## wpiea2019177-print-pdf

## Source details

**Canonical URL:** [wpiea2019177-print-pdf](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019177-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019177-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019177-print-pdf.pdf.json)

---

### Introduction and research question
- Central theme: enhancing tax collection for development, particularly in low-income countries (LICs).
- Key motivation: delivering on the SDG agenda would require additional spending in 2030 of US$0.5 trillion (15 percentage points of GDP) in LICs.
- Research question: Does causality run from state institutions → tax capacity, or from tax capacity → state institutions?
- Definitions:
  - Tax capacity: ratio of tax revenue to GDP (Gaspar et al., 2016a).
  - State institutions: rule of law, effectiveness of government, corruption control — broadly, traditions by which authority is exercised.

### Data, sample, and institutional measures
- Sample and panel:
  - Panel of 110 non-resource-rich countries covering 1996 to 2017.
  - Panel dimensions: 22 years and 110 countries.
- Income-group disaggregation:
  - LICs: 23 countries
  - LMICs: 29 countries
  - UMICs: 27 countries
  - HICs: 31 countries
- Tipping-point split:
  - Below tipping point: 18 countries
  - Above tipping point: 92 countries
  - Tax tipping point estimated at 12¾ percent of GDP.
- Institutional measures:
  - Primary indicators: government effectiveness (GE), control of corruption (CP), freedom to trade internationally (FTI).
  - Composite institutional index constructed using principal component analysis (PCA) with 13 institutional indicators.
  - PCA-derived weight shares: about 35 percent to corruption perception index, 34 percent to governance indicators, 31 percent to economic freedom index.
- Data sources (Appendix I):
  - IMF (World Economic Outlook database for tax revenue data)
  - World Bank (World Development Indicators)
  - Transparency International (corruption perception index)
  - Fraser Institute (economic freedom index)

### Methodology and econometric approach
- Rationale:
  - Institutions variables are highly persistent; emphasis on long-run causality rather than short-run dynamics.
- Tests and methods:
  - Panel unit root tests: Levin, Lin and Chu (LLC); Im, Pesaran and Shin (IPS).
  - Cointegration: Johansen (1988) combined with Fisher (1932) panel approach (Johansen-Fisher panel cointegration).
  - Panel VECM (vector error correction model) following Hurlin and Venet (2001) to detect long-run causality where time series are limited.
  - Engle and Granger (1987) error correction mechanism used for robustness (Appendix IV).
  - PVAR motivations and use of lag lengths: optimal lag length ranges used: 1 to 3 for all variables.
- Hypotheses tested via ECTs (error-correction term coefficients):
  - H0_1: β_e = 0 (no long-run causality from tax capacity to institutions)
  - H0_2: β_y = 0 (no long-run causality from institutions to tax capacity)
  - H0_3: β_z = 0 (no long-run causality involving GDP per capita)
- Interpretation rules:
  1. Reject H0_1 and accept H0_2 → unidirectional causality from tax capacity to institutions.
  2. Accept H0_1 and reject H0_2 → unidirectional causality from institutions to tax capacity.
  3. Accept both H0_1 and H0_2 → no long-run causal relationship.
  4. Reject both H0_1 and H0_2 → feedback (bidirectional) causal relationship between institutions and tax capacity.

### Main empirical findings (long-run causality and cointegration)
- Cointegration:
  - Panel tests detect two cointegrating equations across many groupings.
- Long-run causality:
  - Strong evidence of long-run bi-directional causality between tax capacity and institutions across all country groups (LICs, LMICs, UMICs, HICs).
  - Across all panels and country groupings, results reject the null that tax capacity does not cause institutions and reject the null that institutions do not cause tax capacity.
  - All reported coefficients are statistically significant at least at 10 percent level.
- Cross-sectional and stylized correlations:
  - Tax capacity and government effectiveness: correlation 0.62.
  - Tax capacity and corruption perception indexes: correlation 0.65.
  - Tax capacity and freedom to trade internationally: correlation 0.52.
  - Tax capacity and derived overall institutions index: correlation 0.60.
- Cross-sectional patterns:
  - HICs tend to collect larger tax revenues (as share of GDP) and have better institutions.
  - Some LICs, LMICs, and UMICs collect above the tax tipping point despite poor institutional quality.
  - Countries below the tax tipping point (12¾ percent of GDP) have weak institutions without exception.
- Revenue-specific institutions (RA-FIT):
  - Overall revenue institution index averaged from RA-FIT indicators (equal weights).
  - General state institutions and revenue-specific institutions are highly and positively correlated in cross-section.

### Impulse responses: magnitudes and dynamics
- Variables are I(1); impulses from the cointegrated VECM have permanent effects.
- General dynamics:
  - Impulse responses generally: positive and lasting, peaking after two to three years and thereafter remaining positive at a constant, increasing, or decreasing rate.
- Full-sample magnitudes after three years:
  - A 1 percent positive shock to institutions → permanent increase in tax capacity of about 0.8 percent after three years.
  - A 1 percent positive shock to tax capacity → permanent increase in institutions of about 0.3 percent after three years.
- Cross-country group magnitudes:
  - A 10 percent improvement in the government effectiveness indicator will lead to average increases of about 0.1, 0.05, 0.02, and 0.14 index point in the overall institutions index in LICs, LMICs, UMICs, and HICs respectively.
  - A 1 percent improvement in tax capacity implies average increases in tax revenue of about:
    - 0.12 percent of GDP for LICs
    - 0.17 percent of GDP for LMICs
    - 0.2 percent of GDP for UMICs
    - 0.25 percent of GDP for HICs
- Average annual growth rates (full sample, 1996 to 2017):
  - Institutions: about 0.4 percent
  - Tax capacity: about 1.35 percent

