Assessing the spillover effects of research and development and renewable energy on CO₂ emissions: international evidence

Abstract

The primary motivation of this paper is the lack of consensus on the impact of renewable energy (RE) and research and development (R&D) expenditure on CO₂ emissions in the literature. Current literature has mostly ignored the spillover effect of R&D on CO₂ emissions by increasing the intensity effect of technology, leading to biased results. Further, little is known about the impact of previous epidemics on CO₂ emissions. This study fills these gaps by evaluating the spillover effects of RE and R&D on CO₂ emissions in a global panel of 54 countries from 2003 to 2017. Using a two-way time- and spatial-fixed-effects panel analysis, we find both income-induced and scale effects of economic growth are present in our panel, though the scale effect is the dominant one. Our findings indicate that economic growth increases CO₂ emissions at a decreasing rate, validating the Environmental Kuznets Curve hypothesis, and that urbanization and foreign trade worsen the environment. We also find that epidemic episodes before COVID-19 had a nonsignificant impact on CO₂ emissions internationally. More importantly, our results confirm the presence of both the intensity and scale effects of R&D, with the intensity effect being the dominant one. We find overwhelming evidence that global R&D investment led to an overall (direct plus spillover) reduction of CO₂ emissions, driven by its spillover effect, through two channels: RE and economic growth. Finally, we find that RE installations assist with reducing CO₂ emissions internationally, though RE composition and state of R&D can lead to different findings. Our findings have significant policy implications for sustainable development. Our RE and R&D-spillover results support the policy recommendation of shifting to high-tech clean energy sources.

Introduction

As the urgency of climate change is rising, many policy makers globally are making the environment and its sustainability a priority. Globally, there has been an increase in energy-related carbon dioxide (CO₂) emissions, the leading greenhouse gas. Fossil fuels such as coal, natural gas, and oil are the primary sources of CO₂ emissions and the main drivers of climate change. Figure 1 depicts worldwide primary energy consumption by fuel source from 2019 to 2021. As shown in Fig. 1, before and after the COVID-19 pandemic, oil remained the single most consumed form of energy, accounting for 31–33% of the total consumption, while the consumption of coal and natural gas accounted for 27% and 24%, respectively. The remainder of the energy consumption was of clean sources of energy such as hydropower (7%), nuclear power (4%), and renewables (6%). Cumulatively, the use of fossil fuels such as coal, oil, and natural gas accounts for a significant share (83%) of the world’s energy consumption.1 Globally, we consumed less dirty fossil fuels, especially oil and coal, and more clean energy sources during the COVID-19 pandemic. However, our fossil fuel usage rebounded as the global economy rebounded in 2021. This is exhibited in Fig. 1 below.

Fig. 1.

Global primary energy consumption by fuel source from 2019 to 2021. Source: BP Statistical Review 2021

According to the International Energy Agency, CO₂ emissions rose by 6% to 36.3 billion metric tons in 2021, the highest level that the world has ever witnessed.¹ The agency further estimates that about 40% (15.3 billion metric tons) of the overall growth in global CO₂ emissions resulted from coal consumption. At the same time, natural gas and oil accounted for 7.5 billion metric tons and 10.7 billion metric tons, respectively. In absolute terms, the increase in CO₂ emissions has offset the decline that resulted from the COVID-19 pandemic (2.3 billion metric tons) (Adebayo et al., 2022a). This trend has dashed hopes of the sustainable recovery that the pandemic promised. The unprecedented CO₂ emissions level was mainly driven by fossil fuel combustion to enable the global economy’s rebound after the pandemic.2 Juxtaposing the pattern of primary energy consumption with the contributions to total CO₂ emissions, a positive correlation can be inferred. Studies such as (Xie et al., Sep. 2022) and (Sajid & Gonzalez Aug., 2021) confirm that the surge in global CO₂ emissions is attributable to the emerging economies, chief among them China, whose economic progress comes at a cost to the ecosystem.

The current rate of CO₂ emissions, coupled with the fact that emerging economies will have to rely heavily on nonrenewable energy to fuel their development trajectory, is worrisome. This impedes efforts by policy makers and international bodies such as the United Nations, as part of its Sustainable Development Goals, to reiterate the need for clean energy, technological innovation, and sustainable consumption and production, among others, to curtail climate change.3 Against this backdrop, studies have been undertaken in this direction, all geared toward identifying the best policies and practices to ameliorate the situation. It is, therefore, not surprising that a substantial amount of the literature on environmental and natural resources is dedicated to assessing how CO₂ can be mitigated. Although not universal, one solution is to shift away from highly polluting energy-based economic activities and toward more environmentally friendly and sustainable ones using low-impact technologies (Mamkhezri et al., 2022a)–(Foxon Oct., 2011). Aside from such costly activities, the energy-economics literature largely confirms that technological innovation is required for both CO₂ emission reduction and renewable energy (RE) advancements. Additionally, the level of research and development (R&D) expenditure is the main determinant of technological innovation. Therefore, R&D expenditure is expected to be crucial in fulfilling the CO₂-emission-reduction objective (Weimin et al., Apr. 2022; Xie et al., Sep. 2022). As such, it is prudent to assess the impact of RE and R&D on CO₂ emissions globally.

The literature contains contrasting findings on the impact of RE and R&D on CO₂ emissions, obfuscating how these two variables influence CO₂ emissions globally.4 It is believed that technology should enhance environmental conditions while not hindering economic growth. However, the effect of innovations on environmental quality, particularly on CO₂ emissions, remains unclear. Further, many governments worldwide are pushing to enact RE policies to combat climate change (Dai et al., 2022; Mamkhezri, 2019). As with the nexus between R&D and CO₂ emissions, the general perception is that RE consumption mitigates CO₂ emissions; however, researchers’ findings do not always support this perception. The literature tends to overlook how highly connected RE, R&D, and CO₂ are and how they may very well have varying levels of spillover effect on one another (Kumar & Agarwala, 2016; Popp et al., 2011). One novel contribution of the current study is that we examine the impacts of these two variables on CO₂ emissions concurrently and consider spatial and time-period linkages and thus their spillover effects globally.

Furthermore, COVID-19 has caused significant damage to human life and has had a significant economic and environmental impact. As of October 2022, there have been more than 617.5 million confirmed cases of COVID-19 with more than 6.5 million fatalities reported to the World Health Organization.5 The COVID-19 pandemic’s lockdown period was significantly more extensive than any other epidemics’ in the last few decades. The worldwide lockdown resulted in a sharp decline in CO₂ emissions in 2020, as mentioned above. The lack of long-term evidence in existing studies about how epidemics affect CO₂ emissions is due to the absence of an accurate index for measuring pandemic prevalence across countries. A discussion of the Pandemics Index, calculated by counting the number of words associated with pandemics as revealed in the Economist Intelligence Unit’s country reports, is now available to quantify epidemic prevalence (Ahir et al., 2018). To this end, we contribute to the body of knowledge by investigating the consequences of epidemics for CO₂ emissions across countries for the first time.

To address the discrepancy in the literature mentioned above, we evaluate the impacts of R&D and RE generation on CO₂ emission by considering the epidemic-prevalence index and its effect (if any) in a global panel of 54 countries from 2003 to 2017. Using a two-way time- and spatial-fixed-effects panel model, we find overwhelming evidence that higher GDP per capita increases CO₂ emissions at a decreasing rate and that higher urbanization and trade-openness levels worsen the environment. We also find that epidemic episodes prior to COVID-19 had a nonsignificant impact on CO₂ emissions in our panel. More importantly, our results confirm the existence of both the intensity and scale effects of R&D (defined below) and suggest that R&D has an overall statistically significant and positive effect on the environment by mitigating CO₂ emissions internationally. Finally, we find that RE installations help with reducing CO₂ emissions internationally.

Our contribution to the existing energy-economics literature is twofold. We investigate the international spillover effects of RE and R&D on CO₂ emissions using state-of-the-art econometric techniques. Previous research that finds a positive R&D-CO₂ nexus tends to overlook the spillover effect of R&D expenditures and is therefore upwardly biased and overstated. We show that the intensity effect of R&D, which is the spillover effect of R&D, dominates the direct scale effect of technology internationally. In addition, we are the first to investigate the impact of epidemic episodes on the environment in the selected countries.

