A comparison of methods to measure emissions
embodied in exports using input-output analysis
Keywords: Emissions embodied in trade – Input-output analysis – Producer vs. consumer perspective
MSc International Economics & Business Sjors van Uden (S2726815)
s.c.van.uden@student.rug.nl
Supervisor : Prof. Dr. H. W. A. Dietzenbacher Date: 18-06-2019 Co-Assessor : K. M. Wacker
Abstract
ii
Acknowledgment
1
Table of Contents
Acknowledgment ii
1. Introduction ... 2
2. Literature review ... 6
2.1 International trade and Global Value Chains ... 6
2.2 Production- versus consumption-based accounting ... 7
2.3 Relevant measures of EEX ... 7
3. Data & Methodology ... 9
3.1 Data source ... 9
3.2 Introduction to standard Input-Output Analysis ... 10
3.3 Overview measures of EEX ... 13
4. Results ... 21
5. Discussion and conclusion ... 28
References ... 30
2
1. Introduction
The unprecedented growth of globalization and international trade over the last decades has led to accelerated economic activity. Continued economic growth has been accompanied by an increase in environmental pressures, where international trade enables these pressures being shifted between countries (Munksgaard & Pedersen, 2001; Wiedmann, 2016). In 1997 the Kyoto protocol was adopted by a group of developed countries, aiming to reduce the level of CO2 emissions1. An
important element of the Kyoto protocol is that it poses restrictions on the amount of emissions produced within a country’s administrative area. Soon it became clear that a way to circumvent this restriction is through international trade. Importing relatively ‘dirty’ products does not contribute to higher emissions within your country, and thus the Kyoto protocol does not pose any restriction on it. This raised attention to quantify the amount of carbon emissions being traded around globally (Wiedmann, 2016). The idea that trade could enable (developed) countries to comply to environmental standards has led to a different perspective on this matter. Besides looking at what amount of emissions are generated within the territory of a nation, how much a country ‘consumes’ is also of concern (Gallego & Lenzen, 2005). These two types of perspective are the starting point on how to measure emissions in trade. Multiple methods exist in the academic literature, which serve different purposes. A distinction should be made between the exports of emissions (XEE), and the emissions in exports (EEX). Calculating XEE is straight forward, namely the emissions generated during the production process of a certain good, are ultimately assigned to the country that consumes the good. With EEX, the emissions are not assigned to the country which consumes the good, but to the country which produces the good. In contrast to calculating XEE, there are multiple methods for calculating EEX. It depends on the question what method is the appropriate one.
The distinction between the two methods becomes clear when taken to a practical example. Imagine a border official working alongside the Dutch-German border, he sees products crossing the border every day, and wonders how much emissions are embodied in these products. Later that day, he sits in a restaurant eating an Argentinian steak. Again, he wonders how much emissions are embodied in his delicious piece of beef. The consumption of the steak has led to emissions being generated in multiple countries, obviously in Argentina itself, but also indirectly in other countries (e.g. via fodder, transport, conservation, etc.). The carbon footprint of this steak measures all global emissions embodied in its consumption (XEE), whereas the products that cross the border only have territorial emissions embodied (EEX). Both measures are correct, but they answer different questions; the XEE can be used for questions such as ‘’Who emits how much for whom?’’, while EEX is more appropriate for a question like ‘’How much emissions are traded globally?’’ In a specific scenario, it could be the case that the EEX between two countries is zero, while the XEE has a positive value. When country A and country B do not trade with each other, their EEX
3 would be zero. However, country A and B both do trade with country C, it could be the case that exports in intermediates from country A to country C, end up in the final demand of country B. In this respect their XEE is not equal to zero, as the emissions which flow via a third country (C) do not go directly to country B, but arise in country A’s XEE to country B indirectly. With XEE the emissions from intermediate exports are found in the trade flow between the exporting country and the country of final use, whereas with EEX the emissions are found in the exports of the producing country.
The difference between the outcome of XEE and EEX has become more prevalent over time, due to the trend of increasing international fragmentation. Different stages of production are often performed in different countries, where intermediate inputs cross borders multiple times before being assembled into a final good (Koopman, Wang & Wei, 2014). The process of fragmentation has become more widespread in recent decades due to the plummeting costs of communication, coordination, and transportation. The decrease in transportation and transaction costs have made it more profitable to slice up the production process, in which stages are allocated according to optimize cost reduction. This has changed the nature of international trade fundamentally, from trade in consumer goods towards more trade in tasks and activities (Timmer, Erumban, Los, Stehrer and De Vries, 2014). It is interesting to note that no distinction existed in measuring EEX and XEE, until the introduction of Global Multiple Region Input-Output (GMRIO) tables. Early work on mapping the structure of trade, like that of Hummels, Ishii, and Yi (2001) mainly sourced data from National Input-Output (NIO) tables, which do not convey indirect trade via third countries. NIO accounts provide a rich description of value chain linkages across industries within a country, but they stop at the border. They lack information on how exports are used abroad. In addition, they do not contain information on how imports are produced. NIO are limited in mapping global value chains (GVCs) across multiple countries (Johnson, 2018). With the use of NIO tables, EEX and XEE are measured equal if it is assumed that exports do not return home again, however with using GMRIO tables this is not necessarily the case. In a GMRIO table there is described from whom each industry sources inputs from any other country in the world, and to whom each industry’s output is sold, domestically or abroad, either as final good or as intermediate. GMRIO models have become very useful in analyzing complex networks of GVCs, because they provide information regarding a sector’s relationships with other sectors, across countries (Cadarso, Monsalve & Arce, 2018). The combination of increasingly complex structures of international trade, and sophisticated GVCs, raises the importance to measure EEX and XEE separately.
4 Consider measuring EEX in a complex network of GVCs, take for example the automobile industry. The automobile industry is characterized by multiple production activities distributed along a GVC across many different countries. To illustrate, iron ores are mined in Brazil, the iron ores are exported to the United States to be used in the production of steel plates. These steel plates are exported to Mexico, where they are cut into car parts. These car parts are exported back to Brazil, to be used in the assembly of cars. These cars are then moved within Brazil, from the factory to a car dealer. This Brazilian car dealer then sells its cars to domestic customers, as well as foreign consumers, for example to the Netherlands2. Measuring EEX in such a network of GVCs is complicated, it depends on the question what measure should be used. What emissions should be included in measuring EEX for this particular example? Do you solely include the emissions which are added during the last production step? Or should there only be included the emissions which are domestically generated (in this case the emissions generated within the territory of Brazil)? Or do you measure the total emissions embodied in the entire production process, making no distinction between foreign and domestic content? There is no one best answer to these questions. It depends on the question what type of measure of EEX is appropriate, since every measure serves a different purpose.
The fact that a growing amount of international trade nowadays is in intermediate inputs, and consists of intermediates being exported and further downstream being re-imported even complicates matters for measuring EEX. Due to these circumstances, the double-counting of emissions occurs. Several researchers have examined the complexity of global value chains and international trade, and have highlighted the potential double-counting in trade statistics (Koopman et al., 2014; Meng, Peters, Wang & Li, 2018). Due to GVCs and changing trade patterns it has become challenging to accurately estimate a country’s EEX. Therefore a theoretical outline on how to account for the double-counting of emissions when estimating EEX provides essential insights and will be investigated.
