The Extensive and Intensive Margins of Exports:

The Role of Innovation

Wei-Chih Chen
y
September 2008
Abstract
One important prediction of trade theories is that countries which innovate more will export more.
However, dierent theories have very dierent implications for how innovation would increase exports.
The main objective of this paper is to examine this issue empirically. Using data on patents granted
by the United States and data on manufacturing exports from 105 countries to the U.S. market
over the period 1975-2001, I investigate the extent to which innovation increases the number of
varieties (extensive margins), raises export quantities (intensive margins) as well as improves product
qualities. I have the following ndings: (1) Innovation has positive and signicant eects on both
extensive and intensive margins. The intensive margin contributes about 70% of the eects, and the
extensive margin accounts for 30%. (2) Further decomposition of the intensive margins reveals that
more innovative countries export more quantities at slightly higher prices, suggesting that innovation
increases product qualities in exports. (3) The eect of innovation on export is stronger in low- income
countries than in high-income countries. (4) Innovation has stronger impact in technology intensive
industries. The intensive margin accounts for a larger part in production-intensive sectors like plastic
products and machinery. The extensive margin plays a more important role in science-based sectors,
such as chemical and electronic equipment. (5) The impact of innovation on export declines over
time. Finally, my ndings are robust to specications that allow for the endogeneity of innovation
and xed exporting costs.

I am very grateful to Susan Chun Zhu for her motivation and guidence. I also thank Jeerey Wooldridge for helpful
discussion and suggestions.
y
Department of Economics, Michigan State University. Address: Marshall-Adams Hall, East Lansing, MI, 48824, USA.
email: chenwei5@msu.edu.
1
1 Introduction
International trade theories show that innovation or technological progress improves countries export
performance. These theories dier in their predictions about how innovation increases export. First,
product innovation can introduce new products to the market, or expand the range of existing commodi-
ties that countries export (Vernon [1966]; Krugman [1979]; Jenson and Thursby [1986]; Dollar [1986]).
These studies emphasize the role of innovation on the number of varieties, i.e. the extensive margin in
exports. Second, process innovation can improve productivity and reduce production cost, which will
increase the quantity of export (Posner [1961]; Diewart[1987]). Third, innovation can raise the quality
of products (Flam and Helpman [1987]; Grossman and Helpman [1991]). The second and third types
of work indicate that innovation strengthens countries competitiveness by increasing productivity or
quality, which leads to higher export value in each product, i.e. the intensive margin in exports.
Despite the various predictions of theoretical work, there is surprisingly little corresponding empirical
research showing the relative importance of each margin regarding the inuence of innovation. This
paper integrates two strands of empirical literature to investigate this issue. The rst looks into the
composition of trade. Traditional international trade studies usually focus on the total value of trade
between countries. But an increasing amount of research looks more deeply into the composition of trade.
The extensive margin basically represents the width of trade, such as the number of trading partners
(Evenett and Venables [2002]) or the number of products being traded with each partner (Hummels and
Klenow [2005]; Funke and Ruhwedel [2001]). The intensive margin shows the depth of trade, such as the
trade value with each partner, the trade value of each product with each partner (Hummels and Klenow
[2005]), or the duration of each trading relationship (Besede and Prusa [2007]). The intensive margins
can further be broken into price and quantity, and unit price is used as an indicator of product quality
(Schott [2004]; Hallak [2006]). Many variables (e.g. country size, growth rate, income level) have been
linked with these dierent margins in exports, but the impact of innovation on each margin has not been
explored yet.
The second strand of literature studies the correlation between innovation and exports empirically.
Early studies compared a single countrys export performance in dierent sectors with innovation and
research activities (Gruber et. al. [1966]; Keesing [1967]). Later studies looked at cross country linkage
between trade performance and innovation (Soete [1987]; Fagerberg [1988, 1996]; Wakelin [1998]). Be-
cause of the limitation of data availability, these studies usually use data from developed countries only.
Recent studies focus on heterogenous rms innovation data to explain their export activities (Bernard
2
and Jensen [1999, 2004];
1
Sterlacchini [1999, 2001]; Bleaney and Wakelin [2002]; Lachenmaier and Woss-
mann [2006]). Although these studies nd positive correlation between innovation and overall export
performance, they do not explain which margin plays the primary role.
The main contribution of this paper is to separate the impact of innovations on export performance
into extensive margin and intensive margin, and further decompose the eect on intensive margin into
quantity and price. The estimated results are compared with predictions of dierent trade theories, to
see the extent to which the relation between innovation and export in reality is explained by each model.
Detailed export data from 105 countries in 12 manufacturing industries over the period 1975-2001 is used
to measure the extensive and intensive margins in exports. I use data on patents granted by the United
States as a proxy of innovation, to exploit the extent it aects dierent margins. Figure 1 is the partial-
regression plots of extensive and intensive margins against innovation.
2
It is clear that countries with
more patents have both high extensive margins and intensive margins in exports: they tend to export
more varieties of commodities, and on average export higher value of each commodity. More details will
be discussed in section 4.
The decomposition of innovations eect has important policy and welfare implications. On one
hand, an increase in dierent margins will lead to dierent terms of trade eects. If innovation increases
the number of varieties in exports, the terms of trade moves in the innovating countrys favor and is
an additional welfare gain (as predicted in Krugman [1979]). If technological progress makes countries
intensively export more of each variety, the terms of trade eects can be ambiguous or negative (Flam
and Helpman [1987]). As discussed in Hummels and Klenow (2005), large-scale Computable General
Equilibrium (CGE) models with dierentiated products showed the welfare eects caused by changes
in trade policies are dominated by the terms-of-trade impacts. On the other hand, in trade theories
government uses research and development (R&D) policies to aect rms exports performance and to
improve welfare (Spencer and Brander [1983]). Each study analyzes the welfare eect based on a specic
form of technological progress (e.g. Spencer and Brander [1983] assumed cost-reducing or process R&D).
The welfare changes in reality may not follow the theoretical expectations if innovation does not aect
export in the way assumed in the theories. It is important to know the relative contribution of the
extensive and intensive margins for determining R&D policies and the welfare inuence of exports caused
by innovation.
The method of decomposition used in this paper is closely related to Hummels and Klenow (2005).
1
They found that new products, which are measured by changes in primary SIC codes, lead to exporting.
2
Other control variables include country size and trade resistances, which will be specied in section 4.1 equation 9.
Industry dummies are also included.
3
They analyzed the relation between country size (measured by income and employment) and the extensive,
intensive as well as quality margins of exports. They found the extensive margin accounts for 60% of
the greater exports of larger economies, and the intensive margin is dominated by higher quantity rather
than higher price. My paper is dierent from their work by three points. First, I include innovation
as an important factor determining countries trade performance in dierent margins. The estimated
result shows that innovation indeed has positive and signicant impact on exports, even after controlling
country size and income level. Second, I use panel data which contains observations from 1975 to 2001,
while Hummels and Klenow (2005) used cross-sectional sample of the year 1995. An advantage for
using panel data has been the ability to control for possible correlated, time-invariant heterogeneity in
industries and countries which are unobservable. I could also estimate the model year-by-year to compare
the change of innovations eects over time. Furthermore, I could estimate a dynamic model to take into
account the persistence of exporting, which is related with xed exporting cost. These estimations can
not be implemented with a cross-sectional sample of one year. Third, I construct the indices of export
performance at industrial level, while Hummels and Klenow (2005) constructed these indices at country
level. Using industrial trade data enables me to compare the eect of innovation on export in dierent
sectors. The result shows the impact of innovation on overall export performance diers signicantly
across sectors; and the contribution of the two margins also varies with industrial characteristics.
The impact of innovation on the dierent margins is investigated using two models of estimation: a
static model estimated by a dummy variable regression controlling two-way or three-way xed eects; and
a dynamic model using Generalized Method of Moment (GMM) estimator which regards innovation as
potentially endogenous. The estimates show that in both models, innovation has positive and signicant
eect on the extensive margin as well as the intensive margin. In the benchmark estimation, a country
with 10% more patents will export 5.4% more. The eect can be decomposed into 1.6% higher extensive
margin and 3.8% higher intensive margin. The intensive margin accounts for 70% of the greater exports
from more innovative countries, and the extensive margin accounts for the rest 30%. These results imply
that trade theories, which predicts that innovation improves exports through only one margin, could not
explain its complete impact, and a model which reconciles innovations ability in expanding the number
of varieties as well as in increasing productivity (or quality) will be required. I further decompose the
intensive margins into price and quantity indices, and nd that more innovative countries export higher
quantities at slightly higher prices. This supports theories which predict innovation increases products
qualities. The higher intensive margins are not purely caused by an increase in export quantity which
could possibly lead to a terms of trade loss.
4
Three additional subjects related to innovation and trade are also analyzed in this paper. First, North-
South trade models (e.g. Krugman [1979] and Flam and Helpman [1987]) imply technological progress
could impose asymmetric impacts on developed and developing countries. I include the interaction
between patent and GDP per capita in the estimation to allow innovations eect varying with countries
income levels. The estimates illustrate the impact of innovation on exports is decreasing with countries
income levels, and the contribution of extensive margin is larger in low-income countries than in high-
income countries. Second, the magnitude of innovations inuence could be dierent across industries.
Moreover, the composition of extensive and intensive margins increased with innovation also varies by
sectors. I classify industries into three groups (supplier-dominated, production-intensive, and science-
based) based on the producer-user relation of innovations and patterns of technological changes in each
sector (see Pavitt [1984] and Dosi et al. [1990]). I then implement an industry-by-industry estimation
to explore this issue. The estimated results show that innovation has a stronger impact in technology-
intensive industries as expected. The intensive margin accounts for a larger proportion in production-
intensive sectors like plastic products and machinery; while extensive margin plays a relatively important
role in science-based sectors such as chemicals and electronic equipment. Third, the elasticities of
exports as well as the composition of increase in exports caused by innovation could change as time
passes. Product life cycle research (Klepper [1996]) pointed out that product innovation dominates when
a product is newly introduced; and process innovation becomes important when the production grows
mature. In terms of the dierent margins in exports, this implies extensive margin will account for a
larger share of innovations impact in earlier stages of product life cycles, and intensive margin will become
more important in later stages. I carry out a year-by-year estimation to discover this subject. The
estimates show the impact of innovation on export has declined as time passes. The relative contribution
of extensive margin declines over time, which is consistent with the product life cycle literature.
The remainder of the paper is organized as follows. Section 2 provides a theoretical background of the
empirical model. Section 3 describes the data set and the decomposition method of export. Section 4
shows the model of estimation as well as the estimated results. Section 5 is sensitivity analysis. Section
6 decomposes the eect on the intensive margins into price and quantity. Section 7 is the conclusion.
2 Theoretical Background
This section reviews two strands of trade theories which motivate the empirical estimation of this paper.
These papers stress the role of innovation and technology progress on exports. Nevertheless, they dier
5
in the predictions about how innovation or technological progress aects export.
To help explain the impact of innovation on the dierent margins of export, consider the following
simple model. Suppose consumers in the Home country purchase products from C countries. Each
country has many rms producing dierentiated varieties. Consumers have CES utility function over
varieties. They face the following maximization problem
:ar
qc
l = (
C
c=1
:
c
0
c

c
)

1
(1)
:.t.
C
c=1
:
c
j
c

c
= 1 (2)
Subscription c is countries. :
c
is the number of varieties produced by country c. 0
c
is the quality of
products from country c.
c
is the export quantity of each variety produced by country c. j
c
is the price
of a variety from country c. o 1 is the elasticity of substitution between any two varieties. And 1
is consumers income level. Equation 1 and 2 imply that rms in the same country are symmetric, i.e.
they produce varieties of equal quality (0
c
) and charge the same price (j
c
). From equation 1 and 2, we
could see quality works as a demand shifter in which higher quality will raise the demand to products.
Consumers will buy every varieties exported to the Home country, and the total value of trade from
country c is
:
c
j
c

c
(3)
This total value of export could be decomposed into the extensive margin and the intensive margin. The
extensive margin measures the width of export. In equation 3, it is captured by :
c
, which is the number
of products exported by country c. The intensive margin measures the depth of export. In equation
3, the intensive margin is measured by j
c

c
, which is the export value of each product from country c.
Because quality is a demand shifter in this model, an increase in product quality will lead to a higher
intensive margin of exports.
In trade literature, innovation is predicted to have important inuence on both the extensive margin
and intensive margins.
(1) The extensive margin
The rst strand of theory emphasizes innovations ability in creating new products or expanding the
range of products that countries could produce and export. For instance, Krugman (1979) developed a
6
North-South trade model in which technological progress (i.e. innovation in the North and technology
transfer in the South) determines exports. This paper showed that in equilibrium, the relative number
of products a country exports is increasing with its rate of technological progress. Other studies of this
type include Dollar (1986), Jensen and Thursby (1986), and Grossman and Helpman (1989, JPE).
(2) The intensive margin
Another strand of literature predicts that successful innovation could improve quality of products,
which induces higher demand to these products. Grossman and Helpman (1991, QJE) built a North-
South model of international trade, in which innovation takes the form of improvements in the quality of
a xed set of products. Other studies, such as Flam and Helpman (1987) and Grossman and Helpman
(1991, RES), also assumed innovation is an important determinant of exports quality.
According to the model of this section, if innovation increases the number of products, we expect to
nd positive correlation between innovation and the extensive margin of exports. On the contrary, if
innovation increases the quality of products, we expect innovation to be positively correlated with the
intensive margin of exports. Despite the various predictions of theoretical works, there is no empirical
study, to the best of my knowledge, tries to estimate and compare the relative importance and contribution
of innovation on the dierent margins of exports. This is the motivation of the estimation in this paper.
3 The Data Set
An empirical study to estimate the relation between industrial innovation and export performance requires
data that falls into two categories. The rst is detailed export data, which could be decomposed
into extensive margin and intensive margin. The second is a variable which could measure countries
innovations or technological changes. I also include variables of gravity model, including country size and
trade cost. This section describes the data source and the construction of both export and innovation
variables.
3.1 Export Data
I obtain data on all positive export ows at detailed commodity level for 1975-2001 from Feenstra (1996)
and Feenstra, Romalis and Schott (2002). The data contains quantity and value of each commodity
exported to the U.S. from other countries. Each commodity is classied according to the 7-digit Tari
Schedule (TS7) of the United States Annotated (TSUSA) in 1975-1988, and it is classied according to
7
the 10-digit Harmonized System (HS10) in 1989-2001.
3
I will treat each commodity in TS7 or HS10
as a "product variety." The data sets also include U.S. Standard Industrial Code (SIC) corresponding
to each product variety, which enables me to identify the extensive and intensive margins of export in
industrial level.
3.2 Method of Export Decomposition
To nd out how innovation aects export, we need to decompose trade value into the two margins. The
decomposition method follows Hummels and Klenow (2005), but I calculate the extensive margin and
intensive margin of each country in industry level, instead of country level as in Hummels and Klenow
(2005).
A countrys overall export share is dened as:
crjort
cit
=

j2Icit

cijt

j2Iit

wijt
(4)
Subscription c is country; i is industry; , represents product variety dened by TS7 or HS10; and t is
year.
cijt
is the export value of variety , in industry i that country c exports to the U.S. in year t.

