The dynamic role of institutional quality, renewable and non-renewable energy on the ecological footprint of OECD countries: do institutions and renewables function as leverage points for environmental sustainability?

Abstract

This article examines for the first time the impact of disaggregated energy sources and institutional quality on the ecological footprint (EF) of 29 OECD countries, by explaining how the diversification in countries’ energy mix and their institutional performance are associated with sustainable environmental performance. We use panel data from 1984 to 2016 and we apply second-generation techniques to arrange the critical issues of cross-sectional dependence and heterogeneity. The applied cointegration tests expose a long-run equilibrium relationship that associates renewable/non-renewable energy consumption, economic growth, institutional quality, and the EF of OECD countries. The robust cross-sectional augmented distributed lag (CS-DL) estimator shows that economic growth and the adoption of non-renewable energies are detrimental to the environment, while the operational quality of institutions adds to ecological sustainability. Concurrently, the negative effect of renewables on EF does not seem to cause a significant beneficial impact on the environment. Moreover, there is evidence that non-renewable energy and institutional quality have a bidirectional causal association with EF. Also, a weak unidirectional causal effect is running from the EF to renewables consumption. The study further demonstrates the inefficient integration of renewable energy forms in OECD countries and the concomitant essential role of institutions on environmental sustainability by providing relevant policy orientations.

Introduction

Environmental sustainability constitutes a fundamental component of the process of sustainable development, along with the economic and social sustainability. The concept of environmental sustainability was first developed by Goodland (1995), who defines it as the ability to ensure human well-being through the maintenance of natural capital. This conceptualization was further formulated in the early twenty-first century, when OECD environmental strategy was adopted by OECD countries (OECD 2001). The strategy introduced four criteria for a sustainable environment. These criteria are the regeneration of natural resources, the substitutability capacity of fossil energy sources, the assimilation of waste, and the avoidance of irreversible adverse effects on ecosystems.

The operationalization of the former criteria for environmental sustainability is imperative for policymakers in order to eliminate climate change, global warming, pollution, and depletion of resources. Since economic growth is directly associated with environmental survival (Sharma 2011), joint policy programs tend to integrate environmental policies into the economic ones in the policy formulation arena (Jordan and Lenschow 2008). In this bidimensional policy framework, this study intends to probe the concomitant role of institutional quality and dissociated energy forms on the ecological footprint (EF) of OECD countries.

In the energy economics literature, the use of energy is a central determinant in fostering environmental efficiency (Charfeddine and Mrabet 2017). Efficient and clean energy use contributes to efficient and environmentally friendly exploitation of resources and effective environmental management (Nathaniel et al. 2020). Renewable energy (RE) exerts a significant role in national energy policies as it provides secure energy, expands countries’ energy autonomy, increases employment, and combats environmental problems (IRENA and REN21 2018; Liu et al. 2020a, 2020b). Although globally the portion of renewables¹ on the total supply of primary energy was 12.9% in 2008, most of the predictions assume a 17% share in 2030 and a 27% share in 2050² (IEO 2019). The electricity sector faces the largest and fastest increase in the use of renewables since the share of RE that covered needs in electrical power reached 24% in 2016. This was more than a double growth from 2015 and the fastest one since 1990 (IEO 2019). It is estimated that due to technological innovations and government RE policies, in 2050, renewables will account for nearly half of the world's total electricity production.

On the other hand, the dominant role of fossil fuels in the majority of developing and developed countries and the increasing public awareness about ecological sustainability intensify the implementation of effective energy policies (Fudge et al. 2008). Fossil energy is considered the main culprit of anthropogenic emissions and the acceleration of clean energy policies is vital for the innovation of the energy system (Edenhofer et al. 2011). However, the level of renewables penetration in countries’ energy mix is strongly correlated with the governmental support policies (IRENA and REN21 2018). The high cost of renewable technology implementation may hinder this penetration for the sake of short-run fast economic growth, as a result of a policy level myopia (Sharif et al. 2020b).

The implementation of effective renewable energy policies on the route for a sustainable environment presumes a robust institutional framework (Panayotou 1997; Bhattarai and Hammig 2001). The administrative and institutional settings enable and support environmental policies, not only at the policymaking level but also at the implementation and monitoring level (Lovei and Weiss 1998). Furthermore, effective institutions may influence the rate of renewables penetration in the countries’ energy mix (Bhattacharya et al. 2017), thus causing an indirect effect on the ecological protection. Iacobuta and Gagea (2010) argue that the main obstacle to sustainable development is not insufficient resources, but inadequate institutional operation. Ostrom (1998) states that institutional quality reflects the governments’ structure and effectiveness which is formulated through their designed policies and takes place in the environmental regulatory framework. Institutions could be categorized into formal which refer to the economic, political and legal institutions, and informal institutions implying societal norms and constraints, and cultural-cognitive procedures (Dasgupta and De Cian 2016). Although both types formulate the social allocation of incentives, assets, and resources (United Nations 2016), this study adopts the “formal institutions” concept in an extended form that captures the overlapping role of governance.

Effective institutions are instrumental in achieving higher economic growth by minimizing the environmental cost (Panayotou 1997). Governments’ capacity to formulate and execute policies and regulations that rule the free market, strengthens the standard of contract performance, security of property rights, rule of law, and political power impartiality (Canh et al. 2019). Governmental stability is crucial for the qualitative performance of institutions and thus for productive environmental management. Political instability can have a negative influence on the rule of law as a result of inefficient governmental performance (Abid 2016). The well-established legal and regulatory framework creates inclusive institutions, enforces a sense of trust to the public, and builds institutional transparency and accountability (Ritzen and Woolcock 2000). Impartial institutions support a strong rule of law accompanied by sound legislation, which has the potential to control the companies’ compliance with the respective environmental standards (Lau et al. 2014; Hassan et al. 2020). On the other hand, corruption may be proved to impede environmental quality indirectly. Corrupted governmental officials could promote lax environmental laws and regulations and impede the deployment of RE policies (Fredriksson et al. 2004; Wang et al. 2018). The limited stringency in environmental legislation increases the intensity of energy use, but the existence of governmental stability could regulate the scale of energy consumption (Damania et al. 2003). Besides, institutional corruption and bureaucratic barriers hinder the development and deployment of RE policies by governmental authorities (IRENA and REN21 2018).

Taking the aforementioned into account, this study is motivated by the strong public awareness about environmental issues and the challenge of formulating effective environmental and energy policies under a robust institutional environment. In a holistic approach to environmental sustainability, we intend to explore the dynamic impacts of RE consumption and institutional quality on the EF of 29 OECD countries. The study adds to the relevant literature in several ways.

Firstly, according to our investigation, this study is the first one that engages institutional measures in the energy-environmental framework of OECD countries. Examining the impact of disaggregated energy sources on their environment, we highlight the instrumental role of institutions. On that account, we employ a well-defined institutional index that takes into account the fundamental dimensions of institutional performance. Secondly, unlike previous studies, our analysis employs extended institutional data coverage for the time period 1984–2016. Thirdly, we utilize the EF as a proxy for environmental performance in OECD countries. This indicator represents not only the atmospheric pollution, but also the overall ecosystem’s damage in a heuristic way. EF encompasses the concept of ecological sustainability and reflects the human demand on nature compared to the biocapacity required for the eco-friendly operation of an economic system. Finally, most of the studies in the relevant literature use first-generation econometric techniques that do not consider the issues of cross-sectional dependency, heterogeneity, endogeneity, and the misspecification of the employed panel data dynamic models. In this work, we mainly employ second-generation tests to address methodological flaws related to these issues. We examine the long-run effects of institutional quality and disaggregated energy in the EF via the recently developed Cross-sectional Augmented Distributed Lag (CS-DL) estimator (Chudik et al. 2016) which produces robust and reliable results. We further apply the Dumitrescu and Hurlin (2012) test to detect causality effects in the panel set. To our knowledge, this study employs for the first time these novel methodological approaches to find out how the institutional environment and the disaggregated energy consumption affect the environmental performance of OECD economies.

