The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Heterogeneous Districts,
Interests, and Trade Policy
WP 23-12
Kishore Gawande
University of Texas at Austin
Pablo M. Pinto
University of Houston
Santiago M. Pinto
Federal Reserve Bank of Richmond
Heterogeneous Districts, Interests,
and Trade Policy∗
Kishore Gawande†, Pablo M. Pinto‡, and Santiago M. Pinto§
This version: March, 2023
Abstract: Congressional districts are political entities with heterogeneous trade policy
preferences due to their diverse economic structures. Representation of these interests in
Congress is a crucial aspect of trade policymaking that is missing in canonical political economy models of trade. In this paper, we underscore the influence of districts by developing a
political economy model of trade with region-specific factors. Using 2002 data from U.S. Congressional Districts, we first characterize the unobserved district-level demand for protection.
Extending the model beyond the small country assumption to account for export interests
as a force countering protection, we develop a model of national tariff-setting. The model
predictions are used to estimate the welfare weights implied by tariff and non-tariff measures enacted nationally. Our supply-side explanation for trade policy, while complementing
Grossman and Helpman (1994), reveals district and industry-level patterns of winners and
losers, central to understanding the political consequences of trade and the backlash against
globalization.
Keywords: Trade Policy, Political Economy, Districts, Tariffs, NTMs, Legislature.
JEL Classification: F13, F14, D72, D78
∗
An earlier version of this paper circulated under the title “Voting, lobbying, and trade policy:
A structural estimation framework,” 2020 (https://www.internationalpoliticaleconomysociety.org/
conference-program). See http://pablopinto.com/working-papers/ for the latest version of the paper. We thank Gail Buttorff, Nathan Canen, Maria Carreri, Soohyun Cho, Bill Clark, Jeff Frieden, Guy
Grossman, Bobby Gulotty, Annie Hsu, Hyeran Jo, Quan Li, John Londregan, Sameer Malik, Liu Mendoza,
Devashish Mitra, Peter Rosendorff, Horacio Rueda, Sebastian Saiegh, Shanker Satyanath, Edoardo Teso,
Agustin Vallejo, Rachel Wellhausen, Sunny Wong, Hye Young You, participants in the 2019 American Political Science Association, 2020 International Political Economy Society, 2020 Southern Regional Science
Association Meetings, 2020 Texas Methods Meeting, 2021 Federal Reserve System Committee on Regional
Analysis, 2021 North America Meetings of the Regional Science Association, 2021 Latin American Polmeth
meeting, Petralia Applied Microeconomics Workshop, and workshops at the University of Houston, and
Texas A&M University for excellent comments and suggestions.
†
University of Texas at Austin; kishore.gawande@mccombs.utexas.edu
‡
University of Houston; ppinto@central.uh.edu
§
Federal Reserve Bank of Richmond; Santiago.Pinto@rich.frb.org. The information and views expressed
in this article are those of the authors and do not necessarily reflect those of the Federal Reserve Bank of
Richmond, or the Federal Reserve System.
1
Introduction
Political economy models of trade posit that a political entity, a “government,” decides how
much trade protection is optimal for every sector of the economy. This may diverge from free
trade because what is politically optimal for the tariff setter may not be optimal for citizens
taken together. A classic model explaining this divergence is Grossman and Helpman (1994)
in which special interests pay the government for protection from imports according to the
willingness of the government to receive. That, in turn, is determined by the weight the
government places on (a dollar of) its citizens’ welfare relative to (a dollar of) campaign
contributions that the government pockets. Thus, protection is endogenous: the payoffs
from protection to owners of specific factors of production (workers and capitalists) who
benefit from trade restrictions incentivize them to try to alter the government’s calculus
by making quid pro quo contributions. Helpman (1997) unifies analytically several models
of endogenous protection in which the government’s calculus is altered by interest groups
(Magee et al., 1989); by political support from producers and consumers (Hillman, 1982); by
competing lobbies (Bhagwati and Feenstra, 1982, Findlay and Wellisz, 1982); or by balancing
domestic and foreign policy motivations (Hillman and Ursprung, 1988).
But who or what is “government?” Few models have allowed the actual process of preference aggregation in trade policymaking a significant role. Grossman and Helpman (1996)
model the determinants of trade policy platforms chosen by representatives competing at
the polls, which sheds light on the importance of ideology, uninformed voters, and special
interest. Even in models featuring electoral competition (Magee et al., 1989, Chapter 6)
or direct democracy (Mayer, 1984, Dutt and Mitra, 2002), incentives faced by members of
the legislature, even the executive, are abstracted (Rodrik, 1995). This sidelines the institutionally most important actors in the tariff game, legislators, who must coalesce in the
formulation of trade policy.
This paper attempts to restore the place of the legislature and the executive in a model
of endogenous protection. Our model brings to focus the preferences of economically heterogeneous districts. These district-level preferences must be aggregated into a national policy,
a process in which representatives form coalitions and engage in bargaining to arrive at a
trade policy that is agreeable to a majority of members of the legislature. The impact of
the process of aggregation of heterogeneous regional, or district-level preferences, is absent
in political economy models of trade like Grossman and Helpman (1994) where a unilateral
decision-maker sets tariffs. Who wins in these legislative bargains is the subject of a large
body of research in political science built around the seminal work of Riker (1962). A prin1
cipal contribution to the legislative bargain literature is Baron and Ferejohn (1989), whose
focus on the role of an agenda setter in the distribution of gains, provides a framework for
characterizing the process of preference aggregation in the making of trade policy (see Celik
et al. (2013)).
Our research builds on a large literature that has sought to explain U.S. protectionism
(Deardorff and Stern, 1983, Marvel and Ray, 1983) and its political economy determinants
(Baldwin, 1985, Ray, 1981, Trefler, 1993). These empirical examinations make the case
that, ultimately, the government dispenses trade protection in response to demands from
economic actors affected by trade. The Grossman and Helpman (1994) model highlights an
important aspect of the demand side of trade policymaking: the influence of special interests.
Examinations of the Grossman-Helpman model of trade protection (Goldberg and Maggi,
1999, Gawande and Bandyopadhyay, 2000) have further advanced the empirical literature on
the influence of special interests. They find that while import-competing interests do exert
influence, the amount of protection they can “buy” is less than what one might expect.
The model in this paper offers a view of these results from a different lens: focusing
on legislators’ incentives and legislative bargaining provides a supply-side explanation of
trade policymaking in the U.S. The paper makes three main contributions. First, we model
tariff determination in the presence of heterogeneous regional interests. The model develops
micro-foundations for an institutional explanation of why tariffs have remained low in the
U.S. in the post-WWII era despite a growing public backlash against globalization. The
analysis retrospectively examines the U.S. tariff structure that was largely determined in
the Kennedy and Tokyo rounds of GATT. Hence, we extend the model to account for the
reciprocal nature of trade liberalization, bringing to the fore the interests of specific factors
in exporting industries that value preferential access to foreign markets, thereby affecting
the calculus of policymakers (see Irwin and Kroszner (1999), Irwin (2017)).
The second contribution of the paper is to integrate legislative bargaining into a structural
political economy model of trade, as in Celik et al. (2013). We model stylized coalitions in
the legislature based on geography and politics. The main result is empirical: we estimate
the implicit welfare weights that members of these coalitions “win” on behalf of specific
factor owners in their districts in the national bargain. The bottom line is that the national
tariffs and non-tariff measures derived from the model depend on the regional structure of
economic activity, the weights representatives place on factor owners in their districts, and
the way district preferences are aggregated. The model is consistent with, and closer to,
institutions under which trade policy is made in the U.S. and captures the give-and-take
2
between Congress and the Executive (Finger et al., 1988, Destler, 2005).
Third, and perhaps most importantly, the estimates of structural parameters, the welfare
weights, provide a theory-based explanation of why U.S. manufacturing tariffs have been
low and remained low even at the onset of the China shock. The results highlight the role
of export interests in making it so, portraying which region’s interests went unfulfilled and
which region’s interests were advanced in the making of trade policy.
The main results from our paper can be summarized as follows. First, we use the model
predictions to estimate district-level tariff preferences. These estimates provide a measure of
–otherwise unobserved– local demands for protection. The contrast between the independent
demand for protection by districts and the protection delivered by the legislature, a measure
of their “unmet" demand, can be stark, particularly in industrially concentrated districts.1
Second, the model provides the structure for estimating the implicit welfare weights that
owners of specific factors of production and mobile factors receive in the process of aggregating district-level preferences into the national tariff. Overlaying a model of the legislative
bargaining process further establishes that the vector of national tariffs proposed by an
agenda setter, such as the House Ways and Means Committee, that would muster support
in Congress is a weighted average of the demand for protection in a majority of districts.
We estimate these weights using tariff data from the early 2000s, a time when the U.S.
economy was transformed by the deluge of manufacturing imports, particularly from China.
The results from this exercise suggest that the underlying political process determining national tariffs places twice as much weight on the aggregate welfare of mobile factors (labor)
relative to the aggregate welfare of sector-specific factor owners seeking protection. Further,
the positive weights on specific factor owners in import-competing industries are distributed
unequally across districts and industries. The aggregate level of protection, including tariff
and non-tariff measures (NTMs), implies that Republican-controlled districts take the lion’s
share of the aggregate weights placed on specific capital owners: they outweigh Democrat
districts by a 2-to-1 ratio.
Finally, parameter estimates accounting for the reciprocal determination of tariffs and
terms of trade effects (the large country case) unveil the strong influence of specific factor
owners in exporting industries: their welfare is weighted as much as the welfare of factor
owners in import-competing industries. Furthermore, when accounting for reciprocity with
the rest of the world in the determination of U.S. tariffs, we find that specific factor owners
in safe Republican districts in states carried by the Republican Presidential ticket and safe
1
The relevance of this finding cannot be understated. It is the source of the China shock, examined in
influential articles, e.g., Autor et al. (2013), that promises to shape the trade policy debate.
3
Republican districts in battleground states receive positive weights. These findings suggest
that the legislative majority enacting the tariffs includes representatives from districts with
a higher concentration of specific factor owners in exporting industries. These are important
and novel results that existing models of the political economy of trade do not capture. In
the ensuing sections, we introduce our models, form an estimation strategy based on the
propositions derived from the models, and present the results from our empirical analyses.
2
District Tariff Preferences: A General Framework
What tariff levels would be set by a decentralized policymaker seeking to represent interests
in her district? This section presents a model of “district tariff preferences.”
A small open economy is populated by two groups of economic agents: owners of factors
specific to the production of good j, or specific “capital” Kj , and owners of a mobile factor L
that is used in the production of all goods. Each individual owns one unit of either L or Kj .
J goods are produced nationally, but their production is dispersed across R districts, where
each district has equal political representation in the nation’s legislature. The composition of
output is heterogeneous across districts and depends on the (exogenous) regional endowment
of factors. We assume that factor owners are immobile across districts, that is, a district is a
local labor market (Topel, 1986, Moretti, 2011, Autor et al., 2014, 2013).2 The non-specific
factor (labor) is mobile across goods while specific factor owners, by definition, are immobile
outside the good (sector) in whose production they are employed. The population of district
PJ
PJ
K
K
L
r is nr = nLr + nK
r =
j=1 njr , where njr specific factor owners in district r are
j=0 njr +
P
employed in producing good j. Aggregate population n = r nr .
Goods j = 1, ..., J are tradable and, under the small country assumption, world prices are
exogenously determined, and taken as given. Domestic prices may be changed by raising or
lowering tariffs. To keep the interpretation of the models simple, negative tariffs and subsidies
are not allowed. There are no transport costs and goods are delivered to consumers at these
domestic prices. Policy-induced price changes affect domestic production and consumption
of goods, and hence the welfare of specific factor owners.
Production. Aggregate population n is distributed across R districts indexed by r, r =
1, ..., R. In each district, output in the non-tradable numeraire good 0 is produced using
only the mobile factor (labor) with linear technology, which fixes labor’s wage in district r
at wr > 0 (across all goods). The output of the numeraire good in district r is q0r = wr nL0r ,
where nL0r owners of labor in district r are employed in producing good 0. Units are chosen
2
The assumption that labor markets are local plays a fundamental role in contributing to the impact of
trade and innovation on manufacturing employment and wages (Autor et al. (2013, 2014)).
4
such that the price of the numeraire good (nationally) is p0 = 1. Prices pj in the J nonnumeraire goods are expressed in these units.
Good j is produced with CRS technology. In district r, the technology combines nLjr
units of labor and the fixed endowment of nK
jr specific factors. These specific factors earn
the indirect profit function πjr (pj ), and labor earns wage wr regardless of its sector (good) of
employment. A district does not necessarily produce all goods. By assumption, when good
L
j is not produced in region r, nK
jr = njr = 0 and πjr = 0. The output of good j in district r
P
′
(pj ) > 0 and its aggregate output is Qj (pj ) = R
is qjr (pj ) = πjr
r=1 qjr (pj ).
Preferences. Preferences are homogeneous across individuals in groups L and K, and repP
resented by the quasi-linear utility function u = x0 + j uj (xj ). These imply (separable)
demand functions xj = dj (pj ) for each individual. The indirect utility of an individual who
P
spends z on consumption is z + j ϕj (pj ), where ϕj (pj ) = vj (pj ) − pj dj (pj ) is the consumer
surplus from good j.3 The total per capita consumer surplus of the consumption of goods
P
j = 1, ..., J is ϕ = ϕj (pj ). The aggregate demand for good j is Dj (pj ) = ndj (pj ), where n
is the country’s population.
Imports, tariffs, and tariff revenue. Mj denotes imports of good j, and is given by
Mj (pj ) = Dj (pj ) − Qj (pj ). Trade policy consists of imposing a specific per unit tariff tj
on the import of goods j, j = 1, . . . , J. Total revenue generated by the tariffs is T =
P
P
(pj − pj )[Dj (pj ) − Qj (pj )] = j (pj − pj )Mj (pj ), where Mj (pj ) is good j’s import demand
function, pj is the world price, and tj = pj − pj . Import subsidies are disallowed. Tariffs on
imports are collected at the country’s border and tariff revenue is distributed nationally on
an equal per capita basis, i.e., each individual receives
T
.
n
Total utility. The total utility of the mobile non-specific factor in good-district {jr} is
L
= nLjr wr+ Tn + ϕ , and
Wjr
the total utility of specific factor owners in good-district {jr}
πjr
T
T
K
is Wjr
= nK
jr nK + n + ϕ . Common to both is the per capita tariff revenue, n , and the
jr
total per capita consumer surplus, ϕ. The expressions differ, however, on the income received
by each factor of production. While a higher tariff increases pj and lowers consumer surplus,
it also raises the return to the specific factor owner employed in {jr}. This group, therefore,
could have a strong interest in demanding a tariff on imports of j.
