Did the EU ETS Make a Difference? An Empirical Assessment Using Lithuanian Firm-Level Data

We use a panel dataset of about 5,000 Lithuanian firms between 2003 and 2010, to assess the impact of the EU ETS on the environmental and economic performance of participating firms. Using a matching methodology, we are able to estimate the causal impact of EU ETS participation on CO2 emissions, CO2 intensity, investment behaviour and profitability of participating firms. Our results show that ETS participation did not lead to a reduction in CO2 emissions, while we identify a slight improvement in CO2 intensity. ETS participants are shown to have retired part of their less efficient capital stock, and to have made modest additional investments from 2010. We also show that the EU ETS did not represent a drag on the profitability of participating firms.


INTRODUCTION
As the post-Doha political debate moves on from the old Kyoto framework towards a new post-2020 global policy regime, flexibility mechanisms, that theoretically facilitate the achievement of policy goals at least cost, take center stage.In this context, emissions trading systems, with the possibility they offer to link emissions reduction efforts across space and time, seem poised to become one of the pillars upon which future cost-effective mitigation efforts will be built.As a consequence, there is considerable interest in understanding how these market-based instruments perform in practice.In particular, there is a growing appetite on the part of policy makers for empirical analyses that shed light on how participation in emissions trading systems affects the economic and environmental performance of regulated entities.To date, however, only a handful of studies have attempted such an analysis.
In this paper we exploit a previously unexplored firm-level dataset to investigate these questions in the context of the European Union Emissions Trading System (EU ETS).A rich dataset of Lithuanian firms spanning the years between 2003 and 2010 allows us to investigate the impact of the EU ETS on emissions, profitability and investment decisions.
The unique features of the data allow us to directly compare emissions between ETS and non-ETS firms, something that, to the best of our knowledge, has not been possible before.While this is the main motivation for our choice of Lithuania as a case study, we believe that our findings can be informative about the behavior of firms elsewhere in the EU ETS.Indeed, Lithuania's ETS coverage is sufficiently broad to suggest that the average treatment effects we estimate as a result of our empirical analysis-for example, as refers to emissions and emissions intensity-are broadly transferrable to firms operating elsewhere in the EU.Lithuania has a well-developed financial sector, it officially joined the Eurozone on 1 January 2015, and it is also an extremely open economy 1 ; its larger firms, such as the ones in our dataset, operate across several markets, including ones outside of the EU. 2 Hence, information about the competitiveness challenges faced by Lithuanian firms is likely to be highly indicative of the challenges facing similar firms elsewhere.Finally, Lithuania has been one of the fastest growing economies in the EU for over a decade and, in this sense, the results from our analysis are also likely to be relevant to policy makers in other fast growing economies contemplating the introduction of emissions trading schemes.
Our analysis adds to the scant ex-post empirical literature on the EU ETS by investigating the causal effects of the EU ETS on firm-level environmental and economic performance. 3Exploiting the peculiar features of our data, we are able to make the most of the design characteristics of the EU ETS 4 , and to construct a reliable counterfactual, i.e. believable estimate of the outcome variables that would have been observed in the absence of the EU ETS.Given that only a subset of firms in each sector were required to participate in the EU ETS, we can directly compare environmental and economic outcomes of ETS firms to those of similar firms outside the scheme.One important advantage of this approach is that the counterfactual estimates are free of the potentially confounding effects of changing economic conditions at country level, industry-wide production trends, and technological change.
The contribution of this paper is threefold.Firstly, we are the first-to the best of our knowledge-to be able to compare the evolution over time of CO 2 emissions by ETS firms to that of firms outside of the EU ETS.We do this within a classic non-experimental program evaluation framework, using matching algorithms to derive causal inferences on the impact of the program.Secondly, by exploiting a richer dataset than previously done in the literature, we are able to investigate the effect of the EU ETS on the economic performance of firms in greater detail.For example, we complement the analysis of firms' profitability with a discussion of investment decisions.Finally, our dataset spans the first and (most of) the second phase of the EU ETS (2004)(2005)(2006)(2007)(2008)(2009)(2010), thus extending and updating previous results.
