Journal of International Economics 86 (2012) 133–140

Contents lists available at SciVerse ScienceDirect

Journal of International Economics

j ourna l homepage: www.e lsev ie r .com/ locate / j i e

The rise of vertical specialization trade☆

Benjamin Bridgman ⁎U.S. Department of Commerce, Bureau of Economic Analysis, Washington, DC 20230, United States

☆ I thank two anonymous referees and seminar partiBoard, Federal Reserve Bank of Kansas City, 2009 InternConference and the North American Summer MeetingBrian Moyer kindly provided a data concordance. Theare solely those of the author and not necessarily thoseAnalysis or the U.S. Department of Commerce.⁎ Tel.: +1 202 606 9991; fax: +1 202 606 5366.

E-mail address: Benjamin.Bridgman@bea.gov.1 Intermediate goods are those used as inputs to fu

input–output tables, they are goods that are shippedthan final demand.

0022-1996/$ – see front matter. Published by Elsevier Bdoi:10.1016/j.jinteco.2011.08.016

a b s t r a c t

a r t i c l e i n f o

Article history:Received 21 June 2010Received in revised form 23 August 2011Accepted 23 August 2011Available online 28 August 2011

JEL classification:F1

Keywords:Trade costsVertical specializationManufacturing trade

Manufacturing and vertical specialization (VS) trade, trade in goods that incorporate imported inputs, havegrown rapidly since the 1960s. I argue that declining trade costs are an important explanation for thesefacts. I present a three stage vertical specialization trade model, with raw materials, manufactured partsand final goods sectors. In the simulated model, falling trade costs explain much of the observed growth inoverall and VS trade. Manufacturing trade grows twice as fast as overall trade. Raw materials trade wasmore important in the 1960s when trade costs were high, since their production is more strongly linked toendowments than manufacturing. Therefore, materials will be traded even when trade costs are high.Trade costs have fallen more for manufactured goods over the last 40 years, leading to a rapid expansion ofmanufactured parts trade relative to materials.

cipants at the Federal Reserveational Industrial Organizations of the Econometric Society.views expressed in this paperof the U.S. Bureau of Economic

rther production. In terms ofto production sectors rather

2 The composition

.V.

Published by Elsevier B.V.

1. Introduction

Trade in manufactured goods has expanded rapidly in the last fiftyyears (Bergoeing et al., 2004). U.S. manufacturing export share of GDPgrew by 140% between 1960 and 2006. The share of manufacturingoutput that is exported quadrupled during that period. This fact ispuzzling given that manufacturing has not grown as a nominalshare of output. Early on, when manufacturing was a large part ofproduction, there was little trade in manufactured goods. Later,when manufacturing declined in importance, trade became dominat-ed by these goods.

At the same time, vertical specialization (VS) trade, trade in goodsincorporating imported inputs, has expanded rapidly (Feenstra, 1998;Hummels et al., 1998, 2001). VS trade share of U.S. exports grew from6% in 1972 to 12% in 1997 (Chen et al., 2005). VS trade growth is notdue to a large increase in the share of intermediate goods trade.1

Chen et al. (2005) find that share of trade accounted for by interme-diate goods has been nearly constant since 1972.

I argue that the rise of manufacturing and VS trade are related:Both are driven by falling costs of trading manufactured parts. Priorto the Kennedy Round, U.S. trade was dominated by raw materials.2

Tariffs were high on manufactured goods, including parts. Materialsfaced high freight costs, since they have a low value to weight ratio.However, they were still imported because the ability to producethem is strongly linked to endowments. Materials cannot reliably bereplaced domestically and were essential for production. Manufac-tured goods are easier to replace with domestic goods since theyare less dependent on endowments.

The Kennedy Round focussed on reducing manufacturing tariffsand was notable both for the size and coverage of these cuts. Sincethen, trade policy has gone from being biased against manufacturedgoods to being more neutral. Since manufactured goods are more re-sponsive to trade barriers, manufacturing trade has grown faster thanmaterials trade. The share of trade in intermediate goods has beenroughly constant, but intermediate goods trade is now dominatedby manufactured inputs.

This paper presents a tractable general equilibrium model withRicardian trade in intermediate goods. There are two countries withthree layers of production: Raw materials are inputs to intermediategoods, which in turn are inputs to final consumption goods. Allthree types of goods may be traded, but incur an iceberg transporta-tion cost and may face tariffs. I calibrate the model and run simula-tions using data on freight costs and tariffs.

The simulated model predicts nearly all of the empirical growth intrade and the change in trade composition from 1967 to 2002.

of intermediate goods trade is documented in detail below.

Fig. 2. Materials share of intermediate goods imports.

134 B. Bridgman / Journal of International Economics 86 (2012) 133–140

Manufacturing trade grows much faster than overall trade growth.While overall share of goods output that is traded more than doublesbetween 1967 and 2002 in the baseline simulation, manufacturingtrade share triples. VS trade also grows rapidly, more than doublingfrom 1972 to 1997. Beginning with the Kennedy Round, manufac-tured goods tariffs fell more than non-manufactured goods tariffs.Lower trade costs on manufactured parts led to VS trade growth.

While VS trade grows rapidly, intermediate goods' share of tradedoes not increase. Intermediate goods trade shifts from being domi-nated by rawmaterials to manufactured parts. Rawmaterials produc-tion tends to depend on local geographical conditions in a way thatmanufacturing does not. Therefore, raw materials will be tradedeven when trade costs are high. Combined with the fact that tradecosts for raw materials fell less, most trade expansion is due to man-ufactured parts.

Examining the impact of tariffs and transportation costs separately,falling tariffs have a stronger effect on the growth of both manufactur-ing and VS trade. Specifically, falling tariffs on manufactured partslead to their offshoring while falling freight costs alone do not.

Other papers have studied the rise ofmanufacturing trade. Bergoeingand Kehoe (2003) find that a monopolistic competition model of tradecannot explain increasing manufacturing trade. Dalton (2009) examinesthe impact of Just-in-Time (JIT) inventories on the expansion of manu-factured goods trade. His model is able to generate a level shift inmanufacturing trade in the early 1980s when JIT is adopted, but doesnot generate the empirical pattern of trade expansion over the periodconsidered in this paper.

