firm competitiveness and the eu ets
TRANSCRIPT
Linn Nordin Spring semester 2018 Master’s thesis, 15 ECTS Master’s program in Economics
Firm Competitiveness and the EU ETS An empirical study using Swedish firm-level data
Linn Nordin
Firm Competitiveness and the EU ETS
- An empirical study using Swedish firm-level data -
Linn Nordin, 2018
Abstract
This paper sets out to study the effect of the EU ETS on economic and financial
indicators through which the impacts on firm competitiveness may be observed.
Examining labor productivity, relative profit before tax, turnover, and value added after
the introduction of the EU ETS offers a method for evaluating the effects of the
program on economic performance. The study is based on a unique dataset of Swedish
firms covering the years 2003-2008. To assess the causal impact of the EU ETS on firm
performance the propensity score matching method is combined with the difference-in-
difference approach. The findings suggest a positive and significant impact on labor
productivity, turnover and value added, while a negative and significant effect can be
observed for relative profit before tax.
Keywords: cap-and-trade, economic performance, EU ETS, difference-in-difference,
firm competitiveness, firm-level data, propensity score matching
Acknowledgements I would like to thank my supervisor, Jūratė Jaraitė-Kažukauskė, for her inspiration,
support, and expertise. In addition, I would like to thank Tommy Lundgren for
providing the dataset used in this paper and the attendants of the thesis seminar for their
valuable feedback and comments.
Contents
1 Introduction 1
2 Background 2
2.1 The EU ETS ............................................................................................... 2
2.2 The EU ETS in Sweden ............................................................................. 4
3 Literature Review 4
4 Empirical Methodology 6
5 Data 9
6 Results 11
6.1 Propensity Score and Matching ............................................................... 11
6.2 Estimation of Average Treatment Effects based on Propensity Scores .. 13
6.3 Testing the Common-Trends Assumption ............................................... 15
7 Discussion and Concluding Remarks 15
References 18
Appendix A: A Placebo DID Matching Test 20
1
1 Introduction
The EU Emissions Trading System (EU ETS) is the world’s first and so far largest
emissions trading scheme for greenhouse gases (GHG). The EU ETS is an essential
instrument of the European climate policy, aiming to reduce GHG emissions cost-
effectively (Convery et al., 2008). Since the introduction of the EU ETS, the program
has received much attention worldwide because of its achievements. The program has
not just succeeded in reducing CO2 emissions cost-effectively, but it has also managed
to create a multinational trading system for the right to pollute (Hanley et al., 2007).
Although the cost-effectiveness of the EU ETS is widely acknowledged, the program
has led to a debate among industrial emitters and policymakers regarding the policy’s
potential adverse effects on the competitiveness of EU firms in the global market. The
discussions concern whether emission-intensive firms in the EU will relocate to regions
with no or less carbon restrictions. A relocation of EU firms could increase the risks of
losing jobs and decreasing market shares relative to companies outside the EU (Chan et
al., 2013; Marin et al., 2017).
Emissions trading systems have as a policy instrument the possibility to achieve
significant emissions reduction efforts. Hopefully, can a trading scheme, like the EU
ETS, inspire countries outside the program to future cost-effective mitigation efforts. To
gain an understanding of how participation in the EU ETS affects economic
performance is therefore of interest to policymakers that aim to develop an emissions
trading program. Many empirical studies have investigated the effects of the EU ETS on
firm performance. Still, these studies have reached various conclusions regarding the
impact of the policy on economic performance (e.g., Abrell et al., 2001; Anger &
Oberndorfer, 2008; Chan et al., 2013; Jaraitė & Di Maria, 2016; Marin et al., 2017; Yu,
2011).
This paper aims to study the effect of the EU ETS on economic and financial indicators
through which the impacts on firm competitiveness may be observed. The study is
based on a panel of Swedish firms covering the years 2003-2008, focusing on firms
whose primary activity is in mining and quarrying and manufacturing. Comparisons of
labor productivity, relative profit before tax, turnover, and value added after the
2
introduction of the EU ETS offers a method for evaluating the effects of the program on
firm performance. To evaluate the causal impact of the EU ETS on firm
competitiveness the propensity score matching method is combined with the difference-
in-difference (DID) approach. As discussed above, many empirical studies have
investigated the impacts of the EU ETS on firm performance, however, to the best of
our knowledge this study differs from the papers mentioned in one important way. This
study is one of the first to use a unique dataset including information on CO2 emissions
when evaluating the causal effect of the EU ETS on firm competitiveness. By using the
unique dataset enables us to provide better matches when applying the propensity score
matching method.
Contrarily to the concerns raised by industrial emitters and policymakers regarding the
negative impacts of environmental policy on economic performance, the results from
this study suggest a positive and significant effect on labor productivity, turnover, and
value added. Still, a negative and significant impact can be observed for relative profit
before tax.
The structure of the paper is as follows. Section 2, provides a background of the EU
ETS. Section 3 reviews earlier literature, while Section 4 contains the empirical
methodology. Section 5 describes the data. Section 6 presents and discusses the
empirical results. Section 7 includes a final discussion and concluding remarks.
2 Background
2.1 The EU ETS The European Emissions Trading Directive was adopted in 2003 and launched in 2005,
with the aim to establish a market for the right to pollute as well as a market price for
this entitlement while reducing GHG emissions cost-effectively (Convery et al., 2008).
