How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis

Energy xxx (2014) 1e17

Contents lists available at ScienceDirect

Energy journal homepage: www.elsevier.com/locate/energy

How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis Bongsuk Sung a,1, Woo-Yong Song b, * a b

Department of International Business Management, Woosong University, 186 Jayang-Dong, Dong-Gu, Daejeon 300-718, Republic of Korea Department of Management and Accounting, Habat National University, 125 Dongseo-Daero, Yuseong-Gu, Daejeon 305-719, Republic of Korea

a r t i c l e i n f o

a b s t r a c t

Article history: Received 31 October 2013 Received in revised form 19 March 2014 Accepted 20 March 2014 Available online xxx

This study explores the long- and short-term dynamic relationships between government policies and exports of renewable energy technologies (RETs) at the subsector level (biomass, wind, and solar energy technologies). This allows a more robust exploration of the relationships, in which differences in cost structures and maturity levels exist for different RETs, without losing the generality of the results. Dynamic panel econometric techniques are employed to analyze the relationships, using data of annual measures for 18 countries during 1992e2008. The vector error correction mechanism (VECM) is used to test the dynamic relationships among government policies, exports, and gross domestic product (GDP) for biomass and wind energy technologies, and the vector auto-regression (VAR) model, for solar energy technologies. The study indicates that each subsector has a unique path-dependent process, showing the presence of different positive feedback mechanisms based on interactions among technology-push policy, market-pull policy, exports, and/or GDP in the short and long run. We suggest some policy implications based on the results of this study. Ó 2014 Elsevier Ltd. All rights reserved.

Keywords: Renewable energy technology Government policy Export Subsectoral analysis Dynamic panel approach

1. Introduction As an engine of the new development paradigm, renewable energy technologies (RETs) are expected to play a bridging role between a prospering economy and environmentally sustainable development. Developments in the global RET market have been monitored in numerous studies [1e5]. Several studies predicted that in 2010, more than 20% of global electricity generation would come from renewables and that the share of renewables in electricity generation would reach 31% in 2035 [3e5]. The global market for RETs and the transboundary movements of its components continue to grow substantially. This is because RETs are regarded as being aimed at achieving environmentally sound and sustainable development (ESSD), wherein economic development would be harmonized with environmental protection. Economic development essentially requires more energy, which demands that close attention be paid to energy security issues. In addition, there is

* Corresponding author. Tel.: þ82 42 821 1336; fax: þ82 42 821 1597. E-mail addresses: [email protected] (B. Sung), [email protected], fi[email protected] (W.-Y. Song). 1 Tel.: þ82 42 629 6648; fax: þ82 629 6649.

increasing global concern over environmental problems, including climate change, and global attention on linkages between trade and the environment. These issues as an international pressure have urged countries to conduct economic activities based on environmental considerations, such as the abatement of greenhouse gases (GHGs). In this context, the RET sector is regarded as being economically strategic and has attracted great political interest worldwide. Governments support RETs in order to achieve the 3E (energy, environmental, and economic) goals [6] by reducing their dependence on imports of ever scarcer and more expensive fossil fuels, helping to stem climate change, and enhancing high export potentials in a growing international market [7]. Government policies have been instrumental in the recent growth in renewable energy [5,8]. Nonetheless, there remains substantial room for cost reduction and performance improvement due to the relative immaturity of renewable energy technologies compared to fossil fuel alternatives [9]. From this perspective, there is likely to be an expansion of government policy interventions to reduce costs and thus, create the potential for technological innovation and diffusion, and thereby a larger market share. Renewable energy subsidies jumped to $88 billion in 2011, 24% higher than in 2010, and need to rise to almost $240 billion in 2035 to achieve the trends projected in the

http://dx.doi.org/10.1016/j.energy.2014.03.082 0360-5442/Ó 2014 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

2

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

New Policies Scenario [5].2 The increasing attention given to energy issues and the national policies needed to achieve sustainable economic growth have stimulated research on the link between government policies and economic performance in the RET sector. As continuous improvements in export performance in ensuring industrial growth become more important than ever, the question of whether renewable energy policies may eventually lead to exports on the global market has driven economists to study the interrelations between government policy and export performance. Existing empirical studies have contributed to the discussion on and the understanding of the relationship between government policies and exports of RETs by performing descriptive analyses and exploring case studies [1,3,10e16], and by conducting crosssectional regression [2], time series [17], and panel analyses [7,18e22]. However, additional exploration can contribute to the existing literature in three ways. First, even though the literature has addressed the question of how government policy affects international trade, a comprehensive review of such studies does not exist at this time. Thus, this study attempts to review and identify the common elements and the implications of previous studies, specifically in terms of RET exports and the role of public policy in the sector. Second, most researchers have employed two-, three-, five-, or six-aggregate data on RETs. However, given the difference in the cost structures and maturity levels among different renewable energy sources, it is likely that the effects of policy measures on exports will vary across energy sources/subsectors [2,12,23]. Hence, in order to implement effective policies suited to each subsector-specific situation, further investigation should be conducted using data on different subsectors of the RET field. In this context, the current study deals with the issue at the RET subsector level, as analysis at this level allows a more robust exploration of the effects of government policy on export performance, without losing the generality of the results [22]. Third, with the exception of Costantini and Mazzanti [22] and Sung and Song [7], literature on the dynamic relationship between government policies and RET exports remains sparse. In particular, it is important to investigate the long- and short-run dynamic relationships between government policies and exports, especially considering that most of the panel data used in such studies are heterogeneous and nonstationary co-integrated, and that there are dynamic effects in exports [24e27], production process [28], policy [29] (i.e., path dependence, which means that contemporary inputs are, to some extent, invested for future outputs), and in the interactions among them [30e35]. This calls for a dynamic approach to exploring the relationships between government policies and RET exports. Notably, it is important to consider that there might be a structural break or cross-sectional dependence. Such issues should be taken into account before establishing an empirical model; the applicability of tests for stationarity and co-integration in panel data depends on whether the panel tests allow for structural breaks and/or cross-sectional dependence. In this regard, a systematic approach is lacking in previous studies. Thus, the current study examines both long- and short-run dynamics while systematically applying the dynamic panel approach. This paper is organized as follows. Section 2 provides a literature review on the relationships between government policies and RET exports. Section 3 discusses the theoretical model and assumptions of this paper and presents a description of the data. The empirical

2 This is one of three scenarios (namely “450,” “Current Policies,” and “New Policies”) presented in the IEA [5], which takes account of broad policy commitments and plans that have been announced by countries, including national pledges, to reduce GHG emissions and phase out fossil energy subsidies, even if the measures to implement these commitments have yet to be identified or announced.

results are presented and interpreted in Section 4. Section 5 summarizes the main findings and lists the implications and limitations of this study. 2. Literature review Three approaches have emerged in empirical studies that analyze how government policies affect international trade in RETs (Table 1). The first approach is based on the multivariate cross-sectional regression model by Jha [2], who proved that an exporting country’s policy support plays a crucial role in promoting its export performance in the RET market. She estimates that a composite variable composed of feed-in tariffs (FITs) and the share of renewable energy to the total energy supply may contribute to an increase in exports of aggregate renewable, solar, and wind energy technologies, and of undenatured ethanol, with the coefficients 0.410, 0.946, 0.976, and 1.710, which are significant at 1% or 5%, respectively. The second approach, based on the vector error correction model (VECM), is used by Algieri et al. [17]. They analyze the trade specialization dynamics of the global solar photovoltaic (PV) sector using time series data. They estimate that a 1% increase in price reduces solar PV exports by 1.15% in the long run and suggest that the relative price should be interpreted as an indicator of competitiveness in solar PV exports. They also indicate that foreign income is one of the major factors driving solar PV exports. According to them, since income elasticity exceeds unity, solar panels are regarded as “superior” goods, and thus, an increase in income is expected to raise demand for exports substantially. They find that although the trend variable is significant, its impact is less pronounced than that of the other variables, and suggest that other factors such as consumer preferences and public incentives play a minor role in encouraging exports. The third trend is the most recent in the literature. It explores the nexus between government policies and RET exports by analyzing panel data using the static and dynamic panel approaches. The static panel approach follows mainly the panel gravity model [18e21]. These studies test the effect of government policies on exports by inputting policy variables (environmental regulations and/or renewable energy supportiveness) and/or national innovation system-related variables of exporting and/or importing countries into the general stochastic formulation of the gravity model. The model has general parameters for trade analysis between two countries, such as incomes and populations of the exporting and importing countries, and distance and existence of common border between the countries. The policy variables include carbon dioxide (CO2) emissions, expenditure on environmental protection and taxes (used as proxy variables for a country’s environmental regulation), public R&D expenditure, existence of incentive tariffs, and obligations and tax measures for RETs (used as proxy variables for renewable policy supportiveness). Overall, we find that the exporting country’s R&D expenditure directed at each renewable energy subsector has a positive effect on exports of solar energy technologies, but does not have a significant effect on exports of wind energy technologies [19e21]. Incentive tariffs of exporting and importing countries also help to promote bilateral trade in solar energy technologies [19,21]. Obligations that require suppliers to provide a specific production quantity or percentage from renewable sources through quota systems or targets, however, do not significantly affect bilateral trade in RETs. Environmental regulations of exporting and importing countries become drivers in promoting bilateral trade in solar and aggregate RETs [18e20]. Gross domestic product (GDP) serves as an indicator of the sizes of the exporting and importing countries. As a socioeconomic variable that represents the market size of RETs based on the demands for exports and imports, GDP has a largely positive

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

Study

Industry

Period

Costantini and Crespi [18]

Two aggregated sectors (renewable energies, and energy savings and management)

1996e2005 20 OECD exporting countries and 147 importing countries

Jha [2]

Algieri et al. [17]

2007 34 Countries Six-aggregated sectors (biomass, small hydro, geothermal, ocean, solar, and wind energy technologies) Three disaggregated sectors (solar, wind energy technologies, and ethanol) Solar PV 1991e2009 USA

Groba [19]

Solar energy technologies

1999e2007

Groba [20]

Solar energy technologies

2000e2007

Costantini and Mazzanti [22]

Five aggregated sectors (air, water and waste pollution abatements, renewable energy, and energy efficiency in building and lighting)

1996e2007

Cao and Groba [21] Two disaggregated sectors (solar PV and wind energy technology components)

1996e2008

Sung and Song [7]

1991e2007

Six-aggregated sectors (biomass, small hydro, geothermal, ocean, solar, and wind energy technologies)

Number of countries

Methodology/model/approach Variables Dependent variable: bilateral export flows in US dollars Independent variables: general parameter for gravity model (e.g., GDP, population, distance, etc.), environmental regulations of importing and/or exporting countries (CO2 emissions per GDP, environmental protection expenditure as a percentage of GDP, revenues from environmental taxes as a percentage of total tax revenues, and public environmental investments as a percentage of GDP), innovation system of importing and/or exporting countries (R&D expenditure, % of GDP), patent applications per 100,000 people, five-year moving average of the number of patent applications to the US Patent and Trademark Office as a percentage of total patents from residents, technology diffusion (ArCo index methodology) Control variables: total foreign direct investment inflows as a percentage of GDP of the importing country, index of rule of law of the importing country (based on Ref. [38]) Multivariate regression model Dependent variable: exports or imports in US dollars (cross-sectional approach) Independent variables: factor (comprising the contribution of renewable energy to the total energy supply and FIT) score calculated by Principle Component Analysis, import tariff, share of global patents Panel gravity model (static panel approach)

VECM (time series approach)

Dependent variable: exports in US dollars Independent variables: industrial production of the foreign country, relative export price Panel gravity model (static Dependent variable: bilateral export flows in dollars 21 OECD exporting panel approach) Independent variables: general parameter for gravity model (e.g., GDP, population, countries and 129 distance, etc.), effective applied tariff to solar energy technology imports from the importing countries importing country to the exporting country, FDI of the importing country, index of rule of law of the importing country (based on Ref. [38]), environmental regulations of importing and exporting countries (CO2 emissions per GDP), energy intensity of the exporting country, R&D expenditure in solar energy technologies in the exporting country, patent stock of exporting country Panel gravity model (static Dependent variable: bilateral export flows in US dollars 23 OECD exporting panel approach) Independent variables: general parameter for the gravity model (e.g., GDP, population, countries and 129 distance, etc.), environmental regulations and renewable energy supportiveness (each importing countries importing country’s effective applied tariff to solar energy technology imports from the exporting country, share of solar electricity generation in the exporting country, share of renewable energy electricity generation in the importing country, energy intensity of exporting and importing countries, R&D expenditure in solar energy technologies in the exporting country, environmental regulation index of importing and exporting countries, existence of incentive tariff, existence of obligations, existence of tradable green certificates) 14 Exporting countries Panel gravity model (dynamic Dependent variable: bilateral export flows in US dollars panel approach) Independent variables: general parameter for gravity model (e.g., GDP, population, and 145 importing distance, etc.), public and private environmental measure (energy tax revenues as countries a percentage of total revenues, environmental tax revenues as a percentage of total revenues, pollution abatement and control expenditures as a percentage of GDP, and number of eco-management and audit scheme initiatives by private firms as a percentage of GDP), public and private innovation measures (knowledge stock of importing and exporting countries) China (exporting) and Panel gravity model (static Dependent variable: exports from China to 43 importing countries in US dollars 43 importing countries panel approach) Independent variables: general parameter for gravity model (e.g., GDP, population, distance, etc.), market size (amount of electricity generated from solar PV and wind energy technologies of importing countries and China), policies (existence of incentive tariff, existence of obligations and tax measures, each importing country’s effective applied tariff to solar and wind energy technology imports from China, patent stock of importing country, China’s patent stock) 18 OECD countries Panel VECM (dynamic Dependent variable: exports in US dollars panel approach) Independent variables: R&D expenditure (technology-push policy), contribution of renewable energy to the total energy supply (demand-pull policy), past export performance in dollars

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

Table 1 A summary of studies that explore how government policies affect international trade in RETs.

