TRADE IN INTERMEDIATE GOODS: TRENDS, EFFECTS, AND DETERMINANTS
by
NINO SITCHINAVA
A DISSERATION
Presented to the Department of Economics
and the Graduate School of the University of Oregon
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
June 2008
11
University of Oregon Graduate School
Confirmation of Approval and Acceptance of Dissertation prepared by:
Nino Sitchinava
Title:
"Trade in Intermediate Goods: Trends, Effects, and Determinants"
This dissertation has been accepted and approved in partial fulfillment of the
requirements for the degree in the Department of Economics by:
Bruce Blonigen, Co-Chairperson, Economics
Ronald Davies, Co-Chairperson, Economics
Glen Waddell, Member, Economics
Michael Pangburn, Outside Member, Decision Sciences
and Richard Linton, Vice President for Research and Graduate Studies/Dean of the
Graduate School for the University of Oregon.
June 14, 2008
Original approval signatures are on file with the Graduate School and the University of
Oregon Libraries.
© 2008 Nino Sitchinava
111
in the Department of Economics
Nino Sitchinava
An Abstract of the Dissertation of
for the degree of
to be taken
IV
Doctor of Philosophy
June 2008
Title: TRADE IN INTERMEDIATE GOODS: TRENDS, EFFECTS, AND
DETERMINANTS
Approved:
Bruce A. Blonigen, Co-Chair
Approved: _
Ronald B. Davies, Co-Chair
A large number of studies search for stylized facts on the rapid growth, impact,
and determinants of international outsourcing of production. The analyses of these studies
are considerably constrained by limitations in the international trade data, which do not
differentiate between trade in intermediate and finished goods. I improve on these data
and develop a trade dataset that draws a clear distinction between trade in intermediate
and finished goods. I use new data to provide an integrated view of the importance of
U.S. global production sharing. I assess the magnitude and nature of global production
sharing, explore its impact on growing U.S. manufacturing wage inequality, and examine
the forces driving the location and volume of this trade. My findings indicate that the
composition of trade has not changed as previously speculated and that trade in
intermediate inputs is just as prevalent as trade in consumer goods. Additionally, my
results indicate that the impact of foreign offshoring of intermediate inputs on the
growing wage gap in U.S. manufacturing industries is larger than previously estimated.
Lastly, I show that quality of contracting environment and thickness of input supplier
markets are important factors for the location and extent of trade in specialized inputs.
v
VI
CURRICULUM VITAE
NAME OF AUTHOR: Nino Sitchinava
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, OR
Whitworth University, Spokane, WA
DEGREES AWARDED:
Doctor of Philosophy in Economics, Spring 2008, University of Oregon
Master of Science in Economics, May 2005, University of Oregon
Bachelor of Arts in Economics and International Business, May 2002, Whitworth
University
AREAS OF SPECIAL INTEREST:
Microeconomics, Econometrics, Industrial Organization, International Economics
PROFESSIONAL EXPERIENCE
Graduate Teaching Fellowship, Department of Economics, University of Oregon,
2003-2008
Independent Course Instructor
Introduction to Econometrics, Spring, Fall 2007, Spring 2008
Issues in Industrial Organization, Spring, Fall 2006
Principles of Microeconomics, Summer 2005, Winter 2006
Teaching Assistant
Principles of Economics, Principles of Microeconomics, Money and
Banking, Intermediate Macroeconomics, Issues in Developing
Economies, Introduction to Econometrics
Vll
Research Assistant, Department of Economics, University of Oregon and Institute
for Water Resources U.S. Army Corps of Engineers, 2005
Research assistantship under direction of Wesley W. Wilson
Foundation for Russian American Economic Cooperation, Seattle, WA, 2002-
2003
GRANTS, AWARDS AND HONORS:
Graduate Teaching Fellowship, University of Oregon, 2003-present
Graduate School Research Award, University of Oregon, 2006-2007
Kleinsorge Research Fellowship, University of Oregon, 2006
PUBLICATrONS:
Sitchinava, Nino, Mark Burton, and Wesley Wilson, "Heterogeneous Products,
Demanders, and Elasticities," Transportation Research Part E, Invited Revision, 2008.
Vlll
ACKNOWLEDGMENTS
I wish to express my sincere thanks to Professors Bruce Blonigen and Ron Davies
for their guidance, insights, and assistance throughout the many stages of this project.
Additionally, I thank Christina Steiger, Glen Waddell, Michael Pangburn, and Kelii
Haraguchi for useful comments. The views expressed herein are those of the author.
MOHM pO)J;HTeJUIM, ryrre IT IllaJIHKy.
To my parents, Gulya and Shalik.
IX
Chapter
I. INTRODUCTION
TABLE OF CONTENTS
Page
1
x
II. STRUCTURE OF U.S.TRADE 4
11.1 Introduction 4
ILlI Existing Measures of Global Production Sharing 8
ILlL1 Trade in Parts and Components 8
II.IL2 Input-Output Tables and Trade 10
11.11.3 Other Measures 12
11.11.4 Key Implications 12
11.111 Data Description 13
II.IIL1 Dataset Construction and Sources 13
II.III.2 New Data Versus Old Measures 17
ILlV Magnitudes, Composition, and Cyclica1ity 20
ILlV.1 Overall Magnitudes 20
II.IV.2 Commodity Composition 21
ILIV.3 Cyclicality 23
ILV International Specialization of Production 25
ILV 1 Across-Product Specialization by Import Share 26
II.V.2 Across-Product Specialization by Product Share 29
II.V.3 Within-Product Specialization by Unit Values 33
II.VI Conclusion 36
III. OUTSOURCING, TECHNOLOGY, AND U.S. WAGE INEQUALITy..... 37
111.1 Introduction 37
IILII Old and New Evidence of Wage Inequality..................................... 42
IILlII Empirical Methodology 45
IILlV Data and Descriptive Statistics 50
IILV Results 56
IILV.1 Preliminary Regression 57
IILV.2 Stage 1 59
IILV.3 Stage 2 64
Xl
Chapter Page
IlLVI Sensitivity Analysis 67
IlLVII Conclusion 67
IV. CONTRACTS, MARKET THICKNESS AND OUTSOURCING 70
IV.I Introduction 70
IV.II The Model 74
IV.ILI Model Set-Up 74
IV.II.2 Partial Equilibrium Analysis 79
IV.III Empirical Methodology and Data Description 82
IV.IILI Specification 83
IV.m.2 Data Sources and Variable Definitions 85
IV.IV Results 88
IV.IV.l Trade in Intermediate Inputs 89
IV.IV.2 Comparison with Non-Intermediate Imports 92
IV.V Conclusion 96
V. CONCLUSION 97
APPENDICES 100
A. MARKET STRUCTURE INDEX OF HTS IMPORTS 100
B. FIGURES AND TABLES FOR CHAPTER II 117
C. DATA APPENDIX TO CHAPTER III........................................................... 135
D. FIGURES AND TABLES FOR CHAPTER III............................................. 142
E. TABLES FOR CHAPTER IV......................................................................... 152
BIBLIOGRAPHY 159
xu
LIST OF FIGURES
Figure Page
B.l. Comparison ofNew and Old Data ofTrade in Intermediate Inputs 118
B.2. U.S. Manufacturing Imports by Category, 1989-2004 119
B.3. U.S. Imports by 3-digit NAICS Industry, 1989-2004 120
D.l. U.S. Wage Inequality, 1963-2005 142
Xlll
LIST OF TABLES
Table Page
A. 1. Market Structure ofImports from the BEA Import-Matrix 103
A.2. Hypothetical Example of Market Structure 109
A.3. Import-Matrix Weights vs. My Weights 110
AA. Types and Frequency ofHTS Clusters 111
B.l. Breakdown ofImports by Utilization Weights, 1989-2004 122
B.2. U.S. Imports Relative Importance, 1989-2004 123
B.3. U.S. Top 10 Industry Imports 124
BA. Pro-Cyclical Behavior of U.S. Imports 125
B.S. List of Sample Countries 126
B.6. U.S. Import Value Market Share by Region, 1989-2004 127
B.7. Largest Gains in Market Share, 1989-2004 128
B.8. Product Penetration by Region, 1989-2004 129
B.9. Largest Gains in Product Penetration by Country, 1989-2004 130
B.I0. Decomposed Import Growth by Region, 1989-2004 131
B.ll. Product Shares by Source Country Income Levels, 1989-2004 132
B.12. Regression of Unit Values on Income, 1989-2004 133
B.13. Regression of Unit Values on Income by Select Industries, 1989-2004.... 134
C.l. Selected Services 141
D.l. Summary Statistics of]\Jon-Structural Variables 143
D.2. Summary Statistics for Structural Variables 144
D.3. Consistency of Data with Equation (IlIA) 145
DA. Stage I - Original Feenstra and Hanson (1999) Specification 147
D.5. Stage I - Full Specification 148
D.6. Stage I - Decomposed Dependent Variable (Refined Measure) 149
D.7. Stage I - Final Specification (Refined Measure) 150
D.8. Stage II - (Refined Measure) 151
E.l. Summary Statistics of Explanatory Variables 152
E.2. Correlation Matrix ofInteraction Terms . 153
E.3. Sample Countries 154
EA. Contracts and Market Thickness . 155
XIV
Table Page
E.5. Sensitivity Results 156
E.6. Comparison ofImport Patterns 157
E.7. Comparison ofImport Patterns By Utilization Rates 158
1CHAPTER I
INTRODUCTION
U.S. trade has increased dramatically over the last four decades, with the export
share of GDP doubling and the import share nearly tripling since the 1970s. Various
evidence suggests that much of this increase stems from the explosive growth in
international outsourcing of intermediate inputs production and finished goods assembly.
The rapid widening of the wage gap that happened concurrently with these developments
drew the attention of politicians and popular press to the potentially adverse affects of
global production sharing. In response, many trade economists have been concerned with
various aspects of such overseas production arrangements, mainly: 1) their relative
importance, 2) their impact on home economies, and 3) their determinants. Unfortunately,
data constraints have hampered progress and it is well understood that the available
measures of offshoring do not accurately reflect the true nature and/or impact of global
production sharing. Furthermore, the mixed evidence suggested by these measures
neither justifies nor dismisses the fear of outsourcing projected by the popular press.
In my dissertation I offer a fresh perspective on international outsourcing and the
three issues troubling trade economists. I accomplish this goal by first addressing the
primary shortcoming of the literature - the unavailability of data on global production
2sharing. With this in mind, I construct data on trade in intennediate goods, which is
integral for gauging the value of international outsourcing. The technical documentation
of this dataset is provided in Appendix A. Beginning with Chapter II of the dissertation, I
employ the new data on intennediate inputs to provide unique insights into the magnitude
and the nature of global production sharing. Next, I utilize this knowledge to examine the
impact of international outsourdng on U.S. manufactming wage inequality during
1989-2004, which I describe in Chapter III. Finally, in Chapter IV, I explore the
detel111inants of global production sharing, with a particular focus on institutions, input
supplier markets, and specific investment. Throughout my analyses, I use trade in final
goods as a benchmark for assessing the relative importance of trade in intennediate
goods. This comparison offers a novel perspective on the relevance of trends, effects, and
determinants of outsourcing and highlights one of the contributions of my research.
The new data and my analyses uncover a wealth of unique findings on the nature
and effects of international outsourcing. In Chapter II, I find that contrary to the common
perception, the composition of u.s. impOlis remained relatively constant, with imports of
intennediate and consumer goods each comprising roughly a third of u.s. imports during
the 1990s and 2000s. Trade in inputs is largely vetiically differentiated, with superior
varieties produced in high-income countlies. In Chapter III, I find that international
outsourcing is a one of the main drivers of the growing wage gap during the 1990s and
that these findings are largely obscured if one uses the old proxies of outsourcing. In
Chapter IV, I reveal that input supplying countries with good quality of legal systems and
thick supplier markets specialize in the production of inputs that are more specialized.
The key finding of my work is that there is not sufficient evidence to conclude that the
patterns of trade in intennediate goods are qualitatively different from trade in finished
goods. However, the impact of the two forces on the U.S. economy is very different.
3
4CHAPTER II
STRUCTURE OF U.S. TRADE
11.1. Introduction
Various evidence indicates that the rapid growth in trade over the last several
decades is driven, to a large extent, by a dramatic increase in the offshoring of
intennediate inputs production and finished goods assembly. A number of studies search
for stylized facts on the magnitude and nature of such overseas production arrangements.
However, the results of these studies are considerably constrained by the fact that
intemational trade data do not make a clear distinction between trade in intermediate and
finished goods. In light of the ongoing political debate on the potentially adverse effects
of offshoring on the already shrinking workforce of U.S. manufacturing, the need for
such distinction continues to be relevant. This chapter provides an analysis of the patterns
of global production sharing, made possible by newly constructed dataset that isolates
intennediate goods and a portion of finished goods assembly from total U.S. trade flows.
Previous literature circumvented the absence of data on international outsourcing,
by relying on two key measures of trade in inte1l11ediate goods. The first measure isolates
trade in intermediate goods by focusing on goods described as "parts of' or "components
of'. Another measure relies on a crude assumption that economy-wide trade can proxy
5for trade in intermediate goods. It is commonly believed, however, that these measures
provide an incomplete or inaccurate view of international outsourcing. Thus, the first
measure neglects a vast array of other intern1ediate inputs, e.g., engines, semiconductors,
etc, which do not contain "parts" or "components" in their descriptions. In this chapter, I
show that it underestimates trade in intermediates by more than three-fold. The second
measure allows noise in the estimates of imported inputs, which may impute a potentially
large bias when used in a regression analysis. Finally, since the data on trade in finished
goods do not exist either, previous studies are not able to estimate the importance of
global production sharing relative to other trade. Using the three measures, previous
studies' findings are limited to observing a dramatic growth in trade in intem1ediate goods
and reorientation to include a larger number of non-traditional trading partners.
In this chapter I use a unique dataset of trade in intermediate goods that offers
considerable improvements over the previous measures of such trade. First, this dataset is
meticulously derived from detailed U.S. trade data, which span over 200 countries and
sixteen years. Second, the detivation is based on clearly defined physical and stage-of-
processing characteristics of the goods, which include both "parts", "components", and a
vast range of other intennediate goods. Third, goods are further decomposed into their
estimated end-use demand, so as to not confuse intermediate goods used in
manufacturing with repair components purchased by consumers. Finally, unlike the other
measures, the new data allow for the direct comparison of the i.ntermediate goods trade
with trade in finished goods.
6I utilize the new data on intern1ediate goods to provide a unique view on the
structure of the U.S. global production sharing. My contributions dramatically expand the
findings of the prior literature. First, I explore the overall trends in the U.S. trade in
intermediate goods with respect to overall magnitudes, commodity composition, and
cyclical behavior. My findings indicate that contrary to the common speculation, the
composition of trade, when measured by import volume shares, has changed little over
the period of 1989-2004. However, the content of intermediate inputs in U.S. imports is
higher than previously thought and is similar in magnitude to that of consumer goods.
Thus, manufacturing materials comprise roughly a third of total U.S. import volumes,
while over 70% of products imported in the U.S. are purchased to some extent by U.S.
manufacturing for use as intennediates. Next I find that while the commodity
composition of materials impor1s remained relatively constant over time, material imports
are rapidly gaining imp011ance relative to U.S. output in a number of key industries. For
example, the impor1s of computer and electronics, primary metals, and electrical
equipment materials imp011s more than doubled relative to U.S. output, constituting 23%,
30%, and 15% of output of respective industries in 2004. Finally, my findings provide
reasonable evidence to suggest that imports of manufacturing intermediates are more
prone to fluctuations in business cycles, than imports of consumer goods.
Next, I use data on materials imports to put to test two alternative predictions of
the Heckscher-Ohlin theory of trade. The theory implies that countries with different
relative endowments specialize in either distinct sets of products or distinct varieties of
ver1ically differentiated products. To examine whether countries specialize in distinct sets
7of materials products, I group countries in regions and income categories that may better
reflect endowments distributions across U.S. trading partners. I then check whether some
U.S. materials imports and products are sourced predominantly from specific
regionlincome groups. Similar to previous studies of total trade pattel11s, I find little
evidence of across-product specialization for trade in intellnediates. Next I use data on
detailed unit values of materials imports to proxy for differences in vertical
characteristics of differentiated products. Using standard panel estimation techniques, I
find a positive relationship between unit values and countries' per capita GDP. This result
implies that countries use their skill/capital endowment advantage, proxied by higher per
capita GDP, to produce vertically superior varieties.
The approach taken to identify within-product specialization of trade is
complementary to that of Schott (2004). Schott (2004) sets out to test the two predictions
of the Heckscher-Ohlin theory of trade using detailed U.S. impOlis unit values for
1972-1994. He finds a positive relationship between countries' endowments and U.S.
import unit values. My analysis is different from that of Schott in that I examine the
extent ofintemational specialization for U.S. imports ofintennediate inputs and for a
more recent period of 1989 to 2004. Additionally, I examine specialization patterns across
selected industries, durable and non-durable manufacturing, and total manufacturing
impOlis. Compared to Schott's results for 1980s, my estimates indicate that the
importance of within-product specialization increased substantially in the 1990s.
The remainder of this chapter is structured as follows. Section II.II documents the
relevant existing empirical research. Section II. III provides a detailed description of my
8dataset. Section II.IV establishes some stylized facts on trade in intermediate goods.
Section II.V explores international specialization of production. Section II.VI concludes.
11.11. Existing Measures of Global Production Sharing
A number of studies attempt to establish stylized facts on the extent of global
production sharing in the forn1 of international outsourcing of manufactUling or
processing of inteffi1ediate inputs and assembly of finished goods. Until recently,
however, data constraints have prevented researchers from gauging the full scope of
international outsourcing, as the existing trade data do not differentiate between trade in
intermediate and finished goods. Consequently, previous research resOlis to crude
estimates of such trade of which three measures stand out the most: 1) trade in parts and
components, 2) proxies based on input-output relationships; and 3) other, i.e. processing
trade. It is commonly believed, however, that these estimates either capture only a subset
oftrade in intermediate inputs, are limited to only a number of countries, and/or fail to
capture the true magnitude of trade altogether. Nevertheless, the explosive growth of
outsourcing during the recent decades suggested by these estimates convey the growing
importance of global production sharing. In the survey below I investigate the current
approaches to decomposing intemational trade into relevant components, the results of
these efforts, and their limitations.
11.11. J Trade in Parts and Components
The most common approach to assess the trade in intem1ediate inputs or global
production sharing is to look at trade in parts and components. This approach was
9pioneered by Yeats (2001), who brought attention to the changes in the SITe system of
trade classification, which greatly expanded the number of product groups identified as
"parts" and "components". One limitation of this approach is that the coverage of these
items is mostly limited to the machinery and transport equipment sector of trade (SITe
7). Another shortcoming is that this approach limits intennediate trade only to that
containing "parts of' or "components of' in the product description. Thus, stylized facts
derived from this approach largely discount global production sharing of other vital
manufacturing sectors, e.g. computers and electronics, and omit a large array of other
processed inputs in machinery and transpOli equipment sector, e.g. internal combustion
engines (Kaminski and Ng 2005).
Despite these shortcomings, previous investigation of trade in parts and
components provide some indication of the patterns and explosive growth of global
production sharing. Thus, studies suggest that, while cross-border fragmentation of
production initially began as North-North trade, it is rapidly transitioning into trade
between the developed and developing countries. For example, in the 1980s and the early
1990s, the U.S. and Japan were the largest exporters of transport and machinery
components and parts in the world, both in total dollar value and as a share of their total
exports, where a large portion of this trade took place between these countries (Yeats
2001). However, the U.S. and Japariese shares of total world exports of parts and
components declined from 22% and 16% in 1987 to 16% and 11 % in 2003, respectively,
while East Asia's share grew from 8% in 1987 to 25% in 2003 (Kimura et al. 2007). A
similar upward trend has been documented for the transition economies of Central
10
Europe (Kaminski and Ng 2005). The recipients of parts and components expOlis of East
Asia and Central Europe are not only North America, Japan, and Western Europe, but
also an increasing number oflow and middle-income countries (Yeats 2001; Kimura et al.
2007).
II.!!.2 Input-Output Tables and Trade
An alternative approach to estimating trade in intermediate goods combines data
on total impOlis with data from input-output tables to determine the extent of an
industry's purchases of intennediate inputs from overseas suppliers. This measure was
originally proposed by Feenstra & Hanson (1996) and, for each industry i, is constructed
as follows: I
~[ . ~~o~ '1
.i....J purchases of intern!. inputsiH . J . ,
) J dom.output)+lInportsJ -exports j
(11.1)
where subscript} refers to an industry supplying input} to industry i, where i,j = J, ...N.
Each product tern1 in equation (II.1) is interpreted as industry i's estimate of imported
material inputs from industry j. The measure in equation (II. 1) is generally represented as
a share of industry i's total expenditure on non-energy intennediates to arrive at industry
imported input share.
Equation (II.1) uses thejth supplier's total import share in total domestic supply to
estimate how much of the ith sector's input purchases are due to imports. The underlying
assumption that total import share is a reasonable proxy for estimating the import share of
intennediate inputs may be flawed. At high levels of supplier industry aggregation at
This formula first appears in Feenstra and Hanson (1996), but has been originally used by the BEA in
construction of imported input purchases for the Import Matrices.
11
which these measures are commonly constructed, total imports and total domestic supply
encompass imports and output of both intermediate and non-intennediate goods. Then the
the import share in domestic supply of all goods used in the numerator of equation (ILl)
may in fact over or underestimate the impOli share in domestic supply of only
intennediate goods. As a result, the measurement error introduced in equation (II.l) may
be potentially very large. I discuss the extent of the measurement error in the next section.
A number of studies use the Feenstra and Hanson (1996) measure of imported
intemlediates to determine the extent and characteristics of vertical fragmentation of
production(e.g. Campa and Goldberg 1997; Feenstra and Hanson 1999,2001). Their
findings indicate that the use of imported intermediates has increased in many industrial
countries since the 1970s. For example, U.S. impOlied intennediate inputs, expressed as a
share of total non-energy intennediates purchases, nearly doubled from 6.5% in 1972 to
11.6% in 1990. On the other hand, Canada and Japan are shown to outsource over 20% of
their total materials purchases to overseas suppliers in the 1980s and early 1990s (Campa
and Goldberg 1997; Feenstra and Hanson 1999,2001). Additionally, the value of
imported intermediates embodied in expOlied goods are shown to have accounted for
30% of the growth in the overall export GDP share between 1970 and 1990 and that it
grew by about 40% between 1970 and 1995 for ten OECD countries when measured
relative to exports (Hummels et al. 2001,2003).
12
11.11. 3 Other Measures
The two measures described above are the primary measures of trade in
intennediate goods used in the literature. There are a number of studies, which focus on a
subset of trade in inte1111ediates, which involves intennediate goods that are imported
(exported) for processing and later are exported (imported) back to the country of origin.
To measure "processing trade", studies either examine trade under special tariff
provisions, which exempt inputs imported for processing from custom duties, or proxy
for such trade by looking at the imported intermediate input content of exports. 2 These
measures are heavily limited in the scope of their country, commodity, and year
coverages (Feenstra et a1. 1998, Chen et al 2005). Additionally, with the wide-spread
adoption of free trade agreements, some special tariff provision are becoming obsolete
(Yeats 2001). For example, U.S. processing reimports declined from 12.2% in 1990 to
8.5% in 1995, where much of the decline is likely to be attributable to the producers'
failure to claim the tariff provisions after the introduction ofNAFTA (USITC 1996;
Feenstra et a1. 1998).
Il.Jl.4 Key Implications
The country, sector, and commodity-level restrictions and shOJi time-spans that
characterize the current data used to estimate the extent of global production sharing
considerably limit the information that is available to us about the current state of global
production sharing. The studies that use these data are able to identify only two primary
2 For studies of "processing trade" under special tariff provision see USITC (1996); Feenstra et al.
(I 998), Egger & Egger (2005), Swenson (2005). For studies of proxies of "processing trade" see
Hummels et al (2001), Yi (2003), Chen et al. (2005)
13
trends of trade in intennediate goods. These trends characterize, for the most pmi, only
trade in the machinery and transportation equipment sectors. First, they show that trade in
intennediate goods increased dramatically over the past several decades, specifically,
dUling the 1980s and 1990s. Second, there is an increasing reOlientation of the developed
countries' trade in inten11ediate goods away from their traditional Western suppliers.
In this chapter I expand our understanding of the extent and characteristics of U.S.
trade in intennediate goods by relying on new data on trade in intennediate goods. These
data significantly improve on the measures of trade discussed above in that they span
sixteen years, a comprehensive set of imported manufacturing commodities, and over 200
U.S. trading partners. Using these data, I uncover new dimensions of global production
sharing in regards to the magnitude and composition of trade, the characteristics of source
countries and commodities.
11.111. Data Description
This chapter exploits a new dataset, which links a recently constructed Market
Structure Index ofHTS Imports (Imports Index) with data on detailed U.S. import
transactions. I discuss the dataset construction and sources and compare the new data to
existing measure of trade in intennediate goods below.
11.JJI.l Dataset Construction and Sources
The Imports Index classifies detailed U.S. manufacturing imports into
manufacturing materials, non-manufacturing supplies, capital goods, and consumer
14
goods. The Imports Index is constructed from a number of official government sources.
The first is a dataset of all U.S. manufacturing import products classified according to the
ten-digit coding of the Hannonized Tariff System of the United States (HTS), and
maintained by the U.S. International Trade Commission. These data cover all
manufacturing products that crossed into the U.S. between 1989 and 2004 inclusive.
The detailed descriptions of HTS products identify the physical characteristic of
products and their stages of processing, or the related industry that has use for the good.
These descriptions allowed me to classify imported products by their final destination
markets, e.g. manufacturing materials. Nevertheless, a large share of products are implied
to serve multiple final destination markets, i.e. manufacturing materials and consumer
goods. I verify the accuracy of the implied final destination markets against three existing
indexes of U.S. domestic and imported production. The first index is the Federal
Research Board (FRB) Market Structure of Industrial Production Index, which classifies
detailed domestic industrial production into manufacturing input, non-manufacturing
input, capital, and consumer end-use markets. The second source is the BLS Stage of
Processing Index, which classifies the detailed manufacturing commodities into various
stages of processing, i.e. crudes, intermediates, consumer goods, and capital goods. The
final data sources are the BEA Imp0l1 Matrix and Input-Output table for 1997, which
provide data on imported and domestic input purchases by industry. I use the input
purchases data from the BEA Import Matrix to assign relative importance weights to the
HTS imports that serve multiple markets. (See Appendix A for full description of the
imp0l1s index and its methods of construction.)
15
As the result of these effOlis, the Impolis Index classifies impoli products by four
final destination markets: manufacturing materials, non-manufacturing supplies, capital
goods, and consumer goods. Manufacturing materials consist of (1) goods that are
incorporated into final goods produced by a manufacturing industry and (2) those that are
used during the production of final goods3• The first category of materials incorporates
intennediate inputs into non-durable manufactming, e~g., flour, vegetable oils, wood
pulp, wood logs, industrial chemicals, plastics, and textiles, and materials for durable
manufacturing, e.g., metal mill products and parts and components of machinery and
equipments. The second category of materials are intennediate inputs that complement
the production of final goods, e.g., processed fuel and lubricants, packaging materials,
some administrative supplies, and others.
Non-mam!facturing supplies are defined as inputs into the non-manufactming
sector; i.e., construction, agriculture, utilities, and other industries. For example,
construction supplies include building lumber, plywood, millwork, glass, plumbing
fixtures, etc, while agricultural supplies consist of feeds, processed fuels, machinery
repair parts, etc.
Capital investment goods consist of products that are used to manufacture or
transport other goods in the manufacturing sector and include goods, such as machine
tools for cutting and stamping metals, other specialized machinery (such as fa1111
machinery and textile machinery), heavy trucks, ships, and boats. In addition, this
grouping includes non-manufacturing industry and non-defense related government
3 Intermediate goods are also commonly referred to as industrial materials or just materials (FRS).
16
products, such as computers, office furniture, and heating equipment, that are used in the
operation of businesses. Defense-related government investment such as military
weapons and transportation equipment are also included in this category.
Finally, consumer goods are defined as nondurable goods and durable goods
purchased by consumers and defense- and non-defense related government supplies4•
Examples of these goods include such nondurable items as foods, children's apparel,
prescription drugs, gasoline, home heating oil, and residential electric power and durable
items as passenger cars, light trucks, household appliances, and home electronic
equipment, to name a few. On the other hand, examples of defense and non-defense
related government supplies include ammunition, repair paI1s for military equipment and
machinery, education-related products, office supplies, and repair pm1s for non-military
equipment and machinery owned by the government.
As noted above, a large p0l1ion of the la-digit HTS products are classified as
serving multiple final destination markets. Table B.l illustrates the shares of imported
manufacturing materials and consumer goods according to their assigned utilization
weights. Thus, Columns I, II, and lII, refer to those imp0l1ed materials, which have final
destination market weights of at least 25%, 50%, 75%, and 100%. As can be seen, only
16% of manufacturing materials products, which comprise roughly 23% of total materials
imp0l1 volumes, are used by U.S. manufacturing alone. However, over 60% of materials
products, which comprise roughly 84% of total materials import volumes, are utilized
4 Consumer nondurables consist of items with a shelf life ofless than three years and that are ready for
final demand. Consumer durable goods include products that have it much longer shelflife than
nondurables (SOP).
17
predominantly as manufacturing materials (rate of utilization is larger than 50%). On the
other hand, less than 0.5% of consumer goods serve the consumer markets alone. The
figures for consumer goods imports show, that consumer goods are generally also
classified as non-manufacturing suppliers (e.g., paper), capital investment goods (e.g.,
computers), and/or manufacturing materials (e.g. tires).
In this chapter, I place my focus on trends in U.S. trade in manufacturing
materials, but contrast them with the trends of trade in consumer goods. As I reveal in the
following section, the two types of trade components represent the largest share of the
U.S. imports. I link the ImpOlis Index to detailed U.S. trade transactions from Feenstra
(2002) and U.S. Census (2005) to derive data on import volumes of intermediate and
non-intemlediate goods for the period of 1989-2004. I then use these data to derive
stylized facts of the trade in intennediate goods, with a particular focus on differences
across region and industries. For industry-level statistics, I aggregate data up to 3-digit
industry classification according to the North American Indushial Classification System
(NAICS). While NAICS was not introduced until 1997, the U.S. Census provides a
NAICS-HTS concordance for HTS codes going as far back as 1989.
Next, I compare the new data on inputs trade with the previously used measures
of such trade and reveal potentially severe measurement elTors in the old measures.
11.111.2. New Data Versus Old Measures
In my comparison of new and old data on offshoring of intermediate inputs, I
focus on the measure of trade in intermediates defined by "parts" and "components"
18
descriptions, and the measure of imported inputs originally proposed by Feenstra and
Hanson (1996). These measures are described in Sections lULl and lUL2, respectively.
Figure B.I illustrates the differences between the new data on imports of
intermediate goods and the old data on imports of goods labeled as "parts" and
"components". The classification of "parts" and "components" was kindly provided by
Schott (2004), and includes a selection of 1989-2001 ten-digit HS codes, which contain
these words in their description. As can be seen, the differences between the two
measures are very distinct and large. The volume of U.S. trade in "parts" and
"components" underestimates by more than three times the total volume of trade in
intermediate goods.
Next, I tum my attention to the second measure of import of intennediate inputs
originally proposed by Feenstra and Hanson (1996) and shown in equation (II. I ). This
measure is useful for assessing the extent to which each domestic industry impOlis
intennediate inputs, which is not identified in raw imports data. As discussed in section
lLII.2, equation (11.1) employs an industry's total import share in totaJdomestic supply to
proxy for the industry's share of imports of inputs in the domestic supply of inputs. If for
some industries, the total import share includes data on both intem1ediate and non-
intennediate goods, the measure of imported inputs in equation (II. 1) may be driven by
variation in the import share ofnon-intennediates. Using the new data on imports of
intennediate goods, I am able to refine the original measure and derive imported inputs of
each industry i as:
19
~ l .. . ] interm. imports (II 2)
L. jJurchasesof znterm. znputs j ·r . .. !. l'j / znterm. dam. output j +mterm. Imports j - znterm. exports}
where as before SUbSCliptj refers to an industry from which industry i purchases its
intennediate inputs, where iJ = 1, ...N. The measure in equation (II.2) differs from the
original measure of imported inputs by the right tem1 of the numerator, where the total
impOli share is replaced by import share of intennediate inputs.
To anive at the two measures of imported inputs, I combine data on imports with
data on inputs purchases. The inputs purchases are obtained from U.S. input-output tables
provided by the BEA. The industries in input-output tables are classified on the three-
digit Standard Industrial Classification (SIC) basis during 1989-1996 and four-digit North
American Industrial Classification System (NAICS) basis during 1997-2004. I aggregate
the imports data up to three-digit SIC and four-digit NAICS industries, using the HS-SIC
and HS-NAICS concordances provided by U.S. Census. I then calculate the original and
the refined measures of imported inputs in equations (ILl) and(II.2), respectively, and
express them as shares in total non-energy materials purchases. Figure B.2 illustrates the
movements in manufacturing weighted averages of imported input shares during
1989-2004, where the discontinuity in the graphs identifies the switch from SIC to
NAICS. The differences between the two measures are distinct, although not as large as
those shown for the first measure of trade in "parts" and "component". The differences
are most pronounced when examining on a detailed industry level, not reported here.
20
In summary, the comparison of the new data on trade in intermediate goods with
the previously used measures of such trade suppOlis the literature's suspicions on the
accuracy ofthe old measures.
