当前位置:首页 >> 能源/化工 >>

A Shapley decomposition of carbon emissions without residuals


Energy Policy 30 (2002) 727–736

A Shapley decomposition of carbon emissions without residuals
Johan Albrechta,*, Delphine Francoisa, Koen Schoorsb a

Faculty of Econ

omics and Business Administration Centre for Environmental Economics and Environmental Management (CEEM), University of Ghent, Hoveniersberg 24, 9000 Ghent, Belgium b University of Oxford, SBS, 59, George Street, Oxford, OX1 2BE, UK

Abstract Conventional decomposition techniques for historical evolutions of carbon emissions present path dependent factor weights of selected variables next to signi?cant residuals. Especially for analyses over long periods with many variables, high residuals make it almost impossible to derive reliable conclusions. As an alternative, we present the Shapley decomposition technique for carbon emissions over the period 1960–1996. This technique makes it possible to present a correct and symmetric decomposition without residuals. The starting point of our analysis was an extended Kaya Identity with nine components. In a study of four countries, the Shapley decomposition showed that the carbon intensity of energy use and the decarbonization of economic growthFvariables that are targeted with current climate policy measuresFhave more e?ect on total emissions than generally suggested in conventional decomposition exercises. Another interesting conclusion from our analysis was that the e?ect of population growth on emissions can be for some countries more important than the decarbonization e?orts. r 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Carbon dioxide emissions; (Shapley) decomposition; Kaya identity

1. Introduction In many ?elds of social sciences, decomposition techniques are used to help disentangle the impact of various causal factors. An analysis of energy-related carbon emission patterns and its driving factors can provide essential information for policy studies on national strategies and the use of ?exible instruments in climate policy. A decomposition of total CO2 emissions over a number of contributing factors sheds some light on the importance of crucial parametersFlike the ongoing decarbonization of energy services or the rate of autonomous energy e?ciency improvementsFthat are used in scenarios to calculate the possible cost of climate policy scenarios for developed countries. While sophisticated forecast technologies are available for the latter type of exercises, traditional decomposition analyses with a limited number of factors still yield important residuals, even over short periods of time. Another problem is that the value of the
*Corresponding author. Tel.: +32-9-264-3474; fax: +32-9-2643599. E-mail addresses: johan.albrecht@rug.ac.be (J. Albrecht), Delphine.francois@rug.ac.be (D. Francois), koen.schoors@sbs.ox.ac.uk (K. Schoors).

contribution assigned to any given factor depends on the order in which the factors appear in the elimination sequence. Factors that are not treated symmetrically lead to an important ‘path dependence’ problem (Shorrocks, 1999). This strongly reduces the relevance of decomposition exercises for studies over longer periods. Therefore, we present in this paper a Shapley decomposition of carbon emissions that eliminates both problems. After an introduction to the Kaya Identity, we ?rst work with four contributing factors or components (carbon/energy, energy/GDP, GDP/population and population) for the period 1960–1996. For the same data set, we present the Shapley decomposition results next to the results from a traditional decomposition. Our calculations are based on data for Belgium, France, Germany and the United Kingdom. In the next step, we decompose the ?rst two factors over three economic sectors (industry, transport and other sectors) and discuss the main ?ndings from this detailed Shapley decomposition.

2. The Kaya identity and a decomposition for four countries The aim of a decompostion analysis is to reveal the importance of distinct components or factors that drive

0301-4215/02/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 3 0 1 - 4 2 1 5 ( 0 1 ) 0 0 1 3 1 - 8

728

J. Albrecht et al. / Energy Policy 30 (2002) 727–736

historical data. The relative weight of each factor in the observed change can be relevant information for policy measures. In the ?eld of climate policy, reliable information on the ongoing decarbonization of industrialized economies and on the sensitivity of energy intensity to energy price shocks is necessary input for policymakers. This information is necessary to evaluate various strategies to achieve the reduction targets of the six Kyoto Protocol greenhouse gases, with or without international ?exibility instruments. There are two broad categories of decomposition techniques: input–output techniques and disaggregation techniques. Both techniques have di?erent data requirements but the latter are more suitable for international comparisons, which explains their widespread use. For methodological details we refer to Liaskas et al. (2000) and Park (1992). A recent overview of methodological issues in cross-country/region decomposition techniques is given by Zhang and Ang (2001). They apply two popular conventional decomposition methods, i.e. the Laspeyres method and the arithmetic mean weight Divisia method, and two recently proposed perfect decomposition methods, i.e. the re?ned Laspeyres method and the logarithmic mean weight Divisia method to two data sets and discuss the di?erences. In their de?nition, perfect methods give complete decomposition results regardless of the data pattern, as opposed to the conventional methods. To stimulate the debate on the best decomposition method, we present in the next sections of this paper another perfect decomposition method. Important residuals constitute the most serious problem with conventional disaggregative decomposition. Liaskas et al. (2000) decompose industrial CO2 emissions for a number of European countries. They work with two periods, 1973–1983 and 1983–1993, and the factors in the decomposition are output, energy intensity, fuel mix and structure. For some countries, the weight of the residual in the decomposition over only ten years exceeds the weight of three of the other four components. For the United Kingdom, the Netherlands, Italy, France, Finland, Spain, Denmark, Belgium and Austria, the weight of the structural componentFor the structural economic e?ect on emissionsFis lower than that of the residual. Obviously, conclusions from this type of analysis are not always straightforward. For longer periods and for analyses with more components, the residual becomes even more problematic. This is illustrated in the next section with data for Belgium, France, Germany and the United Kingdom. We ?rst present a mathematical expression of total emissions by means of the Kaya Identity (Kaya, 1990). This equation provides a useful tool to decompose total national carbon emissions (C): C ? ?C=E??E=GDP??GDP=P?P ?1?

