GLOBAL REDUCTION OF CO

1

GLOBAL REDUCTION OF CO

EMISSIONS THROUGH TECHNOLOGY

DIFFUSION

Master Thesis Economic Development & Globalization Faculty of Economics and Business

University of Groningen January 7th, 2020

2

GLOBAL REDUCTION OF CO

EMISSIONS THROUGH TECHNOLOGY

DIFFUSION

Luuk Appels

Abstract: This paper calculates the global potential for CO2 emission reduction, assuming

best-practice technologies could diffuse across countries and sectors. We use data from the World Input Output Database (WIOD) and a recently created database by Timmer et al. (2019) containing information on functional specialization, which relates to the degree to which countries/industries specialize in certain functions, or parts, of the production process. By using input output analysis, we estimate the global emission reduction potential and show that according to our most comprehensive estimate, a global reduction in CO2 emissions of 59%

could be achieved through the diffusion of best-practice technologies. Furthermore, our findings show that accounting for functional specialization is associated with a sizeable reduction of emission reduction potential compared to calculations where functional specialization is ignored, underlining the important role functional specialization plays in CO2

emissions reduction calculations. The main contribution of this study is that it is the first to take into account the notion of functional specialization in emissions research.

Keywords: Input output analysis, CO2 emissions, technology diffusion, functional

3 Table of Contents

1 Introduction . . . 2 Literature Review . . . 2.1 Global differences in CO2 intensities . . .

2.2 Technology and diffusion . . . 2.3 Global Value Chains and functional specialization . . . 2.4 Relevance and expectations . . . 3 Methodology . . .

3.1 Data . . . 3.2 Input Output Analysis . . . 3.3 Calculating emission reduction potential: conventional ways . . . 3.4 Calculating emission reduction potential: accounting for functional

4 1. Introduction

The recent global climate strike, which took place between the 20th and 29th of September 2019,

is evidence of the increasing public pressure on politicians to take the bull by the horns when it comes to climate change and its consequences for the planet and humanity. Research has shown that the largest contributor to the changing, and particular warming, of the climate over the long term is emissions of carbon dioxide (CO2) (Peters et al., 2012). If CO2 emissions are responsible

for a substantial share of climate warming, what then drives CO2 emissions? Feng et al. (2015)

quantify the drivers of US CO2 emissions between 1997 and 2013 and find that the most

important drivers of CO2 emissions were the growth of population and per capita consumption.

Evidently, both these drivers require production to rise which, in turn, increases the need for energy, fuel and many other resources that produce CO2 emissions when generated, burned, or

used in another way. One of the main forces working to depress CO2 emissions is technological

change, which can reduce CO2 emissions in two ways, namely through innovation and

diffusion. An innovation could for instance be the development of a new technology enabling more efficient ways of generating power, reducing CO2 emissions. In the context of innovation,

diffusion relates to the process of adopting a new technology, or replacing an older technology with a newer one (Hall, 2004).

In this paper we will focus on the channel of technology diffusion to explore the potential for global CO2 emission reduction. This means we do not consider the invention of new

technologies, but focus on already existing technologies. In particular, the focus will be on the diffusion of the most emission-efficient (i.e. cleanest) technology available, which is assumed to be the one used in the country emitting the lowest amount of CO2 per dollar of output. More

specifically yet, this paper will attempt to answer the question: By how much could global CO2

emissions be reduced if technology diffusion could facilitate the adoption of best-practice technologies in each sector of each country? Due to data availability the year under investigation is 2007. This may lead to an over- or underestimation of the emission reduction potential of more recent years depending on (i) the degree to which technology diffusion has taken place between 2007 and 2019 and (ii) the extent to which best-practice technologies have been improved between 2007 and 2019. As technology diffusion is the main channel of CO2

reduction being investigated, other factors such as consumption and its composition, investment levels and trade patterns are assumed to remain unchanged, in order to isolate the maximum emission reductions from technology diffusion. In addition, as this paper takes a global approach, we define best-practice technology at the national level. Evidently, within a country, some firms will be cleaner than others and even more emission reductions could be achieved if technology diffusion could take place between firms, too. Therefore, this paper will underestimate the potential for emission reduction to the extent that within each country, between-firm differences exist in terms of cleanness of production.

In climate research, ample studies have been conducted on CO2 emissions, both on the national

and the sector level. Part of this literature focuses on the quantification of CO2 emissions in a

specific sector or country which is useful for envisioning the size the problem. Other scholars have gone one step further by attempting to calculate the total potential for CO2 emission

5

in the sense that it helps understand the potential for improvement on a small (country/sector) scale and may aid policy making and the determination of targets, it is less useful for understanding the global reduction potential.

The most straightforward way to explore the potential for global CO2 emission reduction is to

look at country-level CO2 emissions per dollar of output. We could then identify the country

using the best-practice technology, and see what could be gained if other countries were just as clean, i.e. adopt this best-practice technology. However, this calculation neglects the sector structure of countries. Industrial economies such as China will emit more CO2 per dollar of

output than service economies such as the US, simply because they are home to more polluting sectors. In order to make this comparison fairer, we could compare sectors across countries, in order to account for these discrepancies in sector structures. This would allow us to calculate, per sector, the potential reduction in CO2 emission reductions through technology diffusion.

Summing these then gives us the total global reduction potential.

However, we would still be overlooking the degree to which countries specialize in certain functions, or parts, of the production process. This ‘functional specialization’ as Timmer et al. (2019) call it, is one of the most recent developments in the world economy and relates closely to the concept of Global Value Chains (GVC’s). A GVC of a final product is ‘all activities that are directly and indirectly needed to produce it.’ (Timmer et al., 2014, p100). Timmer et al. (2019) define four functions (marketing, R&D, fabrication, and management) and find large differences in specialization patterns within sectors across different countries. In other words, within a given GVC, the chemical industry in China may in fact be performing activities that differ greatly from the US chemical industry. Indeed, the data of Timmer et al. (2019) show that China’s chemical industry is heavily specialized in fabrication with 62% of labor income being earned in this function, and other functions being only marginally important. Much to the contrary, the US displays an even division of labor income across the four functions, underlining the differences in functional specialization patterns between these countries in the chemical sector. To realize how these two sectors form part of a GVC, note that it is likely that the US performs R&D activities in the chemical sector, the outcomes of which are then fabricated in China, generating value added in both countries.

The functional specialization of trade has implications for comparing sectors across countries. That is, if we compared sectors across countries and made inferences based on the results, we would be ignoring the fact that country A is responsible for marketing in the car industry, for instance, whereas country B performs the fabrication. Evidently, the former will produce much less CO2 emissions than the latter. Therefore, we would be wrong in concluding that country B

could adopt the best-practice technologies used by country A, because they perform entirely different activities and, by extension, use different types of technology. For this reason, functional specialization plays an important role in the goal of this research as it imposes a restriction on which countries can reasonably be compared to one another within a sector and in doing so, will have an impact on the potential for emission reduction.

