international differences in emissions intensity and ... differences in emissions intensity ......
TRANSCRIPT
International Differences in Emissions Intensity
and Emissions Content of Global Trade
Stratford Douglas# and Shuichiro Nishioka*
September 2009
Abstract: Understanding international differences in the emissions intensity of trade and production is essential to understanding the effects of greenhouse gas limitation policies. We develop data on emissions from 48 industrial sectors in 32 countries and estimate the CO2 emissions intensity of production and trade. We find no evidence that developing countries specialize in emissions-intensive sectors; instead, emissions intensities differ systematically across countries because of differences in production techniques. Northern and Western European countries have the lowest emissions-intensity, while Southern and Eastern European countries and China have the highest emissions-intensity. Developed countries such as Japan and the United States whose trading partners are mostly developing countries import the most emissions. F18: Trade and Environment Q27: Renewable Resources and Conservation: Issues in International Trade Q56: Environment and Trade; Environmental Accounting Keywords: Heckscher-Ohlin; Emissions Technique; CO2 Emissions; Environment
# Department of Economics, PO Box 6025, West Virginia University, Morgantown, WV, 26506 6025, Tel: +1(304) 293-7863, Fax: +1(304) 293-5652, E-mail: [email protected]. * Department of Economics, PO Box 6025, West Virginia University, Morgantown, WV, 26506 6025, Tel: +1(304) 293-7875, Fax: +1(304) 293-5652, E-mail: [email protected].
1
International Differences in Emissions Intensity and Emissions Content
of Global Trade
1) Introduction
Policies that are designed to limit anthropogenic global climate change must limit
greenhouse gas emissions, particularly carbon dioxide (CO2). Although the global CO2
sink is more or less a commons today, current and future international agreements
designed to limit CO2 emissions are expected to incorporate some form of cap-and-trade
mechanism that will cause emissions to carry a price. The impact of a carbon dioxide
price on trade, and therefore on the nature and distribution of industry worldwide, is an
open question whose answer will largely determine the distribution of gains and losses
from any international agreement to limit carbon dioxide. It may also determine the
agreement's prospects for success, as participation in international agreements is
voluntary.
Before any reasonably accurate assessment can be made of the global impact of
carbon dioxide emissions pricing on trade and industry, at least two questions must be
addressed. First, which countries currently take greatest advantage of the global
commons; that is, which countries' production and trade is most emissions-intensive?
Countries whose exports embody the most emissions will presumably feel the greatest
impact to their industrial base from the enclosure of this global commons, but countries
whose imports are most emissions-intensive will also feel an impact on their real income.
Second, and more fundamentally, what determines the emissions-intensity of production?
If a country's industrial emissions-intensity is primarily a function of the industrial
sectors in which it specializes, we would expect the cost burden from emissions pricing
to fall primarily on emissions-intensive sectors and the countries (“pollution havens”)
that specialize in them. If, on the other hand, emissions-intensity is a function of
production techniques, we would expect a wider distribution of the burden, and perhaps a
lighter overall burden, as emissions pricing would speed the adoption of less-intensive
technologies through every industry.
2
In this paper we employ the tools of the empirical trade literature to address both
questions. First, we address the question of the determinants of emissions-intensity by
comparing the results of two models of trade. The Hecksher-Olin-Vanek (HOV) model
explains the distribution of net exports by reference to factor endowments, holding
production technique constant worldwide within any given sector. The HOV model
modified in accordance with the Dornbusch-Fischer-Samuelson (1980), or DFS, model
(Davis and Weinstein, 2001) performs the same task, but allows techniques within an
industry to vary across countries. Because the DFS model relaxes the assumptions of
identical techniques and factor price equalization (FPE) in the standard HOV model, it is
better able to predict international differences in emissions-intensity of trade that are
driven by international differences in production techniques. We show that the DFS
model’s predictions are both very different from, and superior to, those of the HOV
model, and conclude that international differences in production techniques, not sectoral
specialization, explain international differences in emissions intensities.
We then use our emissions-intensity estimates to measure the current distribution of
emissions-intensive production and trade among the 32 countries in our data set. We find
evidence of an inverse relationship between level of development and the emissions-
intensity of production, with some outliers. Northern and Western European countries
have the lowest emissions-intensity, while Southern and Eastern European countries and
China have the highest emissions-intensity. Because of the size of its trade deficit the
United States imports the most emissions, although its imports are found to be less
emissions-intensive than its own products. We also find that Western and Northern
European countries import from less emissions-intensive (neighboring) countries while
East Asian, Pacific, and North American countries import from more emissions-intensive
countries.
The paper is organized as follows. In section 2 we describe our data set and our
method for allocating emissions among the ISIC industrial sectors used in the input-
output tables. In section 3 we examine the data on emissions intensity by industrial
sector and country, and we analyze the relationships between emissions-intensity and
capital-intensity, and between emissions-intensity and total factor productivity. In
3
section 4 we develop alternative empirical models of emissions content of trade and
evaluate their predictive power. Section 5 concludes.
2) Data and Emissions Technology
To estimate the emissions content of international trade, we require data on
production techniques and bilateral trade by country and by industry. We concentrate on
CO2 emissions, which constitute 77% of all greenhouse gases according to the World
Resources Institute. We focus on industrial production rather than household
consumption and transportation because we are interested in measuring emissions content
of trade. Our data set consists of observations from 32 countries1 and 48 industrial
sectors (listed in appendix table A) for the year 2000. Our sources are the OECD Input-
Output database (2007), the OECD STAN bilateral trade database (2006), and World
Resources Institute (WRI) CAIT-UNFCCC Climate Analysis Indicators (2008). The
WRI database provides industry-level data on CO2 emissions using the UNFCCC (1996)
Common Reporting Framework (CRF) of the Intergovernmental Panel on Climate
Change (IPCC).2 The OECD Input-Output database provides production data.
Unfortunately, the industrial sector definitions used by the CRF are different from
the ISIC Rev. 3 classifications used by the OECD Input-Output database. In fact,
emissions data using an industry classification system consistent with our production data
are simply unavailable for most countries. Even among the countries that provide such
data, the definitions of allocation methodology in emissions and industry classifications
differ from country to country. For example, Japan, Norway, the United Kingdom, and
the United States provide industry-level emissions data, but differ as to how they allocate
emissions resulting from electricity production and consumption. Turner, Lenzen,
Wiedmann, and Barrrett (2007) and Wiedmann, Lenzen, Turner, and Barrrett (2007)
provide a comprehensive review of data and estimation methodology.
1 Argentina, Australia, Austria, Belgium, Brazil, Canada, China, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, India, Indonesia, Ireland, Italy, Japan, Korea, the Netherlands, New Zealand, Norway, Poland, Portugal, Slovak Republic, Spain, Sweden, the Switzerland, Turkey, the United Kingdom, and the United States.
2 Details of classification definitions can be found at http://www.ipcc-nggip.iges.or.jp/public/gl/invs1.html.
4
Some previous studies (e.g., Lucas, Wheeler, and Hettige, 1992; Levinson, 2009)
have gotten around the problem of matching emissions to industries in different countries
by estimating emissions requirements for industries in a single country (typically the
United States) for which data were available, and assuming that industry-specific
emissions requirements were identical in all countries. This assumption of identical
production techniques imposes a strong restriction, and if that restriction is incorrect it
will bias the results. For example, it implies that not only U.S. exports but also U.S.
imports from other countries employ the same techniques and therefore have the same
emissions-intensity. One of the contributions of the current paper is the relaxation of this
assumption in order to allow different emissions requirements for each country and
industry. To accurately identify different production techniques in different countries
requires a method for constructing consistent and accurate emissions data by industry for
each country. We use a method similar to Ahmad and Wyckoff (2003) to incorporate
emissions from industrial sectors identified in the WRI/CRF data into industrial sectors
identified in the OECD Input-Output data base.3 Details of our method are described in
the data appendix.
Validation of Industry-Specific CO2 Emissions Estimates
We checked the validity of our emissions allocation methods by comparing them
with available official allocations obtained by national governments using survey
methods. In particular, Japan and Norway provide official data on CO2 emissions by
industry suitable for this purpose. In Japan, the “Law Concerning the Promotion of the
Measures to Cope with Global Warming” went into effect in 1998, following the
adoption of the Kyoto Protocol. As a result, the CO2 emissions data from 14,227
business units and 1,439 transport businesses became available for the public. Since the
Japanese government provides CO2 emissions by industry for the year 2006, we compare
them with our predictions4 by using the I-O table from 2006. We find our predictions
work well for Japan. The across-industry correlation between actual CO2 emissions and
3 Ahmad and Wyckoff studied the CO2 content of trade for non-service industries from 13 aggregated industries for 24 countries, and developed their estimates of emissions techniques from IEA data.
4 See equation (A-6) in the data appendix A.
5
predicted emissions is 0.792. Moreover, if we divide emissions by the corresponding
industry’s output (national currency basis) to obtain emissions intensity figures, the
correlation of emissions intensities increases to 0.926. As shown in figure 1-1, “non-
metallic mineral products” and “iron and steel” are the two most emissions-intensive
sectors in Japan, and our methodology precisely predicts these across-industry characters
of emissions.
