A Decade of EU ETS Emissions
Allowance Allocation Policy: Does the
Independence Property hold?
University of Groningen
Faculty of Economics and Business
Msc Master Thesis International Economics and Business
By: VS Baker (s2136465)
Thesis Date: June 13
th2016
Abstract
This paper set out to investigate the validity of the independence property of the allowance allocation policy in the European cap and trade program (EU ETS) that is implied by the Coase theorem. The empirical results presented in this paper find no strong evidence to suggest that the independence property is, indeed, not featured within the EU ETS.
Table of Contents
1 Introduction
12 Literature Review
42.1 Mechanics of a cap and trade program 4
2.2 The state of the EU ETS 5
2.3 Cap and trade and Coase theorem 7
1
1 Introduction
The global efforts to reduce greenhouse gas emissions in the epic battle against climate change have inspired a rich literature devoted to designing and evaluating policy instruments that pursue this end. Much of the current discussion revolves around the carbon cap and trade system, which has been popularized and put into practice throughout the world.
In a carbon cap and trade system a central authority controls the quantity emissions it allows by issuing a finite amount of emission allowances. Polluters are obliged to obtain these emission allowances and hand them over to the authority to pay for their right to pollute. The economic intuition behind such a system is quite straightforward.
By introducing a limited supply of emission allowances and by allowing these to be traded freely among polluters, those who place most value on the emission allowance will pay more for the right to pollute than others. Likewise, those who gain too little from polluting to justify purchasing emission allowances, or those who find that reducing their emissions is cheaper than buying allowances, would reduce their emissions in order to avoid having to purchase emission allowances. This mechanism ensures that the easiest emissions are cut before anything else. The cap can then be gradually lowered over time in order to smoothly reduce overall carbon emissions.
The EU Emission Trading System (EU ETS) is one of such cap and trade systems. It went into effect back in 2005 and is currently in phase III of the program. In phase I, which covered the period 2005 – 2007, some ten thousand polluting installations covering 24 EU countries were enrolled in the program. In this phase, nearly all allowances were allocated freely among the installations based on their historical emission data. This policy of free allocation of all emission allowances also known as ‘grandfathering’ was aimed at fully compensating the participating installations in order to ease them into the program.
Phase II, which ran during the period 2008 – 2012, continued this process of grandfathering be it somewhat more generously than intended. Though the plan was to gently reduce the cap and, hence, tighten the screws as to stimulate emission abatement, the great recession cut carbon emissions through depressed output in such a manner that it created a massive oversupply of emission allowances.
2
mainly not freely, but through auctions. In 2013, some 60% of the allowances were distributed freely while the rest was auctioned off. This figure is set to increase over the years such that, by 2020, the bulk of the allowances will be auctioned off instead of given away for free.
The crucial difference between freely allocating the allowances and auctioning them is that installations are no longer getting a free ride in the latter scenario. Yet, given that the manner in which the allowances are distributed is unrelated, in principle, to the emission cap set by the EU, the equilibrium allocation of allowances should be unaffected by this change in distribution policy assuming that trading allowances bears no significant transaction costs, as is implied by the Coase theorem (Coase 1960). The only difference, then, would be that installations were getting a subsidy in the form of free emissions allowances before and, now, are not. The cost of emitting one ton of CO2 equivalent (CO2e) would still be equal to the price of one emission allowance and, hence, the
monetary incentive to reduce emissions would be the same under either allocation mechanism. In other words, the emission allowances would still end up covering the most valued emissions no matter how the allowances were initially allotted.
However, in the presence of significant transaction costs, the initial allocation does matter. When the basic trading mechanism that works to allot the emission allowances to the most valued emissions is impeded (e.g. by the presence of transaction costs or non-cost-minimizing participants), the initial allocation may affect how the emission allowances will be distributed afterwards. Hence, in the presence of significant transaction costs, the final allocation may no longer be equal to the efficient allocation – the latter being the one that minimizes the cost of cutting emissions.
Instead, those participants that receive more free allowances than they would end up using could end up consuming too many allowances. On the flipside, those that receive fewer free allowances than needed for their ideal emission levels could end up paying a higher price for cutting their emissions than some of the more fortunate – those with the excess in free allowances – participants would.
In the case of significant transaction costs for emission allowances, then, the end of the grandfathering phases I & II marked by the start of phase III could have implications for the emission cutting behavior of all participants involved.
3
4
2 Literature review
2.1 Mechanics of a cap and trade program
The cap and trade system consists of two crucial elements. First, the authority sets a ceiling on how much pollution it allows to occur by introducing a fixed quantity of pollution allowances that polluters need to obtain to cover their pollution i.e. the ‘cap’. Second, the authority lets polluters compete with one another for the right to pollute by allowing the pollution allowances to be transferred between polluters i.e. the ‘trade’.
The cap creates a scarcity and, hence, puts a price on polluting while the trade ensures that those who place the highest value on polluting will can continue to pollute (i.e. by purchasing pollution allowances from those that place less value on them). Consequently, pollution will be cut for as long as the marginal cost of cutting one unit of pollution is lower than the price of a pollution allowance.
Taken together, the cap and trade mechanics ensure that, first, the authority has direct control over the maximum level of pollution and, second, the pollution reduction needed to not exceed this maximum is achieved by cutting pollution at the lowest societal cost.
Although this basic mechanic of cap and trade system is fairly straightforward and intuitive, fleshing it out into a proper policy is somewhat more involved. A proper system of verification and enforcement is of paramount importance, but there are many other choices that, perhaps more fundamentally, affect the way the program functions. As is often the case: the devil is in the details.