6

country. This is the first data source that we use. In order to correctly estimate the total potential emission reduction from technology diffusion, the second data source we use is the World Input Output Database (WIOD) (Timmer et al., 2015). When technology diffuses across sectors and countries, not only will CO2 emissions drop, but sectors will also require less inputs from the

power generation sector, as their demand for electricity drops as a result of more energy-efficient production processes. This is what we call the energy efficiency effect of technology diffusion. In addition, a sector specialized in fabrication not only emits more CO2, but also uses

more energy than a sector specialized in marketing. Only by using input-output analysis we can account for the interdependent structure of production processes across countries (Dietzenbacher et al. 2013). The data on CO2 emissions will come from the environmental

accounts, which are part of the WIOD.

7 2. Literature Review

The scientific literature around climate change has firmly established the undeniable existence of it (Anderegg et al, 2014). Perhaps more importantly, Cook et al. (2013) review a total of 11,944 climate abstracts to confirm the leading role human activities take in the changing climate. The changing, and in particular warming, of the climate can be largely attributed to the emission of greenhouse gases into the atmosphere, notably CO2. The most important drivers of

CO2 emissions have been reported in the literature to be the growth of consumption (per capita)

and production (Raupach et al., 2007; Feng et al., 2015). In the field of climate research, many scholars have successfully attempted to quantify the potential for CO2 emission reduction. For

instance, Worrell et al. (2009) look at the industrial sector, which contributes about 37% of global greenhouse gas emissions and estimate the mitigation potential of industrial CO2

emissions by complementing previously developed estimates with their own, based on the assumption of best practice technologies being deployed by all plants in 2030. This study by Worrel et al. (2009) relates closely to our own as both calculate emission reduction potential through technology diffusion. The main difference lies in the scope: Worrell et al (2009) focuses on a specific sector whereas we will offer a worldwide estimate.

2.1 Global differences in CO2 intensities.

International data shows that differences in CO2 emission can be quite substantial for countries.

Figure 1 displays the CO2 emissions per dollar of output for eight countries, with the aim of

covering a wide range of development levels and size. Clear differences are visible. India, Russia, Bulgaria and China seem to emit far more than Germany, France, Sweden or the USA. One of the reasons this is the case might be found in the different energy mixes these countries are known to use. France, for instance, is known for its heavy reliance on nuclear energy, deriving 75% of its electricity from this source (World Nuclear Association (WNA), 2019), which is extremely clean in terms of CO2 emissions. Similarly, China’s economy being still

very industrial will contribute to its relatively high emissions. Other emission research shows that sizeable differences exist in the CO2 emissions across countries for a given sector. Kim &

Worrell (2002) show that large differences exist in CO2 intensities for the iron and steel industry

in the countries of Korea, Mexico, Brazil, China, India and the US. Related, Graus & Worrell (2011) calculate CO2 intensities in the most polluting sector worldwide, namely power

generation, and find large differences across a global panel of countries.

Figure 2 confirms the idea that differences in CO2 emissions exist across countries for a given

sector and graphs these differences for a rather CO2 intensive sector: Chemicals & Chemical

products sector. A similar picture to that of figure 1 emerges, including large differences between countries. Of the eight countries depicted in figure 2, it seems that Sweden is home to the best-practice technology for producing chemicals, as Sweden has the lowest emission coefficient (emissions per dollar of output). Evidently, many factors are at play here, and suggesting the differences in figure 2 are driven solely by the notion of functional specialization (i.e. the fact that China’s specializes heavily in fabrication in this sector, amongst others) would be premature and most likely incorrect. Nevertheless, empirical evidence exists for both differences in nation-wide CO2 emissions per dollar of output, and for sector-specific

8

Figure 1 (left): CO2 emissions per $ of output across countries.

Figure 2 (right): CO2 emissions per $ of output across countries for the chemical sector.

So what may drive the differences in figure 2? First, international differences between sectors may be driven by dissimilarities in product mixes across countries. In other words, differences in the total range of products they produce, within a sector. Even though the WIOD provides a disaggregated overview of the production structure of the world economy, ‘Chemicals & Chemical Products’ consists of many types of products. Although large economies will generally produce most of them, discrepancies in countries’ product mixes can drive differences in emissions per dollar of output within a sector as products vary in CO2 intensity. In addition,

this research considers CO2 coefficients in terms of gross output rather than value added, which

introduces another possible driver of differences in coefficients. The problem with CO2 to

output coefficients is that output can be equal for two companies (or sectors) while one may only assemble a product using inputs from elsewhere and sell it, whereas the other may produce all inputs itself, and then assemble and sell. The latter will obviously cause much more CO2

emissions than the former. Nonetheless, we use CO2 to output coefficients rather than CO2 to

value added coefficients because using the latter would introduce a host of complications in our further calculations. Dealing with these complications lies outside the scope of this research. 2.2 Technology and diffusion

Another important driving force of the differences in figure 2 can be the use of best-practice technologies, which is driven by technology diffusion. In this research, best-practice is defined as emitting the least CO2 per dollar of output. For instance, in the energy sector it is likely that

nuclear energy is the best-practice technology as our definition ignores other negative effects such as, in this case, the nuclear waste issue. Technology diffusion refers to the process by which climate technologies can spread across countries and sectors, facilitating a reduction in CO2 emissions. In the context of innovation, diffusion relates to the process of adopting a new

technology or replacing an older technology with a newer one, by firms and individuals in an economy (Hall, 2004). Only a few of the world’s richest countries account for the lion’s share of new technologies being developed, which means technology diffusion is one of the main determinants of worldwide technological change (Keller, 2004). However, the diffusion and adoption of new technologies is not always a smooth process, much less a costless one. Cohen

Note: The vertical axis is measured in kiloton of CO2 per million $ of output. Countries included: Bulgaria,

China, Germany, France, India, Russia, Sweden, and the USA. Source: World Input Output Database (WIOD), 2013.

0.0 0.5 1.0 1.5

BGR CHN DEU FRA IND RUS SWE USA CO₂emissions per dollar of output

Chemicals and Chemical Products

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

9

& Levinthal (1989) argue that firms in fact ‘pay’ to develop their ability, which they call ‘absorptive capacity’, to assimilate external information by investing in R&D. Related, Nelson & Winter (1982) state that only a small share of the required knowledge to implement a new technology can be acquired from blueprints or textbooks. The reason is that much of this knowledge is tacit of nature meaning it cannot be easily transferred by writing it down or even verbalizing it, and instead has to be acquired in a learning by doing fashion.

