WORLD BANK LATIN AMERICAN AND CARIBBEAN STUDIES Going Viral COVID-19 AND THE ACCELERATED TRANSFORMATION OF JOBS IN LATIN AMERICA AND THE CARIBBEAN Guillermo Beylis, Roberto Fattal Jaef, Michael Morris, Ashwini Rekha Sebastian, and Rishabh Sinha Going Viral L AT I N A M E R I C A A N D C A R I B B E A N S T U D I E S Going Viral: COVID-19 and the Accelerated Transformation of Jobs in Latin America and the Caribbean Guillermo Beylis, Roberto Fattal Jaef, Michael Morris, Ashwini Rekha Sebastian, and Rishabh Sinha © 2020 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington, DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org Some rights reserved 1 2 3 4 23 22 21 20 This work is a product of the staff of The World Bank with external contributions. The findings, inter- pretations, 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 of the data included in this work. 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 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 This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) http://creativecommons.org/licenses/by/3.0/igo. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following conditions: Attribution —Please cite the work as follows: Beylis, Guillermo, Roberto Fattal Jaef, Michael Morris, Ashwini Rekha Sebastian, and Rishabh Sinha. 2020. Going Viral: COVID-19 and the Accelerated Transformation of Jobs in Latin America and the Caribbean. Latin America and Caribbean Stud- ies. Washington, DC: World Bank. doi:10.1596/978-1-4648-1448-8. License: Creative Commons Attribution CC BY 3.0 IGO. Translations —If you create a translation of this work, please add the following disclaimer along with the attribution: This translation was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. Adaptations —If you create an adaptation of this work, please add the following disclaimer along with the attribution: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank. Third-party content—The World Bank does not necessarily own each component of the content con- tained within the work. The World Bank therefore does not warrant that the use of any third-par- ty-owned individual component or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting from such infringement rests solely with you. If you wish to reuse a component of the work, it is your responsibility to determine whether permission is needed for that reuse and to obtain permission from the copyright owner. Examples of components can include, but are not limited to, tables, figures, or images. All queries on rights and licenses should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; e-mail: pubrights@worldbank.org. ISBN (paper): 978-1-4648-1448-8 ISBN (electronic): 978-1-4648-1460-0 DOI: 10.1596/978-1-4648-1448-8 Cover image: © Alfred Pasieka / Science Photo Library. Used with the permission of Alfred Pasieka / Science Photo Library. Further permission required for reuse. Cover design: Bill Pragluski, Critical Stages, LLC. Library of Congress Control Number: 2020935042. Contents Foreword ix Acknowledgments xi About the Authors xiii Abbreviations xv Introduction......................................................................................................................................................1 Analyzing structural transformation................................................................................................ 2 Implications for economic policy..................................................................................................... 4 Notes............................................................................................................................................... 5 References....................................................................................................................................... 5 1 What is structural transformation?..........................................................................................................7 What drives structural transformation?......................................................................................... 10 Conclusions................................................................................................................................... 21 Notes............................................................................................................................................. 22 References..................................................................................................................................... 23 2 Productivity in the LAC region: A sectoral view................................................................................... 25 Productivity in agriculture............................................................................................................. 27 Productivity in industry and services............................................................................................. 38 Taking stock: The scope for raising allocative efficiency and the expected pace of structural change....................................................................................................................... 42 Conclusions and policy implications.............................................................................................. 45 Notes............................................................................................................................................. 48 References .................................................................................................................................... 48 v vi   C o n t e n t s 3 Economic transformation, skills, and the future of work.................................................................... 53 The labor market is already changing............................................................................................ 54 Looking into the future: Automation, tasks, and skills................................................................... 68 Looking into the future: Digital platforms and the nature of work................................................ 81 Conclusions and policy implications.............................................................................................. 82 Notes............................................................................................................................................. 84 References..................................................................................................................................... 85 4 Conclusions ............................................................................................................................................ 87 Structural transformation: Past and future.................................................................................... 87 Looking forward........................................................................................................................... 89 References..................................................................................................................................... 91 Boxes 1.1 Measuring structural transformation.............................................................................................. 8 2.1 Does technological change benefit small and large farms equally? Evidence from Mexico........... 33 3.1 What are workers doing?.............................................................................................................. 60 3.2 When automation creates jobs...................................................................................................... 70 Figure 1.1 Structural transformation by sector, selected LAC countries and rest of world............................... 9 1.2 Patterns of industrialization across LAC and high-income countries............................................ 13 1.3 Value-added and employment shares by sector: Selected LAC countries, 1950–2010.................. 14 1.4 Absolute total level of employment by sector: Selected LAC countries, 1950–2010..................... 17 1.5 Premature deindustrialization: LAC region (average), 1950–2010............................................... 18 1.6 Relative prices and real consumption per capita: LAC region (average), 1950–2010................... 19 1.7 Labor allocation in manufacturing: Selected LAC countries, 1950–2010..................................... 20 2.1 Output per worker by sector in LAC region relative to that of United States: Selected countries, 2010................................................................................................................ 26 2.2 Agricultural output and TFP growth: LAC region, 1981–2014.................................................... 28 2.3 Correlation between output growth and TFP growth: LAC countries, 2001–14.......................... 28 2.4 Growth decomposition: Latin America by region and United States, 2005–14............................ 29 2.5 Relationship between value added and employment in agriculture: Selected LAC countries, 2017..................................................................................................................... 30 2.6 Sources of agricultural productivity growth.................................................................................. 30 2.7 Histogram of metatechnical efficiency, Peru by region.................................................................. 31 2.8 Effectiveness of growth in different sectors at reducing poverty................................................... 38 2.9 Labor productivity growth in industrial and services sectors: Latin America and United States, 1950–2010............................................................................................................. 39 2.10 Labor productivity in services sector relative to industrial sector: Latin America and United States, 1950–2010...................................................................................................... 40 2.11 Services Trade Restrictions Index, selected LAC countries............................................................ 47 2.12 Logistics Performance Index and its components: 16 LAC countries relative to best performer.........................................................................................................48 C o n t e n t s   vii 3.1 Development of goods and service occupations, LAC and rest of world...................................... 56 3.2 Evolution of task content of jobs (mean change): 11 LAC countries, 2000–2014........................ 62 3.3 Evolution of task content of jobs in industrial sector: 11 LAC countries, 2000–2014.................. 64 3.4 Decomposition of task content in industrial sector: 11 LAC countries, 2000–2014..................... 65 3.5 Evolution of task content of jobs in services sector: 11 LAC countries, 2000–2014..................... 67 3.6 Distribution of automatability across methodologies, Chile......................................................... 76 3.7 Distribution of automatability across methodologies, Colombia.................................................. 77 3.8 Distribution of automatability across methodologies, Bolivia....................................................... 78 3.9 Risk of automation by LAC country, based on four methodologies............................................. 79 3.10 Automation risk by selected characteristics, LAC region.............................................................. 81 Tables 2.1 Misallocation in manufacturing, selected developing and developed countries............................. 41 3.1 Reallocation of occupations within sectors over development process......................................... 57 Foreword Latin America and the Caribbean is the future, and consequently making the policy region most affected by the COVID-19 pan- reforms needed to help create more and bet- demic, with health and economic challenges ter jobs even more urgent. as large as those in advanced nations but More vibrant job creation requires over- without the necessary resources to protect coming the region’s chronically low levels employment and sustain economic activity. It of productivity growth. This will demand is a complex and painful scenario in which Investments in smart infrastructure, the millions are suffering through the huge daily adoption of new technologies, the promo- challenges facing the region and the devastat- tion of competition and product upgrading, ing consequences on their jobs and earnings. and the removal of market distortions that In addition to the new challenges that impede the growth of the most productive the pandemic and the policy response to it firms. Also, the region would benefit from have wrought, the current crisis has, sadly, increasing international trade not only in exposed and deepened some of the old prob- goods but, perhaps more importantly, in ser- lems the region was already facing. A seg- vices. The enormous potential of Latin Amer- mented labor market and social protection ica and the Caribbean will only materialize if system have been able to protect the jobs and the right policies are put in place. earnings of formal workers while leaving At the same time, the region needs to many informal workers unprotected and fac- invest in the human capital of its workforce. ing the dire choice of confronting health risks The jobs of the future will require a very or being unable to sustain their families. different skillset, especially when compar- In Going Viral: COVID -19 and the ing with the many informal jobs available Accelerated Transformation of Jobs in Latin at present. Countries have to prepare their America and the Caribbean, the authors dig children and teenagers by investing now in into the underlying trends that were trans- schools and universities, and by improving forming the labor market even before the the learning content of education. But coun- pandemic. Unfortunately, the current eco- tries also need to adopt retraining and job nomic crisis associated has only accelerated placement programs for the adults that have these trends, bringing the region nearer to the seen their jobs disappear. ix x   F o r e w o r d Finally, the region must rethink its labor the context of shifting sectoral employment regulations and social protection systems for and technological evolution. The huge eco- them to promote the creation of jobs while nomic and social costs created by the pan- encouraging the formalization of workers. demic have accelerated the transformation Already plagued with high informality and of jobs and make the challenge more urgent with trends indicating a future of work with than ever. But, inclusion through better more freelancing and independent workers, jobs is unavoidable if we want more equal new regulations must not only help create societies. That will be the key measure of new jobs but also expand the benefits of success. social protection to larger segments of the workforce. Carlos Felipe Jaramillo Perhaps one of the greatest challenges Regional Vice President for Latin America for Latin America and the Caribbean will and the Caribbean be the creation of new and better jobs in World Bank Acknowledgments This book was prepared by a team led by T he team was for t u nate to receive Guillermo Beylis. The core team also con- ­ xc el lent adv ic e a nd g u id a nc e f rom e sisted of Roberto Fattal Jaef, Michael three distinguished peer reviewers: Jorge Morris, Ashwini Rekha Sebastian, and A raujo, E rnesto L opez- Cordova, and Rishabh Sinha. The team received excellent Richard R ­ ogerson. Although the team is research assistance from Julian Eduardo very grateful for the guidance received, Diaz Gutierrez and Maria Ignacia Paz these reviewers are not responsible for Cuevas de Saint Pierre. The work was con- any remaining errors, omissions, or inter- ducted under the general guidance of Carlos pretations. Additional insights from Rita Vegh, former Chief Economist for the Latin A l meida, Sa muel Pien k nag u ra, M a rc America and the Caribbean (LAC) region of Schiffbauer, Francisco Carneiro, Oscar the World Bank, and Martin Rama, current Calvo-Gonzalez, and other ­ p articipants Chief Economist for the LAC region, with in a workshop held on April 3, 2018, are substantial inputs from Daniel Lederman, gratefully acknowledged. former LAC Deputy Chief Economist and Sabra Ledent was the editor. Patricia Elena Ianchovichina, current LAC Deputy K atayama (acquisitions editor), Mary Fisk ­ Chief Economist. (production editor), and Orlando Mota (print Background papers were prepared by coordinator) of the World Bank’s Formal Guillermo Beylis, Julian Eduardo Diaz Publishing Program were responsible for Gutierrez, Roberto Fattal Jaef, Steven Helfand, managing the editing, design, typesetting, Maria Ignacia Paz Cuevas de Saint Pierre, and and printing of the book. Last, but not Rishabh Sinha. We are very grateful for their ­ least, the authors thank Ruth Eunice Flores original and outstanding contributions, as well and Jacqueline Larrabu re for superb as the many insightful conversations with them. administrative support. xi About the Authors Guillermo Beylis is a research economist Caribbean region. He conducts research and in the Office of the Chief Economist, Latin is involved in the preparation, implementa- America and the Caribbean, at the World tion, and evaluation of lending operations. Bank. He specializes in labor markets, His areas of expertise include agricultural with a focus on skills, gender, and inequal- policy, marketing systems and value chain ity. He has published on many different development, and agricultural innovation ­ topics, including energy, international capital systems. Before joining the World Bank, he flows, inequality, and skills. He holds an MA spent 16 years with the Mexico-based Inter- and PhD in economics from the University national Maize and Wheat Improvement of California, Los Angeles (UCLA), and a Center (CIMMYT). He holds a BA from BA and MA from the University Torcuato di Amherst College and an MSc and a PhD Tella, Buenos Aires, Argentina. from Michigan State University. Roberto Fattal Jaef is an economist with Ashwini Rekha Sebastian is an econo- the Macroeconomics and Growth Team of mist at the World Bank, currently mapped the World Bank’s Development Research to the Latin America and the Caribbean Group. His research interests include vari- region. There she works on themes related ous areas of macroeconomics, with a special to agriculture and food systems, environ- emphasis on economic growth. He has pub- mental conservation, rural livelihoods, jobs, lished in leading journals such as the Amer- and poverty reduction. Prior to joining the ican Economic Journal: Macroeconomics, World Bank, she worked as an economist Journal of Development Economics, and at the United Nations Food and Agricul- Journal of International Economics. Prior ture Organization’s Economic and Social to joining the World Bank, he worked in Development Division on the Protection the International Monetary Fund’s Research to Production (PtoP) team. She previously Department. He holds a PhD in economics collaborated with the International Food from UCLA. Policy Research Institute, including on Michael Morris is a lead agriculture research related to environmental migration economist with the World Bank, currently and labor market integration. Ashwini holds mapped to the Latin America and the a BA in economics and mathematics from xiii xiv   About the Authors Bryn Mawr College, an MSc and PhD in include issues involving structural trans- agricultural and natural resource economics formation, occupational choice, financial and an MSc in economics from the Univer- development, intergenerational mobility, sity of Maryland, College Park and fragile and violent economies. Before Rishabh Sinha is an economist with the joining the World Bank in 2015, Rishabh Macroeconomics and Growth Team of worked at the Federal Reserve Bank of the World Bank’s Development Research Kansas City and in the private financial Group. His interests lie in understanding sector. He holds an MS in economics from the role of allocative efficiency in deliv- the Indian Statistical Institute (Kolkata) ering economic growth. He has studied and an MS and a PhD in economics from this relationship in diverse settings, which Arizona State University. Abbreviations AI artificial intelligence ATM automated teller machine CEDLAS Center for Distributive, Labor and Social Studies (Universidad de la Plata in Argentina) DOT Dictionary of Occupational Titles ENHRUM Mexico National Rural Household Survey FDI foreign direct investment GDP gross domestic product GGDC Groningen Growth and Development Centre GVC global value chain ICT information and communications technology IoT Internet of Things ISCO International Standard Classification of Occupations LAC Latin America and the Caribbean MxFLS Mexican Family Life Survey NR-CA nonroutine cognitive analytical NR-CP nonroutine cognitive interpersonal NR-MP nonroutine manual physical OECD Organisation for Economic Co-operation and Development O*NET Occupational Information Network PIACC Programme for the International Assessment of Adult Competencies PPP purchasing power parity xv xvi   A b b r e v i a t i o n s PtoP Protection to Production R&D research and development RBTC routine-biased technological change RC routine cognitive RoW rest of world RM routine manual SBTC skill-biased technological change SEDLAC Center for Distributive, Labor and Social Studies (Universidad de la Plata in Argentina) SOC Standard Occupational Classification STEP Skills Toward Employability and Productivity TFP total factor productivity Introduction C OVID-19 started as a health emer- by type of job. Formal sector workers, who gency, but it is rapidly evolving into get a paycheck at the end of the month, can an employment crisis. The year 2020 be more easily reached by social protection could well witness the biggest contraction in programs than informal sector workers, who economic activity that the region has expe- make their living on a daily basis. Some of rienced since the Great Depression. Lower these differential effects may recede as the external demand; a protracted period of epidemic is contained, but others may have quarantines and lockdowns; short-term long-lasting effects. liquidity constraints evolving into solvency These changes in the level and compo- problems for firms; and in some cases, finan- sition of employment are taking place in a cial crises are undermining the demand for region that was already undergoing a sig- labor and putting an increasingly large num- nificant transformation of its labor markets. ber of jobs at risk. The limited fiscal space For a long time, the hope was that Latin enjoyed by many countries in the region also America and the Caribbean would become makes it difficult for governments to support a more industrial region, one where wage economic activity. There is still uncertainty employment and formality would be increas- about how severe the economic impact of the ingly prevalent. Instead, the actual trends pandemic will be. However, the drag on the have included premature deindustrialization, region’s employment could last longer than a plateauing level of formality, and a steady the epidemic itself. growth of independent work. The COVID-19 crisis is affecting not only Globalization and technology lagged the level of employment but also its compo- behind this transformation in the employ- sition. Different sectors of activity have been ment structure of the region, leading to the impacted by quarantines and lockdowns to expectation that the trends would continue different extents. Services that could be deliv- over time. The COVID-19 crisis, however, ered electronically have held well, or even could actually accelerate them, bringing thrived, while sectors that require human the future much closer than anticipated and contact to be provided have struggled the possibly calling for new and better-adapted most. The consequences have also differed economic policies. Understanding what to 1 2   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n expect, and how to react, requires a deep the importance of agriculture in the economy grasp of the underlying trends and the ways declines with the level of development, both in in which they may be amplified in the com- terms of employment and value-added shares; ing years. This report analyzes how the eco- it declines with the level of development while nomic structure of the region has evolved in that of the service sector increases. Perhaps recent years and how this transformation is less well-known is the fact that the industrial affecting both productivity growth and the sector follows a “hump shape” or inverted-U nature of jobs. pattern, initially growing at lower levels of Following the so-called Golden Decade GDP per capita during the industrialization (2003–13) of rapid development and strong phase and then declining at higher levels of improvements in social indicators, economic income during the deindustrialization phase. growth had stalled across most of Latin This report focuses on the premature America and the Caribbean. In the few years deindustrialization experienced by Latin preceding the COVID-19 crisis, the external America and the Caribbean. As shown by environment no longer provided tailwinds to Rodrik (2016), developing economies are foster an economic rebound. Foreign direct entering the deindustrialization phase at investment had moderated, trade had slowed lower levels of income per capita, and they amid elevated geopolitical tensions, and com- are achieving lower peaks of industrial modity prices were mostly flat. The region shares relative to developed countries. This needed to find internal sources of growth and is concerning, because in most countries focus on a productivity-enhancing reform the industrial sector has the highest level agenda. That need is even more urgent today, of labor productivity and the highest rate as the region struggles with the consequences of productivity growth. When premature of the pandemic and the dramatic lockdown deindustrialization occurs, labor moves measures that many countries adopted to away from the industrial sector into lower contain it. productivit y g row th sectors — usually Although increasing productivity may services—reducing overall productivity, sound like an abstract concept, it translates with negative consequences for real income in practice into creating more and better jobs. growth and standards of living. 2 Countries In Latin America and the Caribbean, the in Latin America and the Caribbean may year preceding the COVID-19 outbreak was actually be at the forefront of this process. a time of intense social unrest. In a dozen Three notable features emerge from this countries, discontent led to violence, leaving analysis. 3 First, there is substantial hetero- large numbers of people dead or wounded. geneity across the countries in our sample. It is difficult to attribute this unrest to any The more developed economies, Argentina single factor, whether economic, social, or and Chile, have been deindustrializing for institutional. Political circumstances specific decades. Countries such as Brazil, Colom- to each country certainly played a role. Yet bia, and Mexico display stagnant or slight a frustration over unmet expectations seems increases in their industrial employment to cut across many of the episodes. In this shares; the least developed nation in our context, a disappointing employment perfor- sample, Bolivia, is still industrializing. Sec- mance can only be a cause for concern. ond, the deindustrialization process is more pronounced in terms of employment shares than in value-added shares. Third, prema- Analyzing Structural ture deindustrialization does not necessarily Transformation imply a contraction of the industrial sector; As a starting point to understand the ongoing the absolute number of jobs in the industrial changes in employment, the report focuses sector—as opposed to the share of jobs—has on structural transformation, analyzing its been fairly stable or even growing. drivers and documenting the experience of The story of deindustrialization in Latin the region.1 A well-known stylized fact is that America and the Caribbean is thus not one I n t r o d u c t i o n   3 of factory closures and mass layoffs. Rather, Given this centrality of the services sec- it is a story of a stunted industrialization pro- tor, the report calls attention to the complex cess whereby the industrial sector has been role it plays in relation to productivity, value unable to grow and create jobs over time as it added, and job creation. At the aggregate did in today’s developed economies. In part, level, the service sector displays lower pro- this may be related to the history behind the ductivity growth than the industrial sector. industrialization phase in the region. Most Yet the sector is composed of a very diverse countries started the industrialization pro- set of subsectors that differ significantly in cess under the banner of old-school indus- their productivity levels and growth rates— trial policies of protectionism and subsidies. and even in their use of skilled labor. In many As globalization took hold, this mostly shel- countries, some service subsectors—such as tered industrial sector did not, for the most telecommunications, finance, and logistics— part, successfully integrate into global value are more productive and skill intensive than chains. As a result, the industrial sector was manufacturing and are increasingly sharing unable to grow, limited by the size of domes- pro-development characteristics that were tic and regional markets. once thought of as unique to manufacturing. An important question is whether the Rapid advances in information and com- observed pattern of deindustrialization munications technologies, and their accel- results from distortions and inefficiencies in erated adoption in the aftermath of the the economy, or whether it rather represents COVID-19 crisis, enable the emergence of an efficient reallocation of resources given service sectors that are no longer limited by the circumstances. Answering this question market size. More and more services can be requires defining an efficient benchmark digitally stored, codified, and easily traded against which to assess the observed patterns (Ghani and Kharas 2010). Meanwhile, in the in the data. A standard model of structural years preceding the epidemic outbreak, the transformation is used to this effect. The deregulation of services markets was accompa- results show that the deindustrialization of nied by large inflows of foreign direct invest- the region that started in the 1980s was inef- ment. Therefore, certain service subsectors ficient. The implied output loss was modest, were increasingly resembling the manufactur- but there appear to be significant distortions ing sector, with exposure to trade and capital in the sector, as reflected in a skewed firm flows, allowing for greater competition, tech- size distribution—with firms in the region nology diffusion, and the benefits of scale. remaining small by international standards. Importantly, many of these services are This finding highlights the importance of emerging as key inputs into industrial and revising policies that may be discouraging the agricultural processes, with numerous for- growth of firms and incentivizing informality. ward linkages to other sectors and substantial Confronting the region is a future in potential to improve aggregate productivity. which the service sector will continue to New evidence is emerging pointing to a “ser- grow and be the main source of job creation. vicification” of the manufacturing sector. The emergence of new labor-saving technol- This refers to the phenomenon where manu- ogies in the manufacturing sector will only facturing is increasing the share of services as deepen and accelerate this trend. At the same inputs to the production process (embodied time, the COVID-19 crisis highlights the het- services), as well as offering more sales and erogeneity of this sector, offering encourag- after-sales services that are bundled with the ing growth prospects to activities that can sales of goods (embedded services). operate remotely, while threatening those The traditional perspective of analyzing that depend heavily on personal contact. The and devising policies for each sector inde- latter could still come back if a vaccine or an pendently therefore is becoming increasingly effective treatment is found. However, the obsolete. The analyses in this report show boost for the former could be more perma- that reducing distortions in the intermediate nent, regardless of how the pandemic evolves. market for services could have an important 4   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n impact on the size of the industrial sector. require different and higher-order capabili- If distortions in the service market were ties and skills. reduced to their historical minimum, the employment share of the industrial sector could increase by 2 to 3.5 percentage points. Implications for Economic Policy The occupational structure is also chang- The findings of this report have several ing within broad economic sectors. The important implications for economic policy. importance of service occupations is increas- Some of these implications are related to the ing in all sectors of the economy. This is productivity challenges that Latin America clearly related to the servicification of man- and the Caribbean was already facing after ufacturing phenomenon described, but it the end of the Golden Decade. If anything, extends beyond manufacturing. Market the social unrest that emerged across the competition and new technologies leverage region in 2019 was a warning that restoring the contribution of workers who produce economic growth and fostering the emer- intangible value added, such as research- gence of more and better jobs were urgent ers, marketers, managers, and designers. By priorities. Other policy implications could favoring telework over personal interaction, see their relevance enhanced by the COVID- the COVID-19 epidemic is bound to further 19 crisis. As sectors are impacted in different increase the leverage of this group. ways and working remotely becomes more As machines replace humans in carry- common, governments need to respond in ing out simpler, more routine tasks, and ways that support a smooth transformation the internet replaces personal interaction, of jobs, one that is socially acceptable and workers will have to adapt. They will need that contributes to productivity growth. to learn how to operate through electronic platforms and devote more of their work Promoting Productivity Growth time to the more complex, higher-order tasks that are harder to automate and that com- A first important message of the report is plement the tasks performed by machines. that policy makers should not focus on sec- These rely on cognitive or analytical skills toral size but rather on productivity growth. (such as critical thinking, creativity, and The emergence of new technologies—under problem-solving), as well as interpersonal the banner of the Fourth Industrial Rev- skills (such as teamwork, negotiation, and olution—suggests that opportunities for management). The report shows that even further industrialization (or reindustrializa- before the COVID-19 crisis, there was a fall tion) are likely to be limited in many devel- in the demand for routine manual tasks and oping countries. Requirements in terms of a rise in the demand for non-routine tasks. the skills mix and the use of electronic plat- The trend is bound to accelerate as remote forms will increase, but these changes tend working becomes more prevalent. to be labor-saving. Overall, the industrial The report evaluates the potential number sector could continue contributing positively of jobs that are at high risk of being auto- to aggregate productivity growth and value mated in the region and concludes that fears added but not as much to job creation, espe- of mass “technological unemployment” are cially for unskilled labor. largely unfounded. Estimates vary widely, Rather than focusing on sector-specific however, depending on the methodology policies, it will be increasingly relevant to used. Many occupations will be affected formulate value chain policies that take and transformed by the emerging technol- into account how sectors interact with ogies. Although the overall number of jobs each other. The servicification of economic may not fall dramatically, the trend could activity in general, and of manufactur- be accelerated by the social distancing prac- ing in particular, offers new opportunities tices the COVID-19 epidemic may foster. for growth. Already the largest employer Importantly, these future jobs and tasks will in the region with over 60 percent of the I n t r o d u c t i o n   5 workforce, the services sector is expected geared to wage earners working in the for- to grow even further and play an increas- mal sector. Much of the regulation focused ingly crucial role as an input provider to the on employer-employee relationships, while larger economy. This calls for a comprehen- social protection programs were job-based. sive set of service sector–oriented policies, This architecture led to rigidity and exclu- with an emphasis on the distortions that sion in an environment where many workers prevent competition and innovation from were self-employed or operated at the mar- occurring at a rapid pace. gins of formality. Premature deindustrialization, the increas- ing servicification of the economy, and the Investing in Human Capital growing reliance on electronic platforms Second, as new technologies are developed raise doubts that wage employment will and adopted, and as remote working becomes increase substantially in the coming years. more prevalent, investing in the human cap- At the same time, new technologies make ital of the workforce should be a priority activities and earnings much more visible to for policy makers. It is no exaggeration to the authorities. For example, social security say that education offers the best insurance contributions based on earnings processed against the risks of automation (World Bank through electronic platforms are becoming 2019). It is the low-paid and uneducated increasingly possible. The last pillar of the workers who are performing the simpler, policy agenda implied by this report thus more routine tasks that are at highest risk of concerns the flexible regulation of the emerg- eventually being replaced by machines. The ing forms of work in a way that encourages same is true of the workers in high-contact employment and supports formalization, activities, such as those characterizing the thereby expanding the coverage of social pro- informal sector of the economy. tection to larger segments of the population. In recent decades, countries in Latin America and the Caribbean have made substantial progress in improving access Notes to secondary education, but the quality of 1. See Herrendorf, Rogerson, and Valentinyi education continues to lag behind that of (2013) for a comprehensive review of the advanced nations and developing country literature. peers in East Asia. What may become more 2. This is known in the literature as Baumol’s disease. important as new automation technologies 3. The countries for which available compara- are adopted in the region is adult learning ble data exist for the analysis are Argentina, and retraining programs. It is possible that Bolivia, Brazil, Chile, Colombia, Costa Rica, transformations in the workplace will hap- Mexico, Peru, and República Bolivariana de pen mid-career for many. Workers will need Venezuela. to adapt and adjust, particularly by chang- ing the set of tasks performed at work. To minimize the adjustment costs borne by References workers, governments should have pro- Ghani, E., and H. Kharas. 2010. “The Ser­ v ice grams that help workers upskill and retrain. Revolution.” Brief 54595, World Bank, Wash- ington, DC. Herrendorf, B., Rogerson, R., and A. Valentinyi. Rethinking Labor Regulations and 2013. “Two Perspectives on Preferences and Social Protection Policies Structural Transformation.” American Eco- nomic Review 103 (7): 2752–89. Last but not least, the accelerated transfor- Rodrik, Dani. 2016. “Premature Deindustrializa­ mation of jobs calls for a rethinking of labor tion.” Journal of Economic Growth 21 (1): 1–33. regulations and social protection policies. World Bank. 2019. World Development Report Countries in Latin America and the Carib- 2019: The Changing Nature of Work. Wash- bean developed an institutional architecture ington, DC: World Bank. What is structural transformation? 