1
The relation between profitability and carbon
emissions in the EU ETS era: A Dutch market
empirical research
Student number: S2755149
Name: Jacco van Goor
Paper: Masters’ Thesis
Study Program: MSc Finance Focus Area: Energy Finance Thesis Supervisor: Dr. P.P.M. Smid Date of submission: 10-01-2019
2
Introduction
The aim of this research is to see how firms and installations are doing in terms of carbon emission reduction and how emission performances affect profitability. The key is to see how the introduction of the European Union Emission Trading Scheme (EU ETS) in 2005 and the price of an European Emission Allowance (EUA) did affect the emissions and emission growth. The hypotheses stated concentrate on the higher the price of an EUA the lower the emission growth and the lower the total emissions. Furthermore the relation between emission performance, the price of an EUA and financial performance on a firm level will be researched. Here the hypotheses will aim at the higher the growth of emissions, the higher the total emission and the higher the price of an EUA, the lower the profitability of a firm will be. The goal is to see how the price of an EUA impacted emission performance and profitability and if carbon emission levels and carbon emission growth impact profitability as well.
Key papers regarding this research will be by Hart and Ahuja (1996) and King and Lenox (2002) Both papers empirically test both on the level of emissions and growth of emissions as on the relation of environmental performances and profitability. Other notable inputs come from the handbook of the European Commission (2015) and the most recent report of the Organization of Economic Cooperation and Development (OECD, 2018). The European Commission has a big impact on the policies of the EU and is the leader in the development of the EU ETS. The OECD reports on the progress of emission reduction from various levels. The documentation on the European level and the level of the Netherlands from the OECD are very interesting and useful in the context of this research.
Th EU ETS is the first and largest carbon allowances trading regime in the world. According to the World Bank the market is worth US $31.76 bln in 2018 and will cover about 2 giga tons of carbon dioxide equivalent (CO2e). The product sold on this market is an emission permit or
3 successfully achieve the goals in the Kyoto Protocol and the Paris Agreements, the cap is lowered at a rate of 1.74% annually until 2020. Phase IV of the EU ETS will be introduced in 2021 and will last until 2030. This phase features less exceptions, more sectors and types of gas, including a faster annual decrease of the cap to a rate of 2.2% annually (European Commission, 2015).
The aim of this study is to see how the policies of the EU ETS performs and if the scheme impacts the emission performance and profitability in the Netherlands. Furthermore the focus will be on how emission performance do against profitability. Is there evidence of the popularized question whether it ‘pays to be green’ as advocated in Kleiner (1991) and Hart and Ahuja (1996). This will be tested within the framework of the EU ETS in several models on several profitability proxies and indicators.
The Kyoto protocol (1997) was the first time countries agreed upon legally-binding emission reductions targets. As reported by the United Nations (UN) this protocol was signed by 37 industrialized countries, including most of the EU countries. In 2003 the EU adopted the directives of the EU ETS (European Commission, 2015). This directive was based on National Allocation Plans (NAPs), where every EU nation had to hand in their own plans to mitigate carbon dioxide emission. The first phase of the EU ETS was launched in 2005 and would be effective until December 31, 2007. Phase I of the EU ETS was called by the EC to be ‘pilot years’, since no infrastructure was in place to monitor, report and verify emissions by the firms who have to confirm to the EU ETS and no price was established for an EUA. The most important features of phase I were: 1) to cover CO2 of power and heat generators and
4 and carbon capture and storage (CCS). The program funded over €2.1 billion of direct subsidies and has attracted over €2.8 billion of funding from private parties. Altered features were a continuous decrease of allowances with even less allocations awarded for free and more sectors and types of gas included in the EU ETS.
The final goal of the parties who signed the Paris Agreement, including all members associated in the European Commission, is to keep global temperature rises under 2⁰C, with the pursue to keep it below 1.5⁰C (European Commission, 2018). To accomplish this, the European Commission set up a climate policy and targets focusing on bringing down carbon emissions. They set absolute emission reduction goals for the years 2020, 2030 and 2050. In 2020 the policy is set as “the 20/20/20 targets” which include: 1) A 20% reduction in greenhouse gas (GHG) emissions compared to 1990, 2) 20% of EU energy coming from renewable sources and 3) 20% improvement in energy efficiency. For 2030 the targets are: 1) to reduce GHG emission by at least 40% compared to 1990, 2) to reach at least a share of 27% of renewable energy and, 3) to have at least a 27% improvement in energy efficiency. The European Commission projects that the whole EU needs about €38 billion of annual additional investments from 2011 to 2030 to reach the targets. In 2050 the European Commission want to be climate-neutral, so having a net GHG emission of zero. This doesn’t mean that GHG emissions are totally excluded from daily operations, however the emissions have to be completely compensated with natural GHG capturers or carbon capture and storage (CCS).
The latest outlook concerning global carbon emissions is not looking very promising. According to studies of the Global Carbon Project, the global carbon dioxide emissions for 2018 are expected to rise by 2.7%. This is the largest jump in global emissions in 7 years, after little growth in the last couple of years. This is not in line with the goals and targets set at the Paris Agreements. The main contributors of Carbon dioxide emissions are the geographic areas China, USA, EU and India, with China emitting the most in 2017 with 27% of the total global emissions. The projections for 2018 is that China, USA and India increase their emissions by respectively 4.7%, 2.5% and 6.3%. The EU is expected to decrease its carbon dioxide emissions in 2018 by 0.7%.
5 price in 2017 was around €5.76 per allowance, which is an increase of €0.50 compared to 2016. One allowance equals the right to emit one ton of CO2e (European Commission, 2015). The NEa (2018) reports that the total number of Dutch participants in the EU ETS is around 435. This number is subject to change due to opening and closing of installations, inclusion of more types of gas and sectors or reaching the threshold of emission for participation in the EU ETS.
This research focuses on the Dutch market of the EU ETS. I will look into the emission performance of Dutch-based installations that have to confirm to the EU ETS since its origin in 2005. The focus will be on total emissions and emission reduction of these installations and whether or not these reductions met the goals set by the European Commission (EC). Furthermore, I will look into the performance of the firms that own the installations and see if they do or do not meet the EC annual emission reduction goals. Part two of the analysis starts with the relation of the price of European Emission Allowances (EUA) on the emissions and emission reduction. After that I will look into the relation of emissions and financial performance of these firms. I will look how emissions and emission reduction affects profitability. Profitability will be measured as return on assets (ROA), EBIT on assets (EBITOA) and net income (NI). This analysis will only look into phase II of the EU ETS and beyond (2008-2017). The models used will be similar to Hart and Ahuja (1996) and King and Lenox (2002). It consists of linear regressions of longitudinal datasets.
The paper starts with a literature review of papers involved in the research of environmental performance and financial performance. It will also dive deeper into the factors impacting emissions and relation of environmental and financial performance. After the literature review the research design of this paper will be presented. This includes a detailed description of the models used and the stated hypotheses. Before the results will be addressed, a short section will cover the data used in this paper and the data sources providing the data. Then the results of the tests will be provided together with the results of the stated hypotheses earlier in the paper. Lastly the conclusion of the research will be presented along with some conclusions regarding robustness tests and a discussion section with future recommended studies.