Our findings suggest that the institutional quality in OECD countries has a healing environmental effect, while RE consumption seems to produce an insignificant effect on EF. Furthermore, the impact of non-RE consumption and economic growth on the EF is evidenced to be positive and significant. The results are in line with the reports of IRENA and REN21 (2018) and IEA, IRENA, UNSD, WB, WHO (2019) regarding the incomplete integration of renewables in the OECD countries’ energy-mix system, and the need for further relevant policy implementation.

The remaining paper has the following structure: The “Environmental sustainability and OECD countries facts and perspectives” section presents some environmental facts and perspectives in the OECD countries. The “The ecological footprint concept” section refers to the concept of ecological footprint, and the “Literature review” section reviews the relevant literature. The “Data and methodology” section describes the data, the model specification, and the methodological approach of the study. The “Empirical findings and discussion” section exhibits and comments on the empirical findings, and finally, the “Conclusions and policy orientations” section provides conclusions along with policy suggestions.

Environmental sustainability and OECD countries facts and perspectives

The increasing environmental threats and the possible irreversible consequences to human existence have facilitated the international community’s awareness about a sustainable environment. Global warming and climate change cause devastating effects on the environment, such as ecosystem collapse, glacier retreat and changes in sea level, biodiversity loss, extreme weather phenomena, desertification, and human health threats. In 2015, the CO₂ concentration in the atmosphere was 400.1 ppm (parts per million), while in 2017 it reached 405.5 ppm, which is a 146% increase above the preindustrial levels (United Nations 2019). Compared with the pre-industrial era, it is estimated that human activities are responsible for a 0.8 to 1.2°C increase in global warming, with the potential to rise above the threshold of 1.5°C in 2030–2052 (IPCC 2018). Anthropogenic emissions have irreversible consequences in the ecosystem’s sustainability, putting the biodiversity in jeopardy (United Nations 2019). The continuous decline in environmental quality and the further direct negative consequences to the social and economic dimensions of human life increased countries’ environmental awareness. This was recently demonstrated by the ratification of the Paris Agreement, in May 2019. The participants of this agreement committed to deploying nationally determined contributions (NDCs) to eliminate climate change (United Nations 2019).

While many developing countries lay less emphasis on the adoption of renewable energies due to technology and funding constraints, most developed countries have increasingly focused on clean energy policies. Until the late 1980s and during the process of their environmental policy integration (EPI), most OECD countries gradually adopted national and sectoral strategies and environmental projects. In this process, they introduced specialized institutions responsible for the monitoring of sustainable development targets (Jordan and Lenschow 2008). In 2011, OECD countries validated the Green Growth Strategy (GGS) in order to formulate a common policy framework for the promotion of efficient energy targets and environmental sustainable pathways (OECD 2012). Although energy use in non-OECD countries is projected to increase by approximately 70% until 2050, the said increase is estimated at only 15% in OECD countries. Concurrently, until 2050, the energy-induced emissions in OECD countries are predicted to decline by almost 0.2% every year, presenting a 14% decline in comparison with the 2005 levels. (IEO 2019).

OECD countries are considered to have played a pioneering role in the architecture of Sustainable Development Goals (SDGs) (Destek and Sinha 2020). In the process of accomplishing SDG 7 which calls for sustainable energy, they have intensified the use of RE sources to mitigate environmental degradation risks (SDG 13). In OECD Europe, the wind and solar renewables generation has a potential to increase from a 20% share in the renewables mix to a 50%, from 2018 to 2050. Concurrently, the ratio of fossil fuels generation for the same period is estimated to decline from about 38% to 18% of the generation mix. Furthermore, most of the energy policies in OECD countries are related to the growth in electricity demand which is mainly met by renewables. (IEO 2019).

From the first stages of their environmental management, OECD countries created national and local independent institutions and environmental agencies (Lovei and Weiss 1998). In an early study, Scruggs (1999) scrutinized the important role of policy-oriented institutions to the quality of environmental performance in OECD countries. In addition, Fredriksson et al. (2004) highlighted the negative role of corruption in formulating effective and efficient energy policies in these countries. Destek and Sinha (2020) observed that most of the OECD countries do not face an environmentally sustainable economic growth, implying the existence of ineffective institutional performance.

Industrialized countries such as OECD member states are increasingly interested in the production and use of renewables (Saidi and Omri 2020). The quality of institutions is also considered to play a vital role in their sustainable development process (Khalil et al. 2007). Moreover, OECD countries are mostly engaged in the design, adoption, and execution of SDGs 7, 8, 13, 15, and 16 (UN SDGs 2015). These goals urge the need for sustainable growth by supporting the use of clean energies and the improvement of institutional quality and capacity on environmental performance and ecological protection. Considering the lack of literature in approaching both the institutional quality and disaggregated energy effects on the environment of OECD countries, this study attempts to figure out how these important factors influence their EF. In particular, we investigate the concomitant role of institutional quality and RE/non-RE energy utilization in the ecological sustainability of OECD countries.

The ecological footprint concept

Most of the existing studies in the literature use CO₂ emissions or other pollutants as indicators that represent environmental performance (e.g., Shafiei and Salim 2014; Abid 2017; Salman et al. 2019; Liu et al. 2020a, 2020b). However, ambient air pollutants capture the environmental damage only partially, since human activities influence also the land and water of the ecosystem. Furthermore, the impact of the energy-driven economic activity on the environment should be assessed through the overall ecosystem’s damage (Destek and Sinha 2020). In this sense, the Ecological Footprint (EF) indicator extends the conventional environmental indicators into a broader perspective of ecological sustainability.

Rees (1992) and Wackernagel and Rees (1996) were the first who conceived and developed the indicator of EF. It is defined as an ecological sustainability measure that encompasses not only the total carbon footprint aspect but also the aspects of cropland, grazing, forestry, and built-up land and fisheries. EF shows how humans, regions, and countries’ demand for resources is compared with the ability of nature to regenerate and absorb the waste generated under the existing technological and resource management processes (Wackernagel and Rees 1996). It is a measure that reflects the biologically productive earth capacity required for a human or population to balance the resources’ consumption with the respective production and absorb the generated waste (Wackernagel and Monfreda 2004).

Since 1970, the world faces a gradual increase in the EF which shows that human demand exceeds the ecosystem’s biocapacity, leading to a substantial ecological deficit. Most of the developed countries face an ecological deficit (Yasin et al. 2020) since the rate of natural regeneration of resources and waste absorption cannot meet the pace of human demand on nature. The EF is an easy monitoring tool for policymakers and it is measured in global land hectares which represent the biologically productive area required for sustainable resource utilization every year (Kitzes and Wackernagel 2009; GFN 2020). In 2016, among the OECD countries, Luxemburg had the highest EF consumption per capita of 12.91 global hectares per capita (gha pc), while Mexico had the lowest one of 2.6 gha pc. Taking into account that for sustainable resource utilization and regeneration of the ecosystem the current biocapacity per person is estimated at 1.7 gha pc (GFN 2020), it is obvious that OECD countries are subjected to an escalated ecological deficit.

