Research Policy 39 (2010) 1264–1276

Contents lists available at ScienceDirect

Research Policy
journal homepage: www.elsevier.com/locate/respol

The innovation–economic growth nexus: Global evidence

Iftekhar Hasan a,b,∗ , Christopher L. Tucci c
a
Rensselaer Polytechnic Institute, Lally School of Management & Technology, Troy, NY 12180, United States
b
Bank of Finland, Helsinki, Finland
c
EPFL College of Management of Technology, Ecole Polytechnique Fédérale de Lausanne, Station 5, CH-1015, Lausanne, Switzerland

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

Article history: This paper extends the line of research attempting to link innovation to economic growth by addressing
Received 29 July 2008 some unexplored questions. Using global patent data, this paper empirically investigates the importance
Received in revised form of both the quantity and quality of innovation on economic growth, controlling for past measures of
12 November 2009
inventive inputs. Moreover, our research examines how innovation inputs can be translated into per
Accepted 12 July 2010
capita growth under the various economic structures and stages of economic development. Based on a
Available online 16 August 2010
sample of 58 countries for the period 1980–2003, our empirical results indicate that countries hosting
firms with higher quality patents also have higher economic growth. Furthermore, we have some evi-
Keywords:
Innovation
dence that those countries that increase the level of patenting also witness a concomitant increase in
Economic growth economic growth.
Patents © 2010 Elsevier B.V. All rights reserved.

1. Introduction (Cameron, 1998). In this paper, we aim to extend this line of

research by arguing that not only the quantity but also the quality of
Economic growth, especially its long-run sustainability, has long innovation matters in promoting economic growth. Furthermore,
been a focal point of academic researchers and policy makers. we are interested in investigating whether the effects of innovation
Numerous attempts have been made to provide a long list of fac- on economic growth largely depend on the economic structure and
tors that may have an impact on economic growth. Pioneering work stage of development in different countries.
on endogenous growth by Romer (1986), Lucas (1988) and others In the models of Romer (1986, 1990) and Stokey (1995), among
emphasizes the role of knowledge as an input to production. In others, industrial innovation activities are an important deter-
their models, it is the technological advancement and industrial minant of economic growth due to their direct impact on the
innovation that drive long-run growth (Grossman and Helpman, production process and also due to positive externalities. Schol-
1994). ars have also argued that “national innovation systems” – which
In addition, since at least the time of Schumpeter (1932), the include aspects of how intellectual property is protected and how
process of industrial innovation has been seen as important to the research and development (R&D) is funded – is a major contributor
economy. When Schumpeter wrote about innovation, he clearly to innovation activities (Nelson, 1993; Freeman and Soete, 1997).
intended to emphasis not only the “destruction” aspect of creative Though technological change forms the engine of long-run
destruction, but the “creative” part as well (Freeman and Soete, growth, accumulation of other types of capital will still play an inde-
1997). In other words, Schumpeter was an early thinker on the rela- pendent role during a transitional phase (Grossman and Helpman,
tionship between industrial innovation and economic growth at a 1994). This notion implies that how innovation activities can be
more macro level. translated into different rates of growth is closely linked to the vari-
Since then, in the field of economics, research on endogenous ation of economic structures and policies (Grossman and Helpman,
growth theory sparked many empirical studies exploring how and 1994). In this paper, we make use of global patent databases to
to what extent innovation might contribute to economic growth. construct measures of innovation and empirically examine the pro-
Empirical evidence taken as a whole points to the fact that inno- posed research questions. We use a panel of data to investigate the
vation tends to make significant contributions to growth, and potential relationship between measures of the innovations, both
there are also significant spillover effects of innovative activities quantity and quality, and economic growth. Based on a sample of
58 countries for the period 1980–2003, our empirical results indi-
cate that those countries that increase the level of patenting also
∗ Corresponding author. Tel.: +1 518 276 2525. witness a concomitant increase in economic growth. Furthermore,
E-mail addresses: [email protected] (I. Hasan), christopher.tucci@epfl.ch we provide some evidence that countries hosting firms with higher
(C.L. Tucci). quality patents also have higher economic growth.

0048-7333/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.respol.2010.07.005
 I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276 1265

The rest of this paper is organized as follows. Section 2 pro- studies find a positive link between innovation and some measures
vides a brief review of related literature. Section 3 explains our of output (Mansfield, 1980; Griliches and Mairesse, 1986).
data collection and discusses some methodology issues. Section 4 Meanwhile, many other studies attempt to investigate the
presents our empirical results. In Section 5, we summarize, discuss spillover effects of innovation. For example, Coe and Helpman
and conclude. (1995) and Bayoumi et al. (1999) have documented that interna-
tional trade can greatly raise a country’s total factor productivity.
There are many reasons for this, but one factor could very well
2. Literature review be knowledge transfer due to international trade. However, there
is a limitation for such spillovers across countries. Audretsch and
Questions about the sources of economic growth have fasci- Feldman (1996) find that innovation spillovers tend to be local-
nated economists for many years. Neoclassical growth models posit ized in the sense that industries with a prevalence of knowledge
that the rate of return on investment is a decreasing function of spillovers have a high propensity to be clustered. For example, there
per capita capital stock, and per capita outcomes across differ- may be important barriers to knowledge flow even between Euro-
ent countries should converge to a steady state in the absence pean countries (Maurseth and Verspagen, 2002; Bottazzi and Peri,
of exogenous technological change. However, these predictions 2003).
are somewhat inconsistent with observations from the real world. The above leads to the following two hypotheses:
Without a doubt, technological advancement has become a major
H1. The higher the level of innovative activity, the higher the rate
factor behind economic growth by, among other factors, providing
of economic growth in an economy.
new means to combine raw materials. It is unrealistic to attribute
all the unexplained part of economic growth to exogenous techno- H2. The higher the quality of innovative activity, the higher the
logical shocks. rate of economic growth in an economy.
Several studies attempt to incorporate industrial innovation
into models to explain economic growth. Romer (1986) showed
3. Methods
that knowledge with increasing marginal productivity could be an
input in explaining long-run growth. In a competitive economic
How to precisely measure innovation is an important issue in the
environment, intentional investments in innovation activities are
empirical endeavor to explore the effect of innovation on growth. In
motivated by market incentives (Romer, 1986; Aghion and Howitt,
the existing literature, research and development (R&D) expendi-
1992; Stokey, 1995). Treating technological changes as endoge-
tures are widely used as a proxy for innovation partly because of the
nous, Romer (1990) presented a model of the growth rate being
availability and reliability of data (Griliches, 1980; Mansfield, 1980;
determined by the stock of human capital, even though new tech-
Audretsch and Feldman, 1996; Goel and Ram, 1994). Many other
nology is assumed to be no better than old (horizontal product
researchers intend to use alternative measures of innovation such
innovation). In contrast, Aghion and Howitt (1992) developed a
as patenting activities. Despite several major problems associated
model in which vertical innovations make existing products obso-
with using patent data (see below), there are at least three rea-
lete, becoming the underlying source of growth through a process
sons why patent statistics can be an important economic indicator
similar to creative destruction in which demand increases for the
of innovation (Griliches, 1990) and thus the fascination of aca-
superior product, more than compensating at the macro level for
demic economists with such statistics. First, patenting databases
the reduction in competitiveness of the product based on the old
are generally more available and richer in the sense that patents
technology.
are public documents with detailed information on the background
The innovation process has its own externalities. The accu-
of assignees and their activities. Second, patents can be viewed as
mulation of technological advancement enlarges the knowledge
the output of an inventive process and link together R&D activities
base and makes sequential innovations available (Stokey, 1995).
and productivity. Finally, using patent data, one is able to construct
Knowledge flows and technological spillovers across economic
both quantitative and qualitative measures of innovation activities,
agents benefit all firms including rival firms as well (Griliches,
and researchers may under certain circumstances trace the citation
1992). Even when technological spillovers do not exist, an agent
records to better gauge the spillover effects of technology changes.1
does not appropriate all the social gains from her innovation
Of course, there are also problems associated with using patents as
unless she can price-discriminate. In addition to the efforts made
innovation indicators; see Section 5 for more detail on these issues.
by profit-maximizing firms, academic research funded by public
To our best knowledge, there are very few comprehensive
resources in universities and other institutions provides substan-
studies examining innovation and economic growth in a cross-
tial inputs and spillovers into the innovation process (Fagerberg,
country setting using patent data (Maurseth and Verspagen, 2002;
1994).
Bottazzi and Peri, 2003). In this paper, we try to address this need
Innovation activities do not only directly influence economy-
by constructing a cross-country sample, and address some unan-
wide productivity, but also promote economic growth through
swered questions with the help of a recently available global patent
spurring new business formation, which further promotes employ-
database. In particular, we attempt to add knowledge to the exist-
ment growth and other outputs (Kirchhoff, 1994; Wennekers,
ing literature by focusing on the effects of both quantity and quality
1999). Innovation encourages and facilitates entrepreneurs to
of innovation on economic growth. Additionally, we explore the
create new organizations in order to enter certain indus-
impact of innovation at different stages of development for differ-
tries characterized by an entrepreneurial technological regime
ent countries.
(Audretsch, 1995). This indirect mechanism has been supported
Empirical research using cross-country data has provided much
by empirical evidence (Francis et al., 2007; Kirchhoff et al.,
insight on the role of institutions in promoting economic growth
2007).
Summarizing the above, innovation can be considered impor-
tant for potential economic growth. So what evidence do we have
1
that it is linked to growth, and at what levels of analysis? Vari- Note: the structure of the US national innovation system in which the applicants
supply many prior art references leads to this interpretation. However, it should
ous studies have been conducted at the level of individual firms, be noted that in the European Patent Office, the examiners provide the prior art
industries, as well as countries. Cameron (1998) surveys the exist- citations and thus any knowledge flow interpretations are unwarranted (Harhoff et
ing literature on this topic and concludes that the majority of these al., 2008).
1266 I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276

