Policy Research Working Paper 11214 Growth, Structural Transformation and Carbon Emissions Alen Mulabdic Gaurav Nayyar Katherine Stapleton Prosperity Vertical A verified reproducibility package for this paper is Office of the Chief Economist available at http://reproducibility.worldbank.org, September 2025 click here for direct access. Policy Research Working Paper 11214 Abstract The environmental Kuznets curve postulates an invert- knowledge-intensive services. The diminishing positive ed-U relationship between environmental degradation and association between emissions intensity and structural economic growth. And economic growth has been synon- transformation towards these sectors is more discernible ymous with structural transformation. How do patterns of for developing economies compared with advanced econ- growth and structural transformation relate to carbon emis- omies. Further, based on sector-specific carbon emissions sions? Based on data across almost 100 countries between across 66 countries between 1995 and 2018, we find evi- 1960 and 2017, we find that the movement of workers dence of convergence in the carbon emissions intensity of into the manufacturing and services sectors is associated production across countries in all sectors, with the potential with a higher carbon emissions intensity of GDP. How- for further reductions in developing economies, especially ever, this positive association diminishes at higher shares of given relatively high indirect carbon emissions through employment in both the manufacturing sector and modern, inter-sectoral linkages. This paper is a product of the he Office of the Chief Economist, Prosperity Vertical. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at amulabdic@worldbank.org, gnayyar@worldbank.org, kstapleton@worldbank.org. A verified reproducibility package for this paper is available at http://reproducibility.worldbank.org, click here for direct access. RESEA CY LI R CH PO TRANSPARENT ANALYSIS S W R R E O KI P NG PA The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Growth, Structural Transformation, and Carbon Emissions Alen Mulabdic, Gaurav Nayyar, and Katherine Stapleton * Authorized for distribution Aart C. Kraay, Chief Economist, Prosperity Vertical, World Bank Group Keywords: Structural transformation, economic growth, carbon emissions JEL classification: O14, O44, Q56 * We are grateful to Guido Ardizzone, Yewon Choi, and Mohamad Nassar for excellent research assistance. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. The authors may be contacted at amulabdic@worldbank.org, gnayyar@worldbank.org, and kstapleton@worldbank.org. 1. Introduction Economic growth has been historically associated with structural changes in national economies. The pioneering work of Fisher (1935), Clark (1940), Chenery (1960), and Kuznets (1971) postulated a set of stylized facts from empirical evidence relating to the now-industrialized countries. They found that in the early stages of economic development, the agriculture sector’s share in both output and employment is overwhelmingly large. Subsequently, as industrialization proceeds, the agriculture sector’s share falls, and the manufacturing sector’s share rises. Once countries have industrialized and reached an advanced stage of economic development, the manufacturing sector’s share declines, and the service sector’s share increases. This process of industrialization has been synonymous with economic growth owing to the movement of surplus labor from (subsistence) agriculture to (capitalist) manufacturing and capital accumulation in the latter (Lewis 1954). This is reflected in large and systematic differences in labor productivity between the agriculture and manufacturing sectors across countries (Caselli 2005; Herrendorf, Rogerson, and Valentinyi 2014; Restuccia, Yang, and Zhu 2008). This process of structural transformation has also been associated with the environmental Kuznets curve that depicts an inverted-U relationship between environmental degradation (i.e., pollution) and economic development (i.e., per capita income). In other words, per capita income and pollution intensity go hand in hand at lower levels of per capita income while pollution intensity begins to decline beyond a threshold level of per capita income. The structural change hypothesis conjectures that economies shift from low polluting agriculture to high polluting industry and eventually shift again to low polluting services (Panayotou 1993). Panayotou et al. (2000) summarize: “At low levels of development, both the quantity and the intensity of environmental degradation are limited to the impacts of subsistence economic activity on the resource base and limited quantities of biodegradable wastes. As agriculture and resource extraction intensity increase and industrialization takes off, [...] structural change towards information-based industries and services can result in a decline in environmental degradation”. Empirical analyses of how structural transformation interacts with the carbon intensity of production are few and far between. Some empirical studies on advanced economies find that a higher share of industry in total GDP is associated with higher environmental pressure or energy intensity (Suri and Chapman 1998, Hettige et al. 2000). Other studies find that shifts toward the services sector have contributed to reducing the energy intensity of economic activity (Duro et al. 2010, Mulder et al. 2014). A further set of studies use decomposition analyses to compare the contributions of structural change and technological change in reducing the pollution intensity of economic activity. On one hand, De Bruyn (1997) finds that structural change is much less important than technological change in explaining the reduction of pollution in the Netherlands and West Germany during the 1980s. On the other hand, Weber (2009) finds that the contribution of structural change in explaining the decline in total energy intensity in the United States between 1997 and 2002 exceeded that of increased energy efficiency. Using data between 1995 and 2009, Marsiglio, Ansuategi and Gallastegui (2015) find that structural change was responsible for more than 50 percent of the change in carbon intensity in many European countries. The literature cited above is limited to advanced economies. Moreover, several developing countries have been characterized by unconventional patterns of structural transformation over the 2 past three to four decades. The shares of manufacturing in employment and value added have peaked at lower levels of per capita income than what occurred in the past in the now industrialized countries (Rodrik 2016). Synonymously, much of the declining share of agriculture in employment and value added has been offset by the services sector. The pattern of growth within the services sector itself has also been changing. Eichengreen and Gupta (2013) identify two waves of services sector growth, with the first wave consisting primarily of traditional services as countries move from “low” toward “middle” income status and the second wave comprising modern, knowledge- intensive services (in finance, communication, and business) as countries move