The faster the increase in renewable energy use, the faster the decrease in carbon emissions?

Abstract

The annual average near-surface global temperature will temporarily exceed 1.5 °C between 2023 and 2027. If this level is irrevocably exceeded, the functions of marine biodiversity, fisheries, and ecosystems may no longer be maintained. In light of this fact, this paper examines whether a shock-like (rapid and substantial) increase in the proportion of renewable energy sources could result in a precipitous reduction in greenhouse gas emissions. In order to accomplish this, the paper analyzes the magnitude and direction of the asymmetric and symmetric impacts of renewables on CO₂ emissions, which represent shock-like and ordinary changes, respectively. Second, the paper examines whether there is a functional complementarity between major renewables in reducing CO₂ emissions, and whether this complementarity increases when renewables experience asymmetric shocks. The conclusion of the paper is that the negative impact of renewables on CO₂ emissions either (i) remains unchanged when they experience asymmetric shocks or (ii) increases by 60–260 percent or 100 percent, respectively, when they experience these shocks simultaneously or separately, respectively. The paper makes two recommendations to prevent the persistent exceeding of 1.5 °C. First, policymakers should intensify their efforts to rapidly and substantially increase renewable energy consumption in total. Second, they should accomplish this increase in all renewable sources in a balanced fashion so as to maximize their carbon-reduction effects by functionalizing their green complementarities.

1 Introduction

The average annual concentrations of carbon dioxide (CO₂), methane, and nitrous oxide have reached 410 parts per million (ppm), 1866 parts per billion, and 332 parts per billion, respectively. This magnitude of exponential change over the last 50 years is greater than any other 50-year period in the past 2000 years¹ (IPCC, 2021). As a result, the United Nations’ World Meteorological Organization (WMO) recently issued a warning that between 2023 and 2027, there is a probability of 66% that the annual average near-surface global temperature will temporarily exceed 1.5 °C above pre-industrial levels. If this threshold is exceeded permanently, the functions of marine biodiversity, fisheries, and ecosystems, as well as the services they provide to humanity, may no longer be sustainable.²

Renewable energy can mitigate climate change and contribute to sustainable development by decreasing reliance on fossil fuels, increasing energy productivity, lowering energy maintenance costs, and assuring the stability and security of energy systems (Behera et al., 2023; Ciappi et al., 2022; Egea et al., 2022; Eslami et al., 2019; Figaj & Żołądek, 2021; Gao et al., 2022; Gonzalo et al., 2022; Izanloo et al., 2021; Noorollahi et al., 2021a, 2021b; Noorollahi et al., 2021a, 2021b; Wang et al., 2023; Wen et al., 2022). The alarming condition of climate change demonstrates that an ordinary increase in renewable energy sources has not prevented the 1.5 °C threshold from being exceeded. Consequently, it is imperative to accelerate the global adoption of renewable energy sources (Cho et al., 2020; Domingos et al., 2017; Ginting, 2017; Khan et al., 2018; Kim, 2020; Maennel & Kim, 2018; Pugesgaard et al., 2014; Rozensky et al., 2020; Ryu, 2010). At this critical juncture, a crucial question must be answered: Can a shock-like (substantial and rapid) increase in the proportion of renewables result in a substantial decrease in greenhouse gas emissions?

In order to answer this question, this paper investigates the impact of both ordinary and shock-like changes in the consumption of renewable energy on greenhouse gas emissions. The distinction between a shock-like change and an ordinary change in an independent variable, renewable energy in this paper, is that the former refers to the effect of a rapid and substantial increase or decrease in an independent variable on a dependent variable, such as GHG emissions. By investigating the effects of both types of changes, it will be feasible to establish whether a shock-type change in renewables results in a greater decrease or increase in carbon emissions than an ordinary kind of change. In accordance with Shin et al. (2014), this paper computes the variations of shock types in renewables by decomposing renewable energy data into its positive and negative partial sums.

A limited number of studies have been conducted on the asymmetric effects of renewable energy on GHG emissions (Koengkan et al., 2020; Karasoy, 2019), exports, imports, economic growth, and technological innovation (Ali & Kırıkkaleli, 2021; Adebayo et al., 2021; Toumi, 2019), as well as on the effects of research in renewable energy on CO₂ emissions (Ahmet et al., 2021a). [There are numerous research studies on the drivers of GHG emissions, including the effects of energy consumption, energy intensity, stocks traded, financial development, oil prices, economic growth, globalization, fossil fuels, FDI, and exchange rate depreciation (Ahmet et al., 2021b; Al-Mulali et al., 2019; Malik et al., 2020; Koengkan et al., 2020; Ahmad et al., 2018)]. Furthermore, there is no research, to the best of the authors’ knowledge, that investigates either the asymmetric effects of the components of aggregate renewable energy such as solid biofuels, biogases, and solar thermal or the effects of the complementarities between them in reducing GHG emissions.

Against this backdrop, this paper makes two contributions to the extant literature. First, the paper aims to answer these interlinked questions: (i) do symmetric or asymmetric shocks to renewables decrease or increase CO₂ emissions and (ii) are the effects of symmetric and asymmetric shocks greater than symmetric effects? In order to answer the latter question, the paper first investigates the symmetric effects of renewables on CO₂ emissions. It then estimates (i) not only the separate effects of each renewable on CO₂ emissions when they incur positive and negative shocks but also their complementary effects in a multivariate setting arising out of potential positive or negative synergies between them in decreasing or increasing CO₂ emissions, and (ii) if their complementary effects are higher or lower when renewables incur asymmetric shocks. For this purpose, the paper introduces symmetric and asymmetric green or dirty complementarity between renewables as an analytic concept in investigating these synergies.

The paper uses the symmetric and asymmetric Panel Autoregressive Distributed Lag (ARDL) model and the bootstrapped Granger (multivariate) causality test (Emirmahmutoğlu & Köse, 2011) to estimate the long-run and short-run effects of renewables on CO₂ emissions, respectively. The paper takes a panel of 29 OECD countries for the period 1997–2018 based on the availability of frequent and statistically consistent data provided by the International Energy Agency (2021). The paper uses the consumption of renewables as a proxy for the quantity of renewable energy use since it is the final consumption of renewables that determines their ultimate effect on GHGs. To investigate green (functional) complementarity between renewables, the paper selects solid biofuels, biogases, and solar thermal for econometric analysis as these three renewables have the largest share of total renewable energy consumption and their data are sufficiently frequent and consistent for a large number of countries.

The research first demonstrates that the negative impacts of renewables on CO₂ emissions are either significantly more than or the same as when they experience positive shocks. Second, renewables have both symmetric and asymmetric complementarities in lowering CO₂ emissions, with the asymmetric effect being bigger than the symmetric effect. Together, these data show that a rapid rise in the percentage of renewable energy in overall energy consumption has the potential to cut CO₂ emissions more rapidly. These findings are important for policymakers (i) to boost their efforts to increase the share of renewables and (ii) to achieve this increase in all renewable sources, such as solid biofuels and biomass, in a balanced manner to increase their carbon-reduction effects by functionalizing green complementarities between them. These insights may be significant for climate change mitigation programs and the accomplishment of sustainable development objectives by indicating how renewables, the cornerstone of this mitigation, should be managed and how their carbon-reducing impacts might be amplified. The results are also significant for renewable energy research because they indicate why the students of renewable energy should pay greater attention, first, to the asymmetric effects of renewables and, second, to the disaggregate analysis of renewable energy resources in terms of their complementary effects in reducing carbon emissions.

