Public Research and Development Funding for Renewable Energy Technologies in Europe: A Cross-Country Analysis

Abstract

The European Union recognizes the vital role that innovation in renewable energy technologies plays in accelerating the energy transition. In the largest ever transnational research and innovation program, Horizon Europe (2021–2027), the EU allocated 35% of the total budget of €95.5 billion to green technology research. This paper examines public research and development (R&D) funding for renewable energy technologies in 17 European countries from 2000 to 2020 by analyzing its extent, composition, and effectiveness. While large economies lead in the size of total available public R&D support, Nordic countries lead when it comes to available public R&D funding relative to the sizes of their respective economies. Thereby, the share of the European Union’s funding within the total public R&D support available in each country is vastly heterogeneous across countries, ranging from 13% to 63%. Furthermore, based on detailed and recent data, the effectiveness of public R&D funding is estimated through a Negative Binomial Regression model with fixed effects. Overall, public R&D funding is confirmed as an effective driver of green innovation. Like in previous studies, limitations stem from restricted data availability and temporal uncertainty of innovation. These limitations are addressed, which shall incentivize future research and policymaking.

1. Introduction

The European Union (EU) has embarked on a committed path toward decarbonization. By 2030, it aims to reduce greenhouse gas emissions by at least 55% (compared to 1990 levels) [1]. Climate neutrality should be reached by 2050. Currently, more than 75% of the greenhouse gas emissions in the EU stem from the energy sector. For this reason, the EU aims to increase the share of energy produced by renewable energy (RE) technologies to at least 32% by 2030 (compared to 18% in 2018) [2].

Since 2000, the consumption of energy produced by renewable energy sources has increased significantly across the member states of the European Union [2]. The increased diffusion of renewable energy sources is found to have had a positive impact on economic growth and on the reduction of both conventional energy consumption and greenhouse gas emissions [2,3]. Thereby, the development of renewable energies differs significantly among countries [4], as well as among different renewable energy sources [5], being partly determined by the distribution of energy sources within countries in the late 1990s, such as whether countries possessed their own fossil fuel sources [6]. Public policies, especially subsidies and incentives, were found to represent another important factor that determined and contributed to the diffusion of renewable energy sources across the European Union [7].

This paper will focus on the role of one public policy in particular, namely, public research and development (R&D) expenditures that target the development of renewable energy technologies. Despite the rapid diffusion of renewable energy sources, the International Energy Agency (IEA) emphasizes the necessity of strengthened engagement of the EU to achieve its 32% target. One of IEA’s recommendations entails accelerating innovation and technology deployment [8]. Recently, the IEA called more explicitly for a ‘new wave of innovations’ in the RE sector [9]. This paper backs up the IEA’s call with current and detailed evidence. It provides a thorough cross-country analysis of the extent, composition, and general effectiveness of public R&D funding for RE in Europe.

The broad recognition by the EU of the importance of public R&D funding for renewable energy technologies indicates the further significance of the analysis carried out in this research. Relevant publications include the EU’s strategy for an Energy Union, launched in 2015. Therein, R&D is described as an essential pillar for the energy transition and for securing competitiveness in providing clean energy [10]. The Strategic Energy Technology Plans (SET-Plans) endorse collaborative R&D engagement. The SET-Plans are led by a Steering Group consisting of representatives from the EU member states and Norway (included in this analysis), Turkey, Island and Switzerland [11,12]. The National Energy and Climate Plans (NECPs) shape the EU’s energy sector governance and ensure that the EU meets its climate and energy targets. All member states were obliged to submit a national energy and climate plan to the European Commission (EC) (by December 31st 2019, although nearly all NECPs had been submitted by the end of May 2020). While the implementation of the NECPs only started in 2020, the plans also address the necessity to align state research and development activities, particularly those that target RE technologies [8,13]. Furthermore, in support of the more recently published European Green Deal [14], the European Commission’s research and innovation program ‘Horizon Europe’ (2021–2027) forms a powerful instrument. With a total budget of €95.5 billion, it is the largest ever transnational research and innovation program. More than 35% of the funding will be allocated to address climate change. The need to mobilize research related to RE sources is explicitly recognized [14]. Finally, in December 2015, the European Union and twenty-four governments signed the Mission Innovation (MI) initiative in conjunction with the Paris Agreement and committed to double public R&D support for clean energy technologies until 2020 [15].

Existing contributions have already analyzed the extent of European public R&D funding for both energy projects in general as well as renewable energy technologies [12,16,17,18,19]. Bointner, Pezzutto and Sparber 2016 [16] analyzed energy R&D expenditures from EU member states and from the European Commission (EC) for 1974 until 2012 and project spending scenarios for later periods. Their analysis compared the RE sector and energy efficiency topics to other sectors such as nuclear energy. Thereby, the authors asserted the growing importance of energy efficiency and renewable energy in the EC’s R&D spending over time. This is also reflected in future projections: the authors’ estimated scenarios predicted a sharp shift in priorities of public R&D funding towards RE technologies. In addition, the authors estimated cumulative energy knowledge stocks induced by public R&D expenditures as a measure of the economic value of knowledge that accounts for spillover effects and depreciation rates. For 2013, these estimates for the energy knowledge stock induced by public R&D funding amounted to 36 billion euros.

For the case of R&D spending that targets renewable energy technologies, Bointner, Pezzutto, Grilli and Sparber 2016 [17] compared, similar to the present paper, the European Commission’s funding to national budgets. Analyzing R&D expenditures from 1987 until 2013, the authors outlined that R&D funding from the European Commission was more equally distributed across different RE technologies than national budgets. Furthermore, they showed that RE R&D funding stemmed, to a far larger extent, from national budgets rather than EC budgets. For both sources, the authors assumed that public RE R&D expenditures would increase after 2013.

Restricting their analysis to solar energy, De Negri, Pezzutto, Gantioler, Moser and Sparber 2020 [12] described, amongst other things, how national R&D budgets are concentrated on a few member states. Similar to the overall picture for RE technologies, the authors found that EC funding for photovoltaics in Europe had steadily increased over time. However, national funding still outweighed EC contributions considerably [12,19].

Pezzutto, Mosannenzadeh, Grilli and Sparber 2016 [18] focused on the European Union’s support for ‘smart cities’. Their study shed light on vast discrepancies in the distribution of European Union funding. For example, it showed that until 2014, only 3% of the EC’s R&D contributions for energy projects had addressed the topic smart cities.

In existing analyses, it has not been possible to include the European Commission’s RE R&D funding on a country-level. This limitation for studying the distribution of the European Commission’s funding across countries is critical, for example, considering the potential of public R&D funding to drive green innovation in countries and considering the EU’s pledged recognition of the importance of R&D regarding renewable energy.

The main contributions of this paper, relative to the aforementioned studies, can be summarized as follows. This paper attempts to fill the gap in terms of country-level data on the European Commission’s RE R&D funding. Data on public RE R&D funding have been collected on a country level for 17 European countries and over a period of 21 years (2000–2020). Due to the restricted availability of data, the analysis presented herein is restricted to the 17 European countries that possess of the largest national R&D budgets for RE technologies. For the first time (to the authors’ knowledge), and in contrast to other sources, data also include country-specific public R&D support from the European Commission that targets renewable energy technologies. Thus, this paper describes public R&D funding issued by both national governments and the European Commission, over time and on a country level. These data provide recent and detailed insights on a country-level to existing findings (e.g., those of Pezzutto, Grilli and Sparber 2016 [17]): the importance of the European Commission’s R&D funding relative to national budgets will be analyzed and compared across countries. Further, the geographical heterogeneity of the distribution of most recent EC funding is shown on a detailed regional level. Finally, public RE R&D funding is described across countries over time. This adds country-level insights to evidence of the development and steady increase of public RE R&D financing, as provided by Bointner, Pezzutto and Sparber 2016 [16].

