International Trade and Industrial Air Pollution: An Analysis of Conditional Effects

Master Thesis

International Trade and Industrial Air Pollution:

An Analysis of Conditional Effects

Faculty of

2

Abstract

Keywords: International trade – Industrial air pollution – Pollution halo/haven – Factor

endowments

Abstract. This paper aims to investigate the impact of international trade on industrial air

pollution (SO2, NOX, Co2). The effects are explored on a large sample of heterogeneous countries, using panel data on 70 countries between 1995 and 2008. The existence and conditionality of trade-induced effects on industrial emissions are investigated with the use of an extensive interaction model, and guided by the pollution haven, factor endowments, and

pollution halo hypotheses. There is a dominance of a factor endowments effect found for

3

Contents

4.1 Income, capital-labor ratio, and trade 20

4.2 Income, capital-labor ratio, education, and offsetting results 23

4

1. Introduction

Environmental issues concern us on a global scale, but the international community has not yet found an answer to these difficulties. Unprecedented environmental problems are felt throughout the world, while the economy keeps growing. ‘The Limits to Growth’ by the Club of Rome (1974) is a well-known report considering the relationship between environmental problems and economic growth and development, warning there is an inherent tragedy in pursuing infinite economic growth on an earth with finite capacities. This report is heavily debated and criticized, but is still recent regarding the concerns about the relationship between economic growth and the environment.

Global trade consists of about 60 per cent of trade in intermediate goods and services (UNCTAD, 2013). The relationship between the environment, and more specifically, pollution, and trade has been often debated in literature, with a convincing consensus still to be reached. It is important to acquire more and conclusive evidence on this relationship, as it then can become clear to what extent collective action is needed. At this point, international climate change treaties are being signed and on the other hand there are international trade treaties. However, if both trade and the environment proves to be intertwined, then institutional trade structures could start to include environmental concerns as well (Millimet et al., 2011). For example, the World Trade Organization (WTO) may start to impede countries from choosing ‘bad’ environmental policies when that turns out to attract trade flows. It remains a question to what extent these problems should be coordinated globally and to what extent this can be done locally. In a world with more pressing environmental concerns, we have also seen an enormous increase in international trade and worldwide interconnectedness.

5

capital-labor-ratio, there is often higher environmental stringency, also resulting in possibly offsetting cost-effects. These potentially offsetting trade-induced effects of both hypotheses could therefore be in practice much smaller as hypothesized (Cole and Elliott, 2003, Antweiler et al., 2001). Finally, the pollution halo hypothesis predicts that openness to trade can function as a vehicle for the transfer of technology (Abdouli et al., 2015), as trading partners become acquainted with cleaner technology and good practices in the field of environmental responsibility. Consequently, trade liberalization can actually influence the upgrade of environmental standards in a country.

As stated, surprisingly little conclusive evidence is found for the consequences of

trade liberalization on pollution. One of the factors contributing to this lack of evidence might be the existence of conditional effects on a possible influence of trade liberalization on industrial pollution (Zugravu-Soilita, 2017). Therefore, the question being asked in this research is: Is the influence of trade liberalization on industrial pollution conditional upon different country characteristics? To distinguish these conditional characteristics, the three hypotheses discussed above will guide the development of possible trade-induced effects on pollution. Conditions imposed upon the amount of trade in a country are relative per capita income, relative capital-labor ratio, and education level. In literature, there has been little research in discovering conditional effects, and if so only one of the three main hypotheses is addressed. Furthermore, focus is most often aimed at only one pollutant. This research will develop models where all three hypotheses can influence their effects on the final effect of trade and pollution. The main, trade-induced effects of the hypotheses outlined above are thus included in a broad interaction model. A large sample of 70 countries will be used in the period of 1995-2008, and estimations will be conducted upon three different pollutants. Moreover, the analysis will be applied for total trade in a country, but also for six different sectors.

It is found that environmental stringency can influence a decreasing effect of

6

2. Literature

The debate between the relationship between trade and the environment starts with the three main hypotheses in the area of globalization and environment. The pollution haven hypothesis is one of the most well-known but also one of the most heavily debated hypotheses. Even though empirical validation appears to be difficult, the line of reasoning behind this hypothesis remains popular. The richer countries move their pollution-intensive production to low-income countries in the form of foreign direct investment (FDI). On the other hand, applying the mechanism to trade liberalization means that low-income countries will specialize in pollution-intensive production. In either way, low-income countries will be faced with environmental degradation as a result of globalization pressures. Empirical validation seems to depend upon the different methodologies adopted, different pollutants included, and different relocating motivations. Taking these differences into account, the existence of pollution havens seems to be dependent upon the level of stringency in environmental regulations (e.g. Candau and Dienesch, 2017; Cole et al., 2006; Xing and Kolstad, 2002). Furthermore, the level of corruption opportunities in a country influence the coming into existence of pollution havens (Candau and Dienesch, 2017), but also the level of human capital (Lan et al., 2012), and levels of political stability. Here, more human capital and more political stability can reduce harmful effects on the environment even though there are pollution havens (Al-Mulali and Ozturk, 2015). On the other hand, the pollution haven hypothesis has also been discarded as a popular myth (Smarzynska and Wei, 2001), and two broad lines of criticism can be summarized. First of all, it is stated that environmental costs represent only a small share of total costs involved when a firm chooses a location. Secondly, and this relates to the factor endowments hypothesis, it is argued that polluting industries are often also the most capital-intensive industries. These industries are actually hard to relocate, as a comparative advantage of countries is hard to shift (Candau and Dienesch, 2017).

Antweiler et al., (2001) and Cole and Elliott (2003) build on this second line of

7

hypothesis we have to find this potential impact as being determined by relative capital and labor endowments. It is well understood that the north is more capital-abundant than the south. However, there is also a strong correlation between a sector’s capital intensity and its pollution intensity (Cole and Elliott, 2005). This means that capital-intensive industries are often also more intensive. Therefore, firms looking for cost advantages in pollution-intensive industries would do best to relocate to countries with a higher capital-abundancy, encountering relatively cheaper capital. This implies that pollution in these countries with higher capital-labor ratios would actually increase. At the same time, this means that the low-income, labor abundant countries would actually specialize in relatively cleaner, labor-intensive production. This reasoning goes against the pollution haven hypothesis, where these low-income countries are said to specialize in pollution-intensive industries.

Finally, and less dominantly present in research is the pollution halo hypothesis. This

hypothesis is an extension from the Porter Hypothesis, which entails that a well-designed stringent environmental policy in a nation can actually boost investments in cleaner technologies, innovation and output growth (GGKP, 2015). Even if multinationals and trading partners are searching for hosts with lower environmental stringency, they often already comply with stricter environmental regulations. Via this reasoning they are likely to export their good practices, and knowledge about cleaner production processes. This can have the consequence of rejecting the pollution haven hypothesis, but instead finding a halo effect after increased openness (Eskeland and Harrison, 2003). Gallagher and Zarksy (2007) have found that FDI can contribute in transferring more efficient production processes, ‘green’ technologies, and sustainable management practices. This line of reasoning can be extended for international trade in the sense that increasing international trade can bring countries the opportunity to import good environmental practices and ‘clean’ technologies in order to comply with an increasing demand for environmental stringency that follows increasing per capita incomes after trade liberalization. Therefore, a halo effect might also be found when analyzing international trade, and exploring the conditions upon which this effect can become present is very interesting.

Numerous researchers have studied the validity of one of these hypotheses, while

8

developed in the hypotheses below. Before developing the hypotheses, more insights in the aspects of the relationship between trade and pollution must be acquired. A common approach in literature is to start with making pollution a function of economic growth. This means pollution is decomposed in scale, technique, and composition effects (e.g. Frankel and Rose 2005, Cole et al, 2003, Grossman and Krueger, 1995). The scale effect refers to the increases in economic activity that follow from the increased market access resulting from liberalized trade. This can have negative environmental consequences (Antweiler, 2001). The technique effect refers to positive environmental consequences as a result of raised incomes due to trade and growth. The composition effect refers to changes in industrial structure after opening up to trade, which depend on the forces of comparative advantage. These three effects constitute the framework through which the mechanisms based on the three hypotheses above can influence pollution in a country.

