Frontiers in Evolutionary Economics: Themenheft 2+3/Bd. 234(2014) Jahrbücher für Nationalökonomie und Statistik 9783110509205, 9783828206021

Citation preview

Frontiers in Evolutionary Economics

Edited by U w e Cantner and Giovanni Dosi

With Contributions by Amendola, Mario, University of Roma La Sapienza Binder, Martin, SPRU, University of Sussex, Brighton and Universität Kassel Birchenhall, Chris, University of Manchester Brenner, Thomas, Philipps University of Marburg Buenstorf, Guido, INCHER, Kassel Chai, Andreas, Griffith Business School, Griffith University Cohendet, Patrick, Mosaic, H E C Montréal and BETA, Université de Strasbourg Duschl, Matthias, Philipps University of Marburg Gaffard, Jean-Luc, OFCE, SKEMA Business School, Sophia Antipolis Garavaglia, Christian, DEMS, University of Milano Bicocca and CRIOS, Bocconi University Geissler, Matthias, INCHER, Kassel Hanappi, Hardy, University of Technology of Vienna Herrmann-Pillath, Carsten, Frankfurt School of Finance and Management Korn, Jenny, University of Illinois at Chicago

Lucius &C Lucius · Stuttgart 2 0 1 4

Llerena, Patrick, BETA, Université de Strasbourg Loasby, Brian J., University of Stirling Luxen, Dennis, Karlsruhe Institute of Technology Malerba, Franco, CRIOS, Bocconi University Moneta, Alessio, Scuola Superiore Sant' Anna Murmann, Johann Peter, University of New South Wales, Sydney Nooteboom, Bart, Emeritus Professor, the Hague Orsenigo, Luigi, CRIOS, Bocconi University and IUSS, Pavia Pezzoni, Michele, CRIOS, Bocconi University, OST, Paris, and CEMI, Lausanne Schimke, Antje, Karlsruhe Institute of Technology Schubert, Christian, BTU Cottbus Senftenberg Simon, Laurent, Mosaic, H E C Montréal Vanberg, Viktor J., Walter Eucken Institut, Freiburg Windrum, Paul, University of Nottingham Worch, Hägen, Swiss Distance University of Applied Sciences

Guest Editors Professor Dr. Uwe Cantner Friedrich-Schiller-Universität Jena Wirtschaftswissenschaftliche Fakultät Carl-Zeiss-Straße 3, Room 5.23 0 7 7 4 3 Jena u we ,cantner@uni-j ena.de Professor Giovanni Dosi Laboratory of Economies and Management St'Anna School of Advanced Studies Via Carducci 4 0 5 6 1 2 7 Pisa, Italy [email protected]

Bibliografísche Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografísche Daten sind im Internet über http://dnb.d-nb.de abrufbar ISBN 9 7 8 - 3 - 8 2 8 2 - 0 6 0 2 - 1

© Lucius Sc Lucius Verlagsgesellschaft mbH · Stuttgart - 2 0 1 4 Gerokstraße 5 1 , D - 7 0 1 8 4 Stuttgart Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlags unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen und Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen.

Satz: inMEDIAlo, Digital- und Printmedien UG, Plankstadt Druck und Bindung: Neumann Druck, Heidelberg Printed in Germany

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

Inhalt/Contents Guest Editorial

116

Abhandlungen / Original Papers Cohendet, Patrick, Patrick Llerena, Laurent Simon, The Routinization of Creativity. Lessons from the Case of a Video-game Creative Powerhouse Loasby, Brian ]., The Evolution of Knowledge and Knowledge of Evolution Buenstorf, Guido, Matthias Geissler, Like Doktorvater, like Son? Tracing Role Model Learning in the Evolution of German Laser Research Nooteboom, Bart, In What Sense do Firms Evolve? Murmann, Johann Peter, Jenny Korn, Hagen Worch, How Fast Can Firms Grow? Duschl, Matthias, Antje Schimke, Thomas Brenner, Dennis Luxen, Firm Growth and the Spatial Impact of Geolocated External Factors .. Amendola, Mario, Jean-Luc Gaffard, Competition and Increasing Returns Garavaglia, Christian, Franco Malerba, Luigi Orsenigo, Michele Pezzoni, Innovation and Market Structure in Pharmaceuticals: An Econometric Analysis on Simulated Data Chai, Andreas, Alessio Moneta, Escaping Satiation Dynamics: Some Evidence from British Household Data Vanberg, Viktor J., Evolving Preferences and Welfare Economics: The Perspective of Constitutional Political Economy Schubert, Christian, Martin Binder, Reconciling Normative and Behavioral Economics: An Application of the "Naturalistic Approach" to the Adaptation Problem Birchenhall, Chris, Paul Windrum, Global Warming: Technology, Preferences and Policy Herrmann-Pillath, Carsten, Naturalizing Institutions: Evolutionary Principles and Application on the Case of Money Hanappi, Hardy, Evolutionary Political Economy in Crisis Mode

