Essays on International Trade, Growth, and Development
A DISSERTATION
SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL
OF THE UNIVERSITY OF MINNESOTA
BY
Sang Min Lee
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Timothy J. Kehoe, Advisor
Manuel Amador, Advisor
July, 2023
© Sang Min Lee 2023
ALL RIGHTS RESERVED
Acknowledgements
I would like to express my profound gratitude and appreciation to all those who have
contributed to the successful completion of this dissertation. I am especially indebted
to my advisors, Manuel Amador, Doireann Fitzgerald, and Timothy J. Kehoe, for their
unwavering support and guidance, which played a pivotal role in shaping my growth as
an economist.
I am grateful to the faculty at the University of Minnesota and the economists at the
Federal Reserve Bank of Minneapolis. Interacting with them enriched my understanding
of economics and provided me with invaluable insights. Additionally, I extend my thanks
to the participants in the Minnesota International Macro and Trade Workshop, whose
involvement has been an integral part of my Ph.D. training.
I would also like to extend my heartfelt appreciation to my dear friends, Oh-Ran
Kwon and Dhananjay Ghei, whose support and help proved indispensable throughout
this entire process.
Lastly, I express my sincere gratitude for the financial support received from the
University of Minnesota Doctoral Dissertation Fellowship and the Morton D. and Artice
E. Silverman Fellowship, which enabled me to focus on my research and complete this
dissertation.
i
Dedication
To my beloved parents and sister, whose love, support, and encouragement have been
the guiding lights throughout my journey.
ii
Abstract
This dissertation consists of two chapters focusing on the role of international trade
and foreign direct investment in the growth and development of participating countries.
The first chapter studies the effect of globalization, i.e., a reduction in international
trade costs, on structural transformation, i.e., the reallocation of GDP from goods to
services. I study the topic with a focus on international trade in services, which was
largely overlooked in the previous literature, although it accounts for more than one-
third of total global trade. First, I construct a multi-country, multi-sector trade model
with non-homothetic preferences, where exogenous changes in sectoral productivities and
bilateral trade costs generate structural transformation. For the quantitative analysis,
I parameterize the model using data for 66 countries from 1995 to 2018. I find that
globalization decelerated structural transformation of the countries whose comparative
advantage in goods was strengthened by globalization and accelerated that of the coun-
tries with weakened comparative advantage in goods. For countries whose comparative
advantage wasn’t significantly affected, globalization had a negligible impact on their
structural transformation. I show that globalization shifts a country’s comparative ad-
vantage if trade costs faced by the country change disproportionately across goods and
services. Illustration of the mechanism is as follows: if goods export trade costs fall
faster than goods import trade costs for a country, its comparative advantage in goods
strengthens. If services export trade costs relative to services import trade costs decline,
the comparative advantage in goods weakens. Therefore, if a country’s export trade
cost relative to import trade cost for goods and services change at different rates, its
comparative advantage shifts.
The second chapter investigates the effect of inward foreign direct investment (FDI)
on the aggregate productivity growth of China’s manufacturing sector. Many attribute
the explosive growth of China’s manufacturing sector to Deng Xiaoping’s economic re-
form. As part of the reform, China opened itself to foreign investment in 1978. To
promote FDI, it gave foreign investors preferential treatments, including lower tax rates,
which lasted until 2008. As a result, the FDI inflow to China was the largest among the
developing countries from 1992 to 2019. This chapter contributes to the literature by
iii
providing a unified framework to study multiple channels through which FDI can con-
tribute to economic growth. First, I develop a firm-dynamics model with heterogeneous
productivities and FDI. In the model, a firm can improve its productivity through foreign
technology adoption, innovation, and spillovers (imitation). Unlike domestic firms, FDI
firms possess foreign technology adoption capabilities. Moreover, they participate in in-
novation with different rates from domestic firms. These features of the model generate
different productivity distributions for domestic and FDI firms. The model is disciplined
using the microevidence from Chinese firms and their patents from 1998 to 2007. By
calibrating the productivity distributions to the dataset, this study shows that the an-
nual growth rate of aggregate productivity would decrease from 8.42% to 7.50% without
the presence of FDI firms. Counterfactual exercises demonstrate that the growth con-
tribution mainly accrues through foreign technology adoption, which explains 0.72p.p.
of the total gain of 0.92p.p.
iv
Contents
Acknowledgements i
Dedication ii
Abstract iii
Contents v
List of Tables ix
List of Figures xi
1 Globalization and Structural Transformation: The Role of Tradable
Services 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.1 Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.2 Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.3 Market Clearing . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.4 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3 Model Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.1 Structural Transformation in a Closed Economy . . . . . . . . . . 21
1.3.2 Structural Transformation in an Open Economy . . . . . . . . . . 22
1.3.3 Input-output Linkages and Structural Transformation . . . . . . 23
v
1.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.1 Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.2 Three Sectors: Goods, Producer Services, and Consumer Services 25
1.4.3 Data Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.5 Parameter Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.5.1 Preference Parameters . . . . . . . . . . . . . . . . . . . . . . . . 31
1.5.2 Technology Parameters . . . . . . . . . . . . . . . . . . . . . . . 33
1.5.3 Other Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.6 Calibration of Productivities and Trade Costs . . . . . . . . . . . . . . . 35
1.6.1 Model Fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.6.2 Patterns of Changes in Trade Costs . . . . . . . . . . . . . . . . . 37
1.6.3 Patterns of Productivities . . . . . . . . . . . . . . . . . . . . . . 42
1.7 Counterfactuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.7.1 Globalization and Structural Transformation of China . . . . . . 43
1.7.2 Globalization and Structural Transformation of 66 Countries . . 46
1.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2 FDI and Aggregate Productivity Growth in Chinese Manufacturing
Firms 58
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.1.1 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.1.2 Outline of the Paper . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.2.1 Household . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.2.2 Composite Good Producer . . . . . . . . . . . . . . . . . . . . . . 63
2.2.3 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.4 Aggregate Productivity, Wage, and Firm Productivity . . . . . . 64
2.2.5 Productivity Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 65
2.2.6 Exit and Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.2.7 Value of a Firm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.2.8 Law of Motion of Productivity Distributions and Balanced Growth
Path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
vi
2.3 Data and Descriptive Evidence . . . . . . . . . . . . . . . . . . . . . . . 69
2.3.1 Hsieh and Klenow (2009) Productivity . . . . . . . . . . . . . . . 72
2.3.2 Foreign-domestic Difference 1: Foreign Technology Adoption . . . 72
2.3.3 Foreign-domestic Difference 2: Innovation . . . . . . . . . . . . . 74
2.3.4 Spillovers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.4 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.5 Counterfactuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
References 83
Appendix A. Appendix for Chapter 1 90
A.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
A.1.1 Details about OECD ICIO . . . . . . . . . . . . . . . . . . . . . 90
A.1.2 Construction of Price Sequences . . . . . . . . . . . . . . . . . . 92
A.2 Estimation and Calibration . . . . . . . . . . . . . . . . . . . . . . . . . 99
A.2.1 Model Fit: Non-targeted Moment Exercise . . . . . . . . . . . . . 99
A.2.2 Comparative Advantage . . . . . . . . . . . . . . . . . . . . . . . 100
A.2.3 Productivity Growth and Structural Transformation . . . . . . . 102
A.3 Counterfactuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
A.3.1 Solution Method for the Baseline and Counterfactuals . . . . . . 103
A.4 Robustness of Quantitative Results . . . . . . . . . . . . . . . . . . . . . 104
A.4.1 Different Definitions of Sectors . . . . . . . . . . . . . . . . . . . 104
A.4.2 Non-tradable Consumer Services . . . . . . . . . . . . . . . . . . 105
A.4.3 Trade Elasticities for Services . . . . . . . . . . . . . . . . . . . . 105
A.4.4 Wedges on Utility and Production Weights . . . . . . . . . . . . 105
A.4.5 Different Assumption on Production Function . . . . . . . . . . . 105
A.4.6 The Role of Exogenous Net Exports . . . . . . . . . . . . . . . . 106
A.5 Additional figures and tables . . . . . . . . . . . . . . . . . . . . . . . . 106
Appendix B. Appendix for Chapter 2 124
B.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
B.1.1 Household’s Optimization . . . . . . . . . . . . . . . . . . . . . . 124
vii
B.1.2 Composite Good Producer’s Optimization . . . . . . . . . . . . . 125
B.1.3 Firm’s Static Optimization and Profit Function . . . . . . . . . . 125
B.1.4 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
B.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
B.2.1 Cross-sections into a Panel . . . . . . . . . . . . . . . . . . . . . 127
B.2.2 Price Indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
B.2.3 Patent Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
B.3 Additional Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . 129
viii
List of Tables
1.1 Estimation Result for Preference Parameters . . . . . . . . . . . . . . . . 33
1.2 Estimation Result for Production Parameters . . . . . . . . . . . . . . . 34
1.3 Summary Statistics for the Impact of Globalization on Structural Trans-
formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
1.4 Result for the 10 Largest Countries in 2018 GDP . . . . . . . . . . . . . 55
2.1 Summary Statistics: Ownership and Innovation Activities (Extensive Mar-
gin) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.2 Data Analogues of Components of Aj(t) . . . . . . . . . . . . . . . . . . 72
2.3 Patenting Behavior (Extensive Margin) . . . . . . . . . . . . . . . . . . . 75
2.4 Comparison among the Calibrated Model and Four Counterfactuals: Im-
pact on the Annual Growth Rate . . . . . . . . . . . . . . . . . . . . . . 81
A.1 Summary Statistics for the Three Metrics for VAPD vs Imputed GOPD 96
A.2 Comparison between Imputed GOPD and VAPD for Select Country-sectors 97
A.3 Summary Statistics for the Three Metrics for VAPD (OECD STAN) vs
Imputed VAPD from UN SDMX and UN AMA . . . . . . . . . . . . . . 99
A.4 Summary Statistics for the Three Metrics for VAPD (OECD STAN) vs
Imputed VAPD from UN SDMX and UN AMA . . . . . . . . . . . . . . 107
A.5 Full Results for 66 Countries + ROW . . . . . . . . . . . . . . . . . . . 109
A.6 Availability of Price Deflators from OECD STAN, UN NA SDMX, UN
NA AMA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
B.1 Correlations between Firm Characteristics and Log-productivity Growth
(Domestic, Joint, and Fully-foreign Firms) . . . . . . . . . . . . . . . . . 130
B.2 Correlations between Firm Characteristics and Log-productivity Growth
(Domestic and Foreign Firms) . . . . . . . . . . . . . . . . . . . . . . . . 131
ix
B.3 Correlations between Firm Characteristics and Log-productivity Level
(Domestic, Joint, and Fully-foreign Firms) . . . . . . . . . . . . . . . . . 132
B.4 Correlations between Firm Characteristics and Log-productivity Level
(Domestic and Foreign Firms) . . . . . . . . . . . . . . . . . . . . . . . . 133
B.5 Patenting Behavior (Intensive Margin) . . . . . . . . . . . . . . . . . . . 134
B.6 R&D Behavior by Ownership Category . . . . . . . . . . . . . . . . . . . 135
x
List of Figures
1.1 Tradedness and Intermediateness of Service Industries in 2018 . . . . . . 26
1.2 Patterns of International Trade . . . . . . . . . . . . . . . . . . . . . . . 27
1.3 Share of Goods in Export and Import Volumes of Countries in 1995 and
2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.4 Sectoral Shares of Production in GDP and Consumption in Final Expen-
ditures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5 Model Fit: Prices and Trade Shares . . . . . . . . . . . . . . . . . . . . . 37
1.6 Box Plots for Trade Costs in 1995 and 2018 by Country Group Pairs . . 38
1.7 Median Trade Costs by Country Group Pairs from 1995 to 2018 . . . . . 39
1.8 Growth Rate of Relative Export Trade Costs and GDP per Capita . . . 40
1.9 China’s Trade Costs and Structural Transformation . . . . . . . . . . . . 44
1.10 Globalization in Producer Services and Structural Transformation . . . . 47
1.11 Income Levels and Impact of Producer Services Globalization . . . . . . 48
1.12 Globalization in Goods and Structural Transformation . . . . . . . . . . 50
1.13 Income Levels and the Impact of Goods Globalization . . . . . . . . . . 51
1.14 Globalization in Both Goods and Producer Services and Structural Trans-
formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.15 Income Levels and the Impact of Goods and Producer Services Globalization 54
2.1 Timeline of an (a, o, i)-type Firm . . . . . . . . . . . . . . . . . . . . . . 67
2.2 Histogram: Foreign Ownership Shares of Firm-years . . . . . . . . . . . 71
2.3 Log-productivity Growth and Level Gap . . . . . . . . . . . . . . . . . . 73
2.4 Productivity Distributions at t = 100 . . . . . . . . . . . . . . . . . . . . 78
2.5 Numerical Evidence of Convergence to the Balanced Growth Path . . . 79
A.1 Price Deflators for Slovakia’s Goods: GOPD vs VAPD vs Imputed GOPD 98
xi
A.2 Relationship between Sectoral Price Levels and Income . . . . . . . . . . 100
A.3 Model Fit: Non-targeted Moments . . . . . . . . . . . . . . . . . . . . . 101
A.4 Sectoral Net Exports of the US . . . . . . . . . . . . . . . . . . . . . . . 102
A.5 Productivity Growth and Relative Prices from 1995 to 2018 . . . . . . . 108
A.6 India’s Trade Costs and Structural Transformation . . . . . . . . . . . . 110
A.7 Vietnam’s Trade Costs and Structural Transformation . . . . . . . . . . 111
A.8 Lithuania’s Trade Costs and Structural Transformation . . . . . . . . . . 112
A.9 Tradedness and Intermediateness of Service Industries in 2018 . . . . . . 115
A.10 Globalization in Both Goods and Producer Services and Structural Trans-
formation (Goods, Highly-tradable Services, and Barely-tradable Services) 116
A.11 Globalization in Both Goods and Producer Services and Structural Trans-
formation (Goods and Services) . . . . . . . . . . . . . . . . . . . . . . . 117
A.12 Globalization in Both Goods and Producer Services and Structural Trans-
formation (Non-tradable Consumer Services) . . . . . . . . . . . . . . . 118
A.13 Globalization in Both Goods and Producer Services and Structural Trans-
formation (Low Trade Elasticities for Services) . . . . . . . . . . . . . . 119
A.14 Globalization in Both Goods and Producer Services and Structural Trans-
formation (High Trade Elasticities for Services) . . . . . . . . . . . . . . 120
A.15 Globalization in Both Goods and Producer Services and Structural Trans-
formation (Wedges for Preference and Production) . . . . . . . . . . . . 121
A.16 Globalization in Both Goods and Producer Services and Structural Trans-
formation (Nest-CES Production Function) . . . . . . . . . . . . . . . . 122
A.17 Globalization in Both Goods and Producer Services and Structural Trans-
formation (nx = 0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
B.1 Relative Frequencies of ASIE Firm-years’ Patent Activity Counts . . . . 136
xii
Chapter 1
Globalization and Structural
Transformation: The Role of
Tradable Services
International trade in services accounts for more than one-third of total global trade.
This paper studies how globalization, i.e., changes in trade costs for both goods and ser-
vices, affects the speed of structural transformation, i.e., the reallocation of GDP from
goods to services. To do so, I construct a multi-country, multi-sector trade model with
non-homothetic preferences, where exogenous changes in sectoral productivities and bi-
lateral trade costs generate structural transformation. For the quantitative analysis,
I parameterize the model using data for 66 countries from 1995 to 2018. I find that
globalization decelerated structural transformation of the countries whose comparative
advantage in goods was strengthened by globalization and accelerated that of the coun-
tries with weakened comparative advantage in goods. For countries whose comparative
advantage wasn’t significantly affected, globalization had a negligible impact on their
structural transformation. I show that globalization shifts a country’s comparative ad-
vantage if trade costs faced by the country change disproportionately across goods and
services. Illustration of the mechanism is as follows: if goods export trade costs fall
faster than goods import trade costs for a country, its comparative advantage in goods
strengthens. If services export trade costs relative to services import trade costs decline,
1
2the comparative advantage in goods weakens. Therefore, if a country’s export trade
cost relative to import trade cost for goods and services change at different rates, its
comparative advantage shifts.
1.1 Introduction
Structural transformation is a secular reallocation of production shares in value added
away from goods towards services. It is a prominent feature of economic development
and has been widely studied.1 Since international trade implies that production and
consumption can be separated, trade, in principle, can play a major role in structural
transformation.
The literature studying the effect of trade on structural transformation, which started
with Matsuyama (2009) and Uy, Yi, and Zhang (2013), has mostly focused on trade in
goods (agriculture and manufacturing). In their review of the literature, Herrendorf,
Rogerson, and Valentinyi (2014) conclude that “an open question moving forward con-
cerns the extent to which increased trade in services will influence the nature of structural
transformation.” Indeed, international trade in services accounts for over one-third of
the total global trade.2 Moreover, many countries exhibit strong cross-sector specializa-
tion. For example, in 2018, goods share of US exports was 50%, whereas its share of
imports was 71%. In contrast, China’s goods shares of exports and imports were 84%
and 70%, respectively.3 Furthermore, world trade in both goods and services has grown
faster than world GDP for the past few decades.4 These data patterns imply that glob-
alization, i.e., a reduction in trade costs for goods and services, can significantly affect
the speed of structural transformation of different countries.
1See Herrendorf, Rogerson, and Valentinyi (2014) for the survey of the literature and the implica-
tions of structural transformation on economic issues, such as aggregate productivity growth and wage
inequality.
2In 2018, world total goods and services exports were $14.2 and $7.6 trillion USD, based on the
author’s calculation from the OECD Inter-Country Input-Output Tables (ICIO). Other data patterns
in the introduction are also from OECD ICIO. The top three service industries in terms of their trade
volumes were wholesale and retail trade; financial and insurance services; and professional, scientific,
and technical services.
3The specialization patterns of the US and China have strengthened over time. See Figure 1.3 in
Section 2.3 for details.
4From 1995 to 2018, the ratios of world export in goods and services to world GDP rose from 13%
and 6% to 16% and 9%, respectively.
3This paper asks, “how do changes in trade costs for both goods and services af-
fect the pace of structural transformation of 66 countries from 1995 to 2018?” The
main conclusion of this paper is that globalization had heterogeneous effects on the
speed of structural transformation of countries. For some countries, such as Vietnam,
globalization decelerated their structural transformation. Others, such as Lithuania,
experienced accelerated structural transformation through globalization. The pace of
structural transformation of the rest of countries, such as China and India, did not get
affected much.5
I further show that the underlying reason for this heterogeneity hinges on how glob-
alization affected countries’ comparative advantage. Globalization strengthened com-
parative advantage in goods for countries with decelerated structural transformation,
weakened that for countries with accelerated structural transformation. Countries whose
comparative advantage did not get altered much by globalization was the ones whose
structural transformation was the least affected.
Globalization affects comparative advantage of a country because comparative ad-
vantage depends not only on the country’s sectoral productivities but also on its trade
costs, when trade costs are assumed to be asymmetric and sector-specific. Asymmetric
trade costs mean that a country’s export trade costs and import trade costs are different.
As Waugh (2010) and others point out, developing countries face higher export trade
costs than advanced countries, giving rise to the asymmetry.6 Given the asymmetry, the
relative changes in trade costs across goods and services determine how comparative ad-
vantage changes. For example, if goods export trade costs fall faster than goods import
trade costs for a country, its comparative advantage in goods strengthens. If services
export trade costs relative to services import trade costs decline, the comparative ad-
vantage in goods weakens. Therefore, if a country’s export trade costs relative to import
trade costs for goods and services change at different rates, its comparative advantage
shifts.
5In the baseline economy, for Vietnam, Lithuania, China, and India, goods share of GDP decreased
by -0.5 percentage points (p.p.), 9.7 p.p., 12.1 p.p., and 9.5 p.p., respectively. In the counterfactual
economy with trade costs fixed at the 1995 level, the decline of goods share was 10.4 p.p., -0.7 p.p., 11.5
p.p., and 8.3 p.p.
6For evidence of the asymmetry, see Kee, Nicita, and Olarreaga (2009) for an example of policy
barriers; Limao and Venables (2001); Mesquita-Moreira, Volpe, and Blyde (2008); and Blum et al.
(2019) for transportation or inventory costs; and Merga (2021) for financial constraints on fixed costs
of trade.
4To answer the aforementioned research question, I first construct a multi-country,
multi-sector Ricardian trade model with input-output linkages similar to those of Cravino
and Sotelo (2019) and Lewis et al. (2021). A representative household in each country has
a non-homothetic preference with non-unitary substitution elasticity (Comin, Lashkari,
and Mestieri, 2021). There are three channels through which the model primitives (sec-
toral productivities and bilateral iceberg trade costs) generate structural transformation.
First, productivity growth or a reduction in trade costs in any sector increases the house-
hold’s real income. The household shifts their expenditure toward more income-elastic
sectors (income effect). Second, changes in productivities or trade costs that differ across
sectors generate relative price movements. If sectoral outputs are complements in the
household preference, and the relative price of a sector rises, then the expenditure share
of that sector increases (price effect). Through these two channels, the household’s ex-
penditure shares change as productivities or trade costs change. This, in turn, alters
the production pattern of a country. Finally, changes in these model primitives generate
changes in comparative advantage, affecting countries’ cross-sector specialization, where
a country net-exports outputs in the sector of comparative advantage and net-imports
outputs in the sector of comparative disadvantage (cross-sector specialization effect).
For the quantitative analysis, I parameterize the model using input-output, trade,
and price data from 1995 to 2018 for 36 advanced and 30 developing economies, which
account for more than 90% of world export and GDP.7 I aggregate 45 industries in
the data into three broad sectors: goods, producer services, and consumer services. I
classify a service industry as producer services if more than 40% of its outputs are used
as intermediate inputs. I show that producer services are the more tradable part of
services, while consumer services are rarely traded. In 2018, 13% and 2% of producer
and consumer services, respectively, were traded internationally.8
There are two main reasons for splitting services into the two sectors. First, the
prices of consumer services suffer from measurement issues because they are mostly
7One technical contribution of this paper is the construction of price sequences (Appendix A.1.2). In
the model with input-output linkages, sectoral prices measure prices of gross outputs, not value added.
However, not all countries report gross-output price deflators (GOPD). I propose a method to construct
GOPD from value-added price deflators (VAPD) utilizing the double deflation method. I verify the
validity of the method by comparing GOPD in data and GOPD imputed from VAPD for countries
where data for both GOPD and VAPD are available. My method corrects the bias of using VAPD as a
proxy for GOPD.
8For comparison, 22% of goods were traded in 2018.
5non-market services.9 Since I use price data to derive trade costs, the split allows me
to mitigate measurement problems of trade costs for tradable services. Second, because
consumer services are rarely tradable, cross-sector specialization (third channel) occurs
mainly between goods and producer services.
Under the sectoral definition of goods, producer, and consumer services, the general
pattern of structural transformation is that as a country develops, its goods share of
GDP decreases and both producer and consumer services shares increase. Consumption
shares measured in final expenditures follow the same qualitative trend.
In the model, the key elasticities that govern how sensitive production shares are
to the model primitives (productivities and trade costs) are preference parameters, pro-
duction parameters, and trade elasticities. For household preferences, income and sub-
stitution elasticities largely determine the size of the income and price effect. From the
model, following Comin, Lashkari, and Mestieri (2021), I derive an equation that exposes
a log-linear relationship among expenditure shares, relative prices, relative incomes, sub-
stitution elasticity, and income elasticities. Using panel data on consumption, prices,
and incomes, I run a generalized method of moments (GMM) estimation and find that
outputs of the three sectors (goods, producer services, and consumer services) are com-
plements and that the two types of services are more income-elastic than goods in the
household’s preference.10 For each sector’s production function with input-output link-
ages, I estimate the substitution elasticity among four inputs (value added along with
intermediate inputs from the three sectors). The degree of substitutability of inputs
in production can amplify or dampen the three channels of structural transformation.
Note that this substitution elasticity in the production function is different from that
in the utility function. From the cost minimization problem of firms, I obtain log-linear
equations that relate relative input shares, relative prices, and substitution elasticities.
Utilizing panel data on input-output tables and prices, I run a GMM estimation and
9See Deaton and Heston (2010) for the discussion on the measurement problem of cross-country
comparison of prices of non-market services.
10This result is consistent with the previous literature. My estimates are similar to the papers that
adopt similar definitions of sectors (goods and two service sectors), such as Cravino and Sotelo (2019)
and Han, Miranda-Pinto, and Tanaka (2022), who utilize US national data and Canadian provincial
data, respectively.
6find that the production functions are close to Cobb-Douglas (unitary substitution elas-
ticities).11 Lastly, I take trade elasticities, which mainly govern how sensitive trade flows
and prices respond to changes in domestic and foreign productivities and bilateral trade
costs, from the literature.
For the model primitives, following Święcki (2017) and many others (e.g. Lewis
et al., 2021), I recover them by inverting the model, which provides expressions for
sectoral productivities and bilateral trade costs in terms of sectoral prices and trade
shares.12 Using these expressions, the primitives are retrieved from the data on prices
and trade shares. Identification of productivities comes from the gap between input
and output prices. If input prices are high and an output price is low for a country-
sector, this implies that the country-sector is converting inputs into output efficiently,
implying high productivity. Sectoral trade costs are identified by sectoral trade shares
and relative prices. Given relative prices, if a destination country imports more from
an origin country, inferred trade costs are lower. Given trade shares, if the price at
the destination country relative to the origin country is low, which means that the
destination country-sector is more efficient at production or it can more easily import
from the other countries, the implied trade cost is low.
From the calculated trade costs, I first document the asymmetry of trade costs for
both goods and producer services: developing countries face higher export trade costs
than advanced countries.13 Second, I further show that the asymmetry is decreasing over
time for both sectors. Export trade costs faced by developing countries were converging
to those of advanced countries. The main contributing factor for this phenomenon is
the significant decline in trade costs from developing to advanced economies. From
1995 to 2018, median trade costs from developing to advanced economies for goods and
11The previous literature mostly uses Cobb-Douglas production functions (e.g. Lewis et al., 2021).
12I take this approach because data to calculate sectoral total factor productivities, such as KLEMS,
are only available for a small subset of countries. Furthermore, this approach can capture over-time
changes of trade costs which might differ across origin, destination, and sector. In contrast, the es-
timation approach based on the gravity model of trade, where trade costs are inferred from variables
such as distance, common language, colonial relationship, and shared land border, is not suitable to
analyze the dynamics of trade costs: the usual gravity variables aforementioned are stable across time
and symmetric across origin and destination.
13Because the trade costs for consumer services are prohibitively high, which reflects the low trad-
ability of consumer services, they have negligible effects on the quantitative results of this paper (Ap-
pendix A.4.2). Therefore, I focus on the cross-sector difference in changes in the trade costs for goods
and producer services.
7producer services decreased by 38.2% and 24.4%, respectively. For comparison, median
trade costs across all country pairs declined by 15.7% and 10.4% for goods and producer
services, respectively. Since the main focus of this paper is on globalization, I discuss
the sectoral productivity growth and its implications on structural transformation in
Appendix A.2.3.
One key aspect of trade where decreasing trade cost asymmetry matters is countries’
effective comparative advantage. Modifying the definition of Deardorff (2014), I define
the strength of country i’s effective comparative advantage in goods relative to the rest
of the world, denoted by r as the following.14
Definition 1.1. Country i’s comparative advantage in goods relative to the rest of the
world is defined by
CAgir ≡
{
(i’s goods productivity)
(i’s producer services productivity)
/
(mean goods productivity in r)
(mean producer services productivity in r)
}
×
{
(i’s mean goods export trade cost to r)
(i’s mean goods import trade cost from r)
}−1/2
×
{
(i’s mean producer services export trade cost to r)
(i’s mean producer services import trade cost from r)
}1/2
,
where mean productivities are average productivities weighted by GDP and mean trade
costs are average trade costs weighted by volumes of trade with i’s all trading partners.
The weights are from 1995, the initial year, to control for the endogenous response of
output and trade to changes in productivities and trade costs.15
Under this measure, ceteris paribus, if a country’s goods export trade costs decline
faster than its goods import trade costs, the country’s comparative advantage in goods
strengthens. If a country’s export trade costs relative to import trade costs in producer
services decline, its comparative advantage in goods weakens. This is one reason why
14Deardorff (2014) defines the effective comparative advantage in an Armington model. In their
definition based on an inequality, one can compare two countries and determine which country has
effective comparative advantage in which sector. I modify their definition so that it can measure the
strength of a country’s comparative advantage in a Ricardian model. For conciseness, I use the terms
“effective comparative advantage” and “comparative advantage” interchangeably. I describe changes in
comparative advantage in terms of goods. For instance, the statement that a country’s comparative
advantage in goods weakens means either that its comparative advantage in goods weakens or that
its comparative advantage in services strengthens. Since consumer services are barely tradable, my
definition does not include productivity or trade cost of the sector.
15For the rest of the paper, for a country’s export and import trade costs, I use the same measure.
8understanding the relative changes in trade costs across goods and producer services
is important. They affect the comparative advantage, hence countries’ specialization
patterns.
Although globalization can affect structural transformation through comparative ad-
vantage. In theory, it can alter the speed of structural transformation even without any
changes in the measure of comparative advantage. For example, a uniform reduction
in trade cost across sectors and country pairs does not change comparative advantage
of any countries, but it increases real income of households, and through the income
effect, it can affect countries’ structural transformation. Another example is a hypo-
thetical scenario where only goods trade costs are reduced uniformly across all country
pairs. Although this example does not incur any changes in comparative advantage, it
will lower goods price relative to producer services, and through the price effect, it can
impact structural transformation. Therefore, how globalization affects structural trans-
formation through shifts in comparative advantage is a quantitative question, which I
explore in the following counterfactual exercises.
With the parameterized model, I run three counterfactual exercises to quantify the
effects of the changes in trade costs for goods and services on the structural transfor-
mation of 66 countries. For all the exercises, I let productivities evolve following the
calculated historical changes. First, to quantify the effect of changes in the trade costs
for services, I fix the trade costs for producer and consumer services at their level in 1995
and let the trade costs for goods follow the calculated sequence.16 Second, to assess the
impact of changes in the trade costs for goods, I fix the trade costs for goods at their
level in 1995 and let the trade costs for the two service sectors follow their computed
levels. Lastly, to quantify the impact of changes in the trade costs for all three sectors, I
fix all trade costs at their initial-year level. I compare the evolution of production shares
across the baseline economy and three counterfactual economies from 1995 to 2018.
To understand how globalization interacts with structural transformation across the
counterfactual and baseline economies, I first study China, the largest developing coun-
try. As for the globalization pattern of China, I show that its export trade costs were
converging to its import trade costs for both goods and producer services, consistent
16Due to the low tradability of consumer services, trade costs for consumer services have negligible
effects on the quantitative results (Appendix A.4.2).
9with the general pattern of decreasing asymmetry of trade costs. The rates of conver-
gence for the two sectors were similar. For goods, export trade costs relative to import
trade costs decreased by 83% from 1995 to 2018. For producer services, the decline was
85%. This means goods globalization strengthened China’s comparative advantage in
goods, whereas globalization in producer services weakened it. However, since goods
and producer services trade costs decreased proportionally, its overall impact on the
comparative advantage was limited.
Regarding the structural transformation pattern of China, its goods share in GDP
decreased by 12.1 percentage points (p.p.) from 1995 to 2018 in the baseline economy.17
The decline would have been 0.1 p.p. if only trade costs for goods had changed and 22.6
p.p. if only trade costs for services had changed. Without any changes in trade costs for
both goods and services, the decline would have been 11.5 p.p., similar to the baseline
number.
The result for China suggests that the relative speed of a reduction in trade costs
for goods versus services matters for structural transformation due to its impact on
comparative advantage. Analyzing the result for all 66 countries, I show that this cer-
tainly is the case. By comparing the baseline and counterfactual #1, I show that for
countries where export trade costs decreased faster than import trade costs for producer
services, the structural transformation was accelerated by producer services globaliza-
tion. Contrasting the baseline and counterfactual #2 yielded a symmetric result: when
a country’s export trade costs declined faster than import trade costs for goods, its
structural transformation got decelerated. In conjunction with the decreasing asymme-
try in both sectors, this trend means that services globalization generally accelerated
the structural transformation of the poorer countries and decelerated that of the richer
ones. The goods globalization had the opposite effect.
For the total impact of globalization on structural transformation, the key is which
of the two forces dominates and how the overall changes in trade costs affect a country’s
comparative advantage. To make this point, I first propose a measure called a pro-
portionality index, which computes how the dynamics of trade costs affect comparative
advantage.
17In the data, the decline was 15.4 p.p. For the model fit, see Section 1.6.1.