It should be no surprise that it matters, greatly, at what level the cap is set. Set the cap too high and the polluters will not feel enough pressure to cut down on their pollution; too low and the societal cost of cutting pollution grows too large. Getting the cap just right, then, is a matter of correctly identifying the pollution that would occur without any restrictions and the cost of cutting that pollution with a certain amount.
5
be justified using the same argument. After all, when participants can retain their unused allowances, they would have an incentive to cut pollution today if future pollution reductions are expected to be more costly . This can even be expanded to create a two way street in which current polluters can borrow allowances to be repaid at a future date.1
The way in which allowances are allocated is also a choice to be made. The most straightforward way of allocating the allowances might be to simply let the authority auction them off to the highest bidders. In practice, a policy of handing out allowances for free to polluters is often pursued instead. Such an approach addresses the reservations of policymakers that believe a cap and trade system might hurt the international competitiveness of their domestic industries while also making the affected industries themselves less resistant to the implementation of the program.
Whichever allocation policy is pursued, the initial allocation should not have any meaningful impact on the way in which emissions are cut (Montgomery 1972);(Stavins 1995). Differently put, regardless of initial allocation scheme, it should still be true that those that place more value on polluting than the cost of an allowance are the ones that will end up having the allowances they need, be it after purchasing it from another holder who places less value on it. Hence, the only determinants of a firm its emission abatement are the price of an allowance – which, in turn, is indirectly set by the chosen emission cap – and its marginal cost of emission abatement.
2.2 The state of the EU ETS
The EU ETS program has, thus far, seen three phases. The first phase of the program was characterized as ‘learning by doing’ and ran from 2005 to 2007. This phase knew a generous allowance allocation scheme in which virtually all allowances were handed out to participating firms free of charge. The allowances handed out in phase I would not be usable in phase II and beyond. Even so, the generous free allocation of allowances continued in phase II. This, combined with the economic downturn that was the Great Recession, led to a large oversupply of emission allowances and, subsequently, a sharp decline in the price of emission allowances.
1 Holland & Moore (2013) show that the banking feature is very commonly incorporated in cap and trade systems,
6
For phase III, which started in 2013, the emissions cap has been set to decline at a somewhat more ambitious rate. Still, it is estimated that banked allowances have accumulated to the tune of some 1.8 billion over the course of phase II, which is not much less than the cap set for 2013 alone. Over the course of 2013 too, the surplus grew; this time by some 220 million allowances.
The over-allocation of allowances during Phase II has been widely criticized2 for not
putting enough pressure on polluters to reduce their emissions. Another point of contention is the way in which the allowances were distributed during phase I and II. The grandfathering scheme – i.e. providing full coverage through free allocation of allowances to polluters – has been criticized for being a de facto subsidy to polluters.
In 2008, for example, a report by Point Carbon (2007) estimated the combined windfall profits for the power generation sector of five member countries (Germany, Italy, Poland, Spain, and the UK) were upwards of 23 billion euros.3 Goulder et al. (2010) make a similar point
regarding windfall profits from excessive free allocation. They estimate that at 100% free allocation, firms would be severely overcompensated.4 Another example is the difficulties in
dealing with new entrants, which, ultimately, culminated in the creation of a new entrant EUA reserve fund that allows member states to allocate free allowances to new entrants.
In phase III, which took off in 2013 and will last until 2020, Europe is taking a different approach. In this phase, the share of freely allocated allowances is greatly reduced – by some 40% in 2013, with further reductions to be made in subsequent years. Also, more efficient installations are awarded additional free allowances, which serves as an additional incentive for firms to reduce their emission levels. Furthermore, phase III harmonizes the system and expands the cap and trade system in some regards, for example by committing to more uniform allocation benchmarking rules and by including additional greenhouse gases.
2 Deutsche Bank Research (2010) provides an excellent summary of the criticisms leveled against grandfathering. 3 Though it should be noted that even their most conservative price estimation was well above the realized price over
phase II.
4 Their estimation is based on a GEM model applied on the US economy. They estimate that, even for the most
7
2.3 Cap and trade and Coase theorem
As pointed out in section 2.1, the allocation of emission allowances under the EU ETS program should have no effect on the final allocation if the allowances in the absence of transaction costs. This is implied by the Coase Theorem as derived from “The Problem of Social Costs” (Coase 1960). Specifically, the Coase theorem predicts that, if economic agents can trade externalities in the absence of transaction costs, they will reach a pareto efficient outcome regardless of who owns the externality initially.
In Hahn & Stavins (2011), this is dubbed the independence property the and the authors describe how not only transaction costs, but also market power, uncertainty, conditional
allowance allocations, non-cost-minimizing behavior and differential regulatory treatment may
invalidate this property in a emission cap and trade system.
Transaction costs
Although the Coase Theorem assumes no transaction costs, Coase himself was explicitly critical of that very assumption in his paper. Instead, he argued that, in many real economic situations, the presence of significantly large transaction costs works prohibitive to reaching a pareto efficient outcome. In other words, when transaction costs are a relevant factor, it no longer holds true that it does not matter who initially owns the externality.
Applied to the EU ETS program, the externality can be interpreted as being the right to emit a quantity of CO2e, granted by owning a bunch of emission allowances. The moment one
installation cashes in its allowances by emitting a certain amount of CO2e, other installations are
limited in their ability to emit CO2e themselves. If the assumption of no transaction costs does not
hold, then installations are facing barriers to trade that hampers them to buy or sell allowances efficiently enough to reach a cost minimizing allocation of allowances.