3

4

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

effect on bilateral trade in solar and wind [19e21], and aggregate RETs [18]. The studies following the dynamic panel approach are based on the panel gravity model [22] and the panel VECM [7], which estimate dynamic models that specify the dependent variable depending on its values in the previous period. Costantini and Mazzanti [22] devise a research model that adds environmental policy variables and variables related to the national innovation system of exporting and importing countries to the general stochastic formulation of the gravity model and test the relationships between public policy and exports. They find that the stringency of environmental regulations and national innovation intensity have a significant positive effect on exports of energy technologies. Sung and Song [7] set up a panel VECM and test the casual relationships between public policy and exports. They find solid and convincing evidence of a long-run equilibrium relationship between R&D expenditure (used as a proxy variable for technology-push policy measures) and exports. From the panel causality tests based on the one-step system generalized method of moments (GMM) estimation results, they also find short-run bidirectional causality between exports and the contribution of renewable technologies to the total energy supply (used as a proxy variable for the demandpull policy measure, especially with FITs). Besides the empirical studies presented in Table 1, other studies have emphasized the role of government policies in promoting exports of RETs, performing case analyses or descriptive analyses, for example, [1,3,10e16]. Lund [11] investigates the impacts of energy policies on growth in the RET industry by synthesizing key case studies. He confirms the correlation among a country’s level of support for government policies, its home market size and industry position, and its share of the world product market. Liu and Goldstein [16] examine the extent to which Chinese firms’ solar PV and wind energy technology successes have been enabled by policy support and whether these policies appear to have been driven by broader goals versus RET export promotion, per se. With respect to the wind turbine manufacturing industry, public policy serves as a key driving force for the evolution of innovation modes as well as market expansion [10,13,15]. De la Tour et al. [14] show that China has become a major player in the global solar PV industry mainly due to its innovation performance driven by national policies. They find evidence to suggest that government policy toward both wind and solar energy technologies originated from the need to promote export-competitive Chinese companies. Larsen and Sønderberg [1] and REN 21 [3] find that most countries with policies promoting renewable energy have been able to increase their exports of these technologies over time, possibly indicating that energy policies do play an important role in increasing exports of these technologies. Although the methods and approaches employed in the abovementioned studies differ, they essentially study the same relationship, namely, exports and the role of government policy in the RETs sector, and reach the same conclusions. First and most important, the majority of these studies regard governments’ renewable energy policies as the strongest extrinsic political force promoting exports of RETs, mainly due to the impetus given to innovation. Notably, these studies show that R&D expenditure and incentive FITs directed at each renewable energy subsector have had a positive effect on exports. This means that the government plays an important role in facilitating coercive and mimetic isomorphism [36,37] in firms’ RET-related innovation activities by encouraging them to undertake R&D, commercialize RETs, and create both local and export markets for RETs. Second, most studies use GDP as the most relevant socioeconomic factor that is likely to affect the extent of controlled exports. More recent studies also use GDP as a proxy variable for many socioeconomic considerations. The income effect on exports of RETs is the most commonly tested relationship [18,39e41]. It is based on the environmental Kuznets

curve theorem, which suggests that demand preferences for environmental products increase with income [42], which in turn is expected to raise the demand for exports substantially [17,19e21]. Then, GDP is used as a proxy variable for sociopolitical pressure that allows supporting regulatory costs to promote RETs [43]. GDP is also used as a proxy variable for the capitalelabor ratio of each country [22] and for the home market size of RETs [10,44]. Third, previous studies use export performance and export dynamics as dependent variables despite the possibility that several exportcompetitiveness indices can be considered when investigating the impact of government policy on exports. Export performance and export dynamics are largely and significantly affected by R&D expenditure, incentive tariffs (i.e., FITs), and GDP. Fourth, the effects of R&D expenditure, incentive FITs, and GDP on export performance vary among RET subsectors, as shown by Jha [2] and Cao and Groba [21]. This means that the sensitivity of RET exports to government policies may vary across specific subsectors. This study considers the aforementioned points from previous studies and employs dynamic panel econometric techniques to analyze the link between government policy and exports of RETs. This study first tests for structural breaks and the presence of crosssectional dependence. The JarqueeBera test for normality, cumulative sum of recursive residuals (CUSUM), and cumulative sum of recursive residuals of squares (CUSUMQ) tests for structural breaks are conducted in each individual time series. To detect the presence of cross-sectional dependence, this study employs the statistic proposed by Frees [45] and the cross-sectional dependence (CD) test of Pesaran [46]. It is well known that when N (the crosssectional dimension) > T (the panel’s time dimension) (as is the case in this study), the two tests enjoy highly desirable statistical properties relative to other tests, and they can be used with balance and unbalanced panels alike. Panel unit root tests are conducted to investigate the order of integration of the series in the panel data, reflecting the results of the tests for structural breaks and the tests for the presence of cross-sectional dependence. A number of panel unit root tests have been proposed in the literature [47e54]. Choosing which one to test for stationarity in panel data depends on whether it allows for structural breaks and/or cross-sectional dependence. Then, if the results of the panel unit root test indicate that the series are nonstationary, the possibility of a long-term equilibrium relationship among the variables in question can be confirmed by performing panel co-integration tests based on the methodologies of Pedroni [55,56], Banerjee and Carrion-i-Silvestre [57], or Westerlund [58]. In so doing, we also need to consider the results of the JarqueeBera, the CUSUM and CUSUMQ tests, and the CD test of Pesaran [46]. Finally, this study devises an empirical model based on the results of the panel unit root tests and the panel co-integration tests, whereupon panel causality tests are undertaken. 3. Theoretical settings and data The current study considers four factors that may strongly affect the direction and robustness of empirical results in the existing literature vis-à-vis the relationship between energy policies and exports of RETs. First, this study focuses on the relationship between government policies and the export performance of RETs, rather than on other export-competitiveness indices. Second, we deal with RETs at the subsector level. Third, this study devises a dynamic model that accounts for most of the panel data used being heterogeneous and non-stationary co-integrated, and for dynamic effects on exports, production processes, policies, and the interactions among them. Fourth, our model includes R&D expenditure as a proxy for the technology-push policy, FITs as a proxy for the government policy variables associated with market-pull policy

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

directed at each RET subsector, and each country’s GDP per capita as the control variable. Accordingly, to investigate the dynamic relationships among exports, government policies, and GDP per capita, this study considers the panel vector auto-regression (VAR) model proposed by Holtz-Eakin et al. [59], which can be expressed as

2 2 3 3 ln EXit a1l mþ1 6 a 7 X6 6 ln RADit 7 6 4 5 ¼ 4 2l 5 þ 4 ln CRESit a3l j¼1 a4l ln GDPit 2 ln EXitj 6 ln RADitj 6 4 ln CRESitj ln GDPitj 2

b11j b21j b31j b41j 3

b12j b22j b32j b42j

b13j b23j b33j b43j

3

b14j b24j 7 7 b34j 5 b44j

2 3 2 3 h1i m1it 7 6h 7 6m 7 7 þ 4 2i 5 þ 4 2it 5 5 h m 3i h4i

(1)

3it

m4it

where h1i, h2i, h3i, and h4i are country-specific effects for the ith individual in the panel, and m1it, m2it, m3it, and m4it are the disturbance terms. ln EX is the log of the export performance (measured in export values); ln RAD refers to the log of R&D expenditures; ln CRES, as a proxy for FITs, is the log of the contribution of each renewable energy to the total energy supply (measured as a percentage of total primary energy supply); and ln GDP is the log of the per capita real GDP (divided by 1000 people). Our motivation for including per capita real GDP in the model is to control for the relationships among higher income, increasing demand preference for exports, increasing sociopolitical pressure to support and promote the use of these technologies, and increasing home market size of RETs. Export performance of the RET sector is closely related to public policies, since a government’s renewable energy policies to promote the RET industry send a clear signal to firms that it endorses R&D and the commercialization of RETs. This leads firms to try to become isomorphic with the government’s expectations by proactively carrying out various activities related to RETs, including R&D and commercialization. It is also equally likely that improved export performance may lead to more government spending on policymaking and implementation to promote RETs. GDP growth is positively associated with public awareness about environmental issues, including clean energy, such as RETs. This sociopolitical force may lead to home market and export growth of RETs by demanding and supporting policy expenditure to promote RETs. It is generally recognized that an increase in exports is the main driver of GDP growth. The aforementioned points imply that export performance, government policies, and GDP may be jointly determined, suggesting that there are path-dependent processes (i.e., dynamic effects) by which contemporary inputs are, to some extent, invested for future output in ln EX, ln RAD, ln CRES, ln GDP, and the interactions between them. In this context, the signs of b (b11j, .., b44j) are expected to be positive. The data utilized in this study comprise annual measures for 17 years during 1992e2008 on 18 countries (Austria, Canada, Denmark, Finland, France, Germany, Italy, Japan, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, Turkey, the United Kingdom, and the United States). The current study uses data on RET export values, renewable energy R&D expenditures, contributions of renewable energy to the total energy supply at the subsector level (biomass, wind, and solar) of each country, and per capita real GDP at the national level. The balanced panel-data set has 306 observations, representing 17 years of data for each of the 18 countries. The countries, subsectors, and periods have been selected on the grounds of data availability. All variables are expressed in logarithmic form. Data on RET export values (EX) for each subsector in each country were extracted from the United Nations Commodity Trade

