knowledge spillovers and intellectual property rights

International Journal of Industrial Organization 31 (2013) 50–63

Contents lists available at SciVerse ScienceDirect

International Journal of Industrial Organization

j ourna l homepage: www.e lsev ie r .com/ locate / i j io

Knowledge spillovers and intellectual property rights☆

Roberto M. Samaniego ⁎Department of Economics, George Washington University, 2115 G St NW Suite 340, Washington, DC 20052, United States

☆ I am grateful to the editor and two anonymous referwell as Alain Gabler, Boyan Jovanovic, Yu Sun, Evangeliaparticipants at the 2011 Midwest Macro Meetings atCIRPÉE-IVEY Conference on Macroeconomics and Entretions. All errors are the authors'.⁎ Tel.: +1 202 994 6153; fax: +1 202 994 6147.

E-mail address: [email protected] See Cohen (2011) for an extensive survey of the ide

appropriability.

0167-7187/$ – see front matter © 2012 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.ijindorg.2012.11.001

a b s t r a c t
a r t i c l e i n f o
Article history:Received 23 March 2012Received in revised form 3 September 2012Accepted 7 November 2012Available online 19 November 2012

JEL classifications:L26O14O16O31O33O34

Keywords:Knowledge spilloversR&D intensityEntrepreneurshipIntellectual property rightsCopyrightLegal formalism

Knowledge spillovers are widely thought to be important for innovative activity, yet theory is ambiguousabout the sign of the relationship. Assuming that knowledge spillovers are more easily exploited where intel-lectual property rights are weakly enforced, this paper uses country–industry data to uncover the link be-tween knowledge spillovers and innovative activity, as well as the birth and death of enterprises. IPRenforcement disproportionately increases innovation spending in R&D intensive industries, as well as bothrates of entry and exit. The results are robust to accounting for financial development, labor market ridigitiesand a number of other institutional factors.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

An intrinsic feature of knowledge is that it is non-rival and imperfect-ly excludable — see Romer (1990). Imperfect excludability is typicallyinterpreted as a technological feature of knowledge that implies thatnew knowledge, once generated, may be used by agents other than theinnovator — a feature commonly known as “knowledge spillovers”. Theterm “knowledge spillovers” may also refer to the ability of an agent toproduce new knowledge by building on prior knowledge, possibly in-cluding the agent's own stock of knowledge. Thus, knowledge spilloversconstitute a factor of technological opportunity— affecting the yield of in-novative effort — and also of appropriability — affecting the ability ofagents to capture the returns of their innovative effort.1

ees for insightful comments asVourvachaki, Neng Wang andVanderbilt University and thepreneurship for useful sugges-

ntification of opportunity and

rights reserved.

Although theory suggests that knowledge spillovers across agentsshould be related to the quantity of innovative activity, the sign of thelink between spillovers and innovation is ambiguous. On one hand,large spillovers might encourage innovation by providing would-beinnovators with something to build upon or by allowing the rapid dif-fusion of new knowledge. On the other hand, large spillovers mightdiscourage innovation because an innovator's competitors also benefitfrom the generation of new knowledge (be it through imitation or in-spiration). In addition, whether new knowledge is primarily a substi-tute or a complement to existing knowledge is ambiguous too. Sinceincumbents are better positioned to have accumulated past knowl-edge, the impact of spillovers on entry and exit may also help refinethe empirically relevant set of theoretical models for understandingthe process of innovation.2

A key observation made in Romer (1990) is that the impact ofspillovers on innovative behavior depends not only on the technologyof knowledge generation but also on institutions. For example, if

2 For example, in the “creative destruction” models of Aghion and Howitt (1992,1997), the primary beneficiaries of knowledge spillovers are entering firms, whereasin Klette and Kortum (2004) knowledge spillovers benefit entrants and incumbentsequally, and in Peretto (1998) they favor large incumbents.

http://dx.doi.org/10.1016/j.ijindorg.2012.11.001

mailto:[email protected]

http://dx.doi.org/10.1016/j.ijindorg.2012.11.001

http://www.sciencedirect.com/science/journal/01677187

51R.M. Samaniego / International Journal of Industrial Organization 31 (2013) 50–63

intellectual property rights (IPRs) are not well-enforced, or if IPR dis-putes are costly and unpredictable, then appropriability is weakerthan otherwise. Moreover, this should be particularly noticeable inindustries in which the technologically determined extent of knowl-edge spillovers across agents is large — i.e. where opportunity isalso high.3 This suggests that exploiting variation across countries inIPR enforcement, together with variation across industries in innova-tion activity, may be useful for uncovering the impact of knowledgespillovers on innovation.

Consider measuring research intensity in a country where IPRs arestrong, and where financial, labor and product markets are relativelyfrictionless. This provides a benchmark for innovative behavior whenthe impact of knowledge spillovers on appropriability is limited.Then, assuming appropriability encourages innovation primarily inindustries with large knowledge spillovers, whether innovative activ-ity in industries that are R&D intensive in the benchmark environ-ment decreases disproportionately with a weakening of IPRs shouldindicate whether these are industries where potential spillovers arevery large.4 Furthermore, whether rates of entry are also dispropor-tionately affected in R&D-intensive industries, and whether the dis-proportionate impact is positive or negative, should indicate therelative importance of entrants and incumbents in taking advantageof these spillovers. Finally, whether the behavior of entry and exit isdisjoint indicates whether the spillover-induced replacement of in-cumbents by innovating entrepreneurs is an important feature ofthe process of innovation.

This paper implements the empirical strategy outlined above, toidentify the link between institutions that limit costly IPR disputesand research intensity, as well as entry and exit. The paper exploitscountry–industry variation in rates of entry, exit and innovation indi-cators to understand whether knowledge spillovers discourage inno-vation, and whether entry or exit play an important role in thisprocess. The paper focuses largely on innovation spending – a mea-sure of the inputs towards innovation – following the “absorptive ca-pacity” hypothesis in Cohen and Levinthal (1990) and Griffith et al.(2004) that spending is necessary to adopt external knowledge, sothat innovative inputs and outputs are positively linked. This con-trasts with the view in Spence (1984) that spillovers are costless, sothat greater spillovers may encourage R&D spending yet lower inno-vative output. However, we also ask whether there is a disproportion-ate sensitivity of industry growth to IPRs in R&D intensive industries,underlining the validity of the absorptive capacity hypothesis.

This paper uses data from Eurostat, which provides internationallycomparable industry data covering the universe of legal firms in 28 Euro-pean countries, including both manufacturing and non-manufacturingindustries.5 Country–industry data provide a natural environment inwhich to search for evidence of a link between IPRs and innovativeentry. Samaniego (2010) finds that country and industry dummies ac-count for almost half the variation in European rates of entry and exit —whereas time dummies account for about 1%. The use of European dataimplies that the countries considered do not significantly differ in theiraccess to natural or human resources, given low barriers to trade andimmigration.

3 Suppose that knowledge in each industry naturally spreads and can be built uponat a certain rate. The hypothesis is that IPRs limit this spread, which will make most dif-ference where that spread would have been large. In order to build on knowledge onemust possess it, and if one possesses it one could also use it for imitation, absent IPRs.

4 In theory the impact of IPR-induced appropriability is also ambiguous. We assumea positive impact of appropriability to be able to provide concrete interpretations of theresults. Assuming appropriability discourages innovation primarily in industries withlarge knowledge spillovers, some of the conclusions would be overturned, as discussedlater. However, the idea that appropriability discourages innovation contradicts otherevidence, to be supplied later.

5 Most studies of entry and exit focus on manufacturing; exceptions include Brandt(2004) and Samaniego (2010), who use earlier Eurostat entry and exit data but do notlook at R&D nor at IPR enforcement.

The main results are as follows. First, comparing across countries,enterprises in weak-IPR countries tend to disproportionately reportdifficulty raising funds, difficulty finding partners for innovation orthe dominance of an established incumbent as obstacles to innova-tion. This suggests that IPR enforcement not only encourages innova-tion, but that it shifts the balance towards entrepreneurs and awayfrom incumbents. Then, we find that effective IPR enforcement in-deed tends to encourage innovation spending in R&D intensive indus-tries. In addition, the same is true of both rates of entry and exit. Theresults are robust to conditioning on a variety of institutional factors,including other forms of property rights or contract enforcement,entry costs and financial development. As discussed below, IPR en-forcement tends to be measured using patent protection measures,and contribution of the paper is to use several institutional indicators,including several different indicators of IPRs.

The results speak in favor of models of economic growth whereknowledge spillovers across firms encourage innovation, and whereentry and exit are important for innovation. For example, inR&D-based models of growth that are close to growth accountingframeworks such as Romer (1990), Jones (1995) and Krusell (1998),growth is driven by knowledge spillovers across firms, but there areno industry dynamics to delimit the scope of spillovers. Our resultssuggest it is important to distinguish between the impact of spilloverson entrants and incumbents. In the basic creative destruction modelof Aghion and Howitt (1992), as well as more recent versions suchas Howitt (1999), knowledge spillovers increase the rate of innova-tion, and this favors entry (and leads to exit) because incumbentsface the obsolescence of their current IP. The key is that, in thesemodels, innovation is a substitute for prior expertise, so that “busi-ness stealing” is an important incentive for innovating that favorsentrants.

Several more recent papers extended such models to allow for in-cumbent innovation, as well as entry and exit.6 In the quality laddermodel of Klette and Kortum (2004) knowledge spillovers occur be-cause a successful innovator raises permanently the productivity ofthe next innovator, whoever it is, and as such spillovers affect en-trants and incumbents similarly. Peretto (1998) argues that the ten-dency should be towards incumbent-dominated R&D and, while ourfindings appear to contradict this conclusion, the model is useful forinterpreting those findings. Peretto (1998) assumes a weak-IPR envi-ronment where there is a tendency to develop large innovative incum-bents because size is a way to internalize knowledge spillovers whenappropriability is weak, and this is consistent with the finding that incountries with weak IPRs innovative entry appears suppressed, aswell as the surveys that report the presence of a dominant incumbentas an obstacle to innovation in such countries. The implication is thatin an environment with strong IPRs innovative entrepreneurshipshould be boosted, as found in this paper.

Acemoglu and Cao (2010) and Akcigit and Kerr (2010) also devel-op models in which both entrants and incumbents may innovate and,while they do not study the impact of IPRs in their models, the modelssuggest reasons why the entrant-bias of IPR-protected innovationcarries over into an environment with incumbent innovators. Inboth models, entrants are more likely to introduce innovations thatare fundamentally different from what is on the market, whereas in-cumbents are more likely to improve existing product lines — an ac-tivity that would depend more on in-house knowledge and (hence)less on the IPR regime.

The results stand in contrast to the view of Teece (1986) and Ganset al. (2002), whereby a strong IPR regime may discourage entry ininnovative sectors because it allows innovators to profit by sellingtheir idea to an incumbent who may have developed complementaryassets (e.g. distribution networks) rather than having to enter to

6 Thompson (2001) has entering and incumbent innovators, but abstracts from exitas it renders the model intractable.

52 R.M. Samaniego / International Journal of Industrial Organization 31 (2013) 50–63

benefit from the idea via entry and production.7 This could be becausecomplementary assets may not be critical in most industries, or are notdifficult for innovators to acquire themselves, but more likely it is sim-ply because weak IPRs discourage innovation – via entry and via themarket for ideas – rather than shifting it between non-market innova-tors and market incumbents.

