Endogenous Pollution Policy and Trade Liberalization: Impact on Environmental Quality, Papers of Microeconomics

Download Endogenous Pollution Policy and Trade Liberalization: Impact on Environmental Quality and more Papers Microeconomics in PDF only on Docsity! 7 Journal of Economic Literature Vol. XLII (March 2004) pp. 7–71 Trade, Growth, and the Environment1 BRIAN R. COPELAND and M. SCOTT TAYLOR 1 Copeland: University of British Columbia. Taylor: University of Wisconsin. We are grateful for helpful com- ments from John McMillan, Gene Grossman, and the ref- erees. Copeland acknowledges funding from the Social Sciences and Humanities Research Council of Canada. 1. Introduction For the last ten years environmentalistsand the trade policy community have engaged in a heated debate over the envi- ronmental consequences of liberalized trade. The debate was originally fueled by negotiations over the North American Free Trade Agreement and the Uruguay Round of GATT negotiations, both of which occurred at a time when concerns over glob- al warming, species extinction, and industri- al pollution were rising. Recently it has been intensified by the creation of the World Trade Organization (WTO) and proposals for future rounds of trade negotiations. The debate has often been unproductive because the parties differ greatly in their trust of market forces and typically value the environment differently. It has been ham- pered by the lack of a common language and also suffered from little recourse to eco- nomic theory and empirical evidence. This is perhaps not surprising, because much of the work in this area is still quite new. The purpose of this essay is to set out what we currently know about the environmental consequences of economic growth and inter- national trade. We critically review both theory and empirical work to answer three basic questions. What do we know about the relationship between international trade, economic growth, and the environment? How can this evidence help us evaluate ongoing policy debates in this area? Where do we go from here? To answer these questions, we discuss both the empirical and theoretical literature with the aid of a relatively simple general equilibrium model where government policy and private sector behavior interact to deter- mine the equilibrium level of pollution. The model is developed in section 2 of the paper and then employed in various guises throughout. Our use of a model to organize our review reflects an overarching theme of our essay: economic theory needs to play a much larger role in guiding empirical inves- tigation, suggesting alternative hypotheses, and disciplining inferences. The vast major- ity of empirical work in this field has little connection to explicit theory, and we argue that this has left important policy debates, which hinge on the relative importance of various theoretical magnitudes, badly informed. The economic literature on these issues came in two waves, with an initial surge of activity in the 1970s and a resurgence of interest stimulated by the policy debates 8 Journal of Economic Literature, Vol. XLI (March 2004) 2 See, for example, William Baumol (1971), Ingo Walter (1973), James Markusen (1975), Rudiger Pethig (1976), and Horst Siebert (1977). 3 Exceptions include Walter’s (1973) empirical paper and Pethig (1976) on the pattern of trade. 4 Other recent surveys that focus more on policy include Michael Rauscher (2001) and Alistair Ulph (1997). Hakan Nordstrom and Scott Vaughan (1999) provide a good comprehensive review of the trade and environment literature. of the past decade.2 Much of the earlier lit- erature was normative, with a focus on issues such as gains from trade and optimal trade or environmental policies.3 A large compo- nent of recent work also focuses on policy analysis, but its most significant feature is its concern with positive issues: generating and attempting to test hypotheses about how trade or growth affects environmental out- comes. We view these latter issues as funda- mental to resolving current policy debates, and so most of our essay will focus on this aspect of the literature.4 We begin our analysis in section 2 by examining the link between incomes per capita and environmental quality. Interest in this link arose from the pioneering work by Gene Grossman and Alan Krueger (1993) on NAFTA, which subsequently led to a burgeoning literature on what has come to be known as the environmental Kuznets curve (EKC). The environmental Kuznets curve hypothesizes an inverse-U-shaped relationship between a country’s per-capita income and its level of environmental qual- ity: increased incomes are associated with an increase in pollution in poor countries, but a decline in pollution in rich countries. This literature is important because many in the trade-policy community have argued that trade and growth may actually be good for the environment. If environmental qual- ity is a normal good, increases in income brought about by trade or growth will both increase the demand for environmental quality and increase the ability of govern- ments to afford costly investments in envi- ronmental protection. Our review of both the theoretical and empirical work on the EKC leads us to be skeptical about the existence of a simple and predictable relationship between pollution and per-capita income. To investigate the EKC we employ a simple pollution demand- and-supply system linking pollution levels to national characteristics (incomes, factor endowments, and technologies) and trading opportunities (comparative advantage and current trade restrictions). Much of the work on the EKC reduces this set of possible explanatory factors to essentially just one— incomes per capita—although it is not clear why we would want to impose these restric- tions on empirical estimation ex ante. This concern receives some support from empiri- cal work that finds that the shape of the esti- mated relationship between pollution and income is sometimes sensitive to functional form, the sample of countries or cities used, and the time period chosen. Despite these concerns, the EKC litera- ture has made two lasting and significant contributions. First, it raised important empirical questions about how trade and growth affect the environment, and launched a significant research agenda. Second, it has provided quite convincing evidence that there is an income effect that raises environmental quality. Moreover, there are strong indications that this income effect works because increases in the stringency of environmental regulation accompany higher per-capita incomes. Therefore an analysis of the effects of trade and growth on the envi- ronment cannot proceed without taking into account endogenous policy responses. Unfortunately, because most of the litera- ture has relied on simple reduced-form esti- mation, there is very little work that isolates this income effect from other factors, such as the scale effects of growth and capital abun- dance. Consequently we still know very little about when this income effect is strong enough to offset forces, such as capital accu- mulation or export-led expansion of pollut- ing industries, that are associated with Copeland and Taylor: Trade, Growth and the Environment 11 8 The model is based on Brian Copeland and M. Scott Taylor (2003), which has its roots in Copeland and Taylor (1994). Martin McGuire (1982) and Rauscher (1997) use similar models. 9 Models with consumption-generated pollution have been somewhat neglected in the trade literature. For one example, see parts of Copeland and Taylor (1995b) and Rauscher (1997). 10 Production externalities are discussed in Copeland and Taylor (1999) and Michael Bennaroch and Henry Thille (2001). time or across countries. We need at least two primary factors of production to provide a simple motive for international trade that is independent of pollution regulation. And finally, we need endogenous pollution policy to examine how pollution may vary across countries with different levels of per-capita income. In what follows, we opt for simplic- ity rather than generality and do not attempt to be exhaustive in our coverage.8 We adopt a static model and focus on pro- duction-generated pollution.9 Pollution from a given firm harms consumers but does not affect the productivity of other produc- ers.10 There are two goods, X and Y, each produced with a constant returns to scale technology using two primary factors, capital (K) and labor (L). Denote the price of X by p, and treat Y as the numeraire. To capture differences in pollution intensity across sec- tors, we assume X generates pollution during production, but Y does not pollute at all. The production function for good Y is simply: Y = H(Ky,Ly). (1) where H is increasing, concave, and linearly homogeneous. Production of good X generates pollution emissions (Z). If firms do not undertake abatement, we assume each unit of output generates one unit of pollution, and that out- put of X is given by F(Kx,Lx), where F is increasing, concave, and linearly homoge- neous. We can think of F as “potential out- put.” If abatement does occur, then for z £ F, output of X is given by: x = za[F(Kx,Lx)] 1-a, (2) 11 One can alternatively start with a joint production technology and then (under some regularity conditions) invert it to obtain a production function that treats pollu- tion as an input. See Siebert et al. (1980) and Copeland and Taylor (2003). 12 Our functional form implicitly assumes that abate- ment has the same factor intensity as production or poten- tial output. This assumption is made for simplicity. 13 If the pollution tax is sufficiently low, firms will not abate at all, and a corner solution will result. At this point z = x, and so e = 1. Referring to (3), this no-abatement solution occurs if t £ ap. where 0 < a < 1. Although pollution is a joint output, one can treat pollution (or environ- mental services) as an input for analytical convenience.11 A firm can reduce pollution and maintain output constant by using more primary factors and adopting less-polluting techniques.12 If governments regulate pollution, we assume that firms face a price t for each unit of emissions that they release. This price may be implemented with either an emis- sions tax t or by a tradable emissions permit system, in which the government sets the total level of pollution Z, and the emissions price t is determined in the market. Firms choose the emissions intensity that minimizes their production costs. Let e = z/x denote emissions per unit of output. The Cobb-Douglas form of the production func- tion implies that the share of emission charges in the value of output is a = tz/px, and hence at an interior solution, we have:13 3) The emission intensity falls as pollution taxes rise; and it rises when the price of the polluting good p rises because the opportu- nity cost of resources used in abatement is higher. To close the production side of the model, we require nonpositive profits in each indus- try, and full employment. These conditions can be solved to obtain outputs as functions of endowments, prices and policy: (4) x x p K L y y p K L = = ( ) ( ) , , , , , , t t e z x pº = £a t 1 12 Journal of Economic Literature, Vol. XLI (March 2004) 14 It is concave in (K, L, and z) and convex in prices. Moreover, outputs and factor prices can be recovered with simple differentiation: x = ¶G/¶p; r =¶G/¶K; w = ¶G/¶L. See Alan Woodland (1982) and Avinash Dixit and Victor Norman (1980) for the standard treatment of national income functions, and Copeland (1994) for an application to environmental problems. For a given pollution tax, it can be verified that this model behaves much like the pro- duction side of the standard Heckscher- Ohlin model of international trade. In particular, an increase in the supply of capi- tal will increase the output of the capital- intensive industry X, and reduce the output of Y. An increase in the supply of labor stim- ulates Y and contracts X. We can summarize the production side of the model with a national income function. Because markets are competitive, the pri- vate sector maximizes the value of national income for any given pollution level z. This allows us to write national income G as the solution to an optimization problem: (5) where T is the feasible technology set. As is well known, the national income function satisfies a number of useful properties.14 Most useful to us is the following: (6) The equilibrium price of a pollution permit is equal to the effect on national income of an increase in allowable pollution; that is, if more pollution is allowed, national income rises by the value of the marginal product of emissions. If we instead think of the effects of a reduction in emissions z, then the cost to the economy is also ¶G/¶z. This is the gener- al equilibrium marginal abatement cost. Hence another interpretation of (6) is that the price of a unit of emissions is equal to the marginal abatement cost, which is a standard result in environmental economics. We assume there are N identical con- sumers in the economy. Each consumer t = ¶ ¶ G z . G p K L z px y x y T K L z( ) max { : ( ) ( )} {x,y} , , , , , ,= + Î 15 Homotheticity allows us to write the indirect utility function as an increasing function of real income. It also ensures that the relative demand for goods is unaffected by income levels. This is a standard assumption in the inter- national trade literature and it allows us to focus on the role of environmental policy and factor supplies in explain- ing trade patterns. maximizes utility, treating pollution as given. For simplicity, we assume preferences over consumption goods are homothetic and the utility function is strongly separable with respect to consumption goods and environ- mental quality.15 The indirect utility func- tion for a typical consumer is (7) where h is increasing and convex, I is per capita income (so I = G/N), b is a price index, and v is increasing and concave. Pollution is harmful to consumers and is treated as a pure public bad (all consumers experience the same level of pollution). The Demand for Pollution. In our approach, we treat pollution as an endoge- nously supplied factor of production. This suggests a natural way to think about the determinants of pollution is in terms of its demand and supply. Notice (6) can be inter- preted as the inverse demand for pollution. We illustrate this demand curve in figure 1. It slopes down because G is concave in z. More intuitively, we can exploit the structure we imposed on technology to write pollution demand as (8) This is the same relation we would obtain by inverting (6) and using our assumptions on technology. From (8), we can see that the demand for pollution slopes down for two reasons: as t falls, firms pollute more both because the emissions intensity e rises, and because the lower tax on pollution makes pro- duction of the dirty good more attractive (so that output of X expands while Y contracts). The Supply of Pollution. Pollution supply depends on the policy regime. If there is no z e p x p K L= ( ) ( , , , )/ t t V p I z v I p h z( , , ) ( ( )) ( )= -/ b Copeland and Taylor: Trade, Growth and the Environment 13 t Figure 1. Pollution Demand and Supply Z ° ° t Gz N•MD Z regulation, then pollution supply is perfectly elastic at t = 0. Pollution in this case is entirely demand driven. If there is an exoge- nous pollution tax to, then supply is a hori- zontal line. Shifts up or down in pollution demand raise or lower emissions. Alternatively, if there is a fixed overall pollu- tion quota in place (as in a tradable emission permit system), then the pollution supply curve is vertical. Shifts in pollution demand raise or lower the price of emissions, but have no effect on overall pollution. In general, we expect pollution policy to be endogenous; and in particular, we expect that changes in per-capita income will lead to an increase in the demand for environ- mental quality, and (if governments are responsive) a tightening up of pollution reg- ulations. The endogeneity of pollution policy plays a key role in both the theory and empirical literature. There are two approaches to modeling the policy process. One is to simply assume a benevolent government chooses policy. Another is to adopt a political economy framework where the interaction of compet- 16 The use of political economy models in the trade and environment literature is still in the early stages. Examples include Arye Hillman and Heinrich Ursprung (1994), Per Fredriksson (1997, 1999), Sumeet Gulati (2001), Carol McAusland (2003), Raucher (1997, ch. 5), and Joachim Schleich (1999). 17 See, however, our discussion of political economy elements in section 3 on policy implications. ing interest groups determines policy.16 We follow the bulk of the literature on endoge- nous policy and adopt a representative agent framework in which the government pro- vides efficient policy.17 To determine the optimal pollution policy, the government chooses the pollution level to maximize the utility of a representative consumer subject to production possibilities and private sector behavior. The govern- ment’s problem is: (9) Because we assume the economy is small in world markets, the government treats the goods price p as given. Hence dp/dz = 0 and the first order condition from (9) becomes: Max V I p z s t I G p K L z N z { ( ( ) ) ( ) }/ , . . , , , /b = 16 Journal of Economic Literature, Vol. XLI (March 2004) 20 In an interesting paper, Erwin Bulte and Daan van Soest (2001) make a related point in the context of renew- able resource exploitation. Using a model of optimal resource use, they consider the relation between the steady state resource stock and income, and point out that the shape of the relation depends on whether income rises via an increase in the resource price or an exogenous increase in non-resource-based income. 