There are four novelties in this study. This is the first study, to the best of the authors’ knowledge, to estimate the joint asymmetric effects of renewables on carbon emissions. Second, the study establishes a novel benchmark by contrasting the magnitude and direction of symmetric and asymmetric carbon emission reduction effects of renewable energy sources. Third, the study establishes green and dirty complementarities as novel concepts to examine the potential positive or negative synergies that may arise between renewable energy sources in terms of carbon emission reduction or increase, respectively. Fourth, the study presents a critical policy suggestion to decision-makers: it is essential to augment the proportion of all available renewable sources in a balanced and simultaneous fashion for fully maximizing the carbon-reduction effects of renewable energy sources.

The second section of the paper presents the literature review and introduces the conceptual framework of the symmetric and asymmetric effects of the components of renewable energy on CO₂ emissions with specific reference to the complementarity between them. The third section introduces the data and empirical methodology. The fourth section presents and discusses the empirical findings. The last section concludes.

2 Literature review

2.1 Literature review on renewable energy-carbon emissions nexus

An extensive review of the literature is presented in Table 1 and Table A1. It is well-established that both aggregate renewable energy (Ali & Kirikkaleli, 2021; Khan et al., 2018; Kim, 2020; Komarnicka & Murawska, 2021; Norton et al., 1998; Vasylieva et al., 2019; Wang et al., 2019) and its components such as renewable solid waste (Domingos et al., 2017; Ryu, 2010), solid biofuels (Meng & McKechnie, 2019), biogas (Cvetkovic et al., 2016; dos Santos et al., 2016; Ginting, 2017; Roubik et al., 2020), and solar thermal (Ampuno et al., 2020; Ribberink et al., 2013) have significant negative effects on CO₂ emissions. Another established fact is that non-renewable energy consumption increases GHG emissions.

Table 1 Literature review on the effects of renewable energy and components on CO₂ emissions

(Ahmed & Shimada, 2019; Akram et al., 2020; Bhat, 2018; Halkos & Gkampoura, 2021). Furthermore, industrial production in petroleum, chemical, electricity, and nonmetallic minerals, construction, and the transport sectors is a leading driver of GHG emissions (Chen et al., 2022; Hu et al., 2021; Nassar et al., 2021). Financial development also have a positive effect (Sharma et al., 2021), negative effect (Ehigiamusoe & Lean, 2019; Zhao & Yang, 2020) and mixed effects on GHG emissions (Acheampong, 2019), or have a positive effect in the short run (Nasir et al., 2021) but a negative effect in the long run (Nasir et al., 2021; Shoaib et al., 2020).

However, there is limited research on the asymmetric effects of renewable energy on CO₂ emissions. Ali and Kırıkkaleli (2021) and Koengkan (2020) find that positive shocks to aggregate renewable energy consumption have a negative effect on CO₂ in Turkey and in 18 Latin American and Caribbean countries. Karasoy (2019) finds that positive shocks to renewable energy cause an increase in CO₂ emissions in Turkey. Adebayo et al. (2021) find that both positive and negative shocks to renewable energy increase CO₂ emissions in Chile, but Toumi and Toumi (2019) find that both positive and negative shocks to renewable energy have adverse effects on CO₂ emissions.

2.2 Literature review on complementarity between renewables

Complementarity between renewables denotes the synergy between them that may develop in time, space or in both domains to achieve a certain purpose. There are mainly three kinds of complementarities identified in the extant literature: spatial, temporal, and spatio-temporal. Spatial complementarity occurs when energy resources complement each other in a certain region (Aza-Gnandji et al., 2018; Cao et al., 2019; da Luz & Moura, 2019). For example, spatially distributed wind generators would create a smoothing effect, the energy production trends of which vary with an increasing or decreasing distance between sites. Temporal complementarity occurs when two or more energy sources exist within the same region (Zhu et al., 2018). For example, wind energy may become abundant in the autumn and winter seasons and solar energy in the spring–summer period. Spatio-temporal complementarity occurs when a single or multiple energy sources exhibit complementary effects simultaneously in time and space (Canales et al., 2020; Lolla et al., 2015). For example, the Brazilian power system and its hydropower resources interconnect the south-southeastern and north-northeastern subsystems (Jurasz et al., 2020). The research on these three kinds of complementarities has established, for example, that the energy predisposition of an area should be the main determinant of the dominant energy source (Ding et al., 2017). Extending renewable energy generation to potential areas can also be beneficial for secure and reliable energy supplies (Radu et al., 2019). A combination of solar and wind may help increase reliable power production on an annual basis (Slusarewicz & Cohan, 2018). An optimum wind-plus-solar power system reduces variability and enhances the stability of monthly power generation (Han et al., 2019; Jerez et al., 2013; Zhang et al., 2018a, 2018b). An optimal selection of threshold levels for renewables can considerably reduce energy exchange costs (Naeem et al., 2019). It has also been found that spatio-temporal complementarity helps to increase the capacity factors of power plants, and the improvement of the energy mix helps minimize balancing costs and storage needs (Ciria et al., 2020; Jerez et al., 2013; Kapica et al., 2021; Ren et al., 2019).

The current body of research on the impact of renewable energy on carbon emissions has three deficiencies. First, existing research has not investigated how asymmetric shocks change the impact of various components of aggregate renewable energy on carbon emissions. This is significant as the effect of aggregate renewable energy can differ from that of its individual components such solid biofuels, biogases, and solar thermal energy. Policymakers need accurate and energy-specific information to create effective climate change mitigation measures during this critical phase of climate change trends. Second, the existing literature does not cover the impact of simultaneous asymmetric shocks on all individual renewable energy sources. This article demonstrates that a simultaneous shock greatly enhances the carbon-reducing impact of each individual renewable energy source, which is essential for addressing climate change. Providing this crucial information to renewable energy policy-makers could strongly encourage them to swiftly and evenly expand the proportion of renewables in the whole energy mix. Third, the existing literature restricts the scope of complementarities to spatial and temporal factors. This article demonstrates that renewable energy sources work together effectively to reduce carbon emissions. This functional complementarity may have a greater carbon reduction benefit than spatio-temporal complementarity. Excluding this factor could significantly reduce the effectiveness of efforts to mitigate climate change.

3 Analytic methodology, data and empirical methodology

3.1 Analytic methodology

Against this backdrop, the paper seeks to explain if an increase of a particular type of shock would cause a significant decline in CO₂ emissions in the short run or long run. For this purpose, the paper first sets up a base model that estimates the symmetric and asymmetric effects of the major components of renewables on CO₂ emissions. The paper disaggregates renewable energy into its major components to illustrate the symmetric or asymmetric effect of separate renewable energy sources on CO₂ emissions as the effect of each may differ. To illustrate if the effects of symmetric or asymmetric shocks are greater than symmetric effects, we will compare the magnitude of the long-run and short-run coefficient values estimated by the symmetric and asymmetric models. In addition, while pinpointing the exact effect of each renewable energy source on CO₂ emissions, we should also demonstrate how this separate effect occurs when a specific renewable energy source interacts with the others in a multivariate setting. The multivariate analysis matters because the effects of renewables do not occur separately but also complementarily. The paper introduces green and dirty complementarities of symmetric and asymmetric kinds to explain the variety of these potential complementarities between renewable energy sources. The complementarity-theoretic approach matters in illustrating how to govern the overall mix of renewable energy resources.