from “middle” toward “high” income status. The authors find that the second wave started at lower levels of per capita income after 1990 than in the preceding four decades. This matters for the interaction between structural change and carbon emissions as knowledge-intensive services are perceived to be less polluting. How do these patterns of growth and structural transformation relate to the carbon intensity of production? We aim to fill this gap in the literature by analyzing patterns of growth, structural transformation, and carbon emissions intensity across more than 100 countries between 1960 and 2017, as well as detailed data on emissions intensities for 45 sectors across 66 countries between 1995 and 2018. Overall, we find that the movement of workers from agriculture into manufacturing or services is positively associated with the carbon emissions intensity of GDP, irrespective of the stage of structural transformation that a particular country is in. This result holds even after controlling for GDP per capita and energy consumption. The effect of an increase in the share of manufacturing employment on a country’s overall carbon emissions intensity is higher than that of an increase in the share of services employment across the distribution of employment shares of both sectors. At the same time, for manufacturing-driven structural transformation, the positive association between the manufacturing employment share and carbon emissions intensity diminishes as the manufacturing employment share increases. There is a more constant relationship between services-driven structural transformation and carbon emissions intensity of GDP. The results for services reflect, at least in part, the heterogeneity of activities within the services sector. Much like the manufacturing sector, the positive association between the employment share in modern, knowledge-intensive services and carbon emissions intensity diminishes as the employment share of these services increases. The more muted relationship between structural transformation and emissions intensity for the services sector is driven by traditional services. We also find heterogeneous effects across countries at different levels of per capita income. The decline in the carbon emissions intensity of production associated with an increase in the employment share of manufacturing and modern knowledge-intensive services is more pronounced for low- and middle-income countries than for high-income countries. This is consistent with evidence on the diffusion of less-carbon intensive production technologies from high-income countries to low- and middle-income countries among tradable sectors. Further, we find important links between growth and the carbon emission intensity of production that go beyond structural transformation. First, there is evidence of convergence between developing economies and advanced economies in the carbon emissions intensity of production 3 across the agriculture, manufacturing, and services sectors. Second, there is still further room for developing economies to reduce their levels of carbon emissions intensity across sectors even relative to countries in the same income group. Third, much of these gains from further reducing carbon emissions in developing countries is likely to come from reducing indirect emissions associated with inputs from other sectors. Our paper contributes, first and foremost, to the literature on structural transformation and growth. Rodrik (2012) shows that labor productivity in (formal) manufacturing exhibits “unconditional convergence” across countries. Therefore, labor productivity in lagging manufacturing sectors, such as those in low- and middle-income economies, tends to rise and eventually converge with the global technological frontier regardless of policy and institutional determinants. Similarly, Duarte and Restuccia (2010) find that high productivity growth in the manufacturing sector explains about 50 percent of the catch-up in relative aggregate productivity across countries. More recent evidence finds that services growth can contribute to lower-income countries’ ability to catch up as well (Herrendorf, Rogerson, and Valentinyi 2022; Kinfemichael and Morshed 2019). Baymol and Sen (2020) validate the positive development effects of structural transformation into manufacturing that also reduces income inequality but find that income inequality increases with the movement of workers from agriculture to services. Our paper also contributes to the literature on structural transformation and climate change. Several studies document how climate change-related extreme weather events are moving workers out of the agriculture sector. In India, a 1-degree Celsius increase in the daily average temperature corresponded to a decrease of 7.1 percentage point decline in agricultural employment, and an increase of 2.0 and 3.4 percentage points, respectively, in manufacturing and services employment, respectively between 2001 and 2007 (Colmer 2021). In Brazil, employment in localities with higher incidence of droughts during 2000-2010 shifted more rapidly from agriculture toward manufacturing (Albert et al. 2021). Using a dataset of 700 million scraped online job adverts from 35 high- and middle-income countries, Bastos et al. (2024) find that low-carbon technology jobs are most likely to be in the manufacturing, construction, sales, and high-skilled white-collar services. Evidence from the United States shows that green activities are biased toward cognitive, ICT, management, technical and engineering tasks (Curtis et al. 2023, Vona et al. 2018), which have reallocated economic activity toward college-educated workers (Sato et al. 2023). Last, but not least, our paper contributes to the broader literature on the environmental Kuznets curve. The inverted U-shaped relationship between pollution or emissions and per capita income may be attributable not only to the structural transformation that accompanies economic growth but also to the impact of regulation on pollution abatement and changes in technology (Grossman and Krueger, 1995). The role of regulations in explaining the decline in pollution as countries grow beyond middle-income status is well documented in the literature (Dasgupta et al. 2002). The role of technological change has also been the subject of much analysis. Cohen et al. (2017) find that the environmental Kuznets elasticity, i.e., the relationship between GDP growth and emissions growth, declined from 1.1 in the post-World War II period from 1946-1982 to 0.66 between 1983 and 2007. This is indicative of an improved technical frontier for energy efficiency and the availability of lower-carbon energy sources. Like other technologies, the development of several key low-carbon technologies has occurred in advanced economies. And costs of many such technologies ─ including solar panels, wind turbines, lithium-ion batteries and electrolyzers used 4 for green hydrogen ─ has fallen rapidly and exponentially over the past four decades (Way et al. 2022, Arkolakis and Walsh 2023). Bastos et al. (2024) find that the share of jobs related to low- carbon technologies doubled in 2022 relative to 2021 and around 70 percent of these new job openings were in advanced economies with the highest numbers in the United States. The remainder of the paper is structured as follows. Section 2 presents a conceptual framework to assess the Kuznets process of carbon emissions intensity changing with structural transformation. Section 3 describes the data and provides descriptive statistics around patterns of structural transformation and carbon emissions intensity. Section 4 presents the empirical strategy. Section 5 discusses results. Section 6 comprises extensions to the preceding analysis that assesses the relationship between growth and emissions beyond structural transformation. In doing so, it analyzes the carbon intensity of growth acceleration episodes over time; documents patterns of convergence in the emissions intensity of production across countries in all sectors; estimates potential reductions in emissions intensity with access to better technologies; and distinguishes between the importance of direct and indirect emission intensities across sectors. Section 7 concludes. 