### Heterogeneity by income group and by tax tipping point
- HICs:
  - Relatively low quantitative interaction between tax capacity and institutions; limited marginal revenue gain from further institution improvements.
- UMICs and LMICs:
  - Larger impacts from shocks in both directions; institution shocks raise tax capacity and tax capacity shocks raise institutions.
- LICs:
  - Largest response of tax capacity to shocks in institutions.
  - A shock in tax capacity has relatively small impacts on institutions compared with LMICs.
  - Suggests scope for initial gains through strengthening institutions and vulnerability to slippages.
- Above vs. Below TP:
  - Impacts of shocks are positive in both groups.
  - Countries below TP exhibit responses similar to LICs; institution shocks have larger impacts on tax capacity than in countries above TP.
  - Findings support existence of a tax capacity tipping point where the level of tax capacity matters for institutions and growth.

### Robustness checks and auxiliary analyses
- Robustness tests employed:
  - Panel disaggregation by income groups and tax tipping point.
  - Composite institutional index using PCA.
  - Panel VECM augmented with a fixed effects error correction mechanism (Engle and Granger, 1987).
  - Inclusion of political instability as an additional variable; results confirm bidirectional causality.
  - Confirmation of Hypothesis 1 (tax capacity → institutions) with different sample periods (1996-2017).
- Additional procedures:
  - Engle-Granger four-step procedure applied (Appendix IV) with reported ECT statistics showing widespread statistical significance (many entries with ***, **, or * as printed).
  - Panel unit root and cointegration test tables reported (Appendix III) with trace and maximum-eigenvalue statistics for groups (numeric values printed in source tables).

### Theoretical channels and hypotheses tested
- Hypothesis 1 (tax capacity → institutions):
  - Mechanisms:
    - Increased tax capacity raises scope for spending to strengthen institutions.
    - More resources allow governments to implement reforms that strengthen institutions.
    - Rising tax revenue creates constituencies demanding accountability, fostering institutional improvement.
  - Empirical note: IMF (2019) estimate quoted: improvement in control of corruption by one-third of a standard deviation associated with an increase of 1.2 percentage points in government revenue as a share of GDP.
- Hypothesis 2 (institutions → tax capacity):
  - Mechanisms:
    - Better state institutions improve revenue administration and enforcement, boosting tax capacity.
    - Weak institutions can hinder fiscal reforms and entrench low-tax equilibria.
  - Empirical literature: several studies support this direction.

### Policy implications, sequencing, and recommendations
- Vicious vs. virtuous cycles:
  - Countries with low tax capacity and weak institutions may be trapped in a vicious cycle: weak institutions → low tax capacity → further weakening of institutions.
  - Simultaneous reforms can create a virtuous cycle: enhanced tax capacity generates funds to improve institutions; better institutions enhance tax capacity.
- Sequencing of reforms:
  - Short-term revenue-specific policy and administrative changes (e.g., tax rate, tax exemptions, and risk management) can boost tax revenue in the short-term.
  - Proceeds from short-term revenue gains can finance more expensive institutional reforms that lead to broader and longer-term improvements in state institutions.
  - The specific composition and sequencing require country-specific analysis; some cases warrant immediate revenue administration institutional reform initiatives.
- Policy prescription:
  - Intensify efforts to develop institutions and raise tax capacity simultaneously, as they reinforce each other.
  - Implement a strong and coherent medium-term revenue strategy (MTRS) as a framework for structural tax reforms.
  - Under Hypothesis 1 emphasis: prioritize reform measures that produce immediate short-term rises in tax capacity (examples noted in source: simplifying the tax system; curbing exemptions; reforming indirect taxes; better managing compliance risks through taxpayer segmentation and strengthening large taxpayer offices).
  - Under Hypothesis 2 emphasis: focus on strengthening institutions to move to a higher-equilibrium tax capacity.
  - Use improved tax proceeds efficiently to support institutional improvement and to increase public demand for accountability and equitable use of tax revenues.
  - Capacity building support from international partners can help.

### Appendix highlights and selected statistics
- Appendix I — Table A1.2: Average Annual Growth Rates in Tax Capacity and Institutions (values printed as percentage change, not percent of GDP):
  - All: Mean(%) 1.35 Median(%) 0.74 Standard Deviation(%) 12.21 ; Institutions Mean(%) 0.42 Median(%) 0.23 Standard Deviation(%) 3.54 Pool Observation 2,310
  - LICs: Mean(%) 3.40 Median(%) 1.86 Standard Deviation(%) 23.14 ; Institutions Mean(%) 0.30 Median(%) 0.34 Standard Deviation(%) 3.95 Pool Observation 483
  - LMICs: Mean(%) 1.40 Median(%) 0.83 Standard Deviation(%) 8.92 ; Institutions Mean(%) 0.72 Median(%) 0.26 Standard Deviation(%) 4.75 Pool Observation 609
  - UMICs: Mean(%) 0.93 Median(%) 1.07 Standard Deviation(%) 6.81 ; Institutions Mean(%) 0.52 Median(%) 0.39 Standard Deviation(%) 3.13 Pool Observation 567
  - HICs: Mean(%) 0.16 Median(%) 0.24 Standard Deviation(%) 3.61 ; Institutions Mean(%) 0.16 Median(%) 0.06 Standard Deviation(%) 1.77 Pool Observation 651
  - AboveTP: Mean(%) 1.13 Median(%) 0.62 Standard Deviation(%) 11.57 ; Institutions Mean(%) 0.41 Median(%) 0.21 Standard Deviation(%) 3.10 Pool Observation 1,932
  - BelowTP: Mean(%) 2.46 Median(%) 1.81 Standard Deviation(%) 15.02 ; Institutions Mean(%) 0.46 Median(%) 0.40 Standard Deviation(%) 5.26 Pool Observation 378
- Appendix II — PCA construction:
  - Scaling: governance indicators (range -2.5 to +2.5) converted to 0–10 by adding +2.5 and multiplying by 2; corruption perception index divided by 10 to convert to 0–10.
  - Institutional index derived from first eigenvectors (loading matrix) as linear combination of Cpi, Efi, Gi with eigenvector weights χ1, χ2, χ3.
- Appendix IV — Engle-Granger ECT results:
  - Reported ECT statistics show widespread rejection of null of no Granger-causality at 1, 5, or 10 percent levels across All countries, LICs, LMICs, UMICs, HICs, Above TP, and Below TP (exact numeric entries appear in printed Appendix tables).
- Appendix III — Cointegration statistics:
  - Panel cointegration tables (trace and max-eigenvalue statistics) reported for All countries and each subgroup with numeric values as printed in the source table.