Table 1
Summary statistics.

Variable N Mean SD Min Max

Annual growth rate in per capita real GDP – GDPPCGR 1162 0.02031 0.04753 −0.19730 0.26550
Gross private capital formation to GDP – GCAPFORM 1162 0.23426 0.09559 0.03450 0.68000
Government consumption to GDP – GCONGDP 1162 0.19010 0.07523 0.06530 0.54450
Exports to GDP – EXPGDP 1162 0.39678 0.27090 0.05060 1.70000
Literacy rate for labor force – LITRATE 1162 0.89633 0.14516 0.44170 0.99730
Foreign direct investment to GDP – FDIGDP 1162 0.02982 0.05718 0.00000 0.79340
Technology index – TECHINDX 1162 0.02906 0.00453 0.01170 0.03570
Research and development (R&D) expenditure to GDP – RNDGDP 1162 1.27869 0.93115 0.01550 5.08010
Total number of patents granted to R&D expenses – TPATR 1162 0.00870 0.04774 7.74863E−9 0.87655
Proportion of patents granted in the USA 1162 0.11392 0.47591 0.0002528 8.50000
Residual of the estimation of TPATR on RNDGDP – MAGIC 1162 0.01679 0.01309 −0.02760 0.09810

The table presents summary statistics of the variables used in this paper. N refers to country-year observations for 58 countries during the sample period 1980–2004.
GDPPCGR: annual growth rate in per capita real GDP is the change in the natural logarithm of real per capita GDP; GCAPFORM: gross private capital formation of a given
sample year; GCONGDP: government consumption to GDP is the expenses in a given year by the government to GDP; EXPGDP: exports to GDP are defined as the ratio of total
exports to GDP; LITRATE: literacy rate of the labor force (secondary education); FDIGDP: foreign direct investment to GDP is the total FDI incoming to the country in a given
year to the GDP; TECHINDX: technology index is an index developed from a number of variables: electric power consumption (kWh per capita), fixed line and mobile phone
subscribers (per 1000 people), personal computers (per 1000 people), radios and telephone sets (per 1000 people); RNDGDP: total R&D to GDP is the total research and
development expenditure at the country level to GDP; TPATR: total number of patent granted as a proportion of million dollars of research and development expenditure;
USAR: ratio of patents granted in the USA as a proportion of total patents granted by the country in a given year; MAGIC is the residual from the regression estimate of the
variable TPATR on RNDGDP.

(King and Levine, 1993; Knack and Keefer, 1995; Rousseau and of GDP is highly correlated with capital formation and moderately
Wachtel, 2000; Wachtel, 2001). However, as mentioned above, very correlated with FDI, RNDGDP, and TPATR (defined below). Interest-
few cross-country studies are associated with any potential asso- ingly, the correlation of the level of GDP is not highly associated
ciation between direct measures of innovation and growth. with the Technology Index (also defined below).

3.1. Data sources 3.2.1. Dependent variables

Our dependent variable, growth, is the growth rate of real annual
The data set was constructed from a number of sources. We per capita GDP in the country. The variable is operationalized as
started with the US Patent database available from the National the change in the log of real per capita GDP. Real per capita GDP is
Bureau of Economic Research and the PATSTAT database encom- defined as per capita GDP deflated to the base year of 1980.
passing international patents, and examined the country of both the
inventor and the assignee. Work by Jaffe et al. (1993) and Jaffe and 3.2.2. Independent variables
Trajtenberg (1999) suggest that patent data can be considered one A major challenge in this paper was to determine what an
measure of innovation. Unlike an exogenous technological shock, appropriate proxy was for innovation. In the past, scholars have
intentional investments in R&D are driven by profit incentives. used R&D as a measure of innovative activities and in firm-level
Among other forms of protecting intellectual property, economic studies argued that R&D is more of an input to the actual out-
agents are likely to file patents to protect the property rights put (Schmookler, 1966). Teece (1986) proposes that “an innovation
generated by their private investment in R&D. Moreover, patent consists of technical knowledge about how to do things better than
documents themselves contain references to prior patent docu- the existing state of the art.” One criterion for obtaining a patent
ments and influential inventions tend to be heavily cited. Patent in many countries is “commercial applicability,” which points to
citations can thus in some circumstances (see footnote 1) be inter- the need for patent protection as a kind of insurance policy against
preted as knowledge flows from one invention to another (Jaffe et appropriation (Hall et al., 2001). In the United States, one require-
al., 1993; Duguet and MacGarvie, 2005), and can be used to iden- ment is that patents must be “useful” (hence the term “utility”,
tify those innovations with breakthrough impact. Therefore, patent see United States Code, 1952). The combination of utility, non-
data may allow researchers to gauge the quantity of invention as obviousness, and novelty is required for a patent to be granted in
well as something akin to the “quality” of innovations. In this par- the US, and in several US industries (but certainly not all, see Section
ticular case, we combine observations of such inventive activity 5), patenting is regarded as the most important form of protecting
from countries as described below, so we have one observation per inventions (Cohen et al., 2000). Further, counts of the number of
country per year. patents have been used as a proxy for innovation (e.g., Ahuja and
GDP and other macroeconomic data for the countries were Katila, 2001) with recognition of their limitations (see below).
obtained from the World Development Index (WDI), while R&D We employ both R&D and innovation variables in our specifica-
figures for individual firms inside countries were found in the tions. The actual R&D variable considered, RNDGDP, is the total real
WorldScope database. research and development expenditure in the country as a ratio to
the GDP of the country. In cases where such an R&D figure was not
3.2. Variables available, e.g., Egypt, we simply summed the R&D spending of all
firms reporting in the WorldScope database. Second, we took the
Descriptive statistics and variable names are shown in Table 1, actual number of patents granted to a respective country regard-
and Table 2 presents the correlation matrix of the variables. The less of where the patentee actually resided as long as the company
mean real GDP growth rate is 2.03% and the standard deviation was headquartered in the country. In other words, if Siemens were
twice as large. The explanatory variables also show a great deal of the assignee of a patent in the US patent database, regardless of
variation. Their range indicates that there are outlier observations whether the lab itself was in China, the patent would be assigned
though no effort was made to exclude such observations other than to the country Germany, which is where Siemens is headquartered.
to include fixed effects for countries in some regressions. The simple In our paper we operationalize the variable by taking the number
correlations with the growth rate of GDP are all modest. The level of patents granted per million dollars of R&D.
 I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276 1267

Next, we focus on the “quality” aspect of the innovations. An

1.00000
often-used measure would have been to get a quality measure
based on Hall et al.’s (2001) method, i.e., forward citations of each

