Putting the green back in greenbacks: opportunities for a truly green stimulus

Can countries reorient their productive capacity to become more environmentally friendly and inclusive? To investigate this question this paper uses a standard Input-Output modeling framework and data from 141 countries and regions to construct a new global dataset of employment, value-added, greenhouse gas (GHG) emissions (disaggregated into CO2 and Non-CO2 elements), and air pollution (including nine categories of air pollutants such as PM2.5) multipliers from supply side investments. We find that many of the traditional sectors in agriculture and industry have large employment multipliers, but also generate male dominant, lower skill employment, and tend to have higher emission multipliers. It is in economies dominated by these sectors that trade-offs to a ‘greener’ transition will emerge most sharply. However, we find a substantial heterogeneity in outcomes, so even in these economies, there exist other sectors with high employment multipliers and low emissions, including sectors that are more conducive to female employment. In addition, we find a high correlation between industries that generate GHG emissions, which cause long term climate impacts, and those that generate air pollution, which have immediate harmful impacts on human health, suggesting that policies could be designed to simultaneously confer longer climate benefits with immediate health improvements. Our results confirm some of the findings from recent research and shed new light on opportunities for greening economies.


Introduction
The aftermath of the COVID induced recession has rekindled interest in a recovery that can be inclusive and deliver growth without significant climate and environmental degradation. Central to these discussions are aspirations to create jobs in sectors that are relatively 'clean' and 'inclusive' . Proponents of a lowcarbon green economy tout the many benefits and argue that such opportunities abound (UNEP 2015). Critics point to the job losses that may ensue especially in the more emission intensive sectors of the economy (e.g. Greenstone 2002, Kahn andMansur 2013). While the pandemic has unleashed an outpouring of fiscal spending, estimated at US$14 trillion so far at global scale, much has been concentrated on measures to address the health emergency and support household and firms stranded by the lockdowns (IMF 2021). As a result, much spending has been on supporting the 'legacy' economy, including sectors (such as coal) that are environmentally harmful (Vivid Economics 2020). This may reflect, and be a rational response to, the uncertainty about the economic implications of investing in greener and lower emissions sectors. Such reticence is perhaps unsurprising during an economic crisis and public health emergency, especially when there is a paucity of information and research on the effectiveness of green stimuli.
This paper seeks to fill this gap in the literature by generating new information that could assist in the design of better informed and greener policymaking. This paper makes three key contributions. First, we construct a new global dataset of employment, valueadded, greenhouse gas (GHG) emissions, and air pollution (including nine categories of air pollutants such as PM 2.5 ) multipliers. Results are at the global scale, and are also disaggregated into 141 countries and regions 3 , using the Global Trade Analysis Project (GTAP) database in a standard Input-Output model.
There are well-known methodological and data challenges in assessing the macroeconomic and environmental effects of green stimulus measures. Ex post evaluations face the difficult task of defining a suitable counterfactual and addressing biases that derive from violations of numerous exclusion restrictions. For instance, since a fiscal stimulus is typically introduced during a recession, spending will be endogenous to the business cycle. Likewise, if 'greener' spending is allocated in a non-random manner and directed towards certain sectors or regions with higher (or lower) transformational potential, this too would bias the estimates. As a result of these and other challenges, previous work has tended to rely on simulated impacts and ex ante evaluations. The contribution of this paper is in this latter category. We provide baseline estimates that can be used to benchmark performance against more detailed country level estimates where these become available.
Second, the global dataset enables a ranking of industries across countries, and within countries, in terms of employment, value-added, GHG emissions, and other air pollutants (e.g. PM 2.5 ). It provides a way of determining whether a country can reorient its productive capacity to cleaner sectors while creating employment that is gender neutral and does not have adverse distributional impacts. Additionally, investments that stimulate rapid increases in value-added (growth) need not coincide with those that generate the highest levels of employment. Our results identify where such tradeoffs may occur and their prevalence across countries.
Third, the study provides estimates of local air pollution as measured through PM 2.5 , which is known to be responsible for the premature loss of around 4 million lives each year-a number greater than deaths due to war and violence, and more than has been lost in the COVID-19 pandemic. Effective design of stimulus packages could provide an opportunity for governments to reduce harmful and even fatal levels of exposure to air pollution, while creating new jobs and stimulating the economy.
Finally, the results highlight the importance of distinguishing between absolute emissions and emission intensities. The focus of much climate change policy is quite reasonably on sectors with high absolute GHG emissions. The absolute emissions in a sector may be high, either due to the size of that sector (i.e. the scale effect), or because of high emissions generated per unit of value added produced in the sector (i.e. the emissions intensity of the sector). If the aim is to green growth, then 3 Regions represent aggregations of small countries. emission intensities provide valuable information as these capture changes due to cleaner production processes (holding value added constant), as well as changes due to increases in value added (holding emissions constant). Both changes are desirable to decouple economic growth from environmental damage. Neglecting the distinction between scale effects and intensity effects could lead to suboptimal policy choices, if it leads to expanding sectors that generate fewer benefits for the same level of emissions. This paper is related to a growing literature on the greening of the economy. In one of the few empirical estimates available Popp et al (2020) find that in the US the green stimulus termed the American Recovery and Reinvestment Act, increased total employment slower than other stimulus investments. Focusing on habitat restoration projects Edwards et al (2013) find that restoration projects are labor intensive and created, on average, 17 jobs per million dollars spent. In perhaps the most advanced empirical assessment Batini et al (2021) show that multipliers associated with green spending are about 2-7 times larger than those associated with non-eco-friendly expenditure.
Complementing these empirical assessments is a growing simulation-based literature that focuses on job creation potential in particular sectors, or geographies with a typical focus on the high emitting energy sector (e.g. Hepburn et al 2020, IEA 2020, Lewney et al 2021.