Furthermore, this paper goes beyond descriptive statistics and aims to contribute to existing evidence on the general effectiveness of public R&D support in terms of driving green innovation. Johnstone, Haščič, and Popp’s 2010 [20] data cover 25 countries and the period of 1978–2003. Their frequently cited contribution outlines two main findings to which this research will contribute: (a) R&D support drives innovation in RE technologies effectively; and (b) the effectiveness of public R&D funding is heterogenous across RE technologies. Previous studies have confirmed these two findings: Marques and Fuinhas 2012 [7] confirmed (a) and (b) when analyzing the impact of RE policies (including public R&D subsidies) on the share of RE in the total energy supply, as a measure for RE development. Lee and Lee 2013 [21] again confirmed (a) and (b) and added that customized policies are required to foster innovation in specific RE technologies. Restricting their analysis to the biofuels sector, Costantini et al. 2015 [22] confirmed (a) and (b), but also found that public R&D support drives innovation only in more advanced technologies. Pitelis 2018 [23] and Pitelis et al. 2020 [24] confirmed findings (a) and (b) as well, emphasizing, however, the importance of the chosen lag structures. Other studies compared the effectiveness of domestic and foreign public policies (including public R&D support) for RE innovation, for the case of solar photovoltaic modules [25] and wind energy [26], as well as for varying levels of market competition [27].

The econometric approach adopted in this paper to evaluate the effectiveness of public R&D support for RE technologies is in line with existing contributions [20,22,23,24]. Namely, the effectiveness of public R&D funding as a driver of green innovation is analyzed through a Negative Binomial Regression Model with fixed effects. Thereby, its effectiveness is estimated considering the stringency of other environmental policy measures that have been implemented in various countries over time, as well as countries’ overall innovative environmental policies, energy consumption and electricity prices. This paper contributes to similar existing studies and scrutinizes the validity of findings (a) and (b) in several ways. Detailed data are required to adequately capture investments and innovations that are environmentally friendly [28]. This research will examine the validity of findings (a) and (b) based on a uniquely detailed panel dataset of environmentally friendly public (national plus EC’s) R&D support on a national level. Furthermore, the data employed in the estimations cover the period of 2000–2015, which makes it possible to evaluate the validity of existing findings on the effectiveness of R&D for very recent public R&D funding. Finally, this paper contributes by formulating a clear outline of the limitations it shares with similar studies and by providing recommendations for future research and policy makers that address these limitations.

In this vein, the paper addresses the following three research questions:

• What was the extent of European public research and development funding for RE technologies in 2000–2020 across EU countries?
• Within public RE R&D support, what was the importance of the European Commission’s funding relative to national budgets in 2000–2020?
• Has recent European public research and development funding generally been effective as a driver of knowledge and innovation in RE technologies?

Section 2 describes the materials and methods. Results that aim to answer the present research questions are presented in Section 3. Section 4 discusses the results obtained, limitations of this study, and provides policy recommendations. Section 5 concludes the paper.

In a nutshell, addressing research question 1, the descriptive analysis of this paper revealed that the most public funding for RE R&D is available in the largest economies. Nordic countries lead relative to the size of their respective economies. Thereby, until 2011, both EC and national budgets increased over time in all countries (as described on a general rather than country-specific level in [16,17]). From 2011 onwards, however, in most countries, the total amount of available public RE R&D support remained relatively stable. Thereby, overall, total EC expenditures contributed to stabilizing decreasing national budgets from 2011 onwards.

Addressing research question 2, it will be shown that the relative importance of EC funding varies heavily across countries. It ranges, for example, from 63% in Belgium to 15% in France. Economically strong regions profited significantly more from recent EC funding for RE R&D.

Addressing research question 3, this paper confirms findings (a) and (b): based on detailed country-level data and for the period of 2000–2015, the Negative Binomial Regression Model with fixed effects confirmed both the general effectiveness of public RE R&D funding and the heterogeneity of its effect across sectors.

2. Materials and Methods

The first and second research questions, which concern the extent and composition of public R&D funding for RE technologies, are addressed descriptively. The analysis of research question 3, which concerns the general effectiveness of R&D funding as a driver of green innovation, is based on a panel dataset that comprises information for 17 European countries and 19 years (2000–2018). The countries included are Austria (AT), Belgium (BE), Czech Republic (CZ), Germany (DE), Denmark (DK), Spain (ES), Finland (FI), France (FR), Hungary (HU), Ireland (IE), Italy (IT), The Netherlands (NL), Norway (NO), Poland (PL), Portugal (PT), Sweden (SE), Slovakia (SK). The choice of these countries was based on two factors. First, data on public R&D funding and control variables were only available for these 17 countries to a sufficiently detailed extent to carry out the present analysis. Second, these countries are those with the largest national RE R&D budgets in Europe, and therefore, were of key interest.

Sixteen of the 17 countries included are member states of the European Union. Norway was added to the analysis due to the significant amount of public RE R&D funding the country received from the European Commission and due to its membership in the Steering Group of the Strategic Energy Technology Plans. Based on this panel dataset (descriptive statistics can be found in Appendix A,

Table A1. Descriptive statistics (for the period included in the estimation: 2000–2015).

Variable	Obs.	Mean	Std. dev.	Min.	Max.
Patents RE	272	86.899	177.944	0	1282.920
Total public R&D expenditures for RE in Mill. Euro	272	55.081	70.599	0.234	372.022
Total public R&D expenditures for RE in % of GDP	272	0.040	0.033	0.001	0.177
Energy consumption	272	96.588	5.625	80.900	113.300
Electricity prices	272	0.126	0.043	0.052	0.235
Feed-in Tariffs	272	2.003	1.912	0	6.000
Standards	272	4.019	1.338	1	6.627
Taxes	272	1.733	0.806	0.637	4.335
Trading Schemes	272	1.497	1.248	0	5.281
Market-based instruments	272	1.744	0.754	0.250	3.433
R&D Stringency	272	2.578	1.460	0.919	6.728
Patents RE Knowledge Stock	272	476.128	1026.011	2.350	6716.002
Patents Solar Energy	272	42.733	105.462	0	762.090
Patents Wind Energy	272	26.661	58.979	0	349.000
Patents Biomass	272	8.632	13.310	0	81.790
Patents Ocean Energy	272	3.721	5.670	0	38.000
Patents Geothermal Energy	272	1.320	3.195	0	27.000
Patents Hydroelectricity	272	3.828	6.896	0	59.830
Total public R&D for Solar Energy in Mill. Euro	272	20.858	30.179	0	140.600
Total public R&D for Solar Energy in % of GDP	272	0.011	0.008	0	0.054
Total public R&D for Wind Energy in Mill. Euro	272	8.205	12.734	0	75.941
Total public R&D for Wind Energy in % of GDP	272	0.007	0.011	0	0.073
Total public R&D for Biomass in Mill. Euro	272	17.245	22.575	0	140.532
Total public R&D for Biomass in % of GDP	272	0.016	0.018	0	0.103
Total public R&D for Ocean Energy in Mill. Euro	272	1.366	2.755	0	15.834
Total public R&D for Ocean Energy in % of GDP	272	0.002	0.004	0	0.032
Total public R&D for Geothermal Energy in Mill. Euro	272	2.002	4.585	0	24.706
Total public R&D for Geothermal Energy in % of GDP	272	0.001	0.001	0	0.013
Total public R&D for Hydroelectricity in Mill. Euro	272	1.273	2.347	0	12.731
Total public R&D for Hydroelectricity in % of GDP	272	0.001	0.003	0	0.029
GDP Deflator	272	145,603.8	173,140.7	5418.9	745,226.0

), the following Equation (1) was created. It builds on existing contributions [20,22,23] and aims at determining the general effectiveness of public RE R&D funding:

$$\begin{array}{l}({\mathrm{patents} \mathrm{RE}}_{\mathrm{i},\mathrm{t}+1})={ ß}_0+{ ß}_1\left(\mathrm{public} \mathrm{R}\&{\mathrm{D} \mathrm{expenditures}}_{\mathrm{i},\mathrm{t}}\right)\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{ß}_2\left({\mathrm{energy} \mathrm{consumption}}_{\mathrm{i},\mathrm{t}}\right)+{ß}_3({\mathrm{electricity} \mathrm{price}}_{\mathrm{i},\mathrm{t}})\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{ß}_4({\mathrm{patents} \mathrm{all} \mathrm{technologies}}_{\mathrm{i},\mathrm{t}})+{ß}_5(\mathrm{feed}\_{\mathrm{in} \mathrm{tariffs}}_{\mathrm{i},\mathrm{t}})\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{ß}_6({\mathrm{standards}}_{\mathrm{i},\mathrm{t}})+{ß}_7({\mathrm{taxes}}_{\mathrm{i},\mathrm{t}})+{ß}_8({\mathrm{trading} \mathrm{schemes}}_{\mathrm{i},\mathrm{t}})\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{\mathsf{\alpha }}_{\mathrm{i}}+{\epsilon }_{\mathrm{i},\mathrm{t}}\end{array}$$

(1)

The dependent variable that captures innovation output is the country-specific number of patents for renewable energy technologies, lagged by one year. i = 1, …, N represent indexes for countries and t = 2000, …, 2015 represent time indexes. The regressor that is of main interest consists of ‘public R&D expenditures’ that directly target the development of renewable energies, measured in percentages of GDP. The estimation includes the additional covariates ‘energy consumption’, ‘electricity prices’ and ‘patents all technologies’, as well as data on the environmental stringency of other public policies than R&D expenditures through the covariates ‘feed-in tariffs’, ‘standards’, ‘taxes’ and ‘trading schemes’. Time-invariant country fixed effects are captured by ${\mathsf{\alpha }}_{\mathrm{i}}$.