11
patent relative to other patents in its technology class in that year,
aggregated up to the country level. We were somewhat successful

0.04040
1.00000
in gathering the citation numbers for most of the sample countries;
10
however, after careful evaluation, we could not defend the reliabil-
ity of the variable for the full sample (see “robustness checks”). We
therefore utilized two alternative quality indicators, as explained

0.02882
−0.01404
1.00000
below.
First, we examined USAR, defined as the Ratio of Patents Granted
9

in the USA as a proportion of total patents granted by the country

The table presents correlation coefficients and the associated Bonferroni-adjusted significance levels of each correlation coefficients. The definitions of the variables are same as in Table 1.
−0.11999*
in a given year, as one proxy for innovation quality.2 For exam-
0.56900
0.06706
1.00000
ple, the country of Brazil granted 26,582 patents in the year 2000
based on their own patent office statistics from PATSTAT, while
8

Italy granted 28,970. We then examine how many patents were

assigned to Brazil and Italy, and granted in the US in that same
0.21609*

0.03495
0.05029

−0.00190
1.00000

year, in this case 98 and 1715, respectively. Thus USAR for Brazil
in 2000 would be equal to 0.37%, while USAR for Italy in 2000
7

would be 5.9%. Our assumption is that the patent applications of

non-US firms granted in the US are usually significant innovations
−0.13353
−0.01968

0.01599
0.01043
−0.00850
1.00000

and represent relative quality. There are several possible explana-

tions for why firms would patent in the US: the “unreliability” of
6

their own national innovation system, for prestige purposes, and for
trade/exporting/sales (direct commercial) purposes. We attempt to
0.71552*
0.25009*
−0.02124

0.04472
0.01602

0.03602
1.00000

tease apart these explanations in the section on “Robustness”.

We also estimate another quality variable, which we call Magic
5

in the Air (MAGIC) by regressing RNDGDP on TPATR (with and with-

out patent stock3 at the end of the year preceding the period) and
0.09552*

−0.06981*
0.09410*

0.10031*

−0.03915
0.05245
−0.03940
1.00000

taking the residual from this regression. Our perspective here is

that while it is plausible that innovation is a product of R&D, it
4

is also possible that innovation may come to an organization or

to a country beyond the proportion of R&D spending. Kortum and
−0.17518*
0.15261*

−0.07967*
0.06380*

−0.06074*
0.03835

0.04737
−0.02380
1.00000

Lerner (1999) note that during the period between 1963 and the
late 1980s, both RNDGDP and patenting quantity increased dramat-
3

ically, but in the 1990s RNDGDP actually fell in most countries while
0.13442*
0.22147*
0.18591*
0.21168*
0.08832*

0.10218*

patenting quantity continued to explode. Thus R&D intensity may

0.00256

0.02186
0.02224
1.00000

not fully reflect all innovation activities. This might be explained by

several factors. One might be that research productivity changes
2

over time in a way described as early as 1964 by Rosenberg &

0.18116*
0.16980*

0.10368*

0.08338*
0.08387*
0.06472*
0.03983
0.03482
0.02789

0.01751
1.00000

Spencer in which they outline changes in the management of R&D

that might make it more productive and commercially oriented.
1

Another explanation might be knowledge spillover effects that

could act as a “multiplier” on R&D expenditures diffused through
Research and development (R&D) expenditure to GDP – RNDGDP

labor force mobility, globalization of companies, or supply chain

“leakage:” e.g., sharing of information with suppliers who also serve
Total number of patents granted to R&D expenses – TPATR

Residual of the estimation of TPATR on RNDGDP – MAGIC

other customers (Tucci et al., 2005). How much innovation is com-

Annual growth rate in per capita real GDP – GDPPCGR

ing from R&D inputs leading to patents and how much of this
Gross private capital formation to GDP – GCAPFORM

inventive activity may also be a product of changes in market struc-

ture or competitive posturing by firms. If the market is competitive,
Government consumption to GDP – GCONGDP

some firms may want to create entry barriers by erecting patent

Foreign direct investment to GDP – FDIGDP

Proportion of patents granted in the USA

“thickets” around certain technologies (Shapiro, 2001). Diffusion

Literacy rate for labor force – LITRATE

and popularization of the Internet may have added to this effect

through the ability of firms to serve markets in a wider geographi-
Technology index – TECHINDX

cal area, thus leading to more competition (Afuah and Tucci, 2003),
Significance levels of less than 5%.

thus giving firms incentives to search out other methods of keeping

Exports to GDP – EXPGDP

competitors at bay, including via intellectual property means.

2
Even though there is a debate in the literature on the standards of the US Patent
Correlation matrix.

& Trademark Office (i.e., are they too “lax” or “aggressive” in granting patents?), we
feel the correct comparison group is not US firms patenting in the US, but non-US
firms patenting both at home and in the US. We argue that due to the motivations
developed below (and suggested by an anonymous referee), the patents filed and
Table 2

granted in the US would tend to be of higher quality than those patented solely at
1
2
3
4
5
6
7
8
9
10
11

home, regardless of whether the US is too aggressively granting patent rights.

*

3
We would like to thank one of the anonymous referees for this suggestion.
1268 I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276

