THE IMPACT OF FOSSIL AND NON-FOSSIL ENERGY CONSUMPTION ON ECONOMIC GROWTH

(1)

THE IMPACT OF FOSSIL AND

NON-FOSSIL ENERGY

CONSUMPTION ON

ECONOMIC GROWTH

Evidence from a Pooled Mean

Group analysis

(2)

(3)

Abstract

This paper investigates the causal relationship between economic growth and fossil and non-fossil energy consumption while controlling for human and physical capital in a panel of 74 countries over 1990 to 2017. The global sample was divided into four categories according to the World Bank income classifications: low-income, lower-middle income, upper-middle income, and high-income. The purpose of the income level subsamples was to further analyze the role of the country’s income level in the nexus between energy consumption and economic growth. This paper uses panel unit root and cross-sectional independence testing and employs the Pooled Mean Group estimator. The main results in this paper reveal (1) bidirectional causality between fossil energy consumption and economic growth for the low income and lower-middle income countries; (2) unidirectional causality from economic growth to fossil energy consumption for the high income and upper-middle income countries. In terms of non-fossil energy consumption, the results are more mixed; there is (3) unidirectional causality from economic growth to non-fossil energy consumption for the low-income and upper-middle income countries; (4) bidirectional causality between non-fossil energy consumption and economic growth for the lower-middle income countries; and (5) unidirectional causality from non-fossil energy consumption to economic growth for the high-income countries. These results have several policy implications that are discussed.

“The earth is a fine place

and worth fighting for.”

(4)

(5)

1. Introduction

The global climate change has created a public debate that has surged over the last couple of years, however, scientists and regulators have been tackling this issue for a rather long time. In 2005 the EU member states launched the EU emissions trading system (EU ETS), a cornerstone in EU’s energy policy of cost-effectively reducing greenhouse gas emissions. Emissions of carbon dioxide (CO2) are the main cause of anthropogenic global warming. Since pre-industrial

times greenhouse gas emissions have increased global temperatures by around one degree Celsius and the rate of increasing temperature has nearly doubled in the last 60 years (NASA, 2010). Under the same period, the world Gross Domestic Product (GDP) has rapidly increased and the average person in the world is 4.4-times richer today, as compared to 1950. Economic growth enables that everyone can become better off, even when considering a growing population globally (Roser, 2020).

However, with growing economies follow more production and consumption of energy. Ritchie and Roser (2019) showed that energy production and consumption is the primary driver of global climate change as the contemporary and historical systems that produce energy is dominated by fossil fuels (i.e. oil, coal, and gas) which cause CO2 and other greenhouse gases.

The risks of climate change are one of the world’s most pressing challenges, therefore it is interesting to consider the relationship between economic growth and energy consumption.

Fig. 1 show the correlation between energy consumption and GDP per capita for the global sample of countries in this paper, categorized according to their World Bank income level. It can be seen that countries with higher GDP per capita exhibit notably higher energy consumption per capita than the countries of lower-income levels. This correlation, however, does not explain the causal relationship, i.e. does energy consumption cause economic growth or is it the other way around? The relationship has been a topic for research for decades and more recently the literature has been extended by examining a possible relationship between renewable energy and economic growth. However, empirical evidence on the relationship in the energy-GDP nexus is mixed.

(7)

disaggregation of fossil and the non-fossil energy consumption is especially important since fossil energy is the dominating global contributor to CO2 emissions and non-fossil energy is the

climate friendly substitute (EIA, 2019). Previous research on the relationship between economic growth and energy consumption is predominantly executed with a bivariate model consisting of real GDP and total energy consumption. This paper implements a neoclassical aggregate production function where real GDP is a function of the physical capital stock, the human capital stock, total fossil, and non-fossil energy consumption. A global sample of 79 countries are divided into four subsamples based on the World Bank income classifications: high, upper-middle, lower-middle, and low. The purpose of the income level subsamples is to further analyze the role of the country’s income level in the energy-GDP nexus.

This paper aims to test the hypothesis of a unidirectional causal relationship running from fossil energy consumption to economic growth in the low-income countries, and a unidirectional causal relationship running from non-fossil energy consumption to economic growth in the high-income countries. This would support the importance of policymaking that substitute fossil energy consumption with non-fossil energy consumption to achieve sustainable growth in the long-run

Regarding policy implications, the causal relationship in Fig. 1 is of growing importance since the direction of causality will narrate policymaking. However, there are complications as the empirical nature of causality is seemingly hard to establish. Inconsistent forms of causality have been found in the previous literature depending on the examined periods, sample of countries, and the econometric approaches applied. If one can prove that energy consumption is causing economic growth, this suggests that policy implications should be arranged to stimulate non-fossil energy and substitute the environmentally harmful non-fossil energy consumption. If, on the other hand, economic growth can be proven to be causing energy consumption, this suggest that regulators should aim to mitigate energy demand.

(8)

pooling. The PMG estimator, proposed by Pesaran et al. (1999), is an intermediate estimator because it involves both pooling and averaging. The intercepts, short-run coefficients, and error variances are allowed to differ across groups while the long-run coefficients are constrained to be the same. The PMG estimator is further motivated in the methodology section.

The remainder of this paper is organized as follows. The following chapter discusses the background. Chapter 3 provides a literature review. Chapter 4 describes the methodology, data, and theoretical framework. Chapter 5 presents the empirical results and chapter 6 provides concluding remarks.

(9)

2. Background

To highlight the global warming issue the Intergovernmental Panel on Climate Change (IPCC) published the 2019 Special Report on Global Warming of 1.5 °C that reveals new information on the risks of global warming. The report shows that unaccustomed actions in every aspect of modern civilization are required to limit global warming to 1.5 °C above pre-industrial levels. The goal of 1.5 °C above pre-industrial levels originates from the 2015 Paris Agreement made in the United Nations Climate Change Conference, where the members of the United Nations countries agreed on the commitment to pursue efforts to limit the global temperature increase (UNFCCC, 2015). The Secretary-General of the United Nations, António Guterres, said in his speech to the UN General Assembly in 2018, ”Climate change is moving faster than we are,”. Indeed, the IPCC report confirms that climate change is already unfolding and negatively impacts ecosystems and the necessities of life all around the world. The report also stresses the new information revealing that climate change does not follow a linear relation to the increase in global temperature, instead, the physical impacts of climate change increase in a non-linear fashion, making the impacts above 1.5 °C even worse than expected (IPCC, 2019).

In the International Energy Agency’s (EIA) report World Energy Outlook 2019 it is clear that the energy sector is by far the largest contributor among the polluters of CO2 emissions that

cause global warming. Both the EIA report (EIA, 2019) and IPCC report (IPCC, 2019) explore what it takes for the energy sector to limit the temperature rise to 1.5 °C above pre-industrial levels and conclude one thing – we need to reach a situation called net-zero. Net-zero is achieved when human-caused CO2 emissions are removed from the atmosphere in a process

(10)

Regarding global environmental sustainability the jury is still out. While scientist coincide about the threats of global warming, consensus on how to accomplish net-zero has yet to be reached. The subject of energy touches everyone as access to energy is a key pillar for economic development, modern infrastructure, industries, homes, leisure, and more. While ensuring that everyone has sufficient access to energy proposes a great challenge, the massive worldwide production of energy damages the environment on which we all depend. This brings up the nexus between continued economic growth and energy consumption. The critical view is that continued economic growth cannot go hand in hand with environmental sustainability since the economic development process enquires a huge amount of both energy- and natural resources, which are accountable for the majority of sources that cause climate change. Especially fossil fuels since they are responsible for a majority of the carbon dioxide emissions (EIA, 2019). If we take the critical road the sure way to accomplish net-zero emissions is to suppress growth.