Literature review
6 invest in cleaner technologies once the marginal costs of polluting exceed the marginal costs of innovation. This is also described in the paper of Noseleit and de Faria (2013). They describe a theory on how investment in R&D and capital expenditures in innovation lead to lower levels of emissions. The marginal costs of investing should therefore be compared to a commodity that is priced. This is achieved by introduction of the EU ETS. There are several means and stages in reducing emissions, all with different levels of marginal costs. The two primary means of reducing emissions are: 1) controlling emissions through pollution controlling equipment and 2) preventing pollutions through better housekeeping, substituting materials, recycling or process innovations (Frosch and Gallopoulos, 1989; Willig, 1994). Smart (1992) describes in its model that emission reduction through pollution prevention is a financially and environmentally better choice over controlling pollution. It is cleaner and cheaper in the long term to prevent pollution from happening rather than cleaning the pollutive gases constantly. Pollution controlling equipment does not prevent pollution but just cleans it. The equipment is a non-productive investment and hence hurts a firm’s performances over the long run. The early stages of emission reduction are focused on behavioral changes of personnel and material input changes in the production process. These are fairly easy to implement and have relatively low marginal costs (Rooney, 1993). Later stages of emission reduction become progressively more difficult, often requiring significant changes of processes and sometimes even completely new production technologies (Frosch and Gallopoulos, 1989). So, as firms closing in on ‘zero pollution’, it becomes more expensive and more technology and capital-intensive (Walley and Whitehead, 1999). This is based on the theory of diminishing marginal returns (Rooney, 1993; Walley and Whitehead, 1999). This ties in with the role of emission performance on profitability. Based on the theory of diminishing marginal returns scholars like Lothe et al. (1994) and Walley and Whitehead (1999) argue that improvements in environmental performances are in conflict with financial goals, since it creates costs and usage of resources. Other scholars reject this statement and rather theoretically support the case of more efficient usage of assets and waste prevention (Kleiner, 1991). Russo and Pogutz (2009) describe a model that show how other factors can impact profitability. This model mainly describes common factors in financial theory, such as growth and leverage. It also includes factors used in the model of Noseleit and de Faria (2013) and in the empirical models of Hart and Ahuja (1996) and King and Lenox (2002). Grubb et al. (2011) discuss the role of growth of revenues on emission performances and the disconnected relation.
7
Figure 1: Development of the EUA prices
Source: Sandbag
the price of an EUA has not been near €30 for an extended period of time. Even in 2018 with a spike in EUA prices, it still has yet to reach €30 at the writing of this paper. This gives firms no incentives to lower emissions or to invest in polluting preventing technology. They also report that the price of an EUA had no impact on the profitability of European companies that are relatively heavy impacted by the EU ETS. The OECD concludes that price should theoretically impact emission performance and profitability, but it does not do it with the current market design.
Other scholars have empirical evidence that other factors than price do impact emission performances. Both Hart and Ahuja (1996) as King and Lenox (2002) find empirical evidence of a significant impact of size, growth of revenues and R&D intensity. These finding do match with the expectations of the theoretical model by Noseleit and de Faria (2013). At the relation between environmental performance and financial performance both scholars find empirical evidence of a significant impact of the variables growth of revenues, size, capital intensity and R&D intensity. There is also both theoretical as empirical evidence that leverage has a significant impact on the return of equity (ROE), but not on other financial proxies as return of assets (ROA) or return on sales (ROS) (Russo and Pogutz, 2009; Hart and Ahuja, 1996). These results correspond with the theoretic model described in Russo and Pogutz (2009).
8 such as GHG emissions (Hart and Ahuja, 1996; King and Lenox, 2000; King and Lenox 2002). Examples of GHG emissions are carbon dioxide, methane and nitrous oxide (United Nations Climate Change). Other scholars also found empirical evidence that indirect environmental proxies, such as environmental assessments or awards handed out by third parties (Klassen and McLaughlin, 1996; Russo and Fouts, 1997) or through the level of investment in environmental technologies (Nehrt, 1996) have impact on profitability. The results of Nehrt (1996) proves the theory in the paper of Noseleit and de Faria (2013).
Research
The research can be split up into three parts: 1) past emission performance on installation and firm level, 2) the factors influencing emission growth and emissions and, 3) the relation between emission performance and profitability. For the second research question I will look into the hypotheses about the higher the price of an EUA is, the lower the growth of emissions and the lower the total emission output is. For the third research question the hypotheses state that the higher the growth of emissions, the more total emissions and the higher the price of an EUA, the lower the profitability will. A more detailed statement about the hypotheses set and the models used will be discussed shortly.
9 impact of the price of an EUA with regards to incentives, but also discusses the role the price of an EUA has on the profitability of firms.
The focus will be on the power and heat generators and the energy-intensive industries and not on commercial aviation, since commercial aviation requires a different type of emission allowance, called the European Emission Allowance Aviation (EUAA) (European Commission, 2015). These allowances are unique for the commercial aviation industry and different from the EUA. They are traded on a separate market at a different price. The emissions for commercial aviation are treated as a separate entity and are not aggregated when it comes to the datasets of the NEa. All installations that are present in the database of the NEa will be used. The firms with a CO2 emission higher than 100,000 ton are eligible for firm-level analysis.
Further details of the data will be described later in this paper. The first part of the research,
Table 1: Descriptive statistics
Variable Mean St. Dev Min Max Skewness N
Emissions 14.99e+5 26.03e+5 0 13.69e+6 2.341 489
Log Emissions 12.905 1.182 4.394 16.432 -0.594 485
ROA 3.556 5.150 -30.549 27.550 -0.787 417
EBITOA 6.461 6.279 -24.365 38.764 -0.082 417
NI 11.28e+6 43.00e+6 -72.37e+6 4.19e+8 5.802 474
Log NI 14.117 2.264 7.692 19.853 -0.035 410
PEUA 9.652 1.865 4.39 23.33 1.602 10
Revenues 20.60e+7 66.50e+7 144669 59.40e+8 6.026 477
Log revenues 17.094 2.051 11.882 22.054 0.120 477
Growth Revenues 3.800 9.342 -39.261 422.139 9.342 427
Capex 20.04e+6 66.04e+6 -542100 60.10e+7 5.916 466
Log Capex 14.503 2.227 7.789 20.214 0.055 465
R&D 21.40e+5 84.62e+5 0 53.17e+6 4.758 409
Log R&D 12.214 2.148 6.908 17.789 0.179 341
Assets 25.50e+7 70.10e+7 55355 64.50e+8 4.876 467
Size 17.423 2.093 10.922 22.587 -0.150 467
Leverage 43.951 20.289 0 144.760 0.733 463
Capital Intensity 8.897 7.306 -2.333 76.868 3.852 466
R&D Intensity 1.023 1.331 0 6.659 1.791 399
10 about past performances, will look at the time frame since the start of the EU ETS in 2005. The second stage of the research will look at the time-frame starting at phase II of the EU ETS in 2008. The European Commission (2015) marks phase I of the EU ETS as an experimental stage. No policy effects can be concluded during this experimental phase, since the EC noted that almost all allowances were given out for free. Penalties were minimum and there was no policy incentive to address emission performance. The third stage of the research will look into the last year before phase II of the EU ETS to the most recent data. This means the analysis is based on the time-frame 2007-2017. I excluded the phase I time frame for this analysis due to the same reason as before, it was an experimental stage. The year 2007 is used as a benchmark to environmental and financial performances before the EU ETS became a market based system. A description of the variables used in this dataset and how they are calculated is given in the Appendix table A1. Table 1 presents the descriptive statistics of all non-dummy variables used in the context of this research. Brooks (2014) argues that skewness of the data can lead to inaccurate conclusions. Therefore, several natural log transformation are used to tackle any skewness issues.
Hypothesis 1
I will research whether the installations and firms owning the installations decrease the emissions over time and which firms decreased emissions the most. I will also look if the decrease in emissions is conform the EU ETS policy benchmark set at that particular phase. The formula for the annual emission growth is given in equation 1:
𝐴𝐺𝑅𝑒𝑖,𝑡 =
𝐸𝑖,𝑡−𝐸𝑖,𝑡−1
𝐸𝑖,𝑡−1 ∗ 100% (1)
were Ei,t denotes the level of ton carbon dioxide equivalent (CO2e) emission at time t of
installation i and Ei,t-1 is the first lag of this variable. I will also aggregate installations of the same firm to look at firm-specific performance. The formula for annual emission growth of a firm is presented in equation 2:
𝐴𝐺𝑅𝑒𝑓,𝑡 =∑ 𝐸𝑖/𝑓,𝑡−∑ 𝐸𝑖/𝑓,𝑡−1
∑ 𝐸𝑖/𝑓,𝑡−1 ∗ 100% (2)
here the subscript f stands for firm. Variable ∑Ei/f,t stands for the sum of the emission of installations i given its owned by firm f. ∑Ei/f,t-1 is the first lag of this variable.
