International Monetary Fund, "Estimating the Employment and GDP Multiplier of Emergency Cash Transfers in Brazil" (IMF Working Paper WP/22/55, March 2022)

Summary

IMF Working Paper WP/22/55, March 2022, by Daniel Cunha, Joana Pereira, Roberto Accioly Perrelli and Frederik Toscani, estimating what Brazil's 2020 Emergency Aid cash transfers to vulnerable households did for jobs and output. Using two-stage least squares across 5,478 municipalities, with transfers and formal job creation summed over April to September 2020, the authors estimate a formal employment multiplier. They convert it into an implied GDP multiplier of between 0.5 and 1.5. The low end sits below most published estimates, which the authors link to pandemic limits on supply chains and spending, and they conclude the transfers still supported employment and GDP.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Cite this document

“International Monetary Fund, ‘Estimating the Employment and GDP Multiplier of Emergency Cash Transfers in Brazil’ (IMF Working Paper WP/22/55, March 2022).” Original at www.imf.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/international-monetary-fund-estimating-the-employment-and-gdp-multipli-a775437f0d/, SHA-256 9f570d1d2c0459cca19bc83487ef26e8e5b271c648b6492afe867d42a42f1638.

Full text

Estimating the
Employment and GDP
Multiplier of Emergency
Cash Transfers in Brazil
Daniel Cunha, Joana Pereira, Roberto Accioly Perrelli, and
Frederik Toscani

WP/22/55

IMF Working Papers describe research in
progress by the author(s) and are published to
elicit comments and to encourage debate.
The views expressed in IMF Working Papers are
those of the author(s) and do not necessarily
represent the views of the IMF, its Executive Board,
or IMF management.

2022
MAR
© 2022 International Monetary Fund WP/22/55

IMF Working Paper
Western Hemisphere Department

Estimating the Employment and GDP Multiplier of Emergency Cash Transfers in Brazil
Prepared by Daniel Cunha, Joana Pereira, Roberto Accioly Perrelli, and Frederik Toscani*

Authorized for distribution by Rishi Goyal and Cheng Hoon Lim
March 2022

IMF Working Papers describe research in progress by the author(s) and are published to elicit
comments and to encourage debate. The views expressed in IMF Working Papers are those of the
author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.

ABSTRACT: We estimate the subnational employment and GDP multiplier of Brazil's 2020 federal cash
transfers to vulnerable households. Using two-stage least squares regressions we estimate a formal
employment multiplier and then apply an analytical transformation to recover an implied GDP multiplier in the
range of 0.5-1.5. The lower bound of this range lies below most estimates in the literature, which may result
from the exceptional constraints imposed by the pandemic on supply chains and consumption. Nevertheless,
even using the lower end of our range implies that federal cash transfers played an important role in supporting
employment and GDP.

RECOMMENDED CITATION: Cunha, D., Pereira, J., Perrelli, R A., and Toscani, F. (2022): “Estimating the
Employment and GDP Multiplier of Emergency Cash Transfers in Brazil,” IMF Working Paper WP/22/55

JEL Classification Numbers: E62, H53, R50

Keywords: Fiscal multipliers, Household cash transfers, Labor informality

dcunha@imf.org, jpereira@imf.org, rperrelli@imf.org, and
Author’s E-Mail Address:
ftoscani@imf.org

* We thank Hassan Adan, Sakai Ando, Andre Brunelli, Pedro Castro, Gabriel Chodorow-Reich, Steven Pennings, Rodrigo da Silva,
and Antonio Spilimbergo for useful comments and discussions. The views expressed herein are those of the authors and should
not be attributed to the IMF, its Executive Board, or its management.
IMF WORKING PAPERS Estimating the Employment and GDP Multiplier of Emergency Cash Transfer in Brazil

WORKING PAPERS

Estimating the
Employment and GDP
Multiplier of Emergency
Cash Transfers in Brazil

Prepared by Daniel Cunha, Joana Pereira, Roberto Accioly Perrelli and
Frederik Toscani

INTERNATIONAL MONETARY FUND 3
Contents
1 Introduction 5

2 Background: The Brazilian Emergency Aid Cash Transfer Program 8

3 Data 9

4 Research Design 12
4.1 Baseline Econometric Specification and Identification . . . . . . . . . . . . . . . . . . 12

5 Results 14
5.1 The (Formal) Employment Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.2 The Implied GDP Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

6 Conclusion 32

A Analytical Appendix 36

B Appendix 37

List of Figures
1 EA Disbursements Over Time and Municipal Heterogeneity . . . . . . . . . . . . . . 10
2 Correlation Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3 Formal Employment Multiplier and Formal Employment Weight by Municipalities . 21
4 Implied GDP Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

List of Tables
1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Baseline Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Interaction with Informality Regression Results . . . . . . . . . . . . . . . . . . . . . 22
2.A Regression Results Using Ethnic Patterns and BF cover as Instruments . . . . . . . 23
3.A Interaction with Informality Regression Results (Alternative Instruments) . . . . . . 24
4 Controlling for The Job Support Program (BEm) and Pandemic-related Lending
(PEAC) Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5 Formal Employment and GDP Multiplier at Different Horizons . . . . . . . . . . . . 30
6 Additional Robustness tests: Changing the Unit of Analysis . . . . . . . . . . . . . 32
7 Baseline Regression Results (OLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
8 Baseline Regressions First Stage Results . . . . . . . . . . . . . . . . . . . . . . . . . 38

4
1 Introduction
In response to the COVID-19 pandemic, the Brazilian federal government provided substantial fiscal
and financial support to the economy. Policies included tax relief for companies and households,
temporary income support to vulnerable individuals, credit facilities for firms (mostly small and
medium sized companies, which as a group employ the largest share of the population), a job-
retention scheme with subsides for furlough workers, and debt relief and cash transfers to states
and municipalities, to name a few. In all, fiscal measures in 2020 had a direct impact on the primary
deficit of over 7 percent of GDP. In addition, public banks opened pandemic-related credit lines
worth close to 5 percent of GDP. In tandem with the fiscal and quasi-fiscal support, Brazil’s central
bank cut the policy interest rate in quick succession to historical lows— equivalent to negative
interest rates in real terms—and announced extensive liquidity and capital relief measures.

The Emergency Aid (henceforth EA) cash transfers program, which cost nearly BRL 300 billion
(4.3 percent of GDP) in 2020, was the most emblematic pillar of the government’s response package.
Offering basic income to informal workers (employed or unemployed) and vulnerable households,
the program was instrumental to stave off a dramatic rise in poverty and inequality which COVID-
19 would have brought by its direct impact on labor markets, as shown by Flamini et al. (2021) and
Cardoso (2020). For several months the EA more than offset labor income losses for households
at the bottom four deciles of the income distribution, effectively increasing their real household
income relative to pre-pandemic levels.

While the primary goal of the EA program was to provide social assistance in the exceptional
context of the pandemic and forced social distancing, data on retail activity, mobility, and growth
performance - as well as anecdotal evidence - suggests that the EA also provided an important
cushion to the economy overall, avoiding a deeper recession in Brazil (with a contraction of around
4 percent of GDP, Brazil’s recession was milder than roughly 75 percent of advanced economies
and emerging markets). In this paper we aim to formalize the impact of EA on growth. We use
monthly private formal employment data at the municipal level to estimate the employment effects
of the EA, taking advantage of the large heterogeneity in the distribution of the EA eligibility
across Brazil’s several thousand municipalities. As in Chodorow-Reich (2019), we first estimate an
employment multiplier (in our case a formal employment multiplier - the number of private formal
sector jobs created by each BRL 100 thousand disbursed by the federal government on EA) and
then propose an analytical transformation to translate the employment multiplier into an output
multiplier. At the municipal level, no high-frequency data for the large informal sector (nearly
50 percent of the labor market) or public employment (5 percent) exists. This necessitates the
choice to estimate a private formal sector employment multiplier rather than a total employment
multiplier.1 To obtain causal estimates, we instrument for EA spending with the pre-pandemic
1 Informality rates vary from 16 to 97 percent across municipalities.

5
share of Bolsa Familia (longstanding conditional cash transfer program) recipients at the municipal
level.

In the baseline estimations we find a formal employment multiplier around 0.5 (for the six-month
window April-September 2020). This implies an annual cost per private formal job of over BRL
400,000 or USD 78,000. An important consideration, however, is the role of informality. In munic-
ipalities with low informality, estimating a private formal sector employment multiplier is similar
to estimating a total employment multiplier. But in municipalities with high informality, the pri-
vate formal sector employment multiplier is likely to be substantially below the total employment
multiplier (in the limit, with informality at 100 percent, the private formal employment multiplier
becomes meaningless and is equal to 0 as cost per job goes to ∞). Analyzing an expansion of the
Bolsa Familia in 2009, Gerard et al. (2021) find that municipalities with a high informality rate
have an underlying cost per formal job ten times larger than regions with low informality, meaning
that the formal employment multiplier of informal municipalities is 1/10 of the ones with lower
informality rate. To account for this heterogeneity in the formal sector employment multiplier,
we also run regressions interacting the EA variable with the pre-pandemic structural informality
rate. Taking a weighted average of the resulting multipliers to obtain a national estimate yields
a private formal employment multiplier over 6 months of around 1.6 - that is, a cost per year-job
of BRL 124,000 (or USD 24,000). This national multiplier is larger than the one obtained in the
baseline regression - as well as a multiplier obtained through a simple conditional mean (around
0.9) - because municipalities with a higher share of formal employment in total formal employment
(i.e., with higher weight) tend to have larger multipliers.

We derive analytical expressions which allow us to transform the estimated formal employment
multiplier into a total employment multiplier and a GDP multiplier, similar to Chodorow-Reich
(2019) but taking into account the share of formal and informal workers and their relative produc-
tivity. This yields a range of 1-3.5 for the annual total employment multiplier (annual cost per job
of USD 6,000-24,000) and a broad range of 0.3-1.8 for the GDP multiplier, with a preferred range
of 0.5-1.5 considering most adequate specifications. Essentially, lower estimates are obtained from
specifications without the informality interaction term, while the upper estimates originate from
specifications with such interaction term.

