The Design Elements of Feed-in Tariff Policies
Estimating the effect of design elements on the dynamic
efficiency of feed-in tariff schemes
Master Thesis Public Administration: Economics and Governance Rianne van Staalduinen, S2090392
Supervisor: dr. M.C. Berg
Capstone: The economic effects of government intervention 11-06-2018
2
Abstract
To mitigate climate change and to promote the production of renewable energy, governments have designed various policy initiatives. One of these initiatives is a feed-in tariff (FIT), which is a fixed renumeration for a set period of time for electricity produced from renewable energy sources, such as wind or photovoltaics. FIT schemes have been an often-discussed topic in the academic literature. This research adds to this debate by examining how different design elements of feed-in tariff policies affect the dynamic efficiency of the policy, thereby testing the theoretical framework of Pablo del Río (2012).
Regression analysis is used to estimate the effect of eleven design elements, the independent variables, on five dimensions of dynamic efficiency, the dependent variables. A sixth dependent variable, the renewable energy share, is added to the model of Del Río to measure the impact of design elements on a total outcome score. The results of the regressions suggest that design elements have an impact on the six outcome variables as discussed in this research. Furthermore, country characteristics also influence policy outcomes. More research is recommended to further specify the impact of design elements and country characteristics.
3
Table of content
List of abbreviations 1. Introduction
2. The theoretical underpinnings of FIT schemes 2.1. Why implement a FIT scheme?
2.2. Literature review 2.3. The downsides of FITs
2.4. The role of FIT policy design elements 2.5. Del Río’s framework: independent variables 2.6. Del Río’s framework: dependent variables 2.7. Hypotheses
3. Data and methodology 3.1. Independent variables 3.2. Dependent variables 3.3. Control variables 3.4. Hypotheses 3.5. Methodology 4. Results and Discussion
4.1. Technological diversity 4.2. Private RD&D expenditure 4.3. Learning effects
4.4. Technological competition 4.5. Total consumer costs 4.6. Renewable energy share 5. Conclusion
5.1. Recommendations for future research 5.2. Implications for policymakers
5.3. Final concluding remarks 6. Bibliography Appendix A Appendix B Appendix C Appendix D Appendix E 4 5 7 7 8 9 11 12 14 16 20 20 22 25 27 29 31 31 32 34 35 36 38 42 42 43 38 45 57 61 63 65 69
4
List of abbreviations
ACER Agency for the Cooperation of Energy Regulators BNEF Bloomberg New Energy Finance
EU European Union
FIT Feed-in tariff
GCI Global Competitiveness Index
GDP Gross Domestic Product
IEA International Energy Agency
IPCC Intergovernmental Panel on Climate Change IRENA International Renewable Energy Association LCOE Levelised costs of electricity
NEA Nuclear energy agency
OECD OLS
Organisation for Economic Cooperation and Development Ordinary least squares
PPP Purchasing power parity R&D Research and development
RD&D Research, development and demonstration REN21 Renewable Energy Network for the 21st Century RPS Renewable portfolio standards
UN United Nations
UNDP United Nations Development Programme UNEP United Nations Environmental Programme
5
1. Introduction
In their communiqué presented after the 2015 Antalya summit, the G20 leaders declared that “climate change is one of the greatest challenges of our time” (G20, 2015). Even though the discussion on the causes and the impact of climate change remains, it is most likely that climate change is caused by an exponential increase in the emissions of greenhouse gasses (Stocker et al., 2013; European Commission, n.d.; Wageningen University, n.d.). One of the main sources of environmental pollution is energy, producing around 60 percent of the world’s greenhouse gas emissions (United Nations (UN), n.d.). Consequently, “ensuring access to affordable, reliable, sustainable and modern energy for all” has been established as the seventh UN’s Sustainable Development Goal (UN, n.d.). To mitigate climate change and to promote the production of renewable energy, governments have designed various policy initiatives. One of these initiatives is a feed-in tariff (FIT). Feed-in tariff policies guarantee fixed prices for a set period of time for electricity produced from renewable energy sources, such as wind or photovoltaics (Couture & Gagnon, 2010, 955).
Ever since the first FIT scheme was implemented in the United States in 1978, FITs have been an often-discussed topic in the academic literature (see for example Barclay, Gegax and Tschirhart, 1989; Morthorst, 1999; Menanteau, P., Finon, D. & Lamy, M., 2003; Lesser and Su, 2008; Ritzenhofen and Spinler, 2016). Various researchers have for example examined the effectiveness of feed-in tariffs and have concluded that feed-in tariffs are a successful policy tool to promote renewable energy (e.g. Haas et al., 2001; Butler & Neuhoff, 2008; Dong, 2012). Numerous researchers have also discussed the importance of studying the design elements of feed-in tariff policies (for example Held et al. (2010) and Couture and Gagnon (2010)). Less research, however, has been dedicated to testing the design of feed-in tariff policies to study how different design elements relate to different policy outcomes. This research therefore aims at studying this literature gap by examining how different design elements of feed-in tariff policies affect the efficiency of the policy.
One author who has covered the topic of design elements of FIT policies extensively is Pablo del Río (Del Río, 2008; Del Río, 2012; Del Río and Mir-Artigues, 2012; Del Río et al., 2017). In his paper on the impact on different design elements of FIT policies, Del Río (2012) provides a literature overview to establish a framework to study the impact of design elements on the dynamic efficiency of feed-in tariff policies. He defines dynamic efficiency as the capacity of
6 a FIT design element to promote continuous technological improvements and cost reductions for existing renewable energy technologies, enable an advancement of new technologies and support technologies with different maturity levels (Del Río, 2012, 139). Del Río consequently outlines five dimensions of dynamic efficiency and describes eleven design elements that influence the dynamic efficiency of FIT policies via these five dimensions. The research by Del Río therewith provides a comprehensive framework to test the effect of design elements of FIT policies. Since another study testing this framework has not been found, the study by Del Río is chosen as the theoretical framework for the study of design elements of this paper. The aim of this research is consequently to analyse how different design elements of feed-in tariff policies affect the dynamic efficiency of the policy, thereby testing the theoretical framework established by Del Río. The research question is therefore constituted as follows:
Following the theoretical framework of Pablo del Río (2012), to what extent do design elements affect the five dimensions of dynamic efficiency of feed-in tariff policies?
To answer this research question, the rest of this paper is structured as follows. Section 2 outlines the theoretical debate regarding FIT schemes and presents an overview of the literature that has been published on the subject so far. Furthermore, section 2 outlines the theoretical framework provided by Del Río and presents the hypotheses for the analysis. Subsequently, the third section of this paper operationalises the concepts outlined in the theoretical framework and introduces an extra outcome variable outside the framework of Del Río. Moreover, it presents the data used for the analysis and further specifies the methodology used in the analysis. Section 4 presents the main results of the analyses and discusses the implications of the main findings. Consequently, the conclusion seeks to contribute to the debate on the design elements of FIT policies by examining their effect on dynamic efficiency.