11 within a country and indicate ‘leaders’ and ‘laggards’. I want to see how the different installations and firms do with the total emissions and the growth of emissions and how these installations and firm do with respect to the policy benchmarks set by the European Commission. I will use descriptive statistics to cover the performance of the growth of emissions per year for both the installations as the firms. I will use this information to look into how the performances were against policy benchmarks. An extra equation is created, see equation 3:
𝐴𝐺𝑅𝑒𝑖,𝑡 < 𝑃𝐵 (3)
This will also be the case for firm-specific performance per year indicated by equation 4:
𝐴𝐺𝑅𝑒𝑓,𝑡 < 𝑃𝐵 (4)
Studying the reasons for emission reduction - whether it is through lower production (at the installation level or in general), new and cleaner production technologies, relocation of installations, bankruptcy or any other reason - is beyond the scope of this research. I will only focus on the empirical evidence of actual emission output and whether this confirms to policy benchmarks.
Lastly and very interesting is to look into the compound annual emission growth of the high-polluting firms in this research. It shows the perspective over a longer time horizon at a non-linear average growth rate. Therefore I will use a formula calculating the compound annual emission growth given in equation 5:
𝐶𝐴𝐺𝑅𝑒𝑓= (𝐹 𝑆 1 𝑌 ⁄ − 1) ∗ 100% (5)
with CAGRef representing the compound annualized growth of emissions of firm f, F representing the final years’ emission, S representing the first years’ emission in this dataset,
Y represents the amount of years between the final and first years emission. Hypothesis 2
12 2015). This can be seen during the first publication in 2006, when the price of the EUA crashed from almost €30 to less than €12 per ton. Firms can also store surplus allowances from past years and transfer them to next years. However the introduction of a new phase of the EU ETS makes the accounts of surplus allowances obsolete. This can be seen in figure 1 between 2007 and 2008. At the end of 2007 the price of an EUA was very close to €0. In January 2008 the price rose back up to over €20. This was due to the fact that at the end of 2007 most firms stored many allowances from past years so that they did not needed to buy any new allowances from the market anymore. In 2008 these old allowance surpluses from the past phase were obsolete and all firms had to start from scratch again. This resulted in an actual liquid, more balanced market being present in 2008 compared to a very illiquid and very unbalanced market at the end of 2007. The highest EUA price reached since the start of phase II was €29.20 in July 2008. After that, the price quickly dropped and remained relatively stable at the €10-€15 range. During phase III up until 2017 the price never exceeded €10 anymore. According to data from the EEX, 2018 appears to be a breakout year with prices consistently above €15 and even reaching almost €25 in September.
This research will look into the relation between EUA price and the growth of emissions and also into the relation between the EUA price and the total emissions. The hypotheses will focus on the following:
1: The higher the EUA price, the lower the growth of emissions 2: The higher the EUA price, the lower the total emissions
Noseleit and de Faria (2013) theoretically describe factors that can influence emissions and emissions growth. Both Hart and Ahuja (1996) and King and Lenox (2000) empirically use specific variables in their studies to explain emissions and emission growth. I will use the control variables used by these scholars. The variables include size, growth of revenues, capital intensity, R&D intensity. I also added a dummy variable for phase III of the EU ETS for the model for growth of emissions, such that any new policy effects are separated from the explanatory variables. See table A1 for the calculations of these variables. For the model of total emissions I use the same control variables size, growth of revenues, capital intensity, R&D intensity and a dummy variable for phase III of the EU ETS again.
Equations 6 and 7 give the linear regression formula’s for both models:
𝐴𝐺𝑅𝑒𝑓,𝑡 = 𝛼𝑓,𝑡 + 𝛽1𝑃𝐸𝑈𝐴,𝑡+ 𝛽2𝑋𝑓,𝑡+ 𝜀𝑓,𝑡 (6)
13 with AGRef,t coming from equation 2, PEUA,t being the average price of an EUA, Xf,t and εf,t being the control variables and the error term of the model. Log(Ef,t) is the logarithmic function of the emissions of one firm. PEUA,t again represents the average price of an EUA. Yf,t and φf,t represent the control variables and error term of this model. The key is to see whether the price of an EUA has any influence on the growth of emissions or the actual verified emission output since phase II of the EU ETS.
Hypothesis 3
Lastly this research will address the relation of a firm’s financial performance and its carbon dioxide emissions. As a measure of financial performance I will use profitability. The return on assets (ROA), EBIT on assets (EBITOA) and net income (NI) will be used as a proxy of profitability. I opted to use ROA as a proxy of profitability and no other variables such as ROS or ROE, since ROA gives the estimate of financial performance through strategy management (Hart and Ahuja, 1996; Russo and Fouts, 1997; King and Lenox, 2002). The installations and factories that are recognized under assets, are the actual emitters of polluting gases. The firms that have to confirm to the EU ETS are firms with relatively high asset levels. It is the best measure of financial performance through pollution mitigation strategies. ROS and ROE are not considered as good proxies in the way this research is conducted due to the fact that ROS is more a measure of profit margins and ROE is a measure of efficient use of equity. Both factors are limited influenced by carbon dioxide emissions and strategies regarding pollution prevention (Hart and Ahuja, 1996; Russo and Fouts, 1997). ROA is calculated as the net income over average total assets over a year. In order to exclude any interest and taxation differences cross-sectionally I also decided to look into EBIT on assets as a financial proxy. As a third proxy of profitability I decided to use the actual achieved net income as an indicator. This way the sole role of emission can be compared to financial performance without compromising to asset bases or other size proxies in the dependent variable. We can see if the disconnection between emission and the financial outcome of a firm is indeed present, as described by Grubb et al. (2011)
14 achieved, especially in environmental work (Oliver et al., 2016). The market capitalization is based on the stock price and is therefore also less useful in the design of this study.
Based on the theoretical models of Russo and Pogutz (2009) and Walley and Whitehead (1999) and the empirical results of Hart and Ahuja (1996) and King and Lenox (2002) on emissions and emission growth impacting profitability and the report of the OECD (2018) on the impact of the price of an EUA, I came up with the following hypotheses:
3: A higher growth of emissions leads to a lower profitability 4: More total emissions lead to a lower profitability
5: A higher EUA price leads to a lower profitability
The corresponding models are presented in equations 8-10:
𝑅𝑂𝐴𝑓,𝑡 = 𝛼𝑓+ 𝛽3𝜑𝐴𝐺𝑅𝑒𝑓,𝑡+ 𝛽4(1 − 𝜑)𝐿𝑜𝑔(𝐸𝑓,𝑡) + 𝛽5𝑃 𝐸𝑈𝐴,𝑡+ 𝛽6𝑋𝑓,𝑡+ 𝜔𝑓,𝑡 (8) 𝐸𝐵𝐼𝑇𝑂𝐴𝑓,𝑡 = ∅𝑓+ 𝜃3𝜑𝐴𝐺𝑅𝑒𝑓,𝑡+ 𝜃4(1 − 𝜑)𝐿𝑜𝑔(𝐸𝑓,𝑡) + 𝜃5𝑃 𝐸𝑈𝐴,𝑡+ 𝜃6𝑌𝑓,𝑡 + 𝑢𝑓,𝑡 (9) 𝐿𝑜𝑔(𝑁𝐼𝑓,𝑡) = 𝜇𝑓+ 𝛾3𝜑𝐴𝐺𝑅𝑒𝑓,𝑡 + 𝛾4(1 − 𝜑)𝐿𝑜𝑔(𝐸𝑓,𝑡) + 𝛾5𝑃 𝐸𝑈𝐴,𝑡+ 𝛾6𝑍𝑓,𝑡 + 𝑣𝑓,𝑡 (10) with the variable AGRe,f,t being similar as equation 2, variables Log(Ef,t) and PEUA,t being similar to equations 6 and 7. Keep in mind that a model with both AGRef,t and Log(Ef,t) is disallowed because both variables represent the same data. Therefore a dummy variable φ is added to the model to only include one of the variable. The variables Xf,t represents the control variables and ωf,t representing the error term of the model 8. For the model 9 Yf,t and uf,t represent respectively the control variables and the error term of the model. Model 10 uses the symbols
Zf,t and vf,t as control variables and error term.
