2,

.
I.
r

L
«

Yd «.wat a“

:1 ‘ i.

x,
1101"... .t
L... .63

:5 ‘.

. a
.3? x. h‘
x. ..
5.4“.“
J.
at...
.33... ...
.A _..

or."-
.. 1mm
4.4..
.1. .7!
3.2» J
, ‘|
ummguuf
w
can

.. z.-
‘ ii! i .
fit

... .
or: q .. .r.
.. rattan 1..
31%....Qfi
X c
.51
‘-

{A1
.1!
“V..-
2. 1|... .
biz.
1.151..
..
VA" «1
EY. A

. 13
. r I

£3“? .
. . afar... :c,

L: ..
.

.n ,
a...

.1... .2. .
r a £9.
. _

s:

 

 

UBRARY
Michigan State
brriversity

 

 

 

This is to certify that the
dissertation entitled

INNOVATION, NETWORK EXTERNALITY, AND
INTERNATIONAL TRADE

presented by

WEI-CHIH CHEN

has been accepted towards fulfillment
of the requirements for the

Doctoral degree in Economics

 

 

Major Professor’s Signature
é-Zé-cj

Date

 

MSU is an Affirmative Action/Equal Opportunity Employer

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 KzlProlecc&PrelelRC/DateDue.indd

 

INNOVATION, NETWORK EXTERNALITY, AND INTERNATIONAL TRADE
By

Wei—Chih Chen

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment Of the requirements
‘ for the degree of

DOCTOR OF PHILOSOPHY
Economics

2009

ABSTRACT
INNOVATION, NETWORK EXTERNALITY, AND INTERNATIONAL TRADE
By

Wei—Chih Chen

The dissertation consists of three essays exploring the effects of innovation on the
extensive margin, the intensive margin, and the duration of exports, as well as the
influence of network externality and products compatibility on firrns’ FDI decisions.

An important prediction of international trade theories is that countries which
innovate more will export more. However, these theories have very different implica-
tions for how innovation would increase exports. In the first chapter, I empirically
investigate the extent to which innovation increases the number of products (the ex-
tensive margin) and the export value of each product (the intensive margin). Using
data on patents granted by the United States and data. on manufacturing exports
from 105 countries to the US. market over the period 1975-2001, I find that innova-
tion has positive and significant effects on both the extensive and intensive margins.
The intensive margin contributes about 70% of the effects. and the extensive margin
accounts for the remaining 30%. The estimated results indicate that the effect of in—
novation on exports is stronger in low-income countries than in high—incmne countries.
It. is also stronger in industries which have relatively more differentiated products than
homogeneous products. Finally, more innovative countries export larger quantities
at higher prices, suggesting that innovation increases the product quality of exports.

The second chapter continues studying the impact of innovation on exports, but

the focus is switched to the dynamic pattern of trade. I use the survival analysis to
investigate the effect of innovation on the duration of exports at the product level.
The estimated results show that countries which have more innovations can sustain
their export relationships for a longer period of time. The findings are consistent.
with the implication of the product cycle model, which stresses the role of innovation
on dynamic trade patterns. The estimates also display that the duration of exports '
increases with country size and decreases with trade costs and barriers.

Traditional trade theories explaining different FDI patterns across firms and sec-
tors seek answers from the production side, such as the proximity-concent ration trade—
off and the factor-proportions hypothesis. The third chapter provides a new expla-
nation for FDI (foreign direct investment) patterns from demand-side properties. I
develop a differentiated-product duopoly model to analyze the effects of network ex-
ternality (demand-side economies of scale) and product compatibility on firms’ F DI
decisions. The model is able to explain the following facts in FDI. First, firms produc-
ing compatible products are more likely to undertake FDI in the same location than
firms producing incompatible products. Second, FDI is more prevalent in industries
with strong network externality effects. This explains why FDI is more important
in high technology industries. Third, a firm is more likely to undertake F DI if it
has a larger installed base (more existing customers) in the foreign market. This is
consistent with the fact. that larger firms or firms with longer history are more likely

to undertake FDI.

To my parents

iv

ACKNOWLEDGMENTS

First of all, I would like to express the deepest appreciation to my advisor, Pro-
fessor Susan Chun Zhu, for her encouragement and patience throughout my doctoral
study. It would have never been possible to complete this dissertation withtmt
her guidance and help. I am grateful to my committee members, Professor Steven
l\-*Iatusz, Professor Thomas Jeitschko, Professor Jay Pil Choi, and Professor Charles
Hadlock. Their insightful questions and valuable comments make this dissertation
more complete. I am indebted to many of my classmates and friends, especially
Meng—Chi Tang and Yi Zhu. The time we spent together make my five years in
Michigan colorful. Finally and most importantly, I would like to thank my parents
and Chia—Ning for their unconditional love and support. They are the people who

make it all worthwhile.

Ill

 

TABLE OF CONTENTS

LIST OF TABLES ............................................................................... vii
LIST OF FIGURES ................................................................................ ix
Chapter 1: The Extensive and Intensive Margins of Exports: The Role of Innovation
1. Introduction ................................................................................ 1
2. Theoretical Framework ................................................................... 6
3. Data ......................................................................................... 10
4. The Extensive and Intensive Margins: Empirical Models and Estimated
Results ...................................................................................... 17
5. Decomposing the Intensive Margin: Innovation and Product Quality...39
6. Conclusion ................................................................................. 44
Chapter 2: Innovation and Export Duration
1 . Introduction .............................................................................. 65
2. Motivation ............................................................................... 67
3. Data ....................................................................................... 69
4. Models of Estimation and Estimated Results .................................. 74
5. Conclusion ................................................................................ 84
Chapter 3: Network Externality, Compatibility, and Foreign Entry Mode
1. Introduction ........................................................................... 104
2. Benchmark Model: No Installed-Base ........................................... 108
3. Model with an Installed-Base ..................................................... 122
4. Conclusion .............................................................................. 128
APPENDICES
A. Innovation Data ........................................................................ 137
B. Data of Factor Endowment and Factor Intensity ............................ 140
C. Data of Price and Quantity in Exports ....................................... 141
D. Kaplan-Meier Estimator ............................................................. 143
E. Equilibrium Outcomes of the Price Game ..................................... 144
F. Proofs of Comparative Static Effects ............................................ 150
REFERENCES ................................................................................... 154

vi

1.1

1.2

1.3

1.4

1.5

1.6

LIST OF TABLES

List. of Countries ............................................................................. 47
List of Industries in 2-digit SIC ............................................................ 48
Estimated Results: Two-way Fixed Effects Model .................................... 49
Estimated Results: Three-way Fixed Effects Model .................................. 50
Estimated Results: Interaction between GDP per capita and Patent..............51

Elasticities of Exports with respect to Patents in Countries with Different GDP
per capita ...................................................................................... 52

1.7 Estimation by Industry Group: Differentiated Products versus Homogeneous

Products ........................................................................................ 53
1.8 Elasticities of Experts with respect to Patents by Industry Group .............. 54
1.9 Estimation by Year: Patents Classified according to the Application Year ...... 55
1.10 Estimation by Year: Patents Classified according to the Grant Year .............. 56

1.11 GMM Estimator: Persistency in Exporting and Endogeneity of Innovation....5

H
l

1.12 Robustness Checks: Two-VVay Fixed Effects ......................................... 58
1.13 Robustness Checks: Three-Way Fixed Effects59
1.14 Decomposing the Intensive Margin into Price and Quantity ........................ 60
2.1 Example of Export Spells ................................... 86
2.2 List of Countries .............................................................................. 87
2.3 List of Industries in 2-digit SIC 87 ....................................................... 88

vii

2.4 Survivor Function (1975—1988) ........................................................... 89

2.5 Survivor Function (1989-2001) ........................................................... 90
2.6 Summary of Statistics: Estimated Duration ............................................... 91
2.7 Percentile of Durations ..................................................................... 92
2.8 Summary of Statistics: Regressors ................. . ...................................... 93
2.9 Estimated Results of the Benchmark Model ............................................ 94
2.10 Estimated Results: Beta Coefficients ..................................................... 95
2.11 Robustness Checks: First-spell only ...................................................... 96
2.12 Robustness Checks: SITC Industry ....................................................... 97
2.13 Robustness Check: Variables in Natural Logs .......................................... 98
2.14 Robustness Check: Including Factor Endowment ....................................... 99
2.15 Estimated Results: Number of Exporters 1975-1988 ................................. 100
2.16 Estimated Results: Number of Exporters 1989—2001 ................................. 101

3.1 Equilibrium price, market share and profit of each entry mode in the
no—installed—base case ....................................................................... 131

3.2 Equilibrium price, market share and profit of each entry mode when firm A
has an installed base ........................................................................ 132

3.3 Comparative statics effects ................................................................ 133

viii

1.1

1.2

1.3

1.4

2.1

2.2

3.1

3.2

3.3

LIST OF FIGURES

The Elasticities of Export with Respect to Innovation by Country Income
Level ............................................................................................. 61

The Elasticities of Export. with Respect to Innovation: Differentiated vs

Homogeneous Products ..................................................................... 62
The Elasticities of Export with Respect to Innovation by Year ..................... 63
The Elasticities of Export with Respect to Innovation by Period ................... 64

The survivor function of TS7 products in 1975-1988 using the Kaplan—Meier
estimator ...................................................................................... 102

The survivor function of H810 products in 1989-2001 using the Kaplan-Meier

estimator ...................................................................................... 103
Nash equilibrium of the entry game of the no—installed-base case ................. 134
Nash equilibrium of entry mode when the installed base is small ............... 135
Nash equilibrium of entry mode when the installed base is large .................. 136

ix

Chapter 1

The Extensive and Intensive Margins of
Experts: The Role of Innovation

1 Introduction

An important prediction in international trade theories is that innovation improves
countries’ export performance. However, these theories differ in their predictions
about. how innovation increases exports. The first strand of the literature predicts
that innovation can introduce new products to the market or expand the range of
products that a country exports (Krugman, 1979; Jenson and Thursby, 1986; Dollar,
1986; Grossman and Helpman, 1989). These studies emphasize the role of innova—
tion in increasing the number of products, which will increase the extensive margin
of exports. The second strand of the literature stresses that innovation can increase
productivity (Eaton and Kortum, 1997a, 1997b, 1999, 2001, 2002) or improve product
quality (Flam and Helpman, 1987; GrOssman and Helpman, 1991a). Innovation en-
hances countries’ competitiveness and leads to a higher export value of each product,
which will increase the intensive margin Of exports.

Although the effect of innovation on trade has been examined extensively in the
theoretical literature, very little empirical work has been done to find the evidence for
the role of innovation in affecting export performance for a wide range of countrits
and. industries. Inspired by these trade theories, in this chapter I will assess the

importance of innovation in determining export performance using a sample of 105

countries and 12 manufacturing industries. My work is also motivated by the firm-
level studies that find ample evidence that good firms export, but scant evidence for
the learning-through-exporting effect (Bernard and Jensen, 1999; Clerides, Lach, and
Tybout, 1998; and Tybout, 2001). However, these firm-level studies mostly focus on
the US. or a few developing countries (e.g. Mexico and Columbia). It is unclear
whether the positive effect of productivity and innovation exists for a wider range of
countries. In addition, my work is complementary to the literature that focuses on
the effect of international trade on technology diffusion and productivity (Coe and
Helpman, 1995; Keller, 1998; Xu and Wang, 1999).

The main object of this chapter is to estimate the effect of innovation on exports
and to decompose the effect into the extensive margin and the intensive margin. In
empirical trade literature, the extensive margin represents the width of exports, and
the intensive margin represents the depth of exports. In this chapter, the extensive
margin is measured by the number of products exported from a country, and the
intensive margin is measured by the export value of each product from a country. I
calculate the extensive and intensive margins of exports using detailed data on 105
countries’ exports to the US. market over the period 1975-2001, and use the data
on patents granted by the US. as a proxy for innovation. My estimates show that.
innovation has a positive and significant effect on both the extensive margin and the
intensive margin, and the intensive margin accounts for a larger proportion of the
impact. This result holds for various specifications. In the benchmark estimation,
a cormtry with 10% more patents will export 5.4% more to the US. The effect

can be decomposed into 1.6% higher extensive margin and 3.8% higher intensive

2

margin. The intensive margin accounts for 70% of the greater exports from more
innovative countries, and the extensive margin accounts for the remaining 30%. I
further decompose the intensive margin into the price and quantity indices. and find
that more innovative countries export higher quantities at slightly higher prices, which
suggests that innovation increases the quality of exports.

The decomposition of innovation’s effect has important welfare implications. An
increase in the extensive and intensive margins may lead to different terms—of-trade
effects. If innovation increases the number of products in exports (an increase in
the extensive margin), the terms—of—trade move in the innovating country’s favor
and innovation unambiguously increases the welfare of the innovating country. If
innovation makes countries export more of each product (an increase in the intensive
margin), the terms-of—trade effects can be ambiguous or even negative. The estimates
of this paper suggest that more innovative countries export more varieties of products
and have higher export prices. Thus, my results imply that innovation improves the
innovating countries" welfare not only by increasing the number of varieties, but also
by improving their terms of trade.

In this paper I also investigate how the effect of innovation on exports differs
across countries, industries, and years. First, North—South trade models (e.g. Krug—
man, 1979; Flam and Helpman, 1987) show that technological progress may impose
asymmetric impacts on developed and developing countries. I include the interaction
between patent counts and GDP per capita to allow innovation’s effect to vary with
countries’ income levels. The estimates show that the impact of innovation on exports
is stronger and the contribution of the extensive margin is also larger for lower-income

3

countries. This result is different from existing literature which largely makes a sim-
plified assumption that deveIOping countries do not innovate. My estimates suggest.
that innovation is an important factor of developing countries’ technology catch—up.
Second, the effect of innovation may be larger for industries with a higher share of
differentiated goods, because innovation should be more important in the production
of differentiated products than that of homogeneous products. Using Rauch’s (1999)
classification of differentiated and homogenous products to categorize industries, my
estimates reveal that innovation has a stronger impact on exports in industries with
a higher share of differentiated products, such as instruments, electronics, and trans-
portation equipment. Third, the effect of innovation and the composition of the
effect. may change over time. The year-by-year estimation shows that the effect of
innovation decreases over time, which is mainly due to a decline in the contribution
of the extensive margin.

The method Of decomposition used in this paper is closely related to Hununels
and Klenow (2005). They analyze the relationship between country size (measured
by GDP per worker and employment) and the extensive, intensive, and quality mar-
gins of exports. They find that the extensive margin accounts for 60% of the greater
exports from larger economies, and the intensive margin is dominated by higher quan-
tities rather than higher unit values. My paper extends their work in the following
ways. First, in addition to country size, I include innovation as an important fac-
tor determining countries’ trade performance in different margins. My estimated
results show that innovation has significantly positive impacts on different margins
of exports, even after controlling for the country size and income level. Second, I

4

construct the indices of export performance at the industry level, while Humrnels and
Klenow (2005) construct these indices at the country level. Using industry-level trade
data enables me to compare the effect of innovation on exports in different sectors.
My results show that the impact of innovation on overall export performance and the
relative contributions of the extensive and intensive margins differ significantly across
sectors. Finally, I use panel data which contains 105 countries’ exports to the US.
from 1975 to 2001, while Hummels and Klenow (2005) use a cross—sectional sample for
the year 1995. Using the panel data allows me to control for the unobserved country
and industry fixed effects, and to compare the change in the effects of innovation over
time. Furthermore, using the panel data allows me to estimate a dynamic model
that. takes into account the persistence of exporting (Roberts and Tybout, 1997).
This paper also contributes to the following two strands Of empirical literature.
The first strand of literature looks into the composition of trade. Tiaditional in-
ternational trade studies usually focus on the total value of trade between countries.
However, recent research looks more deeply into the composition and the different
dimensions of trade, such as the number of varieties, the volume of exports, the
quality of exports, and the duration of export relationships (Funke and Ruhwedel,
2001; Hummels and Klenow, 2005; Schott, 2004; Hallak, 2006; Besedes and Prusa,
2007). Many variables (such as the country size, growth rate, income level, and dis-
tance) have been linked with these different margins in experts, but the impact of
innovation on the extensive and intensive margins has not been explored yet. The
second strand of literature examines the correlation between innovation and exports

empirically. These studies find a positive correlation between innovation and export

5

performance either within a country (Gruber et. al., 1966; Keesing, 1967) or across
countries at the aggregate level (Soete, 1987; Fagerberg, 1988, 1996; VVakelin, 1998).
However, they do not estimate in which margin innovation has the largest impact,
which is the focus of my paper.

The remainder of this chapter is organized as follows. Section 2 provides a
theoretical background of the empirical model. Section 3 describes the data source
and the construction of both the exports and innovation variables. Section 4 shows
the models of estimation and presents the estimated results. Section 5 decomposes

the effect on the intensive margins into price and quantity. Section 6 concludes.

2 Theoretical Framework

This section reviews two strands of trade theories which motivate the empirical esti-
mations in this paper. These papers stress the role of innovation and technological
progress on exports. Nevertheless, they differ in the predictions about how innovation
affects exports.

To help explain the impact of innovation on the different margins of exports,
consider the following simple model. Suppose consumers in the US. can purchase
products from C countries. Each country has a large number of firms producing
different. products. Consumers have a CBS utility function over products. They
face the following maximization problem

0—1 a

maxch U = (29:17169qquyv—4 (1)

st. EEZI'ncpcj-ch = I (2)

where subscription e is country, j is product variety, nc is the number of products
produced by country c, dcj is the quality of product j from country c, M is the
export quantity of product j from country c, and pcj is the unit price of product j
from country c. 0 > 1 is the elasticity of substitution between any two products,
and I is consumers’ income levels. Equations 1 and 2 imply that firms in the same
country are symmetric: they produce products of equal quality (6,33- : 0(3) and charge
the same price (pg 2 pg). By solving the maximization problem in equation 1 and

2, I get the export value of each product from country c

chc = (176)1—09gq) (3)

where (I) = 2921(ncp61—063). Consumers buy every variety produced in country c,

and the export value of country c is

"CPCQC (4)

This total value of exports can be decomposed into the extensive margin and the
intensive margin. The extensive margin measures the width of export. In equation
4, it is captured by nc, which is the number of products exported by country c.
The intensive margin measures the depth of export. In equation 4, this margin is

measured by pcqc, which is the export value per product from country c. In equation

3, quality works as a demand shifter, and the intensive margin of exports is increasing
with product quality.

In trade literature, innovation is predicted to have important influence on both
the extensive margin and intensive margin.

(1) The extensive margin

The link between innovation and the extensive margin is introduced by theories
like Krugman (1979), Dollar (1986), Jensen and Thursby (1986), and Grossman and
Helpman (1989). These studies emphasize the role of innovation in increasing exports
by creating new products or expanding the range Of products exported. For instance,
Krugman (1979) develop a North-South trade model in which technological progress
determines exports. In equilibrium, the relative number of products a. country exports
is increasing with its rate of technological progress. More innovations enable a country
to export a wider range of goods, which increase the extensive margin of exports. In
terms of equation 4, innovation will increase exports by increasing nc.

(2) The intensive margin

Two strands of trade literature stress the ability of innovation in increasing the
intensive margin of exports. First, innovation can raise product quality, which in-
creases consumers demand of these goods. Grossman and Helpman (1991a) develop
a quality-ladder model in which successful innovations can improve quality of prod—
ucts. Other studies such as Flam and Helpman (1987) and Grossman and Helpman
(1991b) also assume that innovation is an important determinant. of exports qual-
ity. In equation 3, the intensive margin (pcq’c) increases with quality (06-), and thus
a higher quality leads to a higher intensive margin of exports. Second, innovation

8

can improve production efficiency. Eaton and Kortum (1997a, 1997b, 1999, 2002)
develop a series of Ricardian models which link innovation, technology, and trade.
Eaton and Kortum (2001) provide a unified framework in which innovation takes the
form of extending the technology frontier and raising productivity. Higher productiv-
ity will reduce the production cost and decrease the export price pc. From equation
3, the export value of each product decreases with price, as long as the elasticity of
substitution 0’ > 1. This means an increase in productivity will reduce the unit
price and increase the intensive margin of exports. In summary, these two strands
of studie. predict that innovation enhances producers’ competitiveness by improving
product quality or raising productivity, which leads to higher intensive margins of
exports (pcqc).

According to these theories, if innovation increases the number of products, we
expect to find a. positive correlation between innovation and the extensive margin of
exports. If innovation increases the quality of products or productivity, we expect
innovation to be positively correlated with the intensive margin of exports. Despite.
the various predictions of theoretical works, empirical studies have not, to the best
of my knowledge, tried to estimate and compare the relative importance and contri-
bution of the different margins of exports regarding innovation’s influence, and this

motivates the estimations in this paper.

3 Data

An empirical study which estimates the relation between innovation and export per-
formance requires data that fall into two categories. The first. is a. variable which
measures countries’ innovations or technological changes. The second is detailed
information of exports which can be decomposes into the extensive margin and the
intensive margin. Variables of a gravity equation, such as country size, distance,
language and border are also included in my estimation. This section describes the

data source and the construction of both the exports and innovation variables.

3. 1 Innovation Data

The key explanatory variable in this paper is each country’s innovation at the industry
level. Patents granted by the US. are used as the proxy for innovations. The patent
data come from Hall, Jaffe and Trajtenberg (2001).1 The data comprise detailed
information on about 3.5 million patents granted between 1963 and 2002 by the
United States Patent and Tiademark Office (USPTO).

In economic literature, a patent is an indicator of an innovation or invention. Some
studies find a. positively correlation between patents and new products, while others
associate patents with the competitiveness or comparative technology advantage (see
a. survey by Griliches, 1990). These works show that patent count is a valid proxy
for innovation which covers the concepts of both product and process innovations in

the trade theories discussed in this paper. A potential problem is that patents vary

 

lData after 1999 is updated by Hall.

10

enormously in their technological and economic significance, and thus simple patent
counts cannot precisely measure innovative outputs. It has been shown in empirical
studies that patent counts weighted by citations are more closely associated with
their value and importance (Trajtenberg, 1990; Jaffe et al., 2000; Hall et al., 2000,
2001). In this paper, patents are weighted by the number of citations received, and
the weighted patent counts (WPC) are used to assess the innovations or inncwative
capabilities of countries. The linear weighting scheme which follows rI‘rajtenberg
(1990) is applied to construct the weighted patent count index:

I.

WPCm-t = 232331 + 9,) (5)

where subscription 0 is country, i is industry, t is year, "cit is the number of patents
issued to country c in industry i during year t, and gj is the number of citations
received by patent j. The natural log of the weighted patent count in equation 5 is
the key regressor in this paper.

Except patent, R&D expenditure is also used as a proxy for innovation in trade
literature. I use patent counts for the following considerations. First, patents are
the output of innovative activities, but R&D expenditure is the input. In trade the-
ories, it is the results of innovation rather than the efforts of innovation that improve
countries’ export performance. Second, the time that an innovation is created is
more accurately measured by the application year of patents than by the year of R&D

0Xpenditures. There is a. time lag between R&D expenditure and innovation output.

Tlllrd, patent data. from a single source can reflect. countries’ innovative capabilities

11

on a relatively comparable base.2 R&D expenditures from different countries can
have different measurements and efficiency. The patent data from USPTO cover
a wider range of countries in a longer time span (1963-2002). On the contrary,
industry-level R&D data are available only among a limited set of countries (such as
OECD member countries), and the time span is shorter. Using R&D expenditure
as the proxy for innovation will restrict the scope of this paper in a small group of
countries with similar properties. Another drawback of using R&D expenditure is
that. it tends to underestimate the innovation contribution of small firms which do
not. have a separate research department.3

Two points should be emphasized when using weighted patent counts as the proxy
for innovation. First, there is no natural scale or value measurement for patents and
citations.4 The number of patents granted and citations made depend largely on the
examination process of Patent Office, which can be different by year and industry.
This issue can be solved by including industry and year fixed. effects in the regressions,
which is a method similar to the fixed-effects approach used by Hall, et. al. (2001)
to compare the citation intensity of patents from different cohorts. Second, each

patent has two indices of time: the application year and the grant year. I classify

 

2I use patents granted by the US. rather than other countries, because intuitively US. is the
technologically most advanced country in the world. It also has the largest market, which should
attract most inventors in the world to apply for patents from (see Mahnood and Singh, 2003).
Empirically, Soete (1987) finds that a patent granted by US. is indeed a superior measurement
Of international innovation than a patent granted by other countries like Japan, France, U.K. and
Germany.

3W'akelin (1998) uses R&D expenditure and patent numbers separately as proxies of relative in-
novation performance, and she finds that the patent proxy captures it better than R&D expenditure,
especially in high technology industries.

4For example, it is not clear whether a country having 100 patents in the year 1990 is more or
less innovative than a country having 300 patents in the year 2000.

12

patents according to the application year because innovation outcomes are reflected
on the year patents are applied rather than the year they are granted. Hmvever,
patent counts are truncated when classified by the application year, especially in the
last few years of the data. Hall, et. a1. (2001) show that the average lag between
the application and grant year in the sample is 1.97 years, and more than 95% of
patents are granted within three years after they are applied. For the time being, I
drOp observations of the last three years (2000—2002) from the data to mitigate this
problem. The truncation issue will be addressed in section 4.4 with a year-by-year
estimation. More details about the patent data and the construction of weighted

patent counts are discussed in appendix A.

3.2 Export Data

The data of exports at the product level for 1975-2001 are constructed by Feenstra
(1996) and Feenstra, Romalis and Schott (2002). The data. contain the quantity and
value of each product that US. imports from other countries. Products are classified
according to the 7—digit Tariff Schedule of the United States Annotated (TS7) in
1975—1988, and are classified according to the 10—digit Harmonized System (H810)
in 1989-2001.5 The data also include US. Standard Industrial Code (SIC) of each
product variety, which enables me to identify the extensive and intensive margins of

exports at the industry level.

 

'3 These are very disaggregated data sets. In 1988, there are 11,198 commodities classified in TS7
exported to the U.S., and in 2001 there are 16,293 commodities based on HS10.

13

3.3 Method of Export Decomposition

To find out how innovation affects exports, I decompose trade values into the extensive
and intensive margins. The decomposition method is similar to Hummels and Klenow
(2005), but. I calculate the two margins of each country at the industry level, instead
of at the country level as in Hummels and Klenow (2005).

