Energy Transition and Directed Technical Change

Asier Alvarez Galdos

S3724069 | a.alvarez.galdos@student.rug.nl SUPERVISOR: T.M. HARCHAOUI, PHD

CO-ASSESOR: M.V. NIKOLOVA, PHD

Energy Transition and Directed

Technical Change

ENVIRONMENTAL POLICY ANALYSIS OF EUROPEAN

COUNTRIES

Master Thesis

University of Groningen

Faculty of Economics and Business

Energy Transition and Directed Technical Change:

Environmental Policy Analysis of European countries

Asier Álvarez Galdós

*

Abstract

Using a sample of 25 European countries over the 1998-2016 period, this paper examines the effect of directed technical change and environmental policies on the energy transition. Our work contributes to the literature along two dimensions: First, we ascertain the effectiveness of different instruments for the diffusion of renewable energy sources. Second, we provide an empirical test to the literature on direct technical change. Using a variety of estimation methods, our results highlight that: both price and market size play a relevant role in determining the direction of technical change. In particular, an increase in fossil fuel prices positively affects the energy transition while an increase in the market size negatively affects the energy transition. This suggests that both clean technologies (solar, wind, etc.) and dirty technologies are close substitutes, making a policy in favor of temporary subsidies preferable to direct innovation towards renewables.

Keywords: Energy Transition | Directed Technical change | Environmental Policies

Contents

1. Introduction ... 4

2. Literature Review ... 6

2.1. Defining Energy Transition ... 6

2.2. Directed Technical Change ... 6

2.3. Market Externalities ... 7

2.4. Policy Intervention ... 9

2.5. Energy Transition and Directed Technical Change ... 10

3. Theory and Model ... 11

3.1. Empirical Method ... 11

3.2. Variables ... 12

3.3. Hypothesis and related work in the literature ... 15

3.4. Period and sample of countries ... 18

4. Data and methodology ... 19

4.1. Data ... 19

4.2. Methodology ... 22

4.3. Statistical analysis of data ... 23

4.4. Model fit ... 24

5. Econometric Analysis and Results ... 26

6. Conclusions ... 29

References ... 31

Annex ... 35

Annex 1: Renewable energy shares at the EU level an in individual Member States ... 35

Annex 2 Relative Supply of College Skills and College Premium ... 36

Annex 3 Number of EPO patent application in renewable technologies ... 37

Annex 4 White Test... 37

Annex 5 Variance Inflation Factor and tolerance values ... 38

Annex 6 Wooldrige Test ... 38

Annex 7 Aurellano-Bond Test ... 38

Tables and Figures

Figure 1 The Energy Transition Process ... 11

Figure 2 Share of Renewable Energy Supply ... 19

Figure 3 Crude Import Prices (USD/bar) ... 20

Figure 4 Distribution of ET and Logged ET ... 23

Table 1 Expected signs of the explanatory variables ... 18

Table 2 Environmental polices (nº instruments) ... 21

Table 3 Descriptive Statistics ... 22

Table 4 Correlation Matrix ... 24

Table 5 Estimated Coefficients of Lagged (and Logged) ET... 25

1. Introduction

Since the end of the 20th century there has been an increasing amount of reports raising the social awareness on climate change. Consequently, increasing governments commitment towards promoting the energy transition (ET) and thus, creating an appropriate policy windows for cross-country cooperation (Kyoto 1998; Paris 2016). However, implemented policies are not reaching the internationally agreed temperature targets (Bretschger, 2017) and the penetration of renewable energy sources although increasing remains limited [Annex 1]. Thereby, reducing the growing consumption of fossil fuels has become one of the most pressing political challenges of today as exceeding the 1.5ºC target by 2030 could have irreversible effects over global temperatures (Hoegh-Guldberg et al. 2018).

Renewable technologies, such as solar, wind, and renewable combustibles, can provide a clean alternative to energy generation (Noailly and Smeets, 2015). Although, in the cases where the marginal cost of implementing these clean technologies, exceed their marginal benefit, that substitution would not happen as it would not be economically viable (Popp et al. 2009). That is why technical change plays an essential role promoting the ET, as it allows replacing fossil fuels at a lower cost.

However, as showed by Acemoglu et al. (2012) technical change is skewed towards dirty technology due to the exitance of market failures Indeed, his theory on directed technical change suggest that two main forces (price effect, market size effect) determine the direction of the innovation, away from clean technologies (solar, wind etc.).Thus, in the absence of policy intervention, to direct technological innovation away from dirty technologies, clean technologies would not be feasible to replace fossil fuels, ultimately leading us to an “environmental disaster”. Moreover, market failures also limit the diffusion of clean technologies making those innovations of little use for the society, hampering the energy transition (Popp et al. 2009). That is why environmental regulation play an essential role not only in directing the technical change towards renewables but also promoting the diffusion of those innovations.

This paper focuses on the role of different environmental policies2 in promoting the ET in European countrieswhile considering the forces that distort the technical change (towards clean technologies). By doing so we seek to sort out the relative importance of environmental policies and directed technical change in the transition towards clean energy. This question is motivated by the fact that, although both directed technical change and environmental policies are key drivers of the energy transition, their interplay has not been thoroughly explored.

As our literature review argues, the importance of either directed technical change (Acemoglu, 2002; Noailly, 2015; Acemoglu et al. 2012; Greaker et al. 2018; Grimaud and

Rouge, 2008) as a determinant of ET is considered independently from policy interventions (Stokes et al. 2018; et al. 2010). On one hand, analyzing the factors that determine the direction of the technical change (price, market size) allows us to understand the type of policy intervention3 needed to promote the transition towards renewables. On the other hand, environmental policies also play an essential role in the implementation of the energy transition, alleviating the existing barriers imposed by market externalities. In other words, environmental policies respond to the type of instruments that are needed to accelerate the energy transition while the directed technical change provides the necessary guidance. Our work contributes to the literature along two dimensions: First, we ascertain the effectiveness of different instruments for the diffusion of renewable energy sources. Second, we provide an empirical test to the literature on the directed of the technical change. While this literature has contributed to advance our understanding of the theoretical underpinnings on the role of technology in the shift towards clean energy, it has not been thoroughly tested in the statistical sense, and still has to make a dent in the empirical literature.

Our econometric analysis uses a variety of estimation techniques applicable to panel data to examine the factors that affect the share of renewable energy supply (ET) for European countries over the 1998-2016 period. To do so we would use IEA World Energy Balances

2018 database to construct the energy transition indicator (share of renewable supply) and

total energy consumption (ktoe) to create the market size indicator. Furthermore, we would use OECD Crude oil import prices indicator 2020 data to construct the price indicator and

EEA database on climate change mitigation policies and measure in Europe to create a

vector of environmental policies (regulatory, soft, market-based, fiscal, framework).

