Trade and Environment

Energy Economics 52 (2015) 55–68
Contents lists available at ScienceDirect
Energy Economics
journal homepage: www.elsevier.com/locate/eneco
The world trade network and the environment☆

Carlos Aller a, Lorenzo Ductor b, M.J. Herrerias c
a
Universitat de Girona, Girona, Spain
b
Middlesex University London, London, United Kingdom
c
University of Birmingham, Birmingham, United Kingdom
a r t i c l e i n f o a b s t r a c t
Article history: This paper analyses the role of the world trade network on the environment. We rely on methods developed for
Received 21 April 2015 social network analysis to identify the most important countries in connecting trade between all the other coun-
Received in revised form 1 September 2015 tries in the world trade network. We then estimate how the network or indirect effects from trade affect the en-
Accepted 5 September 2015
vironmental quality of a country. As the trade networks are endogenously determined by trade and
Available online 25 September 2015
environmental conditions, we use as instrumental variables the growth in the population of trade partners and
JEL classification:
the growth in the population of trade partners' partners to exploit exogenous variation in the world trade net-
C33 work. Once we simultaneously estimate the environmental, trade, income, and network equations using a
F18 three-stage least square procedure, we find that network effects harm the environmental quality of developed
O50 countries but improve the environment of developing countries.
Q56 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Keywords:
World trade network
Carbon dioxide emissions
Centrality measures
1. Introduction agriculture to service is expected to decrease the level of emissions);