21 Some empirical support for this view can be found in William Harbaugh, Arik Levinson, and David Wilson (2002) and other papers, as we discuss later. where exL < 0 is the elasticity of X output with respect to the endowment of human capital and sw > 0 is the share of human cap- ital in national income. Note exL < 0, follows from the Rybczinski theorem of internation- al trade: human capital accumulation stimu- lates the clean industry Y, which draws resources out of the dirty industry X and lowers pollution. Hence when growth occurs via accumulation of the factor used inten- sively in the clean industry, there is a nega- tive monotonic relationship between pollution and income. This simple example highlights how dif- ferent sources of growth will in general trace out different income and pollution paths.20 Theory suggests there may well be a stable relation between pollution and various prim- itives such as technology and primary factors of production, and between income and these same variables. But unless all countries grow in exactly the same way, there is little reason to expect that there will be a simple relationship between pollution and income.21 Accordingly, all theories that gen- erate an inverse-U-shaped environmental Kuznets curve must proceed by imposing more structure than even our simple pollu- tion demand and supply model contains. There are four main explanations for the EKC. Each explanation places restrictions on preferences and technology to make equilibrium pollution a function of per capi- ta income, and to generate the desired shape of the income-pollution relation. We classify these explanations by the key mech- anism driving their results. These are: (1) sources of growth; (2) income effects; (3) 22 Grossman and Krueger (1995) and others cite Moshe Syrquin’s (1989) discussion of the structural transforma- tion in an economy during the development process (for example, for agriculture to manufacturing to services) as motivation for the view that predictable changes in the sources of growth during the development process could affect the pollution-income relationship. 23 This approach has been relatively neglected in the empirical literature until recently. Lewis Gale and Jose Mendez (1998) include relative capital abundance in their estimation of an EKC for sulfur dioxide, and Theodore Panayotou, Alix Peterson, and Jeffrey Sachs (2000) use a measure of the capital stock in their study of CO2 emis- sions. Hermamala Hettige, Muthukumara Mani, and David Wheeler (2000) decompose industrial water pollu- tion into scale, composition and pollution intensity effects, and find that the manufacturing’s share of output follows a hump-shaped pattern, but that it is not strong enough to yield an EKC pattern for water pollution. threshold effects; and (4) increasing returns to abatement. Although all of these explana- tions describe forces that could interact, we will isolate the key features of each in our presentation. Sources of Growth. To obtain the rising and then falling portions of an EKC, even in the absence of any environmental policy, we could place restrictions on the growth process across all countries. This yields one of the commonly mentioned explanations for the EKC, although it seems to lack a for- mal development in the theoretical litera- ture.22 Very simply, suppose policy is not very responsive to income (i.e. a restriction on pollution supply), but countries grow primarily via capital accumulation in the early stages of development and by human capital acquisition in later stages (i.e. restrictions on the time profile of demand shifters). Then pollution will rise and then fall with growth in per-capita income, as composition effects driven by the factor growth drive the profile for pollution. Composition effects are key here, because we have assumed a zero policy response eliminating technique effects, and the com- position effects of factor accumulation dominate scale effects in this model. Given these assumptions, changes in the sources of growth are reflected in the pattern of pollution.23 Copeland and Taylor: Trade, Growth and the Environment 17 24 Ramon Lopez (1994) provides an early treatment of this approach, as he shows how non-homotheticity in pref- erences between consumption and environmental quality can lead to an EKC. Kishore Gawande, Robert Berrens, and Alok Bohara (2001) provide an interesting variation on the income effect approach—in their model, agents are freely mobile and so income effects induce a sorting equi- librium in which higher income agents avoid polluted areas. The source of growth explanation for the EKC is important to our discussion for two reasons. First, it demonstrates how the pol- lution consequences of growth depend on the source of growth. Therefore, the analogy drawn by some in the environmental com- munity between the damaging effects of economic development and those of liberal- ized trade is, at best, incomplete. Second, the sources of growth explanation demon- strates that a strong policy response to income gains is not necessary for pollution to fall with growth. Hence the shape of the EKC need not be driven by income gains making pollution policy more stringent. Income Effects. An alternative widely cited explanation for the EKC is that its shape reflects changes in the demand for environmental quality as income rises.24 To illustrate this theory, suppose governments set policy efficiently, and consider the effects of neutral technical progress. Let l be a shift parameter representing technology, and again normalize the population so that N=1. With neutral technical change, we can write our GNP function as lG(p,K,L,Z). Pollution is determined by: (21) and differentiating with respect to l and rearranging yields: (22) where D > 0, and eMD,R is the income elas- ticity of marginal damage. Neutral techno- logical progress shifts both the demand and supply of pollution. Demand shifts because dz d MD R l = -1 e , D l lG p K L z MD p G p K L z p zz ( ) ( ( ) ( ) ), , , , , , , / ,= b 25 Similar results are obtained by considering the effects of neutral factor accumulation on pollution. Lopez (1994) shows how the effects of factor accumulation on the environment depend on interaction between the elasticity of substitution between pollution and nonpollution inputs and the income elasticity of marginal damage. Copeland and Taylor (2003, ch.3) explore these issues in greater detail. 26 While the income effect theory is illustrated here in terms of a benevolent government in a representative agent economy, it can be modified to allow for political economy motives. For example, Lopez and Siddhartha Mitra (2001) show how the presence of corruption will move the turning point of the EKC to the right. Some empirical studies, such as Scott Barrett and Kathryn Graddy (2000) include measures of political freedom as an extra shift variable in their EKC regressions; and find that all else equal, increased freedom is associated with a clean- er environment. However, the theory outlined above would imply that the political freedom variables should be interacted with income variables since one would expect political freedom to influence the strength of the policy- induced technique effect. the marginal product of pollution rises; sup- ply shifts because real income has grown. Whether pollution rises or falls with real income changes depends on the income elasticity of marginal damage.25 If the elas- ticity is less than one, then the supply shift is swamped by the demand shift and pollution rises; if it is greater than one, the opposite occurs.26 Because the EKC has both an increasing and decreasing segment, this pure income- driven explanation requires a variable income elasticity of marginal damage to gen- erate the required shape. As an example, suppose indirect utility is given by: (23) where x > 0 (and R is real income). The key characteristic of (23) is that the income elas- ticity of marginal damage is simply R/x. Using (22), we obtain an inverse-U relation between real income and pollution: pollu- tion rises with neutral growth if R > x and falls with neutral growth if R < x. Environmental quality is a normal good throughout, but at low incomes, pollution rises with growth because increased con- sumption is valued highly relative to envi- ronmental quality. As income rises, the willingness to pay for environmental quality V p I z c c R h z( ) exp( ) ( ), , /= - - -1 2 x 18 Journal of Economic Literature, Vol. XLI (March 2004) 27 See Shafik (1994) and Douglas Holz-Eakin and Selden (1995). However, Richard Schmalensee, Thomas Stoker, and Ruth Judson (1998) do find a within-sample peak for carbon emissions per capita. rises and increasingly large sacrifices in con- sumption are made to provide greater envi- ronmental benefits. The income-effect explanation of the EKC follows from two assumptions: neutral growth and a particular assumption on pref- erences. Neutral growth restricts the magni- tude of shifts in pollution demand as growth proceeds; while the rising income elasticity of marginal damage ensures ever-stronger technique effects. Composition effects play little or no role. This explanation suggests that the rela- tionship between pollution and income should vary across pollutants according to their perceived damage. For example, we might expect a very low x for directly life- threatening pollutants such as contaminated drinking water. In this case, the EKC would be (almost) monotonically declining throughout, as was found by Shafik (1994) and Grossman and Krueger (1995). Alternatively, x might be very high for pollu- tants whose harm is uncertain or delayed. Carbon emissions may fit this category, and indeed most studies have found that carbon emissions per capita tend to increase monot- onically with per-capita income.27 Threshold Effects. An alternative explana- tion for the EKC is based on threshold effects. Threshold effects can arise in either the political process as in Larry Jones and Rudolfo Manuelli (1995), or in abatement opportunities as in Andrew John and Rowena Pecchino (1994), Selden and Song (1995), and Nancy Stokey (1998). Threshold effects lead to a very different relationship between income and pollution in early ver- sus later stages of development. At low lev- els of economic activity, pollution may be unregulated entirely or regulation may have little impact on the profitability of abate- ment. Pollution therefore at first rises with growth. But after some threshold has been breached, and policy is either implemented or starts to bind, these models predict pollu- tion declines with income—provided appro- priate assumptions are imposed on tastes and technology. There are at least two possible ways to ensure regulation is ineffective in checking pollution at low levels of income. The first is to assume an abatement production function where the marginal product of abatement is bounded. In this case, there will exist a set of relatively low pollution taxes for which firms choose the zero abatement option; conse- quently, even though taxes may rise with growth over some range, this has no affect on abatement, and pollution rises with eco- nomic activity. An abatement function of this type was used in Copeland and Taylor (1994) and is implicit in Stokey (1998). The model we presented in section 1 contains this attribute because we note from (3) that if t < ap, no abatement occurs. This means that there is no technique effect to offset the scale effect of growth when incomes are low. In models with only one good, this ensures that pollution must rise with growth at low levels of income. In multi-good models, the pollution tax can still play a role by altering the composition of output at low levels of income, but it becomes much more effective once the abatement threshold is breached. Alternatively we can assume a fixed cost to either abatement or policy. Suppose there is a fixed cost CR of setting up a pollution reg- ulation system. When national income is low, the aggregate willingness to pay to reduce pollution to its first best level may be less than the fixed regulatory cost CR, in which case it is not worth setting up a regulatory system. With no system in place, pollution rises lock-step with output. Both threshold theories need two further assumptions to generate an EKC. First we need an assumption on the growth process to restrict composition effects. This is typi- cally done by adopting a one good frame- work or by limiting substitution possibilities by functional form assumptions. We will Copeland and Taylor: Trade, Growth and the Environment 21 33 Lurking in the background of this study is a compo- sition effect operating through changes in the fleet of cars. This composition effect is not investigated in the paper, although it may be responsible for the jump in lead per gallon of gasoline use at low income levels shown in figure 2 of the paper. Despite the proliferation of papers in this area, very little work has gone into evaluat- ing the various hypotheses offered for the EKC, or more generally in examining how the interaction between different sources of growth interact with income and other effects to determine the relation between growth and pollution. Unless we can clarify the causal mechanisms involved, the work will be of little use in helping us understand how growth or trade affect the environment. A few recent studies are useful steps in this direction. Hank Hilton and Arik Levinson (1988) examine the link between lead emissions and income per capita using a panel of 48 countries over the twenty-year period 1972–92. This study is important because it finds strong evidence of an inverted U-shaped relationship between lead emissions and per-capita income, and then factors the changes in pollution into two different components. The first is a technique effect that produces an almost monotonic negative relationship between lead content per gallon of gasoline and income per capita. The second is a scale effect linking greater gasoline use to greater income.33 This study is the first to provide direct evidence on two distinct processes (scale and technique effects) that together result in an EKC. To interpret the empirical evidence as reflecting scale and technique effects, one needs to rule out other possibilities. Although the authors do not couch their analysis in this context, their analysis implic- itly presents the necessary evidence. First, they document a significant negative rela- tionship between the lead content of gaso- line and income per capita (post 1983). This relationship shows up quite strongly in just a simple cross-country scatter plot of lead con- 34 To be precise we should note that since lead content per gallon is an average, and cars differ in their use of lead- ed versus unleaded gas, the composition of the car fleet is likely to be changing as well. Therefore, the fall in average lead content may reflect an income-induced change in the average age of the fleet (which would lower average lead content) plus a pure technique effect. tent against income per capita. We have depicted this in figure 2 below. Since lead content is arguably pollution per unit output, it is difficult to attribute the negative relationship in this figure to much other than income-driven policy differ- ences.34 Our interpretation is simply that regulation is tighter in higher-income coun- tries and this is driving down lead content (or e, emissions per unit output, in our framework). Second, the authors find a hump-shaped EKC using data from the post-1983 period, but in earlier periods they find a monotoni- cally rising relationship between lead emis- sions and income. The declining portion of the EKC only appears in the data once the negative health effects of lead become well known. The emergence of the declining por- tion in the income–pollution relationship is very suggestive of a strong policy response to the new information about lead. The fact that this appears late in the sample makes it difficult to attribute the decline in lead to other factors that could be shifting the demand for pollution. For example, if the declining portion of the EKC was due to increasing returns to scale in abatement, then it should appear in both the pre and post-1983 data. If it was due to shifts in the composition of output arising naturally along the development path, why would it only appear in the post-1983 data? While it is possible to think of examples where these other factors are at play, the scope for mis- taking a strong policy response for some- thing else is drastically reduced in this study. We are therefore led to conclude that the declining portion of the EKC post-1983 reflects a strong induced policy response that more than offsets the scale effect. 