## 5.1    Inequality Overhang

## Source details

**Canonical URL:** [5.1    Inequality Overhang](https://www.imf.org/-/media/files/publications/wp/2017/wp1776.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp1776.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp1776.pdf.json)

---

### Key findings
- Pervasive evidence of nonlinearities in the relationship between income inequality and economic development.
- The slope of the relationship between inequality and economic development turns from positive to negative at a net Gini of about 27 percent, indicating the inequality overhang occurs at low levels of income inequality.
- When financial inclusion is widespread and income is highly concentrated, the negative effects from income inequality become larger.
- A context of high female labor participation can effectively contribute to reduce the negative impact of income inequality and strengthen the positive impact.

### Contribution and empirical strategy
- Paper objective:
  - Tests for nonlinearities in the inequality–development relationship using a recently developed empirical approach.
  - Identifies an inequality overhang level (the level at which the impact of income inequality on economic development becomes negative).
  - Analyzes how the shape of the identified relationship changes for different levels of financial inclusion and female labor participation.
- Main regression starting point (pooled OLS form):
  - y_{i,t} = α_i + Σ_{j=0}^P β_j X^j_{i,t} + u_{i,t}, where y_{i,t} is real per capita GDP and X_{i,t} is the net Gini coefficient.
- Two-step approach (Al Masri and Pedroni (2016) method):
  - Step 1: Static linear time series regressions for each country:
    - y_{i,t} = α_i + β_i X_{i,t} + u_{i,t}.
    - Superconsistency of slope parameters provides robustness to endogeneity.
  - Step 2: Cross-section regressions for each time period on the estimated country-specific slopes:
    - ˆβ_i = Σ_{j=0}^P C_{j,s} X_{i(s)}^j + E_s R_{i(s)} + v_i.
    - Polynomial order selected via a general-to-specific (GTS) procedure starting at order three; most years deliver a polynomial of order two (quadratic); rare years select order three (cubic).
- Constructed fitted polynomial of order P+1 for each time period to trace evolution; average relationship computed via mean group coefficients.
- First derivative used to assess marginal impact of net Gini on per capita GDP and to identify inequality overhang.
- Interaction effects included by augmenting cross-section regressions with terms Z_k,i (e.g., financial access, female labor participation).

### Data, sample, and inclusion criteria
- Net Gini series from SWIID 5.1; real income and population from Penn World Tables (PWT) 9.0.
- Preference for net Gini (post-redistribution) over market Gini for analysis of post-redistribution income and development.
- Data exclusion criteria to obtain non-stationary continuous time series for net Gini:
  - Criterion 1: Exclude non-continuous time series; retain the longest run for each country.
  - Criterion 2: Exclude countries whose longest run is shorter than 20 observations (T threshold set to 20).
  - Criterion 3: Drop countries for which net Gini is stationary.
- Final remaining dataset:
  - 77 countries
  - Minimum of 20 years per country
  - Total of 1,597 data points
- Note on criterion 3: loss of 9 countries (Denmark, Estonia, Georgia, Netherlands, Romenia, Senegal, Slovenia, Spain, and Uruguay) or 11 percent of the dataset obtained after applying the second criterion.
- Robustness checks:
  - Logistic transformation y_{i,t} = ln[n_{i,t} / (1 − n_{i,t})] applied to net Gini as robustness; does not affect results.
  - Human capital proxies included only in robustness tests due to missing observations.

