1 s.m.mous@student.rug.nl
with the European Union Emissions Trading System
Steven Mous
1MSc thesis Economics
University of Groningen
June 2017
Abstract
Interaction effects between the EU Emissions Trading System and overlapping renewable energy policies have been described extensively in the theoretical and simulation-based literature. Yet, it is remarkable that an empirical test of the so-called waterbed effect is virtually absent in this debate. Using a time series model for the permit price, this research provides empirical evidence that increases in renewable energy generation decrease the EU ETS permit price. Furthermore, a panel data approach to sectoral emissions illustrates that, through the functioning of the EU ETS, emissions in other sectors increase in response to increases in renewable energy generation. Therefore, we suggest policy makers to be cautious in combining renewable energy policy instruments with the EU ETS.
Keywords: Emissions trading, waterbed effect, policy interaction, renewable energy JEL-code: Q48, Q58
2 1 Introduction
With the introduction of the European Union Emissions Trading System (EU ETS) in 2005, the EU installed a key tool for reducing greenhouse gas emissions in a cost-effective way (European Commission Climate Action, 2015). The system currently caps emissions of over eleven thousand installations in the twenty-eight EU countries plus Iceland, Norway, and Liechtenstein, amounting to around 45 percent of EU CO2 emissions. Next to the EU ETS, a range of policies exists in Europe that promote the use of renewable sources in energy generation. On the international level, for example, the European Commission (EC) dictates on average a share of twenty percent renewable sources in energy generation by the year 2020 (EC, 2009). In a national setting, governments may consider the early closure of coal-fired plants or the subsidization of renewables alongside the twenty percent target (Frondel et al, 2010; Kloosterhuis and Mulder, 2015). The increase in renewable energy generation is expected to decrease the permit price in the EU ETS, and emissions are expected to merely displace away to other sectors within the EU ETS. The interaction effects between the EU ETS and overlapping renewable energy policies have been described extensively in the theoretical and simulation-based literature. Yet, it is remarkable that an empirical test of the so-called waterbed effect is virtually absent in this debate. An increased empirical understanding of the interaction between policy instruments provides both academics and policy makers with more clarity on the mechanisms at work under the waterbed effect in the EU ETS.
In the string of research concerned with the overlap between the EU ETS and renewable energy policies, a number of simulation-based studies have demonstrated the waterbed effect arising as a result of policy interaction (Bohringer and Rosendahl, 2011; Morris et al, 2010; Sijm, 2005). These simulation-based studies conclude that combining renewable energy constraints with the EU ETS does not bring about additional environmental benefits compared to the market outcome of the EU ETS alone (Fisher and Preonas, 2010; Traber and Kemfert, 2009; Van Den Bergh et al, 2013). Moreover, the costs of reaching the same environmental effect are higher as a result of the extra interference in the market (Abrell and Weigt, 2008; Boeters and Koornneef, 2011).
3
permits in the EU ETS’? The hypothesis is that as increased use of renewables depresses the demand for emission permits, a negative relationship exists between renewable energy generation and the permit price in the EU ETS. A time series model of the permit price is developed to evaluate the effect of renewable energy generation on the permit price. Furthermore, market fundamentals such as EU production levels and energy input prices are also included. The second research question is: ‘To what extent does the permit price affect sectoral emissions within the EU ETS’? The hypothesis is based on the assumption of profit-maximizing behavior by EU ETS participants, and states that a negative relationship exists between the permit price and sectoral emissions. Over the period 2005-2016, a random effects panel data estimation is applied on all sectors of the EU ETS excluding the power generation sector. Relevant emissions function variables such as sectoral production and energy input prices are used as control variables.
The most important result of this research is the significant estimate of -0.25 for the ‘renewables elasticity’ of the permit price. In other words, a 1% increase in renewable energy generation decreases the EU ETS permit price by 0.25 percent. This estimate empirically confirms the price mechanism of the waterbed effect. Furthermore, the permit price model estimates a strong, more-than-unit elastic relationship between production in European industry and the permit price. The permit price estimation fails to confirm the energy input prices hypotheses. Perhaps due to the low data quality, the sectoral emissions model fails to confirm the hypothesis of a negative relationship between the permit price and sectoral emissions. However, using an alternative specification a renewables elasticity of sectoral emissions of around 0.55 is estimated. Hence, the analysis suggests that that a 1 percent increase in renewable energy generation leads to a 0.55 percent increase in emissions of the other EU ETS sectors. The strongest driver of sectoral emissions is sectoral production, and the elasticity estimate of 0.65 suggests that a full decoupling of emissions and production is still far away from reality.
4
European Commission. We see directions for future research as to how these actions may impact the validity of the waterbed effect, as currently no certainty exists.
This paper is structured in the following way. Section 2 discusses the relevant literature on emissions trading, the waterbed effect, permit price determinants, and sectoral emissions. Section 3 and 4 respectively exhibit the data and the methodological approaches used to address the research questions. The results and their interpretation are presented in section 5. The considerations of this research are discussed in section 6, with section 7 concluding.
2 Literature review
Emissions trading and the EU ETS specifically have received a considerable amount of academic interest over the last one and a half decade. In this section emissions trading and the waterbed effect in general are discussed first, then the waterbed effect in the EU ETS follows. The vast majority of the existing literature on the waterbed effect in the EU ETS makes use of theoretical or simulation-based approaches, where this research contributes by empirically addressing both the permit price effect and the sectoral emissions effect of the waterbed effect. Permit price determinants are evaluated using a theoretical framework of supply and demand, and sectoral emissions are analyzed to observe emissions shifting in the waterbed effect. For the interested reader, further details on the origin and background of the EU ETS are discussed in Convery and Redmond (2007), Ellerman and Buchner (2007), Ellerman et al (2016), and Laing et al (2013). The former two publications consider to a greater extent the set-up and early years of the system, whereas the latter two are more of a survey character, collecting the existing literature on the EU ETS and extensively explaining a number of debates.
2.1 Emissions trading
To better understand the background of this research, this subsection reviews the theory on emissions trading systems and the waterbed effect. Then, we specifically discuss the EU ETS waterbed effect, market characteristics, and relevant recent policy developments.
2.1.1 Emissions trading theory and the waterbed effect
5
trading is that heterogeneity exists in the abatement options and abatement costs across participants (Tietenberg, 2006). The tradable nature of emission permits, together with sufficiently low transaction costs and limited market power of participants, then ensures that emission reduction occurs there where it is most economical. Participants with a preferred path of emissions higher than their current stock of permits can either buy more permits from other market participants, or invest in measures to reduce their emissions. From the supply and demand of emission permits market prices arise, a key feature of the market-based nature of emissions trading, which can be interpreted as signals of the equalized marginal costs of abatement in the system (Baumol and Oates, 1971; Dales, 1968). Compared to command-and-control policies with mandatory reduction targets for participants, emissions trading systems can achieve identical environmental effects at lower costs.
