## wpiea2021203-print-pdf - 8.4 percent decline  associated with the war.

## Source details

**Canonical URL:** [wpiea2021203-print-pdf - 8.4 percent decline  associated with the war.](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021203-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021203-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021203-print-pdf.pdf.json)

---

### Main contributions and scope
- Comprehensive analysis of 586 recession episodes from 115 countries over 1957-2019.
- Decomposes persistent output losses using supply-side components of GDP: capital stock (per person engaged), number of persons engaged (employment-population ratio), and total factor productivity (TFP).
- Differentiates recessions by:
  - Type: typical recessions, financial crises, past modern pandemics/epidemics (Hong Kong flu, SARS, H1N1, MERS, Ebola, Zika), natural disasters, and violent conflicts.
  - Depth: loss in real GDP per capita between peak and trough (measured as loss between peak and first year of recession).
  - Duration: number of years between peak and trough.
- Uses a unified regression framework with interaction terms to account for co-occurrence of crisis types.
- Provides analysis of expected medium-term output losses from COVID-19 by comparing IMF WEO forecast vintages (January 2020 vs April 2021) for the level of real GDP forecast for 2024.

### Data sources and definitions
- Primary macroeconomic data: Penn World Table (PWT) 10.0 for real GDP per capita (constant 2017 US dollars), capital stock, employment-population ratio, and total factor productivity.
- Recession dating: Harding and Pagan (2002) algorithm applied to annual real GDP per capita with a window of 1 year, minimum phase length of 1 year, and minimum cycle length of 2 years.
- Financial crisis incidence: Laeven and Valencia (2018) for 1970 onwards; Reinhart and others (2016) for pre-1970. Financial crises include banking crises, currency crises, and sovereign debt crises.
- Past modern epidemics/pandemics: identified for countries with reported cases (Furceri and others 2020; Cockburn, Delon, Ferreira 1969).
- Disasters: EM-DAT, defined as disasters that led to damages exceeding 1% of GDP or affected 5% of population (including deaths).
- Conflict: defined as years with battle-related deaths exceeding 100 people per one million population (Novta and Pugacheva 2021).
- Forecast comparison for medium-term output losses: IMF WEO forecasts, measuring medium-term loss as the difference between the level of real GDP forecast for 2024 in January 2020 vintage and the April 2021 vintage.
- Additional variables in forecast-revision analysis: tourism share of GDP (World Travel and Tourism Council), service sector share of GDP (World Bank WDI), fiscal policy responses (IMF Fiscal Monitor Database of Country Fiscal Measures in Response to the COVID-19 Pandemic).

### Methodology
- Local projections (Jordà 2005) estimate cumulative impulse response functions for horizons h = 0,...,7 (years after recession onset).
- Baseline regression specification:
  - Dependent variable: cumulative growth in log points of real GDP per capita (or other dependent variables) between t-1 and t+h.
  - Key regressors: dummy for recession onset (D_i,t), dummies for crisis event types (E_i,t) and interactions D_i,t * E_i,t-2,t+2 to capture crises occurring within t-2 to t+2 of a recession.
  - Controls: two lags of dependent variable’s growth rate, one lag of log GDP in constant US dollars, and two lags of credit-to-GDP ratio.
  - Fixed effects: country and year fixed effects.
  - Left-hand-side variables winsorized at 0.5/99.5 percentiles.
- Impact interpretation:
  - Impact of a typical recession = β1_h.
  - Impact of a recession associated with crisis E = β1_h + β2_{E,h} + β3_{E,h}.

### Key empirical findings — aggregate impacts and channels
- Recessions are associated with negative and mostly statistically significant cumulative impacts across horizons, indicating average persistent output losses.
- Recessions associated with financial crises lead to more negative outcomes than typical recessions.
- The path of output after past modern epidemic or pandemic recessions lies between that of typical recessions and financial-crisis recessions.
- Natural disasters and violent conflict produce negative and severe immediate impacts with effects persisting for several years; statistical significance weakens at later horizons.
- Supply-side channel results:
  - Real GDP per Capita: cumulative negative impacts across horizons for typical, past pandemic/epidemic, and financial crisis recessions.
  - Total Factor Productivity: recessions impair TFP with negative cumulative coefficients across horizons.
  - Capital Stock per Worker: capital per worker declines following recessions; cumulative effects are negative and statistically significant across many horizons.
  - Employment-Population Ratio: declines following recessions with negative cumulative effects.
- Sample and estimation details:
  - Number of recession observations: 586.
  - Number of countries: 115.
  - Standard errors clustered at the country level.
  - Reported example R^2 values in panel summaries: 0.57 and 0.22 (with further R^2 reported for h = 0 and h = 7 depending on dependent variable).

