On the E/ects of Market Incumbency on Innovative Behaviour in the Pharmaceutical Industry Maria Pluvia ZUNIGA and Gautier DUFLOS y University of PanthØon-Sorbonne, TEAM & CNRS z March 15, 2005 Abstract This paper evaluates the e/ects of market dominance on rmspropensity to innovate and productivity in a panel of 77 pharmaceutical rms for the period 1975-1991. While a higher market-share leads to a weaker propensity to innovate, incumbents rms are found to be persistent innovators and notably, persistent sources of quality-patenting. Neverthe- less, incumbent rms are subject to decreasing R&D productivity likely related to possible diseconomies of scale and organizational inertia. Keywords: Market Incumbency, market share, innovation, Pharmaceutical Industry. JEL: L25, L31, O31, O33. University of Paris1, PanthØon-Sorbonne, TEAM-CNRS (ESA 8059). y University of Paris1, PanthØon-Sorbonne, TEAM-CNRS and Paris-HEC. z Maison des Sciences Economiques, 106-112 Bd de lHpital 75647 Paris cedex 13 FRANCE. O/. +33(0)1.44.07.82.66 - Fax. +33(0)1.44.07.82.67 Mail: [email protected] (correspond- ing author). 1
35
Embed
On the E⁄ects of Market Incumbency on Innovative Behaviour ... I/I.B/ZUNIGA.pdfOn the E⁄ects of Market Incumbency on Innovative Behaviour in the Pharmaceutical Industry ... Although
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
On the E¤ects of Market Incumbency on Innovative Behaviour
in the Pharmaceutical Industry
Maria Pluvia ZUNIGA�and Gautier DUFLOSy
University of Panthéon-Sorbonne, TEAM & CNRS z
March 15, 2005
Abstract
This paper evaluates the e¤ects of market dominance on �rms�propensity to innovate
and productivity in a panel of 77 pharmaceutical �rms for the period 1975-1991. While a
higher market-share leads to a weaker propensity to innovate, incumbents �rms are found
to be persistent innovators and notably, persistent sources of quality-patenting. Neverthe-
less, incumbent �rms are subject to decreasing R&D productivity likely related to possible
�University of Paris1, Panthéon-Sorbonne, TEAM-CNRS (ESA 8059).yUniversity of Paris1, Panthéon-Sorbonne, TEAM-CNRS and Paris-HEC.zMaison des Sciences Economiques, 106-112 Bd de l�Hôpital 75647 Paris cedex 13 FRANCE. O¤.
Since the seminal work of Schumpeter (1939; 1942), scholars have been interested in identifying
what kind of market structure promotes rapid technological progress. Schumpeter advanced
(1942) the now familiar hypothesis that large �rms with market power accelerate the rate of
innovation: larger-dominant �rms possess superior economic and technological advantages al-
lowing them to dominate an industry through the process of creative destruction. However, this
statement has proved di¢ cult to verify empirically by the intensive literature.1 Our perceptions
about the role of market power on �rms�incentives to innovate and the degree of competition
necessary to ensure persistent innovation in industries remain still inconclusive. Are dominant
�rms powerful and persistent engines of technological progress? Do market power and incum-
bency in�uence in the same way innovative e¤ort? More precisely, how do incumbent �rms
persistently innovate?
It has been shown that in high-technology industries, entrants are more likely to introduce
radical innovations (Henderson, 1993; Lerner, 1997). This phenomenon seems to predominate in
the Pharmaceutical Industry particularly since the mid 1980s with the emergence of molecular
biology (Henderson and Cockburn, 1996; 1998). As documented previously (Arora and Gam-
bardella, 1990; Orsenigo, 1998; Orsenigo et al., 2001), new biotechnology �rms are frequently
source of technological breakthroughs. Nevertheless, some established dominant pharmaceutical
�rms appear also as persistence creators or developers of important innovations. Incumbent
�rms have sustained high levels of growth and R&D expenditures by means of in-house inno-
vation, technological and therapeutic market specialization (Achilladelis and Antonakis, 2000).
Although their technological competences may erode over time, face to the process of creative-
destruction, incumbent �rms may adapt to technological change by relying on their downstream
competences and incurring in cooperation with entrants or through mergers and acquisitions of
companies within and outside the pharmaceutical industry (Orsenigo, 1991; Stern, 1995; Lerner
and Merges, 1998; Gans and Stern, 2001). The determinants or forces leading to persistent
technological performance by these incumbent �rms remain however not clear.2
The distinction between incremental and radical innovation constitutes the basis for un-
derstanding the response of large-established �rms to innovation. The incentives to innovate
for incumbent and entrant �rms may di¤er depending upon the nature of innovation. On the
one hand, it has been argued that dominant �rms have stronger incentives to conduct radical
innovation (Schumpeter, 1942; Gilbert and Newbery, 1982; Bud et al.� 1993) because of the
e¢ ciency-e¤ect related to �rms�ex-ante market power. Financial considerations also support
Schumpeter�s proposition: �rms possessing some market power are more able to use internal
(from past innovation successes) and external funds to support their innovative activity, and
better access quali�ed-labour market and science-networks (Cockburn and Henderson, 1998).
On the other hand the replacement-e¤ect facing a monopolist, states that market power �rms
1Comprehensive surveys on the topic are Kamien and Schwartz (1982); Baldwin and Scott (1987); Levin andCohen (1989); Scherer (1992) and Symeonidis (1996).
2 .As technological-based competences by incumbents can depreciate over time, incumbent �rms may rely ontheir downstream chain value capabilites to cooperate with entrants and bring new discoveries to the market(Teece, 1986; Raethormel, 2001).
2
will be declined to win a stochastic patent race too quickly as her existing stream of rents will
be displaced by an entirely new stream of rents (Arrow, 1962; Reinganum, 1982; 1983). R&D
ine¤ectiveness by incumbent �rms to bring radical innovations is also addressed by agency the-
ories. De�ciencies in R&D management (R&D monitoring, projects�choice, coordination) may
lead to a preference incongruity between principal and agent in innovation production deterring
�rms�innovation performance (Berle and Means, 1932; Holmstrom, 1989).3
In many ways, the empirical research to date con�rms the basic thrust of the theoretical
literature: the incentive to bring radical innovation persistently by incumbents is rather un-
certain. While it may be to some extent correlated to incumbency, market power has also an
ambiguous in�uence on �rms�incentives to persistently innovate.4This paper evaluates empiri-
cally these counter-acting views on incumbent �rms�incentives to innovate and their repercus-
sions on �rms� industrial performance in the Pharmaceutical Industry. Our sample concerns
a panel of 77 pharmaceutical �rms for the period 1975-1991. Our results suggest that market
incumbency and market power in�uence in a di¤erent way �rms�innovative behavior. Above
all, incumbent �rms in the pharmaceutical industry show a persistent propensity to introduce
quality-patenting. While a higher market-share seems to deter innovative e¤orts, incumbents
are persistent innovators and responsible of important (leader) patents. Nevertheless, incumbent
�rms are subject to decreasing R&D productivity likely related to possible diseconomies of scale
and organizational inertia.
The study is organized as follows. Section 2 makes a brief review of the literature. Section
3 presents an overview of innovation trends in the Pharmaceutical Industry and describes our
data and methodology. Section 4 presents our empirical models on innovation and productivity.
Section 5 discusses our results and the last section concludes.
2 Theoretical background
An extensive theoretical literature on the relationship between market structure and innovation
contends that incumbent �rms often face di¤erent incentives for innovation than entrants or
small �rms.