The level of emissions is determined only by the total supply of emission permits, given that this supply is lower than the ‘business-as-usual’ level of emissions (Bohringer and Rosendahl, 2011; Sijm, 2005; Tietenberg, 2006). Under such a binding cap, scarcity of permits and positive permit prices exist as abatement must occur somewhere in the system. Profit-maximizing behavior of participants then implies that emissions do not fall short of the cap. On the other hand, total emissions cannot exceed the supply of permits as participants must comply with the system. Hence, the policy maker directly controls the level of emissions through the cap. Note that the policy maker may allow participants to bank permits to different periods within the emissions trading system. Such a design feature is described as welfare improving in the literature, as participants can efficiently spread uncertain marginal abatement cost shocks and expected future price movements over time (Abrell et al, 2011; Fell et al, 2012; Godby et al, 1997; Yates, 2002). This reduces price volatility and the expected welfare costs of the policy. The result that bankability is welfare improving is contingent on the presumption that the marginal damage of CO2 emissions is sufficiently close to invariant over time. The implication of bankability for the level of emissions is that lifetime emissions of the emissions trading system are equal to the aggregated lifetime caps of the system, whilst annual emissions may deviate from the annual supply.
6
strategy exists for participants, giving rise to the demand for permits (Morris et al, 2010; Tietenberg, 2006). Extra policy interference may for example include the mandatory closure of coal-fired plants or subsidization of energy generation from renewable sources (Frondel et al, 2010; Kloosterhuis and Mulder, 2015). For the non-economist, a more detailed description of such promotion schemes and a basic explanation of how coexistence with an ETS may lead to interaction can be found in Del Rio Gonzalez (2007). Such additional policies alter the abatement decision and decrease the demand for emission permits from the constrained participants below the cost-effective demand (Morris et al, 2010; Van Den Bergh et al, 2013; Zeng and Mulder, 2016). Lower demand is expected to translate itself into lower permit prices as long as prices are nonzero, making emission less costly for unaffected installations. Hence, the abatement decision of unaffected participants now includes more emission-intense processes compared to the market outcome of only an emission trading system. Numerous authors have described the new market outcome, in which emissions merely displace away from the affected installations to unaffected installations as a result of the lower price (Bohringer and Rosendahl, 2011; Fischer and Preonas, 2010; Frondel et al, 2010; Lehmann and Gawel, 2013; Morris et al, 2010; Pethig and Wittlich, 2009; Sijm, 2005; Traber and Kemfert, 2009; Van Den Bergh et al, 2013; Zeng and Mulder, 2016).
The presence of the waterbed effect brings about important policy considerations, as the overlap of instruments increases welfare costs without decreasing emissions further. Participants are hindered in reaching the cost-effective market outcome if both policy instruments are binding, hence the costs of reduction are higher (Boeters and Koornneef, 2011; Sijm, 2005; Tietenberg, 2006). Pethig and Wittlich (2009) formally present a general equilibrium model to prove that using only an emissions trading system to achieve emissions reduction has substantial welfare benefits over a combination of policy instruments. If the constraint from the additional policy is so strict that the permit price drops to zero and the emissions cap is no longer binding due to the strong decrease in demand for permits, additional policy may bring about environmental benefits albeit at high cost (Abrell and Weigt, 2008; Bohringer and Rosendahl, 2011; Zeng and Mulder, 2016).
2.1.2 The waterbed effect in the EU Emissions Trading System
7
percent of European energy must come from renewable sources by the year 2020. Although other interfering policies are also conducted at the national level, this research focusses on renewable targets as these are imposed directly by the same policy maker as the EU ETS. The imposed constraints are binding to a certain extent, as by 2015 the majority of participating member states is short of their target. Also, positive permit prices have prevailed in the EU ETS apart from the final months of 2007, thus both policy instruments are binding. The cost-effective market outcome is distorted, as a number of installations are obliged to satisfy a higher than optimal share of renewables in the generation of energy (Bohringer and Rosendahl, 2011; Morris et al, 2010; Sijm, 2005).
The existing literature consists of theoretical and simulation-based work, and demonstrates the presence of the waterbed effect in the EU ETS. Sijm (2005) uses a simple and intuitive numerical example of two hypothetical countries to show that emissions are not reduced under a policy overlap, whilst welfare costs are increased. Bohringer and Rosendahl (2011) simulate a number of policy combinations of renewable targets and a cap-and-trade system representing the EU ETS, concluding that emissions are not reduced but welfare costs are higher when both constraints are binding. Abrell and Weigt (2008), Boeters and Koornneef (2011), and Morris et al (2010) use different types of computable general equilibrium models to estimate that a 20 percent renewables target next to the EU ETS decreases the permit price and leaves overall emissions unchanged. Traber and Kemfert (2009) model the European electricity market to obtain the result that feed-in-tariffs for renewables in Germany decrease the permit price, but do not affect overall emissions. Fisher and Preonas (2010) use a more formal approach to conclude that it is ‘politically popular’ but incorrect to refer to environmental benefits as a rationale for renewable energy policies next to any ETS. Van Den Bergh et al (2013) use a partial equilibrium model to determine a strong price decrease and emissions displacement as a result of the overlapping policy instruments. Zeng and Mulder (2016) use a two-country model of the electricity market to illustrate the existence of the waterbed effect in the EU ETS. In short, increases in renewable energy generation increase emissions in other sectors through the permit price.
8
(2008-2012) data and find mixed results for the relationship of renewables with the permit price, mainly because of differences in their methodological approaches. The main question for the permit price component of this research is: ‘To what extent do market fundamentals, and specifically renewable energy generation, affect the price of emission permits in the EU ETS’? The next subsection discusses the theory and hypotheses for this research question. No empirical work exists to the author’s knowledge that evaluates emissions distribution within the EU ETS. The research question concerned with the emissions component of the waterbed effect is: ‘To what extent does the permit price affect sectoral emissions within the EU ETS’? Subsection 2.3 presents the hypotheses for this research question. More empirical understanding of both elements underlying the waterbed effect in the EU ETS provides academics and policy makers with more confidence on determining the implications of applying multiple instruments in European climate policy.
The method of allocation of emission permits in the EU ETS is not expected to affect the validity of the waterbed effect. Emission permits are currently distributed using a system of auctions and free allocation to participants, with a shift to more auctioning over time (EC, 2012b). Free allocation is often referred to as grandfathering of permits. Grandfathering of permits may improve the political acceptability of an ETS, as full auctioning might come about with opposition from polluting participants (Vollebergh et al, 1997). Carbon leakage, the shifting of emissions away from the EU ETS, may also rationalize the grandfathering of permits to some extent. However, in the existing literature a strong preference exist for more auctioning of permits, as this mitigates strategic incentives for participants (Anderson and Di Maria, 2011; Grubb et al, 2005; Hepburn et al, 2006; Hintermann, 2011; Smale et al, 2006). For the waterbed effect it is important to note that grandfathered permits still have an opportunity cost to the holder, namely the market price at which it can be sold (Reinaud, 2004). Therefore, the allocation method is not expected to affect the price and emissions mechanisms occurring under the waterbed effect.
9
permits away from the affected sector to unaffected sectors. Convery and Redmond (2007) compute the Herfindahl Index for phase 1 (2005-2007) of the EU ETS, and conclude that the power generation sector, the largest participating sector, does not appear to possess any real market power in the permit trading market. Heindl (2012) estimates transaction costs for German participants in 2009 and 2010 to be around 9 million euros, with roughly 70 percent for the measurement, reporting, and verification of emissions. This leaves a little less than 3 million euros of transaction costs for trading and search in the German market, a figure that in itself appears of small magnitude in the overall EU ETS. Furthermore, commitment problems mitigate trading market power if the initial endowment of permits is lower than intertemporal emissions (Hintermann, 2011; Liski and Montero, 2011). Combining these insights leads to the proposition that market power issues in the permit trading market are likely not of strong concern, even more so because the amount of grandfathered permits in the EU ETS is decreasing.