10
Definition 1.2. Proportionality index for country i is defined by
∆t
(
log
(
i’s goods export trade cost at t
i’s goods import trade cost at t
))
−∆t
(
log
(
i’s producer services export trade cost at t
i’s producer services import trade cost at t
))
,
where ∆t(zt) ≡ z2018 − z1995. i’s import and export trade costs are weighted averages of
export and import trade costs across all i’s trading partners. The weights are the trade
flows in 1995 to control for the endogenous response of trade flows to changes in trade
cost.
If the index is further away from zero, it means that the trade cost changes are more
disproportionate across sectors. A negative (positive) value of the index means that
the changes in trade costs strengthened (weakened) the country’s effective comparative
advantage in goods.
I measure the impact of globalization on structural transformation by comparing the
speed of structural transformation in the baseline and counterfactual #3, where all trade
costs are fixed at the 1995 level. For instance, from 1995 to 2018, if a country’s GDP
shares in goods, producer, and consumer services change by -5 p.p., +4 p.p., and +1
p.p. in the baseline economy, and by -2 p.p., +1 p.p., and +1 p.p. in the counterfactual
economy. I measure the impact of globalization as a triple of +3 p.p., -3 p.p., 0 p.p.18
I find that the closer the index of a country is to zero, the closer the gap between the
speed of structural transformation between the counterfactual and the baseline economy.
This means that the proportionality implies a small impact of globalization on structural
transformation. Countries that went through proportional globalization include China
and India. Furthermore, I find a strong linear relationship between the index and the
effect of changes in trade costs on structural transformation from goods to producer
services. For countries with negative indices (e.g., Vietnam), globalization slowed down
18An alternative measure would be to calculate how much of the structural transformation pattern
is explained by globalization. However, this incurs a comparison between small numbers for advanced
economies that had already undergone significant structural transformation by 1995. For example, the
goods share of GDP for the US was 21% in 1995 and decreased by 3.7 p.p. from 1995 to 2018. In
the no-globalization counterfactual, the share fell by 2.0 p.p. Therefore, although the absolute effect is
small at +1.7 p.p. (-2.0 p.p. - (-3.7 p.p.)), if I measure the relative effect, it is large (+1.7 p.p./(-3.7
p.p.)). For this reason, I use the absolute measure to discuss the quantitative results.
11
their structural transformation: their decrease in goods share of GDP and increase in
producer services share of GDP were slower in the baseline economy. The opposite holds
for countries with positive indices (e.g., Lithuania). These effects were more pronounced
for countries with indices further from zero, showing that globalization played a major
role in the structural transformation from goods to producer services of countries that
experienced more disproportionate trade cost changes across goods and producer ser-
vices. As for the consumer services, which is barely tradable, globalization had a limited
impact on how its share of GDP changes over time.
Unlike the analyses of goods globalization and services globalization in isolation,
globalization in both goods and services did not exhibit any trend with regards to the
income levels of countries. Despite a reduction in export trade costs relative to import
trade costs showing strong correlation with the income level of countries in both goods
and producer services, the proportionality index which captures how closely goods and
services trade costs co-moved, did not show any trends with regards to income level, and
hence there was no clear trend of how globalization differentially affects developing and
advanced countries.
The main contribution of this paper is as follows. First, this paper is one of the
first to analyze and find the significant role of services trade in structural transforma-
tion.19 Second, I contribute to understanding why trade has idiosyncratic effects on the
structural transformation of countries.20 Regarding the first two points, I show that the
relative speed of changes in trade costs in goods versus services, which affects countries’
comparative advantage, is the key to understand globalization’s impact on structural
transformation. Third, in regards to the literature on trade costs, I identify structural
transformation as another macroeconomic phenomenon where trade cost asymmetry
matters. Fourth, I deepen the understanding of asymmetric trade costs by showing that
asymmetry exists in producer services trade as well and that the asymmetry is decreas-
ing for both goods and producer services. Finally, I contribute to the literature on the
importance of understanding structural transformation with models with different types
of services. I find that trade mainly affects structural transformation from goods to
19Another such paper is Han, Miranda-Pinto, and Tanaka (2022) who study the role of intranational
and international trade in services of Canadian provinces on their structural transformation.
20Previous studies that show the cross-country idiosyncrasy of the impact of trade on structural
transformation include Święcki (2017) and Sposi, Yi, and Zhang (2018).
12
producer services, the more tradable segment of services, but not so much for that from
goods to consumer services, the less tradable service sector.
The rest of the paper is organized as follows. Section 1.1.1 provides the literature
review. In Section 1.2, I lay out the model. Section 1.3 illustrates how changes in
productivities and trade costs generate structural transformation. Section 2.3 describes
the data sources and the data patterns of international trade and structural transfor-
mation. Section 1.5 shows how I estimate the elasticities. Section 2.4 describes how I
calibrate productivities and trade costs and shows the data fit of the calibrated model.
Furthermore, I discuss the patterns of the historic changes in sectoral trade costs and
productivities. Section 2.5 contains the main quantitative results. Lastly, I conclude in
Section 2.6.
1.1.1 Related Literature
This paper is related to three strands of the literature. The first and most closely related
literature is on structural transformation in an open-economy. In their decomposition
exercises, Święcki (2017) shows that differential productivity growth was the main mech-
anism of structural transformation and that the role of globalization was heterogeneous
across countries, and Sposi, Yi, and Zhang (2018) demonstrate that different mechanisms
played different roles in countries’ structural transformation. I complement these papers
by demonstrating why globalization had heterogeneous effects on countries structural
transformation. The underlying reason for the heterogeneity comes from relative speed
of trade cost reduction in goods versus services and its impact on structural transfor-
mation through shifts in comparative advantage. This conclusion is also an answer to
the point made by Herrendorf, Rogerson, and Valentinyi (2014) that the role of services
trade is an open question in the literature.
My finding that relative speed of globalization in different sectors is important in
understanding structural transformation complements the research by Betts, Giri, and
Verma (2016). They study Korea and discuss how cross-sector differences in trade poli-
cies between agriculture and manufacturing affected the speed of the country’s indus-
trialization. I generalize their argument to changes in total trade costs for goods and
services by utilizing cross-country variation in the multi-country analysis. By doing so, I
show that cross-sector differences are indeed the key mechanism of how changes in trade
13
costs affect structural transformation.
My paper is also related to papers that ask different research questions with multi-
country models of structural transformation with tradable services.21 Other related
papers include those study the role of trade for specific countries and the ones with
theoretical analysis of structural transformation in an open economy.22
Second, this paper furthers the international trade literature on the importance of
asymmetric trade costs in explaining macroeconomic phenomena.23 Waugh (2010) ar-
gues that for a gravity model to be consistent with both trade and price data, developing
countries must face higher export trade costs than advanced countries. He further shows
that this asymmetry is quantitatively important in understanding cross-country income
differences. I complement their paper and other papers on trade cost asymmetry by
demonstrating that there also exists trade cost asymmetry in services, that the cross-
country asymmetry is declining over time for both goods and producer services, and
that this dynamics is important in explaining the impact of globalization on structural
transformation.
Thirdly, this paper contributes to the literature on structural transformation with
different types of services (e.g. Jorgenson and Timmer, 2011; Buera and Kaboski, 2012;
Duernecker, Herrendorf, and Valentinyi, 2017; and Duarte and Restuccia, 2020). This
paper adds to the literature by showing that highly tradable services (producer services)
are the ones that are mainly affected by international trade.
21Cravino and Sotelo (2019) study the role of trade cost changes in rising skill premiums by affecting
the pace of structural transformation. Lewis et al. (2021) study the converse of my research question and
show that that structural transformation led to the recent global trade slowdown. Sposi, Yi, and Zhang
(2021) show that sector-biased productivity growth and trade cost reductions explain deindustrialization
and industry polarization, the two prominent features of structural transformation.
22Han, Miranda-Pinto, and Tanaka (2022) study the role of intranational and international trade in
services of Canadian provinces on their structural transformation. Kehoe, Ruhl, and Steinberg (2018)
show that structural transformation of the US mainly comes from differential productivity growth,
not trade deficits. Reyes-Heroles (2018) decomposes structural transformation of the US into multiple
mechanisms including globalization. Teignier (2018) studies the role of trade in transition from agri-
cultural to non-agricultural sectors in South Korea and Great Britain. Matsuyama (2019) studies the
interconnectedness of closed-economy mechanisms and globalization.
23There is ample evidence that the export costs faced by developing countries are higher than ad-
vanced countries. See Kee, Nicita, and Olarreaga (2009) for an example of policy barriers; Limao and
Venables (2001); Mesquita-Moreira, Volpe, and Blyde (2008); and Blum et al. (2019) for transportation
or inventory costs; and Merga (2021) for financial constraints on fixed costs of trade.
14
1.2 Model
There are I countries and S sectors. The sets of countries and sectors are denoted by
I and S, respectively. Time is discrete, and a representative household in each country
inelastically supplies a unit of labor every period. The household has non-homothetic
preferences with non-unitary substitution elasticities. In each sector, there is a unit con-
tinuum of tradable varieties. Production of a variety requires labor (value added) and
intermediate inputs.24 Productivities differ across countries, sectors, and time. Cross-
border trade of a variety incurs an iceberg trade cost, which differs across origin, des-
tination, sector, and time. Exogenous dynamics of productivities and trade costs drive
structural transformation in the model through three channels: first, when income rises,
the household shifts their expenditure to more income-elastic sectors (income effect);
second, the household’s expenditure shares increase for sectors with increasing rela-
tive prices if sectoral consumptions are complements (price effect); and lastly, countries
specialize in sectors with stronger comparative advantage (cross-sector specialization ef-
fect). Since the model does not have endogenous dynamics, I omit time notations and
reintroduce them later to define an equilibrium.
1.2.1 Technology
Production of Varieties
In each sector, there is a unit continuum of tradable varieties. Each country has technolo-
gies for producing all varieties. Production of a variety requires labor and intermediate
inputs. The production function for variety z ∈ [0, 1] in sector k ∈ S and country i ∈ I
is
yki (z) = A
k
i α
k
i (z)
{(
ψLki
) 1
ρk
(
Lki (z)
) ρk−1
ρk +
∑
h∈S
(
ψhki
) 1
ρk
(
Qhki (z)
) ρk−1
ρk
} ρk
ρk−1
,
where Aki and α
k
i (z) are sector-wide and sector-variety-specific productivities, respec-
tively. Lki (z) and Q
hk
i (z) are labor employed and sector-h aggregates used. ψ
Lk
i and
24Since the model abstracts from capital inputs, labor inputs in the model should be interpreted as
value-added inputs.
15
ψhki denote input weights on labor and sector-h aggregates. Note that production of
a variety in sector k requires inputs from all sectors, reflecting input-output linkages.
ρk is the elasticity or substitution among labor and intermediate inputs.25 Sector-wide
productivity Aki is exogenously given each period. Following Eaton and Kortum (2002),
variety-specific productivity αki (z) is drawn from a Frèchet distribution with cumula-
tive distribution function F k(z) = e−a−θ
k
. It is drawn at the initial period and stays
constant.26 I assume that the variety markets are perfectly competitive.
Cost minimization by a variety producer yields the production cost given by rki (z) =
Rki
Aki α
k
i (z)
, where Rki is the price of an input bundle. R
k
i is given by
Rki =
(
ψLki W
1−ρk
i +
∑
h∈S
ψhki
(
P hi
)1−ρk) 11−ρk
,
where Wi is the wage rate and P ki is the sector-k aggregate price in country i.
International Trade and Production of Aggregates
Varieties can be traded internationally subject to iceberg trade costs. For an origin-j,
sector-k variety producer to export its output to destination i, an iceberg trade cost of
τkji is incurred. For one unit of the variety to arrive in country i, τ
k
ji ≥ 1 units need to
be shipped. I assume frictionless domestic trade: τkii = 1 ∀i ∈ I, k ∈ S. Furthermore,
re-exporting of a variety is not allowed: it is forbidden to import a variety and sell it to
another country.27
25The quantitative result is robust to an alternative specification of production functions such as a
nested CES production function, where an inner nest aggregates intermediate inputs and an outer nest
combines labor and the aggregated intermediate inputs. See Appendix A.4.5 for details.
26An alternative is to assume that αki (z)’s are drawn each period. The two assumptions give the
exact same equilibrium conditions, hence the same macroeconomic implications.
27Eaton and Kortum (2002) assume the following triangle inequality: τij < mink ̸=i,jτikτkj . However,
since the calibrated trade costs violate the triangle inequality, I adopt an alternative assumption. For
goods trade costs in 2018, 9% of country-pairs violated the triangle inequality.
16
In each sector, a unit continuum of tradable varieties are aggregated into a non-
tradable aggregate. The aggregates are then used for the variety production and house-
hold consumption. The production function for aggregates is given by
Y ki =
(∫ 1
0
qki (z)
ηk−1
ηk dz
) ηk
ηk−1
,
where Y ki is the amount of an aggregate produced, q
k
i (z) is the amount the variety z
used as inputs, and ηk governs substitutability among varieties. The aggregation is
done by a perfectly competitive firm. The aggregate producer buys a variety from the
producer with the lowest total cost (gross of production and trade cost). Therefore,
pki (z) = minj∈I τ
k
jir
k
j (z).
The cost minimization of the aggregate firm yields the expressions for the aggregate
prices and import shares.28 The price of aggregates is given by
P ki = γ
k
∑
j∈I
(
τkjiR
k
j
Akj
)−θk
− 1
θk
, (1.1)
where γk = Γ
(
θk+1−ηk
θk
) 1
1−ηk and Γ(·) denotes a gamma function.29 The import share
of country-j sector-k varieties in country-i sector-k aggregate production, πkji is given by
πkji ≡
Xkji
P ki Y
k
i
=
(
τkjiR
k
j
Akj
)−θk
∑
o∈I
(
τkoiR
k
o
Ako
)−θk , (1.2)
whereXkji denotes country-i’s aggregate producer’s expenditure on varieties from country
j: Xkji ≡
∫ 1
0 1
k
ji(z)p
k
i (z)q
k
i (z)dz, where 1
k
ji(z) is a shorthand notation for an indicator
function which equals one if the aggregate producer buys the variety from j and zero
otherwise. The two equations capture how domestic and foreign productivities, costs of
input bundles, and trade costs affect domestic equilibrium outcomes.
28For the derivations of Equations (1.1) and (1.2), see Eaton and Kortum (2002).
29As in Eaton and Kortum (2002), I impose a restriction of ηk < θk + 1 for well-defined prices.
17
1.2.2 Preferences
Each country’s representative household has a non-homothetic constant elasticity of
substitution (non-homothetic CES) preferences as in Comin, Lashkari, and Mestieri
(2021, CLM, hereafter).30 The utility of the household in country i, Ui, is implicitly
defined by
∑
k∈S
(ϕki )
1
σU
1−σ
σ
ϵk
i (C
k
i )
σ−1
σ = 1, (1.3)
where Cki and ϕ
k
i denote the consumption and the utility weight parameter of sector-k
aggregates, respectively. σ denotes substitution elasticity, and ϵk is a non-homotheticity
parameter for sector-k aggregates.31 The household inelastically supplies one unit of
labor and makes exogenous net transfers to the outside world.32 It maximizes utility
given the budget constraint,
∑
k∈S P
k
i C
k
i =Wi−NXi, where NXi denotes an exogenous
net transfer. Incorporating the exogenous net transfers in the model is due to the large
trade imbalances of countries in the data.
The solution to the household maximization problem is given by the budget con-
straint along with the following two equations.
Pi =
{∑
k∈S
ϕki (P
k
i )
1−σU (1−σ)(ϵ
k−1)
i
} 1
1−σ
, and
P ki C
k
i
PiUi
= ϕki
(
P ki
Pi
)1−σ
U
(1−σ)(ϵk−1)
i . (1.4)
The first equation is the definition for a non-homothetic CES price index, or the price
30Non-homothetic CES preferences, unlike Stone-Geary preferences, feature Engel curves that do
not level off as an income grows. CLM provides both micro and macro evidence of the non-leveling-off
Engel curves. In Section 1.4.3, I reiterate CLM’s point for the data used for this paper. As for the
assumption of a representative households, in their Online Appendix B, CLM demonstrates that it is
straightforward to extend the model to households with heterogeneous incomes within a country.
31As Hanoch (1975) shows, if ϵk > 0 ∀k and σ ∈ (0, 1) ∪ (1,∞), this utility is globally monotone
and quasi-concave yielding a well-defined utility function. Estimation in Section 1.5.1 shows that this
parameter restriction is satisfied.
32The net transfers are made in units of a numeraire. For the calibration, I set the price of value-
added in the US as a numeraire. In Appendix A.4.6, I show that the quantitative result of this paper
is robust in the case without the net transfer terms.
18
of a unit of utility, Pi. The second equation is the demand function for a sectoral
aggregate. Note that non-homothetic CES preferences nest homothetic CES preferences
if ϵh = ϵk ∀h, k ∈ S.
Non-homothetic CES preferences feature two channels through which structural
transformation can occur. First, due to relative price changes along with non-unitary
substitution elasticity, households reallocate expenditures across sectors. The effect of
relative price changes on the relative expenditure shares is given by
∂ log
(
P ki C
k
i
P hi C
h
i
)/
∂ log
(
P ki
P hi
)
= 1− σ.
If σ ∈ (0, 1), sectoral aggregates are complements, and if σ > 1, they are substitutes.
Second, the expenditure reallocation also happens through real income growth or,
equivalently, utility growth along with non-homothetic preference. The effect of utility
growth on relative expenditure shares is given by
∂ log
(
P ki C
k
i
P hi C
h
i
)/
∂Ui = (1− σ)(ϵk − ϵh).
If σ ∈ (0, 1), then the household reallocates expenditures towards sectors with higher
ϵk’s if their real income grows. The reallocation of the household’s expenditure can, in
turn, bring about the reallocation of production shares in value added mediated by trade
openness and input-output linkages.
1.2.3 Market Clearing
Labor markets clear: Li =
∑
k∈S
∫ 1
0 L
k
i (z)dz, ∀i ∈ I. Aggregate markets clear: Y ki =
LiC
k
i +
∑
h∈S
∫ 1
0 Q
kh
i (z)dz, ∀i ∈ I, and ∀k ∈ S. Variety markets clear: yki (z) =∑
j∈I 1
k
ij(z)τ
k
ijq
k
j (z) ∀i ∈ I,∀k ∈ S, and ∀z ∈ [0, 1]. Net transfers sum to zero:∑
i∈I LiNXi = 0.
1.2.4 Equilibrium
This section collects equilibrium conditions for each time period and introduces time
notations. Time-varying model primitives are country-sector productivities(
{Aki,t}i∈I,k∈S,t∈T
)
, country-destination-sector trade costs
(
{τkij,t}i,j∈I,k∈S,t∈T
)
, labor
19
endowments ({Li,t}i∈I,t∈T ), and exogenous net export terms ({NXi,t}i∈I,t∈T ), where T
denotes the set of time periods.
For each t ∈ T , the equilibrium conditions from the household maximization are
(Demand)
P ki,tC
k
i,t
Pi,tUi,t
= ϕki
(
P ki,t
Pi,t
)1−σ
(Ui,t)
(1−σ)(ϵk−1) , ∀i ∈ I, k ∈ S; (H1)
(Price index) Pi,t =
{∑
k∈S
ϕki (P
k
i,t)
1−σU (1−σ)(ϵ
k−1)
i,t
} 1
1−σ
∀i ∈ I; and (H2)
(Budget constraint)
∑
k∈S
P ki,tC
k
i,t = Pi,tUi,t =Wi,t −NXi,t, ∀i ∈ I. (H3)
The conditions from the variety-producing firm’s problem are
(Unit cost) Rki,t =
(
ψLki W
1−ρk
i,t +
∑
h∈S
ψhki
(
P hi,t
)1−ρk) 11−ρk
, ∀i ∈ I, ∀k ∈ S; (F1)
(Value-added share) Wi,tLki,t =
ψLki W
1−ρk
i,t(
Rki,t
)1−ρk ∑
j∈I
πkij,tP
k
j,tY
k
j,t, ∀i ∈ I,∀k ∈ S; and
(F2)
(Intermediate share) P hi,tQ
hk
i,t =
ψhki
(
P hi,t
)1−ρk
(
Rki,t
)1−ρk ∑
j∈I
πkij,tP
k
j,tY
k
j,t, ∀i ∈ I, ∀h, k ∈ S,
(F3)
where Qhki,t =
∫ 1
0 Q
hk
i,t (z)dz.
The equilibrium conditions from international trade and global net transfers are
20
(Price for aggregates) P ki,t = γ
k
∑
j∈I
(
τkji,tR
k
j,t
Akj,t
)−θk
− 1
θk
∀i ∈ I,∀k ∈ S (G1)
(Import share) πkji,t =
(
τkji,tR
k
j,t
Akj,t
)−θk
∑
o∈I
(
τkoi,tR
k
o,t
Ako,t
)−θk , ∀i, j ∈ I, k ∈ S (G2)
(Net transfer)
∑
k∈S
P ki,tY
k
i,t =
∑
k∈S
∑
j∈I
πkij,tP
k
j,tY
k
j,t − Li,tNXi,t, ∀i ∈ I; and (G3)
(Global market)
∑
i∈I
Li,t ·NXi,t = 0. (G4)
Finally, the market clearing conditions are
(Aggregate goods) Y ki,t = Li,tC
k
i,t +
∑
h∈S
Qkhi,t , ∀i ∈ I,∀k ∈ S; and (M1)
(Labor market) Li,t =
∑
k∈S
Lki,t, ∀i ∈ I, (M2)
where Lki,t =
∫ 1
0 L
k
i,t(z)dz.
Definition 1.3. The competitive equilibrium at time t is prices
({
Wi,t,
{
P ki,t
}
k∈S
}
i∈I
)
and allocations
({
Cki,t, L
k
i,t, Y
k
i,t
{
Qhki,t
}
h∈S
}
i∈I,k∈S
)
that satisfy the above 12 conditions
given (i) productivities
({
Aki,t
}
i∈I,k∈S
)
, (ii) bilateral trade costs
({
τkij,t
}
i,j∈I,k∈S
)
,
(iii) net transfer terms
({NXi,t}i∈I), and (iv) populations ({Li,t}i∈I).
21
1.3 Model Mechanism
In this section, I explain how the dynamics of productivities (Aki,t’s) and bilateral trade
costs (τkij,t’s) induce structural transformation in a simplified version of the model. I
assume a two-country (countries a and b), two-sector (goods denoted by g and services
denoted by s) world. Furthermore, I abstract away from input-output linkages, i.e.,
ψLki = 1, ψ
hk
i = 0 ∀i ∈ I, h, k ∈ S. Assume further that sector-g and s aggregates
are complements (i.e., σ ∈ (0, 1)) and services are more income elastic than goods (i.e.,
ϵs > ϵg).33
1.3.1 Structural Transformation in a Closed Economy
Suppose that cross-border trade costs are infinite (τkij,t =∞ ∀i ̸= j). In this closed econ-
omy, production shares in value-added equal expenditure shares. From Equations (H1)
to (H3) and normalizing country a’s wage to one, expenditure share on sector s relative
to g in country a is given by
P sa,tC
s
a,t
P ga,tC
g
a,t
=
ϕsa
ϕga
(
P sa,t
P ga,t
)1−σ
(Ua,t)
(1−σ)(ϵs−ϵg) , (1.5)
where Ua,t solves
1
Ua,t
=
{∑
k∈S
ϕka(P
k
a,t)
1−σU (1−σ)(ϵ
k−1)
a,t
} 1
1−σ
. (1.6)
From Equation (1.5), expenditure share on s increases when the price of s relative to g,
P sa,t
P ga,t
, or utility, Ua,t, increases. Note that Ua,t is a strictly decreasing function of P sa,t and
P ga,t. Prices of g and s are given by P si,t =
γs
Asa,t
and P gi,t =
γg
Aga,t
. Therefore, structural
transformation from g to s happens through two channels. First, it occurs through
differential productivity growth: if productivity of g increases faster than s, then the
relative price of s increases. Second, it arises through income growth: if productivity of
either g or s increases, then the utility increases. Following the literature, I refer to the
33Estimation in Section 1.5.1 shows that goods and services are complements and that services are
more income-elastic.
22
two channels as price and income effects. Note that the strength of the two effects is
governed by the substitution elasticity σ and the income elasticity ϵk’s.34
1.3.2 Structural Transformation in an Open Economy
Now assume finite trade costs (τkij,t <∞ ∀i ̸= j). Unlike the closed-economy setting, in
the open economy, production and consumption shares can be different due to special-
ization. Defining ωki as country i’s expenditure share on sector k, country a’s production
share in s measured in value-added is given by
vsa,t ≡
Wa,tL
s
a,t
Wa,tLa,t
= ωsa,t︸︷︷︸
Domestic expenditure share in s
+
πsab,tω
s
b,tWb,tLb,t − πsba,tωsa,tWa,tLa,t
Wa,tLa,t︸ ︷︷ ︸
Net export in s relative to a’s GDP
.
(1.7)
Changes in productivities and trade costs affect prices, relative wages, and import
shares. Price and import shares are determined by the following two equilibrium condi-
tions.
P ki,t = γ
k

(
Wi,t
Aki,t
)−θk
+
(
τkji,tWj,t
Akj,t
)−θk
− 1
θk
∀j ̸= i,∀i ∈ {a, b}, ∀k ∈ {g, s}.
πkji,t =
(
τkji,tWj,t
Akj,t
)−θk
(
τkji,tWj,t
Akj,t
)−θk
+
(
τkoi,tWo,t
Ako,t
)−θk , ∀o ̸= j,∀i, j ∈ {a, b}, k ∈ {g, s}.
Wages of the two countries are endogenously determined by supply and demand changes
induced by productivities and trade costs.
Changes in wages, prices, and import shares affect the value-added shares through
three channels which affect the right-hand side of Equation (1.7). First, it can change
relative prices (price effect). It then affects expenditure shares, ωsa and ωsb . For example,
34To be precise, real-income elasticity of sector k is (1− σ)(ϵk − 1). For conciseness, I refer to ϵk as
an income elasticity.
23
if trade cost reductions are faster in g than s, it can increase the relative prices of s.
Second, it can alter countries’ real income by affecting prices and wages (income effect).
This channel also affects the expenditure shares. Third, changes in prices, wages, and
import shares affect the net export term (cross-sector specialization effect).
On top of the closed economy model, there are trade elasticities (θk’s) that govern
the sensitivity of prices and import shares to productivities and prices. They, in turn,
affect the strength of the price, income, and cross-sector specialization effects.
1.3.3 Input-output Linkages and Structural Transformation
So far, for the exposition of the model mechanisms, I have abstracted away from input-
output linkages and assumed that labor (value added) is the only production input.
Without input-output linkages, demand for an output of a sector solely comes from
final demand by households. Therefore, domestic and foreign households’ final demand
determines the sectoral production shares in value-added.
In contrast, with input-output linkages, demand for the sectoral output comes from
both final demand and intermediate input demand. This changes how the model prim-
itives (productivities and trade costs) affect structural transformation in two ways.35
First, sensitivity of prices, wages, and import shares to the primitives depends on input-
output linkages. For example, if a sector uses intermediates from the same sector more
intensively, the sectoral price reduction from the sectoral productivity increase will be
greater. Second, relative price effects from intermediate demand can mute or amplify
the relative price effects from final demand, depending on whether inputs are substi-
tutes (i.e., ρk > 1) or complements (i.e., ρk ∈ (0, 1)) in production. For example, if
ρk > 1, ∀k ∈ {g, s}, and σ ∈ (0, 1), intermediate demand shares move in the opposite
direction from the final demand shares given the price changes. In this case, the price
effect of structural transformation is dampened.
35For how cross-country differences in input-output linkages affect structural transformation, see
Sposi (2019).
24
1.4 Data
In this section, I describe data sources, define sectors, and show data patterns on inter-
national trade and structural transformation.
1.4.1 Data Sources
The yearly data covers 66 countries and the rest-of-the-world aggregate for years 1995 to
2018. The sample countries account for more than 90% of world exports and GDP. The
data on sectoral bilateral trade and input-output tables are from OECD Inter-Country
Input-Output (ICIO) Tables (OECD, 2021). It contains data for 45 different industries. I
aggregate them into three sectors: goods, producer services, and consumer services. The
rationale behind three sector aggregation is given in Section 1.4.3. For the measurement
of services trade flows, ICIO utilizes a wide array of data sources, such as national
accounts, balance-of-payment, and trade-in-services statistics, from IMF, OECD, and
WTO.36 Total employment of a country-year comes from Penn World Table.
For within-country price variation, I use gross-output price deflators (GOPD). They
are obtained from from OECD STAN, UN National Accounts, and National Statistics of
the Republic of China (Taiwan). For countries which only report value-added price de-
flators (VAPD), I construct GOPDs through applying the relationship between GOPD
and VAPD in the double deflation method. For countries without full sectoral details, I
split an aggregate sector into disaggregate ones using sectoral gross outputs of disaggre-
gated sectors as weights. For each imputation step, I confirm the validity by comparing
original price deflators and imputed price deflators for countries which provide original
price deflators. For example, for the countries that provide both GOPD and VAPD, I
impute GOPD from VAPD and compare the imputed GOPD with the actual GOPD.37
This mitigates measurement concerns for the countries that only report value-added
price deflators.
Cross-country sectoral gross-output price levels are from the Productivity Level
36In general, goods trade can be precisely measured given detailed product-level customs data; ser-
vices trade is more prone to measurement errors. See Francois and Hoekman (2010) for more details.
37This is a technical contribution of this paper. I show that my method of imputing GOPD corrects
the bias of using VAPD as proxies for GOPD, especially so for the country-sectors whose VAPD and
GOPD are very different. For example, Figure A.1 in Appendix A.1 shows the performance of the
method for the goods sector of Slovakia.
25
Database 2005 Benchmark of Groningen Growth and Development Centre (Inklaar and
Timmer, 2014). It provides price-level information for 42 countries and 35 industries.
The 35 industries are mapped into aggregate sectors of goods, producer services, and
consumer services. For 24 countries that are not in this dataset, the gross-output price
levels are imputed. In the data, sectoral price levels and income exhibit strong linear
relationship where countries with higher per-capita income have higher sectoral prices
(Figure A.2 in Appendix A.1), and I use this for the imputation. With the cross-country
price levels for a single year and within-country price deflators across years, I generate
full price sequences for all countries.
Appendix A.1 contains full details about data sources and construction, such as the
detailed procedure to construct sectoral prices from data.
1.4.2 Three Sectors: Goods, Producer Services, and Consumer Ser-
vices
In this section, I aggregate 45 industries in the OECD ICIO data into three sectors:
goods, producer services, and consumer services.38 I allocate each service industry to
producer services and consumer services based on how much of each industry’s output
is used as intermediate inputs. Service industries where more than 40% of their outputs
are used as intermediate inputs in 2018 are classified as producer services. Figure 1.1
shows that producer services are also the more tradable segment of services.
The rationale behind splitting the services sector into producer services and consumer
services is threefold. First, the prices of consumer services have measurement issues, since
many of them are non-market services as pointed out by Deaton and Heston (2010). This
is because with the absence of market prices, the prices are constructed using input mea-
sures. The aforementioned paper further claims that the measurement problems affect
macroeconomic aggregates and international price comparison. Since I am calibrating
the trade cost using prices, splitting services into the two types (highly-tradable producer
services whose prices are better measured and rarely-tradable consumer services whose
prices suffer from measurement issues) mitigates measurement concerns for trade costs
for services. Second, the split allows a distinction between the services that are more
affected by cross-sector specialization and those that are less affected. In data, because
38See Appendix A.1 for the list of the industries.
26
Figure 1.1: Tradedness and Intermediateness of Service Industries in 2018
Construction
Wholesale and retail & repair of motor
Land transport
Water transport
Air transport
Warehousing
Postal services
Accommodation & food
Publishing, audiovisual, broadcastingTelecom
IT
Financial & insurance
Real estate
Professional, scientific, technical
Admin
Public Education
Health
Art, entertainment, recreationOther services
Household
Goods
0.00
0.25
0.50
0.75
0.0 0.1 0.2 0.3 0.4
Total export / total gross output
To
ta
l i
nt
er
m
e
di
at
e 
us
ag
e 
/ t
ot
al
 g
ro
ss
 o
ut
pu
t
* Source: Author’s calculation from the OECD Inter-Country Input-Output Tables.