3
The index r is dropped from the demand functions because they do not change across districts (prices
are nationally determined). Technical Appendix B considers heterogeneous tastes forPthe two types of
m m
agents. This model assumes preferences are described by the utility function um = xm
0 +
j uj (xj ), where
m = {L, K} indexes
of labor
owners of the specific factor, yielding demand functions dm
j (pj ) and
P owners
P and
m
m
consumer surplus j ϕj (pj ) = j [vj (pj ) − pj dm
(p
)].
j
j
5
2.1
District Tariff Preferences
Tariffs are, of course, decided at the national level. Our framework seeks to understand how
a policymaking body comprising representatives from each district –like the U.S. House of
Representatives– arrives at these national tariffs. We approach this problem by answering
two questions. First, if individual districts were granted the authority to choose tariffs for
the entire nation, what would their preferred tariffs be? Second, how are these preferences
aggregated across districts into nationally determined tariffs?
This section answers the first question. Consider the case in which a representative of
district r chooses (national) tariffs to maximize the district’s welfare, defined as a weighted
sum of the welfare of each factor owner in the district. We begin with a general framework
where the welfare weights differ across districts, goods, and the two groups of factor owners.
We will later apply sensible restrictions to identify the weights in the estimation. In district
r, a unit of a specific factor employed in producing good j gets welfare weight ΛK
jr , and a
unit of labor in good j gets welfare weight ΛLjr . District r’s aggregate welfare is
Ωr =
X
L
ΛLjr Wjr
+
j
X
K
ΛK
jr Wjr ,
j
m
where the total welfare of type-m factor owners in district r, Wjr
, depends on the vector
of domestic prices p = (p1 , ..., pJ ). The small open economy assumption means there is a
one-to-one relationship between the tariff tj and price pj (the world price pj is exogenous),
m
for the two types of factors are functions of tariffs. Then, district r’s
and total welfare Wjr
aggregate welfare may be decomposed as
Ωr =
X
ΛLjr nLjr
j
Ç
å
Å
ã X
T
πjr T
K K
wr + + ϕ +
Λjr njr
+ +ϕ .
K
n
n
n
jr
j
(1)
The first parenthesis in (1) defines welfare per a non-specific factor owner, and the second
parenthesis defines welfare for a specific factor owner. The first expression on the right-hand
side weights the sum of per capita wage, tariff revenue, and consumer surplus to arrive at
the aggregate welfare of owners of L residing in district r. The weights are the product
of ΛLjr , the welfare weight assigned to each non-specific factor employed in producing good
j, and the number of district r’s non-specific factors employed in producing the good, njr .
The second expression differs in the first component: the per capita returns to owners of
good j-specific factor,
weights
K
ΛK
jr njr
πjr
.
nK
jr
The three components in the expression are aggregated using the
to obtain the welfare of district r’s specific-factor owners.
6
Noting that T , ϕ and πjr are functions of tj , the tariffs preferred by district r are obtained
by maximizing (1) with respect to each tj . Denote the aggregate welfare weights on factor
PJ
PJ
K K
L
L L
owners in district r as λK
r =
j=1 Λjr njr and λr =
j=0 Λjr njr , respectively, and their sum
as λr = λLr + λK
r . Then, district r’s preferred tariff on good j, tjr , is
tjr
ñ
Ç å
ô
K
n ΛK
qjr
Dj Mj
jr njr
= − ′
+
,
−
Mj
λr
n
n
nK
jr
for r = 1, . . . , R, where
Dj
n
j = 1, . . . , J,
(2)
Mj
n
is the country’s
is the country’s per capita demand for good j,
per capita imports of good j, and
Mj′
≡
∂Mj
∂tj
< 0. The representative chooses trade policy tjr
defined by (2). This equation captures both the interests of producers in district r and the
welfare of consumers nationally, given the assumption of identical tastes. The first term in
the square brackets indicates that the tariff increases with r’s output of good j through the
tariff’s positive impact on profits.4 The second term shows that the tariff declines with the
nation’s per capita demand via the negative impact of the tariff on consumer surplus. The
third term indicates the tariff increases with national imports through its impact on tariff
revenue, which is redistributed lump-sum to the nation’s residents.
An institutional interpretation is that (2) determines the tariff preferred by a representative of a district r, which is one among a federation of districts. This representative is
accountable to district r’s residents;
the choice of the nation’s tariff in good j represents
K
ΛK
n
q
the local interests via jrλr jr njr
in (2). The tariff reduces the consumer surplus of the
K
jr
representative national consumer via
−Dj
n
as a lump sum back to all consumers via
in (2), and revenue from the tariff is distributed
Mj
.
n
In a majoritarian electoral system, such as in
the U.S., a member of the House of Representatives faces incentives to choose a trade policy
tjr defined by (2) that maximizes the welfare function (1) for the district the member represents.5 The following proposition describes the level of protection in terms of ad-valorem
tariffs:
4
By the envelope theorem the derivative of profits with respect to price is output, reflecting the impact
of the tariff on returns to owners of sector-specific factors in district r. With labor perfectly mobile across
goods within district r, wjr = wr for all j, where wr is determined by labor’s productivity in the numeraire
good. Any change in tariff tj does not affect labor income.
5
The district is institutionally constrained, being part of the federation of districts, to distribute import
tariff revenue equally across all districts in the federation. Further, the market for each good clears at the
national level. District r considers the impact of higher tariffs on district r’s consumers; some effects are
“washed out” on the consumer side because preferences across groups are assumed identical. The vector
of tariffs enacted by Congress for the nation then reflects the weights on different factors, industries, and
districts, implied by a legislative bargaining process, given regional output-to-import ratios and import
elasticities.
7
Proposition 1 District r’s effective demand for tariff protection in good j is:
ΛK
τjr
jr nr
=
1 + τjr
λr
where τjr =
Å
qjr /Mjr
−ϵj
ã
Å
Qj /Mj
−
−ϵj
ã
,
(3)
tjr
pj
is the ad-valorem tariff proposed by district r as the tariff applicable to the
nation’s imports of good j, and Mjr = Mj × nnr .
Äp ä
Proof Using good j’s import demand elasticity ϵj = Mj′ Mjj , the market clearing condition
Dj = Qj +Mj , and defining ad-valorem tariffs as τjr =
tjr
pj
or
τjr
(1+τjr )
=
tjr
,
pj
(2) may be written
as:
τjr
n
=
1 + τjr
−ϵj Mj
Ç
K
ΛK
Qj
jr njr qjr
−
K
λr njr
n
å
ΛK
jr n
=
λr
Å
qjr /Mj
−ϵj
ã
Å
Qj /Mj
−
−ϵj
ã
.
(4)
Assuming Mj is distributed according to districts’ populations, district r’s imports of j,
Mjr are Mjr = Mj × nnr . (3) then predicts tariffs with (district) output-to-import ratios,
enabling comparison with the Grossman and Helpman (1994, GH) model. □
Just as in the GH model, good j’s tariff is determined by the output-to-import ratio in
the sector and its import demand elasticity, represented by
qjr /Mjr
.
(−ϵj )
The models differ in that
(3) is the “national” tariff on imports of j that is preferred by the representative of district
L
r. In (3), if ΛK
jr = Λjr = Λr , that is, if all factor owners in district r get equal weight, the
coefficient on
qjr /Mjr
(−ϵj )
equals 1 and
ã Å
ã
Å
Qj /Mj
qjr /Mjr
> 0, if
>
−ϵj
−ϵj
τjr
ã Å
ã
Å
Qj /Mj
qjr /Mjr
1 + τjr
= 0, if
≤
,
−ϵj
−ϵj
(5)
where we impose the non-negativity constraint on tariffs (i.e., no import subsidies allowed).
From (5) it becomes apparent that even when special interests, that is, specific factor owners,
have the same welfare weight as labor, tariffs can be positive. If, for example, production of
good j is concentrated in district r, qjr = Qj and τjr > 0. Expression (3) shows the implicit
demand for tariffs by district r given the institutions. The national tariff schedule aggregates
the tariff preferences, given by (3), of districts. The aggregation of district preferences into
national trade policy is discussed in the next section.
Relationship to the GH model. In the GH model, the welfare of specific factors employed
in good j is given the weight 1j + a, where 1j is a binary indicator equal to one if sector j
8
is politically organized to lobby and zero otherwise. The parameter a represents the weight
given to consumers in the model so that
(1+a)
a
is the relative weight on the welfare of organized
specific factors and reflects their influence on tariff-making. Adapting these weights to our
model with districts, let ar be the weight placed by district r’s representative on the welfare
of labor and 1jr + ar the weight placed on the welfare of specific capital owners, where 1jr
equals one if sector j in district r is politically organized to lobby (the representative) in
district r and zero otherwise. That is, ΛLjr = ar and ΛK
jr = 1jr + ar . Then (3) may be written
as
Å
ã Å
ã
τjr
qjr /Mjr
(1jr + ar ) nr
Qj /Mj
= PJ
−
PJ
K
L
1 + τjr
−ϵj
−ϵj
j=1 (1jr + ar )njr +
j=0 ar njr
Å
ã Å
ã
qjr /Mjr
Qj /Mj
(1jr + ar ) nr
−
.
= PJ
K
−ϵj
−ϵj
j=1 1jr njr + ar nr
Let
αrK denote the fraction of district r’s population that is politically organized, αrK =
P
J
j=1
1jr nK
jr
nr
; this expression is the district-equivalent of GH’s αL . Then,
τjr
1jr + ar
= K
1 + τjr
αr + ar
Å
qjr /Mjr
−ϵj
ã
Å
Qj /Mj
−
−ϵj
ã
.
In the GH model, if everyone is politically organized, lobbies contribute but they nullify each
other and there is free trade in all goods. In our model with everyone organized, αrK = 1
and we get the result in (5).6
2.2
Some Counterfactual Results
Equation (3) may be used to predict the unobservable demand for protection, that is, the
vector of tariffs at the line level preferred by each district. Another use for which the model
may be put is to estimate the counterfactual welfare weights, separately for each district, that
would deliver the observed U.S. tariff data. We construct a spatial data set with industrydistrict output (qjr ), imports (Mj ), import demand elasticities (ϵj ), and ad valorem tariffs
(τj ). Data we collect, from a variety of sources described below, are as disaggregated at the
industry level as possible with public census data.
Data and sources. Output and employment data are from the Census Bureau (County
Business Patterns (CBP), 2002); import and tariff data are from the United States Interna6
Note that (5) would result as well if nobody is politically organized, i.e.,
Q
q
In the GH model, where the district is the nation, Mjrjr = Mjj , and τjr = 0.
9
1jr = 0 for all j, r, and αrK =0.
tional Trade Commission’s DataWeb.7 . Ad valorem tariffs, from USTradeOnline, are based
on duties collected at customs and measured at HS 10 digits. Import elasticities at 6-digit
HS are from Kee et al. (2008). Output and employment data from CBP were converted
to the NAICS 3-digit level, and mapped from Metropolitan Statistical Areas and Counties
onto 433 congressional districts for the 107th Congress (the year 2002).8 The year 2002 is
chosen also for the window it provides at the inception of the “China’s shock,” the subject
of intense recent research.9 The share of workers in district r who own specific capital in
any sector,
nK
r
nr
is measured in two steps. A significant part of the compensation of white-
collar (non-production) workers is rent due to their specificity, while blue-collar (production)
workers, who are not “stuck” to a specific sector earn wages. The Census of Manufacturing
provides data on national manufacturing employment and the proportion of production
and non-production workers
nK
n
in each NAICS industry. The ratio
nK
r
nr
nL
n
is computed as the
average of the national proportions using district r’s sectoral manufacturing employment
as weights. Alternative measures of specific factor ownership by industry, based on the
classification of occupations in manufacturing and services (Autor and Dorn, 2013), have
ratios similar in magnitude to those used in our estimations. Those measures, however, are
not available at the district level. District r’s sectoral manufacturing employment is from
the 2000 County Business Patterns, in turn, from the Geographical Area Series of the 2000
Census of Manufacturing.
District-specific results. Using (3) we conduct two counterfactual exercises about (unobserved) district tariff preferences. The first exercise estimates the relative welfare weights
ΛK
jr
ΛL
jr
under the counterfactual that the observed national tariff τj is the preferred tariff on good
j for every district. We proceed with sensible restrictions that identify the welfare weights.
Assume that in district r, weights on owners of K and L are invariant across goods, that
K
L
L
is, ΛK
jr = Λr and Λjr = Λr . The first assumption is satisfied if representatives who “assign”
these weights are influenced equally by specific factor owners, for example, if they are politically organized in all industries. Another possibility is for specific factors in a district to
get equal weight based on their (equal) voting strength, but more weight than labor whose
7
See USITC DataWeb.
Due to non-disclosure restrictions we lose data for two of the 435 congressional districts. In other cases
(approximately 17% of the sample), we can impute missing district-industry output data using available
district-industry employment data. Documentation of the data and imputations where confidentiality issues
prevent the Census from publicly reporting district output data is provided in Appendix C.
9
The implications of these results for research on the China shock are in a companion paper.
8
10
wage is not influenced by policy. Then (3) may be written as
τjr
=
1 + τjr
L
If ΛK
r > Λr , the coefficient on
Å
1
nK
r
nr
+
nL
r
nr
qjr /Mjr
−ϵj
Ä ΛL ä
r
ΛK
r
qjr /Mjr
−ϵj
ã
−
Qj /Mj
.
