International Trade in Used Vehicles: The Environmental Consequences of NAFTA
International Trade in Used Vehicles: The Environmental Consequences of NAFTA
International Trade in Used Vehicles:
The Environmental Consequences of NAFTA
July 2009
Abstract
Over the last two decades an unprecedented increase in private vehicle
ownership has taken place in the developing world. This growth is due, in part,
to increased international trade in used vehicles. This paper examines the
environmental consequences of this trade pattern focusing on the deregulation of
U.S.‐Mexico trade in used cars and trucks following the North American Free
Trade Agreement. Since trade restrictions were eliminated in 2005, over 2.5
million used vehicles have been exported from the United States to Mexico.
Using a unique, vehicle‐level dataset, we find that traded vehicles are dirtier than
the stock of vehicles in the United States and cleaner than the stock in Mexico, so
trade leads average vehicle emissions per mile to decrease in both countries.
Overall, trade leads to an increase in greenhouse gas emissions, primarily
because trade gives new life to vehicles that otherwise would have been
scrapped. These results are relevant for judging the international carbon
“leakage” consequences of proposed U.S carbon mitigation legislation.
Key Words: Trade and the Environment, Used Vehicles, NAFTA, Climate Change Mitigation
JEL: F18, H23, Q54, Q56
1. Introduction
Over the last two decades an unprecedented increase in private vehicle ownership has
taken place in the developing world. The total number of registered vehicles in non‐OECD
countries increased from 110 million to 210 million between 1990 and 2005, and, by some
estimates, is forecast to increase to 1.2 billion by 2030.1 Rising income explains a large share of
this growth. Another important but rarely discussed factor is international trade in used
vehicles. High‐income countries export large quantities of used vehicles to low‐income
countries. The scope for continued expansion of trade is enormous. For example, in 2007 there
were 768 total vehicles per 1000 people in the United States compared to 30 per 1000 in China
and only 12 per 1000 in India.
In this paper we examine the environmental consequences of international trade in used
vehicles. Vehicles play a central role in the production of local and global pollutants. Perhaps
most importantly, vehicles are a major source of carbon dioxide, the principal greenhouse gas
associated with climate change. Trade in used vehicles raises policy issues at the intersection of
international free trade and global efforts to mitigate the production of greenhouse gas
emissions. For example, as part of the June 2009 American Clean Energy and Security Act (HR
2454), consumers may trade in their old, gas‐guzzling vehicles and receive vouchers worth up
to $4,500 to help pay for new, more fuel efficient cars and trucks.2 The cost effectiveness of this
program hinges on what would have become of these vehicles in the absence of this purchase
program.
We document that international trade between rich and poor countries has acted as a
substitute for an explicit “cash for clunkers” program. Differences in operating costs and
willingness‐to‐pay for quality imply that used vehicles will tend to be traded from high‐income
countries to low‐income countries. Because retirement rates are lower in low‐income countries,
this leads vehicles to be used for many more years than they would have otherwise. In addition
1 Ward’s Automotive Group (1992‐2008), Dargay, Gately, and Sommer (2008).
2 http://energycommerce.house.gov/Press_111/20090505/cashforclunkers.pdf
2
to this scale effect, trade affects the composition of vehicles in all countries. Older vehicles tend
to emit substantially higher levels of pollutants, so trade may have a large impact on
environmental quality in both importing and exporting countries.
Our study focuses on the deregulation of U.S.‐Mexico trade in used cars and trucks
following the North American Free Trade Agreement (NAFTA). In accordance with the
conditions of NAFTA, in August 2005 Mexico issued a decree allowing 10‐15 year‐old vehicles
to be imported from the United States and Canada. This represented a dramatic break from the
previous policy that prohibited entry for all used vehicles except for certain vehicles used in
agriculture. Virtually overnight a vigorous trade flow emerged and we document that between
2005 and 2008 over 2.5 million used vehicles were exported from the United States to Mexico.
To evaluate the environmental consequences of this trade pattern, we assemble the most
comprehensive dataset ever compiled on North American trade in used vehicles and vehicle
emissions. Our dataset allows us to identify, at the vehicle level (using VIN numbers), which
vehicles were traded. The results show that traded vehicles are higher‐emitting per mile than
the stock of vehicles in the United States, but lower‐emitting per mile than the stock of vehicles
in Mexico. As a result, trade has led to a decrease in average vehicle emissions per mile in both
countries.
We also show that vehicles that fail emissions testing in the United States are much more
fail emissions testing twice in California are more than 3 times more likely to be exported. This
provides evidence for the pollution havens hypothesis, the idea that trade liberalization causes
pollution to move to countries with lax environmental standards.
Overall, the analysis indicates that trade has led to an increase in total greenhouse gas
emissions. We show that trade led to no discernible decrease in the number of vehicles in
circulation in the United States, while leading to a considerable increase in the number of
vehicles in Mexico. Over the long‐run this direct effect is exacerbated by differences in vehicle
retirement rates between the two countries. We document that vehicle retirement rates are
3
substantially lower in Mexico and illustrate how this can have a large impact on lifetime carbon
emissions from vehicles.
Our results highlight the importance for policy makers of recognizing the synergies
between carbon mitigation legislation and free trade legislation. Our findings suggest that
unilateral policies in the United States aimed at reducing carbon emissions will increase exports
of fuel‐inefficient vehicles to Mexico, consistent with an emerging literature on carbon leakage.3
If the goal of U.S carbon policy is to minimize global emissions, then such trade effects must be
anticipated.
2 Related Literature
Previous studies of trade and the environment have emphasized how trade affects
where goods are produced. See, e.g., Grossman and Krueger (1993), Copeland and Taylor
(1994), Antweiler, Copeland and Taylor (2001), Copeland and Taylor (2003), and Levinson and
Taylor (2008). Our analysis illustrates that trade also affects where goods are consumed. Trade
affects prices and choice sets and thus changes the type and quality of goods in use, potentially
with serious implications for the environment. With vehicles, how and where they are
consumed is potentially even more important than how and where they are produced. For
example, only 7% of the total lifetime carbon emissions for vehicles come from production and
assembly whereas 93% come from fuel usage.4
pollution (Copeland and Taylor, 1995; McAusland, 2008; Holladay, 2008), but empirical studies
in this area have typically been limited by the lack of available data. Holladay (2008) explains
that, “high quality measures of emissions from consumption rarely exist, making empirical
3 This literature explores the idea that with incomplete regulation carbon reductions in regulated areas may be offset
by increases in unregulated areas. For example, Fowlie (forthcoming) examines the effect of a source‐based carbon
cap and trade program on the California electricity sector, finding substantial scope for leakage to unregulated states.
See also, Goulder, Jacobsen, and van Benthem (2009) that uses simulations to evaluate interactions between state and
federal‐level average vehicle fuel economy standards, showing that emission reductions in states with stringent
requirements are largely offset by increases elsewhere.
4 This estimate is from Massachusetts Institute of Technology Energy Laboratory (2000) based on a total lifetime
driving distance of 300,000 kilometers.
4
work difficult”. Our focus on vehicles is valuable because the emissions from vehicles are well
understood and vehicles have VIN numbers that allow them to be tracked consistently across
countries. Moreover, vehicles are one of the central sources of consumption‐based pollution so
our results are relevant for evaluating consumption‐based pollution more generally.5
Another important feature of our analysis which distinguishes it from much previous
work is our focus on both local and global pollutants. The leading empirical studies in the trade
and environment literature have emphasized effects on local pollutants such as sulfur dioxide
in judging the consequences of trade (see, e.g., Antweiler, Copeland and Taylor 2001 and
Copeland and Taylor 2003). We agree that local pollutants are important, but given the
paramount importance of the issue of climate change it is also important to investigate how free
trade affects this global environmental criterion. With global pollutants the location of
consumption is irrelevant, but the magnitude of lifetime consumption is not. As a result,
policies aimed at reducing greenhouse gas emissions may not achieve aggregate gains when
fuel inefficient durable goods can be traded.