Some of the barriers that can slow or kill the process of technology diffusion altogether are discussed by Hall & Khan (2003). They start their paper with stating that, much contrary to the invention of new technology which is often a sudden jump, diffusion of technology typically is a slow and continuous process. In particular, a process that can be decelerated by the sunk nature of costs of adopting technology, uncertainty associated with the technology, a firm’s lacking of complementary skills and inputs, the absence of invention of new uses of the technology, and much more. In the context of global efforts countering climate change, technology diffusion is also an important tool, as it is needed to make widely available the breakthrough technologies that can help in reducing atmospheric concentrations of greenhouse gases (Hoffert et al., 2002). These breakthrough technologies can consist of a wide variety of options, some of which already exist. For instance, electrical means of transportations, particularly cars, are helping reduce the demand for fossil fuels today. Technology could arguably do much more yet by opening doors to emission reduction in the fields of renewable energy, the storage of energy, or food production.

However, for the reasons discussed earlier, it is not difficult to see how technology diffusion could be impeded and, by extension, not difficult to see how the lack of technology diffusion could put a drag on emission mitigation. Indeed, challenges exist in the diffusion of technology for climate change mitigating purposes. Jaffe, Newell & Stavins (2004) argue that, in the absence of adequate emission reduction policy, the combination of market failures associated with both environmental pollution, and the innovation and diffusion of new technology are likely to cause investment in development and diffusion of new green technologies to be lower than socially desirable. Similarly, González (2005) states that the private sector typically lacks incentive to develop or adopt clean technologies which leads to a lack of adoption and diffusion of green technologies. With current rates of diffusion of green technologies at sub-optimal levels, it seems reasonable to wonder what we could gain if diffusion was enhanced. Indeed, an OECD issued report on green growth states that the door is ‘wide open’ for technology diffusion of green technologies which could ‘help reduce environmental impacts at lower costs.’ (OECD, 2017).

2.3 Global Value Chains and functional specialization

10

Figure 3: Schematic representation of a Global Value Chain.

Figure 3 shows a schematic representation of a GVC. Assume this represents the GVC of a car, which spans three countries and where the car’s final stage of production takes place in country 3, the so-called country-of-completion. To facilitate its production, factor inputs are required from country 3 in the form of capital and labor, producing value added in country 3. Additionally, intermediate inputs, such as the engine or tires, are needed. Some of these intermediate inputs are produced domestically, but others are imported from country 2. As in country 3, factor inputs are required to produce the intermediate goods in country 2, generating value added. However, factor inputs in country two are not only used by industries producing the directly exported intermediate goods, the so-called first-tier suppliers in the production of the final product, but also by second-tier suppliers. These second-tier suppliers are industries in country 2 that produce intermediates, such as metal parts, that are used in the production of, for instance, tires, by first-tier suppliers. Finally, these second-tier suppliers are also located in country 1, where factor inputs are also required for the production of their goods, adding value to country 1’s economy. Figure 3 shows that different intermediates are sourced from different countries, depending on where they can be produced most cost effectively. This, in turn, depends on which country is specialized in which intermediate good required to produce the final good. In other words, different countries specialize in different sectors. It is therefore essential to take into account this sector structure of an economy because it has implications for its CO2 emissions, since a country specialized in mining will evidently emit more CO2 than a

country specialized in financial intermediation.

This ‘fragmentation’ of the production process across countries as Jones and Kierzkowski (1990) call it is a widespread phenomenon nowadays. The great advances in notably communication and coordination costs of the last decades have made it ever more profitable to

11

split the production process so that each stage is carried out at its lowest-cost location (Timmer et al., 2014). Other drivers of international fragmentation are trade liberalization and falling transport costs, both of which are also phenomena of the last decades. Indeed, consensus exists that production processes have become ever more fragmented into separate activities, with countries specializing to an increasing degree in specific stages of production (Los, Timmer & De Vries, 2015). This fragmentation implies more trade in intermediate inputs, which have to be transported across countries more intensively in order to finally reach the destination where the production process is finalized. Here too, scholars share the opinion reflecting a densification of the network of trade in intermediates (Los et al., 2015). However, this is not the full story of specialization in GVC’s.

In recent literature, Timmer et al. (2019) introduce the concept of functional specialization, which reflects the degree to which a sector in a country specializes in one of four functions: fabrication, R&D, management, and marketing. Functional specialization allows for the existence of dissimilarities in a given sector across countries by the extent to which they specialize in one of the four functions. To illustrate this, they provide an example of a manufacturing industry and show that China specializes in the fabrication function, whereas the USA’s main contribution in the sector lies in the management function. In the words of figure 3, it is likely the US imports products fabricated in China and adds to this capital and labor in the form of management activities. Yet another country may have been responsible for the R&D to produce the good, etc. The reason countries specialize in different functions in a given sector can be traced back to standard theories of international trade. Given the relative abundance of low-skilled labor in some developing countries, firms will outsource fabrication to these countries as this is most cost effective. Similarly, the relative abundance of high-skilled labor in many developed countries makes for the availability of more talent to conduct R&D or management activities, for instance, which causes firms to locate these activities to countries such as the US and Germany. Functional specialization is important to this research because it implies that within a given industry, countries may differ substantially, meaning not all countries can be compared to one another in that industry. We have showed earlier that China specializes heavily in fabrication in the chemical industry, whereas the US is more evenly specialized across functions. As fabrication activities typically cause more CO2 emissions than

the activities of the other functions, comparing China to the US would be incorrect because it is unreasonable to suggest that the best-practice technologies employed in the US could be implemented in China too.

2.4 Relevance and expectations

Even though the studies mentioned above, and many others, do manage to quantify the potential for CO2 emission reduction, all of them focus on a specific sector and/or country. However, we

12

fabrication specialized car manufacturing sectors with R&D specialized ones. Even though the US is much cleaner than Bulgaria in the chemical industry, comparing them would be unfair because the chemical industry in Bulgaria, which is heavily specialized in fabrication, cannot learn much from the technologies used in the US, which is more evenly specialized across functions. This is true simply because different functions use different technologies, which are characterized by their CO2 coefficient in this research. Evidently, the technology Bulgaria uses

(heavy machinery) cannot be improved by learning from the technology used in the US, a laptop used for designing an advertisement for instance.

13 3. Methodology

3.1 Data

The data used for the analyses comes from two sources. First, the World Input Output Database (WIOD) and one of its satellite accounts, namely the environmental accounts, will serve as source of the data used to calculate emissions per dollar of output. In particular the World Input-Output Tables (WIOT’s) of the WIOD provide the information about the production structure of the world economy, and the output each country generates. We will use the 2013 release of the WIOD for this research rather than the more recent 2016 release because the data by Timmer et al. (2019) features the same sectors as the 2013 release and can thus not be used in junction with the 2016 release, which uses a different set of sectors per country. In addition, the environmental accounts feature ‘Emissions to Air’ data, which supplies the CO2 emissions

data. Both the WIOT’s and the CO2 data cover the same range of countries and industries, an

extensive overview of both can be found in tables A1 an A2 of the appendix.In summary, they cover 40 countries consisting each of 35 industries, including all 27 EU members (as of January 1st, 2007), and 13 other large economies including the BRIC countries, accounting for over 85%

of world GDP. The next source of data that we will use is constructed by Timmer et al. (2019), which is an occupations database built from detailed survey and census data, conveniently disaggregated into industries in the same fashion as the WIOD. Per country-industry the total income earned is divided in income shares across four functions: marketing, management, R&D, and fabrication, such that the total for each country-industry equals one.