Literature review

Previous researchers have divided the energy-economics literature in two ways (Awaworyi Churchill et al., xxxx; Petrović & Lobanov, 2020). One strand of energy-economics literature investigates various leading macroeconomic determinants, including but not limited to economic growth captured through GDP, energy use, open economy captured via trade openness, level of urbanization, and technology advancement driven by R&D, among other variables. Most of these studies find that technological advancements benefit the environment by reducing CO₂ emissions. This strand of literature sheds light on the inverted-U-shaped relationship between economic growth and CO₂ emissions, known as the Environmental Kuznets Curve (EKC). The second strand of the energy-economics literature is mainly devoted to panel-data analyses of the impact of R&D and/or RE on CO₂ emissions in a set of countries or worldwide. As with the first strand, the findings in the second strand of the literature are highly mixed. Since these two strands of the literature are not mutually exclusive, we discuss the two and provide a brief overview of the mixed findings in the energy-economics literature below. We also outline a synopsis of the previous findings regarding the effects of epidemics on CO₂ emissions.

Research and development and CO₂ emissions

(Fei et al., 2014) examine the relationship between technological advancements and CO₂ emissions. They show that R&D investments increased the use of clean energy in New Zealand and Norway, yielding lower CO₂ emissions. (Fernández Fernández et al., 2018) demonstrate that R&D investments contributed positively to CO₂ emission reductions in developed countries. In a study that sought to find the effect of GDP, RE, export quality, and technological innovation in 22 developed countries, (Rahman et al., Nov. 2022) confirm the EKC hypothesis and find that a negative shock in technological innovation increases CO₂ emissions. They also find bidirectional causality between technological innovation and CO₂ emissions. Ibraheim (Awosusi et al., 2022) and (Salman et al., Oct. 2019) confirm the negative relationship between technological innovation and CO₂ emissions in Egypt and 7 ASEAN countries, respectively. In a study that focuses on the BRICS countries, (Rafique et al., Jul. 2020) find a significant negative impact of technological innovation, financial development, and foreign direct investment on carbon emissions. A bidirectional causal relationship between variables such as financial development, technological innovation, trade openness, urbanization, and energy use on carbon emissions is also found. A similar study, (Ojekemi et al., 2022), finds a negative nexus between consumption-based CO₂ and technological innovation in the BRICS economies. The authors find that an increase in RE consumption translates into an average 0.62% reduction. Analyzing how technological innovation, GDP, and other variables affect CO₂ emissions in 35 OECD countries between 1996 and 2015, (Cheng et al., 2021) find a negative relationship between CO₂ emissions and technology. They also confirm the EKC hypothesis and argue that technological innovation offsets the benefits of economic growth slightly by lowering energy intensity. (Ostadzad Oct., 2022) finds among 29 EU countries that innovation has a negative relationship with emissions. (Kuang et al., Sep. 2022) and (Cai et al., 2021) also confirm a negative relationship between green technology and carbon emissions using 25 and 30 provinces, respectively, in China. According to the latter study, the impact was, however, heterogeneous, depending on the level of economic development of the region. (Ge et al., Jun. 2022) also find that an innovative pilot-city policy in 285 cities in China contributed significantly more to CO₂ emission reduction in pilot cities than nonpilot cities. The effect is, however, heterogeneous, with the impact being noticeable in central cities. The policy had spatial and spillover effects, with the latter only being significant within 150–250 km.

In contrast, (Shahbaz et al., 2018) results indicate that energy-research innovations had a negative effect on CO₂ emissions in France. (Awaworyi Churchill et al., xxxx) study the nexus between R&D intensity and CO₂ emissions in the G7 countries for the past two centuries and find a positive relationship only from 1955 to 1990 (35 years) and a negative one otherwise. They explain these findings by stating that R&D affects CO₂ emissions by simultaneously affecting GDP and the energy efficiency of production, as well as through free trade. (Du et al., Sep. 2019) find among 71 economies that green-technology innovations do make a significant contribution to carbon emission reduction in countries whose income levels are below a certain threshold but the effect on carbon reduction is positive for above-threshold countries.

(Yu & Xu, 2019) study shows that R&D resulted in significant CO₂ emission reductions nationally and regionally in East Asia. The study by Yu and Xu finds support in a similar study by (Yu & Du Jan., 2019), who proxy technology by R&D expenditure. Focusing on China, they find a negative relationship between technological innovation and CO₂ emissions. The finding by Yu and Du did not change even after categorizing the countries into a high-speed-growth group and low-speed-growth group. The effect of R&D also varies depending on the economic development of the country or the form that the innovation takes (Cai et al., 2021; Fethi & Rahuma Jun., 2020; Salman et al., Oct. 2019). The findings of (Dauda et al., May 2019) validate the EKC hypothesis in both countries in the MENA region and G6 countries, but, in contrast, innovation increases CO₂ emissions in the MENA region but reduces emissions in the G6 countries. (Weimin et al., Apr. 2022) confirm that positive shocks in innovation reduce per capita carbon emissions while a negative shock degrades the quality of the environment. A crop of studies has also focused on technological innovations and CO₂ emissions at the firm level. Using some 17 national petroleum companies, Fethi and Rahuma (Fethi & Rahuma Jun., 2020) find that spending on eco-innovation as proxied by R&D, training, and investment in the environment leads to a significant reduction in CO₂ emissions at the company level. (Wen et al., 2020) study construction CO₂, a major source of CO₂ emissions in China. Using 30 provinces in China, they find that patent application and technological innovation have a positive effect on construction-CO₂ mitigation. They also find significant spatial dependence and clustering features in provincial construction-CO₂ emissions. There is a positive effect on CO₂ emissions from surrounding regions.

According to some studies, technological advancements may worsen the quality of the environment. This implies that, although advanced technologies might improve resource-use efficiency, their economic scale also quickly grows, demanding more natural resource inputs, resulting in higher CO₂ emissions (Cheng et al., 2019). (Zhai & Song, 2013) discover that R&D had no discernible influence on China’s long- and short-term CO₂ emissions. According to (Cheng et al., 2019), technological innovations had a negligible effect on CO₂ emissions in OECD countries, but the effects were positive in countries with moderate CO₂ emission levels. (Chen & Lee, 2020) discover that technological advancements had no evident effect on global CO₂ emissions. According to (Petrović & Lobanov, 2020), increased R&D investments had a negative average long-run impact on CO₂ emissions in 17 OECD countries from 1981 to 2014. The authors find a positive individual effect in 6 to 7 (depending on the model) countries they examine, necessitating the estimation of specific regressions. Their short- and long-run findings indicate that the effect of R&D investments on CO₂ emissions could be positive, negative, or nonsignificant. Apart from (Ahir et al., 2018) and (Ge et al., Jun. 2022), who consider spatial dependency at the provincial level in China, all the references cited here ignore the spatial dependency of countries on each other. That is, R&D in countries with high technology capacity and CO₂ emissions can decrease CO₂ emissions in neighboring countries, while the results are reversed in countries with lax environmental regulations. We provide overwhelming evidence in the current study that this oversight can result in incorrect and biased conclusions.

Renewable energy and CO₂ emissions

Similarly to the R&D-CO₂ literature, the energy-economics literature surrounding whether RE consumption reduces CO₂ is mixed; most studies find a negative nexus, while some find a negligible or even positive link. For example, (Bilgili et al., 2016) observe that 17 OECD nations decreased CO₂ emissions by employing RE. Additionally, they disprove the EKC hypothesis in their panel of countries. (Adebayo et al., 2022a) find similar findings in the UK. According to the researchers, a positive shift toward RE use leads to a reduction in carbon emissions and vice versa. They also find that a unidirectional causal impact exists from RE, fossil fuel, and COVID-19 to CO₂ emissions. An estimated regime-switching model also reveals that in the presence of COVID-19, RE reduces carbon emissions but an increase in fossil fuel usage is found across all regimes. Studies by (Yuping et al., Nov. 2021), (Adebayo et al., 2022b), (Du et al., Aug. 2022), and (Bandyopadhyay et al., Jun. 2022) all conclude in a similar fashion. (Yuping et al., Nov. 2021), for instance, find that, RE consumption and globalization tended to reduce emissions while nonrenewable emissions boosted emissions in Argentina. While RE and globalization jointly affected emissions negatively, globalization and non-RE consumption had a positive effect on emissions. Their study confirms also the EKC hypothesis in Argentina. The findings of Yuping et al. are confirmed in a similar study by (Adebayo et al., 2022b) using the MINT (Mexico, Indonesia, Nigeria, and Turkey) countries. (Nakhli et al., 2022) also find that in the US, regarding RE and non-RE consumption and economic policy uncertainty, there is unidirectional causality from electricity consumption to economic policy uncertainty but bidirectional causality between carbon emissions and policy uncertainty.

