《产业经济学》课程教学资源（文献资料）Internal and external factors in innovationpersistence知识累积

RoutledgeTaylorFrancisGroupEconomics of InnovationandNewTechnologyAOATechnolISSN:1043-8599(Print)1476-8364(Online)Journalhomepage:http://tandfonline.com/loi/gein20InternalandexternalfactorsininnovationpersistenceCristianoAntonelli,FrancescoCrespi&GiuseppeScellatoTo cite this article: Cristiano Antonelli,Francesco Crespi &Giuseppe Scellato (2013) Internaland external factors in innovation persistence,Economics of Innovation and New Technology,22:3,256-280,DOl:10.1080/10438599.2012.708135To link to this article: http://dx.doi.org/10.1080/10438599.2012.708135Published online: 02 Aug 2012.中8Submit your article to this journal C[.lil Article views: 603View related articles QCiting articles: 16 View citing articles FullTerms&Conditions ofaccess and usecanbefoundathttp:/tandfonline.com/action/journallnformation?journalCode=gein20Download by: [Wilfrid Laurier University]Date: 02 August 2016, At: 00:11

Full Terms & Conditions of access and use can be found at http://tandfonline.com/action/journalInformation?journalCode=gein20 Download by: [Wilfrid Laurier University] Date: 02 August 2016, At: 00:11 Economics of Innovation and New Technology ISSN: 1043-8599 (Print) 1476-8364 (Online) Journal homepage: http://tandfonline.com/loi/gein20 Internal and external factors in innovation persistence Cristiano Antonelli , Francesco Crespi & Giuseppe Scellato To cite this article: Cristiano Antonelli , Francesco Crespi & Giuseppe Scellato (2013) Internal and external factors in innovation persistence, Economics of Innovation and New Technology, 22:3, 256-280, DOI: 10.1080/10438599.2012.708135 To link to this article: http://dx.doi.org/10.1080/10438599.2012.708135 Published online: 02 Aug 2012. Submit your article to this journal Article views: 603 View related articles Citing articles: 16 View citing articles

EconomicsofInnovationandNewTechnology,2013RoutledgeVol.22,No.3,256280,http://dx.doi.org/10.1080/10438599.2012.708135InternalandexternalfactorsininnovationpersistenceCristiano Antonelliab, Francesco Crespib.c* and Giuseppe ScellatobdaDipartimento di Economia,Universita di Torino,Torino,Italy,bBRICK (BureauofResearch onInnovation,ComplexityandKnowledge),CollegioCarloAlberto,Torino,Italy;DipartimentodiEconomia,Universita Roma Tre,Rome,ItalyDipartimentodi Ingegneria Gestionale e dellaProduzione,Politecnico di Torino,Torino, Italy(Received14September2011;finalversionreceived25June2012)This paper contributes to the analysis of the persistence of innovation activities, asmeasured bytotal factor productivity (TFP),and explores its internal andextemal deter-minants stressing its path-dependent characteristics.The external conditions,namely thequality of local knowledge pools and the strength of the Schumpeterian rivalry, alongwith the internal conditions (the actual levels of dynamic capabilities, as proxied bywagelevels and firm size)exert a specific and localised effect upon the persistent intro-duction of innovations.Amultiple transition probability matrixes (MTPMs)approachhasbeen implemented tocapturethecontingenteffectsofexternal factorson long-terminnovation persistence.The empirical analysis of the dynamics of firm-level TFP forasampleofapproximately7000Italianmanufacturingcompaniesobservedduringtheyears1996-2005isbasedonboththe comparison of differenttransitionprobabilitymatrixes and on dynamic discrete choice panel data models.The evidence provided bythe test ofMTPMs in sub-periods suggests that innovation persistence is path-dependent,as opposedtopast-dependent.Keywords: knowledge cumulability;knowledge extermalities; innovation; persistence;sequential Markov chains; path dependence; TFPJELClassification:O31C23;C25;L201.IntroductionAccording to the conventional economic wisdom, innovation is an exogenous randomshock.The economics of innovationimpingesupontheviewthat innovation isthedeliberate and intentional result of the ability of firms togenerate newknowledge and to applyit to new products,processes, organisational methods, combinations of inputs and newmarkets (Nelson and Winter 1982; Dosi et al. 1988; Fagerberg, Mowery, and Nelson2005).This approach leads to two quite distinct explanations of innovation persistence. Thefirst, consistent with the resource-based theory of the firm, contends that innovation per-sistence is the result of intrinsic characteristics of the firm. Innovation capabilities are*Corresponding author.Email:crespi@uniroma3.it2013Taylor&Francis

Economics of Innovation and New Technology, 2013 Vol. 22, No. 3, 256–280, http://dx.doi.org/10.1080/10438599.2012.708135 Internal and external factors in innovation persistence Cristiano Antonellia,b, Francesco Crespib,c* and Giuseppe Scellatob,d aDipartimento di Economia, Università di Torino, Torino, Italy; bBRICK (Bureau of Research on Innovation, Complexity and Knowledge), Collegio Carlo Alberto, Torino, Italy; cDipartimento di Economia, Università Roma Tre, Rome, Italy; dDipartimento di Ingegneria Gestionale e della Produzione, Politecnico di Torino, Torino, Italy (Received 14 September 2011; final version received 25 June 2012) This paper contributes to the analysis of the persistence of innovation activities, as measured by total factor productivity (TFP), and explores its internal and external deter￾minants stressing its path-dependent characteristics. The external conditions, namely the quality of local knowledge pools and the strength of the Schumpeterian rivalry, along with the internal conditions (the actual levels of dynamic capabilities, as proxied by wage levels and firm size) exert a specific and localised effect upon the persistent intro￾duction of innovations. A multiple transition probability matrixes (MTPMs) approach has been implemented to capture the contingent effects of external factors on long-term innovation persistence. The empirical analysis of the dynamics of firm-level TFP for a sample of approximately 7000 Italian manufacturing companies observed during the years 1996–2005 is based on both the comparison of different transition probability matrixes and on dynamic discrete choice panel data models. The evidence provided by the test of MTPMs in sub-periods suggests that innovation persistence is path-dependent, as opposed to past-dependent. Keywords: knowledge cumulability; knowledge externalities; innovation; persistence; sequential Markov chains; path dependence; TFP JEL Classification: O31; C23; C25; L20 1. Introduction According to the conventional economic wisdom, innovation is an exogenous random shock. The economics of innovation impinges upon the view that innovation is the delib￾erate and intentional result of the ability of firms to generate new knowledge and to apply it to new products, processes, organisational methods, combinations of inputs and new markets (Nelson and Winter 1982; Dosi et al. 1988; Fagerberg, Mowery, and Nelson 2005). This approach leads to two quite distinct explanations of innovation persistence. The first, consistent with the resource-based theory of the firm, contends that innovation per￾sistence is the result of intrinsic characteristics of the firm. Innovation capabilities are *Corresponding author. Email: crespi@uniroma3.it © 2013 Taylor & Francis Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016