1 S tructural transformation, a distinctive productivity, this report concentrates on the feature of economic growth, occurs production side measures. when a sustained period of rising The sizable literature documenting the pat- income and living standards coincides with terns of structural transformation in devel- changes in the distribution of economic oped countries has established three stylized activity across three broad sectors of an facts. First, at lower income levels agriculture ­ e conomy—­ a griculture, industry, and ser- accounts for a dominant share of resources vices.1 Structural transformation is of inter- and output. Second, as an economy grows, est to analysts because of its intimate ties to the agriculture sector shrinks in terms of both trends in productivity, regional income con- employment and value-added share (figure 1.1, vergence, labor force participation, urban- panel a), while the other two sectors, indus- ization, business cycles, wage inequality, and trial and services, increase in prominence many other facets of development. These ties (figure 1.1, panels b and c). Initially, both the often open avenues for policy interventions industrial and services sectors expand. How- contending that the existing allocation of ever, unlike services, which continue to grow activity across sectors is inefficient. at higher levels of income, the industrial sector Economic activity at the sectoral level is eventually reaches a peak and then begins con- generally measured through employment tracting (see figure 1.1, panels b and c).2 This shares, value-added shares, and final con- pattern is commonly known as the “hump sumption expenditure shares. Although these shape” of industry. 3 More recent evidence measures are related and broadly display the uncovered by Buera and Kaboski (2012a, same patterns, they are in fact distinct. Both 2012b) suggests that growth in services value employment and value-added shares  refer added accelerates at around $8,100 (1990 to production side measures, whereas final international dollars). Moreover, this acceler- consumption expenditure shares refers to ation coincides with a decrease in the nominal the consumption side. Box 1.1 highlights value-added share for the industrial sector. the main issues associated with the differ- Although the lack of long time series data ent measures of sectoral economic activity. has limited analysis of the structural trans- Because of data availability and a focus on formation of developing countries, there is 7 8   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n BOX 1.1  Measuring structural transformation Structural transformation involves changes The measures relating to production through time in the amount of economic may also contain different information. For activity occurring in sectors. But what is the example, in 1966 Kuznets noted that during best way to measure economic activity? And the early days of the US economy the share does it matter which units are used? of employment in services increased, while Three measures are conventionally used the value-added share remained almost to measure economic activity at the sectoral constant. More recently, Rodrik (2016, 2) level: employment shares, value-added notes that “in the United States manufac- shares, and final consumption expenditure turing industries’ share of total employment shares. Two of these measures—­ employment has steadily fallen since the 1950s, coming shares and value-added shares—are measures down from around a quarter of the work- of production, whereas final consumption force to less than a tenth today. Meanwhile, expenditure shares are a measure of con- manufacturing valued added has remained sumption. Although the three measures are a constant share of gross domestic product sometimes thought to be interchangeable, at constant prices—a testament to differ- some important differences should be noted, entially rapid labor productivity growth particularly for empirical work. As Herren- in this sector.” Both observations point dorf, Rogerson, and Valentinyi (2013) point to the different effects that technological out, even though the measures often display progress can have on the different mea- the same qualitative behavior, the quantita- sures of structural transformation. The rise tive implications can be very different, and of labor-saving automation technologies at times even the qualitative behavior can may deepen this pattern, further reducing differ. employment shares while maintaining or Perhaps the starkest conceptual dis- increasing value-added shares. tinction between these measures is that Beyond conceptual distinctions, each of production versus consumption. This measure presents some additional limita- distinction can be traced back to the tions. Data availability generally drives difference between the concepts of value researchers to measure employment shares added and final output and how national by calculating the number of workers in accounts are constructed. In their exam- each sector. However, employment may ple, Herrendorf, Rogerson, and Valenti- not reflect changes in true labor input. For nyi (2013) illustrate clearly the distinction example, systematic differences in hours between the measures. The entire cost of worked or in human capital per worker a cotton shirt is recorded as a final con- across sectors vary with the level of devel- sumption expenditure of manufacturing opment. Finally, as noted by Herrendorf, because it is a good as opposed to a service. Rogerson, and Valentinyi (2013, 7) “for However, the accounting of value-added the case of value-added and consumption attributes one component to the agriculture expenditure shares, a key issue arises from sector (the cotton used in the shirt), another the need to distinguish between changes in to the industrial sector (the transformation quantities and prices. This is often difficult of cotton into a shirt), and yet another to the empirically because reliable data on relative services sector (the distribution and retail price comparisons across countries are hard services where the shirt was purchased). It to come by. In addition, consumption and follows that both quantities and prices may production need not coincide because of the differ between the value added and the final presence of investment and of imports and expenditure, suggesting there is no reason exports, so that neither measure alone is to expect the implied shares to exhibit sim- sufficient.” ilar behavior. W h a t i s s t r u c t u r a l t r a n s f o r m a t i o n ?    9 FIGURE 1.1  Structural transformation by sector, selected LAC countries and rest of world a. Agriculture Value-added share Employment share 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 6 7 8 9 10 11 12 6 7 8 9 10 11 12 Log of GDP per capita, PPP, 2011 US$ Log of GDP per capita, PPP, 2011 US$ b. Industrial Value-added share Employment share 1.0 0.6 0.8 0.4 0.6 0.4 0.2 0.2 0 0 6 7 8 9 10 11 12 6 7 8 9 10 11 12 Log of GDP per capita, PPP, 2011 US$ Log of GDP per capita, PPP, 2011 US$ c. Services Value-added share Employment share 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 6 7 8 9 10 11 12 6 7 8 9 10 11 12 Log of GDP per capita, PPP, 2011 US$ Log of GDP per capita, PPP, 2011 US$ Rest of world Argentina Bolivia Brazil Chile Colombia Costa Rica Mexico Peru Venezuela, RB Sources: Original calculations for this publication. Value-added and employment data: Groningen Growth and Development Centre (GGDC), 10-Sector Database (Timmer, de Vries, and de Vries 2015); GDP: Penn World Tables (Feenstra, Inklaar, and Timmer 2015). Note: The graphs plot the sectoral value-added shares and sectoral employment shares against the log of GDP per capita (in 2011 US dollars, PPP-adjusted). Data are for nine coun- tries (Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Mexico, Peru, and República Bolivariana de Venezuela) in the LAC region and 31 countries in the rest of the world. Data are plotted for every five years from 1961 to 2011. GDP = gross domestic product; LAC = Latin America and the Caribbean; PPP = purchasing power parity. 10   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n now substantial evidence that the patterns The first mechanism postulates that the of structural transformation evident in income elasticity of demand varies across developed countries broadly hold true for sectoral goods. Put differently, consumers’ developing countries as well. In poorer coun- preferences are nonhomothetic (Kongsamut, tries, larger shares of employment and value Rebelo, and Xie 2001). Thus as an economy added are devoted to the agriculture sector, becomes wealthier, the sectoral allocation whereas lower-middle-income countries are of activity changes in response to changes industrializing, and higher-middle-income induced by shifts in the household expendi- countries are deindustrializing as the growth ture, which moves away from agricultural of their services sector has accelerated. goods as subsistence food requirements are Although the broad patterns of structural fulfilled. In other words, in the early stages transformation in developing economies are of development households spend most of consistent with the experience of developed their budget on food. As countries grow economies, there is substantial heterogene- and income per capita increases, house- ity among developing countries. Further- holds that have already met their needs for more, recent evidence points to a disruption food begin to purchase industrial goods or stunting of the traditional development and services. At higher levels of income, ladder because developing countries appear households devote a larger share of expen- to be starting the deindustrialization process ditures to services. This mechanism is often at earlier stages of development—that is, referred to in the literature as the income at lower levels of gross domestic product effect. (GDP) per capita—and achieving lower peak The second mechanism posits that exog- industry shares. This phenomenon has been enous technological growth differs across dubbed “premature deindustrialization,” and sectors, which generates long-term changes it is explored at length later in this chapter. in the relative prices of sectoral goods What explains these stylized facts of (Baumol 1967; Ngai and Pissarides 2007). structural transformation? Is this change According to this mechanism, economic in the economic structure of countries’ activity moves away from the agriculture growth maximizing or desirable? Many sector because technological growth in analysts have argued that the transition the sector outpaces technological growth toward the lower productivity and lower elsewhere, making agricultural goods productivity grow th services sector is cheaper over time. Under the assumption problematic because productivity growth that sectoral goods are complements in will stall and with it the growth of an consumption, the relative decline in agri- economy. The next section turns to these cultural prices implies a lower allocation questions. of the household budget to agricultural goods. The main implication of this theory is that higher relative productivity growth What drives structural in one sector pushes workers toward the transformation? lower productivity growth sectors (again, Determining whether the observed patterns under the assumption of complementarity of transformation are efficient requires under- of sectoral goods in consumption and in standing what forces drive the process. Over closed economy models). Therefore, higher the last two decades, research has taken big growth in the productivity of the indus- steps toward identifying the fundamental trial sector would tend to push workers mechanisms underlying structural transforma- toward the services sector. Because of the tion. Although the search for newer explana- association with changes in relative prices, tions continues, two contrasting mechanisms this mechanism is known as the price have attracted most of the scrutiny. effect.4 W h a t i s s t r u c t u r a l t r a n s f o r m a t i o n ?    11 Structural transformation and the Sinha (2019a) analyzes the relative strength role of trade of five different channels—including trading patterns—in accounting for the observed Observers from academia and policy insti- share of industrial employment in econo- tutions argue that a rise in global trade mies in the Latin America and the Caribbean is fostering deindustrialization in many (LAC) region. The author finds limited sup- countries. Economies that lack a comparative port in favor of the comparative advantage advantage in industrial production import hypothesis. Countries with high industrial these goods and therefore allocate productive shares in employment do not derive a large resources to the other sectors. enough share of the value added from indus- Although the bulk of the literature has trial exports relative to countries with a low concentrated on closed economy models, industrial employment share to justify the some studies highlight how trade influences glaring industrial share gap between the two the pattern of transformation. A country groups. According to the analysis, differences experiencing an improvement in comparative in trade shares account for only a tenth of the advantage on the back of relatively high tech- 11 percentage point gap between the LAC nological growth in one of the sectors will sample and a set of benchmark advanced see greater allocation of activity to the sector countries over the 1995–2011 time frame. (Matsuyama 2009). Thus trade opposes the Sinha notes, however, that even though lit- change in allocation induced by the price tle evidence supports the comparative advan- effects mechanism just outlined. Perhaps the tage hypothesis, other trade-induced forces most critical implication of the open economy may be at play. For example, trade may inter- model is that it allows the behavior of pro- act with preferences on the consumption side duction side measures of transformation— of the economy as household expenditure value-added shares and employment—to shares adjust to trade. For instance, trade may deviate from the behavior of the consump- introduce new products and varieties in the tion expenditure share. This discrepancy in domestic market that in turn utilize domesti- the two types of measures of transformation cally produced industrial inputs. Thus when is observed in the data and is pronounced for consumption shifts toward these new prod- some countries.5 ucts and varieties, a share of the household Few studies have quantitatively appraised expenditure also moves toward domestically the role of trade, and when they have, produced industrial inputs indirectly. Simi- dissection of the transformation of the larly, sectoral linkages and productivity gaps Republic of Korea has been at the forefront. may also respond to trade shocks. Uy, Yi, and Zhang (2013) and Teignier (2018) find the trade-induced mechanism Structural transformation and the to be critical to understanding the Korean role of intermediate goods experience. Nevertheless, comparing the labor allocation in Korea to allocations in Following the bulk of the literature on countries at a similar level of development structural transformation, this study team that industrialized before trade was prevalent adopted the assumption that each economic does not reveal a remarkable deviation. An sector has a production function that takes analysis of a larger set of countries reveals labor and capital as inputs for production. that, barring some cases, trade has played a Recently, studies have begun looking at the secondary role to the main mechanisms just relationship between sectoral linkages and described (Świę cki 2017). Trade, however, structural transformation. Specifically, the occupies center stage in smaller economies new models explicitly account for the fact and accordingly has been crucial in determin- that the output of one sector is often used ing their path of structural transformation. as an input for another sector. This section 12   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n explores the potentially important role of allows for price effects on the production side intermediate goods in explaining patterns of as well by allowing for a nonunitary elastic- structural transformation. ity of substitution across intermediates from Berlingieri (2013) documents that shifts in different sectors. Thus sector-biased produc- input-output relationships could account for tivity growth affects changes in intermediate roughly a quarter of the decline in the man- expenditures and changes in final consump- ufacturing employment share in the postwar tion expenditures. United States. Using input-output data for Sinha (2019b) also documents that, as the United States for 1947–2002, Berlingieri in the United States, in Latin American finds that 50 percent of the employment economies shifts in intermediate expendi- growth and 94 percent of the GDP growth ture shares are significant and often larger of the services sector are explained by the in magnitude compared with the shifts in growth of service subsectors—professional final consumption shares. Two import- and business services and finance and real ant results from this exercise are worth estate—in which final demand plays a rela- highlighting. First, changes in distortions tively small role.6 He highlights two import- create a contractionary pressure on the ant channels that help explain the decline of industrial employment in five countries in manufacturing and the rise of services in the the sample. Second, distortions in the use United States: changes in the composition of of service inputs relative to industrial inputs intermediates and their sourcing mode. Spe- explain 80–90 percent of the counterfac- cifically, he suggests that service activities that tual change. When service input distortions were performed within a manufacturing firm are held at their historical minimum (over are now being outsourced to firms specializ- the 1995–2011 time frame of the analysis), ing in these services. In this setting, changes the industrial sector gains a share of in intermediate demand lead to a reallocation 2.5 percentage points on average because sec- of labor across sectors. He concludes that toral inputs are estimated to be complements the sole evolution of the input-output struc- in the production function. Thus reducing ture of the economy accounts for 36 percent distortion in the intermediate service market of the total increase in service employment makes the inputs relatively cheaper, and so and 25 percent of the decline in manufactur- all sectors will tend to increase the share of ing. Sinha (2019a) uses a similar accounting nonservice inputs. In summary, the analysis framework and finds that differences in sec- establishes that distortions play a quantita- toral linkages could account for a third of the tively significant role in intermediate mar- gap in the industrial share of employment kets in determining the sectoral allocation of across the LAC region and advanced compar- labor in the LAC region. ator economies over 1995–2011. In a background paper for this report, The LAC experience Sinha (2019b) analyzes the sectoral alloca- tion of labor in eight Latin American econ- Although the stylized facts of structural omies with an emphasis on whether changes transformation are robust across coun- in distortions in intermediate markets can tries, the patterns of transformation are far have a quantitatively meaningful impact on from identical. The cross-country variations the industrial share of employment. In line vary systematically across certain dimen- with the structural transformation literature, sions, which supports the argument that the main thrust of his model is that economic the sectoral allocation may not be efficient sectors experience different rates of exoge- in some instances. This section turns to the nous productivity growth. This sector-biased experience of the LAC economies and how productivity growth leads in turn to changes they compare with the path observed for in the relative prices of sectoral outputs. today’s developed countries and for their The pivotal extension of the model is that it developing economy peers. W h a t i s s t r u c t u r a l t r a n s f o r m a t i o n ?    13 Since the seminal paper by Rodrik (2016), for the manufacturing sector and can confirm much attention has been devoted to the notion both visually and statistically that the path of the premature deindustrialization of devel- for LAC countries differs from that of today’s oping countries, particularly Latin American developed nations. As stated earlier, LAC economies. The concept of deindustrializa- economies have entered the deindustrializa- tion is not new. Advanced economies have tion phase earlier in the development process been deindustrializing for decades and have (lower GDP per capita) and have achieved shifted into a postindustrial phase of devel- lower peaks (see figure 1.2).7 opment. However, Rodrik (2016, 2) docu- Although a broad pattern of premature ments a “less noticed trend over the last three deindustrialization is true for LAC countries, decades which is even more striking, and there is heterogeneity among them. Consis- puzzling, a pattern of de-industrialization in tent with their level of development, LAC low- and middle-income countries. . . . The economies are at different stages of deindus- hump-shaped relationship between industri- trialization (see figure 1.3, panel a, for real alization (measured by employment or output value-added shares and panel b for employ- shares) and incomes has shifted downwards ment shares). At one end are countries such as and moved closer to the origin.” In other Argentina and Chile with the highest devel- words, the industrial share at its peak in these opment levels and a clear downward trend countries is lower than the ones achieved in in the share of employment in industry. At the past by developed countries, and the peak the other end, Bolivia, with the lowest level materializes at lower levels of income per of development in the sample, is still indus- capita. The study team replicated the analysis trializing—that is, the share of employment FIGURE 1.2  Patterns of industrialization across LAC and high-income countries 0.35 0.30 Employment share in manufacturing 0.25 0.20 0.15 0.10 0.05 0 6.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10 Logarithm of GDP per capita (1990 international $) All data All data tted LAC LAC tted High income data High income tted Sources: Original calculations for this publication using Groningen Growth and Development Centre (GGDC)’s 10-Sector Database (Timmer, de Vries, and de Vries 2015); Maddison Database (Bolt et al. 2018). Note: Graph depicts the share of manufacturing on the logarithm of GDP per capita (expressed in 1990 international dollars). Solid lines are simulated shares from a quadratic fit. Data cover the period 1950–2012 for 40 countries. Graph is based on the World Bank’s 2012 classification of countries by income. GDP = gross domestic product; LAC = Latin America and the Caribbean. Share of value added Share of value added Share of value added Share of value added 14   0 0.2 0.4 0.6 0 0.2 0.4 0.6 19 0 0.2 0.4 0.6 19 0 0.2 0.4 0.6 0.8 19 19 50 50 50 50 19 19 19 19 55 55 55 55 19 19 19 19 60 60 60 60 19 19 19 19 65 65 65 65 19 19 19 19 70 70 70 70 19 19 19 19 75 75 75 75 19 19 19 80 80 80 19 80 19 19 19 85 85 85 19 85 19 19 19 a. Real value-added shares 90 90 90 19 90 19 19 19 95 95 95 19 95 20 20 20 00 00 00 20 00 20 20 20 05 05 05 20 Agriculture 20 05 10 20 20 20 10 10 10 Brazil Chile Bolivia Argentina Share of employment Share of employment Share of employment Share of employment Industrial 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 19 0 0.2 0.4 0.6 0.8 19 19 0 0.2 0.4 0.6 0.8 50 50 19 50 50 19 19 19 19 55 55 55 55 19 19 19 19 60 60 60 60 Services 19 19 19 19 65 65 65 65 19 19 19 19 70 70 70 70 19 19 19 19 75 75 75 75 19 19 19 19 80 80 80 FIGURE 1.3  Value-added and employment shares by sector: Selected LAC countries, 1950–2010 80 19 19 19 19 85 85 85 85 b. Employment shares 19 19 19 19 90 90 90 90 19 19 19 95 95 19 95 95 20 20 20 00 00 20 00 00 20 20 20 05 05 20 05 05 20 20 20 10 10 20 10 10 G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n figure continues next page W h a t i s s t r u c t u r a l t r a n s f o r m a t i o n ?    15 FIGURE 1.3  Value-added and employment shares by sector: Selected LAC countries, 1950–2010 (Continued) a. Real value-added shares b. Employment shares Colombia 0.6 0.6 Share of employment Share of value added 0.4 0.4 0.2 0.2 0 0 50 55 60 65 70 75 80 85 90 95 00 05 10 50 55 60 65 70 75 80 85 90 95 00 05 10 19 19 19 19 19 19 19 19 19 19 20 20 20 19 19 19 19 19 19 19 19 19 19 20 20 20 Costa Rica 0.8 0.8 Share of employment Share of value added 0.6 0.6 0.4 0.4 0.2 0.2 0 0 50 55 60 65 70 75 80 85 90 95 00 05 10 50 55 60 65 70 75 80 85 90 95 00 05 10 19 19 19 19 19 19 19 19 19 19 20 20 20 19 19 19 19 19 19 19 19 19 19 20 20 20 Mexico 0.6 0.6 Share of employment Share of value added 0.4 0.4 0.2 0.2 0 0 50 55 60 65 70 75 80 85 90 95 00 05 10 50 55 60 65 70 75 80 85 90 95 00 05 10 19 19 19 19 19 19 19 19 19 19 20 20 20 19 19 19 19 19 19 19 19 19 19 20 20 20 Peru 0.8 0.6 Share of employment Share of value added 0.6 0.4 0.4 0.2 0.2 0 0 50 55 60 65 70 75 80 85 90 95 00 05 10 50 55 60 65 70 75 80 85 90 95 00 05 10 19 19 19 19 19 19 19 19 19 19 20 20 20 19 19 19 19 19 19 19 19 19 19 20 20 20 Agriculture Industrial Services Sources: Original calculations for this publication. Employment and value-added data: Groningen Growth and Development Centre (GGDC), 10-Sector Database (Timmer, de Vries, and de Vries 2015); GDP: Penn World Tables (Feenstra, Inklaar, and Timmer 2015). Note: All value-added values are computed at 2005 local currency units. GDP = gross domestic product; LAC = Latin America and the Caribbean. 16   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n in industry is actually growing. Somewhere workers. Instead, it is a story of a stunted in between are countries such as Brazil and industrialization process whereby the indus- Mexico, which display small declines and trial sector was unable to grow and create stable shares of employment in industry, jobs over time as it did in developed econo- respectively. To be clear, this finding does mies. In part, this story may be related to the not imply that premature deindustrializa- history behind the industrialization phase in tion is not occurring or is more muted than LAC countries. Economies began the indus- expected. Given their development levels, trialization process under the banner of the Brazil and Mexico have industrial sectors old school industrial policies of protectionism that are smaller than what developed econo- and subsidies. As globalization evolved, the mies achieved at Brazil and Mexico’s income industrial sector of LAC countries largely did level. Moreover, countries such as Bolivia not successfully integrate into global value and Peru should be industrializing at a much chains. The industrial sector was therefore higher pace. unable to grow, limited by the size of domes- The changes in employment shares are tic and, in some cases, regional markets. much more pronounced than the changes in real value-added shares. This is consistent Is premature deindustrialization with the experience of the United States, a problem? where the employment share drop has been significantly more pronounced than the drop Since the work of Rodrik (2016), scholars and in real value added. This evidence points to policy makers have become concerned about the rapid growth of labor productivity in the onset of premature deindustrialization in the industrial sector. As noted by Rodrik countries undergoing their earlier stages of (2016, 2), “in the United States manufactur- structural transformation. The reallocation ing industries’ share of total employment has of resources out of industry into services is steadily fallen since the 1950s, coming down starting at lower levels of development and from around a quarter of the workforce to at lower peaks than in developed nations. less than a tenth today. Meanwhile, manufac- This property of structural transformation turing valued added has remained a constant is interpreted as reflecting an inefficient share of GDP at constant prices—a testament reallocation of labor due to some underlying to differentially rapid labor productivity distortions—that is, Latin America should be growth in this sector.” deindustrializing at a slower pace (and, for One final clarifying point is that analysis the least developed countries, industrializing of structural transformation and deindustri- at a faster pace). However, there is no theo- alization is based on comparing the relative retically grounded benchmark of efficiency importance of sectors and not absolute levels against which to compare the data that of employment or value added. In fact, the justifies the conclusion of inefficiency. number of employed people in the industrial Could it be that resources are flowing out sector has grown over time in almost all LAC of industry sooner than in other countries countries, including Argentina and Chile (see because the underlying drivers of structural figure 1.4), where deindustrialization has change are efficiently calling for such a pat- been under way for decades. However, the tern of deindustrialization? number of people employed in the services On the one hand, scholars such as Rodrik sector has skyrocketed—even in less devel- (2016) posit that the decline in industry oped economies such as Bolivia—leading shares is not good news for developing to a falling relative share of employment in countries because it blocks the main avenue industry. for economic convergence. This assertion is Therefore, the story of deindustrializa- rooted in the fact that manufacturing (the tion in the LAC region is not one of shut- main component of the industrial sector) not tered factories and mass layoffs of factory only has higher productivity, but also higher W h a t i s s t r u c t u r a l t r a n s f o r m a t i o n ?    17 FIGURE 1.4  Absolute total level of employment by sector: Selected LAC countries, 1950–2010 a. Argentina b. Bolivia 15,000 3,000 Employment (thousands) Employment (thousands) 2,500 10,000 2,000 1,500 5,000 1,000 500 0 0 50 55 60 65 70 75 80 85 90 95 00 05 10 50 55 60 65 70 75 80 85 90 95 00 05 10 19 19 19 19 19 19 19 19 19 19 20 20 20 19 19 19 19 19 19 19 19 19 19 20 20 20 c. Brazil d. Chile 80,000 6,000 Employment (thousands) Employment (thousands) 5,000 60,000 4,000 40,000 3,000 2,000 20,000 1,000 0 0 50 55 60 65 70 75 80 85 90 95 00 05 10 50 55 60 65 70 75 80 85 90 95 00 05 10 19 19 19 19 19 19 19 19 19 19 20 20 20 19 19 19 19 19 19 19 19 19 19 20 20 20 e. Colombia f. Costa Rica 15,000 1,500 Employment (thousands) Employment (thousands) 10,000 1,000 5,000 500 0 0 50 55 60 65 70 75 80 85 90 95 00 05 10 50 55 60 65 70 75 80 85 90 95 00 05 10 19 19 19 19 19 19 19 19 19 19 20 20 20 19 19 19 19 19 19 19 19 19 19 20 20 20 g. Mexico h. Peru 40,000 8,000 Employment (thousands) Employment (thousands) 30,000 6,000 20,000 4,000 10,000 2,000 0 0 50 55 60 65 70 75 80 85 90 95 00 05 10 50 55 60 65 70 75 80 85 90 95 00 05 10 19 19 19 19 19 19 19 19 19 19 20 20 20 19 19 19 19 19 19 19 19 19 19 20 20 20 Agriculture Industry Services Source: Original calculations for this publication using the Groningen Growth and Development Centre (GGDC)’s 10-Sector Database (Timmer, de Vries, and de Vries 2015). Note: LAC = Latin America and the Caribbean. 18   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n productivity growth. Indeed, there is evidence general equilibrium model with income and that manufacturing plays a critical role in the relative price effects in consumption. The catch-up process because it exhibits uncon- author derives from this model a benchmark ditional convergence in labor productivity of efficiency against which to characterize the unlike other sectors of the economy (Rodrik pattern of deindustrialization in the data. 2012). Moreover, there is general skepti- Using this model of structural transforma- cism that services can serve as an alternative tion, Fattal Jaef identifies the efficient baseline engine for growth. Although high-productiv- by identifying the paths of labor allocation ity and tradable services are available such as across sectors implied by the model after feed- information and communications technology ing it estimated paths of sectoral labor pro- (ICT) and finance, they are generally highly ductivities and the observed growth in real skill-intensive and cannot absorb large num- expenditure per capita. The author then evalu- bers of unskilled workers in the way that ates the inefficiency of the observed premature manufacturing does (or at least did in the deindustrialization hypothesis by comparing past). Other service subsectors tend to be the labor allocations in the data against the less dynamic (lower productivity growth) or model. In this context, it is possible to char- nontradable, which limits their ability to be acterize a pattern of deindustrialization as an engine of growth because they are con- premature and inefficient if the decline of strained by the size of the domestic market. employment of manufacturing occurs at a rate On the other hand, the deindustrialization faster than the one predicted by the bench- patterns observed in LAC countries may be mark model. For some countries, manufactur- the result of changes in the underlying driv- ing activity is still on the rise. In this case, the ers of structural transformation and thus the industrialization is labeled sluggish if manu- efficient (growth-maximizing) path. In other facturing employment in the data increases at words, efforts to stop or reverse deindustrial- a slower pace than is predicted by the model. ization patterns would create distortions in the The theory, definitions, and estimates economy that ultimately would result in lower of sectoral productivity growth, and the overall growth. In a background paper for this observed path of aggregate real expenditure report, Fattal Jaef (2019) evaluates the pat- per capita, give rise to evidence of prema- terns of structural change in the LAC region ture deindustrialization in Latin America through the lens of a standard three-sector starting in the 1980s (see figure 1.5). For the period 1950–1980, the benchmark model of structural change tracks the observed manu- FIGURE 1.5  Premature deindustrialization: LAC region (average), facturing employment share very closely. For 1950–2010 1980 onward, however, the model predicts that manufacturing employment should 0.24 have continued to expand. Instead, the data indicate a reversal of the trend as employ- Employment share 0.22 ment began to decline. Figure 1.5 shows that, as suggested by 0.20 Rodrik (2016), there is evidence of prema- 0.18 ture deindustrialization occurring in Latin America as of the mid-1970s, years in which 0.16 the model continues to exhibit a rising share of manufacturing employment, whereas the 50 60 70 80 90 00 10 19 19 19 19 19 20 20 share begins to decline in the data. Data Nonhomothetic To make sense of the model’s predicted dynamics, figure 1.6 illustrates the estimated Source: Fattal Jaef 2019. Note: Figure shows average employment shares in industry. The nonhomothetic line refers to paths of relative prices and real consump- model fit. LAC = Latin America and the Caribbean. tion per capita for Latin America (average). W h a t i s s t r u c t u r a l t r a n s f o r m a t i o n ?    19 FIGURE 1.6  Relative prices and real consumption per capita: LAC region (average), 1950–2010 a. Relative price, industrial goods to agricultural goods b. Relative price, industrial goods to services 2.5 1.1 1.0 2.0 0.9 1.5 0.8 1.0 0.7 0.6 0.5 0.5 0 0.4 1950 1970 1990 2010 1950 1970 1990 2010 c. Real consumption per capita (1950 = 1) 3.0 2.5 2.0 1.5 1.0 1950 1970 1990 2010 Sources: Original calculations for this publication using the Groningen Growth and Development Centre (GGDC)’s 10-Sector Database (Timmer, de Vries, and de Vries 2015); Penn World Tables (Feenstra, Inklaar, and Timmer 2015). Note: LAC = Latin America and the Caribbean. The relative price of industrial to agricultural that here is called premature deindustri- goods remains stagnant between 1950 and alization—specifically, in the sense that it 1970, rising notably thereafter (panel a). At implies an inefficiency. the same time, industrial goods cheapen rela- Investigation of each of the seven countries tive to services throughout the entire period, in the sample reveals substantial heterogene- albeit with a slowdown in the mid-2000s ity in the patterns of structural change, with (panel b). Finally, real consumption grows manufacturing activity still rising in some steadily until the 1980s, remaining stagnant countries (see figure 1.7). Argentina is per- until the end of that decade after which it haps the most salient case of premature dein- resumes its growth trajectory (panel c). dustrialization, followed by Chile and Peru. In the benchmark model, the low-income The model seems to follow the data closely elasticity of agricultural goods (relative to for Colombia and Brazil, whereas Mexico industrial goods) induces a reallocation is an outlier in the sense that its observed from agriculture to the industrial sector. industrialization is higher than the bench- Quantitatively, it follows that the relative mark prediction in contrast with the pattern strengths of these channels make the model of the average. and the data remarkably close, at least until In addition to assessing the degree of the 1980s. Thereafter, because relative premature deindustrialization, Fattal Jaef price trends do not change and real con- (2019) offers a quantitative assessment of the sumption continues to grow, the model pre- aggregate output costs associated with this dicts that the region should have continued process. He finds that premature deindustri- to industrialize. The data, however, show alization is not very costly to the LAC region a significant decline. It is this divergence in terms of output and aggregate produc- between data and the benchmark model tivity. For the region as a whole, the output 20   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 1.7  Labor allocation in manufacturing: Selected LAC countries, 1950–2010 a. Argentina b. Bolivia 0.35 0.30 Employment share Employment share 0.30 0.25 0.25 0.20 0.20 0.15 0.15 0.10 0.10 1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010 c. Brazil d. Chile 0.25 0.35 Employment share Employment share 0.30 0.20 0.25 0.15 0.20 0.15 0.10 0.10 1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010 e. Colombia f. Mexico 0.25 0.30 Employment share Employment share 0.20 0.25 0.20 0.15 0.15 0.10 0.10 1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010 g. Peru 0.25 Employment share 0.20 0.15 0.10 1950 1960 1970 1980 1990 2000 Services Nonhomothetic Source: Fattal Jaef 2019. Note: LAC = Latin America and the Caribbean. costs are on the order of 0.1 percent—that is, deindustrialization do not give policy makers aggregate output would be 0.1 percent higher enough incentive to justify implementation if the region had followed the path predicted of industrial policies aimed at remedying by the model. Obviously, the cost estimates it. Taking into account how difficult it is to are highly dependent on the choice of the identify the policy that will undo the ongoing efficient benchmark. Thus the results should distortion, together with the stickiness of the be interpreted with caution. subsidies and benefits put in place to incen- At first sight, under the given trade-offs of tivize manufacturing activity, the available the underlying model, one interpretation of gains do not seem to justify the costs. Alter- these findings is that the costs of premature natively, these findings could be interpreted W h a t i s s t r u c t u r a l t r a n s f o r m a t i o n ?    