A positive answer to the question of whether symmetric or asymmetric shocks to the components of renewable energy have larger negative effects on CO₂ emissions would further consolidate and augment the efforts to increase the share of renewables, which would be critical to reducing or keeping GHGs at very low levels to curb climate change. A negative answer to this question would evidently demonstrate a weak possibility of noticeably reducing GHG emissions or keeping GHG emissions at very low levels in the short or long run by a sudden increase in total usage of the renewable energy. A negative answer may originate in the fact that renewable energy sources have a number of limitations such as having a diminishing marginal effect on GHGs (Liobikiene & Butkus, 2021); causing GHG emissions, even though they are much lower than or half of the level of fossil fuels (Boluk & Mert, 2014; Dincer, 2000; Meng & McKechnie, 2019); having low net energy ratios, requiring a quantity of energy that might be higher than that to produce them; the absence of massive technologies for storing them; and the challenges of adopting, upgrading or extending renewable technologies such as the need for greater capital investment than non-renewables, despite having lower operational costs (Heal, 2010; Timmons et al., 2014).

Complementarities—CMTs can be defined as the mutual reinforcement among a certain group of variables in part or all of a social structure that improves or worsens clustering relative to alternative configurations (Hall & Soskice, 2001). In the context of CMTs, there might emerge four types of relationships between renewables; (i) a ‘green’ symmetric or asymmetric complementarity, (ii) a ‘dirty’ symmetric or asymmetric complementarity, (iii) the conflation of the first two options, and (iv) the lack of any relationships. Green and dirty CMTs between renewables turn out when the effects of a certain number of renewables converge in reducing or increasing CO₂ emissions, respectively. Green and dirty CMTs gains a systemic function when the majority of present CMTs converge in reducing and increasing the emissions, respectively. Fragmented complementarity denotes a situation when the majority of CMTs does not converge in reducing or increasing the emissions.

The first criterion in selecting an econometric technique in investigating the nexus between the components of renewables and CO₂ emissions is to be able estimate the long-run and short-run causal relationships between the variables simultaneously in a multivariate setting. The second criterion is to investigate the nexus by estimating the both symmetric and asymmetric effects of renewables on CO₂ emissions using the same econometric technique. The long run matters when it comes to establishing the existence or the lack of a stable and dynamic relationship between the two or more variables, which is an indication of a systemic or fragmented mode of complementarity. Systemic complementarity can be inferred when there is a mutual reinforcement between the majority of the renewables in reducing or increasing CO₂ emissions in the long-run and short-run. The majority matters because we can expect a noticeable and sustained increase or decrease in CO₂ emissions when the majority of the renewables functionally complement each other or converge to the extent of reducing or increasing pollution, respectively. Fragmented complementarity can be inferred when there is a mutual reinforcement between the few but not the majority of the renewables in increasing or decreasing pollution in the long-run or short-run.

The econometric technique to cover the two points noted above is multivariate panel data cointegration that estimates the long-run relationship between variables. Establishing the cointegration between renewables and carbon emissions will show that renewables have a trending effect on carbon emissions over time. The term "trending effect" refers to the continuous and enduring impact of renewables on carbon emissions. The identification of the inverse relationship between renewable energy sources and carbon emissions will furnish policymakers with a strong basis for formulating measures that seek to increase the share of renewables in the overall energy composition in a timely, substantial, and enduring manner, with the ultimate goal of attaining a sustainable decline in carbon emissions. This study employs multivariate analysis to integrate all renewable energy sources into a single model. Establishing the interrelationships among various renewable energy sources in affecting carbon emissions is crucial in order to ascertain whether their collective impact on carbon emissions reduction surpasses that of any individual renewable energy source. Discovering a stronger carbon reduction impact due to a negative complementary relationship between renewables can offer policymakers a solid foundation for increasing the proportion of all renewable energy sources in a well-balanced way, resulting in a greater mitigation of their adverse effects.

We developed a model which can be written in the following formal algebraic forms. The variables represented by the acronyms have been defined in Table 2.

$${\text{lnCO2}}_{{{\text{it}}}} = {\text{f}}\left( {{\text{lnsbiof}}_{{{\text{it}}}} ,{\text{lnbiogs}}_{{{\text{it}}}} ,{\text{lnsther}}_{{{\text{it}}}} ,{\text{lnrrnew}}_{{{\text{it}}}} ,{\text{lnnrnew}}_{{{\text{it}}}} ,{\text{ lnind}}_{{{\text{it}}}} ,{\text{lnfmd}}_{{{\text{it}}}} ,{\text{e}}_{{1,{\text{it}}}} } \right)$$

(1)

$$\begin{aligned} {\text{lnCO}}2_{it} & = {\text{f}}({\text{lnsbiof}}_{{{\text{it}}}}^{ + } ,{\text{ lnsbiof}}_{{{\text{it}}}}^{ - } ,{\text{ lnsbiogs}}_{{{\text{it}}}}^{ + } ,{\text{ lnsbiogs}}_{{{\text{it}}}}^{ - } ,{\text{lnsther}}_{{{\text{it}}}}^{ + } ,{\text{ lnsther}}_{{{\text{it}}}}^{ - } ,{ } \\ & {\text{lnrrnew}}_{{{\text{it}}}}^{ + } ,{\text{lnrrnew}}_{{{\text{it}}}}^{ - } ,{\text{ lnnrnew}}_{{{\text{it}}}} , {\text{lnind}}_{{{\text{it}}}} ,{\text{lnfmd}}_{{{\text{it}}}} ,e_{{2,{\text{it}}}} ) \\ \end{aligned}$$

(2)

$$\begin{aligned} {\text{lnCO}}2_{{{\text{it}}}} & = {\text{f}}({\text{lnsbiof}}_{{{\text{it}}}}^{ + } ,{\text{ lnsbiof}}_{{{\text{it}}}}^{ - } ,{\text{lnbiogs}}_{{{\text{it}}}} ,{\text{lnsther}}_{{{\text{it}}}} , \\ & {\text{lnrrnew}}_{{{\text{it}}}} ,{\text{lnnrnew}}_{{{\text{it}}}} , {\text{lnind}}_{{{\text{it}}}} ,{\text{lnfmd}}_{{{\text{it}}}} ,{\text{e}}_{{3,{\text{it}}}} \\ \end{aligned}$$

(3)

$$\begin{aligned} {\text{lnCO}}2_{{{\text{it}}}} & = {\text{f}}({\text{lnsbiof}}_{{{\text{it}}}} ,{\text{lnsbiogs}}_{{{\text{it}}}}^{ + } ,{\text{ lnsbiogs}}_{{{\text{it}}}}^{ - } ,{\text{lnsther}}_{{{\text{it}}}} , \\ & {\text{lnrrnew}}_{{{\text{it}}}} ,{\text{lnnrnew}}_{{{\text{it}}}} , {\text{lnind}}_{{{\text{it}}}} ,{\text{lnfmd}}_{{{\text{it}}}} ,e_{{4,{\text{it}}}} ) \\ \end{aligned}$$

(4)

$$\begin{aligned} {\text{lnCO}}2_{{{\text{it}}}} & = {\text{f}}({\text{lnsbiof}}_{{{\text{it}}}} ,{\text{lnbiogs}}_{{{\text{it}}}} ,{\text{lnsther}}_{{{\text{it}}}}^{ + } ,{\text{ lnsther}}_{{{\text{it}}}}^{ - } , \\ & {\text{lnrrnew}}_{{{\text{it}}}} ,{\text{lnnrnew}}_{{{\text{it}}}} , {\text{lnind}}_{{{\text{it}}}} ,{\text{lnfmd}}_{{{\text{it}}}} ,{\text{e}}_{{5,{\text{it}}}} ) \\ \end{aligned}$$