wijt
=
c

cijt
is the total value of variety , that world (all countries in the sample) export to the U.S.
1
it
is the set of varieties that are classied to industry i in year t; 1
cit
is a subset of 1
it
in which country c
has positive export to the U.S. in year t; i.e.
cijt
0. The rst dependent variable, crjort, is the ratio
of a countrys export value relative to the worlds export value. Since it is measured at industry level, a
country can have high export share in one industry but low export share in another.
The rst component of export is the extensive margin, an index which measures the range of products
a country exports:
crtc::ic
cit
=

j2Icit

wijt

j2Iit

wijt
(5)
The extensive margin is the number of varieties that a country exports to the U.S., divided by the total
number of varieties that the world exports to the U.S. Each variety is weighted by its world export value,

wijt
. If all varieties in an industry are of equal importance, the extensive margin is simply the fraction
of varieties in which country c exports to the U.S.
4
3
These are very disaggregated data sets. In 1988, there are 11,198 commodities classied in TS7 exported to the U.S.,
and in 2001 there are 16,293 commodities based on HS10.
4
Note that in this paper, the extensive margin is dened as the coverage rate of varieties in each industry. This is
dierent from some other literatures, in which extensive margin means the number of destinations that a country trades
with.
8
The corresponding intensive margin is dened as
i:tc::ic
cit
=

j2Icit

cijt

j2Icit

wijt
(6)
i:tc::ic
cit
equals the ratio of country cs export value relative to all countries exports value in those
varieties that country c has positive export.
By the denition of equation 5 and 6, the overall export share in equation 4 equals the extensive
margin times the intensive margin as demonstrated in Hummels and Klenow (2005):
crtc::ic
cit
i:tc::ic
cit
=

j2Icit

wijt

j2Iit

wijt


j2Icit

cijt

j2Icit

wijt
=

j2Icit

cijt

j2Iit

wijt
= crjort
cit
(7)
Note each of these three variables are measured at industry level and are made relative to all countries
in a given year, and by denition
c
crjort
cit
= 1.
5
For each country-industry which has positive export
values to the U.S., I construct overall export, extensive margin and intensive by equation 4, 5 and 6.
These are the dependent variables in the estimation.
3.3 Innovation Data
The key explanatory variable in this paper is the proxy of each countrys industrial innovation. I use
patents granted by the U.S. as the proxy of innovations. The patent data comes from Hall, Jae and
Trajtenberg (2001).
6
The data comprises detailed information on about 3.5 million patents granted
between 1963 and 2002 by United States Patent and Trademark Oce (USPTO).
In economic literature, patents are indicators of "innovation" or "invention" counts. Some stud-
ies found positively correlation between patents and new products, while others associate patents with
competitiveness or comparative technology advantage (see the survey by Griliches [1990]). These works
shows that patent is a valid proxy of 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
enormously in their technological and economic signicance, and thus simple patent counts (SPC) could
not precisely measure innovative outputs. A method applied is to use patent citations as an index of the
importance of patents. It has been shown in empirical studies that patent counts weighted by citations
are more closely associated with their value and importance (Trajtenberg [1990]; Jae et al. [2000] Hall
5
By denition, export; extensive, and intensive are ranged in the interval (0,1]. Note although export
cit
= 1,
cextensive
cit
and cintensive
cit
does not have to equal 1. Consider an extreme case: every country exports all varieties
to the U.S. in an industry. In this case, extensive
cit
= 1 for all c, and thus cextensive
cit
6= 1.
6
Data after 1999 is updated by Hall.
9
et al. 2000, 2001]). In this paper, I weight patents by citations received and use the weighted patent
counts (WPC) to assess the economic and technological value of innovations. I apply the linear weighing
scheme which follows Trajtenberg (1990) to construct the weighted patent count indices:
\1C
cit
=
ncit
j=1
(1 + c
j
) (8)
Where :
cit
is the number of patents issued to country c in industry i during year t, and c
j
is the number
of citations received by patent ,.
7
Except patent, R&D expenditure is also used as a proxy for innovation in literatures. In this paper,
I use patent for the following considerations. First, patents are the output of innovative activities, but
R&D is the input. In trade theories, it is the results of innovation rather than the eorts of innovation
that improve countries export performances. Second, the time that an innovation is created is more
accurately measured by the application year of patent than by the year of R&D expenditures. Innovation
has lagged when measured by R&D expenditure. Third, Patent data from a single source of each industry-
year can reect countries innovative capabilities on a relatively comparable base.
8
R&D expenditures
from dierent countries could have dierent measurements and eciency. The patent data from USPTO
covers a wider range of countries in a longer time span (1963-2002). On the contrary, industry level R&D
data is available only among a limited set of countries (such as OECD), and the time span is short. This
restricts the scope of studies in a small group of countries with similar properties. Another drawback of
R&D expenditure data is that it tends to underestimate the innovation contribution of small rms which
do not have a separate research department.
9
Each patent has two indices of time: the application year and the grant year. I classify patents
according to the application years because countries innovation capability are reected on the year
patents are applied rather than the year they are granted.
10
More details about the patent data and the
construction of weighted patent counts are discussed in the data appendix.
7
A patent which receives one citation is considered to be as twice important as a patent which does not recieve any
citation; A patent which recieves two citations is as treble important as a patent which is never cited, and so on.
8
I use patents granted by the U.S. rather than other countries, because intuitively U.S. 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) found that patents granted by U.S. is indeed
a superior measurement of international innovation than patents granted by other countries like Japan, France, U.K. and
Germany.
9
Wakelin (1998) used R&D expenditure and patent numbers, separately, as proxies of relative innovation performance,
and he found that the patent proxy captures it better than R&D expenditure, especially in high technology industries. The
bottom line is using patent to measure innovation for my purpose should be no worse than using R&D expenditures.
10
Patent data is truncated when classifying with application years. This issue is addressed in the data appendix. I
will compare results using grant year and results using application year in the year-by-year estimation, and the comparison
shows there is no material dierence.
10
3.4 Other Controls
Five other controls are included alongside the patent data. In gravity model, GDP is usually used to
control for country size. This is the key independent variable in Hummels and Klenow (2005).
11
In this
paper, I decompose GDP into population and GDP per capita. The population and GDP per capita
data are from Penn World Table 6.2. I also include distance, dummy variables of common language and
common border in the estimation, which represents trade costs and resistances. These data are from
Jon Havemans International Trade Data.
The complete sample of my empirical analysis contains 105 countries, each has 12 manufactured
industries (2-digit SIC) over the period 1975-2001. A list of these countries is in table 1. The 12
manufacture industries selected are the industries that patents could be classied according to the con-
cordance provided by Hall, Jae and Trajtenberg (2001). Table 2 contains a list of these industries.
The balanced-panel sample size should be 1051226 = 32, 760. However, not every country exports to
the U.S. in every industry-year. Since the focus of this paper is to analyze to eect of each regressor on
dierent margins of export, I only keep observations which have positive export values. The total sample
size is 27,450, which means about 16% observations have zero export to the U.S. and are dropped out of
the sample. The estimated results of this paper should be interpreted as the eect of each explanatory
variable on export, conditional on export occurs.
4 Model of Estimation and Results
4.1 Benchmark Estimation Model
My model of estimation is an extension to Hummels and Klenow (2005) in which I consider innovation
as an additional and important determinant of export performance. The motivation comes from the
theoretical models in section 2. Nevertheless, the object of the empirical work here is not to formally
test each model. These models are polar cases, and it is not surprising if none of them could individually
cover the complete eect of innovation in reality. The intension of this section is to estimate the relative
contribution of innovation on each margin. This could help developing a new theory which describes
innovations eect on exports closer to what happens in the real world. The benchmark model of
estimation is:
11
They also use employment and GDP per worker to measure country size.
11
ln(tradc
cit
) = ,
0
+ ,
1
ln(jatc:t
cit
) + ,
2
ln(joj
ct
) + ,
3
ln((
G11
joj
)
ct
) (9)
+,
4
ln(di:t
c
) + ,
5
|a:q
c
+ ,
6
/ord
c
+ j
i
+ `
t
+ -
cit
ln(tradc
cit
) in equation 9 represents the three dependent variables: the natural logs of overall export,
extensive margin and intensive margin in exports. I will run three separate regressions to estimate the
eect of innovation on each margin. These variables are dened in equations 4-6, and are calculated
on an industry-year base.
12
The dependent variables may change with time and industries,
13
and
thus industry-specic eects (j
i
) and year-specic eects (`
t
) are included in the model.
14
The key
explanatory variable, ln(jatc:t
cit
), is the natural log of weighted patent counts (WPC) described in
equation 8. Since the number of patents granted and citations made depend largely on the examination
process of USPTO, this variable should also be adjusted by industry and year, to make observations from
dierent industries and years comparable. This eect is taken into account in a log linear model once
industry and year dummies are included.
15
ln(joj
ct
) and ln((
GDP
pop
)
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 dummies controls for this eect. In addition to innovation and country size, I
also include distance, common language and common border as regressors which proxy trade costs and
barriers. ln(di:t
c
) is the natural log of the distance between country c and the U.S. |a:q
c
is a binary
variable, which equals one if a country has English as its primary language. /ord
c
is a binary variable,
which equals one if a country has a common border with the U.S.
Each estimated ,
i
is interpreted as the elasticity of the dierent margins in exports with respect
to each regressor. Because industry and year dummy variables are included in all estimations, the
coecients represent the impact of innovation as well as other regressors on export within each industry
and year. An advantage of taking natural logs is that we can decompose the impact of each explanatory
12
For instance, export
cit
= 0:1 means country c
0
s export value is 10% of the total value that all countries in the sample
exports to the U.S. in industry i and in year t:
13
Although these variables are calculated relatively in a industry-year base, they could still change with time and industry.
For example, in earlier years there were fewer countries export to the U.S. than in later years, and thus the export share in
earlier years is likely to be higher than that of later years. Similarly, the number of countries which have positive exports
could dier by industries.
14
A complete control should include an interaction between industry dummies and year dummies. It turns out the
estimated results are almost identical, and thus I include industry dummies and year dummies separately in the estimation
without taking interactions.
15
For example, I divide the weighted patent counts of each observation by the total number of weighted patent counts
granted to all countries in the same industry and year, to measure the "relative" innovative capability of a country among
all countries in the same group (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 dummy variables.
12
variable on export performance into dierent margins additively:
b
,
ex
i
=
b
,
em
i
+
b
,
im
i
, where
b
,
ex
i
,
b
,
em
i
and
b
,
im
i
are the estimated coecients in the three equations using overall export, extensive margin and
intensive margin as the dependent variable, respectively.
16
I can then measure the relative contribution
of the two margins to overall export with respect to each independent variable. Because innovation
should improve exports, we expect the estimated coecients on log patent to be positive. According
to Hummels and Klenow, coecients of country size (i.e. ln(joj) and ln(
GDP
pop
)) should be positive, too.
The coecients of common language and common border are supposed to be positive, and that of the
distance is expected to be negative.
I estimate equation 9 using least squares dummy variable regression (LSDV).
17
I refer to this esti-
mation as the two-way xed eects case since it controls industry-specic and year-specic xed eects.
Standard errors are made robust to heteroskedasticity and clustered by country. The estimated results
are in table 3. The key explanatory variable, innovation, has positive and signicant eect on overall
export as well as both extensive and intensive margins. The coecients reveal that on average, a coun-
try which has 10% more patents will export 5.4% more. This eect can be decomposed into dierent
margins: a 10% increase in innovation will raise the extensive margin by 1.6%
18
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.
19
The next two rows in table 3 verify that large countries export more. A country which has 10%
more population will on average export 5.5% more varieties and 5.0% higher value in these varieties. A
10% increase in GDP per capita will raise extensive margin by 8.9% and the intensive margin by 6.1%.
With regard to these two variables, the extensive margins account for a larger part (52% with respect to
population and 59% with respect to income level) than that of the intensive margins (48% with respect
to population and 41% with respect to income level). The relative contribution of the two margins with
respect to country size in this paper is similar to the nding in Hummels and Klenow (2005). They used
cross-sectional data in country level and found the extensive margin accounts for about two third of the
16
Because export share = extensive margin intensive margin, ln(export share) = ln(extensive margin) + ln
(intensive margin).
17
Since trade is between [0,1], ln(trade
cit
) must be non-positive. Because I use OLS to estimate, the tted values could be
positive. But it turns out only a small proportion of the tted values have this problem. For instance, only 103 observations
(out of 25,268, which is less than 0.05% of the sample) has positive velues in log export. And these observations are from
3 countries: Canada, Germany, and Japan. For extensive margin, about 6% of the sample has positive tted values. For
intensive margin no observation has this problem.
18
The denition of extensive margin I use in this paper actually represents the "relative" coverage share of products of
each industry-year. It can not, nevertheless, reect an increase in absolute number of products. For example, suppose in
1975 there are totally 100 varieties in industry SIC 28. If Germany exports 90 varieties in 1975, the extensive margin is
90%. If in 2000, total varieties in this industry becomes 1000 and Germany exports 900, the extensive margin is still 90%.
19
The percentages are derived from: 0:179 0:651 = 27%; 0:472 0:651 = 73%:
13
more export that larger countries export.
The last part in table 3 shows that countries locating close to the U.S., using English as the primary
language, and sharing a common border with the U.S. has higher overall export to the U.S.. Among
these variables, only distance is statistically signicant to overall export. When we look at elasticities of
each margin, all have correct signs except that of the extensive margin with respect to border.
20
Most of
these coecients are individually insignicant, but they are jointly signicant in all the three regressions.
Next, I consider the situation that there could be unobserved country characteristics which are related
to export performance. I estimate a three-way xed eects model using LSDV estimator controlling
industry, year, and country specic eects:
ln(tradc
cit
) = ,
0
+ ,
1
ln(jatc:t
cit
) + ,
2
ln(joj
ct
) + ,
3
ln((
G11
joj
)
ct
) (10)
+j
i
+ `
t
+ c
c
+ -
cit
where c
t
is the country xed eect, and all other variables are as dened in equation 9. Since distance,
language and border are time-invariant country characteristics, they will be dropped out once country
xed eect is included. Standard errors are again made fully robust. The estimates of this three-way
xed eect case are listed in table 4. The eects of innovation on overall export and the two margins
remain statistically signicant, and the eects are stronger than in the two-way xed eect case. A
country which has 10% more patents in an industry-year exports 2.0% more varieties and 3.9% higher
value of these varieties on average, leading to a 5.9% higher overall export. The relative contribution
of the two margins are similar to the two-way xed eect case, with extensive margin accounts for 34%
and intensive margin accounts for 66%. The coecients of country size are dierent from two-way
xed eect case. Intensive margin now accounts for a slightly larger part than extensive margin with
respect to population. Richer countries still export more, but the eect on the intensive margin is
insignicant. An explanation is that countries "relative" GDP per capita position does not change
much across years. Once country xed eect is included, the variation of GDP per capita becomes small
and leads to insignicant eect. The bottom line is the eect of innovation on export is signicant in
both cases, and the relative importance of the two margins are also similar.
The results in table 3 and 4 reveal that trade theories predicting innovation can improve export
20
In this sample only two countries, Canada and Mexico, have common border with the U.S. and thus the variation of
this variable is small.
14
through only one margin do not explain the full eects of innovations. Models like Krugman (1979)
which assume innovations could only increase the number of varieties but not productivity or quality fail
to explain the high intensive margins from more innovative countries. Models assume innovations could
only increase productivity or quality capture the higher intensive margins well, but they miss the eect
on extensive margins which contribute for more than one forth of the total impacts. 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 cover the complete inuence of innovation on
exports.
According to theoretical works reviewed in the introduction, the contribution from the extensive
margins will raise exporting countries welfare. The welfare change caused by the intensive margins
depends on whether innovation increases export quantities or qualities. If the high intensive margins
are induced by higher export quantities only, the expansion in exports may be at the cost of a terms of
trade loss. If the higher intensive margins are caused by an increase in quality, the welfare gain will be
larger. To identify the source of intensive margins, a further decomposition of the intensive margins will
be implemented in section 6.
4.2 Dynamic Panel Estimator and Endogenous Innovation
In this section, I estimate a dynamic model using a Generalized Method of Moment (GMM) estimator.
The model of estimation is as follows.
ln(tradc
cit
) = c + ln(tradc
ci;t1
) + ,
1
ln(jatc:t
cit
) + ,
2
ln(joj
ct
) + ,
3
ln((
G11
joj
)
ct
) (11)
+,
4
ln(di:t
c
) + ,
5
|a:q
c
+ ,
6
/ord
c
+ j
ci
+ `
t
+ -
cit
where j
ci
is country-industry xed eect.
21
This dynamic model diers from equation 9 and 10 in which
it includes lag export variables, ln(tradc
ci;t1
), as an additional regressor. The inclusion of lag export
variables has economical meaning, that it accounts for the persistence in exports which is related with
the xed exporting cost (Robert and Tybout [1997]). The estimated coecient on patent in a dynamic
model shows the eect of innovation on exports after controlling the exports in the previous year.
Using dummy variable regressions to estimate the dynamic model in equation 11 will generate incon-
sistent estimates, because the lag dependent variable is endogenous. I use the Generalized Method of
21
This model is more exible than equation 9 and 10, because it allows each country-industry to have its own xed eect.
15
Moment (GMM) estimator for dynamic panels proposed by Arellano-Bover (1995)/Blundell-Bond (1998)
to estimate the model. The estimator builds a system of two equations, a level equation and a dierence
equation, and is known as system GMM estimator.
22
It uses lag dependent and independent variables
dated t-2 or earlier as instruments, to account for the endogeneity problem.
23
An advantage of using the
dynamic GMM estimator is that it incorporates the problem that one or more regressors (other than lag
dependent variable) can potentially be endogenous. The explanatory variable of interest in this paper,
innovation, could possibly be endogenous due to reverse causality or simultaneity.
24
I will estimate
the dynamic model using system GMM estimator by two cases. In the rst case, I assume the only
endogenous variable in equation 11 is ln(tradc
ci;t1
). All other regressors, including patent, are strictly
exogenous. The estimated results is presented in column (1)-(3) in table 5. In the second case, I assume
patent is potentially endogenous and is instrumented. I list the estimates of this case in column (4)-(6)
in table 5. Standard errors in both cases are made fully robust.
In both cases, the estimated coecients on lag exports are positive and signicant. This shows
country-industry which export more in year t 1 are more likely to export more in year t. This result
supports the hypothesis of xed exporting costs. The estimated coecient in the overall export
equation is the largest, followed by that in the intensive margin equation, and then that in the extensive
margin equation. The persistency in intensive margins is stronger than in extensive margins.
The estimated elasticities of patent reect that even after controlling lag export variables, the eect of
patents on each margin in exports remains positive and very signicant. In the exogenous-innovation case,
a 10% increase in patent will increase the extensive margins by 1.1% and the intensive margins by 2.4%.
In the endogenous-innovation case, a 10% increase in patent will raise the extensive margins by 1.2% and
the intensive margins by 3.6%.
25
The elasticities of the two margins in exports with respect to patent
are larger when assuming innovation could potentially be endogenous than when assuming innovation
is strictly exogenous. In both cases, the elasticity of the intensive margins is larger than that of the
extensive margins. In summary, the estimation of a dynamic model in equation 11 using the GMM
22
Arellano-Bover (1995)/Blundell-Bond (1998) system GMM estimator augments the Arellano-Bond (1991) dierence
GMM estimator in which it allows more instruments and can improve eciency. The system GMM allows the inclusion of
time-invariant regressors (e.g. distance). These variables will be dropped out when using dierence GMM.
23
A necessary condition for lag variables to be valid instrument is there is no autocorrelation of order 2 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 valid instruments.
24
First, there could be variables which are not controlled in the model, but can aect innovation and export simultaneously
(e.g. change in government policies). Second, export and innovation could aect each other. On one hand innovation can
imporve competitiveness and increase export; but on the other hand trade can help the transfer of knowledge (Coe and
Helpman [1995]) which motivates more innovations.
25
Since the coecients are no longer estimated by OLS, the eect can not be decomposed additively, i.e.
b

