Role of renewable energy on industrial output in Canada

Energy Economics 81 (2019) 626–638

Contents lists available at ScienceDirect

Energy Economics journal homepage: www.elsevier.com/locate/eneeco

Role of renewable energy on industrial output in Canada Christoffer Wadström a, Emanuel Wittberg b, Gazi Salah Uddin a, Ranadeva Jayasekera c,⁎ a b c

Department of Management and Engineering, Linköping University, Sweden Institute for Analytical Sociology and Centre for Municipality Studies, Linköping University, Sweden Trinity Business School, Trinity College, Dublin, Ireland

a r t i c l e

i n f o

Article history: Received 24 September 2018 Received in revised form 23 March 2019 Accepted 25 April 2019 Available online 13 May 2019 JEL classification: C22 O49 Q43

a b s t r a c t Several scholars have highlighted the idea that energy consumption in general and consumption of renewable energy (RE) in particular may be a potential driver of economic growth. In this paper, we examine the relationship between RE production and economic activity in Canada between May 1966 and December 2015. By applying quantile causality (Troster, 2018), we adopt a nonlinear approach considering all quantiles of the distribution and analysing monthly data consisting of RE production and the Canadian Industrial Production Index (IPI). We find evidence of a nonlinear relationship in Canada, an important result that widely-used linear models fail to capture. Our main findings imply a unidirectional relationship going from the IPI to RE production, which supports the Conservation hypothesis. The directionality between RE and economic growth is sensitive to the market conditions in Canada. © 2019 Elsevier B.V. All rights reserved.

Keywords: Renewable energy Economic growth Nonlinear Granger causality Canada

1. Introduction This paper addresses the highly debated, double-edged role of energy, where, on the one hand, energy is seen as a critical factor in the provision of prosperity for a modern society, whereas, on the other hand, it is a potential source of great environmental harm. Today there is general agreement regarding the detrimental impact of fossil fuels on the global climate, and yet most of the world's energy demand is met by fossil fuel production. Therefore, mitigation of the ongoing climate change requires large reductions in the production of nonrenewable energy, either by a reduction in energy consumption or by a transformation to a renewable energy (RE) system. However, the former solution, may possibly have drastic consequences for the world economy. Though energy is excluded from the traditional neoclassic growth model, some researchers regard energy services as a vital input for economic growth (e.g., Ayres and Warr, 2009). If this notion is accepted, it would essentially mean that mitigation of climate changes can only be combined with economic growth to a very limited extent in the future world economy, which would indeed paint a bleak outlook. Another, more optimistic, solution is a major shift towards a RE system, ⁎ Corresponding author. E-mail addresses: [email protected] (C. Wadström), [email protected] (E. Wittberg), [email protected] (G.S. Uddin), [email protected] (R. Jayasekera).

https://doi.org/10.1016/j.eneco.2019.04.028 0140-9883/© 2019 Elsevier B.V. All rights reserved.

replacing fossil energy with RE sources such as wind and solar power. The option of a RE system is compelling in theory but far from being a reality. The International Energy Agency (IEA) predicts that non-RE will still be the dominant form of energy by 2040 and, to make matters worse, global energy demand is projected to have increased by 30% in 2040 (International Energy Agency, IEA, 2017b). Clearly, powerful measures will be required to change the course of this development. This potential dilemma related to energy consumption implies that it is important to examine the actual roles of energy in general and RE in particular within the world economy. Extensive research has been conducted regarding both overall energy consumption and consumption of renewable energy. The notion of energy as a vital input for economic growth has garnered some support through empirical evidence [see Omri, 2014 and Tiba and Omri, 2017 for research overviews] but the collected evidence is, nevertheless, inconclusive. Moreover, the potential importance of RE has meant that a growing number of scholars have taken special interest in the impact of RE on the economy and the environment [see Şener et al. (2018) for an overview]. From a theoretical standpoint, Rifkin (2015) claims that distributed production of green energy, such as wind and solar energy, could possibly be a vital component of a highly-efficient economic system with increased productivity and a marginal cost of production close to zero. If this should be the case, then the transition towards a RE system is not only beneficial for mitigating climate changes but is also an opportunity to develop a competitive economy with high productivity, efficiency, and

C. Wadström et al. / Energy Economics 81 (2019) 626–638

627

Energy Structure Canada 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

2005

2006

2007

2008

2009

2010

Fossil Fuels

Nuclear

2011

2012

2013

2014

2015

Renewables

Fig. 1. Energy Structure Canada illustrating composition of energy production 2005 to 2015 in Canada. Data are collected from the US Energy Information Administration.

sustainable growth. However, the empirical evidence on the impact of RE on economic growth is inconclusive due to the different market conditions with respect to different economic structures and compositions of energy sourcing. The IEA also reports that the increasing energy demand is distributed unevenly on a global level. Most developed regions, such as North America and Europe, show a negative trend in primary energy demand. In this global context, Canada is still a major energy consumer and producer, and the actions it takes will have a significant impact on the future development of global CO2-emissions. In 2015, Canada was in the top ten countries with regards to CO2 emissions from fuel combustion (World Bank, 2014), thus making Canada a very important player in the drive to reduce the enviornmental damage caused by fossil fuels. These facts call for further investigation of the relationship

between RE production and economic activity in Canada. This is the subject of this paper. RE in Canada consists mostly of hydropower. Canada is still a party to the Paris Agreement, and, in 2015, the regional leaders agreed on the ‘Responsible Resource Development plan’ to mitigate CO2 emissions (International Energy Agency, IEA, 2015). In May 2015, Canada announced new targets for 2030 to cut greenhouse gas emissions by 30% below the 2005 levels. Fossil fuels like coal and oil have been the dominant energy sources in Canada for a long time. Fig. 1 shows that the share of fossil fuels in the total energy production in Canada has been stable, at around 70%, between 2005 and 2015. In addition, RE as a share of total energy production has also been stable, at around approximately 20%, during this period. Fig. 2, which illustrates Canada's primary energy structure in

Canada primary energy structure in 2015

Solar Energy Production 0.1%

Wind Energy Production 1.2%

Hydroelectric Power Production 17.1%

Nuclear Electric Power Production 5.3%

Biomass Energy Production Coal Production 0.6% 5.7%

Natural Gas (Dry) Production 27.1%

Petrolium and other Liquids 42.9% Fig. 2. Primary Energy Production in Canada by Source. Data source: U.S. Energy Information Administration (EIA) (2018). International Energy Statistics. Coal, Electricity, Petroleum, Hydrocarbon Gas Liquids, Biofuels, Natural Gas, Biomass & Waste.

628

C. Wadström et al. / Energy Economics 81 (2019) 626–638

108

470

106

460

104

450

102

440

100

430 420

98

410

96 94

400 2017

2018

2019

2020

2021

Renewable electricity capacity

2022

Renewable capacity (GW)

Total Renewable generaon (TWh)

Renewable electricity generation forecast in Canada 480

2023

Total renewable electricity generation

Fig. 3. Renewable electricity forecast in Canada. Source: International Renewable Energy Agency, IRENA, 2018.

Biofuel production forecasts Canada (Billion litres) 3 2.5 2 1.5 1 0.5 0 2017

2018

2019

2020

Ethanol

2021

2022

2023

Biodiesel

Fig. 4. Biofuel forecasts Canada. Source: International Renewable Energy Agency, IRENA, 2018.

2015, gives a more detailed insight into the major energy sources in Canada at that time. Natural gas and petroleum liquids accounted for 43 and 27% of total energy production, and hydropower was the dominant source of renewable energy. Although there has been growth in electricity production from other RE sources, such as solar power and wind power, in recent years, the combined share from these sources only accounted for 1.3% during 2015, but as we can see in Fig. 3 the projected share of renewable electricity is increasing. On average, energy demand has increased by 0.8% per year for the past 15 years, and this growth rate is expected to continue under current policy scenarios (IEA-Energy Efficiency Potential in Canada to 2050, 2018). Also, Canada has distinguished itself as the largest consumer of electricity per capita, as the annual per-capita electricity demand was around 15,000 kWh per capita in 2017, which is the highest per capita consumption rate in the world (International Energy Agency, IEA, 2018). On the other hand, Canada's RE capacity is expected to increase by 8% (8 GW) between 2016 and 2023. The largest contribution to this comes from onshore wind power (4.4 GW) followed by hydropower (2.3 GW), and solar PV1 and bioenergy are anticipated to generate 200 MW. Also, Canada's development of advanced biofuel plants constitutes 18% of global novel advanced biofuel projects and in Fig. 4 we can see actual and forecasted biofuel production (IESRenewables, 2018).

1 Solar PV in Canada is projected to increase by 39%, to a total of 4GW, but the projection has been revised because of changes in Solar PV investments and projects in Alberta (IEARenewables, 2018).