8
Figure 1 – Phase I & II EUA prices 2005-2007
Notes: In this figure, the futures prices for phase I (brown/triangle) and phase II (blue/cross) EUA are plotted over the span of phase I Source: Ellerman & Joskow (2008)
9
Figure 2 – Yearly trading volume of EU allowances 2005-2012
Notes: yearly trading volume are presented in millions of allowances. The totals are the sum of exchange trading, over-the-counter trading (i.e. directly between two parties) and auctions (managed by national authority).
Source: EU ETS Handbook (Accessed May,2016)
Figure 2 shows the yearly total trading volume of European emission allowances for the phases I & II in millions of allowances . Having started at a modest 94 million allowances in the first year of the program, the market ballooned to a volume of nearly 8 billion allowances in 2012. To put that number into perspective: in 2012, roughly 2 billion tons of emissions were recorded by the ETS program. This means that, even if all allowances required in 2012 would have to be obtained through trade (one allowance covers one ton of CO2e), the amount of trade necessary to
fulfill the ETS obligations is a factor four smaller than the observed 8 billion. When one considers that the bulk of the allowances were allocated for free in 2012, the numbers become even more staggering.
10
transaction costs, however, it is unlikely that allowances would be traded so frequently and abundantly for reasons other than securing the necessary allowances. Hence, the huge size of the market relative to the yearly emissions covered by the program are indicative of the absence of significant transaction costs.
Market Power
In the presence of firms with market power, the final allocation of emission allowances may be different from the efficient allocation if firms use their market power to manipulate the price of emissions allowances to their benefit i.e. exhibiting monopolistic or monopsonistic behavior. In the presence of market power in the emission allowance market, net sellers of emission allowances could find it rewarding to limit their supply of allowances to increase the price. Likewise, net buyers may choose to limit their demand in order to lower the price they face. In the EU ETS program, the top four accounts cover 13.27% of total emissions and have received 9.85% of the total freely allocated emission allowances. Market power as measured by the Herfindahl-Hirschman index (HHI) scaled between 0 and 1 with regard to total emissions and free allowances are 0.008 and 0.005, respectively.5
These statistics suggest that market power in the EU ETS program is of little significance. Hence, the independence property is not likely to be violated through a presence of market power in the EU ETS program.
Uncertainty
Since the inception of the EU ETS program, there has been great volatility in the prices of allowances. The wild rollercoaster ride that was the phase I EUA price development is a testament of the inability – or even unwillingness – of the authorities to manage the price development in this early period. It is also illustrative of the unpredictability of the level of carbon emissions themselves; though the carbon cap for the 2005-2007 period was well known from the get-go, the price of EUA still peaked at little over 30 euro in early 2006. Only afterwards it became apparent that the carbon emission levels would be much lower than the supply of phase I allowances, which resulted in the slide of the price to near zero during the course of 2006.
11
Phase II saw similar price swings. Most notably, the aftermath of Great Recession caused a sudden drop in phase II EUA prices – even though these allowances would be bankable for use in later phases – as the drop in economic activity meant that the predetermined cap was set too high once again.
These experiences show that, while the price of emission allowances is very much exposed to the actions of the authorities and the levels of carbon emissions that will be realized, both factors have proven to be highly unpredictable.6 Hence, one can suspect that uncertainty regarding future
price developments is very much present.
If such suspicions are valid, then the independence property might be violated. Uncertainty regarding future prices of emission allowances could cause firms to seek to limit their exposure to future allowance prices by adjusting their efforts on decreasing emissions. Net buyers, for example, may choose to ramp up their efforts to hedge against the risk of rising allowance prices. Net sellers, on the other hand, may choose to put less effort than otherwise in reducing their emissions in order to hedge against lower future allowance prices. Hence, in the face of uncertain future allowance prices, it could very much matter how the allowances are initially distributed when it comes to incentives to reduce emissions. This implies that it is no longer necessarily true that the cheapest-to-cut emissions would in fact be cut when firms are differently exposed to the risk of future price developments of emission allowances.
Conditional allowance allocations
When the future free allocation of allowances is to some extent dependent on current emissions, there is an incentive to adjust emissions in such a manner as to secure more free emissions than would otherwise be the case. For example, if reducing current emissions would lead to fewer free emissions in a future period, firms might be discouraged to reduce their emissions. Conversely, if reducing emissions is rewarded with even more free allowances in the future, then firms might cut their emissions beyond what is efficient from a societal perspective.
The EU ETS program addresses the problem that arises by making future free allocation of allowances depended on current emissions by, as a rule, determining the level of free allocation
6 This is not to say that a change in allocation scheme from grandfathering to auctioning will reduce uncertainty
12
based on historical emissions only once and adjusting future allowances based on across-the-board rules instead of on an ad hoc basis. Nevertheless, there are some important exceptions to this rule.
For one, completely closing down a polluting installation forfeits all future free allocation of allowances bound to that specific installation. This is, perhaps, unsurprising as it makes little sense to keep subsidizing firm that has exited the market. However, when a firm owns multiple polluting installations – indeed, this a common occurrence within the ETS program – closing down one installation does not mean the firm itself is not active anymore. In such a scenario, a firm might be deterred from closing down an inefficient installation because it forfeits the freely allocated allowances bound to that installation.
In addition, as of 2011, freely allocated allowances will also be scaled back when a particular installation cuts back on its operational activities by a significant amount. This means that installations running on half their normal capacity might see their future free allowances reduced as a result.
