From Struggle to Opportunity The profile of Brazil’s working poor and implications for economic inclusion Matteo Morgandi, Katharina Fietz, Malin Ed, Gabriel Oliveira Social Protection and Jobs Global Practice Matteo © 2022 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contribution. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy, completeness or currency of the data included in this work and does not assume responsibility for any errors, omissions or discrepancies in the information, or liability with respect to the use of or failure to use the information, methods, processes, or conclusions set forth. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be construed or considered to be a limitation upon or waiver of the privileges and immunities of The World Bank. All of which are specifically reserved. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. ii Acknowledgments This technical note was prepared by a team in the Social Protection and Jobs Global Practice, Latin America and Caribbean Region, composed of Matteo Morgandi (Task Team Leader), Katharina Fietz, Malin Ed, and Gabriel Lyrio de Oliveira, and with inputs by Raquel Tsukada, Tiago Falcao, Luiz Henrique Ferreira Cruz E Superti. The team thanks Josefina Posadas (Senior Economist), Joana Silva (Senior Economist), Pablo Acosta (Program Leader) for their ample and constructive comments. The team is grateful for inputs and feedback provided at different stages by the several teams in the Ministry of Social Development of the Government of Brazil. Cite as: Morgandi, M; Fietz, K; Ed, M; and De Oliveira, G. 2022. “From Struggle to Opportunity. The profile of Brazil’s working poor and implications for economic inclusion”. The World Bank, Washington DC. The policy note was prepared as part of the Externally Financed Output 1627 “Brazil Social Protection Policies During the Recovery from COVID-19” financed by the Agence Francaise de Developpement. Contents List of Acronyms ............................................................................................................................................ 1 Overview ....................................................................................................................................................... 3 Introduction ................................................................................................................................................ 14 Chapter 1: Describing the poor and vulnerable’ s engagement in the labor market ................................. 15 1.1. A conceptual framework ........................................................................................................ 15 1.2. Dimensions of labor market vulnerability: engagement and constraints .................................. 17 1.3. The profile of poor and vulnerable in the labor market ........................................................... 25 1.3.1. Self-employed: subsistence workers or entrepreneurs? .......................................................... 25 1.3.2. Formal workers in Cadastro Único: more vulnerable than conventional workers ................... 32 1.3.3. Unemployment in poor households: a torn safety net ............................................................ 40 1.3.4. The Out of Labor Force population is hindered by caretaking duties or location .................... 44 Chapter 2: Identifying key groups and potential interventions .................................................................. 47 2.1 The in-work population........................................................................................................... 47 2.2 Out-of-work population .......................................................................................................... 49 2.2 Conclusion of LCA: interventions for the largest groups ........................................................... 52 Chapter 3: Implications for an economic inclusion agenda ........................................................................ 56 3.1. Employability services should go beyond formal technical training ......................................... 56 3.2 From single programs to economic inclusion strategies............................................................ 63 3.3. Critical institutional decisions are needed to move forward with economic inclusion in Brazil . 65 3.5. Better use of data and a renewal of Cadastro Único can support the territorial implementation of economic inclusion ................................................................................................................... 67 References .................................................................................................................................................. 73 Annex 1: Additional Tables and Charts ....................................................................................................... 75 Annex 2: Regression Results ....................................................................................................................... 80 Annex 3: Methodology Latent Class Analysis ............................................................................................. 81 Model Specification...................................................................................................................... 81 Sub-group: Individuals in vulnerable work .................................................................................... 82 Sub-Group: Individuals out of work............................................................................................... 84 Output of selected Latent Class model .......................................................................................... 86 Annex 4: Data Description .......................................................................................................................... 90 iv List of Acronyms AB Auxilio Brasil AE Auxilio Emergencial AIC Akaike Information Criterion ALMP Active labor market program ALP Adult literacy program BEm Benefício Emergencial de Preservação do Emprego BIC Bayesian Information Criterion BF Bolsa Familia BPC Benefício de Prestação Continuada BSM Brasil Sem Miseria CAGED Cadastro Geral de Empregados e Desempregados CLT Consolidação das Leis do Trabalho CNAI Cadastro Nacional de Auditores Independentes CNPJ Cadastro Nacional da Pessoa Jurídica CRAS Centro de Referência da Assistência Social FGTS Fundo de Garantia por Tempo de Serviço FST Formal short training HH Household IBGE Instituto Brasileiro de Geografia e Estatística IGD Decentralized Management Index INSS Instituto Nacional do Seguro Social LAC Latin America and the Caribbean LCA Latent Class Analysis MEI Microempreendedor Individual MoE Ministry of Education MoC Ministry of Citizenship MW Minimum wage 1 NIS Número de Identificação Social OECD Organisation for Economic Co-operation and Development PEI Productive Economic Inclusion PIS Programa de Integração Social PLMP Passive labor market program PNAD-C Pesquisa Nacional por Amostra de Domicílios Continua PRONATEC Programa Nacional de Acesso ao Ensino Técnico e Emprego RAIS Relação Anual de Informações Sociais RFB Receita Federal do Brazil SD Seguro Desemprego SICON Sistema de Condicionalidades Sistema Integrado de Pagamento de Impostos e Contribuições das Microempresas e Empresas de SIMPLES Pequeno Porte SINE Sistema Nacional de Emprego SUAS Sistema Único de Assistência Social 2 Overview Introduction The purpose of the note is to inform the design of policies and instruments that can enhance labor market outcomes of Brazil’s poor and vulnerable populations. Global and regional experiences show that active labor market programs, and more broadly economic inclusion interventions, both at the strategic level and for territorial implementation, require population-specific labor market diagnostics. And aggregate labor statistics do not portray adequately the specific situation of the poor and vulnerable. This note studies how Brazil’s poor and vulnerable engage in the labor market and in public labor market policies, or fail to do so, according to individual, family and location characteristics. We focus on two broad populations of interest: work-able adults in households living below the Cadastro Único poverty line (“the poor”), and its subset of beneficiaries of the conditional cash transfer Bolsa Familia (BF), the country’s largest social program in 2019, and currently named Auxilio Brasil (AB). The study leverages survey and administrative data routinely produced in Brazil. Sources include the 2019 survey Pesquisa Nacional por Amostra de Domicílios Continua (PNADC), and four administrative registries: Cadastro Único, BF payrolls, the formal labor registry Relação Anual de Informações Sociais (RAIS), and registry of microform owners Cadastro Nacional da Pessoa Jurídica (CNPJ). To our knowledge, this is the first study utilizing jointly these four administrative datasets to study the Brazilian labor market1. The study tried to elicit structural labor market challenges of the poor, and does not intend to explore the exceptional effects of the pandemic on labor market outcomes, which were explored in other recent WB publications (see World Bank 2022a). By relying on datasets that are regularly produced, however, this note intends to generate diagnostics that can be replicated on forthcoming data to understand the ‘new normal’, as well as on specific populations or geographic locations. How do adults in Cadastro Unico engage in the labor market? Most adult beneficiaries in BF engage in the labor market, even if with low remuneration and job quality. In 2019, 34 million adults—around 30 percent of those who could be considered work-able2 in Brazil—lived below the poverty line (0.5 minimum wage [MW] per capita). Within this large group, adults in BF are more likely to be in the labor force (70 percent) and have greater employment rates (57 percent) than the rest of the poor not in BF (Figure 0.1). They exhibit high propensity to be either in informal wage employment or in informal self-employment, with poor women in particular exhibiting highest propensity to be in informal work (Figure 0.2). Most of the working poor earn little or nothing—41 percent of working BF adults were either unpaid or earning less than the hourly MW (4.5 BRL in 2019). Adults not in BF but poor are less active in the labor market. The poor outside BF display lower employment rates than those in BF, and a much higher share is in unemployment. This group for instance comprises working-age pension recipients, including early retirees or disabled. However, those who are in-work in Cadastro Único tend to have better jobs than those in BF: in 2019, nearly half of them were in 1 Note that PNADC and Cadastro Único micordata are not matched. 2 Work-able population is defined as adults ages 15–64 years, not reporting to be in full-time education and not reporting to be out of work due to health reasons. The definition is applied to PNADC and, to the extent possible, also in Cadastro Único. Poor are defined as those living below 0.5 MW per capita, according to income reported at the household level in PNADC, and at the family level in Cadastro Único, in 2019. 3 the formal labor market. This can be considered the second tier of vulnerable workers: ineligible for social assistance, but still not making enough income to lift their family out of poverty. Figure 0.1: Labor market status Figure 0.2: Occupation of the employed by sex Employed Unemployed Out of labor force Formal Employee Formal Self-employed Informal Employee Informal Self-employed 100% 100% 21% 14% 33% 30% 20% 22% 27% 80% 80% 30% 36% 8% 22% 15% 13% 60% 20% 60% 15% 37% 25% 18% 52% 33% 40% 40% 9% 10% 71% 57% 50% 47% 7% 20% 47% 20% 32% 35% 7% 24% 14% 0% 0% Female Male Female Male Female Male Working Age P.c. Income < Self-declared BF Population 0.5 MWs Working Age P.c. Income < 0.5 Self-declared Population MWs Bolsa Familia Source: Authors based on PNAD 2019 Most of the gap in employment rates between poor and nonpoor is explained by the lower employment outcomes of poor women. Women represent 80 percent of all adults in BF that remain out of the paid labor force. In contrast, the average man in a BF household is nearly as likely to work as the average man in Brazil, even if he earns far less. Moreover, even when in the labor force, poor women face higher unemployment rates than poor men, despite being better educated. Any policy to raise employment of the poor and AB beneficiaries must primarily focus on addressing women employment and constraints. Over time, the education levels of low-income adults increased. Better education is associated with higher chances of earning above the MW and having social protection. In poor families, achieving secondary education remains the main avenue to escape low-wage work, even if often this is not sufficient. For instance, 29 percent of BF beneficiaries with secondary education land a formal wage job, twice the rate of those with completed primary education. Even when working informally, three-quarters of secondary educated adults in BF earn at least the hourly minimum wage. Youth in Cadastro Único are better educated than older adults in the labor market, consistent with the national rise in education rates in the last two decades, so many of the better-off workers in poor families are relatively young. Childcare duties and lack of locally available jobs are the chief reported reasons for the poor’s inability to work; receiving BF does not seem to disincentivize working. The principal reason reported by women for being out of the labor force is home and caretaking duties. The constraint is more binding as the number of children rises. Home duties have been reported as a negligible constraint by men. Usually, men without home and caretaking duties as well as women without children report the lack of work opportunities in their localities as the main reason for not being employed. Importantly, adults in BF report the highest interest in working more hours than they currently do and are the least likely, among the Brazilian population, to state being “not interested in working.” A body of empirical literature generally points that the BF benefit does not constant or disincentivize work. Youth under 35 represent the bulk of active jobseekers. Starting with the 2014 crisis, a growing share of the total unemployed is constituted by jobseekers transitioning from education to the world of work (Morgandi et al. 2020). This situation is also reflected in the Cadastro Único population, with almost 60 percent of the unemployed being below age 34. 4 Only 3 percent of unemployed in BF and Cadastro Único rely on public or private intermediation systems to find work. While 20 percent of adults in BF seek employment through family and friends, 74 percent directly contact employers. Formal intermediation channels remain poorly used even among the nonpoor; the poor, more socially and geographically isolated, pay a particularly high price for this policy deficit. Formal but precarious wage workers According to administrative records (the RAIS database), about 8 million adults in Cadastro Único (2 million of them in BF), had a formal labor relation at some point in 2019. This is equivalent to 15 percent of all formal registered jobs. While significant, the poor’s participation in the formal labor market has been falling: 25 percent of the poor adults in BF had a formal job in 2013, down to 14 percent in 2019 (Figure 0.4). Families in Cadastro outside BF have much higher representation in the formal labor market than those in the program. Youth, women and the better-educated are overrepresented among the formal working poor. Young people in BF are better educated than their parents, and unsurprisingly 60 percent of all formal employees are below the age of 35. 55 percent of the formally employed poor are women (as there are many more adult women than men in the program). Men in the formal sector prevalently work in agriculture, construction, and retail sectors, while women work in retail, trade, and public administration. Adults in Cadastro Único but not in BF tend to have better-paying formal jobs, and to be better educated, than formal workers in BF; this is consistent with the eligibility criteria of the program. Figure 0.3: Percentage of Cadastro Unico work-able adults Figure 0.4: Distribution of tenure (months) in formal in RAIS (with formal job) any month of 2013, 2016, 2019 employment for adults in Cadastro Unico 2013 2016 2019 31 31 28 25 18 15 14 10 8 Extreme poor in BF Poor in BF (Per In Cadastro Único (Per capita Income capita Income not in BF below BRL 89) below BRL 178) Returns to education in the formal sector for the poor are positive, but not linear. The average monthly wage of formal workers in Cadastro Único is about 1.5 MW (BRL 1,605), higher than those in BF (BRL 1,394). Except for tertiary graduates, earnings of BF recipients hardly increase with education: the differences in mean wages among those with primary and completed secondary is less than BRL 50. On the other hand, experience (proxied by age) has a significant premium, especially among Cadastro Único adults. The gender wage gap remains significant, reaching 17 percent among adults in BF, even if women tend to be better educated. 5 BF beneficiaries in the formal sector have increasingly held short-tenured jobs. The Brazilian labor market tends to have high job turnover, and the poor experience this the most. Jobs held by workers in extreme poor households last, on average, only 9 months, compared to 43 months for other adults registered in Cadastro Único, and 69 months for nonregistered workers (Figure 0.4). Around 42 percent of the formal employed in BF were dismissed involuntarily in 2019, and only 20 percent of them found a new job the same year. Young people below 24 experience the shortest tenure—7 months. The gap in average tenure between the extreme poor and other Brazilian formal workers has widened since the 2014 crisis. It is possible that the shorter duration of workers in BF is due to the increase in data cross-checks— this forces formal workers to exit social assistance when their new income is found to be above the eligibility line. Analysis by Fietz et al. (2021) shows that few workers receiving BF invoked Regra de Permancia in 2019 to avoid expulsion from the benefit. In 2019, 30 percent of BF beneficiaries worked in temporary or fixed-term labor contracts, twice the average in the rest of the formal workforce. Brazil also experienced a rise in temporary work after the 2017 labor reform, though temporary contracts remain generally more restricted than in other countries. Temporary work offers opportunities to include those at the margin in the formal labor market, due to their lower cost to employers, but temporary workers also have fewer rights. Formal workers in BF have lower access to severance pay and unemployment insurance. This underscores the importance for BF to be accessible without waiting lists. When a worker is dismissed ‘without just cause’, it costs the employers severance pay. Being dismissed ‘without just cause’ is also a requirement for workers to apply for unemployment insurance Seguro Desemprego (SD). Only half of BF formal workers were dismissed ‘without just cause’ in 2019, compared to 60 percent of all workers. And 39 percent of BF workers in RAIS in 2019 exited employment due to expiration of their contracts, twice the national average. Even among those with open-ended contracts, those in BF were three times more likely to be dismissed before the end of the probation period, when firing costs are lower and there is no provision for unemployment insurance. The data also show that among workers in Cadastro Único who were dismissed in 2019, almost none entered the BF benefit during the rest of the year (unless they were already in BF before losing their job). This illustrates how the combination of labor regulations, contract types, the benefit eligibility criteria, and the BF waiting list can leave households with a single formal worker stranded without income support during a shock. The discrepancy between RAIS and Cadastro Único on formal wage employment participation is significant. Through data cross-checks, we estimate 9 percent of adults in BF had an active contract at some point in 2019, while only 2 percent of BF recipients are listed as formally employed in Cadastro Único at the time of registration or recertification in 2019. Clearly, many update their records in times of greatest financial need and between employment spells. However, even when comparing records of the same month, mismatches remain and more likely depend on the delays in the generation of employment data. The self-employed: few entrepreneurs, many in subsistence work Most of the poor self-employed are informal, male, and live in urban areas. About one-third of BF beneficiaries in the labor market are self-employed, more than 80 percent of them informal (in PNADC 2019). Two-thirds live in urban areas; they are predominantly middle-aged or elderly, and with disparate levels of education. Construction and agriculture are prevalent occupations among men, domestic and retail services among women. On average, even though women are better educated, monthly earnings are lower for self-employed women (BRL 470) than men (BRL 796); this is correlated with their strong occupational segregation, and underscores the importance of information that can nudge occupational choices according to returns, including in self-employment. 6 Few self-employed match the earnings of formal wage workers in Cadastro Único. As shown by their meagre average earnings, most self-employed are not dynamic microentrepreneurs, but more similar to subsistence workers. The two most prevalent industries where the formal poor self-employed work are agriculture and domestic services—these industries were object of government policies to extend social protection, but they are unlikely to be markets for individual to grow a firm. The minority of the self- employed, especially those who completed secondary education, operate as formal microenterprises and can match the earnings of BF workers with a formal labor contract (BRL 1,300 per month). Less than 5 percent of the self-employed are employers. Such data challenge the idea that “entrepreneurship” should be a primary avenue to escape poverty, and instead points at the importance of addressing individual productivity gaps and local constraints to economic growth. Self-employment is captured more superficially in Cadastro Único than in PNADC. Cadastro Único displays a much weaker correspondence with PNADC for self-employment than other forms of work, even on basic indicators (with Cadastro Único largely overestimating self-employed) and share of women (much higher in Cadastro Único than in PNADC). Moreover, the PNADC survey collects sufficient information to classify the self-employed, including formal registration as a firm (CNPJ), and individual contributions to Instituto Nacional do Seguro Social (INSS). Improving the definition of self-employment in Cadastro Unico is thus a priority to identify recipients of potential support. Through data cross-checks, we estimate that 3 percent of work-able adults in BF and 5 percent work- able adults in Cadastro Único out of BF are registered as MEI. Importantly, nearly 42 percent of those who are classified as employers in Cadastro Único have an MEI. But otherwise, MEI registrants are almost equally prevalent in Cadastro Único among informal dependent workers, self-employed, and formal employees. On the one hand, this corroborates the abovementioned hypothesis of poor classification of self-employment in Cadastro Único. On the other, it is also likely that many MEI owners are simply disguised dependent workers (since MEI has few nonwage labor costs). Vulnerable workers are likely less aware of the penalties for employers who hire employees as MEIs. In general, average declared earnings in Cadastro Unico do not differ substantially between workers with and without an MEI; this is consistent with findings in the recent literature on a weak relationship between formalization of MEI and firm growth (Hus Rocha and de Farias 2021; Ulyssea 2018). Priority groups for public interventions This note’s conceptual framework identifies three sets of constraints to entering productive employment: employability barriers, participation constraints, and frictions in matching demand and supply. We use PNADC to map the prevalence of these constraints on the in-work and out-of-work poor population, and then apply latent class analysis (LCA) to identify a few synthetic groups (or classes) of individuals who share similar characteristics. The classes do not perfectly fit all populations and characteristics but help suggest packages of interventions to serve individuals who frequently have similar and overlapping constraints. Among the in-work poor, five groups with similar constraints and capabilities are identified by the model—80 percent belong to the one of the three largest groups: • Full-time formal employees (32 percent). This is, apparently, the least vulnerable cluster. They hold a formal contract job, half of them have completed secondary education, but still with too many dependents to lift the family out of poverty. The majority are single-earners in the family, and less than half receive BF, consistent with lower-income vulnerability. Such families remain vulnerable in case of employment loss, because most workers have short tenures, as discussed above. These formal workers could benefit from policies to improve awareness of their labor 7 rights, proper contractual framing to increase access to social protection programs. Greater access to on-the-job training while working, and labor intermediation when out of work, would also support a reduction in their unemployment spells. • Uneducated informal underpaid workers (25 percent). In this group, 90 percent earn less than the hourly minimum wage, and nearly all receive BF. About 60 percent are self-employed, and live mostly in small urban or rural areas. Two-thirds are male, nearly all are in the age group 35–64 years: the combination of age, location, and low education make this a particularly difficult group to serve through regular training programs. Such workers would benefit from economic inclusion interventions that combine tailored skills training at appropriate schedules compatible with work as well as support to access markets and acquire capital or inputs, depending on the rural or urban context. In addition, interventions should promote financial literacy, access to savings instruments and eventually credit, and greater awareness of the protections deriving from being formalized through the MEI program. • Educated underemployed mothers in the informal sector (23 percent). Women, mostly below age 34 and with young children, comprise 77 percent of this cluster. The majority completed secondary school but still work in the informal sector as employees or self-employed, usually less than 35 hours a week and with low wage. This cluster has constrained potential and should be targeted with an increase in the supply of caretaking facilities in combination with labor intermediation measures and skills development programs. Figure 0.3: In-work poor individuals by clusters (latent classes) Low educated Unpaid low educated informal Educated male full- informal male in BF, dependent female workers, time formal 15.0% 4.8% employees, 32.2% Educated informal Low-educated female workers in underpaid informal BF, 23.2% workers in BF, 24.8% Source: Authors, based on PNAD C 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education. The out-of-work poor are more heterogenous, reflecting very different degrees of labor market distance. With seven classes identified by the model, the four largest classes capture about 70 percent of the out-of-work poor: • Inactive low-educated women (20 percent). Mostly ages 35 to 54 and married, they have been out of work for more than a year, do not receive BF, and rely on spousal labor income or, in half of the cases, other benefits such as Benefício de Prestação Continuada (BPC) or pensions. About 90 8 percent do not have secondary education, and a third live in rural areas. These are difficult targets for activation measures. • Medium-educated young jobseekers (19 percent). Young men and women, most of whom have completed secondary education but are unemployed (usually for more than six months) and tend to live in urban areas. The majority are unmarried and not in BF. Targeted active labor market programs, including labor intermediation and temporary wage subsidies for employers, can help gain a first formal employment and acquire work experience (Morgandi et al. 2020). • Educated female caretakers in BF (18 percent). Mostly below 35 years of age, 97 percent of this group live in a household with either young children or a person with disability. Nearly half of them have completed secondary school, 60 percent receive BF, 77 percent are married and rely on the income of other adult earners in the family. About 40 percent of them are looking for work or wish to work but are discouraged, and they live mostly in small urban or rural areas. In addition to childcare, this group would benefit from policies to connect them with employment opportunities, or support for income-generating activities at the household level. • Short-term urban unemployed (13 percent). A heterogenous group of women and men of different ages (between 18 and 54 years) who became recently unemployed. About 56 percent have secondary education. One-third of them have young children in the family and 40 percent received BF in 2019. Despite their better employability traits, almost half of this group are in families without a single earner, and nearly all live in urban areas that exhibit very low local employment rates. Such groups will benefit from intermediation, programs that support labor mobility, and in some cases training programs. The remaining clusters are described in the note. While the model identified such groups at the national level, the prevalence of each group across territories will vary: this underscores the importance of mapping population traits and characteristics also at the territorial level at the stage of policy design. Figure 0.4: Out-of-work poor individuals by clusters (latent classes) Short term, low- educated Inactive low educated unemployed, 7.3% women , 20.8% Unmarried prime age inactive men , 10.1% Better educated Rural discouraged young jobseekers, women in BF, 11.6% 18.9% Short term urban Educated mothers, unemployed , 13.2% 18.0% Source: Authors based on PNAD C 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education. 