II.IV. Magnitudes~ Composition~ and Cyclicality
In this section, I characterize U.S. imports of manufacturing materials along
several dimensions. First, I examine the overall magnitudes of such trade over the period
of 1989-2004, with a focus on import volumes, number of traded products, and their
respective growth rates. Next, I examine the composition of materials trade by industry
and highlight recent trends of primary traded commodities. Finally, I identify a distinct
pro-cyclical behavior of U.S. materials imports, which is distinguished fi'om that of other
components of U.S. trade.
ll.lVl Overall Magnitudes
The relative composition of trade is speculated to have changed over time in favor
of intenl1ediate goods. Figure B.2 utilizes the new data to reveal the relative composition
of U.S. imports during 1989-2004. The imports of materials and consumer goods appear
to maintain a roughly similar volume and growth during 1989-2000. The early 2000s saw
the trends in the two types of trade diverge, wherein materials imports declined, while
consumer goods imports continued to grow roughly at the same rate. Table B.2 shows
that materials imports nearly tripled in size, from 131.7 in 1989 to 351.9 billion U.S.
dollars in 2004, which corresponds to roughly a third and 29% oftotal imports in 1989
and 2004, respectively. Additionally, the economic significance of materials trade
21
continued to grow, as the share of materials imports relative to total manufacturing output
increased from 4.8% to 8.4%.
The composition of imported products also changed little over time. Table B.2
shows that the number of products used as manufacturing materials increased from 8497
in 1989 to 10615 in 2004. However, their share in total import products increased only
from 71.2% to 73.9%, respectively. The number of import products used as
manufacturing materials grew roughly at an average rate of 1.5% per year, slightly ahead
of the 1.2% growth of consumer products.
These findings indicate that, contrary to the common speculation, the composition
of trade, when measured by import volume shares, has changed little over the period of
1989 to 2004. However, the content of intermediate inputs in U.S. imports is higher than
previously thought and is similar in magnitude to that of consumer goods.
Il.lV.2 Commodity Composition
In this section I examine which commodities are ofplimary importance for trade
in manufacturing matelials and whether their relative standing has changed over time,
both with respect to other commodity imports and U.S. domestic output. First, I illustrate
the relative imp011ance of materials imports in U.S. imports by three-digit NAICS
commodities. Figure B.3 shows that materials imports comprise a significant share of
most major imported commodities. Exceptions to these are Food/Beverage/Tobacco (311
& 312), Apparel & Leather, & Allied (315 & 316), Printing (323), Fumiture (337), and
22
Miscellaneous (339) industries, which tend to to be non-manufacturing supplies- or
consumer goods-heavy.
In Table B.3 I report statistics of market shares of top commodities of materials
and consumer goods imports in their respective total imports. The commodity
composition of materials imports remained relatively constant during 1989-2004, with
computers and electronics and transportation equipment commodities topping the list.
Additionally, computer and electronics matelials imports are the only category of imp011s
that show a significant growth in market share, from 19% of total materials import in
1989 to 24% in 2004. In the consumer goods markets, the composition of imports has
also remained relatively stable over time. The apparel, leather and related products and
transportation equipment imp011s remain the most heavily demanded consumer goods
from overseas, although their market shares declined by 4% and 3% during 1989-2004.
On the other hand, the market share of consumer chemicals more than tripled in
magnitude, from 4% of total consumer goods import in 1989 to 13% by 2004.
Furthel111ore, it appears that the relative importance of materials and consumer
goods with respect to U.S. output has increased significantly for many 3-digit NAICS
commodities. For example, materials imports of computer and electronics, primary
metals, electrical equipment, machinery, and fabricated metal products have roughly
doubled their share in U.S. output of the respective industries during 1989-2004. These
trends are even more pronounced for consumer goods imports. For example, apparel,
leather and related consumer goods imp011 share grew from 50% of U.S. output in 1989
to a striking 258% of the output in 2004.
23
The key point to take away from these initial findings is that the commodity
composition of materials imports remained relatively constant over time, with the
exception of the computer and electronics industry which gained further dominance as a
leading source of materials imports. Despite relatively little change in their composition,
however, material imports in most commodity groupings have increased significantly
relative to U.S. output, with some industries more than doubling their relative share.
I1.1V3 Cyclicality
The period of 1989 to 2004 covered in my sample contains the longest U.S.
business cycles ever recorded. According to the NBER's Business Cycles Dating
Committee, the 1991-2001 business cycles lasted for 128 months, when measured from
trough to trough (NBER). Figures B.2 and B.3 illustrate the volatility of imports over
time. It seems that U.S. material imports are perhaps more volatile with respect to
fluctuations in the U.S. economy than imports of consumer goods. For example, during
the troughs of 1991 and 2001, materials imports grew at negative rate of 1.49% and
12.51 %, respectively. During the same years, however, imports of consumer goods
experiences a positive a positive rate of 1.41 % and a negative rate of 0.54%, respectively.
The correlations and simple regression analyses shown in Table B.4 shed some
light on the sources of imports volatility. In these tables, I use real U.S. sectoral output
and real GDP as measures of manufacturing-wide and economy-wide business cycles.
Output data is obtained from the BEA and covers 18 three-digit NAICS industries. The
correlations shown in Table BA reveal that U.S. imports of intermediates are more
24
correlated with fluctuations in the manufacturing output, rather than economy-wide
fluctuations. The opposite is true, however, for imports of consumer goods.
To explore this issue more rigorously, I proceed to estimate the elasticity of the
response of impOlis to business cycle fluctuations, as
.d In (Imports g .il) == f3 0+ f3 1.d In (Output if) + f3 2 Mat g *.d In (Output il ) +E g .il , (IL2)
where Imports g.il are imports of materials or consumer goods, deflated by the CPI
deflator, sourced from industry i at time t, Output il is real output of U.S. industry i, and
Mat g is a variable that takes a value of 1 if imports are manufacturing materials and 0
otherwise. Column I ofTable BA reports the estimated elasticity of response of imports
to output fluctuations. The coefficient on the materials dummy interaction is positive and
statistically significant, indicating that changes in materials impOlis are more elastic with
respect to changes in business cycle fluctuations, relative to changes in consumer goods.
In Column II, I include changes in national GOP instead of the sectoral output. The
coefficient on the materials dummy interaction is statistically insignificant in this
specification.
The findings described above provide evidence to suggest that imports of
manufacturing intermediates are more prone to fluctuations in U.S. manufacturing rather
than economy-wide business cycles, and more so than impOlis of consumer goods. This
is consistent with what is known about the firms' response to business cycles, where
25
industries forecast changes in demand by slashing of their inventories in times of
recessions and increasing them during recovery.
ILV. International Specialization of Production
In this section of the chapter, I use data on materials import to put to test two
alternative predictions of the Heckscher-Ohlin theory of trade. The theory implies that
countries with different relative endowments specialize in either distinct sets of products
or distinct varieties of vertically differentiated products. Thus, the first prediction of the
Heckscher-Ohlin theory would imply, for example, that the labor-abundant Philippines
export labor-intensive apparel, labor and capital abundant Ireland exports labor- and
capital-intensive chemicals, while capital abundant Japan focuses on capital-intensive
machinery. On the other hand, given the same set of countries and a hypothetical product,
such as a television set, the second prediction would imply that Philippines exports
televisions made with color tubes, Ireland exp011s television sets made with rear-
projection, and Japan exports television set made with plasma displays, given their
relative endowments and relative sophistication of the television production technologies.
To examine whether countries specialize in distinct sets of materials products, I
group countries in regions and income categories, that may better reflect endowments
distributions across U.S. trading partners. I then check whether some u.s. matelial
imports and products are sourced predominantly from particular region/income groups.
Similar to previous studies of total trade patterns, I find little evidence of across product
specialization for trade in intelmediates. Next I use data on detailed unit values of
26
materials imports to proxy for differences in vertical characteristics of differentiated
products. Using standard panel estimation techniques, I find a positive relationship
between unit values and countries' per capita GDP. This result implies that countries use
their skill/capital endowment advantage, proxied by higher per capita GDP, to produce
vertically superior varieties.
Other studies testing the predictions of the Heckscher-Ohlin theory find scant
evidence in favor of endowment-driven trade at either the industry level (e.g., Bowen et
al. 1987, and Trefler 1995) or detailed product level (Schott 2004). On the other hand,
Schott (2004) perfoTI11s an empirical test of the second prediction on detailed U.S.
impOlis unit values during 1972-1994 and finds a positive relationship between countries'
endowments and U.S. import unit values. The approach taken in this section is
complementary to that of Schott (2004). My analysis is different from that of Schott, in
that I examine the extent of international specialization for U.S. imports of intermediate
inputs and for a more recent period of 1989 to 2004. Additionally, I contrast the extent of
intemational specialization of trade in intennediate goods to that of trade in consumer
goods. Finally, I examine specialization patterns across selected commodities, i.e.,
computer and electronics, transportation equipment, chemicals, machinery, and electrical
equipment, durable and non-durable products, and as a whole.
II. Vi Across-Product Specialization by Jmport Share
In an attempt to reveal trends in specialization across materials products, I first
explore the impOli shares of U.S. trading partners across products. Large asymmetries in
27
impOli shares across products should serve as a sign for across-product specialization. I
find these asymmetries to be prevalent in some materials producing industries more than
others, pointing to some across-product specialization in materials trade. At the same
time, however, I find little suppOli of specialization across consumer goods products.
To facilitate the comparison of trading partners, I make use of the country-region
assignments provided in Table B.S. Three aspects of how countries are assigned to
regions deserve mention. First, Latin America includes all of the countries of Central and
South America, excluding Mexico. Second, I define ASEAN as its current 10 member
countriesand includes South Korea & Bhutan. Third, OECD comprises of its 18
founding members, excluding Canada and the U.S., and includes the more recent
members Finland, Australia, and New Zealand. I exclude Mexico, Canada, and Japan
from OECD, as I intend to keep them as independent categories. Additionally, I define
China as China (mainland), Hong Kong, Taiwan, and Macao. Finally, the OTHER
category consists of the remaining countries. The resulting set of countries is intended to
capture regions, according to a more unifoll11 mix of wageslincome levels and common
cultural characteristics. The Other category serves as an exception, since it groups high-
income Israel with low income India.
Table B.6 reports the U.S. market share of U.S. major trading partners in te1111S of
import value by industry, for the first and last years of the sample. A partner's market
share by industry is calculated as the sum of U.S. imports from that industry and region
as a share of U.S. total imports within the industry. At first glance, it appears that more
than half of U.S. imports of intermediate goods is supplied by the world's most developed
28
economies, Canada, Japan, and OECD, although this share decreased from 70% in 1989
to 51 % of total manufacturing impOlis in 2004. The less developed Mexico and China are
the primary source countries gaining from the loss of the market share of the traditional
partners. The market shares of Mexico and China are roughly equal and, when combined,
increased from 12% in 1989 to 26% of total U.S. manufacturing imports in 2004. These
trends come in contrast to those of consumer goods. First of all, Canada, Japan and
GECD demand a much smaller share of U.S. impOlis of consumer goods that is only 50%
in 1989 and 42% in 2004. All of the loss in the combined market share, in fact, stems
only from Japan. Furthermore, Mexico barely competes with China for a share in U.S.
imports, sourcing 8% and 25% of total U.S. materials imports in 2004, respectively. In
the end, nearly 50% of consumer goods are sourced from China and OECD, and these
countries appear stable in their positions as equal leaders.
At second glance, Table B.6 reveals some heterogeneity in market shares of
trading partners of materials across industries. For example, it appears that imports of
nondurable intennediates are heavily concentrated in the hands of Canada and GECD.
Both of these countries are U.S. leading sources of chemicals matelials during
1989-2004. In durable manufacturing, Japan and OECD continue to supply the U.S. with
the majority of machinery intermediates, while neighboring China and ASEAN are taking
the leading positions in sourcing computer and electronics paIis and components. By
2004, Mexico takes the lead in the U.S. imports market of electrical equipment materials.
At the same time, however, transportation equipment intermediates are sourced roughly
equally from Canada, Japan, Mexico, and GEeD. In summary, these findings suggest that
29
across-product specialization is more prevalent in some materials producing industries
than others.
The patterns of specialization of materials imports are quite different from those
of consumer goods. With the exception of the transportation equipment, China and
OECD appear to dominate the market of both durable and nondurable consumer goods
impOlis. Furthennore, OECD's leadership continues to grow at the expense ofASEAN
countries and China's leadership at the expense of Japan in the nondurable and durable
consumer goods markets, respectively.
To shed more light on the U.S.'s most dynamic trading partners, Table B.7 reports
the countries with top ten absolute changes in imports market share between 1989-2004.
China and Mexico top the lists in both materials and consumer goods imports.
II. V2 Across-Product Specialization by Product Share
I examine intemational specialization across materials products further by
exploring differences in import product penetration of U.S. trading pminers. Each cell in
Table B.8 reports the percentage of products in each industry imported in the U.S. from a
region. Regional penetration is 100% if every product in the industry is sourced from at
least one country in the region and 0% if no country in the region sources any of the
industry's products to the U.S. One would expect that product penetration should not be
high by each region, since each region should specialize in a set of goods.
30
As indicated in Table B.8 intermediate imports product penetration by OECD is
over 90% and by other countries/regions is between 40% and 80% during 1989 and 2004.
The same pattern emerges for product penetration of consumer goods imports. However,
both Japan and OECD have experienced slight declines in the product penetration
between 1989 and 2004, while the less developed U.S. trading partners are seeing an
increase in their product penetration. Table B.9 ranks countries with the biggest absolute
gains in penetration between 1989 and 2004. The same countries that had the highest
absolute change in market share top the list for highest absolute change in product
penetration. The high product penetration of countries/regions reported in Table B.8
poses fmiher evidence against the implications of the Heckscher-Ohlin theory that
countries specialize in a distinct mix of products (Schott 2004).
Increases in import market share occur through increasing imports of incumbent
products and an increase in the number of products imported. I decompose growth in
imports of intermediate goods into those parts that are attributable to growth of imports of
the continuously produced goods (intensive margin) and growth of imports from the entry
and exist of new products (extensive margin). Table B.1 0 shows the results of the
decomposition growth of U.S. impOlis from each source country/region for overall
manufacturing and by industry. In contrast to previous tables in this section, I use eight-
digit HS codes rather than ten-digit HS codes in the decomposition. This is due to the fact
that the HS code classification has undergone many revisions over the period of
1989-2004 due to both methodological and tariff schedules changes, which may assign
the same commodities different HS codes. In my experience of dealing with the HS
31
codes, these changes are reflected primarily in the last two digits of the ten-digit HS
codes. Thus, restricting attention to eight-digit HS product codes may circumvent the
discrepancies in the HS classification over time at least partially.
As indicated in Table B.1 0, the relative contribution of the extensive versus
intensive margins varies across industries and import types. The extensive margin is
significantly more important for computer and electronics industry across both
intennediate and consumer goods import growth. On the other hand the extensive margin
is more important for intermediate goods impOlis and intensive margin is more important
for consumer goods imports of electrical equipment industry. The reverse is true in the
machinery sector. All in all, however, it is the intensive margin that is relatively more
important in the growth of imports of intermediate and consumer goods.
The message of the discussion above is relatively clear: when trade is divided into
thousands of products, there is little evidence over time of endowment-related
specialization across products for both intennediate and consumer goods when looking at
largest trading countries/regions with the U.S. As a final test against across product
specialization, I follow Schott (2004) and break countries into relative income cohorts to
examine the share of products sourced from low-, middle-, high-income countries at the
same time. Income levels are commonly used as an indicator of level of endowments, i.e.
skill and capital, with low and high income levels representing less skilllcapital and more
skilllcapital endowments, respectively. A large share of products sourced fi:om low and
high income countries levels at the same time should serve as the final test against across
product-driven specialization.
32
I use per capita GNI (pcGNI) data from World Bank classification to group
countries in relative income cohorts. I also show whether results are sensitive to the use
of alternative relative classifications of income levels, e.g. low income countries are
defIned if country income is below 40% income percentile relative to world income
distribution. Next, I classify imported products according to the source country income
classification. Low (L), Middle (M), and High (H) products originate solely in low-,
solely in middle-, or solely in high-income countries, respectively. Products are Low and
Middle (LM) or Middle and High (MH) if they are sourced simultaneously from at least
one country of each type. Finally, a product is Low, Middle, and High (LMH) if it is
sourced from at least one low and at least one high-income country. I exclude China from
the analysis, as its inclusion significantly inflates the contribution oflow income
countries.
Table B.11 reports product share by source country income groupings according
to various income breakdowns. The share of import products sourced hom only high-
income countries and the share of import products sourced from both middle and high
income countries is diminishing during 1989-2004. At the same time, however, the share
of import products sourced simultaneously from at least one low and one high-income
country is increasing rapidly during 1989-2004. As a further robustness check, I exclude
LMH products sourced from just one low-wage country, which are indicated by a star in
Table B.11. The fact that the share of products sourced from the LMH countries increases
regardless of the income breakdowns is remarkable, as mentioned in Schott (2004) for
total U.S. imports.
------.- .._-----
33
The results presented in this section offer compelling evidence against
intemationa1 specialization across products during 1989 and 2004.
II. V 3. Within-'Product Specialization by Unit Values
I now tum my attention to the altemative prediction of the Heckscher-Ohlin
theory of the importance of within-production specialization. To test the prediction, I
examine whether there is a positive relationship between unit values of U.S. imports of
intermediates and source country per capital GDP. Unit values are measured as import
volumes divided by import quantity. I use real per capita GDP from the World
Development Indicators (2007). Following Schott (2004), I regress log unit values of
U.S. imports ofmanufacturing materials on source country log per capita GDP across
ten-digit HS products, source country, and year for all manufacturing imports during
1989-2004.
10g( uvg.) =(X g +(XI +(X/" + [3 *log (GDPpcCI )+ Eg . C1 (IlA)
where uv gCI is unit value of materials or consumer goods, sourced from country c in
year t, GDPpc CI is country CiS real per capital GDP in year t, and (Xg, (XI' and (X/" are
product, year, and region fixed effects. In an altemative specification, I pool the unit
values for materials and consumer goods, and regress equation (II.4) with an interaction
term of the consumer goods dummy with per capita GDP. The latter allows me to gauge
whether the extent of intemationa1 specialization within materials products is different
from the extent of specialization within consumer goods products. I estimate these
34
differences for selected industries, as well as non-durable and durable manufacturing
imports.
The results in Table B.12 Column I show that unit values of U.S. imports
materials are positively and significantly related to countries' per capita GDP. The
coefficient in Column I implies that 10% increase in per capita GDP is associated with
1.3% increase in unit values of imported materials. Columns II-IV restrict the sample to
only those products, that are used more than 25%, 50%, 75% as manufacturing materials.
The estimates in these subsamples are of the expected signs and significant, and are larger
than the ones in the full sample. This evidence is indicative of the fact that as materials
become more specialized, intemational specialization along vertical dimensions in fact
increase. Finally, Column V reduces that sample to only those products that are sourced
simultaneously from at least one low income and one high income country. The
coefficients remain unchanged from that of full sample.
Next I break up the sample into non-durable and durable manufacturing materials
as shown in Table B.12. Additionally, I include unit values of consumer goods and use a
dummy variable to gauge whether there is a statistically significant difference between
the effect of per capita GDP on unit values of materials and that of consumer goods. The
coefficients in the regressions on each sample are still positive and statistically
significant, however the effect diminishes as the sample gets smaller. This comes in
contrast to the increasing effect found in the full sample. Furthermore, there is a
statistically significant difference ofthe effect of per capital GDP on unit values of
35
consumer goods, which in fact increases as products' utilization by final destination
markets is more concentrated.
Finally, 1 break the sample up further into selected industries, and estimate (IlA)
on each industry products samples, as reported in Table B.l3. Chemicals manufacturing
exhibits the largest coefficient of any of the selected industries and industry groupings,
implying that intemational specialization across vertical dimensions is greatest for
chemicals, compared to other industry groups. The changes in the coefficients with the
extent of utilization of chemical materials are also intuitive. When chemicals imports are
predominantly used by u.s. manufacturing (measured by more than 75% utilization rate
in Column IV), these imports include primarily crudely processed chemicals, which tend
to exhibit a lower degree of vertical differentiation. This implies a lower intemational
specialization across the vertical dimension and is reflected by a smaller estimate of the
effect of GDP on unit values in Column IV, relative to other columns. On the other hand,
chemicals imports used heavily by consumer markets (measured by more than 75%
utilization rate in Column IV), refer to consumer pharmaceuticals, which generally
exhibit a large degree of vertical differentiation. Thus, as the rate of utilization increases,
differences in the effects of GDP on unit values of chemical materials and unit values of
consumer goods also increase, as shown by the estimate on the consumer dummy. The
pattems of the effects of GDP on selected durable industries are also intuitive and
opposite of those for chemicals. Thus, intemational specialization in varieties of
machinery, computers, and electrical equipment materials becomes more pronounced as
36
inputs become more specialized and customized in nature, reflected by higher rates of
utilization in Column III.
The evidence linking unit values and income presented in this section supports an
old trade theory interpretation of U.S. trade. The results are consistent with endowment-
abundant countries using their relative endowments to manufacture vertically distinct
varieties that use those endowments more intensively.
II.VI. Conclusion
Previous attempts to shed light on the nature of trade in intennediate inputs have
been largely constrained by the fact that trade data do not differentiate between trade in
intel111ediate and finished goods. In this chapter, I introduce a newly constructed dataset
that clearly distinguishes between U.S. imports of manufacturing materials, consumer
goods, and others, during 1989-2004. With these data in hand, previous findings on the
nature of trade in intennediate inputs can be confinned, revised, and added to.
Using the new data, I reveal that the magnitude of U.S. trade in intennediate
goods is larger than previously thought, averaging roughly a third of total U.S. import
during 1989-2004. Furthennore, imports of intennediates exhibit sharp pro-cyclical
tendencies, which are distinguished from those of other imports components. Finally, I
find that while inputs are largely vertically differentiated, with superior vaIieties
produced in high-income countries, the within-production specialization is in fact more
pronounced for trade in consumer goods.
37
CHAPTER III
OUTSOURCING, TECHNOLOGY, AND U.S. WAGE INEQUALITY
I. Introduction
It has been well documented that U.S. wage inequality rose dramatically during
the 1980s, when the wages of both the most skilled and moderately skilled workers
increased and the wages of least skilled workers dropped. A large literature spans the
debate on the detenninants of this rise in the wage inequality. A common consensus
points to the on-going growth of the demand for high-skilled workers, of which skill-
biased technical change (SBTC) and international trade are the often cited sources. While
much empirical evidence supports the hypothesis of the effect of SBTC on wages, the
evidence for the impact of trade on wages is mixed. Only one U.S. study finds robust
estimates of the effect of international trade, specifically, trade in intern1ediate inputs, on
the 1980s wage inequality, and many others arrive at inconclusive evidence of the effects
oftrade.5
Surprisingly, the literature has focused almost exclusively on data from the late
1970s and the 1980s. The few studies that have examined this issue using data from the
1990s find mixed evidence on the overall patterns of wage inequality during this period
5 See Feenstra and Hanson (2003) for the survey of trade's impact on wages.
--------_ ... ----_.
38
and merely speculate on its detenninants. 6 At the same time, there is growing evidence to
suggest that both technology and trade gained further prevalence during the 1990s and
early 2000s, as finns finally learned to reap the full benefits of the computer revolution
and established extended networks with the low-wage countries.
Prior literature examining the effect of trade on wage inequality has two
shortcomings that this chapter will focus on. First, virtually all previous papers have
focused on the period of the late 1970s and 1980s, with no work examining the 1990s and
2000s. This seems primarily due to the fact that the National Bureau of Economic
Research (NBER) Productivity Database used for these studies ends in 1996.
Nevertheless, there is a general perception in the literature that the growth in wage
inequality has subsided. This calls into question how strongly trade forces may be
affecting the U.S. skilled-unskilled wage gap, since evidence suggests that the 1990s and
2000s saw a dynamic growth of trade. 1 find that this perception regarding the fall in the
wage gap within U.S. manufacturing to be false. I document a significant rise in wage
inequality in 1990s and a decline in the 2000s, which closely corresponds to the
movements of trade in intennediate inputs over the same period.
A second signifIcant shortcoming of the previous literature is its measurement of
impOIied intennediate inputs, i.e. materials offshoring. Given available data, previous
literature has used input-output relationships to detern1ine the extent of a sector's
intem1ediate inputs purchases from an input supplier. Then the suppliers' total imports
share in the U.S. supply is used to estimate how much of the sector's input purchases are
6 See the survey in Autor et al. (forthcoming 2008) and Lemiuex (2007).
39
due to imports. Thus, it is assumed that total import share is a good proxy for estimating
inputs import share. As shown in Appendix A of this chapter, this assumption introduces
significant measurement error.
I address these shortcomings in the following fashion. First, I update the NBER
Productivity Database through the year 2005. Using these data, I first document that
while the gap between skilled and unskilled workers continued to rise during the 1990s, it
fell significantly after 2000. Next, I use standard data construction techniques and
empirical specifications utilized in product-price literature to estimate the effect of trade
on the skilled-unskilled wage gap for this later period (1989-2005) and find a significant
effect of materials offshoring on the wage gap. However, this effect is not robust to the
inclusion of alternative measures of trade and computerization, which calls into question
the validity of previous findings; e.g., Feenstra and Hanson (1999) who find that
materials offshoring explains up to 25% of the rise in the skilled-unskilled wage gap for
their earlier sample covering the years, 1979-1990.
I then turn to recently constructed trade data on U.S. imports ofintennediate
goods to develop a refined measure of materials offshoring. Using the refined measure, I
find a very large and robust effect of offshOling on the skilled-unskilled wage gap of
1989-1996 and a large, albeit insignificant, effect on wages of 1997-2004. Fmihennore,
offshoring of business services appears to playa large role in the widening of the wage
gap during 1989-1996, although services offshOling contributes to the closing of the gap
during 1997-2004.
40
Other findings indicate that one must take caution in interpreting all technological
change as skill-biased. I find that computer adoption contributed significantly to the lise
in the skilled-unskilled wage gap during 1989-1996, by increasing the non-production
wages and decreasing, albeit statistically insignificantly, the production wages. On the
other hand, the estimates show that office equipment diffusion has a overall neutral effect
on relative wages, while other high-tech technological change is biased towards the
unskilled during 1997-2004. Additionally, the failure to identify the effect of computers
on the wage gap of 1997-2004 may be indicative of the diminishing role of computer
technologies in U.S. manufacturing.
This work is paIi of the growing theoretical and empirical debate on the effects of
technology and international trade on the increase in the relative demand for skill. A
plethora of studies document a striking conelation between the adoption of computer-
based technologies and the increased use of college-educated labor within detailed
industries and finns andacross plants within industries. 7 In contrast, the evidence of the
impact of trade on the demand for skill is much more conf1ic.ting. 8 A number of studies
argue that a constant trade to GDP ratio, increasing product prices ofleast-skilled
industries, and within-industry changes in labor composition of developed countries are
indicative of a relatively minor role of trade in the prediction of relative wages.9
Proponents of trade effects, on the other hand, retaliate by pointing to a rising trade to
7 Katz and Autor (1999) summarize this literature.
8 See Feenstra and Hanson (2003) survey of the literature on trade and wages.
9 See Kfilgman (1995) for a discussion of relative magnitudes of trade; Slaughter (2000) for a discussion
ofliterature on relative-price changes; and Bel111an et al. (1994) on within vs. between industry lsbor
shift.
41
value-added ratio, growing relative domestic prices, and aggregation issues of industry-
level data on labor composition. Furthem10re, recent studies argue that the growing share
of trade in intennediate inputs may shift the relative demand for skill in the same manner
as SBTC does (Feenstra and Hanson 1999,2003). Recently, however, these findings have
been called into question, as the alleged decline in relative wages during 1990s does not
appear to coincide with the dynamic growth of technology and trade of the1990s (e.g.,
Card and DiNardo 2002). One of the contributions of this work is to attempt to shed more
light on the roles of technology and trade in the changing nature of wage inequality of the
1990s and 2000s.
In addition to the contribution discussed above, this work also contributes to the
methodology of the product-price literature (see Slaughter 1999). There are only a
handful of other studies on underlying factors causing changes in prices and productivity,
which then are linked to wage changes. These studies find mixed contributions of trade-
related forces, i.e. materials offshoring, trade barriers, and transportation costs, on U.S.
wage changes of the 1970s and 1980s (Feenstra arid Harlson 1999,Haske1 and Slaughter
2003). I contribute to their methods by using more recent data for 1989-2004 and
exploring a broader set of causal factors, which include more refined measures of trade.
The chapter is organized as follows. Section III.II documents relative wages
during 1989-2005. Section III.III presents empirical methodology. Section III.IV
describes data. Section III.V presents empirical results. Section III.VI discusses
sensitivity analysis and section III.VII concludes.
42
IILII. Old and New Evidence of Wage Inequality
The rapid growth of US. wage inequality of 1980s has been well documented
within both U.S. manufacturing and for the U.S. as a whole. While no papers have
analyzed trends in wage inequality within U.S. manufacturing during 1990s and 2000s
due to data limitations, a few studies have examined the growth in relative wages using
U.S.-wide micro data. These studies find conf1icting evidence, suggesting a changing
nature of the 1990s US. wage inequality, which may not correspond to the dynamic
growth of trade and SBTC that occurred during the same period. In this section, I use new
industry-level data to document movements of wage inequality within U.S.
manufacturing over the period of 1989-2005. The new data show a significant rise in
wage inequality in the 1990s and a decline in the 2000s, which correspond to the pattems
of trade and SBTC referenced in the literature (e.g., Autor et al. 2003; Feenstra and
Hanson 2003).
Prior studies of wage inequality rely on two primary datasets, the eamings data of
workers from all U.S. industries compiled in Current Population Surveys (CPS) and the
wages of workers in U.S. manufacturing available through the NBER Productivity
Database (NBER PD). During the 1980s, these data show a significant rise in wage
inequality. According to the CPS data, between 1979 and 1989, the real wages of workers
with sixteen or more years of education rose by 3.4%, of full-time workers with twelve
years of education fell by 13.4%, and of workers with less than twelve years of education
fell by 20.2%.10 Within U.S. manufactUling alone, the total wages of nonproduction
lOA detailed discussion of basic facts concerning wage movements in the U.S. during 1980s is provided in
Katz and Autar (1999).
43
workers relative to production workers rose by an average of 0.72% per year over the
period of 1979-1990 (Feenstra and Hanson 1999).11
The early 2000s saw a rise in the studies of wage inequality of 1990s, which paint
a mixed picture ofthe changing nature of U.S. wage inequality and the sources of these
changes. For example, Card and DiNardo (2002) explore CPS data and find no noticeable
change in wage inequality between 1988 and 2000. This finding leads them to question
the validity ofthe previously estimated effects that SBTC and trade forces have on wage
inequality during 1980s. On the other hand, Autor et a1. (forthcoming 2008) use similar
data for 1989-2005 to show polarization in wages, where the wages in very low and very
high skill occupations increased, while those in moderately skilled occupations
contracted. 12 No papers document the wage inequality of 1990s and 2000s for the U.S.
manufacturing, as NBER PD data ends in 1996.
In order to illustrate the trends in U.S. manufacturing wage inequality over the
period of 1989-2005, I expand the NBER PD from 1997 to 2005 (see Appendix A for
data and methods description). I use the wages of nonproduction and production workers,
which are often used as proxies of skilled and unskilled labor wages, to construct a
measure of wage inequality.J3 I follow the literature to define this measure as log of the
ratio of nonproduction wages per worker to production wages, where real wages denote
11 In the wage literature, nonproduction and production workers are commonly used to proxy for skilled
and unskilled workers in manufacturing.
12 According to Autor et a1. (Forthcoming) the rising wage inequality in the lower half of wage
distribution was an event confined to the 1980s.
13 Nonproduction wages are constructed as total nonproduction wages divided by total nonproduction
worker employment, whereas production wages are constructed as total production wages divided by
total production hours worked. Data on total nonproduction hours worked is not available.
44
wages per worker. 14 Figure D.l plots 1963-2005 wage inequality for the entire U.S.
manufacturing and as industries' average, where weights for the latter are constructed as
shares of the industry wage bill in total manufacturing shipments. As can be seen, wage
inequality slowly declined from the late 1960s through the 1970s, and began to increase
during the 1980s. Perhaps the most rapid widening of the wage gap can be observed
during the 1990s, when it was also the most steady. Wage inequality decreased
dramatically during the 2001-2002 U.S. recession and fluctuated during the recovery
years that followed.
Table D.1 provides more detail on the growth of workers' wages over the last
three decades. During the period of 1979-1990 covered in most previous studies, the
wages of production workers and nonproduction workers increased at an average 4.99%
and 5.42% per year, such that the relative nonproduction wage rose by an average 0.43%
per year. During 1989-1996 covered in this chapter, production and nonproduction wages
increased at an average 2.67% and 3.78% per year, respectively, leading to a marked rise
in the relative wages of 1.11 % per year. Although both wages continued to grow during
1997-2005, the average annual decline in relative wages of this period amounted to
0.74%, much of which occurred during 2001-2002.
14 This is a common measure of wage inequality in labor economics studies. e.g. Autor et al.
(Forthcoming); Card and DiNardo (2002); etc. Other measures of wage inequality have been used in the
past. For example, Feenstra and Hanson (1999), Haskel and Slaughter (200 I. 2003), and others employ
the ratio of total nonproduction wages to total production wages, which estimates wage inequality in
nominal terms. I find little difference in my measure and this measure of wage inequality.