22 Journal of Economic Literature, Vol. XLI (March 2004) Figure 2. Real GDP Per Capita, 1992 L ea d pe r ga llo n (g ra m s) 0 10000 20000 0 3.5 Another paper that attempts to determine whether an income effect is responsible for the downward turn of an EKC is Kishore Gawande et al. (2000). They estimate an EKC for hazardous waste sites in the United States, and find that it is hump-shaped, although only a small percentage of counties are on the downward-sloping portion. Because it is very expensive to clean up haz- ardous waste sites, they argue that the income effect would be reflected in net out- migration rates. They find evidence that the number of hazardous waste sites in a region increases the net out-migration rate but only after a threshold of income is reached, which is consistent with an income effect driving the downward portion of the EKC. Moreover, the income threshold they esti- mate is indistinguishable from the peak of their estimated EKC. Gale and Mendez (1998) attempted to assess the importance of composition effects in predicting cross-country differences in pollution levels. They re-examine one year of sulfur dioxide data drawn from Grossman and Krueger’s (1993) study. They regress pollution concentrations on factor endow- ment data from a cross-section of countries together with income-based measures designed to capture scale and technique effects. Their results suggest a strong link between capital abundance and pollution concentrations even after controlling for incomes per capita. Their purely cross-sec- tional analysis cannot differentiate, however, between location-specific attributes and scale effects. Nevertheless, their work is important because the strong link between factor endowments and pollution suggests a role for factor composition to affect pollu- tion demand. That is, even after accounting for cross-country differences in income lev- els that may determine pollution supply, other national characteristics matter to pol- lution outcomes. Finally, two recent studies attempt to assess the relative importance of scale, tech- nique and composition effects in accounting for changes in pollution. Selden, Anne Forest, and James Lockhart (1999) compare emissions of six air pollutants in the United States in 1970 and 1990 and decompose the observed changes in pollution into changes in scale, composition of economic activity Copeland and Taylor: Trade, Growth and the Environment 23 35 They actually consider more than one technique effect. They measure changes in energy intensity and a compositional effect reflecting changes in sources of ener- gy as well as an “other technique effect.” 36 They also investigate compositional effects within the manufacturing sector by regressing per capita income on the average pollution intensity in manufacturing, holding the pollution intensity in manufacturing subsectors con- stant. They find that the average pollution intensity first falls with income, and then levels out. (due to changes in sectoral output shares), and changes in emissions per unit of out- put.35 Although this is simply a measure- ment exercise based on aggregate data for a single country at two points in time, the study is nevertheless important because it takes seriously the need to investigate the relative strength of the three effects. They find that technique effects were an impor- tant factor in explaining the fall in emissions. Although composition effects were present, they were not strong enough to account for the downturn in aggregate emissions during this period. Hettige, Mani, and Wheeler (2000) use panel data on industrial water pollution from twelve countries and try to isolate com- position and technique effects, and explain how they vary with income. They decom- pose pollution into the manufacturing pollu- tion intensity, the share of manufacturing in the economy, and total output, and then sep- arately regress firm level pollution intensi- ties, the average pollution intensity in manufacturing, and the manufacturing share on per capita income.36 They find a hump-shaped relation between the share of manufactures and per capita income; how- ever, they find this composition effect is small in magnitude relative to the impact of scale effects. Conversely, they find a strong technique effect: the income elasticity of the pollution intensity is about –1. Overall, they find that industrial water pollution tends to initially rise with income and then flatten out, with the strong technique effect being responsible for offsetting the scale effect of growth. Summary. The EKC literature is impor- tant in several respects: it brought the empirical study of aggregate pollution levels into the realm of economic analysis; it debunked the commonly held view that environmental quality must necessarily decrease with economic growth; and it pro- vided highly suggestive evidence of a strong policy response to pollution at higher income levels. The literature expanded rap- idly because of the ease of estimation and the potential relevance of its findings. Studies replicating or extending the methods of early contributors have played a useful role in providing a check on the original work, but further work along these lines has limited usefulness. Investigators must now move beyond the methods that sparked the literature to develop methods useful in revealing the causal mechanisms underlying the relationship. To proceed further more guidance from theory is surely needed. We would expect that scale, technique and composition effects all play a role in determining the rela- tionship between growth and the environ- ment. This suggests the focus on reduced forms linking only per capita income to pol- lution is unlikely to be fruitful. If we are to ask more detailed questions of the pollution data, we will need different methods. We suggest a step back from the EKC methods to consider theories determining the equi- librium level of pollution as a function of a relatively few factors. An approach that tries to disentangle the scale, technique and com- position effects, and which allows these to vary across countries has much more sup- port from theory and is more likely to gener- ate an increased understanding of what drives the relationship between growth and the environment. 3. Trade Liberalization and the Environment We now turn to the impact of interna- tional trade on the environment. We draw 26 Journal of Economic Literature, Vol. XLI (March 2004) Figure 3. Fixed Emission Intensity Y C B Z1 Z2 Z0 X Z =e0x q0 p A q1 comparative advantage. If a country has a comparative advantage in clean industries, then clean industries expand with trade; and conversely, if it has a comparative advantage in polluting industries, then dirty industries expand with trade. Fixed Emission Permits. Now suppose the government uses a marketable emission per- mit system to regulate pollution, and that it does not adjust the supply of permits in response to changes in the trade regime. Earlier, we noted the equivalence of permit 44 Rod Falvey (1988) obtains a similar non-equivalence result when comparing import tariff reform with import quota reform in models with multiple trade distortions. and tax systems as a method of implement- ing the first best. But if we hold policy instruments fixed in the face of shocks to the economy, this equivalence breaks down.44 If X is imported, then producers shift towards the clean industry when trade is lib- eralized, as in the previous example. This reduces the derived demand for pollution. Copeland and Taylor: Trade, Growth and the Environment 27 45 Although full-blown marketable permit schemes are relatively rare in practice, similar results would be obtained if the regulator enforces ambient air or water quality standards. If trade liberalization increased the derived demand for pollution, a regulator enforcing a rigid air quality standard would respond by tightening up regu- lations, which would raise the shadow price of pollution. 46 The welfare effect of trade liberalization in the pres- ence of environmental distortions has been a major theme of the literature: see Baumol (1971), Pethig (1976), Siebert (1977), Kazumi Asako, (1979), and Copeland (1994). 47 If an inefficient instrument is used to control pollu- tion then there are three distortions to worry about: the inefficient choice of instrument; the inefficient level of pollution; and the inefficiently low level of trade. We assume the government uses efficient instruments; extend- ing our analysis to the inefficient instrument case is left to the reader as an exercise. For given emission intensities, this would reduce pollution. But with a fixed supply of pollution permits, pollution will not change; instead the permit price falls. Producers switch to dirtier production techniques. This negative technique effect completely offsets the beneficial impact of the shift towards producing cleaner goods. Similarly, if X is instead exported, trade liberalization leads to an (upward) adjustment in pollution permit prices, but has no effect on pollution. An important implication of our analysis is that with rigid pollution taxes or emissions intensities, the environmental effects of trade liberalization may be quite substantial. But if pollution quotas are in place, the envi- ronmental effects of trade liberalization may be negligible.45 Welfare Effects of Trade Liberalization. The welfare analysis of trade liberalization in the presence of environmental problems draws heavily on the theory of the second best (Richard Lipsey and Kelvin Lancaster 1956).46 Prior to trade liberalization, there are two types of distortions: trade barriers and inefficient pollution policy.47 Reductions in trade barriers can either alle- viate or exacerbate the problems caused by inefficient pollution policy. Consequently, standard gains from trade theorems do not apply. To determine the welfare effects of trade liberalization, consider the effects on the 48 If we instead model trade barriers as tariffs (t), the welfare effects of trade liberalization can be written as dV/VI = tdM + (t - MD)dz. A reduction in tariffs will raise imports, so the term tdM represents the gains from trade. As well, there is the effect on pollution noted above. The magnitude of the gains from trade term differs according to the type of trade barrier in place; but the basic point that the welfare effects of trade liberalization depend on the interaction between the trade and environmental dis- tortion is robust. 49 The standard “double dividend” literature (see Lans Bovenberg and Ruud de Mooij 1995; and Don Fullerton and Gilbert Metcalf 1997) considers the effects of pollu- tion regulation in the presence of distortionary taxes on labor supply. That literature is also based on the interac- tion between two distortions. utility of the representative consumer of a small fall in the trade friction d. Replace p with pd in (7), and differentiate to obtain: (26) Trade liberalization has two effects on wel- fare: the standard gains from increased trade and the induced change in the environment.48 The standard gain from trade effect is pos- itive for both importers and exporters. If Home imports X, then M > 0 and the domes- tic price of X falls with trade liberalization. As a result Mdpd < 0. If Home exports X, then M < 0 but the domestic price of X rises with trade liberalization. Once again we find Mdpd < 0. Changes in pollution, however, can under- mine the benefits of trade liberalization. To see why, suppose emission intensities are constant and regulation is lax (t < MD). Then if home exports X, pollution rises with trade. Because the pollution tax is less than marginal damage, this increase in pollution is welfare-reducing (t – MD)dz < 0. The net welfare effect of liberalization is now ambiguous: the costs of increased pollution have to be compared with the benefits of increased goods consumption. If pollution is sufficiently damaging, pollution costs will dominate. On the other hand, if home imports X, trade liberalization may yield a double divi- dend by reducing pollution and generating increased consumption.49 With lax pollution dV V Mdp MD dzd 1 = - + -( )t 28 Journal of Economic Literature, Vol. XLI (March 2004) 50 The main point here is that if pollution is regulated with quotas, trade liberalization cannot exacerbate the pol- lution distortion. A similar point was made earlier in the literature on piecemeal trade policy reform, where Falvey (1988) showed that in an economy with multiple trade dis- tortions, alleviating one trade distortion will not exacerbate other trade distortions if import quotas are the instrument of protection. See Copeland (1994) for further details on trade policy reform in a world with many goods and pollu- tants. With pollution quotas in place, uniform tariff reduc- tions will increase welfare, but with pollution taxes (or fixed emission standards), then the welfare effects of trade liberalization depends on whether trade protection is biased towards clean or dirty goods. John Beghin, David Roland-Holst, and Dominique van der Mensbrugghe (1997) use a similar approach to investigate consumption- generated pollution, and Arja Turnen-Red and Alan Woodland (1998) consider multilateral reforms. regulation (t < MD), the economy gains from reduced pollution and the standard gains from trade. The instruments used also play an impor- tant role in determining the welfare effects of trade liberalization. If pollution regulation takes the form of a binding aggregate pollu- tion quota, then trade must always raise wel- fare, even when marginal damage is high and pollution regulation is lax. Referring to (3.6), as long as the pollution quota is bind- ing, pollution does not change with trade lib- eralization, and hence we have (t – MD)dz = 0. This leaves only the standard gains from trade, which as we have already shown must be positive.50 In summary, if pollution regulations are unresponsive, then the welfare effects of trade liberalization depend on the pattern of trade, the type of policy instrument used, and the existing stringency of pollution reg- ulation. If the number of pollution permits is held fixed during trade liberalization, then freer trade has to raise welfare and has no environmental consequences. But if emis- sion intensities are unchanged with trade, then trade increases pollution in countries with a comparative advantage in dirty goods, and decreases it in countries with a compar- ative advantage in clean goods. And when pollution policy does not fully internalize externalities, countries with a comparative 51 The analysis of the effects of trade liberalization on environmental quality has received less attention than the effects on welfare. Lopez (1994), Rauscher (1997), and Copeland and Taylor (2003) consider these issues in more detail. advantage in dirty goods may lose from trade. 3.2 Flexible Policy Now suppose government policy adjusts optimally in response to changes in the trade regime. Pollution is determined by setting the marginal benefit of polluting equal to marginal damage as in (11), where the goods price “p” is interpreted as the domestic price pd. When trade is liberalized, pd changes. By differentiating (11), we obtain, after some manipulation: (27) where D > 0, each of the elasticities is posi- tive, and the change in pd should be inter- preted as coming from a change in the trade friction d. Trade liberalization yields both an income effect (the first term) and substitu- tion effects.51 First, consider the income effect. Because trade liberalization raises real income (dI = - Mdpd > 0 when trade is liberalized) and environmental quality is a normal good, the income effect will always tend to reduce pol- lution. The pollution supply curve shifts back with liberalization and the strength of this income effect depends on the income elasticity of marginal damage, eMD,I. The substitution effects of trade liberaliza- tion, however, move in opposite directions for dirty-good importers and exporters. There are two substitution effects, one in production and the other in consumption. On the production side, an increase in the price of the dirty good stimulates production of the dirty good, and this tends to increase the demand for pollution. On the consump- tion side, an increase in the price of the dirty dz p M I dp dp d MD I d G P MD P dz= é ë ê ù û ú + -é ë ê ê ù û ú ú e e e, , , , D D Copeland and Taylor: Trade, Growth and the Environment 31 Figure 4. Trade Liberalization P A RS RSw RS T X/Y X/YX c/Yc X /Y C RD A T PA PT PA the North. This contracts dirty-good produc- tion X in the North and stimulates it in the South. Moreover, because we have assumed that each country holds pollution taxes fixed, pollution moves in the same direction as X. Pollution rises in the country with weak pol- lution policy (here the South), and falls in the country with strict pollution policy (here the North). Trade induced by pollution poli- cy differences creates a pollution haven in the country with weaker policy. The welfare effects of such trade depend on the stringency of pollution policy, as we discussed in the previous section. If pollu- tion policy is too weak, North must gain from trade, both because of an increase in purchasing power and because of the fall in pollution. South, however, may lose. Its income rises, but so does pollution. And if externalities are not fully internalized, the increase in pollution is harmful to the South. The predictions of this simple pollution haven model are consistent with some criti- cisms of freer trade. North gains from trade by offloading some of its polluting produc- tion onto the South. Moreover, because the dirtiest industry is shifted to the parts of the world with weaker environmental policy, this “global composition effect” tends to raise world pollution. Endogenous Pollution Havens: Income- Induced Policy Differences. A weakness of pollution haven models with exogenous pol- icy is the exogenous policy. Although authors often motivate the pollution policy differ- ences by income differences—such as a North-South income gap—pollution policies in these models do not respond when trade alters income levels. At best we should think of these as short-run models; at worst they contain a logical contradiction. The lack of a policy response affects both the positive and normative effects of trade even within a pol- lution-haven trading situation. To address this issue, we use our model to illustrate a simple version of the Copeland/Taylor (1994) pollution haven model where endogenous income-induced policy differ- ences create, and, respond to, trade. Consider two countries differing only in the scale of their endowment vector. That is, K = lK and L = lL , where l > 1. We also assume each country has the same number of consumers (which we normalize to one), so increases in L should be thought of as an increase in the supply of effective labor. 32 Journal of Economic Literature, Vol. XLI (March 2004) 55 That is, we assume that the regulator does not employ pollution policy to strategically manipulate the terms of trade in the goods market. This assumption is rea- sonable if we think of our model as a proxy for a world with many small Northern and Southern countries. We will turn, however, to the strategic trade policy issues later in this essay. 