There is no clear-cut consensus on the calculation of the EEX, and it is most likely that previous efforts include, at least to some extent double counted emissions. This thesis will add to existing literature by adding a well-defined comparison of multiple methods of EEX. In this comparison every measure of EEX is defined using the same variables and using a clear, consistent design. Such a comparison is novel since it is non-existent in the current academic literature, to the best of author’s knowledge. Since every method serves a different purpose, this thesis will discuss what measure to use in which situation. This thesis will answer the following research question:
What measures of EEX are proposed in the literature, and how do they differ – analytically and numerically?
How they differ analytically will be shown by providing a theoretical outline for each method, where the implications of each method will be discussed and compared. By consistently using data
5 from the World Input Output Database (WIOD) for the calculations, it can be stated that all the deviation in the amount of CO2 emissions embodied in trade is purely caused by the chosen
methodology, and is not affected by differing data from multiple databases. In accordance the numerical differences of each individual method will be set out.
6
2. Literature review
The overview of literature on the topic of emissions embodied in trade involves three different sections. First, section 2.1 outlines the current day setting of international trade and its implications. Thereafter, section 2.2 briefly discusses and compares two commonly used accounting principles related to measuring emissions embodied in trade. Section 2.3 concludes with a brief review of relevant methods to estimate EEX, as proposed in the literature.
2.1 International trade and Global Value Chains
The increase in international trade, and as a result, the growing volumes of emissions embodied in exports have been quantified by multiple studies over the past decades (Peters et al. 2011; Sato, 2014; Wiedmann, 2016; Xu & Dietzenbacher 2014). In an extensive survey on the literature on carbon emissions embodied in trade, Sato (2014) finds that the emissions embodied in trade entail large and growing volumes. For example in 2004, 4-6 Gt of CO2, which is 15-25 % of the total
annual emissions, were embodied in international trade. According to Peters, Andrew and Lennox (2011) in 2008, emissions in trade already increased to 7.8 Gt, which was equal to 28% of the total annual emissions at that time. Wiedmann (2016) compares multiple studies on impacts embodied in trade and finds that the studies have varying results. The emissions embodied in trade as a share of the total annual emissions vary from 22% to 33%. It is clear that an increasing amount of international trade, leads to growing volumes of emissions embodied in trade, but it seems that it is hard to quantify by how much exactly. To estimate the total emissions embodied in trade, only the emissions in exports need to be considered. It is not needed to calculate the emissions in imports as well, by definition exports equal imports in any bilateral relation.
The above estimations are in line with the ongoing globalization and the growing evidence that GVCs have become more prevalent in the current day economy (De Backer & Miroudot, 2014). In addition to the dramatic increase in international trade, trade has become more complex and fragmented through different countries. The production processes of many products are sliced up into small steps, which are linked through global supply chain networks. The classic example of a GVC is the production of an Apple smartphone. The iPhone is designed in the United States, it has its screen and processor produced in Japan and South Korea, it is assembled in China, and thereafter shipped to all over the world to be sold in one of Apple’s stores (Meng et al., 2018).
7
2.2 Production- versus consumption-based accounting
The distinction between EEX and XEE is built upon the difference between two accounting principles: a production-based accounting of emissions (PBA) and a consumption-based accounting of emissions (CBA). In the literature the two main accounting principles, CBA and PBA, are the common thread, therefore their differences will be elaborated on further. The two approaches differ in the allocation of responsibilities, PBA allocates the emissions to the territory of a nation where they production takes place, while CBA allocates the emissions to the territory of a nation where the final products are consumed. CBA extends on the production-based emissions, by removing the export of emissions, and by adding the import of emissions (Peters, 2008; Su, Huang, Ang & Zhou, 2010). International trade thus links the consumption in one country to emissions generated in other countries, as a consequence the emissions produced in one country are not the same as the emissions that are necessary for its consumption. This discrepancy is exactly portraited by the difference between the PBA- and the CBA principle (Serrano & Dietzenbacher, 2010).
From a policy perspective, EEX (measured according to PBA) is more informative and relevant than XEE (measured according to CBA). XEE is based on calculations, which extend beyond the jurisdiction of a country, where responsibility should be taken for emissions generated in a foreign country. Policymakers in general only have mandate within the borders of their own country, which makes it difficult to base policies on estimates which are based on XEE (Peters, 2008). If all nations in the world together would cooperate, this would be less of an issue, however the current system of nations makes such a cooperation highly unlikely (Rocchi, Serrano, Roca, & Arto, 2018). For estimating emissions embodied in trade, EEX takes a more down to earth ‘border perspective’; it measures the emissions that are embodied in the products which cross a country’s border, regardless of where they are consumed. Environmental regulations nowadays are based on territorial emissions, for example the Paris Agreement, an international agreement on reducing global CO2 emissions and keeping the warming of the earth well below 2 °C, poses a restriction on
the amount of greenhouse gasses allowed to be generated within a country (UNFCC, 2015). This study primarily focuses on EEX, XEE can still provide useful as a benchmark for the different measures of EEX. The next section briefly introduces several measures of EEX which are proposed in the literature.
2.3 Relevant measures of EEX
8 The varying results of emissions embodied in trade to be found in the academic literature (Peters et al., 2011; Sato, 2014; Wiedmann, 2016) are caused by several factors. Different databases are used and different time frames are considered. Also not all studies as in Wiedmann (2016) estimate EEX, some do estimate XEE, making it hard to compare these studies. This thesis will only make use of one and the same database and time frame, to obtain consistent and comparable results for each measure of EEX.
9
3. Data & Methodology
In section 2 relevant literature has been reviewed, and four measures of EEX are identified. This section provides an extensive comparison of these four measures. First, section 3.1 provides a brief overview on several GMRIO databases, and more specifically on the WIOD (Timmer, Dietzenbacher, Los, Stehrer & de Vries, 2015); the database used to perform the analysis. Second, an introduction to input-output (IO) analysis is given in section 3.2, followed by the theoretical and mathematical details of all four measures of EEX in section 3.3.