2. Conceptual Framework: The Kuznets Environment Curve and Structural Transformation The structural change hypothesis proposes that the process of economic growth is associated with a shift in production from low- to high-polluting economic activities. The Kuznets process of emissions intensity changing with structural transformation can be described as the sum of two processes: (a) the between-sector effect that comprises the movement of workers from less to more emissions-intensive sectors or vice-versa, and (b) the within-sector effect which consists of the changes in emissions intensity within sectors. If both processes reinforce each other, i.e., if the movement of workers is from a sector with lower emissions intensity to a sector with higher emissions intensity and if the decline in emissions intensity of the latter is slower than of the former, then structural transformation will unambiguously increase emissions intensity. However, if the movement of workers is to a sector with higher emissions intensity but also a larger decline in the emissions intensity over time, then overall emissions intensity will not necessarily increase. We provide suggestive evidence on these “between-sector” and “within-sector” effects by distinguishing between the movement of workers from agriculture into either the manufacturing sector or the services sector. This is based on detailed data from the International Monetary Fund on carbon emissions intensities for 45 sectors across 66 countries between 1995 and 2018. Are the effects of manufacturing-driven structural transformation on a country’s overall intensity of carbon emissions likely to be different than that for services-driven structural transformation? We first consider between-sector differences in carbon emissions intensity. For this component to increase with structural transformation, emissions intensity in the sector absorbing labor from agriculture must be higher than the emissions intensity prevailing in agriculture. Pooling observations across countries and over time, it is not always the case that the carbon emissions intensity in the manufacturing sector exceeds that in the agriculture sector for country-year pairs (Figure 1a). Similarly, it is not always the case that carbon emissions intensity in the services sector 5 exceeds that in the agriculture sector for country-year pairs (Figure 1b). However, among the middle-income countries in the sample, the carbon emissions intensity in the manufacturing sector typically exceeds that in agriculture. This implies that structural transformation out of agriculture toward manufacturing is likely to increase the overall carbon intensity of production. Figure 1: Intensity of carbon emissions, by country-sector pair, 1995-2018 (a) Agriculture versus manufacturing (b) Agriculture versus services Source: Authors’ calculations based on IMF Climate Dashboard Note: Kazakhstan is excluded for presentation purposes. The diagonal line represents a 45-degree line. We next consider within-sector differences in carbon emissions intensity. This component of overall emission intensity may decline or increase with structural transformation owing to technological change or the changing composition of industries within broad sectors. There is no clear relationship between employment shares and carbon emissions intensity in the manufacturing sector across countries and over time (Figure 2a). If anything, there is a positive relationship between the two past a threshold share of the manufacturing sector in total employment both in high- and middle-income countries. In contrast, there is a weak negative relationship between employment shares and carbon emissions intensity in the services sector across both high- and middle-income countries (Figure 2b). This suggests that the Kuznets argument, which proposes that the movement of workers from agriculture to non-agriculture will be exacerbated by the within-sector component of emissions intensity, is likely to hold more for the manufacturing sector than the services sector. 6 Figure 2: Intensity of carbon emissions and employment share, by sector, 1995-2018 (a) Manufacturing sector (b) Services sector Source: Authors’ calculations based on IMF Climate Dashboard 3. Data and Descriptive Statistics: Patterns of Structural Transformation and Emissions Intensity We assess the relationship between patterns of structural transformation and the carbon emissions intensity of GDP using a database that covers almost 100 countries between 1961 and 2017. Data on employment by sector is drawn from the World Bank Productivity Database and Groningen Growth and Development Center Database. The corresponding data on carbon emissions intensity of GDP is based on the Our World in Data database. Data on the carbon emissions of GDP is more limited in its country coverage for the earlier decades in the database. This ranges from around 10 countries in the 1960s and 40 countries in the 1970s to 50 countries in the 1980s. Most countries have data on the carbon emissions of GDP starting in 1995. Figure 3: Carbon emissions per GDP, 1995 and 2017 Source: Authors’ calculation based on Our World in Data database There is variation both across countries and over time. The carbon emissions intensity of GDP declined in most countries between 1995 and 2017 (Figure 3). In terms of structural transformation, the share of employment in the services sector increased between 1995 and 2017 across almost all countries and accounts for as much as two-thirds and three-fourths of 7 employment in high-income countries (Figure 4a). The share of employment in the manufacturing sector declined between 1995 and 2017 in several countries – reflecting conventional deindustrialization in today’s high-income countries as well as premature deindustrialization in developing countries. The share of employment in the manufacturing sector increased in a handful of developing countries during this period (Figure 4b). Figure 4: Sectoral shares of employment, 1995 and 2017 a) Services sector b) Manufactruing sector Source: Authors’ calculations based on World Bank Productivity Database and Groningen Growth and Development Center Database We first look at the overall relationship between manufacturing-driven structural transformation and emissions intensity across countries and over time, then by country group. In the overall sample, we see a positive relationship between the manufacturing employment share and carbon emissions intensity. Beyond a threshold level, there is a mild deceleration in the carbon emissions intensity as the employment share of the manufacturing sector increases (Figure 5a). By country group, we see this mild deceleration in the carbon emissions intensity as the manufacturing employment share increases for high-income countries. For low- and middle-income countries, there is a positive relationship between manufacturing-driven structural transformation and carbon emissions intensity (Figure 5b). 