14 3.2 Input-Output Analysis

The aim of this study is to calculate the global potential for emission reduction if clean (i.e. best-practice) technologies could diffuse throughout sectors and countries. So how can we define clean? For this research we quantify clean through two channels, which are also the channels enabling emission reduction. First, technology diffusion may reduce emissions by the lowering of emission coefficients, which are defined as the CO2 emissions per dollar of output.

Second, we have to consider the indirect channel through which technology diffusion can reduce emissions, which consists of indirect effects. The most prominent example of such an indirect effect is the indirect energy effect or energy intensity effect, which relates to the potential decrease in energy use per dollar of output and associated decrease in CO2 emissions

that technology diffusion could bring about. For example, the use of more energy efficient machinery would reduce energy use in manufacturing sectors, reducing the need for electricity generation by the electricity sector, ultimately reducing emissions in the electricity sector. Many of these indirect effects exist. Another example of such an indirect effect is that technology diffusion could reduce the need for plastic packaging, reducing CO2 emissions in the plastics

sector.

Ideally, we would account for all the indirect effects that exist but on account of the extensiveness of this calculation it is considered out of scope for this research. Instead, we only focus on the indirect energy effect. The reason is that this is by far the most important indirect effect. At a global average of 3.37 kilotons of CO2 per million USD of output, the electricity

sector, ‘Electricity Gas and Water Supply’, has the highest emissions per dollar of output, far surpassing the sector with the second highest emission intensity, which is ‘Other Non-Metallic Mineral’ at 1.2 kilotons of CO2 per million USD of output (WIOD, 2013). Moreover, the energy

sector produces over 44% of worldwide CO2 emissions. Putting this in perspective, the second

most emitting sector is ‘Basic Metals and Fabricated Metal’ at 7.5%. The rest of the industries on average account for about 1.5% of global emissions. It stands to reason, thus, that if country-industries could lower their energy intensity much can be gained in terms of CO2 emissions.

15

Figure 4: Schematic overview of a World Input-Output Table (WIOT).

For this research, we are interested in CO2 emissions for which we need to add to the table in

figure 4 the CO2 emission data from the WIOD. In particular, we start from the following

vector: 𝐞 = ( 𝑒₁1 ⋮ 𝑒_𝑁𝑀 )

Where 𝐞 is the vector with CO2 emissions and each element 𝑒_𝑖𝑚 represents the emissions in

kilotons of CO2 per country 𝑚 and industry 𝑖. We have 41 countries, including the rest of the

world, with each 35 industries so 𝐞 is a column vector of dimension 1435x1. In order to account for indirect emissions made by industries to facilitate production in other country-industries we have to use the Leontief Inverse in our calculations. The Leontief Inverse matrix shows output changes in each country-industry due to a dollar’s increase in final demand in a given country-industry (Miller & Blair, 2009) and can be represented as follows:

𝐋 = (𝐈 − 𝐀)−𝟏

Where 𝐈 represents the identity matrix with one’s across its main diagonal and zero’s elsewhere, such that: 𝐈_𝐧 = 1 0 0 … 0 0 1 0 … 0 0 0 1 … 0 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 … 1

In this analysis, 𝐈 is a square matrix of the dimension 1435x1435 so that its dimensions agree with those of 𝐀, which is the matrix with all combinations of input coefficients across country-industries and looks as follows:

𝐀=

𝑎₁₁11 … 𝑎_1𝑁1𝑀

⋮ ⋱ ⋮

𝑎_𝑁1𝑀1 … 𝑎_𝑁𝑁𝑀𝑀 Source: Timmer et al. (2015)

(1)

(2)

(4) (3)

16

In this matrix, element 𝑎_𝑖𝑗𝑚𝑛 represents the input of industry 𝑖 in country 𝑚 needed to produce one dollar of output in industry 𝑗 in country 𝑛, and is calculated as 𝑎_𝑖𝑗𝑚𝑛 = 𝑧_𝑖𝑗𝑚𝑛 / 𝑥_𝑗𝑛. Here, 𝑥_𝑗𝑛 is a cell in vector 𝐱 in figure 4 and represents the output in country 𝑛 its industry 𝑗. 𝑧_𝑖𝑗𝑚𝑛 is a cell in matrix 𝐙 in figure 4 and represents the inputs of industry 𝑖 in country 𝑚 needed to produce the output in industry 𝑗 in country 𝑛. To put things more simply, each cell in A is found by dividing the appropriate cell in 𝐙 by the corresponding cell in 𝐱. This means that to create A, each cell in 𝐙 is divided by its column total.

The first channel of emission reduction we will consider operates through the emissions per dollar of output, or emission coefficients, given by:

𝐝 = ( 𝑑₁1

⋮ 𝑑_𝑁𝑀

)

Where 𝐝 is the vector with emission coefficients and each element 𝑑_𝑖𝑚 represents the emissions per dollar of output in country 𝑚 its industry 𝑖, which is calculated as 𝑑_𝑖𝑚 = 𝑒_𝑖𝑚 / 𝑥_𝑖𝑚.

The emissions of 𝐞 can be derived from the input-output model in the following fashion: 𝐞 = 𝐝̂𝐋𝐟

Here, 𝐝̂ is the diagonal matrix of 𝐝, which means it is a matrix with all values of 𝐝 on its main diagonal, and zeros elsewhere. Furthermore, 𝐟 is a vector that results from summing 𝐅 in figure 4 over countries, such that each value corresponds to the final demand per country-industry and 𝑓_𝑖𝑚 represents the total final demand in millions of dollars for the product of industry 𝑖 in country 𝑚. (6) is the basis of our analysis and thus deserves a little further discussion. First, throughout this entire study, 𝐟 is assumed to remain unchanged. As has been clarified before, this research focuses on the diffusion of best-practice technology as channel of emission reduction and assumes other factors influencing emissions, such as final demand, to remain constant. In addition, it should be clear that it is essential to use 𝐋. To see why, consider 𝐀, which captures the direct inputs that are required between country-industries. However, to produce these inputs, other inputs are needed, for which even more inputs are needed, etc. We have seen this in figure 3. To account for all these inputs, it is vital to use 𝐋, which has the convenient property of including all further upstream requirements. Using (6), we can calculate the emissions required to satisfy a given level of final demand, if emissions per dollar of output were lower.

(5)

17 3.3 Calculating emission reduction potential: conventional ways

For the first part of our analysis, called analysis 1, calculating the global potential emission reduction can be done simply by comparing countries’ emission coefficients. By summing both CO2 emissions and output over industries for each country to arrive at a national level of output

and CO2 emissions we can calculate the emission coefficient of a country 𝑖 by means of

𝑒𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠𝑖 / 𝑜𝑢𝑡𝑝𝑢𝑡𝑖. This allows us to see which country has the lowest emission coefficient,

and is thus the cleanest, i.e. home to the best-practice technology. Table 1 reports the five cleanest countries in the upper panel, and three large and much less clean countries in the lower panel.

