internal geography, international trade, and regional...
TRANSCRIPT
Internal Geography, International Trade, and Regional Outcomes
A. Kerem CosarU. of Chicago Booth
Pablo D. FajgelbaumUCLA
December 26, 2012
Abstract
When trade is costly within countries, international trade leads to concentration of eco-
nomic activity in locations with good access to foreign markets. Costly trade within countries
also makes it hard for remote locations to gain from international trade. We investigate the
role of these forces in shaping industry location, employment concentration and the gains from
international trade. We develop a model that features comparative advantages between coun-
tries, coupled with differences in proximity to international markets across locations within a
country. International trade creates a partition between a commercially integrated coastal re-
gion with high population density and an interior region where immobile factors are poorer.
Reductions in domestic or international trade costs generate migration to the coastal region
and net welfare losses for fixed factors in the interior region. We present motivating evidence
from the U.S. which shows that export-oriented industries are more likely to locate closer to
international ports. Then, we use the model to measure the importance of international trade
in concentrating economic activity, and of domestic trade costs in hampering the gains from
international trade.
1 Introduction
International trade can be an important determinant of the geographic distribution of economic
activity within countries. Some recent examples illustrate this. In China, the export growth of the
last few decades occurred jointly with movements of rural workers and export oriented industries
toward coastal regions (World Bank, 2009). Similarly, after the U.S.-Vietnamese trade agreement,
employment expanded in Vietnamese comparative-advantage industries located closer to major
seaports (McCaig and Pavnick, 2012). In these cases international trade caused migrations toward
coasts, whereas in others it led to larger employment near borders, as it was observed between the
U.S. and Mexico during a time of economic integration (Hanson, 1996). More generally, it is well
known that the distribution of employment is skewed toward coasts: half of the world population
lives within 100 kilometers of coastlines or navigable rivers, while 19 out of the 25 largest cities in
the world are coastal.1 While the appeal of coastal sites derives partly from their resources and
amenities, a reason for their primacy is that they are well suited for trading with other countries
and regions.
These examples showcase the effects of international trade on regional economic activity as much
as the importance of costly trade within countries. The larger the domestic trade costs, the stronger
the incentives for export oriented activity to concentrate in places with good access to international
markets. Costly trade within countries also makes it diffi cult for remote locations inside countries
to share the gains from international trade.2 Indeed, recent empirical evidence highlights that
proximity to international trade gateways facilitates regional development.3 Therefore, accounting
for within-country heterogeneity in access to foreign markets may be relevant to understand how
the gains from international trade are determined.
In this paper, we develop a theory of international trade with costly trade within countries
to study how international trade and the internal geography of countries interact to shape the
concentration of economic activity and the gains from international trade. The model allows local
specialization patterns to depend on market access, and it implies that changes in internal and
international trade costs lead to migration of workers to coastal regions which host comparative-
advantage industries. It also implies that the welfare effects of trade on immobile factors are small
or negative in distant locations, and that the national gains from trade decrease with domestic
trade costs. We use the model for a quantitative analysis of U.S. data, where we measure the effect
of international trade on the concentration of employment as well as the role of domestic trade
costs in hampering the gains from trade.
We focus on classical comparative advantages across countries as the driver for concentration
of economic activity in places with good market access. Because countries have incentives for
inter-industry trade, export-oriented industries tend to locate closer to international gates, which
1Authors’calculation from Harvard CID. The reported number is for 1995.2Atkin and Donaldson (2012) provide measurement of internal trade costs in several developing countries.3For example, Storeygard (2012) shows that regions in sub-Saharan Africa with better access to port cities benefit
more from terms of trade shocks. In the U.S., Blonigen and Cristea (2012) exploit changes in regulation to show thatincreased access to air services has a positive impact on regional development.
2
in turn leads to higher employment density in these locations. This approach complements ex-
planations based on economies of scale, as in the New Economic Geography literature related to
Krugman (1991), where firms or products within an industry locate closer to demand to exploit scale
economies. Here, cross-country comparative advantages drive concentration through the location
of export-oriented industries in places with favorable foreign market access.
The emphasis on these forces is natural in the light of existing evidence. The location of
comparative-advantage industries in coastal or border regions is a common feature of the case
studies cited above where international trade led to internal migrations. In Section 2, we present
additional motivating evidence from U.S. data. We find that the distribution of employment in
industries with higher export-output ratios at the national level is biased toward counties situated
closer to gates for international trade such as airports, seaports or land crossings. Controlling
for industry and county fixed effects, employment at the county-industry level decreases with the
interaction between industry export orientation at the national level and county distance to interna-
tional gates. Moving inland from a representative port for 150miles, employment in export-oriented
industries relative to import-competing industries shrinks by between 4% and 12%.
In the theory, we furnish the canonical 2-sector model of international trade driven by compar-
ative advantages with a geography within countries. Locations are arbitrarily arranged on a map
and differ in distance to international gates such as seaports, airports or land crossings. Within
the country trade is costly, and international shipments must cross through the international gates
to reach foreign markets. To produce, each location uses both mobile and immobile factors. De-
creasing returns in production lead to congestion, so that it is not optimal to concentrate activity
in a single location. For quantitative purposes, we also allow for complementary forces that lead
to congestion or agglomeration, such as demand for non-tradables and external economies of scale.
An important aspect of the analysis is that comparative advantages are defined at the national
level. This is the standard assumption in Ricardian and Hecksher-Ohlin models of international
trade, where countries are defined by their set of technologies or endowments. We assume that
all locations within a country share the same relative productivities with respect to the rest of
the world, although some locations may be more productive than others in every industry. This
is consistent with the notion that technology flows are relatively costless within countries, and
also with the view that international differences in institutions are an important determinant of
specialization.4 The key implication that stems from this assumption is that only the distance
separating each location from its nearest international gate matters for equilibrium outcomes. This
allows to represent a two-dimensional geography on the line, leading to closed form characterizations
for aggregate equilibrium outcomes and making the model tractable for counterfactuals.
We use the model to study regional patterns of specialization and employment in open economy.
Whenever the economy is not fully specialized in what it exports, two distinct regions necessarily
4Nunn (2007) shows that institutional quality has as much explanatory power on the pattern of internationaltrade as physical capital or skilled labor. Manova (2008) and Cuñat and Melitz (2012) emphasize, respectively, therole of financial and labor market institutions in determining the pattern of trade. Arguably, these institutions varylittle within countries compared to the cross-country differences that motivate trade.
3
emerge. The equilibrium features regions near international gates that specialize in export-oriented
goods, followed by interior regions that are incompletely specialized and do not trade with the rest
of the world. Relative to international autarky, trade causes higher population density close to
international gates. Thus, even though trade costs are uniform across space, international trade
creates a partition between a commercially integrated coastal region with high population density
and an interior region where immobile factors are poorer. The geographic extension of coastal and
interior regions, as well as their shares in employment and income, depend on international and
domestic trade costs.
Then, we analyze the impact of international and domestic trade costs on internal migrations.
Reductions in international trade costs lead to migration of mobile factors toward the locations with
good international market access. Lower international trade costs also cause the boundary between
the coastal and the interior regions to move inland. Marginal locations switch from autarky to
specialization in export oriented industries and they start trading with the rest of the world, so
that the geographic extension of the integrated region increases. As a result, the level of economic
activity and the share in national income increases in the commercially integrated regions located
closer to international gates relative to the interior regions. Reductions in domestic trade costs
cause net migration away from the interior region, but they also cause population density to shrink
in coastal locations close to ports.
We conclude the theoretical analysis with the impact of internal geography on the gains from
international trade. International trade has opposing welfare effects on immobile factors located
in different points of the country. Immobile factors in the interior regions loose from reductions in
international or internal trade costs, while some fixed factors in the coastal region necessarily gain.
In turn, the national gains from trade are a weighted average of the gains from trade across these
heterogeneous locations. They can be decomposed into a familiar term that captures the gains from
trade without internal geography and a term that captures the effect of domestic trade frictions.
The first component depends on the terms of trade, as in classical models of trade, while the second
is given by domestic trade costs, the size of the trading region, and the distribution of land and total
productivity across locations. Since larger domestic frictions cause the trading region to shrink,
the gains from international trade decrease with domestic trade costs. In the quantitative section
we measure this complementarity between international and internal trade costs.
Thus, the model reproduces the salient facts on the interaction between international trade and
regional outcomes: economic activity and comparative-advantage industries concentrate in areas
with good access to international markets, and international trade integration is correlated with
migration toward these areas. In Section 4 we study the quantitative relevance of the theory.
For that, we calibrate the key parameters (domestic trade costs, decreasing returns to scale, and
international comparative advantages) to match moments from the U.S. data that are independent
from the geographic distribution of employment. Then, we use the calibrated model to predict
the concentration of economic activity relative to the data, and to measure the impact of internal
geography on the gains from trade.
4
In a first exercise we consider a version of our model with tradable goods only where interna-
tional trade is the only possible determinant of concentration. Then, we bring in additional forces
that lead to congestion and to agglomeration, such as consumption of non-tradables and external
economies of scale. In each case, we measure the role of trade in concentrating economic activity by
comparing the concentration in coastal areas predicted by the calibrated model with its empirical
counterpart. In a final exercise we also consider heterogeneity in fundamentals across locations. We
treat fundamentals as a residual by choosing their distribution to match the coastal concentration of
economic activity observed in the data, and we assess the role of trade by measuring concentration
in counterfactual autarkic scenarios.
Depending on the quantitative approach, we find that 10% to 40% of the coastal concentration
of the U.S. population can be accounted for by the interaction between U.S. comparative advan-
tages and domestic trade costs. The remaining part is explained by the combination of amenities,
agglomeration, and congestion forces. In our baseline calibration we measure the gains from inter-
national trade in the U.S. to be in the order 0.1%. But, keeping all other parameters the same,
when domestic trade costs are suppressed these gains rise to 4.6%. These results on population
density and on gains from trade hold even though domestic trade costs in the U.S. are quite small.
In the calibration, they represent 1.5% of the f.o.b price of exported products.
Given that the baseline model does not include a number of additional forces such as within-
country differences in comparative advantages, these magnitudes are likely an upper bound for the
effect of domestic frictions. Still, the large potential impact of small domestic costs suggests that
local trade frictions might play an even more important role in poor countries, where infrastructure
is presumably less developed than in the U.S.
Relation to the Literature Few studies consider an interaction between international and
domestic trade costs. In a New Economic Geography context, Krugman and Livas-Elizondo (1996)
and Behrens et al. (2006) present models where two regions within a country trade with the rest
of the world. Henderson (1982) and Rauch (1991) embed system of cities models in open economy
frameworks. Rossi-Hansberg (2004) studies the location of industries on a continuous space with
spatial externalities. These papers are based on agglomeration forces, whereas we focus on the
interaction between heterogeneous market access within countries and comparative advantages
between countries.
Matsuyama (1999) studies a multi-region extension of Helpman and Krugman (1985). He
focuses on home-market effects under different spatial configurations, but does not include factor
mobility. In neoclassical environments, Bond (1993) and Courant and Deardorff (1993) present
models with regional specialization where relative factor endowments may vary across discrete
regions within a country. These papers do not include heterogeneity in access to world markets.
Venables and Limao (2002) study geographic specialization across regions trading with a central
location but do not allow for factor mobility.
More recently, Ramondo et al. (2011) study the gains from trade and ideas diffusion allowing
5
for multiple regions within countries. They do not focus on differences in world market access
across locations or on labor mobility within countries. In their framework, domestic trade costs
do not interact with the gains from trade. Redding (2012) extends the framework in Eaton and
Kortum (2002) with labor mobility across regions. In his analysis, there are multiple goods within
a tradeable industry, and for each good there are independent productivity draws across locations.
