Master Thesis
Universiteit van Amsterdam Department of Economics
Environmental policy and innovation in renewable energy:
Are governments on the right track?
Karlijn Kersten Student Number: 6077390 Supervisor: Carmine Guerriero Second Reader: Sander Onderstal
2
Statement of originality
This document is written by Student Karlijn Kersten who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Acknowledgements
I would like to thank Carmine Guerriero from the UvA for his supervision in writing this master thesis. In addition, I would like to thank SEO Economic Research for giving me the opportunity to write this master thesis and Aloys Kersten for his useful comments and suggestions.
4
Abstract
This paper analyses the effect of different public environmental policies on innovation in renewable energy. More precisely, this paper studies the relation between environmental policy and environmental patenting activity in the area of four renewable energy technologies (i.e. wind, solar, ocean and biomass & waste). It builds upon previous literature by extending the time period, i.e. 1980-2012, and in addition assesses a different methodology (i.e. instrumental variables approach) that allows for possible simultaneous determination of environmental policy and environmental innovation.
The results show that tax incentives, feed-in tariffs and obligations are significantly influencing environmental innovation. Surprisingly, tradable certificates do not seem to have an impact on innovation. Accordingly, this suggests that financial incentives are better stimulators on environmental innovation compared to regulatory instruments. However, including time fixed effects suggests that unobserved time variant factors are also important determinants of innovation. Moreover, the results under a dynamic negative binomial fixed effects model and an instrumental variables approach seem to differ, suggesting there is evidence of simultaneous causality between environmental policy and environmental innovation.
5
Table of contents
1. Introduction 6
2. Literature review 8
2.1 Theories on the effect of environmental policy on environmental innovation 8 2.2 Environmental policy and environmental innovation: empirical evidence 9
2.3 Simultaneous causality 11
3. Hypotheses 12
4. Data and empirical strategy 13
4.1 Data and descriptive statistics 13
4.2 Empirical strategy 19
5. Results 23
5.1 Results negative binomial model 24
5.2 Results instrumental variables approach 27
5.3 Robustness check 31
6. Discussion 32
7. Conclusion 34
8. Limitations and future research 35
References 37 Appendix I 41 Appendix II 42 Appendix III 44 Appendix IV 45 Appendix V 46
6
1. Introduction
“Today, we can see with our own eyes what global warming is doing. In that context it becomes truly irresponsible, if not immoral, for us not to do something."
[Joe Lieberman1]
Global warming and, more broadly, climate change is a worldwide phenomenon that receives an increasing amount of attention. Especially in the last two decades this debate has been more and more on the political agenda, both nationally as internationally, illustrated by the Kyoto Protocol (1997) and the Bali Road Map (2007). During these and other conventions most countries agreed upon a certain reduction of emissions in the future. However, several countries, for example The Netherlands and the United Kingdom, do not meet their targets (European Commission, 2015).
In fighting climate change scholars claim that technological (environmental) innovation is of great importance (e.g. Stern, 2008). Energy emissions represent about two third of total emissions worldwide. Moreover, renewable energy, which is considered more environmental friendly, represents 13.2% of the total primary energy supply (IEA, 2014). Hence, technological innovations in especially the area of renewable energy can significantly contribute to the accomplishment of public environmental objectives (Johnstone et al. 2010).
In addition to the contribution of public environmental objectives, increasing the share of renewable energies could also contribute to other public objectives. The market for fossil fuels is uncertain, illustrated by, for example, the tension with Russia of the last couples of years. Increasing the share of renewable energies could, therefore, lead to greater (national) energy security.
This paper, therefore, examines if public policies that aim to fight climate change resulted in increased innovative activities in renewable energy in seven North-Western European countries. Environmental innovation is approximated by the number of patents in each renewable energy technology (i.e. in energy from wind, solar, ocean and biomass & waste). Moreover, this paper evaluates seven different environmental policies.
“Greenhouse gas (GHG) emissions are externalities and represent the biggest market failure the world has seen” [Stern 2008, p. 1]. This quote nicely illustrates why government intervention is highly important when it comes to climate change. Costs of abatement and thus the price of GHGs are not internalized by the market, justifying government intervention. Likewise, with respect to environmental
1 Source: Joe Lieberman. (n.d.). BrainyQuote.com. Retrieved July 15, 2015, from BrainyQuote.com Web site: http://www.brainyquote.com/quotes/quotes/j/joelieberm175646.html
7
innovation there exist technological externalities that justify government intervention (Greenwald & Stiglitz, 1986). Because spillover effects reduce the potential return, firms tend to underinvest in innovation. Production costs of renewable energy, therefore, are higher than for substitute fossil fuels. Consequently, without government intervention the deployment of renewable energy remains limited.
In general, government stimulation of renewable energy fall into three categories: financial instruments, regulatory instruments and voluntary instruments. Two theoretical approaches explain the effectiveness of these instruments by emphasizing if and how they solve the previously explained market failures. Section 2 extensively describes these two approaches.
Empirical evidence is mixed and limited, although since the last decade scholars are more interested in estimating the relation between environmental policies and environmental innovation. This paper builds upon existing literature by extending the paper of Johnstone et al. (2010). Likewise, this paper studies the effect of different types of policy instruments on the number of patents in renewable energy technologies. However, there are several differences.
Firstly, the focus of this paper is only on North-Western European countries. Secondly, the analyses of this paper is extended with eight years. Finally, and most importantly, the empirical strategy of this paper allows for potential endogeneity between environmental policy and environmental innovation. Most empirical studies do not acknowledge that environmental policy and environmental innovation could be simultaneously determined. Carrion-Flores and Innes (2010), however, stress that the relation between environmental policy and environmental innovation can go both ways. Also McCain (1978) and Downing and White (1986) indicate that environmental policy responds to environmental innovation. Innovations lead to lower electricity generation costs which in turn may lead to further inducement of more stringent environmental policies. This paper, therefore, uses an instrumental variable approach. Correspondingly, this paper evaluates to what extent results differ when controlling for possible simultaneous determination of environmental policy and environmental innovation.
The main findings of this paper are the following. Firstly, for certain renewable energy technologies some of the implemented environmental policies seem to positively influence environmental patenting activity. In particular financial stimulation, such as tax incentives and feed-in tariffs, seem to function well. The impact of regulatory instruments on environmental innovation is less clear. On the one hand obligations have a significant impact on some technologies. On the other hand, tradable certificates do not influence innovation. Secondly, the results suggest there is suspicion that environmental policy and environmental innovation are simultaneous determined.
The rest of this paper is organized as follows. Section 2 provides an overview of the literature related to the environmental policy – environmental innovation relation, both theoretically and empirically. Section 3 presents the hypothesis of the research questions. Section 4 describes the data and the empirical
8
strategy of this paper. Section 5 presents the results, which follows by the discussion in section 6. Section 7 concludes. Finally, section 8 discusses the limitations and in addition gives some suggestions for future research.