A traditional interpretation of the innovation-marked dominance or incumbency correlation
is that concerning �nancial aspects. Firms possessing some market power are more able to use
more internal and external funds to support their innovative activity as well as to better access
quali�ed-labour market (Bhattacharya and Ritter, 1985). Failures in �nancial markets force
�rms to rely on their own supra-normal pro�t to �nance R&D activities. In addition, given
the imperfectness of markets (e.g. asymmetric information, moral hazard), incumbent �rms
may enjoy some reputation which enable them to better access and negotiate �nancial sourcing
(ibid). In the Schumpeterian-perspective, large incumbent �rms may enjoy R&D advantages
3Lastly, organizational theories suggest that incumbent �rms are more imcompetent to answer to technologicalbreakthroughs than entrants given some R&D diseconomies of scale (Arrow, 1974; Nelson and Winter, 1982) andincumbent�s organizational inertia (Mintzberg, 1979; Thusman et Anderson, 1986)
4Scherer (1967, 1980), and recently by Aghion et al (2001) or Aguion et al (2002), suggests that competitionmay have positive e¤ect on innovation up to an in�exion point after which its e¤ect decreases.
3
in the production of technology (Schumpeter, 1942; Galbraith, 1952) that facilitates their tech-
nological leadership : �rms�technological experience allow �rms the building of economies of
scale and scope related to R&D production and R&D diversi�cation (Nelson, 1959; Chandler,
1990; Cockburn et Henderson, 1998), a larger spreading of �xed costs over more R&D output
(Teece, 1980, Cohen and Levin, 1989; Cohen et Klepper, 1996) as well as a better appropriation
of external knowledge spillovers (Henderson and Cockburn, 1996).5
In interpreting the e¤ects of market share, other authors have considered the strategic in-
centives arising from current and expected product market power. The distinction between
incremental and radical innovation (or how easily technologies are imitated) constitutes the
basis for understanding the response of large-established �rms to innovation (Henderson, 1993).
A �rst view on this question is that of Gilbert and Newbery (1986). Accordingly, incumbent
monopolists have greater incentives to conduct innovation than potential entrants because total
industry pro�ts decrease when more �rms share the market. This e¢ ciency e¤ect is internalized
by the incumbent but ignored by the entrant. Thus the monopolists will tend to innovate �rst
and industry evolution will be characterized by persistent dominance (Budd et al., 1993; Kamien
and Swartz, 1972). Further, since incumbency may imply better knowledge of market conditions
and identi�cation of the best submarkets to enter, dominant are better positioned to be �rst
innovators (Adner and Levinthal 2002).
Another view is that of Arrow (1962) and Reinganum (1982) or Scherer and Ross (1990)
who model cases where incumbent �rs recognize a radical innovation but �nd it too costly or
risky to pursue and, consequently, ignore the innovation altogether or attempt to thwart the
innovators through blocking-entry strategies such as pre-emptive patenting (Farrell and Saloner,
1986; Gilbert and Newberry, 1982). Under technological uncertainty; incumbent monopolists
will rationally invest less in radical innovation than entrants will for fear of cannibalizing the
stream of rents from their existing products (Reinganum, 1982; 1983). When the competitive-
ness of new innovators or the actual value of the innovation is misestimated and innovations
are radical, incumbents may be displaced by entrants6 Dominant �rms have stronger (fewer)
incentives to invest in incremental innovation (radical) than entrants do, if all �rms have similar
(heterogeneous) research capabilities. Incumbent �rms�motivation to conduct radical innova-
tion will depend on the the size of the revenues related to the existing innovations and the degree
of heterogeneity across �rms�R&D capabilities (Klette, 1996; Klette and Griliches, 1998).7
Subsequent research has established that whether incumbent monopolists or entrants have
greater incentives to invest in innovation is a function of the degree to which innovation destroys
5Others are: managerial competences (logistic, distribution, marketing, etc.), experience in the production ofinnovation (Chandler, 1991; Freeman, 1986), superior access to capital and skilled labour markets (Schumpeter,1945, Galbraith, 1952), greater abilities to appropriate innovation (legal experience, an larger budget on legalpractice, etc.).
6Gilbert and Newbery�s �ndings are only valid in the case of incremental innovation and when these ones arenot related to incumbents�R&D. Incumbent �rms with market power will retain their dominant position in anindustry whose technology evolves incrementally.
7Some empirical studies have shown that persistent R&D di¤erences (e.g. dynamic increasing returns in R&D)seems to explain the distribution of market shares across companies (Gerosky et al, 1997; Blundell et al, 1999).
4
existing market-power and of the uncertainty surrounding the innovative process (Gilbert and
Newbery, 1984; Reinganum, 1984; Bresnahan, 1985). The in�uence of these elements can �nally
in�uence di¤erently the innovation behaviour according to the maturity of industries and the
degree of obsolescence of technologies (Klette, 1997).8
Although the Schumpeter hypothesis has come in for some hard empirical knocks, at least
one strand of the literature suggests that it should not be rejected altogether. Theory shows that
market power can stimulate technological progress by market leaders when these ones posses
a superior technological capacity relative to entrants. Segerstrom and Woniereck (1999) or
Hörner (2001) show that if the technological di¤erence between two competitors is su¢ ciently
signi�cant, the company in dominant technological position has stronger incentives to continue
its e¤orts in order to maintain its technological leadership.9 Further, large dominant �rms may
bene�t from dynamic increasing returns in the production of innovations in the form of learning-
by-doing or learning-to-learn, which may lead to strong innovation persistence (Chandler, 1990;
Freeman, 1987). In emphasizing the importance of complementary assets when commercializing
an innovation, Teece (1986) views the fully integrated incumbent as the �rm best positioned to
bene�t from innovation. Though radical technological change generally leads to the deprecia-
tion of incumbents�technological competences, the non-technological assets (e.g. downstream
competences) of incumbent �rms may become more valuable, notably when they are specialized
with respect to commercializing the new technology (Leonard-Barton, 1992).
However, a growing body of work suggest that in certain industries, radical innovations are
actually pursued by market entrants to the detriment of incumbent �rms that presently hold
market power yet lack competence in this innovation (Foster, 1986; Henderson, 1993; Tushman
and Anderson, 1986; Henderson, 1993; Lerner, 1997). The ability to �free-ride�on �rst-mover
investments as well as technological and market uncertainty make it di¢ cult for the incumbent
to adapt to environmental change and may lead a �rm to delay entry into a new sub�eld
(Lieberman and Montgomery 1988).10
The organizational literature provides a number of theories with di¤erent implications for
incumbent innovation. One explanation argues that incumbents may attempt to produce the
innovation but are less successful than challengers due to inertia (bureaucratic process, higher
costs of processing routine information across departments) associated with organizational age
and/or size (Hannan and Freeman, 1989; Miller and Chen, 1994). Another explanation notes
that new knowledge associated with the innovation may render an incumbent obsolete and thus
incompetent in its production (Henderson, 1993; Henderson and Clark, 1990; Tushman and
8For instance, in emergent industries marked by a great turbulence, technology changes very quickly, uncer-tainty is signi�cant and the barriers at the entry are weak (Abernaty and Utterback, 1978; Klepper, 1997). Marketdominance is transistory and the technological advantages of innovating �rms then can be only temporary sincethe innovator in place soon is caught up by a competitor.
9On the other hand, if the di¤erence between the competitors is not too signi�cant, the technological followercan have interest to accentuate his innovating e¤orts in order to prevent further technological lagging (Hörner,2001).10 If a product can be easily imitated, Katz and Shapiro (1987) argue that a dominant �rm is better o¤ by being
a follower into a new submarket. The delay permits the dominant �rm to imitate the �rst mover�s product andavoids having its own products imitated.
5
Anderson, 1986). Lastly, several authors have shown that insulation from competitive pressures
discourages innovation and growth, while refutations of Gibrat�s Law imply that smaller, not
larger �rms tend to innovate more than proportionately to their size (Klette and Griliches, 1998).
