CAF - Geography and Agricultural Productivity -The Case of Latin America and the Caribbean

C A F - W O R KI N G PA P E R # 2 0 2 5 / 0 4

F i r s t v e r s i o n : J u l y 3 1 , 2 0 2 5

Geography and Agricultural Productivity: The Case of Latin America and the Caribbean Tasso Adamopoulos1 1Professor, Department of Economics, York University.aadamo@yorku.ca I quantify the role of current land quality and geographic conditions as well as projected future climate change for agricultural productivity differences across and within Latin American and Caribbean (LAC) countries. I combine geo-spatial data on potential yields by crop and grid-cell, with a spatial accounting framework. If LAC countries produced their crops in the locations they produce them with potential yields rather than actual, the 18 percent aggregate yield deficit relative to the richest countries would be reversed to an 18 percent surplus. While there is considerable cross-country and within-country heterogeneity, overall LAC countries have favourable natural land productivity. With improved input application and cultivation practices most LAC countries can double agricultural productivity, with substantial structural change implications. Climate change will reduce average yields in most LAC countries, but because of its heterogeneous effects across regions, there is more scope for yield gains from the spatial reallocation of production than under current conditions.

K E Y W O R D S

Agriculture, Land Quality, Climate Change, Productivity, Spatial Allocation, Crop Choice, Latin America and Caribbean 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. ©2025 Corporación Andina de FomentoC A F - D O C U M E NT O D E T R AB A J O # 2 0 2 5 / 0 4

P r i m e r a v e r s i ó n : 3 1 d e j u l i o 2 0 2 5

Geografía y productividad agrícola: El caso de América Latina y el Caribe Tasso Adamopoulos1 1Professor, Department of Economics, York University.aadamo@yorku.ca Cuantifico el rol de la calidad actual de la tierra y las condiciones geográficas, así como el cambio climático futuro proyectado, para las diferencias de productividad agrícola entre y dentro de los países de América Latina y el Caribe (ALC). Combino datos geoespaciales sobre rendimientos potenciales por cultivo y celda, con un marco de contabilidad espacial. Si los países de ALC produjeran sus cultivos en los lugares donde los producen pero con rendimientos potenciales en lugar de los reales, el déficit de rendimiento agregado del 18 por ciento en relación con los países más ricos se revertiría a un superávit del 18 por ciento. Si bien existe una considerable heterogeneidad entre países y dentro de cada país, los países de ALC tienen en general una productividad natural de la tierra favorable. Con una mejor aplicación de insumos y prácticas de cultivo, la mayoría de los países de ALC pueden duplicar la productividad agrícola, con importantes implicaciones de cambio estructural. El cambio climático reducirá los rendimientos promedio en la mayoría

de los países de ALC, pero debido a sus efectos heterogéneos entre regiones, hay más margen para ganancias de rendimiento a partir de la reasignación espacial de la producción que en las condiciones actuales.