2.1. Public Research and Development Funding

Data on total public R&D funding were collected as accumulated expenditures from two sources: the national budgets of individual countries and the European Commission’s budget.

National R&D budgets are accessible through the International Energy Agency’s (IEA) data browser. The ‘Detailed country RD&D budgets’ report contains the required country- and sector-specific data [29], expressed in millions of Euros and real values (2020 prices and exchange rates), which is why inflationary adjustment was not necessary. Therefore, only funding within ‘Group 3: Renewable Energy Sources’ was considered [30].

Data on the European Commission’s (EC) RE R&D funding were accessed through the Community Research and Development Information Service (CORDIS) [31]. CORDIS provides detailed information on the EU’s framework programs for research and technological development (FPs). Approximately 85% of the R&D investments for the renewable energy sector from the EU have been issued through the EC’s Framework Programmes (FPs) [12,17]. Therefore, and to avoid the risk of double-counting, the present analysis is limited to these FPs, in particular, to FP5 (1998–2002), FP6 (2002–2006), FP7 (2007–2013) and Horizon 2020 (2014–2020). Only for these FPs were the most recent framework program data available to a sufficiently detailed extent.

The analysis presented in this paper is the result of an extensive data collection effort, consisting of three main steps. In a first step, the FP-specific datasets accessed through CORDIS were attributed to the RE-sector and subsectors through a detailed, automated keyword search. Each project’s title and objective description were automatically skimmed for matches with a list of 99 keywords. A keyword method could be biased. It might include irrelevant projects or exclude relevant ones [20]. However, the keyword search method applied is likely reliable. The risk of excluding relevant projects has been mitigated through a long list of different keywords (nearly a hundred), which could also match parts of terms used in the project titles or descriptions. The risk of including irrelevant projects has been mitigated through a manual examination of descriptions of matched projects’ abstracts. In a second step, the categorized projects were attributed to countries. This was done through another set of FP-specific data accessed through CORDIS. Separate files contain information on the precise amounts each country received through its participation in EC R&D projects. Based on the unique project ID numbers, this information on how much R&D funding each country received within each relevant EC project was merged with the previously described attribution of projects to the RE sector that was obtained from the keyword method. The project-level data were then accumulated on a country- and year-level, depending on the project start dates. In a third and final step, the raw data on EC R&D funding, which were expressed in nominal values, were adjusted for inflation. This adjustment was carried out through ‘Harmonised Indices of Consumer Prices’ (HICP), published by the European Statistical Office [32]. HICP provide comparable measures for inflation across Europe, both on a country level and over time. The year 2020 was used as a reference year.

2.2. Private Research and Development Funding

Private R&D funding from businesses and other private sources, such as philanthropic organizations, plays an essential role in the total share of R&D funding for renewable energy technologies. For solar energy, private R&D funding estimates range from 60% to 70% of total R&D funding [12]. For 2011, the European Commission provided comparable sector-specific figures and estimated that in 2011, 55% of total R&D funding for solar energy technologies stemmed from the private sector [33]. For Italy, survey data on renewable energy R&D funding from the private sector were exceptionally available for 2013–2018: approximately 20–40% of total R&D funding for RE technologies was estimated to stem from the private sector [30]. A recent IEA report [34] highlights how companies which are active in renewable energy technologies intensified their R&D efforts more strongly than other firms in the energy technology sector. Between 2010 and 2019, global private R&D expenditures for RE technologies increased by approximately 74%.

Collecting private R&D data for the present analysis proved to be an impossible task. Not only is the availability of data limited to a few specific RE technologies, but even if technology-specific data were fully available, the complex structure and distribution of companies across the various legal entities that are in place in different countries would make a precise attribution of private R&D investments challenging or even impossible [17].

2.3. Patents in Renewable Energy Technologies

Data on the dependent variable ‘patents RE’ were extracted from the OECD’s Environment Database for Technology Development [35]. The OECD’s patent statistics were constructed using algorithms and avoiding double counting. The data were sorted by year in which the patent was filed (being the closest moment to the actual invention) and by the inventor’s country of origin. The latter makes it more likely that patents can be traced back to policy measures in a country (such as public R&D support). Note that because patents on biofuels are listed separately in the OECD’s Database, they had to be added to patents on renewable energy generation.

In line with existing literature [20,22], the dependent variable is treated with a minimal lag of plus one year. Using lags takes into account the fact that innovation needs time, and it reduces possible endogeneity (such as if the number of patents determined the amount of public R&D funding available, and not merely the other way around) [22]. The existing literature supports the use of a small time-lag and suggests that a statistically significant effect of R&D on patenting occurs in the first and, to a smaller extent, in the second year, but not later [36]. Several authors go further and neglect any substantial lag between public R&D support and patent applications [25,37,38].

For different reasons, patent data are not necessarily a reliable measure of innovation. First, the filing of a patent does not necessarily reflect technology adoption. Indeed, most patents have little commercial value, and the adoption of the invention is often not widespread [39]. This research focuses on sets of patents that were signed in at least two jurisdictions, which corresponds to a ‘family size’ equal to two or greater. Thus, it was possible to restrict the analysis to higher quality patents only [40]. Not only would it be unnecessarily costly to file patents for worthless inventions in more than one country [41]. Also, evidence suggests that the family size of patents is, in general, strongly correlated with the economic value of an invention [42].

Furthermore, not all inventions are protected by patents. For example, a firm might prefer to keep its invention secret [43]. While this highlights how patent data are not a perfect measure, other advantages still support their use. Griliches et al. 1990 [44] emphasized the strong relationship between private R&D spending (as a measure for innovative activity) and patents. Also, patent statistics permit a disaggregation to a detailed country and technology level [39], an advantage that holds for this research.

2.4. Additional Covariates

Another argument against the use of patent data stems from differences between countries’ patenting regimes and propensity to patent. In one jurisdiction, a single patent might be sufficient to protect an invention. In other countries, several patents for the same level of protection might be necessary [20]. The covariate on patents in all technologies, retrieved from the OECD’s 2021 Environment Database for Technology Development [35], accounts for this. The variable captures differences and changes over time in countries’ innovative environments and propensity to patent.

Energy consumption and electricity prices represent additional covariates. Data on these variables were retrieved from the Eurostat’s Database [32], which provides datasets on country-specific ‘electricity prices for domestic consumers’ and ‘final energy consumption’. To avoid double-counting policy measures, electricity prices exclude taxes and levies. In addition, to permit a cross-country comparison, prices are expressed in terms of Purchasing Power Standards. Prices of fossil fuels as alternative factor inputs are an essential determinant of innovation in renewable energy technologies sources. Higher prices for electricity produced by fossil fuels plausibly incentivize the adoption of and innovation in renewable energy technologies. The size of demand represents another critical driver of innovation and is measured through electricity consumption. In growing markets where demand is large, RE innovation is incentivized because it is easier to compensate for initial investment costs. [20].

Indices for the policy measures feed-in tariffs, standards, taxes and trading schemes control for changes in other policies that affect green innovation as well. The OECD’s (2016) environment statistics database [45] provides internationally comparable and country-specific Environmental Policy Stringency (EPS) indices. The environmental stringency of various policy measures is determined by their effect on the explicit or implicit price of environmentally harmful behavior. The OECD environment statistics database provide an EPS index for public research and development funding as well. The OECD’s definition of ‘public’, however, is the same as IEA’s: it excludes EC sources. Contrary to existing research, this study includes EC sources.

2.5. Econometric Approach

To examine whether public R&D funding was correlated with RE patenting, even after controlling for other factors and policy measures introduced, the previously described Equation (1) was estimated as a fixed effects Negative Binomial Regression Model (NBRM). The estimation included all 17 countries subject to this analysis but was limited to the period of 2000–2015. Although public R&D data would have been available until 2020, limited availability of patent data and the evaluation of different lag-structures made it necessary to restrict the main analysis to 2000–2015.