The pollution haven hypothesis suggests that income per capita in a country is a crucial element in finding a pollution effect of increasing international trade. The low-income countries have a lower level of environmental stringency, which will boost specialization towards the pollution-intensive industries. On the other hand, high-income countries have a higher level of environmental stringency, leading to a specialization in cleaner production. This relationship between per capita income and pollution is the subject of the Environmental Kuznets Curve (EKC). This theory states that there is an inverted U-shaped relationship between per capita income and pollution (e.g. Cole 2003, Grossman and Krueger). An increase in industrial pollution will follow an initial scale expansion of the economy after an increase in income. However, at a certain point where income is ‘high’ enough, demand for environmental stringency will start to increase as individuals start to attach more value to a cleaner environment. Supply of environmental regulations will follow this increase in demand, and a decrease in pollution will follow. This shows there is a strong correlation between income and environmental stringency. There is also criticism on this reasoning, as there is evidence suggesting that the scale effect of economic growth dominates in pollution, which is not sufficiently compensated by increased environmental stringency (Stern, 2017). This concern will be accounted for in this research, as an extra condition will be imposed on the influence of pollution when there are rises in income per capita and international trade. Therefore, at this point income per capita will be used as a sufficient indication for the level of environmental stringency.

9

that countries with a higher per capita income. The higher per capita income a country has, the more they will specialize in non-pollution-intensive industries as the high costs associated with pollution (due to high environmental stringency) make these industries too expensive. The influence of trade liberalization on pollution is therefore likely to be conditional upon income per capita in a country. This mechanism can be brought about by the composition effect which is concerned with comparative advantage, and is summarized in Hypothesis 1:

H1: There is a decreasing effect on pollution as a result of increased trade when a country

has a relatively higher per capita income.

Furthermore, the factor endowments hypothesis suggest that the impact of trade liberalization on pollution depends on the capital-labor ratio of a country. Following this reasoning, when a country is capital-abundant, increasing engagement in trade will lead to specialization in capital-abundant industries. As stated before, pollution-intensive industries are often also capital-intensive. According to the theory of comparative advantage, these industries will be located in countries which are capital abundant. This means that the effect of trade on pollution will likely be dependent upon the capital-labor ratio of a country, implying that countries that are capital abundant will witness an increase in pollution after trade liberalization. The factor endowment effect would also become visible in the composition effect of pollution, as industrial output will change according to a specialization pattern:

H2: There is an increasing effect on pollution as a result of increased trade when a country

has a relatively higher capital-labor ratio.

10

decreasing pressure of pollution. A high per capita income implies more environmental stringency, and a high capital-labor ratio implies easier access to capital-abundant cleaner technologies. However, in the next hypothesis the focus will remain on forces that influence the composition effect of pollution, and therefore the next hypothesis will be tested:

H3: There is an increasing effect on pollution when a country has a relatively high

capital-labor ratio and a relatively high per capita income.

Finally, based on a trade-induced halo effect, increased openness to trade can actually be beneficial for the environment. Trade liberalization will improve the possibility for countries of importing less pollution-intensive technologies, which can result in decreasing pressures on pollution. Besides acquiring more access to ‘clean’ technologies, increased openness also provides the ability of learning from other countries and reaching a cleaner production process altogether or more effective environmental stringency. This halo effect can thus establish itself through the transfer of environment-friendly techniques and practices. This can be achieved in multiple ways. First of all, an increase in education can increase awareness among the public of environmental issues, who then start to make more environmentally deliberate choices with respect to trade. Second, increasing environmental stringency (increasing income per capita) can be designed and implemented more efficiently if education is higher in a country when a country is open to trade. Education increases absorptive implementation capacity, and openness to trade increases the access of effective environmental stringency policies of other countries. Third, countries that are more capital-abundant have the opportunity to more easily purchase capital-abundant, ‘clean’ technology. A higher education level in the country then ensures more efficient implementation and improvement of production processes. A trade-induced halo effect will function through the technique effect influencing pollution.

H4a: There is a decreasing effect on pollution as a result of increased trade when a country

has a higher level of education.

H4b: There is a decreasing effect on pollution as a result of increased trade when a country

has a higher income per capita and a higher education level.

H4c: There is a decreasing effect on pollution as a result of increased trade when a country

has a higher capital-labor ratio and a higher education level.

11

elaborate and complex empirical models have been established to deal with potential endogeneity concerns. For example, He (2006) developed an interesting simultaneous equation model of the scale, technique, and composition effects on pollution, but these approaches are also very constricted in data availability. Also, models using instrumental variables are only as good of quality as their instrument is. These models are also strongly limited due to difficulty in acquiring the necessary data. This makes country and cross-sectoral comparisons nearly impossible. In this research, the aim is not to validate the pollution haven hypothesis, the factor endowment hypothesis or the pollution halo hypothesis. Rather, these hypotheses form the basis of hypothesizing trade-induced effects on pollution. The models developed in this research will therefore not be limited to the mechanisms of just one hypothesis in the trade-environment literature, but it allows both (or all) hypotheses to become present in the form of trade-induced effects on pollution. This allows the adoption of a broader dataset as well as cross-country comparisons, adding to one of the biggest limitations in the trade-environment literature. Zugravu-Soilita (2017) recently developed one of the first attempts to capture all these potential effects in one model. They analyze FDI-induced effects on pollution in a causal, interaction model applied to a broad sample. The model that will be developed in this research builds upon their model, but in this research trade is the central variable of interest.

Another limitation of existing research in the field is that besides the focus on just one

12

Moreover, the hypotheses outlined above will also be applied to trade in different

sectors, to find out if the conditional trade-induced effects on pollution differ between sectors. Doytch and Uctum (2017) recently published a research showing there is a differing effect of FDI on pollution for different sectors. This provides a reason to think that different trading sectors also could be subject to different results, and therefore worth analyzing.

In this research, an interaction effect model will be developed based upon the

13

3. Methodology

From the literature review follows that trade liberalization can exert influence in the host country in three ways. First of all, it can directly raise the size of the economy. Second, it can bring know-how and different technologies beneficial to the environment. Finally, these flows can change the composition of the industrial structure. The first effect is expected to occur more or less uniformly throughout host countries, as more production ceteris paribus generates more pollution. The trade-induced technique and composition effects can be expected to be conditional on host country characteristics, as also captured in hypotheses (1-4). The PH-induced income effect would mean that the composition effect is dominated by the relative stringency of environmental regulations. A country with a lower level of environmental regulations will then have a comparative advantage in pollution-intensive production. Domination of the factor endowments-induced effect would mean that the composition effect is determined by the capital-labor ratio of a country. The pollution halo-induced effect can exert a positive downward pressure on pollution through the technique effect. This means international trade and pollution can be subject to opposing forces in the composition effect as shown by 1. The pure PH-induced effect would see the composition effect determined by the relative stringency of environmental regulations. A country with a lower level of environmental regulations will have a comparative advantage in pollution-intensive production. 2. The factor endowment induced effect means that countries that are capital abundant will see an increase in pollution-intensive production, whereas labor abundant countries will see a specialization in relatively cleaner, labor-intensive production. Finally, the technique effect can bring about downward pressures on pollution as trade can be used to access cleaner technology and production processes.

3.1 Empirical model

14

Antweiler et al. (2001), and Cole and Elliott (2003). This means that in this research, comparative advantage of a country will be driven by capital and labor endowments, as well as differences in environmental regulations. In allowing these conditions to influence the effect of trade on pollution, relative concepts will be used as comparative advantage is also a relative concept (Cole and Elliott, 2003). As comparative advantage of a country is decided based upon one country’s endowments relative to the others, the concepts representing the comparative advantage in this research will also be expressed as relative concepts. This means that countries with increasing relative income per capita, have an increasing income per capita relative to the average income per capita in the rest of the world. The relative stringency of environmental regulations is proxied within the model by relative income differences. As outlined in the literature review, there is a strong correlation between higher per capita income and environmental stringency. Also, it has been proven difficult to construct a reliable exogenous measure for the level of environmental regulations. In short, problems lie in the area that environmental stringency is subject to multidimensionality which makes it hard to capture stringency in one measure; simultaneity as some countries with economies of a certain shape or bad pollution problems may impose the strongest regulations; industrial composition influence the stringency of regulations; and capital vintage where new pollution sources also bring about more stringent regulatory standards (Brunel and Levinson, 2013). Therefore, income per capita will be adopted as an indication of environmental stringency.

