The effect of the EU ETS on the environmental and productive performance of EU public power plants Jurate Jaraite and Corrado Di Maria November 14, 2009 Abstract This study explores whether the European Union’s CO2 Emissions Trading Scheme (EU ETS) could provide incentives for the public power plants responsible for about one third of EU-27 CO2 emissions, to improve their environmental and productive performance. The existing ex post literature only partially addresses this issue - productive performance is measured partially, or environmental performance is neglected. This paper develops an approach for measuring the overall impact of the EU ETS on the public power plants of the EU member states during 1996-2007. Firstly, it measures environmental efficiency and total factor productivity growth taking into account environmental variables. For this purpose, Data Envelopment Analysis is used as the principle methodological base. Secondly, the impact of the EU ETS, fossil fuel prices and other variables on the environmental efficiency and productivity indicators are estimated by using econometric techniques. Two findings emerge from this analysis: i) the EU ETS had a positive effect on the environmental efficiency and technical change of the public power plants; however, ii) the looseness of this policy reflected in the overallocation of the grandfathered permits led to the deteriorating performance.

Work in progress, do not quote!

1

Introduction

The major advantage of emissions trading is that it provides freedom and flexibility for firms in choosing their compliance strategies. The larger the differences in marginal cost, the greater the potential opportunities for gain from trade and the lower the total welfare cost relative to command and control for participants in reaching the abatement goals set by regulators. In nutshell, emissions trading ensures that environmental goals are reached in a cost-effective manner. However, from the perspective of environmental efficiency, that can be defined as a firm’s ability to improve on pollution only, emissions trading programmes “...can, at least partly, explain and even provoke environmental inefficiencies...” (Tyteca, 1996, p. 299). That is firms with high marginal costs of pollution abatement can meet their emissions constraints by buying permits on the market without improving its environmental efficiency, and this in perfect conformity with regulations. Tyteca names this situation as “the apparent paradox”. Of course, the economic context and other external factors such as energy prices, market structure, and design and stringency of emissions trading may also explain a firm’s motivation not to increase its environmental efficiency. In this paper we look at the trend in environmental efficiency and the changes of total factor productivity of the public power plants in the EU-24 member states in the period 1996-2007 to understand whether this paradox was present in the EU ETS. The main question is whether environmental efficiency and productive performance of energy generating public thermal plants and district heating plants (hereafter public power plants) have improved in the period 1996-2007 and whether these improvements can be explained by the EU ETS and other external factors. To answer to these questions we make use of non-parametric Data Envelopment Analysis and Malmquist Productivity Index approach to calculate environmental efficiency and total factor productivity change measures using aggregated country-level data of the EU-24 member states (MS) during 1996-2007. The impacts 1

of other factors, such as an allocation of emissions, energy prices, fossil fuel abundance and industry structure, on environmental efficiency and productivity change are estimated by using regression techniques. This paper begins with a short characterisation of the public power generation in Section 2. Section 3 presents used methodology. Section 4 describes the data set. In Section 5 the results are discussed. Finally, Section 6 concludes.

2

Characterisation of public power plants

Greenhouse gas (GHG) emissions from public electricity and heat production in the EU-27 member states account for one third of total EU-27 GHG emissions (Source: EEA GHG data viewer). 99% of these emissions are CO2 emissions which were covered in the first phase of the EU ETS (20052007) if they were emitted by power (electricity and heat) producing activities, so called combustion installations, with a rated thermal input exceeding 20 Megawatts. In the first phase of the EU ETS, the combustion installations, compared with the other installations, had a more stringent CO2 permit allocation (see Figure 1). The combustions installations were net short (shortfall between allocated European Union Allowances1 (EUA) and required) of around 50 thousand of EUAs while other installations were net long (allocated more EUAs than required) by approximately 190 thousand. The more stringent allocation for combustion installations was based on the fact that power generating industry is not directly exposed to international competition (Convery et al., 2008). The national net long and net short positions of the combustion installations show that nine countries had a shortage of allowances, while the remaining EU member states had the net long positions (Figure 1)2 . The combustion installations with the long positions either were overallocated or were able to reduce their emissions either because of declining production or by installing carbon-saving technologies. While the combustion installations with the short positions, depending on their marginal costs of pollution abatement and the prevailing market price of CO2 had three options to meet their carbon constraints: (1) install carbon saving technologies and/or switch to less polluting fuels, (2) buy allowances on the market, or (3) reduce carbon-intensive production. (2) and (3) compliance options suggest that installations could comply with environmental regulation without reducing emissions per unit of production. As the market price of CO2 collapsed in the first phase of the EU ETS and as most of the member states experienced a rapid economic growth in 2005-2007, we could expect that most of the combustion installations bought a shortage of allowances on the market. This could be facilitated by the international uncompetitive nature of energy market allowing energy producing firms to pass-through the CO2 opportunity costs more easily into consumer prices. The first empirical study on that shows that opportunity cost pass-through rates vary between 60% and 100% for the wholesale electricty market in Germany and the Netherlands (Sijm et al., 2006). By no means, the abatement option is not neglected. That CO2 emissions abatement occurred in the first phase of the EU ETS is a fact which is well documented in several studies. Ellerman and Buchner (2008) in their study conclude that between 130-200 Mt of CO2 were abated in 2005, and 140-220 in Mt in 2006 for all member states. However, they do not provide sectoral breakout of these results. Delarue et al. (2008) focus on the European power sector CO2 short-term abatement possibilities through fuel switching. Using both a non-calibrated and a historically calibrated simulation model the authors estimates of abatement are between 34.4 and 63.6 Mt in 2005, and 19.2 and 35 Mt in 2006 in the power sector alone. Authors note that there is no single constant relationship between the price of CO2 and abatement. The abatement also depend on the load level of the system (a carbon price will have its greatest effect at relatively low load levels when more lower emitting capacity is available) and the ratio between natural gas and coal prices. In the most recent study, Anderson et al. (2009) perform an evaluation of the first EU ETS phase by estimating a business as usual scenario 1 2

1 EUA corresponds to 1 tonne of CO2 . Bulgaria is excluded.

2

Figure 1: Net long and short positions for 2005-2007 at the national level, Kt CO2 . Source: EEA, CITL viewer. and by comparing it to allocated and verified EUAs to identify abatement and over-allocation. Their abatement estimates are 47 Mt of CO2 , 37 Mt and 23 Mt for 2005, 2006 and 2007, respectively, and reflect the downward sloping development of CO2 market price in this trading period. In general, the emissions intensity of carbon dioxide from public electricity and heat production, despite an increase in the amount of electricity and heat produced, has decreased substantially on the EU-27 level and on the national level since 1990 (EEA, 2008a). Some of this improvement might be explained by an improvement in energy efficiency, measured as a ratio of fuel output to input, due to closure of old inefficient plants, improvements in existing technologies and the installation of new, more efficient technologies, often combined with a switch from coal power plants to more efficient combine cycle gas-turbines. However, the rapid growth in fossil-fuel based electricity production outweighs some of the environmental benefits of the efficiency improvement (EEA, 2008b). Thus, reducing the emissions per unit of electricity and heat produced (emissions intensity) of these plants is regarded as the main strategy. The above emission intensity and energy efficiency indicators do not reflect all inputs and outputs used in power generation. For instance the inverse of energy efficiency indicator (a ratio of fuel input to output) so called energy intensity treats energy output and energy input as the only input and output. The efficiency and productivity measures based on Data Envelopment Analysis allow to evaluate efficiency within a multiple inputs and outputs production framework. Since there exist substitution effects among capital stock, labour force and energy inputs, a lower energy intensity may arise from the transformation of energy inputs to non-energy inputs in producing power rather than from energy efficiency improvement. These substitution effects cannot be reflected in the aggregate energy intensity, but they could be captured by the multiple inputs DEA models which might result in the contradiction between aggregate efficiency indicators and DEA-based efficiency measures (Zhou and Ang, 2008). The DEA provides an opportunity to highlight best practice, rather than average practice. In a semi competitive sector such as power generation with incentive problems, average performance may be well in the interior of the production possibility set. The multi-input, multi-output specification of the technology also increases the informational value of the benchmarking, in addition to the avoidance of a priori assumptions on the production possibility set. DEA, as any parametric or non-parametric 3

Figure 2: CO2 emissions intensity and energy efficiency of public power generation in the EU-24, Source: Eurostat. production function, assumes that the observed productions belong to the same production possibility set, which is a prerequisite for comparability. Given the diversity of power generating industries in the EU with respect to technology, fuel mix, extent of cogeneration, this would pose some classification problems for a parametric method. In addition, the nonparametric models are easy to compute and most of their statistical properties are well established through use of bootstrap methods3 . The main drawback is that their are very sensitive to outliers, and that noisy data are not allowed.

3

Methodological issues

3.1

DEA as a measure of relative efficiency

Consider a production process in which desirable and undesirable outputs are jointly produced by consuming inputs. Assume that x, y and u are the vectors of inputs, desirable outputs and undesirable outputs (pollutants), respectively. Conceptually, this production technology can be described as T = (x,y,u) : (x) can produce (y,u)

(1)

T provides a complete description of all technologically feasible production plans, that is, relationships between inputs, desirable outputs and pollutants. In production theory, T is assumed to be closed and bounded set, which guarantees the output closeness and implies that finite amounts of inputs can only produce finite amounts of outputs. In addition, inputs and desirable outputs are often assumed to be strongly disposable. To reasonably model the joint production of both desirable and undesirable outputs it is often considered that there exists null-joint production, and that inputs and desirable outputs in T are weakly disposable. The concept of null-jointness is used to model the idea that good and bad outputs are produced jointly. It implies that the only way to eliminate all the undesirable outputs is to cease the production process. Formally, ‘null jointness’ is defined as: if (x,y,u) ∈ T and u = 0, then y = 0 3

(2)

See Simar and Wilson (2008) for the comprehensive discussion about the statistical inference in nonparametric frontier models.

