Regional Science and Urban Economics 33 (2003) 383–400 www.elsevier.com / locate / econbase

The geography of US pollution intensive trade: evidence from 1958 to 1994 Matthew E. Kahn* Associate Professor of International Economics, Fletcher School of Law & Diplomacy, Tufts University, Medford, MA 02155, USA Received 11 October 2000; received in revised form 12 June 2002; accepted 24 June 2002

Abstract Over the last thirty years, the United States has experienced a reduction in the share of manufacturing workers, increased international trade in manufacturing, and rising environmental quality. One possible factor driving these three trends is increased imports of dirty goods from ‘pollution havens’. As the volume of world trade grows, the United States has the opportunity to consume goods that are not produced domestically. This paper uses US trade data from 1958 to 1994 to study trends in dirty and clean trade. Bilateral trade regressions results show that poorer, non-democratic nations are not US pollution havens. The pollution content of Africa’s exports to the US are much higher than other continents.  2002 Elsevier Science B.V. All rights reserved. Keywords: Trade; Pollution havens; Geography JEL classification: F1; Q4

1. Introduction Over the last thirty years, the United States has made great progress in reducing its air and water pollution levels. Technique and composition effects have offset the increase in the scale of economic activity brought about by ongoing population *Fax: 1617-627-3712. E-mail address: [email protected] (M.E. Kahn). 0166-0462 / 02 / $ – see front matter  2002 Elsevier Science B.V. All rights reserved. doi:10.1016/S0166-0462(02)00042-X

384

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

and income growth. Technologies such as the catalytic converter for vehicles and electric utilities installing scrubbers have sharply reduced emissions per unit of activity. The ongoing sectoral transition from manufacturing to services has greened the economy. Between 1969 and 1999, the share of all US workers in manufacturing fell from 28.8 to 13.9%, while the service sector’s employment share grew from 23.5 to 37.4% (source: http: / / fisher.lib.virginia.edu / reis / region.html). Many environmentalists are concerned that wealthy nations have enjoyed improved environmental quality because they are ‘exporting their trash’. A familiar argument is that these nations are clean because through global trade they are increasingly importing dirty manufacturing goods from abroad and in particular from poorer nations. If this pollution havens hypothesis were true, then growing world trade would be a key causal factor in determining which nations have high levels of environmental quality. In particular if poorer nations are specializing in dirty trade that is exported back to richer nations, then the Environmental Kuznets Curve (EKC) will be a ‘sharp’ U relative to in a world under autarky.1 In this globalized economy, industrializing nations would experience sharp reductions in environmental quality while richer nations would be very clean (for some evidence on trade’s role in determining the shape of the EKC see Suri and Chapman, 1998, and Antweiler et al., 2001). The World Trade Organization has taken popular concerns about the pollution havens hypothesis seriously (see http: / / www.wto.org / english / tratop]e / envir]e / environment.pdf). After all, the logic behind this hypothesis is simply comparative advantage. Academic economists have been less persuaded that there is much evidence in favor of this hypothesis. Survey evidence shows that environmental regulatory costs rank low in terms of factors that influence firm locational choice (Panayotou, 2000). Concerns about reputation effects and expectations of future regulatory stringency increases in developing countries discourage multinational corporations from using dirty techniques abroad (Summers, 1992). Foreign direct investment is not flowing to industries in nations with lax environmental standards (Eskeland and Harrison, 2002). The costs of environmental regulatory compliance are low relative to labor costs and the Porter hypothesis goes as far as to suggest that regulatory costs are negative (Jaffe et al., 1995). Finally, a growing US empirical literature has documented that there are domestic pollution havens. Since US environmental regulation is not spatially uniformly enforced, regulated firms have a tendency to migrate to less regulated areas (Becker and Henderson, 2000; Greenstone, forthcoming; Kahn, 1997; Gray, 1997; Henderson, 1996). Such domestic pollution havens could substitute for international pollution havens. 1 The EKC graphs environmental quality on the vertical axis and national per-capita GNP on the horizontal axis. A ‘U’ indicates that development is first an ‘enemy’ of environmental quality but past some turning point, growth and the environment are positively correlated. For empirical estimates of this curve see Grossman and Krueger (1995) and Harbaugh et al. (forthcoming).

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

385

This paper contributes to the empirical pollution havens literature by studying changes in the composition of United States trade from 1958 to 1994. Trends in dirty and clean trade are examined. Using simple bilateral gravity trade regressions, this paper tests whether the greatest dirty trade growth has taken place with poorer non-democratic nations. By stratifying the bilateral trade regressions by continent, I test whether Africa is becoming a more important pollution haven over time. While in the past the United States imported more pollution intensive goods, the average pollution content of imports has been falling over time. Poor nations and non-democratic nations are not major exporters of pollution intensive goods to the United States. Relative to South America, Asia and Europe, Africa’s exports to the United States are the most pollution intensive. The next section briefly sketches the factors that determine whether dirty manufacturing will migrate abroad. Section 3 describes the aggregate US trade data and presents results concerning total US trade patterns from 1958 to 1994. Section 4 describes the bilateral trade data and presents gravity regressions using data from 1972 to 1994. Section 5 concludes.