The formula links energy-related carbon emissions (C) to energy (E), the level of economic activity (GDP=gross domestic product) and population (P). C=E denotes the carbon intensity of energy use, E=GDP is the energy intensity of economic activity and GDP=P is the per capita income. At any moment in time, the level of energy-related carbon emissionsFnext to emissions that result from changes in land-useFcan be seen as the product of the four Kaya Identity components. For small to moderate changes in the Kaya components between any two years, the sum of the percent changes in each of the variables closely approximates the percent change in carbon emissions between those two years: d?ln C?=dt ? d?ln C=E?=dt ? d?ln E=GDP?=dt ? d?ln GDP=P?=dt ? d?lnP?=dt: ?2? The historical trends in the Kaya Identity components provide a reference point for evaluating current and future climate policy projections of carbon emissions as well as the key economic, demographic and energy intensity factors leading to those emissions. With the availability of detailed data, the impact of for instance the replacement of coal in electricity generation by natural gas or nuclear power can be compared to the impact of economic growth on energy-related emissions. The Kaya Identity can reveal interesting di?erences between emission patterns of developed and developing countries. For an analysis based on the Kaya Identity of the implications of emission trading under the Kyoto Protocol for the US economy, we refer to Dougher (1999). 2.1. Kaya in the International Energy Outlook 2001 A global view is given in Table 1 with the Kaya Identity components for three world regions. A historical analysis for the period 1970–1999 is complemented with the reference case projections for 1999–2020 from the International Energy Outlook 2001 (EIA, 2001). Positive annual average growth rates of carbon emissions between 1970 and 1999 are found for developed as well as for developing countries. For all countries, economic growth and population growth outpaced declines in energy intensity and carbon intensity of energy use. The average annual decline of carbon emissions of 5.4% in the 1990s in Eastern Europe and the Former Soviet Union presents a special case. This decline is the result of a severe drop in economic output per capita (?4%/yr). The IEO2001 reference case projections illustrate that reductions of carbon emissions require accelerated declines in energy intensity and/or carbon intensity. Such changes may in turn require signi?cant changes in the existing energy infrastructure. It remains questionable whether these necessary changes can be realised in one or two decades.

J. Albrecht et al. / Energy Policy 30 (2002) 727–736 Table 1 Average annual percentage change in CO2 emissions and the Kaya Identity components by region, 1970–2020a History Parameter Industrialized World C/E (%) E/GDP (%) GDP/P (%) P (%) C emissions (%) Developing World C=E (%) E=GDP (%) GDP=P (%) P (%) C emissions (%) 1970–1980 ?0.5 ?1.1 2.4 0.9 1.7 1980–1990 ?0.7 ?2.0 2.2 0.7 0.2 1990–1999 ?0.5 ?0.7 1.6 0.6 1.0 Reference case projection 1999–2010 0.0 ?1.3 2.2 0.5 1.4

729

2010–2020 0.1 ?1.3 2.0 0.4 1.1

?0.8 ?0.4 3.5 2.2 4.6

?0.2 0.9 1.7 2.1 4.5

?0.7 ?1.0 3.1 1.7 3.1

?0.1 ?1.4 3.7 1.7 3.9

?0.1 ?1.4 4.2 0.8 3.5

Eastern Europe and the Former Soviet Union C=E (%) ?0.8 E=GDP (%) 1.4 GDP=P (%) 2.4 P (%) 0.9 C emissions (%) 3.9
a

?0.3 0.6 0.6 0.7 1.6

?1.0 ?0.5 ?4.0 0.0 ?5.4

?0.2 ?2.4 4.1 0.0 1.4

?0.3 ?2.6 4.5 0.0 1.5

Source: Energy Information Administration (2001).