The paper contributes to the historical measurement of the struc-ture of trade protection. Examples include Anderson (1972) andIrwin (2007). It presents estimates of trade costs of goods by finaland intermediate uses. Supplementary tables used in the calculationof the input–output (IO) tables provide estimates of trade costs byIO commodity. These supplementary tables can be combined withthe IO tables to generate estimates of the structure of protection.U.S. foreign trade statistics do not provide detailed data on freightcosts before 1974, so historical data are very thin (Hummels, 2007).

There is a large literature investigating postwar trade growth, in-cluding Rose (1991), Krugman (1995), Baier and Bergstrand (2001),Bergoeing and Kehoe (2003) and Alessandria and Choi (2010).Models incorporating VS trade, such as Yi (2003) and Bridgman(2008) have been successful at resolving the puzzle that tariffs havenot fallen enough to generate the observed trade growth given esti-mates of the Armington elasticity (Armington, 1969), the aggregateelasticity of substitution between domestic and foreign goods. How-ever, they have not emphasized the structure of trade expansion.While Bergoeing et al. (2004) speculate that a VS model could gener-ate that change in composition, they do not pursue the issue.

A number of papers have examined the importance of intermedi-ates trade for a number of issues including development (Jones,2008; Goldberg et al., 2008; Estevadeordal and Taylor, 2008), firm

Fig. 1. Materials share of U.S. intermediate goods imports, 1925–2005.

productivity (Amiti and Konings, 2007), trade elasticities (Ramanarya-nan, 2006), business cycle co-movement (Kose and Yi, 2001), and theborder effect in gravity equations (Yi, 2010). Grossman and Rossi-Hansberg (2008a, 2008b) examine the growth of trade in intermediateservices. A number of papers have used input–output tables to examinethe factor content of trade, including Trefler and Zhu (2000) andReimer (2006). Theoretical models of vertical specialization trade in-clude Dixit and Grossman (1982) and Sanyal (1983). Unlike these pa-pers, I examine the change in the composition of intermediates trade.

2. Intermediate goods trade and trade costs facts

This section documents the change in the composition of interme-diates trade and the structure of trade costs for goods by use.

2.1. Composition of intermediate goods trade

Intermediate goods trade has shifted from being dominated byraw materials to manufactured parts. Fig. 1 shows the nominalshare of materials (agricultural and mining products) of U.S. interme-diate goods imports.3 Imports are dominated by such raw materialsearly in the period. After the 1950s, the composition of importsbegan to shift significantly. Materials fell from over half of importedintermediate goods to less than a quarter in the 1990s.

These data likely underestimate the real decline in the importanceof materials in intermediate goods trade. The data are reported in cur-rent dollars so they are vulnerable to swings in commodity prices, es-pecially oil. The run-up in materials share in the 2000s is driven by oilprices: non-fuel materials share shows a slight decline during this pe-riod. (Data constraints do not allow removing fuels from the full timeseries.) The spike in 1982 is also likely driven by high oil prices.

The decline in the importance of raw materials is not restricted tothe United States. Fig. 2 shows similar data for three major econo-mies.4 These data are reported in constant prices, so are not vulnera-ble to variations in commodity prices. (No such data exist for theUnited States.) All three show a decline in the importance of materialsimports.

2.2. The structure of protection

I now turn to the structure of protection from tariffs and transpor-tation costs for intermediate and final goods. I use input–output

3 Up to 1955, estimates are share of natural resource goods in non-final manufac-tured imports using data from Vanek (1963). From 1967 on, the estimates are theshare of imported intermediate goods used by goods producing (agriculture, miningand manufacturing) industries that originate from materials (agriculture and mining)industries using IO tables. Sources and full details of the estimates are given inAppendix A.

4 In the following Figure, the estimates are the share of imported intermediate goodsused by goods producing industries that originate from materials industries usingOECD constant currency input–output tables. See Appendix A for details.

Table 1Weighted U.S. import costs.

Variable 1967 1972 1992 1997 2002

All importsTariff 7.1 5.9 2.6 2.2 0.7Freight 7.4 5.3 4.0 3.3 3.4

Interm. (mfg.)Tariff 7.1 5.8 2.7 2.0 0.8Freight 7.3 4.9 4.1 4.4 3.9

Interm. (non-mfg.)Tariff 4.1 3.1 0.4 0.9 0.1Freight 10.8 9.9 10.9 7.1 3.5

FinalTariff 8.6 6.4 2.6 2.4 2.1Freight 5.8 4.7 3.4 2.8 2.3

135B. Bridgman / Journal of International Economics 86 (2012) 133–140

tables to split goods by use and estimate their trade costs. The tariffand transportation margins on imports are calculated as a supple-mentary table in the compilation of the input–output tables, sincethe margins need to be allocated to their producing industries:Wholesale trade for tariffs and transportation services for transporta-tion. This table is not reported for all benchmark years, but they arefor 1967 (pre-Kennedy Round) and 1972 (post-Kennedy Round).They can also be calculated for 1992, 1997 and 2002.

These margins are matched to the IO tables.5 I assume thatimported commodities are used at the same rate for intermediateand final production as aggregate supply of that commodity. This as-sumption is equivalent to assuming that the imported share of a com-modity is the same for both final and intermediate goods.6 The tradeweighted import cost is given by:

∑iτi yImpi sUsei

∑i yImpi sUsei

ð2:1Þ

where τi is the tariff rate, yiImp is imports and siUse is the share of the

domestic supply of commodity i that for that use (intermediate orfinal). Freight costs fi are weighted in a similar fashion.

As can be seen from Table 1, tariffs prior to the Kennedy Roundprotected manufacturers and allowed raw materials to enter at rela-tively low tariffs. (The Kennedy Round was agreed to in 1967 andimplemented over the next five years, so the 1967 to 1972 compari-son gives an indication of its effects.) This tariff structure was a longstanding feature of trade policy (Irwin, 2007). Since then, trade policyhas become more neutral with all goods facing similar, low tariffs.