2.1 Empirical Evidence
The empirical evidence on the in�uence of market dominance on innovation remains inconclu-
sive. They point out however, that this relationship depends upon industry-speci�c factors.11
First cross-sectional studies of the relationship between �rm size, market power and innovative
activity have in general found no systematic relationship (Baldwin and Scott, 1987; Cohen and
Levin, 1989). Some works such as Geroski (1995), Blundell, Gri¢ th and Van Reenen (1995),
Nickell (1996), Levin et al., (1985), and Cohen et al., (1987) have found disproportionate evi-
dence against the Schumpeterian hypothesis. In the opposite, Acs and Audretsch (1987) suggest
that concentration may stimulate innovation in industries highly innovative. Largest �rms have
a superior advantage in the concentrated industries whereas small and medium enterprises may
be more productive in innovation in less concentrated industries. The works by Scherer (1967,
1980), and Aghion et al., (2001) and Aguion et al., (2002), market product competition may
have a positive e¤ect on innovation up to an in�exion point after which its e¤ect decreases. Ac-
cordingly to Aghion et al., (2001), more market competition may increase the incremental pro�ts
from innovation (as it may reduce �rms�pre-innovation rents by more than it reduces its post-
innovations rents), and thereby encourages R&D investment aimed at �escaping competition�.12
Blundell et al., (1993) �nd for a panel of U.K. �rms that, while higher market share �rms
innovate more, �rms in competitive industries tend to have a greater pro�tability of innovat-
ing. Blundell et al., (1999) found a robust and positive e¤ect of market share on observable
headcounts of innovations and patents in the British Industry, when unobserved heterogeneity
is controlled for through the use of pre- entry innovation stocks.13 In the opposite, Crepon et
al., (1995) report weak evidence of Schumpeterian e¤ects on innovation in the French Industry
: they are found to be far less important than the sectoral ones (i.e. product imitation, techno-
logical opportunities and industry-demand pull).14 In a sample of early-twentieth-century U.S.
�rms experiencing disrupted market shares produced fewer patent grants on average, while they
also encountered an unfavourable stock market response to their product market performance.
Furthermore, �rms with strong market positions tended to have higher patent counts especially
11Scherer (1980) have shown that the e¤ect of market concentration on R&D investment depends on thedegree of technological opportunity in the industry: higher concentration in low (high) technological opportunitiesincreasing (decreasing) R&D e¤orts.12This "escape competition" e¤ect dominates for low initial levels of competition, whereas the Schumpeterian
e¤ect dominates at higher levels of competition. The higher the average degree of �neck-and-neckness� in anindustry and that is, the degree in which oligopolistic �rms face more similar production costs, the steeper theinverted-U relationship between product market competition and innovation in that industry.13Though, they also report that increased market competition tends to stimulate innovative activity: while
industry concentration is negatively associated with the number of innovations, import penetration is positivelyassociated with innovation counts.14Although they con�rm that innovation increases proportionally with �rm size but no e¤ect is detected from
concentration on innovation activity.
6
in the citation-weighted and scope-weighted (claims) patent production function.
The empirical work evaluating explicitly the role of incumbency on �rms�propensities to
introduce radical innovation arrives also to mixed conclusions. Henderson (1993) and Lerner
(1997) both provide evidence of weak persistence by incumbents, and therefore a constant re-
newal of technological leadership in the American Photolithographic Alignement Equipement
industry and the Disk-Driver industry, respectively Likewise, Cockburn and Henderson (1994)
found no evidence of technology racing behavior (�winner takes all�) in the case case of research
in the Pharmaceutical Industry.
Henderson�s �ndings (1993) show that established �rms invest more in incremental innova-
tion than entering �rms.15 While incumbency seems not to explain larger market shares (Gilbert
and Newbery�s rejection), incumbent �rms are found to have a smaller share of the market for
new products incorporating innovations that were radical in a technological organizational sense.
Lerner (1997) shows for the disk drive industry, that companies who are close-followers to the
leader innovate more, which implies a constant replacement of innovators.16 Lastly, Cockburn
and Henderson (1994) show for a sample of pharmaceutical �rms that the possibility of multi-
ple prizes or multiple patents to innovation makes R&D-rivalry insigni�cant to explain �rms�
innovative e¤ort. Hence there is no evidece of technology-racing a¤ecting �rms� innovative
performance. The discovery of a treatment seems to make easier, for the competitors, to dis-
cover another. Furthermore, spillover between �rms both from internal and external rival R&D
projects, are found to play a major role in increasing research productivity.
The di¤erent views presented in the literature suggest that �rms� innovative behaivor is
a¤ected by co-existing opposing e¤ects related to market dominance and incumbency. Market
dominance and market-structure may actually in�uence in a di¤erent way innovation incentives.
Furthermore, in some industries incumbent �rms could be persistent innovators in the long
run, whereas entrants may be sources of technological breakthrough but with frequently a short
life. One important shortcoming in most of the empirical works (excepting Blundell et al.,
1995; 1999; Duguet et al., 1995) is their aversion to deal with endogeneity arising from �rms�
market power (or market structure) on innovation. In a Schumpeterien economy, �rms gain
market share by innovating but lose it through creative destruction. A second problem relates
to unobserved heterogeneity, deriving, for example, from the di¤erent technological opportunities
and appropriability conditions facing �rms. As explained in the previous section, permanent
technological heterogeneity across �rms contributes to explain di¤erences in innovative behavior.
Lastly, few empirical works consider the role of knowledge dynamic-e¤ects and deal thereby with
the auto-correlation problems imposed by pre-determined variables (Crepon and Duguet, 1997;
Cincera, 1998; Blundell et al., 1995).
15 Investment in incremental R&D appears as 75% superior for old-dominant �rms and even larger when con-sidering organizational innovations (89%).16 Indeed, greatest technical progress is made by the immediate follower - those having e¤ectiveness of 25 and
74% - of the product leader.
7
2.2 The Pharmaceutical Industry
To study incumbency and innovative behavior in this sector is particularly appealing for several
reasons. Firstly, the nature of knowledge is unique in the Pharmaceutical Industry. Prior
research shows that a strong link exists between research and development, namely basic research
and innovation in the pharmaceutical and biotechnology industries. For �rms that develop new
drugs, investing in basic research brings tangible bene�ts. In the US, measured as a percentage
of US sales, R&D expenditures rose from less than 12% in 1970 to over 20% in the late 1990s
(see Fig. 1). In 2001, research-based pharmaceutical companies invested US$ 30.5 billion in
R&D (PhRMA, 2002).
Secondly, appropriability is very high in the pharmaceutical industry. Patent protection
is extremely e¤ective (Mans�eld, 1986; Levin et al., 1987) and the rate of imitation is slower
for ethical drugs than for other products (Mans�eld et al., 1981). Thirdly, the industry has
high entry barriers because buyers have low bargaining power, because drug development is a
complex process requiring specialized knowledge, and because extensive human trials are needed
to satisfy regulatory requirements (Pisano, 1997; Scherer, 1994). The after-tax R&D cost per
new chemical entity that is placed in the market has been estimated at between $194 million
and $241 million (Dimasi et al., 1991). Other studies report (PhrMA, 2001) that it requires
an average of $500 million to introduce a new marketed medicine. These costs per marketed
product have risen considerably in real terms in the last decade. A more severe regulatory
framework has contributed to some extent to raising costs of developing new pharmaceutical
solutions.17 Number of studies has highlighted that there is a decreasing R&D productivity
particularly accentuated in pharmaceuticals relative to other manufacturing industries (DiMasi
et al., 1991; Henderson and Cockburn, 1998).
Third, although patents are essentiel to innovate, there is mixed evidence about an e¤ec-
tive patent race behaviour (Cockburn and Henderson, 1994; Henderson and Cockburn, 1995).
