Tica Susi PxhjVhvHXVHvvxHVx

8/10/2019 Tica Susi PxhjVhvHXVHvvxHVx

1/21

Strategic Management Journal

Strat. Mgmt. J.,32: 11281138 (2011)

Published online EarlyView in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.925

Received 16 July 2009;Final revision received 5 October 2010

RESEARCH NOTES AND COMMENTARIES

ESTIMATING THE PATENT PREMIUM: EVIDENCE

FROM THE AUSTRALIAN INVENTOR SURVEY

PAUL H. JENSEN,* RUSSELL THOMSON, and JONGSAY YONGMelbourne Institute of Applied Economic and Social Research, and Intellectual PropertyResearch of Australia (IPRIA), The Uniersity of Melbourne, Par!ille "ictoria, Australia

In this paper we use novel survey data on 1,790 Australian inventions to estimate the averagepatent premium, which provides an important benchmark for technology managers and IPprofessionals. Our data are drawn from the Australian Inventor Survey, which was sent to allAustralian applicants who submitted a patent application to the Australian Patent Office from1986 to 2005. Since some patent applications were unsuccessful, we have information about the

private value of both patented and unpatented inventions. Our results suggest that the presence ofa patent increases the returns to an invention by around 40 to 50 percent regardless of how wedefine value. Copyright 2011 John Wiley & Sons, Ltd.

INTRODUCTION

The introduction of new products andprocesses plays a pivotal role indriving firms productivity,profitability, and competitiveadvantage (Cecca-gnoli, 2009).However, innovators face animportant challenge in choosingamong different appropriationstrategies including patenting and

keeping ahead of the opposition, thechoice of which can have far-reaching consequences for firmprofitability. Survey evidence hassuggested

Keywords: patents; invention;commercialization; patent premium*Correspondence to: Dr. Paul H. Jensen,Melbourne Institute of Applied Economic andSocial Research, and Intellectual Property

Research Institute ofAustralia (IPRIA) Level 7,Alan Gilbert Building, The

University of Melbourne,Parkville VIC 3010,Australia. E-mail:[email protected]

that, on average,patents are considered

relatively less

important than other

appropriation

strategies, but that the

value of patenting

varies consider-ably

across technology

areas (Levin et al.,

1987; Harabi, 1995;

Cohen, Nelson, and

Walsh, 2000). In part,

this observation may

reflect the fact that not

all technologies are

patentable. In this

paper we use novel

survey data on 1,790

Australian inventions

to estimate the

average patent


2/21

premium, which pro-vides an

important benchmark for technology

man-agers and intellectual property

(IP) professionals.

To understand our estimates of the

patent pre-mium, it is important todistinguish between the value of an

invention and thevalue of patent pro-

tection. The value ofan invention isdefined as thediscounted flow of

profits generated

over the course ofthe inventions

economic life, whilethe value of patentprotection is definedas the return

Copyright 2011 John Wiley &Sons, Ltd.


3/21

over and above that which could havebeen gen-erated by the second bestmeans of appropriation (Schankerman,

1998; Lanjouw, 1998).1 The value ofpatent protection can be defined aseither the incremental return (i.e., indollar terms) or as the proportional

increase in value to an invention due topatent protection. The latter is referredto as the patent premium by Arora,Ceccagnoli, and Cohen (2008) and isanalogous to Schankermans (1998: 95)equivalent subsidy rate.

Existing studies of the value of patentprotection can be divided into threestreams. The first relates to surveys offirms and inventors, which show thatpatenting ranks near the bottom of allappropria-tion mechanisms (see Levin

et al., 1987; Cohen et al., 2000).Similarly, Mansfield and collabo-rators(Mansfield, Schwartz, and Wagner,1981; Mansfield, 1986) found patentprotection to be important in thecommercialization of a minor-ity ofinnovations. Arora et al. (2008)developed a model of innovation andpatenting to directly esti-mate the patentpremium and its impact on research anddevelopment (R&D) investment. Usingsurvey data, they modeled the patentpremium as con-sisting of a fixed-firmcomponent as well as an idiosyncraticcomponent and then jointly estimatedthe firms R&D productivity, patentpropensity, and patent premium. Theyshowed that the mean premium ispositive in only a few industries, butthat conditional on patenting, theaverage patent premium is 0.5.

The second stream of studies involves

imputing the value of patents from patent

renewal decisions. The seminalcontribution is Schankerman and Pakes

(1986), while more recent studies include

Sampat and Ziedonis (2004), Deng

(2007), and Gronqvist (2009). The

premise of this approach is that firms only

renew their patents if the value of patent

protection is greater than the cost of

renewal. The method involves estimating

param-eters that describe the distribution

of the initial value of patented inventions

as well as depreciation rates. It is

generally acknowledged that the highly

skewed distribution of patent values make

finite sample estimation unreliable (see

Bessen, 2009). Schankerman (1998: 95)

generated an estimate of the equivalent

subsidy rate by dividing his esti-mates of

the value of patent protection by the total

1 This implies that, in the absence of patent

protection, firms are able to generate some returnsusing alternative means of appropriation.

Copyright 2011 John Wiley & Sons, Ltd.