Similarly, Allen and Arkolakis (2012) explore the spatial distribution of economic activity in an
arbitrary topography of trade costs where products are differentiated by origin. The distinctive
aspect of our analysis is the focus on industry location and specialization as drivers for concen-
tration. Industries choose their location based on comparative advantages at the national level,
driving activity to places with good market access.
Finally, a large literature in urban economics studies determinants of industry location. Holmes
and Stevens (2004) offer a summary assessment of the forces determining industry location in
the U.S., while Hanson (1998) presents a literature review centered on the role of international
trade. Our explanation for industry location is in line with his view and complements common
explanations such as natural advantages or pure agglomeration effects.
Structure of the Paper We start in Section 2 with motivating evidence for the U.S.. In section
3 we lay out the model and characterize the general equilibrium and comparative statics . In Section
4 we present the quantitative assessment. Section 5 concludes. Proofs and description of the data
are in the appendix.
2 Motivating Evidence
The central feature of our approach is that concentration of employment follows from the loca-
tion export-oriented industries in places with good access to foreign market. As it was highlighted
in the introductory examples, the location of export-oriented industries in places with good foreign
market access is a feature common to many developing countries. To motivate our analysis further,
we show that the positive correlation between industry export orientation at the national level and
proximity to international gates is also present in U.S. data.
We build a location-specific measure of market access by computing the shortest distance be-
tween each of the 48 continental U.S. states or 3108 counties and the gateways for international
trade. There are 288 international ports (airports, seaports or land crossings) in the mainland
for U.S. goods trade. We use data on trade volume by port to identify the 49 largest ports that
account for 90% of total U.S. trade. These ports are located in 38 different counties.5 For each U.S.
state and county, we then calculate the great-circle distance between its population center and the
population center of the nearest county where one of these 49 ports is located. We also compute
an alternative measure that drops the inland airports.6 Summary statistics are reported in Table
5See Figure 8 in the Data Appendix for a map of port locations.6The list of 49 largest ports includes three inland airports: Atlanta, Dallas and Salt Lake City. They account for
3.6% of total trade.
6
1. Counties that contain a port have distance equal to zero. There is considerable variation in
distance to the nearest port at both levels of geographical aggregation.
Table 1: Summary Statistics of the Distance Measure (miles)
States CountiesBaseline Excluding Baseline Excluding(49 ports) airports (49 ports) airports
Mean 158 198 194 233St. Deviation 114 154 121 145Median 140 145 175 215Maximum 500 560 571 637Minimum 15 15 0 0
A natural implication of the forces under consideration is that regions with better market access
should export a larger share of their output and employ more workers in export-oriented activities.
We plot these variables at the state level against state distance in the two panels of Figure 1. Both
export intensity and employment in export oriented goods decline with distance. The correlation
coeffi cient between each of these measures and distance is negative and statistically significant. The
left panel implies that a reduction in distance to international gates from 400 miles to less than
100 miles results in a three-fold increase in the export to value added ratio at the state level.
Figure 1: Export Intensity and Distance to Ports Across U.S. States
AL
AZ
AR
CA
CO
CT
DE
FL
GA IDIL
IN
IA
KSKYLA
ME
MD
MA
MI
MN
MSMO
MT
NE
NV
NH
NJ
NM
NY
NC
ND
OH
OK
OR
PARI
SC
SD
TN
TXUT
VT
VA
WA
WV
WI
WY
0.2
5.5
.75
11.
251.
5
Expo
rts/V
alue
Add
ed in
trad
able
s
0 100 200 300 400 500
Dis tance (miles)
Slope : 0.001266tvalue : 4.03R2 : 0.21
ALAZ
AR
CACO
CT
DE
FL
GA
ID
ILIN
IA
KS
KY
LAME
MD
MA
MI
MN
MS
MO
MTNE
NV
NH
NJ
NM
NY
NC
ND
OH
OK
OR
PA
RI
SC
SD
TN
TX
UT
VT
VA
WA
WVWI
WY
0.0
5.1
.15
.2.2
5.3
.35
.4
% o
f man
ufac
turin
g em
ploy
men
t rel
ated
to e
xpor
ts
0 100 200 300 400 500
Dis tance (miles)
Slope : 0.000134tvalue : 1.91R2 : 0.05
Notes: Exports data for states on the left panel is from the rep ort on the Origin of Movement of U.S. Exports by State issued by the Census Bureau .GDP data by state is from the Statistical Abstract of the United States, a lso issued by the Census Bureau . Data for the vertica l ax is on the rightpanel com es from the Census Exports from Manufacturing Establishments rep ort. A ll data is for 2008. The d istance m easure in the horizontal ax isis the baseline m easure describ ed in the text and the app endix . The fitted lines use states’ population as weights.
We consider whether this increase in export shares toward ports reflects industry composition.
In Figure 2 we plot the export/output ratios at the national level for the 85manufacturing industries
in the 4-digit North American Industry Classification System (NAICS) against each industry’s
7
distance to international gates. We define industry distance as the employment-weighted average
of county distances: di =∑
c (eic/ei) dc, where dc is county c’s distance to its nearest port, eic is
industry i’s employment in county c, and ei is total employment in industry i. Therefore, the larger
is di, the farther away industry i locates from ports. The figure shows that, on average, industries
with higher export/output ratios at the national level are situated closer to ports.7
Figure 2: Export Intensity and Average Distance to Ports Across U.S. Industries
0.1
5.3
.45
.6.7
5
Indu
stry
Exp
ort/O
utpu
t
50 100 150 200
Distance (miles)
Slope : 0.0025tvalue : 3.20R2 : 0.16
Notes: The vertica l ax is uses industry export data from the USITC together w ith output data from the NBER-CES database. There are 85manufacturing industries at four-d ig it NAICS classification . A ll data is averages over 1989-2000. Table 5 in the data app endix rep orts theindustries and their exp ort/output ratios. See the text for the definition of the d istance m easure in the horizontal ax is. The fitted line usesindustries’ employment as weights.
For a more systematic analysis, we estimate the following specification with county-level data:
ln(eic) = ψc + ηi + θ ∗ ln(distc)× TradeBalancei + εic, (1)
where eic again is employment of industry i at county c, distc is the measure of market access by
county, and TradeBalancei is the export orientation of industry i which equals one if the trade
balance of industry i is positive and zero otherwise. If θ < 0, it means that industries with positive
trade balance are more likely to locate closer to ports. We also include county fixed effects ψc which
account for forces such as amenities which drive workers in all industries into specific locations, and
industry fixed effects ηi which account for national labor demand by industry.
Employment and U.S. trade data are both available in the same NAICS classification system
at the county level from the Bureau of Labor Statistics (BLS) and the USITC, respectively. The
Quarterly Census on Employment and Wages (QCEW), published by the Bureau of Labor Sta-
tistics (BLS) reports county-level employment at various levels of industry aggregation. We use
employment in 85 manufacturing industries (4-digit NAICS) in 3108 counties for the year 2003.
7See Table 5 in the data appendix for a list of these industries with their distances and export/output ratios.
8
Table 2: Distance to Ports and Industry Employment Across Counties
Dependent variable: ln(emp)I II III IV
ln(dist)× TradeBalance -0.0164∗∗ -0.0130∗ -0.0614∗∗∗ -0.0540∗∗∗
(0.00785) (0.00770) (0.00761) (0.00736)Distance measure Baseline Excl. airports Baseline Excl. airportsUndisclosed observations Imputed Imputed Dropped DroppedIndustry fixed effects Yes Yes Yes YesCounty fixed effects Yes Yes Yes YesN 259590 259590 207498 207498Adjusted R2 0.419 0.419 0.564 0.564
The trade orientation variable TradeBalancei equals one if average industry exports exceed aver-
age imports over 1997 − 2000. Out of the 85 industries, 27 are defined as net exporters over this
time period.
In using county-level employment data we face some data disclosure limitations. The QCEW
suppresses certain district-industry cells to protect the identities of a few large employers in the
area. These undisclosed cells can be distinguished from true zeros in each industry-county. In the
regression, we either drop them as missing observations or we fill in a uniform number imputed from
the difference between total employment in our data and aggregate manufacturing employment at
the national level. We use 2003 employment levels due to the relatively high coverage compared
to other years for which data is available. In that year, disclosed cells make up 60% of U.S.
manufacturing employment.
Table 2 reports the results. In the first two columns we use the data with imputed values
for nondisclosed observations, while in the last two columns we drop these observations. We also
consider specifications where we exclude the counties that only contain airports from the list of
counties that contain a port, and we compute the distance measure accordingly. Each specification
yields a significantly negative slope for the interaction term. To get a sense of magnitudes, moving
toward the nearest port by 150 miles increases employment in export oriented industries by between
3.6% and 11.8% relative to import competing industries.
The positive correlation between industry proximity to ports and export status suggests that
domestic trade costs and international comparative advantages may play a role in the location of
industries. Next, we present a theory where international comparative advantages lead to concen-
tration of employment in locations with good market access through regional specialization.
3 Model
Geography and Trade Costs The country consists of a set of locations arbitrarily arranged on
a map. We index locations by `, and we assume that only some locations can trade directly with
the rest of the world. Goods must cross through a port to be shipped internationally. As it will
9
be clear below, given the nature of our model only the distance separating each location ` from its
nearest port matters for the equilibrium. Therefore, we assume without loss of generality that `
represents the distance separating each location from its nearest port, and we denote all ports by
` = 0. We let ` be the maximum distance between a location inside the country and its nearest
port.
There are two industries, i ∈ X,M.8 International and domestic trade costs are industry
specific. The international iceberg cost in industry i between ` = 0 and the rest of the world (RoW)
is eτi0 . Within the country, iceberg trade costs are constant per unit of distance. Therefore, the
cost of shipping a good for distance d in industry i equals eτi1d. This implies a cost of international
trade equal to eτi0+τ
i1` in industry i from location `.
Given this geography, we can interpret each location ` = 0 as a seaport, airport, or international
land crossing. What is key is that not all locations have the same technology for trading with the
RoW. This will drive concentration near points with goods access. Internal geography vanishes
when τ i1 = 0 for both industries.
Endowments There are two factors of production, a perfectly mobile factor and a fixed factor.
We refer to the mobile factor as workers, and to the fixed factor as land. We choose units such
that the national land endowment equals 1, and we let λ (`) be the total amount of land available
in locations at distance ` from their nearest port.9
Land is owned by immobile landlords who do not work and who spend their rental income
locally. Locations at distance ` are also endowed with a level of public amenities m (`).
We let N be the measure of mobile factors in the country. We refer to this factor as labor, so
that N is the labor to land ratio at the national level. We let n (`) denote employment density in
`, which is to be determined in equilibrium.
Preferences Workers and landlords consume in the same location as they live. Utility of an
agent who lives in ` is proportional to
m (`)CT (`)βT CN (`)1−βT ,
where CT (`) is a consumption basket of tradables that includes X and M , CN (`) is consumption
of non-tradables, and m (`) is the level of amenities at `. Indirect utility of a worker who lives in `
therefore is
u(`) = m (`)w (`)
E(`), (2)
8The analysis can as well accomodate a continuum of industries as in Dornbusch, Fischer and Samuelson (1977).Our results would carry on in that case.
9 If the distribution of land is uniform, λ (`) represents just the measure of locations at distance ` from theirnearest port.