The discriminatory tariff rates are to a large degree undone byhigher freight costs for non-manufactured goods. Most raw materialsare bulky and of low value. This finding is consistent with those ofYeats (1977). As found in Hummels (2007), freight rates have notfallen as rapidly as tariffs. There are significant differences acrosstypes of goods. Freight costs for manufactured goods have fallen bymuch more than for raw materials. Manufactured goods freightcosts fell in half while raw materials show no downward trend. Thisfinding is consistent with the containerization revolution reducingthe cost of non-bulk items (Levinson, 2006).

The overall protection profile (tariffs plus freight) has gone fromsomewhat protecting manufacturing and final goods producers toprotecting raw materials producers. The tariffs on all goods have de-clined nearly to zero. Freight for manufacturing has fallen significant-ly, especially for final goods. Freight costs remain relatively high formaterials.7

5 Appendix A provides detail on data sources and calculations.6 This assumption is widely used in the literature. For example, the OECD uses it to

split the IO tables into domestic and foreign sources.7 The significant decline in non-manufacturing intermediate freight costs in 2002 is

largely due to the run up in oil prices. Excluding oil products raises the freight rate to5.7%. Bridgman (2010) shows that freight rates for oil are negatively related to oilprices, since rates are charged by volume.

3. Model

The model features two countries with representative households.Production occurs in three stages. Each country produces a rawmate-rial unique to that country. These materials are inputs to a continuumof intermediate goods that are common to both countries. The inter-mediate goods are inputs to country specific final consumption goods.All three types of goods may be traded, but incur an iceberg transpor-tation cost and may face tariffs.

3.1. Households

There are two countries each with a representative household.Households have preferences over a consumption good representedby:

U C i1;C

i2

� �¼ ϕi

1 Ci1

� �ρ þ ϕi2 Ci

2

� �ρh i1ρ ð3:1Þ

where Cji denotes consumption good j∈{1,2} for country i∈{1,2}.

The associated prices are Pc, ji . ϕj

i is a home bias parameter, whereϕji=ϕ if j= i and ϕj

i=1−ϕ if j≠ i. Each country is endowed withlabor N i. The wage is Wi.

3.2. Raw materials sector

Each country can use labor Nmi to produce a raw material good Mj

i

with a price Pm, ji . Each country can only produce the good with its

name: j= i. Output is given by Ymi =Am

i Nmi .

3.3. Manufactured parts sector

There is a continuum of manufactured parts xi(z) with a price Px, ji (z)for z∈ [0,1]. Each country is endowed with technologies that combinematerials inputsMj

i, j∈{1,2} and laborNxi(z) to produce parts. Total out-

put of part z is given by:

Yix zð Þ ¼ Ai

x zð Þ Nix zð Þ

� �α ∑j

Mij zð Þ

� �σ !1σ

0@

1A1−α

: ð3:2Þ

The productivity parameters are given by A1 zð Þ ¼ 1

1þ zð Þθand

A2 zð Þ ¼ 1

2−zð Þθ, a variant of the mirror image technology in Bridgman

(2008) which is based on Dornbusch et al. (1977) and Eaton and Kor-tum (2002). The parameter θ governs the relative comparative ad-vantage of the two countries.

3.4. Consumption goods sector

Manufactured parts can be assembled into consumption goods usinglaborNc

i. Aswithmaterial goods, each country can only produce the goodwith its name: j= i. The total output is given by the technology:

Yic; j ¼ Ai

c Nic

� �αc ∫1

0ln xi zð Þ� �

dz� �1−αc

ð3:3Þ

for i=1,2 and j= i. The associated price is Pc, ji .

3.5. Transportation sector

The countries may trade the goods they produce with each otherby incurring an iceberg transportation cost specific to that good: fkfor k∈{m,x,c}.

136 B. Bridgman / Journal of International Economics 86 (2012) 133–140

3.6. Government

The countries each have a government that can impose an advalorem (net of transport fees) tariff τki on traded goods k∈ {m,x,c}.The government gives the domestic representative household trans-fers Ti and maintains budget balance.

4. Equilibrium

4.1. Definition

Households sell labor and purchase goods. They choose C1i and C2

i

to maximize U(C1i,C2i) subject to the budget constraint

∑jPic; jC

ij ¼ WiNi þ Ti

: ð4:1Þ

Materials firms buy labor and sell materials. They face competitivemarkets and solve:

MaxPim;iAimN

im−WiNi

m: ð4:2Þ

Manufactured parts firms face competitive markets and solve:

MaxPix;i zð ÞAix zð Þ Ni

x zð Þ� �α ∑

jMi

j zð Þ� �σ !1

σ

0@

1A1−α

−WiNix zð Þ−∑

jPijM

ij zð Þ:

ð4:3Þ

For j= i, consumption goods firms solve:

Max Pic;iA

ic Ni

c

� �α c ∫1

0ln xi zð Þ� �

dz� �1−αc−WiNi

c−∫1

0Pi zð Þxi zð Þdz: ð4:4Þ

Transportation firms buy domestic goods and sell exports. Mate-rials exporters face competitive markets and solve:

MaxP−im;iM

−ii −Pi

m;iMii 1þ fmð Þ ð4:5Þ

where Pm, i− i is the price of the materials in the other country. Parts and

consumption goods exporters solve a similar problem.Feasibility for each consumption good requires that for j=1,2:

f jcCj−j þ ∑

i¼1;2Cij ¼ Y j

c ð4:6Þ

where − j is the other country. The term fcjC− j

j is the amount of con-sumption used to pay the iceberg cost to ship the good. There is a cor-responding feasibility constraint for parts that are exported andmaterials production. Labor feasibility requires that labor sums tothe total population.

Ni ¼ Nic þ Ni

m þ ∫1

0Ni

x zð Þdz: ð4:7Þ

The definition of equilibrium is standard.

Definition 1. Given tariffs, an equilibrium is consumption, parts andmaterials goods allocations and prices in each period such that:

1. Households solve their problem,2. Materials, parts, consumption goods and transportation firms

solve their problem,3. The government balances its budget,4. The allocation is feasible.