Moreover, innovators are confronted to an increasing therapeutic competition (rival patents)
before patent expiration. As evidenced by Lichtenberg (2002), the increasing between patent
competition imposed by rival patents covering similar therapeutic solutions (and the�me too
drugs�) has almost the same e¤ect on �rms�market erosion as the �within�patent competition
coming from generics. Although patenting activity in this sector is rather intensive relative to
other industries, quality of these patented-innovations is increasingly doubtful. The motiva-
tion to conduct low-quality re�ecting incremental innovations but intensive patenting may be
in some way related to �rms�ine¤ectiveness to introduce radical innovations over time (Hender-
son, 1993), strategic-purposes (block-entry, cross-licencing, etc.) or simply �rms�reluctance to
Lastly, the link between drug discovery and basic science has increased over time (Pisano,
1997; Cockburn et al., 1999). Firms have moved away from randomly screening a large number
17For instance, the 1962 Kefauver-Harris Amendments to the Food, Drug and Cosmetic Act18 It is increasingly acknowledged that in some industries patents are frequently used for strategic purposes. For
instance : semiconductors (Hall and Ziedonis, 2001), software (Bessen and Hunt, 2004). See also Grindley andTeece (1997); Saphiro (2000) or Arora et al (2001).
8
of potentially useful compounds, towards a more systematic approach called �rational drug de-
sign�. This involves exploiting knowledge about the biochemical mechanisms causing a disease
to identify or develop chemicals that inhibit such mechanisms (Pisano, 1997). Firms are begin-
ning to use biotechnology as a process technology as well as a research tool to improve their
search for new drugs (Cockburn et al., 1999).19 Furthermore, the advent of molecular biology in
the mid-1980 has lead to a new organization of innovation activities, namely the emergence of
a dense network of collaborative relationships among �rms of di¤erent types and other research
institutions.20
The structural break in technology raised by molecular biology has had important implica-
tions on incumbent innovation strategies and �rms�performance. The emergence of biotech-
nology can be interpreted as a technological discontinuity that broke the barriers to entry into
the pharmaceutical industry (Tushman and Anderson, 1986). Consequently, many new biotech-
nology �rms emerged to commercialize this technological breakthrough.21 While the value of
upstream activities (technology-based competences) of incumbent �rms may depreciate in an
environment of radical technological change (Tushman and Anderson, 1986), their downstream
competences (commercialization capabilities, FDA regulatory management, established reputa-
tion and �nancing) may become more valuable, notably when they are specialized with respect
to commercializing the new technology (Rothaermel, 2001; Arora et al., 2001). They may in-
curring in cooperation with entrants (or through mergers and acquisitions of companies), which
frenquently lack of complementary capabilities to put innovations into the market (Orsenigo,
1991; Stern, 1995; Lerner and Merges, 1998; Gans and Stern, 2001). As explained by Rothaer-
mel (2001), there is a symbiotic coexistence between incumbent �rms and new entrants in the
biopharmaceutical industry following radical technological change. The existence of these com-
plementary linkages in the biopharmaceutical industry has been empirically corroborated by
Arora and Gambardella (1990), Lee and Burrill (1994) and recently by Rothaermel (2001).
There are numerous examples of this phenomenon. The cooperation between Genentech and
Eli Lilly is a case in point, as Genentech has preferred to license its human insulin based on
recombinant DNA (Humulin) to Eli Lilly instead of commercializing it on its own (Lee and
Burrill, 1994).22
There is some evidence con�rming the market and technological leadership of incumbent
�rms in the Pharmaceutical Industry. In a study about the evolution of innovation in the
Pharmaceutical industry, Achilladelis and Antonakis (2000) show that most of innovative activity
19A major advance, was the discovery and application of biotechnology processes recombinant DNA and mon-oclonal antibodies in the production of physiological proteins used in therapy or diagnosis of many diseases,particularly cancer, viral and age debilitating diseases for which relatively little progress was made that far.The�rst biotechnology drug, Humulin (human insulin), received �nal FDA approval to be marketed in 1982.20An extensive literature has documented some of the consequences of such technology revolution on the orga-
nization of innovative activities in the Pharmaceutical Industry (Orsenigo, 1989; Henderson, 1994; Gambardella,1995; Orsenigo et al, 2001).21Between 1970 and 1997 alone, 1049 companies entered the industry to commercialize biotechnology. On an
average, 37 companies entered the industry per year in this time period, with 89 entries in 1992 alone (BioScan,1997).22The cooperation between Biogen and Schering-Plough in commercializing Intron A, the �rst biotech-interferon
product approved for cancer treatment, or the cooperation between Chiron and Merck to commercialize the drugEngerix-B for the prevention of hepatitis B, are other examples of these arrangements (Lee and Burrill, 1994).
9
remains concentrated among few large companies: 30 companies introduced more than 70% of all
the innovations of the sample over the period 1950-1990. These �rms have sustained high levels
of growth and R&D expenditures by means of in-house innovation, technological and therapeutic
market specialization, and mergers and acquisitions of companies (Achilladelis and Antonakis,
2000). Even with the emergence of biotechnology, incumbent �rms success to remain market
leaders. In 1993, 6 of the top 10 selling biotechnology drugs where marketed by incumbent
pharmaceutical �rms and not by the NBFs that had developed the products. Those six drugs
alone accounted for almost 60% of the revenues for the top-10 selling biotechnology drugs (Lee
and Burrill, 1994). By 1996 and 1997, the number of top-10 selling biotechnology drugs marketed
by incumbent pharmaceutical �rms had grown to seven.
2.3 Data and Methodology
Our study has two research goals. First, it evaluates empirically the impact of market incum-
bency and market share as well as the in�uence of past-innovation on �rms�propensity to inno-
vate. Second, it assess the e¤ects of the di¤erent knowledge stocks (R&D stocks, non-adjusted
and quality-adjusted patents stocks) owned by incumbents to explain �rms�market sales. Thus
this study attempts to distinguish more adequately incremental innovations from true pioneer
innovations by using patent-citation data.23 If incumbent �rms�tend to innovate incrementally
over time a larger rate of return relative to important and or quality-adjusted patents should be
expected.
To analyze the pharmaceutical inventive activity we use two data sources. First, we ex-
tracted patent data from the NBER patent data �le, by Hall et al. (2001) concerning USPTO
(United States Patent and Trademark O¢ ce) utility patents granted in the classes 514 and
524 over the period 1973-1994 (Drugs Industry). Second, we gathered �rms�patent data and
built �rms�patent portfolio. Third, we match this patent data to �rms�accounting data from
the Standard & Poor�s COMPUSTAT Annual Industrial Files Database. by using the CUSIP
identi�er (Committee on Uniform Security Identi�cation Procedure) code of the �rm.24 Data
are therefore reduced to a sample of 77 Pharmaceutical �rms in a unbalanced panel patenting
at the USPTO between January 1975 and December 1994 and quoted in the American Stock
Exchange. In order to reduce the bias imposed by citation and patent truncation, even after im-
plemented the correction proposed by Hall et al. (2001) to normalize citations, for our empirical
study we reduce our analysis to the period 1975-1991.
The original database on pharmaceutical patents concerns more than 9600 patents referred
in the 514 and 424 industrial USPTO patent classi�cations. Micro-Level Data from COMPU-
STAT concerns R&D spending, capital expenditures, number of employee, net sales, cash �ow,
operating and net income (constant US$94). As noted in Hall et al. (1986), the organization
23To the extent that quality across patents, within a �rms�patent portfolio and across �rms may largely di¤er,it is important to weigh patents�value relative to other innovations in the technological race (Tranjtenberg, 1990;Scotchmer, 1991).24Matching between two sources has been made through an intermediary �le the containing CUSIP code (Hall et
al, 2001) and the corresponding corporate company name. Since Compustat includes �rms that are traded in theUS stock market (manufacturing �rms) only, foreign pharmaceutical patents are not matched, and consequently,our �nal database concerns mainly American Pharmaceutical Firms.