4

Weak disposability of outputs implies that the reduction of undesirable output is not free but the proportional reduction in both desirable and undesirable outputs is feasible. In other words it means that reduction of pollutants may not be a cost-free activity, but involves a cost which is measurable as reduction in good outputs. Formally, this assumption can be expressed as: if (x,y,u) ∈ T and 0 ≤ β ≤ 1, then (x,βy, βu) ∈ T (3) We formulate the above production technology within a non-parametric DEA framework. The resulting technology could therefore be termed as an environmental DEA technology. Assume that we observe a sample of K entities whose environmental efficiency and overall technical efficiency are to be measured, and for the kth entity the observed data on inputs, desirable and undesirable outputs are xk =(x1k ,. . . ,xN k ), yk =(y1k ,. . . ,yM k ) and uk =(u1k ,. . . ,uJk ), the environmental DEA technology exhibiting constant returns to scale can be expressed as T = (x,y,u) :

K X k=1 K X k=1 K X

λk ymk ≥ ym , m = 1, . . . , M λk ujk ≤ uj , j = 1, . . . , J

(4)

λk xnk ≤ xn , n = 1, . . . , N

k=1

λk ≥ 0, k = 1, . . . , K Note that the convexity constraint λk = 1 is not included in the above model, and that undesirable outputs are treated as strongly disposable inputs4 . There are few reasons for assuming non-CRS reference technology. Zhou et al. (2008) find that about half of the studies assessing energy and environmental issues assumed that the reference technology exhibits CRS, although non-CRS might be a more appropriate assumption. One possible reason is that output-orientated radial efficiency measure is just reciprocal of the input-orientated radial efficiency measure under the CRS assumption. As a result, the choice between input-orientated and output-orientated DEA model becomes indifferent. In addition, this could be partially explained by the popularity of Malmquist productivity index and the fact that the MPI based on the CRS assumptions can be interpreted as a total productivity index. Likewise, the choice of the returns to scale of reference technology depends on the object of the analysis. Coelli and Rao (2005) use a CRS technology in their productivity analysis in agriculture by using aggregate country-level data. The main argument of this choice is following (Coelli and Rao, 2005, p. 120): “given that the analysis involves the use of aggregate country-level data, it does not appear to be sensible to consider a VRS technology. That is, how it is possible for a sector to achieve scale economies? (...) The use of a VRS technology when the summary data are expressed on an “average per farm” basis might be sensible, since the scale economies of the “average per farm” could be discussed, but when dealing with aggregate data (as in the case of this study) the use of a CRS technology is the only sensible option.” Moreover, Yang and Pollitt (2009) note that non-CRS constraint might screen some of the effects of uncontrollable variables if regression analysis is used in the second stage. On the basis of (4), we introduce the following non-radial DEA-type programming model for calculating an environmental efficiency performance index: 4

Initially, the authors included a constraint for undesirable outputs with weak disposability. However, under this assumption, the results were the same as in the model with strong disposability of undesirable outputs.

5

EN V (x0 ,y0 ,u0 ) = min s.t.

1 J

J X

θj

j=1 K X

λk ymk ≥ ym0 , m = 1, . . . , M

k=1 K X k=1 K X

λk ujk ≤ θj uj0 , j = 1, . . . , J

(5)

λk xnk ≤ xn0 , n = 1, . . . , N

k=1

λk ≥ 0, k = 1, . . . , K where the subscript “0” represents the entity to be evaluated and λk is a vector of coefficients which represents the intensity levels for entities in the construction of the reference efficiency frontier. This model can be treated as a Russell-type DEA model in the context of environmental efficiency measurement. Examples of similar models in the context of environmental or energy efficiency measurement include Picazo-Tadeo and Garca-Reche (2007) and Zhou and Ang (2008). It can be seen that (5) attempts to non-proportionally adjust undesirable outputs as much as possible for a given level of inputs and desirable outputs. As a result, this programme allows some undesirable outputs to increase so that other undesirable outputs achieve larger reductions in order to reach its ideal benchmarking point in the frontier of the best practice. Since EN V is essentially the minimum average of the ratios of the expected undesirable outputs to the actual undesirable outputs, we may refer to EN V as an average environmental performance index. In this programme the input constraints guarantee that, at the optimum, entity 0 will make use of no fewer inputs than the efficient productive entity it is compared with. The desirable output constraints ensure that under its environmentally efficient production plan, entity 0 produces no more desirable outputs than the technological reference at the frontier, while the undesirable output constraints make sure that, at the optimum, entity 0 pollutes no less than the efficient productive entity it is compared with. The inequality constraints on the desirable-outputs side and on the undesirable-outputs side imply that these outputs are freely disposable. To measure the overall technical efficiency, we make use of the following input-oriented DEA model: ECON (x0 ,y0 ,u0 ) = min θ s.t.

K X k=1 K X k=1 K X

λk ymk ≥ ym0 , m = 1, . . . , M λk ujk ≤ θuj0 , j = 1, . . . , J

(6)

λk xnk ≤ θxn0 , n = 1, . . . , N

k=1

λk ≥ 0, k = 1, . . . , K (6) attempts to proportionally contract the amounts of inputs and undesirable outputs as much as possible for a given level of desirable outputs. (6) adopts radial efficiency measure, which provides a pure technical efficiency index for measuring efficiency performance of energy industries.

3.2

DEA as a measure of productivity

(6) considers performance analysis at a given point in time. Information on changes in technical efficiency over time only tells the “catch-up” part of the productivity story. Total factor productivity (TFP) change can also appear in the form of technical change (or frontier shift). The Malmquist Productivity Index provides performance analysis over a period of time and decomposes productivity

6

change into technical change and technical efficiency change. It measures the TFP change between two data points (e.g. those of a particular country in two adjacent periods) by calculating of each data point relative to a common technology. According to input-orientated production technology of energy generators, the current study employs the nonparametric input-oriented MPI. Input-orientation refers to the emphasis on the equiproportionate reduction of inputs within the context of a given level of outputs. Note that below the subscripts denoting a variety of inputs and outputs are excluded, and that the undesirable outputs are treated as inputs. The input-orientated Malmquist TFP change index between periods s (the base period) and period t is given by "

MtI (yt , xt , ys , xs )

DsI (yt , xt ) DtI (yt , xt ) = × DsI (ys , xs ) DtI (ys , xs )

#1/2

,

(7)

where the superscript I indicates an input-orientation, M is the productivity of the most recent production point (xt , yt ) (using period t technology) relative to the earlier production point (xs , ys ) (using period s technology), D are input distance functions, and all other variables are as previously defined. Values greater than unity indicate positive TFP growth between the two periods. Following F¨are et al. (1994), an equivalent way of writing this productivity index is "

DI (yt , xt ) DsI (yt , xt ) DsI (ys , xs ) MtI (yt , xt , ys , xs ) = It × Ds (ys , xs ) DtI (yt , xt ) DtI (ys , xs )

#1/2

,

(8)

where the ratio outside the square brackets measures the change in the input-oriented measure of technical efficiency between periods s and t. That is the efficiency change is equivalent to the ratio of the technical efficiency in period t to the technical efficiency in period s. The remaining part of the index in equation (8) is a measure of technical change or technical progress as measured by shifts in the frontier measured at period t and period s (the geometric mean of the two ratios in the square bracket). Following F¨ are et al. (1994), and given that suitable panel data are available, the required distance measures for the Malmquist TFP index are calculated using DEA-like linear programs. For the “0” country, four distance functions are calculated in order to measure the TFP change between two periods, s and t. This requires the solving of four linear programming problems. They are: [D0t (xt ,yt ,ut )]−1 = min θ s.t.

K X k=1 K X k=1 K X

t t λtk ymk ≥ ym0 , m = 1, . . . , M

λtk utjk ≤ θutj0 , j = 1, . . . , J

(9)

λtk xtnk ≤ θxtn0 , n = 1, . . . , N

k=1

λtk ≥ 0, k = 1, . . . , K [D0s (xs ,ys ,us )]−1 = min θ s.t.

K X k=1 K X k=1 K X k=1

s s λsk ymk ≥ ym0 , m = 1, . . . , M

λsk usjk ≤ θusj0 , j = 1, . . . , J λsk xsnk ≤ θxsn0 , n = 1, . . . , N

λsk ≥ 0, k = 1, . . . , K 7

(10)

[D0t (xs ,ys ,us )]−1 = min θ s.t.

K X

t s λtk ymk ≥ ym0 , m = 1, . . . , M

k=1 K X

λtk utjk ≤ θusj0 , j = 1, . . . , J

k=1 K X

(11)

λtk xtnk ≤ θxsn0 , n = 1, . . . , N

k=1 λtk ≥

0, k = 1, . . . , K

[D0s (xt ,yt ,ut )]−1 = min θ s.t.

K X k=1 K X k=1 K X

s t λsk ymk ≥ ym0 , m = 1, . . . , M

λsk usjk ≤ θutj0 , j = 1, . . . , J

(12)

λsk xsnk ≤ θxtn0 , n = 1, . . . , N

k=1 λsk ≥

0, k = 1, . . . , K

The four linear programming problems are to be calculated for each entity in the sample. Solution of (9)-(12) provides the input-orientated DEA measured of technical efficiency under CRS in period s and t, respectively, from which the Malmquist productivity index and its components can be calculated.

4 4.1

Data Variables used to measure environmental efficiency and TFP

The present data is based on data drawn from the Euromonitor International, the European Environment Agency (EEA), the Eurostat, and the International Energy Agency (IEA). The following are some of the main features of the data series used (see Table 1 for descriptive statistics). 1. Country and time coverage The study includes 24 EU member states. Cyprus, Luxembourg and Malta are excluded due to unavailability of some data series and due to the fact that these countries are relatively small compared to other countries in the sample. Results are presented for the period 1996 to 2007. 2. Output series Since we are interested in the environmental and productivity effects of controlling CO2 emissions by the most polluting industry in the EU ETS, we restrict our attention to public thermal power plants and district heating plants that mostly rely on combustible CO2 -intensive fuels, such as oil, solid fuels and gas. According to the Eurostat definitions, public thermal power stations generate electricity and/or heat for sale to third parties, as their primary activity. District heating plants produce heat used for process or space heating in any sector of economic activity including the residential sector. Only heat sold to third parties is included. Gross electricity generation of public thermal power plants is measured in Gigawatt hours (GWh) and in thousands tons of oil equivalent (TOE), and is taken from the Eurostat. Gross heat production from public thermal power plants and district heating plants is measured in terra joules (based on the net caloric value) (TJ(ncv)) and in TOE, and is taken from the Eurostat. Because electricity output and heat output are available in the same measurement unit (TOE), we aggregate them into one output variable. 3. Input series

8

Variable

Unit

Mean

Power generation Fuel Installed capacity Labour CO2 emissions SO2 emissions

Thousands tons of oil equivalent (TOE) Thousands tons of oil equivalent (TOE) Megawatts (MW) No. in thousands Thousands of tonnes Thousands of tonnes