Our results show that (1) both price and market size play a relevant role in determining the direction of technical change and (2) an increase in fossil fuel prices positively affects the energy transition while an increase in the market size negatively affects the energy transition. Thereby we might infer that both clean technologies (solar, wind, etc.) and dirty technologies are close substitutes and thus a temporary subsidy would be enough to direct innovation towards renewables. (3) increasing the number of instruments does not always increase the share of renewable energy supplied (4) market-based instruments do not seem to be more effective at promoting the energy transition than other instruments.

The study is structured as follows. Section 2 defines the concept of energy transition and provides a review on the main literature of the topics covered so far (directed technical change, market externalities, policy intervention). Section 3 defines the variables and

introduces the hypothesis. Section 4 presents the methodology and explains the data. Section 5 presents and discusses the results. Section 6 concludes the analysis.

2. Literature Review

2.1. Defining Energy Transition

Despite its importance and its increasing relevance on the policy agenda, the concept of energy transition is still elusive. The international renewable energy agency (IRENA, 2020) refers to it as “a pathway toward the transformation of the global energy sector from fossil-based to zero-carbon” (IRENA, 2020). O´Connor (2010) provides a much broader definition “a particularly significant set of changes to the patterns of energy use in a society” but also focuses on the transformation process “can affect any step in this chain, and will often affect multiple steps”.

According to Sovacool (2017), energy transition refers to “the time that elapses between the introduction of a new primary energy source (energy forms used to generate energy supply), and its rise to claiming a substantial share of the overall energy market” (Sovacool, 2017). Despite not having a standard or commonly accepted definition there is a common theme that prevails, the involvement of a change in the energy system to a distinct source or technology. This is precisely the focus of this study, to measure the effects of environmental policies on the energy system change to renewable technologies. Thereby, in order to proxy the ET, we would use the renewables share over the overall energy supply as our indicator.

2.2. Directed Technical Change

The role of technical change has always received much of the attention within the field of environmental economics (Jaffe et al. 2003). First, because of its capacity to influence environmental impact, especially in the long run. The directed technical change determines which technologies are being developed and thus it rather facilitates or mitigates pollution. Second, due to its ability to reduce the cost of the transition towards clean energy sources, as this might have quantitatively important consequences in the cost-benefit analysis of policy intervention (Popp et al. 2009).

In this context, Acemoglu (2002) developed his theory on directed technical change, which suggests technical change is not neutral, it tends to benefit some factors more than others. To illustrate this, Acemoglu compares the relative supply of skills with their return [Annex 2] showing that college returns kept increasing despite the large increase of college supply. Thereby suggesting that the technologies developed in this period have been skilled biased, that is that they have benefited more skilled workers than unskilled workers.

hand, market size effect drives technical change towards the abundant factor (natural resources) to take advantage of the scale effects.

Furthermore, Acemoglu (2002) indicates that the intensity of these two forces would primarily be determined by the elasticity of substitution between sectors. For example, in the case suggested by Acemoglu et al. (2012) where we have two sectors: clean and dirty. We might expect that when the elasticity of substitution between the two sectors weak (e<1), price effect (of exhaustible resources) should be relatively big as clean technologies would not be able to replace dirty technologies and thus their price will continue rising due to their lack of replacement. Otherwise, when the substitution is strong (e≥1) market size effect should be relatively big as the increasing returns to scale of dirty technologies would overcome the usual substitution effect4 and thus increase the relative reward for the abundant factor.

Ultimately, Acemoglu et al. (2012) suggests that taking into account the direction of the technical change would not only explain why some sectors are more developed than others but also provides useful information to design an appropriate policy intervention. As when both sectors are weak substitutes (e<1), that is when the intensity of the price effect is higher than the market size effect, clean technologies are not able to replace dirty technologies. Therefore, we might need to permanently subsidize clean energy to allow it to compete against dirty energy (Stern 2009) as otherwise dirty technologies would grow at the same rate as clean technologies (Acemoglu et al. 2012, Equation 19).Instead, when both sectors are strong substitutes(e≥1), temporary policies would be enough to direct innovation incentives towards clean technologies.

More recently, Noailly and Smeets (2015) provide an empirical analysis of the direct technical change by looking at the two main forces (price, market size) that affect the rate of innovation in the field of electricity generation. They show that increases in fossil fuel market prices and in fossil fuel market size all widen the gap between clean and dirty technologies. Building on Acemoglu (2002) ´s framework Grimaud and Rouge (2008) study the effects of different economic policies in an endogenous growth general equilibrium framework. Their results show that the technical change is biased towards the dirty sector (fossil fuels) and that environmental policies allow us to redirect the research efforts towards the green sector (renewables).

2.3. Market Externalities

According to Jaffe et al. (2005) market externalities are “an economically significant effect of an activity, the consequences of which are borne (at least in part) by a party or parties other than the party that controls the externality producing activity”. Following this

definition, several authors (Jaffe et al. 2005, Grimaud and Rouge 2008, Popp et al. 2009, Acemoglu et al. 2012) stress the importance of taking into account market externalities in order to correct the related distortions. For example, when the costs of pollution are not internalized, firms would lack the incentive to reduce their emissions ultimately transferring that cost to society, therefore suggesting that we might need to internalize those cost in order to correct that market failure.

Jaffe et al. (2005), as well as other authors (Grimaud and Rouge 2008, Popp et al. 2009), identify the exitance of three main externalities related with technological development: (a)

Knowledge externality occurs when firms have little incentives to innovate due to their

inability to capture all the benefits derived from it. That is, when the profits of implementing a new technology not only benefit the one implementing it but also other firms. (b)

Incomplete information occurs when the firms that have the opportunity to assess the

potential of clean technologies are the ones that have lower incentives at developing them. On the contrary, firms that have the incentive to invest in clean technologies do not have enough information and thus have many uncertainties regarding the returns on R&D investment. (c) Adoption externality occurs when the fixed cost associated with the innovation are relatively high and thus, they need economies of scale to profit from it. The problem here as suggested by Jaffe is that the diffusion of new technologies tend to be gradual and thus it might take time until they can cover the fixed cost.

Moreover, Popp et al. (2009) highlights that the dynamic increasing returns to adoption represent a big impediment for the development of clean technologies. In other words, due to the scaling effect, new entrants (clean sector) would find a hard time entering the market as incumbents (dirty sector) already benefit from the scale effects which allows them to produce energy at a much lower average cost. Thereby, Popp et al. (2009) analyze the time lag between invention and the diffusion of clean technologies, showing that the adoption rate tends to be relatively slow even in cases when the upfront cost and the payback period are relatively small.

