## Appendix point to a statistically signif icant impact of pandemics on income inequality with

## Source details

**Canonical URL:** [Appendix point to a statistically signif icant impact of pandemics on income inequality with](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021127-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021127-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021127-print-pdf.pdf.json)

---

### Impact on other indicators of distribution
- Pandemic events raise income shares of the upper-income quintile and reduce shares of the lower-income quintile.
- Quantitative examples:
  - The share of income going to the top two deciles is 46 percent on average; five years after the pandemic, this share increases to nearly 48 percent.
  - The share of income going to the bottom two deciles is 6 percent; five years after the pandemic, this share falls to 5.5 percent.
- Similar effects are found for the top (bottom) 10 and 40 percent (referenced as Figure A9 in the Online Appendix).
- Employment effects by education:
  - Those with non-basic levels of education are scarcely affected.
  - The employment-to-population ratio of those with basic levels of education falls significantly, by more than 5 percent in the medium term—the effect is statistically significant at 5 percent.

### Pandemics vs. financial crises and other recessions
- Empirical specification augments pandemic analysis with financial crisis and recession indicators (financial crises from Laeven and Valencia 2019; recessions defined as years of negative real GDP growth).
- Key comparative findings:
  - Distributional effects of pandemics are larger than those associated with financial crises or recessions.
  - Financial crises do not have a significant effect on inequality.
  - The Gini increases by about 0.05 following a typical recession—compared to more than 0.4 for pandemics.
  - Recessions seem to result in higher top income shares, while top shares tend to decline in the medium term following a financial crisis.
  - The medium-term effects on the employment-to-population ratio of those with basic education fall significantly in all types of crises, indicating that employment channel alone does not explain the greater inequality impact of pandemics.

### Heterogeneity across episodes depending on severity of pandemics
- Two approaches to assess heterogeneity: a continuous severity measure (cases per population) and a regime-weighted specification with severity indicator z and weighting function F(z) with γ = 3.5.
- Continuous-measure findings:
  - A one percent increase in the measure of severity implies a rise in net Gini of about 0.4 (Figure 5 - Panel A).
  - The effect of an average pandemic—based on the average infection rate in the dataset (0.80 cases per 1000 inhabitants)—is associated with a medium-term increase in the net Gini of about 0.1.
  - As of November 30th, 2020, COVID-19 (average 7.5 cases per 1000 inhabitants, maximum rate of about 52 infections per 1000) would lead to a medium-term increase in the Gini of about 0.5.
- IV approach:
  - Instrument constructed by interacting a time-varying global dummy for pandemic outbreak years and a country-specific exogenous term (average temperature).
  - IV results confirm adverse distributional effects and yield larger magnitudes than corresponding OLS estimates.
- Regime-weighted (F(z)) findings:
  - Episodes with larger cases-to-population ratios (e.g., Croatia, H1N1, 2009) show statistically significant effects larger than the average: medium-term effect on Gini increases from 0.4 to about 0.8.
  - Episodes associated with small outbreaks (e.g., Philippines, SARS, 2003) are not statistically different from zero.
  - Episodes associated with low growth (e.g., Korea, MERS, 2012) have larger medium-term effects (about 0.7), while episodes associated with high growth (e.g., China, H1N1, 2009) are not statistically different from zero.
- Additional heterogeneity note:
  - The rise in inequality is higher when fiscal policy is tighter, while when the fiscal response is supportive, inequality barely increases.

### Conclusion and policy-relevant implications
- Historical evidence indicates pandemics have:
  - increased the Gini coefficient,
  - raised top income shares,
  - lowered employment-to-population ratios for those with basic education relative to higher education groups.
- Severity matters: more severe pandemics (by cases or output effects) produce larger distributional consequences, implying COVID-19 may have larger effects than earlier pandemics in the sample.
- Policy implications highlighted:
  - Supportive fiscal policies can mitigate the rise in inequality following pandemics.
  - Absence of supportive policies to protect the vulnerable risks significant increases in inequality.
- Areas for future research:
  - Cross-country variation in distributional effects depending on country characteristics, initial income distribution, containment stringency, and policy responses.
  - Differential economic effects of COVID-19 across population segments (race, age, gender).
  - Higher human cost of pandemics among low-income groups with poorer access to health services.

