The Institutional Sources of Knowledge and the
Value of Academic Patents
E. Sapsalis and B. van Pottelsberghe de la Potterie
This paper puts forward new potential determinants of patent value which are
mainly related to the identification of institutional sources of knowledge and the
geographic scope of patenting strategy. The impact of these new indicators is
evaluated through an empirical analysis that focuses on the number of forward
citations received by 208 patent families applied for by six Belgian Universities.
The new indicators provide a more in-depth view on the way non-patent
citations, backward patent citations, co-assignees and the geographical scope for
protection determine patent value. The policy implications induced by these
results are the positive impact of collaboration between public research
organisations and the need to focus on academic researchers with a high
scientific profile in terms of publications in order to crystallize their tacit
knowledge into high value academic patents.
JEL Classifications: L33, O33, O34
Keywords: Patent value, patent indicators, knowledge sources.
CEB Working Paper N° 04/003
2004
Université Libre de Bruxelles – Solvay Business School – Centre Emile Bernheim
ULB CP 145/01 50, avenue F.D. Roosevelt 1050 Brussels – BELGIUM
e-mail: [email protected] Tel. : +32 (0)2/650.48.64 Fax : +32 (0)2/650.41.88
The Institutional Sources of Knowledge
and the Value of Academic Patents
Elefthérios SAPSALIS’ and Bruno van POTTELSBERGHE DE LA POTTERIED
Université Libre de Bruxelles - Solvay Business School
Forthcoming in Economics of Innovation and New Technology
June 2005
ABSTRACT
This paper puts forward new potential determinants of patent value which are mainly related
to the identification of institutional sources of knowledge and the geographic scope of
patenting strategy. The impact of these new indicators is evaluated through an empirical
analysis that focuses on the number of forward citations received by 208 patent families
applied for by six Belgian Universities. The new indicators provide a more in-depth view on
the way non-patent citations, backward patent citations, co-assignees and the geographical
scope for protection determine patent value. The policy implications induced by these results
are the positive impact of collaboration between public research organisations and the need to
focus on academic researchers with a high scientific profile in terms of publications in order
to crystallize their tacit knowledge into high value academic patents.
JEL: L33, O33, O34
Keywords: Patent value, patent indicators, knowledge sources.
Acknowledgments: This paper was partly written when Bruno van Pottelsberghe was Visiting Professor at the
Institute of Innovation Research (IIR) of Hitotsubashi University, from July to December 2003. Financial
support was provided by the SSTC, Belgium. We would like to thank Professors Sadao Nagaoka, Akira Goto,
Hiroyuki Odagiri, and the participants to the following seminars/conferences for their helpful comments: Tokyo
University, 24 October 2003; MEXT, NISTEP, 27 October 2003; SPRU Conference, Sussex University, 19
November 2003. We are very grateful to two anonymous referees for their constructive comments.
’
FNRS Research Fellow, Université Libre de Bruxelles - Solvay Business School – Centre Emile Bernheim,
ULB, 50 Av. FD Roosevelt CP145/1, 1050 Brussels, BELGIUM, E-mail: [email protected].
D
Université Libre de Bruxelles – Solvay Business School – Centre Emile Bernheim and DULBEA, Solvay SA
Chair of Innovation, ULB, 50 Av. FD Roosevelt CP145/1, 1050 Brussels, BELGIUM and CEPR, E-mail:
[email protected].
1
1. Introduction
Patent counts are increasingly used to measure innovation performances in technico-economic
studies (see e.g., Griliches (1990) and Griliches et al. (1986)). Patents are frequently used as
an indicator of R&D output, as a vector of knowledge spillovers, as a tool to assess research
direction or strategy, and as a macroeconomic indicator of technological performance.
However, the simple patent counts provide a biased measure of the innovation performances
(see Scherer and Harhoff, 2000). On average, it is said that only one to three patents out of
one hundred yield significant financial returns. This skewed distribution of patent value is at
the origin of a small but burgeoning stream of economic research that attempts to identify the
determinants of patent value.
This paper aims to contribute to this literature. Its main objective is to put forward, develop
and validate new potential determinants of patent value. These new indicators consist of
refined traditional ones (non-patent citations, backward patent citations, co-assignees and
family size) obtained by formally identifying the institutional sources of knowledge.
In order to test the validity of the traditional and the new determinants of patent value,
academic patents are focused on. Most of the existing literature focuses on business-related
patent applications. In the present analysis, the sample is composed of 208 patent families
applied for at the EPO (European Patent Office) by six Belgian universities.
This particular focus on academic patents finds its justification in the radical change observed
in the patenting activity of Belgian universities since the early nineties. The number of patents
has exploded since the late nineties, as witnessed by the filing of 142 patent families by the
six universities, more than five times the number of patent families applied for during the late
eighties. Beyond this revolution, the motivation in analysing academic patents is induced by
the increasing role that policy-makers and corporations want academia to play in the
knowledge society (e.g., Mansfield (1991, 1995), Mansfield and Lee (1996) and Etzkowitz
(1998)).
The paper is structured as follows. Section 2 summarizes the existing literature on patent
value. The number of forward patent citations (the number of citations a patent receives from
more recent patents) is generally identified as a major determinant of patent value. It reflects
the effective use or the importance of the patented technology to new inventions. The other
most important determinants are the family size (geographical scope for protection) and
backward patent citations (the number of citations to previous patents; this illustrates the
technological state of the art on which the patented technology relies). In the light of this
literature, section 3 suggests four new sets of potential indicators of patent value. The
database and descriptive statistics are presented in section 4. Section 5 is dedicated to the
empirical implementation and defines the various models that are used to identify the value
determinants of academic patents. The empirical results are presented and interpreted in
section 6. The last section concludes and provides some policy implications.
The empirical results suggest that the new indicators related to the institutional sources of
knowledge improve the understanding of what determines the value of patents. Using more
qualitative variables instead of purely quantitative ones does indeed improve the fit. It seems
that it is not the number of backward citations that matters, but what type of institution the
patent is citing. It is not the number of co-assignees that matters, but the mix of institutions
involved in developing the patent. It is not the number of non-patent citations that matters, but
2
whose work the patent is using. It is not the number of family members that matters, but to
where you are expanding geographical protections.
2. The literature on patent value
Most of the empirical analyses attempting to better understand the factors underlying the
skewed distribution of patent value are described in table 1, and their results are summarized
in table 2.
Several empirical strategies have been used to approximate the value of a patent. They rely on
different datasets (e.g. all patent applications in a regional office; a particular sector such as
biotech; a sample of firms in a given country), they cover various time spans and use different
data sources. In addition, the functional forms of the empirical models vary from one study to
the other. Some authors rely on the monetary value of each patent (Harhoff et al., 1999, 2003)
or on the present value evaluated by experts on a value scale (Reitzig, 2003), on forward
patent citations (Lerner, 1994), on a composite indicator (Lanjouw and Schankerman, 1999),
on the probability to get a patent granted (Guellec and van Pottelsberghe, 2000), on patent
opposition and renewal data (Pakes and Simpson, 1989, Lanjouw and Schankerman, 1997),
and on whether a high-tech start-up has been created or not on the basis of the codified
invention (Shane, 2001). As a matter of fact, there are as many potential methodologies to
approximate the value of patent as the number of existing investigations.
