JOBS, JOBS, JOBS: A “NEW” PERSPECTIVE
ON PROTECTIONISM
Arnaud Costinot
University of California, San Diego
Abstract
This paper analyzes the determinants of protectionism in a small open economy with search
frictions. In this environment, jobs generate rents whose access depends on the level of trade
protection. By raising the domestic price of a good, a government may attract more firms
in a particular industry. This raises the probability that workers will find jobs in this sector,
and in turn, will benefit from the associated rents. Though simple, this channel may help
explain a variety of stylized facts on the structure of trade protection and individual trade-policy
preferences. (JEL: F130, F160)
1. Introduction
One very robust finding of the empirical literature on trade protection is the posi-
tive impact of unemployment on the level of trade barriers. The same pattern can
be observed across industries, among countries, and over time; see, for example,
Trefler (1993), Mansfield and Busch (1995), and Bohara and Kaempfer (1991),
respectively. These findings are echoed by recent empirical studies of individual
trade-policy preferences emphasizing the prevalence of labor market concerns;
see, for example, Scheve and Slaughter (2004).
Motivated by the previous evidence, we develop a simple theory of endoge-
nous trade protection with search frictions and relate it to various stylized facts
on protectionism across countries, industries, and individuals. In particular, we
show that the introduction of search frictions may offer a strong rationaliza-
tion of the positive correlation between unemployment and trade protection.
In our model, any parameter which increases (respectively, decreases) unem-
ployment also increases (respectively, decreases) the equilibrium trade tax. The
The editor in charge of this paper was Roberto Perotti.
Acknowledgments: I thank Eli Berman, Carl Davidson, Gordon Hanson, Nir Jaimovich, Navin
Kartik, Miklos Koren, Giovanni Maggi, Marc Muendler, Gary Ramey, Jim Rauch, Frederic Robert-
Nicoud, Esteban Rossi-Hansberg, the Editor Roberto Perotti, and four anonymous referees, as well
as seminar participants at UCSD, UCSC, UBC, and MSU for helpful comments.
E-mail address: Costinot: [email protected]
Journal of the European Economic Association September 2009 7(5):1011–1041
© 2009 by the European Economic Association
1012 Journal of the European Economic Association
same logic may also help explain why trade barriers tend to be higher in all
low-productivity industries—in contrast to the Grossman and Helpman (1994)
predictions—and why both high- and low-skilled workers tend to be less pro-
tectionist in more developed countries—in contrast to the Heckscher–Ohlin
predictions.
We start with a small open economy with multiple sectors, each of them
subject to search frictions à la Pissarides (2000). There is a continuum of workers,
each endowed with one unit of sector-specific human capital,
1
and a continuum of
firms, each free to choose the sector in which they want to post a vacancy. Workers
and firms come together randomly. Once a worker and a firm are matched, wages
are determined by Nash bargaining. In equilibrium, jobs generate rents whose
magnitude—the intensive margin—may depend on the level of trade protection.
This is reminiscent of the impact of trade taxes on the price of sector-specific
factors in the Ricardo–Viner model. The distinct feature of our model, on which
the rest of our analysis focuses, is that trade protection may also affect the access
to those rents—the extensive margin. By raising the domestic price of a good,
a government may attract more firms in a particular industry. This raises the
probability that workers will find jobs in this sector, and in turn, will benefit from
the associated rents.
The first part of our paper investigates how the extensive margin of trade
protection may affect the structure of trade protection. We assume that govern-
ments aim to maximize aggregate social welfare, but restrict the set of available
policy instruments to specific trade taxes. We view this assumption as a natural
and tractable benchmark. We do not deny the existence of political–economy
motives in practice, but note that they are not necessary for our argument. In
our model, the chance of a worker to find a job in a given industry depends
on the total number of vacant firms and unemployed workers present in this
industry, which creates trading externalities. There is a priori no reason why
wages, determined by Nash bargaining, would internalize these externalities. As
a result, a government may increase social welfare by imposing a small trade tax
or subsidy.
2
Of course, we do not aim to suggest that trade protection should be used
to correct this distortion. Bhagwati’s (1971) classical argument applies to our
environment. Here, trade policies are at most a second-best policy and the opti-
mal policy intervention should involve a tax-cum-subsidy addressed directly to
offsetting the source of the distortion. In this paper, we adopt a purely positive
1. See, for example, Neal (1995) and Parent (2000) for evidence pointing to the importance of
sector-specific human capital in practice.
2. See Davidson, Martin, & Matusz (1994) for a general discussion of the possibility for Pareto
improvements in dynamic models of unemployment.
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1013
perspective. Conditional on trade taxes being the only policy instruments avail-
able,
3
we investigate how variations in the magnitude of search frictions affect
protectionist incentives across countries and industries.
Our main findings regarding the structure of trade protection can be summa-
rized as follows. In a cross-section of industries, parameters which are positively
correlated with unemployment—workers’ bargaining power, sector size, and
turnover rate—should also be positively correlated with trade taxes. The converse
is true for parameters which are negatively correlated with unemployment—
world price and workers’ productivity. These predictions accord well with various
empirical studies reviewed by Rodrik (1995). For example, our finding regard-
ing workers’ productivity is consistent with the observation that trade barriers
tend to be higher in labor-intensive, low-skill, low-wage industries. It may
also help explain why protection is higher in periods of recession and in poor
countries.
Intuitively, an increase in the probability of finding jobs creates more jobs if
the pool of unemployed workers is initially large. This explains why the marginal
benefits from raising trade taxes are higher in sectors with more unemployment,
and in turn, why their trade taxes are higher in equilibrium. Although the intu-
ition behind our predictions is simple, they stand in sharp contrast to those of
standard trade models. In the Grossman and Helpman (1994) model, which has
become the workhorse of the profession, the level of trade barriers for organized
sectors increases with the level of domestic output. Alternative political econ-
omy approaches based on the Ricardo–Viner model, for example, Findlay and
Wellisz (1982) and Hillman (1982), lead to the same prediction. By focusing on
the extensive rather than the intensive margin of trade protection, our theory is
able to generate the opposite result.
The second part of our paper analyzes how the extensive margin of trade pro-
tection may affect individual trade-policy preferences. To this end, we extend our
model by allowing workers to vary by skills. We assume that skills depend on the
level and specificity of workers’ human capital, that they are perfectly observable
by firms, and that firms may only search for one type of workers. Because high-
skilled workers generate larger amounts of output, a larger number of firms search
for them, which increases their chances of finding jobs. We then consider a hypo-
thetical episode of trade liberalization where trade taxes are uniformly decreased
across sectors. Whether or not an individual should favor this policy change
depends on the trade-off between the benefits from freer trade—higher consumer
surplus net of changes in trade revenues—and the associated costs—destruction
of existing jobs and difficulty of finding new jobs once unemployed.
3. Although this is admittedly an ad hoc assumption, this is not an unusual one. As Rodrik (1995,
p. 1476) already put it a decade ago: “A sufficiently general and convincing explanation for this
phenomenon [the use of trade policy over more efficient instruments] has yet to be formulated.”
Offering this explanation is beyond the scope of the present paper.
1014 Journal of the European Economic Association
Our model predicts that workers with less general human capital are more
likely to be protectionist, as in the Ricardo–Viner model, but more so in com-
parative disadvantage industries. In addition, our model predicts that if workers
mostly care about their current incomes, then less productive workers are more
likely to be protectionist. In this situation, the main determinant of workers’ trade-
policy preferences is the probability of losing their jobs. Hence, less productive
workers—who are more likely to become unemployed—also are more likely to
be protectionist. This prediction, in contrast to those of the Heckscher–Ohlin
model, may help explain why (i) low-skilled workers tend to be more protection-
ist than high-skilled workers, irrespective of their countries of origin; and why
(ii) workers in less developed countries tend to be more protectionist, irrespective
of their skill level (see, e.g., Beaulieu, Benarroch, and Gaisford 2001; O’Rourke
and Sinott 2001; Scheve and Slaughter 2004; Mayda and Rodrik 2005).
The rest of the paper is organized as follows. Section 2 discusses the rela-
tionship between our paper and the previous literature. Section 3 describes our
model. Section 4 and 5 analyze the structure of trade protection and individual
trade-policy preferences. Section 6 offers some concluding remarks. All proofs
can be found in Appendix A.
2. Relation to the Previous Literature
Our paper contributes to two branches of the trade policy literature.
The structure of trade protection. Although there is a large normative litera-
ture analyzing the impact of various market imperfections on the optimal trade
policy—from Bhagwati (1971) to Helpman and Krugman (1989)—the positive
literature has, for the most part, focused on the “political” incentives of gov-
ernments in a perfectly competitive environment; see Helpman (1998) for an
overview. The first contribution of our paper derives from a simple observation:
There is a priori no reason why the “economic” incentives emphasized by the
normative literature shall have no effect on the variations of trade policies across
countries and industries in practice. Our paper carefully analyzes how search
frictions may affect governments’ incentives and shows that, even in the absence
of political-economy considerations, they may help explain a variety of stylized
facts regarding the structure of trade protection.
Among previous political-economy papers, Bradford (2006) and Matschke
and Sherlund (2006) are most closely related to ours. Both papers introduce labor
market imperfections into standard political-economy models. Like us, Bradford
starts from a Pissarides (2000) model which he combines with a polity where
governments maximize votes. The theoretical model then motivates an empirical
analysis that aims to uncover the impact of unionization and turnover rates on
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1015
the structure of trade protection. Unlike our paper, Bradford does not offer clear
theoretical predictions relating the exogenous parameters of the Pissarides model
to the equilibrium trade taxes.
Matschke and Sherlund (2006) emphasize labor market considerations by
introducing trade-union lobbying in the Grossman and Helpman (1994) model.
Their model predicts that the level of trade protection should be higher if the
trade union lobbies, but capital owners do not; and conversely, that the level of
trade protection should be lower if capital owners lobby, but the trade union does
not. Compared to our paper, their main focus remains on political incentives. In
particular, the authors do not try to relate in a systematic manner the level of trade
protection to the magnitude of labor market imperfections.
Finally, our results on productivity and trade protection are related to recent
work by Baldwin and Robert-Nicoud (2007). Using a dynamic version of the
Grossman and Helpman (1994) model, the authors provide an intuitive explana-
tion for the “loser’s paradox.” In expanding industries, policy-created rents attract
new entry that erodes the rents. By contrast, sunk market-entry costs protect these
rents in declining industries. As a result, firms in the latter industries lobby harder,
which explains why a decrease in productivity leads to more protection. Though
our model shares the same focus on economic rents, it provides an alternative
explanation of the “loser’s paradox” based on labor market imperfections. In our
model, low productivity leads to more protection because it increases the unem-
ployment rate, which makes the number of jobs more responsive to changes in
trade taxes.
4
Individual trade-policy preferences. The second contribution of our paper is
to show that search frictions may also shed a new light on the determinants
of individual trade-policy preferences. The previous literature on this topic is
mainly empirical (see, e.g., Magee 1980; Rogowski 1987; Beaulieu, Benarroch,
and Gaisford 2001; O’Rourke and Sinott 2001; Scheve and Slaughter 2001, 2004;
Mayda and Rodrik 2005; Magee, Davidson, and Matusz 2005). The typical paper
compares the attitudes toward free trade of different groups of individuals: If
preferences tend to vary by industry, then the authors conclude in favor of the
Ricardo–Viner model; if they tend to vary by other individual characteristics,
capital versus labor or skilled versus unskilled, then the authors conclude in favor
of the 2 × 2 × 2 Heckscher–Ohlin model. Although this is definitely a valuable
exercise, the “Ricardo–Viner versus Heckscher–Ohlin” dichotomy does not speak
to one salient feature of the survey data: the prevalence of labor market concerns.
In order to address this issue, one needs a theoretical framework without full
employment. By introducing search frictions, we are able to rationalize these
4. Bagwell and Staiger (2003) offer an alternative theory of the countercyclical nature of trade
protection based on the role of self-enforcement in trade agreements.
1016 Journal of the European Economic Association
concerns and, more importantly, to offer a simple and intuitive explanation for
the relationship between human capital and protectionist attitudes observed in the
data.
From a theoretical standpoint, our analysis is closely related to Davidson,
Martin, and Matusz (1999) who consider an economy with search frictions and
two factors, capital and labor.
5
They demonstrate how the turnover rate of an
industry may affect preferences toward trade liberalization across factors of pro-
duction. In sectors where turnover is large, their model predicts that capital-owners
and workers should have opposite preferences, as in the Heckscher–Ohlin model;
whereas in sectors where turnover is low, they should have similar preferences,
as in the Ricardo–Viner model. Unlike our paper, the “Ricardo–Viner versus
Hecksche–Ohlin” dichotomy is still at the heart of their analysis. In particular,
they do not investigate the relationship between human capital and protectionism,
which is our main focus.
3. The Model
We consider a small open economy with i = 0 ,...,n sectors, each of them
subject to search frictions à la Pissarides (2000).
3.1. Workers
There is a mass 1 of workers. Each worker is endowed with 1 unit of sector-
specific human capital, which is the only factor of production.
6
We denote by l
i
the proportion of workers with human capital specific to sector i. Each worker is
in one of two states, employed or unemployed, and aims to maximize her expected
lifetime utility
E
+∞
t=0
δ
t
u(c
t
)
,
where δ is the common discount factor, c
t
= (c
t
0
,...,c
t
n
) is the vector of con-
sumptions at time t, and u(c
t
) = c
t
0
+
n
i=1
ϕ
i
(c
t
i
) is a quasilinear utility function.
We assume that the sub-utility functions ϕ
i
(·) satisfy standard regularity condi-
tions: ϕ
i
> 0 and ϕ