7
2. The theoretical underpinnings of FIT schemes
FIT schemes have been often-discussed in the academic literature and consequently, there are many explanations of the core concepts. Therefore, it is necessary to clearly explain and define these concepts, before analysing the impact of FIT design elements on the dynamic efficiency of FIT policies. Therefore, the following section firstly identifies the theoretical debate regarding FIT schemes and provides a literature review. Secondly, the theoretical framework provided by Del Río is explained, from which the hypotheses for the analysis are drawn.
2.1. Why implement a FIT scheme?
Feed-in tariffs is the classification used for a set of policies aimed at supporting the development of renewable energy. By a feed-in tariff scheme, producers of renewable energy are paid either a set rate or a premium to ‘feed’ their electricity into the grid. The first version of a FIT scheme was implemented in 1978 in the United States. Since then, the scheme has been developed, revised and altered, and is currently employed in many different forms in 110 countries, either on state level or on regional level (Dong, 2012; Ritzenhofen and Spinler, 2016; REN21, 2017). FIT schemes differ on various design elements, such as the relation to the energy market, the diversification of technologies, or the duration of the support. As the design elements are the main topic of analysis in this paper, the different design elements of FIT schemes currently employed are extensively discussed later. Nevertheless, to characterise as a FIT scheme, a policy must guarantee a set payment for electricity generated by pre-specified, renewable energy sources (Dong, 2012; Ritzenhofen and Spinler, 2016; REN21, 2017).
FIT schemes are based on a number of economic arguments. First and foremost, FIT schemes are implemented to make renewable energy sources compatible with non-renewable, conventional energy sources, such as fossil fuels. Fossil fuels cause pollution, which is not internalised in the price of the fuels, therewith creating negative externalities (Ragwitz and Steinhilber, 2014). Since renewable energy sources do not create these negative externalities, it is argued that the benefits of no pollution should be supported by government intervention. By providing financial benefits for suppliers of renewable energy, FIT schemes can be used to incentivise the supply of renewable energy sources (Marques and Fuinhas, 2012; Stokes, 2013).
8 The second economic argument for the implantation of FIT schemes is the reduction of risks involved with investing in renewable energy sources. The production of renewable energy can namely be characterised by high capital costs, but low variable costs, such as operation and maintenance costs. Consequently, the investment costs and the associated risks are high. Since FITs guarantee a certain payment level for a set duration of time, the risks of investing in renewable energy technologies are significantly lowered. By providing operators and investors with a certain level of certainty, investment in new technologies grows and the supply of renewable energy increases. This argument is further supported by the so-called learning effects of renewable energy sources, meaning that by running a new technology, learning by doing will occur, leading to a decrease in the production costs of new technologies in the long run. Thereby, the barriers for developing technologies with high potential, but with high development costs compared to non-renewable energy sources, are removed. By reducing the risks of investing in new renewable energy technologies, FITs therefore also stimulate innovation and support the creation of new energy sources (Couture and Gagnon, 2010; Davies and Allen, 2014; Ragwitz and Steinhilber, 2014; Stokes, 2013).
Politically, there are also multiple arguments to support the implementation of a FIT scheme. Firstly, FITs are not only available for small scale energy project, but in many countries, they are also offered to small scale individual producers, for example house-owners who want to install solar panels on the roof of their houses. Consequently, citizens are able to directly benefit from the policy, which makes it more likely that they will support the implementation of a FIT scheme policy compared to other policies aimed at promoting renewable energy (Stokes, 2012). Secondly, FIT schemes stimulate the deployment of nationally available energy sources. Therewith, FIT schemes also stimulate national job growth in the renewable energy sector, as for example extra construction workers have to be employed to build new energy plants. Furthermore, FIT schemes thereby also stimulate the production of national energy, which reduces the dependence of a state on foreign energy supplies (Marques and Fuinhas, 2012; O’Sullivan, Edler, Ottmüller and Lehr., 2013).
2.2. Literature review
The aforementioned arguments in favour of FIT policy have often been tested in the literature on FIT policies, out of which a few examples are discussed here. Firstly, multiple authors have sought to examine whether FIT schemes are an effective policy for increasing a country’s renewable energy production. A number of authors has developed comparative descriptive
9 studies to discuss the role of FIT policies in promoting renewable energy and has concluded that FIT schemes have been an efficient tool to promote renewable energy development in Europe (among others Butler and Neuhoff, 2008; Haas et al., 2011; Verbruggen and Lauber, 2012; Davies and Allen, 2014). Other authors have quantitively examined the efficiency of FITs. Dong (2012) has for example compared the ability of both FITs and Renewable Portfolio Standards (RPS), a quota system for the promotion of renewable energy, to increase the production of wind energy by using panel data. He concluded that FIT schemes increase a country’s wind capacity to a larger extent than RPS. Marques and Fuinhas (2012) also use panel data from 23 European countries to examine the effect of public policies supporting renewables and conclude that feed-in-tariffs have been an effective policy for fostering renewable energy supply.
Not only the effect of FIT schemes on renewable energy supply has been discussed, authors also sought to examine the relation between FITs and innovation, as theory suggests that FITs stimulate innovation in the renewable energy sector. Firstly, Johnstone, Haščič and Popp (2010) use patent data to identify the effect of different public policies on renewable energy innovation and conclude that feed-in tariffs are able to induce innovation on more costly technologies, such as solar panels. Furthermore, research by Böhringer, Cuntz, Harhoff and Asane-Otoo (2017) shows the positive relation between FITs and induced innovation in Germany. On a larger scale, Lindman and Söderholm (2016) have concluded from their analysis of data from four countries between 1997-2009 that feed-in tariffs positively affect innovation of the production of wind energy.
2.3. The downsides of FITs
Besides the aforementioned arguments in favour of FIT schemes, many authors have also discussed the disadvantages of FITs. Firstly, it is discussed whether FITs are necessary to lower pollution and reach the climate change mitigation goals. As pollution in the energy sector is caused by negative externalities of non-renewable energy sources, an alternative market intervention is to cap the emissions of fossil fuels, or to tax the production of polluting energy sources. It is argued that these policies, when implemented well, are able to lower carbon emissions and are more cost-effective policy options. This leads to the argument that further market interventions to stimulate the production of green energy are unnecessary (Ragwitz and Steinhilber, 2014; Palmer and Burtraw, 2005; Frondel, Ritter, Schmidt and Vance, 2010). The main reason for the debated cost-effectiveness of FITs is the high costs involved in the
10 development of renewable technologies. In many FIT schemes, these costs are charged to electricity consumers via higher electricity prices, which has for example been the case in Denmark, Germany, Spain and Cyprus (Frondel et al., 2010; Pyrgou, Kylili, and Fokaides, 2016; Davies and Allen, 2014).