9 Expanding opportunities for the working poor A number of institutions already exist in Brazil that provide services to strengthen employment outcomes of the working poor. Services provided across the federation include (a) a large network of federally funded technical schools (Institutos Federais) and adult education programs (Educacao Jovens e Adultos, EJA); (b) public labor intermediation offices (SINE); (c) employer-funded vocational training schools and intermediation centers (Sistema S); (d) state-level agricultural extension programs; (e) financial institutions with dedicated microcredit programs. The analysis of federal budget lines suggests that federal labor policies most targeted to the poor emerged in scale with the Brasil Sem Miseria (BSM) strategy. After the 2014 crisis, budget cuts defunded or closed most of such specialized interventions. What remains largely in place at the federal level are vocational education institutions and the autonomous Sistema-S. Training is also reaching a small share of the adult poor. Novel analysis of a special module in PNADC 2019 revealed that 1.7 percent of Brazilians between ages 18 and 65 took part in a formal short training program that year, equivalent to more than 1.8 million adults. This rate is lower, 0.9 percent, among BF beneficiaries and 1.3 percent of others in Cadastro Único. Importantly, training programs were used more often, but not exclusively, by the poor who are also better educated: this shows the supply of programs already covers different educational ranges in the population, but there is room for further adaptation. The data shows that Sistema-S and public federal institutions are the principal institutions attended by the medium- and low-educated. Past evaluations showed that short technical training courses in Brazil, including those targeted to low- income youth, can be effective, if appropriately designed. Almeida et al. (2014) found positive returns to short-term training compared to nonparticipants in Brazil, but only for those who attend private or Sistema-S institutions. PRONATEC programs, when designed with clear demand-driven features, such as offering training only in skills that were expressly requested by employers in the territory, showed positive impacts on employment rates (SAGI 2018). A second strategic service to improve the foundations for future learning are adult literacy programs. One of the reasons why vocational training programs often fail for poor adults is that candidates lack the foundational skills to engage in technical learning. Second-chance education programs are thus strategic to support the high share of uneducated workers and jobseekers in Cadastro Único to eventually access new skills training. Evidence on the effectiveness of such programs remains scarce at the global level, but fundamental. In Brazil, adult literacy programs are provided as part of the federal education system. Such programs should be systematically considered part of the ‘toolkit’ of interventions that social assistance and labor officers can offer their clients. Appropriate screening tools of foundational competences should be made available to identify those in need for reinforcement. Brazil should also revamp its public employment services, by adding services that go beyond job matching, especially for hard-to-place unemployed. The toolkit should go beyond formal in-class training available today. It should encompass instruments to assess candidates skills, job search assistance and soft skills, targeted wage subsidies for the first job, and programs to reduce supply-side barriers such as childcare and transportation allowances. In many localities with limited labor demand, programs for the urban self-employed, or more sophisticated investment to develop value chains and support the poor in participating into these are promising. Above all, the SINE network should start operating in logic of partnership, referrals and as a broker of opportunities on the territory, especially by training its officers on modern profiling and case managements. This needs not to increase operating costs, yet could radically change SINE’s value added. 10 Reforms of the traditional, cash labor market benefits can identify fiscal space to finance such active labor market interventions. With total federal and employers’ spending on labor programs above the resource allocation averages for Organisation for Economic Co-operation and Development (OECD), institutional arrangements seem more important constraints than resources alone. As argued in other publications in more detail (Morgandi et al. 2020), improving the quality of active labor market policies is not fiscally unfeasible for Brazil, provided long-overdue reforms of its contributory labor benefit systems are carried out. Critical institutional decisions underlie successful economic inclusion strategies A major agenda for the future is the development of delivery systems for labor market policies, ideally led by the federal government. These comprise instruments, protocols, and institutions for the coordination and governance of specific interventions. The central government has a critical role to play in establishing the ‘rules and tools of the game’: regulations and financial incentives, a ‘menu’ of interventions that can be customized locally, monitoring systems to verify the implementation on the ground and what works, and operational systems to manage client intake, including by drawing relevant labor profiles of individuals and communities to make treatment decisions. Some of these components can build on existing information systems, such as Cadastro Único. The Federal Government is best placed to develop such delivery system, as this would allow to reap Brazil’s large economies of scale that such investments require. The delivery system can be based on the successful experience of building Sistema Único de Assistência Social (SUAS), the CRAS and the development of a performance-linked funding mechanism – such as the Decentralized Management Index (IGD)3. With the appropriate governance tools in place, funding can be transferred at the subnational level to help contract out labor interventions appropriate to the profiles of local beneficiaries. The delivery of economic inclusion policies, however, should occur at the subnational level. Economic inclusion services are more complex than cash transfers, they cannot be only coordinated at the federal level or implemented only digitally. An important decision will be to identify the institution at the local level that should serve as “entry points” and local coordinator of many potential interventions. Such institutions remain underdeveloped in Brazil, as this reduces the potential to enhance the governance of the system. Well developed “entry point” institutions manage the client treatment cycle, including registration, profiling, referral to services, and follow-up to verify completion. They also serve as local planners to support the identification of necessary interventions based on the territorial vocation. In high income countries, labor offices have often played this role. In Brazil, Accessuas Trabalho, was the institution that tried to play such role for a short period. This function should be revived, based on lessons learned from the past. Even in such a setup, SINE would maintain its core comparative advantage in labor intermediation, and act as one of several labor market services “in the menu”. 3 A strategic index used to incentivizing states and municipalities in Brazil to improve their management of the BF program and Cadastro Único 11 Figure 1: Proposed division of responsibilities across institutional levels Federal Territorial Government Coordinator Entry point Interventions Implementing Study labor and Client registration and Take charge of clients regulations economic opportunities profiling after referral Registry of beneficiaries Identify and certify services to be part of the Referral to interventions Provide service Define menu of 'delivery chain' interventions Define coordination Feed information system Follow up and collect protocols between (attendance Capacity building and data after service actors /completion) Communication Monitor quality of Coordinate with other Reflect on results Monitoring systems services and manage local services information, improve point of service Better use of data for planning economic inclusion The use of individual-level data in Cadastro Único and other public registries can help inform planning and execution of economic inclusion. Before the pandemic, data cross-checks were commonly used to ‘audit’ records of applicants to cash transfer programs. During the pandemic, Auxilio Emergencial (AE) stepped up the use of interoperability in administrative sources to proactively identify potential targets of cash transfers. This note explores the potential of such data for a third, unexplored, objective: to inform labor policies and support economic inclusion planning at the territorial level. The applications include: (a) providing strategic understanding at the national level of the labor market conditions of the population served by social safety nets; (b) better mapping of local conditions, allowing for the design of appropriate measures and targeting criteria among a menu of interventions; and (c) supporting the local implementer by using data as part of the diagnostic phase of individual beneficiaries (profiling) . Through matching and comparing different data sources, the note identified scope for improvements and refinements for Cadastro Unico’s labor data. Recommendations span across many domains (Figure 0.5), including: • align definitions of employment type with PNADC, especially to distinguish the self-employed from other informal wage workers and rural farmers • draw from the national registry of firms and contributors to social insurance, to identify formal labor relations and firm ownership, while maintaining the possibility of overriding this information • include basic questions on recent job search and interest in working, to identify the unemployed. Replace self-reported education with data in the national education registry, adding information on short training programs. • Display at the individual level information on receipt of social protection programs, such as Seguro Desemprego, pensions, and disability benefits, which affect incentives to enter formal employment. • Consider integrating in CadU fields from prontuario SUAS, the social assistance client database, which already collects information on household level constraints. 12 Figure 0.5: Administrative data sources with potential to increase accuracy of Cadastro Único Current information available in Cadastro Único (i) Worked last week - (ii) Type of work (not between formal and informal self-employed) (iii) Self-reported education data (iv) Self-reported incomes Strenghten profile of individuals and families Active Labor Other social protection Labor market Education Market programs Policies Receita Federal MoE Formal SINE SD Declaration IRPF Education Registration Vocational Short formal RAIS + CAGED Pensions Training training Public pre- Disability CNPJ / MEI school and (BPC) childcare Contribtions (GFIP/ Esocial) 13 Introduction Economic inclusion strategies aim to enhance income-generating capacities of low-income individuals. How to do so differs distinctly based on context and characteristics of beneficiaries. In broad strokes, these strategies encompass one of these three approaches: (a) strengthening the participation in the labor market and access to quality employment, (b) increasing the access to product markets for existing or new businesses, and (c) increasing productivity of household-based activities. Economic inclusion policies start from the premise that poor and vulnerable individuals may not necessarily have equal opportunities to take part in economic growth. Addressing individual, household, and institutional constraints that prevent this inclusion is the central aspect of any such strategy. For this reason, this note focuses on the description of structural labor market challenges of the poor, rather than discussing the impact of the recent COVID-19 crisis and policies for countercyclical responses. Information about beneficiaries’ labor market profiles can be used to inform several stages of the policy cycle of economic inclusion. Such stages include early planning, targeting, selection of interventions, monitoring, and evaluation. This note focuses on the application of labor profile data mainly with two purposes: (a) identify the degree of labor market engagement of the poor in the social registry, as well as their constraints for more productive participation, for the purpose of strategic planning of policies and programs and (b) explore potential applications of data analytics to target such policies at the local and individual levels. The analysis leverages both survey and administrative data, namely the 2019 Pesquisa Nacional por Amostra de Domicilios Continua (PNADC) the social registry Cadastro Único, BF payrolls, the formal labor registry Relação Anual de Informações Sociais (RAIS), and registry of microform owners Cadastro Nacional da Pessoa Jurídica (CNPJ) 4. The emphasis on regularly produced data sources is deliberate, with the idea of enabling replication of such analysis over time or in specific populations. Through a comparative approach, the note also tries to identify advantages of each data source, and areas for improvement in administrative data. The note is structured in three chapters. Chapter 1 describes the extent and forms of labor market engagement of the ‘work-able poor’ in Cadastro Único, and the observable constraints, organized according to a conceptual framework. Chapter 2 aims to explore potential applications of data analytics to inform planning of economic inclusion policies at the national and local levels. At the national level, we use LCA to identify groups of individuals who face similar constraints to better economic inclusion. In chapter 3, we provide examples of how the administrative data could be utilized to develop local labor market profiles and how the analysis and available data can be used to inform an economic inclusion strategy for Brazil. 4Both surveys offer a rich database which can be explored in different aspects. Nevertheless, they also face obstacles, which need to be taken into account. See Annex 4 for details on the advantages and constraints of the two datasets. 14 Chapter 1: Describing the poor and vulnerable’ s engagement in the labor market Key messages of chapter 1 • Both demand and supply side constraints determine the labor market outcomes of poor and vulnerable. Supply side constraints include participation barriers (e.g. care duties, poor health, mobility), employability barriers (e.g. low education, lack of cognitive skills) and labor market frictions (e.g. salaries, incentives to work). • Chapter 1 aims to identify different labor market profiles of two groups of poor work-able adults: those in BF and other adults in Cadastro Unico. • 70 percent of BF beneficiaries of working age participate in the labor market (albeit in low quality of work), and 57% were employed in 2019. Other poor adults in Cadastro Único not in BF are less active, but when working they have slightly better quality jobs. • The out-of- work receiving BF report two main barriers to being in employment: caretaking duties and being a in a location without jobs: Women represent 80 percent of all adults in BF that remain out of the paid labor force. • Formal employees: About 9 percent of BF beneficiaries had a formal job during 2019.This share was significantly higher among individuals in Cadastro Único not in BF (28 percent). BF beneficiaries in the formal sector hold short-tenured jobs, often with temporary or fixed-term labor contracts. Access to unemployment insurance and severance pay is lower than among the non-poor due to the type of contracts the poor are hired on. • Subsistence entrepreneurs: Most self-employed are informal and male, between 25 and 44 years old living in urban areas. Most poor self-employed work in low-wage sectors, such as agricultural and domestic services, and only about 5 percent have employees. Rather than being entrepreneurs, these are mostly workers who could not find better wage employment • Unemployment in poor households: Among those that are unemployed, BF is a more important source of income than unemployment insurance. Most unemployed are young and first-time jobseekers. Most poor jobseekers directly contact employers or leverage personal contacts to find work, very few use formal intermediation services. 1.1. A conceptual framework Labor market vulnerability among the poor can take different dimensions, often defined as gradients of ‘labor market distance’. These are very heterogenous and range from being entirely out of the labor force to being adequately paid in the formal sector (see Figure 2). Some adults may be entirely out of work, and for this reason poor; others might be in vulnerable forms of work that does not provide sufficient income or security. Finally, a minority may be in better labor market conditions (such as a formal job) but living with other household members in more vulnerable situations, and thus unable, on their own, to lift the household out of poverty. Figure 2: Degrees of labor market vulnerability 15 Out of labor Long-term Recent Out of work Discouraged force unemployed Unemployed Underpaid or Involuntarily Vulnerable work Unpaid underemploy Informal ed Adequately Better work Formal paid Such outcomes are normally the result of supply-side and demand-side constraints. Supply-side factors have proven to be important determinants of labor market outcomes of the poor, and these are largely the focus of this note. It is, however, important to remember that constraints to labor demand and lack of economic growth at the local level in general can also be very important barriers. In our framework, supply constraints can be grouped into three domains (see Figure 3). Figure 3: Three groups of labor market constraints Employability Participation Barriers Constraints Labor market frictions • Participation constraints are conditions at the individual and household levels that make labor supply, including search for work, difficult or impossible. They include care duties for other family members, poor health, mobility constraints, duties to do household chores, or simply lack of decision-making. Many of these constraints imply individuals are using their time intensively, but not on the labor market; overcoming such constraints often requires incurring a private cost. • Employability barriers are all factors that do not allow a person potentially available to work to seize available labor market opportunities. Typical constraints include low education or lack of basic cognitive skills, as well as socioemotional skills and, often, also prior work experience. A major market failure associated with this constraint relates to the lack of credit or information on returns to invest in training and skills acquisition; such a constraint is much more binding for the poor who have high discount rates. • Labor market frictions capture a variety of factors that prevent even suitable demand and supply matches from taking place. These include lack of appropriate incentives to work (due to high unearned income or disincentivizing benefit rules) and lack of information on existing opportunities, but also signaling failures on the true ability of candidates and labor market discrimination. In the case of Brazil, recent analyses confirmed that labor market disincentives are unlikely due to the benefit’s design (Fietz et al. 2021). 16 1.2. Dimensions of labor market vulnerability: engagement and constraints This section aims to illustrate, through a labor market profiling exercise, the incidence of different forms of labor market vulnerability and the constraints that contribute to labor market outcomes. Four datasets from 2019 are used: the household survey PNADC, the administrative datasets Cadastro Único, RAIS and MEI. In both surveys, we compute a host of indicators that follow the framework presented in section 2, for two main populations: adults recipients of Bolsa Familia (‘BF recipients’), and the ‘nonbeneficiary poor’, defined as those living in households below the Cadastro Único eligibility line, but not in BF. Note that all families in BF also need to be registered in Cadastro Único. Means of the full population are also shown as a benchmark comparison. These analytical groups have been further restricted to individuals that are of working age (18–64 years), not enrolled in full-time education and able-bodied, to represent the population that is potentially able to be in employment. Implementing these definitions requires different sets of assumptions in the two datasets, as explained in the methodology in Annex 4. The analysis focuses on adults living under the Cadastro Único income eligibility line. This per capita income limit allows to capture a broad spectrum of the poor and vulnerable in the labor market. Brazil does not have an official poverty line and administrative eligibility lines have often been used to identify different welfare groups. The Cadastro Único eligibility line reaches the bottom 30 percent of the per capita income distribution and is thus adequately representing not only the poorest but also many of the vulnerable5. The advantage of adopting such definition to defined the analytical groups in our study is that it allows some approximate comparison between the group of poor defined in the administrative registry, and those with similar per capita income levels in the national household survey. About one-third of the Brazilian working-age able-bodied population lives below the poverty line of Cadastro Único. Around 136 million individuals are between ages 18 and 64 years, and, among them, 115 million are potentially work-able (with no stated inability to work due to health reasons and not in full- time education). Among them, around 30 percent (34 million individuals) live below 0.5 minimum wages per capita (Figure 4)6, which is the analytical population described in this note. 5 In 2019, the Cadastro Único threshold, set at half a minimum wage per capita, was about 2.5 higher than the eligibility line for the poverty targeted program BF, and close in value to the international upper poverty line adopted by the World Bank for middle-income countries. One limitation of this analytical approach is that it does not benchmark the labor market outcomes of poor with those in the lower-middle class living above the Cadastro Unico poverty line, who are also susceptible to falling into poverty during shocks. Table 2 in annex 1 provides a comparison with the vulnerable lower middle class and shows the significant difference in outcomes between the groups. 6Cadastro Único line corresponded in 2019 to BRL 500, a value that, in purchasing power parity (PPP) US$ per capita, is just above the World Bank’ s international poverty line of US$ 5.5 (PPP) per day deemed suitable for middle-income countries like Brazil. 