2.4. Policy Intervention

Building on the market externality literature, Acemoglu et al. (2012) indicate that policy intervention is indeed necessary as under laissez faire (no regulation) energy markets would direct resources towards the dirty sector, fostering environmental degradation and ultimately causing an “environmental disaster”. Thus, Acemoglu et al. (2012) suggests using carbon taxes and research subsidies in order to redirect technical change towards renewables in an effort to avoid an environmental disaster. Their results show that when both sectors (dirty and clean) are weak substitutes and that permanent intervention would be needed to correct for the market externalities, as clean technologies would not be able to replace fossil fuels without permanent subsidies. Furthermore, when both sectors are strong substitutes, temporary policies would be enough to overcome the scale effect and thus to replace dirty technologies. Moreover, they also showed that delaying this intervention would have significant cost to the society as this would widen the technological gap between clean and dirty sector and thus extend the transition period. This intuition is based on the assumption that during the transition period countries experience a slow growth and, therefore, a longer transition implies longer periods of slow growths.

More recently, Stokes et al. (2018) highlighted the role of political intervention as the most important barrier to the energy transition. Their intuition is that technical change is biased towards certain factors not only because market externalities but also due to ineffective policy response. They suggest that by analyzing the interaction between politics and environmental regulation they would be able to identify policymaking patterns and thus understand which factors spur and detract the development of these policies. Their results showed that, environmental agreements (Kyoto, Paris) favor the development of environmental policies as they provide windows of opportunity and political cover for policymakers.

From an empirical perspective, De Vries and Withagen (2005) analyzed the effects of environmental regulation over the patenting application regarding SO2. Their results showed that strict environmental policies lead to more innovation although they recognized that using dummy variables to model environmental policies does not fully reflect policy variability. Later on, Johnstone et al. (2010) also pointed at environmental policies as the main driver to correct the distortions generated by market failures in innovation. Although their focus was more into policy instruments and not that much into political process.

would find that policies are more effective at promoting wind technologies than other types of technologies.

2.5. Energy Transition and Directed Technical Change

As noted above, most of the previous literature focuses primarily on the role played by technological innovation, in promoting the energy transition. Most of those, consider that the existence of market externalities distorts that technical change towards dirty technologies and even delays its diffusion. While most of those authors highlight the need for policy intervention in order to direct the technical change towards renewables, we find it surprising that only Johnstone et al. (2010) includes environmental policies in their research. Consequently, we would like to contribute to the literature by including both directed technical change and policy intervention to the analysis of the energy transition. By doing so, our research would be able to capture not only the direction of the technical change5 but also the policy intervention needed to overcome those externalities and thus prevent an environmental disaster.

Furthermore, the bulk of the literature primarily stresses the importance of innovation in the process of replacing dirty energy sources. To do so, most of them use patent application, as their response is variable, to proxy for technological innovation. Although, if we look at the literature on market externalities, we will understand that increasing the number of patents does not necessarily imply the diffusion of those technologies. For that reason and others6 Johnstone et al. (2010) indicate that patent application would imperfectly measure technical change. Therefore, we suggest using the share of renewable energy supply instead, in order to analyze both innovation and diffusion of clean technologies. Furthermore, this way we would also allow for cross country comparison as those statistics are being harmonized by the international partnership initiative7.

Ultimately, and linked with the previous statement, not many authors (De Vries and Withagen (2005), Johnstone et al. (2010) have examined environmental instruments using a cross country approach. And they both suggest that using dummy variables to model environmental policies does not fully reflect policy variability. Thereby in this paper we suggest using Knill et al. (2012) density approach, which accounts for the degree of penetration of policy types by the number of instruments used, to allow the modeling of environmental policies as continuous variables. By doing so we expect to contribute to the existent literature and provide useful information for policymakers to accelerate the energy transition.

5 _{Allows us to understand the type of answer (Nordhaus, Stern, Greenpeace) that would be needed in order to} promote the transition towards renewables. See Acemoglu et al. (2012)

6 _{The variation in the propensity to patent across countries and sectors. See Johnstone et al. (2010)}

7_{Cooperation between energy and environmental statistics world leaders (IEA, Eurostat, EEA, UNDESA,}

3. Theory and Model

3.1. Empirical Method

Our empirical method, illustrated in figure 1, draws insights from four stands of the literature discussed above. First, in line with Acemoglu (2002), we assume that technical change is biased towards dirty technologies due to the price effect and market size effect forces. Furthermore, we accept that those forces would primarily be determined by the elasticity of substitution between both sectors (clean and dirty) (Panel 1).

Second, we assume that without market intervention, that corrects the existent market failures, the innovation would be directed towards the dirty sector leading us to an “environmental disaster” (Acemoglu et al. 2012). Thus, we assume that with an appropriate policy intervention we would be able to direct technical change towards clean technologies. Moreover, we accept that delaying intervention would be costly as this might increase the gap between dirty and clean technologies and thus longer the transition (Acemoglu et al. 2012) (Panel 2).

Figure 1 The Energy Transition Process

Elasticities of subtitution Weak Substitutes Price Effect Strong Substitutes Market Size Effect Technical Change Clean Technologies Laissez Faire No Diffusion Environmental Disaster Policy Intervention Energy Transition Dirty Technologies Policy

Third, we assume that market externalities limit the diffusion of clean technologies. Therefore, we would need policy intervention, directed towards the correction of those externalities, to promote the diffusion of clean technologies and to achieve the energy transition (Popp et al. 2009) (Panel 3). Fourth, we assume that some environmental policies would be more effective than others at promoting the diffusion of clean technologies, thereby representing a key ingredient in the energy transition (Johnstone et al. 2010).

3.2. Variables

Energy Transition

Given the importance of promoting renewable technologies in the process of accelerating energy transition, identifying reliable measures to track the diffusion of clean technologies with respect to dirty technologies has gained interest among environmental economists. Most of the previous literature has focused on patent data which has proven to be a useful indicator of technological development. Although due to the existence of several market failures the economic incentives to adopt those technologies might fall short 8and, thus, the society is not likely to benefit from them.

Thus, we argue in favor of using the share of renewable energy supply (ECO139) in order to account for the amount of energy that has been actually deployed. Furthermore, this indicator, unlike patents10_{, provides a useful way to track the progress across countries}11 _{and over time}

due to the consistent way of measuring energy supply (Ktoe).

Directed Technical Change

Following Acemoglu (2012), there are two main forces affecting the direction of technical change to curb environmental damages: price effect and market size effect. The price effect drives the technical change towards the more expensive factors to economize their use, while market size effect drives technical change towards the abundant factor (large market) to benefit from economies of scale. To determine the direction of the technical change, we need to compare the relative magnitude of these effects, driven by the elasticity of substitution between clean and dirty technologies.

8_{See section 2.3}

9 _{International Atomic Energy Agency, United Nations Department of Economic and Social Affairs,}

International Energy Agency, Eurostat, European Environment Agency. Energy indicators for sustainable development: guidelines and methodologies. Vienna: IAEA; 2005.