### Depth and duration heterogeneity
- Recession duration in sample: range 1 to 10 years.
  - 60 percent of recessions last one year.
  - 90 percent of recessions last not more than three years for both AEs and EMDEs.
- Depth metric: median recession is associated with a 2.2% decline in per-capita output in the first year. Recessions above the median are classified as high depth; below the median as low depth.
- Four recession types by depth and duration used in interaction analysis:
  1. High depth, one-year duration (short).
  2. Low depth, one-year duration (short).
  3. High depth, duration > one year (long).
  4. Low depth, duration > one year (long).
- Heterogeneous outcomes:
  - Advanced economies: deep but short-lived recessions are associated with 'V-shaped' recoveries and no permanent output loss after several years.
  - Emerging market and developing economies: deep recessions—especially when longer duration—are more likely to lead to persistent output losses.

### EMDEs — medium-term losses, channels, and employment
- High depth and long duration recessions tend to lead to the most adverse effect seven years after recession onset.
- Low depth and long duration recessions are negative up to seven years following the shock.
- Low depth and short duration recessions have a statistically significant negative effect immediately, which starts to lose significance and move closer to zero after five years, with faster recovery in EMDEs.
- For EMDEs, historical medium-term losses in GDP per capita for typical recessions are primarily attributable to losses in TFP.
- Employment per capita declines before recovering somewhat in the medium term.
- For financial crisis recessions in EMDEs there is significant, persistent damage to:
  - TFP (column 8)
  - Capital-to-worker ratio (column 13)
  - Employment per capita (column 18)
- In modern era epidemics and pandemics:
  - Productivity losses were the main contributor to output losses in AEs.
  - Capital stock losses played the largest role in EMDEs.

### Short-duration recessions — representative coefficients and model fit
- Representative regression coefficients for short-duration, typical recessions (exact reported values by horizon h):
  - h = 0: -5.933***, -2.813***, -4.683***, -2.170***, 0.754*, 0.474**, -1.175***, -0.666***
  - h = 1: -4.212***, -2.431***, -2.253***, -1.306***, 1.057*, 1.056***, -1.835***, -1.483***
  - h = 2: -3.941***, -1.978***, -1.909*, -0.931, 0.668, 1.099**, -1.694***, -1.571***
  - h = 3: -4.994***, -1.778**, -2.705**, -0.993, 0.185, 0.993*, -1.627***, -1.452***
  - h = 4: -3.655**, -1.602*, -1.338, -1.227, -0.463, 1.019, -1.418**, -1.259**
  - h = 5: -5.282***, -0.852, -3.065, -0.773, -1.295, 0.818, -0.934, -1.027*
  - h = 6: -6.111***, -0.451, -4.161*, -0.574, -1.773, 0.772, -0.574, -0.847
  - h = 7: -5.064**, 0.280, -3.253, 0.020, -2.254, 1.003, -0.117, -0.744
- Model fit and sample indicators:
  - Number of Observations (per dependent variable column): 586
  - Number of Countries: 115
  - R2 (for h = 0) by dependent variable columns: 0.59, 0.53, 0.65, 0.37
  - R2 (for h = 7) by dependent variable columns: 0.40, 0.34, 0.57, 0.22

### COVID-19: unique features and forecast revisions
- COVID-19 could produce larger scarring than past recessions because:
  - Some high-contact sectors may shrink permanently.
  - Widespread school closures with disproportionately adverse impacts on schooling in low-income countries and those less prepared to switch to virtual learning.
  - Potentially larger productivity-decreasing resource mismatches across sectors and occupations.
  - Decline in competition if market power of large companies increases due to small business closures.
  - Offsetting forces: increased digitalization and innovation may offset adverse productivity shocks in some countries; others lack prerequisite widespread and reliable connectivity.
- Evidence cited:
  - Bloom and others (2020) estimate private sector TFP to be 5 percent lower at the end of 2020 than it would have been, and likely to remain 1 percent lower in the medium term.
  - New business formation in the United States reached a record high in Q3 2020 (Brown 2020) — as reported in the source text.
- IMF April 2021 WEO forecasts envision output losses, relative to pre-pandemic projections, of about 3 percent for the world economy by 2024.
- By comparison, lasting damages over a comparable period from the global financial crisis were almost 8.7 percent for the world as a whole.
- Losses are expected to be much lower in advanced economies than in emerging market and developing economies, owing to differences in scale of policy support and access to vaccines and therapies.