17 Figure 4: Population groups 300 Millions 209 200 136 115 100 34 19 0 Whole Population Working age (18-24) Working age (18-24) not in education and able bodied Poor working age (18-24) not in education and able bodied BF beneficiaries at working age (18-24) not in education and able bodied Source: PNAD 2019. Note: Poor defined as living below 0.5 minimum wages per capita. The work able poor are very heterogenous, including in terms of educational attainment. Among BF beneficiaries, around 30 percent have not completed elementary school (5 years of schooling) and half of all work-able BF beneficiaries have not completed middle school (9 years of schooling). Individuals in Cadastro Único not receiving BF have slightly higher education levels, 60 percent of individuals have at least completed middle school. Such education rates, however, mask important differences across age cohorts: within the recipients of BF, about 5 out of 10 of ages 18–24 potentially in the labor market have completed upper secondary education, whereas this is the case for only 1 out of 10 adults above the age of 45 (Figure 5). Figure 5: Work-able poor, by education level and age 100% 80% 60% 40% 20% 0% All Age 18-24 24-34 35-44 45-54 55-64 All Age 18-24 24-34 35-44 45-54 55-64 Groups Groups Self-declared BF P.c. Income < 0.5 MWs No instruction or up to 1 year Incomplete Basic Complete Basic Incomplete secondary Complete secondary Incomplete superior Complete Superior Source: PNADC 2019. Note: Groups are restricted to individuals of working age (18–64) and not education and able bodied. On average, about 70 percent of work-able poor participate in the labor force, and more than 50 percent are working, though with important gaps by gender and age. About 77 percent of males in BF are 18 employed, only 41 percent of women work. Among others living below the 0.5 MW per capita line but not receiving BF, employability is slightly lower: 61 percent of males and 37 percent of females work. Unemployment rates are higher in this group, with unemployment being 21 percent in males and 18 percent in females. In comparison, among male BF beneficiaries, 11 percent are unemployed and among female BF beneficiaries, 14 percent are unemployed. Though better educated, a large share of work-able youth in BF is out of the labor force. The poor face lower employability rates. Among the entire Brazilian population, 71 percent are employed, only 47 percent of the poor not receiving BF, and 57 percent of those that self-declare to receive BF are employed. Unemployment rates and the share of people out of labor force is higher among the poor. Those circumstances are especially present among the young that receive BF. In the 18–24 years age group, only 45 percent are employed compared to 63 among the 35– 44 years group (Figure 6, Figure 7, and Figure 8). Figure 6: Labor market status Figure 7: Labor market status by sex 100% 100% 21% 80% 33% 30% 80% 8% 13% 60% 60% 20% 40% 40% 71% 20% 57% 20% 47% 0% 0% Female Male Female Male Female Male Working Age P.c. Income < 0.5 Self-declared BF Population MWs Working Age P.c. Income < Self-declared Population 0.5 MWs Bolsa Familia Employed Unemployed Out of labor force Employed Unemployed Out of labor force Figure 8: Labor market status by age 100% 80% 60% 40% 20% 0% 18-24 25-34 35-44 45-54 55-64 18-24 25-34 35-44 45-54 55-64 Self-declared Bolsa Familia P.c. Income < 0.5 MWs Employed Unemployed Out of labor force Source: PNADC 2019. Note: Groups are restricted to individuals of working age (18–64) and not education and able bodied. The most common form of employment of the poor is informal dependent work, followed closely by informal self-employment. In the whole working-age population, 35 percent work informally: 17 percent are informal self-employed, and 18 percent are informal employees. The share of informal workers is significantly higher among people living below 0.5 MWs per capita—with 27 percent of informal self- employed and 30 percent of informal employees—and especially high among BF beneficiaries. Among BF beneficiaries, more than 70 percent of work-able individuals in the labor market work informally (Figure 9). Looking at the share of informal/formal employment by gender, it stands out that especially women 19 work informally: 79 percent of female BF beneficiaries are informal compared to 69 percent of male BF beneficiaries (Figure 10). Figure 9: Occupational status as share of employed Figure 10: Occupational status as share of employed by sex 100% 100% 50% 50% 0% Working Age P.c. Income < Self-declared BF 0% Population 0.5 MWs Female Male Female Male Female Male Working Age P.c. Income < Self-declared Informal Self-employed Informal Employee Population 0.5 MWs Bolsa Familia Formal Self-employed Formal Employee Informal Self-employed Informal Employee Formal Self-employed Formal Employee Source: PNADC 2019. Note: Groups are restricted to individuals of working age (18–64) and not education and able bodied. Most of the in-work poor are out of a decent job, defined as work that does not provide access to social protection or earnings equivalent to at least one gross minimum wage.7 Table 1 shows that among the working BF beneficiaries almost 5 percent receive no pay for their work, and 36 percent earn below the hourly minimum wage. The poor that are not receiving BF but have a per capita income below 0.5 MWs are slightly better off with 3 percent of individuals doing unpaid work and 28.3 percent of being underpaid, with significant difference to the rest of the population. Furthermore, more than one-quarter of people in BF would like to work more hours than they currently do. The poor not receiving BF have the highest share of unemployed individuals (long-term and short-term), which is consistent with the assumption that the extreme poor (those in BF) cannot afford to be unemployed for prolonged periods. The highest share of discouraged individuals, meaning individuals that are out of the labor force but would actually like to work can be found among BF beneficiaries (33 percent), followed by other poor not receiving BF (Table 1). Table 1: Degrees of labor market attachment in % Non-beneficiary poor Bolsa Familia Rest of the population (income above 0.5 MW per capita) Out of work 53 43 21 Out of labor market 41 37 63 Discouraged 21 33 13 Long-term unemployed a 16 13 11 Recent Unemployed b 22 17 13 In Work c 47 57 79 Unpaid 3 5 1 7Defined as minimum wage plus nonwage labor costs. The rationale for such definition is that at times informal employment may be voluntarily negotiated with employers, especially when workers do not value the benefits provided with formality; a gross minimum wage assumes that the cash value of forfeited social security rights is paid back to the employee. 20 Non-beneficiary poor Bolsa Familia Rest of the population (income above 0.5 MW per capita) Low pay 28 36 5 (below hourly MW) Underemployed 21 27 9 (involuntary few hours) Without formal contract 55 73 27 or registration Source: PNAD 2019 Note: Population restricted to individuals of working age (18–64), not education and able bodied. a. Long-term unemployed is defined as individuals being without a job and trying to find a job for one year or longer. b. Short-term unemployed is defined as individuals being without a job and trying to find a job for less than one year. c. Subgroups do not sum up to a 100 percent due to possible double counting. Education level remains the strongest predictor of being in a job that pays better or with social protection. The share of adults who work in formal dependent work—and thus are considered to have a more stable education level—is increasing with education levels. Among self-declared BF beneficiaries that have up to one year education, almost all are informal employees. Among those that completed secondary education, almost one-third of BF beneficiaries work as formal employees. This share increases to 43 percent for people that have completed higher education levels (Figure 11). The share of individuals that are un- or underpaid among working individuals decreases with education level. Among individuals with no education or up to one year, 6 percent of working BF beneficiaries are unpaid, only 3.2 percent of individuals that completed secondary education are unpaid. The decrease is more marked for underpaid work (below minimum wage): 53 percent of working BF beneficiaries with no instruction or up to one year are underpaid and only 28 percent of those that completed secondary education ( Table 2) are underpaid. Figure 11: Occupational status of self-declared BF-beneficiaries by education level 100% 80% 60% 40% 20% 0% No instruction Incomplete Complete basic Incomplete Complete Incomplete Complete or up to 1 year basic secondary secondary superior Superior Formal Employee Formal Self-employed Informal Employee Informal Self-Employed Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. 21 Table 2: Workers who are underpaid or unpaid, by education (%) Unpaid Work Earning below hourly MW Non- Bolsa Rest of the Non- Bolsa Rest of the beneficiary Familia population beneficiary Familia population poor (income above poor (income above 0.5 MW per 0.5 MW per capita) capita.) No instruction or up to 5 6 2 52 53 24 one year Incomplete Basic 4 5 2 36 42 11 Complete Basic 3 5 2 30 31 6 Incomplete Secondary 3 4 1 26 31 7 Complete Secondary 3 4 1 19 28 5 Incomplete Superior 6 3 1 22 26 3 Complete Superior 4 2 0 13 15 1 Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. The administrative data in Cadastro Único captures a lower share of total employment than PNADC. In this administrative dataset, it is possible to identify whether or not individuals worked the previous week. However, a distinction between being in unemployment (for example, engaging in job search) and out of the labor force is not possible. About 42 percent of BF beneficiaries and 45 percent of non-BF beneficiaries in Cadastro Único indicated that they had worked during the previous week (Figure 12). When disaggregated by sex, employment rates of BF beneficiaries are 36 percent for women and 54 percent for men, and roughly the same in Cadastro Único (Figure 13). Employment rates in Cadastro Único (Figure 12) seem to be lower than in the PNADC household survey (Figure 6), especially in the case of men: In the Cadastro Único data 54 percent of male BF beneficiaries stated to be working compared to 77 percent of male BF beneficiaries in PNADC (Figure 12 and Figure 7). The survey may capture fewer poor families without male earners (as PNAD is not designed to reproduce the Cadastro Úniconico population), or there could be fewer men registered in Cadastro compared to those actually present in poor families. Figure 12: Labor market status in Cadastro Único Figure 13: Labor market status by sex in Cadastro Único Bolsa Cadastro Male 53% 47% Único Cadastro Único 45% 55% Female 39% 61% Male 54% 46% Familia Bolsa Familia 42% 58% Female 36% 64% Worked last week Did not work last week Worked last week Did not work last week Source: Cadastro Único 2019. Note: Groups are restricted to individuals of working age (18–64) and not education and able bodied Another major discrepancy in Cadastro Único is the higher prevalence of self-employment and underestimation of informal dependent work. The registry contains basic categories of employment (self-employed, agricultural workers, formal wage employee, and informal worker), with no distinction 22 between informal and formal self-employment (something that PNAD allowed to identify in recent years and is discussed further below). Unlike PNADC, where most poor report to work as informal wage employees, a much larger share in Cadastro Único declare to be self-employed, for instance 64 percent of the extreme poor and 76 percent of the poor receiving BF, compared to only 39 percent of the employed receiving BF in PNADC (Table 3). Ownership of physical assets and housing is also an important determining factor of labor market vulnerability. For instance, owning a bike, car, or motorcycle can allow an individual to participate in the gig-economy. Internet access together with a computer or a tablet can help allow workers to work in localities where jobs are limited but also reduce job-search costs. According to data from the forthcoming Brazil Poverty Assessment (Lara Ibarra et al, forthcoming.), only 24 percent of poor own a car in Brazil. In rural areas, the ownership of motorcycles of poor is 45 percent which is higher than the average motorcycle ownership at the national level of 26 percent. Table 3: Employment status of work-able individuals in Cadastro Único, 2019 BF - Extreme Poor BF - Poor Cadastro Único no BF Work-Able Adults (Thousand) 15,528 2,525 16,772 Work-Able Adults (% HH members) 59% 49% 68% Out of work 59 51 55 In work 41 49 45 Occupation (% employed) Self-employed 64 76 45 Formal private worker 2 8 35 Formal public worker 0 2 6 Informal wage worker 4 6 6 Temporary rural 26 7 6 Unpaid workers 5 1 2 Employers - - 0.2 Average labor income in BRL 99.8 347 705 Source: Cadastro Único 2019 Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. The extreme poor in Cadastro Único have similar employment rates as the poor, but their labor market vulnerability is higher. Almost all work-able adults receiving BF are in households classified as extreme poor in Cadastro Único ( Table 2). Surprisingly, extreme poor families have more work-able adults than poor families in 2019 (59 percent versus 49 percent), but less than other families in Cadastro Único not receiving BF. Moreover, the share of work-able adults in extreme poor families, who are working, is 41 percent—only a few percentage points lower than those in Cadastro Único. Rather higher degrees of labor market vulnerability characterize the extreme poor: higher share of temporary rural work, and earnings lower than one-third of workers in a poor household, and far lower than the minimum wage. This are several explanations for observed discrepancies between the survey and Cadastro Único data, linked to the different objectives of these two data sources. First, Cadastro Único captures information at the time when the family income is reported. As this is tied to benefit eligibility verification, this is likely a period when income is lower than average and in the context of high fluctuation of informal earnings (Morgandi et al. 2021). Second, as observed above, it is possible that registration in Cadastro Único of some family members is done more strategically than in PNADC. Since Brazil lacks any official registry of family units, Cadastro Único remains the best public source for this information, albeit imperfect. Third, 23 voluntarily reported income for those without BF show limited discrepancies between survey and administrative data.8 Fourth, Cadastro Único information can be as old as two years, due to update rules; however, the discrepancies with household survey data remain the same even when excluding households that did not update the registry in 2019. Box 1: Employability and Race Race plays an important role when analyzing the Brazilian labor market. Almost half of the working-age population are parda (colored), with a higher share among the poor. About 64 percent of self-declared BF beneficiaries are parda and 58 percent of those in Cadastro Único not receiving BF (Figure 14). A difference by race can be observed in employment rates. Parda and indigenous people have the lowest employment rates among all population groups (Figure 15). The quality in employment differs by race (Figure 16). About 60 percent of informal workers in Cadastro Único not receiving BF (self-employed and wage-employed) are parda, whereas only 54 percent and 50 percent of formal self-employed and formal wage-employed are parda. This difference decreases for BF beneficiaries; however, keep in mind that a large share of BF beneficiaries is parda. Figure 14: Population groups by race Figure 15: Employment rates by population group and race 100 80 80 60 60 40 20 40 0 20 Working Age P.c. Income < Self-declared BF Population 0.5 MWs 0 White Black Yellow Working Age P.c. Income < Self-declared BF Population 0.5 MWs Colored Indigenous Others White Black Yellow Colored Indigenous 8Mean labor income per capita of families living under 0.5 MW per capita is BRL 229 per month in PNADC, and BRL 249 in Cadastro Único. Higher discrepancy is observed for families in BF (BRL277 versus 60) 24 Figure 16: Occupation status by population groups and race 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Worker (CLT+Public+Mil Informal Self-Employed Informal Dependent Worker (CLT+Public+Mil Informal Self-Employed Informal Dependent Retired (Receives a Unemployed Formal Self-employed + Other individuals out of Retired (Receives a Formal Self-employed + Unemployed Other individuals out of Formal Dependent Formal Dependent pension) pension) labor force labor force Worker Worker Employer Employer P.c. Income < 0.5 MWs Self-declared BF White Black (Preto) Yellow Colored (Pardo) Indigenous Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. Differences by race not only in labor market outcomes but also in education access and outcomes is highlighted in the literature. Venturini et al. (2020) point out that out of the whole population, more than 75 percent of parda and 76 percent of black children in elementary school do not have access to a computer with internet, 53 percent of white children in elementary school have access. Those conditions are high in the north and northeast of the country—regions with higher poverty rates. 1.3. The profile of poor and vulnerable in the labor market 1.3.1. Self-employed: subsistence workers or entrepreneurs? Most of the self-employed, as defined in PNADC, are informal, male, between 25 and 44 years old and live in urban areas. While a large share of those already working is engaged in self-employment, the majority of individuals work as informal self-employed. Among BF beneficiaries, more than 80 percent of the self-employed are informal, and this rate is only slightly lower (77 percent) among individuals living below 0.5 MWs per capita. Around two-thirds of self-employed are men; interestingly, women are overrepresented within the formal self-employed. In addition, the preponderance of self-employed increases with age. Most of the self-employed are between 35 and 44 years, followed by the 25–34 age group (Annex 1). Formality rates do not differ substantially between urban and rural areas. In addition, formality rates are higher for women (21 percent) than for men (17 percent). Among other poor in Cadastro Único not receiving BF the formality rates are slightly higher (Table 4). Table 4: Formality rates of the self-employed in PNAD C % of Self-Employed who are formal Self-declared BF Rural 16 Urban 19 25 % of Self-Employed who are formal Female 21 Male 17 Total 18 Poor (Per capita Income < 0.5 MWs) Rural 26 Urban 23 Female 26 Male 21 Total 22 Source: PNAD 2019. Formal are defined as self-employed reporting to be registered as firm (CNPJ) or to contribute to INSS. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. Formal and informal self-employed operate in very different sectors, and most of the formal self- employed still appear in low-wage sectors. Overall, a large share of self-employed in PNADC work in agriculture cattle farming, retail, and vehicle repairs. However, there are differences between formal and informal sectors. The two most prevalent sectors where formal self-employed work are agriculture and domestic services (Figure 17, Figure 18, Figure 19 and Figure 20). These are not the typical sectors associated with formal work, and likely the findings show the effect of policies to extend social protection rights, such as to farmers, in the previous decade, or potentially misclassified dependent employment relations in domestic work. There are also important gender sectoral segregations: construction and agriculture are particularly common among men, while domestic services are a key source of self- employment for women. These findings are in line with the literature looking at gender differences by sector (Araujo and Lombardi 2013; Bruschini and Lombardi 2000). Silveira and Leão (2020) refer to ‘occupational ghettos’, meaning that a concentration of a certain race and sex exists in some sectors in Brazil. 26 Figure 17: Female formal self-employed by sector Figure 18: Male formal self-employed by sector Figure 19: Female informal self-employed by sector Figure 20: Male informal self-employed by sector Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. Cadastro Único is more likely to misclassify workers into self-employment. The total share of workers declaring to be self-employed is much higher in Cadastro Único than what is declared in PNADC (Table 5). Moreover, in PNADC around one-third of self-employed are female, whereas the data in Cadastro Único are almost the opposite. There is also a weak correspondence with regard to sector of employment. Such findings suggest the importance of aligning definitions and leveraging all administrative data available on the self-employed in the future. There is little correspondence between the self-employed in Cadastro Único with registration in the CNPJ registry of formal entrepreneurs (MEI). Cadastro Único does not collect information for formal registration of the self-employed, unlike the case of formal wage employees. Through data cross-checks, we estimate that only 3 percent of adults in BF, and 5 percent for those in Cadastro Único out of BF, are registered in MEI in the firm registry (CNPJ registry) of the tax authority (Receita Federal do Brazil [RFB]) (see Figure 21). Such registration rates are much lower than observed in PNADC (where about 20 percent of self-employed with BF or poor declare to be ‘formal’). At the same time, almost half of the adults found in the MEI registry and in Cadastro Único report other work states than self-employment or employer. 27 When disregarding the declared work status, the total number of MEI is similar to the total formal self- employed in PNADC. This may have several possible explanations.9 Table 5: Descriptive statistic - Characteristics of adult self-employed in Cadastro Único, December 2019 In Bolsa Familia Not in Bolsa Familia Number (in thousands) 5,266 3,515.5 % of all work-able adults 34 21 % within categories: Female 61 51 Male 39 49 18–24 10 9 25–34 31 24 35-44 32 27 45–54 19 24 55–64 8 15 Rural 21 12 Urban 79 88 None 5 3 Incomplete Elementary 14 10 Elementary/Incomplete Middle 31 28 Middle/Incomplete High school 23 21 High school/Incomplete Tertiary 26 35 Tertiary 1 4 Source: Cadastro Único 2019. Note: Groups are restricted to individuals of working age (18-64) and not in education and able-bodied. 9 One is that in recent years MEI has been used by firms to formalize dependent workers instead of a labor contract. In fact, informal wage employees in Cadastro Único earn twice as much as the self-employed. Second, PNADC contains specific questions regarding registration as a firm (CNPJ) that allows more correct classification of work status at the time of interview. Nearly 42 percent of employers in Cadastro Único are MEI, followed by 7 percent of all self-employed and 6 percent of all who declare to be informal dependent employees. 28 Figure 21: Share of work-able adults in Cadastro Único 2019 registered as MEI and mean declared income in Cadastro Único 45 1,558 1600 Mean labor income declared 40 1434 1400 35 1,255 1275 % registered as MEI 1216 1200 30 1,083 974 1000 25 890 903 835 846 800 20 733 628 585 600 15 548 528 490 10 410 357 408 400 260 5 205 196 200 116 99 0 0 23 0 % with MEI (Left) Declared income (MEI) Declared income (not MEI) Source: Team computation based on Cadastro Único 2019 and CNPJ Database 2019. Becoming an employer is a typical sign of having a more established business: only a small share of self- employed are employers. Only 5.4 percent of formal self-employed in BF are employers. This share is— as expected—lower for informal self-employed, where 2.3 percent are employers. Among those with a per capita income below 0.5 MWs the share of formal employers is slightly higher (7.5 percent) (Annex 1). Importantly, administrative data suggest that nearly half of the self-declared employers in Cadastro Único are registered as MEI—this indicates a better identification of this category than other work statuses. Earnings increase with education levels, but only among formal self-employed. According to data in PNADC, formal self-employed have a higher take-home pay than informal self-employed and income increases with education levels (Figure 22 and Figure 23). Evidence from the profiles of users of the Progredir platform also shows how knowledge of basic business accounting principles is correlated with other more advanced business practices and better credit access (see Box 2). The gap in earnings is also visible in the administrative data, especially between self-employed and employees. There is also a significant earning gap between men and women. Monthly earnings are lower for female than for male self-employed. This is the case even though women are better educated than men on average. 29 Figure 22: Monthly take-home earnings for self- Figure 23: Self-employed in Bolsa Familia employed in Bolsa Familia in BRL, by Sex monthly income by education level 1,608 1,208 1,305 1,126 1,184 995 780 849 854 796 614 660 723 486 589 Male 1,282 Incomplete Incomplete basic Complete basic Complete secondary No instruction or up Incomplete superior Complete superior secondary 470 Female to 1 year 833 Informal Formal Formal Informal Bolsa Familia Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. Earnings from self-employment represent the largest share of total income for families in BF. Among self-employed BF beneficiaries, income from their work in 2019 represents 77 percent of total household income, followed by the BF benefit, which represents 11 percent of the income (Figure 24). While essential, such labor income is subject to high income volatility from quarter to quarter (Morgandi et al. 2021). For such families, the BF benefit is particularly valuable as it represents a stable source of income over time. In Cadastro Único, the share of income from self-employment (57 percent) is less than in PNADC (Figure 26). Such discrepancy with PNAD could have several reasons: part of reported incomes in Cadastro pertain to previous years and were not updated; households could be more likely to enroll in Cadastro Único in months when their volatile labor income is the lowest. The fact that households already report such a high share of total income, albeit informal, is already remarkable compared to the experience in other countries. Figure 24: Income sources for self-employed Bolsa Familia families Figure 25: Share of self-employed in Bolsa Familia for which the BF benefit represents 1% the respective proportion of income (in %) 0% 1% 0% Labor income 1% 2% Bolsa Familia 7% 5% 1% BPC 11% (0-0,25) Unemployment Insurance 16% Pension [0,25-0,5) Other Benefits 77% [0,5-0,75) Food Alimony 78% [0,75-1] Rent Income Other Income Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. 