15 These variables are relevant in carbon dioxide emission as they represent relative investments in production assets (capex) and products and production processes (R&D). Investments can either be as expansion of the production facilities and product lines or at pollution-preventing measures. The sign and significance of these variables could be very interesting in conclusion to this research. The next variable is leverage, another common control variable in financial literature (Hart and Ahuja, 1996; King and Lenox, 2002; Russo and Pogutz, 2009). Lastly I added 2 dummy variables to account for the financial crisis and the different policy in phase III of the EU ETS. The effective time period of this research is limited to only 10 years (2008-2017) and there was a financial crisis present during quite an extensive period within this time frame. There is evidence that this affected the profitability and EUA price (OECD, 2018). It would be best to add a dummy variable to account for this event. The financial crisis dummy is active during the whole stint of the financial crisis (2008-2011) (Elliot, 2011). Phase III of the EU ETS included more sectors and types of gas, so it would be best to add a dummy variable to separate any policy effects on the explanatory variables. I did not include the control variable advertisement intensity, as in Hart and Ahuja (1996) due to a lack of theoretical background explanation and the exclusion of this variable by other scholars (King and Lenox, 2002; Russo and Pogutz, 2009).
Most of this research is conducted with an unbalanced longitudinal dataset. This has consequences for the test that can be conducted regarding any cross-sectional and time effects (Brooks, 2014). It is evident that both cross-sectional as time effects can be relevant with this dataset. Cross-sectional differences are recognized as differences amongst different firms. Time effects are recognized as differences over time. The proper test to conduct with a longitudinal dataset with cross-sectional and time effect concerns are the redundant fixed effects test and the Hausman test. The redundant fixed effects test is conducted to see whether any fixed effects are present in the data, both on a cross-sectional level as a time level. If the test results of a redundant test is significant there is evidence that the constant is not equal either cross-sectional or over time. The Hausman test is conducted to see whether any random effects are present in the data, both on a cross-sectional level as a time level. If the test results of a Hausman test is insignificant there is evidence that the error term is not equal either cross-sectional or over time.
Data
16 2005. A total number of 590 unique accounts were registered by the NEa over a time period of 13 years (2005-2017). Along with the carbon dioxide emission data I used the ownership documents of the permit accounts of the installations by the NEa. If a firm or installation changes ownership, the permit account gets transferred to the new owner. Every year the NEa publishes the data on changed ownership of all installations during that year. This helps me to determine the firm associated to the installations registered by the NEa. Financial and accounting data used in this paper comes from the Datastream source of Thomson Reuters and annual reports of certain firms. Not all the firms’ data was (fully) available on Datastream. Then the firms’ annual reports are used to gather these data. If firms report their numbers in a different currency than the Euro, than the exchange rate end of year will be used to translate these accounts into Euro denominated numbers. This research also makes use of the spot price of the EUA since 2008. This data comes from the European Energy Exchange (EEX), the NEa and Bloomberg New Energy Finance. Finally a dummy variable was added to factor in the effect of the financial crisis. Another dummy variable is added to address the introduction of new policies in phase III of the EU ETS. This way the effect of the introduction of new industries and addition of more types of gas in phase III will be separated from the independent variables in the model.
I decided to use a minimum carbon dioxide emission threshold level of 100.000 ton per year per firm to get included in the analysis. The main reason for this was that lower levels of emissions is quite often produced by smaller firms. It is harder to gather any public financial data from these types of firms. It is also done to keep the analysis clear. Including all firms would lead to an overflow of information which could not be handled by the researcher in the given time frame of the research. Adding the availability of financial data of these firms I ended up with 49 unique firms. A description of the firms included in this paper can be found in at the end of the paper in Appendix table A2.
In order to address skewness of certain variables, as depicted by the descriptive statistics in table 1, a log transformation has been applied. This is done with the variables emissions, net income, revenues, capital expenditures, R&D expenditures and assets.
Results
As mentioned earlier, the Netherlands is doing a very poor job when it comes to emission reduction (Oliver et al, 2016; OECD, 2018). As a country the Netherlands emitted 3% more CO2
17
Figure 2: Total emissions of installations in the Netherlands (in ton CO2)
Source: NEa
a group and the these larger European countries are likely to reach the goals set at the Kyoto protocol (1990) and the targets set by the European Commission in 2020. The Netherlands is most likely not able to reach a 20% CO2 emission reduction by the year 2020. If we zoom in to
the CO2 emissions in the Netherlands that fall under the EU ETS, as depicted in figure 2, we
see an increase in total emissions 2017 compared to 2005. This is partly due to more industries and types of gas included in later stages of the EU ETS, but this is a minor difference (NEa, 2018) Since the introduction of phase III of the EU ETS emissions in the Netherlands rose by 4.42 million tCO2, or more that 5%. Even though we see a decrease in emissions in 2016 and
2017, the prediction is still that the Netherlands is not able to reach the target of a 20% reduction (Oliver et al, 2016, European Commission, 2018; NEa, 2018).
Zooming even further in to an installation-level basis, we can see that the biggest emitting installation in 2017 was RWE Eemshaven Centrale emitting a total of 7,587,197 ton of CO2.
Other big emitters that year were Uniper Centrale Maasvlakte (7.3m), Tata Steel IJmuiden
(6.8m), Nuon Power Velsen (4.0m), Shell Nederland Raffinaderij (3.8m), Essent Amercentrale
(3.6m), ENGIE centrale Rotterdam (3.4m), Nuon Centrale Hemweg (3.4m), BP raffinaderij
Rotterdam (2.1m) and the ESSO raffinaderij Rotterdam (2.1m). All installations emitted more
than two million ton of CO2.
If we change our perspectives from total CO2 emission to emission reduction, we can see how
the individual installations are progressing towards a cleaner production. The results of the annual emission reductions of the installations starting from 2005 are shown in table 2. The number of observations is steadily improving, since more and more industries and types of
18
Table 2: Descriptive statistics of growth of emissions per year of installations based in the Netherlands 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Average -0.06 2.59 -0.20 -0.01 20.82 0.00 0.04 2.44 0.22 0.13 0.46 0.28 S.E. 0.02 2.59 0.04 0.04 20.14 0.09 0.13 1.98 0.15 0.08 0.23 0.21 Median -0.04 0.00 -0.05 -0.04 0.07 -0.06 -0.05 -0.02 -0.04 0.01 0.00 0.01 Minimum -1.00 -1.00 -1.00 -1.00 -1.00 -1.00 -1.00 -1.00 -1.00 -1.00 -1.00 -1.00 Maximum 1.95 536.33 2.04 12.92 7612.90 31.83 44.21 690.39 54.66 20.68 60.10 87.25 N 205 207 204 368 378 375 374 358 448 441 437 423 N < PB 126 226 104 260 235 188 268 187 193 172 62% 61% 28% 69% 63% 53% 60% 42% 44% 41% N <PB & >-1 87 217 95 256 216 119 257 180 176 167 43% 59% 25% 68% 58% 33% 57% 41% 40% 39%
The values given in this table represent the growth of emissions per year of the installations in the Netherlands from 2005-2017. Note that the values are base values (unless otherwise indicated), they are not multiplied by 100% to keep the table fitting on page. The S.E. is the standard error. N is the amount of installations that decreased emission that year. PB represents policy benchmark. The values under N < PB and N < PB & >-1 represent the percentage total of installations that reached the policy benchmark out of the total sample size.
gas are included within the EU ETS (European Commission, 2015). Therefore more and more installations have to confirm towards the scheme. In 2006 a total of 205 installations managed to change their carbon dioxide emission. This increased to 423 installations in 2017. We see that ever since 2010 the average emission reduction was positive, indicating that on average the installations emit more than the previous year. Especially for the year 2010 this was the case. Before 2010 we had 3 years of a negative emission growth, indicating that on average the installations decreased their emission. Since average emission growth can be skewed by an outlier, a better proxy for installations’ emission growth is to look at the median growth. This eliminates extreme averages caused by outliers, for instance in 2010. The median growth is negative in 7 out of 12 years, indicating that the median installation did decrease the emissions year-over-year. The minimum increase is -1, since you cannot decrease emission by more than 100%. Maximum increases are quite often above 3,000%. The year 2010 marks the year with the highest maximum increase in emission with an installation emitting more than 761,000% more CO2 emissions than the previous year.