The identification of knowledge spillovers at the industry level hastypically involved employing a measure of knowledge flows, such aspatent citations (e.g. see Jaffe et al., 2000) or survey responses (Cohenet al., 1987; Klevorick et al., 1995), while Bernstein andNadiri (1988) es-timate joint production functions for a small set of hi-tech industries,measuring knowledge using the depreciated stock of R&D spending.The strategy here is complementary and relies instead on the identifica-tion of knowledge spillovers through their interaction with intellectualproperty rights enforcement. The idea that IPRs are of different valueto different industries was explored by Mansfield (1986), again usingsurvey data, and the survey of Rockett (2011) finds mixed evidencethat IPRs affect R&D investments (for example, while Arora et al.(2003) find a positive relationship in certain industries, Qian (2007)finds only a weak relationship in one of those industries, pharmaceuti-cals). However, IPRs tend to be measured using patent protection mea-sures, and Levin et al. (1985)find thatmostfirms do not view the patentsystem as an important way of protecting intellectual property. Thispaper findsmore significant effects of IPR regimes on innovative and en-trepreneurial behavior using different measures — specifically, copy-right enforcement intensity and legal formalism (which is related todelays and inconsistency in legal dispute resolution, see Djankov et al.,2003), and a contribution of the paper is to underline the usefulness ofthese alternative measures. In particular, copyright enforcement isbroader than patent enforcement, and legal formalism could be impor-tant because IPR disputes when they occur tend to be costly8 and, ifthe outcome is uncertain (as in a high-formalism environment), theitself could become a means of “business stealing”.9 This underlinesthe importance for innovation not just of patent protection, but the pro-tection of IP more broadly, including for example design features ortrademarks.10

Evidence to distinguish among types of growth models typically re-lies on aggregate data — see Laincz and Peretto (2006), Ha and Howitt(2007) and Madsen (2008). Instead, this paper examines the links be-tween innovation spending, entry, exit and spillovers, which are thekey elements of the mechanisms underlying creative destructionmodels. As a result this paper underlines the empirical relevance of cre-ative destructionmodels not just for interpreting aggregate data but forunderstanding the industrial organization of innovation. As such, it sug-gests that the most useful creative destruction models are those whereknowledge spillovers benefit entrants and where new knowledge ismainly a substitute for knowledge rather than a complement.

More broadly, the paper bridges the literature on the determi-nants of R&D, the determinants of entry and exit, and institutions.The sense that there should be a link between technical change,entry and exit goes back at least as far as Schumpeter (1934), andGeroski (1989), Audretsch (1991) and others study the link empiri-cally, but none of these papers studies the role of institutions in theprocess of entry and exit, nor of innovative activity. Claessens and

7 Gans et al. (2002) support this with evidence, but our contrasting results could bedue to their focus on a few patent-intensive industries, where the existence of industrystandards, network effects and patent clusters may imply that innovators and incum-bents have incentives to collaborate. The data in the current paper cover almost the en-tire economies of the countries concerned.

8 See Rockett (2011).9 Indeed, Lerner (2005) conjectures that research behavior could respond to legal

formalism.10 A recent example of the importance of non-patented IP is the recent UK suitHC11C03050 through which Apple Inc. attempted to ban the Samsung Electronics(UK) Galaxy Tab 10, a competitor of its iPad: the Judge ruled that the Tab 10 did notinfringe Apple's registered design for the iPad because it was “not as cool.”

Laeven (2003) find that property rights (including intellectual prop-erty rights) enhance growth through improvements in resource allo-cation. The present paper finds evidence for one channel throughwhich this might occur: the replacement of incumbents by innovativeentrants. The idea that property rights encourage entrepreneurshipdates back at least to De Soto (1987), but the emphasis on intellectualproperty rights is novel in this context to the author's knowledge.

Section 2 presents the empirical strategy and introduces the dataused in the paper. Section 3 reports empirical results concerning in-stitutions and industry entry, exit and innovation spending. Section4 examines the robustness of the result to different statistical proce-dures and to the inclusion of other institutional variables, as well asexamining the impact of industry characteristics such as firm size ordifferent moments of the industry R&D distribution. Section 5 con-cludes with a discussion of potential future work.

2. Method and data

2.1. Empirical approach

Consider an environment with strong intellectual property rights(IPR) enforcement— call it country S. In this environment, innovatorshave a reasonable degree of control over what happens to theknowledge they create. In particular, outright imitation without alicensing agreement is likely to be forbidden. On the other hand, theuse of knowledge for inspiration – i.e. for creating further newknowledge – is unlikely to be limited to the extent that it does notdirectly infringe the intellectual property it is building upon. Thus,an innovator's competitors may have access to the innovator's knowl-edge (e.g. by reading the patent, examining the new product, etc.),but they cannot use it to produce (except through licensing), soobserved research activity is related to technologically determinedknowledge spillovers (spillovers due to institutional weakness arelimited). Define “desired” or “benchmark” R&D intensity as the extentto which firms in this environment find it profitable to conductresearch towards the generation of new products or processes.

Nowconsider an environmentwithweak IPR enforcement— countryW. In this environment, an innovator's competitors may be able to imi-tate a new product or process without penalty, or may create a closelyrelated innovation that would have been forbidden or require licensingin country S. Our hypothesis is that this ability to imitate that is providedby weak IPRs is correlated with the underlying magnitude of technolog-ical knowledge spillovers — the “ease”with which a competitor can ac-cess an innovator's knowledge in that industry. If these spillovers tendto encourage innovation, then we should see innovation spending dropin weak-IPR environments, especially in industries with high desiredR&D. On the other hand, the reverse should be true if these spilloverstend to discourage innovation.

Moreover, we replace innovation spending with entry rates as adependent variable, to assess the role of entrants vs. incumbents inthe process of innovation. For example, if entry is disproportionatelysuppressed by weak IPRs in research intensive industries, it suggeststhat entry plays a key role in research-driven innovation. Finally, wesee whether the same is true of rates of exit, which would support aframework where innovation often favors entrants and may displaceincumbents — as opposed to being neutral or favoring incumbents.

To test for these patterns, we adopt the differences-in-differencesapproach pioneered by Rajan and Zingales (1998). Let yi,c be the rateof innovation spending for industry i in country c. Let αc and δi denotecountry and industry indicator variables, respectively. RNDi measuresdesired R&D intensity in industry i, and Ic is a measure of institutionalstrength in a country c, such as intellectual property rights enforcement.We estimate the equation:

yi;c ¼ αc þ δi þ βRNDRNDi � Ic þ εi;c ð1Þ

12 Eurostat indicates the sampling method of earlier surveys may not be uniform


In specification (1), all country- and industry-specific factors af-fecting rates of innovation spending are removed by the fixed effectsαc and δi . Thus, any policies or regulations that affect innovation atthe country level are accounted for, as are all industry-specific factors.Instead, the existence of an impact of IPR enforcement on innovationin R&D intensive industries is identified by asking whether innova-tion yi,c is particularly susceptible to IPRs in industries with highRNDi — whether βRND>0. As in Rajan and Zingales (1998), to dealwith the common problem of heteroskedasticity in fixed effectpanels, we apply a White (1980) heteroskedasticity-consistent esti-mator, which allows the variance of the residual εi,c to vary by countryand by industry (as well as by RNDi× Ic).

We also askwhetherβRND>0when the dependent variable yi,c is therate of firm entry, and also the exit rate. For example, if βRND>0 whenthe dependent variable is innovation spending, we would concludethat technological spillovers are indeed largest in R&D-intensive indus-tries. Furthermore, if βRND>0 when the dependent variable is entry,then we would conclude that entrants play a particularly importantrole in spillover-induced innovation, whereas if βRNDb0 then wewould conclude that incumbents are the main beneficiaries of knowl-edge spillovers. Finally, if the sign of βRND varies depending onwhetherthe dependent variable is entry or exit, we would view exit as beingunimportant for the process of innovation.

We think of desired R&D spending as the optimal R&D spendinggiven the technological parameters of the industry in an environmentwith a strong IPR environment. The hypothesis of the paper is that ac-tual R&D spending is determined by both technological and institu-tional parameters. The maintained assumption behind (1) is thatdesired R&D intensity is an industry characteristic, the ranking ofwhich persists across countries. For example, if in the US firms inChemicals are more R&D intensive than firms in Textiles, our assump-tion is that the same would hold true in, say, Spain or Estonia, if theIPR environment were similar.

A potential concern is endogeneity of IPR enforcement: if there is alot of R&D-driven entry in the country, it may be that this encouragesgreater IPR enforcement through political economy channels. Wehandle this possibility in several ways. First, the fact that the depen-dent variable is defined at the level of the country–industry pair(whereas institutions are country variables) itself should reduce thepossibility of endogeneity. This is precisely the motivation for theRajan and Zingales (1998) differences-in-differences approach: allcountry-specific factors affecting innovation, entry, etc. are capturedby the country indicator αc, and identification depends only on indus-try differences in the dependent variable across countries. Second, weuse more than one measure of institutions, including both de jure andde facto measures. Third, for robustness, we estimate Eq. (1) using in-strumental variables. We use the standard set of instruments for in-stitutional variables, based on legal origin – English, French, Germanor Scandinavian – as well as an additional indicator variable forwhether the country in question is a post-socialist transition econo-my.11 We draw legal origin from the CIA World Factbook: see LaPorta et al. (1998) for more on the use of the legal origin instruments.The distribution of RNDi turns out to be quite skewed: as a result, wecorrect all standard errors for heteroskedasticity and check the ro-bustness of results by bootstrapping, among other methods describedlater.

Regarding the use of entry as a dependent variable, there is a ques-tion as to whether R&D intensity might be determined by rates ofentry (reverse causality). The empirical and theoretical literature sur-veyed in Geroski (1989), Klenow (1996), Cohen (2011) and Ngai and

11 Regressions of the IPR enforcement on the legal origin variables tend to yieldstrongly significant results. This suggests that these are reasonable instruments froma statistical perspective. In practice we regress interactions of the legal origin variableswith RNDi against interactions of the financial development variables with RNDi (seeWooldridge, 2002), which are even more strongly correlated.

Samaniego (2011) argues against this, with industry R&D rankingsbeing determined primarily by technological factors (opportunity andappropriability). The Rajan and Zingales (1998) methodology alsoavoids using actual innovation spending as an independent variable.

2.2. Data on industry–country pairs

2.2.1. Innovation expendituresInnovation expenditures are based on the European Community In-

novation Survey IV, 2002–2005, which was conducted by the EuropeanCommission and which is available through Eurostat.12 The survey re-ports innovation expenditure relative to net sales over the period,which has been used as a measure of innovation intensity since atleast Carlin and Mayer (2003). The survey defines an innovation as:

“a new or significantly improved product (good or service) intro-duced to the market or the introduction within an enterprise of anew or significantly improved process. Innovations are based onthe results of new technological developments, new combinationsof existing technology or the utilization of other knowledge ac-quired by the enterprise. Innovations may be developed by the in-novating enterprise or by another enterprise. However, purelyselling innovations wholly produced and developed by other en-terprises is not included as an innovation activity. Innovationsshould be new to the enterprise concerned. For product innova-tions they do not necessarily have to be new to the market andfor process innovations the enterprise does not necessarily haveto be the first one to have introduced the process.”