21 as strengthening the case for improving Conclusions fundamentals rather than seeking to change the course of structural change through Evidence suggests that trade has not been a policy instruments. One improvement that major factor in explaining the observed pat- would bring about a slowdown of deindus- terns of the changing economic structure of trialization, or a strengthening of ongoing the LAC region. Although trade is import- industrialization, is increasing the productiv- ant in some specific cases, such as Korea and ity of the services sector. some smaller countries, little evidence points A second interpretation of the findings is to the comparative advantage hypothesis in that the model leaves out many features of the explaining the structural change patterns in manufacturing sector that make it desirable the LAC region. to subsidize its operations. Thus the output The role of intermediate inputs and the cost calculations are just the lower bounds input-output relationships between sectors of the full costs of premature deindustri- appears to have some quantitative impor- alization. Rodrik (2016) outlines various tance in explaining the observed economic reasons why manufacturing is special rela- structure in LAC and other developing tive to other sectors of an economy—features countries. Input-output relationships seem that have not been explicitly captured in the to vary systematically according to the model. One of these features that would have degree of development, with more advanced first-order welfare effects is the prominent nations having more interconnected sec- role typically played by manufacturing in tors. In a background paper for this report, absorbing low-skilled workers. A premature Sinha (2019b) finds that distortions in inter- bypassing of industrial activity would reduce mediate markets may play a quantitatively the demand for low-skilled workers, bringing important role in explaining the size of the with it an increase in the skill premium and industrial sector. Distortions in the sourcing therefore an increase in inequality. of intermediate service inputs may have a All this being said, the findings suggest particularly important role. Keeping inter- that expensive industrial policies that intro- mediate service input distortions at their duce more distortions in the economy are observed minimum over the 1995–2011 not readily warranted. Because of the com- period would imply a larger industrial sector plexities of implementing such policies, their by 2–2.5 percentage points—a large effect stickiness, and the discretion underlying considering the actual industrial share of the choice of winners and losers, confront- about 20 percentage points. ing these risks may not be worthwhile. This The process of structural transforma- is not to say that there is no space for gov- tion documented in detail for a sample of ernment policy to address latent issues in LAC economies for which internationally the industrial sector. In fact, as developed comparable data exist confirms the find- further in chapter 2, labor productivity in ings of Rodrik (2016). The process known the LAC industrial sector significantly lags as deindustrialization has in fact begun at that of the United States. In particular, it lower levels of GDP per capita (relative to appears there are significant distortions in the experience of advanced nations) and the the sector that would result in a firm size share of manufacturing in total value added distribution heavily skewed toward small has peaked at lower levels than for advanced and microenterprises. In addition, as new nations. technologies are incorporated into the pro- Some additional features of this process duction processes, complementary invest- are important for the policy debate. First, ments in human capital and infrastructure, the deindustrialization process is reflected as well as modernization of regulatory more acutely in terms of the share of employ- frameworks, will be central to the ongoing ment rather than the share of value added. competitiveness of the industrial sector. This pattern is similar to that observed for 22   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n the United States and may be related to the the actual output costs associated with this introduction of labor-saving technologies process were relatively small (on the order of that increase productivity (and help sustain 0.1 percent of output). the share of value added) but do not foster Thus the premature deindustrialization job creation. process observed in the LAC region since Second, consistent with the differences the 1980s has not been very costly in terms among countries in development level, there of output loss. But this is not to say that is substantial heterogeneity in where they there is no room for improvement or that stand in the deindustrialization process. For there is no scope for government policies to the more developed economies in the LAC improve the allocation of resources across region such as Argentina and Chile, the the economy. The focus should be on rais- deindustrialization process is marked and ing productivity in all sectors and facilitat- has been ongoing for decades. Less devel- ing the transition of workers and resources oped countries such as Brazil, Colombia, and among sectors. Indeed, as shown in the next Mexico exhibit stagnant or slight increases chapter, there are productivity issues in all in their share of industrial employment. At sectors of the economy. Special attention the other end, the least developed economy should be paid, however, to understanding in the study sample, Bolivia, is still industri- the specific productivity issues in the ser- alizing, with a growing share of industrial vices sector. Not only is this sector already employment. This finding does not imply the main employer in LAC economies, the that deindustrialization is not occurring in expectation is that it will continue to grow these countries. And yet relative to the per- as countries continue to develop. Moreover, formance of advanced nations and given their the dearth of data specific to the services level of development, these countries should sector is particularly worrisome, as there is be industrializing at a much faster pace. little evidence on the issues that affect firms Third, an analysis based on shares may in that sector. present a distorted view of reality. The abso- lute number of jobs in the industrial sector is steady or increasing in most countries Notes in the region, including in Argentina and 1. Agriculture refers to agriculture, forestry, and Chile, which have been deindustrializing for fishing. Industry refers to mining and quarry- decades. At the same time, the number of ing, utilities, construction, and manufacturing. jobs in the services sector has skyrocketed, For most countries, manufacturing is the larg- est component of the industrial sector. Services leading to a declining share of industrial jobs. are all other industries. Is the deindustrialization process in the 2. This phase of development is known as LAC region “premature” in the sense of deindustrialization. being inefficient? Empirically, it is clear 3. Studies in the 1950s and 1960s such as those that peak manufacturing shares achieved in by Chenery (1960), Clark (1951), and Kuznets LAC economies were lower relative to those (1966) contributed to documenting these styl- achieved by high-income countries, and that ized facts of the structural transformation the declining shares of industry are happen- process. More recently, Herrendorf, Rogerson, ing at lower levels of GDP per capita. Miss- and Valentinyi (2014), using data from mul- ing from this empirical assessment, however, tiple sources, presented a detailed account of is an evaluation of whether this was an the process covering many countries across the global income distribution. They also provided efficient (growth maximizing) process or the a comprehensive survey of both the theoretical result of inefficiencies or distortions in the and the empirical literature. economy. A background paper prepared for 4. Sector-biased technological change is not a con- this report by Fattal Jaef (2019) asserts that, dition needed for relative prices to change. Rel- although on average the LAC region has pre- ative prices can also change if instead sectors maturely and inefficiently deindustrialized, differ in how intensively they use certain inputs W h a t i s s t r u c t u r a l t r a n s f o r m a t i o n ?    23 over others and if the relative supply of these Caselli, F., and W. J. Coleman II. 2001. “The inputs changes over time. Caselli and Coleman US Structural Transformation and Regional (2001) have explored the shift in the relative Convergence: A Reinterpretation.” Journal of abundance of low- and high-skilled labor, and Political Economy 109 (3): 584–616. Acemoglu and Guerrieri (2008) have studied Chenery, H. B. 1960. “Patterns of Industrial the trend in the relative availability of capital Growth.” American Economic Review 50 (4): and labor during the transformation process. 624–54. 5. Researchers have also explored the influence Clark, C. 1951. The Conditions of Economic of other factors such as the costs associated Progress. London: Macmillan. with the movement of goods (see Adamopou- Dekle, R., and G. Vandenbroucke. 2012. “A los 2011; Gollin and Rogerson 2014) and labor Quantitative Analysis of China’s Structural (see Dekle and Vandenbroucke 2012; Lee and Transformation.” Journ al of Economic Wolpin 2006) across sectors. A recent but Dynamics and Control 36 (1): 119–35. growing literature is exploring how sectoral Fattal Jaef, R. N. 2019. “A Quantitative Eval- linkages interact with the transformation pro- uation of the Premature Deindustrialization cess (Berlingieri 2013; Sposi 2019). Hypothesis in Latin America.” Working paper, 6. In 2002 roughly 83 percent of the output of World Bank, Washington, DC. professional and business services was sold to Feenstra, R. C., R. Inklaar, and M. P. Timmer. firms as intermediate inputs, compared with 2015. “The Next Generation of the Penn 44 percent for the economy as a whole. World Table.” American Economic Review 7. With the possible exception of Argentina. It 105 (10): 3150–82. Available for download achieved a peak of about 27 percent in manu- at www.ggdc.net/pwt. facturing, which is comparable to that achieved Gollin, D., and R. Rogerson. 2014. “Productivity, by today’s high-income countries. Transport Costs and Subsistence Agriculture.” Journal of Development Economics 107: 38–48. References Herrendorf, B., R. Rogerson, and A. Valenti- Acemoglu, D., and V. Guerrieri. 2008. “Capital nyi. 2013. “Two Perspectives on Preferences Deepening and Nonbalanced E conomic and Structural Transformation.” American Growth.” Journal of Political Economy Economic Review 103 (7): 2752–89. 116 (3): 467–98. Herrendorf, B., R. Rogerson, and A. Valentinyi. Adamopoulos, T. 2011. “Transportation Costs, 2014. “Growth and Structural Transforma- Agricultural Productivity, and Cross-Country tion.” In Handbook of Economic Growth, 2: Income Differences.” International Economic 855–941. Amsterdam: Elsevier. Review 52 (2): 489–521. Kongsamut, P., S. Rebelo, and D. Xie. 2001. Baumol, W. J. 1967. “Macroeconomics of Unbal- “Beyond Balanced Growth.” Review of anced Growth: The Anatomy of Urban Crisis.” Economic Studies 68 (4): 869–82. American Economic Review 57 (3): 415–26. Kuznets, S. 1966. Modern Economic Growth: Berlingieri, G. 2013. “Outsourcing and the Rise in Rate, Structure, and Spread. New Haven, CT: Services.” CEP Discussion Paper 1199, Centre Yale University Press. for Economic Performance, London School of Lee, D., and K. I. Wolpin. 2006. “Intersectoral Economics. Labor Mobility and the Growth of the Service Bolt, J., R. Inklaar, H. de Jong, and J. Luiten Sector.” Econometrica 74 (1): 1–46. van Zanden. 2018. “Rebasing ‘Maddison’: Matsuyama, K. 2009. “Structural Change in an New Income Comparisons and the Shape of Interdependent World: A Global View of Man- Long-run Economic Development.” Maddi- ufacturing Decline.” Journal of the European son Project Working Paper 10, Groningen Economic Association 7 (2–3): 478–86. Growth and Development Centre (GGDC), Ngai, L . R., and C . A . Pissarides. 20 07. Groningen, The Netherlands. “Structural Change in a Multisector Model ­ Buera, F. J., and J. P. Kaboski. 2012a. “The Rise of Growth.” American Economic Review of the Service Economy.” American Economic 97 (1): 429–43. Review 102 (6): 2540–69. Rodrik, D. 2012. “Unconditional Convergence Buera, F. J., and J. P. Kaboski. 2012b. “Scale and in Manufacturing.” Quarterly Journal of the Origins of Structural Change.” Journal of Economics 128 (1): 165–204. Economic Theory 147 (2): 684–712. 24   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n Rodrik, D. 2016. “Premature Deindustrializa- Świę cki, T. 2017. “Intersectoral Distortions and tion.” Journal of Economic Growth 21 (1): the Welfare Gains from Trade.” Journal of 1–33. International Economics 104: 138–56. Sinha, R. 2019a. “Distortions in Intermediate Teignier, M. 2018. “The Role of Trade in Markets and Structural Transformation in Structural Transformation.” Journal of Latin America.” Working paper, World Bank, Development Economics 130: 45–65. Washington, DC. Timmer, M. P., G. J. de Vries, and K. de Vries. Sinha, R. 2019b. “What Explains Latin America’s 2015. “Patterns of Structural Change in Devel- Low Share of Industrial Employment?” Policy oping Countries.” In Routledge Handbook of Research Working Paper 8791, World Bank, Industry and Development, edited by J. Weiss Washington, DC. and M. Tribe, 65–83. Abdingdon-on-Thames, Sposi, M. J. 2019. “Evolving Comparative UK: Routledge. Advantage, Sectoral Linkages, and Structural Uy, T., K.-M. Yi, and J. Zhang. 2013. “Structural Change.” Journal of Monetary Economics 103 Change in an Open Economy.” Journal of (May): 75–87. Monetary Economics 60 (6): 667–82. Productivity in the LAC region: A sectoral view 2 C hapter 1 described the phenomenon main avenues for productivity growth are known as premature deindustrial- (1) improving the technical efficiency of pro- ization and its impact on aggregate ducers in view of the existing technology productivity growth in economies in the and (2) pushing out the production possi- Latin America and the Caribbean (LAC) bilities frontier by shifting to new, improved region. This chapter analyzes the produc- technologies. Study of which policies are the tivity dynamics in each sector and its impli- most relevant should be conducted at the cations for the future economic structure in subnational level. Although soybean pro- the region. duction in some areas of Brazil and Argen- Unfortunately, the picture of produc- tina appears to be operating at the efficiency tivity in the region is worrisome. Pervasive frontier, low productivity subsistence farm- productivity issues are affecting all sectors ers can be found in other regions of those of the region’s economy (see figure 2.1). countries. On average, the region displays the larg- As for the industrial sector—specifically est productivity gap (relative to the United the manufacturing sector—evidence clearly States) in the agriculture sector. Even though shows a substantial degree of misallocation the gap is smallest in the industrial sector, between firms. The firm-size distribution is output per worker represents less than skewed toward small and microenterprises. 40  percent of the productivity in the US This finding points toward distortions in the industrial sector. Perhaps most worrisome is market that are preventing the consolidation the gap in the services sector because more and growth of the most productive firms. It than 60 percent of the workforce is employed thus calls for government action instituting there. The productivity of the services sector policies that foster competition— such in the LAC region is about 25 percent that of as international trade and deepening of the United States. regional trade agreements. It also calls Although agricultural productivity has for revisions of size-dependent policies increased in the LAC region, it still rep- (or enforcement) that appear to hamper the resents less than 20 percent of the out- growth of productive firms and incentivize put per worker in the United States. Two informality. 25 26   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 2.1  Output per worker by sector in LAC region relative to that of United States: Selected countries, 2010 0.7 Output per worker relative to United States 0.6 0.5 0.4 0.3 0.2 0.1 0 LAC Argentina Bolivia Brazil Chile Colombia Costa Rica Mexico Peru Agriculture Industrial Services Sources: Original calculations for this publication using Groningen Growth and Development Centre (GGDC)’s 10-Sector Database (Timmer, de Vries, and de Vries 2015); Maddison Database (Bolt et al. 2018). Note: Graph shows the relative output per worker in eight countries in the LAC region. Sectoral output by country is computed by weighing the total gross domestic product in 2011 international dollars by the share of sectoral value added. LAC = Latin America and the Caribbean. One of the main messages of this report has been accompanied by large inflows of is that a comprehensive set of policy actions foreign direct investment (FDI). Therefore, is urgently needed to address productiv- certain service subsectors are looking more ity issues in the services sector. The sector and more like the manufacturing sector already employs more than 60 percent of the (with exposure to trade and inflows of FDI), workforce in the LAC region, and current allowing for greater competition, technol- trends indicate it will continue to grow and ogy diffusion, and benefits of scale. One be the main source of job creation in the caveat is that these services generally require future. Although the scarcity of data on the high-skilled workers and so require in turn services sector is an obstacle to a clearer significant investments in the human capital diagnosis, the existing evidence indicates of the workforce. that this sector has a higher degree of misal- Services also provide inputs for the rest location relative to manufacturing. of the economy. According to Alvarez et al. The shift toward the services sector is (2019), the services sector in the LAC not all bad news. This sector is increasingly region has the highest degree of f ­orward sharing pro-development characteristics linkages (also referred to as “push”). In once thought to be the unique domain of other words, the services sector is heav- manufacturing. The rapid advances in infor- ily intertwined with the rest of the econ- mation and communications technology omy and is the most important sector (ICT) have enabled the emergence of ser- in terms of supplying inputs. The recent vice subsectors that are no longer limited by trend of “servicification” of manufactur- market size because more and more services ing indicates that more services are being can be digitally stored, codified, and easily used as inputs in the production of goods traded (Ghani and Kharas 2010). Mean- (embodied services), and more services while, the deregulation of services markets are provided to customers bundled in the P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    27 sale of goods (embedded services). There- growing more rapidly than total inputs, TFP fore, the increased productivity of back- is said to be increasing. If total output is bone services—such as logistics, ICT, and growing more slowly than total inputs, TFP business services—could ripple through- is said to be decreasing. out the economy, having larger impacts on Several authors have estimated the long- overall aggregate productivity. In fact, in term average annual growth in agricultural a background paper for this report, Sinha TFP for individual LAC countries or for (2019b) finds  that reductions in the cost the region as a whole. Recently, Trindade of service inputs could have quantitatively and Fulginiti (2015) used two different important effects on the size of the indus- m ethods—stochastic production frontier ­ trial sector. and the Malmquist Index—to estimate the The structure of the LAC economy is growth in agricultural TFP for a subset of changing and the requirements for produc- Latin American countries over 1969–2009. tivity growth within sectors are increasing. The results from the two approaches were This chapter turns first to analyzing pro- similar, showing TFP growth averaging about ductivity in the agriculture sector before 2.3 percent a year during the first decade of documenting the productivity dynamics in the 21st century. Common among these more the industrial and services sectors. It con- recent studies and those that preceded them cludes with a discussion of policy interven- is a finding of positive average annual growth tions to enhance productivity growth in the in agricultural TFP in the LAC region. future. Historically, TFP growth has been a major driver of output growth. Figure 2.2 reveals that, beginning in the late 1980s Productivity in agriculture and continuing for more than two decades, What has been the historical performance TFP growth rose steadily in the LAC region of growth in agricultural productivity in before peaking in 2005. Throughout this the LAC region? The question is not an easy period, output grow th moved mostly one to answer because agricultural pro- in tandem. After 2005, however, T FP ductivity and its determinants have often growth dropped sharply, accompanied by been inaccurately measured and imperfectly a slowdown in output growth. However, understood. Many regions have achieved out put g row th d id not decelerate as significant gains in agricultural labor produc- sharply as TFP growth because producers tivity over time, but a large proportion of the compensated for slowing TFP growth by gains came from more intensive use of other resorting to input intensification, especially complementary inputs such as fertilizers, land expansion (see Fuglie et al. 2012). machinery, energy, and irrigation. Because As expected based on the r ­ elationship more intensive use of these other inputs raises shown in figure 2.2, when TFP growth costs, partial productivity measures such is plot ted against output g row th for as land and labor productivity are likely to ­ i ndividual countries, a strong positive overstate the welfare effects of productivity correlation emerges. This relationship holds ­ change. For this reason, a broader concept up not only for countries with modernized, of agricultural productivity is desirable. The ­ t ech nolog ically adva nced ag ricu lt u re most widely used broader measure is total sectors, but also for countries with large factor productivity (TFP). TFP is defined as numbers of s ­ ubsistence-oriented producers the ratio of aggregate output to aggregate (see figure 2.3). In major agricultural inputs, and so it takes into account all factors producers such as Brazil, Chile, Mexico, involved in the production process (such and Peru, high T FP growth correlates as land, labor, capital, and other material strongly with high output growth, but the resources) and compares them with the total same is true in less developed countries crop and livestock output. If total output is such as Guatemala, Haiti, Honduras, and 28   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 2.2  Agricultural output and TFP growth: LAC region, 1981–2014 0.045 0.035 0.040 0.030 0.035 0.025 0.030 Output growth (%) TFP growth (%) 0.025 0.020 0.020 0.015 0.015 0.010 0.010 0.005 0.005 0 0 81 84 87 90 93 96 99 02 05 08 11 14 19 19 19 19 19 19 19 20 20 20 20 20 Agricultural output growth Agricultural TFP growth Source: Original calculations for this publication using US Department of Agriculture’s agricultural TFP growth indexes database (Fuglie 2015; Fuglie et al. 2012), smoothed using Hodrick-Prescott filter lambda = 6. LAC = Latin America and the Caribbean; TFP = total factor productivity. FIGURE 2.3  Correlation between output growth and TFP growth: LAC countries, 2001–14 0.06 GTM 0.05 HTI PRY URY PER SUR 0.04 BOL BRA NIC HND BHS DOM 0.03 CRI ARG Output growth (%) MEX GUY 0.02 ECU VEN COL PAN CHL SEV 0.01 BLZ PRI GUF 0 JAM –0.01 TTO CUB LCA –0.02 –0.03 –0.02 –0.01 0 0.01 0.02 0.03 0.04 0.05 Linear regression TFP growth (%) Source: Original calculations for this publication using US Department of Agriculture’s agricultural TFP growth indexes database (Fuglie 2015; Fuglie et al. 2012). Note: For country abbreviations, see International Organization for Standardization (ISO), http://www.iso.org/obp/ui/#search. LAC = Latin America and the Caribbean; TFP = total factor productivity. P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    29 Nicaragua. The latter ­ countries have large converted from agricultural to nonagricul- populations still employed in agriculture tural uses. sectors that are modernizing. The strong positive correlation is absent only in two Sources of future agricultural countries, Bolivia and Belize. Bolivia exhib- productivity growth in the LAC region ited very strong average output growth but negative TFP growth, whereas Belize Regardless of whether productivity in the recorded output growth despite registering agriculture sector is higher, equal to, or negative TFP growth. lower than productivity in other sectors, One drawback of analyzing productiv- the ability of agriculture to contribute to ity at the regional level is that the regional productivity growth in the overall econ- data conceal a large amount of variability omy depends on the size of the agricul- among countries. This variability can be ture sector and the rate of agricultural seen in figure 2.4, which decomposes by productivity growth. For that reason, it is subregion the agricultural growth recorded important to consider the size of the agri- between 2005 and 2014. The slowdown in culture sector in LAC countries, as well regional TFP growth after 2005 was driven as potential sources of future agricultural mainly by slower growth in the Southern productivity growth. Cone and Andean regions; TFP growth To what extent have LAC agriculture remained robust in Central America, the and food systems contributed to eco- Caribbean region, and the Northeast. Mean- nomic growth and diversification? The while, land contributed strongly to overall importance of agriculture in a country’s output growth in the Southern Cone and the economy is traditionally measured as the Northeast, where the agricultural frontier direct contribution of primary production continued to expand rapidly. By contrast, in activities to overall gross domestic product Central America and the Caribbean region (GDP).1 Measured this way, the importance the land frontier contracted as land was of primary agriculture as a share of the FIGURE 2.4  Growth decomposition: Latin America by region and United States, 2005–14 0.05 0.04 0.03 Growth (%) 0.02 0.01 0 –0.01 –0.02 America, Caribbean Central SA, SA, SA, United States LAC America Andean Northeast Southern Cone LAC Land Inputs TFP Output Source: Original calculations for this publication using US Department of Agriculture’s TFP growth indexes database (Fuglie 2015; Fuglie et al. 2012). Note: America, LAC, refers to average of entire region; LAC = Latin America and the Caribbean; SA = South America; TFP = total factor productivity. 30   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 2.5  Relationship between value added and employment in agriculture: Selected LAC countries, 2017 20 18 HTI 16 NIC Agriculture value added (% of GDP) 14 GUY HND 12 BOL BLZ 10 GTM ECU 8 VCT COL PER DOM JAM 6 URY ARG VEN CRI SLV BRA 4 CHL MEX 2 BRB LCA BHS TTO 0 5 10 15 20 25 30 35 40 45 Employment in agriculture (% of total employment) Source: Original calculations for this publication using World Bank’s World Development Indicators database, 2019 (https://datacatalog.worldbank.org/ dataset​/world-development-indicators). Note: For country abbreviations, see International Organization for Standardization (ISO), http://www.iso.org/obp/ui/#search. GDP = gross domestic product; LAC = Latin America and the Caribbean. FIGURE 2.6  Sources of agricultural productivity growth overall economy has declined in many LAC countries, but agriculture and food systems Output 1 remain a significant contributor to growth (see figure 2.5). Rate of agricultural productivity growth in LAC countries As just noted, the ability of agriculture to contribute to productivity growth in the overall economy depends not only on the size of the agriculture sector, but also on the rate of agricultural productivity growth. As shown in figure 2.6. two conceptually distinct sources of growth can be distin- X0 X1 guished: (1) that achieved by improving the technical efficiency of producers using the 0 Output 2 existing technology and (2) that achieved Moving to the frontier Shifting out the frontier by pushing out the frontier of production possibilities by shifting to new, improved Source: Original calculations for this publication. technology. P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    31 Moving toward the production possibilities incentives needed to catch up with the most frontier efficient producers. What scope exists to unlock future pro- What are the entry points for helping ductivity growth in agriculture by moving producers improve their technical efficiency inefficient producers closer to the produc- and move closer to the existing production tion possibilities frontier? Recent work possibilities frontier? A large empirical lit- carried out in Peru by Espinoza et al. (2018) erature provides insights into factors that using a stochastic production metafron- can influence technical efficiency at the tier approach suggests that the potential is farm level. Factors that show up consistently likely to vary significantly by region, farmer as playing a key role are described in the type, and production system (figure 2.7). following sections. In the Costa region, which is dominated Land. In the LAC region as elsewhere, land by technologically advanced, highly pro- markets are often imperfect. Transfers of land ductive commercial agriculture, the vast tend to be subject to cultural, political, or majority of farmers operate at high levels institutional factors that can raise transaction of efficiency and are clustered close to the costs and influence outcomes. Ownership efficiency frontier. In the Sierra region, of agricultural land is often unequally dis- which is dominated by subsistence-oriented tributed, and in many countries large num- smallholder systems characterized by lim- bers of very small farms coexist with small ited use of improved technology and pur- numbers of very large farms. If farm size chased inputs, efficiency levels are more were unrelated to productivity, it might not variable and centered farther from the fron- matter, but if farm size influences productiv- tier. In the Selva region, which contains a ity, to the extent that land markets prevent mixture of technologically advanced com- the consolidation or division of agricultural mercial plantations and technologically landholdings, productivity could be affected. lagging subsistence farms, the distribution What is the relationship between farm is very flat and dispersed, indicating the size and productivity? Finding the answer to presence of great variability in efficiency this question has proved to be a perennial levels. These results reveal that in con- puzzle in development economics (Barrett, texts such as the Sierra and Selva regions, Bellemare, and Hou 2010; Eastwood, considerable scope still exists to accelerate Lipton, and Newell 2010). Building on ideas productivity growth by moving inefficient first articulated by Schumacher (1973) in his producers closer to the frontier—that is, by classic work Small Is Beautiful: Econom- giving them the knowledge, resources, and ics as If People Mattered, many empirical FIGURE 2.7  Histogram of metatechnical efficiency, Peru by region a. Costa b. Sierra c. Selva 5 4 4 4 3 3 Density Density Density 3 2 2 2 1 1 1 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 Technical e ciency Technical e ciency Technical e ciency Source: Espinoza et al. 2018. 32   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n studies spanning a broad range of contexts which is highly variable in Latin America and have established the stylized fact that farm often differs between regions and between size and productivity are inversely related. farmer typologies within the same country. Leading explanations of this phenomenon Coelli and Rao (2003) estimated the con- include imperfections in labor, land, and tribution of inputs to TFP growth in agri- credit markets (Eswaran and Kotwal 1989; culture for the period 1980–2003. Using Sen 1966), moral hazard between employ- a cross-country approach, they calculated ers and hired agricultural labor (Feder shadow prices and shadow shares of inputs 1985), aversion to price risk (Barrett 1996), to shed light on factors influencing produc- and measurement and identification issues tivity growth. For land and labor, the shadow (Assunção and Braido 2007; ­ Benjamin 1992; shares appear to be meaningful and consis- Carletto, Savastano, and Zezza 2013). How- tent with the expected factor endowments ever, because the existing theoretical expla- of the countries. Shares of purchased inputs, nations fail to fully explain the observed including fertilizers, tractors, livestock, and inverse relationship, the discussion contin- irrigation, are also plausible and appear to ues over the nature and the strength of the support the overall underutilization of these relationship. Indeed, it has influenced the resources in different countries. The general debate over land reform in the LAC region insight emerging from this work is that input and has highlighted constraints to agricul- cost is often a limiting factor for agricultural tural productivity in the region, as well as productivity growth in LAC countries, and opportunities for unleashing faster produc- the price of labor plays an important role as tivity growth. countries develop. The relationship between farm size and Other authors have explored the same productivity, measured as TFP, may be issue using micro approaches. For example, dynamic, evolving over time and across Solis, Bravo-Ureta, and Quiroga (2009) agricultural regions. Comparing the results studied productivity among hillside farmers of studies of regions within Brazil, Helfand in El Salvador and Honduras using a and Taylor (2016) find that in some regions household-level, input-oriented stochas- the inverse relationship between farm size tic distance frontier. They concluded that and productivity has persisted, whereas in differences in the use of purchased inputs other regions it has become U-shaped. Most (including seeds, fertilizer, pesticides, and interestingly, in a few rapidly modernizing hired animal power) explain differences regions a direct positive relationship has in productivity levels among farmers. Fur- begun to replace the inverse relationship. thermore, purchased inputs have a higher Adding further complexity to the issue, impact on productivity among farmers that recent work in Mexico suggests that not only use purchased inputs at low levels, suggest- could the relationship between farm size and ing that degree of access to inputs affects productivity evolve over time, but technolog- productivity growth. These findings are con- ical change could occur at differential rates sistent with those of other microlevel studies across the farm size spectrum and be accom- that have concluded that budget constraints panied by changes in efficiency levels because often oblige small-scale farmers to employ producers in different farm size categories suboptimal amounts of inputs. The general vary in their ability to keep up (see box 2.1). conclusion emerging from this work is that Inputs. Differences between realized out- access to inputs can significantly affect agri- put and potential output can result from cultural productivity growth, and there is a the failure of producers to use the optimal high level of variability in the level of access amount of inputs. To do so, producers must to inputs in LAC countries. have access to the inputs, as well as the means Extension. Agricultural extension services to acquire them. Both conditions are influ- could be defined as the delivery of information enced by farmers’ access to input markets, inputs to farmers. These services can play an P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    33 BOX 2.1 Does technological change benefit small and large farms equally? Evidence from Mexico Helfand and Taylor (2016) explore the relationship year dummy variables, generate coefficients that are between farm size and productivity in Mexico and largely consistent in indicating (1) the existence of identify factors associated with inefficiency. Earlier a strong inverse relationship between farm size and work by Kagin, Taylor, and Yuñez-Naude (2016) using frontier TFP and (2) the existence of positive tech- data from the Mexico National Rural Household Sur- nological change—that is, the frontier is increasing vey (ENHRUM) found evidence of an inverse rela- over time. The stochastic frontier analysis thus finds tionship between farm size and productivity, driven in positive technological change at the frontier even part by larger farms being further from the efficiency though, on average, TFP is not observed to rise. frontier (that is, smaller farms were more efficient). The interactions between farm size and the sur- Using a different data set Mexican Family Life Survey vey year dummy variables identify a positive and (MxFLS),a Helfand and Taylor (2016) expand on the significant relationship between farm size and tech- findings by Kagin, Taylor, and Yuñez-Naude by explor- nological change, suggesting that such change has ing how the relationship may have changed over time. been biased toward larger farms and that the inverse Helfand and Taylor (2016) find an inverse relationship along the frontier has become less steep relationship between farm size and land productivity over time. Similarly, the interactions between farm over the entire range of farm sizes—a relationship size and the survey year dummy variables reveal a that is consistent over time and across samples. dynamic relationship between farm size and technical In each year, land productivity falls rapidly up inefficiency. Although inefficiency has increased over to approximately 1 hectare. Around 1 hectare, time across the entire farm size distribution, it has the relationship levels become relatively flat up to increased faster among larger farms. The differential approximately 20 hectares, at which point land changes in inefficiency across the farm size distribu- productivity once again dramatically declines. tion have caused the farm size–inefficiency relation- T he aut hors t a ke t wo complement a r y ship to disappear in later waves of the MxFLS. approaches to exploring the relationship between Helfand and Taylor find that technological change farm size and total factor productivity.b In the first, on Mexican farms has been accompanied by increas- they use an average production function to estimate ing technical inefficiency. This finding suggests that average total factor productivity (TFP). In the the majority of farms have been unable to achieve second, they use a stochastic production frontier to the same rate of TFP growth as the most productive estimate (1) TFP along the frontier and (2) technical farms, particularly at the upper end of the farm size inefficiency, identified as deviations from the distribution. Increasing technical inefficiency result- frontier. Using the average ­ p roduction function ing from the inability of nonfrontier households to approach, they find a statistically significant keep up is driving the decline of TFP over time iden- inverse relationship between farm size and TFP. tified in the average production function estimates. Tests of several alternative specifications highlight, These findings are consistent with the time-invari- however, the need to assume a flexible functional ant inverse relationship between farm size and TFP. form to fully understand the farm size–productivity Along the frontier, the inverse relationship between relationship because the linear specification does farm size and productivity is becoming less pro- not capture all of the subtleties. nounced because technological advances are favor- The authors’ analysis using the average produc- ing larger farms. At the same time, technological tion function is complemented with analysis using advances have been offset by growing inefficien- a stochastic production frontier to identify produc- cies among larger farms. Inefficiency was initially tivity at the technological frontier, as well as the smaller for larger farms, but this is no longer true in sources of production inefficiencies. The stochastic later waves of the MxFLS. The combination of these production frontier approach generates insights into two forces has led to a farm size–TFP ­ relationship the dynamics of technological change that are diffi- that has been relatively stable over time. cult to detect using the average production function In summary, the work by Helfand and Taylor approach. Alternative specifications of the stochastic (2016) suggests that more rapid technological production frontier, including some that use survey change at the upper end of the farm size distribution Box continues next page 34   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n BOX 2.1 Does technological change benefit small and large farms equally? Evidence from Mexico (continued) indicates an advantage for some larger farms in har- ­ epresentative of