There are also a small number of empirical studies that focus specifically on the
environmental impact of trade within North America. Working prior to NAFTA, Grossman and
Krueger (1993) use a cross‐country regression to show that ambient levels of sulfur dioxide and
particulates increase with per capita GDP at low levels of GDP, suggesting that NAFTA would
improve environmental quality via an overall increase in income. Harbaugh, Levinson, and
Wilson (2002) argue that these results are fragile. Levinson and Taylor (2008) examine the effect
of U.S. environmental regulations on trade with Canada and Mexico, finding a robust
relationship between abatement costs and import levels. For the average industry, a 1% increase
in abatement costs is associated with a .2% increase in net imports from Mexico.
Our study is also relevant to a small existing literature on trade in used durables.
Previous studies provide two complementary explanations for the observed pattern of trade in
which used vehicles tend to be traded from high‐income countries to low‐income countries. Sen
5 In the United States 20.7% of carbon dioxide emissions are derived from motor gasoline. United States Department
of Energy, ʺAnnual Energy Review 2007ʺ, DOE/EIA‐0384 (2008), Table 12.3, Carbon Dioxide Emissions From Energy
Consumption by Sector by Energy Source, 2006.
5
(1962) argues that older durable goods require larger labor inputs, making them relatively more
desirable in countries where wage rates are low. Thoumi (1975), Bond (1983), and Navaretti,
Soloaga, and Takacs (2000) extend this argument to incorporate transportation costs,
heterogeneous buyers and skill constraints. In this operating cost framework, high‐income
countries consume capital‐intensive new goods. As these goods grow older they become labor‐
intensive and thus less desirable in high‐income countries. The trade pattern is sensitive to
wage levels, but also to energy prices, availability of replacement parts, access to specialized
repair equipment, and supply of experienced repair technicians. In addition, environmental
regulations affect operating costs by, for example, requiring costly emissions testing.
An alternative and complementary explanation for trade is non‐homothetic preferences.
If demand for quality is increasing in income then free trade will cause used vehicles to be
exported from high‐income countries to low‐income countries. A growing literature in trade has
recognized the importance of income effects in explaining trade patterns. See, e.g. Flam and
Helpman (1987), Stokey (1991), Murphy and Shleifer (1995), Matsuyama (2000), Fieler (2007)
and Grossman, Fajgelbaum and Helpman (2009). A straightforward implication of non‐
homothetic preferences is that low‐quality goods are relatively more desirable in low‐income
countries. High‐income countries satisfy their demand for high‐quality by purchasing new
goods. As these goods depreciate they are traded to low‐income countries that prefer lower
quality. Overall, the testable implications of the operating cost and non‐homothetic preference
models are very similar.
3 Background: The North American Free Trade Agreement
NAFTA came into effect January 1, 1994, immediately eliminating tariffs on many goods
and establishing a timetable for tariff reductions on many other goods. Restrictions for used
vehicles imports were immediately eliminated in Canada and the United States. Mexico did not
initially eliminate restrictions but agreed to a timetable under which restrictions would be
eliminated in five phases beginning January 1, 2009 and ending, with complete liberalization,
6
January 1, 2019.6 Gearing up to meet these changes, Mexican President Vicente Fox surprised
many industry insiders by accelerating the deregulation process, eliminating trade restrictions
for a large class of used vehicles beginning August 22, 2005. Under the new rules, 10‐15 year old
vehicles could be imported to Mexico from the United States and Canada virtually tax free.
Trade restrictions remained in place for used vehicles less than 10 years old and more than 15
years old. The decision to allow this particular set of vintages was a political compromise. While
recognizing the potentially large gains from trade, vehicles over 15 years old were perceived to
be too high‐emitting to warrant importation. At the same time, by continuing to restrict the
importation of used vehicles less than 10 years old the Mexican government hoped to appease
the powerful Mexican Association of Automobile Distributors (MAAD) and other political
factions within Mexico with a vested interest in new vehicle sales.7
This removal of restrictions marked a substantial change from previous Mexican policies
which allowed used vehicles to be imported only for agricultural use. During the following
three years over 2.5 million used vehicles were imported by Mexico from the United States and
Canada. This represents a small fraction of the vehicle stock in the United States (232 million in
2005 according to R.L. Polk) and Canada (19 million in 2005 according to Ward’s Motor Vehicle
Data 2006), but represents a substantial fraction of the vehicle stock in Mexico (22 million in
2005 according to INEGI). The used car import market that has evolved in response to this
policy change is highly decentralized, with thousands of car dealers and private citizens
bringing vehicles into Mexico.8
This pattern of trade was foreshadowed by Berry, Grilli and López‐de‐Silanes (1993), “If
free trade in used cars is permitted, the relatively poor Mexican consumers would become a
6 Our study is germane to a small literature that examines the determinants of trade restrictions for used vehicles.
Pelletiere and Reinert (2004) find that among 132 countries for which data are available, 74 have some kind of
restrictions on used vehicle imports. Using cross‐country regressions, Pelletiere and Reinert (2002) and Pelletiere and
Reinert (2004) find that the most significant factor determining used automobile protection is the presence of
domestic automobile production.
7 Vehicles exceeding 4,500 kilograms (e.g. buses and semi trucks) are not eligible. Diesel vehicles are eligible, but less
than 1/5th of 1 percent of all imports during this period were diesel vehicles.
8 Although Central America is a large market for used vehicles from the United States, very few of these vehicles
were likely re‐exported to Central America. Vehicles headed from the United States to Central America can bypass
the expense of legalizing the vehicle in Mexico by shipping the vehicle by boat, or by driving through Mexico with a
temporary permit.
7
major source of demand for used cars from the United States and Canada. This would
substantially drive up the price of used cars and lead wealthier consumers (in all countries) to
trade in their old cars more frequently. In this case a more complicated trading pattern might
emerge, with the increase in North American demand for new cars coming largely from U.S.
and Canadian consumers, while a large portion of Mexican demand is satisfied by used cars.”
While these changes were occurring in the used car market, trade policy for new
vehicles did not change.9 Since 1994, NAFTA has allowed duty‐free trade of new vehicles and
new vehicle parts within North America as long as certain content restrictions are met. In 2005,
the most recent year for which data are available, Mexico exported 506,000 new vehicles to the
United States and 49,000 vehicles to Canada.10 Since 1994, all new vehicles sold in Mexico must
meet U.S. emissions standards.
In March 2008, Mexico’s trade policy changed again, when trade restrictions were
reinstated for all 11‐15 year old vehicles. Thus, after the change the only used vehicles that
could be imported were vehicles that were exactly 10 years old. At the same time, the
government increased the tariff on used vehicles entering Mexico from 3% to 15%. This return
to trade restrictions was a political response to pressure from MAAD who pointed to alleged
decreased sales of new vehicles and argued vigorously for trade restrictions, claiming that the
change in policy was needed to, “stop the accelerated conversion of our country into the
world’s biggest automotive garbage dump.”11
It is worth noting that there is much precedent for high‐income countries exporting used
vehicles to lower‐income trading partners. Japan exports vehicles to over 100 different
countries, South Korea exports vehicles to Vietnam and Russia, and the United Kingdom
9 In the short‐run there is limited scope for producers of new durable goods to respond to competition from used
goods. For example, trade in used durable goods may change incentives for producers by altering the market
structure. As first articulated by Coase (1972) and more recently analyzed by Esteban and Shum (2007), the market
for used durable goods limits market power for producers. A trade pattern in which used durable goods are traded
from high‐income to low‐income countries will increase the ability of producers in the high‐income country to
exercise market power, and decrease the ability of producers in the low‐income country to exercise market power.
10 See INEGI, Industria Automotriz en México, 2007, Table 3.3.2, Volumen de Las Exportaciones de Automóviles y Camiones
por Continente y País de Destino.
11 Associated Press, “Mexico Bans Imports of Most Used Cars” (March 2, 2008) reports that this policy change has
caused a surge in prices for vehicles that are exactly 10 years old. In 2008 dealers in South Texas reported $500‐$800
premiums for 1998 model vehicles.