30 Figure 26: Income sources for self-employed in Bolsa Família according to Cadastro Único 1% 1% 0% 0% 2% Labor Income Bolsa Família SD Pension 38% Alimony 57% Donation Others Source: Cadastro Único 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. In this calculation the Bolsa Família benefit was set to the average family benefit observed in 2019. Box 2. Profiles of current and aspiring self-employed in the Progredir Platform Progredir served, until the pandemic, as a digital platform to broker credit and technical support for aspiring and current entrepreneurs in Cadastro Único by territorial service providers. The main objective was to leverage digital technology to reduce information asymmetry preventing poor adults from taking advantage of locally available services and jobs. The platform offered three main service lines: (a) support workers in labor market insertion (by teaching them how to prepare a resume and how to submit it, and by linking individuals with the local employment office); (b) intermediation of local training programs; (c) intermediation to banks offering microcredit loans — financial support to individuals who wish to become entrepreneurs or those who already own a business and wish to expand it. According to data from the Ministry of Citizenship (MoC), nearly 50,000 individuals in Brazil between 2017 and 2021 expressed interest in receiving a microcredit loan through the platform. About 70 percent of them were aspiring entrepreneurs, the remainder reported already having a microbusiness. The highest concentration of users was in São Paulo (15.3 percent), Pará (10.9 percent), and Bahia (10.6 percent). Practically all prospective entrepreneurs (99 percent) believe that they need a microcredit loan, though 68 percent declared not having a collateral guarantee (which was required at the time). Moreover, 76 percent were interested in acquiring more knowledge on how to run a business, and the most requested topics by the aspiring entrepreneurs were financial management (40 percent), marketing (29 percent), followed by human resource management (6 percent). Respondents were mostly interested to engage in the retail sector (62 percent), followed by services or other industries (22 percent each). Few (7 percent) are interested in manufacturing. Most of existing ‘entrepreneurs’ are actually recent self-employed, unsatisfied with their current business line. About half of the existing entrepreneurs reported their businesses to be very recent (48 percent for less than two years), and only 13 percent have it for more than five years. Consistent with the profile of all self-employed in Cadastro Único in this study, very few (15 percent) had an employee. The reported number of activities was vast, but 90 percent declared that they considered changing the sector of activity, which can be interpreted as dissatisfaction with the results of their current efforts. About 44 percent of entrepreneurs rely on some other source of funding, besides their business revenue. Only 11 percent had a diverse source of financing. About a third of respondents lacked basic knowledge or practice of business accounting: Some 13 percent of respondents have no control over inflows and outflows of money in their business, 12 percent do not separate 31 business purchases from own consumption, and 42.2 percent of entrepreneurs do not calculate the cost per unit of the product or service provided (either because they do not find it relevant or because they cannot). About 67.4 percent do not know how to differentiate fixed and variable cost of their product/service. Finally, 7.8 percent do not know how much they produce per month; 34 percent know how much they produce but their businesses make no profit; 58.2 percent know how much they produce and make profits. Users with limited business knowledge also tend operate with very basic business practices. The ability to calculate the cost per unit of production can be used as a proxy of generally limited business knowledge or ability. The share of low-knowledge self-employed is higher in poorer states, such as Piauí (53 percent). Low-knowledge self-employed much more often only rely on personal incomes (58 percent) or savings (34 percent) as a source of financing. More than half purchase inputs exclusively from the local market, and thus reduce the scope for profit margin. Such respondents are also more likely to advertise uniquely by word of mouth (74.15 percent), and less through Facebook (36.1 percent) or Whatsapp (21.65 percent). About 89 percent believe that what their businesses need to thrive is more money, however 35 percent of these entrepreneurs had never applied for a loan. Only 21 percent of low- knowledge entrepreneurs believe that they need a qualification course for their businesses to thrive. Source: Ministério da Cidadania 1.3.2. Formal workers in Cadastro Único: more vulnerable than conventional workers The linking of several administrative datasets allows exploring the profiles of formal employees in low- income households and in BF. The present analysis aims to understand qualitatively how BF beneficiaries engage in the formal labor market, and to what extent they are able to escape the vulnerable status that are observed for other forms of work described in this note. The qualitative aspect of labor market engagement by the poor received limited attention in the literature, compared to the more researched issue of whether BF affects incentives to participate in the formal sector. On the latter, it is important to remember that the empirical evidence rejects the idea that BF generates disincentives to participating in the formal labor market10. For the purpose of this analysis a database containing all members of Cadastro Único who were formally dependent employed at some point in 2019 was constructed. The administrative dataset Cadastro Único was merged with the annual social information report (RAIS) based on identification numbers (Número de Identificação Social [NIS] or Programa de Integração Social [PIS]) of the individual. The RAIS database, which is administered by the Ministry of Labor and Employment, contains information about formal public and private labor relationships. RAIS allows individuals to have several employment links if employment was changed during the given year. For the present analysis the most recent employment link or the link with the higher wage was chosen. Furthermore, individuals with a negative wage or a wage of zero were excluded from the database. Finally, the sample was restricted to people between 18 and 64 years, who were able bodied and not enrolled in full-time education. 15 million adults enrolled in Cadastro Único had a formal labor contract at some point in 2019. Among work-able BF recipients, about 9 percent held a job recorded in RAIS at some point in 2019. Among individuals in Cadastro Único not receiving BF, the share of people in RAIS was substantially higher—28 percent. The participation in the formal labor market declined during the last decade: it was 18 percent in 2013, decreasing to 12 percent in 2016, and eventually reaching 9 percent in 2019. 10While selected studies find no short-term effects of BF on formal labor market participation (De Brauw et al. 2015; De Oliveira and de Araújo Carvalho 2009), others find positive impacts of BF (Fruttero, Leichsenring, and Paiva 2020). Studies on long-term effects are still rare, however, De Oliveira and Chagas (2020) find positive long-term effects on labor market participation, which are stronger for boys, smaller cities, and families with never formally employed parents 32 However, BF beneficiaries represented only 4 percent of all formal workers in 2019, and other individuals registered in Cadastro Único not receiving BF represented an additional 11 percent. Considering those that participated at some point of the formal labor market, the share in the formal employed strongly differs across extreme poor, poor, and not poor (Table 6). Table 6: Presence of Cadastro Único adults in RAIS and classification as formal worker in Cadastro Único Extreme poor in BF Poor in BF (Per All BF In Cadastro Work-able adults in Cadastro Único: (Per capita Income capita Income beneficiari Único not in BF below BRL 89) below BRL 178) es % in RAIS (2019) 8 14 9 28 % in RAIS 2016 10 18 12 31 % in RAIS 2013 15 25 18 31 % classified as formal worker in 1 5 2 15 Cadastro Único (2019 updates) % of group in all RAIS jobs (2019) 3 1 4 11 Source: RAIS 2019, Cadastro Único December 2019, Folha de pagamento 2019. Note: Cadastro Único sample is restricted to individuals of working age (18-64) and not in education and able-bodied. RAIS in 2019 counted a total of 56 million adults. The discrepancy between RAIS and Cadastro Único on formal employment is significant, and in part depends on the frequency of registry update. Figure 27 compares the labor status on the records in Cadastro Unico in December 2019, and the participation in a formal job, according to RAIS, in the same month. About 70 percent of self-declared formal employees (private or public) in CadU indeed appear in the RAIS, but so do also 30 percent of those classified as working without “carteira assinada”. Mismatches can have several explanations: (i) the delays in the generation and transmission of employment data; (ii) timing of Cadastral update by beneficiaries, which often happens when labor income falls and needs increase, and (iii) classification error or undeclarations. Further analysis shows that time mismatches are likely the primary reason. Figure 66 in the Annex 1 shows that the coincidence between RAIS December 2019 and labor status in Cadastro Unico in December 2019 improves for the subset of adults who updated their record in the same month. And the share of BF adult beneficiaries in RAIS at some point in 2019 was as much as 9%, but this drops to only 4% among those individuals that updated their Cadastro in December 2019. Improving channels to reduce discrepancies matters, because undetected formal labor income in CadU leads to elimination from BF at the time of registry verifications. The Regra de Permanencia offers beneficiaries an option to increase their formal labor income and remain in the BF program for two years: however, only 7 percent of exits from BF happened in 2019 due to completion of two years in the Regra de Permanencia. Instead, more than 50 percent of BF exits in 2019 happened after a registry verification process that found discrepancy in declared income (Fietz et al. 2021). An automatic update of CadU to take into account new events in individuals’ working lives, coupled with automatic entry of the family in Regra de Permanencia, could reduce the negative impact of cross-checks. 33 Figure 27: Share of work-able adults in Cadastro Único in December 2019 who were employed according to RAIS in the same month; and mean declared income in Cadastro Único 80% 2,000 Mean labor income declared 70% 1,800 1,600 60% 1,400 50% 1,200 % in RAIS 40% 1,000 30% 800 600 20% 400 10% 200 0% 0 % in RAIS (Left) Mean labor income (RAIS) Mean labor income (not in RAIS) Note: Mean labor income is based on self-declared values in Cadastro Único, disaggregated according to sample of individuals who had one formal job in RAIS and those that did not. Most Cadastro Único and BF adults who participate in the formal labor markets are women and young people. BF recipients are on average younger than individuals not receiving BF. Consistent with the fact that young people in BF are on average better educated than their parents, youth below the age of 35 represented 60 percent of all formal employees, with men being slightly older than women. This pattern is less pronounced among those in Cadastro Único, where youth below 35 years represent 48 percent of all formal employees. However, formal employees in Cadastro Único are better educated with more than 50 percent having at least a high-school degree. Among BF beneficiaries, only 44 percent have at least a high-school degree. Also formally employed women are again better educated than formally employed men (Table 7). Table 7: Formal wage employed sex, age, and education distribution Cadastro Único work-able adults in formal wage work in 2019 (RAIS) Extreme poor in Bolsa Poor in Bolas Familia (Per In Cadastro Único not Familia (Per capita income capita income below BRL in Bolsa Familia below BRL 89) 178) Men 47 42 46 Women 53 58 54 Age All Male Female All Male Female All Male Female 18–24 23 23 22 20 21 18 16 18 15 25–34 39 36 41 35 29 41 32 30 33 35–44 26 25 27 30 33 32 29 27 30 45–54 10 12 8 8 13 8 17 17 16 55–64 3 4 2 1 4 1 7 8 6 34 Education None 3 4 1 2 4 1 1 2 1 Literate 8 11 4 7 10 4 5 7 3 Compl. Elementary 22 27 18 22 26 18 17 21 13 Compl. Middle school 23 23 24 27 26 27 18 20 17 Compl. High school 42 33 50 41 32 47 51 45 55 Compl. Tertiary 2 1 3 2 1 3 8 5 11 Source: RAIS 2019, Cadastro Único December 2019, Folha de pagamento 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. Education computed according to Cadastro Único data. Most men in BF and with a formal wage job are engaged in the agriculture, construction, trade and retail sectors, while women receiving BF are largely in the retail, trade, and public administration sectors. A difference by sex in terms of sectors can be observed (see Figure 28, Figure 29, Figure 30 and Figure 31). In addition, the sector involvement differs substantially between people receiving BF and other poor. For male BF beneficiaries, agriculture plays a substantial role. Whereas other poor male in Cadastro Único also work in agriculture—here the largest share works in retail. For women, the picture is slightly more homogeneous. A large share of women works in retail, public administration, and diverse services. Figure 28: Firm sector code of male BF beneficiaries Figure 30: Firm sector code of female BF beneficiaries Figure 29: Firm sector code of male Cadastro Único Figure 31: Firm sector code of Female Cadastro Único no no BF BF Source: RAIS 2019, Cadastro Único December 2019, Folha de pagamento 2019. 35 Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. The average wage for full-time employees increases with household income, age, and is higher for men. Extreme poor male BF beneficiaries earn about 17 percent more than extreme poor women, for instance. Older poor BF beneficiaries (55–64 years) earn around 10 percent more than beneficiaries between 25 and 34 years. In addition, earnings increase with education levels (Table 8). Table 8: Average monthly wage in BRL (full-time employees only) Extreme poor in Bolsa Poor in Bolas Familia In Cadastro Familia (per capita (per capita income Único not in income below BRL 89) below BRL 178) Bolsa Familia All 1,394 1,394 1,605 Male 1,506 1,504 1,757 Female 1,287 1,312 1,467 18–24 1,329 1,336 1,394 25–34 1,387 1,409 1,562 35–44 1,436 1,399 1,667 45–54 1,431 1,434 1,714 55–64 1,452 1,424 1,767 None 1,331 1,321 1,447 Literate/Incomplete Elementary 1,389 1,378 1,474 Elementary/Incomplete Middle school 1,392 1,364 1,526 Middle school/Incomplete High school 1,377 1,363 1,543 High school/Incomplete Tertiary (college / technical course) 1,394 1,417 1,602 Tertiary 1760 1,772 2,159 Source: RAIS 2019, Cadastro Único December 2019, Folha de pagamento 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. Average monthly wage is shown for individuals working between 36 and 44 hours per week. Wage regressions show a premium for education, experience, and for male workers. Regressions to understand factors explaining higher hourly wages (Annex 2) identify a significant ‘male’ wage premium, even after controlling for education, sector of employment, age, and location. This is more pronounced among recipients of BF. Education is associated with an increase in the hourly wage, but particularly so for those with tertiary degrees. Again, this effect is stronger for individuals in Cadastro Único compared to individuals in BF. While having tertiary education completed increases the hourly wage on average by 26 percent for individuals being in BF, other poor individuals not in BF see a wage increase of almost 41 percent. There is also a significant return to experience, as proxied by age. 36 Figure 32: Wage distribution of individuals in Cadastro Único Source: RAIS 2019, Cadastro Único December 2019, Folha de pagamento 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. BF and Cadastro Único workers have a higher likelihood of being in non-standard, temporary employment contracts. We estimate in RAIS the share of workers in BF and Cadastro Único with non- standard contracts, which include several modalities.11 Temporary employment contracts in Brazil are much more restricted than in other countries (Silva et al. 2021). They can offer opportunities to include workers at the margin in the formal labor market because of their lower costs and also because they carry fewer rights. For instance, in Brazil, fixed term and temporary work contracts do not give the right to request SD or severance pay (multa). Table 10 shows that about 30 percent of those in BF and 15 percent of those in Cadastro Único work in non-standard contracts, which is more than in the general population. About half of these non-standard workers are in fixed term contracts. The share of open-ended contracts is higher for individuals in Cadastro Único not receiving BF than for BF beneficiaries. While almost 70 percent of individuals in Cadastro Único not receiving BF have a CLT open-ended contract, the share of extreme poor and poor receiving BF is 60 percent and 64 percent, respectively. The majority of employment relations for our analytical groups are less than a year, but tenure increases with total household income. The average job tenure of extreme poor BF beneficiaries is 9 months, compared to 14 months for poor BF beneficiaries and as much as 43 months for people in Cadastro Único not receiving BF. Job tenures are even longer for those outside Cadastro Único, presenting an average of 11Trabalhador avulso (freelancer), temporary worker (Law No. 6019), Urban worker linked to a legal entity employer by employment contract governed by the CLT, for a determined period or right work, Urban worker linked to an individual employer for employment contract governed by the CLT, for a determined period or right work, Rural worker linked to a legal entity employer by employment contract (Law No. 5,889/1973), for a period determined, Rural worker linked to an individual employer for employment contract ( Law No. 5,889/1973) for a period determined, Fixed-Term Employment Agreement (Law no. 9,601), Fixed-Term Employment Agreement (Law no. 8,745; amended by Law No. 9,849), Fixed-Term Employment Agreement, governed by State Law, Fixed-Term Employment Agreement, governed by Municipal Law. 37 69 months. Moreover, young people tend to have the shortest tenures: the few older workers in the BF sample had a tenure of 20 and 28 months (extreme poor and poor, respectively), the 18–24 years held their jobs on average only 7 to 8 months (Table 9). Figure 65 (Annex 1) also shows that for the extreme poor, tenure remains short or very short regardless of age, while in less vulnerable groups tenure increases with age. Table 9: Mean tenure of individuals in Cadastro Único in months In Cadastro Extreme Poor in BF Único not in poor in BF BF All 9 14 43 Male 9 15 41 Female 9 13 44 18–24 7 8 15 25–34 8 11 27 35–44 11 17 45 45–54 12 25 70 55–64 20 28 104 None 10 25 49 Literate/Incomplete Elementary 10 26 53 Elementary/ Incomplete Middle school 10 14 46 Middle school/Incomplete High school 9 11 35 High School/Incomplete Tertiary (college / technical course) 9 14 41 Tertiary 10 13 57 Source: RAIS 2019, Cadastro Único December 2019, Folha de pagamento 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. The average tenure in a formal job for adults in BF is significantly shorter than for those in Cadastro Único, and this gap has been widening over time. Out of all adults in Cadastro Único that were employed at some point in 2019 in RAIS, only 68 percent maintained an active work relationship by December 2019, with BF beneficiaries being under average (58.2 percent) and poor non-BF beneficiaries above average (71 percent). Figure 33 unpacks the full distribution of the tenure for three different groups: (a) the extreme poor in BF, (b) the poor in BF, and (c) other poor in Cadastro Único not receiving BF.12 This shows that a much larger share of workers in BF found their work very recently (which is an indication of high turnover). The figure also shows that the tenure of the extreme poor in BF used to be more similar to that of less vulnerable workers in Cadastro Único before the 2014 crisis, but the gap in tenure has been widening over time. It is however possible that this is due to higher entry in the formal labor market of such vulnerable workers compared to the past, in light of more flexible labor regulations. 12To control for the fact that more young people may be present in group (a) and (b), the sample was restricted to people ages between 25 and 64. 38 Figure 33: Tenure in formal employment for extreme poor in BF, poor in BF, and other registered in CadU 2013 2016 2019 Source: RAIS 2019/2016/2013, Cadastro Único December 2019/2016/2013, Folha de pagamento 2019/2016/2013. Note: Groups are restricted to individuals of working age (25–64) and not in education and able-bodied. Note that the first age group (18–24 years old) was excluded from the sample above. Considering the reasons for job separation, as many as 40 percent of workers in BF and in RAIS exit formal employment due to expiration of their term contracts. Conversely, fewer workers in BF are dismissed ‘without just cause’ by their employer: this matters in Brazil because only such dismissals open the right to the unemployment insurance SD. The share of those dismissed without just cause in fact rises as the level of workers’ vulnerability falls, and is the highest for those not in Cadastro Único (Table 10).13 Zooming in on those workers that have an open-ended CLT contract, workers in BF, and to a lesser extent in Cadastro Único, are much more likely to be terminated early, before the end of the statutory probation period, compared to those outside Cadastro Único. During this early termination period, the right to SD is unavailable (Table 10). This is consistent with the finding in the literature of the effect of this particular regulatory protection on employment prospects of the most vulnerable workers. Termination of those having an open-ended CLT contract with the dismissal cause ‘Termination of employment contract’ spikes at months 3–5. Table 10: Contract type and dismissal cause (%) In In RAIS Extreme Poor in Cadastro not in poor in BF Único not Cadastro BF in BF Único Contract Type Open-ended contract 60 64 69 69 Public 8 7 13 16 Fixed-term and temporary 30 27 16 14 Other (for example, Apprentice) 1 2 1 1 Reason of dismissal All Contract types Termination without just cause at the employer's initiative 41 42 59 51 13For an overview of the requirements to receive SD please see: https://www.caixa.gov.br/beneficios-trabalhador/seguro- desemprego/perguntas-frequentes/Paginas/default.aspx 39 Termination of employment contract 39 34 22 18 Termination with just cause at the employee's initiative or 22 dismissal upon request 18 21 15 Other 3 3 4 10 Open-ended CLT contracts Termination without just cause at the employer's initiative 50 50 70 60 Termination of employment contract 24 23 10 8 Termination with just cause at the employee's initiative 23 22 16 22 Others 3 4 4 10 Open-ended CLT contracts and tenure > 5 months Termination without just cause at the employer's initiative 77 76 83 69 Termination of employment contract 3 1 1 1 Termination with just cause at the employee's initiative 15 17 12 19 Others 5 6 4 12 Source: RAIS 2019, Cadastro Único December 2019, Folha de pagamento 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. Average monthly wage is shown for individuals working between 36 and 44 hours per week. Among formal workers in BF who were involuntarily dismissed, only 20 percent were able to find a new job within the same year. Out of all formally employed in 2019, 42 percent of BF beneficiaries were involuntarily dismissed and 20 percent of those found another CLT employment within 2019. Among those in Cadastro Único not receiving BF, the share of workers involuntarily dismissed is lower (32 percent), whereas the share of people that got reemployed within the same year is higher (26 percent). Individuals in Brazil, that have a per capita income above 0.5 MWs and are therefore not registered in the social registry Cadastro Único, show an involuntary dismissal rate of 23 percent (well below the BF involuntary dismissal rate). In addition, almost 30 percent of those found reemployment in the same year. Table 11: Involuntarily dismissed and formally reallocated in 2019 Involuntarily Dismissed Involuntary Dismissed and reemployed in 2019 BF 42 20 Cadastro Único – no BF 32 26 Not in Cadastro Único 23 29 Source: RAIS 2019, Cadastro Único 2019. Note: This table presents the percentage that were dismissed at some point in 2019 for reasons out of their control, that is, employer’s decision or end of contract period (usually the temporary workers), and the percentage of individuals that after involuntary dismissal found another formal job and were employed in December 2019. 1.3.3. Unemployment in poor households: a torn safety net The nature of unemployment among the poor is best explored using PNADC. Cadastro Único does not provide information on participation in active job search, which differentiates statistically the unemployed from other out-of-work individuals. Within the BF population, youth and women on average have the highest unemployment rates and also represent majority of the unemployed. Though being on average better educated than men, 14 percent of adult female BF beneficiaries are unemployed, compared to 10 percent of men. The opposite occurs among the nonrecipients of BF (22 percent of men are unemployed, compared to 18.5 percent of women) (Figure 35). Most of the unemployed are young, between 18 and 34 years old, underscoring the fundamental challenge of entering the first employment after completing or abandoning education for this age group. 40 Most jobseekers directly contact employers or use personal contacts (relatives, friends, or colleagues) to find work. This reliance is higher among BF and Cadastro Único individuals than in the rest of the population. Among BF beneficiaries’ 18.5 percent of jobseekers rely on personal contacts, while only 5.7 percent of the rest of the population do so. As expected, only a small share of people consulted with the municipal or state employment agency or with the National Employment System (SINE). Interestingly, among BF beneficiaries, 2.5 percent of jobseekers consulted with employment agencies, while 1.9 percent among other poor in Cadastro Único, and 1.8 percent among the rest of the population did so (Figure 34). Figure 34: Principal way of searching for work among the unemployed 100% Other such as answered a job in a newspaper 90% or consulted private employment agency 80% Took steps to start its own business (financial 70% resources, place for installation, equipment, 60% legalization, etc.) 50% Have consulted or registered with a municipal or state employment agency or with the 40% National Employment System (SINE) 30% Consulted a relative, friend or colleague 20% 10% Contacted the employer (in person, by phone, by email or through the company portal, 0% including sending a resume) Self-declared BF P.c. Income < 0.5 MWs Rest of working age population Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied and unemployed. Qualitative work conducted as background for this study reveals the extent of race discrimination by employers against young Afrodescendant jobseekers, who are the majority of youth in BF. Recent qualitative work conducted as background to this report highlights widespread discrimination that Afrodescendant youth, with varying levels of education, experience when approaching employers in search of a first opportunity due to their physical appearance. Such accounts are consistent with several recent estimates of labor market discrimination. Afrodescendant women earn about 30 percent less than white males. And even when controlling for education level, location, and sector of employment, women and Afrodescendants are paid less. Such discrimination in turn hampers the opportunity to realize the returns from hard-gained education, and reduces the perception that education opens up new opportunities. In fact, compared to white men, the return for an additional year of education is 0.5 percent lower for white women, 2.5 percent lower for Afrodescendant men, and about 2.9 percent lower for Afrodescendant women. Among all those declaring to be unemployed in Brazil, BF is a more important source of income support than Seguro Desemprego. Comparing the number of annual claims and the number of unemployed computed by the Statistical Institute of Brazil (Instituto Brasileiro de Geografia e Estatística [IBGE]), Morgandi et al. (2020) estimated that on average 18 percent of the unemployed in Brazil receive SD. Coverage of SD is also traced in PNADC, though as it occurs for other benefits, underestimated (possibly due to voluntary underreporting). Nonetheless, the survey suggests that SD covers only a small percentage of active jobseekers (4 percent), and even fewer (less than 1.5 percent) jobseekers living in 41 poor families. Coverage of the unemployed by BF is much higher (25 percent), especially among poor unemployed (43 percent) (Figure 36). This emphasizes the fundamental role of the benefit in supporting individuals looking for work in Brazil, and the potential relevance of measures that support active job- seeking efforts among recipients. Figure 35: Unemployed as a share of working-age Figure 36: Share of unemployed poor receiving Seguro population by gender Desemprego and Bolsa Familia 45 22% 40 18% 35 30 14% 25 20 11% 15 10 5 0 All Unemployed Unemployed and Unemployed and P.c. Income > 0.5 P.c. Income < 0.5 MWs MWs Female Male Female Male P.c. Income < 0.5 MWs Self-declared BF Receiving SD Receiving BF Both Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. The ability of BF to serve as an absorber from labor market shocks, as a complement or a substitute of SD, however, depends on the availability and timeliness of its delivery. As shown above, 10.6 percent of the individuals in RAIS are in Cadastro Único, but not already in BF. The average income for those families is understandably higher and this explains the ‘no qualification for the benefit’ while in work. However, when income falls, these families could qualify for the benefit. Of those RAIS workers (in Cadastro Único) that got dismissed in the first quarter of 2019, 34 percent were already in BF. About 63 percent of workers did not enter BF in the later quarter of the same year and only 2 percent entered BF at some point during 2019 (1.6 percent in the second quarter, and 0.1 percent in the third and fourth quarters) (Figure 37). Among the few who make it into the program, the waiting time between dismissal and entry is about two months (Figure 39). 