2. Why could the spatial distribution of dirty trade change over time? A profit maximizing firm, whose customers are within the United States, will choose a production location by comparing each location’s production costs and its transportation costs. Since labor is a major cost of production, relative wages across areas will play a key role in manufacturing’s locational choice (Crandall, 1993). Rising real US wages provide an incentive for labor intensive manufacturing to migrate abroad. Even in the absence of environmental regulation, workers’ demand for being employed in ‘green’ jobs could create an international pollution haven effect. Labor economists have found ample evidence that ‘dirtier’ jobs (whether measured in job safety, long hours, or pollution exposure) must pay higher wages to attract workers (Garen, 1988; Hammitt et al., 2000; Hamermesh, 1999). This literature has also found evidence of an income elasticity for ‘green’ jobs that is larger than 1. Rising real wages in the United States will increase the wage premium that such workers require for doing onerous tasks. US environmental regulation has sharply increased since the early 1970s. Environmental regulation increased pollution abatement expenditure as a percentage of the United States’ GNP from 1.74% in 1972 to an estimated 2.6% in the year 2000 (Jaffe et al., 1995). Between 1985 and 1998, employment at the Environmental Protection Agency grew by 36.3% while overall Federal civilian employment fell by 10.8%. While the Porter Hypothesis posits that this regulation can actually lower a firm’s cost of production through encouraging innovation, most economists believe that increased regulation could push regulated firms to less regulated areas either within the United States or abroad. Whether this regulatory increase would significantly encourage pollution havens abroad depends

386

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

on changes in other nations’ environmental stringency over time and the willingness of US regulators to slacken their enforcement of environmental regulation (Ederington and Minier, 2002). Given that a polluting manufacturing firm migrates abroad, where will it go? All else equal, this industry would like to locate in a nation where labor is cheap and environmental regulation is lax. A poor non-democratic nation would offer this opportunity. In such a nation, a manufacturing firm would not have to worry about a muckraker press and could produce with ‘brown’ technology without being held accountable except for a side payment to the nation’s leader. If the firm’s output is cheap to ship back to US customers then this industry will have a larger number of possible locations to choose from.

3. Dirty trade trends based on aggregate us trade data This section reports new evidence on ‘clean’ and ‘dirty’ US trade from 1958 to 1994. The raw trade data are from the NBER trade data base (Feenstra, 1996, 1997). For each four digit SIC manufacturing industry between 1958 and 1994, the data set includes information on the United States’ domestic production, and total imports and exports (www.internationaldata.org). To convert the units into real units, I deflate the trade data using the Bartelsman and Gray (1996) shipments deflator. This yields the quantity of trade measured in millions of 1987 dollars. Two different measures of industry pollution content are used. The first is energy consumption per dollar of value added. The data source is the NBER Manufacturing Industry Productivity Database (Bartelsman and Gray, 1996; see www.nber.org). Energy consumption is positively correlated with a number of environmental problems such as air pollution, water pollution and greenhouse gas production. This energy consumption index varies across four digit SIC industries and for any one industry changes over time. The energy index is highly correlated with measures of industry air pollution emissions per unit of output. Using data for 360 four digit SIC industries from the World Bank’s IPPPS Pollution Intensity database (http: / / www.worldbank.org / nipr / data / ippsdown.htm), the correlation between the World Bank measures and this energy index is 0.45.2 In all of the tables reported below, this pollution intensity measure is called ‘Energy’. Over time, there is a great deal of persistence with regard to which industries are energy intensive. The correlation between the ranks of industry energy intensity in 1958 and 1994 is 0.71. The second measure of industry dirtyness is an industry’s carcinogenic toxic releases per dollar of output. I use the 1994 Toxic Release Inventory (see www.epa.gov / opptintr / tri) and focus on those chemicals that are known car2 Since the IPPS data is missing for many industries, I chose not to use this as an index of industry dirtyness.