To suppress the notation, consider export data of products in the same industry
and year. In equation 4, the export value of country c is ncpcqc. The empirical
counterpart of country c’s export value is made relative to the world export value
(i.e., the total export value of all products from all countries in the sample), and so

country c’s overall export share is

EjES'C (pCjQCj)
EjES (pijwj)

 

(6)

erportc 2

Subscription c is country, j is product variety defined by TS7 or HS10, w indicates
world which represents all countries in the sample. pchcj is the export value of
product j that. country c exports to the US. p1,, jqwj = Ecpchcj is the world export
value of product j to the US. S is the set of products that the world exports to
the US. Sc is a subset of S in which country c has positive exports to the US. (
pchcj > 0). The sum of all countries’ export shares equals one (Ecerportc = 1).
The empirical counterpart of the extensive margin is an index which measures the
relative number of products a country exports. If each product has a equal weight,
the extensive margin is the fraction of products in which country c has positive exports

(i.e., -nC/nw,where he is the number of products exported by country c (j E 5,), and

14

nu, is the total number of products exported to the US. from all countries (j E S )).
If each product is weighted by its world export value, pquwj, country c‘s extensive

margin of experts will be

2IJ’jES'c (pqule)
szS (pijwj)

 

(7)

ertensivec :

The extensive margin increases with the number of products that country c exports
(j 6 SC).

The empirical counterpart of the intensive margin in equation 4 is defined as

Ejesc (chQCj)
ZjESC (puijqwj)

 

(8)

intensiveC =

It equals the ratio of country c’s export value relative to world’s export value of those

products that country c has positive exports.6

From equations 6, 7 and 8, country c’s export share equals the product of its

extensive margin and intensive margin:

Ejesc (page) = EjESC (ij‘le) X 2has (peg-qty) (9)
£3ng (pquwj) Ejes (ijqwj) ZjGSc (ijqwj)
\_q,__/ \—v——’ hw_—/

 

 

 

erportc err.ensiveC intensivec

Equation 9 indicates that the overall export share can be decomposed linearly into

 

" As the export share, the extensive margin and intensive margins are smaller than one. Nev-
ertheless, the sum of extensive and the sum of intensive do not equal 1. Consider an extreme
case: if every country exports all varieties to the US, extensive = 1 for all e, and thus the sum of
e.r:tensive 75 1.

15

the extensive and intensive margins after taking natural logs:

ln erportc = ln ertensivec + 1n intensivec (10)

In the estimations of this paper, I construct the logs of overall export, the extensive
margin, and the intensive margin of each country—industry-year according to etpiations
6, 7 and 8. These three variables will be the dependent variables in the estimations.
An advantage of decomposing exports as in equation 10 is that when I regress each
margin on the same regressors using a linear estimator (such as OLS), the estimated
coefficients of each regressor will also have a linear relation: the coefficient of the
overall export regression equals the sum of the coefficient of the extensive margin
regression and that of the intensive margin regression. I can estimate the relative

contribution of each margin regarding the impact of each regressor.

3.4 Other Controls

Five other controls are included alongside with the patent data in the benchmark
estimation. Population and GDP per capita are included to control for country size.
The data. are from Penn World Table 6.2. I also include distance, common language,
and common border in the estimation, which represent trade costs and resistances.

These data are from Jon Haveman’s International Trade Data.

16

3.5 The Sample

The complete sample of my empirical analysis contains observations of 105 countries
in 12 manufactured industries (2-digit SIC) over the period 1975-2001. Lists of these
countries and industries are in tables 1.1 and 1.2. Since the focus of this paper is
to analyze the effect of innovation on different margins of exports, the sample only
contains country-industry—years which have positive export flows to the US. The
complete sample is thus an unbalanced panel which has 27,450 observations. To
mitigate the truncation problem of patent counts as mentioned earlier, observations
after 1999 are dropped in the benchmark estimations, and the sample used in the

benchmark estimations has 25,268 observations.

4 The Extensive and Intensive Margins: Empirical

Models and Estimated Results

4.1 Benchmark Estimation Model

The model of estimation in this section is an extension to Hummels and Klenow’s
(2005) model in the sense that innovation is considered as an additional and important
determinant of export performance. The motivation comes from the theoretical
literature reviewed in section 2. Nevertheless, the object of the empirical work here
is not to formally test each model. These theoretical models are polar cases, and it is
not. surprising if none of them can individually cover the complete effect of innovation

in the real world. The intension of this section is to estimate the relative impact,

17

of innovation on the extensive and intensive margins. The benchmark model of

estimation is:

GDP
111(t7'ad9cit) = [9’0 + 51 ln(pate'ntcit) + /32 ln(p0pct) + 331““ pop

 

)ct) (11)

+84 ln(dz'stc) + £35langc + figbordc + [57”,- + [38M + 5m

ln(t7‘adem-t) in equation 11 represents the three export variables: the natural logs
of the overall export, extensive margin, and intensive margin of exports. I will
run three separate regressions to estimate the effect of innovation on each margin.
These variables are defined in equations 6—8, and are calculated on the industry-year

7 and thus

base. These dependent variables may change with industries and years,
industry-specific effects (lti) and year—specific effects (At) are included in the model.8
The key explanatory variable, 111(paten,tc,;t), is the natural log of weighted patent
counts (WPC) in equation 5. As discussed in section 3, weighted patent counts can
change systematically by industry and by year because of the classification system
and the examination process of US. Patent Office. The patent counts should also be

adjusted by industry and year to be comparable in different industries and years. This

adjustment is taken into account in a log linear model once industry and year fixed

 

7Although these variables are calculated relatively in a industry-year base, they can still change
with time and industry. For example, in earlier years there were fewer countries export to the US.
than in later years, and thus export shares in earlier years are likely to be higher than that of later
years. Similarly, the number of countries which have positive exports can differ by industry.

5A complete control should include an interaction between industry and year fixed effects. It
turns out the estimated results are almost identical, and thus I include industry and year dummies
separately in the estimation without taking interactions.

18

GDP
P0P

 

effects are included.9 ln(popct) and ln(( )ct) are natural logs of population and
GDP per capita of country c in year t, and these two variables will change over time.
Similarly, the inclusion of year dummy variables controls for this effect. ln(distc)
is the natural log of the distance between country c and the US. langc is a. binary
variable, which equals one if country c uses English as its primary language. bordc is
a binary variable, which equals one if country c has a common border with the US.
These three variables are included in the estimation as proxies for trade costs.

The coefficients on ln(patentc,:t) and other leg variables are interpreted as the
elasticity of different margins in experts with respect to these regressors. An ad-
vantage of taking natural legs is: because OLS is a linear operator, the regressimls
can decompose the impact of each regressor on overall exports into different margins
additively: 3,63: = 3:7” + 3:7”, Where 3:31;, fifmand 3:7” are the coefficients in the three
equations using the overall export share, extensive margin, and intensive margin as
the dependent variable, respectively. The relative contribution of the extensive and

Aem aim
intensive margins can be measured by calculating %and %?nr, respectively. Because
2 2
innovation is supposed to improve exportss, the coefficients on ln(patentmjt) should
be positive. According to Hummels and Klenow (2005), coefficients on country size
(i.e., ln(p0p) and ln(%%p—P)) should be positive, too. The coefficients on common

language and common border are supposed to be positive, and that on the distance

is expected to be negative.

 

9To measure the relative innovation of a country in an industry-year, the weighted patent count
of each country should be divided by the total number of weighted patent counts of all countries in
the same industry-year. Observations from a same industry-year will be divided by a same number,
and after taking log this will be obsorbed by the industry and year fixed effects.

19

Equation 11 is estimated using OLS.10 This estimation is referred to as the
two-wey—fimed-efiects case since it controls for industry-specific and year—specific fixed
effects. Standard errors are made robust to heteroskedasticity and are clustered by
country. The estimated results are presented in table 1.3: the estimates of the export
share regression, the extensive margin regression, and the intensive margin regression
are reported in columns (1), (2) and (3), respectively. The key explanatory variable,
111(patent), has positive and significant effects on overall export as well as on both the
extensive and intensive margins. The coefficient in column (1) reveals that a country
which has 10% more patents will export 5.4% more. This effect can be decomposed
into different margins: a 10% increase in patent counts will raise the extensive margin
by 1.6(’“/(,-.-11 and the intensive margin by 3.8%. This means 30% of the higher export
from more innovative countries occurs on the extensive margins, and 70% occurs on
the intensive margins.12

The next. two rows in table 1.3 verify that large countries export more. A coun-

try which has 10% more population will on average export 5.5% more varieties and

 

H’Since‘ by definition export shares, extensive margins, and intensive margins are between zero and
one, ln(t'ra.de) must be non-positive. Because I use OLS to estimate the model, the fitted values
can be positive. It turns out that only a very small proportion of the fitted values has this problem.
For instance, only 103 observations (out of 25,268, which is less than 0.05% of the sample) have.
positive values in log export. And these observations are from 3 countries: Canada, Germany, and
Japan. For the extensive margin, about 6% of observations have positive fitted values. For the
intensive margin, no observation has a positive value.

11The definition of the extensive margin used in this paper actually represents the products cov-
erage share of each country in each industry-year. It cannot, nevertheless, reflect. an increase in the
absolute number of products. For example, suppose in 1975 there are in total 100 varieties, and
each variety has a equal weight. If Germany exports 90 varieties in 1975, the extensive margin is
90%. If in the year 2000, the total number of varieties in this industry increases to 1,000 and Ger-
many exports 800 of them, the extensive margin of Germany will drop to 80%. Although Germany
exports more products in the year 2000 than 1975, its relative coverage share of products falls and
thus the extensive margin decreases.

12The contribution of the extensive margin is (0160/0537) 2 30%, and the contribution of the
intensive margin is (0378/0537) 2 70%.

20

5.0% higher value in these varieties. A 10% increase in GDP per capita will raise
the extensive margin by 8.9% and the intensive margin by 6.1%. With regard to
population and GDP per capita, the extensive margin accounts for a larger part (52%
with respect to population and 59% with respect to GDP per capita) than that of the
intensive margin (48% with respect to population and 41% with respect to GDP per
capita). The relative contribution of the two margins with respect to country size in
this paper is similar to Hummels and Klenow (2005), in which they use cross-sectional
data at the country level and find that the extensive margin accounts for about two
thirds of the increased exports from larger countries.

The remaining part of table 1.3 shows that countries located close to the U.S.,
using English as the primary language, and sharing a common border with the U.S.
have higher overall exports to the U.S. Among these variables, only the coefficient
on distance is significantly different from zero. All coefficients have correct signs
except that of border in the intensive margin regression.13 Most of these coefficients
on trade costs are individually insignificant, but they are jointly significant in each of
the three regressions.

In addition to the industry—specific and year-specific effects, there can be unob-
served country characteristics which are related to export. performance. A three-
way—flred-eflects model controlling for industry, year, and country-specific effects is

estimated using OLS:

 

UIn this sample only two countries, Canada and Mexico, have common border with the U.S. and
thus the variation of this variable could be small.

21

DP

 

111(t’7‘ad6c-zt) : .130 + $81 111(patentc-it) + fi2 1119901761.) + #33111“ )(ft) (12)

+£34Hz' + 135/\t + 565C + Ecz't

where (it is the country fixed effect, and all other variables are as defined in equatirm
11. Since distance, language, and border are country—specific characteristics, they
are dropped out. once the country fixed effect is included. Standard errors are again
made fully robust. The estimates of this three—way-fixed-effects case are presented
in table 1.4. The effects of innovation on overall export shares and the two margins
remain statistically significant, and the effects are stronger than in the two-way-
fixed-effect case. A country which has 10% more patents exports 2.0% more varieties
and 3.9% higher value of these varieties on average, leading to a 5.9% higher overall
export share. The relative contribution of the two margins is similar to the two-
way-fixed-effects case. The extensive margin accounts for 34% and the intensive
margin accounts for 66%. The coefficients on country size are different. from that
of the two-way-fixed-effects case. The intensive margin now accounts for a, slightly
larger proportion than the extensive margin, but the effect on the intensive margin
is insignificant. An explanation is that countries’ relative sizes do not. change much
across years. Once the country fixed effect is included, the variation in country size
becomes small and leads to insignificant coefficients. The bottom line is the effect. of
innovation on exports is significant in both cases, and the relative importance of the

two margins is also similar.

22

The results in tables 1.3 and 1.4 reveal that trade theories predicting innovation
can improve exports through only one margin do not explain the full effects of innova—
tions. Models like Krugman (1979) which assume that innovations can only increase
the number of varieties but not productivity or quality fail to explain the high inten-
sive margins from more innovative countries. Models which assume that innovation
only increases productivity or quality capture the higher intensive margins, but they
miss the effect on the extensive margin, which contributes nearly one third of the total
impact. From this point of view, a theoretical model which reconciles innovations’
ability in expanding the number of varieties as well as in increasing productivity or
quality is required to explain the complete influence of innovation on exports.

According to theoretical works reviewed in the introduction, the contribution of
the extensive margin will raise exporting countries’ welfare. The welfare change
caused by the intensive margins depends on whether innovation increases export
quantities or product quality. If the high intensive margin is mostly induced by
higher export quantities, the expansion in export volume may be at the cost of a.
terms—of—trade loss. If the higher intensive margin is mainly driven by an increase
in product quality, the terms of trade can improve and welfare will be higher. To
identify the welfare impact and terms of trade effect, a further decomposition of the

intensive margin will be discussed in section 5.

23

4.2 Interaction between Patent and GDP per capita

The benchmark estimations in the previous section assume that the effect. of inno-
vation on exports is the same across countries. However, trade theories which use
North-South models display that technological progress can impose asymmetric im-
pacts on developed and developing countries (eg. Flam and Helpman, 1987). To
explore this issue, I estimate a different version of equation 11 which adds an inter-

action between patent counts and GDP per capita.

ETD/P
ln(tradem) = {30 +,[311n(patentm-t) + ,821n(patentcit) x ln(( lot) (13)
pOp

GDP
+5 111 pop; +6 ln
3 ( ct) 4 (( pop

 

 

)ct) + ,85 ln(distc)

+551angc + 13700Tdc + '88/1'2' + fight + Ecit

GDP per capita in the interaction term is divided by the mean of all countries’

6513
pop

 

GDP per capita. of the same year (i.e., ln(( )ct) is the demeaned log GDP per

capita). In equation 13, the elasticity of export with respect to the patent is 61 +

N

[32 111((Q-Q£)Ct), which varies with countries’ income levels. This model is estimated

1901’
by OLS and the estimates are reported in table 1.5. The coefficients on ln(patem‘)
are positive in all regressions, and the coefficients on the interaction term are all
negative. This means the influence of innovation on the different margins of export

decreases with countries’ income levels. Though the coefficient on the interaction

term in the intensive margin regression is individually insignificant, the impact of

24

patents is jointly significant with the coefficient on ln(patent).

To have a better quantitative intuition about how elasticities of exports vary by
countries’ income levels, consider three countries which are identical in all aspects
except their GDP per capita: the high-income country’s GDP per capita is twice as
much as the sample average; the middle-income country’s GDP per capita equals the
average; and the low-income country’s is half as much as the average. Using the
estimates in table 1.5, I can calculate the elasticity of each margin with respect to
innovation of these countries, and the results are reported in table 1.6.14 In each
margin, the low-income country has the highest elasticity, followed by the middle-
income country, and then the high-income country. For example, a 10% increase
in patent counts will increase the low—income country’s overall exports by 7.9%; but.
it will increase the middle—income country’s overall exports by 6.6%, and the high-
income country’s exports by only 5.4%.

The composition of the higher exports led by innovation also changes with income
levels. In table 1.5, the coefficient on the interaction term in the extensive margin
regression is larger in absolute value and statistically more significant than that in
the intensive margin regression. Though the impact of innovation falls with the
income level in both margins, it falls more quickly in the extensive margin than in

the intensive margin, and thus the relative importance of the extensive margin also

 

1" T he estimated elasticity of export with respect to innovation equals 31 + B211’1(demea.ned GDP
per capita). For example, the high-income country’s elasticity of overall export is 0663—0. 178 x In 2 :
0.540. We can calculate the elasticity of countries with any income level. For example, in 1999
Norway’s GDP per capita is $32,355, and the mean GDP per capita is $10,609. This means Norway
has demeaned GDP per capita = 3.05, or ln(demeaned GDP percapita.) = 1.12. The estimated
elasticity of expert with respect to innovation for Norway will be 0.663 — 0.178 x 1.12 = 0.464.

25

declines with GDP per capita. In table 1.6, the extensive margin accounts for 40%
of innovation’s impact in the low-income country, while it accounts for 30% of the
effect in the high-income country. These findings are also presented in figure 1.1.
The height of each bar is the elasticity of overall exports with respect to innovation,
and the composition of each bar shows the relative contribution of the extensive and
intensive margins.

This section is concluded by the following two points. First, patents have stronger
effects on low-income countries than on high-income countries. This difference is
significant in the overall exports and the extensive margins but relatively insignificant
in the intensive margins. Second, the extensive margins in low-income countries
contribute a higher proportion than that in high-income countries to an increase of

exports caused by innovations.

4.3 Differentiated and Homogeneous Products

The estimations up to this point implicitly assume that the effect of innovation on
exports is identical across industries. In this section, I estimate an alternative model
specification which allows innovation’s effect on exports and the composition of this
effect to differ by industry. More specifically, I investigate whether innovation imposes
different impacts on differentiated products and homogeneous products.

Ranch (1999) classifies products into two types: homogeneous products which have
a reference price and differentiated products which have no reference price. Based

on Rauch’s classification, I calculate a difierentinted-product share of each industry.

26

which equals the number of differentiated products divided by the number of total
products in each industry. Industries with higher differentiated—product shares have
relatively more differentiated products than homogeneous products compared with
industries with lower shares. The differentiated-product shares of the 12 manufac-
turing industries in my sample are listed in table 1.2. I then categorize industries
into two groups. The first group contains six industries which have differentiated-
product shares higher than the sample median. These industries are defined as the
differentiated-product industries, including electronics, instrument, transportation,
machinery, fabricated metal, and glass—stone—clay. The second group includes six
industries which have differentiated-product shares lower than the sample median.
These industries are food, textile, plastic, chemicals, rubber, and primary metal, and
are defined as homogeneous-product industries.

To explore whether innovation imposes asymmetric effects on the export perfor—
mance of differentiated-product and homogeneous-product industries, I estimate the

following model:

ln(tra.de(.it) = 30,- + 131 ln(patenta-t) + 521n(patentC,-t) X differentiated, (14)

GDP
+5 ln(popt +13 111
3 c) 4 (( pop

 

)ct) + [351n(distc)

+136langc + 67b0rdc + (38 difierentz'ated, + 69A,: + Em

The variable difierentz’ated, is a dummy variable which equals one if industry 2' is a

differentiated—product industry and equals zero if industry 2' is a homogeneous—product,

27

industry. The interaction between ln(patent) and differentiated allows innovation’s
impact on exports to be different across industry groups. The elasticity of exports
with respect to innovation in the differentiated industries is Bl + 52, while that in
the homogeneous industries is [3’ 1. We expect that innovation has a. stronger impact
on exports in differentiated industries than in homogeneous industries, because the
production of differentiated products requires more innovations to create new varieties
or to increase the quality of products than that of homogeneous products.

I estimate equation 14 using OLS, and the estimated results are presented in
table 1.7. The coefficients on the interaction term (132) are positive and significant. in
all three regressions, indicating that innovation has stronger impacts in differentiated
industries than in homogeneous industries. The estimated elasticities of exports with
respect to innovation of differentiated—product and homogeneous-product. industries
are reported in table 1.8. A 10% increase in patents will increase the overall exports
of differentiated—product industries by 6.5%, and will increase the overall exports
of homogeneous-product industries by only 2.6%. The elasticity of overall experts
with respect to innovation of differentiated industries is more than twice as large as
that of homogeneous industries. Next, I compare the effects of innovation on the
extensive and intensive margins in the two types of industries. The elasticity of the
extensive margin in differentiated and the homogeneous industries are 0.15 and 0.12,
respectively. The difference is statistically significant but economically not large
(only 0.03). This is in contrast to the intensive margin, in which the elasticities of
the two types of industries differ greatly (0.50 for differentiated industries and 0.15 for
homogeneous industries). In differentiated industries, the intensive margin accounts

28

for 77% of innovation’s effect on exports, which is higher than that in homogeneous
industries (56%). These results can also be found in figure 1.2.

In summary, the estimates of this section show that innovation has different de-
grees of impacts on export performance across sectors. The impact is much stronger
in industries which have relatively more differentiated products, and the difference
in the intensive margin is the primary reason which causes the large difference in

innovation’s impact on exports across industries.

4.4 Estimation by Year

In this section I split the sample by year to carry out. a separate estimation of each
year. This work has two objects. First, the magnitude of innovation’s effect on
exports may change over time, and a year-by—year estimation helps to detect it. Sec-
ond, as mentioned in section 3, patent counts calculated according to the application
year are truncated especially in the last few years of the sample. In previous sections
I drop observations from the last three years to reduce the possible bias. By estimat—
ing the model by year, I can use observations of these dropped years to find out. if
they generate very different estimates. I can also compare the results using patent.
counts calculated by the application year and the grant year (which does not have the
truncation problem) to see if the two measurements lead to different conclusions.

A version of equation 11 is estimated separately by year t:

29

GDP
ln(tradem't) = [30, + ,8“ ln(patenta't) + 52t1n(P0Pct) + [33t1n(( pop

 

)ct) (15)

+fl4t111(distc) + [35tlangc + [36tbOTdc + [37t/J’i + Ecit

The estimates will contain twenty-seven sets of coefficients. I list the coefficients
on ln(patent) in table 1.9, and depict the elasticities of each margin against year in
figure 1.3. The change in estimates with time does not show a consistent pattern.
To summarize the changes in elasticities over time, I separate observations into three
periods and estimate the average elasticities of experts with respect to innovation
in each period: the seventies (1975-1979), the eighties (1980-1989), and the nineties
(1990—1999). The results are given in the second panel of table 1.9. The average
elasticity of overall exports in the seventies is 0.65; this average drops to 0.53 in the
eighties, and then slightly drops to 0.52 in the nineties. Innovation’s influence on
overall exports falls through the late eighties, but it fluctuates during the nineties
and climbs up after 1998. Figure 1.4 summarizes these results of estimation.

How can the change of innovation’s effect on exports be explained? First, the
falling influence may be related to the foreign direct investment (FDI). Compared
with domestic firms, a subsidiary of a multinational firm depends less on the host,
country’s technology level. For instance, a subsidiary of Dell in Mexico can produce
more advanced products than domestic Mexican firms. As the proportion of exports
accounted for by multinational firms increases with time, the explanatory power of

innovation becomes weaker, because innovation and export do not always take, place

30

in the same country. However, this answer cannot explain why the elasticity stopped
falling in the nineties, while FDI kept increasing during that period. The second
explanation is the increased pace of knowledge and technology transfer. The cen—
tral concept of technology gap theory is that innovation gives countries a temporary
advantage in competition. If technology and knowledge transfer more quickly and
more easily, the advantage created by innovation fades more quickly and becomes
relatively impotent. To identify whether or not the decreasing effect. of innovation on
exports is driven by these two reasons, we need data of multinational firms’ exports
and technology transfer, which will be explored in future research.

In table 1.9, the average elasticities of the extensive margin with respect to in-
novation in the seventies, eighties, and nineties are 0.23, 0.16 and 0.14, respectively.
The average elasticities of the intensive margin in these three periods are 0.42, 0.37,
and 0.38, respectively. The decrease in the elasticity of overall exports from the sev-
enties to the eighties occurs primarily on the extensive margin, and the change in the
intensive margin is comparatively small. This indicates that the relative importance
of the extensive margin regarding the effect of innovation decreases over time.

Finally, I implement the year-by-year estimation again, but now patents are classi-
fied according to the grant year rather than the application year. Classifying patents
by the grant year does not have the truncation problem, but it measures countries’
innovation capabilities with lags. The export elasticities with respect to patent using
grant year are reported in table 1.10. In most years, estimations using grant-year
patents have higher elasticities than that of using application-year patents, but the

differences are not large. More interestingly, the elasticity of overall exports starts

31

to climb after 1998 in both cases. Since the reverse trend occurs in both cases, I am
cautious to attribute it to the data truncation and need data of more recent. years
to reach a conclusion. In summary, table 1.10 basically tells us the same story as
table 1.9: the impact of innovation on exports decreases until late eighties, and the

contribution of the extensive margin falls with time.

4.5 Sensitivity Analysis

In this section, the robustness of the main findings is checked by estimating models
under different specifications and data construction methods: ( 1) considering the
persistency in exporting, (2) taking the endogeneity of innovation into account, (3)
including factor endowment and factor intensity, (4) controlling for tariffs and freight
costs, (5) using lag patents instead of current patents as the proxy for innovation,
(6) using different citation weights to calculate the weighted patent counts, (7) using
variables in level forms rather than in log forms, (8) constructing export variables
under different levels of aggregation. The estimates are stored in tables 1.11, 1.12
and 1.13. None of these variations alters the main conclusions of this paper. More
innovations induce countries to export more products and higher values of these
products, and the intensive margin accounts for a greater proportion of the effect

than the extensive margin.

4.5.1 Persistency in exporting

Trade literature shows that prior exporting experience affects later decisions of par-

ticipating in foreign markets. The persistency in trade patterns is related to the sunk

32

cost of exporting (Roberts and Tybout, 1997; Bernard and Jensen, 2004; Das et. al.,
2007). Although most of the evidence is presented at the firm level, the persistency
in exporting can still exist at more aggregate levels. I consider the following dynamic

model:

ln(tradem) 2 oz + 'y ln(tradem;¢_1) + 61 ln(patentm) + 521n(p0pct) (16)

, GDP
+£331n((
P0P

 

)ct) + 541n(distc) + [351angc

+flGb07‘dC + (37Hic ‘1' 38M + 5cit

Compared with the benchmark model (equation 11), the dynamic model includes the
lag export variable (ln(tradeci’t_1)) in the regression as an additional regressor.15
Using OLS or regular panel data estimators to estimate this model will generate
inconsistent. estimates, because the lag dependent variable is endogenous. I use the
Generalized Method of Moment. (GMM) estimator for dynamic panels proposed by
Arellano—Bover (1995) / Blundell—Bond (1998) to estimate this model. The system.
GAIM estimator uses lag dependent variables in level or difference forms dated t — 2

or earlier as instruments for period t, to account for the endogeneity of lag dependent.

variable.16

 

1"3This model is more flexible than equations 11 and 12, because it allows each country-industry
to have its fixed effect.

16A necessary condition for lag variables to be valid instruments is that there is no autocorrelation
of order 2 or above in the idiosyncratic error term. Using the Arellano and Bond autocorrelation
test, my sample shows symptoms of AR(2) but not of AR(3). So only variables dated t-3 or earlier
are used as instruments.

33

The estimated coefficients on lag exports and patents are recorded in panel (A)
in table 1.11. The coefficient on lag export variable of each regression is positive and
significant. This finding verifies the persistency in exporting at the industry level.
The coefficients on ln(patent) are smaller than that of the benchmark estimation.
But the result shows that even after controlling for lag export variables, the effect of
patents on each margin remains positive and very significant, and the elasticity of the

intensive margin is larger than that of the extensive margin.