Another disadvantage of FIT schemes is the information asymmetry between producers and the government, as the government does not exactly know what specific costs are involved in the development of renewable energy plants. This complicates the adequate setting of the FIT rate or premium. A FIT rate that is set too high leads to an over-stimulation of renewable energy and can lead to the development of inefficient technologies which are not sustainable in the long run. Furthermore, this leads to higher costs for consumers and may consequently weaken support for the policy. On the other hand, FIT rates that are set too low, the production levels of renewable energy stagnate, and the effect of the FIT scheme is limited (Davies and Allen, 2014; Stokes, 2012; Sakah, Diawuo, Katzenbach and Gyamfi, 2017). Empirical research by Dusonchet and Telaretti (2010) on support mechanisms for photovoltaics of eastern European Union (EU) member states therefore concludes that only the states which have been able to create support measures which are adequate to cover investment costs have been able to increase trade in photovoltaic installations. Consequently, in the Baltic states and Hungary, impact of the FIT scheme was limited.
A third argument against FITs is the inequality that can be created by FIT schemes. As explained earlier, in most countries, small scale energy projects are supported by FIT policies, enabling for example house-owners to install solar panels on their roof. Thereby, FIT policies benefit high income households more than low income households, as they can afford to invest in renewable energy sources. This inequality is amplified by the fact that in most countries, the costs of FIT schemes are charged to consumers, which leads to lower-income consumers having to pay higher electricity bills without having the possibility to invest in renewable energy (Davies and Allen, 2014). This argument is empirically examined by Grover and Daniels (2016), who conclude that FITs have led to an unequal distribution of households investing in solar panels in Great-Britain.
The last argument against the implementation of FITs given here is the crowding out effect they might cause. As explained in the advantages earlier, an often-mentioned argument in favour of FITs is the increased investment in renewable energy technologies, supporting innovation, and
11 the associated job creation in the renewable energy sector. These effects may however be offset by the crowding out of investment in other sectors, which leads to job losses in these sectors (Heindl and Voigt, 2012). Research on the employment effects in Germany for example conclude that, when taking into account job losses in other sectors, the net employment effect of the German FIT scheme has been negative (Frondel et al., 2010).
2.4. The role of FIT policy design elements
Despite the aforementioned downsides of FITs, many authors have concluded that FIT schemes are an effective tool to promote the production of renewable energy (among others Butler and Neuhoff, 2008; Haas et al., 2011; Verbruggen and Lauber, 2012; Davies and Allen, 2014; Dong, 2012; Marques and Fuinhas, 2012). That is why the Renewable Energy Network for the 21st Century (REN21) concluded that in 2016, FITs “remained the most prominent form of regulatory policy support for renewable power promotion” (REN21, 2017, 122). Furthermore, the fact that there is discussion on the effect of FITs only increases the relevance of this paper, as it is important to analyse the effect of policy design elements in order to better design these schemes in the future. Consequently, this paper does not focus on testing the effects of FIT policy schemes, but instead seeks to analyse the effect of FIT policy design elements.
According to Ragwitz and Steinhilber (2014), the success of a FIT policy is largely influenced by the concrete design elements chosen for the policy (Ragwitz and Steinhilber, 2014, 217). That is why the design elements of FIT policies have been discussed by various authors. Lesser and Su (2008), Held et al. (2010) and Couture and Gagnon (2010) qualitatively compare the FIT schemes employed in different countries to describe the different elements of FIT policy that can be chosen, such as renumeration, the duration of support and the degression of tariffs. The effects of design elements on a policy outcome have been empirically tested by for example Ritzenhofen and Spinler (2016), who analyse the impact of design elements on a firm’s investment decision and find that the more a FIT scheme is left to the market, the more investment is negatively affected. A second quantitative study focusing on the testing on design elements is the study by Jenner, Groba and Indvik (2013), who concluded that a poorly designed policy does not necessarily have a better effect on wind and solar energy production than not having a policy at all. The last study discussed here is the study by Devine, Farrell and Lee (2017), who present a simulative case study of Ireland to analyse the effects of policy design elements on market price risk sharing. They conclude that an efficient division of the market price between investors and consumers becomes of increasing importance when the deployment
12 or renewable energy grows. Therefore, policy makers should include flexible measures in FIT policies in order to adjust the price when necessary.
Another author who has focused on the impact of design elements is Del Río (Del Río, 2008; Del Río, 2012; Del Río and Mir-Artigues, 2012; Del Río et al., 2017). In his paper on the dynamic efficiency of FIT schemes (2012), he outlines a theoretical framework for the analysis of the effect of design elements of FIT policies on the dynamic efficiency of the policy. Thereby, he defines dynamic efficiency as the capacity of a FIT design element to promote continuous technological improvements and cost reductions for existing renewable energy technologies, enable an advancement of new technologies and support technologies with different maturity levels (Del Río, 2012, 139). More authors have discussed the dynamic efficiency of FIT policies in their work. Verbruggen and Lauber (2008) for example argue that dynamic efficiency enables cost reductions in the long run via innovation. Furthermore, Ragwitz and Steinhilber (2014) stress the importance of continuous technological improvements, as technologies which seem unsustainable today have the possibility to become highly useful in the future. Nevertheless, the framework presented by Del Río extends these discussions by offering an extensive explanation of the links between policy design elements and the dynamic efficiency of FIT policies, explained by five different dimensions of dynamic efficiency. A complete study testing the influence of design elements on the dynamic efficiency of FIT policies has not yet been performed. Therefore, this framework is chosen as theoretical basis of the regression analysis employed in this paper.
2.5. Del Río’s framework: independent variables
Del Río outlines eleven design elements of FIT schemes that influence the dynamic efficiency of the policies via five dimensions that are discussed later. These eleven design elements are chosen as the independent variables for this research and are explained more extensively below.
The first design element is the choice between a tariff and a premium. FIT schemes can grant a total amount of support to the electricity that is fed into the grid, which is called a tariff. It is also possible that a FIT scheme grants an additional amount above the electricity price to the producers of renewable energy sources, which is called a premium. With a tariff, the total support is known beforehand, which therefore is likely to reduce the risks of investment. With a premium, on the other hand, the support is mostly linked to the electricity price (Couture and Gagnon, 2010; Del Río, 2012; REN21, 2017).
13 The second design element discussed is whether the tariff or premium is technology specific. A technology specific FIT allows for an adaptation of support to the cost levels of different technologies. Therewith, a larger number of technologies can be supported. However, a technology specific FIT scheme could also increase the information asymmetry, as for each technology, the appropriate level of support needs to be decided (Couture and Gagnon, 2010; Del Río, 2012; Stokes, 2012; Davies and Allen, 2014).