−ϵj
(6)
is greater than 1 (and conversely). For each of the
433 districts, we regress 2002 U.S. manufacturing tariffs at HS 8-digits, the tariff line level
at which policymakers determine the schedule, on
qjr /Mjr
−ϵj
and
Qj /Mj
,
−ϵj
the latter constrained to −1. We then back out the relative weights
coefficient on
with the coefficient on
ΛK
jr
ΛL
jr
from the estimated
qjr /Mjr
.
−ϵj
K
Figure 1: Implicit relative weights on specific to mobile factors ( ΛΛL ) by CDs
Figure 1 maps the distribution of the estimates of
ΛK
jr
ΛL
jr
. The estimate of
ΛK
r
ΛL
r
is lower
than one in 75% of the 433 districts, implying that the nationally set tariffs go against the
interests of specific factor owners. A takeaway is that it is hard for the majority of districts
to even have the voices of their specific factors heard in the determination of national tariffs,
let alone receive their tariff preference.
The next exercise answers the counterfactual of matching district tariff preferences with
what districts succeed in obtaining once their preferences are aggregated (in Congress) into
11
Figure 2: Predicted district-level tariffs (τr ), by NAICS 3-digit industries
the tariff schedule that governs U.S. trade policy.10 This second counterfactual exercise fixes
the ratio
ΛK
jr
ΛL
jr
equal to one for all j, r, and predicts the vector of tariffs τr for district r =
1, . . . , 433. Figures 2 summarize these results.11 Figure 2 clearly shows that the distribution
of district-level demand for protection, τjr , varies widely across districts and industries.
While predicted industry-district tariffs can be large as shown in figure A.1.2 in Appendix
q
10
/M
As a prelude, Figure A.1.1 in Appendix A.1 shows the distribution of the variable jr−ϵj jr for the twenty
NAICS 3-digit industries. Note that this exercise does not attempt to calculate the optimal district-preferred
tariff since this would require, among other things, estimating the variables qjr , Mj , Qj and ϵj at the tariff
level decided by the district. In this counterfactual exercise, we evaluate equation (6) at the observed data
for the year 2002. Thereby, we intend to provide an idea of the magnitude of the preferred district-sector
tariff that would rationalize the observed outcomes.
11
Figure A.1.2 in Appendix A.1 shows the distribution, across 433 CDs, of the positive tariff estimates
(τjr > 0) for the twenty NAICS-3 industries.
12
A.1, the overwhelming majority of districts are predicted to demand zero tariffs in most
industries, as shown in Figure A.1.2 and Table A.1.1.. in Appendix A.1. Importantly, in
districts with positive tariff estimates, the implied demand for protection dwarfs the level of
protection granted to the industry in 2002 (Table A.1.1 in Appendix A.1).
A message from these counterfactual exercises is that district representatives have little
chance of getting their preferred tariffs. For an individual district,
qjr /Mjr
−ϵj
>
Qj /Mj
−ϵj
only
if the output of j is concentrated. A coalition C of districts with output-to-import ratio
qjr /Mjr
−ϵj
>
Qj /Mj
−ϵj
for all r ∈ C has a better chance of obtaining at least some protection
(than if each r went alone) in the legislative bargain over the national tariff schedule. The
bargain ultimately determines the welfare weights the winning coalition earns for specific
factors in their districts relative to other coalitions. The aggregation of district preferences
into national tariffs is the subject of the remainder of the paper.
3
National Tariffs in a Small Open Economy
How are district tariff preferences aggregated into national tariffs? We draw on the legislative
bargaining literature (Baron and Ferejohn, 1989, Eraslan and Evdokimov, 2019, Celik et al.,
2013) to answer this question. The solution provides a foundation for estimating welfare
weights implied by national tariffs.
3.1
The Baron-Ferejohn solution
The Baron-Ferejohn model (henceforth BF) predicts the distribution of gains in a legislature
under different voting rules. In the canonical model, a fixed amount of money A is to
be distributed among n (homogeneous) districts.12 An agenda-setter proposes a specific
distribution of A across n districts, with the motion holding if a majority of the districts
votes in favor. Under a closed rule, if the proposal by the agenda setter is rejected, the
session terminates.
We extend the canonical model to include districts with heterogeneous tariff preferences.
The framework presented in this section is a variation of the model in Celik et al. (2013).13
The starting point of the BF version of our model is the vector of tariffs preferred by each
district given by equation (3) in Section 2.1. As described above, these are the tariffs district
r would choose if it could impose its preferences over the other districts. In reality, however,
individual districts do not have that power: they will need to form coalitions and hope to
12
Eraslan and Evdokimov (2019) review this literature.
The main difference with the Celik et al. (2013) model is that in our solution tariffs determined by the
winning coalition depend not only on the geographic concentration of economic activity, but reflect welfare
weights placed on different factor owners in the districts and nationally.
13
13
be part of the majority needed to move their joint preferred tariffs. This approach explains
how district tariff preferences tjr , for r = 1, . . . , R, may be aggregated into national tariffs t.
One-period, three-region BF model. Consider a one-period BF bargaining model with
three districts, each with the same number of residents nr = n/3 (so that each district has
the same national representation). Suppose district r is randomly selected to be the agenda
setter. Then, r’s proposal is implemented if at least one other district, district r′ , joins to
form a majority coalition.
The objective of the agenda setter, district r, is to attract one other district to the
coalition, for instance, r′ , at the minimum cost possible cost for r.14 The problem can
be thought of as being solved in two stages. In the first stage, the agenda setter chooses
the vector of (specific) tariffs tr = {t1r , . . . , tjr , . . . , tJr } that maximizes district r’s welfare
Ωr (tr ) subject to Ωr′ (t)r ≥ Ωr′ (t), r′ ̸= r, where t is the vector of existing (status-quo)
′
′
tariffs. Denoting by trr the solution tariff vector for each r′ , district r receives utility Ωr (trr ).
In fact, district r maximizes the Lagrangian Lr = Ωr (tr ) + ρr′ [Ωr′ (tr ) − Ωr′ (t̄)] with respect
to tr , where ρr′ ≥ 0 denotes the Lagrange multiplier for each r′ ̸= r. Specifically, ρr′ =
î ∂Ω /∂t ó
Max − ∂Ω r′ /∂tjj , 0 . At an interior solution, when the constraint is binding, the numerator
r
and denominator have opposite signs: conceding a higher tj to satisfy r′ lowers r’s welfare.
The size of ρr′ depends on the rate of this trade-off at the constrained maximum. The
solution to this problem gives the vector of specific tariffs that district r would propose to
district r′ , and district r′ would accept. For each j = 1, . . . , J, the solution tariff, denoted
′
by trjr , is given by
ñ
′ ô
r′
r′
λK
λK
Qj (trjr )
n
jr qjr (tjr )
jr′ qjr′ (tjr )
=
αr
+ (1 − αr )
−
,
′
λr nK
λr′ nK
n
−Mj′ (trjr )
jr
jr′
′
trjr
(7)
′
where αr = λr /(λr + ρr′ λr′ ) ≥ 0. Note that the endogenous variables are evaluated at trjr .
The following proposition summarizes the result using ad-valorem tariffs.
Proposition 2 In the three-district case, the ad-valorem tariff on good j proposed by the
district-r agenda setter to (the representative of ) district r′ that would be accepted by r′ ,
′
r′
τjr
/(1 + τjr ) = trjr /pj , for each j = 1, . . . , J, is given by
′
r
τjr
′
r
1 + τjr
ñ
ô
r′
r′
r′ )
λK
λK
Qj (τjr
n
jr qjr (τjr )
jr′ qjr′ (τjr )
=
αr
+ (1 − αr )
−
r′ )M (τ r′ )
λr nK
λr′ nK
n
−ϵj (τjr
j jr
jr
jr′
(8)
In general, the agenda setter r would need to offer tariff concessions to on another district r′ in the
economy, resulting in a tariff vector that in general does not coincide with the agenda setters’ unconditional
most preferred tariff vector.
14
14
K K
where λK
jr = Λjr njr is the aggregate welfare weight placed on special interests in district r,
P P
m
λr = ΛL0r nL0r + m j Λm
jr njr is the aggregate welfare weight on the district r’s population,
and the weight αr =
λr
λr +ρr′ λr′
is a function of the Lagrange multiplier ρr , and 0 < αr < 1.
If the supply for good j at each region r is fixed and Mj′ is constant, then, observing (4),
′
′
′
r
trjr and the ad-valorem tariff τjr
= trjr /p̄j can be expressed as the weighted average of their
respective district-preferred tariffs.15 Specifically,
′
′
trjr = αr tjr + (1 − αr )tjr′ ,
′
r
and τjr
= αr τjr + (1 − αr )τjr′ .
(9)
′
r
r
For τjr
/(1 + τjr
), it follows, under the previous assumptions, that
′
r
τjr
= αr
r′
1 + τjr
|
Ç
Ç
å
å
1 + τjr′
1 + τjr
τjr
τjr′
,
+
(1
−
α
)
r
′
′
r
r
1 + τjr
1 + τjr′
1 + τjr
1 + τjr
{z
}
|
{z
}
α
er
(10)
(1−e
αr )
so the ad-valorem district-preferred tariffs receive weights α
er and (1 − α
er ), respectively.
In the second stage, the agenda setter r chooses to form a coalition with district r′ , and
′
′
′′
′
implement the tariff vector trr if: (i) Ωr (trr ) ≥ Ωr (trr ), for r′′ ̸= r, r′ , and (ii) Ωr (trr ) ≥ Ωr (t).
′
Alternatively, r decides to maintain the status quo tariffs if Ωr (t) > Max{Ωr (trr ), for r′ ̸= r}.
The tariffs that emerge from this political bargaining process can be characterized as
follows. First, note that tariffs proposed by district r to district r′ (either (7) or (8)) can
be expressed as a weighted average of per capita specific factor output in districts r and
r′ . Second, this solution is a convex combination of the unconstrained tariffs preferred by
districts r and r′ . Third, the distance of the status quo tariff tj from r′ ’s unconstrained
maximum is a key consideration in the first step. The larger this distance, the larger the
multiplier ρr′ , and the greater the weight (1 − αr ). But this adversely affects r’s welfare
′
Ωr (trr ), and in the second step pushes the agenda setter to coalesce with the district less
constrained by the status quo.
In sum, since individual districts do not have the political power to impose their unconstrained preferred tariffs, they are required to coalesce with other districts. The vector of
tariffs ultimately approved and implemented at the national level is shaped by the influence
of districts belonging to the winning coalition. The form of the solution in equation (8)
generalizes to more than three districts (see Technical Appendix B).
15
See the Technical Appendix for detailed derivations.
15
Institutional background. The institutional setting under which U.S. tariff policymaking
has unfolded in recent history lends credibility to the model presented in previous sections.
Through the 1960s and 1970s, negotiating multilateral tariff cuts required each GATT member country to believe that the agreed-upon reciprocal cuts would be legislated by all their
GATT trading partners (Bagwell and Staiger, 1999, Irwin, 2017). In the U.S. such credibility resulted from the authority that Congress extended to President Kennedy via the
1962 Trade Expansion Act; this statute set the scope of the tariff cuts in manufacturing
and explicitly limited the liberalization of agriculture. Once the U.S. Trade Representative
(USTR) completed GATT negotiations on behalf of the Executive, the President brought
the proposal to Congress for a final up-or-out vote. This precedent prevailed when Congress
legislated the Trade Act of 1974, and (as in the 1962 Act) granted “fast track,” delegating
authority to President Ford to determine the tariff cuts to be negotiated during the Tokyo
Round. Fast-track, as in the canonical BF model, was subject to a closed rule vote –the
fast-track procedure meant the motion by the president would receive an up-or-out vote by
Congress, not subject to amendment.
In the ensuing sections, we introduce a centralized solution, where a “government” chooses
a vector of tariffs that maximizes a national welfare function. The solution, a vector of tariffs,
aggregates district tariff preferences in a way that is analytically tractable for estimating
district-specific welfare weights for owners of L and K. This is the primary goal of the
empirical analysis following the model’s prediction. The solution(s) also plausibly reconcile
the counterfactual district tariff preferences obtained in section 2.1.
3.2
A General Model
Trade policy is determined through a political process that aggregates the preferences of
districts, where welfare weights capture the political influence of districts and economic
actors. The political process is assumed to maximize the weighted sum of the individual
utilities of the population of factor owners:
Ω =
XX
r
K
ΓK
jr Wjr +
XX
r
j
L
ΓLjr Wjr
,
(11)
j
m
where Γm
jr is the weight attached to the welfare Wjr of the owner of m ∈ {L, K} employed in
producing good j in district r. The weights capture the impact of rich regional heterogeneity
m
in production and factor ownership on tariff-making. Wjr
depends on domestic prices p. In
this section, we consider the small country case, where (specific) tariff tj has no impact on
world price p̄j , and domestic price pj = p̄j + tj . Welfare for the two types of factor owners
16
are therefore fully functions of tariffs t. Expressing national welfare in (11) as the sum of its
three components yields
XX
Ω =
r
where
T
n
ΓLjr nLjr
j
Ç
å
ã XX
Å
T
πjr T
K K
Γjr njr
w0r + + ϕj +
+ + ϕj ,
n
n
nK
jr
r
j
(12)
is per capita tariff revenue and ϕj is per capita consumer surplus from the con-
sumption of all goods. Expression (12) is essentially a weighted sum of the district welfare
functions. National tariffs are obtained by maximizing (12) with respect to each tj . The
resulting per-unit (specific) tariff on imports of each good j is given by:
tj
P
where
r
"
#
Ç å
K
n X ΓK
qjr
Dj Mj
jr njr
= − ′
−
+
,
Mj r
γ
n
n
nK
jr
j = 1, . . . , J,
(13)
K
ΓK
jr njr
γ
is the share of the total welfare weight received by the nation’s owners of
P P
P P
K
specific factors employed in good j, γ L = j r ΓLjr nLjr , and γ K = j r ΓK
jr njr are the
aggregate welfare weights on non-specific and specific factors, respectively, and γ = γ K + γ L .
Dj
n
∂M
Mj
n
is per capita imports of good j, and Mj′ ≡ ∂tjj < 0.