42 Figure 37: Workers in Cadastro Único dismissed in Figure 39: Time between dismissal and BF entry for first quarter 2019, by BF status in Q2, Q3, Q4 individuals in Cadastro Único not receiving BF at the entered in was point of dismissal BF, 1.8% already in BF, 33.3% did not enter BF, 63.0% Figure 38: Workers in Cadastro Único dismissed in first quarter 2016, by BF status in Q2, Q3,Q4 entered in BF, 5.7% was already in BF, 36.9% did not enter BF, 51.7% Source: Cadastro Único 2019, RAIS 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. Even among formal workers, the majority do not have the necessary job tenure to receive SD after their current job. To be eligible for SD, a work tenure of 18 months within the last 24 months needs to be achieved. Figure 40 indicates that only a minority of individuals achieved this target in their current job. Among the extreme poor BF beneficiaries, only 7 percent have a tenure of 18 months or above. For individuals in Cadastro Único not receiving BF, 34 percent have a tenure of 18 months or above (Figure 40). Nevertheless, the 18 months do not need to be in the same job, meaning that the accumulated tenure of multiple jobs may account for the necessary condition to receive SD. However, it may be assumed that only a small fraction of the extreme poor is eligible for SD. Box 3: Enhancing financial inclusion among workers to build resilience against income shocks The recent World Bank study “Increasing the resilience of low-income workers in Brazil – financial instruments and innovations” seeks to inform the design of a financial product and complementary public policy actions that can strengthen the resilience of families to income shocks, in particular for families in the Bolsa Família and Cadastro Único who work in the informal economy. The Report proposes a three-folded policy mix: (a) leveraging the Cadastro Único and Bolsa Família as platforms for financial inclusion – for example with the creation of preventive savings accounts, focus on grading beneficiaries, and using Cadastro Único as a multi-provider partnership platform; (b) design voluntary savings instruments for low-income families to build resilience – fostering the ability to save through 43 behavioral stimuli, monetary incentives (counterparts, rewards for goals) and integrated financial education resources in digital bank accounts; and support (c) complementary financial interventions – such as encouraging transparency measures to allow identifying better value credit, delivering financial education for self-employed workers, and packaging microinsurance against rare but catastrophic risk events with other financial products. Source: Morgandi et al, 2021. Figure 40: Job tenure in current job (below and above 18 months) 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Extreme poor in Bolsa Poor in Bolas Familia (P.c. In Cadastro Único not in Familia (P.c. Income below Income below BRL 178) Bolsa Familia BRL 89) Tenure below 18 months (in last job) Tenure equal or above 18 months (in last job) Source: Cadastro Único 2019, RAIS 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. 1.3.4. The Out of Labor Force population is hindered by caretaking duties or location Around one-third of the poor and vulnerable are out of the labor force. According to PNADC, among BF beneficiaries and other poor individuals with a per capita income below 0.5 MWs, 30 percent are not working or actively looking for employment. This is twice the rate of inactivity among the rest of the population (16 percent). Even though better educated, 81 percent of the people out of labor force are women (Figure 41 and Figure 42). Traditionally, women are more involved in unpaid work—such as housework or caretaking duties, to enable their spouse to work in paid labor (Biroli 2016). 44 Figure 41: Out of labor force by gender Figure 42: Out of labor force by education and gender Self-declared P.c. Income < Male 0.5 MWs 19% 26% Female Male BF 81% 74% Female 0% 50% 100% No instruction or up to 1 year Incomplete basic Self-declared BF P.c. Income < 0.5 MWs Complete basic Incomplete secondary Female Male Complete secondary Incomplete superior Complete Superior Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. A major reason reported by women for being out of labor force are caretaking duties, most likely for children. As shown in Figure 43, such constraint is seldom reported by men. This constraint becomes more prevalent with the number of children (0–5 years old) living in the same household (Figure 44). Contrary to expectations, the presence of elderly in a household is not correlated with labor supply of women in the data. A likely reason is that the household survey PNADC does not allow to distinguish healthy elderly from those in need for care or ill. Healthy elderly could actually favor female labor supply by caring for young children, and thus offset the negative effect of elderly in need for care. The demographic change and increasing dependency ratios in Brazil thus have an ambiguous impact. 45 Figure 43: Reason to be out of labor force Figure 44: Reason for female BF beneficiaries to be out of labor force, by number of young children Cadastro Único 100% Male Female 50% Bolsa Familia Male 0% Female No 1 child 2 children3 children 4+ children children 0% 20% 40% 60% 80% 100% Other Job proposal to start after the reference week Studyig but not enrolled in school Waiting for a response of measure Home and Caretaking duties Could not find an adequat job No work in the locality No Professional experience Too young/ too old Too young/ too old No Professional experience No work in the locality Home and Caretaking duties Could not find an adequat job Studyig but not enrolled in school Waiting for a response of measure Other Job proposal to start after the reference week Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. The lack of work opportunities in the locality is the main reason reported by most men, and many women, to be out of work and discouraged,. This is reported most often by men in rural areas (70 percent) compared to urban dwellers (37 percent). Such self-reported information is corroborated by the data: those who report that lack of jobs explains their state of joblessness are more likely to be surrounded by other adults out of work14. This shows that, beyond individual constraints, policies that support either mobility, migration, or connectivity of depressed areas with value chains and economic development remain key to enhance labor outcomes. 14 We identify ‘neighbors’ labor market status by estimating the average labor market state of other adults located in the in the same primary sampling unit of the respondent, and excluding the respondent. 46 Chapter 2: Identifying key groups and potential interventions Key messages of chapter 2 This chapter uses Latent Class Analysis (LCA) based on PNADC to identify and inform the development of tailored policies for the most common and homogeneous groups (classes). Using a LCA, the note identifies groups (or classes) of individuals with similar characteristics. This analysis helps suggest package interventions to serve individuals who frequently have similar and overlapping constraints. Among the in-work poor, 80 percent belong to three larger groups: - Full time formal employees (32 percent) - Uneducated informal underpaid workers (25 percent) - Educated underemployed mothers in the formal sector (23 percent) Among the out-of-work, 70 percent belong to the four more heterogenous largest groups: - Inactive low-educated women (20 percent) - Medium educated young jobseekers (19 percent) - Educated female caretakers (18 percent)Short-term urban unemployed (13 percent) From the LCA it becomes evident that some groups are easier to include in the labor markets, and some will require greater and multiple interventions A tailored set of policies are suggested for each of the three biggest classes of the out-of-work and in-work population. For the in-work population formalization interventions and daycare services are priorities. For the out-of-work population, more complex ALMPs and daycare facilities are possible ways forward to activate these classes. The previous profiles in Cadastro Único revealed the high heterogeneity of potential labor market participants and of potential beneficiaries of economic inclusion policies, with multiple overlapping characteristics. This section will use LCA based on PNADC to identify and inform the development of tailored policies for the most common and homogeneous groups (clusters) (see Annex 3 for a detailed overview of the methodology). We develop two models, one for the out of work population, and the other for those in work. Both models help picture the relationship between different labor market outcomes and constraints, as illustrated in the analytical framework in chapter 1 (see Figure 1 and Figure 2). 2.1 The in-work population The LCA for the in-work population in Cadastro Único identified five clusters (Figure 45): 47 Figure 45: In-work poor individuals by clusters Low Unpaid low educated Educated educated informal dependent male full-time informal female workers, 4.8% formal male in BF, employees, 15.0% 32.2% Educated informal Low-educated female underpaid workers in BF, informal 23.2% workers in BF, 24.8% Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education. • Educated male full-time formal employees. The first and biggest cluster (32 percent) consists of rather well-educated, middle-aged males. Over 50 percent of the cluster is between 35 and 54 years and almost half (49 percent) have secondary education or above. The great majority (81 percent) have a full-time dependent work contract, which is in line with more than 50 percent not being in BF. The majority in the cluster are the only earners in the household. For 33 percent of the cluster, a second earner contributes to the family income. This group has the closest attachment to the labor market with the most secure jobs (formal dependent work). This group should be able to cope with temporary external shocks through the availability of other social protection measures—for example, unemployment insurance. Nevertheless, if workers in this group do not have sufficient job tenure to qualify for unemployment insurance, they are in danger of falling into poverty, since labor market income represents almost 90 percent of their household income. Crises, such as the COVID-19 pandemic, could become a serious threat for this group. • Low-educated underpaid informal workers in BF. The second largest group consists of 25 percent of the working poor. The education level of this group, two-thirds of whom are men, is low: 84 percent of workers have an education level equal to or lower than primary education. Almost 90 percent are informal workers (self-employed and employees) that are underpaid. While being a rather old cluster (91 percent between 35 and 64 years), most live in urban areas outside the capital and pericapital or in rural areas and do not have children. This group of people consists of workers who are very vulnerable to income volatility. The informality of this groups excludes them from other social protection programs. The age and education profile make this group particularly difficult to serve with mainstream training or labor programs; dedicated pedagogical approaches would be needed. Also, enabling access to financial instruments could help this group to increase stability in case of external shocks. 48 • Educated informal female workers in BF. The third cluster (23 percent of all working poor) comprises 77 percent women. Most of them have children (70 percent) and are rather well educated. The majority live in families that receive BF, and they are mostly working in informal sectors. Two-thirds of this group are between 18 and 34 years and thus a rather young cluster. Almost 60 percent of them work below 35 hours a week. This cluster should be targeted with an increase in the supply of caretaking facilities in combination with formalization measures. In particular, the low work intensity is an indicator that these women do not work due to caretaking responsibilities. • Low-educated informal male in BF. The fourth cluster (15 percent) consists of middle-aged men, with low education levels. Almost 100 percent of this cluster is married, and 90 percent has children. Again, most of them are working informally; however, the majority is not underpaid or underemployed. In addition, a large share of this group lives in rural or semi-urban areas, which indicate fewer opportunities. For this cluster, it may be recommendable to pair formalization campaigns with basic educational modules on entrepreneurship, financial literacy modules, and support to commute to different localities with higher employability chances. • Unpaid low-educated informal dependent female workers. Finally, the last cluster, which is a rather small cluster (5 percent of working individuals) is mainly female. These workers are informal dependent workers, who are unpaid and live in rural areas. The education level of these workers is low, and the majority have children, are married, and receive BF. This is probably the group with the most precarious labor market conditions. Being a relatively old cluster compared to the previous ones, a functioning social assistance system is crucial. Providing social pensions will be necessary due to the lack of paid employment history of this cluster. 2.2 Out-of-work population The same exercise was repeated for the out of work population. For this group, seven different clusters have been identified (Figure 46): • Inactive low-educated women. The first and biggest cluster is dominated by female individuals (87 percent) in the age group 35–54. Most of these are out of the labor market and did not work within the past year. They are low educated, do not receive BF and are married, and therefore rely most likely on their spouses. This group has a large distance to the labor market and is dependent on other family members and pensions. They have limited opportunities for activation. • Better-educated young jobseekers. The second group (19 percent of the out of work) are young male and female long-term unemployed. They are rather well educated and live in urban areas. This group, which does not have care responsibilities and is not in BF, consists most likely of first- time jobseekers. Targeted active labor market programs or wage subsidies—such as the contrato verde amarelo—could help this group gain employment (Morgandi et al. 2020). • Inactive and discouraged female caretakers. The third cluster of the out-of-work (18 percent) consist mainly of young women below 35 living in a household with children or disabled (50 percent of individuals in this cluster have a child between 0 and 5 years old) and most likely have care responsibilities. Most are in BF. In addition, 96 percent live in households without older people—who commonly support women in the caretaking responsibilities—and rely on a second earner for most of their income. About 40 percent of them are looking for work or wish to work but are discouraged. They live mostly in small urban or rural areas. In addition to childcare, this 49 group would benefit from policies to connect with local employment and income-generating programs that are compatible with other responsibilities. • Short-term urban unemployed. The fourth group consists of 13 percent of the out-of-work and are mainly short-term unemployed. Those women and men of different ages (between 18 and 54 years old) have worked within the last year and became recently unemployed. Most of them are rather well educated and live in urbanized areas, with employment rates between 40 and 60 percent. In comparison to the in-work poor, this employment rates are lower and indicate labor market frictions for those individuals. Creating job opportunities to reintegrate those individuals—who are mainly not in BF—in the labor market is important. • Rural discouraged women in BF. The fifth group (12 percent) consists of low-educated discouraged women with care responsibilities in BF. About 63 percent of this cluster lives in rural areas with extremely low local employment rates. • Unmarried prime age inactive men. Cluster 6 represents 10 percent of the out-of-work work-able adults. It consists mainly of inactive, prime age males who do not receive BF. Their education level is low and 25 percent of them receive BPC. Identifying the labor market possibilities for these men is crucial to offer targeted trainings to integrate this group in the labor market. • Short-term, low-educated unemployed. Finally, the analysis identifies a group representing 7 percent of the out-of-work, which consists of short-term unemployed men. A key difference from cluster 4 is the low education level of this group. This group of middle-aged men (35–54 years old), of which the majority is not married, lives mainly in rural or semi-urban regions with rather low employment rates. Upskilling these individuals to provide them access to employment, which is long-lasting is important to provide stability for this group. Only 50 percent of them receive BF, that is, 50 percent of these individuals will be very vulnerable to future shocks. Figure 46: Out-of-work poor individuals by clusters Short term, low- Inactive low educated educated unemployed, women , 20.8% 7.3% Unmarried prime age Better educated young inactive men , 10.1% jobseekers, 18.9% Rural discouraged women in BF, 11.6% Short term urban Educated mothers, unemployed , 13.2% 18.0% Inactive low educated women Better educated young jobseekers Inactive and discourage women Short term urban unemployed Rural discouraged women in BF Unmarried prime age inactive men Short term, low-educated unemployed Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education. While all these groups are out of work, they are at different distances from the labor market, as shown by an employment chance index. The index uses the characteristics in each class to predict employment 50 probabilities, based on similarity with poor individuals who are already in employment.15 Figure 47 show the overall employment chances of the cluster (dots), and the contribution of each sociodemographic characteristic to the probability of being employed (bar). Variables that increase the chances of being employed (such as education levels above ‘none’) are above the zero line, and factors that reduce the probability of being employed (such as receiving social assistance) are below the zero line. The lowest employment chances are in clusters 1 and 3—inactive low-educated women and inactive and discouraged women, respectively. Closest to the labor market is cluster 7 (short-term, low-educated unemployed). Positive factors for employability in this cluster are the high share of males and age group of 35–54 years (Figure 47). Figure 48 repeats the exercise but estimates how close the individuals in each of the clusters are to being in formal wage employment. The chances of being formally wage employed are significantly lower throughout clusters. Receiving BF decreases the chances of being formally employed to a larger extent than to be employed in general. Again, cluster 1 is further away from the labor market while cluster 7 has the highest employability chances (Figure 48). Figure 47: Decomposition of employability barriers according to clusters’ relative shares of LCA variables, out-of- work poor Inactive low Educated mothers in Rural discouraged Short term, low-… educated women BF women in BF Weighted factors of employment 100% Short term urban Unmarried prime age 1.5 Better educated Average employment probability unemployed inactive men young jobseekers chance (bars) 70% 0.5 (dot) 40% 10% -0.5 Clusters Male Bolsa Familia BPC 25-34 35-54 55-64 Complete primary or incompleted secondary Complete Upper Secondary or Tertiary Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education. The number of clusters refer to the clusters introduced above. The dots represent the expected probability of each cluster of beneficiaries of out- of-work poor to be in employment, according to their characteristics. Bars represent the decomposition of the effect of different factors on the probability of being into employment. Baseline values: gender (female), education (primary), age (18–34 years). 15The employment index was created by estimating a probit model on the total work-able population (inside and outside of Cadastro Único). See Annex 3 for the outcomes of the probit model. The relative contribution of sociodemographic characteristics measured in the Latin America and Caribbean (LAC) region on the change of being (formally) employed were estimated. Covariates included in the model were (a) sex, (b) receiving BF, (c) receiving BPC, (d) age groups, and (e) education levels. To construct the weighted factors of employment chances, the saved probit coefficient of the respective independent variable was multiplied with the respective share of that variable in the given cluster. The overall employability index was estimated by predicting the employment chance of every individual based on the outcomes of the probit model and taking the average employability chance by cluster. Therefore, the index not only takes into account the impact of different variables on employment chances but also its frequency in a given cluster. 51 Figure 48: Decomposition of employability barriers for a formal job, according to clusters’ relative shares of LCA variables, out-of-work poor population Short term, low- Better educated Educated Mothers in educated young jobseekers, BF, 22% 40% unemployed, 34% 1.5 19% Formal employment probability (dot) Weighted factors of employment chance (bars) 35% 1 30% 0.5 25% 20% 0 15% -0.5 10% -1 5% 0% -1.5 Clusters Unmarried prime age Inactive low Rural discouraged inactive men , 33% Short term urban women in BF, 28% educated women , unemployed , 23% 10% Male Bolsa Familia BPC 25-34 35-54 55-64 Complete primary or incompleted secondary Complete Upper Secondary or Tertiary Married Individual receives BPC Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education. The number of clusters refer to the clusters introduced above. The dots represent the expected probability of each cluster of beneficiaries of out- of-work poor to be in employment, according to their characteristics. Bars represent the decomposition of the effect of different factors on the probability of being into employment. Baseline values: gender (female), education (primary), age (18–34 years). 2.3: Priorities for the largest classes identified in the LCA Some groups are easier to include in the labor markets, and the more complex will require greater resources and multiple interventions. The figure below (Figure 49: Labor market interventions according to beneficiaries’ employment barriers and participation constraintsFigure 49) illustrates interventions based on beneficiaries’ level of distance from the labor market. The distance is decided by employability barriers, such as skills and experience, or participation constraints, such as distance from jobs and markets, care and empowerment. While a worker with low skills and high participation constraints requires very complex interventions at the family and individual level, a worker with higher skills and closeness to jobs and markets only needs training and skills development. 