*Source: IMF working paper content (wpiea2019177-print-pdf).*

### REFERENCES ___________________________________________________________________________________ 25

### REFERENCES (wpiea2019177-print-pdf)

### Introduction and research question
- Central theme: enhancing tax collection for development, particularly in low-income countries (LICs).
- Key motivation: delivering on the SDG agenda would require additional spending in 2030 of US$0.5 trillion (15 percentage points of GDP) in LICs.
- Research question: Does causality run from state institutions → tax capacity, or from tax capacity → state institutions?
- Definitions:
  - Tax capacity: ratio of tax revenue to GDP (Gaspar et al., 2016a).
  - State institutions: rule of law, effectiveness of government, corruption control — broadly, traditions by which authority is exercised.

### Data and empirical approach
- Sample: panel of 110 non-resource-rich countries covering 1996 to 2017.
- Income-group disaggregation:
  - LICs: 23 countries
  - LMICs: 29 countries
  - UMICs: 27 countries
  - HICs: 31 countries
- Tipping-point split:
  - Below tipping point: 18 countries
  - Above tipping point: 92 countries
  - Tax tipping point estimated at 12¾ percent of GDP.
- Institutional measures:
  - Three primary indicators used: government effectiveness, control of corruption, freedom to trade.
  - Composite institutional index constructed using principal component analysis (PCA) with 13 institutional indicators.
- Methodology:
  - Panel vector error correction model based on Hurlin and Venet (2001) to detect long-run causality in panel data where long time series are limited.
  - Multivariate causality testing focused on long-run relationships rather than short-run dynamics.

### Main empirical findings
- Long-run causality:
  - Strong evidence of long-run bi-directional causality between tax capacity and institutions across all country groups.
  - Interpretation: simultaneous improvement in institutions and tax capacity yields best results; reinforcing feedback exists.
- Heterogeneity by income group:
  - Relatively low quantitative interaction between tax capacity and institutions in HICs.
  - Generally higher effects in developing countries (LICs, LMICs, UMICs).
- Tipping-point relevance:
  - Findings support the existence of a tipping point in tax-to-GDP levels (Gaspar et al., 2016a), above which significant acceleration of growth and development is observed.
- Stylized correlations:
  - Tax capacity and government effectiveness: correlation 0.62.
  - Tax capacity and corruption perception indexes: correlation 0.65.
  - Tax capacity and freedom to trade internationally: correlation 0.52.
  - Tax capacity and derived overall institutions index: correlation 0.60.
- Cross-sectional patterns:
  - HICs tend to collect larger tax revenues (as share of GDP) and have better institutions.
  - Some LICs, LMICs, and UMICs collect above the tax tipping point despite poor institutional quality.
  - Countries with very low tax capacity (apparently below 12¾ percent of GDP) have weak institutions without exception.
- Revenue-specific institutions:
  - In cross-section analysis using RA-FIT, general state institutions and revenue-specific institutions are highly and positively correlated—countries with good state institutions tend to have good revenue institutions as well.

### Theoretical channels and hypotheses tested
- Hypothesis 1 (tax capacity → institutions):
  - Mechanisms:
    - Increased tax capacity raises scope for spending to strengthen institutions.
    - More resources allow governments to implement reforms that strengthen institutions.
    - Rising tax revenue creates constituencies demanding accountability, fostering institutional improvement.
  - Empirical literature: mixed evidence; some case studies and regional evidence provide tentative support but effects are not guaranteed and depend on reform character.
- Hypothesis 2 (institutions → tax capacity):
  - Mechanisms:
    - Better state institutions improve revenue administration and enforcement, boosting tax capacity.
    - Weak institutions can hinder fiscal reforms and entrench low-tax equilibria.
  - Empirical literature: several studies support this direction; IMF (2019) estimate quoted: improvement in control of corruption by one-third of a standard deviation associated with an increase of 1.2 percentage points in government revenue as a share of GDP.

### Policy implications and recommendations
- Vicious vs. virtuous cycles:
  - Countries with low tax capacity and weak institutions may be trapped in a vicious cycle: weak institutions → low tax capacity → further weakening of institutions.
  - Simultaneous reforms can create a virtuous cycle: enhanced tax capacity generates funds to improve institutions; better institutions enhance tax capacity.
- Policy prescription:
  - Intensify efforts to develop institutions and raise tax capacity simultaneously, as they reinforce each other.
  - To enhance tax capacity and enable institutional building, implement a strong and coherent medium-term revenue strategy (MTRS) as a framework for structural tax reforms.
  - Under Hypothesis 1 emphasis: prioritize reform measures that produce immediate short-term rises in tax capacity (examples noted in the source: simplifying the tax system; curbing exemptions; reforming indirect taxes; better managing compliance risks through taxpayer segmentation and strengthening large taxpayer offices).
  - Under Hypothesis 2 emphasis: focus on strengthening institutions to move to a higher-equilibrium tax capacity.