The optimistic perspective emphasizes that continued economic growth does not have to be conflicting environmental sustainability since the technological progress will eventually solve the human-made emission problems. This perspective tells that the continued economic growth will lead to technological progress that makes way for green and other alternative ways of consumption and production, resulting in medium-term reducing, and in the long term eliminating the harmful climate change. Thus, continued economic growth would be necessary to reach net-zero, a view that is supported by the Environmental Kuznets Curve (EKC).

(11)

3. Literature review

3.1. The energy-GDP nexus and extensions

There is a broad literature examining the causal relationship between economic growth and energy consumption and perhaps an even more recently common subject is the causal relationship of economic growth and CO2 emissions. Bruns and Gross (2012) metastudy of 44

studies and 534 causality tests found that energy and GDP are strongly coupled. However, despite the numerous papers issued a consensus on the direction of causality in the energy-GDP nexus is yet to be reached.

The energy-GDP nexus can be divided into four hypotheses; neutrality hypothesis suggests that there is no causal relationship, growth hypothesis involves causality running from energy to GDP, conservation hypothesis states causality running from GDP to energy and feedback hypothesis is bidirectional causation between energy and GDP. Each of these four hypotheses involves important policy implications that do not coincide, hence, there is an urge to clarify the causal direction. For instance, the growth hypothesis implies that reducing energy consumption will slow down economic growth. While the conservation hypothesis state that reducing energy consumption will not impact economic growth.

(12)

(1988) two-stage procedure that extended the analysis of cointegrated variables and causality, commonly used in error correction models (ECM).

The succeeding generation of literature followed Johansen’s (1991) multivariate ECM approach to testing causality while also controlling for other inputs such as capital, prices, and labour (e.g. Asafu-Adjaye, 2000; Stern, 2000). The Johansen (1991) methodology made analyzing short-run adjustments possible while also allowing for more than two variables in the cointegration relationship. Using this methodology on Korea data Oh and Lee (2004) found no evidence for short-run causality between GDP and energy and unidirectional causality between GDP and energy in the long-run. However, this methodology did not consider country and time-specific effects, Therefore, the literature published from around 2004 commonly corrects this and allows for the country and time-specific effects and examines the causal relationship between GDP and energy with univariate or multivariate panel data analysis.

At this point it is worth mentioning the importance of panel data structure. The use of panel data offers multiple advantages over pure time-series and cross-sectional data as panel datasets controlling individual heterogeneity enables more efficient and robust statistical tests. Panel data greatly increases sample size, enabling time-specific effects and cross-section to be combined, we get more variability and less collinearity among the variables (Baltagi, 2008). Unit root and cointegration tests are usually characterized by low statistical power, a problem that is solved by the higher degrees of freedom in panel datasets (Rapach and Wohar, 2004). However, the previous standard of examining the relationship between economic growth and energy consumption with a bivariate model has been criticized and Saboori and Soleymani (2011) showed that the framework suffers from omitted variable bias. With this knowledge the multivariate panel data tools have dominated the recent literature and extensions of the modeling have been conducted by controlling for physical and human capital. Human capital is often measured by the educational level in the population and physical capital is commonly measured by gross fixed capital formation. However, these studies are more limited as in many cases complete data is hard to find.

(13)

running from GDP to energy consumption. Aside from GDP and energy consumption, the model included gross capital formation. Al-Irani (2006) investigate the GDP-energy relationship in the six oil-exporting countries of the Gulf Cooperation Council and find evidence for unidirectional causality running from GDP to energy consumption, suggesting that energy conservation policies will not harm economic growth. Lee and Chang (2007) use a panel-based vector error correction model (VECM) that include real GDP, energy consumption, capital stock, and labour for 16 Asian countries over the period 1971-2002. They find no evidence of causality in the short-run and unidirectional causality running from energy consumption to economic growth in the long-run. Mahadevan and Asafu-Adjaye (2007) consider a trivariate panel VECM model that considers real GDP, energy consumption, and energy price, based on the argument of energy prices crucial role in energy consumption. They categorize countries according to the net energy trade balance and level of development. Among both the energy exporters and importers in developed countries they found bidirectional causality between economic growth and energy consumption in both the short-run and long-run. While among both energy exporters and importers in developing countries there was bidirectional causality from energy consumption to economic growth only in the short-run.

(14)

1990-2007. They find bidirectional causality between both renewable and non-renewable energy consumption and economic growth in both the short- and long-run.

Payne (2011) extended the renewable energy research by adopting the Toda-Yamamoto (1995) causality test on U.S. annual data from 1949-2007 to investigate the relationship between biomass energy consumption and economic growth. Unidirectional causality running from biomass energy consumption to real GDP is found, suggesting evidence of the growth hypothesis. Menegaki (2011) uses a random-effects model to study the relationship between renewable energy and economic growth in 27 European countries in a multivariate panel also including final energy consumption, greenhouse gas emissions, and level of employment. Only a weak relationship between economic growth and the renewable energy consumption is found, in support of the neutrality hypothesis.

With the addition of disaggregating energy consumption into renewable and non-renewables, the literature on the energy-GDP nexus more commonly also started to include CO2 emissions.

Ozcan (2013) test 12 Middle East countries over the period 1990 to 2008 and found unidirectional causality running from economic growth to energy consumption in the short-run and unidirectional causality running from energy consumption and economic growth to CO2

emissions in the long-run. Caraiani et al. (2015) study emerging European countries over the period 1980 to 2013 and the findings show causality running from GDP to renewable, gas, and coal energy consumption for Hungary, Poland, and Turkey. While there was causality running from coal consumption to GDP in Romania. Ohler and Fetters (2014) perform a panel error correction model to study the causal relationship between economic growth and renewable energy consumption for 20 OECD countries over the period 1990 to 2008. Bidirectional causality between renewable energy and economic growth is found while causality running from biomass, hydroelectricity, waste, and wind energy to GDP is present in the long-run. Hydroelectricity and waste generation exhibit bidirectional causality with GDP and their results suggest that energy conservation policies will stimulate GDP growth if there are decreases in biomass or waste energy combined with an increase in hydroelectricity and wind energy.

(15)

approach to study the causal relationship of economic growth, renewable energy consumption, capital and labour among new EU member countries over the period 1990 to 2009. The empirical results show positive causal effect running from renewable energy consumption to economic growth for Bulgaria, in support of the growth hypothesis. For the Czech Republic there was causality running from economic growth to renewable energy consumption, in support of conservation hypothesis, and neutrality for the other countries. Inglesi-Lotz (2016) base the study on fundamental economics of production and employ the general form of the traditional Cobb-Douglas production function (Cobb and Douglas, 1928) with measures of GDP, renewable energy consumption, gross fixed capital formation, employment and R&D expenditure for each country to examine the relationship of renewable energy consumption and economic growth. When investigating 34 OECD countries the results show that an increase in renewable energy consumption will increase economic growth. Asafu-Adjaye et al. (2016) employ the Pooled Mean Group estimator on panel data of 53 countries over the period 1990 to 2012 to investigate the relationship between economic growth, fossil, and non-fossil fuel consumption. Using proxies for human and physical capital and dividing the countries into four categories: developing importers, developing exporters, developed importers, and developed exporters. Bidirectional causality between fossil energy consumption and economic growth is found in all subsamples except for developing importers. The results of non-fossil energy consumption are more contrasting. For developed importers bidirectional causality between non-fossil energy consumption and economic growth is found in both the short and long-run. For developed exporters there is bidirectional causality only in the long run and no causality in the short-run. Developing exporters show negative causality from non-fossil energy consumption to economic growth in the long-run, developing importers show a positive causal effect from non-fossil energy consumption to economic growth in the long-run.