Energy use in the transport sector will continue to depend on oil since there are currently few economical alternatives. And if such an alternative would soon arrive (e.g. ethanol, bio-methanol, hydrogen, fuel cell technology, etc.), reshaping the current fuel delivery infrastructure would take a long time. Furthermore, we need to consider actual political decisions that will later have important consequences for energy infrastructures and energy-related emissions. Several European countries already committed to a complete phase-out of domestic nuclear power generation. This decision will slow down the decline in carbon intensity as a result of the increased use of fossil fuels. 2.2. An analysis for four countries In Table 2, we present for four European countries data for the period 1960–1996. Instead of working with average annual percentage changes as in Table 1, we still use the Kaya Identity but subdivide the total period in three subperiods: the early years before the oil crisis (1960–1973), the oil crisis years (1974–1986) and recent history or the period after the oil crisis (1987–1996). Information on the used sources and calculations that were necessary to compile Table 2 are given in the appendix. Carbon or CO2 emissions did increase in the four countries over the period 1960–1996 but the di?erences are remarkable. Total emissions increased strongly in Belgium, Germany and France, but only modestly the United Kingdom (UK). It is important to note that the

di?erent situation in the UK seems to be the result of what happened in the early years of the analysis. From 1960 to 1973, emissions in the UK did grow by only 13.8% while emissions in Germany and France more than doubled (+123% resp. +109%). Emissions in Belgium did increase by 85.1% from 1960 to 1973. These divergent post-war evolutions depend on strategic (and even environmental) fuel choices for electricity production and on the strong and uneven development of energy-intensive industries like iron and steel production, chemical manufacturing and mining. London experienced a sequence of ‘killer smogs’ in the late 1940s and early 1950s, leading to the Clean Air Act of 1956 that imposed much lower sulfur dioxide concentrations (Elsom, 1997). New technologies were installed and coal gradually was replaced by gas and oil. The result was a decline of carbon and sulfur dioxide emissions. A good example of the di?erent evolution in energyintensive industries is the metals industry in the UK that did grow by only 0.06% per year during the period 1954–1973. For that period the average annual growth rate of ‘all manufacturing’ was 0.88% with the highest growth rates found for instruments (+1.25%), electrical engineering (+1.39%) and vehicles (+1.38%). For the period 1973–1986, the metals industry faced an average annual growth of ?0.73% while the average for UK manufacturing was ?0.47% (Oulton and O’Mahony, 1994). For countries like Belgium, the strong growth of the iron and steel industry was one of the driving economic forces in the post-war era.

730

J. Albrecht et al. / Energy Policy 30 (2002) 727–736

Table 2 Carbon emissions (percentage changes per period) 1960–1996 (%) Carbon emissions Belgium France Germany United Kingdom Carbon/Energy Belgium France Germany United Kingdom Energy/GDP Belgium France Germany United Kingdom GDP/Population Belgium France Germany United Kingdom Population Belgium France Germany United Kingdom +86 +102 +96 +9.7 ?24.3 ?26.1 ?21.6 ?23.1 ?15.9 ?12.1 ?8.4 ?37.3 +162 +145 +143 +103 +11.4 +27.8 +12.7 +12.2 1960–1973 (%) +85.1 +109.7 +123.3 +13.8 ?12.4 ?10.4 ?10.4 ?12.7 +13.1 +17.9 +43.6 ?12.6 +75.1 +74 +59.9 +39.1 +6.8 +14.1 +8.6 +7.3 1974–1986 (%) ?12.8 ?12.6 +1.6 ?6.3 ?10.8 ?14.7 ?5.5 ?5.1 ?19.4 ?18.9 ?15.9 ?21.4 +19.9 +19.2 +30 +24.2 +1.2 +5.9 ?1.6 +1.1 1987–1996 (%) +22.3 +15.2 ?9.7 +6.5 ?1.8 ?1.7 ?6.4 ?5.9 +3.2 ?1.1 ?20.4 ?3.7 +17.7 +13.4 +14.9 +14 +2.6 +4.6 +5.4 +3.1

During the oil crisis years, emissions decreased in all countries with the exception of Germany (+1.6% over the period 1974–1986). After 1986, the reverse happened: German emissions decreased with 9.7% as a result of the closing down of parts of the former Eastern German economy in the early and mid 1990s, while emissions in other countries did increase. The growth of emissions in the UK is again more modest than in Belgium and in France. With the Kyoto Protocol of December 1997, the year 1990 became the base year for the necessary reductions of emissions. Especially Germany will bene?t from this choice since its emissions increased strongly in the period 1960–1990, followed by a sharp reduction as a result of the German uni?cation. The impact of the oil crisis on the energy intensity of production (energy/GDP) provides an indication of the potential of prices instruments (like energy or CO2 taxes) to reduce energy use in developed economies. Table 2 shows that during the period 1974–1986, the energy intensity in the four countries has been reduced by 15.9–21.4%. The oil crisis did clearly lead to a similar reaction in the four countries but it is interesting to note that from 1960 to 1974, the energy intensity in the UK decreased by 12.6% while in the other countries a completely di?erent evolution did take place. The increase of energy intensity of the post-war German economy between 1960 and 1973 is striking (+43.6%).

In the period after the oil crisis years, the energy intensity of production more or less stabilized in France and even increased in Belgium (+3.2%). The latter evolution can be ascribed to the strong growth of the basic chemical industry in Belgium. Over the total period 1960–1996, energy intensity decreased by 37.3% in the UK while the reduction in the other three countries was between 8.4% and 15.9%. Table 2 also shows that pro capita economic growth over the period 1960–1996 was strongest in Belgium, followed by France and Germany. The di?erence with the growth rate for the UK is mainly the result of slower income growth in the UK in the period before the oil crisis. Before the oil crisis years, pro capita growth was in all countries higher than during or after the oil crises when Europe faced long periods of economic recession. Finally, from 1960 to 1996 the population increased in all countries and especially in France. The total French population growth is more than double of that in the three other countries. We will later show that the French population growth is an important component in the decomposition of total emissions growth. Only since 1987, Germany faces a higher population growth than France. The data in Table 2 are used in the analysis in the next sections. There is however an important innovation in Section 4 where we will also include a structural element, this is the result of changes in the structural composition of the economy.