45
JlLIII. Empirical Methodology
The empirical studies estimating the effect of trade and technology on wage
inequality have typically used a methodology derived from the Stolper-Samuelson
theorem (SS theorem), which links product price changes to changes in factor prices,
under zero-profit conditions. IS This methodology relies on the production side of the
Heckscher-Ohlin model which considers an economy with multiple sectors of different
factor intensities and factors with complete mobility across sectors 16. In this framework,
aggregate demand for skilled workers relative to unskilled workers is horizontal and
aggregate relative labor supply is upward sloping J7 . The aggregate relative labor demand
is horizontal since a change in the demanded quantity of skilled (unskilled) labor can
potentially be absorbed by a change in output in an unskilled (skilled) sector, and thus
may be independent of relative wages l8 . Relative wages, in tum, are detennined by
product prices and/or productivity under zero profit conditions, which in tum are driven
by exogenous forces, i.e. trade or technological innovation. When changes in exogenous
forces alter intersectoral profitability, relative wages change to restore zero profits,
factors flow to other sectors, and the relative aggregate demand curve shifts.
15 Deardorff (1994) surveys all statements of the SS Theorem that have appeared during the past 50-plus
years. One of the statements is the following: "For any vector of goods price changes, the
accompanying vector of factor price changes will be positively correlated with the factor intensity-
weighted averages of the goods price changes."
16 This is different from labor studies which assume that factors are immobile (HaskeI 1999).
17 Note, that the relative demand curve in each sector is still downward-sloping, while the aggregate
demand curve is flat.
18 This is the so-called Rybczynski effect (Rybczynski. 1957)
46
This process can be formalized by supposing that the economy, which in this case
is U.S. manufactudng, produces J different traded goods, associated with J industries.
Each industry employs some combination ofJ primary factors and M intermediate inputs.
Under constant returns to scale technology, zero profit conditions for industry i can be
written as
P,= L PmiGm'+ L wjiaji
mEM jEJ
where P, is the domestic price of one unit of output, Pml is the unit cost of mth
(IlL1 )
intermediate input, G m' is the quantity of rnth input required for production of one unit of
output, w;; is the unit cost ofjth primary-factor, and aji is the quantity ofjth factor
required for production of one unit of output. Totally differentiating to express everything
in instantaneous changes and allowing for changes in the technology of production,
equation (III.1) can be rewritten as
"v·, L" ° "P· = w··--TFPI Jf .11 , ,
jEJ
(IlL2)
where p/A= P,- L P"miOmi is change in value-added prices, TFP,=- Lit jiBji is the
IJlE 114 jEJ
primal measure of total factor productivity, and Bml and Bji are the cost shares of
intern1ediate inputs and primal factors in total costs of industry i, respectively.
Since all factors are mobile across sectors, changes in wages of primary factors
can be assumed to be equal across sectors. Then the existing differences between the
industry wage changes and the manufacturing-wide changes are assumed to arise from
47
the variations in factor qualities across sectors 19. Expressing industry wage changes in
equation (III.2) as differentials from manufacturing-wide changes, I obtain
p;A= L v..'j8 j;-TFP;+ L (li'j;-v..'J8ji,
jEJ jEJ
(III. 3)
.-
where W j is the effective manufacturing-wide wage change of primary factor} and
Wji- i1~j is industry i's wage change differential of}th primary factor. I combine industry
wage differentials with changes in TFP and refer to them as changes in effective TFP,
such that
Li In p;;A +Li In ETFP;, =L Li In Wj ~ (8 jil-l +8 j J ,
JEJ
(IlIA)
where instantaneous changes are expressed in first-log-difference and primary factor cost
shares are averaged over two peliods.
Equation (IlIA) shows how manufacturing-wide factor prices adjust to changes in
value-added product prices andlor effective productivity to restore zero profits in all
sectors. This equation captures the wage adjustments to shifts in aggregate relative labor
demand described above. Value-added price andlor effective productivity increases in a
sector tend to raise (reduce) the relative wages of factors employed relatively intensively
(unintensively) in that sector, where intensity is defined by ~ (8ji,_, +8ji,) . Note, that
productivity changes can be factor-biased or factor- neutral, as long there are changes in
net productivity (or by duality net costs), which raises sectoral profitability and so
necessitates wage changes2o .
19 See Feenstra and Hanson (1999) discussion on pg. 911.
20 This is different from labor studies focus, where only factor-biased teclmica1 change affects wages since
it changes the relative productivity of factors within a sector. See Haskel (1999) for discussion.
48
In the framework discussed above, value-added prices and effective productivity
changes are assumed to be exogenous. In a large country-setting, however, prices and
productivity changes are detennined by domestic and foreign forces. To model the
endogeneity of prices and productivity changes, Feenstra and Hanson (1999) developed a
two-stage procedure, in the first stage changes in prices and productivity are regressed on
exogenous factors, which are then linked to changes in wages. I {onow this procedure, as
described it below.
In the first-stage, I regress changes in value-added price and effective productivity
on a set of J( causal factors, which are hypothesized to drive thes~ changes over time:
(IlLS)
where zi/a is the kth causal variable, :Y k is a coefficient on kth causal variable, and 17
'1 IS
a disturbance tenn that captures all other shocks to the value-added price and
productivity, which are assumed Olihogonal to Zi/,/' Changes in a causal factor can affect
changes in either only value-added prices, or both value-added prices and effective
productivity. In addition to its direct effect on both prices and productivity changes,
LJ Z Ikl can affect price changes indirectly through its impact on productivity changes,
which are "passed through" to product prices (Feenstra and Hanson 1999; Krugman
2000).21 Assuming a 100% pass-through rate, effective productivity changes are neutral
if one finds ;Y k equal to zero.
21 The latter result stems from the fact that productivity changes distort equilibrium in the goods market,
by shifting goods supply, which in tum affect product prices (Haske I 1999). These changes in goods
supply are possible either because the country in question is large in world markets or because the
productivity shocks are common across countries (Krugman 2000)
49
Given the results ofthe first-stage regression (IlLS), one can decompose the total
change in value-added prices and effective productivity into those components due to
each structural variable, namely y k6 Z ilk. These decomposed changes, when individually
regressed on the primary factor cost-shares, yield coefficients interpreted as predicted
factor price changes due to that structural component. The second-stage regressions for
each structural variable k is expressed as:
Yk 6z ikr= L Ojk~(eir-l+eir)+Uil(/.
jE.!
(III.6)
The coefficients 0 jk obtained from these regressions can be seen as the economy-wide
change in the price ofjth primary factor that would have OCCUlTed ifthe change in kth
structural variable had been the only source of changes in prices and effective
productivity.
Only a handful of studies have used the two-stage procedure to identify causal
factors of changes in prices and productivity and link them to wages. These studies find
. mixed contributions of trade-related variables, i.e., foreign outsourcing of materials, trade
barriers, transpOliation costs, and changes in international product prices, on U.S., U.K.,
and Mexico's wages. For example, Feenstra and Hanson (1999) find that a rise in foreign
outsourcing of materials accounts for 15%-25% of the rise in U.S. wage inequality in the
1980s. On the other hand, Haskel and Slaughter (2003) fail to identify a significant
impact of other trade-related variables on U.S. wages of the 1970s and 1980s, although
stronger results are found for U.K. and Mexico's wages (Haskel and Slaughter 2001;
Robertson 2004). A number of studies have also looked at the effect of technology on
50
wage inequality, where both factor-biased, i.e. skilled-biased technological change
(SBTC), and sector-biased technological changes are considered. Feenstra and Hanson
(1999) find that SBTC due to office equipment and computer investment explain over
35% of the rising U.S. wage inequality in the 1980s. On the other hand, industry
innovation contributed the most to the increase in the skilled-unskilled U.K. wage gap
during 1996-1990 (Haskel and Slaughter 2003). I contribute to their methods by using
most recent data for 1989-2004 and exploring a broader set of trade- and technological
change-related factors.
ilLIV. Data and Descriptive Statistics
I apply the estimation technique described in the previous section to U.S.
manufacturing industries for the period of 1989-2004. This sample period encompasses
the changing nature of the U.S. wage inequality debated in the literature, which occuHed
after 1989, when the wage inequality either polarized (Autor et. al. FOlihcoming) or
substantially declined (Card and DiNardo 2002). One feature of the sample is that the
data are classified under the Standard Industrial Classification (SIC) during 1989-1996
and North America Industrial Classification System (NAICS) during 1997-2004. This
forces me to split the sample along the classifications distinction and run the estimation
separately on each of the subsamples. While working with shorter time-series is less
ideal, this approach circumvents the differences in the definition of manufacturing
embedded in the classifications.22 It is impOliant to note that most industry-level studies
22 Other than the classifications differences, I have reasons to believe that the subsamples are roughly
similar, in that they contain equal time-series panels of eight years and both encompass recession and
post-recession recovery periods.
51
of the U.S. wage dispersion span the period of no later than early 1990s, thus I am able to
go far later than the existing literature.
The data for prices, total factor productivity, and cost shares are obtained from the
Bartelsman and Gray (1996) NBER PD for the period of 1989-1996 and the extended PD
for the period of 1997-2004, which I constructed for the purposes of this chapter (see
Appendix A for description of the extended PD). The descriptive statistics for these
variables are reported in Table D.1, which also includes the data for 1979-1990 used in
most previous studies as a basis of comparison. As shown in Table D.1, the period of
1997-2004 experienced the slowest growth in total factor productivity and value-added
prices compared to the prior periods. Services appear to have gained more prominence by
early 2000s.
Now I turn to data description of trade and technology-related causal factors. The
trade-related variables that I identify include offshoring of materials, offshoring of
selected business services, and finished goods imports openness. The set of technology-
related variables consists of computer, office equipment, and other high-tech capital
shares.
To measure offshoring of materials, I rely on standard construction methods,
originally proposed by Feenstra and Hanson (1996, 1999), and an alternative method,
which refines the original formula by utilizing new and previously unavailable data on
trade in intermediate goods. To arrive at the original measure of offshoring, I combine
data on total imports with data on inputs purchases. The data on U.S. imports for the
52
period of 1989-2004 come from Feenstra (2002) and the Census Bureau. The inputs
purchases are obtained from U.S. Input-Output tables provided by the BEA. For each
industry i, the original measure of materials offshoring is constructed as follows: 23
"\' r ., 'I imports
£....; l purchases of znterm. In]Jut.\'ij ·r . j I
" dam. output j+ll17ports j -exports jI ' ,
Total Nonenergy Interm. Purchases;
(III.7)
-----------
where subscript} refers to an industry supplying input} to industry i, where i,j = 1, ... N.
Each product tenn in the numerator of equation (IIL7) is interpreted as industry i's
estimate of imported material inputs from industry j. Then equation (III.7) represents an
industry's share of total imported intennediate inputs in the industry's total expenditure on
non-energy intennediates. This measure is commonly refelTed to as a broad measure of
materials offshoring. One can obtain a nalTOW measure of offshoring, by restricting
attention to only those inputs that are purchased from the same two-digit SIC industry or
three-digit NAICS industry as the good being produced. 24 1 will include the nalTOW
measure of offshoring and the difference between the broad and narrow measures as
separate variables in my estimation. When averaged over all industries, the original
measure of offshOling, defined nan-owly and as a difference, increased at an average
0.29% and 0.23% per year during 1989-1996, and declined at an average 0.19% and
0.13% per year during 1997-2004, respectively, as is apparent in Table 0.2.
23 This formula first appears in Feenstra and I-Ianson (1996), but has been originally used by the BEA in
construction of imported input purchases for the Import Matrices.
24 The narrow measure is assumed to capture the precise definition of foreign outsourcing, which refers to
the contracting out to overseas suppliers those production activities that can be done within a company
(Feenstra and Hanson 1996, 1999).
53
The original measure of materials offshoring suffers from potentially serious
measurement enor. The measurement error arises from the inclusion of economy-wide
import share to proxy for imports of intennediate goods. Since the total imports share
consists of goods unrelated to intel111ediate inputs, the levels and changes of the
offshoring measures are over or underestimated by the levels and variation of the share of
the unrelated goods (see Chapter II). Therefore, the inclusion of the original offshoring
measure as an explanatory variable may bias coefficient estimates.
In this chapter, I make use of unique data on imports of intennediate goods to
refine the cunently used measure of materials offshoring. These data are made possible
as a result of a recently constructed Market Structure Index of HTS ImpOlis (the ImpOlis
Index), which classifies imports into intermediate and finished goods (See Appendix A). I
combine the Imports Index with detailed imports data obtained from Feenstra (2002) and
the Census Bureau for 1989-2001 and 2002-2004, respectively, to derive impOlis of
intel111ediate goods. 25 These are then incorporated into the following modified version of
original measure of offshoring:
'" [ . . ] interm. imports. .LJ purchases of znterm. in uts ·1 . " '. I
i . . P 1/ mterl17. c!O!17. output; + Il1term.lmports; -lI1term. exports;
Tota! Nonenergy Jnterm. Purchases;
(III. 8)
where subscript} refers to an industry from which industry i purchases its intennediate
inputs, where i,j = 1, ... N. This refined measure of offshoring differs from the original
measure by the right term of the numerator, where I use the share of imports of
25 The imports ofintennediate goods include imports of parts, components, and raw materials, as well as
final goods assemblies tbat go through the domestic industries before they enter the retail markets.
These data provide a near perfect estimate of imports of goods subject to otfshoring, in that they
exclude imports of otfshored assemblies of final goods, which enter the U.S. retail markets directly.
54
intermediate goods in the domestic supply of intermediate goods in place of the share of
total imports in the total domestic supply. Comparing the original with the refined
measure of offshoring, there appear to be considerable differences between the measures,
as shown in Table D.2.
Another trade-related causal factor considered in this chapter is offshoring of
services, which has recently attracted much interest in both academic and popular press
circles. The services subject to offshoring commonly include information technology
services; professional, scientific, and technical services; and administrative and support
services (Amiti and Wei 2006). The construction of the measure follows the same
formula as shown in equation (111.8), where intermediate inputs are now replaced with
inputs of selected services. The data for services inputs and services imports come from
the BLS input-output tables and are described in Appendix A. As shown in Table D.2,
offshoring of services grew substantially in 1989-1996, with an average change of 0.04%
or roughly a ten percent growth of the average level of 0.42%. During 1997-2004,
however, the average growth of services offshoring was relatively stagnant.
Following Feenstra and Hanson (1999), I expect to find positive effects of
materials offshoring on changes in value-added prices and effective productivity in the
first-stage and the skilled-unskilled wage gap in the second-stage. Offshoring of services
is likely to have a similar effect in the first-stage, if imported services stir the technology
of production away from nonproduction workers in a productivity enhancing manner.
This then should lead to a negative impact of services offshoring on the skilled-unskilled
wage gap in the second-stage. However, if offshoring of services is merely an alternative
55
to domestically outsourced services, then one should find a price reducing and negative
effect of services offshoring in the first stage. The skilled-unskilled wage gap will
increase (decrease) if sectors experiencing declining product prices are skilled-intensive
(unskilled-intensive).
The measure of openness to imports of finished goods is constructed as the
finished goods imports to industry value-added ratio. During 1989-1996 imports of
finished goods constituted an average of 29.89% of industry value-added, while by
1997-2004 this percentage went up to 47.16%. Competition arising from imports of
finished goods is expected to put a downward pressure on domestic product prices across
all sectors of the economy in the first stage estimation. The skilled-unskilled wage gap
will increase (decrease) if the sectors experiencing declining product prices are skilled-
intensive (unskilled-intensive).
Finally, the technology-related variables are constructed from three measures of
high-technology capital stock; i.e., (l) computers, 2) office, computing, and accounting
machinery (office equipment); and (3) communications equipment; science and
engineering instruments; and photocopy and related equipment (other high-tech
equipment).26 Combining these capital stock measures with "ex post" and "ex ante" user
costs yields "ex post" and "ex ante" measure of services rendered by office equipment,
computer, and other high-tech capital, or in other words, the opportunity cost of capital
26 Previous literature incorporated investment in computer capital in the studies of the 1980s wages (Autor
& Katz 1998, Feenstra and Hanson 1999, 2003). During the 1990s, these data were compiled only
during 2002-2004, which makes it impossible to incorporate computer investment in this chapter.
However, the inclusion of the computer services share variable should reasonably proxy for the impact
of computerization on productivity, prices, and wages.
56
possession (Berndt and Morrison 1995, 1997; Feenstra and Hanson 1999).27 I express
these measures as shares in total capital services and use the first-difference of the "ex
post" capital shares as the primary technology-related explanatory variables. I check the
robustness of the results to the "ex ante" measures in the sensitivity analysis. The data for
the construction of the technology variables are courtesy of the BLS and more detailed
discussion of the construction methods can be found in Appendix A. As shown in Table
D.2, the computer share increased continuously throughout the sample period. At the
same time, office equipment share steadily declined, while other high-tech share rose
during 1989-1996 and declined during 1997-2004. Previous studies found the
technological change attributable to high-tech equipment diffusion as productivity
enhancing and ski]]-biased (Berndt and Morrison 1995, 1997; Feenstra and Hanson
1999). I test the robustness of these findings in the section below.
III.V. Results
The estimation is performed over 458 U.S. manufacturing industries at the four-
digit SIC level for the period of 1989-1996 and 473 six-digit NAICS industries for the
period of 1997-2004. I utilize two methods of variable construction. The first method
uses variables expressed as differences over 1989-1996 and 1997-2004 periods, divided
by the number of years in each period to obtain annualized differences. The estimation
then reduces to a cross-sectional analysis, which is common in the product-price
27 The ex post user costs reflect the internal rate of return in each industry and capital gains on each asset,
and the ex ante user costs reflect a "safe" rate of return (the Moody rate of Baa bonds) and excludes the
capital gains on each asset. Feenstra and I·Janson (1999) comment that ex ante measures might be
prefeITed because they do not reflect the capital gains on the assets and the internal rates of return in the
industry.
57
literature and is motivated by the log-run nature of the Heckscher-Ohlin theory and is
often used to circumvent the limited availability of yearly data (Haskel and Slaughter
2001).
I contrast the results from the "annualized differences" estimation to those where
vmiables are expressed as first-differences. Estimation is then perfol1ned using panel
estimation techniques with fixed effects to control for year-specific unobservables. As
will become apparent, the differences in the magnitudes of estimates from the two
methods are considerable. These differences arise from the fact that the first-difference
estimation captures both industry trends in the data and the time-series variation around
these trends. On the other hand, the annualized differencing approach weeds out the time-
series variation by construction and evaluate the coefficients based on industry trends
alone. Thus, the additional noise captured by the first-di fferences estimation should yield
smaller coefficients, which could potentially be interpreted as short-run estimates. Then
the larger estimates from the annualized differences estimation could be evaluated as
long-run effects.
111. Vi. Preliminary Regression
Before tuming to estimating the two-stage procedure of linking price changes to
wage changes, I check the consistency of equation (4) against the data. Table D.3, part b)
presents the regressions of changes in value-added prices plus effective productivity on
the average cost shares of production and non-prOduction workers and capital.
Regressions are run for changes in variables measured as annualized differences and first-
58
differences, as discussed above. The estimated coefficients can be compared with the
annual average changes in the prices of these primary factors shown in Table D.3, part a).
Similar to the results reported in Feenstra and Hanson (1999) for the 1980s, the estimated
coefficients are extremely close to the actual factor price changes and the regressions fit
nearly perfectly. The wage of nonproduction labor rises faster than production labor
during 1989-1996, indicating an increase in wage inequality, and slower during
1997-2004, indicating a decrease in wage inequality.
In Table D.3, Part c) I examine whether changes in value-added prices, changes in
effective TFP or both are responsible for the increase in the skilled-unskilled wage gap
during 1989-1996 and the decline in the wage gap during 1997-2004. Taking the
differences between the predicted coefficients on non-production and production cost
shares, it appears that changes in prices are concentrated in the unskill-intensive sectors
in both periods, as they result in a relative decrease of the skilled-unskilled wage gap. On
the other hand, changes in the effective productivity are concentrated in the skill-
intensive sectors, as they result in the relative increase in the wage gap during both
periods. 28 This contradicts the findings of Leamer (1998) who finds that both changes in
prices and changes in productivity were concentrated in the skill-intensive sectors of U.S.
manufacturing during 1980s.
28 Leamer (1998) runs similar regressions, but use changes in prices and changes in TFP to predict factor
price changes for the U.S. during 1981-1991. He finds that both changes in prices and productivity are
skilled-labor intensive. I rerun the regressions in Table 3, part c) using the same dependent variables,
and find similar results to Table 3, part c), except that changes in TFP in fact decrease the skilled-
unskilled wage gap during 1997-2004.
59
The results of these regressions are robust to the inclusion of market power
controls, i.e. output/capital ratios and market concentration measures, and to the
exclusion of the computer industry. The results of Table D.3 solidify theoretical
predictions of the SS theorem of the link between prices and productivity and wages.
111 V2. Stage 1
In this section, I report the first stage estimation results of the two-stage
procedure, where I regress changes in value-added prices plus effective productivity on
trade- and technology-related causal factors. The key variables of interest are the
measures of outsourcing of intermediate goods in equations (III.7) and (III. 8). As it will
become apparent, these measures which are comparable to those used in the existing
literature produce coefficients of varying magnitudes and significance, where the
estimates on the refined measure are more robust to various specifications.
There are four estimation issues to be addressed. First, while the dependent
variable is available only at a highly disaggregated level, the SBTC variables are
available only at two-digit SIC level and three-digit NAICS levels in the respective
periods, and the outsourcing variables are available only at three-digit SIC and four-digit
NAICS levels. I cluster the errors at the most aggregated groups to avoid the possibility
that errors are correlated within the more aggregated industry groups (Moulton 1986;
Feenstra and Hanson 1999). Second, since the dependent variables in the second-stage
regressions embody the same estimated coefficients, the standard errors of the second-
stage coefficient estimates need to be corrected.29 I follow the steps outlined in Dumont et
29 If not corrected, the second-stage regressions provide conditional estimates of the residuals that
60
al. (2005) to correct the standard elTors of the second-stage estimation.'o Third, if
industries are not perfectly competitive then the measure of total factor productivity is
biased because the capital share includes pure profits. I include the log change in the
output-capital ratio as a regressor to absorb the market power effect (Domowitz et al.
1988; Feenstra and Hanson 1999; Haskel and Slaughter 2001,2003). Finally, caution
needs to be taken in comparing the coefficients from the 1989-1996 and 1997-1996 data
samples, due to differences in SIC and NAICS classification during the respective
periods. These classification vary considerably in their definition of U.S. manufacturing,
and thus may change the behavior of manufacturing-specific variables across the two
periods.
Table DA presents estimation results from the first-stage regression using the
original specification proposed by Feenstra and Hanson (1999), which includes only
materials offshoring and high-tech capital shares, excluding the computer share. Columns
1-IV contrast estimates for original (I & III) and refined (II & IV) measures of offshoring,
where variables are constructed either via annualized differences or first-differences
methods using the 1989-1996 data sample. Similarly, Columns V-VIII present estimates
for the period off 1997-2004. As mentioned earlier, I include year fixed effects in the
estimation with first-differenced variables to account for time-varying unobservables.
incorporate the additional variance of the residuals from the first-stage estimation. To test the
significance of the second-stage coefficients, unconditional estimates of the standard en-ors accounting
for this additional variance have to be computed.
30 Feenstra and Hanson (1997, 1999) propose a correction procedure which has been disputed in most
recent work by Dumont et al. (2005), since the correction does not require that the computed variances
are positive and may impose a negative bias on the standard errors. The procedure developed by
Dumont et al. (2005), in turn, does guarantee positive variances of the second-stage estimates.
61
As is apparent from Table D.4, the signs and statistical significance of the
coefficients are relatively robust to various specifications within each period. On the
other hand, the magnitudes of the estimates vary considerably across specifications and
sample periods. The most striking differences in magnitudes appear across annualized
differences and first differences specifications, in particular for estimates on materials
offshoring. These differences persist when year fixed effects are excluded from the first-
differences estimation, not shown in Table D.4, although an F-test confirms the necessity
of year fixed effects. Additionally, the negative sign on the office equipment share comes
in contrast to the findings of Feenstra and Hanson (1999). The general lack of
significance of the impact of offshoring measures during 1997-2004 is troubling.
In Table D.5 I present results of specifications with a full set of causal factors. The
inclusion of other controls reveals the severity of the measurement error introduced in the
original measure of materials offshoring. Unlike the estimates on the refined measure, the
estimates on the original measure become insignificant in all specifications and shrink in
magnitudes compared to those in Table D.4. As a result of such poor performance, I turn
my focus to the specifications using the refined measure of materials offshoring of
Columns II, IV, VI, and VIII.
Turning to trade-related causal factors first, the estimates on these factors come
through with mixed signs and significance. The effect of materials offshoring, per refined
measure, changes across time. While offshoring, defined as a difference of broad and
narrow measures, drives the growth in changes in value-added prices and effective
productivity during 1989-1996, it is the narrow measure of offshoring (within closely-
62
related industries) that appears to have a significant effect during 1997-2004.
Furthermore the effect of materials offshoring changes to negative, albeit very small, in
the first-difference estimation of Column VIII. Services offshoring appears to have a
negative impact on changes in value-added prices and effective productivity during
1989-1996, and positive effect during 1997-2004. Openness to finished goods has a very
small and insignificant coefficient.
In order to make sense of the result in Table D.S, I find it useful to separate the
dependent variable in the first-stage estimation into its respective components. Table D.6
shows independent regressions of changes in value-added prices and changes in effective
productivity on causal factors. The first-differences specifications reveal a consistent
picture, where trade-related variables, with the exception of services offshoring in the
1997-2004 sample, increase productivity and reduce prices. This is consistent with prior
expectations that trade-driven market competition puts a downward pressure on prices
and production-related inefficiencies in the short-run. In the long-run, expressed by
annualized differences, however, the results are less consistent. Thus, materials
otIshoring appears to mostly increase both prices and productivity, services offshoring
appears to decrease productivity with mixed effect on prices, and openness to impOlis has
mixed effects on both prices and productivity across the two sample periods. The latter
results may indicate perhaps that it is hard to predict a consistent impact of trade on
prices and productivity when too many things are at play, e.g.. , contracting and expansion
of sectors, restructuring of production technologies, etc.
63
Next I tum attention to the effects of technology-related causal factors on changes
in value-added prices and effective productivity, as shown in Table D.S. The inclusion of
the computer share in the 1989-1996 specifications considerably affects the magnitudes,
signs, and significance of the coefficients on other technology-related variables compared
to those in Table DA. The estimates on the computer share are, in tum, large and highly
significant. However, the effect of computers goes away by 1997-2004, while office
equipment and other high-tech capital shares retain their signs and significance. These
results may be indicative of a changing role of computer technologies in U.S.
manufacturing. While the computer revolution of late 1980s-early 1990s changed the
technology of production in a productivity enhancing manner during 1989-1996, the
Internet revolution ofthe late 1990s and early 2000s may in fact have introduced little
change to the existing manufacturing processes. On the other hand, by the late 1990s,
advances in computerization may have penetrated other high-tech technologies leading to
higher productivity gains, shown by the estimates on other high-tech share in Columns V-
VIII. These interpretations are also confirmed by larger productivity gains from other
high-tech capital share and lower productivity gains from computer share of Table D.6.
Under zero-profit conditions, these estimated changes in value-added prices and
effective productivity can be linked to changes in factor prices. In Table D.7 I rerun the
regressions, only retaining those causal factors that had a non-neutral impact on the
dependent variable in the full specification of Table D.6. Only significant coefficients
signal actual changes in prices and productivity, which will then mandate changes in
factor prices under the zero profit condition (Slaughter 2000).As can be seen, the
64
coefficients on the remaining trade- and technology-related variables are robust to these
changes. I use these final specifications in the second-stage analysis discussed below.
III V3. Stage 2
Before turning to the second stage of the estimation procedure, I first decompose
the dependent variables ofthe first-stage regressions from Table D.7 into those
components due to each causal factor. I then use these components as dependent variables
in the second-stage regressions. The second stage regressions are run without a constant
and are weighted by the average industry shipment in total manufacturing shipments. The
standard errors are corrected using the Dumont et al. (2005) correction procedure, as
discussed above.
The results of the second-stage estimation are presented in Table D.8. Consider,
first, the changes in value-added product price plus effective productivity due to
technological change and induced changes in factor prices. It appears that upgrading of
computer capital is the only technical change variable that is skill-biased, that is it leads
to a negative, albeit insignificant, change in production wages and a positive large change
in the non-production wages during 1989-2004. In contrast, the office equipment share
raises both the production and nonproduction wages in relatively equal amounts, while
other high-tech share increases production wages and decrease non-production wages
during 1997-2004. Taking the difference between the predicted changes in the
nonproduction and production wages due to computerization, the relative wage of
nonproduction labor increased by an astounding average 1.725% per year measured in
65
the long-run, and 1.058% per year, measured in the shOli-run. In contrast, other high-tech
share is responsible for an average 0.310% per year decline in relative wages measured in
the long run, and 0.221 % per year decline measured in the short-run during 1997-2004.
The estimates of Table D.8 can, in fact, be compared with the actual increase in
relative non-production wages. Recall, that the average annual change in log non-
production and log production real wages is 3.839% and 2.666% during 1989-1996
measured by annualized differences and 3.784% and 2.668% measured by first
differences, as reported in Table D.3, Pmi a). The difference between these figures
provides the actual increase in the relative wages of nonproduction to prod.uction workers
of 1.173 % and 1.116% per year, respectively over 1989-1996. Thus, computerization can
individually account for over 147% and 95%, respective of the differencing approaches,
of the observed annual increase in the relative wage of nonproduction labor during
1989-1996. During 1997-2004, the actual relative nonproduction wages declined by
0.256%, when measured in annualized differences, and 0.265%, when measured in first
differences. Then the high-tech equipment diffusion explains over 119% and 85% of the
actual decline in relative non-production wages, respectively, during 1997-2004.
Next, I consider the predicted changes in relative nonproduction wages due to
changes in trade-related variables. Using the above approach of comparing the predicted
wage changes to the actual wage changes, the changes in product price plus productivity
due to materials offshoring explain 51 % of wage changes, when measured in annualized
differences, and 7% of wage changes when measured in first-differences, during
1989-1996. Materials offshoring fails to impact wages in a significant way during
66
1997-2004. At the same time, however, services offshoring has a strikingly large positive
effect on the skilled-unskilled wage gap during 1989-1996, yet a large negative effect on
the wage gap during 1997-2004. These findings are contradictory to each other and leave
me puzzled, since the service offshoring comprise a very small percentage of total
services outsourcing over the period of 1989-2004.
In summary, I find a very strong link between trade and technological change and
relative wages. This link, however is highly sensitive to the nature of the trade and
technology forces in play and the time period under inspection. I find a very strong and
robust effect of materials offshoring on the skilled-unskilled wage gap during 1989-1996,
but its effect during 1997-2004 appears to be statistically insignificant. Similarly,
computerization is found to be the main driver of the relative wage inequality during the
first half of the period, whereas other technological change plays the main role in
determining wages during 1997-2004. Furthermore, one must be careful in considering
all technological change as skill-biased. I find that other high-tech diffusion significantly
raises wages of the unskilled and in fact lowers wages of the skilled dming 1997-2004.
These findings may be indicative of the diminishing role of computers in U.S.
manufacturing, and a growing role of computerization of other high-tech equipment
which works to enhance the productivity of the unskilled, thus raising their relative
wages. The large role of services offshOling in both raising the skilled-unskilled wage gap
during 1989-1996 and then reducing it during 1997-2004 is surprising due to its relatively
low prevalence in manufacturing.
67
III.VI. Sensitivity Analysis
There are a number of points worth noting about my estimation in Tables DA-D.8.
First of all it may be argued that the computer industry has experienced an unusual
productivity growth over the past decades and should be excluded from the industry-level
analysis (Leamer 1998, Feenstra and Hanson 1999). I rerun the estimation without the
computer industry and find that the coefficients are not qualitatively different from the
ones presented in Tables DA-D.8. Another potential concern that may arise is that trade
and technology regressors in the first-stage estimation may be endogenously determined
with value-added prices and productivity. I follow the previous literature in assuming that
they are exogenous. Additionally, I check that the estimation is not sensitive to the
weights employed in the analysis, by using employment and wage bill weights. The
results are qualitatively the same. Furthermore, one may argue that both value added
prices and cost shares need to be deflated by appropriate deflators to net out inflationary
forces over 1989-2004. I rerun the estimation, using manufacturing wide producer price
indexes to deflate product prices and wages, and find no significant changes in coefficient
estimates. Finally, I check the sensitivity of the results using the alternative ex ante
measures of technological change variables. The results are not qualitatively different and
are available 011 request.
III.VII. Conclusion
This study is the first study of the impact of trade and technology on U.S. wages
of 1990s. Using recently available data on industry statistics, I am able to document a
68
near-continuous growth in the 1990s wage inequality within the U.S. manufacturing,
where by some measures, the wage gap is growing more rapidly than that recorded in
1980s. I use these data to contribute to the on-going debate of the effects of trade and
technology on U.S. wages.
My findings indicate that the relative contribution of trade is sensitive to the data
and the type of variables used in the estimation. My preliminary estimation indicates that
the standard measure of offshoring of mateli aIs, proposed by Feenstra and Hanson (1996,
1999) and commonly used in the literature, suffers from severe measurement errors that
prohibit the estimation of the impact of trade in intemlediate inputs on the wage
dispersion of the 1990s. I address this issue by developing an improved measure of
materials offshoring, which remarkably improves the perfonnances of offshoring and
other vmiables across all specifications. FUlihennore, various trade-related variables have
radically different effects on U.S. wage inequality of 1989-1996 and 1997-2004. Thus, I
find that trade in intennediate inputs contribute dramatically to the increase in the wage
inequality during 1989-1996 and 1997-2004, although the effect dUling the latter period
is insignificant. On the other hand, trade in services inputs either raises or reduces the
demand for skilled workers, and these effect are strikingly large.
Looking at the technology-related variables, I find that computelization remains
the most appropriate measure of skill-biased technological change as it adversely affects
the demand for the unskilled and positively impacts the demand for skilled labor.
