The complementary effects of environmental policy and oil prices on innovation: evidence from OECD countries

ABSTRACT This paper examines the single and the joint influence of environmental policy stringency and oil prices on green innovation, admitting the possibility of different magnitudes of the response of innovation depending upon whether oil prices are increasing or decreasing, and accounting for endogeneity of policies. A panel data set of OECD countries is used over the period 1990–2016. Results suggest that increasing the stringency of environmental regulation can, beyond inducing green innovation, shield the effect of oil prices on innovation. In addition, a more stringent environmental policy reduces the asymmetric response of innovation when oil price increases or decreases. Thus, environmental policy and oil prices can be complements when inducing green innovation. Exploiting these complementarities requires an interdependent use of environmental and energy policies, through dynamic adjustments of subsidies and taxes on oil prices alongside reasonable levels of stringency in environmental policy.


Introduction
Over the last years, the global energy mix has, to some extent, shifted away from fossil fuels and towards clean energies.The majority of countries is now on a quest for a low-carbon energy system.This energy transition is expected to help achieve a higher degree of energy supply security and mitigate climate change and other environmental externalities associated with fossil fuels.
As a result, green growth has become an important development strategy for several countries and international organizations.The Paris Agreement, adopted in 2015, aims to keep the increase in global average temperature well below 2°C above pre-industrial levels.In 2018, the Intergovernmental Panel on Climate Change (IPCC) approved a special report highlighting the impacts of global warming of 1.5°C above pre-industrial levels and the related global greenhouse gas emission pathways, with the objective of strengthening the global response to the threat of climate change (IPCC 2018).In 2021, another assessment report from the IPCC declared that human activities have 'unequivocally' warmed the natural system.Weather and climate extreme events, caused by 'human-induced climate change', are now occurring in every region across the globe (IPCC 2021).More recently, in April 2022, an updated global assessment of climate change mitigation progress stated that policies appropriately designed to domestic circumstances and technological characteristics have successfully induced low-carbon innovation and technology diffusion (IPCC 2022).
Yet even though countries are moving away from fossil fuels toward cleaner alternatives, the predominance and dependency on fossil fuels remain.One can observe in Figure 1 how fossil fuels are still by far the prevailing source of primary energy (the countries presented in this figure are the ones included in our analysis).Renewable energy, although displaying in some cases considerable progress over the last 20 years, is still far from replacing fossil fuels in the energy mix.This represents a challenge not only due to the global need to reduce carbon footprint but also because most countries are dependent on foreign, and often geopolitically unstable, regions to fulfil their energy needs (see Figure 2 for the net energy imports of a selected group of OECD countries, most of them net importers, as Figure 3 shows).As a result, fossil-fuel-dependent economies are also vulnerable to changes in fossil fuel prices, namely oil.As with any other commodity, the price of oil fluctuates considerably depending upon the supply and demand (see Figure 4 for the evolution of crude oil price).Several events can cause oil shocks, from geopolitical events, such as  the war in Ukraine, to natural disasters or pandemics.Thus, the need to shield countries from these shocks by increasing their energy independence.
Accordingly, the concern and need for green growth and energy security led to several studies focusing on green innovation and technological change (e.g.Wang, Sun, and Guo 2019).Indeed, environmental policies in tandem with energy prices are defined in the literature as the two principal drivers of energy technology innovation, in both creation and adoption (Popp 2005;Johnstone, Haščič, and Popp 2010a;Dechezleprêtre et al. 2011;Cheon and Urpelainen 2012).Well-designed environmental policies can foster green innovations by altering the relative prices of factors of production.By imposing a price on the costs of pollution emissions (or by changing the opportunity costs of using a certain resource), a more stringent environmental policy will motivate firms to search for new production methods, since firms always aim to meet policy objectives at the least possible cost (Johnstone, Haščič, and Kalamova 2010b).Energy prices induce innovation in a similar way since energy commodities are broadly used in production processes.The higher the price of energy inputs, the higher the incentive firms have to innovate in order to reduce the usage of that input or to find alternatives (Triguero, Moreno-Mondéjar, and Davia 2014).
Environmental resources are needed for almost every economic activity, consequently some environmental degradation is inevitable.Thus, policymakers face the challenge of changing established technological systems in ways that mitigate their environmental impact, while minimizing  social disruption (Unruh and Carrillo-Hermosilla 2006).The answer to this challenge lies within technological innovation.
The main goal of this paper is to analyse the single and the joint influence of environmental policy stringency and oil prices on green innovation, admitting the possibility of different magnitudes of the response of innovation whether oil prices are increasing or decreasing.We use a panel data set of OECD countries to investigate the effects of environmental policy and oil prices on green innovation over the period 1990-2016.A dynamic econometric specification that accounts for policy endogeneity is used to estimate the relationship between the development of patents for environment-related technologies, oil prices, environmental policy stringency, international environmental agreements, and other related independent variables.As noted by Carrión-Flores and Innes (2010), innovation and policy are often jointly determined and so estimates of induced innovation effects that fail to account for the joint endogeneity of innovation and policy are likely to be biased.This paper overcomes a research gap in the literature regarding energy innovation by assessing the joint effects of the two main drivers of energy technology innovation (energy prices and environmental policies) and by exploring how these complementarities may impact green innovation and the path to decarbonization.
The results obtained show a positive relationship between green patents and oil prices and between green patents and environmental policy.In addition, a more stringent environmental policy diminishes the effect that oil prices have on innovation and reduces the asymmetric response that occurs on innovation upon variations in oil prices.
We contribute to the literature on drivers of technology innovation in the energy area by including a study period in which several oil shocks with different magnitudes and severities took place.Our results have significant policy implications since the existence of complementarities between environmental policy stringency and energy prices when inducing green innovation suggests that environmental policies should be combined with energy policies to be more effective. 1Since oil prices are affected by energy policies, the negative effects that decreasing oil prices have on green innovation could be minimized when environmental policy stringency is properly adjusted.For instance, dynamic adjustments of subsidies and taxes on energy prices should be considered when oil prices are decreasing and, consequently, incentives for green innovation decrease.In addition, increases in tax revenues or decreases in public spending, resulting from these dynamic adjustments, could be used to subsidize environment-related technologies.
The remainder of the paper is organized as follows.Section 2 provides a literature overview on the two main drivers of energy technology innovation: environmental policies and energy prices.Section 3 presents the empirical approach, including the data sources, the definition of the variables, and descriptive statistics.Section 4 presents the econometric results.Section 5 highlights the policy implications of the paper and section 6 concludes.