Literature review

Environment—renewable energy relationship

In the energy and environmental economics literature, the association between energy consumption and environmental quality has been the topic of intensive research. Most of the relevant studies examine energy consumption at an aggregated form (Menyah and Wolde-Rufael 2010a, 2010b; Arouri et al. 2012; Abul et al. 2019; Abbasi et al. 2020; Ansari et al. 2020; Nawaz et al. 2020) and therefore do not consider the distinct effects of disaggregated energy sources on the environment. Since this study decomposes energy consumption in renewable and non-renewable, we focus on the distinctive effects of these two forms of energy on the environment.

Most of the studies demonstrate that the use of renewables has a substantial contribution to environmental sustainability. Dogan and Seker (2016a) followed an Environmental Kuznets Curve (EKC) approach from 1980 to 2012 in the European Union (EU). They concluded that RE and trade decrease CO₂ emissions in the EU, while non-renewables exert a respective increasing effect. The same authors (Dogan and Seker 2016b) presented similar findings for the top 23 RE-consuming countries. Lee (2019) investigated the dynamic relationship between RE consumption and CO₂ emissions in the EU from 1961 to 2012. He concluded that RE has been used successfully to mitigate the EU environmental deterioration in both the short and the long term. Paramati et al. (2017) showed the positive contribution of RE on mitigating CO₂ emissions using data from the eleven next generation emerging economies (N-11)³ in the period 1990–2012. Sharif et al. (2019) also found that RE promoted environmental quality in 74 nations during the period 1990–2015.

In a single-country study, Danish et al. (2017) conducted the Autoregressive Distributed Lags (ARDL) bound test for Pakistan from 1970 to 2012. They concluded that RE consumption is environmentally friendly, as opposed to non-RE, which is considered to be the main culprit for environmental degradation. Besides, they reported a bidirectional causality between the two types of energy consumption and CO₂ emissions. Dogan and Ozturk (2017) also adopted an ARDL approach in an EKC model and confirmed that RE consumption exerts a negative influence on CO₂ emissions in the US. In the case of Malaysia, Bekhet and Othman (2018) used the Vector Error Correction (VECM) Granger causality, the Dynamic Ordinary Least Squares (DOLS), and Fully Modified Least Squares (FMOLS) methodologies. Their findings showed that RE mitigates CO₂ emissions reporting a unidirectional causal effect from the latter to RE use.

Sharif et al. (2020b) investigated the relationship between RE and environmental quality using the Quantile-on-Quantile methodological approach for the 10 mostly polluted countries. Though they concluded to a two-way causality between RE and environmental performance, they provided mixed results. In 15 countries where RE occupies a critical role in their energy mix, Saidi and Omri (2020) showed the positive effect of RE on environmental quality, positing no long-run causality. Nathaniel S.et al. (2020) used data from the MENA⁴ countries and applied the Augmented Mean Group (AMG) method. They pointed out that factors such as financial development, urbanization, growth, and fossil fuel energy use degrade the environment and RE does not exert a substantial impact on eliminating this degradation. Similarly, Menyah and Wolde-Rufael 2010a, 2010b found that the adoption of RE does not have a substantial impact on the US environmental quality. Lin and Moubarak (2014) found no causal association between RE consumption and CO₂ emissions in China either in the short or in the long run. In contrast with the previous studies, Apergis et al. (2010) examining 19 countries of different stages of development found that though nuclear energy consumption mitigates CO₂ emissions, RE consumption has the opposite effect. Furthermore, Boluk and Mert (2014) indicated that both renewable and fossil energy use is harmful to the environmental quality of 16 EU countries. Using a panel fixed effects model, they showed that the effect of renewables on the environment is half per energy unit harmful than the effect of non-renewables. The challenge of investigating environmental sustainability in a broad ecological perspective gave birth to studies that adopted the ecological footprint (EF) variable as a novel environmental indicator. Destek et al. (2018) examined the existence of the EKC using the EF of 15 EU countries. They found that RE and trade openness have a positive environmental impact while non-RE a negative one. Danish et al. (2019) explored how the real GDP, urbanization, natural resources, and RE affect the EF consumption in BRICS⁵ countries. They found that RE consumption decreases the EF of the examined countries. In the same line, Wang and Dong (2019) using the AMG estimator concluded an adverse relationship between RE and the EF of 14 Sub-Saharan Africa (SSA) countries. They also reported a long-run unidirectional causal effect from RE use to their EF. In a study for Turkey, Sharif et al. (2020a) used the Quantile ARDL (QARDL) to investigate how RE and non-RE consumption influence its EF. Findings showed that RE influences negatively Turkey’s EF confirming a bidirectional causal effect between them.

The promotion of RE forms in OECD countries during the last decades and their endeavor of promoting environmental quality through changes in their energy policies have been demonstrated by some recent studies. For a sample of 25 OECD countries over the period 1980 to 2011, Apergis and Payne (2014) showed that RE reduces per capita CO₂ emissions in the short-run. In a study by Alvarez-Herranz et al. (2017) for 17 OECD countries from 1990 to 2012, findings revealed the important role of RE adoption and energy innovation in mitigating greenhouse gas emissions. Shafiei and Salim (2014) utilized a STIRPAT⁶ model for 29 OECD countries over the same period. They confirmed the positive role of non-RE and the negative role of RE on CO₂ emissions in the long-run. Zaghdoudi (2017) probed how oil prices and RE affected the environment in 26 OECD countries from 1990 to 2015 in an EKC framework. Findings showed that low oil prices expanded the use of fossil fuels which harms the environment and scrutinized the significance of promoting RE with a view to eliminate environmental pollution. In a recent study, Destek and Sinha (2020) used second-generation techniques to investigate the role of RE and non-RE consumption, together with the impact of trade openness, on the EF of 24 OECD countries. Investigating the period 1980–2014, the results confirmed the contribution of renewables to promoting environmental sustainability by decreasing the EF of OECD countries.

Environment—institutional quality relationship

While several studies investigate the factors that influence environmental performance, there is little evidence in the literature about the role of institutions in promoting environmental sustainability. Approaching the environmental-social dimensions of sustainable development, Lehtonen (2004) emphasized the important role of institutions at the national level as they have to arrange the appropriate framework for local activities that promote environmentally sustainable development. In a political economics approach, Scruggs (1999) investigated the nature of the relationship that associates the political-economic institutions of 17 OECD countries with their environment, from 1970 to the 1990s. He found that the corporatist institutions⁷ are associated with better environmental quality, whereas pluralistic institutions⁸ do not have a significant effect on it.

Bhattarai and Hammig (2001) showed the critical role of political institutions and governance in the deforestation process of 66 Latin American, African, and Asian countries. They found that institutional and governance improvements reduce the levels of deforestation in Latin America and Africa but increase them in Asia. Culas (2007) presented evidence that in 14 countries of Latin America, Africa, and Asia, improvements in institutions related to secure property rights for forests could eliminate deforestation. He showed that the contract enforceability of government and the efficiency of bureaucracy can promote forestland conservation. Dutt (2009) examined the impact of institutional indicators in an EKC model of 124 countries from 1984 to 2002. Findings revealed the crucial role that political institutions, governance, and socioeconomic conditions play in reducing CO₂ emissions. Examining 28 EU countries, Castiglione et al. (2012) reported the principal role of a strong rule of law in achieving environmental sustainability. Gani (2012) showed that rule of law, political stability, and control of corruption promote the environmental quality of 99 developing countries, while government effectiveness and regulatory quality do not have a significant effect on it. Ibrahim and Law (2016) applied the system Generalized Method of Moments (SYS-GMM) in a panel of 40 Sub-Saharan countries, identifying the importance of institutional arrangements in environmental performance. They also showed the important institutional influence in the relationship between trade and the environment. Abid (2016) demonstrated that political stability, effective governance, and the reduction of corruption decrease CO₂ emissions in 25 SSA countries, while the well-defined regulatory framework and rule of law increase them.