i
< 0. Good 0 is used as the numéraire good with world and
5. Our paper is also related, though less closely, to recent models analyzing the labor market
impact of exogenous episodes of trade liberalization in environments with costly mobility (see, e.g.,
Chaudhuri and McLaren 2007), or firm heterogeneity (see, e.g., Davis and Harrigan 2007; Egger
and Kreickemeier 2006; Helpman and Itskhoki 2007; Janiak 2006).
6. One could extend our model to include physical capital; see Pissarides (2000), Section 1.6. As
long as there are constant returns to scale and a perfect second-hand market for capital goods, this
extension would leave our results unchanged.
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1017
domestic price equal to one. We call p
∗
i
the exogenous world price of good i, and
p
i
its domestic price.
7
The demand for good i is denoted by d
i
(p
i
) ≡ (ϕ
i
)
−1
(p
i
).
In turn, the indirect utility of a worker is given by
E
+∞
t=0
δ
t
[x
t
+ s(p)]
,
where x
t
is the worker’s income at date t, p = (p
1
,...,p
n
) is the vector of
domestic prices, and s(p) =
n
i=1
ϕ
i
[d
i
(p
i
)]−
n
i=1
p
i
d
i
(p
i
) is the surplus
derived from the consumption of these goods. We assume that x
t
= w
i
+ τ + ω if
the worker is employed in sector i at date t , and x
t
= τ + ω if she is unemployed.
w
i
corresponds to the wages paid by firms in sector i; τ + ω corresponds to the
income that each worker, employed or not, derives from government transfers τ
and firms’ dividends ω.
3.2. Firms
There is a large mass of firms with access to the same constant return to scale
technology. Each firm can employ at most 1 worker
8
and is in one of three states:
inactive, unfilled vacancy, and filled job. In any period, a firm with a filled job in
sector i generates revenues equal to p
i
a
i
. We refer to a
i
as workers’ productivity
in sector i. A firm with an unfilled vacancy does not generate any revenues and
must pay a recruiting cost k per period.
9
An inactive firm obtains a pay-off of
zero. Each firm chooses in which industry to post a vacancy (if any) in order to
maximize its expected discounted profits
E
+∞
t=0
δ
t
(π
t
− kn
t
)
,
where π
t
are the firm’s net revenues at date t and n
t
∈{0, 1} is the number of its
unfilled vacancies. By definition, π
t
= p
i
a
i
− w
i
if the firm employs a worker
in sector i, and zero otherwise.
7. Because of quasilinear preferences, specific factors, and exogenous world prices, there are no
general equilibrium effects in our model. This guarantees the separability of the government’s max-
imization program. In our model, as in Grossman and Helpman (1994) and many others in the trade
policy literature, the equilibrium levels of trade protection will be independent across sectors; see
equation (14).
8. Under constant returns to scale, this assumption is without any loss of generality.
9. It is worth emphasizing that k does not vary by industry. In particular, recruiting costs are not
proportional to revenues, p
i
a
i
. This is an important feature of the model. If recruiting costs were
always proportional to p
i
a
i
, then trade policy and productivity would have no effect on sectoral
unemployment; see equation (8).
1018 Journal of the European Economic Association
3.3. Labor Market
Firms and workers come together randomly. At the beginning of each period, the
number of matches taking place is given by
m(l
i
v
i
,l
i
u
i
) = min{l
i
v
i
,l
i
u
i
},
where v
i
and u
i
are the vacancy and unemployment rates in sector i, respectively.
Throughout this paper, we assume that v
i
<u
i
for all i.
10
Hence, firms with
unfilled vacancies find workers with probability one, and unemployed workers
“wait at the gate” and find jobs with probability θ
i
= v
i
/u
i
. We further discuss
this assumption and its implications in the next section. When a firm and a worker
are matched, wages are determined by Nash bargaining
w
i
= arg max(W
i
− U
i
)
β
i
(J
i
− V
i
)
1−β
i
,
where U
i
and W
i
are the expected lifetime utility of, respectively, an unemployed
and an employed worker in sector i; V
i
and J
i
are the expected discounted profits
of, respectively, a firm with an unfilled vacancy and a filled job; and β
i
∈ (0, 1)
is workers’ bargaining power in sector i. Finally, we assume that existing jobs
are randomly destroyed following a Poisson process. At the beginning of each
period, workers may move from employment to unemployment with probability
λ
i
to which we refer as the turnover rate in sector i.
11
3.4. Steady-State Equilibrium
We focus on the steady-state equilibrium of this economy.
12
For each industry,
this equilibrium includes the four value functions (U
i
,W
i
,V
i
,J
i
), the wage w
i
, the
unemployment rate u
i
, and the vacancy rate v
i
. The expected lifetime utilities of
unemployed and employed workers satisfy the following Bellman equations:
U
i
= τ + ω + s(p) + δ[θ
i
W
i
+ (1 − θ
i
)U
i
], (1)
W
i
= w
i
+ τ + ω + s(p) + δ[λ
i
U
i
+ (1 − λ
i
)W
i
]. (2)
10. We provide sufficient conditions such that this inequality is satisfied in Section 4.
11. Because job creation and job destruction occur simultaneously, unemployment rates before and
after matching takes place are equal in a steady-state equilibrium.
12. The obvious benefit of this approach is its tractability; its cost is that it does not allow us to
disentangle the short-term from the long-term effects of trade liberalization on unemployment (see,
e.g., Trefler 2004). Because we are mostly interested in cross-industry and cross-country evidence,
we believe that the benefit outweighs the cost. Focusing on the steady state equilibrium is admittedly
more problematic when we discuss evidence related to trade protection over time (see Bohara and
Kaempfer 1991).
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1019
In each period, unemployed workers receive utility τ + ω + s(p) and become
employed with probability θ
i
, whereas employed workers receive utility w
i
+τ +
ω + s(p) and become unemployed with probability λ
i
. Similarly, the expected
discounted profits of firms with unfilled vacancies and filled jobs are given by
V
i
=−k + δJ
i
, (3)
J
i
= π
i
+ δ[λ
i
V
i
+ (1 − λ
i
)J
i
]. (4)
Firms with vacancies pay recruiting costs equal to k and find workers with proba-
bility one. Meanwhile, firms with filled jobs receive profits π
i
and have to search
for new workers with probability λ
i
next period. Because of Nash bargaining, we
have
W
i
− U
i
= β
i
i
, (5)
where
i
= W
i
+ J
i
− U
i
− V
i
is the total surplus generated by a job. Free entry
of firms implies
V
i
= 0. (6)
Finally, the sectoral unemployment rate in the steady-state satisfies
u
i
=
λ
i
λ
i
+ θ
i
. (7)
We have a system of seven equations with seven unknowns. We can directly solve
for the equilibrium values of
i
and θ
i
. This leads to
θ
i
=
p
i
a
i
(1 − β
i
) − k
˜
λ
i
kβ
i
, (8)
i
=
k
δ(1 − β
i
)
, (9)
where
˜
λ
i
= 1/δ − 1 + λ
i
. Equations (8) and (9) completely characterize the
steady-state equilibrium; U
i
, W
i
, V
i
, J
i
, w
i
, u
i
, and v
i
can be computed by simple
substitutions.
From equation (8), we see that the domestic price of good i affects the
tightness of the labor market in industry i.Asp
i
goes up, more firms enter,
13
which raises the probability θ
i
that workers find jobs in sector i, and in turn,
increases total employment in that industry.
14
This is what we call the exten-
13. We refer to “firm entry” as the source of new jobs, but it should be clear that we do not
necessarily mean the creation of new legal entities or plants in practice. “Firm entry” in the model
refers to new vacancies being posted; whether or not these vacancies are actually posted by new or
existing firms is irrelevant for our purposes.
14. The formal mechanism through which p
i
increases θ
i
is slightly more subtle. Because of free
entry, the value of a vacant firm must be zero. Because firms find workers with probability one, the
value of a firm with a filled job is in turn determined by recruiting costs alone. Hence, any increase
in p
i
must be offset by an equal increase in w
i
, which can only be consistent with Nash bargaining
if θ
i
—and hence workers’ outside option—goes up.
1020 Journal of the European Economic Association
sive margin of trade protection. Equation (8) also implies that θ
i
increases with
workers’ productivity, a
i
, and decreases with workers’ bargaining power, β
i
,
and the turnover rate, λ
i
. Although the exact functional form clearly depends
on our particular matching function, it is worth emphasizing that these quali-
tative insights do not. With any other matching function with constant returns
to scale, the monotonicity of θ
i
with respect to p
i
, a
i
, β
i
, and λ
i
would be the
same.
By contrast, our predictions on
i
are very specific to the Leontief matching
function. According to equation (9), the domestic price of good i has no effect on
the surplus generated by a job in sector i. With any other matching function with
constant returns to scale, this would not be true. Generically, trade protection
raises the magnitude of the rents of the factors employed in a given industry.
Assuming that workers wait at the gate shuts down the intensive margin of trade
protection.
The Leontief matching function is admittedly a strong assumption. We view it
as a useful expositional device that allows us to focus on a channel largely ignored
by the previous literature: the extensive margin of trade protection. Because of our
Leontief matching function, trade policy can affect the number of jobs in a given
industry, but it cannot affect the rents associated with these jobs, as in a standard
Ricardo–Viner model. This stark feature of our model admittedly narrows the
scope of our analysis, but leads to a clear and intuitive picture of what we believe
is a new and robust determinant of trade protection in an open economy with
search frictions: the ability to create new jobs (or save old ones). Of course, the
Leontief matching function presents another advantage. Unlike general matching
functions, it provides closed form solutions for the steady-state equilibrium, which
greatly improves the tractability of the model.
4. The Structure of Trade Protection
The previous section describes the equilibrium of the economy, taking domestic
prices as given. We now analyze how the government’s trade taxes endogenously
determine these prices.
4.1. The Government’s Maximization Program
We restrict the set of policy instruments available to the government to specific
trade taxes: t
i
= p
i
− p
∗
i
for i = 1,...,n. If good i is imported, t
i
represents
a specific import tariff; if good i is exported, it represents an export subsidy.
Throughout this paper, we assume that the government may only choose t
i
in
[t
,
¯
t] such that
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1021
max
i=0,...,n
k
˜
λ
i
a
i
(1 − β
i
)
− p
∗
i
≤ t < 0 <
¯
t ≤ min
i=0,...,n
k(
˜
λ
i
+ β
i
)
a
i
(1 − β
i
)
− p
∗
i
.
This series of inequalities guarantees that θ
i
is between 0 and 1 in all sectors
i = 0,...,n. We further assume that the government chooses trade taxes in
order to maximize aggregate social welfare
G =
n
i=0
G
i
, (10)
where G
i
= l
i
u
i
U
i
+ l
i
(1 − u
i
)W
i
is the welfare associated with the workers of
sector i. Because all trade revenues are redistributed uniformly to workers, the
net lump-sum transfer to each worker is given by
τ =
n
i=1
t
i
m
i
(p
i
), (11)
where m
i
(p
i
) = d
i
(p
i
) − y
i
(p
i
) and y
i
(p
i
) = l
i
(1 − u
i
)a
i
are the net imports
and domestic output of good i. The lifetime income that each worker derives from
firms’ dividends is given by
ω
1 − δ
=
n
i=0
[l
i
u
i
V
i
+ (1 − u
i
)l
i
J
i
]. (12)
Using equations (1), (5), (6), (10), and (12), and the definition of
i
, we can
rearrange the government’s objective function as
G =
τ + s(p)
1 − δ
+
n
i=0
l
i
1 − δ
p
i
a
i
−
k
˜
λ
i
1 − β
i
+
n
i=0
l
i
i
(1 − u
i
). (13)
4.2. Equilibrium Policies
Let us consider a marginal increase in the trade tax of sector i = 1,...,n.By
differentiating equation (13) with respect to t
i
, we obtain
∂G
∂t
i
=
1
1 − δ

