## ANNEX A

## Source details

**Canonical URL:** [ANNEX A](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021018-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021018-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021018-print-pdf.pdf.json)

---

### Data sources and descriptive statistics (Table A1)
- Output performance:
  - Output performance 1 — Source: IMF — Obs: 97 — Mean: -17.60 — Std. dev: 12.44
  - Output performance 2 — Source: IMF — Obs: 97 — Mean: -16.67 — Std. dev: 13.15
- Health factors:
  - Health condition — Source: JHU — Obs: 191 — Mean: 40.58 — Std. dev: 14.41
  - Containment stringency — Source: OxCGRT — Obs: 183 — Mean: 0.59 — Std. dev: 0.20
  - Deaths per capita (log) — Source: JHU — Obs: 175 — Mean: -11.02 — Std. dev: 1.95
- Policy support:
  - Liquidity — Source: IMF — Obs: 194 — Mean: 1.25 — Std. dev: 3.50
  - Policy rate cut — Source: IMF — Obs: 194 — Mean: 87.29 — Std. dev: 134.04
  - Fiscal stimulus — Source: IMF — Obs: 194 — Mean: 4.26 — Std. dev: 6.09
- Regulation:
  - Labor market regulation — Source: FI — Obs: 157 — Mean: 6.46 — Std. dev: 1.36
  - Credit market regulation — Source: FI — Obs: 157 — Mean: 8.18 — Std. dev: 1.55
  - Business regulation — Source: FI — Obs: 157 — Mean: 6.75 — Std. dev: 1.27
- Macroeconomic factors:
  - Financial globalization — Source: KOF — Obs: 180 — Mean: 63.19 — Std. dev: 19.67
  - Trade globalization — Source: KOF — Obs: 183 — Mean: 56.41 — Std. dev: 20.37
  - Current account (% of GDP) — Source: IMF — Obs: 132 — Mean: -1.72 — Std. dev: 8.36
  - Financial system deposit (% of GDP) — Source: FSD — Obs: 163 — Mean: 59.93 — Std. dev: 50.73
  - Government debt (% of GDP) — Source: FSD — Obs: 115 — Mean: 56.34 — Std. dev: 37.13
  - Domestic credit (% of GDP) — Source: FSD — Obs: 165 — Mean: 57.31 — Std. dev: 43.31
  - Bank concentration — Source: FSD — Obs: 160 — Mean: 65.69 — Std. dev: 19.30
  - Exchange rate regime — Source: IMF — Obs: 192 — Mean: 2.07 — Std. dev: 0.87
  - Average GDP growth, 17-19 — Source: IMF — Obs: 199 — Mean: 2.97 — Std. dev: 3.60
- Sectoral composition:
  - Tourism (% of GDP) — Source: WTTC — Obs: 174 — Mean: 13.96 — Std. dev: 13.07
  - Service (% of GDP) — Source: WDI — Obs: 193 — Mean: 56.60 — Std. dev: 13.36
  - Industry (% of GDP) — Source: WDI — Obs: 202 — Mean: 25.43 — Std. dev: 12.45
- Development and other controls:
  - Share of population over 65 — Source: WDI — Obs: 190 — Mean: 8.34 — Std. dev: 5.88
  - Population (log) — Source: WDI — Obs: 211 — Mean: 15.29 — Std. dev: 2.41
  - GDP per capita (log) — Source: WDI — Obs: 208 — Mean: 8.83 — Std. dev: 1.50
  - Democratization — Source: Polity IV — Obs: 152 — Mean: 3.87 — Std. dev: 3.92
  - Population density — Source: WDI — Obs: 209 — Mean: 454.25 — Std. dev: 2085
  - Gini coefficient — Source: SWIID 7.1 — Obs: 165 — Mean: 0.39 — Std. dev: 0.08
  - Social fractionalization — Source: Alesina et al. (2003) — Obs: 179 — Mean: 0.44 — Std. dev: 0.19
  - Informality — Source: WDI — Obs: 143 — Mean: 29.30 — Std. dev: 14.25
  - Remittance inflow to GDP (%) — Source: FSD — Obs: 180 — Mean: 4.68 — Std. dev: 6.61
- Notes:
  - Output performance 1 is the difference between the observed cumulative real GDP growth in 2020H1 and the cumulative growth that was expected before the onset of the pandemic for the same period—based on the IMF World Economic Outlook 2020 January forecast for 2020H1.
  - Output performance 2 is the difference in cumulative real GDP growth between the first half of 2020 (2020H1) and the first half of 2019 (2019H1).
  - Data source abbreviations: IMF; JHU; OxCGRT; FI; KOF; FSD; WDI; WTTC.

### Bivariate relationships from Figures A1–A6 (selected regression lines and R²)
- Output performance vs. Deaths per capita (log): y = -0.771x -23.948 — R² = 0.0326
- Output performance vs. Containment stringency: y = -8.9567x -10.721 — R² = 0.0518
- Output performance vs. Health condition: y = -0.0529x -13.281 — R² = 0.0066
- Output performance vs. Service (% of GDP): y = -0.2249x -2.8807 — R² = 0.0661
- Output performance vs. Industry (% of GDP): y = 0.3072x -24.09 — R² = 0.0987
- Output performance vs. Tourism (% of GDP): y = -0.4145x -11.102 — R² = 0.1671
- Output performance vs. Credit market regulation: y = -1.4515x -3.0507 — R² = 0.0393
- Output performance vs. Liquidity: y = -0.1037x -15.971 — R² = 0.0008
- Output performance vs. Policy rate cuts: y = 0.0157x -17.477 — R² = 0.0315
- Output performance vs. Fiscal stimulus (% of GDP): y = -0.184x -15.151 — R² = 0.0077
- Output performance vs. Average GDP growth, 17-19: y = -0.8301x -13.475 — R² = 0.0361
- Output performance vs. Government debt (% of GDP): y = -0.0624x -12.635 — R² = 0.0559
- Output performance vs. Democracy: y = -0.227x -14.878 — R² = 0.0137
- Output performance vs. GDP per capita (log): y = 0.0523x -16.589 — R² = 7E-05