In line with existing similar literature [20,22,24], an NBRM was chosen. This choice was justified as follows, and partly based on the underlying assumptions of the model. An NBRM accounts for the possibility of values of the dependent variable having a lower bound at zero, while other models (such as Ordinary Least Square-OLS models) may lead to negative results that make no sense for the non-negative dependent variable [46]. Furthermore, while the NBRM builds on the Poisson Regression Model (PRM), the NBRM, in contrast to the PRM, does not exclude so-called ‘conditional overdispersion’ of the dependent variable by assumption. More specifically, the PRM assumes conditional equidispersion $(\mathrm{Var}({\mathrm{y}}_{\mathrm{i}}|{\mathrm{x}}_{\mathrm{i}})=\mathrm{E}({\mathrm{y}}_{\mathrm{i}}|{\mathrm{x}}_{\mathrm{i}})):$ the conditional variance of ${\mathrm{y}}_{\mathrm{i}}$ is assumed to be equal to the conditional mean. Conditional overdispersion violates this assumption, since it entails that after accounting for all predictors, the variance of the number of patents changes dependent on whether the mean of public R&D funding is high or low within a country. In other words, all else equal, in years in which public R&D support is high in a country, RE patents in that country would vary more heavily than in years in which the country’s public R&D support was low. This is plausible to expect, for example, due to the uncertainty innovation processes inhibit. The presence of conditional overdispersion would not bias PRM estimates in their sizes, but would bias standard errors and p-values downward [22]. In contrast, the NBRM is flexible enough to account for conditional overdispersion. More in detail, in an NBRM, it is assumed that the dependent variable follows a negative binomial distribution. Like in the PRM, the dependent variable is still assumed to follow a Poisson process. The difference to the PRM is that the NBRM introduces the possibility of heterogeneity in the variance of patents. It does so through the introduction of an unobserved error parameter η. More precisely, in this study, the variance of the dependent variable is specified as $\mathrm{Var}\left({\mathrm{y}}_{\mathrm{i}}|{\mathrm{x}}_{\mathrm{i}}\right)=\mathrm{E}({\mathrm{y}}_{\mathrm{i}}|{\mathrm{x}}_{\mathrm{i}})\times \left(1+\mathsf{\eta }\right)$, whereby $\mathsf{\eta }$ is positive. Compared to a PRM, where $\mathrm{Var}\left({\mathrm{y}}_{\mathrm{i}}|{\mathrm{x}}_{\mathrm{i}}\right)=\mathrm{E}({\mathrm{y}}_{\mathrm{i}}|{\mathrm{x}}_{\mathrm{i}})$ must be true, the specification $\mathrm{Var}\left({\mathrm{y}}_{\mathrm{i}}|{\mathrm{x}}_{\mathrm{i}}\right)=\mathrm{E}({\mathrm{y}}_{\mathrm{i}}|{\mathrm{x}}_{\mathrm{i}})\times \left(1+\mathsf{\eta }\right)$ is more flexible, given that $\mathsf{\eta }$ can take on any positive value. This inherent flexibility is one reason for which an NBRM is chosen over a PRM [46].

As a third justification of the NBRM,

Table A2. Negative binomial regression estimates for the effect of public research and development funding on innovation in renewable energy technologies compared to Poisson regression estimates and Ordinary Least Square regression estimates (Equation (1)) (dependent variable: patents in the renewable energy sector (lag 1)).

	(1)	(2)	(3)
Variable	NBRM 2000–2015	PRM 2000–2015	OLS 2000–2015
Total public RE R&D in% of GDP	6.058 ***	8.701 ***	832.800 *
	(0.00)	(0.00)	(0.05)
Electricity prices	4.292 **	−0.058	−263.800
	(0.00)	(0.98)	(0.18)
Energy consumption	0.035 ***	0.021	−0.418
	(0.00)	(0.23)	(0.74)
Patents all technologies	0.000	0.000 ***	0.136 ***
	(0.70)	(0.00)	(0.00)
Feed−in tariffs	0.044	−0.021	−9.674
	(0.08)	(0.62)	(0.26)
Standards	0.153 ***	0.158 ***	−1.098
	(0.00)	(0.00)	(0.84)
Taxes	−0.034	−0.093	−14.670
	(0.66)	(0.25)	(0.48)
Trading schemes	0.158 ***	0.140 ***	5.817
	(0.00)	(0.00)	(0.13)
Constant	−3.376 ***		−373.200 **
	(0.00)		(0.00)
Observations	272	272	272
BIC	2088.7	4701.5	3163

Note. Table A2 compares negative binomial regression estimates (NBRM) for Equation (1) with poisson regression estimates (PRM) and ordinary least square estimates (OLS). All available years (2000−2015) are included. As in Table 1, the dependent variable is the by the OECD constructed number of renewable energy patents per year and country, lagged by one year and restricted to patents of at least family size 2, which excludes low quality patents. The additional covariates include: final energy consumption and electricity prices for domestic consumers (expressed in Purchasing Power Standards) as measures of demand; the number of patents in all technologies (restricted to at least family size 2) as measure of the propensity to patent; OECD indices for the environmental policy stringency of feed−in tariffs, standards, taxes and trading schemes. Estimates for ‘Total public RE R&D’, the variable of main interest, indicate the effect of public R&D support for renewable energy technologies (measured in% of GDP) on patents in the renewable energy sector. The estimates reveal that the finding of a positive and statistically significant effect of public R&D funding is independent of the chosen model. BIC−values as measures for model−fit are displayed at the bottom. p-values are displayed in parentheses: *** Significance at the 0.1 percent level; ** Significance at the 1 percent level; * Significance at the 5 percent level.

in Appendix B displays the estimation results of Equation (1) as an NBRM, a PRM and as an OLS model. Importantly, the Table also provides model-specific values for the Bayesian Information Criterion (BIC) [47], a measure of model fit that is commonly used for model selection [48]. The lowest BIC value is attributed to the NBRM estimation of Equation (1). This again supports the choice of the NBRM.

In summary, the choice of the NBRM was justified by its prevalent use in similar literature, its ability to account for zero truncated values, its flexibility in accounting for conditional overdispersion, as well as its comparably low BIC values.

Notably, the NBRM is combined with a fixed effects method. Controlling for country fixed effects ensures that any change in the dependent variable stems from influences other than time-invariant characteristics of countries. Unobserved characteristics of countries that do not change over time may be correlated with both public R&D expenditures and the innovation outcome. If they are not included in the regression, they may bias estimated coefficients (‘omitted variable bias’). An example would be that the covariate on patents in all technologies is not sufficient to fully capture all differences between countries in innovative environments. In the fixed effects method employed in this paper, any influence from such country-specific factors that do not change over time (fixed effects ${\mathsf{\alpha }}_{\mathrm{i}}$) is eliminated through differencing [49].

3. Results

3.1. National Funding

Across all countries, national R&D expenditures in the RE sector increased from 2000 until 2011 and decreased after that (Figure 1). Norway, the Netherlands and Belgium (and Portugal, where public R&D funding for RE was very low) were the only countries whose national R&D budgets for renewable energy technologies did not decrease between 2011 and 2020. Country specific shares of national RE R&D expenditures summed up for the whole period, instead of being shown over time, are displayed in Figure A1 (Appendix A). As Figure A1 suggests, the following four countries provided more than sixty percent of all R&D funding for renewable energy technologies that stemmed directly from national budgets (for 2000–2020 and among the 17 countries analyzed): Germany (23%), France (21%), Italy (10%) and the Netherlands (9%) lead national R&D investment in RE. Portugal, Belgium and Hungary are among the smallest investors.

3.2. European Commission Funding

Figure 2 summarizes these R&D expenditures from the European Commission. High volatility due to legislative fractionalization is frequent in the realm of public R&D expenditures on new energy technologies [50]. In the case of EC R&D expenditures, high volatility and sudden drops (most clearly visible for the year 2014) stem from transition periods between framework programs (FPs). In the first year of new FPs, less funding is paid out as most projects are still in the application phase.