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

387

cinogens (as defined by OSHA).3 To convert the plant level TRI data into an industry pollution index, I sum all carcinogenic chemical releases by a four digit SIC industry. This yields a measure of 1994 industry emissions measured in pounds. Dividing this by the industry’s domestic real value added in 1994 yields a measure of carcinogenic toxic releases per million dollars of output for each four digit industry. A ‘dirty’ industry has a high value of this index.4 By simply adding the total pounds of the TRI carcinogens, this pollution index imposes that any two chemicals that weigh the same have the same pollution content. Since the TRI data is from 1994, I am implicitly assuming that industry rankings do not change over time (i.e. that a dirty industry in 1992 was a dirty industry in 1972). In all of the tables reported below, this pollution intensity measure is called ‘TRI’. The correlation between Energy and TRI is 0.14. To show how these two pollution indices vary across manufacturing industries, Table 1 reports two regressions using the 1994 measure of TRI and the 1994 Energy index. Each of the pollution indices is regressed on 20 two digit SIC fixed effects. The R2 of 0.33 for the TRI regression and 0.46 for the Energy regression indicates that there is substantial within variation in industry pollution content. Some of the high pollution industries, based on both measures, include primary metals, stone, clay and glass products, petroleum and coal, chemical and allied products and Textile Mill products. The dirtiest three industries based on the TRI index are SIC 2812 (Alkalies and Chlorine), 3732 (boat building and repairing), 2611 (pulp mills) and based on the Energy index are 2813 (industrial gases), 3334 (primary aluminum), 2812. To begin to investigate the trade data, Fig. 1 reports how the average TRI content of US domestic production, imports and exports has changed from 1958 to 1994. For all of the figures reported in this paper, average pollution content is calculated as: Average pollution content in year t 5

O s( j, t)*X( j, t)

(1)

In Eq. (1), s( j, t) represents the share of economic activity in industry j in year t. In any given year, the shares sum to one. X( j, t) represents the pollution index for 3

The Superfund Amendments and Reauthorization Act of 1986 requires EPA to establish an inventory of toxic chemical emissions from certain facilities. The purpose of this data set is to inform the public of the presence of chemicals in their communities and releases of these chemicals into the environment. The data set covers over 600 chemicals. The reporting requirement applies to owners and operators of facilities that have 10 or more full time employees and are in SIC 20–39 (manufacturing) and manufacture, import, process, or otherwise use a listed toxic chemical in excess of specified threshold quantities. The EPA provides information on monitoring and enforcement of this self reported data. 4 Hettige et al. (1992) use this data to form their pollution index. Capitalization studies have measured the extent that home prices are lower in high TRI exposure communities relative to the observationally identical home in a low TRI area (Bui and Meyer, 1999).

388

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

Table 1 Variation in pollution content by two digit industry Log (TRI)

b Tobacco products 20.5152 Textile mill products 1.6098 Apparel and other textile products 20.0873 Lumber and wood products 2.0723 Furniture and fixtures 3.0724 Paper and allied products 1.0864 Printing and publishing 0.5532 Chemicals and allied products 4.0354 Petroleum and coal products 2.1359 Rubber and miscellaneous plastics 2.1707 products Leather and leather products 0.6909 Stone, clay, and glass products 2.0914 Primary metal industries 3.7572 Fabricated metal products 3.8459 Industrial machinery and equipment 2.2181 Electronic and other electric 3.1642 equipment Transportation equipment 4.0896 Instruments and related products 2.9688 Miscellaneous manufacturing 2.5212 Constant 0.5152 Observations 447 Adjusted R2 0.3316 The omitted industry is Food and kindred products.

Log (Energy) S.E.

b

S.E.

1.0085 0.4525 0.4397 0.5480 0.6068 0.5480 0.5480 0.4622 0.9108 0.8394

21.1100 0.3887 20.8050 0.1208 20.5756 0.1699 21.1944 0.3243 0.6966 20.0660

0.3339 0.1498 0.1456 0.1814 0.2009 0.1814 0.1856 0.1530 0.3016 0.2779

0.6485 0.4732 0.4732 0.4288 0.4062 0.4256

20.8333 0.6572 0.9132 20.2534 20.7617 20.8180

0.2147 0.1567 0.1567 0.1420 0.1345 0.1409

0.5480 0.6068 0.5169 0.2824

20.8584 21.0813 20.6753 23.1918 448 0.4636

0.1814 0.2009 0.1711 0.0935

industry j in year t. Since the TRI index is based on 1994 technique, any changes observed in Fig. 1 represent composition effects. The average TRI content of domestic production, exports and imports can only decline if the share of activity in high TRI industries declines. Has the pollution content of imports increased while the pollution content of US production fallen from 1958 to 1994? As shown in Fig. 1, in 1958 the average pollution content of US imports was over twice as high as exports and production. Between 1958 and 1994, convergence has taken place. The average pollution content of imports has declined much more than the average pollution content of domestic production and exports. Fig. 2 recalculates Eq. (1) using the Energy index as the pollution intensity indicator. The picture is almost identical to Fig. 1 but convergence is fully achieved by the late 1980s. The average energy content of imports had been very high at the start but has fallen sharply over time. It is important to note that unlike in Fig. 1, Fig. 2 is based on an index that changes over time. Thus, the average energy content of manufacturing can change due to composition effects (changes in the s( j, t)) or technique effects (changes in X( j,

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

389

Fig. 1. Trends in the United States manufacturing’s toxicity.

t)). Fig. 3 presents trends using the Energy index from 1972 (for any industry the s( j, 1972) weights do not change over time). Thus, like Fig. 1, any observed changes in average energy content represent composition effects. From 1958 to 1980, Fig. 3 looks quite similar to Fig. 2, imports had a higher pollution content but it has fallen sharply over time. The only surprising finding is that starting in

Fig. 2. Trends in the United States manufacturing’s energy intensity.