The literature on economic growth and energy consumption might not have reached a consensus, but important progress has been made and it is evident that further research is required. As the most recent research commonly implements the CO2 emissions in the analysis,

(16)

3.2. The environmental Kuznets curve

The research on EKC began with Grossman and Krueger (1991) followed by Shafik (1994) and was made known widely by the World Bank World Development Report 1992: Development and the Environment. Empirical studies on EKC mainly measure environmental destruction with the quality of the atmosphere, water, waste, and energy use. Among studies that use air quality as a measure of environmental destruction we find Selden and Song (1994), who confirm the inverted-U shape relationship between pollution and economic development using cross-national panel data. Grossman and Krueger (1995) find evidence that environmental quality indicators of air pollution, oxygen regime in river basins and metal contamination in river basins deteriorates in the initial phase of economic growth, followed by a turning point of continuous environmental improvement for almost all environmental indicators, in support of EKC. Cole et al. (1997) use additional environmental indicators and suggest that meaningful EKC exists for local air pollutants, whilst the indicators with global impacts increase monotonically with per capita income.

However, results from the studies that investigate the more global pollutants such as CO2 are

more uncertain about reaching EKC formation. Shafik (1994) examine over 100 countries between 1960 and 1989 and find EKC formation on some indicators but not when only looking at carbon dioxide emissions. He explains this as a classic free-rider problem since there are no major local costs associated with carbon dioxide emissions, there are no incentives to reduce carbon emissions. Holtz-Eakin and Selden (1992) employ panel data on the relationship between CO2 emissions and economic growth and find evidence of EKC formation but with a

very high turning point. They suggest careful attention to be paid to policymaking in support of improving the global distribution of income, as EKC formation is supported in developed economies. Forrest (1995), Stern et al. (1996), and Ekins (1997) investigate the EKC hypothesis and their conclusions coincide, increased GDP per capita increases the environmental indicators: urban air quality, urban sanitation, and access to clean water. However, municipal waste and CO2 emissions continue to rise when GDP per capita increases.

(17)

statistical properties such as serial dependence, stochastic trends, and problems of multicollinearity. Dinda (2004) provided a survey of the EKC in previous literature and found a large variation in the estimated turning point when examining local pollutants, it ranged from $3000 to $10 000 (1985 US dollar at a constant price). With the improvements of econometrical techniques, the literature on EKC started to take cross-sectional and parameter heterogeneity issues into account. Jaunky (2011) applies panel data unit root and cointegration tests to investigate the EKC hypothesis for 36 high-income countries over the period 1980-2005. Using a vector error correction mechanism unidirectional causality running from GDP per capita to CO2 emissions is uncovered. Jaunky (2011) suggests that there is no evidence for EKC,

however, CO2 emissions are stabilizing over time in rich countries.

In a similar approach as Jaunky (2011), Arouri et al. (2012) use panel unit root tests and cointegration techniques to test the EKC hypothesis for 12 Middle-East and North African countries over the period 1981 to 2005. Applying the Pooled Mean Group estimator by Pesaran et al. (1999) results show that energy consumption has a positive significant impact on increasing CO2 emissions in the long-run. The estimated coefficients in most countries satisfy

the EKC hypothesis in the long-run, however, with a large variety in the turning points the evidence in support of EKC is poor. Hamit-Haggar (2012) also took cross-sectional dependence and heterogeneity into account to examine the causal relationship between greenhouse gas emissions, energy consumption, and economic growth. Studying Canadian industrial sectors over the period 1990 to 2007 findings consistent with the EKC hypothesis is evident with an inverted U-shape relationship between greenhouse gas emissions and economic growth. Also, in the long-run there is a weak bidirectional causality running from energy consumption and economic growth to greenhouse gas emissions.

3.3. Contribution of this paper

(18)

research methods could answer the questions and propose vital policy implications for the regulators.

This paper can be defined as complementary to the previous literature. It differs from existing empirical papers by disaggregating total energy consumption into fossil and non-fossil energy consumption in line with Asafu-Adjaye et al. (2016), as opposed to previous research using renewable and non-renewable total energy consumption. This study uses proxies for human and physical capital and divides the global sample into four subsamples based on countries income level. By doing this, it is possible to further link the findings to the EKC. The subcategories represent the observations along the horizontal axis on the EKC since countries of all four World Bank income groups are included. Also, the disaggregation of energy consumption will reflect environmental degradation since fossil energy consumption is the main contributor to environmentally harmful CO2 emissions (EIA, 2019).

This disaggregation of energy can also be informative in the commonly flimsy classification of nuclear energy. The polarized attitudes towards nuclear energy are reflected in national energy policies (Ho, 2016) and while nuclear energy is not classified as renewable, however, nuclear energy stands in line with most renewable energy sources when looking at CO2 emissions

(19)

4. Methodology

4.1. Data description

The data analyzed in this paper is a sample of 74 countries from around the world are chosen for this study for the following reasons. World Bank divides all world countries into 4 subcategories according to their level of income per capita, the subcategories are: low-income, lower-middle income, upper-middle income, and high-income. To capture the different characteristics in all these four subcategories approximately 20 countries from each income group are chosen. However, due to lack of data, from mainly educational level and real gross fixed capital, some subsamples have less than 20 countries.

Annual data on 74 countries (see Appendix, Table A.1) from 1990 to 2007 were obtained from the World Bank, the U.S. Energy Information Administration (EIA), and Our World in Data. Data on Real GDP (constant 2010 USD), real gross capital formation (constant 2010 USD) and total labour force were collected from the World Bank World Development Indicators (WDI) database, while data on fossil and non-fossil energy consumption (measured in trillions of British thermal units) were derived from the EIA. The EIA does not publish data disaggregated into fossil and non-fossil energy consumption, therefore, the two variables were obtained in the same fashion as Asafu-Adjaye et al. (2016). The fossil energy consumption was calculated as the sum of coal, natural gas, and petroleum consumption, while the non-fossil energy consumption was calculated as total energy consumption less fossil energy consumption. Human capital was estimated with the average number of years of total schooling across all educational levels, for the population aged 25 and above, obtained from Our World in Data.

(20)

second-energy consumption) is highest in the low-income group, at a first glance this is somewhat surprising as the low-income countries might not be focusing their resources towards non-fossil energy consumption. However, this could be a result of relatively low electricity use and relatively large use of primitive biofuels in low-income countries. The average years of schooling for the population aged 25 and above are also the lowest in the low-income group with 3.84. As compared to the lower-middle, upper-middle, and high-income groups exhibiting 6.91, 8.85, and 11.78 average years of schooling.

According to the hypothesis in the introduction, the analysis in this paper expects that fossil energy consumption will have a positive significant effect on economic growth among low-income countries, but not vice versa. Further it is expected that non-fossil energy consumption will have a positive significant effect on real GDP among the high-income countries, but not vice versa.

Following this, the theoretical framework will be presented in the next section. After the motivation for the theoretical framework the panel data analysis starts with investigating the

Table 1

Summary statistics for the sampled countries, 1990⎼2017.