5

Statistics (UN COMTRADE) database based on the Harmonized Commodity Description and Coding System (HS) [60]. The topologies of RETs and components are well defined by Jha [2], starting from the classification HS [60] (as described in Table A3 of Appendix A). Data on the renewable energy R&D expenditure (RAD) of each subsector in each country were obtained from the International Energy Agency’s (IEA) Energy Technology Research and Development database. The contribution of renewables to the total energy supply (CRES) of each subsector in each country is used as a proxy for FITs because of the lack of a reference database for FITs. Our logic is that FITs have a positive effect on the renewable energy share in the grid. Considering that CRES and FITs constituting a composite variable are highly correlated at 0.7 [2], and that the production of renewable energy is driven largely by FITs, which are the most common and effective policy measure currently under implementation [2,3], CRES seems a suitable proxy for FITs [7]. Data on the CRES of each subsector in each country are calculated based on data obtained from the IEA’s Renewable and Waste Energy Supply database and the US Energy Information Administration’s International Energy Statistics. Data on real GDP per capita for each country are obtained from the database of the Organization for Economic Cooperation and Development for economic, environmental, and social statistics. The variables EX, RAD, and GDP are calculated at constant 2009 prices and international purchasing power parity. 4. Empirical analysis 4.1. Testing frameworks To test whether sample data have the skewness and kurtosis matching a normal distribution, we conduct the JarqueeBera test. The test statistic JarqueeBera (JB) is defined by JB ¼ n/ 6(S2 þ (k  3)2/4), where n is the number of observations (or degrees of freedom in general), S is the sample skewness b 3 =m b 3=2 b m m3=2 ðS ¼ m 2 Þ, which is an estimator of 1 ¼ 3 = 2 , k is the b 4 =m b 22 Þ, which is an estimator of b2 ¼ m4 =m22 , sample kurtosis ðk ¼ m and m2 and m3 are the theoretical second and third central moments, P b j ¼ 1=n ni¼ 1 ðxi  xÞj ; j ¼ 2; 3; 4. respectively, with estimates m JB is asymptotically chi-squared distributed with two degrees of freedom because JB is just the sum of squares of two asymptotically independent standardized normal distributions (see Ref. [61]). That means: H0 (¼ normality) has to be rejected at level a if JB  c21a;2 . Table A1 of Appendix A summarizes the basic statistics of the variables for the subsectors (biomass, solar, and wind energy technologies) during the research period. The JarqueeBera normality test results show that almost all these series do not deviate substantially from the normal distribution, demonstrating that the null hypothesis of normality cannot be rejected at the 1%, 5%, or 10% significance levels in each individual time series. We also apply CUSUM and CUSUMQ tests proposed by Brown et al. [62] to test the stability of each individual series. The null hypothesis is that the coefficients are the same in every period, i.e., there is no structural break. The recursive least squares estimates of b are based on estimating yt ¼ b0t xt þ et , t ¼ 1, ., n by least squares recursively for t ¼ k þ 1, ., n giving n  k least squares estimates b Þ. Recursive least squares estimates are efficiently ; .; b ðb T

kþ1

b computed using the Kalman Filter. If b is constant over time, then b t should quickly settle down to a common value. The recursive repffiffiffiffi b 2 ½1 þ x0t ðXt0 Xt Þ1 xt , siduals are defined as wt ¼ yt = ft , ft ¼ s 0 b are the vector of ordinary least squares (OLS) where x , y , and b t

t

t

estimates for the regression parameters based on data t ¼ 1, ., n, wt

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

6

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

are recursive Chow Forecast “t-statistics” with n2 ¼ 1, and if bi changes in the next period, then the forecast error will not have mean zero. The CUSUM and CUSUMQ test statistics are constructed Pt b b b2 based on these residuals: CUSUMt ¼ j ¼ kþ1 w j = s w , s w ¼ 1=n  k P P Pn 2 t n b2 b2 t ¼ 1 ðwt  wÞ ; CUSUMt ¼ j ¼ kþ1 w j = j ¼ kþ1 w j . Under the null hypothesis that b is constant, CUSUMt has mean zero and variance that is proportional to t  k  1, and CUSUMQt behaves like c2(t) and confidence bounds can be easily derived. Hence, if the graphical plot of CUSUM and CUSUMQ stays within the 5% significance level, then coefficient estimators are said to be stable. The CUSUM and CUSUMQ tests (results available upon request from the authors) also suggest that almost all the series are stable over time, showing that the null hypothesis of the absence of structural breaks cannot be rejected at the 5% level of significance in each individual time series of the subsectors. Overall, the results imply that almost all the series in biomass, solar, and wind energy technologies are stable over time. To detect the presence of cross-sectional dependence, we carry out the Pesaran CD test [46] and Frees’ tests [45]. The standard 0 panel-data model is expressed as: yit ¼ ai þ b xit þ uit, i ¼ 1, ., N and t ¼ 1, ., T where xit is a K  1 vector of regressors, b is a K  1 vector of parameters to be estimated, and ai represents time-invariant individual nuisance parameters. Under the null hypothesis, uit is assumed to be independent and identically distributed over the period and across cross-sectional units. Under the alternative, uit may be correlated across-sections, but the assumption of no serial correlation remains. Thus, the hypothesis of interest is H0: rij ¼ rji ¼ Cor(uit, ujt) ¼ 0 for i s j versus H1: rij ¼ rji s 0 for some i s j where rij is the productemoment correlation coefficient of the PT PT 2 1=2 disturbances and is given by rij ¼ rji ¼ t ¼ 1 uit ujt =ð t ¼ 1 uit Þ PT ð t ¼ 1 u2jt Þ1=2 . The number of possible pairings rises with N. The pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pesaran CD statistic is as follows: CD ¼ 2T=NðN  1Þ PN1 P r ij Þ. The CD statistic follows the standard normal ð N1 i¼1 j ¼ iþ1 b distribution under the null hypothesis of no cross-sectional dependence. In Frees’ tests, the statistic is based on the sum of the squared rank correlation coefficients and equals P PN1 2 b R2ave ¼ 2=NðN  1Þ N1 i¼1 j ¼ iþ1 r ij , where if {ri1, ., riT} is defined to be the ranks of {ui1, ., uiT}, such that the average rank is (T þ 1/2), then Spearman’s rank correlation coefficient equals rij ¼ rji ¼     PT   PT  1 1 2 . A funcrjt  T þ 12 t ¼ 1 rit  T þ 2 t ¼ 1 rit  T þ 2 tion of the statistic shown by Frees [45] follows a joint distribution of two independently drawn c2 variables. The null hypothesis is rejected if R2ave > critical value (for more details on testing for cross-sectional dependence in the panel-data model, see Ref. [63]). The Pesaran CD test [46] for fixed effects regression residuals in each subsector strongly rejects the null hypothesis of no crosssectional dependence, showing residual mean absolute correlations (0.386, Pesaran CD statistic ¼ 9.432 (p < 0.000) for biomass energy technologies; 0.366, Pesaran CD statistic ¼ 12.287 (p < 0.000) for wind energy technologies; and 0.382, Pesaran CD statistic ¼ 7.983 (p < 0.000) for solar energy technologies). The statistic proposed by Frees [45] is 1.980, 2.292, and 2.536 for biomass, wind, and solar energy technology, respectively, which are all beyond the critical value (1%). Frees’ tests [45] also reject the null hypothesis of cross-sectional independence. We perform three panel group-wise heteroskedasticity testsdLagrange multiplier, likelihood ratio, and Wald testsdto ascertain whether the residuals present the problem with groupwise heterogeneity. The panel group-wise heteroskedasticity tests show that the null hypothesis of homoskedasticity within crosssectional units is rejected at the 1% significance level (see

Table A2 of Appendix A). The results imply that there is group-wise heteroskedasticity and cross-sectional dependence in the panel. Having established that the series correlate cross-sectionally, the next step is to implement a panel unit root test that accounts for the presence of cross-sectional dependence. Cross-sectional dependence biases the panel-data unit root test results towards the alternative hypothesis [64]. The occurrence of cross-sectional dependence may reflect the mixed results observed over the different panel unit root tests. One such test is the cross-sectionally augmented version by Pesaran [54] of the Im et al. [52] test, known as the ImePesaraneShin (IPS) test. Pesaran’s panel unit root test [54] is favored over all others for its simplicity and clarity and we opt for this test. Let yit be the export performance of country i in period t. The model could be represented by the dynamic AR (1) panel-data model for heterogeneity in the intercept but not in the autoregressive parameters yi0 ¼ d0 þ d1hi þ yi0, yit ¼ diyit1 þ uit, uit ¼ (1  a)hi þ yit where a1 ¼ .aN ¼ a for each i ¼ 1, ., N, t ¼ 2, ., T. The series have a unit root (or are integrated of order 1) if ai ¼ 1 and are stationary if ai < 1. In case of independence across countries, the error term statistics E(hi) ¼ 0, E(yit) ¼ 0 for i ¼ 1,., N, and t ¼ 2, ., T, in which a test for the presence of a unit root in the panel is presented by the null hypothesis H0: a ¼ 1, and E(yityis) ¼ 0 for i ¼ 1, ., N, and t s s. Pesaran [54] builds the assumption that the error terms yit follow a single common factor structure yit ¼ liftþεit. The common factor is assumed to be stationary and to impact the cross-section by a fraction determined by the individual-specific factor loading li. Because of the common factor, cross-sectional dependence arises and can be approximated by the cross-section P mean yt ¼ 1=N N i ¼ 1 yit . As usual, the εit are assumed to be independent and identically distributed across i and t with zero mean and variance s2, and E(εit)4 < N. Furthermore, εit, ft, and li are mutually independently distributed for all i. The augmented DickeyeFuller regression proposed by Pesaran [54] takes the following form:

Dyit ¼ ai þ bi yit1 þ ci yt1 þ

q X

bij Dytj þ

j¼0

q X

dij Dyitj þ eit ;

j¼1

P PN 1 where yt1 ¼ N 1 N i ¼ 1 yit1 and Dy ¼ N i ¼ 1 Dyt . The test statistic to test the presence of unit roots is defined as P CIPSðN; TÞ ¼ N 1 N i ¼ 1 ti ðN; TÞ, where ti(N,T) is the t-ratio of the OLS estimate of bi. The results of Pesaran’s test [54] for ln EX, ln RAD, ln CRES, and ln GDP (Table 2) show that the hypothesis of the series containing a unit root is confirmed at the 1% or 10% significance levels. This implies that the series are non-stationary. The results of the panel unit root test shown in Table 2 indicate that there can be a longterm equilibrium relationship among the variables. Hence, the current study implements the heterogeneous panel co-integration tests proposed by Westerlund [58], which allow for cross-sectional dependence. Since Westerlund’s tests [58] are based on structural rather than residual dynamics, there is no one common factor restriction. The Westerlund panel co-integration test [58] has the following error correction model (ECM):

Dyit ¼ d0i dt þ ai yit1 þ l0i xit1 þ

pi X j¼1

aij Dyitj þ

pi X

gij Dxitj þ eit ;

j¼qi

where t ¼ 1, ., T and i ¼ 1, ., N index the time series and crosssectional units, respectively, dt contains the deterministic compo0 nents, for which there are three cases (dt ¼ 0; dt ¼ 1; dt ¼ (1, t) ), and k-dimensional vector xit is modeled as a pure random walk, such that Dxit is independent of eit, and these errors are independent

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17 0

across both i and t, and where yit1 þ li xit1 is equal to

ai ðyit1 

b0i xit1 Þ,

l0i

0 ai bi .

and hence, ¼ The parameter ai determines the speed at which the system corrects and reverts to the 0 equilibrium relationship yit1  bi xit1 after a sudden shock. If ai ¼ 0, there is no error correction, which implies that yit and xit are not co-integrated (H0: ai ¼ 0 for all i). If ai < 0, there is error correction, and thus, co-integration. The alternative hypothesis depends on what is being assumed about the homogeneity of ai. Two of the tests (GT, Ga), called group mean tests, do not require the values of ai to be equal, which means H0 is tested versus H1g : ai < 0 for at least one i. The group mean tests are computed in the PN P b b b b following way: GT ¼ N1 N i ¼ 1 a i =SEð a i Þ and Ga ¼ 1=N i ¼ 1 T a i = a i . The second pair of tests (pT, pa), called panel tests, assume that ai is equal for all i, and are, therefore, designed to test H0 versus H1p : ai ¼ a < 0 for all i. The panel statistics are computed in the following b i =SEð a b i Þ and pa ¼ T a b i . In these cases, SEð a b i Þ is the way: pT ¼ a b i . These tests have limiting normal conventional standard error of a distributions and are consistent. Table 3 shows the results of the Westerlund panel co-integration tests [58] that include an intercept, as well as those with an intercept and a linear trend. For biomass energy technologies, the results indicate significance for the Ga, Pt, and Pa statistics in both the cases, the constant, and the constant and the trend. For wind energy technologies, the results indicate significance for the Ga, Pt, and Pa statistics only in the constant and the trend. However, the results for solar energy technologies do not indicate significance for any statistics. Overall, the results of Table 3 indicate that, on balance, there is at least some evidence of co-movement among the variables for biomass and wind energy technologies, and there is no evidence of co-integration among the variables for solar energy technologies.

