band; Termination of employment; Responsible for debt collection and enforcement; Provision of tax policy advice
- Large Taxpayer Office (LTO) (Weight: 1)
  - An office or a program dedicated to large taxpayers is in operation (1)

- Construction: Each sub-index is constructed by making averages of responses to the binary questions in ISORA. For questions with sub-questions, averages of sub-question responses are used to calculate the sub-index.

### Correlation between sub-indices (2015 and 2017)
- Compliance risk management: 0.8677
- Use of third party data: 0.8439
- Degree of digitalization: 0.7290
- Service orientation: 0.8559
- Public Accountability: 0.8825
- Autonomy: 0.8319
- Est. of LTO: Status the same for 93 percent of tax administrations
- Tax admin staff out of labor force: 0.9561
- Active taxpayer out of labor force: 0.9267

### Cross-section regression outcomes — key findings (Appendix Table 1)
- Operational Strength is positively associated with tax collection and is the single most relevant factor in the regressions.
- Selected coefficient estimates (OLS with robust standard errors; *, **, *** denote statistical significance at the 1, 5, and 10 percent levels):
  - PerCapitaGDP_PPP: 0.0270*** (column 3); 0.0281*** (column 4)
  - Terms of Trade: -0.0254* (column 1)
  - Oil export (as a % of GDP): -0.2606*** (column 4)
  - Operational Strength: 0.1129*** (column 1); 0.0973*** (column 3)
- Observations: 96 (columns 1 and 3) and 122 (columns 2 and 4)
- R-squared: 0.5651 (col 1), 0.4061 (col 2), 0.5373 (col 3), 0.3877 (col 4)
- Note: Period average used for dependent variable (tax to GDP (excluding trade tax)) and independent variables (2010-18) due to limited availability of historical ISORA series.

### Staffing and taxpayer base impact on tax collection (Appendix IV)
- Staff level effect
  - Staff level is positively associated with tax collections, but with diminishing returns; quadratic term negative and significant.
  - Threshold at which marginal increase turns negative: 0.25 percent of labor force.
  - Appendix Table 2: Tax gains from increasing staff by 0.01 percent of labor force:
    - Initial Level of Staff (% of labor force) -> Additional Tax (% of GDP)
      - 0.05  -> 0.67
      - 0.10  -> 0.49
      - 0.15  -> 0.31
      - 0.20  -> 0.13
      - 0.25  -> -0.06
  - EMs and LIDCs generally below the threshold; some AEs pass the threshold. LICs appear understaffed.
- Taxpayer base effect
  - The share of active taxpayer out of labor force is positively and significantly associated with tax collection.
  - Active taxpayer measured as sum of active taxpayers in PIT, CIT, PAYE, and VAT.
  - Advanced economies have larger taxpayer base than emerging economies, followed by low income economies.
  - Many LIDCs have significantly low shares of active taxpayers, indicating scope to expand taxpayer bases.
  - Correlation: the share is correlated with sub-indices for compliance risk management and the use of third-party data; therefore not included in baseline regression but inclusion does not change regression outcomes.

### Second-stage regression outcomes — Operational Strength index (Appendix V)
- All Countries (selected coefficients)
  - Operational Strength: ranges from 0.0901* to 0.2317*** across specifications (standard errors reported)
  - #Tax.Staff/LaborForce: positive and significant in specifications where included (e.g., 0.1059***)
  - Sq(#Tax.Staff/LaborForce): negative and significant where included (e.g., -0.0239***)
  - #ActiveTaxpayer/LaborForce: positive and significant where included (e.g., 0.0329***)
  - R-squared range: 0.2049 to 0.5934 depending on specification
  - Number of countries: between 62 and 101 depending on specification
- Emerging and Low-income Countries (selected coefficients)
  - Operational Strength: ranges from 0.1060** to 0.1585*** across specifications
  - #Tax Staff/LaborForce: positive and in some specs significant (e.g., 0.0861***)
  - Sq(#Tax Staff/LaborForce): not consistently significant
  - ActiveTaxpayer/LaborForce: significance appears in some specs (e.g., 0.0404**)
  - R-squared range: 0.1130 to 0.4333
  - Number of countries: between 46 and 80

### Second-stage regression outcomes — Single variates (Appendix VI)
- Individual sub-index effects on tax outcomes (selected coefficients):
  - Compliance Risk Management: 0.1052***, 0.1154***, 0.1192*** (columns 1–3)
  - Third Party Data: 0.1027***, 0.1136***, 0.1196*** (columns 2–3)
  - Digitalization: 0.0540**, 0.0830***, 0.0853*** (columns 3–5)
  - Service Orientation: 0.1041**, 0.1175***, 0.1242*** (specifications reported)
  - Public Accountability: 0.0661**, 0.0874***, 0.0900*** (specifications reported)
  - Autonomy: 0.0610*, 0.0809**, 0.0791** (specifications reported)
  - Est. LTO: -0.0405, -0.0245, -0.0231 (not statistically significant)
- R-squared examples:
  - Compliance Risk Management specifications: 0.2682, 0.2835, 0.2733
  - Third Party Data specifications: 0.2702, 0.2843, 0.2802
  - Digitalization specifications: 0.0566, 0.1515, 0.1465

