## wpiea2021066-print-pdf

## Source details

**Canonical URL:** [wpiea2021066-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021066-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021066-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021066-print-pdf.pdf.json)

---

### Introduction and context
- The analysis focuses on Brazil using microdata from the quarterly PNAD-Continua and the monthly PNAD-COVID household surveys.
- Framework: employs the Alfaro et al (2020) accounting framework to assess ex-ante vulnerability of Brazil’s labor market to the COVID-19 shock.

### Ex-ante vulnerability and employment-at-risk estimates (three-stage scenario)
- Shutdown stage (immediate impact, Q2 2020):
  - Ex-ante employment-at-risk close to 38 million jobs (40.7 percent of total employment in Q2 2019).
  - Total employment in Q2 2019 was approximately 93.3 million jobs.
- Partial opening stage:
  - Employment-at-risk reduced to 22 million jobs.
- Advanced reopening stage:
  - Employment-at-risk reduced to 7 million jobs.
- Changes across stages:
  - Employment-at-risk drops by roughly 50 percent in the initial reopening phase (shutdown → partial opening).
  - Employment-at-risk drops by another 70 percent in the advanced reopening phase (partial → advanced reopening).
- Composition and concentration:
  - Employment-at-risk concentrated in less teleworkable, more contact-intensive sectors such as trade, transport and hospitality.
  - Informal workers face almost double the employment risk of formal workers.

### Observed labor market impacts (aggregate and sectoral)
- Aggregate job losses and participation:
  - As of August 2020, Brazil lost roughly 12 million jobs relative to the same period in 2019.
  - Participation rate fell to a historic low of 55 percent.
  - At the peak in May 2020, job losses in Chile, Colombia and Mexico exceeded 20 percent of total pre-pandemic employment; Brazil’s losses were lower in comparison.
- Sectoral employment changes (year-on-year reductions observed):
  - Construction, household services and hospitality experienced reductions in employment of over 18-34 percent year-on-year.
- Hours and wages:
  - Average weekly hours decreased from 39 hours pre-COVID to 27 hours post-COVID.
  - Table A2 (May changes, All Employed):
    - Monthly Income Change (%) −17.9
    - Change in Hours (%) −30.3
    - Hourly Income Change (%) 17.9
  - Formal employees (May):
    - Change in Hours-Formal Employees (%) −25.9
    - Change in Income-Formal Employees (%) −6.4
    - Hourly Income Change-Formal Employees (%) 26.3

### Distributional impacts and role of Emergency Aid (EA)
- EA program design and scale:
  - EA had a cost of about 4 percent of GDP in 2020.
  - EA initially covered 67.7 million eligible beneficiaries (around one-third of the population).
  - For May through August beneficiaries received monthly cash payments of R$600 (150$US); women head-of-households received twice the amount.
  - The program was extended through end-2020 with benefit reduced to R$300 for the last four months; eligibility tightening led to about 50 million beneficiaries by end-year.
- Coverage (May basis):
  - Initially covered as much as 38.6% of all households (26.3 million out of 68 million) — numbers based on May.
  - By May 2020 the government analyzed 52 million individuals in the Unified Civil Registry, of which 29.7 million were approved.
- Distributional outcomes (May):
  - Bottom income decile: monthly labor income decreased from R$502 to R$327, a decrease of 35 percent (loss of about 1 percentage point of their percentile share).
  - Top decile: labor income decreased by 17 percent, increasing their percentile share by approximately 3.2 percentage points.
  - More than 70 percent of households in the bottom decile received EA; bottom three deciles received 53.9 percent of benefits; top three deciles received about 10.7 percent.
  - Transfers increased the average income of the bottom 40 percent of households by 20 percent relative to pre-COVID income.
- EA impact on poverty and inequality (May counterfactuals):
  - Without EA:
    - Gini index would have increased from 0.53 pre-COVID to 0.58 post-COVID.
    - Poverty headcount ratio (extreme poverty line of R$128 per month) would have increased from 4.7 percent to 13.9 percent in May — corresponding to approximately 15.7 million individuals.
    - For households headed by single earner women with at least one child, poverty rate would have increased from 11 to 30.4 percent.
  - With EA (May):
    - Poverty headcount ratio fell to 4.4 percent.
    - Gini coefficient fell to 0.51.
    - For households headed by single earner women with at least one child, poverty rate temporarily fell to 8.4 percent.
  - Depending on the poverty line used, EA prevented between 7.8 and 16.4 million people from falling into poverty at the peak of the crisis.
- Dynamics and expiration risk:
  - Progressive decrease of post-COVID poverty headcount ratios between May and October; slight increase in ratio including EA in October reflects lower benefit in September.
  - EA expired at end-2020; a continued substantial improvement in the labor market will be necessary to avoid a potentially sharp increase in poverty and inequality in 2021.

