## 1. Overall Strength of Tax Administration

## Source details

**Canonical URL:** [1. Overall Strength of Tax Administration](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023231-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023231-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023231-print-pdf.pdf.json)

---

### Data
- ISORA (International Survey on Revenue Administration):
  - Surveys TA features through both numerical and categorical questions with 982 data points.
  - Covers 154 countries, including 36 Advanced Economies (AEs), 71 Emerging Markets (EMs), and 46 Low-Income Developing Countries (LIDCs).
  - Four vintages exist; analysis uses ISORA 2016 and ISORA 2018 only.
  - Constructed indexes: an overall operational strength (OS) index as the equally weighted average of seven sub-indexes; seven sub-indexes are:
    - i) compliance risk management (CRM) approach
    - ii) the use of third-party data
    - iii) digitalization of services
    - iv) service orientation
    - v) public accountability
    - vi) the degree of autonomy
    - vii) the establishment of LTO or large taxpayer program
  - Additional ISORA-derived indicators used: ratio of tax staff over the labor force; active taxpayers over the labor force.
- TADAT (Tax Administration Diagnostic Assessment Tool):
  - Provides assessments starting in 2013, covering 86 countries with 11 repeated assessments between 2013 and 2019.
  - Focuses on nine performance outcome areas (POAs): i) integrity of the registered taxpayer base, ii) effective risk management, iii) supporting voluntary compliance, iv) timely filing of declarations, v) timely payment of taxes, vi) accurate reporting in declarations, vii) effective dispute resolution, viii) efficient revenue management, ix) accountability and transparency.
  - Nine POAs are compiled using 32 high-level indicators (HLIs) constructed from 55 measurement dimensions; overall performance assessment report (PAR) index is the equally weighted average of the nine POAs.
- Revenue variable and exclusions:
  - Dependent variable: tax-to-GDP ratio (excluding trade taxes and social security contributions).
  - Exclusions rationale: to account for volatility of commodity prices, exclude resource revenues, and because other agencies than the TA may collect those revenues.

### Methodology
- Two complementary empirical approaches are applied:
  - Cross-country panel analysis:
    - Uses slow-moving indicators of TA strength and accounts for endogeneity of TA reform impact.
    - Specifically applies the Hausman-Taylor cross-country panel regression with endogenous time-invariant variable.
    - Controls for macroeconomic, structural, and tax policy determinants identified in prior literature.
  - Event-study design of reform episodes:
    - Studies TA reform episodes in LIDCs and EMs after the global financial crisis (GFC).
    - Employs three recent extensions to the canonical difference-in-differences (DID) to estimate revenue yields and their evolution over time.
- Additional operational output:
  - A comprehensive database and an operational tool were developed in parallel (Box 1) to identify gaps in TA and systematically estimate contributions of past or future TA reforms.

### Key findings and estimated impacts
- Overall estimated yields:
  - An increase in TA strength moving a country from the 40th percentile to the 60th percentile is associated with:
    - up to 1.4 percentage points (pp.) of GDP increase in tax revenues for all countries.
    - 1.8 pp. of GDP increase in tax revenues for EMs and LIDCs.
  - Event-study estimates for comprehensive TA reforms after the GFC indicate additional revenues between 2 to close to 3 pp. of GDP.
- Timing and dynamics:
  - Revenue yields from far-reaching and comprehensive reforms increase over time, reaching more than 3 pp. of GDP after the 6th year following reform.
  - Significant TA reforms may lead to some initial gains, but full results may take time to materialize.
- Specific TA practices with high revenue correlations:
  - Using ISORA:
    - Compliance risk management improvements have the highest revenue yields.
    - Public accountability and LTO also play a more significant role in increasing revenue in EMs and LIDCs.
  - Using TADAT:
    - Accountability and transparency, and timely filing of tax declarations matter the most for raising revenues.
  - High correlations among ISORA and TADAT sub-indexes indicate TA functions support and reinforce each other.
- Comparative magnitude:
  - Point estimates from cross-country panel analysis are broadly consistent with event-study estimates: cross-country up to 1.4–1.8 pp., event-study between 2 and close to 3 pp., and dynamic effects exceeding 3 pp. by the 6th year for comprehensive reforms.

### Stylized facts on TA strength, practices, and characteristics
- Main overview:
  - Uses ISORA and TADAT datasets to construct aggregate indexes and sub-indexes; focus period: 2010-2019 (assessments after 2020 are not included).
- a) TA strength is slow-moving:
  - ISORA and TADAT indexes "only slightly changed" over the years; TA strength, practices, and characteristics are slow-moving variables.
  - Example: correlations between ISORA indexes for 2014 and 2017 are elevated; "eight out of the ten ISORA’s indexes have a coefficient of correlation higher than 0.8."
  - Exceptional cases with significant improvement: Grenada, St. Vincent and Grenadines, Uruguay, Greece, Ghana, Poland, and Myanmar.
- b) TA strength varies with economic development:
  - Aggregate finding: TA strength is on average much stronger in AEs, while LIDCs have the weakest TA.
  - ISORA detailed patterns:
    - AEs have higher scores than EMs and LIDCs, except for the establishment of LTO.
    - EMs have higher scores on average than LIDCs, except for autonomy and establishment of LTO where EMs and LIDCs are almost similar.
    - Weaknesses in EMs and LIDCs: use of third-party data and digitalization, and compliance risk management.
  - TADAT detailed patterns:
    - Differences between AEs, EMs, and LIDCs are more pronounced than in ISORA.
    - AEs have higher scores across all practices and characteristics, especially for timely filing of tax declarations and timely payments of taxes.
    - EMs score higher than LIDCs on average but have weaker scores for effective risk management, accurate reporting in declarations, timely filing, and efficient revenue management.
    - LIDCs have the lowest scores across all practices and characteristics, especially for timely filing of tax declarations, efficient revenue management, accurate reporting in declarations, and effective risk management.
- c) Heterogeneities by income groups:
  - AEs tend to collect more tax revenues with stronger TA.
  - LIDCs have, on average, weaker TA and collect lower tax revenues.
  - EMs exhibit large variation: some EMs resemble LIDCs in weak TA, others resemble AEs in strong TA.
- Correlation with revenues:
  - "The strength of TA and tax revenues are positively correlated."
  - ISORA and TADAT indexes correlate highly with tax revenues (excluding trade taxes and social contributions), though there is high variance among peer group countries.
- d) Relationship between ISORA’s OS and TADAT’s PAR:
  - The two aggregate indexes are correlated: "the coefficient of correlation is 0.61."
  - TADAT’s PAR is generally lower than ISORA’s OS for almost all countries; the gap is higher for LIDCs.
  - Possible reasons: ISORA is self-reported by authorities; TADAT assessments are by external experts. Different aggregation methodologies: ISORA uses equally weighted additive components; TADAT uses M1 and M2 methods and non-additive MD components.
- e) Correlation among sub-components:
  - High correlations among ISORA and TADAT sub-indexes.
  - Implications:
    - i) Good practices/characteristics in one dimension are generally associated with good practices in others.
    - ii) Reforms should be comprehensive and implemented in an integrated manner.
  - Caveat: High correlations across sub-indexes make it difficult to identify the relative importance of individual practices or characteristics for revenue gains or to analyze their interactions.

### Empirical methodology (summary)
- Slow-moving nature of TA measures:
  - Hypothesis: indexes measuring TA strength and practices remain constant over 2010 to 2019.
  - Practical implementation:
    - ISORA’s OS and seven sub-indexes: use 2017 values for all years in 2010-2019.
    - TADAT’s PAR, POAs, and HLIs: use the latest assessment value (or average when repeated) for all years in the period.
  - Robustness check: re-estimate baseline models over 2015–19 for ISORA only (TADAT lacks sufficient data); results similar for range estimates though point estimates for Models 2 and 3 are lower.
- Estimation approach:
  - Employ model by Chatelain and Ralf (2021) to estimate effects of endogenous time-invariant variables correlated with individual effects.
  - Uses a pretest procedure to determine exogenous internal instruments among time-varying variables; then applies Hausman and Taylor (1981) estimator (HT) if not under-identified.
  - Internally selected time-varying instrumental variables (after pretest): i) terms of trade, ii) the share of agriculture to GDP, iii) control of corruption, iv) the proxies of tax policy.
- Endogeneity concerns and controls:
  - TA practices and characteristics may be endogenous—policy changes can be anticipated and correlated with past economic activity and revenue performance.
  - Control variables X include:
    - i) macroeconomic variables: per capita GDP and its square, inflation, trade openness, external debt to GDP, terms of trade.
    - ii) structure and institutions: oil exports to GDP, share of agriculture to GDP, control of corruption.
    - iii) proxies of tax revenue policy (three approaches described; preferred proxy: difference between budgeted revenues in year (t) and actual revenues of year (t-1), both relative to respective GDP).