both rural and nonrural Mexican is r nessing more modern agricultural practices. And yet households. Thus use of the data set to study Mexican ­ this advantage has not been widespread enough to agriculture must include the important caveat that it likely underrepresents the larger commercial agricultural translate into higher TFP because of the inability of operations to the degree that they are not family farms. the nonfrontier households to keep up. A comparison with the 2007 Agricultural Census reveals that in both the census and MxFLS less than 5 percent of a. For more detailed information on the MxFLS farms are larger than 50 hectares. However, these “large” ­ composition, longitudinal panel nature of the data, farms are not necessarily the same as those in the census ­ representativeness, and sample size and ­characteristics refer because they are family-run farms and do not include to Helfand and Taylor (2016). Although not ­ representative corporate-run commercial agricultural operations. ­ exican agriculture sector per se, the MxFLS of the M b. For details, see annex 1 in Helfand and Taylor (2016). important role in teaching farmers how to Based on their comprehensive review of improve their productivity and in moving the literature, Anderson and Feder (2003) the products of research—typically infor- conclude that the record of the impacts of mation and technical innovations—from the extension on farm performance is actually laboratory to the field. Anderson and Feder quite mixed. (2003) argue that productivity improvements Finance. Finance can be used by farmers are possible only when there is a gap between to bring input levels closer to the optimal actual and potential productivity. They level, allowing them to approach the pro- describe two types of “gaps” that contribute duction frontier and increase productivity to the productivity differential: the technol- and production. A large body of empirical ogy gap and the management gap. Extension evidence from around the world shows that can help to reduce the differential between improved access to finance is associated with potential and actual yields in farmers’ fields increased technical efficiency and higher by accelerating technology transfer (reducing productivity in agriculture. For example, the technology gap) and by helping farmers based on a review of more than 30 studies become better farm managers (reducing the from 14 developing countries, Bravo-Ureta management gap). and Pinheiro (1993) concluded that use of A large empirical literature documents credit has a positive and statistically signif- many cases in which extension services have icant impact on technical efficiency at the had a measurable impact on agricultural farm level. Espinoza et al. (2018) used sto- productivity (for summaries, see Alston chastic frontier analysis to explore sources of et al. 2000; Anderson 2007; Birkhaeuser, the variability in farm-level productivity and Evenson, and Feder 1989; Evenson 2000). efficiency in Peru. These authors concluded A  practical problem is that most studies that access to credit was associated with have examined the joint impacts of research reduced inefficiency. and extension because the two are often Interestingly, although there is abun- cofinanced and coimplemented. Relatively dant evidence that greater access to credit is few studies have assessed the impacts of associated with higher agricultural produc- extension services alone. Generalizing tivity and production in credit-constrained across the empirical literature, it is clear households, higher agricultural productivity that extension services can significantly and production do not always translate into accelerate agricultural productivity growth. higher net income. Carter (1989) found in The effect is not guaranteed, however. Nicaragua that, although the use of credit P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    35 had a positive impact on production, it did available. For example, when farmers face not translate into increases in net income the risk of unpredictable weather or erratic measured at market prices. rainfall, they may choose not to invest in Education. The positive impact of edu- high-yielding varieties or fertilizers that cation on agricultural production and could boost agricultural productivity. Using efficiency has been confirmed by many survey data from Honduras, Nicaragua, empirical studies. Education improves farm- and Peru, Boucher, Carter, and Guirkinger ers’ decision-making skills and enables (2008) show that in the absence of insur- them to choose a different mix of inputs ance to protect against losses, lenders tend and allocate resources more efficiently—the to pass on risk to borrowers, which results so-called allocative effect. Education also can in borrowers withdrawing from the credit have a “worker effect” or “technical effect” market. This outcome reduces investment in which farmers are simply able to use a by farmers and negatively affects agricul- given amount of resources more efficiently tural productivity. (Reimers and Klasen 2013). Accord ing to evidence from L atin Empirical studies have documented how America, the demand for agricultural insur- more years of schooling frequently result ance is strong, and farmers with access to in higher levels of agricultural produc- insurance tend to engage in larger agricul- tion. Reimers and Klasen (2013) analyzed tural investments. Moreover, farmers with the impact of education on agricultural insurance make riskier production choices productivity across 95 developing countries than those who do not have insurance—that from 1961 to 2002 and found a 3 percent is, they invest more in the face of uncertainty. increase in agricultural productivity for each Because it gives farmers an incentive to take additional year of schooling. on more risk, insurance leads to the adoption Risk management. Risk is associated of technology, resulting in higher productiv- with all production processes, especially in ity and returns over the long run. agriculture. Risk stems from uncertainty, Connectivity. The effect of connectivity which originates in imperfect knowl- on agricultural productivity has received edge. Risk consequently can be thought more attention in recent years. Helfand of as exposure to uncertain consequences and Levine (2004) studied the determi- that result from imperfect knowledge nants of productive efficiency in agricul- (Hardaker et al. 2015). In agriculture, ture in Center-West Brazil and found that imperfect knowledge can apply to many access to markets facilitated by new infra- factors, including agroclimatic condi- structure is an important determinant of tions, market conditions, policy regimes, agricultural  efficiency. Where produc- and the behavior of key players. A large ers are isolated, transportation-induced empirical literature has found that risk transaction costs depress productivity by aversion is common among all groups of altering relative prices in such a way that farmers, especially among smallholders input use is reduced—for evidence from who have few resources on which to rely Latin America, see, for example, Goyal and in time of production shortfalls. Level of González-Velosa (2012) and Calderón and income and various socioeconomic vari- Servén (2010). ables typically influence farmers’ attitudes Transportation-induced transaction costs to risk, which in turn affect their adoption can also affect productivity by influenc- of technology and therefore productivity. ing crop choice. High transportation costs In the absence of well-functioning insur- push farmers to grow food crops that can be ance markets, farmers often have difficulty stored easily instead of perishable cash crops sharing or pooling risks. As a result, they such as vegetables and fruits that cannot may choose to invest less (or differently) be sold easily in distant markets. Stifel than they would have done were insurance and Minten (2008) found that farmers far 36   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n from markets grow more low-value staple sec tors have consistent ly fou nd t hat crops than high-value perishable cash countries that invest more in agricultural crops. Switching from high-value fruits and R&D achieve higher agricultural produc- vegetables to lower-value cereals and pulses tivity growth (Craig, Pardey, and Roseboom tends to depress agricultural productivity. 1997; Evenson and Fuglie 2009; Evenson Lack of connectivity can also affect and Kislev 1975; Thirtle, Lin, and Piesse agricultural productivity through a third 2003). Fuglie et al. (2020) summarize the channel—amplifying price variability in iso- results of studies that econometrically lated areas, thereby forcing farmers to adopt estimated the impact of R&D on agricul- coping mechanisms that lead to lower pro- tural TFP growth in one or more devel- ductivity. In isolated areas where farming oping countries. The elasticities appear to households may have few opportunities to show systematic variation in the elastici- diversify their income sources with off-farm ties of R&D among regions. R&D spill- activities, farmers who know that prices for ins from national R&D systems of other agricultural products will be low during countries appear to be relatively unim- the postharvest period and high during the portant for developing countries, unlike in subsequent “hungry season” may insure developed countries where cross-country themselves by expanding production to technology transfer has been found to be less fertile land and investing less in inputs, significant (Fuglie 2018; Schimmelpfennig reducing agricultural productivity. and Thirtle 1999). A possible explanation is that agricultural R&D in developing Pushing out the frontier countries may be more location-specific A second potential source of agricultural pro- (Fuglie 2018). Latin America may be an ductivity growth is expansion of the produc- exception, however. Along with national tion possibilities frontier. What are the entry R&D, international R&D spill-ins and pri- points for helping to push out this frontier, vate R&D appear to have made significant and how effective are they? Two in particular contributions to agricultural productivity stand out: innovation and education. growth in the LAC region. Innovation. Innovation that produces Because R&D spending is usually only changes in technology is a major factor a small fraction of agricultural GDP, the driving technological change leading to marginal benefits implied by the elasticities TFP growth in agriculture. However, inno- of each dollar of R&D spending tend to be vation is difficult to define and measure large. Many studies report internal rates of because successful innovation has multiple return for public agricultural research spend- ingredients, including new technology, an ing. They compare costs to benefits, taking effective technology transfer mechanism, into account the lag time between invest- a target population with the requisite ment in R&D and its effect on productivity. knowledge and skills needed to take up In a meta-analysis of returns to agricultural the innovation, availability of associated research, Alston (2010) found that public inputs, and favorable economic incentives. agricultural research in developing countries Despite the inherently complex nature of earned a median internal rate of return of innovation, it is clear that a major driver 39 percent. More recent work by Hurley, of innovation is research and development Rao, and Pardey (2014) using a modified (R&D), and the ability of R&D to boost internal rate of return suggests that although productivity in agriculture has received returns have not been as high as had long much attention. been reported (median of 9.8 percent a year), In agriculture, the evidence linking they are still substantial. investment in R&D to productivity growth Education. Education plays a ­ significant is compelling. Studies comparing the long- role in the adoption and use of technological term performance of national agriculture advances in ag ricu lt u ral produc tion. P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    37 Educated farmers are more likely to adopt this finding to a dominant role for agricul- technology, thereby contributing positively tural growth in poverty reduction has waned, to productivity growth. Specifically, among however. The share of agriculture in most smallholder rice farmers in Bangladesh developing economies has declined; econ- an additional year of schooling shifted the omies have become more open as the result rice production frontier by 3–7 percent of globalization; and food has become more (Asadullah 2009). Education of farmers not tradable. Many policy makers believe that only enhances agricultural productivity fol- because productivity in agriculture, especially lowing technological adoption as discussed smallholder agriculture, is now so low com- in getting to the frontier, but also promotes pared with productivity in other sectors and adoption itself by creating more informed because food is sufficiently tradable, poverty producers. reduction is much more likely to come from More investment in research can have urbanization. In the LAC region, this view a n a mpl i f y i ng ef fec t on ag ricu lt u ra l has been reflected in policies designed to facil- productivity when paired with higher levels itate migration out of agriculture, promote of education. Evenson and Fuglie (2009) industrialization, place greater reliance on found that education without improve- food trade, and transform the agriculture ments in research capacity is not associated sector by introducing mechanized large-scale w it h i ncreased produc t iv it y g row t h. farming. Furthermore, in a study of eight East Asian This view may not be equally relevant economies Luh, Chang, and Huang (2008) to all countries, however. Recent work found that domestic R&D and its inter- summarized in Christiaensen and Mar- action with human capital have the most tin (2018) suggests that it may not always significant effects on progress in agricul- be correct to assume that when it comes to tural technology. This finding suggests that reducing poverty, growth from any sector t he generat ion a nd d issem i nat ion of has the same effect. Citing results from a improved technologies should be coupled coordinated series of studies that used differ- with farmer education to have a maximum ent methodological approaches, the authors impact on agricultural productivity. argue that growth from agriculture is in general two to three times more effective at reducing poverty than equivalent growth Agricultural growth and generated outside agriculture. An import- poverty reduction ant caveat, however, is that even though the Policy makers are interested in productivity advantage of agriculture over nonagriculture because productivity growth drives growth in reducing poverty is large for the poorest of the overall economy, resulting in higher in society, the effect diminishes as incomes incomes, reduced poverty, and improved rise and ultimately disappears as countries welfare. In considering the process of struc- become richer (figure 2.8). The implication tural transformation, which is characterized is that promoting agricultural growth can by shifts in resources between sectors, it is be particularly effective as a strategy for therefore relevant to ask: Does the sector reducing poverty in low-income countries, in which growth occurs matter for poverty but it will be less effective in middle-income reduction? This question is especially relevant countries and relatively ineffective in in developing countries in which a large pro- high-income countries. portion of the population lives in poverty. I n interpreting this finding that as Many empirical studies have concluded that econom ies develop and the relatively growth in agriculture has been more effective g reater ef fec tiveness of g row t h from in reducing poverty than growth outside agri- agriculture at reducing poverty declines, it culture, especially in closed economies where is important to keep in mind that the work food is not tradable. The policy relevance of summarized by Christiaensen and Martin 38   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 2.8  Effectiveness of growth in different sectors for Latin America is the relative pace of at reducing poverty productivity growth in the agriculture, industrial, and services sectors. With this 1.0 in mind, this section evaluates productivity dynamics in industry and services in Latin 0.5 America, benchmarking it against the dynam- ics observed in advanced economies and spec- 0 ulating about sources of future growth. Measuring economic activity is an intrin- Poverty change –0.5 sically difficult task, even more so when it concerns the services sector. How does one –1.0 appropriately account for the value added of services that do not operate through a –1.5 market transaction (government services, the digital economy)? What types of busi- ness services are offered in Latin America –2.0 compared with those in more advanced economies? Are they similar? Because of –2.5 190 760 3,040 the increasing participation of the services sector in aggregate production, the conse- Percent of GDP per capita quences of these challenges are even more Agriculture Log. (Agriculture) pronounced. Industry Log. (Industry) Another layer of difficulty lies in the Services Log. (Services) lack of data needed to compute total ­ factor Source: Ivanic and Martin 2018. productivity by sector and across coun- tries. The main data source used for this type of analysis, the Groningen Growth (2018) focuses on growth generated only and Development Centre’s 10-Sector Data- by primary agriculture. To the extent base, offers internationally comparable t h at g row t h i n pr i m a r y a g r ic u lt u re data on real value added and employment, generates value added and employment making it suitable for the task of measur- through forward and backward linkages, ing labor productivity. However, it does not the impacts on poverty reduction would offer information on sectoral capital stocks. decline more slowly as economies develop. Hereafter, then, every reference to sectoral Although it remains uncertain, the fact productivity will refer to the dynamics of that as incomes rise the demand increases labor productivity. for processed foods and meals eaten away from home—creating new jobs in the food processing and food services ­ i ndustries— Labor productivity in services and suggests that growth from agriculture may industry: Latin America and the United still have a relatively large effect in terms States of reducing poverty even in middle- and A natural starting point for assessing the high-income countries. performance of Latin America in terms of the productivity growth of its industrial and services sector is to compare it with the Productivity in industry and performance of advanced economies. ­ services The first step is to evaluate the produc- An insight stressed throughout this report tivity dynamics of sectors individually. Are is that an important driver of the patterns sectors in the LAC region catching up with of structural change observed in the data the global frontier or lagging further? In the P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    39 second step, because this relative pace of d ­ isappointing is the performance of Mexico, growth guides the sectoral allocation of especially after 1980, when a healthy pace resources in theory, it is instructive to con- of technological upgrading in both sectors trast the performance of Latin America and was interrupted, leaving it on track to be one the United States in terms of the relative of the worst performers in the region. The growth of the productivity of the industrial most worrisome case is Bolivia, whose level and services sectors. of productivity declined over the course of six Figure 2.9 reports the dynamics of labor decades. productivity in the industrial and services T he underperformance of many of sectors for Latin America and the United Latin America’s services sectors is also States between 1950 and 2010, measured as wor r i some i n v ie w of t he fa st pac e value added per worker. The salient prop- of dei ndust ria l i zat ion i n t he reg ion. erty of the  figure is that, although there Although there is scope for slowing down is some evidence of convergence in indus- deindustrialization by addressing the trial productivity, the performance of the d istortions that underlie it, the prospects ­ services sector in Latin America is wors- of sustaining growth accelerations in the ening relative to that of the United States. region becomes grimmer because the lower Chile and Brazil, the best performers, pace of productivity growth in services managed to outpace the United States in precipitates the reallocation of economic terms of industrial productivity growth, activity toward that slow-growing sector. indicative of convergence. However, even In short, a reversal of the disappointing for these best performers of the region, the rate of growth in the services sectors will pace of productivity growth in the services play an important role in Latin America sector could not keep up with productivity moving up from its designation as a sticky, growth in the United States. Surprisingly low-middle-income region. FIGURE 2.9  Labor productivity growth in industrial and services sectors: Latin America and United States, 1950–2010 a. Industrial sector b. Services sector 4 4 Value added per worker (1950 = 1) Value added per worker (1950 = 1) 3 3 2 2 1 1 1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010 United States Argentina Bolivia Brazil Chile Mexico Peru Source: Fattal Jaef 2019. 40   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n Relative prices: Relative productivity Sources of productivity growth: The growth between services and industry potential of improving allocative efficiency Evaluation of the productivity dynamics of each sector independently is useful to fully One way or the other, every evaluation appreciate that underwhelming productivity of the long-term behavior of an economy growth is a widespread phenomenon in Latin ends up discussing the conceivable drivers America. In terms of helping understand of productivity growth. How will Latin structural change, however, it is the relative America improve the productivity of its pace of productivity growth between sectors services sector? Among the long list of can- that acts as a driving force. For this reason, didates is the role of allocative efficiency, figure 2.10 illustrates how Latin America which is the focus here. What is the scope has fared in relation to the United States in in Latin America for raising total factor terms of the pace of the relative productivity productivity by means of allocating pro- growth of the industrial and services sectors. ductive resources more efficiently across Although it is a shared feature of almost firms? every country that productivity growth in T he ju s t i f ic at ion for fo c u si ng on industry outpaces services, this pattern is misallocation lies primarily in that it not most pronounced in Latin America. Inter- only constitutes a conceptually plausible preted from standard theories of structural source of productivity growth in the future, change, this relative decline in services sector but also has been shown to constitute a productivity is a primary driver of a reallo- barrier to productivity growth in many cation of economic activity toward that sec- economies. Because most studies focus on tor. It is this driving force, combined with manufacturing productivity, it is useful to the lethargic pace of productivity growth in revisit the literature and uncover findings services, that substantiates the concern about on misallocation in the services sector, long-term growth in the region. especially in Latin America. To appreciate the mechanisms through which resource misallocation harms total factor productivity, consider the follow- FIGURE 2.10  Labor productivity in services sector relative to ing scenario. Two producers, A and B, industrial sector: Latin America and United States, 1950–2010 provide an identical service, albeit using different technology. For a given number 1.2 of hours worked, producer A can supply (ratio of services sector to industrial sector) more units of the service than producer B. 1.0 Subject to diminishing marginal products, Value added per worker 0.8 the output maximizing rule will allocate workers across firms until their marginal 0.6 products are equalized. Because producer A is more productive than producer B, 0.4 A  will ultimately operate a larger firm. Now suppose a distortion in the e ­ conomy 0.2 interferes with the efficient rule. For exam- ple, bigger firms may be subjected to higher 0 tax rates than smaller firms. This policy 50 60 70 80 90 00 10 would discourage producer A from achiev- 19 19 19 19 19 20 20 Brazil Chile Argentina Peru ing an efficient size and force a cut in the United States Mexico Bolivia labor force to below the optimal level. Despite some wage adjustment because of Source: Fattal Jaef 2019. the excess supply of labor, the equilibrium P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    41 allocation will result in fewer workers between 10 percentage points (Chile) and for producer A and more for producer B. 50 percentage points (Mexico). Because of the productivity differences This report takes on the question of what across producers, and for a given size of the countries could do to alleviate misalloca- labor force, lower output will be the out- tion. In the meantime, it is instructive to come in the aggregate. recall that if countries were to find a way The logic just described provides a to reap the gains from reversing misalloca- strategy for measuring the degree of mis- tion, the subsequent growth in industrial allocation in a country. Specifically, effi- productivity would, unless accompanied by cient allocation carries the strong tes­t able an equal or stronger force in services sector implication that, among comparable goods productivity growth, result in a deepening of and services (for example, in the same the reallocation of employment toward the four-digit Standard Industrial Classifica- services sector. On the one hand, growth in tion industry code), the value of marginal industrial productivity increases aggregate products across firms should be equal- income. Through income effects, expenditure ized. Otherwise, workers could be real- moves away from agriculture to industry and located toward high marginal product services. On the other hand, the relative price firms and increase the aggregate amount channel, directly connected to the widening of output. Therefore, a sufficient statistic of the gap in productivity growth between of the degree of misallocation in a narrow industry and the services sector, further industry or sector is the standard deviation contributes to the reallocation of employ- of the marginal revenue products. Because ment to the services sector. Therefore, the of the increasing availability of firm-level data sets, it is now possible to measure TABLE 2.1  Misallocation in manufacturing, the deviation between actual and efficient selected developing and developed countries allocation from the data. Standard deviation, TFP gain, revenue ­efficient How pervasive is misallocation in Country productivity ­allocation (%) industry and services? United States 0.49 42.9 Bowing to data constraints, most of the China 0.63 86.6 literature on misallocation focuses on India 0.67 127.5 manufacturing industries. The evidence Colombia 1.21 50.5 thus far provides compelling evidence that misallocation is prevalent in the developing Venezuela, R.B. 1.28 64.7 world and is preventing these economies from El Salvador 0.64 60.6 reaping substantial gains in TFP. Table 2.1 Chile 0.72 53.8 summarizes the statistics on misallocation Uruguay 0.97 60.2 for several countries in Latin America and Bolivia 0.88 60.6 Sub-Saharan Africa, alongside the counter- factual gains in TFP that would accrue if the Ecuador 0.62 57.6 efficient allocation were to be implemented. Argentina 0.62 60.0 As a baseline, the table also shows misalloca- Mexico 0.82 95.0 tion for the United States.2 Ethiopia 0.78 66.6 The main message conveyed by the table Ghana 0.95 75.7 is that the manufacturing misallocation in Latin America is almost as severe as it is Kenya 1.52 162.6 in China, India, and many Sub-Saharan Côte d’Ivoire 0.65 31.4 economies, and the potential TFP gains from Sources: Data, United States, India, and China: Hsieh and Klenow (2009); reverting misallocation to the US level range Sub-Saharan Africa: Cirera, Fattal-Jaef, and Hibret Maemir (2020). Note: TFP = total factor productivity. 42   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n degree to which the reversal of misallocation so far. In those studies, misallocation is will contribute to structural transformation measured as deviations from a theoretically hinges critically on the pattern of misalloca- prescribed benchmark of efficiency, as in tion in the services sector. the pioneering work of Hsieh and Klenow What about misallocation in services? The (2009). Crespi, Tacsir, and Vargas (2016) evidence here is much thinner. Representa- base the indicator of allocative efficiency tive firm-level data sets on services sectors on Olley and Pakes (1996), which measures are scarcer than they are for manufacturing, allocative efficiency as the degree of correla- constituting the main hurdle for achieving tion between the value-­ added share of a firm a comparable density of research. Further- and its relative productivity with respect to more, the inherent difficulties in measuring that of the average firm in its industry. the value added of various services make Even using a different methodology, the measurement of misallocation more prone study by Crespi, Tacsir, and Vargas (2016) to measurement error, thereby hindering its confirms that a ­ llocative efficiency in Latin validity as a useful diagnostic tool. America is lower in the services sector than Still, based on the existing inquiries into in the industrial sector. This evidence points the role of misallocation in services, the once again to distortions in the business conclusion is that distortions in the ser- environment that have a disproportion- vices sector seem to be more prevalent and ate effect on firms in the services sector. damaging than they are for the industrial This distinction is important in thinking sector. Dias, Marques, and Richmond (2019) about what type of friction or policy may estimated misallocation and the associated be behind the observed misallocation. For counterfactual TFP gains from its resolution example, if credit market frictions are the in Portugal’s economy between 1996 and culprit, then it must be that external finance 2011. They found that the potential for TFP dependence is more prevalent in services growth through efficient reallocation is about than in industry. Another candidate is state- twice as large in the services sector as in the owned enterprises, which tend to be more industrial sector. common among services. A similar quantification was performed by Garcia-Santana et al. (2016) for Spain’s econ- omy between 1995 and 2007. In this case, Taking stock: The scope for the construction sector exhibited the lowest raising allocative efficiency and allocative efficiency, but again misallocation the expected pace of structural was worse in the services sector than in the change industrial sector. This chapter has revealed that there is scope The conclusion about the relative severity in Latin America for raising total factor of misallocation in services versus industry productivity by improving the allocative was also confirmed in the Latvian economy efficiency in an economy. Furthermore, during the financial crisis and its aftermath, a review of the existing evidence points as reported by Benkovskis (2015). toward a higher prevalence of misalloca- Turning to Latin America, the evidence tion in the services sector than in the indus- is confined to a study by Crespi, Tacsir, trial sector. To connect back to the primary and Vargas (2016). The authors perform objective of this study, structural change, a more general investigation of the drivers what is the implication of the observed mis- of the low level of productivity in Latin allocation for the future pattern of sectoral America. A  factor here, however, is that resource allocation? the methodology for the characterization Answering the question requires going of misallocation underpinning the work by back to the fundamental drivers of struc- Crespi, Tacsir, and Vargas (2016) is different tural change in a country. As discussed ear- from that used in the studies summarized lier in the presentation of the theoretical P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    43 framework, these drivers stem from the in developing countries. Specifically, the incentives to move expenditures toward introduction of new labor-saving technolo- income elastic goods and services as countries gies is reducing the importance of low wages become richer and the incentives to reallocate as a determinant of comparative advantage. resources toward sectors with relatively lower In other words, labor costs are becoming less productivity growth. important, whereas quality, reduced time to The fact that misallocation is severer in market, faster innovation, and scale econo- services than in industry implies that there mies are becoming more relevant. is a channel through which productivity in New technologies are enabling suppliers services will catch up with productivity in to produce higher-quality goods at lower the industrial sector. In the short to medium prices, and thus suppliers using older tech- term, as misallocation is progressively nologies will need to adapt or they will reversed, theory implies that deindustrializa- not survive. However, the adoption of new tion will slow and aggregate growth will go technologies requires complementary invest- up. In the long run, however, once the gains ments in infrastructure (particularly in ICT from efficient reallocation have been reaped, technologies) and human capital, as well the pace of deindustrialization and the long- as modernization of the regulatory system run growth in the economy will again be to address issues of intellectual property, determined by the long-run forces driving privacy, and ownership of data. Mean- productivity growth in each sector. Unless while, the expectation is that global value the resolution of misallocation translates into chains (GVCs) will shorten, and there will a permanent change in the rate of technolog- likely be fewer entry points in the future ical progress, the boost in industrial activity Hallward-Driemeier and Nayyar 2018). (­ and aggregate growth will be temporary. In their book, Hallward-Driemeier and This permanent effect of dismantling Nayyar (2018) analyze in detail the feasibility misallocation on productivity growth is not of expanding production by industrial sub- an unreasonable possibility. It is quite possible sector. In their analysis, they consider aspects that, once firms are confronted with distor- such as the relative magnitude of automa- tions that damage their profitability, not only tion (measured by the density of robots per will resources flow out of the most productive 1,000 workers), export concentration, service firms (static misallocation effect), but also intensity, and the extent to which goods in firms will be more reluctant to invest in inno- a subsector are internationally traded. What vations that would make them even more pro- follows are the two conclusions most relevant ductive (dynamic effect through innovation). to the LAC economies. In short, at the very least the current First, despite the changing globaliza- relatively low levels of allocative efficiency tion patterns and the emergence of new in the services sector are an opportunity to labor-saving technologies, in some man- slow down the deindustrialization of Latin ufacturing industries there is room for American economies and boost aggregate insertion or expansion. Examples are com- growth. The perpetuation of these trends will modity-based processing manufacturers depend to a large extent on the credibility of that are less automated, less concentrated in the reforms that are implemented to boost terms of export locations, and less intensive allocative efficiency and that would induce in the use of professional services. Also, for firms to innovate and invest in technology. industries such as textiles, garments, and footwear, which are labor-intensive and tradable, countries with low unit labor costs The future in manufacturing may retain a comparative advantage. There The new technologies of the so-called Fourth also may be scope to serve domestic or Industrial Revolution are threatening the regional markets for lower-quality, lower-­ potential for large-scale industrialization price manufactures across industries. 