*Source: Appendix of wpiea2021127-print-pdf (IMF working paper).*

### ANNEX A — Scope, data coverage, and descriptive statistics
- Sample coverage and period:
  - Impulse response estimations use a sample of 177 countries over the period 1960-2019.
  - Subsamples: 1995-2019 restricted sample; 1981-2017 sample for income shares (64 countries); 1990-2017 sample for E/P ratios (76 countries).
- Key descriptive statistics (Table A1):
  - Gini Market (SWIID 8.3): Obs 5,472; Mean 45.39; Std. Dev. 6.59; No. of Countries 177.
  - Gini Net (SWIID 8.3): Obs 5,472; Mean 38.38; Std. Dev. 8.73; No. of Countries 177.
  - Top 40% Income Share (WDI): Obs 1,444; Mean 67.77; Std. Dev. 6.65; No. of Countries 64.
  - Top 20% Income Share (WDI): Obs 1,444; Mean 46.28; Std. Dev. 7.83; No. of Countries 64.
  - Top 10% Income Share (WDI): Obs 1,444; Mean 30.85; Std. Dev. 7.31; No. of Countries 64.
  - Bottom 40% Income Share (WDI): Obs 1,444; Mean 17.12; Std. Dev. 4.56; No. of Countries 64.
  - Bottom 20% Income Share (WDI): Obs 1,444; Mean 6.31; Std. Dev. 2.19; No. of Countries 64.
  - Bottom 10% Income Share (WDI): Obs 1,443; Mean 2.44; Std. Dev. 1.02; No. of Countries 64.
  - E/P ratio – Basic Education (ILO): Obs 1,340; Mean 42.51; Std. Dev. 16.22; No. of Countries 76.
  - E/P ratio – Non-basic Education (ILO): Obs 1,340; Mean 57.49; Std. Dev. 16.22; No. of Countries 76.
  - Financial Crises: Laeven and Valencia (2020) — 289 episodes; No. of Countries 177.
- Pandemic and epidemic episodes (Table A2):
  - Total Pandemic and Epidemic Events: 225.
  - SARS (2003) — 27 countries; Total Deaths 774; Total Cases 8,094; Total Mortality rate 9.56 percent; Average Cases/Pop (*100,000) 1.25; Average Mortality rate (%) 9.77.
  - H1N1 (2009) — 149 countries; Total Deaths 19,091; Total Cases 6,502,779; Total Mortality rate 0.29 percent; Average Cases/Pop (*100,000) 122.69; Average Mortality rate (%) 4.02.
  - MERS (2012) — 22 countries; Total Deaths 572; Total Cases 1,453; Total Mortality rate 39.37 percent; Average Cases/Pop (*100,000) 0.24; Average Mortality rate (%) 35.95.
  - Ebola (2014) — 6 countries; Total Deaths 8,767; Total Cases 24,809; Total Mortality rate 35.34 percent; Average Cases/Pop (*100,000) 74.37; Average Mortality rate (%) 16.34.
  - Zika (2016) — 21 countries; Total Deaths 20; Total Cases 198,122; Total Mortality rate 0.01 percent; Average Cases/Pop (*100,000) 76.21; Average Mortality rate (%) 0.03.
- Country-level case/death listings appear in Table A3 (countries with at least one reported case and data on income inequality).