10
remaining in the sample show a size distribution naively tilted toward the larger �rms in the
original universe. We build �rms�quality-innovation adjusted portfolio of patents as the sum of
total patents weighted by their number of received normalized citations.25 Stocks of R&D, and
non-weighted and weighted patents are built following the perpetual inventory method using a
15% depreciation rate (Griliches, 1989; Hall and Mairesse, 1995).
There is an emerging empirical literature using patent citations to study knowledge transfer
and innovation mechanisms. In particular, the front of a patent details other patents which
contributed to the knowledge underlying the new patent (Hall et al., 2001). Patent citation
information can be used therefore to detect the �technological value�of a patent and the �orig-
inality�of innovations: a patent which is cited many times is more likely to be valuable than a
patent that is rarely cited (Ja¤e et al., 1999, Hall et al., 2001).26 As explained by Hall et al.
(2001) and Ja¤e and Trajtenberg (1996), patent citations are themselves as a means to tackle
the underlying innovation (or economic value) heterogeneity across patents as well as a way to
trace spillovers.27
We address the problem of truncation related to patents and citations by implementing the
methodology proposed by Hall et al. (2001). Hall et al. propose to estimate the shape of the
citation lag distribution, i.e. the fraction of lifetime citations (de�ned as the 30 years after the
grand date) that are received in each year after patent grant. Accordingly, the total citations
for any patent (for which we observe a portion of its citation life) is estimated by dividing the
observed citations by the fraction of the population distribution that lies in the time interval for
which citations are observed. It is assumed that this distribution is stationary and independent of
overall citation intensity. In addition, there is an important lag between patent applications and
patent grants (in average about two years).and consequently, we observe only a small fraction
of the patents applied for that eventually will be granted as we approach the last year of patent
data. Count of patents should then be corrected using weighting factors, according to the
estimated application-grant empirical distribution (ibid).We use patent data from 1975, patent
stocks are built for those �rms patenting before this year. Our sample is restricted to the period
1975-1991 in order to avoid any bais related to truncation and patent application lags on our
data.25A point to note is that all patents are granted in USA but there is di¤erent nationality of �rms even if an
American �rm represent 75% of the sample (in the universe of the database). NBER data concerns all patentsgranted in USA meaning by �rms having an international activity and a critical size (over the period 1975 and1995).26Number of researchers have used patent citations as a �paper �ow� to track the way knowledge spills over
between organisations and areas (Henderson et al, 193; Ja¤e and Trajtenberg, 1998).27Nevertheless, citations have also some shortcomings. Frequently, many of the citations reported in the patent
are not really issued by the �rm, but they are added by the inventor�s patent attorney or the patent examiner,representing thus inventions that were wholly unknown to the citing inventor. In addition, identi�cation of originalpatents or placebo patents, is also not quite clear in the citations made (Hall et al. 2001).
11
3 Patent quality Propensity and Industrial performance
3.1 Propensity to innovate: distinguishing nature of patents
A key focus of interest in this paper is the estimation of an innovations equation (Iit). Following
Hausman et al. (1984) and Blundell et al. (1995), for a latent variable of innovation (Iit)
consider the knowledge production function 28,
Iit = f(xit; �i) (1)
where xit is a vector of �rm i characteristics (e.g. past R&D investments) and unobservables
individual �rm-speci�c factors (to the econometrician) are represented by the term �i: It rep-
resents permanent unobservable di¤erences across �rms a¤ecting the production of innovation
�rms (appropriability conditions, marketing or �nancial characteristics). This relationship is
derived as the outcome a �rms�optimal search rule for innovation (Reinganum, 1989; Blundell
et al. 1995). The search process is assumed to generate innovations in future periods. Iit in-
novations can be de�ned by the observed number of patent applications made by the �rm i:
Because the number of patents is a non-negative integrer, we adopt count data regressions as
described by Hausman et al. (1984) and Blundell et al. (1999). We model the conditional mean
as a multiplicative or log link function of explanatory factors as:
E(Iitjxit; �i) = exp(x0it� + �i)
= �itvi; for i = 1; :::::; N and t = 1::T;
where �it = exp(x0it�); vi = exp(�i); xit is a vector of observed explanatory variables, and � is
the corresponding vector of parameters to be estimated including lagged values of �rms�market
share, R&D expenditures and stock of knowledge, etc. These are assumed to be stationary. The
for i = 1; :::::; N and t = 1; :::::; T ; where Iit= fNAit; CAit; Litg
Hence in this speci�cation, �rms� innovation propensity is conditional on previous market
power, measured by the lagged value of market-shareMSit�1, previous R&D e¤ortR&Dit:Considering
the dynamics of innovation, we introduce the lagged knowledge stock variables GNA;CA;Lt�1;
where subscripts NA refer to non-adjusted patent stock (GNA;t�1); CA to citation-adjusted
patent stock (GCA;t�1); and L to leader patents stock into the knowledge production function.29
28See also Hausman et al. (1984); Winkelmann and Wimmermann (1992); Crepon and Duguet, (1997); Blundellet al (2000), amongs others.29Poisson and negative binomial with �xed e¤ects have been extended to this approach (Hausman et al, 1984),
Winkelmann and Zimermann (1992). Semi parametric methods letting the distribution of residual unspeci�edhave been proposed by Gouriéroux et al. (1984).
12
A patent is considered as leader or important patent if it has made less citations than the mean
value of citations made by all patents in the current year (application year) and received more
citations (normalized) than the average of the patents that same year. Thus, in a cross-sectional
basis (yearly), we identify the pionner patents, which are those that determine the technological
path followed by �rms in the pharmaceutical industry. These "important" patents are assumed
to generate the process of creative destruction in a technological sense, since they open a new
To identify incumbent �rms�technological superiority on innovation we include the dummy
INCi: It reefers to those �rms that we considered as technological incumbents : those who were
present in 1975 and reported consecutive patenting at least T>3 periods (19 �rms). Further, we
include an interactive term INCi � log(Git�1); to identify the relative e¤ect of incumbent �rms�stocks of knowledge on �rms�propensity to innovate. If incumbency implies stronger permanent
incentives to innovate, the incumbent-dummy should enters positively �rms�propensity to inno-
vate (Gilbert and Newbery, 1982; Segerstrom and Woniereck, 1999; Hörner, 2001). As exposed
in the previous section, incumbent �rms may enjoy several economic and technological advan-
tages facilitating the production of innovation as well as in the extent of patenting (�nancial
strength or R&D economies of scale and scope, amongst others). Inversely, they might su¤er
of diseconomies of scale related to size, in�exible R&D (specialization) and organizational iner-
tia, hampering their abilities to answer to radical innovations by others (technological change).
However, this e¤ect should di¤er across the di¤erent quality levels of patenting.