Conceptually, our estimates yield a cross-region transfer multiplier. Pennings (2021) shows that
a transfer multiplier should be smaller than a purchase multiplier, depending on how large the
marginal propensity to consume (MPC) is and on how much of the transfers are being spent
locally (how open the economy - here municipality - is). The higher the MPC and the less open
municipalities are, the higher will be the transfer multiplier.2 . In addition, how permanent transfers
2 Considering a corner example in which the households of a given municipality have a “home-bias” preference to

fully spend their transfers in locally-produced goods/services, then the transfer multiplier would be 1, similar to a
situation in which the government directly purchases goods/services from this municipality.

6
are also plays an important role, unless the share of hand-to-mouth households is very large. For
the US, Pennings (2021) finds a cross-regional transfers multiplier of 1.5 for permanent transfers
and 1/3 for one-off transfers.

There exists only limited empirical evidence on the cross-regional transfer multiplier in emerging
markets. Given the wide range of theoretically plausible multipliers, we see as one of the main
contributions of our paper that it provides a benchmark for the case of Brazil. It is also among
the first studies of the impact of Covid-related response programs on economic activity in emerging
markets. Last, we build on the literature by carefully considering the role of informality in obtaining
multipliers in an emerging market. Closely related studies include Sadoulet et al. (2001), who find a
multiplier range of 1.6-2.5 analyzing the PROCAMPO program in Mexico, and Egger et al. (2019),
who estimate a local cash transfer multiplier of 2.6 using novel experimental evidence from Kenya.
Using panel regressions Denes et al. (2018) estimate that the Bolsa Familia multiplier was as high
as 4 in Brazil in the 2004-2010 period.3

Corbi et al. (2019) estimate formal employment purchase multipliers at the municipal level in Brazil.
The authors exploit a discontinuity in the allocation of federal transfers to municipalities, based
on population size thresholds, to study the impact of externally financed municipal government
spending on formal employment over the period 1999-2014. They find a cost-per-annual-formal-job
of USD 8,000-13,000 and a preferred GDP multiplier when accounting for the informal sector of
2.4.4 This is at the very top of the range of direct spending (purchase) multipliers generally obtained
in cross-sectional studies for the US. Chodorow-Reich (2019) cite a multiplier range of 1-2.5, with a
preferred point estimate of 1.8, and Serrato and Wingender (2016) find a local multiplier between
1.7 and 2 using a Census shock to map expenditure changes that depend on the local population
size.

In all, the empirical evidence for emerging markets has so far shown large local multipliers, with
both purchase and transfer multipliers significantly above 1. The upper end of our estimated GDP
multiplier range (0.5-1.5) corresponds to estimates previously found in the literature while the
lower end lies substantially below most cross-sectional GDP multiplier estimates. Of course, the
exceptional nature of the Covid-19 pandemic might plausibly explain lower estimates. It seems
intuitive that forced social distancing and substantial restrictions on the supply of various services,
with a (likely associated) sharp increase in aggregate household savings, would lead to lower mul-
tipliers ceteris paribus5 . We nonetheless interpret our estimated local EA multiplier as providing
3 Estimating national transfer multipliers in emerging markets, Bracco et al. (2021) find a general multiplier of 0.9

for Latin America, compared to 0.3 for developed economies. The difference is mainly explained by the larger share
of hand-to-mouth households in EMs economies.Neri et al. (2013) find an implied GDP multiplier of Bolsa Familia
of 1.8.
4 Corbi et al. (2019) calibrate the productivity ratio of informal to formal workers to ρ = 0.55. Updating their

estimation using recent work by Ulyssea (2018), we obtain ρ = 0.81 and, thus, an implied GDP multiplier of 3.5.
5 Auerbach et al. (2020) provide evidence for the US that points in a similar direction. Looking at the average

7
a lower bound for what the corresponding national multiplier would be.6 Chodorow-Reich (2019)
for instance argues that local spending multipliers lay out a rough lower bound for the aggregated
national (deficit financed) spending multiplier for a closed economy when monetary policy is pas-
sive. Similar factors are at play in this study. Brazil is a relatively closed economy, and surely more
closed than its individual municipalities. In addition, monetary policy was accommodative in 2020.
When real rates do not rise in response to higher government spending, the standard multiplier
measured at the national level rises, but this effect is ”netted” out in cross-sectional regressions, as
all municipalities are equally affected by monetary policy.7

The remainder of this paper proceeds as follows. Section 2 provides more details on the EA program,
motivating the subsequent discussion of data and research design in sections 3 and 4. Section 5
presents the results and section 6 presents brief concluding considerations. Additional tables and
figures not presented in the main text can be found in the Appendix.

2 Background: The Brazilian Emergency Aid Cash Transfer
Program
The EA was a means tested program of monthly disbursements, which covered roughly 60 percent of
the total workforce in the initial months. It was initially also very generous, providing an estimated
replacement rate of 40 percent of the average income in the informal sector and increasing the real
household income of the bottom four deciles of the income distribution by 20 percent (at least in
May and June 2020) according to Flamini et al. (2021). The original design (Law no. 13.983, April
2, 2020) offered support during 2020Q2. Given the continued outbreak of COVID-19, however, the
authorities extended the program twice in 2020, first through end 2020Q3 and later for the whole
year (at that stage with tighter eligibility and a 50 percent reduction in stipends).

Eligibility. The EA initially offered a monthly basic income of BRL600 (and twice as much for
single parents) to all contributors to Brazil’s public social security system (INSS), participants of
the national single registry (Cadastro Único), citizens registered as individual micro-entrepreneurs
(MEI), and informal workers not registered in other federal assistance programs. In addition, Bolsa
Famı́lia 8 beneficiaries could temporarily migrate to the EA program. The eligibility age was 18
years or older. Participants must belong to a household with per capita monthly income of no
impact of Covid-related fiscal response measures in a large group of countries during 2020, Deb et al. (2021) also
find that demand policies (including, though not restricted to, cash transfers) had less impact on economic activity
during stringent lockdowns
6 A national EA multiplier cannot be obtained given the very short time period under consideration.
7 Chodorow-Reich (2019) considers the case when monetary policy response is restricted by a zero lower bound

scenario instead. Similarly, real rates fall in that scenario following the increase in government spending.
8 Bolsa Famı́lia was Brazil’s most important social assistance program. Prior to the pandemic it covered around

14 million households, paying a monthly benefit of less than BRL 200, at a total annual cost of 0.4 percent of GDP.

8
more than BRL 522 (half the minimum wage) or total monthly income up to BRL 3,135 (thrice
the minimum wage). Finally, participants could not have had annual taxable gross income greater
than BRL 28.5 thousand in 2018.

Financial inclusion. A vast number of EA participants did not have a banking account at the
time of the first transfer. Caixa Econômica Federal (CEF) thus offered digital banking accounts,
debit cards, and cell phone apps to include them in the system. Participants with accounts in other
institutions could decide which bank to use to withdraw the cash. Importantly, banks could not
withhold EA transfers to citizens with outstanding debt or past due overdraft accounts. The social
safety net got wider and digital.

Fiscal costs. The EA transferred approximately BRL 40 billion (1/2 percent of GDP) per month
to recipients during 2020Q2 and 2020Q3. Owing to tighter controls over claimants’ eligibility and
reduction in monthly stipend, the cost of EA transfers fell to half the amount in 2020Q4.

3 Data
Brazil is a federal republic with three levels of government - federal, state and municipal. The unit
of analysis in this paper are Brazil’s municipalities. Our main (independent) variable of interest
is the amount of Emergency Aid (EA) transfers disbursed by the federal government (Ministry of
Citizenship) to the public bank CEF, which in turn was responsible for disbursing payments to
individuals.

Ideally, we would like to use a measure of total municipal employment (or even a municipal GDP
proxy) as the dependent variable. However, no suitable GDP proxy is available, and total employ-
ment data (covering both the formal and informal sector) is not available at the municipal level,
even at annual frequency.9 The dependent variable is thus private formal employment as measured
by the government’s administrative CAGED dataset. To account for extreme outliers linked to
likely measurement problems, we exclude the top and bottom 0.5 percent of municipalities in terms
of employment growth and EA per capita.10 Below we briefly summarize the main data sources.
Table 1 presents summary statistics while Figure 2 shows correlations between the main variables.

Emergency Aid : Data on EA payments at the municipal level are provided by the Brazilian
Ministry of Citizenship. The first round of EA payments - six installments from April to Septem-
ber/2020 at the full original amount - represented 80 percent of total year-transfers (Figure1).
9 The continuous household survey PNAD provides data at the national and state level for both the formal and

informal sectors, but is not representative at the municipal level. Only the population census - conducted once every
ten years - provides a full picture of employment at the municipal level.
10 Regression results are not meaningfully different when outliers are retained, but parameter stability decreases

somewhat.

9
For each round, new beneficiaries were informed ex ante about the payment schedule of future
installments.

Figure 1: EA Disbursements Over Time and Municipal Heterogeneity

(a) Monthly Disbursements (b) Regional (municipal) Distribution

Formal Employment: CAGED is an administrative database that covers in principle the universe
of private formal workers in Brazil. There was a significant change in its methodology at the end
of 2019 and, thus, we chiefly focus on the series from 2020 onward to ensure consistency. We use
CAGED’s November 2021 vintage which offers the most up to date payroll series, after incorporating
ex-post notifications of hiring/layoffs and correcting methodological issues.

Mobility : The google mobility index tracks more than 2,400 Brazilian Municipalities. We use
the average of mobility indices for groceries and pharmacies, parks, retails and recreation, transit
stations, and workplaces.

COVID-19 Deaths: New daily Covid-19 related deaths per million were computed using the
administrative data from the Brazilian Ministry of Health.