Following the literature review, pollution (z) can be decomposed into scale,

composition, and technique effects. This can be summarized in the following equation:

𝑧𝑧̂ = 𝑆𝑆̂ + 𝜎𝜎� + 𝑒𝑒̂ + â (1)

Here, ^ denotes percentage change, s captures the scale effect and represents the change in

emissions that would occur if the size of the economy changed. The next variable 𝜎𝜎 captures

the composition effect and it represents the share of a pollution-intensive sectors in total

output. Finally, 𝑒𝑒 represents the technique effect and it shows the pollution intensity of the

dirty industry.

The scale effect and the technique effect will be captured by income per capita. The

15

pollution in two ways. First of all, there is an autonomous effect, where investment in cleaner technologies occurs due to exogenous reasons. The second is an ‘induced’ effect, where technological change happens as a result of increases in environmental stringency. This rise in environmental stringency is then endogenous, as the willingness to pay for better environmental quality rises with an increased per capita income. As outlined before, income per capita functions as an indication of the level of environmental stringency in an economy, and therefore this variable also captures the induced technique effect. Capturing the exogenous effects of the technique effect are subject to strong limitations in the broad sample that will be adopted here. There are a lot of data constraints to constructing such a measure as detailed value added data in production is necessary, but even if the data availability is high on for example R&D expenses, this does not automatically mean that there indeed will be effective investments in cleaner technologies. Income per capita is therefore initially a good indicator of the technique effect. Later, education will be added to let a trade-induced technique effect be influenced by the ability to efficiently implementing technological progress. Finally, the composition effect will be proxied by the capital-labor ratio per country. This follows common practice in literature (e.g. Antweiler et al., 2001, Cole and Elliott, 2005), and using this ratio has the strong advantage that it is for most countries available There are some concerns about using the capital-labor ratio (Dinda, 2006), with the main argument that some industries with high capital abundancy could also be the owner of cleaner capital-intensive technology, leading to less pollution. However, the construction of environmental performance indicators of different industries require detailed data on value added and sectoral pollution intensities, which are more limited available. Therefore, in this research the capital-labor ratio will be used to capture the composition effect, but these limitations will be kept in mind while interpreting the results. Finally, in all estimations a variable of trade will be included (â) to capture a possible direct and linear effect of trade on pollution. However, following Antweiler’s (2001) theory, this coefficient can be zero as there are no suggestions in literature that trade on itself has a direct impact on pollution. Rather, it is

more likely that the effect of trade depends upon certain country characteristics. Based on

this methodology, the model on which Hypothesis 1 will be estimated is the following: ln (𝐸𝐸𝑘𝑘𝑘𝑘) = 𝛼𝛼0+ 𝛼𝛼1ln (𝐾𝐾𝐿𝐿𝑘𝑘𝑘𝑘) + 𝛼𝛼2(𝐼𝐼𝑘𝑘𝑘𝑘) + 𝛼𝛼3𝑙𝑙𝑙𝑙(TRADE𝑘𝑘𝑘𝑘)+𝛼𝛼4ln(𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝐸𝐸𝑘𝑘𝑘𝑘) ln(𝑇𝑇𝐼𝐼𝑘𝑘𝑘𝑘) +

16

Where 𝐸𝐸_{𝑘𝑘𝑘𝑘} is industrial pollution per capita for year t in country k. 𝐾𝐾𝐿𝐿_{𝑘𝑘𝑘𝑘} captures the

capital-labor ratio and 𝐼𝐼_{𝑘𝑘𝑘𝑘} represents income per capita. TRADEtk represents trade intensity and is

constructed by exports plus imports divided by GDP. RIkt represents the relative income of a country and is constructed by dividing the per capita income by the world average per capita income. The interaction term captures possible income effects embedded in the trade-induced composition effect. t captures possible time effects in emissions per capita, and 𝜀𝜀_{𝑘𝑘𝑘𝑘} represents the error term. By interacting trade with relative income per capita the effect that trade exerts on pollution is captures when there is a certain level of environmental stringency. This will provide an indication if there is indeed a specialization in the direction of less pollution-intensive production when environmental stringency is higher. The interaction term will provide insights in this mechanism as it will show what the effect of trade on pollution will be for countries with higher relative incomes. According to the hypothesis, we expect coefficient 𝛼𝛼4 < 0, representing a decrease in pollution for countries with increasing relative per capita

incomes due to higher environmental stringency.

ln (𝐸𝐸𝑘𝑘𝑘𝑘) = 𝛼𝛼0+ 𝛼𝛼1ln (𝐾𝐾𝐿𝐿𝑘𝑘𝑘𝑘) + 𝛼𝛼2(𝐼𝐼𝑘𝑘𝑘𝑘) + 𝛼𝛼3𝑙𝑙𝑙𝑙(TRADE𝑘𝑘𝑘𝑘)+

𝛼𝛼4ln(𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝐸𝐸𝑘𝑘𝑘𝑘) ln(𝑇𝑇𝐾𝐾𝐿𝐿𝑘𝑘𝑘𝑘) + 𝑘𝑘 + 𝜀𝜀𝑘𝑘𝑘𝑘 (3)

Equation (3) will be used to explore the second hypothesis. RKLkt represents a country’s relative labor ratio. This relative ratio is constructed by dividing a country’s capital-labor ratio by the world average capital-capital-labor ratio. In this way, we can acquire insights in what happens with pollution when a country with a higher relative capital-labor ratio engages in international trade. The interaction term captures the trade-induced composition effect following the prediction of the factor endowments hypothesis. As this mechanism predicts a specialization in pollution-intensive industries for countries with a higher capital-labor ratio when engaging in trade, it is expected that 𝛼𝛼₄ > 0.

ln (𝐸𝐸𝑘𝑘𝑘𝑘) = 𝛼𝛼0+ 𝛼𝛼1ln (𝐾𝐾𝐿𝐿𝑘𝑘𝑘𝑘) + + 𝛼𝛼3ln (𝐼𝐼𝑘𝑘𝑘𝑘) + 𝛼𝛼5ln (𝐾𝐾𝐿𝐿𝐼𝐼𝑘𝑘𝑘𝑘) + 𝛼𝛼6ln (TRADE 𝑖𝑖𝑘𝑘𝑘𝑘) +

𝛼𝛼7ln(TRADE 𝑖𝑖𝑘𝑘𝑘𝑘) ln( RKL𝑘𝑘𝑘𝑘) + 𝛼𝛼8ln(𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝐸𝐸𝑖𝑖𝑘𝑘𝑘𝑘) 𝑙𝑙𝑙𝑙𝑇𝑇𝐼𝐼𝑘𝑘𝑘𝑘 +

𝛼𝛼9ln(𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝐸𝐸𝑖𝑖𝑘𝑘𝑘𝑘) ln(𝑇𝑇𝐾𝐾𝐿𝐿𝑘𝑘𝑘𝑘) ln (𝑇𝑇𝐼𝐼𝑘𝑘𝑘𝑘) + 𝑘𝑘 + 𝜀𝜀𝑖𝑖𝑘𝑘𝑘𝑘 (4)

17

country’s capital-labor ratio will be on pollution. This will shows what happens when these variables change, without being influenced by trade. The interaction terms of the previous two equations are also included in this model, as well as a three-way interaction term. This final term will give more specific insights in potentially offsetting trade-induced effects within the composition effect. This term can show what will happen to the effect of trade on pollution when a country has a high relative income as well as a high relative capital-labor ratio, and it can either be positive or negative depending on the direction of the offsetting effect.