4.2. Solution

The two countries are mirror images in manufactured parts pro-duction. There is a symmetric equilibriumwith a closed form solutionwhen the parameters are the same in the two countries. Specifically,if the parameters N i,τki, Ak

i for k∈{m,x,c} and are constant acrossthe two countries, there exists an equilibrium where C1

1=C22,

C12=C2

1, Pm, 11 =Pm, 2

2 , W1=W 2, Pc, 21 =Pc, 12 and Pc, 1

1 =Pc, 22 . Prices and

quantities in the parts and materials sectors across the countries mir-ror each other: Px1(z)=Px

2(1−z), etc. In the rest of the paper, I exam-ine this symmetric equilibrium.

I denote the common parameters and quantities (for example, Ni

and Wi) by omitting the i superscript (for example, τ1=τ 2=τ) andnormalize price of country one's material good to one (Pm, 1

1 =1).

This implies that the wage W1 ¼ 1Am

. Define zi as the cutoff industry

in country i such that manufactured parts z N z1 and z b z2 will beimported. Given the functional forms,

z1 ¼ 1− z2 ¼ 2 1þ τx þ fxð Þ1θ−1

1þ τx þ fxð Þ1θ þ 1: ð4:8Þ

Parts exports are given by:

z21þ τx þ fxð Þ

AmN þ Tð Þ 1þ fx þ 1þ τx þ fxð Þ 11−ρ

h i1þ τx þ fx þ 1þ τx þ fxð Þ 1

1−ρ

h i : ð4:9Þ

Consumption goods exports are given by:

C12 ¼ C2

1 ¼ AmN þ T

Pc 1þ τc þ fc þ 1þ τc þ fcð Þ ϕ1−ϕ

� � 11−ρ

� � ð4:10Þ

where Pc, 11 =Pc, 2

2 =Pc.Tariffs in the United States are collected on the FOB value of goods

(the value before transport costs are added). Therefore,

T ¼τmAmN

11þ τm þ fm

� � 11−σ

1þ 11þ τm þ fm

� � σ1−σ

� � 1−αð Þ z1 þ 1þ fxð Þ z2ð Þ

þ NAmτx 1− z1ð Þ þ AmNτc

1þ τc þ fc þ 1þ τc þ fcð Þ ϕ1−ϕ

� � 11−ρ

� � : ð4:11Þ

5. Results

This section calibrates the model and presents the results of thesimulations. In the calibration, I follow the convention of Yi (2003)and interpret the two countries as the United States and other indus-trialized countries (the EC plus Japan). I will generally use U.S. data toselect parameters since these data are easier to obtain. In particular, Ionly have structure of protection data for the United States. In addi-tion, aggregating some of the data concepts, such as vertical speciali-zation trade, is difficult across multiple countries. The industrialstructures of these countries are very similar and production param-eters do not appear to vary significantly across even countries withvery different industrial structures. (For example, see Gollin(2002).) Therefore, using U.S. parameters should be a reasonableproxy.

The model abstracts from the service sector since vertical special-ization over the period I examine is dominated by goods trade. There-fore, I use goods GDP as the data concept that matches the model'sGDP, with agricultural and mining sectors matching the model's

8 Irwin (2007), using the closely related Trade Resistance Index, estimates that theratio in 1960 was 1.74 which suggests the bias hasn't changed too much over the sam-ple period.

Table 2Baseline parameters.

Variable ρ θ α αc σ Am Ac ϕValue 0.85 0.24 0.7 0.3 −1 1.075 1 0.54

137B. Bridgman / Journal of International Economics 86 (2012) 133–140

materials sector and manufacturing matching the manufactured partsand final goods sectors.

The model also abstracts from capital. The industrial countries areat similar levels of development so differences in capital are unlikelyto be a significant factor in the impact of falling trade costs on tradegrowth. Since there is no capital, labor is the only source of valueadded. In the calibration, the model's labor stands in for all empiricalvalue added.

5.1. Calibration

The model's parameters are selected as follows.Waugh (2010) examines data for a number of countries (including

the United States) and finds that the value added share ofmanufacturing gross output is 0.33. I set the share of intermediategoods in consumption goods production αc equal to 0.33.

In the model, the only intermediate inputs into manufacturedparts production are raw materials. In I–O tables, the largest sourceof an industry's intermediate shipments is typically itself. To matchthe value added share of parts manufacturing α, we need to re-move the non-raw materials portion of intermediates. Let produc-tion function of parts in the I–O table be x ¼ nαmβx1−α−β . I will set

α ¼ αα þ β.

Based onWaugh (2010), I set Pα ¼ 0:33. To calculatePβ, I calculate

the share of intermediate shipments to manufacturing originatingfrom materials industries (agriculture and mining). Using the 1967U.S. direct transactions I–O table, this share is 21%. Therefore, I calcu-late that β ¼ 0:21⁎ 1− αð Þ ¼ 0:21⁎ 1−0:33ð Þ ¼ 0:14. Therefore, α ¼

0:330:33þ 0:14

¼ 0:7.

There is little information on materials elasticity parameter σ.Rotemberg and Rotemberg (1999) survey the business cycle litera-ture and find a range of elasticities from 0.1 to 0.7, which implies avalue of σ between −0.4 and −9. I use the value of −1 suggestedby Jones (2008), which implies an elasticity midway between Cobb–Douglas and Leontief. Below, I examine the robustness of the resultsto variations in this parameter.

The consumption good productivity parameter Ac is normalized to1. The materials productivity parameter Am is set to 1.075 to matchthe non-manufacturing share of goods GDP in 1967 (14.0%).

The Armington parameter ρ is set to match the long run tradeelasticity of 6.4 estimated in Ruhl (2005). While the model does notinclude the fixed costs featured in Ruhl (2005), his estimate isdesigned to capture the elasticity with respect to permanent changesin trade costs that this paper examines. I discuss the robustness of theresults to changes in ρ below.

The comparative advantage in parts parameter θ and home biasparameter ϕ are selected by grid search to match the level of VStrade in 1972 (6% of exports) and share of manufacturing outputthat is exported in 1967 (9%) respectively given the other parameters.Model VS trade is measured as the sum of the three sources of VS trade:Materials imports that are exported in parts ( 1− z2ð ÞPm;2

1 M21), imported

parts in exported final goods ( 1− z2ð ÞP12C

12) and imported materials in

domestic parts used in exported final goods P11C

12

Pm;21 M 2

1

P1x 0ð Þx1 0ð Þ z2

!. Note

that this definition does not include goods that are exported and reim-ported. While this is an important source of VS trade (see Johnsonand Noguera (2008)), it is omitted from the data sources I use.