Note that, under this definition of innovation, a firm need not createa “new-to-the-world” innovation. This is consistent with the fact thatwe wish to allow firms other than the one that produced something“new-to-the-world” to innovate, through imitation or adoption.Cohen and Levinthal (1990) argue that a significant function of R&D isto implement innovations developed at other firms, and the definitionof “innovation” in the Eurostat data encompasses this function of R&D.This is also consistent with creative destruction models whereby inno-vation is the adoption of a technology thatwas revealed by amovementin the technology frontier due to spillovers from previous innovation.

The sampling population for the innovation data includes all en-terprises with 10 or more employees, as well as many smaller enter-prises. Responding firms comprised 45% of the universe of firms in thebusiness registries, totalling 181,838 firms. Eurostat reports industryinnovation expenditures across enterprises that reported some inno-vation, which is about 40% of responding firms, varying somewhatacross countries.

From these datawe construct twomeasures of R&D spending. One isthe ratio of innovation expenditures to sales reported in Eurostat for in-dustry i in country c, called Innovi,c

RAW. As mentioned, these “raw” dataonly cover innovating firms. The other measure is Innovi,cRAW multipliedby the share of innovators in each country, which we call Innovi,cADJ.

2.2.2. Entry and exitRates of entry and exit are drawn from the Eurostat Business De-

mography database,13 as are data on industry expenditures on innova-tion. The data cover 28 countries over the period 1997–2006. Eurostatreports data gathered by the national statistical agencies of themembercountries of the European Union concerning the universe of enterprises

across countries.13 The unit of observation is the “enterprise”, which is similar to the US Census Bu-reau definition of a “firm”, except that mergers and changes of legal status are distin-guished from “true” entries and exits. Included countries are those that reported toEurostat at the time of the study: participation in the data collection exercise wasnot mandatory so some countries report entry data but not innovation data and vice-versa.


in the business register, following a common methodology, so that thedata are comprehensive and internationally comparable.

Entering the business register is required to legally produce and sellgoods and services. If an enterprise ceases operations, by law itmust no-tify the business register within a matter of months. Mergers andchanges of legal form are not counted as entry, nor are temporaryshut-downs counted as exit. Thus, the data should adequately reflectentry and exit rates in the formal sector of each country. As well as cov-erage and comparability, an advantage of using European data is thatthe relatively skilled workforces of European economies, along withthe cross-bordermobility of labor and goods, imply that bottlenecks ex-perienced by would-be entrepreneurs are not likely to be driven by thelack of existence or availability of certain skills or resources, but ratherby the inability or unwillingness to acquire themdue to the institutionalenvironment. The survey data will support this presumption.

We study the same 41 industries as Samaniego (2010).14 This in-cludes 15 manufacturing industries and 26 non-manufacturingindustries.

For industry i in country c, the variable Entryi,c is the proportion ofenterprises active at a given date t that entered since date t-1, and thevariable Exiti,c is the number of enterprises that closed between t-1and t, divided by the number of enterprises active at date t. The vari-able Turnoveri,c is the sum of these two variables. All of these are av-erage rates over the sample period for each country–industry pair,to abstract from short term conditions. For much of the paper wewill focus on the variable Turnoveri,c, but also check that results arerobust to considering Entryi,c and Exiti,c separately.

2.3. Data on industry characteristics

2.3.1. Research intensityIn what follows, desired R&D intensity is viewed as an industry

characteristic. Cohen et al. (1987) find that industry dummies ac-count for over half of the variation in research intensity across firmsin their sample, and Ilyina and Samaniego (2011) find that the indus-try ranking by R&D intensity is stable across decades. We require anindicator of the “technological” or “desired” degree of research inten-sity in an industry, as defined earlier. In particular, the ideal indicatorshould not be contaminated by financing or other constraints. Wedraw on data on publicly traded US firms. The presumption is thatthese firms operate in highly liquid capital markets in an environ-ment with strong IPR enforcement, so any constraints on profitableinvestment projects should be minimal except perhaps in times ofcrisis — see Rajan and Zingales (1998) and Ilyina and Samaniego(2011). As a result, the R&D activity of a typical firm drawn fromthis environment should adequately reflect the technological tenden-cy of firms to perform R&D in that industry when the direct rewardsfrom innovation largely accrue to the innovator.15

R&D intensity at the firm level is defined as R&D expenditures di-vided by value added (DATA 46 divided by DATA 12 in Compustat).For each firm, we compute this value over the period 1997–2006.The industry measure of R&D intensity is the median firm value,which we call RNDi.16 See Appendix A, for the industry values.

14 Samaniego (2010) contains additional details regarding the construction of theEurostat entry and exit data, but uses an earlier edition of Eurostat with fewercountries.15 Ilyina and Samaniego (2011) find that R&D spending measures reported by theNSF correlate very highly with this measure of R&D intensity when the two are com-puted for a common industry grid.16 We do not use the CIS IV data to construct measures of “fundamental” industrytendency to perform R&D. The main reason is that (as discussed later) these numbersdo not represent a “clean”measure of the technological requirement for research, sincefinancing constraints in different countries may affect their innovation spending. In ad-dition, the innovation measures are not available for some service sector industries.Still, it is worth noting that RNDi and Innovi,c

RAW are positively correlated in 20 of the22 countries for which Innovi,c

RAW was available.

2.4. Descriptive statistics

Table 1 compares industry R&D intensity with entry, exit andturnover rates, averaged across countries.17 Entry, exit and turnoverare very highly correlated across industries, as known since Dunneet al. (1988) reported this finding for US Manufacturing. The correla-tion between RNDi and entry/exit is negligible. Comparing RNDi withindustry entry rates in each country, the correlation between RNDi

and entry varies from only 0.06 to −0.27. This implies that any inter-action between R&D and IPRs leading to differences in turnovershould not be due simply to the fact that R&D is itself a determinantof turnover, but rather due to the impact of institutional factors onthe ability of firms to conduct business or to pursue R&D.

The simplest interpretation of the lack of correlation between R&Dand entry is that this contradicts models (such as the creative destruc-tion framework), where innovations are introduced through R&D-driven entry. However, this is not necessarily so. In those modelsthere are several parameters that affect R&D and innovation. For exam-ple, in a quality-ladder model, it could be that R&D intensive industriesare so because the steps in the quality ladder are large, but the probabil-ity of success is small, so that entry rates themselves are not significant-ly different from other industries. In this case, if IPRs encourageinnovation mainly in high-spillover industries, we would expect tosee more entry in those industries too.18 A contrasting hypothesis isthat, as in Teece (1986), in a high-IPR environment innovation andentry are de-linked because innovators can sell their innovations to in-cumbentswithout going through the trouble of entering and producing,whereas in a weak-IPR environment themarket for technologieswouldbe impaired. In this case, wewould expect disproportionate amounts ofentry into the industries with the greatest spillovers when IPRs areweak.

2.5. Data on countries

2.5.1. InstitutionsThe strategy for identifying spillovers involves the institutions

that avoid costly intellectual property rights disputes. However, theliterature distinguishes between two broad classes of institutions:contracting institutions and property rights institutions. See North(1984) and Acemoglu and Johnson (2005). In principle, both ofthese could matter for effective IPR enforcement. There may be insti-tutions that clearly allocate the ownership of intellectual propertyand such institutions lower the likelihood of IPR disputes arising inthe first place, ensuring that the profits from an innovation accrueto the innovator. In addition, there may also be institutions that affecthow costly any given dispute will be once it arises. If IPR disputes arecostly then, regardless of the likelihood, this should also discourageinnovation.

Since it is not clear what specific institutional measures may be ofimportance, we survey a variety of measures of contract enforcementand property rights to check that it is specifically the institutions re-lated to intellectual property disputes that interact with knowledgespillovers, and not just general institutional quality.

17 To be precise, the industry index of entry, exit or turnover is based on the industryfixed effect in a regression of country and industry dummy variables. If yi,c is entry, exitor innovation spending in industry i in country c; we estimate:

yi;c ¼ αc þ δi þ εi;c ð2Þ

where αc and δi are country and industry dummy variables. The index of entry for in-dustry i is then the coefficient δi, added to the coefficient αc for the median country.18 A calibrated version of the basic creative destruction model has this feature and isavailable upon request.

Table 2

Table 1Cross-industry correlations between turnover measures and industry variables. Ratesof turnover, entry and exit are based on industry fixed effects δj in Eq. (2). Standard er-rors are in parentheses. In all tables, one, two and three asterisks represent significanceat the 10%, 5% and 1% levels respectively.

Industry indicator

Entry Exit RNDi

Turnover 0.97⁎⁎⁎ 0.90⁎⁎⁎ –0.15(0.042) (0.071) (0.158)

Entry – 0.75⁎⁎⁎ –0.11(0.106) (0.159)

Exit – – –0.18(0.158)


Property rights institutions determine the extent to which entrepre-neurs can control the use and transfer of the firm's physical or intangi-ble assets. If agents cannot credibly transfer physical or intangible assetsin the event of default then they cannot use them as collateral (seeClaessens and Laeven, 2003) nor will they be able to fight off IPR dis-putes with incumbents nor credibly engage in partnerships with otherfirms (whomight steal their IP). In addition, even though certain intan-gible assetsmay be inherently difficult to collateralize, intellectual prop-erty rightsmay ensure that the revenues from an intangible investmentaccrue ultimately to the firm, as opposed to a competitor. An invest-ment project is more valuable, and more likely to survive, when the in-tangibles that delimit the project (the business plan, the product, etc.)are less likely to be stolen or copied by competitors. 19

We use the following property rights indicators:

1. PTNTc: patent enforcement. This applies to patentable (e.g. scientif-ic) knowledge, such as new products or processes. It is measuredas in Ginarte and Park (1997), a de jure measure of IPR intensity,updated in 2000. See Property Rights Alliance (2007).

2. COPYc: copyright enforcement. It is based on the Office of the Unit-ed States Trade Representative “Special 301” watch list, reflectingpiracy rates in the business software, entertainment software, mo-tion picture, record and music industries. We take it to indicate theprotection of new products and processes where some aspect ofthem might require copyright, as well as proxying more generallyfor the extent to which violations of any kind intellectual propertyis protected.20 See Property Rights Alliance (2007).