Our analysis starts by assessing the environmental consequences of the EU ETS.We first look into actual emissions reductions-so called "abatement"-by comparing actual CO 2 emissions with counterfactual CO 2 emissions.Only a handful of previous studies have documented aggregate emissions abatement in the first phase of the EU ETS.Ellerman and Buchner (2008) calculate that 130-200 Mt of CO 2 were abated in 2005 and 140-220 Mt of CO 2 2006, across all EU member states.Anderson and Di Maria (2011) improve on these results using more refined data for 2005-2007 and estimate overall abatement at 247 Mt of CO 2 during the first phase.They also find evidence of emissions "inflation," however, they show that several countries had actual emissions in excess of the counterfactual.Delarue, Ellerman, and D'Haeseleer (2010), Pettersson, So ¨derholm, and Lundmark (2012), Linden, Ma ¨kela ¨, and Uusivuori (2013), and Widerberg and Wra ˚ke (2009) explicitly focus on abatement in power generation.Delarue, Ellerman, and D'Haeseleer (2010) analyze the power sector's CO 2 short-term abatement possibilities through fuel switching.The authors estimate abatement of the European power sector to be in the range of 34.4-63.6 Mt of CO 2 in 2005, and 19.2-35 Mt in 2006. Pettersson, So ¨derholm, and Lundmark (2012) use a Generalized-Leontieff model of the electricity sector calibrated on data from eight western European countries-Austria, Belgium, Germany, Ireland, Italy, Netherlands, Spain and the UK-over the time period 1980-2004 to support the view that the elasticity of substitution, especially between oil and gas, may be rather high in the short run.Their model suggests that an allowance price of US$90 would lead to a two per cent decrease in CO 2 emissions, via fuel-switching only (using 2004 as the base year, this is equivalent to 18Mt of CO 2 ).Linden, Ma ¨kela ¨, and Uusivuori (2013) find that the EU ETS has short term impacts on the fuel mixes of the energy plants in Finland.The elasticity of substitution between fossil and non-fossil fuels is larger under the ETS.Widerberg and Wra ˚ke (2009) look at the effect of the carbon price on the CO 2 emissions intensity of the Swedish electricity sector for the period 2004-2008.They find no statistically significant link between the price of CO 2 and CO 2 emissions, and conclude that it is unlikely that there are significant volumes of low-cost CO 2 abatement possibilities with short response times in the Swedish electricity sector.Finally, Abrell, Faye, and Zachmann (2011), use CITL data to test whether the EU ETS induced acceleration in emissions reductions, and find some evidence to the positive.In the current paper, for the first time we are able to look at abatement at the firm level, thanks to the unique characteristics of our dataset.
We then investigate the economic consequences of carbon pricing.Emission reductions generally entail costs as resources need to be reallocated from production activities to emissions reduction.In principle, these costs depend on the stringency of the carbon constraint, the tighter the cap the higher the marginal abatement cost or CO 2 price, and the greater the cost, all else being equal.Since the EU is the only region in the world in which a carbon price is applied on such a scale, the EU ETS might have serious implications in terms of the loss of competiveness it causes to ETS firms versus both European firms outside of the EU ETS, and their non-European competitors.We address the impact of the EU ETS on competitiveness by looking at a firm's ability to generate profits.In doing this, we can also gauge the implications of the free permits allocation for firms' profitability.Our dataset also allows us to investigate whether ETS firms exhibit a different behavior relative to their non-ETS counterparts when it comes to investment in tangible capital.This is particularly important in terms of the future competitive position and profitability of regulated firms.
To date, only a few studies have analyzed the impact of the EU ETS on firm competitiveness, profitability or general economic performance ex-post. 5Potentially negative impacts on competitiveness have understandably been the main concern of firms within the EU ETS, and have been the focus of most empirical analyses.Energy-intensive and trade-exposed industrial activities namely cement, glass, iron and steel, paper and pulp, refining and aluminum, are arguably the most exposed among the EU ETS sectors in terms of international competition.Empirical research using trade data (imports into the EU as well as exports from the European Union), however, has found The rest of this paper proceeds as follows.Section 2 provides some background on Lithuania's experience with the EU ETS; Section 3 outlines the research design and the data sources used in this analysis; Section 4 presents and discusses the empirical results; finally, Section 5 summarizes and concludes.