### Unit root and cointegration (panel results)
- Panel unit root tests suggest both net Gini and ln per capita real GDP are integrated of order one.
- Selected panel unit root test statistics (levels and first differences):
  - Levin, Lin, and Chu (2002):
    - Net GINI: Intercept = -1.344; Intercept and trend = -1.110; First differences: Intercept = -12.061**; Intercept and trend = -6.933**
    - Ln per capita real GDP: Intercept = -1.419; Intercept and trend = -2.58; First differences: Intercept = -23.297**; Intercept and trend = -19.438**
  - Im, Pesaran, and Shin (2003):
    - Net GINI: Intercept = -0.696; Intercept and trend = -2.048*; First differences: Intercept = -18.162**; Intercept and trend = -15.552**
    - Ln per capita real GDP: Intercept = 8.137; Intercept and trend = -1.070; First differences: Intercept = -22.258**; Intercept and trend = -19.927**
  - Maddala and Wu (1999):
    - Net GINI: Intercept = -0.427; Intercept and trend = -1.489; First differences: Intercept = -16.224**; Intercept and trend = -14.141**
    - Ln per capita real GDP: Intercept = 7.685; Intercept and trend = -0.524; First differences: Intercept = -20.329**; Intercept and trend = -17.768**
- Panel cointegration tests (Kao (1999) and Pedroni (1999, 2004)) indicate net Gini and ln per capita real GDP are cointegrated.

### Evolution of the unconditional relationship (1975–2010)
- GTS procedure returns a quadratic relationship for most years and a cubic relationship in a very few years.
- Coverage: starting in 1990, at least 95 percent of the sample of 77 countries is represented.
- Phases identified:
  - Before the 2000s: relationship is positive for low levels of income inequality (up to a net Gini of less than 30 percent), and negative for higher levels (net Gini higher than about 30 percent).
  - For very high levels of income inequality (net Gini higher than about 50 percent), the relationship flattens and becomes positive in 1990.
  - From 2000 onward: the relationship is strictly negative for all levels of inequality and features an increasingly negative slope as net Gini rises.
- Notes: early years have far fewer observations.

### Derivative results and timing
- Derivative computed as in equation (8); shape evolves over time:
  - Until 1990: derivative described by a quadratic function (polynomial relationship often cubic).
  - After 1990: derivative becomes linear (polynomial relationship typically quadratic).
- Findings on marginal impact:
  - Impact of income inequality on growth is positive for low levels of net Gini until 1995.
  - Starting in 1995, the derivative becomes a negatively sloped linear function, indicating the impact of income inequality on growth is negative for all countries in the sample and increasingly negative as inequality rises.
  - Exception: only in 1990 does the polynomial relationship show positive growth effects from changes in inequality at very high levels of income inequality.

### Inequality overhang threshold and average conditional relationship (1990–2010)
- Average relationship (1990–2010) allowing for control variables is best described by a cubic function; corresponding average derivative is quadratic and concave.
- Inequality overhang threshold:
  - Solutions of the quadratic derivative suggest the inequality overhang occurs at a level of net Gini of about 27 percent.
  - Interpretation: for net Gini below about 27 percent, an increase in income inequality has a positive effect on growth; for net Gini above 27 percent, any increase in income inequality has a negative effect on growth, strengthening as inequality rises.
  - The other root of the quadratic equation is negative and does not have an economic meaning.

### Growth-dividend experiments and country-level impacts
- Two experiments to quantify potential growth dividends from reducing income inequality:
  - Experiment 1: every country reduces inequality by the sample average of the change in net Gini, i.e. 0.13 percentage points (pp).
    - Results: growth dividends are relatively small, up to about 0.05 pp; for countries with initial net Gini below the overhang level the effect would be slightly negative but very small.
  - Experiment 2: every country reduces net Gini by one pp.
    - Results: larger effects on growth, ranging from about zero percent to about 0.43 percent.
- Distribution of benefits:
  - Largest benefits would accrue to Western Hemisphere countries plus Zimbabwe (ZWE), China (CHN), and India (IND).
- Country example and feasibility note:
  - Argentina displays an average net Gini of 44 percent over 1990–2010 (coefficient never below 40 percent); a reduction from 44 percent to 27 percent would be dramatic and unrealistic for many countries.
- Cross-country sensitivity: countries with higher initial levels of income inequality are more sensitive to changes in income inequality.