257EconomicsofInnovationandNewTechnologytime-invariant endowments that display their effects.Innovation persistence isfully drivenby the initial allocation of innovation capabilities: firms are able to learn (Penrose 1959;Stiglitz1987;TeeceandPisano1994;LangloisandFoss1999).The second explanation posits that innovation persistence is a path-dependent processin which the probability of introducing an innovation at time t is influenced by the intro-duction of an innovation at time t -1. However, the transition probability might changeovertimebecause of the effects of contingent events and specifically because of changinglevels of knowledge externalities.Thegeneration of newknowledge and the introductionof innovations are the conditional results of a creative and localised reaction that occurswhen firms faceunexpected events inbothfactor and product markets.Some contextualand ever-changing conditions,however,arenecessarytomakethereaction creativesothatit leads to the introduction of an innovation, as opposed to an adaptation. In the latter case,the lack of contextual characteristics would enable firms to change techniques in a giveneetechnical space but would not lead to the introduction of a productivity-enhancing novelty(Schumpeter1947)To contend with unexpected events in factor and productmarkets and the subsequentout-of-equilibrium conditions, firms try and mobilise their internal stocks of knowledgethroughlearningprocesses.Theprobability that thefirm's reaction leads to the successfulintroductionofan innovationdependsonaccesstoavailableexternalknowledge.In otherwords, the firm's reaction to unexpected events becomes creative when the competenceaccumulated through internal learning processes and access to external knowledge poolscombine(Antonelli2008,2011)According tothis view,proximity,knowledge governanceand the communicationchannels that link firms might enhance the process of knowledge generation, favouringinteractions among agents with diverse knowledge bases. Indeed, firms cluster mainly forthese specific reasons(Swann, Prevezer, and Stout 1998; Baptista and Swann 1999).Long-distance coordination among agents andknowledge interactions can also berealised throughorganised proximity (Torre and Rallet 2005). In this context, knowledge governance mech-anisms and the characteristics of knowledge structureare particularly relevant (Quatraro2012)Beginning with the seminal contribution by Griliches (1979), a rich and detailed arrayof empirical studies confirm the pervasive role of technological spillover in favouring theeconomicperformances ofclusteredfirms in terms ofoutput,employment, labour produc-tivity and total factor productivity (TFP).The subsequent literature has interpreted theseempiricalfindings as reliableclues to assessing the positive effects ofknowledge externali-tiesonthe rate of introduction oftechnologicalchangesbyfirms thatare ableto use externalknowledge as an input in their own innovation processes (Acs, Anselin, and Varga 2002;Fritsch 2002, 2004; Fritsch and Franke 2004).Building on this literature, we advance the hypothesis that innovation persistence ispath-dependent, as opposed to past-dependent, because it is the result of not only theinternal characteristics offirms,astheresource-based theory ofthefirm claims,butalsothechangingcharacteristicsofthecontextinwhichfirmsarelocated.Knowledgeexternalitiesarestrictlynecessaryforfirmsreactionstobecomecreative.Whenandifthecharacteristicsof the context change,the results of the innovative effortalso change.Hence,innovationpersistence can no longerbe regarded as theresult of an intrinsic capability ofthefirm thatbehaves as an endowment,given once and lasting forever; rather, it should be regarded asthe conditional resultofa systemic and interactiveprocessthatkeeps changingovertime(Antonelli and Scellato, forthcoming)

Economics of Innovation and New Technology 257 time-invariant endowments that display their effects. Innovation persistence is fully driven by the initial allocation of innovation capabilities: firms are able to learn (Penrose 1959; Stiglitz 1987; Teece and Pisano 1994; Langlois and Foss 1999). The second explanation posits that innovation persistence is a path-dependent process in which the probability of introducing an innovation at time t is influenced by the intro￾duction of an innovation at time t − 1. However, the transition probability might change over time because of the effects of contingent events and specifically because of changing levels of knowledge externalities. The generation of new knowledge and the introduction of innovations are the conditional results of a creative and localised reaction that occurs when firms face unexpected events in both factor and product markets. Some contextual and ever-changing conditions, however, are necessary to make the reaction creative so that it leads to the introduction of an innovation, as opposed to an adaptation. In the latter case, the lack of contextual characteristics would enable firms to change techniques in a given technical space but would not lead to the introduction of a productivity-enhancing novelty (Schumpeter 1947). To contend with unexpected events in factor and product markets and the subsequent out-of-equilibrium conditions, firms try and mobilise their internal stocks of knowledge through learning processes. The probability that the firm’s reaction leads to the successful introduction of an innovation depends on access to available external knowledge. In other words, the firm’s reaction to unexpected events becomes creative when the competence accumulated through internal learning processes and access to external knowledge pools combine (Antonelli 2008, 2011). According to this view, proximity, knowledge governance and the communication channels that link firms might enhance the process of knowledge generation, favouring interactions among agents with diverse knowledge bases. Indeed, firms cluster mainly for these specific reasons (Swann, Prevezer, and Stout 1998; Baptista and Swann 1999). Long￾distance coordination among agents and knowledge interactions can also be realised through organised proximity (Torre and Rallet 2005). In this context, knowledge governance mech￾anisms and the characteristics of knowledge structure are particularly relevant (Quatraro 2012). Beginning with the seminal contribution by Griliches (1979), a rich and detailed array of empirical studies confirm the pervasive role of technological spillover in favouring the economic performances of clustered firms in terms of output, employment, labour produc￾tivity and total factor productivity (TFP). The subsequent literature has interpreted these empirical findings as reliable clues to assessing the positive effects of knowledge externali￾ties on the rate of introduction of technological changes by firms that are able to use external knowledge as an input in their own innovation processes (Acs, Anselin, and Varga 2002; Fritsch 2002, 2004; Fritsch and Franke 2004). Building on this literature, we advance the hypothesis that innovation persistence is path-dependent, as opposed to past-dependent, because it is the result of not only the internal characteristics of firms, as the resource-based theory of the firm claims, but also the changing characteristics of the context in which firms are located. Knowledge externalities are strictly necessary for firms’ reactions to become creative. When and if the characteristics of the context change, the results of the innovative effort also change. Hence, innovation persistence can no longer be regarded as the result of an intrinsic capability of the firm that behaves as an endowment, given once and lasting forever; rather, it should be regarded as the conditional result of a systemic and interactive process that keeps changing over time (Antonelli and Scellato, forthcoming). Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016