Today the system operates in 28 EU countries plus Iceland, Liechtenstein, and Norway,
covering approximately 13,000 power stations and manufacturing plants. As from 2012,
the system also includes aviation (Swedish EPA, 2017).
3
The EU ETS is a cap-and-trade system aiming to reduce GHG emissions by 21 percent
below 2005 levels before 2020. The system is an essential instrument of European
climate change policy and works in the following way. First, each Member State (MS)
sets an annual cap limiting total CO2 emissions. The government then divides the cap
into individual allowances to emit one tonne of CO2 emissions and distributes them
between firms covered by the system. Allowances can be auctioned or allocated for free
(grandfathering). When the distribution of permits is complete, firms can start trading
emission allowances with each other. The participating firms are allowed to purchase
extra allowances if they emitted more, selling or saving allowances if they emitted less
(Convery et al., 2008).
The development of the EU ETS can be divided into different phases. The first trading
period (2005–2007) can be seen as a trial phase to gain experience as to achieve CO2
reductions. Industrial plants were given a free allocation of allowances under the
national allocation plants (NAPs) during the first and second (2008–2012) trading
periods, based on historical emissions and production forecasts (Convery et al., 2008).
Many MS, including Sweden, had a surplus of allowances during the first trading
period. As a consequence, the European Commission decided to reduce the number of
allowances by 6.5 percent in the second trading period. In the third phase (2013–2020)
an EU-wide cap on emissions was introduced, which will be reduced by 1.74 percent
each year. In this period, there will also be a progressive shift from free allocation of
allowances towards auctioning (European Commission, 2016).
In 2005 the average price of allowances was EUR 22 per tonne of CO2. In April 2006,
emissions data for 2005 was released, which showed that many MS had a surplus of
allowances. Because of this, the price fell sharply (almost zero by mid-2006) and
remained very low during the rest of the first period (Convery et al., 2008). During the
second trading period, the price of allowances was relatively more stable compared to
the first phase. However, the price fell from about EUR 30 per tonne in 2008 to less
than EUR 10 per tonne in 2012 (Marin et al., 2017). In April 2018, the price of
allowances was around EUR 13 per tonne of CO2 (Market Insider, 2018).
4
2.2 The EU ETS in Sweden
Since the oil crisis in the 1970s, the use of oil as energy supply has decreased in
Sweden, while the use of clean energy power (e.g., hydropower and nuclear power) has
increased. Between 1970 and 1998, the Swedish energy sector has successfully
succeeded in reducing CO2 emissions by approximately 40 percent (Yu, 2011).
The Swedish environmental policy is based on 16 environmental quality objectives (one
of the objectives is to reduce climate impact). Besides the environmental targets, the
Swedish climate strategy consists of instruments, follow-up and evaluation of the
development towards established targets. There are different instruments available for
policymakers aiming to combat climate change. To reduce CO2 emissions, Swedish
policymakers decided to introduce a CO2 tax in 1991 as a supplement to the energy tax.
CO2 taxes are levied on fuels that are used as motor fuels or for heating purposes. To
support the expansion of electricity production and effectively control emissions, the
Tradable Green Certificate (TGC) was introduced in the energy sector in 2003 and is
planned to be in effect until 2030 (Swedish EPA, 2012).
Sweden has been included in the EU ETS since the program launched in 2005. In 2012,
the trading scheme included 853 Swedish installations, corresponding to 264 firms1
(Coria & Jaraitė, 2018). Before the introduction of the EU ETS, Swedish firms needed
to pay the CO2 tax for the right to emit CO2 emissions. However, after the program
launched in 2005, the sectors covered by the EU ETS both needed to buy EU
allowances (EUA – unit price for one tonne of CO2) and pay the CO2 tax for the right to
emit CO2 emissions (Löfgren et al., 2013). Since 2008, Swedish ETS firms are entirely
or partially relieved of the energy tax and the CO2 tax for certain fuels (Swedish Tax
Agency, 2018).
3 Literature Review
Many empirical studies analyze the effects of the EU ETS on firm economic
performance and competitiveness. Earlier literature covering the subject differ in terms
1 Some of the firms covered by the EU ETS owned several installations.
5
of empirical strategy, sector and geographical extent and the selection of indicators they
evaluate.
Some academic studies cover a more extensive selection of European countries. Abrell
et al. (2011) assess the effect of the EU ETS at a firm level on profit margins, value
added, and employment during 2005-2008. Their results do not show any significant
impact on the participating firms’ profits, employment or added value during the first
trading period and the beginning of the second trading period. By using a panel of 5,873
firms in ten European countries during 2001–2009, Chan et al. (2013) analyze the
impact of the EU ETS on material costs, employment, and revenues in the power,
cement, iron and steel sectors. The authors find no significant effect of the EU ETS on
any of the three variables in cement and iron and steel sectors, but a significant positive
impact on material costs and revenues in the power sector. Additionally, their findings
do not substantiate any concerns over carbon leakage, job losses, and industry
competitiveness. Based on an extensive panel of European firms, Marin et al. (2017)
study the effect of the EU ETS on a broader set of indicators of economic performance
in its first and second trading periods. The authors conclude that the program did not
affect economic performance negatively and that firms reacted by passing-through
additional costs to the customers, resulting in a higher markup and an increase in
turnover.