3.1 Data source
In recent years several GMRIO databases have been developed, and have become publicly available. In a GMRIO model, NIO tables, representing financial transactions between sectors within a country and trade flows between countries, showing the value of exports and imports by country and industry, are linked together in one coherent accounting framework (Wiedmann, Wilting, Lenzen, Lutter & Palm, 2011). Tukker & Dietzenbacher (2013) provide an overview on five widely used GMRIO databases. These five databases are: the EORA (Lenzen, Moran, Kanemoto & Geschke, 2013), the GTAP (Aguiar, Narayanan & McDougall, 2016), the EXIOBASE (Tukker et al., 2009), the OECD inter-country IO tables (Yamano, 2016) and the WIOD (Dietzenbacher, Los, Stehrer, Timmer & de Vries, 2013; Timmer et al., 2015). Every database has its own strength and weaknesses. All of them cover different years, countries and vary in their environmental and/or socio-economic accounts (Xu, forthcoming). The use of different databases results in varying outcomes since each database uses different environmental accounts, energy data, system boundaries, assumptions and definitions (Andrew et al., 2009). The database considered in this thesis is the WIOD, which is a publicly available GMRIO database. By only using the WIOD, varying outcomes can only be attributed to each measure’s methodology. The WIOD contains extensive data for 40 countries and the ‘Rest of the World’ (RoW), for the period 1995-2011. The RoW region is used as a summative region to include all other countries which are not in the dataset itself. In Table 1 below all the countries included in WIOD are listed:
10 The WIOD covers annual time-series of world IO tables and factor requirements (Timmer et al., 2015). The WIOD is a suitable database for this research since it includes a set of environmental accounts, where ‘Emissions to air’ is particularly useful. The dataset involves emissions to air by sector and pollutant, where carbon dioxide will be of use in this thesis. Data on these environmental accounts are only covered for the period 1995-2009, so this will be the time-span covered in this thesis. This data allows for making calculations regarding EEX. Also, the WIOD provides time series of IO tables in current and previous year’s prices. This is especially useful when one would like to do a structural decomposition analysis, where it is needed to single out the price effect (Xu, forthcoming). The environmental accounts are included in all other IO databases, whereas IO tables in both current and previous year’s prices make the WIOD rather select.3
3.2 Introduction to standard Input-Output Analysis
For this thesis IO analysis is performed, therefore a brief introduction to the offered methodology is required. The IO analysis in this thesis is inspired by the early work of Leontief & Ford (1970). A world input-output table (WIOT) entails a summary of all transactions in the global economy, between industries (intermediate deliveries) and final users across countries (final goods). A schematic outline of a WIOT is shown in Figure 1 below:
Figure 1: Schematic outline of a WIOT (Timmer et al., 2015).
Before going into detail, it is useful to note that a bold upper case letter refers to a variable which is shaped as a matrix (Z), a bold lower case letter refers to a vector (x). A vector is a column by definition, a row vector is obtained by transposition, and indicated by a prime (x′). Scalars are indicated by italicized lower case letters (𝑧𝑖𝑗𝑅𝑆). The subscripts and superscripts indicate the location
11 of the scalar, considering its rows and columns. The row is mentioned first, the column second (in 𝑧𝑖𝑗𝑅𝑆 R and i refer to a row, where S and j refer to a column). A circumflex indicates a diagonal matrix with the elements of any vector on its diagonal, with all other values equal to zero (x̂). The core of the IO model is the intermediate deliveries, or inter-industry requirements matrix, referred to as the Z matrix. It includes a series of rows and columns, where the rows refer to the production of industry i within country R and where the columns refer to industry j in country S using these inputs. So the data in each column corresponds to the amount of inputs being used for production by that industry. Every industry in every country has a row (how much it produces and where the output goes to) and a column (how much they buy from other industries for their production). The Z matrix is a Mn x Mn matrix, since in the WIOD 41 countries (M = 41) with 35 industries (n = 35) per country are included. The whole Z matrix consists of 1435 (41x35) rows and 1435 (41x35) columns (Timmer et al., 2015). Its element 𝑧𝑖𝑗𝑅𝑆 indicates the intermediate deliveries (in million US $) of industry i in country R that is sold to industry j country S (with i, j = 1, 2, …, n | with R, S = 1, 2, …, M). The intermediate use can be either at domestic markets, with n x n submatrices 𝐙𝑅𝑅 on the diagonal, or at foreign markets with off-diagonal submatrices 𝐙𝑅𝑆,
where R≠ 𝑆 (Meng et al., 2018). All values in WIOT are in millions of US $, so when in this section there is referred to ‘one unit’, this equals one million of US $. The typical element of the Z matrix 𝑧𝑖𝑗𝑅𝑆 can be found in context in Figure 1 above.
The original final demand matrix in WIOD has the size of Mn x Mc matrix, where c refers to the number of final demand categories (final consumption by households, final consumption by non-profit organizations serving households, final consumption by government, fixed capital formation, changes in inventories and valuables). In the analysis no distinction will be made between the five different final demand categories of WIOD, they are summed for every country, so that a final demand matrix F with Mn rows and M columns (1435x41) is obtained. The element 𝑓𝑖𝑅𝑆 then gives the goods and services produced by sector i in country R that are bought by the summation of the five final demand categories. As opposed to the Z matrix, trade flows in the F matrix have left the production process of the industry/country under consideration. The vector x, with Mn elements as well, represents the total output of each industry, either used by final (F) or intermediate (Z) purchasers. Note that for each industry i in country R, the total value on its row, describing to what industries and final demand categories it is sold, equals the sum of the values in its column, showing by what industry they are used. This ensures that the total output x can be obtained by summing each row, and by summing each column, which gives x′.
The production technology matrix A can be derived from the Z matrix, it has the exact same Mn x Mn structure (1435x1435). Its typical element 𝑎𝑖𝑗𝑅𝑆 is defined as 𝑎
𝑖𝑗𝑅𝑆 = 𝑧𝑖𝑗𝑅𝑆 / 𝑥𝑗𝑆 , where the element
in the Z matrix is divided by the element of the gross output vector x. The columns of the A matrix reflect the input from each sector in each country, required to produce one unit of output (in millions of US $) in the respective sector in a certain country (Timmer et al., 2015; Xu & Dietzenbacher 2014). The model is given by the following equation, where F is summed to a vector f using a summation vector i of M elements (Fi = f):
12 Rewriting this as (I-A)x = f gives the solution of eq. (1) as:
x = (I-A)-1 f = Lf eq. (2)
Where (I-A)-1 is the Leontief inverse, also known as the multiplier matrix L. It accounts for the output that is necessary for a certain final demand. Its element 𝑙𝑖𝑗𝑅𝑆 gives the production in industry
i in country R that is necessary for one unit of final demand of sector j in country S. It does not only account for all the direct production requirements necessary for one additional unit of final demand, but also the indirect input deliveries. The direct inputs require inputs in their production, which in their turn also need intermediate inputs to be produced, which in their turn again have their production requirements, and so forth. So the total production necessary to satisfy a final demand yields the direct and all the indirect effects. The L matrix has, just as the A matrix, the exact same Mn x Mn size as the Z matrix.
IO analysis can be extended for environmental applications, like quantifying EEX. In order to estimate emissions generated with the production of a certain good/service, a direct carbon emission coefficient vector d is introduced. Its element 𝑑𝑗𝑅 indicates the direct carbon emission in
kilotons (Kt) per unit of sector j’s output in country R. The carbon emissions coefficients are obtained by dividing an industry’s total carbon emissions (as of WIOD’s environmental account) by its total output xj. An important issue in this thesis is how much emissions are generated by
producing a certain final demand. In order to find out the following equation follows:
𝐝̂(𝐈 − 𝐀)−1 = C eq. (3)
The carbon emission intensity matrix C is obtained and shows how much emissions are generated to satisfy one unit of final demand, instead of satisfying one unit of output, which is the case for d. Its element 𝑐𝑖𝑗𝑅𝑆gives the carbon emissions in industry i in country R that are necessary for satisfying one unit of final demand in industry j in country S.