### Cross-country regression correlates of forecast revisions
- Cross-sectional regression setup: ΔŶit = α + βXi + γΓi + εi, for years t = 2022–2024.
- Regressors of interest Xi include:
  - (i) income group (advanced, emerging, low-income)
  - (ii) share of GDP from tourism and transportation in 2019
  - (iii) share of GDP from services in 2019
  - (iv) fiscal support during COVID-19 crisis up to December 2020
- Key quantitative correlates reported:
  - A one-standard deviation increase in tourism and travel share of GDP is associated with a 2.5 percent reduction in expected output in 2022; exposure through tourism fades but remains close to 2 percent in 2024.
  - Economies with larger service sectors are likely to experience larger output losses, with a ½ percent reduction in expected output in 2022.
  - Countries with larger pandemic-related above-the-line fiscal measures are projected to experience smaller losses, all else equal.
- Revisions between forecast vintages:
  - Comparing October 2020 WEO and April 2021 WEO, favorable news on vaccines and a stronger-than-expected second half of 2020 improved projections more in advanced economies than in other income groups.

### Policy recommendations to limit scarring and support recovery
- Maintain measures that support credit provision while ensuring balance sheet resilience and adequate buffers to prevent scarring from future financial instability.
- Maximize use of limited fiscal space by tailoring responses and targeting support to the most-affected sectors and firms.
- Implement policies that reverse setbacks to human capital accumulation, boost job creation, and facilitate worker reallocation.
- Promote competition, innovation, and technology adoption to lift productivity growth and boost investment.
- Multilateral cooperation on vaccines to ensure adequate production and timely universal distribution is crucial to prevent worse scarring in developing economies.

*Source: Authors' calculations and analysis in wpiea2021203-print-pdf - 8.4 percent decline  associated with the war.*

### 8.4 percent decline  associated with the war.

### wpiea2021203-print-pdf - 8.4 percent decline  associated with the war.

### Main contributions and scope
- Comprehensive analysis of 586 recession episodes from 115 countries over 1957-2019.
- Studies channels of persistent damage by analyzing supply-side components of GDP: capital stock (per person engaged), number of persons engaged (employment-population ratio), and total factor productivity (TFP).
- Differentiates recessions by:
  - Type: typical recessions, financial crises, past modern pandemics/epidemics (Hong Kong flu, SARS, H1N1, MERS, Ebola, Zika), natural disasters, and violent conflicts.
  - Depth (loss in real GDP per capita between peak and trough, measured as loss between peak and first year of recession).
  - Duration (number of years between peak and trough).
- Uses a unified regression framework with interaction terms to account for co-occurrence of crisis types.
- Provides analysis of expected medium-term output losses from COVID-19 by comparing IMF World Economic Outlook (WEO) forecast vintages (January 2020 vs April 2021) for the level of real GDP forecast for 2024.

### Data sources and definitions
- Primary macroeconomic data: Penn World Table (PWT) 10.0 for real GDP per capita (constant 2017 US dollars), capital stock, employment-population ratio, and total factor productivity.
- Recession dating: Harding and Pagan (2002) algorithm applied to annual real GDP per capita with a window of 1 year, minimum phase length of 1 year, and minimum cycle length of 2 years. Recessions so identified for the United States match NBER dates.
- Financial crisis incidence: Laeven and Valencia (2018) for 1970 onwards; Reinhart and others (2016) for pre-1970. Financial crises include banking crises, currency crises, and sovereign debt crises.
- Past modern epidemics/pandemics: identified for countries with reported cases (Furceri and others 2020; Cockburn, Delon, Ferreira 1969).
- Disasters: EM-DAT, defined as disasters that led to damages exceeding 1% of GDP or affected 5% of population (including deaths).
- Conflict: defined as years with battle-related deaths exceeding 100 people per one million population (Novta and Pugacheva 2021).
- Forecast comparison for medium-term output losses: IMF WEO forecasts, measuring medium-term loss as the difference between the level of real GDP forecast for 2024 in January 2020 vintage and the April 2021 vintage.
- Additional variables used in forecast-revision analysis: tourism share of GDP (World Travel and Tourism Council), service sector share of GDP (World Bank WDI), fiscal policy responses (IMF Fiscal Monitor Database of Country Fiscal Measures in Response to the COVID-19 Pandemic).