However, this effect could only be estimated in the 1989-1996 sample, as the extent of
computerization failed to have a non-neutral effect on productivity during 1997-2004.
69
Furthem10re, the changes in the share of other high-tech capital, i.e., communications
equipment, photocopy equipment and various scientific and engineering instlUments, in
fact, are found to increase wages of production workers and decrease wages of
nonprodllction workers during 1997-2004.
In summary, I find much support for the hypothesis that both trade and technology
are some of the factors responsible for the growing wage gap during the 1990s. A
different type of technological change, in turn, is responsible for the declining gap during
the 2000s.
70
CHAPTER IV
CONTRACTS, MARKET THICK1~ESSAND OUTSOURCING
IV.I. Introduction
There is a growing theoretical1iterature on the determinants of the extent and
location of offshoring of intennediates' production (Grossman and Helpman 2002, 2005,
Antras 2003, 2005). These models are built on the transaction costs and property rights
literature, paliicularly because of the necessity of relationship specific investment (RSI)
and the presence of incomplete contracts. Thus, the production of specialized inputs,
tailored to the specific needs of a final goods producer, requires a specific investment
fi'om the supplier. Because of contract incompleteness, final goods producers fear being
held up and choose locations of outsourcing where the probability of hold up is the
lowest. All in all, these models show that the location of outsourcing is sensitive to
market thickness, quality oflegal systems, and extent ofrequired specificity of
investment.
SU11)risingly, the existing empirical studies of outsourcing strategies fail to take
the theoretical predictions described above into consideration all together. These studies
model the determinants of the location and extent of outsourcing by exploring
heterogeneities in countries' production costs, e.g. wages, trade costs, transportation cost,
71
etc. Furthennore, due to the limited availability of data, a large share of their analyses
considers the particular case of an industry or firm. 31 There are a number of empirical
studies in the trade literature, however, which draw on the implications of the incomplete
contracts literature to model the quality of legal systems as a source of comparative
advantage in trade of final goods (e.g., Levchenko 2007 and Nunn 2007). These studies
assume that final goods producing sectors, which require RSI from their input suppliers,
rely on countries' legal systems more than others. As such, they empirically show that
countries with superior legal systems specialize in exports of the final goods that are
more institutionally-dependent in nature. To my knowledge, no empirical study considers
the detenninant role of market thickness on trade or outsourcing activities.
In this chapter of the dissertation I present the first empirical test of the
detenninants of international outsourcing described in models of incomplete contracts of
Grossman and Helpman (2002, 2005) and Antras (2003, 2005). I evaluate the role of the
quality of legal systems, specific investment, and market thickness in the location and
extent of U.S. intemational outsourcing of intermediate inputs, while controlling for other
country-level and industry level characteristics. While my focus is on the outsourcing
strategy of the U.S., I improve on the data-constrained analyses of the existing studies by
exploring a large cross-section of industries and countries which source intermediate
inputs to the U.S. This is made possible due to a recently constructed comprehensive
dataset of u.s. offshoring of intennediate inputs, which spans imports sourced £1'om over
270 industries and over 170 countries. This study is the first study to test the impact of
31 See Ginl1a and Gi:irg (2004) for the United Kingdom (U.K.) manufacturing industries, Swenson (2004)
for the United States (U.S.), Kimura (2001) and Tomiura (2005) for Japanese manufacturing firms and
Holl (2007) and Diaz-Mora and Triguero (2007) for the Spanish economy.
72
market thickness on trade data and also the first to analyze the detenninants of U.S. trade
in intennediate inputs. Finally, I evaluate whether the detenninants of the location of
source countries and the extent ofD.S. international outsourcing are different than those
of U.S. imports of final goods. As such, this paper is the first to quantify whether
outsourcing is just a fonn of trade or a qualitatively different phenomenon all together.
In the first section of the chapter, I present a partial equilibrium model of the
detenninants of the location of outsourcing and derive a number of testable hypotheses.
This model is in spirit of the general equilibrium model of outsourcing proposed by
Grossman and Helpman (2005). The model incorporates three essential features of a
modem outsourcing strategy. First, final-goods producers in the North must search for
input suppliers either in the North or South with the expertise that allows them to produce
specialized inputs. Second, they must convince the potential suppliers in a region to
customize products for their own specific needs. Lastly, final goods producers must
induce the necessary investment in customization in an environment with incomplete
contracting in both the North and South. In a partial equilibrium setting, where
outsourcing happens in both the North and South, improvements in the quality of the
contracting environment and/or market thickness affect the probability for each
specialized final good producer of fInding a suitable partner and successfully engaging in
a contractual relationship. Since the expected profits of each specialized final producer
depend on the profitability of matching, the improvements in contracting environment
and/or market thickness in a country increase the prevalence of outsourcing in that
country.
73
The central implication of the model is that differences in the quality of legal
systems and the thickness of input supplier markets are important determinants of
international outsourcing of inputs which require RSI. I test this prediction with the
recently constructed data on u.s. import of intermediate inputs by country, which I
disaggregate by six-digit industry in accordance with the Bureau of Labor Analysis 1-0
classification. I use a factor content of trade methodology, originally developed by
Romalis (2004), and recently used by Levchenko (2007) and Nunn (2007) to test the
institutional content of trade. The latter two studies test whether countries that have good
quality of institutions capture larger shares of U. S. imports of final goods that exhibit
institutional dependence. This chapter takes this specification and augments it with
variation in industry-level measure of dependence of inputs on RSI and country level
measures of legal system quality and market thickness.
The two main findings of the chapter are as follows. First, I find that the cross-
country differences in the quality of contracting environment and market thickness are
just as important in determining the location and extent of U.S. international outsourcing
as factor endowments. Second, I find that the quality of contracting environment explains
more of the patterns of trade in intermediate inputs, relative to patterns of trade in final
goods. However, the opposite is true for market thickness and factor endowments.
The structure of the chapter is as follows. Section IVII describes the theoretical
model. Section IVIII explains the empirical methodology and the data. Section IVV
presents results. Section IVVI concludes.
74
IV.II. The Model
In this section I develop a number of testable hypothesis of the detenninants of
outsourcing within the context of a partial equilibrium model. This model borrows
heavily from the general equilibrium model of outsourcing proposed by Grossman and
Helpman (2005).
IVJJ.l Model Set-Up
Since outsourcing may include both domestic and international outsourcing,
consider a setting with two countries, North and South. There are two types of consumer
goods, a homogeneous good z and a differentiated good yO,Z), where YO'z) represents the
j-th variety of a continuum of varieties of an I-type good. The I-type good is associated
with point Ion the circumference of a unit circle.
Consumers in both countries share identical preferences and view the varieties of
the good y as differentiated. Letting z and yO. I) be consumption of the homogeneous good
and the j-th variety of the I-type differentiated good, preferences of the representative
consumer are of the form,
I iii I) .Ii
U=ZI-illf f y(j,l)"djdl]" , 0<()(,f)<1
o 0
where fl (I) is the endogenously detennined measure of varieties of the I-type
(IV 1)
differentiated good. Consumers allocate an optimal share /3 of their spending on the
differentiated goods. The elasticity of substitution between any pair of varieties of good y
is f = II (1- ()() .
75
There are two types of producers: final-goods producers and suppliers of
intennediate inputs. Northern and Southern final producers ofthe homogeneous good z
may enter their respective markets and incur the cost of Wi per unit of output, where Wi
is the wage rate in country i and i=N, S. On the other hand, only the NOlihern final
producers have the know-how to produce any varieties of good y. Such a finn must bear a
fixed cost of product design and development, wNIn' where w N is the NOlihern wage
rate and In is the fixed labor requirement. Additionally, the Northem final producer
needs one unit of a specialized input per unit of output, the production of which they
must contract out to a local or overseas input supplier.
The entry of an input supplier in country i requires investment in expertise and
equipment, which I refer to as a production know-how, the cost of which is w' /:/1 , where
i=N, S. Due to high relative costs of entry, only a limited number of suppliers, 11/ U;/I) ,
enter a given market and each supplier serves multiple final producers in equilibrium.12 A
supplier's know-how is represented by a point on the unit circle, spaced at an equal
distance 1/ m' from the next supplier's know-how. Final producers do not know the exact
location of the supplier's know-how on the circle, but consider the nearest supplier to be
at a random distance x from the producer's own production technology know-how, where
x follows a uniforn1 distribution on the [0,1/2 mil interval.
Finally, any supplier must develop a prototype before it can produce the
customized inputs needed by a particular final producer. The full cost of this investment,
Cim' i
32 1 assume that Cil':" <0 . For simplicity, 111 is assumed to be a continuous variable.
76
w' J/ x, varies directly with the distance in expeliise. Furthennore, the input supplier's
compensation for the investment and the actual induced investment are subject to
negotiation and depend on the nature of the contracting environment in country i. Once
the prototype is completed, input suppliers employ one unit of local labor per unit of
output.
The setting is one of incomplete contracts in both North and South. In particular, I
assume that in country i, an outside pmiy can verify a fraction y' <t of the investment
in customization undertaken by an input supplier for a potential specialized final good
producer. In other words, i captures the quality of the legal system in country i; the
greater is i, the more complete are the contracts that can be written there.
When a final producer approaches a potential supplier in a given market,
negotiations between the final producer and input supplier involve bargaining over an
investment contract and an order contract. When two firms negotiate an investment
contract, they specify the extent of the supplier's investment in a prototype, l' (x) , and
the amount of compensation that the customer will pay for the investment, pi (x) .
Assuming Nash Bargaining, one can deri ve the equilibrium outcome of negotiations over
pi(X) and 1'(x) .
Proposition 1: Let 5' denote the profits that the parties will share if they reach a stage
where a suitable prototype exists and ~rthe two parties subsequently reach agreement on
an order contract. Then Nash Bargaining results in/inal goods producer's payment to the
input supplier ofcountry i of
{I . S5 S\
P'(X)= -2 W 'Il'x if -5-\<X< \ S \'
- 2wp 2wJ.l'(]-y)
o otherwise
and the induced investment level of
{
. . S5
t(x)= w' 11' x if x.s \ v \
2w J.l (1- Y ) .
o otherwise
(lV2)
(IV3)
77
Proof: An enforceable contract stipulates an investment of i]./ / x and payment of pi.
Under Nash Bargaining, the input supplier expects to receive a prospective profit of
Si /2. Then, if S /2<(1-;/)"v' IJ'x , the input supplier's perspectlve profits are not large
enough to cover the cost of unenforceable investment and there is no incentive to engage
in the development of a full prototype. If (1 -;/l w' IJ' x :::;S' /2 , on the other hand, the
supplier has an incentive to fully invest in a prototype. Furthe1111ore, if S /2"2. Wi IJ' X , the
supplier's prospective profits cover the full cost of investment, which means that the
supplier is willing to proceed with the full investment even if there is no contract and no
initial payment whatsoever. Finally, if (1 - i) w' J/ X :::;Si /2 <Wi IJ' x, the input supplier
commits to full investment only if the investment contract exists and there is a
sufficiently large payment of P'. In case of the latter, under Nash bargaining, joint
surplus S'-w'IJ'x is split equally and P'=~v' ~/ x/2. 0
Proposition 1 indicates that the investment contract and the induced investment
behavior depend on the final producer's, prototype requirements w'll l and the distance
between the supplier's production know-how, x. Furthennore, x depends on the the size of
78
the potential profits that would be generated by an efficient order contract, and, in case of
the Southern suppliers, the quality of the contracts in their home country, y S •
Once the input supplier has invested in the prototype, the parties negotiate an
order contract. Equal profit sharing ensures that the partners have equal interests
concerning the production and marketing of the final good. The preferences in equation
(IVl) provide a constant-elasticity demand function, which means that profits are
maximized by fixed mark-up price pi= w'/ C<. .33 Then, the optimal quantity of final
goods/inputs specified by the order contract is
i -c
vi=A(~)
• c<.
and the maximum joint profits, net of manufacturing costs, are
i 1-(
S' = (l - c<. ) A (~ )
c<.
(IVA)
OVS)
Now, I consider the search problem facing a typical final producer. The firm must
decide whether to search for a supplier in the NOlih or in the South. Suppose the firm
searches in country i. The producer expects to acquire specialized inputs and earn profits
only ifhe obtains a suitable prototype, that is ifhis distance, x, from an input supplier's
production technology know-how is within the range indicated by equation (lV3). Recall
that the final producer considers x to be a random draw from a uniforn1 distribution over
the range from 0 to 1/2m'. Then for the limitations on contracting to have real effects,
the range of distance, x, indicated by equation (lV3) must be binding; that is lay within
33 The preferences in (l) imply that the producer of thej-th variety of the I-type good y faces a demand
given by)' (.i. I) =Ap (j, I f' when it charges the price p(j, I) , where
''\' I Ilfl rhlll I' 'II 'A=[fJ L.., E' I ". II P (.i, I) - dj dl and E' denotes spending on consumer goods in country i.
79
the [0,1I2m i ] interval. This ensures that not every final producer finds a suitable input
supplier which agrees to and fully invests in a prototype. Let / denote the greatest
distance in input space between any producer that does not exit after having searched for
a partner in country i and its supplier. Then the final producer engages in a relationship
with an input supplier and expects to earn operating profits only if x E[ 0, ri ] , where
i
r = .. .
2w' Ji' (1- y') (IY.6)
The probability, with which a final good producer earns non-zero operating profits
in country i, is equal to the density of suitable input suppliers on each side of the final
good producer's production know-how or 2 f~ mi dx . It follows that the expected profits
of a final good producer who searches in country i are
(Iy.7)
Similar to above, the probability, with which an input supplier from country i earns non-
zero operating profits, is equal to the density of final good producers searching in that
country on each side of the input supplier's expertise or 2 f~' ni dx . The supplier's
expected operating profits are
rr:II =2n
i f~ [~'+pi(X)-WIJ1iX]dx.
IV11.2 Partial Equilibrium Analysis
(IY.8)
In this chapter I consider an equilibrium, where outsourcing occurs in both
countries and there is free entry. While a number of other equilibria exit, this one allows
so
me to explore the trade-offs of outsourcing in the South relative to the North. For
outsourcing to occur in both countries, the final producers face equal expected profits in
both North and South, TT~ :::::TT~ . To ensure zero expected profits, the expected operating
profits for a typical final producer equal the fixed costs of entry, that is
i N[f f] .TTJ1::::: W . 17+. S ,z=N,S. (IV9)
Similarly, the expected operating profits for a typical input supplier equal the fixed costs
associated with the investment in production know-how in country i, that is
i ifi. NSTTI1J= W. 11J' z= , .
The volume of outsourcing can be defined as the totaLinputs/final output
(IV] 0)
produced by all final producer-input supplier pairings in country i, that is Vi =2 m' ni 1" Vi .
In equilibrium, substituting in equations (IVA)-(lV6), (IVS), and (IV] 0) for n i r i y' , the
extent of outsourcing in country i is
, 4 LX ]- / ' riv:::::-- m
I ] . I1J- LX] , (i)1
-y -"2 y (IV]I)
Thus, outsourcing is a nonlinear function of the extent of contracting environment,
market thickness, and the fixed cost of acquiring the production know-how in the region.
I assume a partial equilibrium setting, where the effects on labor and goods
markets, other than specialized inputs markets, are ignored. From equation (IV] I ), I can
derive the following set of comparative statistics as well as the main results. Then, it can
be shown that
611' ~ i(l-iy')
6y' I-LX [I-y'-Hiff m'I:" >0 (IV 12)
81
6 Vi 4 Q( 1- y' '[1 + r5 m' f:,,] > < 0
-- -- m ---- or
6/:/1 1-Q( l-Y'-~(y'J2 r5 r;" m'
where y'<l and om' fM <0.
2 i5 [' ,
• 11/ In
(IV13)
(IV 14)
The intuition behind the comparative statics of equations (IV12)-(IV14) is as
follows. An increase in y' , everything else held constant, affects the probability, for each
specialized final good producer, of finding a suitable partner and successfully engaging in
a contractual relationship. Since the expected profits of each specialized final producer
depend on the profitability of finding a pminer, the improvements in contracting
environment in country i increase the prevalence of outsourcing in that country. In the
same vein, market thickness affects the probability of matching between final goods
producers and input suppliers, which in turn raises the volume of outsourcing. On the
other hand, the input supplier's fixed cost of acquiring the production technology know-
how, /:/1, has an ambiguous effect on the volume of outsourcing in country i, as shown
in (IV14). As the fixed cost of entry increases, the market thickness in country i declines,
which, in turn affects the relative profitability of search in country i. The volume of
outsourcing in country i declines ifmarket thickness is elastic with respect to fixed costs
of entry.
With these predictions in hand, lnow turn to data of U.S. outsourcing of
intennediate goods to test their plausibility.
82
IV.III. Empirical Methodology and Data Description
The basic model described in the previous section illustrates the detenninants of
the extent of outsourcing, with a particular focus on the supplier's country contracting
environment, market thickness, and the fixed cost of obtaining the production know-how,
i.e. expertise and equipment. The impact of institutions on trade has been a topic of much
interest in the recent empirical literature. These studies rely on implications of the
incomplete contracts literature to examine the role of institutions as a source of
comparative advantage in trade of final goods (e.g., Levchenko 2007; Nunn 2007). The
starting point oftheir analysis is the assumption that some sectors rely on institutions
more than others. This would be the case, for example, in sectors which cannot rely on
spot markets for inputs, and instead require establishing complex relationships with the
input suppliers. Then, institutionally superior countries specialize in exports of the
institutionally dependent final goods. The commonly used empirical test of institutional
comparative advantage of trade in final goods relies on the factor content of trade
specification o(Romalis (2004). However, since data on trade in final goods is not
available, these studies test the importance of institutions using data on bilateral total
trade flows.
Unlike Levchenko (2007) and Nunn (2007), I aim to examine the role of
institutions and market thickness as sources of comparative advantage in international
outsourcing of intennediate inputs. I then contrast my finding to those for trade in final
goods. Similar to prior studies, the empirical strategy exploits variation in country
characteristics, i.e., quality of contracting environment and market thickness, and
83
industry characteristics, i.e., dependence on institutions. Using data on U.S. imports
disaggregated by industry and country, my analysis reveals stark differences in the
detenninants of patterns of trade in intennediate inputs and trade in final goods.
IVIII.l. Specification
The empirical framework I follow was developed by Romalis (2004). In this
model, endowments of skilled labor, unskilled labor, and capital across countries are
interacted with the production intensities of these factors across industries. The model
tests the Heckscher-Ohlin prediction, which states that countties specialize in the
production of those goods that use factors in which they are most abundant. Following
Levchenko (2007) and Nunn (2007), I augment Romalis (2004) model to include
institutional intensity. Unlike the previous studies, however, my specification also
includes interactions with country-level measures of market thickness. Thus, I estimate
where i indexes industries and c countries. In particular, In (imports t denotes U.S.
imports of intennediate inputs from industry i of country c, nonnalized by the average
U.S. imports from country i. Country-level variables contr c and thick c measure the
quality of contracting environment and market thickness. These vmiables are interacted
with CONTR i , which denotes the industry-level measure of contractual dependence. I
assume that the cost of acquiring production know-how from equation (IV I I ) is highly
correlated with the country endowments of skilled labor and capital. Consequently, per
Romalis (2004), I include the interactions of country-level measures of skilled labor and
84
capital endowments, sk c and caPe, with industry-level measures of skill and capital
production intensities, SKj and CAP j . These interaction terms are meant to test the third
theoretical prediction ofthe importance of the suppliers' ability to acquire expertise and
equipment, expressed in equation (lV14). Finally, I include country and industry fixed
effects, '\ and T] i , respectively.
Motivated by equations (lV12) and (lVl3), I am most interested in the
coefficients on the contracting environment and market thickness. Positive estimates of
6] and 62 would provide evidence consistent with the predictions of the model: inputs
requiring complex specialized relationships with suppliers (as proxied by CONTR j )
originate from countries governed by good contracting envirolUnents ( cantr e ) and
countries with a large number of specialized inputs suppliers ( thick c). As a fmiher test of
the importance of contracting environment and market thickness, in some specifications
I include an additional interaction term, cantrethick, CONTR j • A negative estimate on
this interaction tenn would indicate that when contracting environment is very good
(markets are thick), the importance of contracting environment (market thickness)
diminishes. This result stems from the model described in the previous section, where
better contracting environment and market thickness improve the probability of matching
with the suitable specialized input supplier and successfully engaging in a contractual
relationship.
85
IV.lll.2 Data Sources and Variable Definitions
I use data on the 1997 U.S. imports classified by 1O-digit Hannonized System
(HS) commodities and country of origin from Feenstra (2002). I disaggregate these data
into manufacturing intennediate input and non-intermediate goods according to the
recently developed Market Structure Index ofHTS Imports (See Appendix F). Using the
BEA mapping of 6-digit 10 industries to 1O-digit HS codes, I aggregate the import data
into 6-digit 10 industries. Overall, there are impOlis data for 171 countries and 315
industries. The dependent variables in each of the specifications, is the natural logarithm
of U.S. imports sourced from country c's industry i. Because a large number of imports
are zero, I replace missing observations on In (imports iC) with zeros. The fit of the model
improves dramatically when zero observations are dropped, however, the signs ofthe
coefficients remain the same, as I show in Table E.5.
Country-level measures of the quality of the contracting environment are adopted
from Kaufmann (2004) measure ofthe rule oflaw.:\4 This measure is meant to capture the
quality of contract enforcement, security of property lights, and predictability of the
judiciary.JS To proxy for input supplier market thickness across countries, I rely measure
of market size, i.e. real GDP (in constant 2000 U.S. dollars) and labor supply from World
Development Indicators. These are available for 108 and 120 countries, respectively.
Labor supply is perhaps a better measure of market thickness, as GDP is likely to be
34 Same measmes of the quality of the contracting environment are used in Levchenko (2007) and Nunn
(2007).
35 My results are robust to the use of other measure of the quality of legal systems, such as those from
Gwartney and Lawson (2007) and World Bank's 120041 Doing Business Database
86
correlated with countries' quality of contracting environment and level of development.
Finally, to measure the input suppliers' cost of acquiling production know-how, i.e.
expertise and equipment, I rely on standard measures of factor endowments, such as
skilled labor and capital. These are adopted from Hall and]ones (1999) and are natural
logarithms of human and physical capital per worker, respectively.36 These measures are
available for 116 countries.
The empirical strategy requires an industry-level measure that captures the
contractual dependence of inteTInediate inputs sourced from overseas. In other words, this
measure captures the importance of relationship-specific investment required in the
production of inteTInediate inputs. To construct this measure, I use 1997 U.S. input-output
tables from the BEA to determine which downstream industries purchase and in what
proportions the imported inputs. Next, I use data from Rauch (1999) to identify which
downstream industIies may require specialized relationships with their upstream input
suppliers. These data classify industry output according to three categories: sold on an
organized exchange, reference-priced in a trade publication, or neither. I follow previous
studies to assume that goods which are neither sold on an organized exchange or
reference price in a trade publication, are more complex (e.g. Berkowitz et al. 2006;
Ranjan and Lee 2004). Then downstream industries with more complex production are
more likely to require relationship-specific investment from their input suppliers. Using
this infoTInation, along with infonnation for the input-output table, I construct for each
36 I test the robustness of results to alternative measure of factor abundance provided by Antweiler and
Tret1er (2002). These measures are available for only 69 countries and considerably reduce the sample
size. The inclusion of these alternative measures does not qualitatively alter the results.
87
J
impOlied input i a measure of contractual dependence as L suR'ther ,where Su is the
1=0
share of input i used by the downstream final good producer}, and R,;eilher is the
proportion of downstream good} that is neither sold on an organized exchange or
reference priced in a trade publication. 37
The measure of contractual dependence desclibed above is similar to a measure
proposed by Nunn (2007), which aims to capture the importance of relationship-specific
investment in explaining the patterns of trade in final goods. Nunn (2007) shows that the
effect of institutional quality on the patterns of trade of final goods is the greatest for
goods that use larger proportions of intell11ediate goods that require relationship-specific
investment. My measure of contractual dependence is symmetrically different from that
ofNunn (2007). Rather than trying to identify the market thickness of upstream
intell11ediate inputs, my measure captures the market thickness of downstream final good
producer that require relationship-specific investment of their input suppliers.
Finally, I control for factor intensity differences in production, which are expected
to capture the extent of required industry-level production know-how. The construction of
factor intensities follows the baseline three-factor model developed by Romalis (2004).
Capital intensity of an industry is measured as 1 minus the share of total compensation in
value added. Skilled labor intensity is then the ratio of nonproduction workers to total
37 The data from Rauch (1999) are classified according to the 4-digit SITC Rev. 2 system. Each industry is
coded as being in one of the following three categories: sold on an exchange, reference prices, or
neither. I aggregated Rauch data into 315 manufacturing industries classified according to the BEA's 6-
digit I-a industry classification. To match each SITC industry to I-a industry, I use a SITC4-HS 10
codes concordance from Feenstra (2002) and the HSIO-I06 concordance from the BEA. Equal weights
are used when in the final SITC4-I06 concordance. The final data contains the fraction of each input
that is neither sold on an organized exchange nor reference priced.
88
employment multiplied by the total share of labor in value added, or I minus capital
intensity. Unskilled labor is not included in the regression because by construction of
capital and skill intensities, it is absorbed into the constant term. These are calculated
from the U.S. manufacturing statistics available in the extension of the NBER
Productivity Database for 1997-2004 (See Appendix F). While all factor intensity
measures are calculated using U.S. data, the estimated coefficients are interpretable as
long as there are no factor intensity reversals (Romalis 2004; Levchenko 2007).
The final sample contains 108 countries and 315 industries, where U.S. imports of
intennediate inputs and imports of non-intern1ediates are sourced from 273 and 303 of
these industries, respectively. Table E.l summarizes the explanatory variables used in the
analysis and Table E.2 provides cOlTelations of interactions employed in the regression
analysis. The countries in the sample are listed in Table E.3.
IV.IV. Results
I now tum to my estimation of (IV16). In addition to presenting the results for
patterns of U.S. imports of intermediate inputs, I contrast these to the estimates for
patterns of U.S. imports of non-intermediate goods. My estimates suggests that judicial
quality, market thickness, factor abundance are important detenninants of patterns of
intern1ediate goods. Furthermore, while market thickness and factor endowments explain
less ofthe patterns of trade in intermediates relative non-intennediates, in most
specification, the quality of contracting environment is more important for trade in
intermediates.
89
IVIV J. Trade in Intermediate Inputs
The baseline estimates are presented in Table EA. Column 1 repOlis a
specification commonly used in the recent literature on the determinants of trade pattems.
The coefficient on the contracting quality interaction is of the expected sign and highly
significant. This estimate changes little with the inclusion of market thickness
interactions in Columns II and III. Because I report standardized beta coefficients, one
can directly compare the relative magnitudes of the contracting quality interaction with
the market thickness interaction. A one standard deviation increase in the contracting
quality interaction increases impOlis by .163 standard deviations, while one standard
deviation increase in market thickness, increases the dependent variable by 0.06 standard
deviations, when market thickness is measured by labor supply in Column III. I consider
these figures as my baseline estimates. These effects are even larger, when I add the
additional interaction of market thickness with the quality of contracting environment..
However, the coefficient on the latter is not statistically significant. The GDP interactions
with contracting quality compare poorly to those of labor supply. As noted earlier, GDP
is, perhaps, a poor measure of market thickness, as it is highly correlated with the quality
of the contracting environment in a country. The correlations on the GDP interactions
with contracts-related measures range from .72 to .92, as shown in Table E.2.
The results presented in Table EA COnfil111 the model predictions that both the
extent of contracting environment and market thickness of suppliers are important
determinants of pattems of trade in specialized inputs. Nevertheless, the combined effects
90
that factor endowments have on the pattern of trade are greater than those of the
contracting quality and market thickness.
To ensure that I am really picking up the effect of institutional quality and market
thickness, I now conduct a number of robustness checks. One concern might be that the
contracting environment measure is a proxy for some other feature of countries with good
contracting environment. For example, perhaps the more specialized inputs require higher
endowments of skilled labor or capital. To address this issue, Table E.5 presents results
for several alternative specifications and subsamples, where I use labor supply as the
ultimate measure of market thickness. In Column I, I repOli results for a full set of
interaction tenns, where contract-dependence measure is interacted with skill and capital
abundance, and factor dependence measures are interacted with the quality of the
contracting environment. The coefficients on the key interactions of interest remain
virtually unchanged. However, the il~teraction tenns of factor dependence and the quality
of contracting envirom11ent seem to pick up all the significance from other interaction
tenns. This suggests that the quality of contracting environment is relatively more
imp011ant than factor abundance for patterns of trade in skill and capital intensive inputs.
To test robustness further, I examine whether the results are dtiven by certain
subsets of the sample. Column II ofTable E.5 reports results where I retain only non-zero
observations of imports volumes. The coefficients retain their sign and significant,
although now the combined effects of contracting quality and market thickness dominate
those of factor endowments. In Column III, I run estimation on a subsample of only
countries of the South, where the South is defined as countries with real per capita GDP
91
of less than 50% of the U.S. level. The list of countries belonging to the North and the
South are provided in Table E.3. It is clear from Column III that the results are not driven
simply by the Northern countries. Neither are they driven by the poorest countries in the
sample, as reported in Column IV where I exclude the countries of Sub-Saharan Africa.
Additionally, I omit China and South East Asia in Columns V and VI to test whether the
effect oflabor supply is driven by China's or South East Asia's large population sizes.
The results remain qualitatively the same. Finally, I check whether the estimates are
driven by outlier industries, such as the top contract-dependent industries. The exclusion
of the top 20 contract-dependent industries does not qualitatively alter the results. 38
One obvious concern is whether the results are sensitive to my choice of variables.
I test the robustness of the coefficients by using alternative measures of contract-
dependence employed in prior studies. For example, Berkowitz et al. (2006) and Ranjan
and Lee (2004) apply Rauch (1999) data directly to total trade volumes, and find that the
effect of insti tutional quality on the volume of trade is greatest for goods that are not sold
on an organized exchange. In unreported findings, I test the effect of this alternative
measure of contract-dependence on patterns of trade in intennediate goods. I find that the
estimated effect of the contract-dependence interaction with the quality of contracting
environment is statistically significant in all specifications, although roughly three times
as small as the baseline results of Table EA. Furthern10re, the estimated effect of the
market thickness interactions are not statistically significant. I attribute this result to the
fact that high levels of aggregation of Rauch's data make them better suited for
38 Neither does the exclusion of top 40 contract-dependent countries, as I attest to in the unreported
findings.
92
characterizing trade in final goods, rather than trade in intermediate inputs.39 Additionally,
I use alternative measures of factor endowments, which qualitatively do not change the
results. 40
IVIV2 Comparison with Non-Intermediate Imports
Next, I aim to examine the importance ofthe determinants of the patterns of trade
in intermediate goods relative to their importance in explaining the patterns of trade in
final goods. The impact of contract enforcement on trade in final goods is tested in recent
works of Levchenko (2007) and Nunn (2007). These studies construct measures of
contractual dependence, as described earlier, specifically for trade in final goods. Due to
the lack of data on trade in final goods, they rely on data on total trade flows and find a
positive impact of institutional quality on contract dependent final goods. The mechanism
of the impact of the contracting environment and market thickness on trade in final goods
is similar to the one for trade in intermediates described in the model in Section IVII. For
example, the model can be extended to allow the South to produce differentiated final
goods, which require outsourcing of specialized inputs. Similar to the North, the Southern
final good producers engage in search of suppliers, with whom the probability of entering
in successful investment contracts depends on the extent of contracting environment and
market thickness in the inputs supply markets. Furthermore, allowing for the Heckscher-
39 Additionally, Levchenko (2007) uses Herfindahl index of inputs usage concentration to determine the
contractual dependence of traded final goods. The idea behind this measure is that the lower
concentration of input usages implies a higher complexity of production technology and higher
contractual dependence. It is unclear whether a symmetrically opposite measure of concentration of
downstream industries purchases is an appropriate measure of contract dependence of intermediate
inputs. Thus, it is ambiguous whether an input that serves more downstream industries is more
relationship-specific than an input that serves a single final producer.
40 These data are from Anteiler and Trefler (2002) and cover 69 countries.
93
Ohlin world with multiple countries, the differences in contracting environments and
market thickness across countries may serve as sources of comparative advantage in the
production of final goods that require varying degree of relationship specificity of
investment.
It is difficult to theoretically pin down the differences in the importance of
contracting environment and market thickness between trade in inputs and final goods.
On the other hand, it is easy to estimate the extent of these differences empirically. Table
E.6 contrasts the separate regression results of equation (IY.16) for U.S. imports of
intennediate and non-intermediate goods. 41 The construction of the contract-dependence
measure for imports of non-intennediate goods follows the one proposed by Nunn
(2007), which is the weighted average of product differentiation of inputs employed by
the source industries (see the data section for more detail). As shown in Table E.6
Column II, in addition to different magnitudes of the coefficients, market thickness
appears to have no effect on imports of non~intennediategoods. In Column III, I replace
the dependent variable of the Column II specification with total U.S. imports and obtain
coefficients similar to those in Column II. The new regression is similar to the one
estimated by Levchenko (2007) and Nunn (2007), where total trade is used as a proxy for
trade in final goods. Finally, in Column IV, I use a weighted average of the contract
41 Imports of non-intermediate goods are calculated as total U. S. imports net of impo11s of intermediate
inputs, and contains the following product categories: consumer goods, non-manufacturing supplies,
and capital investment goods.