LITHUANIA IN THE EU ETS
Lithuania's overall CO 2 emissions increased between 2000 and 2007-peaking at 15.81 Mt-and subsequently decreased to just below 14Mt in 2010 (See Table 1).In per-capita terms, emissions increased from 3.37 tonnes in 2000 to 4.12 tonnes in 2010.On average, the ETS sectors contributed around 43 per cent of total CO 2 emissions between 2005 and 2010.
Under the Kyoto agreement, Lithuania had a target to reduce its CO 2 emissions by eight per cent relative to the 1990 level by 2008-2012.In reality, the fact that Lithuania gained independence from the Soviet Union in 1990 and subsequently underwent major structural changes during its transition to a market economy explains why emissions in Lithuania have been significantly  (Zapfel 2007).
Table 2 summarizes the first period permit allocation for Lithuania.In the first period, the NAP allocated over 36 million EUAs to 93 installations.As a consequence, Lithuanian installations had a net long position of 15.40 million EUAs (almost 45 per cent when compared with the initial allocation excluding unexecuted allowances). 7 The significant over-allocation of permits to Lithuanian's firms came about as the result of wrong expectations in terms of the impact of changes in the Lithuanian electric power market following the closure of the first reactor unit of the Ignalina Nuclear Power Plant (Ignalina NPP) on 31 December 2004.Indeed, at the time the NAP was being drafted, there was a widespread belief that this would imply a large increase in fossil-fuel-based generation.This was the main reason behind the increase in the number of EUAs allocated to Lithuanian energy enterprises.The extent of this increase is illustrated by S ˇtreimikiene (2008): in 1998, CO 2 emissions from powergenerating firms participating in the EU ETS accounted for 5.19 Mt, while the annual first period allocation of those firms was 7.59 Mt of CO 2 .In fact, during the first trading period, the actual verified emissions for combustion installations, which are mainly owned by power generating firms, were on average about 3.63 Mt of CO 2 per annum.
The two main reasons underlying this shortfall in emissions were that the second unit at Ignalina NPP operated very efficiently throughout 2005, and the time spent for regular maintenance was kept to a minimum, and the concomitant increase in natural gas prices in 2005.Given the increased cost of domestic gas-fired generation, the growing electricity demand was met by electricity imported from Russia and Estonia, rather than by increases in domestic supply (S ˇtreimikiene 2008).Electricity produced by Ignalina NPP decreased from 15,102 GWh in 2004 to 9,914 GWh on average during the period 2005-2009, but electricity import increased from 4 ,293 GWh in 2004 6.According to GHG inventory data under the United Nations Framework Convention on Climate Change, in 2007 GHG emissions in Lithuania were down 53 per cent on the 1990 level.During this period, real GDP increased by 101 per cent.
7. The CITL does not incorporate the data on allocated reserves.See McGuinness and Trotignon (2007) for a detailed discussions on how information on the reserves can be combined with the CITL data.Source: Adapted from Ellerman and Trotingnon (2009).
to 5,546 GWh on average in the same period.It is also worth mentioning that Lithuania remained a net exporter of electricity through 2009 (See Figure 1).As a consequence of the generous allocation of permits, Lithuanian firms were able to sell large amounts of allowances and benefit from wind-fall profits.Table 3 reproduces information from Ellerman and Trotingnon (2009) and summarizes information on the flows of EUAs from Lithuania.Not surprisingly, Lithuania was a net exporter of allowances in the first trading period, exporting almost 11 million EUAs.Three quarters of the Lithuanian EUAs were transferred in 2007 when the price of carbon was at its lowest.Assuming that surrendered allowances were acquired at the average price for the intervals before and between the end of April-dates when allowances were to be surrendered and at the time when these flows were monetized-we conclude that Lithuanian installations accrued most of their revenue from selling EUAs in 2006 when the price of EUAs averaged at 9.57 EUR.We can confirm that Lithuanian installations monetized their surpluses at very high prices thanks to the information collected by the National Audit Office of Lithuania on Lithuanian firms partly controlled by the Lithuanian Government.For these firms, we also have information about the way they used the revenue from allowances' sales in the first trading period.Table 4 shows that the average price received per EUA ranges from 28.1 EUR to 5.79 EUR.The Lithuanian power plant, which was expected to cover the energy shortage due to a closure of unit 1 at Ignalina NPP, sold the 4.65 Mt of EUAs out of received 7.40 Mt in total.It is also important to note that only about a tenth of this revenue was used for emission abatement.
The operation of ETS firms during the first period provided useful information that was used as a basis for the second period NAP (2008)(2009)(2010)(2011)(2012).Consequently, the annual second period allocation for Lithuanian installations was significantly reduced, even allowing for the envisioned closure of the Ignalina NPP from January 2010 onwards.The NAPs for the second trading period were approved before the start of 2008.
Despite the tighter second period allocation, the compliance results for the second trading period (2008)(2009)(2010)(2011)(2012) reveal that in total ETS firms in Lithuania were still significantly over-allocated (see Table 5).During 2008-2012 ETS firms in Lithuania received 39.64 million of EUAs.Lithuanian installations had the net long position of 10.03 million of EUAs (25.31 per cent when compared with the initial allocation).Again, this suggests that the second period allocation did not create strong incentives for ETS firms to mitigate their CO 2 emissions, especially in light of the low level of EUAs prices over most of this period and of the increasing realization of over-supply of permits after the financial crisis of 2008-2009.