The value of the comparative advantage parameter θ is set to 0.24.This parameter is not far from that used in the heterogeneous tradeliterature. Eaton and Kortum (2002) suggest a range of 0.08 to 0.28as reasonable for this parameter for traded goods, close to the rangeof 0.1 to 0.25 in Alvarez and Lucas (2007). Waugh (2010) uses avalue of 0.18 for all traded goods for all countries. These models donot map exactly (they only have one layer of traded production,

among other differences), but it suggests that the calibration is notstrongly different from the literature. I discuss the robustness of themodel to changes in θ below.

Tariffs and freight rates are taken from Table 1. I use non-manufactured intermediate goods for raw materials, manufacturedintermediate goods for parts and manufactured final goods for finalproduction.

Since these are trade weighted measures, they suffer from somewell-known limitations. High trade cost goods are likely to be tradedless than low trade cost goods. A particular issue with this measure inthis context is that there has been significant trade growth along theextensive margin: trade in new goods (Kehoe and Ruhl, 2003). There-fore, there are a significant number of goods whose trade costs arenot measured in the early years. Bridgman (2010) shows that forfreight, lower trade costs induce lower value goods to be tradedwhich mask changes in trade costs.

A measure of the size of trade weighting bias is the MercantilistTrade Resistance Index (MTRI) proposed by Anderson and Neary(2003), which is the estimated uniform tariff equivalent that gener-ates the observed level of trade. I scale up trade costs by 1.69, theratio of MTRI that Kee et al. (2005) estimates to trade-weighted tariffsfor the United States in 2002.8 These estimates only cover tariffs. I amnot aware of any MTRI estimates for transport costs. Anderson andvan Wincoop (2004) note that transport costs are similar to tariffsin magnitude and variability, so a tariff based estimate is likely to bea reasonable proxy for bias in transport cost measures.

The baseline parameters are summarized in Table 2.

5.2. Simulations

This section presents the results of the calibrated model. In inter-preting the results, I identify the raw materials sector as non-manufacturing output and the manufactured parts and final goodssectors as manufacturing output.

The model is able to match a number of trade growth facts. It gen-erates both the empirical growth in trade and the change incomposition.

As can be seen from Fig. 3, the model does a good job of matchingoverall empirical trade growth. The share of goods production that isexported in the model grows 174% from 1967 to 2002, not far fromthe actual growth in export share of 135%. There are only predictionsfor a few years, so the model has less to say about the year to yeartime series. However, those few predictions are consistent with thedata.

The model generates a doubling of the trade share with a relativelymodest fall in trade costs due to the rapid expansion of manufacturingtrade. The share of manufacturing output that is exported in themodel grows much faster than total trade, growing by 350% between1967 and 2002. This growth is somewhatmore than the 317% empiricalgrowth in the share of manufacturing output. This growth ismostly dueto increasing trade in manufactured parts. Of the three types of goods,manufactured parts grow the fastest. In 1967, there is no trade inparts. By the 1990s, this category is over half of manufacturing trade.This prediction is consistent with the finding that parts and componenttrade has grown more rapidly than manufacturing trade (Yeats, 2001).

Fig. 3. U.S. exports/value added, model and data 1967–2006. Fig. 4. U.S. materials exports/value added, model and data 1967–2006.

138 B. Bridgman / Journal of International Economics 86 (2012) 133–140

VS trade also grows rapidly. Table 3 compares the model's predic-tions with the calculations from Chen et al. (2005). In the model, VStrade increases from 6% of exports in 1972 to 17% in 1997. Themodel is broadly consistent with the estimates for the United Statesin Chen et al. (2005), though it overpredicts the increase in VS trade.

While VS trade grows rapidly, intermediate goods trade does notincrease significantly. This prediction matches the data: Intermediategoods share of trade is roughly constant. The model predicts that52.3% of exports are intermediates (materials and parts) in 1967which is close to its prediction of 58.5% in 2002. Therefore, the riseof VS trade in the model is not driven by a relative increase in inter-mediates trade. The level is similar to estimates in Chen et al.(2005). They estimate intermediates were about half of trade (50.4%in 1972 and 51.9 in 1997), not too far from the model's predictions(45% in 1972 and 54% in 1997).

As can be seen in Fig. 4, the model predicts very little growth inmaterials trade. This prediction matches the data. There is little per-manent growth in materials trade, though commodity price swingslead to temporary spikes. The share of U.S. materials productionthat is exported increases by 10.6% from 1967 to 2002, very close tothe model's prediction of 12.1%.

One might be concerned that shifts in industrial structure drivethe results. Table 3 reports the materials share of U.S. goods GDP.While there has been a shift away from goods production to services,within goods the share devoted to materials production has been es-sentially constant. The model also predicts no trend in materials shareof goods GDP.

Raw materials production tends to depend on local geographicalconditions in a way that manufacturing does not. Mines can only besited where ore exists naturally. A steel plant can be placed anywhere.Therefore, raw materials will be traded even when trade costs arehigh. Combined with the fact that trade costs for raw materials fellless, most of the new trade in goods is due to manufactured parts.This feature of the model is consistent with empirical finding thatgoods lower down the supply chain have lower price–trade elastici-ties (Balassa and Kreinin, 1967).

It is not the case that geography does not matter for manufactur-ing, but it is less tied to geographic endowments relative to raw ma-terials. Even industries that use inputs that are closely tied to

Table 3Model moments.

Variable 1967 1972 1992 1997 2002

VS trade (model) 5.1 5.8 16.2 17.0 19.7VS trade (data) 5.9 12.3Interm. trade share (model) 52.3 45.1 53.8 54.0 58.5Interm. trade share (data) 50.4 51.9Mat. share of goods GDP (model) 14.0 14.1 14.1 14.1 14.1Mat. share of goods GDP (data) 14.0 15.6 14.9 13.5 15.2

natural endowments are often placed far from the sources of those in-puts. For example, the center of cane sugar refining in the UnitedStates was New York City. New Orleans, a major port close both to do-mestic and imported raw sugar sources, was a minor producer (Glae-ser, 2005).