10
where w(`) is the wage at `, and E(`) is the cost of living index which includes the price index for
tradables ET (`) as well as the price of non-tradables PN (`),
E (`) = ET (`)βT PN (`)1−βT . (3)
The price index of tradables ET (`) ≡ ET (PX(`), PM (`)) is defined over the prices in sectors X and
M . We let p(`) ≡ PX(`)/PM (`) be the relative price of X in `. Since preferences are homothetic,
there exists an increasing and concave function e(p(`)) that depends on the relative price of X such
that
ET (`) = PM (`)e (p (`)) .
For owners of fixed factors, income equals rents r(`) per unit of land and utility is therefore
increasing in m (`) r (`) /E(`). Landowners are immobile, but workers decide where to live.
To characterize the equilibrium we will need to use consumption of non-tradables in each loca-
tion. Demand for non-tradables per unit of land in location ` is
cN (`) = (1− βT )y (`)
PN (`), (4)
where y (`) = w (`)n (`) + r (`) is income generated by each unit of land at `.
Technology Production in each sector requires one unit of land to operate a technology with
decreasing returns to scale in labor. We let ni(`) be employment per unit of land in industries
X,M or in the nontradable sector N at location `. Profits per unit of land in sector i = X,M,N
at ` are
πi (`) = maxni(`)
Pi(`)qi (ni(`))− w(`)ni(`)− r (`) . (5)
The production technology is
qi (ni(`)) = κini(`)
1−αi
ai (`), (6)
where ai (`) is the unit cost of production in industry i in sector ` and κi ≡ α−αii (1− αi)−(1−αi)
is just a normalization constant that helps to save notation. Decreasing returns to scale 1 − αimeasure the labor intensity in sector i, acting as congestion force. From (6) it follows that the
aggregate production function in sector i at ` is Cobb-Douglas with land intensity αi.10
To simplify the exposition, we assume that land intensity is the same across tradeable sectors,
αX = αM ≡ αT ,
but may differ between the tradeable and non-tradeable sectors, αN 6= αT . We discuss implications
of differences in land intensity across tradeable sectors below.11 For future use, we define the10 It would have been equivalent to include consumption of two non-tradable sectors, housing (which only uses land)
and services (which only uses labor), with expenditure share αN on land. Our current notation is more compact.11See section (3.4.2). For our quantitative analysis, where we interpret the two tradeable sectors as export-oriented
11
consumption-weighted average of sectorial land shares,
α = βTαT + (1− βT )αN .
In the nontradable sector, unit costs aN (`) may vary across locations. In the tradeable sec-
tor, industry-specific production costs [aM (`) , aX (`)] may vary across locations subject to the
restriction that the relative cost of production is constant across the country,
aX (`)
aM (`)= a for all ` ∈
[0, `]. (7)
Therefore, while some locations might be more productive than others in every industry or
hold comparative advantages in non-tradables, comparative advantages in tradables are defined
at the national level.12 In turn, a differs across countries, creating incentives for international
trade. In this way, we retain the basic structure of a Ricardian model of trade, where countries are
differentiated by their comparative advantages.
The solution to the firm’s problem yields labor demand per unit of land used by each sector i
in location `,
ni(`) =1− αiαi
(Pi(`)
ai (`)w(`)
)1/αifor i = X,M,N . (8)
Finally, we let λi(`) be the total amount of land used by sector i = X,M,N at `.
3.1 Local Equilibrium
We first define and characterize a local equilibrium at each location ` that takes prices PX (`) , PM (`)and the real wage u∗ as given.
Definition 1 A local equilibrium at ` consists of population density n (`), labor demands ni (`)i=X,M,N ,
patterns of land use λi (`)i=X,M,N , non-tradeable consumption and price cN (`) , PN (`), andfactor prices w (`) , r(`) such that
1. workers maximize utility,
u(`) ≤ u∗, = if n (`) > 0, (9)
with demand of non-tradables cN (`) given by (4);
2. profits are maximized,
πi (`) ≤ 0, = if λi (`) > 0, for i = X,M,N, (10)
where πi (`) is given by (5);
and import-competing industries in the U.S., setting αX = αM is an appropriate assumption. See Section 4.12When we introduce externalities, the levels of [aM (`) , aX (`)] will depend endogenously on population density
in each location. See section (3.4.1).
12
3. land and labor markets clear, ∑i=X,M,N
λi(`) = λ(`), (11)
∑i=X,M,N
λi(`)
λ(`)ni(`) = n (`) ; (12)
4. the non-tradeable market clears,
cN (`) = qN (nN (`)) ; and (13)
5. trade is balanced.
Conditions 2 to 5 constitute a small Ricardian economy extended with a non-tradeable sector.
In addition, in each local economy ` the employment density n (`) is determined by (9).
We let pA be the autarky price in each location. By this, we mean the price prevailing in the
absence of trade with any other location or with the rest of the world, but when labor mobility is
allowed across locations. We first note that, as in a standard Ricardian model, the autarky price
pA is the same and equal to a in all locations. Using (10), location ` must be fully specialized in
X when p (`) > pA, and fully specialized in M when p (`) < pA. Since each location takes relative
prices as given, a location that trades with either the rest of the world or with other locations is
(generically) fully specialized. Only if p (`) happens to coincide with pA a trading location may
be incompletely specialized. This logic also implies that an incompletely specialized location is
(generically) in autarky.
To solve for the wage w(`) we note that whenever a location is populated, the local labor
supply decision (9) must be binding. Together with (8) and the clearing conditions, this gives the
equilibrium population density in each location,13
n (`) =
1−αα
[zX(`)u∗
(p(`)
eT (p(`))
)βT ]1/αif p (`) ≥ a,
1−αα
[zM (`)u∗
(1
eT (p(`))
)βT ]1/αif p (`) < a.
(14)
where zi (`) denotes the fundamentals of each location,
zi (`) ≡ m (`)
aN (`)1−βT ai (`)βTfor i = X,M . (15)
Expression (14) conveys the various forces that determine the location decision of workers.
Agents care about the effect of prices on both income and cost of living. Since preference are
homothetic, agents employed in the industry-location pair (i, `) necessarily enjoy a higher real13 In Appendix A we present the solution for the local equilibrium for the more general case with differences in
factor intensities across sectors. Equation (14) follows from that derivation.
13
income when the local relative price of industry X is higher in location `. That is, the positive
income effect from a higher relative price offsets cost-of-living effects. At the same time there are
congestion forces, so that mobile workers avoid places with high employment density. Congestion
depend on the intensity of land use in non-tradables αN and in tradables αT . The larger the
congestion, the smaller the population density. In the quantitative exercise, we will measure these
parameters to assess the effects of trade on concentration of economic activity. Naturally, agents
also prefer locations with better fundamentals zi (`).14
Trade affects density through the effect on the relative price p (`). When p (`) 6= pA, locations
are fully specialized and necessarily export. In this circumstance n(`) increases with the relative
price of the exported good. Also, regardless of whether a location trades or stays in autarky, keeping
relative prices constant an increase in the national real wage u∗ causes workers to emigrate from `.
Using these effects, we characterize the general equilibrium below. We summarize the properties of
the local equilibrium as follows.
Proposition 1 (Local Equilibrium) There is a unique local equilibrium where location ` is fully
specialized in X when p (`) > pA and fully specialized in M when p (`) < pA. Population density is
given by (14), so that it increases with the relative price of the exported good, and it decreases with
the real wage u∗.
3.2 General Equilibrium
We have characterized the local equilibrium independently from a location’s geographic position.
We move on to study how market access matters for the employment density and the specialization
pattern in general equilibrium. We study a small economy that takes international prices P ∗X , P ∗Mas given, and we let
p∗ =P ∗XP ∗M
be the relative price at RoW. We assume that every port ` = 0 faces the same international price
p∗. We also define the average international and domestic iceberg cost across sectors,
τ j ≡1
2
∑i=X,M
τ ij for j = 0, 1.
No arbitrage implies that for any pair of locations ` and `′ separated by distance δ ≥ 0, relative
prices in industry i satisfy
Pi(`′)/Pi(`) ≤ eτ
i1δ for i = X,M . (16)
This condition binds if goods in industry i are shipped from ` to `′. A similar condition holds with
respect to RoW. Since all locations ` = 0 can trade directly with RoW and face the same world
14 In the last quantitative exercise of Section 4 we treat zi (`) as a residual to match the empirical distribution ofemployment.
14
relative prices, (16) implies
e−2τ0 ≤ p(0)/p∗ ≤ e2τ0 . (17)
The first inequality is binding if the country exports X to RoW, while the second is if it imports
X. Therefore, for any location ` we have
e−2τ1` ≤ p(`)/p(0) ≤ e2τ1`, (18)
where the first inequality binds if ` exports X to RoW, and second does if ` imports X.
We are ready to define the general equilibrium of the economy.15
Definition 2 (General Equilibrium) An equilibrium in a small economy given international
prices P ∗X , P ∗M consists of a real wage u∗, local outcomes n (`) , ni (`)i=X,M,N , λi (`)i=X,M,N ,
cN (`) , w (`) , r(`) and goods prices Pi(`)i=X,M such that
1. given Pi(`)i=X,M and u∗, local outcomes are a local equilibrium by Definition 1 for all
` ∈[0, `];
2. relative prices p(`) satisfy the no-arbitrage conditions (17) and (18) for all `, `′ ∈[0, `]; and
3. the real wage u∗ adjusts such that the national labor market clears,
∫ `
0n (`)λ (`) d` = N. (19)
To characterize the regional patterns of specialization, we first note that the no-arbitrage con-
ditions rule out bilateral trade flows between any pair of locations within the country. Intuitively,
since all locations share the same relative unit costs, there are no gains from trade within the
country. To see why, suppose that there is bilateral trade between locations `X , `M at distance
δ > 0. If `X is the X-exporting location of the pair, part (i) from Proposition 1 implies that
p (`M ) ≤ pA ≤ p (`X). But, at the same time, the no-arbitrage condition (16) implies that the
relative price of X is strictly higher in `M , which is a contradiction. This implies that the country
is in international autarky if and only if all locations are in autarky and incompletely specialized.
This result is a spatial impossibility theorem, in the tradition of Starrett (1978).
With this in mind, we are ready to characterize the general equilibrium. We can partition the
country into the set of locations that trades with RoW and those that stay in autarky. It follows
that if the country is not in international autarky there must be some boundary b ∈[0, `]such
that all locations ` < b are fully specialized in the export industry. In turn, all locations ` > b
do not trade with the RoW and stay in autarky. All locations at distance b from the nearest gate
15Since Definition 1 of a local equilibrium includes trade balance for each location, trade must also balance at thenational level.
15
are indifferent between trading or not with RoW.16 Therefore, an internal boundary b divides the
country between a trading "coastal region" comprising all locations ` ∈ [0, b) whose distance to the
nearest international gate is less than b and an autarkic "interior region" comprising the remaining
locations ` ∈ (b, `].