Methods and data
To evaluate the economic and environmental implications of fiscal policies we use an Input-Output (I-O) modeling framework. Two types of policies can be examined using this modeling approach-those that capture the impacts of exogenous increases in sectoral final demands and those that evaluate the consequences of exogenous increases in sectoral investments. Most of the existing applied research in this field has evaluated the demand multipliers by sector, even though much of this research (misleadingly) refers to demand multipliers as investment multipliers. This section provides clarity on the theoretical background and differences between the demand and investment multipliers. Though our focus in what follows is on investment multipliers 4 .
We present results on employment and income multipliers which are the most common measures used to summarize many economic impacts, as well as measures of impacts on the gender and the skill distribution of employment pattern. We complement these with measures of environmental implications that includes CO 2 , Non-CO 2 GHG emissions, and 4 Upon request from the authors results from the demand multipliers are available. These are not presented here due to our interest in exploring the potential of new sectoral investments in generate more environmentally benign employment and growth. a variety of other air pollutants (Chepeliev 2020a(Chepeliev , 2020b. This range of indicators provides information that is relevant in assessing the wider social and environmental implications of investment policies beyond the usual macroeconomic indicators. The theoretical foundation of using I-O tables to measure changes in final demand or investment has been widely discussed in the literature (e.g. Miernyk 1967, Richardson 1972, 1985, Pleeter 1980, Miller and Blair 1985 and used to evaluate multipliers at both regional and national levels (Lenzen 2001, Bekhet 2011, Cassar 2015, Liu and Liang 2017, Bivens 2019).

Income and employment multipliers
Using the I-O table of each country, we determine the following four sectoral multipliers: • dlm i : Employment multiplier for sector i due to an increase in its final demand by one million USD • dim i : Income multiplier for sector i due to an increase in its final demand by one million USD • slm i : Employment multiplier for sector i due to an investment in that sector by one million USD • sim i : Income multiplier for sector i due to an investment in that sector by one million USD Annex A fully describes the mathematical derivation process of each of these multiples.

Emission multipliers
Emissions' accounting could be considered across three different scales as suggested in the literature 5 . The most restrictive, Scope 1, considers only direct emissions from a single source (e.g. emissions associated with fuel combustion in a boiler). Scope 2 refers to indirect emissions associated with purchase of a typical good or service (e.g. purchase of electricity). Scope 3 takes into account a remaining part the entire supply chain from a life cycle perspective. In developing emission multipliers, we compute the sum of Scopes 1, 2 and 3 emissions which is the most comprehensive and policy appropriate measure that can also be readily computed from I-O tables (Peters 2008). For each type of emission (CO 2 , each element of Non-CO 2 GHG emission, and each element of air pollution) we develop two sets of multipliers: • ed er i : Demand multiplier of emission type e in region r for an increase in the final demand of sector i by one million USD • es er i : Supply multiplier of emission type e in region r for an increase in the final demand of sector i by one million USD A full derivation process of these emission multipliers is discussed in annex A.
It is important to note that the investment emission multiplier is a relative measure that compares changes in emissions induced by changes in the value chain of a sector, due to an investment of 1 million USD. Hence, an investment in what might be deemed a 'dirty' sector with high absolute emissions that also generates high value added, may lead to a small emission multiplier and vice versa. Having information on the magnitude of emissions that result from an investment of a given amount (i.e. the emission multiplier and not just absolute emissions), is one of the vital pieces of information that would be needed to determine the design of efficient 'green' policies that seek to decouple economic growth from environmental damage.
Data for the employment, income, and emission multipliers are from the GTAP-Power database version 10, which includes the global economy in 2014 and represents the I-O tables of 141 countries and regions, where regions represent aggregations of small countries by continents (Chepeliev 2020c).

Sectoral and geographical aggregation
The sectoral aggregation is presented in annex B. As shown in this annex, economic activities are classified into 37 sectors. The GTAP Power database divides the global economy into 141 countries. In this paper, we cover all countries presented in this database.

Results
This section presents some of the key findings from the simulations. We begin by ranking sectors by their employment, value-added, GHG and PM 2.5 emission multipliers to provide a global picture of the potential for job creation, economic growth, and 'greening' the economy through lower emission investments 6 . In the annexes we also provide regionally disaggregated estimates using the World Bank's classification of regions 7 .
We exclude the primary and secondary energy sectors (fossil fuels and electricity) that have been intensively examined in previous work. Overall, the analysis finds that investment in service sectors such as health, education, communication, and a few nonservice sectors such as forestry and construction are associated with relatively high job multipliers and lower GHG emission multipliers. In some countries agriculture also falls into this category, with crops (with the exception of rice) and non-ruminant livestock, being associated with lower emission and high employment multipliers. Conversely there are more severe trade-offs in other sectors where relatively high job growth is associated with large emission multipliers. As expected, the transport sectors have high GHG emission multipliers, with water transportation in particular exhibiting especially large GHG emissions. These are also sectors, where much of the benefits of job creation accrue to low-skill males. The direct employment multipliers of these sectors are usually small, but with large spillovers of jobs to the rest of the economy. In contrast, the 'greener' education and health sectors tend to have high direct job multipliers, but limited spill-over to the rest of the economy, resulting in lower indirect job multipliers.
Beyond these broad trends there is considerable heterogeneity in the job creation potential, valueadded, and emission intensities of economic activities across sectors and countries. Regardless of the global ranking of economic activities, the results suggest that the ratios of GHG emissions to job creation, or valueadded, vary significantly across regions for each sector and across sectors within each country. This suggests that generalizations can be misleading. It also points to the fact that opportunities could be available for greening many sectors and making them more inclusive. Different emission intensities of economic activities across countries typically reflect the fuel mix in an economy, and the reliance on fossil fuels. Without a fundamental shift towards cleaner sources of energy, economic activities in fossil fuel dependent countries will continue to generate higher levels of GHGs than in those with cleaner fuel sources.
Finally, it is important to note that access to national I-O tables that provide more detailed sectoral disaggregation could help identify tradeoffs between employment and emissions more effectively and accurately at the country level. The GTAP database used in the present study could be viewed as a first step in developing these more granular assessments of opportunities and trade-offs.