390

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

Fig. 3. US manufacturing’s energy intensity (1972 index).

1980, all three indicators start sloping up. While the levels in the early 1990s are still below the 1958 levels, the figure shows that the US has increased its imports and its exports of energy intensive industries after 1980. To further investigate trends in pollution intensive trade, I create a dependent variable defined as Y that equals (US Imports /(US Consumption)). The reduced form trade regression is presented in Eq. (2). Log(Yjt) 5 B1*Dirty 1 B2*Dirty*time 1 B3*High Emissions Industry 1 U (2) In Eq. (2), the unit of observation is an industry / year. Dirty is a two dimensional vector which includes log(Energy) and log(11TRI). Time represents dummy variables for calendar years. I divide the period 1958 to 1994 into three periods; 1958–1971, 1972–1982, 1983–1994. The first time period, 1958–1971, captures the ‘pre-regulation’ era. If the growth of the US Environmental Protection Agency’s regulation of industry is the key causal determinant of the pollution haven movement, then I would expect to see a discrete jump in the pollution content of imports when this legislation was enacted. The variable ‘High Emissions Industry’ in Eq. (2) is a dummy variable that equals one if it was listed in Greenstone’s (forthcoming) Appendix Table 2 as an industry with high emissions of carbon monoxide, nitrogen dioxide, vocs, sulfur dioxide and particulates. This variable is interacted with time dummies. Evidence that would support the pollution haven hypothesis would be if estimates of Eq. (2) yielded positive coefficient estimates of B1, B2 and B3. This

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

391

would indicate that higher pollution industries have a larger import share and that this share is growing over time. Each column of Table 2 reports an estimate of Eq. (2). The omitted category is 1958–1971. Regression 1 shows that the import share elasticity with respect to the toxic release inventory measure has increased over time. In the 1958–1971 period, a 1% increase in an industry’s TRI levels increased import shares by 0.03%. This elasticity grew to 0.11% by the 1983–1994 period. While small in size, this is evidence in favor of the pollution havens hypothesis. The energy consumption index tells the opposite story and the quantitative magnitude is much larger. In the 1958–1971 period, a 1% increase in energy consumption increased imports by 0.26%. This ‘pollution haven’ effect has vanished over time such that by the 1990s, there is a negative relationship between import shares and energy intensity. Regression 2 estimates the same specification but adds the additional measure of industry ‘dirtyness’. While it is difficult to measure environmental regulation’s intensity it seems reasonable that under the Clean Air Act that these high emitting industries would face the greatest Table 2 National trade regressions using data from 1958 to 1994 The dependent variable is the log (imports / consumption) Regression Explanatory variables log(TRI) log(TRI)*1(1972–1982) log(TRI)*1(1983–1994) log(Energy) log(Energy)* 1(1972–1982) log(Energy)* 1(1983–1994) High Air Emissions High Air Emissions* 1(1972–1982) High Air Emissions* 1(1983–1994) Observations Adjusted R2

1

2

3

4

0.0310 (0.0093) 0.0567 (0.0139) 0.0803 (0.0134) 0.2639 (0.0219) 20.3223 (0.0345) 20.4255 (0.0342)

0.0405 (0.0095) 0.0556 (0.0141) 0.0840 (0.0137) 0.2874 (0.0220) 20.3155 (0.0359) 20.3883 (0.0354) 20.2705 (0.0522) 20.0001 (0.0793) 20.2001 (0.0778) 15 794 0.1806

0.0675 (0.0153) 0.0456 (0.0222) 0.0787 (0.0214) 0.4805 (0.0312) 20.1875 (0.0473) 20.2200 (0.0460)

0.0835 (0.0161) 0.0258 (0.0233) 0.0631 (0.0225) 0.5003 (0.0318) 20.2128 (0.0498) 20.2388 (0.0488) 20.2454 (0.0771) 0.3062 (0.1171) 0.2374 (0.1138) 6873 0.1483

15 794 0.1742

6873 0.1469

Each column reports a separate regression estimate of Eq. (2) in the text. High Air Emissions, 1(1972–1982), and 1(1983–1994) are all dummy variables. Robust standard errors are presented in parentheses. The omitted category is trade in 1958–1971 in a low air emissions industry. Regressions 3 and 4 are estimated using data only on the Light to ship industries. The unit of analysis is a four digit SIC / year.