K E Y W O R D S

Agricultura, Calidad de la Tierra, Cambio Climático, Productividad, Asignación Espacial, Elección de Cultivos, América Latina y el Caribe Pequeñas secciones del texto, menores a dos párrafos, pueden ser citadas sin autorización explícita siempre que se cite el presente documento. Los resultados, interpretaciones y conclusiones expresados en esta publicación son de exclusiva responsabilidad de su(s) autor(es), y de ninguna manera pueden ser atribuidos a CAF, a los miembros de su Directorio Ejecutivo o a los países que ellos representan. CAF no garantiza la exactitud de los datos incluidos en esta publicación y no se hace responsable en ningún aspecto de las consecuencias que resulten de su utilización. ©2025 Corporación Andina de FomentoADAMOPOULOS 2 1 | INTRODUCTION Behind the large differences in real GDP per capita across countries lie substantial differences in economic structures. Lower income countries not only engage a large share of their labor force in agriculture but are also particularly unproductive in producing farm goods relative to advanced economies (Gollin et al., 2002; Restuccia et al., 2008; Caselli, 2005). Understanding the source of the large measurable differences in real agricultural productivity across countries is a fundamental question with important implications for structural change, development, poverty reduction, sustainability and welfare. Broadly speaking, there are two possible explanations for the agricultural productivity disparities across countries. First, they may reflect differences in production possibilities, related to the inherent geography, of land quality, climate, and topography. Geography would matter more in agriculture than any other sector of the economy because farming depends on the local biophysical environment in which it takes place. Second, agricultural productivity disparities may reflect differences in economic choices across countries, resulting from differences in institutions, frictions or constraints that affect either the operated technologies or the level and allocation of inputs. What the source of agricultural productivity differences is, is important and has different implications for policy. Most of the research in economics has focused on the second set of factors, related to economic choices and the economic or institutional environment in which they are made. Adamopoulos and Restuccia (2022) however, examine the role of land quality and geography in accounting for the differences in agricultural yields across countries. In this paper, I study the role of current land quality and geographic conditions in understanding the observed agricultural productivity differences across countries and across sub-national administrative units within countries (states or provinces), with an emphasis on Latin America and the Caribbean (LAC). In addition, I study the implications of spatially granular climate change projections for future potential agricultural productivity differences across Latin American and Caribbean countries. Understanding the determinants of agricultural productivity is particularly important for Latin American and Caribbean countries, given that large populations still earn their living on the rural side of the economy with agriculture accounting for a sizable portion of employment and output (e.g., Bolivia, Ecuador, Guatemala, Honduras, Nicaragua, Peru); many countries are key exporters of major agricultural goods (e.g., Argentina, Brazil, Colombia, Mexico); and, there is remarkable bio-diversity and diverse agro-ecosystems not only across countries, but also across regions within countries (e.g, Amazon basin, Andean regions, Pampas, Yucatán Peninsula).

Understanding the implications of climate change is particularly important, given that the challenges of food security and sustainable growth are emerging as a key policy priorities worldwide. Following the methodological approach of Adamopoulos and Restuccia (2022), I combine gridded micro-geography data at the pixel level from the Global Agro-Ecological Zones GAEZ v3.0 (GAEZv3.0, 2012) project of the Food and Agricultural Organization (FAO) with a spatial accounting framework that decomposes aggregate potential yield gains into a production component, a spatial component and a crop composition component. In addition to the geographic focus on Latin American and Caribbean countries, my analysis extends the work of Adamopoulos and Restuccia (2022) by: (a) conducting sub-national analysis, across states/provinces, within all the major Latin American economies; (b) estimating the effect of climate change scenarios on aggregate potential yields across these economies. GAEZ assembles spatially detailed land quality, climate and terrain topography attributes at the 5-arc minute resolution. Picture a grid covering the entire globe where the size of a cell in the grid is about 10 × 10 kilometers. GAEZ combines these detailed cell-ADAMOPOULOS 3 specific geographic conditions with well established state-of-the-art crop-specific agronomic models for all major crops to estimate the maximum potential amount of output per unit of land that can be produced in that cell for that crop, what is called a potential yield. The GAEZ database provides crop-cell potential yields for different levels of cultivation inputs and water conditions for all cells on earth. I use two potential yields for current geographic conditions, the low inputscenario and the mixed inputscenario. The low input scenario assumes the lowest level of intermediate inputs and cultivation practices and assumes rainfed water supply conditions, capturing closest the natural suitability of the land in a cell

for the production of a particular crop. The mixed input scenario assumes an intermediate input application and targeted cultivation practices by land quality with both irrigated and rainfed water supply conditions, what GAEZ considers a reasonable approximation of current production practices. In order to study the role of climate change I use potential yield projections from GAEZ v4.0 (GAEZv4.0, 2021), which make use of the most recent climate change scenarios predicted by climatologists, accounting for their impact on crop-cell level production conditions. I examine the predictions under two climate models for the 2050s, which involve moderate temperature increases and carbon fertilization effects. The spatial accounting framework aggregates yields for all crops across all cells within an administrative unit. The aggregate actual yield, i.e., the total value of output per unit of harvested land, can be expressed as a weighted average across all crops and cells, where the weights are the shares of land allocated to a crop and cell. This expression can be used to construct counterfactual yields. My main counterfactual estimates the aggregate production potential yield, whereby the same crops are produced in the same locations but with the potential crop-cell yields rather than their actual ones, keeping the allocation of land across crops and cells to the observed one. This production potential yield can be thought of as an aggregate summary measure of the natural productivity of the land, given the current land quality and geography of the administrative unit, and the current actual distribution of production across space and crops. In terms of actual productivity, LAC countries are closer to the richest countries in the world, than the poorest. On average, the aggregate actual yield of LAC countries is at 82 percent of that of the highest income countries. However, there is considerable variation across LAC, with the actual yield of countries like Costa Rica, Chile, Colombia, Jamaica, Peru exceeding that of rich countries, with most somewhat below, and countries