3. An introduction to the Shapley decomposition Table 2 provides information on the evolution of Kaya Identity factors for three di?erent periods. Changes in these factors have caused changes in total carbon emissions. It is therefore interesting to know the relative importance of each Kaya Identity factor in explaining the growth of total emissions. If we take the case of Belgium, we ?nd in Table 2 for the period 1960– 1996 that C=E has been reduced by 24.3% while the decline of E=GDP was 15.9%. Total emissions increased by 86% so what is the impact of these speci?c factors reductions? How would emissions evolve if all other variables were held constant and C=E was reduced by 24.3%? Is the impact of the reduction of E=GDP with 15.9% more or less important? To answer these questions, we need to decompose total emissions over a number of factors. More than 100 decomposition studies in energy and environmental studies are listed in a survey by Ang and Zhang (Ang and Zhang, 2000). They indicate that the methods reported prior to 1995 always leave a residual after decomposition. This residual was sometimes omitted, causing a large estimation error. In other models the residual was regarded as the interaction e?ect, which still leaves a new puzzle for

J. Albrecht et al. / Energy Policy 30 (2002) 727–736

731

the reader (Sun, 1998). Some methods proposed after 1995 are perfect, i.e. do not leave a residual term in the results (Ang and Zhang, 2000). One of these perfect decomposition methods is the one introduced by Sun (Sun, 1998). In his method, referred to as the re?ned Laspeyres index method (Ang and Zhang, 2000), the interactions (residual) are distributed equally among the main e?ects based on the ‘jointly created and equally distributed’ principle. In order to do this, one has to assume that ‘there is no reason to assume contrary’. However, no further proof of the validity of this assumption is given. Sun and Ang (2000) apply the same principle to the Paasche and Marshall-Edgeworth forms. Contrary to the Laspeyres index model, which adopts a prospective view, the Paasche index adopts a retrospective view. The Marshall-Edgeworth index adopts a compromising view based on the Laspeyres and Paasche indices (Sun and Ang, 2000). The authors prove that when the ‘jointly created and equally distributed’ principle is applied to the Paasche and Marshall-Edgeworth models, the decomposition results are identical to the Laspeyres form results. But the method implies the assumption made by Sun. Another perfect decomposition method discussed by Ang and Zhang (2000) is the logarithmic mean Divisia method, proposed by Ang and Choi (1997). They replaced the arithmetic mean weight function used in the arithmetic mean Divisia index method by a logarithmic mean weight function. This re?nement results in a perfect decomposition, but one has to take into account the fact that using a logarithmic weight function implies the assumption of a constant growth rate. Furthermore, this re?ned Divisia method is based on the normalization of the weight function, because the sum of the weight function over all sectors is not unity, but by de?nition always slightly less than unity (Ang and Choi, 1997). We add to this literature by proposing a perfect decomposition of carbon emissions based on the Shapley value. Indeed, the decomposition problem has formal similarities with a classical problem in cooperative game theory. Shapley (1953) was the ?rst to give a formula for the real power of any given voter in a coalition voting game with transferable utility. This is commonly referred to as the Shapley value. The Shapley value is the mathematical expression of the real power of a player when all orders of coalition formation are equiprobable. The Shaply value distributes the real power among the players, satisfying three axioms, namely symmetry, no inessential players and additivity. Symmetry means that every player should be treated symmetrically in the estimation. No inessential players means that players that do not contribute to the power of any coalition do not receive any power. Additivity means that the power derived from every single possible coalition can be added to ?nd the total real power.

Since 1953, the Shapley value has been used in a number of cost allocation models. The properties of symmetry and no inessential players are very useful in this context. A clear and simple explanation of how to use the Shapley value in cost allocation problems is given in Hamlen et al. (1977). Young et al.(1982) proposed to use the Shapley value to allocate costs in water resources development projects. Kattuman et al. (1999) proposed to use the Shapley value to allocate the costs of electricity transmission losses in the network between several electricity generators. In these Shapleyvalue based cost allocation models everything happens as if the cost drivers enter the equation one by one, each of them receiving their marginal contribution to the total cost. All orders of entering the cost equation are considered and receive the same weight 1=n! in the computation of the ultimate allocation of costs. Shorrocks (1999) points at the formal similarity between the original Shapley value coalition problem and the general problem of allocating a certain amount of any output or cost among a set of agents, bene?ciaries or cost drivers. Shorrocks builds on this similarity to construct a general decomposition procedure based on the Shapley value (Shapley, 1953). Basically, the technique involves estimating the impact of eliminating each factor in succession, repeating this exercise for all possible elimination sequences (the symmetry property) and then for each factor averaging its estimated impact over all the possible elimination sequences (the additivity property). Let us consider what this concretely implies for the decomposition. In a simple ceteris paribus type decomposition one calculates the impact of each variable, leaving other variables constant. Because of the interactions between several variables, this gives rise to a residual. The literature has come up with several ways to avoid or allocate this residual (see higher). One simple method is to calculate the contribution of one variable, and then add cumulatively more and more variables. The result is a perfect decomposition without residuals. However, the order in which we include variables largely determines their calculated contribution because the allocation of the interaction e?ects depends on the order of inclusion of the variable. Since the results depend on the order by which variables enter the calculation, this cumulative approach is path dependent and hence biased. The underlying problem is that variables are not treated symmetrically. The Shapley decomposition iterates the cumulative approach for every possible order (permutation) of variables. With n variables, we need to make n! calculations, with each calculation based on another order for including new variables. The Shapley value implies that taking the average of the n! estimated contributions for every variable, yields the true contribution of each variable. As a result, the Shapley decomposition has three major advantages. First of all,