7318 15346 15382 58 55214 241

Standard deviation 506 1128 1080 4 4245 18

Minimum

Maximum

569 996 569 5 1910 1

39679 89896 71072 255 345673 1361

No. obs. 288 288 288 288 288 280

Table 1: Descriptive statistics of outputs and inputs Given the constraints on the number of input variables that can be used in a DEA analysis, this analysis considers three groups of input variables: labour, fuel inputs, and net installed electrical capacity. Labour data refers to the economically active population in electricity, water and gas supply industry (ISIC-68, division 4), and is measured in thousands of employees. It is taken from the Euromonitor International. The more disaggregated data, which could correspond to the number of employees in public thermal power plants and district heating plants, is not available. Therefore, we have made some adjustment for the available labour data (see Appendix A). Fuel is measured in TOE, and includes all varieties of fuel utilised by the public power plants: coal, oil, gas and renewable fuels. The Eurostat provides a detailed disaggregation of each fuel category used for electricity and heat production. As fuel input data are available in the same measurement units, they can be aggregated into one indicator. Net installed electrical capacity of thermal power plants is measured in Megawatts (mw), and is taken from the Eurostat and the IEA. The Eurostat provides the net installed electrical capacity data for thermal power plants only, that is there is no separation between public thermal power plants and autoproducer thermal power plants. This distinction is made by the IEA, but it is available only for the OECD countries. (See an Appendix B for details of how the net installed electrical capacity of thermal power stations is allocated between public thermal power plants and autopoducer thermal power plants for the remaining non-OECD countries in the sample.) Generating capacity of power stations is used as a proxy for capital. However, some studies use capital expenditure to find out not only the generating capacity of a plant, but also the extent to which plants have invested in equipment to reduce pollution (Yaisawarng and Klein, 1994). Unfortunately, disaggregated capital expenditure or capital stock data are not available for electricity and heat industry alone. Likewise, it should be noted that the capacity data are not available for district heating plants. 4. Environmental variables Finally, CO2 emissions from public electricity and heat production are available from the EEA. According to the Revised 1996 IPPS Guidelines for National Greenhouse Gas Inventories, these emissions correspond to a sum of emissions from public electricity generation, public combined heat and power generation, and public heat plants. Public utilities are defined as those undertakings whose primary activity is to supply the public. They may be in public or private ownership. Emissions from own on-site use of fuel should be included. Emissions from autoproducers (undertakings which generate electricity/heat wholly or partly for their own use, as an activity which supports their primary activity) should be assigned to the sector where they were generated. Autoproducers may be in public or private ownership. CO2 are measured in thousands of tons and corresponds to the national emissions reported to the UNFCCC and to the Greenhouse Gas Monitoring Mechanism. When measuring environmental efficiency we also take into account SO2 emissions that influence climate change indirectly and hence they have to be reported to the UNFCC Secretariat. These are available from the EEA and are measured in thousands of tons.

9

of

4.2

Determinants of environmental efficiency and TFP

An attempt is made to analyse the determinants of environmental efficiency, TFP and its components through the use of econometric techniques. We group the variables thought to influence environmental efficiency and TFP change under five broad categories. 1. Environmental and energy policies “Climate policies affect the power sector in both short and long run. In the short run, these policies rearrange production merit order by forcing generators to take emission costs into consideration. In the long run, they are expected to drive the generation mix toward less carbon-intensive technologies” (Chen and Tseng, 2008, p. 31). A CO2 price of 20e/t adds roughly 40% to power generating costs for coal and 20% to power generating costs for gas (Graus and Worrell, 2009). The impact of carbon emission constrains depends heavily on the allocation of the EUAs. The windfall profits received from selling the surplus of the EUAs could encourage “stickiness” to carbon-intensive fuels and an inefficient use other resources. Therefore, we use the average annual CO2 market price as an indicator of the overall stringency of the EU ETS, while a ratio of the initial permit allocation to verified emissions is used to represent a country-level stringency of the EU ETS. 2. Fuel prices The fuel choice in public power plants responds to changes in fuel prices. Fuel costs account for approximately 40% of power generating costs for coal-fired power plants and 60% for gas-based power plants. Therefore high fuel prices, as high CO2 prices, are an incentive to both short-term fuel switching, providing that the capacity to switch between fuels exists (S¨oderholm, 2001), and implentation of new technologies. If public power sector has a possibility to switch to more energy efficient and less-dirty fuels in the short run, we could expect environmental efficiency and productivity improvement to realise spontaneously. While the gains from installing new power generating capacity might be postponed for few years. As the country-level fossil fuel price data is not available for all countries for the period of interest, we use the annual growth rates in the real market prices for coal, crude oil and natural gas. 3. Fossil fuel abundance Sachs and Warner (1995, 2001) analyse the impact of resource abundance on economic growth. Their hypothesis - the “resource curse” - is that growth decelerates in resource abundant countries. Hoffmann and Voigt (1905) analyse this hypothesis in hard coal fueled electricity generation. Their empirical results suggest that the more hard coal resources a country possesses, the less efficient is its electricity generation. As Hoffmann and Voigt (1905), we will use a ratio of dirty fossil fuel exports to total primary energy supply (TPES) to represent the fossil fuel abundance. Dirty fossil fuels include coal and coal products, peat, crude oil, natural gas liquids and feedstocks, and natural gas. TPES of a natural resource results from production of the respective resource plus imports, stock changes, and reserves stored in bunkers etc. minus exports. 4. Economic growth Overall economic growth, as an indicator for the velocity of economic progress, is supposed to affect both environmental efficiency and productivity in the sense that faster growing economies adopt technological innovations and thus implement more sophisticated and more efficient plants. Likewise fast economic growth encourages power generators to use their resources more efficiently in order to meet the increasing demand for power. Inflows of foreign direct investments (FDI) might lead to knowledge transfers and thus to knowledge spillovers. In the case of power generation this might initiate the introduction of more productive and more emissions and energy efficient technologies. Growth in productivity index is used to measure economic growth and a ratio of FDI inflows to gross domestic product is used to measure an extent of foreign knowledge transfers and spillovers. 5. Technological characteristics Dominance of specific power generation technology might also explain environmental and productivity differences in power generation across countries. Countries that specialise in thermal power generation could be more productive than countries that produce only a small share of power by thermal power plants. Also we expect that the higher share of cogeneration (CHP) leads to more 10

productive exploitation of resources. CHP typically convert 75-80% of the fuel source into useful energy. In contrast, in conventional separate electricty and heat generation, overall efficiency is only 60% (IEA, 2008). Likewise, the gas based power generation is more efficient in terms of emissions and energy than coal or oil based heat generation (Graus and Worrell, 2009). The share of electricity produced by public thermal power plants in total electricity generation is used as a measure of specialisation. In order to account for CHP generation we create a dummy variable, which is equal to 1 if a portion of heat of total power generated by public thermal power plants is more than a third, and 0 if it less than one third 5 . To measure the extent of gas-based power generation, we use a proportion of gas fuel in total fuel used in power generation by public power plants.

5

Results and discussion

5.1 5.1.1

Environmental efficiency Environmental efficiency scores

Before interpreting the results, it should be understood that “efficiency” is a relative concept: that is the efficient countries, with the efficiency score equal to 1, are the best in the sample. Accordingly, the environmental efficiency scores lower than 1 should be interpreted as measures of relative efficiency, as they were computed comparing each country in the sample with the countries with the best observed performance. Also, the influence of atypical observations should be tested. We have checked that efficiency scores do not depend upon one or two countries repeatedly enveloping other countries, but rather upon the set of efficient countries. In addition, we perform a sensitivity test which show that the results changed very little when the analysis was run with the efficient countries shaping the frontier being removed once at a time. Also, we followed the procedures suggested by Wilson (1995) which allows detecting whether an observation in the efficient subset is really efficient relative to other observations in the sample, and, at the same time, testing how much the presence of this observation in the efficient subset affect the measured efficiency of other observations. Environmental efficiency scores for CO2 and SO2 emissions and their average for period 1996 to 2007 are reported in Table 2, Table 3, and Table 4, respectively. The countries are listed according to the magnitude of average environmental efficiency scores. Figure 3 shows an average performance of these scores. The results suggest that substantial reductions in CO2 and SO2 emissions are possible in most of the countries. In 2007, the average CO2 and SO2 environmental efficiency is 0.42 and 0.37, respectively, implying that, on average, it is possible to reduce these air pollutants by a maximum of 58% and 63% while still maintaining inputs and desirable outputs. This potential of emissions reduction in fossil fuel based power generation is confirmed by the statements of the European Commission on that that the largest and cheapest GHG emission reductions can be achieved in power generating industry: “. . . the power generation sector is the one, at global scale, that would experience most of the fuel switch [. . . ] due to its relatively high technological flexibility (CEC, 2009, p. 65).” The environmental efficiency scores vary across countries reflecting fuel mix and technology used in power generation: the average relative environmental efficiency for gas-based (the share of gas in fuel mix is larger than 50% in 2007) power generation is almost twice as big as the relative environmental efficiency of coal-based (the share of solid fuels in fuel mix is larger than 50% in 2007) power generation (see Figure 4). Also, the countries, such as Austria, Denmark, Finland, Latvia and the Netherlands, shapping the frontier, rely on gas-based power generation together with a considerable 5

The available data does not give a possibility to distinguish between power produced by traditional power plants and CHP plants. However, it distinguishes the portions of electricity and heat produced by all (traditional and CHP) power plants. Since only CHP produce heat, we use its share in total production of power by all power plants as a proxy of the extent of CHP.

11

Austria Finland Latvia Netherlands Sweden Denmark Hungary Lithuania Italy Belgium Germany United Kingdom Slovakia Ireland Slovenia Czech Republic Portugal Poland Estonia Spain Romania Bulgaria France Greece Mean

Base fuel Other Coal Gas Gas Other Coal Gas Gas Gas Gas Coal Coal Coal Gas Coal Coal Coal Coal Coal Coal Coal Coal Other Coal

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

1.000 0.526 1.000 1.000 1.000 1.000 0.186 0.409 0.216 0.223 0.219 0.182 0.298 0.180 n/a 0.242 0.170 n/a 0.163 0.132 0.229 0.179 0.079 0.117 0.398

0.747 0.601 1.000 1.000 1.000 1.000 0.168 0.390 0.178 0.185 0.186 0.161 0.231 0.142 n/a 0.207 0.139 n/a 0.139 0.112 0.194 0.119 0.056 0.092 0.366

0.935 0.391 1.000 1.000 1.000 0.385 0.182 0.335 0.177 0.215 0.226 0.171 0.255 0.152 n/a 0.224 0.158 n/a 0.149 0.117 0.218 0.137 0.079 0.101 0.346

0.770 1.000 1.000 1.000 1.000 0.373 0.162 0.306 0.155 0.235 0.205 0.183 0.202 0.132 n/a 0.341 0.141 n/a 0.130 0.101 0.176 0.117 0.083 0.090 0.359

0.649 1.000 1.000 1.000 1.000 1.000 0.143 0.327 0.164 0.209 0.215 0.156 0.150 0.128 0.104 1.000 0.129 0.147 0.117 0.092 0.153 0.100 0.087 0.083 0.381

1.000 0.349 1.000 1.000 1.000 1.000 0.162 0.345 0.213 0.235 0.237 0.165 0.224 0.136 0.109 0.202 0.135 0.146 0.127 0.098 0.155 0.104 0.098 0.089 0.347

