CAF - Out of Agriculture, Still Vulnerable - Rural Labor Markets in Latin America de 2026

First version: September 3, 2026

Small sections of text that are less than two paragraphs may be quoted without explicit permission as long as this document i s acknowledged. Findings, interpretations and conclusions expressed in this publication are the sole responsibility of its author(s) and cannot be, in any way, attributed to CAF , its Executive Directors or the countries they represent. CAF does not guarantee the accuracy of the data included in this publication and is not, in any way, responsible for any consequences resulting from its use.

C A F - W O R K I N G P A P E R # 2 0 2 6 / 10

Out of Agriculture, Still

Vulnerable: Rural Labor

Markets in Latin America

Dolores de la Mata1 | Mariana Marchioni2 | Ivan Albina3 | Juan Barnes4

1 CAF --Banco de Desarrollo de América Latina y el Caribe. mdelamata@caf.com 2 CEDLAS - IIE - Universidad Nacional de La Plata and CONICET. marchionni.mariana@gmail.com

3 CEDLAS - IIE - Universidad Nacional de La Plata and

CONICET.

ivan_albina@hotmail.com 4 CAF --Banco de Desarrollo de América Latina y el Caribe. jbarnes@caf.comC A F - W O R K I N G PA P E R # 2 0 2 6 / 1 0

F i r s t v e r s i o n : S e p t e m b e r 3 , 2 0 2 6

Out of Agriculture, Still Vulnerable: Rural Labor Markets in Latin America Dolores de la Mata Mariana Marchioni† Ivan Albina‡ Juan Barnes§ Abstract Rural poverty has declined substantially in Latin America over the past two decades, yet vulnerability remains high and rural–urban gaps persist. This paper examines the factors associated with this persistence using harmonized household survey microdata for 14 Latin American countries over 2003–2023. We document three main patterns. First, persistent rural disadvantage does not appear to reflect a worsening composition of the rural population: rural–urban gaps narrowed in household demographics, educational attainment, and access to basic services. Second, rural employment shifted markedly away from agriculture, with the agricultural share falling from 53% to 43%, while informality also declined. Yet convergence in job quality was much more limited: rural workers remain more informal, less concentrated in high-complexity occupations, and lower paid than urban workers, with lower returns to education. Importantly, these disadvantages extend to the increasingly prevalent non-agricultural rural jobs. Third, vulnerability is not confined to agricultural households: poverty remains substantial among non-agricultural rural households, which now account for nearly half of the rural poor. Overall, the evidence highlights a central distinction between sectoral reallocation and improvements in job quality: moving out of agriculture has not, by itself, eliminated rural vulnerability.

Keywords: Rural labor markets, income diversification, structural transformation, Latin America, vulnerability. CAF –Banco de Desarrollo de América Latina y el Caribe. delamata@caf.com †CEDLAS - IIE - Universidad Nacional de La Plata and CONICET. ivan_albina@hotmail.com ‡CEDLAS - IIE - Universidad Nacional de La Plata and CONICET. marchionni.mariana@gmail.com

§CAF –Banco de Desarrollo de América Latina y el Caribe. jbarnes@caf.com Small sections of text that are less than two paragraphs may be quoted without explicit permission as long as this document is acknowledged. Findings, interpretations and conclusions expressed in this publication are the sole responsibility of its author(s) and cannot be, in any way, attributed to CAF, its Executive Directors or the countries they represent. CAF does not guarantee the accuracy of the data included in this publication and is not, in any way, responsible for any consequences resulting from its use. ©2026 Corporación Andina de FomentoDE LAMATA ET AL. 1 1|INTRODUCTION Over the past two decades, Latin America has experienced a substantial decline in poverty. The regional poverty rate fell by roughly half between the early 2000s and the eve of the COVID-19 pandemic, from 42.4% in 2003 to 23.6% in 2019 (Gasparini et al., 2023). After the temporary increase associated with the pandemic, poverty resumed its downward trend (SEDLAC, 2025). These gains reflect a combination of sustained economic growth, improvements in labor market conditions, and the expansion of social protection systems across the region (Alvaredo and Gasparini, 2015). Despite these broad-based improvements, a large rural-urban divide persists. Rural areas—home to approximately 120 million people, about 18% of the population in Latin America (CELADE, 2024)—continue to exhibit significantly higher levels of poverty and economic vulnerability than urban areas. As shown in Figure 1, poverty declined somewhat faster in rural areas, falling from 65% in 2003 to 34% in 2023. Yet rural poverty remains high in absolute terms and about twice as high as urban poverty (17%) by the end of the period. Thus, despite substantial progress, rural Latin America continues to combine high levels of

