## wp1901

## Source details

**Canonical URL:** [wp1901](https://www.imf.org/-/media/files/publications/wp/2019/wp1901.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wp1901.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wp1901.pdf.json)

---

### I. Introduction and scope
- Renewed interest in improving governance and fighting corruption globally; African Union theme for 2018: “Winning the fight against corruption”.
- SSA countries generally lag other regions in corruption perceptions and governance:
  - "Eighty percent of SSA countries (36 out of 45) score below the global average in the Transparency International’s Corruption Perceptions Index (CPI and only 3 of the 30 SSA countries included in the International Country Risk Guide’s (ICRG) governance indicator have above average scores."
- Definitions and measurement conventions used in the study:
  - Corruption: “the abuse of public office for private gain”.
  - Governance: “institutions, mechanisms, and practices through which government power is exercised in a country, including for the management of public resources and regulation of the economy. This includes processes at the country level, including institutions-level structural arrangements” (IMF, 2017).
  - Indicators normalized 0 (worst) — 100 (best) for governance; corruption perceptions normalized 0 (best) — 100 (worst).

### II. Data and empirical strategy
- Sample and period:
  - 190 countries, 5-yearly observations over the period 1984-2015.
- Estimation approach:
  - System generalized method of moments (SGMM) model; robust two-step SGMM with country and time fixed effects, parsimonious instrument set, lags restricted to 2, instruments collapsed.
- Governance and corruption indicators used (normalized 0-100):
  - ICRG’s Political Risk Rating (ICRG).
  - Aggregate Kaufmann and Kraay Worldwide Governance Indicator (WGI).
  - Kaufmann and Kraay Control of Corruption Indicator (CCI).
  - Transparency International Corruption Perceptions Index (CPI).
- Key controls (X) and data features:
  - Initial GDP per capita (log, PPP), Gross capital formation (percent of GDP), Education (years), High inflation dummy (1 if inflation > 15 percent), Terms of trade (percent change).
  - Annual data averaged into nonoverlapping five-year periods in baseline regressions.
- Sensitivity and robustness checks include:
  - Alternative governance indicators, sample periods (3-year averages, annual), country groupings, specifications controlling for omitted variables bias, tests for endogeneity (IV and lagging governance), and serial correlation/Hansen tests.

### III. Main empirical findings (stylized facts and magnitudes)
- Overall correlation between governance/corruption and growth:
  - Weak governance and higher corruption perceptions are associated with lower GDP per capita growth; the correlation is both statistically and economically meaningful and larger in SSA.
- Magnitude of estimated growth effects (baseline SGMM):
  - Bringing governance to the world average: associated increase in GDP per capita growth of about 1 to 2 percentage points for an average SSA country (improvement of about one-standard deviation in the SSA sample).
  - Bringing corruption perceptions to the world average: associated increase in GDP per capita growth of about 1 percentage point for an average SSA country (improvement of about one-standard deviation in the SSA sample).
  - One standard deviation in the SSA sample for both governance and corruption perceptions is equivalent to 10 percentage points if all scores are normalized to 0-100.
- Key numerical baseline interpretations (System GMM results):
  - A 10-point improvement in governance (one standard deviation in SSA) associated with higher growth by between 1¼ percentage point (WGI) and 2 percentage points (ICRG).
  - A 10-point improvement in corruption perceptions in SSA associated with higher growth by 0.4-0.6 percentage points in the long run.
- Relative magnitudes across regions:
  - Impact of weak governance on growth in SSA is two to three times larger than for the average country in the rest of the world and stronger than impacts estimated for LAC and MENAAP.
- Nonlinearities and distributional aspects:
  - Nonlinear (quadratic) specifications: β1 positive and β2 negative in most specifications, implying stronger adverse correlations at low governance/high corruption levels.
  - SSA country-years with governance scores below (or corruption perceptions above) 60 — about 75 percent of SSA country-years in both samples — stand to benefit most from addressing governance and corruption.
  - Non-linear models indicate growth gains could be twice as large as baseline for SSA countries at extreme tails (e.g., 5th and 95th percentile of distributions).
- Specific governance channels (unbundled measures):
  - Rule of law, regulatory quality, and government accountability are positively correlated with growth.
  - Estimated growth gains from a 10-point improvement in specific governance components:
    - Rule of law: ¾ percentage point.
    - Voice and accountability: about 1 percentage point.
  - Gains are non-additive; complementarities across channels may amplify overall impact.

### IV. Empirical specification, identification, and endogeneity mitigation
- Baseline model specification:
  - git = β0 + β1 GOVit + β2 GOVit·SSA + β3 SSA + φ′Xit + τt + μi + νit
  - Coefficients of interest: β1 (governance effect) and β2 (additional marginal effect in SSA).
- Instrument strategy and diagnostics:
  - SGMM instruments: lagged values of endogenous and predetermined variables; collapsed instruments to limit proliferation.
  - Tests reported: absence of second-order serial correlation; Hansen test of overidentifying restrictions; χ2-tests for joint null β1 = 0 and β2 = 0.
- IV robustness (cross-section IV using 1984-2015 averages):
  - Instruments include ethnic fractionalization; settler mortality; fraction of years independent since 1776 (Independence); absolute value of latitude.
  - IV results: ethnic fractionalization, Independence, and latitude yield coefficients with expected signs and statistically significant β1 + β2 for ICRG and WGI in many specifications; settler mortality informative for WGI only.
- Additional endogeneity checks:
  - Lagging governance by one period: interaction term retains expected sign and statistical significance though overall effect smaller.
  - Model withstands most tests when sample size sufficient; typically reject null that governance coefficients are jointly zero.

