The implications of international trade on the environmental Kuznets Curve

(1)

The implications of international

trade on the environmental Kuznets

Curve

A Thesis submitted in

partial fullment

of the requirements of the degree

M.Sc. Economics

at the University of Groningen - Faculty of Economics and

Business

by Bastien Haller

under the supervision of dr. J.P. Elhorst

18.06.2014

Master's Thesis Economics EBM877A20

Email b.haller@student.rug.nl

(2)

Abstract

The majority of the studies on the environmental Kuznets curve (EKC) consider production-based indicators for environmental quality, in this way they ignore the consequences of international trade. This study applies methods of input-output analysis to obtain CO2 consumption-based indicators to investigate the relationship between economic growth and CO2, using a spatial econometric framework. By making use of CO2 consumption gures, one can control for the shift of emissions via international trade. Furthermore, the traditional approach will be complemented by a structural decomposition analysis to gain an insight into the driving forces behind the EKC. The results show that an EKC can be obtained for consumption and production-based indicators. However, the former starts to decline at much higher levels of economic growth, indicating that rich countries consume still relatively large amounts of CO2, while the emissions emitted during the production process have already declined. The shift of emissions via international trade can account for the dierence.

Key Words: environmental Kuznets curve, input-output analysis, consumption-based, spatial econometrics, international trade

JEL classication: C21, O13, Q56, R15

Acknowledgements

(3)

1 Introduction

Climate change represents a serious global risk to society and economy. Increasing global average temperatures, melting of the polar ice caps and glaciers, and rising global average sea level indicate that the climate is already changing. The ecological and environmental consequences will be devastating if greenhouse gas (GHG) emissions continue to rise, whereby 84% of them consist of CO2, the largest component of the GHG. The emission of GHG in to the atmosphere disturbs the earth's radiative balance which leads to an increase of the surface temperature (OECD, 2014). Increasing temperatures pose a risk to the basic elements of life on a global scale such as the access to drinking water, health, food supply, and physical and natural capital. Inadequate recognition of the consequences by policy makers could have important implications on human well-being and economic growth (OECD, 2012).

Stern (2008) estimated that the costs of not taking action will correspond to at least 5% of global GDP each year, from today onwards. Though this would be the best case scenario, the incorporation of a wider range of risks and impacts could cause the damage to rise to 20% of global GDP or more, whereas the costs of reducing GHG emissions to avoid the worst outcomes of climate change would amount to only 1% of global GDP. Strong and early environmental policy decisions would outweigh the costs of not acting, while a delay of appropriate action could have irrevocable implications on the environmental quality, resulting in the turning point of the Environmental Kuznets curve (EKC) to disappear or recede into the distance (Bimonte, 2002).1

The United Nations Framework Convention on Climate Change (UNFCCC) Conference in Cancún in 2010 set the goal to reduce GHG emissions in order to limit the global average rise in temperature to two degrees Celsius (2°C) above pre-industrial levels; the level of 2°C represents a critical threshold for the Earth's ecological system. However, the worldwide CO2 emissions reached a level of 28.84 gigatonnes (Gt) in 2009 and are estimated to increase by 70% in 2050. This estimate would imply that the global average temperature is likely to exceed the prior specied goal of the UNFCCC by 2050; the temperature would be 3°C to 6°C higher than the pre-industrial levels (OECD, 2012).

Rising CO2 emission caused by high economic growth in some of the major emerging economies such as China and India are central to the debate. The prevalent approach to investigate the relationship between economic growth and the environment is the EKC. In its most basic form, an indicator for environmental degradation is regressed on income per capita and squared income per capita. The EKC theory predicts an inverted-U relationship between environmental deterioration and per capita income, i.e. environmental quality deteriorates up to a certain income level as the economy grows, but starts to improve again after the so-called turning point has been reached (Fig.1). For

(5)

the climate change debate, this could imply that the environmental pressure will decline after emerging countries have suciently grown (Dinda, 2004). The rst studies on this matter appeared in the early 1990s (Grossman and Krueger, 1991; Bandyopadhyay and Shak, 1992; Panayotou, 1993), but the literature is ever increasing even today. The relationship between economic growth and changes in the environment is still drawing the attention of academics and policymakers, and it does not seem to diminish, because of its relevance regarding questions about the climate change, the scarcity of raw materials and other environmental problems (Kaika and Zervas, 2013).

Figure 1: Patterns of the EKC (Wagner, 2010)

The driving forces behind the shape of the EKC are, according to empirical literature and theoretical models, an increasing demand for environmental quality, international trade, technological progress, structural change towards service based-activities, envir-onmental regulations and market mechanism (Dinda, 2004). The topic of international trade requires greater attention, since trade allows a country to partly disconnect its do-mestic consumption from its ecological system by importing pollution-intensive products. Environmental quality deteriorates in the exporting country and not in the importing country where the consumption takes place (Dietzenbacher and Mukhopadhyay, 2007). International trade illustrates a shifting of environmental problems from developed to developing countries, rather than a solution. The majority of the studies acknowledge trade as a possible explanation for the EKC, but they do not pay further attention to its implications. The amount of CO2 emissions embodied in international trade cannot simply be ignored; they amounted to 3.46 Gt in 2009, which corresponds to 12% of the global emissions.

(6)

ter-ritorial emissions. The geographic separation of consumers and pollution emitted during the production process makes it possible to shift emissions to a remote country. This may be considered a rational option for local pollutants, since it allows a country to pursue an unsustainable consumption without bearing the consequences, but it is certainly not reasonable for global pollutants such as CO2; people worldwide will bear the consequences of climate change regardless of the country of production (Peters et al., 2007). This is aggravated by the fact that the net eect of shifting emissions via international trade has negative consequences on the global environment, since the developed country could produce the same products in a more environmentally friendly manner.

Naturally, one would except that the design of national and international environmental regulations would take into account the implications of international trade. However, the majority of proposals for appropriate climate policies consider emissions trade schemes, technology cooperation, reforms of fossil fuel support subsidies and carbon pricing (Stern, 2008; OECD, 2012), and only a few proposals paid direct attention to the consequences of international trade (Peters et al., 2011).

The Kyoto Protocol2 _{and the Doha Amendment to the Kyoto Protocol}3 _{focus solely} on the reduction of territorial greenhouse gas emissions. The implications of trade are only briey mentioned in the second article of the Kyoto Protocol, which states that the participating parties shall implement measures to reduce their emissions in such a way that the adverse eects of international trade on other parties are minimized, but they are not obliged to publish corresponding numbers. In addition, the Kyoto protocol recognizes that developed countries are mainly responsible for the high levels of greenhouse gas emissions as a consequence of over a 150 years worth of industrial activity, therefore the reduction commitments are imposed only on them. While developing countries do not have to comply with such commitments to allow for economic growth. It seems to be a fair, and reasonable regulation considering the EKC theory, but it may facilitate the shift of pollution-intensive production to developing countries.