732

J. Albrecht et al. / Energy Policy 30 (2002) 727–736

the decomposition is perfect, meaning that the sum of the impacts, allocated to each of the explanatory variables, equals the observed change in the decomposed variable. One does not need to make any assumptions or e?ort to allocate the residual, as the solution is free from residuals. Secondly, the Shapley decomposition is symmetric (or anonymous): the factors are treated in an even-handed manner, without making any further theoretical assumptions. Thirdly, the Shapley decomposition allows for very complex decompositions that would otherwise be troublesome because of very high residuals and subsequent interpretation problems. We will illustrate the Shapley decomposition of a more complex identity in the next section.

mendations, this type of information is essential, especially for analyses with extended time horizons. By including these e?ects in our analysis, we have nine components for the decomposition: three sectoral carbon intensity e?ects (C=Eindustry ; C=Etransport ; C=Eother ; denoted as Cji in (3)) three sectoral energy intensity e?ects (E i in (3)), the e?ect of pro capita GDP (Ppc in (3)), the population e?ect (P in (3)) and the o structural e?ect. Name ajt the share of a sector j at time t in total production, the ?nal equation for the change in carbon emissions over n sectors is presented in (3): P  P  n pc n pc i i i i j?1 ajt Cjt Ejt Pt Pt ? j?1 aj0 Cj0 Ej0 P0 P0 P  C? ?3? n i i aj0 Cj0 Ej0 Ppc P0 j?1 0 As a result of the inclusion of sectoral and structural e?ects, there are some modest di?erences in the data, when comparing to the data used in Table 2. We explain our data in the appendix. We illustrate the problem of residuals in the basic decomposition case with only four components: carbon intensity, energy intensity, GDP and structure of the economy. These components are used in most decomposition exercises (see Liaskas et al., 2000). Notice that these four factors are not the Kaya Identity factors. The Kaya equation does not include a structural e?ect. With nine components the analysisFand especially the residualsFwould even be more troublesome. The results of the non-Shapley decomposition are presented in Table 3, panel (A). Panels (B) and (C) present the results of a Shapley decomposition without residuals.

4. Sectoral and structural e?ects in the decomposition Starting from (1) and the data in Table 2, we add sectoral and structural e?ects to our analysis over the period 1960–1996. The change in carbon intensity of energy use and the change of energy intensity of production can be due to changes within sectors (sectors become more or less intensive in energy and carbon) or changes between sectors (sectors that are intensive in energy or carbon become more or less important in total production). For simplicity, we work with three sectors: industry, transport and other sectors. For these three sectors, changes between 1960 and 1996 in the carbon intensity of energy and in the energy intensity of the production are calculated. For climate policy recom-

Table 3 Three decompositions of carbon emissions (A) With residuals (four components) Components Belgium France Germany UK C=E (%) ?25.6 ?28.4 ?22.6 ?23.9 E=GDP (%) ?16.3 ?10.2 ?14.5 ?36.1 GDP (%) 192.2 212.7 174 122.5 Structure (%) 2.4 0.2 6.4 ?1.9 Sum (%) 152.6 174.3 143.2 60.6 Real (%) 84 107.1 98.8 4.6 Residual (%) 68.6 67.2 44.4 56

(B) Shapley decomposition without residuals (four components) Belgium France Germany UK ?42.9 ?55 ?39.4 ?41.1 ?27.5 ?16.3 ?25.5 ?71.4 154.1 176.4 152.4 124.4 ?0.02 2 11.5 ?8 84 107.1 98.8 4.6 84 107.1 98.8 4.6 F F F F

(C) Shapley decomposition without residuals (nine components) C=E (%) Industry Belgium France Germany UK ?19.6 ?23.1 ?17.3 ?12.9 Transp. ?1.9 ?3.5 ?3.5 ?3.1 Other ?21.4 ?28.4 ?18.6 ?25.1 E=GDP (%) Industry ?31.2 ?26 ?12.5 ?20.4 Transp. 12.6 5.9 ?9.7 4.7 Other ?8.9 3.8 ?3.3 ?55.7 144 148 140 111 10.1 28.4 12.4 13.4 ?0.02 2 11.5 ?8 84 107.1 98.8 4.6 GDP=P (%) POP (%) Struct. (%) Sum (%)

J. Albrecht et al. / Energy Policy 30 (2002) 727–736

733

The information in Table 3 (A) should in principle answer the question ‘how do carbon emissions change if one component changes and other components are ?xed?’ The reliability of the answer when applying a decomposition method with residuals, is however very limited since the sum of changes strongly exceeds the real change of emissions. An ‘overexplanation’ between 44% and 68% is found for the four countries in our sample. There is no reliable way to attribute these residuals to the four components and hence interpretation becomes very cumbersome.