### V. Robustness checks and heterogeneity analyses
- Frequency and sample variants:
  - 3-year averages: increases observations (e.g., Observations: 822; Number of countries: 119) and yields strong governance × SSA coefficients (e.g., Governance x SSA: 0.108 ***, 0.108 **, 0.149 *** across specifications).
  - Annual data: mixed results; SSA effect remains but overall sample less consistent; example ICRG annual panel: Observations: 2,197; Governance x SSA: 0.073 ***, 0.070 **, 0.111 ***, 0.065 *** in various columns.
- Regional benchmarking and country groups:
  - Models considering LAC, MENAAP, Non-SSA show governance-growth correlation present but generally weaker than in SSA.
  - Governance effect especially strong for LICs in SSA and SSA countries with below-median governance.
- Alternative controls and robustness:
  - Including conflict, resource intensity, agriculture share, openness, government consumption, different inflation thresholds: core results remain with expected signs.
  - Drop-one-country sensitivity (2 x 119 = 238 regressions): model remained remarkably stable; overall effect significant in all regressions.
- Narrow governance indicators:
  - Narrow ICRG and WGI components (voice and accountability; government effectiveness; regulatory quality; rule of law) remain positively and significantly correlated with growth in SSA; estimated overall effects in range 0.07-0.1 for channels.

### VI. Summary statistics and sample characteristics
- Overall sample (Obs = 470):
  - Growth: Mean 2.3 percent; Std. Dev. 2.6; Min. -11.8; Max. 12.2
  - GDP per capita (Log (2011 PPP$)): Mean 9.2; Std. Dev. 1.2; Min. 5.9; Max. 11.9
  - Investment (Percent of GDP): Mean 22.1; Std. Dev. 8.5; Min. 2.0; Max. 61.0
  - Education (Years): Mean 3.5; Std. Dev. 1.8; Min. 0.1; Max. 8.6
  - Inflation (Percent): Mean 6.8; Std. Dev. 9.0; Min. -6.4; Max. 84.2
  - Terms of trade (Percent): Mean 0.5; Std. Dev. 3.9; Min. -15.9; Max. 18.0
  - ICRG (0-100): Mean 67.0; Std. Dev. 14.4; Min. 25.5; Max. 96.4
  - WGI (0-100): Mean 55.5; Std. Dev. 21.9; Min. 3.2; Max. 99.1
  - CCI (0-100): Mean 53.4; Std. Dev. 22.8; Min. 1.9; Max. 95.5
  - CPI (0-100): Mean 57.7; Std. Dev. 22.8; Min. 1.9; Max. 91.8
- SSA sample (Obs = 91):
  - Growth: Mean 1.8 percent; Std. Dev. 2.8; Min. -11.8; Max. 8.1
  - GDP per capita (Log (2011 PPP$)): Mean 7.6; Std. Dev. 0.9; Min. 5.9; Max. 9.9
  - Investment (Percent of GDP): Mean 17.0; Std. Dev. 8.0; Min. 2.0; Max. 46.2
  - Education (Years): Mean 1.5; Std. Dev. 1.0; Min. 0.1; Max. 4.1
  - Inflation (Percent): Mean 8.2; Std. Dev. 10.6; Min. -6.4; Max. 84.2
  - Terms of trade (Percent): Mean 0.7; Std. Dev. 5.3; Min. -15.9; Max. 12.4
  - ICRG (0-100): Mean 55.9; Std. Dev. 11.1; Min. 25.7; Max. 78.7
  - WGI (0-100): Mean 39.7; Std. Dev. 13.5; Min. 3.2; Max. 70.4
  - CCI (0-100): Mean 68.6; Std. Dev. 11.9; Min. 35.1; Max. 95.5
  - CPI (0-100): Mean 72.2; Std. Dev. 10.6; Min. 41.0; Max. 84.8
- Note: one standard deviation in the SSA sample is just above 10 units for the four normalized indicators.

### VII. Interpretation, caveats, mechanisms, and policy implications
- Causality caveat:
  - Estimated correlations do not prove causation; endogeneity concerns are addressed via SGMM and robustness checks, but causality is not definitively established.
- Potential mechanisms for adverse effects of corruption/weak governance:
  - Higher tax evasion and lower tax revenue.
  - Increased central bank financing of government deficits.
  - Changes in composition and quality of government spending.
  - Poor lending practices and weak financial supervision.
  - Lower investment.
- Possible amplifiers in SSA:
  - Unmeasured characteristics such as limited institutional capacity and rent-seeking in resource-rich sectors may partly explain amplified growth impacts in SSA.
- Exceptions and heterogeneity:
  - Some poorly governed countries have experienced strong growth driven by other factors (e.g., natural resource wealth); good governance does not guarantee strong growth.
- Policy implications and recommended priorities:
  - No one-size-fits-all approach; country-specific strategies reflecting interactions among government, institutions, civil society, media, and the private sector.
  - Policies and institutional features that can facilitate improvements:
    - Independent and credible institutions that improve transparency and accountability and enable better monitoring and prosecution of corrupt behavior.
    - Enforcing laws aligned with best practices that level the playing field and reduce scope for rent-seeking.
    - An independent judiciary.
    - Improving regulatory quality and government effectiveness.
    - Strengthening fiscal institutions, fiscal transparency, and controls.
  - Empirical implication:
    - Significant long-term dividends from improving governance and reducing corruption are likely; however, the pace and timing of gains depend on country-specific starting points and political commitment.
    - Prior literature suggests institutional reform and political/economic liberalization may take at least three years to affect governance and corruption perceptions (Giavazzi and Tabellini, 2005).