Income group

Variable Global sample Low Lower-middle Upper-middle High GDP per capita (2017) $13,586 $718 $2,008 $8,294 $43,197 GDP growth rate (1990-2017) 2.78% 3.82% 4.23% 4.62% 2.10% Physical capital per capita

(2017) $42,292 $1,888 $4,488 $37,842 $136,194 Physical capital growth rate

(1990-2017) 2.53% 3.58% 3.08% 2.48% 2.55% Labour force (average, 2017) 29,020,313 6,489,989 21,366,994 66,687,340 23,029,117 Labour force growth rate

(1990-2017) 1.24% 2.51% 2.41% 0.97% 0.91% Schooling (average, 2017) 7.91 3.84 6.91 8.85 11.78 Fossil energy consumption

(2017, in % of total energy

consumption)

83.30% 75.02% 91.23% 86.61% 79.23%

Non-fossil energy consumption (2017, in % of total energy

consumption)

16.70% 24.98% 8.77% 13.39% 20.77%

(21)

stationarity of the variables with unit root tests. If the variables are non-stationary the next step is to examine the presence of cointegration and its magnitude. Finally, the panel error correction model will be applied to estimate the dynamics and significance of the variables in the short and long-run.

4.2. Theoretical framework

This paper adopts the theoretical frameworks with the fundamental economics of the augmented neoclassical production function in an effort to evaluate the causal relationship between economic growth and energy consumption. The general function takes the following form:

𝑌#$ = 𝑓(𝐾#$, 𝐻#$, 𝐹𝐸#$, 𝑁𝐹𝐸#$) (1)

Where 𝑖 = 1, … , 𝑁 for each country observed over the periods 𝑡 = 1, … , 𝑇; 𝑌_#$ represents real GDP; 𝐾_#$ is the physical capital stock; 𝐻_#$ represents the human capital stock; 𝐹𝐸_#$ is total fossil energy consumption; and 𝑁𝐹𝐸#$ is total non-fossil energy consumption. 𝑌#$ is obtained in its

form directly from the database. 𝐹𝐸_#$ is aggregated from fossil energy consumption sources coal, natural gas, and petroleum. Since EIA does not publish information on non-fossil energy consumption the 𝑁𝐹𝐸_#$ variable is calculated as the total energy consumption less fossil energy consumption. The motivation for disaggregating energy consumption into fossil and the non-fossil energy consumption is that it extends the previous literature on the energy-GDP nexus, mainly focusing on total energy consumption or renewable and non-renewable energy consumption. The vast literature on the subject has yet to reach consensus, therefore, examining new ways of modeling the relationship could contribute to achieving a more unanimous view on the approach we should adopt in the climate issue.

(22)

Physical capital in previous literature is often represented by real gross fixed capital formation, however, this is rather a measure for investments as opposed to the actual stock of physical capital. This paper adopts the same procedure as Asafu-Adjaye et al. (2016) in an attempt to extend previous literature by including a more informative measure of capital stock for each country. This is done using the simplified perpetual inventory method (OECD, 2009), accumulating the real gross fixed capital formation and using a constant depreciation rate. Physical capital stock 𝐾 in year 𝑡 is the investment in year 𝑡 adding the accumulated and depreciated stock of physical capital in year 𝑡 − 1. As follows,

𝐾_$ = 𝐼_$+ (1 − 𝛿)𝐾_$9: (2)

Where 𝛿 represents the depreciation rate of the fixed physical capital, set at 4% according to the aggregate depreciation rate estimate in Berlemann and Wesselhöft (2014). The 4% value is the median from weighted averages obtained from depreciation rates of residential, private nonresidential, and government fixed assets of 22 OECD countries over the period 1980-2010. Following, the benchmark physical stock, 𝐾;, is estimated as the sum of depreciated previous

and current investments. As follows,

𝐾_; = 𝐼_;+ (1 − 𝛿)𝐼_;9:+ (1 − 𝛿)<_𝐼

;9<+ (1 − 𝛿)=𝐼;9=+ ⋯ (3)

After this, the growth rate of the volume of investment is assumed to be equal to the long-run real GDP growth, 𝑔, implying that 𝐼_; = (1 − 𝑔)𝐼_;9:. Now, substituting this into Eq. (3) gives,

𝐾; = 𝐼;@1 + A 1 − 𝛿 1 + 𝑔B + A 1 − 𝛿 1 + 𝑔B < + A1 − 𝛿_{1 + 𝑔}B = … C (4)

It can be shown that the physical capital stock benchmark can be approximated as follows,

𝐾;=

1 + 𝑔

(23)

Hence, given the depreciation rate, the long-run GDP growth rate, and the estimated physical stock benchmark, a measure for each country’s physical capital stock is created. The variable for human capital, 𝐻_#$, was constructed as follows,

𝐻_#$ = 𝐿_#$ℎ_#$ (6)

Where ℎ#$= 𝑒GHIJ. 𝐿#$ represents the total labour force, 𝑠#$ depicts the average years of

schooling for the population aged 25 and above. 𝑟 represents the return to education set at 10%, motivated by Pritchett (2001).

Proxies for human capital and physical capital are included in the model to control for potential omitted variable bias. As mentioned in the literature review (Chapter 3) the previous research on the causal relationship between economic growth and energy consumption is common of bivariate nature, Mankiw et al. (1992) motivate the use of human and physical capital when estimating economic growth in the production function framework. They provide evidence of the importance of human and physical capital when modeling economic growth. Lütkepohl (1982) also addresses the issue of omitted variable bias when testing for causality. Using Canadian income, money, and interest rate he demonstrates that non-causality in a bivariate system can occur from omitted variable bias.

4.3. Panel unit root tests

(24)

unit root test does not control for this. Therefore, the Pesaran (2007) unit root test is adopted in this paper, see Table 2. In the Pesaran (2007) approach an Augmented Dickey-Fuller (ADF) regression is estimated as follows,

∆𝑦_#$ = 𝑐_#+ 𝛼_#𝑦_#$9:+ 𝛽_#𝑦R_$9:+ S 𝛾_#U∆ V UW; 𝑦R_#9U + S 𝛿_#U∆ V UW: 𝑦R_#$9U+ 𝜀_#$, 𝑖 = 1, … , 𝑁 ; 𝑡 = 1, … , 𝑇 (7)

Where 𝑐_# is a pre-determined term, 𝑦R_$ depicts the cross-sectional mean at time t, and p represents the lag order. Now, let 𝑡_#(𝑁, 𝑇) represent the corresponding ratio of 𝛼_#. Following from this, the average of the t-ratios is denoted by cross-sectional augmented IPS (CIPS), as

𝐶𝐼𝑃𝑆(𝑁, 𝑇) =∑^#W:𝑡#(𝑁, 𝑇)

𝑁 (8)

Pesaran (2007) further presents a panel unit roots test that is truncated to deal with problems of moment calculation for mean-deviations of 𝑡-ratios. This truncated 𝑡-ratio is defined as follows,

𝑡_#∗_{(𝑁, 𝑇) = `}𝑡_−6.19,#(𝑁, 𝑇),

2.61,

if − 6.19 < 𝑡_#(𝑁, 𝑇) < 2.61 if 𝑡_#(𝑁, 𝑇) ≤ −6.19,

if 2.61 ≤ 𝑡_#(𝑁, 𝑇), (9)

So, the truncated panel unit root test is,

𝐶𝐼𝑃𝑆∗_{(𝑁, 𝑇) =}∑^#W:𝑡#∗(𝑁, 𝑇)

𝑁 (10)

Corresponding critical values to the limiting distributions of the 𝐶𝐼𝑃𝑆(𝑁, 𝑇), and its truncated counterpart 𝐶𝐼𝑃𝑆∗_{(𝑁, 𝑇) are listed in Pesaran (2007). It should be mentioned that they are}

(25)

4.4. Cross-section dependence

When conducting panel unit root tests, it is important to consider the assumption of cross-sectional independence. Many unit root tests are not robust to this assumption and can lead to spurious results if degrees of error cross-sectional dependence are not considered (Liu, 2013).

A common test for cross-sectional independence (CD) is the Pesaran (2004) test. This paper applies the Pesaran et al. (2008) bias-adjusted LM test of error cross-section independence, which is an extension of the Pesaran (2004) CD test. The bias-adjusted LM test is shown to be consistent even when the Pesaran (2004) CD fails to be (Pesaran et al., 2008).