392

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

regulation. Given that the Clean Air Act was enacted in the early 1970s, I was surprised to find no evidence that the interaction terms ‘High Emissions*1972– 1982 dummy’ are not positive. High air emissions industries are not increasing their import share between 1972 and 1994 relative to the pre-regulation period. It is intuitive that cheap to ship manufacturing goods are more likely to be produced in pollution havens and then shipped back to the United States relative to heavier goods. To proxy for industry ‘lightness’, I use data from the 1977 Census of Transportation that reports median dollars of output per pound by industry. Using this data, I create a dummy variable that equals one if an industry’s dollars of output per pound is greater than the median across the 450 four digit SIC industries. This dummy variable is called ‘Light’. In Table 2, columns 3 and 4 report estimates of Eq. (2) for the subset of ‘Light’ industries. The pollution elasticities are much larger for these industries than for the entire sample. In the 1983–1994 time period, the TRI elasticity is 0.14. The energy elasticity was 0.48 in the earliest time periods and has fallen over time to 0.26 in the 1983–1994 period. These results show that light to ship polluting industries have always been imported from abroad. It is interesting that the TRI index suggests a small growth in pollution havens while the energy index shows a sharp decline in pollution haven effects. Regression 4 is identical to regression 2 but this is estimated only for the ‘Light’ sub-sample. The interesting finding here is that for this sub-sample there is evidence of a displacement effect in the post-EPA era. The large negative coefficient on ‘High Air Emissions’ in the 1958–1971 period vanishes and becomes positive from 1972 to 1982. For the light industries, high emissions industries were increasingly imported during this time period. In addition to the results presented in Table 2, a number of robustness checks were performed. The results are robust to including SIC fixed effects or industry controls such as the capital to labor ratio. Including the IPPS World Bank emissions factors did not change the basic results. Finally, a Democratic President dummy was coded and interacted with the dirty variables to test whether pollution havens have grown during Democratic administrations. This hypothesis was rejected.

4. Dirty trade trends based on US bilateral trade data In this section, US bilateral trade data from 1972 to 1994 are used to further explore changes in the spatial distribution of dirty trade over time. Throughout this section, the TRI and energy pollution indices are used. Unlike the previous section, this section’s empirical work is based on disaggregated bilateral trade data so a data point would be US trade with Mexico in 1984 in SIC 3312. Many environmentalists have been concerned that Mexico’s proximity to the United States and its relatively low level of environmental regulation make it a

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

393

Fig. 4. The toxicity of US trade between 1972 and 1994.

natural pollution haven. With the signing of NAFTA only accelerating trade between these nations, it is important to establish how trade trends have evolved. Figs. 4 and 5 are based on Eq. (1). For both Canada and Mexico, the pollution content of US imports and exports is presented. Fig. 4 uses the TRI pollution indicator and Fig. 5 uses the Energy pollution indicator. Both figures tell the same

Fig. 5. The energy intensity of US trade between 1972 and 1994.

394

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

story. In 1972, Canada, not Mexico, has been the major exporter of dirty goods to the United States. Between 1972 and 1994, Mexican imports have become cleaner and so have Canadian imports. Both Figs. 4 and 5 show that US exports to Mexico are on average dirtier than US imports from Mexico and this relationship has not changed over time. The pollution intensive trade partners of the United States do not change much over time. The correlation between national average energy imports in 1972 and 1994 is 0.63. In 1972, the average energy content of imports was highest from Zimbabwe, South Africa and Jamaica. In 1994, the dirtiest trading partners of the United States were Cyprus, Nigeria, and Venezuela. Similar to the previous section, trade regressions are run to generate new pollution havens facts. Unlike the previous section, the bilateral data allows for an investigation of how trading partner attributes affect the pollution intensity of trade. Eq. (3) presents the trade regression. log(Ylmt)5nation1controls1B1*Dirty1B2*Dirty*GNP 1B3*Dirty*Dictator1U

(3)