ln (𝐸𝐸𝑘𝑘𝑘𝑘) = 𝛼𝛼0+ 𝛼𝛼1ln (𝐾𝐾𝐿𝐿𝑘𝑘𝑘𝑘) + + 𝛼𝛼3ln (𝐼𝐼𝑘𝑘𝑘𝑘) + 𝛼𝛼5ln (𝐾𝐾𝐿𝐿𝐼𝐼𝑘𝑘𝑘𝑘) + 𝛼𝛼6ln (TRADE 𝑖𝑖𝑘𝑘𝑘𝑘) +

𝛼𝛼7𝑙𝑙𝑙𝑙(𝐸𝐸𝑇𝑇𝐸𝐸𝐸𝐸𝑘𝑘𝑘𝑘) + 𝛼𝛼8𝑙𝑙𝑙𝑙(𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝐸𝐸 𝑖𝑖𝑘𝑘𝑘𝑘) 𝑙𝑙𝑙𝑙( 𝐸𝐸𝑇𝑇𝐸𝐸𝐸𝐸𝑘𝑘𝑘𝑘) + 𝛼𝛼9ln(TRADE 𝑖𝑖𝑘𝑘𝑘𝑘) ln( RKL𝑘𝑘𝑘𝑘) +

𝛼𝛼10ln(𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝐸𝐸𝑖𝑖𝑘𝑘𝑘𝑘) 𝑙𝑙𝑙𝑙𝑇𝑇𝐼𝐼𝑘𝑘𝑘𝑘 + 𝛼𝛼11ln(𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝐸𝐸𝑖𝑖𝑘𝑘𝑘𝑘) ln (𝑇𝑇𝐼𝐼𝑘𝑘𝑘𝑘)𝑙𝑙𝑙𝑙(𝐸𝐸𝑇𝑇𝐸𝐸𝐸𝐸𝑘𝑘𝑘𝑘) +

𝛼𝛼12ln(𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝐸𝐸𝑖𝑖𝑘𝑘𝑘𝑘) ln(𝑇𝑇𝐾𝐾𝐿𝐿𝑘𝑘𝑘𝑘) ln (𝐸𝐸𝑇𝑇𝐸𝐸𝐸𝐸𝑘𝑘𝑘𝑘) + 𝑘𝑘 + 𝜀𝜀𝑖𝑖𝑘𝑘𝑘𝑘 (5)

Equation (5) includes all possible conditioning effects on international trade to acquire the broadest insights as possible. Education is added as an extra condition here, in order to test for trade-induced technique effects (the pollution halo effect). This effect will be captured by the EDUC variable, representing the education level and returns on education in a country. An

expectation of 𝛼𝛼₈<0 shows the possible effect of increased education resulting in more

deliberate choices in trade due to increased awareness about the environment, captured in Trade*EDUC. Two three-way interaction variables are included, as education can exert different influence what happens in a country with a higher relative income, but also when a country has a higher relative capital-labor ratio. The interaction variable Trade*RI*EDUC will give an insight on H4b, where increased education can capture an increased awareness of the importance of environmental regulations, but also increased absorptive capacity to effectively implement environmental regulations. A higher demand for environmental stringency is one step, but possibly a certain education level in a country to reach the ability of effectively and efficiently implementing environmental regulations, resulting in an

expectation of 𝛼𝛼₁₁<0. The other three-way interaction term (Trade*RKL*EDUC) gives

18

3.2 Data

The country sample used is composed out of industrialized economies and emerging industrial economies as classified by UNIDO. This consists out of 90 countries, and it countries with a population below a threshold of 30000 will be dropped. Furthermore, for some countries there are no trade data at all available, so these countries will be dropped from the sample as well. The countries used in the analyses can be found in Appendix 1 accompanied by descriptive statistics. The selection of countries according to this classification is based on the reason that it is difficult to acquire detailed trade data for lower-income countries. A lot of these data is either unreliable, or only available for a short period of time. Also, data on pollution is less available for the low-income countries. In this research availability of detailed trade data, as well as on pollution is very important. That is why it is chosen to leave the low-income countries out of the analysis. For these two categories of countries, data is available for a reasonably broad time-period, while still showing income and capital-labor ratio differences. The time-period is 1995-2008, which is selected according to the broadest data availability.

The analyses will use three different pollutants: SO2, NOx and Co2. This allows

insights in robustness of mechanisms inherent in trade and pollution. Data on emissions per capita on SO2 and NOx are retrieved from the UNEP, and Co2 data from the World Bank. In conducting analyses on cross-country groups, it is common to use per capita emissions as a means of comparison among groups and over time. The relationship between economic variables and pollution can differ depending on whether pollutants are measured in emissions or concentrations as indicated on the literature on the EKC. Emissions data provide more information on wider environmental issues, whereas concentrations data gives insights in for example health effects of pollution (Cole and Elliott, 2003). When using concentrations data, a number of dummy variables have to be included in order to eliminate certain site-specific effects and time-varying effects. However, a benefit of concentrations data is the opportunity to separate the scale and technique effects (Antweiler et al., 2001). This is not the aim of this research however, as the main focus lies on trade-induced effects of pollution. Therefore, emissions data will be used.

Data on trade flows and sectoral trade data are collected from the UNCTAD Statistics

19

specific sectors in the total goods and services sector: total goods; total services; chemical sector; crude materials; manufactured goods; and financial services. In the sectoral analyses, some countries will be dropped from the sample as the sectoral trade data is not always available for every country. This results in balanced panel data for the analyses on total trade, and in unbalanced panel data for the analyses for sectoral trade.

Data on GDP per capita is retrieved from the World Bank. The relative income is then

constructed by taking the ratio of GDP per capita to the world average GDP per capita. The yearly world average GDP per capita is the result of the average of all countries included in the World Bank database. For the capital-labor ratio data from the Penn World Tables 9.0 is used. This ratio is constructed by adjusting the capital stock by the yearly capital depreciation rate before dividing it by the number of persons employed. In the Penn World Tables the capital stock is a combination of information on four assets. Relative capital-labor ratios are constructed by dividing the country’s capital-labor ratios by the world average. The yearly world average is constructed by all the countries available in the PWT. Data on education levels is also retrieved from the PWT, with the advantage that return on education is also included in the measurement.

In literature, trade and income are sometimes identified as endogenous factors in

20

4. Empirical results

4.1 Income, capital-labor ratio, and trade

Table 1: Single conditional effects for the influence of trade on industrial pollution

(1) (2) (3) (4) (5) (6)

VARIABLES lnSO2 lnNOX lnCo2 lnSO2 lnNOX lnCo2

lnKL -0.0382 0.0867** 0.0580 -0.144** 0.0464 0.0286 (0.0543) (0.0429) (0.0514) (0.0710) (0.0540) (0.0527) lnGDP -0.0677 0.0970*** 0.0381 -0.0396 0.119*** 0.0651** (0.0743) (0.0358) (0.0274) (0.0836) (0.0312) (0.0255) lnTrade -0.253* 0.0850 -0.0683 -0.234 0.121** -0.0298 (0.142) (0.0615) (0.0801) (0.144) (0.0569) (0.0785) lnTrade*lnRI -0.256*** -0.0967*** -0.102*** (0.0704) (0.0274) (0.0365) lnTrade*lnRKL -0.348*** -0.173*** -0.124** (0.0852) (0.0587) (0.0531) t -0.0255*** -0.0239*** 0.00465 -0.0245*** -0.0246*** 0.00296 (0.00676) (0.00464) (0.00407) (0.00796) (0.00412) (0.00411) Constant 11.29*** 8.435*** 14.66*** 12.29*** 8.731*** 14.78*** (0.809) (0.479) (0.624) (1.079) (0.570) (0.615) Observations 948 948 962 948 948 962 R-squared 0.121 0.291 0.117 Number of country 69 69 70 69 69 70

Country FE NO NO YES YES NO YES

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

21

mixed coefficients, and only with significant results in column (1) and (5). The mixed coefficients and the lack of significance is unsurprising, as all mechanisms outlined imply that trade has to be conditional upon a feature before there is a clear effect predicted. Finally, the time trend shows significant and negative results for SO2 and NOX emissions, which also follows expectations. In the time period of the analysis a lot of attention has been focused on reducing SO2 and NOX emissions, but not yet so much towards the reduction of Co2 emissions.