* Note: Red rectangles and blue circles denote industries belong to producer services and consumer
services, respectively. For full names of the industries, refer to Appendix A.1.
consumer services are rarely tradable, cross-sector specialization occurs mainly between
goods and producer services. Third, the trade barriers facing producer services and
consumer services are conceptually different. Consumer services trade consists mostly of
consumption abroad, unlike producer services trade. In 2018, direct purchases abroad
account for 6% and 78% of producer and consumer services trade, respectively. The
vast majority (94%) of trade in producer services is through other modes, such as cross-
border supply. Lastly, the defining characteristics of a sector are input-output linkages
and trade barriers in the model. Producer services are used more as intermediate inputs
and traded more.39
Appendix A.4.1 provides robustness analyses for two alternative definitions of sectors.
First, I classify all producer service industries and the two tradable consumer service
39Intermediate usage has important quantitative implications for the sensitivity of equilibrium prices
and allocations to the model primitives, since sectors whose outputs are more used as inputs have
amplification effects when their productivities or trade costs change. See Yi (2003), for example.
27
industries (“Accommodation and food” and “Art, entertainment, recreation”) as highly-
tradable services and the rest of consumer service industries as barely-tradable services.
Second, I explore the case of two-sector setup (goods and services). Under the two
alternative sectoral definitions, the main result of the paper still holds.
1.4.3 Data Patterns
International Trade
In this subsection, I discuss the patterns of international trade in goods, producer ser-
vices, and consumer services. Figure 1.2 shows the over-time changes in trade patterns.
The followings are the key data patterns.
Figure 1.2: Patterns of International Trade
CS
G
PS
0.05
0.10
0.15
0.20
0.25
1995 2000 2005 2010 2015
Year
To
ta
l e
xp
or
t /
 to
ta
l g
ro
ss
 o
ut
pu
t
(a) Total export/total gross output
CS
G
PS
0e+00
5e+06
1e+07
1995 2000 2005 2010 2015
Year
To
ta
l e
xp
or
t (m
illio
n c
urr
en
t U
SD
s)
(b) Total export
CS
G
PS
13
14
15
16
1995 2000 2005 2010 2015
Year
Lo
g(t
ota
l e
xp
or
t i
n 
m
illi
on
 c
ur
re
nt
 U
SD
s)
(c) Log(total export)
* Source: Author’s calculation from the OECD Inter-Country Input-Output Tables.
* Note: G, PS, and CS denote goods, producer services, and consumer services, respectively. Total
trade is measured as sum of exports of all economies in million current USDs.
Pattern 1.1. Goods and producer services are highly tradable, whereas consumer services
are rarely tradable.40
40Following Betts and Kehoe (2001), I use tradedness (the ratio of total world trade to world total
28
From Panel (a), in 2018, the percentage of total world export among total world
gross output for goods, producer services, and consumer services were 21.7%, 13.0%,
and 1.8%, respectively. Over time, there have been sizable changes in the tradability of
goods and producer services. In contrast, tradability of consumer services remains low.
Pattern 1.2. Trade in services has grown parallel to trade in goods.
Panel (c) plots logarithms of total world export by year. The volumes of both
producer services and consumer services trade grew parallel to goods trade. This is
suggestive evidence that there have been changes in trade costs in services that might
be comparable to goods.
Pattern 1.3. After 2008, there has been a slowdown in global trade, especially in goods.
Panel (a) shows that tradability of goods dropped sharply after 2008. The tradability
of producer services leveled off. For the total trade volume (Panels (b) and (c)), there
has been a fast growth from 2000 to 2008, but after 2008, the growth became slower.
Pattern 1.4. There are significant cross-sector specialization for individual countries.
Figure 1.3 shows goods share of export and import for the sample countries for
1995 and 2018. Unlike Panel (b), where countries’ goods share in imports are closer
to the world average, Panel (a) shows that goods share of exports are more dispersed.
This means that countries are specializing in their sector of comparative advantage.
Furthermore, the specialization patterns are changing over time. For example, from
1995 to 2018, goods share of US exports decreased from 60% to 50%, and that of China
increased from 71% to 84%.41 This pattern implies that globalization, i.e., a reduction
in trade costs for goods and services, can significantly affect the speed of structural
transformation of different countries by affecting their specialization patterns.
Structural Transformation
Figure 1.4 describes production shares in GDP and consumption shares of country-years
across their level of development (real GDP per capita at chained PPP in 2017 USD).
output) as a measure of tradability.
41Goods share of imports for both countries were stable across years at the range of 70% to 75%.
29
Figure 1.3: Share of Goods in Export and Import Volumes of Countries in 1995 and
2018
ARG
AUS
AUT
BEL
BGR
BRA
BRN
CAN
CHE
CHL
CHN
COL
CRI
CYP
CZE
DEU
DNK
ESP
EST
FIN
FRA
GBR
GRC
HKG
HRV
HUN
IDN
IND
IRL
ISLISR
ITA
JPN KAZKHM
KOR
LAO
LTU
LUX
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZL
PER
PHL
POL
PRT ROU
ROWRUS
SAU
SGP
SVK
SVN
SWE
THA
T N
TUR
TWN
USA
VNM
ZAF
TOT
ARG
AUS
AUT
BEL
BGR
BRABRN
CAN
CHE
CHL
CHN
COLRI
CYP
CZE
DEU
DNK
ESP
EST
FIN
FRA
GBR
GRC
HKG
HRV
HUN
IDNIND
IRL
ISL
IS
ITA
JPN
KAZ
KHM
KORLAO
LTU
LUX
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZL
PER
PHL POL
PRTROU
ROWUS
SAU
SGP
SVK
SVN
SWE
THA
TU
TUR
TWUSA
VNM
ZAF
TOT
(a) Goods fraction of total exports (b) Goods fraction of total imports
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
0.2
0.4
0.6
0.8
Fraction of goods (1995)
Fr
a
ct
io
n 
of
 g
oo
ds
 (2
01
8)
* Source: Author’s calculation from the OECD Inter-Country Input-Output Tables.
* Note: Vertical and horizontal dashed lines denote the goods fraction of world trade. Diagonal lines
represent 45-degree lines.
Pattern 1.5 (structural transformation). Country-years with higher per capita real in-
come exhibit higher value-added shares in producer and consumer services and lower
shares in goods.
Although each country has idiosyncratic patterns of changes in the production shares,
structural transformation is a general trend where a higher level of development implies
higher shares in both types of services.
Pattern 1.6. Country-years with higher per capita real income exhibit higher expenditure
shares in producer and consumer services and lower share in goods. Expenditure shares
change linearly; they do not level off as incomes grow.
This observation is the reason for adopting the assumption of non-homothetic CES
preferences. As Comin, Lashkari, and Mestieri (2021) pointed out, shares of final con-
sumption of services do not level off as incomes grow. Panel (b) of Figure 1.4 shows
30
Figure 1.4: Sectoral Shares of Production in GDP and Consumption in Final Expendi-
tures
(a) Production
Goods Producer services Consumer services
8 9 10 11 12 8 9 10 11 12 8 9 10 11 12
0.0
0.2
0.4
0.6
Log real GDP per capita (PPP, 2017US$)
Sh
ar
e 
of
 G
DP
(b) Consumption
Goods Producer services Consumer services
8 9 10 11 12 8 9 10 11 12 8 9 10 11 12
0.2
0.4
0.6
Log real GDP per capita (PPP, 2017US$)
Sh
ar
e 
of
 fi
na
l e
xp
en
di
tu
re
* Source: Author’s calculation from the OECD Inter-Country Input-Output Tables. Total real GDP is
expenditure-side real GDP at chained PPPs in 2017 USD from the variable “rgdpe” from PWT 10.0.
* Note: Each dot represents a country-year. There are total of 1584 country-years in each panel (66
countries from 1995 to 2018). The solid lines and shaded areas represent the regression line and 95%
confidence interval, respectively.
31
that their insight from agriculture-manufacturing-services setup holds in goods-producer
services-consumer services setting. Therefore, Stone-Geary preferences would not be an
adequate assumption.
1.5 Parameter Estimates
In the section, I estimate the key elasticities (income elasticities, substitution elastic-
ities for preferences, substitution elasticities for production) that govern how sensitive
production shares are to the model primitives (productivities and trade costs). I take
trade elasticities, which mainly govern how sensitively trade flows and prices respond
to changes in domestic and foreign productivities and bilateral trade costs, from the
literature.
1.5.1 Preference Parameters
Preference parameters (substitution elasticity, σ, and sectoral income elasticities {ϵk}k∈S)
govern the sensitivity of price and income effects in the model.42 To estimate them, I
follow the approach of CLM and use cross-country panel data on prices and sectoral
final expenditures.
Note that in non-homothetic CES preferences (Equation 1.3), any positive monotone
transformation of Ui leads to observationally equivalent utility functions. Therefore, for
the estimation, normalization of {ϵk}k∈S is required and I normalize ϵg = 1. From
Equation (1.4), expenditure on sector k ∈ {ps, cs}, relative to sector g can be expressed
as the follows: for k ∈ {ps, cs},
log
(
P ki,tC
k
i,t
P gi,tC
g
i,t
)
= (1− σ) log
(
P ki,t
P gi,t
)
+ (1− σ)(ϵk − 1) log(Ui,t) + log
(
ϕki
ϕgi
)
. (1.8)
Note that in the above equation, the utility of a household, Ui,t, is not a data observable.
Using the demand function (equation 1.4) for g, Ui,t can be expressed in terms of data
42Precisely speaking, the income elasticity is (1 − σ)(ϵk − 1), but for conciseness, I refer to ϵk as a
income elasticity.
32
observables:
log (Ui,t) =
1
1− σ log
(
1
ϕgi
)
+ log
(
Pi,tUi,t
P gi,t
)
+
1
1− σ log
(
P gi,tC
g
i,t
Pi,tUi,t
)
(1.9)
Note that Pi,tUi,t =Wi,t −NXi,t is an income of a household, which is an observable.
Adding error terms and merging Equations (1.8) and (1.9) yield the following esti-
mation equation: for k ∈ {ps, cs},
log
(
P ki,tC
k
i,t
P gi,tC
g
i,t
)
= (1− σ) log
(
P ki,t
P gi,t
)
+ (1− σ)(ϵk − 1) log
(
Pi,tUi,t
P gi,t
)
+ (ϵk − 1) log
(
P gi,tC
g
i,t
Pi,tUi,t
)
+ Ski + ν
k
i,t,
(1.10)
where Ski = (ϵ
k − 1) log
(
1
ϕgi
)
+ log
(
ϕki
ϕgi
)
.
I run a two-step GMM estimation jointly for the two estimating equations above
(one for k = ps and the other for k = cs) using a country-sector fixed effect term to
control for Ski . I use the data for 66 countries from 1995 to 2018 for the estimation. The
rest-of-the-world aggregate is dropped because its prices are imputed. Identification of
σ, ϵps, and ϵcs comes from within-country variations in prices, income, and sectoral final
expenditures.43 The identifying assumption is that the shocks to the relative demand
(νki,t) are uncorrelated with prices, incomes, and expenditures.
44 Table 1.1 shows the
estimation result.
43For this reason, cross-country price levels do not affect the result of the estimation, except for the
fixed effect terms.
44Due to this assumption, there can be endogeneity problems. However, the estimation result is
similar to the literature where studies utilized different datasets. For example, using the US country-
level data from 1977 to 2012, Cravino and Sotelo (2019) estimate the substitution elasticity as 0.58,
non-homotheticity parameter for skill-intensive services and non-skill intensive services as 1.82 and
1.42, respectively. Utilizing the 1992-2017 data from the Canadian provinces, Han, Miranda-Pinto, and
Tanaka (2022) find substitution elasticity to be 0.59 and non-homotheticity parameter for services to
be 1.59. CLM adopt agriculture-manufacturing-services structure and find that the outputs of three
broad sectors are complements and services are more income-elastic, from both US micro-level data
and cross-country data. Furthermore, CLM shows that the estimates from the micro-level data using
IV estimation and those from the macro-level data are similar, which further mitigates the endogeneity
concern of this estimation.
45I report bootstrapped confidence intervals instead of the ones calculated from standard errors of
GMM estimation. This is because the standard errors are high due to the collinearity issue of the
33
Table 1.1: Estimation Result for Preference Parameters
Estimate
(95% confidence interval)
σ
0.52
(0.43, 0.59)
ϵps
1.65
(1.53, 1.80)
ϵcs
1.41
(1.25, 1.54)
Observations 1584
* Note: The 95% bootstrapped confidence intervals
are computed with 1,000 runs of resampling with
replacement from the original sample at the same
sample size.45
Four remarks are in order for the estimation result. First, consistent with the struc-
tural transformation literature, I find that the sectoral outputs are complements and
that services are more income-elastic than goods. Second, I find that producer services
were more-income elastic than consumer services. This reinforces the point made by
Duarte and Restuccia (2020), who estimate income elasticities for different service sec-
tors. Third, the parameter restrictions on the non-homothetic CES preference are met,
that is, ϵk > 0 ∀k and σ ∈ (0, 1) ∪ (1,∞).46 Lastly, the utility weights {ϕki }k∈S are nor-
malized so that their sum equals one and are obtained from the estimated fixed effects
and non-homotheticity parameters.
1.5.2 Technology Parameters
In previous literature, it is common to assume the Cobb-Douglas production function
(e.g. Sposi, Yi, and Zhang, 2021 and Lewis et al., 2021). However, since input-output
linkages can amplify or dampen the three mechanisms of structural transformation (Sec-
tion 1.3.3), I explore whether inputs are substitutes or complements for production.
From Equations (F2) and (F3) and adding error terms, the input shares relative to
country-sector fixed effect and the country’s predictor variables. For example, in data, Brunei’s income,
prices, and sectoral expenditure shares are very stable across time, and this blows up the standard
errors.
46I imposed the strict positivity of ϵk’s by estimating log(ϵk). Estimating ϵk directly without imposing
strict positivity yielded similar point estimates of σ, ϵps, and ϵcs.
34
goods for sector k production are given by the following. For k ∈ {g, ps, cs},
log
(
P hi,tQ
hk
i,t
P gi,tQ
gk
i,t
)
= (1− ρk) log
(
P hi,t
P gi,t
)
+ log
(
ψhki
ψgki
)
+ χhki,t , h ∈ {ps, cs}, and (1.11)
log
(
Wi,tL
k
i,t
P gi,tQ
gk
i,t
)
= (1− ρk) log
(
Wi,t
P gi,t
)
+ log
(
ψLki
ψgki
)
+ χLki,t . (1.12)
For each sector k, I jointly estimate three equations above by running a two-step
GMM estimation. The result is provided in Table 1.2. The estimation result shows
that the production functions are close to the Cobb-Douglas case (ρk = 1).47 The
production weights {ψLki , {ψhki }h∈S}k∈S are normalized so that their sum equals one
and are obtained from the estimated fixed effects.
Table 1.2: Estimation Result for Production Parameters
Estimate (standard error)
ρg 1.21 (0.04)
ρps 0.91 (0.04)
ρcs 0.82 (0.04)
Observations 1584
* Note: Each line comes from a separate estimation.
1.5.3 Other Parameters
There are two sets of remaining parameters: trade elasticities θk’s and elasticities of
substitution of varieties ηk’s. Trade elasticities mainly govern how sensitive trade flows
and prices respond to changes in domestic and foreign productivities and bilateral trade
costs. I set θg = θps = θcs = 4. Trade elasticity of goods, θg, is the estimate of
Simonovska and Waugh (2014). I assume that trade elasticities of services are equal
to the elasticity for goods, because there is no consensus on the range of services trade
elasticities.48 For example, Muñoz (2022) finds that the services trade elasticity is 1.1
47For the estimation of sectoral production functions for the US, Cravino and Sotelo (2019) and
Atalay (2017) find that the elasticities of substitution across intermediate inputs are close to zero.
If I run the estimation using the US data only, I obtain a similar result; however, incorporating the
within-country variations of other countries yields the elasticities closer to one.
48For Melitz-type trade models, there are estimates from firm-level studies. See footnote 7 of Benz
and Jaax (2022) for more details.
35
from Europe’s cross-border job posting policies, and Sposi (2019) finds it to be 6.2 from
the gravity equation analysis using the price data from the International Comparison
Program. Appendix A.4.3 provides robustness check for the cases of θps = θcs = 2
and θps = θcs = 6. As for the elasticities of substitution among varieties, ηk, for all
ηk < θk+1, the equilibrium prices and allocations remain the same given the calibration
strategy.49
1.6 Calibration of Productivities and Trade Costs
To calibrate sectoral productivities, Aki,t, and bilateral trade costs, τ
k
ij,t, I take the
“wedges” approach of Święcki (2017), which has its root in business cycle accounting
by Chari, Kehoe, and McGrattan (2007).50 Specifically, I derive two structural equa-
tions from the model that relate productivities and trade costs to prices and import
shares. From the data on prices and import shares, I back out the productivities and
trade costs. The two equations are
Aki,t = γ
k
(
P ki,t
)−1
Rki,t
(
πkii,t
) 1
θk ∀i ∈ I, k ∈ {g, ps, cs}, and (1.13)
τkij,t =
(
πkij,t
πkii,t
)− 1
θk P kj,t
P ki,t
∀i, j ∈ I, k ∈ {g, ps, cs}. (1.14)
The two equations are from Equations (G1) and (G2).
Measurement of productivities comes from the gap between input and output prices
adjusting for Ricardian selection. Hypothetically, if a country does not import from any
other country πkii,t = 1, the input and output price gap directly implies productivities,
49The reason why I can choose any ηk < θs + 1 is as follows. Suppose that there two choices of ηk:
ηˆk and η˜k. Let hat and tilde denote corresponding model objects. In the model inversion for sectoral
productivity (Equation (1.13)),
Aˆki,t
A˜ki,t
= γˆ
k
γ˜k
. From Equations (G1) and (G2), this in turn implies that
Pˆ ki,t = P˜
k
i,t and πˆkij,t = π˜kij,t, and all the rest of equilibrium prices and allocations remain the same across
the two equilibria.
50An alternative approach is to derive sectoral total factor productivity (TFP) from data sources,
such as KLEMS datasets, and to derive trade costs from gravity equation estimation. This alternative,
however, has the following limitations. First, data sources to estimate sectoral TFPs are not available
for all 66 countries. Second, the estimation of trade costs through gravity equations does not allow for
origin-destination-sector specific trade costs. This mechanically prohibits the analysis of how changes
in trade costs affect the effective comparative advantage of countries.
36
since the gap reflects how efficiently the domestic variety firms can turn inputs to outputs.
In an open economy, where πkii,t < 1, the sectoral price P
k
i,t also reflects prices of imported
varieties. Therefore, the adjustment for the specialization, or Ricardian selection, should
be made to the output price.51
Measurement of trade costs comes from the import shares and prices. Given fixed
relative prices, if country j imports more from i, it means that the trade cost from j
to i must have been lower. Given constant import shares, if the relative price exporter
(country i) is lower, which roughly means that i is more efficient at producing the sectoral
goods, the trade cost must have been higher.
The two other model primitives are net transfer terms (NXi,t) and population (Li,t).
Since net exports in the data are nominal, I take the price of US value-added as a
numeraire. Therefore, real net transfer nxi,t =
NXi,t
Wus,t
, where NXi,t are the nominal total
net export in the data. This calibration strategy assumes that the country’s net transfers
relative to other countries remain constant across data, baseline, and counterfactual
economies. Li,t is taken from the variable “Number of persons engaged (in millions)”
from PWT 10.0.
1.6.1 Model Fit
Since the utility and production functions are estimated with error terms (Equations
(1.10) to (1.12)), the model does not fit the data perfectly, despite model primitives
being calibrated through structural equations from the model.52 To check the model
fit, I compare prices and trade shares between the baseline model and the data. This
is because I calibrated the model primitives (productivities and trade costs) from the
prices and trade shares in the data. As Figure 1.5 demonstrates, the baseline model shows
close-to-perfect fit with the data. It confirms that errors in the estimating equations for
the utility and production functions do not generate a misfit of a model.
Due to the calibration strategy, the prices and trade shares are indirectly targeted
moments. Appendix A.2.1 offers the model fit for non-targeted moments by checking
51See Finicelli, Pagano, and Sbracia (2013) for more discussions about the Ricardian selection effect.
52If I let utility weight terms (ϕki ) and production weight terms (ψhki and ψLki ) be time-varying
to account for the error terms, then the calibrated model fits the data perfectly. One downside of this
alternative approach is that since production functions vary over time, the productivity sequences are not
interpretable. The result of this paper is robust to this alternative model specification (Appendix A.4.4).
37
Figure 1.5: Model Fit: Prices and Trade Shares
(a) Prices (b) Trade shares
* Note: Each dot represents a country-year. For prices each year, the price of US value added (labor)
was used as a numeraire.
the model’s performance in explaining the structural transformation pattern of the US
under a different calibration strategy.
1.6.2 Patterns of Changes in Trade Costs
Since consumer services’ tradability is very low (?? 1.1), which implies prohibitively high
trade costs, the quantitative results of the paper do not significantly depend on changes
in trade cost for consumer services.53 Therefore, for brevity, I show the results for the
goods and producer services only.
Figure 1.6 shows the two patterns of trade costs. First, in both 1995 and 2018 and
for both goods and producer services, there exist trade cost asymmetry. Developing
countries tend to face higher export trade costs than advanced countries. (Definition
of developing and advanced economy follows IMF, 2022.) Developing countries’ export
trade costs are higher when they are exporting to advanced economies than to developing
economies. Second, in both sectors, the asymmetry is decreasing over time. Whereas
53See Appendix A.4.2 for details.
38
Figure 1.6: Box Plots for Trade Costs in 1995 and 2018 by Country Group Pairs
(a) Goods (b) Producer services
ADV−ADV ADV−DEV DEV−ADV DEV−DEV ADV−ADV ADV−DEV DEV−ADV DEV−DEV
0
5
10
15
20
25
Origin−destination
Tr
a
de
 c
os
ts
Year
1995
2018
* Note: ADV-DEV denote trade costs from advanced to developing economies. Other country group
pairs (ADV-ADV, DEV-ADV, DEV-DEV) are defined in a similar fashion . The top of the box plots
are truncated for illustration purposes, since maximum trade costs are very high.
the distributions of the advanced countries’ export trade costs remain stable, those
for developing countries were shifting towards zero. Decreasing trade costs were more
pronounced for the trade costs from developing to advanced economies. Figure 1.7 shows
the median trade costs for the country group pairs for all sample years.
Three notes about the calibrated trade costs are as follows. First, regarding the
underlying reasons of the trade cost asymmetry, there is ample evidence for goods sector.
See Kee, Nicita, and Olarreaga (2009) for an example of policy barriers; Limao and
Venables (2001); Mesquita-Moreira, Volpe, and Blyde (2008); and Blum et al. (2019)
for transportation or inventory costs; and Merga (2021) for financial constraints on
fixed costs of trade. Second, the iceberg trade costs are high, which is in line with
the previous literature. See Anderson and Van Wincoop (2004) for the discussion on
high trade costs in the gravity literature. Apart from trade barriers discussed by the
aforementioned paper, such as policy barriers, information costs, and costs from using
different currencies, it is important to note that the trade costs are the average costs for
39
Figure 1.7: Median Trade Costs by Country Group Pairs from 1995 to 2018
ADV−DEV
ADV−ADV
DEV−DEV
DEV−ADV
ADV−DEV
ADV−ADV
DEV−DEV
DEV−ADV
(a) Goods (b) Producer services
1995 2000 2005 2010 2015 1995 2000 2005 2010 2015
5
10
Year
M
ed
ia
n 
tra
de
 c
os
t
* Note: In panel (a), Q1, Q2, and Q3 denote first quartile, median, and third quartile of all bilateral
trade respectively. Mean denotes trade-weighted means.
all varieties. Some varieties that face high trade costs affect the average measure. For
the real-world example of high trade cost varieties, China does not allow US subscription
video-on-demand services operators, such as Netflix, to operate in its domestic market
(USITC, 2022). Another example is a license requirement. Many countries do not allow
lawyers from a different country to practice in their countries. For goods, there are
sanitary or phytosanitary measures or technical regulations that block imports of certain
goods. The last notable pattern is the increase of median trade costs from advanced to
developing economies. This partly reflects the slowdown of global trade after the 2008
Financial Crisis (?? 1.3).
Both Figure 1.6 and Figure 1.7 are based on the distribution of trade costs for
country-group pairs across years. To show the decreasing asymmetry for each country,
in Figure 1.8, I plot each country in terms of their log GDP per capita in 1995 and
the growth rate of export trade relative to import cost for each sector. Export and
import trade costs are measured as trade-flow-weighted average of trade costs across
40
all trading partners of a country. Trade flows are from year 1995 to control for the
endogenous response of trade flows to changes in trade costs. For the rest of the paper,
for calculating each country’s export and import trade costs, I use the same method. The
general pattern in Figure 1.8 is that the poorer countries relative export cost decreased
faster than the richer countries.
Figure 1.8: Growth Rate of Relative Export Trade Costs and GDP per Capita
ARG
AUS
AUT
BEL
BGR
BRA
BRN
CAN
CHE
CHL
CHN
COL
CRI
CYP
CZE
DEU
DNKESP
EST
FIN
FRA
GBR
GRC
HKG
HRV
HUN
IDNIND IRL
ISL
ISR
ITA
JPN
KAZ
KHM
KOR
LAO
LTU
LUX
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZLPER
PHL
POL
PRT
ROU
ROW
RUS
SAU
SGP
SVK
SVN
SWE
THA
TUN
TUR
TWN
USA
VNM
ZAF
ARG
AUS
AUT
BEL
BGR
BRA
BRN
CAN
CHE
CHL
CHN
COL
CRI
CYP
CZE
DEU
DNK
ESP
EST
FIN
FRA
GBR
GRC
HKG
HRV
HUN
IDN
IND
IRL
ISL
ISR
ITA
JPN
KAZ
KHM
KOR
LAO
LTU
LUX
LVA
MAR
MEX
MLT
MMR
MYS
NLDNOR
NZL
PERPHL POL
PRT
ROU
ROW
RUS
SAU
SGP
SVK
SVN
SWE
THA
TUN
TUR
TWN
USA
VNM
ZAF
(a) Goods (b) Producer services
6 8 10 6 8 10
−2
−1
0
1
Log(GDP per capita) in 1995
G
ro
w
th
 ra
te
 o
f (e
xp
or
t t
ra
de
 c
os
ts
)/(
im
po
rt t
ra
de
 c
os
ts
)
Trade Cost Asymmetry and Comparative Advantage
In the previous section, I described the decreasing asymmetry of trade costs in both goods
and producer services, which means that it has become easier for developing countries
to export both goods and producer services relative to importing them. This pattern
matters for one key determining factor of international trade, comparative advantage of
countries.
Building on the definition of effective comparative advantage by Deardorff (2014), I
define the strength of country i’s effective comparative advantage in goods relative to
the rest of the world, denoted by r, in terms of relative productivities and bilateral trade
41
costs as the following.54
Definition 1.4. Country i’s comparative advantage in goods relative to the rest of the
world at time t is defined as
CAgc,t ≡
(
Agc,t
Apsc,t
/
A˜gr,t
A˜psr,t
)(
τ˜ gcr,t
τ˜ grc,t
/
τ˜pscr,t
τ˜psrc,t
)− 1
2
,
A˜gr,t is GDP-weighted productivity of r and τ˜
g
cr,t is trade-flow-weighted export trade costs
from c to r. The weights are from 1995, the initial year to control for the endogenous
response of output and trade to changes in productivities and trade costs. Other terms
are defined in a symmetric manner.
Under this measure, ceteris paribus, if a country’s goods export trade costs decline
faster than its goods import trade costs, the country’s comparative advantage in goods
strengthens. If the country’s export trade costs relative to import trade costs in producer
services declines, its comparative advantage in goods weakens. This is one reason why
understanding the relative changes in trade costs across goods and producer services
is important. They affect the comparative advantage, hence countries’ specialization
patterns.
Although globalization can affect structural transformation through comparative ad-
vantage. In theory, it can alter the speed of structural transformation even without any
changes in the measure of comparative advantage. For example, a uniform trade cost
reduction across sectors and country pairs does not change comparative advantage of
any countries, but it increases real income of households, and through the income effect,
it can affect countries’ structural transformation. Another example is a hypothetical
scenario where only goods trade costs decline uniformly across all country pairs. In this
54See Appendix A.2.2 for the definition of the effective comparative advantage by Deardorff (2014)
and its theoretical meaning. They define the effective comparative advantage in an Armington model.
In their definition based on an inequality, one can compare two countries and determine which country
has effective comparative advantage in which sector. I modify their definition so that it can measure
the strength of the comparative advantage in a Ricardian model. For conciseness, I use the terms
“effective comparative advantage” and “comparative advantage” interchangeably. I describe changes in
comparative advantage in terms of goods. For instance, the statement that its comparative advantage
in goods weakens means either that its comparative advantage in goods weakens or that its comparative
advantage in producer services strengthens. Since consumer services are barely tradable, my definition
does not include productivity or trade cost terms related to the sector.
42
case, although it does not incur any changes in comparative advantage, it will lower goods
price relative to producer services, and through the price effect, it can impact structural
transformation patterns. Therefore, how globalization affects structural transformation
through shifts in comparative advantage is a quantitative question, which I explore in
the following counterfactual exercises.
1.6.3 Patterns of Productivities
Since the main focus of this paper is on globalization, I discuss the sectoral productivity
growth and its implications on structural transformation in Appendix A.2.3. To briefly
discuss the main finding, I show that productivity growth in goods and producer ser-
vices was faster than that of consumer services, when measured in labor productivity.
This differential productivity growth is significant in understanding the price effect of
structural transformation. My finding resonates with the studies that find differential
productivity growth among services industries (e.g., Jorgenson and Timmer, 2011; and
Duernecker, Herrendorf, and Valentinyi, 2017).
1.7 Counterfactuals
With the estimated parameters and calibrated primitives, I run three counterfactual
exercises to quantify the effects of trade cost changes in goods and services on the
structural transformation of 66 countries. For all three exercises, I let productivities
(Aki,t’s), population (Li,t’s), and net transfers (nxi,t’s) evolve following their calibrated
levels. Let τkij,t and τ
k
ij,t denote trade costs in the counterfactual economy and in the
calibrated levels. The three counterfactuals are as follows: for t = 1995, ...2018,
• Counterfactual #1 (changes in only goods trade costs): τ gij,t = τ
g
ij,t, τ
ps
ij,t = τ
ps
ij,1995,
τ csij,t = τ
cs
ij,1995
• Counterfactual #2 (changes in only services trade costs): τ gij,t = τ
g
ij,1995, τ
ps
ij,t = τ
ps
ij,t,
τ csij,t = τ
cs
ij,t
• Counterfactual #3 (no change in trade costs for both goods and services): τ gij,t =
τ gij,1995, τ
ps
ij,t = τ
ps
ij,1995, τ
cs
ij,1995 = τ
cs
ij,1995
43
I compare how production shares in value added evolve from 1995 to 2018 in the coun-
terfactual economies with the baseline economy. In the comparison between the baseline
and counterfactual #1, I assess the effect of changes in the trade costs for services on
structural transformation. Likewise, by contrasting the baseline and counterfactual #2,
I quantify the impact of changes in the trade costs for goods. Lastly, through the dif-
ference between the baseline and counterfactual #3, I evaluate the effect of changes in
both goods and services trade costs on structural transformation.
To understand how globalization interacts with structural transformation in details,
I first study China, the largest developing countries. Then, I proceed to study all 66
countries in my data.
1.7.1 Globalization and Structural Transformation of China
Figure 1.9 describes the globalization and structural transformation patterns of China.
Regarding the main inputs to the counterfactual exercises (trade costs), I show how
China’s trade costs for goods and producer services changed in Panel (a) of Figure 1.9.55
The weighted average of export and import trade costs, where weights are trade flows
in the initial year, 1995, exhibits the following patterns. For both goods and producer
services, China’s export cost relative to import cost decreased. Furthermore, the over-
time changes are proportional across sectors: sectoral relative export costs decrease in
an almost parallel fashion as evidenced by Panel (a)-(ii).
Panel (b) compares the evolution of production shares from 1995 to 2018 for China
in the baseline and the three counterfactual exercises. Comparing the baseline and coun-
terfactual #1, we can see that changes in services trade costs had a strong effect on the
speed of China’s structural transformation. In the baseline, China’s GDP share of goods
decreased from 57% to 44% over 23 years. In contrast, in counterfactual #1, the goods
share in 2018 is 56%, approximately equal to the level of 1995. This means that if only
goods trade costs had changed, China would not have gone through structural trans-
formation at all. This possibly reflects the fact that China’s changes in services trade
costs weakened China’s comparative advantage in goods. Another noticeable pattern is
55Because consumer services trade costs are prohibitively high, which reflects low tradability of
consumer services, changes in consumer services trade costs have negligible effects on the quantitative
results (Appendix A.4.2). For brevity, I omit trade cost statistics for consumer services.
