NAFTA`s and CUSFTA`s Impact on North American Trade

NAFTA’s and CUSFTA’s Impact on
North American Trade
John Romalis∗- University of Chicago GSB, April 2002 (First Draft:
February 2001).
Abstract
This paper finds that NAFTA and CUSFTA have had a substantial impact on North American trade. The paper focuses on where the US sources its
imports of almost 5,000 different commodities and compares this to where the
European Union (EU) sources its imports of the same commodities. It identifies
the impact of NAFTA using a differences in differences strategy, exploiting the
substantial variation across commodities and time in the US tariff preference
given to goods produced in Canada and Mexico. The paper finds that the recent
rapid growth in Mexico’s share of US trade would have been much slower without NAFTA while Canada’s share may not have increased without CUSFTA.
Useful products of the empirical work are estimates of consumer willingness to
substitute between different varieties of a commodity, an important parameter in welfare analysis of trade liberalization. Estimated average elasticities of
substitution typically range from 4 to 7. Elasticities of this magnitude imply
that even modest trade liberalizations will have a pronounced effect on trade
volumes. One alarming result is that NAFTA may have produced substantial
trade diversion, because the largest tariff preferences are often in industries
where imports from outside North America represent a substantial proportion
of domestic absorption.
∗
I would particularly like to thank my advisors, Daron Acemoglu, Rudi Dornbusch and Jaume
Ventura. Thanks are also due to Mark Aguiar, Christian Broda, Gita Gopinath, Roberto Rigobon,
Alwyn Young and participants at seminars and lunches at Chicago GSB, EIIT Conference 2001,
Federal Reserve Bank of New York, University of Michigan, MIT and University of Pennsylvania.
All errors are my own.
1
1
Introduction
Preferential Trade Areas (PTAs) have received a great deal of analytical and empirical
attention since Viner (1950) distinguished between the trade creationary and trade
diversionary effects of preferential tariff liberalization. Much of this attention is driven
by the ambiguous welfare implications of PTAs. Favorable effects (“trade creation”)
result from removing distortions in the relative price between domestically produced
commodities and commodities produced in other members of the PTA. Unfavorable
effects (“trade diversion”) come from the introduction of distortions between the
relative price of commodities produced by PTA members and non-members (Frankel,
Stein and Wei 1996). Research has also been motivated by the political economy
of PTAs, such as whether PTAs help or hinder movement towards the first best of
global free trade (for example, Baldwin 1996, Levy 1997, Bagwell and Staiger 1999).
Much empirical work has been devoted towards evaluating trade and welfare effects
of PTAs (Baldwin and Venables 1995). This paper extends the empirical literature.
It uses differences in differences estimation on very detailed trade and tariff data
for the US and the EU to estimate the extent to which the world’s second-largest
PTA has affected trade, and to estimate a critical parameter in all analyses of trade
liberalization; the elasticity of substitution between different varieties of a product.
On January 1, 1994 the North American Free Trade Agreement (NAFTA) between
the United States, Canada and Mexico entered into force and incorporated the prior
Canada-US Free Trade Agreement (CUSFTA). NAFTA has been described as the
most comprehensive free trade pact, short of a common market, that has ever been
negotiated between regional trading partners (Hufbauer and Schott, 1993). It is by
far the largest free trade pact outside of the European Union and is the first reciprocal
free trade pact between a substantial developing country and developed economies.
Further expansion is in prospect following the April 2001 Summit of the Americas.
Ministers from almost all North and South American nations have been directed to
negotiate the Free Trade Area of the Americas (FTAA) by January 2005.
Since the advent of NAFTA one of the more striking occurrences has been the
rapid increase in Mexican trade. Mexico has become the US’s second largest trading
partner, accounting for 11.5 percent of US merchandise imports in 2001 and 13.9
percent of US exports, up from 6.9 and 9.0 percent respectively in 1993. Only Canada
is a partner for more US trade. Mexico now accounts for a larger share of US trade
than Korea, Thailand, Singapore, Malaysia, Hong Kong and Taiwan combined.
Empirical studies often have great difficulty in identifying an effect of NAFTA.
For example, Krueger (1999, 2000) does not attribute the increase in Mexican trade
to NAFTA, but to the real depreciation of the Mexican exchange rate in 1994 and
to Mexico’s unilateral reduction of tariffs and quantitative trade restrictions after its
entry into GATT in 1986. By contrast, this paper finds that NAFTA has had a
substantial impact on North American trade. It does so by focusing on where one of
the NAFTA partners, the United States, sources its imports of almost 5,000 6-digit
2
Harmonized System (HS-6) commodities and compares this to the source of European
Union (EU) imports of the same commodities.
Figure 1A shows that Mexico’s share of US imports has increased most rapidly
in commodities for which it has been given the greatest increase in tariff preference,
defined as the difference between the US tariff on a commodity sourced from Mexico
and the US’s Most Favored Nation (MFN) tariff rate for the same commodity.1 For
the 389 commodities where the US tariff preference for Mexican goods has increased
by at least 10 percentage points, the simple average of Mexico’s share of US imports
has risen by 223 percent since 1993. For the 2663 commodities where Mexico’s tariff
preference has not increased, its share has risen by a more modest 22 percent. Was
this NAFTA or was this the result of some other factor such as the exchange rate
or Mexico’s earlier trade liberalization? The timing and cross-commodity pattern of
Mexico’s trade increase are themselves highly suggestive that trade was very responsive to NAFTA’s tariff preferences, and Figure 1B further supports the case. Figure
1B shows Mexico’s share of EU imports from 1989-2000. Without the benefit of a
free trade agreement until late 2000, the evolution of Mexico’s trade with the EU
has been very different. Its share of EU imports of commodites with high NAFTA
preferences declined by 76 percent, while its share of EU imports of commodities
where NAFTA did not increase preferences rises by 82 percent. This paper estimates
that approximately one-third of the post-1993 increase in US imports sourced from
Mexico can be attributed to Mexico’s preferential treatment.
Canada’s share of US imports has also increased since CUSFTA came into effect in 1989, and Figures 2A to 2C also suggest that CUSFTA/NAFTA was partly
responsible. For commodities where there was no increased preference for goods of
Canadian origin, Canadian goods now account for a 4 percent smaller share of US
imports than they did in 1988. But where the preference increased by at least 10
percentage points, Canada’s share of US imports increased by 95 percent. The timing and cross-commodity pattern again suggest that CUSFTA is at work. Figure
2B shows Canada’s share of US imports from 1980 to 2000. For most of the 1980s,
Canada’s share of US imports is declining in all tariff classes, but just before CUSFTA,
Canada’s share begins to rebound for commodities where large tariff preferences were
negotiated. Figure 2C provides a comparison with Canada’s trade with the EU, which
does not have a preferential trade agreement with Canada. For the commodities with
no CUSFTA preferences, Canada’s share of EU imports has increased by 15 percent.
For commodities with high CUSFTA preferences, Canada’s share of EU imports has
declined by 15 percent. Figures 1A to 2C together suggest that NAFTA/CUSFTA
have had a substantial impact on North American trade, and even though US tariffs
are typically low, trade appears to be quite sensitive to even small trade preferences.
Most studies examining the impact of actual PTA’s are either ex-ante simulations
using Applied General Equilibrium (AGE) models or are ex-post studies examining
1
The MFN tariff is the tariff applicable to imports from countries that have normal trade relations
with the US.
3
changes in the direction of aggregate trade between countries or regions following the
introduction of the PTA. AGE models capture key production, demand and trade
barrier details both for the economies that are party to the PTA and for the rest of
the world. They can generate predictions for the welfare, price, output and trade
consequences of trade liberalization. Examples of AGE modelling of NAFTA are
Kehoe and Kehoe (1995), Brown, Deardorff and Stern (1995), Cox (1995) and Sobarzo
(1995). All models predicted welfare gains for NAFTA members, with approximate
gains of 0.1 percent for the US, 0.7 percent for Canada and 5 percent for Mexico.
Examples of ex-post studies that use aggregate trade data for NAFTA are Gould
(1998) and Garces-Diaz (2001). Gould finds that NAFTA has increased US-Mexico
trade, but has had no effect on US-Canada or Mexico-Canada trade. Garces-Diaz
finds that Mexico’s export boom is not attributable to NAFTA.