Literature review and research hypotheses
This section presents relevant literature and analyses the state of the art concerning energy technological innovation and asymmetric effects, in order to formulate the research hypotheses.Accordingly, subsection 2.1.starts by presenting the two main drivers of energy technological change (environmental policy stringency and energy prices), disclosing the mechanisms of how these drivers induce green innovation and highlighting the main results of previous literature.This will lead to the first research hypothesis to be tested in this paper.Then, in subsection 2.2., the state of the art regarding innovation asymmetric effects, a rather novel concept in energy innovation literature, is presented, once again displaying the main results of previous articles and presenting the second research hypothesis at the end.

Drivers of technological change in the energy area
Government policies are considered an important driver of technology innovation in the energy area.Its importance has increased due to environmental sustainability and energy security concerns.In the evolution of a technological system, governments' intervention may be directed for instance towards removing market uncertainty about the direction of technological development and deployment (Cheon and Urpelainen 2012;Veugelers 2012;Ambec et al. 2013).In fact, as Johnstone, Haščič, and Popp (2010a) noted, most policies implemented do not explicitly support technological innovation.Nevertheless, these policies are able to provide increased returns on the development of more efficient forms of electricity generation using renewable sources, either by decreasing the relative price of the use of renewable energy relative to fossil fuels or by increasing the demand for electricity generated from renewable sources.
The broad notion that stricter environmental policies can induce innovation is well documented (see Newell, Jaffe, and Stavins 1999;Popp 2002;Crabb and Johnson 2010).Induced innovation may be stimulated by regulation, as policies that restrict pollution or waste generation encourage the development of clean energy technologies (Newell, Jaffe, and Stavins 1999;Popp 2002;Crabb and Johnson 2010;Cheon and Urpelainen 2012;Veugelers 2012;Ambec et al. 2013;Costantini, Crespi, and Palma 2017;Bel and Joseph 2018).This relationship is known as the Porter hypothesis (Porter 1991;Porter and van der Linde 1995).Its weaker version (see Jaffe and Palmer 1997) states that well-designed environmental policies can foster green innovations as environmental policies alter the costs of pollution.This effect follows the Hicksian concept that changes in the relative price of factors of production provide incentives for firms to search for new production methods.Thus, if governments can use policy to alter relative input prices, changing the opportunity costs associated with the use of certain resources, firms will seek technologies that allow savings on these inputs, and innovation is induced (Johnstone, Haščič, and Kalamova 2010b).
Thus, policy-induced technological change lowers the costs of complying with environmental regulations and facilitates the implementation of stricter environmental policies, shifting the economy to a less-polluting pathway.Since the innovation process responds to economic incentives, increasing the stringency of environmental regulation motivates firms to develop new, or less costly, ways of production with the goal of eliminating particular types of emissions (Jaffe and Palmer 1997;Ambec et al. 2013;Fabrizi, Guarini, and Meliciani 2018).
More flexible regulatory instruments give firms greater incentives to innovate.This suggests that market-based instruments (such as pollution taxes, deposit-funds schemes, or tradable permits) are more likely to induce innovation than direct forms of regulation (such as performance standards).Actually, once a standard is satisfied there are no incentives to develop or adopt cleaner technologies, while in the presence of market-based instruments the higher the emissions reduction the higher are the benefits in terms of subsidies, revenues from tradeable permits, or tax reductions (Johnstone, Haščič, and Kalamova 2010b;Cheon and Urpelainen 2012;Ambec et al. 2013;Fabrizi, Guarini, and Meliciani 2018).
Accordingly, policy stringency plays a significant role in inducing innovation.A more stringent policy will provide greater incentives for polluters to search for ways to avoid the costs imposed by the policy.However, inducing environmental innovation is a challenge to policymakers.Pollution involves a negative externality and innovation involves a positive externality.A firm that invests or implements a new technology may create benefits for others while incurring all the costs.Consequently, the firm lacks the incentive to increase those benefits.So, without public policy designed to overcome these two market failures, firms pollute too much and innovate too little, compared with what would be the social optimum (Jaffe, Newell, and Stavins 2005;Johnstone, Haščič, and Kalamova 2010b;Veugelers 2012;Bel and Joseph 2018).
Many environmental policies attach a price to pollution.As a result, the more stringent the policy the greater the effect it has on innovation and on reducing pollution emissions.Since markets often fail to put a price on environmental resources, the opportunity cost of many environmental assets is to a large extent established by government regulation (Johnstone, Haščič, and Kalamova 2010b).Technological innovation is important for environmental policy, and analysis of energy and environmental policy can benefit from the perspective of the economics of technological change (Jaffe, Newell, and Stavins 2005;Costantini, Crespi, and Palma 2017).
Energy prices also lead to technological change and induced innovation in the sector.High energy costs, namely high fossil fuel costs, resulting from rising prices, encourage firms to adopt cleaner strategies and search for green technologies.Ultimately, innovation is induced allowing firms to reduce the usage of fossil fuel energy (Dechezleprêtre et al. 2011;Triguero, Moreno-Mondéjar, and Davia 2014).For instance, fossil fuel prices display a positive relationship with innovation regarding renewable power generation, i.e. when prices increase innovation activity also increases (Johnstone, Haščič, and Popp 2010a;Cheon and Urpelainen 2012).Moreover, Noailly and Smeets (2015) show that high fossil fuel prices significantly increase alternative energy patenting, while Ley, Stucki, and Woerter (2016) show that energy prices and green innovation activities are positively related and energy prices have a significantly positive impact on the ratio of green-innovations to non-green innovations.Thus, even though rising prices can reduce a firm's profitability and resources for R&D, it can also make them react by patenting new and cleaner environmental technologies (Hussain et al. 2022;Lin and Chen 2019).
Cheon and Urpelainen (2012) argue that when oil prices remain at sustainable high levels, countries adjust their policy and innovation strategies, which may increase the development of alternative solutions.The deployment of renewable energy technologies becomes 'accidentally' linked with the countries' innovation system.Thus, it is not possible to separate out the impact that oil prices have on innovation for alternative energy technologies without controlling for environmental policy and without accounting for and dealing with the endogeneity between innovation and environmental policy.Not accounting for this endogeneity may produce biased estimates of the environmental regulation effects on innovation activity, as noted by Rubashkina, Galeotti, and Verdolini (2015).
Therefore, we analyse the joint effects of these two drivers in order to assess if oil prices and environmental policy complement each other when inducing green innovation.Thus, the following first hypothesis is tested: H1: Are oil prices and environmental policy substitutes or do they complement each other when inducing green innovation?This is a relevant question given that green technology innovation has been pointed to as one of the main energy transition drivers, as well as a solution to the crisis of global energy governance.