Thus, a key feature of the model is that the distance separating each location from the nearest
international gate ` = 0 is the only local fundamental that matters for specialization. This justifies
our initial statement that locations may be arbitrarily arranged on a map, as well as our decision
to index locations by their distance to the nearest port. In addition, every bilateral trade flow in
the country either originates from ports, or is directed toward them.17 These features allow us
to represent a two-dimensional geography on the line, leading to closed form characterizations for
aggregate equilibrium outcomes and making the model tractable for counterfactuals.18
Since all locations ` ∈ (b, `] are in autarky, they are incompletely specialized and their relative
price is pA = a. Given this price in the autarkic region and the regional pattern of production,
the no-arbitrage conditions (17) and (18) give the price distribution depending on the position of
b. Using (20) and (22) we have
p(`) =
p∗e−2(τ0+τ1min[`,b]) if the country is net exporter of X,p∗e2(τ0+τ1min[`,b]) if the country is net exporter of M.(20)
From now on, we assume that the economy is net exporter of X and net importer of M . Below,
we provide conditions such that this is the case. Using (14) in the aggregate labor-market clearing
condition (19) we can solve for the real wage,
u∗ =
[1− αα
1
N
∫ `
0λ (`) z (`)1/α
(p(`)
eT (p(`))
)βT /αd`
]α. (21)
Since the relative price function in (20) depends on b, so does the real wage. To find the location
of the boundary b we use the continuity of the relative price function:
p(b) ≥ pA, = if b < `. (22)
16To determine which locations belong in each set, we note that if the country is exporter of X then all locationsthat trade with RoW must also export X. Therefore, all locations ` such that e−2(τ0+τ1`)p∗ < pA must stay inautarky, for if they specialized in X then the relative price of X would be so low that it would induce specialization inM . In the same way, all locations ` such that e−2(τ0+τ1`)p∗ > pA must specialize in X, for if they stayed in autarkythen the relative price of M would be so high that it would induce domestic consumers to import from abroad,violating the no-arbitrage condition (18).
17 In the model, international shipments depart from coastal locations near international gates, while interiorlocations only ship locally. This is broadly consistent with the view in Hilberry and Hummels (2008) that shipmentsin the U.S. are highly localized. In the calibration, we target the domestic shipping costs of export-bound shipmentsin the U.S. to measure τ1.
18Naturally, bilateral trade flows unrelated to foreign trade would arise if we allowed for product differentiationwithin industries or for differences in comparative advantages across locations. With these features, representing thetwo-dimensional geography on the line would be generically unfeasible.
16
When p(`) > pA then b = `, so that the interior region does not exist. The general equilibrium
is fully characterized by the pair u∗, b that solves (21) and (22). All other variables follow fromthese two outcomes.
Figure 3: Relative Prices and Population Density over Distance
Figure 3 illustrates the structure of the equilibrium when the economy exports good X but
is not fully specialized. On the horizontal axis, locations are ordered by their distance to their
nearest port. As we have already established, this is the only geographic aspect that determines
the local equilibrium. In the left panel, the relative price of the exported good shrinks away from
the port until it hits the autarky relative price, and remains constant afterward. The economy is
fully specialized in X in the coastal region, but incompletely specialized in the interior. Only the
coastal locations ` ∈ [0, b] are commercially integrated with RoW.
In the right panel, we plot population density assuming that the fundamentals z (`) defined
in (15) are constant across locations, so that international trade is the only force that shapes the
distribution of population density. Since the relative price of the export industry decreases away
from international gates, so does population density in the coastal region until it reaches the interior
region.
We summarize our findings so far as follows.
Proposition 2 (Population and Industry Location in General Equilibrium) There is a
unique small-country equilibrium, where: (i) if the country trades internationally but is not fully
specialized in what it exports, there exists an interior region[b, `]that is incompletely specialized
and a coastal region [0, b) that trades with RoW and specializes in the export-oriented industry; and
(ii) if z (`) is constant, the distribution of population is uniform under international autarky, but
it increases toward international gates if the country trades.
These results demonstrate that international trade drives concentration of economic activity
and industry location. When the economy trades but is not fully specialized, two discrete regions
17
emerge: a coastal region surrounding international gates that is densely populated, connected to
international markets and specialized in the export-oriented industry; and an interior region that
is sparsely populated, disconnected from the rest of the world and incompletely specialized.
In our reasoning so far we have assumed a given trade pattern at the national level. We establish
the conditions on the parameters that determine the national trade pattern and existence of the
interior region.
Proposition 3 (National Trade Pattern and Existence of Interior Region) (i) The coun-try exports X if pA/p∗ < e−2τ0; in that case, the interior region exists if and only if e−2(τ0+τ1`) <
pA/p∗; (ii) the country exports M if e2τ0 < pA/p
∗; in that case, the interior region exists if and
only if pA/p∗ < e2(τ0+τ1`); and (iii) the country is in international autarky if e−2τ0 < pA/p∗ < e2τ0.
These results imply that domestic trade costsτX1 , τ
M1
, while capable of affecting the gains
and the volume of international trade, are unable to impact the pattern or the existence of it. In
other words, the conditions that determine when international trade exists as well as the direction
of international trade flows are the same as in an environment without domestic geography. The
second implication is that, when then country trades, the interior region exists when trade costs
τ1, τ0 or the extension of land ` are suffi ciently large, or when comparative advantages, capturedby pA/p∗, are not suffi ciently strong.
3.3 Impact of Changes in International and Domestic Trade Costs
We use the model to characterize the impact of international and domestic trade costs on the
concentration of economic activity and the gains from trade. In the quantitative section we measure
the importance of these effects.
3.3.1 Trade Costs and Internal Migrations
Our motivating examples from the introduction show that international trade integration is
associated with shifts in economic concentration and industry location. In our model, population
density varies across locations based on proximity to international gates, and population density
in the coastal region relative to the interior region is endogenous. We summarize the impact of a
discrete change in trade costs on these outcomes as follows.
Proposition 4 (Internal Migration) A reduction in international or in domestic trade costs
moves the boundary inland to b′ > b and causes migration from region[c, `]into region [0, c] for
some c ∈ (b, b′). A lower τ0 causes population at the port n (0) to increase, but a lower τ1 causes
n (0) to decrease.
The direct impact of a reduction in trade costs is that the relative price of the exported industry
increases in the coastal region. In the case of a reduction in τ0, the shift is uniform across locations,
while a lower τ1 results in a flattening of the slope of relative prices toward the interior. In both
18
cases, the change in prices causes the relative price at b to be larger than the autarky price pA,
so that locations at the boundary now find it profitable to specialize in export industries and the
boundary moves inland.
What are the internal migration patterns associated with these reductions in trade costs? As
we show below, a consequence of lower trade costs is an increase in the real wage u∗. Since in the
interior relative prices remain constant, this causes labor demand to shrink. As a result, workers
migrate away from locations that remain autarkic toward the coast and relative population density
increases in the coastal region.
Figure 4: Effects of Changes in Domestic and International Trade Costs
Figure 4 illustrates the effects. The solid black line reproduces the initial equilibrium from
Figure 3. The solid red line represents a new equilibrium with lower international trade costs. The
price function shifts upward and the intercept increases from p (0) to p1 (0), increasing population
density at the port to n1 (0). Locations in [b, b′] start trading, but the newly specialized locations
[c, b′] loose population. The dashed blue line shows the effect of a reduction in domestic trade
costs. Prices at the port stay constant, but the slope flattens. As a result, population density
at the port shrinks to n2 (0). In relative terms, locations at intermediate distance become more
attractive when domestic trade costs decline. In both cases, population density is higher in [0, c] in
the new equilibrium.
These results reproduce the cases that we highlight in the introductory paragraphs of the pa-
per: as trade costs decline, employment migrates to coastal areas that host comparative-advantage
industries. The empirical literature also highlights the importance of proximity to ports for the
level of economic activity. In our model, statements about the distribution of employment density
apply as well to the distribution of real income per unit of land across locations, since both are
proportional. In the quantitative section, we use the model to measure the contribution of interna-
tional trade to the excess concentration of employment in coastal relative to interior areas in the
U.S.
19
3.3.2 Internal Geography and the Gains form Trade
We move to the impact of domestic trade costs τ1 on the gains from international trade. We
study the impact of trade costs on the real wage u∗, but we note that average real returns to land
as well as national real income are proportional to the real wage u∗.19 Therefore the impact of
trade costs on the average real returns to fixed factors is the same as the impact on mobile factors.
We can consider two extreme cases. As τ1 → ∞, domestic trade becomes prohibitive so thatb → 0 and the country approaches international autarky. In that case, all locations face the same
relative price p (`) = pA. We let ua be the real wage in that circumstance. In the other extreme,
when τ1 = 0 then b = ` and all locations face the relative price p (`) = p (0). In that case, the real
wage is
u =
(1− αα
Z
N
)α( p(0)
eT (p(0))
)βT,
where
Z =
∫ `
0λ (`) z (`)1/α d`
measures the distribution of land land and total productivity across locations. As in a standard
Ricardian model, the real wage is increasing in the terms of trade. Here, it also depend on the
share of tradables in total expenditures, while congestion causes the real wage to decrease with the
national labor endowment.
Using the solution for the real wage u∗ from (21), the actual gains of moving from autarky to
trade can be decomposed as follows,
u∗
ua= Ω (b; τ1) ∗
u
ua, (23)
where we define the potential gains of moving from autarky to international trade as
u
ua=
(p (0) /eT (p (0))
pA/eT (pA)
)βT,
and where the effects of domestic frictions is
Ω (b; τ1) =
∫ `
0
λ (`) z (`)1/α
Z
[p (`) /eT (p (`))
p (0) /eT (p (0))
]βT /αd`
α. (24)
The actual gains from trade, u∗/ua, equal the potential gains from trade without domestic
19Adding up real returns to labor and land across locations gives the aggregate production function,
Y ≡∫ `
0
w (`)n (`) + r (`)
E (`)λ (`) d` =
1
α
∫ `
0
λ (`) z (`)1/α(
p (`)
eT (p (`))
)βT /αd`
α(N
1− α
)1−α,
which in turn implies that real income per worker is proportional to the real wage, Y/N = u∗/ (1− α). Following thesame steps, we find that the real returns to land deflated at local prices equal (α/1− α)Nu∗ and are also proporitionalto the real wage.
20
trade costs, u/ua, adjusted by Ω (τ1, b). This function captures the impact of internal geography
on the gains from trade. It is a weighted average of the losses caused by domestic trade costs in
each location. The weights across locations correspond to the importance of their fundamentals
and land endowments. In turn, the location-specific losses from domestic trade costs are captured
by the reduction in the terms of trade. The overall friction Ω (τ1, b) is strictly below 1 as long as
τ1 > 0, and it equals 1 if τ1 = 0. In the quantitative section we measure the magnitude of each
component in (23) to assess the importance of domestic trade costs for the gains from trade.
How do the gains from trade depend on domestic trade costs? Intuitively, the larger the size of
the export-oriented region, the more a country should benefit from trade. Since τ1 causes the export
oriented region to shrink, we should expect the gains from trade to decrease with domestic trade
costs. A lower τ1 makes exporting profitable for locations further away from the port, allowing
economic activity to spread out and mitigate the congestion forces in dense coastal areas.
To formalize this, we define the elasticity of the consumer price index,
ε(p) =deT (p)/eT (p)
dp/p,
and we also define the share of location ` in total employment,
s (`) =n (`)λ (`)
N.
Using these definitions, we have the following result for the change in welfare when there is a change
in the environment. We let x represent the proportional change in variable x when there is a change
in the environment.
Proposition 5 (Gains from International Trade) Consider a shock to p∗, τ0 or τ1. Then,the change in the real wage is
u∗ = βT
∫ b
0[1− ε(p(`))] s (`) p(`)d`. (25)
Therefore, the gains from trade are decreasing with domestic trade costs τ1,
d(u∗/ua)
dτ1< 0. (26)
Expression (25) describes the aggregate gains from a reduction in trade costs, either domestic or
international, as function of the relative price change faced by export-oriented locations weighted by
their population shares s (`). Reductions in domestic or international trade costs cause the relative
price of the exported good to increase. This has a positive effect on revenues and a negative effect
on the cost of living. The latter is captured by the price-index elasticity ε(p (`)), and mitigates the
total gains. In this context, (25) implies that the gains from an improvement in the terms of trade,
caused by either lower τ0 or larger p∗, are bounded above by the share of employment in export-
21
oriented regions. It also implies that domestic and international trade costs are complementary, in
that the gains from international trade are decreasing in domestic trade costs. Larger τ1 causes
relative export prices to decrease faster toward the interior, reducing the gains from trade.
