## wp17126 — Summary Statistics

## Source details

**Canonical URL:** [wp17126 — Summary Statistics](https://www.imf.org/-/media/files/publications/wp/2017/wp17126.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17126.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17126.pdf.json)

---

### Introduction and theoretical framework
- Focus: relationship between education expansion and income inequality; education expansion seen as instrument to reduce income inequality and inequality of opportunity and to support growth.
- Human capital earnings approximation: log(Y) = log(Y0) + r S + u (schooling S, return r).
- Theoretical decomposition implications:
  - Increase in education inequality, Var(S), unambiguously increases income inequality.
  - Impact of higher average schooling, S̅, depends on Cov(r, S):
    - If Cov(r, S) = 0 or Cov(r, S) > 0 → average schooling increase unambiguously increases income inequality.
    - If Cov(r, S) < 0 → higher average schooling can attenuate or reverse inequality increase.
  - Early-stage expansion dynamics: initial education expansion can increase income inequality before later reducing it as access broadens.

### Data and empirical strategy
- Baseline country-panel specification: I_it = β0 + β1 E_it + β2 σ_it + β3 X_it + α_i + ε_it (I = income Gini, E = average years of education, σ = education Gini).
- Key data:
  - Income Inequality (I): Gini coefficient for disposable income, five-year intervals, based on Bastagli, Coady and Gupta (2012), 1980–2010.
  - Average Education (E): average years of schooling for population aged 25+, Barro and Lee (2013), five-year intervals.
  - Education Inequality (σ): Gini coefficient of years of education from Barro and Lee (2013).
- Econometric concerns and preferred estimators:
  - Persistence of income inequality → include lagged inequality; use dynamic panel methods.
  - Endogeneity of education outcomes → use instrumental and GMM estimators.
  - Preferred estimator: Sys-GMM (Blundell and Bond, 1998); complemented by Diff-GMM, Ins-GMM (instrumenting with parents’ schooling), and long-difference IV (Hahn, Hausman and Kuersteiner, 2006).
  - Tests reported: AR(1), AR(2), Hansen and Sargan over-identification tests; Hausman test for weak exogeneity discussed.

### Descriptive patterns
- Regional income inequality (1980–2010):
  - Advanced (ADV): average income inequality lowest but increased steadily over four decades.
  - Emerging Europe (EE): step increase between 1990 and 1995; stable thereafter.
  - Asia and Pacific (AP) and Middle East and North Africa (MENA): higher than ADV and EE and increased steadily.
  - Sub-Saharan Africa (SSA) and Latin America and the Caribbean (LAC): relatively large levels; SSA slowly declining; LAC sharply declined since peak in 2000 but remains most unequal.
- Education inequality vs years of schooling:
  - Inverted-U relationship overall: education inequality rises at low schooling levels then declines.
  - MENA and SSA show sharp early increases in education inequality as average schooling expands from low levels; at around 4 years average schooling, education inequality higher in MENA and SSA than in AP or LAC.

### Empirical results — baseline OLS and fixed effects (selected)
- OLS and fixed-effects comparisons (selected coefficients from Table 1):
  - OLS (column 1): Education -2.909 (p-value 0.77); Schooling -1.716*** (p-value 0.00); Adjusted R2 = 0.22; N = 873.
  - OLS Time dummies (column 2): Education -18.094* (p-value 0.09); Schooling -3.101*** (p-value 0.00); Adjusted R2 = 0.29.
  - OLS Time and Country dummies (column 3): Education 0.000 (p-value 1.00); Schooling -0.395 (p-value 0.58); Adjusted R2 = 0.85.
  - SURE (column 5): Schooling -0.980* (p-value 0.09); Adjusted R2 = 0.82; N = 418.
  - Fixed effects (FE, column 6): Schooling -0.795* (p-value 0.09); Education 3.300 (p-value 0.68); Adjusted R2 = 0.17; N = 324.
- Interpretation:
  - Simple OLS and time-dummy specifications find negative associations between schooling and income inequality; controlling for country fixed effects weakens significance.
  - Education inequality coefficients often insignificantly estimated in OLS/FE and signs vary.