4/21

Research Notes and Commentaries

R&D expenditure used to producethose patents (in practice this ismeasured by aggregate R&Dexpenditure in the previous year) givingan esti-mated effective subsidy rate of

0.25.2

The third stream of research attempts

to infer the value of patent protectionfrom its effect on the market value offirms (typically Tobins Q). Exam-plesof this approach include Griliches(1981); Hall, Jaffe, and Trajtenberg(2005); Hall and Mac-Garvie (2006);Bosworth and Rogers (2001); and,Ceccagnoli (2009). Bessen (2009)argued that, in principle, theincremental value of patents can beisolated if we can control for the firmstotal quality-adjusted technology stock.

However, this approach relies on theefficient markets hypothe-sis, which isnot universally considered to be arobust assumption.

In this paper, we adopt a new approach

to this issue based on analysis using data

drawn from the Australian Inventor

Survey 2007 (AIS-07),3which was sent to

all Australian applicants who submitted a

patent application to the Australian Patent

Office from 1986 to 2005. A major

difference between the AIS-07 and other

inventor surveys (see Harhoff et al., 1999;Giuri et al., 2005; and Gambardella,

Harhoff, and Verspagen, 2008) is that we

sur-vey patent applicants rather than

patentees. Since some patent applications

were unsuccessful, we have information

about the private value of both patented

and unpatented inventions. Moreover,

there is considerable variation in the

commercial-ization outcomes across

patented and unpatented inventions. This

variation is the key to our empir-ical

identification of the patent premium. We

also attempt to separate the patent

premium from observable quality

characteristics of the invention.4

Our paper makes a number of

contributions. First, we present new

estimates of the aver-age patent premium

based on a novel empirical approach. All

of the inventions in our counterfac-tual

sample are potentially patentable, as

indicated by the decision of the applicants

to apply for a patent.5Second, we present

estimates of the private

2 See similar analysis by Lanjouw (1998)for West German patents during the period 19531988.3 For more details of the AIS-07, see Webster andJensen (2009).

4Ideally, we might use independent evaluations ofinvention quality (e.g., Moser, 2007) to achieve this.However, such data are not available for our sampleof inventions.

5 Moreover, note that since our estimated

patent premium is conditional on a patent

application being made, inventions for which it is

optimal to appropriate returns via secrecy are not

included.

Strat. Mgmt. J.,32:11281138 (2011)

DOI:10.1002/sm

j


5/21

1130 P. H. Jensen, R. Thomson, and J. Yong

value of

Australian

inventions.

Finally, our paper

provides estimates

of the

heterogeneity ofthe patent

premium across

technology areas.

In this light, our

results have

important

implications for

the strategic

management of

inventive activity.

Estimating the

patent premium

also has important

public policy

implications.

Governments

around the world

employ an array

of policies

designed to

stimulate

innovative

activities. Forexample, across

the Organisation

for Economic Co-

operation and

Development

(OECD),

governments

typically fund

between 10 and

20 percent of total

business invest-

ment in R&D.6

Estimates of the

patent premium

enable policy

makers to

consider the size

of the implicit

subsidy provided

to inventors via

the patent system

vis-a`-vis other

subsidies.

SUR

VEY

DAT

A

AND

EMPI

RICA

L

APP

ROA

CH

The AIS-07involvedsending aquestionnaire

to everyAustralianinventor whosubmitted apatentapplication tothe AustralianPatent Officebetween 1986

and 2005.7Therelationshipbetween inven-

tor andinvention ismany to many.To deal withthis, we sentthe survey toeach listedinventor on apatent

application.8

Where aninventor hadmultipleinventions, thesurvey askedquestionsregarding amaximum offive inventions.We con-ductedtwo separatemailings of thesurvey in Julyand Decemberof 2007.

In total,

there were43,200inventor-applicationpairs in thepopulation thatrelated to

31,313 uniquepatentapplications(i.e.,inventions). Onthe basis of the number of surveys returnedto us unopened (and a post-enumeration sur-vey ofnonrespondents), we estimatethat therewere 5,446 inventions with valid addresses at

the time of the mailing.9 In total, we received6

Figures for 2005or nearest availableyear, calculatedfrom OECD (2008a,2008b).

7As in the PatVal-EU survey, we sentthe survey to theinven-tor rather thanthe owner of theinvention. While itis possible that theinventor knows lessthan the ownerabout commercial-ization outcomes,Gambardella et al.(2008) comparedthe value estimatesfrom a sample ofFrench inventorsand patent ownersand found that thebias introduced wasnegligible. We haveno reason to believethat such a bias isany larger inAustralia.

8Where multiplesurveys werereturned on thesame invention,only one (randomlyselected) responsewas used.

9 The low

proportion of

inventions with

valid inventor

addresses is a result

of the fact that the

survey included

patent applica-tions

dating back to 1986.

Since the Australian

Patent Office

Copyright 2011 JohnWiley & Sons, Ltd.


6/21

completedquestionnairesrelating to 3,736uniqueinventions.

The data sethas 2,448observations

with non-missinginvention values.Of these, 645were stillpending a patentofficeexaminationdecision at thetime the surveywasadministered.These appli-

cations wereexcluded fromthe analysisbecause ourfocus here is onthe estimation ofthe pre-miumassociated with a

patent grant.10 Afurther 13observations hadmissing

information onother explanatoryvariables andwere alsoexcluded,leaving a finalsample of 1,790observations.