Liu X. et al. (2020) emphasized the three forms of governance (political, economic, and social-institutional) in the context of the EKC hypothesis for 5 high polluting countries (China, India, Japan, Russia, and the USA). They utilized the DOLS and FMOLS methods providing evidence for the catalytic role of efficient governance on environmental performance. Abid (2017) argued that the level of environmental quality is vitally connected with the nature of economic growth, and institutions play a key role in this process. Using data from 58 MEA⁹ and 41 EU countries, he found that institutional quality reduces the environmental degradation of EU countries but has no effect on the environment of the MEA region. Ali S. et al. (2020) examined the impact of institutional efficiency, along with other factors such as urbanization, foreign direct investment (FDI) inflows and trade, on the EF of 47 OIC¹⁰ countries. They observed that except for institutions all the other factors have a positive relationship with the EF.

Environment—energy consumption—institutional quality relationship

The third strand of the literature incorporates both energy utilization and institutional measures in the examination of the effects of economic development on the environment. Tamazian and Rao (2010) used panel data of 24 transition economies and showed that institutional quality promotes environmental protection, while financial liberalization exerts a positive influence in the environment only in the presence of strong institutions. Salman et al. (2019) incorporated the factors of institutional efficiency and energy consumption in the examination of the relationship between economic growth and CO₂ emissions in 3 East Asian countries. Using “law and order” as an institutional indicator, they found that the quality of local institutions is important to lessening environmental degradation in these areas. In a panel of 47 developing countries, Ali H.S. et al. (2019) applied the system GMM estimation which revealed that economic growth, trade openness, energy use, and urbanization increase CO₂ emissions while institutional quality produces the opposite effect.

Bhattacharya et al. (2017) highlighted the positive role of institutions and RE consumption in the environmental quality of 85 countries by applying the system GMM and FMOLS estimators. Sarkodie and Adams (2018) used an ARDL model to analyze how renewable, fossil, and nuclear energy along with economic growth, urbanization and institutions affect the environment of South Africa. Findings showed that in the long-term, nuclear energy provokes an insignificant effect on South Africa’s environment, while non-RE consumption is harmful to it. On the contrary, RE use and robust political institutions exert a beneficial effect on the environment. Al Mulali and Ozturk (2015) employed the FMOLS to find how energy consumption and political stability along with other factors, such as urbanization and industrial development, influence the EF of 14 MENA countries. They concluded that although energy consumption and the other factors increase EF consumption, political stability reduces it. Likewise, in an EKC framework, Yasin et al. (2020) established that political institutions promote the environmental sustainability of 53 developed and 57 developing countries, while energy consumption increases their EF.

In contrast with the abovementioned studies, Azam et al. (2020) explored the effects of institutional quality on 5 environmental indicators and energy consumption of 66 developing countries using the system GMM method. They concluded that the quality of institutions seems to increase environmental degradation, as well as promotes the use of oil and fossil-based energy sources. In the same line, Hassan et al. (2020) established a positive relationship between institutional quality, energy consumption, and CO₂ emissions in Pakistan applying an ARDL model. Le and Ozturk (2020) used the Common Correlated Effects Mean Group (CCE-MG), AMG, and Dynamic CCE (DCCE) estimators to conclude that both institutional quality and energy use raise the environmental pollution in 47 EMDEs¹¹. Charfeddine and Mrabet (2017) employed an EKC model and used the EF as an environmental proxy in 15 MENA countries. Their results, based on DOLS and FMOLS estimators, revealed that energy consumption and institutional quality (represented by political rights and civilian freedom) increase the EF in these regions.

Obviously, the studies that incorporate institutional and energy measures in the identification of potential factors that influence environmental performance are mostly recent. However, as regards the role of institutions and renewables on the environment, a divergence is observed in the relevant empirical findings. This could be explained by the special socioeconomic conditions of the country groups or regions being examined, the differences in the time span, model specifications, and methodologies being implemented as well as the environmental indicator employed. Regarding the OECD countries, the literature, except for the study of Scruggs (1999), lacks relevant empirical research. Furthermore, although the energy sector is strongly connected with the level of sustainable development, no study establishes a common framework in examining the impact of disaggregated energy utilization and institutional quality on the environment of OECD countries. This study attempts to fill this gap by providing empirical evidence through the employment of recent econometric techniques. Using second-generation unit root tests and cointegration methods is able to account for the critical issues of cross-sectional dependence and heterogeneity.

Data and methodology

Data and model specification

The study provides an insight into the role of institutions in a disaggregated energy-environmental model for OECD countries. We use annual data of 29 OECD countries (Appendix Table 12) over the period 1984–2016, and we focus on a second-generation panel data analysis approach. Unlike the majority of existing studies, and similar to Al Mulali and Ozturk (2015), Charfeddine and Mrabet (2017), Destek et al. (2018), Danish et al. (2019), Sharif et al. (2020a), and Ali et al. (2020), we employ EF as an environmental indicator. As described previously (“The ecological footprint concept” section), this indicator expresses more broadly and inclusively the level of environmental performance. Moreover, in our model specification, we use the real GDP, RE consumption, non-RE consumption, and institutional quality as independent variables. Thus, in line with Bhattacharya et al. (2017), we formulate the following model to capture the distinct impact of the disaggregated energy sources and the influence of institutions in the environment of OECD countries during the process of economic growth:

$$ {EF}_{it}=f\ \left({GDP}_{it},{RE}_{it},{NRE}_{it},{IQ}_{it}\right) $$

(1)

In Eq. (1) EF_it is the ecological footprint consumption in total global hectares (gha); GDP_it is the real GDP in billions of constant 2010 US dollars as a proxy of economic growth; RE_it and NRE_it are the renewable electricity net consumption¹² and fossil fuels electricity net consumption¹³ in billion kilowatt-hours; and IQ_it is an institutional quality index created through the Principal Component Analysis (PCA) of 6 basic institutional sub-indices: Socioeconomic Conditions (SEC), Government Stability (GS), Corruption (COR), Law and Order (LO), Democratic Accountability (DA), and Bureaucracy Quality (BA).

The institutional quality index constitutes a composite measure of the fundamental institutional dimensions of the OECD countries. These dimensions are reflected by the respective social, constitutional (political and democratic), and administrative-judicial systems (basic institutional quality). The PCA approach produces an index that comes from the weighted factor loadings of originally correlated variables. The i_thprincipal component is the result of the linear combination of m original variables:

$$ {IQ}_i={w}_{i1}\ast {X}_1+{w}_{i2}\ast {X}_2+\dots +{w}_{im}\ast {X}_m $$

(2)

where ${w}_{i1}^2$+${w}_{i2}^2+\dots +{w}_{im}^2=1,\kern0.5em i=1,2,\dots, m$; w_ij is the weight of the j variable in the i_thcomponent; X_m are the original indicators; and m the number of variables. PCA overcomes the problem of collinearity between individual indicators and forms a composite one that encompasses the most crucial information from the original indicators and expresses the highest possible variation of the original dataset (OECD 2008).