and the technique effect that predicts a positive effect of trade on the en-
The significant increase of trade flows among developed and devel- vironment through the use of cleaner techniques of production.1 Since
oping countries have led to a more integrated and globalised interna- these are effects of trade between a particular country a and other coun-
tional market (De Benedictis and Tajoli, 2011). Globalisation may tries in the world on the environment of country a, we call them direct
exert a positive effect on economic growth by facilitating specialisation effects from trade. These direct effects from trade on environmental deg-
among countries according to their comparative advantage and facili- radation has been widely examined in the empirical literature. Howev-
tating the transfer of resources across countries (Antweiler et al., er, there is no agreement as to whether the relationship between trade
2001). On the other hand, the increase of trade flows may also have det- and environmental quality is positive, negative, or non-existent.2
rimental effects on the environment. In theory, the effect of trade on the Both traditional trade models and empirical evidence on interna-
environment is ambiguous. According to the traditional theoretical lit- tional trade analyse the relation between each couple of countries in
erature, trade affects the environment through three main different
channels (Antweiler et al., 2001): the scale effect states that higher
GDP leads to higher pollution; the change in the sectoral composition 1
Jayadevappa and Chhatre (2000) introduce a new effect based on property rights, e.g.
of a country as a consequence of trade, composition effect, could affect countries with a well-defined property rights are more able to allocate resources efficient-
positively or negatively the environmental conditions of a country ly and reduce their level of emissions.
2
(e.g. a change from agriculture to industry may lead to higher energy Antweiler et al. (2001) found that emissions increases as GDP rises (i.e. positive scale
effect) and decreases as trade openness increases. Frankel and Rose (2005) use an instru-
consumption and air pollution while a change from industry or mental variable strategy to control for the reverse causality between trade and the envi-
ronmental quality measure and found that trade reduces the level of SO2 emissions. In
☆ Authors are grateful to Fran Sánchez Samblas for his research assistance at Universitat contrast, Cole and Elliot (2003) found a negative relationship between income and emis-
de Girona, for helpful comments to participants at VI AERNA conference, and for the sions (i.e. negative scale effect) and a positive relationship between trade openness and
comments and encouragement of three anonymous referees. All remaining errors are emissions. Managi et al. (2009) uses a simultaneous equation model and instrumental var-
ours. iable to estimate the interdependence relationship between trade, income, and the envi-
E-mail addresses: [email protected] (C. Aller), [email protected] (L. Ductor), ronment. They found that trade improves the environmental quality of OECD countries
[email protected] (M.J. Herrerias). but increases the level of SO2 and CO2 emissions in non-OECD countries.
http://dx.doi.org/10.1016/j.eneco.2015.09.008
0140-9883/© 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
56 C. Aller et al. / Energy Economics 52 (2015) 55–68
isolation (Chaney, 2014). Thus, for a pair of countries a and b, they ig- measures and trade using the growth in the population of trade partners
nore the effects that other countries' (let say c and d) trade relations and the growth in the population of trade partners' partners. The main
may have in the trade relation between a and b. The effects of the idea is that the higher the population growth of trade partners and
trade relations between c and d on the trade relations between α and trade partners' partners, the higher the trade of these trade partners
b are what we call indirect effects from trade. In this paper, we propose or trade partners' partners with the country. We simultaneously esti-
two indirect trade (or network) effects: the congestion externalities, mate the environmental equation, together with the network, trade,
which we define as the reduction in the capacity to import environmen- and income per capita equations, using the three-stage least square
tally friendly goods, energy-efficient technology, or services from trade estimator.
partners owing to an increase in trade between trade partners or trade The effect of trade on the environment may also depend on the level
partners' partners. Countries whose trade partners are trading signifi- of economic development of the country. Trade can increase environ-
cantly with other trade partners or with trade partners' partners will mental pollution in developing countries as a result of the trade activity
have lower opportunity to import goods, since resources are limited. from developed countries, that is, the so-called pollution haven hypothe-
Thus, these countries will have to increase their national production, sis (Taylor, 2005).6 However, trade facilitates the specialisation of coun-
thereby increasing their level of CO2 emissions.3 The second indirect tries according to their comparative advantage; countries abundant in
trade effect is related to market power. Choi et al. (2014) show that capital (developed countries in general) will tend to specialise in the
the role of a node (country) in connecting the trade between two production and export of pollution-intensive goods, leading to an in-
other countries defines market power and consequently pricing and ef- crease in the level of pollution in these countries (Antweiler et al.,
ficiencies. Higher trade between two countries, b and c, that pass 2001; Managi et al., 2009; Termushoev, 2006). This is the factor endow-
through another country, a, may increase the market power and ment hypothesis, which states that factor endowments are the main
bargaining power of the intermediary country, which could be translat- determinants of trade and its relationship with the environment. In
ed into cheaper imports or greater market access and better environ- our empirical analysis, we split the sample of countries into two ‘high-
mental quality.4 These indirect effects may be beneficial for the income and low-income countries’ in order to analyse whether the
environment if the market power gains outweigh the congestion exter- pollution haven hypothesis or factor abundant hypothesis is found in
nalities or detrimental if the congestion externalities outweigh the mar- our sample of countries, once we account for the indirect or network
ket power gains. effects of trade on the environment.
The central focus of this paper is to estimate the indirect effects of The findings show that once country characteristics have been con-
trade on the environment. Previous studies that focus only on the di- trolled for, network centrality measures have significant environmental
rects effects from trade, ignoring the indirect effects, might be implications. More specifically, having a relevant position in the
underestimating or overestimating the total effect of trade (direct and network (as measured by betweenness centrality) is positive in envi-
indirect effects) on the environment. Understanding the role of the indi- ronmental terms for developing countries but detrimental to the envi-
rect effects on the environment is important for policy makers on the ronment for developed economies. Moreover, we find when we focus
design of trade and environmental policies. For example, if for develop- on the export network that the closer a country is to its trade partners
ing countries the direct effects from trade are positive but the indirect as an exporter, the lower the level of carbon dioxide emissions for de-
effects are negative, an open trade policy might be still beneficial for veloped economies. Finally, we find some support for the pollution
the environment in developing countries. haven hypothesis in low-income countries as increases in foreign direct
To estimate the indirect effects of trade, we first represent all the investment lead to higher levels of emissions.
trade relations between all the countries, the world trade network, in The structure of the paper is as follows. In Section 2, we present
a graph, using a rich dataset on bilateral trade from 1996 to 2010.5 We the conceptual framework. Data and methodology are described in
then measure the degree of connectivity of a country in the world Section 3, while the results are presented in Section 4. Conclusions are
trade network using five different measures of centrality: out- drawn in Section 5.
closeness, in-closeness, closeness, betweenness, and eigenvector.
Higher centrality is related to a better strategic position in the world 2. Conceptual framework
trade network. Finally, we estimate the indirect (network) effects of
trade on the environment using these centrality measures. As the posi- Climate-change policy to prevent global warming has been one of
tion of the country in the trade network may be endogenously related to the most important concerns among economists, environmentalists,
trade, income, and the environment, we instrument the centrality and policy makers over the last four decades. In addition, gradual
trade liberalisation has opened up debate about its environmental con-
sequences among developed and developing countries since the United
3
This network effect is based on the co-authorship network formation model of Jackson Nations Stockholm Conference on Development and Environment in
and Wolinsky (1996). In their model, the utility driven by the formation of a link between 1972 (Jayadevappa and Chhatre, 2000). Moreover, the progressive
two players, declines as each player forms new links with other players. The idea is that if introduction of environmental aspects into international trade agree-
your co-authors are busy because they are working on many projects at the same time, ments, such as NAFTA, provides evidence of the awareness of environ-
they have less time to devote to your project, so you will have to devote more time to
the collaboration and hence have less time to do other work. Ductor (2015) presents em-
mental quality in a more globalised world.
pirical evidence about the existence of these externalities in co-authorship networks. Initial works in this area of research started in the context of the En-
4
These indirect effects from trade on the environment are defined in details in vironmental Kuznets Curve (EKC), which predicts an inverted U-shaped
Section 2. relationship between environmental pollution and economic develop-
5
Besides the countries finally included in the analysis (Table 2), we consider the follow-
ment. However, it has been argued that the EKC provides a limited ex-
ing in order to construct the world trade network: Afghanistan, Algeria, Angola, Antigua
and Barbuda, Aruba, Azerbaijan, Bahrain, Barbados, Belarus, Belize, Benin, Bermuda, planation about this relationship as it does not account for the trade
Bhutan, Bosnia and Herzegovina, Burkina Faso, Burundi, Cambodia, Caiman Islands, patterns in the original framework (Antweiler et al., 2001). Subsequent
Central African Republic, Chad, Dem. Rep. Congo, Cuba, Dominica, Equatorial Guinea, works have introduced trade into the analysis, creating a large body of
Ethiopia, Fiji, Sudan, Gambia, Georgia, Greece, Guinea, Guinea-Bissau, Guyana, Haiti, Hong empirical research, but no conclusive evidence has been found
Kong, Iraq, Israel, Jamaica, Rep. Dem. Korea, Kyrgyz Rep., Lao, Lebanon, Liberia, Libya, Ma-
cao, Macedonia, Madagascar, Maldives, Marshall Islands, Mauritania, Mauritius,
Micronesia, Myanmar, Nepal, New Caledonia, Nicaragua, Nigeria, Oman, Papua New
6
Guinea, Qatar, Romania, Rwanda, Samoa, Seychelles, Solomon Islands, St. Kitts and Mani and Wheeler (1998) provide empirical evidence consistent with the pollution
Nevis, St. Lucia, St. Vicent and the Grenadines, Suriname, Tajikistan, Turkmenistan, haven hypothesis. In contrast, Janicke and Binder (1997) find no evidence to suggest that
United Arab Emirates, Uzbekistan, Vanuatu, Yemen, Yugoslavia, and Zimbawe. the environmental regulations are a determinant of trade patterns.
C. Aller et al. / Energy Economics 52 (2015) 55–68 57
(Kleemann and Abdulai, 2013). On the one hand, the pollution haven differ as the relative factors prices and therefore the source of compara-
hypothesis (PHH) can partially explain a reduction in the level of pollu- tive advantage are affected. Trade frictions include extensive and inten-
tion in high-income countries given that there is a transfer of pollution sive margins of trade (the volume and the number of trade partners).
through international trade to developing countries with more flexible Helpman et al. (2008) and Chaney (2014) formalised this model, con-
environmental regulations (Cole, 2004; Wagner, 2010). The specialisa- sidering trade frictions and economic geography, respectively, arguing
tion of trade can account for this phenomenon as developed countries that traditional estimates of trade are biassed in the omission of the ex-
display comparative advantage in the production of clean products tensive margins of trade and asymmetric trade flows among countries.
and developing ones do have pollution-intensive production (Wagner, We show in this work that trade frictions captured by the extensive
2010). On the other hand, the factor endowment hypothesis (FEH) pos- and intensive margins of trade are linked with the world trade network
tulates that countries that rely on a comparative advantage in capital and therefore with our centrality measures, in a similar fashion to
and technology will pollute more as their economies become more Felbermayr and Koheler(2005).
open to international trade because these factors are used intensively In this paper, we propose two network (indirect) effects, namely,
in high-polluting industries (Mani and Wheeler, 1998). This is contrary congestion externalities and market power. On the one hand, for example,
to the PHH's prediction and therefore one of the sources of the mixed an increase in exports from country a to country b may reduce the ca-
evidence in the relevant literature (Birsdsall and Wheeler, 1993; Cole pacity of other trade partners and trade partners' partners to import
and Elliot, 2003; Copeland and Taylor, 2004; Lucas et al., 1992). In addi- from country a, because resources are limited. These indirect effects
tion, this mixed evidence in the literature has confronted two different from trade (or network effects) are what we call congestion externalities
economic policy perspectives: the defenders of trade liberalisation, (Jackson and Wolinsky, 1996). Countries that have trade partners trad-
who support the reduction of trade barriers among countries to protect ing among themselves will have fewer opportunities to import environ-
the environment, and those who argue for stronger environmental reg- mentally friendly goods or technology that leads to energy efficiency.
ulation and a more inward-looking trade policy orientation. Environ- They may also have to increase their national production as a conse-
mentalists, who support the latter view, argue that increases in quence of the reduction in their capacity to trade. Therefore, countries
income and consumption levels lead to an increase in levels of pollution. suffering from congestion externalities, i.e. countries whose trade part-
Thus, an outward-looking trade policy can damage the environment. Of ners trade significantly with other trade partners or trade partners' part-
course, these conflicting points of view depend on the source of com- ners, will have higher levels of carbon dioxide emissions, holding other
parative advantage and the stage of economic development across factors constant.
countries, which might clarify the net effect of international trade on The world trade network may also reveal information about the
the environment and therefore whether the PHH or FEH explains such market power of each trader. Following Choi et al.'s (2014) model, sup-
a phenomenon. pose that there is a source node, S, and a destination node, D. A path be-
Among the most influential works on the relationship between trade tween these two nodes is a sequence of edges connecting them. The
and the environment were those published by Grossman and Krueger world trade network is defined by all the nodes (i.e. all the countries
(1991, 1995) and later on by Antweiler et al. (2001), who argue that in the world) and paths between them. The trade of goods from source
there are three potential factors through which international trade can to destination may involve intermediaries that post a price to the prod-
lead to variations in carbon dioxide emissions. First, the scale effect, uct, thereby increasing the total cost of the good traded between the
often measured by per capita GDP, which refers to the effect of an source and the destination node. Choi et al. (2014) show that the role
increase in production emissions. Second, the technique effect, of a node in connecting the source and the destination nodes defines
which can operate through the process of openness (foreign direct in- market power and, consequently, pricing and efficiencies. Specifically,
vestment [FDI] and trade) as a result of knowledge transfer and the Choi et al. (2014) define a node as critical if it lies on all the paths be-
use of less-polluting technologies, preventing environmental degrada- tween the source and the destination nodes. Critical nodes have greater
tion (Herrerias et al., 2013). Third, the composition effect, captured by market power in the trade network, a higher bargaining power in trade
the share of industrial output over GDP, which can lead to positive or negotiations, and more opportunities to reduce their carbon dioxide
negative effects on environmental quality. For example, one should ex- emissions at the expense of the least critical countries.
pect that a change from agriculture to industry leads to an increase in To fix ideas, consider the agricultural trade agreements between
the level of energy consumption, and therefore an increase of the level Russia and the European Union. In 2014, the European Union imposed