In the majority of developed countries CO2 emissions have stabilized or even reduced, whereas developing countries reported steeply increasing numbers. Even if the domestic emissions have reduced, the global CO2 consumption either increased or declined at a much slower pace, suggesting that the reduction in domestic emissions was partly driven by international trade at the expense of developing countries (Peters et al., 2011).

CO2 consumption shows the emission embodied in the domestic nal demand of a country; irrespective of where the polluting activity initially took place, thereby it is possible to control for the implications of international trade (Arto, 2012). The dierence between the domestic CO2 emissions and the CO2 consumption reect the emissions embodied in international trade (Fig.2). Numbers on CO2 emissions embodied in trade

(7)

Figure 2: Trade causes the dierence between an EKC for consumption and production (Wagner, 2010)

and consumption are usually not readily available nor ocially calculated, therefore they play only a minor role in climate policy discussion, despite their importance (Wagner, 2010).

This paper estimates an EKC for domestic CO2 emissions and consumption. CO2 emissions lead to environmental degradation at the place of production and not neces-sarily at the place of consumption, for this reason it may be possible that there is only evidence for a production-based EKC. Furthermore, by comparing the production and consumption-based EKC, one can investigate the consequences of trade on the pollution income relationship and whether economic growth leads to a sustainable CO2 consump-tion. If not, policy makers should act by taking into account the implications of interna-tional trade in their environmental policy decisions, in addition to the current regulations on territorial emissions. Eventually, environmental quality will only improve worldwide if CO2 consumption approaches sustainable dimensions.

(8)

vari-ables and geographic location, therefore spatial econometric methods could be extremely benecial, since they allow to incorporate spatial dependence in the regression analysis.

To the extent of my knowledge, this is the rst paper using a consumption-based approach on the EKC, apart from a few papers using the ecological footprint; an aggregate measure of environmental quality (Bagliania et al., 2006; Caviglia-Harris et al., 2008; Wang et al., 2013), and a study from Wager (2010) on energy use. Wagner (2010) uses constant 1995 US energy intensities, i.e. the amount of energy required for one unit of output, to calculate energy consumption gures. Leading him to indirectly assume that the energy intensities are the same for all the countries included in his study and do not change over time. These assumptions are unlikely to hold, and more importantly, they do not account for less advanced production technologies in developing countries, and ignore technological improvements which could have led to a reduction in the energy intensities and are important determinants for the shape of the EKC.

Eventually, a structural decomposition analysis is adapted to shed some light on the driving forces behind the changes in the environment, thereby responding to the cri-ticism of Stern (2004); that reduced-form models alone are not suitable to formulate policy recommendations, since only income is used as a catch-all variable. The struc-tural decomposition analysis and the calculations of CO2 consumption are based on the input-output methods developed by Leontief (1970). The basic idea is that the changes in CO2 emission levels can be decomposed into their single elements, i.e. technological progress, structural progress, domestic nal demand and exports. This approach is a useful complement to the traditional EKC approach, since one is able to quantify some of the most important determinants of the EKC.

The plan of the paper is as follows. Section 2 gives an overview of the literature on the EKC with a focus on the possible causes and the criticism on the existing approaches. Section 3 illustrates the employed data and gives some insights into the trade and CO2 consumption patterns of the countries included in this study, while section 4 presents the underlying models and methods. The empirical analysis and the discussion of the results are carried out in section 5. Finally, section 6 includes the conclusion and a discussion for future research.

2 Literature review

2.1 Origin of the EKC

(9)

curve, implying that the income distribution is less equal in the early stages of economic growth, but becomes more equal after a certain level of per capita income is reached. His result is generally referred to as the Kuznets Curve. In the 1990s the Kuznets Curve had a revival in the context of environmental quality and economic growth; it was suggested that environmental degradation rises with income up to a threshold level, referred to as the turning point, and that beyond this level environmental quality starts to improve with income (Lieb, 2004).

The relationship between economic growth and environmental quality has been an im-portant topic in economic research for many years. In 1972, the Club of Rome's Limits to Growth launched a huge discussion about the sustainability of economic growth. They argued that the niteness of environmental resources would constrain the possibilities for economic growth. For this reason they demanded a steady-state economy with zero growth, to avoid that the world economy reaches its physical limits regarding nonrenew-able resources and excessive pollution (Kaika and Zervas, 2013). The view was soon challenged on empirical and theoretical grounds. Auty (1985) found evidence that the relationship between the intensity of metal use and income has the shape of an inverted-U curve. His result is known as the intensity-of-use hypothesis and illustrates that the con-sumption to income ratio of some metals was declining for developed countries; thereby it opposed the view of the Club of Rome (Dinda, 2004).

In the 1990s data on several pollutants became readily available through the Global Environmental Monitoring System, the environmental data of the OECD and the Oak Ridge National Laboratory. The wide range of new data triggered a whole series of empirical studies on the validity of the inverted-U curve relationship between income and environmental degradation (Dinda, 2004). Three working papers on the EKC appeared independently, in close succession. Grossman and Krueger (1991) were the rst who found evidence for an EKC for SO2. In the following years Bandyopadhyay and Shak (1992) and Panayotou (1993) found further evidence for an inverted-U relationship for local air pollutants and deforestation, whereby the term Environmental Kuznets Curve was coined rst by Panayotou (1993).

Over the years a large body of research on the EKC has emerged, and the following subsections will give an overview of the most important studies which are relevant for this paper.4 _{The literature review starts with the results of the empirical studies, followed by} the possible causes for the EKC. The last two subsections cover the theoretical models and the criticism on the EKC approach.

4_{The reader is referred to Dinda (2004), Lieb (2004) and Kaika and Zervas (2013) for surveys on the}

(10)

2.2 Empirical Evidence

Empirical EKC studies examine the eect of economic growth on several indicators for environmental degradation, and the data is usually in the form of cross-sections of coun-tries, cross sectional panels and time series. Three main categories of indicators can be distinguished: air pollutants, water pollutants and other environmental degradation indicators (Dinda, 2004).

The inverted-U relationship is well established for local air quality indicators, such as sulphur dioxide SO2, suspended particulate matters (SPM), oxides of nitrogen NOx and carbon oxide CO (Grosmann and Krueger, 1991; Selden and Song, 1995; Wu, 1998; Lieb, 2004). The existence of an EKC for local air pollutants is not an unexpected result, since they are relatively easy to detect and have an immediate eect on the local environment, and health of nearby living citizens. Furthermore, their abatement costs are comparatively low. It seems that the pollutants with the highest short-term risks to health, such as SO2 and SPM, are addressed at the lowest levels of income (Lieb, 2004). In contrast, the support for an EKC for global air pollutants such as CO2 emissions, which are characterized by a relatively small direct impact on health, is small. Coondoo and Dinda (2008), Aslanidis and Iranzo (2009) and Iwata et al. (2011) nd evidence for a monotonic increasing relationship between CO2 and income, whereas Cole (2004) and Lee et al. (2009) nd evidence for an inverted-U and N-shaped relationship, respectively. The empirical studies on water pollutants do not come to a clear result about the shape of the curve. The concentration of pathogens in water, amount of heavy metals and degradation of water oxygen regime are commonly used as measures for water quality (Lieb, 2004). Some studies nd a signicant EKC, but they reveal large dierence in the turning points (Cole et al., 1997 and Grossman and Krueger, 1995), while other studies do not nd evidence for signicant pollution income relationship (Bandyopadhyay and Shak, 1992;5 _{Vincent, 1997). However, Bandyopadhyay and Shak (1992) found a} N-shaped relationship for fecal coliform which counts as a pathogen. The explanation they oer for the nal upturn of the curve, is that individuals living in high income countries are less concerned about the quality of rivers, because they are not directly dependent on them for water, due to the improvements in the water supply systems.