Thirdly, support could be linked to the electricity prices. Thereby, the support follows the electricity market developments and does not distort the energy market. However, not linking support to the electricity price is expected to offer greater investment security, as support levels are known in advance (Couture and Gagnon, 2010; Del Río, 2012; Devine et al., 2017).
The fourth design element is the way in which the costs of a FIT scheme are financed. The costs can either be charged to the electricity consumers, or they can be paid by the government budget, by which they are consequently charged to taxpayers. When the costs are charged to consumers, the support for renewable energy production is not linked to changes in the government budget. A system in which the costs are paid for by taxes, on the other hand, allows the government to differentiate the budget for each technology (Couture and Gagnon, 2010; Del Río, 2012).
Fifthly, a degression rate can be installed in the FIT scheme, meaning that the support level for new plants decreases over time. Degression therewith is expected to enhance competition and stimulate cost reductions. Degression rates can either be set beforehand or can be flexible. Flexible degression rates enables policymakers to adjust the degression to market circumstances and therewith provide the opportunity to contain the costs of a FIT policy (Del Río, 2012; Pyrgou et al., 2016).
The sixth and seventh design element is the installation of either a cap or a floor price for renewable energy. Caps put a limit on the tariff and therewith ensure that the consumer costs are limited, which can positively affect the support for the policy. Floor prices on the other hand ensure a certain payment level for producers and consequently are expected to reduce the investment risks (Del Río, 2012; Devine et al., 2017).
14 Eighthly, policy makers can decide to reduce the support for an existing plant over time. Since the costs for installing a plant are high in the initial phase, but decrease over time, reduction over time allows for higher levels of support at the beginning phase of a new plan. Reduction therewith could especially allow support for immature technologies (Del Río, 2012; Ragwitz and Steinhilber, 2014).
The ninth design element is a limited plant size, meaning that only small-scale plants can receive support. A maximum plant size allows for participation from a greater variety of investors. Furthermore, it is likely to stimulate small-scale projects, which do not profit from economies of scale, to become more efficient. Nevertheless, as large projects do gain from scale effects, the costs for small schemes are generally expected to be higher and consequently, the costs of a FIT scheme with a maximum plant size are likely to be higher as well (Couture and Gagnon, 2010; Del Río, 2012).
The tenth design element discussed by Del Río is the possibility to install a maximum capacity per technology. Given the information asymmetry that exists between government and producers when setting a tariff or a premium, a capacity cap enables the government to keep the total costs under control. Nevertheless, a maximum capacity could hamper the stimulation for owners of existing plants to produce more efficiently, compared to for example a cap price level (Del Río, 2012; Frondel et al, 2010).
The last design element discussed by Del Río is the duration of the contract. The longer the duration of the contract, the more security is likely to be granted to producers and the lower the risks of investment become. On the other hand, a longer contract increases the burden on consumers for a longer period of time. Furthermore, a longer contract also could reduce the flexibility of the policy and complicate the alteration of a policy in case of unforeseen circumstances (Del Río, 2012; Jacobs et al., 2013).
2.6. Del Río’s framework: dependent variables
Besides eleven design elements, Del Río outlines five dimensions that contribute to the dynamic efficiency of a policy. Del Río combines the literature on learning effects and systems of innovation to outline these dimensions. The study of learning effects in renewable energy production studies the mechanisms that lead to costs reductions in the future. The literature on systems of innovation focuses on the interdependencies between technology, networks of
15 actors, and institutions, which together form a system of innovation (Del Río, 2012). By combining these two concepts, Del Río develops five dimensions of dynamic efficiency, namely technological diversity, private research, development and demonstration (RD&D), learning effects, technological competition and the total consumer costs for electricity. These five dimensions are chosen as the dependent variables for this research and are more extensively discussed below.
The first dimension of dynamic efficiency is technological diversity. As explained with the arguments in favour of FIT policies, FIT schemes have the ability to lower the risks for investment in new technologies. In order to induce continuous technological improvements of renewable energy technologies, it is important that a FIT policy supports a balanced number of technologies at different stages of maturity. In order to avoid neglecting immature technologies with a large cost saving potential, support for a variety of technologies is important. On the other hand, if the available funds are distributed among too many technologies, there is a risk that the individual producers receive too little funds to sufficiently build up new energy plants. Therefore, it is important that the design elements ensure a balanced variety of technologies (Markard and Truffer, 2008; Del Río, 2012).
The second dimension of dynamic efficiency is the support for private RD&D investments. In order to promote continuous cost reductions of renewable energy technologies, it is important that innovation is supported by not only public support, but also by private funds. Therefore, the more private investment is supported by the design of a FIT policy, the dynamically efficient a policy becomes (Del Río, 2012; Ragwitz and Steinhilber, 2014).
Thirdly, learning effects are important for the dynamic efficiency of FIT policies. These learning effects run through three different mechanisms. Firstly, producers learn by doing, meaning that the repetitious production of a product leads to efficiency improvements in the production process. Secondly, producers learn by using, as user experiences can be used to improve the product. Lastly, different actors in the field of renewable energy sources learn from networking with other actors in the field, which is called learning by interacting. Considering the learning effects of new technologies, FIT design elements supporting technological development and innovation for a longer period of time should result in better learning effects and consequently, a higher dynamic efficiency (Ek and Söderholm, 2010; Stokes, 2012; Ragwitz and Steinhilber, 2014; Del Río, 2012).
16 The fourth dimension of dynamic efficiency is technological competition. Competition between producers of renewable energy incentivises renewable energy suppliers to produce more efficiently, thereby lowering their production costs. Once again, a balanced tariff must be sought, high enough to attract investment, but low enough to stimulate competition. Consequently, FIT schemes that manage to stimulate competition between producers are more likely to induce future cost reductions than FIT schemes that do not (Lesser and Su, 2008; Del Río, 2012; Cowart and Neme, 2013).
The last dimension of dynamic efficiency as described by Del Río are the total consumer costs of the promoted renewable energy sources. As discussed earlier, dynamically efficient FIT policies promote a variety of technologies, in order to also support new technologies with large cost saving potential. Consequently, new technologies might be more expensive in the short run but become less expensive over time. Therefore, policymakers should find a balance between short and long-term costs in the design of a FIT scheme, in order to avoid overall inappropriate support (Del Río, 2012).
2.7. Hypotheses
In his paper on the effect of design elements, Del Río (2012) also describes the expected effects of the eleven design elements on the five dimensions of dynamic efficiency. These expectations are explained below and form the hypotheses of this research.
As it is expected that the investment risks are lower with a tariff than with a premium, Del Río argues that technological diversity is higher under a tariff system. Furthermore, the risks associated with a premium lead to the assumption that RD&D expenditure is to be higher with a tariff, and that learning effects are negatively affected by a premium. Regarding the fourth dimension, technological competition, it is hypothesised that a premium is more market-compatible, and that it consequently stimulates competition more. Lastly, it is unsure which of the two options leads to lower consumer costs, as the support levels under a premium are expected to be higher, however it is also expected that a tariff is a more effective form of support (Del Río, 2012).