3. RULEc: Acemoglu and Johnson (2005) interpret property rightsprotection as protection from expropriation by the government.We capture this aspect of property rights with the Rule of Lawindicator developed by Kaufman et al. (2007) — an overall indica-tor of the confidence of the populace in governmental institu-tions, perception of freedom from corruption, etc., and isconstructed from a variety of sources.21 Conditioning on RULEc

19 Acemoglu and Johnson (2005) define property rights institutions as those that spe-cifically guard against expropriation by the executive. The definition here is slightlydifferent, but still depends on whether property rights are credibly assigned andenforced, independently from the identity of a potential expropriator.20 IP can include patents, copyright, trademarks and registered designs. The pre-sumption is that copyright infringement rates are correlated with infringement ratesof the other types of IP.21 Acemoglu and Johnson (2005) use the de jure Polity IV indicator of constraints onthe executive to measure protection from expropriation by the government. There isinsufficient variation in this indicator across the countries in Eurostat to be able touse it in the present study. Hence, the results of the present paper may be interpretedin terms of the impact of finance on the process of creative destruction, conditional ona certain level of protection from expropriation by the government. However, it isworth noting that the values of RULE in the countries studied in this paper cover thetop half of values among countries in the Kaufmann et al. (2007) dataset (as well asmatching the range covered by the top 50% of countries), indicating a wide variety ofinstitutional outcomes.

ensures that our results are not simply due to overall institutionalquality.

4. PHYSc: a measure of physical property rights protection. It is drawnfrom the World Economic Forum Global Competitiveness report,and is based on the response to a survey question on the clarityof definition of property rights, including over financial assets.

Contracting institutions support the effectiveness of private agree-ments. Countries may differ in terms of the likelihood of contract dis-putes, differences in efficiency, and the weight they give the principal(e.g. the borrower, the entrepreneur) as opposed to the agent (thelender, or financier) in the event of disputes or default. They may af-fect IPR enforcement because disputes regarding intellectual propertymay be more or less costly to resolve (or avoid in the first place)depending on contracting institutions. Contracts are less likely to besigned if principal-agent problems are severe, transparency is weakor if enforcement is lax or costly. We use the following measures:

1. INVPc: investor protection intensity, from Djankov et al. (2008). Itaggregates together three measures: the extent of disclosure, theextent of director liability and the ease of shareholder suits. Thedata come from a survey of corporate lawyers and are based on se-curities regulations, company laws and court rules of evidence.

2. DISCc: disclosure requirements, from La Porta et al. (2006). It re-flects the transparency of contracting relations.

3. ENFORc: contract enforcement costs, also from Djankov et al.(2003). It is the proportion of a claim, including attorney fees,that must be paid to successfully sue a buyer to pay for deliveredgoods worth 200% of GDP when the goods are alleged not to beof adequate quality.

4. INFORc: Djankov et al. (2003) find that the degree of formalism inthe legal system is a strong predictor of delays in dispute resolu-tion, as well as lower consistency/fairness in outcomes. FollowingAcemoglu and Johnson (2005), we use the negative of theDjankov et al. (2003) formalism index as a measure of efficiencyin dispute resolution.

Correlations among different forms of contract enforcement insti-tutions are all positive but mostly low, suggesting that they capturedifferent provisions. Only the investor protection and disclosurevariables are strongly related to each other. By contrast, propertyrights measures are mostly very closely related to each other. SeeTable 2.

Cross-country correlations between institutional measures. P-values are in parenthe-ses. In all tables, one, two and three asterisks represent significance at the 10%, 5%and 1% levels respectively.

Contract enforcement

INVP DISC INFOR

ENFOR 0.12 0.08 0.04(0.560) (0.691) (0.870)

INVP – 0.82⁎⁎⁎ 0.34(0.000) (0.104)

DISC – – 0.15 (0.477)

Property rights

PTNT COPY PHYS

RULE 0.29 0.65 0.80⁎⁎⁎

(0.162) (0.000) (0.000)PTNT – 0.41⁎⁎ 0.24

(0.049) (0.241)COPY – – 0.77⁎⁎⁎

(0.000)


2.6. Preliminary evidence: survey results

As well as reporting innovation spending data, the Community In-novation Survey IV (CIS), asks firms what kind of factors severelyhamper innovation. We begin by asking whether the answers provid-ed to these questions are in any way correlated with cross-countrydifferences in institutions, providing suggestive evidence as to theimpact of institutions on innovation. Again, we look at a number ofdifferent measures of contract enforcement and property rights tomake sure it is not broad contract enforcement or property rightsthat matter, but rather specific aspects thereof.

Only four institutional variables appear significantly related to anydifficulties in innovation: copyright enforcement (a broad notion ofintellectual property rights), informality of the legal system (viewedas an indicator of relatively costless dispute resolution mechanisms),physical property rights and rule of law. The weakest of the four arerule of law, which is a broad indicator of institutional quality, andphysical property rights. This suggests that specific institutions (IPRenforcement and/or low costs of dispute resolution), rather thanoverall institutional quality, have an impact on the ability or willing-ness of firms to innovate.

Firms in economies with high levels of COPY or INFOR are less like-ly to report lack of external funds or the costs of innovation as being aproblem — see Table 3. This suggests that lenders are unwilling to fi-nance projects where the firm may have weak control over the re-quired intellectual property. Tellingly, they also appear less likely toreport dominance by “established enterprises” as a factor discourag-ing innovation, consistent with the idea that IPR enforcement stimu-lates competition by facilitating innovation by entrepreneurs. Thiswould explain the inability of innovators to raise funds for innova-tions (since incumbents are more able to quash them). They also re-port difficulty finding partners for innovation, suggesting theunwillingness of established enterprises to cooperate with potential

Table 3Correlations between reported significant difficulties in financing innovation amongfirms and financial development measures. Answers include (1) lack of own funds(2) difficulty of raising external funds (3) high costs of innovation (4) difficulty of find-ing qualified personnel (5) difficulty of adopting information technology (6) lack of in-formation about market conditions (7) uncertainty about the demand for innovation(8) difficulty of finding partners for innovation (9) no need to innovate due to the pres-ence of a recent innovation and (10) presence of a dominant incumbent. Source —

Author's calculations and the Community Innovation Survey IV.

Limitations on innovation Property rights institutions

PHY S PTNT COPY RULE

Lack of own funds −0.21 0.03 −0.20 −0.18Lack of external funds −0.32 −0.27 −0.71*** −0.33Innovation costs too high −0.24 −0.23 −0.47** −0.14Lack of qualif. personnel −0.35* −0.13 −0.39* −0.35*IT adoption costs high −0.37* −0.04 −0.28 −0.33Lack of info. on market −0.38* −0.03 −0.29 −0.32Uncertain demand for innov −0.37* −0.14 −0.41* −0.22Lack of innovation partners −0.49** −0.14 −0.48** −0.42**No need to innovate −0.34 −0.05 −0.19 −0.31Dominance by incumbents −0.53*** −0.23 −0.56*** −0.41**

Limitations on innovation Contracting Institutions

INV P DISC ENFOR INFOR

Lack of own funds −0.15 0.11 0.15 −0.30Lack of external funds −0.39* −0.24 −0.15 −0.61***Innovation costs too high −0.33 −0.06 −0.00 −0.47**Lack of qualif. personnel 0.14 0.22 −0.25 −0.35*IT adoption costs high −0.05 −0.06 −0.34 −0.28Lack of info. on market 0.06 0.00 −0.31 −0.17Uncertain demand for innov −0.25 −0.15 −0.30 −0.24Lack of innovation partners −0.20 −0.07 −0.34 −0.40*No need to innovate 0.04 −0.05 −0.35 −0.22Dominance by incumbents 0.01 0.09 −0.26 −0.52**

competitors, or the fact that a weak IPR system makes it hard totrust potential partners. All of this is consistent with weak IPRs de-creasing the ability of innovators to control the use of their knowl-edge by others and thus suppressing innovation, as hypothesized inthis paper.

3. Country–industry regressions

We estimate Eq. (1) for each institutional measure above. Also, tosee which are the “dominant” institutional measures, we include sev-eral interaction terms in the same regression. Suppose that we are in-terested in comparing the interactions of a set K of institutionalmeasures. Let Ick equal the value of institutional measure k∈K. We es-timate

yi;c ¼ αc þ δi þ ∑k∈K

βRNDRNDi � Ikc þ εi;c: ð3Þ

In this case, a potential concern is that the institutions are mea-sured with different amounts of error, so the winner of a “horserace” among institutions might simply be the variable with the leasterror. Thus, we report results instrumenting the interaction termsusing Legal Origin indicators, following Acemoglu and Johnson(2005). Results are similar without instrumenting, however.

Results are reported in Tables 4–6. We find that the “dominant”contract enforcement measure is INFORc, the “informalism” of thelegal system: none of the other measures display a strong interactionwith RNDi. By contrast, all the property rights measures display a sig-nificant interaction with RNDi — except for the de jure measure ofpatent enforcement PTNTc. Although PTNTc is a commonly used mea-sure of IPR enforcement, the tendency to patent seems to be concen-trated in a few industries (see Cohen, 2011; Rockett, 2011 forexample), so it is likely too narrow a notion of IPRs for most indus-tries. Levin et al. (1985) report that most firms do not view the patentsystem as an important way of protecting intellectual property.

In terms of statistical significance, however, COPYc and RULEc standout. If we estimate Eq. (3) including all the property rights interac-tions simultaneously, COPYc is the only one with a significant coeffi-cient, when we examine the role of entry and exit. When thedependent variable is innovation spending the same is true exceptthat the interaction of PTNTc takes on a significant coefficient of thewrong sign that is due to collinearity. If we include only the interac-tions of COPYc and RULEc in that regression, then neither is significantbut the coefficient on COPYc is close to statistical significance at the10% level whereas that on RULEc is around 85%, and in the case ofInnovRAW it is negative. Thus, the institutions that interact most ro-bustly with the technological determinants of R&D intensity appearto be the definition and protection of intangible assets (COPYc), andthe degree of formalism of court proceedings (INFORc). When weput interactions of R&D intensity with COPYc and with INFORc in thesame equation (Table 7), INFORc dominates when we look at innova-tion spending as a dependent variable, whereas COPYc dominateswhen we focus on enterprise turnover.

These results as indicate that, as suggested by the survey results,the most important factors leading institutions to affect entrepre-neurship and innovation – the creation of new firms and of new prod-ucts or methods – is the protection of intellectual property broadlydefined against disputes, and the efficient resolution of disputesabout the ownership of these assets when they do occur. Further-more, the impact of these institutions is disproportionately positiveon innovation spending, entry and exit in R&D intensive industries,suggesting that these are industries with large technological knowl-edge spillovers and where innovation by entrepreneurs and the re-moval of incumbents play an important role in the process ofinnovation. This suggests that, assuming that IPRs encourage innova-tion in industries where knowledge spillovers are large, industries

22 They also find evidence that this is the case, in that industries with higher TFPgrowth in the US tend to display more rapid value-added growth.

Table 4Effect on turnover, entry and exit of the interaction between R&D intensity and contract enforcement, based on estimating Eq. (1). Country and industry fixed effects are omitted forbrevity. Heteroskedasticity-corrected standard errors are reported in brackets. R&D intensity is measured as the ratio of research spending to net sales at the median firm inCompustat (RND). Observations for turnover, entry and exit are 869, 916 and 875.