Next, we turn to a discussion of the results in columns (1-3). The capital-labor ratio is

only significantly positive for NOX emissions per capita. A higher capital-labor ratio results in increasing NOX emissions. This follows the factor endowment prediction, that more capital-abundancy is a sign for increased specialization in capital- and pollution-intensive production. Income per capita is also positively significant for NOX emissions, showing the scale effect where increased income has a positive effect on NOX emissions when there is no trade. However, when there is trade liberalization, Trade*RI shows that there is an offsetting effect of the positive income effect on pollution when the relative income of a country is higher. In fact, this interaction term is negatively significant for all three pollutants, following the prediction that higher relative per capita income exerts a downward pressure on industrial pollution. This is the result of increased environmental stringency that accompanies an increase in relative per capita income. Regardless that there is no direct significant income effect found on emissions of SO2 and Co2, higher trade intensity and a higher relative income indicate there is an offsetting effect on industrial pollution. With respect to NOX emissions, engagement in trade for countries with higher relative incomes results in a downward pressure on the increased pollution brought about by the scale effect.

Then, exploring the results of columns (4-6), we see that a higher capital-labor ratio

22

capital-abundant will specialize in pollution-intensive industries. Rather, this shows an indication that the capital-abundant countries can invest and import more in capital-intensive and cleaner production processes.

Table 2: Conditional effects of trade on industrial pollution

(1) (2) (3) (4) (5) (6)

VARIABLES lnSO2 lnNOX lnCo2 lnSO2 lnNOX lnCo2

lnKL 0.645 0.866*** 0.399* 0.924** 0.898*** 0.382 (0.398) (0.166) (0.208) (0.413) (0.172) (0.234) lnGDP 0.906* 1.167*** 0.517* 1.213** 1.129*** 0.512* (0.518) (0.211) (0.267) (0.495) (0.200) (0.290) lnKL*lnGDP -0.0849* -0.0902*** -0.0403* -0.108** -0.0919*** -0.0396 (0.0440) (0.0177) (0.0224) (0.0436) (0.0174) (0.0242) lnTrade -0.311* 0.102 -0.0789 0.667* -0.0525 -0.188 (0.158) (0.0673) (0.0795) (0.392) (0.157) (0.276) lnTrade*lnRI -0.153** -0.0251 -0.0589* -0.867** -0.301* -0.107 (0.0760) (0.0470) (0.0332) (0.370) (0.179) (0.257) lnTrade*lnRKL -0.166** -0.0714 -0.0258 1.297*** 0.471* 0.0843 (0.0734) (0.0978) (0.0537) (0.486) (0.250) (0.306) lnTrade*lnRKL*lnRI -0.0217 -0.0244 0.00944 (0.0477) (0.0277) (0.0277) lnEduc -2.430** 0.607 0.409 (1.131) (0.414) (0.722) lnTrade*lnEduc -1.012*** 0.0943 0.110 (0.363) (0.137) (0.247) lnTrade*lnRI*lnEduc 0.688* 0.241 0.0504 (0.352) (0.167) (0.230) lnTotal*lnRKL*lnEduc -1.412*** -0.482* -0.179 (0.509) (0.254) (0.288) t -0.0188** -0.0223*** 0.00539 -0.00240 -0.0201*** 0.00362 (0.00798) (0.00456) (0.00419) (0.0120) (0.00412) (0.00589) Constant 3.486 -0.763 10.63*** 2.253 -1.248 10.36*** (4.664) (1.915) (2.452) (4.777) (2.008) (2.805) Observations 948 948 962 879 879 893 R-squared 0.309 0.144 0.374 0.246 0.161 Number of country 69 69 70 64 64 65

Country FE YES NO YES YES YES YES

23

4.2 Income, capital-labor ratio, education and offsetting affects

Table 2 shows the results for estimations on Equation (4) columns (1-3), and for estimations on Equation (5) columns (4-6). Significant estimations of the main interactions are in black. Here, the coefficients of the capital-labor ratio and income per capita are positively significant in all estimations (except for column 6). This implies that with everything else being equal, a higher capital-labor ratio will result in an increase in pollution. This shows a prevailing factor endowments effect, where countries that are capital-abundant specialize in pollution-intensive industries. Changes in the scale of an economy, reflected by higher income per capita will also lead to an increase in pollution, when nothing else changes. When there are both increases in per capita income and in capital-labor ratio (KL*GDP), a downward pressure on the increase in pollution becomes visible. This shows that countries with a higher per capita income demand more environmental stringency, and if these countries have a higher capital-labor ratio, they can then purchase capital-abundant clean technology. The capital-capital-labor ratio becomes now subject to a prevailing technique effect, instead of a factor endowments effect. Trade intensity is only negatively significant for SO2 emissions, showing that with everything else being equal, an increase in trade results in lower SO2 emissions. For NOX and Co2 such an effect is not found, following the prediction in the previous section that there is not necessarily a direct effect of trade on industrial pollution.

Next, the interaction term Trade*RI, shows that countries that have a higher relative

income, see a decrease in pollution when they engage in more trade. Note that this interaction implies there is no change in relative capital-labor ratio. This implies there is an offsetting effect of the initial scale effect when a country engages in trade. When a country has a relatively higher per capita income, its environmental stringency is also higher. Opening up to trade can thus bring the positive effect of increasing access to measures of environmental stringency, leading to a decrease in emissions. Significance for this term is found for SO2 and Co2 emissions per capita.

24

capital-abundant, clean technologies. This interaction term thus shows that with no trade in a country, the factor endowment reasoning prevails, but when a country opens up to trade, there is a halo effect associated with international trade if a country has a higher relative capital-labor ratio.

The fourth interaction term is a three-way interaction term (Trade*RKL*RI). This does not deliver significant results, indicating that results found for Trade*RI and Trade*RKL are not influenced by changes in the third variable. Moreover, there is no offsetting or extra effect on the interaction KL*GDP when a country witnesses trade liberalization. As there is no significance found here of possible effects, this variable is dropped out of the estimations in (4-6).

In the models (4-6), education is added to the model. The direct effect of education on

25

endowments effect of trade bringing about more specialization in the pollution-intensive, capital-abundant sector. When education is also allowed to influence this effect, (Trade*RKL*Educ) shows negative and significant coefficients for SO2 and NOX emissions. This implies that rising education levels exert a pollution halo effect on a higher relative capital-labor ratio and trade. This means that there is an offsetting effect when education is high on the factor endowment trade-effect. For the halo effect to come to show, certain capabilities are necessary to efficiently implement the clean technologies and production processes that can be imported. It shows that a certain level of education needs to be reached, before the ability to import these technologies/processes also have an actual effect on a decrease in pollution.

4.3 Sectoral results

In this section, the robustness of the results found in Table 2 will be checked for different trading sectors. Results are reported in Tables 3-5, and significant results of the main interactions are in black. The sectors that are included are trade in total goods, manufacturing, chemical, crude, total services, and the financial services sectors. The direct effects of the capital-labor ratio and income per capita variables keep the same direction as found for total trade in Table 2. For all different sectors, in an economy where there is no trade, increases in either capital-abundancy or in income per capita lead to an increase in pollution. The variables showing the effect of education and/or trade is sometimes negative and sometimes positive, as well as differencing in significance, indicating that this effect should be interpreted in combination with the conditional effects added. The interaction KL*GDP remains negatively significant for all sectors, showing that increases in both variables exert an offsetting influence on the emissions following the same reasoning as outlined above.

In Table 3 columns (1-3), trade in total goods is used for estimations. The signs of the

26

higher relative income per capita. Interesting here is that there is no significant effect of increasing relative incomes and increased trade in manufacturing sector when there is no education in the country. This shows that increases in relative per capita income (increases in environmental stringency) together with trade, does not necessarily need to have a downward pressure on the scale effect of increased income.