### Methodology
- Local projections (Jordà 2005) estimate cumulative impulse response functions for horizons h = 0,...,7 (years after recession onset).
- Baseline regression specification (summarized):
  - Dependent variable: cumulative growth in log points of real GDP per capita (or other dependent variables) between t-1 and t+h.
  - Key regressors: dummy for recession onset (D_i,t), dummies for crisis event types (E_i,t) and interactions D_i,t * E_i,t-2,t+2 to capture crises occurring within t-2 to t+2 of a recession.
  - Controls: two lags of dependent variable’s growth rate, one lag of log GDP in constant US dollars, and two lags of credit-to-GDP ratio.
  - Fixed effects: country and year fixed effects.
  - Left-hand-side variables winsorized at 0.5/99.5 percentiles.
- Impact interpretation:
  - Impact of a typical recession = β1_h.
  - Impact of a recession associated with crisis E = β1_h + β2_{E,h} + β3_{E,h}.

### Key empirical findings — aggregate impact and channels
- Recessions are associated with negative and mostly statistically significant cumulative impacts across horizons, indicating average persistent output losses.
- Recessions associated with financial crises lead to more negative outcomes than typical recessions.
- The path of output after past modern epidemic or pandemic recessions lies between that of typical recessions and financial-crisis recessions.
- Natural disasters and violent conflict also produce negative and severe immediate impacts with effects persisting for several years; statistical significance weakens at later horizons, potentially due to post-disaster reconstruction effects and data/sample limitations for fragile states.
- Supply-side channels (from Table 1 and Figures):
  - Real GDP per Capita: cumulative negative impacts across horizons for typical, past pandemic/epidemic, and financial crisis recessions (coefficients across h = 0,...,7 show persistent negative values; results mostly significant at conventional levels).
  - Total Factor Productivity: recessions impair TFP with negative cumulative coefficients across horizons.
  - Capital Stock per Worker: capital per worker declines following recessions; cumulative effects are negative and statistically significant across many horizons.
  - Employment-Population Ratio: employment-population ratio declines following recessions with negative cumulative effects.
- Sample and estimation details:
  - Number of recession observations: 586.
  - Number of countries: 115.
  - Regressions estimated separately for each horizon; same fixed sample across horizons.
  - Standard errors clustered at the country level.
  - Reported R^2 (for main regressions) examples: 0.57 and 0.22 in panel summaries (table columns), with further R^2 reported for h = 0 and h = 7 depending on dependent variable.

### Depth and duration heterogeneity
- Recession duration in sample: range 1 to 10 years.
  - 60 percent of recessions last one year.
  - 90 percent of recessions last not more than three years for both AEs and EMDEs.
- Depth metric: median recession is associated with a 2.2% decline in per-capita output in the first year. Recessions above the median are classified as high depth; below the median as low depth.
- Four recession types by depth and duration used in interaction analysis:
  1. High depth, one-year duration (short).
  2. Low depth, one-year duration (short).
  3. High depth, duration > one year (long).
  4. Low depth, duration > one year (long).
- Key heterogeneous outcomes:
  - Advanced economies: deep but short-lived recessions are associated with 'V-shaped' recoveries and no permanent output loss after several years.
  - Emerging market and developing economies: deep recessions—especially when longer duration—are more likely to lead to persistent output losses (less evidence of full recovery).

### Measurement of COVID-19 medium-term losses
- Medium-term output loss (or gain) is measured as difference between IMF WEO forecast for level of real GDP in 2024 in the January 2020 vintage and the April 2021 vintage.
- IMF WEO forecasts provide vintages up to five years ahead for 194 countries with regular revisions (around April and October, with interleaved updates around January and July), enabling analysis of forecast revisions at different pandemic stages.