### Baseline results and estimated revenue yields
- Results are presented for a sample of all countries (columns 1–3) and EMs and LIDCs only (columns 4–5).
- The overall strength of TA (ISORA’s OS; TADAT’s PAR) is positively and statistically significantly associated with higher tax revenues in all specifications, with slightly higher magnitude in EMs and LIDCs.
- Point estimates for feasible reforms (increase in TA strength equivalent to the change between the 40th percentile and the 60th percentile) show:
  - ISORA’s OS: revenue yields of up to 1.8 pp. of GDP (95 percent confidence interval of 0.5-2.6 pp. of GDP).
  - TADAT’s PAR: revenue yields of up to 1.4 pp. of GDP (range of 0.2-2.1 pp. of GDP).
- Table 1 summary (Percentage points of GDP):
  - All countries, Using ISORA OS: 1.4 (0.5 ‒ 2.2)
  - EMs and LIDCs, Using ISORA OS: 1.8 (1.0 ‒ 2.6)
  - All countries, Using TADAT PAR: 1.3 (0.2 ‒ 2.0)
  - EMs and LIDCs, Using TADAT PAR: 1.4 (0.2 ‒ 2.1)
- Empirical goodness of fit: adjusted R-squared indicates models could explain up to 56 percent of the variation of tax revenues over the period 2010–19.

### Staffing, taxpayer base, and other determinants
- Staffing of TA — nonlinear effects:
  - There is an inverted-U relationship between tax revenues and #Tax Staff/LaborForce for all countries:
    - Marginal return turns negative past a threshold of tax staff between 0.15 to 0.22 percent of the labor force (depending on specification).
  - This inverted-U relationship does not hold in EMs and LIDCs in the sample.
- Taxpayer base and interaction with TA strength:
  - Coefficients for Active Taxpayer/Labor Force are positive but non-significant when controlling for TA strength and staffing; become significant when not controlling for TA strength or staffing.
- Other macroeconomic and structural determinants:
  - Trade openness (lagged) is positively and statistically associated with higher revenues.
  - External debt to GDP shows mixed effects; in some specifications increasing external debt to GDP leads to higher revenues.
  - Oil exports (as a % of GDP), lagged, are positive and statistically significant in multiple specifications.
  - Higher inflation (CPI, lagged) is associated with lower revenues in some specifications.
  - Stronger control of corruption is associated with higher revenues in TADAT specifications.
  - Tax policy proxy (Budgeted_Actual, Nontrade_tax, Combined) is positively and significantly associated with tax revenues.
  - Variables not significantly impacting tax revenues (after controls): real GDP per capita and its squared, terms of trade, share of agriculture in GDP (in many specifications).

### Selected magnitudes from regression tables
- ISORA (Table 2) — Operational Strength, 2017 coefficients (selected):
  - Column (1) All: 0.1333*** (0.0386)
  - Column (4) EMs & LIDCs: 0.1658*** (0.0356)
- TADAT (Table 3) — TADAT: PAR coefficients (selected):
  - Column (1) All: 0.1328*** (0.0319)
  - Column (4) EMs & LIDCs: 0.1376*** (0.0324)
- Selected staffing coefficients (TADAT, Table 3):
  - #Tax Staff/LaborForce in some All-country specifications: 0.1304*** (0.0437)
  - Sq(#Tax Staff/Labor Force) negative and significant in some specifications: -0.0429*** (0.0159)
- Adjusted R-squared examples:
  - ISORA models: Adj. Overall R2 up to 0.4553.
  - TADAT models: Adj. Overall R2 up to 0.5664.
- Sample sizes and country counts (examples from tables):
  - ISORA Table 2 observations: 768, 551, 482, 682, 470, 416
  - ISORA number of countries: 92, 75, 65, 80, 63, 55
  - TADAT Table 3 observations: 536, 315, 299, 518, 297, 281
  - TADAT number of countries: 60, 41, 39, 57, 38, 36

### Robustness, caveats, and empirical considerations
- Robustness checks: findings are robust to dropping outliers/influential data points (coefficients slightly lower but variable of interest remains similar).
- Baseline findings robust to excluding all commodity exporters (results available upon request).
- Caveats on tax policy measure and revenue fluctuations:
  - Revenue fluctuations are not proportional to GDP (commodity price swings, supply shocks, recessions/booms); potential multicollinearity could increase standard errors.
  - Use of estimated actual revenues in year (t-1) could differ from factual revenues; concern only if estimation error is systematic.
  - Nowcasting tax revenues projected in the budget for year (t) can be affected by bias or serial correlation.
- Practical policy caution: TA reforms work best as well-designed, properly sequenced packages accounting for interactions and country-specific constraints; small and isolated improvements may not translate into higher yields and could pose fiscal risk if expected yields are not achieved.
- For fiscal projections, prudence is advised: use the lower econometric bound of estimated yields (95 percent confidence lower bound) to avoid overestimation; lags between reform implementation and results may be country specific.

### TA-AYT: IMF FAD Tax Administration - Assessment and Yield Tool (operational implications)
- TA-AYT is an Excel-based tool developed to estimate contribution of TA reforms and benchmark TA strength across IMF member countries using ISORA (ISORA 2016 and ISORA 2018 vintages) and other sources.
- Structure and modules:
  - Module 1: Benchmarking TA performance over time and comparators; overview of TA strength overall and by practices; correlations with tax revenues.
  - Module 2: Presents goodness of fit of the model for the country of interest and comparators; helps assess predictive accuracy.
  - Module 3: Allows user inputs on potential outcomes and duration of TA reforms to construct new index values and estimate a range and cumulative revenue yields using regression results.
- Use guidelines:
  - TA-AYT helps identify bottlenecks and weaknesses but macro-level analysis may mask country-specific specifics; should be complemented by in-depth country review.
  - For fiscal projection use, prefer the lower 95 percent confidence bound and prepare for upside risk; validate yields with granular, country-specific characteristics.

### Multivariate models — specific components and caveats
- Overview and caveats:
  - Multivariate models include all sub-indexes simultaneously to assess relative importance of practices; two main caveats:
    - High correlation among components (multicollinearity) makes it difficult to distinguish individual impacts.
    - Singling out one component without others risks overestimating its impact (omitted variable problem).
  - Despite caveats, multivariate specifications yield higher adjusted R-squared compared with aggregate-index models.
- Empirical findings from multivariate models:
  - ISORA results:
    - Compliance Risk Management: coefficients range (examples) 0.0555*** (0.0212) to 0.0714*** (0.0256) across columns; consistently positive and statistically significant.
    - Public Accountability: positive and significant in several EMs & LIDCs specifications (e.g., 0.0437* (0.0262)).
    - Est. LTO: significant in EMs & LIDCs (e.g., 0.0688*** (0.0192)).
    - Third Party Data and Digitalization: coefficients not consistently statistically significant.
    - #Tax Staff/LaborForce: positive in some specifications with Sq(#Tax Staff/Labor Force) negative indicating diminishing returns.
  - TADAT results:
    - POA4: Timely Filing of Tax Declarations — positive associations (examples: 0.0635* (0.0328); 0.0786* (0.0417)).
    - POA9: Accountability and Transparency — positive and statistically significant (examples: 0.1081*** (0.0388); 0.1098*** (0.0406)).
    - #Tax Staff/LaborForce: positive with negative square term (e.g., 0.1181*** (0.0429) and Sq = -0.0390** (0.0166)).
    - Trade openness, external debt, oil exports, and Budgeted_Actual often positive and significant.