0.936 0.384 1.000 1.000 1.000 1.000 0.183 0.400 0.248 0.267 0.263 0.189 0.218 0.163 0.123 0.187 0.153 0.148 0.146 0.113 0.147 0.116 0.115 0.100 0.358

1.000 1.000 1.000 1.000 1.000 1.000 0.199 0.503 0.290 0.298 0.303 0.201 0.247 0.199 0.142 0.209 0.167 0.163 0.146 0.129 0.162 0.128 0.124 0.108 0.405

1.000 1.000 1.000 1.000 1.000 0.853 0.210 0.469 0.313 0.266 0.284 0.197 0.238 0.191 0.133 0.195 0.170 0.154 0.146 0.130 0.145 0.118 0.116 0.102 0.393

1.000 1.000 1.000 1.000 1.000 0.865 0.337 0.442 0.365 0.260 0.276 0.194 0.223 0.167 0.129 0.172 0.157 0.143 0.144 0.129 0.127 0.107 0.107 0.092 0.393

0.858 1.000 1.000 1.000 1.000 0.716 0.431 0.439 0.348 0.278 0.263 0.180 0.195 0.171 0.163 0.153 0.145 0.128 0.135 0.129 0.118 0.099 0.094 0.089 0.380

1.000 1.000 1.000 1.000 1.000 0.699 0.452 0.431 0.429 0.338 0.268 0.205 0.203 0.187 0.174 0.155 0.146 0.129 0.127 0.127 0.113 0.097 0.095 0.093 0.394

Table 2: Average environmental efficiency scores, 1996-2007

Austria Finland Latvia Netherlands Sweden Denmark Hungary Lithuania Italy Belgium Germany United Kingdom Slovakia Ireland Slovenia Czech Republic Portugal Poland Estonia Spain Romania Bulgaria France Greece Mean

Base fuel Other Coal Gas Gas Other Coal Gas Gas Gas Gas Coal Coal Coal Gas Coal Coal Coal Coal Coal Coal Coal Coal Other Coal

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

1.000 0.523 1.000 1.000 1.000 1.000 0.347 0.714 0.343 0.307 0.296 0.313 0.565 0.292 n/a 0.359 0.292 n/a 0.273 0.244 0.387 0.344 0.124 0.203 0.497

0.494 0.579 1.000 1.000 1.000 1.000 0.314 0.646 0.291 0.260 0.247 0.275 0.440 0.242 n/a 0.308 0.243 n/a 0.237 0.209 0.338 0.231 0.089 0.161 0.436

0.870 0.459 1.000 1.000 1.000 0.477 0.343 0.608 0.280 0.295 0.269 0.295 0.481 0.262 n/a 0.306 0.279 n/a 0.255 0.218 0.374 0.265 0.124 0.176 0.438

0.539 1.000 1.000 1.000 1.000 0.439 0.309 0.552 0.246 0.290 0.233 0.310 0.384 0.230 n/a 0.280 0.248 n/a 0.229 0.188 0.311 0.228 0.130 0.160 0.423

0.375 1.000 1.000 1.000 1.000 1.000 0.272 0.548 0.251 0.258 0.200 0.266 0.253 0.216 0.196 1.000 0.229 0.235 0.205 0.169 0.270 0.196 0.135 0.145 0.434

1.000 0.426 1.000 1.000 1.000 1.000 0.301 0.572 0.313 0.278 0.211 0.270 0.380 0.224 0.196 0.269 0.236 0.237 0.220 0.179 0.274 0.203 0.144 0.155 0.420

0.872 0.458 1.000 1.000 1.000 1.000 0.338 0.669 0.345 0.300 0.229 0.306 0.373 0.260 0.221 0.264 0.264 0.243 0.251 0.207 0.261 0.225 0.163 0.175 0.434

1.000 1.000 1.000 1.000 1.000 1.000 0.364 0.765 0.359 0.335 0.260 0.321 0.423 0.300 0.253 0.297 0.279 0.262 0.254 0.234 0.288 0.247 0.182 0.191 0.484

1.000 1.000 1.000 1.000 1.000 0.705 0.367 0.735 0.385 0.317 0.247 0.301 0.408 0.295 0.235 0.281 0.290 0.256 0.250 0.236 0.257 0.228 0.171 0.180 0.464

1.000 1.000 1.000 1.000 1.000 0.729 0.381 0.642 0.361 0.289 0.225 0.272 0.381 0.250 0.217 0.244 0.261 0.230 0.242 0.232 0.236 0.208 0.153 0.162 0.446

0.716 1.000 1.000 1.000 1.000 0.432 0.390 0.621 0.358 0.301 0.215 0.241 0.331 0.247 0.201 0.218 0.236 0.215 0.228 0.230 0.218 0.191 0.128 0.156 0.411

1.000 1.000 1.000 1.000 1.000 0.404 0.386 0.646 0.365 0.314 0.213 0.253 0.342 0.259 0.192 0.221 0.241 0.213 0.217 0.224 0.205 0.187 0.129 0.163 0.424

Table 3: Carbon dioxide efficiency scores, 1996-2007

12

Austria Finland Latvia Netherlands Sweden Denmark Hungary Lithuania Italy Belgium Germany United Kingdom Slovakia Ireland Slovenia Czech Republic Portugal Poland Estonia Spain Romania Bulgaria France Greece Mean

Base fuel Other Coal Gas Gas Other Coal Gas Gas Gas Gas Coal Coal Coal Gas Coal Coal Coal Coal Coal Coal Coal Coal Other Coal

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

1.000 0.529 1.000 1.000 1.000 1.000 0.025 0.103 0.090 0.140 0.143 0.051 0.031 0.067 n/a 0.125 0.048 n/a 0.054 0.020 0.070 0.013 0.033 0.031 0.299

1.000 0.622 1.000 1.000 1.000 1.000 0.021 0.135 0.065 0.110 0.124 0.046 0.023 0.042 n/a 0.107 0.034 n/a 0.041 0.015 0.051 0.008 0.023 0.023 0.295

1.000 0.323 1.000 1.000 1.000 0.294 0.020 0.062 0.075 0.136 0.183 0.048 0.029 0.042 n/a 0.141 0.036 n/a 0.042 0.017 0.062 0.008 0.033 0.027 0.254

1.000 1.000 1.000 1.000 1.000 0.307 0.016 0.060 0.064 0.181 0.178 0.056 0.021 0.033 n/a 0.403 0.034 n/a 0.032 0.015 0.042 0.006 0.036 0.020 0.296

0.924 1.000 1.000 1.000 1.000 1.000 0.015 0.106 0.076 0.160 0.230 0.046 0.047 0.040 0.012 1.000 0.030 0.059 0.029 0.015 0.036 0.005 0.040 0.021 0.329

1.000 0.273 1.000 1.000 1.000 1.000 0.023 0.119 0.113 0.192 0.264 0.061 0.067 0.049 0.023 0.135 0.033 0.054 0.035 0.016 0.036 0.006 0.051 0.023 0.274

1.000 0.311 1.000 1.000 1.000 1.000 0.027 0.130 0.150 0.234 0.296 0.073 0.063 0.067 0.024 0.110 0.042 0.054 0.040 0.019 0.033 0.007 0.066 0.025 0.282

1.000 1.000 1.000 1.000 1.000 1.000 0.034 0.240 0.222 0.261 0.347 0.081 0.071 0.098 0.030 0.120 0.056 0.063 0.039 0.023 0.036 0.008 0.066 0.025 0.326

1.000 1.000 1.000 1.000 1.000 1.000 0.054 0.204 0.241 0.215 0.320 0.093 0.067 0.087 0.031 0.108 0.049 0.053 0.042 0.024 0.033 0.008 0.062 0.024 0.321

1.000 1.000 1.000 1.000 1.000 1.000 0.293 0.242 0.369 0.231 0.327 0.116 0.065 0.084 0.041 0.100 0.053 0.056 0.046 0.026 0.019 0.007 0.060 0.022 0.340

1.000 1.000 1.000 1.000 1.000 1.000 0.472 0.256 0.338 0.255 0.310 0.119 0.059 0.095 0.126 0.089 0.053 0.041 0.042 0.028 0.018 0.007 0.059 0.022 0.349

1.000 1.000 1.000 1.000 1.000 0.994 0.517 0.215 0.492 0.362 0.323 0.156 0.064 0.116 0.157 0.089 0.051 0.045 0.036 0.029 0.021 0.008 0.061 0.024 0.365

Table 4: Sulphur dioxide environmental efficiency scores, 1996-2007

Figure 3: Mean environmental efficiency scores

13

Figure 4: Mean environmental efficiency scores for coal-based and gas-based power generation share of renewable fuels. Among them Sweden is an exception since its public power generation is mainly based on renewable fuels (74% in 2007). The average environmental efficiency of gas-based power generation has been increasing since 19966 . It visibly improved in the years of the first trading period of the EU ETS, while the oposite is true for coal-based power generating industries. To what extent the latest developments in environmental efficiency might be attributed to the EU ETS? 10 out of 14 coal-based power generating industries had a net long positions of EAUs. On average, the environmental efficiency of these countries deteriorated in 2005-2007 (Figure 5). In none of these countries environmental efficiency improved. Denmark experienced the largest decrease in CO2 efficiency. This drop might be associated with an increase of coal use in power generation: the share of it increased from 50% in 2005 to 60% in 2007. The deterioration of the environmenal efficiency scores in the first trading period migth suggest that the net long posistions of the EUAs is a consequence of the overallocation rather than emissions abatement. On average, the environmental efficiency of the remaining 4 coal-based countries (Greece, Sloevenia, Spain and UK) with the net short positions of the EUAs somewhat improved but most of this improvement occured in SO2 efficiency rather than CO2 efficiency suggesting that there was a marginal switch from coal to gas. Another implication of this result is that a portion of the shortage could be covered by buying the EUAs on the market. The somewhat similar pattern might be observed in the environmental efficiency for gas-based power generating countries with the net long positions of EUAs: the environmental efficiency for these countries did not improve in 2005-2007, on average. For the countries with the short positions (Belgium, Ireland and Italy) we observe a slight improvement. However, it occurred only in 2007 after a slight deterioration in 2006. Again, this drop might be supported by the low price of CO2 , while the improvement by the movement to less emission intensive fuel-mix. Austria, France and Sweden fall neither under the solid-fuel based power generation category nor under gas-based. Austria and Sweden are ones of the most efficient countries in the sample in 2007. Austria uses one third of renewable fuels in power generation, while in Sweden this share amounts to more than two thirds. France’s power generation ranks as the most environmentally inefficient after Greece. This might be explained by the fact that the share of power produced by public power plants is very low in France, and this might suggest that benefits from environmental efficiency improvement would be relatively low compared to costs. 6 A direct comparison between years is not straightforward, since efficiency scores are relative to the best performing power generating industries in each year. The fact that some of the countries remain on the frontier over the period of interest let us do some comparison like this.