of pollution. However, to the extent that an economy moves away economic sanctions on Russia over the Ukraine crisis. Russia responded
from industry to the service sector, one should expect the level of emis- by imposing a trade ban on imports of most of the agricultural products
sions to decrease and decouple from economic activity. Jayadevappa from the European Union, the United States, and Canada. This embargo
and Chhatre (2000) added an additional relevant factor, namely, the en- shaped the world agricultural trade network: the immediate effect was
vironmental externalities coming from common property rights. These that the agricultural trade links between Russia and the banned coun-
authors argued that countries with a lower presence of the state sector tries were removed, so Russia lost criticality in the world trade network.
and well-defined property rights are able to allocate resources in a more Second, as indirect effects, other countries, such as Israel, Argentina, and
efficient way, leading to an increase in income and a reduction in the former Soviet republics in the Caucasus and Central Asia increased
environmental problems. By contrast, countries with poorly defined their agricultural exports to Russia, while agricultural importers from
property rights tend to exploit and misallocate resources, leading to a Russia, such as Kazakhstan, had to decrease their agricultural imports
reduction in environmental quality. from Russia and increase their own production (i.e. Kazakhstan suffered
One of the shortcomings from traditional trade models is that they from the congestion externality). Thus, trade relations between, for ex-
ignore that international trade is more than an exchange of goods. It is ample, Russia and the United States, should not be analysed in isolation;
a complex structure in which countries are connected to each other in we should take into account the effect of the European Union or other
a world trade network, and bilateral trade among two countries not countries in the relationship between Russia and the United States.
only affects those two countries but also other trade partners or trade The structure of interdependence relations between all the countries
partners' partners. Explicit consideration of the network can provide in the world trade could be represented by a graph in which countries
new insights and can challenge the predictions of the traditional trade are nodes and links are defined by the volume traded among all the
models and our understanding of the relationship between trade and countries. These interdependence relations can be analysed using
global emissions. There are a few attempts in international trade theory methods developed for social network analysis.
to capture the degree of interdependence of countries. Krugman (1980) Network externalities and the world trade network structure have
argued that under the presence of trade frictions, trade patterns can often been neglected in the empirical literature of international trade,
but there are exceptions: Bernard and Jensen (1995) and Bernard et al. of connectivity and criticality that we use in our analysis are the
(2007) found that is a high heterogeneity in trade flows, partners, and following:
links across countries. Serrano and Boguñá (2003) show that the
1. Out-closeness measures how close a country is to any other country in
world trade network satisfies the properties of a complex (and not ran-
the world trade network, where the distance between the countries
dom) network. De Benedictis and Tajoli (2011) examine the features
is weighted by bilateral exports. Formally, the out-closeness central-
and evolution of the world trade network. They show that the trading o
ity, Ca,t, is the inverse of the average weighted distance of a country to
system is becoming more and more interconnected at the same time
other countries within the largest component of the network and is
as heterogeneity among countries is increasing. More recently, De
defined as
Benedictis et al. (2014) analysed the centrality or criticality of countries
in the world trade network.7
nt −1
However, to our knowledge, there is no study in the empirical trade C oa;t ¼ X
dða; b; Gt Þ
literature on how the world trade network structure and network exter-
b≠a
nalities affect key macro-magnitudes, such as environmental quality.
Our contribution is to incorporate social network analysis into the em-
where nt is the size of the largest component in year t in the world
pirical international trade literature and estimate the direct and indirect
trade network Gt.
effects of trade on environmental quality. Unlike the traditional ap-
2. In-closeness, Cia,t, is defined like out-closeness but using a weighted
proach adopted in the literature, we do not treat the trade relationship
distance based on bilateral imports. Notice that the two centralities,
between two countries, a and b, in isolation; we also take into account
in-closeness and out-closeness, may differ. A country with higher
how the trade between countries c and d affect the relations between
out-closeness centrality than in-closeness centrality is closer to its
countries a and b. Understanding these trade interdependences is cru-
trading partners as an exporter than as an importer (De Benedictis
cial to accurately estimating the impact of trade on the environment.
and Tajoli, 2011).
We expect that changes in the trading network structure will help to
3. Closeness centrality, Ca,t, is defined like out-closeness but using a
further clarify the role played by the PHH or the FEH, as well as the com-
weighted distance based on total bilateral trade. The higher the
position and technique effects in the relationship between trade and the
closeness centrality of country a is, the closer is the connection be-
environment.
tween country a and the rest of the countries in the trade network.
Whether countries should open their economies to trade or promote
The relationship between closeness centralities and the environment
a more inward-looking strategy to prevent damage to the environment
is ambiguous. On the one hand, the closeness centrality of country a
is an interesting question to investigate, not only for trade policy but
may increase as a result that this country has new trade partners (ex-
also for the relationship between trade and the environment. The intro-
tensive margin of trade). Having more trade partners may facilitate
duction of the novel aspect of the world trade network in the relation-
the transfer of technology from these new trade partners to country
ship between trade and the environment can help policy makers to
a, resulting in a reduction of carbon dioxide emissions, as a conse-
design and link trade and environmental policies for the sustainability
quence of a more efficient production (see Keller, 2004, for the effect
of economic growth. However, these effects may differ across countries,
of international diffusion of knowledge). On the other hand, close-
making it necessary to separate the analysis into high- and low-income
ness centrality may also increase if the country is exporting (or
economies.
importing) more to another country (intensive margin of trade). A
higher volume of exports is associated with higher pollution levels
in low-income countries because they have weaker environmental
3. Data and methodology
regulations. An increase in exports would also lead to higher carbon
dioxide emissions in capital-abundant countries if the FEH holds, be-
3.1. The world trade network
cause these countries would specialise in the production and expor-
tation of pollution-intensive goods.
World trade can be represented as a weighted network, G. Formally,
4. Betweenness centrality, Ba,t, is another centrality measure that cap-
a weighted network can be depicted as consisting of nodes, N = 1, …, n,
tures the criticality of a country in the trade between two other coun-
and a real-valued n × n matrix g (the adjacency matrix), where gab,t rep-
tries. A country that is intermediary in the trade between many other
resents the intensity of the relation between a and b at time t. Following
countries will have a high betweenness centrality and a high market
De Benedictis and Tajoli (2011), in our world trade network, two nodes
power and bargaining power in trade negotiations, as they are key
(countries) a and b have a weighted link gab,t = Eab,t in which gt equals
countries in the trade of goods between two other countries (Choi
the exported volume from country a to b, Eab,t.8 The world trade net-
et al., 2014). For example, France is a critical country in connecting
work Gt is directed since it is possible that gab,t ≠ gba,t. In other words,
the trade between Spain and continental Europe or the United
the volume exported from country a to country b does not necessarily
Kingdom, so it is expected that the betweenness centrality of
coincide with the volume exported from country b to country a.9
France is higher than the centrality of Spain. France as an intermedi-
We consider for our network analysis the total bilateral exports.
ary country may have market power in the products exported from
With this information, we construct for every time period t the world
the Iberian peninsula and may have more opportunities to import
trade network. We use the world trade network and methods devel-
pollution-intensive goods at the expense of less central countries.
oped for social network analysis to measure the criticality of each coun-
Thus, we expect that the more central or more critical countries
try in the world trade network and to estimate the indirect effects of
will have lower levels of carbon dioxide emissions.
trade on the environment. We say that there is a path between a and
Formally, the betweenness centrality of country a is the frequency of
b in Gt if gab,t ≠ 0, i.e. if there exists a sequence of edges that connects
shortest paths passing through node a and is calculated as
both nodes, a and b. The geodesic distance between two nodes in net-
work Gt, d(a, b; Gt), is the length of the shortest path. The main measures
X τ ab;c ðGt Þ
Ba;t ¼
7
Other studies analysing the world trade network are Garlaschelli and Loffredo (2005), τ Gt;s
b≠c:b;c≠a b;c
Kali and Reyes (2007), Fagiolo et al. (2009) and Serrano et al. (2007).
8
Notice that the exported volume from country a to b is equal to the imported volume a
by country b from a.
where τb,c (Gt) is the number of shortest paths between b and c in Gt
9
In a directed network, the matrix g that contains all the connections between the dif- that pass through node a, and τb,c(Gt) is the total number of shortest
ferent countries is not symmetric. paths between b and c in Gt.
Table 1
Descriptive statistics.
High income (836 obs.) Low income (416 obs.)
Variable Mean Std. Dev. Min. Max. Mean Std. Dev. Min. Max.
Per capita GDP 8.674 0.739 6.195 10.55 6.258 1.214 3.818 8.904
CO2 emissions (kt per capita) 7.453 5.372 0.053 38.161 0.816 1.183 0.02 8.076
Exports 17.501 1.763 12.452 21.315 15.161 1.606 11.406 19.232
Imports 17.536 1.623 13.357 21.477 15.443 1.431 11.884 19.657
Betweenness 3.537 3.847 0 10.737 1.307 2.648 0 8.613
Eigenvector 0.114 0.177 0 0.881 0.011 0.021 0 0.144
Out-closeness standardised 0.663 0.104 0.285 0.871 0.498 0.197 0.06 0.855
In-closeness standardised 0.446 0.052 0.284 0.547 0.374 0.081 0.081 0.534
Closeness standardised 0.828 0.094 0.481 1.008 0.664 0.181 0.187 0.995
Foreign direct investments (% of GDP) 11.626 1.719 4.463 15.718 9.587 1.581 5.379 13.826
Industry output (% of GDP) 31.608 8.048 15.783 66.22 28.924 9.684 14.218 100
Economic institutional quality 0.488 0.812 −2.342 1.888 −0.742 0.538 −2.029 0.639
Notes: GDP, CO2 emissions, trade and centrality measures are computed in logs.
5. Eigenvector. This centrality measures the proximity of a country to industry over GDP, the technique effect is covered by FDI, and the scale
many other ‘central’ countries, so the eigenvector centrality of a effect is captured by per capita GDP. FDI is probably one of the most im-
country depends on the centrality of the largest trading partners. portant channels for knowledge transfer from developed and developing
Formally, the eigenvector centrality is defined as, countries; healthier institutions lead to higher innovation capacity and
stronger environmental regulations, while per capita GDP is a traditional
λC e ðGt Þ ¼ gC e ðGt Þ ð1Þ factor in the relevant literature on trade and environment. We introduce
the potential non-linearities between environment and per capita in-
where λC e(Gt) = ∑j gij,tC e(g), i.e. the centrality of a country is propor- come (the EKC) into our model by adding the square of per capita GDP.
tional to the sum of the centralities of the trading partners, the propor- Finally, we perform a set of robustness analyses by considering alterna-
tionality factor is given by λ. From Eq. (1), we can observe that C e(G) is tive measures of environmental quality, such as SO2 emissions. In the
an eigenvector of the adjacency matrix g, and λ is its corresponding ei- Appendix, we report the details of the data used in this paper.
genvalue. We use as a measure of the centrality of a country in the As we cover a large sample of countries and one of our aims is to test
network the eigenvector C e(G) associated with the largest eigenvalue. whether the PHH or FEH is present in our data, we split the sample of
The idea is that countries with high eigenvector centralities are those countries into two. First, we take into account the low-income countries
which trade substantially with many other countries that are, in turn, ($4085 or less) and then high-income ($4086 or more) economies ac-
trading considerably with many other countries, and so on. cording to the World Bank February 2014 gross national income per
In the next section, we describe the data and rank countries in terms capita classification.
of their centrality or criticality in the world trade network. Descriptive statistics are reported in Table 1. We observe that, on aver-
age, high-income countries are significantly more polluting. They have
3.2. Data and descriptive statistics better institutions and the weights of FDI and industry output are slightly
larger. Moreover, high-income economies trade more and are more cen-
In order to analyse the direct and indirect effects of trade on environ- trally in the world trade network, considering the five indicators outlined.
mental quality, we use data from different sources for 177 countries for For example, the betweenness centrality is on average three times larger
the period from 1996 to 2010. for high-income countries than it is for low-income economies.
We first consider per capita carbon dioxide emissions as a proxy var- Table 2 lists all the countries included in the final estimations,
iable of environmental quality. These data come from the World Bank. ranked by the trade centrality measure. The most central countries in
Carbon dioxide emissions are generally accepted as a consistently defined terms of betweenness centrality are the United States, Germany,
measure of pollution across a significant set of countries such as ours. France, the Russian Federation, and Japan. These countries are ‘regional
Second, to be able to capture the direct and indirect effects of trade hubs’; the United States plays an important role in American trade,
and the environment, we build the world trade network and five Germany and France likewise in European trade, the Russian Federation
centrality measures based on data coming from bilateral trade flows in trade among the central Asian former Soviet republics and the coun-
of 177 countries. We obtained the data from the United Nations tries of the Caucasus, and Japan in Asian trade. Notice that richer econ-
COMTRADE, via World Integrated Trade Solution (WITS).10 omies are better ranked in terms of centralities even though the
We use the following as control factors in the estimation of network rankings of the various centrality measures differ significantly. For in-
effect and environment: per capita GDP, the stock of FDI over GDP, the stance, Russia ranks 4th for betweenness centrality while it is the 30th
share of industry over GDP, and the quality of institutions. Per capita for in-closeness; Vietnam is the third economy for out-closeness and
GDP comes from the World Bank, FDI from the United Nations Confer- 47th for betweenness; and Kazakhstan is second for out-closeness,
ence on Trade and Development (UNCTAD), and the share of industry 42nd for eigenvector, 51st for in-closeness, and 95th for betweenness.
comes from the Quality of Government dataset (QOG). The quality of in- This disparity justifies the use of several measures of centrality, because
stitutions measured by economic institutional quality was taken from they capture different characteristics of the network.
Kunčič (2014). The composition effect is captured by the share of Country-specific information obtained by averaging CO2 per capita
emissions and trade and centrality measures across time is shown in
10
Bilateral trade is recorded in nominal US dollars by WITS. Imports information is more Fig. 1. We observe a positive correlation between volume of trade
complete than exports information is. For this reason, the network is constructed by taking (which increases with per capita income) and carbon dioxide emissions.
the imports value, recorded in cif (cost, insurance, and freight). Whenever this information A similar pattern is observed for out-closeness and in-closeness central-
is missing, we take the exports value, if it exists, recorded in fob (free on board). Some ob- ity measures. However, for eigenvector and betweenness, the relation-
servations are recorded as a zero and others (the majority) are missing. In this paper, we
will consider both kinds of observation as no trade, because there is limited consensus in
ship is not so clear.
the literature about the share of true zeros and missing values (Sato and Dechezleprêtre, Fig. 2 shows time-specific information by averaging CO2 per capita
2013). emissions and trade and centrality measures across countries. Results
Table 2
Country Sample and Centrality Measures.
Country (1) Eigenvector (2) Betweenness (3) Out-closeness (4) In-closeness (5) Closeness Income group
United States 1 1 5 5 3 HIGH

