Disponible à / Available at permalink :

132  Download (0)

Texte intégral

(1)

- - -

- - -

Dépôt Institutionnel de l’Université libre de Bruxelles / Université libre de Bruxelles Institutional Repository

Thèse de doctorat/ PhD Thesis Citation APA:

de Rassenfosse, G. (2010). Essays on the propensity to patent: measurement and determinants (Unpublished doctoral dissertation). Université libre de Bruxelles, Faculté des sciences sociales, politiques et économiques – Sciences économiques, Bruxelles.

Disponible à / Available at permalink : https://dipot.ulb.ac.be/dspace/bitstream/2013/210130/5/fd1bed5f-9b8e-42e3-8f84-a7229d5b7ee5.txt

(English version below)

Cette thèse de doctorat a été numérisée par l’Université libre de Bruxelles. L’auteur qui s’opposerait à sa mise en ligne dans DI-fusion est invité à prendre contact avec l’Université (di-fusion@ulb.ac.be).

Dans le cas où une version électronique native de la thèse existe, l’Université ne peut garantir que la présente version numérisée soit identique à la version électronique native, ni qu’elle soit la version officielle définitive de la thèse.

DI-fusion, le Dépôt Institutionnel de l’Université libre de Bruxelles, recueille la production scientifique de l’Université, mise à disposition en libre accès autant que possible. Les œuvres accessibles dans DI-fusion sont protégées par la législation belge relative aux droits d'auteur et aux droits voisins. Toute personne peut, sans avoir à demander l’autorisation de l’auteur ou de l’ayant-droit, à des fins d’usage privé ou à des fins d’illustration de l’enseignement ou de recherche scientifique, dans la mesure justifiée par le but non lucratif poursuivi, lire, télécharger ou reproduire sur papier ou sur tout autre support, les articles ou des fragments d’autres œuvres, disponibles dans DI-fusion, pour autant que :

Le nom des auteurs, le titre et la référence bibliographique complète soient cités;

L’identifiant unique attribué aux métadonnées dans DI-fusion (permalink) soit indiqué;

Le contenu ne soit pas modifié.

L’œuvre ne peut être stockée dans une autre base de données dans le but d’y donner accès ; l’identifiant unique (permalink) indiqué ci-dessus doit toujours être utilisé pour donner accès à l’œuvre. Toute autre utilisation non mentionnée ci-dessus nécessite l’autorisation de l’auteur de l’œuvre ou de l’ayant droit.

--- English Version ---

This Ph.D. thesis has been digitized by Université libre de Bruxelles. The author who would disagree on its online availability in DI-fusion is invited to contact the University (di-fusion@ulb.ac.be).

If a native electronic version of the thesis exists, the University can guarantee neither that the present digitized version is identical to the native electronic version, nor that it is the definitive official version of the thesis.

DI-fusion is the Institutional Repository of Université libre de Bruxelles; it collects the research output of the University, available on open access as much as possible. The works included in DI-fusion are protected by the Belgian legislation relating to authors’ rights and neighbouring rights.

Any user may, without prior permission from the authors or copyright owners, for private usage or for educational or scientific research purposes, to the extent justified by the non-profit activity, read, download or reproduce on paper or on any other media, the articles or fragments of other works, available in DI-fusion, provided:

The authors, title and full bibliographic details are credited in any copy;

The unique identifier (permalink) for the original metadata page in DI-fusion is indicated;

The content is not changed in any way.

It is not permitted to store the work in another database in order to provide access to it; the unique identifier (permalink) indicated above must always be used to provide access to the work. Any other use not mentioned above requires the authors’ or copyright owners’ permission.

(2)

propensity to patent and déterminants

Gaétan de Rassenfosse

TlKse présentée en \Tie de l'obtention du grade de Docteur en Sciences économiques et de gestion

ai 2010

(3)

Contents

Acknowledgments v

Introduction vii

How much do we know about flrms’ propensity to patent? 1

1.1 Introduction... 2

1.2 The propensit}' to patent... 3

1.2.1 Why does the propensity to patent inatter?... 3

1.2.2 Known déterminants of the propensity to patent... 6

1.3 Data and econonietric methodology... 9

1.3.1 Independeut v'ariables... 9

1.3.2 Econonietric methodology... 11

.1.4 Empirical results... 13

1.4.1 Descriptive statistics... ... ., 13

1.4.2 Empirical results... :... 18

1.4.3 Potential sélection biases and other sample-induced biases... 24

1.5 Concluding remarks ... 25

A Appendices... 27

A.l Potential pitfalls... 27

A.2 Industry classification ... 30

A.3 Details on the econometric methodology... 31

A.4 Sample sélection... 33

Bibliography ... 35

2 The two faces of the R&D-patent relationship 39 2.1 Introduction... 40

2.2 Patent statistics as économie indicators... 41

2.3 The Model... 43

2.4 Empirical Implémentation ... 47

2.4.1 Patent count ... 47

2.4.2 Value indicators ... 49

(4)

2.4.3 Sample ... 50

2.5 Results... 52

2.5.1 Descriptive statistics... 52

2.5.2 Productivity and propeusity... 53

2.5.3 Interprétation and discussion of the results ... 58

2.6 Concluding remarks ... 60

A Appendix... 63

A.l Description of value indicators ... 63

Bibliography ... 64

3 The rôle of fees in patent Systems 69 3.1 Introduction... 70

3.2 Historical perspective on the setting of fees ... 71

3.3 Stylized facts... 72

3.4 Impact of fees on applicants’ behavior... 76

3.4.1 Descriptive papers and early hypothèses... 76

3.4.2 Survey studies: never “cheap” enough... 77

3.4.3 Pre-grant fees... 78

3.4.4 Post-grant fees... 80

3.5 In search for an optimal fee polic}' ... 81

3.6 Concluding remarks ... 84

A Appendix... 86

A.l The complex fee schedule of patent Systems... 86

Bibliography ... 88

4 On the price elasticity of demand for patents 93 4.1 Introduction... 94

4.2 Methodological approach... 96

4.3 Descriptive statistics... 98

4.4 Econometric methodology... 100

4.4.1 Partial adjustment model... 102

4.4.2 Error correction model... 102

4.5 Empirical results... 103

4.6 Concluding remarks ... 107

A Appendices... 109

A.l Working assumptions... 109

A.2 Data sources ... 110

Bibliography ... 111

(5)

ESSAYS on THE PROPENSITY TO PATENT:

MEASUREMENT and DETERMINANTS

Gaétande Rassenfosse

Thèse présentée en vue de l’obtention du grade de Docteur en Sciences économiques et de gestion.

Mai 2010

J

Université libre de Bruxelles

Faculté des Sciences sociales et politiques - Solvay Brussels School of Economies and Management

(6)

Université libre de Bruxelles Av FD Roosevelt 50 - CP 114 1050 Bruxelles

Belgique

Email: gderasse@ulb.ac.be

Thèse de doctorat présentée en séance publique le 28 mai 2010 à TUniversité libre de Bruxelles.