3.3.3 Distributive Effects of Trade between Immobile Factors
The aggregate gains from trade hide distributional effects between immobile factors located in
different points of the country. The real returns to fixed factors at location ` are v (`) ≡ r (`) /E (`).
The change in real returns to immobile factors at ` when there are marginal changes in international
or domestic trade costs is
v (`) =βTα
[1− εT (p (`))] p (`)− 1− αα
u∗. (27)
The first term measures the impact of relative price changes through both revenues and cost of liv-
ing. The second part considers the economy wide increase in real wages, which captures emigration
of mobile factors.
Consider first the interior locations ` ∈ (b, `]. In these places, p (`) = 0 because distance
precludes terms of trade improvements, but mobile factors emigrate with trade reform because of
the real wage increase. As a result, immobile factors in the interior (b, `] region loose from lower
trade costs. However, some immobile factors located in the coastal areas ` ∈ [0, b) necessarily
gain. Hence, reductions in both international and domestic trade costs generate redistribution of
resources away from the interior to the coastal region.
However, outcomes for immobile resources in the coastal region are not uniformly positive.
While, on average, the coastal region necessarily gains from improvements in international or do-
mestic trade conditions, reductions in domestic trade costs τ1 necessarily hurt immobile factors
located at the port. In other words, coastal areas are better off if places further inside the country
have poorer access to world markets.
These results echo the well known distributional effects in specific factor models, where the
factors that are specific to goods whose relative price increases gain from trade. Here, the new
margin is that immobile factors experience differential relative price changes due to international
and domestic trade reform depending on their geographic position. We summarize these results as
follows.
Proposition 6 (Distributive Effects of Trade) Real income of immobile factors located in theinterior region (b, `] decreases with reductions in international or internal trade costs, while on
average the coastal locations [0, b) gain. There is some ε < b such that real income of immobile
factors in locations [0, ε) decreases with reductions in domestic trade costs.
22
3.4 Extensions
3.4.1 Economies of Scale
Urban economists emphasize that agglomeration forces play an important role for the concen-
tration of economic activity. Therefore, for quantitative purposes we include a simple form of
external economies of scale. In our quantitative exercises we set the strength of the externality to
match the estimates from the empirical literature. We introduce the externality in a tractable way
that does not affect the structure of the equilibrium. We model a general urbanization economy
whereby productivity in all sectors may be increasing on population density:
ai (`) = ain (`)−ζ for i = X,M,N . (28)
In more populated locations the level of productivity in all sectors is higher.20 So far, we allowed
ai (`) for i = X,M to vary across locations subject to (7). Here, aX (`) , aM (`) , aN (`) varyendogenously but are still consistent with (7).21
As before, in each local economy, there is a unique equilibrium with positive density. This
equilibrium is stable, in the sense that the real wage decreases with population density, as long
as congestion forces outweigh external effects, so that α > ζ. In this case, using (14), in the new
equilibrium with external effects we have that population density in a location that produces X is
n (`) =
[(1− αα
)α m (`) /u
a1−βTN a
βTX
(p (`)
eT (p (`))
)βT ]1/(α−ζ). (29)
Comparing (14) and (29) we see that, formally, the only effect of ζ is to reduce the congestion
caused by the use of land. The solution for the real wage (21) is modified accordingly. In the
quantitative section, we parametrize ζ following the estimates from the empirical literature and
and we measure α directly from data. Using the calibrate model, we investigate the importance of
international trade on the concentration of economic activity and effect of internal geography on
the gains from trade.
3.4.2 Sectorial Differences in Land Intensity
So far we have proceeded under the assumption that decreasing returns αi are the same across
tradeable sectors. When αX 6= αM , a few issues arise. First, each local economy becomes a
small Hecksher-Ohlin economy so that relative differences in factor abundance across locations
may motivate internal trade. Second, imposing (7) on relative productivities is no longer suffi cient
for our key feature that the autarky price pA is the same in all locations. This property allowed us
20We could also allow ζ to vary between tradeable and non-tradeable sectors without affecting the structure of theequilibrium implied by (7).
21This generic formualtion of the external effect can be derived from different microfoundations. For example, itis equivalent to assuming economies to scale at the firm level as in Krugman (1980) in a non-tradeable intermediateinput used with same intensity in both X and M . See Abdel-Rahman and Fujita (1990).
23
to represent the model on the line.
We establish the condition on the underlying distribution of fundamentals which guarantees
that the economy retains the same equilibrium structure that we presented so far for the more
general case with αX 6= αM . For that, we note that the proper notion of fundamental comparative
advantages in each location is no longer aX (`) /aM (`), as in the basic Ricardian model, but rather
the following expression which involves fundamental productivities, amenities, and differences in
factor intensities across sectors:
a (`) ≡(aN (`)1−βT
m (`)
)αX−αM [aX (`)βTαM+αN (1−βT )
aM (`)βTαX+αN (1−βT )
].
Using this expression we have the following.
Proposition 7 When αM 6= αX , (i) pA (`) is independent from the distribution of land endow-
ments λ (`); and (ii) pA (`) = pA for all ` ∈[0, `]if and only if a (`) is constant for all `.
Part (i) states that the land endowment does not affect autarky prices. For this, labor mobility
is key. When αM 6= αX , conditions 2 to 5 of the local equilibrium represent a small Hecksher-Ohlin
economy where the autarky price in principle depends on factor endowments. However, (9) implies
that the labor density is higher in places with more land abundance. This offsets factor proportions
effects, turning each local economy into a small Ricardian economy. Hence, when amenities m (`)
and productivities ai (`) are constant across locations, an arbitrary distribution of land is consistent
with common autarky price pA (`) even when αM 6= αX .
Part (ii) characterizes the variation in m (`) , aN (`) , aM (`) , aX (`) across locations that isconsistent with a constant autarky price pA (`) when αX 6= αM . If this condition holds, we ensure
that our economy retains the coastal-interior structure that we have described, as well as the effect
of trade on population and price gradients.
4 Quantitative Analysis
We propose a quantitative methodology to measure the impact of market access and comparative
advantages on economic outcomes. First, we calibrate the parameters of the model to match features
of the U.S. data related to international trade and domestic shipping costs. Second, we compare
the model prediction for the concentration of population with its empirical counterpart. Finally,
we compute the counter-factual gains from trade under alternative levels of domestic trade costs.
4.1 Calibration Strategy
For the calibration, we assume that tradable goods are consumed with constant elasticity of
substitution σ and a share parameter γX , which implies the following price index:
ET (`) =[γXPX (`)1−σ + (1− γX)PM (`)1−σ
] 11−σ
.
24
The model lends itself to several normalizations. Since the distribution of land endowments
λ (`) does not affect the outcome for n (`), we take it as uniform, i.e. λ (`) = 1/`. By proper choice
of units, we can normalize to one the national labor to land ratio N and the maximum distance
`.22 We set aM = 1 and write aX = a. Since the relative price at the port p(0) = p∗e−τ0 cannot be
separately identified from the relative cost a, we can set p(0) = 1.
Some of the remaining parameters can be set equal to their empirical counterparts or to values
commonly used in the literature. Using 1997 U.S. input-output tables, Valentinyi and Herrendorf
(2008) impute the cost share of land in manufacturing as 0.03 with no difference in factor shares
between imports and exports. Therefore we set αT = αX = αM = 0.03. The land cost share in the
production of non-tradables is αN = 0.06 which is higher than its counterpart in tradables mainly
due to residential housing. The elasticity of substitution σ is set equal to 3, which is the mid-value
between the average and median elasticities reported by Broda and Weinstein (2006, table 4). The
expenditure share on tradables is calibrated as βT = 0.3 based on the share of goods in Personal
Consumption Expenditures of the U.S. National Income and Product Accounts from the Bureau
of Economic Analysis. In our exercises below we also allow for external economies of scale as in
section (3.4.1). To discipline the scale of external economies ς, we conservatively use the lower
bound of values reported by Rosenthal and Strange (2004) and set ς = 0.043.23
Table 3: Parameters for Calibration
Parameter Description Value Source/Target
(αX , αM ) Land intensity in tradables 0.03 Valentinyi and Herrendorf (2008)
αN Land intensity in non-tradables 0.06 Valentinyi and Herrendorf (2008)
βT Expenditure share of tradables 0.3 BEA National Income Accounts
σ Elasticity of substitution 3 Broda and Weinstein (2006)
ς Agglomeration economies 0.043 Rosenthal and Strange (2004)
a Relative cost of production 0.924 X/Q in net-export industries (15%)
γX Preference parameter 0.408 Expenditure share of net-export industries (45%)
τ1 Domestic shipping costs 0.938 Domestic shipping costs for exports (1.5%)
The remaining parameters (a, γX , τ1) are pinned down by solving the model and targeting
moments from the data. The parameter a captures the strength of comparative advantages. As
a standard target we include the export to output ratio in export-oriented industries of the U.S.
The average for this statistic over the years 1989-2000 is 15%. We pick the preference parameter
γX such that the model replicates the share of final expenditures in net-export industries equal to
45% over the same time period.
These data would be suffi cient to measure gains from trade in a model without internal geog-
22Given our calibration strategy, choosing a different value for ` would only scale the calibrated τ1 proportionallywithout affecting model outcomes such as the land and population shares of the coastal region.
23Reviewing the evidence on agglomeration economies, Rosenthal and Strange (2004) report that doubling densityincreases productivity by an amount that ranges between 3-8%. We calibrate ς by solving 2ς = 1.03. This satisfiesthe parameter restriction ζ < α necessary for stability.
25
Figure 5: Population Density over Distance
01
23
45
67
89
1011
Den
sity
(nor
mal
ized
)
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
Distance (normalized)
raphy. In our case, we also need to measure the domestic trade cost τ1. We choose τ1 so that the
model matches the share of domestic freight costs in the f.o.b value of export-bound shipments in
the U.S. equal to 1.5%.24 Details about the calibration of τ1 are reported in Section C.2.
Table 3 lists the parameters. The calibrated values for (a, γX , τ1) vary across the calibration
exercises that we perform. Reported values correspond to our baseline calibration below.
4.2 Density in Coastal Areas
It is well known that the U.S. population is heavily concentrated in coastal areas (Rappaport
and Sachs (2003)). Figure 5 shows a fitted spline for population density in the U.S. by county
distance. The horizontal axis shows the distance to the nearest port that we introduced in Section
2 relative to the maximum distance in the data, which equals 637 miles. The line is a median cubic
spline on county population densities normalized by the average U.S. population density, which
equals 94 inhabitants per square mile. This is the empirical counterpart to the model prediction in
Figure 3 (right panel), where we showed the theoretical distribution of population over distance.
This empirical density-distance profile is qualitatively consistent with the theory. To what
extent can the forces that we highlight account for this concentration quantitatively? To address
this question, we compare the model prediction for the higher concentration in the low distance
regions with the actual concentration that we see in the data.
First, we assess the model prediction for density at the port. We let nm0 be the model-generated
density at ` = 0 (n(0) in the theory). We denote its empirical counterpart by ne0. The empirical
density at the port includes the 35 counties where the major U.S. ports (excluding 3 major interior
24This implies that shipping goods between the most distant pair of locations within the country costs 1 −exp(−τ1) = 60% of the f.o.b. value in ad-valorem equivalent terms. This magnitude is comparable to the internaltrade costs estimated by Ramondo et al. (2011).