44
Figure 1.9: China’s Trade Costs and Structural Transformation
(a) Trade cost patterns
Import cost (PS)Import cost (G)
Export cost (G)
Export cost (PS)
5
10
1995 2000 2005 2010 2015
Year
Tr
a
de
 c
os
ts
(i) Trade costs (weighted means)
G
PS
0.0
0.5
1.0
1.5
2.0
1995 2000 2005 2010 2015
Year
Lo
g(e
xp
or
t c
os
t/i
m
po
rt 
co
st
)
(ii) Log(export cost/import cost)
(b) Structural transformation in the baseline and counterfactuals #1 to #3
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Pr
od
uc
tio
n 
sh
ar
es
Baseline (G, PS, CS)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 1 (G)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 2 (PS & CS)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 3 (none)
Year
* Note: Trade costs reported in Panel (a) are weighted averages of trade costs across China’s trade
partners. The weights are trade flows in 1995.
45
that the changes in trade costs for services mainly induce production share differences in
goods and producer services, the two highly-tradable sectors. Consumer services, which
are rarely tradable, are affected little.
In contrast, changes in the costs for goods trade had the opposite effect. Contrasting
the baseline and counterfactual #2, changes in the trade costs for goods decelerated
structural transformation of China. In the counterfactual, the goods share declines to
34% by 2018. This is partly due to the decline of the cost of exporting goods relative to
importing them strengthened China’s comparative advantage in goods. The changes in
trade costs for goods mostly affect production shares of the two highly-tradable sectors,
just like the case for the changes in services trade costs.
The resemblance of counterfactual #3 to the baseline shows that the net effect of the
reduction in trade costs in all sectors on structural transformation is minimal. With-
out any changes in goods and services trade costs, the goods share decreases to 44%,
approximately equal to the baseline. The interpretation of this result is as follows.
China’s services export cost reduction relative to import cost has been commensurate
with that of goods. This proportionality implies that China’s comparative advantage
did not change much. This, in turn, quantitatively implies that changes in the trade
costs played a minor role in the structural transformation of China.56
For China, proportional trade cost changes across goods and producer services, which
induced a small change in their comparative advantage, had little effect on their patterns
of structural transformation. However, it is theoretically possible that proportional
changes can have a large impact on a country’s structural transformation through the
income or price effect, as discussed in Section 1.6.2. In the next section, using the
cross-country variation in the quantitative results for 66 sample countries, I show that
globalization mainly affected structural transformation through comparative advantage.
56Given the proportional changes in trade costs, then, why are China becoming a larger net ex-
porter of goods as evidenced by Figure 1.3? It is because sectoral productivities also shift a country’s
comparative advantage and its production patterns.
46
1.7.2 Globalization and Structural Transformation of 66 Countries
First, by comparing the baseline and counterfactual #1, I show that countries whose
producer services export trade costs relative to import trade costs decreased experi-
enced slower structural transformation from goods to producer services in the baseline.
In contrast, those with increased relative export trade costs in producer services went
through faster structural transformation from goods to producer services in the baseline.
Furthermore, the speed of structural transformation to consumer services did not get
affected much for an either case.
To make this point in Figure 1.10, for each country, I measure the growth rate of
export trade costs relative to import trade costs for producer services in log points,
which I plot on x-axis. A negative growth rate for a country means that globalization
in producer services weakened the country’s comparative advantage in goods, which was
defined in Definition 1.4. A positive index means the opposite. On y-axis, I measure
the impact of globalization in services on structural transformation by subtracting the
changes in production shares from 1995 to 2018 in the baseline economy from those in
counterfactual #1. For instance, from 1995 to 2018, if a country’s GDP shares in goods,
producer, and consumer services change by -5 p.p., +4 p.p., and +1 p.p. in the baseline
economy, and by -2 p.p., +1 p.p., and +1 p.p. in the counterfactual economy. I measure
the impact of globalization with a triple of +3 p.p., -3 p.p., 0 p.p.57
Figure 1.10 shows that for countries whose comparative advantage in goods got weak-
ened by globalization in producer services, structural transformation from goods to pro-
ducer services was accelerated. The opposite holds for the countries with strengthened
comparative advantage in goods. Structural transformation to consumer services was af-
fected little, just like the case of China. In conjunction with the decreasing asymmetry in
producer services outlined in Figure 1.8, the result means that globalization in producer
services generally accelerated structural transformation from goods to producer services
of the poorer countries, and decelerated that of the richer countries (Figure 1.11).
57An alternative measure would be to calculate how much of the structural transformation pattern
is explained by globalization. However, this incurs a comparison between small numbers for advanced
economies that had already undergone significant structural transformation by 1995. For example, the
goods share of GDP for the US was 21% in 1995 and decreased by 3.7 p.p. from 1995 to 2018. In
counterfactual #1, the share falled by 4.2 p.p. Therefore, although the absolute effect is small at -0.5
p.p. (-4.2 p.p. - (-3.7 p.p.)), if I measure the relative effect, it is large (-0.5 p.p./-3.7 p.p.). For this
reason, I use the absolute measure to discuss the quantitative results.
47
Figure 1.10: Globalization in Producer Services and Structural Transformation
ARG
AUS
AUT
BEL
BGR
BRA
CAN
CHECHL
CHN
COL
CRI
CYP
CZE
DEU
DNK
ESP
EST
FI FRAGBRGRC HKG
HRVHUN
IDN
IND
IRL
ISLISR
ITA
JPN
KAZ
KHM
KOR
LAO
LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLDNOR
NZL
PERPHLPOL
RT
ROU
ROW
RUS
GP
SVK
SVN
SWE
THA
TUN
UR
TWN
USA
VNM
ZAF
∆(τps) strengthens CAg∆(τps) weakens CAg
∆(τps)
decelerates
str. trans.
from G
∆(τps)
accelerates
str. trans.
from G
−40
−20
0
20
40
−2 −1 0 1 2
Growth rate of relative export trade costs (goods)
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG
AUS
AUT
BEL
BGR
BRA
CAN
CHECHL
CHN
COL
CRI
CYPCZE
DEUDNK
ESP
EST
FI FRAGBR
GRC
HKGHRV
HUN
IDN
IND
IRL
ISL
ISR
ITA
JPN
KAZ
KHM
KOR
LAO
LTU
LVA
MAR
MEXMLT MMRMYS
NLDNOR
NZLPER
PHLO
PRT
ROU
ROW
RUS
SGP
SVK
SVN
SWE
T A
TUN
TUR
TWN
USA
VNM
ZAF
∆(τps) strengthens CAg∆(τps) weakens CAg
∆(τps)
accelerates
str. trans.
to PS
∆(τps)
decelerates
str. trans.
to PS
−40
−20
0
20
40
−2 −1 0 1 2
Growth rate of relative export trade costs (goods)
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARGAUS AUTBEL
BGR BRACAN CHECHL
CHN COL
CRI
CYP
ZE
DEUDNKESP
EST
FIN FRAGBRGRC HKG
HRV
HUN IDNINDIRL
ISL
ISR ITA JPNKAZ KHM KORLAO
LTU LVA
MARMEX
MLT
MR
MYS
NLDNORNZLPERPHLPOL
PRT
ROU
ROW
RUS
SGP
SVK
SVN SWETHA
TUN
UR TWN
USA
VNM
Z
∆(τps) strengthens CAg∆(τps) weakens CAg
∆(τps)
accelerates
str. trans.
to CS
∆(τps)
decelerates
str. trans.
to CS
−40
−20
0
20
40
−2 −1 0 1 2
Growth rate of relative export trade costs (goods)
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #1) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
48
Figure 1.11: Income Levels and Impact of Producer Services Globalization
ARG
AUS
AUT
BEL
BGR
BRA
CAN
CHECHL
CHN
COL
CRI
CYP
CZE
DEU
DNK
ESP
EST
FINRAGBRGRC HKG
HRVHUN
IDN
IND
IRL
ISLISR
ITA
JPN
KAZ
KHM
KOR
LAO
LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLNORNZLPERPHL POL
PRT
ROU
ROW
RUS
GP
SVK
SVN
SWE
THA
TUN
TUR
TWN
USA
VNM
ZAF
∆(τps)
decelerates
str. trans.
from G
∆(τps)
accelerates
str. trans.
from G
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG
AUS
AUT
BEL
BGR
BRA
CAN
CHECHL
CHN
COL
CRI CYPCZE
DEUDNK
ESP
EST
FINRAGBR
GRC
HKGHRV
HUN
IDN
IND
IRL
ISL
ISR ITA
JPN
KAZ
KHM
KOR
LAO
LTU
LVA
MAR
MEX MLTMMR YS NLNOR
NZLPERPHL POL
PRT
ROU
ROW
RUS SGP
SVK
SVN
SW
THA
TUN
TUR
TWN
USA
VNM
ZAF
∆(τps)
accelerates
str. trans.
to PS
∆(τps)
decelerates
str. trans.
to PS
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS AUTBELBGR BRA CAN CHECHL
CHN COL
CRI
CYP
CZE
DEUDNKESP
EST
FINRAGBRGRC HKG
HRV
HUNIDNIND IRL
ISL
ISR ITA JPNKAZKHM KORLAO
LTU LVA
MAR MEX
MLT
MMR
MYS
NLNORNZLPERPHL POL
PRT
ROU
ROW
RUS
SGP
SVK SVN SWETHA
TUN
T R TWN
U AVNM ZAF
∆(τps)
accelerates
str. trans.
to CS
∆(τps)
decelerates
str. trans.
to CS
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #1) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
49
Symmetric argument holds for goods globalization. As for the globalization in goods,
I compare the baseline economy and counterfactual #2 in the same manner. Figure 1.12
demonstrates that changes in goods trade costs affect structural transformation as pre-
dicted by their impact on comparative advantage. Figure 1.13 shows that goods glob-
alization decelerated structural transformation from goods to producer services of the
poorer countries and accelerated that of the richer countries.
To assess the impact of changes in trade costs for both goods and services on struc-
tural transformation, I first construct a measure of how globalization affects countries’
comparative advantage, defined in Definition 1.4, and call the measure the proportion-
ality index.
Definition 1.5. The proportionality index for the changes in trade costs from t1 to t2
for country i is defined as
(Proportionality Index)i,t1,t2 = ∆t log
(
τ˜ gir,t
τ˜ gri,t
)
−∆t log
(
τ˜psir,t
τ˜psri,t
)
,
where ∆t(zt) ≡ zt2 − zt1 , τ˜kir,t and τ˜kri,t denote weighted mean of the country’s export
trade costs and import trade costs in sector k for trade with the all the other countries,
denoted by r. The weights are trade flows at t1 to control for the endogenous response
of trade flows to trade cost changes.
The index measures how trade cost changes affect a country’s comparative advantage in
goods. It abstracts from consumer services, which are rarely tradable.58 If the index is
negative, it means that overall changes in trade costs strengthened the country’s effective
comparative advantage in goods. A positive index implies the weakening of the country’s
comparative advantage in goods.
Figure 1.14 describes the relationship between the sample countries’ proportionality
indices for changes in trade costs from 1995 to 2018 and the measured impact of trade cost
changes on the speed of their structural transformation. I find that the closer the index
of a country is to zero, the closer the gap between the speed of structural transformation
between the counterfactual and the baseline economy. For countries whose changes in
trade costs were proportional (e.g., China and India, whose indices were close to zero),
58See Appendix A.4.2 for the robustness check on the role of consumer services trade costs.
50
Figure 1.12: Globalization in Goods and Structural Transformation
ARG
AUS
AUT
BEL
BGR
BRACAN
CHE
CHL
CHN
COL
CRI
CYP
CZE
DEU
DNKSP
EST
FIN
FRAGBR
GRC
HKG
HRV
HUNIDNIND
IRL
ISLIS ITA
JPN
KAZ
KHM
KOR
LAO
LTULVA
MAR
MEX
MLT
MMRMYS NLD
NOR
NZL
PER
PHL
POL
PRT
ROU
ROW
RUS
SGP
SVK
SVN
SWE
THA
TUNT R
TWN
USA
VNM
ZAF
∆(τg) weakens CAg∆(τg) strengthens CAg
∆(τg)
decelerates
str. trans.
from G
∆(τg)
accelerates
str. trans.
from G
−40
−20
0
20
40
−2 −1 0 1 2
Growth rate of relative export trade costs (goods)
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG
AUS
AUT
BEL
BGR
BRACAN
CHE
CHL
CHN
COL
CRI
CYP
CZE
DEU
DNKSP
EST
FIN
FRAGBR
GRC HKGHRV
HUNIDNIND
IRL
ISLISR ITA
JPN
KAZ
KHM
KOR
L O
LTU
LVA
MAR
MEX MLT
MMRMYS NLD
NOR
NZLPE
PHL
POL
PRT
ROU
ROW
RUS
SGP
SVK
SVN
SWE
THA
TUNT R
TWN
USA
VNM
ZAF
∆(τg) weakens CAg∆(τg) strengthens CAg
∆(τg)
accelerates
str. trans.
to PS
∆(τg)
decelerates
str. trans.
to PS
−40
−20
0
20
40
−2 −1 0 1 2
Growth rate of relative export trade costs (goods)
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARGAUS AUTBEL
BGR
BRACAN CHECHLCHN COLCRI
CYP
CZE
DEUDNKSP
EST
FINFRAGBRGRC
HKG
HRV HUIDNINDIRL ISLISR ITA
JPNKAZ KHM
KOR
LAOLTULVA
MAR
MEX
MLTMMRMYS NLDN R NZLPER
PHLPOL PRTROU ROWRUS
SGP
SV SVN
SWE
T A
TUNT R TWNU A
VNM
Z F
∆(τg) weakens CAg∆(τg) strengthens CAg
∆(τg)
accelerates
str. trans.
to CS
∆(τg)
decelerates
str. trans.
to CS
−40
−20
0
20
40
−2 −1 0 1 2
Growth rate of relative export trade costs (goods)
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #2) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
51
Figure 1.13: Income Levels and the Impact of Goods Globalization
ARG
AUS
AUT
BEL
BGR
BRA CAN
CHE
CHL
CHN
COL
CRI
CYP
CZE
DEU
DNKESP
EST
FIN
FRAGBR
GRC
HKG
HRVHUN
IDNIND
IRL
ISLISR ITA
JPN
KAZ
KHM
KOR
LAO
LTU LVA
MAR
MEX
MLT
MMR MYS NL
NOR
NZLPER
PHL
POL
PRT
ROU
ROW
RUS
SGP
SVK
SVN
SWE
THA
TUN TUR
TWN
USA
VNM
ZAF
∆(τg)
decelerates
str. trans.
from G
∆(τg)
accelerates
str. trans.
from G
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG
AUS
AUTBEL
BGR
BRA CAN
CHE
CHL
CHN
COL
CRI
CYP
CZE
DEU
DNKESP
EST
FIN
FRAGBRGRC HKGHRV
HUNIDNIND
IRL
ISLISR ITA
JPN
KAZ
KHM
KOR
LAO
LTU LVA
MAR
MEX MLT
MMR MYS NL
NOR
NZLPER
PHL
POL
PRT
ROU
ROW
RUS
SGP
SVK
SVN
SWE
THA
TUN TUR
TWN
USA
VNM
ZAF
∆(τg)
accelerates
str. trans.
to PS
∆(τg)
decelerates
str. trans.
to PS
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS AUTBELBGR BRA CAN
CHECHLCHN COL CRI
CYP
CZE DEUDNKESPEST
FINRAGBRGRC
HKG
HRVHUNIDNIND IRLI LISR ITA JPNKAZKHM KORLAO LTU LVA
MAR
MEX MLTMMR YS NLNORNZLPER
PHL POL PRROUROW RUS SGPSVK SVN
SWTHA
TUN T R WN USA
VNM
ZAF
∆(τg)
accelerates
str. trans.
to CS
∆(τg)
decelerates
str. trans.
to CS
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #2) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
52
Figure 1.14: Globalization in Both Goods and Producer Services and Structural Trans-
formation
ARG AUS
AUT BEL
BGR BRA
CAN
CHE
CHL
CHNCOL
CRI
CYPCZE
DEUDNK
ESP
EST
FIN
FRAGBR
C
HKG
HRV HUN
IDN
IND
IRL
ISL ISR
ITAJPN
KAZ
KHM
KORLAO
LTU
LVA
MAR
MEX
MLT
MMR
MYS
N D
NOR
NZL PER
PHL
POLPRT
ROU
ROWRUS
SGP
SVK
SVN
SWE
THA
TUN
TUR TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS AUT
BELBGR BRA
CAN
CHE
CHL
CHNCOL
CRI
CYP
CZE
DEUDNK
ESP
EST
FIN
FRAGBRC HKGHRV
HUN
IDN
IND
IRL
ISL
ISR
ITAJPN
KAZ
KHM
KORL O
LTU
LVA
MAR
MEX
MLT
MM
MYS NLD
NOR
NZL PER
PHL
POL
PRT
ROU
ROW
RUS
SGP SVK
SVN
SWE
THA
TUN
TU
TWN
USA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to PS
∆(τ)
decelerates
str. trans.
to PS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS
AUT
BEL
BGR
BRA CAN
CHECHL CHNCO
CRI
CYP
CZE
DEUDNK ESP EST
FINFRAGBR
C
HKG
HRV
HUN
IDN IND
IRL
ISL
ISRITAJPN
KAZ KHM
KORLAO LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR NZL PER PHL
POL
PRT ROU
ROW
RUS
SGP
SVKSVNSWETHA
TUN
TUR
TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to CS
∆(τ)
decelerates
str. trans.
to CS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
53
globalization had a limited impact on their structural transformation. Furthermore, I
find a strong linear relationship between the index and the effect of trade cost changes on
structural transformation from goods to producer services. For countries with negative
indices (e.g., Vietnam), trade cost changes slowed down their structural transformation
from goods to producer services: the decline in goods share of GDP and the increase of
producer services share of GDP were slower in the baseline economy. The opposite holds
for countries with positive indices (e.g., Lithuania). These effects are more pronounced
for countries with indices further from zero, showing that trade played a major role in the
structural transformation from goods to producer services of countries that experienced
disproportionate changes in trade costs across goods and producer services.59
This cross-country result implies that despite the model having multiple channels
(income effect, price effect, and cross-sector specialization effect along with input-output
linkages), a key summary statistic that explains the impact of changes in the trade costs
on the speed of structural transformation is the cross-sector proportionality index.
Unlike the analyses of goods globalization and services globalization in isolation,
globalization in both goods and services did not exhibit any trend with regards to the
income levels of countries (Figure 1.15). Despite a reduction in export trade costs relative
to import trade costs showing strong correlation with the income level of countries
in both goods and producer services, the proportionality index which captures how
closely goods and services trade costs co-moved, did not show any trends with regards
to income level, and hence there was no clear trend of how globalization differentially
affects developing and advanced countries.
Furthermore, globalization mostly affect production shares of goods and producer
services, the two highly-tradable sectors. Table 1.3 reports summary statistics for abso-
lute effect comparing the baseline and counterfactual #3 (y-axes of Figure 1.14). The
table shows that consumer services, the rarely-tradable sector, is the least affected by
the changes in trade costs.60
Until now, I have explained the result in terms of the developing countries. For
59See Appendix A.5 for the details on the globalization and structural transformation patterns of
Vietnam, Lithuania, and India. The section also includes the results for all 66 countries across the four
equilibria (baseline and counterfactuals #1 to #3.)
60Under an alternative sectoral specification of goods, highly-tradable services, and barely-tradable
services, the point that globalization mainly affects structural transformation among tradable sectors is
even stronger (Appendix A.4.1).
54
Figure 1.15: Income Levels and the Impact of Goods and Producer Services Globalization
ARG AUS AUTBELBGR BRA
CAN CHE
CHL
CHN COL
CRI CYP
CZE DEUDNKESP
EST FIN
FRAGBRGRC HKGHRVHUN
IDN
IND
IRL
ISLISR ITA JPNKAZKHM KOR
LAO
LTU
LVA
MAR
MEX
MLT
MMR
YS
NL
NOR
NZLPER
PHL
POL PRT
ROU
ROW RUS
SGPSVK
SVN SWE
THA
TUN
TUR WN USA
VNM
ZAF
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS AUTBELBGR BRA
CAN
CHE
CHL
CHN COL CRI
CYP
CZE DEUDNKESP
EST FIN
FRAGBRGRC HKGHRVHUN
IDN
IND IRL
ISL
ISR ITA JPN
KAZ
KHM KORLAO
LTU
LVA
MAR
MEX MLT
MMR MYS NL
NOR
NZLPER
PHL
POL PRT
ROU
ROW RUS
SGPSVK
SVN SWE
THA TUN
TUR TWN USA
VNM
ZAF
∆(τ)
accelerates
str. trans.
to PS
∆(τ)
decelerates
str. trans.
to PS
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS AUTBELBGR BRA CAN CHECHLCHN COL
CRI
CYP
CZE
DEUDNKESPEST FINRAGBRGRC HKG
HRVHUNIDNIND IRL
ISL
ISR ITA JPNKAZKHM KORLAO LTU LVA
MAR MEX
MLT
MMR
MYS
NLNORNZLP RPHL POL
PRTROUROW RUS SGPSVK SVN SWETHA TUN
T R TWN U AVNM ZAF
∆(τ)
accelerates
str. trans.
to CS
∆(τ)
decelerates
str. trans.
to CS
−30
0
30
6 7 8 9 10 11
Log(GDP per capita) (1995)
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
55
Table 1.3: Summary Statistics for the Impact of Globalization on Structural Transfor-
mation
Observations Mean Q1 Q2 Q3
Goods 67 4.9 1.0 2.3 6.1
Producer services 67 4.3 0.8 1.9 5.0
Consumer services 67 2.0 0.4 1.1 2.5
* Note: Summary statistics are for the absolute effect of globalization
on structural transformation. Absolute effect measures (changes in GDP
shares in counterfactual #3) - (changes in GDP shares in the baseline
economy).
the purpose of presenting a result for advanced countries, Table 1.4 reports the results
in Figure 1.14 for the 10 countries with largest GDP in 2018. As evidenced by the
proportionality index in the fourth column, all 10 countries experienced more or less
proportional trade cost changes. This, in turn, means that trade cost changes had
quantitatively little effect on the production shares of countries.
Table 1.4: Result for the 10 Largest Countries in 2018 GDP
Shares of GDP
in 1995 (%)
Change in shares
from 1995
to 2018 (p.p.)
Absolute effect
of globalization
(p.p.)
Ctry GDP DV Index G PS CS G PS CS G PS CS
USA 20.6 N 0.2 21.0 38.3 40.7 -3.7 -0.1 3.8 -1.7 0.9 0.8
CHN 13.9 Y 0.1 56.5 22.6 20.9 -12.1 4.1 7.9 -0.6 0.4 0.2
JPN 5.0 N -0.0 28.0 35.1 37.0 -1.0 -0.2 1.2 0.0 -0.4 0.4
DEU 4.0 N -0.1 26.2 33.9 39.9 1.0 -0.6 -0.4 1.7 -1.2 -0.5
GBR 2.9 N -0.0 20.1 38.3 41.5 -3.6 0.3 3.3 -1.0 0.4 0.6
FRA 2.8 N 0.0 20.6 37.0 42.4 -3.2 -0.9 4.1 -0.9 -0.3 1.2
IND 2.8 Y -0.1 48.8 24.4 26.9 -9.5 3.8 5.7 -1.2 1.2 -0.1
ITA 2.1 N -0.0 24.9 35.9 39.2 -0.6 -1.8 2.4 0.5 -0.9 0.4
BRA 1.9 Y -0.1 27.2 31.7 41.1 2.0 -0.1 -1.9 1.2 -0.8 -0.4
KOR 1.7 N 0.3 34.8 31.3 33.9 -2.1 -0.3 2.4 -0.4 0.2 0.2
* Note: GDP is in trillion USDs in 2018. “Ctry” denotes a country. “DV” denotes whether a country
is a developing or an advanced country based on the IMF classification. Index is the proportionality
index for trade cost changes from 1995 to 2018. G, PS, and CS denote goods, producer services, and
consumer services, respectively. Absolute effect measures (changes in GDP shares in counterfactual #3)
- (changes in GDP shares in the baseline economy).
56
To reiterate, the main findings from the counterfactual exercises can be summarized
in the following two points. (i) Despite the multiple channels through which globalization
can affect structural transformation, the cross-sector proportionality index is a summary
statistic that precisely describes why changes in trade costs greatly impact structural
transformation in some countries and not in others. The index measures the relative
speed of trade cost reductions in goods and producer services, hence the impact of
globalization on a country’s comparative advantage. (ii) Globalization mainly affects
structural transformation from goods to producer services. Rarely-tradable consumer
services sectors are the least affected.
1.8 Conclusion
How does globalization affect the pace of structural transformation? I answer this ques-
tion with a focus on the role of tradable services. I show that the relative speed of changes
in trade costs for goods versus producer services, the highly tradable segment of services,
is the key to understand why the impact of globalization on structural transformation
is different across countries. This also provides an answer to the “open question . . . [on]
the extent to which increased trade in services will influence the nature of structural
transformation[,]” pointed out by Herrendorf, Rogerson, and Valentinyi (2014).
I reach this conclusion by first constructing a multi-country, multi-sector model of
trade with non-homothetic preferences and input-output linkages. I then parameterize
the model with the data for 66 countries from 1995 to 2018. I document that the trade
cost asymmetry, where developing countries face higher export trade costs than advanced
countries, decreased over time in both goods and producer services. I further show that
this pattern matters for how comparative advantage of a country changes. Using the
parameterized model, where multiple channels (i.e., income, price, and cross-sector spe-
cialization effect) can generate structural transformation, I quantitatively demonstrate
that the crucial factor to comprehend the heterogenous effect of globalization on struc-
tural transformation is how the relative speed of changes in trade costs for goods versus
producer services affects a country’s comparative advantage.
Future research is on the exact nature of the trade costs. This paper models trade
57
costs as resource costs in an iceberg form. Therefore, the trade costs entail both tech-
nology and policy barriers that hinder international trade. Discovering how much of
the changes in trade costs come from policies would enrich the analysis by enabling the
model to assess how trade liberalization efforts affected structural transformation.
Chapter 2
FDI and Aggregate Productivity
Growth in Chinese Manufacturing
Firms
This paper develops a firm-dynamics model with heterogeneous productivities and for-
eign direct investment (FDI). In the model, a firm can improve its productivity through
foreign technology adoption, innovation, and spillovers (imitation). Unlike domestic
firms, FDI firms possess foreign technology adoption capabilities. Moreover, they par-
ticipate in innovation with different rates from domestic firms. These features of the
model generate different productivity distributions for domestic and FDI firms. The
model is disciplined using the microevidence from Chinese firms and their patents from
1998 to 2007. By calibrating the productivity distributions to the dataset, this study
shows that the annual growth rate of aggregate productivity would decrease from 8.42%
to 7.50% without the presence of FDI firms. Counterfactual exercises demonstrate that
the growth contribution mainly accrues through foreign technology adoption, which ex-
plains 0.72p.p. of the total gain of 0.92p.p.
58
59
2.1 Introduction
Many attribute the explosive growth of China’s manufacturing sector to Deng Xiaoping’s
economic reform. As part of the reform, China opened itself to foreign investment in
1978. To promote foreign direct investment (FDI), it gave foreign investors preferential
treatments, including lower tax rates, which lasted until 2008. As a result, from 1992
to 2019, the value of FDI inflow to China was highest among developing countries every
year except for 2000 (UNCTAD, 2020).1
One unresolved debate in academia regarding the inward FDI of China is over how
it contributes to economic growth. Several reduced-form studies, such as Du, Harri-
son, and Jefferson (2012), focus on two mechanisms: own-firm effects, where FDI brings
superior technology to recipient firms, and spillovers (inter-firm externalities),2 where
foreign firms’ existence improves domestic firms’ productivities. However, to the best of
my knowledge, there is no unified model framework to assess each channel’s relative im-
portance. Studies in the policy arena share the same limitation. They provide narrative
evidence for each channel but do not quantify its macroeconomic impact. For example,
USITC (2010) asserts that FDI led to counterfeiting by domestic firms, but is silent on
measuring its impact.
Furthermore, academic and policy studies overlook one important driver of growth:
innovation. Foreign firms may be growing faster not because they are adopting foreign
technologies, but because they are innovating or, equivalently, developing new technolo-
gies in China. (The empirical section of the paper investigates the innovation channel
and shows its importance.) To that end, this paper separates the own-firm effect into
foreign technology adoption and innovation.
This research aims to answer two questions: First, how much was the contribution
of inward FDI to aggregate productivity growth in Chinese manufacturing firms? Sec-
ond, which mechanism among foreign technology adoption, innovation, and spillovers is
quantitatively most important?
1In 2000, Hong Kong ranked first.
2Spillovers in the FDI literature can be understood as imitation or diffusion in the growth and
development literature. This paper uses the three terms interchangeably. Spillovers can happen when
domestic firms hire a labor force trained in FDI firms or imitate foreign firms’ business practices. See
Harrison and Rodríguez-Clare (2010) for a detailed explanation behind the spillover channels.
60
To answer the questions, this paper develops a firm-dynamics model with heteroge-
neous productivities and FDI and disciplines it with the microdata of Chinese firms and
their patents from 1998 to 2007. In the model, a firm improves its productivity through
foreign technology adoption, innovation, and spillovers. There are two foreign-domestic
differences: First, foreign technology adoption is only available for FDI firms. Second,
they participate in innovation activities at different rates.
Moreover, this study provides empirical evidence of the two differences: First, FDI
firms (equivalently, foreign firms) exhibit faster growth after controlling for information
relevant for growth, such as industry, location, and innovation activities. Second, the
proportion of foreign firms participating in innovation activities such as patenting and
R&D is higher than domestic firms.
The differences produce separate productivity distributions for foreign and domestic
firms in the model. The distributions of the balanced growth path were calibrated
to match three moments of the Chinese dataset: aggregate productivity growth rate,
labor productivity differences between foreign and domestic firms, and those between
patenting and non-patenting firms. The calibrated model predicts that if foreign firms
are replaced with domestic ones, China’s average annual productivity growth rate from
1998 to 2007 will decrease from 8.42% to 7.50%, quantifying the total gain from FDI as
0.92p.p. Furthermore, through counterfactual exercises, this research shows that foreign
technology adoption is quantitatively most important, explaining 0.72p.p. of the total
contribution. Innovation and spillovers account for 0.17p.p. and 0.31p.p., respectively.3
The paper’s main contribution is the development of the quantitative model. To
the best of my knowledge, this paper is the first to quantify the growth effects of FDI
by calibrating the productivity distributions to a firm-level dataset. The model brings
empirical content to the theoretical framework of Benhabib, Perla, and Tonetti (2017).
It has two advantages over the reduced-form studies on FDI and productivity:4 First, it
enables aggregate productivity analysis. Second, counterfactuals can isolate and quantify
each channel through which FDI affects productivity growth.
3Note that the three channels’ contributions do not add up to the total gain because the three
are interdependent, where foreign technology adoption and innovation indirectly affect growth through
spillovers.
4For a detailed survey of the reduced-form research on the topic, see Harrison and Rodríguez-Clare
(2010).
61
Another contribution is the empirical discovery of two patterns of inward FDI in
the Chinese manufacturing industry. First, FDI firms were more innovation-intensive
in an extensive margin. FDI firms were more likely to apply for a patent, get a patent
grant, have a patent with citations, and participate in R&D. This salient data feature
is why the model incorporates innovation. The analysis was made possible by matching
patent data from Google Patents, which encompasses patent data from 105 patent offices
worldwide, to the Chinese firm-level data. Second, the gains from FDI mostly came
from joint ventures. Foreign technology adoption effects were much larger for joint-
ownership firms than fully-foreign-owned firms. Furthermore, joint firms showed higher
participation rates in patenting and R&D than domestic firms, whereas fully-foreign
firms were not so different from the domestic ones. This finding possibly reflects the
Chinese governments’ policy of only allowing joint ventures in certain industries, such
as automobile manufacturing, to benefit from foreign technology adoption.
2.1.1 Related Literature
This study contributes to the literature on growth, innovation, and imitation. It disci-
plines a model similar to Benhabib, Perla, and Tonetti (2017) with detailed firm-level
data. Also, it builds upon the spillover (imitation, diffusion) technology of König et al.
(2020).
The paper is also closely related to the firm-dynamics literature quantifying interna-
tional trade’s role in technology diffusion. It includes Perla, Tonetti, and Waugh (2015)
and Buera and Oberfield (2020). This study differs in that its focus is on another aspect
of international economics: foreign investment.