∂τ
∂t
i
+
∂s
∂t
i
+
l
i
a
i
1 − δ
− l
i
i
∂u
i
∂t
i
. (14)
The first two terms are fairly standard. ∂τ/∂t
i
= t
i
m
i
(p
i
) + m
i
(p
i
) and ∂s/∂t
i
=
−d
i
(p
i
) correspond to the marginal changes in trade revenues and consumer
surplus, respectively. It is easy to check that ∂τ/∂t
i
+ ∂s/∂t
i
= t
i
m
i
(p
i
) −
1022 Journal of the European Economic Association
y
i
(p
i
)<0. In other words, increasing trade taxes always reduces the sum of
trade revenues and consumer surplus. The second term, l
i
a
i
, captures the marginal
increase in total wages in sector i.
The most novel feature of the model is the third term, −l
i
i
(∂u
i
/∂t
i
).In
an economy without search frictions,
i
would be equal to zero, l
i
a
i
would
be equal to total output in sector i, and free trade would always be optimal:
∂G/∂t
i
= t
i
m
i
(p
i
)<0. In an economy with search frictions, however, imposing
trade taxes has one extra benefit: creating jobs in the targeted industry. By raising
the level of trade protection in sector i, a government may reduce unemployment,
∂u
i
/∂t
i
=−[λ
i
/(λ
i
+ θ
i
)
2
](∂θ
i
/∂t
i
)<0,
and in turn increase workers’ expected income. Lemma 1 formally describes the
determinants of trade policies in our model.
Lemma 1. Suppose that φ

i
≤ 0 and k(1/δ − 1)/a
i
(1 − β
i
)>
¯
t for all i =
1,...,n. Then there exists a unique vector of equilibrium policies (t
0
1
,...,t
0
n
)
such that, if t
0
i
is interior (t <t
0
i
<
¯
t), then it is a unique solution to
a
i
l
i
λ
i
λ
i
+ θ
i
+
a
i
l
i
λ
i
(λ
i
+ θ
i
)
2
β
i
1
δ
− 1 −
a
i
t
i
(1 − β
i
)
k