vulnerability with sizable gaps relative to urban areas. 2003 2008 2013 2018 2023 0% 10% 20% 30% 40% 50% 60% 70%Poverty rate 64.9 55.6 45.9 39.7 34.2 37.3 27.9 21.4 17.1 16.8 Rural Urban 1 F I G U R E 1Rural and urban poverty in Latin America, 2003–2023.Notes:Poverty is defined as the share of the population living in households with income below a threshold of USD 6.85 per person per day (PPP).Source:Authors, based on data described in Section 2. This raises the question of what explains the persistence of vulnerability in rural Latin America and how it relates to the broader process of economic development. By documenting the evolution of rural household characteristics, labor-market outcomes, and income composition over the first quarter of the twenty-first century, this paper seeks to shed light on the factors associated with persistent rural vulnerability. To this end, we use microdata from harmonized household surveys for 14 countries spanning the period 2003–2023, drawn from the Socio-Economic Database for Latin America and the Caribbean (SEDLAC), a joint initiative of CEDLAS and the World Bank (CEDLAS and The World Bank, 2023). This broad spatial and temporal coverage allows us to construct a consistent and detailed picture of rural labor markets and household income structures across the region. Our analysis proceeds in three steps. We first examine whether persistent rural dis-DE LAMATA ET AL. 2 advantage is associated with persistent gaps in the observable characteristics of the rural population, comparing the evolution of demographic characteristics, human capital, and living conditions across rural and urban households. We then ask whether changes in rural

labor markets may have altered the economic opportunities available to rural workers, documenting the evolution of the sectoral structure of employment, labor arrangements, and job quality, as captured by informality, occupational complexity, and hourly earnings. Finally, because household vulnerability depends not only on individual employment outcomes but also on how different income sources are combined within households, we examine the composition of labor and non-labor income across household types and along the rural income distribution. Our findings reveal a sharp contrast between substantial transformation and persistent vulnerability. First, the persistence of rural disadvantage does not appear to be driven by the composition of the rural population. Rural households became smaller, dependency ratios declined, educational attainment improved substantially, and access to basic services expanded, narrowing rural–urban gaps along all of these dimensions. In particular, the narrowing of education gaps is important from a labor-market perspective, as it indicates that rural workers did not fall increasingly behind in one of the key observable determinants of employment and earnings. These patterns suggest that persistent rural vulnerability cannot be accounted for by a worsening composition of the rural population and point instead to the economic opportunities available in rural areas. The labor-market evidence reveals a similar tension between substantial change and persistent disadvantage. Rural employment shifted markedly away from agriculture between 2003 and 2023, with the agricultural share falling from 53% to 43% and non-agricultural activities expanding accordingly. Over the last decade, non-agricultural employment became, for the first time, the main source of rural jobs. This reallocation was accompanied by some improvements in employment conditions: informality declined from 74% to 67%, largely through reductions in low-skilled self-employment and unpaid work. Yet these changes brought only limited convergence toward urban labor markets. Rural informality remains substantially higher than urban informality (67% versus 45% in 2023), occupational upgrading has been modest, and sizeable earnings gaps persist even after accounting for