The effects of inward FDI on Chinese firms are analyzed in Jeon, Park, and Ghauri
(2013), Du, Harrison, and Jefferson (2012), Hu, Jefferson, et al. (2002), Lin, Liu, and
Zhang (2009), and Wang and Wang (2015).5 These papers study signs and magnitudes
of spillovers and own-firm effects of inward FDI through reduced-form methods. The
motivation behind this work is to formulate a model that encompasses both spillovers
and own-firm effects.
This study abstracts from creative destruction, where entrants replace low-productivity
5For the reduced-form studies of FDI and productivity in other countries, see the comprehensive
surveys by Alfaro (2017) and Harrison and Rodríguez-Clare (2010).
62
incumbents. This aspect hinges on the recent finding of Garcia-Macia, Hsieh, and
Klenow (2019) that aggregate productivity growth comes mostly from incumbent firms’
productivity improvement rather than the creative destruction.
The model’s static environment comes from the seminal work of Hsieh and Klenow
(2009), which provides a framework to calculate an individual firm’s productivity and
allows aggregate productivity analysis. The dynamic extension of the environment in
this paper is similar to David and Venkateswaran (2019) and König et al. (2020).
2.1.2 Outline of the Paper
The remainder of the paper is organized as follows. The model of FDI and growth is
set up in Section 2.2. Section 2.3 introduces the data of Chinese firms and patents and
discusses the descriptive evidence. Section 2.4 presents the methodology and result of the
calibration. Section 2.5 lays out the counterfactual exercises, which are the main result
of the paper. Concluding remarks are in Section 2.6. An appendix contains additional
details behind the theory and the data and supplementary tables and figures.
2.2 Theory
2.2.1 Household
Time is continuous and denoted by t. The representative household’s utility is
U =
∫ ∞
0
e−ρt log(C(t))dt,
where C(t) is the consumption of a composite good at t. The household inelastically
supplies one unit of labor and owns domestic firms at every t.6 He or she maximizes the
utility subject to the following per-period budget constraint:
C(t) ≤W (t) + Πd(t),
where W (t) and Πd(t) are a wage rate and total profits from domestic firms in terms
of composite goods at t, respectively. More details on the household problem are in
6This paper abstracts from population growth.
63
Appendix B.1.1.
2.2.2 Composite Good Producer
The composite good is a version of Dixit and Stiglitz (1977). It is produced by a
competitive producer with a constant elasticity of substitution technology:
Y (t) =
(∫ 1
0
Yj(t)
σ−1
σ dj
) σ
σ−1
,
where Yj(t) is a firm j’s output at t. The price index for this composite good is
P (t) ≡
(∫ 1
0
Pj(t)
1−σdj
) 1
1−σ
,
where Pj(t) is a price of a variety j at t.
Let the composite good be a numeraire, i.e., P (t) = 1 ∀t ≥ 0. Cost minimization of
the composite good production (Appendix B.1.2) yields a demand for a variety j:
Pj(t)Yj(t) = Pj(t)
1−σY (t).
2.2.3 Firms
A firm is defined by its access to production technology for a particular differentiated
variety, ownership type, and innovation state. For the brevity of exposition, let j denote
a generic firm’s identity for the remainder of the paper. A production function of a firm
is linear in labor:
Yj(t) = Aj(t)Lj(t),
where Aj(t) is a firm’s productivity and Lj(t) is the labor hired by the firm at t.7 Firm’s
static optimization problem is in Appendix B.1.3.
Each firm’s log-productivity lies on an evenly-spaced quality ladder à la Grossman
7Aj(t) captures both preference shocks and technology levels for good j. (See Luttmer (2007).) This
paper abstracts from the distinction and calls Aj(t) productivity.
64
and Helpman (1991):
log (Aj(t)) = na˜,
for some n ∈ N and where a˜ is a step size of the ladder. For notational simplicity, define
productivity level
aj(t) ≡ log (Aj(t))
a˜
.
Firm j is either owned by domestic households (oj = d) or foreigners (oj = f). A
firm’s ownership oj does not change over time. A firm’s profits belong to its owner.
A firm’s innovation state at t, ij(t), is high (h) when a firm is innovating and low
(l) when it is not. The innovation state of an o-ownership-type firm follows a Markov
chain, where the rate of the transition from low to high is λol and from high to low is
λoh. Note that the transition rates depend on whether a firm is foreign or domestic.
To summarize, firm j is defined as a triple of its productivity level, ownership type,
and innovation state, i.e., (aj(t), oj , ij(t)). For the rest of the model description, it is
useful to define the following notations:
• f(t, a, o, i): measure of (a, o, i) firms at t;
• fo,i(t, o, i) ≡
∑∞
a=1 f(t, a, o, i): measure of (o, i) firms at t;
• fa(t, a) ≡
∑
o
∑
i f(t, a, o, i): measure of (a) firms at t; and
• Fa(t, a) ≡
∑a
a′=1 fa(t, a
′): marginal cdf of (a) firms at t.
2.2.4 Aggregate Productivity, Wage, and Firm Productivity
Under the outlined static environment, the aggregate productivity A¯(t) such that Y (t) =
A¯(t) and wage W (t) are given by
A¯(t) =
(∫ 1
0
Aj(t)
σ−1dj
) 1
σ−1
, and
W (t) =
σ − 1
σ
A¯(t),
65
respectively.
With algebraic manipulation à la Hsieh and Klenow (2009), a firm’s productivity is
given by
Aj(t) =
(Pj(t)Yj(t))
σ
σ−1
W (t)Lj(t)
Y (t)
1
1−σW (t). (2.1)
This is the one-sector version of Hsieh and Klenow (2009) productivity, and all the
elements on the right-hand side have direct analogues for the data. Calculating the
productivity from the data is explained in Section 2.3.1.
2.2.5 Productivity Dynamics
Successful improvement of firm j’s productivity level at t (aj(t) ≡ log(Aj(t))/a˜) is given
by
aj(t+∆t) = aj(t) + 1.
The improvement can happen through foreign technology adoption, innovation, and
spillovers.
Foreign technology adoption is available only for f -ownership-type (FDI) firms. The
success rate for this mechanism is γf , which is non-negative.
Only h-innovation-type firms can improve their productivities through innovation.
The non-negative rate of success is γi. Recall that the Markov processes govern the
transition between h and l innovation types. This simplification of the innovation process
is in the spirit of Benhabib, Perla, and Tonetti (2017).8
The spillover technology follows König et al. (2020) and is a variant of König, Lorenz,
and Zilibotti (2016). Unlike foreign technology adoption and innovation, spillovers are
available for all firms in the economy. If a firm randomly meets another firm whose pro-
ductivity is higher than its own with probability 1−Fa(t, a), the improvement happens
with a non-negative rate q. Hence, the rate of successful improvement through diffusion
8In their model, a high-type firm can endogenously choose innovation intensity. This paper abstracts
from the intensive margin.
66
at t is given by
q(1− Fa(t, a)).
This technology implies that less productive firms are improving their productivities
faster through spillovers. For example, the most productive firm in the economy would
enjoy no spillovers, whereas the least productive firm experiences spillovers at rate q.
For simplicity, assume that all three technologies incur 0 costs. It leads to the absence
of a firm’s dynamic choice of adopting each improvement method.
2.2.6 Exit and Entry
All firms face a random death at rate δ every time period. If a firm j dies at t with
(aj(t), oj , ij(t)), then at the same time, a new firm who has access to the technology of
producing variety j is born. It belongs to the owner of the old firm and keeps the inno-
vation type of the old firm. However, the new firm does not inherit the old productivity
level. It draws its starting productivity level aˆj(t) from fa(t, a). To summarize, when a
firm j with (aj(t), oj , ij(t)) dies, a new firm j with (aˆj(t), oj , ij(t)) is born, where aˆj(t)
is drawn from fa(t, a). This implies that the total measure of firms remains constant at
one and the mass of (o, i) firms at t, fo,i(t, o, i) does not get affected by exit and entry.
This process is a technical assumption for the existence of a balanced growth path
(Definition 2.1). It implies that a domestic (foreign) firm dies and can be reborn with
a foreign (domestic) productivity level. It ensures that foreign firms’ productivity dis-
tributions do not diverge from those of domestic firms. Without this assumption, the
stability of the balanced growth path is only possible for the knife-edge case of γf = 0.
γf > 0 generates a situation where the foreign productivity distributions get infinitely
far away to the right from the domestic ones as time goes.
2.2.7 Value of a Firm
Figure 2.1 summarizes how the state of a firm with productivity level a, ownership type
o, and innovation state i evolves at t.
Given the law of motion for an (a, o, i)-firm’s state and the random exit process, its
67
Figure 2.1: Timeline of an (a, o, i)-type Firm
State at t Transition rates State at t+∆t
(a, o, i)
1{o=f}γf
(a+ 1, o, i)1{i=h}γi
q(1− Fa(t, a))
λoi (a, o, i
′), where i′ ̸= i
δ (death)
Note that 1 is an indicator function.
value at t, V (t, a, o, i) is given by
r(t)V (t, a, o, i)
= π(t, a) (per-period profits)
+
{
1{o=f}γf + 1{i=h}γi + q(1− Fa(t, a))
}
(success rates from a to a+ 1)
× {V (t, a+ 1, o, i)− V (t, a, o, i)} (productivity improvement)
+ λoi
(
V (t, a, o, i′)− V (t, a, o, i)) (transition to i′-type)
− δV (t, a, o, i) (random death)
+ ∂tV (t, a, o, i) (capital gains),
where i′ ̸= i, 1 denotes an indicator function, and r(t) and π(t, a) are a real interest rate
and profits of a firm with productivity level a at t, respectively.9 The value function
does not include the value of a new firm being born with access to the same variety j
when a firm j dies.
9For the analytical expressions of the two, see Appendix B.1.
68
2.2.8 Law of Motion of Productivity Distributions and Balanced Growth
Path
The productivity distribution of (o, i)-firms (the probability distribution of (o, i)-firms’
productivity levels) is given by
f(t, a, o, i)
fo,i(t, o, i)
.
Since fo,i(t, o, i) =
∑∞
a=1 f(t, a, o, i), the productivity distribution is fully characterized
by f(t, a, o, i).
Given the productivity dynamics and exit-entry process, the law of motion for the
measure of (a, o, i)-firms, f(t, a, o, i), is given by
∂tf(t, a, o, i) (2.2)
=− {1{o=f}γf + 1{i=h}γi + q(1− Fa(t, a))} (success rates from a to a+ 1)
× f(t, a, o, i) (measure of (a, o, i) firms)
+
{
1{o=f}γf + 1{i=h}γi + q(1− Fa(t, a− 1))
}
(success rates from a− 1 to a)
× f(t, a− 1, o, i) (measure of (a− 1, o, i) firms)
− λoi f(t, a, o, i) (outflow to i′ type)
+ λoi′f(t, a, o, i
′) (inflow from i′ type)
− δf(t, a, o, i) (random death)
+ δfa(t, a)fo,i(t, o, i) (entry),
where i′ ̸= i, and a ∈ N \ {1}.10 This law of motion is a system of ordinary differential
equations and can be easily solved numerically.
The balanced growth path (BGP, henceforth), whose productivity distributions will
be used for calibration, is defined below:
Definition 2.1. A balanced growth path (BGP) with a growth rate g is an equilibrium
10The case of a = 1 is trivial. The only difference from the a > 1 case is that the term for the inflow
from (a− 1, o, i) firms does not exist.
69
defined in Appendix B.1.4 with
f(t, a, o, i) = f˜(a− gt, o, i) ∀t, a, o, i for some f˜ .
Note that g and f˜ are endogenously determined by model parameters and the mea-
sures of foreign and domestic firms. Although this paper does not prove convergence to
the BGP, it provides numerical evidence of the convergence in Section 2.4.
In the BGP, the productivity distributions of four (o, i) types (foreign-high, foreign-
low, domestic-high, and domestic-low) shift at the same rate g. Furthermore, the mea-
sures of the four types remain constant over time. The four distributions differ from
each other because of foreign technology adoption (γf ) and innovation (γi) terms.
It is worth noting how the three technologies interact to give rise to BGP. h-
innovation-type and f -ownership-type firms push the upper end of the distributions
to diverge. However, the firms located at the lower end of the distributions catch up
through faster diffusion. The forces of divergence and convergence coincide and generate
the BGP distribution.11
2.3 Data and Descriptive Evidence
Chinese firm-level dataset is from the Annual Survey of Industrial Enterprises (ASIE,
henceforth) from 1998 to 2007 conducted by the National Bureau of Statistics (NBS)
of China.12 It is a survey of all private firms with sales higher than five million yuans
(around 700 thousand USDs) and all public firms in the secondary industry, which
includes mining, manufacturing, and utility sectors. Despite the reporting threshold
for private firms, the survey is representative of the whole secondary industry: When
compared to the census of all firms conducted by NBS in 2004, ASIE accounted for more
than 90% of both sales and output (Brandt, Van Biesebroeck, and Zhang, 2014).
The ASIE cross-sections were merged into an unbalanced panel of 562,732 firms.13
The number of firms increased from 165,077 in 1998 to 336,723 in 2007, and the total
number of firm-years was 2,223,433. The information from ASIE used for the analysis
11For theoretical details, see König et al. (2020).
12Brandt, Van Biesebroeck, and Zhang (2014) provides a more detailed description of the dataset.
13Appendix B.2.1 describes the methodology.
70
is a firm’s ownership structure (paid-up capital), value-added, wage bills, four-digit in-
dustry code (Chinese Industry Classification, CIC), six-digit area code, age, and R&D
expenditure. The R&D expenditure variable was reported only for the surveys from
2005 to 2007.
Firm’s foreign ownership percentage was calculated from the paid-up capital infor-
mation:
(Foreign ownership) =
(Foreign capital)+ (Hong Kong, Macao, and Taiwan capital)
(Total paid-up capital)
.
(2.3)
If the paid-up capital information is missing or contains errors, a registration code was
used to determine a firm’s ownership type, instead. The qualitative result of the paper
is robust to whether it includes Hong Kong, Macao, and Taiwan capital for the foreign
ownership calculation.
The ASIE dataset was merged with three other datasets, Chinese Patent Data
Project (He et al., 2018), Google Patents Public Data (Google and IFI CLAIMS, 2020),
and Google Patents Research Data (Google, 2020).14 The first dataset contains ASIE
firms’ patent applications in the Chinese patent office, National Intellectual Property
Administration. The latter two Google datasets provide application, grant, and citation
information in the 105 patent offices worldwide. Merging the datasets allowed the re-
searcher to see a firm’s global patent activities. More details for the patent data are in
Appendix B.2.3.
Note one difference from the theory section: foreign firms are further divided into
two categories: fully-foreign and joint firms. Fully-foreign (joint) firms are defined as
firms with foreign ownership greater than 95% (greater than 5% and less than or equal to
95%). The distinction among foreign firms is due to their different behaviors, which will
be described in detail. For all the empirical analyses, the foreign ownership category is
used instead of the continuous foreign ownership variable (Equation (2.3)). It is because
most firm-years are either domestic or fully-foreign, as shown in Figure 2.2, and using the
continuous measure will lead to under-representation of joint firms. Table 2.1 provides
the summary statistics of the dataset.
14For the introduction to the two Google datasets, see https://cloud.google.com/blog/products/
71
Table 2.1: Summary Statistics: Ownership and Innovation Activities (Extensive Margin)
% of Total
Total firm-years (n = 2,223,443) 100.00
Domestic 80.85
Joint 8.81
Fully-foreign 10.34
Patent application 1.04
Patent grant 0.72
Patent citation 0.90
R&D (2005-2007) 10.04
R&D expenditure information is only available for years from 2005 to 2007, and 10.04% of firm-years
in these three years had positive R&D expenditures.
Figure 2.2: Histogram: Foreign Ownership Shares of Firm-years
Foreign ownership share
%
 o
f t
oa
l
0.0 0.2 0.4 0.6 0.8 1.0
0
20
40
60
80
10
0
72
2.3.1 Hsieh and Klenow (2009) Productivity
Recall the expression for Aj(t) from Equation (2.1):
Aj(t) =
(Pj(t)Yj(t))
σ
σ−1
W (t)Lj(t)
Y (t)
1
1−σW (t).
Following Hsieh and Klenow (2009), it is assumed that σ = 3. The data analogue of each
term on the right-hand side is explained in Table 2.2. All the data objects presented
in the table are real. Price indices described in Appendix B.2.2 are used to convert the
nominal values.
Table 2.2: Data Analogues of Components of Aj(t)
Model Data
Pj(t)Yj(t) Value-added of a firm
W (t)Lj(t) Wage bills of a firm
Y (t) Total value-added of the economy
W (t) Average wage of the economy
The total value-added is the sum of value-added of all firms in the dataset. The average wage was
calculated by dividing the sum of wage bills by the total number of employees hired by the sample
firms.
2.3.2 Foreign-domestic Difference 1: Foreign Technology Adoption
In the model, foreign technology adoption term γi implies that foreign firms’ produc-
tivities are higher and growing faster than domestic firms in the balanced growth path.
Figure 2.3 provides empirical evidence. It plots conditional mean differences in log-
productivity levels and growth between foreign and domestic firms. Each point is a
regression coefficient of an ownership category when log-productivity or its growth rate
(yearly change of log-productivity) is regressed on firm characteristics, including a patent
application indicator (to control for innovation activities).
In Panel (a), there is a general trend that foreign firms are growing faster than
domestic firms with the exceptions of fully-foreign firms in 2002 and 2006, and this is
gcp/google-patents-public-datasets-connecting-public-paid-and-private-patent-data.
73
Figure 2.3: Log-productivity Growth and Level Gap
(a) Growth gap
0.00
0.05
0.10
0.15
1998 1999 2000 2001 2002 2003 2004 2005 2006
Year
Co
ef
fic
ie
nt
s 
of
 o
w
n
e
rs
hi
p 
ca
te
go
rie
s
Ownership
Joint
Joint+Fully−foreign
Fully−foreign
(b) Level gap
0.1
0.2
0.3
0.4
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Year
Co
ef
fic
ie
nt
s 
of
 o
w
n
e
rs
hi
p 
ca
te
go
rie
s
Ownership
Joint
Joint+Fully−foreign
Fully−foreign
Both panels plot regression coefficients of ownership categories over the years. Each point is a conditional
mean difference from that of domestic firms. Panel (a) is from regressing productivity growth (log-
productivity difference) on ownership controlling for industry, province, age, log-productivity, and patent
application dummy for each two-year balanced panel. Panel (b) is from regressing log-productivities on
ownership controlling for industry, province, age, and patent application dummy for each year’s cross-
section. The coefficients for joint and fully-foreign come from models in Table B.1 and Table B.3 and
the ones for joint + fully-foreign are from models in Table B.2 and Table B.4.
74
especially conspicuous for joint firms. Joint firms were growing approximately 18.0p.p.
(4.8p.p.) faster than domestic firms in 1998 (2006). Panel (b) shows that the foreign
firms’ productivities are higher than those of domestic firms. Here, again, joint firms are
the ones that stand out. In 2007, they were 27.4% more productive than their domestic
counterparts. The pattern that foreign firms have both growth and level advantages
over domestic ones in China is coherent with the findings of studies that utilized total
factor productivities estimated with econometric methods. These include Jeon, Park,
and Ghauri (2013), Du, Harrison, and Jefferson (2012), Hu, Jefferson, et al. (2002), and
Lin, Liu, and Zhang (2009).
The growth gap can exist for multiple possible reasons other than foreign technology
adoption, such as better corporate governance and government subsidies. However,
disentangling it is out of this paper’s scope. Hence, this research abstracts from the
distinction and calls the growth gap foreign technology adoption.
Another thing to note is that the level gaps are decreasing, even though foreign firms’
productivities are improving faster. This happens because two-year balanced panels are
used for the growth analysis, and cross-sections are utilized for the level analysis. Entry
and exit explain this seemingly contradictory feature. For example, domestic firms’ exit
rates are higher than those of the foreign firms every year in the data.
2.3.3 Foreign-domestic Difference 2: Innovation
In the model, foreign firms’ innovation types follow a Markov process different from that
of domestic firms, i.e., the transition rates are different: (λfl , λ
f
h) ̸= (λdl , λdh). This section
provides an empirical rationale for this formulation.
Table 2.3 summarizes the extensive margin patenting behaviors of firms by ownership
types. Joint firms’ participation in patent activities stands out compared to the other
ownership categories. On average, 1.74% of joint firm-years were applying for patents,
whereas only 0.97% (1.05%) of domestic (fully-foreign) firm-years had patent applica-
tions. Furthermore, Model (1) in Panel (b) shows that joint firms are 0.8p.p. more
likely to apply for patents than domestic firms. Since patent applications do not convey
any quality information, other measures such as patent grants and patent citations are
considered. Even with these measures, the qualitative result remains the same. For
more details about the patent variables, see Appendix B.2.3.
75
Table 2.3: Patenting Behavior (Extensive Margin)
(a) Proportion of firm-years with patent activities
Application Grant Citation
Domestic 0.0097 0.0067 0.0083
(0.0001) (0.0001) (0.0001)
Joint 0.0174 0.0124 0.0151
(0.0003) (0.0003) (0.0003)
Fully-foreign 0.0105 0.0069 0.0090
(0.0002) (0.0002) (0.0002)
Joint + Fully-foreign 0.0137 0.0094 0.0118
(0.0002) (0.0001) (0.0002)
(b) Correlations between ownership and patent activi-
ties (linear probability models)
Dependent variable:
Application Grant Citation
(1) (2) (3)
Joint 0.008∗∗∗ 0.006∗∗∗ 0.007∗∗∗
(0.0002) (0.0002) (0.0002)
Fully-foreign −0.001∗∗ −0.001∗∗∗ −0.001∗∗∗
(0.0002) (0.0002) (0.0002)
Industry effects Y Y Y
Province effects Y Y Y
Age effects Y Y Y
Year effects Y Y Y
Observations 2,216,697 2,216,697 2,216,697
R2 0.018 0.014 0.017
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Standard errors are in parentheses. In panel (a), joint + fully-foreign denotes firms in either joint or
foreign category.
76
Merging joint and fully-foreign firms for the regression analysis as in Panel (b) of
Table 2.3 yields the coefficient of 0.004. It means that a foreign firm is 0.4p.p. more
likely to apply for patents than its domestic counterpart. However, this number can
be understood to come from two observations: First, the numbers of joint and fully-
foreign firm-years are close enough. (They are 8.81% and 10.34% of the total firm-
years.) Second, joint firms are 0.8p.p. more likely to apply for patents than domestic
firms, whereas foreign firms are 0.1p.p. less likely to do so. To sum up, foreign firms are
more likely to engage in patenting, but this is mainly driven by joint firms.
Appendix B.3 contains the results of the intensive margin analysis of patents (Ta-
ble B.5) and the intensive and extensive margin analyses of R&D (Table B.6). Extensive
margin analysis of R&D shows the same qualitative result with that of patent: joint firms
exhibit higher participation rates than domestic or fully-foreign ones. Ownership did not
explain the intensive margins of patents and R&D well. The analyses had extremely low
R2. The possible reason behind the weak power is the high variance of patent variables.
Most of the patenting firms are in the lower ends of the patent activity distributions
(Figure B.1); however, there are few “superstar” firms with extremely high patent activ-
ity counts. For instance, one firm in 2007 had 4,550 patent applications with more than
9,000 total forward citations.
The innovation participation rate difference and the flow of improvement through
innovation (γi) generates the second channel of FDI’s growth gain. The empirical ev-
idence for γi > 0 is provided in Table B.1 and Table B.2: Firms applying for patents
are growing faster than those that do not for every year. For instance, in 2006, patent-
ing firms grew approximately 27.5p.p. faster than non-patenting ones. The qualitative
result still holds when R&D is used in place of patents.
2.3.4 Spillovers
The spillover technology q(1− Fa(t, a)) characterizes the situation where lower produc-
tivity firms are growing faster. Table B.1 and Table B.2 show that this was the case in
China every year. Although this does not provide conclusive evidence, it suggests the
existence of spillovers in the economy.
77
2.4 Calibration
The model produces four distinct productivity distributions for foreign-high, foreign-
low, domestic-high, and domestic-low firms. The differences among the means of the
distribution were used for calibration.
To simplify the calibration, some of the parameters are directly obtained from the
data. The measure of foreign firms were set as 0.1915. The exit rate δ = − log(0.8702) =
0.1390 to match yearly exit rate of 12.98%. The transition rates for the Markov process
governing the innovation type (λdl , λ
d
h) = (0.0132, 1.3427) and (λ
f
l , λ
f
h) = (0.0102, 0.7354)
so that in the stationary distribution 0.97% and 1.37% of domestic and foreign firms are
high-type, respectively (Panel (a) of Table 2.3), and yearly transition probability from
low to high type is 0.72%.
The remaining parameters (γf , γi, q), which controls the efficiency of foreign tech-
nology adoption, innovation, and spillovers, respectively, are jointly calibrated to match
the following three data moments:
• Mean log-productivity growth: 0.0843,
• Log-productivity difference between patenting and non-patenting firms in 2007:
0.8618, and
• Log-productivity difference between domestic and FDI firms in 2007: 0.1294.
The first moment is an average of mean growth rates of two-year balanced panels. The
last two moments are the conditional mean differences obtained from the regression in
Model (10) of Table B.4.
For calibration, the step size of the quality ladder, a˜, was set as 0.1. The calibration
algorithm is as follows:
1. Start with an initial guess of (γf , γi, q).
2. For t = 100, compute the law of motion (Equation (2.2)) given the initial produc-
tivity distributions.
3. Check the convergence of growth rates and standard deviations of the distributions.
78
4. If the model moments match the data moments, stop. Otherwise, update the guess
and repeat 2-4.
The resulting parameter values (γf , γi, q) = (0.30, 13.32, 1.52). The model moments
matched those from the data well: data moments = (0.0843, 0.8618, 0.1294), model
moments = (0.0842, 0.8619, 0.1293). The distribution at t = 100 is given at Figure 2.4.
The algorithm was robust to the selection of initial guesses and distributions. Further-
more, it showed fast convergence to the balanced growth path. The numerical evidence
of the convergence is presented at Figure 2.5.
Figure 2.4: Productivity Distributions at t = 100
The figure shows the final time period productivity distributions of the calibrated model.
2.5 Counterfactuals
In this section, I use the quantitative model to study the growth effects of FDI and
the relative importance of three channels: foreign technology adoption, innovation, and
spillovers. The model calibrated in the previous section is used as a baseline. The
comparisons among the models are between the balanced growth path growth rates.
To measure the contribution of FDI to aggregate growth, the two foreign-domestic
79
(a) Productivity distributions at t = 80 and t = 100
(b) Convergence of yearly growth rates (c) Convergence of standard deviations
Figure 2.5: Numerical Evidence of Convergence to the Balanced Growth Path
The initial distribution used for the simulation is where all firms are at productivity level 1. y-axis of
Panel (b) denotes (log-productivity growth) × 10.
80
differences are removed:
1. removing foreign technology adoption (γf = 0) and
2. removing innovation intensity difference (setting the values of (λfl , λ
f
h) as (λ
d
l , λ
d
h)).
This exercise is in effect to replace foreign firms with domestic firms. The resulting
annual growth rate is 7.50%. Compared to the baseline growth rate of 8.42%, this means
that the total contribution of FDI to the annual growth rate is 0.92p.p., explaining more
than one-tenth of the total growth.
The decrease of the growth rate comes from two direct and one indirect effects: First,
foreign firms’ productivities are growing slower without foreign technology adoption.
Second, since the proportion of high-type among foreign firms decreases from 1.37% to
the level of domestic firms, 0.97%, there will be fewer innovating firms in the economy,
hence the slower growth. Lastly, the direct effects affect growth through spillovers. The
direct effects would incur the situation where there are fewer firms in the higher rungs
of the quality ladder. This in turn lowers the success rates of spillovers for all firms in
the economy.
The second counterfactual exercise is to quantify the contribution of foreign tech-
nology adoption. Unlike the previous experiment, only foreign technology adoption is
removed, i.e., γf = 0. This yields the annual growth rate of 7.70%. The total gain from
FDI decreased from 0.92p.p. (8.42%−7.50%) to 0.20p.p. (7.70%−7.50%). This implies
that 0.72p.p. out of the total of 0.92p.p. are explained by this channel.
The third experiment is to measure the impact of innovation intensity difference.
Here, only innovation intensity difference is removed: (λfl , λ
f
h) are set at the values of
(λdl , λ
d
h). The growth rate in this scenario is 8.25%, suggesting that it explains 0.17p.p.
of the total gain.
The fourth and last is to evaluate the effect of spillovers. The spillovers from foreign
to domestic firms are blocked. Unlike the previous experiments, where only parameter
values were changed, the functional form of spillovers needs to be changed. Foreign
firms’ spillovers remain the same: they can learn from all firms in the economy. Their
technology is given by
q(1− Fa(t, a)).
81
Unlike foreign firms, domestic firms can only learn from themselves. The spillover effi-
ciency q remains the same, but the random matching only happens with domestic firms,
i.e., the learning technology is
q
(
1−
∑
i
∑a
a′=1 f(t, a
′, d, i)∑
i fo,i(t, d, i)
)
.
This model variant yielded a growth rate of 8.11%, accounting for 0.31p.p. of the total
contribution.
The results for the counterfactual exercises are summarized at Table 2.4. Note that
the growth gains explained by the three channels do not add up to the total gain of
0.92p.p. It is due to the interdependence among the mechanisms: foreign technology
adoption and innovation indirectly affects growth through spillovers.
Table 2.4: Comparison among the Calibrated Model and Four Counterfactuals: Impact
on the Annual Growth Rate
Model variant Growth rate (%) Difference (p.p.)
Baseline 8.42 0
All foreign differences removed 7.50 0.92
Foreign technology adoption removed 7.70 0.72
Innovation difference removed 8.25 0.17
No foreign-to-domestic spillovers 8.11 0.31
Growth rates are the annual growth rates in the balanced growth paths. Differences are the growth rate
differences from that of the baseline.
The sequence of counterfactual exercises highlights two important aspects of inward
FDI in China: First, more than one-tenth of growth can be explained by FDI. Second,
most of the gains from FDI come from foreign technology adoption.
2.6 Conclusion
This article assesses the impact of inward FDI on the aggregate productivity growth
in the manufacturing industry of China. Analysis of Chinese firms and their patents
from 1998 to 2007 reveals three salient patterns: First, foreign firms’ productivities are
82
growing faster than those of domestic firms even after controlling for relevant observables,
such as industry, location, and firm age. Second, FDI firms exhibit higher participation
rates in innovation activities (patenting and R&D). Third, lower-productivity firms are
growing faster than higher-productivity ones, which indicates the existence of spillovers.
These features are incorporated into a firm-dynamics growth model, where firms im-
prove their productivities through foreign technology adoption, innovation, and spillovers.
The model produces different productivity distributions for foreign and domestic firms.
The distributions of the balanced growth path were calibrated to the Chinese data. The
model predicts that the annual growth rate of aggregate productivity would significantly
decrease from 8.42% to 7.50%, if foreign firms are replaced with their domestic coun-
terparts. Most of the gains from FDI (0.92p.p.) incurred through foreign technology
adoption, which explained 0.72p.p. Innovation and spillovers (imitation) accounted for
0.17p.p. and 0.31p.p, respectively. This result suggests that imitation might not be as
important as previously claimed by studies like USITC (2010).
The calibration and counterfactual exercises in this paper rely heavily on the use of
balanced growth paths. This precludes an analysis of a transition path. As shown in
Figure 2.3, there are clear transition patterns in the Chinese manufacturing sector: the
productivity level and growth gaps between FDI and domestic firms decrease over time.
Therefore, the analysis of transition remains an important task for future research.
References
Alfaro, Laura. 2017. “Gains from foreign direct investment: Macro and micro ap-
proaches.” The World Bank Economic Review, 30(Supplement_1): S2–S15.
Alvarez, Fernando and Robert E Jr Lucas. 2007. “General equilibrium analysis
of the Eaton–Kortum model of international trade.” Journal of monetary Economics,
54(6): 1726–1768.
Anderson, James E and Eric Van Wincoop. 2004. “Trade costs.” Journal of Eco-
nomic literature, 42(3): 691–751.
Atalay, Enghin. 2017. “How important are sectoral shocks?” American Economic
Journal: Macroeconomics, 9(4): 254–280.
Baumol, William J. 1967. “Macroeconomics of unbalanced growth: the anatomy of
urban crisis.” The American economic review, 57(3): 415–426.
Benhabib, Jess, Jesse Perla, and Christopher Tonetti. 2017. “Reconciling models
of diffusion and innovation: A theory of the productivity distribution and technology
frontier.” National Bureau of Economic Research.
Benz, Sebastian and Alexander Jaax. 2022. “The costs of regulatory barriers to
trade in services: New estimates of ad valorem tariff equivalents.” Economics Letters,
212: 110057.