94
dependence of source industries and find that both the institutional quality and market
thickness are impOliant detenninants of the pattems of total U.S. impOlis. 42
Next, I combine the samples containing U.S. imports ofintennediate and non-
intennediate goods, in order to gauge the relative importance of the institutional quality,
market thickness, and factor endowments for the two types of goods. Similar to Column
IV of Table E.6, the measure of contract dependence of the source industry is the
weighted average of product differentiation of inputs employed by and the final output of
the source industry. The results are presented in Table E.7, where specifications include
interactions with a dummy, which takes a value of 1 if goods are intennediate in nature
and 0 otherwise (Materials Dummy). It can be be seen from Columns I and II that there
are statistically significant differences in the determinants of the pattems of imports of
intem1ediate input relative to non-intermediates. In Column II, the sample is restricted to
only those industlies that source both intem1ediate and non-intennediate goods. The
common picture that emerges is that the quality of contracting environment, market
thickness, and factor endowments explain less of the pattems ofD.S. import of
intermediates relative to non-intennediates. However, the magnitudes of the coefficients
are relatively small. This indicates that the pattems of imports of intennediate inputs and
ofnon-intem1ediate goods are explained reasonably well by the same set of detenninants.
42 The weighted average of contract dependence of source industries is measured as a weighted average of
contract dependence of inputs employed by the industry and contract dependence of final production of
the industry. The weight for inputs is calculated as the share of input costs in total output, while the
weight for final production is calculated as one minus the inputs weight or the share of value-added in
total output.
95
Next, I test the robustness of the results of Columns I and II to the nature of the
imported products. As such, I restrict the sample to only those imported products that are
predominantly intermediate or non-intermediate in nature. 43 Columns III-IV, V-VI, and
VII-VIII ofTable E.7 present results for samples where 25%,50%, and 75% of the
volume of each imported product enter either intermediate or non-intennediate U.S.
markets, respectively. As can be seen, market thickness and factor endowments continue
to explain less of the pattems of U.S. imports ofintellnediate goods relative to non-
intennediate goods, irregardless of the nature of the imported products. However, as
imported products become more intermediate or non-intennediate in nature, the quality of
contracting environment appears to explain more of the pattems of imports of contract-
dependent intermediates, relative to imports of contract-dependent non-intellnediates. If
one restricts attention only to those industries which source both intennediate and non-
intennediate goods, the contracting environment is roughly 50% more important in
explaining the pattems in U.S. imports of intennediate goods relative to non-intem1ediate
goods as shown in Columns IV, VI, and VIII. A one standard deviation increase in the
contracts interaction term, increases U.S. impOlis of non-intennediate goods by
0.103-0.183 standard deviations and U.S. imports ofintennediate goods by 0.155-0.295.
As can be seen, the importance of the quality of contracting environment grows as
imported products become more intelmediate or non-intermediate in nature.
43 According to the Imports Index, a large share of imported commodities are purchased, to some extent,
by both U.S. manufacturing and final goods markets, e.g. consumers, govemment, etc. Some examples
of these goods are tires, repair parts, fabric. The Import Index assigns weights to commodities that serve
multiple end-use markets, allowing one to deterrnine how much of import volumes of a given good
enter a specific market. Thus, it is possible, that the similarity in the determinants of the pattems of
imports of intermediates and final goods stems from the fact that most commodities tend to serve both
intermediate and final goods markets.
96
In summary, my findings suggest that while patterns of imports of intermediate
inputs and of non-intermediate goods are explained well by the same set of determinants,
the institutional quality explains more of the patterns of trade in those goods that tend to
be contract dependent and mostly intermediate in nature.
IV.V. Conclusion
In this chapter I have tested whether a country's contracting environment, market
thickness, and factor endowments explain the pattern of U.S. trade in specialized
intermediate inputs. I found that countries with good contract enforcement and thick
markets of inputs suppliers specialize in intermediate inputs for which relationship-
specific investments are most important. Contract enforcement and market thickness are
equally important determinants of the patterns of U.S. imports of intermediates, as are
countries' endowments of skilled labor and physical capital. Furthermore, my findings
indicate that countries with better quality of contracting environment specialize in the
production of goods that tend to be more intermediate, e.g., manufacturing inputs, than
non-intermediate, e.g. consumer goods, in nature. Market thickness and factor
endowments, on the other hand, explain less of the patterns of trade in intermediate
inputs, relative to patterns of trade in non-intermediates. This study is the first
comprehensive study to analyze the determinants ofD.S. trade in intermediate inputs and
to do so in the context relative to trade in final goods. Furthermore, this is the first paper
to 'analyze the impact of source country market thickness on international trade.
CHAPTER V
CONCLUSION
----- _._-- - --- ~-- --~ ~~~--------
97
This dissertation examines trends, effects, and detenninants of U.S. imports of
intennediate goods, otherwise known as outsourcing in popular press. Previous attempts
to shed light on the nature of outsourcing of intermediate production have been largely
constrained by the fact that trade data do not differentiate between trade in intermediate
and finished goods. I introduce a newly constructed dataset that clearly distinguishes
between U.S. imports of manufacturing materials, consumer goods, and others, during
1989-2004. With these data in hand, previous findings on the nature of trade in
intem1ediate inputs are confirmed, revised, and added to. For example, in Chapter II I
find that contrary to common speculation the magnitude of US. international outsourcing
of intern1ediate production is larger than previousl y thought. In Chapter III, I use new
data to refine the existing industry-level of measure of imported inputs to find a
significant impact of international outsourcing on the U.S. skilled-unskilled wage gap of
1989-2004. Finally, in Chapter IV, I find SUppOlt for existing theoretical predictions and
reveal that intennediate inputs which require relationship-specific investment are sourced
from countries with better institutional quality and thick supplier markets.
98
Much hype in the popular press and political circles revolves around adverse
effects that intemational outsourcing of production may have on U.S. economy. These
claims rely on a prevailing notion that offshoring is a phenomenon distinct from trade in
finished goods. A major contribution of this dissertation is to objectively analyze the
validity of this assumption. My results suggest that while the differences between the
pattems of intemational outsourcing of intennediate production and the pattems of trade
in finished goods do exist, they are relatively small. However, the effects that the two
types of trade have on importers' home markets are distinct.
The reasoning behind these results is straightforward. The similarities in the
pattems of trade are driven by the fact that intemational specialization of production is
guided by the same mechanisms ilTegardless of whether goods are intelmediate or
finished in nature. This explains why in Chapter II I find that superior varieties of
intem1ediate and consumer goods are produced in high-income countries; and why in
Chapter IV, I find that specialized intermediate and consumer goods are sources from
countries with better quality of institutions, more inputs suppliers, and larger skill and.
capital endowments. At the same time, the effect of trade on U.S. labor markets depends
on the nature of the commodities. This stems from the fact that trade in finished goods
affects all worker in all sectors of U.S. manufacturing, while imports of intennediate
goods, due to their lesser skill requirements, are substitutes to lesser-skilled workers. This
explains why in Chapter III I find that intemational outsourcing is a one of the main
dri vers of the growing wage gap during the 1990s, while trade in finished goods has no
effect on the wage gap whatsoever.
--_.. __._-----
99
Despite the significant contribution to the current literature on outsourcing, the
analyses and results offered in this dissertation should be deemed as mostly preliminary
in nature. A number of issues come to mind. First, the analysis perfonned in Chapter II
relies on the commonly accepted assumption that observed prices of imports, i.e., unit
values, are a good predictor of the quality or the extent of vertical differentiation of
goods. Since in addition to quality, prices reflect costs of production and exchange rate
mechanics, they may be poor instruments in gauging the differences between the
detenninant of international specialization of production of intennediate goods and
finished goods. Second, the empirical methodology of Chapter III relies on the long-run
general equilibrium theory of trade, which holds labor supply as constant. However, the
empirical methodology fails to hold constant the supply oflabor over the period of
1989-2004 and may be the reason why the predicted changes in the wage gap exceed the
actual changes in the U.S. wage gap. Finally, Chapter IV, relies on the assumption that the
characteristics of U.S. source industries and countries are taken exogenous of their
expolis to the U.S. Some recent studies show the reverse causality between country
characteristics, i.e., quality of institutions, and trade which points to a potentially severe
endogeneity in the estimation perfoDned in Chapter IV. These and other issues are
commonly ignored in the CUlTent methods employed in trade-related research. Addressing
these issues is a goal of my future research.
100
APPENDIX A
MARKET STRUCTURE INDEX OF HTS IMPORTS
A.I. Introduction
The Market Structure Index of Manufacturing Impolis (the Imports Index)
decomposes U.S. manufacturing imports into two market groups, (1) finished products
and (2) materials. Finished products are subdivided into consumer goods, equipment, and
nonindustrial supplies (which are inputs to nonindustrial sectors). Materials are industrial
inputs in the manufacturing of finished products. The ImpOli Index contains market
structure information on over 22,000 manufacturing import codes described by the
Hannonized Tariff Schedule of the United States (HTS). The HTS import codes follow a
hierarchical structure for describing all goods in trade for duty, quota and statistical
purposes, and, at ten-digit level of disaggregation, contain a detailed description allowing
one to gauge an impoli's end use market(s). The ImpOli Index covers all HTS codes
describing U.S. manufacturing imports over the period 1989 to 2004. Manufacturing
imports refer to HTS codes that map into the manufacturing industries included in the
NOlih American Industrial Classification system (NAICS) definition of manufacturing.
CUlTently, this does not include those industries such as logging and newspaper,
periodical, book and directory publishing that have traditionally been considered to be
manufacturing and included in the industrial sector under the Standard Industrial
Classification system (SIC). This appendix describes the methods and source data used in
the construction of the market structure classification and relative importance weights of
the Import Matrix.
The methods used in this appendix have been inspired, for the most part, by the
methods utilized in the construction of the Industrial Production Index (lP index),
published in G.17 Statistical Release of the Federal Reserve, and the End-Use
Commodity Classification System (End-Use system), published by the BEA. 44 The
Industrial Production Index classifies U.S. industrial production into market groupings
based on the concept of end-use demand, i.e. intennediate and final demand. The U.S.
input-output tables are used to refine the end-use demand fmiher into industrial materials
and non-industrial supplies, equipment and consumer goods. The input-output data are
also used to construct the relative impOliance weights for each of these market groups,
44 The Industrial Production Index can be found on
http://www.federalreserve.gov/releases/G17lip_notes.htm
101
when goods are assigned more than one end-use market. The FR uses the market detail
provided by the IP index to illuminate structural developments in the economy. The BEA
used the same concept of end-use demand to classify commodity trade data provided by
the U.S. Census Bureau into broad end-use categories, such as industrial supplies and
materials; capital goods, except automotive; automotive vehicles, pmis and engines;
consumer goods; food, feeds, and beverages, and other goods, including government
defense imports. Both the IP Index and the End-Use system are available at a relatively
aggregated industry and imported commodity levels (six-digit NAICS and five digit End-
Use codes, respectively).
Following the use of the input-output tables in the construction of the IP Index by
FR, my construction of the Import Index relies heavily on the 1997 U.S. Import-Matrix
published by the BEA. The import-matrix is a supplementary table to the U.S. input-
output accounts and shows the value of imports of that same commodity used by each
industry. The data from the BEA import-matrix can be used both to define the market
structure of each of the HTS import codes and to derive the relative importance weights
for each market group that comprises the U.S. imports end-use market structure. I
supplement the BEA import-matrix data on commodity market classification with stage-
of-processing data on products comprising these commodities. I obtain these data from
the Stage-of-Processing Index (SOP) provided by the BLS, which classifies major
products comprising six-digit NAICS commoditieslindustries into a relevant stage of
processing, i.e. crude materials, intermediate materials, finished consumer goods, and
capital equipment.45 The BLS products descriptions parallel the overall content of the
descriptions of the HTS import codes and the products' stage of processing infonnation is
used to supplement the infonnation on the end-use markets obtained from the detailed
HTS imports descriptions.
The construction of the market structure classification ofHTS imports involved
an individual examination of the over 14,000 code descriptions, which document the
physical nature of the imported product and its stage of processing or the industry
categories associated with its production. As the result of these efforts, I am able to
construct a market classification ofHTS imports that decomposed imports into industrial
materials, non-industrial supplies, consumer goods, and capital investment. The relative
impOliance weights for multiple market groups were derived from the BEA import-
matrix by setting up and solving a set of constrained matrix equation problems.
The ImpOli Index can be used in a wide variety of research projects where trade
data by the type of impOlis is needed. For example, trade in intell11ediate imports has
been reported to have increased dramatically in the past four decades. Some economists
attIibute it to the rising levels of foreign outsourcing, where U.S. finns contract
intermediate pmis and components at arms-length from foreign suppliers. However,
previous proxies of foreign outsourcing incorporate an estimate of imported intennediate
45 The Commodity-Based Stage of Processing Index can be found on
ftp://ftp.bls.gov/pub/special.requests/ppi/sopnew07.txt.
102
goods based on the share of total u.s. imports in domestic supply. Thus, the Import Index
allows the derivation of more accurate measure of imp011ed intermedi ate goods.
The Import Index is likely to be updated as more data become available or as
difficulties with the data are noted and clarified. For example, currently the Import Index
relies on data from the 1997 BEA import-matrix for the derivation of the relative
importance weights of market groups over the period of 1989-2004. It is reasonable to
assume that industrial technologies and consumer demand is subject to fluctuations, and
the market structure of imported production changes with time. I intend to update the
relative importance weights as new imp011-matrices become available in the future.
The remainder of the appendix is structured as follows. A.2 describes the methods
and data used to derive the market structure classification ofHTS imports. A.3 describes
the methods and data used to derive the relative importance weight of market groups. AA
contains a discussion of some of the conceptual and practical problems involved in
deriving the Import Index. A.S concludes.
A.2. Classification Data
A.2.1 BEA Import-Matrix Data
The BEA develops the import-matrix as a supplementary table to the input-output
accounts in order to distinguish domestic production from imp011ed production, which
the input-output accounts do not do. However, since the data on the use of imports by
industries and final uses are not available, the BEA develops its import matrix by making
the assumption that imports are used in the same proportion across all industries and final
uses. As described below, this is a major sh011coming with the BEA's approach. A
commodity's imports are then decomposed into inputs for an industry and final uses by
multiplying the share of the commodity imports in total domestic supply (domestic
shipments or receipts plus imports less exports and change in private inventories) by each
ofthe commodity inputs in the input-output tables. As a result, the market structures of
the import-matrix and input-output tables are conceptually the same. When looking at the
import-matrix, however, the relative magnitudes of inputs in the import-matrix are
different than the ones in the input-output tables. I attribute these differences to the BEA's
ability to recognize that some imp011s may not serve the same markets as the domestic
inputs. As the result of these small differences, I use the 1997 imp011-matrix, rather than
the 1997 benchmark input-output table, as the basis for construction of the Import
Index.46
The BEA import-matrix is used to derive the market structure classification and
relative importance weights for each market for the six-digit commodity codes (I/O
codes) used in the import-matrix. Table A.l describes the market structure layout, which I
borrow from the definition developed by the Federal Reserve Board (FRB) and used in
46 In the case of a small number of commodities, I find that the data from the input-output tables do a
better job of describing the market structure of the HTS imports. As a result, I substitute the data from
the import-matrix with the data from the input-output tables and record these commodities in AA.
103
Table AI: Market Structure of Imports from the BEA Import-Matrix
Import Type
mtermediate Goods
Industrial Materials
Final Goods
Non-Industrial Supplies
Consumer Goods
Capital Goods
Fnd-use Demand
mtermediate Demand
Industrial Sectors
Manufacturing
Final Demand
Non-Industrial Sectors
Agriculture, forestry, fIsheries
Mining
Construction
Transportation
Communications
Utilities
Trade
Finance, insurance and real estate
Services
Government
Special Industries
Consumption
Personal consumption expenditures
Federal government consumption
State and local government consumption
Fixed investment
Private investment in equipment, software, and structures
Federal government investment
State and local government investment
the Industrial Production Index. The market structure classification is based on the
concept of end-use demand, comprised of intermediate and final demand. The
intermediate demand consists of end-user industries that belong to the U.S.
manufacturing sector, while the final demand is comprised of end-users in the U.S. non-
industrial sector, i.e. private consumers and the government. Businesses and government
investments into equipment, software, and structures are also included in the final
demand market group to comply with national accounting standards even though
equipment and software are inputs into production of final goods.47The imported
commodities purchased by the suggested end-use markets can then be decomposed into
intermediate goods, i.e. industrial materials, and final goods, i.e. non-industrial supplies,
consumer goods, and capital goods.48
47 Imports of structures constitutes only a small percentage of capital investment goods and include
imports of mobile homes, bridge sections, and others. Imports are not distributed to the change in
private inventories. See U.S. Department of Commerce (2006), pg. 12-6 for more detail.
48 See Industrial Production Index developed by FRB for more detail, which can be found on The
Industrial Production Index can be found on http://www.federalreserve.gov/releases/G17/ip_notes.htm
104
It is important to note that an imported commodity may be included in more than
one end-use market group. This occurs because some goods are used by businesses as
inputs and are also purchased by consumers for personal consumption, ego gasoline. At
the BEA import-matrix commodity level, the relative importance weights of each of
market group are derived as the share of imports going to that market in total imports. I
find that the markets captming the largest amount of imported inputs in the impOli-matrix
closely compare to the markets indicated in the BEA's End-Use Classification, which is a
5-digit coding system of U.S. exported and imported merchandise that judges the
principal final use of the traded commodities.49
I use the market structure classification and the relative importance weights fi'om
the BEA import-matrix as the basis of the market structure classification for the HTS
imports falling into the I/O commodity codes of the import matrix. An additional helpful
feature ofthe import matrix is that it provides information on all the detailed industries
consuming an imported commodity. This more refined detail on market structure
compliments the description ofthe HTS codes and validates the end-use markets
suggested in the HTS description.
One ofthe disadvantages of using the BEA impOli-matlix to derive the volume of
imported commodities by type, i;e. materials, supplies, consumer goods, and capital
goods, is that the import-matrix itself provides only a rough estimate of imported inputs.
As mentioned earlier, since the data of import purchases by industry are not available, the
BEA assumes that the ratio of total imports to domestic supply is the same as the ratio of
imported inputs to total inputs produced in an industry. This methodology is rather crude,
since imports of an import-matrix six-digit commodity consist of a range of products that
may relate to their domestic supply in the same way as the total impOlis relate total
domestic supply. In other words the assumption made by the BEA in the impOli-matIix
may grossly over- or under-estimate the impOlied inputs end-use demand. In order to
arrive at a more accurate decomposition of impOlis by the type of end-use markets they
serve, I need to examine the imported commodities at a higher level of detail, i.e. the ten-
digit HTS coding system used to record products on U.S. custom forms.
A.2.2 BLS Stage a/Processing Index Data
Another important data source used in the construction of the market structure
classification of imports is the Stage-of-Processing Index used by the BLS in its efforts to
compile the Producer Price Indexes (PPI). The BLS constructs the Producer Price Index
(PPI) to measure change over time in the selling of domestic producers of goods and
services. The product price indexes are developed by identifying one or multiple major
commodity(s) produced by each of the four-digit SIC industries prior to 1997 and six-
digit NAl CS industries as of 1997. I will refer to the BLS commodities as products, in
order to avoid the confusion with BEA six-digit commodities, which the products
49 The End-Use Classification divides imports and exports into the following markets: industrial supplies
and materials; capital goods, except automotive; automotive vehicles, parts and engines; consumer
goods; food, feeds, and beverages, and other goods, including govemment defense imports.
105
comp11se. These products are classified by six-digit BLS product codes and price
information is collected on each of the products from a sample of industries. Over 10,000
different price indexes are offered by the BLS for individual products and services and
their groupings. One set of such groupings is the products aggregated by their stage-of-
processing (SOP).
The BLS had identified three SOP categories that consist of crude materials for
fmiher processing; intermediate materials, supplies, and components; and finished goods.
The crude materials for further processing include products that are entering the market
for the first time and have not been processed. The intem1ediate category includes
partially processed materials that require fUliher processing and components that require
only assembly or installation. In addition, this category includes fuels and lubricants,
containers, and supplies consumed by businesses as inputs into the production of outputs.
Final goods are those that are ready to be sold to consumers for personal consumption or
to businesses as capital investment. As can be seen, the SOP methodology follows the
same line of reasoning as the end-use demand concept developed in the FR Industrial
Production Index. Unlike the FR, however, which considers non-industrial supplies as
final goods, the BLS refers to them as intermediate goods. This distinction is rather
arbitrary, thus I develop the market structure classification of impOlis in such a way as to
allow the practitioners to draw the line between intermediate and final goods as they
deem appropriate.
The SOP index of product end-use markets compliments the data from the BEA
import-matrix in that it provides infOlmation on the product composition of BEA
commodities which are missing from the BEA impOli-matrix. Using the concordance
provided by the BLS, I combine the I/O commodity codes fi-om the impOli-matrix and the
BLS product codes from the SOP index to compare the market structure composition of
commodities and products from each of the data sources, respectively. I find that when
aggregated, the BLS products structure parallels that of the BEA import-matrix
commodity end-use demand structure. Additionally, I are fortunate to find that the BLS
products description can be easily matched with the HTS imports descriptions. Thus,
while the BEA import-matrix gives us a rough idea of market structure of the HTS
imports at the six digit I/O commodity level, the BLS SOP index provides further detail
on the market structure of individual products that closely match HTS imports within the
I/O commodities.
A.2.3. HTS Codes and Descriptions Data
The U.S. Census Bureau is the gatekeeper of the Hannonized System of
commodity classifications, which records imports and expOlis as they cross U.S. customs
boundaries. The Harmonized System of commodity classification comprises a
hierarchical structure for describing all goods traded for duty, quota and statistical
purposes. The HS was developed under the auspices of the International Customs
Cooperation Council, which sought to establish an intemationally accepted standard for
the classification of internationally traded goods in order to eliminate one source of nOl1-
106
tariff trade barriers. Currently, the HS system is administered by the World Customs
Organization in Bmssels, which assigns 4- and 6-digit HS product categories to all
products traded world-wide. The U.S. subdivides these products further into 1O-digit non-
legal statistical reporting categories. The particular application of the Hannonized System
to U.S. imports is called the Harmonized Tariff Schedule (HTS). Currently over 12,000
HTS codes describe U.S. imports of manufacturing goods. Each code is supplemented by
a highly detailed description of the physical nature of products and their stage of
processing or the industry categories associated with their production.
I compile a dataset ofmanufacturing HTS import codes over the period of
1989-2004 from two sources. I make use of the HTS imports data from the NBER Trade
Database for 1989-2001 and the Census Bureau imports data for 2002-2004, from which
I derive manufacturing HTS codes and their detailed descriptions.50 I use HTS-NAICS
concordance files from Census Bureau to identify the manufacturing HTS codes as those
falling into the NAICS definition of manufacturing sector.
I proceed with constmcting the market stmcture classification by studying the
detailed description of over 14,000 codes and classifying them by end-use market type,
i.e. materials, supplies, consumer goods, and capital goods. This is done by cross-
referencing my classification with the overarching market stmcture of the six-digit
commodity codes derived from the BEA import matrix and stage-of-processing struchii-e
of the matched BLS products descriptions. 51 I find that roughly 75%-85% of the HTS
import descriptions provided sufficient detail on the physical nature of impOlis, including
their stage of processing and/or industry-related infOlmation, to establish the HTS
impOlis end-use demand with reasonable certainty. Another 10%-15% of the HTS codes
descriptions entailed highly technical specification which permitted us to tum to
specialists in the relevant fields for help in classifying their end-use demand. The
remaining portion of the descriptions rendered little if any infonnation on products end-
use demand. The most easily accessible type of product descliptions referred to
commodities such as textiles, household goods and appliances, heavy machinery and
transport equipment, anns and ammunition, raw materials, and items described as "parts"
or "in a retail package meant for the ultimate consumer". The next level of description
difficulty referred to commodities such as instruments and appliances for technical uses,
feliilizers and agricultural chemicals, construction materials, and some food items.
Commodities such as chemicals, paper products, some electronic and mechanical
equipment and their parts and accessories, wood products, and some other miscellaneous
manufactured goods required us to tum to the help of specialists in these fields. Many of
the phannaceutical, food items and products labeled as "others" provided us with little
50 ] use the Census Bureau HTS concordances for 2000 and 2006 to derive the code descriptions over the
period of 2002-2004, since these are not included in the Census Bureau imports data (HTS
concordances: http://www.census.gov/foreign-trade/reference/codes/index.html).
5] The BEA 1/0 codes are mapped to the ten-digit HTS codes by Llsing the ]/O-HTS concordance provided
by the BEA.
107
infonnation on the impOlis' end-use and had to be assigned the market structure of the
overarching six-digit commodity code from the impOli-matrix.
As the result of these effOlis, I am able to construct a Market Structure Index of
HTS Imports that decomposed imports into industrial materials, non-industrial supplies,
consumer goods, and capital investment. I describe each of these import categories in
more detail below.
Industrial materials consist of goods that are incorporated into final goods
produced by a manufacturing industry and those that are used during the production of
final goods. The first category of industrial materials incorporates materials for non-
durable and durable manufacturing. For example, materials for food manufacturing
encompass processed foods such as flour, vegetable oils, and confectionery materials.
Materials for other nondurable manufacturing include wood pulp, lumber, industrial
chemicals, plastics, textiles, and others. Metal mill products and pmis and components of
machinery and equipments are example of materials for durable manufacturing. The
second category of industrial materials are materials that compliment the production of
final goods. These materials include processed fuel and lubricants, packaging materials,
some administrative supplies, and others.
Non-industrial supplies consist of materials for construction, agriculture, utilities,
and other businesses. Materials for construction include a wide range of commodities, i.e.
lumber, plywood, millwork, glass, plumbing fixtures, water heaters, and furnaces.
Materials for the agricultural industry include feeds, processed fuels, machinery repair
parts and so forth. Other business supplies include telecommunication supplies,
packaging materials, office supplies, repair pmis, and processed fuels.
Consumer goods consist of consumer foods, other nondurable goods, durable
goods, and defense- and non-defense related govemment supplies. The consumer foods
category is made up of a range of processed food items, examples of which include
bakery products, processed meats, canned and frozen items, and so on. Other consumer
nondurable goods consist of items with a shelf life ofless than three years and that are
ready for final demand (SOP). Some examples of these goods are children's apparel,
prescription drugs, cosmetics, sanitary papers, and energy goods such as gasoline, home
heating oil, and residential electric power. Consumer durable goods include products that
have a much longer shelf life than nondurables. Items in this category include passenger
cars, light trucks, household appliances, and home electronic equipment. Govemment
supplies consist of defense and non-defense related consumption. Examples of defense-
related supplies include ammunition, certain chemicals compounds, repair parts for
military equipment and machinery, and others. Items in the non-defense-related supplies
include education-related products, office supplies, and repair pmis for non-military
equipment and machinery owned by the government, e.g. snowplows.
Capital goods consist of products that are used to manufacture or transport other
goods in the manufacturing sector and includes machine tools for cutting and stamping
108
metals, other specialized machinery (such as faIm machinery and textile machinery),
heavy trucks, ships, and boats. In addition, this grouping includes non-manufacturing
industry and non-defense related government products, such as computers, office
furniture, and heating equipment, that are used in the operation of businesses. Defense-
related government investment such as military weapons and transportation equipment
are also included in this category.
A.3. Market Weights Data and Methods
Having obtained the market structure classification ofHTS imports, I proceeded
to derive relative importance weights for the end-use markets that define each import.
HTS imports that have only one end-use market are assigned a weight of one. HTS
imports that are purchased by more than one market are assigned market weights that
sum up to one. The initial weights at the six-digit I/O commodity level are derived from
the 1997 BEA impOli-matrix and are used to calculate weights at the ten-digit HTS
import code level. Ideally, I would like to have an impOli-matrix weight for every year in
my sample, 1989-2004. However, so far only the 1997 impOli-matrix had been made
publicly available by the BEA. Since the impOli-matrix weights and market structure
parallel those of the U.S. input-output tables, I can also use U.S. input-output table to
construct the weight. However, the most disaggregated input-output data is available only
for benchmark years during which the U.S. Census is conducted. During my sample
period, the benchmark data are available for 1992 and 1997 years only, as the 2002 data
have yet to be published. Therefore I focus only on the 1997 import-matrix data and leave
the revision of the Import Index that incorporates the 1992 and 2002 input-output data for
future work. When incorporating the 1992 data, which is available at SIC industry detail,
I plan to expand the index to include industries that match the SIC definition of
manufacturing sector as well.
In the derivation of the market weights, I limit my focus to HTS codes that have
non-zero imports in 1997, which, theoretically, are the imports that are embedded in the
1997 U.S. input-output tables and import-matrix. Additionally, I utilize the 1997 NBER
import data to derive the relative importance of each of the HTS codes within the I/O
commodity code. The relative importance is calculated as the share of 1997 imports
volume of an HTS code import in total imports ofHTS codes within the I/O commodity
code.52
The first step in deriving the relative importance weights for each of the HTS
codes is to recognize that many of the HTS codes within each I/O commodity code have
identical market structure compositions. For example, a number ofHTS codes within the
same overarching I/O code are classified as materials and supplies, and another set may
be classified as supplies and consumer goods. I assume that HTS codes with identical
market composition within the same I/O code are purchased by end-use markets in the
52 I restrict attention to imports based on the "general imports" classification of imports at foreign port
values, similar to the imports in the import-matrix.
109
same proportion relative to each other. I have to make this assumption, as there are data
on the unique market shares of each of the HTS codes which do not exist.
Using the assumption discussed above, I aggregate HTS import codes with
identical market structure classification within the same six-digit I/O commodity code to
form a HTS cluster. I find that each I/O commodity code contains at most seven HTS
clusters. My original intention was to calculate the relative importance weights for HTS
clusters within each of the I/O commodity code by setting up a matrix equation and
solving it for the market weights. However, since each unique market weight represents
an unknown, I found that for some of the I/O commodity codes, the matrix equation does
not have a solution, since there are more unknown market weights to be calculated than
there are equations (the matrix subject to inversion is not full rank).
I proceeded to solve this problem by netting out the known market weights from
the import-matrix weights and then distributing the remaining I/O code weights to
markets in the same proportion as they appear in the import-matrix weights. It is best to
illustrate this using an example. Suppose a hypothetical I/O commodity code 333333 is
found to have three HTS clusters with markets assigned as shown in Table A.2. Table A.2
also shows the percentage of imports that each cluster contributes to total imports.
Additionally, not shown in Table A.2, the market structure of the I/O commodity in the
BEA import matrix is represented by "materials, supplies, and consumer goods" with
relative weights of 0.35,0.60, and 0.05, respectively.
Table A 2: Hypothetical Example of Market Structure
I/O Commodity HTS Cluster Markets bnport Share
333333 1 Mat. 0.25
333333 2 Mat.- Supp. 0.60
333333 3 Mat. - Supp. - Cons. 0.15
Data Sources: HTS Cluster are HTS codes wilhing the 1/0 code aggregated by
sim ilarity in market composition; markets have been assigned as described in
Section 3; import share isFom 1997 NBER Trade Dataset and is calculated as
importsjor the HTS cluster over total importsjor the 110 com modity code.
From the market structure in Table A.2, one can see that the HTS cluster l's
materials market has a weight of one, and the HTS cluster 3's consumer goods market
has a weightofO.05/0.15:::::0.33. The remaining import-matrix weights then become 0.10
for materials [=0.35-0.25], still 0.60 for supplies and 0 for consumer goods. These
remaining weights are used to establish the proportions in which materials and supplies
weights will be calculated in clusters 2 and 3. Thus, the weight for materials becomes
0.14 [=0.10/(0.1 0+0.60)] and 0.86 for supplies. In cluster 2, the materials and supplies
weights then become 0.14 and 0.86, respectively. In cluster 3, one must subtract 0.33 of
the consumer goods weight, which results in roughly 0.66 ofjoint weight distributed to
materials and supplies. Multiplying this by 0.14 and 0.86, I obtain the actual weights for
110
materials and supplies. Multiplying this by 0.14 and 0.86, I obtain the actual weights for
materials and supplies in cluster 3, which are and supplies weights become roughly 0.09
and 0.57, respectively. I check that the weights assigned to clusters' markets add up to
those of the import-matrix in Table A.3 as the last step of the process.
Table A.3: Import-Matrix Weights vs. My Weights
Weights
Import-Matrix
Imports Index
Materials Supplies Consuroor Goods
0.35 0.60 0.05
1*0.25+0.14*0.60+0.09*0.15=0.35 0.86*0.6+0.57*0.15=0.60 0.33* 0.15=0.05
This method of calculating market weights worked well for all combinations of
clusters encountered in the market structure classification of imports. Table AA lists the
frequency of six-digit I/O commodity imports with one, two, and three clusters
(excluding multiple clusters containing single markets), the market composition of and
the number ofHTS imported products within the clusters in 1997. As can be seen, my
method of calculating the relative importance weights applies to all of the HTS imports in
1997.
Having figured out a way to calculate market weights, as described above, I ran
into another somewhat anticipated hurdle. Since the data in the BEA imports matrix are
roughly estimated by assuming that the ratio of total imports to domestic supply is the
same as the ratio of imported inputs to total inputs produced in an industry, the weights
derived from the import matrix result in crude approximations of the actual weights of
markets that the imports serve. Consequently, in the process of estimating relative
importance weights at the level of detail provided by the 1O-digit HTS codes, I am bound
to run into situations where the BEA import-matrix weights need to be slightly modified,
moderately adjusted, or completely replaced by a better weight measurement. I found that
HTS imports within 16 1/0 commodity codes do not serve a market indicated by the
commodity code, and two do not serve multiple markets53 • I remedy this situation by
distributing the weights of the missing market to other markets, while maintaining the
proportions in which the other markets relate to each other. Additionally, HTS imports
within 26 I/O commodity codes have a market either under- or over-represented by the
import-matrix weights, and multiple markets are under or over-represented for HTS
imports within six other I/O codes54 • I deal with the former situation by redistributing the
53 No materials:335222, 336212, 337121, 337124; no equipment: 313100, 335929 , 336211,333299,
325180,334613; no supplies: 313310; no consumer goods: 325212,333611,334113; no supplies or
consumer goods: 331312, 336611.