### Key statistics and reference values (preserved exactly)
- Sample span illustrated: 1975–2010.
- Sample coverage benchmark: starting in 1990, at least 95 percent of the sample of 77 countries is represented.
- Net Gini reference points: less than 30 percent; about 30 percent; about 50 percent.
- Inequality overhang (threshold): about 27 percent net Gini.
- Average net Gini (used in examples): about 37 percent.
- Average sample reduction in net Gini used in Experiment 1: 0.13 percentage points (pp).
- Experiment 2 reduction in net Gini: one pp.
- Growth dividend magnitudes:
  - Up to about 0.05 pp (Experiment 1).
  - Ranging from about zero percent to about 0.43 percent (Experiment 2).
- Country example: Argentina average net Gini of 44 percent over 1990–2010 (coefficient never below 40 percent).

### Regression evidence (mean group estimations summary)
- Selected coefficients from Table B.1 (coefficients with t-statistics in parentheses; ***, **, and * indicate significance at 1, 5, and 10 percent):
  - Net Gini:
    - (1) 0.0024* (1.646)
    - (7) 0.0039*** (3.441)
    - (8) 0.0182*** (6.070)
  - Net Gini squared:
    - (1) -0.0001** (-2.504)
    - (7) -0.0001*** (-4.391)
    - (8) -0.0003*** (-7.341)
  - Initial (log) real per capita GDP:
    - (1) 0.0087** (2.452)
    - (6) 0.0246*** (20.733)
    - (7) 0.0108*** (14.510)
  - Trade openness:
    - (1) 0.0005*** (9.235)
    - (4) 0.0004*** (7.637)
  - Investment in percent of GDP:
    - (1) 0.0019*** (3.593)
    - (6) 0.0012*** (12.115)
  - Population growth:
    - (1) 0.0073*** (2.826)
    - (7) 0.0093*** (18.297)
  - Financial access (column 6):
    - (6) -0.1292*** (-2.791)
    - Financial access * net Gini (6) 0.0132*** (5.246)
    - Financial access * net Gini squared (6) -0.0003*** (-9.835)
  - Female labor participation (column 7):
    - (7) 0.0081*** (7.199)
    - Female labor participation * net Gini (7) -0.0003*** (-5.027)
    - Female labor participation * net Gini squared (7) 0.0000*** (4.854)
  - Observations by column:
    - (1) 1449; (2) 1449; (3) 1412; (4) 1397; (5) 1397; (6) 1395; (7) 1352; (8) 1352

*Source: Authors’ calculations (wp1776 - 5.1 Inequality Overhang).*

### 5.1    Inequality Overhang   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .

### 5.1    Inequality Overhang

### Section identification
- Section number: 5.1
- Section title: Inequality Overhang
- Location in source: page 9

*Source: wp1776 - 5.1 Inequality Overhang (IMF Working Paper PDF)*

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

### wp1776 - References

### Key findings
- Pervasive evidence of nonlinearities in the relationship between income inequality and economic development.
- The slope of the relationship between inequality and economic development turns from positive to negative at a net Gini of about 27 percent, indicating the inequality overhang occurs at low levels of income inequality.
- When financial inclusion is widespread and income is highly concentrated, the negative effects from income inequality become larger.
- A context of high female labor participation can effectively contribute to reduce the negative impact of income inequality and strengthen the positive impact.

### Contribution summary
- The paper tests for nonlinearities in the inequality–development relationship using a recently developed empirical approach.
- It identifies an inequality overhang level (the level at which the impact of income inequality on economic development becomes negative).
- It analyzes how the shape of the identified relationship changes for different levels of financial inclusion and female labor participation.