8
exports vehicles to Ireland.12 Another prominent example of a vehicle importing country is
Peru, where over 80% of the vehicles in circulation were imported as used vehicles from the
United States and Japan.13 Although these accounts suggest that the total volume of
international trade in used vehicles is large, we are not aware of any comprehensive measures
of global trade. For example, WTO (2007) tracks “automobile products” but does not
distinguish between new and used vehicles.
4 Empirical Analysis: Scale and Composition
In this section we examine the environmental consequences of used vehicle trade
between the United States and Mexico during the period 2005‐2008. Following the existing trade
and environment literature, we distinguish between scale and composition effects, recasting
these mechanisms to focus on trade in used vehicles. In section 4.1, we focus on scale. The scale
effect of trade is the change in pollution due to the overall change in economic activity, or in our
case, the total number of vehicles on the road. In section 4.2, we characterize composition. The
composition effect of trade is the change in pollution due to adjustments in the mix of goods. In
our case, composition will be assessed by comparing the average emissions of traded vehicles to
the average emissions of the stock of vehicles in both countries. In section 4.3 we test for
differences in emissions levels at the vehicle‐level. Our dataset allows us to identify, at the
vehicle level (using VIN numbers), which vehicles were traded. This degree of detail is crucial
because, for example, it allows us to test for whether vehicles that fail emissions testing in the
United States are more likely to be traded.
Our approach emphasizes documenting changes in quantities rather than changes in
prices. The previous studies of trade in durable goods described in Section 2 imply that free
12 According to the Wall Street Journal, “Driving Change: How Japan’s Second‐Hand Cars Make Their Way to Third
World” (January 8, 2004), Japan exports one million used vehicles annually to countries all over Asia, Africa, and the
Middle East. According to The Korea Herald, “Exports of Used Cars Hit Record in 2007” (April 17, 2008), South Korea
exported over 220,000 used cars and trucks in 2007. See also The Irish Times, “Used Car Imports from UK hit 50,000 a
Year” (July 4, 2007).
13 See Centro de Investigación y de Asesoría del Transporte Terrestre del Perú, “Cuarto Informe de Observación Pública:
Externalidades Negativas Generadas por la Importación de Vehículos Usados Sobre la Salud y la Vida de la
Población en Perú”, April 2005, for details.
9
trade will cause prices for used cars to increase in the United States and decrease in Mexico.
Examining prices would be a valuable test of these models, and would provide indirect
evidence about trade flows. Nonetheless, for several reasons we chose not to pursue this
approach. First, the U.S. used car market is very large, so we did not expect to find large price
effects. Moreover, identification of price changes would be based primarily on changes in prices
over time for vehicles of different vintages, and thus could be difficult to separate from other
time‐varying factors that affect prices. Second, no publicly‐available data exist for used car
prices in Mexico.14 Third, our objective is to document changes in emissions of local and global
pollutants brought about by international trade. Given our focus, it is crucial to track quantities.
4.1 The Effect of Trade Deregulation on Scale
Tables 1 and 2 describe the vehicles exported from the United States to Mexico between
November 2005 and July 2008. These data were collected by the Mexican Customs Agency, a
branch of the Mexican Ministry of Finance and represent the universe of vehicles that were
imported from the United States during this period. Table 1 describes vehicles by year traded,
vintage and vehicle manufacturer. Table 2 describes the top 10 most traded models.15 The
customs records describe in detail the substantial flow of used vehicles from the United States
to Mexico during this period. Overnight the flow of used vehicles from the United States to
Mexico changed from virtually zero to over 75,000 vehicles per month. Overall, 2.45 million
vehicles were traded during the almost three year period.
Newer vehicles are heavily represented among traded vehicles. Vehicles 10‐15 years old
were eligible to be traded but 10‐11 year old vehicles were much more often traded than 14‐15
14 In a different context, Davis (2008) examines used car prices in newspaper advertisements from Mexico City.
However, advertised prices may be a poor proxy for transaction prices. Davis also finds that driving restrictions in
the Mexican capital led to an increase in the number of vehicles in circulation, as drivers acquired additional vehicles
to circumvent the restrictions. Nonetheless, using the Mexican Public Vehicle Registry, http://www.repuve.gob.mx/,
we find that only approximately 7% of the vehicles imported from the United States are registered in Mexico City,
despite the fact that Mexico City has 12% of all registered vehicles nationwide, consistent with Mexico City’s
emissions testing program serving as an effective deterrent for these older vehicles.
15 The most commonly traded vehicle model is the Ford Explorer, a vehicle that fell largely out of favor with U.S.
consumers after a highly‐publicized recall of Firestone tires in August 2000 and claims that the large number of tread
separations observed with these tires might be related to vehicle design. See Krueger and Mas (2004) for details.
10
year old vehicles. This may reflect the travel and administrative costs associated with importing
a vehicle.16 The fact that relatively newer vehicles are heavily represented provides suggestive
evidence about what trade in vehicles would look like under different trade policies. According
to NAFTA, the North American market for used vehicles will be completely deregulated by
2019. The observed pattern suggests that the policy might lead to substantial trade in vehicles
less than 10 years old.
Another striking feature of these data is the prevalence of trucks and other large
vehicles. There are several explanations for this pattern. First, this reflects the fact that only U.S.
and Canadian‐produced vehicles were eligible to be imported. In Section 4.3 we compare traded
vehicles to the stock of eligible vehicles in the United States, finding evidence that this is indeed
an important explanation. In addition, gasoline prices in the United States were high during the
period 2005‐2008, making these fuel‐inefficient vehicles relatively desirable in Mexico where
gasoline prices are set by PEMEX, the national petroleum company at less than $2.50 per
gallon.17 Moreover, many of the imported vehicles are destined for rural areas within Mexico
where larger vehicles are disproportionately valued because they can be used on rough roads.
4.2 The Effect of Trade Deregulation on Fleet Composition
Although scale is clearly important, the environmental impact of trade also depends
critically on the type of vehicles that are traded. In this subsection we compare the emissions
characteristics of traded vehicles with the emissions characteristics of the stock of vehicles in the
United States and Mexico. For the fleet in the United States, we obtained data from R.L. Polk
that describe the distribution of registered vehicles in the United States as of 2005 by
manufacturer and vintage. For the fleet in Mexico, we filed the equivalent of a Freedom of
Information Act request with the Mexican Ministry of Public Safety via the Mexican Federal
16 The Alchian‐Allen, or “Shipping the Good Apples Out” theorem points out that fixed transportation costs decrease
the relative price of high‐quality goods in importing countries, causing consumption to shift toward these goods. See
Alchian and Allen (1964) and Borcherding and Silberberg (1978) for details.
17 According to the Mexican Energy Information System, the average price per gallon of regular unleaded gasoline in
Mexico was $2.11 in 2005, $2.39 in 2006, $2.40 in 2007, and $2.49 in 2008 compared to $2.31, $2.62, $2.84, and $3.30 for
the United States according to the Department of Energy, Energy Information Administration.
11
Institute for Access to Public Information. 18 In response to our request, we were provided with
a document describing the distribution of vehicles in Mexico by state, vintage, and
manufacturer as of 2008.19 In this subsection we describe the stock by vintage and manufacturer,
rather than by vintage and model, because data for Mexico at the model level are not available.
The analysis in section 4.3 adds model fixed effects for comparing traded vehicles to the stock in
selected states in the United States.
Vehicle level data on emissions and vehicle attributes come from the California and
Illinois Environmental Protection Agencies. These data describe emissions results for 7.2 million
vehicles that were tested in 2005 under California’s Smog Check program.20 For Illinois, these
data describe emissions results for 835,000 vehicles that were tested in 2005. We use these
records to estimate average emissions levels for vehicles of different manufacturers and
vintages. For vehicles which were tested multiple times, we use data on the first vehicle
emissions test for each vehicle. In Section 4.3 we return to these multiple tests and are able to
test directly if traded vehicles are more likely to have failed emissions testing.