In Eq. (3), Ylmt is (11imports) from nation l in industry m in year t. The regressions include nation fixed effects. Taking a log of the dependent variable, Eq. (3) represents a gravity model (see Wall, 1999). In Eq. (3), GNP represents two dummy variables indicating whether a nation’s 1972 real GDP per-capita was in the world’s middle 1 / 3 of the income distribution or lower third of the income distribution. The omitted category is nations in the upper 1 / 3 of the world income distribution. The data source is the Penn World Tables (http: / / pwt.econ.upenn.edu / ). The bilateral trade regressions include 98 US trading partners. The names of these nations and their GNP income categories are given in Appendix A.5 The final important variable included in Eq. (3) is an indicator of a nation’s level of dictatorship in each year. The data source is Polity 4 (http: / / www.bsos.umd.edu / cidcm / polity / index.html). This data set includes a variable called ‘democracy’ which takes on the values 0–10 which represents a nation’s general openness of political institutions. The variable ‘Dictator’ is defined as 102Democracy. Table 3 presents five regressions to test the hypothesis that the typical nation that is a growing pollution haven is a poor dictatorship. Across the five regressions, this hypothesis is rejected. The omitted category is trade with a democratic, rich nation between 1972 and 1981. The coefficients on Log(TRI)*Poor Nation and Log(Energy)*Poor Nation are either negative or small positive and statistically insignificant. Across the five regressions in Table 3, the

5 Some nations were dropped from the regression estimates of Eq. (3) due to incomplete data on either trade with the US, real GNP per-capita, or the democracy indicators.

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

395

Table 3 Bilateral trade with the United States 1972–1994 The dependent variable is log(11Imports from nation l from industry m in year t) Whole Whole Whole Whole sample sample sample sample

Light sample

Regression

1 b

2 b

3 b

4 b

5 b

log(TRI) log(TRI)*Post-1982 log(TRI)*Poor Nation log(TRI)*Dictator Index log(TRI)*Dictator Index *Poor Nation log(TRI)*Middle Income Nation log(TRI)*Light log(energy) log(energy)*Post-1982 log(energy)*Poor Nation log(energy)*Dictator Index log(energy)*Dictator Index *Poor Nation log(energy)*Middle Income Nation log(energy)*Light Observations Adjusted R2 Fixed effects

0.0480 0.0125 20.1249 20.0029 0.0069

0.0482 0.0099 20.1207 20.0036 0.0072

0.0395 0.0095 20.1164 20.0031 0.0074

0.0307 0.0119 20.1282 20.0032 0.0074

0.0452 0.0227 20.0800 20.0036 0.0046

20.0750

20.0749

20.0715

20.0616

0.0440 20.0713 0.0220 0.0007* 0.0010

0.0582 20.0687 20.0005* 0.0033 0.0009

0.1091 20.0627 0.0048* 0.0027 0.0009

20.0783 0.0281 20.0232 20.0708 0.0316 20.0008 0.0010

0.0537

0.0817

0.0851

0.0578

0.0297

391237 0.3263 Nation

379832 0.166 None

0.1810 342250 0.335 Nation

170571 0.297 Nation

379832 0.188 SIC two digit

0.1505 20.0178 20.0065 20.0022 20.0015

Each column reports a separate regression estimate of Eq. (2) in the text. The omitted category is trade with a rich, democratic nation, from 1972 to 1981. The Table only presents OLS regression coefficients because almost all of the coefficients are statistically significant at the 1% level. A * indicates not statistically significant. The variables (Post-1982), Poor Nation, Middle Income Nation, and Light are all dummy variables. The variable Dictator Index takes on the values 0–10 where 10 indicates a dictatorship. Each regression includes year controls, and the log of US production of industry l in year t. Their coefficients are suppressed. Regressions 2 and 3 do not include nation fixed effects. Instead, controls for whether the nation shares a common language with the US, same currency area, land area, and its distance from the US are included.

hypothesis that dirty trade takes place with dictatorships is rejected.6 Similar to the results in Table 2, light industries feature greater pollution elasticities. As shown in regression 5, even for light to ship industries, the hypothesis that poor nondemocratic nations are the major pollution havens is rejected.7 6 It is important to note that nation fixed effects are included in each regression. Thus, changes over time in a nation’s political structure from dictator to democracy identify the political effect. 7 In Table 3, other controls in the regressions include year dummies, an industry’s total production in a given year in the United States and in regressions 2 and 3 national attributes. The data source is Adam Rose’s web page http: / / faculty.haas.berkeley.edu / arose / RecRes.htm. The control variables are listed at the bottom of Table 3.

396

Regression log(TRI) log(TRI)*Post-1982 log(TRI)*Dictator Index log(energy) log(energy)*Post-1982 log(energy)*Dictator Index Observations Adjusted R2 Fixed effects

The dependent variable is log(11Imports from nation l from industry m in year t) Africa South America

Asia

All

Light

All

Light

All

Light

All

Light

1 20.004* 20.0108 0.0003* 0.2932 20.0121 20.0185 45 545 0.1197 Nation

2

3

4

5

6

7

8

0.0308 20.0006* 20.0035 0.4583 20.031 20.037 19 315 0.1670 Nation

0.0876 20.0117 20.0076 0.1453 20.1013 20.0033 99 516 0.4185 Nation

0.0949 0.0067* 20.0091 0.1993 20.1023 0.0018* 44 875 0.4322 Nation

0.0883 20.0029* 20.0113 20.2175 20.0117 0.0138 88 063 0.2905 Nation

0.0944 0.0212 20.0106 20.0119* 20.0054* 0.0026* 37 646 0.2364 Nation

0.0385 0.0138 20.0053 20.0212 20.0521 0.01265 142 676 0.2802 Nation

0.0371 0.0237 20.006 0.1344 20.0323 0.0015* 62 009 0.2228 Nation

Europe

Each column reports a separate regression. Regressions 2, 4, 6 and 8 and are estimated using data only on the Light to ship industries. The table only presents OLS regression coefficients because almost all of the coefficients are statistically significant at the 1% level. A * indicates not statistically significant. The variable (Post-1982) is a dummy variable.