56 In particular, if the elasticity of marginal damage with respect to income is less than or equal to one, then pollu- tion rises with trade. Therefore, North’s workers are more highly skilled than South’s, but the ratio of capital to effective labor is the same in both. This means that North is richer than South, but because the K/L ratios are the same across countries, there is no incentive to trade in the absence of pollution policy. We assume the regulator acts as a price taker in world goods markets when choosing pollution policy.55 Because environmental quality is a normal good, the country with higher income chooses a higher pollution tax for any given goods price, and these differ- ences in environmental policy create an incentive to trade. To demonstrate, refer to the relative supply curve given by (28). North and South have the same K/L ratio, but North’s higher income means that its pollution tax is higher (t > t ). Consequently, North’s relative supply of X is to the left of South’s for any given p. Figure 4 can therefore be used to infer the trade pattern again. North’s high income gives it a comparative advantage in the clean good. When trade is opened, North will export the clean good (Y) and import the dirty good (X). The polluting industry will contract in the North and expand in the South. The low- income country becomes a pollution haven. The effects of trade on pollution can be inferred from our earlier results. Pollution falls in the North, as both the substitution and income effects of trade liberalization induce the policy maker to choose less pol- lution. Pollution will rise in the South as long the income effect is not too strong.56 Moreover, if income effects are not too strong, world pollution can rise with trade as well because the dirtiest industries shift to 57 See Copeland and Taylor (1994, 2003) 58 That is, we assume that t/p is identical across coun- tries. the country with weaker policy.57 But because both North and South fully inter- nalize pollution externalities, trade liberal- ization is welfare-increasing for both: income-induced policy differences are an efficient source of comparative advantage. Overall, pollution haven models are con- sistent in their prediction that freer trade leads the country with weaker pollution pol- icy to export the dirty good. In a model with endogenous policy, they predict that the low-income country has weaker policy and therefore the low-income country will export the dirty good. The effects of such trade on pollution and welfare, however, depend on the policy regime, as we have discussed above. A major weakness of the pollution haven models, however, is that they assume that policy differences are the only motive for trade. Factor Endowments Hypothesis. The main alternative to the pollution haven hypothesis is what we have called the factor endowments hypothesis. We illustrate this in figure 5. Relative demand (RD) is as before. To isolate the pure factor endowment hypothesis, assume pollution taxes are iden- tical and exogenous across countries, but relative factor endowments differ.58 Specifically, suppose North is relatively cap- ital abundant so that K/L > K /L . Let RS denote South’s relative supply curve. Because X is capital intensive and emission intensities are held constant, then North’s capital abundance pushes its relative supply curve (RS) out to the right of South’s. Hence the autarky relative price of X is lower in the North than in the South. With identical emission intensities across coun- tries, the capital-abundant country (North) exports the capital-intensive (dirty) good. Trade expands the polluting, capital-inten- sive industry in the capital-abundant country (the North), and pollution rises in the North. Copeland and Taylor: Trade, Growth and the Environment 33 Figure 5. The Factor Endowments Hypothesis P A RSRSwRS T X/Y X/YX c/YcX /Y C RD A T PA PT PA Conversely, pollution falls in the capital- scarce country (the South) as the polluting industry contracts there. Although we have illustrated the factor endowments hypothesis with a very simple example based on capital abundance, the key insight is that the impact of trade on the envi- ronment depends on a country’s underlying production capabilities. Countries relatively abundant in factors used intensively in pol- luting industries will on average get dirtier as trade liberalizes, while countries that are rel- atively abundant in factors used intensively in clean industries will get cleaner with trade. The predictions of this theory contrast sharply with those of the pollution haven hypothesis. If the factor endowments hypothesis is correct, and if a poor country is abundant in factors used intensively in clean industries, then its pollution will fall as trade is liberalized. Factor Endowments and Endogenous Policy Differences. We have illustrated the pollution haven and factor endowment hypotheses in isolation, but of course coun- tries differ in both their pollution policy and 59 The interaction between income differences and rel- ative factor endowments in determining the pattern of trade is analyzed in Copeland and Taylor (1997, 2003), Yves Richelle (1996), and Werner Antweiler, Copeland, and Taylor (2001). in their factor endowments. Rich Northern countries are likely to be both capital abun- dant and have stricter pollution policy than poorer Southern countries. North’s strict pollution policy will tend to make it a dirty- good importer, but its capital abundance tends to make it a dirty-good exporter. The pattern of trade depends on which of these effects is stronger.59 If relative factor endowments are similar but North is richer than South, then pollu- tion-haven effects dominate and North exports the dirty good. But if relative factor- endowment differences dominate relative income differences, then North will export the dirty good, despite having more strin- gent environmental regulation than the poor South. Since this result reverses the pattern of trade under the pollution haven hypothesis, it has a number of important implications. 36 Journal of Economic Literature, Vol. XLI (March 2004) TABLE 1 RANKING OF THE DIRTIEST MANUFACTURING INDUSTRIES Rank Air Water Metals Overall 1 Iron and steel Iron and steel Nonferrous metals Iron and steel 2 Nonferrous metals Nonferrous metals Iron and steel Nonferrous metals 3 Nonmetal. min. Pulp and paper Indust. chemicals Indust. chemicals 4 Petro. coal prod. Misc. manufac. Leather prod. Petro. refineries 5 Pulp and paper Indust. chemicals Pottery Nonmetal. min. 6 Petro. refineries Other chemicals Metal products Pulp and paper 7 Indust. chemicals Beverages Rubber products Other chemicals 8 Other chemicals Food products Electrical products Rubber products 9 Wood products Rubber products Machinery Leather products 10 Glass products Petro. products Nonmetal. min. Metal products Source: Mani and Wheeler (1997, p. 4) While this method is clearly not ideal, it has its strengths. For example, the set of dirtiest manufacturing industries appears to be fairly stable across both countries and pollutants. Therefore, identifying a dirty industry may not be that difficult. For future reference, we present the top ten dirty (manufacturing) industries ranked by air, water, and metals discharges. The data in table 1 are drawn from Muthikumara Mani and David Wheeler (1997). Given the similarities in the rankings across air, water, and metals discharges, it appears that identifying the dirtiest manu- facturing industries is relatively simple. The five dirtiest sectors often selected for inten- sive study are: iron and steel (371), nonfer- rous metals (372), industrial chemicals (351), pulp and paper (341), and nonmetal- lic mineral products (369). Using the same methods to identify clean sectors, Mani and Wheeler (1997) classify textiles (321), non- electrical machinery (382), electrical machinery (383), transport equipment (384), and instruments (385) as the five cleanest sectors in U.S. manufacturing. Under the assumption that this categoriza- tion of manufacturing industries into dirty and clean holds across both time and space, researchers construct the cross-country data needed for their analysis. Taking these new data as their dependent variable, these stud- ies proceed by linking variation in dirty- and clean-good trends to country characteristics such as income, income growth, or open- ness. The analysis may employ simple sum- mary statistics such as the Balassa revealed comparative-advantage measure, or employ regression analysis to explore the sensitivity of the series to several potential determi- nants. In all of these studies, researchers are searching for pollution havens; and there- fore, income differences, income growth rates, and measures of openness are the prominent explanatory variables. An immediate limitation of these studies is apparent: by measuring trends in dirty- industry output rather than pollution levels, Copeland and Taylor: Trade, Growth and the Environment 37 63 See table 6-2, p. 94. Developing country reliance on dirty-good exports varies by region. In Eastern Europe the percent rises from 21 percent to 28 percent; in Latin America from 17 percent to 21 percent; in South-East Asia the share of dirty-good exports in total exports is flat at 11 percent; and in West Asia it rose from 9.2 percent to 13 percent. they have necessarily assumed that changes in the composition of a country’s output cor- respond to changes in environmental quality. But if the techniques of production change over time because of trade, income growth, or technological progress, then a greater share of dirty-good output is consistent with both greater and lesser pollution levels. And as section 3.1 showed, changes in the com- position of output tell us relatively little about environmental outcomes except per- haps in the short run if emission intensities are fixed. Since many of these studies cover quite significant stretches of time, skepticism is in order. A second concern is that since the compo- sition of national output is affected by many factors, researchers in search of pollution havens run the risk of attributing any change in the composition of output to pollution- haven-driven trade rather than some alto- gether distinct domestic process. This risk is magnified by the avoidance of theory— which would naturally suggest alternative hypotheses—and an almost single-minded focus on income levels as a determinant of changing trade patterns. Despite these limitations, several authors have employed these methods to conclude, sometimes tentatively, that the rise in envi- ronmental control costs in the developed world has led to the creation of pollution havens in the South. For example, Patrick Low and Alexander Yeats (1992) find that over the 1965–88 period, the share of dirty goods in exports from industrial countries fell from 20 percent to 16 percent, but over this same time period the share of dirty goods in exports from many poor developing countries rose.63 64 Xinpeng Xu (1999) finds little statistical evidence for a change in competitiveness based on a primarily OECD sample. The raw data, however, indicate that for OECD countries, environmentally sensitive goods have fallen from 24 percent of exports in 1965 to 18 percent in 1995; whereas for the “other” category these shares have moved from 18 percent in 1965 to 22 percent in 1995. See table 1, p. 1219. 65 See Ratnayake (1998), tables 1 and 2, p. 82. 66 See Lucas et al. (1992, p. 80). Other researchers, employing slightly dif- ferent country groups and methods, corrob- orate these findings.64 Ravi Ratnayake (1998), in a study of New Zealand’s trade patterns, notes that in 1980, 96 percent of its imports of dirty goods came from the OECD, but by 1993 this had dropped to 86 percent. At the same time, the share of dirty goods imported from the developing coun- tries increased from 3 percent to 11 per- cent; but their share in clean-good imports only increased from 9 percent to 13 percent over this same period. Similarly, New Zealand’s exports of dirty goods to develop- ing countries fell from 59 percent of exports to only 46 percent, while its exports of clean goods rose.65 Similar results are presented in Robert Lucas, Wheeler, and Hettige (1992). They examine the toxic intensity of manufacturing output and GDP for over eighty countries during the 1960–88 period. They note that while toxic releases per unit of GDP fall as countries become richer, this only occurs because the composition of output in richer countries becomes cleaner. Coupling this with a finding that the greatest toxic intensi- ty growth occurred in the poorest countries leads the authors to conclude that all of their results are consistent with the view that “stricter regulation of pollution-intensive production in the OECD countries has led to significant locational displacement, with con- sequent acceleration of industrial pollution intensity in developing countries.”66 Nancy Birdsall and Wheeler (1992), in a study of pollution havens in Latin America, reach similar conclusions. They state: “Our 38 Journal of Economic Literature, Vol. XLI (March 2004) evidence is strongly consistent with the dis- placement hypothesis: Pollution intensity grew more rapidly in Latin America as a whole after OECD environmental regula- tion became stricter” (p. 167). Finally, in one of the most carefully constructed of these studies, Mani and Wheeler (1997) examine the production and consumption of dirty goods for several developing-country regions plus Europe, North America, and Japan over the 1965–95 period. Their con- clusions are more tentative than most, but they note (p. 28): Our cross-country evidence has found a pattern of evidence which does seem consistent with the pollution haven story. Pollution-intensive output as a percentage of total manufacturing has fallen consistently in the OECD and risen steadily in the developing world. Moreover, the periods of rapid increase in net exports of pollution-inten- sive products from developing countries coincid- ed with periods of rapid increase in the cost of pollution abatement in the OECD economies. The trends these authors identify are not really at issue, but the inferences drawn from them are. To underscore this point, we make use of the models from section 3.1 to illustrate how the findings are consistent with all of the first three hypotheses outlined at the beginning of this section. First suppose the pollution haven hypoth- esis is correct. Then as trade was liberalized throughout the sample period, we would expect to find each country specializing more in the goods in which it has a compar- ative advantage. The share of dirty goods in exports would fall in richer countries and rise in poorer countries, which is consistent with what has been found in these studies. Next suppose that a pollution haven effect is present, but not necessarily strong enough to make South a net dirty-good exporter. Then consider the effects of a tightening of pollution policy in the North, as it was throughout this time period. Referring to figure 5, North’s relative sup- ply curve would shift to the left, as would 67 We only need that Southern growth be biased towards the dirty sector, while Northern growth is either slower or not biased in this direction. world relative supply, pushing up the world price of dirty goods. As a result, some dirty- good production would shift from North to South, again causing the share of dirty goods in exports to rise in the South and fall in the North, as is consistent with the trends found by many authors in this branch of the literature. Finally, suppose there is no pollution- haven effect at all, and the factor endow- ments hypothesis is correct. That is, suppose that pollution policy plays little or no role in influencing trade flows. As well, suppose there is capital accumulation in the South during this time period. This alternative scenario is presented in the two panels of figure 6 below. In the left panel we again depict North’s and South’s relative supply curves. We assume that North is capital abundant and has a comparative advantage in the dirty industry. The initial world equilibrium is shown in the right panel at A with a price p0. Now consider the impact of an increase in South’s capital stock.67 Since capital accu- mulation favors the dirty industry, South’s relative supply curve shifts outwards from RS to RS 1. World relative supply shifts out as well from RSW to RSW1. The world price of dirty goods falls from p0 to p1 with the new equilibrium at point B in the right panel. We note, using the left panel, that the North specializes less in dirty goods since N1 < N while the South specializes more in dirty goods since S1 > S. Therefore, capital accu- mulation in the South implies a develop- ment path concentrating on capital-intensive dirty industries. Heightened competition in world markets caused by Southern develop- ment leads the North to shift out of these sectors, thereby concentrating on relatively clean manufactures. As a result, figure 6 roughly mimics the trends reported by many authors—although the reasons for these Copeland and Taylor: Trade, Growth and the Environment 41 72 If a pollution-haven effect were operative, but North had a comparative advantage in dirty goods, then the model’s predictions in this regard are more ambiguous. One the one hand, if South has a comparative advantage in clean goods, then in the absence of any environmental pol- icy in the North, more closed Southern countries would have a dirtier mix of goods. But if North introduces and tightens environmental policy, then dirty-good production would shift to countries in the South that are both open and have relatively weak environmental regulations. This could mean that if the pollution-haven effect is operative, more open Southern countries may have a dirtier mix of goods than more closed Southern countries. 