10
relatively low permit price may also feed concerns that the supply of emission permits is too generous.
The European Commission has taken temporary measures affecting the timing of permit allocation to address supply concerns in the EU ETS, but these do not affect the validity of the waterbed effect. First, a total of 900 million permits have been withdrawn from auction in the years 2014-2016 and were to be added back to auction in 2019-2020 (EC, 2012b). Hence, the total supply of permits is unaltered but the allocation time path is as a result of the back loading measure. Second, a Market Stability Reserve (MSR) is to be installed in the EU ETS at the end of 2018 (EC, 2015a). The MSR functions in the following way:
i) If permits in circulation < 400 million then: release 100 million permits unless the Market Stability Reserve is empty
ii) If 400 million < permits in circulation < 833 million then: no intervention is made by the Market Stability Reserve
iii) If permits in circulation > 833 million then: withdraw a number of permits equal to 12 percent of permits in circulation from the next auction and place them in the Market Stability Reserve
The MSR is initially filled with the 900 million permits that were back loaded to auction in 2019-2020. Given that the current reserve in the EU ETS is over 1.5 billion permits, it is likely that the MSR will grow in volume upon introduction. Note that under the waterbed effect renewable policies decrease the demand for permits, which may result in a larger influx into the MSR than in the market outcome without renewable policies. However, these permits are eventually released into the market as long as positive prices exist and therefore the lifetime supply of permits in the EU ETS is unaffected. Hence, we conclude that both the back loading of permits and the establishment of the MSR do not impact the validity of the waterbed effect theory. Potential future policy measures of a more permanent nature that may affect the strength of the waterbed effect are discussed in section 6.
11
ETS. To form hypotheses on our first research question, we now turn to the theory on permit price determinants in the EU ETS.
2.2 Explaining the permit price
Over the years, several authors have aimed at better understanding the underlying drivers of the permit price in the EU ETS, however the inclusion of renewables has been limited. A detailed overview of the literature on the permit price in phase 2 of the EU ETS is provided by Hintermann et al (2016). Relevant contributions on the determinants of the permit price in both phase 1 and 2 are discussed here, while no work to our knowledge exists using data from a number of years from phase 3 (2013-2020). Bredin and Muckley (2011) and Creti et al (2012) agree that the market has matured more in phase 2, as the fundamentals drive the permit price stronger from January 2008 onwards compared to phase 1. This research provides new insights as new phase 3 data is incorporated in the estimation. In this subsection the theoretical framework used to evaluate potential determinants of the permit price is described first. Then, a brief discussion follows evaluating the literature on the relationship between the permit price and production levels, international credits, energy input prices, and renewable energy generation in relation to the permit price.
2.2.1 Theoretical framework
To develop hypotheses of how market fundamentals may affect the permit price, we discuss how these variables may affect the demand and supply of emission permits in theory. The overall supply of emission permits is fixed, however trade occurs in the secondary market. Therefore, this subsection refers to supply as the supply in the secondary market. We assume that buyers of emission permits are indifferent between permits from the primary or secondary market, as their characteristics are the same. David and Garcés (2009) depict the market demand and supply functions as:
𝑄𝑖 = 𝑄𝑖𝐷(𝑃𝑖, 𝑤𝑖𝐷, 𝑢𝑖𝐷; 𝜃𝐷) (1)
𝑄𝑖 = 𝑄𝑖𝑆(𝑃𝑖, 𝑤𝑖𝑆, 𝑢𝑖𝑆; 𝜃𝑆) (2)
12
𝜃𝐷 and 𝜃𝑆 are parameters accompanying these factors. Combining equations (1) and (2) under the requirement that demand equals supply in equilibrium obtains1
𝑃𝑖 = 𝑃(𝑤𝑖𝐷, 𝑢𝑖𝐷, 𝑤𝑖𝑆, 𝑢𝑖𝑆; 𝜃𝐷, 𝜃𝑆) (3)
the reduced-form2 equation for price. Finally, assume that, ceteris paribus, increased demand (supply) boosts (depresses) price:
𝜕𝑃𝑖/𝜕𝑄𝑖𝐷 > 0; 𝜕𝑃𝑖/𝜕𝑄𝑖𝑆 < 0 (4)
From equation (4) it is clear that to formulate hypotheses one must understand how factors may affect the demand or supply of emission permits. In the secondary market for emission permits, the separation between suppliers and demanders of emission permits is not very distinct, making them potentially interrelated. However, the objective is not to estimate demand or supply separately, but the market outcome of their combination (David and Garcés, 2009). Therefore, determining unambiguous hypotheses and estimations is possible if demand and supply are affected in opposing directions, as their effect on price is opposing. More specifically, no problem exists if a variable is both in 𝑤𝑖𝐷 and 𝑤𝑖𝑆, as long as its sign in 𝜃𝐷and 𝜃𝑆 is opposing, as only the market outcome 𝜃 is estimated. We now turn to theory and empirical literature on potential determinants of the permit price, and discuss how these may affect the demand and/or supply of emission permits.
2.2.2 Production levels
Theory suggests that the level of production by participants of the EU ETS is an important determinant for the demand for emission permits, and it has been included in various ways in the permit price literature. The relationship can be expressed as: increases (decreases) in the production volume boost (depress) the demand for emissions and therefore emission permits (Creti et al, 2012; Koch et al, 2014; Rickels et al, 2015). In terms of our theoretical framework, it is also possible that the production level may be a determinant for the supply of emission permits in the secondary market, as more activity by permit holders decreases their willingness to sell permits. Remarkably, in the literature no work exists that uses data directly production
1 By solving the single equation 𝑄
𝑖𝐷(𝑃𝑖, 𝑤𝑖𝐷, 𝑢𝑖𝐷; 𝜃) = 𝑄𝑖𝑆(𝑃𝑖, 𝑤𝑖𝑆, 𝑢𝑖𝑆; 𝜃) for price
2 Reduced-form entails that the endogenous variable price is expressed in terms of only exogenous
13
levels. Most authors use the level of economic activity in the EU to represent the level of production of EU ETS installations in their estimations. Creti et al (2012), Hintermann (2010), Koch et al (2014), and Rickels et al (2015) use broad stock market indexes and a number of regression techniques to estimate a significant and positive relationship between economic activity and the European Union Allowance (EUA) price, specifically in phase 2. On the other hand, Aatola et al (2013) use end-product prices, such as the electricity price and the steel price, to reach the same conclusion on the relationship between production levels and the permit price. However, this requires the potentially problematic assumption that end-product prices and the volume of production are positively related. Finally, the oil price is used by Koch et al (2014) and Aatola et al (2013) in their sensitivity analyses to reach similar conclusions on the relationship between economic activity and EUA prices. Hence, although methodological approaches differ, strong consensus exists on the empirical relationship between production levels and the permit price. The combination of theory and empirical evidence leads to the clear-cut hypothesis that a positive relationship exists between the level of production and the permit price.
2.2.3 International credits
14
hypothesis that a negative relationship exists between the use of international credits and the permit price.