Other widely used environmental indicators for EKC studies are access to safe drinking water, municipal solid wastes and urban sanitation. The available evidence seems to suggest that environmental problems which pose a high risk on the human health, such as the lack of clean drinking water and the lack of sanitation, tend to decrease with income or to have very low turning points. Those problems are addressed as soon as the necessary resources become available (Bandyopadhyay and Shak, 1992; Lieb, 2004; Dinda, 2004). However, municipal waste seems to increase steadily with income; the

(11)

curve does not fall for even high levels of income (Dinda, 2004). Bandyopadhyay and Shak (1994) explain this result by the fact that waste can be deposited outside the cities where nobody, or only poor people with little political inuence, live.

Wu (1998) argued that environmental quality is perceived as a whole by the society and cannot be measured by a single indicator for pollution. For this reason, a few studies started to employ an aggregate measure for environmental quality such as the ecological footprint. The ecological footprint was developed by Mathis Wackernagel (1992) for his PhD dissertation under the supervision of William Rees. This footprint measures the amount of land and water needed to produce the resources, and to absorb the waste generated by an individual's consumption. Footprints belong to the consumption-based measures of environmental quality. The majority of the studies used production-based indicators despite the fact that recent research suggested that consumption-based indic-ator can give a new insightful perspective (Rothman, 1998).6 _{Rothman (1998), Bagliani} et al. (2008) and Harris et al. (2009) do not nd evidence for an inverted-U relationship between income and the ecological footprint.

2.3 Spatial econometric approaches

The majority of EKC studies implicitly assume that a region's environmental quality is independent from the pollution in its neighboring countries. This view has been recently challenged by new evidence showing that environmental degradation has a spatial dimen-sion. However, spatial econometric approaches still depict a minority (Aslanidis, 2009). This section discusses the few studies employing a spatial econometric framework, while the advantages of these methods are illustrated in the criticism of the EKC and in the section about spatial econometrics.

Rupasingha et al. (2004) were one of the rst using a spatial econometric framework; they investigated the relationship between per capita income and toxic pollution for 3,029 US counties. Applying a spatial error model, they found evidence that the decline in pollution is only temporary and starts to rise again for higher levels of income, i.e. an N-shaped relationship. Maddison's (2006) approach was more sophisticated from a spatial econometric point of view and has been a motivation for future studies in this eld. His paper focused on the investigation of an EKC for local air pollutants. He discriminated between a spatial error and a spatial lag model, on the basis of robust LM tests, whereby the tests were in favor of the spatial lag model. Moreover, he considered four dierent specications of the spatial weight matrix. Although he did not nd any empirical proof for an inverted-U relationship, he discovered that a country's SO2 and NOx emissions are inuenced by the emissions of its neighboring countries; his results suggest that NOx emissions are reduced by adjacent high income countries.

(12)

McPherson and Nieswiadomy (2005) studied an EKC for threatened birds and mam-mals for 113 countries. They found evidence for an EKC curve and spatial autocorrela-tion, while Tevie et al. (2011) did not nd supporting evidence for such a relationship for biodiversity risk in the US. However, the spatial lag eect is highly signicant in their model, which suggests that biodiversity risk in one state can spill over to bordering states; cooperation between the federal states is necessary to eectively reduce the biodiversity risk.

Wang et al. (2013) investigated the ecological footprint for consumption and pro-duction, using a spatial Durbin model. They did not nd supporting evidence for an inverted-U relationship between the ecological footprint and income, but their paper gave new insights into the spillover eects of environmental quality. Their ndings indic-ate that the ecological footprint is inuenced by income, biocapacity and the ecological footprint in neighboring countries. Donfouet et al. (2013) and Burnett and Bergstrom (2010) employed a dynamic spatial panel data approach to control for time and space in their studies about Europe and the US. They both found proof for an EKC and show that CO2 emissions are positively inuenced by bordering regions.

2.4 Possible explanations for the EKC

The EKC reects a dynamic process, as income starts to rise over time, most environ-mental quality indicators deteriorate initially, but after income has reached its threshold level, environmental quality starts to improve. The EKC hypothesis suggests that it is possible for a country to grow out of its environmental problem. However, the process should not be considered as automatic, enhancement of the environmental quality does not come by economic growth itself (Kaika and Zervas, 2013). The following subsections give an overview of the driving forces behind the EKC. One should note that the driving forces are often closely related, and in reality it is not always easy to distinguish them from each other.

2.4.1 Income elasticity of environmental quality and consumer preferences

(13)

of purchases of more environmentally friendly products and donations to environmental organizations (Roca, 2003).

Martini and Tiezzi (2013) found evidence that the income elasticity of the willingness to pay for environmental quality is close to unity, across all income groups. According to their ndings environmental quality is a normal good, i.e. the demand for environmental quality increases proportional to income. An EKC can still be obtained, since an income elasticity of environmental quality greater than one is not a necessary condition for the existence of an EKC (Lieb, 2002). Consumer preferences may play an important role for environmental quality, since without consumers exercising sucient environmental eort, the government would not impose stricter regulations and companies would not have the necessary incentives to adapt new abatement technologies.

However, even the high-income households in the United States have not reached an income level at which their demand for environmental quality is sucient to trigger improvements in the environmental quality. Moreover, consumer preferences can depend on spatial and time conditions. For instance, residents of a city may not be concerned about health eects of waste deposited in a less populated area outside the city, or about pollutants which harm the environment in a remote future. In such circumstances environmental improvements occur only if individuals genuinely care about the whole environment (Khanna and Plassmann, 2004).

2.4.2 Equality Considerations

(14)

as a consequence the turning point of the EKC could disappear. According to Bimonte (2002) a more equal distribution of income, education and better access to information can increase citizens' concerns about environment and make growth sustainable.

2.4.3 Government and regulations

The government's willingness to implement stricter environmental standards is an im-portant factor in explaining the pollution income relationship. Panayotou (1997) argues that when the economy starts to grow, the government will respond to the increasing demand for environmental quality by imposing new regulations to reduce pollution. A high quality of government and public institutions can help to reduce environmental de-gradation, at low income levels. Institutional variables such as secure property rights, eective environmental regulations and the better enforceability of contracts can atten the EKC, thereby reducing the cost that economic growth imposes on the environment. Furthermore, public spending on environmental R&D rises with income, and though this alone may not account for large changes in the environment, it can serve as a positive ex-ample for the private sector, to spend more on the development of abatement technologies (Komen et al., 1997).

