## annexch2

## Source details

**Canonical URL:** [annexch2](https://www.imf.org/-/media/files/publications/weo/2020/october/english/annexch2.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/weo/2020/october/english/annexch2.pdf.md)
- [Structured JSON version](/-/media/files/publications/weo/2020/october/english/annexch2.pdf.json)

---

### Sample coverage and data sources
- Cross-country sample varies between 22 and 52 countries based on data availability.
- High-frequency indicators:
  - Job postings sample: 22 countries.
  - Mobility sample: 128 countries.
- Subnational mobility: 422 units for 15 G20 countries.
- Vodafone data: limited to Italy, Spain, and Portugal.
- Infections analysis sample: 89 countries with information on temperature, humidity, public information campaigns, testing, and contact tracing.
- Subnational infections sample: 373 units for G20 15 countries.
- Lockdown stringency data source: Coronavirus Government Response Tracker of the University of Oxford. The stringency index is a simple average of nine sub-indexes (school closures, workplace closures, cancellations of public events, gathering restrictions, public transportation closures, stay-at-home requirements, restrictions on internal movement, controls on international traveling, and public information campaigns). The analysis constructs the stringency index excluding public information campaigns.

### Cross-country specification and key variables
- Forecast error definition: deviation of real GDP growth from the January 2020 World Economic Outlook projections (sum of real GDP in the first two quarters of 2020 and growth relative to the same sum a year ago).
- Main estimated specification (equation (2.1)):
  - y_i = α + β lock_i + γ cases_i + ε_i
  - y_i alternately: forecast error of real GDP, real consumption, real investment in the first half of 2020; average growth of industrial production and retail sales; average change in manufacturing PMI and services PMI in the first three months after a country’s epidemic started.
  - lock_i: average lockdown stringency over the same period as y_i.
  - cases_i: log of per capita COVID-19 cases at end of the period used for y_i.
- Findings:
  - Lockdowns are associated with lower economic activity.
  - Impact remains significant whether or not the spread of the virus is controlled for.
- Comparative metric:
  - Coefficient β is scaled by (standard deviation of lockdowns / standard deviation of the economic activity indicator) to compare across indicators.

### High-frequency mobility and job-postings methodology
- Mobility indicator: average of Google mobility indexes for groceries and pharmacies, parks, retail and recreation, transit stations, and workplaces (China: Baidu). Available for over 130 countries at daily frequency since early February.
- Job postings: Indeed data available for 22 countries (18 advanced economies and 4 emerging market and developing economies).
- Estimation approaches:
  - Local projections (Jordà, 2005) and panel regressions.
  - Regressions include mob_i,t+h, ln Δ cases_i,t−p, lock_i,t−p, lags of dependent variable (week worth of lags), country and time fixed effects; standard errors clustered at country level.
- Robustness:
  - Replacing log daily cases with log daily deaths does not change results.
  - Driscoll-Kraay (1998) correction yields similar standard-error adjustments.

### Key empirical findings on mobility
- Response to a full lockdown:
  - Mobility declines after a week by almost 25 percent relative to pre-lockdown level (Figure 2.2 panel 1).
  - Effect dies off over a month as restrictions are eased.
- Effect of COVID-19 cases on mobility:
  - A doubling of COVID-19 cases leads to a decline in mobility of about 2 percent after 30 days (Figure 2.2 panel 2).
  - Using COVID-19 deaths: mobility declines by 28 percent a week after lockdown introduction; a doubling of COVID-19 deaths leads to a reduction in mobility by 1.2 percent after 30 days.
- Subnational identification (422 units, 15 G20 countries):
  - Mobility response shape similar to national data but magnitude about 10 percentage point larger.
  - Doubling of COVID-19 cases leads to contraction in mobility of 1.7 percent after 30 days (subnational).
- Location-specific results:
  - Across retail and recreation, groceries and pharmacies, parks, transit stations, workplaces: a full lockdown leads to a reduction in mobility between 23 and 28 percent about a week after introduction.
  - A doubling in COVID-19 cases leads to a decline in mobility between 1 and 2.8 percent after 30 days.
- Controls for public-information campaigns, contact tracing, and massive testing:
  - Including these controls yields: a full lockdown reduces mobility by 24 percent after a week, and a doubling of COVID-19 cases leads to a reduction in mobility by 1.9 percent after 30 days.
- Heterogeneity tests:
  - Interaction specifications test whether population density (2019), rule of law (2018), trust or altruism (2012) modify effects.
  - Results indicate impacts of lockdowns and voluntary social distancing are not statistically different across different population densities, strength of rule of law, and levels of trust and altruism.

