## _wp14206 - 2013. The results indicate that such conditionality had a positive impact on tax revenue, with

## Source details

**Canonical URL:** [_wp14206 - 2013. The results indicate that such conditionality had a positive impact on tax revenue, with](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14206.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14206.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14206.pdf.json)

---

### I. Introduction — purpose and high-level findings
- Research question: Does revenue conditionality in IMF-supported programs affect tax revenue collection, and does the design of conditionality (tax type targeted, policy vs administration, specific vs general) matter?
- Main empirical findings:
  - Revenue conditionality had a positive impact on tax revenue, with strongest improvement on taxes on goods and services, including the VAT.
  - Revenue conditionality matters more for low-income countries, particularly those where revenue ratios are below the group average.
  - Revenue conditionality appears to be more effective when targeted to a specific tax.
  - Results hold after controlling for potential endogeneity, sample selection bias, and when revenues are adjusted for economic cycle.
- JEL Classification Numbers: C33; E62; F33; H2
- Keywords: Tax revenue; structural conditionality

### II. Data, scope, and descriptive context
- Sample and period:
  - Unbalanced panel of 126 low- and middle-income countries over the period 1993-2013.
  - T=21 years and N=126 in the panel context.
- IMF-program and conditionality prevalence:
  - Since 1993, 96 of the 126 countries had an IMF-supported program for at least 1 year.
  - Over the entire sample, about 43 percent of the time countries had IMF-supported programs.
  - IMF-supported programs were more frequent in low-income countries (about 63 percent).
  - Over the sample period, over 1,500 revenue conditionalities were met in the 441 newly approved IMF-supported programs.
  - Over the entire sample period, about 20 percent of the time countries included some type of revenue conditionality; excluding non-IMF-supported program countries, this figure is 50 percent.
  - On average, countries had 5 years with revenue conditionality and 3 years of consecutive revenue conditionality.
- Composition of revenue conditionality:
  - 80 percent structural benchmarks; 20 percent prior actions.
  - Share of conditionality by main tax: G&S 56 percent; Income 32 percent; Trade 12 percent.
- Aggregate descriptive revenue performance:
  - Average increase in tax revenue collection by about 2 percentage points of GDP over the sample period.
  - Middle-income countries increased revenue by around 3 percentage points of GDP until 2008, about 1 percentage point of GDP more than low-income countries; after 2008 low-income countries strengthened collection while middle-income countries experienced lower tax-to-GDP ratios, producing near convergence.

### III. Data sources and variable construction
- Tax revenue sources: IMF’s Government Finance Statistics (GFS), IMF’s World Economic Outlook (WEO), OECD’s Revenue Statistics in Latin America.
- Tax series included (relative to GDP, expressed in logs in regressions): total tax revenue, taxes on goods and services (includes VAT, excise taxes, and other consumption-related taxes), VAT, corporate income tax (CIT), personal income tax (PIT), taxes on international transactions (Trade).
- IMF program and conditionality data: IMF’s Monitoring of Fund Arrangements (MONA) database.
- Binary revenue conditionality indicator: equals 1 if a country in a given year had an IMF-supported program with met revenue conditionality and 0 otherwise. If conditionality was not met, the dummy equals 0. For general (non–tax-specific) conditionality, assumed to apply to all taxes in that year.
- Lag structure: mostly the first lag of the revenue conditionality dummy is used.

### IV. Empirical specification and estimation
- Baseline dynamic specification:
  - Includes lagged dependent variable, lagged revenue conditionality dummy D, control vector X, country and time fixed effects.
  - Separate estimations for Total Tax, G&S, VAT, Income, CIT, PIT, and Trade.
- Estimator: System-Generalized Method of Moments (system-GMM), Blundell and Bond (1998) variant.
- Controls (X): GDP per capita, share of agriculture in value-added, trade openness, inflation, external indebtedness, and other standard determinants.
- Instruments and diagnostics:
  - Instruments for selection and endogeneity include international reserves in months of imports (lagged), change in bilateral exchange rate to US dollar (lagged), overall fiscal balance (lagged), and lags of tax changes.
  - Diagnostics reported: Hansen and Sargan statistics, Arellano-Bond (M1 and M2) tests, difference-in-Hansen test; robust standard errors used.
- Alternative estimators reported in appendices: fixed-effects, Heckman two-stage, inverse probability weighting (Hirano et al., 2003).

### V. Main regression results — IMF program vs. revenue conditionality (Table 2 summary)
- Key coefficients and diagnostics (Table 2, Columns 1–2):
  - Tax, lagged: 0.6749*** (0.1117) in (1); 0.6981*** (0.1135) in (2)
  - IMF Program, lagged: 0.0471 (0.0302) — not statistically significant in Column 1
  - IMF Program with revenue conditionality, lagged: 0.0539** (0.0241) in Column 2
  - IMF Program without revenue conditionality, lagged: 0.0391 (0.0261) in Column 2 — not statistically significant
  - Observations: 1851; Number of countries: 122
  - M1 (p value): 0.001; M2 (p value): 0.454 (1); Hansen-Over-identification (p value): 0.674 (1); Diff-in-Hansen (p value): 0.250 (1)
- Interpretation:
  - IMF-supported programs per se do not have a significant impact on tax revenue.
  - IMF-supported programs with revenue conditionality have a statistically significant positive effect; estimated magnitudes imply a raise in tax revenue of about 0.7 percentage points of GDP (interpretation from Table 2 discussion).

### VI. Revenue conditionality effects by tax component (Table 3)
- IMF Conditionality, lagged estimated coefficients (Table 3):
  - Total Tax: 0.0310** (0.0154)
  - G&S: 0.0503*** (0.0209)
  - VAT: 0.0483** (0.0218)
  - Income: -0.0141 (0.0222) — not significant
  - Trade: -0.0410 (0.0281) — not significant
- Sample sizes and fit:
  - Observations: Total 1851; G&S 1599; VAT 629; Income 1718; Trade 1582
  - Number of countries vary by specification (e.g., Total 122; VAT 81)
- Quantitative implications stated in source:
  - IMF revenue conditionality could raise tax revenue by about ½ a percentage point of GDP in a given year.
  - About ¼ percentage point of GDP of the single-year gain is explained by taxes on goods and services.
  - With an average of five years of IMF-supported programs with revenue conditionality, implied cumulative revenue gain is about 2 ½ percentage points of GDP.
  - With persistence and an average of 3 years of consecutive revenue conditionality, implied gain is about 1 percentage point of GDP by the third consecutive year, with ¾ of the gain explained by taxes on goods and services.
  - On average during a program three revenue conditionalities were met, suggesting about a 1 percentage point of GDP revenue gain over the duration of a program.
- Interpretation emphasized in source:
  - Positive and statistically significant effect concentrated in taxes on goods and services, particularly VAT; effects on income taxes and trade taxes are not statistically significant.