The data for EF_it were obtained from the Global Footprint Network (GFN), the data for GDP_it from the World Development Indicators (WDI) (2020), and the data for RE_it and NRE_it from the US Energy Information Administration (EIA) (2020). The data for the 6 original institutional indicators which formed the IQ_it index were taken from the International Country Risk Group (ICRG) (Table 1). Table 2 presents some descriptive statistics of the model’s variables. The US exhibits the highest EF in 2005, whereas Luxemburg the lowest one in 1984. Also, the US has the largest portion in RE utilization in 2016, while the lowest user of RE is Israel in 1989. It is also observed that the US has the highest level of non-RE consumption in 2007, while Norway is the lowest consumer of fossil fuel energies in 1991. Switzerland seems to have the most qualitative institutional framework in 1984, and Turkey the least qualitative one in 1990. The standard deviation statistics imply a heterogeneous behavior across cross-sectional units. In our model, all variables except for the IQ_it are expressed in natural logarithmic form, thus equation (1) can be written in the following log-linear form:

$$ {LEF}_{it}={\beta}_{0i}+{\beta}_{1i}{LGDP}_{it}+{\beta}_{2i}{LRE}_{it}+{\beta}_{3i}{LNRE}_{it}+{\beta}_{4i}{IQ}_{it}+{\varepsilon}_{it} $$

(3)

where i = 1,...,N the cross-sectional units (countries); t = 1,... T, the time period (1984-2016); β_1i, β_2i, β_3i, and β_4i parameters that represent the coefficient estimates and e_itis the error term. The study is based on a panel data analysis that takes into account the heterogeneity and cross-sectional (CS) dependency and focuses on second-generation methodologies and econometric approaches.

Table 1 Variables description

Table 2 Descriptive statistics

Table 3 Principal component analysis of the IQ index

Table 4 Cross-section dependence tests results

Table 5 Slope heterogeneity test results

Table 6 Unit root tests results (first generation)

Table 7 Unit root tests results (second generation)

Table 8 Pedroni cointegration tests results

Table 9 Westerlund Cointegration tests results

Table 10 Long-run estimates

Table 11 D-H panel causality test results

Cross-sectional dependency and slope heterogeneity

One of the principal factors that directly influence the adopted methodological strategy in panel data analysis is the issue of errors’ CS dependence (De Hoyos and Sarafidis 2006). In the era of globalization, this is a predominant phenomenon in panel data estimations due to socioeconomic linkages among countries or hidden common effects and regional characteristics (Chudik and Pesaran 2013). When CS correlation is present, the conventional estimation methods are criticized for providing inconsistent and unreliable results (Kapetanios et al. 2011). For this reason, our first step is to examine the existence of CS dependency in the residuals of the model through the Breusch and Pagan (1980) LM test, Pesaran Scaled LM test (2004), Pesaran CD test (2004), and Pesaran bias-adjusted LM test (2008).

Considering a typical panel data model of the following form:

$$ {y}_{it}={a}_i+{\beta}_i{\boldsymbol{X}}_{it}+{\varepsilon}_{it} $$

(4)

where i = 1,...,N and t = 1,... T express the cross-sections and time escalation respectively; a_i and β_i represent the individual intercepts and slope coefficients; and X_it is a k × 1 vector of the independent regressors. The null hypothesis when testing the existence of CS dependence is Cov(ε_itε_jt) = 0 and shows CS independency and the alternative is Cov(ε_itε_jt) ≠ 0 and indicates that at least two CS units are correlated with each other. The Breusch and Pagan (1980) LM test which follows the asymptotic χ² distribution is appropriate when the dimension of N is fixed and T is large and it is expressed by the following LM statistic:

$$ LM=\sum \limits_{i=1}^{N-1}\sum \limits_{j=i+1}^N{T}_{ij}{\hat{\rho}}_{ij}^2\sim {\chi}_{N\left(N-1\right)/2}^2 $$

(5)

where ${\hat{\rho}}_{ij}$ are the values of the pair wise correlation of the errors of Eq. (4) using ordinary least squares (OLS).

Pesaran (2004) introduced a test (the so-called scaled LM test) to overcome the limitation of small N, i.e., for N →  ∞ , T →  ∞ , which is asymptotically normally distributed and is denoted by the following statistic:

$$ {LM}_S=\sqrt{\frac{1}{N\left(N-1\right)}}\ \sum \limits_{i=1}^{N-1}\sum \limits_{j=i+1}^N\left({T}_{ij}{\hat{\rho}}_{ij}^2-1\right)\sim N\left(0,1\right) $$

(6)

Due to the fact that the LM_S test is subjected to size distortions when T is small and N is large, Pesaran (2004) also suggested the CD test which is based on the average of the ${\hat{\rho}}_{ij}$ values. Unlike the previous tests, the CD test addresses the problem of size distortions, it is consistent for fixed T and N values and it performs well even in the case of small T and N (Pesaran 2004). The CD statistic is described as follows:

$$ CD=\sqrt{\frac{2T}{N\left(N-1\right)}}\ \left(\ \sum \limits_{i=1}^{N-1}\sum \limits_{j=i+1}^N{\hat{\rho}}_{ij}\right)\sim N\left(0,1\right) $$

(7)

Nevertheless, the CD test lacks power in cases that the population mean pair-wise correlation, in contrast with the individual ones, is zero. For this reason, Pesaran et al. (2008) introduced the bias-adjusted LM test as follows:

$$ {LM}_{adj}=\sqrt{\frac{2}{N\left(N-1\right)}}\ \sum \limits_{i=1}^{N-1}\sum \limits_{J=i+1}^N\frac{\left(T-k\right){\hat{\rho}}_{ij}^2-{\mu}_{T_{ij}}}{V_{T_{ij}}}\sim N\left(0,1\right) $$

(8)

where ${\mu}_{T_{ij}}$ and ${V}_{T_{ij}}$ denote the exact mean and variance of $\left(T-k\right){\hat{\rho}}_{ij}^2$ respectively.

In accordance with the CS dependence test, another crucial issue that defines the next methodological steps is the issue of slope homogeneity across the individual units of the panel. In most cases the CS homogeneity hypothesis does not correspond to the real world (Urbain and Westerlund 2006). For this reason, we utilize the $\overset{\sim }{\varDelta }$ test of Pesaran and Yamagata (2008) which is based on the modified Swamy’s (1970) statistic ($\overset{\sim }{S}$), and it is appropriate in panels where both the N and T dimensions could be large in their relative expansion dynamics. In the case of non-normal errors and exogenous regressors, and under the premise that$\sqrt{N}/{T}^2\to 0$ as (N, T) →  ∞ ,the ($\overset{\sim }{\varDelta }$) test follows the standard normal distribution:

$$ \overset{\sim }{\varDelta }=\sqrt{N}\left(\frac{N^{-1}\overset{\sim }{S}-k}{\sqrt{2k}}\right) $$

(9)

where $\overset{\sim }{S}$ is the Swamy statistic¹⁴ and k the explanatory variables. The bias-adjusted ${\overset{\sim }{\varDelta}}_{adj}$ test meliorates the small sample properties of the $\overset{\sim }{\varDelta }$ statistic under the premise that errors are normally distributed. This test requires no restrictions on the expansion dynamics of N and T, as (N, T) →  ∞ :

$$ {\tilde{\varDelta}}_{adj}=\sqrt{N}\left(\frac{N^{-1}\tilde{S}-E\left({\tilde{z}}_{it}\right)}{\sqrt{\mathit{\operatorname{var}}\left({\tilde{z}}_{it}\right)}}\right) $$

(10)

where $E\left({\tilde{z}}_{it}\right)=k$ and $\mathit{\operatorname{var}}\left({\tilde{z}}_{it}\right)=\frac{2k\left(T-k-1\right)}{T+1}$ are the mean and variance respectively. The null hypothesis of these tests assumes slope homogeneity (H₀ : β_ι = β),whereas the alternative one assumes heterogeneity of the CS_s slope coefficients (H₁ : β_ι ≠ β_j, i ≠ j) for a non-zero fraction of the pairwise slopes. For dynamic panels, this test produces consistent results when T ≥ N (Pesaran and Yamagata 2008), thus it is considered suitable for our study.