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

Table 4 Bilateral trade with the United States 1972–1994 stratified by continent

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

397

It is possible that the extent of the pollution haven effect differs by continent. To explore this, I estimate Eq. (3) stratified by continent. The regressions are run for the whole sample and the ‘Light’ sub-sample. As shown in Table 4, the largest US pollution haven effects are found in Africa (see regressions 1 and 2). The energy elasticity is twice as big as the next largest continent (0.46 versus 0.20 for South America). The African energy elasticity has fallen slightly over time. While the energy elasticity for Africa is large, the TRI elasticity is basically zero. In Africa, the hypothesis that dirty trade takes place with dictatorships is rejected. It is important to note what cannot be learned from regression estimates of Eqs. (2) and (3). This approach cannot decompose whether changes in dirty trade are taking place due to changes in US environmental regulation or changes in trading partner environmental regulation or whether other factors such as changes in relative wages are driving the observed patterns. To directly answer whether increased US environmental regulation increases the imports of dirty goods would require data on environmental regulation in each nation for each industry in each year. Even if one collected all of this, there would still be an issue of the endogeneity of environmental regulatory enforcement (Deily and Gray, 1991; Ederington and Minier, 2002).8

5. Conclusion Environmentalists have voiced fears that growth in world trade will lead poorer non-democratic nations to become major pollution havens. Trends in US international trade suggest that these concerns are misplaced. This paper has explored how the composition of trade has evolved over time. Between 1958 and 1994, the average pollution content of US manufacturing imports has fallen. Comparing the pre-Environmental Protection Agency (EPA) era to the post-EPA era, there is no evidence that dirty trade increased. Overall, poorer non-democratic nations are not pollution havens. One piece of evidence that supports the pollution haven hypothesis is that the average African nation is exporting energy intensive goods to the United States. All else equal, the elasticity of trade with respect to pollution content for ‘light’ industries is much higher than for ‘heavy’ industries. Shipping costs may be playing a key role in why the evidence for the pollution haven hypothesis is weak. As Eastern Europe integrates into the European Union, it may be no accident that 8

Ederington and Minier (2002) argue that research that has studied regulation’s impact on dirty trade underestimates the pollution haven effect induced by regulation. For example, if countries tend to relax environmental regulation on those industries facing strong import competition, then net imports and the level of environmental regulation may appear to be only weakly correlated across industries, even if stringent environmental regulations are a major source of comparative disadvantage.

398

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

these nearby ex-communist nations are required to harmonize their environmental regulatory standards with Western Europe’s to reduce the likelihood that they will become major pollution havens. Recent research has argued that trade liberalization improves the techniques used in production by poorer nations and thus reduces pollution externalities (Dasgupta et al., 2002). Future research should study differences in production techniques around the world. This paper’s pollution indices were based on United States production processes. Due to technological differences, and differences in capital stock vintages, other nations may be able to cleanly produce goods that would create a high level of pollution in the United States.

Acknowledgements I thank the editor, two reviewers and seminar participants at Columbia, Tufts and the Federal Reserve Bank of New York for useful comments. A previous draft of this paper was titled ‘US pollution intensive trade trends 1972–1992’.

Appendix A Countries included in the bilateral trade regressions by income level Poorest Third of World Real GDP Per-Capita

Middle Third of World Real GDP Per-Capita

Richest Third of World Real GDP Per-Capita

ANGOLA BENIN BANGLADESH BURKINA FASO BURUNDI CAMEROON CHAD CHINA EGYPT

ALGERIA BOLIVIA BRAZIL COLOMBIA CONGO COSTA RICA CZECHOSLOVAKIA DJIBOUTI DOMINICAN REPUBLIC ECUADOR GABON GUATEMALA HONDURAS IVORY COAST

ARGENTINA AUSTRALIA AUSTRIA BELGIUM CANADA CHILE CYPRUS DENMARK FINLAND

ETHIOPIA GAMBIA GHANA GUINEA GUINEA BISAU

FRANCE GREECE IRAN IRAQ IRELAND

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

HAITI INDIA INDONESIA KENYA LIBERIA MADAGASCAR MALAWI MALI MAURITANIA NEPAL NIGER NIGERIA PAKISTAN PANAMA ROWANDA SENEGAL SIERRA LEONE SRI LANKA SUDAN TOGO TUNISIA UGANDA