Non-cost-minimizing behavior by firms
The cap and trade mechanics follow the assumption that firms are cost minimizing entities. However, this need not be the case. The presence of non-cost-minimizing behavior could lead to non-efficient outcomes regardless of the initial allocation. Still, this does not mean that the initial allocation is necessarily unimportant when considering the effects of non-cost-minimizing behavior.
Indeed, the way in which allowances are initially allocated could, in itself, induce or exacerbate non-cost-minimizing behavior. In particular, a form of non-cost-minimizing behavior that could exist in the presence of free allocation is that mere ownership of allowances increases a holder its perceived value of these allowances. This phenomenon is also known as the endowment effect7 in the field of behavioral science. The result of such an endowment effect would be that
participants would sooner hold on to their allowances, using them to maintain less valuable emissions. Meanwhile, those looking to buy additional allowances would face higher prices due
7 In Requant & Ellerman (2008), the term ‘endowment effect’ is, somewhat confusingly, used to refer to the effect
13
to the endowment effect-induced reluctant sellers, and be forced to abate more valuable emissions than otherwise would be the case.
Differential regulatory treatment of firms
A particular manner in which a differential regulatory treatment of firms could undermine the independence property may be when buying allowances or getting them for free is consequential to a firm its income statement. Regulations may differ from industry to industry, and, consequently, the relevance of getting to write the acquisition of allowances up as an expenditure may differ as well. Hahn & Stavins (2011) illustrate how it may matter in the often heavily regulated industry of power generation.
Although some of the aforementioned factors may be more prevalent in the EU ETS program than others, a significant presence of just one of these would be enough to undermine the independence property. If so, the efficient allocation of allowances would not be reached, as those that receive too many free allowances – that is, more allowances than they would hold under efficient allocation – end up holding too many allowances. Consequently, this group would abate its emissions at a lower-than-efficient rate. On the flip side, the firms that receive too few allowances would abate their emissions at a higher-than-efficient rate.
Requant & Ellerman (2008) investigate the validity of the independence property in the EU ETS program by exploiting the way in which Spanish coal plants were allotted free allowances.8 They find no signs of the independence property being violated and they conclude
that the manner in which allowances are distributed may be inconsequential to the manner in which emission reductions are achieved under the EU ETS program. To this writer his knowledge, thus far, no study has been able to present evidence that does suggest a consequential role of the allocation scheme to the emissions reduction incentives of EU ETS participants.
8 The way in which these plants were allotted allowances resulted in differences in allocations that were not linearly
14
3 Methodology
The change in the allocation process introduced in phase III provides a natural experiment for measuring the effects of the initial allocation across firms. Hence, this paper exploits new data compared to, for example, Requant & Ellerman (2008) in investigating the validity of the independence property in the EU ETS. There are two important characteristics, in particular, upon which the experiment hinges.
First, in phases I and II, firms benefitted in varying degrees from the grandfathering process. While the grandfathering phase was aimed to fully compensate polluting installations through free emission allowances, the actual emissions covered by freely allotted allowances differed widely between firms, with some receiving too many and others too few allowances. Though such a discrepancy is to be expected over time as the free allocation is supposed to remain fixed despite the fact that installations cut emissions at a different rate, it already very much existed in the early stages of the program.
Figure 3 helps illustrate this point. This figure shows, for 2008, the ratio of free allowances to verified emissions (which, in this paper, will be referred to by the term ‘coverage’) of the four largest industry groups within the ETS program in terms of total verified emissions. The green (striped) bars display these ratios for the least covered half of emissions – that is, the 50% of emissions associated with accountholders9 with the lowest coverage – while the yellow bars
represent the ratios for the most covered half of emissions.
9 Installations enrolled in the EU ETS are tied to accounts that receive and surrender emission allowances. The
15
Figure 3 - Free Allowances / Verified Emissions ratios 2008
Notes: Ratio of free allowances-to-verified emissions in 2008. Ratio = 1 is marked by the red broken line. The breakdown between least and most covered 50% is in terms of emissions (tons of CO2e) tied to the most and least covered account-holders.
Source: Created for this paper, based on data taken from European Union Transaction Log, retrieved on April 2016.
Although not shown in the figure, it should be stressed that significant differences within industry groups remain even on a country level. Clearly, firms – even among their domestic industry peers – have not benefitted from the free allocation scheme equally.10
Second, the start phase III cut the level of free allocation of emissions across the board by some 40%.11 This major shift in allocation policy could have affected the rate at which firms
reduce their emissions if the independence property of the Coase theorem is invalid.
10 It should be noted that, in terms of number of accountholders, the amount that received less than full coverage (i.e.
have a ratio smaller than 1) in 2008 was relatively small. Out of 6446 accountholders active in 2008, only 1767 have a ratio less than 1 and only 658 a ratio smaller than 0.8. On the other hand, 2778 accountholders received allowances at a ratio of more than 1.2 and 1586 accountholders at a ratio larger than 1.5.
11 Although some installations were spared this tragedy.
0 1
Combustion of fuels Production of lime or
16
In particular, one could expect that those firms that benefitted more from free allocation in phase II would have cut their emissions at a lower-than-efficient rate in that period, whereas, by definition, other firms would have cut their emissions at a higher-than-efficient rate. By changing the allocation process in phase III such that free allocations become less relevant, the efficiency discrepancy induced by the free allocation of allowances should diminish. In other words, the change would cause previously privileged firms to start cutting their emissions at a higher rate, closer to the efficient rate. At the same time, the other firms would decrease their abatement efforts and move closer to the efficient rate as well. This is captured in (H.1).