Looking at how well the installations managed to improve their CO2 emission to the targets
19
Table 3: Descriptive statistics of growth of emissions per year of firms with installations in the Netherlands who emit >100.000 ton per year.
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Average -0.03 -0.01 0.12 -0.06 0.15 0.23 -0.11 1.64 0.28 -0.06 -0.03 0.03 S.E. 0.01 0.02 0.15 0.04 0.08 0.26 0.03 1.01 0.31 0.05 0.04 0.06 Median -0.03 -0.02 0.01 -0.07 0.05 -0.02 -0.04 0.04 -0.01 -0.02 0.00 -0.02 Minimum -0.37 -0.43 -0.99 -0.56 -0.73 -0.68 -0.96 -0.96 -0.87 -1.00 -1.00 -0.78 Maximum 0.13 0.27 4.59 1.26 2.87 11.50 0.24 38.54 14.57 0.90 0.72 2.48 N 36 36 36 40 44 45 45 44 47 47 46 45 N < PB 12 30 15 23 32 14 22 23 18 22 33% 75% 34% 51% 71% 32% 47% 49% 39% 49% N < PB & >-1 10 30 15 23 32 14 22 21 17 22 28% 75% 34% 51% 71% 32% 47% 45% 37% 49%
The values given in this table represent the growth of emissions per year of firms that have installations in Netherlands and emit >100.00 ton per year from 2005-2017 Note that the values are base values (unless otherwise indicated), they are not multiplied by 100% to keep the table fitting on page. S.E. is the standard error. N is the amount of firms that decreased emission that year. PB represents policy benchmark. The values under N < PB and N < PB & >-1 represent the percentage total of firms that reached the policy benchmark out of the total sample size.
CO2 emissions, or a growth of -1.3%. For phase III of the EU ETS the benchmark was set at a
reduction of 1.74% annually, or a growth of -1.74%. I added a segment that excluded the -100% inputs, since it was quite often the case that an installation just did not report its
emission for the next year. Ultimately this results in an decrease of emission by 100%. First looking at the results without the -100% restriction in table 1. We see that about 40% to 62% of installations reach the EC targets over the complete time frame. For 2010 it is even lower with only 28% of the installations reaching the benchmark. From 2015 to 2017 the percentage of installations that reached the policy benchmark was well below 50%. With the restriction the number of installations hitting the benchmark drops even further. However with the exceptions of 2008 and 2013 the difference is not that big. The reason for the big differences in 2008 and 2013 is due to the introduction of a new phase, phases II and III started in 2008 and 2013. This resulted in quite some installations not having to report anymore and hence leading to omitted datapoints. Including the restriction, since the introduction of phase III of the EU ETS never did more than 50% of the installations reached the policy benchmark of a CO2 emission reduction of 1.74%. Now focusing on the performance of the growth of
20 installations it is better to look at the median firm to counter any extreme outliers. For most years the median value shows a decrease in emission growth. The minimum again is capped at -1, as again you cannot decrease you emission by more than 100%. Major reasons for firms decreasing their emissions by 100% is bankruptcy or leaving the country by selling all installations owned. The maximum growth of emissions per year are way less extreme at a firm level than on the level of individual installations. The highest increase in CO2 emission was
in 2013 by BASF. They increased their carbon dioxide emissions by 3,854%. This is mainly due the introduction of phase III of the EU ETS, which included more sectors and types of gas. BASF had to report more emitting installations in phase III compared to phase II, which increased their EU ETS CO2 emission quite extensively. This proves the argument of a dummy variable
for phase III of the EU ETS in the models later in this paper.
In table 3 the results regarding performance against policy benchmarks is also shown. As with the installations, the policy benchmarks were a reduction of 1.3% annually in phase II and a reduction of 1.74% annually in phase III. And like the installations-level CO2 emissions, I
decided to add the same restriction to cross out any -100% data points. Looking at the results for the firms during phases II and III, the results are quite bad. Only twice since 2008 did more than 50% of the firms in the Netherlands reached the EC targets regarding CO2 emission. That
was 2009 and 2012 with 75% respectively 71% of firms reaching the EU policy benchmark. Adding the restriction on firm-level performances did not change the outcome very much. This is due to the fact that not a lot of firms completely reduced their emissions or left their Dutch installations.
When looking which firms did well and which firm did not, equation 5 is used to calculated the compound annualized carbon dioxide emission reduction of all firms that are represented in table 3. The results are presented in figure 3 and a more details can be found in Appendix A3. We cannot compare the results of the annualized CO2 reduction to the policy benchmarks,
because the firms have different years in which they operate Dutch installations that have to confirm to EU ETS and the policy benchmark is not constant over the years. 26 out of the 49 firms (53%) of the firms managed to decrease its CO2 emission. 22 out of the 49 (45%) firms
increased their CO2 emission. Crown van Gelder was the only firm to hit a compounded annual
CO2 emission growth of 0.00%. Great performances came from K+S Group 25.0%), Uniper
(-17.9%), Veolia (-13.6%), GasUnie (-13.6%), Norske Skog (-11.3%) and DELTA (-11.1%). An honorable mention will be handed to AGC Inc., who managed to decrease the CO2 emission
21
Figure 3: Compound annual emission growth of selected firms (2005-2017)
This figure depicts the compound annualized carbon dioxide emission growth (eq. 5) of the 49 firms in this research. For the full details of every firm’s performance, see table A3.
Continuing with the second part of the analysis the focus shifts to the relation of the price of an EUA to CO2 emission reduction and total CO2 emissions. Table 4 gives the correlation matrix
for the variables used in the models of equations 6 and 7. The highest absolute correlations can be found between size and log emissions at 0.35, R&D intensity and log emissions and PEUA
and growth of revenues with a correlation of 0.25. These values are relatively high but do not indicate any additional measurement steps to be undertaken. However, for safety reasons
Table 4: Correlation matrix for models in equation 6 and 7 AGRe Log emissions Growth revenues Capital intensity R&D intensity Size PEUA AGRe 1.00 Log emissions -0.04 1.00 Growth revenues -0.01 0.06 1.00 Capital intensity 0.18 -0.19 -0.08 1.00 R&D intensity 0.01 -0.25 -0.08 -0.13 1.00 Size -0.21 0.35 0.09 0.14 0.07 1.00 PEUA -0.07 -0.01 0.25 -0.03 0.03 -0.10 1.00
This table gives the correlation matrix of the variables used in models of equations 6 and 7. The calculation of the variables are given in table A1. Some variables are abbreviated for fitting purposes. Log em. Is log emissions, g. rev. is growth of revenues, cap. Int. is capital intensity and R&D int. is R&D intensity.