*Source — Excerpts from the References, Introduction, Sections III–IV, Appendix, and selected tables of the IMF working paper (content unit: wp1901).*

### References _______________________________________________________________41

### References _______________________________________________________________41

### I. Introduction and scope
- Renewed interest in improving governance and fighting corruption globally; African Union theme for 2018: “Winning the fight against corruption”.
- SSA countries generally lag other regions in corruption perceptions and governance.
  - "Eighty percent of SSA countries (36 out of 45) score below the global average in the Transparency International’s Corruption Perceptions Index (CPI) and only 3 of the 30 SSA countries included in the International Country Risk Guide’s (ICRG) governance indicator have above average scores."
- Definitions and measurement conventions used in the study:
  - Corruption: “the abuse of public office for private gain”.
  - Governance: “institutions, mechanisms, and practices through which government power is exercised in a country, including for the management of public resources and regulation of the economy. This includes processes at the country level, including institutions-level structural arrangements” (IMF, 2017).
  - Indicators normalized 0 (worst) — 100 (best) for governance; corruption perceptions normalized 0 (best) — 100 (worst).

### II. Data and empirical strategy
- Sample and period:
  - 190 countries, 5-yearly observations over the period 1984-2015.
- Estimation approach:
  - System generalized method of moments (SGMM) model.
- Governance and corruption indicators used:
  - ICRG’s Political Risk Rating (ICRG).
  - Aggregate Kaufmann and Kraay Worldwide Governance Indicator (WGI).
  - Kaufmann and Kraay Control of Corruption Indicator (CCI).
  - Transparency International Corruption Perceptions Index (CPI).
- Sensitivity and robustness checks include:
  - Alternative governance indicators, sample periods, country groupings, specifications that control for omitted variables bias, and tests for endogeneity.

### III. Main empirical findings
- Overall correlation between governance/corruption and growth:
  - Weak governance and higher corruption perceptions are associated with lower GDP per capita growth; the correlation is both statistically and economically meaningful and larger in SSA.
- Magnitude of estimated growth effects:
  - Bringing governance to the world average: associated increase in GDP per capita growth of about 1 to 2 percentage points for an average SSA country (improvement of about one-standard deviation in the SSA sample).
  - Bringing corruption perceptions to the world average: associated increase in GDP per capita growth of about 1 percentage point for an average SSA country (improvement of about one-standard deviation in the SSA sample).
  - One standard deviation in the SSA sample for both governance and corruption perceptions is equivalent to 10 percentage points if all scores are normalized to 0-100.
- Relative magnitudes across regions:
  - The impact of weak governance on growth in SSA is two to three times larger than for the average country in the rest of the world and stronger than impacts estimated for LAC and MENAAP.
- Nonlinearities and distributional aspects:
  - Nonlinear correlation: stronger adverse correlations with weaker governance or higher corruption perceptions.
  - SSA country-years with governance scores below (or corruption perceptions above) 60 — about 75 percent of SSA country-years in both samples — stand to benefit most from addressing governance and corruption.
  - Non-linear models indicate growth gains could be twice as large as baseline for SSA countries at extreme tails (e.g., 5th and 95th percentile of distributions).
- Specific governance channels:
  - Rule of law, regulatory quality, and government accountability are positively correlated with growth.
  - Estimated growth gains from a 10-point improvement in specific governance components:
    - Rule of law: ¾ percentage point.
    - Voice and accountability: about 1 percentage point.
  - Gains are non-additive; potential complementarities across channels may increase overall impact.

### IV. Comparison with prior literature and quantitative benchmarks
- Consistency with other empirical results:
  - Gyimah-Brempong (2002): one standard deviation improvement in corruption in African countries would increase per capita income growth by about ¾ percentage point.
  - Gyimah-Brempong and de Camacho (2006): roughly identical estimate.
  - Ugur and Dasgupta (2011): one-unit improvement in corruption perceptions (≈ one standard deviation if measured by CCI) increases annual GDP per capita growth by about 0.6 percentage point in LICs and almost 0.9 percentage point in a mixed sample of LICs and non-LICs.
- Paper’s baseline estimates are in the same general range as other global investigations, including those that control for model uncertainty (IMF, 2018).

### V. Interpretation, caveats, and mechanisms
- Causality caveat:
  - Estimated correlations do not prove causation; endogeneity concerns are addressed via SGMM and robustness checks, but causality is not definitively established.
- Potential mechanisms for adverse effects of corruption/weak governance:
  - Higher tax evasion and lower tax revenue.
  - Increased central bank financing of government deficits.
  - Changes in composition and quality of government spending.
  - Poor lending practices and weak financial supervision.
  - Lower investment.
- Possible amplifiers in SSA:
  - Unmeasured characteristics such as limited institutional capacity and rent-seeking in resource-rich sectors may partly explain amplified growth impacts in SSA.
- Exceptions and heterogeneity:
  - Some countries perceived as poorly governed have experienced strong growth driven by other factors (e.g., natural resource wealth); conversely, good governance does not guarantee strong growth.
  - Studying exceptions could illuminate mitigating or amplifying factors but is beyond this paper’s scope.