Jury: Bruno Cassiman, lESE Business School, Espagne Michèle Cincera, Secrétaiie, ULB

Catherine Dehon, Présidente, ULB

Keld Laursen, Copenhagen Business School, Danemark Carine Peeters, ULB

Bruno van Pottelsberghe, Promoteur, ULB

(7)

Acknowledgments

J’aimerais remercier chaleureusement Bruno van Pottelsberghe, mon promoteur, pour tout ce qu’il m’a apporté. Non seulement pour m’avoir communiqué sa passion pour la recherche, mais aussi pour la liberté qu’il m’a donnée. J’apprécie également sa grande disponibilité et réactivité. Ses nombreux conseils et recadrages ont été un inestimable guide tout au long de mon parcours de doctorant. Je le remercie, enfin, pour m’avoir fourni de très bonnes condi­

tions de travail, notamment en me donnant la possibilité de participer à un grand nombre de séminaires et conférences ; son attitude bienveillante m’a été bénéfique.

J’aimerais également exprimer ma gratitude à l’égard d’autres personnes cini ont utilement commenté mes travaux. Je pense en particulier à Karin Hoisl et Nicolas van Zeebroeck pour leurs commentaires prom]jts et pertinents. Vielen herzlichcn Dank Karin fur Deine sehr hil- freichen und stels zeitnahen Kommeniare. L’œil affûté de Jérôme Danguy m’a également été très utile. Je remercie chacun des membres de mon jury pour leur présence et leurs commen­

taires constructifs. Tak for at du vil vœre en del af min PhD komite og for dine konstruktivc kommentarer. Ik dank u dat u aanvaard heeft lid te zijn van mijn jury en voor uw conslruc- tieve commentaren.

Je suis reconnaissant envers le FNRS pour la bourse qui m’a été attribuée ainsi que pour les financements additionnels qui m’ont permis de participer à divers séjours de recherche et conférences.

Les amitiés que j’ai nouées durant mon doctorat, tant avec des personnes d’ECARES que du CEB et du DULBEA, ou avec des personnes extérieures à l’ULB, me sont également très précieuses et m’ont permis de passer 4 très belles années. J’aimerais également saisir l’opportunité pour remercier les personnes concernées de m’avoir permis de partir à l’INSEAD.

Je profite de l’occasion pour remercier tous mes proches et amis, notamment Alexis et Ben­

jamin pour leur amitié de longue date et Caroline et Serge pour mes années à Solvay.

(8)

Je remercie vivement les membres de ma famille, plus particulièrement mes parents, ma sœur et ma grand-mère, ^^es parents pour m’avoir fourni un cadre de vie épanouissant et pour leurs encouragements continus. Ma sœur, Céline, pour la grande complicité qui nous lie, et ma grand-mère pour sa bonne humeur et sa joie de vivre.

Enfin, j’aimerais adresser mes derniers remerciements à Sophie, pour un milliard de petites et grandes choses, mais surtout pour son amour profond et le bonheur qu’il me procure.

Bruxelles, Mai 2010.

(9)

Introduction

Patent Systems hâve been désignée! to foster innovative efforts. Tire monopoly power conferred by patent right aims to compensate firins for the weak appropriability of revenues accruing frbm their inventions, thereby pushing private investment in research and development (R&D) doser to the socially optimal level. Patent policy is particularly effective in industries such as pharmaceuticals and Chemicals, where it is generally admitted that most new products would not hâve been developed had patent protection not existed (see, e.g., Mansfield, 1986).

The effectiveness of patent protection is not unchallenged, however. For instance, Anton and Yao (1994) show that an inventer can appropriate a sizeable share of the market value of his invention for which no property rights exist with a well-designed contract. Heller and Eisenberg (1998) explain how the fragmentation of patent rights may hamper innovation in biomédical research.

Furthermore, patenting is not the sole means to protect the profits potentially accruing from inventions. Secrecy, first mover advantage and complernentary assets hâve also been found to be important appropriability mechanisms (Cohen et al., 2000; Levin et al., 1987).

Conversely, protection from imitation is not the sole motivation to apply for a patent. There is evidence that patenting is being used in non-traditional ways. Firms take out patents for varions reasons, such as to facilitate the raising of capital, to block competitors or to protect their freedom of operation (Blind et al., 2006; Cohen et al., 2000; Giuri et al., 2007).

In this context, it is particularly important to understand how and why firms use the patent System and how intellectual property (IP) policy influences firms’ behavior. According to Mansfield (1986, p. 173), understanding ‘ïo what extent l...j firms make use of the patent System, and what différences exist among firms and industries /.../ in the propensity to patent”

is one “of the most important questions conceming the patent System. ” It is ail the more true today, more than 20 years later, with the great success that patent data hâve had with scholars and policymakers. Indeed, patents hâve become an essential source of information for studying the innovation process and assessing technological performance (see, e.g., Griliches, 1990).

The central, unifying theme of the dissertation is the analj'sis of different facets of the propensity to patent, which we defiue in broad terms as the extent to which firms rely on the patent System, In particular, the objectives of the thesis are (i) to understand how the propensity to patent affects patent statistics, (ii) to iinprove the measurement of the propensity to patent, and (iii) to assess its déterminants.

(10)

OverView of the chapters

In Chapter 1 we analyze survey data on the proportion of inventions patented — a potential metric for the propensity to patent — from an international sample of manufacturing firms.

As background to the study, in the first part of Chapter 1, we explain the pitfalls associated with the use of patent production functions to study the invention process. In brief, because inventions are not easily observable, scholars usually rely on patent data to study the invention process. However, it is well known that patents are iinperfect indicators of inventive activity.

Not ail inventions are patentable and not ail patentable inventions are actually patented. In addition, patented inventions differ in their quality and scope, and strategie behaviors by firms add further noise to the data. Consequently, the classic approach of using patent production functions to study the invention process is subject to an identification problem. Economet- ric analysis cannot disentangle the productivity effect (which leads from research efforts to inventions) from the propensity effect (which leads from inventions to patents). In this re­

spect, a Sound understanding of the invention process necessarily requires an understanding of the propensity to patent. The data used are particularly useful in that they allow a proper analysis of the déterminants of the propensity to patent. The next part of Chapter 1 seeks to explain the proportion of inventions patented using an appropriate econometric framework.

The empirical analysis provides evidence for four drivers of a firm’s decision to patent: its size, its attitude towards patenting, market wealth and patent policy.

The fundamental identification problem, wdiich motivâtes the analysis presented in Cha]5- ter 1, is explicitly addressed in Chapter 2. In the words of Lanjouw et al. (1998, p. 413),

“one of the longeât lasting debates in the history of économie measurement has been whether the noise and the biases in patent count measures can be made small enough to make patent counts useful measures of innovative output in économie studies. ” We propose a methodologj' to filter out the noise induced by varying patent practices in the R&D-patent relationship.

The methodology décomposés the patent-to-R&D ratio into its components of productivity and propensity. It is designed for cross-country analyses of patenting performances at the industry level and leads to results that hâve intuitive interprétations. Under some general assumptions, we hâve shown that the density of patent value can be used to identify produc­

tivity and propensity effects. The methodology is then applied to a novel data set of patent applications in which each priority filing is fractionally allocated to its inventors’ countries and to the technological areas to which it belongs.' We performed the analysis for six industries and four countries over the period from 1988 to 1992.