### Box: State institutions and revenue-specific institutions (RA-FIT cross-section)
- Data source: IMF Revenue Administration Fiscal Information Tool (RA-FIT).
- Construction: overall revenue institution index averaged from RA-FIT indicators with equal weights.
- RA-FIT indicators used (summarized list):
  - Key compliance risk; automated risk profiling for payments and returns; use of electronic payment methods; formal public service delivery standards; published strategic and business plans; annual reports; use of external auditors and internal assurance mechanisms; enterprise-wide risk policy; existence of code of conduct; provision of e-services (integrated taxpayer accounts, online taxpayer applications, electronic invoicing); use of computer-based information systems for processing employer, financial institution, government agency, international exchange, insurance, online trading, assets leasing, and VAT invoicing data.
- Result: positive and statistically significant cross-sectional correlation/regression linking general state institutions to revenue-specific institutions.

*Source: IMF working paper content (wpiea2019177-print-pdf).*

### 14. Table 2 shows several notable features reflecting expected variations in tax capacity

### 14. Table 2 shows several notable features reflecting expected variations in tax capacity and state institutions

### Variation in tax capacity and state institutions (summary of Table 2 and text)
- Computed standard deviation (across time and cross-sections) are significantly different from each other, suggesting that much of the variations in both tax capacity and state institutions are cross-country.
- Variations in tax capacity are also significantly larger than state institutions.
- Clear variation across country groups in tax capacity, with LICs and HICs having the lowest cross-section and time variations respectively.
- State institutions remain highly persistent over time and with not much cross-country variations among country groups.
- Looking at countries above and below the tax tipping point, the variations suggest that many of the LICs are still operating below the tax threshold and low state institutions.
- Period covered for the Table 2 summary: 1996-2017.

### Key statistics and groupings reported in the source text
- Panel dimensions: 22 years and 110 countries.
- Income groups referenced: LICs, LMICs, UMICs, HICs.
- Country grouping by tax tipping point: AboveTP, BelowTP.
- Optimal lag length ranges used: 1 to 3 for all variables.
- Unit root and integration order reported: institutions, tax capacity and GDP per capita are I(1) variables.
- Cointegration test outcome: tests detect two cointegrating equations across many of the groupings.
- Statistical significance threshold referenced for panel-causality results: at least at 10 percent level.

### Major empirical tests and methodology
- Panel techniques and motivations:
  - Panel vector autoregressive (PVAR) methodology examined to reveal contemporaneous effects; PVAR follows Hurlin and Venet (2001) and Vidangos (2009).
  - Homogenous causality hypothesis from Hurlin and Venet (2001) is adopted (assumes coefficients identical for all countries and different from zero; PVAR specification therefore does not include fixed effects).
  - Multivariate panel causality approach: institutions, tax capacity (tax revenue to GDP ratio), and real GDP per capita are included as a vector to account for common driving factors.
- Unit root and cointegration testing:
  - Panel unit root tests applied: Levin, Lin, and Chu (2002) (LLC) and Im, Pesaran, and Shin (2003) (IPS).
  - Cointegration test employed: Johansen (1988) combined with Fisher (1932) panel approach developed in Maddala and Wu (1999) (Johansen-Fisher panel cointegration).
  - Engle and Granger (1987) error correction mechanism used for robustness checks capturing fixed effects (results presented in Appendix IV according to text).
- Rationale for focusing on long-run causality:
  - Institutions variables are highly persistent; short-run/bivariate causality techniques (e.g., Dumitrescu and Hurlin (2012)) are limited and may be implausible for slow-moving institutions.
  - The paper therefore emphasizes long-run causality testing via panel Vector Error Correction Model (VECM).

### Panel VECM specification and inference
- Variables in system (as given):
  - Institutional variables: government effectiveness (GE), corruption perception (CP), freedom to trade internationally (FTI), and the composite institutional index (GE_INDEX) — represented generally as institutional variables.
  - Tax capacity: tax revenue to GDP ratio.
  - GDP per capita.
- PVAR representation and PVAR equations (1)-(3) are specified for institutions, tax capacity, and GDP per capita with lagged vectors and error terms as in the source.
- Panel VECM equations (4)-(6) are estimated in first differences with lagged error-correction terms (ECTs) to capture long-run adjustment:
  - The ECT coefficients (β_e, β_y, β_z) represent the rate at which dependent variables adjust back to long-run equilibrium after shocks.
  - Error correction terms are defined explicitly as linear combinations of lagged levels (long-run cointegrating relationships) with long-run coefficients (σs).
- Hypotheses tested via ECTs:
  - H0_1: β_e = 0 (no long-run causality from tax capacity to institutions)
  - H0_2: β_y = 0 (no long-run causality from institutions to tax capacity)
  - H0_3: β_z = 0 (no long-run causality involving GDP per capita)
- Interpretation of hypothesis test combinations (as provided):
  1. Reject H0_1 and accept H0_2 → unidirectional causality from tax capacity to institutions.
  2. Accept H0_1 and reject H0_2 → unidirectional causality from institutions to tax capacity.
  3. Accept both H0_1 and H0_2 → no long-run causal relationship.
  4. Reject both H0_1 and H0_2 → feedback (bidirectional) causal relationship between institutions and tax capacity.
- Identification of cointegrating vectors:
  - Two cointegrating equations detected; the paper focuses on identifying Equation 4 and 5 (institutions and tax capacity) with four restrictions imposed to fully identify cointegrating vectors (Equation 7 shows the normalized long-run part).