ex
i
6=
b

em
i
+
b

im
i
.
16
estimator which accounts for possible endogeneity in innovation does not change the main conclusion
found in the previous section: innovation has positive and signicant impacts on both margins in export,
and the impact on the intensive margins is relatively stronger.
4.3 Interaction between Patent and GDP per capita
The models of estimation in equation 9, 10 and 11 assume the eect of patents on export is the same across
countries. However, technological progress could impose asymmetric impacts on exports in developed
countries and developing countries (e.g. Flam and Helpman [1987]). To explore this issue, I add the
interaction between patent and GDP per capita in the model. The model of estimation becomes
ln(tradc
cit
) = ,
0
+ ,
1
ln(jatc:t
cit
) + ,
2
ln(joj
ct
) + ,
3
ln((
G11
joj
)
ct
) (12)
+,
4
ln(di:t
c
) + ,
5
|a:q
c
+ ,
6
/ord
c
+,
7
ln(jatc:t
cit
) ln((
^
G11
joj
)
ct
) + j
i
+ `
t
+ -
cit
This model is estimated by dummy variable regressions, with the control of two-way xed eects.
26
To
make estimated coecients on the interaction terms more intuitive, GDP per capita is divided by the
mean of all countries in each year (
] GDP
pop
) before interacting with log patents.
27
In equation 12, the elasticity of export with respect to patent is ,
1
+ ,
7
ln((
] GDP
pop
)
ct
), which varies
with countries income level. The estimated coecients and standard errors are reported in table 6.
The estimates of patent (,
1
) are positive in all equations. This means innovation has positive eect
on overall export and both margins in a country which has GDP per capita equals sample mean, i.e.
(
GDP
pop
)
ct
= (
GDP
pop
)
t
. The estimated coecient of the interaction term (,
7
) is negative, showing the
inuence of innovation is decreasing with countries income levels. Both ,
1
and ,
7
are individually and
jointly signicant in each equation, so the eect of patents on export is still signicant after including
the interaction term.
To have a more intuitive interpretation about how the eects of innovation on export diers across
income levels, consider three countries which are identical in all aspects except their GDP per capita.
26
Each regression uses the natural logs of overall export, extensive margin and intensive margin as the dependent variable.
Standard errors are made robust to heteroskedasticity and clustered in country.
27
This means ln(
^GDP
pop
)ct = ln(
GDP
pop
)ct ln(
GDP
pop
)t, where (
GDP
pop
)t is the mean of all countries GDP per capita in year
t. Replacing the main eect with ln(
^GDP
pop
)ct will not change the estimated results because the demeaned term will be
absorbed by the year dummies.
17
I dene these countries as high-income, middle-income, and low-income countries. The high-income
countrys GDP per capita is twice as much as the sample mean i.e.(
GDP
pop
)
ct
= 2(
GDP
pop
)
t
; the middle-
income countrys GDP per capita equals the mean ( (
GDP
pop
)
ct
= (
GDP
pop
)
t
); and the low-income countrys
income is half as much as the mean ((
GDP
pop
)
ct
=
1
2
(
GDP
pop
)
t
).
28
I then calculate the elasticities of overall
exports, extensive margins and intensive margins with respect to innovation of each country, using the
coecients in table 6. The elasticities of these countries are reported in table 7.
29
A 10% increase
in patent will increase low-income countrys overall export by 7.9%. But a 10% increase in innovation
will increase the middle-income countrys overall export by 6.6%, and the high-income countrys overall
export by only 5.4%. The eects of an increase of innovation on extensive margins and intensive margins
are also strongest in the low-income country, followed by the middle-income country, and last the high-
income country. These results are depicted in gures 2.1-2.3, with log exports on the vertical axis and log
patens on the horizontal axis. Slopes of the lines are the elasticities of the dierent margins in exports
with respect to patent. All lines are upward sloping, showing countries which have more patents export
more. In each gure, the low-income countrys line is the steepest, followed by middle-income countrys,
and last the high-income countrys. This indicates the positive impact of patent on export is strongest
in the low-income country, and weakest in the high country.
I then compare the composition of the higher export led by innovations of countries with dierent
income level. For the low income country, the extensive margins account for 40% and the intensive
margins account for 60%. For the middle-income country, the extensive margins account for 36% and
the intensive margins account for 64%. For the high-income country, the extensive margins contributes
30% and the intensive margins contributes 70%. The relative importance of extensive margin decreases
with GDP per capita, while the contribution of intensive margin increases with it.
I included the interaction between patent and income in the dynamic model in equation 11. The
model is estimated using the same Arellano-Bover (1995)/Blundell-Bond (1998) GMM estimator with two
cases, one assumes innovation is strictly exogenous and the other assumes it is potentially endogenous.
The estimates are reported in table 8. The coecient on patent is positive and signicant in all margins
and in both cases. The estimated coecients on the interaction are negative. Although the interaction
28
For example: in the year 1999, the average GDP per capita of all countries in the sample is $10,609. Countries which
have income levels close to twice, once and half as much as the sample mean are Turkey ($5,460), Hungary ($10,821), and
Finland ($21,585), respectively. These countries are at the 45, 60, and 80 percentile.
29
The estimated elasticity of export with respect to innovation equals
b