The e¤ects of market share on innovation propensity are ambiguous. Strategic considerations
imply that the marginal bene�t of an innovation (an additional increment to the �rm�s innovation
stock, Gi;t) may be more valuable to a leader with high market share than a follower with low
market share (Gilbert and Newbery, 1982). In the opposite, high-market share �rms may be
more reluctant to innovate as they �nd more pro�table to exploit existing innovation.31
Individual knowledge stock variables are obtained by accumulating past patents: Gi;t =
Ii;t + (1 + �Git�1). The stock of knowledge is continually increased by the addition of new
patents but is also continually decreased by a constant depreciation rate.� or obsolescence rate
((15%).32 Knowledge-stocks should enter positively innovation-propensity as there are dynamic
returns in the production of innovation. Cumulated technological experience facilitates in some
way current R&D productivity : R&D economies of scale (Teece, 1980, Cohen and Levin, 1989;
Chandler, 1990), learning-by doing, and learning-by learning e¤ects (Rosenberg, 1987; Cohen et
Klepper, 1996). An elasticity superior to one, suggests that there increasing dynamic returns in
cumulated knowledge to further innovate. Gi;t enters the exponential mean multiplicatively.33
30Although, the process of creative destruction in drugs is to some extent limited, both economically andtechnologically. First, as we mentioned previously (Henderson and Cockburn, 1993; 1994), pharmaceutical �rmsmay enjoy each di¤erent "patent" prize as patents may cover similar technological innovations (e.g. patentingaround, evergreening patenting). Second, economically �rms may keep their market power in speci�c therapeuticdomaines through lead time, brand reputation strategies, and other �rst-mover advantages.31 .They may also rely on alternative strategies to keep market power (�rst mover, lead time etc.).32 Initial values of Gi;t correspond to the pre-sample stock of patents for �rms patenting before 1975, otherwhise
patent�stocks for �rms appearing after 1975 are calculated according to the perpetual inventory.33Our speci�cation recalls the standard model used in the R&D- patents literature focuses mainly on the
13
The di¤erent parameterization between the multiplicative feedback model (MFM) and the liner
feedback model (LFM) (Blundell et al., 1999; Cincera, 1998) implies that, in comparison to the
LFM, the coe¢ cients �E(Iit) in the MFM measures the short run impact of a change in Xit.34
In a similar study, Blundell et al. (1999) implement both speci�cations and both models are
found to yield similar estimates for the main parameters of interest.35
We also include three additional lagged terms referring to �rms�innovation degree of gen-
erality, �rms�citations (received) share and the level of market concentration. The two former
constitute the mean value of �rms�patent portfolio (NBER patent �le, Trajtenberg et al., 1997;
Hall et al., 2001). Firms�citations share represents the part of the �rm in total (received) cita-
tions in the pharmaceutical classes 514 and 324. It measures the par of knowledge generated by
the �rm in the previous period or the degree of technological leadership. If technology-leaders
have stronger incentives to remain in the innovation-race, it should enters positively �rms�in-
novation propensity (Segerstrom and Woniereck, 1999; Hörner, 2001). Nonetheless, its impact
must di¤er à priori according to the quality level of patenting (Henderson, 1993).36
Innovation generality constitutes our measure of �rms�R&D diversi�cation. As shown by
Cockburn and Henderson (1998) and Henderson and Cockburn (1996), pharmaceutical �rms
which are more diversi�ed have been found to enjoy a stronger R&D productivity (R&D economies
of scope). The measure of originality is an Her�ndhal concentration index (Trajtenberg et al.,
1997; Henderson et al., 1998): =1-NiXk=1
�Ncitingi;kNcitingi
�2, avec Nciting=nombre of citing patents,
K=number of class (all citing patents). It denotes the share of citations made by patent i that
belong to patent class n, out of Ni patent classes. Thus, if a patent cites previous patents that
belong to a narrow set of technologies, the generality score will be low, whereas citing patents
in a wide range of �elds would render a high score.
3.2 Weakly exogenous variables and dynamics
The introduction of predetermined variables (E(xit; "it+j) = 0; j �; 0 but E(xit; "it�s) = 0; s �1) in the right hand side of the equation in combination with unobserved heterogeneity requires a
special treatment for estimation. Under weak exogeneity of variables, estimation by Conditional
Maximum Likelihood (CML) developed by Hausman et al. (1984) allowing for individual-speci�c
(�xed) e¤ects give inconsistent estimates.37 Likewise, more �exible models allowing for data
distributed lag of past R&D investments. Assumming that historic R&D investments are combined through aCobb Douglas technology to produce knowledge Qit = R&DitR&Dit�1:::vi:, the conditional mean speci�cation ina multiplicative distributed lag model is the standard model: E(yitjxit; xit�1; ::::; xit�m; vi) = exp(x0it�1+x0it�1�2+:::x0it�m�m)vi:34As explained by Blundell et al (1999), in the MFM model the nonlinear dynamic makes it di¢ ccult to judge
the stability properties of the model and the potentially important feedback parameter is sensititive to largeinnovation counts.35Although a non-nested comparison has suggested a preference for the LFM speci�ciation.36Likewise, a weak citation-share indicates that other �rms or competitors own a larger part of the new tech-
nology. Hence competitive spillovers may have a negative rivalry e¤ect on a �rms�likelihood to apply for a patent: the more competitors invet in R&D, the less a �rm is likely to invent a new technology (Loury, 1979; Lee andWilde, 1982).37PoissonThis estimator is the log-link equivalent of the within group sestimator for the linear panel data
modelSemi parametric methods where the distribution of the residual is unspeci�ed have been proposed byGouriéroux et al (1984):
14
overdispersion while controlling for unobserved heterogenity such as the negative binomial model
(Cameron and Trivedi, 1986; Winkelmann and Zimmermann, 1991) also rely on the strong
assuption that the �rms�unobserved e¤ets are independent of the explanatory variables.
To deal with predetermined regressors while allowing for correlation of unobserved individual
e¤ects with regressors, Chamberlain (1992) and Wooldridge (1997) have proposed a non-linear
GMM estimator. It uses the equation in quasi-di¤erences that eliminate the �xed e¤ect from
the multiplicative model, generating orthogonality conditions that can be used for consistent
estimation in count data models with predetermined regressions.38
Considering �it = exp(X 0�); the quasi-di¤erenced transformation by Wooldridge (1991;
1997) for the model may be written :
sit(�) = Iit � exp((xit � xit+1)0�)Iit+1 (3)
Instruments date t or before will be orthogonal to sit(�) and therefore the moment conditions
E(sitxis) = 0 for t > s (4)
The precision of the GMM quasi-di¤erences estimator requires that functions of the xit; xit�1:::xi0,
and Iit�1; Iit�2::::Ii0; are good instruments for the di¤erence term exp((xit � xit+1)0�)Iit+1:39
3.3 Performance and Quality-Adjusted Patents
To evaluate the impact of non-adjusted and quality adjusted patent stocks, we consider now a
where yit is a measure of industrial performance, market sales of �rm i in year t, Lit is log
employment, Kit is log capital stock, Git is the knowledge stock (Git NA: non-weighted patents
stocks, Git CA: citations-weighted patents stocks and Git L and GitF , leader and follower patents
stocks) and t is a year speci�c intercept. Of the error components �i is an unobserved �rm
speci�c e¤ect whereas vit is a possibly autoregressive (i.e. productivity/performance) shock
and ait re�ects serially uncorrelated measurement errors. Constant returns to scale would imply
�n + �k = 1; but is not necessarily imposed.
Besides dealing with simultaneity, dynamic panel data methods (IV-GMM estimation) treat
the inherent problem of serial autocorrelation of residuals and the potential heteroscedasticity
38Application of the quasi-di¤erenced GMM estimator to count data models with predeterminated regresorshas been made by Montalvo (1997), Crépon and duguet (1997) and Cincera (1997). Blundell et al. (2000) providean extension of the GMM estimation by including the sample-mean count of patents as a measure to deal withpersistent unobserved heterogeneity.39Blundell et al (1995) have developed an estimator in which the pre-sample information on innovation replaces
the unobservable �rm �xed e¤ect.
15
within panels. We implement the system GMM outlined in Arellano and Bover (1995), and fully
developed by Blundell and Bond (1999), both improving e¢ ciency of GMM-estimation, relative
to standard IV techniques and the Arellano and Bond GMM estimator. As known, estimation
using the Arellano and Bond (1993) GMM estimator presents some shortcomings. It provides
very small and imprecise estimates of �L and �K (Blundell and Bond, 1999). Overidenti�cation
restrictions are typically rejected as fxi;t�2; yi;t�2g are not valid instruments. Further, it su¤ersof the problem of weak instruments in dynamic moders where regressors in �rst di¤erences are
weakly autocorrelated. Lastly, �rst di¤erence transformation also eliminates the cross-sectional
variation an it is suject to the problem of measurement error in input.