### Main empirical findings — impact of pandemics on inequality (regression summaries)
- Baseline ADL estimates for Net Gini (Table A4, Panel A) — coefficients on pandemic dummies by horizon k:
  - D_{i,t}:
    - k=0: 0.017 (std. err. (0.028))
    - k=1: 0.065 (0.054)
    - k=2: 0.135** (0.065)
    - k=3: 0.232*** (0.086)
    - k=4: 0.325*** (0.109)
    - k=5: 0.414*** (0.128)
  - D_{i,t-1}:
    - k=0: 0.038 (0.029)
    - k=1: 0.105** (0.046)
    - k=2: 0.237*** (0.067)
    - k=3: 0.309*** (0.100)
    - k=4: 0.400*** (0.119)
    - k=5: 0.534*** (0.150)
  - D_{i,t-2}:
    - k=0: 0.037 (0.028)
    - k=1: 0.115* (0.060)
    - k=2: 0.233*** (0.080)
    - k=3: 0.316*** (0.103)
    - k=4: 0.346*** (0.130)
    - k=5: 0.401** (0.166)
  - Dynamics (lags of y):
    - y_{i,t-1}: coefficients range from 1.629*** to 2.624*** across horizons.
    - y_{i,t-2}: coefficients range from -0.651*** to -1.911*** across horizons.
  - Observations: 5,110 (k=0) down to 4,228 (k=5). R^2 up to 1.000 (k=0) down to 0.990 (k=5).
- Baseline ADL estimates for Market Gini (Table A4, Panel B):
  - D_{i,t}:
    - k=0: 0.050 (0.036)
    - k=1: 0.113** (0.052)
    - k=2: 0.182*** (0.070)
    - k=3: 0.219** (0.090)
    - k=4: 0.314*** (0.119)
    - k=5: 0.413*** (0.152)
  - Observations identical pattern: 5,110 down to 4,228. R^2 from 0.999 to 0.980.
- AIPW Average Treatment Effects (Table A6):
  - Panel A (ATE, restricted θ_1^h = θ_0^h):
    - k=0: 0.06*** (0.02)
    - k=1: 0.13*** (0.03)
    - k=2: 0.25*** (0.04)
    - k=3: 0.26*** (0.05)
    - k=4: 0.17*** (0.06)
    - k=5: 0.25*** (0.06)
    - Observations: 1,826.
  - Panel A (ATE, unrestricted θ_1^h ≠ θ_0^h):
    - k=0: 0.02 (0.02)
    - k=1: 0.16*** (0.04)
    - k=2: 0.27*** (0.05)
    - k=3: 0.36*** (0.06)
    - k=4: 0.38*** (0.08)
    - k=5: 0.62*** (0.08)
    - Observations: 1,826.
  - Panel B (alternative propensity score specification):
    - Restricted ATE:
      - k=0: 0.08*** (0.01)
      - k=1: 0.15*** (0.03)
      - k=2: 0.28*** (0.04)
      - k=3: 0.35*** (0.05)
      - k=4: 0.38*** (0.06)
      - k=5: 0.47*** (0.06)
      - Observations: 3,298.
    - Unrestricted ATE:
      - k=0: -0.01 (0.02)
      - k=1: 0.09** (0.04)
      - k=2: 0.16*** (0.06)
      - k=3: 0.15** (0.07)
      - k=4: 0.13 (0.09)
      - k=5: 0.27*** (0.10)
      - Observations: 3,298.
  - Notes: Empirical sandwich standard errors (clustered by country) in parentheses. Significance: *** p<0.01, ** p<0.05, * p<0.1.
- IV first-stage (Table A8):
  - Coefficient on IV term across horizons:
    - k=0: -0.002*** (t-statistic -2.770)
    - k=1: -0.002*** (t-statistic -3.570)
    - k=2: -0.003*** (t-statistic -3.530)
    - k=3: -0.003*** (t-statistic -3.840)
    - k=4: -0.004*** (t-statistic -3.850)
    - k=5: -0.004*** (t-statistic -4.340)
  - Observations: 4,888 (k=0) down to 4,075 (k=5).
  - Centered R-squared: 0.012 (k=0) up to 0.040 (k=5).
  - Kleibergen-Paap rk Wald F-statistic: 7.676 (k=0), 12.77 (k=1), 12.45 (k=2), 14.72 (k=3), 14.82 (k=4), 18.86 (k=5).
- F-tests for differences (Table A7) — selected entries:
  - Pandemics vs Financial Crises (F-test difference):
    - k=2: 3.976**
    - k=3: 5.807**
    - k=4: 7.106***
    - k=5: 8.473***
  - Pandemics vs Recessions:
    - k=3: 2.794*
    - k=4: 3.832*
    - k=5: 4.545**
  - Interaction with Cases-Population ratio:
    - k=5: 3.575*
  - Interaction with GDP Growth:
    - k=4: 3.142*
    - k=5: 3.727*
  - Significance notation: *** p<0.01, ** p<0.05, * p<0.1.

### Robustness checks and alternative specifications
- Alternative inequality measures:
  - Figure A3 compares Gini POVCAL (69 countries – 763 observations) and Gini WIDER (66 countries – 641 observations) with 90 percent confidence bands.
- ADL and Δy_{i,t} specifications, restricted samples (1995-2019), additional controls (GDP, GDP squared, population density, share urban, KOF trade globalization, KOF financial globalization), country-specific time trends, placebo tests, and income-share decile and employment outcomes are presented in Figures A4–A12 and show robustness of main results across samples and specifications.