### Targeting counterfactuals and scenarios (October analysis)
- Baseline extreme poverty headcount ratios:
  - Income pre-COVID: 4.23 percent
  - Income post-COVID: 6.65 percent
  - Income post-COVID with EA: 1.93 percent
- Targeting and redistribution scenarios (headcount ratios):
  - “Redist from top 30’ile” (costless): 1.17 percent
  - “Redist.from top 30’ile Costly” (3 percent administrative targeting cost): 1.23 percent
  - “Targeting criteria” (only eligible receive aid, no targeting cost): disbursed amount would be R$3.1 billion lower in October; headcount ratio 0.83 percent
  - “Targeting Criteria Costly” (3 percent targeting cost): headcount ratio 0.88 percent
  - Adding exclusion of those receiving other welfare benefits (except Bolsa Família): headcount ratio would rise to 4.7 percent
- Reduced-transfer scenarios (unit transfer reduced to R$200, “2/3 of EA”):
  - “2/3 of EA” (holding original distribution): would save R$4.9 billion in October and raise extreme poverty headcount ratio to 2.7 percent
  - “2/3 of EA from top 30’ile”: extreme poverty ratio 1.9 percent
  - “2/3 of EA from top 30’ile Costly” (3 percent targeting cost): extreme poverty ratio 2.0 percent
- Interpretation:
  - With better targeting, two thirds of the observed transfer could produce a similar extreme poverty rate as observed in October, even after accounting for administrative targeting costs.

### Heterogeneity: who was hardest hit (stylized facts and regression evidence)
- Groups disproportionately affected:
  - Informal workers, women and young cohorts were disproportionately affected.
  - Older cohorts retained employment more often but experienced large reductions in hours worked and monthly wages.
  - Households in the bottom deciles were most affected; extreme poverty and labor-income Gini increased sharply at the height of the pandemic in May 2020.
- Formally observed changes (May and Q2/Q3 comparisons):
  - Q3 2020: reduction of approximately 11 million informal jobs relative to 3 million jobs lost for formal workers.
  - Informality rate fell by 6.1 percentage points in 2020 Q2 relative to 2019 Q2.
  - Wage gap (informal vs formal): approximately 47 percent in 2020 Q2 from 53 percent in 2019 Q2.
  - Female participation (2015 Q1–2020 Q1 averages): about 52 percent; female unemployment averaged about 13 percent. 2020 Q2 female participation fell to 46 percent.
  - Hours contraction (year-on-year, Q2 2020 vs Q2 2019): females −24 percent (rebounded to −6 percent in Q3); males −16 percent in Q2 and +4 percent in Q3.
- Specific labor-market effects (selected magnitudes reported):
  - Workers in sectors subject to a higher probability of lockdown:
    - worked 3.4 percent fewer hours
    - earned 5.5 percent lower monthly wage
  - Workers in contact-intensive sectors:
    - worked 1.1 percent less hours
  - Workers in teleworkable sectors:
    - worked 1.9 percent less hours
    - earned 2 percent more than those in less teleworkable ones
  - Firm-size differences (vs large firms above 30 employees):
    - Small (<10 employees): earned 7 percent lower monthly wage and worked 4.1 percent less hours
    - Medium (10–30 employees): earned 5.8 percent lower monthly wage and worked 3.4 percent less hours
  - Informality:
    - Informal workers worked 1.3 percent more hours than formal workers
    - Informal workers earned 1.3 percent lower monthly wage than formal workers
- Regression evidence (selected interaction coefficients, Post COVID × characteristic):
  - Post COVID × High Exposure: Employed −0.0129*** (0.00287)
  - Post COVID × Secondary Educ: Employed −0.0154*** (0.00367); Log(Wage) 0.0271*** (0.00971)
  - Post X Female: Log(Hours per week) −0.0251*** (0.00442)
  - Post COVID × age 14-34: Employed −0.0134*** (0.00335); Log(Hours per week) 0.0543*** (0.00848)
  - Post COVID × Pr. Lockdown: Employed −0.0554***; Log(Wage) −0.0338***
  - Post COVID × Teleworkability: Log(Wage) 0.0201***; Log(Hours per week) −0.0192***
  - Post COVID × <=10 employees: Employed −0.0697***; Log(Hours per week) −0.0410***
  - Post COVID × Informal: Employed −0.0127*; Log(Wage) 0.0126**
  - Regression sample sizes:
    - Employed regression: 1,494,975 observations; R-squared 0.469.
    - Log(Wage) regression: 691,065 observations; R-squared 0.139.
    - Log(Hours per week) regression: 710,014 observations.
- Interpretations drawn from regressions:
  - Individuals in high exposure areas were 1.3 percentage points less likely to be employed post-COVID.
  - Younger cohorts (14–34 and 35–54) were 1.3 and 2.7 percentage points less likely to be employed than the 54+ cohort.
  - Workers with secondary education were 1.5 percentage points less likely to be employed than tertiary educated workers.
  - Employed females worked 2.5 percent fewer weekly hours (even if not significantly less likely to be employed).
  - Black workers were 1.3 percentage points less likely than white workers to be employed post-COVID.

### Ex-ante vs ex-post: explanatory factors for the gap
- Ex-post realized job losses were only about one-third of ex-ante employment-at-risk, though sectoral distribution of ex-post losses strongly correlated with ex-ante risk.
- Factors explaining lower observed losses relative to jobs-at-risk:
  - Government employment retention program covered around 10 million formal workers (cost estimated at around 0.6 percent of GDP) and allowed subsidized reduction in hours or temporary suspension of employment relationships.
  - Approximately 13 million workers were temporarily removed from work due to social distancing but officially remained employed (5 million of which unpaid); including these would bring “de facto” employment losses closer to ex-ante estimates.
- Key uncertainty:
  - Whether the employment retention program preserved employment relationships permanently or merely delayed job losses until program expiration.