The results may explain why trade among industrial countrieshas increased, despite having similar industrial structures. Whentrade was dominated by goods that depend heavily on endowments,less developed countries — economies dominated by raw materialsproduction — made up more of world trade. This explanation doesnot rely on increasing returns or agglomeration economies, as inKrugman (1980), to explain the concentration of trade among simi-lar countries. In fact, it is precisely because productivity differencesin parts production between industrialized countries are small that rel-atively small declines in trade costs have such a large impact on tradegrowth. Since the productivity differences in tradeable goods betweenrich and poor countries are large (Herrendorf and Valentinyi, 2007),even high trade barriers (such as those used by import substitution pro-grams) are not sufficient to prevent poor countries from specializing inmaterials production.

5.2.1. RobustnessThe Armington parameter ρ and the materials elasticity σ were

assigned rather than calibrated directly. I examine the robustness ofthe model to changes in these parameters below. I also examine theimpact of changes in the comparative advantage parameter θ.

There is controversy over the proper value of the Armington pa-rameter ρ. Simonovska and Waugh (2011) show that one of theprominent estimates of this parameter is subject to bias in small sam-ples that tends to overestimate it. (The estimate I use, from Ruhl(2005), is not subject to this bias.) Simonovska and Waugh (2011)obtain estimates close to 4, lower than the 6.4 that is used in the base-line estimate. I check the robustness of the model by setting ρ ¼1− 1

4 ¼ 0:75 and redo the calibration. The only parameter thatchanges is that the home bias parameter ϕ increases to 0.6.

The qualitative predictions of the model are unchanged.Manufacturing trade still grows faster than overall goods trade andthe other moments of the model are almost unchanged. The amountof trade expansion falls somewhat. The predicted increase inmanufacturing output exported falls from 350% to 297%. Predictedgoods export share increases 146% compared to 174% in the baselinemodel. The model still predicts trade expansion very close to what isobserved in the data.

The empirical range for the materials elasticity parameter σ isquite wide. I experimented with alternative values for this parameterand it has a negligible effect on the results. Even large changes in σhave a small effect on materials trade. Since the growth of materialstrade is a very small part of the increase in overall trade, changes inthe materials sector have a tiny impact on the predictions of themodel.

Table 4Counterfactuals.

Variable 1967 tariffs 1967 freight 1967 parts tariffs

Total trade growth 1967–2002 20.6% 78.0% 73.4%Mfg. trade growth 1967–2002 35.2% 156.9% 137.4%z1 1 0.89 1

139B. Bridgman / Journal of International Economics 86 (2012) 133–140

The estimate of the comparative advantage parameter θ was nearthe top of the range found in the literature. Low values of θ reducepredicted trade growth. There is little difference in parts productivityacross the two countries. Therefore, even modest levels of trade costsshut down manufactured parts trade, closing off a major source oftrade growth.

5.2.2. Offshoring: tariffs or freight?The model allows us to decompose the importance of the various

trade costs for the increase in offshoring, the shift of production fromdomestic to foreign sources. Falling tariffs are widely cited as the pri-mary reason for increasing trade and offshoring. Others, such asLevinson (2006), have suggested that improvements in shippingtechnology such as containerization are a first order source of increas-ing trade.9

The amount of offshoring is measured by the cut-off z. The mea-sure of offshored industries in the symmetric equilibrium is givenby 1−z1 ¼ z2. The baseline model predicts that all possible domesticindustries operate in 1967. As trade expands, the set of industries thata country operates contracts. By 2002, 26% of domestic partsmanufacturing industries have closed. (In terms of the model,z1 2002ð Þ ¼ 0:74.)

To examine the relative importance of these two forces, I runcounterfactual simulations holding trade costs at their 1967 levels.The first counterfactual simulation (1967 Tariffs) reported in Table 4holds all tariffs at their 1967 levels. Freight costs fall as they do inthe baseline simulation. The next simulation (1967 Freight) doesthe opposite: it holds freight rates at their 1967 levels while tariffsfall as they do in the baseline. Trade growth is much stronger, indicat-ing a stronger role for tariffs.

Falling freight costs alone generate very little trade growth. Theycannot induce parts trade while falling tariffs do. Table 4 shows thatthe 1967 Tariffs counterfactual does not cause any of the manufac-tured parts to be traded, while there is parts trade in the 1967 Freightcounterfactual. In terms of model quantities, z1 does not fall from onein the 1967 Tariffs counterfactual while it falls to 0.89 in the 1967Freight counterfactual. In fact, simply maintaining tariffs on manufac-tured parts at their 1967 levels (the counterfactual named “1967Parts Tariffs” in Table 4) is sufficient to prevent trade in parts through2002 (z1 1). While manufactured goods trade still grows significantlydue to growing finished goods trade, there is no trade in parts.

6. Conclusion

This paper shows that trade costs can explain the change in thecomposition of international trade. However, it does not consider al-ternative causes of VS trade growth. Improvements in technology,both production (allowing better standardization) and communica-tion (allowing better coordination across locations), may have had arole. Financial liberalization has encouraged foreign direct invest-ment, allowing firms to offshore while keeping production withinthe firm. Trade among affiliated firms within multinationals hasbeen an important source of trade growth. However, the strength ofthe results suggests that trade costs would remain a significant source

9 Technological change may improve transportation in ways that are not reflected inprice, such as increasing reliability (Hummels, 2007). The importance of timeliness isemphasized by Harrigan and Venables (2006).

of the rise in VS and manufacturing trade even if other sources wereconsidered.

Appendix A. Data

A.1. Fig. 1

A.1.1. 1925–1955Data are drawn from Vanek (1963). Natural resource share of U.S.

imports (Table 5.11) divided by intermediate goods share of imports:1 minus final manufactured goods (Table 5.8) and manufactured food(Table 5.6) import share.

A.1.2. 1967Import data are from U.S. Department of Commerce (1977),

Table 1b. Benchmark I–O table is the 85-industry total requirementstable from the BEA website. Imports are estimated by multiplying im-port share of gross output by total requirements table. Materials shareis imported shipments from mining and agriculture industries (I–Ocodes 1–10) to goods producing industries (mining, agriculture andmanufacturing: I–O codes 1–10 and 13–64) over all imported ship-ments from goods producing industries to goods producingindustries.