258C.Antonellietal.Thepresent paperbuilds on therecognition thatexternal technologicalknowledge repre-sentsan augmentingandfacilitatingfactorintheintroductionoftechnological innovationsand extends this concept by arguing that external knowledge is a keyfactor in determiningapath-dependentinnovationpersistencecharacterisedbycontextual andconditionalrecur-sive feedbacks. The paper elaborates the hypothesis that the introduction of innovations isthe persistent,emerging property ofan economic system characterised by knowledge cumu-lability and complementarity,both inside and outside offirms.Indeed,the introduction ofnewtechnologies and new organisational methods affects the system in two ways: it engen-ders further series of unexpected events and Schumpeterian rivalry and makes availablenewknowledge spillovers that add to the existing stock of external knowledge.Knowledge cumulability consists of the inter-temporal, diachronic indivisibility ofknowledge.Itis well-known that the arrovian economics ofknowledge assumes that knowl-edge is characterised by indivisibility and non-exhaustibility.Knowledge vintages add onand build up a stock of knowledge that is not exhausted because of repeated use.Indivisibilitymanifests in cumulability and complementarity among the different unitsofknowledge.Inparallel totheunits of knowledgethatareinternallypossessedby eachagent,external units of knowledge possessedby other agents also play a central role.Thegeneration of newknowledge is possibleonlyby‘standing on the shoulders of giants',thatis, through access to and use of the existing stock ofknowledge.The existing knowledgebase, however, is located both inside each firm and in the other agents that belong to thesamesystem (ColombelliandvonTunzelmann2011).As Peter Swann has convincingly shown,the structure of the system changes endoge-nouslybecauseofthe changing modes of interaction amongfirms,their entry and exitand their growth.The introduction of innovations is itself a major factor of change in thearchitecture of the system.The external conditions in which firms are embedded are simul-taneouslya consequenceanda cause oftherecursivefeedback that supports thepersistenceofinnovation activities (Swann, Prevezer,and Stout1998;Baptista and Swann 1998,1999;Beaudryand Swann2009)Internalandexternalknowledgecumulabilityaffectthedynamicsofeconomicprocessesbecause the knowledge base that each firm can access and use internally and externallyshapes the probability of the generation of new knowledge. Such effects can change overtime because the rates ofaccumulation and the conditions of access are notfixed. Inventionsand scientific breakthroughs can make some portions of the stock ofknowledge obsolete.Changes in the structure of interactions and transactions can modify access to externalknowledge.As such, the effects of internal and external knowledge cumulability are typi-cally path-dependent ratherthan past-dependent. In theformer case, the effects ofhysteresisare qualified and shaped bythe contingentchanges that occur in theprocess.In thelattercase, the process is shaped by the initial conditions only.The dynamics of the process areinfluenced by a weak irreversibility that allows changes in the process to alter both therate and the direction of the dynamics with typical path-dependent effects (David 1997,2007).With this approach in the background, the aim of this work is threefold. First, we con-tribute to the literature on the persistence of innovation by providing an empirical analysisbased onTFP measures.Second,wequalifythecharacteristicsof thispersistenceandexplore its external determinants by specifically examining the role of regional contextand the characteristics of the product markets in shaping this process. Third, we discuss indetail the methodological and theoretical implications of theuse of TPMs with reference toMarkovchains theory

258 C. Antonelli et al. The present paper builds on the recognition that external technological knowledge repre￾sents an augmenting and facilitating factor in the introduction of technological innovations and extends this concept by arguing that external knowledge is a key factor in determining a path-dependent innovation persistence characterised by contextual and conditional recur￾sive feedbacks. The paper elaborates the hypothesis that the introduction of innovations is the persistent, emerging property of an economic system characterised by knowledge cumu￾lability and complementarity, both inside and outside of firms. Indeed, the introduction of new technologies and new organisational methods affects the system in two ways: it engen￾ders further series of unexpected events and Schumpeterian rivalry and makes available new knowledge spillovers that add to the existing stock of external knowledge. Knowledge cumulability consists of the inter-temporal, diachronic indivisibility of knowledge. It is well-known that the arrovian economics of knowledge assumes that knowl￾edge is characterised by indivisibility and non-exhaustibility. Knowledge vintages add on and build up a stock of knowledge that is not exhausted because of repeated use. Indivisibility manifests in cumulability and complementarity among the different units of knowledge. In parallel to the units of knowledge that are internally possessed by each agent, external units of knowledge possessed by other agents also play a central role. The generation of new knowledge is possible only by ‘standing on the shoulders of giants’, that is, through access to and use of the existing stock of knowledge. The existing knowledge base, however, is located both inside each firm and in the other agents that belong to the same system (Colombelli and von Tunzelmann 2011). As Peter Swann has convincingly shown, the structure of the system changes endoge￾nously because of the changing modes of interaction among firms, their entry and exit and their growth. The introduction of innovations is itself a major factor of change in the architecture of the system. The external conditions in which firms are embedded are simul￾taneously a consequence and a cause of the recursive feedback that supports the persistence of innovation activities (Swann, Prevezer, and Stout 1998; Baptista and Swann 1998, 1999; Beaudry and Swann 2009). Internal and external knowledge cumulability affect the dynamics of economic processes because the knowledge base that each firm can access and use internally and externally shapes the probability of the generation of new knowledge. Such effects can change over time because the rates of accumulation and the conditions of access are not fixed. Inventions and scientific breakthroughs can make some portions of the stock of knowledge obsolete. Changes in the structure of interactions and transactions can modify access to external knowledge. As such, the effects of internal and external knowledge cumulability are typi￾cally path-dependent rather than past-dependent. In the former case, the effects of hysteresis are qualified and shaped by the contingent changes that occur in the process. In the latter case, the process is shaped by the initial conditions only. The dynamics of the process are influenced by a weak irreversibility that allows changes in the process to alter both the rate and the direction of the dynamics with typical path-dependent effects (David 1997, 2007). With this approach in the background, the aim of this work is threefold. First, we con￾tribute to the literature on the persistence of innovation by providing an empirical analysis based on TFP measures. Second, we qualify the characteristics of this persistence and explore its external determinants by specifically examining the role of regional context and the characteristics of the product markets in shaping this process. Third, we discuss in detail the methodological and theoretical implications of the use of TPMs with reference to Markov chains theory. Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016