(Moutinho & Robaina, 2016) establish concrete evidence for the EKC hypothesis’s validity and explain how RE may help reduce CO₂ emissions. (Zoundi, 2017) demonstrates that RE plays a significant role in African countries’ efforts to reduce CO₂ emissions. According to (Kahia et al., 2017), RE reduced CO₂ emissions and boosted economic development in MENA nations. Using the Method-of-Moments Quantile Regression, (Awosusi et al., Nov. 2022) also find that in the BRICS economies, the use of biomass energy reduces the ecological footprint. Studies by Du et al. Aug. (2022); Awan et al., Jul. 2022; Chen et al., 2019), and (Xie et al., Sep. 2022) have all established negative relations between the use of RE and carbon emissions. (Awan et al., Apr. 2022), for instance, find that among 10 emerging countries including China, Argentina, and Mexico, RE consumption has a negative impact on carbon emissions. The study also provides evidence for the EKC hypothesis in these economies. The finding by (Awan et al., Jul. 2022) using 107 countries is not different from the above. Their finding is confirmed in the case of China by (Xie et al., Sep. 2022) and in the case of Malaysia by (Jahanger et al., Jul. 2022), who not only confirm the EKC hypothesis but also find that an increase in RE consumption and technological innovations have a negative impact on carbon emissions. (Hu et al., 2018) investigate the contribution of RE to the CO₂ emissions of 25 developing nations. Their results suggest that using RE sources resulted in a reduction in CO₂ emissions. (Chen et al., 2019) reach similar findings and demonstrate that RE decreased carbon emissions. Similarly, (Mamkhezri et al., 2021) find that RE investments reduce CO₂ emissions in New Mexico. According to (Bekun et al., 2019), CO₂ emissions dropped in 16 European Union member countries due to increasing RE use. In a study that spans from 1990 to 2018, (Miao et al., Mar. 2022) find that among newly industrialized economies, RE usage tends to reduce ecological footprint.

In contrast, several experimental studies show that RE use has no impact on CO₂ emissions or even hampers the environment by increasing greenhouse gas emissions. For example, (Apergis & Payne, 2010) demonstrate that RE does not contribute to short-run emission reductions in the 19 developing and industrialized countries they analyze. (Farhani, 2013) identify unidirectional causality between RE production and CO₂ emissions in the short run but not in the long run. (Ben Jebli et al., 2015) find no evidence of a causal link between RE use and CO₂ emissions. According to (Ben Jebli & Ben Youssef, 2017), RE consumption increased CO₂ emissions in 5 North African nations. (Silva et al., 2018) discover that CO₂ emissions hinder the use of RE in the 17 sub-Saharan nations they evaluate. (Saidi & Omri, 2020) find that RE investments reduced CO₂ emissions in France, Finland, Japan, Canada, Belgium, the UK, Sweden, Germany, the United States, Switzerland, and the Czech Republic. However, similar investments boosted CO₂ emissions in South Korea and the Netherlands. (Tiwari et al., Sep. 2022) assess hydropower consumption’s ecological footprint in Brazil and China and find contrasting results for the two countries. Hydropower usage reduced ecological footprint in China, while it had a nonsignificant impact in Brazil. (Nguyen & Kakinaka, 2019) study the nexus between RE consumption and CO₂ in 107 countries from 1990 to 2013 and show that RE consumption increased CO₂ emissions in low-income nations. Using variables such as financial development, coal consumption, GDP, and RE, (Adebayo et al., Sep. 2021) also find an insignificant correlation between RE usage and CO₂ emissions in South Africa. They further assert that this is because RE usage forms a small component of energy usage in the South African economy.

COVID-19 and CO₂ emissions

COVID-19 has caused significant damage to human life and has had a significant economic and environmental impact. As a result of the virus’s prevalence and the contraction in industrial and transportation activities, CO₂ emissions decreased compared to the pre-outbreak period (Quéré et al., 2020; Wang et al., 2020). Most recent economic studies on COVID-19’s economic effects examine its negative impact on variables such as unemployment status (Couch et al., 2020) and household incomes (Akesson et al., 2020). There is, however, no conclusive evidence that the disease affects air quality. (He et al., 2020) discovers that lockdowns in Chinese cities resulted in a significant improvement in air quality. The finding by He et al. aligns with the claim by (Nguyen et al., 2021) that China witnessed close to an 18% reduction in energy-related carbon emissions. (Gillingham et al., Jul. 2020) find that in the US, COVID-19 reduced the consumption of jet fuel and gasoline by 50% and 30% respectively and electricity decreased by 10%. This translated into a 15% decline in carbon emissions between March and June 7, 2020. These findings confirm those of (Brodeur et al., 2021) for the USA.

Other studies also find the impact of COVID-19 to be based on the sector, the severity of the pandemic in a country, and lockdown policies (Bazzo Vieira et al., xxxx; Gillingham et al., Jul. 2020; Schulte-Fischedick et al., 2021; Sikarwar et al., 2021). In a panel study using industries in the USA, 28 countries from the EU, China, and India, (Sikarwar et al., 2021) find a reduction of 1,749 million ton in total carbon emissions but the magnitude of the reduction depends on the sector. The transportation sector contributed the most (58%) in carbon emissions reductions, followed by the coal-generation sector (29%) and industry (10%). In another study on 23 EU countries, (Andreoni xxxx) finds that the manufacturing, wholesale, retail-trade, transport, accommodation, and food-service sectors accounted for about 93.7% of the reduction in CO₂ in 23 EU countries and 10 economic sectors. Countries such as Spain, Italy, and France who were adversely affected by the pandemic witnessed the largest reduction in carbon emissions. According to (Gillingham et al., Jul. 2020), COVID-19 reduced the amount of fuel consumed in the form of jet fuel and gasoline by about 50% and 30% respectively in the USA. The amount of electricity consumed also declined by less than 10%. These reductions have translated into some 15% reduction in carbon emissions in the USA. Gillingham et al. argue that the long-term effect occurs in two scenarios. The best scenario is one where COVID-19 comes under control and the positive impact on emissions will only be short-lived; the worse scenario is one where COVID-19 gets out of control. In the latter scenario, there could be a direct effect of short-run emissions decline and an indirect effect where there will be changes in behavior and investment in eco-friendly energy. Tian et al. (Tian et al., 2022) support the latter finding in their study. They assert that the economic impact of the pandemic has led to volatility in RE markets since the declining fossil fuel prices have weakened the competitiveness of RE. Reduction in government subsidies has also meant investment in RE projects will be halted. Aktar et al. (Most et al., 2021) find that, even though there was a noticeable impact of COVID-19, its impact differently manifested in different regions. Electricity demand in China dropped by 6.6%, but Italy, India, France, Spain, the northwestern region of the United States, and the United Kingdom, where the impact of the pandemic was severe, witnessed about a 15% reduction. Reductions in emissions therefore manifested more in the USA, China, and the EU, which are leading emitters of CO₂. (Andreoni xxxx) arrives at a similar conclusion in a study involving 23 European countries. Spain, Italy, and France, who were largely affected by the pandemic, had the largest carbon reductions, accounting for more than 55% of the total emission change. (Sajid & Gonzalez Aug., 2021) find COVID-19 to have had heterogeneous effects on carbon emissions in Asia–Pacific countries under different conditions. They find that in Bangladesh, India, and Pakistan, apart from a worst-case scenario where it is assumed that travel restrictions last for six months, no other scenario witnessed a negative impact on carbon emissions. In China and Indonesia, however, the pandemic had a negative effect on carbon emissions irrespective of the scenario. Using energy, government policy, and activity data in 69 countries, 50 US states, and 30 provinces in China, Quéré et al. (Hu et al., 2018) find that restrictions that came with the pandemic led to daily global carbon emissions declining.

On the other hand, (Almond et al., May 2021) highlight that COVID-19 had an ambiguous effect on pollution in China. Due to coronavirus’s severity, the results can become more ambiguous when the focus shifts away from the coronavirus and toward other types of epidemics. The quarantine time for the COVID-19 pandemic is much longer than for any previous epidemic in the last few decades. The scarcity of long-term evidence on how epidemics effect CO₂ emissions stems from the lack of an appropriate index for tracking pandemic incidence across nations. To measure epidemic prevalence, a discussion of the Pandemics Index, created by measuring the number of terms related to pandemics disclosed in the Economist Intelligence Unit’s nation reports, is now available (Ahir et al., 2018). To that end, for the first time, we contribute to the body of knowledge by evaluating the effects of epidemics on CO₂ emissions across nations.

Data description and methodology

We assess the effects of CO₂-emission determinants on environmental quality in this study using annual data from 2003 to 2017 in 54 countries. Our sample of countries includes Argentina, Armenia, Austria, Azerbaijan, Belarus, Belgium, Brazil, Bulgaria, Canada, China, Colombia, Costa Rica, Croatia, Czech Republic, Denmark, Egypt Arab Republic, Finland, France, Germany, Greece, India, Ireland, Israel, Italy, Japan, Kazakhstan, Republic of Korea, Kuwait, Kyrgyz Republic, Latvia, Lithuania, Mexico, Mongolia, Netherlands, Norway, Panama, Poland, Portugal, Romania, Russian Federation, Singapore, Slovak Republic, Slovenia, South Africa, Spain, Sweden, Tajikistan, Thailand, Tunisia, Turkey, Ukraine, United Kingdom, United States, and Uzbekistan. The period covered and the data availability were the primary determinants of the sample of countries.