2010 Census-based Indicators: Informality at the municipal level is defined as the ratio of
informal employment to total employment (considering the occupied population over 10 years of
age). The share of non-white population covers all individuals that did not declare themselves as
white or Asian. Urbanization is calculated as the ratio of urban population to total population in
each municipality. Agriculture as a share of total employment captures employment in agriculture
as a share of total employment. Last, services as a share of total employment were computed
taking into account the sectors outside of the perimeter of agriculture, manufacturing, and public
administration.

10
Bolsa Familia Recipients: The share of the population receiving Bolsa Familia benefits in
each municipality was calculated using administrative data from the Ministry of Citizenship, as of
December 2019.

Job Protection Program: The Emergency Benefit for Preserving Employment and Income
(BEm) was launched in April 2020, allowing for temporary reductions in working hours or con-
tract suspension in the formal sector, by mutual agreement between employers and employees. The
program backed workers by partially compensating for the associated salary losses, in an amount
proportional to the unemployment insurance to which the employee would have been entitled to
if she lost the job (i.e., pro-rated by the percentage reduction in working hours). Importantly,
each employee could have more than one BEm agreement, either because she worked for more than
one firm or for agreeing first to a cut in working hours and later to a contract suspension, or vice
versa. The BEm data, from the Ministry of Economy, shows the number of monthly agreements
aggregated at the municipal level.

Pandemic-related credit to SMEs: Besides the EA and BEm, pandemic-related lending facili-
ties were the third largest pillar of the government’s fiscal response. Among those, the Emergency
Program for Access to Credit (PEAC), launched in June 2020, was the largest pandemic-related
credit line to the real sector (≈ 1.2 of the GDP). The program unlocked credit to SMEs, associ-
ations, private foundations, and cooperatives with underlying 2019 revenues between BRL360,000
and BRL300 million. The federal government provided guarantees to the credit lines, covering 80
percent of the face value of each operation. The PEAC data, from the Brazilian Development Bank
(BNDES), shows monthly disbursements aggregated at the municipal level.

11
Table 1: Summary Statistics

Statistic N Mean St. Dev. Min Median Max

∆ Formal Employment per Thousand 5,483 −0.4 13 −203.6 −0.3 348
Emergency cash transfers (EA) per capita 5,483 1,120 322 212 1,118 2,016
Share of Bolsa Familia Receivers 5,483 9 6 0 7 27
Share of non-white Population 5,472 52 24 1 57 99
Bolsa Familia (BF) Cover Ratio 5,472 1.08 0.23 0.22 1.09 2.90
Job Agreements per Formal Worker 5,333 0.31 0.28 0 0.23 9
PEAC Disbursements per capita 3,843 154 314 0.0 44 6,309
∆ Covid-19 Deaths per million 5,483 2 2 0 2 18
Urbanization Rate 5,479 64 22 4 65 100
Agriculture as Share of Employment 5,479 35 18 0 35 87
Services as Share of Employment 5,479 53 12 8 53 87
Informality Rate 5,478 64 17 16 67 97
GDP per capita (Thousand) 5,483 23 24 5 17 575
Population (Thousand) 5,483 38 225 1 12 12,325

Note: The unit of analysis is municipalities, the lowest administrative level in Brazil. Data on formal employment
creation, EA cash transfers, and number of job protection agreements under the BEm program refer to the sum
over April-September 2020. The share of the population receiving Bolsa Familia benefits in each municipality was
calculated using administrative data from the Ministry of Citizenship, as of December 2019. The Bolsa Familia
cover ratio is defined/estimated as the number of poor households receiving Bolsa Familia benefits as of end-2012
as a share of the estimated number of poor households according to the 2010 Census. Covid-19 deaths are measured
as the average of daily new deaths over April -September 2020. GDP per capita is taken from the 2018 municipal
accounts. At the average 2020 exchange rate to the USD, mean GDP per capita of BRL 23,000 corresponds to about
USD 4,000. Population figures are collected from the Brazilian Institute for Geography and Economics (IBGE). We
exclude the top and bottom 0.5 percent of municipalities in terms of the change in formal employment and emergency
aid per capita.

4 Research Design

4.1 Baseline Econometric Specification and Identification
To estimate the cross-sectional formal employment multiplier we follow a standard approach, as
set out for example in Chodorow-Reich (2019). Specifically, we regress the total change in private
formal employment per capita between April and September 2020 on total EA disbursements per
capita at the municipal level over the same period - which is the period with the bulk of EA
disbursements.11
11 In robustness exercises we change the window of analysis to the initial disbursement period (April-June) or the

full 9-month period of EA disbursements (April-December).

12
By design, EA disbursements per capita are determined by the share of the local population which
is eligible for the program. Given that eligibility at the individual (and, thus, municipal) level
was determined based on data as of end-2019, it could be viewed as largely exogenous. However,
selection into or out of the program can give rise to endogeneity concerns. In particular, households
might either opt to not seek the EA even when eligible and non-eligible households could find a
way around the proposed targeting, with the latter having occurred especially at the beginning of
the program according to reports. We thus instrument for EA disbursements with the share of
the population receiving conditional cash transfers under the Bolsa Familia program pre-pandemic.
As explained above, being a Bolsa Familia recipient is one of the criteria for being eligible for EA
disbursements. Figure 2 presents the correlations between the main variables used in the analysis.
As a first takeaway, we note that there is a high correlation between EA disbursements per capita
and the share of pre-pandemic Bolsa Familia recipients at the municipal level, already suggesting
that the inclusion restriction for the instrument is likely to be satisfied.

There is a body of literature - e.g. Barrientos et al. (2016) and Ribas et al (2011)- pointing out the
exogenous features of Bolsa Familia figures at the municipal level. According to Barrientos et al.
(2016), even though the selection of Bolsa Familia receivers depends on their per capita household
income, the placement at the municipal level depends mostly on the pre-programme poverty level of
the municipality. As such, the pre-pandemic Bolsa Familia program assignment can be considered
exogenous concerning the EA transfers and control variables.

We estimate the following set of equations:

h
X h
X
0 m
f em s
t =α +µ+β eaˆm
t +γ X + εm (1)
t=0 t=0
h
X
eaˆm S
t = α + φ0 + φ1 BF
m
+ φ02 X m + ξ m (2)
t=0

where αs and αS are state fixed effects, f emt denotes the monthly change in formal employment
per capita at time t for municipality m, and eaˆm t indicates EA transfers per capita. BF
m
stands
m
for the share of individuals entitled to Bolsa Familia benefits pre-pandemic, X represents the
set of control variables which in the baseline specification comprises the daily new COVID-19
deaths per million and formal employment trends (by taking into account the change in formal
employment in Q1 2020 and in 2019). εm and ξ m are robust standard errors. Last, we control for
the municipal informality rate, given the direct link between formal job creation and the structural
level of informality.

13
Figure 2: Correlation Matrix

Note: The matrix shows correlations for the variables summarized in Table 1. Correlations marked
with “x” are not significant at 95 percent confidence levels.

5 Results
5.1 The (Formal) Employment Multiplier
Table 2 presents the results from the baseline specification (column (1)) as well as a number of
robustness exercises and extensions. Specifically, column (2) adds the share of services in total
employment as a control variable, column (3) controls for the urbanization rate, and column (4)
includes google mobility as a control variable. The first stage F-statistic is highly significant across
specifications (Table 8 in the Appendix shows the first stage results).

In the baseline specification we find a formal employment multiplier of 0.53 - in other words 100,000
of EA payments create 0.53 private formal jobs for a 6 month period. This implies an annual cost-
per-formal-job of BRL 378,813 =(100,000/0.528)*2. At the average 2020 USD-BRL exchange rate,
this corresponds to a cost per job of around USD 73,000. Comparing this with the simple OLS
results presented in Table 7 of the Appendix, we see that the 2SLS multiplier estimate is around

14
three times as large as the OLS one. This is consistent with the bias discussed in section 4.1,
whereby worse pandemic outcomes and thus worse labor market outcomes would lead to higher EA
per capita.

It is instructive to map the estimated coefficient for the formal employment multiplier to a total
employment multiplier. This will allow to gain a fuller picture, facilitate comparisons with the
literature, and obtain a GDP multiplier estimate, which we’ll discuss in section 5.2.

The private (formal) employment multiplier βF E is defined in the standard way as the change
in private formal employment dF E for a given change in government spending dG . The total
employment multiplier is defined equivalently and the expression can be rearranged as follows

dE
dE = βE dG =⇒ βE = dEGt (3)
Et

where Et denotes the total number of jobs. Multiplying and dividing by FdFEEt and collecting terms
gives

1
βE = ΘβF E (4)
ωf e

where Θ is the elasticity of total employment to private formal employment and ωf e is the formality
rate. The ratio ω1f e aims to adjust the private formal employment multiplier by its relative size
to total employment, considering the informal and public (formal) employment shares. We recover
ωf e = 0.45 from the 2010 census data, which has the most comprehensive account of employment
formality (using the number from the household survey PNAD12 , 0.46, would not make a big
difference, however), yielding an adjustment factor of ω1f e =2.22.

The elasticity of total employment to private formal employment, Θ, can be expressed as the sum
of three underlying components:

Θ = ωf e + Ψωif + ηωpe (5)
.

ωif = 0.5 is the informality share and ωpe = 0.05 is the share of public employment. Additionally,
we take into account the elasticity of informal employment to private formal employment, denoted
by Ψ = 2.3513 and obtained from PNAD (which includes monthly data on both formal and informal
12 We do not rely on PNAD data for our regressions given that the PNAD survey is not representative at the

municipal level).
13 From 2019 to 2020 (after the labor reform of 2017/2018 was implemented) the mean of Ψ has been relatively

stable around 2.4, with an underlying coefficient of variation close to 0.8

15
employment) for the same 6-month window used in the regressions (April-September 2020). We
assume that the elasticity of public employment to changes in private formal employment, given
by η, is zero considering that the public sector dynamics were most likely driven by the healthcare
response and, thus, orthogonal to EA transfers. As a result, we find that Θ equals to 1.63.