Since technology specific FITs differentiate the support, a system with a technology specific FIT is expected to support a larger variety of technologies more effectively. Therefore, technology specific FITs are hypothesised to lead to more technological diversity and to better
17 learning effects. As technology specific FIT rates are expected to be better adapted to the actual costs of each technology, it is also hypothesised that RD&D investment is higher under a technology specific scheme. On the other hand, such a scheme reduces the technological competition between technologies, as each technology receives different support. Lastly, technology specific FITs are more likely to lead to cost reductions, if the support scheme is well adapted to the different costs associated with each technology (Couture and Gagnon, 2010; Del Río, 2012).
The third design element, linking the support to the electricity price, is hypothesised to lead to higher investment risks. Consequently, technological diversity becomes lower. Nevertheless, the overall support may be higher when support is linked, as electricity prices tend to rise. Therefore, it is uncertain whether RD&D, learning effects and technological competition are negatively affected by a linkage to the electricity price. The effect of linking on the total consumer costs is expected to be negative, as it is unpredictable whether the electricity price follows the renewable energy costs developments, which are likely to increase over time (Del Río, 2012).
Fourthly, it is expected that charging the costs of a FIT scheme to consumers lowers the investment risk, since it is less likely that the budget for FIT schemes is cut in periods of economic downturn. These lower risks consequently positively affect technological diversity, RD&D investments and learning effects. As cost reductions are not further stimulated in either of the schemes, it is hypothesised that the choice of whom to charge the costs of the FIT scheme to does not differently affect consumer costs. Furthermore, technological competition is also expected to be similarly stimulated in both schemes (Couture and Gagnon, 2010; Del Río, 2012).
The fifth design element, degression, is expected to incentivise technological efficiency and cost reductions, which consequently stimulates technological competition and leads to lower consumer costs. Degression therewith also stimulates learning effects. If degression rates are known beforehand, investors’ risks are lowered, which can stimulate RD&D investment. However, degression also reduces the producer surplus and can therefore also lower RD&D investments, so the effect on investment is uncertain. As degression does not stimulate the development of immature technologies, but also not discourages it, it is hypothesised that a
18 degression rate does not affect technological diversity differently than a FIT scheme without a degression (Del Río, 2012).
Cap prices discourage the deployment of more expensive technologies and are consequently expected to lower technological diversity and private RD&D expenditures. Subsequently, learning effects are also negatively affected by cap prices. On the other hand, cap prices enable policy makers to better control the costs of a FIT scheme, and therefore lead to lower consumer costs. Furthermore, cap prices stimulate producers to reduce costs, and therefore, more technological competition is expected (Del Río, 2012).
Floor prices, on the other hand, ensure a minimum payment level and therewith reduces investors’ risks, which leads to higher RD&D investments, more learning effects and more technological diversity. However, cost reductions and technological competition are expected to be negatively affected by a floor price, as a minimum level of support is already granted (Del Río, 2012).
The eighth design element, reduction over time, grants higher support to plants in the beginning of production and therewith promotes immature technologies. Technological diversity and learning effects are therefore expected to be higher in schemes with a reduction. As support is differently distributed, but the total amount remains the same, it is not expected that reduction stimulates cost reductions and competition, and therefore, no difference is expected between schemes with and without reduction over time. Furthermore, a reduction does not alter the risks of a FIT scheme and therefore, it is also hypothesised that RD&D investments are not extra stimulated (Del Río, 2012).
Ninthly, a maximum plant size is expected to lead to more competition and efficiency gains for smaller plants. However, this effect is likely to be offset by lower effectiveness for large plants. Therefore, it is uncertain whether a maximum plant size lowers consumer costs and leads to overall better learning effects. Since a maximum plant size limits the eligibility of new plants, technological diversity and technological competition are hypothesised to be negatively affected. Lastly, the effect on RD&D investments is uncertain, as the market for small plants is stimulated, but the market for large markets is not (Del Río, 2012).
19 A longer duration of support lowers the investment risks, which is consequently expected to stimulate RD&D investment, technological diversity and learning effects. However, it also stimulates the continuation of an old plant rather than building new ones, therewith negatively affecting technological competition. The effect on the consumption costs is uncertain, as annual support might be higher with a shorter duration and might therefore lead to higher total costs.
Lastly, a capacity limit increases investors’ risks, as it is uncertain whether all capacity of a plant can be deployed within the FIT scheme. Consequently, private RD&D investment and learning effects are negatively affected. Furthermore, a capacity limit stimulates the production of more efficient technologies, which therewith promotes technological competition. Moreover, a capacity limit enables the government to contain the costs of the scheme and therefore lowers consumer costs. The effect on technological diversity is uncertain, as it is hard to estimate what the technology mix would have been without the cap.
This section has discussed the literature on the effectiveness and the design elements of FIT schemes. Furthermore, it has outlined the eleven design elements and the five dimensions of dynamic efficiency as outlined by Del Río. Lastly, the expected relations between the design element and the dimensions have been presented as the hypotheses of this research. Consequently, the next section operationalises the dependent and independent variables.
20
3. Data and methodology
The previous section outlined the theoretical framework of Del Río and explained that there are five different dimensions contributing to the dynamic efficiency of a FIT policy. These five dimensions are the dependent variables for this research. The eleven different design elements as explained by Del Río are the independent variables chosen for this research. This section outlines the indicators that have been chosen to operationalise the independent and the dependent variables as outlined in the theoretical framework. Furthermore, this section explains why a sixth dependent variable is added outside the framework of Del Río. Lastly, this section presents the methodology employed in this paper to examine how different design elements of feed-in tariff policies affect the dynamic efficiency of the policy.
3.1. Independent variables
The eleven different design elements as outlined by Del Río and explained in section two are chosen as the independent variables for this research. In order to create the independent variables, the FIT policies of 110 countries as indicated in the report of REN21 (2017) have been analysed. Thereby, a large variety of sources has been used, such as national laws, data from the International Energy Agency’s (IEA) policies and measures database (n.d.), and qualitative analyses provided by other scholars. Consequently, not all 110 states have been listed in the database. Firstly, many of the dependent variables, which are discussed more extensively later, are only available on national level. Therefore, only countries with a FIT policy implemented on national level have been included in the database and countries with FIT schemes only on regional level have been excluded. Secondly, since the latest outcome variables are only provided for 2017, the effect of policies implemented after 2016 cannot yet be determined. Therefore, countries which implemented a FIT policy after 2016 have been excluded from the database. Lastly, the FIT policies of countries for which limited information was available and for which consequently not all the design elements could be distinguished have also been excluded from the database. Subsequently, an overview of the FIT policies of 57 countries has been created, which is the base sample of this research. This database, and the sources on which it has been based, can be found in Appendix A.