Äp ä
Using good j’s import demand elasticity, ϵj = Mj′ Mjj , the market clearing condition Dj =
is per capita demand for good j,
Qj + Mj , and defining τj =
tj
,
pj
we have the following result about the ad-valorem national
tariff for good j.
Proposition 3 In terms of ad-valorem tariff, protection to good j is given by:
τj
n
=
1 + τj
−ϵj Mj
where
τj
1+τj
=
tj
pj
K
X ΓK
Qj
jr njr qjr
−
K
γ njr
n
r
!
Å
ã Å
ã
R
K
X
ΓK
qjr /Mj
Qj /Mj
jr njr n
=
−
,
γ
−ϵj
−ϵj
nK
jr
r=1
(14)
is the ad-valorem tariff applied to imports of good j.
A comparison with its district counterpart in (3) shows that (14) aggregates district tariff
preferences as a weighted sum. While the framework abstracts from how weight shares in (14)
are determined, the equilibrium national tariff aggregates district preferences like the BaronFerejohn solution in Section 3.1. A comparison with (8) shows their essentially similar form:
a weighted sum of the output-to-import ratio scaled by absolute import elasticity across
districts
the agenda-setter’s tariff proposal. The term
λK that form a majority and legislate
K nK
Γ
λK
αr λjrr in (8) is the counterpart to jrγ jr in (14). In words, λjrr is the share of district r’s
total welfare weight received by specific factors employed in the production of good j and
αr is district r’s share of aggregate (national) welfare weight. Their product is equal to the
17
aggregate welfare weight received by specific factors located in district r and employed in
the production of good j, or
K
ΓK
jr njr
.
γ
If the status quo utility for district r′ is relatively low compared to the utility it would
get under r’s proposal, then the Lagrange multiplier in (7) ρr′ is close to zero. That is, it
is “cheap” for district r to attract r′ to the coalition. As a result, αr = 1, and r proposes a
vector of tariffs that is the same as its preferred tariff vector (4), which r′ accepts. If, on the
other hand, the multiplier ρr′ > 0 and the constraint for r′ is binding, then 0 < αr < 1. This
means the agenda setter’s proposal must place positive weight on specific factors in district
r′ in the tariff determination.
As a final remark, suppose welfare weights are equal for all factors, goods, and districts,
so that Γm
jr = Γ, that is, political economy considerations do not influence the outcome.
Then, tariffs are zero and there is free trade.16 In Appendix A.2 we compare this result
with the Grossman and Helpman (1994) model predictions. As shown in the appendix,
removing districts from the model provides an interpretation of the GH coefficient a with
the parameters in our model.
3.3
Estimation Strategy in the Small Country Case
A primary goal of the paper is to estimate the welfare weight shares derived in Proposition 3.
The estimated weights would reveal which group of agents, districts, and goods were influential in determining the vector of tariffs prevailing in 2002. No doubt, history had much to do
with these tariffs –the Kennedy and Tokyo Rounds of tariff cuts through the 1960s and 70s
are reflected in the commodity composition of U.S. tariffs to this day.17 On December 27,
2001, President Bush signed a proclamation establishing permanent normal trading relations
(PNTR) with China, putting an end to the annual reviews of U.S.-China relations mandated
by the Jackson–Vanik amendment to the Trade Act of 1974. The authority to normalize
relations between the U.S. and China and the certification of the terms of China’s accession
to the WTO resulted from an act of Congress. Congress was aware that the decision to grant
MFN status to a large country like China (which took effect on January 1, 2002) effectively
moved U.S. tariffs out of their existing political-economic equilibrium. Subsequently, legislators introduced bills to terminate China’s MFN trade access to the U.S. market. During
the 107th Congress, for instance, H. J. Res. 50 terminating China’s conditional trade access
P P
m
L
K
To see this note that γ m = Γ r s nm
sr = Γn , γ = Γ(n + n ) = Γn, and (14) reduces to
P K
P
n
τj
Qj
jr qjr
n
= −ϵ1 j ( r qjr − Qj ) = 0. This result does not necessarily hold in the districtr n nK − n
1+τj = −ϵj
16
jr
preferred tariff case.
17
The 2007 World Trade Report (WTO 2007, Ch II.D) details the process of tariff cuts. See, for example,
Whalley (1985) for a CGE analysis of the process of tariff cutting in the Tokyo Round.
18
to the U.S. market was referred to the Ways and Means Committee, negatively reported to
the floor, and ultimately defeated by a 169-259 vote.18 Hence, while U.S. trade policy is
rooted in the reciprocal concessions negotiated under successive GATT Rounds, the vector
of tariffs prevailing in 2002 reflects the will of the legislative coalition at that time.
We attempt to characterize those coalitions and estimate the welfare weights on industries
and districts in the small country case, which has been the setting for the majority of
empirical studies of trade protection. The building block of the empirical strategy is to
estimate the welfare weight shares
3.3.1
K
ΓK
jr njr
(γ K +γ L )
using equation (14).
Specification and Identification
L
In (14) the number of parameters {ΓK
jr , Γjr }, r = 1, . . . , R, j = 1, . . . , J, is excessive. The
forming of coalitions resolves this problem.19 We estimate the industry and district-level
weights that would result from bargaining among plausible legislative coalitions. We consider
two stylized coalitions (i.e., aggregation of districts) founded, respectively, on (i) political
geography, reflecting the spatial clustering of industries in districts; and (ii) purely political
coalitions, based on the competitiveness of the state in the presidential election, and whether
the district’s election is competitive or safe for incumbent Democratic or Republican representatives. The latter grouping is intended to capture differential electoral incentives faced
by local representatives, parties, and the president. Without loss of generality, we continue
to use R to denote the number of coalitions of districts, or “regions” and r to index the
regions.
Expressing the demand-for-protection term in (14) with region r’s output-to-imports
ratios
qjr
,
Mjr
the tariff equation (14) can be rewritten in a form resembling the GH prediction
with regional output-to-import ratios. Since preferences are homogeneous, the imports of j
by region r (which are unobserved, only national imports of j are observed) are approximated
by distributing national imports of j according to r’s population share as Mjr = Mj × nnr .
18
See Congressional Research Service, CRS Report RL30225, “Most-Favored-Nation Status of the People’s
Republic of China,” June 7, 2001–July 25, 2001: Link (accessed 10/2022).
19
Public output data for districts is most completely available at NAICS 3-digits. NAICS-332 Printing
and Related Support Activities, a largely non-tradable industry, is dropped, leaving twenty manufacturing
industries, the upper bound on the number of estimable parameters.
19
Then, (14) may be written as:20
τj
1 + τj
Å
ã Å
ã
R
K
X
ΓK
qjr /Mjr
Qj /Mj
jr njr nr
=
−
.
γ nK
−ϵj
−ϵj
r
r=1
(15)
For the small country case, (15) provides the basis for industry-region welfare weights implied
by the observed vector of tariffs.21 We estimate the relative welfare weights
K
ΓK
jr njr
γ
using the
econometric specification
Å
ã
Å
ã
R
X
τj
qjr /Mjr
Qj /Mj
=
βr
+α
+ uj ,
1 + τj
−ϵ
−ϵ
j
j
r=1
(16)
with βr ≥ 0.22 The coefficient α on the national output-import ratio scaled by absolute
import elasticity is constrained to −1. The relative welfare weights are under-determined:
K
the R parameters βr do not identify the 2 × (J × R) industry-region welfare weights ΓK
jr njr
and ΓLjr nLjr . As in the district-specific counterfactual exercises, we assume the welfare weights
for specific factor owners have no within-region variation. That is, specific factors employed
K 23
If all specific factor
in all goods j produced in region r are treated the same, ΓK
jr = Γr .
owners were politically organized, or weights were assigned based on their (equal) voting
strength, this assumption is plausible. The corresponding assumption for owners of labor,
ΓLjr = ΓLr , is due to their mobility. Then, the coefficient βr is
βr =
where
nr
nK
r
K
K
ΓK
ΓK
n
r nr nr
r nr
P
P L L Kr ,
=
K
K
K
γ nr
( r Γr nr + r Γr nr ) nr
(17)
is the inverse of the proportion of region r’s population who are specific factor
L
owners. There are 2R parameters, ΓK
r and Γr , but for our purpose it is sufficient to recover
K
(R + 1) parameters: R welfare weights on specific capital in each region, ΓK
r nr , and the
P
collective economy-wide welfare weight on labor, γ L = r ΓLr nLr . This is straightforward
20
We abstract from the role of lobbying to focus on the two-level process by which U.S. trade policy is
determined. Lobbying may be incorporated as done in prior work testing the GH model (Goldberg and
Maggi, 1999, Gawande and Bandyopadhyay, 2000) as we show in Online Appendix B Section 1.3. Future
research can move the literature by measuring lobbying at the district-good level. This framework would
need to allow for lobbies to emerge endogenously, as in Mitra (1999). Lobbying could influence policy stances
at both district and national levels.
21
Modeling institutions that aggregate preferences, frame legislative bargaining rules, and regulate instruments of protection are a potential research agenda.
22
Import subsidies (negative tariffs) are disallowed. In the U.S. they are rarely, if at all, used in manufacturing. Any subsidies may be incorporated by admitting negative weights.
23
Lobbying structure distinguishing specific capital across goods is a potential research direction.
20
with estimates of βr in hand.
Arguably, the regressors
qjr /Mjr
−ϵj
are endogenous: in the specification (16), shocks to
the tariff τj can move the output-to-import ratio
qjr
Mjr
in region r. Shocks that increase
the tariff can lower Mjr and increase qjr ; negative tariff shocks, by liberalizing trade, can
have the opposite effect. This endogeneity can cause OLS estimates of the R coefficients
βr , r = 1, . . . , R to be biased.
Our strategy to identify coefficients on the endogenous regressors
qjr /Mjr
−ϵj
employs Bartik-
like instruments (Bartik, 1991, Goldsmith-Pinkham et al., 2020). To construct Bartik instruments (BIVs), we start by decomposing good j’s overall imports-to-output ratio using
the accounting identity
Mj
Mj1
Mj2
MjR
= zj1
+ zj2
+ . . . . . . + zjR
,
Qj
qj1
qj2
qjR
where zjr is region r’s share of output Qj , where for each j,
Pr=R
r=1
zjr = 1. The weights {zjr }
are constructed using district output data (aggregated up to “regions” that form political
coalitions). Let us construct the BIV for the (endogenous) variable for region 1,
qj1
.
Mj1
Rewrite
the identity as
1 Mj
zj2 Mj2
zjR MjR
Mj1
=
−
− ...... −
.
qj1
zj1 Qj
zj1 qj2
zj1 qjR
Now decompose both region r’s import penetration
Mj
Qj
Mjr
qjr
(18)
and the nation’s import penetration
as
Mjr
Mr fi
Mjr
Mj
M g
Mj
=
+
, and
=
+
,
qjr
qr
qjr
Qj
Q
Qj
where
Mr
qr
is region r’s overall import-output ratio and
component; similarly,
M
Q
g
Mjr
is the idiosyncratic good-region
f
M
is the nation’s aggregate import-output ratio and Qjj the idiosyn-
cratic component. The BIV for
Mj1
qj1
qjr
is constructed by substituting the non-idiosyncratic
components into the RHS of (18):
Å
Mj1
qj1
ãBIV
=
1 M
zj2 M2
zjR MR
−
− ...... −
.
zj1 Q
zj1 q2
zj1 qR
The BIV addresses the correlation between the idiosyncratic component and the structural
Mjr
error uj . For example, an unobservable variable that shocks both
and τj generates
qjr
M
endogeneity (Goldsmith-Pinkham et al., 2020, p. 2593). The general BIV for regressor qjrjr
21
is
Å
Mjr
qjr
ãBIV
d=R
X zjd Md
1 M
=
−
,
zjr Q
z
jr qd
d=1
(19)
where the sum is taken over d ̸= r.
The identifying assumptions may be cleanly described with two regions (i.e., r = 1, 2).
Then zj1 = 1 − zj2 and the BIV is
Å
Mj1
qj1
ãBIV
1
=
zj1
Å
M
M2
−
Q
q2
ã
+
M2
q2
The research design inherent in this 2-region case is that the (inverse) share
1
zj1
measures
exposure to a “policy” that affects region 1, and where the difference between the national
ÄM
ä
import-output ratio and region 2’s import-output ratio of good j, Qjj − Mq22 , is the size of
M
the policy ( Mq22 is a constant and does not vary with j). Instrumenting qj1j1 in the first stage
Ä M äBIV
with qj1j1
achieves identification from the differential exogenous exposure z1j1 .
We assume strict exogeneity of the inverse share, which is necessary for the Bartik estimator to be consistent. The identifying assumption in the 2-region example is that the
differential effect of higher exposure of one region only affects the change in the outcome
τj through the endogenous variable
Mj1
qj1
and not through any confounding channel. This
is clearly spelled out in the theory from which specification (16) is derived. Note that the
Ä
ä
Q
policy shock M
− Mq22 is constant, so the identifying variation comes solely from differential
exposure for each region separately.
In our more general case, we have R endogenous variables. Each is associated with the
BIV (19). We report the first stage estimates in Appendix A.1, which provides information
about the exposure design.24
4
National Tariffs in a Large Country
The political economy of trade literature, with few exceptions, has sublimated the role of
specific factors employed in exporting. The presumption has been that the primary trade
policy influencers must be import-competing producers since they stand to gain the most.
In the large country model, world prices are no longer exogenous. Tariffs can lower world
prices and worsen the terms of trade for exporters. Grossman and Helpman (1995) model
the interaction between two large countries and make the case for the terms of trade motive
24
q
q
BIV
In (15) the R output-to-import ratios Mjrjr are instrumented using the Bartik IVs Mjrjr
Ä
Pd=R zjd Md ä
z
1/ z1jr M
−
, Identifying variation comes (nonlinearly) from output share ratios zjd
.
d=1 zjr qd
Q
jr
22
=
for tariffs (in addition to the special interest motive). Bagwell and Staiger (1999) view the
emergence of trade liberalizing institutions like the GATT as a commitment by countries
to avoid a global race to the bottom where countries impose terms of trade externalities on
each other.