### VI. Policy implications and recommended priorities
- No one-size-fits-all approach; country-specific strategies are required, reflecting interactions among government, institutions, civil society, media, and the private sector.
- Policies and institutional features that can facilitate improvements:
  - Independent and credible institutions that improve transparency and accountability and enable better monitoring and prosecution of corrupt behavior.
  - Enforcing laws aligned with best practices that level the playing field and reduce scope for rent-seeking.
  - An independent judiciary.
  - Improving regulatory quality and government effectiveness.
  - Strengthening fiscal institutions, fiscal transparency, and controls.
- Empirical implication:
  - Significant long-term dividends from improving governance and reducing corruption are likely; however, the pace and timing of gains depend on country-specific starting points and political commitment.
  - Prior literature suggests institutional reform and political/economic liberalization may take at least three years to affect governance and corruption perceptions (Giavazzi and Tabellini, 2005).

*Italic: Source — Excerpts from the References and Introduction section of the IMF working paper (content unit: wp1901).*

### section III looks at some stylized facts; section IV focuses on the empirics using aggregate

### wp1901 - section III looks at some stylized facts; section IV focuses on the empirics using aggregate

### III. Stylized facts — correlations and transmission channels
- Cross-country patterns:
  - Unconditional positive correlation between governance (WGI) and real GDP per capita; unconditional negative correlation between corruption perceptions (CCI) and real GDP per capita. Scatterplot trendlines reported:  
    - ln(y) = 0.0335gov + 7.5734 (governance vs ln GDP per capita)  
    - ln(y) = 0.0362gov + 6.3599 (alternate trendline reported)  
    - ln(y) = -0.0322corr + 11.142 (corruption vs ln GDP per capita)  
    - ln(y) = -0.035corr + 10.149 (alternate trendline reported)
  - SSA countries on average have weaker governance, higher corruption perceptions, and lower real GDP per capita relative to the rest of the world.
- Governance / growth correlations:
  - Positive (negative) unconditional correlation between governance (corruption perceptions) and GDP per capita growth in SSA; correlations are weaker and sometimes opposite-signed outside SSA. Reported fitted relationships for growth:  
    - growth = -0.0253gov + 4.0349  
    - growth = 0.028gov + 0.9847  
    - growth = 0.0272gov + 1.1907  
    - growth = -0.0402gov + 4.8064
- Fiscal and investment channels:
  - Higher corruption perceptions are negatively associated with education spending (Education Spending/GDP) and with public investment efficiency (Efficiency index 0 to 1). Reported regression lines:  
    - educ = -0.0233corr + 5.7253  
    - educ = -0.0503corr + 8.2234
  - The regression framework in Section IV includes schooling and total investment, capturing indirect corruption effects through these variables.
- Governance channel granularity:
  - Other governance dimensions (voice and accountability, government effectiveness, regulatory quality, rule of law) are positively correlated with GDP per capita growth and strongly correlated with corruption perceptions (see Table 1 correlations).
  - Table 1 correlation excerpts (SSA, 1995-2015): high pairwise correlations and significance levels: e.g., ICRG–WGI 0.93***; WGI–Rule of law 0.98***; CCI correlations negative as expected (e.g., WGI–CCI -0.91***). (Table shows full matrix with *** denoting significance at 1-percent level.)
- Regional attributes and potential drivers:
  - Many SSA countries are low-income countries (LICs) or resource-intensive; resource intensity and weak institutions may exacerbate corruption (resource curse hypothesis).
  - Sample attribute share chart notes: WGI below median, LICs, Resource-rich — proportions compared across SSA, MENAAP, LAC, World ex-SSA/LAC/MENAAP.

### II. Brief literature review — mechanisms and quantitative magnitudes
- Mechanisms by which corruption harms growth (literature synthesis):
  - Corruption acts like a distortionary tax on private agents (Shleifer and Vishny, 1993).
  - Corruption diverts resources away from human-capital-enhancing spending toward wasteful or capture-prone items (Tanzi and Davoodi, 1997; Mo, 2001).
  - Corruption raises business costs and uncertainty, depressing private investment (Mauro, 1995).
  - Corruption weakens fiscal institutions, undermines tax compliance, and can precipitate fiscal/political crises in extreme cases.
  - Corruption can lower education and health spending, disproportionately harming the poor and entrenching inequality and poverty.
- Quantitative magnitudes from prior studies:
  - Gyimah-Brempong (2002): a one standard deviation improvement in corruption in Africa would increase per capita income growth by about ¾ percentage point annually.
  - Ugur and Dasgupta (2011) meta-analysis: a one-unit improvement in corruption perceptions (one standard deviation as measured by CCI and ICRG) would increase annual GDP per capita growth by about 0.6 percentage point in LICs, and almost 0.9 percentage point in a mixed sample of LICs and non-LICs.
  - Regional heterogeneity noted: impact of corruption on growth stronger in SSA and LAC (Gyimah-Brempong and de Camacho, 2006).