Similarly, the type and number of explanatory variables that have been used as determinants
of patent value are heterogeneous across studies. Table 2 shows that the most frequently used
determinants are the number of forward patent citations (when it is not used as a dependent
variable), the number of backward patent citations and the geographical scope for protection
(number of countries in the patent family). Other variables rely on the concepts of opposition
procedures, renewal data1, application scope (the number of claims) and non-patent citations.
*** Insert table 1 around here ***
Despite this strong heterogeneity across studies, some similarities emerge. The most
important one is probably the fact that the number of forward patent citations (FPC) is closely
associated with the value of a patent. All studies using forward patent citations reach this
conclusion. In a similar vein, indicators of the geographical scope of protection of a patent (or
family size) are closely related to patent value (see bottom line of table 2).
Lanjouw and Schankerman (1999) develop a composite quality index significantly related to
the decisions to renewal and to oppositions. They point out that forward citations and family
size are important determinants of renewal decisions, but the number of claims and the
number of backward citations are not. In the field of Computed Tomography scanners,
Trajtenberg (1990) finds a close correlation between indicators based on forward patent
citations and independent measures of the social value of innovation in that field. Harhoff et
al. (1999) also establish for a set of German patented inventions that patents with greater
economic value are more likely to be cited in subsequent patents. Hall et al. (2000) highlight
that “citation-weighted patent stocks are more highly correlated with market value than
1
Schankerman and Pakes (1986) developed a model of patent renewal, considering that patent renewal decisions
illustrate the value of a patent protection. Within four technology groups, Lanjouw et al. (1996) also used patent
renewal to estimate a patent value index.
3
patent stocks themselves and that this fact is mainly due to the high value placed on firms that
hold very highly cited patents” [Hall et al., 2000, abstract].
A third indicator that has received substantial empirical validation is based on backward
patent citations (BPC, the number of former patents cited by the inventors). Three out of six
studies using this indicator found that it has a positive and significant impact on patent value.
*** Insert table 2 around here ***
Besides these three main indicators, the other potential determinants that have been
incorporated in patent valuation models have either led to conflicting results or received very
little empirical validation. The biggest bone of contention concerns the scope of the invention
(approximated by the number of claims and/or the number of technological classes associated
with the invention).
For instance, Lerner (1994) examines how patent scope affects the valuation of new
biotechnology firms. The author finds that an increasing scope (measured by the number of
four-digit IPC - International Patent Classification - classes into which patents are classified)
of patent protection is associated with higher valuations. Focusing on MIT’s patent
applications, Shane (2001) also demonstrates that the scope, the radicalness and the
importance of MIT’s patents have a significant and positive impact on the probability to
create new technology-based firms. However, the valuable impact of the patent scope is
challenged by many authors, including Harhoff et al. (2003), Harhoff and Reitzig (2002),
Guellec and van Pottelsberghe (2000, 2002) and Lanjouw and Schankerman (1997). Scope is
sometimes inversely correlated with patent value, at least when measured with the number of
listed technological classes.
Tong and Frame (1994) put forward the number of claims as a measure of the size of an
invention. Lanjouw and Schankerman (1997, 1999) show, however, that this indicator is
related to the probability of litigation but not to the probability of being renewed.
The indicator based on non-patent citations (NPC), or citations to the scientific literature, has
not been used extensively so far. Meyer (2000) underlines how important scientific tacit
knowledge is to inventive activity in certain fields. However, the author argues that nonpatent citations hardly represent a direct link between cited paper and citing patent.
Experimental results vary in this respect. Analysing a set of German patents applied for in
1977 and fully renewed until 1995, Harhoff et al. (2003) establish a significant positive
impact of scientific citations on the patents’ monetary value. Harhoff and Reitzig (2002),
investigating patents filed in the biotechnology and pharmaceutical fields, observe a non
significant influence of the number of non-patent citations on the probability that a patent will
be opposed (considered a value indicator).
In a nutshell, it clearly appears from the literature survey that the number of forward patent
citations is one of the most important determinants of patent value. It is closely followed by
family size and backward patent citations. The scope of an invention has an ambiguous
impact, whereas non-patent citations received little empirical support. The next section
presents new and more detailed indicators that might improve our understanding of what
determines patent value.
4
3. New indicators of patent value
The central hypothesis that we put forward is that a more precise definition of the components
of the existing value determinants might improve our understanding of how patent value is
determined. Most indicators reflect a broad source of knowledge, to some extent, non patent
citations are deemed to approximate the scientific base of an invention; backward patent
citations reflect the codified origin of the technological knowledge used for an invention; the
co-assignees indicator measures the extent to which different sources of tacit knowledge have
been merged into a single research project; and patent family size witnesses the financial
resources used to extend the intellectual property abroad.
Figure 1 illustrates how these four indicators might be improved through a more precise
‘institutional’ typology. If these indicators broadly reflect the sources of knowledge that have
been used to invent a patent, they might be improved through a formal identification of their
institutional origins. Similarly, the number of countries chosen for the international protection
phase can be disaggregated into strategic geographical choices.
*** Insert figure 1 around here ***
Scientific Knowledge (non-patent citations)
The scientific sources of knowledge, or non-patent citations, have been little investigated in
the literature, and no conclusive result has emerged. An important distinction can be made
between self and non-self citations to the scientific literature (see the Eastern quadrant of
figure 1). Self citations to the scientific literature witness an invention that is based on the
researcher’s (or the research team’s) own scientific and tacit knowledge. Non-self citations
would have a lower potential impact on the value of an invention for two main reasons. First,
it is a citation that is available to all inventors and is therefore open to worldwide competition
(scientific publications are in the public domain). Self citations also refer to the public domain
but are implicitly associated with a large potential tacit knowledge characterized by a
substantial – and recognised – experience in the field of research. It can be assumed that an
inventor who uses his or her own tacit knowledge (part of which has been published) is more
likely to translate it into a successful invention. Second, it is well known that a substantial
proportion of citations is imposed by patent examiners. These citations, although indubitably
related to the field of research, would not have any, or at most very little, direct influence on
the quality and hence the value of an invention. Self citations would always appear in patents
(few scientists or researchers would forget to cite their own work), without any intervention
from a third party.
Technological Knowledge (backward patent citations)
Backward patent citations (the Western Quadrant of figure 1), or the technological source of
knowledge, have been validated as an indicator of patent value. The inventions that cite
backward inventions are either architectural (i.e., merging different technologies) or
incremental (improving an existing invention). In the former case, one can expect a higher
value of the invention. In the latter it depends on the incremental step. A small increment
would probably be associated with a lower return (and less forward citations) than a large
increment. The only information that differentiates cited inventions relates to their
institutional origins. A cited patent can be applied for by a public research institution (the
invention comes from the scientific community), by a private firm (the invention comes from
5
an invention that might be exploited by a firm), or by the research team itself (self backward
citations).
The relationship between these three institutional sources of technological knowledge and the
quality or value of an invention (whether incremental or architectural) is far from being clearcut, a priori. It can be expected, however, that patents that cite other patents invented by
public research institutions (PRI) are more related to the scientific field and face less potential
competition on the final product market than cited patents that have been applied for by a firm
(the business sector using patents for exploitation or licensing purposes). Implicitly, self
backward citations would relate more to incremental inventions and would in turn be
associated with a lower potential value than architectural inventions.