2.2.4 Energy input prices
Installations may make use of a number of high-emitting and low-emitting inputs in their production process. Aatola et al (2013) show formally that under the assumption that participants are profit-maximizing firms, a price increase of the high-emitting (low-emitting) input is expected to decrease (increase) the price of emission permits as producers switch to the low-emitting (emitting) input. This is an intuitive effect, as a higher price of the high-emitting (low-high-emitting) input decreases the demand for the high-high-emitting (low-high-emitting) input and therefore the demand for emissions decreases (increases). The causalities between energy input prices and the permit price are depicted graphically in Keppler and Mansanet-Bataller (2010). For a complete model, all possible high-emitting and low-emitting inputs for all participants are to be included in the observed factors wi.
15 2.2.5 Renewables
In terms of our theoretical framework, renewable sources are very low emitting inputs in the generation of energy, and therefore increases in their use are expected to decrease the permit price. As discussed in the previous subsection, the simulation-based literature confirms the negative relationship between renewables and the permit price. Morris et al (2010) use a computable general equilibrium model for the United States, and estimate that a 20 percent RES standard by 2020 and 80 percent CO2 emissions decrease from 1990-2050 decreases the price of emission permits by 20 percent per year. Abrell and Weigt (2008) conclude that renewables negatively affect the price of emission permits, although they do not quantify the magnitude of this effect. Also, Boeters and Koornneef (2011), Bohringer and Rosendahl (2011), Fischer and Preonas (2010), Sijm (2005), and Traber and Kemfert (2009) conclude that the stronger the use of renewables becomes the lower the price for CO2 emission permits falls. Van Den Bergh et al (2013) use a partial equilibrium model of the electricity sector of West- and Southern-Europe using data on renewable energy generation for twelve countries in the EU ETS, and simulate that renewable policies may have decreased prices by an outer limit of 15-46 euros for the years 2007-2008. However, their interpretation becomes more troublesome for later years as the outer limit goes to levels at which renewables deployment may actually be optimal from a market point of view. They estimate that the price of emission permits must be at least 100 euro for onshore wind and bio sources, and up to 1000 euro for solar technology.
The string of empirical work on the relationship between the price of emissions and renewables is smaller than the simulation-based work. Koch et al (2014) use data on the deployment of renewables in eight European countries to examine the relationship with the price of permits. They estimate a significant and negative relationship between renewables deployment and the price of permits, though of a smaller magnitude than the simulation-based literature. Rickels et al (2015) empirically confirm the negative effect using data on Norwegian hydropower reservoirs. Estimating a negative relationship between renewable energy generation and the permit price demonstrates the price component of the waterbed effect. In short, overall consensus exists that renewables adoption in energy generation decreases the price of emission permits, although estimates for the strength of this effect are different between simulation-based and the two empirical studies.
16
et al, 2005). For example, Mansanet-Bataller et al (2007) and Keppler and Mansanet-Bataller (2010) find that unexpected deviations from mean temperature significantly affect the permit price, although Alberola et al (2008) find weather evidence only for a number of sub-samples of phase 1. Since our dataset contains information on the utilization of renewable energy sources, there is no theoretical basis for including a weather variable in the analysis. However, seasonality in the generation or consumption of energy may be addressed through the use of seasonal indicators. No hypotheses for the seasonal indicators are defined, as both Koch et al (2014) and Rickels et al (2015) find no significant seasonal effect To conclude this subsection, table 1 below provides an overview of all hypotheses for our permit price model.
Table 1: Hypotheses for the permit price model
Variable Relationship Production levels + International credits - Coal price - Gas price + Renewables - Seasonal indicators +/-
2.3 Explaining sectoral emission levels
Having discussed the literature and theory concerning the permit price, we now turn to the second question of this research. Under the waterbed effect, emissions shift away from the energy generation sector to unaffected sectors as a result of the constraints on renewables. Quantifying this shift without using a simulation-based approach is troublesome, as the market outcome of the EU ETS alone is counterfactual and therefore unobservable. To address this complication, we estimate the relationship between sectoral emissions of the unaffected EU ETS sectors and a number of variables using a panel data approach. It is remarkable that this relationship has not been empirically examined before, as the displacement of emissions is a key implication of the waterbed effect.
2.3.1 Emissions functions
17
𝑒𝑖(𝑤𝑖, 𝑢𝑖; 𝜃) (5)
Where 𝑒𝑖 is the level of sectoral emissions, 𝑤𝑖 are observed variables affecting the emissions function, 𝑢𝑖 are unobserved variables affecting the emissions function, and finally 𝜃 a vector of parameters accompanying 𝑤𝑖 and 𝑢𝑖. The hypothesis for each fundamental in 𝑤𝑖 or 𝑢𝑖 is then determined by its expected sign in 𝜃. Note that no work exists on sectoral emissions functions, but that a number of authors have described emissions determinants on a national level (Ang, 2008; Omri et al, 2014; Schandl et al, 2016; Stern et al, 1996)
2.3.2 Sectoral production
We follow the literature on national emission levels that dictates a positive relationship between production levels and emission levels. The volume of emissions increases for production volume increases as long as the marginal effect on emissions of production is positive. On a national level, the intuitive relationship is confirmed for different groups of countries with different methodologies by Ang (2008), Diakoulaki, and Mandaraka (2007) and Omri et al (2014). This is similar to the industrial production levels discussed in the previous subsection. Stern et al (1996) and Schandl et al (2016) describe the Environmental Kuznets Curve, which entails that an inverted U-shaped relationship may exist between environmental pollution and economic growth at a country level. Diakoulaki, and Mandaraka (2007) find limited evidence for the decoupling of industrial growth and emissions in the EU manufacturing sector. Unless production and emission are fully decoupled, a positive relationship exists between these two variables. The results are extended to the sectoral level, to hypothesize that production increases (decreases) boost (depress) emission levels.
2.3.3 Permit price and renewables
18
emissions. The hypotheses for the direct relationship between renewable energy generation and emissions of other sectors is positive under the waterbed effect.
2.3.4 Energy input prices
Substitution effects in the production functions of participants lead to the hypotheses of negative relationships for high-emitting input prices with emission levels, and positive for relatively low-emitting input prices. The reasoning is similar to the logic of the permit price hypotheses for input prices. In the production function of a participant typical inputs are capital, labor, energy and occasionally materials, also known as KLEM production functions (Kemfert and Welsch, 2000; Stern and Cleveland, 2004). Profit-maximizing behavior by participants suggests that the volume of inputs in the production process is based on their respective prices, with emission intensity differing per input. We assume that energy is a high-emitting input, and therefore expect a negative relationship between energy input prices and emission levels. We make use of the prices of the inputs gas and oil, and exclude the price of coal from the sectoral emissions model as the power generation sector is not included in our panel. Also, the electricity price is not included as it is strongly related to the price of gas.
2.3.5 Learning effects
Finally, the presence of learning effects in CO2 abatement may lead to a decreasing level of emissions over time, regardless of variations in the other explanatory variables. Manne and Richels (2004) and Miketa and Schrattenholzer (2004) discuss so-called ‘learning-by-doing’ (LBD) effects in the costs of abating emissions for firms that are affected by emissions reduction policies. Through for example technological advancement over time, firms improve their ability to reduce emissions. Therefore, a negative time trend is expected because of LBD effects. To conclude this section, table 2 contains the hypotheses for the sectoral emissions model.