### Policy implications and fiscal considerations
- Fiscal scale and sustainability:
  - EA cost about 4 percent of GDP and is not sustainable beyond the short-term; program terminated in December 2020.
  - Any additional social assistance spending should be financed within expenditure limits imposed by the federal expenditure ceiling to preserve debt sustainability, macroeconomic stability and market confidence.
  - Given budget rigidity, financing additional social assistance requires fiscal structural reforms aimed at flexibilizing the budget process and reducing mandatory spending to create fiscal space.
- Practical policy finding:
  - Further reducing the transfer by one-third (to R$200 a month per individual) combined with improved targeting would keep the poverty headcount ratio close to October levels even after accounting for administrative targeting costs.
- Targeting insight:
  - Counterfactual targeting analysis (October): with better targeting, two thirds of the observed transfer could produce nearly the same extreme poverty outcome, indicating scope for more cost-effective designs for renewed emergency transfers.

### Data, methodology and calibration notes
- Main data sources:
  - PNAD-Continua (quarterly/annual; available Q1 2012 to Q2 2020).
  - PNAD-COVID (special edition available since May 2020; telephone survey ~48 thousand households per week).
- Key methodological elements of ex-ante job-loss probability:
  - Job-loss probability π_it = min { 1, ((Supply Shock_it + Demand Shock_s) * AD) }.
  - Supply shock depends on sectoral lockdowns, firm size, teleworkability, contact-intensity, and upstream/downstream lockdown probabilities.
  - Demand shock includes direct and indirect losses; direct loss figures adopt consumption shocks to Sweden mapped to Brazilian sectors.
  - Aggregate demand multiplier AD = 1.682 (Alfaro et al. (2020)) in one section and AD=1.692 in annex calibration.
  - Employment-at-risk equals sum of all individuals employed prior to the pandemic weighted by their job-loss probabilities.
- Calibration highlights (selected table entries from Table A1):
  - Lock_s values: I Accommodation and Food Services 1; H Transportation and Storage 0.647; G Wholesale and Retail 0.561.
  - Direct loss (selected): I Accommodation and Food Services 0.540; H Transportation and Storage 0.526; G Wholesale and Retail 0.188.
  - Input-Output table: sourced from WIOD for Brazil.
  - Shares of teleworkability and contact intensity: Dingel and Neiman (2020) and Leibovici et al. (2020) applied via crosswalks.

*Source: wpiea2021066-print-pdf (IMF).*

### References _________________________________________________________________________ 29

### References

### Introduction and context
- The COVID-19 pandemic triggered major economic disruptions and severely affected labor markets worldwide, with Latin America among the worst affected regions.
- The analysis focuses on Brazil using microdata from the quarterly PNAD-Continua and the monthly PNAD-COVID household surveys.
- The analysis employs the framework of Alfaro et al (2020) to assess ex-ante vulnerability of Brazil’s labor market to the COVID-19 shock.

### Ex-ante vulnerability and employment-at-risk estimates
- Ex-ante employment at risk during the shut-down stage of the pandemic was close to 38 million jobs (40.7 percent of total employment in Q2 2019).
- Total employment in Q2 2019 was approximately 93.3 million jobs.
- Ex-post realized job losses were only about one-third of ex-ante employment at risk, though sectoral distribution of ex-post losses strongly correlated with ex-ante risk.
- More contact-intensive and less teleworkable jobs were more at risk.

### Observed labor market impacts (high-level)
- Employment fell sharply in vulnerable sectors; reductions in employment of over 18-34 percent year-on-year were observed in construction, household services and hospitality.
- Informal workers, women and young cohorts were disproportionately affected.
- Older cohorts retained employment more often but experienced large reductions in hours worked and monthly wages.
- Households in the bottom deciles of the income distribution were most affected; extreme poverty and labor-income Gini increased sharply at the height of the pandemic in May 2020.
- Social mobility indices dropped by about 50 percent when most regional governments closed schools and imposed restrictions around late March/early April; mobility levels recovered substantially by June when many measures were relaxed. Schools started reopening only in September.

### Distributional impacts and role of Emergency Aid (EA)
- The government’s large-scale temporary cash transfers (Auxílio Emergencial – Emergency Aid – EA):
  - Had a cost of about 4 percent of GDP in 2020.
  - Initially covered as much as 38.6% of all households (26.3 million out of 68 million) (these numbers are based on the month of May).
- Employment retention scheme:
  - Cost estimated at around 0.6 percent of GDP.
  - Around 25 percent of all formal private employment was covered by the program.
  - Program’s coverage of around 10 million workers is mentioned as indicative.
- Poverty and inequality counterfactuals:
  - Without EA, inequality measured by the Gini index would have increased from 0.53 pre-COVID to 0.58 post-COVID.
  - Without EA, the poverty headcount ratio would have increased from about 4.7 percent to 13.9 percent—corresponding to 15.7 million individuals.
  - With EA, the poverty headcount ratio fell to 4.4 percent and the Gini coefficient to 0.51—both lower than pre-COVID levels.
  - Depending on the poverty line used, EA prevented between 7.8 and 16.4 million people from falling into poverty at the peak of the crisis.
- Counterfactual targeting analysis (October survey): with better targeting, two thirds of the observed transfer would have produced the same extreme poverty rate as observed in October, even after accounting for administrative cost of better targeting—indicating scope for more cost-effective designs for renewed emergency transfers.

### Heterogeneity and regression evidence
- A regression analysis summarizes heterogeneity of COVID-19 impacts across wages, hours worked per week and informality status.
- Multiple margins of heterogeneity (formality, age, sector, gender, education) mattered; no single factor explained all variation.