Emission coverage in phase II is positively correlated to improvements in the rate of emission abatement in phase III relative to phase II. (H.1)
3.1 Model
To measure the effect of the allocation policy change, the average rate of emission abatement occurring in phase II will be contrasted with the average rate of abatement in phase III for each accountholder12. Specifically, the average yearly rate of abatement in emissions in phase
III is divided by the average yearly rate of abatement in phase II to arrive at a final ratio that expresses the abatement rate change that occurred in phase III relative to phase II.
ARC
4
3
(E.1)Here, ARC is a measure of abatement rate change, with values smaller than one indicating an acceleration in phase III relative to phase II. Ei resembles the emissions of an accountholder in year i.
The main independent variable measures the free allocation coverage in phase II. That is, the percentage of emissions that the accountholder can cover using freely received allowances.
12 Observations on an accountholder level are obtained by summing the installation level data corresponding to each
accountholder. A more detailed description of the summing process can be found in 3.3 Accountholder-level
17
Accountholders that are very successful in cutting their emissions would see their coverage improve over time as their amount of free allowances, generally, remains more or less fixed. To the extent that high emission reductions in phase II lead to a higher coverage, this may pose a problem of endogeneity. That is, emission reductions taking place in phase II would affect both the dependent variable coverage and the independent variable rate of abatement change. In order to prevent such endogeneity, the coverage variable will be calculated as the ratio of free emission allowances in 2008 and the average emissions taking place in the period 2005-2007.
Given that the emission allowances are calculated using historical emissions, this measure seems appropriate. The construct is as follows:
(E.2)
Were CII is the measure of coverage during phase II, A08 are the free allowances granted in 2008 and ρ corresponds to the number of years that the accountholder has been active during 2005-2007. Note also that instead of including free allowances granted in all years for phase II, only the first year is used. The reason for this is that the EU ETS policy13 for dealing with installation capacity changes is to adjust allocation of free allowances under certain conditions, which, again may lead to bias14 (e.g. capacity reductions leading to fewer allowances in later years would inappropriately inflate the measure of coverage in a similar fashion as basing the measure on total phase II emissions would).
In addition, the model will include a measure of coverage in phase III using the free emission allowances in 2013 and the average emissions during phase I. Since, in phase III, the free allocation of allowances is partly based on (carbon efficiency) performance benchmarks, it can be expected that those firms that have performed well in terms of past abatement receive a relatively higher coverage compared to their peers. In the case that the independence property is violated, one might expect that those that receive relatively more free allowances in phase III will abate at a lower-than-efficient rate during phase III. Hence, hypothesis (H.2) can be formulated as follows:
13 Also see conditional allowance allocation (p.11).
14 In Appendix (p.ii) Table 2 presents alternative results based on coverage (c
II) measured as average allowances
18 Phase III emission coverage is negatively correlated to improvements in the rate of
emission abatement in phase III relative to phase II. (H.2)
To mitigate problems arising from suspected heteroscedasticity, the variables are transformed into natural logarithmic forms.15 The basic OLS regression model is as follows:
(E.3)
Here, arc is the logged value of the abatement rate change, cII and cIII are the logged values of the coverage in period II and III, respectively, and s is the logged value of size16. The model is
completed with dummy variables Dij where i is an index for countries and j for industry groups.
3.2 Data
The data is taken from the European Union Transaction Log17 which comprises of
observations on all – little more than 15,000 – stationary installations enrolled in the EU ETS program over the years 2005-2015 for 2418 of the 31 participating countries. These observations
include the amount of freely allocated allowances as well as the verified emissions in tons of CO2e
for each year the installation is enrolled in the program. In addition, there are a couple of general descriptors for each installation, such as the administrating country, the accountholder (one accountholder may be tied to several installations), and the industry type.
The bulk of the observed installations have been enrolled in the ETS program since phase I. However, a sizable group only entered after the starting year 2005. Furthermore, (temporary) closures and new entrants throughout phases II & III appear frequently in the dataset. As a result,
15 The logged variables ‘cII’ and ‘cIII’ are calculated such that a value of zero corresponds to an accountholder that
received zero free allowances. This is achieved by adding 1 to CII and CIII before taking their log.
16 Size is estimated by summing the total emissions of the accountholder over 2008-2015 (i.e. phases II & III). 17Available at http://ec.europa.eu/environment/ets. Note that the downloads are for some reason limited to 3000
observations per file and the data is presented in long format . In practice, this means that for the dataset used in this paper, more than 300 files have to be retrieved (individually). Furthermore, a few of the files contained inconsistent ordering of columns and, in one case (Liechtenstein), had columns missing.
18 Bulgaria, Croatia, Cyprus, Iceland, Liechtenstein, Malta and Norway have been excluded due to the lacking of
19
roughly half the installations are missing data for one or more of the recorded years. On an accountholder-level, 4096 have sufficient data out of a total of 8948 accountholders.
3.3 Accountholder-level observations
This section will briefly outline the reason for using accountholder-level observations instead of installation-level observations and manner in which the account-holder level data is represented in the model variables.
The aim of this paper is to investigate the validity of the independence property with regards to the allowance allocation scheme. Although amounts of free allowances are determined per installation and not per accountholder, the independence property (or lack thereof) is only relevant in the context of market transactions between economic actors. To the extent that installations belonging to the same firm are not engaging in market transactions amongst each other, it is not appropriate to treat them as separate economic actors. Hence, the use of accountholder-level observations is nothing more than an attempt to group those installations that belong to a single economic actor.19
An accountholder-level observation is, essentially, just the sum of the installation-level data tied to one accountholder. However, for the model, it might prove problematic to include installations that are only active in a limited number of years. In particular, the coverage variables might be contaminated as the coverage is based on historical emissions.