3.2.3. Control variables autoregressive (AR) test, which examines the hypothesis that the
We use a number of independent variables popularly used in error term is not serially correlated in both the difference regres-
the literature in cross-country growth estimations. Initial GDP Val- sion and the system difference-level regression. By construction,
ues are useful to understand the relative changes from sample the differenced error term is allowed to be first-order serially cor-
countries exhibiting a wide range of development stages. Capital related, but the second-order serial correlation of the error term
Formation (GCAPFORM) and Literacy Rate (LITRATE) are common will violate the assumption of the GMM procedure.
proxies for capital and human capital, respectively, of the coun- We also present equations with annual data estimated with the
tries. So is Government Spending (GCONGDP). Exports and Foreign Blundell and Bond dynamic panel data estimation technique, i.e.,
Direct Investments (FDI) are considered crucial to economic growth. the two-step system GMM estimations. We treat all of the financial
We also include a Technology Index variable to control for the tech- and institutional variables as endogenous and the baseline vari-
nological sophistication of the respective countries. This index was ables as exogenous. As described above, the instruments for the
developed from a number of variables: electric power consump- regression in levels are the lagged differences of the corresponding
tion (kWh per capita), fixed line and mobile phone subscribers (per variables, and the instruments for the regression in differences are
1000 people), personal computers (per 1000 people), radios and the lagged levels. We also report the Wald Chi-square test statis-
telephone sets (per 1000 people). tic, the p-value of the Hansen test, and the p-value of the AR(1)
and AR(2) tests. In all instances, the p-values of the Hansen test
3.3. Statistical approach and the AR(2) test are larger than 0.05, which indicates failure to
reject the null hypotheses of over-identification and second-order
Our model builds on the approach to growth equations intro- serial correlation of error terms. In other words, the specification
duced by Barro and Levine (1991). As mentioned earlier, the tests support the validity of the instruments, thus bolstering the
baseline equation includes crucial variables such as the conver- interpretation of the estimated coefficients as being free from endo-
gence effect (log of initial real GDP), the human capital investment geneity bias.
variable (literacy) and the export ratio (openness). We also add Another approach to the endogeneity problems commonly
a few other relevant variables. The baseline regression provides found in the growth literature is to use a multi-year average growth
a reasonable framework for analyzing growth in a cross-country rate as the dependent variable and to use initial year values for all
environment. We then add measures of research and development the independent variables. For example, Levine et al. (2000) use
and innovation (patent) proxies. We start with OLS regressions with non-overlapping 5-year average data in their GMM specifications.
robust standard errors. We then follow the same procedure except Similarly, we also estimate but do not report the GMM estima-
the dependent variable takes 3-year averages of growth as the mea- tions based on 3-year average growth rates due to the similarity
sure of country growth. All estimates control for year and country of the results with the annual data. These regressions are avail-
fixed effects. It is well known that OLS estimates are biased and able upon request. All these regressions meet the specifications
inconsistent when there are dynamic effects and simultaneities in tests, and there is no evidence of second-order serial correlation.
the specification. To account for these effects, the recent literature Furthermore, the regressions pass the Hansen specification test.
(e.g., Levine et al., 2000) has employed the Generalized Method of
Moments (GMM) technique developed by Arellano and Bond and
others for panel estimation and we utilize this technique as well. 4. Results
The Arellano–Bond GMM technique is specifically designed to
address the econometric problems induced by unobserved group- Tables 3–6 show the results of our regression analyses. Each
specific effects and joint endogeneity of the explanatory variables column in the table represents a different model. Let us begin with
in lagged-dependent-variable models, such as growth regressions. Table 3, which shows results for per capita GDP growth in relation to
Similar to Levine et al. (2000), we employ an augmented GMM the innovation and growth variables. Table 3 is grouped into three
procedure outlined in Arellano and Bover (1995) and developed in different groups, with three models per group. As discussed above,
Blundell and Bond (1998), which combines the regression in differ- the first six rows of the table contain control variables considered
ences with the regression in levels (see Bond, 2002). In the Blundell to be important in the economic growth literature, plus the control
and Bond GMM estimator, the instruments for the regression in lev- variable Technology Index, plus fixed effects for Year and Country,
els are the lagged differences of the corresponding variables, and as shown in Models 1, 4, and 7. The first three models are for the
the instruments for the regression in differences are the lagged entire combined sample of all 58 countries over the 25-year period.
levels.4 These are appropriate instruments under the additional The second and third set of three models split the sample along the
assumption that there is no correlation between the differences of lines of income level of the countries, with Models 4–6 showing
these variables and the country-specific effects, while correlation the results for upper-income countries and Models 7–9 showing
between levels of the right-hand side variables and the country- results for the lower-income countries.
specific effect is allowed. There are two tests to test the validity Each of the regression tables shows the same model specifica-
of the instruments, as suggested by Arellano and Bover (1995) and tions. Absolute values of t-statistics are reported in parentheses in
Blundell and Bond (1998). The first is the Sargan test or Hansen all the regression tables and *, **, *** indicate significance at the
test of over-identifying restrictions, which tests the overall validity 10%, 5% and 1% levels respectively. The first equation reported in
of the instruments by analyzing the sample analog of the moment Table 3 column 1 reports the baseline model. The second equation
conditions used in the estimation process.5 The second test is the shows the baseline growth model augmented by the two measures
of innovation related variables, RNDGDP (Research and Development
as a proportion of GDP) and PATR (Patents Granted per Million of R&D
4
More specifically, we use the two-step GMM instead of one-step because Spending). Similarly, equation three (column 3) introduces a dif-
two-step is asymptotically more efficient, meanwhile we also compensate for the ferent proxy for innovation (MAGIC) that takes the residual from
potentially downward biased two-step standard errors by making a finite-sample a separate regression where TPATR is regressed on RNDGDP. The
correction to the two-step covariance matrix derived by Windmeijer (2005). next six columns follow the same order for the two sub-samples.
5
As for the over-identifying restrictions, we conduct the Hansen test instead of
the Sargan test, because the Sargan statistic is not robust to heteroskedasticity or
We also estimate a similar regression excluding the USA from the
autocorrelation while the Hansen statistic, which is the minimized value of the two- sample (not reported, but with highly similar results). We see that
step GMM criterion function, is robust. the baseline regression provides some evidence of a convergence
Table 3
OLS regressions of real per capita GDP growth, annual data.

Combined sample High-income and upper middle-income Lower middle-income and low-income
countries countries

1 2 3 4 5 6 7 8 9

Constant −0.0152 −0.0132 −0.0355 **

−0.0259 −0.0085 −0.0333 −0.0184 0.0040 −0.0541**
(1.39) (0.91) (2.65) (1.28) (0.33) (1.34) (0.78) (0.14) (2.09)

Log of initial per capita real GDP 0.1716*** 0.1617*** 0.1643*** 0.0842 0.0585 0.0941 0.1799*** 0.1675*** 0.1681***
(5.78) (5.61) (1.26) (0.89) (1.42) (5.39) (5.03) (5.11)

Gross private capital formation to GDP 0.0759*** 0.0628*** 0.0669*** 0.0851*** 0.1185*** 0.0757** 0.0702*** 0.0602*** 0.0699***
– GCAPFORM (5.05) (3.54) (3.81) (2.96) (3.06) (2.05) (3.91) (2.96) (3.45)

Government consumption to GDP – 0.0119 0.0122 0.0201 0.0262 0.0318 0.0239 0.0118 0.0493 0.0069

I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276

GCONGDP (0.62) (0.59) (1.06) (1.00) (1.24) (0.92) (0.37) (1.45) (0.22)

Exports to GDP – EXPGDP −0.0040 −0.0046 −0.0023 0.0529*** 0.0675*** 0.0496*** −0.0001 0.00209 −0.0029
(0.78) (0.46) (3.27) (4.03) (3.05) (0.02) (0.35) (0.52)

Literacy rate for labor force – LITRATE 0.0129 0.0230 0.0118 0.0168 0.0235 0.0144 0.0058 0.0229 0.0150
(0.94) (1.63) (0.84) (1.11) (1.52) (0.91) (0.19) (0.75) (0.50)

Foreign direct investment to GDP – 0.0468* 0.0439** 0.0294 0.0701* 0.0784** 0.0541 0.0389 0.0370 0.0201
FDIGDP (1.91) (1.80) (1.21) (1.81) (2.05) (1.38) (1.23) (1.18) (0.67)

Technology index – TECHINDX 0.5378 0.8436 0.2017 0.5548 0.2956 0.6509 0.5374 1.2227 0.8961
(0.12) (1.55) (0.44) (0.76) (0.41) (0.90) (0.67) (1.26) (1.12)

Research and development (R&D) – 0.0059*** – – 0.0490*** – – 0.0075*** –

expenditure to GDP – RNDGDP – (2.87) – – (3.29) – – (3.04) –

Total number of patents granted to – 0.0599** – – 0.6629** – – 0.0561* –

R&D expenses – TPATR – (2.10) – – (2.55) – – (1.91) –

Proportion of patents granted in the – 0.0107** 0.0043 – 0.0104 0.0016 – 0.0107** 0.0032
USA – (2.43) (1.01) – (1.08) (0.19) – (2.02) (0.64)

Residual of the estimation of TPATR on – – 0.6058*** – 0.4924** – – 0.6270***

RNDGDP – MAGIC – – (5.69) – (2.44) – – (5.21)

Year dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes
Country dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 1104 1104 1104 213 213 213 891 891 891
Adj. R2 0.0586 0.0690 0.0863 0.1327 0.1783 0.1502 0.0478 0.0594 0.0756
F-Statistic 10.81 9.18 12.58 5.63 5.60 5.16 7.39 6.62 9.09

The table presents the OLS regressions of real per capita GDP growth based on the annual data at the country level with White heteroskedastic-consistent standard errors. The dependent variable is real per capita GDP growth,
and we include log of initial per capita real GDP, gross private capital formation, government consumption, and literacy rate as the base variables in the regressions. Definitions of variables are same as in Table 1. N refers to
number of observations included in the estimation. Absolute values of t-statistics of the coefficients of the independent variables are reported in parentheses.
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.

1269
 1270
Table 4
OLS regressions of real per capita GDP growth (3-year average data).