Multiplying the estimated βF E as set out in equation (4) yields a total employment multiplier of
1.6 for the 6-month window. In other words, around 1.6 jobs (0.5 formal and 1.1 informal) were
generated for each BRL 100,000 of paid EA. Equivalently, the annual cost-per-job is around BRL
103,000, or USD 20,000.

Sectoral employment structure as a possible confounder While the baseline specification
controls for state fixed effects and important structural municipal characteristics such as informality,
one additional concern might be that a different local economic structure (for example in terms
of sectoral composition) is correlated with the share of pre-pandemic Bolsa Familia recipients,
while also impacting the sensitivity of formal employment to the pandemic shock. One plausible
mechanism might be that structurally poorer municipalities have a larger services sector which
in turn suffered more during the pandemic. At first sight the correlations shown in Figure 2 do
not suggest that this is a particularly pronounced correlation, but in a robustness exercise we
nevertheless include the share of services employment as a control14 . In addition, we control for the
urbanization rate; another potentially important structural municipal characteristic which could
impact how the pandemic affected employment creation.

Looking across columns (2) and (3) of Table 2, the estimated formal employment multiplier drops
marginally when adding service employment as a control variable and drops somewhat more to 0.44
when also adding the urbanization rate, but remains significant. Both additional control variables
have the expected sign, with more services-intensive and urbanized municipalities experiencing
weaker job growth.

Controlling for Google Mobility Google mobility data - for which we would like to control given
the pandemic-induced variability in mobility and, hence, economic activity - is only available for
2,210 (slightly less than half of all) municipalities. However, these 2,210 municipalities account for
89 percent of national GDP, 93 percent of total formal employment, and 77 percent of total EA
disbursements. Including mobility in the regressions thus has two effects - (i) a composition effect,
whereby we exclude municipalities which are on average smaller, poorer, more informal, more rural
and more dependent on agriculture, and (ii) the direct impact of adding the control variable for a
constant sample.

The change in the estimated multiplier is large when including Google mobility data, roughly dou-
bling in the restricted sample (column (4)). As mentioned above, municipalities in the sample for
14 We also ran exercises controlling for employment in industry and in commodity-sensitive sectors, which yielded

similar results to the ones shown in Table 2.

16
which mobility data is available are larger and richer than those without mobility data. The change
in the coefficient is entirely due to this composition effect and not because of the importance of
mobility as a control variable. To formally test this, we re-run the regressions with the reduced
sample but without mobility as a control, concluding that this only marginally changes the coeffi-
cient. The coefficient on mobility has the expected sign, with higher mobility associated with larger
formal employment creation.

Intuitively, whether mobility data, as captured by Google, is available or not for a specific munic-
ipality suggests important differences in its level of development or other characteristics, beyond
the control variables we include. These differences could be perhaps along dimensions we cannot
observe such as internet connectivity, smart phone usage and related factors which could deter-
mine how much remote work, for example, is feasible. We investigated the difference in coefficients
between the full and restricted sample further (including by controlling for additional observable
variables which differ between the two groups) but could not obtain a clear explanation - beyond
the hypothesis that there might be a true difference in treatment effect based on some unobserv-
able sample characteristic. We continue to refer to the results for the full sample as the baseline,
but given that the restricted sample accounts for around 90 percent of national GDP and formal
employment we do not discard the higher estimates it yields.

Interacting pre-pandemic informality with EA As mentioned above, there exists a direct
relationship between the informality rate in a municipality and the number of formal jobs created
per capita over any time period - a marginal formal job in a highly informal municipality likely
requires a larger change in local economic activity. This is an important concern for our research
design, which might be even more important during the pandemic since informal and formal jobs
were affected at varying degrees by lockdowns and social distancing. To allow for a difference
in treatment effect between more and less formal municipalities we include an interaction term
between the pre-pandemic informality rate and the EA disbursement per capita (instrumented by
the interaction between the pre-pandemic share of Bolsa Familia recipients and the pre-pandemic
informality rate).

Table 3 repeats the order of Table 2 in terms of the control variables included in each column
but adds the interaction term between EA and pre-pandemic informality to all specifications. The
national formal employment multiplier is obtained by taking a weighted average of the sum of
the main effect and interaction effect across all municipalities (using the municipalities’ share in
national formal employment as the weights).

The coefficient on the interaction term is highly significant and negative, indicating that more
informal municipalities create less formal jobs for each BRL 100,000 of EA payment - an intuitive
result. More interestingly, the average weighted formal employment multipliers we obtain are
significantly higher than the ones in Table 2. The fact that more formal municipalities with a

17
Table 2: Baseline Regression Results

Cumulative Change in Formal Employment per capita

Instrument: Share of Population Receiving Bolsa Familia (Pre-pandemic)

Baseline Adding Services Empl. Adding Urbanization Rate Adding Mobility

(1) (2) (3) (4)
∗∗∗ ∗∗∗ ∗∗∗
EA per capita (BRL 100K) 0.528 0.443 0.415 0.666∗∗
(0.138) (0.138) (0.135) (0.267)
Covid-19 Deaths 0.002 0.063 0.123 0.261
(0.115) (0.115) (0.112) (0.185)
Informality Rate −0.0005 −0.0001 −0.006∗∗ −0.010∗∗∗
(0.002) (0.002) (0.003) (0.004)
∆ Formal Employment Q12020 −0.436∗∗ −0.439∗∗ −0.436∗∗ −0.270
(0.220) (0.218) (0.218) (0.236)
∆ Formal Employment 2019 0.007∗∗∗ 0.004∗ 0.003 0.002
(0.002) (0.002) (0.002) (0.006)
Share of Services in Employment −0.011∗∗∗ −0.008∗∗∗ 0.002
18 (0.002) (0.002) (0.003)
Urbanization Rate −0.007∗∗∗ −0.010∗∗∗
(0.001) (0.002)
Overall Mobility 0.026∗∗∗
(0.003)
Constant −0.007∗∗∗ 0.0003 0.006∗∗∗ 0.005
(0.002) (0.002) (0.002) (0.003)

States Fixed effects Y es Y es Y es Y es
Implied Number of Year-Formal Jobs Created 633,557 531,614 498,444 799,007
Implied Cost per Year-Formal Job (BRL) 378,813 451,454 481,497 300,372
Implied Cost per Year-Formal Job (USD) 73,413 87,491 93,313 58,211
Implied Number of Year-Jobs (Formal and Informal) Created 2,287,845 1,919,720 1,799,939 2,885,306
Implied Cost per Year-Job (USD) 20,329 24,228 25,840 16,120
First Stage F statistic 417∗∗∗ 406∗∗∗ 400∗∗∗ 91∗∗∗
Wu-Hausman 12.53∗∗∗ 8.53∗∗∗ 6∗∗ 2
Observations 5,478 5,478 5,478 2,210
Residual Std. Error 0.013 (df = 5446) 0.013 (df = 5445) 0.013 (df = 5444) 0.011 (df = 2175)

∗
p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Note: The unit of analysis is municipalities, the lowest administrative level in Brazil. Data on formal employment creation, and EA transfers refer to the sum over April-September
2020. Google mobility and Covid-19 deaths are measured as the average over April-September 2020. The share of population receiving Bolsa Familia is as of December/2019.
higher multiplier also account for a large share of total formal employment, leads to a national
cross-sectional multiplier which is larger than the simple average relationship implied. Figure 3
makes this point visually by showing that municipalities with a higher multiplier also have a larger
weight. Columns (1)-(4) show private formal employment multipliers of 1.5-1.7, implying an annual
cost per formal job of BRL 132,000-117,000 (USD 23,000-26,000). The implied total employment
multipliers are in the range of 5-6, with an annual cost per job as low as BRL 30,000 (around USD
6,000).

Instrumenting EA transfers using municipal ethnic patterns and BF cover ratio As
an additional robustness exercise, we use the share of non-white population (NWP) measured by
the 2010 Census and the estimated Bolsa Familia cover ratio of poor households to instrument EA
transfers per capita15 . Historically, the NWP has limited access to higher education in Brazil16 and
thus larger unemployment rates than white Brazilians. Figure 2 shows a high correlation between
EA disbursements per capita and the share of non-white population at the municipal signalling the
potential strength of the instrument. Aiming to capture some idiosyncratic components of municipal
poverty features that are fully mapped by ethnics patterns, we use as an additional instrument the
estimated Bolsa Familia cover ratio of poor households. The ratio takes into account the number
of Bolsa Familia beneficiaries at end-2012 and the estimated number of poor Households according
to 2010 Census17 .

Table 2.A is built such as Table 2 with the difference that EA transfers per capita are instrumented
by the share of NWP instead of the share of Bolsa Familia receivers. The formal employment
multiplier is, on average, twice as large as that detailed in Table 2. Moreover, the results shown
by Table 2.A are in line with Table 3, pointing to a multiplier around 1.2. The tests reported in
Table 2.A provide evidence of the strength of our instruments, indicating absence of endogeneity
or over-identification issues.

Interacting pre-pandemic informality with EA using alternative instruments As high-
lighted before, it is useful to estimate the formal employment multiplier directly taking into account
the formality level of different municipalities. Thus, like in Table 3, we add to Table 2.A’s specifi-
cations an interaction term between the pre-pandemic informality rate and the EA disbursements
per capita (instrumented by the interaction between the pre-pandemic share of NWP and the
pre-pandemic informality rate, keeping the instruments already used in Table 2.A18 ).
15 We also considered using either NWP or the BF cover ratio as single instrument to the EA transfers. However,

considering the interaction with informality, the combined use of NWP and BF cover outperformed the isolated
specifications as a feasible instrument.
16 Mello (2021) provides a detailed discussion on this topic.
17 Rougier et al (2018) argue that the municipal-level cover ratio of Bolsa Familia is a good indicator of the

municipality’s capacity to identify, enroll and register poor individuals. As a result, the authors claim that the ratio
is a good exogenous predictor of the cross-municipality variation of Bolsa Familia to GDP ratio.
18 We ran regressions instrumenting the interaction of informality with EA by the interaction of the BF cover

19
Similar to Table 3, Table 3.A shows the national formal employment multiplier taking a weighted
average of the sum of the main effect and interaction effect across all municipalities. On average,
the obtained formal employment multiplier is 20 percent larger than in Table 3.