### Robust drivers across countries (Figures A7–A12 — qualitative magnitudes)
- Variables consistently associated with worse output performance (negative effect from 25th to 75th percentile):
  - Average GDP growth, 17-19
  - Credit market regulation
  - Tourism (% of GDP)
  - Containment stringency
  - Deaths per capita (log)
  - Government debt (% of GDP)
  - Social fractionalization (in some specifications)
  - Informality (in some specifications)
- Variables associated with better output performance (positive effect from 25th to 75th percentile):
  - Fiscal stimulus
  - GDP per capita (log)
  - Financial globalization (in selected extended covariate models)
  - Remittance inflow to GDP (%) (in some interaction specifications)
- Interaction with income level:
  - Effects of liquidity, remittance inflow, democracy, credit market regulation, containment stringency, deaths per capita (log), tourism (% of GDP), and GDP per capita (log) vary with continuous and dummy definitions of income level.

### ANNEX B

### Research question, sample, and outcome measures
- Context:
  - The COVID-19 recession is described as unprecedented and larger than the Global Financial Crisis (IMF, 2020).
  - For a sample of 60 advanced, emerging and developing economies, unexpected growth shocks in the first semester of 2020 range from a few percentage points (e.g., Korea) to more than 30 percentage points (e.g., Peru).
- Research aim:
  - Identify robust drivers of heterogeneous output losses in the acute phase (first semester of 2020).
- Two measures of output performance:
  - (i) actual growth in the first semester of 2020 minus the January 2020 IMF growth forecast for this period;
  - (ii) growth in the first semester of 2020 minus the growth rate for the first semester of 2019.
- Quarterly GDP growth observations: 96.
- Baseline regressors: 30 variables grouped into six categories: (i) Public health; (ii) Sectoral composition; (iii) Fiscal and monetary response; (iv) Macroeconomic characteristics; (v) Regulation; (vi) Development level, Demographics and Institutions.
- Robustness extension: up to 34 (and later 36) regressors with a more limited sample.

### Empirical framework and identification strategy
- Base specification: Linear reduced form Y_i = α + βX_i + μ_i, where X is a k-vector of covariates.
- Econometric challenges: (i) limited number of observations; (ii) large set of potential correlated regressors; (iii) unknown true model.
- Model-averaging techniques:
  - Weighted Average Least Squares (WALS) as baseline (advantages: orthogonal transformation of auxiliary regressors, Laplace prior for auxiliary parameters, computationally efficient).
  - Bayesian Model Averaging (BMA) as robustness check (Gaussian prior for auxiliary parameters; considers full model space).
- Scope of computation:
  - More than 1.07 billion regressions via model-averaging.
  - BMA considers the full model space of 2^30 = 1,073,741,824 models.
- Variable inclusion rules:
  - WALS: regressor considered robust if |t-statistic| > 1.
  - BMA: variables with high posterior inclusion probabilities deemed robust.

### Potential determinants examined (variables and measurement)
- Public health indicators:
  - log of deaths per capita—cumulative deaths as of June 30 relative to population;
  - containment stringency index from OxCGRT, normalized from 0 to 1;
  - Global Health Security Index from Johns Hopkins University.
- Industry shares:
  - Services, manufacturing, tourism (2019 shares in value added; agriculture excluded).
- Fiscal and monetary policy measures (IMF Covid-19 Policy Tracker; Dec 2019–June 2020):
  - total fiscal stimulus (above and below the line) deployed (or announced);
  - cumulative change in the policy interest rate;
  - liquidity injected by central banks (percent of GDP).
- Regulation (Fraser Institute Index; range 0–10, higher = less regulation):
  - Credit market deregulation; labor market deregulation; business deregulation.
- Macroeconomic fundamentals and financial variables:
  - Current account balance (% of GDP); general government debt-to-GDP ratio; financial system deposits (% of GDP); bank concentration; domestic credit (% of GDP); trade and financial globalization indices (KOF); exchange rate regime (1=fixed, 2=intermediate, 3=flexible); three-year average pre-crisis GDP growth.
- Development, demographics and institutions:
  - (log) GDP per capita; Gini coefficient (after-tax income); share of population over 65; (log) population; population density; remittances (% of GDP); ethnic and religious fractionalization; informality indicator; Polity IV democracy score.

### Main empirical findings (WALS robust determinants)
- Robust negative correlates of output performance (associated with larger output losses):
  - Lower GDP per capita.
  - More stringent containment measures.
  - Higher deaths per capita.
  - Larger tourism share in value added.
  - More liberalized credit markets (less stringent credit market regulation).
  - Higher pre-crisis growth.
  - More democratic political regimes.
- Additional relationships (one measure or less robust):
  - Lower fiscal stimulus (correlated with larger output loss for at least one measure).
  - Higher social fractionalization (correlated with larger output loss for at least one measure).
- Quantitative magnitudes (inter-quartile / percentile comparisons via WALS coefficients):
  - A country at the 75th percentile of GDP per capita has a 7-percentage-point smaller growth surprise than a country at the 25th percentile.
  - Moving from the 25th to the 75th percentile of containment or deaths is associated with an increase in output losses of 4½-5 percentage points.
  - A large tourism share implies about a 3½ percentage point larger-than-expected output loss across the inter-quartile range.
  - Differences in democracy scores are associated with about a 4 percentage point difference in output growth losses.