We present a model highlighting the role of domestic specific factors employed in producing export goods as a countervailing influence against protecting domestic import-competing
goods. The threat of retaliation by partners, reflected in the tariffs and incidence of granting
China preferential access presented on Tables 1 and 2 of the 2001 CRS Report, and the consequent worsening of terms of trade for U.S. exporters is the primary motive for exporters
to force trade liberalization.25
4.1
The Model
Consider a world with two countries, U S and RoW , and three types of goods: a numeraire
(good 0), importable, and exportable goods. From the perspective of country U S, there are
J import goods (M -sector) indexed by j, j ∈ M, and G export goods (X-sector) indexed
0
M
X
by g, g ∈ X . The three sectors employ nL = nL + nL + nL units of labor, where
P P
P P
P
0
M
X
X
X
M
0
M
0
nL = r nrL , nL = r j∈M nLjr , nL = r g∈X nLgr , and nK = nL + nL + nL
P P
P P
M
M
X
X
units of specific factors, where nK = r j∈M nK
and nK = r g∈X nK
jr
gr . Total
M
X
employment is n = nL + nK + nK .
On the demand side, consumer surplus from the M and X sectors are ϕj = uj (dj ) − pj dj
and ϕg = ug (dg ) − pg dg . In this two-country world, imports of good j, Mj (respectively,
exports of good g, Xg ) by U S are equal to exports of good j, Xj∗ (respectively, imports of
good g, Mg∗ ) by RoW . Therefore, the market clearing conditions are Dj − Qj = Q∗j − Dj∗
(> 0), and Dg − Qg = Q∗g − Dg∗ (< 0), where asterisks refer RoW ’s output and demand in
exporting and import-competing goods.
U S may impose an ad valorem tariff τj =
(pj −pj )
pj
on imports of good j, so that the
domestic price of j in U S is pj = (1 + τj )pj . Tariffs generate a tariff revenue of T =
P M M
i τi pi Mi , where T ≥ 0, since export subsidies are not allowed. As before, tariff revenue
is distributed back to all domestic residents in a lump-sum way.
The world price of good j, pj , is implicitly determined by the market clearing condition,
Mj [(1 + τj )pj ] − Xj∗ (pj ) = 0, making pj a function of τj . Export subsidies are disallowed,
so the domestic price prevailing in RoW is simply p∗j = pj .26 Reciprocally, if RoW imposes
25
Relevant at that time, the Jackson-Vanik amendment and Title IV procedure provided Congress with
a statutory basis for continuing in force or (unilaterally) withdrawing China’s MFN status. Back-of-theenvelope calculation of losses to exporters, if China retaliated, are in CRS Report.
26
U S chooses τj ≥ 0. In RoW , τj∗ = 0 since it does not subsidize its exports of j.
23
tariff τg∗ on U S exports of good g, its price in RoW is p∗g = (1 + τg∗ )pg , where pg is g’s world
price determined by market clearing, Mg∗ [(1 + τg∗ )pg ] − Xg (pg ) = 0. The price of good g in
the U.S. is the world price, pg = pg .
Aggregate welfare in U S is the sum of welfare of owners of the mobile factor and owners
0
M
X
M
X
of specific factors, or Ω = ΩL + ΩK = ΩL + ΩL + ΩL + ΩK + ΩK , where
!
L0
X
ΩL =
L0
Γr n0r w0r +
r
Ω =
LM
LM
Γjr njr w0r +
j∈M
"
K
X
X X
r
M KM
ΓK
jr njr
Ç
j∈M
X
LX
LX
Γgr ngr w0r
+ γ L Υ,
(20)
g∈X
M
(pj )
πjr
M
nK
jr
å
+
X
X KX
ΓK
gr ngr
Ç
g∈X
X X
(pg )
πgr
X
nK
gr
å#
+ γ L Υ.
P
P
P P
T
X X
KM KM
K
The previous expression uses Υ = j∈M ϕM
j (pj )+
g∈X ϕg (pg )+ n , γ =
r
j∈M Γjr njr +
P P
P L0 L
P P
P P
KX KX
L
LM LM
LX LX
r
g∈X Γgr ngr , γ =
r Γr n0r +
r
j∈M Γjr njr +
r
g∈X Γgr ngr , and γ =
γ L + γ K . To estimate the welfare weights, we will assume they differ between the importable
0
M
M
X
X
and exportable sectors, but not within each sector. That is, ΓLr = ΓLjr = ΓLr , ΓLgr = ΓLr ,
M
M
X
X
K
K
ΓK
, and ΓK
for all j ∈ M, g ∈ X .
jr = Γr
gr = Γr
Nash Bargaining. Tariffs are determined in a Nash bargaining game between U S and RoW
that makes explicit the possibility of a retaliatory response to a tariff. The equilibrium vectors
Ä
ä Ä
ä
US σ
RoW (1−σ)
of tariffs {τ , τ ∗ } maximize ΩU S − Ω
ΩRoW − Ω
, where τ = (τ1 , ..., τj , ..., τJ ),
and τ ∗ = (τ1∗ , ..., τg∗ , ..., τG∗ ). The FOCs (at an interior solution) with respect to each τj chosen
by U S and τg∗ chosen by RoW are (taking the tariffs of the other country as given):
τj : Ä
τg∗ : Ä
where
dΩU S
dτj
=
∂ΩU S ∂pj
∂pj ∂τj
ä
US
dΩU S
(1 − σ)
dΩRoW
ä
+Ä
= 0,
RoW
dτj
dτj
ΩRoW − Ω
ä
US
dΩU S
(1 − σ)
dΩRoW
Ä
ä
+
= 0,
RoW
dτg∗
dτg∗
ΩRoW − Ω
σ
ΩU S − Ω
σ
ΩU S − Ω
+
∂ΩU S
∂τj
and
dΩU S
dτg∗
=
dΩU S
dΩU S
−
dτj
dτg∗
If U S is a small country,
∂pj
∂τj
ñ
∂ΩU S ∂pg
.
∂pg ∂τg∗
Rearranging and taking the ratio,
ô
dΩRoW /dτj
= 0.
dΩRoW /dτg∗
= 0, eliminating any interaction such as in (21).
24
(21)
We will consider the large country case where U S exports a single good g.27 To gain
insight into (21), suppose U S and RoW come to an agreement that when U S raises a tariff
on RoW ’s exports of j, RoW is entitled to increase its tariff on U S exports of g to keep
RoW ’s utility at its pre-existing level (i.e., before the increase in tariffs). The amount by
dΩRoW /dτ
which RoW increases τg∗ to keep ΩRoW at its status quo is given by − dΩRoW /dτ j∗ =
g
dτg∗
.
dτj
The
change in RoW ’s tariff on U S exports of g in reaction to the U S tariff increase is a gauge of
the “bargaining strength” of U S relative to RoW with respect to τj , denoted by µj , where
µj =
dτg∗
.
dτj
The equilibrium τj and τg∗ under such an agreement are determined endogenously
by (21) (and the corresponding expression for RoW ).28
The separate influence of specific factors in the export sector requires the welfare weights
of specific factors employed in import-competing sectors to differ from the welfare weights
of specific factors employed in the export sector. We denote these welfare weights for each
district r by ΓK
r
M
X
and ΓK
r , respectively.
Decomposing the impact of a change in τj . In the import-competing sector, a change
in τj indirectly affects ΩU S through its impact on the domestic price pj :
∂ΩU S X K M K M
Γr nr
=
∂pj
r
where nK
r
M
Å
qjr
M
nK
r
ã
−
γ
γ
′
Dj + τj pM
j Mj ,
n
n
(22)
is employment of specific factors in the M sector in district r. The first term
in (22) captures the impact of a change in pj on producers, the second term, its impact on
27
The model generalizes to many export goods (Online Appendix B). The counterpart to (21) is
ñ
ô
X dΩU S
dΩU S
dΩRoW /dτj
− P
= 0.
RoW /dτ ∗
dτj
dτg∗
g
g dΩ
g
RoW can retaliate by potentially increasing tariffs, τ ∗ , on all U S exports. The (negative of the) term in
dΩRoW /dτj
RoW /dτ ∗ .
g
g dΩ
square brackets represents U S bargaining strength with respect to τj , µj ≡ − P
28
A rise in τj by U S reduces RoW ’s utility. The logic of the “agreement” is that it allows RoW to
compensate for this decline: RoW is allowed to increase its tariff τg∗ on U S exports to keep RoW ’s utility
constant before the increase in the U S tariff. Let ΩRoW (τj , τg∗ ) denote the indirect welfare function for RoW ,
b RoW = ΩRoW (τj , τg∗ ) for an
where ∂ΩRoW /∂τj < 0 and ∂ΩRoW /∂τg∗ > 0. The agreement would state that Ω
RoW
b
agreed-upon status quo utility Ω
(reciprocally also for U S). Then,
∂ΩRoW ∗
∂ΩRoW
dτj +
dτg = 0
∂τj
∂τg∗
⇒
dτg∗
∂ΩRoW /∂τj
.
= − RoW
dτj
∂Ω
/∂τg∗
In general, dτg∗ /dτj represents the slope of RoW ’s reaction function evaluated at the equilibrium tariffs
(dτg∗ /dτj is the quotient of the two expressions immediately before (21), but for RoW instead of U S). Since
bargaining strength µj is not measurable, Online Appendix A.3 provides a sensitivity analysis to a range of
its possible values.
25
consumer surplus, and the third term the (indirect) effect on tariff revenue T = τj pj Mj . A
change in τj also affects T , and consequently ΩU S , both directly and indirectly through its
impact on the world price pj as follows:
Å
ã
∂pj
γ
∂ΩU S
γ ∂T
γ M
p Mj + τj Mj
=
=
.
∂τj
n ∂τj
n j
n
∂τj
(23)
Finally, the change in tariffs by U S triggers a response by RoW : RoW modifies the tariff
on U S exports of good g, τg∗ , which in turn affects g’s equilibrium world price. The latter
has an impact on producers and consumers of g scattered across U S districts, which is given
by
∂ΩU S X K X K X
Γr nr
=
∂pg
r
Å
qgr
X
nK
r
ã
−
γ X
D ,
n g
(24)
X
where nK
is the employment of specific factors in the X sector in district r,
r
KX
KX
per unit of specific factor, which gets a welfare weight Γr nr , and
γ
n
qgr
X
nK
r
is output
is the welfare weight
on the representative consumer. A decrease in the world price of U S export good g due to
a (retaliatory) tariff increase by RoW is the negative of this expression. The solution to the
Nash bargaining game is stated in this proposition.
Proposition 4 The tariff on good j in the two-country bargaining game satisfies
ã X
ã
ãÅ
ã
Å
M KM Å
X KX Å
R
R
X
ΓK
nr
ΓK
n
n
qjr /Mj
qgr /Mj
r
r nr
=
+
µj θjg
M
X
γ
−δj
γ
−δj
nK
nK
r
r
r=1
r=1
Å
ã
ã
Å
Qj /Mj
1
Dg /Mj
− µj θjg
−
+
,
(25)
X∗
−δj
−δj
1 + ϵj
τj
1 + τj
(pj −pj )
pj
where τj =
P
r
M
ΓK
nK
r
r
γ
is the ad-valorem tariff applied to imports of good j,
τj
(1+τj )
=
(pj −pj )
,
pj
M
is the share of the national welfare weight received by specific capital employed in
producing the nation’s import-competing goods, and
P
r
X
ΓK
nK
r
r
γ
X
is the share of the national
welfare weight received by specificcapital employed
in producing the nation’s export good.
∂X ∗ p
∗
∂M p
1
L
K
M
M
Further, γ = γ + γ , δj = ϵj ϵX ∗ + 1 < 0, ϵj = ∂pjj Mjj < 0, ϵX
= ∂p j Xj∗ > 0,
j
θjg =
∂pg /∂τg∗
∂pj /∂τj
j
j
< 0, and µj =
dΩRoW /dτ
− dΩRoW /dτ j∗
g
j
> 0.
Proof Result (25) is obtained by substituting expressions (22), (23), and (24) into (21), and
isolating τj . We then divide both sides by (1 + τj ) =
26
pj
pj
complete elasticities. These expres-
sions use the results
∂pj
∂τj
=
ϵX
j
∗
> 0, and
M
ϵX∗
j −ϵj
∂pg
∂τg∗
∗
=
ϵM
g
∗
X
ϵg −ϵM
g
< 0, obtained by differentiating
the market clearing conditions Mj [(1 + τj )pj ] − Xj∗ (pj ) = 0 and Mg [(1 + τg∗ )pg ] − Xg (pg ) = 0
∗
X
and the elasticities ϵM
j and ϵj .
□
The two terms on the right-hand side of the imports-only (small country) case (14) also
appear in (25), except that the absolute import elasticity −ϵM
j is now replaced by −δj . In
the large country case, −δj incorporates the response along RoW ’s export supply function as
the international price pj changes. The tariff τj is lower than it would be in the small country
case (−δj > −ϵM
j ). Three additional terms for the large country case appear in (25). The
Ä q /M ä
X KX
P ΓK
gr
j
n
< 0, is the demand by specific capital owners
first term, r r γnr
µ
θ
j jg
−δj
nK X
r
in the export sector for a reduction in τj in response to the threat of retaliation by RoW
on exports of g (θjg < 0). The second term,
1
∗,
1+ϵX
j
accounts for the impact of tariffs on the
Ä D /M ä
g
j
equilibrium world price of good j, and the third term, −µj θjg −δ
> 0, is the (beneficial)
j
effect of a retaliatory tariff by RoW (in response to an increase in τj ) for U.S. consumers of
the exportable.
4.2
Estimation Strategy in the Large Country Case
How significant were U.S. export interests in the minds of policymakers determining 2002 U.S.
tariffs? The share of the aggregate welfare weight received by specific capital employed in
X
producing the export good g,
ΓK
nK
r
r
γ
X
, quantifies the impact of export interests in liberalizing
trade. By estimating this expression, we provide a possible answer to this key question in
the political economy of trade policy literature.