4.2. Model specification and empirical testing In the last phase, dynamic panel causality tests are conducted taking into account the results of panel co-integration tests. Since evidence of co-integration is found in the case of biomass and wind energy technologies, the Engle and Granger [65] approach can be used to estimate an ECM. Hence, for biomass and wind energy technologies, dynamic panel causality tests based on the VECM are employed to evaluate the short-run and long-run directions of causality between the examined variables. The Granger causality model among the variables in question, based on the panel VECM to be examined, can be expressed as follows:

2 3 b11j ln EXit m 6 X 6 ln RADit 7 6 b21j 4 5 ¼ 4 b31j ln CRESit j¼1 ln GDPit b41j 2 3 2

b12j b22j b32j b42j

b13j b23j b33j b43j

32

3

b14j ln EXitj 6 ln RADitj 7 b24j 7 76 7 b34j 54 ln CRESitj 5 ln GDPitj b44j 3 2 g1i Dm1it 6 g2i 7 6 Dm2it 7 7 7 6 þ6 4 g3i 5½ECTit1  þ 4 Dm3it 5 g4i Dm4it (2)

where D is the first difference operator, ln EX is the log of exports, ln RAD is the log of R&D expenditure, ln CRES is the log of the contribution of renewable energy to the total energy supply and serves as a proxy for FITs, ln GDP is the log of per capita real GDP, ECTit1 is the error correction term lagged one period that comes from the lagged residuals derived from the long-run co-integrating relationship, bijs are the short-run adjustment coefficients, and mits

7

are disturbance terms assumed to be uncorrelated with each other and to have mean zero. As for biomass and wind energy technologies, after having established a co-integrating relationship, it is necessary to estimate the long-run equilibrium relationship given by the error correction term. The long-run equilibrium coefficients can be estimated by using various single equation estimators, such as the fully modified OLS procedures (FMOLS) proposed by Pedroni [66], the dynamic OLS (DOLS) estimator of Saikkonen [67] and Kao and Chiang [68], and the pooled mean group estimator (PMG) proposed by Pesaran et al. [69]. One may also use system estimators as panel VARs, estimated with GMM or Quasi Maximum Likelihood. Single equation approaches assume homogeneity between cross-section units for the long-run relationship, whereas short-run dynamics may be panel specific across cross-sections. Although this restriction may seem too severe for some variables, allowing all parameters to be panel specific would reduce the appeal of the panel-data approach considerably [70]. The FMOLS estimator is consistent and efficient in estimating long-run co-integrating coefficients. It also allows for endogenous regressors and serial correlation. As demonstrated by Kao and Chiang [68], the DOLS outperforms the FMOLS estimator in terms of mean biases. Despite the differences between the two methods, the estimates from both the FMOLS and the DOLS are asymptotically equivalent for more than 60 observations [71]. Hence, to determine the long-run equilibrium relationship among the variables in question, the present study performs the DOLS and PMG procedures developed by Kao and Chiang [68] and Pesaran et al. [69] for estimating the residuals to be included in the panel VECM as the error correction terms. For solar energy technologies, the study takes the differences in the VAR structure (1) based on the results of the panel cointegration tests, wherein there is no co-movement among the series. Thus, the empirical model to test the causal relationship among the variables in question is based on the panel VAR model in the first difference, which can be expressed as follows:

2 3 b11j ln EXit m b21j 6 ln RADit 7 X6 6 4 5¼ 4 b31j ln CRESit j¼1 ln GDPit b41j 2

b12j b22j b32j b42j

b13j b23j b33j b43j

32

3

2

3

b14j ln EXitj Dm1it 6 7 6 7 b24j 7 76 ln RADitj 7 þ 6 Dm2it 7 b34j 54 ln CRESitj 5 4 Dm3it 5 ln GDPitj b44j Dm4it (3)

where D is the first difference operator, ln EX is the log of exports, ln RAD is the log of R&D expenditure, ln CRES is the log of the contribution of renewable energy to the total energy supply and serves as a proxy for FITs, ln GDP is the log of per capita real GDP, bijs are the short-run adjustment coefficients, and mits are disturbance terms assumed to be uncorrelated with each other and to have mean zero. However, for VECM structure (2) and VAR structure (3), differencing introduces a simultaneity problem, because the lagged endogenous variables on the right-hand side correlate with the new differenced error term. In addition, heteroskedasticity exists in the genuine errors across industries. To deal with these problems, Arellano and Bond [72] propose a difference-GMM approach, in which the lags in the explanatory variables at different levels are used as instruments. For the instruments to be valid, no serial correlation must exist among the error terms. The optimal lag length, m, is likewise selected so that no serial correlation is observed in the residuals. This assumption may be tested while taking into account the fact that if the disturbances are not serially correlated, there should be evidence of significant negative firstorder serial correlation and no evidence of second-order serial correlation in the differenced residuals. Arellano and Bond’s [72]

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

8

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

Table 2 Panel unit root tests. Variables

LnEX DlnEX LnRAD DlnRAD LnCRES DlnCRES LnGDP DlnGDP

Biomass energy technologies

Solar energy technologies

Wind energy technologies

With trend

Without trend

With trend

Without trend

With trend

Without trend

3.003 8.127*** 1.941 4.491*** 3.676 3.540*** 1.253 1.565*

0.552 8.892*** 1.074 5.653*** 2.564 4.448*** 2.718 2.562***

6.693 6.228*** 0.442 7.932*** 4.320 3.841*** 1.253 1.565*

2.581 6.856*** 1.189 4.259*** 8.120 5.219*** 2.718 2.562***

6.274 7.655*** 2.654 6.244*** 0.365 2.894*** 1.253 1.565*

3.140 6.379*** 5.782 4.414*** 2.790*** 4.163*** 2.718 2.562***

Notes: The numbers denote the Pesaran cross-sectional augmented DickeyeFuller (CADF) test z [t-bar] statistics. The test of the null hypothesis of non-stationarity is based on the mean of individual DickeyeFuller (DF) (or augmented DF) t-statistics of each unit in the panel. To remove the cross-sectional dependence, the standard DF (or ADF) regressions are augmented with the cross-sectional average of lagged levels and the first differences of the individual series (CADF statistics). The lag lengths for the panel test are based on those employed in the univariate ADF test. The normalized z-test statistic is calculated using the t-bar statistics. *** and * denote significance at the 1% and 10% levels, respectively.

statistic, mj, is used to test the null hypothesis of there being no jth order correlation in the differenced residuals. For the overidentifying restrictions, both Sargan’s test [73] and Hansen’s J test [74] are conducted, and an inference is made chiefly by analyzing the Hansen test results. This is because the Sargan test is not robust against heteroskedasticity or autocorrelation. On the other hand, the Hansen test, which gives the minimized value of the system GMM criterion function, is robust. In this context, the different sources of causation can be identified by testing H0: bhij ¼ 0, h, j ¼ 1, 2, 3, 4 with h s i. For biomass and wind energy technologies, the significance of the coefficients of the ECTit1 variables represents the long-run causality of the speed with which deviations from the long-run equilibrium are eliminated following changes in each variable, which are examined by testing H0: gji ¼ 0, cj, i. On the basis of the difference-GMM estimation results, causality is determined by running Wald tests on the coefficients of variables. The test statistics follow a c2 distribution with k  m degrees of freedom. The results of the panel causality tests for biomass, wind, and solar energy technologies are presented in Tables 4e6, respectively. Panel A of Table 4 shows the DOLS and PMG results obtained using the co-integration equations for biomass energy technologies, which can be interpreted as long-run coefficients. From these results, the R&D elasticity estimates are 1.708 and 1.215, and are statistically significant at the 1% level. The CRES elasticity estimates are 0.714 and 2.591, and are statistically significant at the 1% level. The GDP elasticity estimate is 0.247 and is statistically significant at the 1% level. Panel B(a) of Table 4 shows the estimates, the Sargan and Hansen test results, and m1 and m2 statistics. As the m1 and m2

statistics show, the selection of one lag is needed in VECM structure (2) to have no serial correlation in the disturbance terms m1it, m2it, m3it, and m4it. The Hansen test does not reject the validity of the instruments. The dynamic regression results in Panel B(a) of Table 4 show that EX has positive effects on EX, RAD, and CRES and that RAD has positive effects on RAD and GDP, and a negative effect on CRES. Moreover, CRES has a positive effect on CRES, and GDP has positive effects on RAD and GDP. However, there is no significant correlation between EX and GDP, and CRES and GDP. RAD and CRES have no effects on EX, and CRES has no effect on RAD. Panel B(b) of Table 4 shows the Wald test results for heterogeneous panel causality obtained using VECM structure (2). This study finds evidence of a positive short-run and strong causal relation running from EX to RAD, and a negative short-run and strong causal relation running from RAD to CRES. There is also evidence of positive short-run and strong bidirectional causality between RAD and GDP, positive strong bidirectional causality between EX and GDP, and a positive short-run causal relation from EX to CRES. There is no evidence of short-run bidirectional causality between CRES and GDP, and EX and GDP, or of a short-run linear causal relation running from RAD to EX, and from CRES to EX and RAD. In the long run, only one coefficient in the model, wherein DCRESit is the dependent variable, is significant, indicating that exports and CRES could be key adjustment factors as the systems depart from the long-run equilibrium. The joint tests of EX and RAD show that exports have a positive effect on public RAD, rejecting the null hypotheses that exports do not promote RAD. There is also a strong Granger causality running from RAD to CRES, which implies that RAD has a negative effect on the contribution of renewable energy to the total energy supply. The joint tests of EX and GDP, and RAD and GDP,

Table 3 Panel co-integration tests. Variables

Gt

Ga

Pt

Pa

Value z-Value Robust p-value Value z-Value Robust p-value Value z-Value Robust p-value Value z-Value Robust p-value

Biomass energy technologies

Solar energy technologies

Wind energy technologies

With trend

Without trend

With trend

Without trend

With trend

Without trend

1.613 5.317 0.206 0.845 7.752 0.022 3.531 7.353 0.001 0.615 6.436 0.077

1.654 2.608 0.360 1.970 5.426 0.075 3.488 4.410 0.002 1.300 3.986 0.083

1.698 4.903 0.635 1.164 7.584 0.432 3.979 6.877 0.153 0.631 6.427 0.385

1.638 2.679 0.521 1.332 5.811 0.944 2.101 5.723 0.890 0.942 4.217 0.807

1.253 7.069 0.822 0.826 7.762 0.068 4.094 6.754 0.009 0.935 6.257 0.099

1.273 4.322 0.678 1.798 5.530 0.185 1.973 5.878 0.106 0.989 4.186 0.195

All distributed statistics are standard normal. The lag and lead lengths are set to 1 and 0, respectively. Choosing too many lags and leads can result in a deterioration of the small sample properties of the test. To control for cross-sectional dependence, robust critical values are obtained through 5000 bootstrap replications.