### Event-study and quasi-experimental evidence (LIDCs and EMs)
- Data and sample:
  - Treatment groups of countries with sustained RA reforms identified using Akitoby et al. (2020); country groups by reform timing: 2008-09, 2010-11, 2012-13; control group: most other EMs and LIDCs.
- Identification challenges addressed using three methods:
  1. Staggered-adoption DID (Callaway and Sant’Anna, 2021) — estimates ATT(g,t) and aggregates by exposure length.
  2. Synthetic Difference-in-Differences with Common Support (SDID) (Arkhangelsky et al., 2021) — constructs synthetic controls with unit and time weights, then applies staggered DID.
  3. Controlled Interrupted Time-series Analysis (CITSA) (Linden, 2015) — verifies parallel trends and estimates level/trend changes with specification:
     - Yt = β0 + β1 Tt + β2 Xt + β3 XtTt + β4 Z + β5 ZTt + β6 ZXt + β7 ZXtTt + ∈t
- Staggered DID and duration effects:
  - First-stage findings: average increase in tax revenues of about 2 pp. of GDP for the three country groups adopting TA reforms.
  - Country group 2008-09 mean ATT: 2.6 pp. of GDP.
  - Country group 2010-11 point estimate: 2.2 pp. of GDP.
  - Country group 2012-13 mean ATT: 2 pp. of GDP.
  - Aggregated duration trajectory (exposure-based):
    - 1st year after TA reforms: about 1.5 pp. of GDP.
    - 3rd year after TA reforms: about 2.6 pp. of GDP.
    - Stabilizes around 2.2 pp. of GDP then increases to just above 3 pp. of GDP after the 6th year.
- SDID aggregated effect:
  - Main aggregated result (Table 5. Aggregated ATE):
    - Yield in %: 2.168
    - Std. Err.: 0.00609
    - t: 3.56
    - P>t: 0.000
  - Interpretation: SDID estimate around 2.2 pp. of GDP, similar to staggered DID.
- CITSA case studies — selected country results:
  - Jamaica (comprehensive reforms 2012–2015):
    - CITSA estimates: sustained difference of 4.9 pp. of GDP relative to control; extrapolated to about 0.8 pp. of GDP on an annual basis; no immediate level effect observed.
    - Key post-intervention trend coefficient: β7 = 0.049 0.0191 2.57 0.010.
  - Rwanda (2010–2014; less LTO/segmentation):
    - Post-reform sustained revenue yield of 4.7 pp. of GDP; immediate uptick possibly influenced by tax policy measures.
  - Senegal (2010–2015; less improving compliance):
    - Estimated revenue yield of 3.3 pp. of GDP; sustained trend improvement (β7 positive and significant).
- Synthesis of event-study evidence:
  - Sustained, comprehensive TA reform packages are associated with revenue increases between 2 to close to 3 pp. of GDP, with effects increasing over several years and exceeding 3 pp. of GDP after the 6th year for far-reaching reforms.

### Concluding synthesis and policy-relevant recommendations
- Measurement and methods:
  - TA indexes are slow-moving and endogenously determined; robust panel and event-study frameworks are required to estimate yields.
- Main quantitative findings:
  - Increase in TA strength between the 40th and 60th percentile associated with up to 1.8 pp. of GDP increase in tax revenues (ISORA) with 95 percent confidence interval 0.5 to 2.6 pp.; TADAT range 0.2 to 2.1 pp.
  - Sustained post-GFC TA reforms led to increases between 2 to 3 pp. of GDP, with effects rising over six years.
- Priority reform areas with higher estimated yields:
  - Strengthening compliance risk management.
  - Enhancing public accountability.
  - Establishing Large Taxpayer Office (LTO).
  - Strengthening accountability and transparency.
  - Enhancing timely filing of tax declarations.
- Policy design and sequencing:
  - Reforms should be comprehensive and sequenced, recognizing complementarities among TA functions.
  - Use conservative yield estimates (e.g., lower 95 percent confidence bound) for fiscal projections, especially where capacity constraints, weaker ownership, or weaker institutions exist.
  - Small, isolated changes are unlikely to achieve large revenue gains and may pose fiscal risk if expected yields are not realized.
- Operational application:
  - The TA-AYT tool can benchmark TA strength, identify gaps, and estimate expected revenue gains from proposed TA reforms; should be complemented by granular country reviews.

*From IMF Working Paper "Quantifying the Revenue Yields from Tax Administration Reforms", Section 1. Overall Strength of Tax Administration (source: wpiea2023231-print-pdf).*

### 1. Overall Strength of Tax Administration ..............................................................................

### 1. Overall Strength of Tax Administration

### Data
- ISORA (International Survey on Revenue Administration):
  - Surveys TA features through both numerical and categorical questions with 982 data points.
  - Covers 154 countries, including 36 Advanced Economies (AEs), 71 Emerging Markets (EMs), and 46 Low-Income Developing Countries (LIDCs).
  - Four vintages exist; analysis uses ISORA 2016 and ISORA 2018 only.
  - Constructed indexes: an overall operational strength (OS) index as the equally weighted average of seven sub-indexes; seven sub-indexes are:
    - i) compliance risk management (CRM) approach
    - ii) the use of third-party data
    - iii) digitalization of services
    - iv) service orientation
    - v) public accountability
    - vi) the degree of autonomy
    - vii) the establishment of LTO or large taxpayer program
  - Additional ISORA-derived indicators used: ratio of tax staff over the labor force; active taxpayers over the labor force.

- TADAT (Tax Administration Diagnostic Assessment Tool):
  - Provides assessments starting in 2013, covering 86 countries with 11 repeated assessments between 2013 and 2019.
  - Focuses on nine performance outcome areas (POAs): i) integrity of the registered taxpayer base, ii) effective risk management, iii) supporting voluntary compliance, iv) timely filing of declarations, v) timely payment of taxes, vi) accurate reporting in declarations, vii) effective dispute resolution, viii) efficient revenue management, ix) accountability and transparency.
  - Nine POAs are compiled using 32 high-level indicators (HLIs) constructed from 55 measurement dimensions; overall performance assessment report (PAR) index is the equally weighted average of the nine POAs.

- Revenue variable and exclusions:
  - Dependent variable: tax-to-GDP ratio (excluding trade taxes and social security contributions).
  - Exclusions rationale: to account for volatility of commodity prices, exclude resource revenues, and because other agencies than the TA may collect those revenues.

### Methodology
- Two complementary empirical approaches are applied:
  - Cross-country panel analysis:
    - Uses slow-moving indicators of TA strength and accounts for endogeneity of TA reform impact.
    - Specifically applies the Hausman-Taylor cross-country panel regression with endogenous time-invariant variable.
    - Controls for macroeconomic, structural, and tax policy determinants identified in prior literature.
  - Event-study design of reform episodes:
    - Studies TA reform episodes in LIDCs and EMs after the global financial crisis (GFC).
    - Employs three recent extensions to the canonical difference-in-differences (DID) to estimate revenue yields and their evolution over time.
- Additional operational output:
  - A comprehensive database and an operational tool were developed in parallel (Box 1) to identify gaps in TA and systematically estimate contributions of past or future TA reforms.

### Key Findings and Estimated Impacts
- Overall estimated yields:
  - An increase in TA strength moving a country from the 40th percentile to the 60th percentile is associated with:
    - up to 1.4 percentage points (pp.) of GDP increase in tax revenues for all countries.
    - 1.8 pp. of GDP increase in tax revenues for EMs and LIDCs.
  - Event-study estimates for comprehensive TA reforms after the GFC indicate additional revenues between 2 to close to 3 pp. of GDP.
- Timing and dynamics:
  - Revenue yields from far-reaching and comprehensive reforms increase over time, reaching more than 3 pp. of GDP after the 6th year following reform.
  - Significant TA reforms may lead to some initial gains, but full results may take time to materialize.
- Specific TA practices with high revenue correlations:
  - Using ISORA:
    - Compliance risk management improvements have the highest revenue yields.
    - Public accountability and LTO also play a more significant role in increasing revenue in EMs and LIDCs.
  - Using TADAT:
    - Accountability and transparency, and timely filing of tax declarations matter the most for raising revenues.
  - High correlations among ISORA and TADAT sub-indexes indicate TA functions support and reinforce each other.
- Comparative magnitude:
  - Point estimates from cross-country panel analysis are broadly consistent with event-study estimates: cross-country up to 1.4–1.8 pp., event-study between 2 and close to 3 pp., and dynamic effects exceeding 3 pp. by the 6th year for comprehensive reforms.

### Policy and Operational Implications
- Targeted reforms in specific TA practices can yield substantive revenue gains; prioritization may depend on country context (e.g., EMs and LIDCs may particularly benefit from LTO and public accountability enhancements).
- Reform sequencing and persistence matter: comprehensive reforms across multiple TA practices yield larger and increasing benefits over time, with notable gains manifested several years after reform implementation.
- The developed database and operational tool (Box 1) can be used to benchmark TA strength, identify gaps, and systematically estimate expected revenue gains from proposed TA reforms.