The next two chapters (3 and 4), jointly written with Bruno van Pottelsberghe, examine the rôle of fees in patent Systems. Understanding the rôle of fees is not only relevant for inforraed policy decision making, but also for the growing number of scholars who use patent data but often ignore the price considérations concerning patenting.

Chapter 3 takes stock of the literature, which usually addresses two main research ques­

tions: the question of optimal fees and the impact of fees on the behavior of applicauts. Patent fees are generally applicant-friendly and setting their level has traditionally been driven by the need to balance the budget, or by the w'illingness to adjust to international standards without rauch welfare considération. Tins is probably due to the fact that discussions on patent fees remain confîned to the circle of civil servants who staff patent offices and patent

priority filing is the first patent application filed to protect an invention, as opposed to second filings which are patents filed in later stage to protect the invention in foreign markets.

(11)

Introduction IX

lawyers. Economists hâve recently realized that the setting of fees can be used to fine tune the patent System and liave been devoting increased attention to the topic. Contributions that look at the optimal fee policy provide grounds for both high and low application fees and renewal fees, depending on the structural context or the political objectives, New statistics gathered for a large number of patent offices show that application fees are, by and large, lower than renewal fees, and renewal fees increase more than proportionally with patent âge.

Studies that focus on the behavior of applicants confirm that patent fees are a deterrent to patenting, though a patent seeras to be a highly inelastic good.

Chapter 4 présents panel-data estimâtes of the price elasticity of demand for patents at the trilatéral offices (that is, in the U.S., Japan and Europe). This exercise is useful for three reasons. First, most existing studies rely on cross-sectional data to estimate the price elasticity, leading to results that are potentially inaccurate. Indeed, this approach implicitly assumes no or low adjustment costs, sucli that a change in any exogenous variable leads to an immédiate adjustment in the number of filings. The estimate of the price elasticity is very sensitive to this issue, as patent fees are volatile by nature. Second, it is useful to look at these three particulai- offices since they face the greatest challenges both in ternis of the increase in patent filings and in ternis of the backlog of applications. Third, on ^ more practical level, studies on fee elasticity are of interest to patent offices, which often are required to be self- financing. We présent a unique data set of patent fees charged in the U.S. (USPTO), Japanese (JPO) and European (EPO) patent offices since 1980. Descriptive sta,tistics show that fees severely decreased at the EPO over the 1990s. The estimation of dynamic panel data models of patent applications suggests that the long-term price elasticity is about -0.30,

Contribution to the literature

The thesis extends in many wa3's two previous studies. The motivation for the first two chapters is provided in de Rassenfosse and van Pottelsberghe (2009). This studj' puts forward the distinction between the productivity effect and the propensity effect in the R&D-patent relationship. It confirms that both dimensions explain heterogeneity in patent performance across countries and shows that high value patent indicators (such'as the count of patents filed simultaneously at the EPO, the USPTO and the JPO, called triadic patents) are reflective of a productivity effect, whereas priority patent applications respond to éléments of propensit}’.

The second study on which the thesis builds, de Rassenfosse and van Pottelsberghe (2007), takes a preliminary look at the price elasticity of demand for patents in the Member States of the European Patent Convention (EPC). Patent fees are found to be an important déterminant of the number of a country’s priority filings, with an estimated price elasticity that ranges froni -0.45 to -0.56.

The présent work introduces varions éléments of novelty to these two studies and to the literature at large, A first kej' contribution that niust be mentioned is methodological.

Although scholars agréé that patents are imperfect indicators of innovation performance, no study has explicitly solved the problem of the identification of productivity versus propensity in the R&D-patent relationship. Several attempts hâve been made to extract nieaningful information from patent data, however, which can be grouped into two distinct approaches.

The first consists in applying a geographical filter to the patent count in order to select the most valuable patents. The count of triadic patents is the most popular of these indicators (Demis et al., 2001; Grupp and Schnioch, 1999). In the second approach, patents are weighted

(12)

by One or several value indicators, sucli as the Family size (i.e., the number of countries covered by the patent), the number of citations, or the length of renewal (see, e.g,, Schankerman and Pakes, 1986; Trajtenberg, 1996). The solution proposed in Chapter 2 is similar in spirit to the existing techniques because it also relies on information on the value of [jatents. However, it differs in ternis of how the information is used and what information is produced. Whereas the existing techniques use arbitrai')' filters and weighting schemes, our identification strategy relies on clear assumptions. These assumptions can be challenged, but they hâve the merit of being made explicit, The resuit of the identification is also easy to interpret, which should be appealing to policymakers.

The effort made in collecting data is a second important aspect of the thesis. Inevitably, the résulta of the erapirical analyses are as reliable as the data on which they are based. This is the reason why a particular effort has been made to collect high quality and novel data. The data set used in Chapter 1 cornes froin an international survey of manufacturing firms and has never been exploited previously. The information collected by the survey is particularly valuable (notably the information on the propensity to patent and on the attitudes towards patents) and makes it possible to conduct a rich analysis of the déterminants of firms’ reliance on patents. Chapter 2 relies on a very detailed count of priority patent applications at the industry level. The large majority of studies that use patent data do so at the firm level or, to a lesser extent, at the country level. Industry-level data are rarely used — without doubt, the fact that patents are difficult to assign to industries has something to do with this. In addition, few, if any, studies that use jiatent data rely on priority filings. The use of priority filings nécessitâtes overcoming a major difficulty: there is a lot of missing information on the nationality of inventons and on the IPC classification.^ This information is essential to allocate patents to countries and industries. We hâve used and extended a technical method proposed by de Rassenfosse et al. (2010) which consists in extracting the missing information from the potentiel second filings of the patent document. Once the information has been recovered, each patent has been allocated to the industries and countries to which it belongs on a fractional basis. The patent count we obtained is thus both innovative and very detailed.

Chapters 3 and 4 also rely on novel data. Chapter 3, for instance, présents a long time sériés of patent application fees at the USPTO, dating back to its inception in 1790. Similarly, the analysis presented in Chapter 4 makes use of an original panel data set of patent fees at the USPTO, the EPO and the JPO. We computed these fees for a period of almost 30 years with a rigorous methodology to make them as comparable as possible over time and across countries.

Each chapter also brings new insights and results to the existing literature. Two readings of the results in Chapter 1 can be made. A first reading is that of the econometric pitfalls associated with using patent production functions to study the invention process. In this context, three findings must be emphasized; the positive impact of company size on the propensity to patent, the non-significant effect of R&D on the propensity to patent, and the low share of the variance in the propensity to patent that can be explained (about a quarter).

Since large companies tend to seek patent protection for a higher proportion of their inventions, patent data provide a picture of innovation performance that is not consistent across ail company sizes. On the contrary, the non-significance of R&D expenditure iinplies that the elasticity of inventions with respect to R&D can be accurately assessed with patent data. Ail

^IPC is the acronym for International Patent Classification. IPC codes are assigned by patent offices to indicate the different areas of technology to which the patent pertains.