Cooperation (co-applications of patents)
In the literature, the number of co-assignees (Southern quadrant of figure 1) has been little
investigated as a determinant of patent value. Nevertheless, it either reflects an active research
collaboration and/or contractual research and/or an independent research project, but whose
intellectual property has been applied for by one of the co-assignees in order to commercialize
or license the invention. Co-assignees can be firms or public research institutions. The former
would witness either an active collaboration in research or contractual research. The potential
effect of these co-assignees is unclear. If it is purely contractual research for the development
of an invention, one might expect a lower potential value (otherwise the research project
would be implemented by the firm to avoid any potential knowledge leakages). On the other
hand, if it is performed under a collaborative framework, where two knowledge bases and
research skills are merged, one might expect a higher impact on the value of the invention. If
the co-assignee is a public research institution, a higher potential value can be expected, as the
knowledge base of the invention would be related to greater research efforts from the
scientific sector.
Patent Protection Strategy (or patent family)
The fourth indicator of patent value is the geographic scope of protection, or family size
(Northern quadrant in figure 1). This indicator suggests that a patent that has been applied for
in several countries is associated with a much higher value than a patent applied for in only
one country or region. However, countries differ markedly in terms of both size and
technological intensity. It might therefore be expected that a patent extended to the US (by
non US applicants), for instance, is more valuable than a patent applied for only to the
Australian market. From a European viewpoint, this assumption can be tested by simply
looking at the patents that have been applied in the United States and/or Japan, by far the
largest ‘homogenous’ markets in the world.
4. The Database on academic patents
Since 1985, Belgian universities have been increasingly concerned by the protection of their
intellectual property. To assess this evolution, patent data have been collected in July 2002
through the DELPHION online database. In a first stage, the patents applied for by the 19
Belgian universities or university faculty centres2 were gathered. In a second stage the focus
2
The universities from the French-speaking community are: Ulg, UCL, ULB, UMH, FUSAGx, FUNDP, FPMs,
FUSL, FUCaM, FUL. The universities from the Flemish community: KUL, LUC, KUB, UA, RUCA, UFSIA,
UIA, UG, VUB. Acronyms are defined in Table A.1
6
was brought to the six most productive universities in terms of patent applications.3 The
database does not include patents that were invented within universities but applied for by
third parties.4
The patents were grouped in families. A patent family is the set of patents characterised by
the same priority number. To take into account only the potentially most valuable patents, we
focused on patent families with at least one patent application at the EPO (European Patent
Office). This selection criterion was preferred to the analogue choice of having at least one
patent application at the USPTO (United States Patent and Trademark Office). This choice
was made considering that EPO applications give a more complete picture of the patenting
behaviour of Belgian universities.
The six most active (in terms of patent applications) Belgian universities (KUL, UG, VUB,
UCL, ULB, Ulg) applied for 208 patent families (with a priority date ranging from 1985 to
1999) at the EPO. Among these, 78 EPO patents were also applied for at the USPTO. Only
seven other patents were applied for at USPTO without being applied for at the EPO. The
period of investigation concerns the priority dates going from 1985 to 1999. This selection is
due to the patenting process and the constraints of data collection. The choice of the upper
bound (1999) can be explained by two factors. First, there is a delay between the reporting
date of the application in a database and the application date at a patent office. Second, the
PCT (Patent Co-operation Treaty) procedure is increasingly used by firms and universities.
This procedure allows the decision to internationally extend the rights of the patent to be
delayed (delay of 30 months after the priority date). Therefore, as far as this paper focuses on
patent applications at the EPO, the extension of the date filed after 1999 creates the risk of
missing patents that are not primarily applied for at EPO but whose protection rights could be
extended to the EPO at the end of the PCT procedure. Before 1985, the patenting activities of
the Belgian universities were limited, as only 9 patents were applied for at the EPO (5 by
UCL, 1 by Ulg and 3 by KUL).
In what follows, unless specified, the use of the term “patent” will refer to a patent family
whose priority date is between 1985 and 1999 and with one member (at least) applied for at
the EPO. Table 3 presents the number of patent applications of the six Belgian universities
over three 5-year sub-periods; the late eighties, the early nineties and the late nineties; and for
the entire period ranging from 1985 to 1999.
*** Insert table 3 around here ***
There has been a spectacular growth of patent applications by Belgian universities. An
increase of 44% occurred between the late eighties (1985-1989) and the early nineties (19901994). This increase was followed by a growth of 264% between the early and the late
nineties (1995-1999). This trend has continued since 1999. The University of Brussels (ULB),
for instance, applied for more than 30 patents since the year 2000, a number larger than the
number of patents applied for over the 15 preceding years.
3
For the period 1985-1999, the KUL, UG, VUB, UCL, ULB and Ulg held 94% of the identified Belgian
academic patents.
4
Saragossi and van Pottelsberghe (2003) show that the risk of having a wrong picture of the number of
patentable inventions is very high when the focus is exclusively put on patents filed by universities. Universities
may use different names to file their patents. For instance, KUL has used many names, such as KULeuven, R&D
Leuven and REGA. These corrections have been implemented for the construction of our database. However,
independent inventors and the patents invented at universities but applied for by firms are not taken into account.
7
*** Insert table 4 around here ***
Table 4 shows the number of forward patent citations received by patents applied between
1985 and 1999. It shows that KUL is currently the university with the highest number of
patents and the highest average patent quality. Conversely, the 5-FPC index shows that
patents at KUL are cited less rapidly on average than those of other universities.5 It is clear
that the average FPC is higher for universities that started to patent their inventions earlier.
*** Insert figure 2 around here ***
Figure 2 shows that the well-known skewed distribution of patent value also holds for
academic patents. Considering that value can be approximated by the number of forward
patent citations, figure 2 shows that out of the 208 patent families included in the dataset,
about 140 were not referenced by any subsequent patent, whereas two received more than 30
forward patent citations. The next section presents the econometric model that aims to explain
this skewed value distribution of patents.
5. Empirical implementation
Two main equations are to be used in order to evaluate the value determinants of academic
patents. The first one includes the ‘traditional’ determinants and the second one decomposes
each traditional determinant into its ‘new’ institutional components. We use the number of
forward patent citations as a proxy for the economic value of patents.
The highly skewed nature of the dependent variable requires a negative binomial model
(count model). A negative binomial distribution is used because of the overdispersion of the
data.6 Individual units yi follow a Poisson regression model (with parameter Oi) with an
omitted variable ui such that exp(ui) follows a gamma distribution with mean 1 and variance
D.
yi ~ Poisson(Pi*)
Pi* exp(xi Ei ui)
exp(ui)~Gamma(1/D,1/D)
where Ei is the vector of parameters associated with the vector of explanatory variables xi.
Alpha (D) is the overdispersion parameter.
The dependent variable of our models is the number of forward patent citations (FPC)
received from patents applied in any patent office. This number was not limited to the amount
of forward citations received within the five years following the priority date because the
delay between the priority date of the university patent and the filing date of the first patent
citing the university was of about 6 years on average.
The basic econometric model is described in equation (1) :
5
More exhaustive descriptive statistics are available in Table A.2 in the appendix.
As opposed to the Poisson model, it allows the conditional mean and variance of the dependent variable to be
different.
6
8
2
4
6
15
Oˆi exp ¦m 1D mCVi m ¦ j 1E jVDi j ¦k 1J kUNIVi k ¦l 1G lYEARil H i
(1)
where Ôi is the estimator of the Poisson parameter explaining the number of forward patent
citations of each patent family [i = 1,…, 208] and Hi is the error term.