*From IMF Working Paper "Quantifying the Revenue Yields from Tax Administration Reforms", Section 1. Overall Strength of Tax Administration.*

### 11. This section presents several stylized facts on the TA strength, its practices, and

### 11. This section presents several stylized facts on the TA strength, its practices, and

### Stylized facts on TA strength, practices, and characteristics
- Main overview:
  - Uses ISORA and TADAT datasets to construct aggregate indexes and sub-indexes; summary statistics are in Annex II.
  - Focus period: 2010-2019 (assessments after 2020 are not included).
- a) TA strength is slow-moving:
  - ISORA and TADAT indexes "only slightly changed" over the years; TA strength, practices, and characteristics are slow-moving variables.
  - Example: correlations between ISORA indexes for 2014 and 2017 are elevated; "eight out of the ten ISORA’s indexes have a coefficient of correlation higher than 0.8."
  - Exceptional cases with significant improvement: Grenada, St. Vincent and Grenadines, Uruguay, Greece, Ghana, Poland, and Myanmar.
- b) TA strength varies with economic development:
  - Aggregate finding: TA strength is on average much stronger in AEs, while LIDCs have the weakest TA.
  - ISORA detailed patterns:
    - AEs have higher scores than EMs and LIDCs, except for the establishment of LTO.
    - EMs have higher scores on average than LIDCs, except for autonomy and establishment of LTO where EMs and LIDCs are almost similar.
    - Weaknesses in EMs and LIDCs: use of third-party data and digitalization, and compliance risk management.
  - TADAT detailed patterns:
    - Differences between AEs, EMs, and LIDCs are more pronounced than in ISORA.
    - AEs have higher scores across all practices and characteristics, especially for timely filing of tax declarations and timely payments of taxes.
    - EMs score higher than LIDCs on average but have weaker scores for effective risk management, accurate reporting in declarations, timely filing, and efficient revenue management.
    - LIDCs have the lowest scores across all practices and characteristics, especially for timely filing of tax declarations, efficient revenue management, accurate reporting in declarations, and effective risk management.
  - c) Heterogeneities by income groups:
    - AEs tend to collect more tax revenues with stronger TA.
    - LIDCs have, on average, weaker TA and collect lower tax revenues.
    - EMs exhibit large variation: some EMs resemble LIDCs in weak TA, others resemble AEs in strong TA.
  - Correlation with revenues:
    - "The strength of TA and tax revenues are positively correlated."
    - ISORA and TADAT indexes correlate highly with tax revenues (excluding trade taxes and social contributions), though there is high variance among peer group countries.
- d) Relationship between ISORA’s OS and TADAT’s PAR:
  - The two aggregate indexes are correlated: "the coefficient of correlation is 0.61."
  - TADAT’s PAR is generally lower than ISORA’s OS for almost all countries; the gap is higher for LIDCs.
  - Possible reasons: ISORA is self-reported by authorities; TADAT assessments are by external experts. Different aggregation methodologies: ISORA uses equally weighted additive components; TADAT uses M1 and M2 methods and non-additive MD components.
  - The comparison should be interpreted as indicating high correlation rather than precise equivalence.
- e) Correlation among sub-components:
  - High correlations among ISORA and TADAT sub-indexes.
  - Implications:
    - i) Good practices/characteristics in one dimension are generally associated with good practices in others.
    - ii) Reforms should be comprehensive and implemented in an integrated manner.
  - Caveat: High correlations across sub-indexes make it difficult to identify the relative importance of individual practices or characteristics for revenue gains or to analyze their interactions.

### Empirical methodology (summary)
- Slow-moving nature of TA measures:
  - Hypothesis: indexes measuring TA strength and practices remain constant over 2010 to 2019.
  - Practical implementation:
    - ISORA’s OS and seven sub-indexes: use 2017 values for all years in 2010-2019.
    - TADAT’s PAR, POAs, and HLIs: use the latest assessment value (or average when repeated) for all years in the period.
  - Robustness check: re-estimate baseline models over 2015–19 for ISORA only (TADAT lacks sufficient data); results similar for range estimates though point estimates for Models 2 and 3 are lower.
- Estimation approach:
  - Employ model by Chatelain and Ralf (2021) to estimate effects of endogenous time-invariant variables correlated with individual effects.
  - Uses a pretest procedure to determine exogenous internal instruments among time-varying variables; then applies Hausman and Taylor (1981) estimator (HT) if not under-identified.
  - Pretest procedure steps:
    - Estimate a random effects estimator including all time-varying and time-invariant variables and averages over time of time-varying variables.
    - Select internal instruments as the subset of time-varying variables whose coefficients associated with their average over time are not statistically significant.
    - HT estimator then estimated specifying the exogenous instrumental variables.
  - Monte Carlo evidence: HT estimator using pretest-selected instruments performs well relative to competing estimators for endogenous time-invariant variables.
- Internally selected time-varying instrumental variables (after pretest):
  - i) terms of trade
  - ii) the share of agriculture to GDP
  - iii) control of corruption
  - iv) the proxies of tax policy
- Endogeneity concerns and controls:
  - TA practices and characteristics may be endogenous—policy changes can be anticipated and correlated with past economic activity and revenue performance.
  - Literature evidence: adoption of LTUs and SARA related to age dependency ratio, size of agricultural sector, non-resources tax-to-GDP ratio, presence of IMF-supported programs, and past TA reforms.
- Model specification (econometric equation):
  - The dependent variable is tax-to-GDP ratio excluding trade taxes and social security contributions.
  - Z_i: set of potentially endogenous time-invariant variables of interest.
  - X1_it: set of control endogenous time-varying variables.
  - X2_it: set of control exogenous time-varying variables identified by the pretest.
  - τ_t: time fixed-effects.
  - μ_i: unobserved country random effects.
  - ε_it: error term.
  - Control variables X include:
    - i) macroeconomic variables: per capita GDP and its square, inflation, trade openness, external debt to GDP, terms of trade.
    - ii) structure and institutions: oil exports to GDP, share of agriculture to GDP, control of corruption.
    - iii) proxies of tax revenue policy.
- Tax policy proxies (three approaches, following Chang et al., 2020):
  - Preferred proxy: difference between tax revenues projected in the budget for year (t) and the actual tax revenues of the preceding year (t -1), both relative to respective GDP. Rationale: budgeted revenues reflect expected tax policy changes and macro fluctuations; one-year horizon limits inclusion of TA reform gains, making approach conservative (if anything understating TA contribution).
  - Sensitivity proxies:
    - (i) difference between budgeted revenues in year (t) and budgeted revenues in year (t-1) to account for projection bias.
    - (ii) top tax rates for VAT, CIT, and PIT.

### Estimation strategy and objectives
- Estimation steps:
  - Estimate equation (1) for aggregate indexes: ISORA’s OS and TADAT’s PAR.
  - Also estimate multivariate specifications including all sub-indexes to identify which sub-components are more critical for raising revenues.
- Purpose:
  - To quantify revenue yields from TA strength and reforms while accounting for endogeneity and the slow-moving nature of TA indicators.

*Source: wpiea2023231-print-pdf — https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023231-print-pdf.pdf*

### 18. Our baseline results are presented in Table 2 and Table 3, using ISORA and TADAT data,

### 18. Our baseline results are presented in Table 2 and Table 3, using ISORA and TADAT data

### Baseline results and estimated revenue yields
- Results are presented for a sample of all countries (columns 1–3) and EMs and LIDCs only (columns 4–5).
- The overall strength of TA (ISORA’s OS; TADAT’s PAR) is positively and statistically significantly associated with higher tax revenues in all specifications, with slightly higher magnitude in EMs and LIDCs.
- Point estimates for feasible reforms (increase in TA strength equivalent to the change between the 40th percentile and the 60th percentile) show:
  - ISORA’s OS: revenue yields of up to 1.8 pp. of GDP (95 percent confidence interval of 0.5-2.6 pp. of GDP).
  - TADAT’s PAR: revenue yields of up to 1.4 pp. of GDP (range of 0.2-2.1 pp. of GDP).
- Table 1 summary (Percentage points of GDP):
  - All countries, Using ISORA OS: 1.4 (0.5 ‒ 2.2)
  - EMs and LIDCs, Using ISORA OS: 1.8 (1.0 ‒ 2.6)
  - All countries, Using TADAT PAR: 1.3 (0.2 ‒ 2.0)
  - EMs and LIDCs, Using TADAT PAR: 1.4 (0.2 ‒ 2.1)
- Empirical goodness of fit: adjusted R-squared indicates models could explain up to 56 percent of the variation of tax revenues over the period 2010–19.