Ordinal ranking of economic activities
To identify common patterns in sectoral emissions and employment we divide each sector into terciles (33.3 percentile) first at the country level and then globally. At the country level, emission multipliers are labeled brown, for the 'dirtiest' industries in the highest tercile of emission multipliers; green representing the cleanest sectors in the lowest tercile of emission multipliers, and yellow for the intermediate tercile. The exercise thus identifies the relatively 'greener' sectors within a country in the first instance 8 . Investment in such green sectors would imply that a country can reduce the average emission 8 A sector with the same emission multiplier in two countries could be classified as Green in one and Brown in the other depending upon the emission performance of other sectors in each economy.
intensity of the economy, as a step towards decoupling economic expansion and pollution emissions on the path to greening the economy.
Turning next to a comparison of sectors across countries, we use a similar ranking procedure. If a sector is classified as green (yellow, or brown) in a plurality (i.e. relative majority) of countries it is labeled as a globally green (yellow or brown) sector 9 . This is one way of illustrating similarities (or differences) in the relative performance of sectors across countries.
Employment multipliers are ranked in an analogous manner: High, Medium, and Low to provide a broad summary across sectors. The High and Low classes represent the sectors in the terciles with the highest and lowest employment multipliers, while Medium includes the intermediate sectors. Similarly, at the global scale, a sector is labeled as High/Medium/Low if it falls in the High/Medium/Low category in a plurality of countries.
In summary, within country rankings provide a snapshot of where low emission and high job opportunities exist within the country, while the distribution of performance across countries can be used to identify outliers where attention may be needed to implement measures that can contribute to reducing emissions. Figure 1 summarizes the outcome of the ranking process for the emission multipliers. It shows that in an overwhelming majority (139 out of 141) of countries the Other Transportation sector is ranked in the highest tercile of GHG emission multipliers. Other brown sectors include: Livestock (excluding non-ruminants), Water Transportation, Water (including sewage), Air Transportation, Paddy Rice, and Non-Ruminant Livestock. There is much variation across countries in some sectors-such as the Non-Ruminant Livestock sector which is brown in a relatively smaller number of countries. It is notable that the ten sectors that fall in the intermediate category, generally exhibits much more heterogeneity across countries than do sectors in the green or brown categories. For instance, Trade with 97 countries and Mining with only 64 countries representing the upper and lower boundaries of a relative majority. On the other hand, the Green category includes: Public Administration, Health, Education, Textile-Wearing-Leather, Construction, Machinery-Equipment, Wood-Paper, Forestry, Communication, and Fishing.

GHG emission multipliers
In interpreting these results it is instructive to distinguish between absolute emissions and relative 9 Note that since there are three categories, a plurality (which identifies the highest number) rather than an absolute majority (which requires more than 50%), is the more appropriate criterion to classify the distribution of sectors globally. However, in most cases the plurality coincides with the absolute majority. emissions. The absolute emissions in a sector may be high, either due to the size of the sector (the scale effect), or because it has a high emissions intensity. The emission multiplier measures the latter-the relative emissions that result from an investment of a given amount. Hence, an investment in a sector with high absolute emissions but a large value chain would lead to a lower emission multiplier than a sector with the same level of emissions but a smaller value-added chain. If the aim is to green growth of an economy, then emission intensities provide valuable information as these capture changes due to (say) cleaner production techniques (holding value added constant), as well as changes due to improvements in productivity (holding emissions constant). Both changes are desirable to decouple economic growth from environmental damage. However, a focus on absolute emissions will ignore the benefits of expanding sectors with lower emission intensities, at the expense of others.
A notable example in figure 1 is the Construction sector which is categorized as green. However, Construction is often viewed as a major source of GHG emissions-but this may be due to the size of the sector in most countries. The value created by the sector is large 10 , with value-added (wages and payments to capital) accounting for a large share of the cost structure. Moreover, the polluting materials that are used in the sector are relatively inexpensive compared to the cost of other inputs. More importantly, the amount of the embodied emissions is low relative to the overall value of sectoral production. Hence, these factors contribute to a low GHG emission multiplier for the Construction sector. Another prominent example of a sector typically perceived as being highly polluting is the Textile-Wearing-Leather industry. While the sector uses emissions-intensive inputs such as cotton, industrial fibers, livestock products, and petrochemical products to operate, like the Construction sector, the embodied emissions in the final product is low relative to the value-added that is produced. This is partly because the combustion emissions due to energy consumption of the sector are not large. Thus, additional investment in the sector generates substantially greater additional output, relative to the emissions that it generates. The forestry sector is another case that may seem anomalous. It emerges as 'green' for a different reason. The GHG multiplier captures improvements in soil organic carbon and forest carbon sequestration due to investment in the forestry sector 11 .
The transportation sectors also highlight the importance of distinguishing between absolute and relative emissions. As shown in figure 1 rankings, these sectors are in the brown category with overwhelming majorities: 139 countries for land transportation (Transport-Oth); 134 countries for water transportation (Transport-Wat); and 130 countries for air transportation. At the global scale, among these sectors Transport-Oth generates the largest GHG emissions in absolute and relative terms. In absolute terms, due to the scale of the sector Transport-Air emits more emissions than Transport-Wat. However, in relative term, Transport-Wat is the larger polluter than Transport-Air. This is because the water transportation sector generates more GHG emissions per unit value added than Air transportation sector since it is heavily reliant on residual fuel oil (Mazut) which is among the dirtiest of liquid fossil fuels. This significantly increases the emissions intensity of this sector compared with other transportation sectors that use cleaner fossil fuels. An exclusive focus on absolute emissions would likely miss this nuance which is relevant for decoupling growth from emissions.
In fact, while the land and air transportation sectors are subject to many regulations across the world, the water transportation sector has not been subject to environmental regulations and hence continues to mainly rely on bunker fuel. National and international regulations could push the water transportation sector to move towards using cleaner fuels. Figure 1 only highlights the number of countries that form the relative majority for each sector and compares those across sectors and across the three categories of brown, yellow and green. However, it does not show the full distribution of countries across these categories for each sector. Figure C1 of annex C provides this complimentary information.
Using the same approach, we rank sectors by all individual GHGs and air pollutants. As an example, we analyzed the PM 2.5 multiplier and its relationship with the GHG multiplier in annex C (figures C2 and C3). The results confirm a robust positive relationship between these two multipliers. Figure 2 divides the non-energy sectors into terciles of the employment multipliers labeled as High, Medium, and Low. There are 12 sectors in the top 11 Note that the forestry sector represents managed forest activities and its products. Since managed forests typically entail managed rotation, these activities do not entail deforestation. tercile with high employment multipliers, with significant majorities, in Finance-Insurance, Health, Education, and Rice sectors. Seven sectors form the Medium category and include Warehousing with 95 countries in the upper bound of this category and Mining with 62 countries in the lower bound. There is much variation in the employment multiplier across countries in the intermediate multiplier sectors. Finally, there are eight sectors in the low employment multiplier category. Electronics in the upper bound with 117 countries and Wood-Paper with 79 countries are the lower bound of this class of employment.