(13)

Introduction XI

iii ail, however, patent data provide only a weak measure of innovation performance: the low degree of prédiction of the propensity to patent suggests that patent is a noisy indicator of inventive activity. A second reading of the results allows a better understanding of firms’

patenting behavior. Firms hâve very heterogeneous motivations to use the patent System, and these motivations hâve a significant impact on the size of their patent portfolio. We find that the majority of firms file patents to protect against imitation by competitors, and that having this motivation increases the propensity to patent, A large number of firms also apjjly for patents in order to protect their freedom to operate. Défensive patenting is thus not only widespread outside the patent System (Henkel and Pangerl, 2008) but also is at its very heart.

However, firms that use this strategj' do not exhibit a higher propensity to patent. One-fifth of the companies in our sample take patents with a view to licensing them out. These firms exhibit a higher reliance on patenting. On the contrary, firms that use patents to raise capital hâve a significantly lower propensity to patent.

Chapter 2 provides two novel insights worth discussing. First, it gives précisé estimâtes of the relative propensity to patent of Belgian, Dutch, French, and German companies. Ac- cording to our methodology, Belgian companies hâve roughly one-third of the propensity to patent of German companies (that is, they would file 30 patents where German companies would file 100). Dutch companies patent at about half the level of German companies, and French companies at about two-thirds the German level. Second,'the results suggest that similar patent-to-R&D ratios may Inde important productivity différences. Indeed, it is fre- quently observed that an industry may exhibit a low number of patents per unit of R&D in one country yet actually be more productive.than the same industry in another country where the patent-to-R&D ratio is higher.

The key finding of Chapters 3 and 4 is that patent fees can be used cis an effective policy tool to fine-tune patent Systems. Applicants are foimd to react to changes in patent fees, aJbeit with an elasticity that is lower than unity. The inelastic demand is actually a windfall for the main patent offices that are currentl3' struggling to keep afloat financially

Links between chapters

AU the chapters hâve as a focal point the propensity to patent and they can be put in the broader context of the use and the usefulness of patent indicators. The attentive reader will make links between the chapters and maj' notice apparent inconsistencies, which aie discussed in this section.

The importance of a country' effect on the propensity to patent goes unnoticed if the first two chapters are read separatelJ^ However, when the first two chapters are taken together, it becomes obvions that the propensity to patent lias a stroiig country component. In Chapter 1, country dumniies account for 12% of the variance of the propensity component. It is twice as large as the industry effect, which is traditionally considered as verj' important. The results presented in Chapter 2 also point in the direction of an important country effect. The ratio of propensities between two countries varies in a narrow range across industries (i.e., the propensity in Belgiuin is around one-third that in Germany regardless of the industry).

Although this resuit is not indicative of différences in propensity levels across industries, it does suggest a significant country effect. The attentive reader will notice that the country-level statistics presented in Figure 1.3 of Chapter 1 differ from the findings of Chapter 2. However, Figure 1.3 should not be put in direct comparison with Chapter 2 as the time period covered

(14)

is different (early 1990s vs. mid 2000s). Besides, the mean propensity rates in Figure 1.3 are based on a very limited number of observations. The standard errors around the mean level are admittedly very large.

Links can also be made between Chapter 1 and Chapter 4. Using different data sets and méthodologies, the two studies confirm the significant effect of patent fees on the propensity to patent. While Chapter 1 présents estimâtes of the effect of fees on the propensity to patent using a large cross-sectional sample, Chapter 4 looks at the effect of fees on the total number of patent filings using a long narrow panel. Despite these important methodological différences, the price elasticity of demand for patents is similar. The elasticity parameter computed from the marginal effects presented in Table 1.5 is about -0.5, which is remarkably close to the parameter estimated in Chapter 4. The reader may be surprised by the spécification of the patent production fonction in Chapter 4, especially in light of Chapters 1 and 2, which call for a good modeling of the propensity and the productivity effects in patent production functions. The focus in Chapter 4 is exclusively on the price elasticity parameter and not on the other déterminants of the patent production function. Unobserved productivity and propensity effects are thus simply captured by country fixed effects and time effects. It is unlikely that patent fees are correlated with missing propensity (or productivity) variables such that blases are unlikely to arise. In addition, since patent fees do not influence the productivity of research (at least if fees are not unreasonably too low or too high), the impact of fees on the propensity to patent can be assessed reliabl3'. The statement in Chapter 3 that fees matter and can be used to fine-tune the patent System may also seem to be at odds with the very low price elasticity' of patents estimated in Chapter 1 and Chapter 4. How can a fee policj' be effective if a patent is an inelastic good? Again, the inconsistency is only apparent, as moderate changes in fees maj' hâve strong observable effects. A tj'pical example is the EPO fee policy of the mid 1990s, which, according to our estimâtes, is responsible for as much as one-fifth of the growth in patent applications over that period. This estimation is still smaller than that of Eaton et al. (2004), who attribute 60% of the increase in the number of patent filings at the EPO during the 1990s to the decrease in the overall cost of seeking protection.

Perspectives

It is always useful to step back and think about research perspectives. The end of a thesis is certainly an appropriate moment to draw up a list of potential research questions. Among the varions projects that I hâve in mind, several are directly related to the présent work.

Chapter 1 could be extended in two important ways. First, it would be worth extending the econometric framework to the analysis of patent production functions. One could think of modeling the number of patent filings taking explicitly into account the propensity to patent.

This would make it possible to gauge the magnitude of the extra information provided by the propensity component and would allow a proper identification of déterminants of the productivity and the propensity effects. Second, another source of bias in patent statistics that is overlooked in the literature could be investigated using the same empirical framework.

Most of the studies that rely on patent data use patents filed at the EPO or at the USPTO.

However, EPO and USPTO patents are only a subset of ail the patents filed by a company, at least as far as European companies are concerned. Since the decision to file at the EPO rather than at the national patent office is probablj' not random, there is a risk that the count of EPO patents may bias the estimation of the parameters of patent production function. Two

(15)

Introduction xiii

studies can be envisaged; one that is specifically concerned with the biases, and another that looks at the déterminants of the géographie location of patent filings.

An obvions extension to Chapter 2 is to perform the analysis at the country level. This exercise is likely to lead to very interesting insights, as the results will suffer less from potential measurenient errors of R&D and misallocation of patents to industrial sectors. Once the country-level productivity and propensity components hâve been computed, it will also be possible to assess their validity. In particular, one could regress the productivity (propensity) component on variables that are known to affect the productivity of research (the propensitj' to patent). The industry level analysis we could conduct in the thesis was limited by the missing information on IPC codes. The missing data, however, should not jeopardize industry-level analyses, w'hich remain promising. One can think of varions ways to identify missing IPC codes. For instance, the IPC codes of backward and forward citations to a patent document could provide reliable information on the missing IPC codes of the patent document itself.

A third extension would be to apply the methodology to firm-level data. A useful starting point would be to collect information on the patent portfolio of finns in the sample used in Chapter 1 and test whether the propensity rates recovered correlate with the propensity rates observed in the sample. A significant enough corrélation would mean that the methodology can be used to separate the productivity from the propensity effect at the firm level, therefore significautly contributing to the usefulness of patent statistics.