52 Figure 49: Labor market interventions according to beneficiaries’ employment barriers and participation constraints Formalization measures and more care-taking facilities can support the three largest classes in the in- work poor population. The three largest groups of the poor in-work population in Brazil consist of educated male full-time employees, low-educated under-paid informal workers and educated informal female workers (Table 12). These groups require very different interventions spanning from day-care services to the informal female workers, short-professional training to the uneducated informal workers as well as simple skills development to the male educated full-time workers. Table 12: The three largest classes of the in-work population and potential labor market interventions Name of Characteristics Category Policy implications cluster • Between 35 and 54 • Able to cope with years old income shocks Educated • Secondary • If not sufficient job male full- education or above • Ready for the tenure they risk falling time • Full-time dependent job market into poverty employees work contract • Training to increase • Only earner in the skills family Low- • 2/3 are men • Vulnerable to income • Ready for educated • Low education fluctuations as they formal underpaid (Fundamental or are excluded from technical informal less) social protection training or workers in • Informal and • Could be targeted to first job BF underpaid formalize through MEI 53 • Live in urban areas experience/di outside capital/rural fficult to train areas • Most without children in family • The low work intensity • Female and well in this group can be a educated Educated result of care-taking • Have children and informal duties receive BF • Restricted female • Can be targeted with • Engaged in informal talent workers in an increase of care- work with low work BF taking facilities and intensity formalization • 18-34 years old measures Active labor market policies and an improvement of daycare can support the three largest classes in the out-of-work poor population. The three largest classes of the out-of-work population are inactive low-educated women, better educated young jobseekers as well as inactive and discouraged women (Table 13). The first and the last class of women need a combination of day care services as well as training and individually tailored interventions. The better educated young job seekers can be targeted with labor market programs or wage subsidies (for instance policies like the CVA subsidy). Table 13: The three largest classes of the out-of-work population and potential labor market interventions Name of Characteristics Category Policy implications cluster • Women between 35 and 54 • A functioning social Inactive assistance net is • Did not work within • Difficult to low- important for this the past year activate and educated group since they are • Do not receive BF reach women dependent on their • Low educated other family member • Married • Young male or • Most likely first-time • Ready for female long-term job seekers Better formal unemployed • Targeted active labor educated technical • Well educated and market programs or young training or live in urban areas wage-subsidies (e.g. jobseekers first job • No children nor in CVA) could support experience BF this group 54 • Middle-aged women • Out of work because • In households with of their care-taking Inactive children or disabled • Restricted duties and • Live without older talent/difficult • An improvement of discouraged individuals in to activate the daycare system women household (who are and reach can help this cluster likely to share care- out on the labor taking duties) market 55 Chapter 3: Implications for Brazil’s economic inclusion agenda Key messages for chapter 3 This last chapter discusses the implication of the analyses of the note for an economic inclusion agenda. These are arranged in three pillars: • Services beyond training are especially important to reach the population with high participation constraints and the rural population. Formal short trainings exist in Brazil and is also offered to the poor, though less intensively. Studies have found positive returns on employment of these services. However, high participation constraints, discrimination and lack of relevant work experience call for an expansion of toolkit of interventions for urban labor markets, including career counselling, childcare, wage subsidies for first time jobseekers, intermedaition and economic support to face the costs of job search. • Institutional arrangements around economic inclusion policies are vital for the success of interventions. Instruments, protocols and institutions for the coordination and governance of individual interventions remain underdeveloped in Brazil. The central government has a key step to take in the development of delivery systems, while the selection and implementation of programs should be coordinated at the territorial level. • Better use of data can facilitate the planning of economic inclusion initiatives. The note suggests improving the interoperability of Cadastro Unico with other sources, and reformulation of part of the questionnaire, to draw a better labor and skills profile of the poor. 3.1. Employability services should go beyond formal technical training Provision of individual programs A number of institutions already exist in Brazil that may provide some of the services to strengthen employability and opportunities, but funding has been increasingly driven by budget rigidities and not by strategic priorities. Services in place, at different scale, include (a) a large network of federally funded technical schools and adult education programs, (b) public labor intermediation offices, (c) employer- funded vocational training centers, (d) state-level agricultural extension programs, (e) financial institutions with dedicated microcredit programs. Detail on the specific courses and target groups reached by these institutions are included in two former flagship reports (e.g. Silva, Almeida and Strokova, 2015, and Almeida and Packard, 2017). Although these pieces are dated, there has been little evolution institutionally since the 2015 crisis. This chapter will therefore give a quick overview of ALMPs followed by the key issues of delivery systems, coordination and use of data, which were not covered by this literature. After the 2014 crisis, budget cuts defunded or closed most of the specialized interventions targeting the urban poor and the low-skilled unemployed. Most of these initiatives emerged and died out with the BSM strategy. Today, active labor market programs (ALMPs) are a very small share of total spending, and 56 almost none of them focus on the poor (Figure 50 and Table 14). Some states decided to continue funding these programs autonomously. Budget to labor intermediation also fell year on year. Recent attempts to introduce innovative programs that targeted the low income and the informal did not pass congressional approval. What remains largely in place at the federal level are programs shielded from discretionary budget decisions: federal vocational education institutions (propped up by the minimum spending rules on education), Sistema-S (funded by employers’ contributions), and benefits for formal workers such as SD (constitutional benefits). Figure 50: Active and passive labor market expenditure financed by federal government and employers Total PLMPs 180.21 Total Total ALMPs 38.01 Sistema S 15.94 expenditure 2018 Training (e.g. Bolsa Qualificação, Pronatec) expenditure 2020 0.01 Services Entrepreneurship support, startup incentives,… 0.28 Labor Intermediation, Labor Registries, services… 0.15 Other ALMPs (Economia Solidária, Agricultura Familiar) 0.05 FGTS - Dismissal Withdrawal 106.69 Seguro Desemprego (Contrib and Noncontrib) 39.99 BEm 33.50 Cash Abono Salarial 19.58 FGTS Complemento 0.04 Salario Familia* 2.00 - 50 100 150 200 Billion BRL Source: Based on budget data from Portal Transparencia data for 2018 and 2020. Note: *Out of budget expenditures financed by mandatory employer contributions. Passive labor market program (PLMP) - Salario Familia, Abono Salarial, SD, Fundo de Garantia por Tempo de Serviço (FGTS), Benefício Emergencial de Preservação do Emprego (BEm). 57 Table 14: Details of active labor market policies in Federal Budget (2020, BRL, billions) • Apoio à Organização Econômica e Promoção da Cidadania de Mulheres Rurais • Fomento ao Desenvolvimento de Instituições de Microcrédito • Fomento e Fortalecimento de Empreendimento Econômicos Solidários e suas Redes de Cooperação • Apoio à Agricultura Urbana • Plano Brasil sem Miséria • Fomento para a Organização e o Desenvolvimento de Cooperativas Atuantes com Entrepreneurship support Resíduos Sólidos /startup incentives (cash and • Aquisição de Alimentos da Agricultura Familiar in-kind grant, microcredit) • Fomento ao Desenvolvimento Local para Comunidades Remanescentes de Quilombos e Outras Comunidades Tradicionais {0.28 billion} • Fomento à Produção e à Estruturação Produtiva dos Povos Indígenas, Povos e Comunidades Tradicionais e Agricultores Familiares • Assistência Técnica e Extensão Rural para Reforma Agrária • Assistência Técnica e Extensão Rural para Agricultura Familiar • Fomento, Capacitação Ocupacional, Intermediação e Assistência Técnica a Empreendimentos Populares e Solidários e a Trabalhadores • Fomento e Fortalecimento da Economia Solidaria Labor intermediation and • Cadastros Públicos na Área de Trabalho e Emprego registries • Sistema de Integração das Ações de Qualificação Profissional com a Intermediação do Emprego e Seguro-Desemprego – SIGAE {0.15 billion} Other Active Labor Market Programs • Fortalecimento da Institucionalização da Política Nacional de Economia Solidária • Promoção e Fortalecimento da Agricultura Familiar {0.05 billion} • Concessão de Auxílio-Financeiro Projovem Training (vocational, life skills, • Qualificação Social e Profissional de Trabalhadores cash for training) • Accessuas Trabalho Ações Complementares de Proteção Social Básica • Elevação da Escolaridade e Qualificação Profissional - ProJovem Urbano e Campo {0.01 billion} • Apoio à Formação Profissional, Científica e Tecnológica Note: Sistema-S (Out of budget) Budget data from Portal Transparencia, Government of Brazil, for 2018 and 2020. Among the existing programs, training is the most widely available to vulnerable workers. Training is largely provided through the federal network of vocational institutes depending on the education ministry, through Sistema-S centers, or by the private sector. According to PNADC, in 2019 about 1.7 percent of Brazilians between ages 18 and 65, and not full-time students in other courses, took part in FST equivalent to more than 1.8 million adults. The share of individuals in Cadastro Único participating in FST courses is below average—0.9 percent of BF beneficiaries and 1.3 percent of others in Cadastro Único. Importantly, about one-third of the whole population attendees had not completed secondary education. Both Sistema-S and public federal institutions are more likely to offer short training programs attended by the medium- and low-educated, compared to the private sector Nonetheless, such training programs tend to cater to the less-vulnerable segment of low-income workers. The average student of formal short training courses in 2019 was 34 years old, 85 percent were already employed and about 56 percent were women (see Box 4). Past evaluations showed that formal short training courses in Brazil, including those targeted to low- income youth, can be effective, if appropriately designed. Almeida et al. (2014) found positive returns of participants in FST compared to nonparticipants in Brazil, but only for those that attended private or Sistema-S institutions (as against federal public technical schools) and for those completing short-term training courses (2.2 percent on average). However, there is large heterogeneity in results. Those who 58 received short-term training provided by public providers alone have returns close to zero. Similarly, evaluation of the Programa Nacional de Acesso ao Ensino Técnico e Emprego (PRONATEC) program, one of the flagship interventions of Brazil’s antipoverty strategy up to 2016, showed that design matters for outcomes. When designed with clear, demand-driven features, such as offering training only in skills that were expressly requested by employers in the territory, short training had positive impacts on employment rates of treated individuals compared to control groups, including for participants benefitting from the BF program (SAGI 2018). Da Mata et al (2021) studied the long-term effect of a vocational training program targeting disadvantaged groups in Brazil. The program offered 18 months of classroom training coupled with 6 months of on-the-job training. When assessing the impact of the program the authors found that female student fared better than male students. Furthermore, the effects were stronger after a few years, six years after the program the author still find a significant and positive effect on employment. Box 4: Use and Supply of Short Formal Training (FST) in Brazil’s household survey The participation in FST of BF beneficiaries and other poor in Cadastro Único is below average. According to PNADC, in 2019 about 1.7 percent of Brazilians between ages 18 and 65 and not full-time students in other courses took part in FST (curso FIC). Participation rates fell to 0.9 percent for BF beneficiaries and 1.3 percent for other poor in Cadastro Único (Figure 51). Among those attending such courses, the majority are unemployed, formal wage workers (in Cadastro Único), or informal workers (in BF) (Figure 52). Figure 51: % of adults attending FST courses Figure 52: Employment status of attendants of FST courses registered in Cadastro Único 2.7% CLT Workers Pop. average = 1.7 % Unemployed Public workers + military 1.3% Formal Self-employed 0.9% Informal Self-employed Informal wage employed Others BF Cadastro Único Seguro 0.0% 5.0% 10.0% 15.0% 20.0% 25.0% 30.0% 35.0% no BF Desemprego Cadastro Unico no BF BF Source: Authors and World Bank 2020, based on PNADC 2019 On average, Sistema-S has the best coverage for low and middle educated. Only 17 percent of the students in Sistema- S have completed tertiary education, while the share in private institutions is about 25 percent (Figure 54). Similarly, the average income of students in Sistema-S is below BRL 2,000, while students in public institutions earn on average BRL 2,733, and those in private institutions around BRL 3,090. Occupation in the labor force also varies significantly for students attending each type of institution: while 20 percent of students in public institutions are employed in manual labor occupations such as manufacturing, construction, and vehicle repairs, the share of students in Sistema-S is double that, at 40 percent. Overall, the data suggest that Sistema-S plays an important role in forming skills for the middle segment of the labor market, thus performing an important public function for both existing and prospective workers. Both enrollment and completion of FST increase with income and work status and depends on providers. The cost of receiving training for adults includes both opportunity costs and tuition. While FST is easier to reconcile with career and adult responsibilities, annual enrollment rates increase monotonically with household income (from 0.9 percent for 59 adults in the bottom decile to 2.9 percent for those at the top). Analysis of service providers shows that most students (56.3 percent) are enrolled in private institutions, presumably self-paying for the training. Only 10 percent receive training through their employer (which is usually free to the worker). Additionally, 17.2 percent report receiving their FST in Sistema-S (which offers both free and fee-based programs), and a similar low share (16.2 percent) received the program in public federal institutions (Figure 55). FST participants also on average have higher earnings. The probability of completing training in Sistema-S is higher than in courses offered by other providers. Finally, regression analysis indicates that, holding all the other characteristics constant, the unemployed and informal salary workers are those with the highest risk of noncompletion. Figure 53: Enrollment in FST, by education Figure 54: Distribution of students by education and institutions that provide the service Figure 55: Main provider of FST courses for vulnerable workers Source: Authors and World Bank 2020, based on PNADC 2019. A complementary policy to improve the foundations for further learning are adult literacy programs. One of the reasons why vocational training programs often fail to work for the poor is that candidates lack the foundational skills to absorb technical learning, either due to poor quality schooling or early interruption. In part, the solution lies in adapting content and delivery modalities; however, foundations in literacy and numeracy remain essential and should be learned before engaging in more advanced training. For this reason, adult literacy programs and second-chance education programs remain strategic to enable a high share of uneducated workers and jobseekers to learn more technical skills (Box 5). 60 Box 5: Adult Literacy Programs: good practices More than 1 percent of today’s young adults (the prime age workers in 2042) declared in 2019 to be illiterate and another 21 percent did not complete primary education: based on test score results, a portion of these little schooled adults is likely functionally illiterate. A recent publication takes stock of ALPs, a little studied but fundamental public policy to enable more complex learning, including digital literacy. Most evaluated programs did not achieve improvement in literacy levels. The main reason for the poor performance of these programs is that teaching methods are incompatible with students' needs and baseline ability levels. For instance, these programs end up improving word recognition but fail to develop actual reading comprehension. An exception that could serve as inspiration for Brazil is Neuroalfa in Mexico. This ALP had a total of 126 hours spread over 12 weeks of training, targeted urban poor population, and was successful in making illiterate students achieve reading comprehension. Some lessons learned from this and other successful programs are the importance of designing the course to set reasonable expectations, training teachers on how adults learn without spending too much focus on the teacher profile, promoting student and teacher motivation, offering flexibility, using information and communication technology (ICT) tools as complements to promote feedback and adaptive learning, and if possible, grouping participants by baseline ability rather than having a mixed-ability target group. The right mix of design and diagnosis is necessary for ALPs to succeed. However, the current knowledge obtained from evaluations is insufficient to conclusively outline ‘what works’ for ALP (World Bank 2019). Going forward, labor market policies should make use a broader toolkit of services than the formal in- class training . The menu of labor market programs for an urban context, or to prepare adults for migration from rural to urban areas could include a combination of the following, chosen according to territorial priorities and individual barriers (Table 15): (a) Job readiness and socioemotional skills training for the low-skilled and long-term unemployed approaching the wage labor market. Such programs teach behaviors and workplace skills demanded by employers, which are rarely taught in formal education and usually depend more on family socioeconomic status and thus penalize less-established labor market entrants. (b) Appropriately funded labor intermediation services, including tools to assess candidates cognitive competences and socioemotional skills to improve chances of appropriate matches through better signals and reduce employers’ tendency to discriminate based on gender and race. (c) Targeted wage subsidies for youth in search of their first job and without sufficient education (see Morgandi et al, 2020b; this is also helpful to reveal abilities and prevent long-term youth unemployment. (d) Referrals and vacancies in childcare centers, reserved for women ready to engage in job search, training, or work. The recent AB reform contains dedicated subsidies for this target group that will require careful implementation. (e) Transport subsidies as part of a job-search assistance, for the isolated unemployed in rural areas, semi-urban settlements, and slums. (f) Comprehensive programs for the self-employed (very context specific, typically a combination of skills development, financial inclusion, grants, or subsidized credit). These programs can be best delivered in the communities where the self-employed operate, with compatible hours and pace. International evidence, particularly metanalysis of impact evaluations of active labor interventions for vulnerable groups, show that different programs have different short and long-term impacts. Specific evidence from studies in LAC countries (Escudero et al, 2019) shows that ALMPs affect in particular the probability of beneficiaries of being employed and the chances of being in the formal sector, compared 61 to non-beneficiaries with similar characteristics. However, even successful programs tend to have modest magnitude in terms of impact, increasing chances of employment by 5 to 10 percentage points compared to control groups. While job search assistance programs show a larger impact in the short-term, “human capital” training programs and private sector subsidies yield larger impact in the medium- and long-run run ( ). The literature also finds that public employment programs (public works) have negligible impacts in terms of employment, so these serve more as a temporary safety net than as a springboard for sustainable employment. Table 15: Type of labor market interventions according to the type of beneficiaries’ constraints Participation Employability Barriers Information asymmetries Lack of labor demand constraints Childcare subsidy Vocational training Job-search assistance and Public works training Transportation subsidy On-the-job training Targeted wage subsidies for (apprenticeship, Job matching employers Individual and family internship) coaching Skills certification Financial and technical Remedial literacy and assistance to self-employed Monetary incentives in numeracy programs Career guidance social benefits Local economic Socioemotional skills development programs training Public procurement programs Comprehensive programs (multiple, sequenced or overlapping, interventions) Source: Authors adaptation from Santos and Rigolini (2019) and World Bank (2023 forthcoming) Furthermore, the impact of ALMPs differs according to the target group: in general, the more disadvantage see greater results. Long-term unemployed often tend to benefit more from ALMPs than other segments of the unemployed, for instance (Card, 2018). Similarly youth-targeted ALMPs show slightly larger impacts on female and disadvantaged youth (ILO & WB, 2023). Vocational skills training specifically has larger impacts of female youth than male (Kemper, Stöterau, Ghisletta 2023). Moreover, training is more impactful when it is offered more intensively, and when targeted to the poor and vulnerable specifically. These services, whose implementation requires adaptation, can take advantage of innovative contractual processes that encourage innovation and focus on outcomes. Performance-based contracts can reward at least in part for results (rather than financing for inputs). In this way, they can be very effective in harnessing the more capable providers and stimulating innovation. Aedo et al. (2020) review the encouraging evidence in developing countries. These processes can also be replicated in Brazil’s public procurement system, but strong guidance from the central government should support managers in local administrations, given the limited precedents. Reforms in passive labor market policies can produce the fiscal space to finance such interventions from the federal level. With total federal and employers’ spending on labor programs at 2.2 percent of the gross domestic product (GDP) in 2020 (above OECD averages), implementing economic inclusion in Brazil is more constrained by legacy allocative decisions and limited institutional capacity than by a resource constraint (see Morgandi et al. 2020). Improving the quality of active labor market policies would 62 contribute to building trust in the safety nets by workers and enable long-overdue reforms of contributory labor benefits, which drain most resources but do not reach the poor and vulnerable. 3.2 From single programs to economic inclusion strategies Well-coordinated economic inclusion strategies have the potential to address the multiplicity of constraints that inhibit income generation. Several countries in the LAC Region, including Brazil, introduced Economic Inclusion strategies, which can be defined as a conjunction of public policies aiming to improve the income-generating capacity and assets building of the poor (Murrugarra and Isik-Dikmelik 2020). Such approaches have mostly targeted the poor in rural or peri-urban areas, and thus these policies are complementary to ALMPs discussed above. EI strategies by design try to address both the productive and the social constraints for the poor and vulnerable. They recognize the gains from jointly addressing these constraints through bundled interventions, which typically include cash or in-kind transfers, skills training or coaching, access to small scale finance, and links to markets (Figure 57). Figure 56: Social and productive constraints and the components of integrated services Source: Mugarra and Isik-Dikmelik 2020. Global evidence shows that bundles of economic inclusion interventions connected with targeted cash transfers have the ability to increase households’ income, assets, and savings, and resilience to shocks. Varghese Paul, Dutta, and Chaudhary (2021) analyze global evidence on the impact and costs of EI programs in 80 different countries. The analysis shows promising short-term results on a wide range of impacts including income, assets, and savings. Interactions between components are likely to drive the program impact as bundles of interventions demonstrate greater impact relative to stand-alone interventions. Further, most programs increase households’ resilience to shocks by diversifying sources of income. Many of the programs are also empowering women by enhancing their economic opportunities and social status. Lastly, the analysis also identifies the business capital or consumption support components as the main drivers of overall cost of interventions (Varghese Paul, Dutta, and Chaudhary 2021). Table 16 below summarizes the different impacts found. The World Bank State of Economic Inclusion Report 2021 finds similar promising impacts on a wide range of outcomes in the short- to medium-term, up to 4 years after enrolling in a program (Andrews et al, 2021). Most programs have heterogeneous impacts with the poorest and most vulnerable experiencing the fewest gains. 