### Empirical results — dynamic/GMM and heterogeneity (selected)
- Addressing persistence and endogeneity (selected results from Table 2):
  - First Difference (col 1): Gini Education 6.213 (p-value 0.42); Schooling -0.390 (p-value 0.36); N = 322.
  - Diff-GMM (col 2): Gini Education 36.816* (p-value 0.05); Schooling 1.087 (p-value 0.32); N = 309.
  - Sys-GMM (col 3, preferred): Gini Education 34.445** (p-value 0.02); Schooling 1.310 (p-value 0.13); N = 402; L.gini_net 0.593*** (p-value 0.00).
  - Ins-GMM (col 4): Gini Education 30.466** (p-value 0.03); Schooling 1.453** (p-value 0.05); N = 402.
  - Long Difference (col 5): Gini Education 12.101 (p-value 0.18); Schooling 0.194 (p-value 0.73); N = 164.
- Diagnostics:
  - AR(1) and AR(2) p-values reported; Hansen J and Sargan test p-values reported and not rejecting instruments in reported GMM estimations.
- Heterogeneity by development:
  - Allowing coefficients to differ: positive coefficient on education inequality is halved in advanced countries.
    - Sys-GMM column 6: Gini Education Advanced -15.740** (p-value 0.02); Schooling Advanced 0.123 (p-value 0.52).
  - Interpretation: education-inequality impact on income inequality stronger in developing countries; reducing education inequality more important for developing countries.

### Cohort (age) effects (selected)
- Rationale: returns to education “fan out” with age; education inequality among older workers more strongly translates into income inequality.
- Selected coefficient estimates:
  - Sys-GMM (column 5): Schooling 1.986*** (p-value 0.00); Schooling Young -2.830** (p-value 0.01); Gini Education Old 63.903* (p-value 0.08).
  - Ins-GMM (column 6): Schooling 1.958*** (p-value 0.01); Schooling Young -2.250** (p-value 0.04); Gini Education Old 69.535* (p-value 0.09).
- Interpretation:
  - Higher education inequality among older cohorts (holding overall education inequality constant) is associated with substantially larger income inequality.
  - Higher education levels for younger cohorts are associated with lower income inequality (negative and significant Schooling Young in GMM/Ins-GMM).

### Policy simulations — historical (1990–2005)
- Method: simulations use preferred Sys-GMM estimates (Table 2, Column 6) to translate observed changes in education inequality and average schooling into changes in income Gini over 1990–2005.
- Key quantitative results (1990–2005 scenario):
  - Advanced Economies: Inequality in 1990 = 28.1 | Inequality in 2005 = 30.4 | Change in Income inequality = 2.3 | From education inequality = -0.9 | From years of schooling = 1.5 | Total change from education = 0.532 | Number of countries = 3
  - Emerging Europe: 25.3 | 33.0 | 7.7 | -2.9 | 1.5 | -1.4 | 11
  - Latin America & the Caribbean: 46.4 | 48.0 | 1.5 | -2.8 | 1.9 | -0.9 | 21
  - Asia & Pacific: 33.5 | 38.6 | 5.1 | -3.8 | 1.9 | -1.8 | 15
  - Middle East & North Africa: 38.7 | 37.2 | -1.6 | -4.8 | 2.2 | -2.7 | 6
  - Sub-Saharan Africa: 49.4 | 43.6 | -5.8 | -3.6 | 1.7 | -1.9 | 17
- Summary of historical simulation results:
  - Decreases in education inequality reduced income inequality by:
    - 1.5 points in ADV and EE (text earlier summarized 1.5 points in ADV and EE),
    - 4.8 points in MENA (table shows -4.8 contribution).
  - Increases in average education increased income inequality by 1.5 to 2.2 points across regions.
  - Net effect:
    - Slight increase in income inequality in ADV.
    - Net decrease in income inequality in all other regions.

### Projections (2005–2025)
- Projection method:
  - Average education projected forward using known education levels for current working cohorts; future working cohorts assumed to have same education outcomes as youngest current working cohort (aged 20–25 in 2005).
  - Changes used to project impacts on education inequality and then on income inequality via Sys-GMM coefficients.
- Policy simulations and projections (2005–2025 scenario) — regional results (Table 2, Column 6 based simulations):
  - Advanced Economies: Inequality in 2005 = 30.4 | From education inequality = -0.7 | From years of schooling = 1.2 | Total = 0.1 | Number of countries = 35
  - Emerging Europe: 33.0 | -1.4 | 1.1 | -0.1 | 12
  - Latin America & the Caribbean: 48.0 | -2.9 | 1.4 | -1.3 | 25
  - Asia & Pacific: 38.6 | -3.0 | 1.3 | -1.5 | 23
  - Middle East & North Africa: 37.2 | -4.1 | 1.6 | -2.3 | 18
  - Sub-Saharan Africa: 43.6 | -3.2 | 1.1 | -2.0 | 29
- High-level projection result:
  - Changes in education outcomes projected to continue having a dampening impact on income inequality in most regions over 2005–2025, except ADV where projected changes will increase income inequality (table shows total = 0.1 in ADV).