Table 1: Country-wide emission coefficients, the five cleanest countries and the USA, China and India, 2007. COUNTRY EMISSION COEFFICIENT LUX 0.0227 SWE 0.0582 IRL 0.0585 FRA 0.0588 AUT 0.0757 USA 0.1820 CHN 0.5141 IND 0.5473

We can see that Luxemburg produces the least CO2 per dollar of output by rather much and that

Sweden, Ireland and France follow with very similar values. Conversely, China and India are much less clean, with emission coefficients over ten times the size of the cleaner countries, indicating much can be gained if technology diffused from the clean countries to China and India, especially given the enormous levels of output of these economies. Even the US, a very much developed country, has much to gain by the looks of its coefficient. We now have to decide which of the clean countries to use as benchmark, which means to decide which country the other countries will be assumed to converge to through technology diffusion. In later analyses, the same process will have to be repeated for each industry, requiring 35 picks. Therefore, let us create a framework for consistently picking the most appropriate country. Even though Luxemburg seems the most straightforward choice, this choice could introduce bias into our calculations because Luxemburg is a very small country and may not be a representative benchmark. One of the reasons this is likely true is because Luxemburg, due to its size, may very well lack certain industries altogether. Indeed, the WIOD reveals that Luxemburg produces virtually zero output in the sectors ‘Coke, Refined Petroleum and Nuclear Fuel’ and ‘Leather and Footwear’.

In other words, Luxemburg produces only in several industries rather than all, whereas larger countries are home to all industries featured in the WIOD. If the industries Luxemburg produces

Note: Countries included from top to bottom: Luxemburg, Sweden, Ireland, France, Austria, the United States of America, China and India. Emission coefficients are reported in kilotons CO2 per million USD of

18

in happen to be very clean in their production, this will introduce a bias in the country’s emission coefficient. The same argument exists on the industry-level: small countries will be home to smaller industries that produce only a few products produced by that industry, whereas the same industry in larger countries will produce all. Moreover, due to the measurement error present in the WIOD, smaller country-industries can show severely distorted emission coefficients. This is particularly relevant for small country-industries because measurement errors are relatively much larger at low levels of emissions and output. To solve these problems, we introduce an important criterion each country/country-industry has to fulfill in order to be qualified as the benchmark. This criterion states that the benchmark country (country-industry) has to account for at least 1% of worldwide output (in the relevant sector.)

For the sake of robustness, we have experimented with other threshold values for the criterion, such as 0.5% and 0.1%, and concluded that differences in outcomes are very marginal. Figure 5 serves as a fitting example of this. According to the 1% threshold, France would be the benchmark for analysis 1. Dropping the threshold would allow Sweden to become the benchmark country which, as can be seen, has an emission coefficient virtually equal to France, implying the results will hardly change. However, according to the 1% rule, the benchmark country for analysis 1 is France. This means each country-level emission coefficient is adapted to equal the emission coefficient of France. The emission coefficients of countries that produce even cleaner than France but do not comply with the 1% criterion are left unchanged. Simply multiplying each countries’ new emission coefficient with their output and summing the results yields the total emissions that would be made worldwide if all countries produced as little CO2

per dollar of output as France. For instance, given that China’s output is approximately 10,741 billion USD, its emissions using the French best-practice technologies would be 10,741 * 0.0588  631,565 kilotons of CO2, a reduction of over 88% compared to their actual emissions

of approximately 5,522,106 kilotons of CO2.

Table 2: Country emission coefficients in the chemical sector, the five cleanest countries and the USA, China and India, 2007.

COUNTRY EMISSION COEFFICIENT MLT 0.0027 IRL 0.0091 CYP 0.0174 LUX 0.0268 DNK 0.0312 USA 0.2336 CHN 0.4099 IND 0.5096

The second part of our analysis, analysis 2, moves from the country to the country-industry perspective. For each industry, the country with the lowest emission coefficient will be determined, while respecting the 1% criterion. For instance, table 2 shows the five cleanest

Note: Countries included from top to bottom: Malta, Ireland, Cyprus, Luxemburg, Denmark, the United States of America, China and India. Emission coefficients are reported in kilotons CO2 per million USD of

19

countries in the chemical industry, including some large countries with much higher coefficients. The cleanest country (using the best-practice technology) complying with the 1% criterion is Ireland, with 0.0091 kiloton CO2 per million USD of output. For all other countries,

the emission coefficient of the chemical industry will be changed to 0.0091, reflecting the diffusion of Ireland’s best-practice technology across the world. Table 2 shows that the emission coefficients of China and India are over 50-fold that of Ireland, signaling great reduction potential. Emission coefficients of countries with chemical industries that produce even cleaner than Ireland’s but that did not generate enough output to comply with the 1% criterion, in this case only Malta, are left unchanged. This process is repeated for each industry. Ultimately, most country-industries will be assigned a new emission coefficient. Multiplying this new emission coefficient with the output in the relevant country-industry yields the CO2 emissions that would be necessary to produce this output, using the best-practice technology. 3.4 Calculating emission reduction potential: Accounting for functional specialization. For the third part of our analysis, analysis 3, the goal is to incorporate the notion of functional specialization in the calculation of the global emission reduction potential. Let us first look at which countries are similar in terms of functions they specialize in. Figure 5 shows the functional specialization in the chemical industry for the same set of countries we have observed before. The black shares of the bars represent fabrication, blue marketing, grey R&D, and green management. Clear differences exist between countries. Russia, China and Bulgaria perform many fabrication activities, India focuses on management, and other countries show an evenly divided specialization pattern. As fabrication activities are likely associated with higher CO2

emissions per dollar of output due to, amongst others, higher energy use, it stands to reason that the specialization pattern of a country is important for how clean the country is (i.e. how low the emission coefficients of this country are). Indeed, inspecting the emission coefficients (in kilotons of CO2 per million USD) of the fabrication intensive countries in figure 5 yields 1.4

for Bulgaria, 0.41 for China, and 1.22 for Russia.

20

Figure 5: Functional specialization in the chemical sector, across eight countries.

Cluster analysis groups data based on information that describes the data and has many practical applications across different fields of research such as biology, geology and marketing. The aim of cluster analysis is to create groups that have minimal within-group variance and maximal between-group variance (Tan, 2018). In other words, clusters are ideally as homogenous as possible within, but as heterogeneous as possible across. The outcomes of the clustering process are essential to this research because they relate directly to its objective: quantifying the potential CO2 emission reduction worldwide. The reason is that comparisons between countries

in a given sector are made based on the clusters they are in, meaning different clustering results could lead to large deviations in the final answer to the research question of this study. It is therefore of great importance that the clustering make sense from an economic perspective and be done in a statistically sound way. For instance, it could not be the case that for the chemical sector China and the US are in the same group, given their widely different specialization patterns in this sector, as shown in figure 5.