The design elements of the FIT policies of the countries in the base sample are measured as follows. Firstly, it is considered whether a FIT scheme grants a tariff or a premium to producers of renewable energy. Some countries combine both a premium amount and a tariff. Therefore,
21 two dummy variables are created to measure the first design element. The policy can either grant a tariff or not, and the policy can either grant a premium or not. Secondly, it is analysed whether the support level differs per renewable energy technology or not, which also constitutes a dummy variable. Thirdly, it is determined whether the support is linked to the electricity price or not, therewith creating another dummy variable. For the fourth element, charging the support to either taxpayers or consumers, no countries have been found which combine charging the support to both taxpayers and consumers at the same time. Therefore, the fourth independent variable is also measured by a dummy variable, constituting 0 if the support is charged to the taxpayers and 1 if the support is charged to consumers. The fifth element discussed is degression. Degression can take multiple forms and degression rates can range from high and fast, to low and gradual. Nevertheless, the purpose of this research is to analyse the effect of the eleven design elements outlined by Del Río. Consequently, it falls outside the scope of this paper to analyse the effect of a variety of degression designs. Therefore, degression is also measured by a dummy variable, constituting 1 in case of degression. The sixth and seventh design element, a cap or a floor price, are also measured with dummy variables, whereby a 0 is given for policies without a cap or a floor, and a 1 for policies which do set a cap or a floor on the support level. Eighthly, considering the explanation given for the measurement of degression, reduction for existing plants is also measured by a dummy variable. For the ninth design element, maximum plant size, it is considered whether plants below a certain maximum capacity can receive FITs, or whether all plants are eligible in a FIT scheme. Countries which do install a maximum capacity limit for a plant are given the value 1, countries which do not are measured by a 0.
The tenth design element, duration of support, is measured numerically. As some countries differentiate the duration of support per technology, the maximum contract duration in number of years is chosen as the value for this independent variable. Lastly, it is examined whether policies place a capacity limit on the total level of energy produced by renewable energy sources. Policies which do impose such a capacity limit are given the value 1, while policies which do not set a limit are valued with a 0. The table below provides a summary of the eleven design elements and the way these are measured. All the independent variables can be found in table 10 in Appendix A. The year of installation of a FIT scheme has also been added to table 10. This is not a design element, but it is used to ensure that countries which only implemented a FIT policy after the date of a certain outcome variable are excluded from the analysis on that outcome variable. This is more extensively explained in the discussion on methodology.
22 Table 1: Eleven design elements of FIT policies (Del Río, 2012)
Design element Description How is this element measured?
1. Fixed-premium vs. fixed-tariff
Tariff: a fixed amount of support is granted
Premium: an additional amount above the electricity price is granted
Two dummy variables
2. Technology-specific FITs
The level of support differs per technology to reflect technology-specific production costs Dummy variable 3. Support is linked to electricity price
Support is either linked to the electricity price or not
Dummy variable
4. Support paid by electricity consumers or by taxpayers
Support can be charged to either consumers or taxpayers
Dummy variable
5. Degression The support level is reduced over time (for new plants)
Dummy variable
6. Cap price Support is capped at a maximum price level
Dummy variable
7. Floor price There is a minimum level of support
Dummy variable
8. Reduction of support for existing plants
Reductions of support over time for existing plants
Dummy variable
9. Maximum size of plants
Only installations below a certain capacity maximum are eligible for support
Dummy variable
10. Duration of support The maximum duration of a FIT contract
Numerical
11. Capacity limit per technology
The level of generation from renewable energy is capped (this can be overall support or support for individual technologies)
Dummy variable
3.2. Dependent variables
The framework provided by Del Río does not outline indicators by which the five dimensions of dynamic efficiency can be measured. Therefore, proxies to measure the dimensions have been sought in existing databases, and the chosen indicators are outlined below. Unfortunately, no dataset has been found which combines the five elements described by Del Río and assigns a total dynamic efficient coefficient to every country. It falls outside the scope of this paper to develop a method to calculate a total dynamic efficient coefficient. Furthermore, the data availability on the five elements of dynamic efficiency is not always sufficient and consequently, the outcome variable data are not available for every country from the base sample. Therefore, a sixth variable, outside the framework of Del Río, has been added to the
23 outcome variables, to measure the total impact of design elements on the development of renewable energy. This variable is also explained more extensively below.
The first dimension of dynamic efficiency is the technological diversity of renewable energy sources, since it should be avoided that immature technologies with a large cost saving potential are not supported. As technological diversity in itself is hard to measure, many authors have used patent data to measure innovation activities on either firm level or state level (among others Verdolini and Galeotti, 2011; Noailly and Smeets, 2015; Kruse and Wetzel, 2016; Lindman and Söderholm, 2016). Following Haščič and Migotto’s working paper (2015) on using patent data in measuring technological activities, the indicator diffusion of climate change mitigation technologies related to energy generation, transmission or distribution from the Organisation for Economic Cooperation and Development’s patents in environment-related technologies database is used to measure technological diversity (OECD, 2018), as technological diffusion is considered the closest estimate for technological diversity. This database provides data for 30 countries from the base sample.
The second dimension of dynamic efficiency is the support for private research, development and demonstration, which complements public technological innovation support in the long run. Unfortunately, as is also stated by the IEA “information on research and development expenditures per technology area is scarce, especially for private sector spending” (IEA, 2017). A database with absolute numbers for private RD&D investment per country has therefore not been found. Therefore, a secondary source is selected as a proxy for private RD&D investment, namely the EY renewable energy country attractiveness index (EY, 2017). This index combines indicators such as the investment climate and the availability of funds to assign an attractiveness score to each country, and consequently publishes the 40 most attractive renewable energy investment countries. Out of the top 40, 24 countries currently employ a FIT scheme and therefore, this database contains data for 24 countries from the base sample.
Thirdly, learning effects are important for the dynamic efficiency of FIT policies. These learning effects run through three different mechanisms, namely learning by doing, learning by using, and learning by interacting. Considering the learning effects of new technologies, FIT policies supporting technological development for a longer period of time should lead to dynamic efficiency. Various authors have tried to quantify learning effects (Chen et al, 2015; Huenteler, 2016; Dismukes and Upton Jr., 2015). Nevertheless, an extensive database covering
24 learning curves multiple countries does not exist. Therefore, a proxy is used to measure learning effects. Del Río argues that dynamic efficiency is mainly caused by learning by interacting. When learning by interacting is poor, innovation is prevented. Consequently, innovation activity is used to measure a state’s learning effects. Therefore, another indicator from the OECD’s patents in environment-related technologies database used. Innovation activity is measured by the indicator percentage of climate change mitigation technologies related to energy generation, transmission or distribution (Del Río, 2012; Haščič and Migotto, 2015; OECD, 2018). This database contains data for 40 countries.