Ic Dependent variable yj,c

Turnover Entry Exit InnovADJ InnovRAW

βRND R2 βRND R2 βRND R2 βRND R2 βRND R2

ENFOR 0.015 0.65 0.034 0.63 −0.020 0.54 −0.105 0.62 −0.365* 0.62(0.113) (0.077) (0.046) (0.086) (0.206)

INVP 0.113 0.65 0.049 0.64 0.078 0.53 0.241_ 0.62 0.385 0.68(0.147) (0.100) (0.062) (0.131) (0.274)

DISC 0.072 0.65 0.063 0.64 0.043 0.53 0.113 0.62 0.274 0.68(0.119) (0.089) (0.049) (0.091) (0.223)

INFOR 0.351⁎⁎⁎ 0.66 0.261*** 0.65 0.091* 0.54 0.352⁎⁎⁎ 0.63 0.888** 0.68(0.107) (0.076) (0.0483) (0.132) (0.431)

Table 5Effect on turnover, entry and exit of the interaction between R&D intensity and property rights, based on estimating Eq. (1). Country and industry fixed effects are omitted for brev-ity. Heteroskedasticity-corrected standard errors are reported in brackets. R&D intensity is measured as the ratio of research spending to net sales at the median firm in Compustat(RND). Observations for turnover, entry and exit are 869, 916 and 875.


Turnover Entry Exit InnovADJ InnovRAW

βRND R2 βRND R2 βRND R2 βRND R2 βRND R2

RULE 0.417⁎⁎⁎ 0.65 0.284 0.64 0.126*** 0.54 0.442⁎⁎⁎ 0.61 0.764*** 0.65(0.112) (0.082) (0.052) (0.118) (0.292)

PTNT 0.308⁎ 0.65 0.180 0.66 0.116* 0.55 0.086⁎ 0.60 − .160 0.65(0.160) (0.100) (0.070) (0.144) (0.551)

COPY 0.462⁎⁎⁎ 0.65 0.305*** 0.64 0.152*** 0.54 0.521⁎⁎⁎ 0.62 1.04*** 0.66(0.105) (0.077) (0.045) (0.111) (.223)

PHYS 0.486⁎⁎⁎ 0.66 0.333*** 0.64 0.134** 0.54 0.291⁎⁎⁎ 0.60 0.414* 0.65(0.107) (0.076) (0.054) (0.097) (0.233)


with high R&D spending are those where knowledge spilloversare most prevalent, and that entrepreneurship along with the re-placement of incumbents form an important part of the process ofinnovation.

How should one interpret the economic significance of the coeffi-cients on the interaction terms? Since fixed effects account for muchof the variation in the data, the R2 values alone do not give much in-dication of the economic significance of the interaction variables. Thefollowing example, however, might give a sense of the economic sig-nificance of the interaction term. Consider Table 5. The country withthe lowest copyright enforcement index is Bulgaria, and the highestis the UK (jointly with the Germany, Sweden and the Netherlands).The coefficients imply that the difference in innovation spendingrates between the industries with the highest and lowest rates ofR&D intensity in Bulgaria is about 8.5% smaller than in the UK. Consid-ering that RNDi ranges from zero to around 30%, (Chemicals), theseare large values. The coefficients also imply that the difference inentry rates between the industries with the highest and lowestrates of R&D intensity in Bulgaria is about 5.5% smaller than in theUK. Since rates of entry vary across industries from about 3% toabout 16%, this represents a substantial difference. Thus, the interac-tion coefficients are such that there is significant compression of in-novation spending and entry rates in countries with weak IPRs,compared to countries with strong IPRs.

4. Robustness

So far we have focused on the impact of spillovers on innovationspending and on entry/exit. As discussed in the Introduction, this em-bodies the “absorptive capacity” hypothesis of Cohen and Levinthal(1990) but, in principle the amount of innovation spending and the

amount of successful innovations could be negatively related ifknowledge spillovers are free (Spence, 1984). This suggests exploringwhether the IPR-induced increase in entry and innovation spendingin R&D intensive industries is related to innovation outcomes. Weask specifically whether it is related to industry value-added growthrates. If goods are substitutes, Ilyina and Samaniego (2012) showthat industries with more rapid value added growth experiencemore rapid productivity growth,22 and that disproportionately rapidvalue added growth in a differences-in-differences framework canbe interpreted in terms of disproportionately rapid productivitygrowth. It turns out that if we estimate (1) with value added growthin industry i in country c as a dependent variable, the coefficient βRND

is indeed positive for both INFORc and COPYc, although it is only statis-tically significant in the case of the latter. Thus, IPRs disproportionate-ly enhance entry, exit, innovation spending, value added growth andproductivity growth in R&D-intensive industries. See Table 8.

As discussed in the Introduction, our interpretation of the resultshinges also on an assumption that low appropriability in the form ofweak IPRs reduces innovation where knowledge spillovers are strong.There are models where the opposite is the case (appropriability dis-courages innovation, e.g. in Spence (1984) too much appropriabilityreduces innovation because of duplication). This would lead to theopposite conclusions regarding the link between benchmark R&Dand spillovers as well as entrepreneurship. For example, suppose in-deed that low appropriability through weak IPRs boosts innovationwhere spillovers are large. In this case, the results imply that it is

Table 6Effect on turnover and innovation of the interaction between R&D intensity and institu-tional indicators, based on estimating Eq. (3). Interactions of R&D intensity with all fourmeasures of contract enforcement (or all four measures of property rights) are includedtogether in the corresponding regression. Interaction variables are instrumented usinglegal origin dummies.

Ic Contract enforcement

Dependent variable yj,c

Turnover InnovADJ

βRND βRND

ENFOR 0.295 0.723(0.929) (0.643)

INVP −0.893 −0.477(0.598) (1.35)

DISC 0.722 −0.707(0.594) (1.91)

INFOR 0.817⁎⁎⁎ 1.03⁎⁎

(0.214) (0.462)Obs 808 470R2 0.660 0.443

Ic Property rights

Dependent variable yj,c

Turnover InnovADJ

βRND βRND

RULE −0.040 0.389(1.69) (0.632)

PTNT 0.988 −0.511***(0.272) (0.178)

COPY −0.028 0.431***(1.90) (0.128)

PHYS 0.334 −0.064(1.094) (0.737)

Obs 763 517R2 0.676 0.611

Table 8Effect on value added growth of the interaction between R&D intensity and the domi-nant institutional variables from each category. Results are based on estimating (3), in-cluding interactions of INFOR and COPY with R&D intensity. Interaction variables areinstrumented using legal origin.

Ic Specification

INFORc only COPYc only

INFORc 0.326 –

(0.429)COPYc – 1.05**

(0.527)Obs 820 851R2 0.138 0.142


the low-RNDi industries where spillovers are large, and theirinnovative spending is disproportionately larger in low-IPR countries.This would result in values of the interaction coefficient βRND of thesame sign, since it would result in a divergence of observed rates ofinnovation spending in strong-IPR countries. This would also implythat low-RNDi industries experience disproportionately high rates ofentry and exit in low-IPR environments. Thus, the conclusion thatentry and exit is important for innovation does not change, only thequestion of whether industries with high benchmark R&D spendingare those where spillovers are “naturally” large or small.

However, several pieces of evidence suggest that our findings areconsistent with IPRs having a salutary rather than detrimental effecton innovation overall. First, recall that in the survey results innovators

Table 7Effect on turnover and innovation of the interaction between R&D intensity and thedominant institutional variables from each category. Results are based on estimating(3), including interactions of INFOR and COPY with R&D intensity. Interaction variablesare instrumented using legal origin.


Turnover Entry Exit InnovRAW InnovADJ

INFOR 0.290 0.190 0.108 2.02** 0.486***(0.194) (0.137) (0.088) (0.931) (0.180)

COPY 0.362*** 0.222** 0.126** −0.152 0.258(0.131) (0.092) (0.052) (0.684) (0.172)

Obs 828 875 834 516 516R2 0.662 0.648 0.543 0.656 0.629

in countries with weak IPRs report a variety of obstacles to success-ful innovation. Second, consider a regression of the innovation indi-cator Innovi,c

RAW on country and industry dummies. The correlationbetween COPYc and the country dummies is 0.44**, and that betweenINFORc and the dummies it is 0.39*. Third, the correlation betweenCOPYc and real GDP per capita in 2007 as reported by the WorldBANK is 0.81*** (or 0.65*** for INFORc). In addition, Branstetter etal. (2006) find that multinationals increase R&D and experiencegreater royalty income when the countries where they operatestrengthen IPR enforcement.

As mentioned earlier, it could be that countries which for some ex-ogenous reason have high rates of entrepreneurship or innovationmight have institutions that better protect intellectual propertyrights, i.e. institutions might be endogenous. Since the dependentvariable is at the country–industry level, not the country level, it isnot clear how this would bias results, which is precisely the rationalein Rajan and Zingales (1998) behind specifications such as (1). None-theless, we performed the estimation of the single-variable regres-sions using Legal Origin variables to instrument for COPY and INFOR,as well as without. Results are generally similar, see Table 9.

One feature of the data is that research activity is not smoothly dis-tributed across industries. For example, the most research-intensive in-dustry (Chemicals) has a ratio of R&D spending to net sales of 32.2%,and the next highest (Computers and Electronic Products) is 13.7%.Also, several industries have zero R&D intensity. To ensure that the re-sults are not driven solely by outliers and that the standard errors are ro-bust to skewness, we estimate several variations of the originalspecification. First, we eliminate Chemicals from the list of industries.Second, we check whether the results hold only for manufacturing, asmany of the industries with zero R&D intensity are service sector indus-tries. Third, we estimate the original specification, with bootstrappedstandard errors. Fourth, we estimate a “median regression,” where

Table 9Effect of the interaction between institutions and R&D intensity based on estimatingEq. (1). Robustness exercises.

Ic measure COPYc INFORc

Dependent variable yj,c Turnover InnovADJ Turnover InnovADJ

Instrumental variables 0.500*** 1.23*** 0.622*** 1.90***(.115) (0.296) (0.157) (0.551)

Without chemicals 0.793*** 1.75*** 0.670*** 0.408(0.208) (0.458) (0.200) (0.840)

Manufacturing only 0.290*** 0.768*** 0.189** 1.00***(0.091) (0.170) (0.096) (0.370)

Bootstrapped standard errors 0.462*** 1.05*** 0.351*** 0.888**(0.104) (0.225) (0.108) (0.417)

Median regression (bootstrapped) 0.435*** 0.575*** 0.263*** 0.730***(0.121) (0.196) (0.092) (0.222)


absolute deviations (rather than squared deviations) are minimized bythe estimation procedure, again with bootstrapped standard errors.This approach weights outliers less than more common “least squares”methods. The results are robust to all of these variations of the originalspecification.