### IV. Empirical analysis — identification strategy and data
- Estimation framework:
  - Baseline model: standard growth regression augmented for governance using an unbalanced panel of 190 countries over 1984-2015. Baseline specification (equation (1)):
    - git = β0 + β1 GOVit + β2 GOVit·SSA + β3 SSA + φ′Xit + τt + μi + νit
    - where g = real GDP per capita growth, GOV = governance or corruption perceptions, SSA = sub-Saharan Africa dummy, X = vector of controls, τ = time fixed effects, μ = country fixed effect, ν = error term.
  - Coefficients of interest: β1 (governance effect) and β2 (additional marginal effect in SSA). Expectations: β1 positive for governance (negative for corruption perceptions); stylized facts suggest β1 + β2 > β1 in absolute terms for SSA.
  - Estimator: robust two-step System GMM (SGMM) per Blundell and Bond (1998) and Roodman (2009). System includes level and first-difference equations; investment and lagged growth treated as endogenous; remaining regressors initially treated as exogenous.
  - Instrument strategy: for differenced equation use lagged values of endogenous and predetermined variables and current/lagged values of differenced exogenous variables; for level equation use lagged differences and current/lagged exogenous variables. Tests reported: absence of second-order serial correlation and Hansen test of over-identifying restrictions.
  - Rationale: SGMM suitable for small T, large N panels with country fixed effects and potentially heteroskedastic idiosyncratic errors. Interaction term GOV·SSA captures region-specific practices of corruption (centralized vs decentralized bribery) that may amplify costs in SSA.
  - Interpretation caveat: coefficients β1 and β2 measure the direct effect of governance/corruption on growth after controlling for human and physical capital and macroeconomic policy quality (some indirect effects captured by controls).
- Data and variables:
  - Sample: unbalanced panel; baseline sample covers 1995-2015 (number of observations varies with availability and specification).
  - Controls in X:
    - Initial GDP per capita: real GDP per capita PPP (log) in the year before each five-year period (to capture convergence).
    - Gross capital formation: total investment (percent of GDP), measured in first year of each five-year period to mitigate endogeneity.
    - Education: per capita years of secondary and tertiary schooling for population aged 15+.
    - High inflation dummy: equals 1 if inflation > 15 percent (top quartile of sample), 0 otherwise.
    - Terms of trade: percentage change in terms of trade.
  - Governance and corruption indicators used (normalized to 0-100; higher = better governance or higher corruption perceptions as applicable):
    - ICRG’s Political Risk Rating (ICRG): time series since 1984; covers many governance aspects; covers ~two thirds of SSA countries.
    - Aggregate Worldwide Governance Indicators (WGI): simple sum of six WGI components (voice and accountability; political stability and absence of violence/terrorism; government effectiveness; regulatory quality; rule of law; control of corruption); covers all SSA countries but available from 1996 (with interpolations for 1997, 1999, 2001).
    - Control of Corruption Indicator (CCI): WGI component, available from 1996.
    - Transparency International Corruption Perceptions Index (CPI): available from 1995 with gradually expanded country coverage.
  - Temporal averaging:
    - Annual data are averaged over nonoverlapping five-year periods in baseline regressions and most robustness tests to reflect delayed effects of governance changes on growth.
- Data caveats and measurement issues:
  - Subjective corruption perception measures may not fully capture actual corruption; cultural differences and distinctions between petty vs grand corruption can affect comparability across countries/regions.
  - High correlations across governance indicators (Table 1) reflect methodological overlap (e.g., corruption perceptions included in aggregate indices) but also support their use as comprehensive proxies in empirical work.
- Model robustness and extensions:
  - The paper tests robustness to alternative control sets and addresses endogeneity concerns (tests for exogeneity of governance/corruption perceptions reported).
  - Interaction specification and inclusion of SSA dummy allows identification of region-specific differential effects using system GMM despite time-invariant nature of the SSA dummy.

*Italic: Source — wp1901 (sections III–IV) as provided*

### Appendix provide more details about the data used in the empirical analysis.

### Appendix provide more details about the data used in the empirical analysis.

### Data and variables
- Dependent variable
  - Growth: GDP per capita growth rate, percent. Data sources: Penn World Tables 9.0, WEO. Sample period: 1980-15.
  - Note: Observations with average annual changes of ±20 percent or more over any five-year period are excluded.
- Right-hand side variables (variable — description — data sources — sample period — expected sign)
  - GDP per capita — Measured at PPP, log — Penn World Tables 9.0, WEO — 1980-14 — (-)
  - Investment — Gross capital formation, percent of GDP — Penn World Tables 9.0 — 1980-14 — (+)
  - Education — Secondary and tertiary education, years — Barro-Lee database — 1980-15 — (+)
    - Observations for the 5-year period 2011-15 were obtained by extrapolating Barro-Lee data.
  - Inflation — 1 if inflation > 15 percent, 0 otherwise — WEO — 1980-15 — (-)
    - The dummy variable takes value 1 if average annual inflation exceeds 15 percent over a given five-year period, and zero otherwise.
  - Terms of trade — Annual percent change — WEO and WDI — 1980-15 — (+)
  - ICRG — Governance indicator, 0-100 — ICRG — 1984-15 — (+)
  - WGI — Governance indicator, 0-100 — Kaufmann and Kraay — 1996-15 — (+)
  - CCI — Corruption perceptions, 0-100 — Kaufmann and Kraay — 1996-15 — (-)
  - CPI — Corruption perceptions, 0-100 — Transparency International — 1995-15 — (-)
    - The normalization of CPI ensures that there is no level shift due to the methodological change introduced by Transparency International in 2012.