MB(t
i
)
=−t
i
d
i
p
∗
i
+ t
i

MC(t
i
)
. (15)
The two inequalities, φ

i
≤ 0 and k(1/δ − 1)/a
i
(1−β
i
)>
¯
t, are sufficient to
derive the strict concavity of G(·) with respect to t
i
. The first one implies that the
right-hand side—the marginal cost MC(t
i
) associated with distorting demand—
is increasing in t
i
. The second one implies that 1/δ −1−a
i
t
i
(1−β
i
)/k > 0,
15
and
so that the left-hand side—the marginal benefit MB(t
i
) associated with improving
labor market conditions—is decreasing in t
i
.
In the rest of this paper, we restrict ourselves to interior equilibria and assume
that the two previous inequalities hold in every industry. Hence, the equilibrium
policy in sector i can be described as in Figure 1. Note that our theory always
predicts import tariffs or export subsidies. Because d
i
< 0, equation (15) implies
t
o
i
> 0 for all i = 1,...,n. This result derives from the particular nature of
the labor market imperfections in our economy. Whereas unemployed workers
exert negative search externalities on other unemployed workers, vacant firms
do not exert any externality on other vacant firms (they always find workers
with probability one). Therefore, the unemployment rate is too high and a small
import tariff or export subsidy that raises the level of employment also raises
social welfare.
15. Economically speaking, this inequality guarantees that the positive rent effect, −l
i
i
(∂u
i
/∂t
i
),
outweighs the negative trade revenues effect, −t
i
y
i
(p
∗
i
+ t
i
)/(1 − δ).
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1023
Figure 1. Equilibrium policy.
This feature of our model shall not be interpreted as a normative case in
favor of trade protection. First, different matching functions may lead to different
predictions on the overall level of trade protection: t
o
i
> 0 is a direct consequence
of our Leontief matching function. Second, trade taxes in this environment are
at most a second-best policy. For example, output subsidies at the rate t
o
i
would
achieve the same level of employment in sector i without distorting consumers’
behavior (see Helpman and Krugman 1989, pp. 20–22). Our focus in this paper
is purely positive. Conditional on trade taxes being the only policy instruments
available, we simply want to ask when, according to our theory, governments
shall have “bigger” incentives to impose trade taxes; use this insight to derive
predictions on the cross-sectional variations of trade protection; and finally, relate
these predictions to available empirical evidence.
4.3. Comparative Statics
We now use equation (15) to analyze the impact of five exogenous parameters
of the model, z
i
∈{a
i
,β
i
,l
i
,λ
i
,p
∗
i
}, on the equilibrium policy t
o
i
. Proposition 1
presents the main findings of our paper on the structure of trade protection.
Proposition 1. Ceteris paribus, equilibrium trade taxes t
o
i
are higher if
(i) output per worker a
i
is low;
(ii) workers’ bargaining power β
i
is high;
(iii) sector size l
i
is high;
(iv) the world price p
∗
i
is low; or
(v) job turnover λ
i
is high.
1024 Journal of the European Economic Association
Proposition 1 crucially relies on one key feature of our model: the absence
of the intensive margin of trade protection. In the standard Ricardo–Viner model,
factors are fully employed and trade taxes can only affect their rents. In our model,
it is the contrary: Factors are not fully employed and trade taxes can only affect
the level of employment.
To understand the role of the extensive margin of trade protection, it is use-
ful to decompose the impact of our five exogenous parameters on the marginal
benefit of a trade tax, MB, into a direct effect and an indirect effect on sectoral
unemployment
dMB
dz
i
=
∂MB
∂z
i
+
∂MB
∂l
i
u
i
∂l
i
u
i
∂z
i
.
Let us first focus on the indirect effect. By equations (7) and (15), we have
∂MB
∂l
i
u
i
= a
i
+
2u
i
a
i
β
i
λ
i
1
δ
− 1 −
a
i
t
0
i
(
1 − β
i
)
k
> 0. (16)
Inequality (16) states that the marginal benefits from raising trade taxes tend to
be higher in sectors with more unemployment. The reason is twofold and does
not depend on our functional form assumptions. First, there is a mechanical tax-
base effect. Holding demand constant, a higher unemployment rate increases the
level of imports, which raises the marginal change in tariff revenues. Second,
there is a more subtle job creation effect. In a steady-state equilibrium, a given
increase in the probability of finding jobs, θ
i
, always has a bigger effect on the
level of employment—and so, on the number of workers benefiting from rents—
if the total number of unemployed workers is initially large. Intuitively, more
job creation, θ
i
l
i
u
i
, must be compensated by more job destruction, λ
i
l
i
(1 − u
i
),
which requires a higher level of employment.
16
According to inequality (16), factors raising sectoral unemployment,
∂l
i
u
i
/∂z
i
> 0, should tend to increase the marginal benefit of a trade tax,
(∂MB/∂l
i
u
i
)(∂l
i
u
i
/∂z
i
)>0, while the opposite should be true for factors low-
ering it. Using equations (7) and (8), we can compute the signs of ∂l
i
u
i
/∂z
i
for
z
i
∈{a
i
,β
i
,l
i
,λ
i
,p
∗
i
}. We find that sectoral unemployment increases with work-
ers’ bargaining power, sector size, and the turnover rate, whereas it decreases with
output per worker and the world price.
17
Because higher marginal benefits call
for higher trade taxes in equilibrium, the previous mechanism suggests a positive
16. Formally, u
i
is a convex function of θ
i
; see equation (7).
17. These are fairly intuitive predictions: an increase in a
i
or p
∗
i
raises profits, which leads to more
entry and a decrease in unemployment; an increase in β
i
raises wages, lowers profits, and increases
unemployment; an increase in λ
i
implies more job destruction, so unemployment must increase for
job creation to catch up; finally, an increase in l
i
mechanically increases sectoral unemployment. As
mentioned in section 3.4, these predictions are robust to changes in the matching function.
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1025
correlation between t
o
i
and β
i
, l
i
, and λ
i
; and a negative correlation between t
o
i
and a
i
and p
∗
i
. This is exactly what Proposition 1 predicts.
What about the direct effect, ∂MB/∂z
i
?Forz
i
∈{l
i
,p
∗
i
}, the situation is
simple, ∂MB/∂z
i
= 0; for z
i
∈{a
i
,β
i
,λ
i
}, unfortunately, it is more complex.
For example, we have ∂MB/∂λ
i
< 0, which goes in the opposite direction as the
indirect effect, ∂l
i
u
i
/∂λ
i
> 0.
18
For those factors, the fact that the indirect effect
always dominates is naturally influenced by our functional form assumptions.
Notwithstanding, our analysis demonstrates that the introduction of search fric-
tions à la Pissarides (2000) can provide a strong rationalization of the stylized facts
offered in the introduction. In a simple version of the Pissarides model—where
only the extensive margin of trade protection is active—any parameter which
increases (respectively, decreases) unemployment also increases (respectively,
decreases) the equilibrium trade tax. Hence, large low-skill industries which are
heavily unionized and face tough competition from abroad accumulate reasons
to receive more protection.
A few comments are in order. First, Proposition 1 shows that search frictions
can offer a strong rationalization of the positive correlation between trade pro-
tection and unemployment, not that they necessarily do. Although the extensive
margin of trade protection would remain active in more general environments,
the introduction of an intensive margin under different matching functions may
blur the sharpness of our predictions. To see this, note that an extra term,
l
i
(1 − u
i
)(∂
i
/∂t
i
), would appear in equation (14). Accordingly, the marginal
benefit from increasing rents would be higher in sectors where the number of
jobs, l
i
(1 − u
i
), is high. This effect, which is reminiscent of the impact of trade
taxes on the price of sector-specific factors in the Ricardo–Viner model, would
go against the main mechanism in our model.
Second, it is worth pointing out that Proposition 1 does not rely on the
existence of tariffs revenues.
19
Without tariffs revenues, equation (15) would
become
a
i
l
i
λ
i
(1/δ − 1)/(λ
i
+ θ
i
)
2
β
i
= d
i
p
∗
i
+ t
0
i
.
Compared to the previous case, the marginal cost of trade protection would be
equal to the decrease in consumer surplus, which would no longer be compensated
by a change in revenues. Similarly, the marginal benefit of trade protection would
no longer include the tax-base effect. Yet, the job creation effect—which is the
main focus of our analysis—would still be present. As a result, the amount of
trade protection would still be increasing in β
i
, l
i
, and λ
i
, and decreasing in a
i
and p
∗
i
.
18. In order to maintain the equality between job creation and job destruction after an increase in
the probability of finding jobs, the level of employment needs to rise less if the turnover rate is high.
This explains why the marginal benefit from increasing trade taxes decreases with the turnover rate.
19. I am grateful to an anonymous referee for bringing this to my attention.
1026 Journal of the European Economic Association
Finally, we want to acknowledge that one could imagine alternative theo-
ries leading to similar insights regarding the relationship between unemployment
and trade protection. Suppose, for example, that governments mostly care about
the “poor.” Then, one should observe more protection in industries with more
poor which, presumably, tend to have higher unemployment as well. However,
we believe that our approach presents one crucial advantage over such theo-
ries: It recognizes the endogeneity of the unemployment rate. As equation (15)
shows, most factors affecting the unemployment rate also have direct effects on
the government’s objective function. In principle, the latter effects may overturn
the positive relationship between unemployment and trade protection. To assess
whether or not this is the case, one needs an explicit model of labor market
imperfections which our paper provides.
4.4. A Raw Look at the Evidence
We conclude this section by discussing how the five predictions of Proposition
1 relate to available empirical evidence. Our goal is not to offer a formal test
of a stylized model, but rather to assess whether the channel emphasized in our
paper—the extensive margin of trade protection—appears to be consistent with
existing “stylized facts” on the structure of protection. Whenever possible, our
“stylized facts” are taken from the chapter by Rodrik (1995) in the Handbook of
International Economics.
20
Prediction (i) states that, ceteris paribus, equilibrium trade taxes are higher
if output per worker is low. This prediction accords well with a large body of
empirical work. In line with our theory, trade protection tends to be higher in
labor-intensive, low-skill, low-wage industry (see Caves 1976; Saunders 1980;
Anderson 1980; Ray 1981; Marvel and Ray 1983; Baldwin 1985; Anderson and
Baldwin 1987; Ray 1991; Finger and Harrison 1994). If we reinterpret the previ-
ous comparative statics exercise in terms of changes over time or across countries,
prediction (i) also is consistent with the fact that trade protection tends to be higher
in periods of recession (see Ray 1987; Hansen 1990; O’Halloran 1994) and in
poor countries.
21
20. While we believe that there is valuable information to be gained from such an exercise, it
presents some obvious limitations. As mentioned in Rodrik (1995), many studies are not directly
comparable: “they use different measures of protection, cover different countries and time periods,
and include different sets of righ-handside variables.” In particular, they may control for variables
that are not indicated by our model, the most problematic of all being unemployment. In our model,
the indirect impact of a
i
, β
i
, l
i
, λ
i
, and p
∗
i
on unemployment is key to derive Proposition 1.
21. Though we have no intention to delve into the empirical debate on country growth and openness
to trade (see, e.g., Rodriguez and Rodrik 1999), our result highlights the potential importance of
reverse causality when interpreting the evidence. In our model, when output per worker goes up, the
government’s incentives to be protectionist go down.
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1027
Predictions (ii) and (iii) are consistent with the empirical findings of Matschke
and Sherlund (2006) and Goldberg and Maggi (1999), respectively. After con-
trolling for the Grossman and Helpman (1994) determinants of trade protection,
these papers find that the unionization rates of industries as well as their size
remain positively correlated with the level of their trade barriers. Similarly, pre-
diction (iv) is consistent with the fact that trade protection tends to increase with
the level of import-penetration in a given industry (see Anderson 1980; Finger
and Harrison 1994).
To the best of our knowledge, Bradford (2006) is the only empirical study
investigating the relationship between the cross-sectoral variations in job turnover
and trade barriers. After controlling for unemployment, the author finds that
higher turnover rates lead to lower trade protection. Although this is cer-
tainly not direct evidence in favor of prediction (v), this does not contradict
it either. Since ∂MB/∂λ
i
< 0, our model indeed predicts that, holding unem-
ployment constant, higher turnover rates should decrease the equilibrium trade
taxes.
22
5. Individual Trade-Policy Preferences
In Sections 3 and 4, we described a small open economy with search frictions
and characterized the structure of trade protection in this environment. We now
investigate the impact of these frictions on individual trade-policy preferences.