Let denote the average emissions level among all vehicles from manufacturer j and
vintage t. Vehicles produced in 1976 or before are grouped together, and vehicles produced
after 2005 are grouped together, so the set of all vintages, T, includes 1976‐2005. We focus on all
manufacturers for which there were more than 1 million total registered vehicles in the United
States as of 2005. Other vehicle manufacturers are included in an “other” category.21 For the
results reported below, we calculated the vector using the emissions data from California,
the largest and arguably the most reliable set of emissions measures. We have also examined
18 Although vehicle registration is required in all parts of Mexico, registries are maintained at the state level and the
Mexican National Statistics Institute compiles total vehicle counts by state and year, but not by vintage or
manufacturer. Moreover, since 2006 the Mexican National Statistics Institute has measured vehicle counts by adding
vehicle sales to the totals for 2005 rather than through direct measurement of vehicle registries, making these data
inappropriate for examining vehicle retirement patterns in Mexico.
19 The Mexican Ministry of Public Safety compiled this national database for the first time in 2008. In the results
which follow, we find that on average traded vehicles are lower emitting than the stock of vehicles in Mexico. Had
we been able to obtain data from 2005 presumably this result would have been reinforced because the emissions
levels for 2006, 2007, and 2008 vintages tend to be lower than the emissions levels for older vehicles.
20 We have also examined emissions testing data from the state of Illinois and the results are similar.
21 Our manufacturers include Acura, BMW, Buick, Cadillac, Chevrolet, Chrysler, Dodge, Ford, GMC, Honda,
Hyundai, Infiniti, Isuzu, Jeep, Kia, Lexus, Lincoln, Mazda, Mercedes‐Benz, Mercury, Mitsubishi, Nissan/Datsun,
Oldsmobile, Plymouth, Pontiac, Saturn, Toyota, Volkswagen, and Volvo.
12
results based on emissions factors calculated using data from the Mexico City Emissions Testing
Program and results are qualitatively similar, providing evidence that our results are not driven
by selection of a particular set of emission factors.
Average emissions are calculated by multiplying emissions factors by the proportion of
vehicles of each type,
Here denotes the proportion of traded vehicles of each manufacturer and vintage,
the United States and Mexico.
Table 3 presents the composition results. The table reports mean emission levels for
hydrocarbons, carbon monoxide and nitrogen oxide, as well as vehicle vintage. Compared to
the stock of vehicles in the United States, traded vehicles emit higher levels of all three local
pollutants. The differences are substantial, ranging from 6% for carbon monoxide to 20% for
nitrogen oxide. Compared to the stock of vehicles in Mexico, traded vehicles emit lower levels
of all three local pollutants. Again the differences are substantial, ranging from approximately
3% for nitrogen oxide to 38% for carbon monoxide. Overall, these results imply that NAFTA led
to a decrease in average emissions of local pollutants per mile in both countries. The table
indicates a clear ordering of vehicles by age. The stock in the United States is newer than the
stock of vehicles in Mexico, with traded vehicles in between.22 Based on the hedonic price for
used vehicles results of Bin and Martins‐Filho (2008), it is clear that the U.S is exporting lower
value vehicles to Mexico.
Older vehicles tend to emit higher levels of emissions because of both vintage effects and age effects, though
22
engineering studies (e.g. Bishop and Stedman, 2008) have tended to find that vintage effects are more important. This
suggests that these differences in local emission levels are likely to be persistent over time.
13
The table also reports results for miles‐per‐gallon, vehicle weight, number of cylinders,
and engine size.23 These measures directly or indirectly measure vehicle fuel‐efficiency and,
therefore, carbon dioxide emissions which are proportional to total gasoline consumption.
Traded vehicles are marginally less fuel efficient on average than the stock of vehicles in both
countries. As a result, trade increases fuel efficiency in the United States, and decreases fuel
efficiency in Mexico City, though the differences are small. Whereas local emissions vary across
columns by as much as 30‐40%, the differences in miles‐per‐gallon vary by less than 3%,
indicating that composition effects are likely to play a much smaller role for carbon dioxide
emissions.
4.3 Exported Vehicles are High Emitting
The previous subsection illustrated that the vehicles exported from the United States to
Mexico were higher‐emitting on average than the stock of vehicles in the United States and
lower‐emitting than the stock in Mexico. Our data, however, allow us to refine the analysis
further. For each vehicle that was emissions tested in California and Illinois in the year 2005, we
know the vehicle identification number (VIN). By merging these records with the customs
records of traded vehicles, we were able to identify the subset of vehicles that were
subsequently exported to Mexico. 24
In this section we test hypotheses concerning differential emission levels between
vehicles that were exported to Mexico and vehicles that were not. In particular, we run
regressions controlling for vintage, manufacturer, and model fixed effects,
y 1 δ ω σ ε .
23 We imputed miles‐per‐gallon for each vehicle using vintage, vehicle weight, cylinders, and engine size using data
from the Department of Transportation, National Highway Traffic Safety Administration. For each vintage 1978‐2005
we estimated a separate regression of miles‐per‐gallon on weight, cylinders and engine size and then used the
estimated coefficients to predict miles‐per‐gallon for each vehicle.
24 Out of 3.6 million vehicles in the California data for the relevant vintages, 83,057 vehicles were exported. This
represents only 3.3% of traded vehicles, while nearly 15% of vehicles registered in the United States are from
California. This would seem low given the proximity to Mexico. However, emissions testing in California is required
only every other year. Moreover, 24 out of 58 counties in California do not require emissions testing, and an
additional six counties require emissions testing only within certain zip codes. Finally, there may be some vehicles
that were intended to be exported and thus not emissions tested in 2005. We believe that these vehicles will tend to be
even higher‐emitting than the vehicles that were tested.
14
We estimate a variety of specifications using as the dependent variable the same
measures of emissions and vehicle characteristics used in the previous subsection. In all
specifications, the coefficient of interest corresponds to an indicator variable for whether the
vehicle was subsequently exported to Mexico. We report results from specifications that control
for vintage indicators, δ , as well as specifications that control for manufacturer/vintage
describes how y varies compared to other vehicles of the same model and vintage.
Table 4 reports coefficients and standard errors based on 36 separate OLS regressions.
Across pollutants and specifications we find evidence that the vehicles exported to Mexico have
higher levels of emissions of local pollutants. Columns (3) and (7) report results with the full set
of model/vintage interactions. Compared to other vehicles of the same model and vintage,
traded vehicles have emissions levels that are significantly higher than vehicles not sent to
Mexico. The effect remains in column (4) after controlling for a quartic in mileage, indicating
higher emissions for exported vehicles even within vehicles of the same vintage, make, model,
and mileage. We also examine odometer readings as a separate dependent variable. Controlling
for model/vintage interactions, vehicles exported to Mexico have on average almost 10,000 more
miles than other vehicles.
The table also reports coefficients corresponding to miles‐per‐gallon and other vehicle
characteristics predictive of carbon dioxide emissions. Controlling for vintage, traded vehicles
are less fuel efficient, heavier, with more cylinders, and larger engines than the stock of vehicles
in California. This is consistent with the pattern observed in table 2 that minivans, SUVs, and
trucks are heavily represented among traded vehicles. However, after controlling for
manufacturer/vintage the coefficients become much smaller. Only U.S. and Canadian‐produced
vehicles were eligible to be imported and the smaller coefficients indicate that these restrictions
are important for explaining the prevalence of minivans, SUVs, and trucks among the list of
25 Model is measured using the 5th digit of the VIN number as assigned by each individual manufacturer. This
classification of model distinguishes between, for example, the Ford Windstar (a minivan) and the Ford F‐Series (a
truck), but does not distinguish, for example, between the Ford F‐150 and the Ford F‐250.
15
most‐traded vehicle models. We do not report coefficients controlling for model/vintage
because there is essentially no variation within vehicle model in these characteristics.