Turning to an academic study that focuses solely on one specific European country,
Anger and Oberndorfer (2008) conduct a study on revenues and employment on
German firms during the first trading period of the EU ETS. The authors do not find
any support that the allowance allocation, within the first trading period, had a
significant effect on revenues and employment. By using firm-level data to analyze the
impact of participation in the EU ETS among Swedish power generating firms in 2005
and 2006, Yu (2011) do not find any significant effect on firms’ profitability in 2005.
However, the author concludes that the program had a significant negative impact in
2006. These findings can according to Yu be a result of the changes in the price of
allowances and potential investments in abatement. Additionally, the analysis suggests
that the EU ETS had a different effect on the profitability of under-cap and over cap-
firms in 2005, but not in 2006. Jaraitė and Di Maria (2016) use a dataset based on
Lithuanian firms between 2003 and 2010 to assess the impact of the EU ETS on
6
investments and profitability. Their findings suggest that the EU ETS led to some
additional investments in new capital equipment and that the firms do not seem to have
suffered from their participation during the early stages of the program. Still, according
to their results the firms might have become less profitable in 2007-2010.
4 Empirical Methodology
This paper aims to estimate the changes in firm competitiveness relative to what would
have occurred if the EU ETS had not been implemented. Propensity score matching
combined with the DID approach are applied to evaluate the causal effect of the EU
ETS on firm performance. The propensity score matching is a popular method used in
observational studies aiming to evaluate policy interventions. In an observational study,
the assignment of subjects to treatment and control groups is not random and as a
consequence, the estimation of the treatment effect might be biased due to the existence
of confounding factors. However, using propensity score matching is a way to adjust
the estimation of treatment effects controlling for the existence of confounding factors
given that the treated and control entities are as similar as possible (Becker & Ichino,
2002).
The propensity score, 𝑝(𝑋) , is defined by Rosenbaum and Rubin (1983) as the
conditional probability of being treated given pre-treatment characteristics:
𝑝 𝑋 ≡ 𝑃𝑟 𝐷 = 1 𝑋 = 𝐸 𝐷 𝑋 , (1)
where 𝐷 = 0,1 is an indicator variable that identifies a firm’s participation in the EU
ETS and 𝑋 is a multidimensional vector of pre-treatment characteristics. Based on the
propensity score, observations from the treated group are matched with observations
from the control group with similar characteristics (Becker & Ichino, 2002). Different
matching methods are available2, however, by following Jaraitė & Di Maria (2016) the
nearest neighbor (NN) and Kernel matching estimators are used in this paper.
By using the NN estimator, each treated unit is matched with the control unit that has
the closest propensity score. In this paper, the NN method is applied with replacement 2 See Caliendo and Kopeinig (2008) for a review about the properties of different matching estimators.
7
since each control unit can be the best match for more than one treated unit. After
matching each treated unit with a control unit the difference between the outcome of the
treated units and the outcome of the matched control units is estimated. By averaging
these differences the Average effect of Treatment on the Treated (ATT) is obtained.
With Kernel matching, each treated observation is matched with several control
observations with weights inversely proportional to the distance between treated and
control observations. The common support is applied to improve the quality of the
matches when using these methods, implying that the matching is restricted based on
the common range of the propensity scores (Becker & Ichino, 2002).
After each treated observation has been matched with observations from the control
group, the next step is to estimate the average treatment effect of the EU ETS on the
desired outcome variables. The ATT is estimated as:
𝐴𝑇𝑇 ≡ 𝐸 𝑌!"(1)− 𝑌!"(0) 𝐷! = 1 (2)
= 𝐸 𝐸 𝑌!"(1)− 𝑌!"(0) 𝐷! = 1,𝑝 𝑋!
= 𝐸 𝐸 𝑌!"(1) 𝐷! = 1,𝑝(𝑋!) − 𝐸 𝑌!"(0) 𝐷! = 0,𝑝 𝑋! 𝐷! = 1 ,
where 𝑖 = firm and 𝑡 = year; 𝑌!"(1) is the potential outcome conditional on participation;
𝑌!"(0) is the potential outcome conditional on non-participation. If firm 𝑖 participates in
the EU ETS it belongs to the treatment group, and 𝐷! is equal to 1. On the contrary, if
firm 𝑖 does not participate in the EU ETS the firm belongs to the control group, and 𝐷!
is equal to 0.
When estimating the propensity score matching combined with the DID approach, the
following two assumptions must hold. First, the assumption regarding the balancing of
pre-treatment variables must be satisfied. If the propensity score is defined as 𝑝(𝑋),
then:
𝐷 ⊥ 𝑋 𝑝(𝑋) (3)
This assumption implies that assignment to treatment is independent of the 𝑋
characteristics, given the same propensity score.
8
Second, the assumption concerning unconfoundedness given the propensity score must
be satisfied:
𝑌!,𝑌! ⊥ 𝐷 𝑝(𝑋) (4)
This assumption implies that the outcomes are independent of treatment, conditional on
the propensity score (Becker & Ichino, 2002).