Survey

responses came

from inventors in

a range of

employmentarrangements:

companies (43.7

percent), public-

sector research

organizations (5.9

percent), and

individuals (50.4

percent). The

inventions in the

sample of survey

respondents

covered a broad

cross-section of

different

technol-ogy

areas, which

were classified

using the

United

Kingdom Officeof Science and

Technology-

Inter-national

Patent

Classification

(OST-IPC)

technol-ogy

concordance.

The distribution

by technology

area suggeststhat our sample

is broadly

represen-tative

of the

population of

patent

applications.

By applying

for a patent,the inventorhas sig-naled

that his or hertechnology ispatentablesubject matter.Since the

patentapplicationpro-cessinvolves

disclosure ofthe technology,

the inventorhas foregone

the possibilityof relying on

secrecy toappropriate

returns. Thisindicates that

our sampleincludes onlyinventions forwhich

patenting was

considered tobe the bestoption (Moser,2007). It also

suggests thatour estimatesare an upperbound of the

ex ante valueof patentprotection

since it isharder toappropriatereturns without

patentprotection once

disclosure hasoccurred

(Horstmann,MacDonald,

and Slivin-ski,1985;

Schankerman,1998).

ESTIMATES

OF

INVENTION

VALUE

The privatevalue of aninvention is thediscountedflow of profitsthat accrue toits owner overthe

does not updateinventors contactdetails, we wereunable to contactinventors whosecontact address hadchanged for anyreason (e.g., jobchange, retirement,or changes inpersonalcircumstances).10

As a robustnesscheck, we alsoundertook theanalysis by

including


7/21

applicationscategorized aspending. This did notchange the substantive

results.

Strat.

Mgmt. J.

32: 1128

1138 (2011)

DOI: 10.1002/smj

course of its economiclife. Invention value isdif-ficult to measure in

practice. Economistswould typicallyapproach such aproblem using pricesobserved in markettransactions. However,the vast majority ofpatents are not traded in

an open mar-ket.11 Anincreasingly popularalternativewhich weuse in this paperis to

ask inventors to esti-mate the value of theirinvention.

Like PatVal-EU, our

survey data include

inven-tions of different

vintages. The initial

PatVal-EU survey, which

was conducted in 2003

2004, includes a sample

of patents with a priority

date between 1993 and

1997. Both PatVal-EUand AIS-07 require

respondents to estimate a

total value including

returns already generated

as well as expected future

returns.12 PatVal-EU

asked inven-tors to report

the price they would be

willing to sell the

invention for at the time

the patent was issued.Our approach is different

in that we sep-arate

historical returns (i.e.,

returns generated up to

the time of the survey in

2007) from the future

value (i.e., the discounted

value of expected future

profits after 2007) of the

invention. This eases the

computational burden

imposed on the inventors

and enables us to

evaluate the sensitivity of

our results to different

measures of invention

value. This also means

that our forward-lookingmeasure of value is in

constant (2007) prices.

We asked inventorsthe following threeques-tions in relation toinvention value:

(1) To date, what isyour estimate ofsales rev-enuefrom products and

processes usingthis invention?

(2) If this patent hasbeen licensed,what is your bestestimate of licensing revenuesto date?

(3) If you wereselling this patentor invention today,

what price wouldyou be willing toaccept for it?

In each case, thesurvey respondent hadto select one of sixpossible value ranges(in Australian dollars):below $100,000,$100,001 $500,000,$500,001$1million,

$1$2 million,11

The few that are subject toobservable trades are highlyselected. Serrano (2006) andSneed and Johnson (2009) haveexamined some such patenttransfers (via auctions) in orderto understand the market forpatents.

12 This is quite

different to Harhoff et al.

(1999), who surveyed one

cohort of patented inventions,

namely all patent applications


8/21

in Germany in 1977. Thus, the

Harhoff et al. (1999) survey

covered revenues actually

generated by the invention.

Copyright 2011 John Wiley &

Sons, Ltd.

Research Notes and Commentaries 1131

$2$10 million, above$10 million. To convertthe responses tonumerical values, wetook the mid-point ofeach interval. Since the

highest value category(above $10 million) isunbounded, we imposedan upper bound of $50

million.13

Questions (i) and (ii)focus on returns thathave already beengenerated, capturingsales revenue

generated14 andlicensing revenues,

respectively. Question(iii) reflects the netpresent value of expected future returns.To estimate inventionvalue, we useinformation fromresponses to all threequestions. SinceQuestion (i) is revenue-rather than profit-based,we set the gross marginat 30 percent for goods

and services producedusing an invention.15

We then added the threecompo-nents of valuetogether to construct theprivate invention value,which we denote

InvVal.16

We argue that it is

important to include both

past profits and expected

future revenue in our

mea-sure of inventionvalue. Since we have a

mix of inventions of

different ages, simply

using data on total

revenue to date will

underestimate the value

of newer inventions. The

value of inventions with

faster rates of

depreciation may also be

system-atically

overestimated. As


9/21

depreciation of inven-

tion value is in part a

function of diffusion, it is

likely to be related to

patent grant status.

However, we

acknowledge that survey

questions relating to

expected future revenuesare perhaps noisier and

less reliable than

questions based on actual

rev-enues. Accordingly,

we also present results

from regressions using

the answer to Question

(i) as the dependent

variable (with controls

for priority year)which

we refer to as SalesRev.17

The distribution ofestimated private inven-tion value (InvVal ) ishighly skewed: themean and medianinvention values in oursample are A$6.6million and A$800,000respectively, which isbroadly consistent withother inventor surveys(see Gambardella et al.,

2008). The difference in13

We also computed all resultsusing values of $100m and$200m as the upper bounds,which did not change theresults.