It is also worth noting that the different energy sources are distributed unevenly for different uses. Hence, though RE sources are an integrated part of Canada's electricity production, the transportation sector is highly dependent on petroleum. Canada is also a major energy exporter. Most of the fossil fuels and uranium produced in Canada are exported to the U.S. (Natural Resources Canada, 2017). Moreover, 9% of the electricity produced in Canada in 2016 was exported to the U.S. This study on the renewable energy-growth nexus for Canada is also motivated by the fact that previous studies on RE and growth assume a linear relationship between RE and economic growth. Empirical evidence from these studies has generated four hypotheses on the relationship between energy and economic activity. These are the Growth hypothesis, the Conservation hypothesis, the Feedback hypothesis, and the Neutrality hypothesis (Menegakia & Tugcu, 2017). However, these hypotheses are based on the flawed assumption that the relationship between energy and economic growth is constant during good and bad market conditions, which, arguably, is not realistic. Recent advances in econometric modelling point to nonlinear analysis as offering a more precise examination of the relationship between energy and economic activity. In this paper, we apply nonlinear models – quantile regression and quantile causality – to investigate the relationship between RE and economic activity over the business cycle. The primary purpose of this is to investigate whether there is a relationship between economic activity and RE production in Canada and whether these relationships vary over different market conditions. Therefore, we apply Granger causality in quantiles (GCQ), developed by Troster (2018), to examine the possibility for nonlinear-based directionality in the different market conditions. Further, as an innovative approach, we apply quantile regression to estimate the immediate effect between RE production and economic

C. Wadström et al. / Energy Economics 81 (2019) 626–638

activity within any given month, which would provide a clearer understanding of the interplay between Canada's RE production and economic activity. We use the IPI2 as a proxy for economic development and are motivated by the fact that industrial production is a vital component for economic growth in the long-run and tends to correlate with economic development over the business cycle. Thus, we examine explicitly the following research questions: first we investigate whether there are any nonlinear causality effects over the quantiles for the RE and economic activity nexus in Canada. Second, based on our results obtained from a superior econometric technique that takes into account the nonlinear dependence structure, we examine which of the four energy-growth hypotheses would be valid for Canada. Our findings indicate that a long-run steady state may exist between RE and economic activity in Canada. In our results, short-run quantile causality is detected from economic activity to renewable energy. Our study contributes to the field of research in several ways. First, this study considers the monthly time-series data to investigate RE production and Industrial Production in Canada during the sample period of 1966–2015. In contrast to most previous studies, with the exception of Troster et al. (2018b), we apply causality analysis to monthly data, which improves the chance of capturing short-run effects under different states. Second, this study employs nonlinear quantile causality analysis to study the causal relations between the production of RE and economic activity under different market conditions (boom vs recession, or normal). The remainder of the paper is organised as follows. Relevant literature is presented in Section 2. In Section 3, we present the methods used in our study. Section 4 contains descriptive statistics and tests, and Section 5 discusses the study results. Section 6 outlines our conclusions and policy implications. 2. Related literature review Georgescu-Roegen (1971) was among the first to introduce the notion of a causal relationship between energy usage and economic performance, and Kraft and Kraft (1978) were among the first to test this relationship empirically. Since then, many scholars have tried to establish whether such a relationship exists. Although several scholars have supported the idea of energy consumption as an important input for economic growth in the long-run (e.g., Ayres and Warr, 2009; Kümmel et al., 2015), the empirical evidence is ambiguous, and no consensus has been reached regarding this matter. A shift in the energy structure towards distributed RE production seems to be a trend in developed economies (IEA – Renewables, 2017). Though RE has the same function as do other forms of energy, several scholars have highlighted the idea that RE may provide additional social and economic benefits, such as cost mitigation and improved productivity and health (Intergovernmental Panel on Climate Change, IPCC, 2012; Rifkin, 2015). The technical potential of RE is theoretically large due to the unrestricted access to solar power and wind power and the potential for a decentralised energy system with synergy effects (Intergovernmental Panel on Climate Change, IPCC, 2012).3 The relationship between RE and economic development is a wellstudied subject and a highly-discussed topic among scholars and policymakers. The four hypotheses that are at the centre of the energy–growth debate are also generally accepted as a reference point in the discussion regarding RE and economic growth. Energy might have a different effect on growth in developing countries than it does in either developed countries or countries that depend on a particular energy mix. For instance, several studies indicate that the impact of RE 2 IPI also covers other aspects, such as technological progression and innovation in the production industry. 3 In practice, the restricted occurrence of required materials for producing solar cells will be a limitation.

629

might be different from the impact of other energy sources (Aspergis and Payne, 2012; Furuoka, 2017). Moreover, several studies have highlighted the possibility that the causal effect of energy might differ in both the short-run and the long-run (e.g., Salim et al., 2014). The possible causal linkages between energy consumption and economic growth are commonly categorised into four types, depending on direction. Each category has been formulated as a hypothesis that brings important implications for energy policies. The Growth hypothesis implies a unidirectional relationship between energy consumption and economic growth. The Conservation hypothesis implies a unidirectional relationship between economic growth energy consumption. The Feedback hypothesis means a bidirectional relationship, and the Neutrality hypothesis means that there is no causal relationship between energy consumption and economic growth. In this paper, we test for the possibility of a nonlinear relationship between RE and economic activity. If this relationship is, in fact, nonlinear, this may indicate varying relationships in different market conditions, in which case more than one hypothesis may be valid in the energy-growth nexus with regards to renewable energy. A few studies have looked at the relationship between energy and growth in Canada. Most of these support the assertion that consumption of energy leads to economic growth. One example is RodríguezCaballeroa and Ventosa-Santaulària (2017), who found evidence for the Growth hypothesis in Canada. Another example is Katırcıoğlu et al. (2016), who found evidence for a long-run relationship between energy consumption, international trade, and real income in Canada. Also, their Granger causality tests suggested that there is a relationship between energy consumption, international trade, and real income in the long-run for the Canadian economy, which is consistent with the Feedback Hypothesis. Similarly, Tugcu et al. (2012) conducted Granger causality analysis with bootstrap simulation and found the Feedback hypothesis to be valid for non-RE in Canada. Finally, by applying Quantile to Quantile analysis, Shahbaz et al. (2018b) also found weak support for the Feedback hypothesis in Canada. All three studies suggest that energy consumption is an important input for economic growth. In contrast, the empirical evidence regarding a relationship between RE and economic growth in Canada is, first and foremost, in favour of the Neutrality hypothesis. Studies by Chang et al. (2015) and Tugcu et al. (2012) suggest that neutrality hypothesis is evident based on panel Granger causality testing approach. Cai et al. (2018) examined the impact of clean energy, including RE consumption and nuclear power, on economic growth. They found no cointegration between real GDP per capita, clean energy consumption, and CO2 emissions. Nevertheless, their results indicate that clean energy consumption leads to real GDP per capita growth in Canada. Previous studies that use time-series analysis to investigate the renewable energy–economic growth nexus vary in results and directions in the causality. The results of these studies reflect specific conditions in individual countries and are, therefore, easier to interpret in comparison to panel studies. On the other hand, time series often have fewer observations that might aggravate statistical inference. Furthermore, most time-series studies up to this point have assumed a symmetrical and linear relationship between economic activity and renewable energy. As we explain later, this assumption is not necessarily realistic. Among the numerous time-series studies in other countries, there are a few worth mentioning. A 2016 study on the Nordic countries Denmark, Finland, Norway, and Sweden (Irandoust, 2016) used a modified version of the non-Granger causality test. Toda-Yamamoto developed the test in 1995 and used it on time series yearly data ranging from 1975 to 2012. The 2016 study found that technological innovation and economic growth Granger-cause the rise in RE in all countries in the long-run. This result is interesting for our study, since it supports the Conservation hypothesis in light of the way that economic activity is causing RE production. It also implies that technological development leads to increased production of renewable energy.

630

C. Wadström et al. / Energy Economics 81 (2019) 626–638

Studies that employ linear models to examine the relationship between RE and economic growth in the U.S. also tend to favour the Neutrality hypothesis (Chang et al., 2015; Payne, 2009). In contrast, studies with nonlinear models indicate that RE is a vital input for economic growth, at least in the short-run. Dogan and Ozturk (2017) found support for the Growth hypothesis for the relationship between RE and economic growth in the U.S .when applying Gregory and Hansen's (1996) cointegration test, which allows for one structural break. Shahbaz et al. (2016) found support for the Feedback hypothesis between RE and economic growth by applying wavelet analysis. Finally, Troster et al. (2018) examined the relationship between RE and Industrial production in the U.S. by applying a Quantile Autoregression (QAR) model. They found the Feedback hypothesis to be valid for the lowest tail of the distribution and the Growth hypothesis to be valid for the highest tail of the distribution. As stated above, it is evident that previous results regarding the relationships of RE and growth are contradictory at best. One plausible explanation for the inconclusive results of previous studies is that the roles of energy and RE might be different for different economies. The effect of energy consumption may be dependent on several country-specific factors; for example, the level of industrialisation, the composition of industrial production, and the energy sources (e.g., Katırcıoğlu et al., 2016). If this is the case, it motivates inquiries into the specific relationship between growth and energy consumption among different types of economies. The country-specific studies allow us to observe distinctions and significant differences in energy dependence between economies. The different effects stemming from energy consumption cannot be explained fully by country-specific factors alone. For any given country, the results vary based on several factors, including the research method and the sampling period. These diverse results indicate to us that there are some issues regarding high validity and low reliability in these studies. From our point of view, a likely explanation for this inconsistency consists partly in the diverse range of methods used and partly in the lack of robustness of the commonly used methods; for example, using linear models when empirical evidence suggests a nonlinear relationship. After examining previous literature, we conclude that many energy– growth nexus bivariate studies use models based on linear assumptions; for example, the Granger non-causality in mean tests, applied either in time series or in panel vector error correction models. Also, as stressed by Troster (2018), these models rest on a potentially erroneous assumption of a linear and non-asymmetric relationship between energy and economic activity. For energy and economic growth studies in general, there is growing evidence of an asymmetric relationship between energy consumption and economic output in some countries. Studies by Chen et al. (2017) and Hatemi-J and Uddin (2012), among others, have shown that the relationship between energy and economic activity can be asymmetric between positive and negative shocks. Similarly, Bowden and Payne (2009) showed that the relationship between energy and real output are asymmetric over different sectors of the U.S. economy. Though an aggregated analysis did not indicate any causal relationship, the Feedback hypothesis was supported by a disaggregated analysis of commercial and residential energy consumption. Moreover, studies on RE and economic growth also indicate asymmetric causality in some countries (e.g., Destek, 2016; Shahbaz et al., 2016; Alper and Oguz, 2016). An overview of asymmetric and nonlinear relationship studies is presented in Appendix 1. There are several reasons why an asymmetric relationship may exist between RE and economic activity. One reason is that a negative shock in either energy consumption or economic activity often has a more severe impact than does a positive shock. Another probable reason is that economic agents, in general, react more to negative information than to positive information, a fact that is well known in psychological research and behavioural economics (Rozin and Royzman, 2001; Tversky and Kahneman, 1992). Until recently, many asymmetric Granger non-causality tests have followed the methodology developed by Hatemi-J A (2012). The methodology isolates positive and negative energy shocks by applying