An important question left open in Chapter 4 concerns the impact of .patent fees on the quality of patent applications. So far, indeed, there is no evndence that higher fees are effective in weeding out low quality patents. Early evidence, discussed by Nicholas (2010), on the substantial drop in patent fees that occurred in Britain in 1883 leads to uncertain conclusions. The sudden rise in application fees at the USPTO at the beginning of the 1980s offers a very interesting natural experimental setting to investigate this question. I am not aware of any study that looks at the impact that this fee increase lias had on the overall quality of applications filed.

References

Anton, J., Yao, D., 1994. Expropriation and inventions: Appropriable rents in the absence of property rights. The American Economie Review, 84(1), 190-209.

Blind, K., Edler, J., Piietsch, R., Schmoch, U. 2006. Motives to patent: Einpirical evi­

dence fi'om Germany. Research Policy, 35(5), 655-672

Cohen, W., Nelson, R., Walsh, J., 2000. Protecting their intellectual assets: Appropriability conditions and why U.S. manufacturing firms patent (or not). NBER Working Paper 7552 de Rassenfosse, G., Demis, H., Guellec, D., Picci, L., van Pottelserghe de la Potterie, B., 2010. A corrected count of priority filings. OBGD Working Paper, fortheoming.

de Rassenfosse, G., van Pottelsberghe de la Potterie, B., 2007. Per un pugno di dollari:

A first look at the price elasticity of patents. Oxford Review of Economie Policy, 23(4), 588- 604.

(16)

de Rassenfosse, G., V'an Pottelsberghe de la Potterie, B., 2009. A policj' insight into the R&D-patent relationship. Research Policy, 38(5), 779-792.

Demis, H., Guellec, D., van Pottelsberghe de la Potterie, 2001. Using patent counts for cross-country comparisons of technology output. OECD STI Review, 27, 129-146.

Eaton, J., Kortum, S., Lerner, J., 2004. International patenting and the European Patent Office: A quantitative assessnient. Patents, Innovation and Economie Performance: OEGD Gonference Proceedings, 27-52.

Giuri, P., Mariani, M., Brusoni, S., Crespi, G., Prancoz, D., Gambardella, A., Garcia-Fontes, W., Geuna, A., Gonzales, R., Harhoff, D., Hoisl, K., Le Bash, G., Luzzi, A., Magazzinia, L., Nestac, L., Nomaler, O., Palomeras, N., Patel, P., Romanelli, M., Verspagen, B. 2007.

Inventors and invention processes in Europe: Results from the PatVal-EU survey. Research Policy, 36(8), 1107-1127.

Griliches, Z., 1990. Patent statistics as économie indicators: A survey. Journal of Eco­

nomie Literature, 28(4), 1661-1707.

Grupp, H., Schmoch, U., 1999. Patent statistics in the âge of globalisation: New legal proce­

dures, new analytical methods, new économie interprétation. Research Policy, 28(4), 377-396.

Relier, M., Eisenberg, R., 1998. Can patents deter innovation? The anticommons in bioméd­

ical research. Science, 280(5364), 698-701.

Henkel, J., Pangerl, S., 2008. Défensive publishing - An empirical study. Technische Uni- versitàt München working paper.

Lanjouw, J., Pakes, A., Putnam, J., 1998. How to count patents and value intellectual property: The uses of patent renewal and application data. The Journal of Industrial Eco­

nomies, 46(4), 405-432.

Levin, R., Klevorick, A., Nelson, R., VVinter, S., 1987. Appropriating the returns from in­

dustrial research and development. Brookings Papers on Economie Activity, 1987(3), 783-831.

Mansfield, E., 1986. Patents and innovation: An empirical study. Management Science, 32(2), 173-181.

Nicholas, T., 2010. Gheaper patents. Mimeo, Harvard Business School.

Schankerman, M., Pakes, A., 1986. Estimâtes of the value of patent rights in European countries during the post-1950 period. The Economie Journal, 96(384), 1052-1076.

Trajtenberg, M., 1990. A penny for your quotes: Patent citations and the value of inno­

vations. The RAND Journal of Economies, 21(1), 172-187.

(17)

C

hapter

1

How much do we know about firms’ propensity to patent and should we worry about it?*

The objective of the paper is twofold. First, it explaiiis the econometric pitfalls that may arise when patent production functions are used to study the invention process. In order to avoid these pitfalls, it is argued that researchers should strive to understand the déterminants of the propensity to patent. Second, the econometric analysis seeks to explain the proportion of inventions that are patented — a potential metric for the propensity. It relies on data from an international survey of industrial firms. Four key dimensions are found to influence the propensity to patent: firm characteristics, firm’s attitude towards patents, market factors and IP policies.

Keywords: fractional response variable, patent production fonction, propensity to patent, proportion data, knowledge production function

JEL Classification: L60, 030, 031, 034

'The author is gratefui to Bruno Cassiman, Michèle Cincera, Jérôme Danguy, Catherine Dehon, Keld Laursen, Georg Licht, Carine Peeters, Bruno van Pottelsberghe and Nicolas van Zeebroeck for early comnients as well as to the participants of the ZEW Workshop in Innovation ir Compétition (Mannheim, Dec. 2009) and two anonymous referees of the Tilburg Conférence on Innovation.

(18)

1.1 Introduction

Patent data are widely used to study the drivers and the conséquences of the innovation pro- cess. In particular. patent statistics are frequently used to investigate firms’ innovative output by means of so-called ‘’patent production functions” which relate a firm’s patent applications to its R&D spending. It is often heard, however, that patents are imperfect proxies for in­

ventions: not ail inventions are patentable and not ail patentable inventions are patented. In addition, patents diflfer in their intrinsic scope and value, and strategie behaviors by firms add further noise to the data. As a matter of fact, the R&D-patent relationship is composed of the RidD-invention relationship and the invention-patent relationship, which capture a productiv- ity effect and a propensity effect respectively. In this respect, a Sound understanding of the invention process necessarily requires an understanding of the déterminants of the propensity to patent, “one of the most important questions conceming the patent System” according to Mansfield (1986).

The objective of the paper is twofold. First, it explains the pitfalls associated with the use of patent production functions to study the invention process. They are related to missing variable bias, identification problem and interprétation of measures of goodness-of-fit. The discussion is supplemented with numerical simulations to gauge the magnitude of the effects.

In order to mitigate these effects, we argue that researchers should strive to better understand the déterminants of the propensity to patent. It is precisely the second objective of the paper.

Using data from an international survey of industrial firms, the econometric analysis seeks to explain the proportion of inventions that are patented — a potential metric for the propensity to patent. Overall, firms in our sample patent roughly 50% of their inventions. Four broad déterminants are found to influence a firm’s decision to apply for a patent: its characteristics, its attitude towards patenting, market factors and intellectual property (IP) policies.

The econometric analysis brings new insights into firms’ reliance on patents. Three main empirical findings are worth noting. First, the propensity to patent increases with firm size.