observable worker characteristics. Returns to education also remain lower in rural than in urban labor markets, suggesting that the substantial gains in rural educational attainment have not been matched by comparable improvements in the rewards to human capital. Importantly, the limitations of rural labor markets extend beyond agriculture. Non-agricultural jobs have become increasingly prevalent in rural areas and generally offer better conditions and higher earnings than agricultural jobs, but they remain considerably more informal, less intensive in high-complexity occupations, and lower paid than urban employment. Thus, the substantial movement out of agriculture has not been accompanied by a comparable narrowing of rural–urban gaps in job quality. These labor-market opportunities and constraints are reflected in the way rural households generate their incomes and, ultimately, in their exposure to poverty. Agricultural households remain the most vulnerable segment of the rural population. They have the highest poverty rates and rely more heavily on agricultural self-employment, with less diversified income portfolios and a greater incidence of non-monetary income, largely associated with own-consumption. Yet rural poverty is far from confined to agriculture. In 2023, poverty reached 41% among agricultural households but remained as high as 27% among non-agricultural rural households. Non-agricultural households now represent 54% of the rural population and account for nearly half (46%) of the rural poor. This reinforces the broader message from the labor-market evidence: the persistence of rural vulnerability reflects not only the continued disadvantage of households tied to agriculture, but also theDE LAMATA ET AL. 3 limited economic opportunities available outside it. This paper contributes to the literature on rural development in three ways. First, it provides updated and comparable evidence on the evolution of rural households, employment, and income structures across Latin America. Extending the analysis of rural labor markets in CAF (2026) and building on the harmonized household-survey tradition of SEDLAC (CEDLAS and The World Bank, 2023), we assemble one of the most comprehensive harmonized rural datasets for the region to date and use it to update earlier cross-country portraits of rural Latin America (de Ferranti et al., 2005; Reardon et al., 2001; Dirven, 2004). Second, the paper highlights the distinction between sectoral reallocation and improvements in job quality. We view the changes documented here as micro-level manifestations of the long-run process of structural transformation—the secular reallocation of labor and output away from agriculture—studied since Lewis (1954) and Kuznets (1966) and synthesized by Herrendorf et al. (2014). Our evidence shows that substantial reallocation away from agriculture can coexist with persistent informality, limited occupational upgrading, low earnings, and weak returns to human capital in rural labor markets (La Porta and Shleifer, 2014). In this sense, we shift attention from whether rural workers move out of agriculture to the quality of the opportunities available to them outside it. Third, we bring a household and distributional perspective to this transformation, documenting how dependence on agriculture, non-agricultural wage employment, and other income sources vary across rural households and along the income distribution. This relates to a longstanding literature on rural livelihood diversification and non-farm employment (Ellis, 1998; Lanjouw and Lanjouw, 2001; Haggblade et al., 2010), while emphasizing that vulnerability remains substantial even among rural households whose livelihoods are no longer primarily tied to agriculture. The rest of the paper is organized as follows. Section 2 describes the data and main definitions. Section 3 examines changes in the composition of the rural population and the evolution of rural–urban gaps in observable characteristics. Section 4 documents changes in the structure and quality of rural employment. Section 5 analyzes household income sources and their distribution across rural households. Section 6 discusses the main implications of

the findings, and Section 7 concludes. 2|DATA 2.1|Data source and sample coverage We draw on microdata from harmonized household surveys from the Socio-Economic Database for Latin America and the Caribbean (SEDLAC), a joint initiative of CEDLAS (Universidad Nacional de La Plata) and the World Bank (CEDLAS and The World Bank, 2023). SEDLAC compiles and processes microdata from official household surveys conducted by national statistical offices across Latin America. A key advantage of SEDLAC is the harmonization of variables across countries and over time, which ensures comparability in core indicators. Our analysis covers fourteen countries—Bolivia, Brazil, Chile, Colombia, Costa Rica, Ecuador, El Salvador, Honduras, Mexico, Panama, Paraguay, Peru, the Dominican Republic, and Uruguay—spanning the period from 2003 to 2023. This extensive spatial and temporal coverage allows us to construct a detailed and comparable picture of income sources, labor market outcomes, and living conditions across rural and urban areas in Latin America. For the sake of exposition, the paper focuses on regional patterns. Unless otherwise noted, regional statistics are computed as simple averages across countries, so that each country receives the same weight. This approach captures the experience of the “typical”DE LAMATA ET AL. 4 Latin American country rather than that of the average individual in the region, which would be dominated by the largest economies, particularly Brazil and Mexico. Countrylevel results documenting the heterogeneity underlying these regional averages are reported in the Appendix. 2.2|Definitions Rural areas.The definition of rural areas follows the official criteria established by each country’s national statistical office. These criteria typically rely on combinations of population thresholds, administrative status, settlement characteristics, and, in some cases, economic activity or access to services, as summarized as summarized in Table A.1 in the