### Additional empirical insights and robustness checks
- Fiscal and monetary policy:
  - As of June 30, 2020, fiscal packages announced in more than 90 countries ranged from 1 to 23 percent of GDP.
  - Policy rate cuts took place in 97 countries from December 2019 to June 2020.
  - In simple associations, only policy rate cuts appear associated with lower output loss; fiscal stimulus variables are not consistently statistically significant in OLS/WALS.
  - WALS indicates higher fiscal stimulus is associated with smaller output losses in some robustness checks (e.g., after winsorizing).
- Regulation:
  - Countries with freer financial/credit markets are associated with larger output losses.
  - Labor and business regulation do not show robust significant relationships in baseline WALS.
- Macroeconomic characteristics:
  - Pre-crisis growth and debt-to-GDP ratio emerge as robust determinants in some specifications (higher pre-crisis growth associated with larger output loss; higher debt-to-GDP ratio associated with weaker performance in some checks).
  - Trade and financial openness were not robust drivers in the baseline.
- Robustness checks:
  - Winsorizing dependent variable at the upper and lower 5 percentiles produces broadly similar results; winsorized results make higher fiscal stimulus a robust driver for the first measure.
  - BMA posterior inclusion probabilities identify containment stringency, tourism share, and deaths per capita as variables with highest inclusion probabilities.
  - Variables with ≥10 percent posterior inclusion typically include pre-crisis growth, credit market regulation, GDP per capita, and democracy.
  - Expanded variable set (up to 36 controls, restricted sample) confirms GDP per capita, containment, deaths, and tourism as most robust; in restricted sample, rule of law, debt-to-GDP ratio, and smaller fiscal stimulus are also associated with higher output losses.
  - Using only Q2 2020 yields similar findings and additionally indicates that higher debt-to-GDP ratio, larger current account surpluses, and more flexible exchange rates tend to be associated with weaker performance.

### Mediating channels and interactions with income level
- Interactions examined between GDP per capita (continuous and dummy definitions) and deaths per capita, containment stringency, and policy response variables.
- WALS interaction results:
  - Output costs of containment and deaths are larger in lower-income countries.
  - Monetary stimulus (particularly liquidity provision) has been less effective in poorer countries.
  - Some evidence that fiscal stimulus effectiveness is lower in poorer countries.

### Policy implications and conclusions
- Main conclusions:
  - Countries experiencing smaller output losses in the acute phase are associated with: higher GDP per capita; less stringent containment; fewer deaths per capita; smaller tourism sectors; less flexible credit markets; lower pre-crisis growth; higher fiscal stimulus; less social fractionalization; and less democratic regimes.
  - GDP per capita is the quantitatively largest robust factor: a 75th vs 25th percentile difference corresponds to a 7-percentage-point difference in output surprise.
  - Death rates and containment stringency have similar quantitative effects (~4½-5 percentage points across inter-quartile changes), implying close links between saving lives and saving the economy.
- Policy messages:
  - Relax containment only when new infections are declining and pair rollback with strong testing and contact tracing.
  - Fiscal stimulus has helped reduce economic losses; premature withdrawal is self-defeating.
  - Improve monetary policy transmission in emerging and developing economies.
  - Targeted support is essential for high-contact sectors (tourism, retail).

### Key numeric results from WALS and BMA summaries (selected)
- From TABLE 7 (WALS, N 60) — selected magnitudes (t-statistic reported in table; bold indicates robust):
  - Health condition: 0.41 (OP1), 0.27 (OP2)
  - Containment stringency: -2.23 (OP1), -2.38 (OP2)
  - Deaths per capita (log): -2.44 (OP1), -2.65 (OP2)
  - Liquidity: -0.81 (OP1), -0.84 (OP2)
  - Policy rate cut: 0.37 (OP1), 0.48 (OP2)
  - Fiscal stimulus: 0.85 (OP1), 1.20 (OP2)
  - Credit market regulation: -1.59 (OP1), -1.63 (OP2)
  - Government debt (% of GDP): -0.92 (OP1), -0.95 (OP2)
  - Average GDP growth, 17-19: -1.05 (OP1), -1.71 (OP2)
  - Tourism (% of GDP): -3.01 (OP1), -2.75 (OP2)
- BMA posterior-inclusion-probabilities (TABLE 9, N 60):
  - Tourism (% of GDP): 0.99 (OP1), 0.85 (OP2)
  - Containment stringency: 0.97 (OP1), 0.88 (OP2)
  - Deaths per capita (log): 0.56 (OP1), 0.54 (OP2)
  - Government debt (% of GDP): 0.17 (OP1), 0.17 (OP2)
  - Democracy: 0.14 (OP1), 0.08 (OP2)
  - GDP per capita (log): 0.11 (OP1), 0.10 (OP2)
  - Fiscal stimulus: 0.08 (OP1), 0.08 (OP2)
  - Credit market regulation: 0.07 (OP1), 0.11 (OP2)
  - Average GDP growth, 17-19: 0.07 (OP1), 0.32 (OP2)

*Italic: Content derived from wpiea2021018-print-pdf — ANNEX A and ANNEX B of the provided IMF working paper PDF content.*

### ANNEX A ................................................................................................................

### ANNEX A

### Table
- Table A1. Sources and descriptive statistics of the variables used in the analysis ................................................................................... 39

### Figures — Output performances and public health, composition, policy, and robustness
- Figure A1. Output performances (%) and public health ........................................................................................... 41
- Figure A2. Output performances (%) and sectoral composition .............................................................................. 41
- Figure A3. Output performances (%) and fiscal and monetary response .................................................................. 42
- Figure A4. Output performances (%) and regulation ............................................................................................... 42
- Figure A5. Output performances (%) and macroeconomics characteristics .............................................................. 43
- Figure A6. Output performances (%) and development, demographic and institutions .......................................... 44
- Figure A7. Robust drivers of output performance across countries—controlling for outliers, magnitude of the effects ............................................................................................................ 45
- Figure A8. Robust drivers of output performance across countries—additional covariates, magnitude of the effects ............................................................................................................ 46
- Figure A9. Robust drivers of output performance across countries—using Q2 deviation as dependent variable .................................................................................................................... 47
- Figure A10. Robust drivers of output performance across countries— interaction with income level (continuous), magnitude of the effects ............................................................................. 48
- Figure A11. Robust drivers of output performance across countries— interaction with income level (dummy 1), magnitude of the effects ............................................................................... 49
- Figure A12. Robust drivers of output performance across countries— interaction with income level (dummy 2), magnitude of the effects ............................................................................... 50

*Content unit: wpiea2021018-print-pdf - ANNEX A*

### ANNEX B ................................................................................................................