8 Figure 5: Overall emissions intensity and the share of manufacturing employment, 1961- 2017 (a) Overall sample (b) By income group Source: Authors’ calculations based on World Bank Productivity Database and Groningen Growth and Development Center Database We next look at the relationship between services employment share and carbon emissions intensity for the overall sample and then by country group. In the overall sample, there is an inverted U-shaped relationship, i.e., beyond a threshold level, there is a decline in the carbon emissions intensity as the employment share of the services sector increases (Figure 6a). By country group, we see a U-shaped relationship for high-income countries, and a positive relationship for middle-income countries (Figure 6b). Figure 6: Overall emissions intensity and the share of services employment, 1961-2017 Source: Authors’ calculations based on World Bank Productivity Database and Groningen Growth and Development Center Database Overall, the scatter plots suggest that there is a clear inverted U-shaped relationship between services-driven structural transformation and carbon emissions intensity, and a largely positive relationship between manufacturing-driven structural transformation and carbon emissions intensity. We now proceed to an econometric analysis of the relationship between structural transformation and carbon emissions intensity. 9 4. Empirical Strategy To assess the relationship between manufacturing-driven structural transformation and services- driven structural transformation, on the one hand, and the carbon emissions intensity, on the other hand, we estimate the marginal impact of an increase in the shares of employment in manufacturing and services on emissions intensity with Equation (1): 2 = 1 + 2 2 + 2 3 + 4 + 5 ℎ + + + + (1) where ‘i’ denotes the country, and ‘t’ denotes the year. CO2intensity is the emissions intensity of GDP; manufacturing, other industry (mining, utilities, and construction) and services are the employment shares of country i in year t in these sectors. Since we are interested in the marginal impact of a sector’s employment share on inequality, we control for the employment shares of the other sectors. X is a vector of other controls, which include per capita GDP and energy consumption, while and are country- and year-fixed effects. We also include squared terms of the manufacturing and services employment shares to allow for a non-linear effect of structural transformation on emissions intensity – as suggested by the Kuznets postulate that emission intensity may first increase, then decrease with structural transformation. Levels of per capita income may have an independent effect on carbon emissions intensity over and above through the effect of structural transformation because the use of low-carbon technologies is likely to be more widespread in higher-income countries. Further, larger countries with higher levels of energy consumption are likely to see higher carbon emissions intensity. To distinguish between different types of services while assessing the relationship between services-driven structural transformation and carbon emissions intensity, we disaggregate the employment share for services into modern (finance, ICT, and professional services) and traditional services (commerce, hospitality, transportation, government, and personal services) in Equation (2): 2 = 1 + 2 2 + 3 + 4 2 + 5 + +6 2 + 7 ℎ + + + + (2) To allow for the effect of manufacturing and services employment shares on carbon emissions intensity to differ by a country’s stage of structural transformation, we interact the manufacturing and services employment shares with a dummy variable that equals 1 whether the country is high- income and zero otherwise (the residual is if the country is low- or middle-income) in Equation (3). 2 = 1 + 2 2 + 2 3 + 4 + 5 ∗ + 6 2 * + 7 ∗ 10 + 8 2 ∗ + 9 ℎ + + + + (3) In Equation (4), we allow for the effect of manufacturing and services employment shares on carbon emissions intensity to vary by country income group, while distinguishing between traditional and modern services. 2 = 1 + 2 2 + 2 3 + 4 + 5 + +6 2 + 7 ∗ + 8 2 * + 9 ∗ + 10 2 ∗ + 11 ∗ + 12 2 ∗ + 13 ℎ + + + + (4) 5. Results We present the results of the set of panel regressions that aim to investigate the relationship between the manufacturing and services employment shares and carbon emissions intensity in Table 1. Columns (1) and (2) present the estimates of Equation (1) with and without the basic control variables. Columns (3) and (4) present the estimates of Equation (2) with country income groups interacted with manufacturing and services employment shares. a. Baseline specification In the baseline specification, the fixed effect estimates in column (1) of Table 1 suggest that an increase in the manufacturing employment share is associated with higher carbon emissions intensity – the coefficient on the manufacturing employment share is positive and significant at 1 percent level of significance. Furthermore, the negative and statistically significant coefficient on the squared manufacturing employment share variable in column (1) suggests that there is an inverted U-shape relationship between manufacturing employment and carbon emissions intensity. The results are qualitatively similar for the services sector. The coefficients on the share of employment in services are positive and statistically significant at the 1 percent level. There is also a clear inverted U-shaped relationship between service employment share and carbon emissions intensity – the quadratic term on services is negative and statistically significant. These results in column (3) of Table 1 remain largely unchanged after controlling for income per capita and energy consumption. The coefficient on GDP per capita is negative and significant, reflecting a lower average carbon intensity for richer countries, while that on energy consumption is positive and significant, reflecting a higher carbon intensity of more energy intensive countries. In nonlinear models, the relationship between the dependent and explanatory variables varies across the range of the relevant explanatory variable. We therefore present the marginal effect of the change in manufacturing and services employment shares on carbon emissions intensity at different levels of these shares. We find that the movement of workers