The IV-GMM estimator proposed by Blundell and Bond (1999) deal with these issues. The
x�(xi;t�1;!it) and fYi0; Li0;Ki0g is in steady state. Predetermined and endogenous variablesin levels are instrumented with suitable lags of their own �rst di¤erences fxi;t�j;yit�j ; j � 2g.And the �rst di¤erences equation is respectively instrumented with lagged levels of inputs and
outputs f�xi;t�j;�yit�j ; j � 1g. These instruments in the SYS-GMM addresses the problems
of omitted variable bias, endogeneity and unit root e¤ects in the choice of instruments (Blundell
and Bond, 1998). Given its di¤erent advantages to deal with the speci�cities of our data, is our
preferred econometric method.40
We implement the Arellano and Bond test for �rst and second (m(1), m(2)) order serial
autocorrelation of residuals. If vit is not serially correlated, the di¤erence residuals should be
characterized by negative �rst-order serial correlation and the absence of second-order serial
correlation. The Hansen test of over-identifying restrictions are reported at the bottom of
regressions. These tests are inspired from Sargan (1985) and Hansen (1982) and conduct test
for the null hypothesis that the remaining theoretical orthogonality restrictions are equal to zero.
Failure to reject the null hypothesis indicates that the instruments are valid, and then supports
the validity of the model speci�cation.
4 Innovation Trends
Figures 1-2 give an overview of the innovation trends in the American Pharmaceutical Industry.
The evolution of the �rm-average R&D intensity, unadjusted and citation-adjusted patenting
in the industry is displayed in �gure 1. During the period considered 1975-1995, �rms�have
dramatically and continually increased their R&D investment relative to sales from 8% in 1975
to 11% at the end of 1991. This trend re�ect both an expansion in the scale of research e¤ort
and the change in its nature, as the adoption of rational drug discovery techniques increased
the demand for sophisticated equipment and more highly skilled researchers (Henderson and
Cockburn, 1996; Cockburn and Henderson, 1998). Likewise, total patenting has also continuing
40As we are confronted to potential endogeneity arising from simultaneity of input-factors and unobservedheterogeneity, both are likely sources of bias in the OLS estimation. Further, OLS produces inconsistent estimatesof the lagged dependent variable (Nickell, 1981; Bond, 2002), since this explanatory variable is positively correlatedwith the error term due to the presence of the individual e¤ects. Alternatively, within groups estimator is alsoinconsistent here, as they are subject to bias in absence of complete exogeneity.
16
Figure 1: Evolution of Pharmaceutical R&D Productivity
augmented: the number of total patents (applied) almost doubled during this decade from just
over 1,000 patents in 1980 to around 5,000 in 1995.
When adjusting patents by their normalized received citations, the �gures show a di¤erent
trend. The level of quality seems to follow a more stable path, although decreasing at the
end of the period. However, �rms�productivity (patents per R&D) has signi�cantly decreased
(Figure 2). This contracting trend is still more accentuated when considering the number of cites
per patent (industry means). Lastly, when distinguishing leader patents from follower patents
(citations received and forward citation lag larger and smaller, respectively than the yearly-
industry mean). Indeed, the number of incremental patents seems to increase over time whereas
leader patents show a slower and contracting movement along the period. A similar tendency
is found for incumbents�patenting activity (�gures 3 and 4). Although patenting by incumbent
�rms relative to the industry remains important all over the period, their R&D productivity is
notably decreasing particularly in quality-patents at the end of the period.
Henderson and Cockburn (1993; 1996) suggest that one possibility of this declining is that the
transition to more rational methods of drug research has led to a change in patenting strategies
and an increase in the signi�cance of each patent. Real research output may have remained
constant or even risen. Another possibility is that the industry is approaching technological
exhaustion (Grabowsky and Vernon, 1990; Scherer, 1994) but rates of investment in research
are continuing to accelerate despite the decline in patenting, perhaps as a strategic response
to actions of competitors, or simply because changes in demand conditions were increasing the
marginal revenue product of research.
Table 1 reports the 12 most cited patents in pharmaceuticals (514-424 UPSTO patent classes)
in the NBER patent data �le (Hall et al., 2001). All of these patents were applied during the 1970
17
Figure 2: Evolution of Pionner and Incremental Patenting in Drugs
and 1980s. The oldest most-cited patent dates from 1969 (application date) and belong to Alza
Corporation. The most recent cited patent (application date) belongs to DuPont de Nemours
and was applied for protection in 1988. Table 2 summarizes the number of patents considered as
�leader patents�to 1976 and then (cumulated) to 1995. This simple frequency of technological
leadership at the beginning and at the end of the period suggest that pharmaceutical important
innovations are introduced by an increasing number of �rms: there are new innovators in the
industry (with small and medium number of patents), whereas the technological leaders in 1976
continue to bring quality-patents in a relative similar percentage: 7,14% of all quality-patents
(to 1995) belong to Merck and Co., 6,79% to Bristol Myers Squibb, 6% to Ely Lilly and Co.
whereas P�zer Inc. and Schering Plough both own around 5,4% of total leader patents. Total
patents and citations in 1975 and 1990 are reported for a small of large pharmaceuticals in table
3.
Figure 1 in Appendix displays the evolution of total citations (received) and total citations
normalized to correct for the bias of few citations attributed to patens in the lasts year of the
period. We use the de�ator provided in Hall et al. (2001) for the pharmaceutical industry.
Although the normalization of citations reduces the striking contracting tendency at the end of
the period, it does not eliminate completely the problem. For this reason, our empirical analysis
will focus on the period 1975 to 1991, in order to minimize this bias.
4.1 Descriptive Statistics
Table 2-4 presents summary statistics describing the data. Averaging across all �rms and years,
our �rms spent on average in R&D 258,550 dollars and has average sales of 2952,86 (USD
1990). Important di¤erences appear across �rms when looking at the sub-sample of continuing
innovators (77) for which at least three years are reported consecutively, our sample of study;
18
Figure 3: Incumbent Firms�Patenting in Pharmaceuticals (1)
Figure 4:
and at the one concerning incumbent �rms (19). We consider an incumbent �rm, if it appears
in 1975 patenting, and continue patenting at least three periods consecutively. Compared to
non-innovators which reports annual (average) sales of 2589.10 (less than the industry average),
persistent patenting �rms enjoy an average annual sales of 3240.70, report the largest R&D
expenditures (343,11) and an average market share of 0.54. For this sample of 77 �rms the
average number of patents is 6.35, of leader patents is 0.48 whereas of follower patents, 5.86.
Their average citation share (annual) is 0.0075
For the incumbent �rms�sample (19 �rms), the statistics change substantially. Their average
sales reported is around 10,480.11, with an average market share of 1.74, and R&D expenditures
of 1076,53. The average number of citations received (normalized) 2055.78. Their average
number of patents is 26.77 (non-adjusted patents), 1.82 leader patents and 25 follower patents.
Their average citation share (annual) is 0.1871. These simple descriptive statistics suggest
that incumbent �rms show both an economic and technological superiority: they have a more
intensive patenting activity, a higher quality-patenting relative to the rest of the industry and
the largest market and citation share.
5 Empirical Results
5.1 Innovation equation results
In this paper it is argued that by looking at the relationship between incumbency and innovation
together with the impact of patents stocks on �rms�pro�tability it is possible to shed further
light on the importance of the incentives to innovate. Furthermore, it attempts to disentangle
�rms� organizational and strategic e¤ects on �rms� propensity to innovate by looking at the
impact of incumbency as a proxy for the former and of market share for the latter.
19
Table 1 reports estimates of the innovation equation for the sample of 77 �rms using the
GMM-quasi di¤erentiated regression. The set of instrumental variables used are reported at
the bottom of the table, as well as the corresponding auto correlation tests (�rst and second
order). The Sargan test for over-identi�cation justi�es the validity of instrumental variables.