Asymmetric responses in the energy area
Concerning the link between energy prices and economic performance, some literature has been devoted to asymmetric macroeconomic responses to energy price increases and energy price decreases.For instance, Mork (1989) shows that the effects of oil price declines and oil price increases on the United States' real GDP growth are different.Mork, Olsen, and Mysen (1994) investigate the correlations between oil price movements and GDP fluctuations for the United States, Canada, Japan, Germany, France, the United Kingdom, and Norway.The authors provide evidence that most countries display asymmetric effects, with Norway as an exception.Kilian (2008) addresses several topics regarding the economic effects of energy price shocks, one of them being the fact that energy price increases seem to cause recessions but energy price decreases do not seem to cause economic expansions.According to the author, this asymmetric relationship between oil prices and macroeconomic aggregates results from the different effects that demand shocks and supply shocks have on energy markets.
However, while the link between energy prices and economic activity is explored in the literature, the link between energy prices and innovation activity is fairly unexplored.The objective of this paper is to fulfil this gap.Indeed, the occurrence of asymmetric effects on innovation depending on whether energy prices are increasing or decreasing is more scarce in the literature.The exception is the work of Nunes and Catalão-Lopes (2020) which investigates if the magnitude of the response of innovation in alternative energies is significantly different depending on whether oil prices rise or fall.In fact, the authors show that the effects are asymmetric and that the impact is stronger when prices go down, i.e. when prices are decreasing the reduction of alternative patent applications is more prominent than the expansion when prices are rising.This result suggests a lack of long-term commitment to finding alternatives to fossil fuels.In line with this outcome, Shah, Hiles, and Morley (2018) suggest that when oil prices are low policymakers should increase financial support in order to guarantee that investment in renewable energies remains at a constant level.
Thus, besides examining if oil prices and environmental policy complement each other when inducing green innovation (Research Hypothesis 1), we also assess if there are asymmetric effects in this complementarity depending on whether oil prices are increasing or decreasing.Hence, we formulate the following second hypothesis: H2: Is the effect of oil prices and environmental policy on green innovation asymmetric depending on whether oil prices are increasing or decreasing?
Hence, this paper investigates the joint effects of environmental policy and crude oil prices on green innovation assuming different asymmetric responses to innovation depending on whether oil prices are increasing or decreasing.

Data and methods
This section describes the empirical approach followed, designed to capture the complementarity effects of environmental policy stringency and oil prices on green innovation, considering the possibility of asymmetric responses.
A panel data of 14 OECD countries is used to investigate these hypotheses.The countries considered are Australia, Belgium, Canada, France, Germany, Italy, Japan, South Korea, the Netherlands, Norway, Spain, Sweden, the United Kingdom, and the United States of America.These are the most innovative OECD countries and together they represent around 88 per cent of the global number of patent applications concerning green technologies (average for the period 1991-2016).As in Wang, Sun, and Guo (2019), the reason for choosing OECD countries is related to, first, the importance of these countries in the world economy, making them a representative sample from a global point of view and, second, with the leading role they have taken in promoting green growth.For these reasons, examining the combined effects of environmental policy stringency and oil prices on green innovation of OECD countries provides a global overview and significant lessons for other countries.
Green innovation in the development of patents for environment-related technologies is explained as a function of environmental policy, crude oil prices, and other relevant independent variables.We discuss each of them in turn in the next two sub-sections.