### Summary statistics (main variables used in the regressions)
- Overall sample (Obs = 470)
  - Growth: Mean 2.3 percent; Std. Dev. 2.6; Min. -11.8; Max. 12.2
  - GDP per capita (Log (2011 PPP$)): Mean 9.2; Std. Dev. 1.2; Min. 5.9; Max. 11.9
  - Investment (Percent of GDP): Mean 22.1; Std. Dev. 8.5; Min. 2.0; Max. 61.0
  - Education (Years): Mean 3.5; Std. Dev. 1.8; Min. 0.1; Max. 8.6
  - Inflation (Percent): Mean 6.8; Std. Dev. 9.0; Min. -6.4; Max. 84.2
  - Terms of trade (Percent): Mean 0.5; Std. Dev. 3.9; Min. -15.9; Max. 18.0
  - ICRG (0-100): Mean 67.0; Std. Dev. 14.4; Min. 25.5; Max. 96.4
  - WGI (0-100): Mean 55.5; Std. Dev. 21.9; Min. 3.2; Max. 99.1
  - CCI (0-100): Mean 53.4; Std. Dev. 22.8; Min. 1.9; Max. 95.5
  - CPI (0-100): Mean 57.7; Std. Dev. 22.8; Min. 1.9; Max. 91.8
- SSA sample (Obs = 91)
  - Growth: Mean 1.8 percent; Std. Dev. 2.8; Min. -11.8; Max. 8.1
  - GDP per capita (Log (2011 PPP$)): Mean 7.6; Std. Dev. 0.9; Min. 5.9; Max. 9.9
  - Investment (Percent of GDP): Mean 17.0; Std. Dev. 8.0; Min. 2.0; Max. 46.2
  - Education (Years): Mean 1.5; Std. Dev. 1.0; Min. 0.1; Max. 4.1
  - Inflation (Percent): Mean 8.2; Std. Dev. 10.6; Min. -6.4; Max. 84.2
  - Terms of trade (Percent): Mean 0.7; Std. Dev. 5.3; Min. -15.9; Max. 12.4
  - ICRG (0-100): Mean 55.9; Std. Dev. 11.1; Min. 25.7; Max. 78.7
  - WGI (0-100): Mean 39.7; Std. Dev. 13.5; Min. 3.2; Max. 70.4
  - CCI (0-100): Mean 68.6; Std. Dev. 11.9; Min. 35.1; Max. 95.5
  - CPI (0-100): Mean 72.2; Std. Dev. 10.6; Min. 41.0; Max. 84.8
- Note: one standard deviation in the SSA sample is just above 10 units for the four normalized indicators.

### Baseline results and magnitudes
- OLS estimates
  - Positive correlation between governance and growth for the overall sample.
  - Coefficient on interaction term for SSA (β2) not statistically significant in OLS, but joint null that both governance coefficients equal zero is rejected.
  - OLS results do not address endogeneity; schooling sign contradicts theory in OLS.
- System GMM baseline (robust two-step SGMM with country and time fixed effects; parsimonious instrument set, lags restricted to 2, instruments collapsed)
  - All coefficients on core controls have expected signs; nearly all statistically significant.
  - Coefficient on interaction term for SSA (β2) positive and statistically significant.
  - Specifications pass tests for absence of serial correlation and validity of instruments.
  - Rejection of null that β1 and β2 are jointly zero.
- Key interpreted magnitude
  - A 10-point improvement in governance—equivalent to a one standard deviation in the SSA sample—would be associated with higher growth by between 1¼ percentage point (WGI) and 2 percentage points (ICRG).
  - This impact is about thrice and twice larger than for the overall sample, respectively.
  - Total impact including indirect effects through covariates (education, inflation) likely larger than β1 + β2.

### Mitigating endogeneity
- Lagging governance by one period
  - Interaction term retains expected sign and is statistically significant; overall effect still statistically significant but smaller.
- Instrumental variables (cross-section IV using averages over 1984-2015)
  - IVs used: ethnic fractionalization; settler mortality; fraction of years independent since 1776 (Independence); absolute value of latitude.
  - In IV regressions initial income (averages over 1980-1983), inflation, education, and terms of trade treated as exogenous; governance and investment treated as endogenous.
  - Findings:
    - Ethnic fractionalization: coefficients have expected sign and β1 + β2 statistically significant for both ICRG and WGI.
    - Settler mortality: informative for WGI only; overall marginal effect statistically significant at 5 percent.
    - Independence: coefficients have expected sign; β1 + β2 relatively large and statistically significant for both ICRG and WGI.
    - Latitude: coefficients have expected sign; overall effect for SSA statistically significant for both ICRG and WGI.
  - Other instruments explored (incidence of malaria, legal origins, religious and language fractionalization) yield weaker and less reliable results, partly due to small samples.
- Summary of endogeneity tests
  - Model withstands most tests when sample size sufficient.
  - Typically reject null that governance coefficients are jointly zero.
  - Estimated correlation between governance and growth in SSA appears robust and larger than for full sample.

### Robustness checks
- Regional benchmarking
  - Models: (2) baseline sample with one region at a time; (3) region-only samples; (4) multiple regions.
  - Regions: Non-SSA (developing and emerging non-SSA), LAC, MENAAP.
  - Findings: strong correlation between governance and growth for all regions in model (2); model (3) broadly holds with lower precision in small samples.
  - Model (4): mixed evidence — interaction term significant for LAC and MENAAP when jointly considered with SSA in WGI, but generally governance appears to hinder growth in other regions to a lower extent than in SSA.
- Country groups (models (5) and (6))
  - Groups: LICs; LICs in SSA; LOW (governance below sample median); LOW in SSA.
  - Overall effect of governance on growth statistically strong across almost all country groups and especially strong for LICs in SSA or SSA countries with below-median governance.
- Alternative controls tested
  - Lagged growth; level of inflation; inflation dummy with 20 percent threshold; resource intensity dummy; conflict; agriculture intensity; openness; government consumption.
  - Overall effect remains significant with expected sign in all specifications.
  - Controlling for resource intensity (resource-rich defined as 20 percent or more of exports primary products) does not change baseline result; coefficient on resource-intensity dummy is negative.
- Data frequency and outliers
  - Using 3-year averages and annual data: overall effect for SSA remains statistically significant with the right sign in most cases.
  - Drop-one-country sensitivity (2 x 119 = 238 regressions): model remained remarkably stable; overall effect significant in all regressions.