### Empirical strategy and methodology
- Main regression starting point (pooled OLS form):
  - y_{i,t} = α_i + Σ_{j=0}^P β_j X^j_{i,t} + u_{i,t}
  - where y_{i,t} is real per capita GDP, X_{i,t} is the net Gini coefficient.
- Two-step approach (Al Masri and Pedroni (2016) method):
  - Step 1: Static linear time series regressions for each country:
    - y_{i,t} = α_i + β_i X_{i,t} + u_{i,t}
    - Superconsistency of slope parameters provides robustness to endogeneity (reverse causality, omitted dynamics, omitted stationary variables).
  - Step 2: Cross-section regressions for each time period on the estimated country-specific slopes:
    - ˆβ_i = Σ_{j=0}^P C_{j,s} X_{i(s)}^j + E_s R_{i(s)} + v_i
    - Polynomial order selected via a general-to-specific (GTS) procedure starting at order three and dropping highest insignificant terms.
    - For most years the cross-section estimation delivers a polynomial of order two (quadratic); in rare cases a polynomial of order three (cubic) is selected.
- Constructed fitted polynomial of order P+1 for each time period to trace evolution of relationship:
  - y_{i,t} = ̂α_i + Σ_{j=0}^P ̂C_{j,s} X_{i(s)}^j X_{i} + ̂E_s R_{i(s)} X_i
- Average relationship computed via mean group coefficients:
  - ̂c_j = T^{-1} Σ_{s=1}^T ̂c_{j,s}; associated t-statistic t_j = T^{-1/2} Σ_{s=1}^T t_{j,s}
  - Average relationship (quadratic case) yields a cubic equation in x_{i,t}.
- First derivative used to assess the marginal impact of net Gini on per capita GDP and to identify inequality overhang:
  - δy_i / δX_i = Σ_{j=0}^P j ̂C_{j−1} X^{j−1}_i + ̂E R_i
  - For cubic relationship: δy_i / δx_i = ̂c_o + ̂e_s r_i(s) + 2̂c_1 x_i + 3̂c_2 x_i^2
- Interaction effects included by augmenting cross-section regressions with terms Z_k,i (e.g., financial access, female labor participation):
  - ̂β_i = Σ_{j=0}^P ̃C_{j,s} X_{i(s)}^j + E_s R_{i(s)} + Σ_{k=1}^K Σ_{j=0}^P D_{k,j} X_{i(s)}^j Z_{k,i} + v_i

### Data, sample, and inclusion criteria
- Net Gini series from SWIID 5.1; real income and population from Penn World Tables (PWT) 9.0.
- Preference for net Gini (post-redistribution) over market Gini for analysis of post-redistribution income and development.
- Data exclusion criteria to obtain non-stationary continuous time series for net Gini:
  - Criterion 1: Exclude non-continuous time series; retain the longest run for each country.
  - Criterion 2: Exclude countries whose longest run is shorter than 20 observations (T threshold set to 20).
  - Criterion 3: Drop countries for which net Gini is stationary.
- Final remaining dataset:
  - 77 countries
  - Minimum of 20 years per country
  - Total of 1,597 data points
- Note on criterion 3: the application implied the loss of 9 countries (Denmark, Estonia, Georgia, Netherlands, Romenia, Senegal, Slovenia, Spain, and Uruguay) or 11 percent of the dataset obtained after applying the second criterion.
- Robustness checks:
  - Logistic transformation y_{i,t} = ln[n_{i,t} / (1 − n_{i,t})] applied to net Gini as robustness; does not affect results.
  - Human capital proxies (years of schooling, education attainment) included only in robustness tests due to missing observations.

### Unit root and cointegration test results (panel)
- Panel unit root tests suggest both net Gini and ln per capita real GDP are integrated of order one.
- Table 1: Panel Unit Root Tests (levels and first differences)
  - Levin, Lin, and Chu (2002)
    - Net GINI: Intercept = -1.344; Intercept and trend = -1.110; First differences: Intercept = -12.061**; Intercept and trend = -6.933**
    - Ln per capita real GDP: Intercept = -1.419; Intercept and trend = -2.58; First differences: Intercept = -23.297**; Intercept and trend = -19.438**
  - Im, Pesaran, and Shin (2003)
    - Net GINI: Intercept = -0.696; Intercept and trend = -2.048*; First differences: Intercept = -18.162**; Intercept and trend = -15.552**
    - Ln per capita real GDP: Intercept = 8.137; Intercept and trend = -1.070; First differences: Intercept = -22.258**; Intercept and trend = -19.927**
  - Maddala and Wu (1999)
    - Net GINI: Intercept = -0.427; Intercept and trend = -1.489; First differences: Intercept = -16.224**; Intercept and trend = -14.141**
    - Ln per capita real GDP: Intercept = 7.685; Intercept and trend = -0.524; First differences: Intercept = -20.329**; Intercept and trend = -17.768**
  - Notes: The null hypothesis of the panel unit root tests is that all panels contain a unit root. Fixed effects always included. The Schwartz Information Criterion is used to select the optimal lag length. ** and * indicate statistical significance at 1 and 5 percent, respectively.
- Panel cointegration tests (Kao (1999) and Pedroni (1999, 2004)) reported in Table 2 indicate net Gini and ln per capita real GDP are cointegrated, consistent with existence of a long-run relationship between the variables.