Employment multipliers
The variation in employment multipliers across countries in a given sector often reflects differences in industry structure and technology. For instance, the Textile-Wearing-Leather sector falls in the Low employment multiplier category in a majority of 95 countries out of 141 countries, where the sector is found to be relatively capital intensive in the GTAP database. Conversely, in five countries the sector is relatively labor intensive with High employment multipliers and is in the intermediate range in 40 countries. Table 1 presents the joint distribution of the GHG and employment multiplier rankings as a general guide for fiscal stimulus packages that aim to maximize employment while controlling GHG emissions. It also points to sectors which have high job creation potential, but where attention may be needed to reduce pollution or GHG emissions, so that short-term efforts of recovery are not pursued at the cost of long-term sustainability.

Intersection between GHG and employment multipliers
The Air and Water Transportation sectors occupy the least attractive spot of this table in the Low-Brown cell which represents low employment and high GHG categories. On the other side sectors such as Construction, Education, Fishing, Forestry 12 , and Health lie in the 'sweet spot' of High employment and Greener multipliers in a large number of countries 13 . Other sectors such as Hotel-Restaurant-Recreation, Mining, Processed Food, and Warehousing fall in the middle ranges of the joint distributions of employment and GHG multipliers, or the edges of table 1 which represent various mixes of employment and GHG categories.

Gender and skills in employment
The distributional consequences of greening economies will be of considerable importance, as the fruits of a green transition will not be evenly distributed and could result in greater inequality. To explore the extent of such risks globally and by sector annex C estimates both the gender distribution of employment as well as the skill mix of new jobs created. In cases where greening investments accentuate prevailing inequalities there would be a need for governments to provide support whether it is through training schemes, regional development support, or social protection programs. The analysis finds that in many cases there are trade-offs between greener Notes: (a) The GTAP database used in this paper aggregates the global economy in 141 countries or regions. We divided these countries into three Low-, Middle-, and High-income groups according to the World Bank data 14 .
(b) Green, yellow, and brown represent terciles of GHG multipliers. That is the colors represent the lower 33.3%, middle 33.3%, and top 33.3% of the sectoral GHG multipliers in each income category.
(c) The first number in each cell shows number of countries in the corresponding tercile by sector in each income group (e.g. for the low-income group, in 46 countries the rice sector falls in the top 33.3 percentile of the sectoral GHG multiplier of this income group).
(d) The second number in each cell represents the average value of GHGEMP index measured in metric ton of CO2 equivalent per employment (e.g. for the low-income group, for the rice sector the value of GHGEMP is 28 metric ton per employment).
(e) The third number in each cell represents the average value of GHG intensity of value added (GHGVAL) index measured in metric ton of CO2 equivalent per 1000 USD (e.g. for the low-income group, for the rice sector the value of GHGVAL is eight metric ton per 1000 USD. investments and the outcomes on women and lower skilled workers. The detailed analyses is provided in annex C (figures C4 and C5).

Cardinal rankings
The ordinal rankings provided thus far are useful but ignore the distance between sectors across the employment and GHG indices. This subsection seeks to address this issue by providing measures that 14 World Bank definitions classify economies into four income categories: low-income, lower-middle-income, upper-middleincome and high-income (https://datahelpdesk.worldbank.org/ knowledgebase/articles/906519-world-bank-country-and-lending -groups). In our country classification, we map World Bank's lowincome and lower-middle-income economies to the low-income group, upper-middle-income countries to the middle-income group and high-income economies to the high-income group.
facilitate comparisons by normalizing the indices to account for differences in the size of economies. A $1 million stimulus will have negligible employment effects in a large country than it would in (say) a small island economy. To allow for meaningful cardinal comparisons across countries this subsection standardizes the multiplier in terms of the emissions per unit of employment and emissions per unit of value added. We labeled these cardinal measures as GHG intensity of the employment multiplier (GHGEMP) and GHGVAL, respectively. Annexes D and E respectively introduce these indices and rank economic activities based on their values. The results provided in these annexes suggest that the employment and value-added multipliers often diverge implying that there are tradeoffs not only between economic and environmental objectives but at times also between growth and employment priorities. The results also recommend that when there is a desire to limit GHGs within a constrained fiscal envelope, selecting amongst sectors with low GHGEMP and GHGVAL ratios may provide a useful way of delivering on both growth and environmental objectives. Fortunately, there is a convergence across sectors with high employment potential and high value-added multipliers, easing somewhat the trade-offs between these economic policy objectives. However, as the graphs suggest, heterogeneity across countries is significant, necessitating more detailed analyses. Additional discussion for selected sectors and regional dimensions is included in annexes D and E.

Regional analyses of the rankings
The previous sections have assessed the greenness of job creation due to sectoral investments and their corresponding economic and environmental effects using the employment, income, and GHG multipliers from a global perspective. This section provides a summary of these analyses by region to assess the extent to which the results may vary across the world. While annex E provides some sectoral analyses by region, this section uses the regional and income classification defined by World Bank to divide countries into three categories of Low-, Middle-and High-income and evaluate the economic and environmental impacts of the sectoral investment for these groups. Table 2 shows the results. This table summarizes a large amount of information which shows the greenness of each sector for each income group by the color codes defined in section 3.1, which rank countries by GHG multipliers. The table also presents the GHGEMP and GHGVAL indices by sector and income group. This table also compares the results by income groups with the global results as well. The results suggest the following important findings: • There is remarkable uniformity in some sectors across the world. Several sectors including crops, rice, livestock, and all transportation sectors remain in the brown category across the three income groups. Various sectors including public administration, health, education, textileswearables-leather, wood-paper, and construction remain green across the three income groups. Four sectors including finance-insurance, trade, mineral-metal, and chemicals-rubber remain in the yellow category across the income groups. The greenness of other sectors varies across income groups. • On the other hand, the GHGEMP and GHGVAL indices vary significantly across industries and income group. Two dominant results emerge: First, the GHGEMP values are usually lower in higher income countries. This reflects the facts that higher income countries: are less labor intensive (more capital-intensive); usually use more advanced production technologies with lower emissions intensities; and implement more advanced and efficient environmental regulations systems. Second, for a given sector, the GHGVAL values are usually lower in the higher income countries. This is also because the value added is higher per unit of employment in the higher income countries, which use more capital-intensive technologies with lower emissions intensities. • The sectors that are labeled brown at the global level are also brown in the low-, middle-, and highincome countries. This suggests that additional investments in these sectors should aim to reduce their emissions intensities worldwide. This requires establishing worldwide mitigation policies for each of these brown sectors.