### Staffing of tax administration (TA) — nonlinear effects
- There is an inverted-U relationship between tax revenues and #Tax Staff/LaborForce for all countries:
  - Staffing is positively associated with revenues at lower levels; returns diminish above a threshold.
  - The marginal return turns negative past a threshold of tax staff between 0.15 to 0.22 percent of the labor force (depending on specification).
- This inverted-U relationship does not hold in EMs and LIDCs in the sample.

### Taxpayer base and interaction with TA strength
- Coefficients for Active Taxpayer/Labor Force are positive but non-significant in all specifications when controlling for TA strength and staffing.
- When not controlling for TA strength or staffing, these coefficients turn significant — suggesting strength and staffing of TA may matter more to raise revenues than broadening the taxpayer base.
- Measurement caveat: Active taxpayer measure follows Chang et al. (2020) and may not capture public servants without tax identification numbers, potentially creating attenuation bias.

### Other macroeconomic and structural determinants
- Trade openness (lagged) is positively and statistically associated with higher revenues.
- External debt to GDP shows mixed effects; in some specifications increasing external debt to GDP leads to higher revenues.
- Oil exports (as a % of GDP), lagged, are positive and statistically significant in multiple specifications.
- Higher inflation (CPI, lagged) is associated with lower revenues in some specifications.
- Stronger control of corruption is associated with higher revenues in TADAT specifications.
- Tax policy proxy (Budgeted_Actual, Nontrade_tax, Combined) is positively and significantly associated with tax revenues.
- Variables not significantly impacting tax revenues (after controls): real GDP per capita and its squared, terms of trade, share of agriculture in GDP (in many specifications).

### Magnitudes from regression tables (selected coefficients and statistics)
- ISORA (Table 2) — Operational Strength, 2017 coefficients (selected):
  - Column (1) All: 0.1333*** (0.0386)
  - Column (4) EMs & LIDCs: 0.1658*** (0.0356)
- TADAT (Table 3) — TADAT: PAR coefficients (selected):
  - Column (1) All: 0.1328*** (0.0319)
  - Column (4) EMs & LIDCs: 0.1376*** (0.0324)
- Selected staffing coefficients (TADAT, Table 3):
  - #Tax Staff/LaborForce in some All-country specifications: 0.1304*** (0.0437)
  - Sq(#Tax Staff/Labor Force) negative and significant in some specifications: -0.0429*** (0.0159)
- Adjusted R-squared examples:
  - ISORA models: Adj. Overall R2 up to 0.4553 (columns vary; shown up to 0.4553).
  - TADAT models: Adj. Overall R2 up to 0.5664 (columns vary; shown up to 0.5664).
- Sample sizes and country counts (examples from tables):
  - ISORA Table 2 observations: 768, 551, 482, 682, 470, 416
  - ISORA number of countries: 92, 75, 65, 80, 63, 55
  - TADAT Table 3 observations: 536, 315, 299, 518, 297, 281
  - TADAT number of countries: 60, 41, 39, 57, 38, 36

### Robustness, caveats, and empirical considerations
- Robustness checks: findings are robust to dropping outliers/influential data points (coefficients slightly lower but variable of interest remains similar).
- Baseline findings robust to excluding all commodity exporters (results available upon request).
- Caveats on tax policy measure and revenue fluctuations:
  - Revenue fluctuations are not proportional to GDP (commodity price swings, supply shocks, recessions/booms); potential multicollinearity could increase standard errors.
  - Use of estimated actual revenues in year (t-1) could differ from factual revenues; concern only if estimation error is systematic.
  - Nowcasting tax revenues projected in the budget for year (t) can be affected by bias or serial correlation.
- Practical policy caution: TA reforms work best as well-designed, properly sequenced packages accounting for interactions and country-specific constraints; small and isolated improvements may not translate into higher yields and could pose fiscal risk if expected yields are not achieved.
- For fiscal projections, prudence is advised: use the lower econometric bound of estimated yields (95 percent confidence lower bound) to avoid overestimation; lags between reform implementation and results may be country specific.

### TA-AYT: IMF FAD Tax Administration - Assessment and Yield Tool (operational implications)
- TA-AYT is an Excel-based tool developed to estimate contribution of TA reforms and benchmark TA strength across IMF member countries using ISORA (ISORA 2016 and ISORA 2018 vintages) and other sources.
- Structure and modules:
  - Module 1: Benchmarking TA performance over time and comparators; overview of TA strength overall and by practices; correlations with tax revenues.
  - Module 2: Presents goodness of fit of the model for the country of interest and comparators; helps assess predictive accuracy.
  - Module 3: Allows user inputs on potential outcomes and duration of TA reforms to construct new index values and estimate a range and cumulative revenue yields using regression results.
- Use guidelines:
  - TA-AYT helps identify bottlenecks and weaknesses but macro-level analysis may mask country-specific specifics; should be complemented by in-depth country review.
  - For fiscal projection use, prefer the lower 95 percent confidence bound and prepare for upside risk; validate yields with granular, country-specific characteristics.

*Source: wpiea2023231-print-pdf (IMF Working Paper).*

### 2. Specific Components of Tax Administration: Multivariate Models

### 2. Specific Components of Tax Administration: Multivariate Models

### Overview and caveats
- Multivariate models are used to assess the relative importance of specific practices or characteristics of tax administration (TA) by including all sub-indexes simultaneously.
- Two main caveats:
  - Each component of TA is correlated with others, making it difficult to distinguish one component's impact from others (multicollinearity).
  - Singling out one component without considering others risks overestimating its impact (omitted variable problem).
- Despite caveats, inclusion of all sub-indexes yields higher adjusted R-squared compared with models using aggregate indexes, suggesting added explanatory power for variation in tax revenues.
- The authors do not use the multivariate model for the TA-AYT overall estimate because multicollinearity may make statistical significance and inference less reliable; single-variate models show almost all sub-index coefficients positive and statistically significant (results available upon request).

### Empirical findings from multivariate models
- ISORA (International Survey on Revenue Administration) results:
  - Enhancing compliance risk management (both for all countries and EMs and LIDCs) can significantly raise revenues.
  - Strengthening public accountability can significantly raise revenues.
  - Establishing a Large Taxpayer Office (LTO) is significant for EMs and LIDCs.
  - These are identified as more critical components to raise revenues and confirm findings in Chang et al. (2020).
- TADAT (Tax Administration Diagnostic Assessment Tool) results:
  - Improving accountability and transparency and timely filing of tax declarations could yield higher revenues.
  - Coefficients associated with POAs are mostly non-significant when controlling for staffing of TA and taxpayers’ base (columns 2–3 and 5–6 in the source tables).

### Data and stylized facts for reform episodes
- Sample: LIDCs and EMs with sustained revenue mobilization efforts identified using Akitoby et al. (2020); used to construct a treatment group of countries with observed TA reforms.
- Many countries pursued tax administration and tax policy reforms in parallel; all countries in the sample saw some increase in revenue ratio in the year after TA reforms.
- Table 4 (source) lists countries with TA reforms after the Global Financial Crisis (GFC) and which TA practices were covered. Country grouping for analyses:
  - Country group 2008-09: Afghanistan, Algeria, Burkina Faso, Burundi, Central African Republic, Guinea-Bissau, Liberia, Nepal.
  - Country group 2010-11: Congo Central Republic, Mauritania, Philippines, Rwanda, Senegal, Sierra Leone, Gambia, Maldives.
  - Country group 2012-13: Malawi, Jamaica, Tuvalu, Uganda, Belize, Armenia.
  - Control group: Most other EMs and LIDCs.
- Note: Tax-to-GDP ratios exclude trade taxes and social contributions.