4.5.2 Endogeneity of innovation

This section takes the endogeneity of innovation into account. Innovation can be
endogenous due to reverse causality or simultaneity. Changes in trading opportunities
can alter the motivation of innovation, and it can even change the incentive to apply
inventions to be patents.17 I estimate equation 16 without including ln(tradcm-JJ)
and considering innovation to be endogenous. The model is estimated using G.\*IM
estimator, and current innovation is instrumented with lag innovations. Panel (B) in
table 1.11 shows the estimated coefficients on ln(patent). The result is very similar to
the benchmark case, which indicates that considering the endogeneity of innovation

does not change the main conclusion of this paper.

 

17First, there can be variables which are not controlled in the model, but can affect innovation
and export simultaneously (e.g. change in government policies). Second, export and innovation
can affect each other. On one hand innovation can improve competitiveness and increase exports;
but on the other hand trade can induce the transfer of knowledge (Coe and Helpman, 1995) which
motivates more innovations.

34

4.5.3 Factor endowment and factor intensity

The Heckscher-Ohlin model (H-O model) predicts that trade patterns are determined
by factor endowments of countries. In this section I estimate a. version of equations
11 and 12 which controls for the factor endowments of countries. I calculate each
country’s skilled labor, unskilled labor, and capital endowment proportions (made
relative to land). The interactions between factor endowment proportions of each
country and the factor intensity of each industry are included in the model of estima-
tion. More details about the data source and how I calculate the factor endowments
and factor intensities are discussed in appendix B. The estimated coefficients of
ln(patent) are displayed in panel (A) of tables 1.12 and 1.13. The result indicates

that. including factor endowments does not change the main conclusion of this paper.

4.5.4 Tariffs and freight costs

In this estimation tariffs and freight costs are included to control for trade barriers.
The trade data described in section 3 have information of freight costs and duties paid
at. U.S. custom of each product. Similar to Hummels and Klenow (2005), I define an
index of trade barrier which equals the sum of duty and freight divided by the total
export value of each country-industry in a given year. The benchmark models in
equations 11 and 12 are estimated by OLS with the inclusion of this index of trade
barrier (ln(duty freight trate)c,~t). The coefficients on ln(patent) are displayed in
panel (B) of tables 1.12 and 1.13. The magnitude and the composition of patents

effect do not alter after including the duty-freight rate.

35

4.5.5 Replace current patent counts by lag patent counts

In this specification, I use lag patent counts instead of current patent counts as the
proxy for innovation. Producers may need some time to adjust their production
in response to a change in the technology level, and the effect on exports can take
even longer. Furthermore, it is more convincing that lag patent counts are exoge-
nous to current exports, and thus using lag patent counts can mitigate the possible
simultaneity problem. ln(patenta-t) in the benchmark models (equations 11 and 12)
is replaced by ln(pritentcit_1) and ln(pa.te'n.tc,-t_2), separately, and the model is es-
timated by OLS. The sample size becomes smaller because some observations are
drOpped out when lag patent counts are used instead of current patent counts. The
coeflicients of ln(pate~ntcz't_1) and ln(pa.tentcit_2) are shown in panel (C) in tables
1.12 and 1.13. Compared with the estimates using current patent counts, the coeffi-
cients in the two—way-fixed-effects case become smaller, but the main results in both

cases remain very similar to that of using current patents.

4.5.6 Different weights of patent citations

In section 3, I discuss the importance of weighting patents by the number of citation
received. The sample applies the linear weighting scheme of Trajtenberg (1990), in
which the weight. of each citation equals 100%. To make sure that the results are not
sensitive to the selection of the citation weight, I calculate the weighted patent counts
with alternative citation weights from 0 (simple patent counts) to 100%. Equations

11 and 12 are estimated with weighted patent counts using different citation weights.

36

Since elasticities with respect to country size and trade resistances are very similar
to those in benchmark case in table 1.3, I only present the elasticities with respect. to
patents in panel (D) of tables 1.12 and 1.13. The elasticity decreases as the citation
weight increases.18 The decrease in the elasticity occurs on both the extensive margin
and intensive margin, but patent’s effects are significant and the relative importance

of the two margins remains similar under different weights of citations.

4.5.7 Variables in levels

In this case the benchmark models are estimated again but all variables are. in level
forms without taking logs. The patent count variable is made relative to the world
level in each industry-year, and population and GDP per capita are made relative

19 The results are shown in panel (E) in tables

to the world level in each year.
1.12 and 1.13. Without taking log on exports, I cannot decompose the impact of
each explanatory variable on export performance into different margins additively.
The bottom line is that all of the coefficients on innovation are still positive and

statistically significant. Switching from a log-linear model to a linear model does not

deny the impacts of innovation on exports.

 

115A possible connection is as follows. High income countries generally have more patents, and thus
their patents are more likely to be cited. This is related to the concept of "self-citation" mentioned
in Hall, Jaffe and Trajtenberg (2001). Assigning higher weight to citations tend to assign higher
weights to patents from rich countries. Since we already show the effect of patent on export is
decreasing with countries’ income level, the outcome in tables 1.12 and 1.13 is not too surprising.

19The proxy of innovation is a country’s weighted patent counts divided by the total number of
weighted patent counts of all countries in an industry-year. Population variable equals a country’s
population divided by total population of all countries in a year. GDP per capita variable equals a
country’s GDP per capita divided by the average of all countries’ GDP per capita in a year.

37

4.5.8 Different aggregation levels in exports

Hummels and Klenow (2005) point out that the relative importance of the extensive
margin and intensive margin is sensitive to the level of aggregation in export data.
In the sample, the number of product categories exported from a. country is identified
by each TS7 or H810 code. To precisely estimate the relative contribution of the two
margins, it is critical to have a disaggregate product classification in the export data.
If products are classified at a more aggregate level, it is more likely that different
products are included in a same category. In this case, an increase in the number
of products within each code cannot be observed, and the effect will be attributed to
the intensive margin (an increased value of this product code). This type of within
product variation is proved to exist and discussed in Schott (2004). By intuition,
the contribution of the extensive margin should fall when the export variables are
calculated at a. more aggregate level. To check this effect in this paper, I follow
Hummels and Klenow (2005) to estimate equation 11 and 12 with difler‘ent levels of
product aggregation.20 Panel (F) in tables 1.12 and 1.13 show that the elasticities
of the extensive margin with respect to innovation under different levels of aggrega-
tion. Using more aggregate product classification will decrease the elasticity and the
contribution of the extensive margin, and will increase that of the intensive margin
at the same time. Except the most. aggregate case (3—digit), innovation’s impacts on
both margins remain statistically significant. By using the most disaggregated trade

data (TS7 and HS10) available, this paper can estimate the contribution of the two

 

2”Other than the TS7 and H810 codes, I calculate ln(emport), ln(ertensive), and ln(intensire)
with each variety j defined by 3 digit to 6 digit of TS or HS codes.

38

margins more accurately than studies using more aggregate data.

5 Decomposing the Intensive Margin: Innovation

and Product Quality

In section 4 I show that the intensive margin accounts for a relatively larger part
of innovation’s effects than the extensive margin does. According to trade theories
reviewed in section 2, innovation can increase the intensive margin in at least two
ways: increasing productivity and / or improving the quality of products. Empirical
studies have shown evidence of the positive correlation between innovation and pro—
ductivity.21 However, the relation between innovation and product quality has not.
been studied broadly.22 A common difficulty faced. by researchers is that product.
quality is not directly observable. In empirical trade literature, quality is usually
inferred from unit values. Some literature uses the export unit price as an index of
quality (Hummels and Skiba, 2004; Hallak, 2006): within each variety, a higher unit
price is equivalent to higher quality. Others studies consider the export price as an
indicator of quality (Hallak, 2006; Hummels and Klenow, 2005; Hallak and Schott,
2008): the unit price reflects product quality and other factors such as production

costs.23 These studies agree that product quality is an important, if not. the only,

 

2’ Early studies are surveyed ,by Griliches (1979). Recent studies mainly use firm level data, such
as VVakelin (2001) and Griffith et al., (2006).

2QIn empirical international trade literature, export quality has been connected to factor endow—
ments (Schott, 2004), income levels (Schott, 2004; Hallak, 2006) and different types of export costs
(Hummels and Skiba, 2004).

23Different export prices can also be caused by within variety differentiation. So it is very
important to have disaggregate trade data to measure product quality more precisely. See Hallak
(2006).

39

factor that affects export prices. In this section I assume that the unit price is pos-
sibly affected by both the product quality and production cost. The object of this
section is to decompose the intensive margin into the price and quantity indices, and
then to look for evidence of the link between innovation and product quality.
Consider two polar scenarios. First, suppose innovation increases productivity
but has no influence on the quality of products. As shown by Schott (2004), countries
with higher productivity will expert at lower prices. In this case, we expect export
prices to be lower and export quantities to be higher in more innovative countries.
Second, suppose innovation enables producers to produce goods of higher quality
(Grossman and Helpman, 1991), and higher quality is linked with higher pr(’)duction
cost (Baldwin and Harrigan, 2007). Innovation induces producers to export goods
of higher quality at higher prices, but the export quantities can increase as well?”1
In summary, if innovation raises the intensive margin mainly by reducing pro-
duction costs, we will observe more innovative countries export higher quantities at.
lower prices. On the other hand, if innovation increases the intensive margin pri-
marily by improving product quality, countries with more innovations will expert. at
higher prices. The logic is similar to Schott (2004), but my focus is on the impact of

innovation, which is not taken into account in Schott (2004).

 

“In a model with differentiated-quality products such as Grossman and Helpman (1991a), the
export quantity is determined by the quality-adjusted price. If the quality-adjusted price increases
(prices increases more than quality), export quantity will decrease. If quality-adjusted price de-
creases, export quantity will increase.

40

5.1 Decomposition of Intensive Margin

As shown in section 2, the intensive margin of exports (pchcj) is measured by the
export value of each product. To estimate innovation’s effect. on price and quantity
separately, the intensive margin has to be further decomposed into the price ([26])
and quantity (ch) components. I adopt the well known Fisher Index to construct.

the empirical counterparts of the price and quantity indices:25

P _ [ZjE-Sc chch ]%[2j65c pquwj ]% (17)
C _
ZjeSC ijch ZjESc pquwj

 

 

ZjESC Page 5 53265.: 29...qu g

Zjesc 10chqu Ejesc ijq'wj

 

 

QC 2 i (18)

As in section 3, consider the price and quantity indices of different countries in the
same industry and year. Subscription c, j, and w represent country, product, and
world, respectively. pcj is the unit. price of product j from country c, and (ch is
the export quantity of product j from country c. pm is the average unit price of
product j from all countries, and qwj is the total export quantity of product j from all
countries. The price index in equation 17 is the geometric average of two elements.
The first element is the Paasche index: it is the sum of country c’s export prices of
PI‘ 0 ducts exported by c divided by the sum of world prices of the same products, with
(Bell product weighted by country c’s export quantity, ch. The second element of
price index is the Laspeyres index: it is the sum of country c’s export prices divided

by the sum of world prices of the same products, with each product. weighted by its

 

 

QC’Hummels and Klenow (2005) use a modified Fisher Index to construct. the price and quantity
indices at the country level which considers the effect of new-entering products .

41

world export quantity, qwj. Using the same method, the quantity index in equation
18 is the geometric average of the Paasche index and the Laspeyres index?“

The trade data (Feenstra, 1996 and Feenstra, Romalis and Schott, 2002) contain
the export value and quantity of each product. The quantity information can be

used as M directly, and the unit price of each product can be inferred by dividing

Because the dataset does

 

the export value by the quantity (i.e., pcj = p333. ).
not have perfect quantity information of each product, some observations will have
missing values in the price and quantity, and the sample size used in this section
is smaller than the original one. More details of the construction of the price and
quantity indices are discussed in appendix C.

The intensive margin in equation 8 can be decomposed into the price index and

quantity index as in equations 17 and 18:

intensivemft = Pcit x Qmjt (19)

5.2 Estimation and Results

Because the sample used in this section is different from that used in section 4.1, the
twaway-fixed-effects model in equation 11 is estimated again using the new sample,
and the results are recorded in columns (1)—(3) of table 1.14. Compared with the
estimates in table 1.3, the coefficient on ln(patent) in the extensive margin regression

increases and that in the intensive margin regression falls. The influence of innova-

 

‘) ' . . - .

*“Bealuse a Single country’s export quantity index can not exceed world export quantity, the
quantity index is smaller than 1. But a country’s export price can be higher or lower than the world
average price, so the price index can be greater or smaller than 1.

42

tion on both margins remains positive and significantly different from zero, and the
intensive margin accounts for a larger proportion. Using the new sample does not
induce any material change in the main conclusions.

To estimate the relative contribution of the price and quantity components to

the intensive margin, I regress the natural logs of the price and quantity indices

GDP)

on ln(patent) and other explanatory variables as in equation 11 (ln(pop), ln( pop

ln(dist), ln(lang), and 1n(b0rd)). Industry and year dummies are controlled, and
standard errors are robust to heteroskedasticity and clustered by country. Since
ln(intensiveCfi) = ln(PCZ-t) + ln(Qm-t), the effect on the intensive margin can be de-

_ . bin).
composed additively: 13,-

’5 fq , éim {5}) {sq
, = [3? + if,- i = 1 ...... 6, where 6,- , [3,, and 5,,- are the
coefficients in the intensive margin, price margin, and quantity margin regressions,
respectively. According to the two assumed polar cases, we have two competing
hypotheses. If innovation increases the intensive margin by reducing production
costs, more innovative countries will export. higher quantities at lower prices, and we
1;,71 eq . . . . _. .
expect to see i 1 < 0 and 51 > 0. If innovation increases the intens1ve margin by
u a u :\p “‘0 ¢q o , -

raismg product quality, we expect to see ,31 Z 0; and the sign of 131 is determined by
qua ity-a. juste price.

The estimated coefficients of the price and quantity regressions are presented in
columns 4 and 5 of table 1.14. A 10% increase in patents will increase export prices by

0.5% and export. quantities by 2.5%. The higher intensive margin led by iiiiitwation

is mainly driven by higher export quantities (84%), and the higher export price ac-

 

‘77- . u . . . .

" If the coeffiCient in the quantity regression is negative, the absolute value should be smaller than
the coefficient in the price regression to have a positive correlation between innovation and intensive
margin.

43

counts for a relatively small contribution (16%). Although innovation does not have
economically strong effect on export prices, the effect is statistically significant. The
results show evidence that innovation raises product quality. Countries which have
more innovations can export higher quantities at higher prices, because their products
have higher quality. The rise in the export quantity implies that the quality-adj usted
price falls. Although the result indicates that innovation increases product. quality,
it does not necessarily contradict other empirical studies which find that innovation
can increase productivity. It is possible that the two effects coexist, and an increase
in productivity can decrease the quality-adjusted price further and contribute to the
increased quantities of exports. The bottom line is that the estimates do not. show
direct evidence that innovations worsen the terms of trade, and thus the increase in

exports caused by innovation is more likely to be welfare—improving.

6 Conclusion

In this paper I divide export flows into the extensive and intensive margins to investi-
gate the extent. to which innovation increases the number of products and the export
value of each product. I use export data of 12 manufacturing industries from 105
countries to the U.S. market. for the period 1975-2001 to construct the extensive and
intensive margins, and use the number of patents granted by the U.S. as the proxy
for innovation, to investigate the relative importance of the extensive and intensive
margins of exports regarding innovation’s impact. The estimates show that. innova-

tion has significantly positive effects on both the extensive margin and the intensive

44

margin of exports. In the benchmark estimation, a country with 10% more patent
counts will export 1.6% more products and 3.8% higher value of each product, which
leads to a 5.4% increase in overall exports. The intensive margin accounts for a.
larger proportion of innovation’s effect (70%) than the extensive margin (30%). The
results are robust to alternative model specifications which consider the persistency
of exporting, the endogeneity of innovation, the inclusion of tariffs and freight costs,
alternative innovation measurements, and the level of exports aggregation. Further
decomposing the intensive margin into the price and quantity indices shows that more
innovative countries export greater quantities at higher prices, suggesting that inno-
vation raises product quality of exports. This result implies that innovation improves
the innovating countries’ welfare not only by increasing the number of varieties but
also by improving their terms of trade.

I also investigate how the effect of innovation on exports varies by country, by
industry, and by year. First, innovation has stronger impacts on countries which
have lower GDP per capita, and the relative importance of the extensive margin is
also greater in lower-income countries than in higher—income countries. This result
points out that innovation may be an important factor which determines developing
countries’ technology catch-up. This is different from most existing literature which
simply assumes that developing countries do not innovate. Second, innovation has a
greater impact on exports in industries which have relatively more differentiated goods
than homogeneous goods. This indicates that innovation’s impact on exports differs
by industry. Except the share of differentiated goods, other industrial characteristics
such as technology intensity and the producer-user relation of innovation may also

45

lead to different impact of innovation on exports. Finally, the year-by-year estimation
displays that the effect of innovation on exports and the relative contrilmtion of the

extensive margin decline over time.

46

Table 1.1: List of Countries

 

Algeria
Argentina
Australia
Austria
Bahamas
Bahrain
Belgium
Belize
Bolivia
Brazil
Bulgaria
Cameroon
Canada.
Chad
Chile
China
Colombia
Congo
Costa Rica
Cyprus
Denmark

Dominican
Ecuador
Egypt

El Salvador
EthiOpia
Germany
Fiji
Finland
France
Gabon
Ghana
Greece
Guatemala
Guinea
Guyana
Haiti
Honduras
Hong Kong
Hungary
Iceland
India

Indonesia
Iran

Iraq
Ireland
Israel
Italy
Jamaica
Japan
Jordan
Kenya.
Korea
Kuwait.
Laos
Liberia
Madagascar
Malawi
Malaysia.
Mali
Mauritania
Mauritius
Mexico

Morocco
Nepal
Netherlands

New Zealand

Nicaragua

Niger
Nigeria
Norway

Oman
Pakistan
Panama

Paraguay

Peru
Philippines
Poland
Portugal
Qatar

South Africa

Romania

Saudi Arabia

Senegal

Sierra Leone
Singapore

Spain

Sri Lanka

Sudan

Suriname
Sweden
Switzerland
Tanzania
Taiwan
Thailand
Trinidad and Tobago
Tunisia

Turkey

Uganda

United Arab Emirates
United Kingdom
Uruguay
Venezuela
Zambia
Zimbabwe

 

47

GOA
GOA
GOA
OGA
NGG
ENG
GOO
GOO
ONO
mmG
AVOG
AGO

wqflsmme SO 3885me
SSA on£¢> A322

mime??? mqoammAwB
$853800 030585

862 6.582% Saginaw
mamamem was wQEoEm gain
$260.5 $3280 .mumfieucoo wmmAU
mflfiom mesmflm

winced EaonHuonA

mAaoAEon 35282

mucnwm 5533on Go mHmAAmEE
as; 8:96 J52 25

32.605 U828 cue muzmasfimfi
#58955 nomaefiommdafit

£58935 2:95an «Em 33583
508953 95 522592: Aflbmsvfi
mposcoa E68 woumotafim

motumfiofi A32: basin

$2605 $8280 can .mmflw SEQ .mcopm
mpuscoa 83me msomaazmomma was Hmnnsm
$2603 Row «:5 Esflobmm

33on won? 98 £35830

$3605 ES 23wa

$360.5 UBUEVA was Goon

mm
um
mm
mm
vm
mm
mm
Gm
ON
mm
mm
Om

 

v.85. poswoafioufiuamaflma

zuowomebrsm oEEexm

coflaflomom OAm

 

 

aw 2m sasa a acéefi a a: ”2 £8.

48

Table 1.3: Estimated Results: Two—way Fixed Effects Model

 

 

 

 

 

 

Dependent Variables ——> In export ln extensive ln intensive

Regressors 1 (1) (2) (3)

111 patent 0.537*** 0.160*** 30% 0.378*** 70%
(0.075) (0.040) (0.043)

In population 1.046*** 0.546*** 52% 0.501*** 48%
(0.089) (0.041) (0.054)

In GDP 1.505*** 0.891*** 59% 0.614*** 41%

WI)

(0.17) (0.080) (0.10)

In distance -0.616** ~0.383*** 62% —0.233 38%
(0.29) (0.13) (0.17)

language 0.339 0.199 59% 0. 140 4 1 Vt
(0.29) (0.14) (0.17)

border 0.829 -0.310 -37% 1.139** 137%
(1.03) (0.57) (0.48)

constant -31.27*** —15.49*** -15.77***
(3.36) (1.49) (2.05)

Industry dummy Yes Yes Yes

Year dummy Yes Yes Yes

Country dummy No No No

Observations 25268 25268 25268

R2 0.61 0.51 0.53

 

Notes: This table reports the estimated results of equation 11 using OLS. Robust standard
Percentages describe the contribution of

errors clustered by country are in parentheses.
each margin to overall exports.

Coefficients of industry and year dummy variables are

skipped. * indicates significance level: *** p<0.01, ** p<0.05, * p<0.1.

49

Table 1.4: Estimated Results: Three-way Fixed Effects Model

 

 

 

 

 

 

Dependent Variables —-> In export ln extensive ln intensive

Regressors i (1) (2) (3)

ln patent 0.584*** 0.196*** 34% 0.388*** 66%
(0.063) (0.025) (0.046)

In population 1.144* 0561* 49% 0.584 51%
(0.64) (0.30) (0.42)

DP ..
In G p 0614* 0.381*** 62% 0.233 38%
120"

(0.32) (0.14) (0.23)

constant -33.98*** ~16.86*** -17.12**
(11.6) (5.41) (7.82)

Industry dummy Yes Yes Yes

Year dummy Yes Yes Yes

Country dummy Yes Yes Yes

Observations 25268 25268 25268

R2 0.72 0.61 0.60

 

Notes: This table reports the estimated results of equation 12 using OLS. Robust standard
Percentages describe the contribution

errors clustered by country are in parentheses.
of each margin to overall exports.

Coefficients of industry, year, and country dummy

variables are skipped. * indicates significance level: *** p<0.01, ** p<0.05, * p<0.1.

50

Table 1.5: Estimated Results: Interaction between GDP per capita and Patent.

 

 

 

 

 

 

Dependent Variables ——> In export In extensive 1n intensive
Regressors 1 (1) (2) (3)
In patent 0.663*** 0.238*** 0.426***
(0.11) (0.044) (0.072)
”515 .
In patent X In G —0.178* -0.110*** —0.0680
P0P
(0.091) (0.035) (0.063)
In population 1.011*** 0.524*** 0.487***
(0.092) (0.043) (0.055)
In fl 1.524*** 0.902*** 0.621***
population.
(0.17) (0.080) (0.100)
In distance -0.629** -0.391*** -0.238
(0.29) (0.13) (0.17)
language 0.309 0.181 0.128
(0.28) (0.13) (0.17)
border 0.794 ~0.332 1 . 126* * *
(0.89) (0.48) (0.43)
constant -30.78*** -15.19*** -15.59***
(3.35) (1.49) (2.05)
Observations 25268 25268 25268
prob>F 0.00 0.00 0.00
R2 0.62 0.51 0.53

 

Notes: This table reports the estimated results of equation 13 using OLS. Robust
standard errors clustered by country are in parentheses. Coefficients of industry and year
dummy variables are skipped. * indicates significance level: *** p<0.01, ** p<0.05, *
p<0.1. The prob>F shows the p—value of testing 731 = [32 = 0 in equation 13.

51

Table 1.6: Elasticities of Exports with respect to Patents in Countries with Different
GDP per capita

 

 

Elasticities with respective to innovation

(1) (2) (3)

 

 

 

 

 

 

Country ratio export extensive intensive
P0P
Low-income 0.5 0.786 0.314 40% 0.473 60%
Middle-income 1 0.663 0.238 36% 0.426 64%
High-income 2 0.540 0.162 30% 0.379 70%
GDP . . . . . . .
Notes: ratio IS each country’s GDP per capita d1v1ded by the mean of all countries

P0P
GDP per capita. Numbers in columns (1)-(3) are estimated elasticities of each margin of

GDP

P0P
ages describe the contribution of the extensive and intensive margin to overall exports.

 

exports with respect to patent, which equals [31+{J’21n(( )) in equation 13. Percent-

Table 1.7: Estimation by Industry Group: Differentiated Products versus Homoge-

neous Products

 

 

 

 

 

Dependent. Variables ——> In export. In extensive ln intensive

Regressors l (1) (2) (3)

1n patent. 0.264*** 0.116*** 0.149***
(0.069) (0.036) (0.038)

111 patent >< differentiated 0.381*** 0.0321** 0.349***
(0.036) (0.015) (0.028)

In population 1.107*** 0.568*** 0.540***
(0.084) (0.040) (0.051)

In GDP 1.614*** 0.938*** 0.676***

p01)

(0.16) (0.077) (0.096)

In distance -0.649** -0.394*** —0.254
(0.29) (0.13) (0.17)

language 0.403 0222* 0.181
(0.27) (0.13) (0.17)

border 0.872 -0.310 I.182** *
(0.96) (0.56) (0.43)

differentiated -2.309*** -0.0196 -2.290***
(0.15) (0.061) (0.11)

constant -33.67*** -16.38*** -17.29***
(3.25) (1.45) (2.00)

Observations 25268 25268 25268

prob>F 0.00 0.00 0.00

R2 0.58 0.45 0.45

 

Notes: This table reports the estimated results of equation 14. Robust standard errors
clustered by country are in parentheses. Coefficients of year dummy variables are skipped.

* indicates significance level: *** p<0.01, ** p<0.05, * p<0.1.

p-value of testing {3’1 : [32 = 0 in equation 14.

53

The prob>F shows the

Table 1.8: Elasticities of Exports with respect to Patents by Industry Group
Elasticities with respective to innovation

(1) (2) (3)

Industry Group export extensive intensive
Differentiated 0.645 0.148 23% 0.498 77%

Homogeneous 0.264 0.116 44% 0.149 56%

Notes: Numbers are estimated elasticities of each margin with respect to patents. The
elasticity of exports in differentiated-product industries equals 61 + 52 in equation 14,
and the elasticity of exports in homogeneous-product industries equals [3 1. Percentages
describe the contribution of each margin to overall exports.