### Endogeneity, identification strategies, and estimation methods
- Augmented Inverse Probability Weighting (AIPW):
  - Purpose: address potential endogeneity of pandemic treatment assignment using propensity-score reweighting combined with regression adjustment to estimate Average Treatment Effect (ATE).
  - Predictive Probit models include: GDP per capita (log), GDP growth rate, average temperature (year average), total health expenditures, general government final consumption expenditure (% of GDP), age dependency ratio (% of working-age population), mortality rate adult (per 1,000 adults), and alternative specifications with past levels/changes of disposable Gini.
  - Selected pooled Probit average marginal effects (Table A5):
    - Age dependency ratio: negative and significant (e.g., -0.046*** (0.016) in column 1).
    - GDP per capita (log): positive and significant (e.g., 0.965** (0.390) in column 1; 1.648*** (0.316) in column 4).
    - Temperature (Year average): positive and highly significant (e.g., 0.253*** (0.080) in column 1; 0.375*** (0.077) in column 4).
    - Mortality rate, adult: small positive effects (e.g., 0.002* (0.001) in column 1).
    - Observations vary: 1,912; 1,850; 3,485; 3,413.
  - Interpretation: pandemic dates are uncorrelated with past levels and changes of inequality but depend on country characteristics such as temperature and GDP per capita.
- AIPW estimator specifics:
  - Baseline model includes country fixed effects α_i^k, time fixed effects γ_t^k, treatment effect Λ_k on D_{i,t}, and controls X_{i,t} (two lags of dependent variable and two lags of pandemic dummy).
  - Propensity scores p̂_t come from Probit models; first-step conditional means I_{j,h}(X_t, θ̂_{j,h}) used in AIPW augmentation.
  - AIPW results (Table A6) are robust and quantitatively close to baseline impulse-response estimates.
- Instrumental Variable (IV) alternative:
  - Continuous severity measure: log10(1 + 1000*countries_cases/country_population).
  - IV constructed as interaction of a time-varying global dummy (1 in pandemic outbreak years for all countries) and average temperature (country-specific).
  - IV first-stage shows negative and highly significant coefficient on instrument at all horizons and Kleibergen–Paap rk Wald F-statistics indicating reasonable instrument strength.
  - IV interpreted in a differences-in-differences framework.

### Quantitative interpretation and timing of effects
- Across ADL, AIPW, and IV specifications, pandemics are associated with increases in inequality measures (Net Gini and Market Gini), with effects emerging and growing over the 1–5 year horizon after pandemic events.
- AIPW restricted estimates (Panel A) indicate ATEs:
  - k=0: 0.06*** (0.02)
  - k=1: 0.13*** (0.03)
  - k=2: 0.25*** (0.04)
  - k=3: 0.26*** (0.05)
  - k=4: 0.17*** (0.06)
  - k=5: 0.25*** (0.06)
- Baseline ADL point estimates on D_{i,t} and its lags indicate effects that become statistically significant at k≥2 and rise through k=5 (e.g., Net Gini D_{i,t} at k=5 = 0.414*** (0.128)).

*Source: ANNEX A of wpiea2021127-print-pdf (IMF).*

### Appendix point to a statistically signif icant impact of  pandemics on income inequality with

### wpiea2021127-print-pdf - Appendix point to a statistically signif icant impact of pandemics on income inequality with

### Impact on other indicators of distribution
- Pandemic events raise income shares of the upper-income quintile and reduce shares of the lower-income quintile.
- Quantitative examples:
  - The share of income going to the top two deciles is 46 percent on average; five years after the pandemic, this share increases to nearly 48 percent.
  - The share of income going to the bottom two deciles is 6 percent; five years after the pandemic, this share falls to 5.5 percent.
- Similar effects are found for the top (bottom) 10 and 40 percent (referenced as Figure A9 in the Online Appendix).
- Employment effects by education:
  - Those with non-basic levels of education are scarcely affected.
  - The employment-to-population ratio of those with basic levels of education falls significantly, by more than 5 percent in the medium term—the effect is statistically significant at 5 percent.

### Pandemics vs. financial crises and other recessions
- Empirical specification augments pandemic analysis with financial crisis and recession indicators (financial crises from Laeven and Valencia 2019; recessions defined as years of negative real GDP growth).
- Key comparative findings:
  - Distributional effects of pandemics are larger than those associated with financial crises or recessions.
  - Financial crises do not have a significant effect on inequality.
  - The Gini increases by about 0.05 following a typical recession—compared to more than 0.4 for pandemics.
  - Recessions seem to result in higher top income shares, while top shares tend to decline in the medium term following a financial crisis (consistent with financial income concentration in upper incomes).
  - The medium-term effects on the employment-to-population ratio of those with basic education fall significantly in all types of crises, indicating that employment channel alone does not explain the greater inequality impact of pandemics.