Granger non-causality tests on the isolated series, consisting of cumulative positive and negative sums of changes in energy and economic growth. One limitation of this method is that the asymmetric causality test only accounts for differences between positive and negative shocks. For example, the methodology does not consider the asymmetrical effects across the distribution. As emphasised by Troster (2018) and Troster et al. (2018), this form of asymmetry concerns the effects in different quantiles of the distribution. By applying GCQ in a QAR framework on U.S. RE consumption data and the IPI, Troster et al. (2018) showed the existence of bi-directional causality between changes in RE consumption and economic production at the lowest and the highest quantiles of the distribution. This nonlinearity implies that the impact of extreme values is different from that of average values. Besides this, Troster et al. (2018) introduced causality analysis between RE production and economic growth on monthly data by using the IPI as a proxy variable for economic activity. We recognise that this approach is advantageous because it allows for the use of monthly data, which include more observations than annual data. Besides, monthly data can be expected to better capture short-run dynamics between economic activity and energy consumption. Other recent methodological innovations concerning estimation of nonlinear relations between energy and economic activity are the Quantile Autoregressive Distributed lag model (QARDL) applied by Shahbaz et al. (2018a) and Quantile on Quantile analysis conducted by Shahbaz et al. (2018b). These studies estimated nonlinear causality between economic growth and energy consumption which further strengthens the notion of the superiority of nonlinear models over linear models concerned with effects in the mean. In summary, growing empirical evidence implies an asymmetric relationship between energy consumption and economic activity. Methods based on linearity assumptions might, therefore, be inadequate for studies on the energy and growth nexus. Previous studies have also shown the occurrence of nonlinear relationships for RE as well. It is therefore appropriate to further investigate the possibility of asymmetric and nonlinear causality between RE and economic activity. Therefore, we account for nonlinearity by applying GCQ in the QAR framework introduced by Troster (2018). Since the data available for all quantiles are limited, we are concerned only with using a reduced model and examining the directionality between industrial production and RE production. QARDL would have been an alternative model enabling estimation of the exact parameters for the long-run and shortrun relationships between the variables. However, this method requires more observations to be reliable. In this study, we test for nonlinear causal relations between RE and IPI. This approach will contribute to the field of research in several ways. The quality of our dataset makes statistical inference more reliable, because our time series contains between 518 and 596 observations. Second, in contrast to most previous studies, we apply causality analysis on monthly data, which improves the chance of capturing short-run effects in our model. Thirdly, to the best of our knowledge, QAR has been applied only to US data; we extend the collective knowledge by including Canada in our quantile analysis on the causal relations between the production of RE and IPI. Our approach allows us to analyse the asymmetric causal relationship associated with different market conditions. This view allows for different directionality at different quantiles of the distribution. Hence, the relationship between RE and economic activity in one country does not have to be ascribed exclusively to any of the four hypotheses presented. 3. Methodological overview 3.1. Quantile regression We also tested for the possibility of immediate correlation between changes in industrial production and RE production by applying quantile regression. This method allows for nonlinear relations and

C. Wadström et al. / Energy Economics 81 (2019) 626–638

causal relations in conditional quantiles. In comparison to earlier studies, it is a novelty that our quantile regression includes the current change rate of the explanatory variable. This is based on the assumption that the influence of RE on industrial production, and vice versa, is immediate rather than delayed. In comparison, quantile causality, which is introduced later, tests for causality from the lagged change of one variable to the current change in the other variable. While this difference may seem small, it is important to bear in mind that quantile regression shows correlation rather than causality. Therefore, its major role is to help us interpret the sign of potential relationships, and it is mainly a complement to quantile causality analysis. As shown by Portnoy (1991), time series need to be stationary for quantile regression to be valid. To deal with non-stationarity, we differentiated the logarithmic series. This means that all coefficients should be interpreted as elasticities. Hence, the coefficients for a given quantile can be interpreted as the percentage change in the dependent variable caused by one percentage change in the independent variable. Apart from this, the estimated coefficients can be interpreted as in an ordinary OLS. With regard to a given market condition, they estimate how increases in the independent variable affect the dependent variable. The estimated quantile regression models can be found in Appendix 2; these are the quantile autoregressive model and quantile causality. We also tested for Granger causality following the methodology developed by Troster (2018). He proposed a parametric test for GCQ in a quantile autoregressive framework. In contrast to most earlier studies, which applied causality with the assumption of linear dependence in the mean, this method allows for nonlinear causalities and causal relations in conditional quantiles. In contrast to our quantile regression model, our test for quantile causality is based on the assumption that earlier values of one variable have explanatory power for the current change rate in another variable. This approach allows for a distinction between causality affecting the median and the tails of the distribution and does not test the assumption of linear dependence. This allows us to test for the causality between RE production and economic activity in different market conditions. Another advantage of this method is that it is less sensitive to non-normality of the data. The quantile causality test strongly resembles ordinary causality tests in interpretation. In our setting, we include lags of the dependent variable to control for spurious relationships. Nevertheless, a drawback with causality tests is their inability to reveal the sign of the estimated relationship. Therefore, as mentioned previously, we complement our study with quantile regression estimations. To test for Granger causality in quantiles we perform the following tests: n h o
i H△IPI↛△RE : E 1 △REt ≤m I△RE ; θ0 ðτÞ I△RE ; I△IPI 0 t t t ¼ τ; a:s:for all τ∈Τ

ð1Þ

versus n h o
i ≠τ; a:s:for some τ∈Τ H△IPI↛△RE : E 1 △REt ≤m I△RE ; θ0 ðτÞ I△RE ; I△IPI A t t t

ð2Þ and n h o
i : E 1 △IPIt ≤m I△IPI ; θ0 ðτÞ I△RE ; I△IPI H△RE↛△IPI 0 t t t ¼ τ; a:s:for all τ∈Τ

against the alternative hypothesis of nonlinear Granger causality. Further, in accordance with the model specification introduced by Troster et al. (2018), we apply the following test statistics [first proposed in Troster, 2018]4: Z Z ST ∶ ¼

ST ¼

n h o
i ≠τ; a:s:for some τ∈Τ; H△RE↛△IPI : E 1 △IPIt ≤m I△IPI ; θ0 ðτÞ I△RE ; I△IPI A t t t

ð4Þ where m(I△IPI ,θ0(τ)) correctly specifies the true conditional QYτ(· │ IYt ), t for all τ ∈ Τ. Hence, the null hypothesis of linear Granger causality is tested

w

jvt ðω; τÞj2 dFω ðωÞd Fτ ðτÞ;