54

Table 1.9: Estimation by Year: Patents Classified according to the Application Year

 

 

 

 

 

Export —’ In export In extensive In intensive
Year

1975 0.698*** 0.223*** 0.475***
1976 0.650*** 0.224*** 0.426***
1977 0.635*** 0.303*** 0.331***
1978 0.644*** 0.208*** 0.437***
1979 0.608*** 0.209*** 0.399***
1980 0.614*** 0.204*** 0.410***
1981 0.632*** 0.207*** 0.425***
1982 0.569*** 0.184*** 0.385***
1983 0.520*** 0.149*** 0.371***
1984 0.525*** 0.149*** 0.376***
1985 0.516*** O.166*** 0.350***
1986 0.491*** 0.113*** 0.378***
1987 0.450*** 0.117*** 0.333***
1988 0.481*** 0.117*** 0.365***
1989 0.490*** 0.134*** 0.355***
1990 0.528f“** 0.159*** 0.369***
1991 0.543*** 0.161*** 0.382***
1992 0.507*** 0.118** 0.389***
1993 0.512*** 0.143*** 0.369***
1994 0.544*** 0.183*** 0.361***
1995 0.552*** 0.151*** 0.401***
1996 0.512*** 0.135*** 0.377***
1997 0.515*** 0.132** 0.383***
1998 0.502*** 0.108** 0.394***
1999 0.522*** 0.0835 0.439***
2000 0.596*** 0.108** 0.487***
2001 0.617*** 0.0913 0.525***

 

 

In export In extensive In intensive

 

1975—79 (avg) _0.647§*—*_" 0.231*** 36% 0.417*** 64%
198089 (avg) 0.526*** 0.155*** 29% 0.371*** 71%
1990—99 (avg) 0.517*** 0.140*** 27% 0.377*** 73%

 

Notes: The first panel of his table reports the estimated results of the year-by-year esti—
mation of equation 11. The estimated coefficients are elasticities of exports with respect
to patents. Estimates of other regressors are skipped. Industry dummy variables are
included. The second panel shows the average elasticities in the 70s, 803, and 905.
Percentages describe the contribution of each margin to overall exports. * indicates
significance level: *** p<0.01, ** p<0.05, * p<0.1.

Table 1.10: Estimation by Year: Patents Classified according to the Grant Year

 

 

 

 

 

 

 

 

 

Export —-> In export In extensive In intensive
Year
1975 0.651*** 0.205*** 0.445***
1976 0.634*** 0.196*** 0.438***
1977 0.672*** 0.323*** 0.350***
1978 0.600*** 0.198*** 0.402***
1979 0.584*** 0.198*** 0.386***
1980 0.565*** 0.176*** 0.389***
1981 0.656*** 0.215*** 0.441***
1982 0.549*** 0.167*** 0.382***
1983 0.490*** 0.135*** 0.355***
1984 0.513*** 0.141*** 0.372***
1985 0.482*** 0.150*** 0.332***
1986 0.434*** 0.0876** 0.346***
1987 0.467*** 0.127*** 0.341***
1988 O.487*** 0.107*** 0.380***
1989 0.435*** 0.128*** 0.307***
1990 0.498*** 0.131*** 0.367***
1991 0.523*** 0.146*** 0.377***
1992 0.491*** 0.123*** 0.368***
1993 0.507*** 0.134*** 0.373***
1994 0.522*** 0.175*** 0.347***
1995 0.520*** 0.144*** 0.376***
1996 0.458*** 0.117** 0.341***
1997 0.502*** 0.128** 0.374***
1998 0.416*** 0.0960** 0.320***
1999 0.457*** 0.0677 0.389***
2000 0.517*** 0.103** 0.414***
2001 0.528*** 0.104* 0.424***

In export In extensive In intensive
1975-79 (avg) 0.627*** 0.221*** 35% 0.406*** 65%
1980-89 (avg) 0.506*** 0.145*** 29% 0.361*** 71%
1990—99 (avg) O.487*** 0.128*** 26% 0.359*** 74%

 

Notes: The first panel of this table reports the estimated results of the year-by-year
estimation of equation 11, with patents counts based on the grant years. The estimated
coefficients are elasticities of exports with respect to patents. Estimates of other regressors
are skipped. Industry dummy variables are included. The second panel shows the average
elasticities in the 70s, 805, and 903. Percentages describe the contribution of each margin
to overall exports. * indicates significance level: *** p<0.01, ** p<0.05, * p<0.1.

Table 1.11: GMM Estimator: Persistency in Exporting and Endogeneity of Innovation

(1) (2) (3)

In export In extensive 1n intensive

 

(A) Persistency in Exporting

 

In patent 0.163*** 0.0913*** 0.118***
(0.019) (0.012) (0.017)

lag export variable 0.377*** 0.123*** 0.252***
(0.026) (0.023) (0.021)

(B) Endogenous Innovation
In patent 0.532*** 0.153*** 0.379***
(0.11) (0.050) (0.087)

Notes: This table reports the estimated coefficients on patents in models which take
the persistency of exporting and the endogeneity of innovation into account using GMM
estimator. Robust standard errors are in parentheses. "‘ indicates significance level: ***
p<0.01, ** p<0.05, * p<0.1.

 

57

Table 1.12: Robustness Checks: Two-Way Fixed Effects

 

 

Two—Way Fixed Effects

 

In export In extensive In intensive

 

 

 

 

 

 

 

 

 

 

(1) (2) (3)
(A) Factor Endowment
In patent 0.336*** 0.0892** 0.247***
(B) Tariffs and Freights
In patent 0.517*** 0.153*** 0.363***
(C) Lag Patents
In patent t-I 0.508*** 0.146*** 0.362***
In patent t—2 0.486*** 0.134*** 0.352***
(D) CitationWeights
0% 0.686*** 0.183*** 0.503***
10% 0.651*** 0.179*** 0.472***
30% 0.607*** 0.172*** 0.434***
50% 0.579*** 0.167*** 0.411***
100% 0.537*** 0.160*** 0.378***
(E) LinearModel
patent 0.470*** 1.651*** 0.504***
(F) Levels of Aggregation
7/10 digit 0.537*** 0.160*** 0.378***
6 digit 0.537*** 0.138*** 0.399***
5 digit 0.537*** 0.111*** 0.427***
4 digit 0.537*** 0.0822** 0.455***
3digit 0.537*** 0.0134 0.524***

 

Notes: This table reports the coefficients on patents.

58

Table 1.13: Robustness Checks: Three-Way Fixed Effects

 

 

Three-Way Fixed Effects

 

In export In extensive In intensive

 

 

 

 

 

 

(1) (2) (3)
(A) Factor Endowment
In patent 0.442*** ‘ 0.161*** 0.280***
(B) Tariffs and Freights
1n patent 0.562*** 0.188*** 0.374***
(C) Lag Patents
In patent t-1 0.589*** 0.192*** 0.397***
In patent t-2 0.575*** 0.185*** 0.390***
(D) CitationWeights
0% 0.869*** 0.278*** 0.591***
10% 0.814*** 0.264*** 0.550***
30% 0.725*** 0.238*** O.487***
50% 0.668*** 0.221*** 0.447***
100% 0.584*** 0.196*** 0.388***
(E) LinearModel
patent 0.538*** 0.906*** 0.577***
(F) Levels of Aggregation
7/10 digit 0.584*** 0.196*** 0.388***
6 digit 0.584*** 0.163*** 0.421***
5 digit 0.584*** 0.129*** 0.455***
4 digit 0.584*** 0.0941*** 0.490***
3digit 0.584*** 0.0111 0.573***

 

Notes: This table reports the coefficients on patents.

59

A.OVQ 4. .mOGVQ .1. A0.0VO 43.3.. ”—0.23 8:85:9m 8.8.5:: * 8853.5
2: 83:25 .3256 :8». 95 33:13 O: 8:255:30 8:85:58: E e8 .2225 .3 .8883: 80:8 EEEEm Ame—20m .839: 52:55
1:: 4.8:: E: 0:: 5.3.2: 9:28.:3 85 .AO :3E8O:581 85:3 a 5A..» .2 :otmsg .3 838: «825.528 85 8.5%; 83.3 mEH ”83.2,.

 

 

 

GmG GAG ENG GANG wad mm
Emma Emma Emma gwmm wGwmm 8058,2820
OZ :2 :2 OZ OZ .3255 409550
8% 8% 8% 8% 8% 49:83: 88%
m0? mm? m®> m®> mc> .AEHESU kuumSUAAH
.3: .88 as: as: 3.2
***ww.v.mn **HMK..O.. ***H©.m#1 ***©N.©Hu ***Nw.fimu HSOAmCOU
.38 38.8 .38 $4.3 see
iEwmA 444mg? 4.3.3: mmmG *mcmA 8:8:
.83 :38 6:: 3.3 :2:
386 23d 22:. ”Ed £8 85:...
:2: 88.8 3:: .23 see
GOAG- nmmGG Amid- FAG- End. 888% E
:2: $8.8 $8.3 38.8 32:
iinvd ***ANA.G ***msm.O 493..sz 4......vwmA E :_
n30
68.8 95.3 83.3 $3.3 $8.8
***GGm.G ivmmod- 4443.56 iESG *44wwGG 853:8: E
:33 68.8 :33 38.8 88.8
***Nnm.o ***vao.o ***©om.o ***®AN.O ***©An.o .Acwfiwm A:
E 3 .3 .3 E A .5888:
33:95 :A 82: :A @3835 :_ @3838 :_ 8:98 :A .I 8388..) EchEQoQ

 

 

433850 O8 main 9:: :QEE @3835 .23 mEchEcer “EA .5389

60

s - 0.786

Elasticity of exports

 

 

high middle low
[WE extensive - intensive]

 

 

Figure 1.1: The Elasticities of Export with Respect to Innovation by Country Income
Level

Note: The height of each bar shows the elasticity of overall export with respect
to innovation in countries with different income levels. "High" represents a high-
income country which has GDP per capita equals twice as the sample mean, "middle"
represents a middle—income country which has GDP per capita equals the sample
mean, and "low" is the low—income country which has GDP per capita equals half
as the sample mean. The numbers in the bars are elasticities of the extensive and

intensive margins, and percentages are the relative contributions of the two margins.

61

0.645

.4
l

Elasticity of exports

.2

   

 

 

homogeneous differentiated

9913;: extensive _ intensive]

Figure 1.2: The Elasticities of Export with Respect to Innovation: Differentiated vs
Homogeneous Products

 

Note: The height of each bar shows the elasticity of overall export with respect
to innovation in different industry groups. "Homogeneous" industries include the
six industries which have differentiated-products shares larger than the median of all
industries; "differentiated" industries represent the other six industries which have
differentiated-products shares smaller than the median of all industries. The numbers
in the bars are elasticities of the extensive and intensive margins, and percentages are

the relative contributions of the two margins.

62

Elastrcrty of exports
4
l
‘ -_
—
31’—
‘3:—
-.—
—
—
x—
—
_
_
_
_
—
=—
:—
1:552—
.—
fi—
_

 

 

 

75767778798081828384858687888998919293949596979899
{‘Zifijr‘ifi'5’2Zi extensive _ intensiveJ

 

Figure 1.3: The Elasticities of Export with Respect to Innovation by Year

Note: The height of each bar shows the elasticity of overall export with respect
to innovation in different years. The numbers in the bars are elasticities of the
extensive and intensive margins, and percentages are the relative contributions of the

two margins.

63

Elasticity of exports

 

 

 

 

fizz-.3} extensive - intensive

 

 

Figure 1.4: The Elasticities of Export with Respect to Innovation by Period

Note: The height of each bar shows the elasticity of overall export with respect.
to innovation in different time periods. "70s" represents the elasticity of 1975—1979,
"80s" represents that of 1980—1989, and "903" represents 1990—1999. The numbers in
the bars are elasticities of the extensive and intensive margins, and percentages are

the relative contributions of the two margins.

64

Chapter 2
Innovation and Export Duration

1 Introduction

Most international trade literature focuses on the cross-sectional trade performance,
such as what types of countries are more likely to become trade partners, what prod-
ucts are traded, or how much do they trade with each other. The duration of trade
relationship is rarely addressed. However, the data. of U.S. imports display that trade
pattern is surprisingly dynamic.28 First, at the product level, the median length of
an export relationship is only two to three years. The short trade duration prevails
in a wide range of countries and industries. Second, the variation in export duration
is large. Even though many trade relationships end in a few years, there is still a
considerable proportion (about 25%—35%) survives at least 12 years. Third, about
30% of trade relationships occur more than once. This means a country enters a.
foreign market, exits it, and then re—enters the market again (and the process can
repeat multiple times). In spite of these unexpected facts, it is still not clear what
the main determinants of the length of trade relationships are, and how to explain the
large variation in export durations. The main object of this chapter is to use survival
analysis to investigate the extent to which innovation accounts for export durations.
Using the number of patents as the proxy for innovation and data. of exports from

105 countries to the U.S. in 1975—2001, I find that innovation has a. significantly pos—

 

2’5 Besedes and Prusa (2006a) first point out that the duration of U.S. import is short.

65

itive effect on the duration of exports. More innovative countries can sustain their
export relationships for a. longer period of time, and this result is consistent. with the
implications of the product cycle model.

The lengh of trade relationships has not been studied extensively. However,
Besedr-is and Prusa (2007) show that for export growth, the survival of existing trade
relationships plays a more important role than building new trade relationships. The
empirical estimation of this chapter is inspired by Besedes and Prusa’s (2006a) pio-
neering work. They first estimate the survivor function of trade durations, and show
that trade relationships are exceptionally short. They then report survival rates of
trade relationships broken out by regions and industries to examine whether trade
durations differ with these factors. This chapter extends their work in the following
ways. First, other than region and industry, I include innovation as well as a variety
of variables (country size, trade barriers and costs, exchange rate, factor endowment,
number of exporters, etc.) in the estimation to analyze the extent to which these
factors can explain the duration of exports. Second, instead of breaking out observa-
tions into groups, I apply a two—stage estimation strategy which can control multiple
explanatory variables at the same time and test the statistically significance level of
each variable. The detailed estimation strategy will be discussed in section 4.

Only a limited number of other empirical studies have examined the durations of
trade. Besedes and Prusa (2006b) analyze the extent to which product differentiation
affects the duration of U.S. imports. Nitch (2007) examines the duration of German
trade and extends previous studies by including more variables, such as contract.
intensity, market share, and two-way trade. Besedes (2008) applies a search model

66

of international trade to explore whether initial trade volume, reliability, and search
cost play important roles in trade durations. My paper contributes to this strand of
literature by offering a new explanation, innovation, to the dynamic trade patterns,
which is based on the product cycle model. This paper also makes contribution to
empirical studies on the dynamics of trade patterns, especially those which are based
on product cycle models (e. g. Feenstra and Rose, 2000). My paper is also related to
literature which stresses zero trade flows (e.g. Helpman et. al., 2007). While zero
trade flows indicate potential trade relationships which are not built, the duration of
trade records the length of time until a trade relationship moves from positive trade
to zero trade.

The remainder of the chapter is organized as follows. Section 2 shows the theo-
retical background of the empirical model. Section 3 describes the data source and
how I compile the export duration and innovation variables. Section 4 displays the
two-stage model of estimation. Section 4.1 presents the model of export duration,
and the estimates of survival function and the length of export relationships. Sec-
tions 4.2-4.4 show the model of regression and the estimated results of the extent to

which innovation affects export durations. Section 5 concludes the paper.

2 Motivation

The trade data demonstrate that expert. durations are short but have high variation.
The duration of trade relationship, however, has rarely been explicitly included in

theoretical models. Standard trade theories appear to imply that. trade patterns

67

should be stable over time. For instance, the Heckscher-Ohlin (H-O) model predicts
that the direction and volume of trade depend on the factor endowment. The short
duration of trade and multiple-spell service then indicate that. countries’ relative factor
endowment change very quickly and switch back and forth. The change of factor
endowment is not a convincing explanation for dynamic trade pattern. Moreover,
theories which consider the dynamic of trade more explicitly also suggest that trade
pattern should be static and persistent. Models which emphasize the importance
of sunk and entry costs (Baldwin and Krugman, 1989) or search cost (Ranch, 2001)
infer that firms tend to export to a market over long periods of time.

Different from these trade theories, the product cycle models can possibly provide
an answer to the highly dynamic trade patterns. These models stress the role of
innovation and technological progress on international trade. Grossman and Helpman
(1991b) construct a. model in which innovation takes the form of quality improvement.
Each product line has numerous potential varieties (producers), but consumers only
buy the variety which has the highest quality.29 The firm (leader) which has the
ability to produce the state-of-art product in each product line can make profit, but
they face the competition from other firms (followers) which produce the second-to—
top products. By investing resources in innovative (or imitative) activities, firms
can increase the probability of success in research, which will improVe the quality of
product. ' Innovation has direct impact on the quality of products, which determines

the pattern of international trade.

 

29Grossman and Helpman (1991b) show that. in equilibrium, this is also the producer which has
the lowest quality—adjusted price.

68

The quality ladder model is capable to explain the short duration and multiple.-
spell of trade relationships. First, a follower can capture the market and become the
leader by conducting a successful innovation. Since innovations take place frequently,
it is not shocking that the leading position can be switched among firms in a short
time, which will result in short durations of exports. Second, the leaders can also
conduct continuous innovations to remain on the top of the quality ladder. This an-
swers why there is still a certain proportion of trade relationships which can last. for
a long time. Finally, previous leaders can also restore their advantageous positions
and take back the markets with successful innovations. This can explain the multiple
spells of service in trade. The model also suggests that countries which have more in-
novations are more likely to move toward (or remain on) the top of the quality ladder,
and are more likely to sustain their exports for a longer period of time. Motivated
by the quality ladder model, the empirical estimation of this chapter focuses on the

effect of countries’ innovations on their export durations.

3 Data

3.1 Export Duration

The duration analysis is based on trade data compiled by Feenstra (1996) and updated
by Feenstra, Romalis and Schott (2002), which contain the export. and import of
the U.S. at the product level. Products are classified according to the 7—digit

Tariff Schedule of the United States Annotated (TS7) before 1988, and are classified

69

according to the 10—digit Harmonized System (H810) in 1989-2001. The dataset.
covers exports from more than 160 countries to the U.S. market. Because I am
interested in the relation between export duration and innovation, the analysis of this
chapter will only focus on products of 12 manufacture industries from 105 countries
whose export and innovation data are both available.30 The sample used in this
paper is thus different from that of Besedes and Prusa (2006a, 2006b) and Besedes
(2008).

The export duration indicates the length of time that an export relationship has
been existed without breaks. Since export data are reported annually, the duration
is measured in years. Table 2.1 lists the export pattern of an example product, gas
meters (HSIO code 9028100000), from 10 countries to the U.S. market between 1989
and 2001. A mark "27" represents that a country exports gas meters to the U.S. in
the given year. For example, Australia exports gas meters to the U.S. in five years:
1997, 1998, 1999, 2000, and 2001. This export relationship lasts for five consecutive
years without a gap, which means the length of this spell of service is five. Argentina
exports this product in twelve years (1989-1999 and 2001). Because of the gap in
2000, Argentina’s export is interpreted as two separate spells whose length are eleven
years (1989—1999) and one year (2001), respectively.

To measure the duration of exports, we need to identify the starting and ending
years of each spell of service. Nevertheless, the data only record exports of each

TS7 product between 1975 and 1988, and each HSlO product between 1989 and 2001.

 

5” Lists of these industries and countries are in tables 2.2 and 2.3.

70

We are not able to observe the exact starting/ ending year of an export relationship
if it starts/ends before or after the record the sample. In table 2.1, the first spell
of Italy’s export starts in 1989 and ends in 1991. The starting year is also the first
year that export data of H810 products are available. We cannot be certain whether
Italy starts exporting the good in 1989 or before it, because we do not have the data
before 1989. Similarly, a country which exports in the last year of the sample (like
Australia’s export in 2001 in table 2.1) does not have an exact ending year of the
trade relationship. Observations like these are inferred to as being censored, and
must be taken into account when we estimate the durations of exports.

Next, a country may re-export a product after it steps exporting it. These cases
are referred to as multiple spells of service. For instance, in table 2.1, Argentina has
two spells; Belgium, China, and Ireland have three spells; and Italy has four spells.
We can treat multiple spells as independent and include all spells in the survival
analysis, but this method implicitly assigns greater weights to products which have
multiple spells. An alternative method is to include only the first spell of each
country—product when estimating the durations of exports (see Besedes and Prusa,
2006a). Estimated results of both cases will be presented in this paper.

The level of product aggregation will change the estimation of survival time. Ex—
port durations are supposed to be longer if we classify products at a more aggregate
level. This paper uses the most disaggregate export data available (TS7 and HS10),
which is important for distinguishing the difference between countries’ export per-
formance in terms of duration. Since there is no appropriate concordance between
TS7 and HS10 codes, the trade durations of the two classification systems must be

71

calculated separately. In the first period (1975-1988), products are classified by TS7;
and in the second period (1989-2001) products are classified by HS10. The longest
observable duration of a TS7 product is 14 years (1975-1988), and that of a H810
product is 13 years (1989—2001). As a comparison, I also calculate the duration of
exports at a more aggregate industry level (5—digit SITC code), which has export.

information covering both periods of 27 years (1975-2001).

3.2 Innovation

The regressor of interest of paper is the number of patents granted by the U.S. Patent.
Office, which is the proxy for countries’ innovations. The patent data come from Hall,
Jaffe and Trajtenberg (2001) and. updated by Hall. Because patents vary enormously
in their technological and economic significance, I weight each patent by the number
of citations that it receives, and the weighted patent counts (WPC) are used to assess
the innovations or innovative capabilities of countries. The linear weighting scheme
which follows Trajtenberg ( 1990) is applied to construct the weighted patent count

index:

WPCJ- = (1 +93) (1)

Subscription j is patent, and gj is the number of citations received by patent j, i.e.
a patent receives no citation has I/VPC = 1, and a patent cited for five times has
IV PC = 6, etc. Patents which are cited for more times are supposed to be more

important and valuable, and receive higher weights when calculating the weighted

72

patent counts.

The advantages of using patent as the indicator of innovation over R&D is de-
scribed in the first chapter and will not be repeated here. I need to construct an
innovation indicator which affects countries’ export durations. I use the number of
weighted patent counts of the first year in each period to measure countries’ innova-
tions; i.e. for TS7 products, I use the number of weighted patent counts of 1975 as
the innovation indicator of the period, and for HSlO products I use that of the year
1989. The weighted patent count is the regressor of interest when I estimate the

effect of innovation on the duration of exports.

3.3 Other control Variables

In selecting variables which may affect durations of trade, it seems intuitive to start
from gravity variables (country size, trade cost, and trade barrier) which successfully
explain trade volume. I include population and GDP per capita to control for country
size. The data are from Penn World Table 6.2. I use the average population and
GDP per capita of each country in the corresponding period. The model of estimation
also contains distance, common language, and common border, which are proxies for
trade costs. These variables do not change over time, and the data are obtained
from Jon Haveman’s International Trade Data. Duty and freight. costs come from
Feenstra et. al. (1996) and Feenstra (2001).

Since this paper highlights the dynamics of trade patterns, the fluctuation of

real exchange rate should be added in the model. The nominal exchange rate and

73

consumer price index (CPI) data. are obtained from World Development Indicator
(WDI) and are used to calculate the change in real exchange rate. Factor endowment.
proportion of each country is based on the skilled-unskilled labor data of Barro and
Lee (2001), and capital and land data are from World Development Indicator (WDI).
I get the factor intensity of each industry from N BER-CES Manufacturing Industry
Database. The interactions between the factor endowment proportion of each country
and the factor intensity of each industry are added in the regression in the robustness
check, which account for their effect on exports as predicted in Heckscher-Ohlin model
(H-O model). Finally, I calculate the number of countries from which U.S. imports
a product. The fact that some products have more exporting countries to the US
market can be connected to higher level of product differentiation, higher U.S. demand
to the product, or higher degree of competition. For each exporting country, I
calculate the average number of exporters of those products that this country exports

to the U.S.

4 Models of Estimation and Estimated Results

Several estimation strategies can be applied to study the effect of innovation on the
duration of exports. First, I can drop censored observations and regress uncensored
export durations on innovation. The estimation does not make use of the corn-
plete sample, and it also causes sample selection bias.31 Second, I can estimate the

survival rate of different groups (e.g. highly innovative countries versus less innova-

 

J’The durations of censored observations are not randomly distributed.

74

tive countries) using the Kaplan—Meier estimator, which accounts for data censoring.
This model uses the information of all observations. However, it can control only
one dependent variable at a time, and the threshold of group classification is rather
arbitrary. As a result, I implement a two—stage process to estimate how innovation
affects the duration of exports. In the first stage, I use the Kaplan-Meier estimator
to estimate the survival function of exports at the product level, and to calculate
the mean and median of export duration of each country-industry. In the second
stage, I regress export durations on innovation, to examine whether more innovative

countries have longer export relationships as predicted in trade theories.

4.1 First Stage: Model of Duration

The export duration indicates the length of time until a country stops exporting the
product. Let T be a discrete random variable which takes on values t k. k = 1, 2, ...... n.
In this paper, T represents the number of years that an export relationship survives.
The probability density function of T is f (1,) = Pr0b(T 2 t,). The survivor function

of T is expressed as

5(1) = WONT Z t) = Ztkzt HQ) (2)

The survivor function of an export relationship describes the probability that an
export relationship survives over t years.
As explained in section 3, we cannot observe the exact startng or ending year of

some trade relationships (e.g. we cannot observe the ending year of a TS7 product.

75

if it is after 1988 because TS7 product information after 1988 is not. available). For
these censored observations, the observed duration length is equal or shorter than the
real duration, and this must be considered in the estimation. I use the K aplan-Meier
estimator which accounts for data censoring to estimate the survivor function:

- (nk — dk)

505) = Htkgt M (3)

nk is the number of subjects which are at risk of being failure right prior to period
tk. It is the number of survivors prior to tk subtracts the number of subjects which
exit the sample (censored observations). In the context of this paper, a failure event
indicates that a country stops exporting a product. So "k is the number of export
relationships which have lasted for tk__1 years, subtract the number of those which
are censored in tk. (1;, is the number of observed failure events which occur at tk. In
this paper, it is the number of export relationships which are active in year tk_1 but
become inactive (fail) in year tk. The Kaplan—Meier estimator is nonparametric and
robust to data. censoring. More details and examples of the Kaplan-Meier estimator
are discussed in appendix D.

Tables 2.4 and 2.5 show the estimated survivor function of export in 1975-1988
and 1989-2001 using the Kaplan-Meier estimator. In table 2.5, 370,616 product
counts (HS10) are exported to the U.S. from the 105 countries in the sample. Some
country-products have multiple spells, so the total number of spells is 550,991. In
the left part of table 2.5, I treat each spells as independent to estimate the survivor

function; and in the right part, I only use the first spells of each country-product

76

in the estimation. The survival curve of the two periods are depicted in figures
2.1 and 2.2, respectively. The properties of export durations can be summarized
by the following points. First, the durations of exports are short in general: only
58%—65% of export relationships survive one year, and 44%-54% of them survive two
years. This indicates about half of trade relationships end within two years, even after
considering censored data. These facts seem to contradict with empirical findings
which show that trade is stable over time at the country level, or even at the firm
or plant level (Roberts and Tybout, 1997; Clerides, Lach, and Tybout, 1998). The
short durations of experts at the product level indicate that although countries or
firms export to the same destination in different years, the products that they export
actually change over time. There exists considerable export turnover at the product
level which is hidden behind the persistent trade pattern at more aggregate levels.
Second, an export relationship can last quite long once it survives the first few years.
The survivor rate in the sixth year is 29%-42%, and it only drops slightly to 25‘7c-40%
in the thirteenth year. This is also demonstrated in figures 2.1 and 2.2 that. the
survival curves become flatter in latter years. Third, the survivor rate estimated
using all spells is lower than that using only the first spells, which implies that the
first spells are generally longer. These properties are similar to the finding of Besedes
and Prusa (2006a). 'As mentioned in the introduction, the main contribution of this
paper to Besedes and Prusa (2006a) is to adopt the two-stage estimation method and
to introduce the influence of innovation on export durations, which will be displayed
in section 4.2.