Combined sample High-income and upper middle-income Lower middle-income and low-income
countries countries

1 2 3 4 5 6 7 8 9
*** *** *** *** *** *** *** ***
Constant 3.8638 4.0346 3.5374 4.7879 4.4372 4.1169 4.1424 4.9980 3.8579***
(14.63) (11.64) (10.81) (10.33) (7.47) (7.18) (6.99) (6.86) (5.85)

Log of initial per capita real GDP 2.7259*** 2.3920*** 2.6908*** 2.5797 2.2240 2.8368 2.6621*** 2.2697*** 2.6193***
(3.65) (3.22) (3.61) (1.55) (1.36) (1.73) (3.17) (2.70) (3.11)

Gross private capital formation to GDP 2.0758*** 1.7637*** 1.7678*** 2.3416*** 1.8358** 1.2697* 1.9838*** 1.8505*** 1.8423***
– GCAPFORM (5.56) (4.07) (4.05) (3.70) (1.99) (1.88) (4.34) (3.68) (3.63)

I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276

Government consumption to GDP – 0.1160 0.6674 0.1400 0.4469 0.5521 0.6128 0.1146 1.0982 0.2230
GCONGDP (0.24) (1.30) (0.29) (0.74) (0.91) (1.01) (0.15) (1.29) (0.28)

Exports to GDP – EXPGDP −0.0941 −0.1083 −0.0837 −0.0473 0.1070 −0.0914 0.1060 0.1531 0.0932
(0.76) (0.88) (0.68) (0.12) (0.26) (0.24) (0.78) (1.13) (0.68)

Literacy rate for labor force – LITRATE 0.3023 0.5801* 0.3967 0.2653 0.5678 0.4010 −0.1395 0.2760 0.0256
(0.93) (1.73) (1.18) (0.77) (1.57) (1.10) (0.20) (0.37) (0.03)

Foreign direct investment to GDP – 0.3160 −0.3933 −0.4001 −0.1312 0.1010 −0.2079 −0.6172 −0.6447 −0.6770
FDIGDP (0.57) (0.72) (0.74) (0.16) (0.12) (0.25) (0.85) (0.89) (0.93)

Technology index – TECHINDX 2.0154 2.2056 1.9026 3.3889** 3.13478 3.0246 2.0039 1.0389 1.3582
(0.18) (0.19) (0.17) (0.16) (0.18) (0.36) (0.10) (0.75) (0.07)

Research and development (R&D) 0.1753*** 0.5343 0.2007***

expenditure to GDP – RNDGDP (3.44) (1.54) (3.23)

Total number of patents granted to 1.8689** 1.9323** 1.1740**

R&D expenses – TPATR (2.46) (2.26) (2.21)

Proportion of patents granted in the 0.2679** 0.14254 0.2225 0.3286 0.2301* 0.0897
USA (2.39) (1.29) (0.92) (1.56) (1.70) (0.68)

Residual of the estimation of TPATR on 3.9674 1.4253** 2.3784

RNDGDP – MAGIC (1.40) (1.98) (0.73)

Year dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes
Country dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 823 823 823 227 162 162 661 661 661
Adj. R2 0.0486 0.0679 0.0511 0.1080 0.1465 0.1384 0.0365 0.0544 0.0351
F-Statistics 7.00 6.99 5.92 3.78 3.76 3.87 4.57 4.80 3.67

White heteroskedastic-consistent standard errors. All independent variables are the initial year of the 3-year period. Absolute values of t-statistics of the coefficients of the independent variables are reported in parentheses.
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
Table 5
Blundell Bond dynamic panel data estimations of real capita GDP growth (annual data), two-step system GMM results.

Combined sample High-income and upper middle-income Lower middle-income and low-income
countries countries

1 2 3 4 5 6 7 8 9

Constant −0.0012 0.0685* −0.0451*** −0.0026 −0.8073 −0.3064 0.0326 4.3197 0.2956
(0.11) (1.94) (3.66) (0.14) (1.31) (1.30) (0.21) (0.81) (1.37)

Log of initial per capita real GDP 0.1514*** 0.0892*** 0.1005*** 0.1271*** 0.0646** 0.1218*** D 1.7124* 3.0913*
(13.71) (4.56) (6.65) (8.60) (2.07) (4.42) (1.82) (1.76)

Gross private capital formation to GDP 0.1042*** 0.1127*** 0.0806*** 0.0918*** 0.0841 0.0852*** 0.7835 0.2761 1.0774**
– GCAPFORM (15.96) (4.63) (5.59) (12.83) (1.19) (3.12) (0.86) (0.86) (2.17)

I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276

Government consumption to GDP – 0.0015 0.0502* 0.0230 0.0307*** 0.8462 0.1718 D D D
GCONGDP (0.07) (1.79) (1.35) (2.70) (1.57) (0.79)

Exports to GDP – EXPGDP −0.0335 −0.0269 −0.0151 −0.0102 −0.0410 −0.0307 0.3943 2.1277* 0.5464
(0.64) (0.79) (1.01) (0.71) (1.11) (1.35) (1.50) (1.66) (0.23)

Literacy rate for labor force – LITRATE 0.0581*** 0.1134*** 0.0014 0.0326 0.6202 0.2636 D D D
(5.18) (3.61) (0.04) (1.25) (0.97) (1.11)

Foreign direct investment to GDP – 0.0318* 0.0055 0.0051 0.0213 0.0264 0.0885 1.2697 1.3850* 1.2353*
FDIGDP (1.68) (0.12) (0.17) (0.77) (0.38) (1.26) (1.48) (1.75) (1.82)

Technology index – TECHINDX 1.6712*** 5.7588*** 0.4595 0.7763 2.0806 2.2311 D D D

(3.33) (3.01) (0.41) (0.60) (0.49) (0.88)

Research and development (R&D) 0.0124* 0.0075* 0.0131*

expenditure to GDP – RNDGDP (1.72) (1.82) (1.84)

Total number of patent granted to R&D 0.1434** 0.0157** 1.5432***

expenses – TPATR (1.96) (2.22) (2.71)

Proportion of patent granted in the USA 0.0054 0.0106* 0.0407* 0.0223* 0.1289 0.0442*
(0.39) (1.75) (1.89) (1.68) (1.03) (1.82)

Residual of the estimation of TPATR on 0.6482*** 0.8850** 0.8251***

RNDGDP – MAGIC (4.73) (2.52) (2.86)

Observations 1104 1104 1104 213 213 213 891 891 891
Wald 2 536.8 3992.0 464.5 252.0 574.5 204.3 2.469 4.647 4.930
p-Value of Hansen test 0.000 0.000 0.000 0.000 0.000 0.000 0.984 1.000 0.998
AR(1) test 0.000 0.000 0.000 0.000 0.000 0.000 0.566 0.211 0.241
AR(2) test 0.162 0.384 0.401 0.285 0.530 0.153 0.710 0.630 0.958

Blundell Bond two-step system GMM results. All independent variables are the initial year of the 3-year period. Absolute values of t-statistics of the coefficients of the independent variables are reported in parentheses. D, dropped.
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.

1271
 1272
Table 6
Blundell Bond dynamic panel data estimations of real capita GDP growth (3-year average data), two-step system GMM results.

Combined sample High-income and upper middle-income Lower middle-income and low-income
countries countries

1 2 3 4 5 6 7 8 9
*** *** *** *** *** *** ***
Constant 1.564 2.976 2.663 2.97 3.453 3.457 2.952 1.763 2.077***
(8.365) (9.056) (7.194) (4.602) (3.728) (4.18) (2.92) (3.02) (3.51)

Log of initial per capita real GDP 1.253*** 1.792*** 2.073*** 1.375* 1.028 1.435* 1.664** 1.845** 2.159*
(3.063) (3.459) (3.252) (1.66) (1.43) (1.88) (2.35) (1.97) (1.85)

Gross private capital formation to GDP 0.885** 0.098** 1.029** 1.275** 1.052** 0.952** 1.262** 1.052** 1.419**
– GCAPFORM (2.37) (2.55) (2.49) (2.063) (2.00) (2.43) (2.39) (2.25) (2.47)

I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276

Government consumption to GDP – 0.045* 0.039 0.058* 0.175* 0.149 0.253 D D D
GCONGDP (1.75) (1.38) (1.69) (1.94) (1.40) (1.46)

Exports to GDP – EXPGDP −0.021 −0.017 −0.010 −0.008 −0.015 −0.054 0.518 1.712* 1219
(0.38) (0.57) (1.23) (0.48) (0.40) (0.51) (0.57) (1.69) (1.52)

Literacy rate for labor force – LITRATE 0.276 0.107* 0.175 0.263 0.191 0.285 D D D
(1.38) (1.69) (1.45) (1.02) (1.29) (1.10)

Foreign direct investment to GDP – 0.015 0.007 0.017 0.028 0.019 0.0712 1.196 1.219 1.001
FDIGDP (1.47) (1.15) (0.31) (1.38) (1.06) (1.05) (1.34) (1.29) (1.64)

Technology index – TECHINDX 1.2854* 0.985 0.751 0.776 2.155 2.189 D D D

(1.76) (1.62) (1.07) (0.60) (0.38) (1.60)

Research and development (R&D) 0.062* 0.079* 0.059*

expenditure to GDP – RNDGDP (1.91) (1.26) (1.76)