To this end, we extend our analysis by allowing workers’ human capital to vary
in terms of both level and specificity.
5.1. Human Capital and Labor Market Outcomes
We index workers by j ∈[0, 1] and assume that workers are endowed with
h
j
units of human capital, out of which (1 − σ
j
)h
j
are general and σ
j
h
j
are
sector-specific.
23
The parameters h
j
> 0 and 1 ≥ σ
j
≥ 0 measure the level
and specificity of worker j’s human capital, respectively. Section 3 corresponds
to the case where h
j
= 1 and σ
j
= 1 for all j ∈[0, 1]. We denote by a
j
i
=
a
i
h
j
(1 − σ
j
i
) the output per period of worker j when matched with a firm in
22. For the same reason, the fact that the United Kingdom has a higher job turnover and is less
reluctant to trade than many continental European countries cannot be taken as evidence against
prediction (v). According to our model, higher turnover rates only lead to more protection if they
are associated with more unemployment. This is not the case for the United Kingdom.
23. We still ignore issues related to the existence of firm-specific human capital. To maintain the
general structure of our model unchanged, we abstract from match-specific productivity differences
that may lead to job rejections and from investments in firm-specific skills (see, e.g., Pissarides 2000,
Chapter 6; Wasmer 2006).
1028 Journal of the European Economic Association
sector i. By definition, σ
j
i
= 0 if worker j has human capital specific to sector i,
and σ
j
i
= σ
j
otherwise. With a slight abuse of notations, a
i
now represents the
productivity of human capital in sector i. In the spirit of Hall and Jones (1999),
one may interpret a
i
as a measure of physical capital per worker and the quality
of social infrastructure, which may vary across countries and industries. We refer
to h
j
(1 − σ
j
i
) as the skill level of worker j in sector i. Finally, we assume that
unemployed workers search for jobs in the sector that maximizes their expected
lifetime utility, that workers’ skill levels are perfectly observable, and that firms
can only search for one type of workers.
Under these assumptions, we can solve for the steady-state equilibrium
as we did in Section 3. Labor markets are segmented by skill levels. Free
entry guarantees that firms are indifferent between searching for high- or low-
skilled workers: V
j
i
= 0 for all j ∈[0, 1]. Irrespective of the mass of workers
per industry, which now is endogenous, the labor market equilibrium for each
type of workers is determined by equations (1) to (7). In turn, the total surplus
and the labor market tightness associated with each worker and industry are
given by
j
i
=
k
δ(1 − β
i
)
, (17)
θ
j
i
=
p
i
a
j
i
(1 − β
i
) − k
˜
λ
i
kβ
i
. (18)
Equation (17) implies that total surplus
j
i
is independent of worker j’s
skill level. As in Section 3, this feature of the equilibrium is an artifact
of our particular matching function. More importantly, equation (18) implies
that the tightness of the labor market θ
j
i
is increasing in the skill level
of worker j . Ceteris paribus, high-skilled workers generate higher surplus
when matched with a firm, which increases the number of firms search-
ing for them, and in turn, their probabilities of finding jobs.
24
This fea-
ture of our model captures in a stylized way the well-known fact that
unemployment rates are higher for less-educated workers (see, e.g., Mincer
1993).
Using equations (5), (17), (18), and (1), we can express the expected lifetime
utility of worker j when unemployed in sector i by
U
j
i
=
1
1 − δ
τ + ω + s(p) + p
i
a
j
i
−
k
˜
λ
i
1 − β
i
(19)
24. Again, it is worth emphasizing that this prediction relies on the fact that recruiting costs, k, are
not proportional to output per worker, a
j
i
.
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1029
and her expected lifetime utility when employed in sector i by
W
j
i
=
1
1 − δ
τ + ω + s(p) + p
i
a
j
i
−
k
˜
λ
i
1 − β
i
+
β
i
k
δ(1 − β
i
)
. (20)
Finally, we can compute the mass of workers per industry by solving max
0≤i≤n
U
j
i
for all j ∈[0, 1].
5.2. Why Are Some People (and Countries) more Protectionist than Others?
25
In order to answer this question, we consider a hypothetical episode of trade
liberalization, dt
1
= ··· = dt
n
= dt < 0. We then compare the expected
lifetime utility of a worker j employed in sector i in the steady states before and
after trade liberalization.
26
We denote by
W
j
i
(respectively
U
j
i
) the expected lifetime utility of worker j
when employed (respectively, unemployed) in sector i after trade liberalization. In
order to avoid a taxonomic exercise, we restrict our attention to situations where:
(i) workers never quit their jobs after trade liberalization,
W
j
i
> max
0≤i
≤n
U
j
i
=
U
j
for all i = 0,...,nand j ∈[0, 1]; and (ii) workers always work in sectors
where they have sector-specific human capital before trade liberalization.
27
The
change in the expected lifetime utility of a worker j employed in sector i is
given by
dW
j
i
=
du
j
i
1 − u
j
i
U
j
− W
j
i
+
1 −
du
j
i
1 − u
j
i
W
j
i
− W
j
i
, (21)
where
du
j
i
=−
a
j
i
(1 − β
i
)λ
i
dt
kβ
i
λ
i
+ θ
j
i
2
> 0
is the change in the unemployment rate induced by trade liberalization. If worker
j loses her job, which occurs with probability du
j
i
/(1 − u
j
i
), the change in her
expected lifetime utility is equal to
U
j
− W
j
i
. If she keeps her job, it is equal
25. The title of this section is borrowed from Mayda and Rodrik (2005).
26. Though we always refer to “trade liberalization,” it should be clear that our analysis equally
applies to foreign productivity gains, dp
∗
1
=··· =dp
∗
n
= dt.
27. Assumption (i) guarantees that changes in unemployment rates are the main determinants
of trade-policy preferences; it requires employment rents
j
i
= k/δ(1 − β
i
) to be large enough.
Assumption (ii) merely is a normalization of σ
j
i
.
1030 Journal of the European Economic Association
to
W
j
i
− W
j
i
instead. According to our theory, a worker j employed in sector i
should declare herself in favor of trade liberalization if and only if dW
j
i
≥ 0.
The next proposition describes the impact of human capital specificity on
individual trade-policy preferences.
Proposition 2. Ceteris paribus, workers are more likely to be protectionist if
the specificity of their human capital σ
j
is high.
The proof is straightforward. By definition, workers with more general human
capital lose less when switching sectors. This implies better outside options once
unemployed, which reduces their incentives to be protectionist. Though simple,
this idea may help explain the negative impact of age on attitudes toward free
trade (see, e.g., O’Rourke and Sinott 2001; Mayda and Rodrik 2005). Over time,
human capital becomes more specific. As a result, workers become less mobile
across sectors, and so more likely to oppose trade liberalization.
Note that the Ricardo–Viner model, absent of any search frictions, leads to
a similar prediction. Namely, the owners of the specific factors should be more
protectionist than the owners of the mobile factor. However, the insights of our
search model are finer. According to our theory, the specificity of human capital
only matters if the decrease in the trade tax is large enough to trigger a reallocation
of workers across sectors. This suggests that the impact of specificity on trade-
policy preferences should be stronger in industries where trade liberalization leads
to a larger decline in domestic prices.
If we reinterpret the specificity of workers’ human capital more generally in
terms of “mobility,” this prediction accords well with the results of Scheve and
Slaughter (2001). Using data from the 1992 National Election Studies survey, the
authors find a positive correlation between home ownership in counties with a
manufacturing mix concentrated in comparative-disadvantage industries and the
support for trade barriers. They interpret this result as evidence of the impact of
asset values, in addition to current factor incomes, on trade-policy preferences.
An alternative interpretation offered by our theory is that: (i) workers in these
counties are more likely to lose their jobs; and that: (ii) once unemployed, home
ownership increases the costs of moving to another sector.
Finally, we can use equations (7), (12), (17), (18), and (20) in order to
rearrange equation (21) as
(1 − δ)dW
j
i
= d[τ + s(p)]+dt
a
j
i
+
(1 − δ)λ
i
a
j
i
(1 − β
i
)
kβ
i
θ
j
i
λ
i
+ θ
j
i
W
j
i
−
U
j
.
The first term, d[τ + s(p)]=
n
i=1
[t
i
m
i
(p
i
) − y
i
(p
i
)]dt > 0, captures the
gains from trade liberalization: higher consumer surplus net of changes in trade
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1031
revenues. The second term captures the losses: difficulty of finding new jobs once
unemployed, a
j
i
dt; and destruction of existing jobs,
(1 − δ)λ
i
a
j
i
(1 − β
i
)
kβ
i
θ
j
i
λ
i
+ θ
j
i
W
j
i
−
U
j
dt.
Our next prediction on the determinants of individual trade-policy preferences
can be stated as follows.
Proposition 3. If δ is small enough, then workers are more likely to be
protectionist if their productivity a
j
i
is low.
When δ is small enough, workers mostly care about their current incomes.
Whether they have general or sector-specific human capital, the main determinant
of their trade-policy preferences is the probability of losing their jobs. As a result,
less-productive workers—who are more likely to become unemployed
28
—also
are more likely to be protectionist.
Proposition 3 directly implies the following.
Corollary 1. If workers mostly care about their current incomes, then the
prevalence of protectionism decreases with:
(i) countries’ level of development a
i
;
(ii) workers’ level of human capital h
j
.
These two predictions are in line with the recent empirical studies by
Beaulieu, Benarroch, & Gaisford (2001), O’Rourke and Sinott (2001), Scheve
and Slaughter (2004), and Mayda and Rodrik (2005). Using data from the 1995–
1997 World Values Survey and the 1995 International Social Survey Programme,
they find that (i) workers in less-developed countries tend to be more protectionist,
irrespective of their skill level; and that (ii) low-skilled workers tend to be more
protectionist than high-skilled workers, irrespective of their countries of origin
(though less so in less-developed countries). This can easily be seen in Table 1
which is constructed from the World Values Survey 1994–1999; see Appendix B
for details.
The second finding has been interpreted as evidence in favor of the
Heckscher–Ohlin model by O’Rourke and Sinott (2001), Scheve and Slaugh-
ter (2004), and Mayda and Rodrik (2005); and as evidence against it by Beaulieu,
Benarroch, and Gaisford (2001). The latter focus on the first part (low-skilled
28. This crucial feature of our model is consistent with the evidence that less-educated workers are
much more likely to be displaced in practice (see, e.g., Kletzer 1998).
1032 Journal of the European Economic Association
Table 1. Proportion of protectionist opinions.
Country
Education High income Rest of the world
Upper 50% 63%
Middle 63% 69%
Lower 70% 76%
Source: World Values Survey 1994–1999.
workers are more protectionist almost everywhere) and the former on the second
part (less so in less-developed countries) while arguing that the least developed
countries, for which low-skilled workers tend to be less protectionist, are not
in the sample. We have little to add to this debate. To us, the first finding is
the most problematic for the Heckscher–Ohlin model: Why would low-skilled
workers in a less-developed country—who win more, or at least lose less,
from free trade according to this theory—ever be more protectionist than their
counterparts in a more developed country? We believe that the introduction of
labor market imperfections may provide a simple and intuitive answer to this
question.
6. Concluding Remarks
This paper analyzes the determinants of protectionism in a small open econ-
omy with search frictions à la Pissarides (2000). By focusing on the extensive
margin of trade protection, our theory generates a rich set of predictions on the
structure of protection and individual trade-policy preferences. First, our model
predicts that in a cross-section of industries, parameters which are positively
correlated with unemployment—workers’ bargaining power, sector size, and
turnover rate—should also be positively correlated with trade taxes. The converse
is true for parameters which are negatively correlated with unemployment—
world price and workers’ productivity. Second, our model predicts that workers
with less general human capital are more likely to be protectionist, and more
so in comparative disadvantage industries. In addition, our model predicts that
if workers mostly care about their current incomes, then more productive work-
ers are less likely to be protectionist, irrespective of the countries and industries
where they are located. Though distinct from the predictions of standard trade
models, these findings appear to accord well with various empirical studies.
To us, this illustrates one key idea: The extensive margin of trade protec-
tion (whether or not workers keep their jobs and the associated rents) may
matter as much in practice as its intensive margin (by how much these rents
vary).
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1033
Appendix A: Proofs
A.1. Proof of Lemma 1
Using equations (7), (9), (8), and the definition of y
i
(·), equation (14) can be
rearranged as
(1 − δ)
∂G
∂t
i
= t
i
d
i
p
∗
i
+ t
i
+
a
i
l
i
λ
i
λ
i
+ θ
i
+
a
i
l
i
λ
i
(λ
i
+ θ
i
)
2
β
i
1
δ
− 1 −
a
i
t
i
(1 − β
i
)
k
.
First note that if t
i
≤ 0, then the upcoming equation (A.1) implies ∂G/∂t
i
> 0.
Because
¯
t>0, any equilibrium trade tax must be strictly positive. We now focus
on t
i
> 0. Consider the second derivative of G with respect to t
i
:
(1 − δ)
∂
2
G
∂t
2
i
= d
i
p
∗
i
+ t
i
+ t
i
d