Controlling for the government’s Job Protection Program (BEm) and pandemic-related
credit to SMEs (PEAC) As discussed in section 3, BEm was explicitly designed to limit formal
sector job losses by allowing for a flexible reduction in working hours with partial income com-
pensation. Trying to explicitly identify the impact of BEm on formal employment runs into an
endogeneity problem that is significantly worse still than that of assessing the impact of the EA
- municipalities in which the pandemic had a large impact would have seen higher selection into
BEm participation. At the same time, the direct and mechanical impact of the BEm has to be
an improvement in formal job dynamics relative to a counterfactual without BEm. PEAC credit
lines face a similar endogeneity issue as the selection into the program was likely impacted by local
economic conditions.

Given this difficulty, we do not aim to retrieve the impact of BEm and PEAC on employment
(even though it is a very interesting question in its own right) and therefore did not include them as
regressors in our baseline specifications. Nevertheless, given the direct link between BEm and formal
employment and the impact of pandemic-related credit on formal employment, we investigate how
our results change when BEm and PEAC are indeed controlled for. We opted for a specification in
which BEm is normalized by total formal employment - in essence, the share of jobs protected by
the program.

Table 4 adds the share of formal jobs covered by BEm as a control variable: column (1) adds the
BEm variable to a specification otherwise identical to column (1) of Table 2, while columns (2),
(3), and(4) do the same for column (1) of Tables 3, 2.A, and 3.A respectively.

The estimated formal employment multiplier drops significantly relative to the baseline without
BEm and PEAC, by over 40 percent in the specification without the informality interaction (col-
umn (1)), by around 25 percent in the version with the informality interaction (column (2)) and
when instrumenting EA transfers by ethnic patterns (column(3)), and about 10 percent in the
specification instrumenting EA transfers by ethnic patterns and the BF cover ratio (column (4)).
The specifications with BEm and PEAC as a control variables thus provide us with the lower end
of our estimated multiplier range. Note that the (non-causal) coefficient of the BEm variable is
highly significant and negative, suggesting that the negative selection effect dominates the positive
mechanical association with formal employment retention.

ratio with the informality rate. The implied formal employment multiplier is about 25 percent larger but in all
specifications we reject the null hypothesis of the Sargan Test that the model is not over-identified.

20
Figure 3: Formal Employment Multiplier and Formal Employment Weight by Municipalities

21
Table 3: Interaction with Informality Regression Results

Cumulative Change in Formal Employment per capita

Instruments: Pre-pandemic Share of Population Receiving Bolsa Familia (BF) and BF*informality

Baseline Adding Services Empl. Adding Urbanization Rate Adding Mobility

(1) (2) (3) (4)
∗∗∗ ∗∗∗ ∗∗∗
EA per capita (BRL 100K) 3.118 3.096 3.003 3.380∗∗∗
(0.434) (0.433) (0.425) (0.648)
EA per capita (BRL 100K)*Informality −3.491∗∗∗ −3.594∗∗∗ −3.502∗∗∗ −3.919∗∗∗
(0.520) (0.522) (0.517) (0.843)
Covid-19 Deaths −0.152 −0.087 −0.026 −0.018
(0.118) (0.118) (0.115) (0.195)
Informality Rate 0.033∗∗∗ 0.034∗∗∗ 0.028∗∗∗ 0.026∗∗∗
(0.005) (0.005) (0.005) (0.008)
∆ Formal Employment Q12020 −0.434∗∗ −0.432∗∗ −0.257
(0.221) (0.218) (0.218) (0.240)
∆ Formal Employment 2019 0.005∗∗ 0.001 0.0003 −0.003
22 (0.002) (0.002) (0.002) (0.006)
Share of Services in Employment −0.012∗∗∗ −0.010∗∗∗ −0.003
(0.002) (0.002) (0.003)
Urbanization Rate −0.006∗∗∗ −0.009∗∗∗
(0.001) (0.002)
Overall Mobility 0.026∗∗∗
(0.003)
Constant −0.031∗∗∗ −0.024∗∗∗ −0.017∗∗∗ −0.016∗∗∗
(0.004) (0.004) (0.004) (0.005)

States Fixed effects Y es Y es Y es Y es
Formal Employment Multiplier - Weighted Average Across Munis 1.624∗∗∗ 1.558∗∗∗ 1.504∗∗∗ 1.703∗∗∗
(0.497) (0.499) (0.487) (0.643)
Implied Number of Year-Formal Jobs Created 1,949,041 1,870,596 1,805,242 2,044,448
Implied Cost per Year-Formal Job (BRL) 123,137 128,301 132,946 117,391
Implied Cost per Year-Formal (USD) 23,863 24,864 25,764 22,750
Implied Number of Year-Jobs (Formal and Informal) Created 7,038,204 6,754,930 6,518,928 7,382,730
Implied Cost per Year-Job (USD) 6,608 6,885 7,134 6,300
First Stage F statistic (EA) 1259∗∗∗ 1272∗∗∗ 1310∗∗∗ 201∗∗∗
First Stage F statistic (EA*informality) 1531∗∗∗ 1568∗∗∗ 2122∗∗∗ 196∗∗∗
Wu-Hausman 28∗∗∗ 23∗∗∗ 21∗∗∗ 18∗∗∗
Observations 5,478 5,478 5,478 2,210
Residual Std. Error 0.013 (df = 5445) 0.013 (df = 5444) 0.013 (df = 5443) 0.011 (df = 2174)

∗
p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Note: The unit of analysis is municipalities, the lowest administrative level in Brazil. Data on formal employment creation, and EA transfers refer to the sum over April-September
2020. Google mobility and Covid-19 deaths are measured as the average over April-September 2020. The share of population receiving the Bolsa Familia benefit is as of December/2019.
To compute the national formal employment multiplier, we calculate individually for all municipalities the implied effect for their specific level of informality and then calculate the
average by weighting with formal employment.
Table 2.A: Regression Results Using Ethnic Patterns and BF cover as Instruments

Cumulative Change in Formal Employment per capita

Instruments: Non-white Share of the Population and BF Cover

Baseline Adding Services Empl. Adding Urbanization Rate Adding Mobility

(1) (2) (3) (4)
∗∗∗ ∗∗∗ ∗∗∗
EA per capita (BRL 100K) 1.312 1.263 1.217 1.158∗∗
(0.302) (0.300) (0.296) (0.525)
Covid-19 Deaths −0.054 0.003 0.066 0.170
(0.110) (0.110) (0.109) (0.214)
Informality Rate −0.005∗ −0.005∗ −0.011∗∗∗ −0.014∗∗
(0.003) (0.003) (0.003) (0.006)
∆ Formal Employment Q12020 −0.548∗∗∗ −0.551∗∗∗ −0.547∗∗∗ −0.274
(0.171) (0.169) (0.169) (0.241)
pre e 2019 0.006∗∗∗ 0.002 0.001 0.002
∆ Formal Employment 2019 0.006∗∗∗ 0.002 0.001 0.002
(0.002) (0.002) (0.002) (0.006)
23 Share of Services in Employment −0.011∗∗∗ −0.008∗∗∗ 0.001
(0.002) (0.002) (0.003)
Urbanization Rate −0.007∗∗∗ −0.010∗∗∗
(0.001) (0.002)
Overall Mobility 0.028∗∗∗
(0.004)
Constant −0.012∗∗∗ −0.005∗∗ 0.001 0.004
(0.002) (0.002) (0.002) (0.004)

States Fixed effects Y es Y es Y es Y es
Implied Number of Year-Formal Jobs Created 1,574,509 1,515,295 1,460,698 1,389,937
Implied Cost per Year-Formal Job (BRL) 152,428 158,385 164,305 172,669
Implied Cost per Year-Formal Job (USD) 29,540 30,694 31,842 33,463
Implied Number of Year-Jobs (Formal and Informal) Created 5,685,727 5,471,898 5,274,744 5,019,216
Implied Cost per Year-Job (USD) 8,180 8,500 8,817 9,266
First Stage F statistic 198∗∗∗ 198∗∗∗ 201∗∗∗ 68∗∗∗
Wu-Hausman 18∗∗∗ 17∗∗∗ 15∗∗∗ 4 ∗∗

∗ ∗
Sargan 2.86 1.32 3.3 1
Observations 5,472 5,472 5,472 2,208
Residual Std. Error 0.013 (df = 5440) 0.013 (df = 5439) 0.012 (df = 5438) 0.011 (df = 2173)

Cumulative Change in Formal Employment per capita

Instruments: Non-white Share of Population(NWP), BF cover and NWP*Informality

Baseline Adding Services Empl. Adding Urbanization Rate Adding Mobility

(1) (2) (3) (4)
∗∗∗ ∗∗∗ ∗∗∗
EA per capita (BRL 100K) 4.073 4.281 4.056 3.554∗∗∗
(0.679) (0.680) (0.672) (0.926)
EA per capita (BRL 100K)*Informality −4.350∗∗∗ −4.769∗∗∗ −4.480∗∗∗ −4.071∗∗∗
(0.761) (0.768) (0.760) (1.074)
Covid-19 Deaths −0.188 −0.133 −0.070 −0.044
(0.117) (0.116) (0.115) (0.229)
Informality Rate 0.040∗∗∗ 0.045∗∗∗ 0.037∗∗∗ 0.027∗∗
(0.007) (0.007) (0.007) (0.011)
∆ Formal Employment Q12020 −0.543∗∗∗ −0.546∗∗∗ −0.543∗∗∗ −0.263
(0.172) (0.169) (0.170) (0.243)
∆ Formal Employment 2019 0.003 −0.001 −0.002 −0.003
24 (0.002) (0.002) (0.002) (0.006)
Share of Services in Employment −0.013∗∗∗ −0.010∗∗∗ −0.004
(0.002) (0.002) (0.004)
Urbanization Rate −0.006∗∗∗ −0.009∗∗∗
(0.001) (0.002)
Overall Mobility 0.026∗∗∗
(0.004)
Constant −0.040∗∗∗ −0.034∗∗∗ −0.027∗∗∗ −0.017∗∗∗
(0.006) (0.006) (0.006) (0.006)