Dependent variable: patents for environment-related technologies
As explained before, the objective of this paper is to measure the joint effects that environmental policy and crude oil prices have on green innovation.Following convention, we use information on patent applications as a proxy for innovation (Popp 2002;Popp 2005 Our dependent variable, the number of environment-related inventions, expressed as a percentage of all domestic inventions in all technologies, is modelled using a panel data set comprised of the 14 OECD countries previously mentioned.This ratio of environment-related patents to the total patents is preferred over a patent count because by doing so we can interpret changes in environmental technological innovation in relation to innovation in general, i.e. in relation to the countries' innovative capacity, in relative rather than absolute terms. Figure 5 shows the evolution of environment-related technologies in our sample countries.Data was obtained from the OECD Green Growth database, considering the indicator Patents: Technology development dataset, developed by Haščič and Migotto (2015), which is available for the period 1990-2016. 2Indicators of technology development are constructed by measuring inventive activity using patent data from the following environment-related technological domains: environmental management (including waste management, air pollution abatement, and environmental monitoring); climate change mitigation technologies related to transport, production processes, and energy generation, transmission, and distribution; water-related adaptation; capture, storage, sequestration or disposal of greenhouse gases; and energy efficiency in buildings.Some of these technological domains are only indirectly influenced by energy prices.Yet, the commitment to sustainability and to more energy independence influences the technological path and innovation system of a country and spills beyond technologies trying to directly replace fossil fuel usage in the generation of energy.Patent data was organized by International Patent Classification (IPC) code, a hierarchical classification system used by patent offices to classify patent documents according to the technological fields they pertain (to consult the technological classes included in this variable, see Haščič and Migotto [2015]).
Figure 6 presents the evolution of patenting activity from the environment-related selected technological domains for the aggregate panel of OECD countries under analysis.Patents for Climate Change Mitigation regarding Energy, Transport, and Production, together with technologies for Environmental Management, are prevalent during the whole period under scrutiny.This predominance may be justified by the fact that environmental policies have consistently targeted the most polluting industries, as noted by Caravella, Costantini, and Crespi (2021).Yet, Climate Change Mitigation regarding Energy, which includes renewable energy technologies, appears by far the dominant technological domain since 2008, which is not surprising given the evidence that innovation and deployment incentives for renewables are an effective environmental strategy (see, for instance, Carrilho-Nunes and Catalão-Lopes [2022]).As for the evolution of these technologies over time, there is an increasing trend until 2010, explained by the increase in the stringency of environmental and technology policies, side by side with growing environmental awareness worldwide.This peak is followed by a downward trend after 2011, resulting possibly from the financial crisis and the supply surplus of innovative equipment resulting from China's mass production (Caravella, Costantini, and Crespi 2021).Finally, one should also notice the peak in patenting activity circa 2014, a period of consistently high energy and oil prices, which was followed by an oil shock and a slump in oil prices.
The counts of patent applications used in this dataset contain only 'higher-value' inventions, with a patent family size higher or equal to 2 (OECD 2017), and allocated to countries according to the country of residence of the inventor(s). 3Even though using patents as an indicator of innovation has some disadvantages (e.g. a patent can be granted to significant and minor inventions equally, and the presence of a patent does not imply technology diffusion), the advantages allow patent data to serve as an acceptable and widely used indicator for technology innovation since it displays the magnitude of knowledge production activities and the investment done (for a further discussion on the advantages and disadvantages of using patent data as a measure of environmental technological innovation see, for instance, Nunes and Catalão-Lopes [2020]).One should also notice that patent data also signals the degree of technological sovereignty held by a country/region (Caravella, Costantini, and Crespi 2021).
Data from the OECD Green Growth database was preferred over counts of patents applications from the World Intellectual Property Organization (WIPO) database because, as noted by Haščič and Migotto (2015), taking simple extractions from WIPO is problematic for two reasons.First, irrelevant patents may be included if an IPC class includes patents that do not hold the desired environmental focus.Second, relevant inventions may be left out and overall innovative activity may be underestimated.

Independent variables
A positive relationship between oil prices and the development of environment-related technologies is anticipated.Data for Imported Crude Oil Real Price was collected from the Energy Information Administration Short-Term Energy Outlook (February 2019). 4A multiplicative dummy equal to Oil Prices × Dummy Price Decreases was constructed to measure the asymmetric impact on the development of environment-related technologies depending on whether oil prices are increasing or decreasing.This variable is equal to 1 when oil prices decrease compared with the previous year and 0 when oil prices increase.
A composite index, the Environmental Policy Stringency (EPS) index, was used as the indicator for environmental policy stringency.This index, provided by the OECD.Stat database and developed by Botta and Kozluk (2014), represents, according to Verdolini, Vona, and Popp (2018), the most complete indicator on environmental policy instruments available for OECD countries.It was developed following a methodology that allows for comparison across countries and over time, reducing the complex set of multidimensional policies into a comparable country-specific proxy.The list of all indicators and correspondent weights used in the aggregation of the EPS composite index are listed in Table 1.
The index takes continuous values from 0 (not stringent) to 6 (highest degree of stringency), where stringency is defined as the degree to which environmental policies put a price (explicit or implicit) on polluting or environmentally harmful behaviour.In the first year of our sample, the average EPS index is 0.84, while by the end the average is higher than 3.A significant increase in the index is clear for all countries.In fact, during our time span all more than double their environmental stringency, and some even present values more than 5 times higher than those recorded in 1990 (see Figure A1 in the Appendix).
Following the Porter hypothesis, a positive relationship is expected between the EPS index and innovation for environment-related technologies, since a tough environmental political stance is expected to perceive the development and establishment of a green technological system as a priority. 5 Two multiplicative variables were constructed to investigate how environmental policy stringency may be additionally used as a tool to buffer the influence that crude oil prices have on green innovation.The first multiplicative variable, equal to Oil Prices × EPS, was constructed to identify the effect of environmental policy on the impact that oil prices have on innovation.The second multiplicative variable, equal to Oil Prices × Dummy Price Decreases × EPS was constructed to examine if environmental policy can influence the different magnitude responses of green innovation to either positive or negative oil price changes.
Other two measures of policy stringency are included in the empirical model, representing a policy mix between demand-pull and technology-push instruments, following the contributions of Costantini, Crespi, and Palma (2017), Costantini et al. (2020), andCostantini, Sciabolazza, andPaglialunga (2022).The first measure is a demand-pull indicator, representing how policy stringency can influence energy prices, via energy taxation, and, consequently, demand patterns (Costantini and Martini 2010).Accordingly, a variable with diesel taxes per litre provided by OECD.Stat was included (as of 2022, the minimum excise duty on fuel-based products in the EU, for instance, is around 0,36 €/ litre, representing a considerable portion of diesel prices).The second measure is a technology-push indicator, intended to capture the effort to support invention activities and promote the deployment of green technologies.To build this indicator we follow the work of Costantini et al. (2020) to compute the stock of environmental R&D.Like the authors, we assume that technological knowledge is cumulative and can be added over time but is subject to an obsolescence rate.Data on government budget allocations for environmental R&D, provided by OECD.Stat, was used in a Perpetual Inventory Method (PIM) to calculate the stock values (Equations 1 and 2): where g i represents the average annual growth rate of environmental R&D during the period under analysis (different for each country), d is an average discount rate (or decay rate) of 15 per cent, i and t represent countries and years respectively (note that t 0 is the first period in the time series, 1991).As with the EPS variable, a positive relationship between the indicators of the policy mix and green patenting activity is anticipated.
To capture the effects of international agreements on green innovations and the expectations about future policy, a dummy variable equal to one after ratifying the Kyoto Protocol was included in the model.Information about the Kyoto Protocol was provided by the Statistics Division of the United Nations' Department of Economic and Social Affairs.A positive relationship is expected, as the Protocol might encourage green innovation through patent fillings for clean energy technologies.
Since innovation activities display high persistence and can exhibit significant path dependence (Hecker and Ganter 2014), a one-year lagged dependent variable was incorporated among the covariates.Like Nicolli and Vona (2019), Rubashkina, Galeotti, and Verdolini (2015), Nesta, Vona, andNicolli (2014), andPopp (2002), a variable controlling for previous patenting activity was included in order to capture innovation experience and persistency in inventive activities.This variable is expected to have a positive influence on the innovation capacity given its enduring nature.
As in Bel and Joseph (2018), annual GDP growth rates, provided by the OECD.Stat, were used as a control variable.A positive relationship between this variable and the dependent one was a priori expected.However, this was not confirmed by the results (see possible explanations and interpretation in Section 4).
A variable for energy dependence, measured as the net energy imports as a percentage of energy use, provided by the World Bank Development Indicators, was also included in the model.While a positive relationship with green patenting activity implies that the more a country imports energy (thus, is energy dependent) the more prone to green innovation that country is, a negative relationship can also be intuitive given that exporting countries (thus, energy independent) might have an advanced innovative energy system given the importance that energy has in their economic activity.
Table 2 provides basic descriptive statistics for the dependent and independent variables (except for dummies) included in the final model, as well as the shorted names of all the variables.