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

9

Table 4 Panel causality tests for biomass energy technologies. Panel A: Panel long-run estimates Estimators

Variables ln RAD

ln CRES

ln GDP

Dynamic OLS Pooled mean group procedure

1.078 (25.070)*** 1.215 (3.090)***

0.714 (7.290)*** 2.591 (2.910)***

0.247 (2.290)*** 0.103 (0.190)

Panel B: BlundelleBond system GMM panel VECM causality test (a) GMM estimation Independent variables

Dependent variables

Dln EXit

Dln RADit

Dln CRESit

Dln GDPit

Dln EXit1 Dln EXit2 Dln RADit1 Dln RADit2 Dln CRESit1 Dln CRESit2 Dln GDPit1 Dln GDPit2

0.785 (0.040)***

0.327 (0.154)**

0.115 (0.185)*

0.021 (0.026)

0.114 (0.098)

0.328 (0.161)**

0.177 (0.296)**

0.080 (0.047)*

0.137 (0.239)

0.397 (0.296)

1.109 (0.185)***

0.083 (0.089)

0.077 (0.102)

0.590 (0.125)***

0.051 (0.125)

0.817 (0.039)***

ETCit1 Sargan test Hansen test m1 m2

0.081 (0.083) 208.470 [0.000] 17.070 [1.000] 2.790 [0.005] 0.610 [0.544]

0.141 (0.099) 232.550 [0.000] 13.770 [1.000] 3.030 [0.002] 1.280 [0.202]

0.137 (0.075)* 181.170 [0.000] 10.960 [1.000] 2.310 [0.012] 1.090 [0.277]

0.056 (0.034)* 179.950 [0.000] 17.490 [1.000] 1.670 [0.094] 1.400 [0.161]

Dln EX

Dln RAD

Dln CRES

Dln GDP

e 1.340 0.330 0.570 0.950 e 1.170 0.130 9.100***

4.500** e 1.790 22.260*** 3.030 20.550*** e 1.220 23.860***

2.440** 4.050** e 0.170 3.380** 0.170 3.890** e 0.380

0.630 2.940* 0.880 e 2.650 3.930** 2.830** 0.240 e

(b) Statistic values for panel causality Independent variables (sources of causation)

Short-run

Dln EX Dln RAD Dln CRES Dln GDP

Long-run Strong (joint)

ECT Dln EX ECT Dln RAD ECT Dln CRES ECT Dln GDP ECT

Dependent variables

Panel A reports the result of model tests wherein EX is the dependent variable. Numbers in parentheses are t-statistics. *** denotes the 1% significance level. Panel B(a) contains the results of tests based on one-step system GMM estimates. *, **, and *** denote the 10%, 5%, and 1% significance levels, respectively. GMM robust standard errors are in parentheses, and p-values are in square brackets. The explanatory variables are assumed to be endogenous and are instrumented in GMM style [75]. Panel B(b) reports the c2statistics. *, **, and *** denote that the null hypothesis of no causation is rejected at the 10%, 5%, and 1% significance levels, respectively.

show that exports and RAD are positively correlated with a country’s national income. Panel A of Table 5 shows the DOLS and PMG results for wind energy technologies. They show that the R&D elasticity estimates are 0.722 and 0.565, and they are statistically significant at the 1% level. The CRES elasticity estimates are 0.197 (significant at the 1% level) and 0.038 (insignificant). The GDP elasticity estimates are 0.832 (significant at the 1% level) and 0.055 (insignificant). Panel B(a) of Table 5 shows the estimates, the Sargan and Hansen test results, and the m1 and m2 statistics. The m1 and m2 statistics show that the selection of one lag, one lag, seven lags, and three lags is needed in VECM structure (2) for there to be no serial correlation in the disturbance terms m1it, m2it, m3it, and m4it, respectively. The Hansen test does not reject the validity of the instruments. The dynamic regression results in Panel B(a) of Table 5 show that EX has positive effects on EX and CRES, that CRES has positive effects on EX, CRES, and GDP, and that GDP has a positive effect on GDP. However, there are no significant correlations between EX, GDP, and RAD; CRES and RAD; EX and RAD; and EX and GDP. GDP has no effect on CRES. Panel B(b) of Table 5 shows the Wald test results for heterogeneous panel causality obtained using VECM structure (2). This study finds evidence of a positive short-run and strong causal

relation running from EX to CRES, and positive short-run causal relations running from CRES to EX and GDP. There is also evidence of positive strong bidirectional causality between EX and GDP. There is no evidence of short-run bidirectional causality between EX and RAD, EX and GDP, RAD and GDP, RAD and CRES, and CRES and GDP. There is no evidence of a short-run causal relation running from GDP to CRES. In the long run, no significant variables could be key adjustment factors as the systems depart from longrun equilibrium. The joint tests of EX and CRES, and EX and GDP, show that exports have a positive effect on the contribution of renewable energy to the total energy supply, rejecting the null hypotheses that EX does not cause CRES and GDP. There is also a strong bidirectional Granger causality between EX and GDP, which implies that exports are positively correlated with a country’s national income. Table 6 presents the results of panel causality tests for solar energy technologies. Panel A of Table 6 shows the estimates, the Sargan and Hansen test results, and the m1 and m2 statistics. As the m1 and m2 statistics show, the selection of one lag, one lag, seven lags, and three lags is needed in VAR structure (3) to have no serial correlation in the disturbance m1it, m2it, m3it, and m4it, respectively. The Hansen test does not reject the validity of the instruments. In

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

10

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

Table 5 Panel causality tests for wind energy technologies. Panel A: Panel long-run estimates Estimators

Variables ln RAD

ln CRES

ln GDP

Dynamic OLS Pooled mean group procedure

0.722 (1.672)*** 0.565 (0.145)***

0.197 (0.875)*** 0.038 (0.048)

0.832 (4.735)*** 0.055 (0.744)

Panel B: BlundelleBond system GMM panel VECM causality test (a) GMM estimation Independent variables

Dependent variables

DlnEXit

DlnRADit

DlnCRESit

DlnGDPit

Dln EXit1 Dln EXit2 Dln RADit1 Dln RADit2 Dln CRESit1 Dln CRESit2 Dln CRESit3 Dln CRESit4 Dln CRESit5 Dln CRESit6 Dln CRESit7 Dln GDPit1 Dln GDPit2 Dln GDPit3

0.544 (0.183)***

0.053 (0.199)

0.197 (0.269)***

0.023 (0.025)

0.056 (0.072)

0.114 (0.548)

0.034 (0.052)

0.008 (0.018)

0.019 (0.009)**

0.170 (0.113)

0.013 (0.003)***

0.031 (0.212)

1.278 (1.536)

0.737 0.022 0.004 0.038 0.001 0.023 0.005 0.138

ETCit1 Sargan test Hansen test m1 m2

0.156 (0.123) 221.230 [0.000] 15.630 [1.000] 2.050 [0.041] 1.570 [0.116]

0.067 (0.678) 229.630 [0.000] 9.170 [1.000] 2.020 [0.043] 1.130 [0.260]

0.062 (0.067)* 190.950 [0.000] 7.320 [1.000] 2.790 [0.005] 1.050 [0.296]

(0.088)*** (0.066) (0.032) (0.031) (0.015) (0.013)* (0.017) (0.199)

0.488 (0.073)*** 0.335 (0.074)*** 0.043 (0.072) 0.014 (0.029) 216.680 [0.000] 17.030 [1.000] 1.840 [0.065] 0.860 [0.391]

(b) Statistic values for panel causality Independent variables (sources of causation)

Short-run

Dln EX Dln RAD Dln CRES Dln GDP

Long-run Strong (Joint)

ECT Dln EX ECT Dln RAD ECT Dln CRES ECT Dln GDP ECT

Dependent variables

Dln EX

Dln RAD

Dln CRES

Dln GDP

e 0.600 3.850** 0.020 1.600 e 1.180 1.820 2.50*

0.070 e 2.270 0.250 0.010 0.030 e 0.100 0.420

7.490*** 0.430 e 0.480 0.850 2.610* 0.670 e 0.260

0.860 0.210 14.430*** e 0.250 5.180** 0.240 0.900 e

Panel A reports the result of model tests wherein EX is the dependent variable. Numbers in parentheses are t-statistics. *** denotes significance at the 1% level. Panel B(a) contains the results of tests based on one-step system GMM estimates. *, **, and *** denote the 10%, 5%, and 1% significance levels, respectively. GMM robust standard errors are in parentheses, and p-values are in square brackets. The explanatory variables are assumed to be endogenous and are instrumented in GMM style [75]. Panel B(b) reports the c2-statistics. *, **, and *** denote that the null hypothesis of no causation is rejected at the 10%, 5%, and 1% significance levels, respectively.

addition, Panel B of Table 6 reports the Wald test for the null hypothesis of no causality. The dynamic regression results in Panel A of Table 6 show that GDP, CRES, and EX have a positive effect on EX; RAD has a negative effect on EX; CRES has a negative effect on RAD and a positive effect on CRES; EX and GDP have positive effects on GDP; GDP has a positive effect on RAD; and there are no significant effects of EX on RAD and CRES, and of RAD on CRES, GDP, and RAD. There is also no significant correlation between GDP and CRES. Panel B of Table 6 shows the Wald test results for heterogeneous panel causality obtained using VAR structure (3). This study finds evidence of positive short-run linear causality running from CRES and GDP to EX, and negative short-run linear causality from RAD to EX and from CRES to RAD. There is also evidence of a positive shortrun linear causal relation running from GDP to RAD and of a native short-run causal relation running from CRES to RAD. There is no evidence of short-run bidirectional causality between CRES and GDP, or short-run linear causality from EX to RAD and CRES, and from RAD to CRES and GDP.

5. Discussion and conclusions 5.1. Summary and policy implications This study tested the dynamic relationships between government policy and exports of RETs for three subsectors (biomass, wind, and solar energy technologies) using panel data for 18 countries spanning the period 1992e2008. The study implemented heterogeneous panel unit root tests and co-integration tests, while taking into account the results of the normality and structural breaks tests for each individual time series and tests for the presence of cross-sectional dependence in the panel. This study found evidence of co-movement among the series of biomass and wind energy technologies. However, there was no co-movement among the series of solar energy technologies. Thus, empirical models to test the causal relationship among the variables in question were based on the panel VEC model for biomass and wind energy technologies, and the panel VAR model in first difference for solar

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

11

Table 6 Panel causality tests for solar energy technologies. Panel A: GMM estimation Independent variables

Dependent variables

Dln EXit

Dln RADit

Dln CRESit

Dln GDPit

Dln EXit1 Dln EXit2 Dln RADit1 Dln RADit2 Dln CRESit1 Dln CRESit2 Dln CRESit3 Dln CRESit4 Dln CRESit5 Dln CRESit6 Dln CRESit7 Dln CRESit8 Dln CRESit7 Dln GDPit1 Dln GDPit2

0.722 (0.071)***

0.103 (0.217)

0.099 (0.183)

0.041 (0.051)***

0.043 (0.025)*

0.217 (0.171)

0.042 (0.032)

0.011 (0.010)

0.032 (0.018)*

0.073 (0.042)*

0.006 (0.008)

0.389 (0.114)***

0.813 (0.232)***

0.994 (0.061)*** 0.056 (0.025)** 0.005 (0.014) 0.017 (0.042) 0.015 (0.041) 0.035 (0.009)*** 0.029 (0.019) 0.012 (0.028) 0.009 (0.012) 0.581 (0.575)

Sargan test Hansen test m1 m2

176.740 [0.000] 16.690 [1.000] 2.620 [0.009] 0.430 [0.668]

196.460 [0.000] 13.740 [1.000] 2.380 [0.017] 1.340 [0.181]

171.900 [0.000] 10.210 [1.000] 2.240 [0.025] 0.700 [0.484]

0.579 (0.051)*** 0.243 (0.031)*** 239.560 [0.000] 17.780 [1.000] 1.830 [0.067] 0.850 [0.395]

Panel B: Statistic values for panel causality Independent variables (sources of causation)

Short-run

Dln EX Dln RAD Dln CRES Dln GDP

Dependent variables

Dln EX

Dln RAD

Dln CRES

Dln GDP

e 3.030* 3.090* 11.560***

0.220 e 2.940* 12.240***

0.290 1.700 e 1.020

15.540***\ 1.080 0.590 e

Panel A contains the results of tests based on one-step robust GMM estimates. *** and ** denote significance at the 1% and 5% levels, respectively. GMM robust standard errors are in parentheses, while p-values are in square brackets. The explanatory variables are assumed to be endogenous and are instrumented in GMM style [75]. Panel B reports the c2-statistics. *** and * denote that the null hypothesis of no causation is rejected at the 1% and 10% significance levels, respectively.