States Fixed effects Y es Y es Y es Y es
Formal Employment Multiplier - Weighted Average Across Munis 2.212∗∗∗ 2.240∗∗∗ 2.139∗∗∗ 1.812∗∗∗
Implied Number of Year-Formal Jobs Created 2,654,545 2,688,926 2,566,991 2,174,648
Implied Cost per Year-Formal Job (BRL) 90,410 89,254 93,494 110,362
Implied Cost per Year-Formal (USD) 17,521 17,297 18,119 21,388
Implied Number of Year-Jobs (Formal and Informal) Created 9,585,858 9,710,012 9,269,689 7,852,897
Implied Cost per Year-Job (USD) 4,852 4,790 5,017 5,922
First Stage F statistic (EA) 132∗∗∗ 132∗∗∗ 134∗∗∗ 47∗∗∗
First Stage F statistic (EA*informality) 187∗∗∗ 185∗∗∗ 181∗∗∗ 52∗∗∗
Wu-Hausman 19∗∗∗ 19∗∗∗ 17∗∗∗ 9∗∗∗
Sargan 3.7∗ 1.8 3.6∗ 0.9
Observations 5,472 5,472 5,472 2,208
Residual Std. Error 0.013 (df = 5439) 0.013 (df = 5438) 0.012 (df = 5437) 0.011 (df = 2172)

∗
p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Note: The unit of analysis is municipalities, the lowest administrative level in Brazil. Data on formal employment creation, and EA transfers refer to the sum over April-September
2020. Google mobility and Covid-19 deaths are measured as the average over April-September 2020. The share of population receiving the Bolsa Familia benefit is as of December/2019.
To compute the national formal employment multiplier, we calculate individually for all municipalities the implied effect for their specific level of informality and then calculate the
average by weighting with formal employment.
Table 4: Controlling for The Job Support Program (BEm) and Pandemic-related Lending (PEAC) Regression Results

Cumulative Change in Formal Employment per capita

Instruments:
Share of Population Receiving BF BF and BF*informality Non-white Share of Population(NWP) and BF cover NWP, BF cover, and NWP*Informality

(1) (2) (3) (4)
∗∗∗ ∗∗∗
EA per capita (BRL 100K) 0.260 2.468 1.265 3.420∗∗∗
(0.198) (0.484) (0.403) (0.851)
EA*Informality −2.928∗∗∗ −3.484∗∗∗
(0.597) (0.912)
Number of BEm Agreements per formal worker −0.010∗∗∗ −0.009∗∗∗ −0.009∗∗∗ −0.009∗∗∗
(0.002) (0.002) (0.002) (0.002)
PEAC per capita (BRL 100K) 0.132 0.203 0.193 0.249
(0.176) (0.176) (0.172) (0.175)
Covid-19 Deaths 0.122 −0.058 −0.022 −0.166
(0.143) (0.146) (0.160) (0.176)
∆ Formal Employment Q12020 −0.120 −0.113 −0.115 −0.112
(0.183) (0.184) (0.187) (0.187)
∆ Formal Employment 2019 0.007∗ 0.004 0.006 0.003
(0.004) (0.004) (0.004) (0.004)
Constant −0.003 −0.023∗∗∗ −0.009∗∗∗ −0.030∗∗∗
(0.002) (0.004) (0.003) (0.007)
25
States Fixed effects Y es Y es Y es Y es
Formal Employment Multiplier - Weighted Average Across Munis 0.260 1.218∗∗∗ 1.264∗∗∗ 1.930∗∗∗
(0.141) (0.410) (0.292) (0.675)
Implied Number of Year-Formal Jobs Created 311,474 1,461,114 1,516,800 2,315,573
Implied Cost per Year-Formal Job (BRL) 770,529 164,258 158,227 103,646
Implied Cost per Year-Formal Job (USD) 149,327 31,832 30,664 20,086
Implied Number of Year-Jobs (Formal and Informal) Created 1,124,767 5,276,245 5,477,333 8,361,791
Implied Cost per Year-Job (USD) 41,352 8,815 8,491 5,562
First Stage F statistic (EA) 1125∗∗∗ 619∗∗∗ 186∗∗∗ 77∗∗∗
First Stage F statistic (EA*Informality) 789∗∗∗ 125∗∗∗
Wu-Hausman 2 22∗∗∗ 11∗∗∗ 10∗∗∗
Sargan 2 2
Observations 3,789 3,789 3,785 3,785
Residual Std. Error 0.012 (df = 3755) 0.012 (df = 3754) 0.012 (df = 3751) 0.012 (df = 3750)

Our approach to obtain a GDP multiplier from the estimated (formal) employment multiplier closely
follows Chodorow-Reich (2019), adjusting for the fact that we estimate a private formal employment
multiplier rather than a total employment multiplier. The adjustment for the formal/informal
dimension takes inspiration from the approach taken in Corbi et al. (2019).

Starting from the definition of formal employment multiplier, we derive (Y denotes the level of
GDP and F E denotes the level of formal employment):

dF E Yt dG
dF E = βF E dG =⇒ = βF E (6)
F Et F Et Yt

The GDP multiplier βY , in turn, is equivalently defined as:

dY = βY dG (7)

which can be rearranged as

dY
Yt
βY = dG
(8)
Yt

Now consider a production function Yt = A(Nt Lt )1−α , where L denotes the stock of effective units
of labor, and N indicates hours worked per worker. We define L = F E + ρIE + P E where IE
equals informal employment, P E indicates public employment, and ρ is the relative productivity
ratio of informal to formal workers. After totally differentiating the production function we obtain:

dY dN dL
=( + )(1 − α) (9)
Yt Nt Lt

Dividing (9) by FdFEEt and rearranging terms yields

dY F Et
= (χ + ΘL )(1 − α) (10)
Yt dF E

where χ is the elasticity of hours per worker (N ) to private formal employment (F E), and ΘL is
the elasticity of effective labor units (L) to F E, which is formally given by

ΘL = ωfLe + ρΨωif
L L
+ ηωpe . (11)

26
Ψ and η are the elasticities of informal and public employment, respectively, to F E, while ωfLe , ωif
L

L
and ωpe are the shares of private formal, informal, and public employment in effective labor.

Equation (11) is similar to equation (5), which defined the elasticity of total employment to formal
employment Θ. However, since ΘL maps the response of effective labor units, total employment
is ’normalized’ by the productivity ratio of informal to formal workers (ρ). Furthermore, we can
rewrite the parameters ωfLe , ωif
L L
and ωpe as a function of their corresponding shares in total em-
ployment, as follows:

Et 1
ωfLe = ωf e = ωf e (12)
Lt 1 − (1 − ρ)ωif

L Et ρ
ωif = ρωif = ωif (13)
Lt 1 − (1 − ρ)ωif

L Et 1
ωpe = ωpe = ωpe (14)
Lt 1 − (1 − ρ)ωif

When ρ equals to 1, the underlying weight components of Lt are equivalent to their respective
shares in total employment given that Et = Lt , and, thus, Θ = ΘL .

Plugging in equation (11) into (10) yields

dY F Et
= (χ + ωfLe + ρΨωif
L L
+ ηωpe )(1 − α) (15)
Yt dF E

Multiplying and dividing (8) by FdFEEt and finally combining (15), (6), and (8), yields:

Yt
βY = (1 − α)(χ + ωfLe + ρΨωif
L L
+ ηωpe ) βF E (16)
F Et

For a given private formal employment multiplier βF E , and initial ratio of output per private formal
worker FYEt t , the GDP multiplier βY increases with the labor share (1 − α), with the elasticity of
hours worked, informal employment or public employment to F E, and/or when private formal
employment has a larger relative weight in effective labor L, either directly (higher share in total
employment) or due to higher relative productivity (higher ρ).

It is worth noting the five differences with the transformation for the total employment multiplier
obtained by Chodorow-Reich (2019). First, we have an explicit expression for the formality rate
in the multiplicative factor. Second, we have an additional term in the multiplicative factor which
captures the elasticity of informal employment to private formal employment adjusted for their
relative productivity and the informality share. Third, we consider an extra term which measures

27
the response of public employment to changes in private formal employment adjusted by the weight
of public employment. Forth, we multiply by output per private formal worker rather than output
per worker. Lastly, we use the elasticity of hours per worker to formal employment instead of total
employment.

We demonstrate that the estimated GDP multiplier obtained from a total employment multiplier
as a starting point is equivalent to the GDP multiplier derived from a private formal employment
multiplier19 . Furthermore, we show that when the informality rate is equal to zero we fall back
L
to the exact equation defined in Chodorow-Reich (2019) given that ωif goes to zero and Lt = Et .
Intuitively, the differences stem from the fact that we only observe a partial employment response,
and the leap from the estimated private formal employment multiplier to a GDP multiplier is thus
somewhat larger.

Calibration

As discussed in section 5.1, the continuous household survey PNAD allows us to observe the change
in both informal and formal employment at the national level. We thus have a sense of how private
formal, informal and public employment developed in both absolute and relative terms during the
period of analysis - informal employment fell by more than formal employment, leading to a decrease
in the informality rate, and public employment remained broadly stable20 .

We use the PNAD survey to calibrate Ψ, finding that for the 6 months period from April-September
2020, Ψ (seasonally adjusted) was equal to 2.3521 . ωifcensus
= 0.5 is the weighted municipal infor-
mality rate from the 2010 Census. Based on the work of Ulyssea (2018), we estimate ρ to 0.81.22
In line with Corbi et al. (2019), we calibrated the labor share, (1 − α), to 2/3 and the elasticity of
hours per worker to total employment, χE , to 0.12. As a result, the elasticity of hours per worker
Lt E
to the stock of effective units of labor can be written as χL = E t
χ = 0.11, implying that the
L L
elasticity of informal to formal employment is given by χ = χ Θ ≈ 0.15. As we mention before,
we set η ≡ 0 considering the orthogonality of the reaction of public sector labor to EA transfers.