ð5Þ

 n  1 X ´   ψ∙jWψ∙j; Tn j¼1

ð6Þ

where W is the T x T matrix with elements wt,s = exp [ − 0,5(It − Is)2, and ψ ∙ j denotes the j-th column of ψ. 4. Data and summary statistics In this paper, we analyse the causal relationship between RE production and economic activity in Canada. Our time series consists of monthly data on RE production and IPIs.5 We believe that the IPI variable both corresponds to and contain, the theoretical aspects and arguments presented in our theoretical model. Therefore, we use the IPI not only as a proxy for GDP and economic growth but also as a variable that accounts for other aspects, such as technologic progression and innovation in the production industry. The economic arguments for this are that RE may be a vital component for high productivity (Rifkin, 2015) and economic growth, and that RE production is dependent on technologic advancement and innovation (Johnstone et al., 2010; Verbruggen et al., 2010). Hence, we can expect IPI to be a valid measure of economic activity in the sectors of the economy that may be expected to be influenced by renewable energy. Moreover, though there may be an endogenous relationship between IPI and RE production, our approach is robust to this problem, since we mainly apply reduced models. We focus on overall directionality rather than exact interpretation of coefficients. Canadian RE data were collected from Statistics Canada.6 We removed the seasonal dependence from the RE data with X12-ARIMA. IPI data were collected from the OECD database. These series are adjusted with X11-ARIMA and contain no seasonal effects. Our time series consists of 596 observations for Canada covering 1966–2015. The Canadian series comprises hydroelectric power, tidal power, wind power, and solar power generation and excludes biomass and waste energy production, because there is no way for us to distinguish renewable biomass and waste power generation from non-RE production. For our analysis, this is a limited problem, because the total share of the biomass and waste production in Canada is small (U.S. Energy Information Administration (EIA), 2018). See Troster et al. (2018b) for an extensive derivation of the QAR model. OECD gives the following definition of IPI: “Industrial production refers to the output of industrial establishments and covers sectors such as mining, manufacturing, electricity, gas and steam and air-conditioning. This indicator is measured in an index based on a reference period that expresses change in the volume of production output.” Source: OECD database. (2018-03-21) “Industrial production”. Available at: https://data.oecd.org/ industry/industrial-production.htm 6 The Canadian series consists of two combined series, one from 1950 to 2007 (Table 127-001), where the first series was terminated, and the measurement continued in another series (Table 127-002). The difference is that, prior to January 2008, hydro included wind and tidal generation, whereas, from January 2008 onwards, wind and tidal generation are reported separately. Hence, we had to include them in the aggregated series. 5

versus

τ

where Fω(·) is the conditional distribution function of a d-variate standard normal vector, Fτ(·) follows a uniform discrete distribution over a grid of Τ in n equally-spaced points,Τn = {τj}nj=1, and the vector of weights of ω ∈ ℝd is drawn from a standard normal distribution. The test statistic in Eq. (5) can be estimated using its sample analog. Let ψ be a T x n matrix with elements ψi,j = Ψτj(Yi − m(IYi ,θT(τj))) and Ψτj(·) be the function Ψτj(ε) ∶ = 1(ε ≤ 0) − τj. Then, the following test statistic is applied:

4

ð3Þ

631

632

C. Wadström et al. / Energy Economics 81 (2019) 626–638

Renewable energy production and Industrial Production Index LOG RE CANADA

LOG IPI CANADA

10.8

4.8 4.6

10.4 4.4 10.0

4.2 4.0

9.6 3.8 9.2

3.6 1970

1980

1990

2000

2010

1970

1980

DLOG RE CANADA

1990

2000

2010

DLOG IPI CANADA

.20

.04

.16 .12

.02

.08 .04

.00

.00 -.04

-.02

-.08 -.12

-.04 1970

1980

1990

2000

2010

1970

1980

1990

2000

2010

Fig. 5. Renewable energy production and Industrial Production Index. Graphs of Canadian renewable series and Industrial Production Index in level and in first difference indicate that our series is non-stationary in level and stationary in first difference. Renewable energy is in MWh and all series are taken in natural logarithm. DLog = ln(Pt) - ln(Pt-1).

Fig. 5 illustrates graphs of IPI and RE production in Canada (1966 onwards). The series on RE production are characterised by larger variance. A break with increased variance for the RE series in Canada is notable around 2010, which can be seen from observing both the series in level and in first difference. This may possibly also be associated with a structural break in the dependence between our variables and is, thus, an argument for our nonlinear approach. When conducting quantile analysis, the large post-2010 deviations for the series in first difference will be included in the largest and the highest quantiles and, therefore, be separated from the effect of the other observations. Moreover, several large shocks can be seen for the indices on Industrial Production. Three of the shifts in the series can be identified, respectively, as results of the early 1980s' recession, the dot-com bubble from the early 2000s, and the Great Recession of 2008. This makes sense, since indices on industrial production are closely related to GDP. The Canadian RE series is characterised by a few shifts, although the variance of the series is increasing since 2010. The series in level appears to be non-stationary, whereas the series in first difference seems

to be stationary around a stable mean. Therefore, we can postulate that the series are integrated of order one I(1). Descriptive statistics in Table 1 are in natural logarithm. Our time series are not normally distributed, as indicated by the Jarque–Bera test, and this gives us reason to suspect asymmetry and non-normality in our data, which could result in less-efficient estimates. Due to this suspicion, we also conducted a BDS-test, which is presented in Table 2. With this post-estimation test, we investigated whether there is evidence of nonlinear aspects in our model. The linear structure is removed by detrending the series by first-difference –ARMA(m,n)-GARCH (p, q) types of process. The results suggest clearly that our data are nonlinear, and this substantiates the use of nonlinear methods in our analysis. Therefore, nonlinear models, like quantile autoregressive models, are a preferable analytical approach compared to ordinary linear models. It is also important to verify the integration order of our series. Results from three different unit root tests are presented in Table 3. As we can infer from examining the graphs in Fig. 5, these tests reveal all series to be stationary in first difference and integrated of order one.

Table 1 Descriptive statistics.

RECan IPICan

Mean

Median

Maximum

Minimum

Std. dev.

Skewness

Kurtosis

Jarque-Bera

Obs.

10.04 4.34

10.15 4.37

10.51 4.74

9.25 3.63

0.30 0.31

−1.07 −0.43

3.11 2.07

114.48*** 40.04***

596 596

Notes: Table over descriptive statistics indicates asymmetry and non-normality in the data. Industrial production index source: OECD database. Renewable energy consumption in Canada source: Statistics Canada and authors' calculations, and the period covered is 1966.05–2015.12. The notations *, ** and *** indicate rejection of the null hypothesis at the 10%, 5% and 1% significance levels.

C. Wadström et al. / Energy Economics 81 (2019) 626–638

633

Table 2 BDS independence test. Fist-difference detrending

REcan

IPIcan

AR-process

GARCH process

m

ε = 0.5

ε = 0.7

ε = 0.9

ε = 0.5

ε = 0.7

ε = 0.9

ε = 0.5

ε = 0.7

ε = 0.9

2 3 4 2 3 4

0,021*** 0,023*** 0,020*** 0,007** 0,009*** 0,009***

0,031*** 0,048*** 0,057*** 0,009*** 0,017*** 0,024***

0,021*** 0,042*** 0,065*** 0,004** 0,009*** 0,017***

0,021*** 0,023*** 0,020*** 0,007** 0,009*** 0,009***

0,032*** 0,049*** 0,058*** 0,009*** 0,017*** 0,024***

0,022*** 0,044*** 0,067*** 0,004** 0,010*** 0,017***

0,021*** 0,023*** 0,020*** 0,007** 0,009*** 0,009***

0,031*** 0,048*** 0,057*** 0,009*** 0,017*** 0,024***

0,021*** 0,042*** 0,065*** 0,004*** 0,009*** 0,017***

Notes: Table presents BDS statistics and the notations. *, ** and *** indicate rejections of the null hypothesis at 10%, 5% and 1% significance levels. All series are in log diff except FSI which is first-difference only and in addition to the first-difference, the series are also detrended by estimating AR(1) and GARCH(0.1) models and the residual series are then tested for iid. ε is the distance for testing proximity of the data points and is calculated as a fraction of pairs with three values 0.5, 0.7 and 0.9, and m is the number of consecutive data points to include in the set. The test includes 596 observations and the p-values are also bootstrapped with 1000 iterations.

Table 3 Stationarity.

I(0)

REcan IPIcan REcan

I(1) IPIcan

Augmented Dickey-Fuller test

Philips-Perron test

ADF(δ)

ADF(φ)

PP(δ)

PP(φ)

KPSS(δ)

Kwiatkowski–Phillips–Schmidt–Shin test KPSS(φ)

−3.101(12) −0.908(12) −10.011(11) ***

−2.138(12) −2.038(15) −10.339(11) ***

0.692(18)*** 0.345(17)***

0.399(20)*

0.041(22)

−6.882(11)***

−2.516(14) −1.946(13) −32.549(21) *** −24.722(13) ***

2.753(18)*** 2.773(17)***

−5.467(14)***

−2.627(20) −1.256(13) −31.080(20) *** −24.731(13) ***

0.096(13)

0.048(13)

Notes: All series are stationary in first difference. ADF(δ). PP(δ) and KPSS(δ) represent test models with intercept. ADF(φ). PP(φ) and KPSS(φ) represent test models with intercept and trend. The notations *, ** and *** indicate the rejection of the null hypothesis at 10%, 5% and 1% significance levels. For ADF and PP the null hypothesis is no unit root process, while the null hypothesis in the KPSS is the unit root process. Numeric values in parentheses indicate selected lag length in ADF and bandwidth in PP and KPSS tests. The tests have been performed with automatic lag selection, with a maximum lag of 14 and AIC as information criteria. All tests indicate stationary series in first difference.