### Empirical results (Panel-causality tests and main findings)
- Cointegration: The panel tests confirm the presence of cointegrating relationships for equations (1), (2), and (3); tests detect two cointegrating equations across many groupings.
- Panel VECM results:
  - Long-run multivariate panel causality tests (via t-statistics on ECT coefficients) reveal bidirectional causality between tax capacity and all four institutional variables.
  - Across all panels and country groupings, results reject the null that tax capacity does not cause institutions and reject the null that institutions do not cause tax capacity.
  - Bidirectional causality is identified across all income groups (LICs, LMICs, UMICs, HICs).
  - Results are robust to grouping by countries above and below the tax tipping point: irrespective of the level of tax capacity and institutions a country operates in, institutions and tax capacity cause each other.
  - All reported coefficients are statistically significant at least at 10 percent level.
- Interpretation and implications from results:
  - Changes in institutional variables cause changes in tax capacity; changes in tax capacity cause changes in institutional variables (feedback loop).
  - The bidirectional causality supports the notion that state institutions are a good reflection of revenue-specific institutions; improvements in state institutions can be expected to lead to changes in tax capacity.
  - Practical implication noted: there may be limited marginal revenue gain from institution improvements in strong-institution countries (e.g., HICs), while weak-institution countries may gain more from additional institution-building efforts.

*Source: wpiea2019177-print-pdf - 14. Table 2 shows several notable features reflecting expected variations in tax capacity and state institutions (excerpt).*

### 28. The results further suggest a sequencing of policy and/or institutional reforms for

### wpiea2019177-print-pdf - 28. The results further suggest a sequencing of policy and/or institutional reforms for

### Sequencing of policy and institutional reforms
- Short-term revenue-specific policy and administrative changes (e.g., tax rate, tax exemptions, and risk management) can boost tax revenue in the short-term.
- Proceeds from short-term revenue gains can finance more expensive institutional reforms that lead to broader and longer-term improvements in state institutions.
- The specific composition and sequencing of reforms requires country-specific analysis; some cases warrant immediate revenue administration institutional reform initiatives.

### Empirical findings on causality between tax capacity and institutions
- The paper confirms a long-run bidirectional causality between tax capacity and state institutions across all country groups.
- Inclusion of GDP per capita in multivariate analysis improves robustness and plays an important role in establishing dual-causality between institutions and tax capacity.
- The analysis focuses on long-term causality; results differ from Prichard and Leonard (2010) due to different empirical approaches (fixed effects with five-year lags vs. long-run causality focus here).
- The study does not investigate a specific tipping point value that accelerates institutional development; rather, it uses the tipping point threshold to detect whether reinforcement occurs when a country operates below or above the tax tipping point.

### Robustness checks
- Panel disaggregation by income groups and tax tipping point.
- Construction of a composite institutional index using PCA.
- Methodological augmentation: panel VECM with a fixed effects error correction mechanism (Engle and Granger, 1987) confirming bidirectional causality.
- Inclusion of political instability as an additional variable confirming bidirectional causality.
- Confirmation of hypothesis 1 (tax capacity → institutions) with different sample periods (1996-2017).
- Note: Results indicate direction of causality, not magnitude.

### Impulse response analysis: magnitude and dynamics
- Variables are nonstationary [I(1)]; impulses from the cointegrated VECM have permanent effects.
- Impulse responses generally reveal positive and lasting effects, peaking after two to three years and thereafter remaining positive at a constant, increasing, or decreasing rate.
- Full-sample magnitudes after three years:
  - A 1 percent positive shock to institutions → permanent increase in tax capacity of about 0.8 percent after three years.
  - A 1 percent positive shock to tax capacity → permanent increase in institutions of about 0.3 percent after three years.
- Cross-country group magnitudes:
  - A 10 percent improvement in the government effectiveness indicator will lead to average increases of about 0.1, 0.05, 0.02, and 0.14 index point in the overall institutions index in LICs, LMICs, UMICs, and HICs respectively.
  - A 1 percent improvement in tax capacity implies average increases in tax revenue of about:
    - 0.12 percent of GDP for LICs
    - 0.17 percent of GDP for LMICs
    - 0.2 percent of GDP for UMICs
    - 0.25 percent of GDP for HICs
- Average annual growth rates (full sample, 1996 to 2017):
  - Institutions: about 0.4 percent
  - Tax capacity: about 1.35 percent

### Heterogeneity by income group and by tax tipping point
- HICs:
  - Both directions exhibit limited effects throughout the period.
  - Suggests that beyond a certain institutional level, further improvements have limited impact on tax capacity and vice versa.
- UMICs and LMICs:
  - Larger impacts from shocks in both directions; institution shocks raise tax capacity and tax capacity shocks raise institutions.
- LICs:
  - Largest response of tax capacity to shocks in institutions.
  - A shock in tax capacity has relatively small impacts on institutions compared with LMICs.
  - Large responses of tax capacity to institutional shocks suggest scope for initial gains through strengthening institutions and vulnerability to slippages.
- Above vs. Below tax tipping point (TP):
  - Impacts of shocks are positive in both groups.
  - Countries below TP exhibit responses similar to LICs; institution shocks have larger impacts on tax capacity than in countries above TP.
  - Findings support the existence of a tax capacity tipping point where the level of tax capacity matters for institutions and growth.

### Policy implications
- For developing countries with low tax capacity and weak institutions:
  - Exit the vicious cycle by starting with short-term tax policy and administrative reforms (e.g., increasing tax rates, removing exemptions, improving auditing and risk management) while working on longer-term institution building.
  - Use improved tax proceeds efficiently to support institutional improvement and to increase public demand for accountability and equitable use of tax revenues.
  - Simultaneous efforts to raise tax capacity and develop institutions can create a virtuous cycle; capacity building support from international partners can help.
  - A Medium-Term Revenue Strategy involving coherent and structured reforms of tax policy, customs administration, and legal measures can guide sustainable enhancement of tax capacity.