14It may, however, be a weak

measure of process inventionvalue since processes are morelikely than product inventionsto be used in-house, andtherefore do not typicallygenerate sales revenue.

15We have also checked ourresults by taking the value of aninvention at 100 percent of therevenue of all goods andservices sold. This variationproduces essentially the sameresults.

16With regard to licensingrevenue, we set the value tozero if the respondent did notanswer the question or wasunsure of the amount. Thus, weprobably understate the actuallicensing revenue.

17 We thank an

anonymous referee for

making this suggestion.

S

t

r

a

t.

M

g

m

t.

J.,

3

2

:

1

1

2

8

1

1

3

8

(

2

0

11

)

DOI: 10.1002/smj


10/21


Figure 1. Distribution of private invention value, by patent grant status

invention value by patent grant statusis pre-sented in Figure 1. From this,we can see that there areproportionately more valuable inven-tions (above A$1 million) amonginventions pro-tected by a patent.However, many inventions that arenot protected by a patent are highlyvaluable. The difference between themean value of the patented and non-

patented inventions is 9.4 percent.18

Thus, patented inventions are morevaluable on average, but unpatentedinventions are far from worthless.

We also undertook a range ofanalyses to probe the robustness ofour value measure. First, it isreassuring that responses to Question(i) and Ques-tion (iii) are positively

correlated.19We also con-sidered thedistribution of invention values by

commercialization stage.20 As onewould expect, private value was

found to be increasing in the stage ofcommercialization attempted.21

18The mean values are A$6.7 million and A$6.1million for patented and unpatented inventionsrespectively. The difference in median inventionsvalues for the two groups, however, is muchlarger.

19 The raw correlation is 0.33. In a log-on-log regression model controlling for year ofapplication (and therefore obsolescence), thecoefficient on the response to Question (iii) is 0.5,with a t-value above 25.20 Webster and Jensen (2009) consider the impactof patent grant on attempts to commercializeinnovations using data from the AIS-07. Theyshow that, on average, patents play a small butimportant role in shaping attempts tocommercialize inventions.

21 Ideally, we

would also test the validity

of our measure of invention

value by comparing them

with other indicators of

patent value, such as the

number of forward

citations. However, this is

not possible in this instance

because disclosure of prior

art is voluntary at the

Australian Patent Office.


EMPIRICAL

MODEL

Let Vij denote the totalprivate value of invention i in technologyarea j , i = 1, . . ., njand j=1,. . ., J. We specify alinear model wherethe value of aninvention depends onwhether a patent hasbeen granted(conditional on apatent applicationbeing made).

ln Vij=Gij+X+j+ij, (1

where Gij is a binary variable taking the valuof unity if a patent hasbeen granted and zerootherwise,22 X is avector of additionalexplana-toryvariables, j is atechnology-specificterm, and ij is theresidual error term.We model thetechnology-specific

term, j, using atechnology-specificdummy variable,which takes the valueof unity if theinvention is intechnology areaj, andzero otherwise.23

One of the mostdifficult aspects ofour study relates toseparating the valueof patent protec-tionfrom the underlyingvalue of thetechnology. If


11/21

inventions that are granted a patentgenerally have higher commercialvalue, our estimates of

22Although we treat this variable as a simplebinary variable, we acknowledge that the claimsin a granted patent may be different from those inthe original patent application. Unfortunately,changes in the claims as a result of interactionbetween the applicant and the patent examiner areunobserved. We thank an anonymous referee forpointing out this issue.

23 We also attempted a second

specification, the random-intercept model, in which

jis assumed to be a random variable and follows an

identical and independent

normal distribution with

mean zero and variance 2.

The results are similar and are

omitted for the sake of

brevity.

Str

at.

Mg

mt.

J.,

32:

11

28

11

38

(20

11)

DOI: 10.1002/smj

the patent premium would be biased.However, it is not clear that commercialvalue and the patent examinationdecision are necessarily corre-lated. Inexamining a patent application, thepatent office considers whether theinvention is useful, novel andnonobvious. Thus, the patent exam-inerevaluates the inventions technologicalmerit, not its expected economic value.Evidence sug-gests that a largeproportion of patented inventions havevery low (or zero) commercial value(see Harhoff et al., 1999; Gambardellaet al., 2008). Conversely, there aremany examples of inventions withsubstantial commercial value that arenot cov-ered by a patent. However, it

remains a possibility that there are othercharacteristics of the under-lyingtechnology that determine the value ofthe invention and are associated with the

patentability requirements.24

Taking these factors into account, weproceed using a reduced-form approachthat includes addi-tional variables tocontrol for characteristics of thetechnology that may be correlated with

invention value.25 First, we include adummy variable, radi-cal invention,which relates to whether the inventorrated his or her invention as radical orincre-mental. Second, we construct avariable related to the number ofproducts and processes for which theinvention was used (denoted as no. ofuses). This variable is expected to bepositively correlated with value.Although the relationship is proba-blynoisy (since there are many single-useinven-tions that have high commercialvalue), it seems reasonable to expectthat value is an increasing function of

the generality of the invention, ceterisparibus.