### VII. Policy vs. administration and specificity of conditionality (Tables 4 and 5)
- Tax policy vs. tax administration conditionality (Table 4):
  - Conditionality on tax policy, lagged: Total 0.0266* (0.0152); G&S 0.0888** (0.0380); VAT 0.0822** (0.0427)
  - Conditionality on tax administration, lagged: Total 0.0275* (0.0165); G&S 0.0816** (0.0381); VAT 0.0765* (0.0438)
  - Interpretation: Tax policy and tax administration conditionalities have similar estimated impacts on total revenue; VAT coefficient for tax policy conditionality is more significant; complementarity when both are present.
- Specific vs. general conditionality (Table 5):
  - Specific conditionality, lagged:
    - G&S: 0.0645* (0.0383)
    - VAT: 0.0723* (0.0458)
    - Income: 0.0923** (0.0407)
    - CIT: 0.1412*** (0.0521)
  - General conditionality, lagged:
    - G&S: 0.0477** (0.0223)
    - VAT: 0.0478** (0.0210)
    - Income: 0.0245 (0.0779) — not significant
    - CIT: -0.0815 (0.0945) — not significant
  - Interpretation: Specific conditionality yields larger revenue gains than general conditionality; specific conditionality has positive significant effects on income taxes driven by CIT.

### VIII. Instrumenting for selection into IMF programs and alternative instruments (Table 6)
- Key IMF Conditionality, lagged coefficients (Table 6 with selection instruments):
  - Total: 0.0615** (0.0303)
  - G&S: 0.1296* (0.0732)
  - VAT: 0.0885** (0.0410)
  - Income: 0.0911 (0.0797) — not significant
  - Trade: -0.0880 (0.0886) — not significant
- Instruments used: first lag of differences in tax; international reserves (months of imports, lagged); exchange rate to US dollar (percent change, lagged); overall fiscal balance (lagged); second lags of levels in differenced equation.
- Interpretation: Results qualitatively identical to Table 3 but with larger estimated coefficients after instrumenting for probability of being in an IMF program; IMF revenue conditionality remains significantly positive for total tax revenue and G&S.

### IX. Robustness, heterogeneity, and cyclically-adjusted results
- Robustness checks reported:
  - System-GMM with instruments, fixed-effects, Heckman two-stage, inverse probability weighting, and cyclically-adjusted revenues.
- Heterogeneity findings:
  - Stronger positive effects of revenue conditionality in low-income countries and in countries eligible for concessional financing.
  - Conditionality is more effective where initial revenue ratios are below the group average and where institutional quality is stronger.
  - Revenue conditionality has no significant impact in countries with weak institutions/high corruption.
- Examples of subgroup coefficient estimates (Table 7 previewed):
  - IMF Conditionality, lagged:
    - Low income: 0.0421** (0.0230)
    - Concessional financing: 0.0325** (0.0158)
    - Middle income: 0.0248* (0.0148)
    - Strong institutions: 0.0413* (0.0245)
    - Weak institutions: 0.0060 (0.0167) — not statistically significant
- Cyclically-adjusted tax revenue (Table 9):
  - IMF Conditionality, lagged:
    - Total Tax: 0.0462*** (0.0170)
    - G&S: 0.0704*** (0.0196)
    - VAT: 0.0446** (0.0235)
    - Income tax: -0.0109 (0.0178) — not significant
    - Trade tax: 0.0076 (0.0250) — not significant

### X. Fixed-effects and treatment-effect evidence (Appendix summaries)
- Fixed-effects (Appendix Table A3) selected estimates (lagged IMF conditionality):
  - Total Tax: 0.0176** (0.0078) and 0.0203* (0.0111) across specifications.
  - G&S: 0.0382*** (0.0128) and 0.0310** (0.0034).
  - VAT: 0.0129** (0.0046) and 0.0128** (0.0076).
  - Income: not statistically significant in reported columns.
  - Trade: not statistically significant; one negative point estimate -0.0116 (0.0155).
  - Autoregressive persistence: Tax, lagged coefficients large and highly significant across specifications.
- Inverse probability weighting ATE on change in Total Tax Revenue (Appendix Table A4):
  - Year 0: 0.0175** (0.0075)
  - Year 1: 0.0195*** (0.0077)
  - Year 2: 0.0123* (0.0077)
  - Year 3: 0.0169** (0.0082)
  - Observations by year: Year 0: 1610; Year 1: 1563; Year 2: 1512; Year 3: 1461

### XI. Key quantitative and programmatic facts (as cited)
- 441 approved IMF-supported programs over the last 20 years (period covered by descriptive discussion).
- In more than 75 percent of cases where tax reform was supported by at least two consecutive years of revenue conditionality, the tax-to-GDP ratio increased compared to the year prior to conditionality.
- Over the entire sample, about 20 percent of the time countries included some type of revenue conditionality; excluding non-IMF-supported program countries, the figure is 50 percent.
- On average, countries had 5 years with revenue conditionality and 3 years of consecutive revenue conditionality.
- Revenue conditionality composition: 80 percent structural benchmarks; 20 percent prior actions.
- Distribution of conditionality by tax: G&S 56 percent; Income 32 percent; Trade 12 percent.
- Average targeted fiscal adjustment in 133 IMF-supported programs was 1.7 percent of GDP during 1993-2001 (IEO, 2003).

### XII. Policy-relevant conclusions and implications
- Revenue conditionality in IMF-supported programs is associated with higher revenue collection, particularly in low-income countries and for taxes on goods and services (including VAT).
- Targeting conditionality to specific taxes (e.g., VAT) or concrete revenue-administration measures tends to enhance effectiveness.
- Combining tax policy and tax administration conditionality is complementary and can yield substantial revenue gains.
- Conditionality is less effective where corruption is high; institutional quality matters for fiscal policy implementation.
- Distributional considerations: VAT can be regressive; mitigation strategies include exempting sensitive food items and adopting turnover thresholds; if VAT revenues finance increased social expenditures, net outcome can be progressive.
- Observed increase in revenue conditionality since 2008 coincided with expanded IMF technical assistance in the revenue area; countries appear to use conditionality to monitor implementation of tax reforms.