54 Less materials: 315200, 315900, 322226, 325520, 33451A, 33712A; less equipment: 336999, 333313,
333315, 339113, 33999A; more materials:323118, 332211, 333293, 333991, 334210, 335129, 335211,
335212,336110,337122,337127; more consumer goods: 325221, 325222; more equipment: 333993,
336120; more or less of multiple markets: 333298, 334210, 334300, 335224, 335228, 335228,325520.
Table A.4: Types and Frequency of HTS Clusters
111
One Cluster # ofI/O # ofHTS Two Clusters # ofI/O # ofHTS Three Clusters # ofI/O # ofHTSComm imports Comm imports Comm imports
ec 13 133 ec, mec I 6 ec, rns, rnsc 1 8
me 2 21 ec, rnse I 16 ec, rnsc, sc I 30
me 6 59 me,rnsc I 60 me, sc, sec I 21
mec I 2 me,rnse 1 15 rns, msc, rnse I 11
rns 18 201 me,rnsec 2 23 rns, sc, sec 6 606
rnsc 61 3399 me, sec I 59 msc, sc, sec 5 252
rnse 12 102 mec, rnse I 43
rnsec II 280 rns,rnsc 28 1876
sc 21 612 rns,rnse 4 25
se 3 30 rns,rnsec 9 459
sec 2 12 ms,sc 4 39
rns, sec I 54
rnsc, rnse 3 229
rnsc,rnsec I 41
rnsc, sc 47 3312
rnsc, sec 6 218
rnse,rnsec I 12
rnse, sc I 16
rnse, se I 2
rnse, sec 4 110
rnsec, sc 2 45
rnsec, se I 22
rnsec, sec 1 36
sC,sec 2 54
Total 150 4851 53 2621 15 928
Note: m - materials, s - supplies, e - equipment, c - consumer goods
extra weight from other markets or to other markets, respectively, while maintaining the
proportions in which the other markets relate to each other. In the latter situation, I
recalculate weights for each HTS cluster based solely on the relationship that the markets
in the cluster hold when compared to each other in the import-matrix. Lastly, some HTS
imports market weights were better described by the weights in the input-output tables,
which, while very similar to the import-matrix weights, contain enough of a difference to
better fit the HTS imports market composition.55 In these cases, I substituted the import-
matrix weights in the corresponding I/O codes with the weights from the input-output
matrix.56
55 These correspond to 1/0 codes 311119, 316200, 326290, 332994, 333120, 333210, 333313,333131,
3335IA,333913, 333921,333992,333924,334516,335212,335224,335228,335999,336999,
339112,339115.
56 Additionally, two HTS imports in 333319 and 339111 I/O codes belong to a different 1/0 code. I made
-------- --------------------
112
Having completed the derivation of relative impOliance weights for HTS imports
in 1997, I assign these weights to all the HTS import codes for the period of 1989-2004.
During this period, some HTS impOlis codes became obsolete and some new ones may
have been introduced due to changes in tariff margins and introduction of new imported
products to the U.S. markets. It is interesting to note that the end-use markets of many of
the pre-1997 HTS codes and post-1997 HTS codes are not different from products
imported in 1997. I find that most of the newly introduced impOlis mimic the old imports
in their general physical characteristics or stages of processing, or industries with which
they are associated. The differences between the new and old import are predominantly
detem1ined by aspects of product differentiation, such as size, incorporation of new
materials (plastics vs. metal), and others. For the most part, this can be explain by the
efforts the Census Bureau makes in improving the detail by which goods are described at
U.S. customs. As the result, even though the number of HTS codes in the Import Index
expands from 15,283 in 1997 to 22,660 during 1989-2004, I find that only 48 HTS codes
have market compositions that differ from other HTS codes within their I/O commodity
code. The weights that I assign to these 48 HTS codes are proportional to the weight of
the corresponding markets from the import-matrix.
A.4. Complications
There are a number of complications that I encountered in the construction of the
ImpOli Index. All of them have to do with the difficulties involved in merging the various
data sources used in the appendix when utilizing the concordance files obtained from the
agencies maintaining the data. All in all I have had to use concordance files for the four
industry and products coding systems discussed in the previous sections: BEA's six-digit
1/0 commodity coding system, the 1997 and 2002 six-digit NAICS industry coding
system, 1987 four-digit SIC industry coding system, and 1989-2004 ten-digit HTS
imports coding system. Other con~.plications involved the finding that the imports
reported by the BEA import-matrix do not conespond to the aggregated HTS imports
from the NBER trade dataset. I describe these issues and my solutions to them in the
subsections below.
A.4.1 BEA 1/0 vs. NA1CS vs. HTS Codes
One of the steps described in the methods of constructing the Imports Index
required merging data from the import matrix to the HTS imports descriptions. The data
in the import-matrix follows an I/O commodity coding system comprised of 382 series
and constructed by the BEA on the basis of the 1997 NAICS. Consequently, the majority
of the 1/0 codes are identical to NAICS codes from which they are derived (262), a
number of them are equivalent to five-digit NAICS (82) and four-digit NAICS (13), and
the remaining codes map into two or more six-digit NAICS codes (25). Since the impOli-
matrix follows a coding system different from the HTS system, I use the 1997 BEA HTS-
1/0 concordance file to establish a mapping of 1/0 codes to the HTS codes. One issue that
comes up when I attempt to use the concordance file is that a number of HTS impOli
codes do not have an 1/0 code mapping, as the concordance file incorporates HTS's
113
applicable only to 1997 import flows. I deal with this problem in two steps. First, I utilize
the Census Bureau HTS-NAICS Concordance File for 1997, 2000-2004 to map
1989-2004 HTS import codes to six-digit NAICS codes. 57·I find that all of the HTS codes,
with an exception of only a handful, have an HTS-NAICS mapping. 58 Next, I use the
BEA NAICS-IO concordance file to arrive at a mapping ofNAICS to I/O codes, and
consequently HTS to I/O codes. This gives an HTS-I/O code mapping that allows us to
merge the import-matlix data to the HTS imports descriptions data. One last step in the
concording sequence is to make sure that the BEA HTS-I/O mapping corresponds to the
HTS-I/O mapping I derive as the result of the two-step procedure discussed above. I find
that a large number of HTS codes that have two sets of I/O codes from the two sources do
not have matching I/O codes (1193 I/O codes). A number of the BEA I/O codes that map
to the HTS do not belong to the manufacturing sector, while the I/O codes from the two-
stage procedure do (143 I/O codes). This is problematic because the market classification
and weights are roughly based on the BEA's data and I would prefer that only the HTS
codes that are considered by the BEA be included in the manufacturing sector are in fact
included. I deal with the issue of mismatching I/O codes by utilizing the hierarchical
structure of the HTS descriptions. By examining the HTS descriptions with mismatching
I/O codes and the descriptions of the neighboring HTSs, I am able to {;orrect the I/O
codes derived from the two-stage concording procedure to the I/O codes derived from the
BEA HTS-I/O concordance file. I find that for a large majority of the HTS codes, BEA's
I/O codes derived from the BEA concordance file, if different, tend to describe the HTS
impOlis more accurately than tIle ones derived from the two-stage procedure. This finding
indicates the the Census Bureau's classification of HTS impOlis according to the NAICS
coding system is incorrect for over 1600 HTS imports codes over the period of
1989-2004.
As tIle result of the process described above, the Import Matrix contains a
concordance of manufacturing HTS codes to BEA's I/O codes and an improved
concordance of manufacturing HTS codes to NAICS codes over the period of 1989-2004.
Manufacturing HTS codes are those falling undefthe I/O and NAICS definition of the
manufacturing sector (codes within the 300000-399999 range). These manufacturing
HTS codes exclude used and second-hand goods, scrap and waste, and goods under
special classification, i.e. reimpOlis.
57 The Census Bureau HTS Concordance Files for 1997 come from Feenstra (2000) and for 2000-2004
from the Foreign Trade Division website_http://www.census.gov/foreign-trade/schedules/b/2004/imp-
code.txt. The NAICS industry coding system had undergone a series of revisions since its introduction
in 1997. The first revision took place in 2002 and focuses on non-manufacturing industries. None of the
manufacturing industries from the 1997 NAICS differ from the 2002 NAICS. The next revision took
place in 2007 and does not afYect my analysis.
58 I was easily able to fill NAICS codes, for the handful on-ITS that had them missing, from the
neighboring HTS codes.
114
A.4.2. NAICS vs. SIC
The HTS codes included in the Import Index cover the period of 1989 to 2004,
during which the U.S. followed two different systems of industrial classification: the
four-digit SIC system during 1989-1996 and six-digit NAICS system during 1997-2004.
The Census Bureau HTS Concordance File for 1997 contains the mapping of HTS codes
to SIC codes. I am able to extend the SIC system to classify all the HTS import codes in
the Import Index by comparing the HTS imports descriptions that do not have SIC
mapping to the HTS imports descriptions that do have an SIC mapping. Currently, the
Import Index does not'contain all the HTS imports that fall under the SIC definition of
the manufacturing sector and not under the NAICS definition of the manufacturing
sector. This limitation is imposed by the fact that I use the 1997 import-matrix that
follows the NAICS system to derive the market structure classification and market
weights. Additionally, since the market classification and weights of the ten-digit HTS
imports are derived on individual basis, I am not concerned that the Import Index is
biased by the differences between the NAICS and SIC system. The conversion weights
included in the Import Matrix should enable researchers to use the index in both NAICS
and SIC industry-based research of intemational trade. However, in future work I do plan
to extend the Import Index to include market classification and weights for the
manufacturing import on SIC-basis by utilizing the 1992 input-output tables and intend to
compare the results to the ones cUlTently derived from NAICS data.
A.4.3. BEA Imports 1'S. HTS Imports
I compare the HTS imports provided by the U.S. Census Bureau and available in
the 1997 NBER imports data with the impOli figures reported in the BEA import-matrix.
I find that when aggregating the HTS imports up to the BEA I/O codes level of
aggregation using the BEA concordance file, the aggregated HTS imports can be
considerably different fI'om the imports figures reported in the BEA impOli matrix. The
difference between the imports estimated by BEA and actual imports reported in the 1997
NBER imports data does not exceed 10% of the value of the latter for 81 % of the six-
digit I/O commodities, with 77.3% of these imports being overestimated by the BEA. For
11 % of the six-digit I/O commodities, the difference constitutes between 10% and 20% of
value of imports from the NBER data, of which 66% are overestimated by the BEA. For
5% of the six-digit I/O commodities, the differences are between 20 and 50% of the value
of imports from the NBER data, with 77% overestimated by the BEA. Lastly, the
difference of the remaining commodities exceed 50% of the import values, with 75%
overestimated by the BEA. In total, however, the BEA overestimates imports for the
manufacturing sector only by 4% of the value of the NBER impOli data. These
differences between import levels of the import-matrix and the HTS imports may lead
some to argue hat the impOli-matrix should not be used for the purposes of constructing
the Import Index, if the differences result in differences in the market structure and
relative impOliance weights. I examine the causes of these differences and potential
consequences for the market structure below.
115
These differences in the data may be attributed to a number of different sources,
all stemming from the methods by which the BEA tabulates u.s. Census Bureau import
data. The primary source for the BEA import-matrix estimates of trade in goods and
services is the International Transactions Accounts (ITAs), which are prepared by the
BEA's Balance of Payments Division (BP division) (U.S. Department of Commerce
1990). The BP division uses the HTS imports data provided by the U.S. census to classify
import data in broad commodity categories based on the concept of end-use demand. This
end-use commodity classification system was developed to make it easier to relate
changes in merchandise trade to production and income data. Additionally, the U.S.
Census Bureau import data are retabulated to correct for time discrepancies, which arise
when exports or imports of goods are reported by the Census Bureau in one period, but
are actually shipped or received in another. Then, the BEA adjusts the data for coverage
and valuation to bring them into confonnity with the BP concepts and for seasonal
variation. The seasonally adjusted and tabulated import data on the end-use basis are then
incorporated into the BEA input-output and import-matrix accounts by mapping the end-
use codes to the BEA I/O codes (U.S. Department of Commerce 1990). Any stage of this
process may result in the BEA import-matrix imports figures deviating from the raw U.S.
Census Bureau HTS data 1use in my impOli-index classification.
However, given the nature of construction of the import-matrix, where the output
of each commodity in the input-output tables is multiplied by the impOli ratio of the
commodity in total domestic supply of the commodity, one should not see differences in
the market structure and market relative importance weights for the commodity from
those ofthe input-output tables (U.S. Department of Commerce 2006). When comparing
the input-output tables with the import-matrix, indeed, I see that the market structure and
the relative importance weights parallel each other, with only small differences for some
of the commodities. Since all that I derive from the import matrix is the market
composition and weights, expressed as shares, I do not worry about the differences
between the import levels of the import-matrix and the HTS data.
A.5. Conclusion
The Market Structure Index of HTS ImpOlis is to date the most complete
classification of imports by their end-use demand, i.e. intennediate and final demand. The
Imports Index spans manufacturing imports for the period of 1989 to 2004 and can be
easily extended to more recent years, adjusting for changes in the HTS as reported by the
US Trade Commission on their website. The Index is highly reliable, as the market
groups are assigned strictly according to the ten-digit HTS imports descriptions, that
specify the imports physical characteristics and stage of processing and/or the related
industries that the imports serve. Additionally, the market classification in the Imports
Index was cross-referenced against official government sources of market classifi cations
of commodities and industries at highly disaggregated levels. The most notable sources
are the 1997 BEA import-matrix, the BLS commodity-based stage-of-processing
classification, the FRB's Market Structure Index of Industrial Production, and the BEA's
End-Use classification of imports.
116
The documentation provided in this appendix should be considered a working
document to be used in conjunction with the Imp011s Index. Over time, changes will be
made to the document as the Imports Index is updated, new years are added,
methodology is changed, and NAICS redefinitions take place. One major revision to be
expected soon is the inclusion of the SIC-based assignment of market classification and
weights for Import Index's years prior to 1997 and expansion of the index to include
imports in the SIC definition of the manufacturing sector.
--- ----------------
117
APPENDIXB
FIGURES AND TABLES FOR CHAPTER II
Figure B.l: Comparison of New and Old Data of Trade in Intermediate Inputs
1) Imports of Intermediates (Refined) vs. "Part" and "Components" (Original)
1---- Refined Measure ---,4-- Original Measure]
2) Refined vs. Original Feenstra and Hanson (1996)' Measure of Imported Inputs
118
--- Refined Iv]easure - ......- Original Measure
Figure B.2: U.S. Manufacturing Imports by Category, 1989-2004
119
0
0
""
fit 0
0 0C')
:.0
.~ 0
<IJ
0
t N
0
D.
E 0
0
,....
--- Materials Imports
-- Business Supplies Imports
----A- Consumer Goods Import~
--- Capital Goods Imports I
._----
Figure B.3: U.S. Imports by 3-digit NAICS Industry, 1989-2004
Non-Durable Manufacturing
,<' ,# ,i;; ;;'>.1777J# J.iP.iP
NAICSJ13&314· i~le Mills&"ftl:l1od
~ ,4' ,~' ,,," # i ,,,, #' ,#Tn-~
........__.._. _.-.__.._--_.-.~--_.<----~--"
/'
NAICS:I!2·Pa~
.._.._~ ..........--.o..-"""" 4_..._·-J,,··-·-_·······.I.··
"
'--'-- _j/ '---.---'----
...$' 4:>.j ...i i ....J. ..i, ,$--'; *~;'-~~--';'-:f i
"
fb"g
"~
j.~
NAIC5315 &316·App.rcl &lealher/l. Allied
!'j
1il ). ••••"' , ,.0.
~~j ",--'
iSt~·- ~ ..-
-:b- ,=' -t' ••• '=.............
~~~;#~#~~~~~##~J
/
~._.~A, , ,;//'.·y~
.//"-~
_~ • .4(,,,11
.... ......
..
~
~ ... t.-_· ....,
f~
Jr....- ....
....... ~,/
,/
NAJCS3" /10 J12-Fcod 11. E\eYeralle& TDbII=
..__ ....,,_.~ ........-o-.....-,.J./
".::..~:::.t.:; ...:t.~=:::::::::::.~=..--::::~~~~""-"--"'.-'
~
'"~~
~
~~
""
--- M!lll!rT~ls lmpDl1.!l --""_•• OJlllUmer Goods rmpo
-- I:v,.n~%pliesImports -- CapllaIGoa15lmpo~
r== MlItl!mlls Impcrls ----4----- ea,w~Gcod.5lmportsll=:=- 8usines8Suppl~1Tnports -- ~itt.lGooo:hlmp;:lftJ L:" MmerialslmpDlt5 ----...-- Cooillmer~...._-- eusllt9'lS,qllllleslmiM!. - Clpb1Goo1ls.Im~ \ MllleriallJ!i'l1~ Cauluml!J~&lmporbl....- ···~t'»5up""ieslmpor3 --CaJJol\31Goot!,lmpol't:J
o1,f' ...#' ,<p7n-7 :# ,<i ,,I. ) ~ i'; j j
~
#~~~JiiiJJJ~#~;J
NAlC5 323 - Prin~ng
~J
~. -,~.±c:::::cc~:~:::::=:=::=:
01l;:'=;=;::::::;:::=;=;:::;:=;=;::::::;:::=;=;::::::;:::=;=;::::;i~j;~j~;~))i;;JJ
'"
"..
"l:
NAICS 3i4- Petroleum &Coal
,I.
/~//-.~Ii /~~'"bk~~,~/::::':!:::/."
..."~- ......---..- ...._-....
s
'"
NArCS 3.."15. ChemIC'll
.',
.'
....
,/ ".A~--'-._--~--~_.,- :.-_~,A -""'-..' .-
",:"""";:::,,,!::~:_~~:.::.:-.:.~.:~:~.- ...._>.- ....-
,"U'," ;,"i #'ii J,..... ##">7
~
NAICSJ26·plllstit:5e;Rubber
J
.' /_.:=~:::~~:.~:::
I=;rialSlmports --.0._ C005~1IlIGoodslmport!ll
~- 8uSlness5ll;1ph~lmports - Cap~alGc;odslm<:orts
r=;:;lll,rmp;ll't! ._-o."-ConllU~eT~
............ Busino!!Ss Sup~;".lmpclfb ~- ClIpiC!l GaldslmpoltJ
~ MlIIl!lilllslmpOo1s c<Y1'UmI!JGoo;jslmpoj
- BusjneMStJp~~lmportJI - Caprtl!lGoodsl~
1 h1ateria1.Im~s ----w--- Cco-1wmer GlDd, Importsl
~ Busilll!!DSLqlplie:flmp;ll't! ~- CapiiaIGolld5lmfD~
>-'
N
o
Figure B.3: U.S. Imports by 3-digit NAICS Industry, 1989-2004 (Cont.)
Durable Manufacturing
, __ __ _..
1)))J;)}JJ;,J;;;
NAtCS 321 - Wood
"-~'-~-><
NAJcsm·~'Y
~~ ~~nM~~Suppl... lmporlll -CapilalGoa::!lI~
~#~~#~~#~~~$#~~#
~: ~i~~t ,,*~ ::ciC:;:C=;~~;;-:~-=:::~~~~~-=:~~
~'miXf'l!! .l--CoIlSUmer~lm~
=::ElusinN Supplies lmpcW. --- c"pillll Goo:b 1m.,.:>?
NAICS332·Fot:ll1l::lll..dMobl
...........-_...-,../
I~;~::=::.
...I"" ,$ $" .f:'''' ,<f<'" ..of" ,l' $''' ,<f'.... ,$ ,J' ,,"" $" .ff -i-'> 4'''
./
"·-."i
NAICS 331 • PnmBry "'!!\ill
-',
,........-...-.......,. ~ ..
....--~ - -.
1--- MMerilllslmparl' -4.----- Ca>!unwGoodslmport<.:'
::.__._. Bo.:sine-.Suppr",,'mF':uls -.~aJIGoodslmiXf'l!!_
2
NAICS 321· NDTVTl""~lIjtMineml
/
E M:ll~rialr. ImpC>'l!l - ConsuIMr Gocdllmpcr1l!
••m __•• Busin""'ll$Jpplfllllmports - C-'Jp;t11GcodslmpcllU;
~-
i:18X":C".<~;~~:l:
. 'i'-I"""#' ..~* 'f'''' ./('''.:,.Il-.$.... .# ,of>'> ...#' ,#.#' ...i
..'" ......
,
if
/.-~~_=::~:'::-~);
" ,/"'
- ...--- .--....~
E~:::~=~~~~ =:=- ~;~;;=;,=j
~
NAICSJ39.Mi$::ll!Il"neou~
-", .... . " ...,;"._..K
/'
,.,.--,---~_.--,,--~
.-'
~-'-"--'''=---f'''--a=--'~
..~~ ..i ..i ~ ii ,J,~ "';'7i; i J 4"i.#' ..#
g
.'•:i:j~
i~,
/ ..../
/
/'
~
~1~/'............
..-- ..........._.....-....,.,-- ...-.__.-..-,---"
oJft;=::"(fr~~~~,~;-;i;~:
~A.lCS3315· Tr:lMp:lrtstion Equipmenl
..ol>" ..# ..~" $'''' ~.:, .:IN ...$,'; i'~--';7'l
~l~"'~~ ,/.,.~--~~'
_yr_.K ...._<o .....~._•• ~ ... ' ..••
~... ...-;:~:~_ ......~..:.---~/.-.-
NAICS33S· EJ~IEqu;l""'enl&Ap;:Ira!lC1!:S
1 I \ .• I j -.-------.---------r--r
"~"",#,#,,,,~
~l:I~,~~-?
NAteS 334 -Compl4l!f &EJecln:>n(;
,fI' ..J ............."'.. $'.;,i,.,..;i ..i.ii ..#~
g
§
~
,',
".." '\
/ _ ..--..." '00"
~':1=~==~~~~:7:,~:::~:.::~:::~<~~-::
g Mal~i1ll1lmpDl1s Qln~m~
~- Bus<ness SUpplieslm~ - Cap«ol Goods Impol15
~._, MaleriRlslmptYls , Ca:wlm""Gcodslm~
----~. BusiI1ess Sup~il!$lmpcrt:S -- Capl!31 Goods Imp:lrts I E-.·-- M;l\r:riotlroimpol'lS .. .- Consumerc-ood~--- Busine:ss Stlpplia, Imports -- DlpblGc.:w;l,lmport3 F,~ Materials Imports .•.. CcoSIJm~r GoOO, Importsll..:==:.- BU!iIl~9JppIi~lmports -- Cap'tllGoods lm.,.:>rts I .--M9\l!rill1slrniXf'l!!~ -.-....... c;g,""marGoods~•.• BllSlness Sup~ias Impcm -- Capital Goods lm.,.:>rt'I I
.......
N
.......
Table B.l Breakdown of Imports by Utilization Weights, 1989-2004
122
u.s. Materials Imports
Products
Year 25% 50% 75% 100%
1989 73 63 40 16
2004 74 64- 40 16
U.S. Consumer Goods Imports
Products
Year 25% 50% 75% 100%
1989 52 45 35 0
2004 49 42 33 0
25%
91
89
25%
91
91
Volume
50% 75%
84 57
80 52
Volume
50% 75%
84 58
85 62
100%
25
21
100"10
o
1
Note: The columns show the share of products (import volumes) in total products (total import volumes)
by each category, that have more than 25%, 50%, 75%,100% utilization rate as intermediate or
consumer goods.
Table B.2: U.S. Imports Relative Importance, 1989-2004
US. Materials Imports
Volume Products
Year $ %~ ShM ShY # %~ ShM
1989 131.7 7.2 33.0 4.8 8497 1.5 71.2
2004 351.9 7.2 29.1 8.4 10615 1.5 73.9
US. Consumer Goods hnports
Volume Products
Year $ %~ ShM ShY # %~ ShM
1989 133.7 8.3 33.5 4.9 9921 1.2 83.1
2004 435.2 8.3 35.9 10.3 11792 1.2 82.1
US. Total Imports
Volume Products
Year $ %~ ShY # %~
1989 399.1 7.9 14.6 11932 1.3
2004 1210.8 7.9 28.8 14366 1.3
Note: Imports are expressed in billions of U.S. dollars. The % fJ. refers to average annual growth.
The Sh M refers to import share in total manufacturing imports and Sh Y refers to import share in
total manufacturing output. # refers to the number of distinct import products.
123
Table B.3: U.S. Top 10 Industry Imports
u.s. Materials Imports
1989 2004
NAICS Description ShM ShY NAICS Description ShM ShY
334 Computer & electronics 19 10 334 Computer & electronics 24 23
336 Transportation equipment 18 6 336 Transportation equipment 16 9
331 Primal)' metals 17 16 331 Primal)' metals 15 30
325 Chemical products 8 4 325 Chemical products 10 7
322 Paper products 7 8 333 MachineI)' 6 8
333 MachineI)' 5 4 332 Fabricated metal products 5 7
332 Fabricated metal products 4 3 335 Elec. eq., appl., & compnts 4 15
335 Elec. eq., appl., & compnts 3 5 322 Paper products 4 9
326 Plastics & rubber products 3 4 321 Wood products 3 11
313 Textile mills & products 3 5 326 Plastics & rubber products 3 6
U.S. Consumer Goods Imports
1989 2004
NAICS Description ShM ShY NAICS Description ShM ShY
315 Apparel, leather & allied 25 50 315 Apparel, leather & allied 21 258
336 Transportation equipment 22 7 336 Transportation equipment 19 12
334 Computer & electronics 13 7 325 Chemical products 13 11
339 Miscellaneous 13 26 339 Miscellaneous 12 39
311 Food, beverage & tobacco 7 2 334 Computer & electronics 12 14
325 Chemical products 4 2 311 Food, beverage & tobacco 6 4
324 Petroleum & coal products 4 4 324 Petroleum & coal products 4 5
335 Elec. eq., appl., & compnts 3 5 335 Elec. eq., appl., & compnts 3 15
333 MachineI)' 2 1 337 Furniture & related prod. 3 18
313 Textile mills & products 1 3 313 Textile mills & products 2 12
Note: Sh M refers to materials/consumer goods import share in total materials/consumer goods imports and Sh Y
refers to materials/consumer goods import share in industry output.
124
Table B.4: Pro-Cyclical Behavior of U.S. Imports
Correlations
~ Materials ~ Cons umer
Imports Imports
125
~ Output
~GDP
0.48
0.26
0.20
0.35
Dependent Variable: ALn (Imports)
I II
~ Ln (Output) 0.127* **
[0.031]
X Materials Dummy 0.071***
[0.039]
~ Ln (GOP) 0.054* **
[0.007]
X Materials Dummy 0.000
[0.009]
ObselVations
Number ofYears
Number of Industries
R-squared
540
15
18
0.11
540
15
18
0.20
Note: Standard errors are given in parenthesis. Dependent
variable is first-difference of natural log of U.S. imports of
intermediate inputs and consumer goods. Sample data
contain U.S. imports from three-digit NAICS industries for
1990-2004. Imports are deflated by CPI. GDP and Output
are measured in chained 2000 dollars.
126
Table B.5:List of Sample Countries
ASEAN SWEDEN EAST GERMANY LITHUANIA SOUTH YEMEN
BRUNEI & BHUT AN SWITZERLAND ECUADOR MACEDONIA SRI LANKA
CAMBODIA UK EGYPT MADAGASCAR ST. KITTS-NEVIS
INDONESIA EL SALVADOR MALAWI ST. PIERRE & MIQ.
SOUTH KOREA OWER EQUATORIAL GUINIMALI SUDAN
LAOS AFGANISTAN ESTON1A MALTA SURINAME
MALAYSIA ALBANIA ETHIOPIA MAURITANIA SYRIA
MYANMAR ALGERIA FALKLAND ISLS. MAURITIUS TAJIKISTAN
PHILIPPINES ANGOLA FIJI MONGOLA TANZANIA
SINGAPORE ARGENTINA FORM. YUGOSLAV. MOROCCO TOGO
THAILAND ARMENIA FRENCH GUIANA MOZAMBIQUE TRINIDAD & TOB.
VIETNAM ARUBA & N. ANT. GABON NEPAL TUNISIA
CANADA AZERBAIJAN GAMBIA NEW CALEDONIA TURKEY
BAHAMAS GEORGIA NICARAGUA T URKMENIST AN
CHINA BAHRAIN GHANA NIGER US OUTL. ISLS.
CHINA (MAINLAND)BANGLADESH GIBRALTAR NIGERIA UGANDA
HONG KONG BARBADOS GREENLAND NORTH KOREA UKRAINE
MACAU BELARUS GUADELOUPE OMAN U. A. EMIRATES
TAIWAN BELIZE GUATEMALA PAKISTAN URUGUAY
JAPAN BENIN GUINEA PANAMA USSR
BERMUDA GUINEA-BISSAU P. N.GUINEA UZBEKIST
MEXICO BOLIVIA GUYANA PARAGUAY VENEZUELA
BOSNIA-HERZEG. HAITI PERU YEMEN ARAB REP.
OEeD BRAZIL HONDURAS POLAND ZAIRE
AUSTRAL BULGARIA HUNGARY PUERTO RICO ZAMBIA
AUSTRIA BURKINA INDIA QATAR ZIMBABWE
BELGIUM BURUNDI IRAN R. OF MOLDOVA
DENMARK CAMEROON IRAQ REUNION
FINLAND C. AFRIC. REP. ISRAEL ROMANIA
FRANCE CHAD IVORY COAST RUSSIA
GERMAN CHILE JAMAICA RWANDA
GREECE COLOMBIA JORDON SAMOA
ICELAND CONGO KAZAKHSTAN SAUDI ARABIA
IRELAND COST A RICA KENYA SENEGAL
ITALY CROATIA KIRIBATI SERBIA & MONT.
LUXEMBURG CUBA KUWAIT SEYCHELLES
NET HERLANDS CYPRUS KYRGYZST AN SIERRA LEONE
NEW ZEALAND CZECH REPUBLIC LATVIA SLOVAKIA
NORWAY CZECHOSLOVAKIA LEBANON SLOVENIA
PORTUGAL DJIBOUTI LIBERIA SOMALIA
SPAIN DOMINICAN REP. LIBYA SOUTH AFRICA
Table B.6: u.s. Import Value Market Share by Region, 1989-2004
us. Materials ImJDrts
Canada CHINA Japan Mexico ASEAN OECD OTHER
-
IndustJY 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004
Chemicals (325) 21 24 3 6 13 9 4 5 2 5 48 38 9 13
MachineJY (333) 11 10 3 7 28 23 3 12 3 3 45 39 6 7
Compo & EIec. (334) 7 5 12 30 35 11 7 11 27 30 10 9 2 3
Electric. Equip. (335) 11 9 11 17 28 12 19 32 5 6 22 20 4 4
Transp. Equip. (336) 32 25 2 5 27 18 10 24 2 3 25 22 2 4
Nondurable Manuf. 35 31 5 9 8 6 3 5 6 6 29 28 14 15
Durable Manu£ 23 18 6 16 24 11 7 14 9 12 23 18 7 12
Total Manu£ 26 21 6 14 19 10 6 12 8 10 25 20 9 13
US. Consumer Goo~ ImJDrts
Canada CHINA Japan Mexico ASEAN OECD OTHER
--
IndustJY 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004
Chemicals (325) 9 7 2 2 14 6 2 2 1 2 63 76 9 5
MachineJY (333) 10 6 7 33 43 22 4 11 8 7 25 18 4 2
Comp. & Elec. (334) 3 1 15 37 43 12 11 19 22 21 5 8 0 1
Electric. Equip. (335) 5 6 36 50 16 4 8 15 15 9 17 12 3 2
Transp. Equip. (336) 28 28 2 3 43 26 3 10 4 7 20 25 1 2
Nondurable Manu£ 5 8 26 21 3 2 3 6 19 10 23 32 21 22
Durable Manu£ 13 12 15 28 33 14 5 10 11 11 18 17 5 8
Total Manu£ 10 10 20 25 20 8 4 8 14 10 20 24 12 14
Note: Figures express import shares of each country/region in total U.S. imports by the specified industry and year. Rows add up to 100 per year. CHINA
refers to mainland China, Hong Kong, Taiwan, and Macao. The six country groupings are mutually exclusive.
........
tv
--J
128
Table B.7: Largest Gains in Market Share, 1989-2004
u.s. :Materials Imports U.S. Consumer Goods Imports
Country :Market Share Ails. b. %b. Country :Market Share Ails. b. %b.
1989 2004 1989 2004
China 1.01 10.27 9.26 919 China 5.88 21.18 15.30 260
Mexico 6.04 11.70 5.66 94 Mexico 4.26 8.27 4.01 94
Russia 0.00 1.65 1.65 N/A Ireland 0.41 3.94 3.53 866
Malaysia 1.47 2.86 1.39 94 Gennany 4.12 5.09 0.97 24
Ireland 0.33 1.13 0.80 247 Vietnam 0.00 0.88 0.88 N/A
India 0.39 0.85 0.46 119 Israel 1.29 1.98 0.68 53
Thailand 0.65 1.10 0.45 70 Indonesia 0.68 1.26 0.58 85
S. Korea 3.21 3.56 0.35 11 Honduras 0.09 0.66 0.57 637
Philippines 0.65 0.99 0.34 53 UK 2.66 3.21 0.55 21
Peru 0.21 0.54 0.33 158 India 1.65 2.15 0.51 31
Note: Here China refers to only mainland China. Percentage changes are not available if a country's share of U.S.
imports in 1989 was zero.