Third, we include a variable, other

inventions used, to proxy for the

complexity of the technology area. This

variable is based on the survey question

that asked the inventor how many other

patents were used to develop the product.

The expected sign of this variable is

unclear since it could be the case that

complex technologies are more valu-able

(i.e., there is a positive association with

value) or it could be that transaction costs

associated

24One important feature of our empirical approach isthat our data only include inventions that, in the viewof the applicant, involve patentable subject matterand have the potential of passing the criteria ofnovelty and nonobviousness.

25 To investigate the possibility of a

selection effect, we also used a two-stage sample

selection model. We failed to reject the null

hypothesis of no sample selection (results are

available from the authors upon request).



with negotiating with other patentowners erode the value of the invention(i.e., there is a nega-tive association

with value). The net effect of these twoforces depends on who owns the otherpatents required to develop the product,which is unob-served in our dataset. Wealso include a dummy variable, PCTapplication, indicating whether theapplication was made through the PatentCooper-ation Treaty (PCT) to accountfor the fact that the invention may havepatent protection in other legal

jurisdictions, even if the patent was notgranted in Australia.


12/21

Attributes of the firm or individualwho owns the invention also affect therevenue that can be appro-priated. Firmsize is a key variable as the poten-tialprofit from an invention is directlyrelated to the resources of the firm.Teece (1986) highlights the importanceof complementary assets, such asmanufacturing capacity, in facilitatingthe appro-priation of innovation returns.Similarly, large ver-tically integratedfirms and conglomerates that operate inmultiple industrial sectors may be betterplaced to capture profits andinterindustry techno-logical spillovers(see Weder and Grubel, 1993). Werelied on information from severalsources to construct firm size dummy

variables.26 Our model includes threedummy variables reflecting large

company, small and medium enterprise(SME), and public sector researchorganization, with indi-vidual serving asthe reference group.A priori, we expectthat large firms and SMEs have greaterincentives and are better placed inappropriating returns than public sectorresearch organizations and individuals.

The dependent variable in our basemodel is the measure of invention value(denoted by InvVal ), whichincorporates information about pastsales revenue and license fees as well asthe forward-looking value (as discussedpreviously). As an alternative, weconsider historical sales revenue

26 From the patent office database, we observe

whether the patent is registered to an individual or anorganization. Public sector research organizationswere identified by searching for any organizationwith words such as institute, university, orresearch in its name. Of those not registered to anindividual or public sector research organization, wematched the name of the organization with aproprietary database maintained by IBISWorld, aprivate Australian market research and analysis firm.The IBISWorld database includes all of the largestfirms in Australia and we use this information toconstruct the firm size dummy variables. Admittedly,a better indicator of firm size would be based onturnover or employment. However, such data are notavailable and we note that this would not be possiblefor nonaffiliated individual inventors.

Strat. Mgmt. J.,32:

11281138 (2011)

DOI:10.1002/smj


13/21


Table 1. List of variables and summary statistics

Variable Explanation Mean Std. dev.

ln(InvVal) Private value of invention (in logarithms) 6.945 2.026No. of uses No. of products/processes for which invention was used 1.331 0.971Other inventions used No. of other inventions used in development 0.378 0.623No. inventors No. of inventors involved 1.549 1.127

Large company dummy, (=1 if invention owned by large company) 0.088 0.283SME dummy, (=1 if invention owned by SME) 0.351 0.478Public organization dummy, (=1 if invention owned by public organization) 0.059 0.235Individual dummy, (=1 if invention owned by nonaffiliated individual) 0.502 0.500Radical invention dummy, (=1 if invention was radical relative to state of art) 0.628 0.483PCT application dummy, (=1 if PCT application) 0.356 0.479

Granted dummy, (=1 if patent application granted) 0.727 0.446

as a dependent variable, which we denote as

SalesRev. The use of these two different

measures enables us to ascertain the

robustness of our results and the ad hoc

assumption that gross profit mar-gins are

uniformly the same (30 percent) across all

inventions. In particular, if profit margins

vary systematically with patent protection,

our use of a fixed proportion may downward

bias our estimate of the patent premium.

Each model also includes patent application

year dummies (equivalent to invention age)

to capture the effects of technol-ogy,

business cycles, and/or obsolescence. Table

1 lists the variables we used in the

estimation and some descriptive statistics.

The patent premium is captured by thecoef-ficient in Equation (1) and is

reported as the coefficient of the variablegrantedin Table 2. This dummy variableequals one for patent applica-tions thatwere granted, and zero if the applicationwas refused, withdrawn, lapsed, orrevoked. As such, the variable takes abroad definition of non-grants in that itincludes all patent applications that wereeither unsuccessful or were removed fromthe examination process by the patentoffice (or the applicant). The key point isthat all the inventions categorized asgranted have patent protection and that

all inventions categorized as non-grantsdo not. The coefficient of the variablegranted tells us whether the protectionoffered by apatent increases (and by howmuch) the inventors returns, ceteris

paribus. As the dependent vari-able is inlogarithms, reflects the proportionalincrease in invention value that isassociated with a patent grant.

We extend the base model by allowingthe effect of a patent grant to vary bytechnology area. Specifically, we

introduce interaction termsCopyright 2011 John Wiley & Sons, Ltd.

between the twovariables Gij andOST technologyarea dummy dj, sothat (1) becomes:

ln Vij=Gij+

j(Gijdj)+

X+j+ij,

j =1, . . . ,J 1

Thus the effectof a patent granton an invention intechnology area jis given by +j,which varies bytechnology area.