To assess the effect of R&D expenditure and RE production on CO₂ emission, we control for widely used variables in the literature cited above to explain CO₂ emissions such as GDP (Awaworyi Churchill et al., xxxx; Li & Wang Dec., 2017; Petrović & Lobanov, 2020; Rahman et al., Nov. 2022) as well as factors related to urbanization (Petrović & Lobanov, 2020) and trade openness (Acheampong et al., 2020; Solarin et al., 2017). We control for epidemics by including the discussion surrounding the Pandemics Index, which is computed by collecting the number of phrases linked with pandemics reported in Economist Intelligence Unit country reports (Ahir et al., 2018). To test the EKC hypothesis, we include squared GDP per capita as a determinant of CO₂ emissions in our empirical analyses. To reiterate, the EKC hypothesis postulates a relationship between economic prosperity and environmental quality in which environmental quality initially degrades before improving as economic expansion increases. Therefore, environmental quality and economic development have an inverted-U-shaped (concave) connection. A statistically significant and negative coefficient on the latter variable proves the EKC hypothesis (Grossman & Krueger May, 1995). To capture the nonlinear nature of our variables, to impose restrictions on negative values, and for ease of interpretation, we transform our variables into logarithmic forms. Given that some independent variables contain values equal to zero in certain years, one is added to the value before transforming them logarithmically. Table 1 summarizes the variables used in the analyses and their corresponding data sources.

Table 1.

Variable definitions

Variable	Variable constructed	Source
${\text{lnCO}}_{2it}$	${\text{lnCO}}_{2it}=\text{log}\left({\text{CO}}_{2it}\right)$ ${\text{CO}}_{2it}={\text{CO}}_2$ emissions per capita (metric tons)	WDI
$ln{\text{GDPP}}_{\mathit{it}}$	$ln{\text{GDPP}}_{\mathit{it}}=\text{log}\left({\text{GDPP}}_{\mathit{it}}\right)$ ${\text{GDP}}_{\mathit{it}}$ = GDP per capita (constant 2010 US$)	WDI
$lnUR{B}_{\mathit{it}}$	${\text{lnURB}}_{\mathit{it}}=\text{log}\left({\text{URB}}_{\mathit{it}}\right)$ ${\text{URB}}_{\mathit{it}}$ = Urban population (as a % of the total population)	WDI
${\text{lnOPE}}_{\mathit{it}}$	${\text{lnOPE}}_{\mathit{it}}=\text{log}\left({\text{OPE}}_{\mathit{it}}\right)$ ${\text{OPE}}_{\mathit{it}}$ = Trade Openness (as a % of GDP(	WDI
${\text{lnRE}}_{\mathit{it}}$	${\text{lnRE}}_{\mathit{it}}=\text{log}\left(1+{\text{RENEW}}_{\mathit{it}}\right)$ ${\text{RE}}_{\mathit{it}}$ = Renewable energies (as a % of Total Electricity Installed Capacity)	IRENA
$lnR{D}_{\mathit{it}}$	$lnR\&D_{\mathit{it}}=\text{log}\left(R\&D_{\mathit{it}}\right)$ $R\&D_{\mathit{it}}=$ R&D expenditure (as a % of GDP)	WDI
${\text{lnDPI}}_{\mathit{it}}$	${\text{lnDPI}}_{\mathit{it}}=\text{log}\left(1+{\text{DPI}}_{\mathit{it}}\right)$ ${\text{DPI}}_{\mathit{it}}$ = Discussion about Pandemics Index	WDI

WDI World Development Indicators, https://databank.worldbank.org/, IRENA International Renewable Energy Agency, https://www.irena.org/Statistics

The summary statistics of the main variables used in the empirical analyses are summarized in Table 2. The standard deviations for most variables are much lower than their means, showing a low amount of temporal volatility in the variables and the lack of outliers despite the relatively long duration.

Table 2.

Summary statistics (54 countries observed between 2003 and 2017)

Variable	Mean	Median	Max	Min	Std. Dev	Obs
${\text{lnCO}}_{2it}$	1.625	1.775	3.351	− 1.184	0.772	810
${\text{lnGDPP}}_{\text{it}}$	9.490	9.506	11.425	6.277	1.183	810
${\text{lnURB}}_{\text{it}}$	4.204	4.254	4.605	3.277	0.279	810
${\text{lnOPE}}_{\text{it}}$	4.357	4.367	6.081	3.072	0.516	810
${\text{lnRE}}_{\text{it}}$	3.057	3.254	4.610	0.000	1.103	810
$\text{lnR}\&{\text{D}}_{\text{it}}$	− 0.182	− 0.133	1.572	− 2.780	0.995	810
${\text{lnDPI}}_{\text{it}}$	2.408	0.000	176.557	0.000	12.183	810

Figure 2 depicts the historical CO₂-emission data in the 54 assessed countries from 2000 to 2019. China, the USA, and India are the three most industrialized countries in the world, with the largest CO₂ emissions. China has been the world’s largest polluter since 2005, after surpassing the United States. The United States has been gradually lowering its emissions, whereas China and India have sharply increased their emissions. Russia and Japan are the fourth- and fifth-largest polluters, respectively, while Armenia, with 6,170 kilotons of CO₂ in 2019, is the smallest emitter in our sample. According to the International Energy Agency, China recorded its highest CO₂ emissions of 11.95 billion tons in 2021, and the expected trajectory of its future emissions shows no signs of decreasing.6 Last, China has substantially led the increase in global CO₂ emissions above prepandemic levels, mainly due to its consumer-goods exports and heavy reliance on coal.

Fig. 2.

Million kilotons CO₂ emission in the assessed countries from 2000 to 2019. is the rest of the world.

Source: World Development Indicator

Econometrics modeling approach

We utilize a few spatial econometrics models to account for the spatial dependence among countries. In particular, we estimate three different models, including a one-way spatial-fixed-effects model, a one-way time-fixed-effects model, and a two-way time-fixed-effects and spatial-fixed-effects model. We also present the results of a Panel Estimated Generalized Least Square (PEGLS) (two-way error-component random-effects) model for robustness-check purposes. To find the optimal model, we employ a battery of diagnostic tests.

Theoretically, Eq. (1) summarizes a two-way time-fixed-effects and spatial-fixed-effects model (Wooldridge, 2021).

$$y_{\mathit{it}}={\pi }_i+{\gamma }_t+\beta x_{\mathit{it}}+{\epsilon }_{\mathit{it}},i=1,...,N\text{and}t=1,...,T.$$ 1

${\pi }_i$ represents the individual-country-specific fixed effect or the unobserved individual-country effect (i.e., heterogeneity). It refers to all time-invariant factors that impact the value of country $i$’s outcome variable. Including such a variable in the model decreases the possibility of omitted variable bias, thereby enhancing the accuracy of the estimation results. In addition, ${\gamma }_t$ is the time fixed effect, which captures the same information as a dummy variable for each time period. This encompasses any factors that are present for every $i$ within a specific time period and has the potential to impact the outcome variable. $x_{\mathit{it}}$ is a vector of explanatory variables of country $i$ at time $t$, $\mathrm{\beta }$ is a vector of corresponding coefficients, and ${\epsilon }_{\mathit{it}}$ is a vector of the unobserved error terms. The two-way time-fixed-effects and spatial-fixed-effects model reduces to a time-fixed-effects model when ${\alpha }_i=0$. Furthermore, if ${\gamma }_t=0$, Eq. (1) reduces to a one-way spatial-fixed-effects model. We estimate our specifications described below using all three possible models.7

By incorporating either spatial or time-period fixed effects, we exclude spatial and/or time-period averages, which in turn yields robust and reliable results (Wooldridge, 2021). These approaches also individually and collectively allow for correlation between explanatory variables and unobserved heterogeneity in nonlinear models, such as those in ours (Wooldridge, 2021). Due to inconsistent results in the literature regarding the influence of R&D and RE on CO₂ emission, a two-way time-fixed-effects and spatial-fixed-effects model is a viable choice since it eliminates structural changes in the model that have the same effect on all countries (Wooldridge, 2021).