120-141 142-157

158-184 185-209 210-233 234-256 257-273

274-298 299-327 328-349

350-365 366-387 388-421 422^440

Buchbesprechungen / Book Reviews Aufderheide, Detlef, Martin Dabrowski (Hrsg.), Effizienz oder Glück? Wirtschaftsethische und moralökonomische Perspektiven der Kritik an ökonomischen Erfolgsfaktoren Maennig, Wolfgang, Andrew Zimbalist (eds.), International Handbook on the Economics of Mega Sporting Events Solow, Robert M., Jean-Philippe Touffut (eds.), What's Right with Macroeconomics?

441 445 446

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

Guest Editorial The diffusion of an evolutionary approach in economics during the past roughly 40 years has been steady and persuasive. In some areas - such as the economics of innovation and technological change, industrial dynamics, the theory of the firm, the economics of development - evolutionary interpretations feature prominently. In others - including the analysis of institutions, economic geography, consumption and the demand side, the relationship between economics and law, and the welfare implications of an evolutionary perspective - the inroads are still somewhat exploratory and pioneering. This special issue is taking stock of some recent developments especially in the latter fields and presents an ensemble of works many of which resonate with the variegated research interests of Ulrich Witt. Let us distinguish in the following four broad groups of papers addressing "Creativity and knowledge accumulation", "Firms, consumers and structural change", "Welfare and evolving preferences" and "The macro and institutional levels".

Creativity and knowledge accumulation Without any doubt, newness and hence invention and innovation are core ingredients of an evolutionary approach. Research into the determinants of newness has always been facing the epistemological caveat that new ideas cannot be foreseen ex ante otherwise they would not be new (Arrow 1991: 473). Certainly, a lot of progress has been done in the identification of the sources of innovative knowledge, its structure, and the procedure of search (for a recent critical survey, see Dosi/Nelson 2010). However, we know much less about the very process of the creation of new ideas and how it relates with the generation, dissemination and evolution of knowledge and competencies. To what extent is "creation" itself routinized? And, how does it relate to organizational set-ups? In "Lessons from the Case of a Video-Game Creative Powerhouse" Patrick Cohendet, Patrick Llerena and Laurent Simon offer an in-depth exploration of the micro-context of the origin of routines (generative, interpretative and evaluative, as suggested in Witt 2009) and of their intimate link with organizational creativity. The authors show how organizational creativity orchestrates continuous interactions between different types of routines, operating at different levels of the organization. The in-depth analysis of organizational creativity is pursued in the world-leading videogame company, Ubisoft. To some extent, Ubisoft is one of the flagships of the "creative industries", in which the clear imperative is to sustain creativity on a permanent basis. With the paper by Brian Loasby the analytical focus shifts from creativity to knowledge. In his paper "The Evolution of Knowledge and Knowledge of Evolution " the author reflects on knowledge being considered the outcome of a process of trial and error, the rate and content of which depends on its organization, both conscious and unconscious. The generation and also the dissemination, transfer and replication of new knowledge crucially depends also on the balance between codification and tacitness. Whenever the latter feature is strong the inheritance of knowledge requires specific mechanisms, one of which is likely to be observational learning, as suggested in Witt (1998). The paper "Like Doktorvater, like Son? Tracing Role Model Learning in the Evolution of German Laser Research" by Guido Buenstorf and Mathias Geissler addresses indeed knowledge transfer in public research (German laser research between 1960 and 2007) and asks whether the relationship between a PhD advisor (Doktorvater) and a Phd student (son) involves observational learning effects with tacit transfers of capabilities affecting the differential performance of doctoral students, conditional on the revealed capabilities of their supervisors. And it significantly does.