Table B.8: Product Penetration by Region, 1989-2004
u.s. Materials ImJXlrts
Canada CillNA Japan Mexico ASEAN OECD OTHER
Industry 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004
Chemicals (325) 60 56 35 74 74 65 31 36 23 39 97 93 45 69
Machinery (333) 94 94 79 89 95 93 59 74 66 82 99 99 66 87
Comp. & Elec. (334) 78 75 94 95 94 90 68 72 86 86 97 94 65 75
Electric. Equip. (335) 83 85 89 95 94 88 69 77 78 85 99 97 64 81
Transp. Equip. (336) 92 91 79 91 86 88 68 75 66 80 98 97 59 84
Nondurable Manuf 59 60 46 66 59 49 30 41 39 49 93 89 49 69
Durable Manuf 85 83 73 86 85 78 56 65 65 73 96 95 59 78
Total Manuf 70 69 57 75 70 61 41 51 50 60 94 92 54 73
U.S. Consumer Good<; ImJXlrts
Canada CillNA Japan Mexico ASEAN OEeD OTHER
Industry 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004 1989 2004
Chemicals (325) 59 55 35 74 73 64 30 35 23 37 96 93 47 69
Machinery (333) 91 92 79 92 93 87 50 64 62 74 98 99 60 82
Comp. & Elec. (334) 65 68 89 93 87 86 51 62 76 79 97 96 56 73
Electric. Equip. (335) 85 86 91 97 95 87 70 78 82 90 99 98 65 83
Transp. Equip. (336) 85 84 72 76 82 81 60 60 63 72 100 98 59 75
Nondurable Manuf 58 62 60 74 55 48 32 46 51 58 92 90 55 74
Durable Manuf 80 81 82 90 84 76 56 65 71 77 96 96 63 80
Total Manuf 66 68 67 80 65 58 40 53 58 65 93 92 58 76
Note: Figures express import product shares of each country/region in total U.S. import products by the specified industry and year. A product is included in
the share if it is imported in the U.S. by at least one country in the region. CHINA refers to mainland China, Hong Kong, Taiwan, and Macao. The six
country groupings are mutually exclusive.
.......
N
\0
Table B.9: Largest Gains in Product Penetration by Country, 1989-2004
US. Materials Imports US. Consumer Goods Imports
Country Market Share Ails. /1 %/1 Country Market Share Ails. /1 %/1
-~--~~
1989 2004 1989 2004
---~--~--~--~----~--
China 34 70 37 108 China 45 76 32 71
Mexico 19 44 24 127 Mexico 22 49 26 116
Russia 6 20 14 223 Ireland 10 25 15 141
Malaysia 8 20 12 143 Gennany 14 27 13 94
Ireland 17 28 12 69 Vietnam 6 19 13 212
India 6 17 11 175 Israel 24 36 13 53
Thailand 41 51 10 25 Indonesia 40 53 12 31
S. Korea 7 17 10 129 Honduras 7 20 12 164
Philippines 11 21 9 83 U.K 33 44 11 34
Peru 31 40 9 30 India 15 23 8 56
Note: Here China refers only to mainland China.
130

132
Table B.ll: Product Shares by Source Country Income Levels, 1989-2004
u.s. Materials hnports
Income Breakdown H M L LMH MH LM
19892004- 19892004- 19892004- 19892004- 19892004- 19892004-
World Bank 22 15 0 0 29 50 48 33 0 0
World Bank* 22 15 0 0 10 17 48 33 0 0
L<40th, 40th::;M<60t\ 60thSH 34 30 0 0 0 0 38 55 27 14 0 0
L<40t\ 40th SM<60t\ 60th$H* 34 30 0 0 0 0 18 25 27 14 0 0
L<30th , 30th:SM<70th , 70thSH 28 21 0 0 31 50 41 28 0 0
L<30'h, 30thSM<70'h, 70th :SH* 28 21 0 0 11 16 41 28 0 0
L<30th, 3Oth:SM<90th , 90th$H 6 7 3 4 0 0 30 49 60 39
L<30'h, 30thsM<90th , 9Oth :SH* 6 7 3 4 0 0 11 16 60 39
L<20t\ 20th:SM<80th , 80tl'SH 18 15 2 2 0 0 7 9 73 74 0 0
L<20th, 20th :SM<80th , 80th$H* 18 15 2 2 0 0 1 2 73 74 0 0
U.S. Cons umer Goods Imports
Income Breakdown H M L LMH MH LM
19892004- 19892004- 19892004- 19892004- 19892004- 19892004-
WDB Breakdown 20 13 0 0 36 57 43 28 0 0
WDB Breakdown * 20 13 0 0 16 27 43 28 0 0
L<40th, 40th:SM<60th , 60th$H 31 25 0 0 0 46 62 23 12 0 0
L<40th, 40thSM<60th , 60th :SH* 31 25 0 0 0 25 35 23 12 0 0
L<30'\ 30th:SM<70th, 70'hSH 26 18 0 0 37 57 36 24 0 0
L<30th, 30th:SM<70t\ 70th$H* 26 18 0 0 17 27 36 24 0 0
L<30t\ 30tl1SM<90th , 90'h:SH 5 5 5 4 0 0 36 56 53 33 2
L<30t\ 30thSM<90tl1 , 90th$H* 5 5 5 4 0 0 17 26 53 33 2
L<20tl', 20th:SM<80th, 80th:SH 16 13 2 2 0 0 12 17 70 68 0 0
L<20th, 20th:SM<80tl" 80th$H* 16 13 2 2 0 0 4 8 70 68 0 0
Note: Figures express import product shares by countries grouped into income levels. T he income level breakdown
follows the one indicated in the first column, where number refer to percentiles and H - high income, M-middle
income, L -low income countries. LMH products originate simultaneously from at least one low- or one high-income
countries. MH products originate from at least one middle- and one high-income countries. LM products originate
from at least one low-income and one middle-income countries. The six source country groupings are mutually
exclusive. • refers to an extra restriction in the construction of LMH products, where products which originate from
only one low-income country are dropped.
133
Table B.12: Regression of Unit Values on Income, 1989-2004
Dependent Variable: Log(Unit Values)
All 25% 50% 75% LMH
I II III IV V
Total Manufacturing
1..0g(Real GDPpc) 0.171 *** 0.189*** 0.191*** 0.186*** 0.170***
[0.032] [0.035] [0.036] [0.034] [0.032]
Obs. 1628940 1203789 1003836 650399 1535165
Null. OfProducts 14082 10533 9123 5625 10727
R-squared 0.08 0.08 0.08 0.10 0.08
Non-Durable Manufacturing
Log(Real GDPpc) 0.171*** 0.157*** 0.123*** 0.124*** 0.169***
[0.027] [0.027] [0.031] [0.032] [0.026]
X Cons umer Dummy 0.002*** 0.019*** 0.071*** 0.061 *** 0.002***
[0.000] [0.007] [0.022] [0.021 ] [0.000]
Obs. 2300591 1558263 1413440 1098902 2184750
Null. OfProducts 12410 12196 11241 7795 9624
R-squared 0.10 0.12 0.14 0.19 0.11
Durable Manufacturing
Log(Real GDPpc) 0.170*** 0.163*** 0.143*** 0.144*** 0.170***
[0.041 ] [0.044] [0.047] [0.049] [0.042]
X Consumer Dummy 0.000* 0.016 0.044* 0.029 0.000*
[0.000] [0.011] [0.026] [0.035] [0.000]
Obs. 1575327 887844 699630 466340 1518529
Null. OfProducts 6872 6020 4961 3478 5667
R-squared 0.10 0.10 0.11 0.11 0.11
Units of observation are product-country-year. Unit Values data comes from Feenstra (2002) and U.S.
Census Bureau (2004), real per capita GDP are from WDI (2007), Consumer dummy takes a value of I if
a product is a consumer goods, and zero otherwise. Each regression includes product and year fixed
effects, as well as regional dummies (see Table B.5 for description of regional breakdown). Columns II-IV
restrict the sample to only those products that have more than 25%,50%,75% use, respectively, as
intermediates or consumer goods. Column V restricts the sample to only those products that are sourced
simultaneously from one low- and one high income country, where income breakdown follows World
Bank classification. Robust standard errors adjusted for source country clustering are noted below
coefficients. Results for fixed effects and constant are suppressed. ***, **, and * refer to statistical
significance at the 1 percent,S percent, and 10 percent levels, respectively.
Table B.B: Regression of Unit Values on Income by Select Industries, 1989-2004
Dependent Variable: Log(Unit Values)
All 25% 50% 75% LMH
I II III IV V
Chemicals .Manufacturing
Log(Real GDppc) 0.235*** 0.239*** 0.237* ** 0.209* ** 0.235* **
[0.041] [0.044] [0.045] [0.044] [0.040]
X Consumer Dummy -0.000* -0.006 0.046* 0.054* -0.000*
[0.000] [0.019] [0.024] [0.033] [0.000]
Machinery.Manufacturing
Log(ReaIODPpc) 0.174*** 0.148** 0.168* ** 0.217*** 0.176***
[0.062] [0.057] [0.063] [0.078] [0.062]
X Cons umer Dummy 0.000 0.052 0.040 -0.268*** 0.000
[0.000] [0.072] [0.091] [0.099] [0.000]
Computers and Electronics Manufacturing
Log(Real ODPpc) 0.146*** 0.156* ** 0.161* ** 0.158*** 0.146* **
[0.042] [0.054] [0.055] [0.055] [0.042]
X Consumer Dummy 0.000 -0.025 -0.039 -0.036 0.000
[0.000] [0.044] [0.047] [0.049] [0.000]
Electrical Equip./Appliances Manufacturing
Log(ReaIODPpc) 0.139** 0.150* * 0.147** 0.199* * 0.139* *
[0.061] [0.064] [0.065] [0.091] [0.061]
X Consumer Dummy 0.000 -0.002 -0.007 -0.099 0.000
[0.000] [0.003] [0.041] [0.075] [0.000]
Transportation Equipment .Manufacturing
Log(Real GDppc) 0.172*** 0.148** 0.147** 0.146* ** 0.175***
[0.060] [0.060] [0.060] [0.038] [0.061]
X Consumer Dummy 0.000 0.077 0.086* 0.116** 0.000
[0.000] [0.047] [0.049] [0.045] [0.000]
Units of observation are product-country-year. Unit Values data comes from Feenstra (2002) and U.S.
Census Bureau (2004), real per capita GDP are from WDI (2007), Consumer dummy takes a value of I if
a product is a consumer goods, and zero otherwise. Each regression includes product and year fixed
effects, as well as regional dummies (see Table 8.5 for description of regional breakdown). Columns II-IV
restrict the sample to only those products that have more than 25%,50%,75% use, respectively, as
intermediates or consumer goods. Column V restricts the sample to only those products that are sourced
simultaneously from one low- and one high income country, where income breakdown follows World
Bank classification. Robust standard errors adjusted for source country clustering are noted below
coefficients. Results for fixed effects and constant are suppressed. ***, **, and * refer to statistical
significance at the I percent, 5 percent, and 10 percent levels, respectively.
134
135
APPENDIX C
DATA APPENDIX TO CHAPTER III
c.l. Productivity Database Extension
Most of the data used in construction of the non-structural variables are obtained
from the NBER Productivity Database (PD). The NBER PD extends as far as 1996 on
1987 SIC basis and incorporates data on shipments, employments, materials, inventory,
energy, investment, capital stock, deflators, and TFP measures for 458 industries. Since
my analysis goes as far as 2004, I extend the NBER PD following the methodology
outlined in Bartelsman and Grey (1996). I describe the construction of each of the
variables of the PD extension and the data issues encountered on the way below. The
final PD extension spans 1997-2005 and in addition to the NBER PD variables, includes
two versions of output price deflators, cost of selected services, and services deflators for
473 six-digit NAICS industries.
C. J. J Industry Statistics
Data on shipments, employment, materials, inventory, energy, and investment
come from the Annual Survey of Manufacturers, which are currently available for
1997-2005 and can be downloaded from the Census website. I have identified two issues
with the ASM data. First, while the industlies in the1997-2001 ASM data follow six-digit
NAICS, the industries in the 2002-2005 data follow NAICS-based code which aggregates
some six-digit NAICS industries into two to five grouped Census-defined industry code.
In order to break down the Census-code industries data into data for each of the
embedded six-digit NAICS industries, I aggregate the data from 2001 ASM into the
corresponding Census code industries. Then, for each industry statistic of six-digit
NAICS industries in 2001, I calculate its share in the respective aggregated industry
statistic of the corresponding Census-code industry of 2001. These shares are then used to
impute the six-digit NAICS industry data from the Census-code industry data in
2002-2005. Since energy data is available as total energy, fuel and electricity purchases, I
first break down fuel and electricity and then aggregate these to create the broken down
total energy purchases. The break-down method for investment, which is subdivided into
structures and equipment investment, is slightly different. I first used the method
described above to obtain total investment for the six-digit NAICS industries. The broken
down structures and equipment investment are constructed by applying the shares of
equipment and structures of the corresponding Census-code industry in its total
investment for 2002-2005 to the broken down total investment for the six-digit NAICS
136
industries within the Census-code. Thus, I assume that the six-digit NAICS industries
embedded in the Census-code industry invest in structures and equipment in the same
proportions as the overarching Census-code industry. I justify this method by noting that
since investment in structures and equipment takes place in discrete amounts, one cannot
assume that propOliions of 2001 will hold up during 2002-2005.
The second issue is similar to the one experienced by Bmielsman and Gray
(1996), where some industries in the ASM data have missing information due to the
disclosure reasons. I were able to approach the issue in two ways. For some missing
observations of six-digit NAICS industries, I were able to subtract the existing data for
other six-digit NAICS from the data of the overarching five-digit NAICS industries. If
data for five-digit NAICS were not disclosed, I used the same method to first obtain the
missing five-digit NAICS data from the overarching four-digit NAICS data. This method
took care of all the missing observations but the ones due to energy and investment,
where multiple industries within a five-digit NAICS would have missing infonnation. I
remedied this issue by first obtaining the aggregated data for the multiple industries with
missing observations by the method of subtraction the existing data of six-digit NAICS
from five-digit NAICS. Then the aggregated data was broken down for total energy, fuel,
and electricity, by the average shares of these variables in the aggregated data of the
nearby years, for which full data was available. The aggregated data for investment,
equipment, and structures, was broken down by the share of the aggregated equipment
and structures in the aggregated total investment of the same years. Once again, I did not
use the data from the nearby years for the investment variables, since investment of one
year does not have to follow the investment pattems of the previous year.
C l. 2 Shipment Price Deflators
In the NBER PD, output price deflators data come from the BEA shipments price deflator
data. While the BEA produces the shipment price deflators for 1997-2005, the data come
with a disclaimer about the lack of precision in the data. This is true because the BEA
basis its shipment price deflator data on the BLS producer price index data for each six-
digit NAICS industry, where 130 observations are missing for some industries and years.
Since the changes in product prices are integral to the two-stage estimation, upon
consulting the BLS, I construct my own output price deflators from the producer price
indexes. I replace the missing observations with the related commodity price indexes or
conveliing the existing SIC indexes into NAICS. While the differences between my
deflators and BEA deflators is notable, the TFP calculations using each of the defl ators
yield near identical values. The PD extension includes my version of the output price
deflators as the default prices, and the BEA shipment price deflators as alternative prices.
Cl.3 Materials Deflators
Materials deflators are constructed for each industry as the sum of materials
supplying industry PPI's weighted by the share of material purchases from that supplying
industry in total matelial purchased of the purchasing industry. The wei ghts are 0 btained
137
from the 1997 input-output tables, since this the only benchmark input-output table
available to date. The 2002 benchmark input-output tables have been released as of the
writing of this dissertation. The six-digit NAICS materials include materials from
manufacturing and non-manufacturing sectors, where the latter includes agriculture,
logging, mining and utilities. The BLS does not post PPI's for the agriculture industry.
Having consulted the BLS staff, I average out the price indices of the commodities
produced by each six-digit NAICS agriculture industry. While the BLS staff had provided
us with the BLS commodity code - NAICS mapping, the concordance does not contain
relative importance weights for multiple commodity codes mapped in the one NAICS
industry. As the result, the constructed agricultural PPI's are the equally weighted average
of commodity price indexes, provided by the BLS. There were a number of six-digit
NAICS, for which some commodities had missing price indexes either partially or
entireli9 . A small number ofNAICS had no commodity price index data, which I
excluded from the material deflator calculations6o . One drawback of the material deflator
construction method described above, which is outlined in Bartelsman and Grey (1996),
is that thePPI data does not contain changes in the shipment and retail margin prices.
This implies that the materials deflator data does not reflect the actual price changes
experienced by the materials purchasing industries.
C 1.4 Services Deflators
Services deflators are constructed for each industry as the sum of services
supplying industry PPI's weighted by the share of services purchases from that supplying
industry in total services purchases of the purchasing industry. The weights are obtained
from the 1997 input-output table, since this is the only benchmark input-output table
available to date. I restrict services to only those related to the infonnation services
(NAICS 5112, 518, 514); professional scientific support services (NAICS 5411-5119);
and administrative and suppOli services (NAICS 5614). PPls are available for only a
limited number of these services (5112, 518, 514, 5411, 5412, 5413, and 5418). Services
deflators are not available in the NBER PD and could not be constructed for years prior
to 1997.
C 1.5 Capital Stock and Investment Deflators
As described in the NBER Productivity Database, the staIiing point for the
process of creating real capital stock series is a set ofless aggregated industry capital
stock estimates. I use PRB 4-digit NAICS net capital stocks as the basis for my 6-digit
NAICS estimates 61 • The PRB 4-digit net capital stock data are based on 4-digit
59 These NArCS codes and their respective commodity codes are listed as follows: 111199:01220415;
111320:01110107; 111334:01110225; 111335:01190105; 111339:01110206; 114111:02230102,
02230103,02230134,02230135; 114112:02230503,02230504
60 The following NArCS do not have a commodity code mapping, which prevents us from constructing
PP1 data: 111160, 111136,1114,111910,111930,111991, 11199R, 112111,112130,111234,112420,
112511,112512,112519,112910, 112920, 112930,112990,114119,113110, ]13220,2213,230320
6] r thank Jaim Stevens of FRB for providing me with these da ta
138
investment series for plant, equipment, and software of the Annual Survey of
Manufacturers, and the 1997 industry-asset type investment flow matrix, producer
durable equipment deflators, and a table of mean service lives by asset type from the
BEA. The 4-digit data are convelied to the 6-digit level by assuming that the industry-
asset type flows are the same for all 6-digit industries within a 4-digit. With this
assumption in mind, I are able to use the FRB 4-digit data on real and nominal
investment by asset type (structures, equipment, software) and create investment
deflators, which I use to create real investment at 6-digit NAICS level. The initial 6-digit
real capital stocks for 1997 are created using the ratio of 6-digit to 4-digit real (net
capital) from the Annual Survey of Manufacturers. I construct the implied "depreciation"
from the 4-digit capital stock and and real investment data by using Kit=( 1-8i) Kit-l+li to.
Now I can successively add real investment in equipment and structures and subtract the
"depreciation" to create real net capital stocks from 1997-2005.
C.2 Non-Structural Variables
C2.1 Factor Cost-Shares
I calculate factor cost-shares by dividing payment to each factor by the value of
shipments, in nominal terms. The factor cost-share of services cannot be derived from the
ASM data. I assume that six-digit NAICS industries have the same share of services costs
as the over-arching four-digit NAICS. The data for the latter comes from the BLS input-
output tables for 1997-2004, which are provided on four-digit NAICS levels. The
services cost-shares for years prior to 1997 are obtained at three-digit SIC level from the
BLS input-output table for 1989-1996.
C2.2 ractor Prices
I proxy prices of unskilled and skilled labor by the ratio of production and
nonproduction wages to the number of production and nonproduction workers employed,
respectively. The price of capital is calculated by dividing the payments to capital in each
industry (which equals value of shipments less payments to labor and materials) by the
quantity of capital. In the specifications where services are netted out from value added
prices and TFP calculations, payments to services are also netted out from the payments
to capital. Materials, energy, and services price deflators are used to calculate log change
in the respective prices.
C2.3 Value-added product prices
The log change value-added product price is measured by the formula provided in
the text, ,6 ln p ;/1 == 1,6 In p il - h(r il - I + I' il) , ,6 In p;;' J ,where r ii-I and r il are the materials
cost-shares of industry i= 1, ... , N, averaged over the two periods and ,6 In p%' is the
change in log price of intennediates. The product price data comes from the output
deflator data, and tIle price of intermediates comes from the materials deflator data from
the NBER PD and PD extension An altemative specification of value-added prices is the
139
. change in log product price net of the average cost-share weighted change in log price of
intennediates and services.
C.2.4 Primal TFP
Primal total factor productivity is constructed as the difference in the growth of
value added (log change) and cost-share weighted growth of primary factors (log
change). The value-added is calculated as the growth in real shipments (log change)
minus the average cost-share weighted growth in real materials payments (log change). In
the alternative specification ofTFP net of services, the growth of value-added is
constructed as the growth of real shipments net of weighted growth of real materials and
services payments.
C.3 Structural Variables
C.3. J Technology
The data I use for technology variables, i.e., office equipment share, other high-
technology share, and computer share, have been supplied to us by Randal Kinoshita of
BLS. These data are available in 2000 constant dollars and distinguish capital by asset
type for 1948-2002 on 2-digit SIC level and 1987-2005 on 3-digit NAICS level.
Berndt and Morrison (1995) define high-technology capital to include office,
computing, and accounting machinery; communications equipment; science and
engineering instruments; and photocopy and related equipment. This definition of high-
technology capital does not incorporate computers. The data currently available to us
breaks assets up slightly differently. On SIC level, the high-technology capital is broken
up into the following: office, computing, and accounting machinery (asset 14) and
communications equipment (asset 16) had stayed the same, while instruments category is
broken up into photocopy and related equipment (asset 27); medical equipment and
related (28); electromedical (29); and other medical (30). On NAICS level, the high-tech
capital is broken up into the following: office and accounting machinery (asset 4);
communications equipment (6), photocopying and related equipment (26); medical
equipment and related equipment (27); electromedical instruments (28); nonmedical
instruments (29). Similarly to Berndt and Morrison (1995), I separate high-technology
capital into office equipment (SIC asset 14 and NAICS asset 4), and other high-tech
capital. I also define computer capital to include SIC assets 32-42 and NAICS assets
33-43, which is not considered in Berndt and Morrison (1995).
To calculate the technology shares, I first calculate the capital services incurred
from each type of high technology capital (office equipment, computer, and other high-
tech capital) by summing the production of the productive stock of assets and the assets'
user costs over all assets in each type of high-technology capital. I then divide the office
equipment, computer, and other high-tech capital services by the total productive stock
services, obtained using the same method. I use two measures of use costs, ex post and ex
ante user costs. Ex post use cost (or internal rental plice) is provided by BLS and are
140
calculated as in Hall and Jorgenson (1967), and reflect the internal rate of return in each
industry and capital gains on each asset. On the other hand, ex ante use cost used by
Berndt and Monison (1995) reflect a "safe" rate ofreturn and excludes capital gains on
each asset. The "safe" rate of return is measured by Moody's Baa Corporate Bond rate,
which I obtain from St. Louise FRB on monthly basis and average out to get the annual
rate.
A practical problem arises when capital income in national accounts (gross
operating surplus) becomes negative or assets undergo a very high revaluation. In such
cases, the measured rental prices using internal rate of return may also become negative,
which is theoretically inconsistent. One way of eliminating such negative rental prices is
to employ an extemal rate of return. Following Harper, Berndt and Wood (1989) I take a
constant rate at 3.5%, which is the difference between nominal discount rate and inflation
rates in the US as calculated by Fraumeni and Jorgenson (1980) (see Harper et a1. 1989 or
Erumban 2004, pg 13). Thus I substitute internal rate of return in rental price fommla
(13) with a 3.5. Note that the 3.5 rate of return is assumed to be a real rate ofretum (net
of capital gains).
C3.2 Outsourcing and Import Openness
The construction of these measures of outsourcing and import openness follows
the descriptions provided in Appendix AA and A.5. The data for the measures come from
the BLS input-output tables, U.S. impOlis from Feenstra (2000) and the Census Bureau,
and the Market Classification of HTS Imports provided in this appendix. Foreign services
outsourcing is constructed using the services inputs and impOlis infonnation from the
BLS input-output tables. The services are limited to infonnation; professional, scientific,
and technical; and administrative and suppOli services. The corresponding NAICS and
SIC industries are provided in the Table C.l.
141
Table C.l: Selected Services
Services Break-Down on 2002 NAICS-oosis
InfOrmation
Software publishers 5112
Internet and other 518 I
Professional, scientific, and technical services
Legal 5411
Accounting, tax preparation, bookkeeping, & payroll 5412
Architectura~ engineering, & related 5413
Specialized design 5414
Computer systems design & related 5415
Managerrent, scientific, & technical cons ulting 5416
Scientific research & developrrent 5417
Advertising 5418
Other professional, scientific, & technical 5419
Administrative and support
Business support services 5614
Services Break-Down on 1987 SIC- oosis
81
872, 89
871
737
874
873
731
732, 733, 738
InfOrmation; professional, scientific, and technical services;
administrative and support
Legal
Accounting, auditing, & related
Engineering, architectural, & related
Computer, data processing, & related
Managen-ent & public relations
Research & testing
Advertising
Miscellaneous business
'Note, that this 2002 NAICS translates to 514 1997 NAICS
Price data found for 51 12,518,5411,5412,5413,5418 only
142
APPENDIXD
FIGURES AND TABLES FOR CHAPTER III
Figure D.l: U.S. Wage Inequality, 1963-2005
200519691963
:,
(j)
OJ
~
"0
o N
'- .
0..""'-
--OJOJ
cu
S I
"0 '
et .,...- I
c ':l~ JL,--~_-----',----------,-----,-----_-,--, --,-----i-,-----,---------,-
1~5 1%1 1%7 1~3 1~n~9
- - Manuf. Sector (SIC) --9-S- Manuf. Sector (NAICS)
- - Manuf. Ind. Wt. Avg. (SIC) - .....""'- Manuf. Ind. Wt. Avg. (NAICS
Table D.l: Summary Statistics of Non-Structural Variables
1979- 1990 1989- 1996 1997-2004
Average Annual Average Annual Average Annual
(percent) change (percent) change (percent) change
Change in log factor prices
Production labor 4.99 2.67 3.02
Nonproduction labor 5.42 3.78 2.76
Capital 3.98 2.91 0.27
Materials 3.29 0.88 1.66
Energy 3.31 2.00 4.55
Selected Services 2.62
Factor cost-shares:
Production labor 13.41 -0.18 12.03 -0.17 11.44 -0.12
Nonproduction labor 10.66 0.01 10.14 -0.15 8.91 0.01
Capital 32.06 0.33 35.12 0.32 38.30 0.25
Materials 53.41 -0.06 52.95 -0.02 50.55 -0.08
Energy 2.45 -0.01 1.86 -0.02 1.83 0.03
Selected Services 2.53 0.02 4.38 0.19
Change in productivity
Primal TFP 0.80 0.70 0.43
PrimaIETFP 0.78 0.68 0.40
Change in product prices
Value-added 1.53 0.67 0.12
Note: Both averages and changes are weighted by the industry share of total manufacturing shipments, except
changes in log primary factor prices, which are weighted by the industry share of total manufacturing payments to
that factor. All variables are computed over 452 four-digit SIC industries in 1979-1988 and 1989-1996 and 472
six-digit industries in 1997-2004. The data come from the NBER PD (Bartelsman and Gray 1996) and the PD
extension of it for 1997-2005 based on the data from Annual Survey of Manufacturers, Federal Reserve Board, and
Bureau of Labor Statistics.
143
144
Table D.2:Summary Statistics for Structural Variables
1979-1990 1990- 1996 1998-2004
Average Annual Average Annual Average Annual
(percent) change (percent) change (percent) change
Trade
Materials Offihoring
Original measure (Br) 14.98 0.52 15.73 -0,32
Original measure (Nr) 7.64 0.29 8,35 -0.19
Original measure (Br- Nr) 7,34 0.23 7,38 -0.13
Refilled measure (Br) 14.56 0.46 17.54 -0.06
Refilled measure (Nr) 7.68 0.23 9.71 -0.02
Refilled measure (Br- Nr) 6.88 0.23 7.84 -0.04
Services Offihoring
Selected Business Services 0.42 0.04 0.51 0.0003
Openness to Imports
Finished Goods hnportslVA 29.89 0.87 47.16 4.61
Technology
With Ex Post User Costs
Computer Share 4.75 0,32 7.17 0.16 12.20 0.48
Office :Equipment Share 0.83 -0.05 0.45 -0.03 0.10 -0.03
Other Hi-Tech Share 4.01 0.20 5.12 0.03 4.76 -0.1 I
With Ex Ante User Costs
Computer Share 2.87 0.23 5.14 0.19 9.67 0.48
Office :Equipment Share 0.48 -0.03 0,33 -0.01 0.08 -0.D2
Other Hi-Tech Share 3.01 0.18 4,32 0.07 4.23 -0.08
Note: Both averages and changes are weighted by the industry share of total manufacturing shipments. All variables
are com puted over 453 four-digit SIC industries in 1989- I996 and 473 six-digit industries in 1997-2004. T he data
come from the BLS input-output tables and Ray Roshita of BLS.
Table D.3: Consistency of Data with Equation (111.4)
a) Descriptive Statistics: Mean Changes in Log Factor Prices
145
1989-1996
Annualized Diff. First Diff.
1997-2004
Annualized Diff. First Diff.
Production labor
Nonproduction labor
Capital
2.666
3.839
2.900
2.668
3.784
2.771
3.025
2.769
0.418
3.022
2.758
0.274
b) Regression of .1 p~A +.1 ETFP j on primary factor cost shares
1989-1996 1997-2004
Annualized Diff First Diff. Annualized Diff. First Diff.
Prod. Cost Share 2.631*** 2.667*** 3.010*** 3.032***
[0.022] [0.013] [0.015] [0.012]
Non-Prod. Cost Share 3.689*** 3.644*** 2.777*** 2.744***
[0.172] [0.161] [0.040] [0.025]
Capital Cost Share 2.941*** 2.798*** 0.422*** 0.275* **
[0.030] [0.029] [0.008] [0.005]
Obs ervations 458 3206 473 3311
R-squared 1.00 0.99 1.00 0.99
Note: Standard errors are in parenthesis. All regressions are weighted by the industry share of total
manufacturing shipments.
Table D.3: Consistency of Data with Equation (4) (Cont.)
c) Regression of L1 P ~A and L1 ETFP i on primary factor cost shares
1989-19% 1997-2004
Annualized Diff. First Diff. Annualized Diff. First Diff.
L1 VA L1 VA VA L1 VAPi L1ETFPi Pi L1ETFPi L1 Pi L1 ETFP j Pi L1ETFPi
Prod. Cost Share 11.516 -8.885 10.317** -7.650* 4.986 -1.976 5.099* -2.067
[9.095] [9.110] [4.396] [4.397] [6256] [6.251] [2.956] [2.953]
Non-Prod. Cost Share -4.037 7.725 -0.970 4.614 -2.713 5.490 -6.781 * 9.525***
[8.659] [8.684] [3.715] [3.719] [4.834] [4.828] [3.675] [3.672]
Capital Cost Share -0.482 3.423 -0.628 3.426 -0204 0.626 0.601 -0.325
[2.757] [2.765] [2.087] [2.087] [2.529] [2.524] [1.395] [1.393]
Observations 458 458 3206 3206 473 473 3311 3311
R-squared 0.06 0.09 0.03 0.03 0.02 0.03 0.01 0.01
Note: Standard errors are in parenthesis. All regressions are weighted by the industry share of total manufacturing shipments.
........
+:.
0"1
Table D.4. Stage I - Original Feenstra and Hanson (1999) Specification
1989-1996 1997-2004
Annualized Difference First Difference Annualized Difference First Difference
Original Refmed Original Refmed Original Refmed Original Refined
Measure Measure Measure Measure Measure Measure Measure Measure
II III IV V VI VII VIII
Trade
Materials Offsh. (Nr) 0.067 -0.021 0.016 0.000 0.136 0.135 0.002 -0.001
[0.083] [0.077] [0.012] [0.017] [0.151] [0.184] [0.003] [0.001]
Materials Offsh. (Br-Nr) 0.440** 0.533* 0.081* 0.061* 0.133 0.296 0.001 0.010*
[0.208] [0.263] [0.046] [0.030] [0.348] [0.218] [0.007] [0.005]
Technology
Office Equip. Share -3.820* -4.835** -3.337** -3.476** -2.903** -2.975*** -1.749*** -1.754***
[2.095] [2.277] [1.530] [1.595] [1.067] [0.799] [0.569] [0.557]
Other Hi-Tech Share -0.322 -0.415 -0.251 -0.262 0.440** 0.560*** 0.332* 0.334*
[0.386] [0.423] [0.325] [0.333] [0.186] [0.193] [0.163] [0.164]
Other Controls
Year Fixed Effects - - Yes Yes - - Yes Yes
Constant 1.I67*** 1.I50*** 1.038*** 1.020** * 0.554*** 0.542*** 0.476*** 0.476***
[0.113] [0.137] [0.151] [0.161] [0.072] [0.045] [0.069] [0.069]
Observations 458 458 3206 3206 473 473 3311 3311
R2 0.18 0.17 0.08 0.Q7 0.14 0.18 0.08 0.08
Note: Standard errors in brackets are robust to heterosckedasticity and correlation in the errors within two-digit SIC industries for 1989-1996 and three-digit NAICS industries
for 1997-2004. Variables expressed as annualized differences are constructed as differences over end-years of each period, divided by the number of years in the period.
Variables expressed as first-difference are constructed as differences over year I and 1-1. Regressions are weighted by an average industry share of the manufacturing shipments.
.....
.j:>.