The Norwegian Economic and Environment Accounts (NOREEA) Project published
industry-level data on CO2 emissions for the year 2001.5 While the environmental
accounts follow the NACE version 1 industry classification, there are several differences
between these environmental accounts and the official emissions data that follow the
IPCC’s common reporting framework (CRF). For example, since the environmental
accounts use the national accounts definition of Norwegian activity and not a
geographical definition of Norwegian territory, ocean transport and international air
transport are included.6 In addition, CO2 emissions from transportation are included in
corresponding sectors. By adjusting for these differences, the across-industry correlation
between official Norwegian CO2 emissions and predictions from our methodology is
0.783. Once we divide them by corresponding industry’s output, the correlation
decreases to 0.714. Even though these correlations are derived due to high emissions
intensities for transportation sectors, our methodology predicts the across-industry
variations in emissions for Norway accurately, as shown in figure 1-2.
Overview of the Country-Level Data
We present a summary of our CO2 emissions data from 32 countries and 48
industries for the year 2000 in table 1. Energy and industrial CO2 emissions for each
country are provided in the sixth column “total (1.+2.)” of table 1. Total emissions from
these 32 countries are 17,530 million tons, or 72% of total world emissions according to
WRI (2008). Of the 32 countries whose emissions we consider, China and the United
States are by far the largest emitters of CO2: China emits 27.4% of the 32-country total
5 See http://www.ssb.no/nrmiljo_en/arkiv/tab-2004-03-29-02-en.html. Detailed methodology is reported in Hass, Sorensen and Erlandsen (2002).
6 Total emissions based on national accounts are 56,493 kt, while the IPCC CRF emissions are 40,000 kt.
6
and the U.S. emits 24.6%. The rest of the columns in table 1 summarize the national data
on CO2 emissions used for our estimates. Since we are interested in production sectors,
our estimates of CO2 emissions do not include emissions generated from households’
electricity and heat consumption or transportation.
Emissions from electricity and heat production from industrial sectors account for
30% of total emissions. The proportion of electricity and heat in national CO2 emissions
varies widely across countries, depending chiefly on the characteristics of electricity
generation plants. For example, according to EIA (2008), electricity production accounts
for only 4% of total CO2 emissions for Switzerland, which generates 55% of its power
from hydro plants, purchases much of the rest from France, and only produces 2% from
fossil fuels. Similarly, electricity generation produces only 9.9% of total CO2 for France
(79% nuclear-powered). At the other extreme, electricity generation accounts for 50% of
emissions in India, where the fossil-fuel share of electric power production is 81% (75%
coal-fired). Similarly, Australia and Poland rely on fossil fuels for 92% and 97%,
respectively, of their electricity production; again, the primary fuel in those countries is
coal and the electricity sector accounts for more than 40% of total emissions in each case.
Overall emissions from industrial activities account for 62% of total emissions in the
countries we consider. They account for a substantial proportion in all countries, but the
exact proportion is quite variable, ranging from 42.7% of total emissions in Switzerland
to 88.6% in China. This variation across countries is observed because of cross-country
variation in factors such as capital-shares in production, emissions from transportation,
size of the manufacturing sector, and modes of power generation.
7
3) Productivity, Capital Intensity, and Emissions Requirements
Figure 2-1 is a scattergram of the association between the total factor productivity
(TFP) index7 and emissions intensity, measured as total industrial CO2 emissions divided
by real GDP. We find a clear decline in emissions intensity as TFP increases. A high-
TFP country requires less physical capital and labor per unit of production than a low-
TFP country does, so a developed country tends to emit less CO2 per unit of production.
The least efficient country in terms of TFP indices, China, employs relatively high-
emissions techniques as well. The United States has the world’s highest TFP and
displays a moderate level of emissions-intensity, but most European countries use much
cleaner techniques than the United States does.
Figure 2-2 shows the association between each country’s emissions intensity and its
capital-to-labor ratio. The relationship between capital intensity and emissions intensity
appears to be non-linear. A developing country such as India or Indonesia that employs
emissions-intensive capital (e.g., coal-fired power plants) may still produce fewer units of
emissions per unit value if it also employs labor-intensive techniques. While capital-
intensive sectors are typically emissions-intensive, capital-abundant developed countries
are typically more efficient producers and therefore less emissions-intensive. Outliers
both above and below the curve in figure 2-2 are explained at least in part by coal usage:
High-side outliers include Poland, with 70% of fossil fuel emissions from coal, China
(82%), the Czech Republic (60%), and the Slovak Republic (41%), whereas low-side
outliers include Argentina (1%), Switzerland (1%), Brazil (11%), and Sweden (16%).
Because capital-intensity is correlated with income, these results weakly support the
7 We first calculate the multilateral TFP index of Caves, Christensen, and Diewert (1982). The multilateral TFP index is defined as
1 1 1 1 1 1 1ln( ) ln( ) ln( ) ln( ) ln( ) 1 ln( ) ln( )2 2
c c c c c c c c c c
c c c cTFP Y Y L L K K
C C C C Cσ σ σ σ
⎡ ⎤⎡ ⎤ ⎛ ⎞ ⎡ ⎤ ⎛ ⎞ ⎡= − − + − − − + −⎢ ⎥⎜ ⎟ ⎜ ⎟⎢ ⎥ ⎢ ⎥ ⎢⎣ ⎦ ⎝ ⎠ ⎣ ⎦ ⎝ ⎠ ⎣⎣ ⎦∑ ∑ ∑ ∑ ∑ c
c
⎤⎥⎦
where C is the number of countries in the dataset, Yc is real value added for country c, Lc is country c’s labor force, Kc is its physical capital, and σc is its labor-compensation share.
8
environmental Kuznets inverse-U hypothesis discussed in Grossman and Krueger (1995)
and Roberts and Grimes (1997).8
Estimating Emissions Requirements in Techniques
The unit emissions requirement, a cz , represents the amount of CO2 emissions
required to produce one unit of net output value in industrial sector (z) in country c.
Previous studies have implemented the assumption that industry-specific production
techniques are identical by using a cross-sectional regression of emissions requirements
on industry-specific dummy variables. Country-specific deviations from industry-
specific emissions techniques appear in the regression as residuals. Following this
methodology we use weighted least squares to estimate the following equation,9
(1) 1 1ln( )c cz za zα ε= + ,
in which azc is industry z’s emissions per unit value of output, and the industry-specific
emissions intensity parameter α1z is estimated as industry z’s dummy variable coefficient.
A large number of studies in the trade literature (e.g., Bowen, Leamer, and
Sveikauskas, 1987; Trefler, 1995; Davis and Weinstein, 2001), have documented that the
standard HOV empirical model performs poorly unless it is modified to take account of
efficiency differences across countries. Those efficiency differences are sometimes
modeled as factor-specific (e.g., Trefler, 1993; Maskus and Nishioka, 2009). To
incorporate efficiency into the estimation, we estimate the equation
(2) 2 2 2ln( )c cz za c
zθ α ε= + + .
Here, θ2c is country c’s efficiency level (estimated as a country-specific dummy variable
coefficient), and the industry-specific dummy variable coefficient α2z corresponds to the
emissions requirement for industry z after adjusting country-specific efficiency
8 For a survey of the environmental Kuznets curve literature, see Dinda (2004) and Stern (2004). Harbaugh, Levinson, and Wilson (2002) and Bertinelli and Strobl (2005), for example, found little support for this hypothesis.
9 We use the weights proposed by Davis and Weinstein (2001), which control for heteroskedasticity due to smaller measurement error variance associated with larger factor shares and higher industry output. Estimation using OLS with robust standard errors provides substantially similar results.
9
differences. It is convenient to normalize efficiency differences to the United States, so
that θ2US=0, or exp(θ2
US)=1.
Because the aggregated data indicate a weakly nonlinear relationship between
emissions-intensity and country capital-to-labor ratio (see figure 2-2), we also estimate a
version of equation (2) that takes account of this relationship as a quadratic function:
(3) 2
3 3 1 2 3ln( ) ln lnc c
c c z zz z c c
z z
K KaL L
czθ α γ γ ε
⎡ ⎤⎛ ⎞ ⎛ ⎞= + + + +⎢ ⎥⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎝ ⎠⎣ ⎦.
Given the positive relationship between capital-intensity and income, if the
environmental Kuznets curve is an inverted U-shape, we would expect γ1>0 and γ2<0.
Table 2 presents estimates of θ c from equations (2) and (3), along with estimates of
γ1 and γ2 from equation (3) and diagnostic statistics from all three regressions. Values
less than zero indicate that a country uses cleaner production techniques than the United
States. Schwarz (SIC) and Akaike (AIC) statistics support adding variables to control for
country-specific efficiency differences, as in equations (2) and (3). The signs of the
capital to labor ratio coefficient estimates γ1 and γ2 in equation (3) are as expected, but
statistically insignificant at the 5% level. Inclusion of the capital to labor ratio appears to
add little information, since the country-specific efficiency coefficient estimates θ2c and
θ3c in equations (2) and (3) are nearly identical.