The most similar papers to this are Clausing (2001) and Fukao, Okubo and Stern
(2001). Clausing was first to exploit tariff variation at the detailed commodity level
using US import data from 1989 to 1994. Clausing finds that US import growth was
related to tariff preferences conferred on Canada and also concludes that NAFTA
was primarily trade creating. Fukao, Okubo and Stern analyze US imports at the
HS 2-digit level for the period 1992-1998. Of the 70 sets of industry regressions they
run, NAFTA tariff preferences had a significant effect on US imports in 15 cases.
Research at an industry level also includes two papers on NAFTA by Krueger (1999,
2000) who studies North American trade patterns at the 3 and 4 digit SIC industry
level. Krueger finds no evidence that NAFTA has had any impact on intra-North
American trade. Head and Ries (1999) study the industry rationalization effects of
tariff reductions and find that on balance, NAFTA has had little net effect on the scale
of Canadian firms. Trefler (2001) finds that Canadian industries that experienced the
largest tariff cuts under NAFTA experienced substantial labor productivity gains,
but a decline in both output and employment. In studies of MERCOSUR, the PTA
formed between Brazil, Argentina, Uruguay and Paraguay, Yeats (1997) finds that
the fastest growth in intra-MERCOSUR trade was in commodities in which members
did not display a comparative advantage, inferred from the lack of exports of these
commodities outside MERCOSUR. This was interpreted as evidence of the trade
diversionary effects of MERCOSUR. Chang and Winters (2000) look to Brazilian
import price data to examine whether preferential tariffs have depressed the prices
of excluded countries’ exports. They find the rather extraordinary result that due to
the tariff preference, Argentinian competition has led to significant and substantial
reductions in American, Japanese, Korean and most other countries’ export prices to
Brazil.
The key differences between this paper and Clausing are the use of a much longer
data series, the use of EU import data to control for unobserved cost shocks, and the
use of an actual tariff schedule that allows the tariff preference to be calculated even
if imports from Canada or Mexico are not observed. Data prior to 1989 suggest that
US import growth was related to NAFTA tariff preferences many years prior to the
4
FTA, so that Clausing’s conclusion that NAFTA was primarily trade creating is put
in doubt. This paper estimates that over 40 percent of the increased Canadian and
Mexican exports to the US are due to trade diversion. Data after 1994 suggest a very
dramatic effect of NAFTA on US trade with Mexico, a result at odds with Krueger
(1999, 2000). I attribute the difference in this paper’s findings from Krueger to two
factors. Two more years of data have become available, but more importantly, this
paper gets very close to the level of commodity detail at which tariffs are set, rather
than at a more aggregate level.2 This allows the use of better tariff data. Much of
the cross—commodity variation in tariff preferences occurs even within quite detailed
industry sectors. Focussing at this detailed level minimizes the loss of variation in
tariff preferences, reduces the problems of aggregating across commodities, and allows
for a greater ability to control for unobserved factors that may be affecting North
American trade.
This paper is organized as follows. Section 2 provides a brief review of NAFTA.
Section 3 introduces a simple model of preferential trade liberalization that is used
to derive the estimating equations. Section 4 describes the data. Section 5 presents
and discusses the empirical results. Section 6 concludes.
2
NAFTA
The Canada-United States Free Trade Agreement (CUSFTA) came in to effect on
January 1, 1989 and provided for the gradual elimination of tariffs and for reductions
in non-tariff barriers to trade. By January 1, 1998, all US and Canadian tariffs on
goods produced in the US and Canada were eliminated, with the exception of overquota tariffs on several hundred agricultural products (primarily sugar, dairy, poultry,
peanuts and cotton). CUSFTA was incorporated into the North American Free Trade
Agreement (NAFTA) on January 1, 1994. NAFTA was designed to increase trade
and investment among the United States, Canada and Mexico. Almost all tariffs
on goods originating in the US, Canada and Mexico will be eliminated by January
1, 2008. NAFTA did not affect the phase-out of tariffs for US-Canada trade under
CUSFTA. Some US tariffs applied to Mexican goods were, however, transitionally
increased by NAFTA. Prior to 1994, Mexico as a developing country was a beneficiary
of the Generalized System of Preferences (GSP). Under the GSP, the US and other
developed countries allow duty-free or concessional access for the output of developing
countries in several thousand HS 8-digit commodities, accounting for just under 10
per cent of Mexican exports to the US in 1993. With NAFTA, the US ceased to
confer GSP benefits on Mexico.
NAFTA covers a much larger amount of trade than any other regional trading arrangement outside of Europe (Baldwin, 1996), and there are prospects for NAFTA’s
2
The reason why I do not go right to the tariff-line level with approximately 10,000 commodities
is that trade data is not harmonized across countries at this level. Most countries’ trade data is
harmonized to the HS 6-digit level.
5
incorporation into a free trade agreement covering the all of the Americas. While
NAFTA is not a “deep” integration like the European Union and the Australia-New
Zealand Closer Economic Relations Trade Agreement, it contains provisions that go
beyond mere removal of tariffs and quantitative trade restrictions, including disciplines on the regulation of investment, transportation and financial services, intellectual property, government purchasing, competition policy, and the temporary entry
of business persons (Hufbauer and Schott, 1993).
3
Theoretical Framework and Empirical Strategy
This paper seeks to exploit the commodity and time variation in the tariff preference
that is afforded to goods originating in NAFTA partners to identify NAFTA’s effect
on North American trade. The paper focuses on where the US and the EU source
their imports of different commodities. It seeks to explain changes in US import
sources using the preference afforded to commodities of Canadian and Mexican origin. The idea is that where Canada and Mexico are afforded no special preference
(where the MFN tariff rate is zero, for instance), NAFTA’s only impact should come
through a general equilibrium effect on factor prices, or through reductions in “border
effects” due to NAFTA provisions that go beyond tariff liberalization. For commodities where NAFTA causes a preference to open up for Canadian and Mexican goods,
the preference should have an additional effect causing US consumers to substitute
towards Canadian and Mexican goods and away from other sources of supply. The
empirical strategy can be derived from a simple model.
A. Model Description
Firms produce commodities under perfectly competitive conditions. Trade is
driven by preference for variety and by commodities being differentiated by country of origin. Countries may impose ad-valorem tariffs on imports. Countries may
then enter into preferential trading agreements whereby each country in the agreement lowers tariffs on imports from partner countries but need not adjust the tariff
on imports from other countries. This causes consumers to substitute towards the
output of preferred countries and away from all other sources of supply, including
domestic production. The model assumptions are set out in detail below.
1. Countries are denoted by c and time by t.
2. There is a continuum of industries z on the interval [0,1]. In each country,
every industry produces a commodity under conditions of perfect competition with
marginal cost at (zc ). Let qtS (zc ) be the production of commodity z in country c.3
3. In every period consumers in each country are assumed to maximize identical Cobb-Douglas preferences over their aggregate consumption of each commodity,
3
The terms ‘industry’ and ‘commodity’ are essentially interchangeable in this framework.
6
Qt (z), with the function of income spent on commodity z being b (z) (Equation 1).
Expenditure shares for each commodity are therefore constant for all prices and incomes. All income is spent so the integral of b (z) over the interval [0, 1] is 1 (Equation
2).
Ut =
Z1
b (z) ln Qt (z) dz.
(1)
Z1
b (z) dz = 1.
(2)
0
0
4. A CES demand structure is assumed. Each commodity is not a homogeneous
good. Although firms in the same country produce identical goods, production is
differentiated by country of origin.4 Qt (z) can be interpreted as a sub-utility function
that depends on the quantity of each variety of z consumed. I choose the CES function
with elasticity of substitution σ z > 1. Let qtD (zc ) denote the quantity consumed of
commodity z produced in country c. Qt (z) is defined by Equation 3:
ÃN
! σ σZ−1
Z
X
σ Z −1
Qt (z) =
qtD (zc ) σZ
.
(3)
c=1
5. There may be transport costs for international trade. Transport costs are introduced in the convenient ‘iceberg’ from; gc0 t (zc ) units must be shipped from country
c for 1 unit to arrive in country c0 ; gct (zc ) = 1, ∀c .5
6. Tariffs: τ c0 t (zc ) is the ad-valorem tariff imposed by country c0 on imports of
commodity z from country c; τ ct (zc ) = 0, ∀c . Tariffs are rebated as a lump-sum to
consumers.
B. Equilibrium
In equilibrium, consumers maximize utility, firms maximize profits and trade is
balanced. Because of the assumption of perfect competition, prices (exclusive of tariffs
and transport costs) are equal to marginal cost, at (zc ) . Consider the consumers in
country 1, which we will call the US. Tariffs and transport costs raise the price paid
by US consumers for goods imported from country c to at (zc ) g1t (zc ) (1 + τ 1t (zc )).