Empirical Framework
The main goal of this paper is to empirically estimate the changes in a number of firms' environmental and economic performance indicators relative to what would have occurred if the EU ETS had not been implemented.Since we want to identify the causal effect of the EU ETS, we exploit the unique design features of the EU ETS to construct tenable and transparent estimates of counterfactual emissions and other outcome variables.As a counterfactual we use econometrically adjusted observed environmental and economic outcome variables at firms that were not subject to the EU ETS over the same period.
A firm in the EU can find itself in one out of two regulatory states: it may be regulated under the EU ETS, or it may find itself outside of the ETS remit and hence be allowed unconstrained CO 2 emissions.Let D be the indicator variable that identifies a firm's participation in the EU ETS.Hence, D i is equal to 1, if the i th firm is in the EU ETS (that is the firm is "treated").A firm, i, that is instead outside of the treated group is assumed to be unaffected by the EU ETS, and is identified by D i = 0.All firms with D i = 0 are said to belong to the control group.Y (1) and Y (0) denote it it potential outcomes at firm i and time t, conditional on participation and non-participation, respectively.
We are interested in estimating the average treatment effect on the treated: where t represents any year following the introduction of the EU ETS and α TT measures the average treatment effect of the EU ETS on the desired outcome variable (e.g. the annual firm level CO 2 emissions).CO 2 emissions and other outcome variables of ETS and non-ETS firms are observed prior to the implementation of the EU ETS and over several years following its introduction.Firm-level emissions data collected from ETS firms during the years following the introduction of the program can be used to identify i.e. we do not know what would have happened to ETS firms, had they not participated in the scheme.To overcome this limitation, we take advantage of a key feature of the EU ETS, i.e. the fact that the EU ETS regulates only a subset of the largest CO 2 emitters located within the 27 EU member states.Moreover, the remaining firms are not subject to any other type of carbon constraint, at least in Lithuania.The incomplete programme participation provides us with a potential comparison group as we are able to use econometrically adjusted observations on outcome variables relative to non-participants, to estimate the unobserved counterfactuals.
The simplest estimate of α TT is obtained using standard differences-in-differences (DiD) estimators.These estimators, however, may be biased if the variables related to firm-level outcomes vary significantly across the treatment and comparison groups.To reduce this potential bias we utilize the observable differences across ETS participants and non-participants to estimate α TT by using semi-parametric matching estimators. 9 Matching estimators, which are used extensively in non-experimental program evaluation, are an extension of standard regression approaches.Our general estimation strategy follows Heckman, Ichimura, and Todd (1997) and Heckman et al. (1998) who introduced the following DID matching estimator: where I denotes the set of program participants; I denotes the set of non-participants; and N is 1 0 1 the number of firms in the treatment group.The participants are indexed by j; the non-participants are indicated by k.The weight placed on individual k when constructing the counterfactual estimate for treated facility j is w jk .Different matching estimators adopt different approaches to defining the weights w jk used to scale the contribution of each participant.In general, when the observable characteristics, X i , of an untreated unit k are closer to the characteristics of a treated facility j (relative to other facilities in the control group), the untreated unit k is weighted relatively more heavily in the construction of a counterfactual estimate for unit j.
Since the seminal work of Rosenbaum and Rubin (1983), propensity scores (i.e., the conditional probability of treatment) are used, rather than conditioning on all the relevant covariates.An important finding in the literature is that, if unconfoundedness holds, conditioning only on the propensity score assures the independence of D i and Y i (0) (Imbens, 2004). 10While a variety of the propensity-score-based matching algorithms are available, 11 in what follows, we use the nearest neighbor (NN) and Kernel matching estimators.The NN estimator is the most straightforward matching estimator.An appropriate facility from the control group is selected as the matching partner for a treated facility on the basis of its exhibiting the closest propensity score.In our case, we use NN matching with replacement since each untreated firm can be used more than once as a match.Matching with replacement involves a trade-off between bias and variance.By allowing replacement, the average quality of matching will increase and the bias will decrease, especially when one has a relatively small group of comparable facilities in the control group, as we do.
The main difference between Kernel matching and NN matching is that instead of using only a few observations from the comparison group, weighted averages of all individuals are used in the control group to construct the counterfactual outcome.That is, it constructs a match for each treated entity using a kernel-weighted average over multiple entities in the control group.Smith 9.As the EU ETS already includes the biggest emitters, it is unlikely to be able to find good non-ETS counterparts in terms of absolute emissions.The criteria according to which firms are included in the EU ETS along with a variety of matching algorithms (see below), however, allows us to find close non-ETS matches.The inclusion criteria are based on energy capacity and production at the installation level.In some cases an installation corresponds to a whole firm, but in many other cases it represents only a part of it.This suggests that there are firms which do not qualify for inclusion in the EU ETS based on the individual installation, but would qualify if the selection criteria were applied to the firm as a whole.Hence, it is the case that some of the non-ETS firms are closely comparable with ETS firms in terms of emissions at the firm level.In what follows, we discuss how we impose the common support property (e.g., Imbens, 2004), thus assuring that we have appropriate controls for even the largest ETS installations.
10. Unconfoundedness, a term introduced by Rubin (1990), indicates the situation in which adjusting for differences in a fixed set of covariates removes biases in comparisons between treated and control entities, thus allowing for a causal interpretation of those adjusted differences.
11. See Caliendo and Kopeinig (2008) for a non-technical discussion about the properties of various matching estimators.
Copyright © 2016 by the IAEE.All rights reserved.
and Todd (2005) note that Kernel matching can be seen as a weighted regression of the counterfactual outcome on an intercept with weights given by the kernel weights.The weights depend on the distance between each control group observation and the treated observation for which the counterfactual is being estimated.Thus, one major advantage of this approach is the lower variance, which is achieved because more information is used.The main demerit of this approach is that it utilizes all observations, including ones that are objectively poor matches.This problem is eased by properly imposing the common support condition.