As a resuit, patent data overlook innovation in small firms. Second, patent fees are a strong deterrent to patenting. To the best of our knowledge, this paper is the first to provide micro- level evidence of the rôle of fees on the demajid for patent. Third, the decision to apply for l>atents is significantly impacted by firms’ attitude towards patent. The majority of firms file patents to protect against imitation bj' competitors. In addition, firms that hâve this motivation hâve a higher pro])ensity to patent. A large number of firms also apply for patents in order to protect their freedom to operate. However, firms that do so do not exhibit a higher propensity to patent. Monetary motivations are of particular importance to SMEs: 40% of SMEs in our sample take out patents to convince iuvestors or banks of the value of their invention. Overall, firms that use patents to raise capital tend to hâve a lower propensity to patent, whereas firms that take out patents with a view to licensing them out exhibit a higher leliance on patenting. Yet, the drivers of the decision to apply for patent remain largely unknown. The best econometric spécification explains at most a quarter of the variance of the propensity to patent.

The paper is organized as follows. The next section discusses the pitfalls that may arise when patents are used as output to the “knowledge production function” and briefly présents the déterminants of the propensity to patent as reported in the literature. The data and the econometric framework are introduced in Section 1.3 and Section 1.4 présents the results of the empirical analysis. Section 1.5 discusses the implications of the results and concludes.

(19)

1.2 The propensity to patent 3

1.2 The propensity to patent

This section starts by defining the propensity to patent. It then explains how tins impacts the estimation of parameters of patent production functions. In so doing, it calls for a good understanding of the déterminants of the propensity. Three ways of assessing its déterminants are then presented, and the main empirical findings are surveyed.

A common définition of the “propensity to patent” is the number of patents per R&D dollar (see, e.g., Griliches, 1990, p. 1678 and Hall and Ziedonis, 2001, p. 102). The patent- to-R&D ratio, however, is sometimes referred to as the “research productivity" (see, e.g., Lanjouw and Schankerman, 2004, p. 441), thereby raising some concerns as to what the ratio really measures and how the terni “propensity to patent” should be understood. To the best of our knowledge, the terni was first coined in Schmookler (1962) and Scherer (1965).

If neither of the two authors explicitly defines it, it seems that tliej' refer to the extent to which firins rely on patent protection to appropriate the returns to invention. For instance, Scherer (1965, p. 1101) writes that “a crude indicator of différences in the propensity to patent is différences in average patent output per unit of engineering input." Clearly, Scherer suggests that the patent-to-R&D ratio is a proxy for the propensity to iratent rather than the propensity per se. Throughout the paper, the term propensity to patent is used to refer to the proportion of inventions that are patented. This use of the term is in line with authors such as Mansfield (1986, p. 178) or Cohen et al. (2000, p. 16). To avoid confusion, the patent-to- R&D ratio should be better termed “patent intensity” or “patent rate.” These ternis are more neutral and do not hide the fact that the R&D-patent relationship actually captures both a productivity effect, which leads from R&D to inventions, and a propensity effect, which leads from inventions to patents.'

1.2.1 Why does the propensity to patent matter?

The management literature has been devoting considérable attention to the means available to protect profits accruing from inventions. Secrecy, first mover advantage and complementary assets hâve been found to be important appropriability mechanisms, together with patent protection (see, e.g., Cohen et al., 2000; Levin et ah, 1987; Teece, 1986). In this respect, understanding the extent to which firins rely on the patent System is an important building block in this literature (Mansfield, 1986). In the following, we focus on an important aspect of the propensity to patent that has been neglected in the literature. In a nutshell, since patent data are often used to study firins’ innovation performance, the propensity to patent may act as a filter that blurs the information on research productivity conveyed by patent data.

We can consider in a very general way that research efforts {R) lead to inventions (7) through a productivity effect A and inventions, in turns, lead to patents (P) as a fonction of the propensity to patent <5. We can write

*Daiiguy et al. (2009) further décomposé the propensity to patent into the appropriability propensity — the share of inventions patented — and the strategie propensity — the number of patents filed to protect a given invention. Although this fonnahzation is conceptualJy appropriate, we nevertheless stick to the traditional view for ease of exposition. Throughout the paper, the term ‘propensity to patent” refers to the share of inventions patented.

(20)

I = (1.1)

P = 61 (1.2)

The parameter A is the rate at which research efforts lead to inventions and 5 e [0,1] is tfie proportion of inventions that are patented. In practice, however, onl)' the patented output is observed such that the following patent production function must be estimated

P = <pR° (1.3)

wliere = A5 is the patent intensity.^ That is, tlie unobserved knowledge production function (1.1) is proxied by the patent production function (1.3). Since / is unobserved, so are A and 5 and there is an identification problein. As Lanjouw and Schankerman (2004, p. 442) put it:

“/..•/ it is useful to break the patent to R&D ratio into its two coniponerit parts: the patent to invention ratio and the invention to R&D ratio. A fait in mcasured research productivity may be real — a declining invention/R&D ratio — or only apparent - a declining patent/invention ratio. Since we do not nonnally hâve information on the number of inventions, there is an identification problem in inlerpreting changes in the patent to R&D ratio. What appears to be technological exhaustion may simply be mismeasurement."

Most researchers implicitly acknowledge the limitation of patent production fonctions with the often-quoted sentence that “not ail inventions are patentable and not ail patentable in­

ventions are patented.” Yet, they often overlook the effects induced by différences in the propensity to patent across firms. In what follows, we explain how looking at the innovation process through the leus of patent data potentially distorts the reality. Generallj' speaking, probleins arise when propensity variables are correlated with productivity variables. The following thought experiment illustrâtes four conséquences that scholars must bear in mind.

They relate to the oinitted variable bias, an identification problem and the interprétation of measures of goodness-of-fit. Although these pitfalls are well-known in standard econo- metric theory, it is useful to briefiy review them in order to frame the empirical analysis.

Without imposing rmy functional form for A and 6, we can write that I = f(L,R,ei) and P = g(f(L,R,Si),'D,Ep) where L and D vepresent the set of déterminants of A and S re- spectively, and a and Sp are indépendant nuisance parameters. Assume that, say, the size of the firm (S) is the only variable that influences the count of patents, beside R (that is, 5 e LUD).

i. If we assume that larger firms are more likely to patent, omitting S in the régression will upwardly bias the estimated elasticity of R, a, to the extent that R and 5 are correlated (the so-called omitted variable bias). It corresponds to a situation where the spécifi­

cation P = g{f{R,Ei),£p) is used to estimate P = g{f{R,ei),S,ep). H is particularly important to include ail the propensity variables that are potentially correlated with the

^More complex spécifications of the patent production function can be envisaged, such as adding knowiedge stock or spiiiovers (see, e.g., Cincera, 1997; Jaffe, 1986), however, they ali boi! dow'n to the simple model presented here. Knowledge stocks, for instance, impact tire productivity of research and are thus implicitly captured by A.

(21)

1.2 The propensity to patent 5

productivity variables to ensure unbiased estimâtes of the déterminants ofthe knowledge production function. Note that if a propensity variable is not correlated with any of the productivity variables, oinitting it will only reduce the performance of the régression model. Table 1.6 iii Appendix A.l illustrâtes the potential bias induced bj' a raissing (correlated) propensity coinponent for a set of realistic paranieter values. Under some circumstances, the bias on a can be as high as 10 percentage points.

ii. If we now assume that larger firms are not only more likely to patent but also more productive (that is, S € L n D), the parameter associated with S will be inflated and the impact of this variable on the production of inventions will be overestimated. In the spécification P = g{f{S,R,£i),S,£p), the estimated impact of S on the knowledge production function will be inflated to the extent that S also increases the propensity to patent. A similar problem arises with knowledge spillovers. Spillovers increase the pro­

ductivity of research, but their existence is au additional incentive to patent, such that the true impact of spillovers on the productivity coinponent is difficult to estimate. In addition, the estimated parameter can turn out to be non-significant if the productivity and the pro]jensity effects cancel out. Indeed, if smaller firms are, say, more productive but larger firms are more likely to patent, no significant impact of 5 on the iratent production function can be observed although it is a crucial déterminant of both effects.