14

Figure 5: Mean environmental efficiency scores for coal-based and gas-based power generation with short and long positions of EUAs 5.1.2

What factors explain environmental performance differences between power generating industries?

The regression model was estimated to investigate the determinants of the aggregate environmental efficiency. Until recently a common practice to analyse such relationship was to employ Tobit regression. However, Simar and Wilson (2007) have shown that DEA efficiency coefficients are biased and serially correlated in a complicated, unknown way, which makes Tobit estimators not appropriate methods for inference. Instead they propose to use bootstrapped truncated regression and show its good performance in Monte Carlo experiments. Here we follow this approach by applying the bootstrapping procedure to improve on inference. After inverting the environmental efficiency estimates obtained in the first stage, we perform the truncated regression analysis7 . Table 5 displays the estimated parameters and their confidence intervals computed according to the single bootstrapping Algorithm 1 of Simar and Wilson (2007). We analyse whether fossil fuel abundance and fuel prices, technological characteristics and the EU ETS have influenced countries’ environmental efficiency. We also include dummies for countries as they are shown to be significantly different from zero after performing Wald test8 . With a 99% confidence level, the price of CO2 emissions statistically and positively influences the environmental efficiency. This is in line with our expectations and confirms the findings of the earlier studies mentioned above. The effect of overallocation, which decreases incentives to reduce emissions, is reflected in a significant coefficient for the ratio of allocated EUAs to verified EUAs. The coefficient for interaction between the price of crude oil and the share of oil used in power generation is negative implying that the increase in crude oil price have the higher negative impact on the environmental efficiency the larger the share of oil-based power generation. This might suggest that a major switch from oil based power generation to less carbon-friendly fuels happens only in the countries where the share of oil-based power generation is solid. In most of the EU countries, the 7

Then the environmental efficiency scores range from one to positive infinity. The larger the value of the transformed environmental efficiency score the smaller the firm’s environmental efficiency. 8 Note that Stata does not allow to perform truncated regression for panel data.

15

Estimated Dep. Var. Environmental efficiency (inverse) parameter

Constant CO2 price Allocation to verification ratio Coal price (log difference)*Coal share Crude oil price (log difference)*Crude oil share Natural gas price (log difference)*Natural gas share Coal price (log difference) Crude oil price (log difference) Natural gas price (log difference) Export-TPES ratio (lagged) Specialisation CHP dummy Coal share Crude oil share Natural gas share

Confidence intervals

4.836 -0.041 0.004 -0.001 0.055

99% lower -4.355 -0.057 0.002 -0.011 0.032

upper 19.126 -0.025 0.006 0.010 0.078

95% lower 2.200 -0.054 0.003 -0.009 0.038

upper 18.385 -0.029 0.006 0.007 0.073

-0.006

-0.021

0.008

-0.018

0.005

-0.108 0.042 1.206 -0.050 -0.032 -0.661 -0.010 -0.013 -0.097

-0.846 -0.430 0.738 -0.072 -0.049 -1.112 -0.041 -0.047 -0.129

0.596 0.475 1.692 -0.034 -0.018 -0.300 0.020 0.016 -0.067

-0.664 -0.309 0.830 -0.064 -0.046 -1.026 -0.035 -0.035 -0.121

0.421 0.380 1.598 -0.037 -0.020 -0.342 0.012 0.008 -0.076

Table 5: Determinants of environmental efficiency. Note: ’truncreg’ command in Stata does not produce R2 , thus, we compute a rough estimate of the degree of association by correlating the dependent variable with the predicted value and squaring the result. The calculated value of 0.88 is a rough estimate of the R2 . share of oil-based power generation has been declining and in most of the cases the oil-based power generating capacity is utilised as a peak-load capacity rather than base-load capacity. In terms of fuel prices, only natural gas price has a negative and significant effect on the environmental efficiency. The increasing price of natural gas might encourage power generation to switch to cheaper but more carbon intensive fuels. The estimated parameter for resource abundance has a significant and positive sign - opposite than expected - meaning that the more fossil fuel resource abundant country is the more environmentally efficient it is. It might suggest that fuel abundant countries utilise more advanced power generation technologies in term of emissions. Likewise, fuel abundant and countries might have more pressures from the society to upgrade their technologies if it is clear that fossil fuel based power generation will be brought in the future. For instance, in Denmark CO2 and SO2 intensity of power production has been decreasing due to national efforts. In 1992 the carbon tax and in 1996 the sulphur tax were introduced to meet the national environmental goals, also, the subsidy scheme was introduced to create the incentives for utilisation of renewable energy in power sector. In Germany the SO2 emission reductions have been mainly due to the extensive use of flue gas desulphurisation technologies. The coefficient for specialisation in thermal power production is significant and positive. This is in line with our expectations that the thermal power production is more efficient the larger its share in the overall country’s power production. We can illustrate this argument with the following example. In France, where most of the power is produced by the nuclear power plants and the share of electricity produced by thermal power plants is around 7% (EU-24 average is near 60%), the environmental efficiency of public thermal power generation is the lowest after Greece in the EU-24. The low dependence on thermal power generation does not provide sufficient incentives to improve the performance of thermal power plants at national level. The remaining coefficients for the natural gas share and the dummy for cogeneration are positive and significant in explaining the environmental efficiency scores. The countries with the large share of gas-based power generation and cogeneration are more environmentally efficient.

16

Finland Latvia Sweden Lithuania Denmark Czech Republic Poland Austria Slovakia Netherlands Estonia Hungary Germany Belgium Slovenia Romania Bulgaria United Kingdom Italy Portugal Ireland Greece Spain France Mean

1996 0.930 1.000 1.000 0.843 1.000 0.800 0.766 0.621 0.582 0.625 0.637 0.522 0.665 0.575 0.561 0.628 0.732 0.517 0.439 0.426 0.539 0.529 0.387 0.510 0.660

1997 0.961 1.000 1.000 0.830 1.000 0.890 0.817 0.573 0.521 0.738 0.702 0.564 0.731 0.622 0.671 0.651 0.650 0.581 0.491 0.441 0.637 0.611 0.506 0.474 0.694

1998 0.974 1.000 1.000 0.755 0.851 0.837 0.792 0.561 0.569 0.691 0.608 0.585 0.677 0.566 0.700 0.620 0.595 0.516 0.430 0.503 0.577 0.624 0.476 0.527 0.668

1999 1.000 1.000 1.000 0.767 0.927 0.920 0.871 0.595 0.618 0.788 0.692 0.690 0.748 0.678 0.772 0.606 0.646 0.655 0.488 0.701 0.721 0.682 0.586 0.511 0.736

2000 1.000 1.000 1.000 0.762 1.000 1.000 0.864 0.615 0.678 0.859 0.726 0.679 0.797 0.715 0.758 0.612 0.658 0.703 0.488 0.668 0.771 0.777 0.659 0.479 0.761

2001 0.983 1.000 1.000 0.745 1.000 0.976 0.832 0.647 0.796 0.812 0.801 0.631 0.759 0.608 0.647 0.599 0.542 0.632 0.518 0.600 0.733 0.774 0.549 0.430 0.734

2002 0.972 1.000 1.000 0.780 1.000 0.831 0.745 0.608 0.735 0.670 0.720 0.548 0.605 0.611 0.588 0.560 0.456 0.539 0.452 0.603 0.520 0.626 0.513 0.446 0.672

2003 1.000 1.000 1.000 0.832 0.931 0.811 0.717 0.669 0.754 0.705 0.726 0.599 0.626 0.583 0.592 0.582 0.526 0.558 0.460 0.487 0.526 0.568 0.422 0.416 0.670

2004 1.000 1.000 1.000 0.792 0.801 0.858 0.786 0.698 0.776 0.744 0.730 0.610 0.654 0.595 0.611 0.575 0.537 0.551 0.539 0.542 0.541 0.610 0.437 0.402 0.683

2005 1.000 1.000 1.000 0.786 0.765 0.928 0.858 0.756 0.802 0.829 0.848 0.619 0.770 0.617 0.678 0.581 0.561 0.579 0.538 0.680 0.536 0.631 0.453 0.489 0.721

2006 1.000 1.000 1.000 0.870 0.786 0.817 0.747 0.684 0.749 0.722 0.688 0.641 0.673 0.624 0.588 0.554 0.705 0.549 0.534 0.500 0.508 0.592 0.457 0.337 0.680

2007 1.000 1.000 1.000 0.852 0.821 0.742 0.727 0.718 0.711 0.708 0.687 0.627 0.620 0.605 0.592 0.578 0.549 0.537 0.537 0.527 0.515 0.512 0.460 0.359 0.666

Table 6: Technical efficiency scores, 1996-2007

5.2

Overall technical efficiency

Before measuring total factor productivity, based on the Eq. 6, we calculate the technical efficiency scores for each country for the 1996-2007 period which are reported in Table 6 and sorted in the descending order according to the size of the efficiency scores in 2007. In 2007, the average technical efficiency score is 0.67 meaning that, on average, the EU public power sector can reduce its inputs by 33% while still maintaining energy production level. In 2007, Finland, Latvia and Sweden were shapping the frontier. The ranking of the technical efficiency scores differs from the ranking of the environmental efficiency scores meaning that while some countries are more efficient in terms of emissions they might be less efficient in terms of overall utilisation of inputs and vice versa 9 . For instance, Czech Republic ranks 16th in terms of the environmental performance, but 6th in terms of the overall technical efficiency. As changes in the technical efficiency only tell the “catch-up” part of the productivity story, and the technical change (or frontier shift) is another source of TFP change, we explore the measures of TFP change, technical efficiency change and technical change.

5.3 5.3.1

Malmquist Productivity Index (MPI) MPI and its components

Table 7, Table 8 and Table 9 present annual TFP change, technical efficiency change (hereafter efficiency change, EC) and technical change (TC), respectively, for all countries during the 19972007 period. The countries in all tables are presented in the descending order of the magnitude of the cumulative TFP changes. The confidence intervals derived from the bootstrap (according the proposed bootstrap algorithm of Simar and Wilson (1999)) show that these changes are significant 9

It should be noted that the technical efficiency scores do not take into account SO2 emissions.