*Source: Authors’ calculations (wp1776 - References).*

### 5.1  Inequality Overhang

### 5.1  Inequality Overhang

### Evolution of the unconditional relationship (1975–2010)
- Analysis uses a GTS procedure to select the polynomial shape that best describes the data; procedure returns a quadratic relationship for most years and a cubic relationship in a very few years.
- Coverage: starting in 1990, at least 95 percent of the sample of 77 countries is represented.
- Phases identified:
  - Before the 2000s: relationship is positive for low levels of income inequality (up to a net Gini of less than 30 percent), and negative for higher levels (net Gini higher than about 30 percent).
  - For very high levels of income inequality (net Gini higher than about 50 percent), the relationship flattens and becomes positive in 1990.
  - From 2000 onward: the relationship is strictly negative for all levels of inequality and features an increasingly negative slope as net Gini rises.
- Notes on data coverage: early years in the sample have far fewer observations (only a few AE and EM have series going back to those years).

### Derivative of real per capita GDP with respect to net Gini (evolution and shape)
- The derivative is calculated as in equation (8).
- Shape over time:
  - Until 1990: derivative described by a quadratic function (polynomial relationship often cubic).
  - After 1990: derivative becomes linear (polynomial relationship typically quadratic).
- Findings:
  - Impact of income inequality on growth is positive for low levels of net Gini until 1995.
  - Starting in 1995, the derivative becomes a negatively sloped linear function, indicating the impact of income inequality on growth is negative for all countries in the sample and increasingly negative as inequality rises.
  - Exception: only in 1990 does the polynomial relationship show positive growth effects from changes in inequality at very high levels of income inequality.

### Average conditional relationship (1990–2010) and the inequality threshold
- Average relationship estimated over 1990–2010 allowing for control variables; the average relationship is best described by a cubic function.
- Shape and interpretation:
  - The cubic function is steep and negative for most net Gini values, but flattens and becomes mildly positive for low levels of inequality.
  - The corresponding average derivative is a quadratic function with a concave shape.
- Inequality overhang threshold:
  - Calculating the solutions of the quadratic derivative suggests the inequality overhang occurs at a level of net Gini of about 27 percent.
  - Interpretation: for net Gini below about 27 percent, an increase in income inequality has a positive effect on growth; for net Gini above 27 percent, any increase in income inequality has a negative effect on growth, strengthening as inequality rises.
  - The other root of the quadratic equation is negative and does not have an economic meaning.

### Relationships between growth rates, levels, and steady-state changes
- Using data-driven polynomials in levels permits retrieving relationships among growth rates and levels without re-estimating.
- Figure 4 experiments:
  - Conditioning on different steady state changes in net Gini (50 percent and 150 percent of avg. ∆ net Gini) shows the relationship between real per capita GDP growth and net Gini steepens for larger steady state changes.
  - Conditioning on different levels of net Gini (50 percent and 150 percent of avg. net Gini) shows the relationship between growth rate and changes in net Gini varies: positive when halving average net Gini, steeper negative when increasing it.
- Average net Gini reference value: about 37 percent (used as “average” in the discussion of conditional relationships).