In each case the log-link model (4.1) is estimated across the di¤erent models of patenting. A
di¤erent instrumentation set is used according to the type of knowledge stock (G): subscriptsNA
refer to non-adjusted patent stock (GNA;t�1); CA to citation-adjusted patent stock (GCA;t�1);
and L to the stock of leader or "important" patents (GL;;t�1): As reported by the AR(1) test,
there is signi�cant serial auto-correlation (Blundell et al., 1999) which re�ects the importance of
dynamic e¤ects on �rms�innovation propensity. Nonetheless, it could also indicate the presence
of permanent heterogeneity across our �rms. The coe¢ cients �E(Iit) of the multiplicative
feedback model (MFM) measures the short run impact of a change in Xit. Regressions include
time-dummies in the instrumental variables set.
Regarding the impact of market incumbency (dummy variable) and market-share, it seems
that they a¤ect di¤erently �rms�propensities to innovate. These di¤erences appear more accen-
tuated when distinguishing patents�average quality and leader-patenting. Two main �ndings
appear from our estimations. Corroborating the hypothesis on �rms�incompetence to persis-
tently innovate, our results indicate that �rms�with higher market share show a weaker propen-
sity to introduce quality innovations. Nevertheless, among this group of �rms, incumbent �rms�
appear as persistent innovators.
Whereas past market share a¤ects positively non-adjusted patenting propensity, it is nega-
tively associated with the number of citation-adjusted patenting (columns 2-4). Evaluated at
their means, the elasticity of patenting with respect to market share evaluated at the mean
market share is 0,4. An 1% increase of market share increases non-adjusted patenting by 0,4%
(0,2821/1-0,6408)*0,0054) but decreases quality-patenting by 0,3%.41
Moreover, this deterring-innovation e¤ect related to market-share is more important for
the expected number of leader patents (columns 2-4). These results suggest that �rms�with
increasing market power are less likely to introduce quality-patents: an 1% increase of market
share decreases by 7,6% the probability of applying for an important patent (leader patents).
Hence it seems that there are strategic motives are at stake. The weaker propensity to have
pioneer patents by dominant �rms may obey to strategic reasons such as the reluctance to
cannibalize existing innovations (Reinganum, 1982; 1983) or to reveal new technological options
for followers (Farrell and Saphiro, 1996). Our results diverge from those reported by Nicholson
(2003), who found that high-market power �rms (those having more than 70% of the market
sales) have 72% higher probability in citation-weighted patenting whereas for medium power
�rms the e¤ect is 75% (those �rms having between 40 and 69% of the market sales).
41This e¤ect is larger than the one reported by Blundell et al. (1999) on major innovations for the U.K. industry:the elasticity at the means value is 0,08 when using the MFM (0,10 when using the LFM) for the whole industrywhereas for the pharmaceutical industry the estimate is dramatically larger: 0,50.
20
Tab. 1 GMM-Wooldridge Estimation on Patenting by Pharmaceutical Firms (1975-1991)Non-Adjusted Patents Citation-Adjusted Patents Leader Patents
AR(1) -2.445** -1.898* -1.507AR(2) 0.7279 -1.230 0.677Sargan Overident. Test 43.293 40.954 16.591DOF 41 41 13Robust Standard Erros reported. IV sets include time dummies.* signi�cant at 10%; ** signi�cant at 5%; *** signi�cant at 1%IV=Stock Capitalt�1�t�2, MSt�2 Incumbency, Citations-Sharet�2,LGR&D2 ;LGR&D2*INC
Our results for the pharmaceutical industry contrast but also con�rm to some extent the
Schumpeterian hypothesis. First, as �rms increase their market-share, they may show lower in-
centives to innovate because of their increased market power and fear to cannibalise innovation.
Nevertheless, not all market-power �rms may be subject to such disincentives. Regarding market
incumbency, two main conclusions stem from our analysis. On the one hand, our �ndings corrob-
orate the Schumpeterian perspective about incumbents�superiority and long-run persistence in
the production of innovations. On the other hand, our results also con�rm the advanced hypoth-
esis that incumbent �rms�R&D productivity is decreasing notably, in the production of major
pacity, which con�rms their increasing incompetence to respond to technological breakthrough
(Tushman and Anderson, 1986).
Though, even under depreciation of their technological capacity, incumbent �rms remain
technological leaders. Radical technological change that undermines incumbent �rms�upstream
value chain activities will cause those �rms to seek out a new source of competitive advantage
within the rede�ned technological framework (Dosi, 1982; Pavitt, 1998). Our results suggest that
technological leadership is in�uenced by other non-technological factors, which are captured to
some extent by the incumbency status. For instance, the non-technological assets of incumbent
�rms may become more valuable for the successful commercialization of new technologies, no-
tably; when they are specialized with respect to commercializing in speci�c therapeutic domains
(Leonard-Barton, 1992; Rothaermel, 2001).
The raising of new technologies (e.g. molecular biology and biogenetics) brought by entrants
does not imply the replacement of incumbent �rms. Actually, face to the threat of creative
destruction as it implies a competences�destroying process, incumbent �rms �nd their way to
adapt to technological breakthroughs by means of cooperating with entrant �rms, which often
lack of production and marketing capabilities to bring discoveries into the market. Arora and
Gambardella (1990), Orsenigo et al (2001) and Rothaermel (2001) have provided evidence on
24
the specialization-based division of labour in commercializing a new technology in the Pharma-
ceutical industry following the biotechnology revolution.
5.3 The Productivity Equation
In this stage we evaluate the returns on �rms�sales from the di¤erent knowledge-stocks. Table 3
displays estimates by IV-GMM method for the whole sample and the 77 patenting �rms. In the
whole sample estimation, the R&D stock does not contribute to �rms�economic performance. In
the opposite, the non-adjusted patent stock and citation-adjusted patent stock contributes both
positively and signi�cantly to enhance industrial sales. The impact of patent stock is almost
the double the e¤ect by citation-adjusted patenting stock or citation-stock. This result suggests
that large patent portfolios contribute to increase �rms�pro�tability. Quality of patent stocks
matters in a lesser extent than non-adjusted patent. Thus pharmaceutical �rms are motivated
to patent intensively regardless of the quality-level of innovations as patents serve to increase
or protect market power (entry-blocking). Regarding the sample of the 77 patenting �rms, the
R&D stock has now a large signi�cant impact on performance but it has smaller e¤ect than the
non-adjusted patent stock : the elasticity of real sales respect to the R&D stock ($10 000) is
0,10. Likewise, the double of the patent stock increases sales by 11% whereas the double of the
citation-adjusted patenting and of citation-stocks lead to increases between 5 and 6% of total
factor productivity.
These �ndings suggest that intellectual-property stocks have a large impact on economic
performance than quality-weighted patenting. As patenting is pro�table, this strategy would
explain the increasing patenting trends previously reported in the Industry. The high importance
of patents in industrial performance has been documented previously. As explained by Hall et
al. (2004), drugs are characterized by discrete product technologies where patents serve their
traditional role of exclusion, and some of them are therefore valuable on an individual basis, as
measured by citations.42 In pharmaceuticals, the impacts of patent yield (Patent stock/R&D
stock) and of citations per patents on �rms�market value are more important than the R&D
stock (to assets). Our results contrast to the �ndings by Hall et al. (2002) and Blundell et al.
(2002) on the impact of knowledge-stocks on �rms�market value. Hall et al. (2002) found that
citation-weighted patent stocks are more highly correlated with market value than patent stocks
themselves. They explain that this result is due mainly to the high valuation placed on �rms
that hold very highly cited patents.
42Regarding pharmaceuticals, Hall et al. (2002) show that the impact of patent yield of R&D (Patentstock/R&D stock) for Drugs on �rms�market value is found to be three times the average e¤ect (0.10 versus.031), still larger than that of Computers. Likewise, the impact of the Citations/Patent stock is over 50% higherthan the average e¤ect in the manufacturing industry.
25
Tab. 3 IV-GMM Estimation on Log(Sales): American Pharmaceutical Firms (1975-1991)Whole Sample Patenting Firms (T>=3)