5. Results from the Shapley decomposition Panels (B) and (C) of Table 3 further illustrate the disturbing impact of the residuals that are found in Table 3. It is shown that the analysis with residuals clearly overestimates the e?ect of economic growth on emissions and underestimates the e?ects of changes in carbon and energy intensity (C=E and E=GDP), the two important ‘target’ parameters in climate policy. Especially the carbon intensity e?ect on emissions is much more important in Panels (B) and (C) than suggested in Panel (A) of Table 3. Our exact decomposition reveals that shifts in fuel mixes in?uence carbon emissions more than suggested in the decomposition with residuals. For the four countries, the real weight of this factor (in Panels (B) and (C)) is almost double of the value suggested in Panel (A). Similar conclusions are valid for the energy intensity factor. A more detailed picture follows when we apply the Shapley procedure for a decomposition with 9 components. This decomposition starts from the Kaya equation and adds the structural factor to it. The results are presented in Panel (C) of Table 3. This analysis without residuals allows us to formulate some further conclusions. For the UK, the analysis is based on the period 1960–1995. First of all, we notice that the structural e?ect is not that important for explaining total emissions growth. For Germany, the structural e?ect did lead to an increase of emissionsFa growth of 11.5% when holding all the other factors constantFwhile for the UK emissions decreased (?8%) as a result of the structural e?ect. In the analysis with four components and with residuals, the weight of the structural component for Germany is lower (+6.4%). For Belgium and France, there is almost no structural e?ect found in Table 3 Panel (C). Does the fact that some energy-intensive sectors did become relatively more or less important without signi?cantly in?uencing total carbon emissions suggest that future structural changes will also have a modest impact on total emissions? Since every economic

sectorFagriculture, industry and servicesFconsumes energy, shifts between sectors have a limited impact especially because energy e?ciency is for most service industries not the same priority as it is for industries like basic chemicals whose pro?t base depends on it to a large extent. Population growth had a positive impact on emissions, especially for France. We notice that for France, the population e?ect is more important that the separate sectoral carbon intensity and energy intensity e?ects. The population e?ect (+28.4%) is for France stronger that the e?ect of the ongoing energy e?ciency improvement of the French economy (?26%+ 5.9%+3.8%=?16.3%). When this population growth is expected to continue in the next decades, this development will negatively impact the possibility for France to achieve the same emission reductions as countries with more stable populations. If this hypothesis would hold, a possible solution would be to base future European burden-sharing agreements on population data. The energy needs that follow from the population growth have a stronger impact on carbon emissions than the e?ort to reduce the energy intensity of the French economy. To reverse this evolution, the increase in energy e?ciency should not be limited to French industry (?26%) but should become visible in transport and other sectors especially housing, hospitals, schools and administrations. These ‘asymmetric e?ciency gains’ seem to be especially valid for France. In Belgium, Germany and the UK, we ?nd that the e?ciency gains of the ‘others’ sector indeed have a negative impact on emissions when we hold all other factors constant. Especially for the UK, the e?ciency gains of the ‘other’ sectors are spectacular (?55.7%) and had a strong impact on total emissions. The shift from less e?cient energy use to more e?cient energy use in households and services is very important. Did this shift already take place in the other three countries or did they just not recognize this potential yet? The e?ect of the average income is, as could be expected, the most important component in the growth of carbon emissions. As a result of the pro capita income e?ect, emissions would ceteris paribus have increased with 111%–144% (Panel (C)). This reveals that the impact of economic growth (pro capita income e?ect plus population growth e?ect) is overestimated in traditional decomposition analyses as is illustrated by the high factor weights in Table 3 (A). Without the e?ect of economic growth, carbon emissions would have decreased in the four countries from 1960 to 1996. In contrast to most developing countries, the growth of emissions for the four countries in Table 3 (B) is lower than the e?ect of GDP growth. The argument that the Kyoto Protocol targets can be achieved at a low cost without impacting economic growth therefore seems to depend on access to low-cost emission reduction opportunities. These low hanging