i
p
∗
i
+ t
i
−
a
2
i
l
i
λ
i
(1 − β
i
)
(λ
i
+ θ
i
)
2
kβ
i
−
∂θ
i
∂t
i
a
i
l
i
λ
i
(λ
i
+ θ
i
)
2
β
i
β
i
+
2
λ
i
+ θ
i

1
δ
− 1 −
a
i
t
i
(
1 − β
i
)
k

.
By definition, we have d
i
(p
i
) ≡ φ
−1
i
(p
i
). Thus, φ

i
< 0 implies that
d
i
(p
i
) = 1/φ

i
φ
−1
i
(p
i
)
< 0,
and φ

i
≤ 0 implies that
d

i
(p
i
) =−
φ

i
φ
−1
i
(p
i
)
φ

i
φ
−1
i
(p
i
)

3
≤ 0.
So, we get
d
i
p
∗
i
+ t
i
+ t
i
d

i
p
∗
i
+ t
i
< 0.
The third term, −a
2
i
l
i
λ
i
(1 − β
i
)/(λ
i
+ θ
i
)
2
kβ
i
, clearly is negative. Similarly,
t
i
<
¯
t<k(1/δ − 1)/a
i
(1 − β
i
) implies 1/δ − 1 − a
i
t
i
(1 − β
i
)/k > 0, which
means that the last term is negative as well. Combining the previous observations,
we get ∂
2
G/∂t
2
i
< 0 for all t
i
> 0. Because ∂
2
G/∂t
i
∂t
j
= 0 for all j = i and [t,
¯
t]
is a compact set, there exists a unique vector of equilibrium policies (t
0
1
,...,t
0
n
).
In particular, any interior equilibrium policy t
<t
0
i
<
¯
t satisfies (∂G/∂t
i
)
t
0
i
= 0
which is equivalent to condition (15).
1034 Journal of the European Economic Association
A.2. Proof of Proposition 1
By definition, the interior equilibrium policy satisfies (∂G/∂t
i
)
t
0
i
= 0. Hence, the
implicit function theorem implies
∂t
o
i
∂z
i
=−
∂
2
G
∂z
i
∂t
i
t
o
i

∂
2
G
∂t
2
i
t
o
i
(A.1)
for all z
i
∈{a
i
,β
i
,l
i
,λ
i
,p
∗
i
}. From the proof of Lemma 1, we already know that
(∂
2
G/∂t
2
i
)
t
o
i
< 0 . Thus, ∂t
o
i
/∂z
i
must have the same sign as (∂
2
G/∂z
i
∂t
i
)
t
o
i
.We
now compute the signs of the cross-derivatives associated with our five exogenous
parameters.
Claim (i). (∂
2
G/∂a
i
∂t
i
)
t
o
i
< 0.
Proof. Consider equation (A.1). Because the first term of the sum does not depend
on a
i
, we only need to show that the last two terms are decreasing in a
i
. By equation
(8), we have
a
i
λ
i
+ θ
i
=
p
i
(1 − β
i
)
kβ
i
−
r
a
i
β
i
−
1 − β
i
a
i
β
i
λ
i
−1
,
which, by inspection, is decreasing in a
i
. Similarly, we have
a
i
(λ
i
+ θ
i
)
2
= a
−1
i
p
i
(1 − β
i
)
kβ
i
−
r
a
i
β
i
−
1 − β
i
a
i
β
i
λ
i
−2
,
which is positive and decreasing in a
i
. Because 1/δ −1−a
i
t
i
(1−β
i
)/k is positive
and decreasing in a
i
as well, the last term also is decreasing.
Claim (ii). (∂
2
G/∂β
i
∂t
i
)
t
0
i
> 0.
Proof. Consider equation (A.1). Following the same logic as in claim (i), we only
need to show that 1/(λ
i
+ θ
i
),1/(λ
i
+ θ
i
)
2
β
i
and 1/δ − 1 − a
i
t
i
(1 − β
i
)/k are
increasing in β
i
. By equation (8), we have
1
λ
i
+ θ
i
=