It is thus impossible to estimate the correct impact of a variable on the productivity if it affects both dimensions. As an illustration, Table 1.7 in Appendix A.l reports the results of simulations in which the impact of the productivity variable is reinforced or weakened. The effect of S on the number of patents may vanish or double, depending on the spécification.

iii. More generally, it is impossible to distinguish productivity from propensity effects. If the researcher lias no a priori knowledge on the expected impact that the size of a firm has, the regiession will not help him identify the channel(s) of action. This problem should not be overlooked, as many variables hâve an ambiguous interprétation. For instance, R&D collaboration may lead to a more productive research but could also simply increase the propensity to patent (see, e.g., Peeters and van Pottelsberghe, 2006;

Czarnitzki et al., 2007). Similarly, a high degree of compétition may force companies to adopt more stringent management practices inside the research lab, thereby raising the productivity of research, but may also increase companies’ reliance on patents to appropriate the returns to innovation. Very few authors conduct formai tests to identify the channel of action. One example is Kortum and Lerner (2000, p. 689). They find a strong relationship between venture capital and patenting but are worried that this relationship might reflect a propensity rather than a productivity effect: “/.../ one major concem remains. In particular, it might be thought that the relationship between venture capital disbursements and patent applications is not indicative of a relationship between venture disbursements and innovative output. It may be that the increase in patenting is a conséquence of a shift in the propensity to patent innovations stimulated by the venture financing process itself." They then présent evidence based on alternative measures of innovation to support their daim. Note, however, that the authors were not able to exclude the possibility that venture capital might also affect the propensity to patent.

iv. Fourth, even though an econometric model accounts for ail the relevant éléments of

(22)

productivity, it could still explain a low share of the variance of paient counts, giviiig the feeling that the innovation process is not well inodeled. A complété spécification of the invention process will lead to a good P? only if the nuisance parameter £p is not too big.

In this respect, tlie propensity to patent acts as a filter that blurs tlie R&;D-invention relationship. A full spécification of tlie innovation process inay translate into a very low rp if the déterminants of the propensity to patent are not accounted for. Figure 1.4 in Appendix A.l show's how neglecting the propensity component affects ineasures of goodness-of-fit. Our simulations suggest that the P? is very sensitive to the variance in the propensity to patent. In some cases, a P? of 0.70 (or a pseudo P? lower than 0.20) can be achieved even though the innov'ation function is well modeled and its parameters are properly estiinated.

As the discussion illustrâtes, patents are potentially biased proxies for inventions and these pitfalls are rarely acknowledged — nor controlled for — in the empirical literature.^ As long as patent data are used to study firms’ innovation performances, the previous discussion suggests that researchers should strive to understand the déterminants of the propensity to patent. According to Mansfield (1986, p. 173), understanding “fo what extent do firms make use of the patent System, and what différences exist among firms and industries /.../ in the propensity to patent” is “one of the most important questions conceming the patent System." It is ail the more true nowadays with the great success that patent data hâve had with scholars and policymakers.

1.2.2 Known déterminants of the propensity to patent

One can think of three broad ways to investigate the déterminants of the propensity to patent.

In an idéal set-up, one would observe both the raw innovative output and the number of patents and would thus be able to assess the déterminants of the probabilitj' that an invention is patented. The first approach is close to the idéal set-up. It consists of matching a set of inventions (such as inventions presented at trade fairs or awarded a jirize) with actual patent data so that researchers can control for both firm level and invention level characteristics."* The second, probably most common, inethod consists in estimating a patent production function where only the inputs to the innovation process and the patented output are observed (see Scherer, 1965 and Hausmann et al., 1984 for pioneering works). Yet, as the discussion in Section 1.2.1 illustrâtes, this approach leads to limited insights. Patent production functions will provide only weeik evidence unless there Is no apparent reason to suspect that the variable has an impact on the productivity of research.® The third approach uses information at the portfolio level such as the proportion of inventions that are patented. Because it cannot account for invention characteristics but ha.s sonie information on non-patented inventions, it lies somewhat in between the two other alternatives. Few authors hâve reported propensity

^This is not to say that scholars are too hasty. Other econoinetric pitfalls related to patent data hâve been extensively discussed in the literature and are generally well-treated (see, e.g.. Crépon et al., 1998, for the sélection bias in innovation surveys).

■*Unfortunately, very few studies hâve taken that direction. Three examples are Moser (2009), Kleinknecht and van der Panne (2009) and Fontana et al. (2009). Although this approach is appealing, it is unclear whether a sélection bias is likely to affect the findings.

^Other econometric methods that link R&D to patenting hâve been used. For instance, Sorensen and Stuart (2000) use Cox models to model the déterminants of the time elapsed until a firm issues a patent. Brouwer and Kleinknecht (1999) and Peeters and van Pottelsberghe (2006) use logistic régressions for predicting the probability that a firm applies for at least one patent. AU these approaches are subject to the same caution.

(23)

1.2 The propensity to patent 7

rates in the past, therefore much can be learned frora this approach. We hâve identified only three indepeudent studies that report data on the proportion of inventions patented: Mansfield (1986), Arundel and Kabla (1998) and Cohen et al. (2000). Among them, onlj' Arundel and Kabla (1998) hâve carried an econometric anal3'sis. They investigate the déterminants of tlie propensity to patent of Europe’s largest industrial firms. This study is complemented bj' that of Duguet and Kabla (1998) who focus on the sample of French manufacturing firms. A third econometric study is that of Arora et al. (2008). In an attempt to quantify the patent premium (the extent to which patenting stimulâtes R&D), they estimate the proportion of inventions patented using data from the Carnegie Mellon Survey by Cohen et al. (2000).

A firm applies for patent protection if the expected benefits from patenting exceed the costs; the déterminants of patenting must thus be analyzed within this framework. The following five dimensions hâve been found to influence the propensity to patent to varions extents.

□ Invention characteristics and the invention process. Little evidence is available as the propensity to patent is rarely studied at the invention level. There are retisons to think that the propensit}' to patent increases with invention value. For one thing, more valuable inventions hâve more chances of arousing interest of competitors. Moser (2009) finds that inventions presented at 19th- and 20th-century world faits were more likelj' to be patented if they were awarded a prize. The invention process may also influence the decision to applj' for patent. Brouwer and Kleinknecht (1999), for instance, estimate that firms engaged in R&D collaboration hâve a higher propensity to patent. In a similar vein, Fontana et al. (2009) find evidence that the propensity to patent increases with the number of inventors involved and Kleinknecht and van der Panne (2009) hâve shown that inventions involving manj' partners for R&D collaboration are more likely to be patented.