Using the estimated survivor rates, I derive the median (in which year the survivor

77

rate reaches 50%) and the restricted mean32 (the area under the survival curve) of
export durations. The summary of statistics is shown in table 2.6. The first panel
shows the summary of T87 products in 1975—1988, and the second panel includes the
summary of H810 in 1989—2001. As a. comparison, I also present. the means and
medians of observed spells length, which do not account for data censoring. In each
period, the mean and median of duration based on the Kaplan—Meier estimator is
longer than that based on observed spell length, which confirms that durations are
underestimated when calculated with observed spell length. The statistics of SITC
products are listed in the last panel of table 2.6. I also report the percentile of spell
length in table 2.7 to display the variation and dispersion of export durations.

The results in tables 2.4, 2.5, 2.6 and 2.7 summarize the properties of export
durations at the product level. To study whether more innovative country-industry
have longer export durations, I need the export durations of each country-industry.
I estimate the survivor function and get the mean and median33 of export duration
of each country-industry following the same procedures. The estimated mean and
median of export durations of each country-industry are the dependent variables in

the second-stage regression.

 

32The mean of export durations should be the area under the survival curve. Due to data
censoring, the survival curve does not reach zero at the longest observed duration (thirteenth year
in the first period and fourteenth year in the second period). The restricted mean is the area under
the survival curve up to the largest observed duration. It. will underestimate the real mean of the
duration.

33Some subjects (country-product) have many censored spells and thus their survival rates never
reach 50% until the longest observed duration. This means the median of export duration is longer
than the maximum observed duration, but we cannot estimate how long it is exactly. In these cases,
I use the longest observed length of spells as the median of export durations.

78

4.2 Second Stage: Model of Regression

The second stage of the estimation is to run a regression to find the correlation

between innovation and export duration. The benchmark model is:

0 DP
P0P

 

darationci = 50 + [fiWPCa- + flgpop’alationc + [i3( )6 (4)

+,[34distancec + ,B5languagec + fifiborderc

+,B7dury_ f reightm- + figerchangec + 391% + +5”:

The dependent variable, durationci, is country c’s median or mean of export duration
in industry i. The median or mean of export. durations are estimated in the first stage
as explained in section 4.1. In the benchmark case, I assume that multiple spells
from the same country-product are independent. Results using durations based on
only the first spells will be shown in the section of robustness analysis.

The regressor of interest, the proxy of innovation, is the weighted patent count.

GDP

(WPCci) of the first year in each period. populationc and ( pop

 

)0 represent the
average population and GDP per capita in each period, respectively. These two
variables are country—specific and do not vary by industry but will change over time.
distancec is the distance between country c and the U.S. languagec, and bordcrc
are binary variables which show whether country c uses the same main language
and share a common border with the U.S., respectively. These three variables are

country-specific and do not change with industry or time. di.1,ry_freight(v,; is the

79

sum of duty paid and freight cost relative to the total export value in that. period.
exchangeC is the percentage change of country c’s real exchange rate between the first
and last years of each period. Finally, I include an industry dummy variable (,ai) to
account for industry-specific duration pattern.34 Basically this model looks at the
cross-country variation in innovation to investigate its influence on the duration of
exports. The summary of statistics of the regressors is listed in table 2.8.

The model in equation 4 is estimated by OLS. For each period, I use both the
median and mean of export durations as the dependent variables. The expectation is
that innovation increases countries’ competitiveness, which enables them to maintain
export relationships for longer. The estimated coefficients and standard deviations
are reported in table 2.9. For the convenience of interpreting the results, I also
report the beta coefficients in table 2.10. Table 2.9 shows that in both periods (75-
88 and 89-01), innovation has a. positive and significant impact on export durations.
During the period of 1975—1988, an increase in initial weighted patent counts by
1,000 will raise the median and mean of export duration of a country-industry by
1.27 and 0.83 year, respectively. Considering that the median of duration is only two
years, the influence of innovation is not. negligible. In 1989—2001, a 1,000 increase
in initial patent counts will increase the median and mean by 0.27 and 0.12 year,
respectively. These coefficients are significant at 1% level. From the beta. coefficients
in table 2.10, an one standard deviation increase in patents will increase the mean and
median of export duration by 0.15-0.47 standard deviation. The estimates confirm

the implications of trade theories that countries which have more innovations are

 

34It is possible that U.S. imports products of an industry for a longer time than that of another.

80

more likely to sustain export relationships for a longer period of time.

Except common language, coefficients on other regressors have expected signs,
although some of them are not statistically significant. Larger countries have longer
export durations, and the effect is statistically significant. This is consistent with the
result in Besedes and Prusa (2006a), in which they find that OECD countries have
longer export. durations than non—OECD countries. Higher trade costs and trade
barriers reduce the durations of export relationships. Countries whose currencies
experience higher rates of depreciation (against U.S. Dollar) have longer export dura-
tions, though the effect is not significant. These findings hold in both periods, either

when the median or mean of export duration is used as the dependent variable.

4.3 Robustness Analysis

I conduct several robustness checks. First, I reestimate equation 4, but this time the
durations of exports are estimated using only the first spell of each country-product
in the first-stage estimation. The estimates are stored in table 2.11. Second, I
estimate the survivor function of each country-industry using the more aggregate
industry level data (5—digit SITC). Different from exports data at the product level
which are recorded by T87 and H810 in two periods, export data of SITC industries
are available throughout 1975-2001. The estimated results are presented in table 2.12.
Using only the first spells or measuring durations at more aggregate do not alter the
finding that innovation has a significantly positive effect on export durations. Third,

I estimate the benchmark model with all variables taken natural logs (except binary

81

variables). Estimated coefficients now represent the elasticity of export duration.
In table 2.13, the estimates indicate that a 10% increase in weighted patent counts
will raise the export duration by 0.4%—1%, and the effect is still statistically very
significant.

Finally, I include the interaction between countries’ factor endowment 1')roportions

and industries’ factor intensities:

 

. ,. GDP . _
durationm- = [30 + [BIW FCC,- + figpopc + 133( pop )c + [34distc (5)

+,B5langc + 561)de + B7dury_freighta- + flgeamhangec

+39endowc X intensity,- + fi 101% + +€ci

For each country c, I develop the endowment ratio of skilled labor, unskilled labor,
and capital stock, all made relative to land. I also calculate the factor intensity of
skilled labor, unskilled labor, and capital stock of each industry i. The interaction
(endowc X intensityi) is intended to account for possible endogeneity problem caused
by omitted variables. The Heckscher-Ohlin (H-O) model predicts that cross~sectional
trade pattern is determined by the factor endowment, and this influence may exist in
the durations of exports. The interaction is not included in the benchmark model
because 21 out 105 countries in the sample do not have complete factor endowment.
information. The estimated results are in table 2.14. Among the estimates of
three factors, only that of the skilled labor has a correct sign and is statistically
significant. But the inclusion of factor endowment does not. change how irmovation

affects export durations: more innovative countries still have longer export durations

82

in both periods.
In summary, the main finding of this paper is robust to alternative estimation of

duration and model specifications.

4.4 The Number of Exporters

The number of providers of each product differs greatly in U.S. imports. This differ-
ence can be related with several factors. First, the number of providers of a. product.
can represent the degree of product differentiation or sophistication. According to
Besedes and Prusa (2006b), the export duration of differentiated products is longer
than that of the homogeneous products. Second, the fact that U.S. imports a prod-
uct from more countries than other products can imply that U.S. has special demand
to that product (e.g. specific design for each buyer or contract relation). Third, a
higher number of exporters may reflect more intense market competition, which may
reduce the durations of exports.

In this section, I construct an exporter index to examine how the number of
exporters affects trade durations. For each product, I identify the number of exporters
in the year that this product firstly enters the U.S. market. For each country, I
calculate the exporter index which equals the average number of providers of those
products that this country exports to the U.S.35 A higher exporter index of a country
means that the products that this country exports have more exporters.

I estimate the following model using OLS:

 

37’Suppose that a country exports three products to the US during 1975-1988. The number of
exporters of these three product in the first year that they are exported to the U.S. are 10, 20, and
30. 80 the average number of exporters faced by this country is (10 + 20 + 30) / 3 = 20.

83

_ GDP
durationcz- = 80 + fiIWPCm- + 32populationc + x33( pop

 

)C + Bjdistancec (6)
+fi5languagec + B6borderc + [i7dury_fr€ighta

+58erchangec + ,39erporterc + /310;i,- + +56,-

The regressor erporterc is the exporter index of country c. All other variables have
the same definition as in equation 4. Tables 2.15 and 2.16 show the estimates. The
coefficients on the exporter index are positive, meaning that a country which exports
products with more exporters will have longer durations of trade relationships. The
result supports the hypothesis that more exporters implies higher degree of product
differentiation. The effect of innovation on export durations remains positive and

significant.

5 Conclusion

Trade pattern is more dynamic than generally thought. Approximately half of export.
relationships last for only two to three years, but another one third survives 12 years
or more. In addition, many countries enter and exit a foreign market. repeatedly.
In spite of the evidence, literature does not offer a clear answer to what. the main
determinants of the length of trade relationships are, and how to explain the large
variation in export durations. Using the number of patents as the proxy for innovation

and data of exports to the U.S. market, I find that more innovative countries have

84

longer durations of exports. The finding provides a new explanation to the volatility
of trade pattern, and is consistent with the implications of the product cycle model.
The length of export relationships also increases with country size, and decreases with

trade barriers and costs.

85

same.» 9: E 8355 may mtoaXm REESE a :33 839:5“ =X:
”+52 .58 was $9 5953: .m.D .2: 3 metacsoo 3 Soc .6me mam .3:on 035me as mo 53.3 “.898 m5 mopetmcoficw 2an $3. ”332

 

H H x assess
N. H X X X X X X X Gmfiwam
2 H X X X X X X X X X X X X X semen.
S v X X X X X X X X X XE:
m. m X X X X X 983.:
N. m X X X X X X X 6:30
ma H X X X X X X X X X X X X X 33250
c m X X X X X X SEwEm
m H X X X X X dfldhumfldw
Q m X X X X X X X X X X X X eamucmwfix
X5550

 

38% £on mo Sow ooom 33 mag Rx: .83 no.2 «4mm: mag 33 SE 093 $3
mot/Em $355: 38w

oooogwmom Havoc 3mm
38on mew ”8300.5

 

mzwmm floaXm— mo cEEeXm AN Bash.

86

Table 2.2: List of Countries

 

Algeria
Argentina
Australia
Austria
Bahamas
Bahrain
Belgium
Belize
Bolivia
Brazil
Bulgaria
Cameroon
Canada
Chad
Chile
China
Colombia
Congo
Costa Rica
Cyprus
Denmark

Dominican
Ecuador
Egypt

El Salvador
Ethiopia
Germany
Fiji
Finland
France
Gabon
Ghana
Greece
Guatemala
Guinea
Guyana
Haiti
Honduras
Hong Kong
Hungary
Iceland
India

Indonesia
Iran

Iraq
Ireland
Israel
Italy

J amaica
Japan
Jordan
Kenya
Korea
Kuwait
Laos
Liberia
Madagascar
Malawi
Malaysia
Mali
Mauritania
Mauritius
Mexico

Morocco
Nepal
Netherlands

New Zealand

Nicaragua
Niger
Nigeria
Norway
Oman
Pakistan
Panama
Paraguay
Peru
Philippines
Poland
Portugal
Qatar

South Africa

Romania

Saudi Arabia

Senegal

Sierra Leone
Singapore

Spain

Sri Lanka

Sudan
Suriname
Sweden
Switzerland
Tanzania
Taiwan
Thailand
rIrinidad and Tobago
Tunisia

Turkey

Uganda

United Arab Emirates
United Kingdom
Uruguay
Venezuela.
Zambia
Zimbabwe

 

87

Goa
GOA
om:
om:
mad
5.0
owd
3.0
end
and
3.0
5.0

wagsmeoz H8 355:me
mfienm 2038/ H302

33.25an mqoamfloh.
E82580 omcofioflm

382 3.5855 vmueoflnem
wficcom was wan—38m XESEA
maoswoka 8.8280 .maofieuaov $30
838m .83de

wficmmm Ezflobom

$83850 $36632

wetneh sesameem mo 223:5
ESE «5:30 J52 33m

mac—.605 @338 98 35858.5
EmEQBUm comumfioamdefi.

EmEQSUm 0805030 was 38:85—
.EmEQBUo Use 385er Embmsgfi
32603 :32: woueomfiafi

matumgofi 3me basin

$0:on 8.05:8 95 £de SEQ .maoum
3035.5 83me msoeqezmommfi was ~25:an
mposwoa 88 was EsmfioSonm

3260.5 @225 was mfieomfimno

@3355 ES 255B

£2603 Umatafi was pooh

mm
mm
mm
mm
vm
mm
mm
on
mm
mm
mm
on

 

Exam poscoaéouesqemta

boweueObsm oEEeXm

somuacommm OHm

 

 

a 2m sass s seas? we a: as sea.

88

.aonHHEmm SEZLEExX e5 .3 33:58 $2605 .3 226:5 5.23:5 2: 2: 3.5%: .833 mEH ”33?.

 

 

amiss: as

ogmo mooo o mooo vH wovmo mooo o mooo vH
ovnmo 33 o 533 mH wovmo 33 o mmHoH mH
ovnmo moo mm mmHHH mH wovmo meH mm msmHH mH
wasmo womm E wmnoH H H Hhvmo mmvo voH mow: H H
mohmo oon wwm moowH oH mwvmo smvm m3 wowom oH
nmwmo onnm o: vmvnm o mvmmo wmvo omo 2: mm o
vHomo mmoH mooH oswwm w mmomo obom oovH msmmm w
nmomo Chm hmmH voomm N. Hnsmo omov vHoH 3an N.
memo moom mooH vommm o vowmo vav mtm Hmvmv o
nmmmo mvwv moon omomv m mwom .o wvoo o3 m wvmnm n
mmomo ommm omvo mmvom v nwmm .o omoo ommo flown» v
moovo ovva nmvo 8me m omwmo HmmmH @1va mHmmoH m
monvo mmom H mnomm wwwomH m Hugo mmmoH mewm nmmmoH m
mmwno ovam HmH HoH wmwmvm H mwwmo H H va omvovH nonmmm H
.:E Stardom umo‘H mam <33 Ewmm mail .85 Hoe/35m 80H mam ~30... Ewmm mag
23% ”$an £on HH<

 

 

$2.33 :26qu 52:5 as 233.

89

.SEECmm Emma/Tcfioex of .3 33:58 3260.5 mo 228:3 Notizm of 2: £593 £23 mEH ”$3.2.

 

 

2838: 2mm

oHovo ommvm o ommvm mH ovvmo ommvm o ommvm mH
oHovo omit o moowm mH ovwmo omvv o moowm mH
oHov.o ”mom Hem omoHv H H oven .o voom Hem onwmv H H
mmovo mva Hom mvovw oH wovmo Nbom own 5an S
ooovo woo."V mg vooov o vowmo omom oHo mmmmm o
moovo omHVN. mm» Nuvmmn w mmmmo mHNo mmHH monvo w
ovHvo omooH moH H woooo N. Nonmo oN.HmH onH mwmoN. N.
vao owmmH oomm smoow o Hwomo wwva mmom woNHoH o
nmmvo HommH hog 333 m wmeo owmmH goo ommomH m
nmmvo momw mon HvHHmH v ovovo wmmmH oome omHmeH v
mmwvo oomoH omme ohmmvH m mvao anwH wommm onHvoH m
vommo ommNH omem HoomoH m mmono mnwom momma nmoonm m
wmvoo mmmmv oomHmH oHooNh H mumoo memo Hmmmom Hooonm H
.:E Sexism «mod 3% H33 Ewmm 085. .Eo Hog/39m 80H mam H33 Ewmm warm
mommm amok 36% :<

 

 

288%: 8:25 833% new 238.

90

622mm :08 E maofigzw fiomxm we :2:va “Ea 588 23 mo mosmfifim 32.298me 9: .32an £23 £23. £8922

 

 

 

 

 

82% m mwmmd 2 2:3 25 26% 222E
23.52 m 32K m £53 23% =<
2833; 895
235 m 2:3 a $82 28 220% ii
235% m 9&3 m 38%. 23% =<
2838: 2%
Manama m 822% 2 28% 28 26% 2222
85.22 m 322 2 8mm.“ 23% =<
AwmeS: $9

fidmfima EGGS Gamfima Gama

£on mo .02 282838 37M LEE: =QO 82,530

 

 

2535.53 wmaaasmm ”mesmsfim mo baafism 6d 2an

91

 

 

 

 

 

R R m 2 2 mm 22 2 2 2 2225 23% 3.222
R 22 m 2 2 222 23 m 2 2 23% 2222
22833: 82.222
222 2.2 m 2 2 m2 3 m 2 2 222s 23% 3222
m2 m2 2.. 2 2 w 23 m 2 2 23% 2222
288-2222: 2222222
222 222 m 2 2 a m. 2 2 2 2225 23% 22222
22 222 m 2 2 2 m 2 2 2 23% 22
622233: EH

05% 0%» 05% $22 2.2: $2222 232 03%. 23% 2.222

8382222 22-2

222M222 23% 33.2.2330

 

 

m22osa222m 20 3222282322 ”Em £an

92

202$ 322320.232 m on 3
2532.525 m2 Awwumhv 222w¢_c->.22€ .26 5222:2522 32: E #522222: gimme: «EH. 2.2323232 .20 murmtfim 922232.233 @222 3.209: 33222 2225 ”9:222

 

 

mm: 222322.23 2222228. $23223 52:22 223$ 33.2 322233 32
2% 258.2222 $22228. $33.22 22233 $2322 32 33232333 32
2222 $382.2 22 £828.22 823322.22 222.-me 2233222222223
«32 82223.2 222328.22. 232822.22 22258.22 222.33 22322242226
E2 2 0 22282.22 23222.22 32:22
«32 2 o 32223.22 $22.22 2223222
$2 2232 322.3 322.35 2332222 8:323
22 mama $3323 82.23% «3.33. 2222-228 3228 32 220
2232 22:32“ 22%;? 22.222; 228.8% $22.32 82228 32 220
92 8382282 528% 2382.222 323%? 2222-232 222
$2 «$2322.82 23.232232 3222222322 2232223222 $22.22: 222
22.22 23.2823 22 33.5% $20.28 2222-232 023
2222 3.2222222 22 ~33: 322.2222 $232.2 0&5
2320 .oz é: 2222.2 22% 232.2 222.25

 

 

m2¢mmw2wmm ”82223.3me 20 fiafiasm ”wN 333.8

93

Table 2.9: Estimated Results of the Benchmark Model

 

1975-1988 (TS7)

1989-2001 (HS10)

 

 

 

Median Mean Median Mean
(1) (2) (3) (4)
patent 1.27*** 0.834*** O.271*** 0.123***
(0.000196) (0.000219) (5.06e-05) (2.83805)
population 0.0136 0.0418*** 0.0437*** 0.0472***
(1.19809) (1.538-09) (1.118—09) (6.149210)
GDP per capita 0.0501*** 0.0948*** 0.0850*** 0.113***
(1.59e—05) (2586—05) (2.10e-05) (1.88805)
dury_freight -0.795*** —1.370*** —0.767*** -1.383***
(0.111) (0.237) (0.175) (0.176)
distance -0.0467 -0.0128 -0.0169 -0.0174
(1.798-05) (2.66e—05) (3.04805) (3.40e-05)
language 0.183 -0. 198 -0.0833 -0. 306
(0.253) (0.242) (0.380) (0.283)
border 3.014** 2.872*** 5.169*** 3.380***
(1.424) (0.415) (1.100) (0.317)
real exchange rate 0.000471 0.00435 0.00583 0.00514
(0.00231) (0.00490) (0.00780) (0.0132)
Constant 1.146*** 3.351*** 0.995*** 3.164***
(0.182) (0.297) (0.276) (0.324)
Observations 984 984 1128 1 128
R2 0.422 0.477 0.373 0.485

 

Notes: The estimated specification is equation 4. Coefficients on patent, GDP per capita,
and distance are multiplied by 1,000; coefficients on population is multiplied by 107.

Robust standard errors clustered by country are in parentheses.

Industry dummies are

included but not reported. * indicates significance level: *** p<0.01, ** p<0.05, * p<0.1.

94

Table 2.10: Estimated Results: Beta. Coefficients
1989-2001 (H810)

1975-1988 (TS7)

 

 

Median Mean Median Mean
(1) (2) (3) (4)

patent 0.477 0.325 0.281 0.153
population 0.0554 0.177 0.266 0.343
GDP per capita 0.175 0.345 0.286 0.456
duty-freight -0. 110 -0. 197 -0.0803 -0. 173
distance -0.00916 -0.0261 -0.0251 —0.0310
language 0.0399 —0.0447 -0.0141 —0.0621
border 0.236 0.233 0.290 0.227
real exchange rate 0.00302 0.0289 0.00975 0.0103

 

Notes: Estimated beta coefficients of the regression model in equation 4.

Table 2.11: Robustness Checks: First-spell only

 

 

First spell

 

TS7 median TS7 mean

H810 median

H810 mean

 

 

(1) (2) (3) (4)
patent 1.44*** 0.917*** 0.266*** 0.141***
(0.000322) (0.000260) (6.89e—05) (3.51e—05)
population 0.0195 0.0484*** 0.0517*** 0.0507***
(1.55409) (1.79509) (1.13e-09) (7.16e—10)
GDP per capita 0.0677*** 0.110*** 0.134*** 0.140***
(2.14505) (2.97e—05) (3.07505) (2.28e-05)
dury_freight -0.869*** -1.379*** -0.620*** -1.292***
(0.138) (0.266) (0.224) (0.200)
distance ~0.0239 -0.0216 -0.0331 —0.0204
(2.26e—05) (2.90605) (4.43e-05) (39073.05)
language 0.201 -0.240 -0.114 -0.321
(0.324) (0.270) (0.496) (0.335)
border 4.178** 3.294*** 6.478*** 3.952***
(1.932) (0.509) (1.238) (0.386)
real exchange 0.00167 0.00505 0.00943 0.00719
(0.00267) (0.00530) (0.0119) (0.0156)
Constant 1.134*** 3.429*** 0.930** 3.168***
(0.247) (0.330) (0.401) (0.364)
Observations 984 984 1 128 1 1 28
R2 0.445 0.477 0.381 0.493

 

Notes: The estimated specification is equation 4. Coefficients on patent, GDP per capita,
and distance are multiplied by 1,000; coefficients on population is multiplied by 107.

Robust standard errors clustered by country are in parentheses.

Industry dummies are

included but not reported. This table reports estimated results in which durations are
calculated based on only the first spell of each country-product. * indicates significance

level: *** p<0.01, ** p<0.05, * p<0.1.

96

Table 2.12: Robustness Checks: SITC Industry

 

 

 

 

 

SITC
SITC median SITC mean
(1) (2)
patent 2.45*** 1.46***
(0.000772) (0.000480)
population 0.0455 0.108***
(3.15e—09) (3.28e-09)
GDP per capita 0.187*** 0.253***
(5.29505) (4.82405)
dury_freight -0.349 -2.600**
(0.518) (1.092)
distance -0.0699 -0.0137
(6.33505) (6.654e-05)
language 0.556 -0.364
(0.982) (0.714)
border 10.43** 8.464***
(5.100) (1.403)
real exchange 0.000705 0.00101
(0.00479) (0.00439)
Constant 0.0204 3.361***
(0.577) (0.654)
Observations 948 948
R2 0.445 0.537

 

Notes: The estimated specification is equation 4. Coefficients on patent, GDP per capita,
and distance are multiplied by 1,000; coefficients on population is multiplied by 107.
Robust standard errors clustered by country are in parentheses. Industry dummies are
included but not reported. This table reports estimated results in which products are
classified by 5-digit SITC code. * indicates significance level: *** p<0.01, ** p<0.05, *
p<0.1.

97

Table 2.13: Robustness Check: Variables in Natural Logs

 

 

1975-1988 (TS7)

1989—2001 (HS10)

 

 

Median Mean Median Mean
(1) <2) <3) <4)
patent 0.0989** * 0.0648*** 0.100*** 0.0386* **
(0.0142) (0.0101) (0.0139) (0.0108)
population 0.0389*** 0.122*** 0.0869*** 0.165***
(0.0101) (0.0125) (0.0151) (0.0147)
GDP per capita 0.0391** 0.168*** 0.0817*** 0.247***
(0.0163) (0.0280) (0.0147) (0.0279)
dury_freight. -0.115*** -0.182*** -0.0781*** -0.102***
(0.0200) (0.0297) (0.0149) (0.0265)
distance -0.0201 —0.0744** —0.0462 ~0.132***
(0.0247) (0.0284) (0.0283) (0.0378)
language 0.0121 0.0163 -0.00131 0.0512
(0.0396) (0.0300) (0.0416) (0.0360)
border 0.243 -0.0936 0.384*** -0.137
(0.156) (0.0783) (0.0885) (0.130)
real exchange rate —0.000588 0.00611 -0.00593 ~0.00117
(0.00577) (0.00715) (0.00596) (0.00732)
Constant -0.172 -1.566*** -0.934*** -2.309***
(0.324) (0.488) (0.352) (0.499)
Observations 983 983 11 25 1 125
R2 0.550 0.665 0.562 0.664

 

Notes: The estimated specification is an alternative version of equation 4 in which all
(except binary) variables are in natural log form.
country are in parentheses. Industry dummies are included but not reported. * indicates

significance level: *** p<0.01, ** p<0.05, * p<0.1.

98

Robust standard errors clustered by

Table 2.14: Robustness Check: Including Factor Endowment

 

 

1975-1988 (TS7)

1989-2001 (HS10)

 

 

 

Median Mean Median Mean
( 1) (2) (3) (4)
patent 1.36*** 0.762*** O.256*** 0.128***
(0.000318) (0.000243) (6.70e-05) (3206.05)
population 0.0215 0.0474*** 0.0525*** 0.0485***
(1.61409) (1.71509) (1.10409) (6.50510)
GDP per capita 0.0780*** 0.119*** 0.147*** 0.151***
(2.63e—05) (3.51405) (3.67405) (2380-05)
dury_freight —0.858*** -1.455*** —0.691** -1.417***
(0.205) (0.408) (0.295) (0.263)
distance ~0.0450 —0.0341 -0.0538 -0.0124
(2.88e-05) (3.45e—05) (5.34e—05) (4.46405)
language 0.316 -0.195 -0.170 -0.443
(0.398) (0.318) (0.604) (0.374)
border 4.036** 3198* ** 6. 243*** 3. 786* * *
(1.873) (0.515) (1.252) (0.416)
real exchange rate 0.00230 0.00453 0.0148 0.00773
(0.00228) (0.00446) (0.0127) (0.0151)
unskill_labor -0.00154 -0.00252* -0.00148 -0.000817
(0.00102) (0.00138) (0.00137) (0.00106)
skill_labor 0.0580 0.110** 0.0238*** 0.0129***
(0.0392) (0.0488) (0.00415) (0.00312)
capital —1.74e—07 —3.85e—07 -1.08e-07** -7.14e—08
(2.28e—07) (3.02407) (5.22408) (5106—08)
Constant 1.084*** 3.509*** 1.008** 3.193***
(0.295) (0.399) (0.435) (0.379)
Observations 840 840 960 960
R2 0.473 0.541 0.423 0.545

 

Notes: The estimated specification is equation 5. Coefficients on patent, GDP per capita,
and distance are multiplied by 1,000; coefficients on population is multiplied by 107.
Robust standard errors clustered by country are in parentheses. Industry dummies are
included but not reported. * indicates significance level: *** p<0.01, *"‘ p<0.05, * p<0.1.