Dutt (2009) claims the reduction of environmental degradation in high income countries is not an automatic process. Rather, better socioeconomic conditions, a greater awareness about the environment and a greater demand for environmental quality, along with better institutions and governance, are responsible for improvements. He provides evidence that countries characterized by favorable socioeconomic conditions, higher investment in education and strong political institutions have lower emission levels. Bhattarai and Hamming (2001) and Culas (2007) results show that institutional factors have more explanatory power for the tropical deforestation process than economic and population growth. Stronger political institutions and regulations could help to atten the EKC for deforestation. However, Panayotou (1997) acknowledges that the quality of a government and institutions are dicult to measure.

(15)

2.4.4 International Trade

International trade is considered one of the most important causes for the EKC. Trade allows a country to partly separate its domestic consumption from its ecological sys-tem, by importing products from other countries. The Heckscher-Ohlin theory predicts that countries with an extensive allowance for emissions have a comparative advantage in producing emission intensive goods, leading to a deterioration of the environmental quality in developing countries. In other words, developed countries with strict regula-tions shift partly their production of dirty products to developing countries with weaker environmental standards. This phenomenon is commonly referred to as pollution haven hypothesis (PHH) or the displacement eect (Dietzenbacher and Mukhopadhyay, 2007). The import of pollution-intensive goods in developed countries can explain the downward sloping part of the EKC, whereas the exports of those goods explain the upward sloping part in developing countries (Kaika and Zervas, 2013).

Cole (2004) and Kearsley and Riddel (2010) do not nd strong evidence in favor of a strong inuence of the PHH on the EKC. Cole (2004) argues that if pollution havens exist, they are likely to be temporary and limited to certain sectors and regions. Cave and Bloomquist (2008) investigate whether the implementation of stricter environmental standards in the European Union led to increased imports of dirty products from devel-oping countries. They nd evidence in favor of the PHH for energy intensive products, but not for toxic intensive goods. The results of Peters et al.'s (2011) input-output analysis for the US suggests that the net eect of shifting emissions through international trade has a negative inuence on the global environmental quality. Simply put, the emissions embodied in the imported dirty goods exceed the emissions if they would have manufac-tured the products themselves. However, they did not take into account that developing countries replace a part of their domestic productions by cleaner imports from developed countries. Dietzenbacher and Mukhopadhyay (2007), in contrast, took that it into ac-count, when they examined the PHH for India. Their results show that the decrease in pollution due to additional clean imports outweighs the pollution caused by the addi-tional dirty exports, indicating that the environmental quality in India improved as a result of a higher trade volume.

(16)

the environmental quality through the transfer of ecological products and production processes (Letchumanan and Kodoma, 2000). Environmental institutions are either weak or absent in emerging nations, but international communities such as the World Bank can help to improve the environmental quality by supporting policy reforms, providing information and public environmental education. Therefore the turning point of the EKC can be decreased. Independent nancially supported research programs may play an important role in understanding the consequences of environmental changes and can help to enforce environmental regulations by providing information about polluters, local environmental quality, abatement technology, etc. (Dasgupta et al., 2002).

The environmental race to the bottom theory can explain the possibility of an N-shaped pollution income relationship. Strict environmental regulations in developed countries impose high cost on polluters, thereby creating an incentive for highly polluting industries to relocate in nations with lower environmental standards. The increasing capital outows force governments to relax their regulations triggering a race to the bottom, and as a consequence, the pollution starts to rise again in high income countries (Dasgupta et al., 2002).

2.4.5 Structural changes and technological progress

Many authors consider structural change and technological progress as the driving forces behind the EKC (Grosmann and Krueger, 1991; Shac and Bandyopadhyay, 1992; De Bruyn et al., 1998; Panayotou, 2003). Structural change describes the gradual transition from pollution-intensive industrial sectors to a more environmentally friendly service in-dustry. Initially a country produces relatively clean agricultural products, but as the economy starts to grow, the production shifts to pollution-intensive industrial products and eventually to clean service-based activities. The transition process is referred to as the composition eect, whereas the technique eect describes technological progress that leads to improvements in the production process, which can either be advances in abate-ment technologies or a more ecient use of inputs. The developabate-ment of new technologies is triggered by investments in R&D, which requires a suciently large level of income; technological progress arises as the result of economic growth (Komen et al., 1996; Kaika and Zervas, 2013).

(17)

Figure 3: Structural change and technological progress (Kaika and Zervas, 2013) technique eect will nally overcome the negative changes in the environment (Fig.3). That leads to the downward sloping part of the EKC (Dinda, 2004).

De Bruyn et al. (1998) argue that the decline in emission levels can be explained better by technological progress and structural changes as opposed to by economic growth itself. Their results suggest that the decrease in the pollution level of SPM and SO2 can be attributed to changes in the underlying production structure and to technology, whereas technological innovations represent the dominant factor. However, one should note that even though reduction of local air pollutants may be the result of technological progress and structural changes, this result reects the political and economic conditions of the country at the time of the research (Grossmann and Krueger, 1995).

2.4.6 Market mechanism

(18)

re-source prices reect their relative scarcity. The following price increase stops the heavy exploitation of the environment. In addition, higher prices create a nancial incentive for companies to invest in more ecient production technologies (Dinda, 2004). Unruh and Moomaw (1998) point out that the rising oil prices in the 1970s led to a shift to alternative energy sources. Market signals may be an alternative explanation for the EKC.

2.5 Theoretical foundation of the EKC

The EKC is mainly regarded as an empirical phenomenon in the literature. Since em-pirical research on the EKC utilizes reduced-form rather than structural form models, the relationship between environmental degradation and income cannot be suciently explained. For this reason, it is important to develop theoretical models to clarify under which circumstances an EKC arises. Theoretical models and stylized facts are necessary to improve the relationship between theoretical and empirical analysis, and to overcome shortcomings of the reduced-form models (Dinda, 2004). There are an immense number of models which describe the pollution income relationship. The following models are chosen to give an overview of the variety topics, and is not necessarily a complete list of all theoretical work on it.7 _{The pollution income relationship in theoretical models is} usually explained by behavioral changes, preferences, institutional changes, and techno-logical progress. All that the models have in common is that they describe a pollution caused trade-o (Kijima et al., 2010).

López (1994) was one of the rst who developed a theoretical model to describe the EKC. He considers environment a production factor. The inuence of economic growth on environmental quality depends crucially on the existence of productive stock feedback eects on the natural resource8 _{or if producers internalize the externality. In case the} producers are obligated by the government to internalize their negative externalities, economic growth will improve the environmental quality. Given that preferences are non-homothetic, which means the valuation of environmental quality by consumers increases with income, an inverted-U shaped relationship between environmental degradation and economic growth is obtained.