### Main analytical findings and diagnostics
- Dynamic panel estimation that controls for persistence, endogeneity and heterogeneity:
  - The positive relationship between education inequality and income inequality becomes substantially larger, statistically significant and stable across estimators.
  - Education expansion reduces income inequality through decreasing the inequality of education.
  - The relationship between income inequality and schooling levels is positive but small and not always statistically significant.
- Statistical diagnostics:
  - Tests indicate the dynamic estimators are consistent and the identifying instruments are valid.
- Net impacts observed:
  - Over the last fifteen years, the net impact of education expansion has been to reduce income inequality, especially in emerging and developing economies.
  - In advanced economies, education expansion is associated with a net increase in income inequality, reflecting smaller impacts from decreasing education inequality being offset by inequality-increasing impacts of rising education levels.

### Policy implications
- Reducing the inequality of education quality (e.g., improving cognitive skills) is important to enhance the income-inequality-reducing impact of education expansion.
- Other policies beyond education expansion will be needed to address rising income inequality as countries develop.
- Emphasis: policies should focus on more equal access to education and raising education quality for disadvantaged groups to maximize distributional benefits.

### Methodological notes and appendix statistics
- Education inequality measure construction:
  - Barro and Lee attainment categories mapped to average years: no education = 1; incomplete primary = 4; complete primary = 7; incomplete secondary = 10; complete secondary = 13; incomplete tertiary = 16; complete tertiary = 19.
  - Education Gini estimated using Lorenz-curve approach with population weights for attainment segments.
- Explanatory variables (sources noted in Appendix 1): GDP per capita PPP (log and square) — World Bank WDI; Social spending (%GDP) — IMF, Government Finance Statistics Yearbook; Trade openness ((Exports+Imports)/GDP); Capital account openness (Cap100) — IMF Annual Report on Exchange Restrictions; Credit to the private sector (%GDP) — IMF Government Finance Statistics Yearbook; Dependency ratio measures (elderly >65; children <16) — World Bank WDI; Inflation (average annual CPI) — IMF Government Finance Statistics Yearbook; Urban population (% living in urban areas) — World Bank WDI.
- Appendix Table 1 summary statistics (Mean | SD | Obs):
  - GDP per capita PPP (constant 2000 US$): 5617.89 | 8872.32 | 1588
  - Social spending (%GDP): 6.30 | 6.10 | 813
  - Trade openness ((Exports+Imports)/GDP): 0.22 | 0.16 | 775
  - Capital account openness (Cap100): 58.53 | 29.38 | 1134
  - Credit to the private sector (%GDP): 37.75 | 38.02 | 1400
  - Elderly population (% total population): 6.05 | 4.06 | 1985
  - Children (% total population): 35.47 | 10.05 | 1985
  - Inflation (% average): 37.35 | 412.93 | 1267
  - Urban (% population): 48.57 | 25.40 | 2150

*Source: wp17126 — "1. Summary Statistics" and associated tables and simulations.*

### References .............................................................................................................

### wp17126 - References

### Tables
- 1. Education and Income Inequality: OLS, SURE and Fixed Effects Estimates ......................................... 10
- 2. Education and Income Inequality: Difference, GMM, Long Distance Estimates ................................ 12
- 3. Education and Income Inequality: Cohort Effects ......................................................................................... 13
- 4. Implications of Changing Education Outcomes for Income Inequality, 1990–2005 ....................... 14
- 5. Implications of Changing Education Outcomes for Income Inequality, 2005–2025 ....................... 15

### Figures
- 1. Disposable Income Inequality by Region, 1980–2010 ................................................................................... 8
- 2. Education Levels and Education Inequality ........................................................................................................ 9

### Appendixes
- 1. Explanatory Variables and their Sources .......................................................................................................... 17
- 2. Constructing a Measure of Education Inequality .......................................................................................... 19
- Appendix Table

*Source: wp17126 - References (wp17126 - References .............................................................................................................).*

### 1. Summary Statistics ..................................................................................................