into the manufacturing and 11 services sectors is associated with an increase in the carbon emissions intensity of GDP, irrespective of the stage of structural transformation that a particular country is in (Figures 7a and 7b). However, the marginal effect of structural transformation on the carbon intensity of GDP is lower at larger shares of both manufacturing and services employment. There are two important differences between structural transformation toward the manufacturing and services sectors. First, the marginal effect of an increase in the share of manufacturing employment on a country’s overall carbon emissions intensity is higher than that of an increase in the share of services employment across the entire distribution of employment shares for the two sectors. Second, the decline in the marginal effect on a country’s overall carbon emissions intensity is much more pronounced with an increase in the share of manufacturing employment compared with an increase in the share of services employment. Table 1: Baseline regressions (1) (2) (3) (4) CO2 intensity Manufacturing 2.012*** 1.774*** 2.666*** 1.841*** (0.384) (0.380) (0.391) (0.418) Manufacturing^2 -1.556* -1.877** -2.654*** -2.120** (0.905) (0.857) (0.885) (0.863) Services 2.547*** 2.287*** (0.256) (0.265) Services^2 -1.289*** -0.853*** (0.172) (0.208) Modern Services 0.077 -0.129 (0.473) (0.484) Modern Services^2 -8.818*** -7.352*** (2.042) (1.919) Traditional Services 1.809*** 1.779*** (0.280) (0.284) Traditional Services^2 -0.120 -0.123 (0.245) (0.265) Other industry -2.358*** -2.335*** -1.794*** -2.193*** (0.493) (0.489) (0.540) (0.529) Log of GDP per capita -0.164*** -0.099*** (0.016) (0.016) Log of energy consumption 0.099*** 0.103*** (0.017) (0.017) Observations 3,793 3,793 3,717 3,717 R-squared 0.793 0.807 0.790 0.803 Country FE YES YES YES YES Year FE YES YES YES YES Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 12 Figure 7: Marginal effects of sectoral employment shares on the overall emissions intensity (a) Manufacturing sector: Column 3, Table 1 (b) Services sector: Column 3, Table 1 (c) Modern services: Column 4, Table 1 (d) Traditional services: Column 4, Table 1 Further, we find differences between modern (finance, business, and professional services) and traditional (commerce, hospitality, transportation and social) services. In column (4) of Table 1, where we control for income per capita and energy consumption, the employment share in both modern services and traditional services is positively associated with emissions intensity of GDP. However, beyond a threshold share of employment, there is a negative association between the employment share in modern services and emissions intensity of GDP. There is no evidence of a diminishing relationship for traditional services. In terms of the marginal effects derived from these coefficients, the marginal effect of an increase in the share of modern services employment on a country’s overall carbon emissions intensity declines much like the manufacturing sector (Figure 7c). However, there are no such diminishing effects with an increase in the share of traditional services employment (Figure 7d). These results are consistent with the fact that modern services are digitally deliverable and therefore less subject to the carbon intensity associated with transportation. b. Heterogeneous effects We find evidence of heterogeneity in the relationship between manufacturing and services employment shares and the carbon emissions intensity of GDP by country income group. The marginal effects discussed below are derived from the coefficients presented in columns (3) and (4) of Table 2 where GDP per capita and energy consumption are included as control variables. 13 For high-income countries, the marginal effect of the manufacturing employment share on emissions intensity is positive and significant and remains largely constant irrespective of the level of manufacturing employment share. For low- and middle-income countries, the marginal effect of the manufacturing employment share on emissions intensity is positive and statistically significant and larger. However, this marginal effect declines considerably with an increase in the employment share of the manufacturing sector (Figure 8a). This suggests that manufacturing- driven structural transformation has a smaller effect on emissions as the manufacturing employment share expands in low- and middle-income countries. The difference in the marginal effect of the services sector’s employment share on emissions intensity between high-income countries and low- and middle-income countries is not statistically significant (Figure 8b). The less conclusive results for the services sector are driven, at least in part, by differences across subsectors. For modern services, much like the manufacturing sector, the marginal effect of an increase in the sector’s employment share is positive and statistically significant but declines considerably with the sector’s expansion in low- and middle-income countries relative to high- income countries (Figure 8c). The difference between high-income and low- and middle-income countries in the relationship between the increasing employment share of traditional services and the carbon intensity of GDP is not statistically significant (Figure 8d). 14 Table 2: Income groups (1) (2) (3) (4) CO2 intensity Manufacturing 3.577*** 3.295*** 4.077*** 3.553*** (0.453) (0.418) (0.454) (0.439) Manufacturing^2 -5.087*** -5.267*** -5.901*** -5.697*** (0.966) (0.860) (0.949) (0.866) Services 2.459*** 2.410*** (0.262) (0.270) Services^2 -1.212*** -1.069*** (0.183) (0.218) Modern Services 4.771*** 5.034*** (0.859) (0.805) Modern Services^2 -32.277*** -32.943*** (5.445) (5.173) Traditional Services 2.016*** 2.022*** (0.273) (0.277) Traditional Services^2 -1.012*** -1.014*** (0.260) (0.275) Other industry -2.603*** -2.331*** -2.080*** -1.997*** (0.490) (0.451) (0.528) (0.470) I(High income) x … … x (Manufacturing) -5.175*** -4.156*** -4.816*** -3.970*** (1.763) (1.479) (1.753) (1.466) … x (Manufacturing^2) 10.005*** 8.093*** 9.169*** 7.832*** (3.243) (2.744) (3.211) (2.668) … x (Services) 1.609 1.331 (1.048) (1.044) … x (Services^2) -1.288 -0.907 (1.093) (1.089) … x (Modern Services) -5.727*** -6.629*** (1.900) (1.962) … x (Modern Services^2) 24.411*** 28.837*** (8.144) (8.100) … x (Traditional Services) 1.095 1.052 (1.068) (1.075) … x (Traditional Services^2) 0.550 0.748 (1.314) (1.324) Log of GDP per capita -0.151*** -0.124*** (0.015) (0.015) Log of energy consumption 0.097*** 0.097*** (0.017) (0.015) Observations 3,793 3,793 3,717 3,717 R-squared 0.804 0.825 0.802 0.822 Country FE YES YES YES YES Year FE YES YES YES YES Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Figure 8: Marginal effects of sectoral employment shares on the overall emissions intensity, high-income countries versus developing economies 15 (a) Manufacturing sector: Column 3, Table 2 (b) Services sector: Column 3, Table 2 (c) Modern services: Column 4, Table 2 (d) Traditional services: Column 4, Table 2 c. Robustness checks We re-estimate Equation (1) by excluding country-year observations that are outliers in terms of the employment share of the manufacturing and services sectors. Specifically, in column (1) we exclude top and bottom 1 percent of observations based on the manufacturing employment share, while in column (2) we exclude the top and bottom 1 percent based on the services employment share. The sign and statistical significance of the coefficients on the manufacturing employment share in Table 1. We also re-estimate Equation (1) by using sectoral shares in value added rather the employment. The results remain qualitatively similar (column 3, Table 3). 