Betts, Caroline M and Timothy J Kehoe. 2001. “Tradability of goods and real
exchange rate fluctuations.” Federal Reserve Bank of Minneapolis Staff Report.
83
84
Betts, Caroline, Rahul Giri, and Rubina Verma. 2016. “Trade, reform, and struc-
tural transformation in South Korea.” USC-INET Research Paper, , (16-20).
Blum, Bernardo S, Sebastian Claro, Kunal Dasgupta, and Ignatius J
Horstmann. 2019. “Inventory Management, Product Quality, and Cross-Country
Income Differences.” American Economic Journal: Macroeconomics, 11(1): 338–388.
Brandt, Loren, Johannes Van Biesebroeck, and Yifan Zhang. 2014. “Challenges
of working with the Chinese NBS firm-level data.” China Economic Review, 30: 339–
352.
Brandt, Loren, Johannes Van Biesebroeck, Luhang Wang, and Yifan Zhang.
2017. “WTO accession and performance of Chinese manufacturing firms.” American
Economic Review, 107(9): 2784–2820.
Buera, Francisco J and Ezra Oberfield. 2020. “The global diffusion of ideas.” Econo-
metrica, 88(1): 83–114.
Buera, Francisco J and Joseph P Kaboski. 2012. “The rise of the service economy.”
American Economic Review, 102(6): 2540–69.
Chari, Varadarajan V, Patrick J Kehoe, and Ellen R McGrattan. 2007. “Busi-
ness cycle accounting.” Econometrica, 75(3): 781–836.
Comin, Diego, Danial Lashkari, and Martí Mestieri. 2021. “Structural change
with long-run income and price effects.” Econometrica, 89(1): 311–374.
Cravino, Javier and Sebastian Sotelo. 2019. “Trade-induced structural change and
the skill premium.” American Economic Journal: Macroeconomics, 11(3): 289–326.
David, Joel M and Venky Venkateswaran. 2019. “The sources of capital misallo-
cation.” American Economic Review, 109(7): 2531–67.
Deardorff, Alan V. 2014. “Local comparative advantage: trade costs and the pattern
of trade.” International Journal of Economic Theory, 10(1): 9–35.
Deaton, Angus and Alan Heston. 2010. “Understanding PPPs and PPP-based na-
tional accounts.” American Economic Journal: Macroeconomics, 2(4): 1–35.
85
Dixit, Avinash K and Joseph E Stiglitz. 1977. “Monopolistic competition and
optimum product diversity.” The American economic review, 67(3): 297–308.
Duarte, Margarida and Diego Restuccia. 2020. “Relative prices and sectoral pro-
ductivity.” Journal of the European Economic Association, 18(3): 1400–1443.
Duernecker, Georg, Berthold Herrendorf, and Akos Valentinyi. 2017. “Struc-
tural change within the service sector and the future of Baumol’s disease.” Available
at SSRN 3082293.
Du, Luosha, Ann Harrison, and Gary H Jefferson. 2012. “Testing for horizontal
and vertical foreign investment spillovers in China, 1998–2007.” Journal of Asian
Economics, 23(3): 234–243.
Eaton, Jonathan and Samuel Kortum. 2002. “Technology, geography, and trade.”
Econometrica, 70(5): 1741–1779.
Finicelli, Andrea, Patrizio Pagano, and Massimo Sbracia. 2013. “Ricardian se-
lection.” Journal of international Economics, 89(1): 96–109.
Francois, Joseph and Bernard Hoekman. 2010. “Services trade and policy.” Journal
of economic literature, 48(3): 642–92.
Garcia-Macia, Daniel, Chang-Tai Hsieh, and Peter J Klenow. 2019. “How de-
structive is innovation?” Econometrica, 87(5): 1507–1541.
Google. 2020. “Google Patents Research Data.” data retrieved from Google BigQuery,
https://console.cloud.google.com/marketplace/product/google_patents_
public_datasets/google-patents-research-data.
Google and Google and IFI CLAIMS Patent Services IFI CLAIMS.
2020. “Google Patents Public Data.” data retrieved from Google BigQuery,
https://console.cloud.google.com/marketplace/product/google_patents_
public_datasets/google-patents-public-data.
Grossman, Gene M and Elhanan Helpman. 1991. Innovation and growth in the
global economy. MIT press.
86
Hanoch, Giora. 1975. “Production and demand models with direct or indirect implicit
additivity.” Econometrica: Journal of the Econometric Society, 395–419.
Han, Yuancheng, Jorge Miranda-Pinto, and Satoshi Tanaka. 2022. “Services
Trade and Structural Transformation.” mimeo.
Harrison, Ann and Andrés Rodríguez-Clare. 2010. “Trade, foreign investment, and
industrial policy for developing countries.” In Handbook of development economics.
Vol. 5, 4039–4214. Elsevier.
Herrendorf, Berthold, Richard Rogerson, and Akos Valentinyi. 2014. “Growth
and structural transformation.” Handbook of economic growth, 2: 855–941.
He, Zi-Lin, Tony W Tong, Yuchen Zhang, and Wenlong He. 2018. “A database
linking Chinese patents to China’s census firms.” Scientific data, 5: 180042.
Holmes, Thomas J, Ellen R McGrattan, and Edward C Prescott. 2015. “Quid
pro quo: Technology capital transfers for market access in China.” The Review of
Economic Studies, 82(3): 1154–1193.
Horvát, Peter and Colin Webb. 2020. “The OECD STAN Database for industrial
analysis: Sources and methods.”
Hsieh, Chang-Tai and Peter J Klenow. 2009. “Misallocation and manufacturing
TFP in China and India.” The Quarterly journal of economics, 124(4): 1403–1448.
Hu, Albert GZ, Gary H Jefferson, et al. 2002. “FDI impact and spillover: evidence
from China’s electronic and textile industries.” the world Economy, 25(8): 1063–1076.
IMF, International Monetary Fund. 2022. “World Economic Outlook 2002.”
Inklaar, Robert and Marcel P Timmer. 2014. “The relative price of services.”
Review of Income and Wealth, 60(4): 727–746.
Jeon, Yongbok, Byung Il Park, and Pervez N Ghauri. 2013. “Foreign direct
investment spillover effects in China: Are they different across industries with different
technological levels?” China Economic Review, 26: 105–117.
87
Jorgenson, Dale W and Marcel P Timmer. 2011. “Structural change in advanced
nations: a new set of stylised facts.” Scandinavian Journal of Economics, 113(1): 1–29.
Kee, Hiau Looi, Alessandro Nicita, and Marcelo Olarreaga. 2009. “Estimating
trade restrictiveness indices.” The Economic Journal, 119(534): 172–199.
Kehoe, Timothy J, Kim J Ruhl, and Joseph B Steinberg. 2018. “Global im-
balances and structural change in the United States.” Journal of Political Economy,
126(2): 761–796.
König, Michael D, Jan Lorenz, and Fabrizio Zilibotti. 2016. “Innovation vs.
imitation and the evolution of productivity distributions.” Theoretical Economics,
11(3): 1053–1102.
König, Michael, Zheng Michael Song, Kjetil Storesletten, and Fabrizio Zili-
botti. 2020. “From imitation to innovation: Where is all that Chinese R&D going?”
National Bureau of Economic Research.
Lewis, Logan T, Ryan Monarch, Michael Sposi, and Jing Zhang. 2021. “Struc-
tural Change and Global Trade.” Journal of the European Economic Association.
jvab024.
Limao, Nuno and Anthony J Venables. 2001. “Infrastructure, geographical disad-
vantage, transport costs, and trade.” The world bank economic review, 15(3): 451–479.
Lin, Ping, Zhuomin Liu, and Yifan Zhang. 2009. “Do Chinese domestic firms ben-
efit from FDI inflow?: Evidence of horizontal and vertical spillovers.” China Economic
Review, 20(4): 677–691.
Luttmer, Erzo GJ. 2007. “Selection, growth, and the size distribution of firms.” The
Quarterly Journal of Economics, 122(3): 1103–1144.
Matsuyama, Kiminori. 2009. “Structural change in an interdependent world: A global
view of manufacturing decline.” Journal of the European Economic Association, 7(2-
3): 478–486.
Matsuyama, Kiminori. 2019. “Engel’s law in the global economy: Demand-induced
patterns of structural change, innovation, and trade.” Econometrica, 87(2): 497–528.
88
Mayerhauser, Nicole and Erich Strassner. 2010. “Preview of the comprehensive
revision of the annual industry accounts.” Surv. Curr. Bus, 90(3): 21–34.
Merga, Roman. 2021. “International Trade, Volatility, and Income Differences.” Un-
published manuscript.
Mesquita-Moreira, Mauricio, Christian Volpe, and Juan S Blyde. 2008. “Un-
clogging the Arteries. The impact of transport costs on Latin American and Caribbean
Trade.” Journal of Globalization, Competitiveness & Governability/Revista de Global-
ización, Competitividad y Gobernabilidad/Revista de Globalização, Competitividade e
Governabilidade, 2(3): 126–140.
Muñoz, Mathilde. 2022. “Trading Non-Tradables: The Implications of Europe’s Job
Posting Policy.” mimeo.
OECD. 2021. “OECD Inter-Country Input-Output Database.” data retrieved from
http://oe.cd/icio.
Perla, Jesse, Christopher Tonetti, and Michael E Waugh. 2015. “Equilibrium
technology diffusion, trade, and growth.” National Bureau of Economic Research.
Reyes-Heroles, Ricardo. 2018. “Globalization and Structural Change in the United
States: A Quantitative Assessment.” mimeo.
Simonovska, Ina and Michael E Waugh. 2014. “The elasticity of trade: Estimates
and evidence.” Journal of international Economics, 92(1): 34–50.
Sposi, Michael. 2019. “Evolving comparative advantage, sectoral linkages, and struc-
tural change.” Journal of Monetary Economics, 103: 75–87.
Sposi, Michael, Kei-Mu Yi, and Jing Zhang. 2021. “Deindustrialization and In-
dustry Polarization.” National Bureau of Economic Research.
Sposi, Michael, K Yi, and Jing Zhang. 2018. “Accounting for structural change
over time: A case study of three middle-income countries.” Vol. 1141.
Święcki, Tomasz. 2017. “Determinants of structural change.” Review of Economic Dy-
namics, 24: 95–131.
89
Teignier, Marc. 2018. “The role of trade in structural transformation.” Journal of
Development Economics, 130: 45–65.
UNCTAD, United Nations Conference on Trade and Development. 2020.
“World Investment Report.” Annex table 01 retrieved from, https://unctad.org/
en/Pages/DIAE/World%20Investment%20Report/Annex-Tables.aspx.
USITC, US International Trade Commission and others. 2010. China: Intel-
lectual Property Infringement, Indigenous Innovation Policies, and Frameworks for
Measuring the Effects on the US Economy, Inv. 332-514. DIANE Publishing.
USITC, US International Trade Commission and others. 2022. “Recent trends
in US services trade: 2022 annual report.”
Uy, Timothy, Kei-Mu Yi, and Jing Zhang. 2013. “Structural change in an open
economy.” Journal of Monetary Economics, 60(6): 667–682.
Wang, Jian and Xiao Wang. 2015. “Benefits of foreign ownership: Evidence from
foreign direct investment in China.” Journal of International Economics, 97(2): 325–
338.
Waugh, Michael E. 2010. “International trade and income differences.” American
Economic Review, 100(5): 2093–2124.
Whelan, Karl. 2002. “A guide to US chain aggregated NIPA data.” Review of income
and wealth, 48(2): 217–233.
Yi, Kei-Mu. 2003. “Can vertical specialization explain the growth of world trade?”
Journal of political Economy, 111(1): 52–102.
Appendix A
Appendix for Chapter 1
A.1 Data
A.1.1 Details about OECD ICIO
List of Countries
The following are the list of developing and advanced economies that the data come
from. The classification of a country into a developing or advanced economy follows the
April 2022 edition of the IMF’s World Economic Outlook. Country names are given
in three-letter country codes (ISO 3166-1 alpha-3 codes), and full names are in the
parentheses.
• 36 advanced economies: AUS (Australia), AUT (Austria), BEL (Belgium) , CAN
(Canada), CHE (Switzerland), CYP (Cyprus), CZE (Czech Republic), DEU (Ger-
many), DNK (Denmark), ESP (Spain), EST (Estonia), FIN (Finland), FRA (Fran-
ce), GBR (Great Britain), GRC (Greece), HKG (Hong Kong), IRL (Ireland), ISL
(Iceland), ISR (Israel), ITA (Italy), JPN (Japan), KOR (Korea), LTU (Lithua-
nia), LUX (Luxemburg), LVA (Latvia), MLT (Malta), NLD (Netherlands), NOR
(Norway), NZL (New Zealand), PRT (Portugal), SGP (Singapore), SVK (Slovak
Republic), SVN (Slovenia), SWE (Sweden), TWN (Taiwan), USA (United States)
• 30 developing economies: ARG (Argentina), BGR (Bulgaria), BRA (Brazil), BRN
(Brunei Darussalam), CHL (Chile), CHN (China), COL (Columbia), CRI (Costa
90
91
Rica), HRV (Croatia), HUN (Hungary), IDN (Indonesia), IND (India), KAZ
(Kazakhstan), KHM (Cambodia), LAO (Laos), MAR (Morocco), MEX (Mex-
ico), MMR (Myanmar), MYS (Malaysia), PER (Peru), PHL (Philippines), POL
(Poland), ROU (Romania), RUS (Russia), SAU (Saudi Arabia), THA (Thailand),
TUN (Tunisia), TUR (Turkey), VNM (Viet Nam), ZAF (South Africa)
List of Industries
45 industries in the OECD ICIO are aggregated into three sectors (goods, producer
services, and consumer services). ISIC Rev. 4 industry classification is given in the
parentheses.
• Goods: Agriculture, hunting, forestry (01, 02); Fishing and aquaculture (03); Min-
ing and quarrying, energy producing products (05, 06); Mining and quarrying, non-
energy producing products (07, 08); Mining support service activities (09); Food
products, beverages and tobacco (10, 11, 12); Textiles, textile products, leather and
footwear (13, 14, 15); Wood and products of wood and cork (16); Paper products
and printing (17, 18); Coke and refined petroleum products (19); Chemical and
chemical products (20); Pharmaceuticals, medicinal chemical and botanical prod-
ucts (21); Rubber and plastics products (22); Other non-metallic mineral products
(23); Basic metals (24); Fabricated metal products (25); Computer, electronic and
optical equipment (26); Electrical equipment (27); Machinery and equipment, nec
(28); Motor vehicles, trailers and semi-trailers (29); Other transport equipment
(30); Manufacturing nec; repair and installation of machinery and equipment (31,
32, 33); Electricity, gas, steam and air conditioning supply (35); Water supply;
sewerage, waste management and remediation activities (36, 37, 38, 39)
• Producer services: Wholesale and retail trade; repair of motor vehicles (45, 46,
47); Land transport and transport via pipelines (49); Water transport (50); Air
transport (51); Warehousing and support activities for transportation (52); Postal
and courier activities (53); Publishing, audiovisual and broadcasting activities (58,
59, 60); Telecommunications (61); IT and other information services (62, 63);
Financial and insurance activities (64, 65, 66); Professional, scientific and technical
activities (69 to 75); Administrative and support services (77 to 82);
92
• Consumer services: Construction (41, 42, 43); Accommodation and food service
activities (55, 56); Real estate activities (68); Public administration and defence;
compulsory social security (84); Education (85); Human health and social work
activities (86, 87, 88); Arts, entertainment and recreation (90, 91, 92, 93); Other
service activities (94, 95, 96); Activities of households as employers; undifferenti-
ated goods- and services-producing activities of households for own use (97, 98)
OECD ICIO Data Preparation
Since the model abstracts away from investment, all consumption and investment vari-
ables in the data are merged into a single consumption term. I condense the input-output
tables into three aggregate sectors. Furthermore, another model assumption was the ab-
sence of taxes and subsidies. In OECD ICIO, prices for outputs are given in basic prices
and there is a separate variable for taxes less subsidies (TAXSUB). To remove TAXSUB
while keeping two key input-output table identities (a country-sector’s output = the
input used by the country-sector, and GDP in production = GDP in expenditure), I
change the input-output tables in basic prices into those in purchaser’s prices. This is
done in two steps. First, TAXSUB on a country-sector’s production is allocated to value-
added of the country-sector. Second, TAXSUB on a country’s sectoral final consumption
is allocated to the country where the output is produced. It is added to the value-added
of the producing country-sector and the final consumption of the consuming country’s
on the producing country-sector. Since the OECD ICIO does not provide information of
which producing country-sector the TAXSUB is imposed on, I assume proportionality.
For example, hypothetically, if half of country a’s final consumption on goods comes
from country b, I allocate half of TAXSUB of country a’s goods consumption on country
b.
A.1.2 Construction of Price Sequences
The goal of this section is to lay out the process that I construct the gross-output price
levels for a country-sector. First, I describe how I add up price deflators of industries
that belong to each of the broad three sectors. Second, I show price deflator data
availability for each country, and how I construct price deflators given data limitation of
some countries. Third, I demonstrate in depth how I impute gross-output price deflator
93
from value-added price deflator, if the former are not available. Fourth, I discuss how I
obtain cross-country price levels. Lastly, I discuss how I impute the prices for the ROW.
Aggregation of Subsectors’ Price Deflators
Chain-linked quantities are not additive. Therefore, to aggregate quantities of sectors
into three broad sectors, I use the aggregation method laid out in Whelan (2002) for
the chain-linked Fisher index. For the chain-linked Laspeyres index, I use the method
from Annex 6 of Horvát and Webb (2020), which provides the Laspeyres-version of the
method by Whelan (2002). After aggregating the industry prices into prices of three
broad sectors, I assume that the prices of three sectors are additive.
Availability of Price Deflators
The price notions that are most coherent with the model structure is gross output
deflators. However, not all countries report gross-output price deflators. Table A.6 lays
out the availability of each country’s price deflator data. STAN (GO) and STAN (VA)
denote the availability for the gross-output and value-added price deflator data from
OECD STAN database. They offer full industry details that match can be mapped
to three broad sectors. As for Taiwan, National Statistics of the Republic of China
offers value-added price deflator data with industry details. For other countries not
in the OECD database, I obtained their value-added price deflators from UN National
Accounts, where they do not offer full industry details. I construct gross output price
deflators (GOPD) from the following steps.
1. For country-sector-year’s where GOPD data is available from OECD STAN, use
them.
2. For country-sector-year’s without GOPD data, I impute GOPD from value-added
price deflators (VAPD).
(a) If VAPD is available from OECD STAN, use them.
(b) If VAPD is not available from OECD STAN, I impute VAPD for three broad
sectors from UN National Accounts.
94
3. After obtaining GOPD in local currencies from data and imputation, adjust for
the nominal exchange rate fluctuations.
The details are given in the following subsections.
Gross-output Price Deflators from Value-added Price Deflators
To construct value-added price deflators (VAPD), U.S. Bureau of Economic Analysis
(BEA) uses double deflation method, where they deflate outputs using output prices
and inputs using input prices (Mayerhauser and Strassner, 2010). For countries for
which I can only obtain value-added price deflators, I utilize the inverse of the double
deflation method to obtain gross-output price deflators (GOPD).
The double deflation method in the three sector setting assuming additivity of prices
are given by the following formula. For sector k ∈ S ≡ {g, ps, cs}, the price deflator for
intermediate inputs is given by
P kINT,i,t =
∑
h∈S X
hk
i,t∑
h∈S Q
hk
i,t
, (A.1)
where Xhki,t and Q
hk
i,t denote nominal and real sector h intermediate usage for sector k
production. Then, the quantity index for the sector k value-added is
QkV A,i,t =
Xki,t
P ki,t
− X
k
INT,i,t
P kINT,i,t
, (A.2)
where XkINT,i,t =
∑
h∈S X
hk
i,t , and X
k
i,t and Q
k
i,t denote nominal and real outputs by
sector k. From the quantity, one can derive the VAPD.
With VAPD and input-output data, I find GOPDs that yield VAPDs consistent with
the data. Specifically, for Equations (A.1) and (A.2), data observables are {Xhki,t }h,k,
{Xki,t}k (from OECD ICIO) and {QkV A,i,t}k (quantity derived through price deflators
from OECD STAN), and I find {P ki,t}k that yields {Qhki,t }h,k that are consistent with
the equations. In essence, this is deriving three GOPDs from three VAPDs using three
equations.
95
Performance of Imputed Gross-output Price Deflators
In this subsection, I show that the imputed GOPD using the method in Appendix A.1.2
returns reasonable estimates for the actual GOPD. I use 23 countries who report both
GOPD and VAPD in the OECD STAN database to verify the validity of the imputed
price deflators. I utilize three ways for to check the imputation performance and demon-
strate that the imputed GOPD outperforms VAPD as a proxy for GOPD. Let P⃗ kGO,i,
P⃗ kV A,i and P⃗
k
IMP,i denote the time series of GOPD, VAPD, and imputed GOPD for
country i sector k. The three comparisons are
• cor(P⃗ kGO,i, P⃗
k
V A,i) vs cor(P⃗
k
GO,i, P⃗
k
IMP,i)
• RMSE(P⃗ kGO,i, P⃗
k
V A,i) vs RMSE(P⃗
k
GO,i, P⃗
k
IMP,i), where root-mean-square error
RMSE(a, b) =
√⟨(a− b), (a− b)⟩/(number of years).
• Regression coefficient β1 from regressing P⃗ kGO,i on P⃗
k
V A,i vs β2 from regressing P⃗
k
GO,i
on P⃗ kIMP,i (regressions are done without intercepts.)
Compared to P kV A,i, if P
k
IMP,i yields higher correlation, lower RMSE, or the regression
coefficient closer to one, then it means that the imputed GOPD outperforms VAPD.
Correlation captures the directional movements, and RMSE and the regression coeffi-
cients complement the correlation measure by bringing-in level-wise similarity across two
price sequences. Among 69 country-sectors (23 countries and 3 sectors), the imputed
GOPD outperformed VAPD in 49, 50, 46 country-sectors in terms of correlation, RMSE,
and the regression. Table A.1 lists summary statistics for the comparisons in the three
metrics. It shows that for goods the imputation provides superior performance. Since
value-added shares are typically lower for goods production, VAPD is not a good measure
for GOPD, as evidenced by the summary statistics. However, the imputation corrects
for this bias. In Table A.2, I show the result for the country-sectors whose VAPD’s
performance is one of the top-10 worst in each of the three metrics. Slovakia’s goods
sector is a good example where the imputed GOPD matches GOPD well (Figure A.1).
96
Table A.1: Summary Statistics for the Three Metrics for VAPD vs Imputed GOPD
Correlation RMSE Reg. coef.
Ind Stats VA IMP VA IMP VA IMP
1 G Mean 0.75 0.91 0.07 0.05 0.98 1.00
2 G Q1 0.71 0.91 0.04 0.03 0.96 0.98
3 G Q2 0.95 0.97 0.07 0.05 0.99 0.99
4 G Q3 0.98 0.99 0.09 0.06 1.02 1.02
5 PS Mean 0.99 0.99 0.02 0.02 1.00 1.00
6 PS Q1 0.99 0.99 0.01 0.01 0.99 0.99
7 PS Q2 0.99 1.00 0.02 0.02 1.00 1.00
8 PS Q3 1.00 1.00 0.03 0.03 1.01 1.01
9 CS Mean 0.99 0.99 0.03 0.02 1.01 1.00
10 CS Q1 1.00 1.00 0.01 0.01 1.00 0.99
11 CS Q2 1.00 1.00 0.03 0.01 1.01 1.00
12 CS Q3 1.00 1.00 0.04 0.02 1.02 1.01
* Note: Columns Correlation-VA and Correlation-IMP denote the
summary statistics for cor(P⃗ kGO,i, P⃗ kV A,i) and cor(P⃗ kGO,i, P⃗ kIMP,i), re-
spectively. Other columns follow symmetric definitions. For column
Ind, G, PS, CS denote goods, producer services, and consumer ser-
vices, respectively. Q1, Q2, and Q3 denote first, second, and third
quartiles, respectively.
Imputing Value-added Price Deflators for Three Broad Sectors when Full
Industry Details are Missing
OECD STAN provides full industry details which I can use for price deflator aggregation
into three broad sectors. However, for countries that are not in OECD STAN database,
I need to impute their sectoral prices from less granular UN National Accounts (UN NA)
data. I extract price deflators in UN NA from two sources: UN SDMX and UN Analysis
of Main Aggregates (AMA). The former reports aggregated price deflators for sectors G
(wholesale and retail trade, repair of motor vehicles), H (transportation and storage),
and I (accommodation and food services) of ISIC Rev. 4 classification. I split GHI into
GH and I by using nominal value-added outputs observed in OECD ICIO as weights.
UN AMA reports aggregated prices for ISIC Rev. 3 J to P as other services. Here I
split this category using the value-added outputs as well. Table A.4 shows correlation,
RMSE, regression measures of how VAPDs from UN data compares with VAPDs from
OECD STAN for countries reporting in both datasets. Imputed VAPDs are good proxies
97
Table A.2: Comparison between Imputed GOPD and VAPD for Select Country-sectors
Correlation RMSE Reg. coef.
Ctry Ind VA IMP Impr. VA IMP Impr. VA IMP Impr.
1 BEL G 0.77 0.96 Y 0.13 0.08 Y 0.91 0.96 Y
2 CHE CS 0.99 1.00 Y 0.05 0.01 Y 1.05 1.01 Y
3 CHE G 0.90 0.81 N 0.02 0.03 N 1.02 1.03 N
4 CZE G 0.66 0.91 Y 0.06 0.03 Y 0.99 0.99 Y
5 DEU G 0.92 0.92 Y 0.04 0.03 Y 0.99 1.00 Y
6 FIN G -0.49 0.63 Y 0.11 0.05 Y 0.94 0.99 Y
7 FRA G -0.08 0.88 Y 0.10 0.06 Y 0.95 0.97 Y
8 GRC G 0.94 0.96 Y 0.07 0.06 Y 1.05 1.03 Y
9 HUN G 0.99 0.99 Y 0.09 0.09 N 1.09 1.08 Y
10 JPN G 0.25 0.25 Y 0.18 0.14 Y 0.88 0.91 Y
11 LUX G 0.98 0.97 N 0.07 0.09 N 1.06 1.09 N
12 NOR G 0.99 1.00 Y 0.07 0.05 Y 0.97 0.98 Y
13 PRT G 0.87 0.90 Y 0.05 0.04 Y 1.03 1.02 Y
14 SVK G 0.37 0.97 Y 0.21 0.04 Y 0.85 0.98 Y
15 SWE G 0.48 0.88 Y 0.09 0.06 Y 0.96 0.98 Y
16 USA G 0.97 0.97 Y 0.09 0.06 Y 0.95 0.98 Y
* Note: Columns Correlation-VA and Correlation-IMP denote the cor(P⃗ kGO,i, P⃗ kV A,i) and
cor(P⃗ kGO,i, P⃗
k
IMP,i), respectively. Column Correlation-Impr. denotes whether imputed GOPD out-
performs VAPD as a proxy for actual GOPD in terms of correlation. Other columns follow symmetric
definitions. For column Ind, G and CS denote goods and consumer services.
for actual VAPDs.
Cross-country Price Levels in 2005
Through the procedures laid out in the previous sections, I obtain within-country sectoral
price changes over time. To have full price level sequences, I use 2005 cross-country
sectoral price level data. It is obtained from the Productivity Level Database 2005
Benchmark of Groningen Growth and Development Centre (Inklaar and Timmer, 2014).
It provides price levels information for 42 countries and 35 industries. The 35 industries
are mapped into aggregate sectors of goods, producer services, and consumer services.
For 42 countries, sectoral price levels exhibit strong linear relationship with income levels
of countries (Figure A.2). Using this relationship I impute sectoral price levels for 24
countries that are not in the dataset.
98
Figure A.1: Price Deflators for Slovakia’s Goods: GOPD vs VAPD vs Imputed GOPD
GO
GO (imputed)
VA
0.7
0.8
0.9
1.0
1.1
1.2
1995 2000 2005 2010 2015
Year
Pr
ic
e 
de
fla
to
rs
 (2
01
5 =
 1)
* Note: GO, VA, GO (imputed) denote GOPD, VAPD, imputed GOPD, respectively.
Imputing Prices for the Rest-of-the-world Aggregate
Since I do not have price level data for the rest-of-the-world (ROW) aggregate that
accounts for approximately 7% of world GDP and 8% of world total export, I impute
their prices. For each sector, I regress log-prices for country-year’s on countries’ log
of GDP per capita with year fixed effect. Using the relationship between income and
prices, I impute the prices for the ROW. ROW’s GDP and population is obtained from
OECD ICIO and PWT. For ROW’s population, I subtracted population of 66 countries
from the population of all sample countries in the PWT.
99
Table A.3: Summary Statistics for the Three Metrics for VAPD (OECD STAN) vs
Imputed VAPD from UN SDMX and UN AMA
Correlation RMSE Reg. coef.
Ind Stats SDMX AMA SDMX AMA SDMX AMA
1 G Mean 0.99 0.98 0.01 0.02 1.00 1.01
2 G Q1 1.00 0.97 0.00 0.01 1.00 1.00
3 G Q2 1.00 1.00 0.00 0.01 1.00 1.00
4 G Q3 1.00 1.00 0.01 0.02 1.00 1.01
5 PS Mean 1.00 0.98 0.01 0.04 1.01 1.02
6 PS Q1 1.00 0.98 0.01 0.03 1.00 1.01
7 PS Q2 1.00 0.99 0.01 0.03 1.00 1.03
8 PS Q3 1.00 1.00 0.01 0.05 1.01 1.04
9 CS Mean 1.00 0.99 0.01 0.03 1.00 1.00
10 CS Q1 1.00 1.00 0.00 0.02 0.99 0.98
11 CS Q2 1.00 1.00 0.01 0.02 1.00 0.99
12 CS Q3 1.00 1.00 0.01 0.03 1.00 1.01
* Note: columns Correlation-UN SDMX denotes the summary statistics for the
correlation between the VAPD from OECD STAN and VAPD from UN SDMX.
Other columns follow symmetric definitions. For column Ind, G, PS, CS denote
goods, producer services, and consumer services, respectively. Q1, Q2, and Q3
denote first, second, and third quartiles, respectively.
A.2 Estimation and Calibration
A.2.1 Model Fit: Non-targeted Moment Exercise
This section provides model fit for non-targeted moments. To do so, I adopt an alterna-
tive calibration strategy for a two-country (USA-ROW) model similar to that of Kehoe,
Ruhl, and Steinberg (2018). Sectoral TFP growth rates of the US are calculated from
the BEA-BLS KLEMS data. Output quantities of the industries belonging to each of the
three broad sectors are aggregated through Törnqvist quantity index. The productivity
sequence for the ROW are derived such that the sectoral prices are equalized between
the two countries. Trade costs are calibrated through the same approach outlined in the
paper.
Figure A.3 provides the model fit for the baseline model for explaining the structural
transformation pattern of the US. The baseline model replicates the pattern well. It
also compares the model’s performance to that of non-tradable services model, where
100
Figure A.2: Relationship between Sectoral Price Levels and Income
ARG
AUS
AUTBEL
BGR
BRA
CAN
CHL
CHN
CYP
CZE
DEUDNK
ESP
EST
FIN
FRAGBR
GRC
HUN
IDN
IND
IRLITA
JPN
KOR
LTU
LUX
LVA
MEX
MLT
NLD
POL
PRT
ROU
RUS
SVK
SVN
SWE
TUR
USA
ZAF
ARG
AUS
AUTBEL
BGR
BRA
CAN
CHL
CHN
CYP
CZE
DEU
DNK
ESP
EST
FIN
FRAGBR
GRC
HUN
IDN
IND
IRL
ITA
JPN
KOR
LTU
UX
LVA
MEX
MLT
NLD
POL
PRT
ROURUS
SVK
SVN
SWE
TUR
USA
ZAF
ARG
AUSAUT
BEL
BGR
BRA
CAN
CHL
CHN
CYP
CZE
DEU
DNK
ESP
EST
FINFRAGBR
GRC
HUN
IDN
IND
IRL
ITA
JPN
KOR
LTU
LUX
LVA
MEX
MLT
NLD
POL
PRT
ROU
RUS
SVK
SVN
SWE
TUR
USA
ZAF
(a) Goods (b) Producer services (c) Consumer services
8 9 10 11 8 9 10 11 8 9 10 11
−2.0
−1.5
−1.0
−0.5
0.0
−1.2
−0.8
−0.4
0.0
−0.6
−0.3
0.0
0.3
Log GDP per capita (2005US$)
Lo
g(p
ric
e 
re
la
tiv
e
 to
 th
e 
US
)
trade costs for producer and consumer services are set to infinity. The two models’
performance are similar before 2009, but after then the tradable model outperforms the
non-tradable one. The likely cause is that the non-tradable model does not capture the
enhanced cross-sector specialization after 2009 (Figure A.4).