Table 5 presents coefficient estimates from additional regressions using as the
dependent variable whether or not the vehicle has failed emissions testing. Vehicles that emit
extremely high levels of pollutants are particularly important for the environment because it
has been shown that these vehicles contribute a large proportion of total emissions. In these
regressions the dependent variable is an indicator variable for whether the vehicle failed
emissions once, twice, and three or more times, as well as whether the vehicle was a “gross
polluter” once, twice, and three or more times. According to California law, a “gross polluter” is
a vehicle that exceeds twice the allowable emissions for at least one pollutant. The results
indicate that exported vehicles are significantly more likely to be super‐emitters. For example,
after controlling for model/vintage fixed effects, exported vehicles are over four times more
likely to have failed emissions testing three times, and over six times more likely to have been
declared a gross polluter three or more times. This pattern could be evidence of differential
asymmetric information, with Mexican buyers less able to identify high‐emitting “lemons”.
Overall the results of these vehicle‐level regressions imply that exports are “browner”
than the average stock in the United States. We conclude based on the California and Illinois
samples that exported vehicles tend to be significantly higher‐emitting than other vehicles. This
effect remains after controlling for vintage, manufacturer, and model.
5 The Behavioral Response: New Vehicle Sales and Vehicle Retirement
In this section we examine vehicle retirement in both countries and new vehicle sales in
Mexico. Adjustments in retirement and purchase behavior are important because they may
mitigate the direct environmental impact of trade. This section has three main results. First,
there is no evidence of a change in the number of vehicles in circulation in the United States,
suggesting that most of the vehicles exported to Mexico are vehicles that would have been
scrapped otherwise. Second, vehicle retirement rates in Mexico are lower than vehicle
retirement rates in the United States so exporting a vehicle to Mexico tends to increase the total
16
number of years the vehicle will be used. Third, there is no evidence that the increased
availability of used vehicles has decreased sales of new vehicles in Mexico.
5.1 Vehicle Retirement in the United States
Figure 4 plots vehicle exit rates by age in the United States before and after NAFTA.26
Annual exit rates increase with vehicle age from less than 2% for vehicles that are 4‐5 years old
to more than 10% for vehicles over 15 years old. The figure reveals modest increases in exit rates
after NAFTA for 10‐15 year old vehicles. Five out of the six differences are statistically
significant at the 1% level. However, the increases imply only approximately 180,000 additional
exits annually among 10‐15 year old vehicles relative to the earlier period.27 This is considerably
smaller than the 1 million vehicles per year that were exported annually to Mexico during this
period. This suggests that most of the vehicles that were exported to Mexico were either
vehicles that would have been retired otherwise or vehicles that were already retired. This is
consistent with the evidence from section 4.3 that showed that very low‐quality vehicles are
more likely than other vehicles to be traded. 28
26 These data were obtained from R.L. Polk & Company in April 2008. Exit rates were calculated by calendar year and
vehicle age as follows. Let v denote vehicle vintage (e.g. 1991) and let t denote calendar year. We calculated the exit
rate for vehicles age t‐v for year t, as the percentage change in the number of vehicles of vintage v between year t‐1
and t.
27 These data measure exits with a lag because vehicles are registered only once a year. Results are similar (315,000
additional exits), however, when we instead define the post‐NAFTA period as October 2006 to October 2007, which
includes exits from October 2006 to October 2007 and delayed exits from October 2005 to October 2006.
28 Michael Wilson, Executive Vice President of the Automotive Recyclers Association, confirmed that since 2005 fewer
vehicles are reaching scrapyards in the United States for recycling. There has been a “significant impact” on their
members with the largest effects on facilities in the Southwest. With over 8000 vehicle recycling facilities in the
United States, it is difficult to obtain comprehensive information about the decrease in volume of vehicles processed.
We examined and did not find evidence of increases in scrap steel prices from the Bureau of Labor Statistics but this
is perhaps not surprising because according to the Steel Recycling Institute vehicles represent only 14 million tons of
recycled steel annually compared to a market of 83 million tons.
17
Figure 4: Exit Rates by Vehicle Age in the United States
.15
.1
.05
0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Before NAFTA (10/03 - 10/05)
After NAFTA (10/05 - 10/07)
Source: R.L. Polk & Company.
The fact that NAFTA appears to have had little effect on the number of registered
vehicles in the United States is perhaps not surprising. The stock of used vehicles in the United
States is large, so increased demand for used vehicles is unlikely to have raised used vehicle
prices more than a few hundred dollars per vehicle. Moreover, capital costs are only one part of
the costs of operating a vehicle (e.g. maintenance, insurance, gasoline, etc) and in the United
States the elasticity of vehicle ownership with respect to these costs is likely very low. In short,
it is hard to imagine trade leading a large number of households in the United States to reduce
the number of vehicles they own.
These results highlight the immense size of the used vehicle market in the United States.
During the period 2003‐2007, 5.9 million vehicles exited vehicle registries annually in the United
States, 2.1 million vehicles 10‐15 years old. This provides an enormous potential stock of
vehicles for export. A relatively small fraction of these castoffs can represent a large number of
vehicles for a smaller country like Mexico.
18
5.2 Differential Vehicle Retirement Rates in the United States and Mexico
Figure 5 plots the age distribution of vehicles in the United States and Mexico. In
Mexico, used vehicles are relatively expensive and vehicle repairs are relatively inexpensive, so
vehicle owners tend to hold on to vehicles longer. In the United States, used vehicles are
relatively inexpensive and vehicle repairs are relatively expensive, so when faced with repair
costs vehicle owners tend to replace rather than repair. Overall, the average age of vehicles in
the United States is 9.4 years compared to 13.5 years in Mexico and the implied mean annual
retirement rate for 10‐30 year old vehicles is 12.2% in the United States and only 3.8% in Mexico.
Lower retirement rates in Mexico mean that when a vehicle is exported from the United States
to Mexico it will tend to be used for many more years than it would have otherwise.
Figure 5: The Age Distribution of Vehicles
1
.8
.6
.4
.2
0
0 10 20 30
Age
5.3 Vehicle Sales in Mexico
The increased availability of relatively high‐quality used vehicles may cause households
in Mexico to substitute away from new vehicles. Indeed, MAAD has been a vocal opponent to
19
liberalization, and claims that NAFTA has reduced sales considerably.29 In this subsection we
examine the evidence from new vehicle sales.30 Overall, we find no evidence of decreased sales.
The one possible exception is subcompact car sales, which indeed decrease during the post‐
NAFTA period, though sales appear to be in decline even prior to trade deregulation. Overall,
the evidence from new vehicle sales suggests that the 10‐15 year old used vehicles were not an
attractive alternative for most buyers of new vehicles.
Figure 6 plots monthly sales of new vehicles in Mexico during the period January 2001
to December 2007, as well as a fitted cubic polynomial in time with intercept at August 2005
when the market for used vehicles was liberalized. Based on the plot it is difficult to make
strong statements about the impact of trade deregulation on sales. Overall, vehicle sales post‐
NAFTA are similar to sales pre‐NAFTA. There is a pronounced season pattern to sales, which
will be taken into account in the regression analysis which follows.
29 Recent publications from the Mexican Association of Automobile Distributors describe trade in used vehicles as an
environmental and commercial threat. See, e.g., “Trade in Very Old Used Cars: A Challenge for Air Quality” (“El
Comercio de Autos Usados de Gran Antigüedad: Un Reto a la Calidad del Aire en las Cuencas Atmosféricas de la Frontera de
EUA y México”), August 2007. Our study is the first full‐scale attempt to measure empirically the impact of NAFTA
on new vehicle sales.
30 Of course, a decrease in vehicle demand could have decreased either prices or quantities. Data on prices are not
available, however, so we focus on quantities. A decrease in vehicle demand will reduce sales unless supply is
perfectly inelastic.