From 2000 to 2020, the EC’s total R&D contribution for RE technologies increased by a factor of approximately 5. This increase was observed consistently across countries. The largest economies received the largest amounts of R&D funding for renewable energy technologies from the European Union. For example, Germany (DE) received 20% (more than one billion Euros). Country specific shares of the European Commission’s RE R&D expenditures summed up for the whole period, instead of being shown over time, are displayed in Figure A2 (Appendix A). A comparison between EC and national funding is particularly striking for Spain: 16% of all RE R&D expenditures from the European Commission target projects in Spain, while the country’s national contributions amount to only 6% of all contributions of all countries. For the case of Spain, this suggests that EC budgets played a more important role than national budgets in the development of renewable energy technologies in the last 20 years. For France, in contrast, the opposite seems to be the case: France’s share of total RE R&D contributions of all countries is 10 percentage points lower for the European Commission’s contributions (11%) compared to national budgets (21%). The next section will analyze the importance of EC funding relative to national budgets across countries even more in detail.

An even more detailed geographical assessment was possible for the EC’s most recent and largest funding programme included in this analysis ‘Horizon 2020′ (2014–2020). The Horizon 2020 Dashboard [51], an interactive data visualization tool provided by the European Commission, makes it possible to visualize the geographical distribution of the European Commission’s RE R&D funding for selected projects that target RE technologies. The territorial classification follows Eurostat’s official ‘Nomenclature of territorial units for statistics’ (NUTS) [52]. The visualization of the distribution of EC funding was possible across countries (Figure 3) and on NUTS 1 level (across major socio-economic regions) (Figure 4) and NUTS 2 level (across basic local regions for the application of regional policies) (Figure 5). Notably, the maps are not limited to the 17 countries for which data were available to a sufficiently detailed extent to carry out the main estimation.

Figure 3 underlines the insight that most of EC’s R&D funding was concentrated in the largest economies. The largest economies were most active in the EC’s Horizon 2020 projects related to renewable energy technologies: Spain (1.27 k participations), followed by Germany (1.14 k participations), Italy (940), and France (845 participations). In contrast, the participation of EU member states in Eastern Europe was comparatively small, such as for Slovakia (27 participations), Hungary (54 participations), and Czechia (75 participations).

Figure 4 and Figure 5 shed light on the heterogeneous distribution across NUTS 1 and NUTS 2 regions. Like on a country level, on a NUTS 1 level, the distribution of EC Horizon 2020 funding for RE technologies varied geographically, but also strongly across regions within countries. This insight became even more prevalent on a more detailed NUTS 2 level, as shown in Figure 5. For most NUTS 2 regions, the distribution of EC R&D funding did not vary significantly and was uniformly low. However, economically strong NUTS 2 regions, such as Catalonia and Madrid (Spain), Bavaria (Germany), Zuid-Holland (the Netherlands), and Paris and the Rhone-Alps (France), profited considerably more from the EC’s Horizon 2020 R&D funding for renewable energy technologies than most of the other regions. This suggests a high concentration of the European Union’s RE R&D funding in regions that are economically strong (in terms of GDP).

3.3. Total Public Funding

At this point, the first research question (What was the size of European public research and development funding for RE technologies in 2000–2020 across countries?) can be answered. EC expenditures are added to the national budgets to obtain a complete picture. As Figure 6 illustrates, the largest economies by GDP lead in terms of average yearly R&D expenditures for renewable energy technologies (Germany and France, followed by Italy, the Netherlands and Spain). However, the picture differs when countries’ GDP is considered (Figure 7). Nordic countries lead in yearly public R&D funding available relative to GDP. In Denmark and Finland, public R&D funding available for renewable energy technologies (in% of GDP) had more than double the size of Germany or France. When expenses are measured relative to GDP, most large economies with high R&D expenditures move to the middle of the ranking. That is particularly striking for the case of Italy (moves from rank 3 to rank 13, among 17 countries included in the analysis) and Germany (moves from rank 1 to rank 9).

The data obtained in this research can also be used to address research question 2 (Within public RE R&D support, what was the importance of the European Commission’s funding relative to national budgets in 2000–2020 across countries?). Figure 8 sheds light on the vast heterogeneity of the relative importance of EC funding. The European Commission’s contributions as shares of total public RE R&D funding played the most critical role for Portugal (73%) (despite being a country with a meagre RE R&D budget), Belgium (63%) and Spain (46%). In the Slovak Republic (13%), France (15%) and Finland (18%), the share of EC contributions constituted the lowest share of total RE R&D funding compared to other countries. See Appendix A for an illustration of total RE R&D expenditures (EC plus national) by country (Figure A3), and a visualization of the shares of total R&D contributions by country over the whole period (2000–2020) (Figure A4).

Still concerning research questions 1 and 2, Figure 9 displays change over time in total public R&D funding for the development of renewable energy technologies (green line). From 2007 to 2011, total R&D funding increased heavily. One explanation for this strong increase is the growing political awareness of the necessity to support the development of RE technologies at that time. For example, the Renewable Energy Directive 2009/28/EC illustrates this increased political concern. It formed an ambitious and, importantly, legally binding policy that aimed at accelerating the promotion and production of renewable energy from 2009 onwards [23]. For all EU member states, the directive included the legally binding requirement that RE sources should provide 20% of gross final energy consumption and 10% of the energy used in transportation [53].

Furthermore, Figure 9 describes the increasingly important and stabilizing role of EC expenditure for total public R&D funding across countries. The European Commission’s funding (blue line) is added to national expenditures (orange line). The result forms the total public RE R&D expenditures over time (green line). From 2011 onwards, national R&D expenditures decreased or remained stable, while the European Commission’s expenses continued to increase in all countries. Visually, these different trends over time result in both total expenditure curves converging from 2011 onwards. After merging both sources, total public R&D expenditures stabilize. Thus, overall, the EC’s expenditures significantly contributed to stabilizing total public R&D support for renewable energy technologies from 2011 onwards. In Norway, the Netherlands, Belgium and Portugal, the only countries that consistently increased their national budgets over time, available public R&D funding for RE technologies even increased. Hence, for all except those countries, stagnating public R&D funding is driven by decreasing national financing, while EC funding compensates for decreasing national budgets. None of the countries could double available public R&D support between 2015 and 2020, as stated in the Mission Innovation initiative in 2015 [15].

The stabilization of total public R&D funding for renewable energy technologies coincided with a substantial decrease in renewable energy patenting activity (Figure 10). However, a descriptive analysis alone is insufficient to draw any conclusions on the general correlation between R&D expenditures and innovation outcomes. Other policy measures with the same environmental objective were introduced. In addition, countries are vastly different in both R&D expenditures and patenting activity. With regards to the period after 2011, the value of discerning a correlation between high-level stagnating R&D expenditures and decreasing patenting activities only based on descriptive analyses would be questionable. As will be outlined in the Discussion Section 4.1, unobserved drivers, such as industry decline in the solar sector, are very likely to have caused the decrease in patenting for renewable energy technologies in that period. The following analysis shall scrutinize the relationship between public R&D support and patenting activity in the RE sector while going beyond descriptive statistics.

3.4. The Effectiveness of Public R&D Funding

To address research question three (Has recent European public research and development funding been generally effective as a driver of knowledge and innovation in RE technologies?), Equation (1) is estimated through a fixed effect NBRM (

Table 1. Negative binomial regression estimates for the effect of public research and development funding on innovation in renewable energy technologies (Equation (1)) (dependent variable: patents in the renewable energy sector (lag 1)).