4
The model could be extended to allow Nct varieties to be produced in country c. This would
have no impact on the analysis.
5
Transport costs are introduced only to clarify one of my identification assumptions, they play
no other important role.
7
Let T1t (z) denote tariff revenue collected in the US on imports of commodity z, let
D
(zc ) denote US consumption of commodity z produced
Y1t denote US income, and q1t
in country c. US income is equal to the sum of firm revenues plus tariff revenue.6
T1t (z) =
X
D
τ 1t (zc ) q1t
(zc ) at (zc ) ,
(4)
c
Y1t =
Z1
at (z1 ) qtS (z1 ) dz +
0
Z1
T1t (z) dz.
(5)
0
US consumers maximize utility subject to expenditure being equal to income in
every period. Due to the unit substitution elasticity between industries, the share of
income spent on commodity z is constant at b (z) :
X
D
q1t
(zc ) at (zc ) g1t (zc ) (1 + τ 1t (zc )) = b (z) Y1t .
(6)
c
Differentiating the Lagrangian for the consumers’ constrained optimization problem with respect to consumption levels of each commodity, we find that the tariff on
imported goods causes domestic consumers to substitute towards domestically produced varieties. The amount of substitution depends on the level of the tariff and on
the elasticity of substitution between varieties:
∀z, ∀c , ∀t ,
D
(zc )
q1t
=
D
q1t (zc0 )
µ
1 + τ 1t (zc0 )
1 + τ 1t (zc )
¶σZ µ
at (zc0 )
at (zc )
¶σZ µ
g1 (zc0 )
g1 (zc )
¶σZ
.
(7)
Equilibrium conditions for all other countries are symmetric, which will be exploited by the empirical work to control for the effect of unobserved cost shocks that
may be correlated with tariff movements. Finally, all commodity markets have to
clear, taking in to account output that melts in transit:7
∀z, ∀c , ∀t ,
qtS (zc ) =
X
qcD0 t (zc ) gc0 t (zc ) .
(8)
c0
C. Empirical Strategy
Equation 7 will form the basis of the empirical examination of NAFTA. Equivalent
equations exist for every other country, specifically, let country 2 be the EU:
6
Revenue from firm sales will all accrue to factors of production (inputs), which are assumed to
be domestically owned.
7
Underlying factor (input) markets are not modelled.
8
∀z, ∀c , ∀t ,
D
(zc )
q2t
=
D
q2t (zc0 )
µ
1 + τ 2t (zc0 )
1 + τ 2t (zc )
¶σZ µ
at (zc0 )
at (zc )
¶σZ µ
g2t (zc0 )
g2t (zc )
¶σZ
.
(9)
Using Equations 7 and 9 we can eliminate the effect of time-varying marginal
costs:
¸
·
D
D
(zc )
(zc )
q2t
1 + τ 2t (zc0 )
1 + τ 1t (zc0 )
q1t
− ln D
= σ z ln
− ln
ln D
1 + τ 1t (zc )
1 + τ 2t (zc )
q1t (zc0 )
q2t (zc0 )
¸
·
g2t (zc0 )
g1t (zc0 )
+σ z ln
− ln
.
g1t (zc )
g2t (zc )
(10)
Taking the time-difference of Equation 10 gives us:
·
¸
¸
·
D
D
q1t
(zc )
(zc )
q2t
1 + τ 2t (zc0 )
1 + τ 1t (zc0 )
∆ ln D
− ln D
= σ z ∆ ln
− ∆ ln
1 + τ 1t (zc )
1 + τ 2t (zc )
q1t (zc0 )
q2t (zc0 )
¸
·
g2t (zc0 )
g1t (zc0 )
+σ z ∆ ln
− ln
.
g1t (zc )
g2t (zc )
(11)
So long as we only examine countries c and c0 for which the EU does not change its
2t (zc0 )
= 0. Furthermore, I do not have detailed transport
relative tariffs, then ∆ ln 1+τ
1+τ 2t (zc )
cost data, so to identify σ Z I assume that theh change in relative transport
costs of
i
g1t (zc0 )
g2t (zc0 )
shipping commodities to the US and the EU, ∆ ln g1t (zc ) − ln g2t (zc ) , is orthogonal to
1t (zc0 ) 8
the change in US tariffs ∆ ln 1+τ
. This produces the basic estimating Equation
1+τ 1t (zc )
12, where σ will be a weighted average of σ z , and εcc0 z is a random disturbance term.
¸
·
¸
·
D
D
(zc )
(zc )
q2t
1 + τ 1t (zc0 )
q1t
− ln D
= σ ∆ ln
+ εcc0 z
∆ ln D
1 + τ 1t (zc )
q1t (zc0 )
q2t (zc0 )
(12)
We can derive the equivalent expression for the change in the relative value of
imports from Equation 12:
8
The assumption may not be completely innocuous. The most significant recent feature of international trade costs has been the relative decline in air-freight costs. This is likely to disproportionately benefit some commodities and some trade routes. See Hummels (1999) for a detailed
examination of international trade costs.
9
¸
·
¸
·
D
D
(zc )
(zc )
at (zc ) q2t
1 + τ 1t (zc0 )
at (zc ) q1t
− ln
= σ ∆ ln
+ εcc0 z
∆ ln
D
D
1 + τ 1t (zc )
at (zc0 ) q1t
(zc0 )
at (zc0 ) q2t
(zc0 )
(13)
Now consider country c to be Canada or Mexico and country c0 to be any other
country. NAFTA’s and CUSFTA’s increase in the US tariff preferences for Canadian
1t (zc0 )
, will increase their share of US consumption and
and Mexican goods, ∆ ln 1+τ
1+τ 1t (zc )
imports relative to their share of EU consumption and imports. A fraction of the
increased share of US consumption comes from reduced output of domestic suppliers
(“trade creation”), and the rest comes from reduced imports from countries outside
NAFTA (“trade diversion”). The size of the increased share in an arbitrary industry
z depends positively on the size of the increased US tariff preference, and positively
on the elasticity of substitution σ between varieties of z.
4
Data Description
Since 1988 the EU and since 1989 the US have collected their trade data according to
the Harmonized Schedule (HS), a schedule that is now standard for many countries
up to the 6-digit level, or 5,109 commodities. The US International Trade Commission (USITC) maintains a database at the 10-digit level of US imports classified by
commodity, country of origin, import program, month and port of arrival. Eurostat
maintains a similar database for the EU. US tariffs are almost invariably set at the
8-digit level, comprising about 12,000 commodities by the year 2000. Changes in HS
commodity classifications lead to some attrition, but we are able to track US and
EU trade in 4,655 6-digit commodities annually from 1989 to 2000. Because Canada
entered into CUSFTA with the US in 1989, it is useful to collect data for earlier years.
Prior to 1989, trade data was collected according to a different commodity schedule,
the TSUSA. Concordances are available for this data, but revisions to the TSUSA
also lead to attrition. We are able to track 4,483 commodities continuously from 1988
to 2000, and 3,592 from 1980 to 2000.
For each year I calculate the share of US imports of each commodity measured
by customs value (that is, exclusive of tariffs, freight and insurance) that originate in
each of the trading partners of the US. The change in the simple average of Canada’s
and Mexico’s share of US imports by commodity is summarized in Figure 3. From
1980 to 1988 Canada’s simple average share of US imports declined by 3.5 percentage
points, but from 1989 to 2000 some of this decline was reversed, with the average share
increasing by 2.1 percentage points. Between 1980 and 1985, Mexico’s share of US
imports barely changed, rising by 0.2 percent. In 1986 Mexico joined the GATT and
unilaterally liberalized is trade regime. Reducing extremely high import protection
helped to promote exports, and Mexico’s share of US imports rose by 1.4 percentage
points from 1985 to 1993. From 1994 with Mexico a part of NAFTA, its share of US
imports increased a further 1.8 percentage points, but the first big increase occurred
10
in 1995, leading to speculation that the devaluation in 1994 and not NAFTA was
responsible.
The data also contains information on physical quantities imported for a large
number of the commodities down to the HS 10-digit level, allowing the calculation of
unit price variables. Where possible, I calculate the price of Canadian and Mexican
goods relative to the price of goods sourced from the rest of the world, denoted
RPt (zc ). These prices will be used to make a very tentative search for NAFTA’s
terms of trade effects.