259Economics of Innovation and NewTechnologyThe remainder of the paper is structured as follows.Section 2 reviews the literature onthis matter. Section 3 outlines the hypotheses and the research design of this study. Section4presentstheeconometricevidence.Theconclusionsummarisesthemainresults2.Priorresearchon innovationpersistenceIn a special issueof theInternational Journal of Industrial Organizationdedicated totheeconomicsofpathdependence,Malerba,Orsenigo,and Petretto(1997)pavethewayto theanalysis of the persistence of innovation activities now explored by a growing literature,which is synthesised in Table 1.Earlier studies can be grouped into two subsets: those that build on the analysis oflargesamples of patents and empirical studies that use data from innovation surveys.The persis-tence of innovation has been addressed by studying various factors,such as technologicalspecialisation(Malerba,Orsenigo,andPetretto1997),cross-countryandcross-sectorevo-lution (Cefis and Orsenigo 2001; Raymond et al.2010; Clausen et al.2012), the empiricalproperties of thedistribution of persistence(Cefis 2003)and thediversetypologiesof inno-vative activities (Roper and Hewitt-Dundas 2008; Martinez-Ros and Labeaga 2009; Peters2009;LeBas,Mothe,andNguyen2011;Antonelli,Crespi,and Scellato2012)Some convergentconclusions appeartohavebeen reached byprevious studies,althoughtheyhaveemergedfromdifferentcontexts.Inparticular,bothinnovatorsandnon-innovatorsshowed a strong tendencyto remain within their states.The evidence shows that the degreeof persistence varies according to the innovation indicator adopted (Duguet and Monjon2004).While the works that have used patents as indicators suggest that persistence isweak, exhibiting strong values only in the case oftop patentees,empirical analyses basedon surveydatafound strongerevidence of innovation persistence.Several factorshavebeen associatedwiththepresenceofpersistenceininnovativeactivities.Among thesefactors,firm size,profitabilityand theintensityofR&D activitieswere shown to be crucial, confirming the hypothesis that theaccumulation of knowledgeovertimetendstoinduceastatedependencein innovativebehaviourandthattheavailabilityof internal funds enhances the ability to continuously engage in innovation (Cefis andCiccarelli 2005; Lathamand Le Bas2006; Peters2009).The evidence suggesting that R&D-based innovation activities tend to be associatedwith higher persistence appears to be of particular importance because it helps to explaintwo important results emerging from the previous literature.First, several contributionshighlighted that innovation persistence is stronger in high-tech, science-based industrieswhereR&Dactivitiesareconcentrated (Raymond etal.2010;Clausen etal.2012).Sec-ond,when different innovation output indicators havebeen considered,product innovationwhich is very often linked to R&D investments (Crespi and Pianta 2007),tends to becharacterised by higher persistence than process innovation (Martinez-Ros and Labeaga2009; Antonelli,Crespi, and Scellato 2012).In thisrespect,the complementarities amongdifferent types of innovation activities emerged as crucial in shaping different patterns ofpersistence (Clausen et al. 2012; Antonelli, Crespi, and Scellato 2012), including the caseoforganisational innovation.In thereviewedstudies,attention hasbeenpaidprimarilytointernalfactorsthatconsiderpersistencetobetheresultoffirmcharacteristics,whiletheroleofexternalknowledgeandlocal context in shaping innovation persistence is almost totally neglected. In this respect,our paper adds totheprevious literaturebecause itis thefirst attemptto consider exter-nal factors in determining innovationpersistence.Building onprevious analyses showingthat successful innovative activity ismore likelyto occur within strong industrial regions

Economics of Innovation and New Technology 259 The remainder of the paper is structured as follows. Section 2 reviews the literature on this matter. Section 3 outlines the hypotheses and the research design of this study. Section 4 presents the econometric evidence. The conclusion summarises the main results. 2. Prior research on innovation persistence In a special issue of the International Journal of Industrial Organization dedicated to the economics of path dependence, Malerba, Orsenigo, and Petretto (1997) pave the way to the analysis of the persistence of innovation activities now explored by a growing literature, which is synthesised in Table 1. Earlier studies can be grouped into two subsets: those that build on the analysis of large samples of patents and empirical studies that use data from innovation surveys. The persis￾tence of innovation has been addressed by studying various factors, such as technological specialisation (Malerba, Orsenigo, and Petretto 1997), cross-country and cross-sector evo￾lution (Cefis and Orsenigo 2001; Raymond et al. 2010; Clausen et al. 2012), the empirical properties of the distribution of persistence (Cefis 2003) and the diverse typologies of inno￾vative activities (Roper and Hewitt-Dundas 2008; Martínez-Ros and Labeaga 2009; Peters 2009; Le Bas, Mothe, and Nguyen 2011; Antonelli, Crespi, and Scellato 2012). Some convergent conclusions appear to have been reached by previous studies, although they have emerged from different contexts. In particular, both innovators and non-innovators showed a strong tendency to remain within their states. The evidence shows that the degree of persistence varies according to the innovation indicator adopted (Duguet and Monjon 2004). While the works that have used patents as indicators suggest that persistence is weak, exhibiting strong values only in the case of top patentees, empirical analyses based on survey data found stronger evidence of innovation persistence. Several factors have been associated with the presence of persistence in innovative activities. Among these factors, firm size, profitability and the intensity of R&D activities were shown to be crucial, confirming the hypothesis that the accumulation of knowledge over time tends to induce a state dependence in innovative behaviour and that the availability of internal funds enhances the ability to continuously engage in innovation (Cefis and Ciccarelli 2005; Latham and Le Bas 2006; Peters 2009). The evidence suggesting that R&D-based innovation activities tend to be associated with higher persistence appears to be of particular importance because it helps to explain two important results emerging from the previous literature. First, several contributions highlighted that innovation persistence is stronger in high-tech, science-based industries where R&D activities are concentrated (Raymond et al. 2010; Clausen et al. 2012). Sec￾ond, when different innovation output indicators have been considered, product innovation, which is very often linked to R&D investments (Crespi and Pianta 2007), tends to be characterised by higher persistence than process innovation (Martínez-Ros and Labeaga 2009; Antonelli, Crespi, and Scellato 2012). In this respect, the complementarities among different types of innovation activities emerged as crucial in shaping different patterns of persistence (Clausen et al. 2012; Antonelli, Crespi, and Scellato 2012), including the case of organisational innovation. In the reviewed studies, attention has been paid primarily to internal factors that consider persistence to be the result of firm characteristics, while the role of external knowledge and local context in shaping innovation persistence is almost totally neglected. In this respect, our paper adds to the previous literature because it is the first attempt to consider exter￾nal factors in determining innovation persistence. Building on previous analyses showing that successful innovative activity is more likely to occur within strong industrial regions Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016