Total number of patent granted to R&D 0.109** 0.0209** 1.109**

expenses – TPATR (1.99) (2.45) (2.04)

Proportion of patent granted in the USA 0.063* 0.021* 0.047* 0.025* 0.128 0.044*
(1.70) (1.89) (1.89) (1.68) (1.03) (1.82)

Residual of the Estimation Of TPATR on 0.696** 0.716** 0.903***

RNDGDP – MAGIC (1.93) (2.13) (2.85)

Observations 823 823 823 227 162 162 661 661 661
Wald 2 507.2 3648 418.0 206.9 523.7 197.7 2.138 4.269 4.642
p-Value of Hansen test 0.000 0.000 0.000 0.000 0.000 0.000 0.864 0.936 0.986
AR(1) test 0.000 0.000 0.000 0.000 0.000 0.000 0.506 0.200 0.229
AR(2) test 0.141 0.309 0.326 0.226 0.506 0.149 0.710 0.630 0.958

Blundell Bond two-step system GMM results. All independent variables are the initial year of the 3-year period. Absolute values of t-statistics of the coefficients of the independent variables are reported in parentheses. D, dropped.
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
 I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276 1273

effect. As the OLS annual and 3-year equations show (Table 4), whether the country is known as an “emerging” or a “developed”
the convergence effect is mostly significant although it is insignif- country market. We used nominal R&D expenditures instead of real
icant in the high- and upper middle-income country sub-sample. expenditures. We also added some additional independent vari-
In both annual and 3-year average OLS estimations, Gross Capital ables used by some authors in the growth literature as control
Formation (GCAPFORM) and FDIGDP variables are consistently sta- variables, e.g., bank credit to private sector as a proportion to GDP
tistically significant. The openness or Export variable is positive and and also private sector to GDP ratio. In these cases, we found strong
statistically significant in a few of the regressions. support for our hypotheses. Moreover, we substituted the quality
Coming to our hypotheses, the first hypothesis proposed that of innovation variable – proportion of patents granted in the USA
innovation activities are important for firms’ economic growth. This – with a new proxy: total number of forward citations of the total
is seen by looking at the second model in each subset of the data granted patents. We could not get consistent data for all sample
(i.e., Models 2, 5, and 8 in each of the two tables). As shown in the countries for all sample years. However, we still found statistically
tables, the ratio R&D Expenditures to GDP has a positive and statis- significant association between our independent variables and the
tically significant relation with growth in per capita GDP. Likewise dependent variable that supports the two hypotheses.
the ratio of Total Patents Granted to Total R&D Expenditures is posi- We also undertook an investigation of “macro gaps” or gaps
tive and statistically significant, indicating an association with GDP in national innovation systems that may confound US patenting
growth. The first measure, as discussed above, can be considered an with quality. The argument goes that poor institutional support in
innovation input measure, while the second measure may be con- the home country may drive firms to patent in the US not because
sidered an innovation output or innovation efficiency measure. Thus quality is so high but that home protection is so poor that a high
we see that innovation input and output are associated with GDP percentage of patents are filed in the US. First, we examined the
growth virtually across the board. Although coefficient estimates applications from non-US companies in the US and found that (1)
and confidence levels vary somewhat from equation to equation they were from the most developed countries, and (2) they were
and across tables, the overall picture is supportive of the fact that also related to the amount of trade they have with the US. We
innovation does matter in explaining economic growth. Turning to also conducted two additional regression analyses, the first was
coefficients on the measures of R&D and innovation variables, we a simultaneous equation (SUR) analysis with annual growth rate
see that RNDGDP has a relatively small coefficient, while TPATR has per capita GDP a function of lagged growth, gross capital formation,
a larger coefficient that is positive and statistically significant. government consumption to GDP, exports to GDP, literacy rate, FDI to
The second hypothesis has to do with whether quality innova- GDP, technology index, R&D Expenditures to GDP, total patents per
tion output matters. This is shown in the same models as above, plus million dollars of R&D, and percentage granted in the US; while per-
the third model of each subset of the data (i.e., Models 3, 6, and 9 in centage granted in the US is simultaneously a function of deviation
each of the two tables). As discussed previously, we take the Propor- from US GDP growth, exports to GDP, literacy rate, technology index,
tion of Patents Granted in the USA as a kind of patent quality measure and R&D Expenditures to GDP. In such specifications, the deviation
at the country level. Overall, there appears to be some support for from US GDP growth is negatively related to percentage granted in
this hypothesis: the coefficient of our quality measure is often, but the US, meaning that rapidly growing (developing) countries tend
not always, positive and statistically significant. However, when we to have a lower percentage patented in the US, not a higher percent-
add the variable MAGIC, which represents a kind of factor in innova- age (results available from authors). We observed the same results
tive output unexplained by R&D spending, the Proportion of Patents with the same functional form but using a 2SLS specification. We
Granted in the USA always becomes statistically insignificant. The therefore conclude that our results are not being totally driven by
overall interpretation offers support for Hypothesis 2, that qual- a macro-gap story, but more likely a direct commercial gain (e.g.,
ity innovation output leads to even higher growth than average exports or US sales) story.
innovation output. We also undertook an analysis of the role of patent stock on our
The GMM annual regressions (Table 5) as well as the 3-year results. For example, it could be that the MAGIC variable represents
average estimations (Table 6) suggest that the results are consis- not a knowledge spillover, but rather a “momentum” effect of prior
tent with the OLS estimations although the relative importance, i.e., output. In other words, when a country has a higher unexplained
the statistical significance of the coefficients of our key indepen- output given their R&D investments, it could be because of all the
dent variables are relatively less robust. As shown in the combined prior research done in the past and not because current knowledge
sample (columns 1–3), both RNDGDP and TPATR show a statisti- is diffusing rapidly. Thus we reran all of our analyses, with patent
cally significant positive impact on economic growth. Additionally, stock included in the first stage in which we generated the MAGIC
when the MAGIC variable is substituted as the alternative innova- variable. In results available from the authors, we find that indeed
tion variable, we still find a strong positive influence of innovation. although the patent stock is statistically significant in the regres-
When the estimations are done separately on upper- and lower- sion that produces MAGIC, the variable MAGIC in the subsequent
income sub-samples, we see the magnitude and strength of the regressions loses none of its power.
results are similar to the combined results except that the lower- Therefore, in all instances examined, there were no alarming
income country coefficients of the focused variables are more qualitative changes to our results or interpretations thereof.
statistically significant.6
In sum, our results show that after controlling for the year and 5. Discussion and conclusions
country-specific effects, there exists a strong, positive link between
innovative success and economic growth. In summary, we find that both the quantity of inventive activity,
Robustness tests. In order to boost confidence in our results, we as well as its quality, are associated with economic growth. Based
performed various robustness tests. The robustness tests address on the OLS results, we may be able to conclude that countries that
several important issues. First, we investigated whether the results have higher levels of patenting activity – as well as those whose
were consistent if we examined different sub-samples, such as patents primarily are filed in the US – tend to be the countries with
higher growth rates. Furthermore, it seems based on panel regres-
sion and country fixed effects that countries that increase the level
6
We should add that in estimating the GMM for this lower-middle-income and
of patenting activity – or increase the proportion filed in the US –
lower-income country sub-sample, the program automatically dropped some vari- tend to be associated with increases in the growth rate. Thus we
ables from the regression. find results consistent with both our hypotheses.
1274 I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276

Table 7
Granger causality tests.

We wanted to investigate the direction of causality between GDP per capita growth with various patent quality measures to see whether quality has an impact on
growth or higher growth causes corporations to be produce higher quality innovations. To this end, we employed the Granger–Sims causality framework (Granger,
1969; Sims, 1972) and estimated the regressions in the following way. We first tried to determine if USAR in the previous year has an impact on the GDPPCGR
variable in the current year: Similarly, we replace USAR with MAGIC.