	(1)	(2)	(3)	(4)	(5)
Variable	2000–2015	2000–2015	2000–2015	2000–2011	2012–2015
Total public RE R&D in % of GDP	12.660 ***	12.200 ***	6.058 ***	5.825 ***	−0.267
	(0.00)	(0.00)	(0.00)	(0.00)	(0.89)
Electricity prices		7.509 ***	4.292 **	8.160 ***	−7.074
		(0.00)	(0.00)	(0.00)	(0.15)
Energy consumption		0.033 ***	0.035 ***	0.020 *	0.024
		(0.00)	(0.00)	(0.02)	(0.07)
Patents all technologies		0.000	0.000	−0.000	0.000 **
		(0.23)	(0.70)	(0.29)	(0.00)
Feed−in tariffs			0.045	0.059 *	−0.085
			(0.08)	(0.04)	(0.11)
Standards			0.153 ***	0.335 ***	−0.029
			(0.00)	(0.00)	(0.64)
Taxes			−0.034	0.007	−0.075
			(0.66)	(0.94)	(0.54)
Trading schemes			0.158 ***	0.094 *	−0.038
			(0.00)	(0.01)	(0.53)
Constant	0.626 ***	−3.373 ***	−3.376 ***	−2.597 **	2.471
	(0.00)	(0.00)	(0.00)	(0.00)	(0.11)
Observations	272	272	272	204	68
BIC	2170.3	2145.5	2088.7	1473.6	382.6

Note. Table 1 shows negative binomial regression estimates for Equation (1). The first row displays estimates for the effect of public R&D support for renewable energy technologies (measured in% of GDP) on patents in the renewable energy sector. The dependent variable is the by the OECD constructed number of renewable energy patents per year and country, lagged by one year and restricted to patents of at least family size 2, which excludes low−quality patents. The additional covariates include: final energy consumption and electricity prices for domestic consumers (expressed in Purchasing Power Standards) as measures of demand; the number of patents in all technologies (restricted to at least family size 2) as measure of the propensity to patent; OECD indices for the environmental policy stringency of feed−in tariffs, standards, taxes and trading schemes. Columns 1, 2 and 3 rely on all available years (2000−2015). Column 2 excludes alternative policy measures. These are included in column 3, which displays a positive and statistically significant association of public R&D funding with patenting. Columns 4 and 5 differ from column 3 only in the periods under scrutiny. They show that the positive and statistically significant association between public R&D support and patenting can only be confirmed until 2011, but not for the period of 2012–2015 in isolation. BIC−values as measures for model−fit are displayed at the bottom. p-values are displayed in parentheses: *** Significance at the 0.1 percent level; ** Significance at the 1 percent level; * Significance at the 5 percent level.

) for 17 countries and the period of 2000–2015. Thereby, different control variables that capture other environmental policy measures, the countries’ innovative environments, and changes in energy consumption and electricity prices are included.

Research question three is answered as follows: For the whole period, public R&D funding (in% of GDP) has a positive and statistically significant effect on RE patenting (which confirms finding (a)). Covariates other than Environmental Policy Stringency (EPS) indices do not affect the R&D estimate in size or statistical significance (compare Columns 1 and 2). The inclusion of EPS indices affects most estimates (Column 3). However, the R&D estimate remains statistically significant. Estimates for EPS indices are statistically significant and positive (except for taxes). A comparison of values for the Bayesian Information Criterion supports the relevance of the inclusion of EPS indices in the model. Despite being a stringent overfitting model test [47], the BIC-value decreases after including EPS indices, indicating a lower penalty related to the inclusion of covariates. The finding of an overall positive and statistically significant effect of public R&D funding on green innovation is confirmed for and independent of whether an NBRM, a Poisson Regression model or an Ordinary Least Square model is chosen (as shown in Appendix B Table A2).

As shown in columns 4 and 5 in Table 1, the present research also accounts for changing trends in public R&D support and patenting after 2011. For the period of 2000–2011, the main results are not strongly affected. Public R&D funding still has a positive and statistically significant effect on RE patenting. However, results for the period of 2012–2015 differ. The estimate for the association between R&D investments and patenting for that short period does no longer result as statistically significant. However, a small within variation and the period of 2012–2015 being very short (only four years and 68 observations) undermine drawing conclusions from a fixed effects method [49]. Furthermore, as will be discussed in the Discussion Section, the results are likely affected by unobserved drivers of patenting activity (e.g., industry decline in the solar sector).

On a sector-specific level (Appendix B,

Table A3. Comparison of NBRM estimates for the effect of public research and development funding on innovation in renewable energy technologies (Equation (1)) between sectors (dependent variable: patents in the renewable energy sector (lag 1)).

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	Biomass	Wind	Solar	Ocean	Geothermal	Hydroelectricity
R&D Biomass	7.906 **
	(0.01)
Electricity prices	4.517	7.996 ***	3.687 *	6.151 *	−3.931	1.934
	(0.05)	(0.00)	(0.03)	(0.01)	(0.42)	(0.46)
Energy consumption	0.029 **	0.024 *	0.044 ***	0.042 ***	0.036	0.016
	(0.00)	(0.02)	(0.00)	(0.00)	(0.05)	(0.20)
Patents all technologies	−0.000	0.000	−0.000	−0.000 ***	0.000 ***	−0.000
	(0.63)	(0.09)	(0.79)	(0.00)	(0.00)	(0.33)
Feed−in tariffs	0.056	0.013	0.056	0.036	0.032	−0.021
	(0.10)	(0.70)	(0.07)	(0.45)	(0.61)	(0.64)
Standards	0.135 **	0.156 ***	0.177 ***	0.209 ***	0.135	0.288 ***
	(0.00)	(0.00)	(0.00)	(0.00)	(0.09)	(0.00)
Taxes	−0.075	−0.055	0.006	−0.118	0.163	−0.212
	(0.43)	(0.61)	(0.95)	(0.31)	(0.40)	(0.12)
Trading schemes	0.225 ***	0.125 **	0.197 ***	0.120 *	0.174 *	0.130 *
	(0.00)	(0.00)	(0.00)	(0.04)	(0.01)	(0.03)
R&D Wind		13.490 *
		(0.03)
R&D Solar			15.080 **
			(0.00)
R&D Ocean				−18.780
				(0.14)
R&D Geothermal					17.130
					(0.77)
R&D Hydroelectricity						−27.300
						(0.23)
Constant	−2.288 *	−2.868 *	−4.672 ***	−3.340 *	−4.173	−0.931
	(0.05)	(0.01)	(0.00)	(0.01)	(0.05)	(0.51)
Observations	272	272	272	272	256	272

Note. Table A3 displays sector−specific negative binomial regression estimates for Equation (1) for the whole period (2000−2015). More specifically, and equally to Table 1 Column 3, this table displays estimates for the effect of public R&D support for renewable energy technologies (measured in % of GDP) on patents in the renewable energy sector (lagged by one year and restricted to at least family size 2, which excludes low quality patents). The additional covariates include: final energy consumption and electricity prices for domestic consumers (expressed in Purchasing Power Standards) as measures of demand; the number of patents in all technologies (restricted to at least family size 2) as measure of the propensity to patent; OECD indices for the environmental policy stringency of feed−in tariffs, standards, taxes and trading schemes. The effectiveness of public R&D funding on patents in the specific RE sectors is estimated separately for all RE sectors (biomass, wind energy, solar energy, ocean energy, geothermal energy, hydroelectricity). The positive and statistically significant effect of public R&D funding is confirmed for all three renewable energy sources which currently provide the most renewable energy (biomass, solar and wind). p-values are displayed in parentheses: *** Significance at the 0.1 percent level; ** Significance at the 1 percent level; * Significance at the 5 percent level.

), the positive and statistically significant effect is confirmed for all three renewable energy sources which provide the most renewable energy (biomass, solar- and wind energy). Across sectors, the size of estimates for public R&D support is heterogeneous, which confirms finding (b). Furthermore, in light of the potential relevance of patent quality, Equation (1) is estimated for patents of various qualities. The statistically significant positive effect of public R&D expenditures on patenting is robust to choices of patent quality (family sizes ranging from one for low to four for very high quality) (Appendix B,

Table A4. Robustness-check of negative binomial regression estimates for the effect of public research and development funding on innovation (Equation (1)) for different patent qualities and temporal treatments of the dependent variable (patents in the renewable energy sector).