In Table 4, columns (1-3) results for trade in the chemical sector are reported. Here, there is an effect for (Trade*RI*Educ), strengthening scale effects of increases in GDP per capita. There are no significant results found for a conditional effect of relative income per capita and education upon trade on the chemical sector. Columns (4-6) show results on estimations for the crude sector, with no significant results for the important interaction variables. There is no effect of trade in this sector conditional upon relative income per capita, relative capital-labor ratio, or education on pollution.

27

Table 3: Trade in total goods and in the manufacturing sector

(1) (2) (3) (4) (5) (6) VARIABLES lnSO2 Total Goods lnNOX Total Goods lnCo2 Total Goods lnSO2 Manufacturing lnNOX Manufacturing lnCo2 Manufacturing lnKL 1.238*** 0.821*** 0.446* 0.771* 0.949*** 0.712** (0.463) (0.193) (0.258) (0.451) (0.230) (0.283) lnGDP 1.724*** 1.028*** 0.623* 1.753*** 0.991*** 0.781** (0.568) (0.211) (0.329) (0.560) (0.204) (0.348) lnKL*lnGDP -0.148*** -0.0810*** -0.0473* -0.135*** -0.0809*** -0.0582** (0.0495) (0.0186) (0.0274) (0.0455) (0.0177) (0.0285) lnTrade 0.407 -0.0980 -0.386 0.226 -0.195 -0.0666 (0.523) (0.179) (0.303) (0.418) (0.162) (0.199) lnTrade*lnRI -0.973*** -0.414** -0.320 -0.258 -0.103 -0.0297 (0.353) (0.174) (0.246) (0.180) (0.0738) (0.106) lnTrade*lnRKL 1.285*** 0.495** 0.214 -0.0854 0.0836 0.0248 (0.479) (0.241) (0.297) (0.182) (0.0872) (0.117) lnEDUC -2.305 0.698 0.651 -2.349 1.063 0.185 (1.449) (0.437) (0.702) (1.751) (0.693) (1.039) lnTrade*lnEDUC -0.700 0.184 0.387 -0.462 0.209 -0.0184 (0.517) (0.164) (0.273) (0.405) (0.162) (0.190) lnTrade*lnRI*lnEDUC 0.900*** 0.397** 0.307 0.337** 0.0926 0.0418 (0.337) (0.170) (0.225) (0.161) (0.0701) (0.0984) lnTrade*lnRKL*lnEDUC -1.507*** -0.513** -0.271 -0.0838 -0.0173 0.0425 (0.486) (0.238) (0.275) (0.181) (0.0876) (0.100) t -0.00971 -0.0247*** -0.00162 -0.0279** -0.0213*** -0.00227 (0.0126) (0.00356) (0.00431) (0.0107) (0.00379) (0.00428) Constant -1.874 -0.618 9.260*** 1.674 -2.154 6.152* (5.418) (2.195) (3.103) (5.623) (2.570) (3.347) Observations 883 883 897 883 883 897 R-squared 0.374 0.156 0.366 0.233 0.152 Number of country 64 64 65 64 64 65

Country FE YES NO YES YES YES YES

28

Table 4: Trade in the chemical and in the crude sector

(1) (2) (3) (4) (5) (6) VARIABLES lnSO2 Chemicals lnNOX Chemicals lnCo2 Chemicals lnSO2 Crude lnNOX Crude lnCo2 Crude lnKL 0.162 0.764*** 0.382 0.624 0.827*** 0.487 (0.421) (0.205) (0.256) (0.475) (0.251) (0.311) lnGDP 1.529*** 0.975*** 0.611* 1.445** 0.883*** 0.717* (0.531) (0.243) (0.354) (0.552) (0.229) (0.369) lnKL*lnGDP -0.110*** -0.0779*** -0.0487* -0.114** -0.0724*** -0.0564* (0.0401) (0.0201) (0.0282) (0.0432) (0.0190) (0.0307) lnTrade 1.170*** -0.0539 -0.0760 0.279 -0.145 0.237 (0.427) (0.163) (0.235) (0.415) (0.146) (0.187) lnTrade*lnRI -0.222 -0.113* -0.107 -0.145 -0.0743 0.00155 (0.151) (0.0610) (0.0858) (0.133) (0.0519) (0.0718) lnTrade*lnRKL -0.0444 0.0671 0.0415 -0.144 0.0380 -0.00680 (0.147) (0.0680) (0.0945) (0.144) (0.0540) (0.0719) lnEDUC -6.543*** 0.933 0.842 -2.385 1.387** -0.436 (2.115) (0.688) (0.980) (2.094) (0.677) (1.059) lnTrade*lnEDUC -1.337*** 0.103 0.149 -0.290 0.196 -0.198 (0.446) (0.155) (0.218) (0.423) (0.154) (0.177) lnTrade*lnRI*lnEDUC 0.312** 0.103* 0.0763 0.174 0.0572 -0.0139 (0.126) (0.0547) (0.0742) (0.120) (0.0485) (0.0686) lnTrade*lnRKL*lnEDUC -0.221 -0.0783 -0.0819 0.0790 -0.0124 -0.00775 (0.156) (0.0740) (0.0847) (0.134) (0.0581) (0.0822) t -0.0130 -0.0228*** -0.000152 -0.0264** -0.0212*** 0.00128 (0.0108) (0.00468) (0.00427) (0.0116) (0.00392) (0.00502) Constant 12.41** 0.0597 10.27*** 4.518 -0.823 10.11** (5.639) (2.445) (3.195) (5.707) (2.895) (4.083) Observations 883 883 897 883 883 897 R-squared 0.389 0.342 0.151 Number of country 64 64 65 64 64 65

Country FE YES NO NO YES NO YES

29

Table 5: Trade in the total services and in financial services sector

(1) (2) (3) (4) (5) (6) VARIABLES lnSO2 Services lnNOX Services lnCo2 Services lnSO2 Financial lnNOX Financial lnCo2 Financial lnKL 0.849* 0.899*** 0.222 1.349*** 0.792*** 0.402* (0.454) (0.224) (0.269) (0.337) (0.181) (0.209) lnGDP 1.554*** 1.078*** 0.773** 2.688*** 0.953*** 0.701* (0.535) (0.212) (0.318) (0.549) (0.257) (0.418) lnKL*lnGDP -0.128*** -0.0870*** -0.0585** -0.212*** -0.0755*** -0.0494 (0.0445) (0.0181) (0.0266) (0.0442) (0.0204) (0.0315) lnTrade 0.254 -0.0701 -0.269 0.418** -0.0613 -0.101 (0.363) (0.145) (0.169) (0.175) (0.0787) (0.121) lnTrade*lnRKL -0.0888 0.106 -0.142 -0.0202 0.0270 -0.0537 (0.197) (0.107) (0.122) (0.105) (0.0377) (0.0558) lnTrade*lnRI -0.134 -0.133 0.0466 -0.0517 -0.0645* -0.0220 (0.176) (0.0840) (0.104) (0.107) (0.0355) (0.0544) lnEDUC -2.141 0.835 1.370* -4.234*** 0.783 0.925 (1.471) (0.510) (0.697) (1.474) (0.550) (0.622) lnTrade*lnEDUC -0.357 0.0797 0.333* -0.363** 0.0417 0.114 (0.363) (0.145) (0.181) (0.175) (0.0754) (0.116) lnTrade*lnRI*lnEDUC 0.171 0.122 -0.0415 0.0875 0.0570* 0.0265 (0.170) (0.0780) (0.101) (0.0935) (0.0315) (0.0441) lnTrade*lnRKL*lnEDUC -0.00833 -0.0916 -0.0192 -0.0211 -0.0161 0.0310 (0.191) (0.104) (0.125) (0.0911) (0.0331) (0.0476) t -0.0200* -0.0225*** 0.000224 -0.0212* -0.0207*** -0.00256 (0.0116) (0.00369) (0.00477) (0.0122) (0.00375) (0.00410) Constant 1.943 -1.504 11.09*** -2.484 -0.436 9.062*** (5.273) (2.466) (3.327) (4.421) (2.149) (2.613) Observations 865 865 879 713 713 726 R-squared 0.331 0.160 0.391 Number of country 64 64 65 59 59 60

Country FE YES NO YES YES NO NO

30

5. Discussion

The estimations in the previous section have delivered interesting results. First of all, different results for conditional effects have been found in Table 1 and 2. In Table 1 significance was found where trade was significantly conditional upon either changing relative income per capita, or changing relative capital-labor ratio. The decreasing pressure on pollution as a result of increasing relative per capita income and trade, shows the effect of increased environmental stringency that accompanies a higher relative per capita income. This delivers evidence for the mechanism hypothesized in H1. Changing relative capital-labor ratio also prove to be a condition for a decreasing influence of international trade on pollution here. A factor endowments effect resulting from an increase in relative capital-labor ratio was predicted in H2, but no evidence was found for this reasoning. Instead, a higher capital-labor ratio ensures that there is a decreasing pressure resulting from international trade on pollution. This provides an indication for a positive pollution halo effect brought about by engagement in international trade and with a higher relative capital-labor ratio, and therefore no evidence was found for H2.