1 − β
i
β
i

p
i
a
i
k
−
r
(1 − β
i
)
− λ
i

−1
,
which, by inspection, is increasing in β
i
. Similarly, we have
1
λ
i
+ θ
i
2
β
i
=
(1 − β
i
)
2
β
i
−1
p
i
a
i
k
−
r
(1 − β
i
)
− λ
i
−2
,
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1035
which is increasing in β
i
as well. Finally, 1/δ−1−a
i
t
i
(1−β
i
)/k also is increasing
in β
i
by inspection.
Claim (iii). (∂
2
G/∂l
i
∂t
i
)
t
0
i
> 0.
Proof. Consider equation (A.1). Because θ
i
does not depend on l
i
, we immedi-
ately get that ∂G/∂t
i
is increasing in l
i
.
Claim (iv). (∂
2
G/∂λ
i
∂t
i
)
t
0
i
> 0.
Proof. Consider equation (A.1). We only need to show that λ
i
/(λ
i
+ θ
i
)
2
and
λ
i
/(λ
i
+ θ
i
) are increasing in λ
i
. By equation (8), we have
λ
i
(λ
i
+ θ
i
)
2
= λ
i
p
i
a
i
(1 − β
i
)
kβ
i
−
r
β
i
−
1 − β
i
β
i
λ
i
−2
,
which, by inspection, is increasing in λ
i
. Similarly, we have
λ
i
λ
i
+ θ
i
= λ
i
p
i
a
i
(1 − β
i
)
kβ
i
−
r
β
i
−
1 − β
i
β
i
λ
i
−1
,
which is increasing in λ
i
as well.
Claim (v). (∂
2
G/∂p
∗
i
∂t
i
)
t
0
i
< 0.
Proof. Consider equation (A.1). By equation (8), θ
i
is increasing in p
∗
i
. So,
1/(λ
i
+ θ
i
)
2
and 1/(λ
i
+ θ
i
) are decreasing in p
∗
i
. We also know that t
0
i
> 0by
equation (15). Thus, we only need to check that d
i
(p
i
) is decreasing in p
∗
i
, which
is true by the proof of Lemma 1.
A.3. Proof of Proposition 2
Consider two workers, j
1
and j
2
, employed in sector i before trade liberalization
such that a
j
1
i
= a
j
2
i
and σ
j
1
≥ σ
j
2
. First, note that a
j
1
i
= a
j
2
i
implies u
j
1
i
= u
j
2
i
,
du
j
1
i
= du
j
2
i
, and W
j
1
i
= W
j
2
i
. Second, note that σ
j
1
≥ σ
j
2
implies
U
j
1
i
≤
U
j
2
i
for all i
= 0,...,n, and so
U
j
1
≤
U
j
2
. Combining these results with equation
(21), we get: dW
j
1
i
≤ dW
j
2
i
.
1036 Journal of the European Economic Association
A.4. Proof of Proposition 3
Let us first introduce some additional notations. We define f(a
j
i
) as
f(a
j
i
) ≡ a
j
i
+
λ
i
a
j
i
(1 − β
i
)
kβ
i
θ
j
i
λ
i
+ θ
j
i
×
⎡
⎣
1
δ
− 1
kβ
i(j)
1 − β
i(j)
−ˆp
i(j)
a
j
i(j)
+ˆp
i
a
j
i
−
kλ
i
1 − β
i
+
kλ
i(j)
1 − β
i(j)
⎤
⎦
,
where i(j) = arg max
0≤i
≤n
U
j
i
, and ˆp
i
and ˆp
i(j)
are the domestic prices of
goods i and i(j) after trade liberalization, respectively. Using equation (18), we
can express the derivative of f with respect to a
j
i
as
∂f
∂a
j
i
= 1−g
1
a
i
j
,δ
⎧
⎨
⎩
ˆp
i(j)
a
j
i(j)
a
j
i
−ˆp
i
+ g
2
(a
i
j
,δ)
⎡
⎣
1
δ
− 1
kβ
i(j)
1 − β
i(j)
+ g
3
a
i
j
⎤
⎦
⎫
⎬
⎭
,
where the three functions g
1
, g
2
, and g
3
are given by
g
1
a
j
i
,δ
=
λ
i
(1 − β
i
)a
j
i
kβ
i
&
p
i
a
j
i
(1 − β
i
) − k
1
δ
− 1 + λ
i
'&
p
i
a
j
i
(1 − β
i
) − k
1
δ
− 1 + λ
i
(1 − β
i
)
'
>0,
g
2
(a
j
i
,δ)=
p
i
(1 − β
i
)
p
i
a
j
i
(1 − β
i
) − k
1
δ
−1+λ
i
+
p
i
(1−β
i
)
p
i
a
j
i
(1−β
i
) − k
1
δ
− 1+λ
i
(1−β
i
)
−
1
a
j
i
>0,
g
3
a
j
i
=ˆp
i
a
j
i
−ˆp
i(j)
a
j
i(j)
+
kλ
i(j)
1 − β
i(j)
−
kλ
i
1 − β
i
.
We also define
g
1
= min
i,j
λ
i
(1 − β
i
)a
j
i
kβ
i
p
i
a
j
i
(1 − β
i
) − kλ
i

p
i
a
j
i
(1 − β
i
) − kλ
i
(1 − β
i
)
> 0,
g
2
= min
i,j
p
i
(1 − β
i
)
p
i
a
j
i
(1 − β
i
) − kλ
i
+
p
i
(1 − β
i
)
p
i
a
j
i
(1 − β
i
) − kλ
i
(1 − β
i
)
−
1
a
j
i
> 0,
g
3
= min
i,j
ˆp
i
a
j
i
− max
i,j
ˆp
i
a
j
i
+ min
i
kλ
i
1 − β
i
− max
i
kλ
i
1 − β
i
< 0,
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1037
and
δ
=
1 + max
i
1 − β
i
β
i
·

1
g
1
+ max
i
ˆp
i
1
g
2
− g
3

−1
> 0. (A.2)
Claim (a.i). If f is decreasing in a
j
i
, then dW
j
i
is increasing in a
j
i
.
Proof. The change in expected lifetime utility of a worker j employed in sector
i is equal to
(1 − δ)dW
j
i
= d[τ + s(p)]+a
j
i
dt +
(1 − δ)λ
i
a
j
i
(1 − β
i
)
kβ
i
θ
j
i
λ
i
+ θ
j
i
W
j
i
−
U
j
dt. (A.3)
Using equations (19) and (20), we can express
W
j
i
−
U
j
as
W
j
i
−
U
j
=
1
(1 − δ)
(1/δ−1)kβ
i(j)
1 − β
i(j)
−ˆp
i(j)
a
j
i(j)
+ˆp
i
a
j
i
−
kλ
i
1 − β
i
+
kλ
i(j)
1 − β
i(j)
.
(A.4)
Combining equations (A.3) and (A.4), we obtain
(1 − δ)dW
j
i
= d[τ + s(p)]+dt ·
a
j
i
+
λ
i
a
j
i
(1 − β
i
)
kβ
i
θ
j
i
λ
i
+ θ
j
i
×
(1/δ − 1)kβ
i(j)
1 − β
i(j)
−ˆp
i(j)
a
j
i(j)
+ˆp
i
a
j
i
−
kλ
i
1 − β
i
+
kλ
i(j)
1 − β
i(j)
(
.
Because d[τ + s(p)] does not depend on a
j
i
and dt < 0, f decreasing in a
j
i
implies dW
j
i
increasing in a
j
i
.
Claim (a.ii). If δ ≤ δ, then ∂f/∂a
j
i
≤ 0 for all a
j
i
.
Proof. Equation (A.2) implies
g
1
)
−max
i
ˆp
i
+ g
2

1
δ
− 1
min
i
β
i
1 − β
i
+ g
3
(
= 1. (A.5)
By construction, we have: g
1
> 0, g
2
> 0, and g
3
(a
i
j
) ≥ g
3
. Hence, equation
(A.5) further implies
g
1
ˆp
i(j)
1 − σ
j
i
−ˆp
i
+ g
2
1/δ
− 1
kβ
i(j)
1 − β
i(j)
+ g
3
(a
i
j
)

≥ 1
1038 Journal of the European Economic Association
for all a
j
i
, where
W
j
i
−
U
j
> 0 implies
(1/δ
− 1)kβ
i(j)
1 − β
i(j)
+ g
3
a
i
j
> 0.
Note that g
1
and g
2
are decreasing in δ. As a result, we have
g
1
a
j
i
,δ
≥
λ
i
(1 − β
i
)a
j
i
kβ
i
p
i
a
j
i
(1 − β
i
) − kλ
i