The t-statistics for the efficiency coefficients θ c indicate that almost half of the
countries, particularly European countries, use production techniques that are statistically
significantly cleaner than the United States. While there are significant differences in
emissions intensities between developed and developing countries, there are also
significant outliers. For example, panel I of table 3 indicates that Australia and New
Zealand are two of the fifteen “dirtiest” countries, while Brazil and Indonesia are among
the “cleanest” countries.
Panel II of table 3 reports emissions techniques, exp(α̂2z), for different industries,
based on estimation of equation (2). According to these estimates, iron and steel,
chemicals, non-metallic mineral products (e.g., cements), and electricity are the four most
10
emissions-intensive industries, while most service industries are less emissions-intensive.
The relatively low emissions-intensity estimates of transportation sectors are somewhat
misleading since liquid fuels used for transportation are not included in our measure. The
iron and steel industry’s highest ranking in emissions intensity is based on both its
electricity usage and its direct-fuel-combustion.
4) Estimating the Emissions Content of Global Trade
With estimates of the emissions intensity of production in hand we can proceed to
analyze the reasons for cross-country differences in emissions intensity, and by extension
we can shed some light on some recurring questions about the relationship between trade
and the environment. It is common in the literature (e.g., Grossman and Krueger, 1991)
to distinguish among three effects of increased trade and development on the
environment. First, the “scale effect” is the increase in emissions resulting from
economic growth, which often accompanies increased trade. Second, the “technique
effect” refers to the changing techniques of production that might occur as a country
develops, in part as a spillover from international trade or technology transfers. Finally,
the “composition effect” is the specialization in emissions-intensive industries that is
often alleged to be linked to a country’s stage of development. The hypothesized trade-
induced shift of polluting industries from high-income countries to less-developed
countries is sometimes referred to as “pollution haven” hypothesis.10 The nature of the
composition effect has been the focus of many previous papers (e.g., Copeland and
Taylor, 1995; Antweiler, Copeland and Taylor, 2001; Cole and Elliott, 2003; Levinson,
2009).
Our analysis focuses on modeling and measuring the technique and composition
effects. Compared to previous work, our method has the advantage of estimating
parameters of the different techniques used in different countries. Building our empirical
work upon the theoretical contribution of Copeland and Taylor (1995), we account for the
10
Lucas, Wheeler, and Hettige (1992) provided evidence that the growth in pollution intensity in developing countries followed a strengthening of pollution regulations in OECD countries. Keller and Levinson (2002) also found evidence for pollution havens. Recent evidence also includes Levinson and Taylor (2008) and Eskeland and Harrison (2003).
11
emissions content of global trade for two cases, distinguished by whether or not the
implicit price of emissions rights is the same across countries. For the FPE case in which
the implicit price of emissions rights equalizes across countries (implying identical
techniques) we employ the Heckscher-Ohlin-Vanek (HOV) model. For the case in which
the price does not equalize (consistent with international differences in production
techniques) we employ the Dornbusch-Fischer-Samuelson (1980) specification proposed
by Davis and Weinstein (2001). The data strongly support the latter specification,
providing evidence that the technique effect dominates the composition effect in
explaining the pattern of emissions content of international trade.
Emissions Content of Trade with Factor Price Equalization
We consider a world economy consisting of C countries and N industries. Countries
differ in their endowments of labor (Lc) and their allowable levels of emissions (Ec).11
Since emissions are produced jointly with output in the production process, the final
output of a product z can be written as a function of CO2 emissions (ez) and labor input
(lz). For tractability, consistent with Copeland and Taylor we assume a constant returns
to scale production function, , in which the parameter α(z) captures
the intensity of emissions for industry z, and industries are ordered according to their
emissions-intensity (0 < z <1). Higher values of z correspond to more emissions-intensive
industries: dα/dz > 0.
1 ( ) ( )( ) ( )zz z zY l eα α−= z
z zα−
Borrowing notation from Copeland and Taylor, we define τ as the price of a unit of
emissions and w as the price of a unit of effective labor, so that the unit cost function is
, where . We also define the
number of units of emissions and labor required to produce each unit of net output (Yz) as
az and bz, respectively. Since we initially assume identical production techniques across
countries, quantities az, and bz are initially the same in all countries. The standard HOV
model requires further assumptions that markets for products are perfectly competitive,
( ) 1 ( )( , ; ) ( ) zc w z z wα ατ κ τ −= ( ) (1 ( ))( ) ( ) (1 ( ))zz z zακ α α− −= −
11
In Copeland and Taylor (1995), pollution targets are implemented with a marketable emissions permit system, in which the government of country c sets its own emissions target, each firm purchases the profit-maximizing number of units of pollution permits, and all revenue is distributed back to consumers via lump-sum transfers.
12
there are no barriers to trade, and that factor endowments are similar so that factor prices
(τ and w) are equalized across countries (factor price equalization, FPE), in order to
measure the emissions content of global trade.
For each country the net-export vector can be obtained as the difference between net
production and final consumption:
(4) ( )c c c= − −T Ι B Q D
where Tc is country c’s N×1 vector of net exports, Qc is its N×1 vector of gross output,
and Dc is its N×1 vector of final consumption. B is an N×N indirect input-output matrix
for the unit intermediate requirements so that (I-B)Qc equals the net output vector Yc, and
(I-B)-1Yc = Qc. To measure the emissions content of net exports we apply our measures
of emissions intensity to the net export vector Tc. Define A as the 1×N row vector whose
elements represent the amount of emissions required to produce one unit of gross output
in each industry. Also define the total technique vector A(I-B)-1 whose elements (az)
represent the amount of emissions required to produce one unit of net output.
Premultiplying the net export vector Tc from equation (4) by the total technique vector
A(I-B)-1 gives country c’s measured emissions content of trade:
(5) 11 ( )c c
mF −= −A I B T
Because it is derived under the assumption of identical techniques across countries,
equation (5) corresponds to the case of factor price equalization (FPE) in the Copeland-
Taylor model.
Pre-multiplying the right-hand side of equation (4) by the total technique vector
A(I-B)-1 gives emissions content of trade as emissions from gross output minus emissions
embodied in final consumption: AQc − A(I-B)-1Dc.12 Assuming identical and homothetic
preferences and identical prices of goods and services, country c’s final consumption
vector is proportional to the world net output vector (Yw):
c cs=D Yw
12
Total domestic emissions AQc =Ec may be set by a regulatory authority, in which case the price of
admissions τ > 0, or if emissions are a pure externality then τ=0.
13
where sc is a scalar representing country c’s share of world expenditure. If, as in the
standard HOV model, production techniques are identical worldwide, then we have
A(I-B)-1Dc = scA(I-B)-1Yw = scEw.
where world emissions Ew=Σc Ec are the sum of all countries’ emissions. Thus, country
c’s predicted emissions content of trade is
(6) c c cpF E s E= − w
c z
cc ⎤⎦
The HOV model thus predicts measured emissions content of trade (Fcm1) from that
country’s emissions (Ec), its final consumption share (sc), and world emissions (Ew).
Emissions Content of Trade without Factor Price Equalization
The assumptions of FPE and identical production techniques across countries are
contrary to real world experience and create problems for the empirical performance of
the HOV model, as has been noted widely in the literature, e.g., Trefler (1993). Some
authors, including Copeland and Taylor (1995) have used the Dornbusch-Fischer-
Samuelson (1980) to relax these assumptions. If factor prices are not equalized, the price
of effective labor will tend to be lower relative to emissions prices in skill-abundant
developed countries, and higher relative to emissions prices in developing countries. As
a result, skill-abundant developed countries export skill-intensive products and import
emissions-intensive products in the DFS specification.
Davis and Weinstein (2001) introduced a DFS specification in which production
techniques vary systematically with countries’ factor abundances, and therefore the factor
contents of imports and exports must be measured separately using the producer
countries’ techniques. To implement this model consistent with the Copeland-Taylor
model, we retain the assumption of constant returns to scale in the production function
but introduce industry- and country-specific factor shares. In particular, we modify the
production function to , which includes country-specific emissions
shares ac(z) for each sector. The technique matrix Ac(I-Bc)-1 then becomes country-
specific as well, so that the measured emissions content of trade becomes:
1 ( ) ( )( ) ( )cc z
z z zY l eα α−=
(7) 1 ' ' ' 1 '2 ' '
( ) ( )c c c cc c cm c c c c
F − −≠ ≠
⎡ ⎤ ⎡= − − −⎣ ⎦ ⎣∑ ∑A I B X A I B M
14
where Xcc’ is an N×1 vector of exports from country c to country c’ and Mcc’ is an N×1
vector of imports from country c’ to country c. We use equation (6) to calculate
predicted emissions content of trade for both the DFS and HOV specifications; see Davis
and Weinstein (2001) for a full discussion of this point.
By comparing and contrasting empirical results from measurement equations (5) and
(7) we can gain insight into the reasons for the patterns of international variation in the
emissions intensity of production. If empirical results support the HOV model, equation
(5), we can conclude that countries with emissions-intensive exports specialize in
emissions-intensive industrial sectors. In other words, the HOV model implies that the
composition effect alone explains the observed international pattern of emissions contents
of trade. On the other hand, if the data do not support the HOV model but instead
support the DFS model, we can conclude that international differences in production
techniques determine the international pattern of the emissions contents of trade.