GDPPCGRt = f(GDPPCGRt−1 , GDPPCGRt−2 ) + ei

GDPPCGRt = f(GDPPCGRt−1 , GDPPCGRt−2 , . . . USARt−2 )

GDPPCGRt = −0.0215(1.23) + 0.1813 GDPPCGRt−1 (6.04) + 0.1526 GDPPCGRt−2 (3.28), R2 = 0.0379

GDPPCGRt = −0.0194(1.38) + 0.1659 GDPPCGRt−1 (5.87) + 0.137 GDPPCGRt−2 (3.09) + 0.0135 USARt−1 (2.11), R2 = 0.0478

The coefficients and t-statistics (in parenthesis) reported above give us a perspective whether USAR was causing growth or whether growth was important in
determining quality of innovation. In this case we were not able to reject the null hypothesis that the USAR had no explanatory power with regards to GDPPCGR. In
fact, we found that the quality variable in year t − 1 was significant at the 5% level. We performed similar regressions using the other alternative quality variable
(MAGIC). The results were reasonably similar, although we obtained the stronger t-statistics (t = 5.42 significant at the 1% significance level).
We then reversed the question and tried to determine whether current or past GDPPCGR performance has an impact on USAR. We estimated the following
regression:

USARt = f(USARt−1 , USARt−2 , GDPPCGRt−1 ) + ei

We obtained the following results:

USARt = 0.0334(1.70) + 0.015 USARt−1 (1.96) + 0.0053 USARt−1 (1.76) + 0.046 GDPPCGRt−1 (1.62), R2 = 0.1354

The significant t-statistics on GDPPCGRt−1 suggests that past performance also has some marginal affect (coefficient 0.046, t = 1.63 significant at 11% level) on
quality measure even after past quality measures are taken into consideration. When we substituted USAR with MAGIC as a quality variable, the results were not
statistically significance at all.
Finally, we performed additional tests (not reported in the text) by incorporating a future period (t + 1) along with the past periods (t − 1, t − 2) and re-estimated the
above-mentioned regressions. We found that in all cases the causality seems to run in both directions except it is significantly stronger and statistically more
significant in the direction from quality of innovation to growth. The overall evidence associated with the Granger–Sims tests suggests that future studies should
consider more careful analysis by including a higher number of lagged values.

An important limitation of this study, of course, is the general Freeman and Soete (1997, pp. 112–120) and Tid and Bessant (2009,
equation of patents with innovation and the use of various deriva- pp. 555–560).
tive measures based on such patents to infer innovation level. This Regarding the dependent variable, we chose GDP growth
is a relatively common practice in the literature at multiple levels because it seems this is the most important variable considered by
of analysis (e.g., Ahuja and Katila, 2001; Bottazzi and Peri, 2003; policymakers. However, in future research we should also consider
Maurseth and Verspagen, 2002) because (1) patents do indeed rep- total factor productivity or other national performance measures,
resent inventive output, and such output is intended to (usually, and perhaps control for educational levels, political risk, and pos-
see below) be commercialized; (2) detailed statistics have been sibly levels of government debt.7
collected and kept over the years; and (3) patents are in general Another question the results raise is one of reverse causality.
costly to obtain and defend, implying some kind of financial benefit That is, how do we know that more patents lead to economic
in return. Thus if innovation = invention + exploitation, as Roberts growth, rather than countries with high economic growth leads
(1988) has proposed, it does appear that the inventive output plus to firms that have more profits to invest, which they then spend on
the intention and incentive to commercialize (exploit) would lead patenting activities? Or it could be that patenting in the US actually
to something akin to innovation. leads to higher returns than patenting in the home country due to
However, there are indeed several problems with equating the size of the US market, which leads to more profits for the firm,
patents and innovation, which we must acknowledge. Griliches leading to growth in the home country (i.e., it is not a question of
(1990, p. 1669) pointed out several, for example, “Not all inven- quality at all, more of path dependency). These questions are diffi-
tions are patentable, not all inventions are patented, . . .” The first cult to tease apart empirically but in an ad hoc analysis, we exam-
of these refers to the fact that there are many innovations that do ined a Granger–Sims causality test (Granger, 1969; Sims, 1972) as
not qualify for legal patent protection (e.g., software products in shown in Table 7. The results of this analysis indicate that, while
many countries of the world, or organizational innovations such as we cannot completely rule out the possibility of growth leading to
adoption of cross-functional teams). The second is that there are higher quality patents, the preponderance of evidence leads us to
many innovations that are never patented (and instead kept as a believe that on average, higher quality patenting precedes growth.
secret), even though they could be patented from a legal point of Another question that this research raises is related to the locus
view (Cohen et al., 2000). Another third issue hinted at above is that of the inventive activity itself. For example, outsourcing has been a
firms may use patents to “block” other companies from commer- hot topic in the early 2000s. Is it possible that a country may have
cializing an invention rather than to commercialize it themselves. A much innovative activity but the economic growth occurs outside
fourth problem may be the different incidences of patenting behav- that country? Put another way, how do we know that if a multi-
ior across different industries (e.g., high propensity in chemicals national firm has R&D centers in various countries, the benefits
compared to lower incidence in software or white goods). flow only to the country in which the firm is headquartered? Thus
Due to the issues described above, patent statistics may be con- another limitation of our study is the inability to tease apart a more
sidered “noisy” and imperfect proxies for innovation. However, fine-grained tracing of patenting and economic behavior. However,
many scholars have concluded that despite their limitations and for a broad picture, the results seem remarkably consistent and
despite the “noise,” patents are acceptable for studying innovation clear.
and in fact are often associated with better measures of innovation
in the limited number of cases where such data are available. For
a more detailed discussion on the advantages and disadvantages 7
We would like to thank Pierpaolo Parrotta and Marianna Marino for these sug-
of using patent data in innovation studies, see Griliches (1990), gestions.
 I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276 1275