Table 2: Hypotheses for the sectoral emissions model
19
3 Data
To address the research questions and hypotheses derived in the previous section, two datasets representing the relevant variables are constructed from a number of sources. This section is divided into two subsections, presenting the datasets of the permit price model and the sectoral emissions model respectively.
3.1 Permit price dataset
The dataset for the permit price estimation consists of monthly observations for the period between January 2008 and December 2016. Phase 1 data is excluded from the permit price analysis, as permits of that period were not bankable (Abrell et al, 2011). The existing empirical work on the permit price does not include extensive phase 3 data. An overview of the data series and their sources is presented in table 3, and their descriptive statistics in table 4. A correlation matrix of the variables in the permit price model is included in table A.1 in the appendix. Finally, figures 1a up to and including 1e display the series throughout this subsection.
Table 3: Data variables and sources permit price model
Series Specification Source
EUA price EU ETS permit price, end-of-year forwards ICE Production levels Index of volume of production for European
industry, 1/2008=100
Eurostat
CERs issued Worldwide Certified Emission Reduction (CER) credits issued, based on UNFCCC data
IGES CDM Project Database
Coal price High-emitting input forward price, API2 Argus/McCloskey index, ARA region
ICE
Gas price Low-emitting input forward price, UK NBP ICE Renewables Renewable energy generation in 19 EU ETS
countries
20
Table 4: Descriptive statistics permit price model, N=108
Variable Min Mean Max Standard
deviation EUA price (€/t CO2) 3.53 10.51 21.17 5.76 Production (%) 80.80 90.33 100.44 3.84 CERs issued (M) 2.28 15.44 62.82 12.20 Coal price (€/MWh) 4.94 8.20 15.83 2.27 Gas price (€/MWh) 9.02 21.37 38.04 5.79 Renewables (TWh) 32.40 52.42 74.77 10.77
The data on the permit price graphically depicted below in figure 1a is taken from the Intercontinental Exchange (ICE), the leading trading platform for emission permits forwards which acquired the European Climate Exchange (ECX) in 2010. Annual December forward prices are used, as the market for forward contracts is far more liquid than the spot market. Their end-of-year maturity corresponds with the compliance period for installations, as permits are surrendered on an annual basis to verify emissions. The series display a decrease over the period observed, with specifically sharp decreases in 2008 and 2011, and a trace of stability around a price of €5 from 2013 onwards.
Figure 1a. EU ETS permit price, end-of-year forwards. Source: ICE
21
series displayed below in figure 1b show a strong decrease to a level of around 80 percent in April 2009, followed by a more gradual path upwards towards almost 95 percent at the end of 2016. As previously discussed, the existing literature mainly makes use of stock market indexes rather than production data (Creti et al, 2012; Hintermann, 2010; Koch et al, 2014; Rickels et al. 2015). Therefore, we include the Stoxx Europe 600 index in the dataset to perform a sensitivity analysis later in this research. This broad-based index represents the value of a portfolio of shares in seventeen European countries, and deviates somewhat from the production level series, specifically from 2012 onwards.
Figure 1b. Index of volume of production for European industry and Stoxx Europe 600 stock market
index. Sources: Eurostat and Stoxx
22
Figure 1c. Worldwide Certified Emission Reduction (CER) credits issued, based on UNFCCC data.
Source: IGES CDM Database
Leading European reference prices for coal and gas are used in the dataset as high- and low-emitting inputs in the production process, respectively. Both series are presented in figure 1d below, and are month-ahead forward prices as these are the most liquid contracts. Although Rickels et al (2015) and Zaklan et al (2012) point out that no universal price of coal exists, the McKloskey API2 index delivered in the Amsterdam-Rotterdam-Antwerpen (ARA) is regarded as the leading price for northwest Europe and used by a number of authors (Koch et al, 2014; Lutz et al, 2013; Rickels et al, 2015). The ICE month-ahead forward contracts for coal are denominated in $/t, therefore the series are converted to €/MWh using exchange rates provided by the European Central Bank (ECB) and the appropriate constant. The gas prices in our dataset are month-ahead prices at the UK National Balancing Point (NBP), the most liquid gas product in the ICE. Similar to the coal price, we translate the gas price from pence/therm to €/MWh. Both series follow roughly the same path, although stronger variation exists in the gas price series. Both series peak in the second half of 2008 and drop to their lowest levels mid-2009. Movements of a more gradual nature follow from 2010 to 2016 onwards.
23
Finally, data on renewable energy generation in nineteen EU ETS countries is retrieved from the European Network of Transmission System Operators for Electricity (ENTSO-E) database, and presented in figure 1e. The sample of nineteen countries represents around 75 percent of renewable energy generation from all EU ETS participants over the observed period, and is selected as observations are missing for the excluded countries. The list of countries together with total energy generation data is included in figure A.1 in the appendix. Renewable energy generation increased in the nineteen countries over the observed period, from around 40 TWh per month in the years 2008 and 2009 to values around 60 TWh from 2013 onwards. Note that this dataset does not specifically distinguish renewables adoption as a result of the policy overlap, but that it does allow for conclusions to be drawn on the relationship between renewables and the permit price.
Figure 1e. Renewable energy generation in 19 EU ETS countries. Source: ENTSO-E
3.2 Sectoral emissions dataset
24
Table 5: Data variables and sources sectoral emissions model
Series Specification Source
Sectoral emissions Verified emissions per EU ETS sector EEA Sectoral production Index of volume of production per NACE Rev.2 sector
in EU28 countries, 2010=100
Eurostat
EUA price EU ETS permit price, year-ahead forwards ICE
Renewables Production of energy from renewable sources in EU28 countries
Eurostat
Gas price Gas price for industrial consumers in EU28 countries Eurostat
Oil price Brent crude (BFOE) oil spot price ECB
Trend variable Annually increasing trend variable -
Table 6: Descriptive statistics sectoral emissions model, N=182
Variable Min Mean Max Standard
deviation Sectoral emissions (MT CO2) 1.05 35.73 168.49 46.01 Sectoral production (%) 75.10 103.87 146.40 13.83 EUA price (€/t CO2) 4.55 11.98 24.44 6.57 Renewables (EJ) 5.02 7.25 8.90 1.24 Gas price (€/GJ) 5.26 7.61 8.92 1.00
Oil price (€/barrel) 39.89 61.38 87.02 16.27
Trend variable 0.00 6.22 11.00 3.41
25
Figure 2a. Verified EU ETS emissions per sector and indices of production volume per sector in
EU28 countries. Sources: EEA and Eurostat
26
Figure 2b. EU ETS permit price year-ahead forwards and the production of energy from renewable
sources in EU28 countries. Sources: ICE and Eurostat
Finally, the energy input prices used in the sectoral emissions dataset are displayed together in figure 2c. To represent the gas price in the input decision making of EU ETS participants, Eurostat’s gas price for industrial consumers in EU28 countries is used. The price of electricity is used in a sensitivity analyses later in this research and is plotted on the same axis as the gas price. The Brent crude oil price is the major worldwide oil reference price, and is taken from the ECB and plotted on the right-hand side y-axis. These energy input prices exhibit strong correlations of up to 0.81. Also, strong correlation exists between the permit price and renewable energy generation. Section 4 discusses the implications of these correlations for the robustness of the results.