If, for example, an installation was active in phase I and was shut down before the start of phase II, then the historical emissions would include that installation while the allowances would have been corrected for that installation no longer being active. Hence, in that case, the coverage would be underestimated.
To prevent such problems, the coverage variables will be based on installations that have no missing data throughout phase II and phase III and have, at least, data for one year of phase I. Furthermore, the average historical emissions of installations during phase I will be calculated
19 Although accountholder-level data is the most practical proxy to capture firm-level activity, there is some
20
prior to deriving the accountholder-level data. This ensures that, in the case that installations have sufficient data but different starting years, the historical emissions will not be underestimated.20
The abatement rate change (arc) variable, in principle, does not exclude installations that are only partially active (i.e. newly entered or exited over the length of the program). After all, in terms of emission management, the closing or opening of installations are as much part of operating under a cap and trade system as is the managing of emissions at existing installations.
To be sure, the closure of an installation or the opening of a new one can lead to an abrupt change on the overall emissions of an accountholder in a particular year. While this is a testament to the importance of including such occurrences in a model concerned with changes in emission levels, one could suspect that any observed correlation in the model hinges on the inclusion of such observations.
For this reason, an alternative measure of the variable abatement rate change that excludes any partially active installations will also be considered:
Alt_ARC
2 ⋅
(E.4)
Hence, this alternative measure is comprised of the same installations as those included in the coverage variables.
Still, it can be expected that accounting for the effects of installation closures and openings amplifies the importance of coverage levels if the independence property does not hold.
Hence, hypothesis (H.3) can be formulated as follows:
When installation closures and openings are taken into account in the calculation of the emission rate change, the observed effect of phase II coverage levels is larger. (H.3)
20 Another point of consideration is that installations might have conflicting categorical data. That is, installations
21
4 Results
Table 1 shows the estimated coefficients along with the explanatory power and number of observations for each of the OLS-regressions. The independent (log) variables are phase II coverage, phase III coverage, and size, respectively. The first three columns represent the results of the main model with the dependent variable as calculated in equation (E.3). The regression results shown in column [1] are based on observations that include partially active installations (i.e. the base model), while column [2] shows the results of the regression that excluded such observations, and column [3] shows the results of a regression run on installation-level data. The final three columns, together, represent the results based on the alternative measure of the abatement rate change variable (E.4). The particular variation of this model is that the abatement rate observed between the years 2012 and 2013 is excluded.
Table 1 Regression results arc (abatement rate change).
The main independent variable, cII (phase II coverage), is not significant at a 10% level in the base model. Furthermore, the other two regressions based on the main measure of arc
arcin [1] arcex [2] arcinst [3] Alt_arcin [4] Alt_arcex [5] Alt_arcinst [6]
Adj. R2 .0553 .0613 .0651 .0540 .0628 .0723 Obs. 4096 4096 6577 4096 4096 6577 cII -.0170 -.0169 -.0140 .0470 *** .0473 *** .0342* (.0134) (.0125) (.0178) (.0126) (.0121) (.0180) cIII .0154 .0220 * .0323** -.0609*** -.0590 *** -.0629*** (.0135) (.0126) (.0138) ( .0127) (.0122) (.0139) s .0070 .0056 .0211*** .0045 .0013 .0178*** (.0050) (.0046) (.0044) (.0047) (.0045) (.0045)
22
(abatement rate change) also show no significant coefficients at a 10% level. Hence, [1] [2] and [3] are not supportive of (H.1).
In addition, the regressions that correspond to the alternative measure for arc – this one excludes the abatement rate observed between 2012-2013 – report significant positive coefficients for cII (at a 1% level for the accountholder-level regressions [4] and [5] and at a 10% level for installation-level regression [6]).
However, as smaller values of arc correspond to larger improvements of the abatement rate in phase III relative to phase II, these positive coefficients are also not supportive of (H.1). That is, the observed positive correlations do not imply that firms with higher phase II coverage perform relatively better in terms of abatement rate change.
The coefficients of cIII (phase III coverage) are also not significant in the main model [1] but are, in accordance with (H.2), positive and significant at a 5% and a 10% level in [2] and [3]. The regressions based on the alternative measure of arc, though, show significant negative coefficients. and the installation-level regression based on the alternative measure [6]. This, similarly to (H.1), is in stark contrast to (H.2) that would predict a positive coefficient.
For the third and final hypothesis, a Wald test using the estimated coefficients for cII in [1] and [2] (and [4] and [5]) is performed. The results do not show any meaningful21 support for (H.3).
In other words, there is no strong indication that the effect of cII on arc is amplified when partially active installations are taken into account.
The coefficients for s (size) are not significant at a 10% level for any of the accountholder-level based regressions. Yet, in the installation-accountholder-level based regressions, s is positive and significant at a 1% level.
All in all, regressions [2] and [3] show some significant coefficients for cIII that would be
supportive of (H.2). Other than that, though, there is simply very little support for (H.1), (H.2) and (H.3) to be found.