### Decomposition: lockdowns versus voluntary distancing
- Extended specification permits coefficients to differ by country group (advanced economies AE, emerging markets EM, low-income countries omitted).
- Contribution calculation (equation (2.84)):
  - C_i,x = Γ̄_g,h × x̄_i
  - C_i,x is the contribution of x ∈ {ln Δ cases, lock} to decline in mobility for country i; Γ̄_g,h is average coefficient over h horizons for country group g ∈ {AE, EM, LIC} with i ∈ g; x̄_i is average of variable during first 90 days of the epidemic. Contributions averaged across countries.
- Aggregate finding (128-country sample):
  - Lockdowns and voluntary social distancing played a similar role in reducing mobility.
  - Contribution of voluntary distancing is larger in advanced economies relative to emerging markets and low-income countries.

### Interaction and asymmetry results
- Interaction with domestic daily cases (equation (2.9)):
  - The impact of lockdowns on economic activity is smaller when cases are relatively higher; interpretation: when cases are high, fear-driven behavior reduces incremental effect of lockdowns.
  - Difference between effects of lockdowns with high and low cases is statistically significant (Figure 2.3.7).
- Interaction with global cases:
  - Replacing domestic with global cases corroborates that lockdowns have weaker impact on mobility when cases are relatively higher globally (Figure 2.3.8).
- Tightening versus loosening asymmetry (equation (2.10), with D indicator for easing periods using seven-day moving average of change in stringency):
  - Introduction of a full lockdown leads to decline in mobility of about 26 percent one week after tightening.
  - Lifting restrictions boosts mobility by about 18 percent over the same period.
  - Tightening and loosening effects are statistically different (Figure 2.3.9).

### Job postings results and sector heterogeneity
- Job postings impulse responses (22-country sample, January 1 to June 28, 2020):
  - Specification mirrors mobility analysis: 7 lags of dependent and independent variables; country and time fixed effects; standard errors clustered at country level (Driscoll-Kraay robust).
  - Full lockdown is associated with a decline in job postings of about 12 percent two weeks after introduction of the lockdown (Figure 2.4 panel 1).
  - Doubling COVID-19 cases leads to a 2 percent decline in job postings after 30 days (Figure 2.4 panel 2).
  - Negative effect robust to leave-one-out, but point estimate declines materially if New Zealand is excluded.
  - Decomposition shows both lockdowns and voluntary social distancing contributed to the drop in job postings; voluntary social distancing contribution particularly large given sample composition.
- Sectoral event study around national stay-at-home orders:
  - Contact-intensive sectors (food, hospitality, personal care) job postings started to decline a few weeks before stay-at-home orders—evidence of voluntary social distancing.
  - Manufacturing sector decline coincided with the introduction of stay-at-home orders.

### Mobility: gender differences and childcare hypothesis
- RD / event-study approach using Vodafone data (share of customers leaving home, disaggregated by gender and age; orthogonalized with respect to day-of-week and province fixed effects; standard errors clustered at the province level).
- Key findings:
  - Introduction of national stay-at-home orders led to a sharp drop in the share of people moving for both men and women.
  - The share of women moving dropped by a larger extent, with the difference being as large as 2 percent in the baseline model.
  - Age group 45–64: effect still significant but smaller—equal to about 1 percent.
  - Italy and Spain sample: difference between men and women equal to 3 percent.
  - Changing bandwidth to 20 days does not affect results.
  - Subsample of five northern Italian regions (local schools closed before national stay-at-home orders): divergence in mobility between men and women starts around school closures (first discontinuity February 23rd; national lockdown March 10th), consistent with women carrying a greater share of childcare responsibilities.

### Mobility: age-group heterogeneity
- Age-group-specific findings (Table 2.4.2 and Figure 2.4.1):
  - Age 18–24: largest drop in mobility because of lockdowns, close to 30 percent.
  - Age 25–44: decline as large as 20 percent.
  - Age 45–64: decline as large as 20 percent.
  - Age 65+: mobility declined by 19 percent; mobility was already significantly lower prior to lockdowns.
- Graphical evidence: magnitude of negative effect becomes smaller for older age groups.