A.2.2 Comparative Advantage
To introduce the definition of comparative advantage by Deardorff (2014) in a Ricardian
model in a simple manner, assume that there are two countries a and b, two sectors g
and s and that there is no input-output linkages (psiLki = 1, ∀i ∈ I ∀k ∈ S.) I will
omit time notations for brevity. A country’s effective comparative advantage is defined
as the following.
Definition A.1. Country a has effective comparative advantage in goods, if
CAgab ≡
Aga
Asa
/
Agb
Asb
√
τ gba
τ gab
/
τ sba
τ sab
> 1.
101
Figure A.3: Model Fit: Non-targeted Moments
Data
NT
T
Data
NT
T
DataNT
T
(a) Goods (b) Producer services (c) Consumer services
1995 2000 2005 2010 2015 1995 2000 2005 2010 2015 1995 2000 2005 2010 2015
0.41
0.42
0.43
0.44
0.45
0.36
0.37
0.38
0.39
0.16
0.18
0.20
0.22
Year
Sh
ar
e 
of
 G
DP
* Note: T denotes the baseline (tradable services), and NT denotes non-tradable services model.
CAgab measures the strength of country a’s comparative advantage in g due to produc-
tivity differences adjusting for the trade costs.
Proposition A.1. Suppose that country a and country b’s consumption patterns and
country sizes are the same. Further suppose that θg = θs. If country a has comparative
advantage in goods, that is, CAgab > 1, then, it is a net exporter of goods.
Proof. (
πgab
1− πgab
/
πgba
1− πgba
)/(
πsab
1− πsab
/
πsba
1− πsba
)
=
{
τ gab
τ gba
(
WaA
g
b
WbA
g
a
)2}−θg/{
τ sab
τ sba
(
WaA
s
b
WbAsa
)2}−θs
.
102
Figure A.4: Sectoral Net Exports of the US
Consumer services
Goods
Producer services
−1000
−750
−500
−250
0
250
1995 2000 2005 2010 2015
Year
N
et
 e
xp
or
t (b
illio
n U
SD
)
The assumptions θs = θg > 0 and CAgab > 1 imply that(
πgab
1− πgab
/
πgba
1− πgba
)/(
πsab
1− πsab
/
πsba
1− πsba
)
> 1.
Since countries do not have differences in their consumption patterns (ωga = ωgb ⇔ ωsa =
ωsb) and sizes (WaLa =WbLb), the above implies(
Xgab
1−Xgab
/
Xgba
1−Xgba
)/(
Xsab
1−Xsab
/
Xsba
1−Xsba
)
> 1, (A.3)
where Xkij denotes trade flows from country i to j in sector k.
Claim: Country a is a net exporer of g, that is Xgab −Xgba = − (Xsab −Xsba) > 0.
If I assume to the contrary that Xgab−Xgba ≤ 0, the left-hand side of Equation (A.3)
is less than or equal to 1.
A.2.3 Productivity Growth and Structural Transformation
This subsection describes the pattern of productivity growth and its implications on
structural transformation. Regarding the productivity growth, I use two measures: total
103
factor productivity
(
Aki,t
)
and measured labor productivity
(
Y ki,t
Lki,t
)
. The reason why
I report labor productivity is because it has more direct implications on the prices
rather than total factor productivity: due to input-output linkages, changes in total
factor productivity is not directly mapped into changes in prices. Figure A.5 describes
how productivities and relative prices evolved over time. Labor productivity increased
fastest in producer services and slowest in producer services, with productivity growth
in goods in between (Panel (b)). This had implications on the relative prices as well.
Relative to goods, producer services prices decreased and those for consumer services
increased (Panel (c)). Since the outputs of the three sectors are complements, this means
that generally price effect was incurring consumer services shares in consumption going
up and those for producer services going down. This reinforces the argument by the
papers (e.g., Jorgenson and Timmer, 2011; and Duernecker, Herrendorf, and Valentinyi,
2017) arguing that Baumol’s cost disease should be revisited with disaggregated services.
Baumol (1967) points out that since goods and services are complements and services
productivities are growing slower than goods, economic activities move away from goods
to services incurring aggregate growth slowdown.
Productivity growth also affects structural transformation through the income effect.
It makes the real income of households, and households shift their expenditure shares
away from income-inelastic goods to income-elastic producer and consumer services.
A.3 Counterfactuals
A.3.1 Solution Method for the Baseline and Counterfactuals
I use a modified version of the solution method from Alvarez and Lucas (2007) to solve
the model for each time period. The algorithms are stated below: Given model primi-
tives,
1. Guess {Wi}i∈I ∈ ∆−→W ≡
{−→
W ∈ RI+ :
∑
i∈IWiLi = 1
}
.
2. From Equations (F1) and (G1), derive P ki , R
k
i .
1
1In the setup of Alvarez and Lucas (2007), this step is a contraction mapping. As for this paper’s
setup, this step is a non-expansive mapping. Numerically, the non-expansive mapping converged to
fixed points.
104
3. From Equations (H2) and (H3), derive Pi, Ui.
4. From Equation (H1), derive Cki
5. From Equation (G2), derive πkij
6. From Equations (F3) and (M1), derive Qkhi and Y
k
i .
7. Modifying Equation (G3), define the excess demand for country i’s products in
terms of wage rates
Zi(
−→
W ) ≡ 1
Wi
∑
k∈S
∑
j∈I
πkijP
k
j Y
k
j − nxiWUSALi −
∑
k∈S
P ki Y
k
i
 .
8. If Zi(
−→
W ) = 0 ∀i, stop. Otherwise, update −→W using the following mapping:
Ti(
−→
W ) =Wi
(
1 + κ
Zi(
−→
W )
Li
)
,
where κ is a small number.
A.4 Robustness of Quantitative Results
A.4.1 Different Definitions of Sectors
This subsection provides robustness analyses for two alternative definitions of sectors.
First, I classify all producer service industries and the two tradable consumer service
industries (“Accomodation and food” and “Art, entertainment, recreation”) as highly-
tradable services and the rest of consumer service industries as barely-tradable services
(Figure A.9). Second, I explore the case of two-sector setup (goods and services). For
each of the two alternative models, I re-estimate the model. Figures A.10 and A.11
demonstrate that the main result in Figure 1.14 holds in the alternative sectoral defini-
tions.
105
A.4.2 Non-tradable Consumer Services
This subsection demonstrates that trade costs for consumer services do not affect quanti-
tative result of the paper much by assuming consumer services are non-tradable through
out the period, i.e., τ csij,t = ∞ ∀i ̸= j ∈ I, t ∈ T . The main result are reported for this
case in Figure A.12.
A.4.3 Trade Elasticities for Services
Since there is no consensus on the trade elasticity for services (θps and θcs), I set it so that
θps = θcs = θg = 4. Trade elasticities matter for the result of the paper for two reasons.
First, in the model inversion, i.e., Equations (1.13) and (1.14), trade elasticities govern
how sensitive derived productivities and trade costs are to import shares. Second, in the
model, they determine how sensitive prices and import shares are to the productivities
and trade costs (Equations (G1) and (G2)). Because of this circular structure of model
inversion and model simulation, it is ex-ante unclear how trade elasticities affect the
quantitative result. Therefore, I explore the cases where trade elasticities for services
are lower at 2 and higher at 6. Figures A.13 and A.14 demonstrate that the main result
of the paper holds under these two scenarios.
A.4.4 Wedges on Utility and Production Weights
As mentioned in Section 1.6.1, since the utility weights (ϕki ) and production weights
(ψhki and ψ
Lk
i ) are time invariant, the baseline model does not fit the data perfectly.
This modeling choice was to ensure that the productivity sequences from the model
inversion are interpretable. In Figure A.15, I provide the result for when I assume that
utility and production weights are time-varying so that the model fits the data perfectly.
A.4.5 Different Assumption on Production Function
The baseline model assumes that the production function for a variety is the following.
yki (z) = A
k
i α
k
i (z)
{(
ψLki
) 1
ρk
(
Lki (z)
) ρk−1
ρk +
∑
h∈S
(
ψhki
) 1
ρk
(
Qhki (z)
) ρk−1
ρk
} ρk
ρk−1
.
106
To show that this functional assumption that labor is aggregated with the intermediate
inputs under the same degree of substitutability, I test another functional form, a nested
CES production function.
yki (z) =A
k
i α
k
i (z)
[(
χLki
) 1
µk
(
Lki (z)
)µk−1
µk
+
(
1− χLki
) 1
µk
{∑
h∈S
(
ψhki
) 1
ρk
(
Qhki (z)
) ρk−1
ρk
} ρk
ρk−1
µk−1
µk
] µk
µk−1
,
With the re-estimated model, I show that the main quantitative result is similar to the
baseline (Figure A.16)
A.4.6 The Role of Exogenous Net Exports
The baseline model assumes that the net transfer to a country is exogenously given as
units of US labor (numeraire). The result of the paper remains robust when I assume
balanced trade by setting nxi,t = 0∀i ∈ I, t ∈ T (Figure A.17).
A.5 Additional figures and tables
107
Table A.4: Summary Statistics for the Three Metrics for VAPD (OECD STAN) vs
Imputed VAPD from UN SDMX and UN AMA
Correlation RMSE Reg. coef.
Ind Stats SDMX AMA SDMX AMA SDMX AMA
1 G Mean 0.99 0.98 0.01 0.02 1.00 1.01
2 G Q1 1.00 0.97 0.00 0.01 1.00 1.00
3 G Q2 1.00 1.00 0.00 0.01 1.00 1.00
4 G Q3 1.00 1.00 0.01 0.02 1.00 1.01
5 PS Mean 1.00 0.98 0.01 0.04 1.01 1.02
6 PS Q1 1.00 0.98 0.01 0.03 1.00 1.01
7 PS Q2 1.00 0.99 0.01 0.03 1.00 1.03
8 PS Q3 1.00 1.00 0.01 0.05 1.01 1.04
9 CS Mean 1.00 0.99 0.01 0.03 1.00 1.00
10 CS Q1 1.00 1.00 0.00 0.02 0.99 0.98
11 CS Q2 1.00 1.00 0.01 0.02 1.00 0.99
12 CS Q3 1.00 1.00 0.01 0.03 1.00 1.01
* Note: columns Correlation-UN SDMX denotes the summary statistics for the
correlation between the VAPD from OECD STAN and VAPD from UN SDMX.
Other columns follow symmetric definitions. For column Ind, G, PS, CS denote
goods, producer services, and consumer services, respectively. Q1, Q2, and Q3
denote first, second, and third quartiles, respectively.
108
Figure A.5: Productivity Growth and Relative Prices from 1995 to 2018
(a) TFP Growth Factors
Q1
Q2
Q3
Q1
Q2
Q3
Q1
Q2
Q3
Goods Producer services Consumer services
1995 2000 2005 2010 2015 1995 2000 2005 2010 2015 1995 2000 2005 2010 2015
1.0
1.2
1.4
1.6
Year
Qu
ar
til
es
(b) Labor Productivity Growth Factors
Q1
Q2
Q3
Q1
Q2
Q3
Q1
Q2
Q3
Goods Producer services Consumer services
1995 2000 2005 2010 2015 1995 2000 2005 2010 2015 1995 2000 2005 2010 2015
1.2
1.5
1.8
2.1
Year
Qu
ar
til
es
(c) Prices Relative to Goods
Q1
Q2
Q3
Q1
Q2
Q3
Producer services Consumer services
1995 2000 2005 2010 2015 1995 2000 2005 2010 2015
0.4
0.5
0.6
0.7
0.8
0.7
0.8
0.9
Year
Pr
ic
e 
re
la
tiv
e
 to
 g
oo
ds
* Note: Growth factor is productivity of a country-sector divided by productivity of the country-sector
in the initial year. Q1, Q2, and Q3, represent first, second, and third quartiles, respectively.
109
Table A.5: Full Results for 66 Countries + ROW
* Note: DV denotes whether a country is developing country; T (G) denotes growth rates of the ratio
of export trade costs to import trade costs for goods (log points); and Index denotes the
proportionality index. The GDP shares are in percents.
Data Baseline C1 C2 C3
1995 2018 1995 2018 1995 2018 1995 2018 1995 2018
Country DV T (G) T (PS) Index G PS G PS G PS G PS G PS G PS G PS G PS G PS G PS
1 ARG Y 0.20 0.68 -0.48 41 31 31 33 34 32 35 32 34 32 26 38 34 32 39 29 34 32 32 34
2 AUS N -0.85 -0.58 -0.27 25 35 21 35 23 36 22 35 23 36 25 32 23 36 17 40 23 36 21 37
3 AUT N 0.22 0.31 -0.10 27 32 24 34 25 34 26 32 25 34 18 39 25 34 32 27 25 34 26 34
4 BEL N 0.00 0.01 -0.01 26 36 18 42 23 40 21 38 23 40 18 40 23 40 23 37 23 40 21 38
5 BGR Y -1.24 -1.01 -0.23 27 30 26 39 35 34 29 34 35 34 41 23 35 34 14 45 35 34 27 34
6 BRA Y -0.56 -0.46 -0.10 28 32 26 34 27 32 29 32 27 32 30 31 27 32 26 34 27 32 28 32
7 BRN Y -1.79 0.45 -2.24 65 17 61 18 52 22 63 17 52 22 48 29 52 22 17 55 52 22 7 60
8 CAN N -0.41 -1.12 0.72 28 30 21 34 26 32 23 33 26 32 35 20 26 32 19 36 26 32 31 23
9 CHE N 0.16 -0.10 0.26 25 39 22 42 23 42 24 42 23 42 22 42 23 42 29 37 23 42 28 38
10 CHL Y -0.67 -0.25 -0.42 34 28 28 34 32 34 31 32 32 34 32 32 32 34 26 37 32 34 27 36
11 CHN Y -1.74 -1.88 0.14 57 22 42 29 57 23 44 27 57 23 56 17 57 23 34 36 57 23 45 26
12 COL Y -0.22 -0.21 -0.01 33 28 30 30 33 27 32 29 33 27 32 29 33 27 31 29 33 27 31 29
13 CRI Y -0.54 -0.86 0.32 38 29 21 38 35 30 25 34 35 30 38 26 35 30 18 39 35 30 31 32
14 CYP N 0.36 0.17 0.19 18 38 11 42 15 38 11 42 15 38 6 31 15 38 20 35 15 38 13 36
15 CZE N -0.77 -0.84 0.07 36 30 32 33 35 33 34 31 35 33 50 21 35 33 19 43 35 33 34 34
16 DEU N 0.46 0.54 -0.08 28 32 27 34 26 34 27 33 26 34 17 40 26 34 35 27 26 34 26 34
17 DNK N -0.02 0.21 -0.22 24 33 19 37 23 35 21 35 23 35 15 40 23 35 23 32 23 35 19 37
18 ESP N 0.06 -0.03 0.10 26 31 20 34 24 32 21 33 24 32 20 34 24 32 23 31 24 32 23 32
19 EST N -0.99 -1.46 0.47 30 31 24 37 30 34 25 37 30 34 51 14 30 34 9 47 30 34 33 28
20 FIN N 0.57 0.17 0.39 32 28 24 31 31 30 25 29 31 30 19 33 31 30 37 19 31 30 31 24
21 FRA N 0.43 0.41 0.01 23 35 16 38 21 37 17 36 21 37 12 41 21 37 24 31 21 37 18 36
22 GBR N 0.21 0.21 -0.00 23 35 15 40 20 38 17 39 20 38 12 42 20 38 21 35 20 38 18 38
23 GRC N -0.12 0.09 -0.21 24 34 19 33 22 33 22 33 22 33 17 34 22 33 22 33 22 33 19 33
24 HKG N 1.06 0.35 0.71 11 51 3 56 9 52 2 52 9 52 1 48 9 52 10 49 9 52 4 52
25 HRV Y -0.34 -0.45 0.11 34 31 25 35 32 34 26 33 32 34 32 31 32 34 23 34 32 34 31 33
26 HUN Y -0.10 -0.62 0.52 35 28 29 34 33 33 30 32 33 33 38 24 33 33 24 39 33 33 35 30
27 IDN Y -0.90 -0.28 -0.62 50 29 44 29 48 29 45 29 48 29 44 30 48 29 37 36 48 29 39 35
28 IND Y -0.91 -0.83 -0.08 52 21 39 30 49 24 39 28 49 24 43 24 49 24 34 33 49 24 40 27
29 IRL N -0.91 -1.27 0.35 30 35 37 39 29 38 33 40 29 38 55 19 29 38 10 61 29 38 40 35
30 ISL N -0.29 -0.08 -0.21 29 31 19 35 28 34 21 32 28 34 25 36 28 34 17 34 28 34 22 37
31 ISR N -0.40 -0.36 -0.05 24 29 17 37 21 33 19 36 21 33 20 33 21 33 16 38 21 33 20 35
32 ITA N -0.10 -0.07 -0.04 28 35 23 35 25 36 24 34 25 36 23 36 25 36 24 34 25 36 24 35
33 JPN N 1.35 1.38 -0.02 30 33 26 35 28 35 27 35 28 35 11 49 28 35 40 23 28 35 27 35
34 KAZ Y -1.29 -1.04 -0.25 39 34 34 38 34 36 35 38 34 36 48 26 34 36 12 60 34 36 33 43
35 KHM Y -0.59 -0.80 0.21 58 19 56 20 61 20 55 22 61 20 63 16 61 20 49 26 61 20 60 20
36 KOR N 0.65 0.34 0.31 35 30 33 31 35 31 33 31 35 31 23 39 35 31 42 24 35 31 33 31
37 LAO Y -0.89 -0.88 -0.02 49 29 45 24 51 27 46 27 51 27 52 22 51 27 40 32 51 27 45 28
38 LTU N -1.51 -2.22 0.71 36 28 25 42 35 32 25 40 35 32 61 7 35 32 7 55 35 32 36 28
39 LUX N -0.05 -1.14 1.09 16 48 8 60 15 50 8 58 15 50 29 30 15 50 10 55 15 50 35 26
40 LVA N -1.64 -1.33 -0.30 35 33 20 39 27 41 22 40 27 41 44 24 27 41 5 52 27 41 19 43
41 MAR Y 0.64 0.37 0.27 38 27 39 25 44 25 37 24 44 25 28 32 44 25 49 15 44 25 40 24
42 MEX Y -0.78 -0.26 -0.51 33 34 29 38 32 36 31 34 32 36 33 33 32 36 22 42 32 36 26 40
43 MLT N -0.43 -0.93 0.50 27 35 11 44 23 40 11 36 23 40 31 35 23 40 7 42 23 40 23 43
44 MMR Y -0.26 -0.26 -0.01 65 18 62 20 66 16 57 22 66 16 56 23 66 16 59 21 66 16 59 22
45 MYS Y -0.41 -0.09 -0.32 38 28 39 31 36 28 38 31 36 28 27 30 36 28 37 31 36 28 31 32
46 NLD N 0.12 0.16 -0.04 25 38 18 44 21 43 20 42 21 43 15 47 21 43 23 40 21 43 19 44
47 NOR N -0.85 0.09 -0.94 31 31 30 28 30 31 32 27 30 31 29 31 30 31 18 39 30 31 17 42
48 NZL N -0.55 -0.39 -0.16 30 34 21 35 26 35 24 35 26 35 23 32 26 35 20 37 26 35 23 35
49 PER Y -0.62 -0.68 0.06 43 27 35 31 39 29 38 29 39 29 41 26 39 29 31 32 39 29 38 30
50 PHL Y -0.06 -0.64 0.58 44 30 41 34 46 29 35 37 46 29 39 31 46 29 40 33 46 29 44 28
51 POL Y -0.61 -0.63 0.02 36 31 28 41 34 36 28 39 34 36 34 33 34 36 20 46 34 36 30 38
52 PRT N 0.17 0.09 0.07 27 33 21 34 25 34 22 33 25 34 21 37 25 34 26 31 25 34 25 34
53 ROU Y -1.36 -1.77 0.41 50 24 30 35 47 26 33 32 47 26 63 16 47 26 16 47 47 26 40 28
54 ROW Y -0.38 -0.34 -0.04 45 25 44 26 46 26 44 25 46 26 44 24 46 26 41 28 46 26 43 26
55 RUS Y -1.56 -1.17 -0.39 36 36 36 32 36 35 35 34 36 35 52 26 36 35 17 51 36 35 34 38
56 SAU Y -1.88 -0.72 -1.16 48 16 46 20 46 19 49 17 46 19 57 12 46 19 17 42 46 19 36 29
57 SGP N -0.01 -0.45 0.44 26 48 20 53 22 49 20 53 22 49 22 41 22 49 24 51 22 49 29 41
58 SVK N -0.72 -1.65 0.93 33 29 29 36 33 33 30 33 33 33 56 10 33 33 15 45 33 33 42 22
59 SVN N 0.06 -0.05 0.11 34 31 29 36 32 33 30 34 32 33 30 34 32 33 31 34 32 33 31 33
60 SWE N 0.44 0.27 0.17 30 30 20 38 28 35 22 35 28 35 15 41 28 35 33 26 28 35 26 33
61 THA Y -0.41 -0.24 -0.17 44 34 43 33 43 34 43 35 43 34 40 35 43 34 37 39 43 34 39 39
62 TUN Y 0.41 1.02 -0.62 38 28 32 31 36 29 34 29 36 29 12 40 36 29 43 23 36 29 23 37
63 TUR Y 0.55 0.46 0.09 41 33 32 35 37 34 32 34 37 34 25 39 37 34 39 28 37 34 35 33
64 TWN N 1.22 0.75 0.47 33 34 34 35 34 34 31 35 34 34 10 52 34 34 55 13 34 34 34 32
65 USA N -0.06 -0.29 0.23 22 36 16 40 21 38 17 38 21 38 17 38 21 38 18 37 21 38 19 37
66 VNM Y -2.43 -1.87 -0.56 58 20 57 21 56 21 56 22 56 21 76 6 56 21 16 54 56 21 45 30
67 ZAF Y 0.01 0.27 -0.26 31 30 25 33 30 33 26 33 30 33 22 37 30 33 30 31 30 33 26 35
110
Figure A.6: India’s Trade Costs and Structural Transformation
(a) Trade cost patterns
Import cost (PS)Import cost (G)
Export cost (G)
Export cost (PS)
5
10
15
1995 2000 2005 2010 2015
Year
Tr
a
de
 c
os
ts
(i) Trade costs (weighted means)
G
PS
1.0
1.5
2.0
2.5
1995 2000 2005 2010 2015
Year
Lo
g(e
xp
or
t c
os
t/i
m
po
rt 
co
st
)
(ii) Log(export cost/import cost)
(b) Structural transformation in the baseline and counterfactuals #1 to #3
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Pr
od
uc
tio
n 
sh
ar
es
Baseline (G, PS, CS)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 1 (G)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 2 (PS & CS)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 3 (none)
Year
* Note: Trade costs reported in Panel (a) are weighted averages of trade costs across India’s trade
partners. The weights are trade flows in 1995.
111
Figure A.7: Vietnam’s Trade Costs and Structural Transformation
(a) Trade cost patterns
Import cost (PS)Import cost (G)
Export cost (G)
Export cost (PS)
0
10
20
30
1995 2000 2005 2010 2015
Year
Tr
a
de
 c
os
ts
(i) Trade costs (weighted means)
G
PS
1
2
3
1995 2000 2005 2010 2015
Year
Lo
g(e
xp
or
t c
os
t/i
m
po
rt 
co
st
)
(ii) Log(export cost/import cost)
(b) Structural transformation in the baseline and counterfactuals #1 to #3
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Pr
od
uc
tio
n 
sh
ar
es
Baseline (G, PS, CS)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 1 (G)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 2 (PS & CS)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 3 (none)
Year
* Note: Trade costs reported in Panel (a) are weighted averages of trade costs across Vietnam’s trade
partners. The weights are trade flows in 1995.
112
Figure A.8: Lithuania’s Trade Costs and Structural Transformation
(a) Trade cost patterns
Import cost (PS)Import cost (G)
Export cost (G)
Export cost (PS)
0
10
20
30
1995 2000 2005 2010 2015
Year
Tr
a
de
 c
os
ts
(i) Trade costs (weighted means)
G
PS
1.0
1.5
2.0
2.5
3.0
1995 2000 2005 2010 2015
Year
Lo
g(e
xp
or
t c
os
t/i
m
po
rt 
co
st
)
(ii) Log(export cost/import cost)
(b) Structural transformation in the baseline and counterfactuals #1 to #3
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Pr
od
uc
tio
n 
sh
ar
es
Baseline (G, PS, CS)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 1 (G)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 2 (PS & CS)
CS
PS
G
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1995 2000 2005 2010 2015
Counterfactual 3 (none)
Year
* Note: Trade costs reported in Panel (a) are weighted averages of trade costs across Lithuania’s trade
partners. The weights are trade flows in 1995.
113
Table A.6: Availability of Price Deflators from OECD STAN, UN NA SDMX, UN NA
AMA
STAN (GO) STAN (VA) UN NA SDMX UN NA AMA
Country Start End Start End Start End Start End
1 ARG 1995 2018
2 AUS 1995 2018 1995 2018 1995 2018
3 AUT 1995 2018 1995 2018 1995 2018 1995 2018
4 BEL 1995 2018 1995 2018 1995 2018 1995 2018
5 BGR 1995 2018 1995 2018
6 BRA 2000 2018 1995 2018
7 BRN 1995 2018
8 CAN 1997 2017 1995 2018
9 CHE 1995 2017 1995 2017 1997 2018 1995 2018
10 CHL 1996 2018 1995 2018
11 CHN 1995 2018 1995 2018
12 COL 2005 2018 1995 2018
13 CRI 1995 2018 1995 2018
14 CYP 1995 2018 1995 2018
15 CZE 1995 2018 1995 2018 1995 2018 1995 2018
16 DEU 1995 2018 1995 2018 1995 2018 1995 2018
17 DNK 1995 2018 1995 2018 1995 2018 1995 2018
18 ESP 1995 2018 1995 2018 1995 2018
19 EST 1995 2018 1995 2018 1995 2018 1995 2018
20 FIN 1995 2018 1995 2018 1995 2018 1995 2018
21 FRA 1995 2018 1995 2018 1995 2018 1995 2018
22 GBR 1995 2018 1995 2018 1995 2018
23 GRC 1995 2018 1995 2018 1995 2018 1995 2018
24 HKG 1995 2018
25 HRV 1995 2018 1995 2018
26 HUN 1995 2018 1995 2018 1995 2018 1995 2018
27 IDN 2010 2018 1995 2018
28 IND 2011 2017 1995 2018
29 IRL 1995 2018 1995 2018 1995 2018
30 ISL 1995 2018 1995 2018 1995 2018
31 ISR 1995 2018 1995 2018
32 ITA 1995 2017 1995 2018 1995 2018 1995 2018
33 JPN 1995 2018 1995 2018 1995 2018 1995 2018
34 KAZ 1995 2018
35 KHM 1995 2018
36 KOR 1995 2018 1995 2018 1995 2018
37 LAO 1995 2018
38 LTU 1995 2018 1995 2018 1995 2018
39 LUX 1995 2018 1995 2018 1995 2018 1995 2018
40 LVA 1995 2018 1995 2018 1995 2018 1995 2018
41 MAR 1995 2018
42 MEX 1995 2018 1995 2018 1995 2018 1995 2018
(Continued in the next page)
114
STAN (GO) STAN (VA) UN NA SDMX UN NA AMA
Country Start End Start End Start End Start End
(Continued from the previous page)
43 MLT 1995 2018 1995 2018
44 MMR 1995 2018
45 MYS 1995 2018
46 NLD 1995 2018 1995 2018 1995 2018 1995 2018
47 NOR 1995 2018 1995 2018 1995 2018 1995 2018
48 NZL 1995 2018 1995 2018
49 PER 1995 2018
50 PHL 1995 2018
51 POL 1995 2018 1995 2018 1995 2018 1995 2018
52 PRT 1995 2017 1995 2018 1995 2018 1995 2018
53 ROU 1995 2018 1995 2018
54 ROW
55 RUS 2011 2018 1995 2018
56 SAU 1995 2018 1995 2018
57 SGP 1995 2018
58 SVK 1995 2016 1995 2018 1995 2018 1995 2018
59 SVN 1995 2018 1995 2018 1995 2018
60 SWE 1995 2018 1995 2018 1995 2018 1995 2018
61 THA 1995 2018
62 TUN 1995 2018
63 TUR 1998 2018 1995 2018
64 TWN
65 USA 1995 2018 1995 2018 1997 2018 1995 2018
66 VNM 1995 2018
67 ZAF 1995 2018 1995 2018
* Note: Price deflators from Taiwan is obtained from National Statistics of the Republic
of China (Taiwan). For ROW, the prices are imputed.
115
Figure A.9: Tradedness and Intermediateness of Service Industries in 2018
Construction
Wholesale and retail & repair of motor
Land transport
Water transport
Air transport
Warehousing
Postal services
Accommodation & food
Publishing, audiovisual, broadcastingTelecom
IT
Financial & insurance
Real estate
Professional, scientific, technical
Admin
Public Education
Health
Art, entertainment, recreation
Other services
Household
Goods
0.00
0.25
0.50
0.75
0.0 0.1 0.2 0.3 0.4
Total export / total gross output
To
ta
l i
nt
er
m
e
di
at
e 
us
ag
e 
/ t
ot
al
 g
ro
ss
 o
ut
pu
t
* Source: Author’s calculation from the OECD Inter-Country Input-Output Tables.
* Note: Orange rectangles and green circles denote industries belong to highly-tradable services and
barely-tradable services, respectively. For full names of the industries, refer to Appendix A.1.