10_{Johnstone (2010) “Differences in patent regimes across countries mean that it is difficult to be certain that}

one is comparing ‘like with like’. For instance, some countries would require multiple patents for the same innovation which could be covered by a single patent in other countries”

11_{Vera and Langlois (2007) The international partnership initiative developed “a set of energy indicators and}

In the case that both sectors are weak substitutes (e<1) we expect price effect to be dominant as this implies that clean technologies are not able to replace dirty technologies, making it possible for the price of fossil fuels to continue to rise . If, instead, the two sectors are strong substitutes (e≥1), we expect the market size effect to be dominant as the increasing returns to scale overcomes the usual substitution effect. This increases the relative reward for the sector that already benefit from economies of scale.

Price effect

As noted by several authors (Hicks 1932; Binswanger 1974; Newell, Jaffe and Stavins 1999;) the relative prices of factors have a direct effect on innovation. More specifically Johnstone et al. (2010) point to the rising prices of fossil fuels as an important incentive for renewable energies. Several authors (Acemoglu 2002; Noailly and Smeets 2015; Acemoglu et al. 2012) identify this force as an important variable to determine the direction of the technical change and consequently the intensity of the regulation needed to avoid the environmental disaster.

According to Acemoglu (2002) when price is the dominant effect, we might expect both sectors to be weak substitutes (e<1) and therefore we would need a permanent policy intervention12 _{as dirty technologies would grow at the same rate as clean technologies}13_{. We}

suggest using a similar approach as the one used by Noailly & Smeets (2015) to measure energy prices. That is, to take country-level prices of the different fossil fuels using IEA (International Energy Agency) crude import prices (USD/BAR).

Market size effect

Similarly, the literature also points out the role played by the demand factors on innovation (Schmookler 1966; Scherer and Harhoff 2000; Popp 2006). Moreover, Johnstone et al. (2010) identify that not only the size of the demand affects innovation but also the demand growth as there might be higher opportunities to make profit.

According to Acemoglu when demand size is the dominant effect, we might expect both sectors to be strong substitutes (e≥1). Thus, policies might only need to be in place temporary. The intuition behind this is that once clean technologies are sufficiently developed, the market would channel renewables more than it does fossil fuels. We suggest a similar approach to the one in Noailly and Smeets (2015) which takes country-level energy consumption (from both sectors), expressed in ktoe (Kilotonne of Oil Equivalent), accounted in IEA (International Energy Agency) database.

Environmental policies

Following the literature on individual policy assessment (e.g. De Vries and Withagen 2005; Johnstone et al. 2010), we assume that the effect of environmental policies varies across

12 _{See Stern (2009)}

instrument types as their focus is set on different market externalities. For example, regulatory, mainly target adoption externalities while public investment mainly targets knowledge externalities. In this paper, we classify environmental policies in six different categories: regulatory, soft, market-based incentives, fiscal incentives, public investment, and framework policy. By doing so, we would not only cover most of the mainstream instruments discussed on climate politics literature (Schaffrin et al. 2015) but also include a great variety of other policies.

Regulatory

Represents the heavy-handed regulation response, and includes instruments like auditing, codes and standards, monitoring, obligation schemes. The objective is to achieve government objectives by compelling change to management decisions by companies which will create the necessary innovation needed to reduce the amount of CO2 that firms are allowed to emit.

As noted by Johnstone et al. (2010) this regulation type will be especially effective at inducing innovation of a less mature nature, but which have higher potential like solar or bio- energy systems. Although, the downside of this policy type, based on “picking winners”, might be that the information asymmetry makes it hard for governments to identify the best strategy and thus increases the risk of making a wrong decision.

Soft

Represents the light-handed regulation response, and includes instruments like negotiated agreements, public voluntary schemes, and unilateral commitments. The objective of these would be to achieve government objectives by using implementation of environmental regulation as a threat, so that firms act towards the government goal without the requirement to set regulatory instruments.

As noted by Morgenstern and Pizer (2007) this regulation type would be especially effective when the cost of non-compliance is relatively high as in those circumstances’ firms might want to enact in order to prevent future sanctions. However, this type of policy might not be effective in the cases where there is a risk for an abuse of market power, as firms would easily avoid those sanctions or transfer the cost to customers.

Market-based instruments

Represents an intermediate regulation response, and includes instruments like GHG emissions allowances, green certificates, white certificates. The objective is to achieve government objectives using incentive schemes and leaving freedom to management decisions.

Fiscal Incentives

Represent a regulatory response that facilitates the substitution between renewables and fossil fuels specially in the cases where the production of energy has a high fixed cost per kWh associated (Johnstone et al. 2010). An example of this type of policy would be feed-in tariffs, grants and subsidies, loans, tax relief, taxes. The problem with this type of instrument is that they are based on a strategy focused on “picking winners”, which makes the risks of making a wrong decision relatively high.

Public Investment

Represents a regulation response that provides explicit support for technological innovation, including instruments like funds to sub-national governments, infrastructure investments, procurement rules, R&D funding. The objective of those would be to lower the cost of the energy transition by promoting the technical change, especially if we direct it towards clean technologies.

As noted by Jaumotte and Pain (2005), this regulation type is especially effective at promoting innovation in cases where assessing the potential returns is complicated and thus there are higher risks on the associated investment. The problem of this type of instrument would be that there would also be a higher risk on obtaining benefits from funds due to the lack of commercial application of those innovations.

Framework Policy

This represents a regulation response that target the information asymmetry problem throughout the creation of strategic plans and institutions. The objective is to set long term objectives and obtain market information to increase policymaker’s knowledge and to make sure that the regulation is met (Johnstone et al. 2010).

Other explanatory Variables

As noted by Johnstone et al. (2010), environmental policies are affected by countrie´s energy portfolio as the effect of specific policies varies by technology. That is, some policies might have a higher effect on some technologies than in others. For example, as noted before markets-based instruments should benefit more technologies that are close substitute to fossil fuels while fiscal incentives should benefit more technologies with high fixed cost. Thereby we suggest using country fixed effects in order to control the difference between country energy portfolios.

3.3. Hypothesis and related work in the literature

H1: Changes in the relative price of fossil fuel prices will have a statistically significant impact in the energy transition. Furthermore, when the price has a positive effect over the energy transition, we might expect both clean and dirty technologies to be strong substitutes and weak substitutes when the price has a negative coefficient.

Starting with Hicks’ (1932) induced innovation hypothesis, several authors have pointed at price changes as an important factor to promote innovation. The intuition here is that when a particular factor becomes more expensive the market would try to economize that factor and thus promote the innovation of technologies that might be able to replace that factor. For that reason, Johnstone et al. (2010) suggest that when fossil fuel prices increase there would be higher incentives to invest in clean technologies as dirty technologies become relatively expensive. So that we might expect fossil fuel prices to have a positive effect over the energy transition.