73 However, future research must proceed carefully. For example, it would be useful to be more precise con- cerning the types of measurement error introduced by the classification of industries into dirty and clean. What assumptions are we making concerning this error? Is it correlated across time, countries, or industries? Southern countries would lower the share of dirty goods in Southern production and not raise it. Under the pollution haven hypothesis, relatively closed Southern coun- tries should have a cleaner mix of indus- tries; it is, after all, trade that is making them dirtier!72 Alternatively, under the Southern eco- nomic development scenario, the observa- tions on the mix of dirty industries have a natural explanation. In this view, North has a natural comparative advantage in dirty goods. Import substitution policies by Southern countries raise the share of dirty goods in Southern production. Therefore, under the factor endowment/economic development explanation, closed Southern economies should be relatively dirtier than their open counterparts. This is consistent with the empirical findings. While our analysis of the role of Southern capital accumulation in explain- ing the evidence is only suggestive, our point is that there are reasonable alterna- tives to the simple pollution-haven explana- tions that need to be considered when interpreting the evidence. The literature focusing on trends in dirty-good production and trade has made a useful start in gener- ating some stylized facts, but is still only in the early stages of explaining them.73 Future analysis must rely more heavily on 74 These were reviewed in more detail by Jaffe et. al. (1995) and Rauscher (1997). theory to suggest alternative hypotheses and discipline inferences from the available data. In particular, we suggest an investiga- tion of the role of factor accumulation in determining the pollution intensity of national output. The search for pollution havens in the data has obscured the role capital accumulation and natural resources must play in determining dirty-industry migration and trade. Environmental Stringency and Inter- national Competitiveness. A second branch of the literature uses data on the stringency of environmental policy to test whether or not environmental policy affects trade flows, foreign investment flows, or plant location choices. In our terminology, these studies can be interpreted as a test for the existence of a pollution-haven effect. These studies came in two waves. Virtually all of the early work (up to about 1997) relied on cross-sectional data. These studies link the cross-sectional variation in trade or investment flows to industry-, country-, or region-specific measures of regulatory costs and other variables that affect trade and investment, such as factor costs. The almost universal conclusion drawn by authors of these studies is that environmental-policy differences across countries or regions have little or no effect on trade or investment flows. A second wave of more recent work explicitly accounts for the endogeneity of pollution policy and unobservable industry- or coun- try-specific variables that may affect trade or investment flows. In sharp contrast to the earlier work, these studies have tended to find that differences in environmental poli- cy do affect trade and investment flows. We will start by briefly reviewing the cross-sec- tional studies,74 discuss some of the prob- lems with this approach, and then review recent work. 42 Journal of Economic Literature, Vol. XLI (March 2004) 75 This approach builds on the empirical trade litera- ture, in particular Ed Leamer (1984), who explains net export flows as a function of factor endowments. Tobey adapts this approach by adding a qualitative indicator of a country’s environmental policies 76 Student’s t for a two-sided test is 2.22 at the 5 percent level when there are ten d.f. We will adopt this level as rep- resenting statistically significant. The studies using trade data are all moti- vated by the Heckscher-Ohlin-Samuelson (HOS) model of international trade. James Tobey (1990) is a widely cited study of this type. He regressed cross-country data on exports of five dirty commodity groups on country-specific measures of factor endow- ments and environmental stringency for a group of 23 countries.75 In all the regres- sions reported, Tobey found the environ- mental stringency variable to be an insignificant determinant of net export flows. Moreover, in a follow-up omitted vari- able test conducted with a larger cross-sec- tion of countries, Tobey was not able to reject the hypothesis that environmental stringency had no effect on net exports. However, while Tobey’s analysis is per- haps the most cited study arguing against a link between environmental stringency and trade flows, its conclusions rest on tenuous foundations. His main results follow from a series of five cross-country regressions, one for each commodity group. Each of these regressions has only ten degrees of free- dom. Not surprisingly then, the vast major- ity of the coefficients estimated are insignificant.76 While the stringency vari- ables are insignificant, so too are 55 of the 65 coefficients estimated. In fact, most of the factor endowments are insignificant most of the time. Only one of the twelve included factor endowments is significant in three commodity groups (capital); and only one other is significant in more than one commodity group (a specific variant of land). Insignificance is the norm in these 77 Further concern follows from an examination of the omitted variable test. Tobey expands his collection of countries from 23 to 58, and excludes environmental strin- gency as a regressor in his cross-country regressions. He then divides his sample of countries into three groups according to development level and compares the propor- tion of positive and negative residuals from his net export regressions within each development group. Tobey is unable to reject the null hypothesis of identical propor- tions of errors; therefore, excluding environmental strin- gency does not lead us to overpredict or underpredict net exports in a systematic way across country groupings. But this test relies on the assumption that the omitted variable (environmental stringency) is orthogonal to the set of included regressors (Tobey 1990, p. 199, assumption A1). This requires that (unmeasured) pollution regulations be orthogonal to all of the country characteristics described by its endowments of capital, land, minerals and oil. But these characteristics determine both a country’s produc- tion structure (i.e. its demand for pollution) and its nation- al income (i.e. the supply of pollution). If we really believed the maintained orthogonality assumption, why would we be interested in grouping countries according to development level to conduct such a test? 78 See also Thomas Osang and Arundhati Nandy (2000), and Cees van Beers and Jeroen van den Bergh (1997). regressions.77 Another approach is to link the cross- sectional variation in trade flows to industry characteristics. Joseph Kalt (1988) and Grossman and Krueger (1993) are examples of this approach.78 A typical study of this type employs U.S. data on the cross-section- al pattern of trade in manufactures together with data on factor shares and pollution abatement costs. The standard study would estimate the following equation: (29) where Tit is a measure of trade flows in industry “i” in year t such as the value of net exports, Sit is an industry specific measure of environmental stringency, and X is a matrix of other controls which differ across studies. These controls are typically the cost shares of labor (sometimes disaggregated by skill class), the cost share of capital, and in some cases tariff rates. The set of industries stud- ied is often the entire U.S. manufacturing sector, but sometimes the analysis is limited T S Xit it it= + + +b b b0 1 m Copeland and Taylor: Trade, Growth and the Environment 43 79 They state, “in this paper, we will argue that proper- ly designed environmental standards can trigger innova- tion that may partially or more than fully offset the costs of complying with them. Such ‘innovation offsets’ … will be common because reducing pollution is often coincident with improving the productivity with which resources are used”( p. 98). An entirely different view of regulation’s effect on productivity is presented in Wayne Gray (1987). to the set of manufacturing industries that are most pollution intensive. The measure of trade flows, Tit, differs across studies. Since the underlying model motivating the equation is the HOS model, net exports is the most comfortable choice. Since industries differ in scale so dramatical- ly, most authors scale the dependent variable by industry shipments, value-added, or domestic consumption. The measure of environmental stringency is pollution abate- ment costs divided by industry value-added. This series is available for both capital and operating costs for all of U.S. manufacturing going back to the early 1970s. Most authors employ the operating cost series. The almost universal finding in this litera- ture is that pollution-abatement costs do not appear to explain the cross-sectional pattern of trade. In some studies, the sign of the coefficient on abatement costs is found to be counterintuitive, suggesting a positive rela- tionship between tighter regulation and net (or gross) exports. This awkward sign on the pollution-abatement cost variable has not led to a wholesale reexamination of the esti- mation methodology, but instead is often cited as evidence in favor of the Porter hypothesis (see Michael Porter and Cees van der Linde 1995). Porter et al. argue that tighter environmental regulation spurs tech- nological innovation, and hence tighter reg- ulation could, in theory, raise exports or lower imports.79 Despite the sometimes troubling sign on the pollution-abatement cost variable, the inference drawn from these studies was that there is little connection between the strin- gency of environmental regulation and trade 80 Another potential problem is suggested by the factor price equalization theorem of international trade. Suppose trade really is driven by differences in pollution policy. Then trade will reduce pressure on the environment in the North (which would tend to lower abatement costs there) and increase pressure on the environment in the South (which would tend to increase abatement costs). That is, trade could lead to some convergence in marginal abate- ment costs across countries, which would tend to make it more difficult to identify the effects of pollution policy on trade flows. flows. The explanations for this finding by the authors of these studies varied, but most often included the fact that pollution-abate- ment costs are only a small fraction of total costs.80 Recent work, however, suggests that there are some more fundamental problems with this approach. Endogeneity of pollution- abatement costs, and unmeasured industry characteristics may well be responsible for the results found. We illustrate the problem with the aid of figure 8. The left panel depicts a pollution demand and supply curve from our model of section 2. The right panel depicts import demand for the dirty good. This is graphed as a function of world prices, and shifts in response to changes in pollution taxes and another variable x which we assume is not observed by the researcher. We suppose increases in x shift in import demand. The existing literature links pollution- abatement costs to imports, conditional on some observable control variables, such as factor endowments or costs. To see the logic behind this approach, first suppose pollution taxes are exogenous, and suppose the pollu- tion tax rises from t0 to t1. Firms abate more intensively and abatement costs as a fraction of value-added rise. Pollution falls to z1 along the given pollution demand curve D0. (The pollution supply curve is not relevant at this point if we treat t as exogenous). In the right panel, the import demand curve shifts outward as less of the dirty good is produced. Imports rise to M1. This exer- cise predicts a positive relationship between 46 Journal of Economic Literature, Vol. XLI (March 2004) 83 Osang and Nandy (2000) employ both industry and time fixed effects which makes it more difficult to attribute their results to unobserved (constant) industry attributes. The endogeneity of regulation, however, could still be responsible for the results. 84 See table 5, p. 150, column 1. framework, both unmeasured natural- resource intensity and industry productivity shift pollution demand rightwards and import demand leftwards, obscuring the simple higher pollution-abatement cost–lower imports link.83 Grossman and Krueger also report coun- terintuitive signs for their pollution-abate- ment cost variable in four of the six cross-industry regressions explaining U.S. imports from Mexico. In only two of these cases is the negative relationship statistically significant at the 5 percent level. Nevertheless, the results are troubling, and the authors themselves note the strange sign on this cost variable may be arising from omitted variable bias. Further support for our interpretation is evident in related literatures. For example, consider Daniel Trefler’s well-known (1993) empirical paper on endogenous trade pro- tection. Trefler notes that empirical research had found only an embarrassingly small impact of tariff reductions on trade flows. He suggests that it arises from the treatment of trade barriers as exogenous. Trefler adopts a cross-sectional regression frame- work quite similar to that discussed above and estimates a standard one-equation model treating import barriers as if they were exogenous. He finds nontariff barriers have a small negative effect on imports. When non-tariff barriers are treated as endogenous, the results are striking: Trefler’s estimate for the impact of trade restrictions on imports is ten times higher.84 More direct evidence is presented in Levinson (1999), who examines the relation- ship between state-to-state hazardous waste shipments and import taxes on hazardous waste. Levinson regresses state-to-state waste shipments on state characteristics, dis- tances, and waste tax rates. With no correc- tion for endogeneity or unobserved heterogeneity, he finds a positive and statis- tically significant relationship between import taxes and imports of disposal waste. Once he accounts for the potential endo- geneity, he finds a strongly negative and sig- nificant relationship. Higher waste taxes deter imports of hazardous waste. These two studies show how endogeneity and unobserved characteristics can lead entire literatures astray. Given the wide vari- ance in results reviewed here, it is likely that similar problems abound in the literature on the effects of pollution policy on internation- al trade flows. Two recent studies have made some progress on this front. Levinson and Taylor (2002) present a simple model with endogenous pollution policy to suggest an empirical strategy testing for the impact of regulations on trade flows. They identify industry size, natural-resource intensity, political-economy concerns, and tariff substi- tution as likely candidates creating a link between regulation and imports. They then estimate a two-equation model adopting a methodology similar to Trefler (1993). The results are similarly striking. In the cross-sec- tion regressions with no correction for endo- geneity there is little relationship between net exports and pollution abatement control expenditures; however, once they instrument for pollution-abatement costs, the results change dramatically. Tighter pollution regu- lations lower net exports significantly. Similarly, Josh Ederington and Jenny Minier (2003) examine the link between pol- lution abatement costs and imports in a set- ting where imports and control costs are determined simultaneously. They motivate their work on the basis of tariff-substitution as outlined above. In their fixed-effects implementation they find a small but statis- tically significant relationship between pol- lution control costs and imports. A 1-percent point change in costs raises import penetra- tion by 0.53 percent points. In contrast, their Copeland and Taylor: Trade, Growth and the Environment 47 3SLS estimates which account for the possi- bility that pollution policy responds to import flows endogenously yields an unbe- lievably large impact, with a 1-percent increase in pollution-abatement costs raising import penetration by 30 percent. Further work along these lines is certainly required. Similar results have been found in the lit- erature on plant location. The main conclu- sion from studies using cross-sectional data from U.S. states was that differences in pol- lution policy across states do not affect plant location decisions (see Levinson 1996 for a review). These studies are subject to the same types of endogeneity problems dis- cussed above (see Vernon Henderson 1996). In response to this problem, several recent studies have used panel data and found neg- ative and statistically significant effects of environmental policy on plant location or manufacturing activity. Randy Becker and Vernon Henderson (2000) used a panel of U.S. county-level data. The measure of the stringency of envi- ronmental regulations was whether or not a county had attained the Clean Air Act’s national air quality standards. Since states are required to enforce strict environmental regulations in counties not in compliance with the national standards, nonattainment status is taken as an indicator of relatively more stringent environmental policy after the Act was passed. This study addresses three problems in the earlier plant-location literature. First, many studies had used a state-level indicator of the stringency of