US tariff rates for the years 1997 to 2000 are available from the USITC and tariff
data for earlier years was extracted from USITC files. Tariffs are almost invariably
set at the HS 8-digit level. While most tariffs are ad-valorem, there are still several
hundred specific tariffs applied. The USITC calculates the ad-valorem equivalent of
any specific tariffs. The distribution of MFN tariffs in 2000 is illustrated in Figure
4A. The simple average of tariff rates is low at 5.5 per cent, but importantly there
is a large amount of dispersion, with the standard deviation of MFN tariff rates
being 12 per cent. Under NAFTA, all but a few hundred of these tariffs have been
eliminated for Canada and are in the process of being eliminated for Mexico, creating
a large variation in the preference given to goods of Canadian and Mexican origin
(Figure 4B). Table 1 shows that much of this variation occurs even within fine product
classifications. Industry-level studies therefore ignore most of the tariff variation.
Complicating matters was the existence of preferential treatment for some Mexican and Canadian goods prior to CUSFTA/NAFTA. In 1965, Canada and the US
negotiated the Auto-Pact, allowing duty-free trade in many automotive goods. The
Auto Pact was incorporated into CUSFTA. Mexico was a beneficiary of the Generalized System of Preferences (GSP), under which the US (and other developed
countries) gave developing countries preferential access to their markets. The US
gave duty free access to the output of developing countries for several thousand HS 8digit commodities, although goods where developing countries may have gained most
from preferential access were often excluded (notably many agricultural items and
textiles, clothing and footwear), and the preference could easily be removed under
“competitive needs limitations” to the GSP. Upon entry into NAFTA, Mexico was
no longer entitled to claim GSP benefits for trade with the US.
For each HS 8-digit commodity, I calculate the preference afforded to Canadian
and Mexican goods as the MFN tariff rate applicable to that commodity in January
1999 less the tariff rate applicable to Canadian and Mexican goods respectively. The
distribution of these preferences is illustrated in Figure 4B. But it is the increase in
the preference that matters. If there was a pre-existing preference under the GSP for
Mexico or the Auto-Pact for Canada, then this preference was simply the applicable
MFN tariff rate. The preferences were aggregated to the HS 6-digit level, see the Data
Appendix for details. One factor complicating the calculation of tariff preferences is
the existence of the maquiladoras, which exploited the fact that for many imported
11
products the US did not charge duty on the US-produced content in imports. For
these products, calculating preferences using the tariff schedule will tend to overstate
the new tariff preferences. Wherever trade in a product is observed, I also calculate
tariff preferences using data on actual import duty paid.
A further complication is the existence of quantitative restrictions on imports of
many textile, clothing and footwear commodities under the Multi-Fibre Agreement
(MFA) and of many agricultural commodities. Many of these restrictions are binding,
though most are not. They are extremely difficult to account for, many restrictions
encompass many HS commodities and most apply bilaterally. Special HS codes are
often created for these restrictions. I take two extreme approaches to the problem.
One is to ignore the problem completely, and treat the tariff as the only measure
of protection for these commodities, and the other extreme is to drop all affected
commodities.
The preferences given to Canadian and Mexican production are systematically
related to some of the characteristics of the commodities. This is to some extent
evident from Figures 1A to 2B showing a systematic negative relationship between
the preference and Canada’s and, to a lesser extent, Mexico’s market share. Given
that the most protected sectors are agriculture and simple manufactures like textiles,
apparel and footwear, the highest preferences are mostly in these sectors, subject to
the existence of quantitative restrictions under the Multi-Fibre Agreement (MFA).
This is especially true for simple manufactures because much agricultural protection
was preserved under NAFTA. Where the preference exceeds 20 percent for either
Mexican or Canadian goods, 70 percent of the commodities are textiles, clothing or
footwear, 17 percent are agricultural commodities, and the remainder are light trucks
(including many SUVs), glassware, bags, brooms and cheap watch movements.
The NAFTA preferences are strongly biased towards commodities in which developed countries have a comparative disadvantage. This effect can be seen in Table
2. This table examines the relationship between the relative price of Canadian and
Mexican exports to the US and NAFTA preferences for Canadian and Mexican goods.
For each commodity, the unit price of Canadian and Mexican exports to the US (exclusive of tariffs) is divided by the unit price of exports from the rest of the world.
The unit import price data contains some very extreme values, therefore I calculate
the median rather than the average relative price of exports, RPt (zc ), for several
arbitrary tariff preference classes. Table 2 strongly suggests that the relative price of
Canadian goods was and is substantially higher in commodities where there is a large
NAFTA preference, a situation that has become noticeably worse where the preference is between 5 and 20 percent. This is consistent with the hypothesis that NAFTA
preferences are skewed towards goods where developed countries have a comparative
disadvantage and suggests that NAFTA may have caused Canada to expand its share
of US imports in commodities where it is a relatively high cost producer.
One interesting feature of Table 2 is the behavior of relative prices for the com12
modities where the greatest preference was conferred. For both Canada and Mexico
the relative price of these goods declines, unlike commodities in other tariff classes.
This last phenomenon admits many interpretations, although it has to be noted that
the sample size is small. One interpretation is that NAFTA may have led to productivity improvements in these very protected industries. An alternative explanation
is that with a larger market under NAFTA, surviving firms producing these commodities have been able to exploit scale economies and move down their average cost
curve. A third explanation is simply that with the benefit of substantial tariff preferences, Canada and Mexico are now able to profitably export to the US very low-value
varieties of these commodities. Why this effect is only evident for the very highly
preferred products is hard to explain. [check EU data...].
The systematic association between NAFTA preferences and some underlying
sources of comparative advantage is of great concern for the empirical analysis because
there is a need to distinguish changes in trade patterns that are due to NAFTA from
changes in trade patterns that would have happened anyway due to shifting comparative advantage. This is done in two ways. The first way comes directly from the
model. Changes in comparative advantage will be reflected in production costs, but
by including EU data in the analysis we can control for the effect of cost shocks using
Equation 10. The second way is to assemble a set of controls for commodity characteristics using USITC data and 6-digit NAICS industry data from the 1997 Economic
Census. The commodity characteristics are the transport costs for the commodity
estimated by dividing Cost including Insurance and Freight (CIF) by Free On Board
(FOB) import values, and the units that the commodity is measured in. A commodity that is expensive to ship may have substantially different characteristics from one
that is cheap to ship. A commodity that is sold by the tonne may have different
characteristics to one sold by the dozen. The industry characteristics include factor
intensity estimates: whether the industry is agricultural; skill intensity measured by
the proportion of non-production workers and by average compensation; and capital
intensity measured by the share of value added that is not total compensation. Other
industry characteristics include firm size measures (average employment per establishment and average assets per establishment) and the proportion of total sales that
is value added.
5
Results
A. Elasticity of Substitution
The mean elasticity of substitution σ is estimated using the following estimating
equation derived from Equation 13:
13
¸
·
¸
·
D
D
(zc )
(zc )
at (zc ) q2t
1 + τ 1t (zc0 )
at (zc ) q1t
− ln
= σ ∆ ln
+ x0z .π c + εcc0 z ,
∆ ln
D
D
1 + τ 1t (zc )
at (zc0 ) q1t
(zc0 )
at (zc0 ) q2t
(zc0 )
(14)
D
(zc ) is the FOB value of US imports of commodity z from country
where at (zc ) q1t
D
(zc ) is EU imports of commodity z from country c at time t;
c at time t; at (zc ) q2t
τ 1t (zc ) is the US ad-valorem tariff on imports of commodity z from country c at time
t; xz is an additional set of controls with effects π c ; and εcc0 z is a random disturbance
term. The parameter σ is of interest because it helps determine the effect of trade
impediments on the volume of trade; it dictates the extent of trade diversion and
trade creation from the preferential trade liberalization; and because it is a critical
ingredient of welfare analysis of trade liberalization. I estimate Equation 14 by OLS,
where country c is alternatively Canada or Mexico, country c0 is the aggregate of all
countries that do not have a free trade agreement with the US or the EU and did not
substantially change their preferential trade relations with either the US or the EU
between 1988 and 1999. A list of these countries is provided in Appendix Table 1.
1988 and 1999 trade data is used for Canada and 1993 and 1999 data for Mexico. It
should be noted that the dependent variable is only defined for commodities where
imports from the relevant country are observed in both years, resulting in a substantial
number of missing observations, especially for Mexico.9 Industry characteristics are
progressively controlled for, in case EU trade data does not provide an adequate
control for unobserved cost shocks that may be correlated with US tariff preferences.
The results are reported in Tables 3A to 3C.