Equation (1) corresponds to the median specification used in the literature (summarized in
table 2) and includes four types of variables. The first ones are control variables (CV). They
take into account the characteristics of the patent (the number of inventors and whether it is
applied in the biotech sector or not). The second type of variables is composed of the value
determinants (VD). They are composed of non-patent citations, backward patent citations, the
number of co-assignees and the family size. The third type of variable is composed of the six
main universities’ fixed effects (UNIV). The universities might indeed have different
endowments and processes to manage knowledge transfer. The estimates also aim to test
whether, taking into account the control variables and the value determinants, a significant
university effect appears. The fourth type of variable is linked to the age of the patent. It is
composed of time dummies.
The vector of unknown parameters is >D1,D 2,E1...,E4,J 1,...,J 6,G1,...,G15 @ . CVm holds for m = 1 to 2:
INV (the number of inventors involved in the patent) and BIO (whether the patent is in the
biotech sector). VDj holds for j=1 to 4: FAM (the family size), BPC (backward patent
citations), NPC (non patent citations), and COAS (the number of co-assignees). UNIVk holds
for k=1 to 6 (the six universities included in the present analysis are UCL, ULB, Ulg, KUL,
UG, VUB; cf. table A.1 in the appendix). YEARl is the matrix of time dummies related to the
priority dates of the patent (year of first filing), which ranges from 1985 to 1999.
Table A.2 in the appendix displays the average value of each explanatory variable for the six
universities and for Belgium as a whole.7
Control variables (CV)
INV is the number of inventors listed in the patent. The second control variable is BIO. The
208 patent families were divided into two broad technological fields: biotech and nonbiotech.8 On average, the patents applied for in biotechnology are associated with higher
citation rates than in other technological fields. The control variable BIO takes the value of
one when the patent’s technological field is the biotech sector, zero otherwise.
Value determinants (VD)
The extent to which a patent is partly or fully based on scientific knowledge can be
approximated with the number of non-patent citations (NPC). In order to assess the origin of
the science-base cited in the patent, the NPC variable can be disaggregated into two variables,
as suggested in section 3: a dummy that reports whether the inventors’ scientific papers (at
least one) have been cited in the patent (self non-patent citations) and a variable for the
7
See Sapsalis and van Pottelsberghe (2003) for a more in-depth description of patent activities within the
Belgian academic sector.
8
The 4-digit IPC classification has been used to identify biotech patents. We consider that a patent is part of the
biotech field if its principal technological class is related to health, to environment, to organic chemistry or to
biology. Table A.3 reports the related field for each 4-digit IPC class that has been found in academic patents.
9
number of citations to scientific papers written by other scientists (non-self non-patent
citations).9
The second traditional value determinant is the number of backward patent citations (BPC).
This variable can be disaggregated according to the institutional origin of cited patents. They
have been clustered into three categories: the number of corporate applicants (firms or
corporate R&D centres), the number of public institutional applicants (universities, hospitals,
public research centres, state departments) and the number of self citations (citations to
previous own inventions). The basic idea with this second approach is to test to what extent
the quality and value of academic patents can be affected by the invention’s institutional
origin. Table A2 in the appendix shows that a majority of the applicants of the backward
patent citations are firms (56%). A non-negligible proportion are self-citations to the
inventors’ own previous patents (7%), whereas 28% of the BPC are made to patents invented
in public institutions.
To assess the impact of collaboration with other knowledge generating institutions, the
number of co-assignees applying with a university was computed (COAS). Only juridical
entities were looked at. In a second stage these co-applicants were disaggregated into two
categories: the corporate co-assignees (industries, corporate R&D centres) and the public coassignees (public institutions such as universities, hospitals, public research centres and state
departments).
The last ‘traditional’ determinant of patent value is the family size (FAM). It is one of the
most relevant indicators of patent value. The variable takes into account the number of patents
that are a member of the family. The alternative approach put forward in the present paper is
to disaggregate this information and to focus on two variables describing whether the patents
have also been applied for at the JPO and/or at the USPTO. These dummy variables take the
value of one if applicable, zero otherwise.
Time variables (YEAR)
Patents are not cited immediately after their filing. The older a patent is, the more forward
patent citations it might receive. This time effect is taken into account through a set of time
dummies that correspond to the priority year of each patent. It is expected that younger
patents receive proportionally less citations than older ones.
6. Empirical results
The estimation strategy is the first to estimate equation (1) with the traditional explanatory
variables. In a second stage, all value determinants are disaggregated into their various
institutional components, as described in section 3 and figure 1.
9
The available information in the database creates a potential bias: the self non-patent citations are
underestimated and non-self non-patent citations are overestimated. Only partial information is available,
because only the name of the first author is referenced in the database. These two variables may be more
relevantly seen as the number of citations to the scientific literature for which one of the inventors is the
principal author and as the number of references to papers for which none of the inventors is the first author. In
health, life and engineering sciences, the first author of a scientific paper is usually the main researcher involved
in the research described by the paper.
10
Table 5 presents the econometric results of the two models. The dependent variable is the
total number of forward patent citations received by each patent. Model 1 relies on a set of
twenty-six explanatory variables. It includes the two control variables, the four traditional
value determinants, the six universities’ fixed effects and fourteen time dummies (1985 to
1999). The subsequent model aims to improve the first model through a more detailed
approximation of the institutional sources of knowledge (NPC, BPC and COAS) and of the
patenting strategy regarding the geographic scope for protection.
The estimates of Model 1, with the traditional value determinants, yield the following results.
The number of inventors (INV) has a negative impact on patent value, contrary to the
conclusions of Guellec and van Pottelsberghe (2000, 2002). This difference might be due to
the fact that they use all patents applications at EPO (i.e., mainly corporate patents), whereas
this paper focuses exclusively on academic patents, which are generally invented by several
researchers. The second control variable, BIO, has a positive and slightly significant (at a
14% probability threshold) impact on the patent value indicator. Biotech patents seem to be
less significantly associated with forward patent citations than expected.
*** Insert table 5 around here ***
Amongst the value determinants, the number of backward patent citations and the number of
co-assignees have a positive and highly significant impact on the value indicator. Non-patent
citations (NPC) have no significant impact. This non significant result regarding references to
the scientific literature differs from the results of Harhoff et al. (2003). This is most probably
due to the fact that all academic patents have a strong tendency to rely on the scientific
literature. The simple count of non-patent citations would therefore not allow to identify the
academic patents with the highest value indicator. University inventions are supposed to
translate into more science-related patents than in the data set of Harhoff et al. (2003), which
is mainly composed of patents applied for by German business firms. This hypothesis is
validated by the results of Harhoff and Reitzig (2002), who focus on biotech and
pharmaceutical patents and observe a non-significant impact of non-patent citations on their
value indicator.
The parameter associated with family size is positive, but not significant. This is rather
unexpected if we compare our results to the existing literature. The explanation might again
be related to the fact that our database is composed exclusively of academic patents.
Universities have fewer resources for patenting applications than do large corporations. The
propensity to patent is probably lower, with a selection of inventions with a high expected
value.
Model 2 advocates for the relevance of identifying the institutional components of the
traditional determinants of patent value. The estimated overdispersion parameter (alpha)
shows that model 2 explains a substantial part of overdispersion, as it decreases significantly
from model 1 to model 2. To not account for the characteristics of knowledge sources is to
leave some of the overdispersion unaccounted for (as compared to a Poisson dispersion).
Moreover, as most of the patents bear little or no value, to explain dispersion is to explain
dispersion on the right-hand side of the distribution, where the strategic and monetary value
of a patent lies.10
10
Other specifications were performed. Poisson regressions were implemented and the results were similar.
Negative binomial models were also estimated with a Quasi-Maximum Likelihood approach; similar results
emerged.