260C. Antonelli et al.Table 1.Summaryof themajorcontributionstothefield of innovationpersistenceDataResultsAuthorsMethodologyPatent data analysesMalerba, Orsenigo, andPatent data fromDynamic panelThe econometricPetretto (1997)OTAF-SPRUdata modelevidence shows thatdata base forinnovative activityfive EU countriesis persistent(19691986)Gerosky, Van Reenen,Patent records andProportionalOnly a minorityand Walters (1997)"major"innovationshazardof firms (majorofa sampleof UKfunctioninnovators)firms (19691988)are found tobe persistentlyinnovativee[eeeTPMCefis and OrsenigoPatent data on aEvidence of weaksampleof 1400(2001)persistence; bothmanufacturing firmslowinnovatorsand(19781993) ingreat innovatorsGermany, Italy,generally remain inJapan, US andtheir classesFranceData on 577 UKTPMCefis (2003)Evidence of littlepatenting firmspersistence(19781991)characterised bya strong thresholdeffect. Only greatinnovators haveastronger probabilitytokeep innovatingCefis and CiccarelliData on 267 UKBayesianThe study shows that(2005)patenting firmseconometriccurrent innovative(19881992)modelsactivity can bepositively influencedby past innovationvia the greateravailability offinancial resourcesInformation on 16,698Time seriesAifranca,Rama,andvonTheevidenceconfirmsTunzelmann(2002)patents granted inanalysisthat global firms inthe USA from 1977this industry exhibitto 1994 to 103 globala stable patternfirms in the food andof technologicalaccumulation inbeverage industrywhich'successbreeds success'.Latham and Le BasPatent data for 3347DurationThe persistence(2006)French firmseconometricof innovation is(19691985)modelstronger amongindividuals thanamong firmsHuang (2008)Patent and R&DDynamicEvidence supportingdata on 246random effectthe existence ofelectronics firmsprobit modelpersistent innovationafter controlling forlisted on theTaiwanStock Exchangefirm heterogeneity(19982003)(continued)

260 C. Antonelli et al. Table 1. Summary of the major contributions to the field of innovation persistence. Authors Data Methodology Results Patent data analyses Malerba, Orsenigo, and Petretto (1997) Patent data from OTAF-SPRU data base for five EU countries (1969–1986) Dynamic panel data model The econometric evidence shows that innovative activity is persistent Gerosky, Van Reenen, and Walters (1997) Patent records and ‘major’ innovations of a sample of UK firms (1969–1988) Proportional hazard function Only a minority of firms (major innovators) are found to be persistently innovative Cefis and Orsenigo (2001) Patent data on a sample of 1400 manufacturing firms (1978–1993) in Germany, Italy, Japan, US and France TPM Evidence of weak persistence; both low innovators and great innovators generally remain in their classes Cefis (2003) Data on 577 UK patenting firms (1978–1991) TPM Evidence of little persistence characterised by a strong threshold effect. Only great innovators have a stronger probability to keep innovating Cefis and Ciccarelli (2005) Data on 267 UK patenting firms (1988–1992) Bayesian econometric models The study shows that current innovative activity can be positively influenced by past innovation via the greater availability of financial resources Alfranca, Rama, and von Tunzelmann (2002) Information on 16,698 patents granted in the USA from 1977 to 1994 to 103 global firms in the food and beverage industry Time series analysis The evidence confirms that global firms in this industry exhibit a stable pattern of technological accumulation in which ‘success breeds success’. Latham and Le Bas (2006) Patent data for 3347 French firms (1969–1985) Duration econometric model The persistence of innovation is stronger among individuals than among firms Huang (2008) Patent and R&D data on 246 electronics firms listed on the Taiwan Stock Exchange (1998–2003) Dynamic random effect probit model Evidence supporting the existence of persistent innovation after controlling for firm heterogeneity (continued) Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016

261Economicsof lnnovationandNewTechnologyTable 1.ContinuedDataResultsAuthorsMethodologyJang and ChenSurvivalPatent data on 125 publiclyEvidence of the state(2011)listed IT firms in Taiwananalysisdependent but transient(1990-2001)nature of the competitiveadvantage attributable toinnovative persistenceSurvey data analysesDuguet andPropensityStrong evidence ofInnovation and censusMonjondata on621Frenchscoreinnovation persistence(2004)firms operating inmatchingassociated with size andmanufacturing sectorsmodelsformal R&Dactivities(1986-1996)e[eoTPMRoper andData on 3604 plantsBoth product and processHewitt-covered by the Irishinnovations arefoundtoDundasInnovative Panelbe strongly persistent(2008)(19912002)Community InnovationTPM andPeters (2009)High levels of persistence inSurvey (CIS) data ondynamicundertaking innovationGerman manufacturingprobitactivitiesand service firmsmodels(19942002)Martinez-RosESEE survey on SpanishRandomEvidence of persistence withand Labeagamanufacturing firmseffect probitrelevant complementari-(2009)(1990-1999)modelsties between product andprocess innovationUnbalanced panel of 2764MaximumRaymond et al.The study finds true(2010)likelihoodenterprises from thepersistence in theDutch Communitydynamicprobability of innovatingInnovation Surveystobit modelsin high-tech industries(1994-2000)and spurious persistenceinthe low-tech categoryClausen et al.Panel database constructedDynamicR&D-intensiveand science-(2012)fromR&Dandrandombased companies areCommunity Innovationeffectsfound tobemore likelytoSurveys in Norwayprobitbe persistent innovatorsmodelsLe Bas, Mothe,MultinomialPanel data on287firmsOrganisational innovation isand Nguyenfrom Luxembourg (CISprobitshown to be a determinant(2011)2006,2008)modelsfactor for innovationpersistenceAntonelli,TPM andData on 451 ItalianClearer evidence ofCrespi, andmanufacturingdynamicpersistence in the caseScellatocompanies observedprobitof product innovation(in press)during the yearsmodelwithrespecttoprocess1998-2006innovation whencomplementarity effectsaretaken intoaccount(Swann, Prevezer, and Stout 1998; Baptista and Swann 1999), we claim that the degree ofaccesstothestockofknowledgeof otheragentsinthesystemislikelytoplayamajorrole inassessing innovationpersistence.Thepersistenceofinnovationisthen determinedbythetwineffectsofknowledgecumulabilityinternaltofirmsandexternal tofirmsbut