*Source: IMF Working Paper (content unit _wp14206 - 2013) — authors’ calculations and analysis based on GFS, WEO, OECD Revenue Statistics in Latin America, and MONA Database, IMF.*

### 2013. The results indicate that such conditionality had a positive impact on tax revenue, with

### _wp14206 - 2013. The results indicate that such conditionality had a positive impact on tax revenue, with

### I. Introduction — purpose and high-level findings
- Research question: Does revenue conditionality in IMF-supported programs affect tax revenue collection, and does the design of conditionality (tax type targeted, policy vs administration, specific vs general) matter?
- Main empirical finding summarized in source opening:
  - Revenue conditionality had a positive impact on tax revenue, with strongest improvement on taxes on goods and services, including the VAT.
  - Revenue conditionality matters more for low-income countries, particularly those where revenue ratios are below the group average.
  - Revenue conditionality appears to be more effective when targeted to a specific tax.
  - These results hold after controlling for potential endogeneity, sample selection bias, and when revenues are adjusted for economic cycle.
- JEL Classification Numbers: C33; E62; F33; H2
- Keywords: Tax revenue; structural conditionality
- IMF Author’s E-Mail Address: ECrivelli@imf.org, Sgupta@imf.org

### II. Context and descriptive evidence
- Scope of study:
  - Unbalanced panel of 126 low- and middle-income countries over the period 1993-2013.
  - T=21 years and N=126 in the panel context.
- IMF-program and conditionality prevalence:
  - Since 1993, 96 of the 126 countries in the sample had an IMF-supported program for at least 1 year.
  - Over the entire sample, about 43 percent of the time countries had IMF-supported programs.
  - IMF-supported programs were more frequent in low-income countries (about 63 percent).
  - Over the sample period, over 1,500 revenue conditionalities were met in the 441 newly approved IMF-supported programs.
  - Over the entire sample period, about 20 percent of the time countries included some type of revenue conditionality; excluding non-IMF-supported program countries, this figure is 50 percent.
  - On average, countries had 5 years with revenue conditionality and 3 years of consecutive revenue conditionality.
- Forms and composition of revenue conditionality:
  - 80 percent of revenue conditionality took the form of structural benchmarks.
  - 20 percent of conditionality took the form of prior actions.
  - Share of conditionality by main tax:
    - G&S (taxes on goods and services): 56 percent
    - Income (taxes on income): 32 percent
    - Trade (taxes on international transactions): 12 percent
- Aggregate revenue performance (descriptive):
  - Figure 2 (described): average increase in tax revenue collection by about 2 percentage points of GDP over the sample period.
  - Until 2008, revenue collection in middle-income countries increased by around 3 percentage points of GDP, on average, about 1 percentage point of GDP more than in low-income countries.
  - After 2008, low-income countries strengthened revenue collection further while middle-income countries experienced lower tax-to-GDP ratios, producing near convergence.

### III. Tabulated average tax revenue changes (Table 1, 1994-2013; in percent of GDP)
- All countries:
  - All countries: 1994-2013 = 0.13; 1994-98 = 0.03; 1999-03 = 0.29; 2004-08 = 0.34; 2009-13 = -0.36
  - IMF Program with Revenue Conditionality: 1994-2013 = 0.30; 1994-98 = 0.16; 1999-03 = 0.46; 2004-08 = 0.24; 2009-13 = 0.20
  - IMF Program without Revenue Conditionality: 1994-2013 = 0.02; 1994-98 = -0.22; 1999-03 = 0.40; 2004-08 = 0.21; 2009-13 = -0.52
  - No IMF Program: 1994-2013 = 0.12; 1994-98 = 0.11; 1999-03 = 0.15; 2004-08 = 0.41; 2009-13 = -0.45
- Middle Income Countries:
  - IMF Program with Revenue Conditionality: 1994-2013 = 0.33; 1994-98 = -0.11; 1999-03 = 0.65; 2004-08 = 0.16; 2009-13 = 0.38
  - IMF Program without Revenue Conditionality: 1994-2013 = -0.09; 1994-98 = -0.16; 1999-03 = 0.38; 2004-08 = 0.28; 2009-13 = -1.66
  - No IMF Program: 1994-2013 = 0.10; 1994-98 = 0.14; 1999-03 = 0.14; 2004-08 = 0.41; 2009-13 = -0.57
- Low Income Countries:
  - IMF Program with Revenue Conditionality: 1994-2013 = 0.36; 1994-98 = 0.51; 1999-03 = 0.18; 2004-08 = 0.33; 2009-13 = 0.12
  - IMF Program without Revenue Conditionality: 1994-2013 = 0.14; 1994-98 = -0.36; 1999-03 = 0.29; 2004-08 = 0.14; 2009-13 = 0.31
  - No IMF Program: 1994-2013 = 0.01; 1994-98 = -0.02; 1999-03 = 0.15; 2004-08 = 0.05; 2009-13 = 0.43

### IV. Data sources and variable construction
- Tax revenue data sources: IMF’s Government Finance Statistics (GFS), IMF’s World Economic Outlook (WEO), OECD’s Revenue Statistics in Latin America.
- Tax series included (relative to GDP, expressed in logs in regressions): total tax revenue, taxes on goods and services (includes VAT, excise taxes, and other consumption-related taxes), VAT, corporate income tax (CIT), personal income tax (PIT), taxes on international transactions (Trade).
- IMF-supported program and conditionality data: IMF’s Monitoring of Fund Arrangements (MONA) database.
- Binary revenue conditionality indicator: equals 1 if a country in a given year had an IMF-supported program with met revenue conditionality and 0 otherwise. If conditionality was not met, the dummy equals 0. For general (non–tax-specific) conditionality, assumed to apply to all taxes in that year.
- Lag structure: mostly the first lag of the revenue conditionality dummy is used to account for delayed revenue responses; contemporaneous inclusion not qualitatively different.