Empirically, we estimate six equations ((2)–(7)) using the latter spatial econometrics models for each equation in this study. We progressively add our variables of interest (the Discussion about Pandemics Index ($\text{lnDPI}$), RE production ($\text{lnRE}$), and R&D expenditure ($lnR\&D$)) and their interaction terms, resulting in six distinct specifications. We do so to avoid the possibility of collinearity among variables. All specifications have logarithms of GDP per capita ($\text{lnGDPP}$), the squared term of GDP per capita (${\text{lnGDPP}}^2$), urbanization ($\text{lnURB}$), and trade openness ($\text{lnOPE}$) in common. The first specification (Eq. (2)) includes the common variables and $\text{lnDPI}$. We substitute $\text{lnDPI}$ with $lnR\&D$ in the second specification (Eq. (3)) and $\text{lnRE}$ in the third specification (Eq. (4)). These two specifications allow us to estimate the direct effect of R&D and RE on CO₂ emissions, respectively. We include $lnR\&D$ and $\text{lnRE}$ in specification 4 (Eq. (5)) simultaneously and their interaction term in the fifth specification (Eq. (6)). For the sixth specification (Eq. (7)), we add the $lnR\&D\times lnGDPP$ interaction term to the fifth specification.

$$lnC{O}_{2it}={\beta }_1+{\beta }_{2}lnGDP{P}_{\mathit{it}}+{\beta }_{3}lnGDP{P}_{\mathit{it}}^2+{\beta }_{4}lnOP{E}_{\mathit{it}}+{\beta }_{5}lnUR{B}_{\mathit{it}}+{\beta }_{6}lnDP{I}_{\mathit{it}}$$ 2

$$lnC{O}_{2it}={\beta }_1+{\beta }_{2}lnGDP{P}_{\mathit{it}}+{\beta }_{3}lnGDP{P}_{\mathit{it}}^2+{\beta }_{4}lnOP{E}_{\mathit{it}}+{\beta }_{5}lnUR{B}_{\mathit{it}}+{\beta }_{6}lnR\&D_{\mathit{it}}$$ 3

$$lnC{O}_{2it}={\beta }_1+{\beta }_{2}lnGDP{P}_{\mathit{it}}+{\beta }_{3}lnGDP{P}_{\mathit{it}}^2+{\beta }_{4}lnOP{E}_{\mathit{it}}+{\beta }_{5}lnUR{B}_{\mathit{it}}+{\beta }_{6}lnR{E}_{\mathit{it}}$$ 4

$$lnC{O}_{2it}={\beta }_1+{\beta }_{2}lnGDP{P}_{\mathit{it}}+{\beta }_{3}lnGDP{P}_{\mathit{it}}^2+{\beta }_{4}lnOP{E}_{\mathit{it}}+{\beta }_{5}lnUR{B}_{\mathit{it}}+{\beta }_{6}lnR\&D_{\mathit{it}}+{\beta }_{7}lnR{E}_{\mathit{it}}$$ 5

$$lnC{O}_{2it}={\beta }_1+{\beta }_{2}lnGDP{P}_{\mathit{it}}+{\beta }_{3}lnGDP{P}_{\mathit{it}}^2+{\beta }_{4}lnOP{E}_{\mathit{it}}+{\beta }_{5}lnUR{B}_{\mathit{it}}+{\beta }_{6}lnR{D}_{\mathit{it}}+{\beta }_{7}lnR{E}_{\mathit{it}}+{\beta }_8(lnR{D}_{\mathit{it}}\times lnR{E}_{\mathit{it}})$$ 6

$$\begin{array}{c}lnC{O}_{2it}={\beta }_1+{\beta }_{2}lnGDP{P}_{\mathit{it}}+{\beta }_{3}lnGDP{P}_{\mathit{it}}^2+{\beta }_{4}lnOP{E}_{\mathit{it}}+{\beta }_{5}lnUR{B}_{\mathit{it}}\\ +{\beta }_{6}lnR\&D_{\mathit{it}}+{\beta }_{7}lnR{E}_{\mathit{it}}+{\beta }_8(lnR\&D_{\mathit{it}}\times lnR{E}_{\mathit{it}})+{\beta }_9(lnR\&D_{\mathit{it}}\times lnGDP{P}_{\mathit{it}})\\ \end{array}$$ 7

$i$ denotes the member country (i $=1,2,...,54$). $\mathit{ln}$ represents the natural logarithm of the variables; the calculated coefficients are therefore elasticities. In Eqs. (6) and (7), the marginal effect of a change in the logarithms of RE production and GDP per capita affects CO₂ emissions as follows:

$$\frac{d\left(lnC{O}_{2it}\right)}{d\left(lnR{E}_{\mathit{it}}\right)}={\beta }_7+{\beta }_{8}lnR\&D_{\mathit{it}}$$ 8

$$\frac{d\left(lnC{O}_{2it}\right)}{d\left(lnGDP{P}_{\mathit{it}}\right)}={\beta }_2+2\times {\beta }_{3}lnGDP{P}_{\mathit{it}}+{\beta }_{9}lnR\&D_{\mathit{it}}$$ 9

As a result, Eqs. (6) and (7) allow us to examine the spillover effects of R&D on how RE and GDP per capita affect CO₂ emissions. The reason for examining such effects is that, in addition to having a direct effect on the expansion of production in countries (energy-consumption scale effect), R&D can also result in a shift toward more environmentally friendly production structures (energy-use efficiency effect) and the production of RE sources with advanced technologies in the studied countries. Therefore, the environmental consequences of production in countries with varying levels of R&D may fluctuate. Such implications can be examined by separating R&D’s direct and spillover effects in Eqs. (6) and (7).

Estimation and analysis of the results

Estimation results

As described in the previous section, we empirically estimate three panel models for each of the six specifications (Eqs. (2)–(7)). The three models are a one-way spatial-fixed-effects model, a one-way time-fixed-effects model, and a two-way time-fixed-effects and spatial-fixed-effects model. We also estimate a two-way error-component random-effects PEGLS model for comparison and robustness-check purposes. Figure 3 depicts the empirical procedure we follow in the current study.

Fig. 3.

Flow chart of the empirical process used in this research

Initially, we examine the stationarity of the variables using four different panel unit-root tests, including the Levin, Lin and Chu t-test; Im, Pesaran, and Shin wald test; ADF-Fisher chi-square test; and PP-Fisher chi-square test. Table 3 summarizes the panel unit-root test results. The results reveal that the majority of the variables are not stationary at level, I(0). These variables become stationary after first difference, I(1).

Table 3.

Panel unit-root test results.

Source: Authors’ estimations

	Method	Level		First difference
		Statistic	Prob	Statistic	Prob
${\text{lnCO}}_{\text{2it}}$	Levin, Lin, and Chu t	− 7.033	0.000	− 24.337	0.000
	Im, Pesaran, and Shin W-stat	0.183	0.573	− 19.032	0.000
	ADF—Fisher Chi-square	121.630	0.175	500.946	0.000
	PP—Fisher Chi-square	147.311	0.007	584.315	0.000
${\text{lnGDPP}}_{\text{it}}$	Levin, Lin, and Chu t	− 10.964	0.000	− 14.665	0.000
	Im, Pesaran, and Shin W-stat	− 3.944	0.000	− 9.199	0.000
	ADF—Fisher Chi-square	175.562	0.000	271.121	0.000
	PP—Fisher Chi-square	305.370	0.000	300.057	0.000
${\text{lnURB}}_{\text{it}}$	Levin, Lin, and Chu t	− 74.532	0.000	− 4.186	0.000
	Im, Pesaran, and Shin W-stat	− 25.017	0.000	1.958	0.975
	ADF—Fisher Chi-square	147.226	0.003	94.860	0.728
	PP—Fisher Chi-square	441.415	0.000	144.615	0.005
${\text{lnOPE}}_{\text{it}}$	Levin, Lin, and Chu t	− 5.090	0.000	− 23.926	0.000
	Im, Pesaran, and Shin W-stat	− 1.517	0.065	− 24.989	0.000
	ADF—Fisher Chi-square	135.740	0.037	477.115	0.000
	PP—Fisher Chi-square	138.476	0.026	721.477	0.000
${\text{lnDPI}}_{\text{it}}$	Levin, Lin, and Chu t	− 15.700	0.000	− 22.227	0.000
	Im, Pesaran, and Shin W-stat	− 11.064	0.000	− 16.542	0.000
	ADF—Fisher Chi-square	211.353	0.000	443.554	0.000
	PP—Fisher Chi-square	333.163	0.000	579.584	0.000
$lnR\&D_{\mathit{it}}$	Levin, Lin, and Chu t	− 3.685	0.000	− 17.408	0.000
	Im, Pesaran, and Shin W-stat	− 0.594	0.276	− 13.551	0.000
	ADF—Fisher Chi-square	152.545	0.003	378.658	0.000
	PP—Fisher Chi-square	110.976	0.403	454.271	0.000
${\text{lnRE}}_{\text{it}}$	Levin, Lin, and Chu t	4.709	1.000	− 9.788	0.000
	Im, Pesaran, and Shin W-stat	10.087	1.000	− 7.895	0.000
	ADF—Fisher Chi-square	45.626	1.000	293.508	0.000
	PP—Fisher Chi-square	47.169	1.000	341.455	0.000

Next, we conduct the Kao panel cointegration test. The Kao panel cointegration test examines the long-run relationships among the assessed variables. Table 4 provides a summary of the cointegration test’s results, which suggest statistical significance at the 5% level. We find evidence that rejects the no-cointegration null hypothesis for all models. This finding demonstrates the presence of a long-run cointegration relationship among variables.