Following, the Pesaran et al. (2008) bias-adjusted LM test of error cross-section independence will be theoretically explained in the general approach. Let’s start by considering the following panel data model,

𝑦#$ = 𝜷′#𝐗#$+ 𝑢#$, 𝑓𝑜𝑟 𝑖 = 1, … , 𝑁 ; 𝑡 = 1, … , 𝑇 (11)

Where 𝑖 represents the cross-section dimension and 𝑡 the time-series dimension, 𝐗_#$ is a 𝑘 𝑥 1 vector of strictly exogenous regressors with unity on its first row. The coefficients, 𝜷_#, are allowed to vary across 𝑖 and are defined on a compact set. For each 𝑖, 𝑢_#$ ~ 𝐼𝐼𝐷𝑁(0, 𝜎_t#<_{), for}

all 𝑡, even though they could be cross-sectionally correlated. The Pesaran et al. (2008) test is based on the following LM statistic,

𝐿𝑀 = 𝑇 S S 𝜌w_#U< ^ UW#x: ^9: #W: (12)

where 𝜌w_#U is the sample estimate of the pairwise correlation residuals. Precisely,

𝜌w_#U = 𝜌w_U# = ∑ 𝑒#$𝑒U$ y $W: (∑y 𝑒_#$< $W: ):/<{∑y$W:𝑒U$<| :/< (13)

(26)

𝑒_#$ = 𝑦_#$ − 𝜷}′_#𝐗_#$ (14)

where 𝜷}_# is the estimates of 𝜷_# computed from the OLS regression of 𝑦_#$ on 𝐗_#$ for each 𝑖, separately. This is a general LM test that does not require a particular ordering of the cross-section units. The following steps to define the bias-adjusted LM test is available in the Appendix under the “Cross section dependence test”. The Pesaran et al. (2008) bias-adjusted LM test statistic is defined as follows,

LM_€•U = ‚ 2 𝑁(𝑁 − 1) S S (𝑇 − 𝑘)𝜌w_#U< _{− 𝜇} y#U 𝜐_y#U ^ UW#x: ^9: #W: . (20)

Table 2 presents the tests for cross-sectional dependence on the panel data in this paper. The tests are conducted on the global sample and each of the four subsamples. There is a significant p-value in all tests, hence, the null hypothesis of cross-sectional independence is rejected across all samples. In other words, the panel data in this paper have cross-sectional dependence. With this knowledge, it is important to implement a unit root test that controls for cross-sectional dependence. This further motivates the use of the Pesaran (2007) unit root test, presented in the previous subchapter.

Table 2

The bias-adjusted LM test of error cross-section independence by Pesaran, Ullah, and Yamagata (2008).

Countries Test statistic Probability value Global sample 79.82 0.00

(27)

4.5. The Pooled Mean Group (PMG) estimator

When working with panel data there are different approaches, where more common ones are the fixed effects, random effects, and pooled OLS. However, it is well known that such static approaches are not consistent in the presence of non-stationary data. Pesaran and Smith (1995) also show that other common panel approaches, such as the generalized method of moments (GMM), give inconsistent and potentially highly misleading estimates when the coefficients differ across groups, i.e. are not static. This paper tries to avoid these potentially spurious results by implementing a panel autoregressive distributive lag model (ARDL). The ARDL model estimated in this paper is an autoregressive model of order 𝑝 in the dependent variable and order 𝑞 in the explanatory variables. The ARDL (p,q,q,q,q) is stated as follows,

𝑦_#$ = S 𝜆_#U𝑦_#,$9U V UW: + S 𝜹′_#U𝐗_#,$9U+ 𝜇_# + 𝜀_#$ ‰ UWŠ (21)

Where 𝐗_#,$9U is the (𝑛 x k) vector of explanatory variables for each country 𝑖, 𝜇_# depicts the country-specific effects, 𝜆_#U is the coefficient of the lagged dependent variable, and 𝜹′_#U are (𝑛 x k) coefficient vectors. Each variable in this paper is in logarithmic form, enabling the estimated coefficients to represent elasticities. The lag length 𝑝 is chosen by the Schwartz Information Criterion (SIC). The Eq. (21) is then reparametrized into a panel error-correction model of the following ARDL (1,1,1,1,1) form,

∆𝑦_#$ = 𝜙_:,#{𝑦_#,$9:− 𝜃′_:,#𝐗_#,$9:| + 𝜹∗_′

:,#∆𝐗#$+ 𝜇#+ 𝜀#$ (22)

where 𝜙_:,# = −(1 − 𝜆_#) and 𝜃_:,U# =∑’‘“”•I‘

:9•I . Notice, above we have the real GDP (𝑦) as the

independent variable and the remaining explanatory variables (𝑘, ℎ, 𝑓𝑒, 𝑛𝑓𝑒) is in the 𝐗_#$ vector. This paper further investigates the relationship of the variables by modeling each variable as a dependent, and the remaining variables as independent ones. Thus, the remaining equations are,

∆𝑘_#$ = 𝜙_<,#{𝑘_#,$9:− 𝜃′_<,#𝐗_#,$9:| + 𝜹∗_′

(28)

∆ℎ#$ = 𝜙=,#{ℎ#,$9:− 𝜃′=,#𝐗#,$9:| + 𝜹∗′=,#∆𝐗#$+ 𝜇# + 𝜀#$ (24)

∆𝑓𝑒_#$ = 𝜙_–,#{𝑓𝑒_#,$9:− 𝜃′_–,#𝐗_#,$9:| + 𝜹∗_′

–,#∆𝐗#$+ 𝜇#+ 𝜀#$ (25)

∆𝑛𝑓𝑒#$ = 𝜙—,#{𝑛𝑓𝑒#,$9:− 𝜃′—,#𝐗#,$9:| + 𝜹∗′—,#∆𝐗#$+ 𝜇#+ 𝜀#$ (26)

where, in each of the equations above, 𝐗_#$ is an (𝑛 x k) vector consisting of the remaining explanatory variables. This paper adopts the Pooled Mean Group (PMG) approach by Pesaran et al. (1999). the short-run coefficient is represented by the 𝜹∗_{, if this coefficient is significant}

this indicates short-run causality from the specific explanatory variable to the dependent variable. While the run coefficients are the 𝜃s, a significant value here indicates a long-run relationship with the dependent variable. Moreover, 𝜙 represents the error-correction term (ECT) coefficient. This coefficient estimates the speed of adjustment from the dependent variable towards its run equilibrium from a change in the explanatory variables. A long-run relationship is present if the condition of 𝜙 < 0 is fulfilled. If significant this is interpreted as evidence of cointegration between the dependent variable and the explanatory variables. However, if the ECT is positive or equal to zero there is no long-run convergence towards the long-run equilibrium.

(29)

5. Results

5.1. Panel unit root tests & cross-section dependence

Table 3 presents the Pesaran (2007) panel unit root test that allows for cross-sectional dependence. The truncated version of the test is used in this paper, this limits the undue influence of extreme values that can occur when the time series is small. Table 3 shows that most of the variables are non-stationary and integrated of order one. This leads the empirical analysis to the Pesaran et al. (2008) bias-adjusted LM test of error cross-section dependence presented in Table 2. With significant p-values across all samples the null of cross-sectional independence is rejected. Hence, the data exhibits cross-sectional dependence which supports the use of the Pesaran (2007) panel unit root test allowing for cross-sectional dependence.

5.1. Pooled Mean Group estimation results

The summary of the results is listed in Table 4. The following subchapters present the results for each sample in declining order, starting with the high-income group.