For the estimate of the ATT to be unbiased it is essential that the treatment does not
affect the firms in the control group. One possible reason for this assumption to fail is
related to spillover effects through general equilibrium impacts. For example, could an
increase in the electricity price, induced by the EU ETS, influence other manufacturing
firms including non-ETS ones (Marin et al., 2017).
To apply the simple estimate of the ATT obtained by using the standard DID can lead to
biased estimators if the variables related to firm-level outcomes vary significantly
across the treatment and control groups. By using the observable differences across ETS
participants and non-participants, this potential bias can be reduced and the ATT can be
estimated by enforcing semi-parametric matching estimators. In the empirical
estimation, the following DID matching estimator (𝛼) is applied:
𝛼 = !!!
𝑌!! 1 − 𝑌!! 0 − 𝑤!" 𝑌!! 0 − 𝑌!! 0!∈!!!∈!! , (5)
where 𝑗 = participants and 𝑘 = non-participants; 𝑁! is the number of firms in the treated
group; 𝐼! is the set of firms that participate in the EU ETS; 𝐼! is the set of non-
participants; and 𝑤!" is the weight placed on firm 𝑘 when constructing the
counterfactual estimate for the treated facility 𝑗. In this paper, the NN and Kernel
matching estimators are used to define the weights 𝑤!". It is common that when the
observable characteristics of an untreated unit are close to the characteristics of a treated
unit, the untreated unit 𝑘 is weighted relatively more heavily in the construction of a
counterfactual estimate for unit 𝑗 (Jaraitė & Di Maria, 2016).
9
5 Data
In this study, data from 2003 to 2008 is used to analyze the effect of the first phase and
the beginning of the second phase of the EU ETS on firm competitiveness. The panel
dataset is collected and owned by Statistics Sweden (SCB)3.
The SCB has identified firms participating in the EU ETS and matched firms across
datasets by using firm-level identifiers. The dataset consists of firms, which are sorted
into sectors based on their NACE code4. Six NACE code industry dummies are
therefore created to identify the participating firms. The dataset consists of firms whose
primary activity is in mining and quarrying (NACE 10-14) and manufacturing (NACE
15-36). As discussed by Marin et al. (2017) firms within the manufacturing sector are
more exposed to international competition compared to firms in the power sector and
other non-tradable sectors. The level of competition might restrict firms’ ability to pass
through the additional cost imposed by the EU ETS to its customers, meaning that the
loss of international competitiveness of EU firms and job losses are of particular
concern when it comes to the manufacturing sector.
To evaluate the impact of the EU ETS on variables through which firm competitiveness
can be observed, the following measures will be examined: labor productivity (value
added divided by the total number of employees), relative profit before tax (profit
before tax divided by turnover), turnover, and value added. Besides the variables
mentioned, the dataset also includes the amount of fossil fuels (CO2 emissions) used by
the firms and the stock of tangible capital assets. The monetary variables are expressed
in SEK and have been deflated to 1990 prices using a producer price index (PPI)
specific to industries.
Table 1 presents the descriptive statistics for both participating and non-participating
firms. The dataset is summarized for the unbalanced sample running from 2003 to 2008.
It is evident that ETS firms, on average, emit more CO2 emissions, create higher added
value per employee, are more capital intensive, and have larger turnovers as well as
3 We would like to thank Tommy Lundgren for providing this data for us. 4 NACE is the Statistical Classification of Economic Activities in the European Community.
10
Tab
le 1
: Des
crip
tive
Stat
istic
s
Non
-ETS
firm
s
Std.
dev
.
6.67
5
395.
4
128.
8
7,71
3.8
77,4
77.8
20,1
91.4
0.10
4
0.33
1
0.48
5
0.25
9
0.39
7
0.40
7
Mea
n
0.78
8
142.
6
0.18
0
526.
4
8,91
7.0
2,60
9.8
0.01
1
0.12
5
0.37
9
0.07
3
0.19
6
0.20
9
Obs
.
14,6
33
351,
355
328,
244
351,
355
351,
355
351,
355
351,
355
351,
355
351,
355
351,
355
351,
355
351,
355
ETS
firm
s
Std.
dev
.
387.
4
583.
9
0.01
1
420,
095.
7
2,18
7,35
4
666,
721.
7
0.18
3
0.23
4
0.49
7
0.44
4
0.29
4
0.23
4
Mea
n
97.0
683.
2
0.00
1
174,
544.
7
1,19
9,70
3
321,
529.
9
0.03
5
0.05
8
0.44
5
0.27
0
0.09
6
0.05
8
Obs
.
636
690
662
690
690
690
690
690
690
690
690
690
All
obse
rvat
ions
Std.
dev
.
81.6
396.
5
128.
7
21,5
41.1
134,
625
38,4
16.6
0.10
4
0.33
1
0.48
5
0.26
0
0.39
6
0.40
6
Mea
n
4.79
8
143.
6
0.18
0
867.
5
11,3
04.8
3,23
4.9
0.01
1
0.12
5
0.37
9
0.07
3
0.19
5
0.20
9
Obs
.