### Heterogeneity across episodes depending on severity of pandemics
- Two approaches used to assess heterogeneity: a continuous severity measure (cases per population) and a regime-weighted specification with severity indicator z (cases-to-population ratio or GDP growth) and weighting function F(z) with γ = 3.5.
- Continuous-measure findings:
  - A one percent increase in the measure of severity implies a rise in net Gini of about 0.4 (Figure 5 - Panel A).
  - The effect of an average pandemic—based on the average infection rate in the dataset (0.80 cases per 1000 inhabitants)—is associated with a medium-term increase in the net Gini of about 0.1.
  - As of November 30th, 2020, COVID-19 (average 7.5 cases per 1000 inhabitants, maximum rate of about 52 infections per 1000) would lead to a medium-term increase in the Gini of about 0.5.
- IV approach:
  - Instrumental Variable (IV) constructed by interacting a time-varying global dummy for pandemic outbreak years and a country-specific exogenous term (average temperature).
  - IV results confirm adverse distributional effects and yield larger magnitudes than corresponding OLS estimates.
- Regime-weighted (F(z)) findings:
  - Episodes with larger cases-to-population ratios (e.g., Croatia, H1N1, 2009) show statistically significant effects larger than the average: medium-term effect on Gini increases from 0.4 to about 0.8.
  - Episodes associated with small outbreaks (e.g., Philippines, SARS, 2003) are not statistically different from zero.
  - Episodes associated with low growth (e.g., Korea, MERS, 2012) have larger medium-term effects (about 0.7), while episodes associated with high growth (e.g., China, H1N1, 2009) are not statistically different from zero.
- Additional heterogeneity note:
  - The rise in inequality is higher when fiscal policy is tighter, while when the fiscal response is supportive, inequality barely increases.

### Conclusion and policy-relevant implications
- Historical evidence indicates pandemics have increased the Gini coefficient, raised top income shares, and lowered employment-to-population ratios for those with basic education relative to higher education groups.
- Severity matters: more severe pandemics (by cases or output effects) produce larger distributional consequences, implying COVID-19 may have larger effects than earlier pandemics in the sample.
- Policy implications highlighted by results:
  - Supportive fiscal policies can mitigate the rise in inequality following pandemics.
  - Absence of supportive policies to protect the vulnerable risks significant increases in inequality.
- Areas for future research emphasized:
  - Cross-country variation in distributional effects depending on country characteristics, initial income distribution, containment stringency, and policy responses.
  - Differential economic effects of COVID-19 across population segments (race, age, gender).
  - Higher human cost of pandemics among low-income groups with poorer access to health services.

*Source: Appendix of wpiea2021127-print-pdf (IMF working paper).*

### ANNEX A

### wpiea2021127-print-pdf - ANNEX A

### Scope, data coverage, and descriptive statistics
- Sample coverage and period:
  - Impulse response estimations use a sample of 177 countries over the period 1960-2019 (unless otherwise noted for subsamples).
  - Subsamples: 1995-2019 restricted sample; 1981-2017 sample for income shares (64 countries); 1990-2017 sample for E/P ratios (76 countries).
- Key descriptive statistics (Table A1):
  - Gini Market (SWIID 8.3): Obs 5,472; Mean 45.39; Std. Dev. 6.59; No. of Countries 177.
  - Gini Net (SWIID 8.3): Obs 5,472; Mean 38.38; Std. Dev. 8.73; No. of Countries 177.
  - Top 40% Income Share (WDI): Obs 1,444; Mean 67.77; Std. Dev. 6.65; No. of Countries 64.
  - Top 20% Income Share (WDI): Obs 1,444; Mean 46.28; Std. Dev. 7.83; No. of Countries 64.
  - Top 10% Income Share (WDI): Obs 1,444; Mean 30.85; Std. Dev. 7.31; No. of Countries 64.
  - Bottom 40% Income Share (WDI): Obs 1,444; Mean 17.12; Std. Dev. 4.56; No. of Countries 64.
  - Bottom 20% Income Share (WDI): Obs 1,444; Mean 6.31; Std. Dev. 2.19; No. of Countries 64.
  - Bottom 10% Income Share (WDI): Obs 1,443; Mean 2.44; Std. Dev. 1.02; No. of Countries 64.
  - E/P ratio – Basic Education (ILO): Obs 1,340; Mean 42.51; Std. Dev. 16.22; No. of Countries 76.
  - E/P ratio – Non-basic Education (ILO): Obs 1,340; Mean 57.49; Std. Dev. 16.22; No. of Countries 76.
  - Financial Crises: Laeven and Valencia (2020) — 289 episodes; No. of Countries 177.
- Pandemic and epidemic episodes listed (Table A2, Table A3):
  - Total Pandemic and Epidemic Events: 225.
  - Examples of aggregate pandemic metrics (Table A2): SARS (2003) — 27 countries; Total Deaths 774; Total Cases 8,094; Total Mortality rate 9.56 percent; Average Cases/Pop (*100,000) 1.25; Average Mortality rate (%) 9.77.
  - H1N1 (2009) — 149 countries; Total Deaths 19,091; Total Cases 6,502,779; Total Mortality rate 0.29 percent; Average Cases/Pop (*100,000) 122.69; Average Mortality rate (%) 4.02.
  - MERS (2012) — 22 countries; Total Deaths 572; Total Cases 1,453; Total Mortality rate 39.37 percent; Average Cases/Pop (*100,000) 0.24; Average Mortality rate (%) 35.95.
  - Ebola (2014) — 6 countries; Total Deaths 8,767; Total Cases 24,809; Total Mortality rate 35.34 percent; Average Cases/Pop (*100,000) 74.37; Average Mortality rate (%) 16.34.
  - Zika (2016) — 21 countries; Total Deaths 20; Total Cases 198,122; Total Mortality rate 0.01 percent; Average Cases/Pop (*100,000) 76.21; Average Mortality rate (%) 0.03.
- Country-level case/death listings appear in Table A3 (countries with at least one reported case and data on income inequality).