An econometric specification to estimate the relative welfare weights
ΓK
r
M
nK
r
γ
M
X
and
ΓrK nK
r
γ
X
based on Proposition 4 is
τj
1 + τj
=
R
X
Å
βr
r=1
Ç
+α
qjr /Mjr
−δj
ã
+β
X
Å
Qg /Mj
µj θjg
−δj
ã
Qj /Mj
1
Dg /Mj
−
+ µj θjg
X∗
−δj
−δj
1 + ϵj
where βr ≥ 0 and β X ≥ 0.29 The (R + 1) coefficients βr =
ΓK
r
M
nK
r
γ
M
å
+ uj ,
nr
M
nK
r
(26)
and β X =
X
ΓK n
γ
are
estimable with our data.
All elasticity measures are from Nicita et al. (2018) (NOP). The variable δj =
ϵM
j
1
ϵX
j
∗
+1
is computed using NOP’s estimates, at HS 6-digits, of the elasticity of RoW ’s export sup29
Weights are constrained to be non-negative. Import subsidies on the j-goods and export tax on good g,
which can lead to negative tariffs, are both disallowed.
27
∗
M
ply of good j to the U.S., ϵX
j , and good j’s U.S. import demand elasticity, ϵj . In (25),
both
Dg
Mj
and
qgr
Mj
are ratios of quantities of different goods, while their data are in values.30
Multiplying by the price ratio
Using
∂pj
dτj
=
∗
ϵX
j
M
ϵX∗
j −ϵj
pg
pj
> 0 and
converts them to ratios of values.
dpg
dτg∗
=
∗
ϵM
g
∗
X
ϵg −ϵM
g
< 0, we denote θjg =
where
dpg /dτg∗
.
dpj /dτj
p
Let θjg = θejg × pgj ,
∗
pj /pj
×
θejg = ∗
pg /pg
X
ϵM
g /ϵg
∗ X
1−ϵM
/ϵ
g
g
1
X∗
1−ϵM
j /ϵj
< 0.
(27)
∗
In this expression, ϵM
is RoW ’s import demand elasticity for good g and ϵX
g
g is its U S export
∗
X
are defined correspondingly for U.S. import good j. Note
supply elasticity, and ϵM
j and ϵj
p
p D
Dg
that θejg is unit-free and pgj converts M
to the ratio of measurables pjgMgj .31 We use NOP’s
j
∗
(RoW ’s import demand elasticity of good g) and ϵX
estimates for ϵM
g (U S export supply
g
32
elasticity of exports of good g to RoW ) to measure θejg .
Additionally, model (26) imposes α = −1. In going from Proposition 4 to (26) we assume
that specific factors employed in producing the export good g coalesce nationally, equalizing
X
X
= ΓK .33 We
welfare weight of each specific factor owner in the export sector, that is, ΓK
r
will estimate the relative welfare weights
M KM
ΓK
nr
r
γ
and
X KX
ΓK
nr
r
γ
by 2SLS using the Bartik-like
IVs described in Section 3.3.1.
5
Results: Trade Policy Influencers
We empirically examine both the small country model (16) and the large country model
(26) using two hypothetical legislative “coalitions” in the determination of U.S. tariff policy.
Case 1 groups the congressional districts into nine geographical regions: New England, MidAtlantic, South Atlantic, East North Central, West North-Central, East South Central, West
South Central, Mountain, and Pacific. Case 2 aggregates districts into nine blocs (R = 9)
according to purely political factors. The classification of districts is based on the electoral
competitiveness of their state in the 2000 presidential election, the competitiveness of the
district in the congressional race closest (and prior) to 2002, and the party that carried the
state in the presidential election and the district in the congressional race. Electoral motives
of the national party drive coalitions for trade policymaking.
30
31
32
Other ratios in (25) have the same good in the numerator and denominator.
∗
See Online Appendix B for more details. The numerator is negative since ϵM
< 0.
g
p /p
The ratio p∗j /pj is set to 1 in the estimation. Results are robust to a range of values.
g
g
33
As discussed in the conclusion, access to highly disaggregated (confidential) geographic area series from
the U.S. Census would allow us to estimate a larger set of parameters.
28
5.1
Geography-Based Coalitions
Table 1 presents descriptive statistics of the variables in the small- and large-country regression models (16) and (26) with geographic coalitions (Case 1). The first two columns show
the number of districts in each coalition and the proportion of the population of workers
–labor and specific factor owners– in each bloc.
Table 1: Descriptive Statistics: Variable Means
Small Country
qjr /Mjr
nr
Districts
n
−ϵj
23
0.060
1.11
65
0.125
1.35
73
0.243
1.22
31
0.067
1.39
75
0.139
1.72
26
0.060
1.59
47
0.096
1.39
24
0.043
1.26
69
0.167
1.11
1.33
New England
Mid-Atlantic
East North Central
West North Central
South Atlantic
East South Central
West South Central
Mountain
Pacific
Qj /Mj
−ϵj
Qg /Mj
−δj
Qj /Mj
1
−
∗
−δj
1+ϵX
j
Large Country
qjr /Mjr
−δj
0.59
0.72
0.63
0.75
0.95
0.82
0.73
0.65
0.58
−0.13
µj θjg
+
Dg /Mj
µj θjg −δ
j
0.31
9, 454
N
8, 735
Notes: (1) nnr is the total employment shares for each region r. (2) In the Large Country case, the export sector NAICS=334
(Computers) is not in the sample, so N = 8735. (3) The 433 districts (out of the 435) for which we assembled output, trade,
protection, and employment data are classified into nine geographical blocs according to the U.S. Census. Division 1: New
England (Connecticut, Maine, Massachusetts, New Hampshire, Rhode Island, and Vermont). Division 2: Mid-Atlantic (New
Jersey, New York, and Pennsylvania). Division 3: East North Central (Illinois, Indiana, Michigan, Ohio, and Wisconsin).
Division 4: West North Central (Iowa, Kansas, Minnesota, Missouri, Nebraska, North Dakota, and South Dakota). Division
5: South Atlantic (Delaware, Florida, Georgia, Maryland, North Carolina, South Carolina, Virginia, District of Columbia, and
West Virginia). Division 6: East South Central (Alabama, Kentucky, Mississippi, and Tennessee). Division 7: West South
Central (Arkansas, Louisiana, Oklahoma, and Texas). Division 8: Mountain (Arizona, Colorado, Idaho, Montana, Nevada,
New Mexico, Utah, and Wyoming). Division 9: Pacific (Alaska, California, Hawaii, Oregon, and Washington). The column
“Districts” indicates the number of districts in each “coalition.”
Table 2 reports 2SLS estimates of coefficients βr in (16) and (26). They are constrained
to be non-negative as import subsidies and export taxes are ruled out. The small country
model (16) requires the coefficient on
Qj /Mj
−ϵj
to be constrained to −1, and the large country
model (26) requires the same of the coefficient on
Qj /Mj
−δj
−
1
∗
1+ϵX
j
+ µj θjg
Dg /Mj
.
−δj
First-stage
statistics indicate that the BIVs do not suffer from a weak-instrument problem.34
The small country estimates indicate positive welfare weights on specific factors (in
import-competing goods) in eight of the nine regions (coalitions of districts).35 The majority
34
First-stage regressions for Table 2 are reported in Tables A.1.2 and A.1.3 of Appendix A.1.
Errors are clustered at the HS 2-digit level of 94 goods. Evidence for clustering of the 9454 HS 8-digit
tariffs at a more aggregate level is in Conconi et al. (2014) and also implied by the huge number of industrylevel studies of protection. Presumably, these are administratively translated to HS 8-digit by replicating
the clustered tariff at this “line level.” Abadie et al. (2023) suggest that the decision to cluster and at
35
29
Table 2: 2SLS Estimates of Coefficients in (16) and (26) for Geography-based Coalitions
Dependent Variable: Applied Tariff, 2002
Small Country
Eq. (16)
0.067
(0.027)
0.163
(0.012)
0.216
(0.025)
0.063
(0.009)
0.140
(0.008)
0.089
(0.020)
0.073
(0.010)
0
β1 : New England
β2 : Mid-Atlantic
β3 : East North Central
β4 : West North Central
β5 : South Atlantic
β6 : East South Central
β7 : West South Central
β8 : Mountain
β9 : Pacific
β X : µj θjg
α:
α:
Qj /Mj
−ϵj
Qj /Mj
−δj
0.214
(0.019)
Qg /Mj
−δj
Qgr
Qr
Large Country
0.21
Eq. (26)
0
0.10
0
0.04
0
0.08
0.292
(0.017)
0.264
(0.020)
0
0.09
0.03
0.12
0.25
0.060
(0.017)
0
0.25
0
3.243
(0.359)
−1
−
1
∗
1+ϵX
j
+
Dg /Mj
µj θjg −δ
j
−1
N
First Stage Statistics
Anderson-Rubin χ2 (10 df)
Anderson-Rubin p-value
Kleibergen-Paap weak IV
9454
8735
2949.0
(0.00)
102.5
2010.0
(0.00)
937.5
Notes: (1) Standard errors in parentheses, clustered at 2-digit HS. (2) α is constrained to equal −1 required by (16) and (26).
Q
(3) (16) and (26) require dropping the constant term in the regressions. (4) Qgr is the share of the output of export industry
r
COMPUTER (3-digit NAICS=334) for coalition r. Larger shares (blue) suggest export coalitions. (6) In the large country
case: (i) unconstrained estimates of β1 , β2 , β3 , β6 , β8 and β9 are negative and constrained to zero to disallow import subsidies
or export taxes. (ii) µj is assumed to equal 1 (equal bargaining strength) for all j. (iii) θjg is calculated as in (27).
what level be determined by both sampling and design. The HS 8-digit sample is the entire population
tariff line products. Unlike field experiments which (randomly) sample micro-units from a few clusters in a
population, our sample includes all clusters of the population of interest. The first step in accounting for
clustering is to determine the clustering in the population. Based on the account of policymakers and the
above studies, it is reasonable to suppose that tariff decisions are taken up in clusters of (the 94) HS 2-digit
level product-groups. That is, “assignment to treatment” by policymakers, which is unobserved, occurs at
HS 2-digits. Abadie et al. (2023) suggest that the decision to cluster standard errors depends on whether this
within-cluster assignment is perfectly correlated (in which case, use clustered standard errors), uncorrelated
(that is, random assignment, in which case use cluster-robust standard errors) or imperfectly correlated (use
the Abadie et al. (2023) bootstrap procedure). We consider the assignment within HS 2 digits to be nearly
perfect (for example, within the HS 2-digit Apparel and Textile group, all HS 8-digit units are assigned to
treatment and receive a positive tariff outcome (which may be different across the 8-digit units). This errs
on the conservative side, so standard errors are overstated compared to the zero correlation or imperfect
30
of empirical studies of protectionism have been predicated on the small country assumption,
most notably the tests of the Grossman and Helpman (1994) model. One interpretation of
the result is that coefficients indicate coalitions of districts that influence tariff-making (positive) versus coalitions of districts that do not move the agenda and are expendable (zero). In
the large country case, protectionist interests find themselves pitted against domestic export
interests who make their presence felt in legislative bargaining. Their anti-protectionism is
due to concern about retaliation by RoW and the terms of trade effects inflicted on them. In
the regression, their inclusion explains both zeros and low tariffs in the data. Facing export
interests, the welfare weights on specific factors employed in import-competing goods are
non-zero in only three of the nine regions. A primary contribution of the paper is this finding that is missing in the majority of studies about protection. The missing export variables
are crucial to any explanation for why U.S. tariffs are low.
What do the estimates imply about the distribution of welfare weights across owners of
specific capital in the nine regions? Table 3 provides the answers. In the small country
model exclusively representing import-competing interests, their aggregate share of welfare
weights is 0.316, with the remainder going to mobile factor owners. With large populations
of specific factor owners, Mid-Atlantic, East North Central, South Atlantic, and Pacific have
the largest weight shares. Specific factors in Mountain get zero weight.
The relative weight
ΓK
r
ΓL
r
on an owner of specific capital versus an owner of a mobile factor
reflects the importance granted to the interests of specific factors in the tariff determination
process. In five of the nine regions, specific factors receive more favorable treatment. The
legislative bargaining interpretation is that it takes these five blocs to create a winning
coalition. Specific capital owners in the Mid-Atlantic, East South Central, South Atlantic,
and Pacific blocs receive the most favorable treatment relative to mobile factor owners.
Viewed through the Baron-Ferejohn lens, the median district belongs to the South Atlantic
bloc. Adding up the number of districts from Table 1 in descending order of
the
218th
ΓK
r
ΓL
r
indicates
district is in region 5. Districts in the remaining regions (3, 4, 7, 8) are inessential
and the preferences of specific factors residing there are ignored. A free-trade bias in the
agenda setter’s tariffs is in evidence, as districts in the industrial East North Central, most
in need of protection, are not in the winning coalition.
The large country model showcases export interests employed in the Computers industry (“COMP”), classified as NAICS 3-digit code 334, who compete with import-competing
interests employed in the remaining 3-digit NAICS industries.36 The “KrM -share” columns
correlation cases.