Although, as noted by Acemoglu et al. (2012), an increase in the price of fossil fuels does not always direct innovation towards clean technologies. According to their theory on directed technical change an increase of fossil fuel prices only benefits renewables when their elasticity of substitution is high enough to replace dirty technologies. That is, substitutability would be the one determining the effect of fossil fuel price increase. Ultimately, Noailly and Smeets (2015) suggested that fossil fuel prices negatively affect the energy transition as their results showed that when prices increase the gap between clean and dirty technologies also tends to increase.

Therefore, in line with the literature we might expect that increasing the price of fossil fuels affects energy market equilibrium and thus the share of renewable energy supplied. The direction of this change would be determined by the elasticity of substitution between both sectors (clean and dirty). So that, when increasing fossil fuel prices ET also increases then we infer that both sectors are close substitutes. Otherwise, we might infer both sectors are weak substitutes when increasing fossil fuel prices reduces the energy transition.

H2: Increases in the relative size of energy markets will have a negative and statistically significant impact in the energy transition.

Similar to the previous hypothesis, although less supported by the literature, some authors suggest that market size is also an important factor to promote innovation. That is that as large markets increase the relative reward for the abundant factor, we might expect that when one resource becomes scarce markets will shift towards abundant factors. For that reason, Johnstone et al. (2010) suggest that in markets with growing energy consumption the incentives to innovate in the clean sector are relatively high, as fossil fuels became relatively scarce and clean technologies represent the abundant factor.

technologies would overcome the usual substitution effect14 and thus increase the relative reward for the abundant factor. That is why we expect that market size increases will have a negative effect over the energy transition as the scale effect of fossil fuels might direct innovation towards fossil fuels more than it does to renewables.

H3: Environmental policy density (nª instruments) will have a positive and statistically significant impact in the energy transition

Building upon the theory on directed technical change, several authors (De Vries and Withagen 2005; Acemoglu et al. 2012; Stokes et al. 2018) indicate that without policy intervention technological innovation would be directed towards the dirty sector. Therefore, the gap between renewable technologies and dirty technologies would increase thus negatively affecting the energy transition (share of renewable energy supply). The intuition is that environmental policies have a positive impact over the energy transition as they redirect the technical change towards renewables and thus accelerate the energy transition.

Furthermore, Knill et al. (2012) point out that using more regulatory targets would in most cases increase the penetration of that field and thus the overall instrument effectiveness. For example, in the case of the energy transition governments might identify several regulatory policy targets (limit CO2 emissions, subsidize renewable energies, promote innovation, foster competition). That is why we expect that when the number of instruments in force increase their effect over their targets should also increase and for that reason the share of renewable energy supply (ET) should also increase.

H4: Market-based instruments will have a significantly higher impact on the energy transition than other type of instruments.

According to several authors (Zerbe 1970; Downing and White 1986; Milliman and Prince 1989; Jung et al. 1996) new technologies have a higher rate of adoption when using market-based instruments than under direct regulation. Their intuition is that market-market-based instruments allow firms to decide which technology to adopt and how and thus provides incentives to adopt technologies that are close to the market and thus that can replace dirty technologies. In contrast, regulatory instruments might not necessarily focus on fossil fuel close competitor technologies and thus have a slower rate of adoption.

Furthermore, De Vries and Withagen (2005) and Johnstone et al. (2010) results also support this assumption that markets-based instruments lead to higher rate of adoption, when comparing the effectiveness of different policy instruments over innovation. In line with the literature we also expect market-based instruments to be particularly effective at increasing the share of renewable energy supply (ET). As firms adopt technologies that are ready to replace dirty technologies therefore have an immediate impact on the energy transition.

Table 1 below reports the expected signs of the coefficients. The signs of the three main explanatory variables refer to the (1-3) hypotheses previously formulated.

Table 1 Expected signs of the explanatory variables

Expected drivers of energy transition

Directed Technical Change

Weak Substitution Strong Substitution Price - + Market - - Policy + + Regulatory + + Soft + + Market-based instruments + + Fiscal Incentives + + Public Investment + + Framework Policy + +

Note: + positive effect, 0 no effect, - negative effect on the share of renewable energy supply

3.4. Period and sample of countries

Regarding the time frame, we cover the period 1998-2016, therefore allowing us to analyze the period between Kyoto agreement and Paris agreement. Furthermore 1998 was also highly relevant for the EU as it was the year in which the European Union burden agreement was passed. This agreement facilitated the 8% GHG reduction target, committed in Kyoto, as it allowed the redistribution of the reduction target among the member states

(

More specifically we would focus on European countries 15(excluding non-OECD countries), as most datasets only include data for OECD countries. Our sample is convenient to study the energy transition, as being the area with the highest share of RE supply (15,26% which is twice as much as in the USA and almost 5% more than the OECD).

Moreover, we expect European Union’s unique political status to favor the introduction of environmental policies, as it provides cooperation mechanisms (e.g. joint support schemes) that reduces the renewable support policy costs (Jacobsen et al. 2014). European countries also showed to be the area with the highest environmental awareness, 93% consider climate change a serious problem in 2019 (Eurostat 2019) while only 45% consider it in USA (Gallup

15_{Austria, Belgium, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland,}

2019). Therefore, this study will focus on the European countries which offer the best conditions to accelerate the energy transition.

4. Data and methodology

4.1. Data

As noted earlier, in this study we would use three main explanatory variables (price effect, market size effect and environmental policies) to examine the factors affecting the energy transition in European countries over the 1998-2016 period. To do so, we would use different data sources (IEA World Energy Balances 2018; OECD Crude oil import prices indicator

2020; EEA database on climate change mitigation policies and measure in Europe; IEA/IRENA Joint Policies and Measures database). First, we used IEA World Energy Balances 2018 data on renewable energy supply (ktoe) and total energy supply (ktoe) to

construct the energy transition indicator (share of renewable supply) and total energy consumption (ktoe) to create the market size indicator.

Figure 2 shows the share of renewable energy supply for the years 1998 and 2016. As expected, Scandinavian countries have the highest counts. Furthermore, we can also see that the degree of dispersion between countries is relatively high, it is for that reason that we might need a model that relaxes the assumption that all individuals have the same coefficients (i.e. FE), or a model that does not require distributional assumptions (one-step difference GMM).

Figure 2 Share of Renewable Energy Supply

Second, we employ OECD Crude oil import prices indicator 2020 data to construct the price indicator16. Figure 3 below displays the small variation in oil prices experienced by our sample countries during the 1998-2016 period. It shows that the variance between countries is relatively small and thus indicates that countries average crude import prices are statistically representative.

Figure 3 Crude Import Prices (USD/bar)

Source: Author’s calculations using data from OECD Crude oil import prices indicator 2020 Third, we utilize EEA database on climate change mitigation policies and measure in Europe

to create a vector of environmental policies (regulatory, soft, market-based, fiscal, framework) and IEA/IRENA Joint Policies and Measures database to include Norway’s environmental policy data, which was not included in the EEA database. Table 2 below shows the average number of environmental policies used to measure policy intensity. It suggests that the dispersion between countries policy intensity (nº instruments) is relatively high. For example, we see that countries like Belgium or Denmark primarily relay on market-based instruments while other countries like Luxemburg or Great Britain rely more in regulatory instruments. For that reason, we might need to use models that account for time-invariant individual characteristics (i.e. FE, OSDGMM).