99

Table 2.15: Estimated Results: Number of Exporters 1975-1988

 

 

1975—1988 (TS7)

 

 

Median Mean Median Mean
(1) (2) (3) (4)
patent 1.26*** 0.829*** 1.07*** O.769***
(0.000194) (0.000218) (0.000135) (0.000209)
population 0.0144 0.0421*** 0.0146 0.0422***
(1.21e—09) (1.54e-09) (1.23e-09) (1.54e—09)
GDP per capita 0.0508*** 0.0952*** 0.0520*** 0.0955***
(1.60e-05) (2.59405) (1.62405) (2.61605)
dury_freight -0.804*** —1.374*** -0.797*** -1.372***
(0.110) (0.236) (0.110) (0.237)
distance -0.00608 —0.0134 -0.00595 -0.0134
(1.79405) (2.66e-05) (1.80e-05) (2.679-05)
language 0.190 —0. 195 0.187 -0. 196
(0.253) (0.242) (0.253) (0.243)
border 2.995** 2.863*** 2.984** 2.860***
(1.415) (0.412) (1.408) (0.411)
real exchange rate —0.0000961 0.00409 -0.0000940 0.00409
(0.00272) (0.00510) (0.00268) (0.00509)
exporter 0.0331* 0.0152 0.0433*** 0.0185
(0.0171) (0.0185) (0.0162) (0.0182)
exporter*pa.t.ent 9.35x10-5*** 2.98x10-5***
(3.17505) (7.294906)
Constant 0.802*** 3.194*** 0.674** 3153*”
(0.292) (0.373) (0.276) (0.372)
Observations 984 984 984 984
R2 0.425 0.477 0.443 0.479

 

Notes: The estimated specification is equation 6. Coefficients on patent, GDP per capita,
and distance are multiplied by 1,000; coefficients on population is multiplied by 107.
Robust standard errors clustered by country are in parentheses.
included but not reported. * indicates significance level: *** p<0.01, ** p<0.05, * p< 0.1.

100

Industry dummies are

Table 2.16: Estimated Results: Number of Exporters 1989-2001

 

 

19892001 (H810)

 

 

Median Mean Median Mean
(1) (2) (3) (4)
patent 0.253*** 0.109*** 0.581*** 0.255***
(4.68e—05) (2.60905) (6.31905) (7.56905)
population 0.0406*** 0.0449*** 0.0400*** 0.0447***
(1.07909) (6.01910) (1.07409) (5.87910)
GDP per capita 0.0804*** 0.110*** 0.0760*** 0.108***
(1.96905) (1.86905) (1.851e-05) (1.85905)
dury_freight -0.747*** —1.368"‘*”< -0.755*** -1.372***
(0.180) (0.168) (0.178) (0.167)
distance -0.0180 -0.0182 -0.0185 -0.0184
(2.88905) (3.27905) (2.76e-05) (3.25405)
language -0.0442 -0.277 -0.0381 —0.274
(0.363) (0.270) (0.353) (0.267)
border 4.842*** 3.139*** 4.828*** 3.133***
(1.059) (0.309) (1.038) (0.304)
real exchange rate 0.00844 0.00706 0.00853 0.00710
(0.00721) (0.0131) (0.00711) (0.0130)
exporter 0.140*** 0.103*** 0.127*** 0.0972***
(0.0448) (0.0356) (0.0450) (0.0357)
export.er*pat.ent -4.55 ><10_5*** —2.02 X 10_5***
(3.70e—06) (7.16e-06)
Constant -0.294 2.215*** -0.111 2.296***
(0.559) (0.451) (0.550) (0.455)
Observations 1128 1128 1 128 1128
R2 0.415 0.518 0.430 0.522

 

Notes: The estimated specification is equation 6. Coefficients on patent, GDP per capita.
and distance are multiplied by 1,000; coefficients on population is multiplied by 107.

Robust. standard errors clustered by country are in parentheses.

Industry dummies are

included but not reported. * indicates significance level: *** p<0.01, ** p<0.05, * p<0.1.

101

TS7 (1975—1988)

 

 

 

analysis time

| ————— all

 

 

f irstl

 

Figure 2.1: The survivor function of TS7 products in 1975-1988 using the Kaplan—
Meier estimator

102

HS10 (1989-2001)

1 .00

0.75
l

 

 

 

 

0.50
1
l
I

 

0.25

 

0.00
l

 

_

j
0 5 10 15

anaiy sis time
l ----- all HQ

 

 

Figure 2.2: The survivor function of H810 products in 1989—2001 using the Kaplan-
Meier estimator

103

Chapter 3

Network Externality, Compatibility, and
Foreign Entry Mode

1 Introduction

Foreign direct investment (F DI) stock grew dramatically in the last 15 years of the
twentieth century. This trend slows down in recent years, but the growth of FDI still
outpaces the growth of world income. Firms undertake FDI for many reasons: sav-
ing trade cost, reducing production cost, or gaining strategic competition advantage.
On the other hand, F DI also induces costs such as loss of economies of scale. and
the costs of disintegration.36 These arguments can be summarized by the proximity-
concentration trade-off” and the factor—proportions hypothesis. These theories are
also used to explain different patterns of FDI across firms, industries and countries.
A deficiency of these theories is that they only focus on. explanations from the supply
side. The impact of the difference from the demand side has not been broadly stud-
ied. This paper develops a differentiated—product duopoly model that. includes the
economies of scale on the demand side (network externality) to analyze the difference
of F DI patterns across industries and firms.

Network externality indicates the utility that a. consumer derives from using a good

 

3bThese theories are summarized in Navaretti and Venables (2004).

“Firms choose to enter the foreign market through FDI when the gains from saving trade
costs outweigh the cost of building a subsidiary in a foreign country. See Horstmann and
Markusen(1987,1992), and Markusen and Venables(1998, 2000). For empirical evidence see.
Brainard(1997).

104

depends on the number of people using the same brand or compatible goods. Katz
and Shapiro (1994) summarize different types of network externality and the sources
of them. Direct network externality is defined as the value of a product directly
increases with the number of people using the same or compatible products. The
most obvious examples are communication products, such as telephone, fax machine,
and E—mail. A telephone that can not call out or receive from other users is not
valuable. The indirect network externality comes from market mediated effects. The
system which has a larger network usually has more varieties of complements and
easily accessible services. A prominent computer operating system attracts more
companies to design software that can be used on it, which increases the utility that
users derive from it. For both types of network externality, consumers benefit from
making the same choice as most other people do.

A central concept related to network externality is compatibility. In this paper,
the level of compatibility measures the degree of interdependency between the network
size of two rivalry systems. Two systems are (partially) compatible if the utility that
an user derives from a system depends not only on the number of agents using the
same system, but also on the number of agents using the other system. A Hotmail E—
mail account is compatible with a Yahoo E-mail account because users having different.
accounts can still send e-mails to each other. By this definition, Yahoo messenger and
MSN messenger are not (perfectly) compatible since agents using different messenger
systems cannot contact each other directly.

This paper highlights the role of network externality and compatibility in explain-
ing the following facts of international investment. First, firms producing compatible

105

products are more likely to undertake FDI in same locations than firms producing
incompatible products. I use two examples to demonstrate that firms’ FDI decisions
differ with the degree of products’ compatibility. The’first example is the personal
computer market. Dell, Hewlett-Packard(HP), and Apple are multinational compa-
nies selling personal computers and laptops worldwide. Dell and HP sell computers
using Microsoft Windows system, while Apple sells computers using Mac OS. The
FDI destinations of these three firms match the compatibility of their products. Dell
mainly invests in India, China, and Canada. HP mainly invests in India, China, and
Singapore. Apple mainly invests in UK, Japan, and Denmark. These FDI destina-
tions show that Dell and HP, which produce more compatible products, choose same
main FDI locations; while Apple, which produces less compatible product, chooses
different F DI locations. The second example is the digital camera market. A digital
camera needs memory cards to store the data. The standards of memory card include
Secure Digital Ca.rd(SD), Memory Stick(MS) and, xD—picture Card(xD). Among
these standards, xD is used only on cameras produced by Olympus and Fujifilm.
The fact is Olympus and Fujifilm also have the same FDI destinations: their largest
FDI destination is China, followed by the U.S. The main FDI locations of other
Japanese companies which have camera as their main products are: Nikon’s main
FDI locations are China and South Korea; Canon’s main FDI locations are Vietnam
and China. These examples reveal the positive correlation between firms’ product
compatibility and their main F DI locations. The factors affecting firms’ FDI decision
can be complicated, but these examples inspire me that network externality and com-
patibility may play a role. Second, FDI is more important in industries with strong

106

network externality. Since high—technology industries usually present. stronger net-
work effect, this result provides a new explanation to a stylized fact in international
trade that FDI is more prevalent in technology-intensive and skill-intensive industries
(Markusen 2002; Navaretti and Venables 2004). Third, firms which have more ex-
isting users (which I refer to a. greater installed base) are more likely to undertake
F DI. Since larger and older firms are more likely to have a greater installed base, the
last finding corresponds to another stylized fact in trade literature that firms which
undertake FDI are on average larger and older than firms which do not undertake
F DI.

The impact of network externality effect on international market has not drawn
much attention. The only paper, to the best of my knowledge, which relates network
externality and international entry mode is completed by Klimenko and Saggi (2007).
They examine the preferences of a foreign firm and the government of the host country
under the effects of network externality and compatibility over two entry modes: the
de novo entry by the foreign firm and the acquisition of the domestic incumbent.
There are at least three main differences between their paper and mine. First,
in their model, firms choose either to enter through greenfield FDI or acquisition
and merger (M&A). In my paper, firms choose between entering through FDI and
exporting. Second, the market structure of their paper is the competition between
one firm in the host country and one firm in the FDI source country. In my paper,
the host country does not produce, and competition is between two firms from foreign
countries. The model is especially suitable to analyze high-technology industries in
which companies from developed countries compete in developing countries which

107

cannot produce the products by themselves. Third, the main object of Klimenko
and Saggi (2007) is to explore how the host country’s government should set optimal
policies to increase welfare under network externality. My paper focuses on how
network externality affects firms’ strategic incentives to undertake FDI.

The remainder of the paper is organized as the follows. In section 2, I outline the
benchmark model to analyze the impacts of network externality and compatibility on
equilibrium prices, market shares, and foreign entry decisions. In section 3, I study
the case that one firm has more existing customers (a larger installed base) in the
foreign country before it directly competes with the other firm. The main result of
this section is to find how an installed base imposes different effects on the incumbent
and. the entrant. Section 4 is the conclusion and I discuss possible extensions of this

model.

2 Benchmark Model: No Installed-Base

There are three countries in the model, denoted as A, B, and Foreign. Two firms, A
and B, are located in country A and B, respectively. The number of firms is fixed due
to the high corporate-level fixed cost. The two firms produce differentiated systems
and sell the products in the foreign country. The foreign country itself does not
produce. The structure of the model is similar to the strategic trade policy literature
(e. g. Brander and Spencer, 1985) that two countries export to a third country. I refer
the model of this section to the symmetric case because the two firms are identical in

all aspects except they produce differentiated systems. This is contrary to the model

108

in section 3, which is defined as the asymmetric case, in which one firm has existing
users (an installed base) in the foreign market before it directly competes with the
rival. The producers play a two—stage game. At the first stage, they determine how
to serve the foreign market. There are two entry modes available to each firm: export
(X) or foreign direct investment (F). At the second stage, the firms compete in the
foreign market by setting prices. The equilibrium is found by backwards induction

beginning with the second stage.

2.1 Preferences of Consumers

In this section, I derive the demand function of a duopoly model in the presence of
network externalities. There is a. unit mass of consumers in the foreign country,
and each consumer has an inelastic demand of one unit of the good (one system).
The total value of a. system is constituted by its stand-alone utility and network
utility. The stand—alone utility a consumer derives from a system depends on his
taste, 2:. The tastes of consumers are uniformly distributed along the interval [0,1].

A consumer who has taste :1: derives the stand-alone utility which equals

11 — as: if he or she buys from firm A (la)

12 — (1(1 — .7') if he or she buys from firm B (1b)

17 is the base value of the systems. A consumer who has a high .7: prefers system

B to system A. The parameter (1. measures the degree of products’ differentiation.

109

Equation 1 shows that the disutility consumers derive from choosing the less preferred
system increases with a. This setting is the same as the Hotelling model. We
can think :1: as consumer’s location on the unit interval and a as the transportation
cost. The locations of firm A and firm B are fixed at the two end points, 0 and 1,
respectively. Consumers who live close to point 0 prefer A to B because buying from
A induces less transport cost.

The network utility of a system depends on its network size; i.e. the number of
agents joining the compatible network. I assume consumers have perfect foresight,
so they can correctly anticipate the network size at the moment they make their

purchase decisions. The network size of firm i is:

yi=Si+7S-; 57:43, 2'45 (2)

y,- denotes the size of network of system i; S,- denotes the equilibrium output level
(which is also the market share) of firm i. I assmne that the base value 1.: is high
enough that in equilibrium the whole market is fully covered by the two producers:
S A + S B = 1. The network size of a system depends not only on its own output level
but also on its rival’s output level, and the degree of interdependency is determined
by 2,. I express 2," E [0.1] as the degree of compatibility between the two systems.
For instance, if y = 0, the two systems are incompatible and the network size of each
system is solely determined by its own market share. If 7 = 1, the two systems
are perfectly compatible and always have an identical network size: yA = yB 2

SA + SB = 1. I allow partially compatible in the range 0 < y < 1. If 7 = 0.5,

110

y A = S A + 0.553. The network size of A depends on A’s and B’s market. shares
jointly, but A’s market share has a greater weight. than B’s market share. I assume
that the degree of compatibility is predetermined. before the firms enter the foreign
market.

Combine the stand—alone utility and network utility, a consumer of type :1: has a

total utility net of price which equals

a = o — (12: + ny A — p A if he or she buys from firm A (3a)

2 v — (1(1 — :27) + nyB — p3 if he or she buys from firm B (3b)

p A and p B are the prices of good A and B, respectively. n is the strength of network
effect. A large 71. means consumers get higher utility from network effect.38 From
equation 3, we can see that the utility consumers derive from product i. increases with
the size of network 3),.

Each agent. purchases the system that maximizes his or her utility. To get the
demand function faced by firm A and B, I find the consumer with type 37" who
is indifferent from buying good A to good B. Type. .73“ must equate 3a and 3b:
v — a:r* + my A — p A = v — a(1 — 16*) + nyB — [23. In equilibrium, consumers who

have type 1: < :L‘* buy from firm A and those who have type :1: 2 1* buy from firm B.

 

3”The setting of the utility is similar to Katz and Shapiro (1985) and Klimenko and Saggi (2007).

111

Solving this equation, I get. the demand functions faced by the two firms

a_n'(1—’)’)—PA+PB

 

 

: * : 4}.
SA x 20. — 2n(l —- 7) ( I)
a-n(1-7)+pA-I)B
=1- * = 4}
SB :1: 2a — 2n(1 —— y) ( ))

2.2 The Second Stage: Price Competition Game

At the second stage, each firm has to set a price simultaneously under the entry
modes that they chose at the first stage. There are four possible combinations of
entry modes at the beginning of the price competition game:

(1) XX: Both firms export.

(2) FF: Both firms undertake F D1.

(3) XF: Firm A exports and firm B undertakes FDI.

(4) FX : Firm A undertakes FDI and firm B exports.

The direct trade—off between FDI and export can be explained by the proximity-
concentration hypothesis. The benefit of investing in the foreign market is to achieve
proximity to consumers. It saves the trade cost, such as transportation cost and
tariff. The benefit of exporting is to achieve concentrating production. It saves the
fixed cost of building a plant in the foreign country. F irms’ profit functions under

these two entry modes are:

112

7n = (p,- — T)S,- if firm i enters through exporting (53.)

77,- = p15,; — f if firm i enters through investing (5b)

Notation 'r is the unit. export cost, and f is the fixed cost of FDI. S2: is market
share of firm i in equation 4. Without loss of generality, I assume the unit production
cost equals zero in all countries.39 A firm enters the foreign country by exporting
has a marginal cost equals 7 and has no additional fixed cost. A firm enters through
FDI has zero marginal cost but has a fixed cost equals f. This cost structure shows
FDI is more likely to be beneficial to the firm capturing a large market share, while
exporting is more likely to be beneficial to the firm has low level of output.

Suppose at the first stage both firms enter through export (mode XX). The
profit functions of A and B are as equation 5a. The firms then choose their prices
simultaneously to maximize the profit of each. By setting the first order derivatives

equal zero, I can get firms’ best—response functions

1

2 (a —- n- + no)
1

2 (a — 71. + m»)

 

fla—n-%A+MrHW)= 0 W)

 

M-n+pA-%m+00 = 0

 

39In reality, an important advantage of making F DI in developing countries is to save production
cost. In my model, assuming a positive production cost will not change the trade-off between export
and FDI. Carrying out FDI still leads to a lower unit cost but higher fixed cost. It is just now the
unit. cost F DI saves includes both the product-on cost and trade cost.

113

Solving the best response functions simultaneously gives us the equilibrium price

of mode XX:

p§X=p§X=a—n0—v)+r c)

The superscript denotes the entry mode: " XX” means that. both firms enter through
exporting. Plug these prices into the demand functions in equation 4 and profit

functions in equation 5a. The equilibrium market shares and profits are

. 1 -
51.21424 2 52ng = § (8)
1
wfx=w§X=§M—ne—4» e)

Similarly, we can find the equilibrium outcomes of the other three modes. In
mode FF, the profit function of both firms is equation 5b. In mode XF and F X, the
exporter’s profit function is equation 5a and the investor’s profit function is equation
51). In each mode, firms maximize their profits by choosing prices. I derive the
equilibrium prices, market shares, and profits from solving the first order conditions.40
The results are listed in table 3.1.

To ensure that both firms have positive prices, market shares, and profits in all four
possible equilibrium entry modes, I assume a > 1137 + n(1 — ’7) and a. > 2f + 77(1 — 2,)
The products differentiation a must be relatively large than the network effect 71(1—7),

otherwise the strong network externality effect will dominate and drive consumers to

 

"0 See Appendix E for detailed steps of finding these results.

114

end up buying from only one firm.

Lemma 1, 2, and proposition 1 summarize the findings from Table 3.1.

Lemma 1 The equilibrium prices of both firms in each entry mode will increase with
(1) an increase in the level of products difierencda), {2) an increase in the level of
compatibilityfl'). and ( 1’) a decrease in the strength of network ertemalityfiz).

8p;Cl (911kl 871k]

Proof. From Table 3.1: W =1 > 0, 77};— = n > 0, 777‘,— = —1 < 0, for Vital = X, F
and i = A, B. I

(1) of lemma 1 verifies the principle of differentiation in industrial organization.41
Under a standard price competition, firms have greater market power when their prod—
ucts are more different and thus they can set higher prices and make more profits. (2)
seems to contradict with (1) if we think greater products differentiation is the same as
lower degree of compatibility. In fact, greater compatibility increases the dependency
of a firm’s profit on the size of its rival’s network, and thus decreases the intensity
of competition between the two firms.42 Contrary to compatibility, (3) shows that

strong network externality effect induces firms to compete more aggressively to get a

large market share.

Lemma 2 Firms will set a lower price when they enter through F DI, and will set a

higher price when they enter through exporting.

Proof. Suppose firm B exports. Firm A will set a higher price when it enters through

exporting than when it enters through F DI: table 3.1 shows pix > pix. Similarly,

 

"l See Tirole (1988) chapter 7'
“See Klimenko and Saggi (2007)

when B enters with F DI, firm A will charge a higher price when it exports than

.XF FF.

invests: p A > p A When the two firms take different actions at the entry game

(mode XF or F X ), the price of the investor is lower than the exporter: pg]: < p‘XF
and pix < pgx. I

Lemma 2 shows the strategic effect of FDI in this model. In a standard price
competition model, the equilibrium prices and sales depend on firms’ marginal costs.
A firm sets a lower price when its marginal cost is lower. From equation 5a and 5b,
we see that firms have lower marginal cost when they enter through FDI. Thus a
firm competes more aggressively in the price competition game if it enters through
F DI in the entry game. Lemma 2 compares the equilibrium prices across different

entry mode and when firms take different entry actions, but lemma 1 describes the

comparative statics for each given entry mode.

Proposition 1 When export and FDI coeaist in the market, the ratio of the FDI
firm's market share to the erporter’s market share increases with the strength of net—

work ezrtern ality ( n )

Proof. Let R denotes the ratio of the output of the FDI firm to the output of

 

the exporting in mode XF and FX. From table 3.1, R = (% + 6(a_nT(1_7)))/(312 —

>0. I

 

 

. T . We can verifv that dB 2 (STU—m")

(1(a—n(1—7))) ., a; (3n—3a+T—3n'T)2
Proposition 1 presents one of the three main finding of this paper. From lemma

2, we know that the investor sets a lower price than the exporter. The lower price

directly attracts more consumers to buy from the investor. But a. strong network

externality strengthens this effect and increases the market. share ()f the investor even

116

further. Network externality effect (direct or indirect.) is especially prominent. in high-
teehnology industries, such as information and telecommunication. From proposition
1, we expect that F DI is more important in high—technology industries, which verifies

a stylized fact in FDI literature.

2.3 The First Stage: Entry Game

In section 2.2, I determined the equilibrium prices and profits for each of the four
possible entry modes. At the first stage, each firm chooses to enter either through
export or through FDI. To find the equilibrium, we need to determine each firm’s

optimal entry decision given its rival’s action.

1. The rival enters through export

Consider the case that the rival enters through exporting. To simplify the analysis,
I introduce the notation Ark to denote the attractiveness of FDI relative to exporting,
i.e. the profit of choosing F DI subtract the profit of exporting, where k = X, F
indicates the entry action that the rival undertakes. Consider the case that. firm B
enters through export. If firm A chooses to export, the outcome is mode XX in
Table 3.1 and A’s profit equals riff. If firm A decides to enter through FDI, the
outcome is mode FX and A’s profit equals nix. Firm A will choose FDI if and
only if AnX(T, f. 7, n) = ”EX — TEX > 0. Similarly, given A exports, B will choose
FDI if AnX (T, f,"y,n) = ngF — HEX > 0. In summary, when the rival exports, a
firm will choose F DI if and only if Aux > O. I analyze the impact. of each of the four

parameters on the entry decision by calculating the comparative static effects. The

117

signs of the derivatives are:43

dAer dAer dAnX dAnX
> 0, —— < 0, —— < 0.

> 0 10
(17 df d7 ' dn ( )

 

 

The first two derivatives demonstrate that Aer increases with expert cost and
decreases with FDI cost. This matches the proximity—concentration argument that
a high export cost and a low F DI fixed cost encourage firms to invest in foreign
markets. The other two derivatives show that the attractiveness of F DI decreases
with compatibility and increases with the strength of network externality, given the
rival enters through export. In lemma 2, we know that the investor in mode XF or
F X competes more aggressively in the price game and captures a larger market share.
How much additional market share it can capture depends on the strength of network
externality effect and the degree of compatibility. Under stronger network externality,
the investor can capture a larger market share. High compatibility, however, tends
to equate the network size and weakens the advantage of the only investor. So the
strategic incentive for a firm to enter through FDI given its rival exports increases
with n and decreases with ’7, because FDI is more likely to be optimal to the firm

with a. high level of output.

2. The rival enters through FDI
Consider a case that the rival directly invests in the foreign market. A firm will

choose FDI if and only if AWF(T, f,'y, n) = nil? — riff 2 NSF — ”EX > 0. The

 

”The proofs are in appendix F-

118

comparative static effects are:44

dArF > 0 time” < o dAer > 0 dAer
(17 ’ df ’ d7 ’ dn

 

 

 

< 0 (11)

The proximity—concentration trade-off still hold: a high export cost and a low FDI
fixed cost make a firm more willing to enter through FDI. The signs of derivatives
with respect to compatibility (q) and network externality effect (n) are different from
the previous case: now F D1 is more attractive when compatibility is higher and when

network externality effect is weaker.45

The intuition is as follows. If my rival
makes FDI, I can either enter with F DI and share the market equally with him, or
I can enter by exporting. The incentive of exporting is to save the FDI fixed cost.
So exporting is optimal when the saving of FDI fixed cost outweights the sum of the
variable export. cost, and this will sustain only when my market share is small enough.
From table 3.1, the market share after switching to export is increasing with n and
decreasing with 7, which means export is optimal when network effect is strong or
when compatibility is low. Or equivalently, FDI is optimal when network effect is

weak or when compatibility is high.

These comparative static effects can be summarized in proposition 2:

 

“The proofs are in appendix F.

45At first glance this result is counterintuitive, because in the previous case I Show that a strong
network externality effect and low compatibility make F DI more attractive. The point is: a firm
can never become the only investor when its rival enters through FDI. In fact, in this case we are
looking for the strategic incentives to be the only exporter. Lemma 2 demonstrates that firms set a
higher price when they export. The only exporter in equilibrium earns a higher marginal revenue
but loses some market share. The smaller output level the only exporter has, the more trade cost it.
saves by entering through export. Thus the attractiveness of entering through exporting increases
with network externality effect and decreases with compatibility. This implies the attractiveness of
entering through F DI decreases with network externality and increases with compatibility.

119

Proposition 2 (I) The attractiveness of entering through F DI always increases with
carport cost (T) and decreases with F DI fired cost ( f ) ( 2) Given the rival exports, the
attractiveness of entering through F DI decreases with compatibility (7) and increases
with the strength of network eflect (n). {3) Given the rival enters through F DI, the
attractiveness of FDI increases with compatibility and decreases with the strength of

network e flect.