The public policy implications of the results are also evident. Aghion, P., Howitt, P., 1992. A model of growth through creative destruction. Econo-
While Lerner (2002) warns that strengthening intellectual prop- metrica 60 (2), 323–351.
Ahuja, G., Katila, R., 2001. Technological acquisitions and the innovation perfor-
erty protection does not do much to spur patent applications in mance of acquiring firms: a longitudinal study. Strategic Management Journal
countries with already advanced systems, it would seem logical 22, 197–220.
that countries with weaker systems could get many more applica- Arellano, M., Bover, O., 1995. Another look at the instrument variable estimation of
error-components models. Journal of Econometrics 68, 29–51.
tions by strengthening their intellectual property rights. Of course, Audretsch, D.B., 1995. Innovation and Industry Evolution. MIT Press, Cambridge.
simply applying for more patents does not necessarily translate Audretsch, D.B., Feldman, M.P., 1996. R&D spillovers and the geography of innovation
into either more patents granted or higher quality patents. Cer- and production. American Economic Review 86 (3), 630–640.
Barro, R.J., Levine, R., 1991. Economic growth in a cross-section of countries. Quar-
tainly, a country should not lower its standards when it comes to terly Journal of Economics 106 (2), 407–443.
granting patents. Maybe governments should subsidize the fees or Bayoumi, T., Coe, D.T., Helpman, E., 1999. R&D spillovers and global growth. Journal
even award prizes for successfully applying for patents in the US? of International Economics 47 (2), 399–428.
Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic
The standards in the US should remain relatively constant over any
panel data model. Journal of Econometrics 87, 115–143.
short time period so the overall quality level should remain rela- Bond, S.R., 2002. Dynamic panel data models: a guide to micro data methods and
tively constant. However, the increased innovation activity to get practice. Portuguese Economic Journal 1, 141–162.
past the application hurdles and perhaps win the prize may lead to Bottazzi, L., Peri, G., 2003. Innovation and spillovers in regions: evidence from Euro-
pean patent data. European Economic Review 47 (4), 687–710.
higher economic growth in the country. Cameron, G., 1998. Innovation and growth: a survey of the empirical evidence.
Thus it would seem that, based on the results of this study, gov- Working Paper. Nuffield College, Oxford University.
ernment schemes that attempt to raise the level of R&D spending Coe, D.T., Helpman, E., 1995. International R&D spillovers. European Economic
Review 39, 859–887.
in a country may be only one of several R&D policy tools associated Cohen, W.M., Nelson, R.R., Walsh, J.P., 2000. Protecting their intellectual assets:
with increasing economic growth. We may consider it to be part of appropriability conditions and why US manufacturing firms patent (or not).
a multi-pronged approach that includes RNDGDP as one piece. As NBER Working Paper 7752. National Bureau of Economic Research.
Duguet, E., MacGarvie, M., 2005. How well do patent citations measure flows of
we know, the goal of most economies is not more R&D spending technology? Evidence from French innovation surveys. Economics of Innovation
per se, but rather a higher rate of growth and national wealth that and New Technology 14 (5), 375–393.
is presumed to result from such investment. Fagerberg, J., 1994. Technology and international differences in growth rates. Journal
of Economic Literature 32 (3), 1147–1175.
As Griliches (1990, p. 1669) also observed, “the inventions Francis, B., Hasan, I., Wang, H., 2007. The role of venture capital on innovation, new
that are patented differ greatly in ‘quality’ in the magnitude of business formation and economic growth. Working Paper. RPI Lally School of
inventive output associated with them.” Regarding such quality, Management & Technology.
Freeman, C., Soete, L., 1997. The Economics of Industrial Innovation. MIT Press,
while RNDGDP was certainly statistically significant and associ-
Cambridge.
ated with growth, we cannot discount the statistical significance Goel, R.K., Ram, R., 1994. Research and development expenditures and economic
of MAGIC as well. There could be several institutional factors at growth: a cross-country study. Economic Development and Cultural Change 42
play that go beyond R&D spending relating to employee mobility (2), 403–411.
Granger, C.W.J., 1969. Investigating Causal Relations by Econometric Models and
(and thus spillovers). On the legal front, these institutional fac- Cross-spectral Methods. Econometrica 37 (3), 424–438.
tors could include labor law, taxation law, immigration law, and Griliches, Z., 1980. R&D and the productivity slowdown. American Economic Review
intellectual property law (especially governing trade secrets). Also 70 (2), 343–348.
Griliches, Z., 1990. Patent statistics as economic indicators: a survey. Journal of
importantly entering into this legal equation could be contract law, Economic Literature 28 (4), 1661–1707.
which has to do with the employment relation, non-compete, or Griliches, Z., 1992. The search for R&D spillovers. Scandinavian Journal of Economics
restraint of trade laws. For example, in jurisdictions in which non- 94 (Suppl.), 29–47.
Griliches, Z., Mairesse, J., 1986. R&D and productivity growth: comparing Japanese
compete clauses are less enforceable, one would expect to see more and US manufacturing firms. European Economic Review 21 (1/2), 89–119.
employee mobility and thus more spillovers (van Caenegem, 2005). Grossman, G.M., Helpman, E., 1994. Endogenous innovation in the theory of growth.
Other factors might include prevalence of trade fairs or technical Journal of Economic Perspectives 8 (1), 23–44.
Hall, B., Jaffe, A., Trajtenberg, M., 2001. The NBER Patent Citations Data File: Lessons,
conferences, availability of venture capital (not from a financial Insights and Methodological Tools. NBER Working Paper No. 8498. National
point of view but a knowledge leakage point of view), prevalence of Bureau of Economic Research.
crowdsourcing over the Internet, strength of university–industry Harhoff, D., Hoisl, K., Webb, C., 2008. European patent citations—how to count and
how to interpret them? Working paper. University of Munich, CEPR and OECD.
ties, and/or cultural norms for exchanging what could be propri-
Jaffe, A.B., Trajtenberg, M., 1999. International knowledge flows: evidence from
etary information. Thus a whole host of institutions and culture patent citations. Economics of Innovation and New Technology 8 (1/2), 105–136.
that affect knowledge sharing and therefore spillovers and there- Jaffe, A.B., Trajtenberg, M., Henderson, R., 1993. Geographic localization of knowl-
fore the ability to generate more growth per dollar spent on R&D. edge spillovers as evidenced by patent citations. Quarterly Journal of Economics
108 (3), 577–598.
So as our final policy prescription/question: perhaps national gov- King, R.G., Levine, R., 1993. Finance and growth: Schumpeter might be right. Quar-
ernments should also focus some of their attention on raising the terly Journal of Economics 108 (3), 717–737.
“magic in the air”? Kirchhoff, B.A., 1994. Entrepreneurship and Dynamic Capitalism: The Economics of
Business Firm Formation and Growth. Praeger, Westport, CT.
Kirchhoff, B., Catherine, A., Newbert, S.L., Hasan, I., 2007. The influence of univer-
sity R&D expenditures on new business formations and employment growth.
Acknowledgements Entrepreneurship Theory and Practice 31, 543–559.
Knack, S., Keefer, P., 1995. Institutions and economic performance: cross-country
testing using alternative institutional measures. Economics and Politics 7 (3),
The authors would like to thank Marco Ceccagnoli, Dominique 207–227.
Foray, W. Curt Hunter, Stéphane Lhuillery, Marianna Marino, Kortum, S., Lerner, J., 1999. What is behind the recent surge in patenting? Research
Pierpaolo Parrotta, Grid Thoma, Editor Fumio Kodama, the two Policy 28 (1), 1–22.
Lerner, J., 2002. Patent protection and innovation over 150 years. NBER Working
anonymous Research Policy referees, and participants at the Asso-
Paper W8977. National Bureau of Economic Research.
ciazione Guido Carli/Fondazione Cesifin Alberto Predieri Fifth Levine, R., Loayza, N., Beck, T., 2000. Financial intermediation and growth: causality
Florence Colloquium on “Money, Derivatives, Innovation, and and causes. Journal of Monetary Economics 46 (1), 31–77.
Lucas, R.E., 1988. On the mechanics of economic development. Journal of Monetary
Growth,” for helpful comments on prior versions of this paper.
Economics 22, 3–42.
Mansfield, E., 1980. Basic research and productivity increase in manufacturing.
American Economic Review 70 (5), 863–873.
References Maurseth, P.B., Verspagen, B., 2002. Knowledge spillovers in Europe: a patent cita-
tion analysis. Scandinavian Journal of Economics 104 (4), 531–545.
Afuah, A., Tucci, C.L., 2003. Internet Business Models and Strategies, 2nd edition. Nelson, R.R. (Ed.), 1993. National Systems of Innovation. Oxford University Press,
McGraw-Hill, New York. New York.
1276 I. Hasan, C.L. Tucci / Research Policy 39 (2010) 1264–1276

Roberts, E.B., 1988. What we’ve learned: managing invention and innovation. Stokey, N.L., 1995. R&D and economic growth. Review of Economic Studies 28 (4),
Research Technology Management 31 (January–February), 11–29. 1661–1707.
Romer, P.M., 1986. Increasing returns and long-run growth. Journal of Political Econ- Teece, D.J., 1986. Profiting from technological innovation. Research Policy 15 (6),
omy 94 (5), 1001–1037. 285–305.
Romer, P.M., 1990. Endogenous technological change. Journal of Political Economy Tid, J., Bessant, J., 2009. Managing Innovation: Integrating Technological, Market and
98 (5), S71–S102. Organizational Change. Wiley, New York.
Rousseau, P.L., Wachtel, P., 2000. Equity markets and growth: cross-country evi- Tucci, C.L., Kaufman, A., Wood, C.H., Theyel, G., 2005. Collaboration and teaming in
dence on timing and outcomes, 1980–95. Journal of Banking and Finance 24, the software supply chain. Supply Chain Forum 6 (1), 16–28.
1933–1957. United States Code, Title 35, Section 101 (Patent Act of 1952).
Schmookler, J., 1966. Invention and Economic Growth. Harvard University Press, van Caenegem, W., 2005. Inter-firm migration of tacit knowledge: law and policy.
Cambridge. Prometheus 23 (3), 285–306.
Schumpeter, J.A., 1932. Capitalism, Socialism, and Democracy. Harper and Brothers, Wachtel, P., 2001. Growth and finance: what do we know and how do we know it?
New York. International Finance 4, 335–362.
Shapiro, C., 2001. Navigating the patent thicket: cross-licenses, patent pools, and Wennekers, S., 1999. Linking entrepreneurship and economic growth. Small Busi-
standard setting. In: Jaffe, A.B., Lerner, J., Stern, S. (Eds.), Innovation Policy and ness Economics 13 (1), 27–55.
the Economy, vol. 1. MIT Press, Cambridge, pp. 119–150. Windmeijer, F., 2005. A finite sample correction for the variance of linear efficient
Sims, C.A., 1972. Money, income, and causality. American Economic Review 62 (4), two-step GMM estimators. Journal of Econometrics 126, 25–51.
540–552.