like Bolivia, El Salvador, Honduras and Nicaragua well below it (although still above the poorest economies of the world). Using the low input scenario, closest to the natural suitability of the land, the production potential yield for LAC countries is on average higher than that of both the richest (by 18 percent) and the most agriculturally productive countries (by over 30 percent) in the world. While there is heterogeneity in potential yields across LAC countries, most exceed the potential yields of the wealthier countries, especially Argentina, Dominican Republic, Guatemala, Portugal and Uruguay. These findings imply that for LAC countries, as a whole, the actual productivity deficit relative to the most productive countries is not due to natural endowment limitations. Appropriately exploiting existing biophysical conditions and geography, LAC countries have the potential to surpass the most productive countries in agricultural productivity. The variation of actual yields across LAC countries is completely uncorrelated with the variation in production potential yields. This implies that the variation in observed agricultural productivity across LAC countries is not driven by variation in land quality and geography. In relative terms, Costa Rica and Colombia produce the closest to their production potential in comparison to the rich group of countries. The findings are robust to including coffee, cocoa and tea in the analysis. The country-level findings mask the sub-national heterogeneity within countries. IADAMOPOULOS 4 assemble maps with variation in the production potential yield (low input scenario) at the cell level, indicating the intensity of the potential yield across all of LAC. Further, I estimate the production potential at the first administrative sub-national level, i.e., for states or provinces, for all major LAC countries. Overall, I find that within countries there are provinces/states with high production potential in levels, that exceed those of the country as a whole. However, there is often a weak correlation between the provinces/states where agricultural production takes place and the potential yield of those regions, with exceptions

such as Argentina. While the correlation of actual yields and production potential yields across provinces/states within countries varies, for all sub-national administrative units taken together across LAC countries the correlation is 0, just like at the cross-country level. This mis-alignment of actual production and potential yields can explain why countries are not fully taking advantage of their natural endowments. I also examine the role that inputs play in projecting land quality and geography into actual agricultural productivity. I estimate production potential yields under the mixed input scenario that assumes an intermediate application of inputs and use of cultivation practices, as well as irrigation. I find that the production potential yields under the mixed input scenario are more positively correlated with the actual yields across LAC countries, with Antigua and Barbuda being an outlier. The closer alignment of actual and potential yields across LAC indicates that the actual suitability of the land when appropriately interacted with inputs can partly account for the variation of agricultural productivity across countries. The production potential yields under the mixed input scenario are higher in levels than the actual yields for all countries. The yield gap, the ratio between the potential yield and the actual yield, is indicative of the aggregate potential yield gains that countries can achieve with more intensive input application given their internal distribution of land quality and geography, and their current allocation of harvested land across space and crops. This yield gap varies considerably across countries, but is smaller for all LAC countries relative to the less developed economies in the world. The largest potential for yield gains are for Bolivia, Honduras, Portugal (yield gap of over 3). The economies with the lowest potential yield gains, implying that they take advantage of their geography the most, are Brazil, Colombia, Costa Rica, Jamaica (yield gap of under 2). I show that these potential yield gains have substantial aggregate implications in terms of structural change and development, particularly for lower income countries. Using the spatial accounting framework and the mixed input potential yields at the cropcell level I also estimate the potential yield gains from the spatial reallocation of production and the change in the set of crops. I implement this through two counterfactuals. The first counterfactual estimates the spatial potentialyield by reallocating production across cells within countries to maximize aggregate output, keeping the total amount of land allocated to each crop to the actual one in the data. The second counterfactual estimates the total potential yield by producing the highest value yielding crop in each cell, effectively allowing for not only a spatial reallocation of production but also a change in the composition of produced crops. I find additional gains relative to the production potential yield for all countries, particularly from the total potential. The largest gains from the spatial reallocation relative to the production potential would be experienced by Portugal (33 percent), Spain (35 percent), Ecuador (27 percent), and Peru (25 percent). The largest gains from the total potential would be experienced by Chile, El Salvador, and Uruguay. Given the increased environmental challenges associated with extreme weather phenomena and climate change, in this paper, I also examine the effect of climate change projections by climatologists on cell-by-cell and crop-by-crop potential yields, and their implications for aggregate future potential yields, with a focus on LAC countries. Under moderate climate change scenarios, I find that aggregate production potential yields in the future are expectedADAMOPOULOS 5 to go down for most (although not all) LAC countries, relative to the potential yields with similar conditions today (mixed input scenario). However, the impact of climate change is highly heterogeneous across cells within countries, with some cells facing challenges for the cultivation of some crops while others presented with opportunities for higher yields. Taking into account the heterogeneous effect of climate change I find that there is more scope for aggregate yield benefits from the spatial reallocation of crop production across cells within countries in the future than under current climate conditions: on average