Economics of Innovation and New Technology 261 Table 1. Continued. Authors Data Methodology Results Jang and Chen (2011) Patent data on 125 publicly listed IT firms in Taiwan (1990–2001) Survival analysis Evidence of the state dependent but transient nature of the competitive advantage attributable to innovative persistence Survey data analyses Duguet and Monjon (2004) Innovation and census data on 621 French firms operating in manufacturing sectors (1986–1996) Propensity score matching models Strong evidence of innovation persistence associated with size and formal R&D activities Roper and Hewitt￾Dundas (2008) Data on 3604 plants covered by the Irish Innovative Panel (1991–2002) TPM Both product and process innovations are found to be strongly persistent Peters (2009) Community Innovation Survey (CIS) data on German manufacturing and service firms (1994–2002) TPM and dynamic probit models High levels of persistence in undertaking innovation activities Martínez-Ros and Labeaga (2009) ESEE survey on Spanish manufacturing firms (1990–1999) Random effect probit models Evidence of persistence with relevant complementari￾ties between product and process innovation Raymond et al. (2010) Unbalanced panel of 2764 enterprises from the Dutch Community Innovation Surveys (1994–2000) Maximum likelihood dynamic tobit models The study finds true persistence in the probability of innovating in high-tech industries and spurious persistence in the low-tech category Clausen et al. (2012) Panel database constructed from R&D and Community Innovation Surveys in Norway Dynamic random effects probit models R&D-intensive and science￾based companies are found to be more likely to be persistent innovators Le Bas, Mothe, and Nguyen (2011) Panel data on 287 firms from Luxembourg (CIS 2006, 2008) Multinomial probit models Organisational innovation is shown to be a determinant factor for innovation persistence Antonelli, Crespi, and Scellato (in press) Data on 451 Italian manufacturing companies observed during the years 1998–2006 TPM and dynamic probit model Clearer evidence of persistence in the case of product innovation with respect to process innovation when complementarity effects are taken into account (Swann, Prevezer, and Stout 1998; Baptista and Swann 1999), we claim that the degree of access to the stock of knowledge of other agents in the system is likely to play a major role in assessing innovation persistence. The persistence of innovation is then determined by the twin effects of knowledge cumulability internal to firms and external to firms but Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016

262C.Antonelli etal.internal totheir localised context ofaction.Accesstotheknowledgebase outsideofeachfirmisnecessaryfortheintroduction oftechnological innovations.Atthesametime,how-ever, external knowledge provided by the location continues to change over time,albeitslowly.The architecture of interactions and transactions that are the carriers ofknowledgeexternalities change gradually over time as a result of thegrowth performances of firms,their entry,decline and exit and ultimately the introduction of innovations (Antonelli andScellato, forthcoming).Furthermore, because evidence of persistence has been shown to be dependent in parton the specific innovation activity scrutinised, we will use TFP growth to obtain a generalmeasure of the extentto which innovation is persistent at thefirmlevel.Theempirical testswill develop the transition probability matrix (TPM)methodology implemented by manyauthors, such as Cefis andOrsenigo (2001),Cefis(2003),Peters(2009),David and Rullani(2008)andAntonelli,Crespi,and Scellato(2012).In particular,weproposean approachthateeconsiders observing different TPMs for specific sub-periods within a longer time interval.This type of analysis enables the identification ofchanges in the transition probabilities andtheinterpretationofthemascluestotheeffectsoftheexternaleventsonpersistence3.HypothesesandresearchdesignThe generation of technological knowledge is an activity characterised by significantindivisibility and learning.Knowledge indivisibility and learning exert strong cumulativeeffects.Withincorporations,thegenerationofnewknowledgeandtheintroductionofinnovations are theresultofthecreation ofnewfunctional routines,ofresearch and developmentlaboratories and of the communication networks that allow access to externalknowledge.Thegeneration of newknowledge and therelated introduction of innovation are shapedbythejointeffectof internal cumulativeforcesand external positivefeedbackexertedbythesystem in which firms are embedded.Therefore, we retain the hypothesis that innovation is a path-dependent, rather than apast-dependent,processdetermined byseveral internal andexternalfactors.External factorsarecharacterised byhighlevels ofcontingency,as such,theirchanges affectthedynamics ofpersistence.Following the resource-based theory ofthefirm, we suppose that thefollowingfactors are important.(A)The size of firms.The generation of technological knowledge is characterised bysubstantial sunk costs. Corporations that have innovated once are more likely to continueinnovating simplybecausethe incremental costs ofthe internal facilitiesdesignedtogener-ate new technologicalknowledge and introduce innovations are low (Penrose 1959;Arrow1974;ConnerandPrahalad1996).(B)The wage level.The well-knowndynamics oftheMatthew effect are likely to applynotonlyto scientists,butalsotofirmsforatleasttworeasons.First,itseemsplausiblethat innovatingfirms are abletopayhigherwagesand,therefore,attractmore creativeandtalentedemployees.Second, innovatingfirms arelikelyto interactwithinnovative suppliersand innovative customers and, therefore, participate in more fertile and productive user-producer interactions.The repeated interaction between the accumulation of knowledgeand the creation of routines to valorise and exploit it eventually leads to the creation ofdynamic capabilities thatfavour the systematic reliance on innovation as a competitivetool(Stiglitz1987;TeeceandPisano1994;LangloisandFoss1999)costmargins onthepersistenceof inno-(C)Price-costmargins.Theeffectsofpricvation are twofold. On the one hand, large price-cost margins should provide access to

262 C. Antonelli et al. internal to their localised context of action. Access to the knowledge base outside of each firm is necessary for the introduction of technological innovations. At the same time, how￾ever, external knowledge provided by the location continues to change over time, albeit slowly. The architecture of interactions and transactions that are the carriers of knowledge externalities change gradually over time as a result of the growth performances of firms, their entry, decline and exit and ultimately the introduction of innovations (Antonelli and Scellato, forthcoming). Furthermore, because evidence of persistence has been shown to be dependent in part on the specific innovation activity scrutinised, we will use TFP growth to obtain a general measure of the extent to which innovation is persistent at the firm level. The empirical tests will develop the transition probability matrix (TPM) methodology implemented by many authors, such as Cefis and Orsenigo (2001), Cefis (2003), Peters (2009), David and Rullani (2008) and Antonelli, Crespi, and Scellato (2012). In particular, we propose an approach that considers observing different TPMs for specific sub-periods within a longer time interval. This type of analysis enables the identification of changes in the transition probabilities and the interpretation of them as clues to the effects of the external events on persistence. 3. Hypotheses and research design The generation of technological knowledge is an activity characterised by significant indivisibility and learning. Knowledge indivisibility and learning exert strong cumulative effects. Within corporations, the generation of new knowledge and the introduction of inno￾vations are the result of the creation of new functional routines, of research and development laboratories and of the communication networks that allow access to external knowledge. The generation of new knowledge and the related introduction of innovation are shaped by the joint effect of internal cumulative forces and external positive feedback exerted by the system in which firms are embedded. Therefore, we retain the hypothesis that innovation is a path-dependent, rather than a past-dependent, process determined by several internal and external factors. External factors are characterised by high levels of contingency; as such, their changes affect the dynamics of persistence. Following the resource-based theory of the firm, we suppose that the following factors are important. (A) The size of firms. The generation of technological knowledge is characterised by substantial sunk costs. Corporations that have innovated once are more likely to continue innovating simply because the incremental costs of the internal facilities designed to gener￾ate new technological knowledge and introduce innovations are low (Penrose 1959; Arrow 1974; Conner and Prahalad 1996). (B) The wage level. The well-known dynamics of the Matthew effect are likely to apply not only to scientists, but also to firms for at least two reasons. First, it seems plausible that innovating firms are able to pay higher wages and, therefore, attract more creative and talented employees. Second, innovating firms are likely to interact with innovative suppliers and innovative customers and, therefore, participate in more fertile and productive user– producer interactions. The repeated interaction between the accumulation of knowledge and the creation of routines to valorise and exploit it eventually leads to the creation of dynamic capabilities that favour the systematic reliance on innovation as a competitive tool (Stiglitz 1987; Teece and Pisano 1994; Langlois and Foss 1999). (C) Price–cost margins. The effects of price–cost margins on the persistence of inno￾vation are twofold. On the one hand, large price–cost margins should provide access to Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016