Besides comparing the measures of EEX mutually, it is also worthy to compare them with XEE. In the latter, the emissions are allocated to the country which eventually consumes the good. A brief overview on how to calculate XEE using the above-described concepts will follow. The export of emissions can be calculated by multiplying C with the diagonal final demand matrix 𝐟̂. The diagonal final demand matrix is obtained by taking the diagonal of f, a 1435x1 column vector with the total final demand for every industry on its rows. XEE is defined as follows:
𝐝̂(𝐈 − 𝐀)−1 𝐟̂ = C𝐟̂ = XEE eq. (4)
XEE is a 1435x1435 matrix which gives the sources of emissions in every industry’s final goods production. Its element 𝑋𝐸𝐸ijRS gives the emissions generated, indirectly or directly in industry i in
13 XEER = ∑ 𝑖 ∑ 𝑗 ∑S≠RXEEijRS eq. (5) It must be stated that some important assumptions are made in IO analysis using WIOD. The production technology for every country and every industry is based on fixed proportions. A given sector uses the same production technology, regardless of for who it is produced (foreign- or domestic demand). The A matrix of input-output coefficients reflects this. Another assumption in IO analysis is that a given sector only produces one good or service. In essence, a column in the A matrix reflects the production recipe of a given product (Lenzen, Kanemoto, Moran & Geschke, 2012). The production recipe for a certain sector, for example construction, is different per country. To produce a product in the German construction sector it requires different inputs than producing a product in the Dutch construction sector.
3.3 Overview measures of EEX
In this section the relevant measures of EEX, their implications and mathematical details are discussed. The following methods are discussed: the emissions embodied in bilateral trade for exports (EEX-1) by Peters (2008); the emissions embodied in exports (EEX-2) by Meng et al. (2018); the emission content in exports (EEX-3) by Johnson (2018) and the emissions embodied in exports (EEX-4) by Serrano & Dietzenbacher (2010) and extended by Xu & Dietzenbacher (2014).
In order to obtain the EEX-1 by Peters (2008) first, the emissions embodied in bilateral trade (EEBT) needs to be calculated. This determines the domestic emissions in one country R generated by 𝐞RS, the exports from country R to country S. Since there is no distinction made between
intermediates or final consumption, 𝐞RS is defined as follows, where i again represents a summation
vector of appropriate length:
𝐞RS = 𝐳RS𝐢 + 𝐟RS eq. (6) In order to obtain EEBT for country R, the Leontief inverse of 𝐀RR instead of A is taken, in order
to exclusively use domestically generated environmental impacts. The domestic environmental impacts embodied in the exports, the EEBT, from region R to region S are as follows:
𝑐RS= 𝐝R′(𝐈 − 𝐀RR)−1𝐞RS eq. (7)
By summing the EEBT for all the regions where is exported to, the total emissions embodied in bilateral trade for exports for region R to every other region S can be determined:
EEX-1 = ∑S≠R𝑐RS eq. (8)
The EEX-1 limits the level of processing to national borders. The measure only distinguishes between intermediate and final consumption on a domestic perspective. For exports no distinction is made between the two, this has an important implication for this measure. In this way, it only measures the domestic CO2 emissions that have been emitted in the last phase of production that
14 case with XEE), but only the emissions generated within a nation’s territory to produce the total exports to a country.
This measure is originally based on an NIO, but can be extended for a GMRIO model. All the 40 regions where is exported to from country R are summed together, so that one trade flow from region R to all the other countries is obtained. This trade flow 𝐞RS contains the exports of region R to all its exporting partners. This is also done in an NIO, which is a strongly aggregated part of a GMRIO for one specific country. The A matrix in an NIO is similar to 𝐀RR in a GMRIO. Using
this information, it can be shown that the EEX-1 in a GMRIO and an NIO model are equal to each other. EEX-1 as defined for an NIO is as follows:
EEX-1 (NIO) = 𝐝R′(𝐈 − 𝐀)−1(∑ 𝐞RS
S≠R ) eq. (9)
Eq. (9) above is equal to inserting eq. (7) in eq. (8). Note that eq. (8) is used for the analysis, any output regarding EEX-1 refers to this definition, and not to the definition in eq. (9).
Second, Koopman et al. (2014) propose a method which breaks up a country’s gross exports into various value-added components, and additional double counted terms. They differentiate between the domestic content and foreign content of gross exports. While doing this they state that official trade statistics must have various double counted terms. Since official trade statistics are measured in gross terms, it includes intermediate as well as final goods, when these intermediate goods cross borders multiple times, their value is counted twice. In a comprehensive framework they show how to decompose and correct for the double counted terms. Although in this model value-added in exports is estimated, the same can be done for emissions in exports. Which is what Meng et al. (2018) have done, they extend the line of research of Koopman et al. (2014), and integrate trade in value-added with embodied emissions in trade.
Meng et al. (2018) propose a method which decomposes the total emissions by a country in five terms. Each term represents a different GVC route. Three of the five terms are used to estimate EEX-2. A fourth term can be added to EEX-2, in order to obtain the EEX-1 as proposed by Peters (2008). This fourth term refers to emissions which are generated in the production of intermediates for foreign countries, but eventually come back home via imports (are re-imported). Meng et al., (2018) exclude the emissions generated in the production of intermediate exports, which are later on re-imported to satisfy domestic final demand (REE). Therefore it is expected that the EEX-2 as proposed by Meng et al. (2018) gives a lower amount of emissions in exports than the EEX-1 as proposed by Peters (2008).