21 H
0 for cII(βy,in - βy,ex = 0) and cII(βy_alt,in - βy_alt,ex = 0) have estimated probabilities of 0.9872 and 09403,
23
5 Discussion
An analysis using three differently constructed datasets – an accountholder-level dataset that includes partially active installations, one that does not, and an installation-level dataset has given no indication that phase II coverage is predictive of the degree to which firms changed their abatement rates in phase III. That is, the change in allocation policy that occurred in phase III – a sudden move away from free allocation in favor of auctioned allocation – seems to have impacted firms with generous phase II emission coverage no differently than firms with much less coverage. This suggests that the independence property holds in EU ETS and, hence, that grandfathering does not lead to less efficient abatement vis-à-vis auctioning.
Interestingly enough, though, the three regressions based on the alternative measure for arc [4]-[6], do indicate a significant relation between abatement rate change and cII (phase II coverage) – just not the one that was hypothesized.
The results, as it appears, very much depend on whether the abatement rate observed between 2012-2013 is taken into account. As the alternative measure is, by necessity, based on only two observation points to approximate the average abatement rate in phase III, having this extra observation point in the main measure might have produced significantly better estimates of the average abatement rate in phase III. Still, this observation point is special as 2012 marks the last year of phase II.
The upshot is that abatement rate in 2012-2013 could, in part, be determined by changing definitions and policies that were introduced in 2013 and, therefore, introduce significant bias to the model. For example, in phase III, the scope of the program expanded to include greenhouse gases N2O and PFC for certain production processes.22 As these greenhouse gases are byproducts
of very particular industrial processes, their inclusion inflate the reported emissions in 2013 in a disproportionate manner across installations. Hence, one would expect the 2012-2013 abatement rate observation point to be biased.
To some extent, the country-industry dummies should absorb such effects, but significant bias could remain if the classifications are not detailed enough, which may be likely. Moreover,
22 To put their effect into perspective: these additional gases correspond to roughly 8% of CO
2e relative to CO2
24
the compiling of the installation-level data into accountholder-level data will have resulted in less accurate country-industry classification as well.
As it appears, the installation-level data based regression [6] provides a better fit than its counterpart [3]. The implication might be that, on even an installation-level, the industry and country dummies are not preventing bias sufficiently enough to warrant that extra observation point.
Hence, although the base model does not suggest that the independence property might not hold in the EU ETS, the results of using the alternative measure can be considered as more meaningful, which makes those results more curious still. After all, even though the an absence of the independence property in the EU ETS was hypothesized (H.1 and H.2) to produce results opposite to those that have been found in [4]-[6], in the presence of the independence property one might not have expected to find any significant relation, regardless of its sign.
Taking a somewhat broader perspective, there are a couple of factors that bear consideration in the context of interpreting the results (or lack thereof).
First, the levels of over-allocation of allowances during phase II has left the system flushed with cheap allowances in both phase II and phase III. The incentives for emission abatement may have been very weak to begin with. In that scenario, the effect of the free allowance allocation scheme would be very weak also. Considering such circumstances, the slightest hint of a violation of the independence property would be a feat in and of itself.
Second, firms may hold multiple accounts and, in this paper, no further effort is made to attribute multiple accounts to a single firm where appropriate. This was done for reasons of practicality due to lack of objective identifiers, but it may have resulted in a weaker dataset compared to one that does correctly assign all installations to the right firm. The problem of not correctly identifying all installations that belong to a single firm is that it leads to having multiple observations in which the coverage might not be accurate within the theoretical framework.
25
abatement measures, but those efforts might only be observable in phase III, where it turns out that those with lower phase II coverage only now start outperforming their peers that received higher coverage.
26
6 Conclusions
This paper set out to investigate the validity of the independence property of the allowance allocation policy in the European cap and trade program (EU ETS) that is implied by the Coase theorem. This independence property states that the initial allocation of allowances does not affect the final allocation of allowances.
A possible absence of the independence property in the EU ETS is an important argument that speaks in favor of the new policy direction that the EU has, since phase III, committed to. That is, if the independence property does not hold, then free allocation of allowances would lead to less efficient abatement (i.e. abatement at a higher societal cost) compared to auctioning23.
Granted, for the EU ETS, the discussion surrounding free allocation versus auctioning of emission allowances seems to be decided in favor of the auctioning proponents already. There are, too, a plethora of other solid arguments that speak in favor an auctioning policy over a free allocation policy, and it seems that European policymakers have taken up on those and decided to work towards full auctioning. Still, possible insights in whether allocation policy matters to the efficiency of abatement can add to discussions currently raging in other cap and trade systems.
Previous literature on this subject is rather sparse and has, thus far, found no indication that the independence property does not exist in the EU ETS. However, this paper has taken a fresh approach by exploiting the change in free allowance allocation policy that was introduced in phase III, and for which data was not available until quite recently.
Furthermore, a key difference between this paper and other literature has been the acknowledgement that installations themselves are not independent actors but, rather, that their owners (i.e. firms) are. This is an important consideration to make as firms quite often control multiple installations.
An examination of what might cause the independence property to fail suggests that there is some reason to suspect that the independence property might not hold up in the EU ETS, though some factors may be more relevant than others. For example, although the allowance market for the EU ETS appears to work very smoothly and without significant transaction costs, and even though there does not seem to be much need for concern regarding firms with market power, some
23 That is, assuming efficient auctioning practices. In Deutsche Bank Research (2010) merits and drawbacks of
27
policies in phase I and II did seem to promote the idea that free allowances could be lost by emitting less.
Yet, the empirical results presented in this paper provide no strong evidence to suggest that the independence property is, indeed, not featured within the EU ETS. Despite the best of efforts, and in addition to uncontrollable factors, some of the results hint at problems with the used models themselves. In particular, while the expected effect of allowance coverage in an absence of the independence property has not been found, it should also not be expected to find a significant effect working in the opposite direction in the presence of the independence property.