Panel unit-root tests

The application of unit root tests is necessary to identify the stationarity properties of the variables. Conventional first-generation tests¹⁵ fail to consider the issues of CS-dependent distribution of the individual units in panels and heterogeneity of slopes. To overcome these problems, Pesaran (2007) suggests the CS augmented Dickey-Fuller (CADF) and CS augmented Im, Pesaran, and Shin (CIPS) unit root tests. These are second-generation tests that produce robust results when panels are subject to CS dependence and heterogeneity.

CADF test extends the standard DF or ADF regressions by adding lagged CS means of the individual units and first differences. In this way, it filters out the observed CS dependence which is caused by a possible unobserved common factor effect. CADF test builds upon the heterogeneous regression below:

$$ {\varDelta y}_{it}={a}_i+{b}_i{y}_{it-1}+{c}_i{\overline{y}}_{it-1}+{d}_i\varDelta {\overline{y}}_{it}+{u}_{it} $$

(11)

where a_i is the constant term, ${\overline{y}}_{i,t-1}$ is the one lag value of the CS mean of y_{i, t}and $\varDelta {\overline{y}}_t$ the first differences of ${\overline{y}}_t$. The number of lags in Eq. (11) can be modified accordingly in order to filter out the presence of serial correlation in the error term (Pesaran 2007). The CIPS test is estimated by the average of individual CADF statistics of the i_th cross-sectional units:

$$ CIPS=\left(\frac{1}{N}\right)\sum \limits_{t=1}^N{CADF}_i $$

(12)

The null hypothesis in CADF and CIPS tests assumes a homogenous unit root for all individuals within a panel, whereas the alternative one supports that there is at least one stationary individual in the panel. Due to the detected CS dependency and slope homogeneity in our panel data sample, this study focuses on the aforementioned second-generation tests to avoid wrong inferences.

Panel cointegration tests

Under the premise that the variables are integrated of order one [I(1)], i.e., they are stationary at first differences, the next methodological step is to trace the existence of a cointegration relationship. Pedroni (1999, 2004) panel cointegration tests take into account the CS heterogeneity and detect if there is of a long-run relationship in the examined model. Considering the following equation:

$$ {y}_{it}={a}_i+{d}_it+{b}_{1i}{x}_{1i,t}+{b}_{2i}{x}_{2i,t}+\dots +{b}_{mi}{x}_{mi,t}+{e}_{i,t} $$

(13)

t is the number of observations; i the number of CS units; m the number of regressors; and a_i, b_i the intercept and slope coefficients respectively, which can vary across individual units. Provided that all the examined variables are I (1), the null hypothesis assumes that the error term e_{i, t} has a unit root and so there is no cointegration. The alternative hypothesis supports that the error term is stationary implying the existence of cointegration. Pedroni suggests seven statistics that consist of four within-dimension and three between-dimension statistics. The main characteristic of within-dimension statistics is that they consider pooled autoregressive coefficients of the residuals which test them for unit roots. In contrast, between-dimension cointegration statistics are group mean statistics calculated by estimators which consider the average of each unit’s individual autoregressive coefficients.

Although the residual-based Pedroni tests are utilized to detect long-run associations in heterogeneous panels, the nature of these tests fails to consider the issue of CS dependence. Westerlund (2007) fills this gap by proposing four panel error-correction based cointegration tests (G_τ, G_a, P_τ, P_α) that overcome any restriction induced by a common factor. According to him, these tests perform greater than the residual-based tests in terms of size accuracy and increase of power. Considering the following error-correction model:

$$ {\varDelta y}_{it}={\boldsymbol{\theta}}_i^{\prime }{d}_t+{z}_i\left({y}_{it-1}-{\omega}_i^{\prime }{x}_{it-1}\right)+\sum \limits_{j=1}^{v_i}{z}_{ij}{\varDelta y}_{it-j}+\sum \limits_{j=0}^{v_i}{\mu}_{ij}{\varDelta x}_{it-j}+{\varepsilon}_{it} $$

(14)

d_t holds the deterministic terms; θ_i is a vector of the relative parameters; and z_i is the error correction term that is tested if is equal to zero or not. The null hypothesis assumes the lack of cointegration when the error correction term is zero, while the alternative one accepts the existence of cointegration. All the test statistics are normally distributed and the G_τ and G_a statistics reveal cointegration to at least one of the cross-sections, while the P_τ and P_α statistics refer to cointegration in the whole panel. In the presence of CS dependence, Westerlund (2007) proposed a bootstrap approach to eliminate inconsistent estimates even when more generic types of CS dependence exist.

Long-run parameter estimates

The investigation of long-run effects in a structural model is of utmost importance as it is joined with the equilibrium state of the model (Chudik et al. 2016). In the presence of CS dependence and heterogeneity in panel data analysis, conventional estimation techniques provide spurious inference (Chudik and Pesaran 2013). The recent dynamic approaches of the CS-ARDL and CS-DL estimators (Chudik et al. 2016) overcome the previous challenges and produce robust long-run estimators. This study focuses on the CS-DL estimator which tends to perform better in small sample sizes where 30 ≤ T ≤ 50, in comparison with the CS-ARDL estimator that is consistent only in large samples where T > 50 (Chudik et al. 2016).

In order to examine the relationship between the x_it regressors and the y_{i, t} dependent variable, we consider an ARDL (ν, ρ) model specification in a panel data context as follows:

$$ {y}_{i,t}=\sum \limits_{j=1}^{v_{y_i}}{a}_{i,j}{y}_{i,t-j}+\sum \limits_{j=0}^{\rho_{x_i}}{\beta}_{i,j}^{\prime }{\boldsymbol{x}}_{i,t-j}+{\varepsilon}_{i,t}, $$

(15)

where x_{i, t}is a k × 1 vector of regressors and ${\varepsilon}_{i,t}={\theta}_i^{\prime }{\boldsymbol{f}}_t+{e}_{it}$, where f_t is a vector of unobserved common factors; and ${v}_{y_i},{\rho}_{x_i}$ are the optimal lag lengths to ensure that ε_{i, t} does not exhibit a serial correlation process across cross-section units. Equation (15) can be expressed as a cointegrating relationship in the following form:

$$ {y}_{i,t}={\boldsymbol{\varphi}}_i{x}_{it}+{\gamma}_i^{\prime }(L){\varDelta x}_{it}+{\overset{\sim }{\varepsilon}}_{i,t} $$

(16)

where φ_i is the vector of the long-run coefficients. The CS-DL estimator is structured as follows:

$$ {y}_{i,t}={c}_{yi}+{\varphi}_i^{\prime }{x}_{it}+\sum \limits_{j=0}^{\rho_{x_i}-1}{\mu}_{ij}{\varDelta x}_{i,t-j}\sum \limits_{j=0}^{v_{\overline{y}}}{\pi}_{y, ij}{\overline{y}}_{i,t-j}+\sum \limits_{j=0}^{\rho_{\overline{x}}}{\pi}_{x, ij}^{\prime }{\overline{x}}_{i,t-j}+{\varepsilon}_{i,t} $$

(17)

where ${\overline{x}}_{it}$ and ${\overline{y}}_{it}$ are the respective average values of x_it and y_it; ${\rho}_{\overline{x}}=\left[\sqrt[3]{T}\right]$ , $\rho ={\rho}_{\overline{x}}$ and ${v}_{\overline{y}}=0$. Except for the main merit of this estimator considering its consistent performance when the sample’s time dimension lies between 30 and 50 periods¹⁶, it is also robust in the presence of unit roots in independent variables and/or unobserved factors. Moreover, this estimator is reliable if serial correlation exists in e_it and f_t and when a number of hidden common factors exist, and it is consistent when there are possible breaks in e_it and dynamic wrong specifications (Chudik et al. 2013).