### ANNEX B

### I. Introduction and key research question
- Context: The COVID-19 recession is described as unprecedented and larger than the Global Financial Crisis (IMF, 2020).  
- Cross-country heterogeneity: For a sample of 60 advanced, emerging and developing economies, unexpected growth shocks in the first semester of 2020 range from a few percentage points (e.g., Korea) to more than 30 percentage points (e.g., Peru).
- Research aim: Identify robust drivers of heterogeneous output losses in the acute phase (first semester of 2020), focusing on two measures of output performance:
  - (i) actual growth in the first semester of 2020 minus the January 2020 IMF growth forecast for this period;
  - (ii) growth in the first semester of 2020 minus the growth rate for the first semester of 2019.

### II. Empirical framework and identification strategy
- Base specification: Linear reduced form Y_i = α + βX_i + μ_i, where X is a k-vector of covariates.
- Econometric challenges: (i) limited number of observations; (ii) large set of potential correlated regressors; (iii) unknown true model.
- Model-averaging techniques used:
  - Weighted Average Least Squares (WALS) as baseline (advantages: orthogonal transformation of auxiliary regressors, Laplace prior for auxiliary parameters, computationally efficient).
  - Bayesian Model Averaging (BMA) as robustness check (Gaussian prior for auxiliary parameters; considers full model space).
- Scope of computation:
  - More than 1.07 billion regressions via model-averaging.
  - BMA considers the full model space of 2^30 = 1,073,741,824 models.
- Variable inclusion rule:
  - WALS: regressor considered robust if |t-statistic| > 1 (criterion based on improvement in adjusted R^2 and MSE).
  - BMA: variables with high posterior inclusion probabilities are deemed robust.

### III. Data, sample and covariate set
- Quarterly GDP growth observations: 96.
- Baseline regressors: 30 variables grouped into six categories:
  - (i) Public health; (ii) Sectoral composition; (iii) Fiscal and monetary response; (iv) Macroeconomic characteristics; (v) Regulation; (vi) Development level, Demographics and Institutions.
- Robustness extension: up to 34 (and later 36) regressors with a more limited sample.

### IV. Potential determinants examined (variables and measurement)
- Public health indicators:
  - (i) log of deaths per capita—cumulative deaths as of June 30 relative to population;
  - (ii) containment stringency index from the Oxford Coronavirus Government Response Tracker, normalized from 0 to 1;
  - (iii) Global Health Security Index from Johns Hopkins University.
- Industry shares:
  - Services, manufacturing, tourism (2019 shares in value added; agriculture excluded).
- Fiscal and monetary policy measures (IMF Covid-19 Policy Tracker; Dec 2019–June 2020):
  - (i) total fiscal stimulus (above and below the line) deployed (or announced);
  - (ii) cumulative change in the policy interest rate;
  - (iii) liquidity injected by central banks (percent of GDP).
- Regulation (Fraser Institute Index; range 0–10, higher = less regulation):
  - Credit market deregulation; labor market deregulation; business deregulation.
- Macroeconomic fundamentals and financial variables:
  - Current account balance (% of GDP); general government debt-to-GDP ratio; financial system deposits (% of GDP); bank concentration; domestic credit (% of GDP); trade and financial globalization indices (KOF); exchange rate regime (1=fixed, 2=intermediate, 3=flexible); three-year average pre-crisis GDP growth.
- Development, demographics and institutions:
  - (log) GDP per capita; Gini coefficient (after-tax income); share of population over 65; (log) population; population density; remittances (% of GDP); ethnic and religious fractionalization; informality indicator; Polity IV democracy score.

### V. Main empirical findings (robust determinants from WALS)
- Robust negative correlates of output performance (i.e., associated with larger output losses):
  - Lower GDP per capita.
  - More stringent containment measures.
  - Higher deaths per capita.
  - Larger tourism share in value added.
  - More liberalized credit markets (less stringent credit market regulation).
  - Higher pre-crisis growth.
  - More democratic political regimes.
- Additional relationships (one measure or less robust):
  - Lower fiscal stimulus (correlated with larger output loss for at least one measure).
  - Higher social fractionalization (correlated with larger output loss for at least one measure).
- Quantitative magnitudes (inter-quartile / percentile comparisons via WALS coefficients):
  - A country at the 75th percentile of GDP per capita (example: Portugal) has a 7-percentage-point smaller growth surprise than a country at the 25th percentile (example: Bangladesh).
  - Moving from the 25th to the 75th percentile of containment or deaths is associated with an increase in output losses of 4½-5 percentage points.
  - A large tourism share implies about a 3½ percentage point larger-than-expected output loss across the inter-quartile range.
  - Differences in democracy scores are associated with about a 4 percentage point difference in output growth losses (less democratic countries like China and Vietnam associated with smaller losses).

### VI. Additional empirical insights and nuanced results
- Fiscal and monetary policy:
  - As of June 30, 2020, fiscal packages announced in more than 90 countries ranged from 1 to 23 percent of GDP.
  - Policy rate cuts took place in 97 countries from December 2019 to June 2020.
  - In simple associations, only policy rate cuts appear associated with lower output loss; however, fiscal stimulus variables are not consistently statistically significant in OLS/WALS—possible reasons include announcement vs. implementation timing, lagged effects, heterogeneous impacts, omitted variable bias, and reverse causality.
  - WALS indicates higher fiscal stimulus is associated with smaller output losses in some robustness checks (e.g., after winsorizing).
- Regulation:
  - Countries with freer financial/credit markets (higher credit deregulation scores) are less resilient—i.e., associated with larger output losses.
  - Labor and business regulation measures do not show robust significant relationships in baseline WALS.
- Macroeconomic characteristics:
  - Among many macro variables considered, pre-crisis growth and debt-to-GDP ratio emerge as robust determinants (higher pre-crisis growth associated with larger output loss; higher debt-to-GDP ratio associated with weaker performance in some checks).
  - Trade and financial openness were not robust drivers in the baseline, in contrast with findings for the GFC.
- Development, demographics, institutions:
  - Inequality and informality show association with larger output losses; evidence confirmed by WALS results.
  - Remittances and population measures show suggestive patterns but are not highlighted as core robust drivers in baseline.