(5)

$$\begin{aligned} {\text{lnCO}}2_{{{\text{it}}}} & = {\text{f}}({\text{lnsbiof}}_{{{\text{it}}}} ,{\text{lnbiogs}}_{{{\text{it}}}} ,{\text{lnsther}}_{{{\text{it}}}} ,{\text{lnrrnew}}_{{{\text{it}}}}^{ + } ,{ } \\ & {\text{lnrrnew}}_{{{\text{it}}}}^{ - } ,{\text{lnnrnew}}_{{{\text{it}}}} ,{\text{ lnind}}_{{{\text{it}}}} ,{\text{lnfmd}}_{{{\text{it}}}} ,{\text{e}}_{{6,{\text{it}}}} ) \\ \end{aligned}$$

(6)

where ${\text{i}}=\mathrm{1,2},\dots {\text{N}}$; ${\text{t}}=\mathrm{1,2},\dots {\text{T}}$. For ${\text{m}}=1,\dots ,6$, all ${{\text{e}}}_{{\text{m}},{\text{it}}}$’s are error terms of each model, respectively.

Table 2 Variables and data sources

Model 1 and Model 2 estimate the symmetric and asymmetric effects of renewables on CO₂ emissions, respectively. Model 2 estimates the effects of all four renewables on CO₂ emissions when all of them incur asymmetric shocks. Model 2 has four sub-models that estimate the effects of four renewables on CO₂ emissions when only one of them incurs asymmetric shocks. Thus, Model 2 has four variants. Models 3, 4, 5 and 6 estimate when positive and negative shocks are applied only to SBIOF, BIOGS, STHER, and RRNEW, respectively. Such a modeling strategy aims to check the robustness of our estimations by demonstrating what would happen under different scenarios when all or any of the renewables incurs symmetric or asymmetric shocks.

3.2 Data

Table 2 illustrates the empirical model data and data sources. We use renewable energy data provided by the International Energy Agency—IEA. The IEA’s data is presented mainly in three categories: production, total supply, and the consumption of renewables. We take consumption as our reference point. Because it is the total final consumption that determines the final effect of renewables on CO₂ emissions. Total final consumption is “equal to the sum of the consumption in the end-use sectors” (IEA, 2021: 14). We take three major renewables that make up a great bulk of total final consumption. These are solid biofuels—SBIOF, biogases—BIOGS, and solar thermals—STHER,³ for which the statistically consistent and frequent data is available both for a sufficiently large number of countries and for a sufficiently long-period of time for panel data cointegration analysis, 22 year-period (1997–2018). We also include the rest of the renewables, RRNEW, apart from SBIOF, BIOGS and STHER as a fourth component of total renewable energy in our analysis so that we could investigate the effect of the total renewables on CO₂ emissions in a disaggregated manner. RRNEW includes hydro, geothermal, solar photovoltaic, tide, wave, ocean, wind, liquid biofuels and renewable municipal waste (IEA, 2021: 51). The components of RRNEW are statistically scattered and infrequent but their total sum, RRNEW, is sufficiently frequent and consistent for being included in our model. Figure 1 depicts the proportion of renewable energy components to total renewable energy consumption. In total, solid biofuels comprised 96% but by 2018 this proportion had consistently decreased to 64%. Between 1990 and 2018, the proportion of solar thermals increased by up to 3%, while the share of biogases fluctuated between 1 and 2%. The share of remaining renewables increased from 2% in 1990 to 30% in 2018 in a consistent manner.

Fig. 1

The share of renewable energy components in total

In addition to renewables, we include non-renewable energy, NRNEW, in our model, too, as a control variable so that we could estimate the effect of all energy sources on CO₂ emissions altogether. Because the effect of renewables on CO₂ emissions occurs at the same time with the effect of non-renewables in a certain time and space. Non-renewable energy is calculated by subtracting the total of renewable energy from total of all energy sources including coal, oil, gas, electricity, heat and others (IEA, 2021: 31). In our case countries, RNEW and NRNEW make up 6.7% and 93.3% of total energy resources on an average. The total of SBIOF, STHER, BIOGS makes up 84% of total renewable energy, the rest of which consists of RRNEW. We take CO₂ emissions as a proxy for GHG emissions as it makes up a great bulk, 80.7%, of total greenhouse gases for the case countries on an average of the period of analysis.

3.3 Empirical methodology

The panel cointegration tests comprise four steps: (i) first, to estimate cross-section dependence between the panels, (ii) second, to run panel unit root tests,⁴ (iii) third, to estimate the cointegration relationship, and (iv) fourth, to estimate long-run and short-run models if the cointegrated relationship is established. Appendix II provides technical information regarding panel unit root, cross-section, and cointegration tests in an effort to conserve space.

According to the results in Tables 3, we fail to reject the null of cross-section independence both for Model 1 and Model 2 because either all or the majority of the tests are not statistically significant at one percent or five percent levels. Given the results of cross-section independence, we use the first-generation panel unit root tests that precondition cross-section independence.⁵

Table 3 Results of cross-section dependence tests for model 1 and model 2

The results from the panel unit root tests using individual intercept and trend models are reported in Table 4. We conclude first that our dependent variable in all six models is non-stationary at level and become stationary at first difference as the results of either all or the great majority of the panel unit root tests are not statistically significant at level and are statistically significant at first differences, respectively. Second, the remaining variables are all either integrated of I(0) or I(1).

Table 4 Results of panel unit root tests

For cointegration estimation, we use Westerlund and Edgerton Cointegration Test with Structural Breaks. The key reason for selecting this test for our analysis is that it allows unknown structural breaks in addition to cross-sectional dependence (as well as heteroskedastic and serially-correlated errors). Besides, the test performs well, even on small samples.⁶ It is noted that the level shift and regime shift models perform better, and that the latter performs better than the former (Westerlund & Edgerton, 2008). As shown in Table 5, all three tests (no shift, level shift, and regime shift) indicate strong cointegration at one percent level both for Model 1 and Model 2.

Table 5 Results of westerlund panel cointegration test

Given the cointegrated relationship, we proceed to estimate long-run and short-run relationships between variables using Linear and Non-Linear Panel Autoregressive Distributed Lag approach. For running Panel ARDL models, we used a dynamic panel estimator, PMG. Main advantages of ARDL methodology are five-fold (Pesaran et al., 1999). First is to be able to estimate cointegrated relationships irrespective of the order of integration, which might be I(0), I(1) or a mix of I(0) and I(1). Panel ARDL cannot, however, be employed when the dependent variable is not I(1) or any of the variables is I(2). As the dependent variable in our model, CO2, is I(1) and there is no variable that is I(2), we are able to run Panel ARDL for our models.⁷ Second, ARDL methodology allows estimating both linear and nonlinear short and long-run coefficients simultaneously (Sheng & Guo, 2016 and Sulaiman et. al., 2015). Third, ARDL methodology allows each variable to have varying number of lags, thereby respecting the potentially diverse relationships between the variables in time. This is significant because variables may exhibit different levels of responsiveness over time (Domingos et al., 2017). Fourth, the ARDL approach makes more efficient and consistent estimation even with a small sample. Fifth, in the long-run, the approach corrects for the endogeneity bias of the regressors and the inability to examine hypotheses on the estimates.