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	lag2	lag3	Family Size Two	Family Size One	Family Size Three	Family Size Four
Total public RE R&D	1.480	−3.070 *	6.060 ***	5.980 ***	6.350 ***	5.470 ***
	(0.299)	(0.034)	(0.000)	(0.000)	(0.000)	(0.000)
Electricity prices	2.070	−1.120	4.290 **	4.650 ***	3.510 *	3.120 *
	(0.111)	(0.397)	(0.002)	(0.000)	(0.016)	(0.041)
Energy consumption	0.040 ***	0.040 ***	0.030 ***	0.030 ***	0.030 ***	0.030 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Patents all technologies	0.0000017	0.0000024	0.0000048	0.0000048	0.0000110	0.0000109
	(0.896)	(0.857)	(0.705)	(0.714)	(0.412)	(0.446)
Feed−in tariffs	0.026	0.007	0.045	0.017	0.040	0.033
	(0.293)	(0.773)	(0.077)	(0.469)	(0.126)	(0.217)
Standards	0.170 ***	0.186 ***	0.153 ***	0.115 ***	0.150 ***	0.176 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Taxes	−0.065	−0.080	−0.034	−0.013	−0.080	−0.097
	(0.361)	(0.244)	(0.656)	(0.848)	(0.312)	(0.227)
Trading schemes	0.189 ***	0.163 ***	0.158 ***	0.159 ***	0.150 ***	0.149 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Constant	−3.185 ***	−2.278 ***	−3.376 ***	−2.382 ***	−2.541 **	−2.434 **
	(0.000)	(0.001)	(0.000)	(0.001)	(0.002)	(0.003)
Observations	272	272	272	272	272	272
BIC	2084.1	2104.6	2088.7	2359.1	1902.3	1770.4

Note. Table A4 analyses the importance of the measurement of the dependent variable. It shows negative binomial regression estimates for Equation (1). The first row displays estimates for the effect of public R&D support for renewable energy technologies (measured in% of GDP) on patents in the renewable energy sector. All available years (2000−2015) are included for all columns. The additional covariates include: final energy consumption and electricity prices for domestic consumers (expressed in Purchasing Power Standards) as measures of demand; the number of patents in all technologies (restricted to at least family size 2) as measure of the propensity to patent; OECD indices for the environmental policy stringency of feed−in tariffs, standards, taxes and trading schemes. Importantly, the columns differ in the measurement of the dependent variable. In Columns 1 and 2, the dependent variable is lagged with plus two and plus three years instead of only one. This different temporal treatment affects the public R&D estimates. This also becomes evident in comparison with Column 3, where patents are lagged with only one year. Note that Column 3 here corresponds to Column 3 in Table 1, on which the main results of this study were based. In this vein, the patents in Column 3 were restricted to those with a family size of at least two. This means that those patents were filed in at least two jurisdictions, excluding low−quality patents. Column 4 does not apply this restriction and includes all renewable energy patents in measuring the dependent variable. Columns 5 and 6 are more restrictive in the quality of patents and include only those that are filed in at least three or four jurisdictions. Columns 3, 4, 5 and 6 show that the main results for public R&D do not depend on the quality of patents, approximated by family size. BIC−values as measures for model−fit are displayed at the bottom. p-values are displayed in parentheses: *** Significance at the 0.1 percent level; ** Significance at the 1 percent level; * Significance at the 5 percent level.

). However, as suggested by Table A4 (Appendix B), the result of a statistically significant and positive effect of public R&D support depends on the temporal treatment of the dependent variable. As noted, both theory and existing literature support the choice of a one- to two-year lag treatment (Section 2) [20,22,25,36,37,38]. Still, this dependence of finding (a) on the temporal treatment of the dependent variable limits the persuasiveness of the results, which is why implications of this inconsistency for future research are discussed in the following section.

4. Discussion

The relevance of a thorough analysis of both the extent and the effectiveness of public RE R&D funding in Europe lies in the potential of public R&D funding to drive green innovation. For different reasons, public research and development (R&D) support tackles undersupply from the private side and drives green innovation. First, the future benefits of investments in environmental R&D are typically highly uncertain. Public support de-risks R&D investments, while otherwise, private firms would have to bear these risks exclusively [43]. Second, public R&D support accelerates the process of reaching commercialization for renewable energy technologies [25,54]. For example, public R&D funding creates innovation networks in which knowledge spillovers are mutually beneficial [55]. Third, public R&D funding is justified by conceiving innovation in renewable energy technologies as a positive externality. Knowledge spillovers, from which society as a whole profits, are not reflected in the firms’ prices. Thus, private returns on R&D investment are smaller than social returns. This leads private firms to invest less than would be socially optimal. Public R&D investment tackles this private underinvestment and increases knowledge and green innovation [25,56]. Fourth, the relative competitiveness of renewable energy technologies is at a disadvantage. While the positive externalities of renewable energy are not reflected in its price prices, also the negative externalities of environmentally harmful energy sources are not fully reflected in the prices of these energy sources [25,56]. For example, the net-negative effective carbon price is estimated to amount to minus $3.44/tCO₂ for 2018 and after taking fossil fuel subsidies into account, for the EU member states and other 22 countries [15]. According to the World Bank, an effective carbon price of plus $40–80/tCO₂ would be necessary to meet the targets of the Paris Agreement (World Bank Group 2019) [57]. All these various reasons justify public funding of R&D activities in the renewable energy sector. In sum, public R&D support as an important driver of green innovation de-risks private investment in RE R&D, accelerates innovation and compensates for positive externalities and competitive disadvantages.

The findings of this paper add to the existing evidence that is described in the Introduction Section. As mentioned, Bointner, Pezzutto and Sparber 2016 [16] assert a growing importance of renewable energy within both EC’s R&D spending and national R&D budgets until 2012. Growing expenditures from the European Commission for R&D have been confirmed for the case of solar energy as well [12,19]. Addressing the first two research questions, this paper adds to this existing evidence, shedding light on the heterogeneity of size and development of both national budgets and EC funding for renewable energy technologies across countries and time (Figure 1 and Figure 2). On a country level, this paper confirms an increase in EC and national R&D funding for renewable energy technologies until 2011. From 2012 onwards, however, national budgets decreased or stagnated for most countries, while the increase in EC funding contributed to a stabilization of total public R&D funding available (Figure 9). Furthermore, the present research adds country-level insights to Bointner, Pezzutto, Grilli and Sparber 2016 [17], who found that national budgets dominate total EC funding. Following up on this research, the present paper shows that the relative importance of EC funding within total public R&D funding for renewable energy technologies varies highly between countries (Figure 8), ranging from 63% (for Belgium) to 15% (for France). In addition, the paper depicts the heterogeneous distribution of recent EC funding (2014–2020) and shows that it predominantly targeted economically strong regions (Figure 5).

Addressing research question three, this research confirms findings (a) and (b) of the existing studies [20,22,23,24] described in the Introduction Section: an overall positive and statistically significant effect of public (EC plus national) R&D support on green innovation has been found, whereby the effect is heterogeneous across sectors. This study supports that insight, since our analysis is based on a detailed and recent country-level dataset which includes, to the best knowledge of the authors and contrary to existing studies, EC expenditures.

The importance of temporal treatment and statistically insignificant results for 2012–2015 inclined the authors to be reluctant regarding quantitative interpretations. This provides a basis for the discussion of the four main limitations which this research shares with existing studies before policy recommendations are formulated.

4.1. Limitations of the Study

The first limitation consists in the temporal uncertainty of innovation outcomes. Evidence suggests that in general, most of the effect of R&D spending on patenting occurs in the first and, to a smaller extent, in the second year [36]. The choice of a one-year time lag is also in line with existing contributions to the effectiveness of R&D spending for RE [20,22]. At the same time, a different temporal treatment affects the results, independent of the chosen model. This reflects a challenge identified by Pless et al. 2020 [28], who stress the long and uncertain times between receiving R&D support and the manifestation of measurable innovation outcomes. Pitelis 2018 [23] justifies his reluctance to a quantitative interpretation of estimates with the dependence of results on chosen lag structures. Future research should address this challenge, for example, through collecting data on the different stages of innovation processes. Different levels of technology advancement imply different time spans for R&D activities to produce innovation outcomes.

A second challenge the evaluation of energy innovation policies, including public R&D funding, commonly faces is the reliable quantification of causal effects. Simultaneity- and selection bias are two common issues discussed in Pless et al. 2020 [28]. For the period from 2012 to 2015, R&D estimates are unreliable for additional reasons. The period is very short, exploitable within variation in total public R&D expenditures is small, and sector-specific price drops and general industry declines are also not included in the analysis. These factors are likely to have caused the decrease in patenting activity for that very recent period instead of stable public R&D funding. For example, between 2006 and 2016, the (inflation-adjusted) price of rooftop photovoltaic systems in Germany decreased by approximately 72% [58], which was mainly caused by massive production by Chinese manufacturers [59]. Alternative measures of innovation outcomes such as academic publications may provide additional insights. Indeed, in contrast to patents, the EU’s share of global academic publications in RE sectors remained constant from 2012 onwards [60]. Pless et al. 2020 propose ways in which the design of new energy innovation policies can facilitate future research. For example, the distribution of public R&D support can be determined through grades or rankings which permits the evaluation through ‘Regression Discontinuity Designs’ [61,62,63].