Guest Editorial - 1 1 7

Firms, consumers and structural change In "In What Sense do Firms Evolve?" Bart Nooteboom adds to the literature on firms seen with a cognitive approach. Herein a firm is understood as a collection of actors with different cognitions (as in Witt 2005) and with different cognitive distances among them. And it is from this angle that the paper tackles the interrelationship among variety generation, selection and transmission, core elements of a (cultural) evolutionary approach in general and the "evolution of firms" in particular. A crucial aspect of the dynamic of a firm is its growth. Johann Peter Murmann, Jenny Korn, and Hagen Worch try to identify the factors that distinguish continuously growing firms from others. The analysis in "How Fast Can Firms Grow?" addresses the maximum rate a firm may be able to grow. The interpretation offers an easy connection to capability-based theories of the firm and also to the paper by Bart Nooteboom as the authors also interpret their results in terms of a cognitive approach to firm development as already envisaged in Witt (1998). In terms of the (open) boundaries of a firm, emphasized by cognitive theories, the knowledge relationships among firms and between firms and other organization also affect firm growth. Matthias Duschl, Antje Schimke, Thomas Brenner and Dennis Luxen contribute to this strand of literature by investigating spatial dimensions of knowledge relationships (such as described in Witt et al. 2012). In "Firm Growth and the Spatial Impact of Geolocated External Factors" the authors focus on the impact of spatial externalities, and shed new light on the interplay between factors internal and external to firms as determinants of their growth. Knowledge accumulation and innovation are intrinsically characterized by various forms of dynamic increasing returns. In turn, the latter in principle can easily lead to a monopolization of the market. The contribution "Competition and Increasing Returns" by Mario Amendola and Jean-Luc Gaffard asks under which conditions innovation and competition among several firms may persist even if production is undertaken under increasing returns. The authors show that internal increasing returns may not lead to concentration but to specialization and hence interdependence among firms given a sequential structure of the production process wherein increasing returns are not relevant at each stage. The evolutionary dynamics of industries driven by heterogeneous agents is addressed by Christian Garavaglia, Franco Malerba, Luigi Orsenigo, and Michele Pezzoni by providing an analysis via an agent-based model out of the modelling framework of history friendly models. The fundamental challenge this type of approach faces is to identify evolutionary processes and their interactions that lead to pattern of structural dynamics matching with empirically observable pattern of industrial dynamics. In the paper "Innovation and market structure in pharmaceuticals: an econometric analysis on simulated data" the authors attempt to track the structural dynamics of this industry which is characterized by low degrees of overall concentration but also first mover advantages exploited by a few innovative firms. Driving forces behind this structural pattern - the authors suggest - are multiple processes of highly uncertain search, the discovery of new submarkets as well as the interactions between patent protection, imitation and price competition. The demand side of markets and the possible emergence of satiation processes are analyzed by Andreas Chai and Alessio Moneta. In their contribution "Escaping Satiation Dynamics: Some Evidence from British Household Data" the authors investigate the so-called satiation-escape hypothesis as formulated in Witt (2001). By looking at how income intensities change with income over time they find that satiation is not a rare case. However, innovation tends to shift upward Engel curves.

118 · Guest Editorial

Welfare and evolving preferences Although envisaged already in the seminal work of Dick Nelson and Sid Winter in 1982, An Evolutionary Theory of Economic Change, a major recent development evolutionary economics, welfare issues have never featured prominently in evolutionary analysis , one of the exception being several works by Ulrich Witt. Of course one of the major difficulties for such an analysis is that in evolutionary worlds preferences change. And any normative dimension requires the understanding of how preferences evolve since it is on the latter that welfare statements are typically based. Three papers in this special issue address these issues from three different directions. Viktor Vanberg, in "Evolving Preferences and Welfare Economics: The Perspective of Constitutional Political Economy", analyzes the approaches from von Weizsäcker, Sugden and Witt, arguing that from a constitutional economics perspective evolving preferences provide no obstacle to normative analyses. A different way to approach evolving preferences is suggested by Christian Schubert and Martin Binder in "Reconciling Normative and Behavioral Economics: An Application of the 'Naturalistic Approach' to the Adaptation Problem". The authors anchor their approach still in the preferences of individuals - as traditional welfare economics does - but look at factors that determine preference formation. The authors suggest a "dynamic naturalistic approach" (as envisaged in Witt 1987) informed by psychological insights into such a formation process. This procedural approach considers the normative status of a certain preference being dependent on the way it was formed and the related capabilities to learn. "Global Warming: Technology, Preferences and Policy" by Chris Birchenhall and Paul Windrum discusses the decisive role of belief, attitude and preference changes of households/citizens, firms and politicians in the solution (or lack thereof) of global warming problems via technology substitution. The authors propose an analytical framework highlighting the fundamental institutional failures driven by path-dependencies. Drawing on earlier work on the interaction between consumers and firms in co-evolutionary processes the paper extends the analysis to the way firms and households/citizens interact with the political domain. Paradigm shifts can only come about via changes in policies next to changes in lifestyles and technologies.