16 Table 3: Robustness checks (1) (2) (3) Excl. Top/Bottom 1% Excl. Top/Bottom 1% Manufacturing Services VA Manufacturing 2.391*** 2.033*** 2.882*** (0.425) (0.395) (0.393) Manufacturing^2 -1.727* -1.722** -2.855*** (0.994) (0.759) (0.630) Services 2.375*** 1.893*** 0.267 (0.269) (0.253) (0.289) Services^2 -0.953*** -0.807*** 0.291 (0.216) (0.242) (0.340) Other industry -2.169*** 1.060*** 1.096*** (0.551) (0.263) (0.081) Log of GDP per capita -0.163*** -0.320*** -0.244*** (0.016) (0.023) (0.021) Log of energy consumption 0.097*** 0.126*** 0.136*** (0.017) (0.026) (0.016) Observations 3,638 2,073 3,717 R-squared 0.793 0.906 0.761 Country FE YES YES YES Year FE YES YES YES Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 In terms of the heterogeneous effects across high-income and low- and middle-income countries, we re-estimate Equations (3) and (4) using country income groups as classified in 1987 (which is the first year of the World Bank country income group classification). The results presented in Table 4 are qualitatively similar to the results presented in Table 2. 17 Table 4: Income groups (historical income group classification) (1) (2) (3) (4) CO2 intensity Manufacturing 3.577*** 3.295*** 4.077*** 3.553*** (0.453) (0.418) (0.454) (0.439) Manufacturing^2 -5.087*** -5.267*** -5.901*** -5.697*** (0.966) (0.860) (0.949) (0.866) Services 2.459*** 2.410*** (0.262) (0.270) Services^2 -1.212*** -1.069*** (0.183) (0.218) Modern Services 4.771*** 5.034*** (0.859) (0.805) Modern Services^2 -32.277*** -32.943*** (5.445) (5.173) Traditional Services 2.016*** 2.022*** (0.273) (0.277) Traditional Services^2 -1.012*** -1.014*** (0.260) (0.275) Other industry -2.603*** -2.331*** -2.080*** -1.997*** (0.490) (0.451) (0.528) (0.470) I(High income) x … … x (Manufacturing) -5.175*** -4.156*** -4.816*** -3.970*** (1.763) (1.479) (1.753) (1.466) … x (Manufacturing^2) 10.005*** 8.093*** 9.169*** 7.832*** (3.243) (2.744) (3.211) (2.668) … x (Services) 1.609 1.331 (1.048) (1.044) … x (Services^2) -1.288 -0.907 (1.093) (1.089) … x (Modern Services) -5.727*** -6.629*** (1.900) (1.962) … x (Modern Services^2) 24.411*** 28.837*** (8.144) (8.100) … x (Traditional Services) 1.095 1.052 (1.068) (1.075) … x (Traditional Services^2) 0.550 0.748 (1.314) (1.324) Log of GDP per capita -0.151*** -0.124*** (0.015) (0.015) Log of energy consumption 0.097*** 0.097*** (0.017) (0.015) Observations 3,793 3,793 3,717 3,717 R-squared 0.804 0.825 0.802 0.822 Country FE YES YES YES YES Year FE YES YES YES YES Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 18 6. Extension: Carbon Intensities within Sectors Over time, the relationship between growth, structural transformation and the overall carbon emissions intensity depends on changing emission intensities within sectors. First, we find countries with higher emissions intensities experienced faster declines than countries with lower initial emissions intensities between 1995 and 2018. This pattern of convergence between developing economies and advanced economies is evident across sectors. Second, we find room for greater convergence in carbon emissions intensities across sectors among developing countries. Third, we find that indirect emissions intensities – that account for input-output linkages between sectors – in developing countries are higher than advanced economies across sectors. a. Convergence in emissions intensity across countries To understand how carbon emission intensities across countries evolve over time, we empirically test for the presence of convergence in emissions intensities at the industry level using the IMF’s database on CO2 emissions, emissions intensities, and emissions multipliers for 45 sectors across 66 countries between 1995 and 2018. Following the economic growth literature (e.g., Rodrik 2013), we estimate the following equation: ∆ ln( ) = ln(−1 ) + + + where we regress the growth in direct emissions, defined as CO2 emissions per US$1 million of output, on the lagged value of direct emissions, a set of sector-year fixed effects ( ), and country fixed effects ( ). The coefficient of interest is , which captures speed of convergence in emissions. A negative and significant coefficient would be indicative of convergence, i.e., countries with higher emission intensities experienced faster declines in emission intensities than countries with lower levels of emissions. The inclusion of country fixed effects allows us to test for conditional convergence, as all time-invariant country characteristics are controlled for, allowing us to identify the convergence coefficient by exploiting within-country variation in emissions across different sectors. The exclusion of country fixed effects allows us to test for unconditional convergence in direct emissions. This equation is estimated for all broad sectors. The results in Table 5a indicate that there is unconditional convergence in emissions intensity across sectors. The estimated convergence coefficients vary between 3.7% and 3.8%, but they are not statistically different across sectors. Given low- and middle-income countries have higher emission intensities, these results suggest that these countries are converging toward the emission intensities of high-income countries across sectors. To understand if this convergence is driven by reductions in emissions in dirtier sectors, we estimate the coefficient for conditional convergence. The results in Table 5b show that all four sectors experienced conditional convergence. Modern services and agriculture have the largest convergence coefficients, indicating that the reduction in emissions intensities in these sectors is driven by faster decreases in emissions in dirtier industries. Industry and manufacturing have relatively smaller convergence coefficients, suggesting that emissions intensities in dirtier industries have not decreased as much as emissions intensities in cleaner industries. 