### V. Empirical specification and identification strategy
- Baseline dynamic specification (general form described in source):
  - Equation includes lagged dependent variable, revenue conditionality dummy D (lagged), control vector X, country and time fixed effects.
  - Eq.(1) estimated separately for Total Tax, G&S, VAT, Income, CIT, PIT, and Trade.
- Estimation method:
  - System-Generalized Method of Moments (system-GMM), Blundell and Bond (1998) variant used.
  - Rationale: appropriate for “small T, large N” panels and unbiased estimation including lagged dependent.
- Controls (X) drawn from literature on tax-to-GDP determinants:
  - GDP per capita, share of agriculture in value-added, trade openness (sum of imports and exports in GDP), inflation, level of external indebtedness, and other standard determinants.
- Addressing endogeneity and selection:
  - Endogeneity concern: IMF loans and revenue conditionality may respond to low tax-to-GDP ratios (reverse causality).
  - System-GMM treated potentially endogenous variables using deeper lags as instruments.
  - Sample selection bias concern: countries in IMF programs may differ systematically; identification follows literature (Barro and Lee, 2005) using instruments.
  - Instruments used for IMF revenue conditionality in system-GMM: international reserves in months of imports, change in the bilateral exchange rate to US dollar, and overall fiscal balance.
  - Diagnostics reported: Hansen and Sargan statistics, Arellano-Bond (M1 and M2) tests for serial correlation, and difference-in-Hansen test of instrument exogeneity. Robust standard errors used due to heteroskedasticity.
- Alternative estimators:
  - Appendix B reports fixed-effects and Heckman two-stage (inverse Mills ratio) specifications for robustness/comparability.

### VI. Robustness and additional analyses (summary as presented)
- Results are robust to:
  - Controlling for potential endogeneity via system-GMM.
  - Accounting for sample selection bias using instruments and Heckman-type approaches in alternative specifications.
  - Adjusting revenues for the economic cycle (cyclically adjusted revenues).
- Heterogeneity findings:
  - Stronger positive effects of revenue conditionality in low-income countries and in countries with initial revenue ratios below the group average.
  - Conditionality targeted to a specific tax tends to be more effective than general conditionality.
  - Largest impacts observed on taxes on goods and services, including the VAT.

### VII. Key quantitative and programmatic facts cited in the source
- 441 approved IMF-supported programs over the last 20 years (period covered by descriptive discussion).
- In more than 75 percent of cases where tax reform was supported by at least two consecutive years of revenue conditionality, the tax-to-GDP ratio increased compared to the year prior to conditionality.
- Over the entire sample, about 20 percent of the time countries included some type of revenue conditionality; excluding non-IMF-supported program countries, the figure is 50 percent.
- On average, countries had 5 years with revenue conditionality and 3 years of consecutive revenue conditionality.
- Over the sample, revenue conditionality composition: 80 percent structural benchmarks; 20 percent prior actions.
- Distribution of conditionality by tax: G&S 56 percent; Income 32 percent; Trade 12 percent.
- Average targeted fiscal adjustment in 133 IMF-supported programs was 1.7 percent of GDP during 1993-2001 (IEO, 2003).

### VIII. Implications highlighted by the authors (as presented in source)
- Revenue conditionality in IMF-supported programs can support implementation of structural tax reforms and is associated with higher revenue collection, particularly in low-income countries.
- Targeting conditionality to specific taxes (e.g., VAT) or concrete revenue-administration measures tends to enhance effectiveness.
- The increase in revenue conditionality since 2008 coincided with expanded IMF technical assistance in the revenue area; countries appear to use conditionality to monitor implementation of tax reforms.

*Source: IMF Working Paper (content unit _wp14206 - 2013) — authors’ calculations and analysis based on GFS, WEO, OECD Revenue Statistics in Latin America, and MONA Database, IMF.*

### Appendix B presents a probit regression for the probability of a country to have an IMF-supported program,

### _wp14206 - Appendix B presents a probit regression for the probability of a country to have an IMF-supported program,

### III. Methods / Identification and Instrumentation
- Appendix B reports a probit regression for the probability of a country to have an IMF-supported program to confirm instrument validity.
- Controlled covariates include GDP per capita and the level of external indebtedness.
- Instrument validity diagnostics used: Hansen-Over-identification (p value), Diff-in-Hansen-test of exogeneity (p value), Sargan test (not shown here but discussed).
- Estimation approaches discussed:
  - Fixed effects estimator with concern for dynamic panel bias due to lagged dependent variable.
  - System-GMM: Eq.(1) taken in differences and the lagged dependent variable instrumented with past levels.
  - Inverse probability weight regression-adjustment method (Hirano et al., 2003) as an alternative to address sample selection bias.
- Notes on instrument proliferation: Both Hansen and Sargan tests discussed; Hansen’s p-value should be high enough to reject correlation between instruments and errors but not too high; Sargan less vulnerable to proliferation but not robust to heteroskedasticity.

### III.A. Effect of IMF-supported Programs on Total Tax Revenue (Table 2)
- Hypotheses tested:
  - H1: IMF-supported program vs. No IMF-supported program
  - H2: IMF-supported program without revenue conditionality vs. No IMF-supported program
  - H3: IMF-supported program with revenue conditionality vs. No IMF-supported program
- Key regression results (Table 2, Columns 1–2):
  - Tax, lagged: 0.6749*** (0.1117) in (1); 0.6981*** (0.1135) in (2)
  - IMF Program, lagged: 0.0471 (0.0302) — not statistically significant in Column 1
  - IMF Program with revenue conditionality, lagged: 0.0539** (0.0241) in Column 2
  - IMF Program, without revenue conditionality, lagged: 0.0391 (0.0261) in Column 2 — not statistically significant
  - Trade Openness: 0.0002 (0.0009) in (1); 0.0010 (0.0007) in (2)
  - Inflation: -0.0073** (0.0036) in (1); -0.0066** (0.0033) in (2)
  - GDP Per Capita (log): 0.1113*** (0.0348) in (1); 0.0964*** (0.0350) in (2)
  - Agriculture share in Value-Added: 0.0048 (0.0031) in (1); 0.0024 (0.0024) in (2)
  - External Debt: -0.0004 (0.0009) in (1); -0.0001 (0.0006) in (2)
  - M1 (p value): 0.001 (both columns)
  - M2 (p value): 0.454 (1); 0.457 (2)
  - Hansen-Over-identification (p value): 0.674 (1); 0.405 (2)
  - Diff-in-Hansen-test of exogeneity (p value): 0.250 (1); 0.792 (2)
  - Observations: 1851 (both)
  - Number of instruments: 74 (1); 110 (2)
  - Number of countries: 122 (both)
- Interpretation:
  - IMF-supported programs per se do not have a significant impact on tax revenue.
  - IMF-supported programs with revenue conditionality have a statistically significant positive effect: estimated coefficient implies a raise in tax revenue of about 0.7 percentage points of GDP.
  - IMF-supported programs without revenue conditionality show no significant impact.