16_{OECD Crude oil import prices indicator 2020 database does not have data for all countries and years. For}

Table 2 Environmental polices (nº instruments)

Regulatory Soft Market based Fiscal incentives Public Investment Framework policy AUT 7,68 0,00 7,42 0,68 0,53 3,79 BEL 15,53 2,68 19,16 4,42 1,16 16,63 CZE 10,79 0,47 9,37 2,47 0,68 4,16 DNK 12,47 3,00 20,84 9,21 0,16 6,21 EST 3,26 0,00 7,11 0,53 0,47 2,84 FIN 11,79 1,00 9,16 4,47 0,16 11,16 FRA 11,42 0,68 10,63 7,95 0,47 8,53 DEU 3,42 0,47 4,63 0,84 0,00 1,42 GRC 12,74 0,00 10,37 6,53 0,00 10,37 HUN 2,79 0,00 2,16 0,58 0,00 2,79 ISL 1,21 0,00 0,00 0,00 0,00 0,00 IRL 5,68 2,58 5,37 1,74 0,00 7,00 ITA 3,11 0,37 1,58 0,47 0,00 0,47 LVA 6,21 0,47 3,68 4,05 0,00 0,89 LUX 16,47 1,63 3,74 2,05 0,42 2,53 NLD 6,05 4,79 3,89 5,00 0,00 1,74 NOR 6,63 0,00 13,95 9,63 1,63 12,00 POL 8,16 0,63 5,79 0,89 1,37 5,05 POR 7,16 0,63 2,84 1,11 0,00 4,47 SVK 3,26 0,00 2,11 0,00 0,00 0,00 SVN 10,84 0,00 14,32 2,84 1,47 15,37 ESP 2,16 0,26 5,21 1,00 0,11 4,47 SWE 9,47 1,68 9,95 6,26 1,00 5,26 CHE 8,89 0,58 3,53 1,32 0,74 3,58 GBR 13,95 1,32 10,42 3,21 0,26 4,84

Table 3 below shows the descriptive statistics of the variables employed in the model. The number of observations indicate that our dataset is strongly balanced, implying that for every variable in the dataset, there is an observation for every country (25) and period (19).

Table 3 Descriptive Statistics

Descriptive

Statistics Obs. Mean Std. Dev. Min Max

ET 475 0,169047 0,17676 0,103077 0.8974612 Price 475 58,93155 32,9533 11,66 115.64 Market 475 47052,42 57.332 159,908 238267,6 Regulatory 475 8,046316 7,40837 0 46 Soft 475 0,930563 1,52663 0 8 Market-Based 475 7,488421 8,03509 0 35 Fiscal 475 3,090526 3,25599 0 15 Investment 475 0,425263 0,90486 0 6 Framework 475 5,423158 6,69497 0 30

4.2. Methodology

To test the hypothesis set out in Section 3 we would use the following regression equation:

(ET

)

(ET

) + 𝛽

+ 𝛽

(PRICE

) + 𝛽

(MARKET

) + 𝛽

(POLICY

) + 𝜀

(1)

where i = 1, . . . , 25 indexes the cross-sectional unit (country) and t = 1,998, . . . , 2,016 indexes time. The dependent variable, energy transition, measured by the share of renewable energy supply. The explanatory variables include the price effect (PRICE it) measured by

crude import prices, the market size effect (MARKET it) proxied by total energy consumption

and a vector of policy variables (POLICY it), which includes the number of instruments in

force in six different policy types. Furthermore, we include the error term (𝜀𝑖𝑡) to capture

the residual variation and energy transition lagged variable (ETit-1) to take into account the

dynamic adjustment process of the energy transition. As several authors (Aghion et al. 2016, Greaker et al. 2018) suggest that the energy transition is path dependent and thus that the share of the actual renewable supply is built upon previous renewable supply levels.

Several methods have been explored to ascertain the robustness of our results. They range from a simple panel data to the more sophisticated GMM method and the quantile regression method.

Source: ET and Market taken from IEA World Energy Balances 2018; Price taken from OECD Crude oil import prices indicator 2020; Environmental policies taken from EEA

4.3. Statistical analysis of data

To assess the normality of the error term we perform a graphical representation of the ET observations (Figure 4) and the result (4.A) show that our dependent variable is positively skewed therefore suggesting the use of a logarithmic model. Furthermore, the results derived from the Jarque-Bera normality test17 _{also support the intuition that the observations are not}

normally distributed. Therefore, we transform ET from linear to a logged variable and the results (4.B) show that the ET values (logged) represent much more the behavior of a normal distribution.

Figure 4 Distribution of ET and Logged ET

Source: Author’s calculations using data from IEA World Energy Balances 2018 Therefore, we would use a log-linear model with a logged dependent variable on one side and absolute values for the independent variables on the other as shown is equation 2.

(lnET

)

(lnET

) + 𝛽

+ 𝛽

(lnPRICE

) + 𝛽

(lnMARKET

) + 𝛽

(POLICY

)

+ 𝜀

(2)

We test for the homoscedasticityassumption. , that is that for each x, the values of e and y are distributed around their mean value following probability distributions that all have the same variance: var(e)= σ2= var(y). To test for the presence of heteroskedasticity we perform the white test (Annex 4) and confirm the hypothesis that errors are heteroskedastic at 0.01 significance level as the p-value reported is equal to zero. Thus, we need to use robust standard errors to account for heteroscedasticity.

We also assess whether the explanatory variables are collinear. The correlation matrix for all our explanatory variables (Table 4). In order to validate this assumption, we would need low correlation between independent variables. Results show that all coefficients are below 0.8,

except form the one relative to framework and market-based policies. In order to measure the magnitude of this problem, we use the Variance Inflation Factor (VIF) indicator to measure the degree of multicollinearity between variables. According to O’Brien (2007) when this indicator is higher than 10 the reliability of the model would be affected by the relationship between the explanatory variables. Although as noted by our results (Annex 5) no VIF value exceed 10 and thus we do not present multicollinearity issues.