Table 3A reports OLS estimates in columns 1-3 and 7-9. The estimates of the
elasticity of substitution typically range between 4 and 7 and are usually reasonably
precisely estimated. Moving across the columns, adding extra industry controls decreases the estimates based on Mexican trade data but increases them for Canada.
One concern with the OLS estimates is that they are based only on those commodities
that were imported from the relevant country in both the base year and the year 1999.
Approximately half of the observations are missing for Canada, and two-thirds for
Mexico. To see if selection bias is an issue, columns 4-6 and 7-9 report results from the
Heckman (1976) selection model, with selection being modelled as a linear function
of the aforementioned industry characteristics and industry dummies. Countries may
tend to specialize in industries that share certain characteristics. While the selection
variables matter for the selection equation (not tabulated), controlling for selection
itself appears to make little difference to the substitution elasticity estimates.
The estimated substitution elasticity was allowed to vary across industry by interacting the NAFTA tariff preferences with other industry characteristics. Once
the elasticity was allowed to vary with the tariff preference itself, no other industry
9
Some of this attrition is due to changes in commodity classifications.
14
characteristic significantly affected the estimates. Table 3B reports results where the
substitution elasticity was only allowed to vary with the tariff preference. For almost
all products these estimates suggest higher elasticities than were estimated in Table
3A. Interestingly, the interaction term is usually significantly negative, suggesting
that goods with higher tariff preferences are less substitutable across source. This
is surprising given that these products are mostly agricultural goods or simple manufactures, and may reflect the existence of quantitative import controls on some of
these items.
Table 3C repeats Tables 3A and 3B but seeks to correct for the effect of the
maquiladoras on Mexican exports to the US by calculating tariff preferences using
US data on actual import duty paid. The estimated substitution elasticities change
little, and for a typical product range between 5 and 7. The interaction term changes
substantially though, with the coefficient always suggesting that the more highly
preferred products are more substitutable across source.
Table 3D reports equivalent results to Table 3A but estimates Equation 15 using
more detailed US trade data at the HS 8-digit level. This roughly doubles the number
of observations but incurs the disadvantage of losing EU trade data as a control for
unobserved shocks. The results are similar, with the typical substitution elasticity
being between 5 and 7.
D
(zc )
q1t
∆ ln D
q1t (zc0 )
=
·
¸
1 + τ 1t (zc0 )
σ ∆ ln
+ x0z .π c + εcc0 z .
1 + τ 1t (zc )
(15)
These elasticities of substitution suggest that consumers are quite willing to substitute between different sources of a commodity. One implication of this willingness
to substitute is that small costs to international trade, whether due to natural barriers such as transport costs or artificial barriers such as tariffs, will have a large effect
on trade volumes. With a substitution elasticity of 4, the median US tariff of 5.5 per
cent will reduce consumption of imported varieties relative to domestic varieties by 20
per cent. With a substitution elasticity of 7, this reduction in relative consumption is
31 per cent. But on some products the effect will be much more dramatic; US tariffs
range up to 350 per cent.
B. Trade Creation and Trade Diversion
The results reported above suggest that NAFTA preferences have had a pronounced effect on the source of US imports. But the results do not tell us whether
the increased imports from Canada and Mexico are the result of new international
trade displacing US domestic production (“trade creation”) or result from displacement of imports from other sources (“trade diversion”). The model presented in
Section 3 predicts that it will be a little of both. The model and the estimates of
15
σ can be used to estimate what proportion of increased imports from Canada and
Mexico is from trade creation, and what proportion is from trade diversion. To get a
tractable expression that can be used to approximate the relative importance of trade
creation and trade diversion it is convenient to hold production costs fixed or, in other
words, to temporarily ignore the terms of trade effects of preferential trade liberalization. With production costs fixed there will be a very small decline in nominal US
national income resulting from the loss of tariff revenue on Mexican and Canadian
imports, and given the assumed substitution elasticities [try nested CES...], nominal
expenditures on each commodity z will be little changed. A measure of the importance of trade creation will be the decline in the share of US expenditure spent on
US production. A measure of the importance of trade diversion is the decline in the
share of US expenditure spent on production from outside NAFTA. From Equation
7 we can derive Equations 16 and 17 for US expenditure shares on commodity z that
are spent on the output of country c0 :
" µ
¶σ µ
¶σ−1 µ
¶σ #−1
D
(zc0 )
a.q1,t=88
1 + τ 1,t=88 (zc0 )
at=88 (zc0 )
g1,t=88 (zc0 )
= Σ
. (16)
D
c
1 + τ 1,t=88 (zc )
at=88 (zc )
g1,t=88 (zc )
Σa.q1,t=88
(zc )
c
"
µ
¶ µ
¶ #−1
D
Dd
a.q1,t=88
(zc ) 1 + τ 1,t=99 (zc0 ) σ 1 + τ 1,t=88 (zc0 ) −σ
a.q1,t=99
(zc0 )
= Σ D
.
D
c a.q1,t=88 (zc0 )
1 + τ 1,t=99 (zc )
1 + τ 1,t=88 (zc )
Σa.q1,t=99
(zc )
c
(17)
Equation 16 gives the expenditure shares prior to CUSFTA and NAFTA, while
Equation 17 is a prediction based on the prior expenditure shares, the extent of preferential tariff liberalization and the substitution elasticity. The change in aggregate
US expenditure E1c on the output of country c is simply a weighted sum of predicted shares in Equation 17 less actual shares in 1988, with the weights given by US
consumption b (z) Y1 :
∆E1c0 =
Z
z
b (z) Y1
"
D
Dd
a.q1,t=88
(zc0 )
(zc0 )
a.q1,t=99
−
D
D
Σa.q1,t=99 (zc ) Σa.q1,t=88 (zc )
c
c
#
(18)
All the expenditure shares and tariffs are essentially observable, all that is needed
to estimate Equation 18 is an estimate of σ, which was estimated above to be approximately 6. Expenditure shares are estimated at the 4-digit SIC industry level because
of the need to utilize US production data. US production data for 1988 is industry
value added data from the NBER productivity database. The amount of US production that is consumed domestically is estimated by subtracting US exports from US
16
production at the 4-digit SIC level using export data from the Center for International Data at UC Davis. 1988 import data is also from the Center for International
Data. This data and the estimated substitution elasticity suggest that CUSFTA and
NAFTA should have caused a 10 percent increase in Canada’s share of US imports
and a 21 percent increase in Mexico’s share of US imports. The actual increases since
CUSFTA and NAFTA have been 11 percent and 66 percent respectively. CUSFTA
and NAFTA appear to be responsible for almost all of Canada’s increased exports to
the US and one-third of Mexico’s increased exports to the US. The estimates also suggest that 59 percent of these increased exports have displaced US production, while
41 percent have displaced exports from non-NAFTA countries. This high proportion
of trade diversion is due to a very simple reason: the largest tariff preferences under
NAFTA and CUSFTA are in industries where imports from outside of North America
represent a substantial proportion of domestic consumption.
C. Import Price Data
Unit import price data also provide some evidence of the original Vinerian concept
of trade-diversion: incurring a real resource cost by paying more for imports from
preferred trading partners (exclusive of tariffs) because of the tariff preference. This
effect is examined by regressing the change in the relative price, ∆RP (zc ), of imports
from Canada and Mexico on the tariff preference extended to them. Table 2 suggests
that the FOB price of US imports from Canada, and to a lesser extent Mexico, have
become relatively more expensive where the NAFTA preference is between 5 and 20
per cent, but any relationship is potentially clouded by relative price declines in a
small number of extremely preferred industries. Equations are estimated of the form:
∆RP (zc ) =
x0z .π c
¶j
µ
J
X
1 + τ 1t (zc0 )
+
δ cj .∆ ln
+ εcz
1 + τ 1t (zc )
j=1
(19)
´
³
1t (zc0 )
where x0z is a vector of controls for industry characteristics, and ∆ ln 1+τ
1+τ 1t (zc )
is the increased US tariff preference for Canadian or Mexican goods due to CUSFTA
and NAFTA. Figures 5A to 5B and Tables 4A to 4D report the regression results separately for Canada and Mexico. Extreme observations are prevalent in the unit import
price data, so median regressions are also estimated in addition to OLS regression
results. Columns 1 to 3 of Tables 4A and 4B for Canada suggest a very weak positive
relationship between NAFTA preferences and the change in relative import prices,
but columns 1 to 3 of Tables 4C and 4D for Mexico suggest a significant negative
relationship, providing no evidence of trade diversion in the unit import price data.