Econometric model
We use a cross-country panel data set for 1990-2016 to examine the relationship between green innovation and a set of relevant independent variables.Our data span is dictated by the fact that the dependent variable, the number of environment-related inventions, provided by the OECD Green Growth database, starts in 1990 and ends in 2016.We lag the independent variables by one year to avoid simultaneity problems with innovation activity.According to Miyamoto and Takeuchi (2019), Nicolli and Vona (2019), Bel andJoseph (2018), andRubashkina, Galeotti, andVerdolini (2015), a lag of at least one year may be enough to reflect the length of time required between invention and patent eligibility.In line with this, we tested different lag structures: lag = 0, lag = 1 and lag = 2, and the model with a lag equal to one was chosen as the preferred model.The data set consists of country-years observations, in which countries, the cross-sectional unit, are represented by i and years by t.
Similar to Ley, Stucki, and Woerter (2016) and Carrión-Flores and Innes (2010), as a first approach we estimate a static Ordinary Least Squares (OLS) model with fixed effects (Column 1 of Table 3), which specification is given by: In this equation m i represents the fixed effects, introduced to capture unobservable country-specific heterogeneity (characteristics that may explain given innovation trajectories, like differences in oil dependency, natural endowments such as.wind or sun, or preference for sustainable development), and 1 i,t represents the stochastic error term. 6Cluster-robust standard errors were used in order to correct for heteroskedasticity and autocorrelation (see, for instance, Cameron and Miller [2015]).Since both green innovation and environmental policy stringency might be co-determined within a country's strategy (as noted by Nicolli and Vona [2019] and Carrión-Flores and Innes [2010]), the identification of policy effects can be problematic.The endogeneity of policy stringency could cause either a downward or an upward bias in the estimations.To overcome potential endogeneity problems presented in the OLS model, a dynamic panel estimation approach is adopted, more precisely the Generalized Method of Moments (GMM) estimators (Blundell and Bond [1998]; Arellano and Bond [1991]). 7 These estimators build a system of two equations, the original equation and the transformed one, and allow the use of lags as instruments for the identification of the parameter on the (endogenous) lagged dependent variable (in our model INNOVATION_1) and on the endogenous regressors (in our model the variable EPS).Thus, this dynamic setup (Column 2 of Table 3) is estimated relying on a GMM estimator that allows for the use of instruments as follows: where we define the exclusion restriction in the case of endogeneity of the regressor as (Z i,t = (1, Xi,t , EPS i,t−t , IV i,t−t ), where Xi,t is the adapted set of independent variables which are considered exogenous, EPS i,t−t is the set of lagged EPS variables in the dynamic model, and IV i,t−t are instruments that serve as additional moment restrictions to account for the endogeneity of policy variables. 8As it is assumed that the innovation process takes time and that environmental policyinduced innovations are translated into patents with at least one (or more) year lag period, we include in the model from one to three-year lagged EPS variables (t = 1,2,3).Thus, in this paper green innovation and environmental stringency dynamics will be examined empirically by using the Arellano-Bover/Blundell-Bond estimation technique.According to Oikarinen and Engblom (2015), this panel model specification is suitable for this type of analysis since it is a general estimator designed for situations with: (1) a linear functional relationship; (2) one lefthand-side variable that is dynamic (i.e.depends on its own past experience); (3) independent variables that are not strictly exogenous (they may be correlated with the error term); (4) fixed individual effects; and (5) heteroskedasticity and autocorrelation within individuals but not across them.