Measured Emissions Content of Global Trade
To measure the emissions content of trade empirically for equations (5) and (7), we
employ the following two equations:
(5') 1 ' 1 '1 ' '
( ) ( )c ccm c C c C
F − −∈ ∈
⎡ ⎤ ⎡= − − −⎣ ⎦ ⎣∑ ∑A I B X A I B Xc c ⎤⎦
c c ⎤⎦
(7') 1 ' ' ' 1 '2 ' '
( ) ( )c c c cc c cm c C c C
F − −∈ ∈
⎡ ⎤ ⎡= − − −⎣ ⎦ ⎣∑ ∑A I B X A I B X
where Xcc’ is an N×1 vector of exports from country c to country c’ and Xc’c an N×1
vector of exports from country c’ to country c. 1( )−−A I B is the average unit emissions
requirements after adjusting for country-specific efficiency differences; each of these
elements corresponds to 2ˆexp( )zα estimated from equation (2).
We estimate emissions content of trade among the 32 countries in our dataset only,
since we do not have data on techniques for the rest of the world.13 In addition, we use
imports from country c' to c (Mcc') rather than exports from c to c' (Xc'c) to avoid biases
13
Ahmad and Wyckoff (2003) estimated the emissions content of trade for the rest of the world by employing U.S. techniques. Also, the choice of proxies for the techniques can induce significant variation in estimates of emissions trade. Refer to appendix B19 “sensitivity to assumptions for non-IO countries” in Ahmad and Wyckoff (2003).
15
generated from issues such as transport costs, tariff rates, non-tariff trade restrictions, and
the statistical adjustments from Purchasing Power Parity (PPP), exchange rates, etc. Also,
aggregate CO2 emissions trades are balanced within the 32-country subset.
The leading exporter of emissions according to the HOV measure, equation (5'), is
China (267 million tons) and the second is Indonesia (63 million tons), followed by India
(54 million tons). The leading importer of emissions is the United States (302 million
tons) and Japan is second (86 million tons), suggesting at first glance that developing
countries specialize in emissions-intensive products and developed countries import these
products. Normalizing a country's emissions by country its labor force (Lc), however,
gives a different picture, as seen in figure 3-1, which shows a scatter plot between Fcm1/Lc
and TFPc. Figure 3-1 shows no significant relationship between the HOV measure of
emissions content of net exports per worker and productivity.
When we use the DFS measure, equation (7'), to allow for technical differences
across countries the volume of emissions trade increases significantly relative to the HOV
case, and the changes are systematic. The leading exporter of emissions using the DFS
measure Fcm2 is still China, but its emissions content of trade more than doubles, from
267 to 559 million tons (30% of emissions from Chinese production). Canada (86
million tons) is the second leading exporter of emissions,14 followed by India (81 million
tons). The leading importer of emissions is still the United States (437 million tons) with
Japan second (149 million tons). Figure 3-2 provides a scatter plot of the relationship
between the DFS measure of Fcm2/Lc and TFPc, showing that the DFS measure of
emissions content of trade per worker has a negative and significant relationship with
productivity.
Assessments of emissions intensity based on either equations (5') and (7') may be
somewhat misleading, however, because both equations measure emissions content of net
exports. Net exports vary with the country's trade balance as well as the emissions-
intensity of its imports and exports. Even if its industries were relatively clean, for 14 Since we restrict our data on bilateral exports to all the combinations of 32 countries, a country’s sum of
bilateral exports from those 32 countries is different from that country’s total exports. In particular, Canada trades mostly with the countries in our data (96.8 percent) so its emissions content of net exports tends to be greater than other countries.
16
example, China's large trade surplus would cause China to rank high among net exporters
of emissions because nearly all exports embody some emissions. To get a more complete
picture of emissions-intensity of trade, therefore, we adapt some additional measures
similar to Ahmad and Wyckoff (2003) and Levinson (2009). Let Mc be the total amount
of imports into country c, let mcz be the share of industry z in Mc, and mcc' be the share of
imports into country c from country c'. Also, let Qc be country c’s gross output, and let
qcz be the share of industry z in total outputs Qc. Recall that exp(α̂2z) measures the
emissions content per unit of output in industry z. We use the expression
(8) ( ) ( )1 2ˆ ˆexp( ) / exp( )N N
c cz z z zz N z N 2
cMix m qα α∈ ∈
= ⋅∑ ∑ ⋅ ,
where NN is the subset of non-service industries, to compare the emissions intensity of the
import product mix to that of domestic production. If Mixc1 is greater than one, then
country c’s import composition is more emissions-intensive than its domestic production.
Panel I in table 4 shows the results from calculation of equation (8). Surprisingly
(but consistent with Levinson, 2009), the United States' import product mix is cleaner
than its domestic production. There is in fact no evidence from equation (8) that
developed countries disproportionately import emissions-intensive composites of
products, as the average Mixc1 score of the 15 countries with the cleanest domestic
technologies (lowest θ̂c in table 3) is nearly identical to the average of the 15 dirtiest
countries (1.045 for the former and 1.049 for the latter). Thus, if there is a systematic
tendency for developed countries to “export pollution” to less developed countries, it
apparently does not operate through the composition effect.
Do cleaner countries import preferentially from countries that employ dirtier
techniques? To investigate this question, we calculate the weighted average of the
emissions efficiencies '2̂exp( )cθ of production techniques used by a country's trading
partners,
(9) ' '2 2'
ˆexp( )c cc c
ccMix mθ≠
= ⋅∑ ,
where c is the subject (importing) country, c' is the index of its trading partners, and '2̂cθ
is measured using equation (2). If country c imports mainly from emissions-intensive
17
countries, we expect a larger number for Mixc2. Since we normalize '
2̂exp( )cθ to the
United States, Mixc2=1 indicates that the countries from which country c obtains its
imports are on average as “clean” as the United States.
Results from calculations using equation (9) are presented in panel II of table 4.
European countries have the cleanest trading partners (the top 9 countries are from
European Union), a result that can be explained in part by the tendency of trade to
increase with proximity (see e.g. Bergstrand, 1989). Asian, North American, and Pacific
countries, on the other hand, tend to import from countries using more emissions-
intensive techniques. In particular, Japan and the United States import heavily from
emissions-intensive developing countries, even though the emissions-intensity of the
products they import is low relative to domestic production.
Predictions of Emissions Content of Trade from the HOV and DFS Models
We use standard test procedures to check the performance of trade tests for the HOV
and DFS models, as in Bowen, Leamer, and Sveikauskas (1987), Trefler (1995), and
Davis and Weinstein (2001). First, a sign test obtains the probability of sign coincidences
between measured emissions content of trade, Fcm1 from equation (5) and Fc
m2 in
equation (7), and predicted emissions content of trade, Fcp in equation (6). If the
specification held perfectly, the sign coincidence would be 100 percent. A slope test
regresses measured emissions content of trade on predicted emissions content of trade
without an intercept. If the HOV specification held without error, the regression
coefficient would be unity. Variance ratios are computed for each factor by dividing the
variance of measured emissions content of trade by the variance of predictions; again, a
successful model would yield a value of unity.
Performance of the standard HOV model in the diagnostic tests is spotty. It yields a
reasonable sign fit of 71.9 percent, but its slope coefficient of 0.040 and its variance ratio
of 0.064 are low. As indicated in figure 4-1, the United States imports significant
emissions (302 million tons), though less than the model predicts (658 million tons), and
China exports 270 million tons of emissions, which is far less than the model predicts
(1,486 million tons). Despite significant over-predictions for China and the United States,
the standard HOV model predicts some countries quite well. For example, India exports
18
a large amount of emissions (54 million tons), which is predicted relatively precisely (72
million tons), and Japan imports 86 million tons that matches with the standard HOV
prediction (136 million tons).
The DFS measure from equation (7) performs much better than the standard HOV in
the diagnostic tests. For the DFS specification, the proportion of correct signs rises
sharply to 84.4 percent, the slope of the regression line is much closer to one (0.645), and
the trade variance ratio (0.202) increases significantly from 0.064 in the standard HOV
model. Figure 4-2 shows that most of the developed countries, including France,
Germany, Japan, the United Kingdom, and the United States, are all measured to be
significant importers of emissions, and most of the developing countries (the Czech
Republic, China, India, Korea, Poland, the Slovak Republic, and Turkey) are both
measured and predicted to be exporters of emissions. Thus, the direction and volume of
the emissions trade is well predicted from the HOV model once it is modified according
to the DFS specification to allow international differences in emissions techniques.
5) Concluding Remarks
All parties to the negotiations on international policies to address anthropogenic
global climate change are sensitive to the distributional impacts of those policies,
particularly on their own countries. Economists have at their disposal some analytical
tools that are useful for assessing those impacts. In particular, we have used the tools of
empirical international trade economics to address some basic questions of whose
production and trade are most emissions-intensive, and why. The answers are important
to the debate because an increase in the price of emissions will disproportionately affect
countries whose production and trade rely upon differences in emissions-intensity. The
issue has a special resonance because there is a negative correlation between a country’s
emissions-intensity and its level of development.