### Identification challenges in estimating TA reform impacts
- Main identification challenges:
  - Variation in treatment timing across countries (staggered adoption).
  - Parallel trends assumption for canonical DID may not hold.
  - Heterogeneity in scope and impact of TA reforms (variation in treatment effect).
  - Co-occurrence of tax policy reforms or exogenous shocks (commodity price hikes, rate changes) complicates attribution.
- Authors address these using three modern quasi-experimental extensions to canonical DID:
  1. Difference-in-differences in a staggered-adoption design (Callaway and Sant’Anna, 2021)
     - Estimates group-time average treatment effect ATT(g,t) = E[Yt − Yt−1 | G=g] − E[Yt − Yt−1 | C=1], where Y is tax-to-GDP ratio and C indicates never-treated control group.
     - Aggregation approaches: average across groups; average by exposure length (elapsed time since reform).
  2. Synthetic Difference-in-Differences (SDID) with common support (Arkhangelsky et al., 2021)
     - Constructs synthetic control units as optimally weighted functions of never-treated countries and pre-treatment periods to achieve parallel pre-treatment trends, then applies staggered DID to the reweighted panel.
     - Estimation steps: i) least squares estimation of country weights for synthetic control using pre-treatment data; ii) least squares estimation of time weights using control data; iii) DID in a staggered design on resulting synthetic panels.
  3. Controlled Interrupted Time-series Analysis (CITSA) (Linden, 2015)
     - Allows verification of parallel trends assumption and adjustment for differences in trend between groups; controls for time-varying confounders and co-interventions.
     - CITSA regression specification used:
       Yt = β0 + β1 Tt + β2 Xt + β3 XtTt + β4 Z + β5 ZTt + β6 ZXt + β7 ZXtTt + ∈t
     - Interpretation of coefficients:
       - β0: intercept (starting level) for control group.
       - β1: pre-intervention slope for control group.
       - β2: change in level immediately post-intervention for control group.
       - β3: change in trend post-intervention for control group.
       - β4: difference in pre-intervention level between treated and control.
       - β5: difference in pre-intervention slope between treated and control.
       - β6: difference in immediate post-intervention level between treated and control.
       - β7: difference in post-intervention trend between treated and control (akin to DID of slopes).

### Controlling for tax policy changes
- DID specifications incorporate a proxy for tax policy changes: the difference between budgeted revenue and actual revenue (budgeted revenue reflects planned tax policies).
- Inclusion of this control did not yield a statistically significant coefficient nor change magnitudes or signs of key coefficients of interest.
- Authors infer the DID specifications absorb most systematic impacts of tax policy changes, though imperfectly.

### Empirical results: staggered DID and duration effects
- Staggered DID first-stage findings:
  - Average increase in tax revenues of about 2 pp. of GDP for the three country groups adopting TA reforms (Table 7).
  - Country group 2008-09 mean ATT: 2.6 pp. of GDP.
  - Country group 2010-11 point estimate: increase in tax-to-GDP ratio by 2.2 pp. of GDP (large variance; statistically less significant).
  - Country group 2012-13 mean ATT: 2 pp. of GDP.
  - Interpretation: 2008-09 cohort had highest ATT likely due to longest exposure and lowest starting base post-GFC.
- Aggregated (by exposure length) ATT (duration results):
  - Yield increases over time from about 1.5 pp. of GDP in the 1st year following reforms to more than 3 pp. of GDP after the 6th year.
  - Specific duration trajectory:
    - 1st year after TA reforms: about 1.5 pp. of GDP.
    - 3rd year after TA reforms: about 2.6 pp. of GDP.
    - Thereafter it goes down a bit and stabilizes around 2.2 pp. of GDP before increasing to just above 3 pp. of GDP after the 6th year.
  - Notes: ATT aggregation uses an exposure-based scheme to mitigate composition-change issues; 2011 and 2013 country groups are dropped from the aggregate ATT after year 7 and year 5, respectively.

*Source: IMF Working Paper — Quantifying the Revenue Yields from Tax Administration Reforms (section: 2. Specific Components of Tax Administration: Multivariate Models).*

### 2.   Synthetic Difference-in -Differences with Common Support

### 2.   Synthetic Difference-in -Differences with Common Support

### SDID aggregated effect and estimation approach
- Methodology:
  - Synthetic Difference-in-Differences with Common Support (SDID) uses both time and unit weights to create a counterfactual.
  - The ATT for countries with TA reforms is obtained by applying the SDID estimator repeatedly, calculating an optimal unit- and time-specific weight vector for each adoption period.
  - The aggregated estimator is a weighted average of the three country-group ATTs, where the weight applied to any country group ATT equals the proportion of countries with TA reforms that originate in that country group.
  - Figure 9 (Panels A–C) presents supporting calculations: left-hand charts show trends in tax-to-GDP ratio over time for treated country groups and the weighted average of control countries (weights used to average pre-TA reform time periods shown at bottom in green); right-hand charts show country-by-country tax-to-GDP ratio differences with dot size indicating weights and a horizontal line showing the weighted average (estimated effect).

- Main aggregated result (Table 5. Aggregated ATE):
  - Yield in %: 2.168
  - Std. Err.: 0.00609
  - t: 3.56
  - P>t: 0.000

- Caveat on sample window:
  - The sample runs until 2018. Countries in the 2011 and 2013 country groups are not observed over a 9-year period and are dropped from the aggregation of ATTs after the 7th and 5th years, respectively.

### Key interpretations from SDID
- The estimated yield of around 2.2 pp. of GDP is very similar to results from the staggered DID (Table 5).
- SDID allows for an intercept in estimation weights; unlike traditional Synthetic Control Method, SDID seeks to mimic the trend of the treatment group even if the two groups have very different intercepts.
- The estimated effect is obtained by applying a DID analysis to the reweighted panel produced by SDID.

### Controlled Interrupted Time-series Analysis (CITSA) case studies — summary of findings
- General:
  - CITSA case studies order tax-to-GDP ratios as a time series for treated countries versus control countries.
  - Yield estimates from CITSA are broadly in line with country-group results.
  - DID methods can partially control for tax policy measures; estimated yields could be upward biased if treated-unit tax policy measures increase tax-to-GDP relative to controls.
  - TA reforms often implemented as a package; establishment of LTO and segmentation and improvement in compliance risk management are essential parts of TA reforms—yield gains of a comprehensive TA reform package would be reduced without those measures.

- Jamaica — Comprehensive TA Reforms (measures 2012–2015):
  - Reforms included: improving compliance, mandatory electronic filing and payment, modernizing tax administration system, semi-autonomous revenue agency, expanding LTO staff, strengthening LTO with increased audit coverage, and legislative changes increasing powers of the tax administration.
  - CITSA key coefficients (Dependent variable: Log (REV/GDP)):
    - β4 - difference in the level of revenue ratio before RA Reforms: 0.568 0.0644 8.82 0.000
    - β5 - difference in the trend in revenue ratio before RA Reforms: -0.038 0.0121 -3.17 0.002
    - β6 - difference in the level of revenue ratio immediately following introduction of RA Reforms: -0.0048 0.0892 -0.05 0.958
    - β7 - difference in the trend of revenue ratio after initiation of RA reforms compared with preinte: 0.049 0.0191 2.57 0.010
  - Estimated effects:
    - Jamaica had a statistically significant sustained difference of 4.9 pp. of GDP in tax-to-GDP ratio relative to the control group.
    - This can be extrapolated to a yield difference of about 0.8 pp. of GDP on an annual basis.
    - No immediate level difference was observed following introduction of TA reforms (suggesting reforms took time to yield gains).

- Rwanda — Comprehensive TA Reforms (2010–2014; less LTO and segmentation):
  - Reforms included: electronic filing and payment systems, electronic registration, improvements in tax compliance, collection of arrears; no observed reforms related to LTO or segmentation.
  - Pre-reform: Rwanda had a lower tax-to-GDP ratio than control groups; statistically significant differences in level and trend could not be established due to large variances in coefficients (possibly driven by the GFC).
  - Post-reform:
    - Considerable uptick in tax-to-GDP ratio in levels relative to control, with a sustained revenue yield of 4.7 pp. of GDP.
    - The large immediate uptick could potentially be due to tax policy measures (increased tax rate for imported construction materials, excise duty on airtime of mobile phones, removal of some VAT exemptions).

- Senegal — Comprehensive TA Reforms (2010–2015; less improving compliance):
  - Reforms included: improvements in taxpayer segmentation, e-filing and e-payment; no observed reforms related to improving compliance.
  - CITSA key coefficients (Dependent variable: Log (REV/GDP)):
    - β4 difference in the level of revenue ratio before RA Reforms: -0.1226 0.068 -1.8 0.072
    - β5 - difference in the trend in revenue ratio before RA Reforms: -0.0196 0.0151 -1.3 0.193
    - β6 difference in the level of revenue ratio immediately following introduction of RA Reforms: 0.1868 0.0917 2.04 0.042
    - β7 difference in the trend of revenue ratio after initiation of RA reforms compared with preinte: 0.0476 0.0194 2.45 0.014
  - Additional coefficients table (Dependent variable: Log (REV/GDP)) reported:
    - β4 difference in the level of revenue ratio before RA Reforms: 0.3178 0.047 96.64 0.000
    - β5 difference in the trend in revenue ratio before RA Reforms: -0.019 0.0118 -1.65 0.099
    - β6 difference in the level of revenue ratio immediately following introduction of RA Reforms: 0.0187 0.074 0.25 0.799
    - β7 difference in the trend of revenue ratio after initiation of RA reforms compared with preinte: 0.0326 0.014 2.27 0.023
  - Estimated effects:
    - Senegal implemented reforms that led to an estimated revenue yield of 3.3 pp. of GDP.
    - No immediate impact following introduction of TA reforms in terms of a sustained level change, but a sustained trend improvement (β7).