### 1. Summary Statistics

### Introduction
- Focus: relationship between education expansion and income inequality; education expansion seen as instrument to reduce income inequality and inequality of opportunity and to support growth.
- Extensions of paper:
  - Addresses endogeneity of education–inequality relationship and persistence of income inequality.
  - Uses updated, comparable income inequality database.
  - Allows heterogeneity across advanced vs developing economies and across working-age cohorts.

### Economic theory
- Human capital model approximation for individual earnings: log(Y) = log(Y0) + r S + u (schooling S, return r).
- Decomposition implies:
  - Increase in education inequality, Var(S), unambiguously increases income inequality.
  - Impact of higher average schooling, S̅, depends on Cov(r, S):
    - If Cov(r, S) = 0 or Cov(r, S) > 0 → average schooling increase unambiguously increases income inequality.
    - If Cov(r, S) < 0 → higher average schooling can attenuate or reverse inequality increase.
- Early-stage expansion dynamics: initial education expansion can increase income inequality (few gain higher education) before later reducing inequality as access broadens.

### Empirical estimation strategy
- Baseline country-panel specification: I_it = β0 + β1 E_it + β2 σ_it + β3 X_it + α_i + ε_it (I = income Gini, E = average years of education, σ = education Gini).
- Key data:
  - Income Inequality (I): Gini coefficient for disposable income, five-year intervals, based on Bastagli, Coady and Gupta (2012), 1980–2010.
  - Average Education (E): average years of schooling for population aged 25+, Barro and Lee (2013), five-year intervals.
  - Education Inequality (σ): Gini coefficient of years of education from Barro and Lee (2013).
- Econometric concerns and methods:
  - Persistence of income inequality → include lagged inequality; use dynamic panel methods.
  - Endogeneity of education outcomes → use instrumental and GMM estimators.
  - Preferred estimator: Sys-GMM (Blundell and Bond, 1998); complemented by Diff-GMM, Ins-GMM (instrumenting with parents’ schooling), and long-difference IV (Hahn, Hausman and Kuersteiner, 2006).
  - Tests reported: AR(1), AR(2), Hansen and Sargan over-identification tests; Hausman test for weak exogeneity discussed.

### Data description
- Gini coefficients used at five-year intervals from 1980 to 2010.
- Regional patterns (Figure 1 summary):
  - Average income inequality lowest in Advanced (ADV) but increased steadily over four decades.
  - Emerging Europe (EE): step increase between 1990 and 1995; stable thereafter.
  - Asia and Pacific (AP) and Middle East and North Africa (MENA): higher than ADV and EE and increased steadily.
  - Sub-Saharan Africa (SSA) and Latin America and the Caribbean (LAC): relatively large levels; SSA slowly declining; LAC sharply declined since peak in 2000 but remains most unequal.
- Education inequality vs years of schooling (Figure 2 summary):
  - Inverted-U relationship overall: education inequality rises at low schooling levels then declines.
  - MENA and SSA show sharp early increases in education inequality as average schooling expands from low levels; at around 4 years average schooling, education inequality higher in MENA and SSA than in AP or LAC.

### Empirical results — baseline and fixed-effects
- OLS and fixed-effects comparisons (selected coefficients from Table 1):
  - OLS (column 1): Education-2.909 (p-value 0.77); Schooling -1.716*** (p-value 0.00); Adjusted R2 = 0.22; N = 873.
  - OLS Time dummies (column 2): Education -18.094* (p-value 0.09); Schooling -3.101*** (p-value 0.00); Adjusted R2 = 0.29.
  - OLS Time and Country dummies (column 3): Education 0.000 (p-value 1.00); Schooling -0.395 (p-value 0.58); Adjusted R2 = 0.85.
  - SURE (column 5): Schooling -0.980* (p-value 0.09); Adjusted R2 = 0.82; N = 418.
  - Fixed effects (FE, column 6): Schooling -0.795* (p-value 0.09); Education 3.300 (p-value 0.68); Adjusted R2 = 0.17; N = 324.
- Interpretation:
  - Simple OLS and time-dummy specifications find negative associations between schooling and income inequality; controlling for country fixed effects weakens significance.
  - Education inequality coefficients are often insignificantly estimated in OLS/FE but signs vary.