Ilyina and Samaniego (2011) show that R&D intensity is positivelycorrelated with external finance dependence. We ask whether the be-havior of RNDi is due to bias stemming from the omission of externalfinance dependence or EFDi. It is worth noting that Rajan and Zingales(1998) developed the EFDi measure assuming that it was a technolog-ically determined characteristic, but without investigating what theunderlying determinants might be. Using the production function asa concept of technology, Ilyina and Samaniego (2011) explore a num-ber of measures of input intensity and find that RNDi is both highlycorrelated with EFDi and outperforms EFDi in industry growth regres-sions, indicating that R&D intensity is a likely technological character-istic that underpins EFDi. Thus, we would expect EFDi to behavequalitatively the same as RNDi, albeit more weakly. Indeed, we findthat replacing RNDi with EFDi in (1) yields statistically significant co-efficients, but only rarely. The main exceptions are the regressionswhere the institutional variable is INFORc and the dependent variableis InnovADJ or InnovRAW: in this case there were positive coefficientsthat were less statistically and economically significant than whenthe industry characteristic is RNDi. 23

Another possibility is that IPRs proxy for an unrelated (but correlat-ed) policy. In particular, Klapper et al. (2006) find that the regulation ofentry hampers firm creation. For the case of investment-specific techni-cal change, Samaniego (2010) suggests that policies that make entrycostly may lead innovations to be introduced by incumbents insteadof entrants. If so, an interaction of RNDj with entry costs might carry anegative coefficient and, if entry costs are negatively correlated withIPRs, the significance of βRND may be due to omitted variable bias. In-deed, startup costs as measured by Djankov et al. (2002) and updatedin World Bank (2006) are negatively related to financial development,however the correlations are weak (−0.31 with COPY and −0.17with INFOR). We did repeat the estimation including an interaction ofentry costs with RND as well as entry costs with COPY or INFOR, findingthat the results are robust and that entry costs do interact with RND butonly when the dependent variable is innovation rather than entry orexit.

Some authors argue that an important channel of knowledge dif-fusion is international trade — e.g. see Coe and Helpman (1995)-and that the ability to absorb knowledge spillovers can depend ontrade openness (Keller, 1998). It could be that IPRs are correlatedwith trade openness, so that the results reflect an increased abilityof industries to absorb knowledge from abroad, and R&D-intensiveindustries are the primary beneficiaries, which stimulates innovativeactivity (particularly by entrepreneurs). This interpretation of the re-sults would still underline the conclusion that R&D intensive indus-tries are those where, in the absence of limitations, knowledgespillovers are large and spur innovation, and that entrepreneursplay an important part in this process. At the same time, the correla-tion between country trade openness developed in Nicoletti et al.(2000) policy are not statistically significantly correlated with COPYcnor with INFORc. An alternative possibility is that industries that aremore R&D intensive tend to be more open to international trade,and that the observed effects are due to IPRs allowing firms more

23 There is a presumption that human capital is required for R&D, and one mightwonder if results reflect the impact of low human capital in financially underdevelopedeconomies. Ilyina and Samaniego (2011) find that industry human capital intensity isnot statistically significantly related to R&D intensity and, in fact, the correlation be-tween COPY and average years of schooling (as reported in Botero et al., 2004) is only0.16 in our sample.

reliable access to licensed or other technologies from abroad. In gen-eral our paper is agnostic as to the source of knowledge spillovers thatIPRs allow firms to access, but it would be interesting to assesswhether the trade channel is important. Using the trade data fromFeenstra et al. (2005) we do not find a significant correlation betweenthe volume of exports or imports in the countries in our data and in-dustry R&D intensity. However, it is important to remember that ourdata are highly aggregated from the perspective of empirical work ininternational trade (and that most of our industries are service indus-tries, for which trade data are rarely available), so a study devoted tothis specific question using much more disaggregated data would bedesirable.

Aghion et al. (2007) find that entry is disproportionately sensitiveto financial development in manufacturing industries that are moredependent on external finance, using a different data set. We askwhether perhaps the results are due to collinearity between financialdevelopment and our institutional measures.

We consider the following measures of financial development.

1. CREc: Our benchmarkmeasure uses the domestic private credit-to-GDPratio. Domestic credit data come from the IMF International FinancialStatistics (IFS); domestic credit allocated to the private sector is IFSline 32d. It is measured at the beginning of the period for which wehave industry data (1997) or else the earliest year in the period forwhich it is available.

2. CAPc: For robustness we also use the domestic capitalization-to-GDPratio, the sum of domestic market capitalization and private credit.Market capitalization is reported by Eurostat. It is measured at thebeginning of the period for which we have industry data (1997) orelse the earliest year in the period for which it is available. AlthoughCAPc is broader than CREc, it may not always accurately reflect theamount of funds raised in domestic financial markets for productiveactivities — due to tax incentives to list on stock exchanges, stockmarket dynamics being driven by factors other than fundamentals,etc. Such distortions likely to be particularly severe for the case oftransition economies. Hence, in what follows we use CREc as ourbenchmark.

3. BANKc: We also use a measure of bank overhead as a share of assetsin 1997. This is an inverse indicator of financial development (seeBeck et al., 2000), as high overhead represents inefficiency in thefinancial sector. Hence, we multiply it by minus one. It is drawnfrom the 2006 update of the Beck et al. (2000) Database on Finan-cial Development and Structure.

4. MARGc: The interest rate margin between borrowing and lending isalso an inverse indicator of financial development. The presump-tion is that high margins reflect high costs of operation, or an un-competitive financial sector. We draw it from the same source asBANKc , and also multiply it by minus one.

5. ACCSc: The World Economic Forum Global Competitiveness Report(GCR) contains a measure of “loan access”. It is based on the surveyquestion “how easy is it to obtain a bank loan with a good businessplan and no collateral?” measured on a scale of 1–7. The questionwas included in the Executive Opinion Survey, which covers over12,000 executives in 134 countries. See Browne et al. (2008) formore details.

6. SOPHc: The GCR also contains a measure of financial market sophis-tication. It grades responses to the question “the level of sophistica-tion of financial markets in your country is (1 = lower thaninternational norms, 7 = higher than international norms).”

Table 10 shows the correlations between the institutional measuresand financial development. Clearly both COPY and INFOR are stronglycorrelated with the financial development measures. Table 11 alsoshows that financial development interacts with R&D intensity both interms of entry/exit behavior and innovation spending, much likeCOPY and INFOR. At the same time, the coefficients are generallylower. If we run a horse-race among the financial development

Table 12Effect on turnover and innovation of the interaction between R&D intensity and finan-cial development, based on estimating Eq. (1). Interaction variables are instrumentedusing legal origin.


Turnover InnovADJ

βRND βRND

CRE 0.065 0.885(0.141) (0.642)

COPY 0.367*** −0.595(0.127) (0.745)

INFOR 0.253 0.181**(0.207) (0.741)

25 We measure Firmsizei,c using average employment at the establishment level in1997, drawn from Eurostat. We use average employment, rather than sales, becausenot all countries in the sample use the Euro so there would be issues of currency con-version, varying inflation rates and possibly simply mis measurement of sales. Al-though firm size may potentially feed back to observed IPR enforcement, the fact

Table 11Effect on turnover, entry and exit of the interaction between R&D intensity and finan-cial development, based on estimating Eq. (1). Country and industry fixed effects areomitted for brevity. Heteroskedasticity-corrected standard errors are reported inbrackets. R&D intensity is measured as the ratio of research spending to net sales atthe median firm in Compustat (RND).

FDc Dependent variable yj,c

Turnover InnovADJ

βRND R2 βRND R2

CRE 0.33*** 0.652 0.56⁎⁎⁎ 0.602(0.115) (0.150)

CAP 0.29⁎⁎ 0.652 0.83⁎⁎⁎ 0.608(0.125) (0.219)

BANK 0.51⁎⁎ 0.649 0.52⁎⁎⁎ 0.586(0.232) (0.179)

MARG 0.45⁎⁎⁎ 0.651 0.57⁎⁎ 0.580(0.180) (0.248)

ACCS 0.46⁎⁎⁎ 0.654 0.63⁎⁎⁎ 0.588(0.129) (0.183)

SOPH 0.44⁎⁎⁎ 0.653 0.70⁎⁎⁎ 0.603(0.137) (0.164)

Table 10Cross-country correlations between institutional indicators and financial development.In all tables, one, two and three asterisks represent significance at the 10%, 5% and 1%levels respectively.

Ic Contract enforcement Property rights

ENFOR INVP DISC INFOR RULE PTNT COPY PHYS

CRE −0.01 −0.13 −0.22 0.45** 0.47** −0.16 0.62*** 0.44**CAP −0.03 −0.13 −0.21 0.49*** 0.58*** −0.04 0.70*** 0.58***BANK −0.06 −0.04 −0.30 0.42** 0.35* 0.01 0.60*** 0.40**MARG −0.07 −0.06 −0.27 0.48** 0.26 −0.12 0.62*** 0.58***ACCS −0.13 0.33 0.09 0.68*** 0.46** 0.05 0.71*** 0.88***SOPH −0.12 0.34* 0.11 0.74*** 0.54*** 0.12 0.81*** 0.87***


variables, COPY and INFOR, the winner is either COPY or INFOR —

Table 12. The fact that COPY and INFOR outperform financial develop-ment in these regressions suggests that one channel through whichweak IPR enforcementmight impact entry and innovation is bymakingit hard for entrants and innovators to raise external funds, consistentwith the survey results in Table 3.

4.1. Firm size

Another question concerns firm size. Conditioning on indus-try dummies and characteristics, larger firms seem to performdisproportionately more R&D, see the survey of Cohen (2011)(see also Peretto, 1998; Thompson, 2001) for theoretical modelswith this feature). It could simply be that IPRs support largerfirms, and that R&D intensive industries tend to have largerfirms. However, we do not find that R&D intensive industrieshave a systematic link to firm size. Using the firm size indicatorfrom Beck et al. (2008) – the share of firms that have fewer than20 employees in the 2007 US Economic Census24 – we do notfind a significant correlation between firm size and R&D intensi-ty. The correlation is −0.23 and not statistically significant. Atthe country level, the cross-industry correlation between firmsize and RNDi is significant at the five percent level in only twocountries, and the average is 0.11.

24 We were unable to locate these data at the desired level of disaggregation for ear-lier years.

At the same time, considering that firm size is endogenous andpossibly affected by R&D spending, it could be that firm size is an im-portant channel through which IPRs encourage R&D — e.g. by facingless imitation, firms in high-spillover, R&D intensive industriesmight become disproportionately larger, and hence more able to in-ternally finance R&D or internalize knowledge spillovers as inPeretto (1998). To examine this, define Firmsizei,c as the averagefirm size in industry in country c.25 First, we estimate (1) withFirmsizei,c as a dependent variable, to see whether indeed firm sizeis systematically affected by the interaction between desired R&D in-tensity and IPR enforcement. It turns out that, although IPRs dispro-portionately encourage entry into R&D intensive industries, theyalso disproportionately increase size in those industries — seeTable 13. Thus, strong IPRs appear to allow incumbents to expand,as well as encouraging entry. At the same time, when we estimatethe original Eq. (1) with turnover or innovation spending as dependentvariables but conditioning on Firmsizei,c, the coefficients on RNDc× Ic re-main essentially unchanged, nor did Firmsizei,c carry a statistically signif-icant coefficient.While the results concerning firm size are suggestive, itdoes appear that the interaction of RNDc and IPR enforcement affectsfirm size, but that firm size is not in itself an important channel leadingto increased innovation.

4.2. Other moments of the R&D distribution

Ourmeasure of R&D intensity has focused on industry medians. Therationale is that the median firm is representative of what a firm in theindustry would do in the absence of financing or other constraints in anenvironment with reasonably well enforced IPRs, whereas industrymeans might be influenced by outliers. Nonetheless, it is interesting toask whether other moments of the R&D distribution might behave ina similar fashion. This is particularly interesting because the nature ofknowledge spillovers is such that the behavior of outliers might turnout to be important for the behavior of other firms: an innovative outli-er could induce other firms to innovate too as they either imitate orbuild on the innovations of the former.