Canada 2 31 36 2 26 HIGH
China 3 6 6 11 5 HIGH
Japan 4 5 7 8 6 HIGH
Germany 5 2 4 6 4 HIGH
Mexico 6 60 8 7 7 HIGH
United Kingdom 7 8 9 10 8 HIGH
France 8 3 22 1 14 HIGH
Korea, Rep. 9 17 10 14 9 HIGH
Italy 10 7 13 12 10 HIGH
Netherlands 11 29 12 13 11 HIGH
Belgium 12 14 11 9 2 HIGH
Singapore 13 16 24 18 18 HIGH
Spain 14 13 17 17 16 HIGH
Malaysia 15 40 14 25 17 HIGH
Saudi Arabia 16 18 1 72 1 HIGH
Switzerland 17 25 16 16 15 HIGH
Australia 18 10 19 19 22 HIGH
Thailand 19 15 23 20 23 HIGH
Brazil 20 12 21 24 21 HIGH
Russian Federation 21 4 20 30 24 HIGH
Ireland 22 79 33 4 31 HIGH
Austria 23 71 15 15 13 HIGH
India 24 9 28 33 30 LOW
Sweden 25 32 32 22 29 HIGH
Indonesia 26 37 27 35 28 LOW
Philippines 27 64 30 32 34 LOW
Poland 28 36 26 21 25 HIGH
Norway 29 23 31 36 35 HIGH
Turkey 30 19 37 29 36 HIGH
Venezuela 31 27 18 37 20 HIGH
Denmark 32 39 35 27 33 HIGH
Czech Republic 33 57 25 23 27 HIGH
South Africa 34 11 41 39 42 HIGH
Vietnam 35 47 3 40 12 LOW
Finland 36 26 39 34 40 HIGH
Hungary 37 45 29 28 32 HIGH
Chile 38 52 43 43 43 HIGH
Portugal 39 22 40 31 37 HIGH
Colombia 40 44 38 38 38 HIGH
Argentina 41 30 53 3 47 HIGH
Kazakhstan 42 63 2 51 19 HIGH
New Zealand 43 24 60 26 58 HIGH
Slovak Republic 44 74 42 42 44 HIGH
Peru 45 86 51 52 52 HIGH
Kuwait 46 91 58 64 55 HIGH
Iran, Islamic Rep. 47 48 57 50 51 HIGH
Costa Rica 48 72 46 44 48 HIGH
Ukraine 49 21 44 41 39 LOW
Pakistan 50 28 56 59 57 LOW
Ecuador 51 49 34 56 41 HIGH
Egypt, Arab Rep. 52 59 63 55 62 LOW
Dominican Republic 53 66 45 45 45 HIGH
Guatemala 54 38 47 46 46 LOW
Morocco 55 93 52 47 53 LOW
Bangladesh 56 51 55 63 59 LOW
Slovenia 57 41 49 48 50 HIGH
Trinidad and Tobago 58 46 48 62 49 HIGH
Tunisia 59 76 54 49 56 HIGH
Honduras 60 55 50 61 54 LOW
Sri Lanka 61 68 61 68 65 LOW
Croatia 62 35 65 53 61 HIGH
El Salvador 63 80 62 58 63 LOW
Bulgaria 64 81 66 57 64 HIGH
Lithuania 65 42 70 54 60 HIGH
Panama 66 92 76 60 67 HIGH
Gabon 67 65 59 81 66 HIGH
Syrian Arab Republic 68 62 64 79 68 LOW
Jordan 69 73 77 69 72 HIGH
Estonia 70 84 67 65 69 HIGH
Cote d'Ivoire 71 33 69 73 70 LOW
Malta 72 58 80 70 76 HIGH
Latvia 73 87 72 66 73 HIGH
Iceland 74 88 78 86 86 HIGH
Uruguay 75 53 71 71 74 HIGH
Table 2 (continued)
Country (1) Eigenvector (2) Betweenness (3) Out-closeness (4) In-closeness (5) Closeness Income group
Ghana 76 34 86 85 87 LOW
Kenya 77 20 81 80 84 LOW
Cyprus 78 96 84 75 80 HIGH
Mongolia 79 77 68 84 75 LOW
Bolivia 80 85 75 83 77 LOW
Paraguay 81 90 79 67 71 LOW
Cameroon 82 50 74 82 83 LOW
Congo, Rep. 83 82 73 89 81 LOW
Tanzania 84 56 89 90 91 LOW
Senegal 85 43 88 77 82 LOW
Zambia 86 54 87 74 78 LOW
Mozambique 87 75 82 76 79 LOW
Moldova 88 70 85 87 88 LOW
Armenia 89 94 91 91 89 LOW
Albania 90 95 83 78 85 HIGH
Uganda 91 61 95 93 92 LOW
Malawi 92 67 93 88 90 LOW
Mali 93 83 96 92 93 LOW
Niger 94 78 90 95 94 LOW
Togo 95 89 94 94 95 LOW
Sierra Leone 96 69 92 96 96 LOW
Notes: Each number represents the position of the country at the different centrality measures rankings. These rankings are constructed by averaging, for every country, the centrality
variable across time.
confirm the strong positive correlation between the volume of trade driven by the transfer of modern technologies across countries. Institit
and CO2 emissions. Eigenvector centrality seems now to have a more controls for the quality of economics institutions; countries with better
clear relationship with emissions. Something similar happens with the economic institutions are likely to have stricter environmental policies
betweenness indicator, which follows the emissions trend but shows, and are more likely to respect international environmental agreements,
at the same time, a higher variability. By contrast, in-closeness and so an increase in quality of institutions will result in a decline in
out-closeness have grown faster than CO2 emissions have during the pollution.11
last decade. The second specification of the emissions per capita model includes
total trade for Eq. (2),
3.3. Empirical strategy
log ðcit Þ ¼ α i þ μ t γ0 þ ρZ it þ θ1 logðTradeÞit þ θ2 log ðGDP Þit
Our aim is to estimate the direct effects from trade (i.e. scale, tech- ð3Þ
þθ3 log ðGDP Þ2it þ θ4 FDI it þ θ5 Indit þ θ6 Instit it þ it ;
nique, and composition effects) and the indirect effects from trade (i.e.
congestion externality, criticality) on environmental quality. In order
where log(Trade)it is the sum of exports and imports of country i at
to distinguish between direct and indirect trade effects, we consider
period t. The trade variable controls for the direct effect of trade,
two different specifications for the per capita emissions model. The
i.e. the effect of increasing the volume of imports and/or exports of
first specification includes as determinants of per capita emissions a
country i on environmental quality while holding income, FDI, insti-
network centrality variable, per capita GDP and its quadratic term,
tutions and indirect (network) effects from trade constant. In Eq. (3),
stock of FDI (as a percentage of GDP), industry output share, and eco-
the network variable, Zit, only captures indirect effects from trade,
nomic institutional quality. Using i to refer to countries and t to refer
such as how an increase in trade between countries h and k affects
to years, the first specification of the emission model is as follows:
the trade relationship between countries i and j, since direct effects
logðcit Þ ¼ α i þ μ t γ 0 þ ρZ it þ β1 log ðGDP Þit þ β 2 logðGDP Þ2it þ β 3 FDIit þ β 4 Indit þ β 5 Instit it þ it ;
are captured by trade.
As trade, income, and the network variable are endogenously deter-
ð2Þ
mined, we estimate a system of structural equations where some of
the equations contain endogenous variables among the explanatory
where cit is the level of CO2 emissions measured in metrics per capita, αi
variables. The income equation is taken from Baghdadi et al. (2013).
refers to country fixed effects and control for time invariant factors, such
We regress per capita GDP, log(GDP)it , on lagged per capita GDP,
as cultural factors and geographical situation, among others, μt refers to
population, investment, human capital formation, year dummies,
year fixed effects and captures world oil prices, changes in technologies,
and country dummies. 12 The trade equation is based on a gravity
environmental regulation, and relevant taxes and subsidies (Judson
model of trade. We regress trade on population, land area, year
et al., 1999). Zit refers to any of the trade network centrality measures
dummies, country dummies, the network variable, trade partners'
described in Section 3.1: eigenvector, betweenness, out-closeness, in-
population, and trade partners' partners' population. We propose
closeness, and closeness. It captures direct and indirect effects from
trade partners' population of country i and trade partners' partners'
trade because the network measure of a country may vary, either be-
population as new instruments in the literature to obtain exogenous
cause its trade partners or trade partners' partners are trading more be-
variation of trade. The idea is that the higher the population of trade
tween themselves (indirect variation) or because the country is trading
partners of country i is, the greater are the needs of these trade
more with others (direct variation). The per capita GDP, log(GDP), and
its quadratic term, ‘scale effect,’ are included to test the EKC hypothesis.
The industry output share of GDP, Indit, controls for the ‘composition ef-
11
fect,’ i.e. the effect caused by trade liberalisation to the structure of the See Kunčič (2014) for a detailed definition of the economic quality of institutions. He
constructed this index using a factor analysis model and information about many different
economy of a country through its specialisation in activities that present variables, including an index of economic freedom, regulatory quality, freedom of the
a comparative advantage. Stock of FDI (as a percentage of GDP), FDIit, press, and regulation of credit, labour, and business, among others.
captures the ‘technique effect,’ i.e. the effect of trade on the environment 12
See Appendix A for a definition of the variables.
CO2 emissions
CO2 emissions
Volume of Exports Volume of Imports