### Empirical results — dynamic/GMM and cohort heterogeneity
- Addressing persistence and endogeneity (selected results from Table 2, columns shown with p-values):
  - First Difference (col 1): Gini Education 6.213 (p-value 0.42); Schooling -0.390 (p-value 0.36); N = 322.
  - Diff-GMM (col 2): Gini Education 36.816* (p-value 0.05); Schooling 1.087 (p-value 0.32); N = 309.
  - Sys-GMM (col 3, preferred): Gini Education 34.445** (p-value 0.02); Schooling 1.310 (p-value 0.13); N = 402; L.gini_net 0.593*** (p-value 0.00).
  - Ins-GMM (col 4): Gini Education 30.466** (p-value 0.03); Schooling 1.453** (p-value 0.05); N = 402.
  - Long Difference (col 5): Gini Education 12.101 (p-value 0.18); Schooling 0.194 (p-value 0.73); N = 164.
- Diagnostics (Sys-GMM and related): AR(1) and AR(2) p-values reported; Hansen J and Sargan test p-values reported and not rejecting instruments in GMM estimations reported.
- Heterogeneity by development:
  - Allowing coefficients to differ: positive coefficient on education inequality is halved in advanced countries (Sys-GMM column 6: Gini Education Advanced -15.740** (p-value 0.02); Schooling Advanced 0.123 (p-value 0.52)).
  - Interpretation: education-inequality impact on income inequality stronger in developing countries; reducing education inequality more important for developing countries.

### Cohort (age) effects (selected results from Table 3)
- Rationale: returns to education “fan out” with age; education inequality among older workers more strongly translates into income inequality.
- Selected coefficient estimates:
  - Sys-GMM (column 5): Schooling 1.986*** (p-value 0.00); Schooling Young -2.830** (p-value 0.01); Gini Education Old 63.903* (p-value 0.08).
  - Ins-GMM (column 6): Schooling 1.958*** (p-value 0.01); Schooling Young -2.250** (p-value 0.04); Gini Education Old 69.535* (p-value 0.09).
  - OLS/FE and other estimators show less precise or insignificant cohort coefficients but signs are consistent with theory.
- Interpretation:
  - Higher education inequality among older cohorts (holding overall education inequality constant) is associated with substantially larger income inequality.
  - Higher education levels for younger cohorts are associated with lower income inequality (negative and significant Schooling Young in GMM/Ins-GMM).

### Policy simulations — historical (1990–2005)
- Using preferred Sys-GMM estimates to simulate impacts of education changes on income inequality over 1990–2005:
  - Regional observed changes: Gini increased by over 5 points in AP and EE on average in sample.
  - Effects of changes in education:
    - Decreases in education inequality reduced income inequality by:
      - 1.5 points in ADV and EE,
      - 4.8 points in MENA.
    - Increases in average education increased income inequality by 1.5 to 2.2 points across regions.
  - Net effect:
    - Slight increase in income inequality in ADV.
    - Net decrease in income inequality in all other regions.
- Policy implication from simulations: education investments that reduce education inequality can play an important role in mitigating increases in income inequality; emphasis on more equal access is key.

### Policy simulations — projections (2005–2025) (summary from text)
- Projection method:
  - Average education projected forward using known education levels for current working cohorts; future working cohorts assumed to have same education outcomes as youngest current working cohort (aged 20–25 in 2005).
  - Changes used to project impacts on education inequality and then on income inequality via Sys-GMM coefficients.
- High-level result:
  - Changes in education outcomes projected to continue having a dampening impact on income inequality in most regions over 2005–2025, except ADV where projected changes will increase income inequality (specific projection magnitudes not provided in excerpt).

*Source: IMF working paper content unit "1. Summary Statistics" (wp17126).*

### 0.1 points. Similar to the period 1990–2005, continued decreases in education inequality will have

### wp17126 - 0.1 points. Similar to the period 1990–2005, continued decreases in education inequality will have

### Key quantitative results (1990–2005 scenario)
- Table metrics: Regions; Inequality in 1990; Inequality in 2005; Change in Income inequality; From education inequality (absolute change); From years of schooling (absolute change); Total change in Income inequality from education; Number of countries
- Advanced Economies: 28.1 | 30.4 | 2.3 | -0.9 | 1.5 | 0.532 | 3
- Emerging Europe: 25.3 | 33.0 | 7.7 | -2.9 | 1.5 | -1.4 | 11
- Latin America & the Caribbean: 46.4 | 48.0 | 1.5 | -2.8 | 1.9 | -0.9 | 21
- Asia & Pacific: 33.5 | 38.6 | 5.1 | -3.8 | 1.9 | -1.8 | 15
- Middle East & North Africa: 38.7 | 37.2 | -1.6 | -4.8 | 2.2 | -2.7 | 6
- Sub-Saharan Africa: 49.4 | 43.6 | -5.8 | -3.6 | 1.7 | -1.9 | 17
- Source note: Simulations based on Table 2, Column 6.