Table 4 Correlation Matrix

Correlation

Matrix lnET lnP lnMkt Reg. Soft

Market

Based Fiscal Invest. Frame.

lnET 1 lnPrice 0.2413 1 lnMarket -0.3520 0.0079 1 Regulatory 0.0650 0.4496 0.0225 1 Soft -0.1317 0.2351 0.1221 0.4786 1 MarketBased 0.1880 0.3897 0.0580 0.7817 0.4281 1 Fiscal 0.2380 0.2852 0.1109 0.5702 0.3934 0.7050 1 Investment 0.1802 0.2412 0.0067 0.4817 0.1390 0.5848 0.3492 1 Framework 0.1460 0.3174 0.0240 0.7599 0.3080 0.8349 0.5941 0.6009 1

4.4. Model fit

Regarding the model selection, one important point that we need to meet is the lack of serially correlated errors. As noted earlier we believe this kind of dynamic to be present in our dependent variable (ET) as several authors noted (Aghion et al. 2016, Greaker et al. 2018). In order to confirm this intuition, we would need to test for the presence of serially correlated errors. One way to do so, might be using the Wooldridge test which examines the relationship between a given variable and a lagged version of itself over various time intervals. Results (Annex 6) allow us to reject the null hypothesis at a 0.01 significance level, meaning that there are serially correlated errors. Thus, we use dynamic panel data methods to find the best estimations for this study.

Another way to test if the model is static or dynamic would be to perform a dynamic panel data and observe if the lagged variable is significantly different from 0. Furthermore, if we also compare the lagged coefficient of the ET with other models (POLS and FE) this would also help us determine whether we need to use a one-step difference GMM (OSDGMM) estimator or a two-step system estimator. If the coefficients of lagged ET in the one step

Source: Data for lnET, lnMarket taken from IEA World Energy Balances 2018, data for lnPrice taken from

difference GMM estimator are close or below the FE coefficient then we would use two- step difference GMM estimator. Otherwise we would use one step difference GMM.

Table 5 displays the three different regressions (POLS-FE-OSDGMM) showing that the OSDGMM seems to be in between the POLS and FE coefficients. Thus, we use the one-step difference GMM model design by Arellano and Bond (1991). As noted by Roodman (2009), under the presence of heteroscedasticity and autocorrelation, this would be the best type of estimator available.

Table 5 Estimated Coefficients of Lagged (and Logged) ET

POLS OLS Model Fixed Effects Model

Arellano-Bond One-Step Difference GMM Model

lnET_L1 0.854*** 0.198*** 0.542**

(23.61) (4.28) (3.67)

Standard errors in parentheses

Source: Data for the construction of ET taken from the IEA World Energy Balances 2018

* p < .05, ** p < .01, *** p < .001

Furthermore, the results from Table 5 also support Wooldridge test results as the coefficient of lagged ET in OSDGMM showed to be statistically significant and different from zero18_.

Therefore, we use a one-step difference GMM, which uses lagged levels of the variables to instrument the difference between variables, to capture the dynamic relationship of our data. Although due to the fact that GMM estimation does not count with indicators (R2, adjusted R2) to measure how good does the model fit the data, we need to run additional test (Arellano Bond, Hansen).

First, the Arellano Bond test (Annex 7) shows that there seems to be autocorrelation in first order (p-value 0.012), therefore supporting the use of a dynamic model to fit the data, but not for the second order (p-value 0.309), which suggests that we might only need to include one time lag and not two. Second, the Hansen test (Annex 8) shows that the instruments used are valid but weakened by many instruments.

While the above-mentioned estimation methods offer a wide-variety of techniques that assess whether our results are robust, they nonetheless put the parameter estimates in a straight-jacket. This suggests that indicators of directed technical change or policy instruments do not vary from one country to the next, despite stringent differences across our sample. We need to allow the parameter estimates to vary along the distribution of the ET variable. The quantile regression technique constitutes a nice alternative which also can also accommodate issues related to endogeneity. For these reasons, its underlying results are of interest.

5. Econometric Analysis and Results

In Table 6 we present our baseline results displayed in six different regressions. The first regression (POLS), assumes coefficients do not differ between individuals although as we have previously seen in Figure 2 (pp. 21) our country selection might not fulfill this assumption. Thereby in our second regression we use a model that does account for unobserved time-invariant individual characteristics as Fixed Effects (FE). However, this model is not suited for dynamic adjustment relationships, that is that if the omitted variable is correlated with the included one’s coefficients would be biased. Therefore, based on the results derived from section 4.4, which confirm that past values of our dependent variable affect current ET values, we might need to find another model that adjust for this dynamic relationship.

For that reason, in our third regression we use a one-step difference GMM estimator to control for the endogeneity of the lagged dependent variable. Although, in order to allow our parameter estimates to vary along the distribution we might need to use less restrictive model, which is why we use a quantile regression (QE) in our fourth- six regressions.

In general results appear to confirm our intuition that the ET transition follow a dynamic adjustment process, that is that the lag of our dependent variable (lnET_1) appears to be useful to estimate the current change in ET. More specifically, column (3) show that an increase in one percent of lagged ET would increase the ET by 0.54 percent. Moreover, looking at the quantile regression (column 4) we see that an increase in one percent of lagged ET would increase the ET by 0.97 percent.

Table 6 Results (1) OLS (2) FE (3) OSDGMM (4) QR (5) QR (0.25) (6) QR (0.75) lnET_L1 0.854*** 0.198*** 0.542*** 0.970*** 0.933*** 0.956*** (0.0362) (0.0462) (0.148) (0.000328) (0.00499) (0.00467) lnMarket -0.0473** -0.664** -0.346 -0.00181*** -0.00975*** -0.00274*** (0.0185) (0.239) (0.281) (0.000421) (0.00210) (0.000844) lnPrice 0.212*** 0.204*** 0.153*** 0.0427*** 0.0759*** 0.0231*** (0.0583) (0.0396) (0.0267) (0.00147) (0.00377) (0.00738) Regulatory -0.00712** 0.0186* 0.0207** -0.000206 0.00303*** -0.00104** (0.00330) (0.0106) (0.00973) (0.000260) (0.000829) (0.000423) Soft -0.0155** 0.0500** 0.0195 -0.000678 -0.00485*** -0.00494*** (0.00735) (0.0236) (0.0152) (0.000455) (0.00128) (0.000648) MarketBased 0.00490* 0.0175** 0.0107 0.00151*** 0.00499*** 0.000292** (0.00261) (0.00793) (0.00903) (0.000132) (0.000455) (0.000143) Fiscal 0.0132** -0.0201 -0.0213 -0.00119*** -3.61e-05 0.00309* (0.00623) (0.0268) (0.0290) (0.000238) (0.000340) (0.00168) Investment 0.00159 -0.0332 0.0203 0.00512*** 0.00403** -0.00381*** (0.0110) (0.0281) (0.0267) (0.000788) (0.00169) (0.00136) Framework 0.000722 -0.0104 -0.0120 -0.00109*** -0.00563*** 0.00128*** (0.00234) (0.0126) (0.0101) (0.000316) (0.00101) (0.000259) Constant -0.688** 3.894 (0.291) -2.389 Observations 474 474 449 474 474 474 R-squared 0.854 0.738 Nº of id 25 25 Nº of groups 25 25 25

Robust standard errors in parenthesis

Source: Data for lnET, lnMarket taken from IEA World Energy Balances 2018, data for lnPrice

taken from OECD Crude oil import prices indicator 2020, data for environmental policies taken from EEA database on climate change mitigation policies and measure in Europe.