36
Our model follows the tradition of one-way trade models (Grossman and Helpman, 1994), where either
31
Table 3: Welfare Weights on Specific Capital Owners: from 2SLS Estimates in Table 2
Dependent Variable: Applied Tariff, 2002
Small Country
Region
1. New England
2. Mid-Atlantic
3. East North Central
4. West North Central
5. South Atlantic
6. East South Central
7. West South Central
8. Mountain
9. Pacific
Agg./Relative Weights
Kr -share
(estimated)
0.023
0.051
0.063
0.019
0.040
0.024
0.023
0
0.073
0.316
Large Country
M
ΓK
r
ΓL
KrM -share
1.136
1.314
0.899
0.941
1.019
1.493
0.766
0
1.300
(estimated)
0
0
0
0.075
0.063
0
0.016
0
0
0.154
ΓK
r
ΓL
X
ΓK
ΓL
K X -share
(imputed)
0
0
0
4.646
2.036
0
0.675
0
0
0.204
3.485
Notes: (1) Small country case: Specific factors employed in import-competing sectors determine tariffs. The proportion of
non-production workers in a NAICS 3-digit industry measures the proportion of specific factors in the industry. The weighted
average of these proportions (weights are region r’s output composition across the NAICS 3-digit industries), measures the
nK
r
nr
. In the Table, (i) Kr -share is the proportion of the
P
ΓK nK
L
national weight placed on region r’s specific capital owners,
= P ΓK rnK r+ΓL nL , where nL = r nL
r and Γ is invariant
rPr
r
across regions. (ii) The aggregate share of weights on specific factors
r γr is 0.316. The remainder, 0.682, is the aggregate
proportion of region r’s population that are specific factor owners
γrK
weight on labor’s welfare γ L . (iii) Relative weights
L
ΓK
r
ΓL
are calculated by dividing Kr -share by the aggregate labor weight share
n
nK
r
. (2) Large country case: Specific factors employed in both import-competing and export-producing
P KM KM
P
X
X
L
sectors. Aggregate weight on agents’ welfare is γ =
nr
+ ΓK
nK
+ r ΓL
r
r
r nr . The proportion of region r’s
r Γr
and multiplying by
X
M
population owning specific capital in the import-competing and export sectors
nK
r
nr
and
nK
r
nr
, respectively, are determined
similarly as in the small country case above. In the Table, (i) KrM -share is the proportion of the national weight placed on region
r’s specific capital owners employed in manufacturing import-competing goods,
ΓK
r
M
nK
r
γ
M
. The welfare-weight share of specific
capital employed in import-competing goods is 0.154 (in contrast to 0.316 in the small-country case). (ii) K X -share is the share
of aggregate welfare weight placed on specific capital employed in the export industry “COMPUTER,”
ΓK
X
nK
γ
is the total employment of specific capital in “COMPUTER.” From Table 2, β̃ X = 3.243, the estimate for
Multiplying by
n
KX
n
X
KX
Γ
γ
, where nK
n
X
from (26).
= 0.063 yields the share 0.204 reported in the bottom row. The remainder 1 − 0.154 − 0.204 = 0.642 is
the aggregate weight share of labor. (iii) The relative weights
M
ΓK
r
ΓL
are calculated as described in the small country case.
indicate zero weights for specific factors employed in import-competing goods in all but the
three regions: West North Central, South Atlantic, and West South Central. The first significantly different finding from the small country case is the sharply lower weight share to K M
owners in the aggregate, equal to 0.154. The second significant finding is the large welfare
weight share to K X owners, equal to 0.204. Specific factors on both sides of tariff protection
get a total welfare weight share equal to 0.358.
An interpretation of the result is that the presence of anti-protection export interests reduces the need to satisfy coalitions of protectionist districts. Thus, the agenda setter needs
to add only “cheap dates” to exporter coalitions and ignore the strong demands for protecthe good/industry is entirely import-competing or exporting, but not both. A significant extension would
model industry with two-way trade in differentiated goods(Krugman, 1981).
32
tion from districts in the East North Central bloc, which receive zero weight. From the
Baron-Ferejohn lens, a strategy for the agenda setter is to form the winning coalition with
export-oriented blocs and then satisfy protectionist coalitions, in the cheapest way possible,
to form a majority. Based on the share of the export industry COMP in the region’s total
manufacturing output ( QQgrr in Table A.1.4 in Appendix A.1), the export coalitions consist of
New England, Mountain, and Pacific, totaling 116 districts. The agenda setter only needs to
satisfy the protectionist demands of regions 4 and 5 (106 more districts) for a majority. Relative to the industrial Midwest (East North Central region) where the demand for protection
is the most intense, the “cheaper dates” produce a majority that puts East North Central
in the losing coalition. The cheap date hypothesis plausibly explains why specific factors in
the less populous West North Central region get a larger-than-commensurate welfare weight
M
(their high
ΓK
r
ΓL
weight).
The third significant finding is the large weight placed on an individual specific-factor
X
owner in Computers relative to labor,
ΓK
ΓL
= 3.485. The implication is that the legislative
bargain determining U.S. tariffs is won by export interests. They handily defeat manufacturing interests in the remaining (import-competing) industries. This representation of export
interests in our model leads to a variable that is a key determinant of low U.S. tariffs, thus
far absent in the literature. The missing variable can account for low overall U.S. tariffs and
the large number
of tariff lines (70 percent) with zero tariffs.
Qj /Mj
Dg /Mj
The term
− 1+ϵ1X ∗ + µj θjg −δ
in (26), whose coefficient is constrained to −1,
−δj
j
j
plays an important role in the results.37 The three terms move tariffs in sometimes opposite
directions. The optimal tariff,
1
∗,
1+ϵX
j
whose values vary between 0.16 and 0.71, would increase
U.S. tariffs by an order of magnitude (its mean is 0.38 compared to the mean U.S. tariffs
equal to 0.029 in 2002). On the other hand, the harm to consumer welfare from tariffs on
imports,
Qj /Mj
,
−δj
calls for lower tariffs. In the net, the sum of the three components varies
between −1.35 and 1.81 with a mean of 0.29. If its variation dominated the variation in
tariffs, then the results would be driven largely by this constraint. That is, the portion of
tariffs explained by import-competing special interest variables would be of second-order
importance relative to concerns about consumer welfare and the optimal tariff. This is the
case with U.S. tariffs and is reflected in the low weights received by special interests in the
import-competing sector. Applying the model to countries with high tariffs (for instance,
37
The coefficient −1 implies that:
1
∗
1+δjX
Qj /Mj
|δj |
lowers tariffs (concern for consumer welfare) on average by 0.81;
raises tariffs (imposition of optimal tariff) on average by 0.38 and
effect of RoW retaliation) on average by 0.14.
33
µj θjg (Dg /Mj )
|δj |
lowers tariffs (TOT
India before its 1990s liberalization) would more appropriately highlight the role of special
interests in India’s protectionism before liberalization, and the influence of export interests
in the liberalization.
5.2
Coalitions Based on Electoral Dynamics
Case 1 ignores the long-held view that the primary motive for building strong parties is
precisely to unify party-based coalitions during legislative bargaining. Case 2 aggregates
districts into stylized electoral coalitions based on how states voted in the 2000 presidential
elections (reflecting incentives faced by the Executive Branch in the formation of trade policy)
and how districts voted that same year in elections to the House of Representatives (home
of the agenda setters such as House Ways and Means and other committee chairs). Districts
are formed into nine blocs (R = 9), combining election outcomes and the party winning
the state or district. Districts in states where a party won more than 52 percent of the
votes in the presidential election are coded as safe for the winning party; they are considered
competitive otherwise. Districts in which a candidate to the House won by more than 52
percent of the vote are considered safe for the winning party. Otherwise, they are considered
competitive in the House elections.
Table 4: Districts, by Political Blocs Based on 2000 Election Outcomes
State-Wide Vote in
Presidential Election
Competitive
Safe Democrat
Safe Republican
Total
Districts in House Elections
Competitive Safe Democrat Safe Republican
17
72
83
[0.03]
[0.16]
[0.22]
(0.09)
(0.09)
(0.09)
8
75
42
[0.02]
[0.16]
[0.09]
(0.12)
(0.27)
(0.15)
5
51
80
[0.02]
[0.11]
[0.20]
(0.05)
(0.12)
(0.06)
30
198
205
Total
172
125
136
433
[1.00]
(0.11)
Notes: (1) Each cell in the 3 × 3 represents “coalition”. A cell contains (i) # districts in the coalition, (ii) proportion of
manufacturing workforce, in brackets, and (iii) proportion of export industry (COMPUTER) output, in parentheses.
Table 4 shows how districts were distributed across the nine political blocs after the 2000
elections. The numbers in square brackets indicate the proportion of the nation’s manufacturing workforce in each bloc. The bottom row indicates there were 205 strongly Republican
districts in 2000, 198 strongly Democrat districts, and just 30 competitive districts. We use
this case with electoral-based coalitions to analyze the determination of the level of protec34
tion granted by both total ad-valorem tariffs and NTMs.38 Institutionally, the authority for
enacting NTMs is distinct from tariffs. It emerges from multiple statutes; further, granting
protection through NTMs faces fewer constraints from international commitments and is
more unilateral.
Table 5: Kr Weight Shares (from 2SLS estimates): Small Country model
Dependent Variable: Applied Tariffs + NTMs, 2002
State-wide Vote in
Presidential Election
Competitive
Safe Democrat
Safe Republican
Total Kr share
Districts in House elections
Competitive Safe Democrat Safe Republican
0
0
0.104
[0]
[0]
[1.560]
0
0.093
0
[0]
[2.100]
[0]
0
0.047
0.073
[0]
[1.576]
[1.212]
0
0.140
0.177
Total
0.104
0.093
0.120
0.317
Notes: (1) N = 8210. (2) Each cell (coalition r) reports Kr -share of total welfare weights and (in square brackets) individual
ΓK
r
ΓL
r
ratio these shares imply. (3) See Notes to Table 2 for computation details.
In the small country setting, the pattern of (estimated) weights reported in Table 5
suggests an interpretation of the trade policymaking process in the 107th Congress in line
with the model.39 Suppose the agenda setter is Representative Cliff Stearns, Chairman
of the Commerce, Trade, and Consumer Protection Subcommittee of the powerful Ways
and Means Committee in the 107th Congress. Stearns represents the 6th CD in Florida,
a Safe Republican district in the most competitive State for the Presidency in the 2000
election. Further, suppose Stearns is to form a legislative majority in support of the status
quo trade policy, which needs to satisfy the protectionist interests of the majority party’s
median representative, and yet be mindful of the externalities imposed on voters in other
districts. To form a winning coalition the agenda setter needs the support of a legislative
majority drawn from the regional groupings used in our estimation. We can observe that
a proposal formed as in equation (8), combining the agenda setter’s status quo level of
protection (tariffs plus NTMs) satisfies special interests in four regions: Safe Republican
States + Safe Republican District (80); Safe Democratic State + Safe Democratic District
(75); Safe Republican State + Safe Democratic District (51) and Stearns’ own group of
Competitive State + Safe Republican District (83). For these groupings of CDs, the relative
M
weights
ΓK
r
ΓL
r
are estimated to be greater than one (square brackets in Table 5). Our estimates
38
(Expression 26) may be used to estimate the political determinants of non-tariff measures (NTMs)
measured as ad-valorem equivalents (AVEs). AVE of core NTMs is defined as the uniform tariff that would
have the same effect on imports as the NTMs. These are measured by Kee et al. (2009). Here, AVE of Core
NTMs is added to ad-valorem tariffs to measure overall protection.
39
The 2SLS estimates for estimating the weights are reported in Table A.1.4 in Appendix A.1.
35
suggest that such a proposal garners enough support of a super-majority in Congress (289
districts), making it presidential veto-proof.
M
Figure 3 depicts the geographic distribution of the estimated relative weights
ΓK
r
ΓL
r
. The
estimates indicate that tariffs and NTMs observed in the data are a winning proposal, and
therefore likely to endure even with manufacturing powerhouses like China getting preferential access. The politically acceptable protection at the national level for any district-good
is lower than the district’s preference.
Figure 3: Estimated
ΓK
r
ΓL
r
Weights – Small Country Case
The large country setting in Table 6 supports an interpretation of legislative bargaining
over trade policy where tariffs and NTMs at home are enacted in the shadow of potential retaliation abroad, and policymakers need to internalize terms of trade resulting from changes
in relative world prices. The estimated weights suggest that the same agenda setter, Trade
Sub-committee Chair representing the coalition of 83 Safe Republican CDs in battleground
states, can propose a vector of tariffs and NTMs that would muster the support of representatives from the 80 Safe Republican CDs in Safe Republican states. The vote of the
additional 55 representatives that would result in a legislative majority could be drawn from
CDs with a large presence of specific factor owners in the export industry, such as those that
are safely controlled by Democratic Congress members in states where the Democrat ticket
carried in the 2000 presidential election. Note that accounting for the reciprocal determination of protection and terms of trade effect, the weights on specific factors in the exporting
36
industry are estimated to be 16.6% of the total welfare weights as shown in Table 6.
Table 6: KrM and K X Weight Shares (from 2SLS estimates): Large Country Model
Dependent Variable: Applied Tariffs + NTMs, 2002
State-Wide Vote in
Presidential Election
Competitive
Safe Democrat
Safe Republican
Total KrM share
Total K X share
Districts in House Elections
Competitive Safe Democrat Safe Republican
0
0
0.081
[0]
[0]
[1.537]
0
0
0
[0]
[0]
[0]
0
0
0.113
[0]
[0]
[2.252]
0
0
0.194
Total
0.081
0
0.113
0.194
0.166
[2.906]
Notes: (1) N = 7675 (export sector NAICS-3=334 (COMP) dropped). (2) Cells above the Total K X share row (coalition
M
r) report (i) share of welfare weights placed on import-competing interests KrM , and (ii) individual
(3) The Total
KX
ΓK
r
ΓL
r
ratio in brackets.
share row reports the aggregate share of welfare weights on export sector interests and (in brackets) the
X
individual
ΓK
ΓL
ratio. (4) See Notes to Table 3 for computation details.
The pattern of protection through tariffs and NTMs in the data would, thus, result in
a winning proposal for a majority in Congress. The relative weight on a specific factor
M
owner in import-competing goods to a mobile factor owner
ΓK
r
ΓL
r
is 1.54 times larger in Safe
Republican Districts located in Competitive Presidential states, and 2.25 times larger in
Safe Republican Districts located in Safe Republican states; the geographic distribution of
relative weights is presented in Figure 4.
The winning coalition, however, is biased toward export interests, in this case, producers
of computers, whose distribution across political coalitions is presented in Figure 5. The
estimated weight on the welfare of a specific factor owner in the exporting sector (nationally)
KX
is estimated to be almost three times that of a mobile factor owner ( ΓΓL = 2.91). The results
show U.S. exporters to be highly effective in countervailing the demand for protection by
domestic interests in import-competing industries. They do so because of the threat of
retaliation, which is internalized by trade policymaking coalitions. It is also an explanation
for why U.S. trade protection is low on average and concentrated in a few industries, facts
that have eluded political economy models of trade policy.