1
+
b

7
ln(
^GDP
pop
): For example, the high-income
countrys elasticity of overall export is 0:663 0:178 ln 2 = 0:540: We could calculate the elasticity of countries with any
income levels. For example, in 1999 Norways GDP per capita is $32,355, and the mean GDP per capita is $10,609. This
means
^GDP
pop
= 3:05; or ln(
^GDP
pop
) = 1:12: The estimated elasticity of export with respect to innovation for Norway will be
0:663 0:178 1:12 = 0:464.
18
is not individually signicant when patent is assumed to be exogenous, it is always signicant joint with
the main eect. These results basically conrm the ndings of the static model using two-way xed
eect dummy variable regression.
The following points conclude this section First, patents have stronger eects improving low-income
countries export performance than high-income countries. This pattern is signicant in the overall
exports as well as the extensive margins, but relatively insignicant in the intensive margins. Second,
the extensive margins in low-income countries contributes higher proportion than that in high-income
countries to an increase of export caused by innovations.
4.4 Estimation by Industry
The estimations so far implicitly assume that the eect of innovation on exports and the composition
of this eect is the same across industries. In reality, the eect of innovation on export could dier by
industries, and the contribution of the two margins might vary with industrial characteristics. In this
section I separate the sample by industry and estimate a by-industry version of the model in equation 9,
except the industry dummies are excluded from the model. The estimations will generate twelve sets of
coecients, each represents the impact of regressors on export in an industry.
It is not surprising that innovation imposes more potent eects on technology-intensive industries than
on traditional industries. The more interesting issue is the composition of increased exports caused by
innovation. Pavitt (1984) and Dosi et al. (1990) proposed a taxonomy which classied sectors into three
categories according to the producer-user relation of innovations. The rst group is supplier-dominated
industries. Most innovations from these sectors are process innovation, so we expect intensive margin
will account most of the eect of innovation. Firms of these industries heavily depend on their suppliers
of materials and equipment, and innovation of their material or equipment suppliers industries could be
more important than their own inventions. In other words, innovation from these industries will have
relatively weaker power in explaining their own export performances. Among the twelve industries in
table 2, food and textile in my sample belong to this category (Amable and Verspagen [1995]). The
hypothesis is that the elasticity of export in this category is relatively smaller and insignicant; the eect
of innovation, if any, goes to the intensive margin rather than the extensive margin.
The second group of sectors is production-intensive. Scale economies and the availability of machinery
are important in these industries, and process innovation is more important than product innovation. In
terms of this paper, we expect the eect of innovation on intensive margin to be stronger than extensive
margin. Based on the classication of Amable and Verspagen (1995), there are eight industries in
19
my sample go to this group: petroleum, rubber-stone-clay, primary metal, fabricated metal, machinery,
transportation, and instrument.
The third group of sectors is science-based. New products are the most important mode of competition
in these sectors. Among the twelve industries in my sample, chemical and electronics are classied in
this category. Because product innovation turns to be more important than process innovation to sectors
of this group, we expect extensive margin accounts for a relative large part than the intensive margin,
compared with sectors in the production-intensive group.
The estimated coecients and signicance level of the industry-by-industry estimation are summarized
in table 9. Since there are twelve sets of estimates, I only show the coecients of patents and skip that
of other regressors. I organize industries according to the producer-user relation described above. The
rst panel is supplier-oriented industries, the second panel is production-intensive industries, and the
third panel is science-based industries. Column (1) shows the overall export elasticity with respect
to innovation in each industry. Transportation has highest elasticity (0.76), followed by industrial
machinery and equipment (0.74), instruments (0.63), and electronics (0.62). The three industries which
have insignicant and also small elasticities are food (0.079), textile (0.11), and petroleum (0.24). The
estimates show an intuitive result: the eect of innovations is stronger and more signicant in industries
that are more technology-intensive.
The result in the rst panel also conrms our hypotheses of the supplier-oriented industries. First,
the two industries belong to this category (food and textile) have smallest and insignicant elasticities
of overall export with respect to patent. Second, intensive margin accounts for all of the eect ( the
coecient of food induustry on extensive margin is even negative), though the eect is insignicant.
Next we look at the estimated results of production-intensive and science-based sectors in the second
and third panels of table 9. Following the discussion above, innovation in these sectors should have more
signicant eects on export compared with supplier-dominated sectors. The hypothesis also indicates
that extensive margin should contribute more in science-based sectors than in production-intensive sec-
tors. Column (1) of the second and third panel displays that innovation has positive and signicant
impacts on overall exports in nine out of the ten industries in these two categories.
30
Among the eight
production-intensive industries, seven of them have signicant elasticities in intensive margin at 5% sig-
nicance level, while ve of them are signicant in the extensive margin. The average contribution of
extensive margin in production-intensive sectors is 31%, which is lower than that in science-based sec-
tors (45%). This conrms the hypothesis that increase in intensive margin caused by innovation plays
30
The only exception is petroleum.
20
a more important role in production-intensive industries than in science-based industries. Or equiva-
lently, extensive margin contributes relative more in science-based industries than in production-intensive
industries
The inter-industry comparison in this section shows that innovation has dierent degree of impact on
export performance across sectors. The impact is stronger in technology intensive sectors. The eect
of innovation on extensive and intensive margins also dier by the characteristics of sectors. Innovation
mostly aects intensive margin in production-intensive sectors which scale economies is important and
process innovation dominates. Innovation has relatively stronger eect on extensive margin in science-
based sectors which new product is the main mode of competition and product innovation plays an
important role.
4.5 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 innovations eect on export can change over time, and estimation by
year could help to explore it. Second, studies of product life cycle such as Klepper (1996)
31
demonstrate
the roles of product and process innovation are dierent in the early and late stages of a product life
cycle. When a new product is introduced, lots of rms enter the market and they oer diversied versions
of the product. The rate of product innovation in this stage is high. As a dominant design emerges
and product becomes standard, more eort is devoted to improving production process, and process
innovation becomes more important. Applying this concept into the framework of this paper implies
the correlation between innovation and the extensive margin will decrease with time as more products
under the current classication system move toward their later stages of product life cycles. This will
be studied in this section with year-by-year estimations.
Another subtle point is that the patent count data I use is calculated according to the application
year, which is truncated (especially in the last few years of the sample). The sample I use so far drops
observations from the last three years to reduce the possible bias accompanied with truncation. If we
estimate the model by year, we can use observations of these dropped years without worrying that they
will aect results of other years. We can then compare the results using patent number calculated by
application year and grant year (which has no truncation problem), to see if the two measurements lead
to dierent conclusion and to nd out if truncated years results in unusual outcomes.
31
For empirical evidence see Utterbak and Abernathy (1975), Anderson and Tushman (1990), Prusa and Schmitz (1993)
and Utterback and Suarez (1993).
21
The same two-way xed eect model in equation 9 is estimated, but now I separate the sample and
estimate the model by year. I will get twenty seven sets of estimated coecients and the year dummies
will be excluded from the model. We have to keep in mind that the results of the last few years are
more likely to suer from a bias caused by patent truncation. I summarize the estimated coecients
of patent (
b
,
1
) in the rst panel of table 10, and depict the elasticities of each margin against year in
gure 3. Figure 3 shows in general the impact of innovation on export declines as time passes in both
margins, especially in the seventies and earlier eighties. To have a better intuition of the numeric change
in elasticities over time, I calculate the average elasticities of the three margins in exports with respect to
innovation in the seventies, eighties, and nineties. These average elasticities are reported in the second
panel of table 10. The average elasticity before the eighties (1975-1979) is 0.65; this average drops to
0.53 in the eighties (1980-1989); and further drops to 0.52 in the nineties (1990-1999). From column (2)
and (3), we can see this falling trend occurs only on the extensive margins. The average elasticities of
extensive margin with respect to innovation in the three periods (i.e. seventies, eighties, and nineties)
are 0.23, 0.15 and 0.14, respectively. The average elasticities of intensive margin in these three periods
are 0.41, 0.38, and 0.39, respectively.
How do we interpret the falling impact of innovation on export? Two explanations are provided.
The rst is the increasing foreign direct investment (FDI). Compared with domestic rms, a subsidiary
of a multinational rm depends less on the host countrys technology level. For instance, a subsidiary of
Dell in Mexico could produce more advanced products than domestic Mexican rms under the technology
level of Mexico. As the proportion of export accounted by multinational rms increases with time, the
explanatory power of innovation becomes weaker, because innovation and export does not always take
place in the same country. Another explanation is the increased pace of knowledge and technology
transfer. The central concept of technology gap theory is that innovation gives countries a temporary
advantage in competition. If technology and knowledge transfer faster and more easily, the advantage
created by innovation fades faster and becomes relatively impotent. To identify whether the decreasing
eect of innovation on exports is driven by these two reasons, we need data of multinational rms exports
and technology transfer. I will leave this part for future research.
Next, I analyze how the relative contribution of extensive and intensive margins changes over time.
The average elasticity of extensive margin drops for more than half since the seventies (from 0.24 to
0.14), while in the same period the intensive margin falls for only less than one fth (from 0.41 to 0.39).
This means the importance of extensive margin relative to intensive margin declines over time. In the
seventies, on average 36% of the higher export caused by innovation is attributed to extensive margin;
22
this percentage drops to 29% in the eighties, and falls to 26% in the nineties. This result is consistent
with hypothesis embedded in product life cycle theories. The relative importance of extensive margin
regarding to the eect of innovation decreases with time, as more products become mature and reach the
late stages in their product life cycles.
Finally, I implement the year-by-year estimation again, but using a new sample classifying patents
according to the "grant" years. Dierent from using application year, using grant year does not cause
truncation problem, but it measures countries innovation capability with lags. The export elasticities
with respect to patent using grant year are in table 11. In most years, estimation using sample of grant
year has higher elasticities than estimation using application year, but the dierence is not large. More
interestingly, the elasticity of export seems starting to climb after 1998. If this happens only when using
application year, we suspect that it is caused by the truncated patent data. But the same pattern remains
when we use patent classied with grant year, which has no truncation problem. We will need data of
more recent years to identify whether this reversing trend after 1999 is due to truncation or because the
eect of innovation climbs. In summary, table 10 and 11 basically tell us the same story: the impact
of innovation on export decreases over time (at least until the year 1998). The decrease occurs on both
extensive and intensive margin, but the contribution of extensive margin declines with time.
5 Sensitivity Analysis
In this section I carry out a number of checks to see if the results are sensitive to changes in data
aggregation, innovation measurement, and model specication.
First, the relative importance of extensive margin and intensive margin is sensitive to the level of
aggregation in export data. In my sample, the number of commodity category exported from a country
is identied by each TS7 or HS10 code. To precisely estimate the relative contribution of the two
margins, it is critical to have disaggregate commodity classication in the export data. If commodities
are classied at a more aggregate level, it is more likely that dierent commodities are included in a same
commodity code. In this case, an increase in the number of varieties "within" each commodity code can
not be observed, and the eect will be attributed to the intensive margin (we will observe an increased
value of this commodity code). This type of within product variation is proved to exist and discussed
in studies such as Schott (2004). By intuition, contribution of extensive margin should be decreasing
when the export data becomes more aggregated. To check this eect in this paper, I follow Hummels
23
and Klenow (2005) to estimate equation 9 with dierent level of export variety aggregation.
32
Table
12 shows the elasticities of extensive margin with respect to innovation and country size under dierent
level of aggregation. As we can see, using more aggregate commodity classication will decrease the
elasticity and contribution of extensive margin with respect to innovation and country size. By using
the most disaggregated trade data (TS7 and HS10) available, this paper could estimate the contribution
of the two margins more accurately than studies using more aggregate data.
Next, I include an additional explanatory variable in the estimation which measures trade barriers
including tari and freight costs. The trade data described in section 3 contains freight cost as well as
duty paid at U.S. custom of each variety. Similar to Hummels and Klenow (2005), I dene an index of
trade barrier which equals the sum of duty and freight divided by total export value of each country-
industry in a given year, i.e. (tradc /arricr)
cit
=

j2I
cit
(dutycit+freightcit)

j2I
cit
vcijt
. I then estimate equation 9
and 10, but this time I include ln(tradc /arricr)
cit
in the estimation. The estimates are displayed in
table 13, in which results of the two-way xed eects case are recorded in column (1)-(3), and that of
three-way xed eects are in column (4)-(6). As expected, duty-freight rates have negative and very
signicant impacts on both extensive and intensive margins. More importantly, the magnitude as well
as the composition of patents eects does not alter after including duty-freight rate.
I also regress exports on lag patent instead of current patent. Producers may need some time to
adjust their production in response to a change in technological level, and the eect on exports could be
even slower. Also, it is more convincing that past patent are exogenous to current export, and thus using
lag patent could mitigate possible simultaneity problem. I replace |:(jatc:t
cit
) with |:(jatc:t
cit1
) and
|:(jatc:t
cit2
), separately, in equation 9 and estimate the model again. The sample size will become
smaller because some observations will be dropped out when we use lag patents. The results using
|:(jatc:t
cit1
) are shown in column 1-3 in table 14, and results using |:(jatc:t
cit2
) are in column 4-6.
Compared with the result using current patent, the elasticities of exports become lower, but the dierence
is not large.
In section 3, I discuss the reason and importance of weighting patents by the number of citation
received. The sample used so far apply the linear weighting scheme of Trajtenberg (1990), in which
the weight of citation equals 100%. To make sure that the result of this paper is not sensitive to the
selection of citation weight, I calculate patent counts with alternative citation weights equal: 0%, 10%,
30%, and 50%. I then estimate equation 9 with patent counts using dierent citation weights. Since
32
Other than TS7 and HS10 codes used so far, I estimate equation 9 with each variety j dened by 3 digit to 6 digit of
TS or HS code.
24
elasticities with respect to country size and trade resistances are very similar to those in benchmark case
in table 3, I only summarize elasticities with respect to patent in table 15. We could see the elasticities
with respect to patent is decreasing as we apply higher weight to citations.
33
The decrease in elasticities
occurs on both extensive margin and intensive margin, and the relative importance of the two margins
remain similar under dierent weight of citation. The bottom line is patent counts always have positive
and signicant impact on exports, and the main result of this paper is robust to the change of citation
weight.
Finally, I regress export, extensive margin, and intensive margin on patent, country size and trade
cost, without taking natural log on any of these variables. Patent, population and GDP per capita
are made relative to the world level.
34
The results are in table 16. Without taking log on export,
we could not decompose the impact of each explanatory variable on export performance into dierent
margins additively. The bottom is all three estimated coecients on innovation are still positive and very
signicant. Switching from a log-linear model to a linear model does not deny the impacts of innovation
on dierent margins in exports.
The analysis in this section shows the results in this paper are robust, and are not sensitive to changes
in level of export aggregation, weight of patent citation, and model specication.
6 Quality and Innovation
The benchmark results in section 4.1 show the intensive margins account for a relatively larger part
(70%) of innovations eects than the extensive margins do (30%). From trade theories, innovations can
increase the intensive margin by at least two ways: reducing production costs and increasing the quality
of products.
35
Empirical studies have shown evidence of the positive correlation between innovations and
productivity.
36
On the contrary, the relation between the quality of exported products and innovation
has not been studied intensively.
37
A common diculty faced by researchers is that product quality is not
33
A 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, Jae and Trajtenberg (2001).
Assigning higher weight to citations tend to assign higher weights to patents from rich countries. Since we already show
the eect of patent on export is decreasing with countries income level, the outcome in table 15 is not too surprising.
34
Patent now equals a countrys weighted patent count divided by the total number of weighted patent count of all
countries, in an industry-year. Population variable equals a countrys population divided by total population of all
countries in the sample of each year. GDP per capita variable equals a countrys GDP per capita divided by the average
of all countries GDP per capita of all countries in the sample of each year.
35
This issue is related to Schott (2004), in which he used unit value to check the validity between old and new trade
theories. This section is dierent from his paper in that I emphasize the role of innovation.
36
Early studies are surveyed by Griliches (1979). Recent studies mainly use rm level data, such as Wakelin (2001);
Grith et al., (2006).
37
In empirical international trade literatures, export quality has been connected to factor endowments (Schott, 2004),
income level (Schott, 2004; Hallak, 2006) and dierent types of export cost (Hummels and Skiba, 2004).
25
directly observable. In empirical trade literatures, it is usually inferred from export price (unit-value).
Some literature used export price as an index of quality (Hummels and Skiba, 2004; Hallak, 2006), i.e.
within each variety, higher export prices are equivalent to products of higher quality. Others studies
considered export price as an indicator of quality (Hallak, 2006; Hummels and Klenow, 2005; Hallak and
Schott, 2008), i.e. export price reects quality of products as well as other factors such as production
cost.
38
These studies agree that product quality is an important, if not the only, factor that determines
export price. In this section I adopt the second view, that unit price is possibly aected by both the
quality of products and production cost. The object of this section is to decompose intensive margin
into price and quantity, and then to look for evidence of quality improvement induced by innovation.
Consider two extreme scenarios. First, suppose innovation could only reduce production cost but
could not improve the quality of products. Lower production cost enhances producers competitiveness,
which enables them to export at lower prices. If innovation increases intensive margin by reducing
production cost, we expect export price to be lower and export quantity to be higher in more innovative
countries. The second polar case is that innovation enables producers to produce goods of higher quality,
but does not decrease production cost (e.g. Grossman and Helpman [1991]). In reality, it is reasonable
to assume that the production cost is increasing with quality. Innovations make producers to export
goods of higher quality at no lower price; but their export volume could increase or fall.
39
According
to these two cases: if innovation increases export quantity but decreases export price, it is very likely
that innovation increases intensive margin by reducing production cost. On the contrary, if both export
price and quantity increase with innovations, we nd some support of quality improvement, i.e. quality
improvement explains why consumers buying more at higher price.
6.1 Decomposition of Intensive Margin
The rst step is to construct the method of decomposition of the intensive margin into price index and
quantity index. In equation 6,
cijt
represents the export value of a commodity , belongs to industry i
exported by country c in year t. The value of export could be expressed as
cijt
= j
cijt

cijt
. j
cijt
is the
average unit price of variety ,, and and
cijt
is export quantity of variety ,. The trade data from Feenstra
(1996) and Feenstra, Romalis and Schott (2002) contains both the value and quantity of each product
variety dened by TS7 or HS10. The quantity information can be used as
cijt
directly, and the average
38
Dierent export price can also be caused by within variety dierentiation. So it is very important to have disaggregate
trade data to measure product quality more precisely. See Hallak (2006).
39
In a model with dierentiated-quality such as Grossman and Helpman (1991), 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 decreases, export quantity will increase with price.
26
price can be inferred from dividing export value by quantity: j
cijt
=
vcijt
qcijt
. Following the same procedure,
I express the world export value of variety as
wijt
= j
wijt

wijt
, where world value
wijt
=
c

cijt
and
world quantity
wijt
=
c

cijt
. I could derive the world average price of variety from j
wijt
=
cvcijt
cqcijt
.
The intensive margin in equation 6 could be re-written as
i:tc::ic
cit
=

j2Icit

cijt

j2Icit

wijt
=

j2Icit
j
cijt

cijt

j2Icit
j
wijt

wijt
(13)
I then calculate the price index and quantity index in industry level following the well known Fisher
Index:
40
1
cit
= [