Table 4.

Kao residual cointegration test results.

Source: Authors’ estimations

	t-Statistic	Prob
ADF	− 1.981	0.024
Residual variance	0.0032
HAC variance	0.004

To determine what model best fits our data, we next carry out three diagnostic tests, including the likelihood-ratio tests, Redundant Fixed-Effects test, and Hausman test. Table 5 presents these tests’ results. We employ two likelihood-ratio tests to determine the possibility of time-period and spatial fixed effects being present in the model. Simultaneous spatial- and time-fixed-effects models are compared to models with either spatial or time-period fixed effects. The null hypothesis precludes models with varying fixed effects in terms of spatial and temporal effects. The high level of significance of the test statistics for all six specifications suggests rejecting the null hypotheses for all six specifications. This implies that both time-period and spatial fixed effects must be considered simultaneously when estimating the equations.

Table 5.

Diagnostic test results.

Source: Authors’ estimations

	Spatial fixed effects		Time-period fixed effects		Redundant fixed-effects tests		Hausman test
Model 1	279.10	(0.000)	2783.50	(0.000)	302.49	(0.000)	21.43	(0.000)
Model 2	281.83	(0.000)	2783.37	(0.000)	302.15	(0.000)	22.71	(0.000)
Model 3	183.56	(0.000)	2489.60	(0.000)	230.49	(0.000)	22.17	(0.000)
Model 4	184.94	(0.000)	2491.63	(0.000)	230.38	(0.000)	24.11	(0.000)
Model 5	171.93	(0.000)	2393.67	(0.000)	206.41	(0.000)	26.76	(0.000)
Model 6	160.37	(0.000)	2395.07	(0.000)	209.82	(0.000)	29.04	(0.000)

Further, we employ the Redundant Fixed-Effects test to determine the joint significance of cross-section effects. The test statistics are all significant at the conventional levels for the six specifications, further validating the likelihood-ratio test results by indicating that a spatial-fixed-effects model cannot be excluded. This also implies that the one-way time-fixed-effects model results are not trustworthy. Last, we employ the Hausman test to examine whether to abandon the fixed-effects model in favor of a random-effects model. The null hypothesis is rejected for all six specifications at a 1% significance level, implying that fixed effects are confirmed as an adequate model specification. Therefore, we move forward by discussing only the two-way time-fixed-effects and spatial-fixed-effects model findings below. The one-way models’ regression results for the six specifications are provided in Table A1 and Table A2 of the Appendix.

Table 6 summarizes the regression results for each of the six specifications (Eqs. (2)–(7)) using the two-way models. The first two (Eqs. (2)–(3)), the second two (Eqs. (4)–(5)), and the last two specifications (Eqs. (6)–(7)) yield similar results. Our findings across the six specifications indicate that a one-percentage-point increase in GDP per capita increases CO₂ emissions between 1.3% and 1.75%. The negative and statistically significant coefficient of squared GDP per capita across specifications suggests an inverted-U-shaped relationship between GDP per capita and CO₂ emissions, confirming the EKC hypothesis in our panel. There are two ways that GDP per capita can affect CO₂ emissions: income-induced technique effect and scale effect (Awaworyi Churchill et al., xxxx; Hübler & Keller Feb., 2010; Petrović & Lobanov, 2020). The former indicates that higher GDP per capita heightens the need for a cleaner environment, leading to stricter environmental regulations, thus reducing CO₂ emissions. The latter suggests that higher GDP per capita results in more energy consumption and hence a higher level of fossil fuel combustion and CO₂ emissions. Depending on which effect is dominant, higher GDP per capita can increase or decrease CO₂ emissions. Both effects are present in our results, though the scale effect is the dominant effect.

Table 6.

Time-period and spatial fixed-effects regression estimation results for Eqs. (2–7).

Source: Authors’ estimations

	Equation (2)		Equation (3)		Equation (4)			Equation (5)		Equation (6)		Equation (7)
$\text{lnGDPP}$	1.713	(0.000)	1.749	(0.000)	1.520	(0.000)	1.559		(0.000)	1.438	(0.000)	1.366	(0.000)
${\text{lnGDPP}}^2$	− 0.063	(0.000)	− 0.065	(0.000)	− 0.051	(0.000)	− 0.053		(0.000)	− 0.047	(0.000)	− 0.045	(0.000)
$\text{lnURB}$	0.755	(0.000)	0.720	(0.000)	0.752	(0.000)	0.708		(0.000)	0.724	(0.000)	0.736	(0.000)
$\text{lnOPE}$	0.035	(0.217)	0.034	(0.224)	0.068	(0.017)	0.067		(0.017)	0.051	(0.080)	0.064	(0.031)
$\text{lnDPI}$	0.000	(0.962)
$lnR\&D$			0.023	(0.189)			0.026		(0.138)	0.102	(0.004)	0.383	(0.007)
$\text{lnRE}$					− 0.080	(0.000)	− 0.080		(0.000)	− 0.075	(0.000)	− 0.070	(0.000)
$lnRE\times lnR\&D$										− 0.026	(0.013)	− 0.030	(0.004)
$lnGDPP\times lnR\&D$												− 0.030	(0.042)
Log-likelihood	788.123		788.99		802.853		803.96			807.09		809.17
$R^2$	0.986		0.986		0.986		0.986			0.987		0.987

Across the six specifications, we find that a one-percentage-point increase in urbanization worsens the environment by increasing CO₂ emissions between 0.71% and 0.76%. This is to be expected, as a higher urbanization level can lead to higher demand for energy, transportation, and natural resource extraction, all of which increase CO₂ emissions. Further, our findings suggest that a higher level of trade openness worsens the environment by increasing CO₂ emission levels. A one-percentage-point increase in foreign trade increases CO₂ emissions between 0.05% and 0.07% (Eqs. (4)–(7)). Openness to trade increases economic growth and thus GDP per capita (Grossman & Krueger May, 1995), which in turn increases CO₂ emissions through the scale effect (Hübler & Keller Feb., 2010; Petrović & Lobanov, 2020). Another reason for such a finding can be linked to the transportation sector through which trades occur. Our findings on GDP per capita, urbanization, and trade openness corroborate previous empirical studies (Awaworyi Churchill et al., xxxx; Bekun et al., 2019; Bilgili et al., 2016; Chen et al., 2019; Hu et al., 2018; Hübler & Keller Feb., 2010; Li & Wang Dec., 2017; Moutinho & Robaina, 2016; Petrović & Lobanov, 2020). Last, we find that previous epidemic episodes in the assessed countries had a nonsignificant impact on CO₂ emissions (Eq. (2)). Thus, we remove this variable from the other specifications.

Now, we turn our attention to the main explanatory variables of interest, RE and R&D. Based on the fourth specification (Eq. (5)), we observe a positive association between the R&D coefficient and CO₂ emissions, and the coefficient is marginally significant at the 14% level. A one-percentage-point increase in R&D expenditure led to increasing CO₂ emissions by 0.026%. We observe a similar association between the two variables in the third specification (Eq. (4)), although the R&D coefficient is nonsignificant. Previous literature finds a positive correlation between R&D and economic growth and foreign trade (Awaworyi Churchill et al., xxxx; Minniti & Venturini Feb., 2017). That is, a higher level of R&D expenditure is associated with higher economic growth and foreign trade. Researchers have also shown that more advanced technologies (higher R&D) can reduce CO₂ emissions intensity, lowering aggregate CO₂ emissions. The latter phenomenon is known as the intensity effect of R&D, while the former is called the scale effect of R&D (Li & Wang Dec., 2017). As explained earlier, we find that GDP per capita and trade openness increase CO₂ emissions through the scale effect. Thus, a positive and statistically significant R&D coefficient indicates that the scale effect of R&D is at play in our panel. In other words, the studied countries use R&D expenditure to cultivate economic growth and trade openness, which in turn increase CO₂ emissions. Although more advanced technologies potentially increase overall energy efficiency, a higher level of output demands more natural resource extraction and exploration, which in turn can increase CO₂ emissions (Awaworyi Churchill et al., xxxx; Li & Wang Dec., 2017). Although these results are incomplete, as explained below, they corroborate some of the existing literature (Adebayo et al., 2022a; Cai et al., 2021; Sajid & Gonzalez Aug., 2021; Weimin et al., Apr. 2022; Xie et al., Sep. 2022; Yuping et al., Nov. 2021).