*Source: wpiea2019177-print-pdf (selected content).*

### REFERENCES

### wpiea2019177-print-pdf - REFERENCES

### Major cited literature
- Governance, corruption, institutions, and taxation are the central literatures cited, including works by Abed & Gupta (2002), Acemoglu & Robinson (2008), Acemoglu, Johnson & Robinson (2001), North (1990), Rodrik, Subramanian & Trebbi (2004), Mauro (1995), Knack & Keefer (1995), Besley & Persson (2009, 2013, 2014), and many IMF and World Bank working papers and staff notes.
- Empirical methods and time-series/panel econometrics references cited include Bai (1993), Engle & Granger (1987), Johansen (1988), Harris (1997), Gonzalo & Ng (2001), Lutkepohl (2005), Dumitrescu & Hurlin (2012), Im, Pesaran & Shin (2003), Levin, Lin & Chu (2002), Maddala & Wu (1999), and Enders (2015).
- Literature on tax capacity, tax effort, and tax revenue mobilization is cited extensively: Bird et al. (2008), Brautigam (2008), Burgess & Stern (1993), Ghura (1998), Gupta (2007), Le, Moreno-Dodson & Bayraktar (2012), Gaspar et al. (2016a, 2016b, 2019), Akitoby et al. (2019), Keen (2012), and others.
- Studies linking energy consumption, environment, and growth are cited: Asafu-Adjaye (2000), Apergis & Payne (2009, 2010), Coondoo & Dinda (2002), Oh & Lee (2004), Abdullah & Morley (2014).
- Experimental and micro-evidence on taxation and accountability: Martin (2013); on state capacity and taxation in Africa: Prichard & Leonard (2010), Bornhorst, Gupta & Thornton (2008).

### Appendix I — Data summary and sources
- Data sources explicitly used:
  - IMF (World Economic Outlook database for tax revenue data)
  - World Bank (World Development Indicators for GDP and governance data)
  - Transparency International (corruption perception index)
  - Fraser Institute (economic freedom index)
- Methodological notes from Appendix I:
  - An unbalance panel estimation technique is adopted.
  - GDP per capita measured in real terms (2010 prices) United States dollars, converted at market exchange rates.
  - The panel is divided by income groups following the 2016 World Bank Atlas method.
  - Panel further divided into countries below and above the so-called tax tipping point.
  - Non-resource rich countries selected by IMF (2012) criteria: on average either natural resource revenue or exports of at least 20 percent of total fiscal revenue and exports, respectively.

### Appendix I — List of countries (as presented)
- High-Income Countries:
  - Austria Greece Poland Uruguay
  - Barbados Hungary Portugal
  - Belgium Ireland Singapore
  - Cyprus Israel Slovak Republic
  - Czech Republic Italy Slovenia
  - Denmark Japan Spain
  - Estonia Luxembourg Sweden
  - Finland Malta Switzerland
  - France Netherlands United Kingdom
  - Germany New Zealand United States
- Upper-Middle-Income Countries:
  - Argentina Gabon Romania
  - Belarus Guyana Serbia
  - Bosnia and Herzegov Jamaica South Africa
  - Brazil Lebanon Thailand
  - Bulgaria Mauritius Turkey
  - China Montenegro Turkmenistan
  - Colombia Namibia Tuvalu
  - Costa Rica Panama
  - Croatia Paraguay
  - Dominican Republic Peru
- Lower-Middle-Income Countries:
  - Armenia India Philippines
  - Bangladesh Jordan Sao Tome and Principe
  - Cabo Verde Kenya Solomon Islands
  - Djibouti Kiribati Tajikistan
  - Egypt, Arab Re Kyrgyz Republic Tunisia
  - El Salvador Lesotho Ukraine
  - Georgia Morocco Uzbekistan
  - Ghana Myanmar Vanuatu
  - Guatemala Nicaragua Vietnam
  - Honduras Pakistan
- Low-Income Countries:
  - Afghanistan Haiti Togo
  - Benin Liberia Uganda
  - Burkina Faso Madagascar Zimbabwe
  - Burundi Malawi
  - Central African Republic Mozambique
  - Comoros Nepal
  - Eritrea Rwanda
  - Ethiopia Senegal
  - Gambia, The Sierra Leone
  - Guinea-Bissau Tanzania

### Appendix I — Table A1.2: Average Annual Growth Rates in Tax Capacity and Institutions
- Note: Values are in percentage change and not percent of GDP.
- Table entries (as printed):
  - Mean(%) Median(%) Standard Deviation(%) Mean(%) Median(%) Standard Deviation(%)
  - All 1.350.7412.210.420.233.542,310
  - LICs3.401.8623.140.300.343.95483
  - LMICs1.400.838.920.720.264.75609
  - UMICs0.931.076.810.520.393.13567
  - HICs0.160.243.610.160.061.77651
  - AboveTP1.130.6211.570.410.213.101,932
  - BelowTP2.461.8115.020.460.405.26378
  - Country Groups Tax Capacity State Institutions (Index) Pool Observation

### Appendix II — Deriving Overall Institutional Index (PCA procedure and weights)
- Indicators included:
  - Average governance indicators: corruption control, political stability, government effectiveness, voice and accountability, rule of law, regulatory quality. Governance indices range from -2.5 to +2.5.
  - Average economic freedom index: size of government, legal system and property rights, sound money, freedom to trade internationally, and regulations. Economic freedom indices range from 0 to 10.
  - Corruption perception index: ranges from 0 to 100.
- Scaling to common range:
  - Selected common range: 0 to 10 (same as average economic freedom index).
  - To convert governance indicators (range -2.5 to +2.5) to 0–10: add +2.5 to all series and multiply by 2.
  - To convert corruption perception index to 0–10: divide all series by 10.
- Aggregation and formula:
  - The first eigenvectors (loading matrix) from the PCA were used as weights.
  - The institutional index was derived from the following linear combination (as printed):
    - CpiEfi
      GiIndexInst
      321
      _χχ
      χ
      ++ =
      (A1)
  - Where 32 1 ,,χχχ are the eigenvectors (weights) from the PCA and Cpi, Efi, Gi, are average governance indicators, average economic freedom index, and the corruption perception index respectively.
  - PCA-derived weight shares (as reported): about 35 percent of the weights is attributed to corruption perception index, 34 percent to governance indicators and the remaining 31 percent to economic freedom index.