### III.B. Revenue Conditionality: Total and Component Taxes (Table 3)
- Purpose: Estimate effect of IMF revenue conditionality on total tax revenue and on components: taxes on goods and services (G&S), VAT, income taxes, trade taxes.
- Key regression results (Table 3, Columns 1–5):
  - Tax, lagged: 0.6847*** (0.1179) Total; 0.8189*** (0.0489) G&S; 0.9073*** (0.0609) VAT; 0.8646*** (0.0677) Income; 0.9083*** (0.0400) Trade
  - IMF Conditionality, lagged:
    - Total Tax: 0.0310** (0.0154)
    - G&S: 0.0503*** (0.0209)
    - VAT: 0.0483** (0.0218)
    - Income: -0.0141 (0.0222) — not significant
    - Trade: -0.0410 (0.0281) — not significant
  - Trade Openness: 0.0012 (0.0009) Total; 0.0026** (0.0013) Trade
  - Inflation: -0.0058* (0.0033) Total; large and not precisely estimated coefficients for G&S, VAT, Income, Trade with large SEs reported
  - GDP Per Capita (log): 0.0991*** (0.0368) Total; other components show mixed signs and significance
  - Agriculture share in Value-Added: small coefficients; VAT: -0.0141** (0.0059)
  - External Debt: small, not significant across columns
  - M1 (p value): 0.001 (Total) and 0.000 (others)
  - M2 (p value): 0.450 (Total); values vary across components
  - Hansen-Over-identification (p value): 0.212 (Total); 0.544 (G&S); 0.753 (VAT); 0.755 (Income); 0.813 (Trade)
  - Diff-in-Hansen-test of exogeneity (p value): values reported per column
  - Observations: 1851 (Total); 1599 (G&S); 629 (VAT); 1718 (Income); 1582 (Trade)
  - Number of instruments: 75 (Total); 113 (G&S); 71 (VAT); 76 (Income); 113 (Trade)
  - Number of countries: 122 (Total); 109 (G&S); 81 (VAT); 114 (Income); 109 (Trade)
- Quantitative implications and interpretation:
  - Estimated coefficient on total tax revenue implies IMF revenue conditionality could raise tax revenue by about ½ a percentage point of GDP in a given year.
  - Half of this gain (about ¼ percentage point of GDP) is explained by taxes on goods and services.
  - Given an average of five years of IMF-supported programs with revenue conditionality, implied cumulative revenue gain is about 2 ½ percentage points of GDP over the sample period.
  - Considering persistence (lagged dependent variable) and an average of 3 years of consecutive revenue conditionality, implied revenue gain is a full percentage point of GDP by the third consecutive year, with ¾ of the gain explained by taxes on goods and services.
  - On average during a program, three revenue conditionalities were met, suggesting a revenue gain of about 1 percentage point of GDP over the duration of a program.
  - The positive and significant effect is concentrated in taxes on goods and services, particularly VAT; effects on income taxes and trade taxes are not statistically significant.

### III.B. Interpretation: Why G&S and VAT respond more strongly
- Large share of revenue conditionality attached to taxes on goods and services (including VAT).
- VAT and broad-base consumption taxes are noted for efficiency and welfare gains and for strengthening tax administration and collection.
- Adoption and strengthening of VAT is consistent with prior literature linking VAT adoption (and IMF involvement) to improvements in tax revenue collection.
- Distributional concerns: VAT can be regressive; mitigation strategies include exempting sensitive food items and adopting turnover thresholds; if VAT revenues finance increased social expenditures, net outcome can be progressive.
- Empirical note: IMF-supported programs have been associated with positive effects on social spending in low-income countries.

### III.C. Tax Policy vs. Tax Administration Conditionality (Table 4)
- Purpose: Assess differential impact of conditionality on tax policy vs. tax administration.
- Key regression results (Table 4, Columns 1–5):
  - Tax, lagged: 0.7004*** (0.1027) Total; 0.3051*** (0.0631) G&S; 0.4750*** (0.0965) VAT; 0.8613*** (0.0542) Income; 0.9092*** (0.0362) Trade
  - Conditionality on tax policy, lagged:
    - Total: 0.0266* (0.0152)
    - G&S: 0.0888** (0.0380)
    - VAT: 0.0822** (0.0427)
    - Income: 0.0019 (0.0256)
    - Trade: 0.0154 (0.0322)
  - Conditionality on tax administration, lagged:
    - Total: 0.0275* (0.0165)
    - G&S: 0.0816** (0.0381)
    - VAT: 0.0765* (0.0438)
    - Income: -0.0055 (0.0237)
    - Trade: -0.0416 (0.0257)
  - M1 (p value): reported per column
  - M2 (p value): reported per column
  - Hansen-Over-identification (p value): 0.436 (Total); others reported per column
  - Observations: 1850 (Total); 1703 (G&S); 629 (VAT); 1718 (Income); 1702 (Trade)
  - Number of instruments: 108 (Total); 109 (G&S); 95 (VAT); 108 (Income); 109 (Trade)
  - Number of countries: 122 (Total); 114 (G&S); 81 (VAT); 114 (Income); 115 (Trade)
- Interpretation:
  - Conditionality on tax policy and on tax administration have similar estimated impacts on total revenue.
  - Coefficient on VAT for conditionality on tax policy is more significant.
  - Combined presence of both types of conditionality in a given year can yield substantial revenue gains, confirming complementarity of tax policy and administration reforms.

### III.D. Specific vs. General Revenue Conditionality (Table 5)
- Purpose: Compare “specific” revenue conditionality (tied to particular tax measures) vs. “general” revenue conditionality.
- Key regression results (Table 5, Columns 1–5):
  - Tax, lagged: 0.7703*** (0.0549) G&S; 0.9071*** (0.0537) VAT; 0.8359*** (0.0673) Income; 0.7775*** (0.0866) CIT; 0.8719*** (0.0429) Trade
  - Specific Conditionality, lagged:
    - G&S: 0.0645* (0.0383)
    - VAT: 0.0723* (0.0458)
    - Income: 0.0923** (0.0407)
    - CIT: 0.1412*** (0.0521)
    - Trade: -0.0440 (0.0695)
  - General Conditionality, lagged:
    - G&S: 0.0477** (0.0223)
    - VAT: 0.0478** (0.0210)
    - Income: 0.0245 (0.0779) — not significant
    - CIT: -0.0815 (0.0945) — not significant
    - Trade: -0.0190 (0.0924) — not significant
  - M1 (p value): 0.000 across columns
  - M2 (p value): values reported per column
  - Hansen-Over-identification (p value): high values reported per column
  - Observations: 1684 (G&S); 620 (VAT); 1699 (Income); 1442 (CIT); 1683 (Trade)
  - Number of instruments: 112 (G&S); 89 (VAT); 72 (Income); 103 (CIT); 93 (Trade)
  - Number of countries: 113 (G&S); 80 (VAT); 113 (Income); 107 (CIT); 114 (Trade)
- Interpretation:
  - Specific conditionality yields larger revenue gains than general conditionality.
  - Specific conditionality has a positive and statistically significant impact on taxes on income explained mainly by a positive effect on corporate income tax (CIT).
  - Estimated coefficient on total tax revenue (aggregating) implies a revenue gain of about 0.6 percentage points of GDP for specific conditionality; half (0.3 percentage points) explained by G&S and half by CIT.