### Data sources and methodology notes
- Main data sources: PNAD-Continua (quarterly/annual; available Q1 2012 to Q2 2020) and PNAD-COVID (special edition available since May 2020).
- PNAD-COVID conducted by telephone for approximately 48 thousand households per week and contains health and labor market modules, including remote work, leave, and income received outside work (EA, Bolsa Familia, BPC-LOAS, retirement, unemployment insurance, rent and others).
- For PNAD-Continua analysis:
  - Effective monthly income from all jobs used; nominal wages converted to real wages using a price index with Q1 2015 as the base quarter.
  - Households reporting income but coded as missing excluded: 0.08% of households for post-COVID income and 0.16% of households for pre-COVID income.
- PNAD-Continua and PNAD-COVID are not perfectly comparable; analyses never combine the two surveys and specify which survey underlies each analysis.

### Related methodological elements
- The ex-ante job loss probability calculation incorporates:
  - Sector-specific demand shock (fear of contagion).
  - Aggregate demand shock (Keynesian multiplier).
  - Sector-specific supply shock associated with lockdowns.
  - Occupation characteristics (contact-intensity, mitigated by teleworkability).
  - Impact of lockdowns and demand shocks on upstream and downstream sectors.
  - Uneven impact of lockdowns on firms of different sizes.
- Employment at risk equals the sum of all individuals employed prior to the pandemic, weighted by their respective job loss probabilities.

*Source: wpiea2021066-print-pdf - References*

### 2019. Looking at a broader measure of labor underutilization encompassing unemployed,

### 2019. Looking at a broader measure of labor underutilization encompassing unemployed,

### Labor market status and pre-COVID slack
- Broader measure of labor underutilization (unemployed, underemployed and discouraged workers) highlights the still high level of slack in the labor market in 2019.
- Underemployed definition: people who worked less than 40 hours a week but are willing and able to increase their working hours. (PNAD Continua)

### COVID-19 shock and aggregate job losses
- As of August 2020, Brazil has lost roughly 12 million jobs relative to the same period in 2019, with most job losses concentrated in March, April and May.
- This exceeds employment destruction of around 4 million jobs during Brazil’s 2015-16 recession.
- Participation rate fell to a historic low of 55 percent; unemployment increased only modestly as many who lost jobs left the labor force.
- At the peak in May 2020, job losses in Chile, Colombia and Mexico exceeded 20 percent of total pre-pandemic employment, close to double the losses for Brazil.
- Rolling quarter July-September saw the first job gains in Brazil after the initial shock; gains continued in August-October and September-November but erased only a small fraction of previous losses and the number of unemployed workers continued to increase through the latest available data.

### Ex-ante / ex-post jobs-at-risk: framework and calculations
- Job-loss probability π_it defined as:
  - π_it = min { 1, ((Supply Shock_it + Demand Shock_s) * AD) }
- Supply shock (Supply Shock_it) functional form includes:
  - S_it * min { 1, (1 − T_i) * Highcontact_i } * min { 1, Lock_s + Lup_s + Ldown_s }
  - Supply shock depends on lockdown stage, firm size (S_it), teleworkability, contact-intensity, and upstream/downstream lockdown probabilities.
- Demand shock (Demand shock_s) defined as:
  - Demand shock_s = Direct loss_s * (finalsales_s / grossspout_s) + Σ Indirect loss_j * (purchases_j:from:s / grossspout_s), j ≠ s
  - Direct and indirect demand shock values adopted from consumption shocks to Sweden applied to sectoral specification here.
- Aggregate demand multiplier AD uses standard Keynesian multiplier: AD = 1 / (1 − MPC).
  - In this exercise, AD = 1.682 (following Alfaro et. al (2020)).
- Employment-at-risk defined as sum of individuals employed prior to the pandemic weighted by their job-loss probabilities.
- The analysis does not incorporate furlough schemes and support programs (such as Brazil’s emergency employment retention program).

### Three-stage scenario results (baseline using 2019 Q2 PNADC)
- Shutdown stage (immediate impact, Q2 2020): ex-ante employment-at-risk close to 38 million jobs (considering intersectoral linkages and supply and demand channels).
- Partial opening stage: employment-at-risk reduced to 22 million jobs.
- Advanced reopening stage: employment-at-risk reduced to 7 million jobs.
- Reductions described:
  - Employment-at-risk drops by roughly 50 percent in the initial reopening phase (shutdown → partial opening).
  - Employment-at-risk drops by another 70 percent in the advanced reopening phase (partial → advanced reopening).
- Remaining 7 million at-risk jobs in advanced reopening driven by lingering demand effects and slower recovery in hospitality, transport and commerce.
- Employment-at-risk concentrated in less teleworkable, more contact-intensive sectors such as trade, transport and hospitality.
- Informal workers face almost double the employment risk of formal workers due to concentration in contact-intensive sectors.

### Ex-ante versus ex-post (observed) outcomes
- Sectoral distribution of ex-ante jobs-at-risk is fairly well estimated relative to actual changes in employment.
- Total observed job losses were significantly lower than the universe of jobs at risk.
  - For large firms, employment losses match predicted jobs-at-risk.
  - Job losses in small and medium-sized firms (below 50 employees) were much lower than jobs-at-risk estimates.
- Contributing factors:
  - Government employment retention program covered around 10 million formal workers and allowed for subsidized reduction in hours or temporary suspension of employment relationships.
  - Approximately 13 million workers were temporarily removed from work due to social distancing but officially remained employed (5 million of which unpaid). Including these would bring “de facto” employment losses closer to ex-ante estimates.
- Key open question: whether the employment retention program will maintain employee-employer relationships permanently or merely delay job losses until the program expires.