The belief that poorer developing countries gain comparative advantage from their
high tolerance for pollution, and are therefore more likely to have emissions-intensive
economies is sometimes referred to as the “pollution haven hypothesis.” Our empirical
findings are consistent with a version of the pollution haven hypothesis in which
countries import more emissions (embodied in goods and services) as they develop. Our
19
findings are not, however, consistent with the usual formulation of the pollution haven
hypothesis, in which the reason for the greater emissions-intensity of developing
countries is the outsourcing of pollution-intensive industries by more developed countries.
That is, our results tend to rule out the hypothesis that developing countries are more
pollution-intensive because they specialize in dirtier industries (i.e., the “composition
effect”).
Instead, our results suggest that technology choice drives the greater emissions
intensity of industry in developing countries. Both the HOV and the DFS models allow
countries to specialize in specific industrial sectors, so if the increased emissions-
intensity of lower- and middle-income countries were a result of specialization in
emissions-intensive industries then the HOV model would be able to predict emissions-
intensity as well as the DFS model does. Because only the DFS model allows for
differences in techniques across countries, and its predictions are both different and more
accurate than the HOV model’s predictions, our empirical research suggests that
differences in emissions intensity must be attributed to differences in country-level
technology choice. A toy factory in France, for example, is likely to be cleaner than a toy
factory in the Czech Republic, because the French use cleaner techniques in production.
Our finding that cross-country differences in emissions intensity are driven primarily
by technology choices suggests that policy makers should emphasize technology transfer
as a tool in addressing global climate change. There is hope that policies that encourage
development of cleaner technologies (presumably in developed countries) and their
implementation worldwide might be able to reduce the overall costs of addressing global
climate change. Further research and practical experience are needed to quantify both the
costs of technological adjustment and the advisability of government intervention to
foster that adjustment, but transferring technology from developed to developing
countries should be less costly and less disruptive to developing economies than changing
the industries in which they specialize.
20
Data Appendix A: Allocation of Emissions Among Industries
The sectors included in the IPCC-CRF are energy (93% of world CO2 emissions in
2000); industrial processes (3.5%), agriculture (0%), waste (0%), and international
bunkers (3.5%).15 Since by far the largest share of CO2 emissions come from energy,
and our interest is in industrial emissions, we concentrate on these two sectors. We
denote country c’s total emissions from energy as ece and those from industrial processes
as ecp. Energy-related CO2 emissions (ec
e) are divided in the UNFCCC reporting
framework into five sub-sectors: electricity and heat (ece1), responsible for 42% of world
CO2 emissions in 2000; manufacturing and construction (ece2), responsible for 18%; other
fuel combustion (ece3), 13%; fugitive emissions (ec
e4), less than 1%; and transportation
(ece5), 20%.16
We allocate emissions from these five energy emissions sub-sectors to our 48
industries using coefficients taken from the input-output tables, as follows. First, the
electricity and heat (ece1) sector includes emissions from electricity producers,
cogeneration plants, and plants whose primary objective is to supply heat for the public.
Since the electricity generation and distribution sector merely supplies the demand for
energy by businesses and households, we allocate the CO2 emissions from the electricity
and heat (ece1) subsector into each industry in proportion to its intermediate spending on
electricity.17 We use the formula
(A-1) ( )1 1 1 1 1/c c c c cz e z zz
e e M F M= ⋅ + ∑
15
These figures are exclusive of land-use change and forestry, which constituted 24% of year 2000 CO2 emissions.
16 Emissions from transportation are excluded from our analysis since country size, geography, composition of transportation and business practices make it difficult to allocate transportation emissions consistently across countries. Emissions from fuel sold to any air or marine vessel engaged in international transport (international bunkers) is excluded from the transportation category (ec
e5) and reported separately. 17 We exclude intermediate spending under “steam and hot water supply” because data for this sector are
not available for most countries.
21
where Mc1z is industry z’s intermediate purchases under “production, collection and
distribution of electricity” in the I-O table and F1c is the final consumption of electricity
by households in country c.
We also use input-output data to allocate CO2 from the manufacturing and
construction subsectors (ece2) into industries. For example, a typical steel production
plant uses iron ore and coal as its main raw materials. Its production process generates
certain gases as by-products that are used as fuel for furnaces or power generation plants
on the premises.18 Even though these by-product gases are fuels similar or identical to
those used in the energy and heat subsector (ece1), the UNFCCC reports emissions
intentionally generated from fossil fuel combustion for energy and heat in the sub-sector
of manufacturing and construction (ece2). We allocate these emissions to our industrial
sectors in proportion to sector purchases of “mining (energy).”19
(A-2) ( )2
2 2 2 2/c c c cz e z zz N
e e M M∈
= ⋅ ∑
where Mc2z is industry z’s intermediate demands for “mining (energy)” and N2 represents
the manufacturing and construction sectors.
Third, we allocate CO2 emissions from “other fuel combustion” (ece3) into each
industry according to its purchases of “coke and refined petroleum products.” We use the
equation
(A-3) ( )3
3 3 3 3 3/c c c c cz e z zz N
e e M F M∈
= ⋅ + ∑
where Mc3z is the intermediate purchases of “coke and refined petroleum products” for
industry z in country c, Fc3 is that for households, and N3 represents industries except
manufacturing and construction sectors.
18
Strictly speaking, some types of secondary emissions (e.g., fuel combustion in coke ovens) are included in “energy and heat.”
19 The allocation of these emissions from manufacturing and construction could be associated with its spending on “coke and refined petroleum products.” Since the spending on “coke and refined petroleum products” is much greater than that on “mining (energy),” the aggregation of these two sectors’ spending reflects strongly the across-industry variation in “coke and refined petroleum products.” As a robustness check, we allocate equation (A-2) by the sum of the two sectors. Results were substantially the same.
22
Next, we allocate emissions from fugitive emissions (ece4), which comprise gases
from leakage or flaring released (intentionally or not) without producing useful energy.
They arise primarily from coal mining and oil and gas production and distribution
operations, and include very little CO2. Since this sector is especially related to coal and
oil production and refining, we allocate fugitive emissions into three sectors, “mining
(energy),” “coke and refined petroleum products,” and “manufacture of gas” according to
their industry outputs.
(A-4) ( )4
4 4 /c c c cz e z zz N
e e Q Q∈
= ⋅ ∑
where Qcz is the industry output of sector z and N4 represents the subset of two sectors.
Finally, emissions from “industrial processes” (ecp) consist of by-product emissions
from industrial processes. Emissions from fuel combustion in an industry are generally
reported under the energy sector where possible. However, if industrial process
emissions result jointly from chemical processes and fuel combustion, as is common in
the cement, limestone, chemicals, pulp and paper, and food industries, it is difficult to
assign these emissions. Since this emissions sector is especially related to the process of
coal and petroleum products, we allocate emissions from industrial processes (ecp)
according to each industry’s procurement of “coke and refined petroleum products:”
(A-5) ( )5
5 5 /c c c cz p z zz N
e e M M∈
= ⋅ ∑ 5
z
where Mc5z is the intermediate purchase of “coke and refined petroleum products” for
industry z in country c and N5 represents the sectors associated with industrial processes.
After allocating the CO2 emissions from each sector into each industry z, we obtain
industry z’s total emissions by summing the five sources of industrial CO2 emissions:
(A-6) 1 2 3 4 5c c c c c cz z z z ze e e e e e= + + + + .
Empirical exercises below were conducted using emissions estimates obtained from
equations (A-1) and (A-6). Since the results do not change much between these two, we
report only the results from equation (A-6). Results from equation (A-1) are available
upon request.
23
Data Appendix B: Data Development of National Accounts
(1) Input-Output Data
Input-output (I-O) tables (total use) for Argentina, Australia, Austria, Belgium,
Brazil, Canada, China, Czech Republic, Denmark, Finland, France, Germany, Greece,
Hungary, India, Indonesia, Ireland, Italy, Japan, Korea, the Netherlands, New Zealand,
Norway, Poland, Portugal, Slovak Republic, Spain, Sweden, the Switzerland, Turkey, the
United Kingdom, and the United States for year 2000 are taken from the OECD Input-
Output Database (2007). These I-O tables employ the ISIC Rev.3 with 48 industrial
groups (Table A).
Input-output matrices, final consumptions, gross outputs, exports, and imports are
drawn from the I-O tables. Final consumption is the sum of final consumption of
households, final consumption and investment of government, gross fixed capital
formation, and changes in inventory. Therefore, the total use table of country c satisfies
the equation Tc=(I-Bc)Qc-Dc where Bc is a 48×48 indirect techniques for the unit
intermediate requirements and (I-Bc)Qc vector equals net output (Yc) by construction. Bc
is obtained by taking input-output data from the I-O tables and dividing inputs in each
sector by the corresponding sector’s gross output.
To convert the dataset into 2000 international dollars, we use country-level PPP rates
from the Penn World Table (PWT) 6.2 (Heston, Summers, and Aten, 2006).
Unfortunately, industry-level PPP rates are not available. This might conceal some of the
cross-industry heterogeneity in techniques for each country. For Argentina, Australia,
India, Ireland, New Zealand, Norway, Portugal, Switzerland, and Turkey, nominal values
in the I-O tables are uniformly multiplied by the growth rates of total nominal GDP to
adjust data from earlier or later years to the year 2000.