environmental regulation. This use of aggre- gate state-level data meant that regulatory differences within states could not be accounted for, and so much potential varia- tion in the data was lost. Second, since the air quality standard was set at the national level, and not in accordance with conditions within individual counties, the endogeneity problem we discussed about is mitigated. Finally, the use of panel data allowed them to control for unobserved heterogeniety across locations. The results are striking, as 85 The issue here is that states that attract polluting industries will have higher average abatement costs even if they face the same regulations as other states, because of the composition effect. Levinson (2001) proposes an index to correct for this. plant births for polluting industries in non- attainment counties were 26–45 percent below those in attainment counties. This suggests that air quality regulations had a significant negative effect on plant location. Other researchers have confirmed these results. Matthew Kahn (1997) found that the growth of manufacturing activity was reduced in non-attainment counties. Michael Greenstone (2002) found a negative effect on plant-level growth and employ- ment in polluting industries in non-attain- ment counties. John List et al. (2003) control for additional endogeneity issues and also find strong evidence that the Clean Air Act affected plant location in the United States. Keller and Levinson (2002) use different data to provide further evidence that envi- ronmental policy affects investment deci- sions, and that endogeneity problems were at least partly responsible for the failure of the earlier literature to find a significant effect. They use U.S. state-level data on for- eign direct investment inflows from 1977–94. Their measure of the stringency of environmental regulations is an index of pol- lutant abatement costs per unit of output in the state (which they adjust to take into account industrial composition effects with- in the state).85 They first replicate the results of earlier studies by using a pooled OLS regression of direct foreign investment on the index of abatement costs, without including state effects. The results indicate that DFI appears to be positively correlated with abatement costs, although in some cases the coefficient is insignificant. This is the same type of result the earlier cross-sec- tional studies found. Once they include state fixed effects, however, they find that the coefficient on abatement costs is negative and significant, as theory would predict.86 48 Journal of Economic Literature, Vol. XLI (March 2004) 86 Although the effect is significant, it is not large—a doubling of abatement costs reduces direct foreign invest- ment by about 10 percent. 87 Attempts to examine the pollution haven effect between rich and poor countries directly are hampered by data limitations. Gunnar Eskeland and Ann Harrison (2003) examine inward investment into four developing countries and find little or no evidence that either emission intensity or differences in U.S. environmental policy across industries affect inward investment into these countries. They do not have data, however, on host-country environ- mental policy, and although they do control for unob- served differences across countries, they have only four different host countries to work with. This strongly suggests that unobserved state- level variables correlated with abatement costs and investments were driving the earlier results. Summary. There is still much more work to be done on this issue. The investment and plant location studies have all used U.S. data, and most have relied on the Clean Air Act.87 Nevertheless, this growing body of work rep- resents a significant reversal of the earlier findings that pollution policy did not affect trade or investment. There are very few studies, however, that explicitly account for endogeneity of pollution policy when exam- ining the impact of tighter pollution regula- tions on trade flows. But even if this work stands up to further testing and scrutiny, it is important to emphasize the evidence found supports the existence of a pollution-haven effect only. The evidence indicates that after controlling for other factors affecting trade and investment flows, more stringent envi- ronmental policy acts as a deterrent to dirty- good production. None of this work presents evidence that this deterrent effect is strong enough to be the primary determinant of the direction of trade or investment flows. It seems likely, however, that differences in pollution policy may have a larger effect on trade and investment flows in the future. Some elements of pollution-abatement costs have been rising quite rapidly over time. For example, in 1984 pollution-abatement capi- tal expenditures represented only 2.8 per- cent of new capital expenditures in all U.S. manufacturing industries, but by 1993 this 88 These figures are drawn from Osang and Nandy (2000). As well, see the discussion of potential errors in reporting capital expenditures on pollution abatement in Jaffe et al. (1995; p.142) 89 Grossman and Krueger (1993) credit a 1990 govern- ment report for the introduction of the scale, composition, and technique effect terminology. share had risen to 7.0 percent. The increase in pollution abatement operating costs is much smaller from .63 percent of total costs in 1984 to only .79 percent in 1993.88 Unless we are willing to assume that these costly investments are somehow undone by Porter’s “innovation offsets,” or merely reflect prob- lems in survey methods, rising pollution- abatement costs must have some effect on trade flows and perhaps world prices. Scale, Composition and Technique Effects. Finally we turn to a group of studies attempting to estimate and then add up the scale, composition and technique effects arising from trade liberalization. Although the notion of scale, composition and tech- nique effects predates Grossman and Krueger (1993), economists did not pay much attention to this conceptual break- down until they employed it to assess the environmental impact of NAFTA.89 As such the Grossman and Krueger study was the first to fashion a logical argument along these lines. On the basis of their estimated EKC for sulfur dioxide, Grossman and Krueger con- cluded that any income gains created by NAFTA would tend to lower pollution in Mexico. This followed since Mexico’s then current per-capita income placed them on the declining portion of their estimated hump-shaped EKC. Since the shape of the EKC was taken to reflect the relative strength of scale versus technique effects, Mexico was literally now over the hump. Future income gains would call forth tighter regulation and lower pollution. To evaluate the composition effect of trade, Grossman and Krueger relied on both the evidence presented in their cross-sec- tional regressions and the results from CGE Copeland and Taylor: Trade, Growth and the Environment 51 Figure 9. relative income (world=1.0) el as tic ity left (reflecting a comparative advantage in clean goods). These elasticity estimates (from Antweiler et al. 2001) are plotted against per-capita income in figure 9 above. If the pollution haven hypothesis were an accurate descrip- tion of world trade in dirty goods, we would expect to see a strong negative relationship in figure 9. Rich countries would have a comparative advantage in clean goods. The opposite should be true for poor countries. Nothing like this pattern emerges from the figure. In fact, the relationship is positive, suggesting the composition effect of trade is more likely to be pollution increasing for high-income countries. And, since capital-to- labor ratios are higher in richer economies, the figure suggests that factor-endowment determinants of trade appear to dominate pollution-haven motives, and richer coun- tries appear to have a comparative advantage in emission-intensive goods. To some extent this result should not be surprising. Although the methods here are different, the conclusions were foreshad- owed in earlier work. Xu (1999) reports that 95 See Walter (1973), p. 67. 1.75 percent of the value of U.S. exports consisted of pollution-related costs, which is about 15 percent higher than the 1.51 percent estimated for U.S. imports. in 1995, almost 80 percent of the world’s exports of environmentally sensitive goods come from the OECD countries. Walter (1973) calculates the direct and indirect environmental control costs in both U.S. exports and imports to find that U.S. exports are actually very dirty by this measure, more than 15 percent more than its imports.95. Figure 9 is also consistent with the CGE evi- dence of Brown et al. (1991) predicting a reallocation of energy-intensive manufac- tures towards the United States and Canada as a result of NAFTA. Although the evidence in figure 9 suggests that there is a composition effect of trade that varies across countries, the estimated effect is quite small. One possible explana- tion for this, which is worth further investi- gation, is that the factor-endowment effects and pollution-haven effects tend to roughly 52 Journal of Economic Literature, Vol. XLI (March 2004) 96 This is not the only means for separating these effects. The authors also exploit variation across time, and cross-country variation in the relationship between GDP and GNP. 97 Country specific estimates differ with some point estimates of the net effect being positive. The hypothesis that scale dominates technique is however rejected by every individual country in the sample. offset each other. High-income countries are capital abundant, which leads to a compara- tive advantage in dirty goods, but they also have stricter environmental policy which works in the opposite direction and would tend to lead to a comparative advantage in clean goods. Therefore a small net effect is consistent with strongly offsetting motives. A second novel aspect of this paper is its method for separately estimating scale and technique effects. Previous work had been unable to separately identify these effects. The problem arises because both scale and technique effects are closely related to income levels. As a result, any increase in income is likely to shift both pollution supply and demand, making identification difficult. Antweiler et al. address this problem by exploiting the within-country variation in their dataset. Under the assumption that pollution policy is set at the national level, pollution supply is common for all cities within a given country. But cities differ in the scale of their output. Therefore, differ- ences within countries across cities in their scale can be used to distinguish between scale and technique effects.96 The estimates indicate that a 1-percent increase in the scale of economic activity raises pollution concentrations by approx- imately .25–.5 percent, but the accompanying increase in income drives concentrations down by approximately 1–1.5 percent via a technique effect. As a result, income gains created by freer trade lead to a net reduction in pollution concentrations from scale and technique effects. The estimated gap between the scale and technique effect seems large, and should be investigated further.97 It is consistent, how- ever, with the work of Sheoli Pargal and 98 See Pargal and Wheeler (1996, table 2, p. 1324.) David Wheeler (1996), who study the link between informal pollution regulation and community characteristics in India. Using plant-level data, they find a 1-percent increase in community income drives pol- lution down by 2.8–4 percent; while a 1- percent increase in output raises pollution by between .6–.7 percent. 98 The strong policy response is also consistent with the results of Hilton and Levinson (1998) on lead. The full effect of trade liberalization on pollution z requires that estimates of scale, composition, and technique effects be added up. Differentiating their reduced form for pollution concentrations with respect to a change in trade frictions b yields: (30) where the pi are estimated elasticities. The first term on the right-hand side is the scale effect, the second the technique effect, and the last the trade-created composition effect. If output and income change propor- tionately, (30) can be simplified to: (31) Equation (31) is very useful, but since Antweiler et al. do not estimate how a fall in trade frictions affects income levels (i.e. dI/db) further information is required. One approach is simply to restrict conclu- sions to the sample-average country. Taking the average value across countries yields p1 = 0.26 and p3 = 1.57 and p4 = -0.38. If trade raises incomes, then dI/db > 0 and this implies that for the average country in their sample, free trade leads to reduced SO2 pol- lution in cities. That is, for this pollutant, free trade is good for the environment. However, the effects are likely to be quite small. Most CGE models predict that even large trade lib- eralizations have only relatively small impacts on income and output. Combining this with small estimated composition effects of trade dz d z dI d Ib b b b= -[ ] +p p p1 3 4 dz d z dS d S dI d Ib b b b b b= - +p p p1 3 4 Copeland and Taylor: Trade, Growth and the Environment 53 on the environment, the net effect of trade on the environment is likely to be quite small. Finally, although this work tends to con- firm earlier work that suggested that the effect of trade on the environment is small, the explanation for this result is different. Earlier work tended to suggest that policy had no effect on trade patterns. In Antweiler et al., policy plays an important role in dampening the factor endowment effect. One final interesting aspect of the Antweiler et al. analysis is its implications for the literature on the environmental Kuznets curve. To isolate the role of international trade in the data, they attempt to distinguish between the pollution consequences of income changes brought about by changes in openness from those created by factor accu- mulation. This means their results can be used to investigate the hypothesis discussed in section 3 that the pollution consequences of economic growth are dependent on the underlying source of growth. As noted above, income gains created by freer international trade are, for an average country, beneficial to the environment. Neutral factor accumulation (an increase in income holding the capital labor ratio con- stant creates only scale and technique effects, and using the estimates mentioned earlier, one can conclude that it, too, is good for the environment in an average country. However, growth via capital accumulation turns out to worsen the environment. To provide some idea of the magnitudes involved, Antweiler et al. conduct a back-of- the-envelope calculation assuming a con- stant capital share in GDP of 1/3. Using the same method as above, the full impact of capital accumulation is given by: (32) where p2 is the elasticity of pollution with respect to a change in a nation’s capital-to- labor ratio. This was estimated to be approx- imately p2 = 1. The scale and technique elasticities p1 and p3 are as given before. If dz dk k z dS dk k S dI dk k I = + -p p p1 2 3 99 This confirms a point made by Grossman and Krueger (1993). capital’s share in income is 1/3, and popula- tion growth is zero over the time period con- sidered, then (32) can be simplified. And employing the estimates given, we obtain: (33) Growth via capital accumulation alone raises pollution concentrations for our average country, even after taking into account the income effect on environmental policy. This is an interesting result, despite its back-of-the-envelope flavor. It suggests that researchers investigating the environmental Kuznets curve need to pay more attention to the sources of economic growth. As well, it suggests that trade liberalization plus capital accumulation is far less environmentally friendly than trade liberalization alone.99 We would stress, however, that once we depart from the standard analysis of trade liberal- ization to include its potential effects on cap- ital accumulation, we should also include its potential role in facilitating technology transfer and accelerating technological change. Unless these are biased towards pol- luting sectors, these induced effects would tend to work in the opposite direction from that of capital accumulation. A couple of other recent studies seek to identify the effect of trade on environmental quality. Matthew Cole and Robert Elliot (2003) use national emission data to investi- gate several pollutants. They are not able to distinguish between scale and technique effects, but use Antweiler et al.’s approach to attempt to isolate the composition effect of trade. They confirm the Antweiler et al results for SO2, and obtain similar results on composition effects for CO2. But they find that BOD and NOx appear to respond dif- ferently, suggesting that it is indeed impor- tant to expand the scope of work to include + -[ ] = >1 13 26 1 57 56 0. . . dz dk k z = + -[ ] =p p p2 1 313 56 Journal of Economic Literature, Vol. XLI (March 2004) (35) (36) where we recall that MD denotes marginal damage. Noting home imports must equal foreign exports (E ), we have and by substituting this into (35) and (36), we can find the optimal policy pair: t = MD (37) (38) where e º pEp/E > 0 is the elasticity of the foreign export supply function. The solution reflects the policy targeting literature: externalities are fully internalized with environmental policy, and tariffs target the terms of trade (see, for example, Avinash Dixit 1985). When governments are unconstrained in their choice of policies, there is no incentive to weaken environ- mental policy to give local firms a competi- tive edge over foreign firms—this can be more effectively accomplished by using tariffs alone. Similar results apply to an exporter of pol- lution intensive goods (the Foreign country in our example). The foreign government’s optimal policy is to fully internalize external- ities and either protect its import-competing industry (Y) or equivalently to tax exports of the polluting good. When both countries use trade policy to improve their terms of trade, the world ends up in a standard noncooperative tariff game. This pushes them inside the global Pareto frontier because trade barriers drive a wedge between prices in the two countries. Both countries can therefore gain from a trade agreement that moves them back to the frontier. This is the motive for entering into a free-trade agreement in our model. t = 1 e dM dt E dp dtp = - + + - =M dp d tp dM d MD dZ dt t t t ( ) 0 - + + - =M dp dt tp dM dt MD dZ dt ( )t 0 101 As is well known, a large country can “win” a trade war in the sense that it is better off in the tariff-ridden equilibrium than in free trade. But because such an equi- librium lies inside the Pareto frontier, even a “winner “ of the trade war has an incentive to negotiate. As Wolfgang Mayer (1981) notes, any point on the Pareto frontier can be implemented with free trade combined with a lump sum transfer. A large country would demand a lump sum transfer (or some other concession) in return for free trade. 