263Economics of Innovation and NewTechnologyinternal funds and favour the innovative efforts of firms;therefore,the effect should be positive.Onthe other hand, however, largeprice-costmargins are clear indicators of barrierstoentryandmarketpower.Firms thatenjoymarketpower have less incentiveto continuefundinginnovationactivities.Therefore,theeffectsshouldbenegative,especiallywhentheprice-cost margin levels are very high (Aghion et al.2005; Antonelli and Scellato 2011).(D)The investment in intangible capital.The intangible assets intensity captures firmsefforts to build innovative competencies.R&D expenditures are the traditional indicatorused to measuretheinternal efforts togeneratenewtechnological knowledge.However.R&D statistics measure onlypart of the overall effort that firms make to introduce newtechnologies.Accountancy rules provide suitable evidence of stocks of intangible capitalthat include capitalised research expenditures,purchasing costs for patents and licencesand the costs incurred to build and implement the brand and know-how (Teece, Pisano, andShuen 1997).In addition to the internal factors that the literature on innovation persistence hasaddressed, we argue that extermal factors play a crucial role.External factors are also con-tingent because the structure of the system in which external knowledge and rivalry occurchange as a result of the introduction of innovations.Ateach point intime,thenetworksof interactionsandthetypesoftransactionsonfactorandproductmarketschange.Yet,ateachpointintime,thearchitectureofthesystemandthemarketexerta strongeffectontheabilityoffirmstoaccessanduseexternalknowledgeandtorelyonitfortheintroductionoffurther innovations as a competitive tool.Because we expect that innovation is a persistentprocess that occurs when external knowledge and external, local feedback play a positiverole,we introduce,in additionto the internalfactors considered sofar,twoexternalfactors:(E)Theaccesstolocalknowledge stockgeneratedbythespilloverofotherfirmsinnovative activity provides a key contribution to the persistence of innovative activitiesSuch effects are typically inter-industrial:knowledge generated in an industry maybe usefulinother activities(Jacobs1969).Hence,weexpectthatthelevelsofTFPoffirmslocated inthe sameregion,irrespective of the industrial sector,favour the persistence of innovationThe higher the levels of TFP of all the firms that are co-localised, the higher we expect theinnovationpersistencetobe.(F)Thelevels ofinnovative activity offirms within the same industrymeasure the extentto which thetypical Schumpeterian rivalry,based upon the introduction ofinnovation, is atwork.The higher the levels of TFP of rival firms are, the stronger the competitive pressureis.The Schumpeterian rivalry pushes firms to innovate to survive.Therefore,we expectthat thehigher the efficiency ofthe rivals within the same industry,thehigher the likelihoodthateach firm relies on the introduction ofinnovation as a competitive tool and the strongerthe persistence of innovation will be (Aghion et al. 2005).These hypotheses are consistentwith the model described byGruber (1992)about the role of sequential product innovationsin maintaining leadership in markets characterised by vertical differentiation.Externalfactorsaddtointernalfactorsandshapethecontextinwhichthepersistenceofinnovation occurs.Theexternal conditions,namely thequality oflocal pools ofknowledgeand the strength ofthe Schumpeterian rivalry,together with the internal conditions(thatis,thelevel ofdynamic ability,asproxied bywagelevelsandfirm size),exerta specificand localised effect onthepersistentintroduction of innovations.Becauseexternalities areinternal to the local system in which firms are embedded, the changing conditions exert apath-dependent effect on the sequence of innovations.To study the persistence of innovation,we rely on a classic indicator such asTFP.Weassumethatinnovationhasamuchbroaderscopethanindicatorsfocusedonthegenerationand introduction of new,science-based technologies suchaspatentstatistics oraimed at

Economics of Innovation and New Technology 263 internal funds and favour the innovative efforts of firms; therefore, the effect should be pos￾itive. On the other hand, however, large price–cost margins are clear indicators of barriers to entry and market power. Firms that enjoy market power have less incentive to continue funding innovation activities. Therefore, the effects should be negative, especially when the price–cost margin levels are very high (Aghion et al. 2005; Antonelli and Scellato 2011). (D) The investment in intangible capital. The intangible assets intensity captures firms’ efforts to build innovative competencies. R&D expenditures are the traditional indicator used to measure the internal efforts to generate new technological knowledge. However, R&D statistics measure only part of the overall effort that firms make to introduce new technologies. Accountancy rules provide suitable evidence of stocks of intangible capital that include capitalised research expenditures, purchasing costs for patents and licences and the costs incurred to build and implement the brand and know-how (Teece, Pisano, and Shuen 1997). In addition to the internal factors that the literature on innovation persistence has addressed, we argue that external factors play a crucial role. External factors are also con￾tingent because the structure of the system in which external knowledge and rivalry occur change as a result of the introduction of innovations. At each point in time, the networks of interactions and the types of transactions on factor and product markets change. Yet, at each point in time, the architecture of the system and the market exert a strong effect on the ability of firms to access and use external knowledge and to rely on it for the introduction of further innovations as a competitive tool. Because we expect that innovation is a persistent process that occurs when external knowledge and external, local feedback play a positive role, we introduce, in addition to the internal factors considered so far, two external factors: (E) The access to local knowledge stock generated by the spillover of other firms’ innovative activity provides a key contribution to the persistence of innovative activities. Such effects are typically inter-industrial: knowledge generated in an industry may be useful in other activities (Jacobs 1969). Hence, we expect that the levels of TFP of firms located in the same region, irrespective of the industrial sector, favour the persistence of innovation. The higher the levels of TFP of all the firms that are co-localised, the higher we expect the innovation persistence to be. (F) The levels of innovative activity of firms within the same industry measure the extent to which the typical Schumpeterian rivalry, based upon the introduction of innovation, is at work. The higher the levels of TFP of rival firms are, the stronger the competitive pressure is. The Schumpeterian rivalry pushes firms to innovate to survive. Therefore, we expect that the higher the efficiency of the rivals within the same industry, the higher the likelihood that each firm relies on the introduction of innovation as a competitive tool and the stronger the persistence of innovation will be (Aghion et al. 2005). These hypotheses are consistent with the model described by Gruber (1992) about the role of sequential product innovations in maintaining leadership in markets characterised by vertical differentiation. External factors add to internal factors and shape the context in which the persistence of innovation occurs. The external conditions, namely the quality of local pools of knowledge and the strength of the Schumpeterian rivalry, together with the internal conditions (that is, the level of dynamic ability, as proxied by wage levels and firm size), exert a specific and localised effect on the persistent introduction of innovations. Because externalities are internal to the local system in which firms are embedded, the changing conditions exert a path-dependent effect on the sequence of innovations. To study the persistence of innovation, we rely on a classic indicator such as TFP. We assume that innovation has a much broader scope than indicators focused on the generation and introduction of new, science-based technologies such as patent statistics or aimed at Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016