RESULTS

The estimationbased on Equation(1) is imple-mented using both

InvVal andSalesRev as thedependentvariable. Theresults are reportedin Table 2. Themain result using

InvValas the valuemeasure suggeststhat inventions thatare protected by apatent are 47percent morevaluable thaninventions withouta patent, ceteris

paribus.27 Incomparison, thesame regression

models imple-mented usingSalesRev as thevalue measureyield estimates ofthe patentpremium of 38percent.Interestingly, thepatent premiumestimates pro-duced using

SalesRev as the

dependent variablehave higherstandard errorsthan thecorrespond-ingestimates producedusing InvVal,suggesting that thehistorical revenue-based measure ofvalue is lessprecise.

The coefficients

of the other

explanatory vari-

ables are consistent

with a priori

expectations. The

variables radical

invention and PCT

application

27 We also estimated a

random-intercept model.

The results, availableupon request, are

remarkably consistent

with those from the

fixed-effects model andare omitted for the sake

of brevity.

Strat.

Mgmt.J.,32:

1128

1138

(2011)

DOI: 10.1002/smj


14/21


Table 2. Estimation results

Dep. variable: ln(InvVal) Dep. variable: ln(SalesRev)coefficient (std. error) coefficient (std. error)

No. of uses 0.4068 0.2471

(0.0460) (0.0548)Other inventions used 0.4931 0.2054

(0.0719) (0.0857)No. inventors in application 0.0789 0.0198


15/21

(0.0457) (0.0544)Radical invention 0.6591 0.3101

(0.0913) (0.1089)PCT application 0.5575 0.3805

(0.1075) (0.1281)Large company 0.3927 1.4052

(0.1752) (0.2088)SME 0.3818 0.9887

(0.1051) (0.1253)Public organization 0.0313 0.4597

(0.2314) (0.2758)Granted 0.4717 0.3790

(0.1047) (0.1248)Intercept 5.4650 4.5903

(0.5341) (0.6366)Adjusted R2 0.208 0.116Number observations 1,790Number tech. classes 28

Notes: (1) Included in all regression models are 19 year dummies, which denote the year in which the patent

application was lodged, and 27 OST technology classification dummies. (2) Significance levels: 10%; 5%;1%.

are both positively associated with value.In addi-tion, the results support the notion

that inventions that have many uses are

positively correlated with higher value,

and that inventions registered to pri-vate

firms (both SMEs and large firms) are

more valuable than those registered to

individuals or to public sector

organizations. It is noteworthy that the

effect of firm size (i.e., the large firm

dummy variable) becomes substantially

larger when we replace InvVal withSalesRevas the dependent variable. This

increase in the firm size effect per-haps

reflects the superior ability of large firms

in product development and

commercialization, and the tendency of

SMEs and individual inventors to rely on

licensing and outright transfer of patents

to appropriate returns.

We extend the base model by allowingthe coef-ficient of grant status to vary by

technology area. The estimation isimplemented using InvVal as thedependent variable, and the results arereported in Table 3. We perform astatistical test of whether the effect of a

patent grant varies by technol-ogy area;the joint test of all interaction terms equalzero gives an F test statistic of 1.3, which


has a p-value of

0.142.28 Thus wefind no evi-dence

to suggest that the

effect of a patent

grant is different

across technology

areas. This is

some-what

surprising given

the empirical

support in other

studies thatsuggests patents

are more valu-able

in pharmaceuticals

and chemicals than

other technology

areas. However,

we note that the

exist-ing literature

also shows some

inconsistencies in

this regard. Forinstance, analysis

by Schanker-man

(1998) finds that

the value of patent

pro-tection in

pharmaceuticals is

less than that in

other technological

fields.

Nonetheless, we

cau-tion that ourresult is likely

caused by the

small number of

observations

within each

technology area.29

28We also performed thesame regressions usingSalesRev as thedependent variable. Theresults, available uponrequest, are similar andare omitted for the sakeof brevity.

29 We alsoestimated the model

using six major OST

technology areas rather

than the 30 technology

area classifications. In

this specification, we

find strong evidence that

the value of a patent

grant varies across

technology areas.

However, this is a much

coarser (and

heterogeneous)

classification of

technology areas.

Strat.

Mgmt.

J.,32:

1128

1138

(2011)

DOI: 10.1002/smj


16/21


Table 3. Estimationresults: effect ofpatent grant, bytechnology area

Dep. variable: ln(InvVal)

No. of uses

Other inventions used

No. inventors in application

Radical invention

PCT application

Large company

SME

Public organization

Granted

G electricaldeviceselectricalengineering

G audiovisual technology

G telecommunications

G information technology

G optics

G analysis, measurement,

control

G medical engineering

G organic fine chemicals

G pharmaceuticals,

cosmetics

G biotechnology

G materials, metallurgy

G agriculture, food

G general processes

G surfaces, coatings

G material processing

G thermal techniques

G basic chemical processing,

petrol

Copyright 2011 John

Wiley & Sons, Ltd. Table 3.(Continued)

Dep. variable: ln(InvVal) Coefficient (std. error)

G environment, pollution(0.6714)0.3961

G mechanical tools(0.8517)0.8348

G engines, pumps, turbines(0.6658)0.2068

G mechanical elements(0.6094)0.8741

G handling, printing(0.6105)

1.0242

G agriculture/food(0.4668)

0.6946machinery

(0.4190)G transport 0.5291

G space technology,(0.4224)0.6602

weapons

(1.5009)G consumer goods & 0.1920

equipment(0.3736)

Intercept 5.1310

(0.5623)Adjusted R2 0.212Number observations 1,790Number technology areas 28

Notes: (1) Included

in the regression are

19 year dummies,

which denote the

year in which thepatent application

was lodged, and 27

OST technology

classification

dummies. (2) The

omitted reference

group for the

interaction terms is

G civil

engineering,

building, mining.