Figure 2c. Gas- and electricity price for industrial consumers in EU28 countries and Brent crude
27 4 Methodology
The research questions investigate both elements of the waterbed effect by examining the determinants of the permit price and sectoral emissions. This section discusses and specifies the functional forms used to evaluate the hypotheses using a number of diagnostic statistical approaches. Section 5 evaluates the results of both approaches using a number of sensitivity analyses.
4.1 Permit price model
The permit price model evaluates the relationships between a number of market fundamentals and the price of EU ETS allowances using time series analysis. The graphical display of the time series in the previous section already hints that stationarity is to be tested. The results of the Augmented Dickey-Fuller (ADF) test with a null hypothesis of a unit root in the series (Dickey and Fuller, 1979) are displayed in table A.3 in the appendix. Non-stationarity is only rejected for the number of CERs issued, therefore the following ordinary least squares functional form is specified for the permit price model:
∆𝑙𝑛(𝑃𝐸𝑈𝐴)𝑡 = 𝛼0+ 𝛽1∆𝑙𝑛(𝐸𝑈 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑖𝑛𝑑𝑒𝑥)𝑡+ 𝛽2𝑙𝑛(𝑖𝑠𝑠𝑢𝑒𝑑 𝐶𝐸𝑅𝑠)𝑡−1+
𝛽3∆𝑙𝑛(𝑃𝑐𝑜𝑎𝑙)𝑡+ 𝛽4∆𝑙𝑛(𝑃𝑔𝑎𝑠)𝑡+ 𝛽5∆𝑙𝑛(𝑅𝑒𝑛𝑒𝑤𝑎𝑏𝑙𝑒𝑠 𝑔𝑒𝑛𝑒𝑟𝑎𝑡𝑖𝑜𝑛)𝑡+ ∑𝑗=3𝑖=1𝛽𝑖𝐷𝑠𝑒𝑎𝑠𝑜𝑛𝑡+ 𝜀𝑡 (6)
28
Alternative assumptions on the distribution of the error term are not incorporated, as a number of tests fail to reject either null of no serial correlation or homoskedasticity. Breusch-Godfrey and Durbin-Watson tests fail to reject first- or higher-order serial correlation in the residuals (Breusch, 1978; Durbin and Watson, 1950; Godfrey, 1978). Heteroskedasticity in the distribution of errors is likely not an issue, as both a White test and a Breusch-Pagan test fail to reject a homoskedastic error distribution (Breusch and Pagan, 1979; White, 1980). Furthermore, collinearity concerns among energy input prices are addressed by calculating the variance inflation factors (VIFs) of equation (6). Although its rule of thumb method is criticized, the highest value of 1.80 in table A.4 indicates that collinearity is not present (O’Brien, 2007; Mansfield and Helms, 1983; Verbeek, 2008). This conclusion matches the result of Keppler and Mansanet-Bataller (2010) that coal and gas prices are not collinear. Finally, serious misspecification is evaluated using the RESET test by Ramsey (1969). This auxiliary regression method fails to reject proper specification at the 10 percent level, signaling that the functional form in (6) is a correct functional form specification.
4.2 Sectoral emissions model
The sectoral emissions model investigates the relationship of sectoral emissions with the permit price and a number of control variables using a panel data approach. It attempts to test the second component of the waterbed effect, namely the emissions increase in other EU ETS sectors. Similar to the previous subsection, it is first determined whether a data transformation is required to overcome unit roots in the variables. The presence of unit roots in the panel series is evaluated using the ADF test with three year lags, in a panel context also known as a Fisher test, and table A.5 in the appendix demonstrates that non-stationarity is strongly rejected only for emission levels and production indices (Choi, 2001; Fisher, 1932; Maddala and Wu, 1999; Verbeek, 2008). Hence, all remaining variables are log-differenced to remove their unit roots. Moreover, emissions and production are also log-differenced to overcome the obstacle that sectoral emissions are in absolute levels, but sectoral production is indexed. This leaves either a fixed effects or a random effects specification, respectively:
29
∆𝑙𝑛(𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠𝑖,𝑡) = 𝛼0+ 𝛽1∆𝑙𝑛(𝑆𝑒𝑐𝑡𝑜𝑟𝑎𝑙 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑖𝑛𝑑𝑒𝑥)𝑖,𝑡+ 𝛽2∆𝑙𝑛(𝑃𝐸𝑈𝐴)𝑡+ 𝛽3∆𝑙𝑛(𝑃𝐺𝑎𝑠)𝑡+ 𝛽4∆𝑙𝑛(𝑃𝑂𝑖𝑙)𝑡+ 𝛽5𝑇𝑟𝑒𝑛𝑑𝑡+ 𝑎𝑖 + 𝜀𝑖,𝑡 (8)
In a fixed effects approach (equation 7) the results are conditional on the intercept terms 𝛼𝑖, which are sector-specific and capture all observed and unobserved time-invariant effects, whereas the error term 𝜀𝑖,𝑡 is sector-independent. On the other hand, a random effects approach (equation 8) is not conditional on individual values of 𝛼𝑖 and is better suited to infer population results, with the error term consisting of a sector-specific component 𝑎𝑖 and a remainder component 𝜀𝑖,𝑡 (Verbeek, 2008; Wooldridge, 2010). This research aims to infer mainly in-sample results as virtually the entire population of EU ETS sectors is included in the in-sample. Therefore, a fixed effects approach may be preferred, although cautiousness is required in the choice. Hence, a Hausman test of no correlation between the explanatory variables and 𝑎𝑖, a property of the random effects specification, is conducted (Hausman, 1978). Relatively little difference exists between the fixed effects and random effects estimators, and with a p-value of 0.72 the Hausman test seriously fails to reject the null of no correlation, suggesting that a random effects specification may be preferred. Although the Hausman test comes with its own considerations such as strict exogeneity, we follow its results and use a random effects specification of equation (8) for the sectoral emissions model (Guggenberger, 2010; Hausman, 1978; Verbeek, 2008; Wooldridge, 2010).
30
The estimation of the random effects sectoral emissions model of equation (8) may yield improper results if collinearity and/or endogeneity are present. First, collinearity among the regressors may lead to unreliable estimation results for individual estimators (Verbeek, 2008). Therefore, similar to the previous subsection, the VIFs of the random effects sectoral emissions model in equation (8) are calculated in table A.6. Inspection of the VIFs takes away the suspicion of serious collinearity among the regressors.
Second, reverse causality or simultaneity between sectoral emissions and the permit price may be present. Such endogeneity leads to biased and inconsistent estimators, and can be resolved by using an instrumental variable (IV) approach (Verbeek, 2008). A Hausman test that compares the IV and random effects estimators can be conducted to test for endogeneity under the assumption that the instruments are both relevant and exogenous (Hausman, 1978). To instrument for the potentially endogenous regressor permit price the generation of renewable and the number of CERs issued are used, their relevance is confirmed in the next section. A Sargan test of overidentifying restrictions fails to reject instrument exogeneity at the 10 percent level (Sargan, 1958; Verbeek, 2008). Hence, the instruments are both relevant and exogenous, and the assumption of instrument validity under the Hausman test is not violated. The Hausman test’s p-value of 0.38 fails to rejects the null of no endogeneity with any confidence, therefore we conclude that endogeneity is not an issue and follow the random effects specification of equation (8) including sector-clustered standard errors.