But adding higher-order terms to the regressions changes this conclusion. Columns 4
to 6 of each table report results where the square of NAFTA preferences have been
added. For both Canada and Mexico it appears that at relatively low levels of preference the change in the relative price of their exports to the US is positively associated
with the tariff preference, providing evidence of trade diversion. But beyond a point
17
this relationship changes. The turning point comes at approximately 15% preference
for Canada and 10% for Mexico, and at very high preference levels relative import
prices appear to have dropped. One interpretation of this result is that for at least a
limited range of industries there have been substantial productivity improvements in
these very sheltered industries brought about by NAFTA. An alternative explanation
is that with a larger market under NAFTA, surviving firms producing these commodities have been able to exploit scale economies. A third explanation is that the
large tariff preference has enabled Canada and Mexico to profitably produce very low
value versions of these products. For Canada the productivity explanation would be
consistent with a finding in Trefler (2001) that the clothing industry has experienced
substantial productivity gains. Mexico, as the only substantial developing economy
with preferential access to the highly protected US clothing and textile market, may
have also witnessed such productivity gains as it became an attractive location to
supply the US market. The regression results do not qualitatively change when further higher order terms are added, the results where a fifth-order polynomial in the
tariff preference are used are depicted in Figures 5A and 5B.
6
Conclusion
This paper seeks to identify an effect for NAFTA by focusing on where the United
States and the European Union source their imports of different commodities from.
NAFTA appears to have had a substantial effect on North American trade. Mexican
and Canadian shares of US imports have increased most rapidly in commodities where
the greatest NAFTA preferences were conferred, even though Canada appears to be a
high cost producer of many of these commodities. The Canadian share of US imports
declined in commodities where it was not given a new preference, while the Mexican
share increased much more modestly. The results of this paper suggest that trade flows
are very sensitive to even small tariff preferences. The NAFTA preferences can be
used to estimate how willing consumers are to substitute between different varieties of
the same commodity. The implied average substitution elasticity is typically between
5 and 7. Consumers are quite willing to substitute between different varieties of the
same commodity. Small changes in trade impediments, whether due to natural or
non-natural barriers, could therefore have substantial effects on international trade
volumes. Preferential liberalization will have substantial effects on the direction of
trade. One disturbing result is that NAFTA appears to have caused a substantial
amount of trade diversion.
18
7
7.1
Appendix
Data
Tariffs were aggregated from the HS 8-digit to HS 6-digit level by solving the following
equation for t in each HS 6-digit category:
σ
(1 + t) =
µP
(1 + ti )σ
iM
Pi
i Mi
¶
,
(20)
where Mi is the value of imports of each HS 8-digit product within the HS 6-digit
category, ti is the HS 8-digit tariff, and σ is the elasticity of substitution between
different sources of HS 8-digit products, which I estimate using US data to be approximately 6.
19
References
[1] Armington P. (1969), “A Theory of Demand for Products Distinguished by Place
of Production”, IMF Staff Papers 16 (March), pp.159-78.
[2] Bagwell K. and R. Staiger (1999), “An Economic Theory of GATT”, American
Economic Review, Vol. 89, No.1, pp.215-248.
[3] Baldwin R. (1996), “A Domino Theory of Regionalism”, in Baldwin R., P. Haaparanta and J. Kiander eds., Expanding the Membership of the EU, Cambridge
University Press.
[4] Baldwin R.E. and A. J. Venables (1995), “Regional Economic Integration”, in
Grossman G.M. and K. Rogoff eds., Handbook of International Economics, Volume 3, North-Holland.
[5] Brown D.K., A.V. Deardorff and R.M. Stern (1995), “Estimates of a North
American Free Trade Agreement”, in Kehoe P.J. and Kehoe T.J. (eds.), Modeling
North American Integration, Kluwer Academic Publishers, Dordrecht.
[6] Chang W. and A. Winters (1999), “How Regional Blocs Affect Excluded Countries: The Price Effects of MERCOSUR”, World Bank Policy Research Working
Paper No. 2157.
[7] Clausing K. A. (2001), “Trade Creation and Trade Diversion in the CanadaUnited States Free Trade Agreement”, forthcoming Canadian Journal of Economics, August 2001.
[8] Cox D.J. (1995), “An Applied General Equilibrium Analysis of NAFTA’s Impact
on Canada”, in Kehoe P.J. and Kehoe T.J. (eds.), Modeling North American
Integration, Kluwer Academic Publishers, Dordrecht.
[9] Deardorff A. and R. Stern (1994), “Multilateral Trade Negotiations and Preferential Trading Arrangements” in A. Deardorff and R. Stern (eds), Analytical
and Negotiating Issues in the Global Trading System, The University of Michigan
Press, Ann Arbor.
[10] Frankel J., E. Stein and S. Wei (1996), “Improving the Design of Regional Trade
Agreements. Regional Trading Arrangements: Natural or Supernatural?”, AEA
Papers and Proceedings, Vol. 86, No. 2, pp.52-56.
[11] Fukao K., T. Okubo, and R.M Stern (2001), “Trade Diversion under NAFTA”,
mimeo, The University of Michigan.
[12] Garces-Diaz D. (2001), “Was NAFTA Behind the Mexican Export Boom (19942000)?”, mimeo (on SSRN), Banco de Mexico, February.
[13] Gould D. (1998), “Has NAFTA Changed North American Trade?”, Federal Reserve Bank of Dallas Economic Review, First Quarter, pp.12-22.
[14] Head K. and J. Ries (1999), “Rationalization effects of tariff reductions”, Journal
of International Economics, Vol. 47(2), pp.295—320.
[15] Heckman J. (1976), “The common structure of statistical models of truncation,
sample selection, and limited dependent variables and a simple estimator for such
models”, The Annals of Economic and Social Measurement, Vol. 5, pp.475-492.
[16] Hufbauer G. and J. Schott (1993), NAFTA: An Assessment, (Washington DC,
Institute for International Economics).
20
[17] Hummels D. (1999), “Have International Trade Costs Declined?”, mimeo.
[18] Kehoe P.J. and Kehoe T.J. (1995), “Capturing NAFTA’s Impact with Applied
General Equilibrium Models”, in Kehoe P.J. and Kehoe T.J. (ed.), Modeling
North American Integration, Kluwer Academic Publishers, Dordrecht.
[19] Kemp M. and H. Wan (1976), “An Elementary Proposition Concerning the Formation of Customs Unions”, Journal of International Economics, Vol. 6, pp.9597.
[20] Krueger A. (1999), “Trade Creation and Trade Diversion Under NAFTA”, NBER
Working Paper No. 7429.
[21] Krueger A. (2000), “NAFTA’s Effects: A Preliminary Assessment”, World Economy, Vol. 23, No.6, pp.761-75.
[22] Levy P. (1997), “A Political-Economic Analysis of Free-Trade Agreements”,
American Economic Review, Vol. 87, No.4, pp.506-519.
[23] Sobarzo H.E. (1995), “A General Equilibrium Analysis of the Gains from NAFTA
for the Mexican Economy”, in Kehoe P.J. and Kehoe T.J. (eds.), Modeling North
American Integration, Kluwer Academic Publishers, Dordrecht.
[24] Trefler D. (2001), “The Long and Short of the Canada-U.S. Free Trade Agreement”, mimeo (on SSRN), University of Toronto, April.
[25] Viner J. (1950), The Customs Union Issue, (New York: Carnegie Endowment).
[26] Wall H. (2000), “NAFTA and the Geography of North American Trade”, mimeo
(on SSRN), Federal Reserve Bank of St. Louis, November.
[27] Yeats A. (1997), “Does Mercosur’s Trade Performance Raise Concerns about the
Effects of Regional Trade Arrangements?”, World Bank Policy Research Working
Paper No. 1729.
[28] Office of the United States Trade Representative (1999), U.S. Generalized System
of Preferences Guidebook, Executive Office of the President, Washington D.C.,
March 1999.