Table 3

Pesaran (2007) panel unit root test.

Global sample _High-income

Upper-middle income Lower-middle income Low-income Var. Level First difference Level First difference Level First difference Level First

(30)

5.1.1. Results for the global sample

Table 5 displays the results for the global sample and Eq. (22) displays the elasticity of GDP with respect to the inputs in the production function framework. All variables are significant in the long-run along with the error-correction term (ECT) also fulfilling the condition of being negative. This gives evidence for a long-run relationship, meaning that real GDP and its explanatory variables are cointegrated. A 1% increase in human capital, ceteris paribus, is associated with a 1.27% increase in real GDP; a 1% increase in fossil energy is associated with a 0.32% increase in real GDP; and a 1% increase in non-fossil energy consumption leads to a 0.03% increase in real GDP. However, a 1% increase in physical capital leads to a 0.2% decrease in real GDP, this result is unintuitive and will be further examined in the following subsample results.

In terms of the energy-GDP nexus, there is evidence of bidirectional causality between fossil energy consumption and real GDP in the short-run and the same result is evident for non-fossil energy consumption and real GDP in the long-run. Also, there is evidence of unidirectional

Table 4 Summary of results. Sample Short-run causality Long-run causality Substitutability

Global sample Feedback, growth FE ↔ Y NFE → Y Growth, feedback FE → Y NFE ↔ Y Short-run and Long-run High-income

countries Feedback, growth FE ↔ Y NFE → Y Conservation, growth FE ← Y NFE → Y (negative) Short-run Upper-middle

income countries feedback, conservation FE ↔ Y NFE ← Y Conservation FE ← Y NFE ← Y Short-run and Long-run Lower-middle

income countries Feedback FE ↔ Y NFE ↔ Y

Feedback FE ↔ Y

NFE ↔ Y (effect from NFE

to Y is negative) Short-run and Long-run Low-income countries Growth, neutrality FE → Y NFE ≠ Y Feedback, conservation FE ↔ Y NFE ← Y None

Feedback is bidirectional causality; growth is causality running from energy consumption to GDP;

(31)

causality running from non-fossil energy consumption to real GDP in the short-run and unidirectional causality running from fossil energy consumption to real GDP in the long-run.

These results support two hypotheses, growth, and feedback. The ECT coefficients for real GDP, fossil energy, and non-fossil energy are highly significant and display the speed of adjustment of 0.12, 0.70, and 0.31. This indicates that the deviation from the long-run equilibrium is corrected by approximately 0.12%, 0.7%, and 0.31% in one year. The Hausman test supports the use of the PMG estimator in all cases except for the Eq. (25), where the MG estimator is applied. The negative coefficients of fossil and non-fossil energy in Eq. (25) and Eq. (26) indicates substitutability in both the short and the long-run.

Human and physical capital is included in this paper’s production function to avoid spurious results due to omitted variable bias. The fact that these variables are in most cases significant in the long-run, as opposed to in the short-run, prove their importance when investigating the

Table 5

Results for the global sample. Independent

variables

Dependent variables

lnY (Eq. (22)) lnK (Eq. (23)) lnH (Eq. (24)) lnFE (Eq. (25)) lnNFE (Eq. (26))

Long-run coefficients lnY 2.57*** (0.09) 0.37*** (0.02) 1.63 (1.37) 0.31*** (0.06) lnK ⎼ 0.20*** (0.03) 0.11*** (0.01) ⎼ 0.59 (0.49) ⎼ 0.08** (0.03) lnH 1.27*** (0.04) ⎼ 0.50*** (0.07) 0.37 (0.71) 0.22*** (0.03) lnFE 0.32*** (0.03) 0.05* (0.03) 0.04** (0.02) ⎼ 0.12*** (0.04) lnNFE 0.03*** (0.01) ⎼ 0.05** (0.02) ⎼ 0.02** (0.01) 0.01 (0.25) ECT ⎼ 0.12*** (0.03) ⎼ 0.06*** (0.01) ⎼ 0.08*** (0.02) ⎼ 0.70*** (0.04) ⎼ 0.31*** (0.03) Short-run coefficients ΔlnY ⎼ 0.07** (0.03) 0.04** (0.02) 0.63*** (0.11) 2.01*(1.16) ΔlnK 0.22 (0.30) 0.01 (0.06) ⎼ 0.27 (1.00) 1.29 (0.92) ΔlnH 0.12 (0.08) 0.09 (0.07) ⎼ 0.01 (0.38) ⎼ 2.25 (1.63) ΔlnFE 0.20*** (0.03) 0.01 (0.01) ⎼ 0.00 (0.01) ⎼ 0.80*** (0.30) ΔlnNFE 0.03*** (0.01) 0.00 (0.00) 0.00 (0.01) ⎼ 0.08*** (0.03) Hausman test stat. 2.94 1.18 1.56 10.15 3.31 Hausman test p-value 0.57 0.88 0.82 0.04 0.51

(32)

relationship between economic growth and fossil/non-fossil energy consumption. Notably, an increase in human capital was found to increase non-fossil energy. This positive relationship suggests that a higher level of education will stimulate the adoption of non-fossil energy.

5.1.2. Results for the high-income countries

Table 6 reports the high-income countries. The Hausman test is insignificant across all equations, hence, the PMG estimator is applied in all cases. Concerning the energy-GDP nexus, there is evidence of bidirectional causality between fossil energy consumption and real GDP only in the short-run. In both the short and long-run there is evidence of unidirectional causality running from non-fossil energy consumption to real GDP. However, the long-run the relationship is negative, an increase in non-fossil energy consumption leads to a decrease in real GDP. This suggests that non-fossil energy consumption increases economic growth in the short-run and decreases economic growth in the long-run. These results imply careful investigation of the relationship between these variables and could imply the importance of further development in non-fossil energies that stimulate economic growth even in the long-run. However, the elasticity is seemingly weak, for example, a 1% increase in non-fossil energy, ceteris paribus, leads to a 0.13% decrease in real GDP.

(33)

Regrading human and physical capital in the high-income countries, both forms of capital are negative and statistically significant in Eq. (25). This means that both human and physical capital reduces fossil energy consumption and suggest that they are substitutes in the production function. This indicates that the countries that have more physical capital and stimulate education in the population will require less fossil energy inputs and/or choose non-fossil energy consumption. One of these two forms of capital is also negative and statistically significant in all the subsamples, except for the low-income countries.

The results for the high-income countries are seemingly more complicated, as opposed to the other income groups, this is further confirmed by the insignificant ECT of the production function in Eq. (22). This implies that there is no long-run relationship between the dependent real GDP variable and its explanatory variables. While the ECTs for fossil and non-fossil energy are highly significant, indicating that a deviation from the long-run equilibrium is corrected by approximately 0.28% and 0.34% respectively in one year. High-income countries differ from

Table 6

Results for the high-income countries. Independent

variables

Dependent variables

lnY (Eq. (22)) lnK (Eq. (23)) lnH (Eq. (24)) lnFE (Eq. (25)) lnNFE (Eq. (26))