energy technologies. For biomass and wind energy technologies, the long-run policy elasticity was computed using the DOLS and PMG techniques. Finally, panel GMM estimations were conducted to determine dynamic relationships within the series of biomass, wind, and solar energy technologies, and to deal with the simultaneity problem introduced by differences and the existence of heteroskedasticity in the genuine error across countries. Then, on the basis of the difference-GMM estimation results, causality was determined by running Wald tests on the coefficients of variables. Based on the causality results from this study, the following main results and implications can be drawn. First, as expected, this study showed that the effects of policy measures on exports vary across each subsector, and that there is a distinct difference between the three subsectors in terms of path dependence, that is, dynamic effects exist. This means that future outputs depend on contemporary inputs (e.g., policies, exports, and GDP) in both the short and long run. The most prominent pathdependent feature for the biomass energy technology sector is positive feedback based on interactions between and among the technology-push policy, exports, and/or GDP in both the short and the long run. The path-dependent process for the wind energy technology sector has a tendency to be based on interactions between and among the market-pull policy, exports, and/or GDP in the short run, and on interactions between and among technologypush policy, exports, and/or GDP in the long run. The solar energy technology sector is affected only by short-term path-dependent processes based on interactions between and among the marketpull policy, exports, and/or GDP. The results support the arguments of Jha [2], Johnstone et al. [23], and Guérin and Schiavo [12] that, given the differences between renewable energy sources in

terms of cost structures, maturity levels, and so on, the effects of policy measures are likely to vary across energy sources. This indicates that the subsector level of the analysis appears to be crucial for providing robust and more in depth explanations of the relationships between export performance, public policies, and other variables, including GDP. Thus, as governments place increasing emphasis on developing a portfolio of energy alternatives, policymakers should understand these differences, so as to devise policies suited to each specific subsector and to enhance the effectiveness and efficiency of policies. Second, this study found solid and convincing evidence of positive long-run equilibrium relationships between RAD and EX and negative long-run equilibrium relationships between CRES and EX for biomass and wind energy technologies. The long-run relationships emerging from the DOLS results indicate that a 1% increase in public RAD increases exports of biomass (wind) energy technologies by 1.078% (0.722%). Moreover, a 1% increase in the contribution of renewable energy to the total energy supply decreases exports by 0.714% (0.197) for biomass (wind) energy technologies. The results showed the absence of short-run linear relationships between policies and exports in the two subsectors, with the exception of positive short-run linear causality running from CRES to EX for wind energy technologies. The results of the short-run and long-run causality tests on the relationship between government policies and exports of biomass and wind energy technologies imply that countries should set up long-term public policies to promote exports of RETs. In particular, considering that the estimated R&D elasticity is positive in the long run, the government’s R&D policy should continue to be directed toward creating reliable and positive long-term policy elasticity with

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

12

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

regard to exports of biomass and wind energy technologies. According to Lewis and Wiser [10] and Haas et al. [76], as a technology matures, transition of the policy focus from the supply side to the demand side is required; that is, governments should provide less support for technology development and manufacturing infrastructure, and should instead concentrate efforts on uptake and deployment. The distinction between mature and immature technologies, however, is far from discrete. For this reason, any renewable energy development policy should consist of a mix of instruments that are designed to create a smooth transition from heavily subsidized, infant technologies, to mature technologies bolstered by competitive, demand-side incentives [12]. This means that in the early innovation phase of the product and technology cycle, government support should focus on dynamic innovationoriented policy, such as R&D and technology subsidy policies [8,77]. Furthermore, in such cases, the technology-push policy is more effective at enhancing competitiveness than the market-pull policy. FITs may lead to specialization and higher export levels, which rely on the successful expansion of the domestic market [11]. Hence, in order to expand the domestic market successfully, FIT policies must lead to a quick and substantial expansion of renewable energy capacity at both the distributed- and utility-scale level [78]. In cases in which considerable room for technological improvement remains, FIT measures do not seem to work in this way and will not have a significant effect on exports in the long run. Even if FITs have positive effects on exports, with increasing exports, FITs tends to entice numerous foreign companies to manufacture and export RET components to the country implementing the policy. Whether the effects of FITs on exports are positive or negative depends on the relative sensitivity of exports and imports of RET components to FITs, and this effect varies across specific subsectors, technologies, market sizes, and so on. [2,11]. Hence, another possible explanation for the negative coefficients is that the sensitivity of imports to FITs is very large compared to that of exports, and this may lower or even reverse export flows [7]. Our results concerning the long run positive effect of RAD on EX and the negative effect of CRES on EX, can be understood in the same context; despite continuous government support, costs for most RETs remain high compared to fossil fuel alternatives. These technologies are relatively immature technically and are thus poised for further (and perhaps significant) cost reductions and performance improvements [9]. In the future, R&D policy will have a profound effect on the enhancement of the export performance of biomass and wind energy technologies. As Lesser and Su [79] have suggested, in the long term, FIT policies should be designed to promote the technological advancement of RETs, so that they can compete directly with “traditional” generation technologies. Third, with regard to the effect of policies on exports in the short run, this study found that there is no short-run linear causality running from RAD and CRES to EX for biomass energy technologies. The results, however, indicated evidence of positive short-run linear relationships running from CRES to EX for wind and solar energy technologies, as seen in Jha [2], Groba [19] and Cao and Groba [21], and of negative short-run linear causality running from RAD to EX for solar energy technologies. It is clear that production of renewable energy is largely driven by FITs [2]. The results back up Guérin and Schiavo’s [12] argument that the optimal application of support policies in maturing technologies, such as solar and wind, can be demand-side incentives, such as FITs and feed-in premiums, compared to relatively immature technology, such as biomass energy. FITs may lead to specialization and higher export levels when they expand the domestic market successfully [11] by a quick and substantial expansion of renewable energy capacity at both the distributed- and utility-scale level [78]. Increasing the share of electricity from renewable energy sources in the shortest time and

at rather low cost can be effected by a well-designed (dynamic) FIT system that includes the most important design criteria, including a carefully calculated starting value, a dynamic decrease of the FIT based on technological learning, and the implementation of a stepped and technology-specific tariff structure [80]. In this context, the results suggest that governments should instead devise and implement short-run FIT measures for wind and solar energy technologies. In other words, increasing the sensitivity of exports of RET components to FIT measures is one of the most important policy issues for any government in the short run. Hence, policymakers should try to create and implement strategically planned FIT measures to ensure that crucial aspects of the policy process become integrated into the country’s domestic market penetration and export promotion strategies in the short run [7]. Fourth, this study found evidence of bidirectional causality between exports and real GDP from the results of the joint tests on the real GDP and the error correction term, and exports and the error correction term, for biomass and wind energy technologies, and the results of the short-run tests on exports and the real GDP for solar energy technologies. As explained in Section 2, the real GDP analyzed in this study represents a variety of meanings, such as the relationship between higher income and increasing demand preference for exports [17e21,39e42], increasing sociopolitical pressure promoting the use of these technologies [34], and increasing home market sizes of RETs [10,44]. Furthermore, the results of the joint or short-run causality tests on exports, policy variables, and GDP suggest that the policy measures have a positive effect on exports or real GDP; thus, EX or GDP is positively affected by the policy variables RAD or CRES. The existence of bidirectional causality between exports and real GDP suggests that government policies to support RETs can make a substantial contribution to achieving environmentally sound and sustainable development in society, by enhancing energy security (e.g., reducing dependence on imports of ever scarcer and more expensive fossil fuels), increasing environmental sustainability (e.g., GHG emissions reductions), and sustaining economic growth (e.g., GDP growth and export expansion). Chien and Hu [81] confirmed a positive relationship between renewable energy use and GDP via increasing capital formation, but not via an increasing trade balance. Thus, they argued that a renewable energy policy related to increasing capital formation (e.g., tax incentives for the establishment of RET industries) would be more efficient than policies related to an increasing trade balance (e.g., increasing the tax on imported fossil fuels). In this context, the results of this study suggest that policymakers should try to identify renewable energy policy measures to increase capital formation and implement mechanisms to form a positive relationship between exports and GDP. Fifth, except for the two equations with RAD as the dependent variable in biomass and solar energy technologies, this study, as in Costantini and Mazzanti [22] and Sung and Song [7], found evidence of dynamic effects in almost all equations for wind and solar energy technologies; the dependent variables depend on their values in the previous period at 1% significance levels. Based on the results of this study and the insights provided by the literature review, we suggest that the use of dynamic estimators in the panel-data structure are more suitable for testing the relationships between government policy and exports of RETs than the other models and methods used in most previous studies. 5.2. Limitations and further research Although this study contributes to a better understanding of the dynamic relationships between government policy and exports of

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

RETs at the subsector level, it has some limitations. First, the variable related to policy strategies, which was not controlled for in this paper, is a relevant factor and is likely to affect the extent of exports. Many countries, such as the United States, the member countries of the European Union, and Japan, implement export promotion policies to help their RET companies explore export potential and feasibility, to expand overseas markets, and to reduce the risk of export trade deficiency [82e86]. Export promotion policies are divided mainly into two groups: (1) international marketing services that focus on international market research, assistance with the export process, support to link buyers and sellers, advisory services, and so on, and (2) export-credit guarantees. Export promotion programs contribute positively to firms’ export performance [87e89] because they increase firms’ informational and experiential knowledge [90], stimulate managers’ positive attitudes and perception towards exports, and increase export commitment

13

[91]. This means, as UNCTAD [32] and Coe et al. [34] stated, that the export orientation of the economy leads to broader learning that influences all aspects of production capabilities. In this context, subsector policy strategies for promoting exports can strengthen the export orientation of RETs. Hence, further research should test the effect of controlling variables related to industry-specific policies and strategies for export promotion, which are likely to affect exports of RETs. Second, this study did not consider the sensitivities of export and import policies. RET policies have an effect on exports as well as imports. Jha [2], Sung and Song [7], and Lund [11] have noted the sensitivity of exports and/or imports to RAD and FIT policies. Although sensitivity of exports and/or imports to such policies may vary across specific subsectors, this study assumed that an increase in government support would have the same effect on exports across disparate subsectors. Given the differences in sensitivity, it is

Appendix Table A1 Descriptive statistics. Country

Variable

Biomass energy technologies Mean

AUT

CAN

DEN

FIN

FRA

GER

ITA

JPN

NED

NZL

NOR

POR

ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln

EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD

2.919 0.822 0.563 4.436 3.020 1.057 0.083 4.408 2.961 0.757 0.671 4.415 2.671 0.958 1.067 4.380 3.516 0.651 0.189 4.379 3.996 0.897 0.311 4.418 3.630 0.800 0.026 4.364 4.092 10143 0.190 40488 3.321 1.105 0.559 4.429 1.616 0.590 0.107 4.270 2.489 0.103 0.512 4.497 0.235 0.708

Solar energy technologies

Wind energy technologies

SD

JarqueeBera

Mean

SD

JarqueeBera

Mean

SD

0.219 0.231 0.153 0.096 0.247 0.318 0.082 0.115 0.471 0.365 0.279 0.112 0.187 0.320 0.068 0.122 0.103 0.510 0.149 0.092 0.191 0.304 0.118 0.083 0.177 0.294 0.395 0.107 0.170 0.441 0.084 0.028 0.289 0.481 0.201 0.130 0.243 0.653 0.067 0.118 0.242 0.207 0.084 0.173 0.150 0.385

1.477 0.159 3.396 0.848 0.649 2.088 1.790 1.135 1.699 3.996 1.440 0.989 2.254 0.069 2.119 1.137 1.512 1.364 1.618 1.046 1.326 1.309 1.364 0.800 1.207 2.207 1.613 1.049 0.557 2.006 1.381 0.213 5.206* 0.559 0.879 1.163 0.643 0.336 1.508 1.019 4.120 0.707 0.307 1.099 0.564 0.810

2.878 0.439 2.124 4.436 3.087 0.779 2.126 4.408 3.377 0.530 1.401 4.415 2.861 0.257 2.625 4.380 3.509 0.871 0.976 4.379 3.745 1.798 1.774 4.418 3.216 1.552 2.098 4.364 3.928 1.8020 1.638 4.488 3.209 1.256 2.102 4.422 1.631 0.475 0.023 4.270 2.319 0.222 2.041 4.497 0.334 0.470

0.249 0.131 0.550 0.096 0.287 0.250 0.067 0.115 0.371 0.187 1.121 0.112 0.210 0.388 0.189 0.122 0.109 0.525 0.043 0.092 0.250 0.096 0.979 0.083 0.202 0.188 0.268 0.107 0.243 0.611 0.820 0.028 0.143 0.103 0.591 0.130 0.239 0.559 0.080 0.118 0.302 0.233 0.118 0.173 0.180 0.497

3.320 0.732 1.991 0.848 1.411 0.691 0.984 1.135 2.171 4.156 2.736 0.989 0.311 0.427 1.538 10137 2.211 1.600 1.389 1.046 1.853 3.055 1.184 0.800 0.926 1.239 34.050*** 1.049 1.239 26.120*** 1.729 0.213 2.801 0.623 1.875 1.163 0.141 0.303 0.079 1.019 2.003 1.597 3.664 1.099 0.435 1.848

2.891 0.609 1.759 4.436 2.966 0.161 1.413 4.408 3.359 0.891 0.891 4.415 2.5898 0.028 1.256 4.380 3.563 0.047 1.838 4.379 3.884 1.340 0.041 4.418 3.303 0.263 1.018 4.364 3.398 0.829 2.520 40488 2.927 0.972 0.050 4.429 1.170 0.933 0.771 4.270 2.116 0.038 1.436 4.497 0.294 1.631