Hypothesis 1

Results from both column (3) and (4) enable us to confirm that changes in the relative price of fossil fuels appears to have a statistically significant (0.01 level) impact on the ET. Furthermore, results from both one step difference GMM and quantile regression show that fuel prices has a positive effect over the energy transition. More specifically, column (3) show that an increase in one percent in fossil fuel prices would increase the ET by approximately 0.15 percent. While using a quantile regression (column 4) we see that an increase of one percent in fossil fuel prices would increase the ET by approximately 0.04 percent. Thereby, we might expect both clean and dirty technologies to be strong substitutes as when fossil fuel price increases the share of renewable energy supply (ET) also increases. This might indeed have important political consequences as when both sectors are close substitutes, subsidies to clean technologies would only need to be in place for a temporary period to direct innovation towards renewables. The intuition behind this is that once clean technologies are sufficiently advanced innovation would be directed towards these technologies without any further intervention (Acemoglu et al 2012).

Hypothesis 2

Results from both column (3) and (4) show that the relative size of energy markets does have a negative effect over the ET, although only results from column (4) seem to be statistically significant. More specifically, column (4) show that an increase of one percent in energy market size would decrease the ET by approximately 0.18 percent. Thereby this would confirm Acemoglu et al. (2012) intuition that increasing returns to scale of dirty technologies increase the relative reward for the abundant factor and thus direct innovation towards fossil fuels. Although in this case the magnitude of these scale effect (- 0.0018) would not be able to overcome the substitution effect (as discussed in section 2.2) as the price effect (0.0427) remains the dominant effect. That is that innovation would be directed towards clean technologies despite the negative sing of the scale effect.

Hypothesis 3

have a positive effect over the ET (2) regulatory and soft policies remain statistically insignificant.

Hypothesis 4

Results from both column (3) and (4) do not enable us to confirm that market-based instruments are more effective that other type of instruments at promoting the energy transition. First, looking at column (3) we see that market-based policies do not seem to have a direct effect on ET at any relevant level of statistical significance. Moreover, their estimated coefficient (0.0107) would still not be more significant than the one of other instruments (i.e. regulatory 0.0207, investment 0.0203, soft 0.0195). Second, looking at column (4) market-based policies became statistically significant at the 0.01 level although their impact was far from being the more significant one (market based 0.0015, investment 0.0051).

6. Conclusions

This paper examines the effect of directed technical change and environmental policies on the energy transition in a cross-section of 25 European countries over the 1998-2016 period. The main contribution of this paper is its emphasis in both directed technical change and environmental policies as the drivers of the energy transition. We have seen that previous studies highlight the role played by the price of fossil fuels and energy market size at driving the technical change away from clean technologies. For that reason, we have argued that environmental regulation plays an essential role not only in directing the technical change towards renewables but also promoting the diffusion of those innovations.

Our analysis starts looking for an appropriate way to measure the energy transition across countries and over time. As discussed above we find that using patent application would be an imperfect measure of the energy transition. Therefore, we use the share of renewable energy supply, which is one of the worldwide indicators developed by the international partnership initiative, to track the progress of our sample countries. Moreover, observing that the two main forces that direct the technical change, are determined by the elasticity of substitution between renewables and fossil fuels, we identity two different scenarios.

Our results, in line with the literature on directed technical change, show that both price and market size play a relevant role in determining the direction of technical change. Therefore, by including these two forces we allow our estimates to measure their effect and by doing so determine the substitutability between both sectors. In this case, increasing fossil fuel prices seems to have a positive effect over the energy transition and increasing the market size a negative one. So that, we might expect clean technologies and dirty technologies to be close substitutes and thus a temporary subsidy would be enough to direct innovation towards renewables. Although unlike we have seen in previous literature, our results do not show market size effect overcome the usual substitution effect, thus indicating that the scale effect would not be strong enough to direct innovation towards fossil fuels when both sectors are close substitutes. Thereby, one important line of research would be to check other indicators, rather than energy consumption, to proxy the market size effect and test the assumption that scale effects overcome the usual substitution effect.

Moreover, our findings also show that increasing the number of instruments does not always increase the share of renewable energy supplied. This implies that, in this case, using more regulatory targets does not necessarily increase their effect. One reason for this to happen is that counting the number of instruments without accounting for their content might be an imperfect way of measuring policy output. Therefore, another line of research would be to combine the use of both policy density (nº instruments) and intensity in order to differentiate between the contents of policies as some might use economic resources while other use normative power.

Ultimately and linked with the results on using the number of instruments to proxy policy output, we find that market-based instruments do not seem to be more effective at promoting the energy transition than other instruments. As explained before, one reason why this might not happen is because of the way we have measured policy output. Thereby, we might need to include policy density in our analysis and test if market-based instruments are in fact relatively more effective than other instruments at promoting the energy transition.

References

Acemoglu, D. (2002). Directed technical change. Review of Economic Studies, 69, 781–810. Acemoglu, D., Aghion, P., & Hemous, D. (2014). The environment and directed technical change in a north-south model. Oxford Review of Economic Policy, 30(3), 513–530. https://doi.org/10.1093/oxrep/gru031

Aghion, P., Dechezleprêtre Antoine, Hémous David, Martin, R., & Van Reenen, J. (2016). Carbon taxes, path dependency, and directed technical change: evidence from the auto industry. Journal of Political Economy, 124(1), 1–51. https://doi.org/10.1086/684581

Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: monte carlo evidence and an application to employment equations. The Review of Economic Studies, 58(2), 277–297.

Bretschger, L. (2017). Is the environment compatible with growth? adopting an integrated framework for sustainability. Annual Review of Resource Economics, 9(1), 185–207. https://doi.org/10.1146/annurev-resource-100516-053332

Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1), 115–143. https://doi.org/10.1016/S0304-4076(98)00009-8

Binswanger, H. P. (1974). A microeconomic approach to induced innovation. The Economic Journal, 84(336), 940–958.

Downing, P. B., & White, L. J. (1986). Innovation in pollution control. Journal of Environmental Economics and Management, 13(1), 18–29. https://doi.org/10.1016/0095-0696(86)90014-8

EEA (2020). Database on climate change mitigation policies and measures in Europe Eurostat (2019). Citizen support for climate action

Gallup (2019). Americans as Concerned as Ever About Global Warming

Greaker, M., Heggedal, T.-R., & Rosendahl, K. E. (2018). Environmental policy and the direction of technical change. The Scandinavian Journal of Economics, 120(4), 1100–1138. https://doi.org/10.1111/sjoe.12254

Grimaud Andrâe, & Rouge, L. (2008). Environment, directed technical change and economic policy. Environmental and Resource Economics, 41(4), 439–463.

Hicks, J. R. (1932). Marginal productivity and the principle of variation. Economica, 35(35), 79–88.