The fourth dimension of dynamic efficiency is technological competition, because competition incentivises renewable energy suppliers to produce more efficiently. Commissioned by the Agency for the Cooperation of Energy Regulators (ACER), consultancy firm IPA Advisory published a report on the competitiveness of the energy markets in 29 European countries (IPA Advisory, 2015). The report combines ten indicators, such as the annual entry and the number of energy suppliers, to create a composite indicator by which the energy markets’ competitiveness is measured. The values of the composite indicator as calculated by the report are therefore chosen to measure competition in the energy sector. Consequently, values for the fourth dimension are provided for 24 countries of the base sample.
The last element of dynamic efficiency are the total consumer costs of the promoted renewable energy sources. As discussed earlier, dynamically efficient FIT policies promote a variety of technologies, in order to also support new technologies with large cost saving potential. Nevertheless, policymakers should find a balance between short and long-term costs, in order to avoid inappropriate support. To measure the total consumer costs, the levelized costs of electricity (LCOE) for wind power is used. The LCOE is a measurement tool used to calculate the unit costs of a technology, taking into account the lifespan of that technology (Bloomberg New Energy Finance (BNEF), 2016). Multiple organisations have calculated the LCOE of various technologies in various countries, but from the calculations that have been found, the report from the OECD, IEA and the Nuclear Energy Association (NEA) (2015) provides the most extensive country-based calculation.1 Therefore, the LCOE for onshore wind energy in various countries as calculated in this report are taken as the independent variable to establish a first indication of the effect of design elements on the costs of energy. This report provides
1 For more calculation of LCOEs, see for example BNEF, 2016, International Renewable Energy Association (IRENA), 2017 and Ram, M, Child, M., Aghahosseini, A., Bogdanov, D. and Poleva, A., 2017.
25 data for 13 countries of the base sample. The data on the five dependent variables can be found in table 11 in Appendix B.
To estimate the effect of the eleven design elements on a total outcome score, a sixth outcome variable is added to the analysis, which falls outside the framework of Del Río. The indicator that has been chosen is the renewable energy as percentage of total energy consumption provided by the World Bank. By adding a sixth outcome variable to the analysis, it is possible to not only estimate the effect of FIT design elements on the five dimensions of dynamic efficiency, but also to estimate the effect of design elements on the total deployment of renewable energy in a country. Furthermore, the World Bank database contains data for all 57 countries in the base sample, which therewith enables a better estimation of the effects of design elements (World Bank, n.d.e.). The data for the renewable energy share can be found in table 12 in Appendix C.
3.3. Control variables
Besides the design elements outlined by Del Río, there are several other factors which could influence the six outcome variables. Therefore, various control variables are added to the analysis. For the first outcome variable, technological diversity, research by Bointer (2014) has shown that there is a positive relation between gross domestic product (GDP) and the accumulation of new technologies via innovation. That is why GDP is taken as control variable for the analysis on technological diversity. GDP data for the countries is the base sample is taken from the World Bank database (World Bank, n.d.b.). The data is measured in international dollars, thereby using the purchasing power parity (PPP) to ensure that the differences in price level between countries are taken into account and do not have to be controlled for separately.
Besides the links between GDP and innovation, the research by Bointer (2014) also examines the link between GDP and research and development (R&D) expenditures, and also identifies a positive effect of GDP on the R&D spending. Therefore, GDP per capita is also added as control variable to the analysis of the second outcome variable. Moreover, Bointer also finds a difference in R&D activities and consequent innovation between small and large countries. Therefore, to control for country and market size, data on both land area surface and energy consumption per capita are added to the analysis. Since the energy consumption per capita does not give sufficient information on a country’s market size, data on the population size are also added as control variable. The data on land area surface, population and energy consumption
26 are all taken from the World Bank database (World Bank, n.d.a.; World Bank, n.d.b.; World Bank, n.d.c.; World Bank, n.d.d.). Furthermore, following the research design of Dong (2012), other policies that might affect the accumulation of RD&D funds are also considered. Consequently, a dummy variable is added to account for countries that offer tax credits for investment in renewable energy. This dummy indicates 1 if the country does offer tax credits and 0 if the country does not. The data for this control variable are taken from the REN21 report (2017).
For the third outcome variable, learning effects measured by focusing on innovation, the control variables are based on the researches by Bointer (2014) and Brutschin and Fleig (2016), who identify the size and the capacity of an economy as factors positively affecting the innovation activities of a country. To measure the size and the capacity of the economy, three control variables are added. Once again, GDP per capita is added to control for the economic performance of a country. Furthermore, to measure the size of the energy market, the energy consumption per capita and the population size are also added as control variables to this analysis.
The fourth outcome variable is competitiveness, for which the researches by Podobnik, Horvatic, Kenett and Stanley (2012) and Kordalska and Olczyk (2016) are followed. Both articles establish a relation between GDP and the competitiveness of a country, that is why GDP per capita is also taken as the first control variable for the fourth regressions. Besides these two researches, the World Economic Forum’s (WEF) Global Competitiveness Index (GCI) stresses market size as one of the components of the GCI (Schwab and Sala i Martin, 2017). To measure market size, three control variables are added, which are the same as for the second outcome variable, namely energy consumption per capita, the size of the population and the land area surface of a country. These data are once again taken from the World Bank database (World Bank, n.d.a.; World Bank, n.d.b.; World Bank, n.d.c.; World Bank, n.d.d.,).
The report on the costs of renewable energy of IRENA (2017) stresses the importance of economies of scale in lowering the costs for renewable energy production. To control for the effect of economies of scale, both land surface and GDP per capita are taken as control variables to measure both the size of the country and the size of the economy. Additionally, as the all the GDP per capita data is conferred to international dollars taking into account the PPP, differences in price levels per countries are simultaneously controlled for.
27 For the last outcome variable, the share of renewable energy in the total energy consumption, four control variables are added. Firstly, various authors have discussed the effect of income on renewable energy employment, as it is expected that higher income countries can better afford the costs of renewable energy technologies and are more able to economically incentivise the deployment of renewable energy (Kilinc-Ata, 2016). The results of these researches are mixed. Authors like Shrimal and Kneifel (2011) and Dong (2012) conclude that income does not have an effect on renewable energy deployment, whereas Jenner et al. (2013) and Aguirre and Ibikunle (2014) do find a positive effect of income of renewable energy deployment. Therefore, to control for a possible effect of income, GDP per capita is added. Moreover, authors have discussed the effect of country characteristics on the potential for renewable energy production, and consequently for the renewable energy share (Shrimal and Kneifel, 2011; Jenner et al, 2013). To control for country characteristics, once again land area surface and population size are added as control variables. Lastly, according to Dong (2012) the energy demand of a country also influences the renewable energy deployment. To account for the energy demand of a country, energy consumption per capita is also added as a control variable to the sixth analysis. The data for the control variables can be found in table 13 in Appendix D.