(2) Industry-Level Data for Factor Inputs
Physical Capital
We first develop country-total capital stock by using the perpetual method from real
gross fixed capital formation in local currencies (GFCFct) from 1985 to 2002. Then, we
allocate these values to each industry according to the compensation for capital (gross
24
operating surplus) obtained from the I-O tables. This procedure is based on the idea that
industry capital compensation flows are proportional to industry capital stocks (e.g., Lai
and Zhu, 2007). To convert GFCF figures into international dollars, we convert real
values of local currency into international dollars by using the price of investment and
nominal exchange rate for year 2000 from the Penn World Table 6.2.
Labor
Sectoral labor inputs (total employment) for the year 2000 are available from the
OECD STAN databases (2005) and the ILO LABORSTA Internet Yearly Statistics for
most of the countries. However, since these databases do not provide the data for all the
countries in this paper, we rather make the most of the I-O tables to obtain industry level
employment data. We first derive the total employment from the World Bank
Development Indicators (2005) and allocate these values into each industry according to
the labor compensations from the I-O tables.
Industry-Level Bilateral Trade
Bilateral trade flows for manufacturing from each of 27 countries and each of all 32
countries and the rest of the world are available from the OECD STAN Bilateral Trade
Database (2006). Bilateral trades for six countries, Argentina, Brazil, China, India, and
Indonesia, are developed from the data from the World Bank Trade, Production, and
Protection (1976-2004). We scale these bilateral trade flows so that bilateral industry
export totals match those from the I-O tables. Because there is no bilateral trade data
available for service industries, we allocate the total service exports for each industry
derived from the I-O tables into each of 32 countries by the share of total manufacturing
exports. In addition, the World Bank Trade, Production, and Protection database does
not report the bilateral trade data for agriculture and mining sectors. Therefore, the
bilateral exports of these two sectors are estimated from the bilateral exports of total
manufacturing for Argentina, Brazil, China, India, and Indonesia.
25
Table. A: List of Industries
name NN (=1) N2 (=1) N3 (=1) N4 (=1) N5 (=1)1 Agriculture 1 0 1 0 02 Mining (energy) 1 0 1 1 03 Mining (non-energy) 1 0 1 0 04 Food products 1 1 0 0 15 Textiles 1 1 0 0 06 Wood 1 1 0 0 07 Pulp and paper products 1 1 0 0 18 Coke and refined petroleum products 1 0 0 1 19 Chemicals 1 1 0 0 1
10 Pharmaceuticals 1 1 0 0 111 Rubber & plastics products 1 1 0 0 112 Non-metallic mineral products 1 1 0 0 113 Iron & steel 1 1 0 0 114 Non-ferrous metals 1 1 0 0 115 Fabricated metal products 1 1 0 0 116 Machinery & equipment 1 1 0 0 017 Office & computing machinery 1 1 0 0 018 Electrical machinery 1 1 0 0 019 Communication equipment 1 1 0 0 020 Medical instruments 1 1 0 0 021 Motor vehicles 1 1 0 0 022 Ships & boats 1 1 0 0 023 Aircraft & spacecraft 1 1 0 0 024 Railroad equipment 1 1 0 0 025 Other manufacturing 1 1 0 0 026 Electricity 0 0 1 0 027 Gas 0 0 1 1 028 Steam and hot water supply 0 0 1 0 029 Water 0 0 1 0 030 Construction 0 1 0 0 031 Wholesale & retail trade 0 0 1 0 032 Hotels & restaurants 0 0 1 0 033 Land transportation 0 0 1 0 034 Water transportation 0 0 1 0 035 Air transportation 0 0 1 0 036 Activities of travel agencies 0 0 1 0 037 Post & telecommunications 0 0 1 0 038 Finance & insurance 0 0 1 0 039 Real estate activities 0 0 1 0 040 Renting of machinery & equipment 0 0 1 0 041 Computer & related activities 0 0 1 0 042 Research & development 0 0 1 0 043 Other Business Activities 0 0 1 0 044 Public administration 0 0 1 0 045 Education 0 0 1 0 046 Health & social work 0 0 1 0 047 Other community services 0 0 1 0 048 Private households 0 0 1 0 0
26
References
Ahmad, Nadim, and Andrew Wyckoff. 2003. “Carbon Dioxide Emissions Embodied in
International Trade of Goods” OECD Directorate for Science, Technology, and
Industry Working Paper DSTI/DOC(2003)15. Available online at
http://www.olis.oecd.org/olis/2003doc.nsf/
Antweiler, Werner, Brian R. Copeland, and M. Scott Taylor. 2001. “Is Free Trade Good
for the Environment?” American Economic Review 91(4): 877-908.
Bergstrand, Jeffrey H. 1989. “The Generalized Gravity Equation, Monopolistic
Competition, and the Factor-Proportions Theory in International Trade,” Review
of Economics and Statistics 71(1): 143-53.
Bertinelli, Luisito and Eric Strobl. 2005. “The Environmental Kuznets Curve Semi-
Parametrically Revisited,” Economics Letters 88: 350-7.
Bowen, Harry P., Edward E. Leamer, and Leo Sveikauskas. 1987. “Multicountry,
Multifactor Tests of the Factor Abundance Theory,” American Economic Review
77(5): 791-809.
Caves, Douglas W., Laurits R. Christensen, and W Erwin. Diewert. 1982. “Multilateral
Comparisons of Output, Input, and Productivity Using Superlative Index
Numbers,” Economic Journal 92: 73-86.
Cole Matthew A. and Robert J.R. Elliott. 2003. “Determining the Trade-Environment
Composition Effect: the Role of Capital, Labor and Environmental Regulations,”
Journal of Environmental Economics and Management 46: 363-83.
Copeland, Brian R. and M. Scott Taylor. 1995. “Trade and Transboundary Pollution,”
American Economic Review 85(4): 716-37.
Davis, Donald R. and David E. Weinstein. 2001. “An Account of Global Factor Trade,”
American Economic Review 91(5): 1423-53.
Dinda, Soumyananda. 2004. “Environmental Kuznets Curve Hypothesis: A Survey,”
Ecological Economics 49: 431-55.
27
Dornbusch, Rudiger, Stanley Fischer, and Paul A. Samuelson. 1980. “Heckscher-Ohlin
Trade Theory with a Continuum of Goods,” Quarterly Journal of Economics
95(2): 203-24.
Energy Information Administration (EIA). 2008. International Energy Annual 2006.
Posted June-December 2008. Available online at http://www.eia.doe.gov/iea/.
Eskeland, Gunnar S. and Ann E. Harrison. 2003. “Moving to Greener Pastures?
Multinationals and the Pollution Haven Hypothesis,” Journal of Development
Economics 70: 1-23.
Grossman, Gene M. and Alan B. Krueger. 1995. “Economic Growth and the
Environment,” Quarterly Journal of Economics 110(2): 353-77.
Heston, Alan, Robert Summers, and Bettina Aten. 2006. “Penn World Table Version
6.2,” Center for International Comparisons of Production, Income and Prices at
the University of Pennsylvania.
Harbaugh, William T., Arik Levinson, and David Molly Wilson. 2002. “Reexamining the
Empirical Evidence for an Environmental Kuznets Curve,” Review of Economics
and Statistics 84 (3): 541-51.
Hass, Julie L., Hunt O. Sorensen, and Kristine Erlandsen. 2002. “Norwegian Economic
and Environment Accounts (NOREEA) Project Report – 2001,” Statistics
Norway.
Keller, Wolfgang, and Arik Levinson. 2002. “Pollution Abatement Costs and Foreign
Direct Investment Inflows to U.S. States,” Review of Economics and Statistics
84(4): 691-703.
Lai, Huiwen and Susan Chun Zhu. 2007. “Technology, Endowments, and the Factor
Content of Bilateral Trade,” Journal of International Economics 71: 389-409.
Levinson, Arik. 2009. “Technology, International Trade, and Pollution from U.S.
Manufacturing,” American Economic Review, forthcoming.
Levinson, Arik and M. Scott Taylor. 2008. “Unmasking the Pollution Haven Effect,”
International Economic Review 49 (1): 223-54.
28
Lucas, Robert, David Wheeler, and Hemamala Hettige, “Economic development,
environmental regulation and the international migration of toxic industrial
pollution: 1960-1988,” in P. Low (Ed.), International Trade and the Environment,
World Bank Discussion Paper No. 159 (Washington, DC: World Bank, 1992),pp.
67-86.
Ministry of the Environment, and Ministry of Economy, Trade and Industry. 2009. “2006
Report on Greenhouse Gases Emissions according to the Law Concerning the
Promotion of the Measures to Cope with Global Warming” (in Japanese).
Maskus, Keith E. and Shuichiro Nishioka. 2009. “Development-Related Biases in Factor
Productivities and the HOV Model of Trade,” Canadian Journal of Economics
42 (2): 519-53.
Roberts, J. Timmons and Peter E. Grimmes. 1997. “Carbon Intensity and Economic
Development 1962-91: A Brief Explanation of the Environment Kuznets Curve,”
World Development 25(2): 191-98.