Conclusions
As countries seek to restore economic growth and employment in a context of mounting debt and limited fiscal space, they confront difficult choices on ways to spend scarce budgets to achieve multiple goals. This paper presents data that could inform these decisions when spending entails sharp tradeoffs between economic objectives and wider environmental concerns. Three broad conclusions emerge from the results of the I-O analysis presented in this paper. First, the analysis finds, predictably, that many of the traditional sectors in agriculture and industry have large employment multipliers, but also generate male dominant, lower skill employment, and tend to have higher emission multipliers. It is in economies dominated by these sectors that trade-offs to a 'greener' transition will emerge most sharply. The good news is that even in these economies, there exist other sectors with high employment multipliers and low emissions, including sectors that are more conducive to female employment.
Second, the results suggest that meeting multiple objectives with a limited budget will typically entail making different choices than would eventuate with a single objective-such as maximizing employment or minimizing GHGs.
Finally, the results uncover significant heterogeneity across countries, especially across the emission multipliers and the gender distribution of employment. This implies that there may be scope for countries that lag in these dimensions to significantly improve their performance over time by emulating those that lead.
Achieving 'catch-up' may require some combination of adopting new technologies and business models and making progress on the supply side of the labor market. Most notably, many countries can close the gap by changing their underlying energy mix, which directly affects GHG multipliers without otherwise changing the production structure of downstream industries. The extent to which changing the energy mix, and more broadly achieving 'catch up' is feasible over the medium term will depend on country capacity, technical endowments, and comparative advantages, as well as domestic political economy. Beyond this, trade may offer a way to promote convergence to a more desirable equilibrium, with the right incentives.
The I-O modeling framework used in this paper is not without limitations. An important caveat is the fixed coefficient property which precludes substitutions that could happen in production, consumption, and trade relationships in the long run. Additionally, since the Social Accounting Matrixes that provide the data for the model are updated by countries with varying frequency, the analysis may not represent the most recent data for each country, in particular the recent attempts to shift the energy mix towards renewables may not be fully reflected.

Data availability statement
The data that support the findings of this study are available upon request from the authors.

Acknowledgments
The authors would like to thank Freddie Taylor, Associate Editor of the journal, and the anonymous reviewers for their helpful comments and constructive suggestions. This study received financial support from The World Bank. The views expressed in this paper are the authors' only.

A1. Income and employment multipliers
The relationship between the demand and supply sides in an I-O framework is described by the following equation: This equation is written in matrix notation, where capital letters represent matrixes or vectors, and their corresponding lower-case letters show their elements. In equation (1) Y is a vector of n × 1 and its elements are: y i , value of production of commodity i which is equal to the value of output of sector y j for i = j; I is an identity matrix of n × n; A is a matrix of n × n and its elements are: a ij = x ij y j , where x ij stands for the intermediate use of commodity i used in sector j; and F is a vector of n × 1 and its elements f i represent final demand for commodity i. Using equation (1), assuming no technological progress (i.e. no change in matrix A), one can determine the impacts of changes in the sectoral final demands (∆F) on the sectoral outputs (∆Y) using the following equation: Define now ∆F i as a vector of n × 1 with a nonzero entry of ∆f i and zeros elsewhere. This implies: In an I-O framework, for a given sector the link between output and each input represents a linear relationship. In this type of production function, in each sector, the ratios of labor (l i ), capital (k j ), and other primary input 15 (o j ) to output (y j ) are fixed. Denote these ratios with γ i , η i , and φ i , respectively. Note that an I-O table provides values of these ratios. Therefore, in an I-O framework the following equations represent the links between the output and primary inputs in each sector: Using equation (3), we can determine changes in sectoral outputs (i.e. ∆y 1 , ∆y 2 , . . . , ∆y n ) for a given ∆F i . Using equation (4) in combination with the changes in sectoral outputs for a given ∆F i we can determine the sectoral labor multipliers (dlm i ) by the following equation: dlm i = j γ j ∆y j , for j from 1 to n.
Using equations (4)-(6) in combination with the changes in the sectoral outputs for a given ∆F i , we can determine sectoral income multipliers (dim i ) using the following equation: In this equation, ω i , τ i , and λ i are per unit sectoral payments to labor, capital, and other primary input.
Note that we refer to dlm i and dim i as demand multiples, because they are determined due to changes in the final demand, ∆F i .
On the other hand, in an I-O framework, the following relationship establishes the links between production and primary production inputs such as labor, capital, and resources: where Y ′ is the transpose of vector Y; V ′ represents transpose of vector of value-added (payments to the primary inputs as mentioned above); and B is a matrix of n × n and its elements are b ij = x ij y i . Note that the elements of matrices A and B are not identical as: x ij y i , except when i = j. Using equation (9), assuming no technological progress, one can determine the impacts of changes in the sectoral value-added on sectoral outputs using the following equation: Define now ∆V ′ i as a vector of 1 × n with a nonzero entry of ∆v i and zeros elsewhere. This implies: In an analogous manner to the demand multipliers, we can use equation (11) in combination with equations (4)-(6) and determine supply multipliers due to changes in the sectoral investment for a given set of sectoral rates of return on capital as explained in what follows (see Miller and Blair 1985). To distinguish between demand and supply multipliers, we show the sectoral supply multipliers of labor and income with slm i and sim i . We also add a superscript (s) to the changes in sectoral outputs due to changes in sectoral investments (∆y s 1 , ∆y s 2 , , . . . , ∆y s n ) to differentiate them from changes in sectoral outputs due to change in final demands.
Using equation (1), we can determine changes in sectoral outputs (i.e. ∆y s 1 ,∆y s 2 , . . . , ∆y s n )for a given ∆V ′ i . Given a set of sectoral rates of return and using equation (4) in combination with the changes in the sectoral outputs for a given ∆V ′ i we can determine sectoral labor multipliers (slm i ) by the following equation: Using equations (4)-(6) in combination with the changes in the sectoral outputs for a given ∆V ′ i , we can determine sectoral income multipliers (sim i ) by we use the following equation:

A2. Emission multipliers
Emissions accounting could be considered across three different scales as suggested in the literature 16 . The most restrictive, Scope 1, considers only direct emissions from a single source (e.g. emissions associated with fuel combustion in a boiler). Scope 2 refers to indirect emissions associated with purchase of a typical good or service (e.g. purchase of electricity). Scope 3 takes into account a remaining part the entire supply chain from a life cycle perspective. In developing emission multipliers, we compute the sum of Scopes 1, 2 and 3 emissions which is the most comprehensive and policy appropriate measure that can also be readily computed from I-O tables (Peters 2008). For each type of emission (CO 2 , each element of Non-CO 2 GHG emission, and each element of air pollution) we develop two sets of multipliers: one for changes in the sectoral final demands and the other for changes in the sectoral investments. Air pollutants are measured in metric tons per million USD.
To calculate the demand emission multipliers, the following conventional equation that has been defined to calculate changes in emissions induced by a given change in the final demand for a given sector can be used: In this equation the subscripts/superscripts of i, j, m, e, and r represent indices for goods (or services), sectors, primary inputs, types of emissions, and regions, respectively. The variable names used in this equation are: The emissions intensity variables defined in this list are obtained from the benchmark data set described in section 3. The I-O simulations mentioned above provide the values of ∆x r ij , ∆p r mj , and ∆y r j . The changes in sectoral outputs (∆y r j ) will be determined by equation (3). The changes in intermediate demands (∆x r ij ) will be determined from the fixed ratio property of I-O models using a ij = x ij y j ∆ ⇒ x ij = a ij ∆y i . The changes in demands for primary inputs of labor ∆p r labor,j = ∆l r j , capital ∆p r capital,j = ∆k r j , and other primary input ∆p r other,j = ∆l r j will be determined by equations (4)-(6). The last component of the above equation represents direct emissions due to final demand for 1 million USD. To avoid double counting there is no need to consider emissions due to exports of i form r to other countries. In an analogues manner, the investment emission multipliers (es er j ) are obtained by simply excluding emissions induced by final demand (i.e. ∆F r i ).

Annex B. Sectoral aggregation
The original GTAP database used in this paper divides all economic activities into 76 categories (sectors). While this detailed database provides an opportunity to develop detailed economic and environmental and analyses in some certain national and regional levels, for a global macro level research, similar to what we developed in this paper, it is not productive nor practical to follow many insignificant small details that the original database provides. Therefore, in this paper, we use an aggregated scheme of the original GTAP database to concentrate on important sources of heterogeneities across countries and economic activities. In the aggregated scheme we combine homogenous activities that represent similar cost and demand structures to reduce the number of sectors. The following table shows the original GTAP sectors and their map to the aggregated scheme used in this paper.
To explain the aggregation process, here we provide an example. The GTAP database represents all cropping activities under 8 different sectors of pdr, wht, gro, v_f, osd, c_b, pfb, and ocr. Each of these sectors represents one or several crops. The descriptions of these sectors are provided in the following table.
In this study we kept the sector of pdr (paddy rice) as an individual activity, as among sectors that produce crops, this sector is the main source of non-CO 2 emissions at the global scale. However, we aggregated other cropping activities in one aggregated sector and labeled that as 'Crops' .
To accomplish the aggregation task, we used the GTAP-AGG program which is designed to accomplish this assignment.

C1. Additional analysis for gender and skills in employment
One such example of distributional impacts governments might consider when making sectoral investments is that on gender equality of jobs multipliers. Indeed, while the total job creation potential of an investment is important, governments may also be concerned about the impact on female labor force participation rates. Figure C4 presents the share of female labor in the employment multiplier, at the global scale (the red line) averaged across sectors and in each sector, mapped to the GHG classification presented in figure 1 (in the main text). On average the share of female labor in the employment multiplier is about 37% globally with significant variation ranging from a low of about 12% in construction to around 65% in the health sector. An especially notable feature is that the sectoral share of female labor is above the global average mainly in some of the Yellow and Green categories including Trade; Financial and insurance; Processed food; Hotels, Restaurants, and Recreational Activities; Health; Education; and Textile-Wearing-Leather sectors. Thus, when designing labor policies which emphasize the importance of female labor, one could unintentionally also generate environmental co-benefits. Next, figure C5 replaces the share of female labor with the share of unskilled labor, which is another area governments may wish for their investments to target. The results indicate that on average the share of unskilled labor in the employment multiplier is about 71% at the global scale (indicated by the blue line). As one might expect, figure C5 shows that the share of unskilled labor is significantly lower than the global average in many service sectors such as Education, Health, and Public Administration (about 41%). On the other hand, the share of unskilled labor is considerably higher than the global average in the Crops and Livestock industries (87% in Livestock and 90% in Crops). The share of unskilled labor is usually close to the global average in many industries that fall in the Yellow category.
A comparison between figures 1 and C5 indicates that the shares of female labor and unskilled labor in the overall employment multiplier do not coincide in many sectors at the global scale, suggesting that policy makers may face tradeoffs between jobs that promote gender parity and those that promote employment amongst the lower skilled. Figure C1. Number of countries that labeled Brown, Yellow, or Green for each non-energy sector based on the sectoral GHG emission multipliers.