44   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n Second, Hallward-Driemeier and Nayyar in the services sector grow faster than those highlight the potentially negative effects with higher initial labor productivity in that of not adopting new technologies. If new sector (Enache, Ghani, and O’Connell 2016; production methods in traded goods render Kinfemichael and Morshed 2016). More- higher-quality goods at lower prices, domes- over, Fagerberg and Verspagen (2002) sug- tic production using older technologies may gest that the services sector has increasingly not be able to compete. This may result contributed to economic growth over the in fewer jobs created or even job losses. last 30 years. Therefore, firms may need to adopt new These results are related to the fact that technologies just to remain globally com- the services sector is increasingly sharing petitive. The authors conclude that “man- pro-development characteristics that were ufacturing will likely continue to deliver once thought of as the unique domain of on productivity, scale, trade, and innova- manufacturing. The huge advances in ICT tion, but just not with the same number of technologies have enabled the emergence jobs. So, its unique desirability in terms of of service subsectors—financial, telecom- the twin wins of productivity and jobs is munications, and business services—that eroding.” can be digitally stored, codified, and more easily traded (Ghani and Kharas 2010). The future in services Meanwhile, the deregulation of services markets has been accompanied by large The traditional view generally holds services inf lows of foreig n direct investment. as an inferior sector that has low productiv- Therefore, certain service subsectors are ity and, perhaps more important, lower pro- looking more and more like the manufac- ductivity growth. As a result, the structural turing sector, with exposure to trade and transformation process that increases the inflows of FDI allowing greater competi- importance of the services sector appears tion, technology diffusion, and the benefits to be terrible news for the region because it of scale. implies a slowdown of aggregate productiv- It is important to note, however, that ity growth. This is known in the economic these service subsectors, which can sub- literature as Baumol’s disease. stantially contribute to increasing produc- Taken as a whole, the services sector does tivity, are also highly skill-intensive. Thus appear to have lower productivity growth their capacity to provide employment for than the industrial sector. However, the ser- unskilled labor may be limited. However, vices sector is composed of a very diverse some service subsectors are intensive in the set of subsectors that differ significantly in use of unskilled labor. Unfortunately, these their productivity levels, in their productiv- subsectors are generally low-productivity ity growth, and even in their use of skilled growth sectors and thus will contribute less labor. In fact, a more disaggregated view to aggregate productivity. of the services sector reveals huge hetero- As noted earlier, Baumol’s disease refers to geneity in which some subsectors are more the phenomenon in which structural change productive and skill-intensive than man- slows down aggregate productivity growth ufacturing. Specifically, there is evidence when it reallocates production to industries that service industries that are intensive in with low productivity growth—see, for knowledge, ICT, and trade, such as telecom- example, Baumol (1967); Nordhaus (2008); munications, finance, and logistics. have and Oulton (2001). The question that follows higher rates of productivity growth than is whether in the future these industries will manufacturing (Jorgenson and Timmer gradually take over the economy and drive 2011). In fact, recent evidence suggests that down aggregate productivity growth. In a there is unconditional convergence; coun- recent paper, Duernecker, H ­ errendorf, and tries with lower initial labor productivity Valentinyi (2017) add a novel feature to the P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    45 standard structural transformation model; activities such as branding and advertising specifically, they disaggregate the services services contribute a large share of total sector into services with high productivity value added, while the intermediate produc- growth and services with low productivity tion stages contribute the least. growth. This approach is a deviation from As noted earlier, the productivity of the literature, which typically considers one services used as inputs in production (such broad services sector and abstracts from the as for design and marketing) or as enablers heterogeneity in service industries. Although for trade (such as logistics and e-commerce the model by Duernecker, Herrendorf, and platforms) are essential to the competitive- ­ Valentinyi (2017). generates the usual struc- ness and growth of the industrial sector. tural change between the goods and services The estimates of Sinha (2019a) suggest that sectors, it also implies structural change if distortions of services as intermediate within the services sector itself. They find goods (embodied services) had been kept that for the postwar United States the cal- at their historical minimum, the industrial ibration of the utility function implies that sector would have been 2–2.5 percentage services with low productivity growth are points larger as a share of the economy. luxuries, high-­ p roductivity services are Moreover, the value added of embodied necessities, and the two service subsectors services, specially distribution and busi- are substitutes. This substitutability between ness services, have contributed more than the two service subsectors limits the impor- a third of the value of gross manufactur- tance of the low-­ productivity subsector in ers’ exports globally (Hallward-Driemeier the economy and thus the future productiv- and Nayyar 2018). According to a grow- Baumol’s disease. ity effects of ­ ing body of evidence, this servicification of manufacturing has raised manufacturing productivity in the Czech Republic, India, Blurring lines and Sub-Saharan Africa (Arnold, J ­ avorcik, Another important trend worth noting is and Mattoo 2011; Arnold, Mattoo, and the “servicification” of manufacturing. Narciso 2008). This term refers to the fact that manu- Recent literature has highlighted the key facturing firms are not only integrating role of services as a supplier of inputs to more services into their production func- the rest of the economy. In fact, Alvarez et tion, but also selling and exporting more al. (2019) find that in the LAC region this services as integrated activities. It is useful sector has the highest degree of forward to distinguish these two aspects of servici- linkages and the highest degree of influ- fication. On the one hand, the increasing ence. Moreover, they find that if several use of services as inputs in the production subsectors in services—such as business process is described as services embodied services, trade, and transport—closed their in goods. On the other hand, embedded productivity gap relative to the Organ- services are those that are bundled with the isation for Economic Co-operation and goods provided to customers, such as sales Development (OECD) average, it would and after-sales services. generate the greatest contribution to aggre- These services are increasingly accounting gate productivity. for much of the value added in a prod- uct’s supply chain. Stan Shih, Acer’s CEO during the 1990s, has described the rela- Conclusions and policy tionship between the stages of production implications and the contribution to total value added as T his chapter has reviewed productiv- a “smiley curve.” Essentially, he is referring ­ ector. ity performance and dynamics by s to the fact that upstream activities such as Although agricultural productivity has product design and R&D and downstream increased in the L AC region, there is 46   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n still room for further improvement. Two of data on the services sector is an obsta- main avenues for productivity growth are cle to a clearer diagnosis, the existing evi- (1)  improving the technical efficiency of dence indicates that there is a higher degree producers using existing technology and of misallocation in this sector relative to (2) pushing out the production possibilities manufacturing. frontier by shifting to new, improved G overn ments should also focus on technologies. increasing competition in the services As for the industrial sector—specifically, ­ s ector by removing distortions in the mar- the manufacturing subsector—the evidence ket and by opening these sectors to interna- clearly shows that there is a substantial degree tional trade. Figure 2.11 shows the results of misallocation between firms. The firm- of applying the Services Trade Restric- size distribution is skewed toward small and tions Index to nine LAC countries. 3 The microenterprises. This finding points toward index measures the degree to which coun- distortions in the market that are preventing tries are open to international trade in five the consolidation and growth of the most service subsectors: telecommunications, productive firms. Thus governments need to financial, transportation, retail, and pro- institute policies that foster competition— fessional services. A score of zero indicates such as international trade and deepening that the country is completely open, and a of regional trade agreements. Revisions are score of 100 implies that it is ­ c ompletely needed as well of size-dependent policies closed. For the region, it appears that tele- (or enforcement) that appear to hamper the communications and professional services growth of productive firms and incentivize are the most restricted sectors, whereas informality. transportation is also relevant for some The future of further industrialization countries. is subject to growing requirements for Recent literature suggests that the services complementary infrastructure, technology sector is heavily intertwined with the rest absorption capacity, and workforce skills. of the economy and is the most important Increasingly, firms will need to adopt new sector in terms of being a supplier of inputs. technologies just to stay competitive. The In addition, the recent trend of “servicifica- introduction of new labor-saving technolo- tion” of manufacturing implies that more gies is reducing the importance of low wages services are being used as inputs in the pro- as a determinant of comparative advantage. duction of goods (embodied services) and Instead, quality, reduced time to market, more services are provided to customers faster innovation, and scale economies are bundled with the goods (embedded services). becoming more relevant. Moreover, emerg- T herefore, the increased productivit y ing technologies (such as 3D printing) are of backbone services—such as logistics, expected to shorten global value chains, ICT, and business services—could ripple limiting the opportunities for entry. There- throughout the economy, having larger fore, opportunities for further industrial- impacts on overall aggregate productivity. ization (or reindustrialization) may be more In fact, as noted, Sinha (2019b) finds that limited and subject to higher requirements reductions in the cost of service inputs could in the future. have quantitatively important effects on the Urgently needed are a comprehensive size of the industrial sector. set of policy actions to address productiv- Figure 2.12 presents the performance of ity issues in the services sector. The sector LAC countries relative to the global best already employs more than 60 percent of performer on the Logistics Performance the workforce in the LAC region, and cur- Index.4 It shows that there is significant rent trends indicate that it will continue to room for improvement across all countries grow and be the main source of job cre- in the L AC region. In particular, the ation in the future. Although the scarcity region  can significantly improve on the P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    47 FIGURE 2.11  Services Trade Restrictions Index, selected LAC countries a. Argentina b. Bolivia c. Brazil Overall Overall Overall 60 50 50 50 40 40 40 Financial Financial Financial Professional 30 Professional 30 Professional 30 20 20 20 10 10 10 0 0 0 Transportation Transportation Transportation Telecommunications Telecommunications Telecommunications Retail Retail Retail d. Chile e. Colombia f. Honduras Overall Overall Overall 50 50 50 40 40 40 Professional 30 Financial Professional 30 30 Financial Financial 20 20 Professional 20 10 10 10 0 0 0 Transportation Telecommunications Transportation Transportation Telecommunications Telecommunications Retail Retail Retail g. Mexico h. Nicaragua i. Peru Overall Overall Overall 70 50 50 60 40 50 Financial Financial 40 Professional 40 Professional 30 Professional 30 Financial 30 20 20 20 10 10 10 0 0 0 Transportation Transportation Telecommunications Transportation Telecommunications Telecommunications Retail Retail Retail Source: Original calculations for this publication using World Bank’s Services Trade Restrictions Database (https://www.worldbank.org/en/research/brief/services-trade-restrictions​ -database). Note: The graphs depict information on five sectors: financial, telecommunications, retail, transportation, and professional services. All datapoints were collected in 2008, except for Mexico and Brazil (2011). The quantitative interpretation of the numbers is as follows: open without restrictions (0 points); virtually open (25 points); existence of major/nontrivial restrictions (50 points); virtually closed (75 points); completely closed (100 points); LAC = Latin America and the Caribbean. customs, infrastructure, and logistics qual- the size of the sector, the expectation that ity components. it will continue to grow, its higher degree Therefore, policy makers in the LAC of misallocation (relative to the industrial region should focus on productivity growth sector), and its role as input provider to and not on the size of any one economic the rest of the economy, the productivity sector. There is room for improvement in agenda for the services sector should be a all sectors of the economy, but a dedicated priority for policy makers in the region. reform agenda is urgently needed for As for the future generally, the structure the services sector. Resources should be of the LAC economy is changing, and the invested in data collection to better under- requirements for productivity growth within stand the specific issues affecting the pro- sectors are increasing. In particular, the ductivity of firms in this sector. Because of demand for skills is changing. What are the 48   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 2.12  Logistics Performance Index and its components: 16 LAC countries relative to best performer 0.20 0.15 Change in relative performance 0.10 0.05 0 -0.05 -0.10 -0.15 LPI Customs Infrastructure International Logistics quality Tracking Timeliness shipments Argentina Bolivia Brazil Chile Colombia Costa Rica Dominican Republic Ecuador Guatemala Honduras Mexico Panama Peru Paraguay El Salvador Uruguay Source: Original calculations for this publication using World Bank’s Logistics Performance Index, 2018 (https://lpi.worldbank.org/) for LAC economies. Note: All values indicate the relative score of any given economy using Germany as a benchmark. LPI refers to the composite Logistics Performance Index; customs refers to the effi- ciency of customs and borders clearance; infrastructure refers to the quality of trade and transport infrastructure; international shipments refers to the ease of arranging competitively priced shipments; tracking refers to the ability to track and trace consignments; timeliness refers to the frequency with which scheduled or expected delivery arrives within expected delivery times; LAC = Latin America and the Caribbean. implications of these changes for jobs and in excess of the US gains are the actual gains the future of work? The next chapter turns from resolving misallocation. This method to these questions. would control only for the measurement error that does not vary systematically across countries. 3. For more information on the index, see Notes https://www.worldbank.org/en/research/brief​ 1. Agriculture refers to the production from /services-trade-restrictions-database. crops, livestock, forestry, and fisheries. 4. For more i n for m at ion on t he i nd ex , 2. The expectation is that misallocation should see https://lpi.worldbank.org/about. be close to zero in the United States, presum- ably an undistorted economy. Table 2.1 reveals that there is indeed misallocation in the US References economy, albeit to a lesser degree. Noting that A lston , J. M . 2010. “T he B enef its f rom part of what the methodology captures as Agricultural Research and Development, misallocation could arise from measurement Innovation, and Productivity Grow th.” error, one could crudely control for measure- OECD Food, Agriculture and Fisheries Paper ment error by attributing all the misallocation No. 31, OECD Publishing, Paris. in the United States to m­ easurement error and Alston, J. M., C. Chan-Kang, M. C. Marra, subtracting the US numbers from those of the P. G . Pa rdey, a nd T. J. Wyat t. 20 0 0. rest of the countries. In other words, only the “A  Meta-analysis of Rates of Return to dispersion in revenue productivity in excess of Agricultural R&D: Ex Pede Herculem?” the dispersion in the United States is actual Research Report Vol. 113, International Food misallocation. Likewise, only the TFP gains Policy Research Institute, Washington, DC. P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    49 Álvarez, F., Eslava, M., Sanguinetti, P., Toledo, Birkhaeuser, D., R. E. Evenson, and G. Feder. M., Eslava, M., Alves, G., and others. 2019. 1989. “The Economic Impact of Agricultural RED 2018: Institutions for Productiv- Extension: A Review.” Center Discussion ity: Towards a Better Business Environ- Paper No. 567, Yale Economic Growth Center, ment. Caracas: CAF. http://scioteca.caf.com​ New Haven, CT. /handle123456789/1410. Bolt, J., R. Inklaar, H. de Jong, and J. Luiten Anderson, J. R. 2007. “Agricultural Advisory van Zanden. 2018. “Rebasing ‘Maddison’: Services.” Background paper, World Devel- New Income Comparisons and the Shape of opment Report 2008. http://­s iteresources​ Long-run Economic Development.” Maddi- . w o r l d b a n k . o r g l I N T W D R 2 0 0 8 I​ son Project Working Paper 10, Groningen ­Resources/27950871191427986785/Anderson​ Growth and Development Centre (GGDC), _AdvisoryServices.pdf. Groningen, The Netherlands. Anderson, J. R., and G. Feder. 2003. “Rural Boucher S. R., M. R. Carter, and K. Guirkinger. Extension Services.” Policy Research Working 2008. “Risk Rationing and Wealth Effects Paper 2976, World Bank, Washington, DC. in Credit Markets-Theory and Implications Arnold, J. M., B. S. Javorcik, and A. Mattoo. for Agricultural Development.” American 2011. “Does Services Liberalization Benefit Journal for Agricultural Economics 90 (2): Manufacturing Firms? Evidence from the 409–23. Czech Republic.” Journal of International Bravo - Ureta, B ., and A . Pin hei ro. 1993. Economics 85 (1): 136–46. “Efficiency Analysis of Developing Country Arnold, J. M., A. Mattoo, and G. Narciso. 2008. Agriculture: A Review of the Frontier Func- “Services Inputs and Finn Productivity in tion Literature.” Agricultural and Resource Sub-Saharan Africa: Evidence from Firm- Economics Review 22 (1): 88–101. Level Data.” Journal of African Economies Calderón, C., and L. Servén. 2010. Infrastruc- 17 (4): 578–99. ture in Latin America. Washington, DC: Asadullah, M. N. 2009. “Returns to Private and World Bank. Public Education in Bangladesh and Pakistan: Carletto, G., S. Savastano, and A. Zezza. 2013. A Comparative Analysis.” Journal of Asian “Fact or Artifact: The Impact of Measurement Economics 20 (1): 77–86. E r rors on t he Fa r m Si z e - Produc t iv it y Assunção, J. J., and L. H. Braido. 2007. “Testing Relationship.” Journal of Development Household-Specific Explanations for the Economics 103: 254–61. Inverse Productivity Relationship.” American Carter, M. R 1989. “The Impact of Credit on Journal of Agricultural Economics 89 (4): Peasant Productivity and Differentiation 980–90. in Nicaragua.” Journal of Development Barrett, C. B. 1996. “On Price Risk and the Economics 31 (1): 13–36. Inverse Farm Size-Productivity Relationship.” Christiaensen, L ., and W. Mar tin. 2018. Journal of Development Economics 51 (2): Agriculture, Structural Transformation and 193–215. Poverty Reduction: Eight New Insights. Barrett, C. B., M. F. Bellemare, and J. Y. Hou. 2010. Amsterdam: Elsevier. “Reconsidering Conventional Explanations Cirera, X., R. Fattal-Jaef, and H. Maemir. 2020. of the Inverse Productivity-Size Relationship.” “Taxing the Good? Distortions, Misallocation, World Development 38 (1): 88–97. and Productivity in Sub-Saharan Africa,” Baumol, W. J. 1967. “Macroeconomics of ­ conomic Review 34 (1): World Bank E Unbalanced Growth: The Anatomy of Urban 75-100. World Bank Group, Washington, Crisis.” American Economic Review 57 (3): DC. 415–26. Coelli, T. J., and D. P. Rao. 2003. “Total Benjamin, D. 1992. “Household Composition, Factor Productivity Growth in Agriculture: A Labor Markets, and Labor Demand: Testing Malmquist Index Analysis of 93 Countries, for Separation in Agricultural Household 1980 –2000.” Draft paper, University of Models.” Econometrica: Journal of the Queensland, Brisbane. Econometric Society 60 (2): 287–322. Craig, B. J., P. G. Pardey, and J. Roseboom. B en kovsk is , K . 2015. “M isa l lo c at ion of 1997. “International Productivity Patterns: Resources in Latvia: Did Anything Change Accounting for Input Quality, Infrastruc- during the Crisis?” Working paper, Bank of ture, and Research.” American Journal of Latvia, Riga. Agricultural Economics 79 (4): 1064–76. 50   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n Crespi, G., E. Tacsir, and F. Vargas. 2016. Journal of Development Economics 18 (2–3): “ Innovation Dynamics and Productivity: ­ 297–313. ­ E vidence for L atin A merica.” I n Fir m Fuglie, K. P. 2015. “Accounting for Growth in Innovation and ­Productivity in Latin America Global Agriculture.” Bio-Based and Applied and the C­ aribbean, 37–71. New York: Palgrave Economics 4 (3): 201–34. Macmillan. Fuglie, K. P . 2018. “R&D Capital, R&D Dias, D. A, C. Robalo Marques, and C. Richmond. Spillovers, and Productivity Growth in World ­ 2019. “A Tale of Two Sectors: Why Is Misallo- Agriculture.” Applied Economic Perspectives cation Higher in Services than in Manufactur- and Policy 40 (3): 421–44. ing?” Review of Income and Wealth, April 3. Fuglie, K. P., M. Gautam, A. Goyal, and W. F. Duernecker, G., B. Herrendorf, and A. K. Valentinyi. Maloney. 2020. Harvesting Prosperity Tech- 2017. “Structural Change within the Service Sec- nology and Productivity Growth in Agricul- tor and the Future of Baumol’s Disease.” Centre ture. Washington, DC: World Bank. for Economic Policy Research, London. Fuglie, K. P., P. Heisey, J. King, C. Pray, and Eastwood, R., M. Lipton, and A. Newell. 2010. D. Schimmelpfennig. 2012. “The Contribution “Farm Size.” Handbook of Agricultural of Private Industry to Agricultural Innova- Economics 4: 3323–97. tion.” Science, November 23, 1031–32. Enache, M., E. Ghani, and S. O’Connell. 2016. García- Santana, M ., E . Moral-B enito, “Structural Transformation in Africa: A His- J. ­Pijoan-Mas, and R. Ramos. 2016. “Grow- torical View.” Policy Research Working Paper ing Like Spain: 1995–2007.” Working Paper 7743, World Bank, Washington, DC. 16098, Bank of Spain, Madrid. Espinoza, M., R. Fort, M. Morris, A. Sebastian, Ghani, E., and H. Kharas. 2010. “The ­ S ervice and L . Villazón. 2018. “Understanding Revolution.” Brief 54595, World Bank, Heterogeneity in Peruvian Agriculture: A ­Washington, DC. Meta-Frontier Approach for A nalyzing Goyal, A., and C. González-Velosa. 2012. Technical Efficiency.” Paper presented to the “Improving Agricultural Productivity and 2018 Conference of International Association Market Efficiency in Latin America and of Agricultural Economists, July 28–August 2, the Caribbean: How ICTs Can Make a Vancouver, BC. Difference?” Working Paper 68255, World Eswaran, M., and A. Kotwal. 1989. “Credit as Bank, Washington, DC. Insurance in Agrarian Economies.” Journal of Hallward-Driemeier, M., and G. Nay yar. Development Economics 1 (1): 37–53. 2018. Trouble in the Making? The Future of Evenson, R E. 2000. “How Far Away Is Africa? Manufacturing-led Development. Washing- Technological Spillovers to Agriculture and ton, DC: World Bank. Productivity.” American Journal of Agricul- Hardaker J. B., G. Lien, J. R. Anderson, and tural Economics 82 (3): 743–49. R. B. M. Huirne. 2015. Coping with Risk in Evenson, R. E., and K. Fuglie. 2009. “Technol- Agriculture: Applied Decision Analysis, 3rd ogy Capital: The Price of Admission to the ed. Wallingford, UK: Centre for Agriculture Growth Club.” Journal of Productivity Anal- and Bioscience International. ysis 33 (3): 173–90. Helfand, S. M., and E. S. Levine. 2004. “Farm Evenson, R. E., and Y. Kislev. 1975. Agricultural Size and the Determinants of Productive Research and Productivity. New Haven, CT: Efficiency in the Brazilian Center-West.” Yale University Press. Agricultural Economics 31 (2–3): 241–49. Fagerb erg , J., a nd B . Ver spagen. 20 02 . Helfand, S. M., and M. Taylor. 2016. “Agri- “Technology-Gaps, Innovation-Diffusion and cultural Productivity and Farm Size in Latin Transformation: An Evolutionary Interpreta- America.” University of California, Riverside. tion.” Research Policy 31 (8-9): 1291–304. Hsieh, C.-T., and P. J. Klenow. 2009. “Misallo- Fattal Jaef, R. 2019. “Quantitative Evaluation of cation and Manufacturing TFP in China and the Premature Deindustrialization Hypothe- India.” Quarterly Journal of Economics 124 sis in Latin America?” Working paper, World (4): 1403–48. Bank, Washington, DC. Hurley, T. M., X. Rao, and P. G. Pardey. 2014. Feder, G. 1985. “The Relation between Farm Size “Re-examining the Reported Rates of Return and Farm Productivity: The Role of Family to Food and Agricultural Research and Labor, Supervision and Credit Constraints.” P r o d u c t i v i t y i n t h e L A C r e g i o n : A   s e c t o r a l v i e w    51 Development.” American Journal of Agricul- Union and United States.” Contemporary tural Economics 96 (5): 1492–04. ­Economic Policy 17 (4): 457–68. Ivanic, M., and W. Martin. 2018. “Sectoral Schumacher, E. F. 1973. Small Is Beautiful: Eco- Productivity Growth and Poverty Reduction: nomics as If People Mattered. New York: National and Global Impacts.” World Devel- Harper Collins. opment 109: 429–39. Sen, A. K. 1966. “Peasants and Dualism with or Jorgenson, D. W., and M. P. Timmer. 2011. without Surplus Labor.” Journal of Political “Structural Change in Advanced Nations: A Economy 74 (5): 425–50. New Set of Stylised Facts.” Scandinavian Jour- Sinha, R. 2019a. “Distortions in Intermediate nal of Economics 113 (1): 1–29. Markets and Structural Transformation in Kagin, J., J. E. Taylor, and A. Yuñez-Naude. 2016. Latin America.” Working paper, World Bank, “Inverse Productivity or Inverse Efficiency? Washington, DC. Evidence from Mexico.” Journal of Develop- Sinha, R. 2019b. “What Explains Latin A ­ merica’s ment Studies 52 (3): 396–411. Low Share of Industrial Employment?” Policy Kinfemichael, B., and A. M. Morshed. 2016. Research Working Paper 8791, World Bank, “Convergence of Labor Productivity across Washington, DC. the US States.” Economic Modeling 76 (C): Solis, D., B. E. Bravo-Ureta, and R. E. Quiroga. 270–80. 2009. “Technical Efficiency among Peasant Luh, Y. H., C. Chang, and F. Huang. 2008. Farmers Participating in Natural Resource “Efficiency Change and Productivity Growth Management Programmes in Central America.” in Agriculture: A Comparative Analysis for Journal of Agricultural Economics 60 (1): Selected East Asian Economies.” Journal of 202–19. Asian Economics 19 (4): 312–24. Stifel, D., and B. Minten. 2008. “Isolation Nordhaus, W. D. 2008. “Baumol’s Diseases: A and Agricultural Productivity.” Agricultural Macroeconomic Perspective.” BE Journal of ­Economics 39 (1): 1–15. Macroeconomics 8 (1). Thirtle, C., L. Lin, and J. Piesse. 2003. “The Olley, G., and A. Pakes. 1996. “The Dynamics of Impact of Research-Led Agricultural Produc- Productivity in the Telecommunications Equip- tivity Growth on Poverty Reduction in Africa, ment Industry.” Econometrica 64 (6): 1263–97. Asia and Latin America.” World Development Oulton, N. 2001. “Must the Growth Rate Decline? 31 (12): 1959–75. Baumol’s Unbalanced Growth Revisited.” Timmer, M. P., G. J. de Vries, and K. de Vries. Oxford Economic Papers 53 (4): 605–27. 2015. “Patterns of Structural Change in Devel- Reimers, M., and S. Klasen. 2013. ­ “Revisiting oping Countries.” In Routledge Handbook of the Role of Education for Agricultural Industry and Development, edited by J. Weiss ­Productivity.” American Journal of ­ Agricultural and M. Tribe, 65–83. Abdingdon-on-Thames, Economics 95 (1): 131–52. UK: Routledge. Schimmelpfennig, D., and C. Thirtle. 1999. Trindade, F. J., and L . E . Fulginiti. 2015. “The Internationalization of Agricultural “Is There a Slowdown in Agricultural Produc- ­ Technology: Patents, R&D Spillovers, and tivity Growth in South America?” Agricultural Their Effects on Productivity in the European Economics 46 (SI): 69–81. Economic transformation, skills, and the future of work 3 C hapter 2 described how technologi- been around for centuries. Perhaps most cal changes, income effects, and con- famously, in England in the early nineteenth sumer preferences are changing the century members of the Luddite movement structure of economies in the Latin America sabotaged new textile machines to defend and the Caribbean (LAC) region. This chap- their jobs. And yet economic history has ter shifts attention to the impacts that this proven these concerns unfounded. Time and economic transformation and the emer- time again, technological innovations have gence of new technologies will have on jobs, spurred dramatic productivity gains that occupations, and the demand for skills. increased standards of living and created Recently, much attention has been devoted many more jobs than they destroyed. to the potential impacts of emerging tech- As will be discussed shortly in more detail, nologies. Under the banner of the Fourth the total impact of automation is hard to Industrial Revolution are technological inno- forecast because the effects of innovations vations such as artificial intelligence (AI), tend to be widespread and ripple through- Internet-of-Things (IoT), and 3D printing. out the economy. In essence, new technolo- Meanwhile, a flurry of reports and books gies that increase productivity have general have appeared aimed at trying to understand equilibrium effects that increase the demand the impact of these technologies on the labor for labor across the economy. The simplistic market and the jobs of the future.1 idea that an economy has a fixed number of Fears around the concept of tasks is known as the “lump of labor fallacy.” “technological unemployment” have made Innovations may generate jobs in the industry headlines and dominate the concerns where they are applied, but also in industries of polic y makers and workers alike. that are connected (through either backward Technological unemployment refers to the or forward linkages) to that industry and notion that technological innovations such even in unrelated industries. History also as AI and automation will take over most of teaches us that innovations can create jobs the production tasks in the economy, leaving that do not even exist today. humans without work. The fear of machines As described in this chapter, however, the taking over jobs is not new and in fact has labor market is already changing. During the 53 54   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n 1990s and the early 2000s, the main concern Finally, this chapter highlights the pol- of labor economists was the rising wage icies that governments must institute to inequality in both developed and developing guard against the potential adjustment economies. Over time, new evidence has costs brought about by technological inno- emerged for advanced economies indicating vations. As some occupations are replaced that jobs and occupations in the middle of by machines, new ones will appear as well. the wage distribution have been shrinking—​ Nevertheless, it is clear that workers will a  phenomenon dubbed labor ma rket interact with more machines and will be polarization. To explain this phenomenon, expected to understand increasingly complex economists rely on a new theory known as technologies. They therefore will need the routine-biased technological change (RBTC), capabilities and skills to adjust to these new which suggests that recent technological demands. change is biased toward replacing labor in Investing in the human capital of the routine tasks. workforce continues to be the best policy Although there is mixed evidence that this to insure against the risk of automation polarization in the labor market has reached and should be a priority for policy makers. developing countries (Maloney and Molina Although investing in early childhood 2016; Messina, Oveido, and Pica 2016), there education generates the highest return on is growing concern that it will reach these ­ investment (World Bank 2019), there is room countries sooner rather than later. This is to improve in every dimension of the educa- of special concern to economies in the LAC tional system. In recent decades, many LAC region, which are already exhibiting high countries have made substantial progress in ­ levels of wage inequality. improving access to secondary education, but This chapter begins by discussing the the quality of education continues to lag that changes in the labor market already under of advanced nations and developing country way in the LAC region. Although there is peers in East Asia. little evidence of labor market polarization What may become more important as new in the region, this study found substantial automation technologies are adopted in LAC changes in the composition of occupations countries is adult learning and retraining in the economy—in particular, a shift away programs. Although the time frame for the from occupations that are intensive in routine adoption of technology is not clear, it is pos- manual (RM) tasks (such as machine operator sible that transformations in the workplace and assembler) toward occupations that are will happen midcareer for many workers, and intensive in nonroutine analytical or cognitive so they will need to adapt and adjust, partic- tasks (such as lawyer, scientist, and manager) ularly to the changing set of tasks they must and nonroutine interpersonal tasks (such as perform at work. To minimize the adjustment teacher, manager, and personal trainer). costs borne by workers, governments should It then discusses the effects of automation support programs that help workers upskill on jobs in general and presents estimates, and retrain for these new jobs and tasks. using different methodologies, of the poten- tial job losses in the LAC economies. It is important to note upfront that the estimated The labor market is already range of potential job losses is very wide, and changing it clearly reflects the limited understanding From production to services of this issue. Perhaps more important, these estimates are based on technological feasi- As argued in this report, potent economic bility rather than economic incentives, and forces are transforming the global econ- these methodologies are designed to capture omy, shifting employment away from pro- only a measure of jobs at risk and not poten- duction (agriculture and industry) and into tial jobs created. services sector. Two factors are at work. the ­ Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    55 First, as incomes rise, consumers tend to fewer goods occupations and more service devote a larger share of their expenditures occupations than expected given the level of to services. Second, technological progress is development. more acute in the agriculture and industrial Table 3.1 presents estimates on how the sectors, thereby pushing workers into the composition of occupations changes over ­services sector. the development process. Clearly, not only Compounding this shift in the economic does economic development bring a shift in structure is a transformation of occupations the total employment levels per sector, but within broad economic sectors. Duernecker also within sectors the input allocation of and Herrendorf (2017) propose a new labor changes. In other words, as economies model of structural transformation that develop, each economic sector employs rela- distinguishes between broad categories of tively fewer people directly in the production occupations instead of broad categories process and more people who are producing of industries. They categorize occupations intangible value added. using the same underlying principle as for Thus two effects are changing the nature industries: goods occupations such as farm of jobs in the same direction. First, techno- workers and machine operators produce tan- logical innovations and rising incomes are gible value added; service occupations such pushing production and workers away from as clerks and managers produce intangible agriculture and manufacturing and toward value added. With this novel classification the services sector. Second, compounding and using 182 harmonized census data for this effect is the change in the composition 67 countries, Duernecker and H ­ errendorf of occupations under way within each broad (2017) show that as gross domestic prod- economic sector. In other words, within the uct (GDP) per capita increases, employment manufacturing and agriculture sectors, occu- in goods occupations decreases, whereas it pations are shifting away from production increases in service occupations. More sur- toward service occupations (that is, more prising, however, is that as GDP per capita managers and professionals and fewer farm increases, the employment share of service workers and machine operators). The picture occupations increases in all economic sec- that emerges is one in which the jobs of the tors. Therefore, workers are shifting toward future will be mostly service occupations that service occupations (producing intangible are increasingly concentrated in the services value added) in the services ­ sector but also sector. Most important for policy makers, in the goods-­ producing ­ s ector. This result service occupations require skills very differ- is intimately related to the “­ servicification” ent from those needed for production-related of manufacturing phenomenon described in occupations. The following section turns to chapter 2. this issue. This study replicates the analysis by Duernecker and Herrendorf (2017) and finds From skill-biased technological change that these shifts are present in LAC coun- to routine-biased technological change tries as well. As Latin American economies have grown over time, the share of workers Recent evidence in the academic literature employed in goods occupations has fallen, points to a labor market that is changing while the share in service occupations has rapidly and significantly. Starting from the risen (see figure 3.1, panels a and b). A more observation that job loss and job creation pat- detailed analysis shows that the decline in terns are not random, labor economists have goods occupations is related to declines in developed different hypotheses to explain the both the agriculture sector and the indus- observed patterns. What is clear is the signif- trial sector. Perhaps related to the premature icant shift in the demand for skills. deindustrialization hypothesis, the graphs During the 1990s and early 2000s, wage reveal that LAC countries appear to have inequality was rising in developed and 56   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 3.1  Development of goods and service occupations, LAC and rest of world a. Goods occupations 1.0 0.8 Share of employment 0.6 0.4 0.2 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 Per capita GDP (1990 US$) b. Service occupations 0.8 0.6 Share of employment 0.4 0.2 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 Per capita GDP (1990 US$) LAC RoW LAC, LOWESS t RoW, LOWESS t Sources: Original calculations for this publication using IPUMS International Database (Minnesota Population Center 2019); Maddison Database (Bolt et al. 2018). Note: Vertical green lines mark the values 1,000, 15,000, and 30,000 for which employment shares appear in table 3.1. GDP = gross domestic product LAC = Latin America and the Caribbean; LOWESS = locally weighted polynomial regression; RoW = rest of world. Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    57 TABLE 3.1  Reallocation of occupations within sectors over development process Goods sector Services sector GDP per capita (1990 US$) 1,000 15,000 30,000 1,000 15,000 30,000 Employment share (%) of   Goods occupations 0.97 0.75 0.60 0.17 0.14 0.11 Service occupations 0.03 0.25 0.40 0.83 0.86 0.89 Source: Original calculations for this publication using IPUMS International Database (Minnesota Population Center 2019). Note: Shares are calculated from fitted LOWESS curves. GDP = gross domestic product; LOWESS = locally weighted polynomial regression. developing economies. In particular, it was have declined. This phenomenon has been documented that the skill premium (the well documented for the United States and extra income earned by educated workers) the United Kingdom, Germany, and other was rising. Labor economists hypothesized major economies of Western Europe.2 that technological innovations were bene- To explain these new patterns in the labor fiting educated workers more relative to less market, economists developed a new theory, skilled workers. This so-called skill-biased RBTC. Essentially, routine tasks are a lim- technological change (SBTC) theory essen- ited and well-defined set of cognitive and tially explained the increase in the wage manual activities that can be accomplished premium of educated workers by suggesting by following explicit rules. For example, that new technologies were making highly picking, sorting, and repetitive assembly are skilled workers more productive. In other RM tasks; record-keeping, calculation, and words, technological innovations were com- repetitive customer service (such as bank tell- plementary to educated workers. As a result, ers) are examples of routine cognitive (RC) they became more productive and more in tasks. Nonroutine tasks are those that cannot demand, ultimately leading to higher wages be easily codified or defined in explicit rules. for the skilled workforce. For a couple of These tasks can be cognitive such as problem decades, the theory and the empirical tests solving, complex communication activities, and evidence worked well in explaining the and forming and testing hypotheses. They patterns observed in the data. can be manual as well—for example, driving Over time, however, new evidence showed and sports activities. a hollowing out of jobs and occupations RBTC models generally posit that com- that were in the middle of the wage distribu- puters and robots are more substitutable for tion in developed economies. Although jobs human labor in carrying out routine tasks were still being created at both ends of the than nonroutine tasks. Routine and nonrou- skill spectrum (low skilled and high skilled), tine tasks are themselves imperfect substi- the middle-skilled jobs were disappearing. tutes, and a greater intensity of routine inputs This phenomenon is known as labor mar- increases the marginal product of nonroutine ket polarization. Specifically, high-paying inputs. According to Autor and Dorn (2013, jobs such as managerial, professional, and 1559), “The secularly falling price of accom- associate professional occupations are expe- plishing routine tasks using computer capital riencing rapid increases in their employ- complements the ‘abstract’ creative, problem-­ ment shares. In addition, the employment solving, and coordination tasks performed shares for ­low-paid service workers such as by highly-educated workers such as profes- domestic helpers, cleaners, security person- sionals and managers, for whom data anal- nel, and those in catering and personal care ysis is an input into production. Critically, have increased. By contrast, the employment automation of routine tasks neither directly shares of middle-paying jobs such as office substitutes for nor complements the core jobs clerks, craft and related trades workers, and tasks of low education occupations—service plant and machine operators and assemblers occupations in particular—that rely heavily 58   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n on ‘manual’ tasks such as physical dexterity economies in Central America, which and flexible interpersonal communication. may receive the manufacturing jobs being Consequently, as computerization erodes the offshored from the United States. wage paid to routine tasks in the model, low- Third, new technologies may lower skill workers reallocate their labor supply to barriers to entry and facilitate informa- service occupations.” tion flows on markets and opportunities, This new theory seems to fit the experience potential products, inputs, and production of developed economies quite well. What has technologies to enable the creation of new been the experience of developing countries? industries such as travel services, finance, Is labor market polarization occurring there tourism, and international marketing of local as well? Will the same patterns materialize? products. Meanwhile, the impact of techno- And where does the LAC region stand in this logical innovations in developing countries is debate? This chapter turns to these questions unclear. Some evidence suggests that adop- next. tion of information and communications technology (ICT) is strongly correlated with job polarization (Michaels, Natraj, and Van Labor market polarization in the Reenen 2013). However, ICT-related capital developing world: Is it coming? stocks are lower in developing countries To date, there is mixed evidence of labor (Eden and Gaggl 2015), and so the displace- market polarization in developing countries. ment effects on jobs directly affected by ICT On the one hand, even if polarization has not may be more muted. yet occurred, it may be around the corner as As will be argued in more detail shortly, technological innovations are dispersed and adoption of labor-saving technologies that adopted around the developing world. On the increase productivity and lower final prices other hand, there are many reasons why the may result in higher employment levels if the experience of developing countries need not demand for these products or services are be the same as that of advanced economies. elastic—­ meaning that an increase in quan- Maloney and Molina (2016) offer several tity demanded will more than compensate possible reasons why labor market polar- for the fall in price. Another argument relates ization may never happen or may be more to the degree to which automation is adopted muted in some countries. First, initial occu- because it depends on several factors: skill pation distributions may be very different in of the workforce, maintenance capacity, and developing economies. For one thing, they technological absorptive capacity, among may not have many workers engaged in the others. Adoption of automation technologies routine tasks commonly associated with may therefore take a long time, depending on manufacturing and routine clerical work in these initial conditions. Meanwhile, polariza- offices. This argument is particularly relevant tion implies higher employment in the types for lower-income countries where industrial- of occupations that complement automation. ization may be limited (and routine manufac- If a country does not have a broad, highly turing jobs are few) and where many workers skilled workforce, then this employment are engaged in primary and elementary occu- growth would never occur, limiting labor pations. Thus few workers engaged largely in force polarization. routine tasks would be displaced. In the con- text of the LAC region, this argument may be Labor market polarization in the relevant to Bolivia, Haiti, and some Central developing world: The evidence American countries. Second, jobs offshored from advanced The different sources of data on employ- economies may be filling in (as opposed to ment, tasks, and occupations have produced hollowing out) the middle-skilled jobs in mixed findings on labor market polar- developing economies. This could be par- ization. Based on harmonized labor sur- ticularly important for Mexico and some veys and Autor’s (2014) classification, the Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    59 World Development Report 2016: Digital response to the introduction of automation Dividends (World Bank 2016, 120) states technologies, workers adjusted their work that time toward tasks complementary to the ones performed by machines. Moreover, analysis there are signs that employment is also of worker-level information on the tasks per- polarizing in a number of low- and mid- formed in an occupation reveals that work- dle-income countries. The average decline ers’ task structures differ remarkably within in the share of routine employment has occupations (Autor and Handel 2013). Cross- been 0.39 percentage points a year, or country differences are relevant as well. 7.8 percentage points for the period. China Messina, Oviedo, and Pica (2016) conclude: is an exception, since the mechanization “Comparing task intensity scores in those of agriculture increased the share of rou- countries and the U.S. shows that while the tine employment. Labor markets in low-­ abstract content of jobs is similar in North- income countries such as Ethiopia, with a large share of employment in manual and South-America, the routine and manual occupations, are also not polarizing; nei- contents are different. We speculate that the ther is employment in Mongolia or Latin reason may be that Latin American occupa- ­ A merican countries where other factors— tions comprise a more heterogeneous set of such as a commodity-driven boom bene- tasks.” Therefore, it is important to consider fiting low-skilled workers—could play a not only the evolution of the occupational larger role in shaping labor markets. structure, but also how the task content of occupations is changing over time. On the other hand, Maloney and Molina (2016), using harmonized census data, do The changing demand for skills in the not find strong evidence of polarization in LAC region developing economies. They find that the key occupational categories associated with rou- What follows is a description of the results tine tasks are not decreasing, even in ­ relative of this study’s analysis of the evolution of terms, for most countries in the s ­ ample. the demand for human skills for 11 LAC However, they do find relative declines countries from 2000 to 2014. The study fol- in these types of occupations in Brazil, lows the methodology proposed by Autor, Indonesia, and Mexico, which could suggest Levy, and Murnane (2003) and updated by potential polarizing forces. Acemoglu and Autor (2011). The approach Both results are based on analyzing the conceptualizes and measures skills by assess- changes in occupational structure within ing the specific tasks associated with differ- countries over time. Following Autor (2014), ent occupations rather than measuring the occupations are classified as low, medium, educational credentials of workers perform- or high skilled. Specifically, medium-skilled ing those tasks. As is standard in the litera- occupations are white-collar clerical, admin- ture, five skills are assessed: routine manual istrative, and sales occupations, as well as (RM), nonroutine manual physical (NR-MP), blue-collar production, craft, and operative ­ routine cognitive (RC), nonroutine cognitive occupations. Therefore, using either har- analytical (NR-CA), and nonroutine cogni- monized census data or harmonized labor tive interpersonal (NR-CP). surveys, the analysis focuses on the relative The analysis relies on the ex post harmo- growth of each occupational category. nized household surveys for each country-­year What is missing from this analysis is the prepared by the World Bank and the Center fact that the set of tasks within an occupa- for Distributive, Labor and Social Studies tion is not fixed over time. Thus the demand at the Universidad de la Plata in Argentina for skills may be changing even if the occu- (SEDLAC). These labor and household pational structure is not changing signifi- income surveys are generally nationally rep- cantly. In fact, Autor, Levy, and Murnane resentative, and they provide information on (2003) and Spitz-Oener (2006) find that in the size of the household, demographics, and 60   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n educational attainment, and, more important, profiles for specific occupations. For exam- detailed information on employment. The ple, teachers in low-income settings are more original (nonharmonized) country-specific likely to lack the tools (especially ICT tools) occupational classifications are reclassified that support innovative teaching than teach- into the International Standard Classification ers in developed countries. Similarly, doctors of Occupations, version 1988 (ISCO-88), or nurses might have access to equipment as developed by the International Labour well as medical knowledge which impacts the Organization. All occupations contained skill content and mix they can bring to bear in the household surveys are then matched in different settings.” The authors then pos- with their respective skill content from the tulate that occupations intensive in nonrou- US Department of Labor’s Occupational tine tasks are probably more skill-intensive Information Network (O*NET) database in more advanced economic settings than in (https://www.onetonline​.org/). lower-income ones. If true, this would sug- An important caveat for this analysis is gest a potential upward bias in the measured that O*NET is taken as the primary reference skill intensity of nonroutine (both analytical because there are no country-year specific and interpersonal) skills.4 catalogs of skill content for LAC countries.3 Measurement of task content is usually Essentially, it is assumed that the skill content based on data that stem from either the of a given occupation is comparable interna- expert-based approach or the worker-based tionally. The validity of this assumption may approach. Box 3.1 describes these approaches differ across certain occupations or country and their pros and cons. contexts. As noted by Aedo et al. (2013, 9), Anyone interpreting the results of this anal- “countries differ in technology and regulatory ysis should note that the values of the indexes contexts which may employ different skill are not strictly comparable across countries. BOX 3.1  What are workers doing? Jobs in an economy are indexed to a set of occupations and industrial psychologists—are interviewed to that develop tasks. These tasks have been categorized weigh in on the importance of a given occupation by in the literature as routinary or n ­ onroutinary—see scoring the importance or intensity of different tasks Acemoglu and Autor (2011) and Autor, Levy, and in the workplace. A common source of information is Murnane (2003). The first category consists of man- the Occupational Information Network (O*NET),a ual and specific activities generally more prone to which covers nearly 1,000 occupations in the United automation and replicability by machines or com- States. At the outset, O*NET operates by providing puters. The second category is ­ composed of more information on work-oriented descriptors (such as complex activities in which abstraction and socio- worker characteristics, worker requirements, and emotional skills play an important role. Measuring experience requirements) and job-oriented descrip- task content, however, is not straightforward, and tors (such as occupational requirements, workforce the two main streams of data that inform research characteristics, and occupation-specific informa- are provided either by a pool of experts on a fixed tion) that account for tasks. These assessments are number of occupations (expert-based approach) or updated periodically to reflect changes in the occu- by workers who identify their task content relying on pational structure of the US economy, using as point their own experience (worker-based approach). of reference the Standard Occupational Classifica- tion (SOC). Expert-based approach The expert-based approach has been widely used This approach hinges on the fact that a group of in studies aimed at understanding occupation dynam- respondents—job incumbents, occupation experts, ics (for example, Acemoglu and Autor 2011; Autor Box continues next page Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    61 BOX 3.1  What are workers doing? (Continued) and Dorn 2009; Goos and ­ M anning 2007; Goos, and math use; physical (manual) requirements; and Manning, and Salomons 2009). One ­ advantage of interpersonal activities at work. this approach is that it serves as a rich source of infor- The main advantage of using the worker-based mation on tasks over time that can also be extrapo- approach is that it avoids the problems of measure- lated to other economies. However, such transition ment error when ascribing data from the United through reference crosswalks has been criticized for States d (via O*NET) to, for example, developing the glaring bias ­produced by assuming that the task economies. Nevertheless, the response bias produced content is the same as in the United States, mainly by the large variance in the computation of task from O*NET data.b indicators within occupations could be problematic. The underrepresentation of occupations because of Worker-based approach small samples that do not cover all economic sec- Unlike the expert-based approach, this way of tors could also hinder comparability vis-à-vis studies measuring tasks is taken from workers using specific based on the expert-based approach. surveys. Workers are interviewed and asked about their cognitive and noncognitive traits at work. Two a. Another reference for the United States is the ­ Dictionary of Occupational Titles (DOT), a previous version of common sources of information are the World Bank’s O*NET also sponsored by the US Department of Labor STEP (Skills Toward E­ mployability and Productivity) but currently outdated in the literature. Skills Measurement Program for developing countries b. See, for example, Hardy, Keister, and Lewandowski and the Programme for the International Assessment (2018) for an application in Eastern Europe and Aedo Competencies (PIAAC) for Organisation for of Adult ­ country comparison using house- et al. (2013) for a cross-­ Economic Co-­ operation and Development (OECD) hold surveys and O*NET. countries.c STEP and PIAAC ask random individuals, c. https://microdata.worldbank.org/index.php/catalog​ /step/about); https://www.oecd.org/skills/piaac/. ages 15–65, about their household characteristics d. Dicarlo et al. (2016) find that data on nonroutine tasks in the areas of health, education, training, and in developing countries are more likely to resemble data employment. Moreover, data are collected on from the United States. A low correlation is reported for cognitive and socioemotional skills that reflect routine tasks, meaning that EA studies overestimate the the complexity and frequency of reading, writing, task content of basic repetitive tasks. Thus a higher value of the index of NR-CA Uruguay, Brazil, and Chile show a slower skills at the endpoint does not imply that more but still important growth rate in the usage of those skills are found in one country than of these skills, whereas the Dominican another. Instead, it implies that the country Republic and El Salvador exhibit slow has changed its occupational structure in favor growth rates. Mexico exhibits a decrease of those skills at a higher rate. Therefore, it is in the use of NR-CA tasks, a surprising possible to compare the rate of change (trends) result that may stem in part from certain across countries over time. data restrictions. Because Mexico changed The general results are for the most its occupational classification in 2008, the part consistent with the findings of the lit- analysis was restricted to the 2000–2008 erature in both developed and developing timeframe. The year 2008 was marked by countries. Figure 3.2, panels a and b, reveal the beginning of the global financial crisis, that most countries in the LAC region have which also may have affected the results. experienced increases in the analytical For NR-CP skills the story is similar. A (panel a) and interpersonal (panel b) tasks first group of countries composed of Costa within the nonroutine cognitive task com- Rica, Ecuador, and Nicaragua lead the pack ponent. In the case of NR-CA tasks, Costa with very strong growth rates, while a second Rica has grown the most, followed by group, comprising the Dominican R ­ epublic, Ecuador, Nicaragua, Peru, and Colombia. Uruguay, El Salvador, and Peru, also show 62   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 3.2  Evolution of task content of jobs (mean change): 11 LAC countries, 2000–2014 a. Nonroutine cognitive analytical tasks (NR-CA) b. Nonroutine cognitive interpersonal tasks (NR-CP) 0.10 0.10 Growth of mean task index Growth of mean task index 0.08 0.05 0.06 0.04 0.02 0 0 a in l Sa ico ile ile a ico Ec ua Ec a ca ca sta r u Sa ru ep r lic r Ni blic sta r n R uay Ur azil Co uay il bi Co do n R ado do bi gu Co do r az Ri Ri e e Ch Ch ub g m E ex ex om P P ua lva u ra ra ua Br Br ica rug ug ica lv lo ep M M ca ca l Co U Ni El in m m Do Do c. Nonroutine manual physical tasks (NR-MP) d. Routine cognitive tasks (RC) 0 0.15 Growth of mean task index Growth of mean task index –0.02 0.10 –0.04 0.05 –0.06 0 –0.08 –0.05 –0.10 –0.10 le n R ico Co ico a Ec ua Ur ca a Co ile a ru Ni Peru Sa y ep r Ur lic M c ca Co dor Ec ay r Ni dor il il n R do bi do bi gu li El gua az i az Ri Ch Pe g Ch ub ub Ri u ica Mex ex m m ica lva ra ua lva Br ra ua ug Br sta ta lo ep u lo ca ca a s m El S Co in in m Do Do e. Routine manual tasks (RM) 0 Growth of mean task index –0.05 –0.10 –0.15 Co blic Sa co ile a a ca r Co uay ru Ec or Ur azil do bi gu Ri d i Pe Ch ex m u ua lva ug ra Br ep sta lo M ca nR Ni El ca ini m Do Source: Original calculations for this publication using Socio-Economic Database for Latin America and the Caribbean (SEDLAC) household surveys, CEDLAS and World Bank (https:// datacatalog​.worldbank.org/dataset/socio-economic-database-latin-america-and-caribbean). Note: CEDLAS = Center for Distributive, Labor and Social Studies; LAC = Latin America and the Caribbean. Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    63 strong growth rates. Behind are Brazil, Industrial sector Mexico, Colombia, and Chile with smaller Figure 3.3, panels a and b, describes the evo- increases. lution of NR-CA and NR-CP in the industrial The time trends for NR-MP in panel c are sector (mining and quarrying, manufactur- consistent with the findings of previous stud- ing, construction, and utilities). Both graphs ies. Throughout the region, there is a marked tell the same story: an increase in the intensity trend of declining NR-MP tasks. The larg- of both NR-CA and NR-CP in the industrial est declines are observed for Costa Rica and sector. Among these, two groups emerge. Peru, followed by Nicaragua and Ecuador. Nicaragua, Peru, El Salvador, the Dominican Panel e paints a similar picture for RM Republic, Ecuador, Costa Rica, and Brazil tasks. The Dominican Republic and Costa show the highest rates of transition toward Rica show the largest declines. Essentially, NR-CA– and NR-CP–intensive occupations. all countries in the region are experiencing a Following at a more modest pace are Chile, shift away from NR-MP and RM, albeit at Colombia, Mexico, and Uruguay. different paces. Panel c of figure 3.3 shows that in the The findings on RC tasks are mixed industrial sector NR-MP labor tasks have (figure 3.2, panel d). They have increased in ­ increased across all countries. The changes many countries of the region (strongly in Peru are most profound in Chile, Ecuador, El and Brazil), while declining in others, most Salvador, and Brazil, and are more moderate noticeably in El Salvador. A similar result is in Costa Rica, Nicaragua, Colombia, and reported by Hardy, Keister, and Lewandowski the Dominican Republic. Panels d and e of (2016) for a sample of 10 Eastern European figure 3.3 describe the evolution of RC tasks countries.5 The authors attribute the different and RM tasks in the industrial sector. The findings across countries to a combination of results closely mirror those for nonroutine varying rates of structural changes and shifts tasks. In most countries, they decrease. toward work with a lower speed of deroutin- These results suggest that production ization. This result contrasts with the experi- processes within the industrial sector of the ence of developed countries, where there is a region are changing, adopting more nonrou- clear and marked decline of occupations with tine cognitive and manual tasks. At the same RC-intensive tasks. This finding should be of time, and consistent with the literature on concern to policy makers in the region. The automation and robotization, the demand for evidence in advanced nations suggests that the skills in the region is moving away from rou- technologies that could replace these types of tine tasks, both cognitive and manual. tasks already exist and could be adopted in At this point a cautionary note is war- the LAC region in the near future. Thus these ranted. As noted earlier, this analysis is based occupations may be at risk of changing or dis- on the O*NET classification of tasks in both appearing in the next decade or so, depending the base year of 2003 and the updated ver- on the rate of technology adoption. sion of 2017. Use of both catalogs allows What follows is a description of the results incorporation into the analysis the possible that emerge from a timeline analysis of the changes in tasks within occupations over two major economic sectors (industrial and time. In other words, workers in the same services) for 11 countries in the LAC region.6 occupation may be performing a different Following the same standardization pro- set of tasks between the two points in time. cedures as earlier, the evolution of the task Adoption of new technologies, for example, component indexes for the sample of work- may replace part of the tasks performed in an ers employed in each sector is described occupation, thereby allowing the workers to separately. Thus the results presented speak spend more time on other tasks and chang- only to the changes in task utilization within ing the task intensity within an occupation. each sector and abstracts from the effects of Thus the assumption in this analysis is that reallocation of labor across sectors. ­ the changes in tasks in the United States have 64   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 3.3  Evolution of task content of jobs in industrial sector: 11 LAC countries, 2000–2014 a. Nonroutine cognitive analytical tasks (NR-CA) b. Nonroutine cognitive interpersonal tasks (NR-CP) 0.8 0.6 Growth of mean task index Growth of mean task index 0.6 0.4 0.4 0.2 0.2 0 0 ua a gu ico ico ca m El S ica ag a y ile Ec ic n R or r a y M e ru ru r ep r Sa c sta l sta l do do n R do ua bi Co azi Co razi El ubli ua ra bi il l r Ri ub ad Pe Pe Ch Ch R ex ex ca om ica om ug ua lva ica lva ug Br Ni in Ecu B M ep Ur N l Ur a l Co Co ica in m Do Do c. Nonroutine manual physical tasks (NR-MP) d. Routine cognitive tasks (RC) 0.25 0 Growth of mean task index Worth of mean task index 0.20 –0.1 0.15 –0.2 0.10 –0.3 0.05 0 –0.4 ep a ico ile Co uay El Peru M a Ni ico Co ua a Ur lic Ur lic ca ca r r y ru ile Ec r Co or Sa zil Ec zil do do do bi n R bi gu ua ub ub Ri Ri Ch d Pe g Ch a a ex ex m ica lom ug ua ua lva lva ra ra ug Br Br ta sta ep lo M ca ca Sa s nR Co Ni El ica in in m m Do Do e. Routine manual tasks (RM) 0 Worth of mean task index –0.1 –0.2 –0.3 –0.4 Ur co a ile Co ua n R Rica r Sa ru Ec y lic Ni dor M il do bi ua az i e Ch ub g ex m P ua lva ug ra Br in osta ep lo ca C El ica m Do Source: Original calculations for this publication using Socio-Economic Database for Latin America and the Caribbean (SEDLAC) household surveys, CEDLAS and World Bank (https:// datacatalog​.worldbank.org/dataset/socio-economic-database-latin-america-and-caribbean). Note: CEDLAS = Center for Distributive, Labor and Social Studies; LAC = Latin America and the Caribbean. Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    65 occurred in LAC countries as well and to the occupations (how tasks have changed in that same extent. occupation), and (3) the interaction between What follows is a simple decomposition these two. This decomposition allows disen- of the overall results for the industrial sector tanglement of some heterogeneous patterns into three components: (1) between occu- that are observed across countries. pations (changes in the occupational struc- Figure 3.4 presents the results of this sim- ture within the industrial sector), (2) within ple decomposition for NR-CA tasks (panel a) FIGURE 3.4  Decomposition of task content in industrial sector: 11 LAC countries, 2000–2014 a. Nonroutine cognitive analytical tasks (NR-CA) 0.9 0.8 0.7 0.6 Share of task content 0.5 0.4 0.3 0.2 0.1 0 –0.1 –0.2 a ile ico a ca r u r lic y il do do bi gu ua r az Ri Ch Pe ub ex om ua lva ra ug Br ta ep M ca Ec l s Sa Ur Co Co nR Ni El ica in m Do b. Nonroutine cognitive interpersonal tasks (NR-CP) 0.7 0.6 0.5 Share of task content 0.4 0.3 0.2 0.1 0 –0.1 a a ico ca ile ru r y r lic zil gu o do bi ua Pe Ri ad Ch ub a ex m ra lva ug Br ta u lo ep M ca Ec Ur s Sa Co Co nR Ni El ica in m Do Between e ect Wthin e ect Total Source: Original calculations for this publication using Socio-Economic Database for Latin America and the Caribbean (SEDLAC) household surveys, CEDLAS and World Bank (https://datacatalog.worldbank.org/dataset/socio-economic-database-latin-america-and-caribbean). Note: Figure shows decomposition of the overall results for the industrial sector into three components: (1) between occupations (changes in the occupa- tional structure within the industrial sector); (2) within occupations (how tasks have changed in that occupation); and (3) the interaction between these two. CEDLAS = Center for Distributive, Labor and Social Studies; LAC = Latin America and the Caribbean. 66   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n and NR-CP tasks (panel b). The patterns just transformation in its production process in described are for the most part the result of which manual tasks are being replaced by within-occupation changes. Interestingly, more cognitive tasks (both routine and non- in five countries—Chile, Ecuador, Mexico, routine). Starkest are the patterns of decreas- Peru, and Uruguay—the changes between ing manual tasks, as economies move toward occupations contributed negatively to the RC and NR-CP tasks, perhaps reflecting overall nonroutine cognitive results. In other increases in more administrative or cleri- words, over time the industrial sector in these cal work, as well as tasks that involve more economies has changed its occupational teamwork or interactions with clients. As structure away from nonroutine c ­ ognitive noted earlier, the increase in the intensity of tasks. This effect is completely reversed, RC skills is somewhat at odds with the lit- however, by the increase in nonroutine cog- erature for developed countries and should nitive tasks within occupations. Thus if one be a red flag for policy makers. The technol- assumes that workers in the LAC region in ogy to replace workers in these types of tasks a specific occupation did not change their already exists, as evidenced by the relative tasks at all (no within-occupation changes), decline of these occupations in the developed then the industrial sector in Chile, Ecuador, world. Therefore, as technology disperses Peru, Mexico, and Uruguay would have seen and reaches LAC economies, it is very likely declines in the use of nonroutine cognitive that workers in these types of occupations tasks. will face competition from machines and are perhaps at risk of losing their jobs. Services sector Figure 3.5 reveals that in the services sector Conclusions NR-CP tasks (panel b) are increasing in all LAC countries except Chile, Colombia, and In general, countries in the LAC region Uruguay, and NR-CA tasks (panel  a)  are appear to be shifting away from occupations increasing strongly in Peru, Nicaragua, and that are intensive in manual tasks (both rou- Ecuador, increasing moderately in El Salvador, tine and nonroutine) and toward occupations Brazil, and Costa Rica, and decreasing in that are intensive in nonroutine cognitive Colombia, Chile, Uruguay, the Dominican tasks (both analytical and interpersonal). The Republic, and Mexico. Ecuador, El Salvador, economywide changes in the occupational Nicaragua, and Peru stand out with the high- structure and therefore in the embedded skill est growth rates in both tasks. Panels c and e intensity of the economy may result from indicate that the services sector is also mov- three related but distinct economic forces. ing away from manual tasks, both routine First, as described in detail in the first and nonroutine. Finally, panel d also reflects chapter of this report, as LAC economies important increases in the intensity of RC develop they are reallocating labor across tasks for all LAC countries. The increase in the broad economic sectors. Although some use of RC tasks is somewhat at odds with the occupations appear in all sectors, in general results for developed countries and the RBTC structural transformation implies changes hypothesis. In fact, Autor, Levy, and Murnane in the occupational structure of an econ- (2003) and Acemoglu and Autor (2011) find omy. In fact, LAC countries experienced that occupations intensive in RC tasks (such substantial structural transformation during as clerical and administrative) are among the the 2000–2014 time frame. In particular, group of occupations that are declining the as documented earlier, most countries in most in the United States, and a similar result the LAC region are experiencing premature has been found for Western European coun- deindustrialization, which implies there are tries (Goos et al. 2014). relatively fewer jobs in the industrial sector, Thus it appears that the services sector in whereas employment in the services sector the LAC region is undergoing an important has increased dramatically. Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    67 FIGURE 3.5  Evolution of task content of jobs in services sector: 11 LAC countries, 2000–2014 a. Nonroutine cognitive analytical tasks b. Nonroutine cognitive interpersonal tasks 0.3 0.2 Growth of mean task index Growth of mean task index 0.1 0.2 0 0.1 –0.1 0 n R xico Co ile Co bia a le n R bia ico a Ur lic ca Ec ca Ni eru Ni dor y ru in olo y M c r Ec r r Sa il sta il gu El ado gu do do li ua ua i El raz Co az ub Ri Ri Ch Ch Pe ub ex m m P ua ra ra lva lva ug ug e Br sta B ep u lo ep M ca ca Ur Sa C ica ica in m m Do Do c. Nonroutine manual physical tasks d. Routine cognitive tasks 0.05 0.4 Growth of mean task index Growth of mean task index 0.3 0 0.2 –0.05 0.1 –0.10 0 m El S bia a ile ico Co ico ep a Co ile a Co ica Ec a r Ni dor ru Co uay Sa ay ru ep r lic Ur lic r Ur azil il do n R do do gu n R bi gu c az Ri Pe Pe Ch Ch ub ub u R ex ex m m ua ua ica alva lva ra ug ug ra Br Br sta sta lo o M M ca ca Ec l Ni El ica in in m Do Do e. Routine manual tasks 0 Growth of mean task index –0.05 –0.10 a Ur ica Ni dor ru ico n R ia lic ile m Co ay Co dor Sa il gu ca mb El az ub Pe u Ch R ex ua ra ug lva Br sta ep ini lo M ca Ec Do Source: Original calculations for this publication using Socio-Economic Database for Latin America and the Caribbean (SEDLAC) household surveys, CEDLAS and World Bank (https:// datacatalog.worldbank.org/dataset/socio-economic-database-latin-america-and-caribbean). Note: CEDLAS = Center for Distributive, Labor and Social Studies; LAC = Latin America and the Caribbean. 68   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n The second force is related to the tech- on jobs. What are the concerns expressed nological progress that is changing the by techno-pessimists, who claim that the nature of production processes within all new technologies of the Fourth Industrial sectors of the economy. Duernecker and Revolution (such as artificial intelligence, Herrendorf (2017) report that service occu- machine learning, Internet-of-Things, addi- pation employment (such as managers and tive manufacturing, and 3D printing) are clerks) grows within the goods-­ producing different from any previous technological sector (agriculture and industry in this innovations? What are all the possible gen- study’s classification) as GDP per cap- eral equilibrium impacts of introducing new ita increases. This is related as well to the technologies into an economy? Which effect phenomenon of servicification of manufac- will be most important? turing described in chapter 2. During this It then turns to how to measure the poten- study’s period of analysis, the LAC coun- tial impact of automation on the total num- tries experienced sustained high growth ber of jobs, followed by estimates of job rates, with higher GDP per capita at the losses due to automation based on different end of the period. Because service occupa- methodologies and data sources for 16 LAC tions differ from goods occupations in skill countries. intensities, a changing skill intensity usage within broad economic sectors should be Automation and jobs: A history of fear expected as well. of machines Third, the adoption of technology in the workplace changes the set of tasks that Concerns about mass technological unem- workers perform as part of their occupa- ployment have been around for centuries. tion. Because automation and robotization When clergyman William Lee applied for a take over the simpler, more routine tasks, royal patent for a knitting machine in 1589, workers have adapted by shifting their work Queen Elizabeth I of England pointed out, time toward the more complex and harder “Consider thou what the invention would to automate tasks. In fact, Autor, Levy, and do to my poor subjects. It would assuredly Murnane (2003) and Spitz-Oener (2006) bring them to ruin by depriving them of found that as a response to the introduction employment” (McKinley 1958). Similarly, of automation technologies, workers adjusted the Qing dynasty of China resisted the their work time toward tasks complementary ­ construction of railways because it was con- to those of the machines. cerned about the potential impact on the luggage-carrying jobs (Zeng 1973). Perhaps most famously, the Luddite movement sab- Looking into the future: otaged new textile machines to defend their Automation, tasks, and skills jobs in England. Since the seminal paper by Frey and Osborne And yet economic history has proven these (2017) claiming that 47 percent of jobs in the concerns unfounded. Time and time again, United States were at risk of disappearing to technological innovations have spurred automation, a flurry of reports and books dramatic gains in productivity that have have stoked fears of mass “technological increased standards of living and created unemployment.” This concern is not new; many more jobs than they destroyed. It is it dates back to the beginning of the First true that some jobs disappeared—machines Industrial Revolution and has been revived replaced many skilled and unskilled work- over time as powerful technological innova- ers over time. However, new jobs, some tions have revolutionized the way goods and related to the new technologies and many services are produced in an economy. not related, have been ­ created over time. As This section begins by organizing the gen- a result, today a higher proportion of a much eral ideas about the impact of automation larger population is working. Thus the lesson Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    69 from history is that technological innovations Daniel and Richard Susskind (2015) posit in have always created more jobs than they have their book about the future of professions: destroyed. 1. What is the new quantity of tasks that Modern techno-pessimists are aware of must be carried out? the lessons from history, but they claim it 2. What is the nature of these tasks? is different this time. In general, those who 3. Who has the advantage in carrying out fear a jobless future point to the increas- these tasks? ingly rapid technological advances driven by digitization (and thus the availability The first question refers to considering all of big data) and the exponential nature the effects produced by introducing a new of computing power.7 At first, advances labor-saving technology into the economy. in automation were limited to routine, Although at first glance it may appear that repetitive tasks that followed regular rules these technologies can only destroy jobs by that could be codified (Autor, Levy, and replacing humans, careful consideration of Murnane 2003). As discussed earlier, this general equilibrium effects may indicate oth- explains the relative decrease in jobs inten- erwise. The simplistic view is that machines sive in RM tasks (mostly in manufactur- replace only workers who perform tasks. It is ing) and RC tasks (for example, clerks and based on the idea that there is a fixed num- bookkeepers). However, recent advances ber of tasks in an economy—the so-called in robotics and AI are threatening to go lump of labor fallacy—and as machines per- beyond routine tasks, encroaching on a set form more and more of these tasks, it comes of tasks that was thought to be the exclu- at the expense of human workers for whom sive domain of humans. In 2003, Autor, there will be fewer tasks to do. History, how- Levy, and Murnane surmised that driving ever, has taught a very powerful lesson: the jobs were relatively safe from automation number of tasks in an economy is not fixed, because driving required far too complex and in fact the total number of tasks has data processing, physical dexterity, situa- increased over time. Why? Several effects tional awareness, and improvisation. Today, must be considered. autonomous self-driving cars have logged If a firm introduces a new technology that thousands of miles on highways and city replaces workers, generally the productivity streets with huge success. Meanwhile, IBM’s of that firm will rise. In competitive markets, Watson has beat the champion on Jeopardy! this higher productivity would result in lower and can identify cancers with more accuracy marginal costs and therefore falling prices. than humans (Brynjolfsson and McAfee In turn, lower prices imply a higher demand 2014). Machines are successfully performing for that product or service. How much the legal searches and writing small journalistic demand increases depends on the specific articles. Techno-pessimists thus argue that price elasticity of that product or service. 