### Appendix III — Unit root and cointegration tests (method and results as presented)
- Unit root testing:
  - Tests used: Levin, Lin and Chu (LLC); Im, Pesaran and Shin (IPS); and related panel unit root tests.
  - Table A3.1: Panel unit roots test results (source: Author’s calculations). The table presents LLC and IPS statistics, first differences and levels for regions/variables, and critical values; entries appear in the printed table as numeric blocks corresponding to variables such as Inst, InstD, FTI, CP, GE, Taxgdp, GDP across All countries, LICs, LMICs, UMICs, HICs, Above TP, Below TP.
- Cointegration testing:
  - Table A3.2: Panel Cointegration test results (source: Author’s calculations).
  - Note printed: "/1 Reports the significant cointegrating ranks. The null hypothesis of no cointegration is rejected when the statistics is greater than the critical values."
  - Table entries (trace statistics, 5% critical value, maximum eigenvalue statistics, 5% critical value) are provided for country groups and variable combinations, including:
    - All countries: multiple rows with numeric values such as 166.93 29.80 112.16 21.13; 54.77 15.49 54.42 14.26; 190.74 29.80 126.26 21.13; 64.47 15.49 64.34 14.26; 296.97 29.80 195.22 21.13; 101.75 15.49 101.75 14.26; 121.72 29.80 98.43 21.13; 23.29 15.49 23.17 14.26.
    - LICs: rows with values such as 82.54 35.19 45.39 22.30; 37.15 20.26 26.93 15.89; 80.99 35.19 46.87 22.30; 34.12 20.26 21.05 15.89; 116.98 35.19 61.79 22.30; 55.18 20.26 30.54 15.89; 60.06 35.19 35.60 22.30; 24.46 20.26 21.77 15.89.
    - LMICs, UMICs, HICs, Above TP, Below TP: analogous sets of trace and maximum-eigenvalue statistics are reported in the table for combinations GDP, Taxgdp & GE; GDP, Taxgdp & CP; GDP, Taxgdp & FTI; GDP, Taxgdp & D_Inst, etc., with numeric values as printed in the source table.

*Source: REFERENCES and Appendices (Appendix I–III) from wpiea2019177-print-pdf*

### Appendix IV. Engle-Granger Procedure for Causality Testing

### Appendix IV. Engle-Granger Procedure for Causality Testing

### Engle-Granger four-step procedure (as applied)
- i. Test for stationarity among the variables. The same procedure of testing panel unit root in the study is adopted.
- ii. Estimate the long run equilibrium relationship among the variables. A panel OLS fixed-effect model is estimated simultaneously for each variable.
- iii. Estimate the error correction model. Using the residual from (ii), an error correction model of Equation 4 and 5 above is estimated. According to Enders (2015), other than the ECT, all procedures developed for a VECM (Equation 4 and 5) is applicable to the system represented by the Engle-Granger error correction model.
- iv. Assess the ECT. The ECT is assess in the same manner as the VECM-ECT.

### Table A4: Engle-Granger Long Run Panel Causality Test Results (excerpted values)
Note: *, ** and *** indicate the rejection of null hypothesis of no granger-causality and with statistical significance at 10, 5, and 1 percent level respectively.

- All countries
  - --  
  - --  
  - ----  
  - -18.27***  
  - --  
  - --  
  - ----  
  - -16.29***  
  - ----  
  - ----  
  - -19.35***  
  - --------  
  - -14.11*  
  - -16.00***-15.94***-17.00***-16.13***  
  - --

- LICs
  - --  
  - ------  
  - -7.55***  
  - --  
  - ------  
  - -4.62***  
  - ----  
  - --  
  - --  
  - -7.96***  
  - ----  
  - --  
  - --  
  - -4.69***  
  - -7.01***-4.39***-6.78***  
  - -6.88***  
  - --

- LMICs
  - --  
  - ----  
  - --  
  - -9.15***  
  - --  
  - ----  
  - --  
  - -8.66***  
  - ------  
  - --  
  - -11.96***  
  - --------  
  - -9.04***  
  - -7.47***-8.02***-8.19***  
  - -7.69***  
  - --

- UMICs
  - ----  
  - ----  
  - -9.08***  
  - ------  
  - --  
  - -8.25***  
  - --------  
  - -2.62***  
  - ----  
  - ----  
  - -7.02***  
  - -9.84***-10.05***  
  - -2.84***-9.59***  
  - --

- HICs
  - ----  
  - --  
  - --  
  - -7.45***  
  - --------  
  - -8.57***  
  - --------  
  - -8.46***  
  - ----  
  - ----  
  - -7.15***  
  - -10.87***-11.05***-10.71***-11.05***  
  - --

- Above TP
  - --------  
  - -14.62***  
  - --------  
  - -14.10***  
  - ------  
  - --  
  - -14.92***  
  - --  
  - ------  
  - -11.00***  
  - -14.81***-14.72***-14.96***  
  - -14.68***  
  - --

- Below TP
  - --------  
  - -4.59***  
  - --------  
  - -4.58***  
  - --------  
  - -9.65***  
  - --------  
  - 6.41***  
  - -5.90***  
  - -5.84***-5.49***-5.96***  
  - --

(Columns correspond to Source of causation: Error correction term (ECT) for dependent variables: Taxgdp_it, GE_it, CP_it, FTI_it, InstD_it, Inst_it as presented in the original table layout.)