Given some form of clustering needs to be applied to the data we let analysis 2 be the foundation of analysis 3, and create country-industry clusters. In other words, for each industry we group countries together based on their specialization pattern across the four functions. Cluster analysis encompasses many types of algorithms and methods, all of which have particular advantages and disadvantages which makes them suitable or unsuitable for a wide range of clustering purposes or types of data. One of the most often used clustering algorithms is the hierarchical clustering algorithm (Tan, 2018). The hierarchical clustering algorithm proceeds in a bottom-up way: each observation (country in a specific industry) forms its own cluster, and the two closest clusters are iteratively combined until only one cluster remains. It is then up to the researcher to evaluate the outcome and decide on the number of clusters he/she thinks is appropriate. The algorithm requires the specification of a distance metric, which is the metric it uses to calculate distances between clusters and so create/inform on clusters, and a method, which the algorithm uses to calculate the distance between two clusters using the distance metric.

Note: The vertical axis measures the share of labor income, summing up to one. Countries included: Bulgaria, China, Germany, France, India, Russia, Sweden, and the USA. Black = fabrication, blue = marketing, grey = R&D , green = management.

21

For this cluster analysis, we will use the Euclidean distance metric, which is the most commonly used and straightforward distance metric in cluster analysis, combined with Ward’s linkage method, which attempts to minimize the sum of squared distances of points from their cluster centers and is one of the most accepted linkage methods (Mooi et al, 2018). Euclidean distance is fairly straightforward conceptually as it can be thought of as a straight line between two points (Tan, 2018). In our case, the calculation of Euclidean distance between two countries in a given industry is based on the sum of squared differences between each of the functions. Essentially, this distance between countries can be thought of as the extent to which they are similar in terms of their specialization patterns. Countries that are similar in terms of specialization patterns (such as Germany and France in figure 5) will have a lower Euclidean distance between them than countries that differ greatly in specialization patterns (such as China and the US).

Ward’s method uses the Euclidean distance only as initial distance between clusters and then continuously merges the closest two based on minimization of total within-cluster variance. Within cluster variance is calculated by computing the sum of squared errors between all observation in a cluster and its mean. (Tan, 2018). Another issue in cluster analysis pertains to the decision on the number of clusters, for which some forms of aid exist. Additional to common sense judgment we will use a graphical representation of the cluster formation process called a dendrogram, which is plotted in figure 6.

Figure 6: Dendrogram of Chemical and Chemical Products industry using Ward’s method and Euclidean distance

Note: Countries are represented on the horizontal axis. The vertical axis represents the dissimilarity between clusters, as calculated by Ward’s method. The red line shows a 5-cluster solution, the green line an 8 cluster solution.

22

By looking at the dendrogram created by hierarchical clustering we can get a feel for the distribution of clusters. At the bottom of the graph, each country forms its own cluster and merging takes place until all countries are in the same cluster, which happens at the top of the graph. The vertical axis measures dissimilarity, meaning the lower on the y-axis countries/clusters are merged, the higher their similarity. This also implies that the larger the leap upwards toward the next fusion of clusters is (which is indicated by a horizontal line merging two clusters), the more is sacrificed in terms of dissimilarity between cluster, and the more dissimilar clusters being merged are. Drawing a horizontal line through the graph cuts the dendrogram and decides the number of clusters, indicated by the number of vertical lines crossed by the cutline. Ideally, the dendrogram is cut as close to the bottom as possible, while achieving a reasonable number of clusters.

Looking at the dendrogram, it seems the first viable solution is using 5 clusters (red line), since using fewer would imply moving much further up the dendrogram, sacrificing a lot of dissimilarity. The next would be 8 clusters (green line), as this allows us to move further down the dendrogram, achieving lower dissimilarity between the clusters. Solutions further down the dendrogram imply a high number of clusters, and many clusters containing only a few countries. The clustering procedure has to be done for each of the 35 industries. As discussing results of all industries is a very time-consuming process we will limit ourselves to discussing the cluster outcome of the chemical sector below. Table 3 presents cluster results of this sector for both a 5 and an 8-cluster solution, as well as the emission coefficients of each country in the chemical sector and the cluster-average emission coefficient based on an 8-cluster solution. Each value in the columns pertaining to the functions (MGT, R&D, MAR & FAB) represents the share of labor income in that function in a country. Shaded areas represent the division of clusters based on the 8-cluster solution.

Table 3: Cluster results Chemicals and Chemical Products.

Country MGT R&D MAR FAB WARD 5 WARD 8 EM COEF AVG EM COEF

23 MEX 0.17 0.17 0.21 0.44 5 4 0.213 JPN 0.03 0.18 0.28 0.50 5 4 0.213 TUR 0.14 0.15 0.22 0.49 5 4 0.287 ROU 0.08 0.25 0.20 0.46 5 4 0.392 CHN 0.07 0.08 0.24 0.62 5 4 0.410 POL 0.11 0.23 0.25 0.42 5 4 0.584 LTU 0.12 0.25 0.24 0.39 5 4 1.198 DNK 0.20 0.38 0.26 0.17 3 5 0.031 0.186 SVN 0.09 0.41 0.24 0.27 3 5 0.060 SWE 0.15 0.38 0.19 0.28 3 5 0.084 FRA 0.15 0.35 0.23 0.27 3 5 0.129 DEU 0.17 0.31 0.24 0.28 3 5 0.143 ITA 0.14 0.29 0.29 0.29 3 5 0.147 BEL 0.24 0.31 0.24 0.21 3 5 0.202 CZE 0.16 0.32 0.22 0.30 3 5 0.689 MLT 0.26 0.23 0.13 0.38 3 6 0.003 0.345 IRL 0.21 0.34 0.15 0.30 3 6 0.009 FIN 0.20 0.34 0.14 0.31 3 6 0.148 HUN 0.16 0.30 0.17 0.36 3 6 0.386 EST 0.07 0.38 0.17 0.38 3 6 0.695 SVK 0.10 0.32 0.18 0.40 3 6 0.829 GBR 0.40 0.26 0.14 0.19 1 7 0.126 0.318 IND 0.52 0.16 0.21 0.10 1 7 0.510 LUX 0.00 0.22 0.78 0.00 4 8 0.027 0.027

Looking at table 3, the 8-cluster solution reveals that the countries in cluster 1 show a somewhat evenly division of labor income across functions. Cluster 2 appears similar but upon closer inspection has higher marketing and lower management shares. Cluster 3 is rather small and is characterized by a very high fabrication and low marketing shares. Cluster 4 is much larger and also features high fabrication shares, but differs from 3 with its higher marketing shares. Cluster 5 mainly specializes in R&D activities. Cluster 6 also features relatively high R&D but combines this with high fabrication shares. Cluster 7 is comprised of India and Great Britain based on their extraordinarily high management shares. Ultimately, Luxemburg is placed in an individual cluster meaning it is seen as an outlier. Its values on the four function indeed confirm so, especially its extremely high marketing and low fabrication share. Generally, looking at the results, the cluster algorithm has yielded fairly reasonable results. In terms of emission coefficients, clear differences within the clusters exist which suggests that taking into account functional specialization will likely lead to a drastic decrease in potential emission reduction. Between clusters, differences exist too. Cluster 3 has by far the highest average emission coefficient and 5, featuring the developed countries, the lowest (not considering the outlier Luxemburg). Many of the other clusters are somewhat similar in their average emission coefficients.