2. Literature
review
This section provides an overview of the existing literature on the effect of environmental policy on environmental innovation. The first part presents an overview of the theoretical evidence. The second part elaborates on the most important empirical literature studying the effect of environmental regulation on environmental innovation.
2.1 Theories on the effect of environmental policy on environmental innovation
In general, there are two strands of theories explaining the relation between environmental policy and innovative activities concerning the environment. These two categories are the ‘induced innovation’ approach and the ‘evolutionary’ approach or ‘Porter hypothesis’. Differences in explaining the determinants of innovation between these two approaches have implications for instrumental policy choice.
The induced innovation approach, first introduced by Hicks (1932), states that firms innovate because of changes in the relative price of input factors. Innovation reduces the need for these input prices. The induced innovation theory thus implies that firms innovate with the intention of producing a new (more) profitable process or product (Jaffe et al., 2002). The nature and magnitude of a firm’s innovative effort depends on the expected increased value of the innovative activity. Stated otherwise, a firm maximizes the expected discounted value of future cash flows. Hence, innovative activities are profit motivated investment decisions.
The induced innovation hypothesis has important implications for environmental policy. The hypothesis suggests that a tightening of environmental policy reduces investment in technology because environmental policy implicitly or explicitly makes environmental inputs more costly (Newell et al., 1998). In other words, environmental regulation reduces profits because they require firms to internalize externalities. This means an unproductively allocation of inputs (Jaffe et al., 1995). Hence, if investing in (environmental) innovative activities would have been profitable in the first place, firms would have chosen to do so even without (environmental) regulation.
The second approach, the Porter hypothesis, recognizes that because of different market failures (i.e. financial and return uncertainties) and unpriced externalities (i.e. pollution) firms underinvest in environmental innovation (Porter 1991; Porter & Van der Linde, 1995a). Moreover, already noted in 1962 by Arrow, because of spillover effects, firms tend to invest less in innovation than they would have done
9
could they capture the full return. Firms are thus not always optimizing. An important implication of this line of reasoning is that the imposition of an external constraint (e.g. a new environmental policy) not necessarily reduces profits (Jaffe et al., 2002). If firms do not optimize, regulation, in theory, could induce firms to optimally invest in innovative activities that otherwise would not occur. Hence, there exists at least the theoretical possibility that environmental regulation leads to an increase in innovative activities whereby pollution decreases and profits increase. This suggests there could be a ‘win-win’ outcome for both the environment and firms. Consequently, properly designed environmental regulation enhances innovation and partially or even more than partially offsets the associated compliance costs (Porter & Van der Linde, 1995a). However, this does not mean that necessarily all environmental regulation increases innovation. Also other win-win theorists point out that environmental regulation should be properly designed in order to maximize innovative activities.
After Porter and Van der Linde (1995a) introduced their hypothesis economists were sceptical about their ‘win-win’ theory. According to these economists (for example Palmer et al., 1995; Oates et al., 1993; Jaffe et al., 1995), it is unlikely that additional environmental constraints or incentives would be beneficial. Firms do the best they can given the available information. In other words, if profitable investment opportunities are systematically being missed, why would environmental regulators, with less information, be in a better position to change this.
Porter and Van der Linde (1995a) specifically refer to ‘environmental regulation’ in terms of additional constraints imposed by the government. Environmental policy, however, is a broader concept which not only implies policy in terms of constraints. Governments also attempt to stimulate environmental protection with financial incentives. In both the induced innovation hypothesis and Porter approach, theoretically, public financial incentives can lead to an increase in environmental innovation. If financial incentives overcome the different market failures, investing in environmental innovation can be profitable. Hence, financial incentives can have a positive influence on environmental patenting activity.
2.2 Environmental policy and environmental innovation: empirical evidence
Since the first publication (i.e. Lanjouw & Mody, 1996) that empirically studies the relation between environmental regulation and environmental innovation, there has been an increasing amount of evidence contributing to this debate. However, the empirical evidence is still rather limited. Lanjouw and Mody (1996) suggest there is a correlation between a firm’s pollution abatement expenditures and environmental related patents with a one to two year lag. However, the authors of this paper do not explicitly test this relation empirically nor do they control for other factors that might spur (environmental) innovation.
Other papers, that test this claim more formally, also explore pollution abatement expenditures. This so called PACE (Pollution abatement and control expenditures) variable is reported expenditures by
10
firms and thus serves as a proxy for governmental abatement pressures (Brunnermeier & Cohen, 2003)2. Both Jaffe and Palmer (1997) and Brunnermeier and Cohen (2003) rely on the PACE variable for their analysis3. While Jaffe and Palmer (1997) use both R&D expenditures and patents as a proxy for innovation, Brunnermeier and Cohen (2003) rely solely on the latter approximation. Furthermore the authors extend the former paper by including only environmentally related patents. In addition, they include pollution related visits by the government as a measure of environmental stringency4. Both papers rely on a dataset from the US manufacturing industry.
The results of Jaffe and Palmer (1997) show no significant relation between pollution abatement expenditures and innovation. Foreign patents and industry valued added, on the other hand, do significantly influence the number of environmental related patents. Contrary to this result, Brunnermeier and Cohen (2003) conclude that the PACE variable is highly significant. The other measure of policy stringency, governmental inspections, only has an impact for certain industries. The coefficient suggests mean patenting increases by 0.04 percent when industry abatement expenditures increases by one million dollar (e.g. the PACE variable). Hence, the effect is significant, though the magnitude is economically very small. Carrion-Flores and Innes (2010) come to the same conclusion that the effect environmental policy is rather small5. A possible explanation for the different results could be that Jaffe and Palmer (1997) include all patents, not only environmental related patents (De Vries & Withagen, 2005).
Above studies focus on all environmental innovations. There are, however, scholars that focus on only innovations in air pollution. Furthermore, the literature presented below are, contrary to above papers, cross-country studies. De Vries and Withagen (2005) investigate the Porter Hypothesis by linking environmental stringency to patents in the field of sulphur dioxide (𝑆𝑂2) abatement. Their measure of
instrumental stringency is whether or not international agreements (e.g. the Helsinki, the Oslo and Gothenburg Protocols) where in effect. The results indicate that the effect of environmental policy stringency on innovation is mixed, depending on the specified model. However, their measure of environmental stringency does not seem to fully reflect policy variability (Johnstone et al., 2010). Moreover, the existence of a protocol does not necessarily mean that the agreements are met. De Vries and Withagen (2005) find similar results with respect to the effect of R&D expenditures on innovation6
2 The PACE variable is collected from a survey conducted by The US Census Bureau.
3 The paper by Brunnermeier & Cohen (2003) is an extension of the paper by Jaffe & Palmer (1997).
4 Brunnermeier & Cohen (2003) also extend the paper of Jaffe & Palmer (1997) by including more control variables. Jaffe & Palmer (1997) control for industry value added and foreign patents, while Brunnermeier and Cohen (2003) extend this by controlling for market structure, by adding industry concentration, capital- and export intensity and include time fixed effects.