8 this new wave of technological innovation If the demand for a specific product is elas- is encroaching on a whole new set of tasks: tic, it may lead to an increase in the level of nonroutine tasks, both cognitive and man- production (that is, the number of tasks to be ual. Therefore, machines could eventually performed) and thus could lead to more jobs (the time frame is not clear) replace humans being created in that firm or industry (see in many if not all tasks in the economy. box 3.2 for some examples). Another important effect to consider is that the increase in productivity and the potential Automation and jobs: General increase in production resulting from higher equilibrium effects demand in the original industry raises the Perhaps the best way to understand the full demand for all industries connected to the impacts of the introduction of new technol- original, both upstream and downstream. ogies is to consider the three questions that Thus new tasks will be created in industries 70   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n BOX 3.2  When automation creates jobs History has witnessed several examples of of fulltime equivalent bank tellers has grown technological innovations that automated since ATMs were widely deployed during production tasks in an industry and led to the late 1990s and early 2000s. Indeed, higher employment in that same industry. since 2000, the number of fulltime equiv- During the Industrial Revolution, the alent bank tellers has increased 2.0% per introduction of new machinery in the textile annum, substantially faster than the entire industry lead to the automation of about 98 labor force. Why didn’t employment fall? percent of the labor required to weave a yard Because the ATM allowed banks to operate of cloth. However, the number of weaving branch offices at lower cost; this prompted jobs actually increased (Bessen 2016). them to open many more branches (their Meanwhile, the productivity gains were so demand was elastic), offsetting the erstwhile significant that they drove the price of cloth loss in teller jobs.” down significantly. Coupled with the highly There are other examples as well. The elastic demand for clothes, it resulted in net number of cashiers in retail has increased job growth in the textile industry, despite the si nc e ba rcode s c a n ners were w idely automation of most production tasks. deployed during the 1980s, even though A similar story can be told about bank the scanners reduced cashiers’ checkout tellers after the introduction of automated times by 18–19 percent (Basker 2015). teller machines (ATMs) in the United States. Electronic document discovery software for The ATM performed many of the tasks per- legal proceedings clearly replaces the work formed by bank tellers such as cash handling of paralegals, and yet even as it has grown and simpler bank operations. As detailed in into a billion-dollar industry the number of the case study of Bessen (2016), “the number paralegals has grown robustly. that supply the original industry, as well as increase in productivity in one industry can industries that may use the products or ser- lead to the creation of new tasks in a com- production. One vices as inputs for their own ­ pletely different area of the economy. example could be the ­ transport and logistics Finally, the emergence of new technologies, industries, which would see more demand for particularly general-purpose ones,9 tends to their services (more tasks to be performed) create new jobs and tasks that do not even because of the increased production in the exist today. In the early 1900s, 41 percent of original industry. the US workforce worked in agriculture. One The potential effects are not even con- hundred years later (and several innovations fined to the industry where the innovation later), employment in agriculture is less than was introduced. The increase in p ­ roductivity 2 percent, and employment in health care, resulting from the adoption of new tech- finance, leisure, and entertainment (much of nology would lead to rises in income in the it in occupations that did not exist 100 years economy. These rises could lead in turn to ago) far outweighs the number of workers an increase in the demand for goods and in agriculture (Autor 2015). A more current services that are completely unrelated to the example is the internet. This innovation has original industry. For example, throughout not only revolutionized access to informa- history increases in the productivity of agri- tion, but also created entirely new indus- culture and manufacturing have led to a tries and jobs that did not exist 30 years ago higher demand for hospitality services such such as search engine optimizers10 or social as restaurants and hotels as well as leisure media managers. By definition, these effects and entertainment activities. Therefore, an are hard to measure and foresee, but history Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    71 teaches that new technologies generally lead It is important to consider not only the to new occupations and tasks that cannot total number of tasks created, but also even be imagined today. whether these tasks are ones that humans One important point relevant to develop- have the advantage in performing (thereby ing nations is that automation in developed creating more employment) or whether these nations may have indirect effects—that is, tasks can also be best performed by machines firms in developed nations adopt automa- (thereby not creating more employment for tion technologies that allow them to reshore humans). (the opposite of offshoring) production. A simple example illustrates this point Thus developing nations may suffer job more clearly. In an industry in which workers losses, or jobs may never emerge in the econ- perform two tasks, A and B, a new technology omy because advanced nations are reshor- is introduced that can fully automate task A. ing production by adopting labor-saving The increased productivity resulting from technologies. adoption of automation technology leads to a Although the evidence on this point is drop in the price of the good (or s ­ ervice) pro- scarce, Artuc, Christiaensen, and Winkler duced, and demand is elastic so that demand (2019) have investigated the labor market for the good increases overall. This increase impacts in Mexico of exposure to US auto- will lead in turn to an increase in the number mation. They find that the ratio of employ- of B tasks used as inputs. To the extent that B ment in the tradable sector to population is tasks are those in which humans have a rel- not affected by exposure to US automation or ative comparative advantage, this advantage by the decline in exports caused by US auto- could lead to more tasks being performed by mation. However, the average effect hides humans. However, if the B task is also suscep- differential effects observed in different local tible to automation, then even if the demand labor markets. On the one hand, areas that for such tasks increases it will not lead to initially had a relatively higher share of man- more employment for humans. ufacturing jobs susceptible to being replaced Thus to understand the total impact of by automation did experience a decline in the automation on employment it is important ratio of manufacturing employment to pop- to consider all three questions together. Not ulation. On the other hand, areas in which only is it important to consider all the possible the fraction of jobs susceptible to being auto- general equilibrium effects that can result in mated was low experienced an increase in the more tasks being created in an economy, but manufacturing employment to population it is equally important to assess whether the ratio. new tasks being created are those in which humans have an advantage in performing them or whether machines can also replace Automation and jobs: Humans working workers in performing them. As detailed in against machines or humans working the models explaining labor market polar- with machines? ization, the introduction of technologies that The previous section established that the automate certain tasks—principally the rou- number of tasks in an economy is not fixed tine tasks of production—raises the value of and the adoption of new technologies can complementary tasks—generally nonroutine in fact lead to more tasks being created. tasks. As long as humans retain the compar- However, assessing whether this implies ative advantage in performing these comple- more employment opportunities for humans mentary tasks, then automation can lead to requires turning to questions 2 and 3 stated new jobs, raising the total employment level. earlier: It is important to note here that jobs and occupations generally do not consist of a sin- 2. What is the nature of these tasks? gle task. Instead, workers perform a whole 3. Who has the advantage in carrying out set of tasks—a bundle of tasks—and thus these tasks (humans or machines)? machines do not generally replace a whole 72   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n job or occupation but rather a subset of tasks, whole occupation will be automated or that allowing workers to perform other sets of the job will disappear entirely. The second tasks. The next section returns to this point approach, developed by Arntz, Gregory, and in more detail, discussing the susceptibility of Zierahn (2016) and called the task-based jobs to automation and how the risk of auto- approach, has produced estimates of the risks mation is measured in the literature. of automation that are significantly lower Finally, although the final result may be (9 percent for the United States). This section an economy that ends up with more tasks to briefly describes each approach. be performed and many of these new tasks will be performed by humans, there are Occupation-based approach likely to be significant adjustment costs. As Frey and Osborne (2017) based their analysis machines become more powerful, dexterous, on the 2010 version of the O*NET database and capable, the subset of tasks in which (box 3.1). This database describes the task humans retain an advantage may shrink over content of 903 occupations in the United time. The evidence indicates that these tasks States. Specifically, Frey and Osborne under- will require more cognitive, analytical, cre- took the following steps: ative, and interpersonal skills. Therefore, pol- icy makers need to consider the urgency of • Provided information on work-­oriented instilling in the workforce of the future these descriptors (such as worker character- higher-order skills. The policy implications istics, worker requirements, and expe- are discussed at the end of this chapter, but rience requirements) and job-­ oriented first the next section looks at how the aca- descriptors (such as occupational demic literature has taken on the challenge of requirements, workforce characteris- measuring how many jobs are at risk of dis- tics, and occupation-specific informa- appearing because of automation. tion) that account for tasks. • Asked experts and researchers of auto- mation technologies (such as machine Measuring the risk of automation: learning and mobile robotics) to Occupation-based versus task-based classify these occupations as either ­ approach automatable or not based on their task Although fears of mass technological unem- structures.11 ployment are not new and actually date • From these, selected only 70 occupa- back centuries, new fears were stoked by the tions about whose labeling the experts research of Frey and Osborne (2017). In their were highly confident. paper, they claimed that up to 47 percent of • Projected the automatability to the jobs in the United States were at risk of being rest of occupations by examining automated. Since then, a flurry of reports whether the classification of experts using different approaches and data have was systematically correlated with produced a wide range of estimates. But why nine objective attributes of occupa- do these estimates differ so much? tions that are related to the identified E s s e nt i a l ly, t h e re a r e t wo bro ad engineering bottlenecks (for example, approaches to measuring the risk of automa- manual dexterity, originality, and social tion of occupations. The first, the occupa- perceptiveness). tion-based approach, was developed by Frey • Applied a series of probabilistic models and Osborne (2017). Subsequent research has to examine the power of these bottle- criticized their approach, recognizing that neck-related attributes in predicting an occupations do not consist of a single task but occupation’s risk of automation. rather a bundle of tasks. Therefore, although • Applied these estimated probabilities a subset of tasks within an occupation may to the occupations that were not confi- be automated, that does not imply that the dently assessed by the experts. Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    73 • Divided occupations into three cat- in the PIAAC database based on the occupa- egories: low risk of automation (less tional codes. One important drawback of this than 30 percent), medium risk (30–70 approach is that only two-digit ISCO codes percent), and high risk (more than are available in the PIAAC database, and 70 percent). thus an assignment issue arises when match- ing occupations with the six-digit codes of Merging this information with the num- SOC. Thus the authors followed an iterative ber of people employed in each occupation in algorithm that assigned each individual in the the United States, Frey and Osborne (2017) data set the automatability with the highest arrived at the estimate of 47 percent of jobs probability based on this method. being at high risk of being automated—mean- This approach is less restrictive than the ing, in their words, that “associated occupa- occupation-based approach because it allows tions are potentially automatable over some for differences in task structures within unspecified number of years, maybe a decade occupations and specifically focuses on or two.” Interestingly, they found that the individual jobs. Moreover, the focus is on risk of automation was higher for low-skilled which tasks are at high risk of automation. workers and for low-wage occupations. Arntz, Gregory, and Zierahn (2016) found The main criticism of this approach is that that the automatability of jobs is lower in it focuses on occupations rather than on tasks jobs with high educational job requirements performed within an occupation. As just or jobs that require cooperation with other noted, occupations do not consist of a single employees or in which people spend more task but rather a bundle of tasks, and it is time i ­nfluencing others. At the other end, tasks that are at risk of being automated, not high-risk tasks are those related to exchang- occupations. Moreover, Frey and Osborne ing information, selling, or using fingers and (2017) implicitly assumed that all workers hands. These results are more in line with the within an occupation perform the same set of literature on tasks in which routine tasks are tasks. Using worker-level information on the susceptible to automation, whereas tasks tasks performed in an occupation reveals that interaction or cognitive tasks related to social ­ a worker’s task structures differ remarkably are less likely to be automated (Acemoglu within occupations (Autor and Handel 2013). and Autor 2011; Autor and Handel 2013). In general, the task-based approach pro- Task-based approach duces estimates that are far below those The alternative approach of Arntz, Gregory, presented by Frey and Osborne (2017). For ­ and Zierahn (2016) to measuring job losses example, although Frey and Osborne find stemming from automation is “based on the that 47 percent of jobs in the United States idea that the automatability of jobs ultimately are at high risk of being automated, Arntz, depends on the tasks which workers perform Gregory, and Zierahn (2016) find that only for these jobs, and how easily these tasks can 9 percent have a high probability (more than be automated.” Arntz , Gregory, and Zierahn 70 percent) of being automated. For OECD (2016) used individual-level PIAAC data, countries, they find that only 9 percent of which contain indicators on demographic jobs are at high risk of being automated. characteristics, skills, job characteristics, and job tasks and competencies. By using indi- Measuring the risk of automation: vidual-level data, the authors were able to Critiques incorporate possible differences in the task structure of workers within an occupation. These approaches are subject to several cri- They estimated the relationship between tiques. First, both approaches to measur- workers’ tasks and the risk of automation by ing the risk of automation are based on matching the automatability indicator of Frey the technical feasibility of automation and and Osborne (2017) to the US observations do not consider the economic desirability 74   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n of adopting these technologies. Thus even consumers may not resist ordering from though certain tasks or occupations may be computers or robots in fast-food restaurants, technically at risk of being automated, those but it is not clear whether they would accept results should not be equated with employ- such innovations in high-end restaurants. ment losses. Adoption of these technologies Similarly, preferences may be skewed toward will depend on the relative factor prices of provision of services by humans in health capital and labor. In fact, both approaches care, nursing, and elder care, for example. suggest that lower-skilled, low-wage occu- Moreover, the set of tasks performed by pations are technically more at risk of being workers is not fixed over time, even within automated (mostly because they perform occupations. For example, Autor, Levy, and more routine tasks). Yet by virtue of being Murnane (2003) and Spitz-Oener (2006) low-wage occupations, the price of capital found that as a response to the introduction will need to drop relatively more to make of automation technologies, workers adjusted automation economically attractive. their work time toward tasks that were com- Second, there is little consideration of plementary to the machines. Thus it is likely the speed of adoption of these technologies. that workers will change and adapt to the This is particularly relevant to developing new technologies to avoid being unemployed. countries. Adoption of new technologies Finally, both approaches are designed generally requires a broad set of complemen- to measure the risk of automation by occu- tarities such as physical and human capital. pations or tasks, but generally they do not In a recent publication, “The Innovation consider all of the general equilibrium effects Paradox: Developing-Country Capabilities described earlier. Neither approach consid- and the Unrealized Promise of Technological ers the possibility that productivity increases Catch-up,” Cirera and Maloney (2017, 2) could translate to higher demand in different state: “If a firm (country) invests in innova- areas of the economy, or that these techno- tion but cannot also import the necessary logical innovations could create a whole new machines, contract trained workers and set of occupations and tasks that do not exist engineers, or draw on new organizational today. techniques, the returns to that investment Now that the caveats associated with this will be low. In turn, the underlying condi- type of analysis have been described, the next tions that impede the accumulation of any section describes the findings of this investi- of these types of capital—such as the cost of gation into the risks of automation in LAC doing business, trade regime, competitive- economies. ness framework, or capital markets, as well as those seen as particular to innovation, Measuring the risk of automation: The such as intellectual property rights protec- LAC experience tion or market failures that disincentivize the accumulation of knowledge—affect the What is the percentage of jobs at risk of returns and hence the quantity of innovation automation in the LAC economies? The investment.” estimates presented here for Bolivia, Chile, Other factors to consider are the legal and and Colombia, and then all LAC coun- ethical barriers that impede or slow down the tries are based on the two approaches just adoption of new technologies. The canoni- described—the occupation-based approach cal example is that of driverless cars, which pioneered by Frey and Osborne (2017) and are facing legal challenges about liability in the task-based approach of Arntz, Gregory, case of an accident (Bonnefon, Shariff, and and Zierahn (2016)—using the PIAAC data Rahwan 2016; Thierer and Hagemann 2015). set for Chile. Estimates are based as well Also, preferences may be skewed toward the on the information available in the Skills provision of services by humans rather than Toward Employability and Productivity robots in certain businesses. For example, (STEP) surveys for Bolivia and Colombia, Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    75 which contain worker-level information on PIAAC data are at a two-digit level of aggre- the tasks performed at their jobs. Here an gation to describe occupations, whereas the approach similar to that of Arntz, Gregory, data from Frey and Osborne are disaggre- and Zierahn (2016) is followed, with some gated to a six-digit level, presented an assign- necessary adjustments because the data sets ment problem. To resolve the problem, the are not strictly comparable with PIAAC. study team followed the approach of Arntz, Because no information is available for Gregory, and Zierahn (2016), using an algo- the rest of the countries in the LAC region, rithm to assign the most likely risk level given estimates for these countries are provided the characteristics of the worker and the job. by imputing the results from the analysis of A comparison of the task-based and occu- Bolivia, Chile, and Colombia. Specifically, pation-based approaches reveals very differ- for each country with worker-level and task- ent risk profiles across jobs and occupations. level data, the percentage of workers within Most strikingly, the range of automation risk an occupation that are at high risk of auto- is highest when imputing the risk number mation (that is, more than 70 percent) is from Frey and Osborne (2017), 46 percent, determined. That number is then applied and lowest when applying the task-based to the other countries using their household approach, 6.5 percent. labor surveys. Although somewhat limited, Consistent with previous findings, the the analysis provides a range of estimates study results indicate that use of the Frey and for each country based on the methodolo- Osborne (2017) methodology generates a gies and data sources available. By imputing bipolar distribution of automation risk with the results from a different country, the only peaks close to the extremes (see figure 3.6). In source of differences among countries stems other words, the occupation-based approach from their differing occupational structures. suggests that a significant number of jobs What follows are the results for the three have a very low risk (less than 30 percent) of countries—Chile, Colombia, and Bolivia— automation and a high number of workers for which worker-level and task-level data have a high risk (more than 70 percent). The are available. Chile’s estimates are based distribution is relatively flat and low for jobs on the PIA AC data set of OECD, and in the middle-risk category (30–70 percent). Colombia’s and Bolivia’s are based on the By contrast, the task-based approach pro- STEP surveys of the World Bank. Although duces a smoother distribution with a peak at the purpose of the surveys is similar—iden- the lower end of the risk spectrum, suggesting tifying the tasks and skills required in the that many jobs are relatively safe from auto- workplace —there are some important mation. Although these estimates suggest that ­ differences in the questions asked and in the few jobs are at high risk of being automated specific responses available. Therefore, the (less than 7 percent), there is a significant results are not strictly comparable among number of jobs in the middle-risk category. these countries. Workers in these types of jobs are generally performing a bundle of tasks, some of which Results for Chile are at risk of being automated, while other The findings for Chile, based on the data tasks are thought to be safe. Thus, although available from the PIACC survey produced by these jobs are not at high risk of disappear- OECD and following closely the task-based ing, it is likely they will be significantly trans- approach developed by Arntz, Gregory, and formed as automation technologies become Zierahn (2016), are in line with the results more powerful. Therefore, workers will find found for OECD countries. In addition, the themselves needing to adapt to a workplace automation risk by occupation calculated with more technology that replaces some of by Frey and Osborne (2017) is imputed and their tasks, and they will need the flexibility applied to the data for Chile. The fact that to perform the tasks complementary to the in Arntz, Gregory, and Zierahn (2016) the work of machines. 76   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 3.6  Distribution of automatability across methodologies, Chile 0.03 Estimated kernel density 0.02 0.01 0 20 40 60 80 100 Automatability (%) Occupation-based approach Task-based approach Source: Original calculations for this publication using Frey and Osborne (2017) values and PIAAC data for Chile from Organisation for Economic Co-operation and D ­ evelopment (https://www.oecd.org/skills/piaac/). Note: In the occupation-based approach, Frey and Osborne values are applied to ISCO occupations in PIAAC’s Chile data, using identical weights for each six-digit SOC occupation within the corresponding two-digit ISCO occupation. ISCO = International Standard Classification of Occupations; PIAAC = ­Programme for the International Assessment of Adult Competencies; SOC = Standard Occupational Classification. Results for Colombia those for Chile and those for Bolivia. In par- The results for Colombia are based on the ticular, the occupation-based approach and worker-level data available in the STEP sur- the task-based approach yield closer results. veys produced by the World Bank. Although Although this finding is not reflected in the these surveys serve the same purpose as the percentage of high-risk jobs—the Frey and PIAAC surveys, there are some significant Osborne (2017) imputation suggests that differences in the specific questions asked 48 percent of jobs are at high risk, whereas and, more important, in the format of the the task-based approach is about half, 24.6 available responses. Therefore, the task-based percent—the risk distribution profiles are methodology of Arntz, Gregory, and Zierahn not quite as dissimilar as they were for Chile (2016) had to be adapted to the specific for- (see figure 3.7). mat of the STEP surveys.12 The task-based approach using STEP data The occupation-based approach gener- generates a distribution with a peak close to ates a similar distribution to that of Chile (but below) the 70 percent cutoff point. Thus and Bolivia: a bipolar distribution, with the it suggests that a large mass of workers are mass concentrated in the low-risk and high- in occupations in which a significant num- risk categories, whereas there is little mass ber of the tasks they perform are at risk of in the middle-risk category. For Colombia, a being automated. The results are somewhat smaller mass of employment is concentrated worrisome because the introduction of new in the low-risk category relative to Chile and technologies will require workers to gain Bolivia. According to this methodology, 48 the flexibility to successfully adapt to and percent of jobs are at risk of being automated. perform the more complex tasks that are The results for Colombia using the task- complementary to machines. As noted later based approach are very different from in this chapter, it would be advisable for Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    77 FIGURE 3.7  Distribution of automatability across methodologies, Colombia 0.04 0.03 Estimated kernel density 0.02 0.01 0 20 40 60 80 100 Automatability (%) Occupation-based approach Task-based approach Source: Original calculations for this publication using Frey and Osborne (2017) values and data for Colombia from World Bank’s STEP Skills Measurement Program (https://microdata.worldbank.org/index.php/catalog/step/about). Note: In the occupation-based approach, Frey and Osborne values are applied to ISCO occupations in STEP data, using identical weights for each six- digit SOC occupation within the corresponding two-digit ISCO occupation. ISCO = International Standard Classification of Occupations; SOC = Standard ­Occupational Classification; STEP = Skills Toward Employability and Productivity. authorities to manage these risks by invest- is concentrated in the high-risk category. ing in education and on-the-job-training pro- The Frey and Osborne approach yields an grams that can help workers adapt to the new estimate that almost 50 percent of jobs will technologies and the significant changes that disappear to automation. Interestingly, the may be coming to their occupations. results for Bolivia are quite different from those for both Colombia and Chile. The Results for Bolivia distribution displayed in figure 3.8 shows a The results for Bolivia are based on the more uniform distribution, with a relatively worker-level data available in the STEP sur- ­ small peak close to (but below) the 70 percent vey produced by the World Bank. Although cutoff. Consequently, the estimate based on these surveys serve the same purpose as the the task-based approach is lower than that PIAAC surveys, there are some significant for Colombia, 16.7 percent. differences in the specific questions asked Although it is hard to pinpoint all the and, more important, in the format of the potential sources for the differences between available responses. Therefore, the task-based Colombia and Bolivia, two factors play a methodology of Arntz, Gregory, and Zierahn major role. First, the occupational structures (2016) had to be adapted to the specific for- of the two countries are different (for exam- mat of the STEP survey.13 The results, how- ple, the manufacturing sector is larger in ever, are comparable between Colombia and Colombia). Second, the task structure within Bolivia. occupations may be different. In particular, As it does for the other countries in the set of tasks performed by workers within the sample, the Frey and Osborne (2017) an occupation appears to be more heteroge- approach produces a bipolar distribution. neous in Bolivia, involving more tasks that However, in the case of Bolivia more mass are difficult to automate. 78   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n FIGURE 3.8  Distribution of automatability across methodologies, Bolivia 0.03 Estimated kernel density 0.02 0.01 0 20 40 60 80 100 Automatability (%) Occupation-based approach Task-based approach Source: Original calculations for this publication using Frey and Osborne (2017) values and data for Colombia from World Bank’s STEP Skills Measurement Program (https://microdata.worldbank.org/index.php/catalog/step/about). Note: In the occupation-based approach, Frey and Osborne values are applied to ISCO occupations in STEP data, using identical weights for each six- digit SOC occupation within the corresponding two-digit ISCO occupation. ISCO = International Standard Classification of Occupations; SOC = Standard ­Occupational Classification; STEP = Skills Toward Employability and Productivity. Results for all LAC countries imputed probabilities derived from Chile This section turns to a larger sample of coun- using the PIAAC data set; and third and tries for which automation probabilities were fourth, the probabilities derived from the imputed by occupation using the results for analysis of Bolivia and of Colombia using Bolivia, Chile, and Colombia. STEP data. For the three countries with Assessment of the number of jobs at task-related data, the numbers are based on high risk of being automated in the larger the specific survey data (STEP and PIAAC), sample of countries is based on the results and, for cross-country comparability, the obtained from the estimations for Bolivia, imputed numbers merged with the household Chile, and Colombia and the results of Frey survey data are included. and Osborne (2017). The estimated risk In interpreting the results, it is import- probabilities by occupation are paired with ant to note that because the probabilities of the occupational structure of each country, automation are imputed by occupation, the which is derived from the harmonized house- differences across countries stem solely from hold survey data available from SEDLAC for the different occupational structures. So, for 16 countries in the LAC region. The results example, the difference between the Frey and identify the percentage of workers within an Osborne (2017) estimates for Argentina and occupation that are at high risk of automa- Uruguay are attributable to Argentina hav- tion according to the methodologies and data ing fewer workers employed in occupations sets used. Thus for each country in the sam- that are at high risk—according to Frey and ple, four different assessments are presented Osborne—of being automated. of the number of jobs at high risk: first, the The results show some clear patterns (see numbers derived from Frey and Osborne’s figure 3.9). Clearly, the occupation-based (2017) occupation-based approach; second, approach consistently produces the highest Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    79 FIGURE 3.9  Risk of automation by LAC country, based on four methodologies 60 Percent of jobs at high risk (>70%) 50 40 30 20 10 0 y ile ico a a ua a u ca a as r r lic ua il y do do liv bi am r in ua az Pe Ch Ri ur ub g ex ag om nt Bo lva ua ra Br ug n nd ta M ep ge r Pa ca Ec Pa l Sa s Ur Co Ho Co nR Ar Ni El ica in m Do Arntz et al., Chile, imputed STEP, Colombia, inputed STEP, Bolivia, imputed Frey and Osborne, imputed Sources: Original calculations for this publication using 2016 PIAAC data from Organisation for Economic Co-operation and Development (https://www.oecd.org​ /skills​/piaac/); data for Colombia and Bolivia from World Bank’s STEP Skills Measurement Program (https://microdata.worldbank.org/index.php/catalog/step​/about); Socio-Economic Database for Latin America and the Caribbean (SEDLAC) household surveys, CEDLAS and World Bank (https://datacatalog​.­worldbank.org/dataset​ /socio-economic-database-latin-america-and-caribbean). Note: Study team calculations follow the methodology of Frey and Osborne (2017) and Arntz, Gregory, and Zierahn (2016). Numbers indicate the percentage of jobs at high risk of being automated, using a cutoff of 70 percent as is standard in the literature. CEDLAS = Center for Distributive, Labor and Social Studies; LAC = Latin America and the Caribbean; STEP = Skills Toward Employability and Productivity. estimates, and thus could be interpreted as that is wider than those for the previous an upper bound. In the sample, the estimates two methodologies. The results indicate across countries range from a minimum of that Argentina, at 18.3 percent, faces the 45.1 percent for Panama to 58 percent for least number of jobs, while Ecuador has the El Salvador; the average for the region is highest, with over 40 percent of the work- 50 percent. Ecuador, Honduras, Mexico, force at high risk. The average for the region and El Salvador seem to have more workers using this methodology is 29.8 percent. employed in occupations that are more likely Argentina, Chile, and Colombia appear to to be automated. At the other end, countries be significantly below the regional average, such as Argentina, Chile, and Panama seem whereas Ecuador, Honduras, Mexico, Peru, to have slightly fewer jobs at risk. Paraguay, and El Salvador are highest in the Use of the probabilities derived from the risk rankings. PIAAC data set for Chile results in the lowest Finally, use of the numbers from the estimates of jobs at risk, and so can be inter- Bolivia STEP survey produces a range that preted as the lower bound. The estimates is not as wide as that using the Colombia range from a low of 6 percent for Bolivia numbers. Once again Argentina, at 19.2 to a high of 12 percent for El Salvador; the percent, has the least number of jobs at risk, average for the region is 9.2 percent. Once whereas El Salvador accounts for the maxi- again, Argentina and Chile are the countries mum, 32.2  percent. The regional average facing below-average risk, whereas Ecuador, is 25.8  percent. The results indicate that Mexico, El Salvador, and Uruguay display Argentina, Brazil, and Colombia have the higher numbers of jobs at risk. least number of jobs at risk, and Ecuador, Apply i ng t he e s t i m ate s u si ng t he Honduras, and El Salvador have the highest Colombia STEP data set produces a range number of jobs at risk. 80   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n Clearly, the methodological approach automate and therefore result in fewer jobs used to assess how many jobs are at risk being at risk. This finding should be of great of automation is of vital importance, and concern to the less developed countries in the the choice is not innocuous. The occupa- region, which, according to the estimates, are tion-based approach developed by Frey and facing higher risks of automation. Osborne (2017) consistently produces the Several patterns also cut across the highest probabilities, and the range of esti- empirical analysis. First, the occupa- mates is narrow, not varying significantly tion-based approach produces a bipolar with the occupational structure of each coun- distribution that concentrates the mass of try. According to these estimates, which are workers in the low-risk category and mostly interpreted as an upper bound, about 50 per- in the high-risk category. Differences among cent of jobs in the LAC region are at risk of the three countries at the heart of the analy- disappearing because of automation. sis are driven solely by their different occu- Beyond the methodological approach, it pational structures. Chile has the lowest is apparent that the data source matters as share of workers at risk with 46.3 percent, well. Use of different surveys, even if all esti- then Colombia with 48.3 percent, and finally mate skills and tasks in the workforce, yields Bolivia with 49.7 percent. different estimates as well. The PIAAC data Second, the risk of automation is nega- set for Chile consistently produces the lowest tively correlated with both education and estimates for each country, and once again income (see figure 3.10, panels a and b). the range of estimates is very narrow, virtu- The  visual results are confirmed by the ally unaffected by the occupation structure statistical analysis. Workers who are less of countries. Thus the results based on this educated and earn less tend to work in occu- methodology and data source are interpreted pations that involve more manual and rou- as the lower bound. In general, these results tine tasks—the very tasks associated with suggest that the fear of mass technological a high degree of automation. On the other unemployment are wildly overblown because hand, workers who are more educated tend only 10 percent of jobs could be at risk. to work in occupations that have higher Data from the STEP surveys, which intensity of cognitive/analytical tasks, as are comparable, yield estimates that are well as complex social interactions such as for the most part similar when comparing teamwork, negotiation, and creative prob- the numbers from the Bolivia analysis or lem solving. the Colombia analysis. The exceptions are F i n a l l y, a lt h o u g h t h e t a s k- b a s e d Bolivia, Ecuador, and Paraguay, where the approach produces a much smaller number estimates differ by more than 10 percentage of jobs at high risk of being automated, all points. three countries display a significant mass of However, cutting across methodologies workers who are close to the cutoff. This and data sources, some patterns emerge. For finding suggests that although the jobs may example, some of the more advanced coun- not be at risk of disappearing, they will be tries in the region, such as Argentina and highly susceptible to the introduction of Chile, consistently display the lowest esti- new technologies. In other words, there are mated number of jobs at risk. At the other many workers whose workday involves a lot end, countries such as Ecuador, Honduras, of tasks that will be automatable in the near and El Salvador, some of the least developed future. Therefore, these workers will need countries in the region, consistently display a to adapt soon to the new technologies and larger number of jobs at risk. It appears, then, shift their task load toward the more dex- that higher levels of development are associ- terous, complex, and cognitive tasks. This ated with an occupational structure in which may require additional skills and capabili- tasks are more complex or more difficult to ties from workers. Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    81 FIGURE 3.10  Automation risk by selected characteristics, LAC region a. Education level b. Income level 100 100 Percent of workers at high risk (>70%) Percent of workers at high risk (>70%) 80 80 60 60 40 40 20 20 0 0 Primary or less Secondary Technical Tertiary <10–25% 25–50% 50–75% 50–75% Level of education Income percentile Sources: Original calculations for this publication using 2016 PIAAC data from Organisation for Economic Co-operation and Development (https://www.oecd.org/skills/piaac/); data for Colombia and Bolivia from World Bank’s STEP Skills Measurement Program (https://microdata.worldbank.org/index.php/catalog/step/about); Socio-Economic Database for Latin America and the Caribbean (SEDLAC) household surveys, CEDLAS and World Bank (https://datacatalog.worldbank.org/dataset/socio-economic-database-latin-america-and​ -­caribbean). CEDLAS = Center for Distributive, Labor and Social Studies; LAC = Latin America and the Caribbean; STEP = Skills Toward Employability and Productivity. Looking into the future: Digital Marketplace, many clusters of rural e-shops platforms and the nature of work have emerged. These entrepreneurs produce goods, agricultural products, and handicrafts The future may bring the emergence of based on their niche competencies. It is esti- another potential disruption in labor mar- mated that Taobao villages have created more kets: the rise of digital platforms as a new than 1.3 million jobs in rural communities avenue for workers to supply labor. For most (World Bank 2019, 39). technological innovations, these platforms Digital platforms not only provide benefits may present significant opportunities, but for entrepreneurs selling products online, but effectively benefiting from them may also also expand market access for professionals present significant challenges. and service providers. Workers can partici- On the positive side, digital platforms pate in multiple online platforms for a rela- reduce the cost of entry for entrepreneurs tively low entry cost and freelance, reaching and expand access to large markets. With millions of customers. This is a huge oppor- only a smartphone and access to the internet, tunity for a region such as Latin America and any entrepreneur can now engage with local, the Caribbean, where most countries share a regional, and even global markets. In turn, language and have similar cultural and insti- successful businesses can scale up quickly tutional backgrounds that can facilitate trade and foster job creation. This development in professional services. may be particularly important and relevant From the perspectives of consumers, for rural communities where employment there are also many potential benefits. outside agricultural activities may be limited. For one thing, greater competition among An example of the huge opportunities for entrepreneurs can result in lower prices. For ­ rural development is the “Taobao villages” another, consumers can now access a better experience in China. On the Taobao.com variety and quality of products and services. 82   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n Apparently, consumers trust these platforms Also important are minimum quality stan- because they can rely on brand certification dards and safety regulations. Moreover, and consumer reviews to make informed policy makers should establish the legal decisions. framework for taxation of transactions One of the major opportunities provided within their country’s border, but also for by digital platforms is expansion of the labor cross-border taxation and liability issues. supply, thereby contributing to regional These are all examples of the regulatory growth. Although data are limited, particu- infrastructure that needs to be in place to fos- larly for developing countries, most workers ter the inclusive growth of digital platforms in advanced nations use digital platforms to while protecting all participants in these earn secondary income. Workers have the markets. flexibility and autonomy to set their own Finally, the greater supply of labor avail- hours according to their needs and the sched- able through digital platforms has opened ules of their main occupations. This flex- an important debate on whether to consider ibility and autonomy may be particularly these workers as employees of the digital important for women who may be out of the platform or as independent contractors or labor force or have limited hours because of freelancers. Moreover, the rise of these alter- their home care duties. native forms of labor supply may threaten the In the context of the LAC region, another sustainability of the traditional social insur- potential benefit of digital platforms is that ance model. As more and more ­ workers— transactions are conducted digitally and both skilled and unskilled—participate in thus create an electronic record. This would these platforms, the social insurance mech- allow—in principle—for the taxation of anism that relies on employer-employee these transactions, many of which occur ­ contributions to finance social protection today on the informal side of the economy, will slowly degrade. From the perspective of thereby escaping taxation. This issue should the LAC region, which already has high lev- not be undervalued because LAC economies els of informality in the labor market, this is are notoriously fiscally constrained, and particularly worrisome. Policy makers in the changes in the labor market may affect the region must think creatively about alterna- sustainability of the traditional social insur- tive social insurance models that do not rely ance system—an issue discussed shortly in on financing and benefits attached to formal more detail. employer–employee relationships. In other For all the potential benefits offered by words, policy makers need to both define digital platforms, they present policy makers the level of social protection and insurance with significant challenges. Obviously, the that will be provided to citizens regardless expansion of digital commerce requires reli- of their labor status and relationship (that able and affordable internet connectivity and is, employee, contractor, or freelance) and high penetration of smartphones. Therefore, find alternative financing mechanisms that success in digital commerce depends on do not depend on the employee–employer countries investing in and expanding their relationship. telecommunications infrastructure—and especially in rural communities if countries would like to replicate the success of Taobao Conclusions and policy villages. implications Another challenge is setting up a regula- In general, this analysis has found that the tory framework that establishes clear and risk of mass technological unemployment is fair rules for all participants. Among sev- low for the LAC economies. In addition, the eral issues is the need for clear rules on the slow adoption rate for these technologies sug- ownership of data and privacy rules for both gests that massive changes in the workplace consumers and providers on these platforms. are not likely to occur in the next decade. Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    83 However, this analysis also suggests that workers performing the simpler, more rou- many jobs will be affected and transformed tine tasks are most at risk of being replaced by the emergence of these technologies. by machines. By virtue of being in lower-paid And although these jobs may not disappear occupations, such tasks may in the short run completely, many of the tasks performed by be less likely to be automated because the humans today will likely be performed by prices of robots and automation technologies machines in the future. Workers will be inter- need to drop further for adoption to be eco- acting with more machines and increasingly nomically desirable. However, in the medium complex technologies. Therefore, they will and long term these tasks are at high risk of need the capabilities and skills to adjust to being fully automated. these new demands. Thus investing in the human capital of There is a growing consensus that the the workforce should be a priority for policy demand for skills in the labor market is makers. Investing in early childhood educa- changing. These changes have been under tion reaps the highest return on investments, way over the last two decades in advanced and the advantages grow over time because economies, and because technology is being learning and skill development are cumu- adopted in developing countries, these lative. In fact, Engle et al. (2011) find that changes are beginning to occur there as well. every additional US$1 invested in quality Reinforcing these changes are new technol- early childhood education programs yields a ogies that are emerging and threatening to return of US$6–$17. When quality and access replace humans, mostly in the simpler, more are ensured, investments in early childhood routine tasks they perform at work. education also increase equity, and there are According to the World Development several examples of successful programs in Report 2016: Digital Dividends, the skills the LAC region. Cash transfers that increase required for the modern economy go beyond the take-up of early childhood education pro- the foundational cognitive skills such as basic grams have fostered language development in literacy and numeracy. Some of the most val- Ecuador and Mexico. Chile’s Crece Contigo ued skills that also have a low risk of auto- program integrates the health, education, mation are the nonroutine, higher-­ o rder welfare, and protection services available as ones. These skills are related to the ability of the first prenatal visit of the mother. to understand complex concepts, learn from Although the LAC region has made sub- experience, adapt to new situations, and stantial progress in improving access to sec- more generally solve problems by using criti- ondary education, the quality of education cal thinking. The need for nonroutine inter- continues to lag that of advanced nations and personal, socioemotional skills is also on the developing country peers in East Asia. Thus rise. As stated in the World Development the focus should be on increasing the quality Report 2016 , “Socioemotional skills (also of secondary education and preparing stu- called soft or noncognitive skills) encompass dents for further education, whether in voca- a broad range of malleable skills, behaviors, tional trade schools or university. attitudes, and personality traits that enable Meanwhile, the demand for higher-­ order individuals to navigate interpersonal and nonroutine cognitive skills is increasing. social situations effectively. These include grit Tertiary education is therefore becoming or the perseverance to finish a job or achieve more important for the future of work. a long-term goal, working in teams, punctu- Not only does it impart the technical skills ality, organization, commitment, creativity, required for certain occupations, but it also and honesty” (World Bank 2016, 213). fosters development of the complex prob- As revealed in this analysis and consistent lem-solving, critical thinking, and advanced with the literature, education continues to be communications skills that are transfer- the best asset to insure against the risks of able across jobs and occupations. Tertiary automation. The low-paid and uneducated education also builds the transferable 84   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n sociobehavioral skills—such as teamwork, which combines learning with on-the-job resilience, and self-­ confidence—that have training, has shown that the probability of also seen higher demand in the labor mar- formal employment and earnings increases in ket. Policy makers should focus on how to the short term, and it has seen the benefits improve the access to and quality of tertiary sustained over time. systems (both trade schools and universi- ties) in order to improve the capabilities of the future workforce. A deep analysis of Notes the tertiary system in the LAC region can   1. See, among others, Brynjolfsson and ­ McAfee be found in the report At a Crossroads: (2014); McKinsey Global Institute (2017); Higher Education in Latin America and the World Bank (2016, 2019); World Economic Caribbean by Ferreyra et al. (2017). Forum (2018). Finally, what may become more important  2. The recent phenomenon of labor market as new automation technologies are adopted polarization has been documented by Autor, in LAC countries is adult learning. Although Katz, and Kearney (2008) and Autor and the time frame for the adoption of technology Dorn (2013) for the United States, and Goos and Manning (2007) for the United Kingdom. is unclear, it is possible that transformations Job polarization has also been documented for in the workplace will happen midcareer for Germany (Dustmann, Ludsteck, and Schön- many workers, and they will need to adapt, berg 2009; Spitz-Oener 2006), and there are particularly by changing the set of tasks per- indications it is pervasive in European coun- formed at work. Governments should have tries (Goos, Manning, and Salomons 2009; programs that help workers upskill and Michaels, Natraj, and Van Reenen 2013). retrain for the new jobs and minimize their   3. The Skills Toward Employment and Produc- adjustment costs. Meanwhile, the design of tivity (STEP) survey is not available for the adult learning programs should take into LAC region. account the constraints often facing adults in   4. Using the STEP survey, Dicarlo et al. (2016) terms of time, financial resources, and com- show that the task content in the United States and developing economies is generally similar peting demands. Meanwhile, behavioral and for high-skilled occupations, while remark- neuroscience research has discovered that ably different for routine-based occupations. the adult brain learns differently.   5. Croatia, Czech Republic, Estonia, ­ Hungary, The success of these types of programs Latvia, Lithuania, Poland, Romania, ­ Slovakia, already in the region is mixed. Argentina’s and Slovenia. Entra21 program is providing adult skills   6. Many of the surveys in the region are urban training and internships resulting in and thus underrepresent the agriculture 40 ­percent higher earnings for its participants sector. (J-PAL 2017). In Peru, a female entrepre-   7. Moore’s Law asserts that computer power neurship program did not generate signifi- doubles every 18 months. cant effects on employment. Similarly, in the   8. If a demand for a product is inelastic, then the increase in quantity demanded does not fully Dominican Republic the Juventud y Empleo compensate for the fall in prices and revenues. program did not increase employment, If demand is elastic, the increase in demand is although it improved noncognitive skills and proportionally higher than the fall in prices, job formality. The evidence suggests that revenues increase, and more workers will be adult learning programs are most successful hired to produce more units. when they are tied to employment opportu-   9. General-purpose technologies are defined as nities. Thus programs that include appren- “deep new ideas or techniques that have the ticeships and internships in partnership with potential for important impacts on many sec- the private sector will generally have more tors of the economy” (Wright 2000). lasting and significant effects. For exam- 10. Specialists who help website providers secure ple, Colombia’s Jovenes en Acción program, high rankings on the results pages of search engines such as Google. Ec o n o m i c t r a n s f o r m a t i o n , s k i l l s , a n d t h e f u t u r e o f w o r k    85 11. Specifically, experts were asked, “Can the tasks Autor, D. H., L. F. Katz, and M. S. Kearney. 2006. of this job be sufficiently specified conditional “The Polarization of the U.S. Labor Market.” on the availability of big data, to be performed American Economic Review 96: 1553–97. by state-of-the-art computer-­ controlled equip- Autor, D. H., L. F. Katz, and M. S. Kearney. ment?” (Frey and Osborne 2017). 2008. “Trends in US Wage Inequality: Revising 12. For technical details, see Beylis and Cuevas the Revisionists.” Review of Economics and (2019). Statistics 90 (2): 300–23. 13. For technical details, refer to Beylis and Autor, D. H., F. Levy, and R. J. Murnane. 2003. ­Cuevas (2019). “The Skill Content of Recent Technological Change: An Empirical Exploration.” Quarterly Journal of Economics 118 (4): 1279–333. References Basker, E. 2015. “Change at the Checkout: Tracing Acemoglu, D., and D. Autor. 2011. “Skills, the Impact of a Process Innovation.” Journal of Tasks and Technologies: Implications for Industrial Economics 63 (2): 339–70. Employment and Earnings.” Handbook of Bessen, J. E. 2016. “How Computer Automation Labor Economics 4: 1043–171. Affects Occupations: Technology, Jobs, and Aedo, C., J. Hentschel, J. Luque, and M. Moreno. Skills.” Law and Economics Research Paper 2013. “From Occupations to Embedded 15–49, Boston University School of Law. Skills: A Cross-Country Comparison.” Policy Beylis, Guillermo, and Maria Ignacia Paz Cuevas Research Working Paper 6560, World Bank, de Saint Pierre. 2020. “Measuring Job Loss Washington, DC. Risk to Automation in LAC: Methods and Arntz, M., T. Gregory, and U. Zierahn. 2016. Data Sources.” Background technical paper, “The Risk of Automation for Jobs in OECD World Bank, Washington, DC. C ou nt r ie s: A C ompa rat ive A n a ly si s .” Bolt, J., R. Inklaar, H. de Jong, and J. Luiten OECD Social, Employment, and Migration van Zanden. 2018. “Rebasing ‘Maddison’: Work i ng Pap er 189, O r ga n i s at ion for New Income Comparisons and the Shape of Economic Co-operation and Development, Long-run Economic Development.” Maddison Paris. Project Working Paper 10, Groningen Growth Artuc, E., L. Christiaensen, and H. Winkler. and Development Centre (GGDC), Groningen, 2019. “Does Automation in Rich Countries The Netherlands. Hurt Developing Ones? Evidence from US Bonnefon, J. F., A. Shariff, and I. Rahwan. and Mexico.” Policy Research Working Paper 2016. “The Social Dilemma of Autonomous 8741, World Bank, Washington, DC. Vehicles.” Science 352 (6293): 1573–76. Autor, D. 2014. “Skills, Education, and the Rise Brynjolfsson, E., and A. McAfee. 2014. The Second of Earnings Inequality among the ‘Other Machine Age: Work, Progress, and Prosperity 99 Percent.’” Science 344 (6186): 843–51. in a Time of Brilliant Technologies. New York: Autor, D. 2015. “Polanyi’s Paradox and the Shape Norton. of Employment Growth.” In Reevaluating Cirera, X. and W. F. Maloney. 2017.  The Labor Market Dynamics, Economic Policy Innovation Paradox: Developing-Country Proceedings,  129–77.  Kansas City, MO: Capabilities and the Unrealized Promise of Federal Reserve Bank of Kansas City. Technological Catch-Up. Washington, DC: Autor, D., and D. Dorn. 2009. “Inequality and World Bank. Specialization: The Growth of Low-Skill Dicarlo, E., S. Lo Bello, S. Monroy-Taborda, Service Jobs in the United States.” NBER A. M. Oviedo, M. L. Sanchez Puerta, and Working Paper 15150, National Bureau of I. V. Santos. 2016. “The Skill Content of Economic Research, Cambridge, MA. Occupations across Low and Middle Income Autor, D., and D. Dorn. 2013. “The Growth of Countries: Evidence from Harmonized Data.” Low Skill Service Jobs and the Polarization of IZA DP7 (102240). the US Labor Market.” American Economic Duernecker, G., and B. Herrendorf. 2017. Review 103 (5): 1553–97. “Structural Transformation of Occupation Autor, D. H., and M. J. Handel. 2013. “Putting Employment.” https://economicdynamics.org​ Tasks to the Test: Human Capital, Job Tasks, /meetpapers/2017/paper_1239.pdf. and Wages.” Journal of Labor Economics Dustmann, C., J. Ludsteck, and U. Schönberg. 31 (S1): S59–S96. 2 0 0 9. “R e v i sit i n g t he G er m a n Wa ge 86   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n Structure.” Quarterly Journal of Economics Maloney, W. F., and C. Molina. 2016. “Are 124 (2): 843–81. Automation and Trade Polarizing Developing Eden, M., and P. Gaggl. 2015. “On the Welfare Country Labor Markets, Too?” World Bank, I mpl i c at io n s of Autom at io n .” Pol i c y Washington, DC. Research Working Paper 7487, World Bank, McK inley, R. A ., ed. 1958. The Cit y of Washington, DC. Leicester: A History of the County of Engle, P. L., L. C. H. Fernald, H. Alderman, Leicester. Vol 4. Victoria County History J.  R.  Behrman, C. O’Gara, A. Yousafzai, Series. Martlesham, Suffolk, UK: Boydell and M. Cabral de Mello, et al. 2011. “Strategies Brewer. for Reducing Inequalities and Improving McKinsey Global Institute. 2017. “Jobs lost, Jobs Developmental Outcomes for Young Children Gained: Workforce Transitions in a Time of in Low-Income and Middle-Income Countries.” Automation.” December. Lancet 378 (9799): 1339–53. Messina, J., A. M. Oviedo, and G. Pica. 2016. “Job Ferreyra, M., C. Avitabile, J. Botero Álvarez, Polarization in Latin America.” Unpublished F. Haimovich Paz, and S. Urzúa. 2017. At a paper. World Bank, Washington, DC. Crossroads: Higher Education in Latin Michaels, G., A. Natraj, and J. Van Reenen. 2013. America and the Caribbean. Washington, DC: “Has ICT Polarized Skill Demand? Evidence World Bank. from Eleven Countries over 25 Years.” Review Frey, C. B., and M. A. Osborne. 2017. “The of Economics and Statistics 96 (1): 60–77. Future of Employment: How Susceptible Are Minnesota Population Center. 2019. Integrated Jobs to Computerisation?” Technological Public Use Microdata Series, International: Forecasting and Social Change 114 (January): Version 7.2 (data set). IPUMS, Minneapolis, 254–80. MN. https://doi.org/10.18128/D020.V7.2. Goos, M., A. Manning, and A. Salomons. 2014. Spitz-Oener, A. 2006. “Technical Change, Job “Explaining Job Polarization: Routine-Biased Tasks, and Rising Educational Demands: Technological Change and Offshoring.” Looking Outside the Wage Structure.” Journal American Economic Review, 104 (8): 2509-26. of Labor Economics 24 (2): 235–70. Susskind, Daniel, and Richard Susskind. 2015. Goos, M., and A. Manning. 2007. “Lousy Professions: How Technology Will Transform and Lovely Jobs: The Rising Polarization of the Work of Human Experts. Oxford, UK: Work in Britain.” Review of Economics and Oxford University Press. Statistics 89 (1): 118–33. Thierer, A., and R. Hagemann. 2015. “Removing Goos, M., A. Manning, and A. Salomons. 2009. Roadblocks to I ntelligent Vehicles and “Job Polarization in Europe.” American Driverless Cars.” Wake Forest Journal of Law Economic Review 99 (2): 58–63. and Policy 5: 339. Hardy, W., R. Keister, and P. Lewandowski. World Bank. 2016. World Development Report 2016. “Technology or Upskilling? Trends in 2016: Digital Dividends. Washington, DC: the Task Composition of Jobs in Central and World Bank. Eastern Europe.” HKUST IEMS Working World Bank. 2019. World Development Report Paper No. 2016 – 40, BS Working Paper 2019: The Nature of Work. Washington, DC: Series, Institute of Structural Research (IBS), World Bank. Warsaw, Poland. World Economic Forum. 2018. The Future of Hardy, W., R. Keister, and P. Lewandowski. Jobs Report 2018. Cologny, Switzerland. 2018. “Educational Upgrading, Structural Wright, Gavin. 2000. “Review of Helpman Change and the Task Composition of Jobs (1998).” Journal of Economic Literature in Europe.” Economics of Transition 26 (2): 38 (March): 161–62. 201–31. Zeng Kunhua. 1973. Zhongguo tie lu shi J-PAL (Abdul Latif Jameel Poverty Action [ The Hi stor y of C hin ese R ailway Lab). 2017. “Skills Training Programmes.” Development]. Vol. 1. Taipei, Taiwan, Cambridge, MA, August. China: Wenhai Press. Conclusions 4 T he Latin America and the Caribbean mass “technological unemployment”—as (LAC) region is facing important these concerns are now labeled—is unlikely, challenges. After a decade of rapid the labor market is undergoing a major growth and strong improvements in social transformation, and government action is indicators, growth has stalled, and external urgently needed to prepare the workforce conditions do not appear to be favorable, at for the future. least in the short and medium term. Trade flows have slowed amid elevated tensions, foreign direct investment (FDI) has fallen Structural transformation: Past off, financing conditions are tightening, and and future all of this is happening in the context of vul- In the LAC region, structural transformation nerable fiscal conditions for governments has contributed negatively to productivity in the region. Commodity prices, which growth. The relative share of employment helped fuel growth during the so-called in services—the sector with the lowest rate Golden Decade (2003–13), are expected to of productivity growth—has significantly remain flat in the short and medium term. increased. In fact, this analysis finds that, The region therefore needs to find internal consistent with the findings of Rodrik sources of growth, suggesting that priority (2016), the structural transformation path should be given to a reform agenda focused followed by LAC countries is systematical- on productivity growth. ly different from the path followed in the At the same time, the world is facing past by what are today developed countries. the huge opportunities and challenges Specifically, the region is entering the de­ that arise with the new technologies of the industrialization phase earlier (at lower lev- Fourth Industrial Revolution. Of particu- els of gross domestic product per capita) and lar concern to policy makers and workers achieving lower peaks of industrial shares rel- is the emergence of automation technolo- ative to developed countries. This “premature gies that threaten to destroy a substantial deindustrialization” is worrisome because number of jobs and risk massive unem- in most countries the industrial sector has ployment. Although this report finds that the highest level of labor productivity and 87 88   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n the highest rate of productivity growth. This is not to say that policy makers should When  premature deindustrialization oc- now ignore the industrial sector. Evidence curs, labor moves from the industrial sector from this analysis clearly shows that signif- to the lower productivity growth sectors, icant distortions remain in the sector. This usually services, reducing overall productiv- is reflected in a skewed firm size distribu- ity (so-called Baumol’s disease), with negative tion in which many firms in the LAC region consequences for real income growth and remain small relative to what is observed for standards of living. the United States. Policies that foster inter- From this analysis of the structural trans- national competition within the region and formation process of nine LAC economies globally should receive priority. Also needed with different development levels, three are changes in the size-dependent policies features stand out. The first is the substan- that are disincentivizing the growth of firms tial heterogeneity across the countries in the and incentivizing informality. Policy makers sample. The more developed economies— should encourage adoption of technology, Argentina and Chile—have been deindus- improve the business environment, and pro- trializing for decades. Brazil, Colombia, and vide the telecommunications, transport, and Mexico display stagnant or slight increases logistics infrastructure required for firms to in their industrial employment shares. The grow. Governments should also continue to least developed country in the sample, invest in human capital development, with Bolivia, is still in the industrialization phase a specific focus on the technical and socio- of development. Second, the deindustrializa- emotional skills that will be demanded by the tion process is more pronounced in terms modern industrial sector. of employment shares than in value-added That said, the region is confronting a shares. Similar to the experience of the future in which the services sector will con- United States, this feature is indicative of tinue to grow and be the main source of job the rapid labor productivity growth in this creation. Meanwhile, the region will have sector. Third, premature deindustrialization to remedy lack of understanding about the does not necessarily imply a contraction of complex role of the services sector in pro- the industrial sector; the absolute number ductivity, value added, and job creation. of jobs in that sector—as opposed to the At the aggregate level, the services sector share of jobs—has been fairly stable or even displays lower productivity growth than the growing in most LAC economies. industrial sector. Yet the sector is composed What are the implications of the changes of a very diverse set of subsectors that differ in industrialization for the future? The significantly in their productivity levels and emergence of new technologies suggests growth rates, and even in their use of skilled that opportunities for further industrial- labor. In many countries, some service ization (or reindustrialization) are likely to subsectors—such as telecommunications, be limited in many developing countries. finance, and logistics—are more productive Requirements in terms of complementary and skill-intensive than manufacturing and infrastructure and skills will increase, and are increasingly sharing pro-development global value chains are expected to shorten, characteristics that were once thought to reducing opportunities for entry. To stay be unique to manufacturing. The rapid competitive, firms will need to adopt many advances in information and communica- of these new technologies, which tend to be tions technology (ICT) have enabled the labor-saving. Overall, the industrial sector emergence of services sectors that are no will likely continue to contribute positively longer limited by market size because more to aggregate productivity growth and value and more services can be digitally stored, added, but not as much to job creation, codified, and easily traded (Ghani and especially for unskilled labor. Kharas 2010). Meanwhile, the deregulation C o n c l u s i o n s    89 of services markets has been accompanied play an increasingly critical role as provider by large inflows of FDI. Therefore, certain of inputs to the larger economy. In short, a service subsectors are looking more and comprehensive set of policies oriented to the more like the industrial sector, with expo- services sector is needed. sure to trade and inflows of FDI, allowing Policy makers should give priority to for greater competition, technology diffu- investing in data gathering and analysis of sion, and the benefits of scale. services sector firms in view of the lack of Many of these services are emerging as data for the sector. Understanding the spe- key inputs into industrial and agricultural cific issues of the sector regarding firm size processes, with numerous forward link- distribution, dynamics, barriers to entry, lack ages to other sectors and huge potential of competition, and restrictions to trade is to improve aggregate productivity. In fact, key to formulating policies that can unleash new evidence is pointing to a “servicifica- productivity growth in this sector. tion” of the manufacturing sector—that Fostering competition and streamlin- is, manufacturing is increasing the share ing regulations in the services sector are of services used in the production process important as well. Governments could (embodied services), as well as bundling more incorporate trade in services as part of sales and after-sales services in the sales of regional integration agreements and work goods (embedded services). Reducing distor- toward establishing common licensing tions in the intermediate market for services and certifications so workers and firms could have an important impact on the size can operate throughout the region. With of the industrial sector. Calculations indi- the emergence of digital platforms that cate that the industrial sector could increase allow workers to supply labor from a dis- by 2–3.5 percentage points if distortions in tance and across borders, establishing the services market were reduced to their common regional regulatory frameworks historical minimum. could unleash important productivity gains Meanwhile, it will be increasingly relevant across the region and spur the creation of to formulate value chain policies in addition new entrepreneurial activities and new jobs. to sector-specific policies—that is, policy As for the future of work, three major makers may have a larger impact on aggre- economic forces are changing the nature of gate productivity by understanding how work and the demand for skills. First, struc- sectors interact with each other rather than tural transformation and the premature by studying isolated sectors (the traditional deindustrialization process described in this approach). It is also important to recognize report imply that job creation in the future that the scale-up of key backbone services will be concentrated in the services sector. may be limited not only by sector-specific dis- Second, accompanying the shift in economic tortions that prevent competition and innova- structure is a transformation of the occu- tion from occurring at a rapid pace, but also pational structure within broad economic by the availability of skilled workers because sectors. Service occupations—those that these sectors are highly skill-intensive. produce intangible value added—are ris- ing in all sectors, implying a huge shift in the demand for skills in the labor market. Looking forward Third, because the simpler, more routine Looking forward, the LAC region should tasks will be automated and performed develop a productivity agenda with a spe- by machines, workers will need to adapt cial focus on the services sector. Already and perform a different set of tasks in the the largest employer in the region with over workplace. Consistent with the empirical 60  percent of the workforce, the services evidence from other countries, in the LAC sector is expected to grow even more and region during the 2001–13 time frame of 90   G o i n g V i r a l : C OVID - 1 9 a n d t h e A cc e l e r a t e d T r a n s f o r m a t i o n o f J o b s i n La t i n A m e r i ca a n d t h e C a r i b b e a n this analysis there has been a decline in the support programs that help workers upskill demand for routine manual tasks and a rise and retrain for these new jobs and tasks. in the demand for nonroutine tasks—both The emergence of digital platforms cognitive/analytical (such as critical think- is another possible disruption of labor ing, creativity, and problem-solving) as well markets. On the positive side, digital plat- as interpersonal (such as teamwork, negotia- forms can significantly expand access to tion, managing). new markets, creating opportunities for Based on analysis of the potential number entrepreneurs, which in turn can create new of jobs at high risk of being automated in the jobs. Consumers will gain access to a wider region, it appears that fears of mass techno- variety of products, to better quality prod- logical unemployment are largely unfounded. ucts, and to lower prices through enhanced Estimates vary widely, however, d ­ epending competition. Workers, especially women, on the methodology used. Nevertheless, may find that such platforms provide auton- many occupations will be affected and omy and flexibility they need for their needs transformed by the emerging technologies. and limitations. Although the overall number of jobs many Yet for these benefits to fully materialize, not fall dramatically, many of the tasks several regulatory and infrastructure obsta- being ­performed by humans today will likely cles need to be overcome. Clearly, access to be performed by machines in the future. affordable and reliable broadband service is ­ Workers will interact with more machines a prerequisite for the success of digital plat- and will be expected to understand increas- forms. Logistics infrastructure is also a must ingly complex technologies. Therefore, future to enable efficient and affordable transpor- jobs and tasks will require different and tation of goods within and across countries. higher-­order capabilities and skills. A regulatory framework that establishes clear Both the World Development Report and fair rules on privacy, ownership of data, 2019: The Changing Nature of Work safety, and minimum quality standards is (World Bank 2019) and this analysis also necessary. ­ c onclude that education offers the best Also arising from the findings of this anal- insurance against the risks of automation. ysis is an important concern: the sustain- Low-paid and uneducated workers are ability of the traditional social protection performing the simpler, more routine tasks, models. The growth of employment in the and so they are at highest risk of eventually services sector stemming from structural being replaced by machines. These results transformation and the emergence of new point to a clear conclusion: investing in the technologies that foster alternative working human capital of the workforce should be a arrangements such as independent contrac- priority for policy makers. While investing tors and self-employment have important in early childhood education generates the implications for that model. Looking into highest return on investment (World Bank the future, it appears that less and less labor 2019), there is room to improve in every will be supplied through the traditional dimension of the educational system. employer– employee relationship. For a What may become more important as region that already struggles with very high new automation technologies are adopted labor market informality, these trends pose in LAC countries are adult learning and a serious challenge to the traditional social retraining programs. It is possible that protection model that is financed through transformations in the workplace will employer-employee contributions. happen midcareer for many workers. They Policy makers in the region must think will then need to adapt and adjust, particu- creatively, then, about alternative social larly by changing the set of tasks performed insurance models that do not rely on at work. To minimize the adjustment costs financing and benefits attached to formal borne by workers, governments should employer–employee relationships. In other C o n c l u s i o n s    91 words, policy makers need to define the level References of social protection and insurance that will Ghani, E., and H. Kharas. 2010. “The Service be provided to citizens regardless of their Revolution.” Brief 54595, World Bank, labor status and relationship (employee, Washington, DC. contractor, freelance) and find alternative World Bank. 2019. World Development financing mechanisms that do not depend Report 2019: The Changing Nature of on the employee–employer relationship. Work. Washington, DC: World Bank. Although there are no clear and obvious Rodrik, D. 2016. “Premature Deindustrial- solutions, the region’s policy makers must ization.” Journal of Economic Growth begin to tackle this issue with urgency and 21 (1): 1–33. creativity. ECO-AUDIT Environmental Benefits Statement The World Bank Group is committed to reducing its environmental footprint. In support of this commitment, we leverage electronic publishing options and print-on-demand technology, which is located in regional hubs world- wide. Together, these initiatives enable print runs to be lowered and shipping distances decreased, resulting in reduced paper consumption, chemical use, greenhouse gas emissions, and waste. We follow the recommended standards for paper use set by the Green Press Initiative. The majority of our books are printed on Forest Stewardship Council (FSC)–certified paper, with nearly all containing 50–100 percent recycled content. The recycled fiber in our book paper is either unbleached or bleached using totally chlorine-free (TCF), processed chlorine–free (PCF), or enhanced elemental chlorine–free (EECF) processes. More information about the Bank’s environmental philosophy can be found at http://www.worldbank.org/corporateresponsibility. COVID-19 started as a health emergency, but it is rapidly evolving into an employment crisis. There is still uncertainty on how severe the economic impact of the pandemic will be. As things go, however, the drag on the region’s employment could last longer than the epidemic itself. Beyond the immediate impacts on the level of employment, the crisis is deepening and accel- erating the transformation of jobs, bringing the future closer. Going Viral: COVID-19 and the Accelerated Trans- formation of Jobs in Latin America and the Caribbean focuses on recent trends in the economies of the region that have been significantly changing the labor market: premature deindustrialization, the servicification of the economy, and the changing skill requirements of jobs as automation advances. The findings of this report have important implications for economic policy. Some of these implications are related to the productivity challenges that Latin America and the Caribbean was already facing after the end of the “Golden Decade” in 2013. Other policy implications see their relevance enhanced by the COVID-19 crisis. As sectors are impacted in different ways, as new tech- nologies are developed and adopted, and as working remotely becomes more common, governments need to respond in ways that support a smooth transforma- tion of jobs—one that is socially acceptable and that contributes to productivity growth, including investing in the human capital of the workforce. The accelerated transformation of jobs also calls for a rethinking of labor regulations and social protection policies. The institu- tional architecture geared to wage earners in the formal sector is quickly becoming outdated. The report calls for the flexible regulation of the emerging forms of work, in a way that encourages employment and supports formalization, thereby expanding the coverage of social protection to larger segments of the population. ISBN 978-1-4648-1448-8 90000 9 781464 814488 SKU 211448