Consequently, we can proceed to examine the series for cointegration. Moreover, differentiation of the series is required for ordinary Granger causality and QAR tests to be valid. However, the results from ADF, PP, and KPSS could be biased in the presence of structural breaks. Consequently, we cross-checked our results with Zivot–Andrews unit root tests, which account for one structural break in intercept and trend. The results from the Zivot–Andrews tests, presented in Table 4, confirm that all series are stationary in first difference and integrated of the first order. Finally, Bai–Perron tests were conducted to look for structural breaks in our data, as shown in Table 5. Three or four statistically significant break points were found for each time series, and the series have several breaks in common. The most recent breaks can be identified as the Great Recession (for the breaks 2008–2009) and the dot-com bubble (for the breaks 2000–2001). These breaks further motivate our nonlinear approach, since they may have resulted in structural shifts, which would make linear models less reliable. We use Dynamic Conditional Correlation DCC-GARCH (1,1) between RE and IP over the investigated sample period.7 Fig. 6 shows the dynamic conditional correlation between RE and IPI and four correlation structure regimes. The first regime is from 1966 to 1973, with relatively small variation {0.1 to 0.16}. The second regime, from 1973 to approximately 2000, is characterised by larger variation {0.16 to 0.31}, and it is followed by a period extending from 2000 to 2010, where the variation is roughly as large as in the previous period but at a lower level {0.09 to 0.28}. The last period, ranging from 2010 to the end of 2015, has both lower levels of correlation {0.17 to 0.06} and lower variation in correlation. We can also see that, from approximately 1966 to 1973, there is a more or less stable correlation structure. However, the following historic correlation structure is more volatile, with peaks in 1977, 1986, and 1992 and clear dips in conditional correlation around 1979, 1990, 2005, 2010, and 2014. It is worth noticing that a rise in 7

DCC was introduced by Engle (2002).

correlation coincides with major economic crises. For example, 1970 energy crisis with the OPEC oil price shock in 1973; the 1979 energy crisis, with the black Monday in 1987; The early 1990s recession followed by the dot -com bubble 2000–2002 and the build up to the financial crisis in 2007. 5. Results and discussion In Table 6, we present the results from the quantile regression for Canada. The effects running from economic activity (IPI) to renewables (RE) are strong β = {0.386 in τ = 0.85 to 0.786 in τ = 0.1} and significant over most quantiles τ = {0.1–0.9} at the 1% significance level. Effects running from RE to IPI are also significant, at the 1% and at the 5% significance levels, respectively, in the same quantiles τ = {0.1– 0.9}; but the effects are much weaker than are the effects running from IPI to renewables. Our interpretation of this is that economic activity has a larger impact on RE production than RE production has on economic activity. Based on the quantile regression alone, and despite the small effects, these results support the Feedback hypothesis, since energy production is a consequence of economic activity, and vice versa. Interestingly, these results for RE resemble the results for overall energy consumption previously estimated by Katırcıoğlu et al. (2016) and Shahbaz et al. (2018b). Hence, there is weak support for a symbiotic relationship between energy production and economic activity in the short-run in Canada, irrespective of the energy source. However, caution should be exercised in these interpretations, since quantile regression shows correlation rather than true causality. As the share of renewables in the energy mix in Canada is fairly constant, the increases in RE are an indication that economic activity is driving RE, rather than a substitution effect. Table 7 reports the p-values of the GCQ test. Considering all the quantiles, there is Granger causality ranging from IPI to RE at the 1% significance level in all models. If we consider only the middle quantiles in the test τ = {0.45–0.55}, it shows no causal relationship running from IPI to RE. However, we find evidence of Granger causality at the upper

634

C. Wadström et al. / Energy Economics 81 (2019) 626–638

Table 4 Zivot-Andrews structural break test. ZA(δ)

Time break

REcan

2.99(12)

IPIcan

−4.63(15)

REcan

10.5(11)***

IPIcan

−6.00(14) ***

I(0)

I(1)

1977 M03 2009 M03 2008 M08 2008 M01

ZA(φ)

−3.98 (12) 4.59(15) 10.58(11)*** −6.23(14) ***

Time break 1983 M12 2000 M05 1998 M08 1996 M02

Notes: Zivot-Andrews Structural Break test indicates that all series is I(1). ZA(δ) indicate model with intercept and ZA(φ) indicate model with intercept and trend. The null hypothesis is no unit root process. The test is performed with a maximum number of lags of 20. The notations *, ** and *** indicate rejection of the null hypothesis at the 10%, 5% and 1% significance levels. Figures in parentheses indicate the selected lag length.

and lower tails of the distribution τ = {0.15–0.35 and 0.65–0.95} at the 1% significance level. These results are also robust for model specifications 1–3 of the auto-regressive model. This indicates that large positive or negative changes in economic activity lead to large or extreme changes in RE production. The results show that there is a disproportionately high demand for RE corresponding to periods of high economic activity. One plausible reason for this could be that Canada's long-run dependency on hydropower has perhaps caused the industry to become reliant on production methods utilising hydro-electric power. Hence, when the demand for energy due to high economic activity crosses beyond a threshold contingent on the optimal available level of hydropower, then this excess would be supplied by a compatible source of energy in the form of RE, thus leading to a surge in the demand for RE during high periods of economic activity. The fact that IPIs do not seem to cause increases in RE in the absolute extreme upper quantile might indicate the occurrence of network congestion effects originating from renewable generation (Ardian et al., 2018). When analysing the causality running from RE to IPI, and considering all quantiles, there is Granger causality in model 1 at the 5% significance level, but there are variations of causality in the distribution. There are a few significant quantiles in the lower tail of the distribution τ = {0.15–0.2} at the 1% significance level, and from one to a few in the upper tail τ = {0.9} depending on model specification, but only at the 5% or 10% significance levels. These results suggest that, in most cases, some of the lowest negative values of RE production Granger-cause negative variations in economic activity. The results are also robust for model specifications 1–3 of the auto-regressive model. In Table 8, we present our main assessment of the quantile analysis for Canada, which is that there exists a unidirectional relationship running from previous changes in IPI to current changes in RE for the lower and the higher quantiles. This contradicts the findings by Chang et al. (2015), which support the Neutrality hypothesis. One possible explanation for our results is that, historically, Canada has been dependent on hydropower, which just happens to be a flexible, efficient, and renewable source of energy. Hydropower generates electricity energy sources directly and, by relying on hydropower, Canada has a substitute for non-RE sources and can effectively adjust its electricity production to compensate for adverse

Table 5 Structural breaks.

RECan IPICan

Sequential Repartition Sequential Repartition

Break1

Break2

Break3

Break4

1977M10 1972M10 1984M05 1971M08

1992M10 1985M10 1997M01 1984M05

2004M11 2004M11 1971M08 1997M05

2008M01 2008M01

Note: Structural breaks identified by Bai-Perron test.

Fig. 6. Dynamic Conditional Correlation between Renewable energy and Industrial output.

market conditions and changes in energy demand. In contrast, the quantile regression indicates a feedback relationship. Both tests combined indicate a weak feedback relationship in the lower quantiles, where the quantile regression exhibits modest effects running from RE to IPI, and causality is supported only in one of the lower quantiles. Potentially, this indicates that RE in Canada leads to changes in economic activity during bad, although not extreme, market conditions. One theoretical explanation for the cointegrating relationship in Canada is that hydropower is well-established and consitutes a relatively large share of the energy mix (policy, environmental, and technological aspects). Moreover, its price is competitive price relative to other non-RE sources (cost-to-price relation). Therefore, we consider hydropower to have surpassed some of the potential barriers related to RE and to be an integrated long-term part of the Canadian energy mix. Though indicating causality, the Granger tests do not give any indications of how the variables affect each other, i.e., either positively or negatively. Therefore, the causality tests are more comprehensible

Table 6 Quantile regression. Δ IPICAN to Δ RECAN

Δ RECAN to Δ IPICAN

Τ

IΔREt = 1

IΔREt = 2

IΔREt = 3

IΔIPIt = 1

IΔIPIt = 2

IΔIPIt = 3

0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95

0.411 0.786*** 0.679*** 0.653*** 0.524*** 0.491*** 0.454*** 0.508*** 0.517*** 0.497*** 0.462*** 0.435*** 0.448*** 0.455*** 0.501*** 0.397*** 0.386** 0.179 0.268

0.464* 0.859*** 0.668*** 0.628*** 0.541*** 0.496*** 0.511*** 0.512*** 0.496*** 0.492*** 0.474*** 0.417*** 0.493*** 0.435*** 0.439*** 0.453*** 0.469*** 0.274 0.144

0.430 0.749*** 0.690*** 0.661*** 0.496*** 0.461*** 0.518*** 0.520*** 0.514*** 0.511*** 0.456*** 0.420*** 0.485*** 0.443*** 0.453*** 0.454*** 0.463*** 0.252 0.182

0.036 0.068*** 0.069*** 0.040** 0.047** 0.047** 0.052** 0.050** 0.060*** 0.060*** 0.054*** 0.056*** 0.054*** 0.060*** 0.057** 0.056** 0.053** 0.051** 0.083***

0.053 0.069*** 0.060*** 0.049*** 0.036** 0.045** 0.042** 0.061*** 0.068*** 0.060*** 0.052*** 0.052*** 0.058*** 0.055** 0.058** 0.057** 0.056** 0.050** 0.087***

0.066*** 0.0746*** 0.0564*** 0.049*** 0.057*** 0.052*** 0.067*** 0.060*** 0.063*** 0.066*** 0.061*** 0.047*** 0.040*** 0.051*** 0.047** 0.048* 0.039* 0.054** 0.082***

Notes: Table shows quantile regression according to Eqs. (1)–(6) (Appendix 2). There are strong and significant effects running from IPI to RE and significant but weaker effects running from RE to IPI over almost all quantiles. The table shows correlation between renewable energy production and Industrial Production Index over almost all quantiles. All series are taken in natural logarithm and then in first difference and so the coefficient should be interpreted as elasticities. Intercepts was included in the regressions. The Huber Sandwich method is used as the estimation method when computing covariances. Bandwidth method: Hall-Sheather. The notations *, ** and *** indicate rejection of the null hypothesis at the 10%, 5% and 1% significance levels.