22 White diagonal standard errors and covariance are added to the models to improve the reliability of any conclusions drawn. Table 5 depicts the results of the linear regressions of equations 6 and 7. Model 1 is a representation of equation 6 and shows that the coefficient of the price of an EUA is negative but insignificant at the 10% statistical level. The coefficient for the control variable R&D intensity is positive and significant at the 10% level. Size is negative and insignificant at the 10% statistical level. All other control variables are positive and insignificant. The adjusted R2 of this model is very weak, only being 0.00. There is no
indication that any of the factors in equation 6 explain the annual growth of emissions outside capital intensity. The price not being significant is according to expectations. The price has no impact on the growth or reduction of carbon emissions. But that none of the control variables have any significant impact is not according to the papers of Hart and Ahuja (1996) and King and Lenox (2002). They find a significance negative relation in the size factor of 10% and 5% respectively. In my model it is just not rejected at the 10% level.
Model 2 in table 5 is a different approach of factoring in the price of an EUA in the emissions of a firm. This time the focus is not on emission growth but on total emissions, so equation 7
Table 5: Growth of emissions and total level of emissions
(1) (2)
AGRe Log emissions
C 88.22 (80.68) 6.97*** (1.03) Price EUA -2.29 (2.21) 0.02 (0.02) Size -7.23 (4.59) 0.37*** (0.06) Growth revenues 0.11 (0.09) -0.00 (0.00) Capital intensity 971.52* (554.61) -4.83 (4.12) R&D intensity 1340.70 (2338.29) -52.42*** (9.19) Phase III 32.41 (25.29) -0.02 (0.25) N 354 355 Firms 45 45 Adj. R2 0.00 0.19
This table depicts the outcome from a leased squared (LS) analysis of a longitudinal dataset of the 49 firms used in this research. The standard error is given in parenthesis. White diagonal standard errors & covariance (d.f. corrected) are used. N is the number of
23 is depicted here. As explained before, the natural log function of total emissions is taken to transform the data. The model shows that the coefficient of the price of an EUA is zero and insignificant. This is in line with results of the OECD (2018), the price appears to not incentivize firms to act on carbon emissions. The constant and the control coefficients for size and R&D intensity are all significant at the 1% level, constant and size being positive and R&D intensity being negative. This is in line with the theoretical model of Noseleit and de Faria (2013) and the empirical findings of Hart and Ahuja (1996) and King and Lenox (2002). A bigger sized firm has a higher emission output and the more a firm invest in R&D to their overall assets, the lower the emission gets. Growth of revenues and capital intensity have no significant impact on the total emission level. This is partly in line with expectation from previous literature. Both King and Lenox (2002) as Hart and Ahuja (1996) find a significant positive relation between growth of revenues and emissions, but it is consistent with the theory of Grubb et al. (2011). Both Hart and Ahuja (1996) as King and Lenox (2002) do not find significant evidence of capital intensity and total emissions. However both papers do discuss the negative sign of capital intensity as a sign of capital investments in less carbon emitting technologies. This is in line with the theoretical model of Frosch and Gallopoulos (1989). The R2 of model 2 is way better
than models 1, but is still at the lower end with 0.19. This is about equal to what other empirical studies have (Hart and Ahuja, 1996; King and Lenox, 2002).
Up next will be the results of the models depicted by equations 8-10. These models focus on environmental performance and profitability on different levels. Table 6 shows the correlation matrix of the variables used in these models. Most correlations are below the absolute value
Table 6: Correlation matrix of models 8, 9 and 10
ROA EBITOA Log
NI AGRe Log em. PEUA Size G. rev. Cap. Int. R&D int. ROA 1.00 EBITOA 0.91 1.00 Log NI 0.36 0.34 1.00 AGRe 0.03 0.04 0.01 1.00 Log em. -0.04 -0.02 0.24 -0.01 1.00 PEUA 0.17 0.27 -0.00 -0.07 0.03 1.00 Size -0.00 0.01 0.90 -0.02 0.29 -0.05 1.00 G. rev. 0.26 0.30 0.17 -0.02 0.09 0.27 0.12 1.00 Cap. Int. -0.14 -0.12 0.04 0.14 -0.23 0.00 0.10 -0.09 1.00 R&D int. 0.02 -0.07 0.10 0.01 -0.25 0.03 0.07 -0.10 -0.14 1.00
24 of 0.30. There are only 2 cases were the correlation was extremely high, being the correlation between ROA and EBITOA and the correlation between log NI and size with respectively 0.91 and 0.90. The high correlation between ROA and EBITOA is not really an issue, since they are not used within the concept of one model. Since both variables are treated as dependent variables, it is likely that both models will hand the same conclusions. The correlation between log NI and size however is concerning, since size will explain a high variation of log NI in the model. However, omitting the size variable from the model did not impact the coefficients nor the significance of any other variable, so I decided to keep the model as-is. Again, like in the models in equations 6 and 7, White diagonal standard errors and covariance are added to the models in equations 8-10 to improve the reliability of any conclusions.
Table 7 depicts the outcome of the last stage of this analysis, focusing on the relation between emissions and the growth of emission, together with the price of an EUA, on the financial performances. These models represent the equations 8-10. In models 1 and 2 the financial performance proxy is return on assets (ROA) or equation 8, models 3 and 4 depict the EBIT on assets (EBITOA) or equation 9 and models 5 and 6 the financial performance indicator is net income (NI) or equation 10. Again a log transformation has been applied for NI. The number of observations and firms is quite steady with around 344 to 355 respectively 44 to 45 for models 1-4. Models 5 and 6 have less observations and firms due to the limitation of taking log functions. Still the number of observations is around 300 and the number of firms equals 43. Also the adjusted R2 is quite consistent with over models 1-4 and consistent with the paper
25
Table 7: Profitability of selected firms
(1) (2) (3) (4) (5) (6)
ROA ROA EBITOA EBITOA Log NI Log NI
C -3.42 (5.24) -1.22 (5.06) -3.36 (5.52) -0.97 (5.43) -2.26 *** (0.51) -2.20*** (0.54) AGRe 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Log emissions -0.33** (0.16) -0.35* (0.19) -0.01 (0.03) Price EUA -0.03 (0.11) 0.04 (0.10) 0.03 (0.14) 0.02 (0.14) -0.00 (0.01) -0.01 (0.01) Size 0.27 (0.24) 0.41 (0.26) 0.29 (0.27) 0.44 (0.29) 0.94*** (0.03) 0.94*** (0.02) Growth revenues 0.04*** (0.01) 0.04*** (0.01) 0.05*** (0.01) 0.05*** (0.01) 0.01*** (0.00) 0.01*** (0.00) Capital intensity 38.75*** (10.82) 36.93*** (10.82) 62.57*** (13.13) 60.68*** (13.15) 6.56*** (1.77) 6.67*** (1.73) R&D intensity 98.13*** (21.47) 78.94*** (22.58) 88.72*** (25.02) 67.52*** (25.81) 14.11*** (4.32) 13.07*** (4.49) Leverage -0.05*** (0.02) -0.05*** (0.01) -0.02 (0.02) -0.02 (0.02) -0.01*** (0.00) -0.01*** (0.00) Financial Crisis 1.86 (1.43) 1.98 (1.41) 1.20 (1.51) 1.36 (1.49) 0.05 (0.17) 0.08 (0.17) Phase III 1.55 (1.20) 1.56 (1.19) 0.70 (1.20) 0.71 (1.18) 0.04 (0.13) 0.04 (0.12) N 354 344 354 355 299 300 Firms 45 44 45 45 43 43 Adj. R2 0.18 0.20 0.17 0.18 0.84 0.84
This table depicts the outcome of a least squared (LS) analysis of a longitudinal dataset of the 49 firms in this research. The dependent variables are profitability, measured as return on assets (ROA), EBIT on assets (EBITOA) and the log of net income (Log NI). The standard error is given in parenthesis. White diagonal standard errors & covariance (d.f. corrected) are used. N is the number of observations. *, ** and *** represent statistical significant at respectively 10%, 5% and 1% level.