High income Low income High income Low income
CO2 emissions
CO2 emissions
Out-Closeness centrality In-Closeness centrality

CO2 emissions
CO2 emissions
Eigenvector centrality Betweenness centrality
Fig. 1. CO2 emissions, trade, and centrality measures by countries. Notes: Each point represents a country. We have calculated the average CO2 emissions per capita and the average
centrality measure across the years obtained in the sample: 1996–2010.
partners to import goods from country i. Thus, we expect that an in- 4. Results
crease in population of trade partners and trader partners' partners
will result in an increase in the total trade of country i. The increase In columns 1 and 2 of Tables 3–5, we present the results of the emis-
in population in trade partners of country i is assumed to be exoge- sions per capita model when we include as a measure of centrality the
nous to the characteristics of country i. Finally, the network equation eigenvector for the whole sample, developed countries and developing
is similar to the trade equation; the dependent variable is a network countries, respectively. Eigenvector measures the degree of connectivity
variable described in Section (3.1) and the regressors are population, of a country to the most central countries in the world trade network;
land area, income per capita, year dummies, country dummies, trade, thus, for example, the United States is the country with the highest ei-
trade partners' population, and trade partners' partners' population. genvector centrality because it trades with many other countries that
The income, trade, and network equations are jointly estimated, to- are, in turn, trading substantially with many other countries. We ob-
gether with the emissions per capita model using a three-stage serve that once we account for the direct effects of trade (column 2), ei-
least squares (3SLS) procedure.13 genvector is positive and significant for high-income economies but
In the next section, we present the results of the emissions per capita negative and statistically significant for low-income countries. High-
equation using a 3SLS to account for the simultaneity between trade, income countries that trade substantially with other countries that
network, income, and emissions per capita.14 also trade with many other countries may not benefit significantly
from their higher centrality. That higher centrality may not be translat-
ed into greater market access because these countries are likely to al-
ready be involved in several trade agreements. On the other hand, an
13
increase in eigenvector for low-income countries, holding trade, in-
We first attempted to estimate the system of equations using the GMM estimator
come, and other factors constant, would imply that the trade partners
since the GMM is more efficient than 3SLS under heteroskedasticity. Unfortunately, given
the large number of parameters to be estimated, we had convergence problems (not con- of these low-income countries are trading with many more countries,
cave function) when we implemented the GMM approach, hence we adopted the 3SLS ap- which are also trading with many other countries. This increase in prox-
proach. As pointed by Altonji and Segal (1996) and Ziliak (1997), GMM estimators that use imity to the most important countries in the world trade network may
many overidentifying restrictions, as in our case, can have very poor finite sample proper- facilitate the creation of new links in the future (thus increasing market
ties. The 3SLS is adopted because it is more efficient than the 2SLS, since the 3SLS takes into
account intertemporal correlation between the error terms of each equation.
access), the imports of environmentally friendly goods and the trans-
14
For the sake of brevity, the income, trade and network estimated equations are not mission of knowledge from central countries in the world trade net-
presented. The results are available upon request to the authors. work, leading to a reduction in the levels of carbon dioxide emissions.
17.5
7.6
7.6
.08
7.55
Exports,Imports
7.55
CO2 emissions
17
.075
CO2 emissions
Eigenvector
16.5
.07
7.5
7.5
.065
7.45
16
7.45
.06
15.5
7.4
.055
7.4
1995 2000 2005 2010

1995 2000 2005 2010 Year
Year
CO2 emissions Exports
CO2 emissions Eigenvector Imports
7.6
7.6
2.6
Out-closeness, In-closeness
.7
2.5
7.55
7.55
CO2 emissions
.6
CO2 emissions
Betweeness
2.4
7.5
.5
7.5
2.3
7.45
.4
7.45
2.2
7.4
.3
2.1
7.4
1995 2000 2005 2010

1995 2000 2005 2010
Year
Year
CO2 emissions Out-Closeness
CO2 emissions Betweenness In-Closeness
Fig. 2. CO2 emissions, trade and centrality measures by year. Notes: Figures are constructed from observations obtained by averaging all countries' CO2 emissions per capita and trade and
centrality measures for each year in 1996–2010.
In the tables, the Wald F-test of joint significance of the instruments and relationship between betweenness and environmental quality for low-
the Hansen J test for overidentifying restrictions imply that the instru- income economies, once we control for the direct effects of trade.
ments used in the 3SLS estimation are enough correlated with the Low-income economies with higher betweenness centrality (higher
eigenvector variable and valid. criticality) may gain higher market power and market access because
The criticality of a country, i.e. the importance of a country as an in- their more important role as an intermediary in the trade of all the
termediary in the trade between other countries, is captured by the be- other countries may facilitate their participation in trade agreements
tweenness centrality (Choi et al., 2014). This variable measures the and increase their bargaining power in trade deals. Thus, their higher
number of shortest paths from all countries to all other countries that market power, as a consequence of their better position in the network,
pass through that country and captures the role of a country as a hub may lead to an increase in imports of energy-efficient technology and
in the trade network (De Benedictis and Tajoli, 2011). Higher between- imports of environmentally friendly goods through two channels:
ness centrality would imply that the country is becoming more impor- greater market access thanks to new trade agreements and higher
tant in connecting the trade of other countries in the world; in other market power that may decrease the final price of the goods imported.
words, their criticality or brokerage power is higher. The results from This transfer of technologies, imports of goods that are more
the betweenness centrality analysis (see columns 3 and 4 of Tables 3) environmentally friendly, and diffusion of knowledge from other coun-
show that greater criticality in the world trade network is associated tries in the world trade network may result in better environmental
with lower carbon dioxide emissions. However, this effect depends on quality.
the level of economic development of the country. We find that, holding Notice that high-income countries have, on average, a centrality that
trade volume fixed, if a high-income country's criticality increases, as a is three times higher than the centrality of low-income economies (see
result of higher trade among trade partners or trade partners' partners, Table 1). The positive effect of betweenness on the levels of carbon diox-
environmental quality is harmed. This lower environmental quality ide emissions for high-income economies and the negative relationship
may be explained by congestion externalities and our instrumental var- between betweenness and carbon dioxide emissions for low-income
iable strategy. Our instrumental variable strategy exploits exogenous economies suggest the existence of a threshold value after which the
variation of trade and betweenness centrality using variation in the betweenness or criticality of a country has a negative effect on environ-
population of the country and in the population of trade partners and mental quality. Countries with high betweenness centrality have less
trade partners' partners. If trade partners of country i are trading more scope to increase their market power from a better strategic position
among themselves or among trade partners' partners because their in the world trade network because they are already involved in many
populations are growing, they have fewer resources and less technology trade agreements, the increase in their bargaining power in trade nego-
available to export to country i; consequently, country i may have to in- tiations is limited and they are likely to already have an energy-efficient
crease its production or import goods that are less environmentally production system. Thus, for high-income economies, the negative ef-
friendly, thereby increasing the level of carbon dioxide emissions. In fects of congestion externalities on the environment exceed the gains
contrast, we find evidence of a positive and statistically significant from market power.
Table 3
CO2 emissions and connectivity: Three-stage least squares estimation.
(1) 3SLS (2) 3SLS (3) 3SLS (4) 3SLS (5) 3SLS (6) 3SLS (7) 3SLS (8) 3SLS (9) 3SLS (10) 3SLS
Log (Eigenvector) −0.462⁎ −0.432