5 Carrion-Flores & Innes (2010) use a firm’s toxic air pollution target to measure policy stringency.
11
Popp (2006) also uses patents for 𝑆𝑂2 abatement, in addition to mono-nitrogen oxides (𝑁𝑂𝑥)
abatement for the analysis. The focus of his paper is on the effect of domestic regulation on foreign patents. The results suggest that, on the one hand, an increase in domestic emissions standards has only a minor impact on foreign patents. On the other hand, there is a significant impact on foreign patents when regulation in the respective home country tightens. Hence, regulatory pressures from abroad increase innovation in the home country. In earlier work Popp (2003) studies the effect of instrument choice on environmental innovation. He compares patenting activity under command-and-control policies and market-based policies7. The results suggest that under the latter policy innovative activity decreased compared to innovative activity under the former.
A paper more closely related to this paper is the one by Johnstone et al. (2010), who study the effect of different policy instruments on innovation in specifically the area of multiple renewable energy technologies. The results show that policy instruments differ in their effectiveness regarding different renewable energy technologies. The authors conclude that broad-based policies, such as tradable certificates, stimulate innovation more when the renewable energy technology is a closer competitor to fossil fuel energy. Targeted subsidies, such as feed-in tariffs, are more effective to stimulate innovation when the energy technology is more costly. Furthermore, the results suggest that for most technologies public R&D expenditures has a positive and significant impact on environmental innovation.
To summarize, most of the empirical evidence suggests, on the one hand, that more stringent environmental policy positively influences environmental patenting activity. On the other hand, the evidence indicates that these effect are in general rather small.
2.3 Simultaneous causality
Previous discussed literature do not take into account that environmental policy might be endogenously determined. Does environmental policy affect environmental innovation or is it also possible that the relation goes the other way around? Moreover, there could be unobserved factors that influence both environmental policy and environmental innovation. Carrion-Flores and Innes (2010) recognize this problem by stating that the causal effect might go in both directions. Therefore, not recognizing this problem may lead to inconsistent estimates. McCain (1978) also emphasises that environmental policy and technical innovation are jointly determined. Moreover, innovation leads to lower electricity generation costs which in turn can lead to further inducement of more stringent environmental policies (Downing & White, 1986). Hence, this means that environmental policy potentially also responds to environmental innovation.
7 The difference between command-and-control policies and market-based policies are that the former leaves little flexibility with respect to how firms can comply with environmental regulation, while the latter does.
12
3. Hypotheses
This section presents the hypotheses for each of the research questions. The hypotheses follow from the literature discussed in the previous section. For the first research question, the effect of environmental policy on environmental innovation, the hypothesis reads as follows:
𝐻1: Environmental policy positively influences environmental innovation in renewable energy
technologies.
Because previous literature overall conclude that environmental policy has a significant positive influence on environmental policy this paper expects to find the same. The latter supports the Porter approach and not the induced innovation approach. Hence, for the policy variables that are regulatory instruments this paper expects a positive relation8. However, governments also use financial incentives to stimulate renewable energy. The main concerning market failures relate to financial incentives. Therefore, this paper expects financial stimulations that aim to correct these market failures to have a significant impact on environmental patenting9. Nonetheless, the question remains if public financial incentives are enough to overcome these market failures. In other words, even though the government financially supports renewable energies it can still be the case that it is not profitable. Therefore this paper expects that relatively higher financial support increase innovation more compared to lower financial support. This means, for example, that higher public R&D expenditures have a greater impact on patenting activity compared to lower R&D expenditures10.
The corresponding hypothesis of the second research question, if environmental policy and environmental innovation are simultaneously determined, is the following:
𝐻2: The results obtained under an instrumental variables approach differ from the results obtained under an approach not allowing for possible simultaneous determination of environmental policy and environmental innovation. .
With respect to this second hypothesis this paper expects that environmental policies and environmental innovations suffer from simultaneous determination, as suggested by McCain (1978),
8 Section 4.1 discusses the different environmental policy variables. However for clarity the following policy variables included in the regression are regulatory instruments: renewable energy certificates (e.g. tradable permits) and obligations (for example quotas and performance standards).
9 The two main market failures are that the price of emissions is not internalized and spillover effects of R&D. 10 Besides public R&D expenditures the following policy instruments included in this paper are financially related: tax incentives, feed-in tariff and investment incentives.
13
Downing & White (1986) and Carrion-Flores and Innes (2010). Consequently, results obtained with a model that does not account for this could be inconsistent. Therefore, they might differ from the results obtained under an instrumental variables approach. The main reason for this claim is that firms respond to environmental regulation by choosing the “best available control technology” (Jaffe et al. (2002, 2003)). This suggests that if, because of innovations, firms can easier comply with environmental policies the government responds by setting even more stringent environmental policies. In terms of financial incentives the argument implies that governments financially support those innovative areas that underperform and thus respond to environmental innovation levels. Overall, this paper expects that the environmental policies variables might be endogenous.
4. Data and empirical strategy
The empirical analysis relies on a panel dataset collected from different sources. The panels are seven northern-west European countries11. The data for each panel is quantified on an annual basis and covers the period 1980-2012. The first part of this section provides a detailed description of the data. The second part presents and discusses the empirical strategy.
4.1 Data and descriptive statistics
This part gives a detailed description of the different variables used for the empirical strategy. The first part describes the dependent variable, followed by a description of the explanatory variables and the instrumental variables. Table 1reports the descriptive statistics for all employed variables.
4.1.1 Dependent variable: environmental innovation
Finding a reliable quantitative measure for innovation is a difficult task for economists. Potential candidates are research and development expenditures and the number of scientific employees (Johnstone et al. 2010). However, since these indicators do not measure the innovative outcome they are imperfect measurements and thus less preferred. This paper, therefore, uses an output measurement: successful environmental patent applications. Several previous studies (e.g. Popp (2002), Brunnermeier & Cohen (2003), Johnstone et al. (2010) and Jaffe & Palmer (1997)) also use successful environmental patent applications as a proxy for environmental innovation.
Griliches (1990) is one of the first researchers that discussed the use of patent application as an economic indicator. According to his study, patents are a valuable source of information reflecting the
11 The following countries are included in the analysis: Denmark, Germany, France, The Netherlands, Norway, United Kingdom and Sweden.
14
innovative performance of firms. Moreover, since patent applications are readily available, discrete, subdivided into technology areas and well documented they are an attractive quantitative measure of innovation. However, there are also some problems with using patents for economic analysis. Firstly, patents differ in their technical and economic significance. This means that each patent value is not fully comparable with each other. Secondly, not all inventions are patented nor are all inventions patentable. Therefore, the propensity to patent across industry and country differs. Lastly, patent regimes are not the same in every country (Johnstone et al., 2010). For example, in some countries you can have multiple patents for the same innovation while in others this innovation is covered by only one patent. This means that it is difficult to compare patents between countries. With respect to the last two mentioned problems this study fortunately can control for country specific differences in the specified regressions. Despite some drawbacks of using patent data it is still the best available proxy for the innovative performance of firms.