11
The estimated parameters of model 2 suggest that the issue that matters is not the total number
of non-patent citations, but rather their origin. If a patent relies on at least one scientific paper
authored by the inventors, it is more likely to be cited (the parameter is significant at a 13%
probability threshold), whereas non-self citations to the scientific literature are associated with
a negative and highly significant parameter. In other words, the universities’ most valuable
patents are those for which the inventors master the related science-base (as witnessed by
their own publications) and decide to crystallize their tacit knowledge into technological
inventions.
A second important determinant of patent value is backward patent citations. In Model 1 it
positively correlates with patent value. Model 2 further investigates the mechanism through
which backward citations affect value, by looking at the institutional origin of the cited
patents. As suggested in the new methodological framework depicted in figure 1,
disaggregated information on backward citations allows for a better understanding of the
influence of their effect on patent value.
The estimates clearly show that patents citing patents invented by at least one similar
researcher (citations to self) are significantly less cited. The explanation might be that self
backward patent citations witness mainly an incremental invention rather than a breakthrough
invention. When the cited patents are invented by public institutions, they are a significant
and positive determinant of patent value, whereas backward patent citations to corporate
patents have no impact. These results validate our expectations; patents that cite other patents
invented by public research institutions are more related to scientific research and face less
potential competition on the final product market than cited patents that have been applied for
by a firm.
Model 1 also illustrates the positive impact of co-assignees on the number of forward patent
citations. Model 2 shows that academic patents co-applied with public institutions are more
likely to be cited.11 Co-assignation with a firm is associated with a positive but weakly
significant (at a 13% probability threshold) impact on forward patent citations. On the one
hand, co-assignation with a private firm might not reveal scientific co-operation but only the
commitments of a research contract sharing the ownership of intellectual rights between the
university (the research partner) and the industry (the financial partner). On the other hand, it
can be assumed that private partners are more aware of the financial value hidden behind an
invention. In this case, and if a disequilibrium occurs between corporate and academic
bargaining powers, one can assume that industry would try to keep full ownership of the
patent and the latter would not appear in our database (the university would not appear as an
applicant). This hypothesis might explain why collaboration with industry seems to be less
fruitful than academic collaboration.
The fourth indicator of patent value relates to family size. Model 1 shows that patent family
size is not an important determinant of patent value. However, as argued in section three,
some ‘members’ of a family might provide a more precise approximation of patent value.
This idea is validated by Model 2, which shows that an application at the JPO and/or at the
USPTO has a positive and significant impact on the probability of a patent to be referenced by
a subsequent one; especially applications at the USPTO, which is one of the most relevant
predictors of patent value (at least for Belgian academic inventions). This result is consistent
11
When the number of public institutional co-assignees is separated into two subclasses (domestic public coassignees and foreign public co-assignees), the number of national public research coassignees has the most
significant and positive impact on the number of forward citations. Proximity might explain why national cooperations seem to be more fruitful than international ones, as it allows for closer and stronger links between
researchers and can therefore induce a more active cooperation.
12
with that obtained by Guellec and van Pottelsberghe (2000, 2002). They found that amongst
all patent applications at EPO, those that had a sound geographic strategy (the designation of
certain European countries for protection) had a significantly higher probability to be granted.
In order to test for the robustness of the estimates presented in table 5, several regressions of
models 1 and 2 were performed with only five universities in the sample (dropping one
university each time). The estimated parameters were stable and similar to those presented in
table 5.
Since self backward citations have a negative and significant impact on the number of
forward citations a patent receives, one could logically wonder whether the dependent
variable should be cleaned of ‘self’ forward citations. Indeed, many inventors might prefer to
see their inventions cited more by other inventors than by themselves. To test whether relying
on the number of self FPC would affect empirical results, similar models were run with the
total number of non-self forward patent citations (instead of total FPC) as a dependent
variable. The estimates of models 1 and 2 are stable (all the parameters have the same sign
and significance), and the conclusions stay basically the same.
One might also wonder whether the recent growth in patent applications by Belgian
universities witnesses a surge in innovative activity, or whether it merely reflects a higher
propensity to patent innovations of lower quality or value. In the USA, Henderson et al.
(1995, 1998) show that patenting by universities has risen dramatically in the last 25 years
and that this increase is clearly associated with an overall increase in university attention to
commercial applications of technology. However, they also notice that “despite the
approximate doubling in the total number of patents after 1980 [Bayh-Dole Act], there is no
increase in the number of very important patents”[Henderson et al., 1995, p23] and suggest
that “their results could reflect either a change in the internal research culture of the US
universities that makes scientists and engineers get involved in more applied research with
less significant patents or the effects of entry into patenting after 1980 by institutions with
little experience and expertise in patenting”. [Mowery et al, 2002b, p74]
These last conclusions are, however, criticized by Mowery et al. (2001, 2002a, 2002b) and
Sampat et al. (2003). They argue that there is a truncation bias in Henderson’s data and
observe that universities with substantial pre-1980 patenting experience display no decline in
importance and generality of their patents after 1980. They conclude that “any changes in the
characteristics of the U.S. university patents after 1980 are due in large part to entry, rather
than to changes in university culture.” [Mowery et al., 2002b, p73] They also stress that
university patenting experience and learning aspects are key elements for applying for
important patents.
The university fixed effects presented in model 2 of table 5 seem to confirm this latter
argument. For instance, the university that has had the largest increase – as well as the oldest
history – in patenting activity is the KUL. The KUL’s fixed effect is in an intermediate
position not significantly different from those of the other five universities. We cannot
conclude, therefore, that the recent surge in patenting activity by Belgian universities
witnesses a higher propensity to patent inventions of lower value. It would rather suggest that
it reflects a higher propensity to patent inventions of high potential value.
13
7. Concluding remarks
In this paper we put forward new potential determinants of patent value and test their impact
empirically. The major improvement on ‘traditional’ indicators consists in a detailed
identification of the institutional sources of the knowledge embodied in the patented
inventions of six Belgian universities.
The scientific knowledge base of an invention, approximated by the number of non-patent
citations, can be disentangled into self and non-self citations to scientific literature. The
technological knowledge, usually approximated by the number of backward patent citations,
can be decomposed into the institutional origins of the cited patents (public research
institution, corporate research and self citations). Potential collaboration between different
institutions can be decomposed into types of collaborative institutions. Finally, an alternative
to the family size of a patent can be approximated by taking into account some of the largest
‘homogenous’ markets targeted for protection, such as the United States and Japan (from a
European viewpoint).
The patent value indicator used for the empirical analysis is approximated by the number of
forward patent citations. The estimates show that, controlling for its age and its technological
field, the most important value determinant of an academic patent is its number of backward
patent citations. As far as academic patents are concerned, the presence of a co-assignee also
seems to improve the patent value. This is probably due to an effective collaboration or an
improved selection process. Another specificity of academic patents is that the total number
of non-patent citations does not seem to be a patent value indicator. This last result might be
due to the fact that most academic patents have a strong propensity to cite the scientific
literature.
However, when traditional indicators are disentangled according to their institutional origin, a
more accurate understanding of what determines the value of patents appears. It is not so
much the number of non-patent citations that matters, but whose work the patent is using. Self
citations (the fact that the patent cites scientific papers whose first author is one of the
inventors) are a positive and significant indicator of patent value, as opposed to non-self
citations. In other words, some of the universities’ most valuable patents are those for which
the inventors attempt to crystallize their tacit knowledge into technological inventions.