across LAC countries, about 30 percent gains in the future relative to half those gains today. The potential gains from changes in the composition of crops produced (total production potential) are at 70 percent, lower than with current conditions which are at 105 percent. This paper contributes to a large literature in macroeconomics and development trying to understand the role of agriculture for structural change and the process of economic development (Kuznets, 1966; Gollin et al., 2007; Caselli, 2005; Restuccia et al., 2008). An important question in this literature is understanding what accounts for the large productivity differences in agriculture across countries and between agriculture and non-agriculture within countries. The vast majority of the macro-development literature on the determinants of agricultural productivity has focused on the role of several frictions, institutions, policies constraining economic choices in agriculture, both across countries as well as within countries: intermediate inputs (Restuccia et al., 2008); farm size (Adamopoulos and Restuccia, 2014); spatial transport connectivity (Sotelo, 2020; Adamopoulos, 2025); land misallocation (Adamopoulos et al., 2022); selection (Lagakos and Waugh, 2013); idiosyncratic agricultural risk Donovan (2021); trade risk (Adamopoulos and Leibovici, 2024); capital intensity (Chen, 2020), among many others. None of this literature however has considered the role of land quality. As in Adamopoulos and Restuccia (2022), I study the role of current land quality conditions and geography for agricultural productivity using a spatial accounting framework and high-resolution geospatial data. I add to this by focusing on agricultural productivity differences across LAC countries, and extend the work by studying the role of

land quality and geography at the sub-national level (across states/provinces) within LAC countries, and by considering the effect of climate change on potential future agricultural productivity. There is an economics and agronomic literature that studies the role of individual geographic attributes, such as temperature, rainfall, soil quality, and topography on crop yields (Levine and Yang, 2014; Schlenker and Roberts, 2009; Cassman, 1999; Kravchenko and Bullock, 2000). Instead, the analysis here accounts for all geographic attributes that impact the biological growth of crops, summarized through potential yields. There is an earlier literature on cross-country growth analyses that use aggregate land quality indices or geography measures (Gallup et al., 1999; Sachs, 2003; Wiebe, 2003). In contrast here, I utilize the explicit spatial nature of the micro-geography data in GAEZ using an accounting framework, to aggregate up to various administrative levels, namely the country and sub-national levels. It is well documented that temperature and climate change are associated with economic outcomes, at the micro, macro and spatial level (Nordhaus, 2006; Dell et al., 2009, 2012; Hsiang et al., 2017; Somanathan et al., 2021; Desmet and Rossi-Hansberg, 2024). There is however no other sector that is affected more directly by climate change than the agricultural sector (Cruz, 2024). Rising temperatures reduce crop yields, especially beyond some threshold, giving rise to non-linear effects (Schlenker and Roberts, 2009; Calzadilla et al., 2013; Burke et al., 2015; Zhao et al., 2017). To cope with climate change in agriculture the literature has examined several adaptation and mitigation strategies: adjusting agricultural