j2Icit
j
cijt

cijt

j2Icit
j
wijt

cijt
]
1
2
[

j2Icit
j
cijt

wijt

j2Icit
j
wijt

wijt
]
1
2
(14)
Q
cit
= [

j2Icit
j
cijt

cijt

j2Icit
j
cijt

wijt
]
1
2
[

j2Icit
j
wijt

cijt

j2Icit
j
wijt

wijt
]
1
2
(15)
The price index in equation 14 is the geometric average of two elements. The rst element is the Paasche
index. It is the sum of country cs unit prices of all varieties in industry i that country c exports, divided
by the sum of world unit prices of the same varieties. The weight of each varietys unit price is country
cs export quantity,
cijt
. The second element of price index is the Laspeyres index. Again, it is the sum
of country cs unit prices divided by the sum of world unit prices, but this time world quantity
wijt
is
used as the weight. Using the same method, the quantity index in equation 15 is derived as the geometric
average of the Paasche index and the Laspeyres.
41
The intensive margin in export can be decomposed
into the price index and quantity index
i:tc::ic
cit
= 1
cit
Q
cit
(16)
6.2 Data
To construct the price index and quantity index of each commodity as in equation 14 and 15, we need
both the value and quantity information of each variety exported. Unfortunately, the trade data does
not have complete and perfect quantity information of each variety. From 1975-2001, about 18% varieties
40
In Hummels and Klenow (2005), they use a modied Fisher Index which considers the eect of new-entering products
to construct the price and quantity indices in country level.
41
Because single countrys export quantity index can not exceed world export quantity, i.e. q
cijt
q
wijt
, the quantity
is between [0,1]. But a countrys export price can be higher or lower than world price, so price index can be greater or
smaller than 1.
27
exported to the U.S. do not have quantity information. These varieties have to be dropped out when
I calculate industrial level price and quantity indices.
42
As a result, the sample used in this section is
dierent from the previous one. Some observations will have dierent values in exports data. Some
other observations, especially those with very few positive export varieties, will be dropped out from the
sample. The sample size used in this section is thus smaller than that of section 4.1 and contains 22,804
observations. 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.
6.3 Estimation and Results
Because the sample used in this section is dierent from that used in section 4.1, I re-estimate the two-
way xed model in equation 9 using the new sample, and record the results in column 1-3 in table 17.
Comparing the coecients of innovation with those in table 3, we can see the elasticity of overall exports
with respect to innovation falls slightly from 0.537 to 0.519. The elasticity of extensive margin increases
(from 0.160 to 0.219) and the elasticity of intensive margin decreases (from 0.378 to 0.300). Though
the elasticities are dierent, the big picture is still very close to that of section 4.1. The inuence of
innovation on each margin remains positive, and they are still statistically signicant.
I then decompose the eects of innovation on the intensive margin into price and quantity. I regress
the industrial price index and quantity index on innovation as well as same explanatory variables as
in equation 9 (population, GDP per capita, and trade cost). Industry dummies and year dummies
are controlled, and standard errors are robust to heteroskedasticity and clustered by country. Since
ln(i:tc::ic
cit
) = ln(1
cit
) ln(Q
cit
), the estimated elasticity of intensive margin with respect to each
regressor could be decomposed additively:
b
,
im
i
=
b
,
p
i
+
b
,
q
i
i = 1......6, where
b
,
im
i
,
b
,
p
1
, and
b
,
q
1
are the
estimated elasticities of intensive margin, price and quantity, respectively. According to the two assumed
polar cases, we have two competing hypotheses. If innovations increase intensive margin only by reducing
production cost, we expect to see
b
,
p
1
< 0 and
b
,
q
1
0. If innovations increase intensive margin only by
raising product quality, we expect to see
b
,
p
1
0; and the sign of
b
,
q
1
is determined by quality-adjusted
price.
43
The estimated elasticities of price and quantity indices are in table 17 column 4 and 5: a 10% increase
42
In addition to the missing data problem, the quantity is recorded with measurement errors (See General Accounting
Oce (1995)). To mitigate this problem, I also estimate the model using an alternative sample which drops observations
with extreme unit value (ve times above or below the mean) and observations with very low quantity (below 50 units).
Similar criteria is applied in Hallak (2006). Since using this sample does not materially alter the conclusion, I will not
report the results in this section.
43
If
b

q
1
is negative, the magnitude should be smaller than
b

p
1
to have a positive correlation between innovation and
intensive margin.
28
in patents will increase price by 0.5% and quantity by 2.5%. High intensive margin led by innovation
is mainly driven by higher export quantity (91%), and price accounts for a relatively small contribution
(9%). Although innovation does not have economically strong eect on export prices, the eect is
statistically signicant. The estimates support trade theories (Grossman and Helpman [1991]) which
assume innovation increases quality: because quality of products becomes higher, consumers are willing
to buy more even at a higher price. The rise in export quantity implies the quality-adjusted price falls.
One thing worth noting is the support to quality-improvement innovation does not necessarily contradict
other empirical studies which nd innovation can increase productivity. It is possible the two eects
coexist, just the quality eect is stronger than the productivity eect which result in a slightly higher
export price. In summary, the estimate shows evidence that innovation increases the quality of exported
goods.
44
7 Conclusion
In this paper I decompose exports of each country into dierent margins, to investigate to which extent
does innovation increase the number of varieties, the volume of each good, and the quality of exported
products. Understanding the inuence of innovation on each margin is important to analyze welfare
eects and R&D policies. I use detailed trade data of 12 manufacturing industries from 105 countries
for the period 1975-2001 to construct the indices of dierent margins in exports. Patent granted by the
U.S. is used as the proxy of innovation. The impact of innovation on the dierent margins is estimated
by two models: a static model using dummy variable regressions, and a dynamic model estimated by a
Generalized Method of Moment (GMM) estimator which regards innovation as potentially endogenous.
I nd that innovation has positive and signicant eect on the extensive margin as well as the intensive
margin in exports. A country with 10% more patents will export 1.6% more varieties and 3.8% higher
value of each variety on average, which leads to a 5.4% increase in overall exports. Since the inuence on
each margin is statistically signicant, trade theories predicting innovation can increase either only the
extensive margins (Krugman [1979]) or only the intensive margin (Flam and Helpman [1987]) could not
explain the full eects of innovations on exports. This motivates the development of a theoretical model
which could reconcile innovations impacts on both margins. A further decomposition of the intensive
44
Other coecients in column 4 and 5 can be summarized as follows. A 10% increase in population leads to a 0.3%
decrease in unit value, and a 5.1% increase in quantity. Price increases by 1.2% and quantity increases by 4.5% when GDP
per capita increases by 10%. Larger countries exports higher quantity at slightly lower price. These nding are consistent
with Hummels and Klenow (2005). Among the three variables represent trade cost, only border has signicant coecients.
Sharing a common border reduces export price and increases export quantity, which is an intuitive outcome.
29
margin into price and quantity components shows that countries with more innovations export higher
quantity at a slightly higher price. This result coincides with theories (Grossman and Helpman [19991])
that innovation raises products qualities.
Other ndings of this paper are summarized as the follows. Innovation has stronger impact on
developing countries, especially on the extensive margins. This points out that technological catch-
up in the developing countries is critical for them to expand the scope of exports, and the eect is a
welfare improvement. In the industry-by-industry estimation, the inuence of innovation is stronger and
more signicant in technology-intensive sectors. The relative contribution of the two margins also dier
across industrial characteristics. Intensive margin accounts for a larger part in production-intensive
sectors like machinery; while the extensive margin is relatively important in science-based sectors such
as chemicals and electronics. The producer-user relation of innovations in Pavitt (1984) is veried with
countries export performances. This nding also implies governments target subsidizing R&D should
take into account that it will lead to dierent impacts in dierent industries. Finally, the year-by-year
estimation displays that the importance of extensive margin falls with time, which supports product life
cycle literature predicting product innovation dominates when a product is newly introduced, and process
innovation becomes important when the production grows mature (Klepper [1996]). The estimation also
illustrates the impact of innovation on overall export as well as each margin has declined as time passes.
Possible explanations include the increase in intra-rm trade and faster technological transfer between
countries. To conrm these hypothesis requires further data and estimations and will be addressed in
future research.
30
References
[1] Anderson, Philip and Tushman, Michael L. "Technological Discontinuities and Dominant Designs: A
Cyclical Model of Technological Change." Administrative Science Quarterly, 1990, 35(4), pp. 604-33.
[2] Amable, B. and Verspagen, B. "The Role of Technology in Market Shares Dynamics." Applied
Economics, 1995, 27(2), pp. 197-204.
[3] Arellano, Manuel and Bond, Stephen. "Some Tests of Specication for Panel Data: Monte Carlo
Evidence and an Application to Employment Equations." The Review of Economic Studies, 1991,
58(2), pp. 277-97.
[4] Arellano, Manuel and Boverb, Olympia. "Another Look at the Instrumental Variable Estimation of
Error-Components Models." Journal of Econometrics, 1995, 68, pp. 29-51.
[5] Bernard, Andrew B. and Jensen, J. Bradford. "Exceptional Exporter Performance: Cause, Eect,
or Both?" Journal of International Economics, 1999, 47, pp. 1-25.
[6] Bernard, Andrew B. and Jensen, J. Bradford. "Why Some Firms Export." The Review of Economics
and Statistics, 2004, 86(2), pp. 561-69.
[7] Bleaney, Michael and Wakelin, Katherine. "Eciency, Innovation and Exports." Oxford Bulletin of
Economics and Statistics, 2002, 64(1), pp. 3-15.
[8] Blundell, Richard and Bond, Stephen. "Initial Conditions and Moment Restrictions in Dynamic
Panel Data Models." Journal of Econometrics, 1998, 87, pp. 115-43.
[9] Branstetter, Lee G. "Looking for International Knowledge Spillovers: A Review of the Literature
with Suggestions for New Approaches." Annales dEconomie et de Statistique, 1998, 49-50, pp.
517-40.
[10] Besedes, T., & Prusa, T. J. "The Role of Extensive and Intensive Margins and Export Growth."
NBER Working Paper Series, 2007, w13628.
[11] Buxton, T.; Mayes, D. and Murn, A. "UK Trade Performance and R&D." Economics of Innovation
and New Technology, 1991, 1(3), pp. 243-56.
[12] Comanor, W. S. and F. M. Scherer (1969). "Patent Statistics as a Measure of Technical Change."
The Journal of Political Economy 77(3): 392-398.
31
[13] Diewert, W.E. "The Eects of an Innovation: A Trade Theory Approach." The Canadian Journal
of Economics, 1987, 20(4), pp. 694-714.
[14] Dollar, David. "Technological Innovation, Capital Mobility, and the Product Cycle in North-South
Trade." The American Economic Review, 1986, 76(1), pp. 177-90.
[15] Dosi, G; Pavitt, K and Soete, L. The Economics of Technical Change and International Trade. New
York: New York University Press, 1990.
[16] Evenett, Simon J. and Venables, Anthony J. "Export Growth by Developing Countries: Market
Entry and Bilateral Trade." Working Paper, 2002.
[17] Eaton, Jonathan and Kortum, Samuel. "Engines of Growth: Domestic and Foreign Sources of Inno-
vation." Japan and the World Economy, 1997, 9, pp. 235-59.
[18] Eaton, Jonathan and Kortum, Samuel. "Technology and Bilateral Trade." NBER Working Paper
Series, 1997, 6253.
[19] Eaton, Jonathan and Kortum, Samuel. "International Technology Diusion: Theory and Measure-
ment." International Economic Review, 1999, 40(3), pp. 537-70.
[20] Eaton, Jonathan and Kortum, Samuel. "Technology, Trade, and Growth: A Unied Framework."
European Economic Review, 2001, 45, pp. 742-55.
[21] Eaton, Jonathan and Kortum, Samuel. "Technology, Geography, and Trade." Econometrica, 2002,
70(5), pp. 1741-79.
[22] Fagerberg, Jan. "International Competitiveness: Errata." Economic Journal, 1988, 98(393), pp.
1203.
[23] Fagerberg, Jan. "Technology and Competitiveness." Oxford Review of Economic Policy, 1996, 12(3),
pp. 39-51.
[24] Feenstra, Robert C. "U.S. Imports, 1972-1994: Data and Concordances." NBER Working Papers,
1996, 5515.
[25] Feenstra, Robert C.; Romalis, John and Schott, Peter K. "U.S. Imports, Exports, and Tari Data,
1989-2001." NBER Working Papers, 2002, 9387.
[26] Felbermayr, G. J., & Kohler, W. "Does Wto Membership Make a Dierence at the Extensive Margin
of World Trade?" CESifo Working Paper Series, 2007.
32
[27] Flam, Harry and Helpman, Elhanan. "Vertical Product Dierentiation and North-South Trade."
American Economic Review, 1987, 77(5), pp. 810-22.
[28] Funke, M., & Ruhwedel, R. . "Export Variety and Export Performance: Empirical Evidence from
East Asia." Journal of Asian Economics, 2001, 12(4), pp. 493-505.
[29] General Accounting Oce, "US Imports: Unit Values Vary Widely for Identically Classied Com-
modities." Report GAO/GGD-95- -90,1995.
[30] Grith, Rachel; Huergo, Elena; Mairesse, Jacques and Peters, Bettina. "Innovation and Productivity
across Four European Countries." NBER Working Paper Series, 2006, 12722.
[31] Griliches, Z. (1979). "Issues in Assessing the Contribution of Research and Development to Produc-
tivity Growth." The Bell Journal of Economics 10(1), pp. 92-116.
[32] Griliches, Z. (1990). "Patent Statistics as Economic Indicators: A Survey." Journal of Economic
Literature 28(4): 1661-1701.
[33] Griliches, Zvi; Hall, Bronwyn H. and Pakes, Ariel. "R&D, Patents, and Market Value Revisited: Is
There a Second (Technological Opportunity) Factor?" Journal of Economics of Innovation and New
Technology, 1991, 1(1), pp. 183-201.
[34] Grossman, Gene M. and Helpman, Elhanan. "Product Development and International Trade." The
Journal of Political Economy, 1989, 97(6), pp. 1261-83.
[35] Grossman, Gene M. and Helpman, Elhanan. "Quality Ladders in the Theory of Growth." The Review
of Economic Studies, 1991, 58(1), pp. 43-61.
[36] Grossman, Gene M. and Helpman, Elhanan. "Quality Ladders and Product Cycles." The Quarterly
Journal of Economics, 1991, 106(2), pp. 557-86.
[37] Gruber, William; Mehta, Dileep; Vernon, Raymond. "The R & D Factor in International Trade
and International Investment of United States Industries." The Journal of Political Economy, 1967,
75(1), pp. 20-37.
[38] Hall, Bronwyn H.; Jae, Adam B. and Trajtenberg, Manuel. "Market Value and Patent Citations:
A First Look." NBER Working Paper Series, 2000, 7741.
[39] Hall, Bronwyn H.; Jae, Adam B. and Trajtenberg, Manuel. "The Nber Patent Citation Data File:
Lessons, Insights and Methodological Tools." NBER Working Papers, 2001, 8498.
33
[40] Hallak, Juan C. "Product Quality and the Direction of Trade." Journal of International Economics,
2006, 68(1), pp. 238-65.
[41] Hallak, Juan C. and Schott, Peter K. "Estimating Cross-Country Dierences in Product Quality."
NBER Working Paper Series, 2008, 13807.
[42] Haveman, Jon "International Trade Data: http://www.macalester.edu/research/economics/page/haveman/trade.resources/tradedata.html".
[43] Heston, Alan; Summers, Robert; Aten, Bettina. "Penn World Table Version 6.2, Center for Interna-
tional Comparisons of Production, Income and Prices at the University of Pennsylvania, 2006.
[44] Hummels, David and Klenow, Peter J. "The Variety and Quality of a Nations Exports." American
Economic Review, 2005, 95(3), pp. 704-23.
[45] Hummels, D., Skiba, A. "Shipping the Good Apples Out? An Empirical Conrmation of the Alchian-
Allen Conjecture." Journal of Political Economy, 2004, 112, pp. 1384-402.
[46] Jae, 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 Commercial Innovation." The Journal
of Industrial Economics, 1998, 46(2), pp. 183-205.
[47] Jae, Adam B.; Trajtenberg, Manuel and Fogarty, Michael S. . "The Meaning of Patent Citations:
Report on the Nber/Case-Western Reserve Survey of Patentees." NBER Working Paper Series, 2000,
7631.
[48] Jensen, Richard and Thursby, Marie. "A Strategic Approach to the Product Life Cycel." Journal of
International Economics, 1986, 21, pp. 269-84.
[49] Keesing, D. B. "The Impact of Research and Development on United States Trade." Journal of
Political Economy, 1967, 75(1), pp. 38-48.
[50] Klepper, Steven. "Entry, Exit, Growth, and Innovation over the Product Life Cycle." The American
Economic Review, 1996, 86(3), pp. 562-83.
[51] Kortum, Samuel S. "Research, Patenting, and Technological Change." Econometrica, 1997, 65(6),
pp. 1389-419.
[52] Krugman, Paul. "A Model of Innovation, Technology Transfer, and the World Distribution of In-
come." Journal of Political Economy, 1979, 87(2), pp. 253-66.
34
[53] Lachenmaier, Stefan and Wossmann, Ludger. "Does Innovation Cause Exports? Evidence from
Exogenous Innovation Impulses and Obstacles Using German Micro Data." Oxford Economic Papers,
2006, 58(2), pp. 317-50.
[54] Mahmood, Ishtiaq P. and Singh, Jasjit. "Technological Dynamism in Asia." Research Policy, 2003,
32(6), pp. 1031-54.
[55] Pavitt, Keith. "Sectoral Patterns of Technical Change: Towards a Taxonomy and a Theory." Re-
search Policy, 1984, 13, pp. 343-73.
[56] Posner, M. V. "International Trade and Technical Change." Oxford Economic Papers, 1961, 13(2),
pp. 323-41.
[57] Prusa, Thomas J. and James A. Schmitz, Jr. "Are New Firms an Important Source of Innovation?"
Economics Letter, 1993, 35, pp. 339-42.
[58] Roberts, Mark J. and Tybout, James R. "The Decision to Export in Colombia: An Empirical Model
of Entry with Sunk Costs." The American Economic Review, 1997, 87(4), pp. 545-64.
[59] Schott, Peter. "Across-Product Versus Within-Product Specialization in International Trade." The
Quarterly Journal of Economics, 2004, 199(2), pp. 646-677.
[60] Soete, Luc. "The Impact of Technological Innovation on International Trade Patterns: The Evidence
Recorded." Research Policy, 1987, 16, pp. 101-30.
[61] Spencer, Barbara J. and Brander, James A. "International R&D Rivalry and Industrial Strategy."
Review of Economic Studies, 1983, pp. 707-22.
[62] Sterlacchini, A. "Do Innovative Activities Matter to Small Firms in Non-R&D-Intensive Industries?
An Application to Export Performance." Research Policy, 1999, 28(8), pp. 819-32.
[63] Sterlacchini, Alessandro. "The Determinants of Export Performance: A Firm-Level Study of Italian
Manufacturing." Review of World Economics, 2001, 137(3), pp. 450-72.
[64] Trajtenberg, M. (1990). "A Penny for Your Quotes: Patent Citations and the Value of Innovations."
The RAND Journal of Economics 21(1): 172-187.
[65] Utterback, James M and Abernathy, William J. "A Dynamic Model of Process and Product Inno-
vation." OMEGA, 1975, 3(6), pp. 639-56.
35
[66] Utterback, James M. and Suarez, Fernando F. "Innovation, Competition, and Industry Structure."
Research Policy, 1993, 22, pp. 1-21.
[67] Vernon, Raymond. "International Investment and International Trade in the Product Cycle." The
Quarterly Journal of Economics, 1966, 80(2), pp. 190-207.
[68] Wakelin, K. (1998). "The Role of Innovation in Bilateral OECD Trade Performance." Applied Eco-
nomics 30(10): 1334-46.
[69] Wakelin, K. (2001). "Productivity Growth and R&D Expenditure in UK Manufacturing Firms."
Research Policy 30, pp. 1079-1090.
[70] Wooldridge, Jerey M. Econometric Analysis of Cross Section and Panel Data. Cambridge, Massa-
chusetts: The MIT Press, 2002.
36
A Appendix: Data
Since my focus is industrial innovation, the rst task is to classify each patent into industry (or industries).
In the patent dataset, most patents could not be classied into a single 2-digit SIC industry. USPTO has
developed a classication system for the technologies to which the patented inventions belong, consisting
of about 400 main patent classes. In Hall et. al. (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 of innovations. This assumes each patent
has equal quality and importance. I take the number of citations received by each patent to measure
quality and importance, and use it as weight to calculate the "weighted patent counts" as a proxy of
innovation. The average number of citation received by each patent is 5, and I set the weight of each
citation received equals 10% of patent counts.
45
This selection of weight seems to be arbitrary, but I will
compare the results using dierent weights in the sensitivity analysis. It turns out the results of using
dierent 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 Oce
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 reected on
the year patents are applied, not the year patents are granted.
46
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
45
For example, a patent which has been cited for ve times will be 1.5 patent counts.
46
For example, a patent applied in 1990 and granted in 1995 should reect the innovation capability of the country in
1990, not 1995.
37
when it is applied. This means if I count the number of patents according to the application year, I
have data truncation: I can not 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 can not accurately measure the
number of patents applied in each year especially in the last few years of the sample.
47
To mitigate the
possible measurement bias caused by truncation, I drop observations of the last 3 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 eect 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 aected by inclusionnexclusion of observations of the last few years.
I will also compare results using grant year to classify patents (which does not suer from the truncation
problem) with results using application year, and there is no material dierence.
47
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.
38
B Appendix: Figures and Tables
-
1
0
-
5
0
5
l
o
g