For this research, country-specific data on international and private R&D funding that targets renewable energy technologies were not available. Country-level data on funding schemes issued by international institutions, such as the UNDP or the World Bank, would add to an even more comprehensive picture and would represent a valuable addition for future analyses. Regarding the lack of data on private R&D funding, recent empirical evidence predominantly supports the so-called ‘crowding-in’ hypothesis of private R&D funding. The hypothesis entails that in general, public R&D support drives innovation in interaction with and through stimulating/‘crowding in’ private R&D investment [64]. This recent evidence suggests that the main result on the general effectiveness of public R&D may not be invalidated by the omission of private data. Provided enhanced availability of private R&D data, future research should analyze this relationship for the case of RE technologies. In addition, private R&D data, as well as data on international funding schemes, could complement the analysis with regard to the first and second research questions.

A fourth challenge stems from the dependence of the impact of public R&D support on the stage of technology development. For example, Grubb 2004 [65] emphasizes that public R&D support is vital during the early phases of the technology development and continues to be effective during the demonstration phases until commercialization in a niche market. Different studies indicate that the stages of development and the generation and advancement of technologies matter for policies’ effectiveness [20,22,65]. For example, Costantini et al. 2015 [22] found that only second-generation technologies in the biofuels sector reacted positively to public R&D support. For this research, it was impossible to clearly discriminate between different stages of technology development. Doing so for RE technologies in general would be a relevant extension for future research.

4.2. Policy Recommendations

The descriptive analysis in this paper revealed a high level of heterogeneity across countries in terms of the relative importance of the European Commission’s R&D contributions to renewable energy technologies. Additionally, across NUTS 2 regions, the sizes of EC’s RE R&D contributions issued through the Horizon 2020 (2014–2020) program varied, although not strongly. The National Energy and Climate Plans (NECPs), which all EU member states must submit, address the necessity to align national support for R&D in renewable energy technologies [13]. However, alignment does not mean equality. Path-dependence and divergence in the governments’ R&D efforts or the allocation of EC funding may be of little concern for green innovation [66]. Overall, drawing normative conclusions from biases of the allocation of EC funding towards certain countries goes beyond the scope of this study. Instead (policy recommendation 1), the accessibility of country-level data on EC and national contributions for research on and the development of specific technologies should be improved to facilitate the transparency and alignment of public R&D efforts across countries. Furthermore, the availability of private R&D data should be enhanced so that the relationship between private and public R&D funding can be analyzed. In relation to this, the vital role international organizations such as the IEA play in providing such data and knowledge services and in mobilizing other agencies to support renewable energy technologies should be recognized [59].

In line with existing contributions, the estimation results on the overall effectiveness of R&D funding for RE are statistically significant, positive, and diverse across sectors. From this follows policy recommendation 2: the size of public R&D support for renewable energy technologies should be determined recognizing the local conditions and policies in place. For example, while Denmark heavily relies on wind power generation, ocean energy generation remains an underdeveloped sector in the country, despite Denmark’s ideal local conditions for ocean energy generation [59]. Another example of a technology-specific factor that affects the impact of R&D support is the complementarity to other energy innovation policies in place. Acemoglu et al. 2012 [67] show that an effective carbon price should complement public R&D subsidies for environmental regulation to be effective. Furthermore, the extent to which R&D support measures should be complementary to demand-pull instruments should, among others, depend on the current stages of development of specific RE technologies [22,55].

Policy recommendation 3 addresses the necessity to quantify marginal effects: the design of energy innovation policies should recognize the need to identify marginal effects and to quantify the effectiveness of measures, such as through implementing rankings or gradings in the distribution of public R&D support. Finally, policy recommendation 4 relates to this and addresses the temporal uncertainty of innovation outcomes. It is recommended that energy innovation policies are provided consistently over time and coupled with reporting processes of innovation impacts maintained for several years. Reporting processes can provide critical data to capture middle- to long-term effects. [28].

5. Conclusions

The EU and its member states widely acknowledge the importance of public research and development support for renewable energy technologies. For example, they committed to double public R&D investment from 2015 until 2020 and to increase the renewable energy share to at least 35% by 2030. This commitment provides the case for this paper’s thorough analysis of the extent, composition, and effectiveness of available public R&D funding for RE technologies in Europe.

This study analyzed public (the EC’s plus national) R&D funding for renewable energy technologies from 2000 to 2020 and across 17 European countries, on a country-specific level and over time. Based on a (to the authors’ knowledge) uniquely detailed panel dataset on a country-level, this paper provides strong support for the relevance of public R&D funding. In addition, it aims to incentivize future research and policy-making. The research questions of this study are answered as follows:

• What was the size of European public research and development funding for RE technologies in 2000–2020 across countries? Concerning this first research question, yearly averages for public R&D funding for RE technologies have been the highest in the largest economies (DE, FR). Nordic countries lead when the sizes of the economies are taken into account. Regarding changes over time, all countries experienced an increase in both national and EC’s public R&D funding until 2011, driven, among others, by ambitious legislation such as the legally binding Renewable Energy Directive 2009/28/EC. From 2012 onwards, in most countries, total public RE R&D support remained relatively stable since 2012. Thereby, a strong increase in EC contributions coincided with and compensated for decreasing national budgets. Hence, this research highlights the vital role of EC expenditures in stabilizing total public R&D support for renewable energy technologies. While relatively stable for most countries, in Norway, the Netherlands, Belgium and Portugal, total available public R&D funding for RE technologies even increased after 2012. However, none of the countries could double available public R&D support between 2015 and 2020, despite their commitment to the MI initiative. The EC’s very recent R&D funding that targeted RE technologies and that was issued through the Horizon 2020 (2014–2020) funding program did not vary strongly across most NUTS 2 regions, although economically strong regions profited significantly more than others.
• Within public RE R&D support, what was the importance of the European Commission’s funding relative to national budgets in 2000–2020 across countries? This research sheds light on the vast heterogeneity of the relative importance of EC contributions (as shares of total public R&D funding for renewable energy technologies): Belgium received 63% and Spain 46% of their total public R&D support for renewable energy technologies from the European Commission. In contrast, France received only 15% and Finland only 18% from the EC.
• Has recent European public research and development funding been generally effective as a driver of knowledge and innovation in RE technologies? Concerning this third research question, estimates for the overall average effect of total public R&D support on green innovation are statistically significant and positive. Therefore, based on precise data for the recent period of 2000–2015 and including the European Commission’s contributions, existing evidence derived from indices and without the consideration of EC contributions is confirmed. Furthermore, the relevance of public R&D support, but also the heterogeneity of its effectiveness, is confirmed for the largest sectors (biomass, solar- and wind energy). For the most recent years 2012–2015, the association between public R&D funding and patenting in renewable energy technologies is statistically insignificant. Statistically, the estimates for that period are unreliable: the period is too short and within variation too small. Also, the decrease in patenting activity contrasts with the development of other innovation output measures such as academic publications on RE technologies. Overall, various factors not included in this analysis may have affected patenting activity in those years, such as the industry decline in the solar sector and less need for patenting due to, for example, technology advancement.

The authors hope to incentivize future research and policy making through the findings of this analysis and a discussion of limitations the paper shares with existing studies:

• The limited availability of private and international data as well as of data that discriminates between various stages of technological development represent limitations in this paper. Accessible data on R&D support that stems from the private sector or international funding schemes and enhanced accessibility and categorization of country-level data on the European Commission’s R&D expenditures are needed. This will provide promising opportunities for future research and an even more comprehensive picture of R&D financing to policy makers.
• The restricted availability of data represents an even more urgent limitation in light of the necessity to align research efforts across countries, a goal that is reflected, for example, in the National Energy and Climate Plans of the EU member states. Efforts need to be aligned so that they recognize the local policies and conditions which are in place. While alignment does not mean equality, the findings of this research strongly suggest a high degree of relevance, but also a vast heterogeneity in terms of the distribution of national and EC budgets for renewable energy R&D across countries and regions. This should be considered in aligning and extending public R&D support policies. Furthermore, this vast heterogeneity of public R&D support should be taken into account in future research that aims at comparing the financing of R&D activities within the renewable energy sector across countries and regions.
• Another limitation which this research shares with similar studies is the general challenge of reliably quantifying causal effects. Again, the design of energy innovation policies can address this challenge and facilitate future research, for example, by determining support by grades or rankings that would permit the application of Regression Discontinuity Designs.
• Finally, also for the present study, the manifestation of innovation outcomes remains uncertain. According to Pitelis 2018 [23], since estimates depend on the chosen lag structures, this may undermine a quantitative interpretation. Energy innovation policies should be designed so that they address this limitation and facilitate future research and policy evaluation. For example, policies should be coupled with consistent reporting processes of innovation impacts that are maintained for several years.