Note: Ward 5 = 5 cluster solution. Ward 8 = 8 cluster solution. EM COEF = emission coefficients in kiloton CO2 per million USD. AVG EM COEF = average emission coefficient for the cluster. The four business function represent the percentage of labor income earned in that function in each country.

24

What do these clustering results imply for our analyses? Consider the example of the chemical industry in table 3. The cleanest country complying with the 1% criterion is Ireland with an emission coefficient of 0.009 kiloton CO2 per million USD of output. Australia, for example,

has a much higher coefficient of 0.398. If this was analysis 2, Australia’s coefficient would be replaced with Ireland’s. Given Australia’s output in the chemical sector of 22.9 billion USD, its new emissions would be 22.9 * 0.009  206 kiloton CO2, a reduction of approximately 97%

compared to its previous emissions of 9120 kiloton. However, the clustering procedure put Australia and Ireland in different clusters. In other words, based on their specialization patterns, we cannot compare Ireland and Australia. Instead, based on the 8-cluster solution, we have to compare Australia to the Netherlands, which is the cleanest country in Australia’s cluster complying with the 1% criterion. This is what happens in analysis 3. Replacing Australia’s coefficient with the Dutch coefficient of 0.235 yields a new Australian emission total of 22.9 * 0.235  5400 kiloton CO2, a reduction of approximately 40%. We see that accounting for

functional specialization makes a large difference in this example. Table 4 shows the differences between the analyses for the entire chemical industry, showing large differences persist for the chemical industry as a whole.

Table 4: Results of analysis 2 and 3 for the chemical sector in megaton (1000 kiloton) of CO2.

Actual (2007) Analysis 2 Analysis 3 5 clusters Analysis 3 8 clusters CO2 emissions 1000 30 370 570 Potential reduction - 97% 63% 44%

In 2007, CO2 emissions in the chemical sector were 1000 megaton (mt). Assuming each country

could produce as clean as the cleanest country in the chemical sector (Ireland) implies a reduction of as much as 97% to 30mt of CO2. Taking into account the degree of specialization

across functions using a 5-cluster solution decreases this reduction to 63%. Allowing for the existence of 8 clusters decreases the reduction even further, to 44%. This is a sensible result because in analysis 2, no clustering is applied, effectively meaning all countries are placed in the same cluster. Increasing the number of clusters decreases the probability of large differences existing within clusters and, by extension, the reduction potential. In the extreme case where each country forms its own cluster no reduction would be possible at all, as each country is then the cleanest in its cluster. But how to decide which number of clusters is optimal?

Table 3 showed that, as we could logically expect, the difference between a 5- and an 8-cluster solution was that some more clusters had been merged. Characterization of the 8-cluster solution is much more difficult just by glancing at the data, because the differences within many of the clusters increase. Table 4 shows that at least for the chemical sector, the number of clusters is rather important for the outcome of our analyses. Further analysis reveals that the chemical sector appears to be an outlier. Even though differences do exist between reduction potentials of different cluster solutions, almost none are as large as that of the chemical sector. In the determination of the number of clusters, an important note is that it is typically better to choose more rather than fewer clusters when in doubt, as it is highly undesirable to force observations into clusters they do not actually belong to (Tan, 2018). Therefore, the chemical

25

sector will be divided in 8 clusters. For the remainder of the sectors, the number of clusters is also decided based on the dendrogram. Most industries are split into 6, 7 or 8 clusters. After formation of the clusters, per industry and cluster the cleanest country is determined using the 1% criterion. Similar to analyses 1 and 2, this value is then assigned to each country in the same cluster, for each industry. Again, emission coefficients of countries that are cleaner than the benchmark countries but do not comply with the 1% rule are left unchanged. Ultimately, the new emission can be calculated by multiplying the new emission coefficient of each country-industry with its respective output.

3.5 Including indirect energy efficiency effects

The aim of the last part of our analysis is to incorporate the indirect effects of energy efficiency into the analysis. We call this analysis 4. What would the global reduction potential look like if, next to lowering emission coefficients, each country-industry could produce as energy efficient as the most energy efficient country-industry in the relevant cluster? Energy efficiency can be measured by means of the energy coefficient, which can be found in 𝐀. Low energy coefficients (e.g. low energy intensity) signal high energy efficiency. In 𝐀, each element of a row corresponding to the industry ‘Electricity, Gas and Water Supply’ reflects the monetary input of the energy sector required to produce one dollar of output in each of the 1435 country-industries. The higher this value, the higher the energy intensity of the country-industry. As we have 41 countries, we also have 41 rows pertaining to the energy sector.

Summing the coefficients of these 41 energy sectors per country-industry yields the total monetary input required of the global energy sector to produce one dollar of output in each country-industry. This is equal to taking the column sum of all energy sectors in figure 4, for each country-industry. For instance, for the production of 1 million USD worth of chemical products in the US, energy is required in the US but also some from Canada, Mexico, and other countries. This is similar to what we have observed in figure 4, namely that each country-industry requires inputs from a variety of other country industries. The difference is that this time we are only looking at the energy industries of each country. Evidently, the vast majority of energy requirement will be sourced domestically. Indeed, calculating country’s energy sector share in the total energy requirement of a country-industry from 𝐀 reveals that most of the country-industries use domestic supply for over 90% of their energy needs.

26

will be divided across country A and B proportionally, meaning A’s inputs are now 0.045 and B’s are 0.005. This leads to a new column of energy coefficients for country-industry X, in this case with only countries A and B. In reality of course, energy inputs will be required from more countries and this exercise will be repeated for all 1435 country-industries. The clusters are equal to those determined in analysis 3, as they are based on functional specialization. Energy coefficients of country-industries that are cleaner than the benchmark in their cluster and do not comply with the 1% criterion, which in this case states that they have to account for at least 1% of output in the energy sector, are left unchanged.