A.1.3. 1972–1990Data are drawn from 1995 edition of the OECD input–output ta-

bles (www.oecd.org/sti/inputoutput/), current dollar imported trans-actions table (USMIOCXX). Materials share is shipments from miningand agriculture industries (industry codes 1–2) to goods producingindustries (mining, agriculture and manufacturing industries: indus-try codes 1–24) over all imported shipments from goods producingindustries to goods producing industries.

A.1.4. 1995–2005Data are drawn from 2006 edition of the OECD input–output ta-

bles (www.oecd.org/sti/inputoutput/), imported transactions table.Materials share is shipments from mining and agriculture industries(industry codes 1–3) to goods producing industries (mining, agricul-ture and manufacturing industries: industry codes 4–25) over allimported shipments from goods producing industries to goods pro-ducing industries. The non-energy series removes shipments origi-nating from mining and quarrying (energy) (Industry 2) from boththe numerator and denominator.

A.2. Fig. 2

A.2.1. 1972–1990Data are drawn from 1995 edition of the OECD input–output ta-

bles, constant price imported transactions table (UKMIOKXX,JPMIOKXX,FRMIOKXX). Materials share is shipments from miningand agriculture industries (industry codes 1–2) to goods producingindustries (mining, agriculture and manufacturing industries: indus-try codes 1–24) over all imported shipments from goods producingindustries to goods producing industries.

A.3. Figs. 3 and 4

A.3.1. Export share of value addedGoods value added and exports from NIPA Tables 1.2.5 and 4.1.

Manufacturing value added from BEA's value added by industry(www.bea.gov/industry/io_histannual.htm). Manufacturing exportsfrom UNCTAD (UNCTAD.org). Materials value added and exports:goods-manufacturing series.

140 B. Bridgman / Journal of International Economics 86 (2012) 133–140

A.4. Table 1

A.4.1. IO tablesBenchmark input–output tables are drawn from the BEA Industry

Economic Accounts website. The 1967 and 1972 tables are the 85-industry total requirements tables. The 1992, 1997 and 2002 are theuse tables at the detailed level after redefinition.

A.4.2. Import margins: 1967 and 1972The imports and trade costs are reported in U.S. Department of

Commerce (1977), Table 1b for 1967 and Ritz et al. (1979), Table Dfor 1972.

A.4.3. Import margins: 1992, 1997 and 2002Import, duties and freight data come from Feenstra (1994) and U.S.

International Trade Commission (dataweb.usitc.gov). This data is con-corded into the IO classification. The 1992 concordance is an unpublishedconcordance provided by BEA's Industry Economic Accounts. The 1997and 2002 concordances are taken from the BEA website.

A.4.4. CalculationCommodities originating from service industries and government

are excluded: two digit IO Industries 65–79 (1967/72/92) and onedigit industries 4–9 and two digit industry 22 (Utilities) (1997/2002).Manufacturing industries are two digit IO industries 13–64(1967/72/92) and one digit industry 3 (1997/2002).

A.5. Table 3

A.5.1. VS and intermediates tradeCalculations from Chen et al. (2005), Tables 1 and 2.

A.5.2. Materials shareBEA's historical GDP by industry: agriculture and mining value

added share of GDP divided agriculture, mining and manufacturingvalue added share of GDP. (www.bea.gov/industry/gdpbyind_data.htm).

References

Alessandria, George, Choi, Horag, 2010. Do falling iceberg costs account for recent U.S.export growth? Working Paper 10–10, Federal Reserve Bank of Philadelphia.

Alvarez, Fernando, Lucas Jr., Robert E., 2007. General equilibrium analysis of the Eaton–Kortummodel of international trade. Journal ofMonetary Economics 54 (6), 1726–1768.

Amiti,Mary, Konings, Jozef, 2007. Trade liberalization, intermediate inputs, and productiv-ity: evidence from Indonesia. American Economic Review 97 (5), 1611–1638.

Anderson, James E., 1972. Effective protection in the U.S.: a historical comparison. Journalof International Economics 2 (1), 57–76.

Anderson, James E., Neary, Peter, 2003. The mercantalist index of trade policy. Interna-tional Economic Review 44 (2), 627–649.

Anderson, James E., van Wincoop, Eric, 2004. Trade costs. Journal of Economic Literature42 (3), 691–751.

Armington, Paul S., 1969. A theory of demand for products distinguished by place ofproduction. International Monetary Fund Staff Papers 16 (1), 159–178.

Baier, Scott L., Bergstrand, Jeffrey H., 2001. The growth of world trade: tariffs, transportcosts, and income similarity. Journal of International Economics 53 (1), 1–27.

Balassa, Bela, Kreinin, Mordechai, 1967. Trade liberalization under the “KennedyRound”: the static effects. Review of Economics and Statistics 49 (2), 125–137.

Bergoeing, Raphael, Kehoe, Timothy J., 2003. Trade Theory and Trade Facts, Staff Report284, Federal Reserve Bank of Minneapolis.

Bergoeing, Raphael, Kehoe, Timothy J., Strauss-Kahn, Vanessa, Yi, Kei-Mu, 2004. Whyis manufacturing trade rising even as manufacturing output is falling? AmericanEconomic Review Papers and Proceedings 94 (3), 134–138.

Bridgman, Benjamin, 2008. Energy prices and the expansion of world trade. Review ofEconomic Dynamics 11 (4), 904–916.

Bridgman, Benjamin, 2010. Market Entry and Trade Weighted Import Costs, WorkingPaper 2010-05, Bureau of Economic Analysis.

Chen, Hogan, Kondratowicz, Matthew, Yi, Kei-Mu, 2005. Vertical specialization andthree facts about U.S. international trade. North American Journal of Economicsand Finance 16 (1), 35–59.

Dalton, John T. 2009, Explaining the growth in manufacturing trade, mimeo, Universityof Minnesota.

Dixit, Avinash, Grossman, Gene, 1982. Trade and protection with multistage produc-tion. Review of Economic Studies 49 (4), 583–594.