Data
Our primary data come from the annual "Sample survey of non-financial enterprises (F-01)" administered by Statistics Lithuania (LS).The survey collects data on the main financial indicators for the sampled enterprises.We use eight waves of the survey (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010), whose sample sizes vary between 8,000 and 17,000 firms.
Sampled firms belong to NACE (Statistical classification of economic activities in the European Community) Revision sectors 10-40.We exclude firms belonging to NACE 4012 (Transmission of electricity), 4013 (Distribution and trade of electricity) and 4022 (Distribution and trade of gaseous fuels through mains) as these firms are service providers such as gas and electricity distributors.
The dataset includes fuel purchases, turnover, capital stock, and profits.Importantly, the dataset also includes a breakdown of fossil fuels expenditures by fuel type.This unique feature allows us to infer quantities purchased from the expenditure data.Applying average emissions coefficients to the estimated quantities, we are able to obtain estimates of CO 2 emissions by both ETS and non-ETS firms (see the Appendix).This allows us to compare changes in CO 2 emissions by ETS firms with the behavior of unregulated firms outside the ETS. 12 Unfortunately, the disaggregated fossil fuels expenditure series have been discontinued from 2008; hence we are only able to conduct our analysis of CO 2 emissions until the end of phase 1, in 2007.
Due to the sampling methodology, the data set is strongly unbalanced.To avoid biasing our estimations, we restrict our analysis using the largest possible balanced panel.As a consequence of this adjustment, some firms drop from our sample.
A complete list of the variables used in the analysis, together with the usual set of descriptive statistics is presented in Table 6.The data are summarized for the balanced sample running from 2003 until 2010, broken down by EU ETS participation status.
The data exhibit notable differences between ETS and non-ETS firms.As expected, ETS firms on average emit more CO 2 emissions, produce more output, and are more capital intensive.On the other hand, ETS firms on average are less profitable than non-ETS ones.Did the EU ETS Make a Difference?/ 79 Copyright © 2016 by the IAEE.All rights reserved.Notes: Seven number of blocks optimally selected, significance level 0.01.
Since the choice of the observable covariates in the propensity score model must satisfy the unconfoundedness assumption, the selection of covariates is crucial.All the important variables that influence both the participation decision and the outcome variables should be included.Hence, both economic theory and the policy setting must be used as a guide.In addition, only variables that are unaffected by participation should be included in the model.To ensure this, we choose variables that are either fixed over time, or measured before participation.
In our study, the propensity scores are measured using data for 2004.The explanatory variables include the amount of fossil-fuel-based energy used by the firm, the stock of tangible capital assets, the firm's turnover, and a dummy identifier for whether the firm belongs to the NACE 40 industries. 13We enforce a common support or overlap condition.This ensures that any combination of characteristics observed in the treatment group can also be observed among firms in the control group.Balance is achieved and there is a significant overlap in the propensity scores of the treatment and comparison groups (see Table 7). 14 Table 8 presents the results of the propensity score measurement.As expected, firms with higher consumption of fossil fuels and more tangible capital assets are more likely to be included in the EU ETS.Additionally, firms belonging to the NACE 40 industries are also more likely to be engaged in emissions trading.The volume of turnover has a negative and significant effect, albeit only at the ten per cent significance level.