Annex D. Cardinal ranking of GHG emissions and employment
We define the GHG intensity of the employment multiplier (GHGEMP), as the ratio of the GHG multiplier to the employment multiplier. It provides a metric, normalized by employment, of the relative 'greenness' of job creation in a sector. In general, a smaller ratio is desirable as it indicates lower emissions per job created for a given investment in the sector. We report four measures across the distribution to compare sector performance-an overall average as well as the low, middle, and high terciles of the GHGEMP index. Figure E1 summarizes the results: • Panel A shows sectors with the smallest overall average of GHGEMP. The index ranges from 0.76 to 4.9 metric tons of GHG per unit of employment in Panel A (see the green bars). This group mainly contains services sectors and a few industries (e.g. Hotels and Recreation), that have low GHG emissions per unit of employment.
• Panel B represents the intermediate sectors. The overall GHGEMP ranges from 5.9 to 23.6 metric tons of GHG per unit of employment (yellow bars). This group is dominated by industries and a few service sectors with significantly higher values of the index than in Panel A. • Panel C shows sectors with the worst performance (see the brown bars). This group comprises the dirtiest industries and services, with GHGEMP values that are extremely large, and significantly higher than the values of the other two groups.
In summary, the cardinal rankings provided in figure D1, identify a wide class of sectors that generate low GHG emissions for a given employment multiplier. These include Health for a given employment multiplier. It is also notable that many of the low emission sectors are also more gender neutral in their employment creation potential, whereas the 'browner' sectors, tend to be in agriculture and the traditional industries that generate more employment for lower skilled males.
The heterogeneity of performance across countries shown in figure D1 could be explained by various factors such as variations in capital intensity, differences in wage rates and prices across countries, variations in production technology, and different energy mixes. These sources of heterogeneity highlight potential opportunities for countries to shift towards cleaner production processes. It also suggests that proper trade policies could limit global emissions by directing demand towards countries with lower emission intensities.

D1. Regional analyses
In this section we provide a sector-by-sector analysis of the relationship between the GHGs and employment multipliers by region. To accomplish this task, we continue to use the index of GHGEMP by sector and country. To highlight regional differences, we divide the whole world into seven regions of East Asia and Pacific (EAP), Europe and Central Asia (ECA), Latin America and Caribbean (LAC), Middle East and North Africa (MENA), North America, South Asia (SAS), and Sub-Saharan Africa (SSA) following the World Bank's definitions. For each sector, this analysis divides countries into five categories of Green (first 20 percentile), Light Green (second 20 percentile), Yellow (third 20 percentile), Light Brown (fourth 20 percentile), and Brown (fifth 20 percentile) based on their GHGEMP index. The analysis concentrates on three measures of GHGRMP in each sector: the Maximum, Minimum and Average values for each percentile category. This annex depicts a figure for each sector representing these data items plus number of countries falling into each percentile by the regional aggregation mentioned above. In what follows, we analyze the results for three representative sectors, one from each panel of figure D1: Education from Panel A, Processed food from Panel B, and Air Transportation from Panel C. The education sector represents a very low GHG intensity of the employment multiplier with a large direct employment effect and low employment spill over impacts on the rest of economy. It has a large female employment share as well. On the other hand, the food processing sector which belongs to the middle tier of GHG intensity of the employment multiplier has strong forward and backward links to other activities, make moderate to strong spill over employment in particular for female employment. Finally, the air transportation sector has a very large GHG intensity of the employment multiplier, very strong forward and backward links to other sector, major spill over employment effect, generating more indirect employment than direct employment. Figure D2 shows the information for the education sector which has a low overall global average of 1.1 metric tons of GHG emissions per unit of generated employment. The min, max, and average values are below 0.5 metric tons per employment in the first two categories of Green and Light Green. Members of LAC, MENA, and SSA mostly fall in these two categories. For example, ten members of LAC, five members of MENA, and seven members of SSA fall in the Green category. As shown in figure D2, the average of GHGEMP in the Yellow category is about 0.75 metric tons of GHG emissions per employment (see the yellow bar). In this category, 14 and 7 countries are members of ECA and SSA, respectively. Only four members of MENA region fall in this category. The average of GHGEMP increases to 1.4 metric tons of GHG emissions per employment in the Light Brown category and most members of this category are from ECA region (17 countries) or SSA region (7 members). Moving from the Light Brown category to Brown category, the average GHG emissions per employment increases sharply from 1.4 to 3. Most members of the brown category are from EAP (ten members) or ECA (nine members). Figure D2 also shows that as we move from the Green to Brown, the max value of GHEMP increases sharply. For example, the max value in the Brown category reaches to 4.5 metric tons per employment, 1.5 times higher than the average of this category.

D2. Education
In summary, for the sector of Education, the members of LAC, MENA, and SAS fall in the low percentiles of the GHGEMP index, while members of EAP and ECA belong to the higher percentiles of this index. Members of SSA appear in all most categories from the Green to Brown. There is a wide range and a large number of countries in EAP and ECA representing considerable higher GHG intensity of the employment multiplier in the sector. In some of these countries the energy intensity of education is higher than other countries. On the other hand, more coal used the mix of consumed energy in some these countries and that leads to higher GHG emissions. Finally, it is important to reemphasize even the GHG intensity of employment multiplier is significantly low for the education sector, even in the countries that fall in the top percentile.

D3. Processed food
From the second panel of figure D1, we further analyze the sector of Processed Food. Figure D3 shows the information for this sector which has a relatively moderate overall global average of 8.2 metric tons of GHG emissions per generated employment. For this sector, the min, max, and average values are below 4.2 metric tons per employment in the first two categories of Green and Light Green. Members of SSA (16 countries), SAS (4 countries), and LCA (4 countries) are the dominate members of Green category of this sector, while members of the Light Green category scattered around the aggregated regions.
As shown in figure D3, the average of GHGEMP in the Yellow category of this industry is about 5.7 metric tons of GHG emissions per employment (see the yellow bar). In this category, 11 and 9 countries are members of LAC and ECA, respectively. Only three members of MENA region fall in this category. The average of GHGEMP increases to 9.2 metric tons of GHG emissions per employment in the Light Brown category and most members of this category are from ECA region (13 countries), EAP region (6 members). Moving from the Light Brown category to Brown category, the average GHG emissions per employment increases sharply from 9.2 to 22.2. In this sector, most members of the brown category are from ECA (15 members) or EAP (5 members). In this category, four countries are from LAC. Figure D3 also shows that as we move from the Green to Brown, the max value of GHEMP increases sharply. For example, the max value in the Brown category reaches to 32.1 metric tons per employment, about 1.5 times higher than the average of this category.
In summary, for the sector of Processed food, the members of SSA, LAC, MENA, and SAS fall in the low percentiles of the GHGEMP index, while members of EAP and ECA belong to the higher percentiles of this index.