To find the Nash equilibrium of the entry game, I need to consider both firms’
optimal entry decisions simultaneously. As indicated in proposition 2, the attrac—
tiveness of FDI depends on T, f, 7, and n. In fact, these four parameters can be
classified into two categories: proximity-concentration factors (T and f) and network

externality factors (7 and n). As we can see in both cases, the two factors in each

 

k
dA: and

category always have opposite impacts on the attractiveness of FDI, i.e.
gdf— have different signs and d‘gfl and dA;_k have different signs for k— — X, F. To
reduce the number of cases to analyze, I will pick one parameter from each. category
to show how these two forces determine equilibrium entry mode.

Figure 3.1 shows the market outcome of the no—installed-base case. The two letters
in each region show the equilibrium entry mode. Curve (i) in figure 3.1 indicates

the parameter values of export cost('r) and compatibilityh) that a firm is indifferent

between exporting and FDI when its rival enters through export (AirX = O). From

 

X
equation 10, we know that d—A—L’: > 0 and am JIX < 0, and thus the locus of Aux = 0
is positive sloping. FDI is optimal in the area above the curve (Arx > 0) and

export is optimal in the region below the curve (AirX < 0). Curve (ii) indicates

120

the parameter values that a firm is indifferent between exporting and F DI when its

. X
rival enters through FDI (AirF = 0). Equation 11 shows that %:— > 0 and

dAnX

(13' > 0, and so the locus of AirF = 0 is downward sloping. F DI is optimal

F > 0) and export is optimal in the region below

in the area above the curve (An
this curve (ArF < 0). In region (I), the market equilibrium is that. both firms
undertake FDI (FF). Any point in this area is above curve (i) (Aer > 0) and (ii)
(AnF > 0), which means FDI is a dominant strategy because the export cost 7 is
relatively high. In region (III), the equilibrium is that both firms enter through
exporting (XX) because the export cost is relatively low. Any point located in this
area is below curve (i)(A7rX < 0) and curve (ii)(A7rF < 0), which means export is
optimal regardless of the rival’s action. In region (II), the Nash equilibrium is that
the two firms make different entry decisions (F X or XF). This is because FDI is
the best action when the rival exports (Aux > 0), but exporting is the best. action
given the rival invests (AnF < 0). From figure 3.1, we can see the height of region

(11) decreases with compatibility, while the height. of area (I) and (III) increases with

compatibility. This brings us the following proposition.

Proposition 3 The two firms are more likely to take same action to enter a foreign
market when compatibility is high, and they are more likely to choose different entry

action when compatibility is low.

This proposition shows an important finding of this paper that. the F DI (.lecisions
of firms producing products with network externality are related to the compatibil—

ity of their products. It reproduces the FDI pattern in personal computer market

121

and digital camera market. Firms producing compatible products tend to choose
same FDI locations, and firms producing less compatible products choose different

locations.

3 Model with an Installed-Base

In this section I address a model in which one of the two firms has an installed base
in the foreign market. An installed base of the incumbent indicates that there are
some existing consumers using the incumbent’s system before the two firms com-
pete directly in the market. This installed—base or lock-in effect appears in many
high-technology industries. In the personal computer operation system(OS) market,
Microsoft Windows has more than 95% market share.46 In the personal computer
CPU market, Intel has about 77% of the market.” In the domestic market, an
installed base can affect consumers’ choices of technology and can change firms’ mo-
tivation of innovation. (Farrell and Saloner 1985, 1986). I name the model in this
section the asymmetric case because only one firm (the incumbent) has an installed
base while the other firm (the entrant) does not. In this section, I analyze how an
installed base imposes different effects on the incumbent’s and the entrant’s entry
decisions in a foreign market.

I assume that the installed base is formed before the two firms compete directly,

and existing consumers who adopted the incumbent’s system do not switch after the

 

”According to ITFacts: in 2007, there are 665 million copies of XP installed worldwide, giving it
74.3% share of the PC installed base. Other versions of Windows account for another 190 million
units, or 21.6% share.

“In 2006, Intel has 77.7% of the x86 CPU market. AMD’s x86 market share is 22.3%.

122

entrant enters the market. Since the installed base is part of the network, new

consumers take it. into consideration when they make their purchase decisions.

3.1 Preferences of Consumers

Let firm A be the incumbent and have an installed base in the foreign country before it
competes with firm B, the entrant. The installed base does not. affect the stand-alone

utility. It enters the total utility function through the network size:

yA = m‘l'SA‘l'TSB (128)

313 = SA+"/’SB (1%)

Notation m measures the size of the installed base, and it is predetermined. we
can think the symmetric case in section 2 (equation 2) as a special case of equation
12 by setting m = 0. All other assumptions and settings are identical to section 2.1.

With the installed base, the demand functions of the incumbent and the entrant are

S = 73* Z a — n(1— y) — pA + 123 +mn(1— 7) (13")
A ‘ 2a — 2n(1 — 7)) (

* a — n(1— ’y) +pA — PB — mn(1— 7)
S = 1— = ‘
B T 2a — 2n(1— 7)) (Mb)

 

 

123

3.2 The Second Stage: Price Competition Game

To find the market outcome of the price competition game, I derive the best—response
functions of the incumbent and the entrant in each of the four entry modes. Following
the same procedures as in section 2.2, I can solve for the equilibrium prices, market.
shares, and profits under different entry modes.48 These results are reported in table
3.2.49

With an installed base, the two firms set different prices and have different profits
even when they choose the same entry action (XX and FF). In addition to lemma

1 and 2, the asymmetric case result also shows the following property:

Lemma 3 The incumbent ’5 price, market share, and profit increase with the size of
installed base m under all entry modes. The entrant ’8 price, market share, and profit

decrease with m.

Proof. From table 3.2: 7%} = :1;'n(1 — y) > 0, pm 2 —:1;n(1 — 7) < 0; .7}? :
kl
6((L—7l(1—’7')) > 07 am — -6(a—TL(1)—’))) < 0, A2,l — X, F I

A greater installed base m makes the incumbent’s product more attractive rela-

tively to the entrant’s, and leads to higher price and market share to the incumbent.

 

4“I assume a — n(1— '7') > (1/3)T + (1/3)mn(l — '7) and (a — (1 +(1/3)m)n(1— 7)) x (a —— (1+
(1/3)m)n(1 — 7)) > 2f(a - n(1 — 7)) to ensure both firms have. positive prices, market shares and
profits in all four possible equilibrium entry modes.

"9 See Appendix E for detailed calculations.

124

3.3 The First Stage: Entry Game

To find the equilibrium of the entry game, I firstly determine each firms optimal entry
decision given its rival’s action. I discuss all four cases, since with the installed base
the two firms are no longer symmetric. To simplify the expression, let. Arr/llC denotes
the attractiveness of FDI to firm i given its rival takes action k, where i = A, B and
k = X, F. For instance, if firm B exports, firm A will undertake FDI if and only if
Ari: 2 Fix — 7734(X > 0. If B enters through FDI, A will undertake FDI if and only if
Ari = riff—n32”? > 0. For firm B, FDI is the best action ifAng = ngF—ng“, > 0
when A exports; F D1 is the best action if Avg 2 NSF —— HEX > 0 when A invests.
In summary, given the rival’s action is k, firm i will undertake FDI if and only if
Anf > 0.
I summarize the signs of comparative static effects in table 3.3.50

The signs of derivatives with respect. to export cost (T) and FDI fixed cost (f) for
both the incumbent and the entrant are consistent with the no—installed-base case.
A higher export cost. and a lower F DI cost make FDI more attractive. The signs
with respect to the size of installed base (m) demonstrate that the chance that the
incumbent enters through FDI increases with m, and the chance that the entrant
enters through F DI decreases with m.

The signs of derivatives with respective to y and n are different for the incumbent.

and the entrant. As shown in Table 3, these signs are also affected by the size of

installed base m. To clarify the impact of network externality and cmrmatibility

 

"’“See Appendix F for proofs.

on equilibrium entry mode, I discuss it with two cases: the small—installed—l)ase case

(0 < m < 2%) and the large-installed-base case (m 2 Z).

1. Small installed base: 0 < m < 2%

With a small installed base (m < 275), the possibility to enter through FDI de-
creases with compatibility (increases with the strength of network effect.) when the
rival enters through export; the possibility of FDI increases with compatibility (de-
creases with the strength of network effect) when the rival enters through FDI. This
conclusion is the same as proposition 2, in which neither firm has an installed base.
Although the installed base makes the equilibrium prices and profits different, the
impact is not strong enough to reverse the direction of marginal effects of 7 and n on
firms’ best response.

Figure 3.2 shows the equilibrium entry mode with a small installed base. Curve (1)
indicates the parameter values of export COSt-(T) and compatibilityh) that makes the
incumbent indifferent between exporting and FDI, given the entrant enters through
export (Ari: = 0). Curve (iii) shows the T and 7 that makes the entrant indifferent
between exporting and F DI, given the incumbent exports(A7rj§ = 0). These two
curves are upward sloping as curve (i) in Figure 3.1. We can see, that curve (i) locates
below curve (iii) for V y < 1: a lower export cost is needed to induce the incumbent.
to enter through exporting. When the two products are perfectly compatible, i.e.
e,- = 1, the entrant can take fully advantage of the incumbent’s installed. base and
thus curve (1) and (iii) merges at 7 = 1.

Curve (ii) indicates the parameter values of T and f that. makes the iI'lcumbent

126

indifferent between exporting and F DI, given the entrant enters through FDI (Ar/I"; =
0). Curve (iv) demonstrates the T and f that makes the entrant indifferent between
exporting and FDI, given the incumbent enters through FDI (Aflg = 0). These two
curves are downward sloping, as curve (ii) in Figure 3.1. Again the incumbent requires
a lower export cost. than the entrant to find exporting to be optimal. Compare the
equilibrium entry mode in figure 3.1 and figure 3.2, the different point is that. in figure
3.2 there are two regions, (II) and (IV), that only the incumbent. undertakes FDI.
The small installed base enables the incumbent to capture a larger market share, and
this makes F DI more likely to be beneficial to the incumbent than to the entrant.
When m —> 0, curve (i) and (iii) will converges and curve (ii) and (vi) will converge,
and figure 3.2 eventual becomes identical to figure 3.1.

2. Large installed base: m > 2%

Figure 3.3 shows the parameter space of equilibrium entry mode when the in-
cumbent has a large installed base. Curve (i)-(iv) have same definitions as in figure
3.2 which indicate the parameter values of export 7' and 1 that satisfy Ari = 0,
AWE = 0, Ang = 0, and Ar; = 0, respectively. We can see curve (iii) is downward
sloping and curve (ii) is upward sloping, which are different. from those in figure 3.2.
With a. large installed base, the incumbent always finds F DI more attractive when
compatibility is low (or when network effect is strong). This effect is independent
of the entrant’s action, because both curve (i) and (ii) are upward sloping. On the
contrary, the entrant always finds FDI more attractive when compatibility is high
(or when network effect is weak), and this condition holds regardless of incumbents
action because both curve (iii) and (iv) are downward sloping.

127

I summarize the result of figure 3.2 and 3.3 in the following proposition.

Proposition 4 (I) The incumbent is more likely to enter through F DI than the en-
trant. (2) The attmctlveness of F D1 to the tnrrumbcnt increases with the installed base

(m), and the attractiveness of FDI to the entrant decreases with the installed base.

In figure 3.2 and 3.3, the regions that the entrant undertakes FDI is smaller than
the set of the incumbent.51 The second part of proposition 4 comes from the third
column ”m” of Table 3.3. The intuition for proposition 4 is: the installed base makes
the incumbent-’5 product more attractive and lets the incumbent to capture a larger
market share. Since F DI benefits firms with high levels of output, it is more possible
for incumbent to find F DI to be beneficial when it has an installed base, and the
benefit increases with the size of the installed base.

Proposition 4 provides an explanation to the fact. in international trade: large
and old companies are more likely to undertake FDI than small and new companies.
Systems of large firms usually have a relatively larger installed base or lock-in effect
than those of small firms. FDI is more attractive to these large firms because the

network externality effect helps them to capture a larger market. share.

4 Conclusion

In this paper I examine firms’ different entry modes in an industry with network

externality. A firm’s optimal entry mode depends on several factors. First, it depends

 

‘3]In figue 2, the regions that incumbant possibly undertakes FDI are 1, II, III, IV, and those of
the entrant are I and III. In figure 3, the regions that the incumbant possibly undertaks F D1 are I,
II, III, IV, V; the regions that the entrant possibly chooses FDI are I and V.

128

on the export cost and FDI fixed cost, as the proximity-concentration hypothesis
predicts. Second, it depends on the strength of network effect and the degree of
cornrmtibility. Third, it depends on the size of the installed base in the foreign
market. The result shows that FDI is more important in industries with strong
network externality, and FDI is more attractive to firms with large installed base. The
model also shows that firms producing compatible products have stronger incentive
to choose same FDI locations than firms producing incompatible products.

There are several ways to extend this paper. The first direction is to build a
dynamic model. In this paper, the size of the installed base is predetermined before
the two firms compete directly in the foreign market. In a dynamic model, we can
let the incumbent. enters the market in the first period as a monopolist. The sales in
the first period form incumbent’s installed base at the beginning of second period. A
multiple-period model also enalbes us to study the optimal timing to enter through
F DI. By intuition, the advantage of an installed base gives the incumbent an incentive
to overprodiure in the first period, and thus the incumbent. may enter earlier through
FDI in an industry with network externality than in an industry with no network
externality.

The second extension is to let the firms choose the degrees of compatibility en-
dogenmisly. In the long run, firms can choose the degree of compatibility by changing
the design or by producing adopters. In section 2.1, I show that if the two firms are
symn‘ietric (no installed base), equilibrium prices and profits increase with compat-
ibility. In an asymmetric case, firms" incentives to make products compatible may
be different. A change in compatibility has two effects. First, a. higher compatibil-

129

ity increases consumers’ willingness to pay as in the symmetric case. Second, higher
compatibility reduces the advantage that the incumbent gains from the installed base.
By intuition, the entrant should always prefer a higher degree of compatibility be-
cause both effects benefit it. The incumbent, however, may have an incentive to
make the products incompatible because a decrease of the importance of the installed
base reduces incumbent’s market. power over the entrant. In this paper, compatibil-
ity further affects firms’ entry modes and thus the problem is more complicated and

requires further studying.

130

Table 3.1: Equilibrium price, market share, and profit of each entry mode in the
no—installed—base case

 

 

 

 

 

 

 

Mode P S 1r
XX p§X=a~G+r S§X=% nix: a_G
p§X=a—G+T S§X=% n§X=—2—(a—G)
FF pfiF = a — G st =§ «if = (“‘G) — f
ng=a_G SSF=% 7rFF_a-G)_f
G 2
XF _ 2 XF XF _ a‘ ‘3’
XF PA —a‘G+3T SA _2_6a—G 7TA r 2a—G
—G+1r 2
XF_ _ 1 XF_1 T XF_(a 3)
PB — a 0+ 3" SB — 2 + 6(a—G) ”B — 2(a—G) ‘f
1 2
FX _ _ 1 FA _ 1 7 F (0 ‘G+3T)
FX PA —“ 0+3” 5 _2+6(a—G) 7rA — 2(a-G) ‘f
—G~1'r
FX , _ 2 FX _1_ r FX " 3

 

 

Notes: To simplify the expressions, I let G = n(1 — 'y).

131

.2. 1 :E: n m as a. 1 a: n e 5:

.mcowmmnzmxm 2: >535? an. ”$32

 

Blew
ml:

Imn
xwm

01% ml 1
$+M+Hkam

starts.

Nb

 

1......” a
Elm” em
Amlsv a RN

I m I.I
mm $+U elk

mm+tm+01snb

m
Na

V
4A

mun

 

 

 

m H 1.5%

E ml at
m +Hlmam

l6”

IGHK

m
kka

.1
RR

mm

 

 

51ch I w I mm

IGHN

l6”

m
KR

ems

KN

 

m alxx
E n1

m +HH 5%.
m

mm10
mm+e
mm1t+e
mm+t+e
a

 

$62

 

momma Ezfimfi as we; < SE can? 358 his :83 m0 .505 was

.985 8x28 63.5 EEEEHEH ”NM mime

132

Table 3.3: Comparative statics effects

 

 

r f m ’y n
m. < 2% m > 2% in < 5% m > 77;,

 

 

mg (+) (-) (+) (—) (—) (+) (+)
mf,‘ (+) H (+) (+) (-) (-) (+)
Mg (+) (-) (-) (-) (+) (+) (->
m; (+) (—> o (+) (+) o o

 

Notes: The (+) and (-) in each cell indicates the sign of derivative of As with
respect. to each of the parameters on the top of the table. For instance, the (+)
. . BAWX
corresponds to row Ar} and column 7 1nd1cates TA— > O; the (-) corresponds
T

a . . 8Ar
to row Ari, and column f 1nd1cates ——4’ < 0.

0f

133

 

 

tau 0.70 -r
(I) FF
0.65 a\§ A
0.60 ~— (11) F X or XF
(i)
055--
(111)xx
0.50 ti¢i¢i:§¢§s§:§+¢;§,{
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
gamma

Figure 3.1: Nash equilibrium of the entry game of the no—installed-base case

134

 

 

 

 

 

 

 

0.70 ~-
tau \
0.65 + (1]) FX \
MK
.. ‘— \b‘
t (11)
060 __ (III) FX or XF
1 (iii)
- __ 1v FX,/”w:—-*-"""""'" f” if
035 .-~(~--”“")" - (I)
«l (V) x,x
0.50=%=%#%:%¢%fi‘+#+rt=tat
0.0 0.] 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
gamma

Figure 3.2: Nash equilibrium of entry mode when the installed base is small

0.8 ‘" \
tau

0,, . (II) FX

t

0,6 __ (111) FX

 
    
 
 

 

 

(ii)
//I\l) Fx //""’7/. iiiiii r-""'f//’
()5 «r /,,,//”"/T' (l)
.. ”’ (v1) xx
04:1:1fie::%:i:e:+‘r::;4i
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 |.0

gamma

Figure 3.3: Nash equilibrium of entry mode when the installed base is large

136

Appendix

A Innovation Data

Since the focus of this paper is innovation at the industry level, the first task is to
classify each patent into an industry (or industries). In the patent dataset, most.
patents could not be classified into a single 2—digit SIC industry. USPTO has de-
veloped a classification system for the technologies to which the patented inventions
belong, consisting of about 400 main patent classes. In Hall et. a1. (2001) NBER
dataset, each patent class has one or more corresponding SIC-sequence code(s).

Each patent SIC sequence corresponds to a 2—digit SIC industry, but not all 2-
digit SIC industries are covered. Also, a 2—digit SIC industry can contain more than
one patent SIC sequence. My sample contains the 12 industries which have at least
one corresponding SIC-sequence code. For patent classes which have more than
one corresponding SIC-sequences, I assume each SIC-sequence has equal share and
calculate the number of "patent count" of each industry. For example, suppose a
patent class has 3 corresponding SIC—sequence: 1, 3 and 4. SIC-sequence 1 matches
with SIC industry 20, and SIC-sequence 3 and 4 match with SIC industry 28. Patents
of this SIC-sequence have 1 / 3 weight going to industry 20 and 2/3 weight going to
industry 28.

Some empirical studies use simple patent count as a proxy for innovations. This
assumes each patent. has equal quality and importance. I take the number of ci-

tations received by each patent to measure quality and importance, and use it as

137

weight to calculate the "weighted patent counts" as a proxy for innovation. The
average number of citation received by each patent is 5, and I set the weight of each

52 This selection of weight seems to

citation received equals 100% of patent counts.
be arbitrary, but I will compare the results using different weights in the sensitivity
analysis. It turns out the results of using different weights are very similar.

There are two indices of year of each patent in the sample: the application year and
the grant year. In general, the application year is earlier than the grant year because
it takes time for the Patent Office to examine each patent and to decide whether to
grant. it or not. The average lag between application and grant year in the sample is
1.97 years, with more than 95% of patents are granted within 3 years after they are
applied. In economic sense, we should classify patents according to their application
year rather than grant year, because countries’ innovation capability or technology
level are reflected on the year patents are applied, not the year patents are granted.53
The patent dataset records patents which are granted until the year 2002. A patent
is recorded in the database only after it is granted, not when it is applied. This
means if I count the number of patents according to the application year, I have data
truncation: I cannot see most patents applied in 2002 in the database, though many
of them should be included in the sample because they are going to be granted in
following years. In other words, although the data covers patents granted until the

year 2002, I cannot accurately measure the number of patents applied in each year

 

”For example, a patent which has been cited for five times will be 1+5=6 patent counts.
"“F or example, a patent applied in 1990 and granted in 1995 should reflect the innovation capability
of the country in 1990, not 1995.

138

especially in the last few years of the sample.“ To mitigate the possible measurement
bias caused by truncation, I drop observations of the last three years (2000-2002) and
use observations in 1975-1999 as my benchmark sample. As I mentioned, more than
95% of patents are granted within three years after they are applied. So the effect. of
truncation to patents applied before 1999 should be relatively small. I will split my
sample by year and use them separately to estimate my model, to show the conclusion
is not affected by inclusion\exclusion of observations of the last few years. I will also
compare results using grant year to classify patents (which does not suffer from the
truncation problem) with results using application year, and there is no material

difference.

 

5" The number of granted patents applied in 1999 is 146,000. This number drops to 107,037 in
2000, to 38,867 in 2001 and to 1,664 in 2002. Obviously, the low numbers of patents in at least the
last tWO years is due to data truncation.

139

B Data of Factor Endowment and Factor Intensity

I consider four factors of production: skilled labor, unskilled labor, capital, and land.
The data of skilled and unskilled labor shares are from Barro and Lee (2001). Skilled
labor is defined as those who have completed the secondary school education and
above, and unskilled labor is defined as those who have attained secondary school but
not completed and below. The Barro and Lee (2001) data contain the percentage
of each level of education in every five years, so I use interpolation to calculate the
percentages of each education level in years between. I multiply the skilled and
unskilled labor shares by population to get. the skilled and unskilled labor endowments
of each country in each year. The capital stock and land data are obtained from
W'orld Development Indicator. Skilled labor, unskilled labor, and capital are divided
by land to get. the factor endowment proportions of each country-year.

The data of production factor intensity of each industry come from NBER-CES
Manufacturing Industry Database. Corresponding to the factor proportions, I cal-
culate the factor intensities of skilled labor, unskilled labor, and capital based on
Romalis (2004). The capital intensity k equals one less the total compensation in
value added. Skilled labor intensity is measured as (1 — k)s, where 3 equals the ratio
of non-production workers to total employment in each industry. Unskilled labor

intensity equals (1 — k)(1 — s).

140

C Data of Price and Quantity in Exports

The trade data of Feenstra (1996) and Feenstra, Romalis and Schott (2002) contain
the export value and quantity of each product variety defined by TS7 or HSIO. The
quantity information can be used as ch directly, and the unit value of each product
can be inferred by dividing the export value by the quantity (i.e., pcj =
The world export quantity of product j is the sum of all countries’ export quantities:
([wt 2 Zcqct, and the world average unit value of product j equals all countries’ total
export. value divided by total quantity: pm 2 chgfécita.

The export value and quantity of each product are required to construct the price
index and quantity index as in equations 17 and 18. Unfortunately, the trade data
does not have perfect quantity information of each variety. From 1975—2001, about
18% of the products exported to the U.S. do not have the quantity information. These
products have to be dropped out when calculating the price and quantity indices. In
addition to the missing data problem, the quantity is recorded with measurement
errors (See General Accounting Office, 1995). To mitigate this problem, I apply
a similar criteria. as Hallak (2006) which drops observations with very low quantity
(below 50 units), which are more likely to be problematic.

As a result, the sample used in this section is different from the one used in
previous sections. Some observations will have different values in exports variables.
Some other observations, especially those with very few positive export products, will
be dropped out from the sample. The sample used in this section contains 22,804

observations, which is fewer than that of the previous one (27,450 observations).

141

Other variables, including innovation, country size and trade cost are from the same

data sources and constructed by the same way as in section 4.1.

142

D Kaplan Meier Estimator

I use table 2.5 to explain the Kaplan-Meier estimator and its application in this paper.
From the left part of table 2.5, the number of spells at risk in the beginning of the
first year is 550,991. This means nl = 550, 991 in equation 3. The number of spells
fail after the first year (i.e. those country-product which export in the first year but
not in the second year) is 205,251, which means (11 = 205,251 in equation 3. The

survivor rate of the first year is 500,959510-‘92g15’251 = 0.6275. This estimate is intuitive:

 

205,251 out of 550,991 spells fail after the first year, indicating that the rest 345, 740
spells survive the first year. So the probability that a trade relationship survive the
first year is W = 0.6275. Without data. censoring, the number of spells at risk
prior to the second year should be 345,740, i.e. those spells which survive the first
year. However, 65, 815 spells are lost after their first year. We know that. these
spells survive at least. one year, but we cannot track their exact fail time since their
data of the second year and thereafter are not available. These spells are not at risk
at the beginning of the second year, and thus 712 = 345,740 — 65,815 = 279, 925.
Among these spells, 55, 893 fail in the second year. Using equation 3, the survivor

rate of the second year is (550959510_929015’251)(W) = 0.5022. The. survivor

 

9

rate of other years can be calculated following the same procedure.

143

E Equilibrium Outcomes of the Price Game
E.1 Symmetric Case (No installed base)

0 Both firms export (XX)

For both firms, plug the demand functions in equation 4 into the profit functions

5a.

a—n—p +1) +72.“
7rA : (pA _ T)(( 2a_2/31(1§7) i))

 

. a—n+p —p +n’y
7TB = (PB — T)(( 2043,03,) ))

 

The first order conditions are:

 

leA_ 1 ’ ' A. —
dl’A ‘Wla-n+r—222A+p3+n,)_o
(lira 1

(1:08 Zmfa—7L+T+PA-2p3+m)=0

Solve the two first order conditions to get the equilibrium prices:

Pix = 10ng

=a—n+7'+n'y
Plug into equation 4 and 5a to get equilibrium market share and profit.

SXX Z SXX : (“"7l—PAJFPB‘l‘m’) = 1
A B 2a—2n(1—'y) 2

 

riff = 7,ng = (a — n + r + 72.7 — T)(%) = %a — $721+ Qny,

Also check the second order condition:

£13” d2” — 1 <0iffa> (1 ) h’-h'1 t'wfid} t'
(“p/1y - leB) — 0'71“)? . n , ,w 1c 1s sa.1s c )yassumpion

a > 2317' + n(1— ’7').
0 Both firms do FDI (FF)

Plug the demand functions in equation 4 into the profit function 5b.

144

_ . (a—n—pA+pB+n'y)
”A _ pA< 2a—2n(1—7) )_ f

 

_ . (a—n+pA—-pB+n/\,r)
”8 — PB( 2a—2n(1—’7) )_ f

The first order conditions are:

fi=m(a-n-2PA+pB+m)_0

(171‘ _ 1 A __

711—)? —— m(a—n+pA—2PB+TI/) —0

Solve the two first order conditions to get the equilibrium prices:
piF = ng = a — n + ny

Plug into equation 4 and 5b to get equilibrium market share and profit.