### Lockdowns and epidemiological outcomes
- National-level panel (77 countries since beginning of January) controls: average temperature, humidity, public information campaigns, massive testing and contact tracing indicators, country-specific linear and quadratic trends, and lagged changes in ln(cases).
- Timing of epidemiological effects:
  - COVID-19 cases start declining 3 to 4 weeks after adoption of a lockdown, relative to a no-lockdown scenario.
  - After a month, cases are about 38 percent lower.
- Subnational analysis (339 units in 15 G20 countries; controls dropped at subnational level; units with the largest number of cases per country and those >20 percent of country’s total cases excluded):
  - After a month since adoption of a lockdown, COVID-19 cases in subnational units under a national lockdown are 58 percent lower than in subnational units without a lockdown.
- Timing of adoption heterogeneity:
  - One fourth of countries tightened lockdown measures within 20 days; half within a month; for the rest it took between a month and a half and four months.
  - Virtually all countries reached maximum stringency before daily cases reached 0.1 cases per thousand people.
  - Comparing early vs. late adopters (split at the median of distributions): countries that tightened lockdown measures early—by time to maximum stringency and by number of weekly cases at maximum stringency—had considerably less COVID-19 infections per thousand people 90 days after the first case.

### Lockdown measures, sequencing, and nonlinearities
- Stringency index aggregates school closures, workplace closures, stay-at-home orders, public event cancellations, gathering restrictions, public transport closures, internal movement restrictions, and international travel controls.
- Collinearity and sequencing:
  - Measures are often introduced in rapid succession, complicating assessment of each individual measure’s effectiveness because of collinearity.
  - Replacing the lockdown stringency index with rescaled indices for individual measures shows:
    - Measures introduced later (e.g., stay-at-home orders or transportation restrictions) display a smaller impact on mobility.
    - Measures introduced first (e.g., international movement restrictions or school closures) are associated with a larger impact.
- Nonlinear effects on mobility:
  - Introducing new measures (or tightening existing ones) when many other measures are already in place has a weaker effect on mobility than introducing them when fewer (or looser) measures are in place.
  - The quadratic term on lockdown stringency in mobility regressions is positive and statistically significant at various horizons.
- Nonlinear effects on epidemiological outcomes:
  - Lockdown measures have an impact on infections if they are introduced on top of existing ones.
  - The quadratic term is negative and statistically significant at various horizons.

### Synthesis and policy-relevant implications
- Empirical synthesis:
  - Lockdowns reduce mobility sharply, with larger declines for younger age groups (especially 18–24) and a larger mobility drop for women relative to men in relevant age brackets (25–44), plausibly linked to childcare responsibilities.
  - Lockdowns lead to reductions in COVID-19 cases with a lag of 3 to 4 weeks; after a month national cases are about 38 percent lower, and subnational estimates show a 58 percent reduction after a month in regions under national lockdown versus those without.
  - Sequencing and timing matter: early adoption of stringent measures is associated with substantially fewer infections per thousand people; measures introduced earlier have larger mobility effects; incremental tightening when many measures are already in place yields smaller mobility returns but can still contribute to reducing infections.
- Policy takeaway:
  - Tighter lockdowns appear to entail modest additional economic costs while bringing considerable benefits in containing the virus.

*Source: annexch2 (technical annex describing cross-country and high-frequency analyses).*

### Annex Table 2.1.1 lists the data sources used in the analysis. The sample coverage for the

### annexch2 - Annex Table 2.1.1 lists the data sources used in the analysis. The sample coverage for the

### Sample coverage and data sources
- Cross-country sample varies between 22 and 52 countries based on data availability.
- High-frequency indicators:
  - Job postings sample: 22 countries.
  - Mobility sample: 128 countries.
- Subnational mobility: 422 units for 15 G20 countries.
- Vodafone data: limited to Italy, Spain, and Portugal.
- Infections analysis sample: 89 countries with information on temperature, humidity, public information campaigns, testing, and contact tracing.
- Subnational infections sample: 373 units for G20 15 countries.
- Lockdown stringency data source: Coronavirus Government Response Tracker of the University of Oxford. The stringency index is a simple average of nine sub-indexes (school closures, workplace closures, cancellations of public events, gathering restrictions, public transportation closures, stay-at-home requirements, restrictions on internal movement, controls on international traveling, and public information campaigns). The analysis constructs the stringency index excluding public information campaigns.