The EEX-2 and its mathematical details will be discussed now. Like stated before, the total emissions by a country are decomposed in five terms, which each represent a different GVC route. Three of these five terms are used to construct EEX-2. Meng et al. (2018) start with the gross output produced by country R, given as:
𝐱R = 𝐀RR𝐱R+ ∑ 𝐀RS𝐱S+ 𝐟RR + ∑M 𝐟RS S≠R M
S≠R = 𝐀RR𝐱R+ 𝐟RR+ ∑MS≠R𝐞RS
15 Where 𝐞R∗ = ∑MS≠R𝐞RS is the total exports of country R. The export 𝐞RS from country R to country S again contain intermediate, and final good exports, just as in eq. (6). Rearranging eq. (10) using the local Leontief inverse gives:
𝐱R = (𝐈 − 𝐀RR)−1𝐟RR+ (𝐈 − 𝐀RR)−1𝐞R∗ eq. (11)
The last part of eq. (11), the exports 𝐞R∗, are then decomposed into exports of intermediate and
final products, and to their destination of absorption. Besides the source country R and the destination country S, exports to third countries T are included. Meng et al. (2018) claim that the decomposition of exports is as follows4:
(𝐈 − 𝐀RR)−1𝐞R∗ = (𝐈 − 𝐀RR)−1(∑MS≠R𝐟RS+ ∑MS≠R𝐀RS𝐱RS) = ∑M 𝐋RS𝐟SR S≠R + ∑ 𝐋 RS𝐀SR(𝐈 − 𝐀RR)−𝟏 M S≠R 𝐟 RR + ∑M 𝐋RR S≠R 𝐟 RS + ∑ 𝐋RS𝐟SS + ∑M 𝐋RS S≠R ∑mT≠R,S𝐟ST M S≠R eq. (12)
In the above equation, the Leontief inverse L is used. It accounts for the output that is necessary for a certain final demand. The matrix 𝐋RS gives the production in country R that is necessary for one unit of final demand in country S. Here, (𝐈 − 𝐀RR)−1, the local Leontief inverse differs from 𝐋RR due to the latter’s inclusion of off-diagonal terms via the inverse operation. The Leontief
inverse, (𝐈 − 𝐀)−1 or L, uses the complete A matrix in the inverse operation, while the local
Leontief inverse only uses the domestic inter-industry requirements 𝐀RR in this process. In this inverse operation of the local Leontief inverse, 𝐀RR is a n x n submatrix of the complete A matrix,
where the off-diagonal submatrices 𝐀RS (R≠S) are equal to zero. A partitioned, graphical
representation of the appropriate A matrix used in the inverse operation is shown below:
𝐀 = [ 𝐀11 ⋯ ⋮ ⋱ 𝐀1R ⋮ ⋯ 𝐀1M ⋰ ⋮ 𝐀R1 ⋯ 𝐀RR ⋯ 𝐀RM ⋮ ⋰ 𝐀M1 ⋯ ⋮ 𝐀MR ⋱ ⋮ ⋯ 𝐀MM]
Implementing eq. (12) in eq. (11), and pre-multiplying the direct carbon emission coefficient diagonal matrix 𝐝̂ with it, results in eq. (13) which gives the total emissions by country R. Like stated before, this equation can be decomposed in five terms. Eq. (13) is given as:
𝐝̂RxR = 𝐝̂R(𝐈 − 𝐀RR)−1𝐟RR + 𝐝̂R(𝐈 − 𝐀RR)−1∑ 𝐀RS∑ 𝐋M ST T 𝐟TR M S≠R + 𝐝̂R∑ 𝐋RR𝐟RS + 𝐝̂R∑M 𝐋RS𝐟SS S≠R M S≠R + 𝐝̂R∑MS≠R𝐋RS∑MT≠R,S𝐟ST eq. (13)
Meng et al. (2018) claim that the sum of the first two terms on the right-hand side in eq. (12) are equal to the second term on the right-hand side of eq. (13). Technical details and proof of this derivation is provided in Wang, Wei & Zhu (2013). The five terms in eq. (13) each represent a different GVC route. All the emissions that are generated in country R are present in one of the five
16 terms. A direct interpretation of the decomposition of the five terms by Meng et al. (2018, p. 38) is as follows:
- ‘’The first term (𝐝̂R(𝐈 − 𝐀RR)−1𝐟RR), captures the domestically produced and consumed
products and services. These are domestic emissions solely used for domestic final demand, they can be considered as country R’s pure self-responsibility.
- The second term (𝐝̂R(𝐈 − 𝐀RR)−1∑ 𝐀RS∑ 𝐋M ST T 𝐟TR M
S≠R ), which captures domestically
produced intermediates, which are exported and used by other countries as intermediate to produce final products, and eventually return back via imports to the source country R to be consumed there (REE).
- The third term (𝐝̂R∑M 𝐋RR𝐟RS
S≠R ), which captures domestically produced final products and
services, that are exported and consumed abroad.
- The fourth term (𝐝̂R∑MS≠R𝐋RS𝐟SS), which captures domestically produced intermediates exported to country S, used to produce final products which are consumed in country S. - The fifth term (𝐝̂R∑MS≠R𝐋RS∑MT≠R,S𝐟ST), which captures domestically produced
intermediates, exported to other countries, used to produce final products and services which are then exported to third countries. ‘’
The third, fourth and fifth terms are summed to obtain EEX-2, Meng et al. (2018) state that the sum of the second, third, fourth and fifth term equal EEX-1 as proposed by Peters (2008). The EEX-2 by Meng et al. (2018) is distinct from EEX-1 by Peters (2008) since it does exclude the portion of emissions that eventually return via imports.
The second term in the above decomposition (REE) will be used to quantify the difference between EEX-1 and EEX-2. In EEX-2 it is required that the emissions associated with a product are consumed in the destination country S, but with EEX-1 there are no such restrictions. EEX-1 is only concerned with where these emissions are generated, regardless of where they are finally absorbed. Meng et al. (2018) claim that this is a source of double-counting when estimating the global emissions generated by a country’s exports. Taking the earlier automobile industry example again, the emissions in the Brazilian iron ores are counted in the exports to the US, but they are also counted in the exports of Brazilian cars to the Netherlands. Since EEX-2 excludes REE, it gives a more accurate view on the domestic emissions generated by a country’s exports. To this extent, EEX-2 overcomes a drawback of EEX-1 by Peters (2008), it excludes double counted terms, where EEX-1 does not.
17 EEX-2 = EEX-1 − (𝐝̂R(𝐈 − 𝐀RR)−1∑ 𝐀RS∑ 𝐋M ST T 𝐟TR M S≠R ) eq. (14) EEX-2 = 𝐝̂R∑ 𝐋RR𝐟RS + 𝐝̂R∑M 𝐋RS𝐟SS S≠R M S≠R + 𝐝̂R∑MS≠R𝐋RS∑MT≠R,S𝐟ST eq. (15)
A drawback of the EEX-2 is its complexity, especially as compared to EEX-1. The EEX-2 by Meng et al. (2018) is derived from complicated matrix algebra, making it much harder to understand than the relatively simple calculation by Peters (2008). For example, the derivation of eq. (12) to eq. (13) is complicated, and not explained in the paper of Meng et al. (2018). Another paper by Wang et al. (2013) needs to be consulted in order to understand. Also, the EEX-1 is based on NIO, which intuitively makes it less complex, though at the cost of being unable to capture REE.
The third measure, EEX-3 from Johnson (2018) estimates the domestic content and foreign content of exports. It is also an extension on the work of Koopman et al. (2014). IO analysis is performed to estimate the value-added content in exports. This measure then separates the domestic content from the imported (thus foreign) content of exports. A decomposition of gross exports results in domestic value-added, foreign value-added and double counted residuals. The value-added coefficients are replaced by emission coefficients in order to obtain EEX-3. The details discussed below are for a two-region IO system, where region 2 should be seen as a gathering of all the other countries S. Region 2 thus has 1400 (40x35) industries, while region 1 has the original 35 industries. Note that this is a way of notation, the sizes of the matrices in essence do not change. Johnson (2018) starts with a reorganization of eq. (1) x = Ax + f where the exports of a country are isolated: x = [𝐀11 0 𝐀21 𝐀22] [𝐱 1 𝐱2] + [𝐟 11 0 𝐟21 𝐟22] + [𝐞 12 0 ] with 𝐞 12= 𝐀12𝐱2+ 𝐟12 eq. (16)
This reorganization removes the intermediate deliveries from country 1 to country 2 (𝐀12𝐱2) from
the inter-industry requirements matrix A and moves them to the variable 𝐞12 which represent
exports from country 1 to country 2. This leaves the modified A* matrix, which in essence is the same as 𝐀RR in Peters (2008) and Meng et al. (2018), only now for a two-region setting. The final
good exports from country 1 to country 2 (𝐟12) are also moved to 𝐞12 leaving a modified F* matrix.