### VII. Robustness checks and alternative specifications
- Outliers:
  - Winsorizing the dependent variable at the upper and lower 5 percentiles produces broadly similar results to the baseline. For the first measure, winsorized results confirm baseline drivers and additionally make higher fiscal stimulus a robust driver.
- BMA checks:
  - BMA posterior inclusion probabilities identify containment stringency, tourism share, and deaths per capita as variables with the highest inclusion probabilities.
  - Variables with ≥10 percent posterior inclusion typically include pre-crisis growth, credit market regulation, GDP per capita, and democracy.
  - For the second output-loss measure, share of elderly shows a 35 percent posterior probability of inclusion.
- Expanded variable set:
  - Adding further controls (trade and capital account measures, rule of law, NPL shares, poverty, central bank reserves in percent of imports) to reach 36 controls (with a more limited sample) confirms GDP per capita, containment, deaths, and tourism as most robust.
  - In the restricted sample, rule of law, debt-to-GDP ratio, and smaller fiscal stimulus are also associated with higher output losses; democracy loses significance in that sample due to correlation with rule of law.
- Alternative period:
  - Re-estimating using only second quarter of 2020 yields similar findings and additionally indicates that higher debt-to-GDP ratio, larger current account surpluses, and more flexible exchange rates tend to be associated with weaker performance.

### VIII. Mediating channels explaining the role of GDP per capita
- Interactions examined: interactions between development measures (level of GDP per capita; dummy for above-sample-average GDP per capita; dummy for advanced economies per IMF definition) and deaths per capita, containment stringency, and policy response variables.
- WALS interaction results highlight:
  - Output costs of containment and deaths are larger in lower-income countries—consistent with more limited social safety nets and larger shares of financially constrained households/firms.
  - Monetary stimulus (particularly liquidity provision) has been less effective in poorer countries, consistent with limited transmission in emerging and developing economies.
  - Some evidence that fiscal stimulus effectiveness is lower in poorer countries.

### IX. Policy implications and conclusions
- Main conclusions:
  - Countries experiencing smaller output losses in the acute phase are associated with: higher GDP per capita; less stringent containment; fewer deaths per capita; smaller tourism sectors; less flexible credit markets; lower pre-crisis growth; higher fiscal stimulus; less social fractionalization; and less democratic regimes.
  - GDP per capita is the quantitatively largest robust factor: a 75th vs 25th percentile difference corresponds to a 7-percentage-point difference in output surprise.
  - Death rates and containment stringency have similar quantitative effects (~4½-5 percentage points across inter-quartile changes), implying that “saving lives” and “saving the economy” are closely intertwined.
- Policy messages emphasized:
  - Rollback of containment should minimize health risks: relax containment only when new infections are declining and pair with strong testing and contact tracing.
  - Fiscal stimulus has helped reduce economic losses; premature withdrawal is self-defeating.
  - Improve monetary policy transmission in emerging and developing economies, as monetary stimulus reinforced resilience more in advanced economies.
  - Targeted support is essential given severe impacts in high-contact sectors (tourism, retail).
- Comparative literature note:
  - Unlike studies on the GFC, trade and financial openness are not found to be important drivers of output loss during the pandemic in the baseline analysis.
- Suggested directions for future research:
  - Whether trade and financial openness will matter more during recovery phases remains an important question.

*Source: ANNEX B (excerpt) from the provided IMF working paper PDF content.*

### References

### wpiea2021018-print-pdf - References

### Major definitions and notes
- Output performance is defined as "the difference between the observed cumulative real GDP growth in 2020H1 and the cumulative growth that was expected before the onset of the pandemic for the same period—based on the IMF World Economic Outlook 2020 January forecast for 2020H1."
- Output performance 1: difference between observed 2020H1 and IMF January 2020 forecast for 2020H1.
- Output performance 2: difference in cumulative real GDP growth between the first half of 2020 (2020H1) and the first half of 2019 (2019H1).
- Tables report OLS and WALS estimates, t-statistics in parentheses; robustness is indicated by regressors with |t-value| > 1. BMA results report posterior-inclusion-probabilities.

### Robust drivers of cross-country output performance (summary of main results)
- From TABLE 7 (WALS, N 60): magnitude (t-statistic reported in table; bold indicates robust):
  - Health condition: 0.41 (OP1), 0.27 (OP2)
  - Containment stringency: -2.23 (OP1), -2.38 (OP2)
  - Deaths per capita (log): -2.44 (OP1), -2.65 (OP2)
  - Liquidity: -0.81 (OP1), -0.84 (OP2)
  - Policy rate cut: 0.37 (OP1), 0.48 (OP2)
  - Fiscal stimulus: 0.85 (OP1), 1.20 (OP2)
  - Credit market regulation: -1.59 (OP1), -1.63 (OP2)
  - Financial globalization: 0.75 (OP1), 0.66 (OP2)
  - Government debt (% of GDP): -0.92 (OP1), -0.95 (OP2)
  - Average GDP growth, 17-19: -1.05 (OP1), -1.71 (OP2)
  - Tourism (% of GDP): -3.01 (OP1), -2.75 (OP2)
  - Selected other estimates included for Service, Industry, Democracy, GDP per capita (log), and institutional and macro variables with their reported WALS coefficients.