Appendix. This is the most feasible approach given the available data, and it has the advantage of aligning with the definitions used by national statistical offices, which are often the basis for policy design in each country. For these reasons, the resulting estimates should be interpreted as reflecting nationally defined rural areas rather than a strictly harmonized cross-country concept of rurality.1 While these limitations may affect the magnitude of measured rural–urban differences, they are unlikely to alter the broad regional patterns that are the focus of this paper, although they should be kept in mind when comparing individual countries or interpreting rural–urban gaps. Labor market outcomes.We classify individuals’ labor market status, distinguishing between employment, unemployment, and inactivity. Among the employed, we further differentiate four labor categories based on their main occupation: employers (those who own a business and hire paid workers), wage employees (including domestic workers, unless the survey clearly identifies them as self-employed), own-account workers (including cooperative members), and unpaid workers (including unpaid family workers, unpaid apprentices, and similar categories). Regarding labor informality, we use an extended legal definition that includes wage earners without pension rights, non-professional self-employment, and unpaid work. This broader measure is particularly relevant in rural areas, where informality often takes the form of low-productivity self-employment and unpaid family work. Finally, we classify occupations into four broad groups based on one-digit codes of the current version of the ILO’s International Standard Classification of Occupations (ISCO-08): high-complexity occupations, including managers, professionals, and technicians (ISCO 1–3); medium-complexity occupations, including clerical, service and sales, skilled agricultural, craft, and machine-operation occupations (ISCO 4–8); low-complexity occupations, corresponding to elementary occupations (ISCO 9); and other occupations, including armed

forces and non-classified categories. Because this classification assigns the same complexity level to all workers within a given one-digit occupational category, variation arises only 1More generally, national definitions of rural and urban areas differ not only in their specific criteria across countries but also in their shared reliance on a binary classification. A growing methodological literature argues that the rural–urban distinction may be better understood as a continuum than as a dichotomy, and has sought to move beyond it—an approach that the public-use household surveys available for Latin America do not allow us to implement, as they lack the geographic detail required to classify territories consistently along such a continuum at the individual or household level. The Degree of Urbanisation (DEGURBA) framework, developed by the European Commission, FAO, UN-Habitat, ILO, OECD, and the World Bank, addresses both concerns: it harmonizes territorial definitions across countries and captures rurality as a gradient, classifying territories using population density, settlement size, and spatial contiguity (European Commission et al., 2021). Other contributions have similarly emphasized the limitations of static and dichotomous definitions: ECLAC and IFAD’s work on rural transformation in Latin America (Gaudin and Padilla Pérez, 2023; Gaudin et al., 2024), and Cattaneo et al. (2021), who propose Urban–Rural Catchment Areas based on travel times to urban centers of different sizes.DE LAMATA ET AL. 5 across occupations. It should therefore be interpreted as a coarse ranking of occupational complexity.2 Country and year coverage for ISCO data is reported in Table A.2 in the Appendix. Agricultural employment.The sector of the main occupation is identified using harmonized industry classifications based on the one-digit level of the International Standard Industrial Classification of All Economic Activities (ISIC Rev. 3). Depending on the country and survey, the original industry variable is either already reported at this level or is available at a more