116
Figure A.10: Globalization in Both Goods and Producer Services and Structural Trans-
formation (Goods, Highly-tradable Services, and Barely-tradable Services)
ARG AUS
AUTBELBGR BRA
CAN
CHE
CHL
CHN
COL
CRI
CYPCZE
DEUDNK
ESP
EST
FIN
FRA
GBR
GRC
HKG
HRV HUN
IDN
IND
IRL
ISL ISR ITA
JPN
KAZ
KHM
KOR
LAO
LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZL P R
PHL
POL RT
ROU
ROWRUS
SGP SVK
SVN
SWE
THA
TUN
TUR TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS
AUTBELBGR BRA
CAN
CHE
CHL
CHNCOL
CRICYP
CZEDEUDNK
ESP
EST
FIN
FRA
GBR
GRC HKG
HRV HUN
IDN
IND IRL
ISL
ISR ITA
JPN
KAZ
KHM
KORLA
LTU
LVA MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZL PER
PHL
POL RT
ROU
ROW
RUS
SGP
SVK
SVN
SWE
THA
TUN
TUR
TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to HTS
∆(τ)
decelerates
str. trans.
to HTS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Highly tradable services
ARG AUS AUTBELBGR BRA CANCHECHL CHNCOL
CRI
CYP
CZE
DEUDNK ESP ESTFINFRA GBRGRC
HKGHRV HUN
IDN IND
IRL
ISL
ISR ITA JPN
KAZ
KHM KORLA LTU
LVA
MAR
MEX MLTMMR
MYS
NLDNOR NZL P PHL
POL PRT ROUOW
RUS
SGP
SVKVNSWETHATUN TUR TWNUSAVNM ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to BTS
∆(τ)
decelerates
str. trans.
to BTS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(c) Barely tradable services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
117
Figure A.11: Globalization in Both Goods and Producer Services and Structural Trans-
formation (Goods and Services)
ARG
AUS
AUT
BEL
BGRBRA
CAN
CHE
CHL
CHN
COL
CRI
CYP CZE
DEUDNK
ESP
EST FIN
FRAGBR
GRC
HKG
HRV HUN
IDN
IND IRL
ISL
ISR ITA
JPN
KAZ
KHM KOR
LAO
LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZL PER
PHL
POL
PRT
ROU
ROWRUS
SGP
SVK
SVN
WE
THA
TUN
TU TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG
AUS
AUT
BEL
BGRBRA
CAN
CHE
CHL
CHN
COL
CRI
CYP CZE
DEUDNK
ESP
EST FIN
FRAGBR
GRC
HKG
HRV HUN
IDN
IND IRL
ISL
ISR ITAJPN
KAZ
KHM KORLAO
LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZL PER
PHL
POL
PRT
ROU
ROWRUS
SGP
SVK
SVN
WE
THA
TUN
TU TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to S
∆(τ)
decelerates
str. trans.
to S
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
118
Figure A.12: Globalization in Both Goods and Producer Services and Structural Trans-
formation (Non-tradable Consumer Services)
ARG AUS AUT
BELBGR
BRA
CAN
CHE
CHL
CHNCOL
CRICYP
CZEDEUDNK
ESP
EST
FIN
FRAGBR
RC
HKGHRV HUN
IDN
IND
IRL
ISL
ISRITAJPN
KAZ
KHM
KORL O
LTU
LVA
AR
MEX
MLT
MM
MYS
N D
NOR
NZL PER
PHL
POLPRT
ROU
ROWRUS
SGP
SVK
SVN
SWE
THA
TUN
TUR TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS AUT
BEL
BGR BRA
CAN
CHE
CHL
CHNCOL
CRICYP
CZEDEUDNK
ESP
EST
FIN
FRAGBRC
HKG
HRV HUN
IDN
IND IRL
ISL
ISRITAJPN
KAZ
KHM
KORL O
LTU
LVA
MAR
MEX
MLT
MM
MYS
NLD
NOR
NZL PER
PHL
POLPRT
ROU
ROW
RUS
SGP
SVK
SVN
SWE
THA
TUN
TUR
TWN
USA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to PS
∆(τ)
decelerates
str. trans.
to PS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS AUT BELBGR BRA CANCHECHL CHNCOL CRICYP
CZE
DEUDNK ESP ESTFINFRAGBRGRC
HKGHRV
HUN
IDN IND
IRL
ISL ISRITAJPN
KAZ
KHM KORLAO LTULVA
MAR
MEX MLTMMMYS NLDNOR NZL P R PHL
POLPRT ROUROW
RUS SGP SVKSVNSWETHATUN TU TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to CS
∆(τ)
decelerates
str. trans.
to CS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
119
Figure A.13: Globalization in Both Goods and Producer Services and Structural Trans-
formation (Low Trade Elasticities for Services)
ARG AUS
AUTBEL BGRBRA
CAN
CHE
CHL
CHNCOL
CRI
CYPCZE
DEUDNK
ESP
EST
FIN
FRA GBR
GRC
HKG
HRV HUN
IDN
IND
IRL
ISL
I R
ITAJPN
KAZ
KHM
K R
LAOLVA
MAR
MEX
MMR
MYS
NLDNZL PER
PHL
POLPRT
ROU
ROW
RUS
SGP
SVNSWE
HA
TUN
TU TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS AUTBEL BGRBRA
CAN
CHE
CHL
CHNCOL
CRI
CYP
CZEDEUDNK
ESP
EST
FIN
FRA GBRGRC HKGHRV
HUN
IDN
IND
IRL
ISL
ISR
ITAJPN KAZ
KHM
K RLAO
LVA
MAR
MEX
MMR
MYS NLDNZL
PE
PHL
POL
PRT
ROU
ROW
RUS
SGP
SVNSWE
THA
TUN
TUR
TWN
US
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to PS
∆(τ)
decelerates
str. trans.
to PS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS
AUTBEL BGRBRA CANCHECHL CHNCOL
CRI
CYP
CZE
DEUDNK ESP ESTFINFRA GBRGRC
HKG
HRV
HUN
IDN IND
IRL
ISL
ISRITAJPN
KAZ
KHM
K RLAO
LVA
MAR
MEX
MMR
MYS
NLDNZL PER PHLPOL
PRT
ROUROW
RUS
SGPSVNSWE
THA
TUN
TUR
TWNUSAVNMZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to CS
∆(τ)
decelerates
str. trans.
to CS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, Saudi Arabia, Lithuania,
Norway, Slovakia, and Malta) are not plotted. Full result including the outlier countries are in Table A.5
in Appendix A.5.
120
Figure A.14: Globalization in Both Goods and Producer Services and Structural Trans-
formation (High Trade Elasticities for Services)
ARG
AUS
AUT BEL
BGR
BRA
CAN
CHE
CHL
CHN
COL
CRI
CYPCZE
DEUDNK
ESP
ESTFIN
FRAGB
GRC
HKG
HRV HUN
IDN
IND IRLISL
ISR ITAJPN
KAZ
KHM
KORLAO
LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZL PER
PHL
POL
PRT
ROU
WRUS
SGP
SVK
SVN
SWE
THA
TUN
TUR TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS AUT
BELBGR
BRA
CAN
CHE
CHL
CHNCOL CRI
CYP
CZE
DEUDNK
ESP
EST
FIN
FRAGBRGRC HKG
HRV
HUN
IDN
IND
IRL
ISL
ISR ITAJPN
KAZ
KHM
KORLAO
LTU
LVA
MAR
MEX MLT
MMRMYS NLD
NOR
NZL PER
PHL
POL
PRT
ROU
ROW
RUS
SGP
SVK
SVN
SWE
THA
TUN
TUR
TWN
USA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to PS
∆(τ)
decelerates
str. trans.
to PS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS
AUT
BEL
BGR
BRA
CAN
CHECHL CHNCOL
CRI
CYP
CZE
DEUDNK
ESP
EST
FINFRAGBR
GRC
HKG
HRV
HUNIDN I D IRL
ISL
ISR ITAJPNKAZ
KHM KORLAO
LTU
LVA MAR
MEX
MLT
MMR
MYS
NLD
NOR NZL PER PHLPOL
P T
ROU
W
RUS
SGP
SVKVN
SWE
THA
TUN
TUR
TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to CS
∆(τ)
decelerates
str. trans.
to CS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
121
Figure A.15: Globalization in Both Goods and Producer Services and Structural Trans-
formation (Wedges for Preference and Production)
ARG AUS
AUT BEL
BGR BRA
CAN
CHE
CHL
CHNCOL
CRI
CYP
CZE
DEUDNK
ESP
EST
FIN
FRAGBR
C
HKG
HRV HUN
IDN
IND
IRL
ISL ISR
ITAJPN
KAZ
KHM
KORL O
LTU
LVA
MAR
MEX
MLT
MMR
MYS
N D
NOR
NZL PER
PHL
POLPRT
ROU
ROWRUS
SGP
SVK
SVN
SWE
THA
TUN
TUR TWN
USA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS AUT
BELBGR BRA
CAN
CHE
CHL
CHNCOL
CRI
CYP
CZE
DEUDNK
ESP
EST
FIN
FRAGBRC HKGHRV
HUN
IDN
IND
IRL
ISL
ISR
ITAJPN
KAZ
KHM
KORLAO
LTU
LVA
MAR
MEX MLT
MM
MYS NLD
NOR
NZL PER
PHL
POL
PRT
ROU
ROW
RUS
SGP SVK
SVN
SWE
THA
TUN
TU
TWN
USA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to PS
∆(τ)
decelerates
str. trans.
to PS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS
AUT
BEL
BGR BRA CAN
CHECHL CHNCO
CRI
CYP
CZE
DEUDNK ESP EST
FINFRAGBR
C
HKG
HRV
HUN
IDN IND
IRL
ISL
ISRITAJPN
KAZ KHM
KORLAO LTU
LVA
AR
MEX
MLT
MMR
MYS
NLD
NOR NZL PER PHL
POL
PRT ROU
ROW
RUS
SGP
SVKSVNSWETHA
TUN
TUR
TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to CS
∆(τ)
decelerates
str. trans.
to CS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
122
Figure A.16: Globalization in Both Goods and Producer Services and Structural Trans-
formation (Nest-CES Production Function)
ARG AUS AUT BEL
BGR BRA
CAN
CHE
CHL
CHNCOL
CRI
CYPCZE
DEUDNK
ESP
EST
FIN
FRAGBR
C
HKG
HRV
HUN
IDN
IND
IRL
ISL ISR
ITAJPNKAZ
KHM
KORL O
LTU
LVA
MAR
MEX
MLT
MMR
MYS
N D
NOR
NZL PER
PHL
POLPRT
ROU
ROWRUS
SGP
SVK
SVN
SWE
THA
TUN
TUR TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS AUT
BELBGR BRA
CAN
CHE
CHL
CHNCOL
CRI
CYP
CZE
DEUDNK
ESP
EST
FIN
FRAGBRC HKGHRV
HUN
IDN
IND
IRL
ISL
ISR
ITAJPN
KAZ
KHM
KORLAO
LTU
LVA
MAR
MEX
MLT
MMMYS NLD
NOR
NZL PER
PHL
POL
PRT
ROU
ROW
RUS
SGP SVK
SVN
SWE
THA
TUN
TU
TWN
USA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to PS
∆(τ)
decelerates
str. trans.
to PS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS
AUT
BEL
BGR
BRA CAN
CHECHL CHNCO
CRI
CYP
CZE
DEUDNK ESP EST
FINFRAGBR
C
HKG
HRV
HUN
IDN IND
IRL
ISL
ISRITAJPN
KAZ
KHM
KORLAO LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLDNOR NZL PER PHLPOL
PRT ROU
ROW
RUS
SGP
SVKSVNSWETHA
TUN
TUR
TWN
USA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to CS
∆(τ)
decelerates
str. trans.
to CS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
123
Figure A.17: Globalization in Both Goods and Producer Services and Structural Trans-
formation (nx = 0)
ARG
AUS
AUT BEL
BGR BRA
CAN
CHE
CHL
CHNCOL
CRI
CYPCZE
DEUDNK
ESP
EST
FIN
FRAGBR
C
HKG
HRV HUN
IDN
IND IRLISL ISR
ITAJPN
KAZ
KHM
KO
LAO
LTU
LVA
MAR
MEX
MLT
MMR
MYS
NLD
NOR
NZL PER
PHL
POLPRT
ROU
ROWRUS
SGP
SVK
SVN
SWE
THA
TUN
TUR TWN
USA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
decelerates
str. trans.
from G
∆(τ)
accelerates
str. trans.
from G
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(a) Goods
ARG AUS AUT
BELBGR BRA
CAN
CHE
CHL
CHNCOL
CRI
CYP
CZEDEUDNK
ESP
EST
FIN
FRAGBRGRC
HKGHRV
HUN
IDN
IND
IRL
ISL
ISR
ITAJPN
KAZ
KHM
KOR
LAO
LTU
LVA
MAR
MEX MLT
MM
MYS N D
NOR
NZL PER
PHL
POL
PRT
ROU
ROW
RUS
SGP SVK
SVN
SWE
THA
TUN
TU
TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to PS
∆(τ)
decelerates
str. trans.
to PS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(b) Producer services
ARG AUS
AUT
BEL
BGR
BRA CAN
CHECHL CHNCOL
CRI
CYP
CZE
DEUDNK ESP EST
FINFRAGBR
RC
HKG
HRV
HUN
IDN IND
IRL
ISL
ISRITAJPNKAZ
KHM
KORL O LTU
LVA AR
MEX
MLT
MMR
MYS
NLD
NOR NZL PER
PHLP L
PRT ROU
ROWRUS
SGP
SVKSVNSWETHA
TUN
TUR TWNUSA
VNM
ZAF
∆(τ) weakens CAg∆(τ) strengthens CAg
∆(τ)
accelerates
str. trans.
to CS
∆(τ)
decelerates
str. trans.
to CS
−20
−10
0
10
20
−1.0 −0.5 0.0 0.5 1.0
Index
Ef
fe
ct
 (p
.
p.
)
(c) Consumer services
* Note: Absolute effect (y-axis) denotes (changes in GDP shares in counterfactual #3) - (changes in
GDP shares in the baseline economy). Outlier countries (Brunei, Luxembourg, and Saudi Arabia) are
not plotted. Full result including the outlier countries are in Table A.5 in Appendix A.5.
Appendix B
Appendix for Chapter 2
B.1 Theory
B.1.1 Household’s Optimization
The household solves the following maximization problem:
max
{C(t)}t≥0
∫ ∞
0
e−ρt log(C(t))dt,
subject to∫ ∞
0
exp
(
−
∫ t
0
r(s)ds
)
C(t)dt ≤
∫ ∞
0
exp
(
−
∫ t
0
r(s)ds
)
{W (t) + Πd(t)}dt, and
C(t) ≥ 0 ∀t,
where r(s) is a real interest rate at time s, andW (t) and Πd(t) are a wage rate and total
profits from domestic firms in terms of composite goods at t, respectively.
Note the assumption of the composite good being the numeraire. Also, the equiva-
lence of the problems under per-period budget constraints and under the present-value
version is trivial.
The Euler Equation from the household problem is given by
r(t) = ρ+DC(t)/C(t).
124
125
B.1.2 Composite Good Producer’s Optimization
min
{Yˆj(t)}j∈[0,1]
∫ 1
0
Pj(t)Yˆj(t)dj
s.t.
(∫ 1
0
Yˆj(t)
σ−1
σ dj
) σ
σ−1
≥ Y (t)
Yˆj(t) ≥ 0 ∀j ∈ [0, 1]
B.1.3 Firm’s Static Optimization and Profit Function
max
Pj(t),Yj(t),Lj(t)
Pj(t)Yj(t)−W (t)Lj(t)
s.t. Yj(t) = Aj(t)Lj(t)
Pj(t)Yj(t) = Pj(t)
1−σY (t)
Lj(t) ≥ 0
Along with W (t) = σ−1σ A¯(t) and Y (t) = A¯(t), this profit maximization yields the
following expression for the profit of a firm with a productivity level a = log(A)/a˜ at t:
π(t, a) ≡ 1
σ
(
exp(aa˜)
A¯(t)
)σ−1
A¯(t).
B.1.4 Equilibrium
Definition B.1. An equilibrium given {aj(0), oj , ij(0)}j∈[0,1] is
• household composite good consumption, {C(t)}t≥0,
• wage rates, {W (t)}t≥0,
• profits, {Πd(t)}t≥0 and {Πf (t)}t≥0,
• interest rates, {r(t)}t≥0,
• composite good producer’s allocations
{
Y (t),
{
Yˆj (t)
}
j∈[0,1]
}
t≥0
,
126
• states of firms,
{
{aj(t)}t≥0 , oj , {ij(t)}t≥0
}
j∈[0,1]
,
• firms’ allocations,
{
{Yj(t), Lj(t)}t≥0
}
j∈[0,1]
, and
• prices,
{
{Pj(t)}t≥0
}
j∈[0,1]
,
such that
• Given {W (t)}t≥0, {Πd(t)}t≥0, and {r(t)}t≥0, {C(t)}t≥0 solves the problem in ap-
pendix B.1.1;
• Given
{
{Pj(t)}t≥0
}
j∈[0,1]
and {Y (t)}t≥0,
{
Yˆj (t)
}
j∈[0,1]
solves the problem in ap-
pendix B.1.2 for all t ≥ 0;
• Given
{
{aj(t)}t≥0
}
j∈[0,1]
, {W (t)}t≥0, and
{Y (t)}t≥0,
{
{Pj(t)}t≥0
}
j∈[0,1]
and
{
{Yj(t), Lj(t)}t≥0
}
j∈[0,1]
solve the problem in
appendix B.1.3 for all t ≥ 0 and j ∈ [0, 1]. (Note that Aj(t) = exp (aj (t) a˜));
• Labor market clears: ∫ 1
0
Lj(t)dj = 1 ∀t ≥ 0;
• Composite goods market clears:
C(t) = Y (t) ∀t ≥ 0;
• Markets for all varieties clear:
Yˆj(t) = Yj(t) ∀t ≥ 0 ∀j ∈ [0, 1];
• Composite good is a numeraire:
(∫ 1
0
Pj(t)
1−σ
) 1
1−σ
= 1 ∀t ≥ 0;
127
• Given {W (t)}t≥0,
{
{Pj(t)}t≥0
}
j∈[0,1]
, and
{
{Yj(t), Lj(t)}t≥0
}
j∈[0,1]
, {Πd(t)}t≥0
and {Πf (t)}t≥0 satisfy
Πd(t) =
∫ 1
0
1{oj=d} {Pj(t)Yj(t)−W (t)Lj(t)} dj ∀t ≥ 0, and
Πf (t) =
∫ 1
0
1{oj=f} {Pj(t)Yj(t)−W (t)Lj(t)} dj ∀t ≥ 0; and
• Given {aj(0), oj , ij(0)}j∈[0,1],
{
{aj(t)}t≥0 , oj , {ij(t)}t≥0
}
j∈[0,1]
evolves following
the exogenous rules outlined in sections 2.2.3, 2.2.5, and 2.2.6.
B.2 Data
B.2.1 Cross-sections into a Panel
ASIE is a collection of cross-sections. Two issues make comparisons across time prob-
lematic. First, a firm ID, which can be used to merge cross-sections into a panel, changes
over time if a firm goes through a legal status change, such as restructuring or M&A.
The ID change is not uncommon: from 1997 to 2008, about 30% of the firms experienced
it (Brandt, Van Biesebroeck, and Zhang, 2014). Second, the 4-digit Chinese Industry
Classifications went through a revision in 2003, rendering the codes incomparable across
time.
To merge the cross-sections into an unbalanced panel, the methods of Brandt et al.
(2017) were utilized. First, I adopted their matching algorithm to track firms across
time. There is one modification that I made to the original algorithm. The original uses
town name in matching key #5. Since the town name variable was not reported in 2003,
I used zip codes instead to mitigate this issue. Regarding the industry classifications,
the harmonization table provided by the authors was used.
B.2.2 Price Indices
For value-added, producer price indices (PPIs) are employed to covert nominal terms
into real ones. Unlike most firm-level datasets, in ASIE, each firm reports its output at
constant prices, the reference prices set by NBS. This variable is available for years until
128
2003. Utilizing the nominal and real outputs reported by firms, for years until 2003, PPI
for each 2-digit industry code was constructed. For years from 2004, PPIs (Ex-Factory
Price Indices of Industrial Products) reported at the Chinese Statistical Yearbook were
used. PPIs for 2004 and 2005 are from the Yearbook of 2006, and those for 2006 and
2007 are from the Yearbook of 2008.1 For wages, consumer price indices (CPIs) are
used. The CPIs for 1998 to 2007 were retrieved from the statistical database of NBS.2
Prices for 1998 were set as one for both PPIs and CPIs.
B.2.3 Patent Data
Chinese Patent Data Project (CPDP) provides ASIE firms’ patent application infor-
mation in the Chinese patent office (National Intellectual Property Administration,
CNIPA). The project collected the data by matching the assignee names of patent ap-
plications to ASIE firms’ names. For the analysis, only invention patents were used
following Holmes, McGrattan, and Prescott (2015). The authors argue that utility
patents are “of negligible importance for foreign firms.”
From CPDP, the researcher observes an ASIE sample firm and its patent applications
in China. By using Google Patents Public Data (GPPD), information on whether the
patent applications were granted and whether they were applied in different patent
offices is retrieved. Google Patents Research Data (GPRD) provides patent citation
information for ASIE firms’ patent applications in both China and other countries (to
be more precise, other patent offices).
To summarize, CPDP, GPPD, and GPRD provide ASIE firms and their patent
family information. A patent family is “a set of patents - taken in various countries -
that protect the same invention (definition from OECD).”3
GPPD and GPRD are dynamic, in the sense that they keep getting updated. There-
fore, I used the stable version of September 2019. These include patent information from
105 patent offices. In patent country codes,4 which are similar to ISO alpha-2, these
offices (countries) are AM, AP, AR, AT, AU, BA, BE, BG, BR, BY, CA, CH, CL, CN,
12006 version is at http://www.stats.gov.cn/tjsj/ndsj/2006/indexeh.htm, and that of 2008 is
at http://www.stats.gov.cn/tjsj/ndsj/2008/indexeh.htm
2http://data.stats.gov.cn/english/easyquery.htm?cn=C01
3https://www.oecd.org/science/inno/33882346.pdf
4https://www.uspto.gov/patents-application-process/applying-online/
country-codes-wipo-st3-table#heading-1.
129
CO, CR, CS, CU, CY, CZ, DD, DE, DK, DO, DZ, EA, EC, EE, EG, EM, EP (European
Patent Office), ES, FI, FR, GB, GC, GE, GR, GT, HK, HN, HR, HU, ID, IE, IL, IN,
IS, IT, JO, JP, KE, KG, KR, KZ, LT, LU, LV, MA, MC, MD, ME, MN, MO, MT, MW,
MX, MY, NI, NL, NO, NZ, OA, PA, PE, PH, PL, PT, RO, RS, RU, SA, SE, SG, SI,
SK, SM, SU, SV, TH, TJ, TN, TR, TT, TW, UA, US, UY, UZ, VN, WO (WIPO), YU,
ZA, ZM, and ZW.
The patent activities used for this research are defined as follows: An application
denotes an application in CNIPA. A grant is defined as a patent application granted by
CNIPA. (It can be defined in many different ways. For instance, an invention (patent
family) whose patent application is rejected by CNIPA can be granted a patent in other
patent offices.) A citation is defined as the number of forward citations of a patent
family in 105 patent offices. Unlike citation , a family citation counts forward citations
only once if the citing patent applications come from the same patent family. This is
called a family-to-family citation in the patent literature. By definition, extensive margin
analyses for citation and family citation are the same. For both citation and family
citation , to mitigate the time bias where older patents tend to get more citations, only
citations within ten years after the application date are counted.
As for the intensive margin analyses of patent activities, all counts are divided by
the number of assignees. For example, if a firm is assigned a patent application with
four assignees and another with three assignees in 2000, its application count in 2000
is 1/4 + 1/3.
B.3 Additional Tables and Figures
130
Table B.1: Correlations between Firm Characteristics and Log-productivity Growth
(Domestic, Joint, and Fully-foreign Firms)
Dependent variable:
lp.1999 -
lp.1998
lp.2000 -
lp.1999
lp.2001 -
lp.2000
lp.2002 -
lp.2001
lp.2003 -
lp.2002
(1) (2) (3) (4) (5)
Joint 0.180∗∗∗ 0.154∗∗∗ 0.075∗∗∗ 0.094∗∗∗ 0.048∗∗∗
(0.012) (0.011) (0.011) (0.010) (0.010)
Fully-foreign 0.141∗∗∗ 0.119∗∗∗ 0.071∗∗∗ 0.099∗∗∗ −0.009
(0.016) (0.014) (0.013) (0.012) (0.011)
Previous year log-productivity −0.345∗∗∗ −0.313∗∗∗ −0.309∗∗∗ −0.297∗∗∗ −0.319∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002)
Patent application dummy 0.342∗∗∗ 0.417∗∗∗ 0.376∗∗∗ 0.318∗∗∗ 0.328∗∗∗
(0.063) (0.053) (0.045) (0.039) (0.031)
Industry effects Y Y Y Y Y
Province effects Y Y Y Y Y
Age effects Y Y Y Y Y
Observations 123,075 124,172 122,779 136,309 148,234
R2 0.162 0.142 0.141 0.134 0.157
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Dependent variable:
lp.2004 -
lp.2003
lp.2005 -
lp.2004
lp.2006 -
lp.2005
lp.2007 -
lp.2006
(6) (7) (8) (9)
Joint 0.102∗∗∗ 0.084∗∗∗ 0.052∗∗∗ 0.048∗∗∗
(0.010) (0.009) (0.008) (0.008)
Fully-foreign 0.039∗∗∗ 0.028∗∗∗ 0.020∗∗∗ −0.029∗∗∗
(0.011) (0.008) (0.007) (0.007)
Previous year log-productivity −0.384∗∗∗ −0.411∗∗∗ −0.329∗∗∗ −0.335∗∗∗
(0.002) (0.002) (0.002) (0.001)
Patent application dummy 0.451∗∗∗ 0.425∗∗∗ 0.332∗∗∗ 0.275∗∗∗
(0.028) (0.024) (0.019) (0.016)
Industry effects Y Y Y Y
Province effects Y Y Y Y
Age effects Y Y Y Y
Observations 153,073 225,262 240,044 268,627
R2 0.180 0.215 0.174 0.179
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Standard errors are in parentheses. Previous year log-productivity is that of a starting year. For example,
in Model (1), it denotes log-productivity in 1998.
131
Table B.2: Correlations between Firm Characteristics and Log-productivity Growth
(Domestic and Foreign Firms)
Dependent variable:
lp.1999 -
lp.1998
lp.2000 -
lp.1999
lp.2001 -
lp.2000
lp.2002 -
lp.2001
lp.2003 -
lp.2002
(1) (2) (3) (4) (5)
Joint + Fully-foreign 0.169∗∗∗ 0.142∗∗∗ 0.073∗∗∗ 0.096∗∗∗ 0.022∗∗∗
(0.010) (0.010) (0.009) (0.009) (0.008)
Previous year log-productivity −0.345∗∗∗ −0.313∗∗∗ −0.309∗∗∗ −0.297∗∗∗ −0.318∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002)
Patent application dummy 0.343∗∗∗ 0.418∗∗∗ 0.376∗∗∗ 0.318∗∗∗ 0.329∗∗∗
(0.063) (0.053) (0.045) (0.039) (0.031)
Industry effects Y Y Y Y Y
Province effects Y Y Y Y Y
Age effects Y Y Y Y Y
Observations 123,075 124,172 122,779 136,309 148,234
R2 0.162 0.142 0.141 0.134 0.157
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Dependent variable:
lp.2004 -
lp.2003
lp.2005 -
lp.2004
lp.2006 -
lp.2005
lp.2007 -
lp.2006
(6) (7) (8) (9)
Joint + Fully-foreign 0.071∗∗∗ 0.053∗∗∗ 0.034∗∗∗ 0.002
(0.008) (0.006) (0.006) (0.005)
Previous year log-productivity −0.384∗∗∗ −0.411∗∗∗ −0.329∗∗∗ −0.334∗∗∗
(0.002) (0.002) (0.002) (0.001)
Patent application dummy 0.453∗∗∗ 0.427∗∗∗ 0.333∗∗∗ 0.278∗∗∗
(0.028) (0.024) (0.019) (0.016)
Industry effects Y Y Y Y
Province effects Y Y Y Y
Age effects Y Y Y Y
Observations 153,073 225,262 240,044 268,627
R2 0.180 0.215 0.174 0.179
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Standard errors are in parentheses. Previous year log-productivity is that of a starting year. For example,
in Model (1), it denotes log-productivity in 1998.
132
Table B.3: Correlations between Firm Characteristics and Log-productivity Level (Do-
mestic, Joint, and Fully-foreign Firms)
Dependent variable:
lp.1998 lp.1999 lp.2000 lp.2001 lp.2002
(1) (2) (3) (4) (5)
Joint 0.455∗∗∗ 0.432∗∗∗ 0.414∗∗∗ 0.359∗∗∗ 0.346∗∗∗
(0.014) (0.014) (0.014) (0.013) (0.013)
Fully-foreign 0.246∗∗∗ 0.172∗∗∗ 0.188∗∗∗ 0.131∗∗∗ 0.156∗∗∗
(0.019) (0.018) (0.017) (0.015) (0.014)
Patent application dummy 1.274∗∗∗ 1.224∗∗∗ 1.284∗∗∗ 1.258∗∗∗ 1.105∗∗∗
(0.080) (0.070) (0.058) (0.052) (0.041)
Industry effects Y Y Y Y Y
Province effects Y Y Y Y Y
Age effects Y Y Y Y Y
Observations 148,404 147,840 150,480 158,543 170,327
R2 0.106 0.109 0.118 0.117 0.117
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Dependent variable:
lp.2003 lp.2004 lp.2005 lp.2006 lp.2007
(6) (7) (8) (9) (10)
Joint 0.289∗∗∗ 0.336∗∗∗ 0.301∗∗∗ 0.297∗∗∗ 0.274∗∗∗
(0.012) (0.011) (0.011) (0.010) (0.010)
Fully-foreign 0.075∗∗∗ 0.054∗∗∗ 0.063∗∗∗ 0.086∗∗∗ 0.040∗∗∗
(0.012) (0.010) (0.009) (0.009) (0.008)
Patent application dummy 1.099∗∗∗ 1.022∗∗∗ 1.044∗∗∗ 0.920∗∗∗ 0.850∗∗∗
(0.034) (0.029) (0.026) (0.021) (0.019)
Industry effects Y Y Y Y Y
Province effects Y Y Y Y Y
Age effects Y Y Y Y Y
Observations 190,048 266,629 264,513 294,889 330,385
R2 0.126 0.098 0.115 0.140 0.154
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Standard errors are in parentheses.
133
Table B.4: Correlations between Firm Characteristics and Log-productivity Level (Do-
mestic and Foreign Firms)
Dependent variable:
lp.1998 lp.1999 lp.2000 lp.2001 lp.2002
(1) (2) (3) (4) (5)
Joint + Fully-foreign 0.390∗∗∗ 0.342∗∗∗ 0.331∗∗∗ 0.265∗∗∗ 0.262∗∗∗
(0.013) (0.012) (0.012) (0.011) (0.010)
Patent application dummy 1.280∗∗∗ 1.233∗∗∗ 1.289∗∗∗ 1.265∗∗∗ 1.109∗∗∗
(0.080) (0.071) (0.058) (0.052) (0.041)
Industry effects Y Y Y Y Y
Province effects Y Y Y Y Y
Age effects Y Y Y Y Y
Observations 148,404 147,840 150,480 158,543 170,327
R2 0.105 0.108 0.117 0.116 0.116
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Dependent variable:
lp.2003 lp.2004 lp.2005 lp.2006 lp.2007
(6) (7) (8) (9) (10)
Joint + Fully-foreign 0.185∗∗∗ 0.183∗∗∗ 0.164∗∗∗ 0.172∗∗∗ 0.129∗∗∗
(0.009) (0.008) (0.008) (0.007) (0.007)
Patent application dummy 1.107∗∗∗ 1.032∗∗∗ 1.054∗∗∗ 0.931∗∗∗ 0.862∗∗∗
(0.034) (0.029) (0.026) (0.021) (0.019)
Industry effects Y Y Y Y Y
Province effects Y Y Y Y Y
Age effects Y Y Y Y Y
Observations 190,048 266,629 264,513 294,889 330,385
R2 0.125 0.096 0.114 0.140 0.153
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Standard errors are in parentheses.
134
Table B.5: Patenting Behavior (Intensive Margin)
(a) Average of firm-years’ patent activity counts
Application Grant Citation Family citation
Domestic 0.0414 0.0278 0.2610 0.1797
(0.0055) (0.0042) (0.0529) (0.0328)
Joint 0.1367 0.0684 0.5356 0.3759
(0.0236) (0.0119) (0.0990) (0.0658)
Fully-foreign 0.0816 0.0480 0.5327 0.3587
(0.0085) (0.0057) (0.0751) (0.0483)
Joint + Fully-foreign 0.1070 0.0574 0.5340 0.3666
(0.0118) (0.0063) (0.0610) (0.0399)
(b) Correlations between ownership and patent activity counts
Dependent variable:
Application Grant Citation Family citation
(1) (2) (3) (4)
Joint 0.007∗∗∗ 0.002 0.084 0.073
(0.002) (0.001) (0.162) (0.101)
Fully-foreign −0.005∗ −0.004∗∗ −0.454∗∗ −0.293∗∗
(0.002) (0.001) (0.160) (0.100)
Industry effects Y Y Y Y
Province effects Y Y Y Y
Age effects Y Y Y Y
Year effects Y Y Y Y
Observations 2,216,697 2,216,697 2,216,697 2,216,697
R2 0.0004 0.0004 0.0003 0.0003
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Standard errors are in parentheses. In panel (a), joint + fully-foreign denotes firms in either joint or
foreign category.
135
Table B.6: R&D Behavior by Ownership Category
(a) Average of R&D variables
R&D dummy R&D intensity
Domestic 0.0945 0.0088
(0.0003) (0.0005)
Joint 0.1687 0.0143
(0.0015) (0.0012)
Fully-foreign 0.0911 0.0092
(0.0009) (0.0018)
Joint + Fully-foreign 0.1199 0.0111
(0.0008) (0.0012)
(b) Correlations between ownership and R&D
variables
Dependent variable:
R&D dummy R&D intensity
(1) (2)
Joint 0.067∗∗∗ 0.002
(0.001) (0.002)
Fully-foreign −0.009∗∗∗ −0.004∗∗
(0.001) (0.002)
Industry effects Y Y
Province effects Y Y
Age effects Y Y
Year effects Y Y
Observations 902,990 902,990
R2 0.067 0.002
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Standard errors are in parentheses. R&D intensity is nominal R&D expenditure divided by nominal
value-added. In panel (a), joint + fully-foreign denotes firms in either joint or foreign category.
136
Figure B.1: Relative Frequencies of ASIE Firm-years’ Patent Activity Counts
0.0
0.1
0.2
0.3
0.4
0.5
0 10 20 30
Number
R
el
at
ive
 fr
eq
ue
nc
y
Variable
Application
Grant
Citation
Family citation
The figure shows the distributions of patent activity counts for participating firm-years. For example,
the distribution for patent application only includes firm-years that applied for at least one patent. The
bin width for each distribution is one.