### Multivariate second-stage outcomes (Appendix VII) — selected coefficients
- Compliance Risk Management:
  - 0.0816***, 0.0830***, 0.0840*** (columns 1–3)
  - 0.0726***, 0.0772***, 0.0778*** (columns 4–6)
  - 0.0955***, 0.0873***, 0.0930*** (columns 7–9)
- Third Party Data:
  - 0.0747***, 0.0763***, 0.0807*** (columns 1–3)
  - 0.0494**, 0.0607***, 0.0644*** (columns 4–6)
  - 0.0492**, 0.0612**, 0.0675*** (columns 7–9)
- Digitalization: coefficients near zero and not statistically significant in multivariate specifications (e.g., 0.0007, 0.0228, 0.0240)
- Service Orientation and Public Accountability: coefficients small and not consistently significant in multivariate specifications; Service Orientation sometimes negative (e.g., -0.0195)
- Autonomy: small positive coefficients, not consistently significant (e.g., 0.0131)
- Staffing and taxpayer variables (when included):
  - #Tax Staff/LaborForce: positive and significant (e.g., 0.0914***)
  - Sq(#Tax Staff/LaborForce): negative and significant (e.g., -0.0216***)
  - ActiveTaxpayer/LaborForce: positive and in some specs significant (e.g., 0.0320**)
- R-squared by column groups:
  - Columns 1–3: 0.4103, 0.4210, 0.4085
  - Columns 4–6: 0.5922, 0.5755, 0.5870
  - Columns 7–9: 0.6525, 0.6492, 0.6744
- Number of countries: 88, 101, 99, 74, 85, 84, 62, 70, 70 across specifications.

### Large Taxpayer Office (LTO) analysis (Appendix VIII)
- Prevalence:
  - 92 percent of tax administrations had an LTO in ISORA dataset.
  - All LIDCs tax administrations indicate the existence of an LTO.
- Statistical relationship:
  - Existence of LTOs did not yield statistically significant relationship with tax revenue performance in cross-section analysis.
  - Variation in LTO resourcing and operations:
    - Proportion of staff allocated to LTO functions ranges from below 1 percent to close to 20 percent.
    - Ratio of corporate taxpayers managed through LTO per staff member ranges from less than one to almost 200.
  - Characteristics examined (relative size of human resources in LTOs, proportion of “large taxpayers,” proportion of revenue collected through the LTO) were not statistically significant.
- Interpretation:
  - Because LTOs are widespread (92 percent), insufficient cross-sectional variation exists to analyze the establishment impact.
  - Effective operationalization (e.g., improved compliance risk management and use of third-party data) likely necessary for LTOs to improve revenue performance.
  - In some EMs and LIDCs, additional staff in LTO is not necessarily associated with larger revenue collection by LTO.

### Normality test of error terms (Appendix IX)
- Skewness/Kurtosis tests for normality (Sktest outcomes from STATA). Null hypothesis: normal distribution. Selected results:
  - Comp. risk mgmt: Obs 96; Pr(Skewness) 0.3677; Pr(Kurtosis) 0.9966; adj chi2(2) 0.83; Prob>chi2 0.6612
  - Third party data: Obs 96; Pr(Skewness) 0.8722; Pr(Kurtosis) 0.4051; adj chi2(2) 0.73; Prob>chi2 0.6934
  - Digitalization: Obs 96; Pr(Skewness) 0.218; Pr(Kurtosis) 0.5923; adj chi2(2) 1.85; Prob>chi2 0.3964
  - Service orientation: Obs 96; Pr(Skewness) 0.0777; Pr(Kurtosis) 0.6974; adj chi2(2) 3.36; Prob>chi2 0.1866
  - Public accountability: Obs 96; Pr(Skewness) 0.0371; Pr(Kurtosis) 0.4808; adj chi2(2) 4.84; Prob>chi2 0.0888
  - Autonomy: Obs 96; Pr(Skewness) 0.1817; Pr(Kurtosis) 0.5836; adj chi2(2) 2.14; Prob>chi2 0.3431
  - Est. of LTO: Obs 96; Pr(Skewness) 0.1355; Pr(Kurtosis) 0.4218; adj chi2(2) 2.96; Prob>chi2 0.2281

*Source: wpiea2020142-print-pdf - Appendix I. Sub-Indices Compiled to Reflect Facets of Revenue Administration*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020142-print-pdf.pdf_