The gross output needed to produce 𝐞12 is then pre-multiplied with the diagonal direct emission
coefficients matrix 𝐝̂ to compute the emission content of the exports of country 1: [𝐱𝐜11 𝐱𝐜21] = 𝐝̂(𝐈 − 𝐀 ∗)−1[𝐞12 0 ] = [ 𝐝̂𝟏(𝐈 − 𝐀11)−𝟏𝐞12 𝐝̂𝟐(𝐈 − 𝐀22)−𝟏 𝐀21(𝐈 − 𝐀11)−1𝐞12] eq. (17)
In eq. (17) xc is the vector of emissions from country R required to produce exports of country S. Summing over all sectors, the total amount of domestic emissions embodied in country 1’s exports is measured by 𝐱𝐜11, and where 𝐱𝐜21 equals the total foreign emissions embodied in country 1’s exports. EEX-1 is the exact same as 𝐱𝐜11, the distinction in this measure is that it also incorporates the foreign emissions embodied in a country’s export 𝐱𝐜21. Summing the vector 𝐱𝐜 results in
EEX-3, the total emissions required to produce the exports of country 1: EEX-3 = 𝐢′[𝐱𝐜11
18 However 𝐱𝐜21 does not give the true value of the foreign content of exports, due to double counted
terms. Again, in the situation where Brazilian iron ores are first exported to the US, used in the assembly process of steel plates, and are later on being re-imported by Brazil as an intermediate, the emissions in the iron ores will be double counted. The emissions related to the iron ores will be double counted if the imported steel plates are used in the assembly of cars, which eventually will be sold to foreign consumers (exports). This double counted term can be identified in eq. (17) above, where the term (𝐈 − 𝐀22)−𝟏 𝐀21(𝐈 − 𝐀11)−1𝐞12 is the vector of country 2’s output needed
to produce the intermediates imported by country 1, used in the production of country 1’s exports. Pre-multiplying this term with 𝐀12 gives the value of the inputs exported from country 1, that are
re-imported and used to produce country 1’s exports. The double counted emissions in the exports of country 1 are defined as follows:
𝐝̂𝟏𝐀12(𝐈 − 𝐀22)−𝟏 𝐀21(𝐈 − 𝐀11)−1𝐞12 eq. (19)
It is interesting to mention that Johnson (2018) only finds double counted terms in the foreign content of exports, and not in the domestic content of exports. Unlike Meng et al. (2018), who claim that double counted terms are present in the domestic content of exports. Both measures are based on EEX-1, but the authors find different double counted terms in the same trade flow. To this extent, the literature on double counted terms is rather contradictive, or at least inconsistent. The double counted terms should be subtracted from EEX-3 to obtain a more accurate estimate of 3. In order to do so, the definition of 3 in eq. (18) is altered. The new definition of EEX-3, without double counted terms in its foreign content, is defined as:
EEX-3 = 𝐢′[𝐱𝐜11
𝐱𝐜21] - 𝐝̂𝟏𝐀12(𝐈 − 𝐀22)−𝟏 𝐀21(𝐈 − 𝐀11)−1𝐞12 eq. (20)
To summarize, EEX-3 estimates the total emissions embodied in a country’s exports. The exports can be decomposed into domestic emission content and imported content. The imported content can then be further decomposed into a foreign emissions content, and a double counted term. As opposed to EEX-1 by Peters (2008) and EEX-2 by Meng et al. (2018), EEX-3 contains domestic and foreign content, while the former two purely focus on domestically generated emissions. The last measure of EEX to be discussed is EEX-4 by Xu & Dietzenbacher (2014). This measure is an extension on the emissions in exports as proposed by Serrano & Dietzenbacher (2010), who estimate EEX for a two-region setting as follows:
EEX (Serrano & Dietzenbacher, 2010) = 𝐝̂1𝐋11𝐟12+ 𝐝̂2𝐋21𝐟12+ 𝐝̂1𝐋12(𝐟21+ 𝐟22) eq. (21)
19 EEX-4 = 𝐝̂K∑M 𝐋KR
K=1 ∑MS≠R𝐟RS + 𝐝̂R∑S≠RM 𝐋RS∑MK=1𝐟SK eq. (22)
Note that the summation over K in the above equation goes from 1 till M, in other words the summation is over all the countries in the dataset, including country R. A direct interpretation of eq. (22) by Xu & Dietzenbacher (2014, p. 12) is as follows:
- ‘’The first part (𝐝̂K∑MK=1𝐋KR∑MS≠R𝐟RS) represents the emissions (generated anywhere in the world) that are embodied in the exports of final goods by country R to be consumed in any other country. The first term in the first part (𝐝̂K∑MK=1𝐋KR) reflects the emissions generated in all countries that are necessary for one unit of final goods produced in a certain sector in country R. The second term in the first part (∑M 𝐟RS
S≠R ) is a 35x1 column vector
which gives each sector’s final goods, produced in country R to be consumed in any other country.
- The second part (𝐝̂R∑M 𝐋RS
S≠R ∑MK=1𝐟SK) represents the export of the emissions generated
in country R’s intermediate production to all the other countries, which are consecutively used in the production of final goods. The first term in the second part (𝐝̂R∑M 𝐋RS
S≠R )
reflects the emissions in country R embodied in one unit of final goods produced in any sector in any other country. The second term in the second part (∑M 𝐟SK)
K=1 is a 1435x1
column vector which represents the final goods produced in country S for final users in all the other countries, including country R. The product of these two terms in the second part of eq. (22) gives the amount of emissions in country R embodied in the intermediates that end up in the final goods produced in any other country S.’’
To summarize, eq. (22) thus gives the global emissions embodied in the exports of final products by country R, and the domestic emissions of country R that are ultimately embodied in the final goods of other countries. The fact that EEX-4 also includes global emissions embodied in the exports of final products makes it distinct from the measures by Peters (2008) and Meng et al. (2018). Since EEX-3 by Johnson (2018) also has a domestic- and foreign part, it is interesting to compare the two, and see how they differ numerically. Though it should be noted that for EEX-4 the foreign content is only found in final good trade, while EEX-3 measures foreign content in final good, as well as intermediate trade.
Separating the domestic emissions from the foreign emissions in EEX-4 enables a comparison with EEX-1 and EEX-2. The second part of eq. (22) already represents the export of emissions generated domestically in country R. Though the first part contains the emissions generated anywhere in the world. To isolate the domestic emissions in the first part of eq. (22), the domestic emissions embodied in exports of final products is altered to 𝐝̂R𝐋RR∑MS≠R𝐟RS. The domestic content of EEX-4 (DC EEX-EEX-4) for country R can now be written as follows:
21
4. Results
Following the discussed methodologies, now follows the empirical results of the different methods. Section 3 discusses the theoretical differences between each measure of EEX, while this section discusses the numerical differences. The implications of these results also follow in this section. This section ends with some limitations which have to be kept in mind when considering the results. First, for the sake of reference, the total emissions emitted by industries, and the total emissions emitted by households over the period 1995-2009 as of WIOD are shown in Figure 2. It is clearly portrayed that an increasing amount of CO2 is emitted over time. From around 22 gigatons (Gt) of
carbon emissions in 1995 to almost 29 Gt in 2009. These numbers coincide with the continued economic growth for this period, which is accompanied by an increase in environmental pressures (Munksgaard & Pedersen, 2001).