28
References
Coase, R. H. (1960). The Problem of Social Cost. The Journal of Law and Economics, III, 1-43. Deutsche Bank Research. (2010). Bidding For the Better - EU Emissions Trading Scheme moves to
auctioning.
Ellerman, D., & Joskow, P. L. (2008). The European Union's Emission Trading System in perspective. European Commission. (2015). Carbon Market Report 2015.
European Commission. (2016). EU ETS Handbook.
Goulder, L. H., Hafstead, M. A., & Dworsky, M. (2010). Impacts of Alternative emissions allowance allocation methods under a federal cap-and-trade program. Journal of Environmental Economics
and Management, 60, 161-181.
Hahn, R. W., & Stavins, R. N. (2011). The Effect of Allowance Allocations on Cap-and-Trade System Performance. The Journal of Law and Economics, 54, 267-294.
Montgomery, W. D. (1972). Markets in Licences and Efficient Pollution Control Programs. Journal of
Economic Theory(5), 395-418.
Moore, M. R., & Holland, S. P. (2013). Market design in cap and trade programs: Permit validity and compliance training. Journal of Environmental Economics and Management, 66, 671-687. Peeters, M., & Weishaar, S. (2009). Exploring Uncertainties in the EU ETS: "Learning by Doing''
Continues Beyond 2012. Carbon & Climate Law Review, 1, 88-101.
Point Carbon. (2008). EU ETS Phase II – The potential and scale of windfall profits in the power sector. Reguant, M., & Ellerman, A. D. (2008). Grandfathering and the endowment effect - An assessment in the
context of the Spanish National Allocation Plan. Center for Energy and Environmental Policy
Research.
Stavins, R. N. (1995). Transaction Costs and Tradable Termits. Journal of Environmental Economics and
i
Appendix
Table 2 - Alternative regression results arc (abatement rate change) based on
alternative coverage measures
arcin [1] arcex [2] arcinst [3] Alt_arcin [4] Alt_arcex [5] Alt_arcinst [6]
Adj. R2 .0562 .0623 .1121 .0528 .0616 .0843 Obs. 4096 4096 6577 4096 4096 6577 cII -.0303 * -.0285 * -.3642 ** .0383 ** .0403 ** -.1983 ** (.0134) (.0125) (.0200) (.0126) (.0121) (.0206) cIII .0293 * .0351 ** .1870 ** -.0550 ** -.0544 ** .0306 * (.0138) (.0129) (.0147) (.0130) (.0125) (.0151) s .0081 .0064 .0354 ** .0054 .0021 .0254 ** (.0050) (.0046) (.0044) (.0047) (.0045) (.0045)
ii
The industry group variable dataset adjustment
In the original dataset as obtained from the European Union Transaction Log, one of the variables groups installations according to their main activity. However, a change in the classification policy in phase II that created a more fine-grained set of activities has not retroactively implemented, leaving some installations with the older labels still attached. As there is no practical way of appropriately classifying these installations using the more fine-grained set of activities, the old labels have been used instead. Table 3 shows the activity types found in the raw dataset and the reclassified labels that are used instead.
Table 3 – industry groups as found in the EUTL dataset and their aggregation
Combustion of fuels
Combustion installations with a rated thermal input exceeding 20 MW 1. Combustion of fuels Production of coke
Coke ovens 2. Production of coke
Production of paper or cardboard Production of pulp
Industrial plants for the production of (a) pulp from timber or other fibrous materials (b) paper and board
3. Industrial plants for the production of (a) pulp from timber or other fibrous materials (b) paper and board
Installations for the manufacture of ceramic products by firing, in particular roofing tiles, bricks, refractory bricks, tiles, stoneware or porcelain Manufacture of ceramics
4. Installations for the manufacture of ceramic products by firing, in particular roofing tiles, bricks, refractory bricks, tiles, stoneware or porcelain
Manufacture of glass
Installations for the manufacture of glass including glass fibre
5. Installations for the manufacture of glass including glass fibre
Production of lime, or calcination of dolomite/magnesite Production of cement clinker
Installations for the production of cement clinker in rotary kilns or lime in rotary kilns or in other furnaces
6. Installations for the production of cement clinker in rotary kilns or lime in rotary kilns or in other furnaces
Installations for the production of pig iron or steel (primary or secondary fusion) including continuous casting
Production of pig iron or steel
Production or processing of ferrous metals
7. Installations for the production of pig iron or steel (primary or secondary fusion) including continuous casting
Refining of mineral oil
Mineral oil refineries 8. Refining of mineral oil
Production of bulk chemicals Production of nitric acid Production of ammonia
Production of hydrogen and synthesis gas Production of soda ash and sodium bicarbonate Production of glyoxal and glyoxylic acid Production of adipic acid
9. Production of bulk chemicals
Production of carbon black 10. Production of carbon black
Production or processing of gypsum or plasterboard 11. Production or processing of gypsum or
plasterboard
Manufacture of mineral wool 12. Manufacture of mineral wool
Production of secondary aluminium Production or processing of non-ferrous metals
Production of primary aluminium 13. Production or processing of non-ferrous metals
Metal ore roasting or sintering
Metal ore (including sulphide ore) roasting or sintering installations
14. Metal ore (including sulphide ore) roasting or sintering installations
Other activity opted-in pursuant to Article 24 of Directive 2003/87/EC 15. Other activity opted-in pursuant to Article 24 of