21
Figure 1A
NAFTA's Impact on Mexico's Share of US Imports
Simple Average of Share of US Imports by
Commodity
0.12
0.10
0.08
0.06
0.04
No new tariff preference (2629 commodities)
0.02
New tariff preference >0% and < 10% (1556 commodities)
New tariff preference >= 10% (298 commodities)
0.00
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
Year
Figure 1B
Mexico's Share of EU Imports
Simple Average of Share of EU Imports by
Commodity
0.008
0.006
0.004
0.002
No new US tariff preference (2629 commodities)
New US tariff preference >0% and < 10% (1556 commodities)
New US tariff preference >= 10% (298 commodities)
0.000
1989
1990
1991
1992
1993
1994
1995
Year
1996
1997
1998
1999
2000
Figure 1C
Mexico's Share of US Imports 1980-2000
Simple Average of Share of US Imports by
Commodity
0.14
1986-2000 GATT/WTO
1994-2000 NAFTA
0.12
0.10
0.08
0.06
0.04
0.02
No new tariff preference (2089 commodities)
New tariff preference >0% and < 10% (1337 commodities)
New tariff preference >= 10% (166 commodities)
19
80
19
81
19
82
19
83
19
84
19
85
19
86
19
87
19
88
19
89
19
90
19
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
0.00
Year
Figure 2A
CUSFTA's Impact on Canada's Share of US Imports
Simple Average of Share of US Imports by
Commodity
0.25
0.20
0.15
0.10
0.05
No new US tariff preference (1551 commodities)
New US tariff preference >0% and < 10% (2540 commodities)
New US tariff preference >= 10% (392 commodities)
0.00
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Year
Figure 2B
CUSFTA's Impact on Canada's Share of US Imports
Simple Average of Share of US Imports by
Commodity
0.35
0.30
0.25
0.20
0.15
0.10
0.05
No new tariff preference (1286 commodities)
New tariff preference >0% and < 10% (2092 commodities)
New tariff preference >= 10% (214 commodities)
19
80
19
81
19
82
19
83
19
84
19
85
19
86
19
87
19
88
19
89
19
90
19
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
0.00
Year
Figure 2C
Canada's Share of EU Imports
Simple Average of Share of EU Imports by
Commodity
0.04
No new US tariff preference (1551 commodities)
New US tariff preference >0% and < 10% (2540 commodities)
New US tariff preference >= 10% (392 commodities)
0.03
0.02
0.01
0.00
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Year
Figure 3
Canada's and Mexico's Share of US Imports 1980-2000
0.10
0.25
0.20
0.08
0.15
0.06
0.04
0.10
Mexico (Right Scale)
0.02
0.05
0.00
1980
1989
Year
Figure 4A
1994
2000
0.00
Simple Average of Share of US Imports
by Commodity (3592 commodities)
Simple Average of Share of US Imports by
Commodity (3592 commodities)
Canada (Left Scale)
US Tariff Preferences for Canadian and Mexican Goods in 2000
4000
Number of HTS8 Commodities
3500
Canada
Mexico
3000
2500
2000
1500
1000
500
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20 >20
MFN Tariff Rate (%) Less Tariff on Imports from Canada, Mexico
Figure 5A: NAFTA Preferences and the Change in the Relative Price of
Canada’s Exports 1988-2000
Fitted values
pct5
pct95
.352256
-.598161
0
nafta_pref
.395
Notes: The change in the log relative unit values of imports from Canada has been regressed on a 5th-order
polynomial in the tariff preference. The fitted values and a 90% confidence interval have been plotted
against the new US tariff preferences for Canadian exports under CUSFTA/NAFTA.
Figure 5B: NAFTA Preferences and the Change in the Relative Price of
Mexico’s Exports 1993-2000
Fitted values
pct5
pct95
.15527
-.570567
0
nafta_pref
.336
Notes: The change in the log relative unit values of imports from Mexico has been regressed on a 5th-order
polynomial in the tariff preference. The fitted values and a 90% confidence interval have been plotted
against the new US tariff preferences for Mexican exports under NAFTA.
Table 1: Percentage of HS 8-digit Tariff Variation Captured by Broader Classifications
Classification Industries
HS-2
HS-4
HS-6
MFN Tariff Canada Preference Mexico Preference
(A)
(B)
(A)
(B)
(A)
(B)
97 0.205 0.351 0.199
0.363
0.181
0.367
1241 0.280 0.470 0.279
0.542
0.251
0.530
5109 0.330 0.555 0.321
0.671
0.289
0.671
Notes: Table 1 reports the percentage of MFN and preferential tariff variation at the tariff-line level that occurs
within broader product classes. Columns headed ‘A’ include 10,178 tariff lines from Chapters 1 to 97. Columns
headed ‘B’ exclude the 20 tariff lines where tariffs are in excess of 100%.
Table 2: Tariff Preference and Relative Prices of Canadian and Mexican Exports
Year 2000 Tariff
Preference %
0
(0,5]
(5,10]
(10,20]
>20
AutoPact
GSP
Canada: Median Log RPX
Mexico: Median Log RPX
1989
1993
2000
No.
0.055 0.067 0.127 1280
0.026 0.034 0.061 1372
0.134 0.205 0.288
661
0.277 0.463 0.536
383
0.730 0.770 0.687
111
0.129 0.187 0.327
236
1989
1993
2000
No.
-0.101 -0.019 -0.029
468
-0.106 -0.025 -0.028
277
-0.184
0.018
0.056
272
-0.201 -0.083 -0.057
173
-0.056
0.147 -0.144
65
-0.154
-0.033
0.007
1273
Notes: By tariff class, the table reports the median log price of Canadian and Mexican exports to the US relative
to the price of exports to the US from the rest of the world. The number of commodities for each calculation is
also reported.
Table 3A: Estimates of the Elasticity of Substitution based on US and EU Import Data
Mexico 1993-1999
Canada 1988-1999
σ
Industry
Characteristics
Industry
Dummies
OLS
Heckitt
(1)
4.23
(1.41)
No
(2)
4.35
(1.45)
Yes
(3)
5.50
(2.09)
Yes
(4)
5.17
(1.20)
No
(5)
4.13
(1.28)
Yes
(6)
5.77
(4.86)
Yes
(7)
7.09
(2.20)
No
(8)
6.53
(2.29)
Yes
(9)
3.82
(1.83)
Yes
(10)
6.85
(1.30)
No
(11)
6.13
(1.43)
Yes
(12)
3.81
(2.31)
Yes
No
No
Yes
No
No
Yes
No
No
Yes
No
No
Yes
Yes
Yes
Yes
No
No
No
Yes
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
No
No
No
Yes
Yes
2096 2054 2054 4890 4890 4890
1477 1444 1444 4890 4890
N
2096
2054
2054
2054
2054
2054
1488 1444 1444 1444 1444
Observed
Notes: robust standard errors appear in parentheses beneath OLS coefficient estimates, regular standard errors
appear beneath Heckitt estimates.
No
Yes
4890
1444
Table 3B: Estimates of the Elasticity of Substitution based on US and EU Import Data
Elasticity varies with NAFTA tariff preference
Canada 1988-1999
Mexico 1993-1999
(1)
9.26
(1.65)
(2)
8.27
(1.83)
(3)
8.01
(2.82)
(4)
9.56
(1.77)
(5)
7.98
(1.94)
(6)
8.27
(6.49)
(7)
13.86
(1.87)
(8)
13.88
(2.14)
(9)
1.38
(3.53)
(10)
13.51
(2.01)
(11)
13.39
(2.35)
(12)
1.40
(3.70)
-11.09
(2.12)
-9.33
(2.28)
-11.07
(6.92)
-11.15
(3.29)
-9.02
(3.43)
-10.98
(18.55)
-14.88
(3.01)
-14.82
(3.17)
4.79
(3.97)
-14.49
(3.33)
-14.14
(3.64)
4.74
(5.78)
No
Yes
Yes
No
Yes
Yes
No
Yes
Yes
No
Yes
Yes
No
No
Yes
No
No
Yes
No
No
Yes
No
No
Yes
Yes
Yes
No
No
No
Yes
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
No
No
No
Yes
Yes
2096
2054
2054
4890
4890
4890
1477
1444
1444
4890
4890
N
2096
2054
2054
2054
2054
2054
1488
1444
1444
1444
1444
Observed
Notes: robust standard errors appear in parentheses beneath OLS coefficient estimates, regular standard errors
appear beneath Heckitt estimates. The estimated elasticity for a given NAFTA tariff preference can be calculated
1 + τ 1t ( zc ' ) .