In addition to the CS-DL estimator, for comparison reasons, this study also employs the group-mean (MG) panel DOLS estimator (Pedroni 2001) which addresses the issues of heterogeneity, endogeneity, and autocorrelation in the detection of long-run relationships. The use of leads and lags which form the parametric nature of DOLS make this estimator perform better as T increases than the nonparametric FMOLS estimator (Wagner and Hlouskova 2010). Concomitantly, DOLS demonstrates greater performance in the case of small samples than the FMOLS estimator (Kao and Chiang 2001). Nevertheless, this estimator is robust only in the presence of CS dependence induced by time factors and is not consistent in the presence of other forms of CS dependency (Pedroni 2001).

Panel causality test

As a final part of the methodological analysis, the study attempts to investigate the existing causal associations in the model applying the Dumitrescu and Hurlin (2012) D-H causality test. This step is critical for implementing relevant policy design. The D-H test is suitable for heterogeneous panels subjected to CS dependence and it is robust in small sample performance (either small T or N). However, it requires all the observed variables to be stationary at the same level. It is implemented on the basis of the Granger (1969) non-causality test by averaging the individual Wald statistics of CS units. The D-H model specification is described as follows:

$$ {y}_{i,t}={c}_i+\sum \limits_{l=1}^L{a}_i^{(l)}{y}_{i,t-l}+\sum \limits_{l=1}^L{b}_i^{(l)}{x}_{i,t-l}+{e}_{i,t} $$

(18)

It is a fixed coefficient VAR model where x and y are the model’s variables examined for N cross-section units under T periods; L is the lag structures; ${a}_i^{(l)}$and ${b}_i^{(l)}$denote the autoregressive parameters and the coefficient slopes that differ across CS groups. The null Homogeneous Non-Causality hypothesis assumes the absence of causal relationships in any of the CS units. The alternative Heterogeneous Non-causality hypothesis considers the existence of causality at least for a subgroup of them (Dumitrescu and Hurlin 2012). The mean Wald statistic which tests the null hypothesis is:

$$ {\overline{W}}_{N,T}=\frac{1}{N}\sum \limits_{i=1}^N{W}_{i,T} $$

(19)

where W_i, T is the individual units’ Wald statistics. Dumitrescu and Hurlin (2012) show that as T initially and N sequentially increase, the test statistic converges to a standard normal distribution. In contrast with previous approaches to the D-H test, this study applies the bootstrap estimation procedure of the test, which is appropriate when CS units are not independent.

Empirical findings and discussion

The construction of an institutional quality index (IQ) that reflects the basic institutional dimensions of the examined OECD economies is provided through the PCA analysis displayed in Table 3. All six individual indicators are assigned equal weights in forming the principal components. The eigenvalues reported in the upper part of this table show the maximum eigenvalue 3.187 of the first factor, 1.087 of the second factor, and 0.657, 0.516, 0.302, 0.247 the respective eigenvalues of the third, fourth, fifth, and sixth factors. These values indicate that the first principal component explains 53.13% of the overall standardized variance, while the proportional variations of the second and third factors are 18.13% and 10.96% respectively. The proportions of the fourth, fifth and sixth factors are found lower than 10%. In the lower part of the table, it is clear that only the first principal component’s loadings exhibit no negative values, so we use this for the construction of our index. The factor loadings suggest the individual contributions to the overall variance of the first principal component. As can be seen, the highest contributions are that of “bureaucracy quality” (with a weight of 0.497), “law and order” (weigh 0.466), “democratic accountability” (weigh 0.412), and corruption (weigh 0.407).

The most critical point that defines the methodological route that is going to be followed in the study is the CS dependency test. Table 4 depicts the results of the four relative tests of Breusch-Pagan (1980) and Pesaran (2004, 2008) mentioned in the previous section. All tests reject the null hypothesis of CS independence which means that unequivocally, in a globalized economic environment, there are common factors that affect OECD countries. In addition, Table 5 shows the findings of the slope heterogeneity test of Pesaran and Yamagata (2008) where the null hypothesis of homogeneity is rejected at a 1% level of significance. The rejection of slope homogeneity reveals a country-specific heterogeneity in the examined panel. This finding implies that the homogeneity restriction in the investigation of the causal relationships between the variables of interest could lead to wrong estimations. The presence of CS correlation and heterogeneity indicates that the second-generation tests probably provide more consistent inferences than the first-generation ones.

For the identification of unit roots in the variables of our model, we employ the CADF and CIPS tests (Pesaran 2007) which are robust to the dependency and heterogeneity of the CS units. However, for comparison reasons, we both apply the first generation unit-root tests of Im, Pesaran, and Shin (2003) (IPS), and the Fisher-Augmented Dickey-Fuller (ADF) (Maddala and Wu 1999). Table 6 presents the results of the first generation unit root tests and Table 7 displays the respective results of the second-generation tests. The outcome in Table 6 shows a mixed order of integration for the variables of the study, as some of them are stationary at levels either with constant or with constant and trend. However, all variables become stationary at their first differences. In contrast with the results from the first-generation tests, the results displayed in Table 7 demonstrate that all the variables of our model contain unit roots at levels but become stationary at first differences. These findings allow us to move forward with detecting if the independent variables of the model have a long-run association with the ecological footprint of OECD countries.

For the detection of a cointegration relationship in the model, we first apply the Pedroni panel cointegration test (1999, 2004) which is applicable for heterogeneous panels. As it is shown in Table 8, almost all the statistics of this test (six out of seven), strongly indicate the existence of cointegration in the panel since the null hypothesis of no cointegration is rejected at the 1% level of significance. Although the residual-based test of Pedroni is pioneering in tracing long-run relationships, it ignores the serious issue of CS dependence. All the Pedroni statistics fail to control the detected cross-sectional dependence in the panel which occurs due to unobserved common factors. Thus, they cannot lead to reliable conclusions about the long-run association between the variables of the examined model. Therefore, we utilize the Westerlund (2007) bootstrap test statistics as described in the preliminary methodological analysis to consider robustly our previous finding of CS dependence. Table 9 reviews the results of this test, where the G_τ, P_τ (at the 5% level of significance), and P_α (at the 10% level) statistics provide clear evidence for the rejection of the null hypothesis of no cointegration. Since the Gτ and Pτ statistics are considered to be most robust when CS correlations exist (Westerlund 2007), the results of the respective bootstrapped p-values provide strong evidence in favor of cointegration. These outcomes validate the Pedroni tests which evidence a long-run relationship that interrelates RE and non-RE consumption, economic growth, and institutional quality with the steady-state of OECD countries’ ecological footprint.

Having established the existence of cointegration in our model, we proceed with the investigation of the long-run coefficients accounting for the issues of heterogeneity and CS dependence. From this perspective, we report the outcomes of the long-run CS-DL estimator in Table 10. The results displayed in this table reveal that the effects of non-RE consumption and real income on the EF are negative and significant. In addition, institutional quality and RE consumption exert a negative effect on EF though the RE effect is not statistically significant. In particular, a 1% increase in non-RE use increases the EF of the examined OECD countries by 0.22% and a 1% increase in the real GDP increases the EF by 0.87%. Moreover, as institutional quality raises by 1%, EF shows a decrease of 1.2%.