The reason why we have chosen ARDL in comparison to alternative methodologies is three-fold. First, Engle and Granger (1987), hereafter EG, suggested a two-step method of testing cointegration. In the first step, residuals from an OLS regression of variable of interest on the rest of the lagged variables of this variable are computed. In the second step, Dickey–Fuller (DF) and Augmented Dickey–Fuller (ADF) statistics are applied to these residuals, assuming no intercept and no linear trend. There are totally five different variants of EG: DF, ADF, Generalized Least Squares (ADF-GLS) proposed by Elliott et al. (1992), Z-GLS proposed by the Phillips (1987a, b), and MSB-GLS proposed by Sargan and Bhargava (1983). However, residual-based tests have a number of shortcomings. In small samples, the test results crucially depend on the choice of the left-hand side variable in the first step. Another shortcoming of residual based tests is that they do not allow for more than one cointegrating relationship. In addition, these tests do not make the best use of available data, and have generally low power. Second, residual-based approaches to panel cointegration which are, Kao (1999), Pedroni (2004) and Westerlund (2005), focus on testing for unit roots in OLS or panel estimates of error terms. As Pesaran (2015) highlights, these tests suffer from deficiencies that can result in size distortions and have very low power. Third, dynamic panel methods such as Fixed Effects (FE), Instrumental Variables (IV), or the Generalized Method of Moments (GMM) estimators proposed by Anderson and Hsiao (1981, 1982), Arellano (1989), Arellano and Bover (1995), and others may produce inconsistent estimates of parameters unless coefficients are identical across countries.

The long-term models have been constructed as follows:

$$\begin{aligned} {\text{InCO}}2_{{{\text{it}}}} & = \upalpha_{{1{\text{i}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upalpha_{{2{\text{i}}}} {\text{Insbiof}}_{{{\text{i}},{\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{i}} = 0}}^{{\text{q}}} \upalpha_{{3{\text{i}}}} {\text{Inbiogs}}_{{{\text{i}},{\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{i}} = 0}}^{{\text{q}}} \upalpha_{{4{\text{i}}}} {\text{lnsther}}_{{{\text{i}},{\text{t}} - {\text{j}}}} \\ & \quad + \mathop \sum \limits_{{{\text{i}} = 0}}^{{\text{q}}} \upalpha_{{5{\text{i}}}} {\text{lnrrnew}}_{{{\text{i}},{\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{i}} = 0}}^{{\text{q}}} \upalpha_{{6{\text{i}}}} {\text{lnnrnew}}_{{{\text{i}},{\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{i}} = 0}}^{{\text{q}}} \upalpha_{{7{\text{i}}}} {\text{lnind}}_{{{\text{i}},{\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{i}} = 0}}^{{\text{q}}} \upalpha_{{8{\text{i}}}} {\text{lnfmd}}_{{{\text{i}},{\text{t}} - {\text{j}}}} + {\text{u}}_{{1{\text{t}}}} \\ \end{aligned}$$

(7)

$$\begin{aligned} {\text{InCO}}2_{{{\text{it}}}} & = \upbeta_{{1{\text{i}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{2{\text{i}}}} {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{3{\text{i}}}} {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{4{\text{i}}}} {\text{lnbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{5{\text{i}}}} {\text{lnsbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{6{\text{i}}}} {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{7{\text{i}}}} {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{8{\text{i}}}} {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{9{\text{i}}}} {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{10{\text{i}}}} {\text{lnnrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{11{\text{i}}}} {\text{lnind}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upbeta_{{12{\text{i}}}} {\text{lnfmd}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + {\text{u}}_{{2{\text{t}}}} \\ \end{aligned}$$

(8)

$$\begin{aligned} {\text{InCO}}2_{{{\text{it}}}} & = \upgamma_{{1{\text{i}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upgamma_{{2{\text{i}}}} {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upgamma_{{3{\text{i}}}} {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upgamma_{{4{\text{i}}}} {\text{lnbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upgamma_{{5{\text{i}}}} {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upgamma_{{6{\text{i}}}} {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upgamma_{{7{\text{i}}}} {\text{lnnrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upgamma_{{8{\text{i}}}} {\text{lnind}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upgamma_{{9{\text{i}}}} {\text{lnfmd}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + {\text{u}}_{{2,1{\text{t}}}} \\ \end{aligned}$$

(9)

$$\begin{aligned} {\text{InCO}}2_{{{\text{it}}}} & = \upkappa_{{1{\text{i}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upkappa_{{2{\text{i}}}} {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upkappa_{{3{\text{i}}}} {\text{lnsbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upkappa_{{4{\text{i}}}} {\text{lnsbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upkappa_{{5{\text{i}}}} {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upkappa_{{6{\text{i}}}} {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upkappa_{{7{\text{i}}}} {\text{lnnrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upkappa_{{8{\text{i}}}} {\text{lnind}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upkappa_{{9{\text{i}}}} {\text{lnfmd}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + {\text{u}}_{{2,2{\text{t}}}} \\ \end{aligned}$$

(10)

$$\begin{aligned} {\text{InCO}}2_{{{\text{it}}}} & = \uppi_{{1{\text{i}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uppi_{{2{\text{i}}}} {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uppi_{{3{\text{i}}}} {\text{lnbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uppi_{{4{\text{i}}}} {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uppi_{{5{\text{i}}}} {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uppi_{{6{\text{i}}}} {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uppi_{{7{\text{i}}}} {\text{lnnrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uppi_{{8{\text{i}}}} {\text{lnind}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uppi_{{9{\text{i}}}} {\text{lnfmd}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + {\text{u}}_{{2,3{\text{t}}}} \\ \end{aligned}$$

(11)

$$\begin{aligned} {\text{InCO}}2_{{{\text{it}}}} & = \uptheta_{{1{\text{i}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uptheta_{{2{\text{i}}}} {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uptheta_{{3{\text{i}}}} {\text{lnbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uptheta_{{4{\text{i}}}} {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uptheta_{{5{\text{i}}}} {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uptheta_{{6{\text{i}}}} {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uptheta_{{7{\text{i}}}} {\text{lnnrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uptheta_{{8{\text{i}}}} {\text{lnind}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \uptheta_{{9{\text{i}}}} {\text{lnfmd}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + {\text{u}}_{{2,4{\text{t}}}} \\ \end{aligned}$$

(12)

where ${\text{InCo}}2$ is the dependent variable for all of six models, the independent variables have been defined in Table 2; ${\text{i}}=1,\dots ,{\text{N}}$ are cross-section units; ${\text{t}}=1,\dots ,{\text{T}}$ are time periods; $p$ and $q$ are optimal lag orders; ${\mathrm{\upalpha }}_{1{\text{i}}}, {\upbeta }_{1{\text{i}}},{\upgamma }_{1{\text{i}}},{\upkappa }_{1{\text{i}}},{\uppi }_{1{\text{i}}}, {\uptheta }_{1{\text{i}}}$ are the group-specific intercepts for all models; ${\mathrm{\upalpha }}_{2{\text{i}}}\dots {\mathrm{\upalpha }}_{8{\text{i}}}, {\upbeta }_{2{\text{i}}}\dots {\upbeta }_{12{\text{i}}},{\upgamma }_{2{\text{i}}}\dots {\upgamma }_{9{\text{i}}},{\upkappa }_{2{\text{i}}}\dots {\upkappa }_{9{\text{i}}},{\uppi }_{2{\text{i}}}\dots {\uppi }_{9{\text{i}}}, {\uptheta }_{2{\text{i}}}\dots {\uptheta }_{9{\text{i}}}$ are long-term coefficients for all models; and all of ${u}_{it}$’s are the error terms.