Results

Figure 4 shows the implied GDP multiplier retrieved from columns (1)-(4) of Tables 2, 3, 2.A,
3.A, and 4. The graph combines uncertainty both from the formal employment multiplier point
estimate - which stems from the range obtained from different regression specifications - as well as
the uncertainty from the confidence interval around each multiplier point estimate.
19 See Lemma 1 in the Analytical Appendix for a detailed discussion.
20 Differently from the private sector, there are several legal protections granted to government employees implying

that layoffs are not easily enacted. This imposes a downward rigidity to the level of public employees.
21 The other parameters of the production function are calibrated using pre-pandemic data due to the lack of high

frequency data to estimate unobservable variables.
22 We proxy the productivity ratio of informal to formal workers (ρ) by the relative share of high skilled workers

in the informal and formal sectors, using Ulyssea (2018)’s estimates for the Brazilian economy.

28
Incorporating the full range of uncertainty implied by the confidence intervals, we obtain a very
wide range of multipliers,g from close to 0 to above 2. But focusing on the point estimates, a
relatively consistent picture emerges. A GDP multiplier around 0.5 is obtained in specifications
which use data from the full sample of municipalities and do not include the informality interaction.
Instead, the estimated multiplier is around 1.5 when informality is accounted for by its interaction
with the EA or when instrumenting EA transfers by ethnic patterns and the BF cover ratio.

Figure 4: Implied GDP Multiplier

Note: βY = (1 − α)(χ + ωfLe + ρΨωif L
) FYEt t βF E is the baseline transformation from formal employment to
GDP multiplier. (1 − α) = 2/3 is the labor share, χ = 0.15 is the elasticity of hours per worker to formal
employment, Ψ = 2.35 is the elasticity of informal to formal employment, ωfLe =0.5 and ωif L
= 0.45 are
municipal private formal and informality rates derived from Census 2010, ρ = 0.81 is the productivity ratio
of informal to formal workers, Yt is the 2020 GDP, and F Et is the stock of formal workers. The vertical
lines represent a 95 percent confidence interval.

29
Importantly, the above results are derived from regressions which estimate the relationship between
the EA and formal employment creation for the six month window between April-September 2020.
As discussed in section 3, this seems the most natural window for the analysis given that it covers
the bulk of EA disbursements. Nevertheless, we re-run the baseline specification with informality
interaction (column (1) in Table 3) for both a three-month window (April-June 2020) and a nine-
month window (April-December 2020, covering the full EA disbursement in 2020). Table 5 presents
the implied formal employment and GDP multipliers for these different time windows. The implied
GDP multiplier (for the whole year) is broadly stable across estimation windows, but the falling
employment multiplier suggests that the impact on economic activity faded over time.23 Taking
this result at face value, a possible explanation would be that transfers were incident on liquidity
constrained consumers (and so had a large immediate effect) at the pandemic onset. However, as
the economy recovered there were fewer liquidity constrained households and so the transfers had
a smaller multiplier even though they became better targeted. The relaxing of liquidity constraints
throughout the pandemic is in line with the implied saving rates from our multipliers.

Table 5: Formal Employment and GDP Multiplier at Different Horizons

Cumulative change in employment per capita

EA time Horizon:
3 months 6 months 9 months

(1) (2) (3)

Formal Employment Multiplier 2.655∗∗∗ 1.624∗∗∗ 1.117∗∗
(0.596) (0.434) (0.468)

GDP Multiplier 1.568∗∗∗ 1.569∗∗∗ 1.485∗∗
(0.352) (0.419) (0.622)

Implied Number of Year-Jobs (Formal and Informal) Created 7,380,900 6,820,800 6,500,940
Implied Cost per Year-Job (USD) 6,301 6,819 7,154
Baseline Controls Y es Y es Y es
Ψ PNAD 3.27 2.35 2.01
First Stage F statistic (EA) 967∗∗∗ 1,259∗∗∗ 1,046∗∗∗
First Stage F statistic (EA*Informality) 1,154∗∗∗ 1,531∗∗∗ 1,202∗∗∗
Wu-Hausman 56∗∗∗ 28∗∗∗ 9∗∗∗
Observations 5,474 5,478 5,478
Residual Std. Error 0.010 (df = 5441) 0.013 (df = 5445) 0.014 (df = 5445)

∗
p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Note: The unit of analysis is municipalities, the lowest administrative level in Brazil. Data on formal employment creation, and EA
transfers refer to the sum over April-June 2020 (3 months), April-September 2020 (6 months), and April-December 2020 (9 months). The
elasticity of informal to formal employment, Ψ, for each time horizon is computed using seasonally adjusted numbers from PNAD.
23 The falling employment multiplier shown in the first row and the stable GDP multiplier are consistent in the

sense that the GDP multiplier throughout the paper refers to an annual concept while the employment multiplier
refers to the estimation window, e.g. 3-months in column (1)).

30
Further discussion

As discussed in the introduction, from a theoretical perspective a large set of cross-sectional transfer
multipliers is plausible ex ante. Here we use our empirical estimate and compare it to a theoretical
multiplier obtained from parameter estimates taken from the literature. From a new Keynesian
model standpoint for a closed economy, Pennings (2021) shows that when monetary policy is passive,
ωT α
the local transfer multiplier collapses to the following equation: βY = 1−ωα where ω is the share
of hand-to-mouth households, ωT is the fraction of transfers targeted at those hand-to-mouth
households, and α is the degree of home bias (a measure of how closed the economy is).

The targeting of transfers and the share of hand-to-mouth households, while by no means fully
observable, can be gauged to some degree from the literature. Bracco et al. (2021) estimate a
share of hand-to-mouth households around half for Brazil (ω ≈ 1/2). Flamini et al. (2021) show
that in the initial months of EA disbursements, the bottom half of the income distribution (which
we can loosely assume here are the hand-to-mouth households) received around 75 percent of total
disbursements, yielding ωT ≈ 3/4. As a first step, we thus take ω and and ωT as given and ask what
α would be consistent with our multiplier estimate. Taking our estimated range of 0.5-1.5 for the
multiplier, we obtain a range of 0.5-1 for α, suggesting relatively closed local economies. Previous
evidence for Brazil does indeed point towards rather closed municipal economies given (i) the large
share of non-tradable services in the Brazilian economy, especially in poorer municipalities, and
(ii) limited (formal) labor mobility. Dix-Carneiro and Kovak (2019) show evidence of imperfect
interregional labor mobility after a negative labor demand shock (brought about by a trade policy
reform). Dix-Carneiro and Kovak (2017) also find minimal effects of regional shocks on inter-
regional worker mobility and Cavalcanti et al. (2019) find an important spatial segmentation of
labor markets. Of course, the values for ω and ωT might not be correctly measured and, moreover,
parameters such as the share of hand-to-mouth households measured outside pandemic times might
not accurately capture the dynamics during the pandemic.

One way to cross-check some of the intuition on how closed local economies are is to compare mul-
tipliers at the municipal level to those at a higher level of aggregation. Using state level data would
lead to under-powered regressions. We thus exploit the fact IBGE provides an intermediate level
of aggregation between municipalities and states, so called microregions. Given possible spillovers
between neighboring municipalities, one would expect a higher multiplier for the microregion level
regressions.

We compare the municipal level multiplier to the microregion level multiplier in two exercises. First,
we focus on the restricted sample of municipalities for which we can observe (and, thus, control for)
social mobility. Second, we consider the full sample of municipalities (not controlling for mobility).
Since at the microregion level we have mobility data for all units of analysis, controlling for mobility
does not give rise to the sample selection effect which arises at the municipal level.

31
Column (1) in Table 6 shows the GDP multiplier at the municipal level when controlling for mobility,
while column (3) shows the municipal multiplier without controlling for mobility. Columns (2) and
(4) show the corresponding microregion level multipliers. In line with intuition on the impact of
openness, the estimated GDP multiplier increases somewhat at the microregion level. Furthermore,
by comparing columns (2) and (4) one concludes that controlling for mobility per se has only a small
impact on the estimated GDP multiplier. Therefore, to gauge the effect of using more aggregated
regional data to estimate the GDP multiplier, the comparison between columns (3) and (4) - which
avoids composition effects - seems most informative. The estimated multiplier increases from 1.45
in column (3) to 1.85 in column (4), a non-negligible difference which suggests that indeed some
spillovers between neighboring municipalities occur.

Table 6: Additional Robustness tests: Changing the Unit of Analysis

Controlling for Mobility Not Controlling for Mobility

Municipalities Microregion Municipalities Microregion

(1) (2) (3) (4)

GDP Multiplier 1.646∗∗∗ 1.733∗∗∗ 1.451∗∗∗ 1.852∗∗∗
(0.628) (0.722) (0.412) (0.749)

Observations 2,208 547 5,478 547

∗
p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Note: For columns (1) and (3), the unit of analysis is municipalities, the lowest administrative level in Brazil.
For columns (2) and (4), the unit of analysis is micro regions. Data on formal employment creation, and EA
transfers refer to the sum over April-September 2020. Columns (1) and (2) use the specification detailed in
column (4) of Table 3. Columns (3) and (4) use the specification detailed in column (3) of Table 3.

All in all, our estimated multiplier range is consistent with a plausible set of underlying parame-
ters, especially when considering forced reductions in consumption due to lock-downs reducing the
marginal propensity to consume.