e
x
t
e
n
s
i
v
e

m
a
r
g
i
n
-4 -2 0 2 4 6
log patent
coef = .15909729, (robust) se = .04503191, t = 3.53
extensive margin
-
1
0
-
5
0
5
l
o
g

i
n
t
e
n
s
i
v
e

m
a
r
g
i
n
-4 -2 0 2 4 6
log patent
coef = .36869146, (robust) se = .04858967, t = 7.59
intensive margin
Figure1: partial-regression plots of export against innovation
Note: The horizontal axes in the two plots are natural log of patent counts of each country-industry.
A country which has more patents is considered as more innovative. The vertical axis in the left plot is
log of each countrys extensive margin of export to the U.S., and that in the right plot is the log intensive
margin. High extensive margin means a country exports more varieties of products, and high intensive
margin means a country exports higher value of those varieties it exports. Each point in these gures
represents the combination of a countrys extensive/intensive margin of export and patent in industrial
level, using the data of the year 1990.
39
-
2
2
-
2
1
-
2
0
-
1
9
-
1
8
-
1
7
l
n

e
x
p
o
r
t
0 1 2 3 4 5
ln patent
low mid
high
Figure 2.1: Predicted patent and overall export in countries with dierent income levels
Note: Figure 2.1-2.3 show the predicted correlation between exports (overall export, extensive margins,
and intensive margin) and patent. All lines are positive sloping, showing countries which have more patent
export more. In each gure, the low-income countrys line is the steepest, followed by the middle-income
countrys, and last the high-income countrys. This indicates the positive impact of patent on export is
stronger in low-income country than in high-income country.
40
-
1
1
-
1
0
.
5
-
1
0
-
9
.
5
-
9
l
n

e
x
t
e
n
s
i
v
e
0 1 2 3 4 5
ln patent
low mid
high
Figure 2.2: Predicted patent and extensive margin in countries with dierent income levels
41
-
1
0
-
9
.
5
-
9
-
8
.
5
-
8
-
7
.
5
l
n

i
n
t
e
n
s
i
v
e
0 1 2 3 4 5
ln patent
low mid
high
Figure 1: Figure 2.3: Predicted patent and intensive margin in countries with dierent income levels
42
0
.
1
.
2
.
3
.
4
.
5
.
6
.
7
.
8
.
9
1
e
l
a
s
t
i
c
i
t
y

o
f

e
x
p
o
r
t
s
1975 1980 1985 1990 1995 2000
year
overall_export extensive
intensive
Figure 3: The Elasticities of Export with Respect to Innovation by Year
43
Table 1: List of countries
Algeria Costa Rica Haiti Laos Pakistan Sweden
Argentina Cyprus Honduras Liberia Panama Switzerland
Australia Denmark Hong Kong Madagascar Paraguay Tanzania
Austria Dominican Hungary Malawi Peru Taiwan
Bahamas Ecuador Iceland Malaysia Philippines Thailand
Bahrain Egypt India Mali Poland Trinidad and Tobago
Belgium El Salvador Indonesia Mauritania Portugal Tunisia
Belize Ethiopia Iran Mauritius Qatar Turkey
Bolivia Germany Iraq Mexico South Africa Uganda
Brazil Fiji Ireland Morocco Romania United Arab Emirates
Bulgaria Finland Israel Nepal Saudi Arabia United Kingdom
Cameroon France Italy Netherlands Senegal Uruguay
Canada Gabon Jamaica New Zealand Sierra Leone Venezuela
Chad Ghana Japan Nicaragua Singapore Zambia
Chile Greece Jordan Niger Spain Zimbabwe
China Guatemala Kenya Nigeria Sri Lanka
Colombia Guinea Korea Norway Sudan
Congo Guyana Kuwait Oman Suriname
44
T
a
b
l
e
2
:
L
i
s
t
o
f
i
n
d
u
s
t
r
i
e
s
i
n
2
-
d
i
g
i
t
S
I
C
8
7
S
I
C
D
e
s
c
r
i
p
t
i
o
n
E
x
a
m
p
l
e
S
u
b
-
C
a
t
e
g
o
r
y
2
0
F
o
o
d
a
n
d
k
i
n
d
r
e
d
p
r
o
d
u
c
t
s
F
l
u
i
d
M
i
l
k
,
C
a
n
n
e
d
F
r
u
i
t
s
,
2
2
T
e
x
t
i
l
e
m
i
l
l
p
r
o
d
u
c
t
s
F
i
n
i
s
h
e
r
s
o
f
B
r
o
a
d
w
o
v
e
n
F
a
b
r
i
c
s
o
f
C
o
t
t
o
n
2
8
C
h
e
m
i
c
a
l
s
a
n
d
a
l
l
i
e
d
p
r
o
d
u
c
t
s
M
e
d
i
c
i
n
a
l
C
h
e
m
i
c
a
l
s
,
B
i
o
l
o
g
i
c
a
l
P
r
o
d
u
c
t
s
,
2
9
P
e
t
r
o
l
e
u
m
a
n
d
c
o
a
l
p
r
o
d
u
c
t
s
P
e
t
r
o
l
e
u
m
R
e

n
i
n
g
,
A
s
p
h
a
l
t
F
e
l
t
s
a
n
d
C
o
a
t
i
n
g
s
3
0
R
u
b
b
e
r
a
n
d
m
i
s
c
e
l
l
a
n
e
o
u
s
p
l
a
s
t
i
c
s
p
r
o
d
u
c
t
s
F
a
b
r
i
c
a
t
e
d
R
u
b
b
e
r
P
r
o
d
u
c
t
s
,
P
l
a
s
t
i
c
s
B
o
t
t
l
e
s
3
2
S
t
o
n
e
,
c
l
a
y
,
g
l
a
s
s
,
a
n
d
c
o
n
c
r
e
t
e
p
r
o
d
u
c
t
s
G
l
a
s
s
C
o
n
t
a
i
n
e
r
s
,
C
o
n
c
r
e
t
e
P
r
o
d
u
c
t
s
3
3
P
r
i
m
a
r
y
m
e
t
a
l
i
n
d
u
s
t
r
i
e
s
S
t
e
e
l
I
n
v
e
s
t
m
e
n
t
F
o
u
n
d
r
i
e
s
,
P
r
i
m
a
r
y
S
m
e
l
t
i
n
g
a
n
d
R
e