Figure 2: The total annual CO2 emissions (in Gt) by industries and households for the period
1995-2009.
The total emissions for every measure of EEX in 2008 are presented in Table 2 below. The year 2008 is chosen since this is the most recent year in the database which is not affected by the financial crisis. In WIOD it is the year with the highest amount of emissions embodied in trade, which can be seen in Figure 2. All results in this section are for the year 2008, unless stated otherwise. In Table 2, the total emissions in trade are presented as a share (%) of the total annual emissions emitted, as well as their absolute amount. The absolute values in Table 2 are CO2 emissions in Gt. Every row represents a measure, referring to an equation as of section 3. As
a benchmark XEE is shown too.
0 5 10 15 20 25 30 35 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Total annual CO
2emissions (in Gt) for the period 1995-2009
22 Table 2: The total emissions in exports for each measure of EEX in 2008, as a share of the total annual emissions (%), and their absolute amount (Gt).
EEX-2 is present twice in Table 2, due to its two definitions (eq. 14 and eq. 15). Meng et al. (2018) claim that these two definitions are conceptually similar and yield the same outcome. From Table 2 it can be concluded that this is not true. It is arbitrary that Meng et al. (2018) claim that these equations yield the same results, while in fact there clearly is a considerable difference. EEX-2 according to eq. (14) yields a total emissions in exports of 9,4 Gt, while EEX-2 according to eq. (15) yields a total emissions in exports of 9,9 Gt, an inconsistency of 0.5 Gt carbon emissions in 2008. A debatable issue are the double counted terms (REE), which are deducted from EEX-1 to obtain EEX-2 (eq. 14). For example Johnson (2018) only accounts for double counted terms in the foreign content of emissions, and not in EEX-1. The fact that eq. (14) and eq. (15) do not coincide makes the REE by Meng et al. (2018) rather arbitrary. Due to this unneglectable inconsistency, both definitions of EEX-2 will be used further in this section.
As shown in Table 2 the emissions embodied in exports entail large volumes, with varying amounts per measure. For example, EEX-3 and EEX-4 have considerable higher total emissions in exports than the other measures. A great part of the difference between these measures can be explained due to EEX-3 and EEX-4 including foreign content, whereas the other measures of EEX strictly contain domestic content. For a more relevant comparison, the four measures should be solely based on domestic emissions. When looking at the domestic content of EEX-3 and EEX-4 (DC EEX-3 & DC EEX-4) the total emissions in exports are much closer to each other. The measures of EEX which capture the domestic emissions traded in 2006 have their share of the total annual emissions vary from 31,8% - 34,2%. Note that EEX-1 and DC EEX-3 are the same, thus portraying the exact same value. This comparison is based on eq. (8), eq. (14) and eq. (23) respectively. Between each measure, there are still differences to be found due to the corresponding assumptions. For example, DC EEX-4 does not exclude any double counted terms, where EEX-2 does. The total emissions in trade for each measure of EEX are higher than what for example is found in Wiedmann (2016), where the highest amount of emissions in trade is reportedly 8.3 Gt for the year 2007, also using WIOD.
The most surprising aspect of the performed IO analysis is the discrepancy between the total annual emissions for every measure of EEX and XEE. The deviation between the estimates of XEE and
Total emissions in exports
Share of total annual emissions (%)
23 the domestic emissions in EEX in 2008 ranges from 3.7 Gt to 4.4 Gt, depending on the measure of EEX. This is a significant difference, which has not been quantified, or at least not specifically pointed out by the relevant literature consulted, to the best of author’s knowledge. What causes this difference is difficult to point out exactly. A plausible explanation is that intermediates which are traded back-and-forth between countries are being captured by EEX, but not by XEE. Intermediates which are produced domestically, are exported, and later on re-imported to satisfy domestic final demand are not captured by XEE. Since these intermediates are ultimately consumed domestically, XEE does not treat them as exports, hence XEE is unable to measure these emissions. Consider the earlier example of the automobile industry again, if the Brazilian iron ores are exported abroad, and used in the production process of cars which are eventually sold to Brazilian consumers, the embodied emissions are not measured by XEE. These domestically produced intermediates do cross the border, as a consequence they are measured by EEX. This phenomena explains the higher amounts of emissions traded around when measuring EEX as opposed to XEE.
In recent years there has been an increasing amount of international fragmentation. Lower costs of transportation, and communication have led to the rise of GVCs with production activities being scattered across countries. The immediate result is a huge increase in intermediate trade over the past few decades (Timmer et al. 2014). Keeping this trend in mind when analyzing the discrepancy between EEX and XEE, it is hypothesized that the difference between the two concepts has become more prevalent over time. Since the increase in international fragmentation goes hand in hand with an increase in international trade, both EEX and XEE have risen over the past few years, but it is expected that EEX has risen by a greater amount. In order to find out if the above is true, XEE, EEX and their difference are analyzed. Figure 3 shows EEX-1, XEE and their difference for the period 1995-2009. All values in Figure 3 are CO2 emissions in Gt. Note that EEX-1 is chosen as
the measure to benchmark with XEE. EEX-1 is the preferred measure due to its simplicity, also it does not face any issues related to multiple definitions (e.g. EEX-2). Besides, all measures of EEX follow a similar trend, so the same conclusions will be derived if comparing XEE with any other measure of EEX5.
24 Figure 3: EEX-1, XEE and their difference (in Gt) for the period 1995-2009.
From Figure 3 it can be concluded that indeed both EEX and XEE have been on the rise for the period 1995-2009. An increase in international trade is clearly visible, almost every year higher volumes of carbon dioxide are emitted through trade, this is true for both EEX and XEE. The difference between the two concepts has become greater over this same period. Where the difference between EEX and XEE was around 2,8 Gt in 1995, this grew to an estimated 3,8 Gt in 2009. This growing difference can be (at least partly) attributed to an increase in international fragmentation (e.g. more trade in intermediates), as discussed by the relevant literature (Timmer et al., 2014; De Backer & Miroudot, 2014). Since more and more intermediates are circling around between countries, it is expected that the difference between EEX and XEE will grow even further apart in the near future.
Table 3 below illustrates an overview of all the measures of EEX for every country. Every column represents a measure, referring to an equation as of section 3, whereas every row reflects the emissions from a certain country. All the values in Table 3 are CO2 emissions in thousands of
Table 3: An overview of all the measure of EEX for every country in 2008. XEE is shown as a benchmark. All values are CO2 emissions in thousands of
Kt. Between brackets the values are shown as a share of the total emissions of each measure (in %).