Similarly, it is not the number of backward citations that matters, but what type of institution
the patent cites. Self backward patent citations indicate a lower patent value, which is
probably due to the fact that these patents witness mainly an incremental invention.
Academic patents that are co-applied for with another institution are more likely to be cited.
However, it is not the number of co-assignees that matters but the mix of institutions the
patent cites. Being involved in the co-application of a patent with a public research institution
has the most significant and positive impact on the number of forward citations. This might be
the result of active collaboration between public institutions that have a strong knowledge
base.
Similarly, it is not the number of family members that matters, but to where one expands
geographic protection. In this respect, extending the patent to the US or to Japan is a good
indicator of patent value, especially to the US which is by far the largest homogeneous market
for technology.
14
Further research is required to evaluate whether the new indicators proposed in this paper are
also valid for corporate patent applications. However, two clear policy implications regarding
academic research emerge from our results. The first one is related to the positive impact of
self citations to scientific literature. It clearly shows that when a patent is invented by a
researcher who uses his own scientific and tacit knowledge, one may expect a potentially
higher economic value. That is, the current policies aiming at fostering knowledge transfer
from universities to industry should try to stimulate academic researchers with a high
scientific profile in terms of publications. Second, the academic patents emerging from a
collaboration with public research organisations are associated with a significantly higher
value than the other academic patents. This result validates the current Belgian Federal S&T
policy that consists in fostering institutional collaboration (as well as the 6th Framework
Programme implemented by the EC Directorate General for Research at the EU-wide level).
15
References
ETZKOWITZ H. (1998), The norms of entrepreneurial science: Cognitive effect of the new
university-industry linkage, Research Policy, 27 (8), 823-833.
GRILICHES Z. (1990), Patent statistics as economic indicators: a survey, Journal of
Economic Literature, 28 (4), 1661-1707.
GRILICHES Z., NORDHAUS W. D., SCHERER F.M. (1989), Patents: recent trends and
puzzles, NBER Working Paper, N°2922, 56p.
GRILICHES Z., PAKES A., HALL B. H. (1986), The value of patents as indicators of
inventive activity, NBER Working Paper, N°2083, 43p.
GUELLEC D., VAN POTTELSBERGHE DE LA POTTERIE B. (2000), Applications, grants
and the value of patent, Economics Letters, 69(1), 109-114.
GUELLEC D., VAN POTTELSBERGHE DE LA POTTERIE B. (2002), The value of
patents and patenting strategies: countries and technology areas patterns, Economics of
Innovation and New Technology, 2002,11(2), 133-148.
HALL B., JAFFE A. TRAJTENBERG M. (2000), Market Value and patent citations: a first
look; NBER Working Paper, N°7741, 40p.
HARHOFF D., NARIN F., SCHERER F. M., VOPEL K. (1999), Citation Frequency and the
value of patented innovation, Review of Economics and Statistics, 81(3), 511-515.
HARHOFF D, REITZIG M. (2002), Determinants of opposition against EPO patent grantsthe case of biotechnology and pharmaceuticals, CEPR Discussion Paper Abstract, p 28.
HARHOFF D., SCHERER F. M., VOPEL K. (2003), Citations, Family size, Opposition and
Value of patent rights, Research Policy, 32(8), 1343-1363.
HENDERSON R., JAFFE A., TRAJTENBERG M. (1995), University as a source of
commercial technology: a detailed analysis of university patenting, 1965-1988, NBER
Working Paper, N°5068, 29p.
HENDERSON R., JAFFE A., TRAJTENBERG M. (1998), University as a source of
commercial technology: a detailed analysis of university patenting, 1965-1988, Review of
Economics and Statistics, 80(1), 119-127.
JAFFE A. (1989), Real effects of academic research, American Economic Review, 79(5), 957970.
JAFFE A. (2000), The U.S. Patent system in transition: policy innovation and the innovation
process, Research Policy, 29 (4-5), 531-557.
JAFFE A., TRAJTENBERG M. (1996), Flows of knowledge from universities and federal
labs: modelling the flow of patent citations over time and across institutional and geographic
boundaries, NBER Working Paper, N° 5712, 31p.
JAFFE A., TRAJTENBERG M., HENDERSON R. (1992), Geographic localization of
knowledge spillovers as evidenced by patent citations, NBER Working Paper, N° 3993, 42p.
16
LANJOUW J. (1993), Patent Protection: of what value and for how long?, NBER Working
Paper, N° 4475, 78p.
LANJOUW J. (1998), Patent Protection in the shadow of infringement: simulation
estimations of patent value, Review of Economic Studies, 65(4), pp 671-710.
LANJOUW J., LERNER J. (1997), The enforcement of intellectual property rights: a survey
of the empirical literature, NBER Working Paper, N° 6296, 32 p.
LANJOUW J.O., PAKES A., PUTNAM J. (1996), How to count patents and value
intellectual property: uses of patent renewal and application data, NBER Working Paper, N°
5741, 33p.
LANJOUW J., SCHANKERMAN M. (1997), Stylised facts of patent litigation: value, scope
and ownership, NBER Working Paper, N° 6297, 43p.
LANJOUW J., SCHANKERMAN M. (1999), The quality of ideas: measuring innovation
with multiple indicators, NBER Working Paper, N° 7345, 39p.
LERNER J. (1994), The importance of patent scope: an empirical analysis, RAND Journal of
Economics, 25(2), 319-332.
MANSFIELD E. (1991), Academic research and industrial innovation, Research Policy,
20(1), 1–12.
MANSFIELD E. (1995), Academic research underlying industrial innovations: sources,
characteristics, and financing. Review of Economics and Statistics, 77(1), 55–65.
MANSFIELD E., LEE J.-Y. (1996). The modern university: contributor to industrial
innovation and recipient of industrial R&D support, Research Policy, 25(7), 1047–1058.
MEYER M. (2000), Does science push technology? Patents citing scientific literature,
Research Policy, 29(3), 409-434
MOWERY D., NELSON R.R., SAMPAT B. N., ZIEDONIS A. (2001), The growth of
patenting and licensing by US universities: an assessment of the effects of the Bayh-Dole act
of 1980, Research Policy, 30(1), 99-119.
MOWERY D., ZIEDONIS A., (2001) The geographical reach of market and non-market
channels of technology transfer: comparing citations and licenses of university patents, NBER
Working Paper, N° 8568, 40p.
MOWERY D., ZIEDONIS A. (2002a), Academic patent quality and quantity before and after
the Bayh-Dole act in the United States, Research Policy, 31(3), 399-418.
MOWERY D., SAMPAT B., ZIEDONIS A. (2002b), Learning to patent: institutional
experience, learning, and the characteristics of the U.S. university patents after the Bayh-Dole
act, 1981-1992, Management Science, 48(1), 73-89
PAKES A., SIMPSON M. (1989), Patent renewal data, Brookings Papers on Economic
Activity, Microeconomics 1989, 331-401.
17
REITZIG M., (2003), What determines patent value, Insights from the semiconductor
industry, Research Policy, 32(1), 13-26
SAMPAT B., MOWERY D., ZIEDONIS A. (2003), Changes in university patent quality after
the Bayh-Dole act: a re-examination, International Journal of Industrial Organization, 21(9),
371-1390.
SAPSALIS E., VAN POTTELSBERGHE DE LA POTTERIE B. (2003), Insight into the
patenting performance of Belgian universities, Brussels Economic Review, 46(3), 37-58
SARAGOSSI S., VAN POTTELSBERGHE DE LA POTTERIE B. (2003), What patent data
reveals about universities - The case of Belgium, Journal of Technology Transfer, 28(1), 4751.