### Cross-country specification and key variables
- Forecast error definition: deviation of real GDP growth from the January 2020 World Economic Outlook projections (sum of real GDP in the first two quarters of 2020 and growth relative to the same sum a year ago).
- Main estimated specification (equation (2.1)):
  - y_i = α + β lock_i + γ cases_i + ε_i
  - y_i alternately: forecast error of real GDP, real consumption, real investment in the first half of 2020; average growth of industrial production and retail sales; average change in manufacturing PMI and services PMI in the first three months after a country’s epidemic started.
  - lock_i: average lockdown stringency over the same period as y_i.
  - cases_i: log of per capita COVID-19 cases at end of the period used for y_i.
- Findings from Annex Table 2.2.1 and Figure 2.1 panel 2:
  - Lockdowns are associated with lower economic activity.
  - Impact remains significant whether or not the spread of the virus is controlled for.
- Comparative metric: coefficient β is scaled by (standard deviation of lockdowns / standard deviation of the economic activity indicator) to compare across indicators.

### High-frequency mobility and job-postings methodology
- Mobility indicator: average of Google mobility indexes for groceries and pharmacies, parks, retail and recreation, transit stations, and workplaces (China: Baidu). Available for over 130 countries at daily frequency since early February.
- Job postings: Indeed data available for 22 countries (18 advanced economies and 4 emerging market and developing economies).
- Local projections approach (Jordà, 2005) and panel regressions used to estimate dynamic responses; regressions include:
  - dependent variable mob_i,t+h (mobility at horizon h),
  - ln Δ cases_i,t−p (log of daily COVID-19 cases, with p lag length) to track pandemic stage,
  - lock_i,t−p (lockdown stringency index),
  - lags of dependent variable (week worth of lags),
  - country and time fixed effects,
  - standard errors clustered at country level.
- Robustness: replacing log daily cases with log daily deaths does not change results; Driscoll-Kraay (1998) correction yields similar standard-error adjustments.

### Key empirical findings on mobility
- Response to a full lockdown:
  - Mobility declines after a week by almost 25 percent relative to pre-lockdown level (Figure 2.2 panel 1).
  - Effect dies off over a month as restrictions are eased.
- Effect of COVID-19 cases on mobility:
  - A doubling of COVID-19 cases leads to a decline in mobility of about 2 percent after 30 days (Figure 2.2 panel 2).
  - Using COVID-19 deaths: mobility declines by 28 percent a week after lockdown introduction; a doubling of COVID-19 deaths leads to a reduction in mobility by 1.2 percent after 30 days.
- Subnational identification (422 units, 15 G20 countries; excludes the unit with the largest number of cases and units with more than 20 percent of country’s cases; US excluded due to state-level policies):
  - Mobility response shape similar to national data but magnitude about 10 percentage point larger.
  - Doubling of COVID-19 cases leads to contraction in mobility of 1.7 percent after 30 days (subnational).
- Location-specific robustness (retail and recreation, groceries and pharmacies, parks, transit stations, workplaces):
  - Across all locations, a full lockdown leads to a reduction in mobility between 23 and 28 percent about a week after introduction.
  - A doubling in COVID-19 cases leads to a decline in mobility between 1 and 2.8 percent after 30 days.
- Controls for public-information campaigns, contact tracing, and massive testing:
  - Including these controls yields: a full lockdown reduces mobility by 24 percent after a week, and a doubling of COVID-19 cases leads to a reduction in mobility by 1.9 percent after 30 days.
- Heterogeneity tests:
  - Interaction specifications test whether population density (2019), rule of law (2018), trust or altruism (2012) modify effects.
  - Results indicate impacts of lockdowns and voluntary social distancing are not statistically different across different population densities, strength of rule of law, and levels of trust and altruism.

### Decomposition: lockdowns versus voluntary distancing
- Extended specification permits coefficients to differ by country group (advanced economies AE, emerging markets EM, low-income countries omitted).
- Contribution calculation (equation (2.84)):
  - C_i,x = Γ̄_g,h × x̄_i
  - C_i,x is the contribution of x ∈ {ln Δ cases, lock} to decline in mobility for country i; Γ̄_g,h is average coefficient over h horizons for country group g ∈ {AE, EM, LIC} with i ∈ g; x̄_i is average of variable during first 90 days of the epidemic. Contributions averaged across countries.
- Aggregate finding (128-country sample):
  - Lockdowns and voluntary social distancing played a similar role in reducing mobility.
  - Contribution of voluntary distancing is larger in advanced economies relative to emerging markets and low-income countries.