The OECD Patent Database provides the data for the variable Patent (OECD, 2015)12. Patents are counted by date of application rather than by the date of grant (i.e. when the patent is officially recognized). The date of application is when the investor perceives he or she made a valuable invention (Jaffe & Palmer, 1997). In addition, the time between the date of application and the date of grant is variable per patent and can take up to four years. For each patent application there is information about the investor’s country, time of application, the office where the application was submitted and the field of technology.
The analysis includes patent applications from the European Patent Office (EPO). In practice, an investor can apply to both the national patent authority and the EPO. When an inventor applies at the EPO the patent is secured within al European countries. Although EPO application are more expensive it is still less costly to apply at the EPO instead of applying in each individual European country (Popp, 2003). Therefore, EPO application are a good reflection of the overall patent activity within the European Union.
This paper focuses on innovation in different (technology) types of renewable energy sources (i.e. wind, solar, geothermal, ocean, hydro and biomass and waste). The OECD Patent Database (OECD, 2015) uses International Patent Classification (IPC) codes to classify the different technology types. This means the number of patents per renewable energy technology depends on the classification per technology. The list of technology types with matching IPC codes and definitions are given in table A2 in Appendix II. A potential concern with IPC codes is that the classification includes irrelevant patents and exclude relevant patents. However, IPC codes for renewable energy sources are straightforward and easy to interpret. Therefore errors are minimal (Lanjouw & Mody, 1996).
15
Table 1: Descriptive Statistics
Variable Obs. Mean Std. Dev. Min. Max.
All renewable 231 54.186 122.025 0 854
Wind 231 14.874 38.235 0 280
Patents Solar 231 23.874 64.669 0 475
Ocean 231 1.489 2.975 0 19
Biomass & Waste 231 9.805 16.643 0 108
Investment Incentives* 231 0.671 0.471 0 1 Tax Incentives* 231 0.545 0.499 0 1 Voluntary Programs* 231 0.134 0.342 0 1 Obligations* 231 0.437 0.497 0 1 REC* 231 0.156 0.363 0 1 REC Percentage (%) 231 1.112 3.224 0 18.200 FIT* 231 0.325 0.469 0 1
Policies FIT Level Wind 231 0.027 0.043 0 0.236
FIT Level Solar 231 0.051 0.122 0 0.564
FIT Level Ocean 231 0.010 0.030 0 0.150
FIT Level Biomass & Waste 231 0.020 0.038 0 0.150
FIT Level All Renewables 231 0.027 0.050 0 0.209
R&D Solar¹ 223 21.753 31.056 0 174.740
R&D Wind¹ 223 11.618 13.165 0.294 88.062
R&D Ocean¹ 231 1.868 4.052 0 34.483
R&D Biomass & Waste¹ 222 13.687 17.357 0.181 110.782 R&D All renewables¹ 224 54.931 57.133 4.236 322.972
EPO fillings (per GDP) 231 4.493 2.307 0.791 11.572
Control Kyoto* 231 0.333 0.472 0 1
Variables ETS system* 231 0.242 0.429 0 1
GDP (% change) 231 2.038 1.979 -5.638 5.989
Electricity Price (% change) 231 4.413 12.365 -27.066 48.445 Electricity Consumption (% change) 231 0.957 3.016 -9.732 10.444 Instrumental Environmental Score Government 231 6.134 3.201 0.703 16.558
Variables Right/Left Wing Government 231 -2.121 9.445 -20.744 21.880
Notes: * Binary variable.
¹ Given in millions USD, 2015 prices and PPP.
4.1.2 Explanatory variables
4.1.2.1 Public policy measures
Seven different public policy instruments are distinguished: investment incentives (e.g. low-interest-rate loans, grant, risk guarantees); tax incentives (e.g. accelerated depreciation, tax exemptions); voluntary
16
programs; obligations (e.g. production quotas, guaranteed market); tariff incentives (e.g. feed-in tariffs); tradable (green) certificates and public R&D support. This paper closely follows the categorization of public policy instruments by Johnstone et al. (2010). The IEA/IRENA Global Renewable Energy Policies and Measures Database (IEA/IRENE, 2015) provides the data on the national environmental policies (Policy), except if it is stated otherwise13.
The Global Renewable Energy and Measures Database is a joint effort by the International Energy Agency and the International Renewable Energy Agency. The database contains a detailed description of each implemented national policy. In addition it includes the starting- and end date of each policy. The IEA/IRENA consults the national authorities on a regular basis to update the database.
It is most preferable to estimate the effect of both the presence and stringency of a policy instrument. This means that instead of including the different policy variables as dummy variables it would be most preferable to include them as continuous variables. However, since some types of policies are impossible to compare across countries it is impossible to construct continuous variables for some of the public policy variables. For example, every country has its own tax system. A one percent tax exemption has not the same monetary value in every country. Therefore, the analysis includes the policy variables investment incentives, tax incentives, voluntary programs and obligations as binary variables. This binary variable is equal to 1 if the specific policy instrument is in effect and 0 otherwise. These variables thus do not represent the relative stringency. However, they do reflect the effect of the presence of such a policy.
For the other three policies (e.g. public R&D expenditures, feed-in tariffs and tradable (green) certificates) this paper constructs continues variables. National public R&D expenditures per type of technology is obtained from the IEA’s Energy Technology Research and Development Database (IEA, 2015). Feed-in tariffs represent the price that is given per technology for every kWh of energy.14 Furthermore, the percentage renewable energy that covers one certificate or that should be generated for one certificate measures the stringency of the tradable certificate policy variable.
4.1.2.2 Other explanatory variables
Besides public policy instruments there are other factors that may influence the innovative performance of firms in the field of renewable energy sources. This paper, therefore, controls for these factors.
13 Source IEA/IRENA: http://www.iea.org/policiesandmeasures/renewableenergy/.
14 For feed-in tariffs this paper thanks Nick Johnstone for granting some of the data. In addition this paper would like to thank the following persons who have helped collecting the data: OECD-EPAU (2013), Renewable Energy Policy Dataset, version February 2013. Compiled by the OECD Environment Directorate’s Empirical Policy Analysis Unit (Johnstone, N., Haščič, I., Cárdenas Rodríguez M., Duclert, T.) in collaboration with an ad hoc research consortium (Arnaud de la Tour, Gireesh Shrimali, Morgan Hervé-Mignucci, Thilo Grau, Emerson Reiter, Wenjuan Dong, Inês Azevedo, Nathaniel Horner, Joëlle Noailly, Roger Smeets, Kiran Sahdev, Sven Witthöft, Yunyeong Yang, Timon Dubbeling).