The last step is to replace the values in 𝐀 pertaining to countries’ energy sectors with their corresponding newly calculated and proportionally divided energy coefficients which leads to a new matrix 𝐀̃. Using 𝐀̃, we can recalculate 𝐋:

𝐋

̃ = (𝐈 − 𝐀̃)−𝟏

Ultimately, a new vector with emissions can be calculated as follows: 𝐞̃ = 𝐝̃̂𝐋̃ 𝐟

Here, 𝐞̃ represents the vector of emissions required in each country-industry in order to satisfy global final demand, assuming each country-industry could lower its emission coefficient and energy intensity to that of the cleanest country-industry in the cluster. Furthermore,𝐝̃̂ represents the vector with emission coefficients used in analysis 3, with each country-industry’s emission coefficient reduced to the lowest in its cluster.

(7)

27

4. Results

Having previewed the results for the chemical sector above, we now present the full results of analysis 2, 3 and 4 in table 5 for a selection of CO2 intensive sectors.

Table 5: Results of analyses 2, 3 and 4 for a selection of sectors.

Sector Actual (2007) Analysis 2 Analysis 3 Analysis 4

CHEMICALS Emissions 1000 30 570 560 Potential reduction - 97% 44% 45% MINING Emissions 970 400 700 650 Potential reduction - 59% 27% 33% REFINED PETROLEUM Emissions 930 260 500 487 Potential reduction - 70% 46% 48%

In the shaded rows, table 5 shows the new emissions required to satisfy final demand for each analysis in megaton. The unshaded rows display the associated potential reduction (%) in emissions relative to the actual value of CO2 emissions in 2007. An example interpretation of

the chemical sector is as follows: assuming that each country in the chemical sector could produce as clean as the cleanest country in the sector, an emission reduction of 97% could be achieved. Accounting for functional specialization and assuming each country could produce as clean as the cleanest country in the cluster, emissions could be reduced by only 44%. Lastly, incorporating the indirect energy efficiency effect, a slightly higher emission reduction potential of 45% could be achieved. We had already seen that, given unchanged output and final demand, for chemicals a very large reduction potential (97%) existed according to analysis 2, and that analysis 3 suggested a sharp decrease in this potential.

In the last column, we can now see that accounting for the energy efficiency effect, allows for a slightly larger potential. Generally speaking, the reduction potential follows a similar trend in each sector. It reduces going from analysis 2 to 3 and picks slightly back up going to analysis 4. However, large differences exist between sectors. Whereas analysis 2 created a large reduction potential in the chemical sector, it seems this potential is far smaller in mining. This signals that for the chemical sector a very clean country exists, whereas the mining sector’s cleanest country is relatively speaking much less clean. A last noteworthy result to discuss is the large difference between analysis 3 and 4 in the mining sector. It seems mining relies on the electricity generation sector much more heavily than the other sectors in table 5. A potential explanation may be found in the heavy reliance of the mining sector on heavy machinery for the mining and transportation process of ore within quarries. It is clear that differences in outcomes of the analyses exist between industries suggesting the existence of heterogeneity in the importance of functional specialization across industries. This begs the question: what about the global picture?

28

Table 6: Global CO2 emission reduction potential.

Actual (2007) Analysis 1 Analysis 2 Analysis 3 Analysis 4

CO2

Emissions 25.3 6.4 3.8 12.6 10.5

Potential

reduction - 75% 85% 50% 59%

Table 6 shows the global potential for CO2 emission reduction according to all four analyses.

Following our analysis of table 5, table 6 shows that generally a large difference of 35 percentage points exists between the reduction potential of analysis 2 and 3. In other words, it appears accounting for functional specialization greatly reduces the global emission reduction potential. Furthermore, the indirect effects of energy efficiency raise the reduction potential by 9 percentage points to 59%, indicating the indirect effects of energy efficiency are substantial and should not be ignored. More specifically, accounting for the functional specialization pattern of countries and indirect energy effects yields a global reduction potential of almost 15 gigaton, or 59%. Lastly, the reduction found using analysis 2 is 10 percentage points higher than the reduction found in analysis 1. In other words, taking into account the fact that different industries have different CO2 intensities yields a higher potential reduction than using the

intuitively more naïve country-wide approach. This might seem like a strange result, because if all countries are assumed to be equal in terms of the sector structure (analysis 1), sector differences in CO2 intensities would be ignored which would paint an incorrect picture of

reduction potential. Analysis 2 alleviated this problem by comparing each country-industry individually to its peers in the same industry, which is evidently more realistic and would yield a smaller reduction potential. As output has remained unchanged throughout the analyses, the only explanation for the difference is that apparently, analysis 2 yielded lower emission coefficients than analysis 1.

A possible explanation for this can be found in the fact that analysis 1, given its country-level perspective, implicitly commands every sector in each country to copy the technology of the cleanest country, which was France. In other words, every sector in each country was effectively assigned France’s emission coefficient of 0.059. However, even though France was the cleanest country when industries were aggregated, it most likely will not be the cleanest country in each sector. Therefore, when in analysis 2 we adopted a sector perspective, cleaner technologies became available from sectors in other countries possibly explaining the lower emission coefficients yielded by analysis 2. Now that we have seen the picture at the sectoral and global level it might be interesting to see if large difference between regions exist in terms of our analyses. Table 7 shows the results of analysis 1, 2, 3 and 4 for the US, China and the European Union, three large economic regions in the world.

29

Table 7: Emission reduction potential in the US, China and the EU.

Country Actual (2007) Analysis 1 Analysis 2 Analysis 3 Analysis 4

US Emissions 4.7 1.5 0.57 3.5 2.9 Potential reduction - 68% 88% 25% 37% China Emissions 5.5 0.6 0.58 2.4 1.7 Potential reduction - 89% 90% 57% 69% EU Emissions 3.5 1.9 1.0 2.0 1.8 Potential reduction - 45% 72% 41% 50%

Generally, a similar picture to that of table 6 emerges. However, some things stand out. First, the total reduction potential according to analysis 4 is by far the highest for China, signaling they have most to gain from the diffusion of clean technologies and that by extension, they are currently using the least clean technologies of the three regions. Conversely, the US has the lowest reduction potential under analysis 4, meaning the country is already using much of the best-practice technologies available. Second, the difference in potential reduction once accounting for functional specialization (i.e. between analysis 2 and 3) is highest for the US, indicating that the US generally shows a high degree of functional specialization throughout its sectors.

At the beginning of this paper we formulated some expectations regarding the outcomes of the analyses. The most important one, which directly relates to the novelty of this paper, stated that accounting for functional specialization would reduce the global potential for emission reduction. Table 6 showed that this suspicion finds strong support and that indeed, a decrease of 26 percentage points (from 85% to 59%) was found once accounting for functional specialization. Figures 5 and 7 confirmed that also on the sectoral and country level, these differences persist. Furthermore, we stated to expect an increase in the reduction potential once taking into account the indirect effects of energy efficiency. Indeed, tables 5, 6 and 7 show that also for this expectation strong empirical support exists, as on both the global, sectoral, and country level clear positive difference between analysis 3 and 4 exist. In particular, on the global level energy, efficiency gains allow for a 9 percentage point increase in emission reduction.

Note: CO2 emissions rounded and displayed in gigaton (gt). EU comprised of all member states as of January