20
Figure 6. Monthly Sales of New Vehicles in Mexico
140 120
Monthly Sales (in thousands)
60 80 40
20 100
Figure 7 plots monthly sales of new vehicles in Mexico by vehicle category. Whereas
sales of subcompacts decrease steadily after August 2005, sales for compacts, luxury cars, light
trucks and SUVs increase. Table 6 reports coefficients and standard errors from an analogous
regression analysis. Monthly sales of new vehicles in Mexico (in logs) are regressed on month
indicator variables, a polynomial time trend, and an indicator for after NAFTA. To control for
aggregate changes in demand, we also include the growth rate of GDP (available quarterly)
from the Mexican National Statistics Institute, National System of Accounts. For second, third,
and fourth order polynomials, the coefficient on NAFTA for total sales is near zero and not
statistically significant. The null hypothesis of a 10% drop in total vehicle sales can be rejected at
the 2% level across vehicle categories and specifications. For subcompacts the coefficient is
negative and near 5%, but not statistically significant. Compact and luxury car sales increase in
all specifications. Overall, the vehicle sales data provide little evidence of substitution away
from new vehicles.
21
Figure 7. Monthly Sales of New Vehicles in Mexico by Vehicle Category
0 10 20 30 40
2001 2002 2003 2004 2005 2006 2007 2008 2001 2002 2003 2004 2005 2006 2007 2008
0 10 20 30 40 50 60
0 2 4 6
2001 2002 2003 2004 2005 2006 2007 2008 2001 2002 2003 2004 2005 2006 2007 2008
In future work it would be valuable to examine trade’s general equilibrium effects based
on structural estimates of household‐level vehicle demand. See, e.g. Berry, Levinsohn, Pakes
(1995), Goldberg (1995), and Bento, Goulder, Jacobsen, and von Haefen (2008). In related work,
Clerides (2008) uses a general equilibrium model of the automobile market to establish the gains
from trade from a 1993 law that relaxed vehicle import restrictions in Cyprus, finding that
welfare gains averaged $2000 per purchaser. One potential challenge of estimating such a model
in this context is the difficulty of acquiring necessary Mexican data. Structural models of vehicle
demand in the United States have been successful, in part, because researchers have access to
detailed micro‐level datasets like the National Household Travel Survey (NHTS) and the
Consumer Expenditure Survey (CES). There is no dataset in Mexico comparable to the NHTS.
The Mexico equivalent to the CES, the National Survey of Household Income and Expenditures
includes basic household demographics and information about expenditures on vehicles,
vehicle maintenance, vehicle insurance, and vehicle licensing and fees, but does not include the
manufacturer, model, or vintage of household vehicles.
22
6 An Illustration of How Trade Can Increase Lifetime Carbon Emissions
In this section we evaluate the implications of international trade in used vehicles
between the United States and Mexico for carbon dioxide emissions. Although our analysis
provides much of the information required for these calculations, these results do require us to
make a number of additional assumptions, particularly about vehicle utilization. As a result, we
refer to these results as an “illustration” and these results should be interpreted with caution.
We consider short‐run and long‐run thought experiments. For the short‐run, we
calculate the annual increase in global carbon emissions associated with Mexico importing a
vehicle from the United States. Short‐run carbon emissions per vehicle are the product of the
number of miles driven annually and emissions per mile:
.
For the long‐run, we calculate the change in lifetime emissions per vehicle,
where λ is the vehicle retirement rate. The long‐run cumulative impact of trade is particularly
important in this context because vehicle retirement rates in Mexico are low, so vehicles tend to
be used for many years. For λ we use .038, the average annual retirement rate for vehicles 10‐30
years from section 5.2. In addition, we assume that the maximum lifespan for a vehicle after
entering Mexico is 20 years. For miles driven, we use the average number of miles driven
annually per vehicle in each country. In Mexico the average vehicle miles travelled (VMT)
annually is 6,100 miles annually compared to 12,400 miles in the United States.31
Table 7 reports the results. Carbon dioxide emissions increase by 2.6 tons per vehicle in
the short‐run (1 year) and 36.5 tons per vehicle in the long‐run (20 years). Using $35 per ton
31 U.S. Department of Transportation, Federal Highway Administration, ʺHighway Statistics
2006ʺ, Section V: Roadway Extent, Characteristics, and Performance, Table VM‐1. No analogous survey‐based
statistic is available for Mexico. However, annual gasoline consumption indicates that vehicles tend to be used less
intensively. Gasoline consumption in Mexico in 2007 totaled 8.6 billion gallons (SIE, 2008). Using average miles per
gallon from Table 3, this implies that the average vehicle in Mexico travels 6,100 miles annually. Vehicles are used
less intensively in Mexico for many reasons including lower income levels, lower quality roads and highways, and
differences in commuting patterns.
23
carbon ($9.9 per ton of carbon dioxide) and a 1.5% annual discount rate following Nordhaus
(2009) these emissions have a social cost of $24 (short‐run) and $308 (long‐run) per vehicle. For
the 2.5 million vehicles exported from the United States to Mexico during the period 2005‐2008
total lifetime social costs would be $770 million.
These results provide a valuable preliminary assessment of the impact of U.S.‐Mexico
trade in used vehicles on carbon dioxide emissions. However, it is important to emphasize that
these results are based on several assumptions, many of which can be only partially verified
empirically. For example, we are assuming that the Mexican vehicle retirement rate stays at .038
despite the influx of used vehicles from the United States. Table 7 reports that if instead the
vehicle retirement rate were to double, long‐run emissions would decrease considerably. In
addition, we have assumed that imported vehicles are driven at the country average 6,100 miles
annually. We would have preferred instead to use measures of miles driven that vary by vehicle
vintage, but no vehicle‐level information on miles driven in Mexico is available. Table 7 shows
that if instead VMT doubled the long‐run effects would be considerably larger. Finally, we are
assuming implicitly that prior to driving an imported used vehicle, Mexicans were generating
no carbon dioxide from transportation. For households substituting from high‐occupancy
public transportation, this is a reasonably accurate approximation. However, for households
substituting away from low‐occupancy public transportation or from shared private vehicles,
there will be an offsetting decrease in vehicle emissions. The final column of Table 7 reports
results assuming that prior to driving an imported vehicle Mexican drivers were riding in a
minibus with average occupancy 8 riders (compared to 1.6 for private vehicles) and fuel
economy one‐half of that for private vehicles.32
It is worth highlighting that these results describe the social cost from carbon dioxide
emissions, but not the social cost from emissions of local pollutants. There is reason to believe
that the social cost from increased emissions of local pollutants could be large. For example,
World Bank (2002) finds that the annual benefits of a 10 percent reduction in ozone and
Average occupancy and fuel economy for minibuses come from International Association of Public Transport,
32
“Millenium Cities Database for Sustainable Transport”, 2007, Brussels and are for the Mexico City metropolitan area.
No comparable numbers are available for Mexico as a whole.
24
particulates in Mexico City would be approximately $882 million (in 2006 U.S. dollars). A more
comprehensive analysis of the social costs of trade in used vehicles would track vehicles after
they enter Mexico, model the relationship between emissions and ambient pollution levels, and
calculate the social costs of the resulting increases. In such an analysis the magnitude of social
costs would depends not only on the level of emissions, but also on the location of emissions.
One of the important themes in a recent literature in the atmospheric sciences is that marginal
damages from emissions can vary dramatically across locations. See, e.g., Mauzerall, et al., 2005.
Thus, a comprehensive analysis of the social costs from local pollution would need to measure
not only the change in total emissions, but also model the relationship between emissions and
ambient pollution levels, and examine changes in levels across particular locations.
7 Conclusion
Wealthy nations demand a range of high‐quality transportation equipment (cars, trucks,
trains, buses, boats, and planes), as well as residential durable goods (air conditioners, clothes
washers), commercial durables (computers, lighting, heating and cooling equipment) and
machinery). These durable goods depreciate in quality over time. Poorer nations want to
purchase similar durable goods but due to income effects desire lower quality. From a societal
perspective, there are gains to trade from shipping used durable goods from rich countries to
poorer countries.