SCHANKERMAN M., PAKES A. (1986), Estimations of the value of patent rights in
European countries during the post-1950 period, The Economic Journal, 96 (384), 1052-1076.
SCHERER F., HARHOFF D. (2000), Technology policy for a world of skew-distributed
outcomes, Research Policy, 29(4-5), 559-566.
SHANE S. (2001), Technological opportunities and new firm creation, Management Science,
47(2), 205-220.
TONG X., FRAME J. (1994), Measuring national technological performance with patent
claims data, Research Policy, 23(2), 133-141.
TRAJTENBERG M. (1990), A penny for your quotes: Patent citations and the value of
innovations, The RAND Journal of Economics, 21(1), 172-187.
18
Appendices
Table A. 1: Universities’ acronyms
Acronym
FPMs
UMH
FUL
FUSL
FUNDP
FUSAGx
ULg
ULB
UCL
LUC
UA
VUB
UG
KUL
KUB
University
Faculté Polytechnique de Mons
Université Mons-Hainaut
Fondation Universitaire Luxembourgeoise
Facultés Universitaires St Louis
Facultés Universitaires Notre Dame de la Paix
Faculté Universitaire des Sciences Agronomiques de Gembloux
Université de Liège
Université Libre de Bruxelles
Université Catholique de Louvain
Limburgs Universitair Centrum
Universiteit Antwerpen (groups the 4 Antwerpen academic institutions UA, UIA, UFSIA, RUCA)
Vrije Universiteit Brussel
Universiteit Gent
Katholiek Universiteit Leuven
Katholiek Universiteit Brussel
19
Table A. 2 : Descriptive statistics on indicators of patent value by university
BELGIUM KUL
UG
VUB
UCL
ULB
Ulg
Forward patent citations (FPC)
% of patents cited at least once (%)
32.2
34.5
13.8
36.8
40.0
32.1
33.3
% of patents cited at least once within 5 year (%)
11.1
8.0
6.9
26.3
13.3
7.1
20.0
% of self FPC
13.7
18.7
9.1
8.7
1.2
5.4
0.0
Average number of FPC per patent
2.1
3.3
0.3
1.1
2.4
1.3
1.3
Average number of non- self FPC per patent
2.0
2.5
0.3
0.9
2.3
1.3
1.3
Average number of FPC per cited patent
6.6
9.4
2.5
2.9
5.9
4.1
4.0
Average number of assignees of FPC
1.2
1.2
1.1
1.2
1.1
1.0
1.0
% of corporate assignees of FPC (%)
53.3
49.9
72.7
60.9
46.9
70.3
90.0
Number of EPO patent
208
87
29
19
30
28
15
Average priority date
1995
1995
1997
1996
1993
1995
1997
Average number of inventors by EPO patent
3.1
3.1
2.5
3.2
3.3
3.6
3.2
% of biotechnological patents (%)
74.5
86.2
55.2
52.6
70.0
78.6
73.3
Average number of NPC per patent
3.3
4.3
1.0
3.4
2.7
2.3
4.9
% of patent with self NPC (%)
11.5
17.2
3.4
10.5
3.6
6.7
11.5
% of Self NPC (%)
12.9
18.4
3.3
10.9
3.7
4.6
8.1
Average number of BPC per patent
2.71
2.77
1.62
3.89
2.77
1.86
4.40
Average number of assignees of BPC
1.1
1.1
1.0
1.0
1.0
1.0
1.0
% of corporate assignees of BPC (%)
55.9
45.0
71.4
63.6
49.4
78.4
70.6
% of public assignees of BPC(%)
27.9
30.6
16.3
29.9
37.9
13.7
20.6
% of self BPC(%)
6.6
10.7
4.1
1.3
5.7
5.9
0.0
% of alone individual/unknown assignees (%)
9.6
13.7
8.2
5.2
6.9
2.0
8.8
Average number of co-assignees per patent
0.43
0.44
0.52
0.53
0.33
0.36
0.47
% patent applied alone (%)
65.4
63.2
58.6
57.9
66.7
82.1
66.7
% of corporate co-assignee (%)
31.1
23.7
33.3
20.0
30.0
40.0
71.4
% of public co-assignees (%)
68.9
76.3
66.7
80.0
70.0
60.0
28.6
Average number of family members
5.9
7.5
3.7
6.4
5.8
3.8
4.8
% of patents applied at EPO (%)
100
100
100
100
100
100
100
% of patents applied at JPO (%)
43.8
43.7
24.1
57.9
60.0
39.3
40.0
% of patents applied at USPTO (%)
37.5
39.1
20.7
52.6
40.0
32.1
Sources: own calculations, Delphion database - 208 EPO patent families, priority date: 1985-1999
46.7
Control variables
Non-patent citations (NPC)
Backward patent citations (BPC)
Co-assignees
Family size
20
Table A. 3 : IPC Codes classified in biotechnological or non-biotech scientific fields
IPC principal BIOIPC principal
BIOIPC Code
BIOCode
NBIO
Code
NBIO
principal
NBIO
*
A01H
B
C01F
N
F21V
N
A01K
B
C04B
N
G01B
N
A01M
B
C07B
B
G01F
B
A23K
B
C07C
B
G01J
N
A23L
B
C07D
B
G01L
N
A61B
B
C07F
B
G01N
B/N**
A61F
B
C07H
B
G02B
N
A61K
B
C07K
B
G06F
N
A61L
B
C08F
B
G06T
N
A61M
B
C08G
B
G10K
N
A61N
B
C08L
B
G21K
N
B01D
B
C10L
N
H01L
N
B01F
N
C11B
B
H01Q
N
B01J
N
C12N
B
H01S
N
B09C
B
C12P
B
H03F
N
B26F
N
C12Q
B
H03K
N
B29C
N
C22B
N
H04B
N
B29D
N
C23C
N
H04J
N
B31D
N
C25D
N
H04L
N
B32B
N
D03D
N
H05H
N
B62D
N
*
B : biotechnology field; N: non biotechnology field
**
The IPC class G01N is a class linked to instrumentation apparatus related to the investigation or
the analysis of materials by determining their chemical or physical properties. We have
considered here that the patent within this classification could be a biotech patent or not. To
determine in which field it should occur. We have looked at the others listed IPC codes. If they
were biotechnology classes (as defined in the table). the patent was considered to be applied in the
biotech field. Otherwise. the patent was considered to be a non biotech patent.