After estimating the long-run model, we construct the short-term (error correction) model as follows:

$$\begin{aligned} \Delta \ln {\text{CO}}2_{{{\text{it}}}} & = \upomega_{{1{\text{i}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upomega_{{2{\text{i}}}} \Delta \ln {\text{sbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upomega_{{3{\text{i}}}} \Delta \ln {\text{biogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upomega_{{4{\text{i}}}} \Delta \ln {\text{sther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upomega_{{5{\text{i}}}} \Delta \ln {\text{rrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} \\ & \quad + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upomega_{{6{\text{i}}}} \Delta \ln {\text{nrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upomega_{{7{\text{i}}}} \Delta \ln {\text{ind}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{{\text{j}} = 1}}^{{\text{p}}} \upomega_{{8{\text{i}}}} \Delta \ln {\text{fmd}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \upomega_{{9{\text{i}}}} {\text{ECT}}_{{{\text{i}},{\text{t}} - 1}} + \varepsilon_{{{\text{it}}}} \\ \end{aligned}$$

(13)

$$\begin{aligned} \Delta \ln {\text{CO}}2_{{{\text{it}}}} & { = }\upvarphi_{{{\text{1i}}}} + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{2i}}}} \Delta {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{3i}}}} \Delta {\text{lnsbiof}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{4i}}}} \Delta {\text{lnbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } \\ & \quad + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{5i}}}} \Delta {\text{lnsbiogs}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{6i}}}} \Delta {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{7i}}}} \Delta {\text{lnsther}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{8i}}}} \Delta {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ + } \\ & \quad + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{9i}}}} \Delta {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}}^{ - } + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{10i}}}} \Delta {\text{lnrrnew}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{11i}}}} \Delta \ln {\text{ind}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \mathop \sum \limits_{{\text{j = 1}}}^{{\text{p}}} \upvarphi_{{{\text{12i}}}} \Delta \ln {\text{fmd}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \upvarphi_{{{\text{13i}}}} {\text{ECT}}_{{{\text{i}}, {\text{t}} - {\text{j}}}} + \varepsilon_{{{\text{it}}}} \\ \end{aligned}$$

(14)

where Δ is the lagged difference operator and ${{\text{ECT}}}_{{\text{i}}}$ represents the error correction term obtained from the long-run model. For space purposes, we write the formal equations of only Model 13 and Model 14 among the short-run models. That the ECT is negative and statistically significant indicates that the short-run deviations adjust to the long-run equilibrium.

Table 6 indicates the results of long-run and short-run estimations along with long-run and short-run results of asymmetry test results. As the ECT for all models estimated is statistically significant and negative, we conclude that short-run deviations from the long-run equilibrium in six models estimated converge to long-run equilibrium and therefore we can interpret the estimated coefficients for the long-run models.

Table 6 Results of symmetric and asymmetric panel ARDL

For short-run analysis, we employ the bootstrap panel Granger non-causality test proposed by Emirmahmutoglu and Kose (2011); technical details are provided in Appendix II. Because Panel ARDL-PMG approach requires pretests for unit roots and cointegration, making estimations biased for short-run causality The bootstrap panel Granger non-causality test overcomes this bias because of not being subject to testing unit root or cointegration. The bootstrap non-causality test is applicable in both bivariate and multivariate setting. Multivariate estimation is relevant because, as noted above, our model is a multivariate one that aims to establish the effects of more than one independent variables on a dependent variable in order to estimate not only the separate effect of any of renewables but also the effect of each renewable on CO₂ emissions in conjunction with the effects of the others simultaneously (Table 7).

$${\text{lnCO}}2_{{{\text{it}}}} = {\text{f}}\left( {{\text{lnsbiof}}_{{{\text{it}}}}^{ + } ,{\text{ lnsbiogs}}_{{{\text{it}}}}^{ + } ,{\text{ lnsther}}_{{{\text{it}}}}^{ + } ,{\text{ lnrrnew}}_{{{\text{it}}}}^{ + } ,{\text{ lnnrnew}}_{{{\text{it}}}} ,{\text{ lnind}}_{{{\text{it}}}} ,{\text{lnfmd}}_{{{\text{it}}}} ,{\text{e}}_{{2,{\text{it}}}} } \right)$$

(15)

$${\text{lnCO}}2_{it} = {\text{f}}\left( {{\text{lnsbiof}}_{{{\text{it}}}}^{ - } ,{\text{ lnsbiogs}}_{{{\text{it}}}}^{ - } ,{\text{ lnsther}}_{{{\text{it}}}}^{ - } ,{\text{ lnrrnew}}_{{{\text{it}}}}^{ - } ,{\text{ lnnrnew}}_{{{\text{it}}}} ,{\text{ lnind}}_{{{\text{it}}}} ,{\text{lnfmd}}_{{{\text{it}}}} ,{\text{e}}_{{2,{\text{it}}}} } \right)$$

(16)

Table 7 Results of symmetric and asymmetric bootstrapped panel causality test

The model 14 is divided into two sub-models (Model 15 and 16) containing positive and negative shocks, respectively. Because the number of estimated parameters increases when higher number of variables in VAR models is included. This may cause the number of observations in time dimension to be insufficient.

4 Empirical findings and discussion

4.1 Long-run

As shown in Table 6 and Figs. 2, 3 and 4, the results of Model 13 indicate that solid biofuels, biogases, and the remaining renewables, which account for 96% of total renewables by 2018, have negative long-term effects on CO₂ emissions, while solar thermal has no statistically significant effect on CO₂. This indicates a green complementarity of the systemic kind between renewables in reducing CO₂ emissions in the long run. Model 14 demonstrates that solid biofuels and the remaining renewables have asymmetric effects on CO₂ in the long run as indicated by the long-run asymmetry tests. This indicates a green complementarity of the fragmented kind between solid biofuels and the remaining renewables in reducing CO₂ emissions as the majority of renewables do not converge in affecting CO₂ in a specific direction. The two renewables account for nearly 95% of the total renewables; therefore, this complementarity, albeit fragmentary, evidence demonstrates a significant negative effect of renewables on carbon emissions when taken together. According to the results of Models 9 and 10, solid biofuels and biogases have long-run asymmetric effects on CO₂, respectively. Models 10 and 12 demonstrate that solar thermal and remaining renewables do not have long-run asymmetric effects on CO₂. Taken together, the symmetric and asymmetric models show a green complementarity of the systemic kind between (i) sbiof⁺, biogs, and rrnew in Model 9, (ii) sbiof, biogs⁺, sther and rrnew in Model 10, and (iii) between sbiof, biogs, and rrnew in Model 11 in reducing CO₂ emissions.