Average Effects of the EU ETS on ETS Firms
This subsection reports the results of the Kernel and NN matching specifications reflecting the EU ETS impacts.Although the European Emissions Trading Directive was ratified in October 2003, the Lithuanian NAP was approved only in the second half of 2004.Therefore, we use the year 2004 as the pre-treatment year.The outcome variables in the year 2004 are then compared with their counterparts in subsequent years (2005, 2006, 2007, 2008, 2009 and 2010 15 ).To better 13.NACE 40: Electricity, gas, steam and hot water supply.14.Our estimation is carried out in STATA, using the "pscore" procedure developed by Becker and Ichino 2002.Adding option "comsup" to the estimation, we ensure that balancing is achieved.The statistical procedure implemented is extremely demanding in that the balancing property is not rejected only in the case that it holds for every component of the conditioning vector (see Becker and Ichino 2002 for details).Additional details on the procedure and the complete test results are available from the authors upon request.
15.Some of the outcome variables, most notably CO 2 emissions, are only available until 2007.
Copyright © 2016 by the IAEE.All rights reserved.understand the dynamics of the effects, we also compute year-on-year changes.Treatment effects are calculated using a "levels" specification.

Total CO 2 Emissions and CO 2 Intensity
As the primary goal of the EU ETS is the reduction in absolute GHG emissions, the first question to be explored is whether the EU ETS effected any significant changes in total CO 2 emissions.Figure 2 illustrates the evolution of average CO 2 emissions among ETS firms between 2003 and 2007.Emissions hovered around the pre-treatment average well into Phase I of the ETS, before increasing in 2007.The results from our matching estimation show that these dynamics are not peculiar to ETS firms, but are rather mirrored by comparable firms outside of the treatment group.The results in Table 9 shows that changes over time in CO 2 emissions among ETS firms, where not significantly different in a statistical sense from changes that occurred among firms in the control group.We can conclude that, in terms of overall CO 2 emissions, the introduction of the EU ETS did not significantly affect the behavior of ETS firms, relative to non-ETS ones.Given that Lithuanian's ETS firms were significantly over-allocated in the first trading periods, and that the price of allowances was already very low by the time allowance trading started to take place in earnest, these finding are hardly surprising.
These absolute measures, while interesting from the point of view of the environmental integrity of the policy, do not allow us to discriminate between changes in production levels and other adjustments the firms might have made, for example, in terms of their fuel mix or their production technologies.To gain some insight into this second group of factors, we next look at changes in CO 2 intensity, measured as the ratio of CO 2 emissions over turnover.
Figure 3 shows that, contrary to the trendless emissions path, the emissions intensity of turnover for ETS firms in our sample continued to creep up until 2005, started decreasing in 2006 and fell markedly in 2007.Our matching exercise (see Table 9) confirms this development.In particular, the year-on-year changes confirm the information derived from the previous picture, and inform us that these changes were not significantly different from the behaviour of non-ETS firms, with the exception of the marked reduction in 2007, which is significantly larger than the comparable change for non-ETS firms.Thus, ETS firms seem to have used the first years of the scheme operation to (slowly) learn how to improve on their environmental efficiency, possibly opting to reduce their use of the most CO 2 intensive fuels,16 purchasing electricity rather than self-generating, and investing in cleaner technology.

Fuel Mix and Electricity Purchase
To investigate which of these strategies have been followed by Lithuanian ETS firms, we start by looking at purchases of fossil fuels by the ETS firms in our sample.
As expected, the vast majority of fossil-fuel based energy generation in Lithuanian is based on natural gas, with coal and oil distant second and third.Figure 4 illustrates that the share of oil has been consistently decreasing over time, while coal has increased slightly.This is suggestive of a progressive shift away from expensive oil into cheaper coal and cleaner gas.The increased reliance on gas by ETS firms is likely to have been the result of the need to compensate for the decrease in electricity output following the closure of Unit 1 of the Ignalina NPP on 31 December 2004.The increase, however, proved to be not as large as it had been previously feared.The natural gas price increase, naturally led to a rise in energy production costs, thus encouraging electricity imports from Russia and Estonia rather than domestic production (S ˇtreimikiene 2008).We find a confirmation of this hypothesis by looking at the increase in electricity purchases by ETS firms over the period 2005-2007 (see Figure 5).This shift away from CO 2 -intensive energy carriers into natural gas, and from fossil-based generation into imports can certainly explain the decrease in CO 2 intensity discussed above.We next look at the role of investment in this process.