*Source: Authors' calculations and analysis in wpiea2021203-print-pdf - 8.4 percent decline  associated with the war.*

### 3. Emerging Markets

### 3. Emerging Markets

### Medium-term output losses by recession type, depth, and duration
- Recessions of high depth and long duration tend to lead to the most adverse effect seven years after recession onset (columns 3, 7, 11 and yellow line in Figure 2, panel 3).
- Low depth and long duration recessions are likewise negative up to seven years following the shock (columns 4, 8, 12).
- Low depth and short duration recessions have a statistically significant negative effect in the immediate aftermath of the recession, which starts to lose significance and moves closer to zero after five years, with faster recovery in emerging market and developing economies (columns 2, 6, 10).
- Recessions associated with financial crises produce worse outcomes for any given depth and duration (not shown in the table but stated in the text).

Key regression diagnostics and sample sizes (as reported)
- R2 (for h = 0): 0.59, 0.71, 0.61
- R2 (for h = 7): 0.40, 0.62, 0.38
- Aggregate counts appearing in the table footnotes: 4,341; 1,337; 2,999

Note on table estimation (methodology and controls)
- The dependent variable is cumulative growth of real GDP per capita in the horizon year h after a recession. Regressions are estimated separately for each horizon.
- All regressions include interaction terms for recession types (financial crisis, pandemic, disaster, conflict) and recession depth (high and low depth recessions are split based on the median loss, with separate interaction terms for recessions that last only one year and those that last longer than one year).
- Controls: two lags of the dependent variable’s growth rate, one lag of log GDP per capita (in constant US dollars), and two lags of credit-to-GDP ratio; country and year fixed effects.
- Statistical significance notation: * p<0.1; ** p<0.05; *** p<0.01.

### Channels of impact (supply-side focus)
- Labor channel:
  - Unemployment may remain higher even after the recession, potentially resulting in a smaller labor force as discouraged workers exit (Blanchard and Summers 1986).
  - Human capital accumulation and future earnings can be affected by skill deterioration during extended periods of unemployment, delayed labor market entry for young workers, and negative effects on educational achievement in the longer term.
  - Footnotes cite evidence that parental job losses can adversely affect children’s schooling and future labor market outcomes, and that in the short-term reduced labor market opportunities can lead to higher educational attainment for high school and college-aged students.
- Capital channel:
  - Weak investment can produce slower physical capital accumulation and slower technology adoption, hampering productivity growth.
  - Greater scarring through the physical capital channel can arise from capital being stranded and corporate debt buildup constraining future investment (IMF 2021b).
- Productivity and firm dynamics:
  - Permanent productivity losses can result from the loss of firm-specific know-how after bankruptcies and their spillovers (Bernstein and others 2019).
  - Declines in research and development and innovation during the recession, and increased resource misallocation, can further depress productivity (Adler and others 2017; Furceri and others 2021).
- Empirical approach to channels:
  - Focusing on the supply-side channels, the chapter analyzes components of the Cobb-Douglas production function.
  - Stand-alone regressions are run for total factor productivity, capital per worker, and the employment-population ratio to show the impact of recessions on each component.

### Empirical patterns illustrated in figures
- Figure 3 presents medium-term output losses and channels of impact across Advanced Economies (AEs) and Emerging Market and Developing Economies (EMDEs) for:
  - Real GDP per Capita (AEs and EMDEs)
  - Total Factor Productivity (AEs and EMDEs)
  - Capital Stock per Worker (AEs and EMDEs)
  - Employment-Population Ratio (AEs)
- The solid lines in the figures represent the estimated cumulative impulse response functions and shaded areas represent 90 percent confidence intervals. Time since the shock (in years) is on the x-axis.
- Past modern pandemics and epidemics considered include Hong Kong flu, SARS, H1N1, MERS, Ebola, and Zika.

*Source: Authors' calculations.*

### 8. EMDEs: Employment

### 8. EMDEs: Employment

### Channels of Medium-Term Output Losses
- Historical analysis shows medium-term losses in GDP per capita for typical recessions are primarily attributable to losses in TFP (column 6 in the source table).
- Employment per capita declines before recovering somewhat in the medium-term (column 16).
- For financial crisis recessions, there is significant, persistent damage to:
  - TFP (column 8)
  - Capital-to-worker ratio (column 13)
  - Employment per capita (column 18)
- For typical recessions and financial crises, channels of impact are broadly the same across country groups, except:
  - Employment per capita losses play a role in AEs, on average, and not EMDEs.
- In modern era epidemics and pandemics:
  - Productivity losses were the main contributor to output losses in AEs.
  - Capital stock losses played the largest role in EMDEs.