Stern, David I. 2004. “The Rise and Fall of the Environmental Kuznets Curve,” World
Development 32 (8): 1419-39.
Trefler, Daniel. 1993. “International Factor Price Differences: Leontief was Right!”
Journal of Political Economy 101(6): 961-87.
Trefler, Daniel. 1995. “The Case of the Missing Trade and Other Mysteries,” American
Economic Review 85(5): 1029-46.
Turner, Karen, Manfred Lenzen, Thomas Wiedmann, and John Barrett. 2007.
“Examining the Global Environmental Impact of Regional Consumption
Activities – Part 1: A Technical Note on Combining Input-Output and Ecological
Footprint Analysis,” Ecological Economics 62: 37-44.
Wiedmann, Thomas, Manfred Lenzen, Karen Turner, and John Barrett. 2007.
“Examining the Global Environmental Impact of Regional Consumption
Activities – Part 2: Review of Input-Output Models for the Assessment of
Environmental Impacts Embodied in Trade,” Ecological Economics 61: 15-26.
29
Tables and Figures
Figure 1-1. Actual and Predicted Techniques for Japan (year 2006)
0
2
4
6
8
10
12ag
ricul
ture
and
fish
ing
min
ing
and
quar
ryin
g
food
, bev
erag
es, a
nd to
bacc
o
text
iles p
rodu
cts
woo
d an
d pr
oduc
ts of
woo
d
pulp
, pap
er a
nd p
aper
pro
duct
s
prin
ting
and
publ
ishin
g
gene
ral c
ham
ical
pro
duct
s
plas
tic p
rodu
cts
prod
ucts
of c
oal a
nd p
trole
um
kiln
pro
duct
s
iron
and
steel
non-
met
alic
pro
duct
s
met
alic
pro
duct
s
gene
ral m
achi
ne
cons
umer
s' el
ectro
nic
mac
hine
acco
untin
g an
d su
bsid
iary
mac
hine
gene
ral m
achi
ne
gene
ral m
otor
veh
icle
s
prec
ision
instr
umen
t
othe
r ind
ustri
al p
rodu
cts
cons
truct
ion
and
repa
ir
elec
trici
ty p
ower
gas a
nd h
eat s
uppl
y
wat
er a
nd g
arba
ge d
ispos
al
com
mer
ce
finan
ce, i
nsur
ance
, and
real
esta
te
trans
ports
com
mun
icat
ion
and
broa
dcas
t
offic
ial d
utie
s
publ
ic se
rvic
e
othe
pub
lic se
rvic
es
rese
arch
and
dev
elop
men
t
othe
r bus
ines
s ser
vice
s
pers
onal
serv
ices
othe
rs se
rvic
e ac
tiviti
es
Emis
sion
s-In
tens
ity in
Pro
duct
ion
ActualPredicted
Figure 1-2. Actual and Predicted Techniques for Norway (year 2001)
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
agric
ultu
re a
nd fi
shin
g
min
ing
and
quar
ryin
g
food
pro
duct
s, be
vera
ges a
nd to
bacc
o
text
iles,
leat
her a
nd fo
otw
ear
woo
d an
d pr
oduc
ts of
woo
d an
d co
rk
pulp
, pap
er, p
rintin
g, a
nd p
ublis
hing
coke
and
refin
ed p
etro
leum
pro
duct
s
chem
ical
s exc
ludi
ng p
harm
a ce
utic
als
iron,
stee
l, an
d m
etal
pro
duct
s
mac
hine
ry &
equ
ipm
ent
trans
port
equi
pmen
ts
othe
r man
ufac
turin
g
elec
trici
ty
colle
ctio
n an
d di
strib
utio
n of
wat
er
cons
truct
ion
who
lesa
le &
reta
il tra
de
hote
ls &
resta
uran
ts
land
tran
spor
t
wat
er tr
ansp
ort
air t
rans
port
supp
ortin
g ac
tiviti
es fo
r tra
nspo
rt
post
& te
leco
mm
unic
atio
ns
finan
ce, i
nsur
ance
, and
real
esta
te
othe
r bus
ines
s act
iviti
es
publ
ic a
dmin
. & d
efen
ce
educ
atio
n
heal
th &
soci
al w
ork
othe
r com
mun
ity se
rvic
es
Emis
sion
s-In
tens
ity in
Pro
duct
ion
ActualPredicted
30
Table 1. Summary of the Statistics (Year 2000)
CO2 Emissions from WRI Climate Analysis Indicators Emissions Estimates (exclude: households and transport) Development Indicators
1. Energy 2. Industrial Process Total (1.+2.) Energy Emissions from Equation (A-1) Industry Emissions from Equation (A-6) GDP/L TFP K/L
(kt) (%, share) (kt) (%, share) (kt) (%, share) (kt) (% to total) (%, share) (kt) (% to total) (%, share) (Int'l $) (U.S.=1) (Int'l $)
Argentina 135400 0.8 3000 0.5 138400 0.8 31426 22.7 0.6 73310 53.0 0.7 11332 0.517 41331
Australia 340000 2.0 3700 0.6 343700 2.0 139038 40.5 2.6 204719 59.6 1.9 25835 0.682 74871
Austria 63700 0.4 1900 0.3 65600 0.4 14625 22.3 0.3 37258 56.8 0.3 27000 0.751 91386
Belgium 119000 0.7 3600 0.6 122600 0.7 21561 17.6 0.4 78556 64.1 0.7 24662 0.791 84266
Brazil 308900 1.8 19500 3.1 328400 1.9 36273 11.0 0.7 186030 56.6 1.7 7194 0.478 15720
Canada 534500 3.2 6300 1.0 540800 3.1 118984 22.0 2.2 286738 53.0 2.7 26821 0.751 74949
China 3037700 18.0 297500 47.1 3335200 19.0 1331653 39.9 25.1 2956554 88.6 27.4 4002 0.278 6363
Czech Republic 118100 0.7 2000 0.3 120100 0.7 47603 39.6 0.9 84632 70.5 0.8 13617 0.508 32522
Denmark 50600 0.3 1000 0.2 51600 0.3 14769 28.6 0.3 25844 50.1 0.2 27827 0.694 73340
Finland 54100 0.3 700 0.1 54800 0.3 21067 38.4 0.4 37730 68.9 0.3 22741 0.643 67171
France 379300 2.2 10000 1.6 389300 2.2 38673 9.9 0.7 175722 45.1 1.6 25045 0.742 83702
Germany 833200 4.9 17600 2.8 850800 4.9 225616 26.5 4.3 454897 53.5 4.2 25061 0.702 79039
Greece 87800 0.5 7700 1.2 95500 0.5 28012 29.3 0.5 52738 55.2 0.5 13982 0.558 44200
Hungary 55500 0.3 1700 0.3 57200 0.3 15220 26.6 0.3 35353 61.8 0.3 11383 0.504 28082
India 972800 5.8 47300 7.5 1020100 5.8 510186 50.0 9.6 848451 83.2 7.9 2644 0.290 3708
Indonesia 277400 1.6 13800 2.2 291200 1.7 62922 21.6 1.2 182598 62.7 1.7 3772 0.436 9214
Ireland 41400 0.2 1300 0.2 42700 0.2 9255 21.7 0.2 22231 52.1 0.2 24948 0.921 56282
Italy 425800 2.5 19400 3.1 445200 2.5 111404 25.0 2.1 258143 58.0 2.4 22487 0.710 71281
Japan 1172200 6.9 40400 6.4 1212600 6.9 332056 27.4 6.3 742954 61.3 6.9 23971 0.576 95580
Korea 424800 2.5 25500 4.0 450300 2.6 126661 28.1 2.4 307617 68.3 2.8 15702 0.484 62616
Netherlands 173700 1.0 1700 0.3 175400 1.0 46333 26.4 0.9 108288 61.7 1.0 26293 0.768 73644
NZ 32400 0.2 500 0.1 32900 0.2 7428 22.6 0.1 16940 51.5 0.2 20423 0.643 53782
Norway 35400 0.2 900 0.1 36300 0.2 7321 20.2 0.1 20170 55.6 0.2 33092 0.767 96954
Poland 292900 1.7 7500 1.2 300400 1.7 136258 45.4 2.6 223388 74.4 2.1 8611 0.434 19245
Portugal 60000 0.4 5200 0.8 65200 0.4 18405 28.2 0.3 39855 61.1 0.4 17323 0.582 47998
Slovak Republic 37500 0.2 1500 0.2 39000 0.2 11459 29.4 0.2 28420 72.9 0.3 9697 0.410 25613
Spain 285600 1.7 19000 3.0 304600 1.7 79382 26.1 1.5 170914 56.1 1.6 19536 0.674 60787
Sweden 53600 0.3 1300 0.2 54900 0.3 5605 10.2 0.1 25813 47.0 0.2 25232 0.704 61262
Switzerland 42200 0.2 1900 0.3 44100 0.3 1790 4.1 0.0 18816 42.7 0.2 28831 0.725 102753
Turkey 202800 1.2 17900 2.8 220700 1.3 57786 26.2 1.1 161559 73.2 1.5 5715 0.479 13838
United Kingdom 525200 3.1 6300 1.0 531500 3.0 151635 28.5 2.9 287851 54.2 2.7 24666 0.758 59820
United States 5724300 33.9 44600 7.1 5768900 32.9 1541514 26.7 29.1 2653742 46.0 24.6 34365 1.000 90872
Total 16897800 100.0 632200 100.0 17530000 100.0 5301919 30.2 100.0 10807831 61.7 100.0 - - -
31
Figure 2-1. Emission Intensity across Countries (TFP)
Argentina
Australia
Austria