D4. Air transportation
Finally, from the third panel of figure D1, we further analyze the sector of Air transportation. Figure D4 shows the information for this sector which has a large overall global average of 176.5 metric tons of GHG emissions per generated employment. For this sector, the min, max, and average values are below 40 metric tons per employment in the first two categories of Green and Light Green. Members of SSA (17 countries) and LCA (7 countries) are the dominate members of Green category of this sector, while members of the Light Green category scattered around the aggregated regions.
As shown in figure D4, the average of GHGEMP in the Yellow category of this industry is about 59.9 metric tons of GHG emissions per employment (see the yellow bar). In this category, ten and six countries are members of LAC and ECA, respectively. Only four members of MENA region fall in this category. The average of GHGEMP increases to 187.1 metric tons of GHG emissions per employment in the Light Brown category and most members of this category are from ECA region (12 countries), EAP region (five members), and MENA (six countries). Moving from the Light Brown category to Brown category, the average GHG emissions per employment increases sharply from 187.1 to 604.4. In this sector, most members of the brown category are from ECA (19 countries) and EAP (5 countries). Figure D4 also shows that as we move from the Green to Brown, the max value of GHEMP increases sharply. For example, the max value in the Brown category reaches to 1249.3 metric tons per employment, more than twice of the average of this category.
In summary, for the sector of Air transportation, the members of SSA and LAC fall in the low percentiles of the GHGEMP index, while members of EAP, and ECA, belong to the higher percentiles of this index and MENA countries across all categories except the lowest tier. Similar graphs are available for other sectors upon the request.

Annex E. Cardinal ranking of GHG emissions and value-added
Investments that have relatively high employment multipliers may not necessarily have high valueadded (income) multipliers. To further explore this relationship, we develop a new index comprising the ratio of GHG emissions to value-added multipliers (henceforth GHGVAL). Analogous to the previous case, we report: the overall global average for each sector, and averages across terciles in the first, second and third 33.3 percentiles. Figure E1 summarizes the outcomes and shows these indices by sector in three panels: • Panel A shows sectors with the smallest overall average value which ranges from 0.06 to 0.37 Kilograms (Kg) of GHG per $ of value-added generated (see the green bars). This group includes service sector industries such as public administration and health that could be expected to have low emission to value-added multiplier ratios and the same industries as identified in the previous ranking of GHGEMP. • In Panel B the intermediate range is dominated by industries with relatively low GHG emissions per unit of value-added, but significantly larger than the values of lowest group presented in Panel A. • Panel C comprising the relatively 'dirtiest' industries have GHGVAL values that are significantly higher than in the preceding groups. These are the sectors that may warrant priority investments in pollution abatement.
Notably, both rankings (GHGEMP and GHGVAL) identify similar industries in the three blocks. Health, Public Administration, Education, Construction, Hotel-Restaurant-Recreation, Trade, Communication, Textile-Wearing-Leather, and Forestry generate lowest GHG emissions per unit of employment and value-added. On the other hand, Crops excluding Paddy Rice, Chemical-Rubber, Non-Ruminant Livestock, Other Transportation, Paddy Rice, Water-Sewage, Livestock excluding Non-Ruminant, Air Transportation, and Water Transportation generate the largest emissions per job created and value-added generated. The implication for policy makers is that when there is a desire to limit GHGs within a constrained fiscal envelope, selecting amongst sectors with low GHGEMP and GHGVAL ratios may provide a useful way of delivering on both growth and environmental objectives. Fortunately, there is convergence across sectors with high employment potential and high value-added multipliers, easing somewhat the trade-offs between these economic policy objectives.
However, as the graphs suggest, heterogeneity across countries is significant, necessitating more detailed analyses. In what follows, similar to the case of GHGEMP index, we analyze the results for the three representative sectors of Education, Processed food, and Air Transportation. Figure E2 shows the information for the sector of education which has a small overall global average Figure E1. Global ranking of non-energy sectors based on generated GHG emissions per generated value-added due to sectoral investment by $1 million. of 0.17 Kg of GHG emissions per generated valueadded. For this sector, the min value is tiny (about 0.005 Kg $ −1 ) but the max value is significantly larger (1.12 Kg $ −1 ).

E1. Education
The average value of GHGVAL for the first four categories of Green, Light Green, Yellow, and Light Brown remains under 0.14 Kg $ −1 . However, the average value of GHGVAL suddenly increases to the value of 0.62 Kg $ −1 in the Brown category. That is basically due to higher values of GHGVAL among the members of SSA region, as shown in figure E2. This figure shows that most members of SSA falls in the Light Brown and Brown categories of GHGVAL. The size of this index in the SSA region is high due to low wage rates. Most members of LAC and MENA regions fall in the Green, Light Green, and Yellow categories of education, while the members of EAC region scattered across all various categories of Green to Brown. Figure E3 shows the information for the processed food sector which has a relatively small overall global average of 0.53 Kg of GHG per $ of value-added generated. For this sector, the min value of GHGVAL is very limited (0.066 KG/$), but the max value is significantly larger (3.11 Kg $ −1 ) compared with the mean value. In this sector the mean value increases gradually from 0.12 Kg $ −1 for the Green category to 0.5 Kg S −1 for the Light Brwon category and then it suddenly jumps to 1.52 Kg $ −1 in the Brwon category. This jump is basically due to low wage rates in several countries in the SSA, SAS, MENA, and EAP regions.

E3. Air transportation
Finally, figure E4 shows the information for the air transportation sector which has a large overall global average of 5.4 of GHG per $ of value-added generated.  For this sector, the min value of GHGVAL index is low (0.97 Kg $ −1 ). However, its max value is significantly larger (20.86 Kg $ −1 ). The mean value of this index for the categories of Green, Light Green, and Yellow remains under 3.7 Kg $ −1 , but it increases to 6.2 Kg $ −1 in the Light Brwon category and then sharply elevates to 13.1 Kg $ −1 in the Brown category. As shown in figure E4 many members of the ECA regions falls in these two high emissions categories. This figure indicates that 10 members of the ECA region belong to the Light Brown category and 14 of them fit in the Brown category. Figure E4 also shows that 12 members of the EAP region fall in the Light Brown and Brown categories.
Similar graphs are available for other sectors upon the request.