21
Table 1 : Literature on patent value
Field
Sample
SCHANKERMAN
and PAKES (1986)
Authors
Post 1950 period,
UK, German and French patents
778
TRAJTENBERG (1990)
Computed Tomography
(1972-1984)
456
Model
- Stochastic model
Dep. Variable
- Patent renewal
- Patent citations may be indicative
of the value of innovation
- Close association between citationbased patent indices and
independent measures of the social
value of innovation
LERNER (1994)
Biotechnology (1973-92),
USPTO patents
1678 - PROBIT
- # Citations
LANJOUW J. et al. (1996)
Precedent data and
results of the authors
LANJOUW and
SCHANKERMAN (1997)
USPTO 1975-1991
5452 - PROBIT
Probability of infringement and
challenge suits
LANJOUW (1998)
Computers, textiles,
combustion engines
and pharmaceuticals
West German patents
(1955-88)
20000 -Dynamic stochastic
discrete choice model
- Renewal decision
HARHOFF et al. (1999)
German and US patents
(1977 expiring 1995)
994
LANJOUW and
SCHANKERMAN (1999)
Pharmaceutical; chemicals;
electronic; mechanical
US patents (1960-91)
8000 -Latent variable Model - Composite index
for a composite measure - Renewal
of quality
- Litigation
- # Years a patent is renewed and
size protection are indicators of
patent value
- OLS regression
- Negative binomial
- Forward citations
- PROBIT for probability
of renewal and litigation
GUELLEC,
van POTTELSBERGHE (2000)
EPO 1985-1992
23487 - PROBIT
- Grant of a patent
SHANE (2001)
MIT patents 1980-1996
8420 - Cox regression
- Creation of new firm
HARHOFF and
REITZIG (2002)
Biotechnology
and pharmaceutics
EPO (1979-1996)
13389 - PROBIT
- Opposition
HARHOFF et al. (2003)
German patents
(1977 expiring 1995 )
772
- Ordered PROBIT
- Monetary patent value
REITZIG (2003)
Patents from a semiconductor
company
127
- Ordered PROBIT
- Present value (scale)
22
Table 2: Determinants of patent value
Patent Value
Authors
PV
QI
Patenting Procedure
SCHANKERMAN
and PAKES (1986)
D
LANJOUW and
SCHANKEMAN (1997)
D
+
LANJOUW (1998)
Others
Time
Owner
char.
Others
+
+
+
+
*
+
+
+
-
D°
+
LANJOUW and
SCHANKERMAN (1999)
D
D
LANJOUW and
SCHANKEMAN (1999)
D
LANJOUW and
SCHANKERMAN (1999)
D
GUELLEC ,
van POTTELSBERGHE
(2000, 2002)
*
+
+
+
+
*
+
/
/
+
*
+
/
+
+
*
D
SHANE (2001)
-
D
HARHOFF and
REITZIG (2002)
HARHOFF et al. (2003)
IPC.
Class
D°
LERNER (1994)
HARHOFF et al. (1999)
Patent characteristics
OP AP GP RP FC FPC BPC NPC Claims Scope Size
D
D
REITZIG (2003)
D°°
CONCLUSIONS
+
+
+
+
+
+
+
*
+
/
/
-
+
+
+
+
/
+
+
*
*
*
*
+
+
+
+
(+)
(+)
+
(+/-)
+
*
+
*
*
*
PV: Monetary patent Value or present patent value; QI : quality index; OP: opposition/ litigation procedure; AP annulment procedure; RP
Renewal data; GP: Granting of patent application; FC : Firm creation. FPC: Forward Patent citations; BPC : Backward Patent citations; NPC:
Non Patent citations. The signs inside the table are D: Dependent variable; +: Positive and significant impact of the explanatory variable on
the dependant one; - :Negative and significant impact of the explanatory variable on the dependant one; / :No significant impact of the
explanatory variable on the dependant one. * :The variable description hides a set of manifold variables. Among those, some may have a
positive or a negative impact and others may have no significant impact on the dependent variable. ° :Model of patent renewal decision is
constructed around the following variables: legal fees, renewal fees, annual return and expected future value. °°: The model relies on
variables like the importance of the patent for current and future technical developments; the difficulty to invent around the patent and to
prove its infringement; the learning value for competitors, the number of competitor and the fact that the patent is the basis for other ones.
These variables have been evaluated by experts on Likert scales.
23
Figure 1 : Institutional sources of knowledge and patenting strategy.1
Patent protection
New determinants
- Protection at USPTO (+)
- Protection at JPO (+)
Patent Family Size
(# Family members)
Technical Knowledge
Backward Patent
Citations
(# Citations)
New determinants
- Industry (?)
- PRI (+)
- Self citations (-)
Scientific Knowledge
Determinants
of
patent value
Non-Patent
Citations
(# Citations)
New determinants
- Self NPC citations (+)
- Non-self NPC citations (-)
Cooperation
Co-Assignees
New determinants
(# co-assignees)
- Industry (?)
- PRI (+)
1. The sign between parentheses indicates the expected relative impact of each characteristic associated with the
main indicator (relative to the average impact of the main indicator). PRI = public research institution.
24
Table 3 : Descriptive statistics of the number of EPO patent applications by six Belgian universities1
Number of EPO patents
UCL
ULB
1985-1989
9
3
1990-1994
7
7
1995-1999
14
18
1985-1999
30
28
1. Sources: Delphion website and own calculations.
ULg
0
2
13
15
KUL
14
18
55
87
UG
1
3
25
29
VUB
0
2
17
19
TOTAL
27
39
142
208
Table 4 : Forward patent citations of Belgian universities (priority date: 1985-1999)
Forward patent citations
UCL
30
2.37
ULB
28
1.32
ULg
15
1.33
KUL
87
3.25
UG
29
0.34
VUB
19
1.05
TOTAL
208
2.12
Number of patents
Average number of
FPC per application1
Average number of
0.33
0.25
0.27
0.09
0.10
0.74
0.22
5-FPC per application2
1
FPC: Forward patent citations. 2 5-FPC: Forward patent citations received within the 5 years after priority date.
Sources: Delphion website and own calculations, Priority date: 1985-1999
Figure 2 : Distribution of forward patent citations to Belgian academic EPO patents
160
# EPO patents
140
120
100
80
60
40
20
0
0
1-9
10-19
20-29
30-35
# forward citations
Source: Number of forward citations, as of July 2002, own calculation, and Delphion database.
25
Table 5 : Econometric results
Negative binomial model
Explanatory variables
Control variables
# Inventors
Biotechnological field
Non-patent citations
# Non-patent citations (Total)
Self non-patent citation (dummy)
#Non-self non-patent citations
Backward patent citations
# Backward patent citations (Total)
# Corporate applicants of cited patents
# Public applicants of cited patents
# Self patent backward citations
Co-assignees
# Co-assignees (Total)
# Corporate co-assignees
# Public co-assignees
Family size
Family size (# different applications)
Application at JPO (dummy)
Application at USPTO (dummy)
University characteristics
UCL (equal 1 if UCL is the applicant)
ULB (equal 1 if ULB is the applicant)
ULG (equal 1 if ULG is the applicant)
KUL (equal 1 if KUL is the applicant)
UG (equal 1 if UG is the applicant)
VUB (equal 1 if VUB is the applicant)
Ln(Alpha)
Wald F² Test
Log Likelihood
Dependent variable: number of forward patent citations
Model 1
Std. Err.
Model 2
Std. Err.
-0.152*
0.639°
0.082
0.431
0.013
0.017
0.091***
0.447***
0.020
-0.190**
0.400
0.078
0.380
0.591°
-0.067***
0.393
0.025
-0.039
0.349***
-0.310*
0.047
0.095
0.186
0.544°
1.097***
0.359
0.278
0.524*
1.246***
0.315
0.361
0.030
0.167
0.028
1.203°
1.307
1.679°
0.916
0.929
2.227**
0.830
0.931
1.082
0.938
1.029
1.057
-0.289
0.076
0.709
-0.365
-0.635
-0.088
0.806
0.828
1.007
0.791
0.918
0.796
0.029
190
-222
0.243
-0.483*
286.5
-205
0.278
Levels of significance (probability threshold): ° <15%; * < 10%; ** <5% ; *** < 1%
Number of observations: 208 patent families applied
between 1985-1999 at EPO. The two models include 14 time dummies.
26