Another interaction

term G

macromolecularchemistry,

polymers is

omitted due to

singularity. (3)

Significance levels:

10%; 5%;

1%.

Theestimated

patent premium

implies that a


17/21

patent increasesthe value of themedian inven-tion by about

A$256,000.30

The results alsohave important

public policy

implications.Since approx-imately 2,000

patents aregranted each yearin Australia, ourresult suggests

that in 2005 thepatent system

provided animplicitsubsidy toinnovators of

about A$4billion. This isapprox-imatelyfive times

larger than thecombined

value of theA$425 million

contributed bythe govern-

ment throughtax incentives(IA, 2007) and

the A$420million viasubsidies,grants, and

procure-ment(ABS Cat.8104.0).

30Given by 0.47/(1

+0#47) (median

patented invention

value).

Strat.

Mgmt. J.,

32: 1128

1138 (2011)

DOI: 10.1002/smj

CONCLUSIONS

This paper sheds new

light on a fundamental

question frequently

asked, but never

definitively answered:

what is the increase in

private value due to the

presence of a patent? We

tackle this question using

a novel empiricalapproach that makes use

of data from a

comprehensive survey of

inventors who applied for

a patent in Australia

between 1986 and 2005.

We use the variation in

patent examina-tion

outcomes to identify the

patent premium. The

results provide strongand robust support for the

existence of a sizeable

patent premium. The

pres-ence of a patent

increases the returns to

an inven-tion by around

40 to 50 percent on

average, regard-less of

how we define value.

While the value of patent

protection is expected to

vary according to

idiosyncratic inventor

and marketcharacteristics

surrounding an invention,

the average effectiveness

reported in this paper

provides a valuable

bench-mark to

technology managers and

IP professionals. This

result should aid firms in

their decisions on the

different appropriation

mechanisms to use as

part of their IP

management strategy.

Further, we also provide

new evidence on the

effectiveness of patenting

by technology area.

We acknowledge that

there are someimportant limitations ofour attempt to estimate

the patent premium.Foremost among theseis that invention value isdifficult to measure. Wehave relied on an

inventor survey to dothis, but we are cog-nizant of the fact thatself-reported

evaluations of inventionvalue can be

problematic. However,


18/21

in the absence of aperfectly functioningmarket for technology(which may never

exist), we have madesome valuable inroadsinto understanding howinventor surveys can be

used to tackle theidentification of the

patent premium. Wealso note that some

recent research hasestablished a strong

statistical relationshipbetween survey-basedmea-sures of inventionvalue and other

measures of valuederived from citationand renewal analysis

(Gambardella et al.,2008). Second, it is

difficult to disentangleinvention quality from

patent value. Ratherthan relying on self-reported proxies ofinvention quality (as we

have done), it may bebetter to rely on

independent evaluationsof the inventions as in

Moser (2007), althoughsuch data are difficult tocome by. Futureresearch on this issue

might consideralternative ways inwhich the underlyingtechnological

characteristics may be



measured. Third, we do

not observe firms

willing-ness to enforce

their patents, which is an

important determinant of

the private returns to

patenting (see Lanjouw1998; Lanjouw and

Schankerman 2001).

ACKNOWLEDGME

NTS

We thank AssociateEditor Tomi Laamanen,two anonymous

referees, Beth Webster,Jim Bessen, AlfonsoGambardella, Georgvon Graevenitz, DirkCzarnitzki, and GeorgLicht for thoughtfulcom-ments on thispaper. Comments andsuggestions fromseminar participants atLudwig MaximiliansUniversity (Munich),

ZEW Mannheim,MERIT (MaastrichtUniversity), and theKatholieke Uni-versityof Leuven are gratefullyacknowledged. Fundingfor this research comesfrom an AustralianResearch Council(ARC) Linkage Grantwith IP Australia(LP0667467).

REFERENCES

Arora A, Ceccagnoli M.2006. Patent protection,comple-mentary assetsand firms incentives fortechnology licensing.

Management Science52:

293308.Arora A, Ceccagnoli M,

Cohen W. 2008. R&Dand the patent premium.


19/21

International Journal ofIndustrial Organization26: 11531179.

Bessen JE. 2009. Estimatesof patent rents from firmmarket value. ResearchPolicy38: 16041616.

Bosworth D, Rogers M.2001. Market value,R&D and intellectual

property: an empiricalanalysis of largeAustralian firms.

Economic Record

77(239): 323337.

Ceccagnoli M. 2009.Appropriability,preemption, and firmperformance. Strategic

Management Journal

30(1): 8198.