Results
Table 3 reports our main empirical findings.Column 1 presents the OLS-FE model as in Equation (3), while Column 2 presents the best specification where a dynamic approach is used. 9 In general, the results confirm that both environmental policy stringency and crude oil prices are determinants of green innovation.The two multiplicative dummies designed to capture the joint effect of environmental policy and oil prices are significant and negative.Thus, a more stringent environmental policy can mitigate the impact that oil prices have on green innovation, and also smooth the uneven response that occurs in innovations due to fluctuations in crude oil prices.When comparing the results of Columns 1 and 2, the effects of oil prices and environmental policy stringency appear larger in the dynamic specification.For instance, not considering the endogeneity of EPS biases its effect downward (by around 57 per cent, as the coefficient for the variable EPS_t-1 in the OLS is 0.3299 and in the dynamic approach it is 0.7736).Additionally, the asymmetric response of innovation caused by variations in oil prices is higher but the ability of environmental policy stringency to correct this asymmetry is also higher.These results empirically emphasize the importance of accounting for endogeneity of political variables.Nevertheless, the specification in model (1) provides a lower bound for our effects of interest.
Detailing for each impact and focusing on the best model specification (Column 2), we see that, unsurprisingly, the results obtained suggest that crude oil prices and the ratio of environmentrelated inventions have a positive and significant relationship, in line with the previous literature.In other words, green innovation increases when oil prices increase, and decreases when oil prices go down.
The dummy measuring the effects of declining oil prices is significative, meaning that the response of innovation is asymmetric when the oil price increases or decreases Similarly, to Nunes and Catalão-Lopes (2020), the impact is stronger when prices go down than when prices go up.That is, the reduction in innovation in the presence of falling oil prices is more pronounced than the expansion in innovation in the presence of rising prices.
Concerning the EPS variable, the results obtained display a positive relationship between a more stringent environmental policy and green innovation, which is aligned with the Porter hypothesis, as expected.Actually, this effect is higher for the year immediately after the implementation of the policies (with a beta coefficient of 0.7736) and, even though it is still significant after two and three years, it has a lower impact (with beta coefficients of 0.5244 and 0.2310).After that (4 or more lags) the effect of EPS on innovation is not significant (the variables EPS_t-1, EPS_t-2, and EPS_t-3 stand for the lagged EPS variable by 1, 2 and 3 years, respectively).
The two multiplicative variables associated with EPS are significant and negative.Increasing the stringency of environmental regulation, beyond from inducing green innovation, can also shield the effect of oil prices on innovation.A 1-unit increase in EPS reduces the effect that oil prices have on innovation from 0.0361 pp to 0.0290 pp.In addition, a more stringent environmental policy reduces the asymmetric response of innovation when oil price increases or decreases.A 1-unit increase in EPS reduces this asymmetric response of innovation from 0.0102 pp to 0.0045 pp.So, for instance, when oil prices increase green innovation increases by 0.0361 pp.Since the response to oil prices variations is asymmetric, when oil prices decrease innovation decreases by 0.0463.However, when oil prices decrease and EPS is used complementarily, green innovation will decrease only by 0.0406 pp.
Thus, a more stringent domestic environmental policy can make the economy less sensitive to the effects of crude oil price fluctuations, which is an important result of our analysis, confirming that environmental policy and oil prices are effective inducing green innovation (notably, when comparing the model without endogeneity correction (1) with the model that accounts for endogeneity (2), one can observe that environmental policy stringency attenuates the exogenous effects of oil prices even more in the second model.Thus, correcting for endogeneity puts in evidence an even stronger role of EPS and oil prices as vehicles for innovation).
The two measures representing a policy mix between demand-pull and technology-push instruments present similar results to the EPS variable, confirming the inducing effect that environmental policy has on green innovation (consistent with Costantini et al. [2020] and Costantini et al. [2022]).Accordingly, the relationship between the diesel tax and green innovation is positive, i.e. the higher the energy tax rate the higher the incentive to innovate.As for the stock of environmental R&D, it also positively impacts innovation, thus the higher the stock of R&D the more prone to green patenting activity the country is.
The ratification of the Kyoto protocol displays a positive and significant relationship with green innovation (in line with Miyamoto and Takeuchi [2019]; Nesta et al. [2014] and Johnstone et al. [2010a]; and contrary to Cheon and Urpelainen [2012]).Hence, energy policies, specifically international treaties, impact the development of the innovation systems as they allow policymakers to attack climate concerns and to comply with environmental targets.Moreover, as noted by Johnstone, Haščič, and Popp (2010a), future policy expectations are relevant to induce innovation.The ratification of an international environmental agreement may create expectations of future stronger policy stringency, thus indirectly promoting more green innovation.
The lagged dependent variable, which adds some form of path dependency into our empirical specification, is positive (consistent with Nicolli and Vona [2019]; Rubashkina et al. [2015] and Nesta et al. [2014]).This result supports the idea that innovation activities are highly persistent and previous innovation experience has a considerable effect on future innovation capabilities.
The estimated model shows a negative relationship between annual GDP growth rates and the ratio of environment-related inventions.One possible explanation might be the role of the Green Economy on boosting the economic activity.In fact, a large-scale penetration of renewable energy installations can increase economic productivity and development, decrease trade deficit, and have positive employment effects (Creutzig et al. 2014).Thus, it seems that as a response to a negative GDP growth many countries invested in innovation for clean sources of energy as a measure to stimulate the economy. 10 Lastly, the variable controlling for energy dependence presents a negative relationship with environment-related patenting activity.This outcome reveals that countries moving towards energy independency are more prone to green innovation.This occurrence can be explained by the important contribution that energy has to the economic activity of these countries.As a result, the energy and innovation systems are more mature, which facilitates patenting activity.In addition, it may also suggest that countries exporting energy might be investing in green technologies to diversify their energy portfolios and make better use of their resources.
The model presents a good fit.The value of the Likelihood Ratio (LR) Chi-Square statistic of model ( 2) is 97213.9with a p-value of 0.0000, indicating that the independent variables chosen provide a suitable approach to explain new environment-related inventions.In addition, we tested for serial correlation in the first-differenced errors using the Arellano-Bond test for serial correlation and, since the null hypothesis at order 2 was not rejected (with a p-value of 0.9628), the moment conditions are valid.