734

J. Albrecht et al. / Energy Policy 30 (2002) 727–736

fruits can probably be found in developing or transitional countries. Table 3 (C) shows that for industry, the decrease in energy intensity had an important impact on total emissions for all countries. This e?ect is strongest for Belgium where the increasing energy e?ciency of industry could reduce emissions by 31.2% when other factors are held constant. For Germany, this e?ect is weakest (?12.5%). There is still no indication of a trend towards increased energy e?ciency in the transport sector. Cleaner and more fuel-e?cient transport equipment can not compensate the strong growth of transport activities in most developed countries. Another explanation can be the declining market share of rail transport in the container market in countries like Belgium. Holding all other components constant, the evolution of the transport energy intensity would lead to increasing emissions in Belgium, France and the UK. Only for Germany, the impact would be negative. With respect to the carbon intensity of energy use, there are only minor di?erences between the four countries. The impact of the change in industrial carbon intensity on total emissions is between ?12.9% (UK) and ?23.1% (France). The impact of carbon intensity changes in transport is very similar: between ?1.9% for Belgium and ?3.5% for France and Germany. This is not a surprise since transport infrastructure is very similar in the four countries. The impact of other carbon intensity changes is more diverging. For Germany, we ?nd the lowest value with ?18.6%. For France this e?ect is most important: ?28.4%.

emissions depends on the evolution of other variables as well. The more interactions are included in the analysis, the more reliable the reported e?ects will be. This explains why we opted for an extended Kaya equation with nine components. Of course, we do not claim that this extended equation captures all relevant interactions. The di?erences in the evolution of the energy intensity of production also correspond to more explicit di?erences in the weight of this factor in the decomposition. The di?erence between Belgium and France is 3.8% (see Table 2: ?15.9% versus ?12.1%) while the di?erence in the total weight of this factor is 11.2% (?27.5% versus ?16.3% in Table 3 (B)). Similarly, compared to the UK, E=GDP is 25.2% lower in France (?37.3% versus ?12.1%). The total weight of this factor is for the UK ?71.4% while it is only ?16.3% for France. These ?ndings illustrate again that similar trends for parameters in Table 2 can be the result of very di?erent evolutions in similar developed economies and vice versa.

7. Conclusions Starting from the Kaya Identity, we presented a Shapley decomposition for carbon emissions for four European countries. This technique makes it possible to present a perfect and symmetric decomposition without residuals. Compared to conventional decomposition techniquesFwith residuals that amount to more than 50% of carbon emission growthFthis is a promising improvement o?ering valid and reliable information on complex questions such as the importance of speci?c energy-related evolutions in total emissions growth. From a limited analysis for four countries, we found that the Shapley decomposition showed that factors like the carbon intensity of energy use and the decarbonization of economic growth have more e?ect on total emissions than suggested in conventional decomposition exercises. In these exercises, the e?ect of economic growth on emissions is overestimated for developed countries since this important variable captures a signi?cant part of the residuals. But the real weight of this variable is lower and the weight of the other variablesFthose that are the essence of climate policyFis higher. These results seem to suggest that fuel mix changes and the ongoing decarbonization can in interaction with other e?ects play an important role in climate policy. Another interesting conclusion from our analysis was that the e?ect of population growth on emissions can be for some countries be more important than the decarbonization e?orts.

6. Growth versus component weight When we compare Table 2 to Table 3 (B) and (C), some points deserve our further attention. For the four countries, the di?erences in the carbon intensity of energy use during the period 1960–1996 seem to be modest in Table 2. Table 3 (B) shows that similar evolutions in carbon intensity (from ?21.6% for Germany to ?26.1% for France, see Table 2) can have a di?erent impact on total emissions (from ?39.4% to ?55%). Precisely this impact provides the most useful information for further policy development. The di?erence in carbon intensity between France and Belgium is only 1.8% (see Table 2) while the di?erence in total weight is 12.1% (see Table 3 (B)). Holding all other components constant, similar reductions in carbon intensity can lead to di?erent reductions in carbon emissions because of structural di?erences between di?erent economies. We therefore need to be aware of all the interaction e?ects that take place inside the economy. The evolution of one single variable can be interesting but the impact of this single variable on total

J. Albrecht et al. / Energy Policy 30 (2002) 727–736

735

Appendix A Population Source: OECD Energy Balances. GDP global Kaya Identity Source: OECD National Accounts, I. We used the Gross Domestic Product (Expenditure) data in US $ at exchange rates and price levels of 1990. sectoral decomposition Source: OECD National Accounts, II, Value Added by Kind of Activity approach (for compatibility reasons with the decomposition of the Energy data). Sectoral decomposition calculations for each separate country were as follows: Belgium (MN FB90 Price); France (MN FF80 Price) Total Industry Sector GDP=Manufacturing+Electricity, Gas and Water+Construction Total Transport Sector GDP=Transport and Storage Total Other Sectors GDP=Gross Domestic Product?Total Industry Sector GDP?Total Transport Sector GDP Germany (MN DM 91 Price) German ‘Value Added’ GDP data were only available from 1991. For the years 1960–1990 we used data from the former Federal Republic of Germany and added 10% as this was estimated to be the GDP share of the former German Democratic Republic. We assumed that the di?erent sectors were equally represented in both parts of the country, knowing that this would lead to only minor distortions of the data. Germany 1991–1996 Total Industry Sector GDP=Manufacturing+Construction Total Transport Sector GDP=Transport, Storage and Communication Total Other Sectors GDP=Gross Domestic Product?Total Industry Sector GDP?Total Transport Sector GDP Germany 1960–1990 Total Industry Sector GDP=Manufacturing+Electricity, Gas and Water+Construction Total Transport Sector GDP=Transport and Storage Total Other Sectors GDP=Gross Domestic Product?Total Industry Sector GDP?Total