inputs (Jagnani et al., 2021); changing patterns of production and trade (Costinot et al., 2016); sectoral reallocation (Nath, 2024); internal migration (Gröger and Zylberberg, 2016);ADAMOPOULOS 6 long-run adaptation (Burke and Emerick, 2016), among others. Here I examine how the high-resolution spatial effects of climate change are expected to affect aggregate agricultural productivity. In addition, taking into account the effect of climate change on future yields by crops and cell, I show that reshuffling agricultural production across space (within countries) and shifting farming to less affected crops, can mitigate the impact of global warming in LAC countries. The rest of the paper proceeds as follows. In Section 2, I briefly examine the sectoral structure of LAC countries and describe the gridded data I use in the analysis. In Section 3, I present the spatial accounting framework. Section 4 presents the aggregate potential yields across countries. Section 5 covers the sub-national analysis across LAC countries, and Section 6 examines the role of climate change on future aggregate yields. I conclude in Section 7. 2 | DATA I first examine the sectoral structures and incomes of Latin American and Caribbean economies, along with Portugal and Spain (LAC for short) and then present the spatial micro-geography data. 2.1 | Sectoral Composition: LAC Countries Before considering the role of geography and land quality for agricultural productivity across LAC countries, I report data for the sectoral composition of their economies and their relative aggregate income. Table 1 displays the share of employment in agriculture (first column), the share of agricultural value added in GDP (second column), and real GDP per capita, PPP in constant 2021 international $ (third column). All data are from the World Development Indicatorsof the World Bank for the year 2010.1 I report the data by LAC country,

their average as well as for Canada, the United States and the top (20 percent richest) and bottom (20 percent poorest) quintiles of the world real GDP per capita distribution. The fourth column reports the ratio of each country’s real GDP per capita relative to the 20 percent richest countries. On average, LAC countries are only at 29 percent of the richest countries in terms of GDP per capita, which is roughly in the middle of the distribution, and well below the other countries in North America. This average confounds the variation across LAC countries, with Portugal and Spain being at 60 and 70 percent of the richest countries, while Bolivia, Honduras and Nicaragua at about 10 percent. The share of employment in agriculture, a measure of the extent of structural change in a country, varies significantly across LAC countries from 30 percent and above in Bolivia, Guatemala, Honduras, Nicaragua to 4 and 6 percent in Spain and Argentina respectively. To put these numbers into perspective, in high income countries like Canada and the United States the agricultural employment share is under 2 percent while in the poorest countries in the world this is over 60. Similar conclusions are drawn by looking at the output share of agriculture. Most major LAC countries, in terms of their sectoral structure, are in the fourth quintile of the income distribution. 1Due to missing or inaccurate data the agricultural employment and output shares for Argentina are from the Groningen Growth and Development Centre’s Economic Transformation Database(de Vries et al., 2021) for the year 2010, and Antigua and Barbuda’s agricultural employment share is from the Food and Agricultural Organization for the year 2013 (FAO, 2015).ADAMOPOULOS 7 TA B L E 1 Role of Agriculture and GDP per capita Agr. Empl. Agr. VA Real GDP pc Real GDP pc Share (%) Share (%) (PPP , int $) Rel. to Rich 20%

North America Canada 1.8 1.5 52474.6 0.86 Mexico 14.6 3.1 20052.6 0.33 United States 1.7 1.0 59799.3 0.98 Central America Belize 18.5 8.8 11868.4 0.19 Costa Rica 12.4 6.5 18668.3 0.30 El Salvador 20.8 7.0 8648.9 0.14 Guatemala 33.5 11.2 9828.4 0.16 Honduras 36.5 11.6 5396.1 0.09 Nicaragua 29.6 17.0 5654.4 0.09 Panama 17.4 3.6 22739.1 0.37 Caribbean Antigua and Barbuda 21.0 1.2 25831.8 0.42 Bahamas 3.3 1.1 32702.0 0.53 Cuba 18.6 3.6 Dominican Republic 12.4 6.1 14853.6 0.24 Jamaica 17.7 5.3 9654.1 0.16 South America Argentina 6.0 7.8 28056.3 0.46 Bolivia 31.7 10.4 7571.7 0.12 Brazil 11.5 4.1 18062.2 0.29 Chile 10.1 3.5 24148.3 0.39 Colombia 18.5 6.3 14012.0 0.23 Ecuador 27.9 8.3 11651.3 0.19