(0.279) (0.278)
Log (betweenness) −0.107⁎⁎⁎ −0.123⁎⁎⁎
(0.022) (0.024)
Log (out-closeness) −1.889⁎⁎⁎ −1.935⁎⁎⁎
(0.579) (0.577)
Log (in-closeness) 1.097 0.855
(0.695) (0.788)
Closeness 1.026⁎ 0.716
(0.531) (0.602)
Log (trade) 0.129 0.036 −0.994 0.067 0.062
(0.098) (0.118) (0.101) (0.111) (0.111)
Foreign direct investment (% of GDP) −0.271⁎⁎⁎ −0.265⁎⁎⁎ −0.272⁎⁎⁎ −0.257⁎⁎⁎ −0.282⁎⁎⁎ −0.276⁎⁎⁎ −0.276⁎⁎⁎ −0.270⁎⁎⁎ −0.263⁎⁎⁎ −0.260⁎⁎⁎
(0.026) (0.026) (0.026) (0.027) (0.028) (0.029) (0.027) (0.027) (0.026) (0.026)
Industry output (% of GDP) 0.002⁎ 0.001 0.002⁎⁎ 0.001 0.002⁎⁎ 0.002⁎ 0.002⁎⁎ 0.002⁎ 0.002⁎⁎ 0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Log (per capita GDP ppp adjusted) 2.123⁎⁎⁎ 1.792⁎⁎⁎ 2.065⁎⁎⁎ 1.774⁎⁎⁎ 1.940⁎⁎⁎ 1.989⁎⁎⁎ 1.776⁎⁎⁎ 1.674⁎⁎⁎ 1.928⁎⁎⁎ 1.783⁎⁎⁎
(0.145) (0.280) (0.150) (0.058) (0.178) (0.293) (0.221) (0.286) (0.153) (0.284)
Log (per capita GDP ppp adjusted squared) −0.079⁎⁎⁎ −0.068⁎⁎⁎ −0.072⁎⁎⁎ −0.055⁎⁎⁎ −0.066⁎⁎⁎ −0.062⁎⁎⁎ −0.064⁎⁎⁎ −0.062⁎⁎⁎ −0.075⁎⁎⁎ −0.069⁎⁎⁎
(0.008) (0.011) (0.008) (0.012) (0.011) (0.013) (0.012) (0.012) (0.008) (0.011)
Economic institutional quality −0.011 −0.022 −0.025 −0.049⁎⁎ −0.010 −0.021 −0.020 −0.023 −0.019 −0.024
(0.016) (0.017) (0.017) (0.020) (0.015) (0.017) (0.017) (0.017) (0.017) (0.017)
Observations 1252 1252 1252 1252 1252 1252 1252 1252 1252 1252
Country FE YES YES YES YES YES YES YES YES YES YES
Year FE YES YES YES YES YES YES YES YES YES YES
Wald F−statistic 14.92 39.73 7.00 69.23 38.37 33.65 53.68 7.45 3.14 40.40
Hansen J test (p-value) 0.2093 0.2551 0.2066 0.3514 0.9427 0.8223 0.2207 0.2911 0.4899 0.6616
Adjusted R2 0.9935 0.9935 0.9813 0.9775 0.9916 0.9912 0.9936 0.9936 0.9933 0.9935
Notes: The results are obtained using three-stage least squares procedure. We regress per capita GDP on lagged per capita GDP, population, investment, human capital formation, year
dummies, and country dummies. Trade volume (sum of exports and imports) is regressed on population, land area, year dummies, country dummies, the network variable, trade partners'
population and trade partners' partners population. The network equation is regressed on population, land area, income per capita, year dummies, country dummies, trade, trade partners'
population and trade partners' partners population. Trade partners' population and trade partners' partners population are the instrumental variables used to obtain exogenous variation
in trade and network. The Wald F statistic is a joint significance test of the instruments. The Hansen J (p-value) is a test of overidentification restrictions in the network equation. These
regressions are obtained using a sample of 96 countries over the period 1996 to 2010. Robust standard errors in parenthesis.
⁎ p b 0.1.
⁎⁎ p b 0.05.
⁎⁎⁎ p b 0.01.
With regard to the out-closeness centrality measure, we find it sig- countries, the effect is insignificant. This result, robust once we consider
nificant and environmentally positive for high-income countries. In par- other types of pollutants, such as SO2, can be understood as part of the
ticular, a 1% increase in out-closeness, i.e. the proximity of the country to disparity found in the literature regarding the contribution of sectoral
its trading partners as an exporter, leads to a decrease of 4.46% in the shifts to energy intensity, as pointed out by Garbaccio and Jorgenson
level of carbon dioxide emissions, holding trade and other factors con- (1999).
stant. Being closer to trade partners as an exporter may facilitate the Moreover, the level of per capita GDP, namely, the scale effect, con-
transfer of knowledge (e.g. production techniques that are more envi- tributes to environmental degradation. Consistent with the EKC, we
ronmentally friendly) from these trade partners to the country, leading find a turning point, captured by the square of per capita GDP, which in-
to a decrease in the level of carbon dioxide emissions. dicates that as long as countries are more developed, there is a
No significant effects are found for the in-closeness centrality, but decoupling behaviour between the level of emissions and per capita
they are for closeness centrality in the case of low-income economies. GDP (Tables 3–5). This result is found not only for the whole sample
Contrary to high-income and out-closeness case, being closer to trade of countries but also for high- and low-income economies. The inverted
partners both as an exporter and as an importer at the same time (close- U-shaped relationship between economic growth and CO2 emissions is
ness centrality) is environmentally detrimental for low-income coun- also found in Cole et al. (1997), Galeotti and Lanza (1999), Apergis and
tries (columns 9 and 10 in Table 5). Payne (2009), Lean and Smyth (2010), and Saboori et al. (2012). Once
The results presented in Tables 3–5 show that the direct effects of we analyse SO2 emissions, we do find some evidence supporting the
trade on environmental quality are negative. In most of the regressions, EKC, but it is not robust for all the specifications considered in our
we find that trade is positively related to the level of carbon dioxide work. This finding suggests that the EKC depends critically on the type
emissions, holding the position of the country in the trade network of pollutant examined.
and other factors fixed. In other words, once we account for the scale, The technique effect is captured by FDI. According to our results, FDI is
composition, technique effects, and network effects from trade, we one of the mechanisms in improving the environmental degradation
still find that trade volume is detrimental to the environment regardless measured by carbon dioxide emissions in the case of the whole sample
of the level of economic development of the country. and high-income countries. The use of advanced technology and better
We have included other relevant factors in our study. The composi- managerial skills by multinationals improves environmental quality.
tion effect is measured by the share of industrial output as in Panayotou This finding is aligned with the results obtained by Mielnik and
(1997). Results show a positive sign when significant for the whole Goldemberg (2002) and Perkins and Neumayer (2008). However, our
sample of countries examined (Table 3) and for the high-income coun- results also support the PHH in the case of low-income countries, as
tries (Table 4), indicating that there is much room for improvement in Acharyya (2009) does in the case of India. We found that in these econ-
the battle against environmental degradation by promoting policies omies, the presence of multinationals leads to a higher level of carbon
aimed at change across sectors. However, in the case of low-income dioxide emissions. These companies often search for new locations for
Table 4
CO2 emissions and connectivity: High-income economies.
Log (eigenvector) −0.466⁎⁎ 0.536⁎⁎

(0.184) (0.232)
Log (betweenness) 0.017 0.110⁎⁎⁎
(0.020) (0.025)
Log (out-closeness) 0.763 −4.457⁎⁎⁎
(0.856) (1.284)
Log (in-closeness) 3.192⁎⁎⁎ 0.159
(0.836) (1.096)
Closeness 1.218⁎⁎ −0.569
(0.519) (0.636)
Log (trade) 0.931⁎⁎⁎ 1.285⁎⁎⁎ 1.078⁎⁎⁎ 0.604⁎⁎⁎ 0.697⁎⁎⁎
(0.140) (0.140) (0.169) (0.153) (0.145)
Foreign direct investment (% of GDP) −0.356⁎⁎⁎ −0.364⁎⁎⁎ −0.347⁎⁎⁎ −0.373⁎⁎⁎ −0.348⁎⁎⁎ −0.372⁎⁎⁎ −0.373⁎⁎⁎ −−0.384⁎⁎⁎ −0.356⁎⁎⁎ -0.367⁎⁎⁎
(0.024) (0.025) (0.024) (0.025) (0.024) (0.025) (0.026) (0.026) (0.024) (0.025)
Industry output (% of GDP) 0.009⁎⁎⁎ 0.004⁎ 0.009⁎⁎⁎ 0.004⁎⁎ 0.009⁎⁎⁎ 0.007⁎⁎⁎ 0.009⁎⁎⁎ 0.005⁎⁎ 0.008⁎⁎⁎ 0.004⁎
(0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.002) (0.002) (0.002) (0.002)
Log (per capita GDP ppp adjusted) 3.095⁎⁎⁎ 0.245 2.802⁎⁎⁎ −0.426 2.760⁎⁎⁎ 0.676 2.500⁎⁎⁎ 1.120⁎⁎ 2.647⁎⁎⁎ 0.985⁎⁎
(0.239) (0.550) (0.230) (1.482) (0.267) (0.475) (0.241) (0.458) (0.235) (0.461)
Log (per capita GDP ppp adjusted squared) −0.145⁎⁎⁎ −0.053⁎⁎ −0.133⁎⁎⁎ −0.046⁎⁎ −0.131⁎⁎⁎ −0.075⁎⁎⁎ −0.118⁎⁎⁎ −0.076⁎⁎⁎ −0.127⁎⁎⁎ −0.072⁎⁎⁎
(0.013) (0.023) (0.013) (0.021) (0.015) (0.020) (0.014) (0.019) (0.013) (0.019)
Economic institutional quality 0.039⁎⁎ −0.006 0.042⁎⁎ 0.003 0.033 0.019 0.028 −0.008 0.034⁎ −0.006
(0.016) (0.020) (0.017) (0.020) (0.023) (0.024) (0.017) (0.019) (0.020) (0.021)
Observations 836 836 836 836 836 836 836 836 836 836
Wald F statistic 42.21 16.98 2.38 34.97 20.21 18.05 185.68 112.23 64.03 45.74
Hansen test (J statistic) 0.1528 0.1658 0.5743 0.4810 0.2728 0.1460 0.1565 0.1723 0.1805 0.1596
Adjusted R2 0.9812 0.9656 0.9800 0.8938 0.9807 0.9510 0.9813 0.9757 0.9801 0.9746
in trade and network. The Wald F−statistic is a joint significance test of the instruments. The Hansen J (p-value) is a test of overidentification restrictions in the network equation. These
⁎ p b 0.1.
⁎⁎ p b 0.05.
⁎⁎⁎ p b 0.01.
their plants where the environmental regulations are less strict and the traditional trade models. We propose two indirect effects from
environmental awareness is lower than in high-income countries. trade: congestion externalities and market power. The explicit consid-
However, when we consider other pollutants that are more locally eration of indirect effects in the world trade network provides new in-
focused, such as SO2 emissions, we do not find any effect (see Table 6). sights into the relationship between trade and the environment. The
Finally, we have included the quality of institutions in our analysis. indirect effects are estimated using the world trade network, construct-
Our results suggest that the quality of institutions is a relevant factor ed using bilateral trade flows from 1996 to 2010, and centrality mea-
in reducing carbon dioxide emissions, especially in low-income coun- sures to establish the degree of inter-connectivity of each country in
tries. This is consistent with previous evidence, such as that of the world trade network. Because these network measures, income,
Panayotou (1997), who argued that policies and institutions can signif- trade and environmental quality are endogenously determined, we
icantly reduce environmental degradation, even if a country's income estimate the environmental, trade, network and income equations si-
level is low. In addition, as pointed out by Farzin and Bond (2006), de- multaneously, using the three-stage least square estimator and instru-
mocracy has been considered a conduit through which agents can exer- mental variables. We also split our sample into high- and low-income
cise their preferences for environmental quality more effectively than countries to test the PHH and FEH. Moreover, we control our estimates
under an autocratic regime. There is an exception in our results for by traditional factors, such as the scale effect, composition effect, the
high-income countries, which is that economic institutional quality in- technique effect, and other country-specific characteristics.
creases the level of emissions when it becomes significant. In these Our results show the relevance of the trade network for environ-
economies, the quality of institutions does not show too much variation mental implications. Not only does the volume of trade affect environ-
and probably cannot offset the effects of further growth, leading to an mental quality but so also does the position of each country in the
increase in the levels of emissions. This unexpected finding is world trade network. We find that indirect effects improve environ-
aligned with relatively recent evidence, like that of Hosseini and mental quality in low-income countries but have a negative impact on
Kaneko (2013), who find that democracy can increase environmental the environment of high-income economies. In addition, we find evi-
degradation. dence of the technique effect captured by FDI. However, results differ
between high- and low-income countries. Our findings suggest that
5. Conclusions the entrance of multinationals into developing countries damages envi-
ronmental quality, supporting the PHH. We also find evidence for the
The relationship between trade and the environment has generated EKC in our sample of countries.
a large body of empirical research in the literature. However, the evi- Finally, the composition effect is the real challenge for environmen-
dence is mixed. We argue in this work that the mixed evidence in the tal policy because when it is significant, it shows a negative effect on the
relationship between trade and the environment can be explained in environment. By contrast, the quality of institutions seems to play a crit-
part by the omission of the indirect (or network) effects of trade in ical role in improving the environmental quality but only in developing
Table 5
CO2 emissions and connectivity: Low-income economies.
Log (Eigenvector) −0.608 −3.176⁎⁎