A — B _ 2a—2n(1—’y) — 2
(WW1) —f= éa— twin-f

Also check the second order condition:

(121% 2 (£271 = ____1____ < 0 iffa > 72(1—7), which is satisfied by assumption
dip/1) d(pB) (L — n+7”. .

a > rlgr +n(1— a).
a One firm exports and the other invests (XF and FX)

Assume firm A exports and firm B invests(XF). Plug the demand functions into

the profit functions

(aun—JDA +pB +7131)
_ T)( 2a—2n(1—’y) )

__ (a — n+pA — p B +717)
”B — PB( 2a—2n(1—7) )— f

The first order conditions are:

D.
:4

_ 1 , . A, _
(PA — 2((L—7t+nef) (a — 71+ T _ 2pA + pB + 11,-) — O
dWB 1 ' (A —

(1,13 = m ("J — ” +PA - 2P3 +12 r) — 0

Solve the two first order conditions to get the equilibrium prices:

145

pA =a—n+:2;7'+n7
ngza—n-l-jlgT-l-R’Y

Plug into equation 4, 5a and 5b to get equilibrium market share and profit.

sXF _ 1 _ T
A _ 2 Zia—iinfiiny
5X17 _ 1 + T
B — Q 6a—6n+6n7
2
XF _ 1 _ _ 1 ,
7rA fl 2(a—n+n7) (a n 3T + n7)
XF_ 1 _ 1 ,2_
7TB — 2(a_n+m) (a n + 37 + 7n) f

Second order condition

(12791 _ d2" — ———1— < 0 which is satisfied b assum tion a > 17+??(1’
(MA)? — d('PB) _ a—n+m. , ~ « y ' p 3 "

E.2 Asymmetric Case (with an Installed Base)

0 Both incumbent A and entrant B enter through export (XX)

 

a—n—p +p +nry+mn—mnry
7rA : (PA _ T)(( A 2(f—n+n7) ) )

 

a—n+p —p +7’ry—7rm+m7w
ma = (m — T><( Aztfim) >>

The first order conditions are

d7r 3 1
dPA = ——2<a_n+n~,..) (a — n + T — 2m + 203 + m + mn — mm) = 0
(177 1

PB :_- W (a —— n + 7' +pA — 2193 + n" — mm + 777.717) 2 0

Solve the first order conditions simultaneously to get equilibrium prices:

pifX = a — n + 7' + my + émn — 111772017

p‘gX = a — n. + 7' + m — :lgm'n. + :lg-nmc,’

Plug prices back into market share and profit. functions

146

. 1_~
Six :%+ nm( ,)

tia—iin+iin’7
SXX _ 1 mull—7')
B — 2 — ticz-iinHin'y
2
XX_ 1 _ ,1 1 ., _1, A,
WA — 2(a_n+m) (a n + n f + 377er 377172,)
2
XX _ 1 __ , ,._1.. 1, , A,
7rB — 2(a_7_l+m) (a n + n, 311172. + 377m ,)

And check the second order condition:

(12% _ (1sz _ _ 1 < 0
(KP/1)“ _ (1(1)B)2 — a—n+n’y

0 Both incumbent A and entrant B enters through F DI (FF)

_ . (a—72,—pA+pB+1w-Hnn—mn'y)
FA _ PA( 2(a—n—Hm) ) _ f

 

_ (a-77'+pA_pB+Tl.’)"—77?72.+Tnn",r)
”B — PB( 2(a—n+n'y) ) _ f

 

The first order conditions are

d97- 1
(31):: = m (a — 71 — 2PA + 113 + my + m-n. — 71171.7) 2 ()
(171' 1
Hing : W (a — 71+ PA - 2133 + n7 — mm + 711717) 2 0

Solve the first order conditions to get
p27: 2 a. — n + 72*," + flimn — Zlgmncy
ng = a — n + 71"," — 113777.71 + :lgmn'y

Plug prices back into market share and profit functions

SFF _ 1 mun—7)
_ 2 + lia—tin+tin'y

__ 1 mn(l—'y)
B _ 2 _ lia—iin+6n*,'

2
FF _ -. ._-1.__ _ , 1 » __ 1., ., a, _
FA — 2(a_n+m) (a n + m + 377m 317212,) f
2
FF _ 1 _ A, _ 1 1, A _
7TB — 2((1—7‘l+n’7‘) (a 71+ 71 , 3771’Il-i- 3mn /) f

The second order condition

d27r d27r 1
d( )2 : d( 2 _a—n+7n < 0

PA PB) _

147

o Incumbent A enters through export and entrant B enters through F DI (XF)

 

a—n—p +p +nqi+mn——mn’y
7rA : (pA — T)(( A2(£_n+,m) ) )

 

a-n+p ——p +n.’y—mn+mnx~/
(( A B )) _ f

”B : PB 2(a—n+n‘,')

The first order conditions are

(177' 1
35?=mm—n‘l‘T—mm'tm;+n7+mn—m7w)=0

d7r B

dp'B 2 mi???) (0 — " +PA - 2m; + m — mn + mm) = 0
Solve the first order condition to get:

pig}? 2 (1. — 71+ I217- ‘l' 71"] + 3%771721 — zlgnzgnry

pfSF = a — 71+ $7 + 717 —:1;mn + Ilgmm

Plug pricesback into market share and profit functions

 

 

SXF _ 1 (T—anrmm)
A _ 2 _ 6a—6n+6n'y
XF _ 1 (T—mn+mn-)

SB _ 2 + 6a—6n+6n'yi

2
”A =(2a—2n+27m;) (a _ Tl — 37' ‘l’ Tl / + 3772,77, 3771.717)

 

2
7rXF_1 1 N _ 1 1 _
”B —(2a_2n+2m) (a — n + 37 + n , 3mm + 3mny) f

The second order condition

_—A2: ——B2: —a—121+n——7, < 0

(MP/1

o Incumbent A enters through FDI and entrant B enters through export (FX)

_ (“-7l-PA+PB +71-7+771.n—m7w)
”A — pA( 2(a—n+7z~,) ) _’ f

 

__ (a—n+pA—pB+n'y—mn+mny)
7TB " (BB _ T)( 2(a—n+n')) )

 

The first order conditions are

d7r 1
fi:m(a—T *2PA+pB+'n"f +mn—7nnel)=()

148

d7r
31—); = W(a—n+r+pA—2pB+n7—mn+mn"y)
Solve the first order condition to get:

piX = a - n + Zlgr + 777 + émn —:1;m'nq/

ng = a — n + %T + 777 —:1;mn + $7727qu

Plug prices back into market share and profit functions

( T+mn—mn"/)

SA — 2 + 6a—6n+67ry

 

SFX 1 (7+mn—m7ry)
B

_ 2 _ 6a—6n+6n'y

FX 1 1 1 1 2
7TA — m (a — Tl + 37' + 71’? + 377171 — 377272121") — f
2
FX _ 1 _ _ 1 ,. _ 1 1 ,
7rB _ (2a-2n+2n'y) (0 n 3T + W) 3mm, + 37mm)

The second order condition

d27r
d(pA)2: (1:312: _a— —7_11+n'y < 0

149

=0

F Proofs of Comparative Static Effects

F.1 Symmetric Case (No installed base)

0 Rival exports:

A firm will choose FDI if and only if

AFX = W(a-n(l—y)+%7)2—f—%(a—n(1—’7)) >0
To find how changes in T, f, 7', n affect one firm’s decision in entry mode when

its rival exports, I check

1A,X 1
g—d:__9a——W(3a—372'+T+3n7)>0

, X
dA’fl' _1 <0

(

 

 

dA’IT‘Y = _ 1 TI. 7'2 < 0
d7 T8 (a.—n+71ry)2
de _ 1 2 -—1
(171 _ —TBT ((L-n+n’y)2 > 0

o Rival invests:

A firm will choose FDI if and only if

2
AWF = %(a, — n(1— 7)) — f — m (a — 72(1— 7) — 2137) > 0

The signs of derivatives are

(lAfl'F _ 1 t _ , , - l _ .
-—-—— —- Q—Q—g—a— 71+ m (3a 377 + 3777 — 7') > O by the assumption a, > 3T+n.(1—1v)

dT ,

F
(lA'fT _1< 0

(

(IATrF 1 72

 

——+- = Tl. > 0
d2" T8 ((i—nJrn'y)2
(1” T8 ((1,—77+n'y)2

150

F.2 Asymmetric Case (Installed Base)

0 Incumbent’s optimal entry mode given entrant enters through expert

The incumbent, firm A, will choose to enter through FDI if and only if

2
A7714: m (a —-77 + 3137' + 777' + $77777 — $777777) —f

2
__1 . , 1, A _ 1
_2(a—77+777) (a — n + TV) + 3771.", 3Wl7l’7) > 0

The comparative statics are:
X

dA7r A.
_4 1 (T+m77—mn ,)
(1T 3 + 9a—977+9777 > 0

(IAWX
WA = —1 < 0
X

(IAWA : _ 1 777' T+2a777 < 0

('7 T8 V ((7—77+77'y)2
lATrX .
(__A _ 1 __ A], T+2a777

(in _ ET“ /) (a—n+777)2 > 0
dAva

_ _ __’l’___
g7lTa_ 71+??? > 0

o Incumbent’s optimal entry mode given entrant enters through FDI

The incumbent. will do FDI if and only if

F 1 1 1 2
”1422217752773 ((7 — 77 + 777 + 377777 — 3777777) -——f

_ l
(2a—277+277~y

 

2
) (a. — 77 — %T + 777 + 31577777 — 313777777) > 0
The comparative statics are:

dAWFA _ 1 T—TI'I.77(l-’)") b h ‘ . 1 1
(r — 3 — W > 0 y t e assumption (7 —— 77(1— 17) > 37 + 377777(1— 7)

(IAWF
WA = —1 < 0
17'3er
f__A .: 1 n7. T—2a7n > 0 lfm < 7" . and

‘17 E (a— 77+ 773:) 23’
dAnF _

1 T—2a777 - 7-
: "T < 0 If m >

‘ '7 IS (a — 77+ 777) 25
(1A7TF
47177 = .1187- (’7 _ 1) T—2a777 < 0 if m < 2%; and

(a — 77 +777)

151

T21187(7—1)—7——2fl?—2>0if777>27—

(a 77+77,)
dATrF
A _ 1 1 1
dm _ QNTa— 77+777>O

o Entrant’s optimal entry mode given incumbent enters through export

The entrant will do FDI if and only if

 

A.X_ 1 __ +1 + ._1 +1 '~2—f
7rB — (2a.—277+2777) a n 37- ”7 377272, 3m‘n' 7

1 1 1 2

The comparative statics are:

dATrX _
(173 = 313+ (gafzgifflgff > 0 by the assumption (7 — 77(1— 7) > 2137' +:1;7m7(1—

 

 

 

 

dATrX
W3— = —1 < 0
dATtX
B _ 1 , T—2a777 - , T .
—— — — IlT——2 < 0 If m < ,and
d7 X E (a—-77+77~) 25
(1A7?
B _ 1 T—2(.7777 - T
— — 77T > 0 If m >
d')" Y T8 ((7—17-1—77'7) 25
dATT‘
1 T—Qnm
= T1—*"——fi2>0ifm<T ;and
d7lY R ( I) (a_ "+1! f) 2—
(ZAW’
B _ 1 _,., T—2a777 ° T
T_-1-8’T(1 ,)m<01f7n>2’5
’l
dAnX
B _ 1. ~_[ —1
(m — 977 Ta— 77+77'y < O

o Entrant's optimal entry mode given incumbent enters through FDI

The follower will do F DI if

AWF_——1— (7—77+77 -1" +1 ’ ‘2—f
B—2((l—72+77.",‘) ‘ 7 '7 3771.72. 13771.71?

 

2
1 1 1, l
— (277—277+277‘/) ((7 — n _ 3T + ”’7 — 317777 + 3772717) > 0
The comparative statics are:

dATTF _..
7&2— —:1;— (13:35:17,297?) > 0 by the assumption (7 — 77(1 — 7) > :1;T + 311777770 —

 

152

_B = 1 ”T T+2am > 0
(1’7 T8 (a—n+n'y)
(IATrF
1 T+2am
= . T ' — 1 < 0
77 T8 (7 ) (a—77+777)
dATTF
12 __ 1 "1—1
dm — 97lTa—77+777 < 0

153

References

[1] Arellano, Manuel and Bond, Stephen. "Some Tests of Specification for Panel
Data: Monte Carlo Evidence and an Application to Employment Equations."
The Review of Economic Studies, 1991, 58(2), pp. 277—97.

[2] Arellano, Manuel and Bover, Olympia. "Another Look at the Instrumental Vari-
able Estimation of Error-Components Models." Journal of Econometrics, 1995.
68, pp. 29—51.

[3] Baldwin, Richard and Harrigan, James. " Zeros, Quality and Space: Trade The-
ory and Trade Evidence." NBER Working Paper Series, 2007, no. 13214.

[4] Baldwin, Richard and Krugman, Paul. "Persistent Trade Effects of Large Ex-
change Rate Shocks." The Quarterly Journal of Economics, 1989, 104(4), pp.
635-54.

[5] Barro, Robert J and Lee, Jong-Wha. "International Data on Educational At-
tainment: Updates and Implications." Oxford Economic Papers, 2001, 53(3),
pp. 541-63.

[6] Bartelsman, Eric J .; Becker, Randy A. and Gray, Wayne B. "Nber—Ces Manu-
facturing Industry Database," 2000.

[7] Bernard, Andrew B. and Jensen, J. Bradford. "Exceptional Exporter Perfor-
mance: Cause, Effect, or Both?" Journal of International Economics, 1999, 47,
pp. 1—25.

[8] Bernard, Andrew B. and Jensen, J. Bradford. "Why Some Firms Export." The
Review of Economics and Statistics, 2004, 86(2), pp. 561-69.

[0] Besedes, Tibor. "A Search Cost Perspective on Formation and Duration of
Trade." Review of International Economics, 2008, 16(5), pp. 835—49.

[10] Besedes, Tibor and Prusa, Thomas. "Product Differentiation and Duration of
Us Import Trade." Journal of International Economics, 2006a, 70(2), pp. 339-58.

[11] Besedes, Tibor and Prusa, Thomas. "Ins, Outs, and the Duration of Trade."
Canadian Journal of Economics, 2006b, 39(1), pp. 266-95.

[12] Besedes, Tibor and Prusa, Thomas J. "The Role of Extensive and Intensive
Margins and Export Growth." NBER Working Paper Series, 2007, no. 13628.

154

[13] Blundell, Richard and Bond, Stephen. "Initial Conditions and Moment Rest ric-
tions in Dynamic Panel Data Models." Journal of Econometrics, 1998, 87. pp.
115—43.

[14] Brainard, S. Lael. "An Empirical Assessment of the Proximity-Concentration
Trade-Off between Multinational Sales and Tirade." The American Economic
Review, 1997, 87(4), pp. 520-44.

[15] Brander, James A. and Spencer, Barbara J. "Export Subsidies and International
Market Share Rivalry." Journal of International Economics, 1985, 18(1-2), pp.
83-100.

[16] Clerides, Sofronis K.; Lach, Saul and Tybout, James R. "Is Learning by Ex-
porting Important? Micro—Dynamic Evidence from Colombia, Mexico, and Mo-
rocco." The Quarterly Journal of Economics, 1998, 113(3), pp. 903-47.

[17] Coe, David T. and Helpman, Elhanan. "International R&D Spillovers." Euro-
pean Economic Review, 1995, 39(5), pp. 859—87.

[18] Das, Sangharnitra; Roberts, Mark J. and Tybout, James R. " Market Entry Costs,
Producer Heterogeneity, and Export Dynamics." Econometrica, 2007, 75(3), pp.
837-73.

[19] Dollar, David. "Technological Innovation, Capital Mobility, and the Product.
Cycle in North—South Trade." The American Economic Review, 1986, 76(1), pp.
177—90.

[20] Eaton, Jonathan and Kortum, Samuel. "Engines of Growth: Domestic and For-
eign Sources of Innovation. " Japan and the World Economy, 1997a, 9, pp. 235-59.

[21] Eaton, Jonathan and Kortum, Samuel. "Technology and Bilateral Trade." NBER.
Working Paper Series, 1997b, 6253.

[22] Eaton, Jonathan and Kortum, Samuel. "International Technology Diffusion:
Theory and Measurement." International Economic Review, 1999, 40(3), pp.
537-70.

[23] Eaton, Jonathan and Kortum, Samuel. "Technology, Trade, and Growth: A
Unified Framework." European Economic Review, 2001, 45, pp. 742-55.

[24] Eaton, Jonathan and Kortum, Samuel. "Technology, Geography, and Trade."
Econometrica, 2002, 70(5), pp. 1741-79.

[25] Fagerberg, Jan. "International Competitiveness: Errata." Economic Journal,
1988, 98(393), pp. 1203.

[26] Fagerberg, Jan. "Technology and Competitiveness." Oxford Review of Economic
Policy, 1996, 12(3), pp. 39-51.

[27] Farrell, Joseph and Saloner, Garth. "Standardization, Compatibility, and Inno—
vation." The RAND Journal of Economics, 1985, 16(1), pp. 70—83.

[28] Farrell, Joseph and Saloner, Garth. "Installed Base and Compatibility: Inno-
vation, Product Preannouncements, and Predation." The American Economic
Review, 1986, 76(5), pp. 940-55.

[29] Feenstra, Robert C. "U.S. Imports, 1972—1994: Data and Concordances." NBER
Working Papers, 1996, no. 5515.

[30] Feenstra, Robert C.; Romalis, John and Schott, Peter K. "U.S. Imports, Exports,
and Tariff Data, 1989-2001." NBER Working Papers, 2002, no. 9387.

[31] Feenstra, Robert C. and Rose, Andrew K. "Putting Things in Order: Trade
Dynamics and Product Cycles." The Review of Economics and Statistics, 2000,
82(3), pp. 369—82.

[32] Fert, Imre and 8003, Karoly Attila. "Duration of Trade of Former Communist
Countries in the Eu Market." Post-Communist Economies, 2009, 21(1). pp. 31-
39.

[33] Flam, Harry and Helpman, Elhanan. "Vertical Product Differentiation and
North-South Trade." American Economic Review, 1987, 77(5), pp. 810—22.

[34] Funke, M., & Ruhwedel, R. . "Export Variety and Export Performance: Em-
pirical Evidence from East Asia." Journal of Asian Economics, 2001, 12(4), pp.
493-505.

[35] General Accounting Office, "US Imports: Unit Values Vary Widely for Identically
Classified Commodities." Report GAO/GGD-95— —90,1995.

[36] Griffith, Rachel; Huergo, Elena; Mairesse, Jacques and Peters, Bettina. "Innova-
tion and Productivity across Four European Countries." NBER Working Paper
Series, 2006, no. 12722.

156

[37] Griliches, Z. (1979). "Issues in Assessing the Contribution of Research and De-
velopment to Productivity Growth." The Bell Journal of Economics 10(1), pp.
92-116.

[38] Griliches, Z. (1990). "Patent Statistics as Economic Indicators: A Survey." Jour-
nal of Economic Literature 28(4): 1661-1701.

[39] Griliches, Zvi; Hall, Bronwyn H. and Pakes, Ariel. "R&D, Patents, and Market.
Value Revisited: Is There a Second (Technological Opportunity) Factor?" Jour-
nal of Economics of Innovation and New Technology, 1991, 1(1), pp. 183-201.

[40] Grossman, Gene M. and Helpman, Elhanan. "Product Development and Inter-
national 'Irade." The Journal of Political Economy, 1989, 97(6), pp. 1261-83.

[41] Grossman, Gene M. and Helpman, Elhanan. "Quality Ladders in the Theory of
Growth." The Review of Economic Studies, 1991a, 58(1), pp. 43-61.

[42] Grossman, Gene M. and Helpman, Elhanan. "Quality Ladders and Product Cy-
cles." The Quarterly Journal of Economics, 1991b, 106(2), pp. 557-86.

[43] Gruber, William; Mehta, Dileep; Vernon, Raymond. "The R & D Factor in
International 'Il'ade and International Investment of United States Industries."
The Journal of Political Economy, 1967, 75(1), pp. 20-37.

[44] Hall, Bronwyn H.; Jaffe, Adam B. and Trajtenberg, Manuel. "Market Value and
Patent Citations: A First Look." NBER Working Paper Series, 2000, no. 7741.

[45] Hall, Bronwyn H.; Jaffe, Adam B. and Trajtenberg, Manuel. "The Nber Patent
Citation Data File: Lessons, Insights and Methodological Tools." N BER Working
Papers, 2001, no. 8498.

[46] Hallak, Juan C. "Product Quality and the Direction of Trade." Journal of Inter-
national Economics, 2006, 68(1), pp. 238-65.

[47] Hallak, Juan C. and Schott, Peter K. "Estimating Cross-Country Differences in
Product Quality." NBER Working Paper Series, 2008, no. 13807.

[48] Helpman, Elhanan; Melitz, Marc and Rubinstein, Yona. "Estimating Trade
Flows: Trading Partners and Trading Volumes." NBER Working Paper Series,
2007, no. 12927.

157

[49] Heston, Alan; Summers, Robert; Aten, Bettina. "Penn World Table Version 6.2,
Center for International Comparisons of Production, Income and Prices at the
University of Pennsylvania, 2006.

[50] Horstmann, Ignatius J. and Markusen, James R. " Strategic Investments and the
Development of Multinationals." International Economic Review, 1987, 28(1),
pp. 109-21.

[51] Horstmann, Ignatius J. and Markusen, James R. "Endogenous Market Struc-
tures in International 'Ifade (Natura Facit Saltum)." Journal of International
Economics, 1992, 32(1-2), pp. 109-29.

[52] Hummels, David and Klenow, Peter J. "The Variety and Quality of a Nation’s
Exports." American Economic Review, 2005, 95(3), pp. 704-23.

[53] Hummels, D., Skiba, A. "Shipping the Good Apples Out? An Empirical Con-
firmation of the Alchian-Allen Conjecture." Journal of Political Economy, 2004,
112, pp. 1384-402.

[54] J affe, Adam B.; Fogarty, Michael S. and Banks, Bruce A. "Evidence from Patents
and Patent Citations on the Impact of Nasa and Other Federal Labs on Commer-
cial Innovation. " The Journal of Industrial Economics, 1998, 46(2), pp. 183-205.

[55] Jaffe, Adam B.; Trajtenberg, Manuel and Fogarty, Michael S. . "The Mean-
ing of Patent Citations: Report on the Nber/Case—Western Reserve Survey of
Patentees." N BER Working Paper Series, 2000, No. 7631.

[56] Jensen, Richard and Thursby, Marie. "A Strategic Approach to the Product Life
Cycel." Journal of International Economics, 1986, 21, pp. 269-84.

[57] Katz, Michael L. and Shapiro, Carl. "Network Externalities, Competition, and
Compatibility. " The American Economic Review, 1985, 75(3), pp. 424-40.

[58] Katz, Michael L. and Shapiro, Carl. "Systems Competition and Network Effects."
The Journal of Economic Perspectives, 1994, 8(2), pp. 93-115.

[59] Keesing, D. B. "The Impact of Research and Development on United States
Trade." Journal of Political Economy, 1967, 75(1), pp. 3848.

[60] Keller, W. "Are International R&D Spillovers 'IYade-Related‘h Analyzing
Spillovers among Randomly Matched Trade Partners." European Economic Re-
view, 1998, 42(8), pp. 1469-81.

158

[61] Klimenko, Mikhail. and Saggi, Kamal "Technical Compatibility and the Mode
of Foreign Entry under Network Externality." Canadian Journal of Economics
40, 176-206, 2007.

[62] Krugman, Paul. "A Model of Innovation, Technology Transfer, and the World
Distribution of Income. " Journal of Political Economy, 1979, 87(2), pp. 253-66.

[63] Mahmood, Ishtiaq P. and Singh, Jasjit. "Technological Dynamism in Asia. " Re-
search Policy, 2003, 32(6), pp. 1031-54.

[64] Markusen, James R. Multinational Firms and the Theory of International Trade.
Cambridge, Mass: MIT Press, 2002.

[65] Markusen, James R. and Venables, Anthony J. "Multinational Firms and the
New Trade Theory." Journal of International Economics, 1998, 46(2), pp. 183-
203.

[66] Markusen, James R. and Venables, Anthony J. "The Theory of Endowment,
Intra-Industry and Multi-National Trade." Journal of International Economics,
2000, 52(2), pp. 209-34.

[67] N avaretti, Giorgio Barba and Venables, Anthony J. Multinational Firms in the
World Economy. Princeton, N .J .: Princeton University Press, 2004.

[68] Nitsch, Volker. "Die Another Day: Duration in German Import Trade." Review
of World Economics, forthcoming.

[69] Ranch, James E. "Networks Versus Markets in International Trade." Journal of
International Economics, 1999, 48(1), pp. 7-35.

[70] Rauch, James. "Business and Social Networks in International Trade." Journal
of Economic Literature, 2001, 39, pp. 1177-203.

[71] Roberts, Mark J. and Tybout, James R. "The Decision to Export in Colom-
bia: An Empirical Model of Entry with Sunk Costs." The American Economic
Review, 1997, 87(4), pp. 545-64.

[72] Romalis, John. "Factor Proportions and the Structure of Commodity Trade."
American Economic Review, 2004, 94(1), pp. 67-97.

[73] Schott, Peter. "Across-Product Versus Within-Product Specialization in Inter-
national Trade." The Quarterly Journal of Economics, 2004, 199(2), pp. 646-677.

159

[74] Soete, Luc. "The Impact of Technological Innovation on International Trade
Patterns: The Evidence Recorded." Research Policy, 1987, 16, pp. 101-30.

[75] Tirole, Jean. The Theory of Industrial Organization. Cambridge, l\IIass.: MIT
Press, 1988.

[76] 'ITajtenberg, M. (1990). "A Penny for Your Quotes: Patent Citations and the
Value of Innovations." The RAND Journal of Economics 21(1): 172-187.

[77] Tybout, James R. "Plant— and Firm—Level Evidence On "New" Trade Theories."
NBER Working Papers, 2001, no. 8418.

[78] Wakelin, K. (1998). "The Role of Innovation in Bilateral OECD Trade Perfor-
mance." Applied Economics 30(10): 1334-46.

[79] Wakelin, K. (2001). "Productivity Growth and R&D Expenditure in UK lVlann-
facturing Firms." Research Policy 30, pp. 1079-1090.

[80] VVooldridge, Jeffrey M. Econometric Analysis of Cross Section and Panel Data.
Cambridge, Massachusetts: The MIT Press, 2002.

[81] World Bank Group, World Development Indicators Online, 2008.

[82] Xu, Bin and Wang, Jianmao. "Capital Goods Trade and R&D Spillovers in the
()ecd." Canadian Journal of Economics, 1999, 32(5), pp. 1258-74.

160

N”

Am

““0
“J
Y”6
Tiilg
Si]

Rm2
EH8
V”o
m“i3
UHO

77777777777177]

3