17
The return of innovating in renewable energy is the potential return and/or the potential market. A growing market or growing (output) prices increases the incentive to innovate. With respect to renewable energy sources this means that electricity prices and electricity consumption are a good reflection of the potential return and potential market, respectively (Popp, 2002; Newell et al., 1998). Since the share of renewable energy sources in the total electricity production is relatively small, the effect of an increase in this share will not have an effect on the price of electricity (Johnstone et al., 2010). Therefore, the price of electricity is assumed to be exogenous.
The variable Consumption is collected from the IEA’s Energy Balances Database (IEA, 2015). This paper uses the sum of the percentage growth of residential electricity consumption and industry electricity consumption to reflect the potential market of renewable energy sources.
In addition, the IEA’s Energy Prices and Taxes Database (IEA,2015) provides the data on electricity prices (labelled as Price). The end-user price of electricity for residents and industry users is weighted by the consumption levels of these two groups. Moreover, the variable represents the yearly growth rate. Unfortunately, for some years the electricity price for either residents or industry were missing in the IEA’s Energy Prices and Taxes Database. For these missing values approximations were constructed by calculating the average yearly growth rate of the countries with no missing values.
The empirical analysis also controls for the state of the economy by including the percentage growth in GDP taken from the World Bank Database. The label of GDP growth rate is GDP.
The general propensity to patent and scientific capacity may differ across countries and therefore may affect the innovative activity of a country. To correct for this, the analysis control for all EPO patent applications per GDP15. Hence, this variable is a ‘trend’ variable as it corrects for the differences in general propensity to patent across country and over time (Johnstone et al. 2010).
Finally, this paper controls for two significant policy changes. Firstly, in December 2002 different countries all over the world signed the Kyoto Protocol (Kyoto), which could be a sign of more stringent policies in the future (Johnstone et al. 2010). Secondly, in 2005 the European Commission introduced the European emission trading system (ETS). This system limits the emission volume per firm by giving allowances for each volume of emission16. If firms have emit more than their allowances cover, they are obliged to buy these allowances on the market, and vice versa. Since the analysis only includes European countries, all countries implemented both policies changes. Therefore, the analysis includes a binary variable which takes on the value 0 prior to implementation and the value 1 after implementation. This paper, ideally, would like to treat these two policies the same as the other environmental policies. All
15 Also the patent data for this variable comes from the OECD Patent Database. 16 Source: http://ec.europa.eu/clima/policies/ets/index_en.htm.
18
countries in the analysis implemented these policies. Therefore, there is no variation in the data and thus no possibility to estimate the influence, so the analysis only controls for the presence of these policies.
Since R&D investments also take place on a firm level basis one ideally would, in addition to public R&D expenditures, like to control for this factor. Unfortunately, firms are reluctant in disclosing R&D expenditures for confidentiality reasons. If firms disclose their R&D related expenditures at all, this is on an aggregate basis and not distinguished by the type or specific amount of the innovative activity. There exist some surveys (e.g. by the OECD or UNESCO) regarding private R&D. These surveys, however are not held on an annual basis and divide businesses according to the International Standard Industrial Classification (ISIC). The latter means that private R&D expenditures are not specified with respect to renewable energy only. As a consequence this paper cannot control for private R&D expenditures in the analysis. Other studies also not control for private R&D expenditures.
4.1.3 Instrumental variables
As described in the empirical strategy section (see hereafter), this paper uses an instrumental variable approach to overcome endogeneity problems. The analysis relies on two instrument. This section describes the data used in the empirical strategy, while the next two sections discusses the validity of both instruments.
Firstly, the instrument Rile indicates whether the government in a particular year is left or right wing oriented. The Manifesto Project Dataset provides the data for this variable Rile17. This dataset contains different indicators of each government of several countries. Under these indicators is the right-left position of each party18. Each party receives a score given by the party’s ideology and program, which in turn indicates the political stance of the corresponding party. The variable Rile is constructed by weighting the average of these scores for each party that forms a coalition. The scores are weighted by the share of votes obtained during the election19. If the elections were in year t, the score on right-left position of the chosen government is in effect from year t+1. This is because the data used in the analysis is on a yearly basis, so if the elections were during the year the score is assigned to the following year. This can theoretically be justified by the fact that there is a period of time between the date of elections and the date the government starts implementing policies.
Secondly, the other instrument used in the empirical analysis is the stance of the government regarding environmental protection. The codebook of The Manifesto Project Dataset describes this
17 The official website of The Manifesto Project Dataset is https://manifesto-project.wzb.eu/. Official reference: Volkens, Andra / Lehmann, Pola / Matthieβ, Theres / Merz, Nicolas / Regel, Sven / Werner, Annika (2015): The Manifesto Data Collection. Manifesto Project (MRG / CMP / MARPOR). Version 2015a. Berlin:
Wissenschaftszentrum Berlin für Sozialforschung (WZB).
18 The right-left position of each party as given in Michael Laver/Ian Budge (eds.): Party Policy and Government Coalitions, Houndmills, Basingstoke, Hampshire: The MacMillan Press 1992.
19
indicator as following [page 16]: ‘General policies in favour of protecting the environment, fighting climate change, and other “green” policies’. For this indicator each political party also receives a score. Like the variable Rile this variable is created by the weighted average of these scores for the parties that formed a coalition. EnvS is the label of this instrumental variable.
4.2 Empirical strategy
This part of section 4 presents the empirical strategy, which consist of three different specification.
4.2.1 Baseline specification
The empirical strategy starts with a reduced-form equation which is similar to the model specification used by Johnstone et al. (2010). The baseline specification is as follows:
(𝑃𝑎𝑡𝑒𝑛𝑡𝑠𝑖,𝑡) = 𝛽0+ 𝛽1(𝑃𝑜𝑙𝑖𝑐𝑦𝑖,𝑡−2) + 𝛽2(𝐶𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛𝑖,𝑡) + 𝛽3(𝑃𝑟𝑖𝑐𝑒𝑖,𝑡) + 𝛽4(𝐺𝐷𝑃𝑖,𝑡) + 𝛽5(𝐸𝑃𝑂𝑖,𝑡) + 𝛽6(𝐾𝑦𝑜𝑡𝑜𝑡) + 𝛽7(𝐸𝑇𝑆𝑡) + 𝜆𝑖+ 𝜀𝑖,𝑡 (1)
Where i = 1,…,7 indexes stand for the cross-sectional unit (country) and t = 1980,…,2012 indexes time. The dependent variable Patent represents the number of patent applications in each of the included renewable energy technologies (i.e. wind, solar, ocean, biomass & waste and all renewables). The explanatory variable is a vector of the different Policy variables as outlined in the previous section: investment incentives; tax incentives; voluntary programs; obligations; tradable (green) certificates; feed-in tariffs and public R&D expenditures. In addition, this paper feed-includes different control variables: electricity consumption (Consumption), electricity price (Price); GDP growth (GDP); total EPO fillings per GDP (EPO); rectification of the Kyoto treaty (Kyoto) and implementation of the ETS system (ETS). Furthermore, 𝜆𝑖 represent a fixed effects term to control for unobserved country-specific heterogeneity.