17

Finland Sweden Slovakia Portugal Spain Netherlands Italy Belgium Greece Austria United Kingdom Germany Estonia Slovenia Hungary Ireland Czech Republic Lithuania France Poland Bulgaria Romania Latvia Denmark Mean

Base fuel Coal Other Coal Coal Coal Gas Gas Gas Coal Other Coal Coal Coal Coal Gas Gas Coal Gas Other Coal Coal Coal Gas Coal

9796 0.944* 1.030* 0.915* 1.010* 1.159* 1.048* 1.003* 0.979* 1.014* 0.963* 0.992* 0.968* 0.993* 1.075* 1.014* 1.038* 0.983* 1.019* 0.926* 0.956* 0.783 1.063* 1.006* 0.886* 0.990

9897 0.955* 1.064* 1.133* 1.234* 1.040* 1.078* 1.002* 1.008* 1.091* 1.040* 1.029* 1.041* 0.919* 1.115* 1.039* 1.012* 0.974* 0.956* 1.019* 0.979* 0.968* 0.997* 0.936* 0.817* 1.019

9998 1.111* 0.998* 0.957* 1.221* 1.140* 0.991* 0.990* 1.124* 0.957* 1.012* 1.104* 0.961* 0.998* 0.963* 1.043* 1.088* 0.966* 0.967* 1.036* 0.968* 0.952* 0.930* 0.901* 1.044* 1.018

0099 1.001* 1.018* 1.037* 0.925* 1.041* 1.013* 0.940* 1.003* 1.131* 0.975* 1.002* 1.005* 1.010* 0.959* 0.925* 1.011* 1.092* 0.979* 0.950* 0.972* 0.974* 1.000* 0.961* 1.086* 1.000

0100 1.134* 1.103* 1.282* 0.968* 0.933* 1.054* 1.175* 0.966* 1.045* 1.141* 1.000* 1.052* 1.172* 0.910* 1.033* 1.045* 1.004* 1.000* 0.985* 1.024* 0.897* 0.999* 1.080* 1.067* 1.045

0201 1.060* 1.053* 0.992* 1.142* 1.132* 1.003* 1.049* 1.114* 0.943* 0.971* 1.030* 0.960* 1.041* 1.057* 0.946* 0.854* 0.956* 1.021* 1.040* 0.970* 0.979* 0.959* 0.949* 1.012* 1.010

0302 1.213* 0.950* 1.036* 0.807* 0.814* 1.019* 0.954* 0.956* 0.980* 1.035* 1.000* 1.071* 1.079* 1.033* 1.057* 0.996* 1.052* 1.012* 1.002* 1.034* 1.091* 0.992* 1.036* 1.083* 1.013

0403 0.972* 1.029* 0.979* 1.078* 1.019* 1.034* 1.146* 1.003* 1.000* 1.009* 0.967* 1.014* 0.935* 1.000* 0.974* 1.005* 0.981* 0.969* 1.004* 1.008* 0.988* 0.981* 0.935* 0.864* 0.996

0504 0.887* 1.027* 0.986* 1.164* 1.044* 1.029* 0.979* 0.987* 0.917* 1.064* 0.993* 1.034* 1.039* 1.007* 1.023* 0.940* 0.968* 0.983* 1.113* 0.994* 1.013* 1.014* 1.014* 0.942* 1.007

0605 1.205* 1.018* 0.944* 0.854* 0.976* 0.936* 1.000* 1.026* 0.998* 0.979* 0.986* 0.983* 0.981* 1.010* 0.989* 1.034* 0.977* 1.038* 0.906* 1.019* 1.037* 0.967* 1.034* 1.096* 1.00

0706 1.018* 0.991* 1.047* 0.950* 0.998* 1.018* 0.997* 1.031* 1.093* 0.951* 1.022* 1.032* 0.951* 0.979* 1.019* 0.985* 1.039* 1.026* 0.985* 0.972* 1.246* 0.961* 0.986* 0.956* 1.011

Table 7: TFP annual change and cumulative change for the 24 countries. Note: * denotes significant differences from unity at 0.01. for most of the countries. From Figure 6 we can see that, on average, TFP change in the European public power generating industry was positive. The average cumulative growth across all countries is around 10% or 0.8% per annum. Both, EC and TC almost equaly contributed to this improvement. Finland, Sweden, Slovakia, Portugal and Spain have experienced the biggest growth in TFP since 1996. It should be noted that these countries (excep Sweden) rely on coal-based public power generation. Figure 7 reveals that most of TFP improvement occurred in the old EU member states10 (the average cumulative growth of 17%) rather than in the new MS (on average, TFP did not increase in these countries). From one side, this might reflect the fact that in some of the new MS public power generation is based on gas (Hungary, Latvia, Lithuania) meaning that these countries are already on or close to the frontier. On the other hand, this might mirror the different structure and developments of power market between the old MS and the new MS. For instance, in Lithuania, the major energy market’s liberalisation has not been started yet. The regulated power market might not provide enough incentives to increase productivity. In the old EU MS TFP change has been mainly driven by technical change. The remarkable growth in TFP in these countries occurred during 1998-2002. TFP change and its components show an interesting pattern in 1999-2002. This pattern is especially strong for the old EU MS. The efficiency change increased in 1999 and 2000, while, at the same time, the technological regress occurred. Technological change was positive and coincided with a reduction in efficiency change in 2001 and 2002 . These dynamics might be explained by the development of fossil fuel prices alone (see Figure 8). In 1999, the price of crude oil increased and it became more cost efficient to produce power using gas. However, this adjustment did not occur instantaneously. Power generating industry continued power production using oil. The high price of oil ensured more efficient production. It could even happen that an oil-fired plant had to be removed 10

By old MS we mean the EU MS which were in the EU before 2004.

18

Cumulative 1996-2007 1.547 1.308 1.293 1.277 1.277 1.238 1.231 1.199 1.156 1.135 1.124 1.120 1.098 1.095 1.055 0.989 0.985 0.966 0.949 0.898 0.870 0.867 0.837 0.818 1.097

Finland Sweden Slovakia Portugal Spain Netherlands Italy Belgium Greece Austria United Kingdom Germany Estonia Slovenia Hungary Ireland Czech Republic Lithuania France Poland Bulgaria Romania Latvia Denmark Mean

Base fuel Coal Other Coal Coal Coal Gas Gas Gas Coal Other Coal Coal Coal Coal Gas Gas Coal Gas Other Coal Coal Coal Gas Coal

9796 1.033* 1.000* 0.895* 1.035* 1.308* 1.180* 1.119* 1.082* 1.155* 0.922* 1.123* 1.099* 1.102* 1.198* 1.082* 1.180* 1.112* 0.985* 0.930* 1.067* 0.888* 1.036* 1.000* 1.000* 1.064

9897 1.014* 1.000* 1.093* 1.141* 0.941* 0.937* 0.874* 0.910* 1.022* 0.979* 0.888* 0.927* 0.866* 1.042* 1.037* 0.907* 0.940* 0.909* 1.112* 0.969* 0.916* 0.952* 1.000* 0.851* 0.968

9998 1.027* 1.000* 1.085* 1.394* 1.230* 1.140* 1.137* 1.198* 1.094* 1.060* 1.269* 1.104* 1.138* 1.103* 1.180* 1.249* 1.100* 1.017* 0.970* 1.099* 1.085* 0.977* 1.000* 1.089* 1.114

0099 1.000* 1.000* 1.098* 0.953* 1.124* 1.091* 0.998* 1.054* 1.139* 1.034* 1.074* 1.065* 1.050* 0.982* 0.983* 1.070* 1.087* 0.993* 0.936* 0.993* 1.019* 1.011* 1.000* 1.079* 1.035

0100 0.983* 1.000* 1.173* 0.898* 0.834* 0.945* 1.063* 0.852* 0.997* 1.052* 0.898* 0.953* 1.103* 0.854* 0.929* 0.951* 0.976* 0.978* 0.898* 0.962* 0.823* 0.978* 1.000* 1.000* 0.962

0201 0.988* 1.000* 0.924* 1.005* 0.935* 0.825* 0.871* 1.004* 0.809* 0.939* 0.853* 0.797 0.900* 0.909* 0.869* 0.709 0.852* 1.047* 1.039* 0.896* 0.842* 0.935* 1.000* 1.000* 0.914

0302 1.029* 1.000* 1.026* 0.807 0.821* 1.052* 1.019* 0.953* 0.907* 1.101* 1.036* 1.035* 1.008* 1.008* 1.092* 1.012* 0.976* 1.066* 0.932* 0.961* 1.152* 1.040* 1.000* 0.931* 0.999

0403 1.000* 1.000* 1.029* 1.113* 1.036* 1.055* 1.171* 1.022* 1.074* 1.043* 0.987* 1.044* 1.005* 1.032* 1.020* 1.028* 1.058* 0.953* 0.966* 1.097* 1.022* 0.988* 1.000* 0.860* 1.025

0504 1.000* 1.000* 1.034* 1.254* 1.037* 1.114* 0.998* 1.036* 1.034* 1.082* 1.051* 1.178* 1.161* 1.109* 1.014* 0.989* 1.081* 0.992* 1.218* 1.092* 1.044* 1.009* 1.000* 0.955* 1.062

0605 1.000* 1.000* 0.887* 0.775 1.015* 0.854* 0.997* 0.981* 0.812 0.950* 0.927* 0.806 0.810* 0.873* 1.014* 0.963* 0.799* 1.084* 0.733 0.847* 0.979* 0.995* 1.000* 1.073* 0.924

0706 1.000* 1.000* 1.053* 0.950* 0.993* 1.020* 0.995* 1.031* 1.155* 0.952* 1.023* 1.085* 1.001* 0.994* 1.022* 0.986* 1.102* 1.021* 0.940* 1.028* 1.284* 0.959* 1.000* 0.957* 1.023

Cumulative 1996-2007 1.075 1.000 1.287 1.175 1.179 1.156 1.216 1.085 1.119 1.100 1.062 1.012 1.080 1.049 1.229 0.942 1.021 1.032 0.662 0.974 0.963 0.882 1.000 0.786 1.045

Table 8: Efficiency annual change and cumulative change for the 24 countries. Note: * denotes significant differences from unity at 0.01.