### Concluding synthesis and policy-relevant findings
- Measurement and methods:
  - Existing TA indexes are slow-moving and TA status is endogenously determined; robust empirical frameworks and detailed case examinations are needed to estimate revenue yields of TA reforms with cross-country data.
  - The paper employed panel methods (Hausman-Taylor) and event-study models and used comprehensive ISORA and TADAT indicators.

- Main quantitative findings:
  - An increase in TA strength equivalent to the change between the 40th percentile and the 60th percentile of IMF member countries is associated with an increase in tax revenues by up to 1.8 pp. of GDP.
  - Range estimates of revenue yields (95 percent confidence intervals) for such an increase in the indexes:
    - Using ISORA: 0.5 to 2.6 pp. of GDP.
    - Using TADAT: 0.2 to 2.1 pp. of GDP.
  - Sustained TA reforms after the GFC led to an increase in tax revenues between 2 to 3 pp. of GDP, with the treatment effect increasing over six years.
  - Event-study analysis: revenue yields increase over time to more than 3 pp. of GDP after the 6th year following a far-reaching and comprehensive reform across several TA practices.

- Policy implications and recommendations:
  - To achieve revenue gains of the magnitudes estimated, a set of TA reforms should be properly implemented and sequenced.
  - Specific TA reform areas with higher yields include:
    - (i) strengthening compliance risks management,
    - (ii) enhancing public accountability,
    - (iii) establishing LTO,
    - (iv) strengthening accountability and transparency,
    - (v) enhancing timely filing of tax declarations.
  - Attribution is not exact; develop TA integrally because different TA functions support and reinforce each other.
  - Use yield estimates conservatively (e.g., lower bound) for fiscal projections where countries have capacity constraints, weaker ownership, or weaker institutions.
  - Small and isolated improvements in some practices may not translate into higher yields and could pose fiscal risk if expected revenue is not achieved.

- Operational tool and further use:
  - The paper develops an operational tool to estimate the contribution of past or future TA reforms for IMF member countries by constructing new index values consistent with reforms and estimating a range of revenue yields.

*Source: IMF Working Paper — section 2 and related case-study material from the provided chapter content.*

### Annex I. Data Descriptions and Sources

### Annex I. Data Descriptions and Sources

### Data definitions and sources
- Operational Strength: It measures the overall strength of the revenue administration across practices and characteristics. — ISORA dataset
- Compliance Risk Management: It measures the formal approach for identifying, assessing, and prioritizing key compliance risks (automated risk profiling, electronic audit, ...) — ISORA dataset
- Third Party Data: It measures the use of computer-based information systems for processing various forms of third-party data and use of third-party in prefilling returns. — ISORA dataset
- Digitalization: It measures the provision of e-services, including e-filing, e-payment, and web-based information and communication services — ISORA dataset
- Service Orientation: It measures the service offerings facilitating compliance and the provisions to enhance taxpayer satisfaction with services offered. — ISORA dataset
- Public Accountability: It measures if key documents such as plans and achievements against plans are published and the use of external auditor — ISORA dataset
- Autonomy: It measures the institutional form, the degree of autonomy in managing expenditure and human resources, and ability to set performance standards, and provision of tax policy advice. — ISORA dataset
- Est. LTO: It takes the value one if there is an office or program dedicated to large taxpayers in operation — ISORA dataset
- PAR: Performance Assessment Report — TADAT dataset
- POA1: Integrity of the Registered Taxpayer Base — TADAT dataset
- POA2: Effective Risk Management — TADAT dataset
- POA3: Supporting Voluntary Compliance — TADAT dataset
- POA4: Timely Filing of Tax Declarations — TADAT dataset
- POA5: Timely Payment of Taxes — TADAT dataset
- POA6: Accurate Reporting in Declarations — TADAT dataset
- POA7: Effective Tax Dispute Resolution — TADAT dataset
- POA8: Efficient Revenue Management — TADAT dataset
- POA9: Accountability and Transparency — TADAT dataset
- Tax revenues excluding trade taxes and social security contributions (% of GDP): Total tax revenues minus trade taxes and social security contributions as a percent of GDP — Authors' calculations using the IMF’s WEO, the IMF’s WoRLD, the OECD’s TRD, and the UNU-WIDER’s GRD.
- Log (PerCapitaGDP_PPP), lagged: One-year lagged value of the log of GDP per capita in purchasing power parity — Authors' calculations using the IMF's WEO dataset
- Square log (PerCapitaGDP_PPP), lagged: One-year lagged value of the squared of the log of GDP per capita in purchasing power parity — Authors' calculations using the IMF's WEO dataset
- Trade openness (out of GDP), lagged: One-year lagged value of trade openness (sum of exports and imports over GDP) — Authors' calculations using the IMF's WEO dataset
- External debt (out of GDP), lagged: External debt as percent of GDP — Authors' calculations using the IMF's WEO and World Bank's WDI datasets
- CPI, lagged: One-year lagged value of inflation, percent change of consumer price index, period average — Authors' calculations using the IMF's WEO and World Bank's WDI datasets
- Terms of Trade (2000=1), lagged: One-year lagged value of terms of trade (2000=1) — Authors' calculations using the IMF's WEO and World Bank's WDI datasets
- Oil exports (as a % of GDP), lagged: One-year lagged value of oil exports as percent of GDP — Authors' calculations using the IMF's WEO dataset
- Log (Agri, %GDP), lagged: One-year lagged value of the log of the share of agriculture in GDP — Authors' calculations using the World Bank's WDI dataset
- Control Corruption, lagged: One-year lagged value of the log of control of corruption index — Authors' calculations using the World Bank WGI dataset
- Budgeted_Actual, Nontrade_tax, Combined: Proxy of tax policy measures: the difference between the budgeted revenue of this year as share of projected GDP and the actual revenue of last year as share of actual GDP — Authors' calculations using the IMF’s WEO, the IMF’s WoRLD, the OECD’s TRD, and the UNU-WIDER’s GRD.
- Top Combined CIT Rate (%), lagged: One-year lagged value of top CIT rate (%) — IMF's Tax Rate Database (DART) and IMF's GFS dataset
- Top Combined PIT Rate (%), lagged: One-year lagged value of top PIT rate (%) — IMF's Tax Rate Database (DART) and IMF's GFS dataset
- Standard VAT Rate (%), lagged: One-year lagged value of VAT rate (%) — IMF's Tax Rate Database (DART) and IMF's GFS dataset
- Budgeted_L.budgeted, Nontrade_tax, Combined: Proxy of tax policy measures: the difference between the budgeted revenue of this year as share of projected GDP and the budgeted revenue in last year as share of projected GDP — Authors' calculations using the IMF’s WEO, the IMF’s WoRLD, the OECD’s TRD, and the UNU-WIDER’s GRD.
- VAT compliance gap: The VAT compliance gap is the difference between the actual VAT revenues, determined on the accrued collections basis, and potential VAT revenues, estimated using the current policy framework. — Authors' calculations using the IMF’s RA-GAP and other sources
- Average daily number of structural conditionalities (SCs) on revenue administration reforms: The number of SCs on RA reforms divided by the program duration in days — IMF's MONA Dataset
- Average daily number of SCs on revenue administration reforms met: The number of SCs on RA reforms met divided by the program duration in days — IMF's MONA Dataset
- The share of SCs on RA reforms met: The number of SCs on RA reforms met divided by the total number of SCs — IMF's MONA Dataset
- Data on Revenue Administration Reforms: Identified countries and years with sustained period of RA reforms — Akitoby et al. (2020); ISORA; TADAT