Fig. 2

The Results of Model 1 (Symmetric) and Model 2 (Asymmetric)

Fig. 3

The Results of Model 9 and 10

Fig. 4

The Results of Model 11 and 12

As illustrated in Table 6 above and in Fig. 5, a positive shock to solid biofuels, biogases, and the remaining renewables in all five asymmetric models all have negative effects on CO₂ when having statistically significant effects. The magnitude of this negative effect for solid biofuels increases from 5% in Model 13 to 8% in Model 14 where a positive shock is applied to all four renewables but remains at the same level as in Model 9 where an asymmetric shock is applied only to solid biofuels. The magnitude of the negative effect of biogases increases from 1% in Model 13 to 2% in Model 10 where an asymmetric shock is applied only to biogases. The magnitude of the negative effect of the remaining renewables increases from 0.3% in Model 13 to 0.8% in Model 14 but does not exist in Model 12 where an asymmetric shock is applied only to the remaining renewables. Solar thermal does not have asymmetric effects on CO₂ either in Model 14 or Model 11 where an asymmetric shock is applied only to solar thermal. In brief, the negative effects of solid biofuels and the remaining variables on CO₂ increases by 60% and 260% when all renewable energy sources incur asymmetric shocks, and the negative effect of biogases increases by 100% when an asymmetric shock is applied only to biogases. In other words, the negative symmetric effects of renewables on CO₂ are likely to increase significantly over time if they continue to experience asymmetric shocks, as demonstrated in Model 14, given that solid biofuels, the remaining renewables, and biogases account for more than 96% of the total renewables.

Fig. 5

The Comparative magnitude of symmetric and asymmetric effects

In contrast, the size of the symmetric effect of solid biofuels remains the same in Model 11. The size of the symmetric effect of biogases also remains the same in Model 9 and 11. The symmetric effect of the remaining renewables increases from 0.3% in Model 13 to 1% in Model 9 and 0.4% in Model 10. Solid biofuels, biogases, and the remaining renewables do not have a statistically significant long-run asymmetric effect on CO₂ in Models 9 and 12, Model 12, and Model 9 and 12, respectively. Thus, the long-run asymmetric effects of renewables are highly likely to increase or not to decline, at least in all cases when they incur asymmetric shocks.

In Model 13, non-renewable energy and industry have positive effects on CO₂, whereas financial development has a negative effect on CO₂. The positive long-run symmetric effect of non-renewable energy declines, although at small ratios in all asymmetric models except Model 9. The positive symmetric effect of industry on CO₂ slightly increases in Model 11 but ceases to exist in all remaining asymmetric models. The negative symmetric effect of financial development on CO₂ declines only in Model 9 by 25%; it remains the same in Models 7, 10, and 11; and increases by 50% in Model 12.

Overall, in the long run, a positive shock to renewables will either significantly increase or not change their CO₂-reducing effects. A key point in this regard is that a green complementarity of the systemic kind between renewables when they incur positive and negative shocks causes a greater decrease in CO₂ emissions than that between renewables when they do not. The trends of change in the symmetric effects of non-renewables, industry, and financial development on CO₂ in asymmetric models indicate that asymmetric shocks to renewables in any of five scenarios are highly likely to reduce the positive effect of non-renewables, although slightly, eliminating the positive effects of industry and increasing the negative effects of financial development on CO₂ in the long run.

4.2 Short run

According to the Panel ARDL estimations, in the short run, none of the renewables has asymmetric effects in any model, illustrated by short-run asymmetry tests. Only solid biofuels have positive short-run symmetric effects on CO₂ in Models 10, 11 and 12, the magnitude of which is the same. The other renewables do not have statistically significant symmetric effects on CO₂ in the short run. Non-renewables all have positive statistically significant effects on CO₂ in all models, whereas industry and financial development do not have any statistically significant effects on CO₂ in any models.

However, according to the results of the Emirmahmutoğlu & Köse panel causality test, in Model 13, biogas has a symmetric causal effect on CO₂ in the short run. In Model 14, all renewables, except sbiof⁺, sther⁺ and rrnew⁻, have causal effects on CO₂. In Model 9 and 10, all renewables have causal effects on CO₂. In Model 11, all renewables except sther⁻ have causal effects on CO₂. In Model 12, all renewables except biogases have causal effects on CO₂. In Model 14, non-renewable energy does not have causal effects on CO₂ in either estimation where we estimated the effects on non-renewables, industry, and financial development, first including only positive shocks of renewables (Model 15) and then including only negative shocks of the renewables (Model 16). Industry has a causal effect on CO₂ in Model 15 but not in Model 16. Financial development has causal effects in both Models 15 and Model 16. Non-renewable energy and financial development have causal effects in Models 9, 10, 11 and 12. Industry has causal effects in Model 9, 10 and 11 but not in Model 12.

Regarding the potential inefficiencies of the short-run Panel ARDL estimations (Emirmahmutoğlu and Köse, 2011), we prioritize the results of the Emirmahmutoğlu and Köse panel causality test. Thus, in the short run, too, a great majority of variables included in asymmetric models have short-run causal effects on CO₂. This indicates that solid biofuels, solar thermal, and the remaining renewables have short-run causal effects on CO₂ when they incur or do not incur positive or negative shocks.

4.3 Comparison of findings with prior research

As Table 8 illustrates, the findings of this paper confirm the negative effects of (i) biogas (Al-Masri & Al-Sharqi, 2020; Budzianowski et al., 2016; Fahmy et al., 2018; Ginting, 2017; Yu et al., 2008; Zhang et al., 2007, 2022); (ii) solar thermal (Ampuno et al., 2020; Greening & Azapagic, 2014; Ribberink et al., 2013; Schnitzer et al., 2007; Treichel & Cruickshank, 2021); (iii) solid biofuels (Cheng et al., 2022; Kougioumtzis et al., 2021); (iv) the remaining renewables and hence aggregate renewable energy (Boluk & Mert, 2014; Cho et al., 2020; Domingos et al., 2017; Kim, 2020; Komarnicka & Murawska, 2021; Norton et al., 1998; Vasylieva et al., 2019; Wang et al., 2019); (iv) the positive complementarity between renewable energy sources (Ciria et al., 2020; da Luz & Moura, 2019; Ding et al., 2017; Karadol et al., 2021; Risso et al., 2019; H. X. Zhang et al., 2018a, 2018b); and (v) the asymmetric effects of renewable energy sources on CO₂ emissions (Toumi & Toumi, 2019; Koengkan, 2020; Ali & Kirikkaleli, 2021). The findings also confirm that the non-renewable energy (Ahmed & Shimada, 2019; Bhat, 2018) and industrial production increase CO₂ emissions (Cheng et al., 2022; Hu et al., 2021; Nassar et al., 2021; Wang et al., 2021) but financial development reduces CO₂ emissions (Khan et al., 2019; Shoaib et al., 2020; Zhao & Yang, 2020), solid biofuels (Cheng et al., 2022; Kougioumtzis et al., 2021).

Table 8 Comparison of the findings on the effects of renewables on carbon emissions

Despite corroboration of the results reported in the existing body of literature, the study’s findings are primarily differentiated along three dimensions. To begin with, this research represents the initial examination of functional complementarity from a multi-country standpoint across OECD member states. Therefore, the results of the research are predicated on an examination of functional complementarity as opposed to spatiotemporal complementarity. Furthermore, the research uncovers notable discrepancies between positive and negative shocks, suggesting that failure to account for these discrepancies could result in the adoption of erroneous policy approaches predicated exclusively on linear estimation. Furthermore, the research highlights the contradictory effects of financial development and industry on the complementarity of renewable energy sources, which are also incorporated into the study. In its entirety, this research adds to the existing body of knowledge by validating previously established conclusions through the implementation of an innovative and all-encompassing approach.