Finland Sweden Slovakia Portugal Spain Netherlands Italy Belgium Greece Austria United Kingdom Germany Estonia Slovenia Hungary Ireland Czech Republic Lithuania France Poland Bulgaria Romania Latvia Denmark Mean

Base fuel Coal Other Coal Coal Coal Gas Gas Gas Coal Other Coal Coal Coal Coal Gas Gas Coal Gas Other Coal Coal Coal Gas Coal

9796 0.915* 1.030* 1.023* 0.976* 0.886* 0.888* 0.896* 0.905* 0.878* 1.045* 0.883* 0.881* 0.901* 0.898* 0.937* 0.880* 0.884* 1.034* 0.995* 0.896* 0.881* 1.026* 1.006* 0.886* 0.935

9897 0.942* 1.064* 1.036* 1.082* 1.106* 1.151* 1.146* 1.108* 1.067* 1.062* 1.160* 1.123* 1.061* 1.070* 1.002* 1.116* 1.036* 1.052* 0.916* 1.011* 1.057* 1.047* 0.936* 0.961* 1.055

9998 1.082* 0.998* 0.883* 0.876* 0.927* 0.870* 0.871* 0.938* 0.875* 0.955* 0.869* 0.871* 0.877* 0.874* 0.885* 0.871* 0.878* 0.951* 1.068* 0.881* 0.877* 0.952* 0.901* 0.958* 0.916

0099 1.001* 1.018* 0.944* 0.971* 0.926* 0.929* 0.942* 0.952* 0.993* 0.943* 0.933* 0.943* 0.962* 0.977* 0.940* 0.945* 1.005* 0.987* 1.014* 0.979* 0.956* 0.989* 0.961* 1.007* 0.967

0100 1.154* 1.103* 1.093* 1.078* 1.119* 1.116* 1.105* 1.134* 1.049* 1.085* 1.113* 1.104* 1.063* 1.066* 1.111* 1.098* 1.029* 1.022* 1.097* 1.065* 1.090* 1.022* 1.080* 1.067* 1.086

0201 1.073* 1.053* 1.074* 1.136* 1.212* 1.215* 1.204* 1.109* 1.165* 1.034* 1.208* 1.206* 1.157* 1.163* 1.089* 1.205* 1.123* 0.975* 1.001* 1.082* 1.163* 1.025* 0.949* 1.012* 1.110

0302 1.178* 0.950* 1.010* 0.999* 0.991* 0.968* 0.937* 1.003* 1.080* 0.940* 0.965* 1.035* 1.070* 1.025* 0.968* 0.984* 1.078* 0.949* 1.075* 1.076* 0.947* 0.954* 1.036* 1.162* 1.016

0403 0.972* 1.029* 0.952* 0.969* 0.984* 0.980* 0.979* 0.982* 0.931* 0.967* 0.980* 0.971* 0.930* 0.969* 0.956* 0.978* 0.927* 1.017* 1.039* 0.919* 0.967* 0.994* 0.935* 1.005* 0.972

0504 0.887* 1.027* 0.954* 0.928* 1.007* 0.923* 0.981* 0.953* 0.887* 0.983* 0.945* 0.878* 0.895* 0.907* 1.009* 0.950* 0.896* 0.991* 0.914* 0.910* 0.970* 1.005* 1.014* 0.986* 0.950

0605 1.205* 1.018* 1.064* 1.102* 0.962* 1.096* 1.003* 1.046* 1.229* 1.030* 1.064* 1.220* 1.211* 1.158* 0.975* 1.074* 1.222* 0.957* 1.235* 1.204* 1.059* 0.972* 1.034* 1.021* 1.090

0706 1.018* 0.991* 0.995* 0.999* 1.005* 0.997* 1.002* 1.000* 0.946* 0.999* 0.999* 0.951* 0.950* 0.985* 0.997* 0.999* 0.943* 1.005* 1.047* 0.946* 0.970* 1.002* 0.986* 0.999* 0.989

Cumulative 1996-2007 1.439 1.308 1.005 1.087 1.083 1.071 1.012 1.105 1.033 1.032 1.058 1.107 1.017 1.044 0.859 1.049 0.964 0.936 1.433 0.922 0.903 0.983 0.837 1.042 1.055

Table 9: Technical annual change and cumulative change for the 24 countries. Note: * denotes significant differences from unity at 0.01. 19

Figure 6: Mean cumulative TFP change, efficiency change and technical change

Figure 7: Mean cumulative TFP change, efficiency change and technical change for new and old MS from service during its conversion to gas-based power plants. During this time the less efficiency capacity had to be utilised, and that could explain the technical regress in 1999 and 2000. After this oil price spike, the share of oil-based power production decreased from 16% in 1998 to 10% in 2002, while the share of gas increased from 24% to 32%. However, increased gas prices in 2005 have led to higher utilisation of existing coal plants in some EU countries (EEA, 2008a). This is reflected in technical change decrease in 2005, which is very visible for coal-based power generating countries (see Figure 9). This return to coal-based power production could be “facilitated” by the surplus of the EUAs. An increase in technical change in 2006 might be attributed to oil and gas price increases and to the existence of the EU ETS. However, the dynamics of TFP change and its components, differently than in the case of environmental efficiency, do not differ much between the countries with the net long positions of the EUAs and the countries with the net short positions. Only gas-based power industries with the net short positions of EUAs experienced a positive TFP change in 2006 and 2007, while, at the same time, the technical regress occurred for the gas-based industries with EUA surplus (see Figure 10). When we look at the annual TFP changes country by country, we observe that TFP growth was positive for the countries with the net short positions of the EUAs. For instance, in Belgium, the annual TFP growth was 2.6% and 3.1% in 2006 and 2007, respectively. In 2006, this growth was driven by the TC, while, in 2007, by the efficiency improvements.

20

Figure 8: Fossil fuel prices, levels and annual changes. Source: Euromonitor International.

Figure 9: Mean cumulative TFP change, efficiency change and technical change for coal-based and gas-based power production

Figure 10: Mean cumulative TFP change, efficiency change and technical change for coal-based and gas-based power production

21

CO2 price Allocation to verification ratio Coal price (log difference)*Coal share Crude oil price (log difference)*Crude oil share Natural gas price (log difference)*Natural gas share Coal price (log difference) Crude oil price (log difference) Natural gas price (log difference) Production index (log difference, lagged) FDI inflow share (lagged) Export-TPES ratio (lagged) Specialisation Coal share Crude oil share Natural gas share CHP dummy Constant No. of obs. No. of countries R-squared F-test (p-value)

Model (1) TFP change 0.0006 -0.0003*** -0.0010 -0.0009 0.0011 0.0315 0.0039 -0.0331 0.3225** 0.0024* 0.0032*** 0.0078*** 0.0005 0.0010 0.0002 -0.0464** -0.4417** 181 21 0.28 0.000

Robust s.e.

Model (2) EC change

Robust s.e.

0.001 0.000 0.001 0.001 0.001

-0.0023** 0.0000 0.0004 -0.0011 0.0006

0.001 0.000 0.001 0.002 0.001

0.044 0.045 0.050 0.135 0.001 0.001 0.002 0.002 0.002 0.001 0.024 0.164

-0.0088 0.1986*** -0.0034 0.6243*** -0.0024 0.0018* 0.0071** 0.0005 0.0015 -0.0003 -0.1265*** -0.3672 181 21 0.26 0.000

0.056 0.058 0.056 0.208 0.002 0.001 0.003 0.002 0.002 0.002 0.040 0.237

Model (3) TC change 0.0029*** -0.0002** -0.0013*** 0.0002 0.0005 0.0407** -0.1951*** -0.0302 -0.3035** 0.0047** 0.0013*** 0.0008 0.0000 -0.0005 0.0005 0.0802*** -0.0743 181 21 0.33 0.000

Robust s.e. 0.001 0.000 0.000 0.002 0.001 0.022 0.054 0.039 0.164 0.002 0.000 0.002 0.001 0.002 0.002 0.023 0.153

Table 10: Determinants of TFP change and its components. Note: Standard errors are displayed in italics. Variables indicated with *, ** and *** show significance at 10%, 5% and 1% levels, respectively. Greece, Ireland and the Netherlands are dropped from the estimation since productivity data was missing for these countries. The larger improvements in TFP in the first phase of the EU ETS also occurred for the countries with the net long positions of the EUAs, such as Bulgaria, Finland, Germany, Lithuania. However, the answer to the question whether these improvements were induced by the EU ETS or other factors is not that straightforward, and is answered in the next subsection. 5.3.2

Analysis of variation in total factor productivity

To explain the TFP change and the dynamics of its components we estimate three linear panel regression equations. After performing a simple F-test, the Breusche-Pagan test and the Hausman test, we select a fixed effects model. Table 10 summarises the estimation results for TFP change, EC and TC models. Each of the three models is significant as a whole according to the probability values of the F-test. For the regression on technical change and efficiency change R2 is higher with technical change as the dependent variable as compared to efficiency change as the dependent variable. This shows that explanatory variables chosen in the model are more significant for the determination of technical change relative to efficiency change. This can be explained due to the unavailability of the data depicting the technological aspects of the thermal power plants and the district heating plants, and the regulatory environment of the power generation industry across countries. For TFP growth CO2 price has an insignificant effect. While it is significant in the EC model and the TC model. However, the sign in the EC model is negative while in the TC model it is positive. The negative impact of CO2 price on EC shows that the presence of the EU ETS encouraged an inefficient use of the existing resources. This would be true for countries that could generate windfall profits from selling the surplus of the EUAs (Sijm et al. (2006) show that at a CO2 price of 20e/t, EU ETS induced windfall profits in the power sector if the Netherlands are estimated at e300-600 million per year). In the TC model the coefficient for CO2 price is positive and significant meaning that 22

CO2 price, despite its variation in 2005-2007, has played an important role in shifting the production function of power generation. The country-level stringency variable, measured as the ratio of the allocated EUAs to the verified EUAs, had a negative effect on TFP change and TC. As in the case of environmental efficiency, the higher initial permit allocation relative to the actual emissions did not encourage technical change in the pilot phase of the EU ETS. Oppositely, the overallocation allowed the use of less emissions and energy efficient fuels that led to technical regress in public power industry. In terms of fossil fuel prices, increases in coal price positively influence TC. The sign is opposite (negative) for the interaction term between coal price and the share of the solid fuel used in power generation. This might suggest that while switching to more energy efficient and less carbon pollutant fuels is easy in the countries with smaller share of solid-fuel based power generation, it does not happen with ease in countries where power production mainly relies on solid fuels. For instance, in Poland the share of solid fuels has remained at 95% since 1996. The negative sign for this interaction might also suggest that, in the periods of increasing demand for power, the countries based on solid-fuel power generation would have to exploit other less efficient solid-fuel based power plants. That the peaks in the power demand might be met by the utilisation of less efficient fuels, is confirmed by the negative coefficients for crude oil prices and the industrial production index in the TC model. The coefficient for the industrial production index growth rate is positive in the TFP change and the EC models. This implies that economic growth increases efficiency of resources used in power generation in order to meet the increasing energy demand. The coefficient for the inflow of FDI is positive and significant in the TFP and the TC models, and supports the hypothesis of the foreign technological transfer. The abundance variable, as in the environmental efficiency model, is significant and positive in all three models and, hence, do not support the so called “resource curse” hypothesis but it rather suggests that fossil fuel abundant countries are more likely to experience growth in total factor productivity led by efficiency change and technical change. This goes in line with the recent findings of Brunnschweiler and Bulte (2008) that the resource curse simply does not exist. At least it is true for the European public power generating industry. The coefficient for specialisation in thermal power plants is significant and positive implying that the more power generating industry is specialised the more efficient it is. The share of gas is not significant in explaining all three dependent variables. The coefficient of CHP dummy variable is significant in explaining the EC and the TC. Though, the signs of these coefficients are different. The sign is negative in the efficiency change equation, while it is positive in the technical change model. The positive coefficient shows that in general CHP plants are more technologically advanced because most of them are gas-based and therefore are more energy efficient and ecological. Also, the new additions to CHP capacity have been more advanced in the last years (Graus and Worrell, 2009).