Finally, an alternative specification of the sectoral emissions model is presented below in equation (9). Although the waterbed effect is expected to occur through the permit price, renewable energy generation is included as then equation (9) directly observes the presence of a relationship between renewable energy generation and emissions of the other sectors. The permit price is excluded from this specification as including both renewable energy generation and the permit price is redundant. Again, sector-clustered standard errors are included as a result of the appropriate tests discussed above. The VIFs of equation (9) are included in table A.7 in the appendix, and suggest that the specification does not suffer from collinearity problems.
∆ln (𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠𝑖,𝑡) = 𝛼0 + 𝛽1∆ ln(𝑆𝑒𝑐𝑡𝑜𝑟𝑎𝑙 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑖𝑛𝑑𝑒𝑥)𝑖,𝑡+
𝛽2∆ ln(𝑅𝑒𝑛𝑒𝑤𝑎𝑏𝑙𝑒𝑠 𝑔𝑒𝑛𝑒𝑟𝑎𝑡𝑖𝑜𝑛)𝑡+ 𝛽3∆ ln(𝑃𝐺𝑎𝑠)𝑡+
31
5 Results
This section presents and interprets the results of both models used to answer the research questions and evaluate the hypotheses. Section 6 discusses a number of limitations in the research and considerations of these results.
5.1 Permit price model
The permit price model as specified in equation (6) estimates to what extent variations in the price of emission allowances are explained by production levels, the use of international credits, the price of coal, the price of gas, renewable energy generation, and the seasons. The results of the ordinary least squares regression without alternative error distribution assumptions are presented in table 7 below.
Table 7: Estimation results permit price model
Coefficient Corresponding variable Estimate Standard error P-value
𝛼0 𝐼𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 0.077** 0.035 0.027 𝛽1 𝑙𝑛(𝐸𝑈 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑖𝑛𝑑𝑒𝑥)𝑡 1.750** 0.798 0.031 𝛽2 𝑙𝑛(𝑖𝑠𝑠𝑢𝑒𝑑 𝐶𝐸𝑅𝑠)𝑡−1 -0.022* 0.012 0.075 𝛽3 𝑙𝑛(𝑃𝑐𝑜𝑎𝑙)𝑡 0.094 0.121 0.437 𝛽4 𝑙𝑛(𝑃𝑔𝑎𝑠)𝑡 0.140 0.088 0.116 𝛽5 𝑙𝑛(𝑅𝑒𝑛𝑒𝑤𝑎𝑏𝑙𝑒𝑠 𝑔𝑒𝑛𝑒𝑟𝑎𝑡𝑖𝑜𝑛)𝑡 -0.255*** 0.097 0.010 𝛽6 𝐷𝑠𝑢𝑚𝑚𝑒𝑟𝑡 -0.043* 0.025 0.086 𝛽7 𝐷𝑎𝑢𝑡𝑢𝑚𝑛𝑡 -0.042* 0.025 0.091 𝛽8 𝐷𝑤𝑖𝑛𝑡𝑒𝑟𝑡 -0.054** 0.023 0.022 R2 0.254 adj-R2 0.194 F 0.000 N 107
Note: * = significant at the 10% level, ** = significant at the 5% level, *** = significant at the 1% level
32
direction of 𝛽5 provides empirical support for the waterbed theory and the simulation-based literature, however the magnitude is substantially smaller. As both the permit price- and renewables series are log-differenced, coefficient 𝛽5 is defined as the ‘renewables elasticity’ of the permit price. The intuitive interpretation that follows from the magnitude of the coefficient is that a 1 percent increase in the level of renewable energy generation decreases the EUA price by roughly 0.25 percent. A very brief and imprecise back-of-the-envelope calculation to illustrate coefficient 𝛽5 suggests that, ceteris paribus, an increase of the 2015 share of 16.7 percent renewables in energy generation in EU28 countries to 20 percent may decrease EUA prices by3 €0.25. This magnitude stands in stark contrast with those simulation-based articles that have specifically projected the impact of renewable targets on the EU ETS permit price (Morris et al, 2010; Van Den Bergh et al, 2013). On the other hand, the magnitude of -0.25 percent is larger than the combined elasticity of solar, wind, and hydro energy of -0.15 estimated by Koch et al (2014). Comparing the magnitude with Rickels et al (2015) is somewhat more troublesome, as their cointegration analysis only estimates the relationships of hydro reservoir levels in France and Norway with the EUA price.
Production levels are estimated to positively affect the permit price in the EU ETS, a result that supports the theory and the hypothesis for coefficient 𝛽1. As production is emissions-intense, higher volumes are accompanied by an increase in the demand for emission permits and accordingly the permit price. The estimate of 1.75 implies that a more-than-unit elastic relationship exists between indexed production levels of European industry and the EUA price, i.e. a 1 percent increase in production volume, ceteris paribus, increases the EUA price by 1.75 percent. Our approach is different from the literature, as most existing work makes use of stock market indexes or even end-product prices to represent production levels (Aatola et al, 2013; Creti et al, 2012; Hintermann, 2010; Koch et al, 2014; Rickels et al, 2015). As a sensitivity analysis, table A.8 presents the results of the permit price estimation using Stoxx Europe 600 stock market data as a variable representing production levels. Although the directions of the effects are similar to the results in table 7 above, the magnitude and significance differ for a number of variables. As the use of a production volume index is closer linked to the theoretical determinant of emissions than a stock market index, we emphasize that future research is to be
3 An increase from a share of renewables in energy generation from 16.7 to 20 percent implies, ceteris paribus,
33
cautious in including stock market indexes to represent production levels of EU ETS participants.
The hypothesis for the use of international credits in the EU ETS is a negative relationship, as their market prices are far lower than EUA prices. The negative estimate of -0.022 for coefficient 𝛽2 hints in the direction of the hypothesis, but the p-value of 0.075 does not provide the estimate with strong confidence. The estimate of -0.022 is not far off the estimates in Koch et al (2014), which is the only work to date that includes an international credits variable in a number of different specifications. A number of considerations to the use of this variable are discussed in the next section.
Both energy input price coefficients fail to confirm their hypotheses, a result that is robust against the maturity of the forward contracts in the price series. Because of switching possibilities in the production process, the hypothesis is a negative (positive) relationship for the price of the high-emitting (low-emitting) input coal (gas). Specifically the estimate 𝛽3 for the coal price is very insignificant, with the gas price estimate 𝛽4 having the expected sign but failing to reject the null of no relationship at the 10 percent level. Rickels et al (2015) find that the selection of price series can strongly affect the estimation results, therefore we perform a sensitivity analysis using year-ahead, rather than month-ahead, forward prices of coal and gas. The results of the sensitivity analysis in table A.9 in the appendix do not differ in sign or significance levels for all variables of interest in the permit price model, with also the model fit at a similar level. Thus, the estimates for both energy input prices appear unaffected by the maturity of the forward contracts in the price series. The discrepancy between the energy input hypotheses and the results is discussed in section 6.