26 change our conclusions, since other scholars also had insignificant coefficients for leverage. Another notable difference is the magnitude of the coefficient of capital intensity, which increase quite some from models 1 and 2 to models 3 and 4. This leads to the conclusion that interest and tax expenditures do factor in when it comes to capital intensity, probably due to the magnitude of its value. But otherwise interest and tax expenditures have no impact on the results of the models.
Models 5 and 6 focus on net income as the dependent variable, so the results of equation 10 are presented here. As mentioned before, the growth of emissions, total emissions and the price of an EUA have no significant impact on net income. All other variables other than the dummy variables are statistical significant at the 1% level. So this time the constant and the size factor turn significant this time compared to models 1-4. The size factor was expected to greatly influence the results of this model, due to the high correlation depicted in table 6 and the model of Russo and Pogutz (2009). This resulted in a high adjusted R2 this time at around
0.84. The relation was expected to be very strong and statistical significant positive and this turned out to be the case. Other control variables are equal in sign and statistical significance as models 1-4. Although the magnitude of both capital intensity and R&D intensity are way smaller in models 5 and 6 mainly due to a smaller distribution of the net income variable. As discussed before, all models could be present with fixed or random effects at either the cross-sectional level or the time level. For example King and Lenox also dealt with fixed effects in their models. There could also be multiple effects present. The models are tested for both fixed effects as random effects on both dimensions of cross-section as time by applying the redundant fixed effect test and the Hausman test. The results of both tests are shown in table 8.
27
Table 8: Redundant fixed effect and Hausman test results
(1) (2) (3) (4) (5) (6)
CS fixed effects 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0.00***
CS random effects 0.00*** 0.00*** 0.00*** 0.00*** 0.01** 0.01**
Time random effects 0.40 0.42 0.59 0.62 0.55 0.60
This table present the results of the redundant test of the cross-sectional fixed effect and the Hausman tests of both the cross-sectional (CS) random effects as the time random effects. *, ** and *** represent statistical significant at respectively 10%, 5% and 1% level.
Table 9: Profitability of selected firm including time random effects
(1) (2) (3) (4) (5) (6)
ROA ROA EBITOA EBITOA Log NI Log NI
C -3.42 (2.59) -1.21 (2.77) -3.36 (3.15) -0.97 (3.37) -2.26*** (0.50) -2.20*** (0.55) AGRe 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Log emissions -0.33** (0.15) -0.36** (0.18) -0.01 (0.03) Price EUA -0.03 (0.08) -0.04 (0.08) 0.03 (0.10) 0.02 (0.10) -0.00 (0.02) -0.01 (0.02) Size 0.27** (0.14) 0.41*** (0.15) 0.29* (0.16) 0.44** (0.18) 0.94*** (0.03) 0.94*** (0.03) Growth revenues 0.04*** (0.01) 0.04*** (0.01) 0.05*** (0.01) 0.05*** (0.01) 0.01*** (0.00) 0.01*** (0.00) Capital intensity 38.74*** (10.17) 36.93*** (10.06) 62.57*** (12.35) 60.68*** (12.24) 6.56*** (1.85) 6.67*** (1.85) R&D intensity 98.13*** (25.26) 78.94*** (26.05) 88.72*** (30.69) 67.52** (31.70) 14.11*** (4.41) 13.07*** (4.60) Leverage -0.05*** (0.01) -0.05*** (0.01) -0.02 (0.02) -0.02 (0.02) -0.01*** (0.00) -0.01*** (0.00) Financial Crisis 1.86* (1.11) 1.98* (1.10) 1.20 (1.35) 1.35 (1.33) 0.05 (0.20) 0.08 (0.20) Phase III 1.55* (0.89) 1.56* (0.88) 0.70 (1.09) 0.71 (1.08) 0.04 (0.16) 0.04 (0.16) N 354 343 354 355 299 300 Firms 45 44 45 45 43 43 Adj. R2 0.18 0.19 0.17 0.18 0.84 0.84
28 can be found in table 9, the results of the models with cross-sectional fixed effects can be found in table 10.
Adding random effects to the models has barely any impact on any coefficients and also barely any impact on the standard errors of all coefficients. A few standard errors increased, such as R&D intensity and a few standard errors decreased, such as the constant and size. However this did not lead to any changes in significance for any variable of any model. It also did not change the R2 of any of the models. However it did improve the reliability of the error terms
since the result of the Hausman tests indicated the inclusion of a time random effect dummy. The Hausman test gave evidence of uncorrelated time random effect with the explanatory variables.
In the models where cross-sectional fixed effects were added we see quite a lot of changes in the coefficients, the signs, significance as in the R2. Most notably the R2 of all models increased
quite a lot, making these models more reliable than the models in the regressions depicted in table 7 and table 9. For models 1-4 the R2 increased from less than 0.20 to 0.55 or more. For
the models 5 and 6 the R2 improved from 0.84 to 0.89. The results for the coefficients are also
different in this case. For all models the constant is was negative and insignificant in models 1-4, now they are all positive and insignificant for models 1-4. This means that for models 5 and 6 the constant went from negative and significant at 1% to positive and significant at 5%. The factors growth of emissions, total emissions and price of an EUA all are insignificant for all six models. Also the sign for the coefficient for total emission changed from negative to positive and significant to insignificant. This is in line with previous literature if you focus on Lothe et al. (1994) and Walley and Whitehead (1999), but not with Hart and Ahuja (1996). Furthermore these results are not consistent with the fixed effects models of King and Lenox (2002). However, the R&D intensity factor turned negative and statistical significant at the 1% level, which is consistent with the fixed effects models of King and Lenox (2002). This is not in line with the theoretic model of Noseleit and de Faria (2013). The coefficient of capital intensity decreased in magnitude in all models and turned negative for model 2. Also the factor is not statistical significant anymore for models 1-4. The standard error for the size effect increase substantially for all models with fixed effects, remaining insignificant in models 1-4 and leading to insignificance for models 5 and 6. The growth of revenues, leverage and the dummy variables were barely affected by the fixed effects and their signs and significance did not change.
Overall the Netherlands do a very poor job when it comes to carbon dioxide emissions. The total CO2 emissions increased by 3% since 1990 (Oliver et al., 2015). The CO2 emissions of the
29
Table 10: Profitability of selected firm including cross-sectional fixed effects
(1) (2) (3) (4) (5) (6)
ROA ROA EBITOA EBITOA Log NI Log NI
C 1.73 (25.89) 0.36 (25.20) 29.81 (30.11) 27.51 (29.08) 10.70** (5.02) 10.18** (4.94) AGRe -0.00 (0.00) -0.00 (0.00) 0.00 (0.00) Log emissions 0.13 (0.27) 0.28 (0.34) 0.05 (0.05) Price EUA 0.06 (0.08) 0.05 (0.07) 0.13 (0.11) 0.11 (0.10) -0.00 (0.01) -0.01 (0.01) Size 0.63 (1.46) 0.62 (1.48) -0.71 (1.70) -0.76 (1.72) 0.26 (0.29) 0.25 (0.28) Growth revenues 0.03*** (0.01) 0.03*** (0.01) 0.04*** (0.01) 0.05*** (0.01) 0.00** (0.00) 0.00** (0.00) Capital intensity 6.02 (11.65) -6.12 (11.37) 17.46 (14.80) 18.25 (14.21) 4.91*** (1.82) 5.24*** (1.74) R&D intensity -218.31*** (109.08) -234.44** (107.06) -271.01*** (113.31) -299.33*** (112.29) -23.86 (19.14) -29.83 (18.70) Leverage -0.20*** (0.04) -0.20*** (0.04) -0.25*** (0.05) -0.24*** (0.05) -0.02*** (0.01) -0.03*** (0.01) Financial Crisis 1.20 (1.07) 1.21 (1.07) 0.24 (1.10) 0.28 (1.10) -0.03 (0.16) -0.01 (0.16) Phase III 0.15 (0.82) 0.09 (0.82) -0.87 (0.78) -0.97 (0.77)) -0.03 (0.13) -0.05 (0.13) N 354 355 354 355 299 300 Firms 45 45 45 45 43 43 Adj. R2 0.55 0.55 0.58 0.57 0.89 0.89
This table depicts the outcome of a least squared (LS) analysis of a longitudinal dataset of the 49 firms in this research with cross-sectional fixed effects included. The dependent variables are profitability, measured as Return on Assets (ROA), EBIT on assets (EBITOA) and the log of net income (Log NI). The standard error is given in parenthesis. White diagonal standard errors & covariance (d.f. corrected) are used. N is the number of observations. *, ** and *** represent statistical significant at respectively 10%, 5% and 1% level.
and looking at the installation-level CO2 emissions in the Netherlands we see an equal pattern.