Robustness checks
To check the robustness of the results we estimated a model using alternative variables for the dependent and the independent variables.This sub-section offers a brief summary of the results obtained, which are presented in the Appendix (Table A1).
First, considering the dependent variable, we used counts of patent applications, provided by the World Intellectual Property Organization (WIPO), in the following technological classes: renewable energy technologies, motor vehicle technologies, and energy efficiency in the residential, commercial, and industrial sectors.Since patent statistics provided by the WIPO database are only available after 1995, this alternative model analyses the period 1995-2016.Given that now the dependent variable is a nonnegative integer or count, we employed a negative binomial specification (see, for instance, Nunes and Catalão-Lopes [2020] on why the negative binomial distribution is preferred over the Poisson when modelling counts of patent applications).As a result, one should be cautious when comparing the coefficients of the alternative model with the main model.
Regarding the main independent variables, we used natural gas prices (Henry Hub Natural Gas Spot Price) instead of oil prices, and diesel end-user prices as a demand-pull instrument instead of the diesel tax.In addition, the complementary effects were assessed using the variable stock of Environmental R&D, instead of the EPS index.The Kyoto dummy was not replaced, because the best alternative would be a dummy ratifying the Paris Treaty but that is unfeasible in our analysis since the ratification started in 2015.
The main conclusions drawn from the baseline model (dynamic specification) remain unchanged.A positive and significant relationship is obtained between natural gas prices and counts of green patent applications.In addition, the response is once again stronger when prices, now natural gas prices, go down than when they go up (the asymmetry hypothesis holds).Moreover, increasing the stock of Environmental R&D can induce green patenting activity and shield the effect of natural gas prices on this innovation.More stock of Environmental R&D reduces the asymmetric response of innovation when natural gas price increases or decreases (the complementary hypothesis holds).Lastly, the response of green innovation to GDP is now positive but non-significant.

Policy implications
Results on complementarity between environmental policy stringency and oil prices have important policy implications.First, as environmental regulation seems to be an effective way of promoting green innovation, aiming at reasonable levels of green growth calls for well-crafted environmental policies, with significant levels of stringency.Second, since our results suggest that environmental policies and oil prices are complements that may enhance the ratio of environment-related inventions, exploiting such complementarities requires the coordination between different policy institutions and a dynamic relationship between oil prices and regulation.
As suggested by Shah, Hiles, and Morley (2018), a more counter-cyclical approach might be required regarding energy policies.Our results confirm this.In countries where the price of oil is subsidized, when oil prices are decreasing policymakers should decrease subsidies.By doing so they ensure that the decrease in green innovation is less pronounced, due to the asymmetric response of innovation, and hence the ratio of environment-related technologies is kept at desirable levels.
For countries where such subsidies do not exist, policymakers should raise taxes on oil prices when these are decreasing.Thus, the negative effects on green innovation due to the decrease of oil prices are minimized.
Additionally, increases in tax revenues or decreases in public spending, resulting from these dynamic adjustments, could be used to subsidize environment-related technologies.In fact, implementing clean energy technologies can provide a useful hedge against the negative effects of volatility and unpredictability in oil prices.
In addition, a policy mix of demand-pull and technology-push instruments is also recommended given that both measures have the ability to induce innovation in green technologies.
Finally, a commitment to international climate agreements might also spur green innovations.The influence of introducing an international framework can stimulate technological change, perhaps driven by future expectations of more stringent regulation or tougher commitments.

Conclusions
This paper investigates the single and the joint effects of environmental policy and crude oil prices on green innovation.More specifically, the paper examines if environmental policy stringency and oil prices are effective drivers of environment-related technologies and whether they reinforce (or hinder) each other's impact.Moreover, the paper investigates if the relationship between these two innovation drivers is different depending upon whether oil prices are increasing or decreasing.This is a pertinent question with relevant policy implications, as energy technology innovation is pointed as a solution to the crisis of global energy governance.
We model the number of environment-related inventions, expressed as a percentage of all domestic inventions in all technologies, using a panel data set comprised of 14 OECD countries.Together these 14 countries represented around 88 per cent of the total number of patenting activity for green technologies, making them a representative sample from a global point of view.Data spans the years 1990-2016.An empirical assessment of the relationship between the ratio of environment-related technologies and crude oil prices, environmental policies, plus other relevant variables is provided using a fixed effects OLS and an Arellano-Bover/Blundell-Bond GMM panel data estimator in order to correct for the endogeneity of policy variables present in the static OLS model.
A positive and significant relationship is obtained between crude oil prices and the ratio of environment-related inventions.Thus, green innovation increases when prices rise and decreases when prices go down.However, the response is stronger when prices go down than when they go up.
According to our results, environmental policy stringency and joining international environmental agreements have a positive and significant impact on green innovation.Increasing the stringency of environmental regulation can, beyond from inducing green innovation, shield the effect of oil prices on this innovation.In addition, a more stringent environmental policy reduces the asymmetric response of innovation when oil price increases or decreases.Thus, environmental policy and oil prices can be complements when inducing green innovation.
In terms of policy implications and given these complementary effects between environmental policy stringency and oil prices, an interdependent combination of environmental and energy policies should be applied, through dynamic adjustments of subsidies and taxes on oil prices, alongside reasonable levels of stringency in environmental policy.This allows to use the development and diffusion of green inventions as a hedge against the negative effects of volatility and unpredictability in oil prices.
This paper contributes to the knowledge of the asymmetric effects that oil price changes and environmental policy have on green innovation by including a research period in which several oil shocks with different magnitudes and severities took place and the growth rate in the world's renewable energy capacity increased considerably.In addition, it contributes to the literature on energy innovation by filling a research gap concerning the complementary effects of the two main drivers of energy technology innovation and whether they reinforce (or hinder) each other's impact.
By using the EPS index, we considered environmental policy stringency as a composite measure of market-based and non-market-based instruments.In this respect, a separate analysis for each type of regulation could be fruitfully addressed in future research.