- Robustness checks controlling for outliers (TABLE 8, N 60) show broadly similar patterns:
  - Tourism (% of GDP): -3.24 (OP1), -2.73 (OP2)
  - Deaths per capita (log): -2.45 (OP1), -2.74 (OP2)
  - Average GDP growth, 17-19: -1.24 (OP1), -1.90 (OP2)
  - Government debt (% of GDP): -0.97 (OP1), -1.10 (OP2)
  - Credit market regulation: -1.72 (OP1), -1.87 (OP2)

- BMA posterior-inclusion-probabilities (TABLE 9, N 60) — variables with highest inclusion probabilities:
  - Tourism (% of GDP): 0.99 (OP1), 0.85 (OP2)
  - Containment stringency: 0.97 (OP1), 0.88 (OP2)
  - Deaths per capita (log): 0.56 (OP1), 0.54 (OP2)
  - Government debt (% of GDP): 0.17 (OP1), 0.17 (OP2)
  - Democracy: 0.14 (OP1), 0.08 (OP2)
  - GDP per capita (log): 0.11 (OP1), 0.1 (OP2)
  - Fiscal stimulus: 0.08 (OP1), 0.08 (OP2)
  - Credit market regulation: 0.07 (OP1), 0.11 (OP2)
  - Average GDP growth, 17-19: 0.07 (OP1), 0.32 (OP2)

### Public health and containment (TABLE 1 & Figure notes)
- Deaths per capita (log):
  - OLS: -0.434 (OP1), -0.655 (OP2)
  - WALS: -0.380 (OP1), -0.496 (OP2)
  - N 85
- Containment stringency:
  - OLS: -11.110** (OP1), -10.041 (OP2)
  - WALS: -8.428* (OP1), -7.460 (OP2)
  - t-statistics reported indicate containment stringency is statistically significant in some specifications (* p<0.05, ** p<0.01).
- Health condition:
  - OLS: -1.011 (OP1), -1.080 (OP2)
  - WALS: -0.602 (OP1), -0.639 (OP2)

### Sectoral composition (TABLE 2 and Figure 3)
- Service (% of GDP):
  - OLS: -0.126 (OP1), -0.168 (OP2)
  - WALS: -0.111 (OP1), -0.148 (OP2)
- Industry (% of GDP):
  - OLS: 0.052 (OP1), 0.074 (OP2)
  - WALS: 0.064 (OP1), 0.091 (OP2)
- Tourism (% of GDP) — consistently robust and negative:
  - OLS: -0.889** (OP1), -0.825** (OP2)
  - WALS: -0.798*** (OP1), -0.723*** (OP2)
  - N 96

### Fiscal and monetary response (TABLE 3 and Figure 4)
- Fiscal stimulus (% of GDP):
  - OLS: -0.079 (OP1), -0.094 (OP2)
  - WALS: -0.047 (OP1), -0.057 (OP2)
- Liquidity:
  - OLS: 0.071 (OP1), 0.034 (OP2)
  - WALS: 0.042 (OP1), 0.020 (OP2)
- Policy rate cut:
  - OLS: 0.002 (OP1), 0.008 (OP2)
  - WALS: 0.001 (OP1), 0.005 (OP2)
- N 96; reported coefficients show limited statistical significance in these baseline specifications.

### Regulation (TABLE 4 and Figure 5)
- Credit market regulation:
  - OLS: -1.207 (OP1), -1.840 (OP2)
  - WALS: -0.798 (OP1), -1.200 (OP2)
- Labor market regulation:
  - OLS: -0.057 (OP1), -0.307 (OP2)
  - WALS: -0.025 (OP1), -0.162 (OP2)
- Business regulation:
  - OLS: 1.297 (OP1), 0.632 (OP2)
  - WALS: 0.820 (OP1), 0.398 (OP2)
- N 94.

### Macroeconomic characteristics (TABLE 5 and Figure 6)
- Average GDP growth (17-19):
  - OLS: -1.055* (OP1), -1.740** (OP2)
  - WALS: -0.677 (OP1), -1.124** (OP2)
  - N 70; indicates countries with stronger pre-pandemic growth experienced larger output shortfalls in some specifications.
- Government debt (% of GDP):
  - OLS: -0.0362 (OP1), -0.0384 (OP2)
  - WALS: -0.0232 (OP1), -0.0248 (OP2)
- Financial system deposit (% GDP):
  - OLS: -0.0235 (OP1), -0.0154 (OP2)
  - WALS: -0.0164 (OP1), -0.0104 (OP2)
- Other macro variables (financial globalization, trade globalization, domestic credit, bank concentration, exchange rate regime) reported with coefficients and t-statistics; select robust results shown in overarching WALS tables.

### Development, demographic and institutions (TABLE 6 and Figure 7)
- Gini coefficient:
  - OLS: -41.40** (OP1), -28.36 (OP2)
  - WALS: -29.44** (OP1), -18.08 (OP2)
  - N 85; Gini shows statistically significant negative coefficient in OP1 (OLS and WALS).
- Social fractionalization:
  - OLS: 10.45** (OP1), 6.586 (OP2)
  - WALS: 7.895* (OP1), 4.540 (OP2)
- Share of population over 65, Population (log), GDP per capita (log), Democracy, Population density, Informality, Remittance to GDP (%) — coefficients reported; notable that some inequality and social fractionalization measures appear in WALS robustness results.

### Additional specifications and sensitivity analyses (selected highlights)
- TABLE 10 (Additional covariates, N 48) shows notable coefficient magnitudes for:
  - Fiscal stimulus: 2.18 (OP1), 1.79 (OP2)
  - Government debt (% of GDP): -1.96 (OP1), -2.10 (OP2)
  - Average GDP growth, 17-19: -1.92 (OP1), -2.49 (OP2)
  - GDP per capita (log): 1.49 (OP1), 1.64 (OP2)
  - Rule of law: -2.01 (OP1), -1.82 (OP2)
- TABLE 11 (Using only Q2 data, N 60) retains:
  - Deaths per capita (log): -2.89 (OP1), -3.20 (OP2)
  - Credit market regulation: -1.98 (OP1), -2.04 (OP2)
  - Tourism (% of GDP): -2.69 (OP1), -2.48 (OP2)

- TABLE 12 (Interaction with income level, N 60) presents interaction coefficients showing how effects vary with income level (Continuous, Dummy 1, Dummy 2 specifications); examples:
  - Deaths per capita (log): -1.78 (OP1, Continuous), -2.54 (OP2, Continuous)
  - Tourism (% of GDP): -3.21 (OP1) across specifications
  - Containment stringency: -1.66 (OP1, Continuous), -1.22 (OP2, Continuous)

### Visual and regression diagnostic notes from figures
- Figures report fitted lines and R² values for scatterplots relating output performance to various covariates. Example reported fit statistics:
  - Death per capita (log): y = -0.4798x -21.468, R² = 0.015
  - Containment stringency: y = -10.84x -10.382, R² = 0.089
  - Tourism (% of GDP): y = 0.5493x -21.922, R² = 0.0087
  - GDP growth, 17-19: y = -0.8301x -13.475, R² = 0.0361
  - Gini coefficient (Figure 7): y = -21.487x -8.858, R² = 0.0556
- Figures uniformly note the definition of output performance tied to IMF WEO January 2020 forecast for 2020H1.