C. Wadström et al. / Energy Economics 81 (2019) 626–638 Table 7 Quantile Causality test.

635

Table 9 Granger quantile causality test at the Volatility level.

ΔIPICAN to ΔRECAN IΔREt = 1

IΔREt = 2

IΔREt = 3

IΔIPIt = 1

IΔIPIt = 2

IΔIPIt = 3

0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 [0.05–0.95]

0.240 0.443 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.058* 0.181 0.511 0.555 0.072* 0.009*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002***

0.679 0.413 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.109 0.353 0.808 0.025** 0.006*** 0.006*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002***

0.704 0.428 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.028** 0.325 0.647 0.036** 0.004*** 0.004*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002***

0.019** 0.366 0.004*** 0.002*** 0.017** 0.094* 0.162 0.302 0.719 0.958 0.723 0.666 0.491 0.543 0.534 0.704 0.330 0.030** 0.191 0.038**

0.136 0.413 0.036** 0.055* 0.058* 0.472 0.236 0.317 0.379 0.949 0.881 0.623 0.270 0.792 0.821 0.615 0.072* 0.025** 0.360 0.325

0.296 1.000 0.387 0.002*** 0.021** 0.638 0.604 1.000 0.862 0.583 0.706 0.606 0.240 0.051* 0.815 0.791 0.242 0.421 1.000 0.377

Notes: This table presents quantile causality with the subsampling p-values of the ST - test in Eq. (6). The results clearly indicate that industrial production leads renewable energy production during times of economic shocks. In addition, there are weak indications of a feedback effect in a few of the low quantiles. ΔIPI is the log-difference of industrial production index. ΔRE is the log-difference of renewable energy production. IΔREt and IΔIPIt are the number of lags of the dependant variables (RE and IPI) under the null hypothesis: No Granger causality in Eq. (1) and Eq. (3). The subsample size is b = 64 for our sample of T = 592 observations. The notations *, ** and *** indicate rejections of the null-hypothesis at 10%, 5% and 1% significance levels.

when analysed in combination with the quantile regressions. From the summary overview, we can see that Canadian economic activity seems to increase RE production in the upper and the lower tails. Though the effects running from Canadian renewable production to economic activity are weaker, they seem to have a positive effect on economic activity in the lower and upper quantiles. To us, this indicates that Canada adheres at least to the Conservation hypothesis, or possibly the Feedback hypothesis. This implies that Canada is dependent on RE, and this dependence has led to a conectedness between RE and economic activity. Canada has relied on hydropower for over half a century and, therefore, already has all the required infrastructure in place. Hence, the marginal costs of production are relatively low. A theoretical explanation for the Conservation hypothesis, which is consistent with the idea that energy services

Table 8 Summation of our study. Net consumption 1

Negative

Share renewables 2

19% IPICAN

RECAN

†††(+) †††(+) †††(+)

†(+) †††(+) †(+)

†††⟹ ⇏ †††⟹

⟸ ⇍ ⟸

3

Quantile regression Lower tail τ = {0.05 to 0.35} Middle τ = {0.40 to 0.60} Upper tail τ = {0.65 to 0.95} Quantile Granger-causality 4 Lower tail τ = {0.05 to 0.35} Middle τ = {0.40 to 0.60} Upper tail τ = {0.65 to 0.95}

ΔVolatility IPICAN to ΔVolatility RECAN

ΔVolatility RECAN to ΔVolatility IPICAN

τ

IΔREt = 1

IΔREt = 2

IΔREt = 3

IΔIPIt = 1

IΔIPIt = 2

IΔIPIt = 3

0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 [0.05–0.95]

0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.820 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.308 0.030** 0.002***

0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.762 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.794 0.002*** 0.002***

0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.756 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.766 0.017** 0.002***

0.002*** 0.002*** 0.002*** 0.002*** 0.004*** 0.023** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.059 0.002*** 0.002***

0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.057* 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.042 0.002*** 0.002***

0.002*** 0.002*** 0.002*** 0.002*** 0.006*** 0.096* 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** 0.023 0.002*** 0.002***

ΔRECAN to ΔIPICAN

Τ

1) Net consumption = Production - consumption. 2) Average of renewable energy production during the period 2005–2015. 3) + or − indicate sign of coefficient. 4) † Notations indicate the number of significant quantiles given robustness in each model (1 to 3): † = at least two 1% significant quantile in all models, †† = at least three 1% significant quantiles in all models, ††† = at least four 1% significant quantiles in all models. An arrow without notations means that there are two or more quantiles with significance levels of 5%. The directions of the arrows indicate causality direction. Bolded (+) signs indicate relatively large coefficients, while small signs indicate small coefficients.

Notes: This table presents quantile causality with the subsampling p-values of the ST - test in Eq. (6). ΔVolatility IPI is the conditional variance in IPI and ΔVolatility RE is the conditional variance in RE. both variables are stationary in level. IΔREt and IΔIPIt are the number of lags of the dependent variable (Volatility RE and Volatility IPI) under the null-hypothesis: No Granger causality in Eq. (1) and Eq. (3). The subsample size is b = 64 for our sample of T = 591 observations. The notations *. ** and *** indicate rejections of the null-hypothesis at 10%. 5% and 1% significance level.

represent a vital input for economic growth, is that economic growth may be driven by increased energy efficiency rather than by increased energy production. Moreover, Ardian et al. (2018) showed that the probability of congestion increases when the production of renewable electricity increases in electricity-exporting regions. From this perspective, it is worth noting that Canada is a net exporter of electricity. Congestion and bottlenecks in the grid system may be a potential technical barrier for RE and may explain why the impact of RE on economic growth in Canada has been insignificant previously. To check the robustness, we applied the quantile causality approach after controlling for conditional variances to detect directionality. Although both DCC and quantile causality methods can detect lead-lag causal relationships and correlation between time-series, there is an important difference. The quantile causality approach is not conditionally dependent on the second moment to derive the aforementioned linkages. Specifically, the DCC model is designed to examine causal dynamics and to capture time varying via modelling conditional volatility and correlation. In Table 9, we see the results from when we test for Granger Quantile causality in the second central moment, where we find a bi-directional relationship between the conditional variance of IPI and the conditional variance of RE, considering all the quantiles and all models. The result differs from the Granger causality test we performed in the first central moment. [Insert a little bit more details and analysis] When looking at the lower tail of the distribution τ = {0.05–0.35} we find evidence of Granger causality at the 1% significant level in all quantiles but τ = {0.30} and in auto-regressive models 1–3. Alas, there is weaker evidence, at 5% and 10% significance levels, of Granger causality running from CondVarRE to CondVarIPI in quantiles τ = {0.30}. Considering the middle quantiles τ = {0.40–0.60} we find evidence of Granger causality in all quantile and auto-regressive models. In the upper tail of the distribution τ = {0.65–0.95} we also find a bi-directional relationship between the conditional variance of IPI and the conditional variance of RE at the 1% significance level in quantiles τ = {0.65–0.85} and in models

636

C. Wadström et al. / Energy Economics 81 (2019) 626–638

1–3. In the extreme quantiles, there is either no indication, τ = {0.90}, or weak evidence of a relationship, τ = {0.95}, at 1% to 5% significance levels in the different models for a relationship. The Granger quantile causality tests of second order moments indicate clearly that there is a relationship between the variance of IPI and RE. Increased volatility of industrial production is associated with increased volatility of RE consumption, and vice versa. Hence, uncertainty in one of the markets spills over to the other market. We consider cost mitigation as the most likely theoretical transmission channel for the observed weak feedback relationship. One explanation could be that hydropower is a flexible source of energy and can serve as a substitute when other forms of energy (e.g., natural gas resources) are dedicated to specific sectors, like housing, commercial interests, industrial production, and transportation. There is also support for the fact that countries with a large share of hydropower in the energy mix adhere to the Feedback hypothesis (Aspergis et al., 2016). The significant upper and lower tails may indicate that RE is a short-term flexible energy source that responds well to shocks in industry production. This explanation is also in line with the characteristics of hydropower, presented by Intergovernmental Panel on Climate Change, IPCC (2012). For example, when there is low economic activity, RE is easy to curtail. When the economic activity is high, RE production is easy to increase at a low marginal cost. 6. Conclusions and policy implications In this paper, we have examined the possibility of a causal relationship between renewable energy (RE) production and the Industrial Production Index in Canada. In contrast to most previous studies of the RE–growth nexus, we tested not only for dependency in mean but also for nonlinear dependence through quantile analysis. In addition, we used a large dataset containing monthly data, with over 500 observations. Our results further strengthen the notion of nonlinear relations previously proposed by Troster et al. (2018), among others. Quantile regression and quantile causality reveal a presence of asymmetry in the relationship between RE and economic activity. The directionality varies over the distribution, validating different hypotheses in different market conditions. Our quantile regression analysis of Canada, for immediate short-run dynamics within one month, supports the Feedback hypothesis for nearly all quantiles. Moreover, even if quantile causality mainly supports the Conservation hypothesis in the short-run – i.e., the lagged period to the subsequent period – there are weak indications of a feedback relationship in the higher and lower quantiles. This result contradicts a previous study by Chang et al. (2015), which supported the Neutrality hypothesis between RE production and economic growth in Canada. Nonetheless, previous studies by Katırcıoğlu et al. (2016) and Shahbaz et al. (2018b) indicated a Feedback relationship between overall energy production and economic growth. On this basis, there might be a general feedback relationship between energy and economic growth in Canada, though the effect is weaker for RE. From this follows the necessity for a more detailed energy policy to overcome current barriers (technological, socio-economic, and institutional) to a transition to an RE system (Verbruggen et al., 2010; Yaqoot et al., 2015). Though the Canadian economic structure seems to have reached a steady state, in which hydroelectricity has become a complement to fossil fuels and accounts for over 60% of the electricity produced, RE is still subordinate to non-RE sources. Also, hydro-generated electricity does not fully possess the specific synergy and efficiency characteristics associated with other RE sources: for example, solar and wind. From the results of this study, and with support from the theoretical