### Growth-dividend experiments and country-level impacts
- Two experiments to quantify potential growth dividends from reducing income inequality:
  - Experiment 1: every country reduces inequality by the sample average of the change in net Gini, i.e. 0.13 percentage points (pp).
  - Experiment 2: every country reduces net Gini by one pp.
- Country example and feasibility note:
  - Argentina displays an average net Gini of 44 percent over 1990–2010, with the coefficient never below 40 percent; a reduction from 44 percent to 27 percent would be dramatic and unrealistic for many countries.
- Results (Figure 5):
  - Experiment 1 (avg. reduction = 0.13 pp): growth dividend are relatively small, up to about 0.05 pp; for countries with initial net Gini below the overhang level the effect would be slightly negative (counterproductive) but very small in magnitude.
  - Experiment 2 (one pp reduction): larger effects on growth, ranging from about zero percent to about 0.43 percent.
  - Distribution of benefits: the largest benefits would accrue to Western Hemisphere countries plus Zimbabwe (ZWE), China (CHN), and India (IND).
- Cross-country sensitivity: countries with higher initial levels of income inequality are more sensitive to changes in income inequality.

### Key statistics and parameters (preserved exactly as in source)
- Sample span illustrated: 1975–2010 (plots shown every five years during 1975–2010).
- Sample coverage benchmark: starting in 1990, at least 95 percent of the sample of 77 countries is represented.
- Net Gini reference points: less than 30 percent; about 30 percent; about 50 percent.
- Inequality overhang (threshold): about 27 percent net Gini.
- Average net Gini (used in examples): about 37 percent.
- Average sample reduction in net Gini used in Experiment 1: 0.13 percentage points (pp).
- Experiment 2 reduction in net Gini: one pp.
- Growth dividend magnitudes:
  - Up to about 0.05 pp (Experiment 1).
  - Ranging from about zero percent to about 0.43 percent (Experiment 2).
- Country example: Argentina average net Gini of 44 percent over 1990–2010 (coefficient never below 40 percent).

*Source: Authors’ calculations (from chapter "5.1 Inequality Overhang").*

### 5.2  Financial Inclusion

### 5.2  Financial Inclusion

### Data and empirical approach
- Financial inclusion proxy: the financial institutions access indicator by Svirydzenka (2016).
- The financial access variable is constructed using the number of bank branches and ATMs per 100,000 adults.
- Empirical strategy: include interaction terms between net Gini (and squared net Gini) and the financial access indicator; estimate conditional coefficients and plot the polynomial relationship for steady state values of financial access.

### Main empirical findings
- Average (left panel of Figure 6):
  - The polynomial relationship between income inequality (net Gini) and real per capita GDP is concave.
  - For average levels of financial access, the concave phase resembles the baseline (no-interaction) relationship.
  - Higher levels of financial access:
    - Move the negatively sloped portion of the relationship towards lower levels of net Gini.
    - Steepen the negatively sloped portion.
  - Lower levels of financial access:
    - Flatten the concavity.
    - Push the negatively sloped portion of the relationship towards higher levels of net Gini.
- Derivative and changes in financial access (middle panel of Figure 6):
  - Increases in income inequality in the presence of a more inclusive financial system have larger negative returns on income.
  - The inequality overhang level (where the derivative switches from positive to negative) is lower when financial access is higher and higher when financial access is lower.
- Changes in net Gini and growth (right panel of Figure 6):
  - Negative effects on growth from changes in income inequality are larger when financial access is easier.
- Three-dimensional evidence (Figure 7):
  - For average levels of financial access, the inequality overhang level of net Gini is about 27 percent.
  - The derivative of real per capita GDP with respect to net Gini becomes more negative as income inequality increases.
  - For high levels of financial access:
    - The negative steepness of the derivative surface is more pronounced, especially when income is highly concentrated.
    - This depicts a larger negative impact of income inequality on economic development and a lower inequality overhang level.
  - For low levels of financial access:
    - The derivative can still be negative, and the overhang occurs at higher levels of income inequality.