McConnel (1997), Andreoni and Levinson (2001) consider relatively simple models, and they both do not include a production function. They focus mainly on the trade-o between marginal utility of consumption and disutility of pollution. McConnel (1997) investigates how the income elasticity of the demand for environmental quality inuences pollution. The income of a representative agent can be spent on either consumption or pollution abatement. His simple model shows that the EKC cannot be necessarily

7_{Kijima et al, (2010) give a detailed overview of a large number of theoretical studies.}

8_{If the intensive exploitation of a resource in the short-run leads to a decrease in productivity in the}

(19)

explained by a high income elasticity of demand alone. Preferences that correspond to a high income elasticity of demand for environment quality can be attenuated by high abatement costs, or by a feedback eect from pollution on production as in the paper by López (1994). Lieb (2002) extends McConnel's (1997) model, whereby he was able to show that normality9 _{of environmental quality is a necessary condition to obtain an} EKC.

Andreoni and Levinson (2001) derive an inverted-U relationship between pollution and income by assuming increasing returns in the technological link (i.e. abating pollution) between consumption and pollution, whereby pollution is considered as an undesired byproduct of consumption. The model is extended by Dinda (2005) who includes en-vironmental quality as a factor of production. The EKC arises as the consequence of a dynamic process. Capital is divided into two parts; one part is used for developments in the environmental sector for pollution abatement and the other part is used in the production process which creates pollution.

The former models focus on pollution and abatement technologies and they underlie assumptions of low factor mobility, and environmental quality as a variable in the utility function. Both assumptions are not always conrmed by empirical evidence, that is, in particular the case for long-lived pollutants (e.g. hazard waste), which are not easily shiftable and for regional cross-sections with free labor mobility (Gawande et al., 2001). Gawande et al. (2001) develop a consumption-based model for long-lived pollutants, which is a general spatial equilibrium model based on labor mobility and location decisions by households. The inverted-U relationship is the result of distancing behavior rather than abatement technologies.

The previous models were static, except for Dinda (2005). Lieb (2002) pointed out that the EKC is an inherently dynamic phenomenon; therefore dynamic models might be more adequate to describe the complex relationship between environment and income. The following models are all dynamic.

One of the rst dynamic models was developed by John and Pecchenino (1994), what they consider an overlapping generation (OLG) model. A representative agent can spend his income for consumption and pollution abatement, both of which determine his lifetime utility. Given that income increases over time due to capital accumulation, later gener-ations can spend more of their income on environmental quality and as a consequence, an EKC is obtained. Selten and Song (1995) are following a similar approach, but they are using a continuous-time model with innitely living agents. Mariani et al. (2010) study an OLG model in which they describe the interrelation between health and envir-onment. Agents who expect a long life-span are more concerned about future impacts on the environment, therefore they spend more on environmental quality; by including human capital in the production function an EKC can be obtained.

(20)

Lieb (2004) extends the model from John and Peccenino (1994) for two dierent types of pollutions, namely stock10 _{and ow pollutions}11_{. He gives a theoretical explanation} for the empirical ndings that the pollution income relationship of ow pollutants follows an inverted-U relationship, whereas stock pollutants are monotonically increasing with income.

Endogenous growth models are employed by Stokey (1998) and Hartman and Kwon (2005). Stokey (1998) uses an AK model and a neoclassical growth model with exogen-ous technological change, while Hartman and Kwon (2005) explore a pollution extended version of the Uzawa-Lucas models in their paper. They assume that human capital, in contrast to physical capital, creates no pollution and physical capital can be substituted by human capital. In the long-run human capital will grow faster than physical capital, which leads to a decline in pollution.

The eectiveness of public policy measures is investigated by Egli and Steger (2007); by assuming increasing returns in abatement, they come to the result that it is optimal to subsidize abatement eort instead of taxing pollution. The model also oers an op-portunity for an N-shaped EKC.

Pyndick (2007) claims that the former deterministic models are not able to capture the existing uncertainties in environmental changes, thereby he triggered a range of real option approaches (Di Vita, 2007; Kijima et al., 2011). Uncertainties can exist regarding the future cost and benets of environmental policy. For instance, environmental damage caused by global air pollution does not necessarily follow a linear relationship. Moreover, environmental policies can be inuenced by uncertainties regarding the discount rate and the long time horizon of the decision. Di Vita (2007) delivers an intuitive explanation for why pollution abatement policies are delayed to later stages of economic development. It is assumed that the marginal return of capital is high at low income levels and decreases at higher income levels. Additionally, a country is only willing to adopt new investments in environmentally friendly projects if the interest rate becomes lower than the discount rate.

Stern (2004) criticizes that many environmental economists treated the EKC as a stylized fact that needs to be explained by theory, although the EKC has never been established for all pollutants. His argument is to oppose the fact that recent research considers the EKC just for certain types of pollution and allows for monotonic increasing, or even N-shaped relationships. Nevertheless, one should keep in mind that it is relatively easy to obtain an EKC under the appropriate assumptions (Stern, 1998).

The huge variety of dierent explanations shows that there is still no agreement about what eventually determines the relationship between income and pollution. Further re-search on the theoretical and empirical side is needed to fully explain this relationship.

(21)

2.6 Criticism of the EKC studies

The EKC approach attracted a lot of criticism for several reasons. Many authors chal-lenged the fundamental assumptions of the EKC studies, while others called into question the empirical methods. These criticisms should particularly be kept in mind for policy recommendations and for the interpretation of the results, since it is tempting to draw conclusions which are too strong.

2.6.1 Distribution of the world income

One aspect all the EKC studies have in common is that they use economic growth as the most important variable to explain environmental degradation. Economic growth is usually measured by GDP per capita which reects the average income of a representative agent. Studies usually draw data on income and environmental quality from a pool of dierent countries, because of the lack of appropriate data. The estimated turning point, at which environmental degradation starts to decline, is at the same income level for each country included in the study. For this reason, the world income should be normally distributed to give the turning point a meaningful interpretation (Kaika and Zervas, 2012).

However, this assumption is likely to be violated, since the world income is highly skewed and the majority has less than the world's average income (Stern, 2004). Kaika and Zervas (2012) show that about 83% of the world's population had an income below the estimated world average of all countries GDP's per capita, in 2005. This implies that the income level at which environmental quality starts to improve cannot be reached by a large share of the representative agents. As a consequence the estimated turning point is of little importance, especially for low-income countries. This raises the question whether GDP per capita is an appropriate measure for economic growth in EKC studies, hence Stern (1996) suggested the use of the median income instead.

2.6.2 Generalizability of the results the EKC

(22)

to repeat the same growth-pattern as developed countries, since those world powers already exist.

The PHH seems to support the reasoning of Roberts and Grimes (1997). Developing countries will start to implement stricter regulations once they have suciently grown and their environmental degradation has reached critical levels. However, the new regulations will tempt pollution-intensive industries to shift their production to countries with lower environmental standards, but as Stern (2004) points out it will become more dicult for the poor countries of today, as they become wealthier, to shift their production of emission-intensive goods to other countries. In the case that developing countries want to adopt similar environmental standards as developed countries, they have to face the more dicult challenge to abate their pollution instead of outsourcing their production. The problem is intensied by the fact that the pollution in developing countries is at least partly created by the imports to developed countries. Hence, it is unlikely that they will follow the same EKC-pattern as today's developed economies (Cole, 2004).