### Unbundling governance (channels and corruption)
- Narrower governance indicators (exclude corruption perceptions and some political/economic outcomes)
  - Narrow ICRG components: military in politics, religious tensions, ethnic tensions, law and order, democratic accountability, bureaucracy quality.
  - Narrow WGI components: voice and accountability; government effectiveness; regulatory quality; rule of law.
  - Narrow measures remain positively and significantly correlated with growth in SSA; magnitude slightly smaller than baseline.
- WGI channels
  - Each of the four selected WGI channels shows a strong correlation with growth in SSA; overall effect in range 0.07-0.1.
  - Because channels have similar weight on WGI, improving one aspect could be associated with higher growth; improving multiple aspects could yield larger payoffs due to complementarities.
- Corruption perceptions (CCI and CPI)
  - Corruption perceptions specifications pass tests for serial correlation and validity of instruments.
  - No statistically significant correlation between corruption perceptions and growth for the average country in the overall sample.
  - Statistically strong negative correlation between corruption perceptions and growth for MENAAP region; stronger and negative correlation for SSA.
  - A 10-point improvement in corruption perceptions in SSA (about one standard deviation) would be associated with higher growth by 0.4-0.6 percentage points in the long run.

### Non-linearity
- Quadratic specification added: g_it = β0 + β1 GOV_it + β2 GOV_it^2 + controls.
  - Expected signs: β1 positive; β2 negative.
  - Estimated models (ICRG governance; CCI corruption) for full and SSA samples:
    - Linear term (β1) positive; squared term (β2) negative in most specifications.
    - Generally reject null that β1 and β2 are jointly zero in most models.
- Implications
  - For governance: non-linear estimates imply positive and stronger effects for SSA at low governance levels relative to linear model.
    - Governance below 60 for over 75 percent of SSA country-years; imposing linearity can understate gains.
  - For corruption: corruption particularly harmful to growth for countries with corruption perceptions above 60 (about 75 percent of SSA country-years); when corruption is moderate or small the impact on growth is less severe.
  - Example magnitudes: for countries at the 5th (95th) percentile of SSA distribution of governance (corruption perceptions)—normalized scores around 30 and 85 respectively—the overall effect would be about twice as large as implied by linear model.

### Concluding remarks
- Strong correlation between governance/corruption perceptions and growth in sub-Saharan Africa; robust across specifications, samples, and endogeneity controls.
- Correlations do not necessarily imply causation but indicate potentially significant economic gains from improving governance and reducing corruption in SSA.
- Countries most likely to gain: SSA LICs and those with the lowest performance in governance and perceived corruption.
- Policy implication: Reducing corruption and strengthening governance remain high-priority and potentially high-payoff objectives, but the process is demanding and time-consuming.

### Appendix: sample period, country coverage, instruments, and estimation notes
- Sample period for baseline regressions: data spanning 20 years (1995–2015), averaged into five nonoverlapping five-year periods.
- Raw sample covers 190 countries; regression samples smaller depending on data availability and specification.
- Country groups: SSA, EMDE, MENAAP, CIS, LIC, etc.; full country lists and ISO/IFS codes provided in table A1.
- Instrumentation and estimation details
  - Baseline SGMM uses a parsimonious number of instruments, lags restricted to 2, instruments “collapsed” to limit instrument proliferation (Roodman, 2009).
  - Example baseline instrument set (17 instruments in baseline model): income, education, inflation, terms of trade, governance, governance × SSA interaction, SSA dummy, time dummies (3), lagged growth and investment (lags 1 and 2), differenced lagged growth and investment (lag 1), and the constant term.
  - Serial correlation tests reported are for second-order serial correlation; Hansen test reported for overidentifying restrictions.
  - IV regressions use robust two-stage least squares; panel regressions use robust two-step system GMM with country and time fixed effects.

*Source: wp1901 - Appendix provide more details about the data used in the empirical analysis.*

### 2015. Africa has a disproportionate number of fragile states, or half of the countries

### 2015. Africa has a disproportionate number of fragile states, or half of the countries identified by the Fund as fragile states (IMF 2015b)

### Main empirical findings on governance and growth
- Controlling for conflicts, there is a statistically valid correlation between governance and SSA growth (column 5).
- Adding traditional controls—share of agriculture in value added, trade openness, and government consumption—leaves the baseline results virtually unchanged (columns 6 to 8).
- Using 3-year averages nearly doubles the number of observations; most controls remain statistically significant and with the expected sign. The overall effect of governance on growth is strong in the full sample and stronger in the SSA sample (Table A7).
- Using annual data yields similar results for SSA but not for the overall sample (Table A8).
- Results should be taken with caution because the authors could not completely remove the presence of serial autocorrelation. The authors consider 5-year averages a reasonable choice given the data and literature practice.