15,2
69
352,
045
328,
906
352,
045
352,
045
352,
045
352,
045
352,
045
352,
045
352,
045
352,
045
352,
045
Mea
sure
men
t uni
t
Kilo
tonn
es
SEK
, tho
usan
ds
SEK
, tho
usan
ds
SEK
, tho
usan
ds
SEK
, tho
usan
ds
SEK
, tho
usan
ds
Dum
my
varia
ble,
1 if
N
AC
E 10
-14
Dum
my
varia
ble,
1 if
N
AC
E 15
-19
Dum
my
varia
ble,
1 if
N
AC
E 20
-22,
36
Dum
my
varia
ble,
1 if
N
AC
E 23
-26
Dum
my
varia
ble,
1 if
N
AC
E 27
-28
Dum
my
varia
ble,
1 if
N
AC
E 29
-35
Not
e: A
ll m
onet
ary
varia
bles
are
in re
al te
rms
Var
iabl
e
Foss
il fu
el C
O2 e
mis
sion
s
Labo
r pro
duct
ivity
Rel
ativ
e pr
ofit
befo
re ta
x
Tang
ible
cap
ital a
sset
s
Turn
over
Val
ue a
dded
Min
ing
and
quar
ryin
g
Food
, bev
erag
es, t
extil
es a
nd c
loth
ing
Woo
d, p
ulp
and
pape
r
Che
mic
als,
min
eral
pro
duct
s and
pl
astic
Met
al, m
etal
pro
duct
s
Mac
hine
ry a
nd e
quip
men
t, el
ectro
nics
11
value added. However, the firms included in the EU ETS are on average less profitable
compared to the firms not participating in the program.
6 Results
6.1 Propensity Score and Matching In this paper, the propensity score is the probability that a firm is regulated under the
EU ETS based on their observable characteristics. The propensity score is measured by
using unbalanced data from 2004 together with a probit model. The explanatory
variables include the amount of fossil fuels used by the firm, stock of tangible capital
assets, turnover, and six dummy identifiers for whether the firm belongs to one of the
six sectors created. By enforcing the common support option the final number of blocks
is eight. This ensures that the average propensity score of treated and controls are
similar in each of the blocks created. Additionally, the combination of X characteristics
discovered in the treated group can also be observed among firms in the control group.
The balancing property is satisfied (see Table 2).
Table 2: Distribution of Control and Treated Entities According to Their
Propensity Scores
Propensity score Non-ETS firms ETS firms Total
0.0011632 751 4 755
0.025 112 6 118
0.05 44 1 45
0.1 24 6 30
0.2 20 8 28
0.4 4 9 13
0.6 2 14 16
0.8 5 30 35
Total 962 78 1 040
Note: The final number of blocks is 8.
Table 3 presents the results of the propensity score estimations satisfying both the
common support condition and the balancing property. The probability of being
12
included in the EU ETS is positively and significantly related to the amount of fossil
fuels used by firms, tangible capital assets, and turnover. The six NACE code industry
dummies, on the contrary, do not correlate with the likelihood of being engaged in
emissions trading.
Table 3: Propensity score estimates
Variables
Fossil fuel CO2 emissions 0.739***
(0.079)
Tangible capital assets 0.146***
(0.055)
Turnover 0.130*
(0.076)
NACE 2
(food, beverages, textiles and clothing)
- 0.782
(0.708)
NACE 3
(wood, pulp and paper)
0.620
(0.676)
NACE 4
(chemicals, mineral products and plastic)
0.127
(0.667)
NACE 5
(metal, metal products)
- 0.269
(0.690)
NACE 6
(machinery and equipment, electronics)
- 0.343
(0.720)
Constant - 5.376***
(0.899)
Number of observations 2,067
LT 𝜒! (8) 426.42
Prob.> 𝜒! 0.0000
Pseudo 𝑅! 0.642
Notes: ***p < = 0.001; **p < = 0.05; *p < = 0.1. Coefficients with standard errors in parentheses. All
monetary variables are in natural logarithms.
13
6.2 Estimation of Average Treatment Effects based on Propensity Scores
This section presents the results of the Kernel and NN matching specifications
reflecting on the effects of the EU ETS. Because the EU ETS launched in 2005, the year
2004 is used as the pre-treatment year. Table 4 presents the results from both using the
outcome variables in 2004 compared with their counterparts in following years (2005,
2006, 2007 and 2008) as well as the year-on-year changes.
The first measure of economic performance to be examined is labor productivity.
According to the matching results in Table 4, the impact of the EU ETS on labor
productivity is positive and significant at low significance level in 2005, 2006, and
2008. By using a dataset based on an extensive panel of European firms, Marin et al.
(2017) find similar results for the second trading period. From their result, the authors
interpret the improvements in labor productivity as a result of capital deepening.
Next, the focus is on assessing the impact of the EU ETS on Swedish ETS firms relative
profit before tax. The matching results presented in Table 4 show that the ETS firms in
Sweden have become less profitable in the early stages of the program (2005, 2006, and
2007). However, the year-on-year changes show that the impact of the EU ETS on
relative profit before tax is positive and significant at a low significance level in 2006.
When analyzing Lithuanian firms, Jaraitė and Di Maria (2016) on the other hand find no
evidence that the ETS firms suffered from their participation in the beginning of the
first trading period. Given of the amount of over-allocation that occurred during the first
trading period, the authors find these results as no surprise. Still, according to their
results, the firms might have become less profitable in 2007-2010.
The matching results presented in Table 4 show that the ETS firms in Sweden have
increased their turnover for the years following the introduction of the program.