63 Table 16: Summary of outcomes identified in EI programs around the world Theme Impact Enhancing Households experience increased income thanks to investment in productive incomes and assets and diversification in employment assets Households are able to increase savings and reduce debt. This impact is more common for programs that offer savings groups. Households diversify incomes as some members shift to other forms of employment Households are able to increase and diversify their asset holding (e.g. livestock). However, estimates of the size vary significantly. Increased Households can protect themselves against destitution resilience and Households increase investments in human capital welfare Women are empowered by improved mobility, social status, household decision making and psychological well-being. Households build resilience to shocks by receiving regular, predictable transfers, facilitating asset accumulation, income diversification, access to low-cost credit. This is especially observed in shock-prone areas. Source: based on Andrews, 2021 Lessons can be drawn for Brazil from LAC’s EI programs. A World Bank synthesis note aiming to document different approaches of EI in LAC countries can provide context-relevant knowledge for the implementation of EI in Brazil. Drawing from five case studies (including the BSM in Brazil), the report offers six key lessons learned: (a) Simplifying the initial design of PEI programs can allow easy and quick implementation. (b) To avoid capacity constraints and other challenges, remaining flexible and constantly adapting the program is important. (c) Due to a heterogenous target population, results will often vary, making it complicated to draw clear conclusions. (d) Social assistance and PEI programs are meant to be complements not substitutes. Therefore, the graduation narrative should be avoided. (e) Establishing output- and delivery-based agreements with partners will ensure better coordination and results. (f) PEI programs show better impact if nested with territorial development plans (Murrugarra and Isik-Dikmelik 2020). The BSM Plan is an eminent example of how better intergovernmental and intersectoral coordination and the bundling of different social policies in Brazil have the potential to increase productive inclusion in the country. The BSM was launched in 2011 as part of Brazil’s productive inclusion strategy. The program consisted of three main axes of action: an income guarantee axis, a productive inclusion axis, and an access to public services axis, which all have the goal of targeting extreme poor families with insufficient production and food security. The BSM relied on a whole-of-government approach coordinating existing policies and programs (for example, BF, Programa Aquisicao de Alimentos) and new instruments (for example, Bolsa Verde, Fomento as Atividades Produtivas Rurais) and packaging them into a PEI bundle. Municipalities acted the key stakeholders in the management of BSM due to their better awareness of local citizens’ circumstances. Cadastro Único and SUAS were both important networks to facilitate the massive coordination efforts of the BSM and to provide a gateway to access all the bundled policies. On the urban labor policy front, the BSM strategy included active campaigns to promote the formalization of self-employed though the MEI program, and, for the first time, provided targeted grants to professional training institutions to include low-income individuals in technical courses). (MDS, 2015) 64 Local economic development policies that can attract investment, or infrastructure and services that connect communities with markets. Brazil has pioneered some of these approaches in the past, especially through the creation of sheltered public markets for small agricultural producers in Cadastro Único, or by developing productive alliances between small farmers and commercial producers. Particularly important are processes that allow for aggregation and scale, for instance via rural cooperatives that allow for the acquisition of capital goods or services. When serving the poor, such processes need to blend together social and technical components. 3.3. Critical institutional decisions underpin economic inclusion in Brazil The institutional setup for the delivery of economic inclusion programs is always country specific. In the delivery of labor market support, most high-income countries put in place cooperation models between labor and social assistance services (or outright integration between the two).16 While these approaches appear institutionally demanding for Brazil, it is important to distill critical functions that should be in place, even with simplified delivery modalities. A major agenda for the future is in the development of instruments, protocols, and institutions for the coordination and governance of individual interventions and potential ‘entry points’, such as CRAS and SINE. A key lesson learned from Brazil’s past policies, including Brasil Sem Miseria, is that serving the vulnerable groups described in this note requires significant adaptation of the supply of service s, and targeting at the right level. To achieve this, it is important that individual services are embedded in an economic inclusion coordination and governance system. The central government is best placed to provide legal and operational mechanisms, data to facilitate planning at the local level, capacity building and information systems to monitor progress. A recent review of economic inclusion strategies shows that the central government has a critical role to play in establishing the ‘rules and tools of the game’: regulations and financial incentives, a ‘menu’ of interventions that can be customized locally, monitoring systems to verify what is happening on the ground and what works, operational systems to manage client intake, draw relevant labor profiles to make treatment decisions. Some of these components can build on existing information systems, such as Cadastro Único. Central governments should also focus on building the capacity of local implementers. On the other hand, most components of the delivery chain of economic inclusion policies should be coordinated territorially. Economic inclusion services are more complex than cash transfers—they cannot be only coordinated at the federal level or implemented only digitally. A major challenge is to identify the institution, or institutions, that has the comparative advantage to deliver the core functions at the local level. Strategic planning and oversight functions can be performed by a single entity over a geographic area, but points of entry are also needed at the client level. These include the following: (a) Managing client treatment cycle, including registration, profiling, referral to services, and follow- up to verify completion. Part of this cycle can be associated with existing processes of client registration for cash transfers, but often requires specialized skills; (b) Identifying appropriate interventions, based on the territorial vocation and the characteristics of the target groups, to include in the menu of offered services; (c) Contracting specific service providers or forging partnerships; 16In the delivery of labor market support, high-income countries put in place cooperation models between labor and social assistance services which, for instance, require social benefit recipients for simultaneous registration in social and labor offices. These models also require coordinated case management. 65 (d) Monitoring implementation of programs, ensuring information on participants’ completion and outcomes are collected and shared at the central level. Figure 57: Typical division of responsibilities across institutional levels Service Federal Territorial Point of Entry providers Government Coordinator (multiple) Implementing Study labor and Take charge of regulations economic Client registration clients after referral opportunities Registry of Partner or contract Provide service beneficiaries Profiling, skills service and include assessment Define menu of in 'delivery chain' Follow up on results interventions Define coordination of intervention protocols between Referral to services Capacity building actors Report outcomes and communication Monitor quality of Follow-up and services and manage ensure data after Coordinate with Monitoring systems point of service service is collected Point of Entry Accessuas Trabalho was Brazil’s institutional solution to the limited reach of labor intermediation systems and the historic segmentation of formal-informal labor markets. In some countries, territorial coordination and point-of-service functions are performed by local labor offices, but this need not necessarily be the case in Brazil. Given the limited capacity and financing of SINE offices, in the past Brazil opted to equip municipalities of dedicated labor units (Accessuas Trabalho) to carry out territorial coordination, and the latter or CRAS also served as ‘points of entry’. This could well be a functional model to be revived, building on lessons learned from BSM’s experience, especially in ensuring that the federal level develops the tools, and the territorial offices perform quality assurance functions. Even in such setups, public employment services maintain a core comparative advantage in labor intermediation. It is premature to delegate to SINE the management of typical clients of CRAS offices. At the same time, initiatives to carry out public labor intermediation outside SINE would be a clear duplication. Instead, SINE could be seen as one of several employment services, involving several public and private actors. For instance, the literature shows that the implementation of training programs for vulnerable populations is more effective if packaged with other services, such as counselling, employers’ wage subsides, internships, soft skills training. Such complex delivery needs to be coordinated and cannot be carried out by service providers themselves. It will be also critical to consider the relationship between Cadastro Único and the registry of jobseekers used by the public employment services in Brazil. It is important to avoid duplicating data collection for the same users in different registries, and instead organize more interoperability protocols. Thus, improvements in Cadastro Único need not go as far in trying to mimic information collected by SINE offices, and rather focus on understanding the key labor traits of users for planning and eligibility screening purposes. 66 3.4 Better use of data and a renewal of Cadastro Único can support the territorial implementation of economic inclusion A major strength of Cadastro Único is its representativity at a municipal level. The administrative dataset allows to create a profile of beneficiaries at the very local level and thus advise local economic inclusion policies. PNADC on the other hand only allows disaggregation at a state level, hence it is not representative at a local level. On a country level, both surveys provide similarity in some variables and differences in others, and this can be used to identify large errors / omissions by households in the registration processes. For example, the distribution by gender differs: While in Cadastro Úncio, 66 percent of BF beneficiaries are female, the share of female BF beneficiaries in PNADC is 54 percent. The distribution by age, education, and urban/rural is similar, and this could be associated with an underreporting of male family members. Through the matching and comparison of different data sources, this note identifies scope for improvements of Cadastro Único. Before the pandemic, interoperability was prevalently used to ‘audit’ records of applicants to cash transfer programs. Auxilio Emergencial stepped up the use of admin sources to proactively identify potential targets of cash transfers. The potential of data to inform labor policies and support economic inclusion, however, remains largely unexplored. Changes to the individual-level module of Cadastro Único could facilitate economic inclusion: (a) strategic understanding at the national level of the labor market conditions of the population served by social safety nets; (b) better mapping of local conditions, and from there identification of appropriate measures and targeting criteria; (c) support to the development of individualized action planning in those municipalities where case management is eventually introduced. An important guiding principle is that reporting obligations by users should not increase through this process, and data should be validated/overridden by human operators at the point of service level, given the high probability of errors that were recorded in the operation of AE for the ExtraCad population (World Bank 2021). 67 Figure 58: Administrative data sources with potential to increase accuracy of Cadastro Único Current information available in Cadastro Único (i) Worked last week - (ii) Type of work (not between formal and informal self-employed) (iii) Self-reported education data (iv) Self-reported incomes Strenghten profile of individuals and families Active Labor Other social protection Labor market Education Market programs Policies Receita Federal MoE Formal SINE SD Declaration IRPF Education Registration Vocational Short formal RAIS + CAGED Pensions Training training Public pre- Disability CNPJ / MEI school and (BPC) childcare Contribtions (GFIP/ Esocial) Changes could gradually include: (a) Aligning the definitions of employment type, to improve accuracy and correspondence with PNADC. It seems particularly important to improve guidance to distinguish the self-employed from other informal wage workers, or to better identify rural farmers (even if not temporary). Consider adding a variable to identify the current sector or industry of employment, consistent with CNAI (even at one-digit level), for the informal. (b) Rely on administrative registries, such as CAGED, the national registry of firms, the E-Social registry of contributors to INSS, to identify formal labor relations (as private dependent worker, autonomous employee, or public employee), and firm ownership (as MEI, SIMPLES or other partnership in firms), while maintaining the possibility of overriding this information by operators if obsolete. (c) Improve information on educational attainment, drawing from the education database, and add information on completion of short cycle training programs, such as those provided by federal institutes or Sistema-S. All such actions would require changing data access regulations by MoE, and improving obligations of data reporting of private entities that benefit from public funding. (d) Draw from the conditionality system (SICON) information on current enrollment in education, which constrains availability to work. Over time, such information should be supplemented to cover child enrollment in preschool and childcare, which is critical to facilitate female labor supply and also to monitor access to services for young children. The extension of AB conditionalities to 68 children between 4 and 5 years, and the introduction of the new childcare subsidy (also in AB) requires such updates. (e) Display at the individual level information on receipt of social protection programs, such as SD, pensions, and disability benefits, which are all highly relevant to identify household level constraints and possible work disincentives. (f) Add information on the sector/industry of employment. Even a crude one-digit classification can differentiate territories with agricultural vocation from those largely in services. For formal employees, such information could derive from RAIS, for the informal this need not to be asked. (g) Add basic questions about recent job search. While Cadastro Único should not be used to identify statistical unemployment, this can give an idea of the degree of distance from the labor market of the out of work and potential interest in ALMPs. Finally, administrative data can also be used as operational instruments to plan economic inclusion policies at the territorial level. As labor market interventions are planned and implemented locally, this note provides an illustration of how administrative data could be processed at the state and micro- municipality level to generate labor market maps of utility for local labor policy planning. The richer the data in Cadastro Único, and the more the number of services included in the registry, the more such maps could be informative for planning and useful for evaluation. The quality of information will depend on the frequency of updates of Cadastro Único and the capacity of local operators to classify employment correctly. Box 6 shows two potential ‘employment maps’ at the municipal level in the state of Piaui based on administrative data. Box 6: Using the social registry to map local labor market conditions of the poor This box illustrates the potential of administrative data to generate labor market insights, applied to two small municipalities (one urban and one rural) in the northeastern state of Piaui. Such statistics could be systematically reproduced for all municipal areas of Brazil (or meso-regions) and support local authorities in selecting and customizing economic inclusion interventions. Out of Piaui’s 3.1 million inhabitants, around 61 percent are registered in Cadastro Único, nearly twice the national share of 36 percent, and 69 percent of them receive BF. Almost half of BF recipients in Piaui live in rural areas (compared to only 32 percent nationwide). The two municipalities reveal the heterogeneity of local labor market conditions compared to state-level averages. The absence of administrative data is a constraint. PNADC is the primary source of labor market information, but its representativity ends at the state level. According to PNADC, in 2019, 34 percent of the state’s working-age population living below the Cadastro Único poverty line worked informally, and 16 percent held a formal wage job, the remainder being out of work. We zoom in on José de Freitas (JDF), an urban municipality in the metropolitan area of the state capital, with 40,000 inhabitants, and on Sao Francisco de Assis (SFA), a small rural municipality with 5,801 inhabitants. Both are mostly poor: In JDF, 67 percent of inhabitants are registered in the social registry and almost half (47 percent) receive BF. In SFA, 84 percent are in Cadstro Único and 66 percent receive BF Some of the key takeaways from the employment indicators that could inform the design of local economic inclusion policies include the following: A minority of Cadastro Único individuals in both municipalities are of working age, and most are in BF. In urban JDF, 66 percent of families have no employed adult. Only 24 percent of the work-able in JDF worked at the time of cadastral update, 17 much lower than the 66 percent in SFA. In both municipalities small children are more prevalent in families with no working no members. JDF has very large employment gaps by gender (40 percent versus 18 percent), these are modest in SFA. In rural SFA, 76 percent of the employed are in unpaid work, those paid report earnings of BRL 37 per month. In urban JDF, 68 percent of workers are self- employed and 16 percent in formal work; earnings are BRL 135 per month on average; however, only 125 adults in Cadastro Único were in the MEI registry. In SFA, almost all family income is from BF; in JDF, labor income is half or more of total income, among families that have an employed member. 17 Individuals that reported that they worked last week or were away from work last week are both considered in this analysis. 69 Among formally employed in SFA, nearly all are in public administration, followed by commerce. In JDF, men work in industry, commerce, and construction, women in hospitality and food. Figure 59: Potentially work-able populations in two example municipalities of Piaui (thousands) Total 39 In Cadstro Único all ages In BF all ages Potentially work able BF beneficaries 26 Potentially work able inidividuals registered in Cadstro Único not receiving BF 19 9 6 5 4 4 2 1 José de Freitas São Francisco de Assis do Piauí Source: Cadastro Único 2019. Sao Francisco de Assis (SFA) - Rural Jose de Freitas (JDF) - Urban Figure 60: BF Family demographic characteristics and income sources in SFA and JDF municipalities Jobless Family 1 employed in family Jobless Family 1 employed in family 2+ employed in family 2+ employed in family Figure 61: BF Family composition in SFA and JDF municipalities % with small children <5 % with children 6-14 % with small children <5 % with children 6-14 % with children 15-18 % with elderly above 65 % with children 15-18 % with elderly above 65 45% 48% 44% 53% 48% 50% 32% 37% 43% 32% 27% 32%35%33% 19% 20% 26% 24% 0% 0% 0% Jobless Family 1 employed in 2+ employed in family family Jobless Family 1 employed in 2+ employed in family family 70 Figure 62: Income sources of families in BF in SFA and JDF municipalities Mean Income in CadU BRL 600 BRL 600 BRL 400 BRL 400 BRL 200 BRL 200 BRL 0 BRL 0 Jobless Family 1 employed in 2+ employed Jobless 1 employed 2+ employed family in family Family in family in family Mean income from other sources Mean income from other sources mean income from labor mean income from labor Mean income from BF Mean income from BF Mean income from pensions Mean income from pensions Figure 63: Employment rates and average earnings of BF families in SFA and JDF municipalities, by age, gender, and education 80% 180 80% 180 70% 160 70% 160 60% 140 140 60% Mean earnings Mean earnings 120 120 % working 50% 50% % workign 100 100 40% 40% 80 80 30% 30% 60 60 20% 40 20% 40 10% 20 10% 20 0% 0 0% 0 All Women 18-29 30-50 51-65 High S. / + Middle S. Men Below Sec Women 30-50 18-29 51-65 High S. / + Middle S. Men Below Sec by by Age by Education All By Gender By Age By EducationTotal Gender Figure 64: Employment type of BF families in SFA and JDF municipalities, by age, gender, education 71 100% 100% 90% 90% 80% 80% 70% 70% 60% 60% 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 0% 0% Formal employee Self Employed Formal employee Self Employed Rural worker Informal employee Rural worker Informal employee Unpaid worker Employer Unpaid worker Employer Table 17: Profile of the formal workers in SFA and JDF municipalities and in Cadastro Único Top Sectors of Formal Employment Top Sectors of Formal Employment Men Women Men Women Public Public Public administration, administration, Transformation administration, 26% 54.76% 27% 39% defense, and defense and social industries defense and social social security security security Commerce: Commerce: Commerce: Commerce: automotive 25% automotive 20.24% automotive 17% automotive 23% reparation reparation reparation reparation Transformation Administrative Construction 18% industries 7.14% activities and Construction 16% 10% complementary Administrative services activities and Public Transformation 16% complementary 5.95% administration, Accommodation industries 14% 8% services defense and social and food security Average tenure (months) Average tenure (months) Men Women Men Women 33 62 31 41 Participation in Formal Registries Participation in Formal Registries Men Women Total Men Women Total % adults in RAIS 6.81% 6.06% 6.43% % adults in RAIS 14 % 5% 9% % adults in MEI 0.04% % adults in MEI 0.91% 0.98% 0.95% Source: Cadastro Único 2019. Note: Groups are restricted to individuals at of working age (18-64) and not in education and able- bodied. 72 References Almeida 2014 Aedo Andrews, Colin; de Montesquiou, Aude; Arevalo Sanchez, Ines; Dutta, Puja Vasudeva; Paul, Boban Varghese; Samaranayake, Sadna; Heisey, Janet; Clay, Timothy; Chaudhary, Sarang. 2021. The State of Economic Inclusion Report 2021 : The Potential to Scale. Washington, DC: World Bank. Araujo, Angela Maria Carneiro, and Maria Rosa Lombardi. 2013. "Informal Work, Gender and Race in Brazil at the Beginning of the 21st century." Cad. Pesqui. 43 (149): 452–477. Biroli, Flavia. 2016. "Sexual Division of Labor and Democracy." Data 59 (3): 719–754. Bruschini, Cristina, and Maria Rosa Lombardi. 2000. "The Bipolarity of Women's Work in Contemporary Brazil." Cad. Research 110: 67–104. Card, David, Kluve, Jochen and Weber, Andrea. 2018. What Works? A Meta Analysis of Recent Active Labor Market Program Evaluations. Journal of the European Economic Association, Volume 16, Issue 3, p. 894–931 Collins, Linda. M., and Stepahnie. T. Lanza. 2010. “Latent Class and Latent Transition Analysis: With Applications in the Social, Behavioral, and Health Sciences. John Wiley and Sons Inc. https://doi.org/10.1002/9780470567333 Da Mata, Daniel., Oliveira, Rodrigo. and Gonzaga da Silva, Diana Lúcia. 2021. Who Benefits from Job Training Programs? Evidence from a High-Dosage Program in Brazil De Brauw, Alan, Gilligan, Daniel O., Hoddinott, John, and Roy, Shalini. 2015. "Bolsa Família and Household Labor Supply." Economic Development and Cultural Change 63 (3): 423–457. De Oliveira, Paulo Felipe Alencar, and José Raimundo de Araújo Carvalho Júnior. 2009. "Desigualdade de gênero na duração do desemprego e seus efeitos sobre os salários aceitos no Brasil." Revista Econômica do Nordeste 40 (4): 833–850. De Oliveira, Gabriel Lyrio, and Andre Luis Squarize Chagas. 2020. “Long-term Effects of Conditional Cash Transfers on Children: The Brazilian Case.” Ph. D Dissertation, University of Sao Paulo. Verónica Escudero, Jochen Kluve, Elva López Mourelo & Clemente Pignatti. 2019. Active Labour Market Programmes in Latin America and the Caribbean: Evidence from a Meta-Analysis, The Journal of Development Studies, 55:12, 2644-266Fietz, Katharina Maria, Tiago Falcao Silva, Maria Concepcion Steta Gandara, and Matteo Morgandi. 2021. Balancing Stability and Transition: A First Assessment of Regra de Permanência in Bolsa Familia (English). Washington, D.C.: World Bank Group. Fruttero, Anna, Alexandre Leichsenring, and Luis Henrique Paiva. 2020. "Social Programs and Formal Employment: Evidence from the Brazilian Bolsa Família Program." Hsu Rocha, Roberto, and Alison de Farias. 2021. "Formality Cost, Registration and Development of Microentreprenuers: Evidence from Brazil." Registration and Development of Microentreprenuers: Evidence From Brazil. November 10, 2021. 73 Lara de Arruda, Pedro; Lazarotto de Andrade, Marina; Falcao, Tiago; Teixeira Barbosa, Diana; Morgandi, Matteo. 2022. Challenges and Potentialities for Implementing Social Protection Responses to Emergency Through Decentralized Administration: Lessons from Brazil’s Auxilio Emergencial. Brasilia: World Bank Group. Lara Ibarra, G. and Campante Vale, R. 2022. Brazil 2020 data update: Methodological adjustments to the World Bank’s poverty and inequality estimates. Washington D.C.: World Bank Lara Ibarra, G. Paffhausen, A. L., Campante Vale, R., Damazceno, R. and Castaneda, C. forthcoming. Brazil Poverty and Equity Assessment: Looking ahead of two crises. Washington D.C.: World Bank Group MDS. 2015. “Brazil Without Extreme Poverty”. MDS: Brasilia Morgandi, Matteo, Tiago Falcao Silva, Marcelo Cortes Neri, Malin Linnea Sofia Ed, Katharina Maria Fietz, Gabriel Lyrio De Oliveira, and Maria Concepcion Lyrio De Oliveira. 2021. Enhancing Resilience of Low-Income Workers in Brazil: Financial Instruments and Innovations. “Morgandi, Matteo; Fietz, Katharina Maria; Ed, Malin Linnea Sofia; De Farias, Alison Rocha; Weber, Michael. 2020a. Enhancing Coverage and Cost-effectiveness of Brazil’s Unemployment Protection System: Insights from International Experience. Washington, D.C.: World Bank Group. Morgandi, Matteo, Katharina Fietz, and Malin Ed. 2020b. “Brazil - The Contrato Verde Amarelo Wage Subsidy: An Initial Assessment.” World Bank, Washington, DC. https://openknowledge.worldbank.org/handle/10986/34111 License: CC BY 3.0 IGO.” Murrugarra, E., and A. Isik-Dikmelik. 2020. “Productive Inclusion in Latin America; Policy and Operational Lessons – A synthesis note.” World Bank, Washington DC OECD, and World Bank. 2016. “OECD-WB Joint Methodological Paper, Final Draft.” SAGI. 2018. “Pronatec: uma análise crítica sobre três avaliações de impacto”. In Cadernos de Estudos Desenvolvimento Social em Debate. N. 30 (2018) . Brasília, DF : Ministério do Desenvolvimento Social; Secretaria de Avaliação e Gestão da Informação. Silva, Joana, Liliana D. Sousa, Truman G. Packard, and Raymond Robertson. 2021. “Employment in Crisis: The Path to Better Jobs in a Post-COVID-19 Latin America.” World Bank Latin America and Caribbean Studies. Washington, DC: World Bank. Silveira, Leonardo Souza, and Natália Siqueira Leão. 2020. "O impacto da segregação ocupacional por gênero e raça na desigualdade de renda no Brasil em três décadas (1986-2015)." Revista Latinoamericana de Población 14 (27): 41–76. Ulyssea, Gabriel. 2018. "Firms, Informality, and Development: Theory and Evidence from Brazil." American Economic Review 108 (8): 2015–47. Varghese Paul, B., P. Vasudeva Dutta, and S. Chaudhary. 2021. “Assessing the impact and cost of economic inclusion programs – a synthesis of evidence.” World Bank, Washington DC Venturini, Anna Carolina, Márcia Lima et al. 2020. "Educational Inequalities and Covid-19." Informative Racial Inequalities and Covid-19. Afro-Cebrap n. 3. World Bank. 2019. “Ending Learning Poverty: What will it take?”