Contrary to the R&D results, the RE coefficient is negatively associated with CO₂ emissions, and the coefficient is statistically significant at the conventional levels. Our findings indicate that a one-percentage-point increase in RE nameplate capacity and thus generation leads to lowering CO₂ emissions by 0.08% (Eqs. (4) and (5)). This finding implies that higher deployment of renewable sources such as hydropower, wind, solar, geothermal, and bioenergy can result in reducing CO₂ emissions in the selected countries. Similar results are found in the literature (Dai et al., 2022; Mamkhezri et al., 2021; Nguyen & Kakinaka Mar., 2019; Rahman et al., Nov. 2022). In addition, the public shares the same perspective on the RE-CO₂ nexus and is supportive of policies that promote integrating higher levels of RE in the energy mix (Mamkhezri, 2019).

Since most RE power plants (e.g., wind and solar) require advanced technology and have significant upfront costs, the interactions (spillover effects) between R&D and RE as well as R&D and economic growth should also be explored. The last two specifications (Eqs. (6) and (7)) take the spillover effect of R&D on RE and GDP per capita into account. Incorporating these interaction terms results in two distinct effects. First, it results in a positive and highly significant R&D coefficient, reinforcing the scale-effect results from the fourth specification. Second, the interaction coefficients are negative and significant in both specifications. Equations (10) and (11) summarize the marginal effect of RE and GDP per capita on CO₂ emissions, respectively, considering R&D’s spillover effect.

$$\frac{d\left({\text{lnCO}}_{2it}\right)}{d\left({\text{lnRE}}_{\mathit{it}}\right)}=-0.075-0.026\times lnR\&D_{\mathit{it}}$$ 10

$$\frac{d\left({\text{lnCO}}_{2it}\right)}{d\left({\text{lnGDPP}}_{\mathit{it}}\right)}=1.366-0.09\times lnGDP{P}_{\mathit{it}}-0.03\times lnR\&D_{\mathit{it}}$$ 11

According to Eq. (10), installing higher RE capacity and generation lowers CO₂ emissions; this reduction is heightened for countries with higher R&D expenditure. In other words, countries that allocate a higher budget for R&D expenditure that is spent on advancing RE technologies and installation, which in turn lowers CO₂ emissions, can counter the scale effect of technology via the intensity effect of R&D. Holding R&D expenditure constant at its mean or median value (see Table 2), Eq. (10) suggests that a one-percentage-point increase in RE nameplate capacity and thus generation can reduce CO₂ emissions by 0.1%.8 Thus, the R&D contribution to the reduction of CO₂ emissions here (0.022%) is approximately identical to the R&D scale effect of 0.026%. This justifies the nonsignificance or marginal significance of the R&D coefficient in the specifications that ignored its spillover effects (Eqs. (3) and (5)). Using the maximum (1.572) and minimum (− 2.78) values of the natural log of R&D presented in Table 2, one can conclude that the impact of a one-percentage-point increase in RE can reduce CO₂ emissions between 0.08% and 0.20%. This is an important finding indicating that the intensity effect of R&D dominates the scale effect of technology in countries with higher-than-average R&D expenditure. The opposite holds for the countries with lower-than-average R&D expenditure spent on RE technologies.9

Equation (11) presents a similar story to that of Eq. (10). While economic growth results in increased CO₂ emissions, the negative coefficient for R&D in Eq. (11) indicates that such increased emissions are lower in countries with higher R&D expenditure. Therefore, it can be concluded that advanced technology and environmentally friendly RE production in countries with high R&D intensity have a positive and significant environmental impact. There are only two papers with similar results that we can find: those of (Khezri et al., 2021) and (Li & Wang Dec., 2017). Previous studies that find that the scale effect of technology is the dominant effect tend to ignore the spillover effect of R&D and thus inaccurately emphasize the positive association of R&D and CO₂ emissions (Fig. 3).

Additionally, evaluating data from the assessed economies gives further insight into the elements that contribute to RE’s increased capacity to mitigate CO₂ emissions. A survey of the components of RE in 54 countries studied from 2003 to 2017 reveals that the share of RE components has shifted significantly in favor of countries with more R&D expenditure. Figure 4 depicts how the average percentage of hydropower energy production in these countries halved from 65% in 2003 to 33% in 2017. This figure demonstrates the intensity effect of R&D in our international panel graphically. Solar-energy production increased from 0.5% in 2003 to approximately 29% in 2017. Similarly, wind-energy generation accounted for 16% of total energy production in 2003 and 24% in 2017. The percentage of bioenergy produced decreased from 18 to 14%. Similar shifts are experienced in the case with R&D index of 1.19. Thus, high-tech RE sources (i.e., solar and wind energy) have displaced the older and more established RE sources (i.e., hydropower and bioenergy) as the primary energy-generation technologies in countries with high levels of R&D.

Fig. 4.

Changes in the various components of RE for different levels of R&D

Nevertheless, the averages remain nearly constant for countries with less R&D expenditure. Such a shift in the composition of RE components in countries with more R&D expenditure indicates that one measure of R&D effectiveness is the transition toward RE with lower pollution levels than traditional types. These changes occurred as a result of a shift in the approach to RE production from an older one (with less technologically advanced methods) to a newer one (with more technologically advanced techniques) in the countries with more R&D. In other words, technical developments have replaced fossil fuels not just with solar and wind energy, but also with other conventional RE sources. These effects manifested as a negative R&D coefficient in the interaction term, implying that one of the spillover effects of R&D is a transition toward RE compounds that result in lower pollution levels and positive environmental dimensions of R&D in the selected countries.

Robustness check

For robustness-check and comparison purposes, we reestimate our six specifications using the two-way error-component random-effects PEGLS model. Table A3 of the Appendix summarizes the regression results for each specification using the PEGLS method. The new set of results has the same signs and similar significance levels to those of the two-way time-fixed-effects and spatial-fixed-effects model findings. Therefore, we conclude that the PEGLS results present the same story and that our two-way time- and spatial-fixed-effects model results are robust. That is, the results confirm the EKC hypothesis and confirm that higher RE installation reduces CO₂ emissions while the intensity effect of R&D dominates the scale effect of technology for the selected countries that include RE in their energy portfolio.

Conclusion and policy implications

The highly mixed findings of empirical research on the impacts of RE and R&D on the environment motivated the current study to assess the spillover effects of RE and R&D on CO₂ emissions in 54 countries from 2003 to 2017. In so doing, we controlled for other CO₂ emissions determinants such as GDP, GDP², trade openness, and urbanization level. In addition, we investigated the relationship between epidemic episodes and CO₂ emissions from 2003 to 2017. The unit-root and cointegration test results demonstrate that all the assessed variables’ series, besides that of urbanization, are I(1). These results confirm the presence of a long-run relationship among the variables. Our diagnostic tests also confirm that the optimal econometric estimation technique is a two-way time- and spatial-fixed-effects model. Our analysis consists of several regression models executed in sequence and uniquely able to build a body of evidence regarding the spillover impacts of R&D and RE on CO₂ emissions in the selected countries.

We found that epidemic episodes before COVID-19 had a nonsignificant impact on CO₂ emissions. This result is justifiable because in recent decades, before COVID-19, there were no similar drastic measures taken to reduce human activities due to epidemic episodes. In other words, previous epidemic episodes were less severe than COVID-19, and thus this finding cannot be generalized to include 2020 and beyond.

Our findings suggest that both the income-induced technique effect and the scale effect of economic growth are present in our panel, though the scale effect is dominant. This result indicates that higher GDP per capita leads to increasing energy demand and thus a higher level of fossil fuel combustion, which in turn increases CO₂ emissions. Our results also confirm the EKC hypothesis internationally and across different specifications. This implies that higher economic growth increases CO₂ emissions at a decreasing rate. In addition, we find that higher levels of urbanization and trade openness harm the environment by increasing CO₂ emission levels, which indicates that the assessed countries have a weak environmental regulatory framework. Therefore, policy makers in the assessed countries and worldwide are advised to endorse sustainable-economic-growth policies that lead to cleaner foreign trade and green urbanization by enacting stricter environmental regulations domestically and internationally. Policies such as a carbon tax for the transportation and electric sectors have the potential to incentivize the reduction of high energy-intensive activities or uncessary transportation for local and international trade. As for green urbanization, policy makers should consider enforcing the installation of rooftop solar on commercial and industrial buildings as well as encouraging community solar. In addition, clean energy sources, such as wind and solar, and advanced technologies with higher efficiency levels (same output with less energy use) such as electric vehicles should become a priority for sustainable international economic growth.