Country / Measure EEX-1 eq. (8) EEX-2 eq. (14) EEX-2 eq. (15) EEX-3 eq. (20) DC EEX-3 eq. (8) EEX-4 eq. (22) DC EEX-4 eq. (23) XEE eq. (5)
26 From Table 3 it can be concluded that some deviation between the measures of EEX is present. This is due to the assumptions made in each measure. For example, every country’s EEX-3 and EEX-4 are in general considerably higher than the other measures of EEX, due to their inclusion of foreign content. EEX-3 is again mostly higher than EEX-4, since the latter only includes foreign content in its intermediate exports, where the former has foreign content embodied in its intermediate and final good exports. By definition 2 (using eq. 14) is always lower than EEX-1, the difference between the two is exactly portraited by the REE for every country. The difference between EEX-1 and EEX-2 for every country seems minimal, however some regions like the United States and RoW do show a notable deviation. Also when there is looked upon the total REE a different perspective is attained. The total double counted terms in 2008 are close to 300 thousand Kt of carbon emissions, which is almost as much as three times the total Dutch EEX-1, or once that of Germany.
Table 3 illustrates that on the country level the measures which solely include domestic emissions in exports (EEX-1, EEX-2, and DC EEX-4) also do not deviate by significant numbers. The mutual differences between these three methods stand in sharp contrast when compared to the discrepancy between EEX and XEE. Closer inspection of Table 3 shows that the region with the greatest difference between all the measures of EEX and XEE is RoW. The share of RoW as of the total emissions in XEE sees a very high deviation with the share of Row in EEX, it is around 15% lower with XEE than in any measure of EEX. Compared to any other region, there is no region that comes closes this deviation. Recall that intermediates which are produced domestically, are exported, and later on re-imported to satisfy domestic final demand are not captured by XEE. It could be that a relatively high share of these intermediates are ultimately consumed domestically (e.g. consumed in RoW), as a consequence XEE being much lower than EEX (in %).
Another explanation is the emission intensity in RoW being high. Relatively high carbon intensity intermediates are produced domestically and then exported to foreign countries. It could be that RoW consumes goods which are produced in relatively clean, developed countries, while exporting ‘dirty’ products. The high levels of emissions produced in RoW are assigned to RoW when measuring EEX, but when measuring XEE they are assigned to the country which consumes the good. If this is the case this would explain RoW’s lower XEE as opposed to EEX since the former assigns emissions to the country which consumes the final good.
27 It is important to bear in mind that there are some limitations to the above values, the results therefore need to be interpreted with caution. These limitations are mostly related to the assumptions made when using WIOD, and performing IO analysis. IO analysis is a great tool for quantifying embodied emissions in trade, and for analyzing the differences between production- based and consumption-based emissions. However, the GMRIO models and environmental accounts used in these applications rest to a certain extent on estimates. To construct a GMRIO model it requires an enormous amount of data from various sources. This data is not always perfect, one can imagine that in some scenarios choices have to be made regarding data coverage, aggregation, etc. For some countries there is not even data available, at least to some extent choices have to be made for these specific cases (Xu & Dietzenbacher, 2014). As a consequence, it has to be noted that some uncertainty is present in the dataset used, and thus in the results.
Furthermore, a problem is that no GMRIO table distinguishes between production for domestic or foreign markets. One average carbon emission coefficient is taken for a certain industry in a certain country, every good produced in this sector generates the same amount of emissions, while in reality this is often not true. Countries which typically produce for foreign markets, like China, have different production techniques for their domestic and foreign market (Johnson, 2018). By having only one emission coefficient per sector per country some biases in the estimates of embodied emission flows could be present.
Another limitation regarding the WIOD is the region RoW being used as a summative region to include all the other countries which are not in the dataset itself. Including more countries to the set of countries, in other words disaggregating RoW, would provide value for the accuracy of the estimates of EEX. Disaggregating RoW would, for example result in more details on the difference between EEX and XEE. Since this is the region with the highest deviation between the two accounting principles, a disaggregation of this region would lead to more information on what exactly causes this discrepancy.
28
5. Discussion and conclusion
This section starts with a brief discussion and interpretation of the findings from section 4. Thereafter, concluding remarks are made regarding the research question of this thesis. This section ends with a few suggestions for future research.
As mentioned in the literature review, a distinction should be made between EEX and XEE. The two methods take a different perspective on how to measure the emissions embodied in trade. With XEE the emissions which are generated during the production process are assigned to the country which consumes the good, but with EEX the emissions are assigned to the country which produces the good. The difference between the two methods has become more prevalent over time, due to international trade becoming more fragmented. The process of fragmentation has become more widespread due to plummeting costs of communication, coordination, and transportation. This decrease in costs has made it more profitable to slice up the production process, where every stage is located in the lowest-cost location. This has changed the nature of international trade fundamentally, from trade in consumer goods towards more trade in tasks and activities (Timmer et al., 2014).
In the academic literature there is consensus on how to calculate XEE, while this is not the case for EEX. There are multiple methods to estimate EEX described in the literature therefore this study was set out to answer the research question: What measures of EEX are proposed in the literature, and how do they differ – analytically and numerically? The objective was to come up with a comparison which accounts for the multiple methods of EEX present in the literature. By defining each measure and by writing each measure’s mathematical details using the same variables and a consistent design, this thesis complements the academic literature. The relevant models in this thesis are as follows: the emissions embodied in bilateral trade for exports (EEX-1) by Peters (2008); the emissions embodied in exports (EEX-2) by Meng et al. (2018); the emission content in exports (EEX-3) by Johnson (2018) and the emissions embodied in exports (EEX-4) by Serrano & Dietzenbacher (2010) and extended by Xu & Dietzenbacher (2014).
29 Future research could potentially tell if it indeed is true that XEE and EEX continue to grow further apart due to an increase in trade in intermediates. A more detailed, and recent view on the considerable discrepancy between XEE and EEX needs to be developed in order to find out. For example, a structural decomposition analysis could help to analyze this gap between the two perspectives for a certain timeframe. Due to scope and time limitations, this was not possible to accomplish in this thesis. It could also be of interest to use a different database for the IO analysis, and see if it yields similar results. Preferably a database which covers more recent data could be used, since the time period used in this thesis is rather outdated. It would be interesting to see what results are obtained if a more recent time period is chosen, sadly the WIOD does not allow to verify this.
Another application for future research could be an extension on the work of Koopman et al. (2014) and Meng et al. (2018) on vertical specialization. Like aforementioned, the domestic content and foreign content in final good and intermediate exports can be used to measure vertical specialization and a country’s position in a GVC. The equations in section 3.3 of this thesis allow to easily identify various measures of vertical specialization, using the foreign content in exports. An interesting finding of this thesis is the fact that eq. (14) and eq. (15) do not coincide, while there is claimed otherwise. Question marks can be put next to the estimations of the double counted terms (REE) by Meng et al. (2018), where not all researchers agree to the derivation of these. Some researchers only account for double counted terms when estimating the foreign content of emissions in trade, where some also claim to find double counted terms in the domestic content of emissions in trade. More research on the double counted terms in trade in emissions is desired. At the moment the academic literature is not sufficient in providing a clear and consistent explanation on where and when to account for double counted terms. Future research should aim to broaden the knowledge regarding this topic. In the long run this will improve the accuracy of the estimations on emissions in trade, for the better of environmental studies and theses like this.
While the focus of this thesis is on CO2 emissions, the same can be done for any environmental
pressure. The WIOD for example has data on CH4, N2O, NOX, SOX, CO, NMVOC, and NH3.
Under the consideration that CO2 emissions are closely related to industrial production, this