102 On this point see William Baumol and Wallace Oates (1988), Markusen (1975), and others. Suppose the two countries reach a binding agreement to eliminate tariffs.101 What happens to environmental policy? Setting the tariff equal to zero in (36) yields: (39) In response to the free trade agreement, Home’s optimal pollution tax diverges from marginal damage. Home still has an incen- tive to protect its import-competing firms, but with trade barriers eliminated, it has to fall back on second best instruments. If envi- ronmental policy is the only other instru- ment available, Home provides an implicit subsidy to its import-competing industry by setting the pollution tax below marginal damage. The discussion above has focused on the incentives facing the importer of polluting goods. The exporter of polluting goods also has an incentive to look for an alternative to trade policy. In this case, as mentioned pre- viously, Foreign has an incentive to reduce the world supply of the polluting good, and hence elimination of tariffs will cause it to tighten up environmental policy: (40) In this model, importers of pollution-inten- sive goods have an incentive to relax envi- ronmental policy to subsidize local production, while exporters have an incen- tive to tighten policy to tax production.102 t t t = - >MD E dp d dz d MD / / . t t t = + <MD M dp d dz d MD / / . Copeland and Taylor: Trade, Growth and the Environment 57 103 See Copeland (1990) for an early model of this process and Kyle Bagwell and Robert Staiger (2001) and Ederington (2001, 2002) for recent examination and sug- gested solutions. 104 Home and Foreign goods may be either homoge- neous or imperfect substitutes—we need Home’s demand to fall when foreign output rises. The result here is a special case of a more general result, which is that when there are multiple policy instruments available to gov- ernments, a free trade agreement that restricts only a subset of instruments is an incomplete contract that can be undermined as governments substitute towards uncon- strained instruments.103 In this context, the option of manipulating environmental policy to improve the terms of trade creates a loop- hole in the trade agreement. The result that governments can use envi- ronmental policy as a substitute for trade policy also appears in both the strategic trade and political economy literatures. In strategic trade papers, there is typically a two-stage game: government policy is set in the first stage, and imperfectly competitive firms move in the second stage. As James Brander and Barbara Spencer (1985) showed, if governments can make binding policy commitments in the first stage, they can give their firms a strategic advantage in the latter stage. In these models, once trade taxes and subsidies are eliminated, govern- ments have an incentive to switch to other instruments, including environmental policy if available (see Barrett 1994; Klaus Conrad 1993; Peter Kennedy 1994). To identify this motive, consider a partial equilibrium model with three countries, East, West and South. East and West each have one firm producing a dirty good that is sold only to the South—that is, all produc- tion is exported.104 And, for clarity, assume only one country, say West, is policy-active. The game proceeds in two stages. In the first stage, West chooses its pollution tax t, and in the second stage, the two firms choose output simultaneously. 105 Note however the role played by no domestic con- sumption in West. 106 Because the Western government moves first, it uses the first-order conditions from the second-stage Cournot game to predict how its policy affects the final outcome. Because the Western firm maximizes profits treating eastern output as given, we have ¶p/¶x = 0, which we have used to get (43). Because the model is partial equilibrium, and there is no domestic consumption of the dirty good, we can write West’s welfare func- tion as: where p is profits of West’s firm, x and x are West and East output of the dirty good, z is West’s pollution (pz > 0), and D is the pollu- tion damage function. If there were no Eastern firm, the West’s government would simply choose the pollu- tion tax so that the marginal benefit of pol- luting equals marginal damage: p z = MD (41) where MDº dD/dz. Moreover, in response to the pollution tax, the firm would choose its emissions level such that t = pz (42) And hence the solution would be imple- mented with a pollution tax set equal to marginal damage (t = MD).105 However, when the Western firm com- petes with its Eastern rival, the Western government has an incentive to use environ- mental policy to help its firm gain a strategic advantage over its rival. In this case, West’s optimal pollution policy is determined by:106 (43) (–) (–) Reducing the pollution tax (and therefore raising emissions) yields two benefits now— there is the direct reduction in the domestic firm’s costs (hence the increase in profits given by pz); but as well, there is a strategic p pz x dx d dz d MD+ = t t / / W x x z D z= -p( ) ( ), , 58 Journal of Economic Literature, Vol. XLI (March 2004) 107 Typically, this literature makes assumptions on demand and cost conditions to ensure that reaction func- tions slope down and a stability condition is satisfied. 108 This linear form ensures the marginal utility of income is unaffected by redistributions across the two groups in society and is a common assumption in the polit- ical economy literature. This simplifies the calculations tremendously but is not necessary for our main point here. effect—a reduction in Home’s pollution tax lowers the Home firm’s costs. This shifts out the Home firm’s reaction function in the output game, and causes the Foreign firm to reduce its output (as long as reaction func- tions slope downward)107. That is, a weaken- ing of Home’s environmental policy allows the Home firm to credibly commit to pro- duce more output, which leads to higher profits for Home. The Home firm always has an incentive to commit to more output, but such a commitment is not credible. A weak- ening of Home’s environmental policy helps the local firm out by making such a commit- ment credible. The optimal pollution policy for the Home country can now be written as: (44) where Home provides an implicit subsidy to the domestic firm by setting the pollution tax below marginal damage. Again we find envi- ronmental policy distorted. And finally, in the political economy liter- ature, governments respond to political pressure and use policies to redistribute income from one interest group to another. If trade policy and environmental policy are the available instruments, then once tariffs are eliminated, governments will manipulate environmental policy to help favored groups. Pollution taxes will be above or below mar- ginal damage depending on the political strength of competing interest groups. To illustrate the implications of this approach for environmental policy, consider a simple political support model. Suppose there are two agents: Labor and Capital, and suppose their utility functions take the form:108 U I p D z= - b( ) ( ) t p t t = - <MD dx d dz d MDx / / where I is income, b is a price index, and D the pollution damage function which is increasing and convex. The production side of the model is the basic competitive model we used earlier in the essay. Assume a small open economy (with goods prices fixed) to eliminate the terms of trade motive for intervention. Suppose that the government places a higher weight on capitalists than workers, and suppose that the only instrument avail- able is the pollution tax. The government chooses pollution to maximize: where UL is the utility of Labor, and UK is the utility of Capital. Then solving as before for the optimal pollution tax now yields: (45) An increase in allowable pollution raises the return to capital and lowers the return to labor in the model of section 3. Consequently, capitalists prefer more pollu- tion than at the socially efficient point (where t = MD); that is, in the relevant range, we have ¶UK/¶z > 0. If the govern- ment gives preferential treatment to capital- ists so that l > 0, then it will subsidize the pollution intensive industry by setting a pol- lution tax below social marginal damage. If the government gives preferential treatment to labor, then l < 0 and the pollution tax is above social marginal damage. Our analysis of tariff substitution has focused on production-generated pollution. Similar issues arise when pollution is gener- ated by consumption (such as with automo- bile emissions), although there are some important differences as well. For production-generated pollution, envi- ronmental policy tends to be directed at local firms, and so most of the scope for manipulating policy for protective purposes lies in either loosening or tightening policy t = - ¶ ¶ = - ¶ ¶ 2b b b dD dz U z MD U z K K l l W U UL K= + +( )1 l Copeland and Taylor: Trade, Growth and the Environment 61 incentive to tax the domestic firms to encourage cartelization by domestic firms. In simple linear models, the taxation motive tends to dominate once there are more than a small number of domestic firms. This implies that once we move to a model with several imperfectly competitive domestic firms, we find the standard result that emerges from competitive models, which is that the government has an incentive to raise the pollution tax above marginal dam- age to tax domestic firms to improve the terms of trade. Criticism also comes from environmental economists who note while trade policy is almost exclusively determined by central governments, much of environmental policy is set at the local, regional, or state level. Therefore, in order to put into place the terms of trade motivation for altering envi- ronmental policy we need a great deal of coordination between different levels of gov- ernment. Moreover, since states, regions, and even cities differ greatly it is not clear this cooperation would be forthcoming since their constituents may well be hurt by a price change that at the national level, would be welfare improving. Political economy models fare better in this regard since local authorities have it in their power to relax environmental standards to help local firms. However, these models are not immune from all of the problems mentioned above. For example, several authors have considered political economy models with both trade and environmental policy, using a Grossman-Helpman (1994) framework. Schleich (1999) pointed out that the usual targeting results hold in this model, so that if tariffs are available, there is no incentive to distort environmental policy (because in these models, the government wants to minimize the social cost of raising the income of favored groups). And as we demonstrated above, the political economy approach does not always imply that envi- ronmental policy will be too weak. If a pol- luting sector has relatively weak political influence, then environmental policy may be tightened in that sector to free up resources for other favored sectors. As Paola Conconi (2003) points out, if environmental groups are sufficiently strong relative to industry groups, then pollution policy may be more stringent than the Samuelson rule requires. Overall, the theoretical literature on tariff substitution does predict that governments may have incentives to manipulate environ- mental policy to help domestic firms in response to trade liberalization. There is no uniform prediction as to whether policy will be too tight or too weak, since predictions are sensitive to assumptions on the set of available instruments and market conduct. This is an area where empirical evidence is badly needed to examine where and when tariff substitution may operate. As long as the bulk of the evidence sug- gested that pollution policy had no measura- ble effects on competitiveness, as was the case at the time of the review by Jaffe et al. (1995), then it was easy to dismiss tariff- substitution as having little practical rele- vance. Recent findings by Levinson (1999), Becker and Henderson (2000), Greenstone (2002), Ederington and Minier (2003), and others demonstrate that pollution policy can indeed be used as an instrument of protec- tion. If this work holds up to further scrutiny, it is important evidence that concerns over tariff substitution need to be taken seriously. It is important to note though that these studies contain evidence that pollution regu- lations matter; they don’t provide evidence that governments alter pollution regulations in order to influence trade outcomes. The empirical evidence on this score is very lim- ited with only a couple of studies attempting to test for policy substitution. Gawande (1999) finds evidence that gov- ernments do substitute nontariff barriers for other instruments of protection, suggesting that the concern about loopholes in trade agreements is well founded. Eliste and Fredriksson (forthcoming) argue that gov- ernments weaken environmental policy to 62 Journal of Economic Literature, Vol. XLII (March 2004) shelter industries newly exposed to freer trade and provide some evidence from the farm sector to support this. Ederington and Minier (2003) use four-digit U.S. manufac- turing data from 1978–92 and find that their measure of the stringency of environmental regulation (abatement costs) has a negative and statistically significant relation with net import flows. This is consistent with the pre- diction that environmental policy is respon- sive to pressures from foreign competition. This is an area where much more empirical work is required to help clarify the policy debates. If governments do engage in tariff substi- tution, then it is still not clear how trade agreements should be modified to deal with the issue. There is a trade off between two different objectives: closing loopholes in trade agreements by constraining the use of domestic policy instruments in an effort to prevent tariff substitution; and allowing gov- ernments flexibility to respond to local changes in local conditions and preferences. In the case of product standards, there has been a gradual shift away from a pure national treatment regime to one that requires increased use of scientific evidence to justify environmental policies that impede trade. This may inhibit tariff substitution, but it reduces the flexibility of governments to implement their own environmental policy. A movement towards increased harmonization of product standards, as in the EU, goes even further in this direction. In the case of production-generated pollu- tion, governments have retained more flexi- bility to adjust their environmental policy, but at the possible cost of increased tariff substitution. This is because explicit export subsidies are subject to countervail laws under the WTO, but the use of implicit sub- sidies by weakening environmental policy are not. One of the main reasons for this seeming inconsistency in the treatment of subsidies is that the informational require- ments of determining the “correct” environ- mental policy are high. Even in our simple framework, the optimal pollution tax varies across pollutants, locations, and depends on preferences and its interaction with other pollutants. Consequently, a regime that attempts to forestall tariff substitution by allowing “green countervail” would likely be unworkable. Bagwell and Staiger (2001) have addressed this issue in their work on self-enforcing trade agreements. If trade agreements are negotiations to allow market access to trad- ing partners, and if these obligations bind, then they show that even if the trade agree- ment does not constrain domestic policy, it nevertheless removes tariff substitution motives. If each government maximizes its objective function subject to its market access commitments, then the solution is to minimize the cost of achieving those objec- tives. This eliminates the temptation to use environmental policy to reduce imports, since some other policy would have to be altered to counteract the negative effects on foreign market access. Although this is an elegant solution to the problem, it requires that we monitor country’s market access commitments very closely to eliminate possi- ble abuses. Since any measure of market access will involve trade flows and prices, shifts in comparative advantage or demand shocks would likely alter measured “market access.” As a result, evaluating whether a country has met its market access commit- ments may be no less difficult that determin- ing its “correct” environmental policy. 