### Selected regression estimates (high-level, as reported)
- From Table A6 (Real GDP per capita growth; WGI):
  - Governance: 0.101 ***, 0.107 ***, 0.108 ***, 0.070 ***, 0.114 ***, 0.112 ***, 0.105 ***, 0.108 ***
  - Governance x SSA: 0.068 *, 0.079, 0.084 *, 0.129 *, 0.075, 0.092 **, 0.091 **, 0.091 **
  - Observations: 469, 468, 470, 470, 362, 451, 470, 470
  - Number of countries: 119, 119, 119, 119, 92, 117, 119, 119
  - Number of instruments: 17, 17, 17, 18, 18, 18, 18, 18
  - Serial correlation test (p-value): 0.941, 0.636, 0.529, 0.668, 0.827, 0.101, 0.468, 0.442
  - Hansen test (p-value): 0.624, 0.161, 0.143, 0.01, 0.343, 0.011, 0.282, 0.276
  - χ2-test for H0: β1 = 0 and β2 = 0: 45.59 ***, 31.62 ***, 39.75 ***, 19.95 *, 28.06 ***, 25.4 ***, 46.68 ***, 48.43 ***

- Alternative presentation in table (lower panel of Table A6):
  - Governance: 0.034 ***, 0.036 **, 0.039 **, 0.017, 0.051, 0.028, 0.035 **, 0.036 **
  - Governance x SSA: 0.072 ***, 0.079 ***, 0.078 ***, 0.077 **, 0.084 ***, 0.087 ***, 0.088 ***, 0.087 ***
  - Observations: 469, 468, 470, 470, 362, 451, 470, 470
  - Number of countries: 119, 119, 119, 119, 92, 117, 119, 119
  - Number of instruments: 17, 17, 17, 18, 18, 18, 18, 18
  - Serial correlation test (p-value): 0.716, 0.416, 0.477, 0.469, 0.698, 0.610, 0.467, 0.431
  - Hansen test (p-value): 0.379, 0.157, 0.130, 0.064, 0.351, 0.001, 0.253, 0.220
  - χ2-test for H0: β1 = 0 and β2 = 0: 21.08 ***, 12.51 ***, 17.58 ***, 7.482 **, 23.07 ***, 12.3 ***, 18.87 ***, 17.8 ***

- Table A7 (3-year averages; ICRG and WGI specifications):
  - Governance: 0.103 ***, 0.099 ***, 0.104 ***, 0.050 ***, 0.030 *, 0.038 **, 0.038 **, 0.032 ***
  - Governance x SSA: 0.108 ***, 0.108 **, 0.149 ***, 0.071 **, 0.073, 0.115 ***
  - Observations: 822 (all columns)
  - Number of countries: 119 (all columns)
  - Number of instruments: 16, 18, 18, 18, 16, 18, 18, 18
  - Serial correlation test (p-value): 0.0193, 0.0192, 0.0221, 0.0187, 0.0396, 0.0385, 0.0384, 0.0370
  - Hansen test (p-value): 0.316, 0.392, 0.417, 0.253, 0.685, 0.741, 0.695, 0.730
  - χ2-test for H0: β1 = 0 and β2 = 0: 29.73 ***, 28.95 ***, 19.12 ***, 23.30 ***, 17.73 ***, 24.11 ***

- Table A8 (annual data; ICRG and WGI specifications):
  - ICRG panel (top):
    - Governance: -0.02 *, -0.02 *, -0.02, -0.03, -0.01, -0.00, -0.00, -0.01
    - Governance x SSA: 0.073 ***, 0.070 **, 0.111 ***, 0.065 ***, 0.066 **, 0.078 **
    - Observations: 2,197 (all columns)
    - Number of countries: 119 (all columns)
    - Number of instruments: 31, 33, 33, 33, 30, 32, 32, 32
    - Serial correlation test (p-value): 0.779, 0.787, 0.799, 0.781, 0.268, 0.330, 0.269, 0.281
    - Hansen test (p-value): 0.00107, 0.100, 0.106, 0.135, 0.00572, 0.110, 0.127, 0.136
    - χ2-test for H0: β1 = 0 and β2 = 0: 9.736 ***, 6.609 **, 13.99 ***, 8.174 **, 4.457, 5.674 *

### Methodology notes and robustness
- Regressions are estimated using robust two-step system GMM estimator and include country and time fixed effects.
- Instruments: the right-hand side controls (lagged), lagged values of dependent variable, and a dummy for resource intensity.
- Serial correlation test refers to second-order serial correlation; Hansen test refers to overidentifying restrictions.
- χ2-test reports test statistics and significance for null hypothesis that coefficients on governance are jointly zero.
- For the 3-year regressions the authors interpolated Barro and Lee data on schooling (available every five years) to mitigate missing observations.
- For yearly regressions the authors lag all control variables, except the inflation dummy, to mitigate endogeneity problems.

### Interpretation and cautions
- Governance improvements are associated with higher real GDP per capita growth, with larger estimated effects in Sub-Saharan Africa (SSA) than in the full sample across several specifications.
- The statistical significance and magnitude of governance coefficients vary across frequency (5-year averages, 3-year averages, annual data) and governance measures (ICRG, WGI).
- Presence of residual serial autocorrelation in some specifications suggests caution in interpreting point estimates; 5-year averages are preferred by the authors given data limitations and comparability with existing literature.

*Source: wp1901 - 2015. Africa has a disproportionate number of fragile states, or half of the countries identified by the Fund as fragile states (IMF 2015b).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wp1901.pdf_