26
airports) are located. Population density in this set of counties is 956 inhabitants per square
mile, implying a (normalized) ne0 = 956/94 = 10.12. Second, we consider the average coastal
density. In the model, the coastal density is the average density for all locations ` ∈ [0, b]. We
let nmC =(∫ b0 n(`)d`
)/b be the model-predicted density in the coast. In parallel, we define coastal
counties in the data as those located within distance b of the nearest port, for the value of b that
we obtain from the calibrated model.
Based on these moments, the following metric quantifies the role of trade in explaining coastal
density:
role of trade =nm − 1
ne − 1. (30)
This statistic attains a minimum value of 0 when the model generates a uniform distribution of
population in autarky.
To highlight the effects of our mechanism, we proceed in stages and perform three exercises that
introduce non-tradable goods, external economies and fundamentals in piecemeal fashion. In each
case, we recalibrate (a, γX , τ1) to match the targets introduced above. Results are summarized in
Table 4.25
We start with a baseline calibration in which the non-tradable sector is shut down (βT = 1),
there are no external economies (ς = 0), and the fundamentals z (`) defined in (15) are constant
and normalized to 1. This case shows the extent in which a model featuring only tradable goods
can explain coastal concentration. In the baseline calibration, the role trade defined by (30) equals
13% for average coastal density and 36% for density at the port.
Table 4: Calibration Outcomes
Definition of Coastal Data Model Role of Trade
Baseline calibration35 counties with a port 10.12 4.3 36%
116 counties within distance b=0.042 10.4 2.2 13%
With scale economies35 counties with a port 10.12 4.94 43%
117 counties within distance b=0.043 10.4 2.39 15%
With fundamentals35 counties with a port 10.12 8.18 21%
46 counties within distance b=0.01 10.33 9.4 10%
In the second exercise – calibration with scale economies– we bring in complementary forces
that account for congestion and agglomeration. We set (βT , ς) equal to their calibrated values
25 In the model, density peaks at the port and decreases monotonically with `. The empirical density displayssome non-monotonicity around the port. In Table 4, normalized density in the 117 counties within distance b = 0.043is higher than the density of the 35 counties with a port. Two factors lie behind this discrepancy: first, countiesare unequally sized so that distance and land area are not perfectly correlated. Second, our distance variable is animperfect measure of market access.
27
so that consumption of non-tradables and external economies generate additional dispersion and
concentration, and we recalibrate the parameters (a, γX , τ1). The results are similar to those
obtained in the baseline calibration.26 In this case the role of trade is 15% and 43% of the average
coastal and port densities, respectively.
Finally, the third exercise– calibration with fundamentals– shuts down external effects but re-
places them with variation in fundamentals. We parametrize z (`) = e−τz`, so that fundamentals
in (15) may display systematic bias toward or away from ports. The slope of τ z is then chosen to
fit the empirical concentration of population, while (a, γX , τ1) are calibrated to the same moments
as before. We calibrate the model twice to match nmC = neC and nm0 = ne0, respectively.
27 Then, we
assess the role of trade through the reduction in concentration generated by taking the country to
international autarky. The counter-factual coastal concentration to be compared with neC is now
calculated using the calibrated b. Since τ z > 0, the model generates a positive density gradient
toward the port in the absence of international trade. Without trade, nmC (ne0) drops from its fitted
value of 10.33 (10.12) to 9.4 (8.18) implying that trade accounts for 1− [(nm − 1) / (ne − 1)] = 10%
(21%) of the empirical concentration.
To sum up, our quantitative exercise suggests that around 10 − 40% of the U.S. population
concentration in coastal regions can be accounted for by market access and comparative advan-
tage forces. As we have shown, the results are robust to adding empirically relevant congestion
and dispersion forces to the baseline calibration, as well as considering the effect of fundamental
differences across locations.
4.3 Effects of Domestic Trade Costs
We conclude with the effect of domestic frictions. For that, we vary τ1 around its calibrated
value while keeping all other parameters fixed. For each value of τ1, we compute the coastal density,
nmC , export share of X-sector output and the percent gains from trade u∗/ua − 1. The results are
shown in Figures 6 and 7. In each case, we highlight the calibrated value of τ1 as well as the level
at which the country is completely specialized. The results correspond to the baseline calibration
above.
As depicted in the left panel of Figure 6, coastal density is increasing in τ1. The market access
advantage of coastal locations looses its importance when domestic trade costs fall. Economic
activity and population spread out more evenly due to decreased demand for proximity to the
port. Concurrently, aggregate exports increase as locations further inside the country gain access
to the foreign market and start exporting (right panel of Figure 6). For low enough τ1, the country
specializes completely and population is distributed uniformly.
Finally, we return to expression (23) for the decomposition of actual gains from trade into
potential gains and domestic frictions. Figure 7 plots the percentage gains from trade (u∗/ua − 1)
26Calibrated values for (a, γX , τ1) are (0.922, 0.408, 0.938).27Calibrated values for (a, γX , τ1) are (0.935, 0.413, 3.125) and (0.938, 0.414, 2.5). The slope of z (`) fitted to the
empirical density is τz = 0.503 and τz = 0.419, respectively.
28
Figure 6: Effect of Domestic Trade Costs on Population Concentration and Exports
0.0396 0.5 0.938 1.5 21
1.5
2
2.2
2.5
τ1
Coa
stal
den
sity
0.0396 0.5 0.938 1.5 2
0.15
0.3
0.45
0.58
τ1
Exp
ort/
outp
ut r
atio
for
Xs
ecto
r
Calibrated valueComplete specialization
Figure 7: Domestic Trade Costs and Gains from Trade
0.0396 0.5 0.938 1.5 2
0.16
2.55
4.64
τ1
Gai
ns f
rom
tra
de in
pct
Calibrated valueComplete specialization
against τ1 and equation (31) below shows the measured values of each component.
u∗
ua︸︷︷︸≈0.16%
= Ω (τ1, b) ∗u
ua︸︷︷︸≈4.64%
(31)
In the baseline calibration the gains from trade are very small, but in the ballpark of studies from
the literature which, in other contexts, offer measurements of the gains from trade with final goods
only, such as Arkolakis et al. (2011). Naturally, in our exercise the actual comparative advantages
of the U.S., captured by the parameter a, are higher than what it would be implied by a model
without domestic trade costs. Therefore, potential gains from trade are higher than the actual
gains from trade, and the maximum welfare gain (u/ua − 1) that would be attained when τ1 = 0
equals 4.64%. Hence the quantitative model suggests that a large part of the gains from trade is
29
lost due to the cost of trade within the U.S.28
5 Conclusion
We developed and quantified a theory to characterize the interaction between international trade
and the geographic distribution of economic activity within countries. The framework combines
standard forces in international trade with an internal geography within countries. Locations within
countries differ in access to international markets, and congestion forces deter economic activity
from concentrating in a single point. The model reproduces salient facts on the interaction between
international trade and regional outcomes, and it implies that the gains from international trade
are larger when domestic trade costs are smaller.
While previous work in economic geography and international trade highlights the importance of
locational fundamentals and scale economies in explaining industry location and the concentration
of economic activity, we offer a novel mechanism through the interaction between comparative
advantages at the national level and domestic trade costs. To motivate the analysis, we use U.S.
data on specialization at the county level which shows that export-oriented industries are more
likely to locate in counties situated at lower distance from international gates such as seaports,
airports or land crossings.
In the model, international trade creates a partition between a commercially integrated coastal
region with high population density and an interior region where immobile factors are poorer.
Reductions in domestic or international trade costs generate migration to the coastal region and
net welfare losses for fixed factors in the interior region. Qualitatively, these stylized outcomes
are consistent with the spatial structure of economic activity in large developing countries such as
China or India in which coastal regions integrated with the rest of the world coexist with remote
and relatively isolated corners.
Finally, we calibrated the model to U.S. data to measure the importance of international trade
in concentrating economic activity in coastal areas, as well as the importance of domestic trade
costs in hampering the gains from trade. The calibrated model implies that 10 to 40% of the coastal
concentration in the U.S. can be accounted for by comparative advantage and market access forces.
It also suggests that domestic trade costs, despite being small, may dissipate a large fraction of the
gains from international trade.
28This result also holds in the extended calibration with non-tradables and external economies. While the magni-tude of the reduction in gains from trade due to domestic frictions (Ω) is robust to the inclusion of these additionalforces, the level of welfare gains is lower in the extended calibration because of the large share of non-tradables inconsumption.
30
References
[1] Abdel-Rahman, Hesham and Masahisa Fujita (1990), "Product Variety, Marshallian External-
ities, and City Sizes," Journal of Regional Science, 30:2 pp. 165-183.
[2] Allen, Treb and Costas Arkolakis (2012), "Trade and the Topography of the Spatial Economy,"
manuscript.
[3] Arkolakis, Costas, Arnaud Costinot, and Andrés Rodríguez-Clare (2011), "New Trade Models,
Same old Gains?", American Economic Review 102:1, pp. 94-130.
[4] Atkin, David, and Dave Donaldson (2012), "Who’s Getting Globalized? The Size and Nature
of Intranational Trade Costs," manuscript.
[5] Behrens, Kristian, Carl Gaigne, Gianmarco I. P. Ottaviano, and Jacques-Francoise Thisse
(2006), "Is Remoteness a Locational Disadvantage?," Journal of Economic Geography 6, pp.
347-368.
[6] Blonigen, Bruce A., and Anca D. Cristea (2012), "Airports and Urban Growth: Evidence from
a Quasi-Natural Policy Experiment," manuscript.
[7] Bond, Eric (1993), "Trade, factor mobility, and income distribution in a regional model with
compensating wage differential," Regional Science and Urban Economics 23, pp. 67-84.
[8] Caselli, Francesco and Wilbur John Coleman II (2001), "The U.S. Structural Transformation
and Regional Convergence: A Reinterpretation," Journal of Political Economy 109:3, pp. 584-
616.
[9] Courant, Paul and Alan V. Deardorff (1993), "Amenities, Nontraded Goods, and the Trade of
Lumpy Countries," Journal of Urban Economics 34, pp. 299-317.
[10] Dornbusch, R., S. Fischer and P. A. Samuelson (1977), "Comparative Advantage, Trade,
and Payments in a Ricardian Model with a Continuum of Goods," The American Economic
Review, 67:5 pp. 823-839.
[11] Eaton, Jonathan and Samuel Kortum (2003), "Technology, Geography, and Trade," Econo-
metrica 70:5, pp. 1741-1775.
[12] Feenstra, Robert C., Robert E. Lipsey, Haiyan Deng, Alyson C. Ma, and Hengyong Mo (2004),
“World Trade Flows: 1962-2000,”NBER WP 11040.
[13] Glaeser, Edward and Janet Kollhase (2004), "Cities, Regions, and the Decline of Transport
Costs," Papers in Regional Science 83, pp. 197-228.
[14] Fujita and Mori (1996), "The role of ports in the making of major cities: Self-agglomeration
and hub-effect," Journal of Development Economics 49, pp. 93-120.
31
[15] Hanson, Gordon (1996), "Economic integration, intraindustry trade, and frontier regions,"
European Economic Review 40, pp. 941-949.
[16] Hanson, Gordon (1998), "North American Economic Integration and Industry Location," Ox-
ford Review of Economic Policy 14:2, pp. 30-44
[17] Helpman, Elhanan and Paul Krugman (1985), Market Structure and Foreign Trade, Cam-
bridge: MIT Press.