4.2 Trade Policy as a Substitute for Environmental Policy The other major policy issue we want to discuss is whether trade policy should be used to achieve environmental objectives. This issue arises most frequently in two con- texts. First there is simply a concern that trade may increase pollution, and therefore restrictions on trade can reduce pollution. Second, there are many issues where people in one country want to achieve an environ- mental objective in another country. Copeland and Taylor: Trade, Growth and the Environment 63 112 An export tax on X is analytically equivalent to an import tariff on Y—we could do the whole analysis in terms of a tariff on Y and get exactly the same result. 113 The pollution tax revenue is included in the nation- al income function G. If we had alternatively written G as a function of the pollution tax t, (instead of as a function of z), then we would have to add the pollution tax revenue to the G function to get total national income. Examples include proposals to ban tropical timber imports to protect rain forests, and the banning of tuna imports into the United States from Mexico to protect dolphins. We examine each of these cases in turn. On theoretical grounds the question of whether we should use trade restrictions to deal with the general issue of trade-induced increases in pollution is quite clear-cut. Trade restrictions are not a first-best instru- ment to deal with local environmental prob- lems. Return to our small open economy model, replace trade frictions with a tax on trade, and assume environmental policy is not optimal. Assume Home exports dirty goods. For concreteness, assume Home levies a specific export tax of t on exports. Then pd in our previous analysis is simply p – t.112 Home’s representative consumer has utility: where (48) and E is exports of X. Home’s pollution level is determined endogenously for any given pollution tax t by113 . (49) Now consider trade liberalization. The effect on welfare of a change in the export tax for given t is: (50) (–) (–) An increase in trade barriers has two effects on welfare. It reduces exports, which is harmful as standard gains from trade are lost. And it reduces pollution because the 1 V dU dt t dE dt MD dz dt1 .= + -( )t t = -G p t K L zz ( ), , , I G p t K L z t E= - +( ), , , U V p t I z= -( ),, , export sector pollutes. Solving for the opti- mal tax on exports we find: (51) (+) If pollution policy fully internalizes exter- nalities (t = MD), the optimal tax on trade is zero. Rather than using trade barriers, coun- tries can more effectively control environ- mental problems with instruments that are finely tuned to deal with the source of the problem, such as pollution taxes or quotas. This well-known result follows from the pol- icy targeting literature (see Dixit 1985). If environmental policy does not fully internalize externalities, then trade policy can be used as a second-best instrument to control pollution. Suppose, for example, the pollution tax is exogenously set to zero. Then if the only other available instrument is a tax on trade, its optimal level is: (52) By restricting trade, the pollution-intensive export sector is prevented from expanding to take advantage of trading opportunities, and this reduces pollution and raises welfare. While the use of trade policy for environ- mental ends seems simple and attractive, there are several problems with the analysis above. The first is simply that because of the complicated general equilibrium effects of trade liberalization determining the optimal second best trade policy to avoid environ- mental damage can be quite complicated (see Copeland 1994). Second, even if environmental policy is imperfect, trade may still be beneficial. As we demonstrated in section 3, the welfare results of trade liberalization depend on both a country’s comparative advantage and the instruments it uses for environmental protection. If a country has a comparative advantage in dirty goods, then the welfare impact of freer trade depends on whether imperfect regulation targets emission t MD dz dt dE dt = >/ / 0 t MD dz dt dE dt = - -( )t / / 66 Journal of Economic Literature, Vol. XLII (March 2004) many of the key issues of second-best policy analysis: free trade need not be welfare improving if environmental costs are not internalized; trade policy can be used as a substitute for environmental policy; and a commitment to free trade may create incen- tives to distort environmental policy. In many ways, however, this early work provid- ed more questions than answers. Optimal second best policy is very sensitive to the market structure, assumptions on technolo- gy, and empirical magnitudes. Consequently, one can draw on this early literature to sup- port a wide spectrum of policy proposals. The major contribution of the recent wave of research (which began in the early 1990’s) has been to shift the literature towards a focus on empirically motivated questions that are important to the resolution of policy debates. While this work is still new, we can identify three conclusions that are emerging. These conclusions should be the subject of further investigation, and some are them rest on more evidence than others, but if correct, they yield important implications for ongoing policy debates. The first conclusion is simply that there is now a great deal of evidence supporting the view that rising incomes affect environment quality in a positive way. This suggests that when we assess the effects of growth and trade on the environment, we cannot simply associate increased economic activity with increased environmental damage. Beneficial changes in environmental policy will likely follow and this leaves the net impact on the environment unclear. This is an important piece of evidence, but it has led the literature to focus almost exclu- sively on the impact of income effects. While incomes per capita are likely to be an impor- tant determinant of pollution policy (or pol- lution supply), actual pollution outcomes reflect the impact of other national charac- teristics as well (since they determine pollu- tion demand). Theory suggests that linking environmental outcomes to income per capi- ta alone is unlikely to be successful, just as predicting the pattern of trade in dirty goods by relative income levels alone is unlikely to be successful. Recent research finding a sen- sitivity of the environmental Kuznets curve to time periods or data may reflect the work- ings of important excluded national charac- teristics. If so, this would echo our concerns with an empirical literature that focuses too heavily on the role of income levels play in explaining the location of dirty good produc- tion and international trade. Future research should move away from estimating highly restricted models of pollution determination to consider alternatives giving a larger role to natural resources, capital abundance and other more conventional factors. Moreover, at a theoretical level it is still not well under- stood how the income effect interacts with the policy process, particularly in the context of political economy influences. A second major finding from the research of the last five years is that the previous con- sensus that environmental policy does not affect trade and investment flows was pre- mature. A number of recent studies find that both trade and investment are influenced by pollution regulations. This work illustrates the benefits of combining theory and empir- ical work. Once we interpret pollution levels and the stringency of regulation as equilibri- um outcomes, then pollution abatement costs are no longer independent of industry attributes. Consequently, measures of trade performance (such as import penetration) and pollution abatement costs are both endogenous variables. Therefore, the com- mon finding of a weak or nonexistent rela- tionship between pollution abatement costs and import penetration is likely to be a symptom of econometric problems and not evidence that environmental regulations are irrelevant. And the occasional finding of a positive relationship between pollution abatement costs and measures of competi- tiveness is surely not prima facie evidence of the “Porter Hypothesis.” Examining this endogeneity problem further should be a major focus of future empirical work. Copeland and Taylor: Trade, Growth and the Environment 67 A third, and more tentative, conclusion is that there is little convincing evidence to support the pollution-haven hypothesis. While there is evidence of a pollution-haven effect, it is only one of many factors that determine trade patterns, and there is no evidence that it is the dominant factor. Although the policy debate is often charac- terized as a conflict between those for and against globalization, it is really a struggle over how the rules governing trade should evolve. The fundamental issues involve the trade-off between allowing governments flex- ibility to pursue independent environmental policies (which sometimes may involve implicit or explicit restrictions on trade), and constraining the ability of governments in order to close loopholes in trade agreements. At present there is little evidence that envi- ronmental policy has been used to substitute for tariff protection. While tightening envi- ronmental standards does have cost and com- petitiveness consequences so too do almost all domestic policies. In the absence of strong evidence of abuse, we come down on the side of flexibility. Environmental policy should not be overly constrained by trade agreements. We also find little reason for trade policy to be used to achieve environmental ends either at home or in foreign countries. While restricting imports from countries with objectionable policies may indeed have envi- ronmental effects, these can easily be nega- tive. Lowering the access of developing countries to developed country markets is likely to lower their incomes and reduce their desire to adopt tighter environmental standards. It may also introduce perverse incentives by reducing the value of natural resources and therefore exacerbate environ- mental problems arising from the lack of property rights enforcement. And while the empirical evidence to date is not conclusive, trade restrictions on imports of developing countries may well lead them to adopt an even dirtier slate of production. This is not to say that concerned citizens in developed countries should sit idly by while the envi- ronment in developing countries worsens; rather it suggests these advocates adopt more efficient policies to enact positive change by supporting the use of direct finan- cial incentives, tied aid and capacity-building exercises in developing countries. And what about the bottom line? Is free trade good or bad for the environment? Most available studies suggest that the effect is small, but an answer to this question requires more careful empirical work guided by theory. Trade affects the environment via scale, composition, and technique effects, and these effects can all be expected to vary across countries. Some recent work has demonstrated how these effects can be iso- lated and estimated. Future work in this area should be attempting to refine, extend, and improve on these methods. Moreover, cur- rent work has looked at only a very few pol- lutants, and there is very little empirical work assessing the effects of trade on renewable resources. Finally, while Antweiler, Copeland, and Taylor (2001) find only small effects of trade on pollution concentrations, they also find relatively large impacts from changes in a nation’s factor composition. This suggests that the effects of capital mobility on environmental outcomes will be different than the effects of trade. More work is need- ed here, but until then it would be unwise for countries to use trade protection as a means to improve their environment. REFERENCES Andreoni, James and Arik Levinson. 2001. “The Simple Analytics of the Environmental Kuznets Curve,” J. Public Econ. 80:2, pp. 269–86. Antweiler, Werner; Brian R. Copeland and M. Scott Taylor. 2001. “Is Free Trade Good for the Environment?” Amer. Econ. Rev. 91:4, pp. 877–908. Arrow, Kenneth; B. Bolin, R. Costanza, P. Dasgupta, C. Folke, C.S. Holling, B.-O. Jansson, S. Levin, K.G. Mäler and C. Perrings. 1995. “Economic Growth, Carrying Capacity, and the Environment,” Science 268, pp. 520–21. Asako, Kazumi. 1979. “Environmental Pollution in an Open Economy,” Econ. Rec. 55:151, pp. 359–67. Bagwell, Kyle and Robert W. Staiger. 2001. “Domestic Policies, National Sovereignty and International Economic Institutions,” Quart. J. Econ. 116:2, pp. 519–62. 68 Journal of Economic Literature, Vol. XLII (March 2004) Barbier, Edward B. and Carl-Erik Schulz. 1997. “Wildlife, Biodiversity and Trade,” Environ. Devel. Econ. 2:2, pp. 145–72. Barrett, Scott. 1994. “Strategic Environmental Policy and International Trade,” J. Public Econ. 54:3, pp. 325–38. ———. 2003. Environment & Statecraft: The Strategy of Environmental Treaty-Making. Oxford: Oxford U. Press. Barrett, Scott and Kathryn Graddy. 2000. “Freedom, Growth, and the Environment” Environ. Devel. Econ. 5:4, pp. 433–56. Baumol, William J. 1971. Environmental Protection, International Spillovers and Trade. Stockholm: Almqvist and Wiksell. Baumol, William J. and Wallace E. Oates. 1988. The Theory of Environmental Policy. Cambridge: Cambridge U. Press. Becker, Randy and Vernon Henderson. 2000. “Effects of Air Quality Regulations on Polluting Industries,” J. Polit. Econ. 108:2, pp. 379–421. Beghin, John; David Roland-Holst and Dominique van der Mensbrugghe. 1997. “Trade and Pollution Linkages: Piecemeal Reform and Optimal Intervention,” Can. J. Econ. 30:2, pp. 442–55. Benarroch, Michael and Henry Thille. 2001. “Transboundary Pollution and the Gains from Trade,” J. Int. Econ. 55:1, pp. 139–59. Bhagwati, Jagdish N. and T.N. Srinivasan. 1996. “Trade and the Environment: Does Environmental Diversity Detract from the Case for Free Trade?” in Fair Trade and Harmonization: Prerequisites for Free Trade. Jagdish N. Bhagwati and Robert E. Hudec, eds. Cambridge, MA: MIT Press, pp. 159–223. Birdsall, Nancy and David Wheeler. 1992. “Trade Policy and Industrial Pollution in Latin America: Where Are the Pollution Havens?” in International Trade and the Environment. Patrick Low, ed. World Bank discuss. paper 159. Washington, DC: World Bank, pp. 159–67. Bovenberg, Lans A. and Ruud A. de Mooij. 1995. “Environmental Levies and Distortionary Taxation,” Amer. Econ. Rev. 86:4, pp. 985–1000. Brander, James A. and Barbara J. Spencer. 1985. “Export Subsidies and International Market Share Rivalry,” J. Int. Econ. 18:1–2, pp. 83–100. Brander, James A. and M. Scott Taylor. 1997. “International Trade Between Consumer and Conservationist Countries,” Resource Energy Econ. 19:4, pp. 267–98. Brock, William A. and M. Scott Taylor. 2003. “The Kindergarten Rule of Sustainable Growth,” NBER work. paper 9597. Brown, Drusilla K.; Alan V. Deardorff and Robert M. Stern. 1992. “A North American Free Trade Agreement: Analytical Issues and a Computational Assessment,” World Econ. 15:1, pp. 11–29. Brown, Gardner M. 2000. “Renewable Resource Management and Use without Markets,” J. Econ. Lit. 38:4, pp. 875–914. Bulte, Erwin H. and Daan P. van Soest. 2001. “Environmental Degradation in Developing Countries: Households and the (Reverse) Environmental Kuznets Curve,” J. Devel. Econ. 65:1, pp. 225–35. Cavlovic, Therese; Ken Baker, Robert Berrens and Kishore Gawande. 2000. “A Meta-Analysis of the Environmental Kuznets Curve Studies,” Ag. Resource Econ. Rev. 29, pp. 32–42. Chichilnisky, Graciela. 1994. “Global Environment and North-South Trade,” Amer. Econ. Rev. 84:4, pp. 851–74. Coate, Stephen and Stephen Morris. 1995. “On the Form of Transfers to Special Interests,” J. Polit. Econ. 103:6, pp. 1210–35. Cole, Matthew A. and Robert J.R. Elliot. 2003. “Factor Endowments or Environmental Regulations? Determining the Trade-Environment Composition Effect,” J. Environ. Econ. Manage., forthcoming. Cole, Matthew A. and Anthony J. Rayner. 2000. “The Uruguay Round and Air Pollution: Estimating the Composition, Scale and Technique Effects of Trade Liberalisation,” J. Int. Trade Econ. Devel. 9:3, pp. 343–58. Conconi, Paola. 2003. “Green Lobbies and Transboundary Pollution in Large Open Economies,” J. Int. Econ. 59:3, pp 399–422. Conrad, Klaus. 1993. “Taxes and subsidies for pollu- tion-intensive industries,” J. Environ. Econ. Manage. 25:2, pp. 121–35. Copeland, Brian R. 1990. “Strategic Interaction Among Nations: Negotiable and Non-negotiable Trade Barriers,” Can. J. Econ. 23:1, pp. 84–108. ———. 1994. “International Trade and the Environment: Policy Reform in a Polluted Small Open Economy,” J. Environ. Econ. Manage. 26:1, pp. 44–65. Copeland, Brian R. and M.S. Taylor. 1994. “North- South Trade and the Environment,” Quart. J. Econ. 109:3, pp. 755–87. ———. 1995a. “Trade and Transboundary Pollution,” Amer. Econ. Rev. 85:4, pp. 716–37. ——— 1995b. “Trade and the Environment: a Partial Synthesis,” Amer. J. Ag. Econ. 77:3, pp. 765–71. ———. 1997. “A Simple Model of Trade, Capital Mobility and the Environment,” NBER work. paper 5898. ———. 1999. “Trade, Spatial Separation, and the Environment,” J. Int. Econ. 47:1, pp. 137–68. ———. 2000. “Free Trade and Global Warming: A Trade Theory View of the Kyoto Protocol,” NBER work. paper 7657. ———. 2003. Trade and the Environment: Theory and Evidence. Princeton: Princeton U. Press. Coxhead, Ian and Sisira Jayasuriya. 2003. The Open Economy and the Environment: Development, Trade and Resources in Asia. Aldershot: Edward Elgar. Dasgupta, Susmita; Benoit Laplante, Hua Wang and David Wheeler. 2002. “Confronting the Environmental Kuznets Curve,” J. Econ. Perspect. 16:1, pp. 147–68. Dean, Judith M. 2002. “Testing the Impact of Trade Liberalization on the Environment: Theory and Evidence,” Can. J. Econ. 35:4, pp. 819–42. Dixit, Avinash K. 1985. “Tax Policy in Open Economies,” in Handbook of Public Economics, Vol.1. Alan J. Auerbach and Martin Feldstein, eds. Amsterdam: North-Holland. Dixit, Avinash K. and Victor Norman. 1980. Theory of