foundation introduced by Rifkin (2015), among others, we can discern that an economy in the process of transitioning towards a RE structure faces several barriers. If a country is to benefit fully from RE technology, it needs to utilise several of the mechanisms in the transmission channels that link RE to economic growth. In the Canadian case, we postulate that the economy has surpassed some of the barriers, such as cost-to-price relation, and has reached a certain state where hydropower is a reliable and competitive energy source. However, realising the full benefits from a decentralised, diverse, and widespread RE system is hindered by institutional and social economical barriers (Verbruggen et al., 2010; Yaqoot et al., 2015). Regarding other types of RE in Canada, such as solar power and wind power, the supply of these energy sources is most likely to be small for the transmission stage. Previous studies by Bohringer et al. (2017) show that changes in electricity prices induced by different designs of renewable power promotion is a key driver of economic impacts at the sector level. Thus, our results offer valuable insights into energy production behaviour during different maket regimes. This is something important for policy construction as the share of renewables in the energy mix increases in Canada. We believe, in accordance with Rifkin (2015) and others, that the potential paradigm shift will come when RE from solar, wind, tidal, and wave technologies are deployed and distributed on a large scale. To determine the impact of RE over the whole business cycle, future studies should continue to examine the possibility of nonlinear dependence between RE and economic activity. On the basis of our results and the theoretical framework summarised in this study, it is also important that future research highlights the special characteristics and potential transmission mechanisms associated with each type of RE source. Though the use of RE sources has been constrained previously by unfavourable cost relations and institutional and technological barriers, this condition is about to change. Technological innovations have enabled new types of energy systems with large potential benefits. The transformation process towards such an energy system is most likely nonlinear and characterised by threshold effects related to infrastructure and institutional factors. In a study by Dato (2018), it is shown that there may exist synnergies between the decision to invest in energy-efficient technology and the adoption of more RE. To understand the future potential of RE, empirical studies on past relations must be complemented by a theoretical understanding of micro-dynamical relations and transmission channels. Another step in the direction towards better understanding of actual transmission mechanisms is to further investigate how the potentially causal relationship between RE and economic growth evolves over time. As shown by a comparison of quantile regressions in our study and other studies, the estimated relationship between RE production and economic activity is not necessarily the same when specified from one year to the next, between months, or as immediate effects within the same month. Acknowledgements Earlier version of this paper was presented at the Economic Division, Linköping University. This paper benefited from the discussions we had with the seminar participants at the Economics Division, Linköping University, Sweden. Our thanks also go to Victor Troster for making the programme code for the Quantile causality test available to us. The third author is thankful for the financial support provided by the Jan Wallander and Tom Hedelius Foundations. Authors are thankful to the Trinity Business School, University of Dublin for the academic facilities provided to Gazi S. Uddin during his stay at Trinity where important parts of this research work was completed.

C. Wadström et al. / Energy Economics 81 (2019) 626–638

637

Appendix 1 Table A.1 Related nonlinear literature. Article, (year)

Country(s)

Period and frequency

Method

Variables

Bowden and Payne (2009)

US

1949–2006 (annual)

Toda–Yamamoto causality tests, Granger-causality test

Hatemi-J and Uddin (2012)

US

1960–2007 (annual)

Asymmetric causality test

Destek Aspergis et al. (2016)

Six NIC

1971–2011

ARDL, Asymmetric causality test

Alper and Oguz (2016) 7 east-European countries

1990–2009 (annual)

ARDL, Asymmetric causality test

CE, RESE, IE, Y CE ⟺ Y RESE ⟺ Y IE ⇒ Y E, Y SR: E− ⇒ Y− E+ ⇒ Y− Y + ⇒ E− Y, K, L, RE LR: RE+ ⇒ Y− in India, Turkey, South Africa and Mexico Y, RE, K, L RE ⇎ Y in 5 countries, Y ⇒ RE in 1 country, RE ⇒ Y in 1 country

Shahbaz et al. (2016)

US

1981–2015 (monthly)

IPI, RE

RE ⟺ Y

Chen et al. (2017)

China

1969–2013

Wavelet Squared Coherence and Multiple Wavelet Coherence Asymmetric causality test

Y, Oil, Coal

Troster et al. (2018)

US

1989–2016 (monthly)

Quantile granger causality

IPI, RE

Oil ⇒ Y Y⇒C C+ ⇒ Y− RE ⇔ IPI (lowest tail) RE ⇒ IPI (highest tail)

Shahbaz et al. (2018a)

Netherlands and Ireland 1970–2015 (quarterly)

QARDL

E, Y

LR: E− ⇒ G+ For both countries

Shahbaz et al. (2018b)

China, the US, Russia, India, Japan, Canada, Germany, Brazil, France and South Korea.

Quantile-on-Quantile approach

E, Y

For Can and U.S.: E⟺Y Although the effect is less strong for the higher quantiles

1960–2015 (quarterly)

Results

Policy implications/conclusions Asymmetric energy dependence over different sectors in the economy Optimal level of energy consumption in the US.

Investments in renewable energy are costly for NIC. Most east-European is mainly dependent on non-renewable energy. A transformation to renewable energy requires large investments Investments in renewable energy are favourable for economic growth. China should limit their dependence on coal. Renewable energy has a positive impact on the US economy in the short run. Energy efficiency hypothesis is valid. Energy conservation is a way to economic growth. Canada and the US have become fuel-efficient economies with decreased energy demand. Peak Energy has been reached.

Notes: Studies focused on asymmetrical and nonlinear dependence between energy and GDP nexus. NIC: newly industrialised countries. Only the most relevant finding is summarised for each study. Notation: Y = GDP, IPI = Industrial production index, E = Energy Consumption, RE = Renewable energy consumption, C = Coal consumption, CE = Commercial energy consumption, RESE = Residential energy consumption, IE = Industrial energy consumption, K = Capital, L = Labour working hours. “+” denotes positive shocks and “–” denotes negative shocks.

Appendix 2. Quantile regression models We estimated the following quantile regression models:
 ΔIPIÞ ¼ CðτÞ þ β1 ðτÞΔREt−1 þ β2 ðτÞΔIPIt þ εt ðτÞ ΔREt │IΔRE Q ΔRE τ τ ; Iτ

ð1Þ


 ΔIPIÞ ¼ CðτÞ þ β1 ðτÞΔREt−1 þ β2 ðτÞΔREt−2 þ β3 ðτÞΔIPIt þ εt ðτÞ ΔREt │IΔRE Q ΔRE τ τ ; Iτ

ð2Þ


 ΔIPIÞ ¼ CðτÞ þ β1 ðτÞΔREt−1 þ β2 ðτÞΔREt−2 þ β3 ðτÞΔREt−3 þ β4 ðτÞΔIPIt þ εt ðτÞ Q ΔRE ΔREt │IΔRE τ τ ; Iτ

ð3Þ


 ΔREÞ ¼ CðτÞ þ β1 ðτÞΔIPIt−1 þ β2 ðτÞΔREt þ εt ðτÞ Q ΔIPI ΔIPIt │IΔIPI τ τ ; Iτ

ð4Þ


 ΔREÞ ¼ CðτÞ þ β1 ðτÞΔIPIt−1 þ β2 ðτÞΔIPIt−2 þ β3 ðτÞΔREt þ εt ðτÞ Q ΔIPI ΔIPIt │IΔIPI τ τ ; Iτ

ð5Þ


 ¼ CðτÞ þ β1 ðτÞΔIPIt−1 þ β2 ðτÞΔIPIt−2 þ β3 ðτÞΔIPIt−3 þ β3 ðτÞΔREt þ εt ðτÞ; ΔIPIt │IΔIPI ; IΔRE Q ΔIPI τ τ τ

ð6Þ

where the β-parameters are specific for each quantile and are estimated by maximum likelihood in an equally spaced grid of quantiles. The left-hand side of the equations mean that the dependent variable is predicted from information sets including its own lagged values and the value of the explanatory variable. Note that the explanatory variable is in its unlagged form in all equations.