*Italic: Content derived from wpiea2021018-print-pdf - References (source PDF content provided).*

### ANNEX A

### ANNEX A

### Data sources and descriptive statistics (Table A1)
- Measure of output performance:
  - Output performance 1 — Source: IMF — Obs: 97 — Mean: -17.60 — Std. dev: 12.44
  - Output performance 2 — Source: IMF — Obs: 97 — Mean: -16.67 — Std. dev: 13.15
- Health factors:
  - Health condition — Source: JHU — Obs: 191 — Mean: 40.58 — Std. dev: 14.41
  - Containment stringency — Source: OxCGRT — Obs: 183 — Mean: 0.59 — Std. dev: 0.20
  - Deaths per capita (log) — Source: JHU — Obs: 175 — Mean: -11.02 — Std. dev: 1.95
- Policy support:
  - Liquidity — Source: IMF — Obs: 194 — Mean: 1.25 — Std. dev: 3.50
  - Policy rate cut — Source: IMF — Obs: 194 — Mean: 87.29 — Std. dev: 134.04
  - Fiscal stimulus — Source: IMF — Obs: 194 — Mean: 4.26 — Std. dev: 6.09
- Regulation:
  - Labor market regulation — Source: FI — Obs: 157 — Mean: 6.46 — Std. dev: 1.36
  - Credit market regulation — Source: FI — Obs: 157 — Mean: 8.18 — Std. dev: 1.55
  - Business regulation — Source: FI — Obs: 157 — Mean: 6.75 — Std. dev: 1.27
- Macroeconomic factors:
  - Financial globalization — Source: KOF — Obs: 180 — Mean: 63.19 — Std. dev: 19.67
  - Trade globalization — Source: KOF — Obs: 183 — Mean: 56.41 — Std. dev: 20.37
  - Current account (% of GDP) — Source: IMF — Obs: 132 — Mean: -1.72 — Std. dev: 8.36
  - Financial system deposit (% of GDP) — Source: FSD — Obs: 163 — Mean: 59.93 — Std. dev: 50.73
  - Government debt (% of GDP) — Source: FSD — Obs: 115 — Mean: 56.34 — Std. dev: 37.13
  - Domestic credit (% of GDP) — Source: FSD — Obs: 165 — Mean: 57.31 — Std. dev: 43.31
  - Bank concentration — Source: FSD — Obs: 160 — Mean: 65.69 — Std. dev: 19.30
  - Exchange rate regime — Source: IMF — Obs: 192 — Mean: 2.07 — Std. dev: 0.87
  - Average GDP growth, 17-19 — Source: IMF — Obs: 199 — Mean: 2.97 — Std. dev: 3.60
- Sectoral composition:
  - Tourism (% of GDP) — Source: WTTC — Obs: 174 — Mean: 13.96 — Std. dev: 13.07
  - Service (% of GDP) — Source: WDI — Obs: 193 — Mean: 56.60 — Std. dev: 13.36
  - Industry (% of GDP) — Source: WDI — Obs: 202 — Mean: 25.43 — Std. dev: 12.45
- Development and other controls:
  - Share of population over 65 — Source: WDI — Obs: 190 — Mean: 8.34 — Std. dev: 5.88
  - Population (log) — Source: WDI — Obs: 211 — Mean: 15.29 — Std. dev: 2.41
  - GDP per capita (log) — Source: WDI — Obs: 208 — Mean: 8.83 — Std. dev: 1.50
  - Democratization — Source: Polity IV — Obs: 152 — Mean: 3.87 — Std. dev: 3.92
  - Population density — Source: WDI — Obs: 209 — Mean: 454.25 — Std. dev: 2085
  - Gini coefficient — Source: SWIID 7.1 — Obs: 165 — Mean: 0.39 — Std. dev: 0.08
  - Social fractionalization — Source: Alesina et al. (2003) — Obs: 179 — Mean: 0.44 — Std. dev: 0.19
  - Informality — Source: WDI — Obs: 143 — Mean: 29.30 — Std. dev: 14.25
  - Remittance inflow to GDP (%) — Source: FSD — Obs: 180 — Mean: 4.68 — Std. dev: 6.61
- Notes:
  - Output performance 1 is the difference between the observed cumulative real GDP growth in 2020H1 and the cumulative growth that was expected before the onset of the pandemic for the same period—based on the IMF World Economic Outlook 2020 January forecast for 2020H1.
  - Output performance 2 is the difference in cumulative real GDP growth between the first half of 2020 (2020H1) and the first half of 2019 (2019H1).
  - Data source abbreviations: IMF; JHU (Johns Hopkins University Coronavirus Resource Center); OxCGRT (Oxford COVID-19 Government Response Tracker); FI (Fraser Institute Economic Freedom Network); KOF (Swiss Economic Institute); FSD (World Bank Financial Structure Database); WDI (World Development Indicators); WTTC (World Tourist & Tourism Council).