Cohen WM, Nelson RR,Walsh JP. 2000.Protecting their

intellectual assets:appropriabilityconditions and why USmanufacturing firmspatent (or not). NBERWorking paper 7552.National Bureau ofEconomic Research:Cambridge, MA.

Deng Y. 2007. Privatevalue of European patents.

European Economic

Review 51: 17851812.Gambardella A, Harhoff D,

Verspagen B. 2008. Thevalue of Europeanpatents.European

ManagementReview 5:6984.

Giuri P, Mariani M, Brusoni

S, Crespi G, Francoz D,Gambardella A, Garcia-Fontes W, Geuna A,Gonza-les R, Harhoff D,Hoisl K, Le Bas C, LuzziA, Mag-azzini L, NestaL, Nomaler O,Palomeras N, Patel P,Romanelli M, VerspagenB. 2005. Everything you

always wanted to knowabout inventors (butnever asked): evidencefrom the PatVal-EUSurvey. LEM Workingpaper 2005/20.Laboratory of Economics andManagement: Pisa, Italy.

S

t

r

a

t.

Mg

m

t.

J

.,

3

2

:

1

1

2

8

1

1

3

8(

2

0

1

1

)

DOI: 10.1002/smj


20/21


Griliches Z. 1981.Market value,R&D, and patents.

Economics

Letters 7(2):183187.

Gronqvist C. 2009.

The privatevalue of patentsby patentcharacteristics:evidence fromFinland. Journalof Technology

Transfer 34:159168.

Hall BH, Jaffe A,Trajtenberg M.2005. Marketvalue and patentcitations. RAND

Journal ofEconomics 35:1638.

Hall BH,MacGarvie M.2006. Theprivate value ofsoftware patents.NBER Workingpaper W12195.National Bureauof EconomicResearch:Cambridge, MA.

Harabi N. 1995.

Appropriabilityof technicalinnovations: anempiricalanalysis.

Research Policy

24: 981992.

Harhoff D, Narin F,Scherer FM,Vopel K. 1999.Citationfrequency andthe value of patented

inventions.Review ofEconomics andStatistics 81:511515.

Horstmann I,MacDonald GM,Slivinski A.1985. Patents asinformationtransfermechanisms: topatent or(maybe) not topatent. Journal

of PoliticalEconomy

93: 167182.

IA. 2007.InnovationAustraliaAnnualReport 2006-07 .

InnovationAustralia:Canberra.

Lanjouw JO.1998. Patentprotection inthe shadowofinfringement:simulationestimationsof patentvalue.

Review of

EconomicStudies 65:671710.

Lanjouw JO,SchankermanM. 2001.Characteristics of patentlitigation: awindow oncompetition.

RANDJournal ofEconomics

32: 129151.Levin RC,

KlevorickAK, NelsonRR, WinterSG. 1987.Appropriating the returnsfromindustrialresearch anddevelopment.

BrookingsPapers on

Economic

Activity3: 783820.

Mansfield E.1986. Patentsandinnovation:an empiricalstudy.

ManagementScience 32:173181.

Mansfield E,Schwartz M,Wagner S.

1981.Imitationcosts andpatents: anempiricalstudy.

Economic

Journal 91:907918.

Moser P. 2007.Why dontinventorspatent?Mimeo,12November,StanfordUniversity,Stanford,CA.


21/21

OECD. 2008a. OECDMain Science and

Technology

Indicators 2008 .Organisation forEconomic Co-operation andDevelopment: Paris,France.

OECD. 2008b. OECD

Science, Technologyand IndustryOutlook 2008 .Organisation forEconomic Co-operation andDevelopment: Paris,France.

Pakes A. 1986. Patentsas options: someestimates of thevalue of holdingEuropean patentstocks.

Econometrica 54:755784.

Sakakibara M,Branstetter L. 2001.Do stronger patentsinduce moreinnovation?Evidence from the1988 Japanesepatent law reforms.

RAND Journal of

Economics 32: 77100.

Sampat BN, Ziedonis

AA. 2004. Patentcitations and theeconomic value ofpatents: apreliminaryassessment. In The

Handbook ofQuantitativeScience andTechnology,Moed H,Glanzel W,Schmoch U(eds). Springer:the Netherlands;277298.

Schankerman M.1998. Howvaluable ispatent protec-tion? Estimatesby technologyfield. RAND

Journal of

Economics 29:77107.

Schankerman M,Pakes A. 1986.Estimates of thevalue of patentrights inEuropeancountries duringthe post-1950period.

EconomicJournal 96:10521076.

Serrano CJ. 2006.The market forintellectualproperty:evidence fromthe transfer ofpatents. Mimeo,

July. Universityof Toronto:Toronto,Ontario, Canada.

Sneed KA, JohnsonDKN. 2009.

Selling ideas:the deter-minants of patent value inan auctionenvironment.

R&D

Management

39: 8794.

Teece D. 1986.

Profiting fromtechnologicalinnovation:implications forintegration,collaboration,licensing andpublic policy.

Research

Policy15: 285305.

Webster E, JensenPH. 2009. Dopatents matter

for com-mercialization?IPRIA Workingpaper 03/09.Univer-sity ofMelbourne:Parkville,Victoria,Australia.

Weder R, GrubelHG. 1993. Thenew growththeory andCoaseaneconomics:

institutions tocaptureexternali-ties.Weltwirtschaftliches Archiv

129: 488513.

Tica Susi PxhjVhvHXVHvvxHVx

Documents