Although, it is not very likely that developing nations follow the historical experience of developed nations, there are new paths which could lead to an EKC. As mentioned before, developing countries can benet from new environmental friendly technologies through FDI, which they could not aord to develop themselves. Furthermore, the threat of global warming has increased mutual eorts to reduce greenhouse gases worldwide (Panayotou et al., 2000).

Another criticism concerns the fact that the EKC cannot be generalized for all types of pollutants. The EKC seems to only be valid for local pollutants with relatively low abatement costs and immediate negative eect on human health, while global pollutants with long-term eect and little impact on health do not follow an inverted-U curve (Arrow et al., 1995; Lieb, 2004; Dinda, 2004). This illustrates that economic growth alone cannot be the solution for all environmental problems. It also casts doubt on the explanation that the demand for environmental quality rises with income, or to put it dierently, it shows that consumer preferences depend on spatial and time conditions (Arrow et al., 1995; Khanna and Plassmann, 2004). Moreover, the studies seem to reect the unique characteristics of the examined pollutants, i.e. spatial, health and abatement costs features. Even if some pollutants may improve, it does not imply that other pollutants will follow the same pattern. Since environmental degradation does not depend on a single pollutant, it is possible that environmental quality declines despite improvements in some local pollutants (Arrow et al., 1995).

2.6.3 Consumption pattern

(23)

side of the economy. Rothmann (1998) claims that consumption is the ultimate driving force behind changes in the environment. As noted above, trade gives the opportunity to separate domestic consumption from consequences of production on the local environ-ment. Therefore, consumption-based approaches are advantageous, since they allow us to control for the PHH.

Cole (2004) suggests analyzing the changes in the income elasticity for pollution-intensive products. When environmental quality improves but the income elasticity does not decline, the demand for pollution-intensive goods is most likely satised by imports. Furthermore, if the income elasticity for ecologically damaging products remains constant, any positive changes in the environment caused by structural changes or technological progress will be oset by an increased demand for ecologically damaging goods (Kaika and Zervas, 2013). Input-output methods as developed by Leontief (1970) prove to be very useful in that regard. They allow us to decompose changes in emissions into their single components, i.e. structural changes, technological progress and changes in the nal demand. It should be noted that a decline in global environmental degradation can only arise from changes in the consumption patterns combined with technological progress and not by trade itself (Dinda, 2004).

Despite considerable progress towards sustainable economic growth in developed coun-tries, the consumption patterns remain unsustainable as shown by increasing municipal waste and CO2 emissions (Panayotou, 2003). Weber (2010) estimates a consumption-based and a production-consumption-based EKC for oil and total energy use, and obtains interesting results. He nds that high income countries use relatively less energy for their indus-trial production, while the consumption of energy-intensive goods is still high. He con-cludes that trade of pollution and energy-intensive goods cause the dierences, while the consumption patterns in developed countries remained unchanged. EKC studies should consider consumption and production-based pollutants to have two complementary views on the relationship between economic growth and environmental degradation (Kaika and Zervas, 2013).

2.6.4 Structural changes and services

(24)

eco-nomy.12 _{In contrast, time-series studies for developed countries are not able to reect} the gradual transition. Service-based activities already represented the largest share of the total GDP in the 1970s (Kaika and Zervas, 2013). Nevertheless, the problem can be solved by longer time series.

Services are believed to be environmental friendly since they are information-based and require fewer raw materials. But on the contrary, the service sector includes a variety of activities which demand large amounts of energy and other inputs. Examples of such services are hotels and restaurants, transport, real estate and health care, among others. In particular, the transport sector is regarded as one of largest fossil-fuel consumers in the world, and the international energy agency (2009) came to the result that the transport sector is responsible for almost 25 per cent of the worldwide CO2 emissions. Nansai et al. (2009) adopted an input-output approach to analyze the role of services in Japan. They came to the result that the service sector is responsible for the rise in CO2 emissions during the 1990s, because of its intense energy and material use within its supply chain. This implies that services became increasingly dependent on energy and that they will play an important role for environmental degradation. Butnar and Llop (2012) came to a similar conclusion for Spain. On these grounds, the assumption that services are less polluting should be reconsidered as a possible cause for the EKC (Kaika and Zervas, 2013).

2.6.5 Econometric concerns

Stern (2004) criticized the econometric approaches based on heteroskedasticity, simultan-eity, omitted variables and cointegration issues. He argued that the empirical research did not come to a robust result and that the use of structural models instead of reduced-form models could produce some relief, and may help to suciently understand the income pollution relationship to formulate suitable policy recommendations.

One major issue in the empirical studies is the availability of appropriate data on environmental degradation. There are not enough sucient time-series data available for all countries, which are long enough to capture structural changes, therefore researchers started to employ panel data. Panel data do indeed allow an increase of the sample size by pooling together data from dierent countries, but they assume homogeneity between countries that implies the EKC has the same slope for all countries in the sample (Dijkgraaf and Vollebergh, 2001; Lieb, 2004). Even though a mutual turning point is found, it does not mean that each country will follow the same pattern (De Bruyn et al., 1998).

Several researchers tested the homogeneity assumption and they all came to the result that this assumption has to be rejected. This means that every country has its own,

(25)

unique EKC (Dijkgraaf and Vollebergh, 2001, 2005; List and Gallet, 1999; de Bruyn, 2000). List and Gallet (1999) even had to reject the homogeneity assumption for federal states in the US regarding NOx and SO2 emissions. Dijkgraaf and Vollebergh (2005) showed that the turning points for single countries dier signicantly when either time series or panel data are used. Hence, it is advantageous to use longer time series data of a single country to study the pollution income relationship (List and Gallet, 1999).

EKC-studies usually assume that causality runs from income to environmental degrada-tion, despite good reason to suspect that environmental degradation can have detrimental eects on economic growth. For instance, Isen et al. (2013) found evidence in a very in-teresting paper, that individuals who were born in counties which adopted the Clean Air Act of 1970 earn comparatively $4,300 more in their lives. They explain this result with the better health of these individuals, which is associated with a higher productivity. The regressions suer from a simultaneity bias, when pollution has an impact on economic growth. However, Cole et al. (1997) and List and Gallet (1999) did not nd evidence for a simultaneity bias. Although pollution surely has an impact on income, it tends to be too small to be statistically signicant (Lieb, 2004; Stern, 2004).