(1.154) (1.414)
Log (betweenness) −0.028 −0.156⁎⁎⁎
(0.020) (0.031)
Log (out-closeness) 0.292 0.416
(0.477) (0.473)
Log (in-closeness) 0.997 0.502
(0.923) (0.955)
Closeness 3.111⁎⁎⁎ 2.231⁎⁎⁎
(0.604) (0.681)
Log (trade) 0.443⁎⁎⁎ 1.067⁎⁎⁎ 0.204⁎ 0.183 −0.034
(0.138) (0.178) (0.113) (0.116) (0.134)
Foreign direct investment(% of GDP 0.275⁎⁎⁎ 0.393⁎⁎⁎ 0.288⁎⁎⁎ 0.521⁎⁎⁎ 0.298⁎⁎⁎ 0.390⁎⁎⁎ 0.293⁎⁎⁎ 0.361⁎⁎⁎ 0.465⁎⁎⁎ 0.529⁎⁎⁎
(0.080) (0.101) (0.080) (0.111) (0.108) (0.120) (0.081) (0.094) (0.101) (0.105)
Industry output (% of GDP) −0.000 −0.001 −0.000 −0.000 −0.000 −0.001 −0.000 −0.000 0.001 0.000
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Log (per capita GDP ppp adjusted) 3.294⁎⁎⁎ 2.661⁎⁎⁎ 3.249⁎⁎⁎ 1.723⁎⁎⁎ 3.286⁎⁎⁎ 2.892⁎⁎⁎ 2.877⁎⁎⁎ 2.655⁎⁎⁎ 2.652⁎⁎⁎ 2.654⁎⁎⁎
(0.439) (0.499) (0.443) (0.552) (0.490) (0.523) (0.540) (0.550) (0.462) (0.494)
Log (per capita GDP ppp adjusted squared) −0.143⁎⁎⁎ −0.121⁎⁎⁎ −0.141⁎⁎⁎ −0.106⁎⁎⁎ −0.145⁎⁎⁎ −0.132⁎⁎⁎ −0.123⁎⁎⁎ −0.117⁎⁎⁎ −0.127⁎⁎⁎ −0.119⁎⁎⁎
(0.027) (0.029) (0.027) (0.030) (0.032) (0.032) (0.031) (0.031) (0.028) (0.028)
Economic institutional quality −0.111⁎⁎⁎ −0.128⁎⁎⁎ −0.112⁎⁎⁎ −0.178⁎⁎⁎ −0.108⁎⁎⁎ −0.129⁎⁎⁎ −0.114⁎⁎⁎ −0.125⁎⁎⁎ −0.103⁎⁎⁎ −0.125⁎⁎⁎
(0.035) (0.035) (0.036) (0.044) (0.037) (0.038) (0.035) (0.035) (0.035) (0.036)
Observations 416 416 416 416 416 416 416 416 416 416
Wald F statistic 9.92 13.86 9.37 6.42 33.07 64.58 3.04 4.03 3.31 26.87
Hansen J test (p−value) 0.4668 0.3832 0.5432 0.5187 0.5696 0.3900 0.5220 0.3384 0.7484 0.5120
Adjusted R2 0.9882 0.9874 0.9869 0.9463 0.9881 0.9878 0.9884 0.9885 0.9787 0.9840
dummies and country dummies. Trade volume (sum of exports and imports) is regressed on population, land area, year dummies, country dummies, the network variable, trade partners'
regressions are obtained using a sample of 34 low-income countries over the period 1996 to 2010. Robust standard errors in parenthesis.
⁎ p b 0.1.
⁎⁎ p b 0.05.
⁎⁎⁎ p b 0.01.
countries. Significant policy implications emerge from our results. First, - Foreign Direct Investments, 1996-2010: Inward and Outward
strategic market power in international trade agreements seems to play Foreign Direct Investment Stock. The outward FDI stock is the
a relevant role. Second, a stronger environmental policy in developing value of the resident investors' equity in and net loans to enterprises
countries for multinationals may help to alleviate environmental degra- in foreign economies. The inward FDI stock is the value of foreign
dation. Third, policies aimed at promoting a shift from heavy to light investors' equity in and net loans to enterprises resident in
industry and/or towards the service sector may help to improve envi- the reporting economy. FDI stocks are measured in USD.
ronmental quality. Finally, policies that foster economic development Source: UNCTADSTAT, United Nations Conference on Trade and
may help countries to decouple from level of emissions. Development.
- Industry output, 1996–2010 (% of GDP): The share of the economy
Appendix A. Description of variables and data ources that stems from industrial production as a percentage of GDP. It
comprises value added in mining, manufacturing, construction,
- GDP per capita (1996–2010): Gross domestic product divided by electricity, water, and gas. Source: World Bank via The Quality of
midyear population. GDP is the sum of gross value added by all res- Government Dataset (version 6 April 2002).
ident producers in the economy plus any product taxes and minus - Economic Institutional Quality, 1996–2010 (relative factor scores):
any subsidies not included in the value of the products. It is calculat- This indicator is constructed as explained in Kunčič (2014)) using
ed without making deductions for depreciation of fabricated assets a set of institutional proxies for: financial freedom, business free-
or for depletion and degradation of natural resources. Data are in dom, regulatory quality, freedom of the press, freedom to own for-
current U.S. dollars. Source: World Bank. eign currency bank accounts, credit market regulations, labour
- CO2 emissions (1996–2010 annual in kilotonnes per capita): Carbon market regulations, business regulations, foreign ownership/invest-
dioxide emissions are those stemming from the burning of fossil ment restrictions, capital controls and investment profile. Source:
fuels and the manufacture of cement. They include carbon dioxide Institutional Quality Dataset.
produced during consumption of solid, liquid, and gas fuels and - Investment share of GDP (%), 1996-2010: The share of capital as a
gas flaring. Source: World Bank. percentage of GDP. Source: The Quality of Government dataset.
- SO2 emissions (2000–2005): based on estimates of emissions com- - Human Capital Index, 1996–2010: average years of schooling.
piled by the Netherlands Environment Assessment Agencys Emis- Source: Penn World Table 8.0.
sion Database for Global Atmospheric Research (EDGAR). The - Land area: country's total area, excluding area under inland water
variable is measured as tons of emissions per populated land. bodies, national claims to continental shelf, and exclusive economic
Source: The Quality of Government Dataset. zones. In most cases the definition of inland water bodies includes
- Bilateral Total Trade Flows (1996-2010): Data obtained from United major rivers and lakes. Source: The Quality of Government Dataset
Nations COMTRADE via World Integrated Trade Solution (WITS). (version 6 April 2002).
Table 6
Robustness check (dependent variable: SO2 emissions).
Log (Eigenvector) 4.393⁎⁎⁎ 4.526⁎⁎⁎