Long-run coefficients lnY 1.44*** (0.07) 0.46*** (0.07) 1.00*** (0.06) ⎼ 0.06 (0.13) lnK 0.87*** (0.11) ⎼ 0.02 (0.04) ⎼ 0.53*** (0.04) 0.01 (0.05) lnH 0.13** (0.06) 0.48*** (0.11) ⎼ 0.29*** (0.04) 0.33*** (0.12) lnFE 0.21* (0.11) ⎼ 0.39*** (0.06) 0.29*** (0.03) 0.06 (0.09) lnNFE ⎼ 0.13*** (0.04) ⎼ 0.06** (0.03) 0.14*** (0.01) 0.00 (0.03) ECT 0.04 (0.03) ⎼ 0.07*** (0.01) ⎼ 0.17*** (0.05) ⎼ 0.28*** (0.07) ⎼ 0.34*** (0.07) Short-run coefficients ΔlnY ⎼ 0.20*** (0.01) 0.10* (0.05) 0.54*** (0.14) 4.88 (4.23) ΔlnK ⎼ 0.56*** (0.15) ⎼ 0.01 (0.18) ⎼ 0.46* (0.27) 2.15 (2.89) ΔlnH 0.10 (0.06) 0.05 (0.07) 0.04 (0.07) ⎼ 2.81 (3.04) ΔlnFE 0.24*** (0.02) ⎼ 0.05** (0.02) ⎼ 0.01 (0.04) ⎼ 2.06** (1.02) ΔlnNFE 0.06*** (0.02) 0.01* (0.01) 0.03 (0.02) ⎼ 0.11*** (0.04) Hausman test stat. 6.21 0.93 2.08 9.02 2.10 Hausman test p-value 0.18 0.92 0.72 0.06 0.71

(34)

the other income groups here, suggesting that further investigation needs to be made before rightful policy implications can be implemented. However, the unidirectional causality from non-fossil energy to real GDP, in both the short and long-run, can be a signal towards the importance of further development in the non-fossil energy adoption.

5.1.3. Results for the upper-middle income countries

Table 7 outlines the findings for the upper-middle income countries. With the energy-GDP nexus in mind, the upper-middle income group deviates, from the other subsamples, as there is evidence of unidirectional causality running from real GDP to both fossil and non-fossil energy consumption in the long-run. This result is also evident for real GDP and non-fossil energy in the short run. Further, there is bidirectional causality between real GDP and fossil energy consumption in the short-run. The results for the upper-middle income countries are less comprehensive: none of the explanatory variables in the production function Eq. (22) is statistically significant in the long-run and only fossil energy consumption is statistically significant in the short-run. This, while the ECTs for real GDP, fossil and non-fossil energy is statistically significant and the deviation from the long-run equilibrium is corrected by approximately 0.02%, 0.22%, and 0.37% in one year. This is evidence that the variables are cointegrated and there is a long-run relationship, however, significant explanatory variables are less common.

(35)

5.1.4. Results for the lower-middle income countries

Table 8 contains the results for the lower middle-income countries. The coefficients of real GDP, fossil and non-fossil energy consumption are highly significant in Eq. (22), Eq. (23) and Eq. (24). There is evidence of bidirectional causality between real GDP and fossil/non-fossil energy consumption in both the short and the long-run, but with one remark. The elasticity of non-fossil energy with respect to real GDP (⎼1.27%) is substantially negative. Hence, an increase in real GDP leads to a decrease in non-fossil energy consumption. This is interesting since the two middle subsamples of countries exhibit larger annual GDP growth than the low and high-income countries, see Table 1. In addition, Table 1 also shows that the lower-middle income subsample exhibits the highest fossil energy consumption in 2017, as a share of total energy consumption. This result could suggest that countries with higher growth rates are less dependent on the often less technologically demanding non-fossil energy consumption, as

Table 7

Results for the upper-middle income countries. Independent

variables

Dependent variables

lnY (Eq. (22)) lnK (Eq. (23)) lnH (Eq. (24)) lnFE (Eq. (25)) lnNFE (Eq. (26)) Long-run coefficients lnY 2.58*** (0.13) ⎼ 0.84 (1.21) 1.05*** (0.15) 0.54*** (0.13) lnK ⎼ 2.79 (2.02) 0.56 (0.37) ⎼ 0.24*** (0.08) ⎼ 0.15* (0.09) lnH 4.40 (2.69) ⎼ 0.35*** (0.09) 0.05 (0.08) 0.20*** (0.05) lnFE 0.88 (0.74) ⎼ 0.09* (0.04) 0.35 (0.33) ⎼ 0.16** (0.07) lnNFE 3.22* (1.92) ⎼ 0.13*** (0.03) 0.31 (0.19) 0.12*** (0.04) ECT ⎼ 0.02** (0.01) ⎼ 0.07*** (0.02) ⎼ 0.33*** (0.06) ⎼ 0.22*** (0.05) ⎼ 0.37*** (0.08) Short-run coefficients ΔlnY 0.35 (0.03) 0.01 (0.04) 0.76*** (0.18) 1.25** (0.60) ΔlnK ⎼ 0.46* (0.26) 1.11 (1.47) ⎼ 0.14 (0.59) 0.11 (0.75) ΔlnH ⎼ 0.01 (0.08) 0.05 (0.09) ⎼ 0.04 (0.33) ⎼ 0.29 (0.45) ΔlnFE 0.26*** (0.05) 0.03 (0.04) 0.11* (0.06) ⎼ 0.50*** (0.19) ΔlnNFE 0.05 (0.04) 0.00 (0.02) 0.03 (0.03) ⎼ 0.01 (0.07) Hausman test stat. ⎼ 2.31 0.42 21.19 5.43 1.90 Hausman test p-value NA 0.98 0.00 0.25 0.75

(36)

opposed to fossil energy consumption. The results could also disclose the attitude against non-fossil energy among the countries with a higher growth rate.

Further, Eq. (25) and Eq. (26) displays evidence of substitutability in both the short and long-run. The Hausman test is insignificant and the PMG estimator is applied in all cases except for in Eq. (24), where the MG estimator is used. The coefficients for human and physical capital are highly significant in all three equations of interest (Eq. (22); Eq. (23); Eq. (24)). This motivates the inclusion of both forms of capital when investigating the energy-GDP nexus. The elasticity of fossil energy, with respect to human capital (⎼0.96) is statistically significant and negative. This observation is also present in the high-income subsample and this suggests that addition in human capital decreases fossil energy consumption. This could be a result of the countries with higher educated populations being more able to implement new technology and enhance non-fossil energy consumption. The ECTs for real GDP, fossil, and non-fossil energy are highly statistically significant and substantially negative. The real GDP, fossil and non-fossil energy variables are cointegrated, there is a long-run relationship between all of them.

Table 8

Results for the lower middle-income countries. Independent

variables

Dependent variables

lnY (Eq. (22)) lnK (Eq. (23)) lnH (Eq. (24)) lnFE (Eq. (25)) lnNFE (Eq. (26)) Long-run coefficients lnY 2.90*** (0.27) 0.50 (0.70) 1.11*** (0.13) ⎼ 1.27*** (0.20) lnK ⎼ 0.17*** (0.05) ⎼ 1.49* (0.78) 0.26*** (0.05) 0.72*** (0.08) lnH 1.25*** (0.08) ⎼ 1.28*** (0.21) ⎼ 0.96*** (0.20) 0.89*** (0.20) lnFE 0.24*** (0.08) 0.43*** (0.12) 0.77 (0.71) 0.08 (0.11) lnNFE 0.05*** (0.02) ⎼ 0.09 (0.06) ⎼ 0.09 (0.24) ⎼ 0.05** (0.02) ECT ⎼ 0.10*** (0.03) ⎼ 0.06*** (0.01) ⎼ 0.27*** (0.07) ⎼ 0.28*** (0.06) ⎼ 0.38*** (0.06) Short-run coefficients ΔlnY ⎼ 0.07 (0.07) 0.02 (0.03) 1.03*** (0.13) 1.52*** (0.43) ΔlnK ⎼ 0.04 (0.14) ⎼ 0.00 (0.14) 0.52 (0.62) 2.62 (1.71) ΔlnH ⎼ 0.04 (0.14) 0.05 (0.10) ⎼ 0.35 (0.35) ⎼ 0.65 (0.80) ΔlnFE 0.10*** (0.02) 0.01 (0.01) 0.02 (0.02) ⎼ 0.45** (0.20) ΔlnNFE 0.03*** (0.01) 0.01* (0.00) 0.01 (0.01) ⎼ 0.08** (0.03) Hausman test stat. 6.39 7.51 13.00 8.59 2.29 Hausman test p-value 0.17 0.11 0.01 0.07 0.68