One way to overcome the scale effect derived from economic growth and promote the income-induced technique effect is to shift away from dirty fossil fuels and toward more environmentally friendly and cleaner energy sources such as advanced small nuclear modular reactors, carbon-capture technology, green hydrogen, and RE sources. Assessed countries should also promote the reduction of energy intensity through enhancing energy efficiency and reducing the consumption of fossil fuels such as coal and oil. Further, our findings confirm that policy makers ought to consider the spatial nature of CO₂ emissions and neighboring economies while enacting regulations.

Our RE and R&D-spillover results further back the recommendation of moving toward high-tech clean energy sources. We found that more RE generation can alleviate the environmental strain by reducing CO₂ emissions. Hence, this finding highlights that the assessed countries and others should reduce reliance on fossil fuels (see Fig. 1) and enact policies that motivate RE installation. Further, we found that R&D expenditure on RE technologies reduces the strain on the environment even further. Our results indicate that R&D has a direct, positive, and statistically significant relationship with increasing CO₂ emissions (i.e., the scale effect of technology). However, the spillover effect of R&D on RE and GDP per capita promotes the intensity effect of R&D such that it dominates the scale effect of technology and leads to an overall reduction of CO₂ emissions. Thus, we conclude that R&D expenditure has a statistically significant and negative overall (direct plus spillover) effect on CO₂ emissions. These results have significant policy implications. First, countries with minimal RE in their generation mix should emphasize reducing CO₂ emissions through advanced carbon-reduction technologies or promoting RE. Second, a positive R&D-investment spillover effect from developed countries has the potential to curtail CO₂ emissions in less wealthy or developing countries. Therefore, developing countries should import advanced technology, especially in their energy sector, to reduce CO₂ emissions. Last, we argue that previous studies that ignore the spillover effects of R&D and find a positive relationship between CO₂ emissions and R&D investment are biased. Considering the spillover effect of R&D on RE generation, we show that the negative RE-CO₂ nexus may vary according to the country’s (green) R&D investment conditions. These findings fill a void in empirical research investigating the effect of R&D and RE on greenhouse gas emissions.

Our study is conditional on some caveats. One of the caveats of the current study is that we use aggregated RE as a variable in our analyses rather than deconstructing RE into its components (i.e., solar, wind, bioenergy, hydropower, and geothermal) (Mamkhezri & Torell, 2022; Khezri et al., 2022). Future research should investigate the spillover effect of different types of RE technologies on CO₂ emissions and other pollutants at the international level when the data become available. Our results cannot be extrapolated to pollutants (e.g., PM_2.5, NOx, SO₂) other than CO₂. Further, since R&D expenditures contain both capital and current expenditures in the primary four sectors of business enterprise, government, higher education, and private nonprofit, future researchers should control for other variables such as different types of related economic-freedom indicators, economic-complexity indexes, foreign direct investment, agriculture’s share of GDP, and other types of pollutants, to name a few. Our results were estimated using two-way time- and spatial-fixed-effects models; future research should estimate similar models using threshold, nonparametric, and other spatial econometrics models such as Spatial Durbin models (Mamkhezri et al., 2020c, 2022b, c; Wang et al., 2022) to capture nonlinearity and spatial heterogeneity in the data. Future studies are also encouraged to explore the CO₂-COVID-19 nexus in the context of RE and R&D when the data become available. Nonetheless, our findings deliver fresh insights for policy makers endorsing R&D and energy-related policies for sustainable global development.

Appendix

See Tables 7, 8 and 9.

Table 7.

Spatial-fixed-effects regression estimation results for Eqs. (1)–(6)

	Equation (2)		Equation (3)		Equation (4)		Equation (5)		Equation (6)		Equation (7)
$\text{lnGDPP}$	1.905	(0.000)	1.793	(0.000)	1.360	(0.000)	1.339	(0.000)	1.126	(0.000)	0.989	(0.000)
${\text{lnGDPP}}^2$	− 0.094	(0.000)	− 0.087	(0.000)	− 0.056	(0.000)	-0.055	(0.000)	− 0.044	(0.002)	− 0.039	(0.005)
$\text{lnURB}$	− 0.272	(0.110)	− 0.042	(0.799)	0.152	(0.306)	0.188	(0.221)	0.248	(0.103)	0.298	(0.050)
$\text{lnOPE}$	− 0.049	(0.102)	− 0.046	(0.123)	0.031	(0.285)	0.032	(0.271)	0.008	(0.775)	0.037	(0.205)
$lnR\&D$			− 0.041	(0.045)			-0.017	(0.364)	0.130	(0.001)	0.724	(0.000)
$\text{lnRE}$					− 0.169	(0.000)	-0.167	(0.000)	− 0.155	(0.000)	− 0.140	(0.000)
$\text{lnRE}\times \text{ln}R\&D$									− 0.050	(0.000)	− 0.057	(0.000)
$\text{lnGDPP}\times lnR\&D$											− 0.063	(0.000)
$\text{lnDPI}$	− 0.001	(0.025)
Log-likelihood	648.574		648.074		711.075		711.489		721.122		728.984
$R^2$	0.980		0.980		0.983		0.983		0.983		0.984

Table 8.

Time-period fixed-effects regression estimation results for Eqs. (1)–(6)

	Equation (2)			Equation (3)				Equation (4)				Equation (5)				Equation (6)					Equation (7)
$\text{lnGDPP}$	1.391	(0.000)		1.409		(0.000)			1.319		(0.000)		1.322	(0.000)			1.219		(0.000)			1.580		(0.000)
${\text{lnGDPP}}^2$	− 0.053	(0.000)		− 0.056		(0.000)			− 0.043		(0.000)		-0.043	(0.000)			− 0.043		(0.000)			− 0.061		(0.000)
$\text{lnURB}$	0.330	(0.001)		0.334		(0.001)			− 0.360		(0.000)		-0.357	(0.000)			− 0.281		(0.001)			− 0.297		(0.001)
$\text{lnOPE}$	0.106	(0.003)		0.122		(0.001)			− 0.076		(0.012)		-0.074	(0.016)			− 0.026		(0.367)			− 0.031		(0.298)
$lnR\&D$				0.050		(0.057)							0.009	(0.667)			− 0.281		(0.000)			− 0.639		(0.004)
$\text{lnRE}$									− 0.306		(0.000)		-0.305	(0.000)			− 0.250		(0.000)			− 0.254		(0.000)
$lnRE\times lnR\&D$																	0.120		(0.000)			0.131		(0.000)
$\text{lnGDPP}\times lnR\&D$																						0.035		(0.099)
$\mathit{lnDPI}$	0.002	(0.183)
Log-likelihood	− 603.625				− 602.695					− 441.948			-441.855					− 389.745					− 388.369
$R^2$	0.563				0.564					0.707			0.707					0.743					0.743

Table 9.

PEGLS regression estimation results for Eqs. (1)–(6)

	Equation (2)		Equation (3)		Equation (4)		Equation (5)		Equation (6)		Equation (7)
$\text{lnGDPP}$	1.676	(0.000)	1.602	(0.000)	1.237	(0.000)	1.224	(0.000)	1.066	(0.000)	0.878	(0.000)
${\text{lnGDPP}}^2$	− 0.079	(0.000)	− 0.075	(0.000)	− 0.047	(0.000)	-0.047	(0.000)	− 0.038	(0.004)	− 0.030	(0.027)
$\text{lnURB}$	− 0.089	(0.581)	0.052	(0.744)	0.166	(0.228)	0.185	(0.195)	0.235	(0.098)	0.275	(0.052)
$\text{lnOPE}$	− 0.046	(0.129)	− 0.044	(0.151)	0.034	(0.228)	0.035	(0.222)	0.021	(0.473)	0.044	(0.132)
$lnR\&D$			− 0.035	(0.094)			-0.010	(0.609)	0.099	(0.010)	0.634	(0.000)
$\mathit{lnRE}$					− 0.180	(0.000)	-0.179	(0.000)	− 0.172	(0.000)	− 0.160	(0.000)
$\text{lnRE}\times lnR\&D$									− 0.037	(0.001)	− 0.045	(0.000)
$\text{lnGDPP}\times lnR\&D$											− 0.056	(0.000)
$\text{lnDPI}$	− 0.001	(0.172)
Hausman Test	21.434	(0.000)	22.712	(0.000)	22.174	(0.000)	24.112	(0.000)	26.762	(0.000)	29.036	(0.000)
$R^2$	0.192		0.192		0.339		0.339		0.352		0.361

Funding

No financial support was received to conduct this study.

Availability of data and materials

Table 1 summarizes the source of the data used.

Declarations

Conflict of interest

The authors declare no competing interests.