23

6

Conclusions and discussion

This paper contributes to the ex-post research on the performance of the first phase of the EU ETS the EU’s policy introduced to facilitate climate change mitigation targets. The public electricity and heat sector is the largest scheme’s participant that received the most stringent CO2 permit allocation, and hence, it was expected to deliver most of emissions abatement. However, the first trading period delivered quite mixed results: the CO2 price collapsed after 16 months due to the large surplus of the allowances meaning that CO2 price was too low to provide strong incentives for abatement. At the same time, fossil fuels prices were approaching the height and the demand for energy was increasing. All this environment could provide sufficient preconditions for public power generating sector not to improve its environmental efficiency, and, hence, could validate the so called Tyteca’s “paradox” of pollution trading programmes. In this paper we investigate whether this paradox was true for public power generation in the first phase of the EU ETS. As the EU ETS could also encourage improvements/disimprovements in the overall use of resources, alongside the environmental efficiency scores, we measure total factor productivity change. We find that, in the case of the EU ETS, this “paradox” is not apparent because of the emissions trading nature itself, but rather because of the allocation design. The looseness of the policy reflected in the overallocation of the grandfathered permits led to the deteriorating performance of the public power sector. Even the low carbon price in the first phase of the EU ETS, had a positive impact on the environmental efficiency. However, the overallocation alleviated some of these improvements by “facilitating” the use of more carbon intensive and less energy efficient fuels. We could expect that this latter effect will vanish once the auctioning of permits for power generating sector will be introduced in the third phase of the EU ETS. The similar findings are in the case of the TFP change. The price of CO2 does not alter the TFP change as its impacts for efficiency change and technical change are with opposite signs and, hence, they cancel each other. The carbon price has a negative effect on the efficiency change, while it is positive on the technical change. The inefficient use of the existing resources could be encouraged by the windfall profits from selling the surplus of the allowances. We believe that most of the occurred technical change in fossil-fuel based public power generation reflects the short-term abatement through fuel switching rather than the long-term abatement achieved through investments in capital stock. The temporal nature of technical change is strengthen by the negative effect of country-level stringency variable on technical change and TFP change. As in the case of the environmental efficiency, the overallocation could encourage stickiness to dirty-fuel based power generation. The effect of fossil fuel prices also confirms the switching potential in the European public power generating industry. In addition, we find that this effect is not evident for coal reliant power generation suggesting that for countries, locked-in with coal use, the transition towards cleaner power production might be longer than for countries where coal-based production is not dominant. However, we do not find that fossil fuel resource abundance could encourage the lavish use of these resources and, hence prevent efficiency improvements. The European Commission has recently forecasted that fossil fuels will continue to be the main input in energy production. Thus, the existing and future climate policies including the EU ETS need to ensure that fossil fuels are used as efficiently as possible. The ultimate goal should be movement towards frontier, not the one shaped by already utilised technologies, but the one built from the best available technologies.

24

A

Disaggregation of labour data

The sufficiently disaggregated labour data are not available neither from public data sources neither from the national statistics offices. The Eurostat, the ILO and the Eurominotor international provide employment data according to the economic activities. The classification of activities is either based on NACE or ISIC classifications. The employment data of our interest is covered under the section Employment in electricity, gas and water supply (a division E, according to the NACE, and a division 4 according to the ISIC). We have decided to adjust these aggregated data by a proportion of electricity and heat generation by public power plants in total electrical energy and derived heat output. This decision is based on the fact that the largest share of employees ascribed to the division E (or 4) belongs to the subdivision Production and distribution of electricity. We use the employment data from the Euromonitor International as it provides the consistent data series from 1996 to 2007 for all EU member states.

B

Adjustment of net installed electrical capacity for 6 (if Malta and Cyprus is excluded) EU member states

Two ways have been chosen for the adjustment of the capacity data for the remaining countries in the sample. First way is to use the net installed electrical capacity data of all thermal power plants as it is provided by the Eurostat. The second option is to adjust the capacity data by using the electricity generation of public thermal power plants as a scale factor. The first adjustment biases the capacity data upwards, while the second option biases the data downwards as it reflects only the utilised capacity. In the first case the IEA data is used for EU 18 countries. For the rest 6 EU member states (Bulgaria, Estonia, Latvia, Lithuania, Romania, Slovenia) the net installed capacity of all thermal power stations is used as a proxy for the net installed capacity of public thermal power plants. This data is taken from Eurostat. This is reasonable as the share of electricity generation of public thermal power plants in electricity generation of all thermal power plants is very high for these countries. We addop this approach in this paper. The capacity data for Bulgaria are not available for the period 1996-1997. However, electricity data are available for all period of interest. The electricity generation data suggests that there were no significant variations in the electricity production in 1996 and 1997. We use the capacity data from the first available year (1998) for the missing years. In the second case, a separation of net installed electrical capacity of all thermal power plants for Bulgaria, Estonia, Latvia, Lithuania, Romania and Slovenia is based on the fact that electricity generation is highly correlated with installed capacity. For instance, for the EU countries for which the separation between public and autoproducer plants was available, the correlation between the capacity and the electricity generation is about 0.9. Correlations for EU 16 only (Italy, Slovakia and Spain are excluded because some data for autoproducers were missing) are calculated. Correlataion is 0.9 in both cases: 1) correlation between electricity generation of public thermal power plants and capacity of public thermal power plants; and 2) correlation between electricity generation of autoproducer thermal power plants vs. capacity of autoproducer thermal power plants. Firstly, we calculate shares of electricity generation from public and autoproducer power plants for the EU 6 member states. Then, by multiplying these shares with net installed capacity of thermal power plants for the EU 6 MS, we get the net installed capacity of public thermal power plants and autoproducer thermal power plants, respectively. We combine these derived data with the IEA data.

25

References Anderson, B. J., Di Maria, C., and Convery, F. (2009). Abatement and allocation in the pilot phase of the eu ets. Brunnschweiler, C. N. and Bulte, E. H. (2008). The resource curse revisited and revised: A tale of paradoxes and red herrings. Journal of Environmental Economics and Management, 55(3):248–264. CEC (2009). Comission staff working document accompanying the communication from the Comission to the European Parliament, the Council, the European Economic and Social Commitee and the Commitee of the Regions, Towards a comprehensive climate change agreement in Copenhagen: Extensive background information and analysis, part 1. Technical report, Commission of the European Communities (CEC), Brussels. Chen, Y. and Tseng, C.-L. (2008). Climate policies and the power sector: Challenges and issues. Journal of Energy Engineering, 134(2):31–32. Coelli, T. J. and Rao, D. S. P. (2005). Total factor productivity growth in agriculture: a Malmquist index analysis of 93 countries, 1980-2000. Agricultural Economics, 32(s1):115–134. Convery, F., Perthuis, C. D., and Ellerman, D. (2008). The European carbon market in action: Lessons from the first trading period. interim report. Technical Report 0802. Delarue, E. D., Ellerman, A. D., and Dhaeseleer, W. D. (2008). Short-term CO2 abatement in the European power sector. Technical Report 0808. EEA (November, 2008a). Indicator: EN09 emissions (CO2, SO2 and NOx) from public electricity and heat production - explanatory indicators. Technical report, European Environmental Agency. EEA (November, 2008b). Indicator: EN19 efficiency of conventional thermal electricity and heat production. Technical report, European Environmental Agency. Ellerman, A. and Buchner, B. (2008). Over-allocation or abatement? A preliminary analysis of the EU ETS based on the 2005-2006 emissions data. Environmental and Resource Economics, 41(2):267–287. F¨are, R., Grosskopf, S., Norris, M., and Zhang, Z. (Mar., 1994). Productivity growth, technical progress, and efficiency change in industrialized countries. The American Economic Review, 84(1):66–83. Graus, W. and Worrell, E. (2009). Trend in efficiency and capacity of fossil power generation in the EU. Energy Policy, 37(6):2147–2160. Hoffmann, T. and Voigt, S. (1905). What Drives the Efficiency of Hard Coal Fuelled Electricity Generation? An Empirical Assessment. SSRN eLibrary. IEA (July 2008). Combined heat & power and emission trading: options for policy makers. Technical report, International Energy Agency. Picazo-Tadeo, A. J. and Garca-Reche, A. (2007). What makes environmental performance differ between firms? Empirical evidence from the Spanish tile industry. Environment and Planning A, 39(9):2232–2247. Sachs, J. D. and Warner, A. M. (1995). Natural resource abundance and economic growth. Working paper 5398. Sachs, J. D. and Warner, A. M. (2001). The curse of natural resources. European Economic Review, 45(4-6):827–838. 26

Sijm, J., Neuhoff, K., and Chen, Y. (2006). CO2 cost pass-through and windfall profits in the power sector. Climate Policy, 6:49–72. Simar, L. and Wilson, P. (2008). The measurement of productive efficiency and productivity growth, chapter 4, Statistical inference in nonparametric frontier models: recent developements and perspectives. Oxford University Press. Simar, L. and Wilson, P. W. (1999). Estimating and bootstrapping malmquist indices. European Journal of Operational Research, 115(3):459–471. Simar, L. and Wilson, P. W. (2007). Estimation and inference in two-stage, semi-parametric models of production processes. Journal of Econometrics, 136(1):31–64. S¨oderholm, P. (2001). Fossil fuel flexibility in west European power generation and the impact of system load factors. Energy Economics, 23(1):77–97. Tyteca, D. (1996). On the measurement of the environmental performance of firms– a literature review and a productive efficiency perspective. Journal of Environmental Management, 46(3):281 – 308. Wilson, P. W. (1995). Detecting influential observations in data envelopment analysis. Journal of Productivity Analysis, 6(1):27–45. Yaisawarng, S. and Klein, J. D. (Aug., 1994). The effects of sulfur dioxide controls on productivity change in the U.S. electric power industry. The Review of Economics and Statistics, 76(3):447–460. Yang, H. and Pollitt, M. (2009). Incorporating both undesirable outputs and uncontrollable variables into dea: The performance of Chinese coal-fired power plants. European Journal of Operational Research, 197(3):1095–1105. Zhou, P. and Ang, B. (2008). Linear programming models for measuring economy-wide energy efficiency performance. Energy Policy, 36(8):2911–2916. Zhou, P., Ang, B. W., and Poh, K. L. (2008). A survey of data envelopment analysis in energy and environmental studies. European Journal of Operational Research, 189(1):1–18.

27