### Bivariate relationships from Figures A1–A6 (selected regression lines and R²)
- Output performance vs. Deaths per capita (log):
  - y = -0.771x -23.948 — R² = 0.0326
- Output performance vs. Containment stringency:
  - y = -8.9567x -10.721 — R² = 0.0518
- Output performance vs. Health condition:
  - y = -0.0529x -13.281 — R² = 0.0066
- Output performance vs. Service (% of GDP):
  - y = -0.2249x -2.8807 — R² = 0.0661
- Output performance vs. Industry (% of GDP):
  - y = 0.3072x -24.09 — R² = 0.0987
- Output performance vs. Tourism (% of GDP):
  - y = -0.4145x -11.102 — R² = 0.1671
- Output performance vs. Credit market regulation:
  - y = -1.4515x -3.0507 — R² = 0.0393
- Output performance vs. Liquidity:
  - y = -0.1037x -15.971 — R² = 0.0008
- Output performance vs. Policy rate cuts:
  - y = 0.0157x -17.477 — R² = 0.0315
- Output performance vs. Fiscal stimulus (% of GDP):
  - y = -0.184x -15.151 — R² = 0.0077
- Output performance vs. Average GDP growth, 17-19:
  - y = -0.8301x -13.475 — R² = 0.0361
- Output performance vs. Financial system deposit:
  - y = -0.0349x -14.171 — R² = 0.0191
- Output performance vs. Domestic credit (% of GDP):
  - y = -0.0034x -16.167 — R² = 0.0004
- Output performance vs. Trade globalization:
  - y = -0.0258x -14.703 — R² = 0.0046
- Output performance vs. Government debt (% of GDP):
  - y = -0.0624x -12.635 — R² = 0.0559
- Output performance vs. Bank concentration:
  - y = -0.0669x -12.156 — R² = 0.0248
- Output performance vs. Financial globalization:
  - y = -0.063x -11.756 — R² = 0.0192
- Output performance vs. Current account (% of GDP):
  - y = 0.2086x -16.262 — R² = 0.0219
- Output performance vs. Exchange rate regime:
  - y = -1.3178x -12.692 — R² = 0.0175
- Output performance vs. Remittance inflow to GDP (%):
  - y = -0.2354x -15.626 — R² = 0.019
- Output performance vs. Share of population over 65:
  - y = -0.1698x -14.298 — R² = 0.017
- Output performance vs. Population (log):
  - y = 0.221x -19.702 — R² = 0.0019
- Output performance vs. Population density:
  - y = -0.0049x -15.262 — R² = 0.0277
- Output performance vs. Democracy:
  - y = -0.227x -14.878 — R² = 0.0137
- Output performance vs. GDP per capita (log):
  - y = 0.0523x -16.589 — R² = 7E-05
- Output performance vs. Gini coefficient:
  - y = -7.1109x -13.52 — R² = 0.0051
- Output performance vs. Social fractionalization:
  - y = 5.7139x -18.209 — R² = 0.0184
- Output performance vs. Informality:
  - y = -0.0801x -13.704 — R² = 0.0223

### Robust drivers of output performance across countries (Figures A7–A12: qualitative magnitudes)
- Variables consistently associated with worse output performance (negative effect) when moving from the 25th to the 75th percentile include:
  - Average GDP growth, 17-19
  - Credit market regulation
  - Tourism (% of GDP)
  - Containment stringency
  - Deaths per capita (log)
  - Government debt (% of GDP)
  - Social fractionalization (in some specifications)
  - Current account (% of GDP) (in extended covariate models)
  - Informality (in some specifications)
- Variables associated with better output performance (positive effect) when moving from the 25th to the 75th percentile include:
  - Fiscal stimulus (positive effect shown in multiple panels)
  - GDP per capita (log)
  - Financial globalization (in selected extended covariate models)
  - Remittance inflow to GDP (%) (positive in some interaction-with-income specifications)
- Interaction with income level findings:
  - Effects of liquidity, remittance inflow, democracy, credit market regulation, containment stringency, deaths per capita (log), tourism (% of GDP), and GDP per capita (log) vary with continuous and dummy definitions of income level (see Panels in Figures A10–A12).
  - Dummy 1: 1 denotes above-average GDP per capita; Dummy 2: 1 denotes advanced economies per World Economic Outlook classification.
- Magnitude charts report differential effects moving variables from the 25th to the 75th percentile based on coefficients robust in column (I-II) of Table 7; “–” denotes a negative effect on output and “+” denotes a positive effect.

### ANNEX B — Model Averaging and WALS methodology
- Objective:
  - Address model uncertainty by (i) running the maximum combination of possible models and (ii) providing estimates and inference that account for variable performance over the whole set of possible specifications.
- Unconditional (model-averaged) estimate formula:
  - β̂_x = ∑_{i=1}^M ω_i β̂_{i x}  (Equation B1)
  - where ω_i denote a measure of goodness of fit of each model.
- Moving averaging technique used:
  - Weighted Average Least Squares (WALS) (Magnus, Powell, and Prüfer (2010)), generalized by De Luca and Magnus (2011) to distinguish focus and auxiliary regressors.
  - Focus regressors: forced to enter every model (in this paper only the constant is a focus regressor).
  - Auxiliary regressors: tested for inclusion across models.
- Linear regression framework:
  - y = X1 β1 + X2 β2 + ε  (Equation B2)
  - y: vector of observations on output performance
  - X1: observations on k1 focus regressors
  - X2: observations on k2 auxiliary regressors
  - Conditional sample likelihood: p(y | β1, β2, σ2, M_i)
- Prior choice:
  - WALS uses a Laplace distribution prior for auxiliary parameters (vs. Gaussian in BMA), reducing risk of prior dominance.
- Orthogonal transformation used:
  - Compute orthogonal k2×k2 matrix P and diagonal k2×k2 matrix ∆ such that P' X2' M1 X2 P = ∆, increasing model-selection space linearly rather than exponentially.
- WALS estimators:
  - β̂1 = (X1' X1)^{-1} X1' (y − X2 β̂2)  (Equation B3)
  - β̂2 = s P ∆^{-1/2} t̅  (Equation B4)
  - where t̅ is the Laplace estimator of the vector of theoretical t-ratios of auxiliary regressors.
- Robustness rule of thumb:
  - Magnus, Powell, and Prüfer (2010) suggest an absolute value of the t-ratio greater than 1 for an auxiliary regressor to qualify as robust. This choice is motivated by the observation that including an auxiliary regressor increases model fit (adjusted R²) and precision of focus-regressor estimators (lower MSE) if and only if the t-ratio of the additional auxiliary regressor is in absolute value greater than 1.

*Source: ANNEX A and ANNEX B of the provided IMF content unit.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021018-print-pdf.pdf_