6
Sensitivity Analysis
Estimates from equation (26) in the large country model (Tables 3 and 6) have assumed
U S and RoW have equal bargaining strength, that is, µ = 1. Here, we investigate the
sensitivity of K X -share, the welfare weight shares of specific factors employed in exports,
37
Figure 4: Estimated
ΓK
r
ΓL
r
Weights – Large Country Case
Figure 5: Output Share Computers (NAICS 334) by Political Coalitions
to a range of µ values.40 A smaller µ implies lower bargaining
strength
h RoW
i for the U.S. Recall
U
S
U
S
dΩ
/dτ
j
from the equilibrium condition (21), given by dΩdτj − dΩ
= 0, incorporates the
dτ ∗
dΩRoW /dτ ∗
g
g
terms of the “agreement.” Suppose, as mentioned in the text, that the agreement allows
RoW to use a retaliatory tariff in response to a unilateral U S tariff increase on imports of
In equation (26), since µ is not separately identified from the price ratio (pj /pj )/(p∗g /pg ) in equation
(27), the thought experiment is to explore sensitivity to µ conditional on (pj /pj )/(p∗g /pg ) = 1.
40
38
j, to keep RoW ’s welfare at the status quo. Then, this retaliatory tariff increase is given
by
dτg∗
dτj
dΩRoW /dτ
= − dΩRoW /dτ j∗ . The magnitude of the retaliation
g
dτg∗
dτj
characterizes U.S. “bargaining
strength,” µ, which is assumed to be constant across goods j.41
Table 7: Sensitivity Analysis of Large Country Results
Bargaining strength
µ
0.33
0.50
0.75
1.00
1.25
1.50
3.00
Geography-based coalitions
X
K X -share
ΓK /ΓL
0.436
11.66
0.318
6.62
0.242
4.40
0.204
3.48
0.181
2.99
0.165
2.67
0.127
1.95
Politics-based Coalitions
X
K X -share
ΓK /ΓL
0.324
7.56
0.214
4.87
0.192
3.51
0.166
2.91
0.150
2.57
0.140
2.35
0.113
1.84
Notes: Results for µ = 1 correspond to estimates from Table 3 and Table 6..
In Table 7, small values of µ imply lower U.S. bargaining strength. These results indicate
that when U.S. bargaining strength is low, the welfare weight on export interests is high.
Export interests and RoW bargaining strength work as complements to discourage U.S.
tariffs. When U.S. bargaining strength is high, the ability of the U.S. to increase welfare
by imposing optimal tariffs diminishes the role of U.S. export interests. Strikingly, even
when U.S. bargaining power is high (µj = 3), the share of the total welfare weight placed on
export interests remains significant, equal to 0.127 in the case of geography-based coalitions
of districts and 0.113 in the case with politics-based coalitions. Quantifying welfare weights
on export interests to counterfactual µj ’s is informative about the role of export interests:
If it is believed that U.S. has lower bargaining strength, export interests have even greater
influence in shaping trade policy.
7
Conclusion
This paper integrates congressional districts into a political economy model of trade. This
is necessary because in the U.S., and in many democracies, trade policymaking is a highly
institutionalized process where elected legislative bodies play a central role. The institutional
process regulating how trade policy is made in the U.S. relies on delegating “fast track”
authority to the Executive branch to negotiate a bilateral or multilateral agreement. Under
“fast track” the trade policy proposal negotiated by the president is subject to an up or down
vote by Congress, without amendments, granting the majority party in Congress agendasetting power over trade policy.
41
Sensitivity analyses for different µj are also possible.
39
Closely related to our model is the protection-for-sale framework of Grossman and Helpman (1994). However, the emphasis of our approach differs: while GH models the demand for
protection by special interests, our setup builds on a political-geography structure to explain
the supply of protection. We are, thus, able to unpack the parameter “a” in the GH model,
the rate at which the government trades welfare for contribution dollars, to account for the
relative influence of local interests in the formation of trade policy. Both approaches feature
special interests, but our present work incorporates congressional districts and legislative
bargaining, the main actors and institutions participating in the legislative processes. The
relative influence of districts is ultimately reflected in the weights received by local economic
actors and interests in the formation of trade policy.
The first step in our framework is to characterize the tariff vectors that each congressional
district would choose if they were to set the national tariff on their own. Estimating the
structural parameters from the model allows us to retrieve the otherwise unobservable local
demand for protection at the industry and congressional district levels. Our results provide
a contrast between the “independent” demand for protection by districts and the protection
delivered after district preferences are aggregated into national trade policy. These findings
are key to understanding the backlash against globalization.
Next, we model the process of preference aggregation as a legislative bargaining protocol,
where an agenda setter proposes a tariff vector that would muster a majority in Congress.
This bargaining process produces welfare weights that are a weighted average of the preferences of the agenda setter and the legislative majority, reflecting the geographic distribution
of economic activity and the institutionally defined process of preference aggregation in the
legislature. Using district-level manufacturing data and national imports and tariff data for
2002, we estimate the welfare weights of specific and mobile factors implied by the model.
We consider two stylized legislative “coalitions,” one based on geography and the other on
political alignments at the state and district levels. They yield substantively similar results:
specific factor owners in import-competing activities located in districts that can deliver a
majority in Congress receive positive welfare weights in the determination of national tariffs.
The large body of research on the political economy of protectionism that the paper
addresses has largely ignored the role of exporters. A key contribution of the paper is to
account for the influence of specific factor owners in exporting sectors. For this, the model
is extended to account for terms of trade effects (the large country case). Using predictions
from the extended model, we estimate a new set of welfare weights separately for specificfactor owners employed in exporting industries and import-competing industries and find
40
that specific-factor owners in exporting sectors receive welfare weights on par with factor
owners in import-competing industries. Further, once we account for exporters, only specific
factor owners located in safe Republican districts in battleground states or in states that
voted Republican in the 2000 presidential elections receive positive weights. The influence
of exporter interests reflects how the political process in the U.S. has internalized market
access concerns in the formation of the country’s trade policy. These are important and
novel results that add significantly to the literature.
By formally integrating districts –whose representatives serve their local economies by
bargaining in the legislature for the trade policies preferred by their constituents– into a specific factors model of trade, our paper builds a bridge between two influential and important
bodies of literature that had remained distant from each other. The model and estimations
provide theoretically motivated and empirically grounded micro-foundations for the low tariffs in the U.S. despite the growing public backlash against globalization in the face of the
surge of Chinese manufacturing imports starting in the late 1990s and culminating in the
“China shock."
Finally, the framework naturally extends in several relevant directions. While labor market effects are abstracted in our model, the paper offers a framework for integrating local
labor market effects into a political economy model of trade. Second, intermediate goods
(see e.g., Gawande et al. (2012)) may be easily incorporated into the model. Accounting for
intermediate goods can result in a more accurate measurement of district tariff preferences
and national tariffs, specifically for district-goods whose output is used intensively in downstream district-goods. Third, the model may be extended to examine the role of lobbies in
determining trade protection.42 The analysis would need to allow for lobbies to organize not
just at the sectoral level, as in previous studies, but regionally or nationally. Within such
a framework, lobbies would emerge endogenously (Mitra, 1999) and target their activities
to influence either the local or the national decision-making process. Their decisions would
depend, among other things, on the relative ability of lobbying efforts to steer policy in their
favor. We hope the paper paves the way for future research in this rich and important area.
42
Online Technical Appendix B (section 1.3) develops an extension with lobbying à la Grossman and
Helpman.
41
References
Abadie, A., Athey, S., Imbens, G. W. and Wooldridge, J. M. (2023), ‘When should you
adjust standard errors for clustering?’, Quarterly Journal of Economics 138(1), 1–35.
Autor, D. H. and Dorn, D. (2013), ‘The growth of low-skill service jobs and the polarization
of the US labor market’, American Economic Review 103(5), 1553–1597.
Autor, D. H., Dorn, D. and Hanson, G. H. (2013), ‘The China syndrome: Local labor market
effects of import competition in the US’, American Economic Review 103(6), 2121–68.
Autor, D. H., Dorn, D., Hanson, G. H. and Song, J. (2014), ‘Trade adjustment: Worker-level
evidence’, Quarterly Journal of Economics 129(4), 1799––1860.
Bagwell, K. and Staiger, R. W. (1999), ‘An economic theory of GATT’, American Economic
Review 89(1), 215–248.
Baldwin, R. E. (1985), The Political Economy of U.S. Import Policy, MIT Press.
Baron, D. P. and Ferejohn, J. A. (1989), ‘Bargaining in legislatures’, American Political
Science Review pp. 1181–1206.
Bartik, T. J. (1991), Who Benefits from State and Local Economic Development Policies?,
WE Upjohn Institute for Employment Research.
Bhagwati, J. N. and Feenstra, R. (1982), Tariff seeking and the efficient tariff, in J. Bhagwati,
ed., ‘Import competition and response’, Univ. of Chicago Press, pp. 245–262.
Celik, L., Karabay, B. and McLaren, J. (2013), ‘Trade policy-making in a model of legislative
bargaining’, Journal of International Economics 91(2), 179–190.
Conconi, P., Facchini, G. and Zanardi, M. (2014), ‘Policymakers’ horizon and trade reforms:
The protectionist effect of elections’, Journal of International Economics 94(1), 102–118.
Deardorff, A. V. and Stern, R. M. (1983), ‘Economic effects of the Tokyo round’, Southern
Economic Journal 49(3), 605–624.
Destler, I. M. (2005), American Trade Politics, Peterson Institute for International Economics.
Dutt, P. and Mitra, D. (2002), ‘Endogenous trade policy through majority voting: an empirical investigation’, Journal of International Economics 58(1), 107–133.
Eraslan, H. and Evdokimov, K. S. (2019), ‘Legislative and multilateral bargaining’, Annual
Review of Economics 11, 443–472.
Findlay, R. and Wellisz, S. (1982), Endogenous tariffs, the political economy of trade restrictions, and welfare, in ‘Import Competition and Response’, Univ. of Chicago Press,
pp. 223–244.
42
Finger, J. M., Hall, H. K. and Nelson, D. R. (1988), ‘The political economy of administered
protection’, American Economic Review 72(3), 452–466.
Gawande, B. K., Krishna, P. and Olarreaga, M. (2012), ‘Lobbying competition over trade
policy’, International Economic Review 53(1), 115–132.
Gawande, K. and Bandyopadhyay, U. (2000), ‘Is protection for sale? Evidence on the
Grossman–Helpman theory of endogenous protection’, Review of Economics and Statistics
82(1), 139–152.
Goldberg, P. K. and Maggi, G. (1999), ‘Protection for sale: An empirical investigation’,
American Economic Review 89(5), 1135–1155.
Goldsmith-Pinkham, P., Sorkin, I. and Swift, H. (2020), ‘Bartik instruments: What, when,
why, and how’, American Economic Review 110(8), 2586–2624.
Grossman, G. M. and Helpman, E. (1994), ‘Protection for sale’, American Economic Review
84(4), 833–850.
Grossman, G. M. and Helpman, E. (1995), ‘Trade wars and trade talks’, Journal of Political
Economy 103(4), 675–708.
Grossman, G. M. and Helpman, E. (1996), ‘Electoral competition and special interest politics’, Review of Economic Studies 63(2), 265–286.
Helpman, E. (1997), Politics and trade policy, in ‘Advances in Economics and Econometrics:
Theory and Applications’, Vol. II, Cambridge Univ. Press, pp. 19–45.
Hillman, A. L. (1982), ‘Declining industries and political–support protectionist motives’,
American Economic Review 72(5), 1180–1187.
Hillman, A. L. and Ursprung, H. W. (1988), ‘Domestic politics, foreign interests, and international trade policy’, American Economic Review 78(4), 729–745.
Irwin, D. A. (2017), Clashing over Commerce: A History of US Trade Policy, Univ. of
Chicago Press.
Irwin, D. A. and Kroszner, R. S. (1999), ‘Interests, institutions, and ideology in securing
policy change: The Republican conversion to trade liberalization after Smoot-Hawley’,
The Journal of Law & Economics 42(2), 643–674.
Kee, H. L., Nicita, A. and Olarreaga, M. (2008), ‘Import demand elasticities and trade
distortions’, Review of Economics and Statistics 90(4), 666–682.
Kee, H. L., Nicita, A. and Olarreaga, M. (2009), ‘Estimating trade restrictiveness indices’,
Economic Journal 90(119), 172–199.
Krugman, P. R. (1981), ‘Intraindustry specialization and the gains from trade’, Journal of
Political Economy 89, 959–973.
43
Magee, S. P., Brock, W. A. and Young, L. (1989), Black hole tariffs and endogenous policy
theory: Political economy in general equilibrium, Cambridge Univ. Press.
Marvel, H. P. and Ray, E. (1983), ‘The Kennedy round: Evidence on the regulation of
international trade in the US’, American Economic Review 73(1), 190–197.
Mayer, W. (1984), ‘Endogenous tariff formation’, American Economic Review 74(5), 970–
985.
Mitra, D. (1999), ‘Endogenous lobby formation and endogenous protection: A long-run
model of trade policy determination’, American Economic Review 89(5), 1116–1134.
Moretti, E. (2011), Local labor markets, in ‘Handbook of labor economics’, Vol. 4, Elsevier,
pp. 1237–1313.
Nicita, A., Olarreaga, M. and Silva, P. (2018), ‘Cooperation in WTO’s tariff waters?’, Journal
of Political Economy 126(3), 1302–1338.
Ray, E. J. (1981), ‘The determinants of tariff and nontariff trade restrictions in the US’,
Journal of Political Economy 89(1), 105–121.
Riker, W. H. (1962), The theory of political coalitions, Yale Univ. Press.
Rodrik, D. (1995), Political economy of trade policy, in ‘Handbook of International Economics’, Vol. III, Elsevier Sc. Pub. Co., pp. 1457–1494.
Topel, R. H. (1986), ‘Local labor markets’, Journal of Political Economy 94(3, Part 2), S111–
S143.
Trefler, D. (1993), ‘Trade liberalization and the theory of endogenous protection: An econometric study of US import policy’, Journal of Political Economy 101(1), 138–60.
USITC DataWeb (2019). US International Trade Commission DataWeb. Using data retrieved from the U.S. Bureau of the Census, dataweb.usitc.gov (accessed 9/17/2019).
Whalley, J. (1985), Trade liberalization among major world trading areas, MIT Press.
44
@FedFRASER