34
5.2 Sectoral emissions model
The sectoral emissions model estimates the relationship between sectoral emissions of all EU ETS sectors excluding the power generation sector and sectoral production levels, the permit price, energy input prices, and an annually increasing trend variable. The results of the random effects estimation of equation (8) with sector-clustered standard errors are presented below in table 8.
Table 8: Estimation results sectoral emissions model
Coefficient Corresponding variable Estimate Standard error P-value
𝛼0 𝐼𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 -0.050 0.043 0.243 𝛽1 𝑙𝑛(𝑆𝑒𝑐𝑡𝑜𝑟𝑎𝑙 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛)𝑖,𝑡 0.679*** 0.260 0.009 𝛽2 𝑙𝑛(𝑃𝐸𝑈𝐴)𝑡 -0.017 0.031 0.590 𝛽3 𝑙𝑛(𝑃𝑔𝑎𝑠)𝑡 0.183 0.141 0.194 𝛽4 𝑙𝑛(𝑃𝑜𝑖𝑙)𝑡 0.002 0.033 0.942 𝛽5 𝑇𝑟𝑒𝑛𝑑𝑡 0.007 0.006 0.264 Within-R2 0.166 Between-R2 0.009 Overall-R2 0.158 Chi-squared 0.000 N 162
Note: * = significant at the 10% level, ** = significant at the 5% level, *** = significant at the 1% level
35
Using the alternative specification of the sectoral emissions model, a positive relationship is estimated between renewable energy generation and the emissions of all but the power generation sectors in the EU ETS. The logic of the waterbed effect dictates that an increase in the use of renewable sources in energy generation is expected to increase emissions in other sectors through the permit price. The direct estimation uses annual Eurostat data on total renewable energy generation in the EU28 area, and the results are displayed below in table 9. A positive and significant relationship between renewable energy generation and the emissions of other sectors is estimated, with the remaining estimates largely unaltered and the model fit moderately improved. The elasticity estimate suggests that a 1 percent increase in renewable energy generation leads to a 0.555 percent increase in emissions of the other EU ETS sectors. The next section addresses the fact that a direct positive relationship between renewable energy generation and the emissions of other sectors is observed, but that the effect through the permit price is not confirmed by the sectoral emissions model.
Table 9: Sectoral emissions model alternative specification: including renewables generation in the place of the permit price
Coefficient Corresponding variable Estimate Standard error P-value
𝛼0 𝐼𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 -0.094* 0.057 0.099 𝛽1 𝑙𝑛(𝑆𝑒𝑐𝑡𝑜𝑟𝑎𝑙 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛)𝑖,𝑡 0.649*** 0.230 0.005 𝛽2 𝑙𝑛(𝑅𝑒𝑛𝑒𝑤𝑎𝑏𝑙𝑒𝑠 𝑔𝑒𝑛𝑒𝑟𝑎𝑡𝑖𝑜𝑛)𝑡 0.555** 0.224 0.013 𝛽3 𝑙𝑛(𝑃𝑔𝑎𝑠)𝑡 0.211 0.149 0.156 𝛽4 𝑙𝑛(𝑃𝑜𝑖𝑙)𝑡 0.002 0.025 0.925 𝛽5 𝑇𝑟𝑒𝑛𝑑𝑡 0.010 0.007 0.170 Within-R2 0.178 Between-R2 0.010 Overall-R2 0.170 Chi-squared 0.000 N 162
* = significant at the 10% level, ** = significant at the 5% level, *** = significant at the 1% level
36
Environmental Kuznets Curve, coefficient 𝛽1 clearly demonstrates that although increases in production are not accompanied one-to-one by increases in emissions, sectoral production and sectoral emissions are far from decoupled. In the future, the relationship may become less strong, as ultimately the aim of the EU ETS is to strongly decrease emissions without decreasing output.
Both estimates 𝛽4 and 𝛽5 of the energy input prices gas and oil respectively do not significantly affect sectoral emissions, with the estimate for oil specifically close to zero and very insignificant. The hypotheses concerning the energy input variables were negative under the assumption that energy sources are emission-intense inputs in a KLEM production function (Kemfert and Welsch, 2000; Stern and Cleveland, 2004). Although insignificant, one must note that the estimate for the gas price is in the opposite direction of the hypothesis. As a sensitivity analysis, table A.10 contains the results of the sectoral emissions model using the electricity price rather than the gas price. The results of the sensitivity analysis are very similar to the results in tables 8 and 9. The discussion in the next section addresses a number of data- and methodological aspects that may play a role in the insignificance of the energy input estimates in the sectoral emissions model.
37 6 Discussion
In this section a number of shortcomings and potential improvements of this research are discussed together with their possible impact on the results of the estimations. Furthermore, it is noted that future policy measures related to the supply of permits in the EU ETS may to some extent impact the occurrence of the waterbed effect.
Perhaps the most room for improvement of this research lies in three aspects of the sectoral emissions model. First, the annual nature of observations in the sectoral emissions dataset provides an average of only 8.1 observations per EU ETS sector over the period 2005-2016. Although data on all explanatory variables used in the analysis are available at a higher frequency, a large amount of variation is lost as the dependent variable emissions is only updated each year. Hence, the availability of more frequent emissions data directly improves the quality of the model, and can provide future researchers with more confident estimation of the relationships between sectoral emissions and the variables of interest.
38
Finally, it is a through-provoking result that the sectoral emissions model fails to confirm a negative relationship between the permit price and sectoral emissions, but that using an alternative specification a positive relationship is observed between renewable energy generation and the emissions of other sectors. In the emissions trading theory, this relationship should work through the permit price of the EU ETS (Tietenberg, 2006). The first step of the waterbed effect, namely the negative relationship between renewable energy generation and the permit price, is confirmed by the permit price model. Although the direct positive relationship between renewable energy generation and emissions of other sectors suggests that the waterbed effect is present, more analysis of sectoral emissions may provide a better understanding of this discrepancy.
Imperfections and improvements for future research also exist in the permit price model, as the theory-derived hypotheses for the prices of coal and gas are not confirmed in the estimation. Hence, either the hypotheses are an incorrect representation of reality or the dataset is not of sufficient quality to test the hypotheses. If the problem lies in the forming of the hypotheses, then the switching between coal and gas may be less straightforward than is assumed, or expectations of future energy input prices may play a more important role than market prices do. In most existing work, the price of the less-emitting input gas is estimated to exhibit a significant and positive relationship with the permit price, with more mixed results for the price of coal (Aatola et al, 2013; Lutz et al, 2013; Hintermann et al, 2016). In Koch et al (2014), the significance of energy input prices is specification-dependent. On the other hand, Rickels et al (2015) illustrate that the estimation results may not be robust against the selection of a different price series with a different delivery location of the energy product. However, this research attempts to use the most representative price series by using the leading price for northwest Europe, and the results are robust against the maturity of the forward contract in the price series. Considering the fact that also the existing literature not always confirms the hypotheses, future research is suggested to gather more clarity on why the theoretical relationships between energy input prices and the EUA permit price may not be present in practice.
39
included as it is unavailable at the moment of writing. Also, a maximum exists on the number of international credits that may be used by participants over a phase of the EU ETS. This is unaccounted for in the analysis, as the maximum limit concerns a number of years and is not imposed on a monthly level. Future research on the relationship between the use of international credits and the EU ETS permit price may provide both academics and policy makers with more clarity on the effects of linking ETS systems.