Notes
1. Energy policy addresses issues such as energy production, energy consumption or energy security, and includes environmental policy, which is concerned only with environmental matters.In the context of the present paper, we refer to energy policy as measures directed to modifying the costs of using oil (e.g., taxes or subsidies).2. OECD Green Growth patent statistics are constructed using data from EPO Worldwide Patent Statistical Database (PATSTAT).Patent data is complex, and its interpretation requires caution, for instance to avoid double-counting of inventions and misinterpreting data, since patents are commonly classified in more than one technology class (Haščič and Migotto 2015).For this reason, we opted for an indicator already developed containing selected environmental technologies from the OECD Green Growth Indicators (ENVTECH) to guarantee the reliability of the data.3. The priority filing of an invention followed by its subsequent patent filing in other jurisdictions forms a patent family which protects the same invention.Thus, patent family size refers to the number of patent authorities where patent applications were filed to protect the same invention.According to Haščič and Migotto (2015), using statistics with a family size equal or greater than 2 (i.e., based on the claimed priorities) is the most appropriate method for international comparisons because only the high-value priority applications are counted.Claimed priorities are associated with inventions of higher value because patenting is costly, and, as a result, a firm will only protect its intellectual property in more than one jurisdiction if its economic viability is expectable.Moreover, considering a family size equal or greater than 2 excludes low-value inventions that usually seek protection only at a single patent office (one-member families).Another methodology for constructing patent statistics is to use triadic families, which consider patents filed in all three major patent offices, the United States Patent and Trademark Office (USPTO), the European Patent Office (EPO), and the Japanese Patent Office (JPO).We opted for the first methodology discussed (family size equal or greater than 2) to apply a more inclusive filter when identifying higher value patents from countries outside the triadic regions.In addition, according to Haščič and Migotto (2015) and Martinez (2010), triadic families place an excessive limitation on narrow technological fields.
4. This variable is common to all countries as the price considered is not country specific.Other variables concerning oil prices were also tested, namely Crude Oil Futures and oil prices moving averages for 2, 3, and 5 years.However, Imported Crude Oil Real Prices turned out to be the most suitable variable.5.It should be noted that, even though a positive relationship between more stringent regulations and green patenting is assumed, firms can react to stern policy by increasing innovation, firms can also offshore the dirtiest stages of production or fail and exit the market if the cost of compliance with stern regulations are too high.6. Models with time fixed effects and time trends were also tested, but the results were not statistically significant.7. A dynamic methodology was preferred over a 2SLS approach with other variables as instrumental variables (IVs) because finding suitable instruments for EPS, other than its own lags, turned out to be very challenging.Some 2SLS models (for instance, Rubashkina et al. [2015] and Nicolli and Vona [2019]) use regulations in adjacent sectors (such as telecommunications) as instruments, but there is no information available regarding policy stringency of other areas besides environment.Another variable used in similar settings is the time length for which a country has had consolidated and durable democratic institutions, since democratic countries tend to approve stricter environmental policies (Nesta, Vona, and Nicolli 2014).However, in our model such variable turned out to be also a weak instrument for EPS, since it displayed a low correlation with the policy variable (a good IV must be highly correlated with the variable that is substituting and not correlated with the dependent variable).8. Models that use IVs (whether these IVs are the lags of the endogenous variables or completely new variables) must meet an exclusion restriction condition, which requires that any effect of the instrument on the dependent variable is exclusively through its potential effect on the endogenous variable.Thus, instruments that directly affect the dependent variable or only indirectly affect the endogenous independent variable must be excluded.9. Other model specifications were tested, namely a static threshold model and a per-period analysis.Regarding the threshold model, the regime dependent variable considered was the previous patenting activity (lagged innovation activity) and the threshold variable considered was trade as a percentage of GDP.The results of this model show that the higher the trade level the lesser the importance of the knowledge stock has on future patenting activity.As for the per-period analysis, quarterly data was used to analyze four periods of oil market instability: the sharp drop following the financial crisis in the countries of Southeast Asia (1998-1999), the fall following September 11 Attacks (2001Attacks ( -2002)), the decrease during the sub-prime crisis ( 2008), and the drop following the slowdown of the Chinese economy (2014)(2015).Results show that even though these four events can be considered as oil crisis, only in the last one, the 2014-2015 oil shock, did oil prices significatively (and negatively) impact environment-related innovation.Nevertheless, following convention and due to robustness considerations, the combination and comparison of static (OLS-FE) and dynamic (Arellano-Bover/ Blundell-Bond) approaches was selected as the best option to present in the paper.10.Besides the counter-cyclical behaviour of green patenting, a possible non-linearity of the relationship between innovation and GDP growth might also explain the sign obtained.In addition, one could expect the elasticity of green innovation to GDP growth to be lower than the elasticity of brown (fossil fuel) innovation.

Figure 1 .
Figure 1.Per capita primary energy mix in 2000.(Source: BP Statistical Review of World Energy June 2021.Note: Others include mostly nuclear energy).

Figure 2 .
Figure 2. Per capita primary energy mix in 2019.(Source: BP Statistical Review of World Energy June 2021.Note: Others include mostly nuclear energy).

Figure 3 .
Figure 3. Net energy imports.(Source: World Development Indicators, World Bank.Note: A negative value indicates that the country is a net exporter).

Figure 5 .
Figure 5.The evolution of environment-related technologies (Source: OECD Green Growth database).

Figure 6 .
Figure 6.Evolution of patenting activity from the environment-related selected technological domains, aggregate panel of OECD countries.(Source: OECD Green Growth database.Note: CCM stands for Climate Change Mitigation).

Table 1 .
Indicators used in the EPS index.

Table 3 .
Estimated results for green innovation.