### III.E. Instrumenting for Selection into IMF Programs / Alternative Instruments (Table 6)
- Purpose: Address selection bias explicitly by instrumenting the probability of being in an IMF program and using alternative instruments in the system-GMM.
- Key regression results (Table 6, Columns 1–5):
  - Tax, lagged: 0.7636*** (0.0857) Total; 0.8292*** (0.0388) G&S; 0.7268*** (0.1136) VAT; 0.8717*** (0.0580) Income; 0.9122*** (0.0288) Trade
  - IMF Conditionality, lagged:
    - Total: 0.0615** (0.0303)
    - G&S: 0.1296* (0.0732)
    - VAT: 0.0885** (0.0410)
    - Income: 0.0911 (0.0797) — not significant
    - Trade: -0.0880 (0.0886) — not significant
  - Trade Openness: 0.0016*** (0.0005) Total; 0.0036*** (0.0013) Trade
  - Inflation and other controls: coefficients and SEs reported per column (some small or not significant)
  - Agriculture share in Value-Added: -0.0040** (0.0017) Total
  - M1 (p value): 0.003 (Total)
  - M2 (p value): 0.267 (Total)
  - Hansen-Over-identification (p value): 0.471 (Total); others reported per column
  - Diff-in-Hansen-test of exogeneity (p value): reported per column
  - Observations: 1851 (Total); 1703 (G&S); 629 (VAT); 1718 (Income); 1683 (Trade)
  - Number of instruments: 113 (Total); 114 (G&S); 70 (VAT); 113 (Income); 112 (Trade)
  - Number of countries: 122 (Total); 114 (G&S); 81 (VAT); 114 (Income); 114 (Trade)
- Instruments used in this specification:
  - First lag of differences in tax
  - International reserves (in months of imports, lagged)
  - Exchange rate to the US dollar (percent change, lagged)
  - Overall fiscal balance (in percent of GDP, lagged)
  - Second lags of their levels in the differenced equation
- Interpretation:
  - Results are qualitatively identical to Table 3 but with larger estimated coefficients after instrumenting for probability of being in an IMF program.
  - IMF revenue conditionality shows a significantly positive effect on total tax revenue and on taxes on goods and services when alternative instruments are used.

### IV. Robustness and Heterogeneity (beginning of Section IV / Table 7 preview)
- Robustness checks include subgroup analysis by development level and institutional quality.
- Table 7 (described) presents results for:
  - Low-income countries (Column 1)
  - Countries eligible for IMF PRGT concessional financing (Column 2)
  - Middle-income countries (Column 3)
  - Countries grouped by ICRG corruption score: strong institutions (score >= 3) vs. weak institutions (score < 3) (Columns 4–5)
- Summary of previewed results:
  - Estimated coefficients on revenue conditionality are significantly positive for low- and middle-income countries and for those eligible for concessional financing.
  - The potential revenue gain is somewhat larger for low-income countries and those eligible for concessional financing (½ a percentage point of GDP) compared to other groups (text truncated before completing exact comparator numbers).

### Policy-relevant conclusions and implications reported in the text
- IMF-supported programs alone are necessary but not sufficient to improve tax revenue; revenue conditionality is a key mechanism associated with revenue increases.
- Revenue conditionality is associated with:
  - An increase in total tax revenue of about 0.0310** (0.0154) in baseline (Table 3 total tax specification).
  - A concentrated positive impact on taxes on goods and services and VAT.
  - Larger gains when conditionality is specific rather than general, and when tax policy and tax administration conditionalities are combined.
- Magnitude implications:
  - Single-year gain of about ½ percentage point of GDP attributable to revenue conditionality (Table 3).
  - Cumulative gain of about 2 ½ percentage points of GDP over five years (average program-years with revenue conditionality = five).
  - Gain of about 1 percentage point of GDP by the third consecutive year of revenue conditionality (average consecutive years = three), with ¾ of that gain explained by taxes on goods and services.
  - During a program, on average three revenue conditionalities were met, implying about 1 percentage point of GDP revenue gain over program duration.
- Distributional considerations:
  - VAT and broad-based consumption taxes can be regressive; mitigation via exemptions for sensitive items and turnover thresholds is noted.
  - If VAT revenues finance increased social expenditures, net distributional outcomes can be progressive.
  - IMF-supported programs have documented positive effects on social spending in low-income countries.

*Document: _wp14206 - Appendix B presents a probit regression for the probability of a country to have an IMF-supported program,*

### 0.4 percentage points of GDP for middle-income countries) in the first period after the

### _wp14206 - 0.4 percentage points of GDP for middle-income countries) in the first period after the

### Main conclusions
- Revenue conditionality in IMF-supported programs has a positive and statistically significant impact on tax revenue, particularly in low-income countries and for taxes on goods and services and the VAT.
- The impact is larger in low-income countries than in middle-income countries and is most pronounced where initial tax-to-GDP ratios are low.
- Revenue conditionality has no significant impact in countries with high corruption (weak institutions).

### Quantitative findings: aggregate and heterogeneous effects
- Potential increase in total tax revenue by the third year after program start:
  - Low-income countries: 1½ percentage points of GDP
  - Middle-income countries: about 1 percentage point of GDP
- Revenue gain in the first year after program start for countries with the lowest tax-to-GDP ratios (below 10 percent):
  - About 0.6 percentage points of GDP in the first year
  - About 1½ percentage points of GDP by the third year
- Small difference in revenue gain (about 0.1 percentage points of GDP) associated with IMF revenue conditionality for countries with tax-to-GDP ratios below the average (compared to above average).