### Policy simulations and projections (2005–2025 scenario)
- Table metrics: Regions; Inequality in 2005; From education inequality (% improvement); From years of schooling (% improvement); Total; Number of countries
- Advanced Economies: 30.4 | -0.7 | 1.2 | 0.1 | 35
- Emerging Europe: 33.0 | -1.4 | 1.1 | -0.1 | 12
- Latin America & the Caribbean: 48.0 | -2.9 | 1.4 | -1.3 | 25
- Asia & Pacific: 38.6 | -3.0 | 1.3 | -1.5 | 23
- Middle East & North Africa: 37.2 | -4.1 | 1.6 | -2.3 | 18
- Sub-Saharan Africa: 43.6 | -3.2 | 1.1 | -2.0 | 29
- Source note: Simulations based on Table 2, Column 6.

### Main analytical findings
- Dynamic panel estimation that controls for persistence, endogeneity and heterogeneity:
  - The positive relationship between education inequality and income inequality becomes substantially larger, statistically significant and stable across estimators.
  - Education expansion reduces income inequality through decreasing the inequality of education.
  - The relationship between income inequality and schooling levels is positive but small and not always statistically significant.
- Statistical diagnostics:
  - Tests indicate the dynamic estimators are consistent and the identifying instruments are valid.
- Net impacts observed:
  - Over the last fifteen years, the net impact of education expansion has been to reduce income inequality, especially in emerging and developing economies.
  - In advanced economies, education expansion is associated with a net increase in income inequality, reflecting smaller impacts from decreasing education inequality being offset by inequality-increasing impacts of rising education levels.

### Projections and implications
- Projections over two decades (2005–2025) indicate:
  - Education expansion will continue to have an inequality-reducing impact in emerging and developing economies.
  - The inequality-reducing impact of falling education inequality is partly offset by the inequality-increasing impact of rising education levels, but the net remains inequality reducing for these economies.
  - The inequality-increasing impact in advanced economies suggests the inequality-reducing role of education expansion in emerging and developing economies will diminish as they converge to advanced-economy levels.
- Policy implication emphasized:
  - Reducing the inequality of education quality (e.g., improving cognitive skills) is important to enhance the income-inequality-reducing impact of education expansion.
  - Other policies beyond education expansion will be needed to address rising income inequality as countries develop.

### Methodological notes (data and variables)
- Education inequality measure:
  - Constructed from Barro and Lee attainment categories by assigning average years: no education = 1; incomplete primary = 4; complete primary = 7; incomplete secondary = 10; complete secondary = 13; incomplete tertiary = 16; complete tertiary = 19.
  - Education Gini estimated using Lorenz-curve approach with population weights for attainment segments.
- Explanatory variables used (sources indicated in Appendix 1):
  - GDP per capita PPP (log and square) — World Bank WDI
  - Social spending (%GDP) — IMF, Government Finance Statistics Yearbook
  - Trade openness ((Exports+Imports)/GDP)
  - Capital account openness (Cap100) — IMF Annual Report on Exchange Restrictions
  - Credit to the private sector (%GDP) — IMF Government Finance Statistics Yearbook
  - Dependency ratio measures (elderly >65; children <16) — World Bank WDI
  - Inflation (average annual CPI) — IMF Government Finance Statistics Yearbook
  - Urban population (% living in urban areas) — World Bank WDI
- Appendix Table 1 summary statistics (Mean | SD | Obs):
  - GDP per capita PPP (constant 2000 US$): 5617.89 | 8872.32 | 1588
  - Social spending (%GDP): 6.30 | 6.10 | 813
  - Trade openness ((Exports+Imports)/GDP): 0.22 | 0.16 | 775
  - Capital account openness (Cap100): 58.53 | 29.38 | 1134
  - Credit to the private sector (%GDP): 37.75 | 38.02 | 1400
  - Elderly population (% total population): 6.05 | 4.06 | 1985
  - Children (% total population): 35.47 | 10.05 | 1985
  - Inflation (% average): 37.35 | 412.93 | 1267
  - Urban (% population): 48.57 | 25.40 | 2150

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17126.pdf_