All of the specifications discussed in this section do not include time fixed effects20. The reason for not including time fixed effects is that year dummies captures a lot of the model’s variation. However, unobserved variation over time could be an important factor explaining increased patent activity. This paper, therefore, checks for robustness by adding clustered time dummies.
Following Johnstone et al. (2010) the analysis continuously uses robust standard errors for each specification. Robust standard errors correct for heteroskedasticity (Stock & Watson, 2012). However, robust standard errors are not valid if there is auto-correlation between the regression errors. Because the regression errors potentially are auto-correlated this paper preferably uses clustered standard errors.
20
Unfortunately, the used software package does not allow using clustered standard errors for one of the specifications21. For the sake of consistency robust standard errors are used.
The patent outcome takes a count form, with no negative values, some zero values and several positive integer values. The literature suggests to use count data models to estimate the number of occurrences of an event (Cameron & Trivedi, 1998). The number of events is the number of successful patent applications at the EPO by each country within each technology area. Examples of count data models are Poisson and negative binomial models. It is assumed that the number of patent applications follows a negative binomial distribution, allowing for over-dispersion. Hausman and Griliches (1984) describe the negative binomial distribution applied to patent counts. This model assumes that the dispersion (e.g. how squeezed the (sample) distribution is) varies across countries but is constant over time. Moreover, the fixed effects parameter (i.e. controlling for unobserved country heterogeneity) may vary between countries. This paper estimates the negative binomial model by conditioning on the sum of patent counts for each country over time. The different parameters are estimated by maximum likelihood under fixed effects.
4.2.1.1 Solving endogeneity problems
As discussed in section 2, a problem that could arise with the estimation of equation (1) is simultaneous determination of the environmental policy variables and the number of patents (i.e. innovation). In addition, there could exist unobserved factors that affect both policy decisions and patent activity. Therefore, the different policy variables are potentially endogenous (Carrion-Flores & Innes, 2010; Smith & Urpelainen, 2014; Besley & Case, 1994; Downing & White, 1986). To correct for this, the analysis includes the different policy variables with a two year lag. The number of patent counts in year t are not likely to influence policy setting in year t-2. Hence, policy in year t-2 is most likely exogenous in equation (1).
Another reason for including the policy variables with a two year lag is because there is a period of time before governmental stimulation of innovation results in a patent application. The literature suggests (e.g. Hall et al., 1984; Popp, 2002; Klaassen et al, 2005; Popp, 2010) that innovation responds with some lag to both environmental policy and R&D.
4.2.2 Instrumental variables approach
4.2.2.1 Endogenous treatment effects model
Given the possibility of policy endogeneity this papers also estimates the environmental policy-environmental innovation relation with an endogenous treatment effects model. The different policies are
21 In the software package Stata it is not allowed to use clustered standard errors in the endogenous treatment effect model (discussed below). Unfortunately, only robust standard errors can be implemented.
21
the endogenous treatments. The endogenous treatment effects model allows for endogenous dummy variables (treatment effects) (Terza, 1998). Using standard regression techniques for the estimating of the effect of a endogenous (dummy) policy variable results in biased and inconsistent estimators (Heckman, 1978). A simple two stage regression estimation solves the endogeneity problem for strictly continuous outcomes. If, however, the endogenous explanatory variable is a binary response, such estimation techniques are invalid because a linear fit is applied to nonlinear data. Heckman (1978) uses, therefore, a maximum likelihood estimation.
The endogenous treatment effects model has the following form:
(𝑃𝑎𝑡𝑒𝑛𝑡𝑠𝑖,𝑡) = 𝛼0+ 𝛼1(𝑃𝑜𝑙𝑖𝑐𝑦𝑖,𝑡−2) + 𝛼2(𝐶𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛𝑖,𝑡) + 𝛼3(𝑃𝑟𝑖𝑐𝑒𝑖,𝑡) + 𝛼4(𝐺𝐷𝑃𝑖,𝑡) + 𝛼5(𝐸𝑃𝑂𝑖,𝑡) + 𝛼6(𝐾𝑦𝑜𝑡𝑜𝑡) + 𝛼7(𝐸𝑇𝑆𝑡) + 𝜆2𝑖+ 𝜀2𝑖,𝑡 (2) 𝑃𝑜𝑙𝑖𝑐𝑦 ∗𝑖,𝑡−2= 𝛿0+ 𝛿1𝑍𝑖,𝑡−𝑥+ 𝑣𝑖,𝑡 (3)
𝑃𝑜𝑙𝑖𝑐𝑦𝑡−2= { 1 𝑖𝑠 𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 𝑖𝑓 𝑃𝑜𝑙𝑖𝑐𝑦 ∗ 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒.𝑡−2 > 0, (4)
Previous studies also rely on above specification (i.e. equation 2-4) when estimating a count model with endogenous treatment effects. For example, Kenkel and Terza (2001) estimate the effect of physician advice on alcohol consumption using a count regression with an endogenous treatment effect. The authors use a similar model as specified above.
Equation (2) differs with respect to equation (1) in that the latter includes a vector of policy variables, while the policy variable in equation (2) represents just one policy instrument. Because the estimation technique only allows for the analyzation of one endogenous dummy variable at the same time, each policy variable is estimated one by one. Each regression, however, includes all of the control variables. Furthermore, since the estimation technique only allows for an endogenous dummy variable and not a continuous endogenous variable, it is impossible to estimate three of the policy variables (i.e. the three continuous policy variables: FIT Level, REC Percentage and R&D). Therefore, the variables FIT Level and REC Percentage are transformed into a binary variable which is equal to one if the value of the corresponding policy variable is larger than its median value, and zero otherwise. This means that the stringency of this policy variables, although different than previously, can still be estimated. For the estimation of the effect of public R&D on the number of patent counts this paper uses another regression technique (e.g. generalized method of moments); which is discussed below.
22
4.2.2.2 Estimation specification for the effect of public R&D expenditures
The previous discussed endogenous treatment effect model does not allow for an endogenous continuous variable. Because all of the countries included in the regression implemented a R&D policy during the majority of the time span of the analysis, it makes no sense to construct a binary variable for R&D expenditures. Therefore, to estimate the effect of public R&D expenditures on the number of patent counts this paper uses another type of instrumental variable approach. This specification has the following form:
(𝑃𝑎𝑡𝑒𝑛𝑡𝑠𝑖,𝑡) = 𝛾0+ 𝛾1(𝑅&𝐷𝑖,𝑡−2) + 𝛾2(𝐶𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛𝑖,𝑡) + 𝛾3(𝑃𝑟𝑖𝑐𝑒𝑖,𝑡) + 𝛾4(𝐺𝐷𝑃𝑖,𝑡) + 𝛾5(𝐸𝑃𝑂𝑖,𝑡)
+ 𝛾6(𝐾𝑦𝑜𝑡𝑜𝑡) + 𝛾7(𝐸𝑇𝑆𝑡) + 𝜆3𝑖+ 𝜀3𝑖,𝑡 (5)
(𝑅&𝐷𝑖,𝑡−2) = 𝜂0+ 𝜂1𝑍2𝑖,𝑡−𝑥+ 𝜂1(𝐶𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛𝑖,𝑡) + 𝜂2(𝑃𝑟𝑖𝑐𝑒𝑖,𝑡) + 𝜂3(𝐺𝐷𝑃𝑖,𝑡) + 𝜂4(𝐸𝑃𝑂𝑖,𝑡) + 𝜂5(𝐾𝑦𝑜𝑡𝑜𝑡) + 𝜂6(𝐸𝑇𝑆𝑡) + 𝜆4𝑖+ 𝜀4𝑖,𝑡 (6)
Equation (5) is the second stage of the instrumental variables approach. The first stage, equation (6), determines 𝑅&𝐷𝑖,𝑡−2 using the instrument. In addition, the first stage also includes the control variables
from the second stage.