19 Table 5: Convergence in the intensity of direct carbon emission across countries Panel A: Unconditional convergence (1) (2) (3) (4) (5) (6) Direct emissions intensity Manufacturin Modern Traditional Agriculture Industry Services g Services Services Log of intensity t-1 -0.037*** -0.037*** -0.038*** -0.039*** -0.039*** -0.038*** (0.007) (0.002) (0.002) (0.002) (0.004) (0.002) Observations 2,828 32,900 24,438 28,402 10,447 17,955 R-squared 0.083 0.098 0.097 0.110 0.105 0.112 Sector-Time FE YES YES YES YES YES YES Country FE NO NO NO NO NO NO Panel B: Conditional convergence Manufacturin Modern Traditional Agriculture Industry Services g Services Services Log of indirect intensity t-1 -0.068*** -0.045*** -0.047*** -0.074*** -0.101*** -0.067*** (0.009) (0.002) (0.003) (0.003) (0.008) (0.003) Observations 2,828 32,900 24,438 28,402 10,447 17,955 R-squared 0.116 0.109 0.110 0.140 0.154 0.137 Sector-Time FE YES YES YES YES YES YES Country FE YES YES YES YES YES YES Note: Robust standard errors, clustered at the country-sector level, are in parentheses. Agriculture includes sectors ISIC 01-03; industry includes sectors ISIC 05-43; manufacturing includes sectors ISIC 10-33; services includes sectors ISIC 49-98; modern services includes sectors ISIC 58-82; and traditional services are non- modern services. *** p<0.01, ** p<0.05, * p<0.1 b. Counterfactual scenarios While low- and middle-income countries are converging toward the emissions intensities of high- income countries, they have the potential to reduce their total emissions by adopting cleaner technologies that are already in use in countries at similar levels of development. Figure 9 illustrates this by presenting three scenarios in which countries adopt cleaner technologies in the manufacturing, services, and electricity and gas sectors. Cleaner technologies are calculated as the 25th percentile in terms of CO2 emissions per US$1 million of output for each sector within the same income group. For instance, if China adopted cleaner manufacturing technologies currently used in other middle-income countries, it could reduce its emissions to 75 percent of current levels. Adopting cleaner technologies in the service sector would lower emissions by an additional 3 percentage points. However, the largest reductions would come from adopting cleaner technologies in the electricity and gas sector, which alone would reduce total emissions by more than manufacturing and services industries combined. For countries such as South Africa, Kazakhstan, the Russian Federation, and Malaysia, adopting cleaner technologies in the electricity and gas sector could reduce emissions by 30-40 percent. On the other hand, countries such as India, Indonesia, and Brazil may achieve limited emissions reductions from adopting cleaner technologies in the electricity and gas sector, as they already use relatively clean production technologies. In these countries, the potential gains are larger from reducing emissions in the manufacturing sector. Among all low- and middle-income countries in 20 the sample, there are limited expected gains from adopting cleaner technologies in the services sector among countries in the same income group. Figure 9: Counterfactual scenarios of emissions intensities with cleaner technologies Authors’ calculations based on IMF Climate Dashboard c. Direct and indirect emission intensities The source of carbon emissions in each sector can be either direct or indirect where the latter draws on inputs used from other sectors. Figure 10 shows the average direct and indirect carbon emission intensities for manufacturing, modern services, and traditional services in developing countries and developed countries. It shows that emission intensities are higher in developing countries, with the difference between high-income countries and low- and middle-income countries being more pronounced for indirect emissions, especially in the manufacturing sector. These differences are statistically significant across sectors for both direct and indirect emission intensities except for direct emissions in manufacturing (Table 6). This suggests that the diffusion of low carbon- intensive technologies across all sectors of the economy is fundamentally important in low- and middle-income countries; emissions intensities across sectors are magnified through upstream inputs used from other sectors. 21 Figure 10: Direct and indirect carbon emissions intensities, advanced versus developing economies Authors’ calculations based on IMF Climate Dashboard Table 6: Intensity of carbon emission across country groups (1) (2) (3) (4) (5) (6) Direct intensities Indirect intensities Manufacturing Modern Traditional Manufacturing Modern Traditional services services services services Developing country 81.845 21.588*** 105.760** 214.225*** 106.969*** 111.130*** (61.199) (2.378) (43.150) (20.867) (14.879) (8.100) Constant 173.328*** 14.740*** 229.333* 123.702*** 58.464*** 87.012*** (53.655) (2.529) (126.341) (8.992) (5.541) (7.161) Observations 1,076 462 790 1,076 462 790 R-squared 0.003 0.127 0.007 0.184 0.162 0.153 Note: The dependent variable is the level of direct or indirect emissions at the sector level. Each column reports the results of a regression that includes a constant and a dummy variable for developing countries, pooling observations across all sectors within each industry. The constant captures the average emission level in developed countries, while the developing country indicator captures the deviation in emissions for developing countries relative developed ones. Robust standard errors, clustered at the sector level, are in parentheses. *** p<0.01, ** p<0.05, * p<0.1 7. Conclusion The process of structural transformation away from agriculture to the manufacturing and services sectors is synonymous with economic growth. This process of structural change is associated with the environmental Kuznets curve where emissions first increase and decline with levels of per capita income. Consistent with the Kuznets hypothesis, we find – based on data across more than 100 countries between 1960 and 2017 – that the movement of workers into manufacturing and modern, knowledge-intensive services is associated with a higher carbon emissions intensity of GDP but this positive association diminishes at higher shares of employment in both sectors. We do not find that the positive association between the employment share in traditional services and the carbon intensity of GDP diminishes at higher employment shares for the sector. Further, the diminishing positive association between structural transformation and emissions intensity is more 22 discernible for developing countries in manufacturing and knowledge-intensive services. These results are indicative of the diffusion of low-emission technologies from high- to low- and middle- income countries in tradable sectors. The relationship between growth and the carbon emissions intensity of GDP also goes beyond structural transformation. First, we find that the carbon intensity of growth acceleration episodes has declined over time after accounting for countries’ levels of per capita income and changing sectoral shares. Second, we find (unconditional and conditional) convergence in the carbon emissions intensity of production across countries across all sectors. Third, low- and middle- income countries can further reduce their carbon emissions intensity by accessing better technologies prevalent among countries even in their country income group. Fourth, low- and middle-income countries particularly lag high-income countries in their intensity of indirect emissions, which draws on input-output linkages across sectors. The findings of this paper show that the structural change and technological change explanations around the environmental Kuznets curve go together. Further, supporting the use of low-carbon technologies across sectors in developing countries is likely to bring larger benefits as the emissions intensity of each sector is compounded through input-output linkages with other sectors. The findings of the paper also provide insights to an emerging literature on structural transformation that distinguishes between manufacturing-led and services-led growth. In particular, it shows that distinguishing between traditional and modern services is not only important for the analysis of productivity, international trade, jobs, and skills but also the intensity of carbon emissions. Overall, these findings have implications for prospective development paths for low- and middle-income countries given that 91 percent of global GDP is now covered by a net-zero pledge by 2050-2060. 