### Heterogeneous impacts — stylized facts

- Labor formality:
  - Q3 2020: reduction of approximately 11 million informal jobs relative to 3 million jobs lost for formal workers.
  - Informality rate fell by 6.1 percentage points in 2020 Q2 relative to 2019 Q2 (hit a record low during the peak).
  - Wage gap (informal vs formal) reached approximately 47 percent in 2020 Q2 from 53 percent in 2019 Q2.
  - Formal workers were more likely to be doing remote unpaid work; probability almost three times as large for older cohort workers than younger cohort workers (logit model).

- Age:
  - Younger cohorts faced the highest reductions in employment; older cohorts experienced larger reductions in hours with milder decreases in employment.
  - Older cohorts more likely removed from work due to social distancing while retaining employment status, resulting in an increase in income per hour worked during Q3 compared to a year earlier (likely effect of job retention program).

- Gender:
  - Between 2015 Q1 and 2020 Q1: female labor participation and unemployment averaged about 52 and 13 percent; male averages about 72 and 10 percent.
  - 2020 Q2: female participation rate fell by about 6 percentage points to 46 percent (from 52 percent in 2020 Q1); continued decreasing in Q3.
  - Female unemployment rate increased from 14.5 percent in 2020 Q1 to about 17 percent in Q3.
  - Male participation decreased by 5 percentage points to 66 percent in 2020 Q2 but rebounded slightly in Q3; male unemployment rose from 10.4 percent in 2020 Q1 to about 13 percent in Q3.
  - Female workers suffered a more pronounced fall in employment than males in both Q2 and Q3; hours worked by women decreased by 24 percent in 2020 Q2 compared to 2019 Q2 and rebounded to −6 percent in Q3.
  - Hours contraction for males: −16 percent in Q2 and +4 percent in Q3.
  - On average income per hour worked for formal workers rose by 26 percent (28 percent for females and 12 percent for males), potentially reflecting job retention program effects.

- Sectoral patterns:
  - Average weekly hours decreased from 39 hours pre-COVID to 27 hours post-COVID.
  - Most contact-intensive industries (recreational activities, personal and domestic services, etc.) experienced largest drops in hours and income.
  - For May, females saw a reduction in monthly hours worked of 36 percent versus 27 percent for males.

- Urban vs Rural:
  - Employment reduction observed in both rural and urban areas.
  - 2020 Q3: urban employment decline reached 12.4 percent relative to same quarter previous year.
  - Those remaining employed in both areas saw little changes in income but sharp reductions in hours worked; fall in hours sharper in urban areas.

### Regression evidence — margins of heterogeneity (specification and main coefficients)
- Regression specification:
  - Y_it = α + β1 X_it + β2 Post_t + β3 X_it × Post_t + γ_uf + γ_industry + ε_it
  - Dependent variables Y_it: employed (binary), monthly wage (log), hours per week (log).
  - X_it: sex, education level, age, firm size, teleworkability, contact intensity, probability of lockdown, race, high exposure.
  - Post_t = 1 for quarter 2 of 2020 and 0 otherwise. State and industry fixed effects included. Controls and survey design/weights accounted for.

- Key regression coefficient estimates (Table 2 summary; standard errors in parentheses):
  - Post COVID × High Exposure: Employed −0.0129*** (0.00287); Log(Wage) 0.00866 (0.00734); Log(Hours per week) −0.00353 (0.00511).
  - Post COVID × Primary Educ: Employed −0.00421 (0.00380); Log(Wage) 0.0725*** (0.0115); Log(Hours per week) 0.0337*** (0.00722).
  - Post COVID × Secondary Educ: Employed −0.0154*** (0.00367); Log(Wage) 0.0271*** (0.00971); Log(Hours per week) −0.00379 (0.00602).
  - Post X Female: Employed −0.000776 (0.00235); Log(Wage) −0.00380 (0.00549); Log(Hours per week) −0.0251*** (0.00442).
  - Post COVID × age 14-34: Employed −0.0134*** (0.00335); Log(Wage) 0.0209** (0.0103); Log(Hours per week) 0.0543*** (0.00848).
  - Post COVID × age 35-54: Employed −0.0265*** (0.00335); Log(Wage) −0.0244*** (0.00947); Log(Hours per week) 0.00752 (0.00774).
  - Post COVID × Black: Employed −0.0129*** (0.00463); Log(Wage) 0.000456 (0.00989); Log(Hours per week) 0.00530 (0.00758).
  - Post COVID × Indigenous: Employed −0.0537*** (0.0194); Log(Wage) −0.0247 (0.0450); Log(Hours per week) 0.00574 (0.0326).
  - Post COVID × Pr. Lockdown: Employed −0.0554***; Log(Wage) −0.0338***.
  - Post COVID × Teleworkability: Log(Wage) 0.0201***; Log(Hours per week) −0.0192***.
  - Post COVID × Contact-Intensity: Log(Wage) 0.00185; Log(Hours per week) −0.0113**.
  - Post COVID × <=10 employees: Employed −0.0697***; Log(Hours per week) −0.0410***.
  - Post COVID × >10 and <=30 employees: Employed −0.0584***; Log(Hours per week) −0.0346***.
  - Post COVID × Informal: Employed −0.0127*; Log(Wage) 0.0126**.
- Observations:
  - Employed regression: 1,494,975 observations; R-squared 0.469.
  - Log(Wage) regression: 691,065 observations; R-squared 0.139.
  - Log(Hours per week) regression: 710,014 observations.
- Notable interpretations from regressions (from text):
  - Individuals in high exposure areas were 1.3 percentage points less likely to be employed post-COVID.
  - Younger cohorts (14-34 and 35-54) were 1.3 and 2.7 percentage points less likely to be employed than the 54+ cohort.
  - Workers with secondary education were 1.5 percentage points less likely to be employed than tertiary educated workers.
  - Workers with primary or secondary education who retained jobs saw statistically significant increases in wage per hour; primary-educated workers worked more hours relative to tertiary educated workers.
  - Females not significantly less likely to be employed post-COVID, but employed females worked 2.5 percent fewer weekly hours.
  - Black workers post-COVID were 1.3 percentage points less likely than white workers to be employed; no significant difference in wage and hours for those who remained employed.
  - Brown workers not less likely to be employed post-COVID vs white workers, but those that remained employed witnessed an increase in their wage.