p
i
a
j
i
(1 − β
i
) − kλ
i
(1 − β
i
)
≥ g
1
,
g
2
a
j
i
,δ
≥
p
i
(1 − β
i
)
p
i
a
j
i
(1 − β
i
) − k(r + λ
i
)
+
p
i
(1 − β
i
)
p
i
a
j
i
(1 − β
i
) − k(r + λ
i
(1 − β
i
))
−
1
a
j
i
≥ g
2
.
Combining the last series of inequalities, we get
g
1
a
i
j
,δ
)
ˆp
i(j)
1 − σ
j
i
−ˆp
i
+ g
2
a
i
j
,δ
(1/δ
− 1)kβ
i(j)
1 − β
i(j)
+ g
3
(a
i
j
)
(
≥ 1
for all a
j
i
.Ifδ ≤ δ, we obtain in turn
g
1
a
i
j
,δ
)
ˆp
i(j)
1 − σ
j
i
−ˆp
i
+ g
2
a
i
j
,δ
(1/δ − 1)kβ
i(j)
1 − β
i(j)
+ g
3
a
i
j
(
≥ 1
for all a
j
i
. This is equivalent to ∂f/∂a
j
i
≤ 0 for all a
j
i
.
Claims (a.i) and (a.ii) imply that if δ is small enough, then workers are less
likely to be protectionist if their productivity a
j
i
is high.
Appendix B: Table 1
All data are from the World Values Survey 1994–1999. “High income” countries
include: Australia, Finland, West Germany, New Zealand, Norway, Puerto Rico,
Republic of Korea, Spain, Sweden, Switzerland, Taiwan, United States. “Rest of
the world” includes: Albania, Argentina, Bangladesh, Bosnia and Herzegovina,
Brazil, Bulgaria, Chile, China, Colombia, Czech Republic, Dominican Republic,
Hungary, India, Macedonia, Mexico, Nigeria, Pakistan, Peru, Philippines, Roma-
nia, Serbia and Montenegro, Slovakia, Slovenia, South Africa, Turkey, Uruguay,
Venezuela. “All Former Soviet Republics”—Armenia, Azerbaijan, Belarus, Esto-
nia, Latvia, Lithuania, Russian Federation, and Ukraine—have been omitted from
the sample. These countries were in the middle of their transition programs at the
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1039
time of the surveys; what may have determined their trade-policy preferences lies
beyond the scope of our paper.
“Upper” levels of education include: some university without degree/higher
education-lower-level tertiary certificate; and university with degree/higher
education-upper-level tertiary certificate. “Middle” levels of education include:
complete secondary school: technical/vocational type/secondary, interme-
diate vocational qualification; incomplete secondary: university-preparatory
type/secondary, intermediate general qualification; and complete secondary:
university-preparatory type/full secondary, maturity level certificate. “Lower”
levels of education include: inadequately completed elementary education; com-
pleted (compulsory) elementary education; and incomplete secondary school:
technical/vocational type/(compulsory) elementary education.
We consider the following question of the World Values Survey: “Do you
think it is better if goods made in other countries can be imported and sold here
if people want to buy them, or that there should be stricter limits on selling
foreign goods here, to protect the jobs of people in this country?” The proportion
of protectionist opinions is computed cell by cell as the number of employed
respondents who think that “there should be stricter limits on selling foreign
goods” divided by the total number of employed respondents.
References
Anderson, Kym (1980). “The Political Market for Government Assistance to Australian
Manufacturing Industries.” The Economic Record, 56, 132–144.
Anderson, Kym, and Robert E. Baldwin (1987). “The Political Market for Protection in Indus-
trial Countries.” In Protection, Cooperation, Integration and Development, edited by A. M.
El-Agraa. Macmillan.
Bagwell, Kyle, and Robert W. Staiger (2003). “Protection and the Business Cycle.” Advances
in Economic Analysis and Policy, 3-1, article 3.
Baldwin, Robert E. (1985). The Political Economy of U.S. Import Policy. MIT Press.
Baldwin, Richard E., and Frédéric Robert-Nicoud (2007). “Entry and Asymmetric Lobbying:
Why Governments Pick Losers.” Journal of the European Economic Association, 5, 1064–
1093.
Beaulieu, Eugene, Michael Benarroch, and James D. Gaisford (2001). “Intra-Industry Trade
Liberalization: Why Skilled Workers in Most Countries Resist Protectionism.” Working
paper, University of Calgary.
Bhagwati, Jagdish (1971). “The Generalized Theory of Distortions and Welfare.” In Trade,
Balance of Payments, and Growth: Papers in International Economics in Honor of Charles
P. Kindleberger, edited by J. Bhagwati, R. Jones, R. Mundell, and J. Vanek. North-Holland
Publishing Company.
Bohara, Alok K., and William H. Kaempfer (1991). “A Test of Tariff Endogeneity in the United
States.” American Economic Review, 81, 952–960.
Bradford, Scott (2006). “Protection and Unemployment.” Journal of International Economics,
69, 257–271.
Caves, Richard E. (1976). “Economic Models of Political Choice: Canada’s Tariff Structure.”
Canadian Journal of Economics, 9, 278–300.
1040 Journal of the European Economic Association
Chaudhuri, Shubham, and John E. McLaren (2007). “Some Simple Analytics of Trade and
Labor Mobility.” NBER Working Paper No. 13464.
Davidson, Carl, Lawrence W. Martin, and Steven J. Matusz (1994). “Jobs and Chocolate:
Samuelsonian Surpluses in Dynamic Models of Unemployment.” Review of Economic
Studies, 61, 173–192.
Davidson, Carl, Lawrence W. Martin, and Steven J. Matusz (1999). “Trade and Search
Generated Unemployment.” Journal of International Economics, 48, 271–299.
Davis, Donald R., and James Harrigan (2007). “Good Jobs, Bad Jobs, and Trade Liberalization.”
NBER Working Paper No. 13139.
Egger, Hartmut, and Udo Kreickemeier (2006). “Firm Heterogeneity and the Labour Market
Effects of Trade Liberalization.” GEP Research Paper No. 2006/26.
Findlay, Ronald, and Stanislaw Wellisz (1982). “Endogenous Tariffs, the Political Economy
of Trade Restrictions, and Welfare.” In Import Competition and Response, edited by J.
Bhagwati. University of Chicago Press.
Finger, Michael, and Ann Harrison (1994). “The MFA Paradox: More Protection and More
Trade?” Working Paper, the World Bank.
Goldberg, Pinelopi K., and Giovanni Maggi (1999). “Protection for Sale: An Empirical
Investigation.” American Economic Review, 89, 1135–1155.
Grossman, Gene M., and Elhanan Helpman (1994). “Protection for Sale.” American Economic
Review, 84, 833–850.
Hall, Robert E., and Charles I. Jones (1999). “Why Do Some Countries Produce so Much More
Output Per Worker than Others?” Quarterly Journal of Economics, 114, 83–116.
Hansen, John M. (1990). “Taxation and the Political Economy of the Tariff.” International
Organization, 44, 527–552.
Helpman, Elhanan (1998). “Politics and Trade Policy.” NBER Working Paper No. 5309.
Helpman, Elhanan, and Oleg Itskhoki (2007). “Labor Market Rigidities, Trade and Unemploy-
ment.” NBER Working Paper No. 13365.
Helpman, Elhanan, and Paul Krugman (1989). Trade Policy and Market Structure. MIT Press.
Hillman, Arye L. (1982). “Declining Industries and Political-Support Protectionist Motives.”
American Economic Review, 72, 1180–1187.
Janiak, Alexandre (2006). “Does Trade Liberalization Lead to Unemployment? Theory and
Some Evidence.” Working Paper, Universite Libre de Bruxelles.
Kletzer, Lori G. (1998). “Job Displacement.” Journal of Economic Perspectives, 12, 115–136.
Magee, Christopher, Carl Davidson, and Stephen Matusz (2005). “Trade, Turnover, and
Tithing.” Journal of International Economics, 66, 157–176.
Magee, Stephen P. (1980). “Three Simple Tests of the Stolper-Samuelson Theorem.” In Issues
in International Economics, edited by P. Oppenheimer. Oriel Press.
Mansfield, Edward D., and Marc L. Busch (1995). “The Political Economy of Nontariff
Barriers: A Cross-National Analysis.” International Organization, 49, 723–749.
Marvel, Howard P., and Edward J. Ray (1983). “The Kennedy Round: Evidence on the
Regulation of Trade in the U.S.” American Economic Review, 73, 190–197.
Matschke, Xenia, and Shane M. Sherlund (2006). “Do Labor Issues Matter in the Determination
of U.S. Trade Policy? An Empirical Reevaluation.” American Economic Review, 96, 405–
421.
Mayda, Anna Maria, and Dani Rodrik (2005). “Why Are Some People (and Countries) More
Protectionist than Others?” European Economic Review, 49, 1393–1430.
Mincer, Jacob (1993). “Education and Unemployment.” In Studies in Human Capital, edited
by Jacob Mincer. Edward Elgar Publishing.
Neal, Derek (1995). “Industry Specific Human Capital: Evidence from Displaced Workers.”
Journal of Labor Economics, 13, 653–677.
O’Halloran, Sharyn (1994). Politics, Process, and American Trade Policy. University of
Michigan Press.
Costinot Jobs, Jobs, Jobs: A “New” Perspective on Protectionism 1041
O’Rourke, Kevin H., and Richard Sinott (2001). “The Determinants of Individual Trade Policy
Preferences: International Survey Evidence.” In Brookings Trade Forum: 2001, edited by
S. Collins and D. Rodrik. Brookings.
Parent, Daniel (2000). “Industry-Specific Capital and the Wage Profile: Evidence from the
Longitudinal Survey of Youth and the Panel Study of Income Dynamics.” Journal of Labor
Economics, 18, 306–323.
Pissarides, Christopher A. (2000). Equilibrium Unemployment Theory. MIT Press.
Ray, Edward J. (1981). “The Determinants of Tariffs and Non-Tariffs Restrictions in the United
States.” Journal of Political Economy, 89, 105–121.
Ray, Edward J. (1991). “Protection of Manufactures in the United States.” In Global
Protectionism: Is the U.S. Playing on a Level Field? edited by D. Greenaway. Macmillan.
Ray, E. (1987). “Changing Patterns of Protectionism: The Fall in Tariffs and the Hise in Non-
tariff Barriers.” Northwestern Journal of International Law and Business, 8, 285–327.
Rodriguez, Francisco, and Dani Rodrik (1999). “Trade Policy and Economic Growth: A
Skeptic’s Guide to the Cross-Country Evidence.” NBER Working Paper No. 7081.
Rodrik, Dani (1995). “Political Economy of Trade Policy.” In Handbook of International
Economics, vol. 3, edited by G. Grossman and K. Rogoff. North-Holland Publishing
Company.
Rogowski, Ronald (1987). “Political Cleavages and Changing Exposure to Trade.” American
Political Science Review, 81, 1121–1137.
Saunders, Ronald (1980). “The Political Economy of Effective Protection in Canada’s
Manufacturing Sector.” Canadian Journal of Economics, 13, 340–348.
Scheve, Kenneth F., and Matthew J. Slaughter (2001). “What Determines Individual Trade-
Policy Preferences?” Journal of International Economics, 54, 267–292.
Scheve, Kenneth, and Matthew J. Slaughter (2004). “Public Opinion, International Economic
Integration, and the Welfare State.” Working Paper, Yale University.
Trefler, Daniel (1993). “Trade Liberalization and the Theory of Endogenous Protection.”
Journal of Political Economy, 101, 138–160.
Trefler, Daniel (2004). “The Long and Short of the Canada-US Free Trade Agreement.”
American Economic Review, 94, 870–895.
Wasmer, Etienne (2006). “General versus Specific Skills in Labor Markets with Search Frictions
and Firing Costs.” American Economic Review, 96, 811-831.
World Values Survey 1994–1999. Ann Arbor, Mich.: Inter-university Consortium for Political
and Social Research.