-...l
Table D 5. Stage I - Full Specification
1989-1996 1997-2004
Annualized Difference First Difference Annualized Difference First Difference
Original Refilled Original Refilled Original Refilled Original Refilled
Measure Measure Measure Measure Measure Measure Measure Measure
II III IV V VI VII VIII
Trade
Materials Offsh. (Nr) -0.030 -0.030 0.010 0.001 0.153 0.361 ** -0.004 -0.004***
[0.066] [0.044] [0.010] [0.015] [0.174] [0.137] [0.005] [0.001]
Materials Offsh. (Br-Nr) 0.260 0.510** 0.066 0.057* 0.071 0.103 -0.007 0.006
[0.171] [0.186] [0.039] [0.031] [0.339] [0.162] [0.008] [0.006]
Services Offsh. -17.100* * -17.263* * -0.993*** -1.039* ** -0.115 5.932** 0.755 0.686
[6.834] [6.136] [0.276] [0.281] [2.493] [2.341] [0.676] [0.639]
Import Openness 0.001 -0.001 0.000 0.000 -0.001 -0.001 0.000 0.000
[0.004] [0.004] [0.001] [0.001] [0.001] [0.002] [0.001] [0.001]
Technology
Office Equip. Share 0.004 -0.129 -1.848 -1.914 -2.438* * -2.300* ** -1.710*** -1.714***
[2.798] [2.827] [1.519] [1.527] [0.903] [0.701] [0.545] [0.534]
Other Hi-Tech Share -0.358 -0.424 -0.280 -0.285 0.372* 0.491 *** 0.326* 0.326*
[0.310] [0.283] [0.289] [0.290] [0.191] [0.159] [0.171] [0.171]
Computer Share 0.567* 0.603** 0.323* ** 0.330*** 0.085 0.077 0.013 0.015
[0.281] [0.242] [0.103] [0.101] [0.088] [0.074] [0.028] [0.027]
Other Controls
Market Power Yes Yes Yes Yes Yes Yes Yes Yes
Year Fixed Effects - - Yes Yes - - Yes Yes
Observations 458 458 3206 3206 473 473 3311 3311
R2 0.33 0.36 0.16 0.16 0.24 0.30 0.09 0.09
Note: Standard errors in brackets are robust to heterosckedasticity and correlation in the errors within two-digit SIC industries for 1989-1996 and three-digit NAICS industries
for 1997-2004. Variables expressed as annualized differences are constructed as differences over end-years of each period, divided by the number of years in the period
Variables expressed as first-difference are constructed as differences over year (and (-1. Regressions are weighted by an average industry share of the manufacturing shipments.
.......
+:>.
00
Table D.6: Stage I - Decomposed Dependent Variable (Refined Measure)
1989-1996 1997-2004
Annualized Difference First Difference Annualized Difference First Difference
L1 VA L1ETFPi L1 VA L1ETFPi L1 VA L1ETFPi L1 VA L1ETFPiPi Pi Pi Pi
I II III IV V VI VII VII
Trade
Materials Offsh. (Nr) -1.603*** 1.573* ** -0.008 0.009 0.093 0.268 -0.175 0.171
[0.404] [0.424] [0.153] [0.147] [1.204] [1.250] [0.113] [0.114]
Materials Offsh. (Br-Nr) 0.070 0.440 -0.698* 0.755* 0.403 -0.300 -0.356 0.362
[0.846] [0.789] [0.390] [0.395] [1.032] [1.078] [0.339] [0.344]
Services Offsh. -8.693 -8.570 -3.014 1.975 155.304*** -149.372*** 81.825*** -81.139***
[24.890] [23.649] [7.457] [7.396] [23.045] [22.050] [10.553] [10.104]
Import Openness 0.Ql5 -0.015 -0.019 0.019 -0.012** 0.011** -0.003 0.003
[0.031] [0.031] [0.012] [0.012] [0,(106] [0.004] [0.005] [0.005]
Technology
Office Equip. Share 3.183 -3.312 2.686 -4.600 1.943 -4.243 -3.457 1.742
[10.265] [9.614] [7.123] [6.231] [3.401] [3.484] [3.004] [3.222]
Other Hi-Tech Share -2.301 * 1.877* 0.912 -1.197 -1.225 1.716 -1.178 1.505
[1.133] [0.912] [1.667] [1.426] [1.632] [1.625] [1.219] [1.264]
Computer Share 0.112 0.491 -0.324 0.654 -0.099 0.176 0.349 -0.334
[0.615] [0.519] [0.561] [0.491] [0.342] [0.369] [0.245] [0.249]
Other Controls
Market Power Yes Yes Yes Yes Yes Yes Yes Yes
Year Fixed Effects - - Yes Yes - - Yes Yes
Obs ervations 458 458 3206 3206 473 473 3311 3311
R2 0.72 0.73 0.22 0.23 0.43 0.42 0.15 0.14
Note: Standard errors in brackets are robust to heterosckedasticity and correlation in the errors within two-digit SIC industries for 1989-1996 and three-digit NAICS
industries for 1997-2004. Variables expressed as annualized differences are constructed as differences over end-years of each period, divided by the number of years in the
period. Variables expressed as first-difference are constructed as differences over year I and I-I. Regressions are weighted by an average industry share of the manufacturing
shipments.
......
.j::.
\0
Table D.7. Stage I - Final Specification (Refined Measure)
1989-1996 1989-1996
Annualized First Annualized First
Difference Difference Difference Difference
II III IV
Trade
Materials Offsh. (Nr)1 0.390** -0.003* **
[0.178] [0.001]
Materials Offsh. (Br-Nr)2 0.433** 0.056*
[0.164] [0.028]
Services Outs. -16.158** -1.086* ** 6.120* *
[6.933] [0.344] [2.751]
Technology
Office Equip. Share -2.394* ** -1.722***
[0.650] [0.549]
Other Hi-Tech Share 0.484* ** 0.327*
[0.154] [0.166]
Computer Share 0.668*** 0.373* **
[0.166] [0.067]
Other Controls
Market Power Yes Yes Yes Yes
Year Fixed Effects Yes Yes
Constant 1.774*** 1.209*** 0.526** * 0.476***
[0.222] [0.090] [0.040] [0.068]
Observations 458 3206 473 3311
R2 0.33 0.12 0.28 0.09
Note: ,., Materials Offshoring measures are constructed using the refined formula. Standard errors in
brackets are robust to heterosckedasticity and correlation in the errors within two-digit SIC industries
for 1989-1996 and three-digit NAICS industries for 1997-2004. Variables expressed as annualized
differences are constructed as differences over end-years of each period, divided by the number of
years in the period. Variables expressed as first-difference are constructed as differences over year I
and I-I. Regressions are weighted by an average industry share of the manufacturing shipments.
150
151
Table D.8: Stage II - (Refined Measure)
Dependent Variable:,1ln pVA +,1ln ETFP explained by causal variables
I I
1989-1996 1997-2004
Materials Services Computer Materials Service Office OtherOffsh. Equip. Hi-Tech
(Br-Nr) Offsh. Share Offsh. (Nr) Offsh. Share Share
Annualized Difference
Prod. Cost Share 0.305* * -2.413*** -0.237 -0.119 0.182 0.243* * 0.270***
[0.122] [0.288] [0.326] [0.161] [0.125] [0.110] [0.073]
Non-Prod. Cost Share 0.898*** 0.546 1.488** 0.066 -0.159 0.234*** -0.040
[0.224] [0.499] [0.656] [0.242] [0.169] [0.090] [0.095]
Capital Cost Share 0.013 -1.131*** 0.137 -0.028 -0.002 0.122*** -0.221 ***
[0.044] [0.146] [0.132] [0.052] [0.035] [0.022] [0.030]
Observations 458 458 458 473 473 473 473
R2 0.59 0.86 0.36 0.04 0.03 0.63 0.61
Net Coefficientl 0.593*** 2.959*** 1.725*** 0.185 -0.341 ** -0.009 -0.310***
[0.180) [00407) [0.518) [0.206) [0.149) [0.100) [0.085)
First Difference
Prod. Cost Share 0.031 -0.172*** -0.112 0.001 0.198*** 0.160* **
[0.025] [0.034] [0.105] [0.008] [0.041 ] [0.023]
Non-Prod. Cost Share 0.113* * -0.004 0.946*** -0.005 0.174*** -0.061 *
[0.058] [0.052] [0.202] [0.014] [0.038] [0.032]
Capital Cost Share 0.005 -0.062*** 0.041 0.001 0.079* ** -0.139* **
[0.007] [0.015] [0.034] [0.002] [0.010] [0.009]
Observations 3206 3206 3206 3311 3311 3311
R2 0.11 0.23 0.20 0.00 0.44 0.50
Net Coefficientl 0.082* 0.168*** 1.058*** -0.006 -0.024 -0.221 ***
[0.045) [0.044] [0.161) [0.011) [0.040) [0.028)
Note: 'Net coefficient refers to the difference between the coefficients on non-production and production cost
shares. Coefficient estimates used to construct the dependent variable for 1989-1996 and 1997-2004 are those
from respective columns of Table 6. Standard errors are in brackets and are adjusted using Dumont et al. (2005)
method described in the text. All regressions are weighted by an average industry share of total manufacturing
shipments.
APPENDIXE
TABLES FOR CHAPTER IV
Table E.!: Summary Statistics of Explanatory Variables
Mean S.D. Min Max
Contractual Dependence 0.72 0.26 0.02 1.00
Skill Dependence 0.10 0.05 0.02 0.28
Capital Dependence 0.66 0.10 0.41 0.95
GDP (in $bill.) 176.18 514.38 0.26 4607.76
Labor Supply (in $mill.) 20.38 77.42 0.14 716.91
Contracts Quality 0.53 0.21 O.ll 0.97
Ln Human Capital/Worker 0.58 0.29 0.07 1.21
Ln Physical Capital/Worker 9.22 1.60 5.76 11.59
152
153
Table E.2: Correlation Matrix of Interaction Terms
I II ill N V VI VII
I (Contr. Dep.) * (Contr. Quality) 1.00
II (Contr. Dep.) * (In GDP) 0.72 1.00
ill (Contr. Dep.) * (Contr. Quality)*(1n GDP) 0.83 0.92 1.00
N (Contr. Dep.) * (In Lab. Supply) 0.21 0.70 0.48 1.00
V (Contr. Dep.) * (Contr. Quality)*(1n Lab. Supp.) 0.45 0.83 0.75 0.89 1.00
VI (Skill Dep.) * (In Skill Endow.) 0.53 0.47 0.54 0.09 0.26 1.00
VII (Cap. Dep.) * (In Cap. Endow.) 0.22 0.29 0.38 -0.12 0.08 0.13 1.00
Table E.3: Sample Countries
154
North
AUSTRALIA
AUSTRIA
BELGIUM/LUX.
CANADA
DENMARK
FINLAND
FRANCE
GERMANY
HONG KONG
ICELAND
IRELAND
ISRAEL
ITALY
JAPAN
NETHERLANDS
NORWAY
SINGAPORE
SWEDEN
SWITZERLAND
U.K.
South
ALGERIA
ANGOLA
ARGENTINA
BENIN
BANGLADESH
BOLIVIA
BRAZIL
BURKINA FASO
BURUNDI
CAMEROON
CHAD
CHILE
CHINA
COLOMBIA
CONGO
COST ARICA
CYPRUS
C. AFRICAN REP.
ECUADOR
EGYPT
FIJI
GABON
GAMBIA
GHANA
GREECE
GUATMALA
GUINEA
GUYANA
GUINEA BISSAU
HAITI
HONDURAS
HUNGARY
INDIA
INDONESIA
IVORY COAST
JAMAICA
JORDON
KENYA
AUSTRALIA
MADAGAS
MALAWI
MALAYSIA
MALI
MALTA
MAURITN
MEXICO
MOROCCO
MOZAMBQ
MRITIUS
NEW GUINEA
NEW ZEALAND
NICARAGA
NIGER
NIGERIA
OMAN
PAKISTAN
PANAMA
PARAGUA
PERU
PHILLIPINES
POLAND
PORTUGAL
ROMANIA
RWANDA
SALVADR
SAUDI ARABIA
SENEGAL
SIERRA LEONE
SPAIN
SRI LANKA
SUDAN
SURINAM
SYRIA
S. FRICA
TANZANIA
THAILAND
TOGO
TRINIDAD
TUNISIA
TURKEY
UGANDA
URUGUAY
VENEZUELA
YEMEN
YUGOSLAV
ZAIRE
ZAMBIA
ZIMBABWE
155
Table E.4: Contracts and Market Thickness
II III IV V
(Contr. Dep.) * (Contr. Quality) 0.161*** 0.117*** 0.163*** 0.042 0.182***
[0.026] [0.034] [0.026] [0.054] [0.037]
(Contr. Dep.) * (In GOP) 0.063** -0.021
[0.026] [0.063]
(Contr. Dep.) * (In Lab. Supp.) 0.060*** 0.096*
[0.019] [0.057]
(Contr. Dep.) * (Contr. Qual.) * (In GOP) 0.129
[0.087]
(Contr. Dep.) * (Contr. Qual.) * (In Lab. Supp.) -0.040
[0.061 ]
(Skill Dep.) * (In Skill Endow.) 0.110** * 0.109*** 0.108*** 0.109*** 0.108***
[0.023] [0.023] [0.023] [0.023] [0.023]
(Cap. Dep.) * (In Cap. Endow.) 0.176*** 0.189*** 0.172*** 0.185*** 0.172***
[0.041 ] [0.042] [0.041] [0.041] [0.042]
Constant -0.252*** 1.217*** 1.098*** -0.450*** 0.671 ***
[0.070] [0.087] [0.095] [0.126] [0.149]
Country & Industry Fixed Effects Yes Yes Yes Yes Yes
Observations 29484 29484 29484 29484 29484
Number of Industries 273 273 273 273 273
Number ofCountries 108 108 108 108 108
R-squared 0.69 0.69 0.69 0.69 0.69
Note: T he regressions are estimates of (IV I 6) and (IV I 7). The dependent variable is the natural log ofD.S.
imports of intermediate inputs sourced from industry i of country c. Standardized beta coefficients are reported,
with robust standard errors clustered around countries in brackets. *, **, *** indicate significance at the 10%,5%,
and I % levels.
Table E.5: Sensitivity Results
II III IV V VI VII
(Contr. Dep.) * (Contr. Quality) 0.168*** 0.366*** 0.118*** 0.178* ** 0.160*** 0.148*** 0.201***
[0.042] [0.052] [0.036] [0.035] [0.026] [0.023] [0.027]
(Contr. Dep.) * (In Lab. Supply) 0.062*** 0.222*** 0.083*** 0.083*** 0.056*** 0.043** 0.081***
[0.019] [0.031] [0.025] [0.023] [0.020] [0.019] [0.020]
(Skill Dep.) * (In Skill Endow.) -0.044 0.320*** 0.003 0.195*** 0.110*** 0.110*** 0.101***
[0.029] [0.057] [0.026] [0.038] [0.024] [0.024] [0.023]
(Cap. Dep.) * (In Cap. Endow.) 0.102* 0.206* 0.122** 0.240*** 0.176*** 0.176*** 0.170***
[0.057] [0.108] [0.053] [0.074] [0.042] [0.043] [0.042]
(Contr. Dep.) * (In Skill Endow.) 0.010
[0.035]
(Contr. Dep.) * (In Cap. Endow.) -0.070
[0.051 ]
(Skill Dep.) * (Contr. Quality) 0.240***
[0.040]
(Cap. Dep.) * (Contr. Quality) 0.119**
[0.058]
Constant -0.483*** -2.362*** 1.337** * -1.420*** 0.501*** 1.008*** 0.787***
[0.131] [0.493] [0.144] [0.221] [0.100] [0.105] [0.108]
Country & Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes
Restrictions None Non-Zero South No Africa No China No S.E. Asia Outlier Ind.
Observations 29484 11984 24024 20748 29211 27846 27324
Number ofIndustries 273 273 273 273 273 273 253
Number ofCountries 108 108 88 76 107 102 108
R-squared 0.69 0.65 0.59 0.68 0.68 0.70 0.69
Note: T he regressions are estimates of (IV.16). T he dependent variable is the natural log of U.S. imports of intermediate inputs sourced from
industry i of country c. Standardized beta coefficients are reported, with robust standard errors clustered around countries in brackets. *, **, ***
indicate significance at the 10%, 5%, and I% levels.
,......
VI
0\
157
Table E.6: Comparison of Import Patterns
Dependent Variable
Intenrediates Non- Total TotalIntenrediates
II III IV
(Contr. Dep.) * (Contr. Quality) 0.163*** 0.206*** 0.161*** 0.256***
[0.026] [0.017] [0.018] [0.025]
(Contr. Dep.) * (in Lab. Supp.) 0.060*** 0.02 -0.005 0.044**
[0.019] [0.014] [0.013] [0.020]
(Skill Dep.) * (in Skill Endow.) 0.108*** 0.094*** 0.096*** 0.093***
[0.023] [0.024] [0.023] [0.023]
(Cap. Dep.) * (in Cap. Endow.) 0.172*** 0.204*** 0.187*** 0.260***
[0.041] [0.044] [0.041] [0.044]
Constant 1.098*** 1.835*** 1.056*** 0.600***
[0.095] [0.098] [0.086] [0.115]
Country & Industry Fixed Effects Yes Yes Yes Yes
Restrictions No No No No
Observations 29484 32724 34020 34020
Number ofIndustries 273 303 315 315
Number ofCountries 108 108 108 108
R-squared 0.69 0.70 0.69 0.69
Note: The regressions are estiImtes of(16). The dependent variables in Columns I and II are
U.S. imports ofintenrediate goods and non-intenrediate goods, respectively. Columns III
and IVare total U.S. imports. The Contract Dependence variable is constructed differently in
each column, as described in the text. All dependent variable are expressed as the natural log
ofimports sourced from industry i ofcountry c. Standardized beta coefficients are reported,
with robust standard errors clustered around countries in brackets. *, **, *** indicate
significance at the 10%,5%, and 1% levels.
~Table E.7: Comparison of Import Patterns By Utilization Rates
All 25%::; 50%::; 75%::;
--
II III N V VI VII VIII
(Contr. Dep.) * (Contr. Quality) 0.I02*** 0.237*** 0.142*** 0.I03*** 0.152*** 0.134*** 0.167*** 0.182***
[0.019] [0.029] [0.027] [0.032] [0.026] [0.028] [0.028] [0.03 I]
X Materials Dunnny -0.024* ** -0.029*** 0.048*** 0.052** 0.059*** 0.071*** 0.095*** 0.113***
[0.009] [O.OIO] [0.017] [0.020] [0.020] [0.024] [0.024] [0.029]
(Contr. Dep.) * (In Lab. Supply) 0.008 0.063** 0.043* 0.048 0.055*** 0.088*** 0.084*** 0.145* **
[0.022] [0.025] [0.022] [0.032] [0.020] [0.029] [0.020] [0.032]
X Materials Dunnny -0.022** -0.030** -0.039** -0.050** -0.042*** -0.054** -0.050** -0.060**
[O.OIO] [0.012] [0.015] [0.020] [0.016] [0.023] [0.020] [0.028]
(Skill Dep.) * (In Skill Endow.) 0.050*** 0.104*** 0.129*** 0.153*** 0.143*** 0.180*** 0.150*** 0.202***
[0.018] [0.025] [0.025] [0.026] [0.025] [0.027] [0.026] [0.029]
X Materials Dunnny -0.014* -0.029*** -0.072*** -0.098*** -0.138*** -0.179* ** -0.158* ** -0.198***
[0.008] [0.009] [0.015] [0.018] [0.021] [0.025] [0.024] [0.029]
(Cap. Dep.) * (In Cap. Endow.) 0.051*** 0.220*** 0.114*** 0.064 0.I07*** 0.095* 0.119*** 0.120**
[0.019] [0.044] [0.043] [0.051] [0.040] [0.048] [0.041] [0.049]
X Materials Dunnny -0.021 *** -0.027*** -0.119*** -0.151*** -0.112** * -0.151*** -0.157*** -0.198***
[0.006] [0.008] [0.017] [0.021] [0.018] [0.024] [0.023] [0.029]
Constant -0.597*** -0.233*** -0.740*** -0.158* -0.618*** -0.394*** -1.115*** 0.017
[0.064] [0.083] [0.121] [0.092] [0.115] [0.120] [0.076] [0.088]
Country & Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
Restricted No Yes No Yes No Yes No Yes
Observations 62208 56376 53292 38915 49541 31672 46568 29607
R-squared 0.65 0.69 0.59 0.58 0.53 0.48 0.53 0.48
Note: T he regressions are estimates of (IY.16). T he dependent variables are drawn from pooled samples of U.S. imports of intermediates and non-intermediates,
expressed as the natural log of imports sourced from industry i of country c. The samples in Columns I and II include all imports, while the samples in Columns III-IV,
V-VI, and VII-VIII contain only those imports that have at least 25%, 50%, 75% utilization rate as intermediate or non-intermediate goods. The samples in Columns
II, IV, VI, and VII are restricted to include only those industries which source both intermediate and non-intermediate goods. Standardized beta coefficients are
reported, with robust standard errors clustered around countries in brackets. '. ", ••• indicate significance at the 10%,5%, and I % levels.
.......
Vl
00
159
BIBLIOGRAPHY
Amiti, Mary, and Shang-Jin Wei, "Service Offshoring and Productivity: Evidence from
the United States," NBER Working Paper No. 11926,2006.
Antweiler, Wemer, and Daniel Trefler, "Increasing Retums and All That: A View from
Trade," American Economic Review, 92 (2002), 93-119.
Autor, David H., Frank Levy, and Richard 1. Mumane, "The Skill Content of Recent
Technological Change: An Empirical Investigation," Quarterly Journal of
Economics, 118 (2003), 1279-1333.
Autor, David H., Lawrence F. Katz and Melissa S. Keamey, "Trends in U.S. Wage
Inequality: Revising the Revisionists," Review o.fEconomics and Statistics,
forthcoming, 2008.
Bmielsman, Eric J., and Wayne Gray, "The NBER Manufacturing Productivity
Database," NBER Technical Working Paper No. 205, 1996.
Berkowitz, Daniel, Johannes Moenius and Katharina Pistor, "Trade, Law and Product
Complexity," Review o.fEconomics and Statistics, 87 (2006), 363-373
Berman, Eli, John Bound, and Zvi Gliliches, "Changes in the Demand for Skilled Labor
within U.S. Manufacturing: Evidence from the Annual Survey of Manufacturers,"
Quarterly Journal o.f Economics, 109 (1994), 367-398.
Bemdt, Emst R., and Catherine J. Morrison, "High-Tech Capital Formation and Labor
Composition in U.S. Manufacturing Industries: An Exploratory Analysis,"
Journal ofEconometrics, 65 (1995), 9-43.
Bowen, Harry P., Edward E. Leamer, and Leo Sveikauskas, "Multicountry, Multifactor
Tests of the Factor Abundance Theory," American Economic Review, 77 (1987),
791-809.
Campa, Jose, and Linda S. Goldberg, "The Evolving Extemal Orientation of
Manufacturing Industries: Evidence from Four Countries," Economic Policy
Review, The Federal Reserve Bank of New York, 3 (1997), 53-81.
160
Card, David and John E. DiNardo, "Skill-Biased Technological Change and Rising Wage
Inequality: Some Problems and Puzzles," Journal o.lLabor Economics, 20
(2002), 733-83.
Chen, Hogan, Matthew Kondratowicz, and Kei-Mu Vi, "Vertical Specialization and Three
Facts about U.S. International Trade," North American Journal o/Economics and
Finance, 16 (2005), 35--59
Deardorff, Alan v., "Overview of Stolper-Samuelson Theorem," in The Stolper
Samuelson Theorem: A Golden Jubilee, Alan V. Deadorff and Robert M. Stem,
eds. (Ann Arbor: University of Michigan Press, 1994), pp. 7-34.
Domowitz, Ian, Hubbard, R. Glenn, and Petersen, Bruce C. "Market Structure and
Cyclical Fluctuations in U.S. Manufacturing," Revievv o.lEconomics and
Statistics, 70 (1988),55-66.
Diaz-Mora, Cannen, and Angela Triguero (2007). "Why Do Some Finns Contract-out
Production?" Studies on the Spanish Economy, 232, FEDEA, 2007.
Dumont, Michel, Glenn Rayp, Olivir Thas, and Peter Willeme, "Correcting Standard
Enors in Two-Stage Estimation Procedures with Generated Regressands," Oxford
Bulletin o.fEconomics and Statistics, 67 (2005), 305-9049.
Egger, Harimut, and Peter Egger, "The Detenninants of EU Processing Trade," The
World Economy, 27 (2005),147-168.
Feenstra, Robert C., and Gordon H. Hanson, "Foreign Investment, Outsourcing and
Relative Wages," in The Political Economy o.lTrade Policy: Papers in Honor 0.1
Jagdish Bhagwati, Robert.C. Feenstra, Gene.M. Grossman, and Douglas A. Irwin,
eds.(Cambridge: MIT Press, 1996), pp. 89-127.
Feenstra, Robert c., "Integration of Trade and Disintegration of Production in the Global
Economy," Journal a.!Economic Perspectives, 12 (1998), 31-50.
Feenstra, Robert C., and Gordon H. Hanson, 'The Impact of Outsourcing and High-
Technology Capital on Wages: Estimates for the United States, 1979-1990',
Quarterly Journal a/Economics, 114 (1999), 907-940.
--, "Global Production Sharing and Rising Inequality: A Survey ofTrade and Wages,"
in Handbook o/International Trade, Kwan Choi and James Harrigan, eds.
(London: Basil Blackwell, 2003), pp. 146-185.
Feenstra, Robert c., and John Romalis, and Peter K. Schott, "U.S. Imports, Exports, and
Tariff Data, 1989-2001," NBER Working Paper No. 9387, 2002.
161
G6rg, Holger, Aoife Hanley, and Eric Strobl, "Productivity Effects of International
Outsourcing: Evidence from plant level data," Canadian Journal o./Economics,
41 (2007), 670-688.
Grossman, Gene M., and Elhanan Helpman, "Integration versus Outsourcing in Industry
Equilibrium", Quarterly Journal 0./Economics, 117 (2002a), 85-120.
Grossman, Gene M., and Elhanan Helpman,"Outsourcing in a Global Economy", Review
0./Economic Studies, 72 (2005), 13 5-159.
Gwartney, James, and Robeli Lawson, "Economic Freedom of the World: 2003
Annual Report," Mimeograph, Fraser Institute, 2003.
Hall, Robert E., and Dale W. Jorgenson, "Tax Policy and Investment Behavior,"
American Economic Review, 57 (1967),391-414.
Harper, Michael J., Ernst R. Berndt, and David O. Wood, "Rates of Retunr and Capital
Aggregation Using Alternative Rental Prices," in Technojogy and Capital
Formation, Dale W. Jorgenson and Ralph Landau, eds. (Cambridge: MIT Press,
1989), pp. 331-372.
Haskel, Jonathan E., "The Trade and Labour Approaches to Wage Inequality," University
of London Working Paper, 1999.
Haskel, Jonathan E. and Matthew J. Slaughter, "Trade Technology and U.K. Wage
Inequality," Economic Journal, 110 (2001), 1~27.
Haskel , Jonathan E. and Matthew J. Slaughter, "Have Falling Tariffs and TranspOliation
Costs Raised U.S. Wage Inequality," Review o/International Economics, 11
(2003), 630-650.
Hall, Robert E., and Charles 1. Jones, "Why do Some Countries Produce So Much More
Output per Woerk than Others?" Quarterly Journal of Economics, 114 (1999),
83-116.
Holl, Adelheid, "Production Subcontracting and Location: Panel Data Evidence from
Spanish Manufacturing Finns," TRP Working Papers 146, University of Sheffield,
2007.
Hummels, David, Jun Ishii, and Kei-Mu Yi, "The Nature and Growth of Vertical
Specialization in World Trade," Journal o/International Economics, 54 (2001),
75-96.
162
Hummels, David, Dana. Rapoport and Kei-Mu Yi, "Vertical Specialization and the
Changing Nature of World Trade", Federal Reserve Bank of New York Economic
Policy Review, IV (2003),79-99.
Jones, Ronald, Henryk Kierzkowski, and Chen Lurong, "What Does Evidence Tell Us
about Fragmentation and Outsourcing?" International Review ofEconomics and
Finance, 14 (2005), 305-316.
Kaminski, Bmilomiej T, and Francis Ng, "Production Disintegration and Integration of
Central Europe into Global Markets," International Review ofEcono/nics and
Finance, 14 (2005), 377-390.
Katz, Lawrence F. and David Autor, "Changes in the Wage Structure and Earnings
Inequality," in Handbook o.lLabor Economics llIA, Orley Ashenfelter and David
Card, eds. (Amsterdam: North-Holland, 1999), pp. 1463-1555.
Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi, "Governance Matters
Ill: Governance Indicators for 1996-2002," Working Paper No. 3106, World
Bank,2004.
Kimura, Fukunari, "Fragmentation, Internalization, and Inter-Finn Linkages: Evidence
from the Micro Data of Japanese Manufacturing Fimls," in Global Production
and Trade in East Asia, L. K. Cheng and H. Kierzkowski, eds. (Boston: Kluwer
Academic Publishers, 2001), pp. 129-152.
Kimura, Fukunari, Yuya Takahashi, and Kazunobu Hayakawa, "Fragmentation and Parts
and Components Trade: Comparison between East Asia and Europe," North
American Journal o.fEconomics and Finance, 17 (2007), 23-40.
Krugman, Paul, "Growing World Trade: Causes and Consequences," Brooking Paper on
Economic Activity, 1 (1995), 327-362.
Krugman, Paul, "Technology, Trade and Factor Prices," Journal o.lInternational
Economics, 50 (2000), 51-72.
Leamer, Edward E., "In Search of Stolper-Samuelson Effects on U.S. Wages," in
Imports, Exports, and the American Worker, Susan Colins, ed. (Washington D.C.:
The Brookings Institution, 1998), pp. 141-202.
Lemiuex, Thomas, "The Changing Nature of Wage Inequality," NBER Working Paper
No. 13523,2007.
Levchenko, Andrei, "Institutional Quality and Intemational Trade" Revievl' o.fEconomic
Studies, 74 (2007), 791-819.
163
Moulton, Brendt R., "Random Group Effects and the Precision of Regression Estimates,"
Journal ofEconometrics, 32 (1986), 385-397.
Nunn, Nathan, "Relationship-Specificity, Incomplete Contracts and the Pattern of Trade",
Quarterly Journal ofEconomics, 122 (2007), 569-600.
Rauch, James E., "Networks versus Markets in International Trade," Journal of
International Economics, 48 (1999),7-35.
Robertson, Raymond. "Relative Prices and Wage Inequality: Evidence from Mexico,"
Journal ofInternational Economics 62 (2004), 387-409.
Romalis, John, "Factor Proportions and the Structure of Commodity Trade", American
Economic Review, 94 (2004), 67-97.
Ranjan, Priya, and Jae Young Lee, "Contract Enforcement and the Volume of
International Trade in Different Types of Goods" University of California, Irvine,
mimeo, 2005.
Rybczynski, T. M. "Factor Endowment and Factor Mobility," American Economic
Review, XLVII (1957), 321-325.
Schott, Peter K., "Across-Product versus Within-Product Specialization in International
Trade", Quarterly Journal ofEconomics, CXIX (2004), 647-678.
Slaughter, Matthew .1., "Globalisation and Wages: A Tale of Two Perspectives," The
World Economy, 22 (1999), 609-629.
Slaughter, Matthew .1., "What Are the Results of Product-Price Studies and What Can I
Learn from Their Differences?" in The Effects of International Trade on Wages,
Robert C. Feenstra, ed. (Chicago: University of Chicago Press, 2000), pp.
129-165.
Slaughter, Matthew l, "International Trade and Labor-Demand Elasticities," Journal of
International Economics, 54 (2001), 27-56.
Swenson, Deborah. L. "Entry Costs and Outsourcing Decisions: Evidence from the U.S.
Overseas Assembly Provision," The North American Journal ofEconomics and
Finance 15(2004), 267-286.
Trefler, Daniel, "International Factor Price Differences: LeontiefWas Right!" Journal of
Political Economy, 101 (1993), 961-987.
164
Tomiura, Eiichi,"Foreign Outsourcing and Finn-Level Characteristics: Evidence from
Japanese Manufacturers," Journal ofthe Japanese and International
Economies, 19 (2005), 255-271.
Yeats, Alexander 1, "Just How Big is Global Production Sharing?" in Fragmentation:
New Production Patterns in the World Economy, Sven W. Arndt, and Henryk.
Kierzkowski, eds. (Oxford: Oxford University Press, 2001), pp. 108-143.
Yi, Kei-Mu, "Can vertical specialization explain the growth of world trade?" Journal of
Political Economy, 109 (2003), 52-102.
U.S. Department of Commerce. Economics and Statistics Administration. Bureau of
Economic Analysis. 1990. Balance o.fPayments o/the United States. Washington
D.C.: Government Printing Office. http://www.bea.gov/scb/pdf/internat/bpa/meth/
bopmp.pdf.
U.S. Department of Commerce. Economics and Statistics Administration. Bureau of
Economic Analysis. 2006. Method and Concepts o.f the u.s. Input-Output
Accounts, by Karen 1 Horowitz and Mark A. Planting. Washington D.C.:
Government Printing Office. https://www.bea.gov/papers/pdf/
Iomanual_092906.pdf.
USITC (United States International Trade Commission). 1997. Production Sharing: Use
o.fUS. Components and Materials in Foreign Assembly Operations, 1992-1995,
USITC Publication 3032, Washington, D.C.
World Bank, Doing Business in 2004: Understanding Regulation (Washington, D.C.:
World Bank and Oxford University Press, 2004).