However, when analyzing the effect of the EU ETS on firm-level economic
performance, Marin et al. (2017) find that the European ETS firms have increased their
turnover only for the second trading period.
14
Tab
le 4
: Ave
rage
Eff
ects
of t
he E
U E
TS
Part
icip
atio
n
2008
Ker
nel
Out
com
e: C
hang
es c
ompa
red
to 2
004
339.
9**
(185
.9)
-0.0
03
(0.0
27)
493,
000*
* (2
83,0
00)
131,
000
(104
,000
)
Out
com
e: y
ear-
on-y
ear c
hang
es
386.
4 (4
43.5
)
0.00
0 (0
.000
)
68,8
21.6
(1
84,0
00)
12,2
48.9
(6
2,06
6.1)
Not
es: *
** p
< 0
.001
; **
p <
0.05
; * p
< 0
.10,
the
p-va
lues
are
est
imat
ed u
sing
one
-taile
d t-t
ests
. Coe
ffic
ient
s with
boo
tstra
pped
stan
dard
err
ors i
n pa
rent
hese
s. A
ll m
onet
ary
varia
bles
are
in re
al te
rms.
The
num
ber o
f ETS
firm
s in
the
treat
ed g
roup
is 7
8. T
he K
erne
l mat
chin
g is
bas
ed o
n 96
2 fir
ms i
n th
e co
ntro
l gro
up. T
he
num
ber o
f firm
s in
the
cont
rol g
roup
diff
ers b
etw
een
the
year
s whe
n us
ing
the
NN
mat
chin
g m
etho
d w
ith re
plac
emen
t5 .
NN
334.
3*
(226
.5)
-0.0
03
(0.6
92)
474,
000
(376
,000
)
90.6
32.5
(1
27,0
00)
292.
0 (7
70.1
)
-0.0
00
(0.0
01)
149,
000
(189
,000
)
5,23
7.6
(80,
379.
5)
2007
Ker
nel
-45.
1 (4
59.5
)
-0.0
03**
(0
.002
)
424,
000*
* (2
53,0
00)
119,
000*
(8
2,09
5.6)
-180
.5
(439
.4)
-0.0
00
(0.0
00)
26,2
35.7
(1
81,0
00)
14,8
81.9
(6
2,96
6.2)
NN
42.2
(5
95.5
)
-0.0
03*
(0.0
02)
325,
000
(843
,000
)
85,3
85.9
(1
08,0
00)
-182
.8
(897
.2)
-0.0
00
(0.0
01)
76,5
41.7
(2
06,0
00)
22,7
77.7
(6
5,57
0.4)
2006
Ker
nel
136.
2*
(90.
5)
-0.0
02
(0.0
02)
387,
000*
(2
69,0
00)
101,
000*
(6
3,89
4.0)
46.2
(8
9.5)
0.00
1*
(0.0
01)
-102
,000
(3
18,0
00)
-109
,000
(1
56,0
00)
NN
237.
0**
(125
.0)
-0.0
03*
(0.0
02)
236,
000
(339
,000
)
59,7
35.1
(9
6,79
8.6)
14.7
(1
35.4
)
0.00
0 (0
.001
)
-99,
200
(319
,000
)
-114
,000
(1
63,0
00)
2005
Ker
nel
89.8
(1
23.7
)
-0.0
03*
(0.0
02)
485,
000*
(3
56,0
00)
208,
000*
(1
50,0
00)
89.8
(1
20.1
)
-0.0
03*
(0.0
02)
485,
000*
(3
46,0
00)
208,
000*
(1
54,0
00)
NN
222.
0*
(151
.9)
-0.0
03**
(0
.002
)
331,
000
(412
,000
)
172,
000
(162
,000
)
222.
0*
(144
.3)
-0.0
03*
(0.0
02)
331,
000
(443
,000
)
172,
000
(148
,000
)
Year
Met
hod
Labo
r pro
duct
ivity
(0
00’ S
EK)
Rel
ativ
e pr
ofit
befo
re ta
x (0
00’ S
EK)
Turn
over
(0
00’ S
EK)
Val
ue a
dded
(0
00’ S
EK)
Labo
r pro
duct
ivity
(0
00’ S
EK)
Rel
ativ
e pr
ofit
befo
re ta
x (0
00’ S
EK)
Turn
over
(0
00’ S
EK)
Val
ue a
dded
(0
00’ S
EK)
5For labor productivity, turnover, and value added, the NN matching with replacement in 2005 and 2006 is based on 33 firms in the control group. In 2007 and 2008, the number of firms in the control group is 30. For relative profit before tax, the NN matching with replacement in 2005 and 2006 is based on 32 firms in the control group. In 2007, the number of firms in the control group is 27 while in 2008 the number of firms is 26.
15
The fourth measure of economic performance to be examined is value added. The
matching results presented in Table 4 show that the ETS firms in Sweden have
increased their value added in the early stages of the program (2005, 2006, and 2007).
Both Abrell et al. (2011) and Marin et al. (2017), however, fail to find any statistically
significant impact of the EU ETS on value added during the first and the second trading
period when using a sample of European firms.