### Short-Duration Recessions: Depth and Channel Breakdown
- Table 4 (short duration, high and low depth) shows the impact is primarily driven by reductions in:
  - Total factor productivity
  - Capital per worker ratio
- Regression specification notes:
  - Dependent variable: cumulative growth of real GDP per capita, total factor productivity, capital per worker, employment-population ratio in the horizon year h after a recession.
  - Controls: interaction terms for recession types and recession depth; two lags of the dependent variable’s growth rate; one lag of log GDP per capita (constant US dollars); two lags of credit-to-GDP ratio; country and year fixed effects.
  - High and low depth recessions are split based on median loss, with separate interaction terms for recessions that last only one year and those that last longer than one year.
- Representative regression coefficients (short duration, typical recession rows shown in source table) include exact reported values such as:
  - h = 0: -5.933***, -2.813***, -4.683***, -2.170***, 0.754*, 0.474**, -1.175***, -0.666***
  - h = 1: -4.212***, -2.431***, -2.253***, -1.306***, 1.057*, 1.056***, -1.835***, -1.483***
  - h = 2: -3.941***, -1.978***, -1.909*, -0.931, 0.668, 1.099**, -1.694***, -1.571***
  - h = 3: -4.994***, -1.778**, -2.705**, -0.993, 0.185, 0.993*, -1.627***, -1.452***
  - h = 4: -3.655**, -1.602*, -1.338, -1.227, -0.463, 1.019, -1.418**, -1.259**
  - h = 5: -5.282***, -0.852, -3.065, -0.773, -1.295, 0.818, -0.934, -1.027*
  - h = 6: -6.111***, -0.451, -4.161*, -0.574, -1.773, 0.772, -0.574, -0.847
  - h = 7: -5.064**, 0.280, -3.253, 0.020, -2.254, 1.003, -0.117, -0.744
- Model fit and sample indicators (as presented):
  - Number of Observations: 586 (per dependent variable column)
  - Number of Countries: 115
  - R2 (for h = 0): 0.59, 0.53, 0.65, 0.37 (columns correspond to Real GDP per Capita, Total Factor Productivity, Capital Stock per Worker, Employment-Population Ratio respectively)
  - R2 (for h = 7): 0.40, 0.34, 0.57, 0.22

### COVID-19: Unique Features and Potential for Larger Scarring
- The impact of COVID-19 could be larger than past recessions because:
  - Some high-contact sectors may shrink permanently.
  - Widespread school closures, with disproportionately adverse impacts on schooling in low-income countries and those less prepared to switch to virtual learning.
  - Productivity-decreasing resource mismatches across sectors and occupations may be larger than in previous crises, depending on permanence of asymmetric losses.
  - Productivity could be negatively affected by a decline in competition if market power of large companies increases due to small business closures in high-contact sectors and more broadly.
  - Offsetting forces: increased digitalization and innovation in production and delivery processes may help offset adverse productivity shocks in some countries; others lack prerequisite widespread and reliable connectivity.
- Evidence cited for potential productivity impacts:
  - Bloom and others (2020) estimate private sector TFP to be 5 percent lower at the end of 2020 than it would have been, and likely to remain 1 percent lower in the medium term (as reported in the source text).
  - Bernstein, Townsend, and Xu (2020) document a “flight to safety” toward known brands and large companies in the US labor market; new business formation in the United States reached a record high in Q3 2020 (Brown 2020) — reported in source text.