Investment
We now ask the question whether Lithuanian ETS firms invested in new technology following the commencement of carbon trading in 2005, and whether their behavioral change can be attributed to the EU ETS.The empirical literature in this area is very scant.Using a survey of Irish ETS firms over the period 2005-2008, Jaraite ˙, Convery, and Di Maria (2010) find suggestive evidence that ETS firms started investing to improve their carbon performance and achieve compliance under the (perceived) tight Irish permits allocation, already in preparation of the first phase.Anderson, Convery, and Di Maria (2010) use the same Irish survey data and find that the EU ETS has influenced the way investments in capital and infrastructure are planned in almost half of the surveyed firms.They also report that during the first phase there was a significant amount of technology adoption.Indeed, 50 per cent of the Irish firms in their sample report employing some form of new machinery or equipment that contributed to decreasing their CO 2 emissions.Lo ¨fgren, Wra ˚ke, Hagberg, and Roth (2013), on the other hand, fail to find any statistically significant impact of the EU ETS on the investment decision of regulated Swedish firms.Their focus is, however, slightly different from the papers above, as they only analyze investment in carbon mitigating technologies in a sample of Swedish firms between 2002 and 2008.Our data do not include investment data per se, but we have information on each firm's total tangible capital assets over time.
In what follows, we use changes in these assets as a proxy for net capital investment.
Figure 6 shows that the average value of tangible assets among ETS firms in our sample declined steadily throughout the first phase.Thus, it seems unlikely that Lithuanian firms improved their carbon efficiency due to the introduction of new technology.
The figure, together with the results reported in Table 9, indicate that this trend was, however, reversed in 2009, and that this change is exclusive to ETS firms in our sample.
These results are consistent with the view that the introduction of the EU ETS did cause the retirement of old and less efficient tangible assets during the first trading years.In the second phase, instead, the evidence indicates that ETS firms started investing in new capital equipment.One factor that might have led to the 2010 increase is the significant legislative change that took place in 2009.On July 7, 2009 the Lithuanian Parliament passed law XI-329 (Seimas of the Republic of Lithuania 2009), which limited the possibility to recycle revenues from the sale of EUAs.The law in fact mandates that all revenues received from the sale of EUAs should be earmarked to be spent on environmental measures.

Profitability
The final part of our analysis focuses on assessing the effect of the EU ETS on Lithuanian ETS firms' profitability.Indeed, one of the main concerns raised by the introduction of the EU ETS Copyright © 2016 by the IAEE.All rights reserved.was the possible deterioration of the competitive position of those ETS sectors most exposed to international competition.It is clear, however, that some of the concerns related to profitability would be alleviated by the fact that within the EU ETS permits are grandfathered, i.e. allocated for free to firms in the EU ETS.In this case, each entity is faced with a clear trade-off between using the permits for compliance purposes (and incurring the opportunity cost of the foregone sale) and selling them on the market (and incurring the cost of abatement).It is this trade-off that generates abatement incentives for efficient firms, who then benefit from the proceeds of the sale of the excess allowances.Revenues from the sale of EUAs are potentially large, and may help bolster the profitability of firms, especially in sectors and countries that received a generous allocation of permits.As discussed in Section 2, Lithuanian firms in all sectors were very generously allocated, and, despite the price collapse, benefited from the sale of EUAs to foreign entities (See Table 3).
Our dataset contains several profit measures, based on financial accounting including gross profit (profit from the main production activities), and profit before tax (profit that takes into account net income from other activities).Although there are as yet no approved accounting standards on how the EUAs should be treated in financial statements, it is recommended to treat them as intangible assets (Rimas ˇauskas 2009).Hence, the net income from buying/selling EUAs should be recorded among the net income from the other activities.As a consequence, we choose to analyze the relative profit before tax, which is measured as profit before tax over turnover.In what follows, we simply refer to this measure as "profit".
The average profitability among ETS firms in our sample (Figure 7) and the matching results presented in Table 9 show that ETS firms in Lithuania do not seem to have suffered from their membership of the emission trading scheme in the early stages of the programme.Our results suggest, however, that they might have become less profitable in both 2009 and 2010.The yearon-year changes are similar but with slightly higher significance already in 2007 and 2008.These findings are not surprising, given the amount of over-allocation enjoyed by Lithuanian ETS firms in the first trading phase.The results are also consistent with the findings of Ellerman and Trotingnon (2009) that ETS firms in Lithuania were able to monetize their EUA surplus in the first trading period.As discussed in Section 2, ETS firms in Lithuania exported most of their unused allowances in 2007, but the highest revenue was earned in 2006 (see Table 2).This might explain the dip in 2007.Subsequent drops might instead be explained by the Lithuanian second period allocation being much tighter than the first one.
In essence, our results run counter the statement that the EU ETS exerted a drag on ETS firms' competitiveness and profitability, at least in the context of our case study.In addition to that, although the available data do not allow us to fully understand the pass-through of opportunity costs, we believe that the above findings do not support the hypothesis that costs have been passed through to consumers.At least four reasons can explain it.Firstly, if there had been the pass-through of the opportunity costs into final consumer prices, we would have observed a persistent EU ETS effect on ETS firms' profits.The second reason relates to structure of the markets in which ETS firms operate.ETS firms operating outside the power generating sector, e.g. in the glass, ceramics, refining and food sectors, are more exposed to domestic and international competition, and thus have limited latitude to pass-through opportunity costs.Thirdly, as most of the Lithuanian firms in the EU ETS had more allowances than they needed, they might not have perceived the full opportunity cost of the freely allocated allowances.Finally, in Lithuania energy end-user prices were regulated in the first trading period and beyond, implying that, in the power-generating sector, firms could not have adjusted end-user prices to CO 2 price fluctuations.Nevertheless, a more detailed investigation of these effects would form an interesting topic for future research.