### Interpretation and suggested mechanism (from the section)
- Counter-intuitive implication: in contexts of high income concentration, widespread financial inclusion can exacerbate the negative effects of income inequality on economic development.
- Proposed explanation based on credit constraints:
  - With widespread access to financial services, the portion of credit-constrained borrowers without financial access is smaller than when access is limited.
  - If financial access relaxes budget constraints but rising income inequality makes those at the lower end relatively poorer and worsens their ability to meet obligations, banks may become less prone to provide credit to those poorer customers who nevertheless have access to services.
- Related evidence: cited study (Blaum (2012)) finds that the effect of income inequality on value added shares becomes larger when financial development improves; Blaum (2012) also finds that for very high levels of financial development, the effect is reversed.

*Source: Authors’ calculations and text in section 5.2 of the provided chapter.*

### Appendix A. Data Sources and Country Groups

### Appendix A. Data Sources and Country Groups

### Data sources and variable scales
- Table A.1 lists all variables used in the paper, along with the source and the scale.
- Variables, sources, and scales:
  - ATMs per 100,000 adults — Financial Access Survey — Units
  - Commercial branches per 100,000 adults — Financial Access Survey — Units
  - Completion rate primary education — World Development Indicators — Percent
  - Completion rate secondary education — World Development Indicators — Percent
  - Female labor participation — International Labour Organization — Percent
  - Financial access — Svirydzenka (2016) — Index
  - Government expenditure — World Economic Outlook — Percent of GDP
  - Education expenditure — IMF’s Fiscal Affairs Department database — Percent of GDP
  - Investment — World Economic Outlook — Percent of GDP
  - Labor productivity — World Economic Outlook — Index
  - Market Gini — SWIID 5.1 — Index
  - Net Gini — SWIID 5.1 — Index
  - Population growth — World Economic Outlook — Percent
  - Real per capita GDP — Penn World Tables 9.0 — PPP
  - Terms of trade — World Economic Outlook — Index
  - Trade openness — World Economic Outlook — Percent
  - Years of schooling — Barro-Lee Database — Units

### Regional groups (according to IMF regional departments)
- Africa: Botswana, Ghana, Malawi, Mali, Nigeria, Rwanda, Sierra Leone, South Africa, Uganda, Zambia, Zimbabwe.
- Asia and Pacific: Australia, Bangladesh, China, India, Indonesia, Japan, Malaysia, New Zealand, Philippines, Singapore, Sri Lanka, Taiwan.
- Europe: Austria, Belarus, Belgium, Bulgaria, Croatia, Cyprus, Czech Rep., Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Latvia, Lithuania, Luxembourg, Norway, Poland, Portugal, Russian Federation, Slovakia, Sweden, Switzerland, Turkey, Ukraine, United Kingdom.
- Middle East and Central Asia: Armenia, Egypt, Jordan, Kazakhstan, Mauritania, Morocco, Pakistan, Tajikistan, Tunisia.
- Western Hemisphere: Argentina, Brazil, Canada, Chile, Colombia, Costa Rica, Dominican Rep., Ecuador, El Salvador, Guatemala, Honduras, Mexico, Panama, Paraguay, Peru, United States.

### Income-level country groups
- Advanced Economies: Australia, Austria, Belgium, Canada, Cyprus, Czech Rep., Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Latvia, Lithuania, Luxembourg, New Zealand, Norway, Portugal, Singapore, Slovakia, Sweden, Switzerland, Taiwan, United Kingdom, United States.
- Emerging Markets: Argentina, Armenia, Belarus, Botswana, Brazil, Bulgaria, Chile, China, Colombia, Costa Rica, Croatia, Dominican Republic, Ecuador, Egypt, El Salvador, Guatemala, Hungary, India, Indonesia, Jordan, Kazakhstan, Malaysia, Mexico, Morocco, Pakistan, Panama, Paraguay, Peru, Philippines, Poland, Russian Federation, South Africa, Sri Lanka, Tunisia, Turkey, Ukraine.
- Low Income Developing Countries: Bangladesh, Ghana, Honduras, Malawi, Mali, Mauritania, Nigeria, Rwanda, Sierra Leone, Tajikistan, Uganda, Zambia, Zimbabwe.