0.207 0.539 2.639 0.096 0.216 0.647 0.812 0.115 0.495 0.139 0.330 0.112 0.327 0.359 0.596 0.122 0.096 0.342 1.151 0.092 0.229 0.168 0.641 0.083 0.232 0.687 1.090 0.107 0.089 0.230 1.948 0.028 0.209 0.125 0.395 0.130 0.317 0.210 0.905 0.118 0.310 0.234 0.847 0.173 0.135 1.299

JarqueeBera 1.116 2.413 4.662 0.848 0.444 1.420 0.450 1.135 1.051 1.376 1.876 0.989 0.446 0.916 1.282 1.137 2.042 1.898 1.081 1.046 2.112 0.624 1.276 0.800 1.443 1.420 2.006 1.049 1.216 2.829 5.624 0.213 0.552 0.395 0.648 1.163 2.875 2.520 1.398 1.019 0.781 0.929 1.526 1.099 0.220 2.552

(continued on next page)

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

14

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

Table A1 (continued ) Country

Variable

Biomass energy technologies Mean

ESP

SUI

SWI

TUR

USA

UK

ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln ln

CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP EX RAD CRES GDP

1.032 4.195 2.750 0.767 0.023 40272 2.921 1.106 0.443 4.424 3.228 0.705 0.389 4.520 0.890 1.655 0.928 3.100 3.982 1.869 0.239 4.493 3.387 0.646 0.100 4.365

Solar energy technologies

Wind energy technologies

SD

JarqueeBera

Mean

SD

JarqueeBera

Mean

SD

JarqueeBera

0.146 0.140 0.268 0.136 0.156 0.158 0.181 0.245 0.259 0.105 0.109 0.125 0.132 0.072 1.299 1.607 0.214 0.949 0.218 0.393 0.020 0.117 0.123 0.176 0.324 0.127

0.753 1.117 0.124 0.682 3.423 1.028 1.283 1.253 1.804 0.888 0.022 0.710 1.178 0.980 1.226 1.768 1.435 1.775 0.110 9.315*** 0.901 1.091 0.871 1.768 1.437 1.092

3.602 4.195 2.656 1.170 1.813 4.272 2.948 0.431 3.281 4.424 2.957 1.303 1.790 4.520 1.336 1.213 0.082 3.100 3.526 2.014 1.842 4.493 2.940 0.733 1.741 4.365

0.270 0.140 0.283 0.158 0.582 0.158 0.201 0.200 2.219 0.105 0.156 0.114 0.310 0.072 1.235 0.951 0.072 0.949 0.222 0.127 0.048 0.117 0.135 0.334 1.537 0.127

0.304 1.117 1.314 9.556* 17.570*** 1.028 0.287 7.046** 18.700*** 0.888 1.799 1.552 0.467 0.980 1.232 1.505 0.890 1.775 0.981 2.202 37.070*** 1.091 1.721 1.369 2.388 1.092

1.688 4.195 2.708 0.653 0.061 40272 2.983 0.410 0.659 4.424 3.011 0.287 3.246 4.520 1.202 1.737 0.874 3.100 3.895 1.394 0.690 4.493 2.631 0.776 0.586 4.365

0.653 0.140 0.355 0.327 0.832 0.158 0.189 0.302 0.525 0.105 0.115 0.295 2.145 0.072 1.389 0.953 0.779 0.949 0.193 0.481 0.385 0.117 0.114 0.490 0.532 0.127

0.742 1.117 3.525 1.004 1.593 1.028 0.520 3.265 1.104 0.888 2.433 10.360** 4.078 0.980 1.400 1.333 1.203 1.775 0.615 22.470*** 1.972 1.091 1.542 1.794 2.003 1.092

The country codes AUT, CAN, DEN, FIN, FRA, GER, ITA, JPN, NED, NZL, NOR, POR, ESP, SUI, SWI, TUR, USA, and UK denote Austria, Canada, Denmark, Finland, France, Germany, Italy, Japan, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, Turkey, the United States, and the United Kingdom, respectively. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. The JarqueeBera statistic is used to determine whether the data come from a normal distribution. The null hypothesis is normality. Table A2 Panel group-wise heteroskedasticity tests. Sectors

Biomass energy Solar energy Wind energy

Lagrange multiplier (LM) test

Likelihood ratio (LR) test

Statistics

p-Value

Statistics

p-Value

Wald test Statistics

p-Value

194.411 117.565 270.569

0.000 0.000 0.000

158.104 132.290 138.041

0.000 0.000 0.000

6330.618 1048.708 566.590

0.000 0.000 0.000

The null hypothesis is that there is homoskedasticity within cross-sectional units. Table A3 HS codes for renewable energy technologies and components. 6-Digit HS code

Product description

Biomass energy technologies 2207.10 Undenatured ethyl alcohol 2207.20 Ethyl alcohol and other spirits 3802.10 Activated carbon 3824.90 Other chemical products and preparations of the chemical or allied industries (including those consisting of mixtures of natural products); not elsewhere specified or included: other 7411.21 Tubes and pipes of copper-zinc base alloys (brass) 7411.22 Tubes and pipes of copper-nickel or copper-nickel-zinc base alloys 7411.29 Other tubes and pipes 8406.81 Steam turbines and other vapor turbines of an output exceeding 40 MW 8406.82 Steam turbines and other vapor turbines of an output not exceeding 40 MW 8411.82 Other gas turbines of a power exceeding 5000 kW 8416.20 Other furnace burners including combination burners 8419.31 Dryers: for agricultural products 8419.40 Distilling or rectifying plant 8419.89 Other machines and mechanical appliances for the treatment of materials by a process involving a change of temperature: other 8479.20 Machinery for the extraction or preparation of animal or fixed vegetable fats or oils 8501.61 AC generators (alternators) of an output not exceeding 75 kVA (kilovolt ampere) 8501.62 AC generators (alternators) of an output exceeding 75 kVA but not exceeding 375 kVA 8501.63 AC generators (alternators) of an output exceeding 375 kVA but not exceeding 750 kVA 8501.64 AC generators (alternators) of an output exceeding 750 kVA Wind energy technologies 7308.20 Towers and lattice masts 8412.90 Other engines and motors: parts 8482.10 Ball bearings

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

15

Table A3 (continued ) 6-Digit HS code 8482.20 8482.30 8482.40 8482.50 8482.80 8483.40 8501.61 8501.62 8501.63 8501.64 8502.31 8503.00 8504.21 8504.22 8504.23 8504.31 8504.32 8504.33 8504.34 8544.49 8544.60 8907.90 9028.30 9030.20 9030.31 9030.39 Solar energy technologies 7009.91 7009.92 7115.90 7304.31 7304.41 7304.51 7322.90 8306.30 8412.80 8419.19 8419.90 8504.40 8541.40 9001.90 9002.90 9005.80

Product description Tapered roller bearings Spherical roller bearings Needle roller bearings Other cylindrical roller bearings Other ball or roller bearings Gears and gearing, other than tooth AC generators (alternators) of an output not exceeding 75 kVA (kilovolt ampere) AC generators (alternators) of an output exceeding 75 kVA but not exceeding 375 kVA AC generators (alternators) of an output exceeding 375 kVA but not exceeding 750 kVA AC generators (alternators) of an output exceeding 750 kVA Other generating sets: wind-powered Parts suitable for use solely or principally with the machines of heading 8501 or 8502 Liquid dielectric transformers having a power handling capacity not exceeding 650 kVA Liquid dielectric transformers having a power handling capacity of 650e10,000 kVA Liquid dielectric transformers having a power handling capacity exceeding 10,000 kVA Electric transformers having a power handling capacity less than 1 kVA Electric transformers having a power handling capacity of 1e16 kVA Electric transformers having a power handling capacity of 16e500 kVA Electric transformers having a power handling capacity exceeding 500 kVA Other electric conductors, for a voltage not exceeding 80 V Other electric conductors, for a voltage exceeding 1000 V Other Electricity meters Cathode ray oscilloscopes Multimeters Other instruments and apparatus for measuring or checking voltage, current or resistance, with a recording device

Glass mirrors, unframed Glass mirrors, framed Other articles of precious metal or of metal clad with precious metals, other Tubes, pipes, and hollow profiles, seamless, of circular cross-section, of cold-drawn/cold-rolled (cold-reduced) steel Tubes, pipes, and hollow profiles, seamless, of circular cross-section, of stainless steel, cold-drawn/cold-rolled (cold-reduced) Tubes, pipes, and hollow profiles, seamless, of circular cross-section, of alloy steel other than stainless steel, cold-drawn/cold-rolled (cold-reduced) Radiators for central heating, air-heaters, hot air-distributors non-electric, other Photograph, picture or similar frames, mirrors, and parts thereof Other engines and motors Instantaneous or storage water heaters, non-electric Other machines and mechanical appliances for the treatment of materials by a process involving a change of temperature: parts Static converters Photosensitive semiconductor devices, including photovoltaic cells whether or not assembled in modules or made up into panels, light-emitting diodes Other (including lenses and mirrors) Other optical elements (including mirrors) Other instruments

Source: Ref. [2].

likely that policies that do not consider such differences would not be effective. To facilitate effective policy implementation, policymakers need to understand the sensitivity of exports and imports to the related policies at the subsector level. Thus, further research should explore this issue. Third, this study tested the effects of the degree of government support for exports, which limited our ability to draw implications for the development of efficient policy measures that encourage innovation and cost reduction. In fact, there remains substantial room for cost reduction and performance improvement due to the relative immaturity of RETs compared to fossil fuel alternatives. This is likely to continue to encourage governments to increase and strengthen their policy support. Under these circumstances, among a government’s most important policy issues would be the enhancement of the dynamic efficiency of its policies, namely, to stimulate and continue incentives for technological progress, innovation, and cost reduction [92,93]. Thus, further research efforts should be directed towards measuring the dynamic efficiency of such policy measures.

Acknowledgements This paper is financially supported by the National Research Foundation of Korea Grant funded by the Korean Government (NRF–2012S1AOAEA03).The authors thank the anonymous reviewers for thouthtful comments. References [1] Larsen H, Sønderberg PL. Energy technologies future options. Risø energy report 6. Denmark: Risø National Laboratory; 2007. [2] Jha V. Trade flows, barriers and market drivers in renewable energy supply goods: the need to level the playing field. Trade and environment issue paper 10. Geneva: International Centre for Trade and Sustainable Development; 2009. [3] REN21. Renewables global status report: 2010. Paris: Renewable Energy Policy Network for the 21st Century Secretariat; 2010. [4] BNEF. Global renewable energy market outlook. London: Bloomberg New Energy Finance; 16 November 2011. [5] IEA. World energy outlook 2012. Paris: International Energy Agency; 2012. [6] Shen Y-C, Chou CJ, Lin GTR. The portfolio of renewable energy sources for achieving the three E policy goals. Energy 2011;36:2589e98.