(37)

5.1.5. Results for the low-income countries

Table 9 presents the results for low-income countries. With regards to fossil energy consumption and real GDP there is evidence of bidirectional causality in the long-run, in line with the lower middle-income subsample. However, in the short-run there is evidence of unidirectional causality running from fossil energy to real GDP, in support of growth hypothesis. Also, in the short-run, there is no causality between non-fossil energy and real GDP, implying that fossil energy cannot explain real GDP, and real GDP cannot explain non-fossil energy. The long-run relationship between non-non-fossil energy and real GDP shows evidence of unidirectional causality running from real GDP to non-fossil energy. Note, the low-income countries are the only subsample where neutrality hypothesis is supported as non-fossil energy and real GDP show no causal relationship. While fossil energy leads to economic growth in the short-run the non-fossil energy seems less important here.

(38)

Table 9

Results for low-income countries. Independent

variables

Dependent variables

lnY (Eq. (22)) lnK (Eq. (23)) lnH (Eq. (24)) lnFE (Eq. (25)) lnNFE (Eq. (26)) Long-run coefficients lnY 2.26*** (0.25) ⎼ 0.11 (0.72) 0.45*** (0.13) 0.46** (0.23) lnK 0.09** (0.04) 0.26 (0.56) 0.38*** (0.08) ⎼ 0.42*** (0.09) lnH 0.97*** (0.06) 0.41** (0.18) 0.24 (0.15) 1.11*** (0.18) lnFE 0.14*** (0.04) 0.03 (0.09) 0.10 (0.16) 0.00 (0.12) lnNFE 0.03* (0.02) 0.03 (0.03) 0.10* (0.06) 0.13*** (0.03) ECT ⎼ 0.29** (0.12) ⎼ 0.05*** (0.01) ⎼ 0.15** (0.06) ⎼ 0.34*** (0.06) ⎼ 0.22*** (0.04) Short-run coefficients ΔlnY 0.02 (0.04) 0.03* (0.02) 0.26 (0.18) ⎼ 0.03 (0.49) ΔlnK 1.80 (1.52) ⎼ 0.62 (0.50) 0.59 (0.42) 0.27 (1.09) ΔlnH 0.31 (0.34) ⎼ 0.05 (0.15) ⎼ 1.52 (1.00) ⎼ 4.85 (5.57) ΔlnFE 0.08** (0.04) 0.03* (0.02) ⎼ 0.00 (0.00) 0.10 (0.09) ΔlnNFE ⎼ 0.01 (0.01) ⎼ 0.00 (0.01) ⎼ 0.01* (0.00) 0.01 (0.05) Hausman test stat. 2.23 5.56 ⎼ 41.32 6.00 2.37 Hausman test p-value 0.69 0.23 NA 0.20 0.67

(39)

6. Conclusions

While economists have been investigating the causal relationship between economic growth and energy consumption for decades, both theoretical and empirical literature is indecisive and consensus on the subject has yet to emerge. This paper set out to investigate the causal relationship between economic growth, fossil, and non-fossil energy consumption. 74 countries over the period between 1990 to 2017 are included in the global sample of this analysis. The countries were then divided into four subsamples according to their World Bank level of income: high-income, upper-middle income, lower-middle income, and low-income. This paper contributes to the previous literature by disaggregating energy into fossil and non-fossil energy consumption and distinguishing between short and long-run effects in the energy-GDP nexus. This enables policy implications on economies depending on their level of income. The multivariate model included improved measures for human and physical capital by combing level of education and total labour force, and by applying the perpetual inventory method on real gross fixed capital. Along with the energy-GDP nexus the EKC is also incorporated in the theoretical outline, this is relevant due to the characterization of countries according to their level of income and the vast implications of CO2 emissions from fossil energy consumption.

(40)

The results for the high-income countries confirm the hypothesis of this paper and exhibit unidirectional causality running from non-fossil energy consumption to real GDP in both the short and long-run. However, the results for the low-income countries do not fall entirely in line with this paper’s hypothesis of exhibiting unidirectional causality running from fossil energy consumption to real GDP. This result is confirmed in the short-run for the low-income countries but not in the long-run, where there is bidirectional causality between fossil energy consumption and real GDP.

These findings have a number of policy implications. First, when investigating the global sample of countries there was evidence for causality running from fossil energy consumption to real GDP in the long-run, i.e. confirming the growth hypothesis. This result suggests that, regardless of the country’s income level, efforts to reduce fossil energy will decrease economic growth. Regulators should implement policies that enhance the development of non-fossil energy to a level where fossil energy can be reduced without the cost of decreasing economic growth. Moreover, there where contrasts between the countries depending on their income level. The results for the high-income countries suggest that non-fossil energy plays an important role in economic growth for these countries, hence, energy policies that improve the efficiency and stimulate the non-fossil energy will enable the high-income countries to experience economic growth, while also improving the environmental quality by mitigating the harmful CO2 emissions.

Further, the results for the lower-income countries provide support for the feedback hypothesis. This relationship between energy consumption and economic growth suggests that energy policies that reduce energy consumption may impact economic growth. Also, shifts in economic growth may very well be impacting energy consumption. As a result, climate change policies that stimulate and enable non-fossil energy development may have an adverse effect on economic growth, but also reduce the CO2 emissions by substituting the fossil energy, since

(41)

possibilities may arise when the country moves up in their income level, hence, this suggests that stimulating the economic growth of these countries might be the most rightful decision.

Also, an important note is that nuclear energy is included in non-fossil energy in this paper. Hence, nuclear energy may very well be a substantial driver for economic growth as non-fossil energy is found to explain the economic growth in the high-income countries. Today, many countries have polarized attitudes and policies regarding nuclear energy and perhaps a more balanced view is indispensable. Mitigation efforts in human-caused climate change should rely on the scientific voice rather than the populistic one. For example, in 1980 the government in Sweden decided to phase out nuclear energy, later in 2010 the parliament then voted to repeal this policy. However, there is still a discriminating tax against nuclear energy making up a third of the operating cost of nuclear energy, and Sweden’s official goal is still to phase out nuclear energy. The polarized attitudes can be argued as problematic since several scientific papers have quantified the global-scale carbon emissions avoided by historical energy production through nuclear energy plants. In 2013 NASA published a report where historical energy production data revealed that if nuclear energy never existed, the energy it produced would instead, almost certainly, have been supplied by fossil energy consumption. (Kharecha and Hansen, 2013). Perhaps there is a need for more distinct energy policies that distinguish carbon emissions as the primary issue over other less vital sources of environmental degradation.

Regarding the environmental Kuznets curve, this paper is not set out to test the EKC. However, the connection with EKC can be drawn by examining the fossil energy consumption, since this leads to environmental degradation through carbon emissions. The results from this paper can be argued to be in line with the empirical observations in the EKC since the countries with lower income levels consume a higher level of fossil energy (as a share of total energy consumption) than the countries with higher income level.

(42)

(43)

Acknowledgements

The author would like to thank Thomas Broberg at the Institution of Economics on Umeå University, for his valuable advice and immense knowledge contributing greatly to enhance the quality of this paper. Any errors, of course, belong to the author.

THE IMPACT OF FOSSIL AND NON-FOSSIL ENERGY CONSUMPTION ON ECONOMIC GROWTH