### Summary statistics and correlations (key figures)
- ISORA (2017) and TADAT summary (selected):
  - Operational Strength: Obs. 1460; Mean 0.617; Sd 0.173; CV 0.280; Median 0.630; Min 0.175; Max 0.951
  - Compliance Risk Management: Obs. 1460; Mean 0.562; Sd 0.305; CV 0.543; Median 0.667; Min 0.000; Max 1.000
  - Third Party Data: Obs. 1460; Mean 0.332; Sd 0.258; CV 0.777; Median 0.273; Min 0.000; Max 0.955
  - Digitalization: Obs. 1460; Mean 0.532; Sd 0.327; CV 0.615; Median 0.583; Min 0.000; Max 1.000
  - Service Orientation: Obs. 1460; Mean 0.719; Sd 0.203; CV 0.282; Median 0.769; Min 0.091; Max 1.000
  - Autonomy: Obs. 1460; Mean 0.713; Sd 0.222; CV 0.311; Median 0.760; Min 0.077; Max 1.000
  - LTO: Obs. 1460; Mean 0.890; Sd 0.313; CV 0.352; Median 1.000; Min 0.000; Max 1.000
  - PAR: Obs. 770; Mean 0.356; Sd 0.188; CV 0.528; Median 0.370; Min 0.021; Max 0.896
  - POA1–POA9: obs. generally 760–770; means range from 0.252 to 0.529; individual min 0.000 and max up to 1.000 for several POAs
- Correlations between ISORA indexes across years (selected):
  - Operational Strength correlation: 0.92 (Obs. 121)
  - Compliance Risk Management correlation: 0.85 (Obs. 121)
  - Third Party Data correlation: 0.88 (Obs. 121)
  - Digitalization correlation: 0.75 (Obs. 121)
  - Service Orientation correlation: 0.83 (Obs. 121)
  - Public Accountability correlation: 0.81 (Obs. 121)
  - Autonomy correlation: 0.83 (Obs. 121)
  - LTO correlation: 0.61 (Obs. 121)
  - PAR: 0.75 (Obs. 11) and POA correlations in TADAT vary; note: low number of TADAT observations cautioned

### Correlation structure (selected pairwise correlations from Table AII .3)
- ISORA internal correlations (examples):
  - (1) Operational Strength with (2) Compliance Risk Management: 0.694***
  - Operational Strength with (4) Digitalization: 0.749***
  - Operational Strength with (5) Service Orientation: 0.776***
  - (3) Third Party Data with (10) #ActiveTaxpayer/LaborForce: 0.644***
- ISORA–TADAT cross-correlations (selected lower-left quadrant):
  - Operational Strength with PAR: 0.590***
  - Operational Strength with POA1: 0.420***
  - Operational Strength with POA2: 0.568***
  - Operational Strength with POA9: 0.672***
- TADAT internal correlations (selected lower-right quadrant):
  - PAR with POA1: 0.812***
  - POA1 with POA6: 0.822***
  - POA3 with POA6: 0.762***

### Multivariate model results — ISORA (Table AIII.1) — key estimated coefficients (Hausman-Taylor estimator; robust SEs in parentheses)
- Compliance Risk Management:
  - Column (1) All: 0.0555*** (0.0212)
  - Column (2) All: 0.0626*** (0.0237)
  - Column (3) All: 0.0714*** (0.0256)
  - Columns (4)–(6) (EMs & LIDCs): coefficients range 0.0596***, 0.0633**, 0.0664** with reported SEs (0.0213), (0.0255), (0.0270)
- Third Party Data: coefficients reported include 0.0218 (0.0293), 0.0372 (0.0356), 0.0265 (0.0435), not statistically significant in columns shown.
- Digitalization: coefficients reported include 0.0160 (0.0176), 0.0173 (0.0196), 0.0181 (0.0205); not statistically significant in columns shown.
- Service Orientation: negative coefficients in some specifications (e.g., -0.0231 (0.0307)), not statistically significant.
- Public Accountability: positive coefficients; in EMs & LIDCs columns shows 0.0437* (0.0262), 0.0407 (0.0281), 0.0593* (0.0315)
- Est. LTO: mixed; Column (4) EMs & LIDCs: 0.0688*** (0.0192); Column (6): 0.0687** (0.0300)
- #Tax Staff/LaborForce: 0.0690** (0.0293) in Column (1); positive with significance in some specifications
- Sq(#Tax Staff/Labor Force): -0.0161* (0.0086) in Column (1) indicating diminishing returns
- Trade openness (out of GDP), lagged: 0.0285*** (0.0089) in Column (1); positive and significant across specifications
- CPI, lagged: -0.0344* (0.0201) in Column (1); negative in some specs
- Oil exports (as a % of GDP), lagged: 0.4289** (0.1694) in Column (1); positive and significant across several specs
- Budgeted_Actual, Nontrade_tax, Combined: 0.1402*** (0.0519) in Column (1); 0.1723*** (0.0560) in Column (3); positive and statistically significant across all columns
- Model fit and sample:
  - Observations: 768, 551, 482, 682, 470, 416 across columns (1)–(6)
  - Number of countries: 92, 75, 65, 80, 63, 55
  - Year FE: Yes
  - Adj. Overall R2: 0.4107, 0.4526, 0.4924, 0.4060, 0.3982, 0.4747

### Multivariate model results — TADAT (Table AIII.2) — key estimated coefficients (Hausman-Taylor estimator; robust SEs in parentheses)
- TADAT: POA4: Timely Filing of Tax Declarations:
  - Column (1) All: 0.0635* (0.0328)
  - Column (2): 0.0786* (0.0417)
  - Column (4): 0.0581* (0.0344)
  - Column (5): 0.0773* (0.0399)
  - Coefficients indicate a positive association in multiple specifications
- TADAT: POA9: Accountability and Transparency:
  - Column (1) All: 0.1081*** (0.0388)
  - Column (4): 0.1098*** (0.0406)
  - Positive and statistically significant in selected specifications
- #Tax Staff/LaborForce: 0.1181*** (0.0429) in Column (1); positive and significant, with Sq(#Tax Staff/Labor Force) = -0.0390** (0.0166) indicating diminishing returns
- Trade openness (out of GDP), lagged: 0.0234** (0.0095) in Column (1); positive and significant across specifications
- External debt (out of GDP), lagged: 0.0176*** (0.0061) in Column (1)
- CPI, lagged: -0.0589* (0.0326) in Column (1); negative in some specifications
- Oil exports (as a % of GDP), lagged: 0.5603*** (0.1303) in Column (1); strongly positive and significant across all columns
- Control Corruption, lagged: 0.0108* (0.0061) in Column (1); positive at the 10 percent level in some specifications
- Budgeted_Actual, Nontrade_tax, Combined: 0.0955* (0.0562) in Column (1); 0.1847*** (0.0685) in Column (3); positive and statistically significant in several specifications
- Model fit and sample:
  - Observations: 536, 315, 299, 518, 297, 281 across columns (1)–(6)
  - Number of countries: 60, 41, 39, 57, 38, 36
  - Year FE: Yes
  - Adj. Overall R2: 0.5749, 0.6028, 0.6192, 0.5535, 0.5510, 0.5740

### Stylized facts and interpretation from annex tables and figures
- ISORA indexes show high within-index stability across available years (e.g., Operational Strength correlation 0.921 across Obs. 121).
- Positive associations with tax revenue (dependent variable) appear for:
  - Compliance Risk Management (ISORA): consistently positive and statistically significant in multiple specifications (coefficients from 0.0555*** to 0.0714*** in columns shown).
  - Public Accountability (ISORA) and Est. LTO in several EMs & LIDCs specifications.
  - #Tax Staff/LaborForce with negative quadratic term indicating diminishing marginal returns to additional tax staff.
  - Trade openness (lagged) consistently positive and significant across ISORA and TADAT models.
  - Oil exports (lagged) large positive coefficients in both ISORA and TADAT models (e.g., 0.4289**, 0.5603***).
  - Budgeted_Actual (proxy for tax policy change) positive and statistically significant across ISORA and TADAT specifications (e.g., 0.1402*** in ISORA Column (1); 0.1847*** in TADAT Column (3)).
- TADAT-specific results highlight POA4: Timely Filing of Tax Declarations and POA9: Accountability and Transparency as POAs with positive and statistically significant associations with revenue outcomes in selected specifications.
- Correlation matrices show strong co-movement among several ISORA indexes (e.g., Operational Strength with Service Orientation 0.776***), and between ISORA and TADAT measures (e.g., Operational Strength with POA9: 0.672***), suggesting related dimensions of administration quality.

*Source: Annex I–IV and accompanying tables and figures from "Quantifying the Revenue Yields from Tax Administration Reforms", Working Paper No. WP/2023/231, IMF.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023231-print-pdf.pdf_