### Forecast Revisions and Cross-Country Heterogeneity
- IMF April 2021 WEO growth forecasts envision output losses, relative to pre-pandemic projections, of about 3 percent for the world economy by 2024.
- By comparison, lasting damages over a comparable period from the global financial crisis were almost 8.7 percent for the world as a whole.
- Losses are expected to be much lower in advanced economies than in emerging market and developing economies, owing to differences in:
  - Scale of policy support
  - Access to vaccines and therapies
- Cross-country regression framework for forecast revisions:
  - ΔŶit = α + βXi + γΓi + εi, where ΔŶit is percentage change in forecasts for output in year t between two forecast vintages; Xi is regressor of interest; Γi is vector of controls.
  - Years t considered: 2022–2024.
  - Regressors of interest Xi include:
    - (i) income group (advanced, emerging, low-income)
    - (ii) share of GDP from tourism and transportation in 2019
    - (iii) share of GDP from services in 2019
    - (iv) fiscal support during COVID-19 crisis up to December 2020
  - Regressions that consider ii–iv include income-group and region fixed effects; ii–iv are standardized to zero mean and unit standard deviation.
  - Standard errors clustered at region level.
- Key quantitative correlates (cross-sectional regression results reported in source):
  - A one-standard deviation increase in tourism and travel share of GDP is associated with a 2.5 percent reduction in expected output in 2022; exposure through tourism fades but remains close to 2 percent in 2024.
  - Economies with larger service sectors are likely to experience larger output losses, with a ½ percent reduction in expected output in 2022.
  - Countries with larger pandemic-related above-the-line fiscal measures are projected to experience smaller losses, all else equal.
- Revisions between forecast vintages:
  - Comparing October 2020 WEO and April 2021 WEO, recent favorable news on vaccines and stronger-than-expected second half of 2020 improved projections more in advanced economies than in other income groups.

### Conclusions (as presented)
- Prospects for scarring from COVID-19 are substantial, even if lower than after the global financial crisis.
- Severe recessions in the past, particularly deep ones, have been associated with persistent output losses.
- Relative financial stability following the COVID-19 shock so far is encouraging, as greatest scarring in the past has occurred with recessions associated with financial crises.
- Experience from previous recessions suggests the productivity channel could be a key driver of persistent output losses.

*Source: Authors' calculations and analysis as presented in the chapter "8. EMDEs: Employment" of the supplied PDF content.*

### 2. Revisions by Income Groups

### 2. Revisions by Income Groups

### Medium-term output losses and divergence
- Medium-term output losses following the pandemic are currently expected to be large but exhibit significant variation across economies and regions.
- Despite higher-than-usual growth as the global economy recovers from the COVID-19 shock, world output is still anticipated to be about 3 percent lower in 2024 than pre-pandemic projections suggested.
- These expected losses are lower than what was seen during the global financial crisis, consistent with the swift policy response that supported incomes and helped contain financial sector disruptions.
- Emerging market and developing economies, in particular, are expected to have deeper scars than advanced economies, partly reflecting their greater sectoral exposure to the pandemic shock and more muted policy response.
- The picture of divergent recoveries, with a larger likelihood and extent of scarring in many of the same countries that have limited fiscal space, suggests a challenging path ahead.

### Channels of scarring and risks
- Recessions have typically been followed by persistent losses to total factor productivity; the current crisis is different from past recessions in many ways, and high uncertainty surrounds the outlook.
- Sectoral exposure: tourism and service sectors are highlighted as particularly important channels affecting countries differently (AEs, EMEs, LICs distinctions noted).
- Financial distress risk: experience from past recessions underscores the importance of avoiding financial distress as the COVID-19 policy response evolves.
- Human capital: setbacks to human capital accumulation are a key channel for long-term output losses and rising inequality.

### Policy recommendations to limit scarring and support recovery
- Maintain measures that support credit provision while ensuring balance sheet resilience and adequate buffers to prevent scarring from future financial instability.
- Maximize use of limited fiscal space by tailoring responses and targeting support to the most-affected sectors and firms.
- Implement policies that reverse setbacks to human capital accumulation, boost job creation, and facilitate worker reallocation.
- Promote competition, innovation, and technology adoption to lift productivity growth and boost investment.
- Multilateral cooperation on vaccines to ensure adequate production and timely universal distribution is crucial to prevent even worse scarring in developing economies.

### Data and sample coverage
- Appendix Table 1 lists economies included in the analysis; the sample used for the medium-term losses exercise is marked with an asterisk (*).
- The sample includes a wide set of economies across income levels (examples from the list: Afghanistan*; Albania*; Algeria*; Angola; Antigua and Barbuda*; Argentina; Armenia; Aruba*; Australia; Austria; Azerbaijan*; Bahamas, The*; Bahrain; Bangladesh*; Barbados; Belarus*; Belgium; ...; Zimbabwe).

*Source: IMF — 2. Revisions by Income Groups (from the supplied PDF content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021203-print-pdf.pdf_