### Results by institutional strength
- Table 7 evidence:
  - IMF Conditionality, lagged coefficients:
    - Low income (column 1): 0.0421** (standard error (0.0230))
    - Concessional financing (column 2): 0.0325** (0.0158)
    - Middle income (column 3): 0.0248* (0.0148)
    - Strong institutions (column 4): 0.0413* (0.0245)
    - Weak institutions (column 5): 0.0060 (0.0167) — not statistically significant
  - Tax, lagged coefficients remain high and significant across groups (e.g., Low income: 0.9626*** (0.0819); Middle income: 0.7599*** (0.0163))
- Conclusion: revenue conditionality has the largest estimated impact in countries with the strongest institutions or lowest corruption; impact for weak-institution countries is not statistically significant.

### Tax-disaggregated and cyclically-adjusted results
- Conditionality affects specific targeted taxes; effects hold after adjusting for the economic cycle.
- Table 9 (cyclically-adjusted tax revenue) shows IMF Conditionality, lagged coefficients:
  - Total Tax: 0.0462*** (0.0170)
  - Taxes on Goods & Services (G&S): 0.0704*** (0.0196)
  - VAT: 0.0446** (0.0235)
  - Income tax: -0.0109 (0.0178) — not significant
  - Trade tax: 0.0076 (0.0250) — not significant
- Tax, lagged coefficients in Table 9 (all highly significant): e.g., Total Tax 0.6515*** (0.1065); VAT 0.9380*** (0.0536).

### Robustness checks and methodology
- Sample: 126 low- and middle-income countries during 1993-2013.
- Multiple robustness checks performed:
  - Splitting sample by income level and concessional financing eligibility.
  - Splitting by initial tax-to-GDP ratio (below average; below 10 percent; above average; above 20 percent).
  - Cyclically-adjusted tax revenue estimation following Fatas and Mihov (2003, 2006) using IV with ΔY(-1) and ΔY(-2) as instruments; residuals used as discretionary tax component.
  - Fixed effects models including inverse Mills ratio to account for selection bias.
- Fixed effects results: statistically significant positive relationship between IMF revenue conditionality and total tax revenue; similar positive results for taxes on goods and services and VAT. Estimated coefficients are smaller than GMM but remain positive; correcting for sample selection bias increases the coefficient for total tax revenue.

### Data and descriptive statistics (selected items from Appendix Table A1)
- Total Tax Revenue, percent of GDP (Obs. 3444): Mean 15.30; Std. Dev. 1.62; Min 0.34; Max 61.39
  - Low-income countries (Obs. 1038): Mean 11.85; Std. Dev. 1.75; Min 0.34; Max 36.54
  - Middle-income countries (Obs. 2406): Mean 17.09; Std. Dev. 1.57; Min 1.14; Max 61.39
- Tax on Goods and Services (G&S), percent of GDP (Obs. 2051): Mean 4.76; Std. Dev. 2.19; Min 0.10; Max 58.26
- Value-added tax (VAT), percent of GDP (Obs. 908): Mean 4.02; Std. Dev. 2.09; Min 0.10; Max 19.30
- Income Tax, percent of GDP (Obs. 3050): Mean 3.68; Std. Dev. 2.14; Min 0.04; Max 50.60
- Trade Tax Revenue, percent of GDP (Obs. 2404): Mean 2.61; Std. Dev. 2.54; Min 0.05; Max 41.50
- IMF Program variable (Obs. 2647): Mean 0.44; Std. Dev. 0.50; Min 0.00; Max 1.00
  - Low-income countries (Obs. 777): Mean 0.63; Std. Dev. 0.48
  - Middle-income countries (Obs. 1870): Mean 0.35; Std. Dev. 0.48
- Revenue conditionality variable on total tax (Obs. 2647): Mean 0.19; Std. Dev. 0.39
  - Low-income countries (Obs. 777): Mean 0.28; Std. Dev. 0.45
  - Middle-income countries (Obs. 1870): Mean 0.15; Std. Dev. 0.36
- ICRG Corruption Score (Obs. 2414): Mean 2.42; Std. Dev. 0.97; Min 0.00; Max 6.00

### Policy-relevant implications
- Revenue conditionality in IMF programs can be instrumental in helping low-income countries overcome implementation challenges and capacity constraints in adopting tax reforms.
- Targeting conditionality to specific taxes is effective: conditionality tied to a specific tax affects that tax’s performance, including income taxes.
- Conditionality is less effective or ineffective where corruption is high; institutional quality matters for fiscal policy implementation.
- Countries with low initial tax-to-GDP ratios stand to gain more from revenue conditionality than countries already with high tax-to-GDP ratios (above 20 percent).

*Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14206.pdf*

### Appendix Table A3. IMF Revenue Conditionality on Tax Revenues – Fixed Effects

### Appendix Table A3. IMF Revenue Conditionality on Tax Revenues – Fixed Effects

### Fixed-effects regression results (summary of Table A3)
- Dependent variables: total tax revenue, revenue from taxes on goods and services (G&S), VAT, income, and trade, each relative to GDP.
- Key regressors reported across 10 specifications (columns (1)–(10)):
  - Tax, lagged:
    - Column (1) Total Tax: 0.5613*** (0.0672)
    - Column (2) Total Tax: 0.5724*** (0.0789)
    - Column (3) G&S: 0.7209*** (0.0397)
    - Column (4) G&S: 0.7302*** (0.0555)
    - Column (5) VAT: 0.6406*** (0.0955)
    - Column (6) VAT: 0.5694*** (0.1199)
    - Column (7) Income: 0.6817*** (0.0369)
    - Column (8) Income: 0.6687*** (0.0480)
    - Column (9) Trade: 0.8002*** (0.0267)
    - Column (10) Trade: 0.7798*** (0.0293)
  - IMF Conditionality, lagged:
    - Column (1) Total Tax: 0.0176** (0.0078)
    - Column (2) Total Tax: 0.0203* (0.0111)
    - Column (3) G&S: 0.0382*** (0.0128)
    - Column (4) G&S: 0.0310** (0.0034)
    - Column (5) VAT: 0.0129** (0.0046)
    - Column (6) VAT: 0.0128** (0.0076)
    - Column (7) Income: 0.0016 (0.0129)
    - Column (8) Income: 0.0162 (0.0167)
    - Column (9) Trade: -0.0116 (0.0155)
    - Column (10) Trade: 0.0091 (0.0162)
  - Trade Openness:
    - Column (1): 0.0014*** (0.0003)
    - Column (2): 0.0012*** (0.0003)
    - Column (3): 0.0006 (0.0005)
    - Column (4): 0.0007 (0.0007)
    - Column (5): 0.0018 (0.0011)
    - Column (6): 0.0023** (0.0010)
    - Column (7): 0.0022*** (0.0005)
    - Column (8): 0.0023*** (0.0006)
    - Column (9): 0.0010* (0.0005)
    - Column (10): 0.0003 (0.0006)
  - Inflation:
    - Column (1): -0.0014 (0.0031)
    - Column (2): -0.0064*** (0.0001)
    - Column (3): -0.0030 (0.0048)
    - Column (4): 0.0501 (0.1049)
    - Column (5): -0.0009 (0.0958)
    - Column (6): -0.1084 (0.1679)
    - Column (7): 0.0132 (0.0092)
    - Column (8): -0.0527 (0.0672)
    - Column (9): -0.0149 (0.0119)
    - Column (10): 0.0638 (0.1084)
  - GDP Per Capita (log):
    - Column (1): 0.1915*** (0.0561)
    - Column (2): 0.1687*** (0.0665)
    - Column (3): -0.0957 (0.0643)
    - Column (4): -0.1212 (0.0847)
    - Column (5): 0.2152** (0.1104)
    - Column (6): -0.0039 (0.0923)
    - Column (7): 0.1477 (0.1007)
    - Column (8): 0.2223* (0.1280)
    - Column (9): 0.0656 (0.0809)
    - Column (10): 0.0826 (0.0925)
  - Agriculture share in Value-Added:
    - Column (1): -0.0002 (0.0020)
    - Column (2): 0.0009 (0.0027)
    - Column (3): -0.0033** (0.0016)
    - Column (4): -0.0030* (0.0018)
    - Column (5): -0.0003 (0.0034)
    - Column (6): 0.0015 (0.0049)
    - Column (7): -0.0014 (0.0021)
    - Column (8): 0.0017 (0.0027)
    - Column (9): 0.0006 (0.0020)
    - Column (10): 0.0007 (0.0024)
  - External Debt:
    - Column (1): 0.0001 (0.0002)
    - Column (2): 0.0006** (0.0003)
    - Column (3): 0.0001 (0.0001)
    - Column (4): 0.0001 (0.0003)
    - Column (5): -0.0002 (0.0008)
    - Column (6): -0.0001 (0.0008)
    - Column (7): -0.0002 (0.0003)
    - Column (8): -0.0001 (0.0004)
    - Column (9): -0.0002 (0.0002)
    - Column (10): 0.0001 (0.0004)
  - Inverse Mills ratio (included in some columns):
    - Reported coefficients (where shown):
      - 0.0119 (0.0102)
      - -0.0066 (0.0091)
      - -0.0090 (0.0110)
      - 0.0089 (0.0104)
      - -0.0018 (0.0126)
  - Constant terms (selected):
    - Column (1): -0.2737 (0.4140)
    - Column (2): -0.1954 (0.4805)
    - Column (3): 1.2522** (0.4986)
    - Column (4): 1.4179** (0.6633)
    - Column (5): -1.1489 (0.8836)
    - Column (6): 0.6045 (0.7934)
    - Column (7): -0.6403 (0.7461)
    - Column (8): -1.2654 (0.9601)
    - Column (9): -0.4841 (0.5933)
    - Column (10): -0.3696 (0.6606)
- Sample sizes and fit:
  - Observations by column:
    - (1) 1850
    - (2) 1457
    - (3) 1703
    - (4) 1382
    - (5) 629
    - (6) 574
    - (7) 1718
    - (8) 1398
    - (9) 1683
    - (10) 1361
  - Number of countries by column:
    - (1) 122
    - (2) 114
    - (3) 114
    - (4) 107
    - (5) 81
    - (6) 76
    - (7) 114
    - (8) 107
    - (9) 114
    - (10) 107
  - R-squared by column:
    - (1) 0.6791
    - (2) 0.7196
    - (3) 0.8759
    - (4) 0.8545
    - (5) 0.8070
    - (6) 0.8632
    - (7) 0.7700
    - (8) 0.7191
    - (9) 0.9150
    - (10) 0.9217
- Notes:
  - Full set of year dummies included in all regressions.
  - Robust standard errors in parenthesis.
  - Significance notation: ***(**,*) indicate significance at 1(5, 10) percent.

### Interpretation of fixed-effects findings (as presented)
- IMF conditionality, lagged, shows positive and statistically significant associations with several tax revenue measures:
  - Positive and significant for Total Tax in columns (1) and (2): 0.0176** and 0.0203*.
  - Positive and significant for G&S in columns (3) and (4): 0.0382*** and 0.0310**.
  - Positive and significant for VAT in columns (5) and (6): 0.0129** and 0.0128**.
  - Not statistically significant for Income in columns (7) and (8): 0.0016 and 0.0162.
  - Not statistically significant (and negative point estimate) for Trade in column (9): -0.0116 and not significant in column (10): 0.0091.
- Autoregressive persistence in tax measures is strong (Tax, lagged coefficients large and highly significant across all specifications).

### Inverse probability weight regression-adjustment method (methodological summary)
- Purpose: address potential sample selection bias in the revenue conditionality variable.
- Approach:
  - First stage: saturated first-stage logit model to predict treatment probability (policy propensity score), using the same observables as used to calculate the Inverse Mills Ratio in the fixed-effects method.
  - Second stage: outcome regression (same linear-projection specification as in other regressions) with regression-adjustment using the propensity score to correct allocation bias.
- References for estimator: Hirano et al., 2003; Angrist et al., 2013; Acemoglu et al., 2004.

### Appendix Table A4. IMF Revenue Conditionality on Tax Revenues – Inverse Probability Weighting Method (treatment-effects estimation)
- Outcome: Change in Total Tax Revenue, in percent of GDP.
- Average Treatment Effect (ATE), Tax Conditionality (Year 0 to Year 3):
  - Year 0: 0.0175** (0.0075)
  - Year 1: 0.0195*** (0.0077)
  - Year 2: 0.0123* (0.0077)
  - Year 3: 0.0169** (0.0082)
- Observations:
  - Year 0: 1610
  - Year 1: 1563
  - Year 2: 1512
  - Year 3: 1461
- Notes:
  - Robust standard errors in parenthesis; ***(**,*) indicate significance at 1(5, 10) percent.
  - First stage uses logit with observables: international reserves (in months of imports, lagged), the exchange rate to the US dollar (percent change, lagged), the overall fiscal balance (in percent of GDP, lagged), and GDP per capita (lagged change).

*Source: Appendix Table A3 and related text, _wp14206 - Appendix Table A3. IMF Revenue Conditionality on Tax Revenues – Fixed Effects*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14206.pdf_