### Main empirical findings — impact of pandemics on inequality (summary of regression results)
- Baseline ADL estimates for Net Gini (Table A4, Panel A — coefficients on pandemic dummies and lags; dependent variable dynamics included):
  - Coefficients on D_{i,t} by horizon k:
    - k=0: 0.017 (std. err. (0.028))
    - k=1: 0.065 (0.054)
    - k=2: 0.135** (0.065)
    - k=3: 0.232*** (0.086)
    - k=4: 0.325*** (0.109)
    - k=5: 0.414*** (0.128)
  - Coefficients on D_{i,t-1}:
    - k=0: 0.038 (0.029)
    - k=1: 0.105** (0.046)
    - k=2: 0.237*** (0.067)
    - k=3: 0.309*** (0.100)
    - k=4: 0.400*** (0.119)
    - k=5: 0.534*** (0.150)
  - Coefficients on D_{i,t-2}:
    - k=0: 0.037 (0.028)
    - k=1: 0.115* (0.060)
    - k=2: 0.233*** (0.080)
    - k=3: 0.316*** (0.103)
    - k=4: 0.346*** (0.130)
    - k=5: 0.401** (0.166)
  - Dynamics (lags of y):
    - y_{i,t-1}: coefficients range from 1.629*** to 2.624*** across horizons.
    - y_{i,t-2}: coefficients range from -0.651*** to -1.911*** across horizons.
  - Observations decline with horizon: 5,110 (k=0) down to 4,228 (k=5). R^2 reported up to 1.000 (k=0) down to 0.990 (k=5).
- Baseline ADL estimates for Market Gini (Table A4, Panel B):
  - Coefficients on D_{i,t} by horizon k:
    - k=0: 0.050 (0.036)
    - k=1: 0.113** (0.052)
    - k=2: 0.182*** (0.070)
    - k=3: 0.219** (0.090)
    - k=4: 0.314*** (0.119)
    - k=5: 0.413*** (0.152)
  - Observations same pattern as Net Gini: 5,110 down to 4,228. R^2 from 0.999 to 0.980.
- AIPW Average Treatment Effects (Table A6):
  - Panel A (ATE, restricted θ_1^h = θ_0^h):
    - k=0: 0.06*** (0.02)
    - k=1: 0.13*** (0.03)
    - k=2: 0.25*** (0.04)
    - k=3: 0.26*** (0.05)
    - k=4: 0.17*** (0.06)
    - k=5: 0.25*** (0.06)
    - Observations: 1,826 (for all horizons reported).
  - Panel A (ATE, unrestricted θ_1^h ≠ θ_0^h):
    - k=0: 0.02 (0.02)
    - k=1: 0.16*** (0.04)
    - k=2: 0.27*** (0.05)
    - k=3: 0.36*** (0.06)
    - k=4: 0.38*** (0.08)
    - k=5: 0.62*** (0.08)
    - Observations: 1,826.
  - Panel B (alternative propensity score specification):
    - Restricted ATE:
      - k=0: 0.08*** (0.01)
      - k=1: 0.15*** (0.03)
      - k=2: 0.28*** (0.04)
      - k=3: 0.35*** (0.05)
      - k=4: 0.38*** (0.06)
      - k=5: 0.47*** (0.06)
      - Observations: 3,298.
    - Unrestricted ATE:
      - k=0: -0.01 (0.02)
      - k=1: 0.09** (0.04)
      - k=2: 0.16*** (0.06)
      - k=3: 0.15** (0.07)
      - k=4: 0.13 (0.09)
      - k=5: 0.27*** (0.10)
      - Observations: 3,298.
  - Notes: Empirical sandwich standard errors (clustered by country) in parentheses. Significance: *** p<0.01, ** p<0.05, * p<0.1.
- IV first-stage (Table A8):
  - Coefficient on IV term (interaction instrument) across horizons:
    - k=0: -0.002*** (t-statistic -2.770)
    - k=1: -0.002*** (t-statistic -3.570)
    - k=2: -0.003*** (t-statistic -3.530)
    - k=3: -0.003*** (t-statistic -3.840)
    - k=4: -0.004*** (t-statistic -3.850)
    - k=5: -0.004*** (t-statistic -4.340)
  - Observations by horizon: 4,888 (k=0) down to 4,075 (k=5).
  - Centered R-squared: 0.012 (k=0) up to 0.040 (k=5).
  - Kleibergen-Paap rk Wald F-statistic: 7.676 (k=0), 12.77 (k=1), 12.45 (k=2), 14.72 (k=3), 14.82 (k=4), 18.86 (k=5) — interpreted as “strong” instrument in context (statistic higher than Stock-Yogo critical value per note).
- F-tests for differences (Table A7):
  - Pandemics vs Financial Crises (F-test difference):
    - k=0: 0.001
    - k=1: 0.895
    - k=2: 3.976** 
    - k=3: 5.807**
    - k=4: 7.106***
    - k=5: 8.473***
  - Pandemics vs Recessions:
    - k=0: 0.075
    - k=1: 0.472
    - k=2: 1.521
    - k=3: 2.794*
    - k=4: 3.832*
    - k=5: 4.545**
  - Interaction with Cases-Population ratio:
    - k=0: 0.159
    - k=1: 0.0301
    - k=2: 0.295
    - k=3: 1.425
    - k=4: 2.642
    - k=5: 3.575*
  - Interaction with GDP Growth:
    - k=0: 0.006
    - k=1: 0.615
    - k=2: 0.539
    - k=3: 2.091
    - k=4: 3.142*
    - k=5: 3.727*
  - Significance notation: *** p<0.01, ** p<0.05, * p<0.1.