3.4. Hypotheses
The hypotheses for the five dimensions of dynamic efficiency have been discussed in the previous section. As the sixth outcome variable, the renewable energy share, is added to the analysis and is not taken from Del Río’s theory, the expected effects of the design elements on the renewable energy share are not specifically hypothesised by Del Río. Nevertheless, his theory is used to further hypothesise the effect of the different design element on the sixth outcome variable. Thereby, the focus is mainly placed on the hypothesised effect of the design element on the investment risks, as it is expected that a lower investment risks leads to more investment in renewable energy and consequently increases the renewable energy share (Del Río, 2012).
Firstly, as investment risks are expected to be lower with a tariff, the first hypothesis is that a tariff leads to a higher renewable energy share than a premium. Secondly, creating a technology specific FIT scheme is mainly expected to increase technological diversity. It does not, however, change the investment risk and therefore, it is expected that a technology specific FIT scheme does not affect the renewable energy share differently than FIT scheme without a technology specific scheme. Thirdly, since linking the support to the electricity price increases
28 investment risks, it is expected that the renewable energy share is higher in countries where the support is not linked. The fourth design element, charging the costs to consumers, is expected to lower investment risks. Consequently, the fourth hypothesis is that charging the costs to consumers leads to a higher renewable energy share. Fifthly, since degression is expected to lower investment risks, the fifth hypothesis is that degression positively affects the renewable energy share. Both a cap and a floor price lower investment risks and therefore it is hypothesised that they both have a positive effect on the renewable energy share. Eighthly, a reduction over time mainly changes the distribution of support, and therewith does not alter the risks associated with investing in renewable energy. Consequently, the hypothesis is that a reduction over time for existing plants does not affect the renewable energy share differently than a FIT scheme without a reduction over time. The same goes for the effect of a maximum plant size, which is also not expected to lead to lower or higher investment risks. Therefore, the ninth hypothesis is that a maximum plant size does not have a different effect on renewable energy share than a FIT scheme without a maximum plant size. The tenth design element, the duration of support, does lower the investment risks, whereby a longer support leads to lower investment risks. Subsequently, it is expected that the duration of the contract positively affects the renewable energy share. Lastly, a capacity limit increases the investment risks, which consequently leads to the hypothesis that a capacity limit negatively affects the renewable energy share. To create a clear overview of the hypotheses, table 2 summarises the hypothesised effects of the design elements on all six outcome variables (Del Río, 2012).
Table 2: Hypotheses Design element Technological diversity RD&D investment Learning effects Technological competition Total consumer costs Renewable energy share Tariff Positive Positive Positive Negative Uncertain Positive
Premium Negative Negative Negative Positive Uncertain Negative
Technology specific
Positive Positive Positive Negative Negative No effect
Linkage to electricity price
Negative Uncertain Uncertain Uncertain Positive Negative
Costs to taxpayers
Negative Negative Negative No effect No effect Negative
Degression No effect Uncertain Positive Positive Negative Positive
Cap price Negative Negative Negative Positive Negative Positive
Floor price Positive Positive Positive Negative Positive Positive
Reduction Positive No effect Positive No effect No effect No effect
Maximum plant size
29
Duration Positive Positive Positive Negative Uncertain Positive
Capacity limit
Uncertain Negative Negative Positive Negative Negative
3.5.Methodology
To estimate the effect of the eleven design elements on the six outcome variables, this paper employs an ordinary least squares (OLS) regression analysis. Thereby, the following regression is built for all six outcome variables:
Yit = α + β DEi + γ Zit + ε
Hereby Y represents the six different outcome variables for every country i in year t and α is the intercept. Furthermore, DE stands for design elements and includes the dummy and numerical variables discussed previously. As the policy design elements discussed in this paper are implemented for a longer period of time, these variables are time-invariant. Policies which have been stopped or drastically altered have consequently not been included in the independent variable database. Z represents the control variables for country i in year t, and ε is a random error term which might contain individual country effects. Lastly, β stands for the coefficient of the design variables, while γ represents the coefficient of the control variables.
Given the relatively small sample sizes of the six different outcome variables, a generalisation of the findings of the regressions is limited. A panel data set with data from multiple years for the same group of countries would of course increase the number of observations. However, many countries in the base sample only started to employ a FIT scheme in the 2010s. Creating a panel data set would consequently also lead to a lower number of cases to be observed. As the number of cases already differs per outcome variable, it would also not be possible to employ the same methodology for each outcome variable. That is why it is chosen to regress the data from one year, and not to instead analyse panel data. The design elements analysed in this paper are time-invariant, which would make a fixed effect panel data analysis impossible. As there is still much discussion in the literature about possible other explanatory variables that could influence the dynamic efficiency of FIT policies, the conditions for a random effects panel data analysis are not completely met, which constitutes the second reason for focusing on
30 data from one year.2 Even though a smaller sample size limits the generalisation of the results, it is considered important to offer a first comprehensive test of the impact of design elements, following the framework by Del Río. That is why this current study employs six single year regressions to estimate the effect of FIT policy design elements.
As explained before, the information on the year of installation of the FIT policy determines which countries are included in each regression. Hereby, the year of installation is lagged and only countries which implemented a FIT scheme in the year before the year from which the outcome variable data are drawn are included in the respective regression. For the six regressions, all data for the control variables are drawn from one year, whereby in most regressions, no lagged variables are used. Unfortunately, the availability of the data does not allow for the same method for all six regressions. Firstly, the database on energy consumption only contains data until 2014. After careful literature review, it is considered important to include energy consumption as a control variable in the regressions as outlined earlier. Given the relatively small variance in energy consumption data in the years before 2014, it is chosen to include the data for the year 2014 for regressions which are based on data from later years (World Bank, n.d.a.). It is not chosen to instead use outcome variables from earlier years, as this would significantly lower the number of countries in the base sample and therefore lower the explanatory power of the model. Secondly, the database on population size and GDP per capita only contains data until 2016 and that is why for the second regression, RD&D investment, lagged variables are used (Word Bank, n.d.b.; World Bank, n.d.d.). Lastly, as land area surface only changes marginally per year, the land area of 2014 is used in every regression which includes land area as control variable (World Bank, n.d.c.).
Following Jenner et al. (2013), Costa-Campi, Duch-Brown and García-Quevedo (2014) and Corsatea, Giaccaria and Arántegui (2014), the natural logarithm is taken from the data on population, GDP, land size, energy consumption and the LCOE of onshore wind energy, in order to reduce the influence of outliers in the dataset.
2 For a discussion of control variables in the analysis of FIT policies, see for example Shrimal and Kneifel (2011), Dong (2012), Jenner et al. (2013), and Kilinc-Ata