This last issue was already shortly mentioned above; most of the papers on the EKC implicitly assume that pollution in a country is unaected by the pollution in neighbor-ing countries, but empirical evidence from spatial econometric approaches has shown that this assumption is unlikely to hold (Aslanidis, 2009). There are several reasons to suspect that a spatial relationship might inuence the pollution income relationship. First, the PHH is likely to be inuenced by common borders and the distance between the trading partners, i.e. rich countries are more likely to shift their dirty production sectors to bordering developing countries (Maddison, 2004). Second, technological diusion is inu-enced by spatial factors; technology spillovers decline with distance (Keller, 2001). Third, governments may mimic environmental policies from neighboring regions to reduce the cost of decision-making, and to legitimize their actions. Fredriksson and Millimet (2001) have shown that the implementation of stricter environmental regulation in one US state, led to higher standards by its neighbors, within a period of two to ve years. On the other hand, governments might also strategically lower their environmental standards to attract capital inows (Markusen et al., 1995).

3 Data

3.1 Database

The data are mainly from the World Input-Output Database (WIOD).13 _{The WIOD} consists of a time-series of world input-output tables which distinguishes between 35

(26)

industries, 59 products and ve dierent categories of nal demand for the 27 EU countries and 13 other major economies covering the period from 1995 to 2011. However, since the environmental accounts include only data on CO2 emissions till 2009, this study is constrained to 40 countries for the period from 1995 to 2009. One advantage of the environmental accounts of the WIOD is that the environmental indicators can be linked directly to the corresponding economic activity that allows us to distinguish between emission caused by production, consumption and trade (Timmer, 2012).

This paper focuses on the investigation of the relationship between dierent indicators for CO2 per capita measured in megagram (MG) and per capita income. The CO2 emission data in the WIOD arise from the combustion of fossil fuels, reneries and other industrial processes such as cement production. The data on CO2 presented in this paper can be classied into three dierent categories. First, domestic CO2 indicators encompass emissions directly generated by polluting activities within a country. This includes the emissions from the production of consumable goods and services, and the direct household emissions such as fuel used for private transportation and heating. These indicators can be directly obtained from the WIOD, only some minor calculations are necessary to obtain the per capita gures. Second, consumption or footprint indicators show the pollution embodied in the domestic nal demand of a country, regardless of the location where the polluting activity took place. Third, trade balance indicators report emissions embodied in international trade which reect the dierence between the footprint and domestic indicators.14 _{The trade indicators are only used to illustrate the displacement eects of} CO2 emissions and not for the estimation of the EKC. The calculations for trade and consumption indicators are located in the appendix D.1 (Arto et al., 2012).

PPP converted GDP per capita expressed in international dollars (I$) is used as a proxy for economic growth, since GDP in market exchange rates would understate the income of relatively poor countries. Some studies estimated an EKC with the human development index (HDI) as a proxy for economic growth, but their results had lower explanatory power compared to GDP. Hence, the use of GDP in PPP seems to be the best option (Lieb, 2004). The data on income and population, for the calculation of the per capita gures, come from the Penn World tables.15

3.2 Data motivation and exploration

CO2 belongs to the greenhouse gases and is one of the most prominent causes for global warming. The global implications of CO2 and its extremely long lifetime in the atmo-sphere made it the main focus of several national and international agreements like the Kyoto protocol (Lieb, 2004). Global CO2 emissions continued to grow, despite the

emer-14_{Pollution caused directly by households has to be subtracted from domestic indicators before, in order}

to obtain the trade balance.

(27)

gence of new national climate policies. In the period from 1995 to 2009 the worldwide CO2 emissions increased by 30.9%.16 The majority of the developing countries showed stabilized or even reduced numbers for CO2per capita emissions. In the US CO2emissions have been reduced by 12%, in Germany by 14.83% and in Sweden by 10.58%, respect-ively. At the same time developing countries experienced a strong increase in their CO2 per capita emissions, Brazil recorded an increase of 15.8%, Indonesia of 50.6%, India of 62.1% and China even of 100.16% (Fig.4).

Figure 4: Domestic CO2 emissions per capita in MG

However, most developed countries reduced their consumption-based emissions slower than they did their domestic emissions. The US for example decreased their domestic per capita emissions by 2.23 MG from 1995 to 2009, while the CO2 consumption decreased only by 1.17 MG. Whereas some countries experienced a decrease in CO2 consumption per capita like Germany (17.85%) and Denmark (19.14%), the majority recorded growing numbers such as Spain (10.24%), Greece (31.82%) and China (101.57%). Additional charts for the footprint and the trade balance are located in appendix D.

The Pollution Haven Hypothesis and the displacement eect suggest that developed countries shift part of their CO2-intensive production to developing countries, which may imply that the stabilization of the emissions was partly at the expense of increasing emissions in developing countries. By comparing the domestic CO2 emissions with the CO2 footprint, one can see the transfer of emissions via international trade. Figure 5 shows the domestic CO2 emissions and footprints from 2009.

The majority of the developed countries exhibit unsustainable consumption patterns. They consume more CO2 than they produce meaning that they run a trade balance decit for CO2, i.e. the excess demand is satised by imports of pollution-intensive products

(28)

Figure 5: CO2 domestic emission and CO2 footprint per capita 2009

from other countries. Austria, Germany, Denmark, Cyprus, Japan and Turkey are the only countries that reduced their CO2 trade balance decits within the period 1995 to 2009. This gives some evidence in favor for the hypothesis that the improvements in domestic CO2 emissions in developed countries were partly driven by increasing imports of pollution-intensive products.

The transfer of emissions via trade does not present real improvements in the environ-mental quality of a country, since the problems of climate change caused by CO2 emis-sions are global. Environmental quality from a global perspective can only be improved by means of sustainable consumption patterns and technological progress. However, the majority of EKC studies take only the domestic side into account; thereby they ignore the fact that an unsustainable consumption may be satised by imports from other countries. For this reasons, I consider CO2 footprints as an additional measure for environmental degradation, and by doing so one can eectively control for the displacement eect of emissions.

4 Methodology

4.1 Regression model

(29)

of income, i.e. an N-shaped curve (Cole, 2003). There is clear empirical evidence for the existence of an N-shaped relationship (Bandyopadhyay and Shak, 1992; Lee at al., 2009); and on that account the following regression function (1) is adopted

EN Vit = β1Yit+ β2Yit2+ β3Yit3+ αi+ ηt+ εit (1) i denotes the country and t the time index. ENVit is an indicator for environmental degradation measured in per capita terms in this case domestic CO2 emissions and CO2 consumption, Yit is GDP per capita denoted in PPP I$, εit is an error term and αi and ηt reect country and time-specic eects, respectively.17 The problem of omitted vari-ables as criticized by Stern (2004) can be addressed by using xed eects to control for unobserved heterogeneity. Time-specic eects are included to collect eects that have a common impact on all countries, but vary over time, such as technological progress, while country-specic eects control for time-invariant factors that are specic to a country, such as climate and resource endowment (Cole, 2004). Fixed eects allow that the pol-lution income relationship shifts across countries and time, but it is still not possible to obtain country-specic growth paths (Wagner, 2010). The use of panel data allows us to have a reasonably large data set that is especially important, since data on consumption-based indicators are still very limited. Another feature of panel data is that countries in dierent stages of economic development can be included, which makes it possible to capture structural changes in a relatively short series of data (Kaika and Zervas, 2013).