Thus, we examine whether the interaction with property rightsremains when we measure research intensity using the industrymean (RNDi

mean), skewness (RNDiskew), kurtosis (RNDi

kurt), and the

that IPRs are a country level variable plus the differences-in-differences specificationwith country and industry dummies should ensure adequate identification just as forother instances of regression (1) — see Rajan and Zingales (1998).

Table 14Correlations between different moments of the industry distribution of R&D intensity.RNDi, RNDi

mean, RNDiskew, RNDi

kurt, and RNDinozero are respectively the industry median,

mean, skewness, kurtosis and share of non-zero entries. Source — Compustat.

RNDimean RNDi

skew RNDikurt RNDi

nozero

RNDi 0.76*** 0.57*** 0.64*** 0.53***(0.560) (0.691) (0.870) (0.477)

RNDimean – 0.33* 0.33* 0.45***

(0.000) (0.104) (0.477)RNDi

skew – – 0.95*** 0.24(0.477) (0.477)

RNDikurt 0.24

(0.477)

Table 15Effect of estimating Eq. (1) with different moments of the industry distribution of R&D.

Ic measure COPYc INFORc

R&D moment Turnover InnovADJ Turnover InnovADJ

RNDi 0.462⁎⁎⁎ 0.521⁎⁎⁎ 0.351⁎⁎⁎ 0.352⁎⁎⁎

(0.105) (0.111) (0.107) (0.132)RNDi

mean 0.196* 0.366 0.228* 0.535(0.111) (0.243) (0.095) (0.334)

RNDiskew 0.258* 0.018 0.260** 0.095

(0.151) (0.1076) (0.130) (0.069)RNDi

kurt 0.214 0.059 0.223* 0.064(0.150) (0.083) (0.124) (0.078)

RNDinozero 0.633*** 0.496*** 0.647*** 0.235*

(0.150) (0.123) (0.158) (0.143)

Table 13Effect of estimating Eq. (1) with Firmsizei,c as the dependent variable.

Ic βRND R2

COPYc 1.30*** 0.461(0.527)

INFORc 1.81*** 0.456(0.563)


share of firms that report non-zero R&D (RNDinozero). This last measure

is interesting because a lot of firms report zero R&D (see Cohen et al.,1987), and if the share of firms that perform R&D turns out to be re-lated to the innovative impact of knowledge spillovers would supportthe hypothesis of Cohen and Levinthal (1990) that R&D is necessaryfor firms to benefit from knowledge spillovers. All of these measuresturn out to be correlated, see Table 14.

Table 15 displays estimates of the interaction of different mo-ments of the R&D distribution with IPRs, based on Eq. (1). All themeasures in Table 14 interact with IPRs in the same manner as the in-dustry mean, although statistical significance is rare and confined tothe case when the dependent variable is turnover rather than innova-tion spending. The exception is RNDi

nozero, especially when IPRs aremeasured using COPYc. This supports the idea of a firm's “absorptivecapacity” depending on whether or not it performs R&D, as opposedto the Spencerian view whereby spillovers are “free.” It also suggeststhat the behavior of outliers is not so important for innovation at theindustry level in general.

5. Concluding remarks

This paper takes a simple yet novel approach to identifying therelationship between knowledge spillovers, innovation and entre-preneurship, exploiting cross-country differences in the institutionsthat limit the use of knowledge spillovers for imitation. The resultsindicate that strong IPR enforcement is associated with dispropor-tionately greater innovation spending, entry and exit in industriesthat have higher “desired” R&D intensity (the R&D intensitydisplayed when IPR protection is strong). This suggests that, aslong as intellectual property can be controlled by the innovator, larg-er knowledge spillovers are associated with more innovation, andwith innovative entry.26 Most importantly, they suggest that new

26 It is interesting that in some creative destruction models equilibrium prices are de-termined by the presence of a competitive fringe that may copy an agent's innovations(e.g. Aghion et al., 2005), whereas the survey data suggest that it is incumbents whoare the main threat to innovating entrepreneurs.

knowledge is often a substitute (not a complement) to some aspectof prior knowledge, so that in a relative sense entrepreneurs benefitmore from larger knowledge spillovers than incumbents.27 It is alsoimportant that patent protection does not seem to play much of arole: intellectual property is much broader than just patentableknowledge, so broader measures of IPR enforcement (copyright pro-tection and non-formalism in the legal system) turn out to be the in-stitutional variables that matter, which protect (or proxy for theprotection of) IP in the form of texts, software, brands, designs, etc.as well as scientific or engineering knowledge.

In future work, it could be interesting to see whether thesource of knowledge spillovers matters for any of these results.For example, most knowledge spillovers are known to be amongfirms in the same industry, which is why we do not distinguishbetween “source” and “benefit” industries: however, patent cita-tion or other data might allow a more careful identification of theimportance of different sources of knowledge spillovers for thegeneration of new knowledge — see Cai and Li (2012) for workin this direction. Also, as mentioned, whether trade linkages in-teract with IPRs could be assessed using more disaggregateddata on trade flows and benchmark R&D. It would be interestingto expand the sample to see whether any additional factors playa role in innovative and entrepreneurial behavior in developingeconomies.

Finally, the results have policy implications, not least that weakIPRs and weak court systems may constitute a drag not only on in-novation but also on entrepreneurship. Also, the findings thatcross-firm spillovers matter and that they favor entrants, implythat the externalities typically identified in creative destructionmodels are present — the externality from the business-stealing ef-fect and the externality from the fact that an innovator may inspireother innovators (or may suffer from imitation). Djankov et al.(2002) find that barriers to entry are strongly related to GDP perperson, whereas traditional steady-state general equilibriummodels for policy evaluation cannot account for such large effects(see Moscoso Boedo and Mukoyama, 2012). The results suggestthis may be because, by reducing entrepreneurship, barriers toentry stifle an important channel of innovation, so that an endoge-nous growth framework may be more suited for policy analysis ofthis kind.

27 This is not to suggest that no aspect of new knowledge is complementary to oldknowledge, nor that incumbent-driven innovation is not important for economicgrowth. An alternative interpretation could be that substitution is dominant in R&D in-tensive industries and complementarity is dominant in non-R&D intensive industries.At the same time, that interpretation still suggests that substitution is most relevantfor studying the phenomenon of R&D.

(continued)

Industry RND Turnover Entry Exit

Information and data processing 3.84 25.60 14.50 11.11Finance (not insurance, trusts) 0.51 19.87 11.05 8.82Insurance, trusts 3.68 12.67 6.85 5.82Real estate 0.00 20.12 12.25 7.87

Table A2. (continued)


Appendix A.

A Summary country and industry data

Table A1. Summary statistics: average annual rates of turnoveracross countries. Source — Eurostat.

Country Turnover Entry Exit InnovRAW InnovADJ

Belgium 14.5 7.0 7.5 2.7 1.4Bulgaria 19.8 10.4 9.4 2.1 0.3Czech Rep. 18.1 9.3 8.8 3.5 1.3Denmark 15.0 7.7 7.3 3.5 1.8Germany – – – 3.3 2.1Estonia 19.1 10.5 8.6 2.4 1.2Ireland – – – 2.4 1.5Greece – – – 6.2 2.2Spain 14.6 8.6 6.0 1.5 0.5France 12.8 7.2 5.6 3.3 1.1Italy 14.0 7.6 6.4 2.8 1.0Cyprus – 4.1 – 4.0 1.8Latvia 21.6 13.6 8.0 – –

Lithuania 20.0 11.9 8.1 2.5 0.7Luxembourg 16.4 9.6 6.8 2.2 1.1Hungary 18.2 10.0 8.2 2.3 0.5Malta – – – 1.7 0.4Netherl. 16.2 8.4 8.2 2.0 0.7Poland – – – 2.6 0.7Portugal 15.3 8.8 6.5 2.1 0.9Romania 25.3 16.9 8.4 3.4 0.7Slovenia 13.5 8.0 5.5 – –

Slovakia 17.4 9.5 7.9 3.2 0.7Finland 13.1 7.0 6.1 – –

Sweden 10.9 6.0 4.9 4.7 2.4UK 21.6 11.1 9.5 – –

Norway 18.4 10.6 7.8 1.8 0.7Switzerland 7.3 3.5 3.8 – –

Rental services 0.00 20.58 11.39 9.19Legal services 0.00 19.93 12.40 7.52Systems design 13.64 24.87 15.37 9.49Technical Services 12.60 21.06 12.41 8.65Waste disposal 2.33 14.91 8.28 6.63Education 0.00 19.09 11.17 7.92Healthcare 0.00 12.61 7.85 4.76Arts, sports, amusement 0.00 22.53 13.22 9.31Hotels 0.00 14.22 7.83 6.39Restaurants 0.00 18.74 9.28 9.46Other services 0.00 18.76 10.68 8.08Median 0.47 14.97 7.87 7.52

Table A2. Summary statistics: annual industry rates of R&D inten-sity and turnover. R&D intensity is the median ratio of R&D spendingto sales. Entry, exit and turnover are industry fixed effects plus themedian country fixed effect. All variables are measured over the peri-od 1997–2006. Sources — Eurostat, Compustat.

Industry RND Turnover Entry Exit

Oil and gas extraction 0.18 13.65 6.65 6.99Other mining 0.82 11.26 5.89 5.37Utilities 0.38 11.19 6.95 1.21Construction 0.22 17.94 10.40 7.53Wood products 0.00 13.91 6.59 7.32Nonmetal products 0.73 12.89 6.38 6.50Primary and fabricated metal prod. 0.77 13.83 7.39 6.43General Machinery 2.77 11.55 5.96 5.58Computers and electronic prod. 13.68 12.65 6.19 6.46Electrical machinery 10.79 5.51 2.65 2.86Transport Equip. 2.28 13.96 7.50 6.45Manuf n.e.c. 11.91 14.97 7.62 7.35Food products 0.56 12.31 5.15 7.16Textiles 1.38 1700 7.63 9.36Leather 0.00 14.24 5.66 8.57Paper, printing, software 0.84 14.71 7.57 7.15Petroleum and coal products 0.47 13.54 7.85 5.68Chemicals 32.18 11.18 5.48 5.71Plastics 1.30 11.00 5.47 5.53Wholesale Trade 0.00 18.95 9.59 9.36Retail Trade 0.00 17.88 8.38 9.50Air Transport 0.00 17.45 9.20 8.26Water transport 0.00 18.26 9.33 8.93Land transport 0.00 15.14 7.43 7.71Transport support 0.00 16.95 9.43 7.52Broadcasting 1.58 27.81 16.80 11.01

References

Acemoglu, Daron, Johnson, Simon, 2005. Unbundling institutions. Journal of PoliticalEconomy 113 (5), 949–995.

Acemoglu, Daron, Cao, Dan, 2010. Innovation by entrants and incumbents. MIT, Mimeo.Aghion, Philippe, Howitt, Peter, 1992. A model of growth through creative destruction.

Econometrica 60 (2), 323–351.Aghion, Philippe, Howitt, Peter, 1997. Endogenous Growth Theory. MIT Press, Cam-

bridge, MA.Aghion, Philippe, Howitt, Peter, Mayer-Foulkes, David, 2005. The effect of financial de-

velopment on convergence: theory and evidence. Quarterly Journal of Economics120 (1), 173–222.