Because the dependent variable in equation (5) is a count variable, the model is estimated with an exponential conditional mean model (Poisson regression model) with an endogenous regressor.
This specification includes the instrument with the same lag as the corresponding policy variable (e.g. two year lag). Moreover, table A1 in appendix I depicts the instrument and its lag used in this specification.
4.2.2.3 Validity of instrumental variables
In equation (3) and (6) the variable Z serves as an instrument for each environmental policy variable. A valid instrumental variable is an instrument which satisfies two conditions: the instrument relevance condition and the instrument exogeneity condition (Stock & Watson, 2012). Instrument relevance means that the instrument has a non-zero correlation with the endogenous (instrumented) variable. Instrument exogeneity implies that the instrument has a zero correlation with the error term of the second stage regression. The latter means that the only effect the instrument may have on the dependent variable (in the second stage regression) is via the endogenous variable. This means the instrument is exogenous.
Applying these conditions to this study means that the chosen instrument must have a direct effect on each policy variable (e.g. relevance condition) but cannot have a direct effect on the number of patent counts (e.g. exogeneity condition). The analysis relies on two instruments: whether the government is left or right wing and the environmental stance of the government. For the estimation of each policy variable
23
either left/right wing government or the environmental score of the government serves as an instrument (e.g. one of the two instruments).
In order to get consistent and unbiased estimates the instrument should be valid. The link between the two instruments and the different policy variables can be justified on the idea that the type of government influences the policies that are set. In general, left wing governments tend to be more concerned with the environment than right wing governments. This means that theoretically speaking left wing governments set more environmental friendly policies compared to right wing governments. In addition, left wing parties are in general more pro-government intervention compared to right wing parties. The same line of reasoning holds for governments with parties who in their programs explicitly state to be concerned with the environment and thus score high on the environmental stance score. Instrumental relevance can also be tested empirically by looking at the first stage F-statistics. The next section discusses the results of this test.
Both instruments are, in theory, also uncorrelated with the error term of the patent equation (e.g. equation (2) and (5)). The only channel through which the type of government influences the number of patenst is via a governments’ policy decision. In other words, whether a government is left or right wing and the degree of environmental responsiveness should not directly affect the number of patent counts.
All instruments are included with the same lag as the environmental policy variables (i.e. with a two year lag). However, the policy variables Obligations and REC target are included with a three- and four year lag, respectively, to increase the fit of the model2223.
5. Results
This section presents the results of the empirical analysis. The first part describes the results of the baseline model (equation (1)). The second part presents the results of two the instrumental variables approaches, followed by a comparison of the results with the baseline specification. The final part describes the results of the robustness check.
Before presenting the results of the regression analysis, this paper first has a look at the raw data. Figure 1 shows that the number of EPO patent applications for renewable energy per country increased
22 For two specific combinations a different instrument lag is used. This concerns the following specifications: 1) when regressing the policy variable Obligation on the number of patent counts in the technology field ocean and 2) when regressing the policy variable FIT level on the number of patent counts in the technology field wind. The instrument is these cases are included with a two- and four year lag, respectively. The reasoning for doing this is because including the instrument with the same lag used in the other technology areas resulted in a weaker instrument. Therefore, including the instrument with a different extra lag improves the fit of the model. Moreover, since the lag that is used in each specification is equal or larger than the lag of the policy variable theoretically this is still consistent. The latter is justified on the idea that a government in time t-x can influence the policy setting in time t. The power of the rest of the instruments per specification is described in section 5.
23 Table A2 in Appendix II also presents the instrument and the corresponding lag that is used for the R&D policy variable.
24
over time. Especially the last decade each country experienced a growth in patents in renewable energy. As discussed in the introduction, attention for environmental protection also increased over time. Moreover, the data on the environmental policy variables suggests that governments increasingly pay attention to the environment. Especially since the nineties governments implemented different environmental policies. From these figures alone, however, one cannot conclude that the increasing number of patents are the result of more (stringent) environmental policies. The following part attempts to answer this question empirically.
Figure 1: Number of EPO patent applications for renewable energy over time per country
5.1 Results dynamic negative binomial model
5.1.1 Results public policy variables
Table 2 reports the estimates for the baseline specification (equation (1))24. The results indicate that each policy instrument has a different impact on patenting in every technology area. Starting with the description of the results of the policy variables which are included as a dummy variable. While investment incentives do not seem to have a significant impact on the number of patents, tax incentives significantly affect innovation in all of the technology areas. Investment incentives are only significant on patenting in wind energy (with a 10% confidence level). Voluntary programs only seem to affect innovation in solar and
24 This and all other specification include the different policy variables with a two year lag Several alternative lag structures were estimated (e.g. one- and three year lag). The results indicate that the estimations are robust using different lag structures.
0 100 200 300 400 500 600 700 800 900 1 9 8 0 1 9 8 1 1 9 8 2 1 9 8 3 1 9 8 4 1 9 8 5 1 9 8 6 1 9 8 7 1 9 8 8 1 9 8 9 1 9 9 0 1 9 9 1 1 9 9 2 1 9 9 3 1 9 9 4 1 9 9 5 1 9 9 6 1 9 9 7 1 9 9 8 1 9 9 9 2 0 0 0 2 0 0 1 2 0 0 2 2 0 0 3 2 0 0 4 2 0 0 5 2 0 0 6 2 0 0 7 2 0 0 8 2 0 0 9 2 0 1 0 2 0 1 1 2 0 1 2 Nu m b er o f p aten ts Year DE DK FR UK NL NO SE
25
biomass & waste, although for the latter the sign is negative (at a 10% significance level). Furthermore, obligations seem to positively influences the number of patents in the area of wind, ocean and biomass & waste, in addition to all renewable energy sources.
Regarding the policy variables for which the effect of policy stringency is also estimated, the results show that on the one hand the implementation of a tradable certificate system (e.g. REC target) has a significant negative impact on innovation in wind and biomass & waste. On the other hand, a more stringent tradable certificate system (i.e. REC Percentage) seems to increase patent activity as the sign of this coefficient is positive. Since the second effect (i.e. the variable measuring stringency) is in magnitude larger than the first effect, the overall effect is positive. However, the second effect is never significant. The negative first effect of implementation of tradable certificates seems to be due to collinearity with the second effect measuring policy stringency25. Overall, the results thus suggest that tradable certificate systems do not influence environmental innovation in renewable energy.