Dornbusch, Rudiger, Fischer, Stanley, Samuelson, Paul, 1977. Comparative advantage,trade, and payments in a Ricardian model with a continuum of goods. AmericanEconomic Review 67 (5), 823–839.

Eaton, Jonathan, Kortum, Samuel, 2002. Technology, geography, and trade. Econometrica70 (5), 1741–1779.

Estevadeordal, Antoni, Taylor, Alan M., 2008. Is theWashington consensus dead? Growth,Openness, and the Great Liberalization, 1970s–2000s, Working Paper 14264, NBER.

Feenstra, Robert, 1994. New product varieties and the measurement of internationalprices. American Economic Review 81 (1), 157–177.

Feenstra, Robert, 1998. Integration of trade and disintegration of production in theglobal economy. Journal of Economic Perspectives 12 (4), 31–50.

Glaeser, Edward, 2005. Urban colossus: why New York is America's largest city. FRBNYEconomic Policy Review 11 (2), 7–24.

Goldberg, Pinelopi K., Khandelwal, Amit, Pavcnik, Nina, Topalova, Petia, 2008. Importedintermediate inputs and domestic product growth: evidence from India, WorkingPaper 14416, NBER.

Gollin, Douglas, 2002. Getting income shares right. Journal of Political Economy 110 (2),458–474.

Grossman, Gene, Rossi-Hansberg, Esteban, 2008a. Task Trade between Similar Coun-tries, Working Paper 14554, NBER.

Grossman, Gene, Esteban, Rossi-Hansberg, 2008b. Trading tasks: a simple theory of off-shoring. American Economic Review 98 (5), 1978–1997.

Harrigan, James, Venables, Anthony, 2006. Timeliness, trade and agglomeration. Journalof Urban Economics 59 (2), 300–316.

Herrendorf, Berthold Valentinyi, Akos, 2007, Which sectors make the poor countries sounproductive? Mimeo, Arizona State University.

Hummels, David, 2007. Transportation costs and international trade in the second eraof globalization. Journal of Economic Perspectives 21 (3), 131–154.

Hummels, David, Rapoport, Dana, Yi, Kei-Mu, 1998. Vertical specialization and thechanging nature of world trade. FRBNY Economic Policy Review 4 (2), 79–99.

Hummels, David, Ishii, Jun, Yi, Kei-Mu, 2001. The nature and growth of vertical special-ization in world trade. Journal of International Economics 54 (1), 75–96.

Irwin, Douglas A., 2007. Trade Restrictiveness and Deadweight Losses from U.S. Tariffs,1859–1961, Working Paper 13450, NBER.

Johnson, Robert C. Noguera, Guillermo, 2008, Accounting for intermediates: productionsharing and trade in value added, mimeo, U.C. Berkeley.

Jones, Charles I., 2008. Intermediate Goods and Weak Links: A Theory of EconomicDevelopment, Working Paper, U.C. Berkeley.

Kee, Hiau Looi, Nicita, Alessandro Olarreaga, Marcelo, 2005, Estimating trade restric-tiveness indices, mimeo, World Bank.

Kehoe, Timothy J., Ruhl, Kim J., 2003. How important is the new goods margin in inter-national trade? Federal Reserve Bank of Minneapolis Staff Report, 324.

Kose, M. Ayhan, Yi, Kei-Mu, 2001. International trade and business cycles: is verticalspecialization the missing link? American Economic Review Papers and Proceed-ings 91 (3), 371–375.

Krugman, Paul, 1980. Scale economies, product differentiation, and the pattern oftrade. American Economic Review 70 (5), 950–959.

Krugman, Paul, 1995. Growingworld trade: causes and consequences. Brookings Paperson Economic Activity 1995 (1), 327–377.

Levinson, Marc, 2006. The Box: How the Shipping Container Made the World Smallerand the World Economy Bigger. Princeton University Press, Princeton.

Ramanaryanan, Anath 2006, International trade dynamics with intermediate inputs,mimeo, Federal Reserve Bank of Dallas.

Reimer, Jeffrey J., 2006. Global production sharing and trade in the services of factors.Journal of International Economics 68 (2), 384–408.

Ritz, Philip, Roberts, Eugene, Young, Paula, 1979. Dollar-value tables for the 1972input–output study. Survey of Current Business 59 (4), 51–72.

Rose, Andrew, 1991. Why has trade grown faster than income? Canadian Journal ofEconomics 24 (2), 417–427.

Rotemberg, Julio J., Woodford, Michael, 1999. The cyclical behavior of prices and costs. In:Taylor, J.,Woodford, M. (Eds.), Handbook of Macroeconomics, 1. Elsevier, Amsterdam,pp. 1051–1135.

Ruhl, Kim J. 2005, Solving the elasticity puzzle in international economics, mimeo,New York University.

Sanyal, Kalyan, 1983. Vertical specialization in a Ricardian model with a continuum ofstages of production. Economica 50 (197), 71–78.

Simonovska, Ina, Waugh, Michael E., 2011. The Elasticity of Trade: Estimates andEvidence, Working Paper 16796, NBER.

Trefler, Daniel, Zhu, Susan, 2000. Beyond the algebra of explanation: HOV for the technol-ogy age. American Econmic Review Papers and Proceedings 90 (2), 145–149.

U.S. Department of Commerce, 1977. Revised Input–Output table for the United States:1967, Staff Paper 29, Bureau of Economic Analysis.

Vanek, Jaroslav, 1963. The Natural Resource Content of United States Foreign Trade,1870–1955. MIT Press, Cambridge.

Waugh, Michael E., 2010. International trade and income differences. American EconomicReview 100 (5), 2093–2124.

Yeats, Alexander J., 1977. Do international transport costs increase with fabrication?Some empirical evidence. Oxford Economic Papers 29 (3), 458–471.

Yeats, Alexander J., 2001. Just how big is global production sharing? In: Kierzkowski, H.,Arndt, S. (Eds.), Fragmentation: New Production Patterns in the World Economy.Oxford Press, Oxford, pp. 108–143.

Yi, Kei-Mu, 2003. Can vertical specialization explain the growth of world trade? Journalof Political Economy 111 (1), 52–102.

Yi, Kei-Mu, 2010. Can multi-stage production explain the home bias in trade? AmericanEconomic Review 100 (1), 364–393.