Aghion, Philippe, Fally, Thibault, Scarpetta, Stefano, 2007. Credit constraints as a barrierto the entry and post-entry growth of firms. Economic Policy 22 (52), 731–779.

Akcigit, Ufuk, Kerr, William R., 2010. Growth through heterogeneous innovations.NBER Working Paper No. 16443.

Arora, Ashish, Ceccagnoli, Marco, Cohen, Wesley M., 2003. R&D and the patent premi-um. NBER Working Paper No. 9431.

Audretsch, David B., 1991. New-establishment survival and the technological regime.The Review of Economics and Statistics 73 (3), 441–450.

Beck, Thorsten, Demirgüç-Kunt, Asli, Levine, Ross, 2000. A new database on financialdevelopment and structure. World Bank Economic Review 14, 597–605.

Beck, Thorsten, Demirguc-Kunt, Asli, Laeven, Luc, Levine, Ross, 2008. "Finance, firmsize, and growth. Journal of Money, Credit and Banking, vol. 40(7). Blackwell Pub-lishing, pp. 1379–1405.

Bernstein, Jeffrey I., Nadiri, M. Ishaq, 1988. Interindustry R&D spillovers, rates of return,and production in high-tech industries. American Economic Review 78 (2),429–434.

Botero, Juan C., Djankov, Simeon, La Porta, Rafael, Lopez-de-Silanes, Florencio, Shleifer,Andrei, 2004. The Regulation of Labor Quarterly Journal of Economics 119 (4),1339–1382.

Brandt, Nicola, 2004. Business dynamics in Europe. OECD Science, Technology and In-dustry Working Paper 2004/1.

Branstetter, Lee G., Fisman, Raymond, Fritz Foley, C., 2006. Do stronger intellectualproperty rights increase international technology transfer? Empirical evidencefrom U.S. firm-level panel data. Quarterly Journal of Economics 121 (1), 321–349.

Browne, Ciara, Bryden, Richard, Delgado, Mercedes, Geiger, Thierry, 2008. ExecutiveOpinion Survey: Capturing the Voice of the Business Community. World EconomicForum, Geneva.

Cai, April J. and Nan Li. 2012. Growth through intersectoral knowledge linkages.Mimeo: International Monetary Fund.

Carlin, Wendy, Mayer, Colin, 2003. Finance, investment, and growth. Journal of Finan-cial Economics 69 (1), 191–226.

Claessens, Stijn, Laeven, Luc, 2003. Financial development, property rights, and growth.Journal of Finance 58 (6), 2401–2436.

Coe, D., Helpman, E., 1995. International R&D spillovers. European Economic Review39, 859–887.

Cohen, Wesley M., 2011. Fifty years of empirical studies of innovative activity and per-formance. In: Hall, B., Rosenberg, N. (Eds.), Handbook of the Economics of Innova-tion, pp. 129–213. Chapter 7.

Cohen, Wesley M., Levinthal, Daniel A., 1990. Absorptive capacity: a new perspectiveon learning and innovation. Administrative Science Quarterly 35, 128–152.

Cohen,Wesley M., Levin, Richard C., Mowery, David C., 1987. Firm size and R&D intensity:a re-examination. The Journal of Industrial Economics 35 (4), 543–565.

De Soto, Hernando, 1987. El Otro Sendero. Editorial La Oveja Negra, Lima.Djankov, Simeon, La Porta, Rafael, Lopez-De-Silanes, Florencio, Shleifer, Andrei, 2002.

The regulation of entry. Quarterly Journal of Economics 117 (1), 1–37.Djankov, Simeon, La Porta, Rafael, Lopez-de-Silanes, Florencio, Shleifer, Andrei, 2003.

Courts. Quarterly Journal of Economics 118 (2), 453–517.Dunne, Timothy, Roberts, Mark J., Samuelson, Larry, 1988. Patterns of establishment

entry and exit in U.S. manufacturing industries. The RAND Journal of Economics19 (4), 495–515.

Feenstra, Robert C., Lipsey, Robert E., Deng, Haiyan, Ma, Alyson C., Mo, Hengyong, 2005.World trade flows: 1962–2000. NBER Working Paper No. 11040.


Gans, Joshua S., Hsu, David H., Stern, Scott, 2002. When does start-up innovation spurthe gale of creative destruction? The RAND Journal of Economics 33 (4), 571–586.

Geroski, Paul A., 1989. Entry, innovation and productivity growth. The Review of Eco-nomics and Statistics 71 (4), 572–578.

Ginarte, Juan C., Park, Walter G., 1997. Determinants of patent rights: a cross-nationalstudy. Research Policy 26 (3), 283–301.

Griffith, Rachel, Redding, Stephen, Van Reenen, John, 2004. Mapping the two faces ofR&D: productivity growth in a panel of OECD industries. The Review of Economicsand Statistics, vol. 86(4). MIT Press, pp. 883–895 (December).

Ha, Joonkyung, Howitt, Peter, 2007. Accounting for trends in productivity and R&D: aschumpeterian critique of semi-endogenous growth theory. Journal of Money,Credit, and Banking 39 (4), 733–774.

Howitt, Peter, 1999. Steady endogenous growth with population and R&D inputs grow-ing. Journal of Political Economy 107 (4), 715–730.

Ilyina, Anna O., Samaniego, Roberto M., 2011. Technology and financial development.Journal of Money, Credit and Banking Journal of Money, Credit and Banking 43(5), 899–921.

Ilyina, Anna O., Samaniego, Roberto M., 2012. Structural change under financing con-straints. Journal of Monetary Economics 59 (2), 166–179.

Jaffe, Adam B., Trajtenberg, Manuel, Fogarty, Michael S., 2000. Knowledge spilloversand patent citations: evidence from a survey of inventors. American Economic Re-view 90 (2), 215–218.

Jones, Charles, 1995. R&D-Based Models of Economic Growth. Journal of Political Econ-omy 103, 759–784.

Kaufman, Daniel, Kraay, Aart, Masturzzi, Massimo, 2007. Governance Matters VI: Ag-gregate and Individual Governance Indicators 1996–2006. World Bank Policy Re-search Working Paper 4280.

Keller, Wolfgang, 1998. "Are international R&D spillovers trade-related?: analyzingspillovers among randomly matched trade partners. European Economic Review,vol. 42(8). Elsevier, pp. 1469–1481 (September).

Klapper, Leora, Luc, Laeven, Rajan, Raghuram, 2006. Entry regulation as a barrier to en-trepreneurship. Journal of Financial Economics 82, 591–629.

Klenow, Peter J., 1996. Industry Innovation: Where andWhy? Carnegie-Rochester Con-ference Series on Public Policy 44, 125–150.

Klette, Tor Jakob, Kortum, Samuel, 2004. Innovating Firms and Aggregate Innovation.Journal of Political Economy 112 (5), 986–1018.

Klevorick, Alvin K., Levin, Richard C., Nelson, Richard R., Winter, Sidney G., 1995. On thesources and significance of interindustry differences in technological opportuni-ties. Research Policy 24, 185–205.

Krusell, Per, 1998. Investment-specific R&D and the decline in the relative price of cap-ital. Journal of Economic Growth 3, 131–141.

Laincz, Christopher, Peretto, Pietro, 2006. Scale effects in endogenous growth theory:an error of aggregation not specification. Journal of Economic Growth, vol. 11(3).Springer, pp. 263–288 (September).

La Porta, Rafael, Lopez-de-Silanes, Florencio, Shleifer, Andrei, 2006. What Works inSecurities Laws? Journal of Finance 61 (1), 1–32.

Lerner, J., 2005. 150 years of patent office practice. American Law and Economics Re-view 7, 112–143.

Levin, Richard C., Cohen, Wesley M., Mowery, David C., 1985. R&D appropriability, op-portunity, and market structure: new evidence on some Schumpeterian hypothe-ses. American Economic Review 75, 20–24.

Madsen, Jakob B., 2008. Semi-endogenous versus Schumpeterian growth models: test-ing the knowledge production function using international data. Journal of Eco-nomic Growth 13 (1), 1–26.

Mansfield, Edwin, 1986. Patents and innovation: an empirical study. Management Sci-ence 32 (2), 173–181.

Moscoso Boedo, Hernan J., Mukoyama, Toshihiko, 2012. Evaluating the effects of entryregulations and firing costs on international income differences. Journal of Eco-nomic Growth 17 (2), 143–170.

Ngai, L. Rachel, Samaniego, Roberto M., 2011. Accounting for research and productivitygrowth across industries. Review of Economic Dynamics 14 (3), 475–495.

Nicoletti, Giuseppe, Scarpetta, Stefano, Boylaud, Olivier, 2000. Summary indicators ofproduct market regulation with an extension to employment Protection Legisla-tion. OECD Economics Department Working Papers, 226.

North, Douglass, 1984. Government and the cost of exchange in history. Journal of Eco-nomic History 44, 255–264.

Peretto, Pietro F., 1998. Technological change, market rivalry, and the evolution ofthe capitalist engine of growth. Journal of Economic Growth, vol. 3(1). Springer,pp. 53–80 (March).

Porta, La, Rafael, Florencio Lopez-de-Silanes, Shleifer, Andrei, Vishny, Robert W., 1998.Law and finance. Journal of Political Economy 106 (6), 1113–1155.

Qian, Yi, 2007. Do national patent laws stimulate domestic innovation in a globalpatenting environment? A cross-country analysis of pharmaceutical patent protec-tion, 1978–2002. The Review of Economics and Statistics 89 (3), 436–453.

Rajan, Raghuram G., Zingales, Luigi, 1998. Financial dependence and growth. AmericanEconomic Review 88 (3), 559–586.

Rockett, K., 2011. Property rights and invention. In: Hall, B., Rosenberg, N. (Eds.), Hand-book of the Economics of Innovation, pp. 315–380 (Chapter 10).

Romer, Paul, 1990. Endogenous technological change. Journal of Political Economy 98(5), S71–S102.

Samaniego, Roberto M., 2010. Entry, exit and investment-specific technical change.American Economic Review 100 (1), 164–192.

Schumpeter, Joseph Alois, 1934. The Theory of Economic Development. Translator:Redvers Opie. Harvard University Press, Cambridge, MA . (Orig. pub. 1911).

Spence, Michael A., 1984. Cost reduction, competition and industry performance.Econometrica 52 (1), 101–122.

Teece, David J., 1986. Profiting from technological innovation: implications for integra-tion, collaboration, licensing and public policy. Research Policy 15, 285–305.

Thompson, Peter, 2001. The microeconomics of an R&D-based model of endogenousgrowth. Journal of Economic Growth, vol. 6(4). Springer, pp. 263–283 (December).

White, Halbert, 1980. A heteroskedasticity-consistent covariance matrix estimator anda direct test for heteroskedasticity. Econometrica 48 (4), 817–838.

Wooldridge, Jeffrey M., 2002. Econometric Analysis of Cross Section and Panel Data.MIT Press, Cambridge, MA.

World Bank, 2006. Doing Business in 2007: How to Reform. World Bank, Washington.

knowledge spillovers and intellectual property rights

Documents