### Summary observations and policy-relevant considerations (as presented)
- The labor market entered the COVID-19 shock with significant pre-existing slack.
- The scale of job losses was unprecedented, but measured unemployment rose only modestly due to large reductions in labor force participation.
- Ex-ante jobs-at-risk analysis indicates a much larger potential loss than observed ex-post; government employment retention program and temporary removal of workers from work while officially employed explain part of the gap.
- Informal workers, younger cohorts, workers in high-exposure areas, workers in small firms, and contact-intensive sectors suffered disproportionate employment risk and losses.
- The persistence of job retention program effects is a critical uncertainty: whether relationships have been preserved permanently or losses are only delayed remains a key question for social outcomes and the speed of economic recovery.

*Source: PNAD-Continua, IBGE Official Statistics, and authors’ calculations as presented in the provided IMF content.*

### 2.3 percent compared to white workers.

### wpiea2021066-print-pdf - 2.3 percent compared to white workers.

### Labor market impacts and heterogeneity
- Workers in sectors subject to a higher probability of lockdown:
  - worked 3.4 percent fewer hours
  - earned 5.5 percent lower monthly wage
- Workers in contact-intensive sectors:
  - worked 1.1 percent less hours
- Workers in teleworkable sectors:
  - worked 1.9 percent less hours
  - earned 2 percent more than those in less teleworkable ones
- Firm-size differences:
  - Workers in small (<10 employees) firms earned 7 percent lower monthly wage and worked 4.1 percent less hours than workers in large firms (above 30 employees)
  - Workers in medium sized (between 10 and 30 employees) firms earned 5.8 percent lower monthly wage and worked 3.4 percent less hours than workers in large firms
- Informality:
  - Informal workers worked 1.3 percent more hours than formal workers
  - Informal workers earned 1.3 percent lower monthly wage than formal workers
- Disproportionate impacts:
  - Economically more vulnerable groups (informal workers, women, the young) were the hardest hit
  - Older cohorts retained employment but experienced high reductions in hours worked and monthly wages
- Jobs-at-risk:
  - Employment-at-risk in the shut-down stage was close to 38 million jobs when accounting for intersectoral linkages and supply and demand channels
  - Employment-at-risk concentrated in less teleworkable and more contact-intensive sectors such as trade, transport and hospitality
  - Informal workers face almost double the employment risk relative to formal workers

### Distributional impacts and the role of Emergency Aid (EA)
- Labor income changes by decile (monthly labor income):
  - Bottom income decile: decreased from R$502 to R$327, a decrease of 35 percent, resulting in a loss of about 1 percentage point of their percentile share in total income
  - Top decile: labor income decreased by 17 percent, resulting in an increase in their percentile share by approximately 3.2 percentage points
- EA program design and coverage:
  - EA initially covered 67.7 million eligible beneficiaries, around one-third of the population
  - For May through August beneficiaries received monthly cash payments of R$600 (150$US); women head-of-households received twice the amount
  - The program was extended through end-2020, with the benefit reduced to R$300 for the last four months of the year
  - Eligibility tightening led to about 50 million beneficiaries receiving the aid by end-year
  - By May 2020, the government analyzed all 52 million individuals in the Unified Civil Registry (Cadastro Único, CU), of which 29.7 million were approved, including 96 percent of Bolsa Família beneficiaries
  - Beneficiaries could receive transfers via digital savings account
- Targeting and distributional outcomes (May analysis):
  - More than 70 percent of households in the bottom decile received the aid, accounting for almost 23 percent of total EA disbursements
  - Bottom three deciles received 53.9 percent of benefits
  - Top three deciles received about 10.7 percent of the aid
  - At the peak of the crisis, the transfers increased the average income of the bottom 40 percent of households by 20 percent relative to pre-COVID income
- EA impact on poverty and inequality (May):
  - The aid corresponded to 150 percent of the pre-COVID labor income in the bottom decile and 40 percent of the pre-COVID overall median labor income
  - Without EA:
    - Gini index would have increased from 0.53 pre-COVID to 0.58 post-COVID
    - Poverty headcount ratio (extreme poverty line of R$128 per month) would have increased from 4.7 percent to 13.9 percent in May — corresponding to approximately 15.7 million individuals saved from falling into extreme poverty
    - For households headed by single earner women with at least one child, poverty rate would have increased from 11 to 30.4 percent
  - With EA in May:
    - Poverty headcount ratio fell to 4.4 percent
    - Gini coefficient fell to 0.51
    - For households headed by single earner women with at least one child, poverty rate temporarily fell to 8.4 percent
  - Depending on the poverty line used, EA prevented between 7.8 and 16.4 million people from falling below the poverty line between pre-COVID and post-COVID income from May until October
- Dynamics and expiration risk:
  - Progressive decrease of post-COVID poverty headcount ratios between May and October; slight increase in the ratio including EA in October reflects the lower benefit granted in September
  - Aid expired at the end of 2020; a continued substantial improvement in the labor market will be necessary to avoid a potentially sharp increase in poverty and inequality in 2021
  - Recovery signs: progressive rebound in hours worked and monthly wages for both males and females between May and October