In Table 2, both relative per capita income and relative capital-labor ratio were

31

Columns (4-6) in Table 2 showed what happens with the results after education is introduced as an extra condition. In a country where there are no changes in relative income per capita or relative capital-labor ratio, higher levels of education and trade liberalization leads to a decrease in SO2 emissions. This provides an indication that where education levels lead to more awareness about environmental problems, trade can function as a channel where more knowledge -about environmental regulation or cleaner processes- can be imported. This delivers evidence for H4a, but the evidence is not very robust as this effect is only significant for SO2 emissions.

Results also show that there are opposing effects at play in a simultaneous manner

with trade liberalization. An increasing relative per capita income in a country engaging in international trade showed a decreasing pressure on pollution, implying at least a partial compensation of the scale effect and following the reasoning that more environmental stringency can lead to a decreasing effect on pollution when engaging in trade. However, when education levels also increase, this effect is offset as follows from the positive results for Total*RI*Educ. This implies that education also leads to an increase in productivity of the workforce (Zugravu-Soilita, 2017). There is no evidence found for H4b, but results do show that an increase in workforce productivity dominates an environmental stringency effect on pollution. There is a negative effect on pollution of increased environmental stringency and trade, but there is also a positive effect again with higher incomes and increased workforce performance. Neglecting the possibility of different conditions influencing trade can thus lead to misleading effects or it covers up some other important underlying mechanisms.

These opposing effects are also present when focusing at the relative capital-labor

32

and confirming H4c. However, the mechanisms found turn out not to be robust throughout the three different pollutants. Most significance was found with respect to SO2 emissions. An explanation can be found in the history of environmental action adopted towards three different pollutants. SO2 was one of the first pollutants that was recognized as being harmful, and that action to reduce SO2 emissions had to be undertaken. In the process of finding regulations to reduce emissions of SO2, education about other forms of pollution also increased indirectly. Therefore, an awareness effect brought about by education on industrial pollution, may have stopped exerting a significant influence after the initial ‘shock’. On the other hand, it can also mean that education takes longer before it will start to exert an influence, that creating a rising sense of awareness takes longer. Regarding SO2 emissions, many countries saw a sharp reduction in emissions starting already in the 1980s (Smith et al., 2004). This was the result of policies that were imposed as early as the 1980s or the beginning of the 1990s. Designing and implementation of regulations takes a while before an effect will become visible. The sample of this analysis started in 1995, and thus effects of these regulations are visible in the sample. NOX emissions became subject of environmental regulation efforts a little bit later, and these regulations came largely into effect in the mid/end of the 1990s (Burtraw and Szambeland, 2009). This can explain why changing significance is found for NOx emissions, as in some areas effective regulations were already in place, but in other areas not. Also, some countries might have been more extensive in their actions in decreasing the emissions than others. In any case, these pollutants have a far longer history in environmental pollution regulations than the final pollutant, Co2. Co2 emissions have been increasing steadily, and have only become subject to environmental regulations relatively recently (Uchiyama, 2016). Co2 is included in the Kyoto Protocol, but this came into effect in 2005, which is almost the end of the time-period used here for the analyses. The lack of insignificance in time effects for Co2 emissions also show that no extensive successful action has been taken to reduce Co2 emissions in the time-period of this analysis. This provides an explanation for the lack of significance for results in estimations on Co2 emissions.

Also interesting is that even though the signs of the estimated coefficients remain

33

34

6. Conclusion

This research has provided new insights on the literature linking international trade to industrial pollution on the basis of allowing conditional country characteristics (relative per capita income, relative capital-labor ratio, and education level) conditioning the influence of trade on industrial pollution on a large sample of 70 countries for three pollutants (SO2, NOX, Co2). The theoretical assumptions made have been empirically validated to a great extent, but interesting additions to these theories have also been found. Evidence was found that an increasing relative per capita income can result in an offsetting pressure of the trade-induced scale effect on pollution. However, when a country also has an increasing education level, an increased productivity level will dominate the offsetting environmental stringency effect. This highlights the importance of allowing different effects to exert an influence on trade and pollution. Just an inclusion of the relative per capita income would have suggested a positive environmental stringency effect, while higher education levels can offset this positive effect. This results in an increase in pollution when a country engages in international trade and has higher education –and thus higher productivity-, even if a country also has stronger environmental stringency. Also, an awareness effect about environmental problems and increased absorptive capacity of implementing new environmental regulations brought about by rising education levels do not appear to be dominant when trade liberalization allows a country to expand its scale.

No evidence was found for the factor endowments prediction that a higher

capital-labor ratio results in a specialization in capital-intensive, pollution-intensive industries after a country engages in international trade. Rather, a trade-induced technique effect seems to prevail here, until education is allowed in the model. Education proves to meaningfully influence the adoption of cleaner and capital-abundant technologies, and improves the efficient implementation of these technologies. In the case that education remains equal in a country, while there is an increase in relative capital-labor ratio, then there can be some confirmation of a factor endowments effect to be found. Also, just relying on education to raise awareness and additional absorptive capacities does not seem to be an effective approach when trying to achieve a decrease emissions. It is shown in this research that it is very important in not just trying to validate one of the main hypotheses in the literature on this topic, but rather combining certain influences. Otherwise, misleading results can be presented and important mechanisms can be missed.

35

actually beneficial for industrial pollution emissions. Rather, the influence of international trade on industrial pollution is subject to certain conditional effects, that can work in opposite directions. This implies that regulations to steer international trade in the direction that it can exert a downward pressure on industrial pollution, different grounds should be covered. Focusing alone on increasing education will not be very effective, while only focusing on increased environmental stringency or increasing capital-abundancy also will not have the desired effect. From the results followed that increased environmental stringency cannot offset the increases in the trade-induced scale effect that follows a stronger productivity. On the other hand, this research did deliver evidence that trade can exert an influence on pollution emission when it is conditional upon country characteristics. This means that it does make sense to incorporate the effect of trade in certain scenarios in regulations aiming to decrease industrial pollution. It also indicates that there is a long way to go to effectively impose stringent measures to steer international trade in a way that is in any case not harmful for industrial pollution. The challenge is to design a comprehensive policy that can cover the main mechanisms.

This analysis is also subject to some limitations, with the most important one being the

somewhat limited time-period of 1995-2008. As shown by the changing significance among pollutants, a broader time-period needs to be adopted to see if indeed a slower pace of environmental regulation is the reason for less significant results. It can be the case that the directions that have proven to be significant for SO2 (and to a lesser extent NOx) do not function in this way for Co2 emissions. However, it can also be that the mechanisms do turn out the same way for all three pollutants and then lessons can be learned. That is why more insights should be acquired in this process. Moreover, another limitation is the country sample where the low-income countries have not been included. According to the pollution haven literature, these countries see the strongest harmful effect of engaging in trade. As this sample did not include this group of countries, we have not been able to check for this effect. Also, environmental stringency was indicated by relative income per capita in this research. It is also important that a correct measure of environmental stringency will be developed to see if results found remain the same. Finally, only a limited selection of trading sectors was included. Limited significance for this selection indicates that the mechanisms related to international trade are not necessarily sector specific mechanisms, but rather as trade for country as a whole. However, as not all trading sectors have been analyzed, this conclusion may be biased.