Testing the Common-trends Assumption
Our analysis relies on the important assumption that the trends in the outcome variables over time should be the same across the ETS and non-ETS firms in our sample.Given the systematic differences in firm size between ETS and non-ETS firms, it is very important to test whether this  The bootstrapped standard errors are in the parentheses.

2.
All monetary variables are in real terms.

3.
There are 41 ETS firms in the treated group.

4.
Kernel matching is based on 312 firms in the control group.

5.
NN matching with replacement is based on 22 firms in the control group.

6.
Did the EU ETS Make a Difference?/ 87 Copyright © 2016 by the IAEE.All rights reserved.
assumption holds for our data sample.Since we have only two years of the pre-treatment data, we run a so-called placebo DID matching test for those two years (2003 and 2004).In this test, we treat the year 2003 as a pre-treatment year and the year 2004-as a treatment year.The results17 of the placebo DID matching test show that there are no statistical differences in the outcome variables (namely, CO 2 emissions, CO 2 emission intensity, tangible capital assets and profitability) between the treatment and control groups for 2003-2004.This suggests that the pre-ETS trends in the outcome variables are the same for the ETS and the non-ETS firms in our sample.As they are the same before the EU ETS was implemented, it will lend some support for the assumption that they are the same after the EU ETS started.In addition to that, these results might indicate that expectations regarding the introduction of the EU ETS were not realized before the year 2005.

CONCLUSIONS
In this paper, we presented new evidence on the effects of the EU ETS on participating firms.In particular, thanks to the features of our dataset, we were able to assess the impact of the EU ETS directly on CO 2 emissions and their intensity at the firm level, and to study the behavior of ETS firms as refers to their investments and profitability.Our results indicate that the EU ETS overall did not cause reduction in CO 2 emissions over the whole first trading period.This is understandable, due to the marked over-allocation of the installations in our dataset.We do observe, however, that CO 2 emission intensity decreased between 2006 and 2007, albeit slightly.Unfortunately, the available data do not allow us to investigate whether this decrease was a one-off effect or continued in the second phase of the EU ETS.We also find that Lithuanian ETS firms shifted out of expensive energy carriers, like oil and gas, into coal, which remained competitive due to the low price of allowances after 2006.We argue that two factors external to the participation in the ETS, namely the closing of the first reactor of the Ignalina NPP and the high gas prices, led ETS firms in Lithuania to increase their imports of cheaper electricity from neighboring countries, causing a possible degree of carbon leakage.
Although the EU-wide emission trading system did not seem have encouraged firms to mitigate their CO 2 emissions in the short-run, our analysis suggests that this policy induced the retirement of old (and less efficient) capital stock during the first trading years, and lead to some additional investments into new capital equipment from 2010.The latter effect was probably compounded by the introduction of law XI-329 (Seimas of the Republic of Lithuania 2009), which required the earmarking of allowance sales' revenues for environmental investments.The injection of new, likely more efficient, capital into the existing Lithuanian capital stock suggests that more substantial emission reductions are to be expected in the near future when capital is fully operational.
In terms of economic effects, our results indicate that the EU ETS did not represent a drag on the profitability of Lithuanian ETS firms.This finding, while derived from a small sample of EU ETS firms, is nevertheless consistent with the analyses of, for example, Ellerman, Convery, and de Perthuis (2010), and contributes to alleviate the competitiveness concerns raised by many industry representatives.At the same time, our findings do not support common speculations that the generous permit allocation generated huge windfall profits for the largest polluters.
Overall, our results lend support to the idea that the stringency of the first two phases of the EU ETS was modest at best, as we find that the EU ETS made very little difference in terms of the environmental and economic performance of the firms involved in the scheme.

Figure 1 :
Figure 1: Lithuania's Electricity Balance, 2000-2010 8 8. It is important to note that Lithuania's Greenhouse Gas Emissions Allowance Registry started functioning only in November 2005.Only since then were ETS firms able to trade their EUAs (National Audit Office of Lithuania 2008).This might explain why most of the EUAs were sold in 2006 and 2007 rather than in 2005.Copyright © 2016 by the IAEE.All rights reserved.

Table 4 : Use of Revenues by Lithuania's ETS Firms from Sold Allowances, 2005-2007
3. GHG investments include transaction costs.4.DHP stands for a district heating plant.Sources: National Audit Office of Lithuania (2008) and the authors' calculations.

Table 5 : Lithuania's Compliance During the Second Trading Period, 2008-2012
Source: European Environmental Agency as of 1 July 2013 and the authors' calculations.