Transport Sector GDP United Kingdom (MN PS Curr. Price) Total Industry Sector GDP=Manufacturing+Electricity, Gas and Water+Construction Total Transport Sector GDP=Transport, Storage and Communication Total Other Sectors GDP=Gross Domestic Product?Total Industry Sector GDP?Total Transport Sector GDP The baseyear and the used currencies di?er from those used in the global Kaya Identity. We therefore multiplied all sectoral GDP data by correction factors (GDP used in global Kaya formula/‘Value Added’ GDP) for each year. Energy Source: OECD Energy Balances. We opted for Total Final Consumption as a basic concept in calculating the Carbon and Energy Kaya Identity Factor. Total Final Consumption is the sum of consumption by the di?erent end-use sectors. The reason for this choice is the fact that Total Final Consumption data can easily be disaggregated into di?erent sectors, more speci?cally the industry sector, the transport sector and the other sectors (Agriculture, Commerce and Publ. Serv., Residential and Non-speci?ed). We did not include Non-Energy Use. A correction factor was included when comparing the sum of the sectoral decompositions with the global Kaya results. Carbon Total Final Consumption data comprise the use of di?erent energy sources, as well as Electricity and Heat. The decomposition of Electricity and Heat is based on data from Electricity Plants, CHP Plants and Heat Plants. Emission factors from fossil fuel combustion were found in the ‘Second Netherlands’ National Communication on Climate Change Policies’ and the ‘Revised 1996 IPPC Guidelines for National Greenhouse Gas Inventories: Reference Manual’.

References
Ang, B.W., Choi, K.H., 1997. Decomposition of aggregate energy and gas emission intensities for industry: a re?ned Divisia index method. The Energy Journal 18 (3), 59–73. Ang, B.W., Zhang, F.Q., 2000. A survey of index decomposition analysis in energy and environmental studies. Energy 25, 1149–1176. Dougher, R., 1999. The Kyoto Protocol: implications of emissions trading scenarios. American Petroleum Institute Research Paper #095, July.

736

J. Albrecht et al. / Energy Policy 30 (2002) 727–736 Park, S.-H., 1992. Decomposition of industrial energy consumption: an alternative method. Energy Economics 14, 265–270. Shapley, L., 1953. A value for n-person games. In: Kuhn, H.W., Tucker, A.W. (Eds.), Contributions to the theory of games, Vol. 2. Princeton University, Princeton, NJ. Shorrocks, A.F., 1999. Decomposition procedures for distributional analysis: a uni?ed framework based on the Shapley value. Mimeo, University of Essex. Sun, J.W., 1998. Changes in energy consumption and energy intensity: a complete decomposition model. Energy Economics 20, 85–100. Sun, J.W., Ang, B.W., 2000. Some properties of an exact energy decomposition model. Energy 25, 1177–1188. Young, H.P., Okada, N., Hashimoto, T., 1982. Cost allocation in water resources development. Water Resources Research 18 (3), 463–475. Zhang, F.Q., Ang, B.W., 2001. Methodological issues in crosscountry/region decomposition of energy and environment indicators. Energy Economics 23, 190–197.

Elsom, D., 1997. Atmospheric pollution trends in the United Kingdom. In: Simon, J. (Ed.), The State of Humanity. Blackwell, Oxford, UK. Energy Information Administration, 2001. International Energy Outlook 2001. Washington DC, March 28. Hamlen, S.S., Hamlen, W.A., Tschirhart, J.T., 1977. The use of core theory in evaluating joint cost allocation schemes. Accounting Review 52 (3), 616–0627. Kattuman, P.A., Bialek, J.W., Abi-Samra, N., 1999. Electricity Trading and Co-operative Game Theory. Proceedings of the 13th Power System Computation Conference, Trondheim, June 28–July 2, 1999, pp. 238–243. Kaya, Y., 1990. Impact of carbon dioxide emission control on GNP growth: interpretation of proposed scenarios. Paper presented at the IPCC Energy and Industry Subgroup, Response Strategies Working Group, Paris, France. Liaskas, K., Mavrotas, G., Mandaraka, M., Diakoulaki, D., 2000. Decomposition of industrial CO2 emissions: the case of European union. Energy Economics 22, 383–394. Oulton, N., O’Mahony, M., 1994. Productivity and Growth. A study of British Industry, 1954–1986. Cambridge University Press, Cambridge.


相关文章:
我国二氧化碳排放的主要特点及减排路径
A shapley decomposition of carbon emissionswithout residuals[J].EnergyPolicy,2002,30: 727-736. 5 Si、Fi、I、R 分别表示能源结构、能源排放强度、能源效率、...
更多相关标签:
carbon emissions | shapley值 | shapley value | shapley | gale shapley算法 | shapley值法 | shapley值法例题 | gale shapley |