In this paper we argued that this pattern has large implications for the environment. As
we discuss in the paper, the effect of trade on emissions of local and global pollutants depends
on the relative magnitude of scale and composition effects. Trade provides rich countries with
an outlet for low‐quality durable goods. Because low‐quality durables are typically also high‐
emitting, this tends to decrease average and total emissions in rich countries. In poor countries,
trade increases the number of durable goods, but may also improve the quality of the stock.
Whether or not this change in composition is large enough to offset the scale effect depends on
the characteristics of the initial stock of durable goods and other factors.
25
In the empirical analysis we focused on the deregulation of the market for used cars and
trucks following NAFTA. Scale effects are immediate and large in magnitude, with millions of
vehicles exported from the United States to Mexico during the first years of trade. Composition
effects are also large. For local pollutants, traded vehicles have significantly higher emissions
per mile than the stock in the United States, and significantly lower emissions per mile than the
stock in Mexico. As a result, trade decreases average emissions of local pollutants per mile in
both countries. For carbon dioxide emissions, composition effects are much smaller, with traded
vehicles marginally less fuel‐efficient than the stock in either country. Total emission levels
decrease in the United States and increase in Mexico, with global carbon dioxide emissions
increasing. Furthermore, because vehicle retirement rates are low in Mexico, these vehicles will
continue to be used for many years. This tendency of trade to increase durable good life spans is
likely to be the case not just for the United States and Mexico, but for international trade more
generally.
The broader conclusion of our paper is that policymakers in conducting economic
analysis of environmental policies ought to take careful account of the implications of their
policies for trade in used durables. This is particularly important for global pollutants such as
carbon dioxide where domestic policies designed to reduce emissions can be easily undermined
by emissions increases abroad. As a result, unilateral policies aimed at reducing greenhouse gas
emissions may not achieve aggregate gains when durable goods can be traded. For example,
any policy that increases the cost of gasoline in the United States will likely increase exports of
used vehicles, leading to increased emissions in importing countries. In contrast, increasing fuel
economy standards for new vehicles or offering “cash for clunkers” (Blinder, 2008) would
increase the price of used vehicles in the United States, decreasing trade.33 Thus, the broader
conclusion of our analysis is that the composition of trade can change sharply as U.S policy
incentives change.
As articulated by Gruenspecht (1982), new vehicle regulations such as fuel economy standards increase
33
the cost of new vehicles, causing households to delay new car purchases. This would increase demand for
used cars in the United States, bidding up the price, and leading fewer used cars to be shipped to Mexico.
26
References
Alchian, Armen A. and Allen William R. University Economics. Belmont, Calif.: Wadsworth, 1964.
Antweiler, W., Copeland, B. and Taylor, S. “Is Free Trade Good for the Environment?”American
Economic Review, 2001, 91 (4), 877‐908.
Bento, Antonio M., Larry Goulder, Mark R. Jacobsen, and Roger H. von Haefen. “Distributional and
Efficiency Impacts of Increased U.S. Gasoline Taxes,” American Economic Review, 2008,
forthcoming.
Berry, Steven, Vittorio Grilli and Florencio López‐de‐Silanes. “The Automobile Industry and the
Mexico‐U.S. Free Trade Agreement,” in P. M. Garber, ed., The Mexico‐U.S. Free Trade Agreement,
Cambridge, MA: MIT Press, 1993.
Berry, Steven, James Levinsohn and Ariel Pakes. “Automobile Prices in Market Equilibrium,”
Econometrica, 1995, 63(4), 841‐890.
Bin, Okmyung and Carols Margins‐Filho. “Vehicle Price and Hydrocarbon Emissions: Evidence
from the Used‐Vehicle Markets,” Applied Economics Letters, 2008, 15(12), 939‐943.
Bishop, Gary A. and Donald H. Stedman. “A Decade of On‐Road Emissions Measurements,”
Environmental Science and Technology, 2008, 42(5), 1651‐1656.
Blinder, Alan S. “A Modest Proposal: Eco‐Friendly Stimulus,” New York Times, July 27, 2008.
Bond, Eric W. “Trade in Used Equipment with Heterogeneous Firms,” Journal of Political Economy,
1983, 91(4), 688‐705.
Borcherding, Thomas E. and Eugene Silberberg. “Shipping the Good Apples Out: the Alchian and
Allen Theorem Reconsidered,” Journal of Political Economy, 1978, 86(1), 131‐138.
Clerides, Sofronis, “Gains from Trade in Used Goods: Evidence from Automobiles.” Journal of
International Economics, 2008, 76, 322‐336.
Coase, Ronald. “Durability and Monopoly” Journal of Law and Economics, 1972, 15(1): 143‐149.
Copeland, Brian R. and M. Scott Taylor, “North‐South Trade and the Environment,” Quarterly
Journal of Economics, 1994, 109: 755‐787.
Copeland, Brian R. and M. Scott Taylor, ``Trade and the Environment: A Partial Synthesis,”
American Journal of Agricultural Economics, 1995, 77: 765‐771.
Copeland, Brian R. and M. Scott Taylor, Trade and the Environment: Theory and Evidence. Princeton:
Princeton University Press, 2003.
Dargay, Joyce, Dermot Gately and Martin Sommer. “Vehicle Ownership and Income Growth,
Worldwide: 1960‐2030,” The Energy Journal, 2007, 28(4): 143‐171..
Davis, Lucas W. “The Effect of Driving Restrictions on Air Quality in Mexico City” Journal of Political
Economy, 2008, 116(1), 38‐81.
Dargay, Joyce, Dermot Gately, and Martin Sommer. “Vehicle Ownership and Income Growth,
Worldwide: 1960‐2030,ʺ The Energy Journal, 2008, 28(4).
Esteban, Susanna and Matthew Shum. “Durable‐goods Oligopoly with Secondary Markets: The
Case of Automobiles,” RAND Journal of Economics, 2007, 38(2), 332‐354.
Feenstra, Robert C. and Gordon H. Hanson, “Foreign Direct Investment and Relative Wages:
Evidence from Mexico’s Maquiladoras,” Journal of International Economics, 1997, 42, 371‐393.
27
Fieler, Ana Cecilia, “Non‐Homotheticity and Bilateral Trade: Evidence and a Quantitative
Explanation,’’ 2007, working paper.
Flam, Harry and Elhanan Helpman, “Vertical Product Differentiation and North‐South Trade,”
American Economic Review, 1987, 77(5), 810‐822.
Fowlie, Meredith, “Incomplete Environmental Regulation, Imperfect Competition, and Emissions
Leakage,” American Economic Journal: Economic Policy, forthcoming.
Goldberg, Pinelopi. Koujianou, “Product Differentiation and Oligopoly in International Markets:
The Case of the U.S. Automobile Industry,” Econometrica, 1995, 63(4), 891‐951.
Goulder, Lawrence, Mark Jacobsen, Arthur Van Benthem, “Impacts of State‐Level Limits on
Greenhouse Gases per Mile in the Presence of National CAFE Standards,” 2009, working paper.
Grossman, Gene M., and Alan B. Krueger. “Environmental Impacts of a North American Free Trade
Agreement” in P. M. Garber, ed., The Mexico‐U. S. Free Trade Agreement, Cambridge, MA: MIT
Press, 1993.
Grossman, Gene M., Pablo Fajgelbaum, and Elhanan Helpman. “Income Distribution, Product
Quality and International Trade,” 2009, working paper.
Gruenspecht, Howard. “Differentiated Regulation: The Case of Auto Emissions Standards,”
American Economic Review, 1982, 72(2), 328‐331.
Harbaugh, William, Arik Levinson, and David Wilson. “Re‐examining the Empirical Evidence for an
Environmental Kuznets Curve,” Review of Economics and Statistics, 2002, 84(3) 541–51.
Holladay, Scott. “Pollution and Consumption and the Trade and Environment Debate,” 2008,
working paper.
Krueger, Alan B. and Alexandre Mas. “Strikes, Scabs, and Tread Separations: Labor Strife and the
Production of Defective Bridgestone/Firestone Tires,” Journal of Political Economy, 2004, 112(2),
253‐287.