4.4 Policy suggestions

Given the paper’s conclusions, three types of major and interdependent policy strategies can be adopted to combat climate change risk by reducing carbon emissions.

First, shock-like development in renewables is the sole source of either a much more rapid decline in GHG emissions or fulfilling the Intergovernmental Panel on Climate Change’s (IPCC) projection of extremely low GHG emissions of 1–1.8 °C without rising GHG emissions. Thus, authorities should accelerate their efforts to boost the share of renewable energy sources in overall energy consumption. There are at least five major policy options. First, authorities can eliminate export or import subsidies for environmentally unfriendly industrial techniques, while supporting the import of renewable energy technology through tariff reductions or public subsidies (Zhang et al., 2022). Second, authorities may ease bureaucratic processes for renewable energy adaption, such as the usage of environmentally friendly automobiles (Cho et al., 2020), provide financial incentives for green investment projects, and levy higher taxes on conventional energy sources (Cho et al., 2020; Liobikiene & Butkus, 2021). Alternatively, they can implement rules in the form of technology-push and market-pull or place regulatory constraints on traditional technologies (Diederich et al., 2016). Fourthly, policymakers may promote a broader and more efficient use of renewables such as solar systems in both the public and commercial sectors by including renewable energy education at all levels of public and private (especially industrial) education and training systems (Schnitzer et al., 2007). Fifthly, authorities may employ public and behavioral techniques to influence public opinion through popular channels such as social media in order to make the usage of renewable energy more attractive (Boluk & Mert, 2014).

Second, the negative impact of renewables should be enhanced by concurrently or evenly increasing their proportion of total energy demand. As the paper showed that when an asymmetric shock is delivered to all four forms of renewables, the negative effects of solid biofuels and the other renewables on CO₂ emissions are at their greatest. To consolidate or increase the present complementarities among renewables, nations must identify their optimal time scale and space combinations to allow the creation of hybrid energy systems. Although regional and national conditions are very likely to vary, governments should optimize their renewable energy mix by identifying regionally-specific complementarity combinations. Moreover, regional planning may enable a more balanced integration of renewable energy sources by boosting complementarity, as well as reduce investment and storage costs. To accomplish all of these procedures professionally, the efficacy and viability of each form of renewable energy may be analyzed through in-depth resource analysis, and specific complementary designs can be developed to enhance their carbon-reducing synergistic effects (Aunon-Hidalgo et al., 2021).

Finally, it is evident that the preceding two results are based on long-term forecasts and that the required policy measures may not be adopted in the near future. Both the achievement of a rapid increase in the proportion of renewables and the adaptation of human behavior, technological infrastructure, production relations, and energy demand and supply conditions to an energy environment in which the proportion of renewables increases much more rapidly than in the past will require considerable time (Boluk & Mert, 2014; Domingos et al., 2017). As such, policymakers should utilize this period of adaptation to at least mitigate climate change. Functionalizing renewable energy governance systems at the micro, meso, and macro levels might be a crucial step in resolving this issue (or at the level of company, sectors, and country-wide). These systems might aid in the planning, execution, modification, and enhancement of all measures designed to increase the breadth and composition of renewable energy consumption, training, technical development, supply chains, etc. The most significant contribution that a multifaceted approach to renewable energy governance may make is to organize all forms of activities aimed at increasing the overall proportion of renewables by fostering positive complementarities not only between renewable energy sources in reducing carbon emissions but also between governments, businesses, households, universities, and the other components of an entire social system in adopting renewable energy more quickly and effectively.

4.5 Limitations and future research

First, the paper included both the aggregate use of renewable energy and its components at the national level, but did not separate their usage at the sectoral level, such as commerce, transportation, residential, etc. Consequently, future research may run the same models evaluating the usage of renewables not only at the level of the entire country, but also at the level of each sector, in order to provide more particular and practicable policy recommendations. Secondly, as a consequence, the paper’s topic matter was studied using theoretical models. This is crucial to offer fundamental insight into whether a quicker growth in renewables might lead to a faster reduction in carbon emissions. Due to the critical nature of this topic, however, it may be necessary to do substantial fieldwork to verify the validity of this study’s conclusions based on micro data. Thirdly, the article analyzed the developed economies based on the availability of data for a sufficiently enough length of time. This approach was necessary due to the requirement of categorizing nations according to their revenues. Each income group, however, has its own renewable or non-renewable energy composition, production and technology endowment structure, governmental and educational system, cultural dynamics, etc. Consequently, the same models should be examined using time-series econometric models at least for independent nations for whom statistically consistent data is available. Fourth, a limited number of yearly observations, 20 per variable, necessitated the inclusion of renewable energy, non-renewable energy, the value-added of the industrial sector, and financial development as independent variables in each of the paper’s six base models. However, future research could include additional independent variables such as green technological development, green levies, and environmental R&D expenditures in similar models in order to provide a more comprehensive explanation of the factors that determine greenhouse gas emissions.

5 Conclusion

This paper aimed to answer two major questions. The first question is if a positive shock to renewables would cause an increase or decrease in CO₂ emissions and if this effect is lower or higher than their symmetric effects of the ordinary kind. The second question is if there was a complementarity between renewables in reducing CO₂ emission and if this complementary effect of renewables is lower or higher when they incur positive shocks. The paper investigated these issues using two empirical methods: the Symmetric and Asymmetric Panel ARDL technique for long-run estimates and the Emirmahmutoglu and Kose Panel Causality Test for short-run estimates. This paper developed six empirical models to estimate the symmetric and asymmetric effects of all renewable energy sources on carbon emissions, including solid biofuel, biogas, thermal solids, and the remainder of renewable energy consumption. The paper used a panel of 29 OECD countries and statistically consistent data from the International Energy Agency for the years 1997–2018.

The paper has first demonstrated that renewables, such as solid biofuels and biogases, have negative long-term effects on CO₂ emissions and short-term causes on CO₂ emissions. In terms of reducing CO₂ emissions, the paper has also demonstrated a green functional complementarity between renewables. Specifically, the negative effects of solid biofuels and the remaining renewables on CO₂ increase by 60% and 260% when all renewable energy sources are subjected to asymmetric shocks, whereas the negative effect of biogases increases by 100% when an asymmetric shock is only applied to biogases. In other words, the negative symmetric effects of renewables on CO₂ are likely to increase significantly over the long term if they continue to experience asymmetric shocks. These findings indicate that a rapid and significant increase in the proportion of renewable energy sources in total energy consumption has the potential to reduce CO₂ emissions in a faster and more effective manner.

The findings of this paper are essential for policymakers (i) to boost their efforts to enhance the proportion of renewables and (ii) to achieve this increase in all renewable sources, such as solid biofuels and biomass, in a balanced manner to raise their carbon-reduction effects by functionalizing their green complementarities. Given the World Meteorological Organization’s warning about the exceeding of 1.5 °C between 2023 and 2027 and its catastrophic consequences for the humanity, the achievement of these two goals might be crucial for attaining the United Nations’ Sustainable Development Goals. Functionalizing renewable energy governance systems at the micro, meso, and macro levels (or at the company, sector, and national levels) may be crucial in achieving these two policy goals. The primary limitation of the article was that it focused solely on the effects of renewables on the economy as a whole, rather than on sectoral effects using IEA data for national economies. Consequently, effectively achieving the two policy objectives may necessitate additional work on collecting micro data on the current complementarities between renewables in various sectors and developing institutional mechanisms to plan, create, execute, and modify these sector-specific complementarities.