36

37

Appendix 1

Industrialized countries: Aruba, Australia, Austria, Bahrain, Belgium, Canada, China, Czech

Republic, Denmark, Estonia, Finland, France, Germany, Hungary, Iceland, Ireland, Israel, Italy, Japan, Korea Republic of, Kuwait, Lithuania, Luxembourg, Malaysia, Malta, Netherlands, New Zealand, Norway, Portugal, Qatar, Russian Federation, Singapore, Slovakia, Slovenia, Spain, Sweden, Switzerland, United Arab Emirates, United Kingdom, United States

Emerging Industrial Economies: Argentina, Belarus, Brazil, Brunei Darussalam, Bulgaria,

Chile, Colombia, Costa Rica, Croatia, Cyprus, Greece, India, Indonesia, Kazakhstan, Latvia, Mauritius, Mexico, Oman, Poland, Romania, Saudi Arabia, South Africa, Suriname, Thailand, TFYR of Macedonia, Tunisia, Turkey, Ukraine, Uruguay, Venezuela.

Summary statistics

(1) (2) (3) (4) (5)

VARIABLES N mean sd min max

SO2 in grams per capita 966 36,041 41,125 2,014 363,139

Nox in grams per capita 966 31,753 22,770 4,731 263,016

Co2 in grams per capita 980 9.432e+06 8.731e+06 844,607 7.098e+07

Income per capita 980 17,492 16,609 381.5 112,852

Relative income 980 1.678 1.476 0.0492 6.792

Capital-labor ratio 980 212,393 129,225 12,794 790,553

Relative capital-labor ratio 980 1.729 1.054 0.114 6.927

Education 910 2.824 0.485 1.601 3.672 Trade intensity 962 0.943 0.648 0.143 4.961 Goods intensity 967 0.740 0.505 0.123 4.107 Manufacturing intensity 967 0.115 0.0728 0.0157 0.544 Chemicals intensity 967 0.0676 0.0566 0.00993 0.464 Crude intensity 967 0.0299 0.0349 0.00203 0.420 Services intensity 949 0.204 0.203 0.0204 2.035

Finances intensity 794 0.0149 0.0953 4.51e-06 1.285

38

Appendix 2

Data definition and sources

Variable Definition Source

SO2 NOx Co2 Population Income per capita GDP K/L RKL I RI Educ Trade intensity

SO2 emissions (tons, metric) NOx emissions (tons, metric)

Co2 emissions (tons, metric)

Used to convert pollution data into the form of per capita emissions

GDP per capita, current USD

Used to create trade intensity

Physical capital stock per worker (in 100000 USD), adjusted by capital depreciation rate Relative KL are expressed relative to the world average. World averages are calculated as the average of all countries for whom data is reported in the PWT (in 100000 USD) Per capita income, logged

Relative income is expressed relative to the world average. World averages are calculated as the average of all countries for whom data is reported in the World Bank, logged. Education level

Ratio of imports plus exports to GDP, everything logged EDGAR v.4.2 UNEP EDGAR v.4.2 UNEP World Bank World Bank World Bank World Bank

Penn World Tables 9.0

Penn World Tables 9.0

World Bank World Bank

39

References

Abdouli M, Hammami S. 2015. The Impact of FDI Inflows and Environmental Quality on Economic Growth: an Empirical Sttudy for the MENA Countries. Journal of the Knowledge Economy. [accessed 10 June 2017];8(1):254-278. http://download.springer.com.proxy- ub.rug.nl/static/pdf/433/art%253A10.1007%252Fs13132-015-0323- y.pdf?originUrl=http%3A%2F%2Flink.springer.com%2Farticle%2F10.1007%2Fs13132-015- 0323-y&token2=exp=1490096593~acl=%2Fstatic%2Fpdf%2F433%2Fart%25253A10.1007%2525 2Fs13132-015-0323-y.pdf%3ForiginUrl%3Dhttp%253A%252F%252Flink.springer.com%252Farticle%252F10.10 07%252Fs13132-015-0323-y*~hmac=6d5ccf4bf10bbc18863df0e5ad223d3455749a607c17d02e15b48dc9174ab17f. Al-Mulali U, Ozturk I. 2015. The Effect of Energy Consumption, Urbanization, Trade Openness, Industrial Output, and the Political Stability on the Environmental Degradation in the MENA (Middle East and North African) Region. Energy; [accessed 19 June 2017];84:382-409. http://ac.els-cdn.com.proxy-ub.rug.nl/S036054421500287X/1-s2.0-

S036054421500287X-main.pdf?_tid=01abf50e-5511-11e7-a257-00000aacb35d&acdnat=1497891899_33b7791acda12fd9303537e4e3633c81.

Antweiler W, Copeland BR, Taylor MS. 2001. Is Free Trade Good for the Environment? The American Economic Review; [accessed 17 June 2017];91(4):877-908.

http://www.jstor.org.proxy-ub.rug.nl/stable/pdf/2677817.pdf.

Berlemann M, Wesselhöft J. 2014. Estimating Aggregate Capital Stocks Using the Perpetual Inventory Method: A Survey of Previous Implementations and New Empirical Evidence for 103 Countries. Review of Economics; [accessed 19 June 2017];65:1-34. http://www.review-of-economics.com/download/Berlemann_Wesselhoeft_2014.pdf.

Brunel C, Levinson A. 2013. Measuring Environmental Regulatory Stringency. OECD Trade and Environment Working Papers; [accessed 17 June 2017]. http://www.oecd-

40

Burtraw D, Szambelan SJ. 2009. U.S. Emissions Trading Markets for SO2 and NOx.

Discussion Paper Resources for the Future; [accessed 17 June 2017].

http://www.rff.org/files/sharepoint/WorkImages/Download/RFF-DP-09-40.pdf.

Candau F, Dienesch E. 2017. Pollution Haven and Corruption Paradise. Journal of Environmental Economics and Management; [accessed 19 June 2017]:85:171-192. http://ac.els-cdn.com.proxy-ub.rug.nl/S0095069617303157/1-s2.0-S0095069617303157-

main.pdf?_tid=dda5e4ce-503f-11e7-8b47-00000aacb360&acdnat=1497362269_9e8ce7ee9aad87b7f2985b8d26fdab12.

Cole MA, Elliott RJR, Fredriksson PG. 2006. Endogenous Pollution Havens: Does FDI Influence Environmental Regulations? Scandinavian Journal of Economics; [accessed 19 June

2017];108(1):157-178.

http://www.jstor.org.proxy-ub.rug.nl/stable/pdf/3877048.pdf?refreqid=excelsior%3Ad53609bf32ad4b6b5dae4cfd04c992 d6.

Cole MA, Elliott RJR. 2003. Determining the trade-environment composition effect: the role of capital, labor, and environmental regulations. Journal of Environmental Economics and Management; [accessed 17 June 2017];46(3):363-383. http://ac.els-cdn.com.proxy- ub.rug.nl/S0095069603000214/1-s2.0-S0095069603000214-main.pdf?_tid=97f984a4-5458-11e7-9fe9-00000aab0f6c&acdnat=1497812694_4ca87057dedee1f837c303dda45b7395. Cole MA, Elliott RJR. 2003. Do Environmental Regulations Influence Trade Patterns? Testing Old and New Trade Theories. The World Economy; [accessed 19 June

2017];26(8):1163-1186.

http://onlinelibrary.wiley.com.proxy-ub.rug.nl/doi/10.1111/1467-9701.00567/epdf.

Cole MA, Elliott RJR. 2005. FDI and the Capital Intensity of ‘Dirty’ Secors: A Missing Piece of the Pollution Haven Puzzle. Review of Development Economics; [accessed 17 June

2017];9(4):530-548.

http://onlinelibrary.wiley.com.proxy-ub.rug.nl/doi/10.1111/j.1467-9361.2005.00292.x/epdf.

Dean JM. 2002. Does Trade Liberalization Harm the Environment? A New Test. The Canadian Journal of Economics; [accessed 19 June 2017]:35(4):819-842.