Levinson, Arik and M. Scott Taylor. “Unmasking The Pollution Haven Effect,” International Economic
Review, 2008, 49(1), 223‐254.
Massachusetts Institute of Technology Energy Laboratory. “On the Road in 2020: A Lifecycle
Analysis of New Automobile Technologies,” October 2000, MIT EL 00‐003.
Matsuyama, Kiminori. “A Ricardian Model with a Continuum of Goods under Nonhomothetic
Preferences: Demand Complementarities, Income Distribution, and North‐South Trade,”
Journal of Political Economy, 2000, 108(6), 1093‐1120.
Mauzerall, Denise L., Babar Sultan, Namsoug Kim and David F. Bradford. ``NOx Emissions from
Large Point Sources: Variability in Ozone Production, Resulting Health Damages and Economic
Costs,ʹʹ Atmospheric Environment, 2005, 39: 2851‐2866.
McAusland, Carol. “Trade, Politics, and the Environment: Tailpipe vs. Smokestack,” Journal of
Environmental Economics and Management, 2008, 55(1), 52‐71.
Murphy, Kevin M. and Andrei Shleifer. “Quality and Trade,” Journal of Development Economics, 1997,
53(1), 1‐15.
Navaretti, Giorgio Barba, Isidro Soloaga, and Wendy Takacs. “Vintage Technologies and Skill
Constraints: Evidence from U.S. Exports of New and Used Machines.” The Word Bank Economic
Review, 2000, 14(1), 91‐109.
28
Newey, W. K., and K. D. West. “A Simple, Positive Semi‐definite Heteroskedasticity and
Autocorrelation Consistent Covariance Matrix,” Econometrica, 1987, 55 (3): 703–8.
Nordhaus, William. “A Review of the Stern Review on the Economics of Global Warming.” Journal of
Economic Literature, 2009, forthcoming.
Pelletiere, Danilo and Kenneth A. Reinart. “The Political Economy of Used Automobile Protection in
Latin America,” World Economy, 2002, 25(7), 1019‐1023.
Pelletiere, Danilo and Kenneth A. Reinart. “Used Automobile Protection and Trade: Gravity and
Ordered Probit Analysis,” Empirical Economics, 2004, 29(4), 737‐751.
Sen, Amartya Kumar. “On the Usefulness of Used Machines,” The Review of Economics and Statistics,
1962, 44(3), 346‐348.
Stokey, Nancy L. “The Volume and Composition of Trade between Rich and Poor Countries,”
Review of Economic Studies, 1991, 58, 63‐80.
Thoumi, Francisco E. “A Theory of the Optimum Age to Import a Durable Good,” Journal of
Development Economics, 1975, 2, 145‐160.
Ward’s Automotive Group. “Ward’s World Motor Vehicle Data,” various editions, Southfield,
Michigan.
World Bank Mexico Air Quality Management Team. 2002. “Improving Air Quality in Metropolitan
Mexico City: An Economic Valuation,” Policy Research Working Paper 2785, World Bank,
Washington, DC.
World Trade Organization. International Trade Statistics 2007 World Trade Publications: Geneva,
2007.
29
TABLE 1
Descriptive Statistics
Vehicles Exported from the United States to Mexico 2005‐2008
Vehicle Manufacturer
Ford .33
Chevrolet .17
Dodge .10
Nissan .07
Jeep .06
Plymouth .05
Mercury .04
GMC .03
Chrysler .03
Pontiac .02
Other .10
Note: This table describes cars and trucks exported from the United States to Mexico
between August 2005 and July 2008. These data were collected by the Mexican
Customs Agency, a branch of the Mexican Ministry of Finance.
TABLE 2
Top 10 Traded Models
Vehicles Exported from the United States to Mexico 2005‐2008
Traded
United States Vehicles Mexico
(1) (2) (3)
Local Pollutants:
Global Pollutants:
Note: This table reports means of the variables listed in the row headings. Column (1) describes the stock of vehicles in the
United States, 232 million total vehicles in 2005 according to R.L. Polk. Column (2) describes the vehicles exported from the
United States to Mexico between August 2005 and July 2008, 2.4 million total vehicles according to Mexican Customs as
described in Tables 1 and 2. Column (3) describes the stock of vehicles in Mexico, 22 million total vehicles in 2005 according to
INEGI. See equations in section 4.2 for details.
TABLE 4
Comparing Traded Vehicles to the Stock in California, at the Vehicle‐Level
With Model
Vintage
With With Make With Model Interactions With With Make With Model
Vintage Vintage Vintage and Quartic in Vintage Vintage Vintage
Fixed Effects Interactions Interactions Mileage Fixed Effects Interactions Interactions
(1) (2) (3) (4) (5) (6) (7)
Local Pollutants:
Global Pollutants:
Note: This table reports coefficients and standard errors corresponding to 1(Exported to Mexico) from 32 separate OLS regressions. The row headings list the dependent
variable used in each regression. The sample includes all vehicles with model years 1991‐1998 in the California sample. Local pollutants are measured in logs. All other
vehicle characteristics are measured in levels. Out of 3.6 million vehicles in the California data for the relevant vintages, 83,057 vehicles were exported. Vehicle mileage is
not available in the Illinois sample.
TABLE 5
Are Exported Vehicles More Likely to be Super‐Emitters?
California
Sample With
Model Vintage
Interactions
1(Failed Emissions Test Once) 1.21
(.013)
1(Failed Emissions Test Twice) 3.15
(.055)
1(Failed Emissions Test Three or More Times) 4.61
(.224)
1(Gross Polluter Once) 1.25
(.027)
1(Gross Polluter Twice) 3.54
(.122)
1(Gross Polluter Three or More Times) 6.07
(.589)
Note: This table reports results from 6 separate OLS regressions. The row headings
list the dependent variable used in each regression. In each row we report how
many times more likely it is for an exported vehicle to be in each category. For
example, controlling for model vintage interactions, exported vehicles are 3.15 times
more likely to have failed emissions testing twice. The sample includes all vehicles in
the California dataset with model years 1991‐1998. In the entire sample, 5.7% of
vehicles failed emissions testing once, 0.47% failed twice, and .04% failed three or
more times. A "gross polluter" is a vehicle that exceeds twice the allowable
emissions for at least one pollutant. In the entire sample, 1.55% of all vehicles were
gross polluters once, .10% were gross polluters twice, and .01% were gross polluters
three or more times.
TABLE 6
The Effect of NAFTA on New Vehicle Sales in Mexico
Note: This table reports estimates that correspond to 15 separate OLS regressions. The dependent variable is monthly sales of new vehicles in Mexico (in
logs) from INEGI, La Industria Automotriz en México, 2002‐2008 for total vehicles or for different vehicle segments as indicated in the column headings.
The table reports coefficients and standard errors for 1(NAFTA) , an indicator variable for the period after the liberalization of the used car market in
August 2005. The sample includes January 2001 to December 2007. Specifications include flexible polynomial time trends as indicated in the row headings
as well as the growth rate of GDP. In accordance with findings from standard diagnostic tests of serial correlation, reported standard errors are estimated
following Newey and West (1987) with a 1‐month lag.
TABLE 7
An Illustration of How Trade Can Increase Carbon Dioxide Emissions
Mexican
Drivers
Mexican Vehicle Intensity of Substituting
Baseline Retirement Rate Use (VMT) Away from
Assumptions Doubles Doubles MiniBuses
Short‐Run (1 year)
Carbon Dioxide Emissions Per Vehicle (Tons) 2.6 2.6 5.1 1.5
Social Cost Per Vehicle (Dollars) $24 $24 $49 $15
Long‐Run (20 years)
Carbon Dioxide Emissions Per Vehicle (Tons) 36.5 27.7 73.1 21.9
Social Cost Per Vehicle (Dollars) $308 $237 $615 $185
Note: This table describes the short‐run and long‐run increase in carbon emissions associated with Mexico importing a vehicle from the United States.