638

C. Wadström et al. / Energy Economics 81 (2019) 626–638

References Alper, A., Oguz, O., 2016. The role of renewable energy consumption in economic growth: evidence from asymmetric causality. Renew. Sust. Energ. Rev. 60, 953–959. Ardian, Faddy, Concettini, Silvia, Creti, Anna, 2018. Renewable generation and network congestion: an empirical analysis of the Italian power market. Energy Journal 39, 3–39. Aspergis, Nicholas, Chang, Tsangyao, Gupta, Rangan, Ziramba, Emmanuel, 2016. Hydroelectricity consumption and economic growth nexus: evidence from a panel of ten largest hydroelectricity consumers. Renew. Sust. Energ. Rev. 62, 318–325. Aspergis, Nicholas, Payne, James E., 2012. Renewable and non-renewable energy consumption-growth nexus: evidence from a panel error correction model. Energy Econ. 34, 733–738. Ayres, Robert U., Warr, Benjamin, 2009. The Growth Engine: How Energy and Work Drive Material Prosperity. Edward Edgar Publishing, Cheltenham. Bohringer, Christoph, Landis, Florian, Tovar Reanos, Miguel Angel, 2017. Economic impacts of renewable energy promotion in Germany. Energy Journal 38, 189–209. Bowden, Nicholas, Payne, James E., 2009. The causal relationship between US energy consumption and real output: a disaggregated analysis. J. Policy Model 31 (2), 180–188. Cai, Yifei, Sam, Chung Yan, Chang, Tsangyao, 2018. Nexus between clean energy consumption, economic growth and CO2 emissions. J. Clean. Prod. 182, 1001–1011. Chang, Tsangyao, Guptab, Rangan, Inglesi-Lotz, Roula, Simo-Kengne, Beatrice, Smithers, Devon, Trembling, Amy, 2015. Renewable energy and growth: evidence from heterogeneous panel of G7 countries using Granger causality. Renew. Sust. Energ. Rev. 52, 1405–1412. Chen, Yiguo, Inglesi-Lotz, Roula, Chang, Tsangyao, 2017. Revisiting the asymmetric causal link between energy consumption and output in China: focus on coal and oil consumption. Energy Sources, Part B: Economics, Planning, and Policy 12 (11), 992–1000. Dato, Prudence, 2018. Investment in energy efficiency, adoption of renewable Energy and household behaviour: evidence from OECD countries. Energy Journal 39 (3), 213–244. Destek, Mehmet A., 2016. Renewable energy consumption and economic growth in newly industrialized countries: evidence from asymmetric causality test. Renew. Energy 95, 478–484. Dogan, Eyup, Ozturk, Ilhan, 2017. The influence of renewable and non-renewable energy consumption and real income on CO2 emissions in the USA: evidence from structural break tests. Environ. Sci. Pollut. Res. 24 (11), 10846–10854. Engle, R., 2002. Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional heteroskedasticity models. J. Bus. Econ. Stat. 20 (3), 339–350. Furuoka, Fumitaka, 2017. Renewable electricity consumption and economic development: new findings from the Baltic countries. Renew. Sust. Energ. Rev. 71, 450–463. Georgescu-Roegen, Nicholas, 1971. The Entropy Law and the Economic Process. Harvard University Press, Cambridge, Mass (Reprint 2014 Edition). Gregory, Allan W., Hansen, Bruce E., 1996. Residual-based tests for cointegration in models with regime shifts. Journal of Econometrics, Elsevier 70 (1), 99–126. Hatemi-J, A, 2012. Asymmetric causality tests with an application. Empirical Economics 43 (1), 447–456. IEA, 2017. Renewables 2017: Analysis and Forecasts to 2022. IEA, Paris. IEA, 2018. Renewables 2018: Analysis and Forecasts to 2023. IEA, Paris. International Energy Agency (2016), 2015. Energy Policies of IEA Countries: Canada Review. Intergovernmental Panel on Climate Change, IPCC, 2012. Renewable Energy Sources and Climate Change Mitigation: Special Report on the Intergovernmental Panel on Climate Change. Cambridge University Press, New York. International Energy Agency, IEA, 2017b. World Energy Outlook. OECD/International Energy Agency, Paris. International Energy Agency, IEA, 2018. Insights Series 2018 - Energy Efficiency Potential in Canada. OECD/International Energy Agency. International Renewable Energy Agency, IRENA, 2018. Renewable Power Generation Costs in 2017. International Renewable Energy Agency, Abu Dhabi. Irandoust, Manuchehr, 2016. The renewable energy-growth nexus with carbon emission and technological innovation: evidence from the Nordic countries. Ecol. Indic. 69, 118–125. Johnstone, Nick, Haščič, Ivan, Popp, David, 2010. Renewable energy policies and technological innovation: evidence based on patent counts. Environ. Resour. Econ. 45 (1), 133–155.

Katırcıoğlu, Salih, Fethi, Sami, Kalmaz Demet, Beton, Çağlar, Dilber, 2016. Interactions between energy consumption, international trade, and real income in Canada: an empirical investigation from a new version of the Solow growth model. International Journal of Green Energy 13 (10), 1059–1074. Kraft, John, Kraft, Arthur, 1978. On the relationship between energy and GNP. Journal of Energy and Development 3, 401–403. Kümmel, Reiner, Lindenberger, Dietmar, Weiser, Florian, 2015. The economic power of energy and the need to integrate it with energy policy. Energy Policy 86, 833–843. Menegaki, A.N., Tugcu, C.T., 2017. Energy consumption and Sustainable Economic Welfare in G7 countries; A comparison with the conventional nexus. Renew. Sustain. Energy Rev. 69, 892–901. Natural Resources Canada, 2017. Energy Fact Book 2016–2017. Government of Canada. Omri, Anis, 2014. An international literature survey on energy-economic growth nexus: evidence from country-specific studies. Renew. Sust. Energ. Rev. 38 (2014), 951–959. Payne, J.E., 2009. On the dynamics of energy consumption and output in the U.S. Appl. Energy 86, 575–577. Portnoy, Stephen, 1991. Asymptotic behaviour of regression quantiles in non-stationary, dependent cases. J. Multivar. Anal. 38 (10), 113. Rifkin, Jeremy, 2015. The Zero Marginal Cost Society. Palgrave Macmillan, New York. Rodríguez-Caballeroa, Carlos Vladimir, Ventosa-Santaulària, Daniel, 2017. Energy-growth long-term relationship under structural breaks. Evidence from Canada, 17 Latin American economies and the USA. Energy Econ. 61, 121–134. Rozin, Paul, Royzman, Edward B., 2001. Negativity bias, negativity dominance, and contagion. Personal. Soc. Psychol. Rev. 5 (4), 296–320 Available at. http://journals.sagepub. com/doi/10.1207/S15327957PSPR0504_2. Salim, Ruhul A., Hassan, Kamrul, Shafiei, Sahar, 2014. Renewable and non-renewable energy consumption and economic activities: further evidence from OECD countries. Energy Econ. 44, 350–360. Şener, Şerife Elif Can, Sharp, Julia L., Anctil, Annick, 2018. Factors impacting diverging paths of renewable energy: a review. Renew. Sust. Energ. Rev. 81 (2), 2335–2342. Shahbaz, Muhammad, Lahiani, Amine, Abosedra, Salah, Hammoudeh, Shawkat, 2018a. The role of globalization in energy consumption: a quantile cointegrating regression approach. Energy Econ. 71, 161–170. Shahbaz, Muhammad, Muhammad, Zakaria, Syed, Jawad, Mantu, Kumar, 2018b. The energy consumption and economic growth nexus in top ten energy-consuming countries: fresh evidence from using the quantile-on-quantile approach. Energy Econ. 71, 282–301. Shahbaz, Muhammad, Hussain, Shahzad, Syed, Jawad, Jammazi, Rania, 2016. Nexus between U.S energy sources and economic activity: time-frequency and bootstrap rolling window causality analysis. Munich Personal RePEc Archive, MPRA Paper No. 68724, 1 January. Tiba, Sofien, Omri, Anis, 2017. Literature survey on the relationships between energy, environment and economic growth. Renew. Sust. Energ. Rev. 69, 1129–1146. Troster, Victor, 2018. Testing for Granger-causality in quantiles. Econ. Rev. 37 (8), 850–866. Troster, Victor, Shahbaz, Muhammad, Uddin, Gazi Salah, 2018. Renewable energy, oil prices, and economic activity: a Granger-causality in quantiles analysis. Energy Econ. 70, 440–452. Tugcu, Can T., Ozturk, Ihlan, Aslan, Alper, 2012. Renewable and non-renewable energy consumption and economic growth relationship revisited: evidence from G7 countries. Energy Econ. 34 (6), 1942–1950. Tversky, Adam, Kahneman, Daniel, 1992. Advances in prospect theory: cumulative representation of uncertainty. J. Risk Uncertain. 5, 297–323. U.S. Energy Information Administration (EIA). (2018-03-11). “Renewable & Alternative Fuels” Available at: https://www.eia.gov/renewable/data.php Verbruggen, Aviel, Fischedick, Manfred, William, Moomaw, Weir, Tony, Nada, Alain, Nilsson Lars, J., Nyboer, John, Jayant, Sathaye, 2010. Renewable energy costs, potentials, barriers: conceptual issues. Energy Policy 38, 850–861. World Bank, 2014. Energy use (kg of oil equivalent per capita).Available at:. Primary source: IEA Statistics, OECD/IEA [Latest accessed: 2019-05-16 18:03]. https://data. worldbank.org/indicator/EG.USE.PCAP.KG.OE?contextual=max&locations= CA&most_recent_value_desc=true. Yaqoot, M., Diwan, P., Kandpal, C.T., 2015. Review of barriers to the dissemination of decentralized renewable energy systems. Renew. Sust. Energ. Rev. 58, 477–490.