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

16

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17

[7] Sung B, Song W-Y. Causality between public policies and exports of renewable energy technologies. Energy Policy 2013;55:95e104. [8] Lior N. Sustainable energy development (May 2011) with some gamechangers. Energy 2012;40:3e18. [9] Arent DJ, Wise A, Gelman R. The status and prospects of renewable energy for combating global warming. Energy Econ 2011;33:584e93. [10] Lewis JI, Wiser RH. Fostering a renewable energy technology industry: an international comparison of wind industry policy support mechanisms. Energy Policy 2007;35:1844e57. [11] Lund PD. Effects of energy policies on industry expansion in renewable energy. Renew Energy 2009;34:53e64. [12] Guérin E, Schiavo J. Chinese renewable energy and technology policies: legal compatibility with WTO rules & economic interactions with other countries’ climate and industrial policies. Working paper no. 02/2011. Paris: Institut du Développement Durable et des Relations Internationales (IDDRI); 2011. [13] Lewis JI. Building a national wind turbine industry: experience from China, India and South Korea. Int J Technol Glob 2011;5:281e305. [14] De la Tour A, Glachant M, Méniére Y. Innovation and international technology transfer: the case of the Chinese photovoltaic industry. Energy Policy 2011;39:761e70. [15] Ru P, Zhi Q, Zhang F, Zhong X, Li J, Su J. Behind the development of technology: the transition of innovation modes in China’s wind turbine manufacturing industry. Energy Policy 2012;43:58e69. [16] Liu J, Goldstein D. Understanding China’s renewable energy technology exports. Energy Policy 2013;52:417e28. [17] Algieri B, Antonio A, Succurro M. Going “green”: trade specialization dynamics in the solar photovoltaic sector. Energy Policy 2011;39:7275e83. [18] Costantini V, Crespi F. Environmental regulation and the export dynamics of energy technologies. Ecol Econ 2008;66:447e60. [19] Groba F. Determinants of trade with solar energy technology components: evidence on the Porter hypothesis?. Discussion paper 1163. Berlin: German Institute for Economic Research; 2011. [20] Groba F. Environmental regulation, solar energy technology components and international trade e an empirical analysis of structure and drivers. Presented at the World Renewable Energy Congress 2011, Linköping, Sweden; 8e13 May 2011. pp. 3670e77. [21] Cao J, Groba F. Chinese renewable energy technology exports: the role of policy, innovation and markets. Berlin: German Institute for Economic Research; 2013. Discussion paper 1263. [22] Costantini V, Mazzanti M. On the green and innovative side of trade competitiveness? The impact of environmental policies and innovation on EU exports. Res Policy 2012;41:132e53. [23] Johnstone N, Has ci c I, Popp D. Renewable energy policies and technological innovation: evidence based on patent counts. Environ Resour Econ 2010;45: 133e55. [24] An G, Iyigun MF. The export technology content, learning by doing and specialization in foreign trade. J Int Econ 2004;64:465e83. [25] Jones M, Coviello NE. Internationalisation: conceptualising an entrepreneurial process of behaviour in time. J Int Bus Stud 2005;36:284e303. [26] Casillas JC, Moreno AM, Acedo FJ. Path dependence view of export behavior: a relationship between static patterns and dynamic configurations. Int Bus Rev 2012;21:465e79. [27] Schmeiser KN. Learning to export: export growth and the destination decision of firms. J Int Econ 2012;87:89e97. [28] Nadeau M-C, Kar A, Roth R, Kirchain R. A dynamic process-based cost modeling approach to understand learning effects in manufacturing. Int J Prod Econ 2010;128:223e34. [29] Smith PC, Street A. Measuring the efficiency of public services: the limits of analysis. J R Stat Soc 2005;168(Series A):401e17. [30] Nemet GF. Demand-pull, technology-push, and government-led incentives for non-incremental technical change. Res Policy 2009;38:700e9. [31] Neuhoff K. Implementing the EU renewables directive. Cambridge: University of Cambridge; 2009. Cambridge working paper in economics no. 0913. [32] UNCTAD. Investment and technology policies for competitiveness: review of successful country experiences. Geneva: United Nations Conference on Trade and Development; 2003. [33] Chaudhry IS, Malik A, Faridi MZ. Exploring the causality relationship between trade liberalization, human capital and economic growth: empirical evidence from Pakistan. J Econ Int Financ 2010;2:175e82. [34] Coe D, Helpman E, Hoffmeister A. North-South R&D spillovers. Econ J 1997;107:134e49. [35] Sun P, Heshmati A. International trade and its effects on economic growth in China. Bonn: The Institute for the Study of Labor; 2010. IZA discussion paper no. 5151. [36] Delmas M, Toffel MW. Stakeholders and environmental management practices: an institutional framework. Bus Strategy Environ 2004;13:209e22. [37] Delmas M, Montes-Sancho M. An institutional perspective on the diffusion of international system standards: the case of the environmental management standard ISO 14000. Bus Eth Q 2011;21:103e32. [38] Kaufman D, Kraay A, Mastruzzi M. Governance matters III: governance indicators for 1996e2002. Washington, DC: The World Bank; 2003. [39] Huang M-Y, Alavalapati J, Carter D, Langholtz M. Is the choice of renewable portfolio standards random? Energy Policy 2007;35:5571e5.

[40] Chang T, Huang C, Lee M. Threshold effect of the economic growth rate on the renewable energy development from a change in energy price: evidence from OECD countries. Energy Policy 2009;37:5796e802. [41] Sadorsky P. Renewable energy consumption and income in emerging economies. Energy Policy 2009;37:4021e8. [42] Dinda S. Environmental Kuznets curve hypothesis: a survey. Ecol Econ 2004;49:432e55. [43] Marques AC, Fuinhas JA. Drivers promoting renewable energy: a dynamic panel approach. Renew Sustain Energy Rev 2011;15:1601e8. [44] Sawhney A, Kahn ME. Understanding cross-national trends in high-tech renewable power equipment exports to the United States. Energy Policy 2012;46:308e18. [45] Frees EW. Assessing cross-sectional correlation in panel data. J Econ 1995;69: 393e414. [46] Pesaran MH. General diagnostic tests for cross section dependence in panels. Cambridge: University of Cambridge; 2004. Cambridge working papers in economics no. 0435. [47] Maddala GS, Wu S. A comparative study of unit root tests with panel data and a new simple test. Oxf Bull Econ Stat 1999;61:631e52. [48] Hadri K. Testing for stationarity in heterogeneous panel data. Econ J 2000;3: 148e61. [49] Breitung J. The local power of some unit root tests for panel data. Adv Econ 2001;15:161e77. [50] Choi I. Unit root tests for panel data. J Int Money Financ 2001;20:249e72. [51] Levin A, Lin CF, Chu C. Unit root tests in panel data: asymptotic and finitesample properties. J Econ 2002;108:1e24. [52] Im KS, Pesaran MH, Shin Y. Testing for unit roots in heterogeneous panels. J Econ 2003;115:53e74. [53] Carrion-i-Silvestre JL, Del Barrio-Castro T, Lópes-Bazo E. Breaking the panels: an application to the GDP per capita. Econ J 2005;8:159e75. [54] Pesaran HM. A simple panel unit root test in the presence of cross-section dependence. J Appl Econ 2007;22:265e312. [55] Pedroni P. Critical values of co-integration tests in heterogeneous panels with multiple regressors. Oxf Bull Econ Stat 1999;61:653e78. [56] Pedroni P. Panel co-integration: asymptotic and finite sample properties of pooled time series tests with an application to the PPP hypothesis. Econ Theory 2004;20:597e625. [57] Banerjee A, Carrion-i-Silvestre JL. Cointegration in panel data with breaks and cross-section dependence. Germany/Switzerland: European Central Bank; 2006. Working paper no. 591. [58] Westerlund J. Testing for error correction in panel data. Oxf Bull Econ Stat 2007;69:709e48. [59] Holtz-Eakin D, Newey W, Rosen H. Estimating vector autoregressions with panel data. Econometrica 1988;56:1371e95. [60] Harmonized commodity description of cording system. World Customs Organization; 1996. 1996 edition. [61] Bowman K, Schenton L. Omnibus test contours for departures from normality based on b1 and b2. Biometrika 1975;62:343e50. [62] Brown RL, Durbin J, Evans JM. Technologies for testing the constancy of regression relationships over time. J R Stat Soc 1975;37(Series B):149e92. [63] De Hoyos RE, Sarafidis V. Testing for cross-sectional dependence in panel-data models. Stata J 2006;6:482e96. [64] Banerjee A, Marcellino M, Osbat C. Some cautions of the use of panel methods for integrated series of macro-economic data. Econ J 2004;7:322e40. [65] Engle R, Granger CWJ. Cointegration and error correction: representation, estimation, and testing. Econometrica 1987;55:251e76. [66] Pedroni P. Fully modified OLS for the heterogeneous cointegrated panels. Adv Econ 2000;15:93e130. [67] Saikkonen P. Asymptotically efficient estimation of co-integration regressions. Econ Theory 1991;12:1e21. [68] Kao C, Chiang M. On the estimation and inference of a cointegrated regression in panel data. Adv Econ 2000;15:179e222. [69] Pesaran MH, Shin Y, Smith RP. Pooled mean group estimation of dynamic heterogeneous panels. J Am Stat Assoc 1999;94:621e34. [70] Breitung J, Pesaran MH. Unit roots and co-integration in panels. In: Matyas L, Sevestre P, editors. The econometrics of panel data: fundamentals and recent developments in theory and practice. USA: Kluwer Academic Publishers; 2008. pp. 297e322. [71] Banerjee A. Panel data unit roots and co-integration: an overview. Oxf Bull Econ Stat 1999;61:607e29. [72] Arellano M, Bond S. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev Econ Stud 1991;58:277e97. [73] Sargan J. The estimation of economic relationships using instrumental variables. Econometrica 1958;26:393e415. [74] Hansen L. Large sample properties of generalized method of moment estimators. Econometrica 1982;50:1029e54. [75] Roodman D. How to xtabond2: an introduction to difference and system GMM in Stata. Stata J 2009;9:86e136. [76] Haas R, Meyer NI, Held A, Finon D, Lorenzoni A, Wiser R. Promoting electricity from renewable energy sources e lessons learned from the EU, US and Japan. In: Sioshani FP, editor. Electricity market reforms. Amsterdam: Elsevier Publishers; 2008. pp. 419e68.

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082

B. Sung, W.-Y. Song / Energy xxx (2014) 1e17 [77] Midttun A, Gautesen K. Feed in or certificates, competition or complementarity? Combining a static efficiency and a dynamic innovation perspective on the greening of the energy industry. Energy Policy 2007;35:1419e22. [78] Cory K, Couture T, Kreycik C. Feed-in tariff policy: design, implementation and RPS policy interactions. Colorado: National Renewable Energy Laboratory; 2009. Technical report NREL/TP-6A2-45549. [79] Lesser JA, Su X. Design of an economically efficient feed-in tariff structure for renewable energy development. Energy Policy 2008;36:981e90. [80] Haas R, Resch G, Panzer C, Busch S, Ragwitz M, Held A. Efficiency and effectiveness of promotion systems for electricity generation from renewable energy sources e lessons from EU countries. Energy 2011;36:2186e93. [81] Chien T, Hu J-L. Renewable energy: an efficient mechanism to improve GDP. Energy Policy 2008;36:3045e52. [82] EREC. European renewable energy export strategy. Brussels: European Renewable Energy Council; 2003. [83] TPCC. Renewable energy and energy efficiency export initiative. Washington, DC: Trade Promotion Coordination Committee; 2010. [84] Jordan-Korte K. International markets for renewable energy technologies and export promotion policies. In: Jordan-Korte K, editor. Government promotion of renewable energy technologies e policy approaches and market development in Germany, the United States, and Japan. USA: Springer Fachmedien; 2011. pp. 139e200. [85] UNCSD, UNCTAD trade and green economy RIOþ20 issues brief no. 1 (March), United Nations conference on sustainable development and United Nations conference on trade and development; 2011.

17

[86] Ilias S, Hanrahan CE, Villarreal MA. US government agencies involved in export promotion: overview and issues for Congress. Washington DC: Congressional Research Service; 2013. CRS report for congress R41495. [87] Genctürk EF, Kotabe M. The effect of export assistance program usage on export performance: a contingency explanation. J Int Mark 2001;9:57e72. [88] Alvarez R. Sources of export success in small and medium-sized enterprises: the impact of public programs. Int Bus Rev 2004;13:383e400. [89] Lages LF, Montgomery DB. The relationship between export assistance and performance improvement in Portuguese export ventures: an empirical test of the mediating role of pricing strategy adaptation. Eur J Mark 2005;39:755e 84. [90] Kotabe M, Czinkota MR. State government promotion of manufacturing exports: a gap analysis. J Int Bus Stud 1992;23:637e58. [91] Marandu EE. Impact of export promotion on export performance: a Tanzanian study. J Glob Mark 1995;9:9e39. [92] Finon D, Menanteau P. The static and dynamic efficiency of instruments of promotion of renewables. Energy Stud Rev 2004;12:53e83. [93] Bökenkamp G, Hohmeyer O, Diakoulaki D, Tourkolias C, Porchia R, Zhu X, et al. WP 9 report on policy assessment of instruments to internalise environment related external costs in EU member states, via promotion of renewables. Flensburg: University of Flensburg. Cost Assessment of Sustainable Energy Systems (CASES); 2008. Project no. 518294 SES6.

Please cite this article in press as: Sung B, Song W-Y, How government policies affect the export dynamics of renewable energy technologies: A subsectoral analysis, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.03.082