### Interaction and asymmetry results
- Interaction with domestic daily cases (equation (2.9)):
  - The impact of lockdowns on economic activity is smaller when cases are relatively higher; interpretation: when cases are high, fear-driven behavior reduces incremental effect of lockdowns.
  - Difference between effects of lockdowns with high and low cases is statistically significant (Figure 2.3.7).
- Interaction with global cases:
  - Replacing domestic with global cases corroborates that lockdowns have weaker impact on mobility when cases are relatively higher globally (Figure 2.3.8).
- Tightening versus loosening asymmetry (equation (2.10), with D indicator for easing periods using seven-day moving average of change in stringency):
  - Introduction of a full lockdown leads to decline in mobility of about 26 percent one week after tightening.
  - Lifting restrictions boosts mobility by about 18 percent over the same period.
  - Tightening and loosening effects are statistically different (Figure 2.3.9).

### Job postings results and sector heterogeneity
- Job postings impulse responses (22-country sample, January 1 to June 28, 2020):
  - Specification mirrors mobility analysis: 7 lags of dependent and independent variables; country and time fixed effects; standard errors clustered at country level (Driscoll-Kraay robust).
  - Full lockdown is associated with a decline in job postings of about 12 percent two weeks after introduction of the lockdown (Figure 2.4 panel 1).
  - Doubling COVID-19 cases leads to a 2 percent decline in job postings after 30 days (Figure 2.4 panel 2).
  - Negative effect robust to leave-one-out, but point estimate declines materially if New Zealand is excluded.
  - Decomposition shows both lockdowns and voluntary social distancing contributed to the drop in job postings; voluntary social distancing contribution particularly large given sample composition.
- Sectoral event study around national stay-at-home orders:
  - Contact-intensive sectors (food, hospitality, personal care) job postings started to decline a few weeks before stay-at-home orders—evidence of voluntary social distancing.
  - Manufacturing sector decline coincided with the introduction of stay-at-home orders.

*Source: annexch2 (technical annex describing cross-country and high-frequency analyses).*

### introduction of stay-at-home orders, suggesting that in less-contact intensive sectors lockdowns

### annexch2 - introduction of stay-at-home orders, suggesting that in less-contact intensive sectors lockdowns

### Summary of methodology and exercises
- Two main empirical exercises:
  - Dynamics of job postings around reopening: panel where time zero denotes the lifting of stay-at-home orders; lifting restrictions led to marginal recovery in job postings, suggesting that raising lockdown restrictions alone is unlikely to provide a sharp boost until the virus is contained.
  - Regression discontinuity (RD) / event-study style analysis around the introduction of national stay-at-home orders to test heterogeneous effects across gender and age groups. The running variable is time with a narrow time window (baseline: 30 days) to address endogeneity and to make unobserved confounders likely similar within the window.
- Mobility data: share of Vodafone customers leaving home, disaggregated by gender and age, orthogonalized with respect to day-of-week and province fixed effects. Standard errors clustered at the province level.

### Mobility: gender differences and childcare hypothesis
- Main RD specification (local linear regression) estimates the percent of people moving in age group 25–44 by province, country, gender, and time, with interaction terms for Female and StayHome and Date.
- Key findings:
  - Introduction of national stay-at-home orders led to a sharp drop in the share of people moving for both men and women.
  - The share of women moving dropped by a larger extent, with the difference being as large as 2 percent in the baseline model (Table 2.4.1, Column (1)); the difference is statistically significant.
  - Robustness and alternative samples:
    - Age group 45–64: effect still significant but smaller—equal to about 1 percent (Column (2)).
    - Italy and Spain sample: difference between men and women equal to 3 percent (Column (3)).
    - Changing bandwidth to 20 days does not affect results (Column (4)).
  - Subsample of five northern Italian regions (local schools closed before national stay-at-home orders): divergence in mobility between men and women starts around school closures (first discontinuity February 23rd; national lockdown March 10th), consistent with women carrying a greater share of childcare responsibilities.