6.3 Testing the Common-Trends Assumption One important assumption that needs to be satisfied when estimating the propensity
score matching combined with the DID approach is that the trends in the outcome
variables should be the same across the ETS and non-ETS firms over time. To test if
this assumption holds, a placebo DID matching test is applied using two years of pre-
treatment data (2003 and 2004). In this test, 2003 is treated as a pre-treatment year and
2004 as a treatment year. According to the results of the placebo DID matching test (see
Appendix A) there are no statistical differences in the outcome variables between the
treatment and control groups for 2003-2004. This result implies that the pre-ETS trends
in the outcome variables are the same for both the firms participating in the EU ETS
and the non-participating firms. Because there are no statistical differences in the
outcome variables before the introduction of the EU ETS, it will lend some support for
the assumption that they are the same after the EU ETS launched in 2005.
7 Discussion and Concluding Remarks
In this paper, we use a unique dataset of Swedish firms to analyze the effect of the EU
ETS on economic and financial indicators through which the impact on firm
competitiveness may be observed. Contrarily to the concerns raised by industrial
emitters and policymakers regarding the negative impacts of environmental policy on
economic performance, the results suggest a positive and significant effect on labor
productivity, turnover, and value added. Still, a negative and significant impact can be
observed for relative profit before tax. Given the data used in this paper, it is difficult to
identify the exact mechanisms behind these results.
16
Using the unbalanced dataset running from 2003-2008 to estimate the propensity score
and the average treatment effects can be considered as a limitation. However, the
balancing property is not satisfied when running the regressions with the balanced
dataset. This limitation could be related to the lack of observations.
Many of the conclusions made in Section 6.2 Estimation of average treatment effects
based on propensity scores are supported by low significance levels. This limitation
may occur since the findings are based on a small number of firms participating in the
EU ETS, which differ in terms of size, technologies, and level of allocation stringency.
Because of this, the effect of treatment differs across ETS firms, and the average effect
may appear insignificant.
As mentioned in Section 2.2 The EU ETS in Sweden, after the program launched in
2005, the sectors covered by the EU ETS both needed to buy EU allowances and pay
the CO2 tax for the right to emit CO2 emissions (Löfgren et al., 2013). It is possible that
this could affect the economic and financial indicators evaluated in this study. Still, both
the ETS and non-ETS firms were covered by the CO2 tax for many years, and therefore
it is likely that the effects found in this paper should be attributable to the EU ETS.
Another limitation related to the method used in this paper is the possible interaction
between the CO2 price and the electricity price. The concern is whether an increase in
the electricity price, induced by the EU ETS, will influence other manufacturing firms
including firms not participating in the emissions trading program. For the average
treatment effect to be unbiased, it is essential that the treatment does not affect the firms
in the control group. Still, it is possible that this requirement fails due to the potential
spillover effects through general equilibrium impacts (Marin et al., 2017). Since the oil
crisis in the 1970s, the use of oil as energy supply has decreased in Sweden, while the
use of clean energy power (e.g., hydropower and nuclear power) has increased (Yu,
2011). Given the amount of clean energy produced along with the surplus of allowances
Sweden had in the first trading period, Swedish firms had no reason to pass through the
CO2 price to the electricity price. Therefore it is assumable that the CO2 price does not
constitute the main component of the energy price.
17
To facilitate future development of cap-and-trade systems and to get a broader picture
of how the EU ETS affects firm competitiveness, the analysis can be extended in at
least two directions. As discussed in Section 2.1 The EU ETS, the first trading period
can be seen as a trial phase to gain experience. Because of this, it is essential to
remember that the study’s estimation results are viewed in consensus with the program
period analyzed. One suggestion for future research is therefore to include data from the
second and the third phase to get a broader picture of the program's impacts on firm
performance. Additionally, a more detailed assessment of the mechanisms analyzed in
the paper is needed. Further research needs to identify these mechanisms and quantify
their contribution to gain an understanding of the potential connection between the EU
ETS and firm competitiveness.
18
References
Abrell, J., Ndoye Faye A., & Zachmann, G. (2011). Assessing the Impact of the EU
ETS using Firm Level Data. Bruegel Working Paper.
Anger, N., & Oberndorfer, U. (2008). Firm Performance and Employment in the EU
Emissions Trading Scheme: An Empirical Assessment for Germany. Energy Policy,
36(1): 12-22.
Becker, S.O., & Ichino, A. (2002). Estimation of average treatment effects based on
propensity scores. The Stata Journal, 2(4): 358–377.
Caliendo, M., & Kopeinig, S. (2008). Some Practical Guidance for the Implementation
of Propensity Score Matching. Journal of Economic Surveys, 22(1): 31-72.
Chan, H.S., Li, S., & Zhang, F. (2013). Firm Competitiveness and the European Union
Emissions Trading Scheme. Energy Policy, 63: 1056-1064.
Convery, F., Ellerman, D., & De Perthuis, C. (2008). The European carbon market in
action: Lessons from the first trading period. Journal for European Environmental &
Planning Law, 5(2): 215–233.
Coria, J., & Jaraitė, J. (2018). Transaction Costs of Upstream Versus Downstream
Pricing of CO2 Emissions. Environmental and Resource Economics.
https://doi.org/10.1007/s10640-018-0235-y.
European Commission. (2016). The EU Emissions Trading System (EU ETS).
European Commission. https://ec.europa.eu/clima/sites/clima/files/factsheet_ets_en.pdf
[Retrieved 2018-05-21].