Banyopadhyay and Shak (1992) suggested to drop the cubic term if it is insignicant, and the same procedure applies to β2 if it remains insignicant in the quadratic specic-ation, therefore equation (1) can yield seven dierent outcomes whereby the EKC is just one of them (Dinda, 2004):

1. β1 = β2 = β3 = 0 insignicant relationship

2. β1 > 0 and β2 = β3 = 0 monotonic increasing relationship 3. β1 < 0 and β2 = β3 = 0 monotonic decreasing relationship

4. β1 > 0, β2 < 0 and β3 = 0 inverted-U shaped relationship, i.e. EKC 5. β1 < 0, β2 > 0 and β3 = 0 U-shaped relationship

6. β1 > 0, β2 < 0 and β3 > 0 N-shaped relationship

7. β1 < 0, β2 > 0 and β3 < 0 inverted-N shaped relationship

Equation (1) is given in levels, but it also can be estimated in logarithms. The estimation in logarithms theoretically makes more sense since, as income approaches innity, pollu-tion would go to minus innity when the regression funcpollu-tion (1) is estimated in levels,

(30)

excluding the cubic term. While, if the logarithm of environmental degradation is used, pollution would approach zero which illustrates a more plausible outcome (Cole et al., 1997). Even though it is important from a theoretical point of view, for empirical research it is more relevant which functional form describes the data the best (Lieb, 2004). Scatter plots will be used for this purpose, and are discussed before the regression results.

One should note that equation (1) is a reduced form model which uses only income as an explanatory variable, and for this reason the model is inappropriate for policy recommendations since the driving forces behind the EKC remain unknown (Stern, 2004). It would be better to rst investigate the impact of income on environmental standards, on technological progress, and on structural changes and afterwards the relationship between these variables and environmental degradation. Even though this approach would help to understand the causes of the EKC better, it is dicult to implement due to the limited availability of data, and as a consequence income is used as a catch-all variable (Lieb, 2004). Moreover, the main objective of this paper is to compare the dierences between consumption-based and production-based indicators of environmental degradation and for this purpose a reduced-form model is sucient. Nevertheless, the regression will be complemented by a structural decomposition analysis to give some insight into the impact of structural changes, consumption and technological progress on environmental changes. Four dierent specications of equation (1) will be estimated: including only a constant, only time-period xed eects, only country-specic xed eects and both country and time-period xed eects. Thus, it can be further investigated whether it is necessary to control for country and time-period xed eects from a statistical point of view by calculating likelihood ratio tests.

4.2 Spatial Econometrics

(31)

goodness-of-t measure R2_.

In contrast, ignoring substantive spatial dependence will yield biased OLS estimates (Anselin and Rey, 1991). In other words, ignoring spatial eects may lead to a wrong spe-cication of the pollution income relationship and the turning point. Recall that spatial eects could be present because of technological diusion, migration of pollution-intensive production and mimicking behavior of politicians (Maddison, 2006). The following para-graph introduces econometric methods to address the problems of spatial dependence.18 For N observations over T time periods, the spatial panel data model including a spatial lag can be written in matrix notation as

Yt = ρW Yt+ xtβ + α + ηt+ εt (2)

The spatial lag model (2) can control for spatial substantive dependence. t is a subscript for time which runs from 1 to T. Y describes an N×1 vector with one observations on the endogenous variable for every country i in the sample, whereby i runs from 1 to N. X denotes an N×K matrix with K explanatory variables, whereas ε is a vector of independently and identically distributed error terms. α19 _{and η}

t are spatial and time-xed eects. ρ is the spatial autoregressive coecient. Finally, W is a prior specied spatial weight matrix which reects the interdependence of the observations. The spatial weight matrix W is a N×N matrix where the rows and columns corresponded to the spatial units in the sample. An element wij of the matrix displays the relative strength of the interaction between the spatial unit i and j. Provided that ρ 6= 0, the spatial lag will be correlated with the error term ε, equation (2) can be easily rewritten to illustrate this

Yt = (I − ρW )−1Xtβ + (I − ρW )−1εt (3) As consequence WY is endogenous which will result in inconsistent OLS estimates. For this reason equation (2) is usually estimated by methods of maximum likelihood or instrumental variables to obtain unbiased parameter estimates. Alternatively, we can control for nuisance dependence by allowing the error term to be spatially correlated. This results into the spatial error model

Yt= Xtβ + α + ηt+ ut (4)

ut= λW u + εt

ut describes the spatially correlated error term and λ is the spatial autoregressive

(32)

coecient.

LeSage and Kelley (2009) argue in favor of the spatial Durbin Model (SDM) since this model is the only one that will produce unbiased coecient estimates when the real data-generating process is a spatial lag, a spatial error or a SAC model.20 _{This makes the} SDM a good point of departure for nding the true data-generation process. Furthermore, it does not impose prior restrictions on the size of the indirect eects (Elhorst, 2010). A change of an explanatory variable in a single spatial unit will aect the unit itself (direct eect), but it will also aect other regions indirectly (indirect eect). The indirect eects are often the main point of interest in spatial econometric research. LeSage and Kelley (2009) acknowledge that the possibility to measure direct and indirect eects of pollution could be useful for the introduction of a classic Pigovian tax, or be important for the implementation of environmental regulations. The SDM includes spatially lagged dependent and independent variables:

Yt= ρW Yt+ Xtβ + W Xtθ + α + ηt+ εt (5) This paper uses the method of the selection of the appropriate model suggested by Elhorst (2010). First, a general-to-specic approach is followed by estimating the OLS model as a benchmark.21 _{Robust LM-tests are calculated in order to test whether the} spatial lag or spatial error model can describe the data better. When the OLS model is rejected in favor of a spatial error or spatial lag model, a specic-to-general approach will be additionally adopted with the SDM as a starting point. Likelihood ratio and Wald tests are calculated to test the hypotheses that the SDM can be simplied to a spatial error or spatial lag model.22 _{The SDM is only used if both hypotheses are rejected.} In the case that the OLS model cannot be rejected in favor of a spatial lag or spatial error model, the OLS model should be estimated again with spatially lagged independent variables (WX), i.e. as the so-called SLX model Y = Xβ + W Xθ + α + ηt+ ε. The OLS model describes the data the best if the hypothesis H0: θ=0 cannot be rejected, which would imply that there is no empirical evidence for spatial interactions. However, if the hypothesis is rejected, the SDM should be estimated to examine whether ρ is signicantly dierent from zero. When ρ appears to be signicantly dierent from zero, a SDM should be adopted otherwise one may conclude that the SLX model describes the data the best. The chosen specication of the spatial weight matrix W is of crucial importance for applied spatial econometrics. The value of the spatial interaction parameters and their signicance levels depend on the specication of W (Leenders, 2002). Spatial econometric

20_{Elhorst (2010) gives a good overview about the relationships between dierent spatial dependence}

models.

21_{All regressions and tests are performed by means of the Matlab routines from dr. Elhorst downloadable}

at http://regroningen.nl/elhorst/software.shtml.

22_H

The implications of international trade on the environmental Kuznets Curve