as σ0 plus the second reported coefficient multiplied by the NAFTA tariff preference
Yes
No
Yes
4890
1444
σ0

1 + τ 1t ( zc ' ) 
 ∆ t ln

1 + τ 1t ( zc ) 

Industry
Characteristics
Industry Dummies
OLS
Heckitt
2
∆ t ln
1 + τ 1t ( zc )
Table 3C: Estimates of the Elasticity of Substitution based on US and EU Import Data
NAFTA Tariff Preferences calculated from actual duty paid to adjust for the Maquiladoras
Mexico 1993-1999
(1)
4.88
(1.27)
(7)
5.82
(1.48)
(8)
6.11
(1.51)
(9)
6.61
(1.53)
(10)
5.92
(1.02)
(11)
6.13
(1.04)
(12)
6.61
(1.01)
3.47
(1.78)
4.01
(1.81)
5.63
(1.97)
3.08
(1.53)
3.71
(1.55)
5.63
(1.60)
No
Yes
Yes
No
Yes
Yes
No
No
Yes
No
No
Yes
No
No
Yes
No
No
Yes
Yes
Yes
No
No
No
Yes
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
No
No
No
Yes
Yes
2096
2054
2054
4890
4890
4890
1477
1444
1444
4890
4890
N
2096
2054
2054
2054
2054
2054
1488
1444
1444
1444
1444
Observed
Notes: robust standard errors appear in parentheses beneath OLS coefficient estimates, regular standard errors
appear beneath Heckitt estimates. The estimated elasticity for a given NAFTA tariff preference can be calculated
1 + τ 1t ( zc ' ) .
as σ0 plus the second reported coefficient multiplied by the NAFTA tariff preference
Yes
No
Yes
4890
1444
σ0

1 + τ 1t ( zc ' ) 
 ∆ t ln

1 + τ 1t ( zc ) 

Industry
Characteristics
Industry Dummies
OLS
Heckitt
(2)
4.93
(1.31)
(3)
5.12
(1.45)
(4)
5.06
(0.93)
(5)
5.05
(0.93)
(6)
5.12
(0.91)
2
No
Yes
Yes
No
Yes
Yes
∆ t ln
1 + τ 1t ( zc )
Table 3D: Estimates of the Elasticity of Substitution based on US HS-8 Import Data Only
Canada 1988-2000
Mexico 1993-2000
σ
Industry
Characteristics
Industry
Dummies
OLS
Heckitt
(1)
6.06
(0.74)
No
(2)
5.09
(0.81)
Yes
(3)
6.55
(1.11)
Yes
(4)
6.07
(0.78)
No
(5)
5.36
(0.84)
Yes
(6)
6.56
(1.06)
Yes
(7)
7.05
(0.95)
No
(8)
6.83
(1.07)
Yes
(9)
2.00
(1.70)
Yes
(10)
6.97
(0.94)
No
(11)
6.78
(1.13)
Yes
(12)
2.01
(1.50)
Yes
No
No
Yes
No
No
Yes
No
No
Yes
No
No
Yes
Yes
Yes
Yes
No
No
No
Yes
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
No
No
No
Yes
Yes
4931 4818 4818 9641 9641 9641
3026 2957 2957 9641 9641
N
4931 4818 4818 4818 4818 4818
3026 2957 2957 2957 2957
Observed
Notes: robust standard errors appear in parentheses beneath OLS coefficient estimates, regular standard errors
appear beneath Heckitt estimates.
No
Yes
9641
2957
Table 4A: Change in Relative Price of Imports from Canada
Dependent Variable: ∆RP89_00
Median Regression Results
RHS Var.
Pref
Pref2
(1)
0.24
(0.30)
(2)
0.19
(0.34)
(3)
0.37
(0.39)
(4)
1.21
(0.64)
-4.20
(2.49)
(5)
1.45
(0.71)
-5.36
(2.64)
(6)
1.27
(1.78)
-4.38
(5.66)
(7)
-1.79
(1.15)
27.0
(10.8)
-69.9
(23.6)
(8)
-0.59
(1.35)
15.8
(12.5)
-47.7
(27.2)
(9)
DNC
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
No
Yes
Yes
No
No
No
Yes
No
No
Yes
No
Pref3
Pref4
AutoPact
Controls
Industry
Characteristics
Industry
Dummies
N
(11)
-4.08
(1.63)
79.4
(25.0)
-376
(120)
498
(174)
Yes
(12)
DNC
Yes
(10)
-4.62
(1.75)
85.9
(26.9)
-398
(130)
518
(189)
Yes
Yes
Yes
No
Yes
Yes
No
Yes
No
No
Yes
Yes
2003 1947 1947 2003 1947 1947 2003 1947 1947 2003 1947 1947
Notes: standard errors appear in parentheses beneath coefficient estimates. “DNC” denotes regressions that did
not converge.
Table 4B: Change in Relative Price of Imports from Canada
Dependent Variable: ∆RP89_00
OLS Regression Results
RHS Var.
Pref
Pref2
(1)
0.23
(0.32)
(2)
0.37
(0.38)
(3)
0.74
(0.54)
(4)
0.89
(0.85)
-2.90
(2.89)
(5)
1.29
(0.94)
-3.99
(3.05)
(6)
2.13
(1.43)
-5.05
(3.94)
(7)
-0.52
(1.59)
12.4
(13.4)
-34.4
(27.8)
(8)
0.23
(1.67)
7.60
(13.9)
-26.0
(29.0)
(9)
1.79
(2.14)
-1.30
(15.7)
-8.36
(31.6)
Yes
No
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
No
No
Yes
No
No
Yes
No
No
Pref3
Pref4
GSP Controls
Industry
Characteristics
Industry
Dummies
N
2003 1947 1947 2003 1947 1947 2003 1947
Notes: robust standard errors appear in parentheses beneath coefficient estimates.
Yes
Yes
(10)
-3.93
(2.53)
78.8
(36.4)
-390
(166)
546
(234)
Yes
No
(11)
-3.46
(2.60)
79.8
(37.4)
-410
(170)
587
(238)
Yes
Yes
(12)
-1.14
(3.10)
56.2
(42.1)
-310
(190)
456
(265)
Yes
Yes
Yes
No
No
Yes
1947
2003
1947
1947
Table 4C: Change in Relative Price of Imports from Mexico
Dependent Variable: ∆RP93_00
Median Regression Results
RHS Var.
Pref
Pref2
(1)
-0.46
(0.26)
(2)
-0.63
(0.29)
(3)
DNC
(4)
1.57
(0.67)
-8.00
(2.65)
(5)
1.22
(0.77)
-6.86
(2.87)
(6)
DNC
(7)
3.56
(1.20)
-30.4
(11.9)
53.5
(27.4)
(8)
2.73
(1.40)
-24.6
(13.3)
43.3
(30.4)
(9)
DNC
Yes
No
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
No
No
Yes
No
No
Yes
No
No
Pref3
Pref4
GSP Controls
Industry
Characteristics
Industry
Dummies
N
(11)
1.07
(2.60)
11.6
(41.3)
-159
(206)
321
(317)
Yes
Yes
(12)
DNC
Yes
Yes
(10)
1.77
(2.03)
5.31
(32.7)
-140
(163)
302
(253)
Yes
No
Yes
No
No
Yes
Yes
Yes
2003 1947 1947 2003 1947 1947 2003 1947 1947 2003 1947 1947
Notes: standard errors appear in parentheses beneath coefficient estimates. “DNC” denotes regressions that did
not converge.
Table 4D: Change in Relative Price of Imports from Mexico
Dependent Variable: ∆RP93_00
OLS Regression Results
RHS Var.
Pref
Pref2
(1)
-0.36
(0.34)
(2)
-0.68
(0.43)
(3)
-0.37
(0.53)
(4)
1.83
(1.00)
-8.99
(3.26)
(5)
1.21
(1.17)
-7.34
(3.52)
(6)
3.21
(1.69)
-12.5
(4.65)
(7)
3.49
(1.97)
-28.3
(17.2)
46.5
(37.2)
(8)
2.59
(2.14)
-23.3
(17.8)
38.1
(38.0)
(9)
6.18
(2.86)
-44.6
(21.3)
74.3
(43.3)
Yes
No
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
No
No
Yes
No
No
Yes
No
No
Pref3
Pref4
GSP Controls
Industry
Characteristics
Industry
Dummies
N
2003 1947 1947 2003 1947 1947 2003 1947
Notes: robust standard errors appear in parentheses beneath coefficient estimates.
Yes
Yes
(10)
3.11
(3.56)
-19.0
(52.5)
-7.55
(241)
89.9
(343)
Yes
No
(11)
2.65
(3.72)
-21.8
(53.7)
25.6
(247)
23.7
(352)
Yes
Yes
(12)
7.15
(4.57)
-62.2
(59.6)
167
(263)
-147
(366)
Yes
Yes
Yes
No
No
Yes
1947
2003
1947
1947