23 References Albert, Christoph, Paula Bustos, and Jacopo Ponticelli. 2021. “The Effects of Climate Change on Labor and Capital Reallocation.” NBER Working Paper 28995, National Bureau of Economic Research, Cambridge, MA. Arkolakis, Costas and Walsh, Conor. 2023. “Clean Growth”. NBER Working Paper No. w31615. Bastos, Paulo, Jacob N. Greenspon, Katherine Stapleton, and Daria Taglioni. 2024. "Did the 2022 global energy crisis accelerate the diffusion of low-carbon technologies?," Policy Research Working Paper Series 10777, The World Bank. Baymul, C. and K. Sen. 2020. “Was Kuznets Right? New Evidence on the Relationship between Structural Transformation and Inequality”. The Journal of Development Studies, 56(9), 1643– 1662. Caselli, Francesco. 2005. “Accounting for Cross-Country Income Differences.” In Handbook of Economic Growth, Vol. 1, edited by Philippe Aghion and Steven Durlauf, 679–741. Amsterdam: Elsevier. Chenery, Hollis B. 1960. “Patterns of Industrial Growth.” American Economic Review 50 (4): 624–54. Clark, Colin. 1940. The Conditions of Economic Progress. London: McMillan. Cohen, G., J. Jalles, P. Loungani, and R. Marto. 2017. “Emissions and Growth: Trends and Cycles in a Globalized World," IMF Working paper Colmer, Jonathan. 2021. “Rainfall Variability, Child Labor, and Human Capital Accumulation in Rural Ethiopia.” American Journal of Agricultural Economics 103 (3): 858–77. Curtis, E. Mark, Layla O’Kane, and R. Jisung Park. 2024. “Workers and the Green-Energy Transition: Evidence from 300 Million Job Transitions.” Environmental and Energy Policy and the Economy 5 (January): 127–61. Dasgupta, Susmita, Benoit Laplante, Hua Wang, and David Wheeler. 2002. "Confronting the Environmental Kuznets Curve." Journal of Economic Perspectives, 16 (1): 147–168. De Bruyn, S.M. 1997. “Explaining the environmental Kuznets curve: structural change and international agreements in reducing sulphur emissions”. Environmental Development Economics 2:485–503. Duarte, Margarida, and Diego Restuccia. 2010. “The Role of the Structural Transformation in Aggregate Productivity.” Quarterly Journal of Economics 125 (1): 129–73. Duro, J. A., V. Alcántara, and E. Padilla. 2010. “International inequality in energy intensity levels and the role of production composition and energy efficiency: an analysis of OECD countries”. Ecological Economics 69:2468–2474. 24 Eichengreen, Barry and Poonam Gupta. 2013. “The two waves of service-sector growth”, Oxford Economic Papers, 65(1): 96–123. Fisher, A. G. B. 1935. “The Significance of Stable Prices in a Progressive Economy.” Economic Record 10 (2): 49–64. Grossman, Gene M. and Alan B. Krueger. 1995. “Economic Growth and the Environment”, The Quarterly Journal of Economics 110(2): 353–377. Herrendorf, Berthold, Richard Rogerson, and Akos Valentinyi. 2014. “Growth and Structural Transformation.” In Handbook of Economic Growth, edited by Philippe Aghion and Steven N. Durlauf, 2:855–941. Handbook of Economic Growth. Elsevier. Herrendorf, Berthold, Richard Rogerson, and Ákos Valentinyi. 2022. “New Evidence on Sectoral Labor Productivity: Implications for Industrialization and Development.” NBER Working Paper 29834, National Bureau of Economic Research, Cambridge, MA. Hettige H., M. Mani, and D. Wheeler. 2000. “Industrial pollution in economic development: the environmental Kuznets curve revisited”. Journal of Development Economics 62:445–476. Kar, Sabyasachi, Lant Pritchett, Selim Raihan, and Kunal Sen, 2013. "Looking for a Break: Identifying transitions in growth regimes," Journal of Macroeconomics 38(B): 151-166. Kinfemichael, Bisrat, and A. K. M. Mahbub Morshed. 2019. “Unconditional Convergence of Labor Productivity in the Service Sector.” Journal of Macroeconomics 59 (March): 217–29. Kuznets, Simon Smith. 1971. Economic Growth of Nations: Total Output and Production Structure. Cambridge, MA: Belknap Press of Harvard University Press. Lewis, W. Arthur. 1954. “Economic Development with Unlimited Supplies of Labour.” The Manchester School 22 (2): 139–91. Marsiglio, S., A. Ansuategi, and M.C. Gallastegui. 2016. “The Environmental Kuznets Curve and the Structural Change Hypothesis”. Environmental Resource Economics 63, 265–288. McMillan, Margaret S., and Dani Rodrik. 2011. “Globalization, Structural Change and Productivity Growth.” NBER Working Paper 17143, National Bureau of Economic Research, Cambridge, MA. Mulder P., H. de Groot, and B. Pfeiffer. 2014. “Dynamics and determinants of energy intensity in the service sector: a cross-country analysis 1980–2005”. Ecological Economics 100:1–15. Panayotou, T. 1993. “Empirical tests and policy analysis of environmental degradation at different stages of economic development”. Working Paper WP238, Technology and Employment Programme, International Labor Office, Geneva. Panayotou, T., A. Peterson, and J. Sachs. 2000. “Is the environmental Kuznets curve driven by structural change? What extended time series data may imply for developing countries”. Center for International Development, CAER II Discussion Paper 80. 25 Pritchett, Lant, Kunal Sen, Sabyasachi Kar, and Selim Raihan. 2016. "Trillions gained and lost. Estimating the magnitude of growth episodes," Economic Modelling 55: 279–291. Restuccia, D., D. T. Yang, and X. Zhu. 2008. “Agriculture and Aggregate Productivity: A Quantitative Cross-Country Analysis.” Journal of Monetary Economics 55 (2): 234–50. Rodrik, Dani. 2016. “Premature Deindustrialization.” Journal of Economic Growth 21 (1): 1–33. Rodrik, Dani. 2012. “Unconditional Convergence in Manufacturing.” Quarterly Journal of Economics 128 (1): 165–204. Sato, Misato, Leanne Cass, Aurélien Saussay, Francesco Vona, Leo Mercer, and Layla O’Kane. 2023. “Skills and Wage Gaps in the Low-Carbon Transition:” Policy Report. London: Grantham Research Institute on Climate Change and the Environment and Centre for Climate Change Economics and Policy, London School of Economics and Political Science. Suri V. and D. Chapman. 1998. “Economic growth, trade, and energy: implications for the environmental Kuznets curve”. Ecological Economics 25:195–208. Vona, Francesco, Giovanni Marin, Davide Consoli, and David Popp. 2018. “Environmental Regulation and Green Skills: An Empirical Exploration.” Journal of the Association of Environmental and Resource Economists 5 (4): 713–53. Way, R., M.C. Ives, P. Mealy, and J.D. Farmer. 2022. 'Empirically grounded technology forecasts and the energy transition'. Joule 6, 1–26. Weber, C.L. 2009. “Measuring structural change and energy use: decomposition of the US economy from 1997 to 2002”. Energy Policy 37:1561–1570. 26