### Notes on Appendix A
- Table A.1 and the country groupings define the dataset used in the analysis and the sample composition by region and income level.

### Appendix B. Regression Results

### Mean Group Estimations — Table B.1 (unconditional and conditional regressions)
- Table presents coefficients with t-statistics in parentheses. ***, **, and * indicate statistical significance at 1, 5, and 10 percent, respectively.
- Coefficients and t-statistics by column (columns labeled in source as (1) through (8)):

  - Net Gini
    - (1) 0.0024* (1.646)
    - (2) 0.0048 (1.218)
    - (3) 0.0050 (1.142)
    - (4) 0.0021 (0.379)
    - (5) 0.0035* (1.678)
    - (6) 0.0031* (1.701)
    - (7) 0.0039*** (3.441)
    - (8) 0.0182*** (6.070)

  - Net Gini squared
    - (1) -0.0001** (-2.504)
    - (2) -0.0001*** (-2.917)
    - (3) -0.0001*** (-2.734)
    - (4) -0.0001* (-1.827)
    - (5) -0.0001** (-2.066)
    - (6) -0.0001* (-1.957)
    - (7) -0.0001*** (-4.391)
    - (8) -0.0003*** (-7.341)

  - Initial (log) real per capita GDP
    - (1) 0.0087** (2.452)
    - (2) 0.0028 (0.788)
    - (3) 0.0052 (1.477)
    - (4) 0.0057 (1.596)
    - (5) 0.0085** (2.199)
    - (6) 0.0246*** (20.733)
    - (7) 0.0108*** (14.510)

  - Trade openness
    - (1) 0.0005*** (9.235)
    - (2) 0.0005*** (9.277)
    - (3) 0.0005*** (8.275)
    - (4) 0.0004*** (7.637)
    - (5) 0.0004*** (3.668)
    - (6) 0.0005*** (4.665)

  - Terms of trade
    - (1) 0.0003 (0.839)
    - (2) 0.0003 (0.751)
    - (3) 0.0002 (0.416)
    - (4) 0.0000 (0.017)
    - (5) -0.0000 (-0.107)

  - Investment in percent of GDP
    - (1) 0.0019*** (3.593)
    - (2) 0.0018*** (3.177)
    - (6) 0.0012*** (12.115)
    - (7) 0.0013*** (12.816)

  - Population growth
    - (1) 0.0073*** (2.826)
    - (6) 0.0081*** (15.549)
    - (7) 0.0093*** (18.297)

  - Financial access
    - (6) -0.1292*** (-2.791)

  - Financial access * net Gini
    - (6) 0.0132*** (5.246)

  - Financial access * net Gini squared
    - (6) -0.0003*** (-9.835)

  - Female labor participation
    - (7) 0.0081*** (7.199)

  - Female labor participation * net Gini
    - (7) -0.0003*** (-5.027)

  - Female labor participation * net Gini squared
    - (7) 0.0000*** (4.854)

  - Constant
    - (1) 0.0777* (1.664)
    - (2) -0.0578 (-0.476)
    - (3) -0.0545 (-0.350)
    - (4) -0.0612 (-0.586)
    - (5) -0.1364 (-1.366)
    - (6) -0.1407 (-1.328)
    - (7) -0.3012*** (-10.711)
    - (8) -0.5698*** (-9.320)

  - Observations
    - (1) 1449
    - (2) 1449
    - (3) 1412
    - (4) 1397
    - (5) 1397
    - (6) 1395
    - (7) 1352
    - (8) 1352

### Notes on Table B.1
- Table B.1 presents the results of the mean group estimations.
- Conditional regressions include interaction terms: between Financial access and net Gini (column grouping), and between Female labor participation and net Gini (column grouping).
- Source: Authors’ calculations.
- Notes: ***, **, and * next to a number indicate statistical significance at 1, 5, and 10 percent, respectively.

*Source: wp1776 - Appendix A. Data Sources and Country Groups*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp1776.pdf_