### Counterfactual targeting and benefit scenarios (October analysis)
- Baseline headcount ratios (extreme poverty line) reported in Figure 19:
  - Income pre-COVID: 4.23 percent
  - Income post-COVID: 6.65 percent
  - Income post-COVID with EA: 1.93 percent
- Redistribution from top 30th percentile:
  - “Redist from top 30’ile” (costless): headcount ratio about 1.17 percent
  - “Redist.from top 30’ile Costly” (3 percent administrative targeting cost): headcount ratio about 1.23 percent
- Replicating EA eligibility criteria:
  - “Targeting criteria” (only eligible receive aid, no targeting cost): the total amount spent would have been R$3.1 billion lower than what was disbursed in October; headcount ratio falls to 0.83 percent
  - “Targeting Criteria Costly” (3 percent targeting cost): headcount ratio 0.88 percent
  - Adding exclusion of those receiving other welfare benefits (except Bolsa Família): headcount ratio would rise to 4.7 percent
- Reduced transfer scenarios (unit transfer reduced by one-third to R$200, “2/3 of EA”):
  - “2/3 of EA” (holding original distribution): would result in R$4.9 billion savings in October (approximately 33 percent of the EA cost in the same month) and a marginal increase in the extreme poverty headcount ratio to 2.7 percent
  - “2/3 of EA from top 30’ile”: extreme poverty ratio would be 1.9 percent
  - “2/3 of EA from top 30’ile Costly” (3 percent targeting cost): extreme poverty ratio would be 2.0 percent
- Interpretation:
  - With better targeting, even two thirds of the original amount of aid would be very effective in reducing extreme poverty to a ratio very close to the one observed in October
  - Targeting costs reduce net aid received and can affect poverty outcomes, but redistribution from wealthier percentiles could substantially lower extreme poverty

### Policy implications and conclusion
- Aggregate impact:
  - COVID-19 resulted in almost 13 million job losses between December 2019 and August 2020
  - Reduction in employment disproportionately affected informal workers, women and young cohorts
- Role of policies:
  - Government employment retention program covered around 10 million formal workers and likely helped reduce actual job losses relative to jobs-at-risk; it allowed subsidized reduction in hours or temporary suspension of employment relationships
  - Key question: whether employment retention program will preserve employer-employee relationships permanently or simply delay job losses once it expires
- Fiscal and sustainability considerations:
  - EA cost about 4 percent of GDP and is not sustainable beyond the short-term; the program was terminated in December
  - Any additional social assistance spending should be financed within expenditure limits imposed by the federal expenditure ceiling to preserve debt sustainability, macroeconomic stability and market confidence
  - Given budget rigidity, financing additional social assistance will require firm implementation of fiscal structural reforms aimed at flexibilizing the budget process and reducing mandatory spending to create necessary fiscal space
- Practical policy finding:
  - Further reducing the transfer by one-third (to R$200 a month per individual) combined with improved targeting would be successful in keeping the poverty headcount ratio close to its October levels even after accounting for administrative targeting costs

*Source: Authors’ analysis based on PNAD COVID and related sections of the provided content.*

### ANNEX I

### ANNEX I

### A. COVID-19 Cases, Government Regulations and Mobility
- Figures referenced:
  - Figure A1. Brazil: COVID-19 Cases and Government Response
  - Figure A2. Brazil: Mobility in Brazil
- Data sources:
  - Government response index: Blavatnik School of Government – University of Oxford. The government response index measures how many of a set of indicators a government has acted upon and to what degree.
  - Mobility data: Google Mobility data.

### B. Ex-Ante Employment Distribution
- Figures referenced:
  - Figure A3. Brazil: Employment Share by Gender Across Teleworkable and Contact-Intensive Jobs
  - Figure A4. Brazil: Employment Share by Gender Across Sectors Defined by Confinement Measures
- Data source and methodology:
  - PNAD-Continua 2019Q2. Authors’ analysis of employment share based on Dingel and Neiman (2020) teleworkability and contact-intensive indices.
  - PNAD-Continua 2019Q2. Authors’ analysis of employment share based on Perez et al (2020) division of sectors based on COVID-19 confinement measures on EU labor markets.

### C. Ex-Ante and Ex-Post Employment Analysis
- Figures referenced:
  - Figure A5. Brazil: Ex-Post Employment Categories (Source: PNAD COVID (May, 2020) and authors’ calculations)
  - Figure A6: Probability of Being Removed from Work Temporarily due to Social Distancing by Age (Authors’ analysis of PNAD COVID. Results from a logit model analysis.)
  - Figure A7: Probability of Doing Unpaid Remote Work by Sector (Authors’ analysis of PNAD COVID. Results from a logit model analysis.)
  - Figure A8: Change in Actual Employment in the Formal Sector
  - Figure A9: Change in Actual Employment in the Informal Sector
- Analytical framework:
  - Authors’ analysis of PNAD Continua using the Alfaro et al. (2020) Accounting framework.