Furthermore, we report the results of the long-run parameters using a first- generation estimator. The DOLS-MG estimator provides similar coefficient signs with the CS-DL estimator and it unveils that all the independent variables have a statistically significant impact on EF. Specifically, this estimator shows that a 1% increase in RE decreases the EF by 0.34%, while a 1% increase in non-RE consumption causes a rise of 0.21% in the EF. Concurrently, a 1% increase in real GDP increases EF by 0.69%, while a 1% increase of institutional quality lessens EF by 2.6%. Although the findings of both estimators have similarities, this study considers the CS-DL estimation procedure as the benchmark methodological approach. This is because the CS-DL estimator performs better when the time dimension is between 30 and 50 observations. Moreover, unlike the DOLS-MG estimator which is consistent only in time factor-induced CS dependency, the CS-DL estimator captures more general forms of CS dependence.

Based on our findings, we substantiate the studies of Tamazian and Rao Bhaskara (2010), Gani (2012), Al Mulali and Ozturk (2015), Ibrahim and Law (2016), Abid (2016), Bhattacharya et al. (2017), Ali et al. (2019), and Salman et al. (2019) regarding the beneficial role of the institutional framework in the environmental performance. In parallel with these studies, via the use of an inclusive institutional quality index and its impact on the more sophisticated environmental indicator of ecological footprint, we provide similar evidence for most of the OECD countries. Concerning the negative role of non-RE consumption on the environmental quality, our study confirms the conclusions reached in the majority of the relevant studies such as those of Shafiei and Salim (2014), Dogan and Seker (2016 a,b), Danish et al. (2017), Sarkodie and Adams (2018) and Wang and Dong (2019). Nevertheless, RE use does not yet seem to demonstrate a substantial diminishing impact on the EF of OECD countries in the long-run. This finding is close to the results of Menyah and Wolde-Rufael (2010a, b) for the US, Lin and Moubarak (2014) for China, and Nathaniel et al. (2020) for MENA countries.

The last part of the study investigates the causal relationships between the respective variables in a robust methodological approach, as described by the Dumitrescu and Hurlin (2012) test. Taking into account the CS heterogeneity and dependence in our panel, we employ the bootstrap application of this test that produces consistent estimates. Table 11 depicts the D-H test results. It is shown that there is a unidirectional causal effect running from economic growth to EF and a weak causal effect from EF to the use of RE. Bekhet and Othman (2018) also found a one-way causality from CO₂ to RE, in the case of Malaysia. In line with the studies of Shafiei and Salim (2014) and Zachoudi (2017) for the OECD countries, we detect no causal effect directed from RE to EF. Furthermore, a bidirectional causal relationship seems to exist between non-RE and EF and between institutional quality and EF.

According to the previous findings, the role of RE consumption in the examined OECD countries seems to provoke an insignificant effect on their EF. Nevertheless, the increase in EF may weakly prompt the adoption of RE in OECD economies. That means that the level of renewables’ utilization in OECD countries is not high enough to cause a substantial reduction in their EF. Concurrently, non-RE use seems not only to upsurge the EF but also to exert a feedback effect to it. Consequently, the share of non-RE still seems to exercise a dominant effect on the environmental status of the examined countries. On the other hand, institutions have a significant healing effect on the environmental performance of OECD countries. This effect is strengthened by the detected interdependence between the institutional operation and the EF. More precisely, our findings testify that the qualitative operation of institutions constitutes the main factor that contributes to environmental quality in OECD countries. This verdict complements the observations of Destek and Sinha (2020), who argue that the institutional performance in most of the OECD countries is not effective enough to ensure environmental sustainability.

The negative but insignificant effect of renewables in reducing the EF of OECD countries could be mainly ascribed to their differences in their available economic and technological-operational dynamics. There is a lot of diversification in the share of RE consumption on their national energy-mix portfolios. This could be attributed to differences regarding the institutional level, the stage of economic growth, and the phase of the fossil fuels decoupling process. The high cost of renewables and the different phases of sustainable targets materialization of OECD countries provoke an insignificant overall effect of renewables on their EF. Moreover, countries’ specific institutional factors such as their legal and regulatory framework and governments’ political will in deploying RE policies could explain this finding.

Conclusions and policy orientations

This study aims at the simultaneous investigation of the dynamic effects of institutional quality and disaggregated energy consumption on the EF of 29 OECD countries during the period 1984–2016. It is the first attempt to incorporate the institutional dimension in a disaggregated energy-environmental framework for OECD countries. For this reason, we generate a constructive institutional index which includes the most critical institutional parameters. The article also adopts a holistic environmental approach, through the use of the ecological footprint indicator, to extend the concept of environmental performance into the overall ecosystem’s performance. Accounting for the detected CS dependence and heterogeneity, though we apply several panel estimation techniques, we base our conclusions on second-generation methodologies, and to the recently developed CS-DL estimator.

The panel cointegration tests reveal the existence of a long-run relationship between the variables in the examined model. The long-run estimates show the stimulative effect of economic growth and non-RE consumption on the EF of OECD countries. Moreover, institutional quality seems to be the most important factor that exerts a positive influence on the environmental performance of these countries. RE consumption has also a negative association with the EF, but it is not proved to be significant enough. In parallel, the D-H causality analysis shows a unidirectional causality from economic growth to the EF and a weak one-way causal effect running from the second to RE consumption. A bidirectional causality is also detected between non-RE and EF and between institutional quality and EF.

The need for environmental policy orientation in OECD countries is not a new project, since Scruggs (1999) was the first to argue it. In this line, this study supports some important policy implications based on the empirical findings. First of all, as OECD countries are the instigators of the SDGs and the Green Growth Strategy (GGS) plan (OECD 2012), they have to reassess their national energy policy portfolios. Most of these countries are subjected to the world’s highest economic growth rates, a fact that makes their need for new energy source exploitation imperative. However, because of their high energy demand, and the broad availability and low cost of fossil fuels energy, it seems that most of their energy needs are still covered by non-renewables. The massive use of fossil fuels inevitably causes a rise in the ecological footprint of these regions.

Therefore, it is evident that OECD countries should give priority to the gradual increase of the renewables share in their national energy-mix, by promoting the substitution of fossil energy from green energy forms. This could be achieved through the establishment of incentive mechanisms for accessible and affordable renewable energies. These incentives could include the implementation of monetary credit channels which can offer lower interest rates on “green enterprises,” tax reductions, green projects certifications, and the provision of RE subsidies on households. Furthermore, the cooperation between public and private entities through the development of Public-Private Partnerships (PPPs) could facilitate the adoption and implementation of renewable technologies.

Nevertheless, the transition to RE forms requires regulatory and institutional efficiency. Institutional quality in OECD countries is associated with a well-defined legislative and regulatory framework that protects property rights and ensures transparent economic processes. Moreover, governmental efficiency and effectiveness act as a warranty on the socio-economic environment of these countries, as governments have the dynamics to implement sustainable environmental policies. Governments should enhance their energy strategies and the relevant policy orientation, in order to overcome institutional barriers that hinder the deployment of renewable energies. These barriers are associated with the proper energy market regulation, the RE installed capacity, the investment implementation in research and development (R&D), and the public information and perception towards the utilization of renewables. Energy regulatory bodies could pursue reforms to reinforce RE, instead of non-RE generation in a level playing field of the energy market. Moreover, institutions should communicate the social dimension of adopting renewables, through an educational approach for the environmental and health benefits.

Finally, OECD governments should strengthen the role of local and national institutions and engage private stakeholders to participate actively in the formulation, monitoring, and evaluation of policies for environmental sustainability. Furthermore, the cooperation among OECD countries (especially in clean technologies applications) is considered critical in the process of materializing their environmental projects in light of their undertaken SDGs. As IEA (2019) argues, the path to a sustainable environment still requires a long way to go, and countries have to undertake a lot of policy measures to integrate renewables into the energy system substantially.

Appendix

Table 12 List of countries