(0.734) (0.738)
Log (betweenness) 0.015 0.018
(0.030) (0.030)
Log (out-closeness) 1.306 3.453⁎⁎
(1.302) (1.533)
Log (in-closeness) 9.300⁎⁎⁎ 8.642⁎⁎⁎
(2.498) (3.085)
Closeness 7.034⁎⁎⁎ 8.719⁎⁎⁎
(1.729) (2.195)
Log (trade) 0.243 0.228 0.578⁎⁎ −0.307 −0.752⁎⁎⁎
(0.205) (0.202) (0.237) (0.255) (0.285)
Foreign direct investment (% of GDP) −0.038 −0.035 −0.040 −0.038 −0.062 −0.055 −0.126 −0.131 −0.097 −0.113
(0.085) (0.085) (0.096) (0.096) (0.086) (0.087) (0.093) (0.100) (0.089) (0.092)
Industry output (% of GDP) 0.018⁎⁎⁎ 0.018⁎⁎⁎ 0.017⁎⁎⁎ 0.018⁎⁎⁎ 0.015⁎⁎⁎ 0.013⁎⁎⁎ 0.016⁎⁎⁎ 0.016⁎⁎⁎ 0.009⁎ 0.006
(0.004) (0.004) (0.004) (0.004) (0.004) (0.005) (0.004) (0.004) (0.005) (0.005)
Log (per capita GDP ppp adjusted) 1.559⁎⁎ 1.151 2.300⁎⁎⁎ 1.934⁎⁎ 2.273⁎⁎⁎ 1.435⁎⁎ 1.993⁎⁎⁎ 2.554⁎⁎⁎ 2.927⁎⁎⁎ 4.468⁎⁎⁎
(0.628) (0.716) (0.754) (0.107) (0.678) (0.119) (0.010) (0.017) (0.727) (1.017)
Log (per capita GDP ppp adjusted squared) −0.081⁎⁎ −0.078⁎⁎ −0.103⁎⁎ −0.101⁎⁎ −0.104⁎⁎ −0.109⁎⁎ −0.097⁎⁎⁎ −0.103⁎⁎⁎ −0.168⁎⁎⁎ −0.202⁎⁎⁎
(0.037) (0.037) (0.046) (0.047) (0.042) (0.042) (0.037) (0.037) (0.046) (0.052)
Economic institutional quality −0.031 −0.031 −0.032 −0.032 −0.032 −0.036 −0.098⁎⁎ −0.102⁎⁎ −0.082⁎ −0.095⁎⁎
(0.041) (0.043) (0.044) (0.047) (0.042) (0.043) (0.048) (0.048) (0.043) (0.043)
Observations 493 493 493 493 493 493 493 493 493 493
Wald F statistic 23.95 34.30 1.97 17.00 12.32 3.65 4.48 31.69 8.52 32.09
Hansen J test (p-value) 0.3631 0.3784 0.3520 0.1308 0.2482 0.2234 0.9441 0.3992 0.825 0.2948
Adjusted R2 0.9855 0.9843 0.9871 0.9864 0.9869 0.9810 0.9837 0.9852 0.9732 0.9715
population, and trade partners' partners population. The network equation is regressed on population, land area, income per capita, year dummies, country dummies, trade, trade partners'
population, and trade partners' partners population. Trade partners' population and trade partners' partners population are the instrumental variables used to obtain exogenous variation
⁎ p b 0.1.
⁎⁎ p b 0.05.
⁎⁎⁎ p b 0.01.
- Population, 1996–2010: Total population is based on the de facto Bernard, A.B., Jensen, J.B., Redding, S.J., Schott, P.K., 2007. Firms in international trade.
J. Econ. Perspect. 21, 105–130.
definition of population, which counts all residents regardless of
Birdsall, N., Wheeler, D., 1993. Trade policy and industrial pollution in Latin America:
legal status or citizenship–except for refugees not permanently set- where are the pollution havens? J. Environ. Dev. 2, 137–149.
tled in the country of asylum, who are generally considered part of Chaney, T., 2014. The network structure of international trade. Am. Econ. Rev. 104 (10),
the population of their country of origin. The values shown are mid- 1–39.
Choi, S., Galeotti, A., Goyal, S., 2014. Trading in networks: theory and experiments. 8.
year estimates. Source: World Bank. Cambridge–INET Working Paper.
- Income group: Countries are classified according to World Bank Cole, M., 2004. Trade, the pollution haven hypothesis and the environmental Kuznets
February 2014 gross national income per capita classification. curve: examining the linkages. Ecol. Econ. 48, 71–81.
Cole, M., Elliott, R., 2003. Determining the trade-environment composition effect: the role
Groups are: low income ($ 4085 or less) and high income ($ 4086 of capital, labor and environmental regulations. J. Environ. Econ. Manag. 46, 363–383.
or more). Source: World Bank. Cole, M., Rayner, A.J., Bates, J.M., 1997. The Environmental Kuznets Curve: an empirical
analysis. Environ. Dev. Econ. 2, 401–416.
Copeland, B.R., Taylor, M.S., 2004. Trade, growth and the environment. J. Econ. Lit. 42,
7–71.
Appendix B. Supplementary data De Benedictis, L., Tajoli, L., 2011. The world trade network. World Econ. 34, 1417–1454.
De Benedictis, L., Nenci, S., Santoni, G., Tajoli, L., Vicarelli, C., 2014. Network analysis of
world trade using the BACI-CEPII dataset”. Glob. Econ. J. 14, 287–343.
Supplementary data to this article can be found online at http://dx. Ductor, L., 2015. Does co-authorship lead to higher academic productivity? Oxf. Bull. Econ.
doi.org/10.1016/j.eneco.2015.09.008. Stat. 77, 385–407.
Fagiolo, G., Reyes, J., Schiavo, S., 2009. World-trade web: topological properties, dynamics,
and evolution. Phys. Rev. E 79, 036115.
References Farzin, Y., Bond, C., 2006. Democracy and environmental quality. J. Dev. Econ. 81,
213–235.
Acharyya, J., 2009. FDI growth and the environment: evidence from India on CO 2 emis- Felbermayr, G., Koheler, W., 2005. Exploring the extensive and intensive margins of world
sion during the last two decades. J. Econ. Dev. 34, 43–58. trade. Rev. World Econ. 142, 642–674.
Altonji, J.G., Segal, L.M., 1996. Small-sample bias in GMM estimation of covariance struc- Frankel, J.A., Rose, A.K., 2005. Is trade good or bad for the environment? Sorting out the
tures. J. Bus. Econ. Stat. 14, 353–366. causality”. Rev. Econ. Stat. 87, 85–91.
Antweiler, W., Copeland, B.R., Taylor, M.S., 2001. Is free trade good for the environment?”. Galeotti, M., Lanza, A., 1999. Richer and cleaner? A study on carbon dioxide emissions in
Am. Econ. Rev. 91, 877–908. developing countries. Energy Policy 27, 565–573.
Apergis, N., Payne, J.E., 2009. CO 2 emissions, energy usage, and output in Central America. Garbaccio, R.F., Jorgenson, D.W., 1999. Why has the energy-output ratio fallen in China?
Energy Policy 37, 3282–3286. Energy J. 20, 63–91.
Baghdadi, L., Martinez-Zarzoso, I., Zitouna, H., 2013. Are RTA agreements with environ- Garlaschelli, D., Loffredo, M., 2005. Structure and evolution of the world trade network.
mental provisions reducing emissions? J. Int. Econ. 90, 378–390. Phys. A 355, 138–144.
Bank, World, 2014. World Bank Development Indicators, World Bank. http://data. Grossman, G.M., Krueger, A.B., 1991. Environmental impacts of North American free trade
worldbank.org/indicator. agreement. National Bureau of Economic Research, Working Paper. 3914.
Bernard, A.B., Jensen, J.B., 1995. Exporters, jobs and wages in US Manufacturing: 1976– Grossman, G.M., Krueger, 1995. Economic growth and the environment. Q. J. Econ. 110,
1987. Brook. Pap. Econ. Act.: Microeconomics 67–112. 353–357.
Helpman, E., Melitz, M., Rubinsteain, Y., 2008. Estimating trade flows: trading partners Managi, S., Hibiki, A., Tsurumi, T., 2009. Does trade openness improve environmental
and trading volumes. Q. J. Econ. 123, 441–487. quality?”. J. Environ. Econ. Manag. 58 (3), 346–363.
Herrerias, M.J., Cuadros, A., Orts, V., 2013. Energy intensity and investment ownership Mani, M., Wheeler, D., 1998. In search of pollution havens? Dirty industry in the world
across Chinese provinces. Energy Econ. 36, 286–298. economy, 1960-1995. J. Environ. Dev. 7, 215–247.
Hosseini, H., Kaneko, S., 2013. Can environmental quality Spread through Institutions? Mielnik, O., Goldemberg, J., 2002. Foreign direct investment and decoupling between en-
Energy Policy 56, 312–321. ergy and gross domestic product in developing countries. Energy Policy 30, 87–89.
Jackson, M.O., Wolinsky, A., 1996. A strategic model of social and economic networks. Panayotou, T., 1997. Demystifying the Environmental Kuznets curve: turning a black box
J. Econ. Theory 71, 44–74. into a policy tool. Environ. Dev. Econ. 2, 465–484.
Janicke, M., Binder, H.M., 1997. Dirty industries: patterns of change in industrial coun- Perkins, R., Neumayer, E., 2008. Fostering environment-efficiency through transnational
tries”. Environ. Resour. Econ. 9, 467–491. linkages? Trajectories of CO2 and SO2, 1980–2000. Environ. Plan. A 40, 2970–2989.
Jayadevappa, R., Chhatre, S., 2000. International trade and environmental quality: a sur- Saboori, B., Sulaiman, J., Mohd, S., 2012. Economic growth and CO2 emissions in Malaysia:
vey. Ecol. Econ. 32, 175–194. a cointegration analysis of the Environmental Kuznets Curve. Energy Policy 51,
Judson, R.A., Schmalensee, R., Stoker, T.M., 1999. World carbon dioxide emissions: 1950– 184–191.
2050. Rev. Econ. Stat. 80, 15–27. Sato, M., Dechezleprtre, A., 2013. Asymmetric Industrial Energy Prices and International
Kali, R., Reyes, J., 2007. The architecture of globalization: a network approach to interna- Trade. CEP discussion paper. 1337.
tional economic integratione. J. Int. Bus. Stud. 38, 595–620. Serrano, M.A., Boguñá, 2003. Topology of the world trade web. Phys. Rev. E 51, 1–4.
Keller, W., 2004. International technology diffusion. J. Econ. Lit. 42, 752–782. Serrano, M.A., Boguñá, Vespignani, A., 2007. Patterns of dominant flows in the world
Kleemann, L., Abdulai, A., 2013. The impact of trade and economic growth on the environ- trade web. J. Econ. Interac. Coord. 2, 111–124.
ment: revisiting the cross-country evidence. J. Int. Dev. 25, 180–205. Taylor, M.S., 2005. Unbundling the pollution haven hypothesis. Adv. Econ. Anal. Policy 3,
Krugman, P., 1980. Scale economies, product differentiation, and the pattern of trade. Am. 1–28.
Econ. Rev. 70, 950–959. Termushoev, U., 2006. Pollution Haven Hypothesis or Factor Endowment Hypothesis:
Kunčič, A., 2014. Institutional quality dataset. J. Institutional Econ. 10, 135–161. Theory and Empirical Examination for the US and China. CERGE-EI Working Paper.
Lean, H.H., Smyth, R., 2010. CO2 emissions, electricity consumption and output in ASEAN. 292.
Appl. Energy 87, 1858–1864. Wagner, G., 2010. Energy content of world trade. Energy Policy 38, 7710–7721.
Lucas, R.E.B., Wheeler, D., Hettige, H., 1992. Economic development, environmental regu- Ziliak, J.P., 1997. “Efficient estimation with panel data when instruments are
lation and the international migration of toxic industrial pollution: 1960–88. In: Low, predetermined: an empirical comparison of moment-condition estimators. J. Bus.
P. (Ed.), International Trade and the Environment, World Bank Discussion Paper. 159. Econ. Stat. 15 (4), 419–431.

Trade and Environment

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Trade and Environment

Uploaded by

Copyright:

Available Formats

Energy Economics 52 (2015) 55–68

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/eneco

The world trade network and the environment☆

1. Introduction agriculture to service is expected to decrease the level of emissions);

High income (836 obs.) Low income (416 obs.)

United States 1 1 5 5 3 HIGH

Volume of Exports Volume of Imports

Out-Closeness centrality In-Closeness centrality

Eigenvector centrality Betweenness centrality

High income Low income High income Low income

1995 2000 2005 2010

1995 2000 2005 2010

Log (Eigenvector) −0.462⁎ −0.432

Log (eigenvector) −0.466⁎⁎ 0.536⁎⁎

Log (Eigenvector) −0.608 −3.176⁎⁎

Log (Eigenvector) 4.393⁎⁎⁎ 4.526⁎⁎⁎

You might also like