With respect to feed-in tariffs, the effect of the variable that measures policy stringency (FIT Level) is again in magnitude larger than the variable that measures the effect of whether the policy is implemented (FIT)26. Therefore, the overall effect of a more stringent feed-in tariff on the number of patent counts is positive, with the exception for wind. The effect, however, is only significant for the technologies solar, ocean and all renewable energy sources. The results show that every cent per kWh increase in feed-in tariffs significantly increase patenting in solar energy by 1.5%27.
Except for ocean, public R&D expenditures seem to significantly raise patent activity in renewable energy technologies. However, although the results are significant (at a 1% significance level), the impact is relatively small for each technology. Every million euro that the government subsidizes R&D in innovation in wind, solar and biomass & waste, increases the number of patent counts with 1.4%, 1.0% and 1.2%, respectively. The magnitude of the impact of public R&D on patenting in renewable energy overall is even smaller: every million euro subsidized by the government increases patent activity by 0.4%.
25 The correlation between the variable REC Target and the variable REC Percentage is as expected high, almost 0.8. Because in terms of econometrics it is preferred to include both variables in the specification. Therefore, the results of both variables are presented and discussed.
26 In addition to the policy tradable certificates, also for the variables FIT and FIT Level of the different technologies it holds that the difference in signs of these affect is most likely due to collinearity. The correlation between these variables ranges from 0.4 to 0.8.
27 The estimates reported in all of the tables included in this section indicates how much a unit in a certain variables changes the logarithm of the number of patents. The latter is not easy to interpret. Therefore, this paper also estimated the incidence-rate ratios (IRR), which have a more simplistic interpretation. The impact of 1.4% is the effect using incidence-rate ratios. The statistical software package, however, did not allow to estimate the IRR for the endogenous treatment effect model (presented below). In order to be consistent, therefore, the estimates reported in each of the tables are not IRR estimates.
26
Table 2: Estimated coefficients of the negative binomial fixed effects model
Wind Solar Ocean Biomass & Waste All renewables Policy variables: Investment Incentive 0.367* (0.203) 0.213 (0.166) -0.334 (0.501) 0.086 (0.204) 0.128 (0.145) Tax Incentive 0.882*** (0.182) 0.337** (0.163) 1.398*** (0.377) 0.299* (0.178) 0.609*** (0.121) Voluntary Programs 0.210 (0.209) 0.698*** (0.153) 0.509 (0.575) -0.383** (0.195) 0.049 (0.139) Obligations 0.430* (0.245) 0.090 (0.275) 1.131* (0.584) 0.597*** (0.228) 0.371** (0.175) REC Target -1.156*** (0.317) 0.105 (0.286) -0.114 (0.461) -0.605* (0.332) -0.438** (0.194) REC Percentage 4.189 (3.229) 0.766 (3.078) -1.905 (4.437) 4.923 (3.134) 1.384 (2.120) FIT 0.208 (0.347) -0.092 (0.118) -0.434* (0.256) 0.081 (0.172) -0.491 (0.120) FIT Level -0.275 (3.914) 1.483*** (0.362) 6.107* (2.360) 1.447 (2.248) 1.988* (1.090) Public R&D 0.014*** (0.004) 0.010*** (0.003) 0.001 (0.013) 0.012*** (0.004) 0.005*** (0.001) Control variables: Electricity Price -0.883* (0.425) -0.686** (0.348) -1.321* (0.804) -0.395 (0.427) -0.848*** (0.298) Electricity Consumption 0.689 (1.710) -1.572 (1.316) 0.757 (2.827) 0.653 (1.551) -0.542 (1.159) GDP -0.051* (0.028) -0.021 (0.021) -0.030 (0.041) -0.034 (0.023) -0.027 (0.017) Kyoto 0.762*** (0.241) 0.523** (0.265) -0.408 (0.501) -0.030 (0.254) 0.558*** (0.160) ETS System 1.186*** (0.175) 0.567*** (0.160) 0.818** (0.350) 1.175*** (0.194) 0.851*** (0.144) Total EPO Fillings 0.185***
(0.046) 0.037 (0.034) -0.183** (0.084) 0.054 (0.040) 0.096*** (0.030) Observations 209 209 217 208 210 Log-Likelihood -483.65 -545.09 -258.79 -503.32 -738.07 χ2 1275.40 2126.31 464.65 1035.86 2397.74 (P > χ2) 0.00 0.00 0.00 0.00 0.00
Notes: 1. Robust standard errors in parentheses. 2. *** denotes significant at the 1% confidence level; **, 5%; *, 10%.
3. All policy variables are included with a two-year lag. 4. Country fixed effects included.
27
To summarize, each type of policy instruments influences patenting in every technology area, in addition to patenting in renewable energy overall, differently. Moreover, the results of the baseline model suggest that especially tax incentives and public R&D have a significant impact on the number of patents in almost all technology areas.
5.1.2 Results other explanatory variables
With respect to the other explanatory variables (e.g. control variables), the results suggest that the price of electricity has a significant negative effect on innovation in wind, solar, ocean and on innovation in renewables overall. This is an unexpected result. Electricity consumption, on the other hand, does not seem to influence innovation in any of the technology areas; all of the coefficient are insignificant. This result is similar for GDP growth as the estimate is only significant for the technology wind. The results, furthermore, indicate that the general propensity to patent (total EPO filings) only significantly influences innovation in energy from wind, biomass & waste and all renewables. The sign of the coefficient for biomass & waste, however, is negative. In addition, the two control policy variables Kyoto and ETS system seem to have a significant positive influence on the number of patent in almost all technology areas. Except for the effect of the Kyoto treaty on patents in ocean and biomass & waste, all other coefficients are significantly positive.
5.2 Results instrumental variable approach
5.2.1 Results endogenous treatment effect model
Table 3 reports the estimates for the endogenous treatment effects model (equation (2) – (4)). Before turning to the results of the different policy variables, this part first discusses the power of the instruments. Table 3 also includes the results of this first stage regression.
5.2.1.1 Results first stage regression
The results in table 3 indicate that the instruments used in the first stage (equation (3)) are most of the time significant (using a 1% significance level), suggesting that the instruments are strong. However, not every instrument is significant at a 1% significance level. The instrument used for the policy variable REC target (i.e. tradable certificate system) is only significant at a 5% confidence level. This also holds for the instrument used for obligations in the technology area solar and ocean and holds for the instrument used for REC percentage in the case of renewable energy overall. Furthermore, this result is similar for the variable FIT level (i.e. stringency measure feed-in tariff) for the technology wind. Moreover, the instrument used for the variable REC percentage is in each technology area only significant at a 10% significance level.