[18] Henderson, Vernon (1982), "Systems of Cities in Closed and Open Economies," Regional Sci-
ence and Urban Economics 12, pp. 325-350.
[19] Hilberry, Russell and David Hummels (2008), "Trade Responses to Geographic Frictions: A
Decomposition using Micro-Data," European Economic Review, vol. 52, pp. 527-550.
[20] Holmes, Thomas J. and John Stevens (2004), "Spatial distribution of economic activities in
North America", Handbook of Urban and Regional Economics, vol. 4, pp. 2797-2843.
[21] Jorgenson, Dale W. and Frank Gollop (1992), "Productivity Growth in U.S. Agriculture: A
Postwar Perspective," American Journal of Agricultural Economics 74:3, pp. 745-750.
[22] Krugman, Paul (1980), "Scale Economies, Product Differentiation, and the Pattern of Trade,"
American Economic Review, 70:5, pp. 950-959.
[23] Krugman, Paul (1991), "Increasing Returns and Economic Geography," Journal of Political
Economy, 99:3, pp. 483-499.
[24] Krugman, Paul and Raul Livas-Elizondo (1996), "Trade policy and the Third World metropo-
lis" Journal of Development Economics 49, pp. 137-150.
[25] Manova, Kalina (2008), "Credit Constraints, Heterogeneous Firms, and International Trade,"
NBER WP 14531.
[26] Matsuyama, Kiminori (1999), "Geography of the World Economy," Kellog CMS-EMS Discus-
sion Paper No. 1239.
[27] Nunn, Nathan (2007), "Relationship-Specificity, Incomplete Contracts, and the Pattern of
Trade," Quarterly Journal of Economics, 122, pp.569-600.
[28] Cuñat, Alejandro and Marc Melitz (2012), "Volatility, Labor Market Flexibility, and the Pat-
tern of Trade," Journal of the European Economic Association 10:2, pp. 225-254.
[29] McCaig, Brian and Nina Pavnick (2012), "Export markets and labor allocation in a poor
country," manuscript.
[30] Rauch, James (1991), "Comparative Advantage, Geographic Advantage and the Volume of
Trade," The Economic Journal 101, pp. 1230-1244.
32
[31] Ramondo, Natalia, Andres Rodriguez-Clare and Milagro Saborio-Rodriguez (2011), "Increas-
ing Returns and Economic Prosperity: How can Size not Matter?," manuscript.
[32] Rappaport, Jordan (2008), "Consumption amenities and city population density," Regional
Science and Urban Economics 38, pp. 533-552.
[33] Rappaport, Jordan and Jeffrey D. Sachs (2003),"The United States as a Coastal Nation,"
Journal of Economic Growth 8, pp. 5-46.
[34] Redding, Stephen (2012), "Goods Trade, Factor Mobility and Welfare," NBER WP 18008.
[35] Rossi-Hansberg, Esteban (2004), "A Spatial Theory of Trade," American Economic Review
95.
[36] Rosenthal, Stuart and William C. Strange (2004), "Evidence on the Nature of Agglomeration
Economies", Handbook of Urban and Regional Economics, vol. 4, pp. 2119-2171.
[37] Starrett, David (1978), "Market Allocations of Location Choice in a Model with Free Mobility,"
Journal of Economic Theory 17, pp. 21-37.
[38] Storeygard, Adam (2012), "Farther on down the road: transport costs, trade and urban growth
in sub-Saharan Africa," manuscript.
[39] U.S. Census Bureau (2006), “Exports from Manufacturing Establishments”.
[40] Valentinyi, Akos and Berthold Herrendorf (2008), "Measuring factor income shares at the
sectoral level," Review of Economic Dynamics 11, pp. 820-835.
[41] Venables, Anthony and Nuno Limao (2002), "Geographical disadvantage: a Heckscher—Ohlin—
von Thunen model of international specialization," Journal of International Economics 58.
[42] World Bank (2009),World Development Report: Reshaping Economic Geography, Washington,
DC: World Bank.
33
A Derivation of Local Equilibrium
We characterize the local equilibrium in the general case with αX 6= αM . The definition of the local
equilibrium is given in (1). First, we define ω (`) ≡ w (`) /r (`) as the wage to rental ratio at `, and we note
that the profit maximization condition (10) is equivalent to
Pi(`) ≤w(`)
ω(`)αiai(`), = if λi (`) > 0 (32)
for i = X,M,N . In the case of tradables, these are the standard zero-profit conditions that underlie the
Stolper-Samuelson theorem. Since non-tradables are necessary consumed we must have that this condition
binds for i = N in every populated location.
In every populated location the indifference condition (9) must be binding, u(`) = u∗ if n (`) > 0. Using
(2) for the definition of u(`) together with the price for nontradables PN (`) resulting from (32) we can solve
for the wage at ` that leaves workers indifferent about staying in that location,
w (`) = ET (`)
(u∗
m (`)
) 1βT[aN (`)
ω(`)αN
] 1−βTβT
(33)
Using (32) and (33) we can solve for the wage-rental ratio if the local economy produces i,
ωi (`) =
[u∗ai (`)
βT aN (`)1−βT
m (`)
(ET (`)
Pi (`)
)βT ] 1βT αi+(1−βT )αN
. (34)
From (32), if λi (`) > 0 then ωi (`) ≤ ωj (`) for i, j = X,M and i 6= j. Also, note that ωX (`) is increasing in
p (`) and ωM (`) is strictly decreasing in p (`), so that the wage-rental ratio increases with the relative price
of the exported good. Therefore, the local economy is fully specialized in X when p (`) > pA (`) and fully
specialized in M when p (`) < pA (`), where pA (`) is the unique value for which ωM (`) = ωX (`). Since the
local economy is incompletely specialized in autarky, pA (`) must be the autarky price.
To solve for population density n (`), we can use (8) and the production function (6) to express the labor
and land market clearing, respectively, as function of sectorial output and unit factor requirements,
∑i=X,M,N
αiai(`)λi (`)qi (`)
ω(`)αi−1= λ (`) ,
∑i=X,M,N
(1− αi) ai(`)λi (`)qi (`)
ω(`)αi= λ (`)n (`) .
Using market clearing for non-tradables (13), when a local economy is fully specialized in i we can solve this
system to obtain the solution for population density at `,
n (`) =
(1
βTαi + (1− βT )αN− 1
)1
ωi (`)for i = X,M . (35)
where ωi (`) is given in (34).
34
B Proofs
Proposition 1 Profit maximization implies that πi (`) = (Pi(`)/ai)1/(1−αi) w (`)
−αi/(1−αi). Therefore,
(10) implies that πX (`) ≥ πM (`)←→ p (`) ≤ a and πX (`) ≤ πM (`)←→ p (`) ≤ a. Since autarkic locationsmust consume, λi (`) > 0 for i = X,M and therefore (10) implies that πX (`) = πM (`)←→ p (`) = pA = a.
Setting αX = αM = αT and using (34) in (35) gives (14).
Proposition 2 For (i) we have that if the country is net exporter of X, all locations that trade with
RoW must produce X. Condition (18) then implies that e−2(τ0+τ1`)p∗ ≤ p(`). Therefore, all locations suchthat e−2(τ0+τ1`)p∗ > pA cannot be in autarky and must specialize in X. In turn, since the country is not
fully specialized, there must be autarkic locations for which e−2(τ0+τ1`)p∗ < pA. Therefore, there must exist
b < ` such e−2(τ0+τ1b)p∗ = pA. For (ii), we have from the main text that in international autarky there is
no trade between locations within the country, Therefore, λi (`) > 0 for i = X,M and p (`) = pA for all `.
This implies that n (`) = N and w (`) = w in all locations. If the country is net exporter of X then n (`) is
given by (14). Since p′ (`) < 0 and p/e (p) is decreasing with p, n′ (`) < 0. If the economy is net exporter of
M , then p′ (`) > 0 so that n′ (`) < 0.
Proposition 3 For (i) and (ii), if pA/p∗ < e−2τ0 but the country is in international autarky or exportsM
then the no-arbitrage conditions (17) and (18) are violated. In that case, equilibrium condition (22) implies
that b < ` ←→ pA/p∗ < e2(τ0+τ1`). Similar reasoning applies when the country exports M . For (iii), if
e−2τ0 < pA/p∗ < e2τ0 but the country exports X or M then the no-arbitrage conditions (17) and (18) are
violated.
Proposition 4 From (22), it follows that b is decreasing with τ1 or τ0. Therefore, b′ > b when either
τ1 or τ0 decreases. Consider the case when the economy is net exporter of X and let p (`)′ and n (`)
′ be
the relative price and density in the new equilibrium with τ ′0 < τ0 or τ ′1 < τ1. Note that∂n(`)′
∂` < 0 and∣∣∣∂n(`)′∂`1
n(`)′
∣∣∣ ≤ ∣∣∣∂n(`)∂`1n(`)
∣∣∣ for all ` ∈ [0, b). Therefore, n (b)′> n (b) for otherwise
∫ `0n (`)
′λ (`) d` < N .
Also note that p (`) = p (`)′
= a for all ` ∈ (b, `] and, as shown in Proposition 5, u∗ is higher in the new
equilibrium, Proposition 1 implies that n (b′)′< n (b′). Hence, we have therefore, there must be c ∈ [b, b′]
such that n (c)′
= n (b). We conclude that n (`)′ − n (`) > 0↔ ` < c, so that population density increases in
[0, c].
When τ0 shrinks, n (0)′> n (0) for otherwise
∫ `0n (`)
′λ (`) d` < N . In contrast, using Proposition 1 we
have that when τ1 shrinks then n (0)′< n (0) since u∗ is higher in the new equilibrium but p (0) = p (0)
′.
Proposition 5 From the labor market clearing condition (19) and the condition (22) for the determination
of b, when there is a change in prices but`, λ (`) , N
remain constant then
∫ `
0
n (`)s (`) d` = 0. (36)
Using (38) in (36) we have:
u∗ = βT
∫ `
0
[1− ε(p(`))] s (`) p(`)d`
35
An increase in the terms of trade implies p(`) = p for all ` ∈ [0, b]. Then, (25) gives
u∗
p= βT
∫ b
0
[1− ε(p (`))]s (`) d` <
∫ b
0
s (`) d`
so that the gains from trade are bounded above by employment in the coastal region. In turn, we have that
dτ i1 > 0 implies p(`) = −`dτ i1 < 0 for all ` ∈ [0, b] so that u∗/dτ1 < 0, implying (26).
Proposition 6 The real returns to immobile factors at ` is v (`) = r (`) /E (`). Using (35) and u(`) = u∗
we can write
v (`) =α
1− αn (`)u∗,
so that the real returns to land at the national level,∫ `0v (`)λ (`) d`, equal
v =α
1− αNu∗. (37)
In turn, using (14), the percent change in population density in each location when there is an arbitrary
change in relative prices or in the real wage is
n (`) =βTα
[1− εT (p (`))] p (`)− 1
αu∗. (38)
so that the percent change in the real return to immobile factors at ` is given by (27) in the text. In turn,
after a change in domestic or international trade costs the change in relative prices is
p (`) =
−2 (dτ0 + `dτ1)
0if
` < b
` > b.
Therefore, v (`) < 0 if ` > b. Since v > 0 from (37), there must be some v (`) > 0 for ` < b. In turn, when
dτ0 = 0 > dτ1 then p (0) = 0 but u∗, implying v (0) < 0. Since v (`) > 0 for ` < b, there is some ε such that
v (`) < 0 for all ` ∈ [0, ε).