### Robustness checks and alternative specifications (figures and tables referenced)
- Alternative inequality measures:
  - Figure A3 compares results using Gini POVCAL (69 countries – 763 observations) and Gini WIDER (66 countries – 641 observations); impulse responses reported with 90 percent confidence bands.
- ADL specification:
  - Figure A4 presents ADL estimates based on Δy_{i,t} specification using 177 countries (1960-2019).
- Restricted sample (1995-2019):
  - Figure A5 reports impulse responses estimated on 177 countries over 1995-2019.
- Additional controls:
  - Figure A6 uses sample of 175 countries over 1960-2017; X includes two lags of dependent variable, two lags of pandemic dummy, level of GDP, GDP squared, population density, share urban, KOF trade globalization, KOF financial globalization.
- Country-specific time trends:
  - Figure A7 includes country-specific time trends (177 countries, 1960-2019).
- Placebo test:
  - Figure A8 randomly attributes D_{i,t} across the sample to test for spurious responses (177 countries, 1960-2019).
- Income-share decile and employment outcomes:
  - Figures A9–A12 show robustness checks for shares of income by deciles (64 countries, 1981-2017) and E/P ratios by education (76 countries, 1990-2017); estimations include 90 percent confidence bands.

### Endogeneity, identification strategies, and estimation methods
- Augmented Inverse Probability Weighting (AIPW) approach (Annex B):
  - Purpose: address potential endogeneity of pandemic treatment assignment using propensity-score-based reweighting combined with regression adjustment to estimate Average Treatment Effect (ATE).
  - Predictive Probit models for pandemic likelihood include: GDP per capita (log), GDP growth rate, average temperature (year average), total health expenditures, general government final consumption expenditure (% of GDP), age dependency ratio (% of working-age population), mortality rate adult (per 1,000 adults), and alternative specifications including past levels/changes of disposable Gini.
  - Table A5 (Pooled Probit average marginal effects) — selected coefficients and signs:
    - Age dependency ratio (% of working-age population): negative and significant across columns (e.g., -0.046*** (0.016) in column 1).
    - GDP per capita (log): positive and significant (e.g., 0.965** (0.390) in column 1; 1.648*** (0.316) in column 4).
    - Temperature (Year average): positive and highly significant (e.g., 0.253*** (0.080) in column 1; 0.375*** (0.077) in column 4).
    - Mortality rate, adult (per 1,000 adults): small positive effects (e.g., 0.002* (0.001) in column 1).
    - Total Health Expenditures not significant in reported columns.
    - Observations: vary by column — 1,912; 1,850; 3,485; 3,413.
  - Interpretation: pandemic dates are uncorrelated with past levels and changes of inequality but depend on country characteristics such as temperature and GDP per capita.
- AIPW estimator specifics:
  - Baseline model (equation (A1)) includes country fixed effects α_i^k, time fixed effects γ_t^k, treatment effect Λ_k on D_{i,t}, and controls X_{i,t} (two lags of dependent variable and two lags of pandemic dummy).
  - Propensity scores p̂_t come from Probit models (Table A5). First-step conditional means I_{j,h}(X_t, θ̂_{j,h}) used in AIPW augmentation.