JAMAICA JORDAN MALAYSIA MOROCCO MOZAMBIQUE MAURITIUS NICARAGUA OMAN PARAGUAY PERU SYRIA TAIWAN THAILAND TURKEY ZAMBIA ZIMBABWE

399

ISRAEL ITALY JAPAN MEXICO HOLLAND NEW ZEALAND NORWAY POLAND PORTUGAL SAUDI ARABIA SINGAPORE SPAIN SWEDEN SWITZERLAND SOUTH AFRICA UK URUGUAY VENEZUELA

Income Categories are based on the Penn World Tables 1972 Real GDP Per-Capita.

References Antweiler, W., Copeland, B., Taylor, S., 2001. Is free trade good for the environment? American Economic Review September 91 (4), 877–908. Bartelsman, E., Gray, W., 1996. The NBER Manufacturing Productivity Database. NBER Working Paper, [ T0205, October. Becker, R., Henderson, V., 2000. Effects of air quality regulation on decisions of firms in polluting industries. Journal of Political Economy 108 (2), 379–421. Bui, L., Mayer, C., 1999. Capitalization and Regulation of Environmental Amenities: Evidence from the Toxic Release Inventory in Massachusetts. Mimeo. Crandall, R., 1993. Manufacturing on the Move. Brookings Institution, Washington, DC. Dasgupta, S., Laplante, B., Wang, H., Wheeler, D., 2002. Confronting the environmental Kuznets curve. Journal of Economic Perspectives 16 (1), 147–168. Deily, M.E., Gray, W., 1991. Enforcement of pollution regulations in a declining industry. Journal of Environmental Economics and Management 21 (3), 260–274. Ederington, J., Minier, J., 2002. Is environmental policy a secondary trade barrier? An empirical analysis. Canadian Journal of Economics. Eskeland, G., Harrison, A.E., 2002. Moving to Greener Pastures? Multinationals and the Pollution Haven Hypothesis. NBER Working Paper [8888, April.

400

M.E. Kahn / Regional Science and Urban Economics 33 (2003) 383–400

Feenstra, R.C., 1996. NBER Trade Database, Disk 1: US Imports, 1972–1994: Data and Concordances. NBER Working Paper no. 5515, March. Feenstra, R.C., 1997. NBER Trade Database, Disk 3: US Exports, 1972–1994, with State Exports and Other US Data. NBER Working Paper no. 5990, April. Garen, J., 1988. Compensating wage differentials and the endogeneity of job riskiness. Review of Economics and Statistics 70 (1), 9–16. Gray, W., 1997. Manufacturing Plant Location: Does State Pollution Regulation Matter? NBER Working Paper [5880. Greenstone, M., forthcoming. The Impacts of Environmental Regulation on Industrial Activity. Journal of Political Economy. Grossman, G., Krueger, A., 1995. Economic growth and the environment. Quarterly Journal of Economics 110 (2), 353–377. Hamermesh, D.S., 1999. Changing inequality in markets for workplace amenities. Quarterly Journal of Economics 114 (4), 1085–1123. Hammitt, J., Liu, J.-T., Liu, J., 2000. Survival is a Luxury Good: The Increasing Value of a Statistical Life. Mimeo, Harvard University School of Public Health. Harbaugh, W., Levinson, A., Wilson, D., forthcoming. Reexamining the Empirical Evidence for an Environmental Kuznets Curve. Review of Economics and Statistics. Hettige, H., Lucas, R., Wheeler, D., 1992. The toxic intensity of industrial production: global patterns: trends and trade policy. American Economic Review 82 (2), 478–483. Henderson, V., 1996. The effect of air quality regulation. American Economic Review 86 (4), 789–813. Jaffe, A., Portney, P., Peterson, S., Stavins, R., 1995. Environmental regulation and the competitiveness of US manufacturing. What does the evidence tell us? . Journal of Economic Literature 33 (1), 132–163. Kahn, M.E., 1997. Particulate Pollution Trends. Regional Science and Urban Economics. February 27 (1), 87–107. Panayotou, T., 2000. Globalization and the Environment, CID Working Paper [53, Harvard University. Summers, L., 1992. Preface. International trade and the environment. In: Low, P. (Ed.). World Bank Discussion Papers, Vol. 159. Suri, V., Chapman, D., 1998. Economic growth, trade and energy: implications for the environmental Kuznets curve. Ecological Economics 25 (2), 195–208. Wall, H., 1999. Using the gravity model to estimate the costs of protection. Federal Reserve Bank of St. Louis Review January / February, 33–40.