. Washington D.C.: World Bank Group 74 Annex 1: Additional Tables and Charts Table 18: Key labor market outcomes among household in Cadastro Unico and middle-class families Cadastro Vulnerable Unico Middle Upper Middle (bottom Class Class (60%- Rich Income distribution: 30%) (30%-59%) 89%) (top 10%) Race White (% adults) 73% 61% 46% 27% Number of adults 18+ 2.39 2.34 2.21 2.02 Family Number of children 0-17 1.91 1.07 0.61 0.52 Composition Number of Elderly 65+ 0.12 0.29 0.33 0.33 Dependency ratio 1.01 0.67 0.48 0.44 South/Southeast (%) 36% 57% 71% 73% Location North/Northeast (%) 59% 35% 20% 18% % adults out of LF 38% 25% 18% 16% Labor % adults unemployed 27% 11% 5% 3% market % adults working formally 11% 35% 55% 63% % working informally 70% 44% 29% 22% Human % w tertiary education (25+) 2% 6% 18% 58% capital % w secondary education (25+) 27% 35% 40% 29% Income Total household income (p.c.) 276 736 1,561 5,848 Source: World Bank 2023: Scenarios for Brazil Future. Based on PNADC 2020 Notes: Income percentiles obtained from SEDLAC data Table 19: Additional information on self-employed in % Formal Informal Self-declared BF 18 82 Female 39 33 Male 61 68 18–24 5 9 25–34 23 27 35–44 34 32 45–54 25 22 55–64 12 11 Rural 30 34 Urban 70 66 Employer 5.54 2.31 Not employer 94.46 97.69 Per capita income < 0.5 MWs 23 77 Female 37 32 Male 63 68 18–24 4 9 25–34 17 22 35–44 33 29 45–54 30 24 75 55–64 16 17 Rural 29 22 Urban 71 78 Employers 8 2 Source: PNADC 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. For sex, age, and rural/urban, columns add up to 100 percent. For share of formal/informal in rural/urban areas, row adds up to 100 percent. 76 Figure 65: Tenure in months by age group 100% 80% 60% 40% 20% 0% 18-24 25-34 35-44 45-54 55-64 18-24 25-34 35-44 45-54 55-64 18-24 25-34 35-44 45-54 55-64 Extreme poor in Bolsa Familia (P.c. Poor in Bolas Familia (P.c. Income In Cadastro Único not in Bolsa Familia Income below BRL 89) below BRL 178) 1 - 5 months 6 - 10 months 11 - 15 months 16 - 20 months 21 - 25 months > 25 months Source: RAIS 2019, Cadastro Único December 2019, Folha de pagamento 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education and able-bodied. 77 Table 20: In RAIS in Dec 2019 & In RAIS any month of 2019 In RAIS in December 2019 Updated CadU in Dec 2019 Decla red labor inco Declared Declared Declare me Declare labor Declared labor d labor (not d labor income labor income income in % in income (not in % in income (not in % in RAIS (in RAIS) RAIS) RAIS (in RAIS) RAIS) RAIS (in RAIS) RAIS) In BF 9.0 263 100 5.7 288 104 3.6 502 103 Not in BF 28.0 836 218 21.7 924 245 17.0 1,115 198 Did not work last week 11.5 69 10 7.4 75 13 3.5 378 12 Worked the last week 26.8 1045 374 21.1 1114 404 20.3 1,240 73 Employees with a formal contract 81.0 1294 1066 69.8 1306 1124 82.6 1,345 1,075 Apprentice 80.3 594 545 49.4 590 579 40.0 924 751 Military or civil servant 80.0 1438 1338 70.1 1476 1282 70.3 1,356 1,253 Employees without a formal contract 38.5 1004 701 28.3 1058 722 25.2 1,015 758 Domestic worker with a formal contract 36.0 1201 994 30.3 1231 997 37.5 1,693 934 Trainee 30.4 1255 756 21.8 1491 745 0.0 - 700 Self-employed (beak, standalone) 12.5 471 355 8.2 481 360 3.4 445 358 Domestic worker without a formal contract 11.1 641 377 7.7 671 384 4.7 1,492 318 Employer 7.0 1575 1543 4.9 1785 1445 0.0 - 1,805 Temporary worker in rural area 5.4 346 185 3.0 1306 189 1.6 423 201 Figure 66: Share of work-able adults registered in Cadastro Unico who also participated to the formal labor market according to RAIS in 2019 78 In RAIS any month of 2019 In RAIS in December 2019 In RAIS in Dec 2019 & Updated CadU In Dec 2019 90% 83% 81% 80% 80% 80% 70% 70% 70% 70% 60% 49% 50% % in RAIS 40% 40% 38% 38% 36% 30% 30% 30% 28% 28% 27% 25% 22% 21% 22% 20% 17% 20% 12% 13% 11% 9% 8% 8% 7% 10% 7% 6% 5% 5% 4% 4% 3% 5% 3% 2% 0% 0% 0% 0% 0% 0% 0% 79 Annex 2: Regression Results Table 21: Second-stage output Heckman Correction Model Dependent Variable = Log Hourly Wage Full Sample Bolsa Familia In Cadastro Único no Bolsa Familia (I) (II) (III) (IV) (V) (VI) Variables Female Male Female Male Female Male Age −0.0004 0.0406 *** 0.0254 0.0225 *** 0.0168 *** 0.0313 *** 0.0020 0.0071 0.0000 0.0043 0.0000 0.0019 Age^2 0.0001 ** −0.0004 *** −0.0004 −0.0002 *** −0.0002 *** −0.0003 *** 0.0000 0.0001 0.0000 0.0001 0.0000 0.0000 Literate/Incomplete Elementary −0.0155 0.0491 *** 0.0239 0.0357 −0.0269 0.0485 ** 0.0208 0.0156 0.0000 0.0222 0.0000 0.0211 Elementary / Incomplete Middle school −0.0011 0.0976 *** 0.0252 0.0491 −0.0092 0.0994 *** 0.0196 0.0146 0.0000 0.0210 0.0000 0.0197 Middle school / Incomplete High school 0.0368 * 0.1476 *** 0.0457 0.0693 ** 0.0334 0.1540 *** 0.0196 0.0149 0.0000 0.0217 0.0000 0.0199 High school/Incomplete Tertiary (college/technical course) 0.0936 *** 0.2219 *** 0.0845 *** 0.1449 *** 0.0941 *** 0.2221 *** 0.0196 0.0147 0.0000 0.0217 0.0000 0.0197 Tertiary 0.3222 *** 0.4066 *** 0.2628 *** 0.2279 *** 0.3161 *** 0.4108 *** 0.0202 0.0179 0.0000 0.0377 0.0000 0.0224 Urban 0.0042 −0.0036 −0.0099 −0.0195 ** 0.0066 −0.0126 0.0065 0.0065 0.0000 0.0093 0.0000 0.0088 Family Size 0.0145 *** 0.0141 *** −0.0169 0.0034 0.0064 *** 0.0185 *** 0.0019 0.0021 0.0000 0.0039 0.0000 0.0024 Number of children between 0 and 3 years old 0.0231 −0.0653 −0.0468 0.0388 0.0426 −0.1217 ** 0.0373 0.0421 0.0000 0.0528 0.0000 0.0611 Number of children between 4 and 7 years old −0.0007 0.0036 0.0378 ** −0.0203 −0.0106 0.0005 0.0373 0.0139 0.0000 0.0233 0.0000 0.0167 Number of children between 8 and 16 years old −0.0120 *** −0.0181 *** 0.0118 −0.0006 −0.0098 *** −0.0082 0.0026 0.0068 0.0000 0.0058 0.0000 0.0042 Constant 3.5123 *** 2.7244 *** 3.5498 ** 3.1267 *** 2.8443 *** 3.1056 *** 0.2767 0.3552 1.2395 0.2485 0.2824 0.2306 Lambda −0.1364 *** 0.1827 0.2649 0.0486 0.0806 *** −0.0409 ** Source: RAIS 2019. Note: *** p < 0.01, ** p < 0.05, * p < 0.10; Model controls for municipal and industry fixed effects; Groups are restricted to working age (18–64) not in education and able- bodied. 80 Annex 3: Methodology Latent Class Analysis LCA is a statistical segmentation method. The method “exploits the interrelations of an array of indicators through a fully-specified (that is, parametric) statistical model for organizing the target population into homogeneous groups” (OECD and World Bank 2016). The LCA in this paper is performed for two subgroups of the population based on PNADC. PNADC is chosen instead of Cadastro Único because Cadstro Único is missing important information on the labor market, such as previous work experience, unemployment, discouragement, and so on (for details see Annex 4). The two subgroups are (a) individuals of working age in vulnerable work having a per capita income below 0.5 MW and (b) individuals of working age who are out of work having a per capita income below 0.5 MW. Model Specification Baseline Model As a first step, the baseline model has to be specified. The baseline model is chosen by including only indicators (Collins and Lanza 2010). Those indicators are assumed to be pairwise independent within the latent classes (local independence assumption). Three different criteria should be taken into consideration when choosing the optimal number of classes: (a) Goodness of fit. The Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AIC) indicate the tradeoff between the “model’s ability to fit the data and the model’s parametrization” (OECD and World Bank 2016). The model with the number of classes that minimize BIC or AIC indicates that the specific model is the best option to balance the abovementioned tradeoff. (b) Misspecification tests. The misspecification test is used to test if the model is violating the local independence assumption. We use bivariate residuals to compare “the observed associations between pairs of indicators with the expected association under the assumption of local independence” (OECD and World Bank 2016). Bivariate residuals greater than 1 indicate residual correlation within latent groups. (c) Classification error. Active Covariates As a second step, active covariates are added to the baseline model because there could be differences in age, gender, and so on, within the latent classes. “Active covariates can be plugged-in to test such differences and, when significant, they can also reduce the classification errors of certain individuals. Although the main role of active covariates is to describe the latent classes, they can also interfere with the actual definition of the latent groups driven by the employment barrier indicators” (OECD and World Bank 2016). 81 Sub-group: Individuals in vulnerable work Baseline Model specification We chose six indicators for the model specification: (a) Children or disabled in the family. A dummy variable indicating 1 if an individual lives in a household with children equal to or below 10 years and disabled people. (b) Education. A categorical variable indicating different education levels of the individuals (primary education or lower; Completed primary or completed secondary; Secondary and above). (c) Low work intensity. A dummy variable indicating if an individual works below 35 hours per week. (d) Under/unpaid. A categorical variable indicating if an individual is (i) unpaid, (ii) paid below the hourly minimum wage, or (iii) paid equal to or above the hourly minimum wage. (e) Bolsa Familia. A dummy variable indicating if an individual lives in a family that receives BF. (f) BPC or pension. A dummy variable indicating if an individual lives in a family where at least one family member receives BPC or a pension. We start the model specification by including the six indicators mentioned above and running the model for 1 to 10 clusters. The goodness of fit indicator BIC is minimized at 5 clusters with a classification error of 29.8 percent (Figure 67). However, the model indicates misspecification due to selected bivariate residuals higher than 1. The misspecification can be resolved by increasing the number of classes or modelling explicitly the local dependencies between pairs of indicators. Due to an increasing classification error with an increasing number of latent classes, we choose to model explicitly the possibility of local dependencies between the pairs of indicators that indicate local dependency. By doing so, the goodness of fit BIC is again minimized at 5 clusters with a classification error of 26.8 percent (Figure 68). However, the 5-cluster model indicates other local residual dependencies within clusters. Also controlling for those dependencies, the goodness of fit would indicate a 4-cluster model, in turn again, with additional local dependencies. The 5-cluster model in this case had a classification error above 30 percent. For this reason, we decided to choose the model with 5 clusters and a classification error below 30 percent as specified in Figure 68. 82 Figure 67: Baseline model percentage variation of Figure 68: Baseline model percentage variation of goodness-of-fit criteria goodness-of-fit criteria including local dependencies 0.1% 50.0% 0.1% 50.0% Classification error (%) Classification error 40.0% 40.0% 0.0% 0.0% 4 5 6 7 8 9 10 4 5 6 7 8 9 10 % var. AIC and BIC 30.0% % var. AIC and BIC 30.0% (%) -0.1% -0.1% 20.0% 20.0% -0.1% -0.1% 10.0% 10.0% -0.2% 0.0% -0.2% 0.0% Var BIC Var AIC Class.Err. Var BIC Class.Err. Var AIC Source: Authors based on PNADC 2019. Source: Authors based on PNADC 2019. Inclusion of active covariates We chose five active covariates: (a) Sex (b) Occupation status (Formal CLT worker, Formal self-employed, Informal dependent worker, Informal self-employed) (c) Region (Capital, Pericapital, Other urban areas, Rural areas) (d) Marial status (e) Age group. A reduction of the classification error due to active covariates indicates that the model alone is not able to explain the latent classes. However, adding active covariates improves the model. In our case including active covariates reduced the classification error for the 5-cluster model to 15.5 percent. In addition, several inactive covariates have been added to the model. These variables do not influence the class definition and serve a descriptive purpose. Ten inactive covariates have been included in the model: (a) Single parent (b) Sector description (c) Share of labor income as a percentage of total income (d) Number of children between 0 and 4 years old (e) Number of children between 5 and 10 years old (f) Single earner (g) Employment rate at a local level (h) Individual receives BPC 83 (i) Household size (j) Number of employed in family. Sub-Group: Individuals out of work Baseline Model specification We choose six indicators for the model specification: (a) Children or disabled in the family. A dummy variable indicating 1 if an individual lives in a household with children equal to or below 10 years and disabled people. (b) Education. A categorical variable indicating different education levels of the individuals (Primary education or lower; Completed primary or completed secondary; Secondary and above). (c) Labor Market Distance. A categorical variable indicating if an individual is (i) out of labor force, (ii) discouraged, (iii) long-term unemployed, (iv) short-term unemployed. (d) Work Experience. A dummy variable that indicates if an individual has worked within the last year. (e) Bolsa Familia. A dummy variable indicating if an individual lives in a family that receives Bolsa Familia. (f) BPC. A dummy variable indicating if an individual lives in a family where at least one family member receives BPC. We start the model specification by including the six indicators mentioned above and running the model for 1 to 10 clusters. The goodness-of-fit indicator BIC is minimized at 9 clusters. (Figure 69). However, the model indicates misspecification due to selected bivariate residuals higher than 1. As mentioned for the other model, the misspecification can be resolved by explicitly modelling the possibility of local dependencies between the pairs of indicators that indicate local dependency. By doing so, the goodness- of-fit BIC is minimized at 7 clusters (Figure 70). With a classification error of 26.4, hence below 30, we prefer this model for our baseline specification. Figure 69: Baseline model percentage variation of Figure 70: Baseline model percentage variation of goodness-of-fit criteria goodness-of-fit criteria including local dependencies 0.1% 30.0% 0.1% 50.0% Classification error (%) Classification error (%) 0.0% 25.0% 0.0% 40.0% -0.1% 4 5 6 7 8 9 10 20.0% -0.1% 4 5 6 7 8 9 10 % var. AIC and BIC % var. AIC and BIC 30.0% -0.2% 15.0% -0.2% 20.0% -0.3% 10.0% -0.3% -0.4% 5.0% -0.4% 10.0% -0.5% 0.0% -0.5% 0.0% Var BIC Var AIC Class.Err. Var BIC Class.Err. Var AIC Source: Authors based on PNADC 2019. Source: Authors based on PNADC 2019. Inclusion of active covariates 84 We chose four active covariates: (a) Male/female (b) Region (Capital, Pericapital, Other urban areas, Rural areas) (c) Marial status (d) Age group A reduction of the classification error due to active covariates indicates that the model alone is not able to explain the latent classes. However, adding active covariates improves the model. In our case including active covariates reduced the classification error for the 6-cluster model to 24.06 percent. In addition, several inactive covariates have been added to the model. Again, those variables do not influence the class definition for a descriptive purpose. In this model, 10 inactive covariates have been included: (a) Share of labor income as a percentage of total income (b) Household receives a pension (c) Number of children between 0 and 4 years old (d) Number of children between 5 and 10 years old (e) Household size (f) Number of adults in family (g) Number of old people in family (h) Number of employed in family (i) Employment rate at a local level (j) Individual receives BPC 85 Output of selected Latent Class model Table 22: Characteristics of in-work individuals in the chosen specification (%) Unpaid low- educated Educated Low-educated Educated informal male full- underpaid informal Low-educated dependent time formal informal female informal male female employees workers in BF workers in BF in BF workers Cluster Size 32% 25% 23% 15% 5% Children (0–10 years old) or disabled are in family 72 39 70 90 53 Family receives BF 44 60 64 79 68 Primary education or lower 31 84 20 60 51 Completed primary or completed secondary education 21 8 26 25 18 Indicators Secondary and above 49 8 54 16 31 Works less than 35 hours per week 6 47 58 25 65 Works equal to or above 35 hours per week 94 53 42 75 35 Paid above equal or above 1 MW per hour 98 40 56 56 7 Unpaid 0 0 0 0 86 Paid BELOW 1 MW per hour 2 60 44 44 7 Female 27 36 77 2 69 Male 73 64 23 98 31 Formal Dependent Worker 81 0 1 3 0 Formal Self-employed + Employer 6 10 8 8 0 Informal Self-Employed 6 51 39 44 0 Active Covariates Informal Dependent Worker 7 39 52 45 100 Capital 24 11 21 8 6 Pericapital 22 13 17 10 8 Other Urban 44 40 48 44 23 Other Rural 9 35 14 38 63 Married 74 65 47 97 66 18–24 12 3 25 12 26 25–34 32 6 39 41 26 35–54 51 68 34 44 40 55–64 5 23 2 3 7 Single parent / caretaker 5 5 12 1 2 Single earner 58 46 36 57 8 Share of labor income in total HH income Below 20% 0 5 3 2 11 Inactive Covariates Between 20% and 40% 1 8 7 6 12 Between 40% and 60% 4 11 10 11 13 Between 60% and 80% 7 15 15 21 17 Between 80% and 100% 88 61 64 60 47 No children below 5 years 59 84 64 47 73 1 child between 0 and 5 years 33 13 29 43 22 2 children between 0 and 5 years 6 2 5 9 4 3 children between 0 and 5 years 1 0 1 1 1 4+ children between 0 and 5 years 0 0 0 0 0 No children between 5 and 10 years 56 77 56 46 66 86 Unpaid low- educated Educated Low-educated Educated informal male full- underpaid informal Low-educated dependent time formal informal female informal male female employees workers in BF workers in BF in BF workers Cluster Size 32% 25% 23% 15% 5% 1 child between 5 and 10 years 34 18 34 41 25 2 children between 5 and 10 years 9 4 9 12 8 3 children between 5 and 10 years 1 1 1 2 1 4+ children between 5 and 10 years 0 0 0 0 0 1 employed in family 58 46 36 57 8 2 employed in family 33 39 48 35 55 3 employed in family 7 11 11 6 21 4 employed in family 2 3 3 2 9 5+ employed in family 0 1 1 1 6 Employment rate at a local level Below 20% 0 1 1 1 1 Between 20% and 40% 8 13 9 14 9 Between 40% and 60% 45 48 45 48 38 Between 60% and 80% 43 35 42 3% 43 Between 80% and 100% 4 4 4 4 9 BPC 2 3 3 3 4 Household of 1 0 8 2 0 1 Household of 2 7 18 11 6 11 Household of 3 26 24 24 27 21 Household of 4 32 23 27 31 29 Household of 5 19 14 18 20 16 Household of 6 9 7 8 8 9 Household of 7 4 3 4 4 5 Household of 8+ 3 3 4 4 8 Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education. 87 Table 23: Characteristics of out-of-work individuals in individuals in the chosen specification (%) Inactive Better Unmarried Short-term, low- educated Short-term Rural prime age low- educated young Educated urban discouraged inactive educated women jobseekers mothers unemployed women in BF men unemployed Out-of-labor force 82 22 61 8 35% 73 7 Discouraged 15 17 24 24 60 20 44 Long-term unemployed 2 45 12 0 5 7 0 Short-term unemployed 0 16 4 68 1 0 49 Children (0–10 years) or disabled are in family 56 47 97 60 77 78 47 Worked in the past 4 0 10 96 20 7 84 Primary Edu. or less 81 18 25 19 66 79 97 Incomplete secondary 8 19 29 26 17 10 3 Secondary and above 10 63 47 55 17 11 0 Family receives BF 38 34 59 39 98 26 51 family receives BPC 8 5 3 4 4 25 5 Female 87 58 99 49 76 33 26 Male 13 42 1 51 24 67 74 18–24 0 40 29 31 16 8 5 25–34 2 25 50 35 29 14 12 35–54 50 31 21 30 52 51 63 55–64 47 4 0 3 3 27 19 Capital 10 28 18 22 0 19 8 Pericapital 16 25 20 20 2 12 11 Other Urban 44 38 44 47 36 52 47 Other Rural 30 8 18 11 63 17 34 Married 74 32 77 44 73 30 59 Family receives pension 34 20 11 15 15 34 19 Individual receives BPC 8 5 3 4 4 24 5 No children below 5 84 78 38 71 61 82 83 1 child between 0 and 5 13 18 49 25 30 15 14 2 children between 0 2 3 12 4 7 3 3 3+ children 0–5 0 0 1 0 1 0 0 No children 5–10 78 77 47 67 55 76 76 1 child 5–10 17 18 40 26 33 18 19 2 children 5–10 4 4 11 6 10 5 4 3+ children 5–10 1 0 2 0 2 1 1 1 adult 10 11 5 13 7 18 17 2 adult 46 35 68 45 56 35 46 3 adult 26 29 15 23 19 27 22 4 adult 12 16 8 12 10 13 10 5+ adult 6 9 5 7 7 8 5 No old person 91 91 96 94 94 86 92 No one employed 47 42 23 46 44 61 61 1 employed in family 42 43 67 42 46 29 31 2 employed in family 9 12 7 10 8 8 6 3+ employed in family 2 3 3 2 2 2 2 Household of 1 3 4 0 5 1 9 11 Household of 2 25 15 5 16 10 21 20 Household of 3 24 24 27 26 22 21 22 Household of 4 22 25 32 24 26 20 21 Household of 5+ Share of labor income in total HH income Below 20 % 43 31 21 33 47 53 49 Between 20% and 40% 5 3 3 4 6 6 5 88 Inactive Better Unmarried Short-term, low- educated Short-term Rural prime age low- educated young Educated urban discouraged inactive educated women jobseekers mothers unemployed women in BF men unemployed Between 40% and 60 % 7 5 6 5 10 7 5 Between 60% and 80 % 6 6 11 7 14 4 6 Between 80% and 100 % 40 54 58 52 23 30 36 Employment rate at a local level Below 20% 5 2 2 2 11 4 6 Between 20% and 40% 21 16 15 13 30 20 24 Between 40% and 60% 49 53 49 50 42 50 47 Between 60% and 80% 23 29 32 33 16 25 22 Between 80% and 100% 1 1 2 2 1 1 1 Source: PNAD 2019. Note: Groups are restricted to individuals of working age (18–64) and not in education. Table 24: Probit on being employed Employed Coef. St.Err. t-value p-value [95% Conf Interval] Sig Male 0.813 0.006 139.68 0.000 0.802 0.825 *** Family receives BF −0.403 0.007 −56.79 0.000 −0.417 -0.389 *** Family receives BPC −0.374 0.225 −1.66 0.097 −0.814 0.067 * 18–24 0.000 . . . . . 25–34 0.470 0.010 48.90 0.000 0.451 0.489 *** 35–54 0.618 0.009 68.09 0.000 0.600 0.636 *** 55–64 0.003 0.011 0.28 0.783 −0.018 0.023 Primary Education or 0.000 . . . . . lower Complete primary or 0.182 0.009 20.50 0.000 0.164 0.199 *** completed secondary Secondary and above 0.380 0.007 55.90 0.000 0.367 0.393 *** Married 0.079 0.006 12.79 0.000 0.067 0.091 *** Individual receives −0.054 0.225 −0.24 0.811 −0.495 0.388 BPC Constant −0.418 0.010 −40.03 0.000 −0.438 −0.397 *** Mean dependent var 0.685 SD dependent var 0.465 Pseudo r-squared 0.125 Number of obs 239,599 Chi-square 37,289.259 Prob > chi2 0.000 AIC 261,449.362 BIC 261,563.616 Sample: Population 18–64, PNAD Continua 2019. Note: *** p<0.01, ** p<0.05, * p<0.1. Table 25: Probit on being formally employed Formally employed Coef. St.Err. t-value p-value [95% Conf Interval] Sig Male 0.479 0.006 84.99 0.000 0.468 0.490 *** Family receives BF −0.933 0.008 −114.1 0.000 −0.949 −0.917 *** 1 Family receives BPC −0.884 0.286 −3.09 0.002 −1.445 −0.323 *** 18–24 0.000 . . . . . 25–34 0.402 0.010 40.58 0.000 0.383 0.422 *** 35–54 0.549 0.009 58.35 0.000 0.531 0.568 *** 55–64 0.121 0.011 10.88 0.000 0.099 0.142 *** Primary Education or 0.000 . . . . . lower Complete primary or 0.336 0.009 37.64 0.000 0.319 0.354 *** 89 completed secondary Secondary and above 0.730 0.007 108.17 0.000 0.717 0.743 *** Married 0.182 0.006 29.91 0.000 0.170 0.194 *** Individual receives BPC 0.335 0.287 1.17 0.242 −0.227 0.897 Constant −1.231 0.011 −112.7 0.000 −1.252 −1.210 *** 9 Mean dependent var 0.405 SD dependent var 0.491 Pseudo r-squared 0.149 Number of obs 239,599.000 Chi-square 48,311.755 Prob > chi2 0.000 AIC 275,099.766 BIC 275,214.020 Sample: Population 18–64, PNAD Continua 2019. Note: *** p<0.01, ** p<0.05, * p<0.1. Annex 4: Data Description Cadastro Único Cadastro Único (Single Registry) is a national registry that was started in 2001 with Bolsa Escola, the previous version of Bolsa Família, and is currently used as a platform for all federal social programs in Brazil. It is managed by the Ministry of Citizenship (MoC) and contains sociodemographic and economic information of all members of vulnerable families in the country, such as education, working status, income, family composition, dwelling location, and accessible public services, among others. As in the present case, the data extracted by the MoC usually refer to the middle of December of each year. This enables one to track families and individuals over the years considering their unique ID (NIS). RAIS Relatório Anual de Informações Sociais is a form that every firm in the country that has employees is obligated to fill and send to the government every year. It contains identified (NIS) information of each employee occupation and monthly compensation. It also comprises other useful data such as the date that the employee entered the firm and eventual dismissal date in that year and the reasons. Information about the firm, such as the economic sector, location, and size (number of employees) are also included. Currently, the Ministry of Economy is responsible for collecting and maintaining the dataset. PNAD Continua Household Survey PNAD Continua is a continuous national household sample survey administered by IBGE that monitors the labor force and other variables that are key for Brazil’s socioeconomic development. The survey produces monthly information related to a restricted set of labor force information, quarterly information on complementary topics related to the work force and annual indicators on complementary topics such as ICT use, income sources, other forms of work, fertility and migration. The monthly information is representative only at the national level and the annual and quarterly data is representative down to municipal level. It is well-documented that the PNADC tends to underestimate participation and coverage of social programs (Lara Ibarra & Campante Vale, 2022). Table 26 and Table 27 depict the main differences in the data sets used. Table 26: Differences in datasets used I. Differences between datasets PNADC Cadastro Único Registered at Cadastro Único Per capita income below 0.5 MWs Officially registered in Cadastro Único 90 Receiving Bolsa Familia Self-declaration of receiving Bolsa Officially registered if receiving Bolsa Familia Familia Able-bodied Question “Reason not searched for a Question “Does the individual have job” with “Health or Pregnancy” any disabilities” answered with “Yes” Work-able • Working-age individuals (18–64 years) • Not enrolled in full-time education • Able-bodied Advantages of survey • Distinguish between formal and • Track of families over time informal self-employed • Inclusion of all families that are • Identify the out-of-labor-force and registered in the social registry their reasons Comparing the poor and vulnerable in both data sets, similarities but also differences can we found. The distribution of age, urban/rural and family composition are very similar across datasets. Simultaneously a difference in sex can be observed. One potential explanation is that men are intentionally left out in Cadastro Único since the benefit is paid out preferable to women. Also, PNADC shows a worse coverage of BF recipients in the South and South-East regions. This is consistent with the fact that BF is less frequent in the richer states and so the survey sample is less likely to represent well this minority group. Last, previous analysis has shown that PNADC does not cover well indigenous people and homeless (Table 27). Table 27 Basic comparison statistics PNADC and Cadastro Único PNAD Cadastro Único Inidivudals living below 0.5 MWs 34,499,543 35,164,600 Share of individuals in Cadastro Único receiving BF 54% 53% Bolsa Familia Recipients (individuals) Sex Male 46% 34% Female 54% 66% Age 18-24 18% 18% 25-34 29% 30% 35-44 28% 27% 45-54 17% 18% 55-64 8% 7% Urban / Rural Rural 32% 32% Ubran 68% 68% Region North 19% 13% Northeast 59% 53% Southeast 13% 24% South 4% 6% Center West 5% 4% Family Composition 0 Children 24% 22% 1 Child 34% 33% 91 2 Children 25% 26% 3 Children 11% 12% 4+ Children 6% 7% Source: PNAD 2019, Cadsatro Único 2019 Note: Sample is restricted to individuals at working age (18-64 years old), not in education and able-bodied 92