### Mobility: age-group heterogeneity
- Simpler specification estimated separately by age group with province-level clustered standard errors.
- Key magnitudes (Table 2.4.2 and Figure 2.4.1):
  - Age 18–24: largest drop in mobility because of lockdowns, close to 30 percent.
  - Age 25–44: decline as large as 20 percent.
  - Age 45–64: decline as large as 20 percent.
  - Age 65+: mobility declined by 19 percent; mobility was already significantly lower prior to lockdowns.
- Graphical evidence: magnitude of negative effect becomes smaller for older age groups.

### Lockdowns and epidemiological outcomes (national and subnational)
- National-level panel specification estimated for 77 countries since beginning of January, controlling for average temperature, humidity, public information campaigns, massive testing and contact tracing indicators, country-specific linear and quadratic trends, and lagged changes in ln(cases).
- Timing of epidemiological effects:
  - COVID-19 cases start declining 3 to 4 weeks after adoption of a lockdown, relative to a no-lockdown scenario.
  - After a month, cases are about 38 percent lower.
- Subnational analysis (339 units in 15 G20 countries; controls dropped at subnational level; units with the largest number of cases per country and those >20 percent of country’s total cases excluded):
  - After a month since adoption of a lockdown, COVID-19 cases in subnational units under a national lockdown are 58 percent lower than in subnational units without a lockdown.
- Timing of adoption heterogeneity:
  - One fourth of countries tightened lockdown measures within 20 days; half within a month; for the rest it took between a month and a half and four months.
  - Virtually all countries reached maximum stringency before daily cases reached 0.1 cases per thousand people.
  - Comparing early vs. late adopters (split at the median of distributions): countries that tightened lockdown measures early—by time to maximum stringency and by number of weekly cases at maximum stringency—had considerably less COVID-19 infections per thousand people 90 days after the first case.

### Lockdown measures, sequencing, and nonlinearities
- Stringency index (University of Oxford) aggregates school closures, workplace closures, stay-at-home orders, public event cancellations, gathering restrictions, public transport closures, internal movement restrictions, and international travel controls.
- Collinearity and sequencing:
  - Measures are often introduced in rapid succession, complicating assessment of each individual measure’s effectiveness because of collinearity.
  - Replacing the lockdown stringency index with rescaled indices for individual measures (Figure 2.6.1) shows:
    - Measures introduced later (e.g., stay-at-home orders or transportation restrictions) display a smaller impact on mobility.
    - Measures introduced first (e.g., international movement restrictions or school closures) are associated with a larger impact.
- Nonlinear effects on mobility (equation with quadratic terms of lockdown stringency):
  - Introducing new measures (or tightening existing ones) when many other measures are already in place has a weaker effect on mobility than introducing them when fewer (or looser) measures are in place.
  - The quadratic term on lockdown stringency in mobility regressions is positive and statistically significant at various horizons (panel 2 of Figure 2.8).
- Nonlinear effects on epidemiological outcomes (equation (2.13) augmented with squared lockdown stringency):
  - Lockdown measures have an impact on infections if they are introduced on top of existing ones.
  - The quadratic term is negative and statistically significant at various horizons (panel 3 of Figure 2.8).

### Synthesis and policy-relevant implications
- Empirical synthesis:
  - Lockdowns reduce mobility sharply, with larger declines for younger age groups (especially 18–24) and a larger mobility drop for women relative to men in relevant age brackets (25–44), plausibly linked to childcare responsibilities.
  - Lockdowns lead to reductions in COVID-19 cases with a lag of 3 to 4 weeks; after a month national cases are about 38 percent lower, and subnational estimates show a 58 percent reduction after a month in regions under national lockdown versus those without.
  - Sequencing and timing matter: early adoption of stringent measures is associated with substantially fewer infections per thousand people; measures introduced earlier have larger mobility effects; incremental tightening when many measures are already in place yields smaller mobility returns but can still contribute to reducing infections.
- Policy takeaway:
  - Tighter lockdowns appear to entail modest additional economic costs while bringing considerable benefits in containing the virus.

*Source: annexch2 - introduction of stay-at-home orders, suggesting that in less-contact intensive sectors lockdowns*

---


_Source: https://www.imf.org/-/media/files/publications/weo/2020/october/english/annexch2.pdf_