n
i
n
g
3
4
F
a
b
r
i
c
a
t
e
d
m
e
t
a
l
p
r
o
d
u
c
t
s
H
e
a
t
i
n
g
E
q
u
i
p
m
e
n
t
,
F
a
b
r
i
c
a
t
e
d
S
t
r
u
c
t
u
r
a
l
M
e
t
a
l
3
5
I
n
d
u
s
t
r
i
a
l
m
a
c
h
i
n
e
r
y
a
n
d
e
q
u
i
p
m
e
n
t
P
r
i
n
t
i
n
g
T
r
a
d
e
s
M
a
c
h
i
n
e
r
y
a
n
d
E
q
u
i
p
m
e
n
t
,
E
l
e
c
t
r
o
n
i
c
C
o
m
p
u
t
e
r
s
3
6
E
l
e
c
t
r
i
c
a
l
a
n
d
e
l
e
c
t
r
o
n
i
c
e
q
u
i
p
m
e
n
t
H
o
u
s
e
h
o
l
d
A
u
d
i
o
a
n
d
V
i
d
e
o
E
q
u
i
p
m
e
n
t
,
T
e
l
e
p
h
o
n
e
A
p
p
a
r
a
t
u
s
3
7
T
r
a
n
s
p
o
r
t
a
t
i
o
n
e
q
u
i
p
m
e
n
t
M
o
t
o
r
V
e
h
i
c
l
e
P
a
r
t
s
,
A
i
r
c
r
a
f
t
,
G
u
i
d
e
d
M
i
s
s
i
l
e
a
n
d
S
p
a
c
e
V
e
h
i
c
l
e
3
8
I
n
s
t
r
u
m
e
n
t
s
a
n
d
r
e
l
a
t
e
d
p
r
o
d
u
c
t
s
S
u
r
g
i
c
a
l
a
n
d
M
e
d
i
c
a
l
I
n
s
t
r
u
m
e
n
t
s
,
I
n
s
t
r
u
m
e
n
t
s
f
o
r
M
e
a
s
u
r
i
n
g
45
Table 3: Dummy variable regressions with two-way xed eects
Dependent Variables ! ln export ln extensive ln intensive
Regressors # (1) (2) (3)
ln patent 0.537*** 0.160*** 30% 0.378*** 70%
(0.075) (0.040) (0.043)
ln population 1.046*** 0.546*** 52% 0.501*** 48%
(0.089) (0.041) (0.054)
ln
GDP
population
1.505*** 0.891*** 59% 0.614*** 41%
(0.17) (0.080) (0.10)
ln distance -0.616** -0.383*** 62% -0.233 38%
(0.29) (0.13) (0.17)
language 0.339 0.199 59% 0.140 41%
(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
1
2
0.61 0.51 0.53
Notes: This table reports the estimated results of equation 9. The estimated coecients are elasticities of
dierent margins in exports with respect to each regressor. Robust standard errors clustered by country are in
parentheses. Percentages describe the contribution of each margin to overall exports. Coecients of dummy
variables are skipped. * indicates signicance level: *** p<0.01, ** p<0.05, * p<0.1.
46
Table 4: Dummy variable regressions with three-way xed eects
Dependent Variables ! ln export ln extensive ln intensive
Regressors # (1) (2) (3)
ln patent 0.584*** 0.196*** 34% 0.388*** 66%
(0.063) (0.025) (0.046)
ln population 1.144* 0.561* 49% 0.584 51%
(0.64) (0.30) (0.42)
ln
GDP
population
0.614* 0.381*** 62% 0.233 38%
(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
1
2
0.72 0.61 0.60
Notes: This table reports the estimated results of equation 10. The estimated coecients are elasticities of
dierent margins in exports with respect to each regressor. Robust standard errors clustered by country are in
parentheses. Percentages describe the contribution of each margin to overall exports. Coecients of dummy
variables are skipped. * indicates signicance level: *** p<0.01, ** p<0.05, * p<0.1.
47
Table 5: Dynamic GMM estimates
Exogenous Innovation Endogenous Innovation
Dependent Variables ! ln export ln extensive ln intensive ln export ln extensive ln intensive
Regressor # (1) (2) (3) (4) (5) (6)
ln patent 0.272*** 0.109*** 0.235*** 0.417*** 0.123*** 0.359***
(0.022) (0.012) (0.018) (0.060) (0.043) (0.058)
ln population 0.682*** 0.460*** 0.417*** 0.456*** 0.433*** 0.285***
(0.042) (0.022) (0.028) (0.060) (0.037) (0.050)
ln
GDP
population
1.014*** 0.772*** 0.543*** 0.578*** 0.719*** 0.277***
(0.066) (0.039) (0.049) (0.11) (0.075) (0.10)
ln distance -0.316*** -0.259*** -0.160** -0.200** -0.244*** -0.0943
(0.091) (0.053) (0.077) (0.084) (0.054) (0.076)
language 0.189* 0.162*** 0.0961 0.0456 0.146** 0.00129
(0.098) (0.060) (0.086) (0.096) (0.063) (0.089)
border 0.719*** -0.131 0.984*** 0.635*** -0.125 0.922***
(0.26) (0.16) (0.24) (0.24) (0.16) (0.22)
ln export t-1 0.355*** 0.448***
(0.026) (0.027)
ln extensive t-1 0.124*** 0.152***
(0.023) (0.022)
ln intensive t-1 0.212*** 0.273***
(0.021) (0.023)
Constant -21.84*** -13.93*** -14.04*** -14.78*** -13.11*** -9.896***
(1.35) (0.71) (0.94) (1.81) (1.13) (1.52)
Observations 22612 22612 22612 22612 22612 22612
Number of country_sic2 1198 1198 1198 1198 1198 1198
Notes: This table reports the estimated results of equation 11. The estimated coecients are elasticities of
dierent margins in exports with respect to each regressor. Robust standard errors clustered by country are
in parentheses. * indicates signicance level: *** p<0.01, ** p<0.05, * p<0.1.
48
Table 6: Two-way xed eects with interaction between income and patent
Dependent Variables ! ln export ln extensive ln intensive
Regressors # (1) (2) (3)
ln patent 0.663*** 0.238*** 0.426***
(0.11) (0.044) (0.072)
ln population 1.011*** 0.524*** 0.487***
(0.092) (0.043) (0.055)
ln
GDP
population
1.524*** 0.902*** 0.621***
(0.17) (0.080) (0.100)
ln patent ln
^GDP
population
-0.178* -0.110*** -0.0680
(0.091) (0.035) (0.063)
ln 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
probF 0.00 0.00 0.00
1
2
0.62 0.51 0.53
Notes: This table reports the estimated results of equation 12. The estimated coecients are elasticities of
dierent margins in exports with respect to each regressor. Robust standard errors clustered by country are
in parentheses. Coecients of dummy variables are skipped. * indicates signicance level: *** p<0.01, **
p<0.05, * p<0.1. The prob>F shows the p-value of testing
1
=
7
= 0 in equation 12.
49
Table 7: Elasticities of exports with respect to patents in countries with dierent incomes
Elasticities with respective to innovation
(1) (2) (3)
Country GDP per capita ratio export extensive intensive
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%
Notes: GDP per capita ratio is (
^GDP
pop
) =
GDP
pop
=
GDP
pop
: Numbers in column (1)-(3) are estimated elasticities of
each margin with respect to patent, which equals
1
+
7
ln((
^GDP
pop
)) in equation 12. Percentages describe the
contribution of each margin to overall exports.
50
Table 8: dynamic GMM estimates with interaction between income and patent
Exogenous Innovation Endogenous Innovation
Dependent Variables ! ln export ln extensive ln intensive ln export ln extensive ln intensive
Regressors # (1) (2) (3) (4) (5) (6)
ln patent 0.282*** 0.131*** 0.227*** 0.376*** 0.132*** 0.371***
(0.024) (0.012) (0.020) (0.047) (0.036) (0.045)
ln patent ln
^GDP
population
-0.0170 -0.0360*** 0.0148 -0.148*** -0.0814*** -0.119***
(0.020) (0.011) (0.019) (0.034) (0.020) (0.039)
ln population 0.680*** 0.456*** 0.417*** 0.474*** 0.443*** 0.285***
(0.042) (0.022) (0.028) (0.046) (0.031) (0.037)
ln
GDP
population
1.020*** 0.784*** 0.536*** 0.738*** 0.794*** 0.366***
(0.067) (0.040) (0.050) (0.077) (0.060) (0.066)
ln distance -0.319*** -0.264*** -0.157** -0.239*** -0.264*** -0.119*
(0.091) (0.053) (0.077) (0.075) (0.052) (0.069)
language 0.187* 0.158*** 0.0969 0.0825 0.158*** 0.0119
(0.098) (0.060) (0.086) (0.083) (0.059) (0.080)
border 0.715*** -0.139 0.987*** 0.552*** -0.144 0.848***
(0.25) (0.15) (0.24) (0.19) (0.14) (0.20)
ln export t-1 0.355*** 0.488***
(0.026) (0.025)
ln extensive t-1 0.123*** 0.166***
(0.023) (0.022)
ln intensive t-1 0.213*** 0.309***
(0.021) (0.023)
Constant -21.85*** -13.94*** -14.00*** -15.78*** -13.70*** -10.25***
(1.36) (0.72) (0.94) (1.36) (0.94) (1.08)
Observations 22612 22612 22612 22612 22612 22612
prob
2
0.00 0.00 0.00 0.00 0.00 0.00
Number of country-sic2 1198 1198 1198 1198 1198 1198
Notes: This table reports the estimated results of equation 11, with the interaction between patent and GDP
per capita included. The estimated coecients are elasticities of dierent margins in exports with respect to
each regressor. Robust standard errors clustered by country are in parentheses. * indicates signicance level:
*** p<0.01, ** p<0.05, * p<0.1. The prob>
2
shows the p-value of the joint test that coecient of patent
and interaction term are both 0.
51
T
a
b
l
e
9
:
E
s
t
i
m
a
t
e
s
b
y
i
n
d
u
s
t
r
y
I
n
d
u
s
t
r
y
a
n
d
c
a
t
e
g
o
r
y
e
x
p
o
r
t
e
x
t
e
n
s
i
v
e
i
n
t
e
n
s
i
v
e
O
b
s
e
r
v
a
t
i
o
n
s
(
1
)
(
2
)
(
3
)
S
u
p
p
l
i
e
r
-
o
r
i
e
n
t
e
d
2
0
F
o
o
d
a
n
d
k
i
n
d
r
e
d
p
r
o
d
u
c
t
s
0
.
0
7
8
6
-
0
.
0
0
5
1
9
-
7
%
0
.
0
8
3
8
1
0
7
%
2
3
8
8
2
2
T
e
x
t
i
l
e
m
i
l
l
p
r
o
d
u
c
t
s
0
.
1
0
9
0
.
0
1
7
2
1
6
%
0
.
0
9
1
7
8
4
%
2
2
1
0
P
r
o
d
u
c
t
i
o
n
-
i
n
t
e
n
s
i
v
e
2
9
P
e
t
r
o
l
e
u
m
a
n
d
c
o
a
l
p
r
o
d
u
c
t
s
0
.
2
3
8
0
.
1
5
6
*
6
6
%
0
.
0
8
2
1
3
4
%
1
5
8
7
3
0
R
u
b
b
e
r
a
n
d
m
i
s
c
e
l
l
a
n
e
o
u
s
p
l
a
s
t
i
c
s
p
r
o
d
u
c
t
s
0
.
5
1
4
*
*
*
0
.
0
8
4
9
*
1
7
%
0
.
4
2
9
*
*
*
8
3
%
1
9
9
5
3
2
S
t
o
n
e
,
c
l
a
y
,
g
l
a
s
s
,
a
n
d
c
o
n
c
r
e
t
e
p
r
o
d
u
c
t
s
0
.
2
6
1
*
*
0
.
0
1
4
1
5
%
0
.
2
4
7
*
*
*
9
5
%
2
0
7
3
3
3
P
r
i
m
a
r
y
m
e
t
a
l
i
n
d
u
s
t
r
i
e
s
0
.
3
3
9
*
*
*
0
.
1
8
6
*
*
*
5
5
%
0
.
1
5
3
*
*
4
5
%
2
0
2
9
3
4
F
a
b
r
i
c
a
t
e
d
m
e
t
a
l
p
r
o
d
u
c
t
s
0
.
6
2
3
*
*
*
0
.
2
1
2
*
*
*
3
4
%
0
.
4
1
1
*
*
*
6
6
%
2
2
4
7
3
5
I
n
d
u
s
t
r
i
a
l
m
a
c
h
i
n
e
r
y
a
n
d
e
q
u
i
p
m
e
n
t
0
.
7
4
0
*
*
*
0
.
0
9
3
4
*
*
1
3
%
0
.
6
4
7
*
*
*
8
7
%
2
2
7
0
3
7
T
r
a
n
s
p
o
r
t
a
t
i
o
n
e
q
u
i
p
m
e
n
t
0
.
7
5
6
*
*
*
0
.
1
3
5
*
*
*
1
8
%
0
.
6
2
0
*
*
*
8
2
%
1
9
4
9
3
8
I
n
s
t
r
u
m
e
n
t
s
a
n
d
r
e
l
a
t
e
d
p
r
o
d
u
c
t
s
0
.
6
3
4
*
*
*
0
.
2
3
9
*
*
*
3
8
%
0
.
3
9
5
*
*
*
6
2
%
1
9
9
0
S
c
i
e
n
c
e
-
b
a
s
e
d
2
8
C
h
e
m
i
c
a
l
s
a
n
d
a
l
l
i
e
d
p
r
o
d
u
c
t
s
0
.
4
3
1
*
*
*
0
.
2
2
7
*
*
*
5
3
%
0
.
2
0
4
*
*
*
4
7
%
2
2
5
9
3
6
E
l
e
c
t
r
i
c
a
l
a
n
d
e
l
e
c
t
r
o
n
i
c
e
q
u
i
p
m
e
n
t
0
.
6
2
1
*
*
*
0
.
2
3
0
*
*
*
3
7
%
0
.
3
9
1
*
*
*
6
3
%
2
2
7
1
N
o
t
e
s
:
T
h
i
s
t
a
b
l
e
r
e
p
o
r
t
s
t
h
e
e
s
t
i
m
a
t
e
d
r
e
s
u
l
t
s
o
f
t
h
e
i
n
d
u
s
t
r
y
-
b
y
-
i
n
d
u
s
t
r
y
e
s
t
i
m
a
t
i
o
n
o
f
e
q
u
a
t
i
o
n
9
.
T
h
e
e
s
t
i
m
a
t
e
d
c
o
e

c
i
e
n
t
s
a
r
e
e
l
a
s
t
i
c
i
t
i
e
s
o
f
d
i

e
r
e
n
t
m
a
r
g
i
n
s
i
n
e
x
p
o
r
t
s
w
i
t
h
r
e
s
p
e
c
t
p
a
t
e
n
t
s
.
E
s
t
i
m
a
t
e
s
o
f
o
t
h
e
r
r
e
g
r
e
s
s
o
r
s
a
r
e
s
k
i
p
p
e
d
.
Y
e
a
r
d
u
m
m
y
v
a
r
i
a
b
l
e
s
a
r
e
i
n
c
l
u
d
e
d
.
P
e
r
c
e
n
t
a
g
e
s
d
e
s
c
r
i
b
e
t
h
e
c
o
n
t
r
i
b
u
t
i
o
n
o
f
e
a
c
h
m
a
r
g
i
n
t
o
o
v
e
r
a
l
l
e
x
p
o
r
t
s
.
*
i
n
d
i
c
a
t
e
s
s
i
g
n
i