264C. Antonelli et al.detecting the specific introductionofnewproductsand processes,as measuredbyinnovationcounts.Innovation consists,moregenerally,ofthesystematiccapabilitytogeneratenewknowl-edge and to apply it to the broad array ofactivitiesthatfirms engage in.Thusfar, our notionofinnovation ismuchbroaderandretainsa strong Schumpeterian flavourbecause itincludesthe introduction of new products and newprocesses aswell as the introductionofchangesin the organisation, in the mix of inputs and in the product and factor markets into whichfirms operate.Hence, we assume that TFP is better able to capture the general increasein efficiency within a firm that has a good command of technological, organisational andcommercial knowledge.Clearly,our hypothesis here is that the ability to introduce an innovation at timet +1depends ontheintroduction ofan innovationattimetandon theeffects ofcontingentforcesthat exert themselves locallytoaffect the sequence of state dependency.eeOur two hypotheses lead to a two-step research design. First, we focus our analysison the determination of innovative activity persistence as measured by TPMs computedusing variations in the levels of TFP. Within the considered time period, we explore thepossibility that relevant external factors may affect the transition probabilities. Herein, weintroducethemultipletransitionprobabilitymatrixes(MTPMs)approach,whichconsistsofanalysing sub-period TPMstotest whethertransition probabilities change withinthetimeperiodconsidered.The MTPMs involve computing a single Markov chain for the full period of timeand comparing those results with the results of computing different Markov chains in therelevant sub-periods. These sub-periods are identified by significant contingent events thatareexpectedtoaffectthetransitionprobabilitiesbetweentheinnovativeandnon-innovativestatusoftheanalysed companies.We suggestthatthis approach,based onthecomparisonof the parametersof theMarkovchains indifferentsub-periodsshould allowabetteridentification of the path-dependent character of the innovation process. In particular, theobservation of different parameters for the Markov chains in different sub-periods mightindicate that the extent of innovation persistence is affected by contingent events and,therefore, that innovation can be qualified as a path-dependent process.Second, we concentrate the analysis on the determinants of innovation persistencebecause we want to qualify the type of persistence at work as well as the role of non-observableheterogeneity.Our main argument here is that several contingent and localisedconditions,both internalandexternalto eachfirm,haveasignificanteffectonthepersistence.The persistence of the innovative activity is therefore path-dependent, not past-dependent(David 1997, 2007).4.Theempiricalevidence4.1.ThedataOuranalysisisbasedon an originaldata setcontainingaccountingdataforasampleofItalianmanufacturingfirms.Thedata set includesfinancial accountingdataforalargesampleofmanufacturingcompanies,observedfrom1996to2005.ThedatahavebeenextractedfromtheAnalisi InformatizzataDelleAziende databaseprovided byBureau Van Dick,whichreportsaccountinginformationforpublicandprivateItalianfirmswithaturnoverlargerthan 0.5 millions of Euros.The companies, included in the analysis, were founded before1995areregisteredas amanufacturingsectoraccordingtothe ItalianATECOclassificationand were active at the end of 2005.The introduction of the latter condition implies that wedo not considermarketexitorentry

264 C. Antonelli et al. detecting the specific introduction of new products and processes, as measured by innovation counts. Innovation consists, more generally, of the systematic capability to generate new knowl￾edge and to apply it to the broad array of activities that firms engage in. Thus far, our notion of innovation is much broader and retains a strong Schumpeterian flavour because it includes the introduction of new products and new processes as well as the introduction of changes in the organisation, in the mix of inputs and in the product and factor markets into which firms operate. Hence, we assume that TFP is better able to capture the general increase in efficiency within a firm that has a good command of technological, organisational and commercial knowledge. Clearly, our hypothesis here is that the ability to introduce an innovation at time t + 1 depends on the introduction of an innovation at time t and on the effects of contingent forces that exert themselves locally to affect the sequence of state dependency. Our two hypotheses lead to a two-step research design. First, we focus our analysis on the determination of innovative activity persistence as measured by TPMs computed using variations in the levels of TFP. Within the considered time period, we explore the possibility that relevant external factors may affect the transition probabilities. Herein, we introduce the multiple transition probability matrixes (MTPMs) approach, which consists of analysing sub-period TPMs to test whether transition probabilities change within the time period considered. The MTPMs involve computing a single Markov chain for the full period of time and comparing those results with the results of computing different Markov chains in the relevant sub-periods. These sub-periods are identified by significant contingent events that are expected to affect the transition probabilities between the innovative and non-innovative status of the analysed companies. We suggest that this approach, based on the comparison of the parameters of the Markov chains in different sub-periods, should allow a better identification of the path-dependent character of the innovation process. In particular, the observation of different parameters for the Markov chains in different sub-periods might indicate that the extent of innovation persistence is affected by contingent events and, therefore, that innovation can be qualified as a path-dependent process. Second, we concentrate the analysis on the determinants of innovation persistence because we want to qualify the type of persistence at work as well as the role of non￾observable heterogeneity. Our main argument here is that several contingent and localised conditions, both internal and external to each firm, have a significant effect on the persistence. The persistence of the innovative activity is therefore path-dependent, not past-dependent (David 1997, 2007). 4. The empirical evidence 4.1. The data Our analysis is based on an original data set containing accounting data for a sample of Italian manufacturing firms. The data set includes financial accounting data for a large sample of manufacturing companies, observed from 1996 to 2005. The data have been extracted from the Analisi Informatizzata Delle Aziende database provided by Bureau Van Dick, which reports accounting information for public and private Italian firms with a turnover larger than 0.5 millions of Euros. The companies, included in the analysis, were founded before 1995 are registered as a manufacturing sector according to the Italian ATECO classification, and were active at the end of 2005. The introduction of the latter condition implies that we do not consider market exit or entry. Downloaded by [Wilfrid Laurier University] at 00:11 02 August 2016