disaggregated two-, three-, or four-digit level and then aggregated to the common one-digit classification. For the purposes of the analysis, the agricultural sector is defined narrowly as agriculture, hunting, and forestry, corresponding to category 1 of the one-digit ISIC Rev. 3 classification. All other sectors, including fishing, mining, manufacturing, construction, commerce, and services, are classified as non-agricultural activities. Agricultural households.According to the main occupation of the household head or spouse, we define agricultural households as a specific subgroup of rural households: those in which the head or spouse reports agriculture as their main occupation. Notice that this category is defined exclusively within the rural population.3 Within agricultural households, we further distinguish four mutually exclusive groups: agricultural employer households, agricultural wage-worker households, self-employed agricultural households, and other agricultural households.4 Income measurement.Household income is constructed by aggregating all monetary and non-monetary sources reported in the surveys. This includes labor income from both wage employment and self-employment, as well as non-labor income such as pensions, transfers, and other private or public sources. Whenever available, income measures incorporate imputed values for self-consumption and in-kind earnings, which are particularly relevant in rural contexts. The availability of these components varies across countries, and no additional correction is applied when they are not captured by the original survey. Income variables are expressed in real terms using country-specific price indices and converted into a common currency using purchasing power parity (PPP) exchange rates (2017 international dollars). Since the analysis relies on a fixed international poverty line, rather than on country-specific poverty lines that may vary across regions or between urban and rural areas, we follow SEDLAC and adjust rural incomes upward by 15 percent to account for the lower cost of living in rural areas. This adjustment is applied uniformly across countries and rural areas, and therefore does not capture differences in local price levels or degrees of

rurality within countries. 2This classification is a proxy for occupational complexity rather than a direct measure of task content. Taskbased measures can capture variation within occupations; for example, Bonavida and Gasparini (2026) use PIAAC data to construct individual-level indices of abstract, routine, and manual task intensity. Such measures are not available at the scale required for our analysis. 3This definition is related to that proposed by Gáfaro et al. (2025), but adopts a more restrictive criterion. While those authors classify a household as agricultural if at least one member is employed in the sector, we focus on the occupation of the head or spouse. This choice is intended to exclude cases in which the household’s link to agriculture is only marginal—for example, when the only member engaged in agricultural activities is a child, possibly as an unpaid worker—and to better capture the household’s primary economic orientation. In practice, however, the share of households classified as agricultural does not differ substantially across definitions. 4The household is assigned to the highest-ranking category observed between the household head and spouse, according to the following hierarchy: employer, wage worker, self-employed worker, and other.DE LAMATA ET AL. 6 2.3|The rural population in Latin America Having defined the rural and urban populations used throughout the paper, it is useful to briefly characterize the evolution of the rural population covered by our data. Figure 2 reports the share of the population living in rural households in the average Latin American country between 2003 and 2023, together with the share of rural households whose head or spouse is primarily employed in agriculture. On average across the region, the share of the population in rural households declined from 31.6% to 25.1%, reflecting the welldocumented process of urbanization in Latin America. Over the same period, the share of the rural population living in agricultural households fell from 56% to 46%, reflecting the

ongoing diversification of rural labor markets away from agriculture. These broad regional trends provide useful context for the analysis that follows. Moreover, although levels vary widely across countries, the decline in both the rural and the agricultural share is observed in most of them, indicating that these regional patterns reflect a widespread phenomenon rather than the experience of a few large economies (see Figure A.1 in the Appendix). 2003 2008 2013 2018 2023 0% 5% 10% 15% 20% 25% 30% 35%% of total population 31.6 29.7 27.4 25.9 25.1 17.5 15.8 14.6 13.6 11.5 Population in rural households Population in agricultural households 1 F I G U R E 2Share of the population living in rural and agricultural households.Notes:Simple averages across 14 Latin American countries.Source:Authors, based on data described in Section 2. 3 |CHANGING RURAL POPULATION AND THE PERSISTENCE OF RURAL VULNERABILITY A natural starting point for understanding the persistence of rural vulnerability is to ask whether it reflects a process of increasing negative selection; that is, whether the rural population has become progressively more disadvantaged relative to the urban population in ways that could account for persistent rural–urban gaps. Table 1 provides little support for this hypothesis. Rural households continue to lag behind urban households along most observable characteristics, but these gaps narrowed markedly between 2003 and