### Appendix V. Long Run Bivariate Panel Causality Analysis — Table A5 (excerpted values)
Note: *, ** and *** indicate the rejection of null hypothesis of no granger-causality and with statistical significance at 10, 5, and 1 percent level respectively.

- All countries
  - --  
  - --  
  - --  
  - --  
  - -2.23**  
  - ----  
  - --  
  - --  
  - -3.61***  
  - --  
  - --  
  - --  
  - --  
  - -14.17***  
  - --  
  - ----  
  - --  
  - -1.21  
  - -11.10***-10.03***-1.00-10.35***  
  - --

- LICs
  - --  
  - ------  
  - -0.45  
  - --  
  - --  
  - --  
  - --  
  - -4.85***  
  - ----  
  - ----  
  - -7.00***  
  - --  
  - --  
  - --  
  - --  
  - -1.12  
  - -6.89***-2.84***-0.87-5.86***  
  - --

- LMICs
  - ------  
  - --  
  - -2.61***  
  - ------  
  - --  
  - -6.25***  
  - ----  
  - ----  
  - -10.40***  
  - --  
  - ------  
  - -1.93*  
  - 4.81***-0.74  
  - -2.74***-5.44***  
  - --

- UMICs
  - --  
  - ----  
  - --  
  - -3.34**  
  - ----  
  - --  
  - --  
  - -5.00***  
  - ----  
  - ----  
  - -6.27***  
  - ----  
  - --  
  - --  
  - -1.2  
  - -4.46***-3.77***-4.62***-4.59***  
  - --

- HICs
  - --------  
  - -2.04**  
  - ------  
  - --  
  - -4.33***  
  - ----  
  - --  
  - --  
  - -6.28***  
  - --  
  - ----  
  - --  
  - -3.31***  
  - -0.92-1.33-1.45-2.50***  
  - --

- Above TP
  - --  
  - ----  
  - --  
  - -2.92***  
  - ----  
  - ----  
  - -1.37  
  - ----  
  - ----  
  - -12.90***  
  - ------  
  - --  
  - -0.54  
  - -9.00***-9.05***  
  - -2.35***-9.21***  
  - --

- Below TP
  - --------  
  - -1  
  - --  
  - ------  
  - -6.44***  
  - --------  
  - -0.39  
  - --------  
  - -1.26  
  - -6.78***-2.25**-6.22**-6.00***  
  - --

(Columns correspond to Source of causation: Error correction term (ECT) for dependent variables as in the original bivariate layout.)

### Appendix VI. Long Run Panel Causality Test (Controlling for Political Stability) — Table A6 (excerpted values)
Note: *, ** and *** indicate the rejection of null hypothesis of no granger-causality and with statistical significance at 10, 5, and 1 percent level respectively. The table excludes political stability and GDP per capita interactions with tax capacity and institutions.

- All countries
  - ----  
  - ----  
  - -8.86***  
  - ----  
  - ----  
  - -10.45***  
  - --  
  - ----  
  - --  
  - -15.23***  
  - --  
  - --  
  - ----  
  - -2.42*  
  - -9.83***-8.58***  
  - -9.11***  
  - -8.85***  
  - --

- LICs
  - ----  
  - --  
  - --  
  - -4.69***  
  - ----  
  - --  
  - --  
  - -5.79***  
  - ----  
  - --  
  - --  
  - -6.34***  
  - --  
  - ----  
  - --  
  - -2.30**  
  - -6.22***-5.40***-4.97***-5.58***  
  - --

- LMICs
  - ----  
  - ----  
  - -5.22***  
  - ------  
  - --  
  - -5.84***  
  - --  
  - ----  
  - --  
  - -6.83***  
  - --  
  - ----  
  - --  
  - -2.27*  
  - -3.90***-3.16***-4.26***  
  - -4.18***  
  - --

- UMICs
  - --  
  - ------  
  - -3.36***  
  - --  
  - ------  
  - -6.16***  
  - --  
  - ----  
  - --  
  - -6.16***  
  - --------  
  - -1.82*  
  - -5.19***  
  - -5.04***-5.42***  
  - -5.49***  
  - --

- HICs
  - --  
  - ------  
  - -3.43***  
  - --------  
  - -4.80***  
  - ------  
  - --  
  - -5.52***  
  - ----  
  - ----  
  - -3.05***  
  - -2.44***-3.36***-3.35***  
  - -2.82**  
  - --

- Above TP
  - ------  
  - --  
  - -8.89***  
  - --------  
  - -8.53***  
  - --  
  - ------  
  - -13.62***  
  - ----  
  - ----  
  - -1.68*  
  - -8.21***-8.36***  
  - -7.45***-8.47***  
  - --

- Below TP
  - --  
  - ------  
  - -3.52***  
  - --------  
  - -6.19***  
  - ----  
  - ----  
  - -7.95***  
  - --------  
  - -1.69*  
  - -4.97***-6.91***-6.92***-7.47***  
  - --

(Columns correspond to Source of causation: Error correction term (ECT) for dependent variables as in the original four-variable layout controlling for political stability.)

*Source: wpiea2019177-print-pdf - Appendix IV. Engle-Granger Procedure for Causality Testing*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019177-print-pdf.pdf_