c
a
n
c
e
l
e
v
e
l
:
*
*
*
p
<
0
.
0
1
,
*
*
p
<
0
.
0
5
,
*
p
<
0
.
1
.
52
Table 10: Estimates by-year: patents classied according to the "application" year
Dependent Variables ! ln export ln extensive ln intensive
Year Observations
1975 0.698*** 0.223*** 0.475*** 886
1976 0.650*** 0.224*** 0.426*** 896
1977 0.635*** 0.303*** 0.331*** 839
1978 0.644*** 0.208*** 0.437*** 920
1979 0.608*** 0.209*** 0.399*** 930
1980 0.614*** 0.204*** 0.410*** 949
1981 0.632*** 0.207*** 0.425*** 954
1982 0.569*** 0.184*** 0.385*** 950
1983 0.520*** 0.149*** 0.371*** 983
1984 0.525*** 0.149*** 0.376*** 1028
1985 0.516*** 0.166*** 0.350*** 1083
1986 0.491*** 0.113*** 0.378*** 1062
1987 0.450*** 0.117*** 0.333*** 1084
1988 0.481*** 0.117*** 0.365*** 1057
1989 0.490*** 0.134*** 0.355*** 1045
1990 0.528*** 0.159*** 0.369*** 1020
1991 0.543*** 0.161*** 0.382*** 1030
1992 0.507*** 0.118** 0.389*** 1037
1993 0.512*** 0.143*** 0.369*** 1058
1994 0.544*** 0.183*** 0.361*** 1069
1995 0.552*** 0.151*** 0.401*** 1060
1996 0.512*** 0.135*** 0.377*** 1081
1997 0.515*** 0.132** 0.383*** 1075
1998 0.502*** 0.108** 0.394*** 1085
1999 0.522*** 0.0835 0.439*** 1087
2000 0.596*** 0.108** 0.487*** 1101
2001 0.617*** 0.0913 0.525*** 1081
export extensive intensive
1975-79 (average) 0.647 0.233 36% 0.414 64%
1980-89 (avarage) 0.529 0.154 29% 0.375 71%
1990-99 (average) 0.524 0.137 26% 0.386 74%
Notes: The rst panel of his table reports the estimated results of the year-by-year estimation of equation 9.
The estimated coecients are elasticities of dierent margins in exports with respect patents. Estimates of
other regressors are skipped. Industry dummy variables are included. The second panel shows the average
elasticities in the 70s, 80s, and 90s. Percentages describe the contribution of each margin to overall exports.
* indicates signicance level: *** p<0.01, ** p<0.05, * p<0.1.
53
Table 11: Estimates by-year: patents classied according to the "grant" year
Dependent Variables ! ln export ln extensive ln intensive
Year Observations
1975 0.651*** 0.205*** 0.445*** 886
1976 0.634*** 0.196*** 0.438*** 896
1977 0.672*** 0.323*** 0.350*** 839
1978 0.600*** 0.198*** 0.402*** 920
1979 0.584*** 0.198*** 0.386*** 930
1980 0.565*** 0.176*** 0.389*** 949
1981 0.656*** 0.215*** 0.441*** 954
1982 0.549*** 0.167*** 0.382*** 950
1983 0.490*** 0.135*** 0.355*** 983
1984 0.513*** 0.141*** 0.372*** 1028
1985 0.482*** 0.150*** 0.332*** 1083
1986 0.434*** 0.0876** 0.346*** 1062
1987 0.467*** 0.127*** 0.341*** 1084
1988 0.487*** 0.107*** 0.380*** 1057
1989 0.435*** 0.128*** 0.307*** 1045
1990 0.498*** 0.131*** 0.367*** 1020
1991 0.523*** 0.146*** 0.377*** 1030
1992 0.491*** 0.123*** 0.368*** 1037
1993 0.507*** 0.134*** 0.373*** 1058
1994 0.522*** 0.175*** 0.347*** 1069
1995 0.520*** 0.144*** 0.376*** 1060
1996 0.458*** 0.117** 0.341*** 1081
1997 0.502*** 0.128** 0.374*** 1075
1998 0.416*** 0.0960** 0.320*** 1085
1999 0.457*** 0.0677 0.389*** 1087
2000 0.517*** 0.103** 0.414*** 1101
2001 0.528*** 0.104* 0.424*** 1081
export extensive intensive
1975-79 (average) 0.628 0.224 36% 0.404 64%
1980-89 (avarage) 0.508 0.143 28% 0.365 72%
1990-99 (average) 0.489 0.126 26% 0.363 74%
Notes: The rst panel of his table reports the estimated results of the year-by-year estimation of equation 9,
with patents counts based on the grant years. The estimated coecients are elasticities of dierent margins
in exports with respect patents. Estimates of other regressors are skipped. Industry dummy variables are
included. The second panel shows the average elasticities in the 70s, 80s, and 90s. Percentages describe the
contribution of each margin to overall exports. * indicates signicance level: *** p<0.01, ** p<0.05, * p<0.1.
54
Table 12: The extensive margin at dierent degrees of aggregation
Elasticity of Extensive Margin
Degree of Aggregation in
1975-1988/1989-1999 ln patent ln pop ln
GDP
population
1
2
7 digit TS/10 digit HS 0.160*** 30% 0.546*** 52% 0.891*** 59% 0.51
6 digit TS/ 6 digit HS 0.138*** 26% 0.497*** 48% 0.821*** 55% 0.50
5 digit TS/ 5 digit HS 0.111*** 21% 0.469*** 45% 0.788*** 52% 0.48
4 digit TS/ 4digit HS 0.0822** 15% 0.419*** 40% 0.713*** 47% 0.47
3 digit TS/ 3 digit HS 0.0134 2.5% 0.305*** 29% 0.529*** 35% 0.39
Notes: This table reports the estimated elasticities of extensive margins, under dierent level of aggregation in
export data. Percentages describe the contribution of extensive margin to overall exports. Estimates of other
regressors are skipped. * indicates signicance level: *** p<0.01, ** p<0.05, * p<0.1.
55
Table 13: Use lag patents as the proxy of innovation
Dependent Variables ! ln export ln extensive ln intensive ln export ln extensive ln intensive
Regressors # (1) (2) (3) (4) (5) (6)
ln patent 0.517*** 0.153*** 0.363*** 0.562*** 0.188*** 0.374***
(0.074) (0.039) (0.043) (0.062) (0.025) (0.046)
ln population 1.081*** 0.555*** 0.526*** 1.147* 0.584** 0.563
(0.088) (0.040) (0.053) (0.62) (0.29) (0.40)
ln
GDP
population
1.535*** 0.898*** 0.637*** 0.659** 0.399*** 0.260
(0.17) (0.080) (0.099) (0.31) (0.14) (0.23)
ln distance -0.486* -0.342*** -0.145
(0.28) (0.13) (0.17)
language 0.318 0.192 0.126
(0.28) (0.13) (0.17)
border 0.255 -0.489 0.744
(1.01) (0.57) (0.46)
ln duty-freight rate -0.779*** -0.246*** -0.533*** -0.736*** -0.224*** -0.512***
(0.067) (0.031) (0.051) (0.048) (0.022) (0.043)
Constant -34.55*** -16.48*** -18.07*** -35.63*** -17.79*** -17.84**
(3.32) (1.46) (2.04) (11.3) (5.30) (7.61)
Observations 25148 25148 25148 25148 25148 25148
1
2
0.64 0.51 0.55 0.74 0.61 0.62
Notes: This table reports the estimated elasticities with the inclusion of ln duty-freight rates.
56
Table 14: Use lag patents as the proxy of innovation
Dependent Variables ! ln export ln extensive ln intensive ln export ln extensive ln intensive
Regressors # (1) (2) (3) (4) (5) (6)
ln patent t-1 0.508*** 0.146*** 0.362***
(0.070) (0.036) (0.042)
ln patent t-2 0.486*** 0.134*** 0.352***
(0.070) (0.036) (0.042)
ln population 0.992*** 0.510*** 0.482*** 1.013*** 0.525*** 0.488***
(0.088) (0.039) (0.056) (0.088) (0.039) (0.056)
ln
GDP
population
1.434*** 0.847*** 0.587*** 1.473*** 0.873*** 0.600***
(0.16) (0.073) (0.11) (0.17) (0.074) (0.11)
ln distance -0.466 -0.288** -0.179 -0.471 -0.301** -0.170
(0.28) (0.12) (0.17) (0.29) (0.13) (0.18)
language 0.262 0.172 0.0892 0.267 0.180 0.0862
(0.28) (0.13) (0.18) (0.29) (0.13) (0.18)
border 1.092 -0.149 1.241*** 1.079 -0.186 1.265***
(0.90) (0.50) (0.43) (0.90) (0.50) (0.43)
Constant -30.99*** -15.49*** -15.50*** -31.70*** -15.67*** -16.02***
(3.22) (1.35) (2.06) (3.26) (1.36) (2.09)
Observations 22612 22612 22612 21569 21569 21569
1
2
0.61 0.51 0.53 0.609 0.517 0.527
Notes: This table reports the estimated results when current patents is replaced with lag patents.
57
Table 15: Elasticities of export with respect to patent: using dierent weights of citations
Dependent Variables ! ln export ln extensive ln intensive
weight of citation # (1) (2) (3)
0 0.686*** 0.183*** 0.503***
(0.106) (0.0574) (0.0572)
10% 0.651*** 0.179*** 0.472***
(0.0977) (0.0532) (0.0534)
30% 0.607*** 0.172*** 0.434***
(0.0884) (0.0479) (0.0491)
50% 0.579*** 0.167*** 0.411***
(0.0830) (0.0448) (0.0466)
100% 0.537*** 0.160*** 0.378***
(0.0753) (0.0401) (0.0430)
Observations 25268 25268 25268
Notes: This table reports elasticities of export with respect to patents using dierent weight of citations.
58
Table 16: Linear model
Dependent Variables ! export extensive intensive
Regressors # (1) (2) (3)
patent 0.470*** 1.651*** 0.504***
(0.011) (0.56) (0.013)
population 0.0710*** 1.762*** 0.0862***
(0.023) (0.29) (0.026)
GDP
population
0.00185* 0.120*** 0.00116
(0.00098) (0.028) (0.0012)
distance 0.000000119 0.000000102 0.000000390
(0.00000023) (0.0000042) (0.00000030)
language 0.00261 -0.0149 0.00421
(0.0026) (0.038) (0.0028)
border 0.0882*** 0.467*** 0.0948***
(0.031) (0.042) (0.028)
Constant 0.0000800 0.0326 0.0209***
(0.0024) (0.037) (0.0037)
Observations 25268 25268 25268
1
2
0.600 0.453 0.408
Notes: This table reports the estimated results of a linear model instead of a log linear model. All dependent
and independent variables are in level form without taking natural log. Patent, population and GDP per
capital are made relative to the world level in each year.
59
T
a
b
l
e
1
7
:
D
e
c
o
m
p
o
s
i
n
g
i
n
t
e
n
s
i
v
e
m
a
r
g
i
n
s
i
n
t
o
p
r
i
c
e
a
n
d
q
u
a
n
t
i
t
y
D
e
p
e
n
d
e
n
t
V
a
r
i
a
b
l
e
s
!
l
n
e
x
p
o
r
t
l
n
e
x
t
e
n
s
i
v
e
l
n
i
n
t
e
n
s
i
v
e
l
n
p
r
i
c
e
l
n
q
u
a
n
t
i
t
y
R
e
g
r
e
s
s
o
r
s
#
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
l
n
p
a
t
e
n
t
0
.
5
1
9
*
*
*
0
.
2
1
9
*
*
*
0
.
3
0
0
*
*
*
0
.
0
4
8
2
*
*
*
0
.
2
5
2
*
*
*
(
0
.
0
7
0
)
(
0
.
0
3
9
)
(
0
.
0
4
1
)
(
0
.
0
1
6
)
(
0
.
0
5
1
)
l
n
p
o
p
u
l
a
t
i
o
n
0
.
9
8
8
*
*
*
0
.
5
1
5
*
*
*
0
.
4
7
3
*
*
*
-
0
.
0
3
3
4
*
*
0
.
5
0
6
*
*
*
(
0
.
0
8
5
)
(
0
.
0
4
2
)
(
0
.
0
5
0
)
(
0
.
0
1
3
)
(
0
.
0
5
6
)
l
n
G
D
P
p
o
p
u
l
a
t
i
o
n
1
.
3
8
4
*
*
*
0
.
8
1
1
*
*
*
0
.
5
7
2
*
*
*
0
.
1
2
1
*
*
*
0
.
4
5
1
*
*
*
(
0
.
1
6
)
(
0
.
0
8
3
)
(
0
.
0
9
8
)
(
0
.
0
2
4
)
(
0
.
1
1
)
l
n
d
i
s
t
a
n
c
e
-
0
.
3
1
5
-
0
.
1
7
1
-
0
.
1
4
4
0
.
0
2
3
5
-
0
.
1
6
8
(
0
.
2
7
)
(
0
.
1
3
)
(
0
.
1
6
)
(
0
.
0
3
3
)
(
0
.
1
7
)
l
a
n
g
u
a
g
e
0
.
2
3
5
0
.
1
4
3
0
.
0
9
1
9
0
.
0
4
1
5
0
.
0
5
0
4
(
0
.
2
7
)
(
0
.
1
4
)
(
0
.
1
6
)
(
0
.
0
5
1
)
(
0
.
1
8
)
b
o
r
d
e
r
1
.
3
6
3
*
0
.
2
2
9
1
.
1
3
4
*
*
*
-
0
.
2
3
2
*
*
*
1
.
3
6
6
*
*
*
(
0
.
7
8
)
(
0
.
4
8
)
(
0
.
3
4
)
(
0
.
0
7
3
)
(
0
.
3
2
)
C
o
n
s
t
a
n
t
-
3
1
.
8
7
*
*
*
-
1
6
.
2
6
*
*
*
-
1
5
.
6
1
*
*
*
-
0
.
7
3
1
*
*
-
1
4
.
8
8
*
*
*
(
3
.
1
1
)
(
1
.
5
3
)
(
1
.
8
2
)
(
0
.
3
3
)
(
1
.
9
7
)
I
n
d
u
s
t
r
y
d
u
m
m
y
Y
e
s
Y
e
s
Y
e
s
Y
e
s
Y
e
s
Y
e
a
r
d
u
m
m
y
Y
e
s
Y
e
s
Y
e
s
Y
e
s
Y
e
s
C
o
u
n
t
r
y
D
u
m
m
y
N
o
N
o
N
o
N
o
N
o
O
b
s
e
r
v
a
t
i
o
n
s
2
2
8
0
4
2
2
8
0
4
2
2
8
0
4
2
2
8
0
4
2
2
8
0
4
1
2
0
.
5
8
0
.
4
9
0
.
4
4
0
.
1
8
0
.
3
9
N
o
t
e
s
:
T
h
i
s
t
a
b
l
e
r
e
p
o
r
t
s
t
h
e
e
s
t
i
m
a
t
e
d
r
e
s
u
l
t
s
o
f
e
q
u
a
t
i
o
n
9
,
w
i
t
h
f
u
r
t
h
e
r
d
e
c
o
m
p
o
s
i
t
i
o
n
o
f
t
h
e
i
n
t
e
n
s
i
v
e
m
a
r
g
i
n
i
n
t
o
p
r
i
c
e
a
n
d
i
n
d
e
x
.
T
h
e
e
s
t
i
m
a
t
e
d
c
o
e

c
i
e
n
t
s
a
r
e
e
l
a
s
t
i
c
i
t
i
e
s
o
f
d
i

e
r
e
n
t
m
a
r
g
i
n
s
i
n
e
x
p
o
r
t
s
w
i
t
h
r
e
s
p
e
c
t
t
o
e
a
c
h
r
e
g
r
e
s
s
o
r
.
R
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
c
o
u
n
t
r
y
a
r
e
i
n
p
a
r
e
n
t
h
e
s
e
s
.
C
o
e

c
i
e
n
t
s
o
f
d
u
m
m
y
v
a
r
i
a
b
l
e
s
a
r
e
s
k
i
p
p
e
d
.
*
i
n
d
i
c
a
t
e
s
s
i
g
n
i

c
a
n
c
e
l
e
v
e
l
:
*
*
*
p
<
0
.
0
1
,
*
*
p
<
0
.
0
5
,
*
p
<
0
.
1
.
60