6 Conclusion
We provide an estimate of the GDP impact of Brazil’s emblematic Emergency Aid (EA) cash
transfer program, implemented from the outset of the Covid-19 pandemic. To the best of our
knowledge, ours is among the first studies to focus on the output effects of fiscal response policies

32
during this period. Although there is considerable uncertainty around the exact multiplier, our
preferred specifications imply that it falls in the range of 0.5-1.5. This is somewhat lower than
estimates found in the related literature for the pre-Covid period - both for the US and EMs, notably
Brazil -, possibly reflecting the effect of lockdowns and social distancing on supply chains and
consumption opportunities (forced savings). We also find that the impact of the EA was strongest
in the first three months, when liquidity constraints were perhaps more pervasive. Still, even when
using our most conservative estimates, the results suggest that the EA played an important role in
cushioning the downturn and facilitating a rapid recovery. The counter-factual without EA would
have been one with at least one million formal sector jobs and two million total jobs less, while
2020 GDP would have fallen by at least 2 percentage points more.

Looking ahead to further work, while progress had been made on understanding the size and
heterogeneity of different types of multipliers in emerging markets, more analysis is needed to allow
policy makers to design policies in the most growth friendly and inclusive way.

33
References
Auerbach, A., Gorodnichenko, Y., McCrory, P. B., and Murphy, D. (2020). Fiscal multipliers in
the covid19 recession.

Barrientos, A., Debowicz, D., and Woolard, I. (2016). Heterogeneity in bolsa famı́lia outcomes.
The Quarterly Review of Economics and Finance, 62:33–40.

Bracco, J., Galeano, L., Juarros, P., Riera-Crichton, D., and Vuletin, G. (2021). Social transfer
multipliers in developed and emerging countries.

Cardoso, B. B. (2020). The implementation of emergency aid as an exceptional measure of social
protection. Revista de Administração Pública, 54:1052–1063.

Cavalcanti, T., Da Mata, D., and Toscani, F. (2019). Winning the oil lottery: The impact of natural
resource extraction on growth. Journal of Economic Growth, 24(1):79–115.

Chodorow-Reich, G. (2019). Geographic cross-sectional fiscal spending multipliers: What have we
learned? American Economic Journal: Economic Policy, 11(2):1–34.

Corbi, R., Papaioannou, E., and Surico, P. (2019). Regional transfer multipliers. The Review of
Economic Studies, 86(5):1901–1934.

Deb, P., Furceri, D., Ostry, J. D., Tawk, N., and Yang, N. (2021). The effects of fiscal measures
during covid-19.

Denes, G., Komatsu, B. K., and Menezes-Filho, N. (2018). Uma avaliação dos impactos macroe-
conômicos e sociais de programas de transferência de renda nos municı́pios brasileiros. Revista
Brasileira de Economia, 72:292–312.

Dix-Carneiro, R. and Kovak, B. K. (2017). Trade liberalization and regional dynamics. American
Economic Review, 107(10):2908–46.

Dix-Carneiro, R. and Kovak, B. K. (2019). Margins of labor market adjustment to trade. Journal
of International Economics, 117:125–142.

Egger, D., Haushofer, J., Miguel, E., Niehaus, P., and Walker, M. W. (2019). General equilibrium
effects of cash transfers: experimental evidence from kenya. Technical report, National Bureau
of Economic Research.

Flamini, V., Toscani, F., and Masri, D. A. (2021). The short-term impact of covid-19 on labor
markets, poverty and inequality in brazil.

Gerard, F., Naritomi, J., and Silva, J. (2021). Cash transfers and formal labor markets: Evidence
from brazil.

34
Mello, U. (2021). Centralized admissions, affirmative action and access of low-income students to
higher education. American Economic Journal: Economic Policy.

Neri, M. C., Vaz, F. M., and Souza, P. H. G. F. d. (2013). Efeitos macroeconômicos do programa
bolsa famı́lia: uma análise comparativa das transferências sociais. Programa Bolsa Famı́lia: uma
década de inclusão e cidadania. Brası́lia: Ipea, 1:193–206.

Pennings, S. (2021). Cross-region transfer multipliers in a monetary union: Evidence from social
security and stimulus payments. American Economic Review, 111(5):1689–1719.

Sadoulet, E., De Janvry, A., and Davis, B. (2001). Cash transfer programs with income multipliers:
Procampo in mexico. World development, 29(6):1043–1056.

Serrato, J. C. S. and Wingender, P. (2016). Estimating local fiscal multipliers. Technical report,
National Bureau of Economic Research.

Ulyssea, G. (2018). Firms, informality, and development: Theory and evidence from brazil. Amer-
ican Economic Review, 108(8):2015–47.

35
A Analytical Appendix
Lemma 1. Equivalence of the GDP multiplier derived from private formal employment
and total employment multiplier
L
Let βE denote the total employment multiplier adjusted for effective units of labor, χL represent
the elasticity of hours per worker to the total stock of effective units of labor, and α, Lt and Yt
represent the same parameters described by equation (9). Then, we can recover the GDP multiplier
shown by equation (16) as follows:

Yt L
βY = (1 − α)(1 + χL ) β (17)
Lt E

Proof. The proof of Lemma 1 comes directly from the definition of the elasticity of hours per
worker to the total stock of effective units of labor, χL , and the transformation of private formal
employment multiplier to a total employment multiplier adjusted for the stock of effective units of
labor.

χL is formally given by

dN Lt dN F Et dF E Lt χ
χL = = = L (18)
Nt d L Nt dF E F Et dL Θ

Using the same steps that were implemented to estimate equation (4), we can translate the private
formal employment multiplier into a total employment multiplier adjusted for effective units of
labor, as follows:

L
βE = ΘL βF E (1/ωfLe ) (19)

After substituting equation (18) and (19) in equation (17), we find that

χ Yt L 1 Yt
βY = (1 − α)(1 + ) Θ βF E L = (1 − α)(χ + ωfLe + ρΨωif
L L
+ ηωpe ) βF E (20)
ΘL Lt ωf e F Et

36
B Appendix
Table 7: Baseline Regression Results (OLS)

Cumulative Change in Formal Employment per capita

OLS

Baseline Adding Services Empl. Adding Urbanization Rate Adding Mobility

(1) (2) (3) (4)

EA per capita (BRL 100K) 0.130 0.132 0.161 0.312∗∗∗
(0.112) (0.112) (0.112) (0.096)
Covid-19 Deaths 0.036 0.086 0.136 0.331∗
(0.122) (0.123) (0.123) (0.178)
Informality Rate 0.001 0.001 −0.005∗∗ −0.008∗∗
(0.002) (0.002) (0.002) (0.003)
∆ Formal Employment Q12020 −0.430∗ −0.432∗∗ −0.430∗ −0.279
(0.222) (0.220) (0.220) (0.233)
∆ Formal Employment 2019 0.008∗∗∗ 0.004∗ 0.004 0.002
(0.002) (0.002) (0.002) (0.006)
Share of Services in Employment −0.011∗∗∗ −0.008∗∗∗ 0.002
(0.002) (0.002) (0.003)
Urbanization Rate −0.006 −0.009∗∗∗
(0.008) (0.002)
Overall Mobility 0.024∗∗∗
(0.003)
Constant −0.003∗ 0.003∗ 0.009∗∗∗ 0.008∗∗
(0.002) (0.002) (0.002) (0.003)

States Fixed effects Y es Y es Y es Y es
Observations 5,478 5,478 5,478 2,210
R2 0.091 0.098 0.102 0.160
Adjusted R2 0.086 0.093 0.096 0.147
Residual Std. Error 0.013 (df = 5446) 0.013 (df = 5445) 0.013 (df = 5444) 0.011 (df = 2175)
F Statistic 17.626∗∗∗ (df = 31; 5446) 18.508∗∗∗ (df = 32; 5445) 18.660∗∗∗ (df = 33; 5444) 12.164∗∗∗ (df = 34; 2175)

37
Table 8: Baseline Regressions First Stage Results

Cumulative Change in EA Transfers per capita (BRL 100K)

Baseline Adding Services Empl. Adding Urbanization Rate Adding Mobility

(1) (2) (3) (4)

Share of BF Receivers in Population 0.039∗∗∗ 0.039∗∗∗ 0.039∗∗∗ 0.032∗∗∗
(0.001) (0.001) (0.001) (0.002)
Covid-19 Deaths 0.123∗∗∗ 0.119∗∗∗ 0.106∗∗∗ 0.129∗∗∗
(0.016) (0.016) (0.016) (0.028)
Informality Rate 0.003∗∗∗ 0.003∗∗∗ 0.004∗∗∗ 0.005∗∗∗
(0.0002) (0.0002) (0.0002) (0.001)
∆ Formal Employment Q12020 −0.003 −0.003 −0.004 −0.005
(0.004) (0.004) (0.004) (0.008)
∆ Formal Employment 2019 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001
(0.0002) (0.0002) (0.0002) (0.001)
Share of Services in Employment 0.001∗∗∗ 0.0003 0.001∗∗
(0.0002) (0.0002) (0.0005)
Urbanization Rate 0.001 0.001∗∗∗
(0.001) (0.0004)
Overall Mobility −0.004∗∗∗
(0.0004)
Constant 0.007∗∗∗ 0.006∗∗∗ 0.005∗∗∗ 0.003∗∗∗

States Fixed effects Y es Y es Y es Y es
Observations 5,478 5,478 5,478 2,210
R2 0.704 0.705 0.709 0.587
Adjusted R2 0.702 0.703 0.707 0.581
Residual Std. Error 0.002 (df = 5446) 0.002 (df = 5445) 0.002 (df = 5444) 0.002 (df = 2175)
F Statistic 417.249∗∗∗ (df = 31; 5446) 406.133∗∗∗ (df = 32; 5445) 400.971∗∗∗ (df = 33; 5444) 91.105∗∗∗ (df = 34; 2175)

38
Estimating the Employment and GDP Multiplier of Emergency Cash Transfers in Brazil
Working Paper No. WP/22/55

File and source

File: u019_bra_english-wpiea2022055-print-pdf-pdf.pdf
Size: 1,821,239 bytes
SHA-256: 9f570d1d2c0459cca19bc83487ef26e8e5b271c648b6492afe867d42a42f1638
Our copy: u019_bra_english-wpiea2022055-print-pdf-pdf.pdf
Original: www.imf.org