The average and median growth in emissions was quite frequently positive. Comparing the growth of emissions to policy benchmark set over the various phases, only 40%-62% reaches the goal. Adding the restriction of excluding omitted datapoints it falls even further. The case for firms is even worse. For them the average and median growth of CO2 emissions looked a
little better, but they had a hard time reaching the policy benchmarks. For most years less than 50% of the firms accomplished the annual targets in the phases II and III of the EU ETS. 53% of the firms in this analysis managed to decrease their compounded annual CO2 emission
30 determining the growth of emissions and the level of emissions. It appears that no factors have a big impact on the growth of emissions. Size and R&D intensity have a positive respectively negative effect on the level of emission. There is no evidence that the price of an EUA affects the growth nor the level of emissions. This comes back in the results of the report of the OECD (2018), who concluded that the prices of CO2 emissions are below estimated
carbon prices and climate costs. The level of emission has a negative significant relation towards the ROA and EBITOA, but not on NI. The factors growth of revenues, capital intensity, R&D intensity and leverage appear to have an significant impact on the financial performance indicators ROA, EBITOA and NI. Size also has an significant impact on NI. This is in line with previous literature. Adding cross-sectional dummy variables in the models resulted in insignificance of total emissions in models 2 and 4. It also led to losing significance for the factors size in models 5 and 6 and capital intensity in models 1-4. Lastly it changed the sign of R&D intensity from positive to negative, but the significance remained for models 1-4.
Conclusion
Looking back at the results we can conclude that we have no evidence to support hypotheses 1 and 2. It appears that the price of an EUA has no impact on both the growth of emissions nor the total emissions output. There is no statistical evidence that the EU ETS impacts the emission strategy of firms in terms of emission reduction nor total emissions. For the hypotheses 3-5 we have mixed evidence. The results presented in table 7 give evidence of a higher level of emissions resulting in lower financial performance. So we do have evidence that supports hypothesis 4. However, we have no evidence that either the growth of emissions nor the price of an EUA have any impact on the profitability proxies and hence no statistical prove to accept hypotheses 3 and 5. The redundant test and Hausman test suggest the presence of fixed effects and time effects. Adding the time random effects to the models did not change the results very much. However, adding cross-sectional fixed effects resulted in the loss of the significance of the total emissions on financial performance. With the cross-sectional fixed effects model we reject all hypotheses 3-5.
To counter any possible problems in the correlation between size and net income, I decided to do a robustness test with models 5 and 6. In the robustness test the control variable size was omitted. This leaded to no new conclusions rather that a lower R2 model. None of the
variables lost or added significance nor any sign changed. Another robustness test that was executed on all models was the way size was calculated. Originally size was calculated as the natural log of assets, as presented in table A1. For the robustness test size was calculated as the natural log of revenues. No model had differences in signs, only the R2 dropped for all
31 significant at 5% but that is it. This draws no new conclusions other than that the natural log of assets is a good indicator of size.
One of the important dependent variable is the log function of net income. This can be viewed as a questionable proxy of profitability, since it neglects the sets of negative net income. In this research it results in a dropout of on average 2 firms and about 50 to 60 observations. I admit that the log function of net income is not the best proxy of profitability, but it can be used to draw conclusion on the basic of profit-hitting years. It is safe to assume that that emissions and the price of allowances have only a limited impact on the corporate results over the year, based on the results of the models involving ROA and EBITOA. It is likely not the case that the production of carbon dioxide and the price of allowance to emit the CO2 will cause a
firm to swing from a profit to a loss. This results to the conclusion that this proxy can still serve its function. Other interesting factors to include in this type of research are the account on surplus of allowances and the percentage of freely awarded allowances per firm per year. The theory on the surplus of allowances is discussed and used in the paper by Russo and Pogutz (2009) and seems to affect profitability. However the NEa, which deals with and monitors these accounts on the purpose of emission coverage at the end of the year, had no permission to share this data with me.
Quite some firms that have installations that have to confirm to the EU ETS have no publicly available financial information. This is mainly the case for small emitters, so the ones emitting less than 100.00 tCO2e. But also in the segment of larger emitters there are quite some firms
who do not publicly publish annual reports. These are predominantly privately-owned firms who are not legally obliged to publish publicly. They are not stock listed and its hard to determine the market value over the full span of the EU ETS. These firms cannot be included in my research and this undermined the availability of data. However, since the number of observations is still large, the conclusions drawn in this paper can be generalized for all firms reaching the 100.000 tCO2e threshold annually.
32 all industries under EU ETS, so including the power and heat generators, the energy-intensive industries and the commercial aviation into one analysis.
Bibliography
Brooks, C. (2014). Introductory econometrics for finance. Cambridge university press. CBC (2018). Global carbon dioxide emission rose almost 3% in 2018. [Accessed 6-12-2018 at https://www.cbc.ca/news/technology/carbon-pollution-increase-1.4934096]
Department of Economic and Social Affairs (2008). International standard industrial
classification of all economic activities (ISIC). Statistical papers series M, No.4/Rev.4. United Nations publications.
Elliot, L. (2011). Global financial crisis: five key stages 2002011. The Guardian. [Accessed 7-12-2018 at https://www.theguardian.com/business/2011/aug/07/global-financial-crisis-key-stages]
European Commission (2015). EU ETS Handbook. [Accessed 9-9-2018 on: https://ec.europa.eu/clima/sites/clima/files/docs/ets_handbook_en.pdf]
European Commission (2018). A clean planet for all, a European strategic long-term vision for a prosperous, modern, competitive and climate neutral economy. [Accessed 7-12-2018 on: https://ec.europa.eu/clima/sites/clima/files/docs/pages/com_2018_733_en.pdf/] Frosch, R. A., & Gallopoulos, N. E. (1989). Strategies for manufacturing. Scientific American, 261(3), 144-153.
Global Carbon Project (2018). Global carbon budget 2018. [Accessed 7-12-2018 at http://www.globalcarbonproject.org/carbonbudget/18/highlights.htm]
Grubb, M., Müller, B. & Butler, L. (2011). The relationship between carbon dioxide emissions and economic growth. Oxbridge studies [working paper]
Hart, S. L., & Ahuja, G. (1996). Does it pay to be green? An empirical examination of the relationship between emission reduction and firm performance. Business strategy and the Environment, 5(1), 30-37.
King, A. A., & Lenox, M. J. (2000). Industry self-regulation without sanctions: The chemical industry's responsible care program. Academy of management journal, 43(4), 698-716. King, A.A., & Lenox, M.J. (2002). Exploring the locus of profitable pollution
reduction. Management Science, 48(2), 289-299.
Klassen, R. D., & McLaughlin, C. P. (1996). The impact of environmental management on firm performance. Management science, 42(8), 1199-1214.
Kleiner, A. (1991). What does it mean to be green? Harvard Business Review, 69, 38-47. Lothe, S., Myrtveit, I., & Trapani, T. (1999). Compensation systems for improving