Belgium
Brazil
Canada
ChinaCzech Republic
Denmark
Finland
France
Germany
Greece
Hungary
India
Indonesia IrelandItaly
Japan
Korea
Netherlands
Norway
Poland
Portugal
Slovak Republic
Spain
SwedenSwitzerland
Turkey
United States
EI = -0.4506TFP + 0.6332R2 = 0.2275
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0 0.2 0.4 0.6 0.8 1.0 1.2
TFP: U.S.=1
Emiss
ions
-Int
ensit
y (C
O2
Emiss
ions
/GD
P)
Figure 2-2. Emission Intensity across Countries (Capital Intensity)
Argentina
Australia
Austria
Belgium
Brazil
Canada
ChinaCzech Republic
Denmark
Finland
France
Greece
Hungary
India
Indonesia Italy
Japan
Korea
Netherlands
Norway
Poland
Portugal
Slovak Republic
Spain
SwedenSwitzerland
Turkey
United Kingdom
United States
EI= -0.0574(ln(K/L))2 + 0.2997ln(K/L) + 0.0815R2 = 0.2118
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
log(Capital/labor)
Emiss
ions
-Int
ensit
y (C
O2
Emiss
ions
/GD
P)
32
Table 2. Estimation Results of Emissions Techniques (Weighted Least Squares)
Equation (1) Equation (2) Equation (3)coef. t-statistics coef. t-statistics coef. t-statistics
Argentina 0.057 0.604 0.067 0.698Australia 0.751 7.979 0.747 7.875Austria -0.155 -1.615 -0.161 -1.669Belgium 0.244 2.568 0.224 2.331Brazil -0.322 -3.580 -0.331 -3.551Canada 0.267 2.901 0.274 2.957China 0.414 4.784 0.393 4.009Czech Republic 0.648 6.743 0.624 6.231Denmark -0.374 -3.736 -0.384 -3.796Finland -0.024 -0.245 -0.041 -0.406France -0.537 -5.948 -0.550 -6.028Germany -0.213 -2.392 -0.229 -2.525Greece 0.674 6.760 0.672 6.262Hungary 0.071 0.723 0.027 0.260India 0.179 2.023 0.157 1.461Indonesia -0.233 -2.539 -0.250 -2.418Ireland -0.796 -7.747 -0.819 -7.170Italy -0.163 -1.806 -0.183 -1.976Japan 0.006 0.071 0.014 0.152Korea 0.349 3.856 0.348 3.827Netherlands -0.266 -2.813 -0.270 -2.836NZ 0.123 1.213 0.104 1.005Norway -0.453 -4.585 -0.446 -4.455Poland 0.760 7.995 0.726 7.107Portugal 0.487 4.982 0.469 4.694Slovak Republic 0.437 4.353 0.401 3.838Spain 0.131 1.433 0.115 1.233Sweden -0.609 -6.348 -0.610 -6.285Switzerland -1.157 -11.514 -1.170 -11.490Turkey 0.508 5.444 0.486 4.774United Kingdom -0.412 -4.552 -0.431 -4.613log(K/L) 0.083 1.486(log(K/L))̂ (2) -0.013 -1.828Obervations (T) 1348 1348 1312Parameters (k) 48 79 81Adjusted R-squared 0.520 0.747 0.749-Log L (LL) -1614 -1165 -1129SIC 2.652 2.151 2.164AIC 2.466 1.846 1.844
Note 1: t-statistics are based on H0: θc = 0. Note 2: Equation (2) assumes no diversity in efficiency across countries, so the restriction θc = 0 for all countries. Note 3: LR tests reject models (2) and (3) at α = .001 or better.
33
Table 3. Industries and Countries with Extreme Emission Intensities
I. Country-Level Average Intensity (U.S.=1)10 Most Emission-Intensive Countries Intensity 10 Least Emission-Intensive Countries Intensity
Poland 2.139 Switzerland 0.314Australia 2.120 Ireland 0.451Greece 1.963 Sweden 0.544Czech Republic 1.912 France 0.584Turkey 1.663 Norway 0.636Portugal 1.628 United Kingdom 0.662Slovak Republic 1.548 Denmark 0.688China 1.512 Brazil 0.725Korea, Republic of 1.417 Netherlands 0.767Canada 1.305 Indonesia 0.792
II. Industry-Specific Intensity10 Most Emission-Intensive Industries Intensity 10 Least Emission-Intensive Industries Intensity
Iron & steel 1.731 Real estate activities 0.080Chemicals 1.057 Finance & insurance 0.091Non-metallic mineral products 0.944 Education 0.126Electricity 0.680 Computer & related activities 0.138Fabricated metal products 0.664 Private households 0.143Non-ferrous metals 0.654 Renting of machinery & equipment 0.143Steam and hot water supply 0.615 Post & telecommunications 0.149Rubber & plastics products 0.581 Other Business Activities 0.149Air transportation 0.509 Public administration 0.165Electrical machinery 0.480 Health & social work 0.174
Table 4. Composition Effects by Products and Countries
I. Average Emission-Intensity of Import Mix Relative to Domestic Production Mix10 Countries with Most Emissions-Intensive Imports Intensity Ratio 10 Countries with Least Emissions-Intensive Imports Intensity Ratio
Indonesia 1.351 Japan 0.751Norway 1.325 Korea 0.852Turkey 1.200 Slovak Republic 0.860Switzerland 1.178 United States 0.902New Zealand 1.174 Ireland 0.902Argentina 1.168 Belgium 0.920Greece 1.147 Brazil 0.941India 1.122 Netherlands 0.947Portugal 1.114 Spain 0.974Denmark 1.101 United Kingdom 0.983
II. Average Emission-Intensity of Imports by Trading Partners 10 Countries whose Trading Partners employ Clean Techniques Average Intensity 10 Countries whose Trading Partners employ Dirty Techniques Average Intensity
Belgium 0.830 New Zealand 1.305Ireland 0.840 Japan 1.155Denmark 0.849 Slovak Republic 1.117Switzerland 0.850 United States 1.107Sweden 0.850 Korea 1.080Spain 0.863 Indonesia 1.064Finland 0.866 China 1.029Norway 0.869 Australia 1.011Italy 0.894 Canada 1.001Argentina 0.894 Brazil 0.966
34
Figure 3-1. Per Capita Pollution Contents of Net Exports with Average Emission Techniques (Heckscher-Ohlin)
United StatesUK
SwitzerlandSpain
Slovak Republic
Portugal
Norway
Netherlands
Korea
Japan
Italy
Ireland
Indonesia
India
Hungary
Greece
Finland
Denmark
Czech Republic
China
Canada
Belgium
Austria
Australia
Argentina
Fcm1/Lc = 529.84TFP - 0.1216R2 = 0.0015
-4000
-2000
0
2000
4000
6000
8000
10000
12000
0.0 0.2 0.4 0.6 0.8 1.0 1.2
TFP (U.S.=1)
Emiss
ions
Tra
de/W
orke
rs
Figure 3-2. Per Capita Pollution Contents of Trade with Producers' Techniques (Dornbusch-Fischer-Samuelson)
Australia
Austria
Belgium
Canada
China
Czech Republic
Finland
Greece
Hungary
India
Ireland
Italy
Japan
Korea
NZ Norway
Poland
Portugal
Slovak Republic
Spain
Sweden
Switzerland
UK
United States
Fcm2= -8507.6TFP + 5201.6R2 = 0.149
-10000
-8000
-6000
-4000
-2000
0
2000
4000
6000
8000
10000
12000
0.0 0.2 0.4 0.6 0.8 1.0 1.2
TFP (U.S.= 1)
Emiss
ions
Tra
de/W
orke
rs
35
Figure 4-1. Measured and Predicted Pollution Contents of Net Exports with Average Emission Techniques (Heckscher-Ohlin)
United States
China
Fcm1 = 0.0399FcpR2 = 0.1577
-1000000
-500000
0
500000
1000000
1500000
2000000
-1000000 -500000 0 500000 1000000 1500000 2000000
Predicted Emission Contents of Trade
Mea
sure
d Em
issio
n Co
nten
ts of
Tra
de
Figure 4-2. Measured and Predicted Pollution Contents of Net Exports with Producers' Techniques (Dornbusch-Fischer-Samuelson)
United States
China
Fcm2 = 0.6449FcpR2 = 0.7125
-1000000
-500000
0
500000
1000000
1500000
2000000
-1000000 -500000 0 500000 1000000 1500000 2000000
Predicted Emission Contents of Trade
Mea
sure
d Em
issio
n Co
nten
ts of
Tra
de
36