Guyana 21.4 28.5 10374.4 0.17 Paraguay 25.6 13.0 12738.6 0.21 Peru 28.0 6.8 12061.2 0.20 Uruguay 11.6 7.2 24960.7 0.41 Venezuela 8.5 5.4 Europe Portugal 11.2 1.9 36670.0 0.60 Spain 4.2 2.4 42657.0 0.70 LAC countries 18.2 7.4 17869.2 0.29 Rich 20% of all countries 3.2 2.0 61312.2 1.00 Poor 20% of all countries 58.3 25.7 2596.0 0.04

Notes: Data from the World Bank’s World Development Indicatorsfor the year 2010, except for Argentina’s agriculture shares (GGDC) and Antigua and Barbuda’s agricultural employment share (FAO).ADAMOPOULOS 8 2.2 | Global Agro-Ecological Zones (GAEZ) Data I use spatial micro-geography data from the Global Agro-Ecological Zones project, GAEZ v3.0 (GAEZv3.0, 2012), of the Food and Agricultural Organization (FAO) and the International Institute for Applied Systems Analysis (IIASA). GAEZ provides gridded datasets on a rich set of current land quality and geographic characteristics at the 5-arc minute resolution.

The spatial unit of observation is a cell or pixel in this grid that is roughly 10 by 10 kilometers, with the mapping from arc minutes to square kilometers depending on the latitude. GAEZ offers a cell-by-cell characterization of the conditions relevant for agricultural production in

terms of: a) soil attributes (e.g., depth, fertility, drainage, texture, chemical composition); b) climate attributes (e.g., temperature, sunshine hours, precipitation, humidity, and wind speed); c) terrain attributes (e.g., elevation, slope). This information is used to identify local biophysical constraints and limitations across cells. GAEZ feeds the cell-specific information on geographic characteristics into well-established state-of-the-art agronomic models for each crop, that account for science-based biophysical growing requirements for each crop. A key output of this procedure is the estimation of a potential yield for each crop for each cell, which captures the highest output per hectare that can be attained in the cell given: the crop’s growing requirements; the cell’s characteristics; and assumptions about water supply conditions and cultivation practices. In sum, potential yields encapsulate the importance of location-specific land quality and geography for the production of particular crops. The crop-specific agronomic model parameters are based on well tested field and lab experiments by agricultural research institutes, reflecting the latest state of scientific knowledge (rather than estimated from reduced-form regressions). The estimated potential yields, for the year 2000, are reported for different water supply conditions (irrigated, rainfed, total) and type of cultivation practices (low, intermediate, high, mixed).2 I focus my analysis on two scenarios: (a) the low inputscenario, with rainfed water supply conditions and the low level of cultivation practices, meant to capture the natural suitability of the land for the production of different crops; (b) the mixed inputscenario that assumes total water supply conditions and mixed level of inputs, which assumes high inputs on the best land, intermediate inputs on moderately suitable land, and low inputs on marginal land, a scenario GAEZ considers a reasonable representation of current cultivation conditions. I use potential yields for baseline historical climate conditions (1960-1990). In

the main analysis, I focus on 18 major crops and commodity groups, 3 and examine the robustness to including cocoa, tea and coffee, which are high value cash crops. The GAEZ database also provides at the 5 arc-minute resolution, for the year 2000, data on crop choice, actual production, actual area cultivated, and actual yield, i.e., tonnes of production per hectare of the crop actually planted. The actual production data for each cell are estimated using a flexible iterative rebalancing methodology that sequential