Hanley, N., Shogren, J., & White, B. (2007). Environmental economics in theory and
practice (2.nd ed.). Basinstoke: Palgrave Macmillan.
19
Jaraitė, J., & Di Maria, C. (2016). Did the EU ETS Make a Difference? An Empirical
Assessment Using Lithuanian Firm-Level Data. The Energy Journal, 37(1): 1–23.
Löfgren, Å., Wråke, M., Hagberg, T., & Roth, S. (2013). Why the EU ETS needs
reforming: an empirical analysis of the impact on company investments. Climate Policy,
14(5): 537-558.
Marin, G., Marino, M., & Pellegrin, C. (2017). The Impact of the European Emission
Trading Scheme on Multiple Measures of Economic Performance. Environmental and
Resource Economics, pp: 1–32.
Market Insider. (2018). CO2 European Emission Allowances in EUR – Historical
Prices. Market Insider. http://markets.businessinsider.com/commodities/historical-
prices/co2-emissionsrechte/euro/1.4.2018_30.4.2018 [Retrieved 2018-05-23].
Rosenbaum, P.R., & Rubin, D.B. (1983). The Central Role of the Propensity Score in
Observational Studies for Casual Effects. Biometrika, 70(1): 41–55.
Swedish EPA. (2012). Instruments to achieve environmental objectives – a mapping.
Swedish Environmental Protection Agency. Report 6415, October 2012.
Swedish EPA. (2017). Utsläppshandel. Swedish Environmental Protection Agency.
https://www.naturvardsverket.se/Miljoarbete-i-samhallet/Miljoarbete-i-
Sverige/Uppdelat-efter-omrade/Utslappshandel/ [Retrieved 2018-05-21].
Swedish Tax Agency. (2018). Utsläppsrätter. Swedish Tax Agency.
https://www.skatteverket.se/foretagochorganisationer/skatter/punktskatter/energiskatter/
verksamhetermedlagreskatt/utslappsratter.4.121b82f011a74172e5880006846.html
[Retrieved 2018-05-23].
Yu, H. (2011). The EU ETS and Firm Profits: An Ex-post Analysis for Swedish Energy
Firms. Department of Economics, Uppsala University, Working paper #2, February.
20
Appendix A: A Placebo DID Matching Test
The propensity score is measured by using unbalanced data from 2003 together with a
probit model. The explanatory variables include the amount of fossil fuels used by the
firm, stock of tangible capital assets, turnover, and six dummy identifiers for whether
the firm belongs to one of the six sectors created. Table A1 presents the distribution of
control and treated entities according to their propensity scores satisfying both the
common support condition and the balancing property.
Table A1: Distribution of Control and Treated Entities According to Their
Propensity Scores
Propensity score Non-ETS firms ETS firms Total
0.0033818 428 9 437
0.1 22 4 26
0.2 14 2 16
0.4 9 11 20
0.6 3 8 11
0.8 1 36 37
Total 477 70 547
Note: The final number of blocks is 6.
Table A2 presents the results of the propensity score estimations satisfying both the
common support condition and the balancing property. The probability of being
included in the EU ETS is positively and significantly related to the amount of fossil
fuels used by firms. Tangible capital assets, turnover, and the six NACE code industry
dummies, on the contrary, do not correlate with the likelihood of being engaged in
emissions trading.
21
Table A2: Estimation of the Propensity Score
Variables
Tangible capital assets 0.092
(0.062)
Fossil fuel CO2 emissions 0.914***
(0.106)
Turnover 0.145*
(0.089)
NACE 2
(food, beverages, textiles and clothing)
- 0.329
(2.061)
NACE 3
(wood, pulp and paper)
1.457
(2.048)
NACE 4
(chemicals, mineral products and plastic)
0.674
(2.041)
NACE 5
(metal, metal products)
0.512
(2.051)
NACE 6
(machinery and equipment, electronics)
- 0.120
(2.097)
Constant - 6.079***
(2.206)
Number of observations 1,317
LT 𝜒! (8) 374.72
Prob.> 𝜒! 0.0000
Pseudo 𝑅! 0.685
Notes: ***p < = 0.001; **p < = 0.05; *p < = 0.1. Coefficients with standard errors in parentheses. All
monetary variables are in natural logarithms.
Table A3 presents the results of the placebo DID matching test, which shows that there
are no statistical differences in the outcome variables (namely, labor productivity,
relative profit before tax, turnover, and value added) between the treatment and control
groups for 2003-2004.
22
Table A3. Placebo DID matching test
Year
Method
2004
NN Kernel
Labor
productivity
(000’ SEK)
11.78
(99.03)
17.89
(102.83)
Relative profit
before tax
(000’ SEK)
0.000
(0.001)
0.000
(0.000)
Turnover
(000’ SEK)
220,000
(375,000)
211,000
(412,000)
Value added
(000’ SEK)
- 133,000
(142,000)
- 113,000
(154,000)
Notes: *** p < 0.001; ** p < 0.05; * p < 0.10, the p-values are estimated using one-tailed t-tests.
Coefficients with bootstrapped standard errors in parentheses. All monetary variables are in real terms.
There are 70 ETS firms in the treated group. NN matching with replacement is based on 22 firms in the
control group, while the Kernel matching is based on 477 firms in the control group.