□ Firms’ characteristics. The literature is not conclusive as to whether propensitj' rate increases with firm size. While some studies report evidence in favor of this hy- pothesis (Arundel and Kabla, 1998; Arora et al., 2008) other studies that use patent production functions find no efîect of firm size (Crépon et ah, 1998; Duguet and Kabla, 1998). Brouwer and Kleinknecht (1999) show that small firms are less likely to patent but that, conditional on patenting, they hâve a proportionally larger patent portfo­

lio. It is however more likelj' that the propensity to patent increases with firm size.

Small companies generally face a high cost of patenting: their lack of expérience in the patenting process and their liinited ability to control and pursue infringement seem to be strong hampering factors (see, e.g., Graham et ah, 2010; Lanjouw and Schankerman, 2004b). Large companies, on the contrarj', maj’ benefit from économies of scope when one invention can be embedded into niany products and may rely more on patents for strategie reasons. A firm’s organizational structure and patent law expertise may also be a strong déterminant of the decision to apply for patents. Somaya et ah (2007) show that firms with in-house patent law expertise hâve on average a larger patent portfolio.

Reitzig and Puranam (2009) find evidence that a firm’s organizational structure affect the speed at which it is granted patents. The impact of the size of the research budget is particularly worth discussing. For a given company size, it is reasonable to assume that an increase in the R&D intensity maj' either increase the number of inventions produced or raise the average value of inventions. In the former, the propensity to patent may

(24)

decrease or, at best, stay constant whereas in the latter it ma}' increase or, at worst, stay constant according to whether the IP budget is flexible. In any case, a signifi- cant impact of R&D expenditures on the propensity rate would necessarily imply that the elasticity of patents with respect to R&D is a biased estimate of the elasticity of inventions with respect to R&D. Empirical evidence is scarce and contradictory. Arun- del and Kabla (1998) find that R&D intensity does not impact the share of inventions patented, Duguet and Kabla (1998) on the contrary provide convincing evidence that R&D expenditures increase the propensity to patent.

□ Firms’ attitude towards/use of patents. Uses of patents that go beyond the tradi- tional protection motive may account for large intra-firm différences in the propensity to patent. Companies hâve been found to use patents in varions ways, such as for pre- venting suits, enhancing réputation or as bartering chips in negotiation (e.g., Blind et ah, 2006; Cohen et ah, 2000; Duguet and Kabla, 1998; Hall and Ziedouis, 2001). Duguet and Kabla (1998) find that firms that apply for patents to acquire a stronger position in technology negotiation and to avoid trials geneially hâve a larger patent portfolio. In this paper, we focus on three uses: tlie use of patents to secure one’s own freedom to operate, to raise capital and to generate licensing revenues. Little is known of the im­

pact that these motivations hâve on the propensity to patent. It is discussed in section 1.3 when we State the empirical hypothèses.

□ Market factors. On the demand side, it is reasonable to assume that the market size and its wealtk play a significant rôle. Firms in larger and richer market hâve more chance to recoup their investment and hence can afford to patent inventions of lower value. We are not aware of studies that confirm or invalidate this argument. On the supply side, the intensity of compétition partly détermines the degree of appropriability.

An intense compétition pushes firms to reinforce their appropriability strategy and, hence, to apply for patent protection. In theory at least. Arora et ah (2008) find that the number of technology rivais, as a measure of the prospective imitators, does not explain the propensity to patent in the U.S. manufacturing industry. Results presented in Duguet and Kabla (1998) go in the saine direction: firms’ market share and the industrial concentration do not impact patent practices. Here, studies that rely on patent production fonctions are of limited use since a significant impact of compétition on patent count may as well be attributed to a productivity effect.

□ IP-related policies. As a matter of fact, IP policies hâve a direct impact on the propensity to patent. For instance, a lowering of the grant standard increases the prob- ability of having its patent gianted and hence raises the expected benefit of patenting.

Conversely, high patent fees are a deterrent to patenting. de Rassenfosse and van Pot- telsberghe (2007, 2008) estimate the fee elasticity of demand for priority filings to be between -0.30 and -0.50. The strength of the patent System is sj'stematically found to increase the propensity to patent, as a strong IP régime mitigates the cost of disclosing technological information by providing a good protection to the patentée.

Note that other déterminants were found to influence the size of a company’s patent portfolio such as the source of funding for research. However, because it is not clear whether the variables induce différences in the productivity of research or in the propensity to patent, they were not reported. Siinilarly, some déterminants of patenting discussed in this section

(25)

1.3 Data and econometric methodology 9

also affect the incentives to conduct research or tlie productivity of research. This is typically the case for the strength of IP protection (see, e.g., Mazzoleni and Nelson, 1998). The présent empirical exercise is useful in that it allows a proper identification of the déterminants of the propensity to patent.

1.3 Data and econometric methodology

The data are collected at the firm level frora an international survey on patent practices. They are supplemented with country-level data collected from other sources.

Firm-level data corne from the EPO Applicant Panel Survey carried ont from June to September 2006.'’ The main purpose of the survey is to provide information on filing inten­

tions for the EPO’s forecasting exercise for budgetary planning purposes. A sample of 2,098 applicants was selected, partly from the largest applicants and partly at random, cov'ering overall about 31% of the total applications at the EPO. Contact details were successfully ' established for 1,524 applicants and 772 responses were returned (leading to a response rate of 51% of the contacted applicants, or 37% of the initial sample). The survey was carried out via téléphoné and mail interviews in German, Prench, Japanese, and Euglish with the pre-estahlished contact persons from roughly 30 countries.

A spécifie part of the questionnaire was dedicated to more qualitative questions. Appli­

cants were explicitlj' asked for the share of inventions that were patented throughout the world in 2005 (variable PROPORTION), together with general background information such as the level of their sales or their research budget. They were also asked to rate a sériés of assertions hetween 1 (totally disagree) and 6 (full}' agréé) related to their motivations to patent. Note that the allocation of applicants into industries follows the EPO industry classification in 14 joint clusters (see Appendix A.2).

The data are completed with other country-specific information from external sources. The number of inhabitants and the GDP per capita in 2005 corne from the IMF World Economie Outlook database. Patent fees are the fees up to the grant for a priority patent application in each of the 26 countries represented in the sample, as reported in de Rassenfosse and van Pottelsberghe (2007, 2009).^ Finally, Park (2008) provides an index of the strength of IP protection.

The econometric analysis seeks to understand the déterminants of the propensity to patent using the proportion of inventions that are patented as dépendent variable (PROPORTION).

Given the nature of the data at hand, we cannot control for invention characteristics. However, we can test for the four broad dimensions that affect the propensity to patent: the firm’s characteristics, its attitude towards patent, market factors and IP policies.

1.3.1 Independent variables

□ Firm’s characteristics. Among the firm’s characteristics (denoted by Xp in the em­

pirical analysis), the impact of its size and its research effort are of particular interest.

®The survey was performed by a private contractor on behalf of the EPO. More information on the sainpling methodology is available on the EPO website (http://www.epo.org). The autlior is grateful to Peter Hingley for having provided access to the data.

^Data are for the year 2003. Fees for Taiwan were not available, they w^ere assumed to be siinilar than those in China.

Figure

Updating...

Sujets connexes :