Proposition 7 From Appendix A we have that the autarky price pA (`) corresponds to the unique value of
p (`) such that ωX (`) = ωY (`), where ωi (`) is defined in (34). Part (i) follows because ωi (`) is independent
from λ (`). Part (ii) follows from setting pA (`) = pA in ωX (`) = ωY (`) and noting that the holds for all `
if and only if a (`) is constant.
36
C Data Appendix
C.1 Data Sources and Details for Section 2
Distance measure: We start with the Schedule D list of U.S. ports downloaded from the Foreign
Trade Division of the U.S. Census Bureau (link). We match this list with data on trade by ports
obtained from the Census USA Trade Online (link). The match generates 365 ports. Matching
this further with ports’zipcode information obtained from Customs and Borders Protection Agency
retains 288 ports accounting for 99.88% of U.S. trade (link). Zipcodes are then matched to counties
using the crosswalk from the U.S. Department of Housing and Urban Development (link). County-
level data is downloaded from the Census Bureau which contains information on population, land
area and coordinates based on the census of 2000 (link). Data on states’ population centroids
based on the census of 2000 is downloaded from the Census Bureau (link). The distance measure
is the minimum great-circles distances between the coordinates of ports and counties/states. The
distribution of trade volumes is heavily skewed toward large ports. Among the 288 ports, there are
many small regional ones with tiny trade volumes. Including all of them in the distance measure
considerably reduces the variation in distance. Thus, we use the top 49 largest ports that account
for 90% of total U.S. trade. These ports are located in 38 different counties plotted in figure 8.
Figure 8: Top U.S. Ports by Export Share
Notes: The dots represent the locations of the top 49 U.S. trade gateways. Red dots are inland airports(Atlanta, Dallas and Salt Lake City) that are excluded in the alternative distance measure.
Industry trade and output data: U.S. exports and imports for manufacturing industries analyzed
in figure 2 and table 2 are downloaded from USITC data (link). There are 85 manufacturing
industries at four-digit NAICS classification listed in table 5. To calculate export/output ratios in
figure 2, we match industry exports with industry shipments from the NBER-CES manufacturing
industry database (Bartelsman, Becker and Gray, 2008) (link). Industry distances are calculated
from state-to-port distances weighted by the employment share of each state in overall industry
37
employment.
County-industry employment data: Employment at industry-county level is obtained from the
Quarterly Census on Employment and Wages (QCEW), published by the Bureau of Labor Statistics
(BLS), downloaded from the database (link). The QCEW suppresses certain country-industry
cells to protect the identities of a few large employers in the area. These undisclosed cells can be
distinguished from true zeros in each industry-county. At the four-digit NAICS level, disclosed cells
make up 60% of U.S. manufacturing employment for 2003. We fill in the missing cells by equally
distributing the gap between total industry employment at the national level and total employment
aggregated up from the county-industry data.
C.2 Calibration Details for Section 4
Three key parameters of the model, (a, γX , τ1), are calibrated to match the following three data
moments:
Export/output ratio of export-oriented industries: We use the same data as in figure 2 (see
previous subsection). The average export/output ratio weighted by industry output over the years
1989-2000 is 15%.
Share of final expenditures in net-export industries: final expenditures in industries where the
U.S. is a net exporter as share of total expenditures in manufacturing. Using U.S. manufacturing
exports, imports and shipments data at four-digit NAICS level, total consumption is C = Y +
IMP − EXP where Y is aggregate gross shipments, IMP is total imports and EXP is total
exports. We then calculate CX as the sum of consumption in all industries that have positive trade
balance. This yields CX/C = 0.454 for the period 1989-2000.
Share of domestic freight costs in the f.o.b value of export-bound shipments equal: The 1997
Commodity Flow Survey (CFS) reports the total f.o.b value for a representative sample of export
shipments. It also informs the number ton-miles for the domestic segment of these shipments. To
find the shipping cost associated with export-bound cargo, we use the cost of shipping per ton-mile
from U.S. Federal Highway Administration data. The cost of a ton-mile shipment varies between 5
to 15 cents in 1997 dollars depending on the size and category of the truck. In the CFS data, $339
billion worth of export shipments by trucks traveled 51 billion ton-miles. At $0.1 per ton-mile,
this makes shipping costs equal to 1.5% of the f.o.b shipment value. In the CFS sample more than
60% of the value of all export bound cargo is carried to ports by trucks, so that we consider this
measure as representative of domestic shipping costs. This way of computing domestic trade costs
as well as the magnitude that we find are similar to Glaeser and Kollhase (2004).
38
Table 5: List of NAICS Manufacturing Industries Used in Section 2
NAICS code Industry Trade Balance EXP/Q Distance3361 MOTOR VEHICLES 0 0.104 733117 SEAFOOD PRODUCTS PREPARED , CANNED AND PACKAGED 0 0.044 833342 COMMUNICATIONS EQUIPMENT 0 0.191 833254 PHARMACEUTICALS AND MEDICINES 0 0.129 843344 SEM ICONDUCTORS AND OTHER ELECTRONIC COMPONENTS 0 0.453 963311 IRON AND STEEL AND FERROALLOY 0 0.095 983366 SHIPS AND BOATS 1 0.083 1023345 NAVIGATIONAL, MEASURING , ELECTROMEDICAL, AND CONTROL INSTRUMENTS 1 0.277 1033343 AUDIO AND VIDEO EQUIPMENT 0 0.626 1043364 AEROSPACE PRODUCTS AND PARTS 1 0.432 1053314 NONFERROUS METAL (EXCEPT ALUMINUM) AND PROCESSING 0 0.386 1053341 COMPUTER EQUIPMENT 0 0.451 1053152 APPAREL 0 0.126 1063256 SOAPS, CLEANING COMPOUNDS, AND TOILET PREPARATIONS 1 0.095 1073346 MAGNETIC AND OPTICAL MEDIA 0 0.252 1073321 CROWNS, CLOSURES, SEALS AND OTHER PACKING ACCESSORIES 1 0.011 1093363 MOTOR VEHICLE PARTS 0 0.213 1103251 BASIC CHEMICALS 1 0.237 1113359 ELECTRICAL EQUIPMENT AND COMPONENTS, NESOI 1 0.240 1113312 STEEL PRODUCTS FROM PURCHASED STEEL 0 0.019 1133255 PAINTS, COATINGS, AND ADHESIVES 1 0.077 1133241 PETROLEUM AND COAL PRODUCTS 0 0.039 1153252 RESIN , SYNTHETIC RUBBER, & ARTIFIC IAL & SYNTHETIC FIBERS & FILIMENT 1 0.234 1173261 PLASTICS PRODUCTS 1 0.080 1173351 ELECTRIC LIGHTING EQUIPMENT 0 0.124 1173113 SUGAR AND CONFECTIONERY PRODUCTS 0 0.041 1183118 BAKERY AND TORTILLA PRODUCTS 0 0.013 1183391 MEDICAL EQUIPMENT AND SUPPLIES 1 0.184 1183379 FURNITURE RELATED PRODUCTS, NESOI 0 0.012 1193353 ELECTRICAL EQUIPMENT 0 0.227 1193122 TOBACCO PRODUCTS 1 0.092 1193352 HOUSEHOLD APPLIANCES AND MISCELLANEOUS MACHINES, NESO I 0 0.157 1193272 GLASS AND GLASS PRODUCTS 0 0.175 1203399 M ISCELLANEOUS MANUFACTURED COMMODITIES 0 0.229 1213365 RAILROAD ROLLING STOCK 0 0.136 1223333 COMMERCIAL AND SERVICE INDUSTRY MACHINERY 0 0.302 1223119 FOODS, NESOI 1 0.062 1223335 METALWORKING MACHINERY 0 0.192 1233336 ENGINES, TURBINES, AND POWER TRANSM ISSION EQUIPMENT 1 0.354 1243231 PRINTED MATTER AND RELATED PRODUCT, NESOI 1 0.049 1243325 HARDWARE 0 0.172 1263315 FOUNDRIES 0 0.016 1263332 INDUSTRIAL MACHINERY 0 0.280 1273326 SPRINGS AND W IRE PRODUCTS 0 0.093 1273114 FRUIT AND VEGETABLE PRESERVES AND SPECIALTY FOODS 0 0.057 1273327 BOLTS, NUTS, SCREWS, R IVETS, WASHERS AND OTHER TURNED PRODUCTS 0 0.035 1273329 OTHER FABRICATED METAL PRODUCTS 0 0.225 1293133 FIN ISHED AND COATED TEXTILE FABRICS 1 0.046 1293274 LIME AND GYPSUM PRODUCTS 0 0.013 1293132 FABRICS 0 0.186 1293259 OTHER CHEMICAL PRODUCTS AND PREPARATIONS 1 0.151 1303323 ARCHITECTURAL AND STRUCTURAL METALS 0 0.016 1303141 TEXTILE FURNISHINGS 0 0.061 1303313 ALUMINA AND ALUMINUM AND PROCESSING 0 0.133 1313339 OTHER GENERAL PURPOSE MACHINERY 1 0.338 1313322 CUTLERY AND HANDTOOLS 0 0.137 1313262 RUBBER PRODUCTS 0 0.141 1323222 CONVERTED PAPER PRODUCTS 1 0.063 1323162 FOOTWEAR 0 0.146 1333159 APPAREL ACCESSORIES 0 0.354 1343151 KNIT APPAREL 0 0.050 1343131 FIBERS, YARNS, AND THREADS 0 0.052 1353121 BEVERAGES 0 0.027 1363372 OFFICE FURNITURE (INCLUDING FIXTURES) 0 0.049 1373324 BO ILERS, TANKS, AND SHIPPING CONTAINERS 1 0.081 1373279 OTHER NONMETALLIC M INERAL PRODUCTS 0 0.098 1373334 VENTILATION, HEATING , AIR -CONDITIONING ETC EQUIPMENT 1 0.170 1373271 CLAY AND REFRACTORY PRODUCTS 0 0.136 1393221 PULP, PAPER, AND PAPERBOARD MILL PRODUCTS 0 0.127 1403161 LEATHER AND HIDE TANNING 0 0.406 1403169 OTHER LEATHER PRODUCTS 0 0.291 1433115 DAIRY PRODUCTS 0 0.018 1433369 TRANSPORTATION EQUIPMENT, NESOI 0 0.170 1443149 OTHER TEXTILE PRODUCTS 0 0.085 1473362 MOTOR VEHICLE BODIES AND TRAILERS 1 0.057 1543219 OTHER WOOD PRODUCTS 0 0.021 1573371 HOUSEHOLD AND INSTITUTIONAL FURNITURE AND KITCHEN CABINETS 0 0.036 1573331 AGRICULTURE AND CONSTRUCTION MACHINERY 1 0.351 1583273 CEMENT AND CONCRETE PRODUCTS 0 0.004 1583253 PESTIC IDES, FERTILIZERS AND OTHER AGRICULTURAL CHEMICALS 1 0.216 1613212 VENEER, PLYWOOD, AND ENGINEERED WOOD PRODUCTS 0 0.060 1673211 SAWMILL AND WOOD PRODUCTS 0 0.095 1703116 MEAT PRODUCTS AND MEAT PACKAGING PRODUCTS 1 0.079 1723112 GRAIN AND OILSEED MILLING PRODUCTS 1 0.124 1733111 ANIMAL FOODS 1 0.054 190
Notes: Trade balance equals one if exp orts exceed imports over 1997-2000, and zero otherw ise. (EXP/Q) is exp ort/output ratio .
39