  - Results reported in Table A6 are robust across specifications and point to statistically significant pandemic impacts on income inequality, quantitatively close to baseline impulse-response estimates.
- Instrumental Variable (IV) alternative identification:
  - Continuous severity measure used: log10(1 + 1000*countries_cases/country_population) — differentiates episodes by severity but raises reverse-causality and measurement error concerns.
  - IV constructed as interaction of a time-varying global dummy (1 in pandemic outbreak years for all countries) and a constant country-specific term (average temperature).
    - Rationale: temperature affects pandemic evolution and can be treated as exogenous; interaction with global outbreak indicator yields cross-country variation in severity.
  - IV first-stage results (Table A8) show a negative and highly significant coefficient on the instrument at all horizons (see values above) and Kleibergen–Paap rk Wald F-statistics indicating reasonable strength of the instrument (values from 7.676 to 18.86).
  - The IV strategy controls for country and time fixed effects and is interpreted in a differences-in-differences framework per Nunn and Qian (2014).

### Quantitative interpretation and timing of effects
- Across specifications (ADL, AIPW, IV), pandemics are associated with increases in inequality measures (both Net Gini and Market Gini), with effects emerging and growing over the 1–5 year horizon after pandemic events in many specifications.
- AIPW restricted estimates (Panel A) indicate ATEs for Net or Market Gini on the order of:
  - 0.06 (k=0), 0.13 (k=1), 0.25 (k=2), 0.26 (k=3), 0.17 (k=4), 0.25 (k=5) — all statistically significant at *** p<0.01.
- Baseline ADL point estimates on D_{i,t} and its lags indicate effects that become statistically significant at k≥2 and rise through k=5 (e.g., Net Gini D_{i,t} at k=5 = 0.414*** (0.128)).

*Source: ANNEX A of wpiea2021127-print-pdf (IMF).*

### References

### References

### Econometrics and Instrumental Variables
- Andrews, Isaiah, James H. Stock, and Liyang Sun. "Weak instruments in instrumental variables regression: Theory and practice." Annual Review of Economics 11 (2019): 727-753.

### Fiscal Policy and Austerity
- Jordà, Ò., & Taylor, A. M. 2016. The time for austerity: estimating the average treatment effect of fiscal policy. The Economic Journal, 126(590), 219-255.

### COVID-19: Containment Measures and Temperature Effects
- Deb, Pragyan and Furceri, Davide and Ostry, Jonathan D. and Tawk, Nour, The Economic Effects of Covid-19 Containment Measures (July 2020). CEPR Discussion Paper No. DP15087, Available at SSRN: https://ssrn.com/abstract=3661431
- Ma, Y., Zhao, Y., Liu, J., He, X., Wang, B., Fu, S., ... & Luo, B. 2020b. “Effects of temperature variation and humidity on the death of COVID-19 in Wuhan, China. Science of The Total Environment, 724: 138226.
- Ujiie, M., Tsuzuki, S., & Ohmagari, N. 2020. Effect of temperature on the infectivity of COVID-19. International Journal of Infectious Diseases. 95: 301-303.

### Food Aid and Conflict
- Nunn, N. and Qian, N., 2014. US food aid and civil conflict. American Economic Review, 104(6), pp.1630-66.

*Source: wpiea2021127-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021127-print-pdf.pdf_