- Table A1. Brazil: Calibration Parameters for the Accounting Exercise
  - Lock_j reflects the impact of different stages of lockdowns in each sector. The following values are assumed:
    - sector_code / sector_name / Lock_s / Lock_s_selective
    - A Agriculture 0 0
    - B, D, E Mining, electricity and water 1 1
    - C Manufacturing 0.681 0
    - F Construction 1 0
    - G Wholesale and Retail 0.561 0.280
    - H Transportation and Storage 0.647 0.647
    - I Accommodation and Food Services 1 1
    - J Information and Communication 0 0
    - K Finance and Insurance 0 0
    - L Real Estate 0 0
    - M, N, R, S Professional services, administrative services, arts and entertainment, etc. 0.114 0.114
    - O Public administration 0 0
    - P Education 0 0
    - Q Human Health and Social Services 0 0
    - T Household as employer etc. 1 0.7

  - it takes the following values:
    - Shutdown period / Partial opening period / Advanced opening period
    - Sectoral Opening<50 percent / Sectoral Opening>50 percent / Sectoral Opening<50 percent / Sectoral Opening>50 percent
    - Small, self-employed, informal: 0.31 0.58 0 0 0
    - Medium: 0.24 0.45 0.225 0.225 0
    - Large: 0.13 0.29 0 0 0

  - Direct loss (direct demand shock): Adopting consumption shock to Sweden to our sectoral classification
    - A Agriculture 0.055
    - B, D, E Mining, electricity and water 0
    - C Manufacturing 0.111
    - F Construction 0
    - G Wholesale and Retail 0.188
    - H Transportation and Storage 0.526
    - I Accommodation and Food Services 0.540
    - J Information and Communication 0
    - K Finance and Insurance 0
    - L Real Estate 0
    - M, N, R, S Professional services, administrative services, arts and entertainment, etc. 0.081
    - O Public administration 0
    - P Education 0
    - Q Human Health and Social Services 0
    - T Household as employer etc. 0

  - AD=1.692, following Alfaro et al. (2020).

  - Data Sources:
    - Input-Output table: sourced from WIOD for Brazil.
    - Shares of high teleworkability and high contact intensity in each occupation: sourced from Dingel and Nieman (2020) and Leibovici et al. (2020), computed on the US occupations with the SOC 2010 classification using the 2017 American Community Survey. The variables are then applied to the national labor surveys through a set of crosswalks from the SOC 2010 to various editions of the ISCO.

### D. Change in Earnings and Actual Hours Worked
- Table A2. Brazil: Change in Hours and Income Statistics for May
  - Columns: (1) All Employed | (2) Employed (excluding zero wage earners and zero hours worked)
  - Monthly Income Change (%) -17.9 -10.8
  - Change in Hours (%) -30.3 -11.2
  - Hourly Income Change (%) 17.9 0.4
  - Change in Hours-Formal Employees (%) -25.9 -7.9
  - Change in Income-Formal Employees (%) -6.4 -4.3
  - Hourly Income Change-Formal Employees (%) 26.3 3.9
  - Change in Hours - Male (%) -26.6 -10.4
  - Change in Hours - Female (%) -35.9 -12.5
  - Change in Income - Male (%) -17.9 -11.2
  - Hourly Income Change-Male (%) 11.8 -0.9
  - Change in Income - Female (%) -17.7 -10.1
  - Hourly Income Change-Female (%) 28.3 2.7

- Note: The table presents the percent change in hours worked per month, wage per month and wage per hour for the month May (comparing pre-COVID and Post-COVID). The first column presents the percent changes among all employed people and the second column presents the results for employed people excluding those that report earning a zero wage and working zero hours.

- Figures referenced:
  - Figure A10. Brazil: Kernel Density Function of Log (Effective Monthly Income)
  - Figure A11. Brazil: Variation of Employment, Income and Hours by Occupation (2020Q2 vs 2019Q2)
  - Figure A12: Variation in Monthly Income across Sectors
  - Figure A13: Variation in Weekly Hours across Sectors
  - Figure A14. Brazil: Change in Weekly Hours across Formal and Informal Employees
  - Source for figures: PNAD Continua and authors’ calculations

### E. Auxilio Emergencial
- Figure referenced:
  - Figure A16. Brazil: Percent of Beneficiary Households by Amount of Transfer
  - Source: PNAD COVID and authors’ calculations

- Table A1. Brazil: Aid Received by Top 2 Deciles Households (based on pre-COVID and current post-COVID income)
  - Columns: mean median min max
  - Auxilio Emergencial 659 600 300 6000
  - Household Members 2.6 2 1 12
  - Pre-COVID Income 5475.6 4600 100 52200
  - Post-COVID Income 5162.6 4300 1250 42000
  - Post-COVID Income with EA 5821.6 5100 1580 42600
  - Source: PNAD COVID and authors’ calculations

- Table A2. Brazil: Households that received aid above R$900
  - Columns: mean median min max
  - Auxilio Emergencial 1913.2 1800 1209 5432
  - Household Members 4.8 4 1 17
  - Pre-COVID Income 2178.4 1500 30 30000
  - Post-COVID Income 1383.5 1000 0 30000
  - Post-COVID Income with EA 3296.7 2845 1212 32400
  - Source: PNAD COVID and authors’ calculations

*Source: ANNEX I, wpiea2021066-print-pdf (IMF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021066-print-pdf.pdf_