2023. Convergence is observed across a broad range of demographic, socioeconomic, and living-condition indicators. Rural households became smaller, dependency ratios declined, educational attainment improved substantially, and access to basic services expanded. For example, the rural-to-urban ratio in the share of household heads with secondary educationDE LAMATA ET AL. 7

or more nearly doubled, from 0.23 to 0.42, while the ratio in access to electricity rose from 0.72 to 0.96. This broad-based improvement was also accompanied by a modest narrowing of the rural–urban gap in average per-capita income, whose ratio increased from 0.47 to 0.65. These relative improvements are particularly notable once we consider that rural-tourban migration could, in principle, work against them. Because migration in Latin America tends to be positively selected—migrants are younger and more educated than those who stay behind (de la Mata and Rosá, 2026)—their departure might worsen the observable composition of the population that remains classified as rural, biasing it toward greater relative disadvantage and making convergence harder to detect. Yet Table 1 shows the opposite: rural households improved relative to urban ones across demographic, humancapital, and living-condition indicators. In part, this may reflect a weakening of the ruralto-urban migration process itself: flows in Latin America have declined over time, both in absolute terms and relative to the size of the rural population (Rodríguez Vignoli, 2008; de la Mata and Rosá, 2026). A separate concern is that the rural–urban classification of localities used by national statistical offices is not fixed over time, and it may bias the observed rural–urban gap in either direction. On the one hand, because these classifications are typically revised only infrequently and in connection with census rounds or other major statistical updates (United Nations, 2005), some localities that have effectively urbanized may continue to be recorded as rural for extended periods. To the extent that these lagging, relatively dynamic areas remain in the rural sample, they improve its measured outcomes and make the observed gap appear narrower than it truly is. On the other hand, when reclassification does occur, it

tends to update precisely the most dynamic localities, removing them from the rural sample and worsening the composition of the population that remains classified as rural, which widens the observed gap. Because the two forces push in opposite directions, their net effect on the measured rural–urban gap is a priori ambiguous. In any case, this reflects how rurality is measured rather than a real change in the underlying population, and the infrequency of reclassification limits its influence on the broad patterns documented here.DE LAMATA ET AL. 8 TA B L E 1Descriptive Statistics by Urban/Rural Area. Average Latin America, 2003 and 2023 2003 2023 Urban Rural R/U Urban Rural R/U Panel A. Demographic characteristics Male household head (%) 71.8 81.0 1.1 56.7 67.4 1.2 Average age of household head 47.1 47.7 1.0 50.9 51.8 1.0 Household size 3.9 4.3 1.1 3.2 3.3 1.0 Children under 15 per household 1.2 1.6 1.3 0.7 0.9 1.2 Dependency ratio 0.7 0.9 1.3 0.5 0.6 1.2 Panel B. Human capital Head with secondary education or more (%) 35.7 8.3 0.2 49.6 21.0 0.4 Secondary educ. completed or more, ages 25+ (%) 38.2 9.4 0.2 52.6 23.8 0.5 Average years of schooling, ages 25+ 8.6 4.6 0.5 10.2 6.8 0.7

School attendance, ages 6–17 (%) 89.9 80.9 0.9 94.7 91.6 1.0 Panel C. Living conditions Access to electricity (%) 96.4 69.4 0.7 99.7 95.5 1.0 Access to piped water (%) 88.3 57.1 0.6 94.3 71.0 0.8 Access to sanitation (%) 55.1 5.9 0.1 63.4 10.8 0.2 Overcrowded household (%) 5.1 10.4 2.1 1.5 3.1 2.0 Panel D. Labor market and economic outcomes Employment rate, ages 15+ (%) 50.7 53.6 1.1 54.4 56.0 1.0 Avg. monthly labor income, employed (2017 PPP USD) 848.2 490.7 0.6 1009.2 736.6 0.7 Avg. monthly per-capita income (2017 PPP USD) 457.1 213.8 0.5 655.3 426.9 0.7 Poverty rate (%) 37.3 64.9 1.8 16.8 34.2 2.0 Extreme poverty rate (%) 14.5 42.4 2.9 4.5 14.6 3.3

Notes: Simple averages across 14 Latin American countries. Household-level variables—the household-head demographics in Panel A, the household head’s education in Panel B, and the dwelling conditions in Panel C—are computed over household heads. The remaining variables

(education and attendance for the indicated age groups, employment, labor income, per-capita income, and poverty) are computed at the individual level. R/U denotes the rural-to-urban ratio.

Income variables are expressed in 2017 international dollars using purchasing power parity (PPP) exchange rates. Poverty and extreme poverty are measured using the $6.85 and $3.65 per-person per-day thresholds, respectively (monthly equivalent). Rural and urban classifications follow each country’s