The macro and institutional levels The major share of work in evolutionary economics has been microeconomic in focus with only weak connections with macro level analysis. Exceptions to that are some analyses of the emergence of institutions and also a more recent interest in micro-macro modelling and related policy analysis. The final two papers are to be seen within these strands of research. In "Naturalizing Institutions: Evolutionary Principles and Application on the Case of Money" Carsten Herrmann-Pillath offers an evolutionary approach towards the emergence and persistence of institutions, exemplified for the case of "money". The author draws on the replicator-interactor duality prominent in the "Darwinian" approach and applies it to a population level dynamics (macro) interacting with cognitive phenomena on the individual level (micro). Macroeconomic policies and how they can be informed by evolutionary economics is at the core of the paper "Evolutionary Political Economy in Crisis Mode" by Hardy Hanappi. In the light of the current economic crisis, the author discusses the (related ?) crisis of economics as a discipline and the purported a lack of an evolutionary economic policy approach (just as in Witt 2003).

Guest Editorial - 1 1 9

References Arrow, Κ. (1991), The Dynamics of Technological Change. Pp. 4 7 3 ^ 7 6 in: OECD (ed.), Technology and Productivity: The Challenge for Economic Policy, Paris. Dosi, G., R.R. Nelson (2010), Technical Change and Industrial Dynamics as Evolutionary Processes Pp. 5 2 - 1 2 7 in: B.H. Hall, N. Rosenberg (eds), Handbook of the Economics of Innovation. Burlington, MA: Academic Press. Nelson, R.R., S.G. Winter (1982), An Evolutionary Theory of Economic Change. Cambridge, Mass. and London: The Bellknap Press of Harvard University Press. Witt, U. (1987), Individualistische Grundlagen der evolutorischen Ökonomik. J.C.B. Mohr (Paul Siebeck). Witt, U. (1998), Imagination and Leadership - The Neglected Dimension of an Evolutionary Theory of the Firm. Journal of Economic Behavior &c Organization 35(2): 161-177. Witt, U. (2001), Learning to consume - A theory of wants and the growth of demand. Journal of Evolutionary Economics 11: 23-36. Witt, U. (2003), Economic policy making in evolutionary perspective. Journal of Evolutionary Economics 13: 77-94. Witt, U. (2009), Propositions about novelty. Journal of Economic Behavior and Organization 70: 311-320. Witt, U. (2005), The evolutionary perspective on organizational change and the theory of the firm. Pp. 339-364 in: K. Dopfer (ed.), The evolutionary foundations of economics. Cambridge: Cambridge University Press. Witt, U., T. Broekel, T. Brenner (2012), Knowledge and its economic characteristics: a conceptual clarification. Pp. 369-382 in: R. Arena, A. Festré, Ν. Lazaric (eds.), Handbook of Economics and Knowledge. Cheltenham: Edward Elgar Publishing. Uwe

Cantner

Giovanni

Dosi

Jahrbücherf. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

The Routinization of Creativity Lessons from the Case of a Video-game Creative Powerhouse Patrick CohendetabI Patrick Llerenab, Laurent Simon3* Mosaic, HEC Montréal, Canada b BETA, Université de Strasbourg, France

a

JEL 031; L23; L82 Organizational creativity; organizational routines; video games.

Summary The aim of this contribution is to proceed to an in-depth exploration of the micro-context of the origin of routines and of their intimate link with organizational creativity. Our view is that organizational creativity orchestrates continuous interactions between different types of routines, operating at different levels of the organization. More precisely we propose distinguishing three types of routines: - First, the routines issued from formal structures or hierarchical working groups in the firm (functional groups, project teams, task force, etc.), for which the context of work and coordination of specialized tasks is defined ex ante by the hierarchy of the firm; - Second, the routines emerging from informal structures, the "knowing communities" which is a "generic term that defines different types of autonomous learning groups of individuals (communities of practice, epistemic communities, and other more or less informal learning groups) united by common beliefs and interests who voluntarily share their resources on a long term basis in order to create and diffuse knowledge"; - Third, the routines that are inherently related to the organizational creativity of the firm, which are essentially corporate routines as expression of patterns of thinking, feeling and acting in the corporate culture. In essence they are the genes of collective identity, and take the shape of project management staging and gating principles and practices, framing collective divergent exploration and convergent production toward a creative goal. The contribution is based on an in-depth analysis of the organizational creativity in the worldleading videogame company, Ubisoft, with a special focus on the studio located in Montréal. T o some extent, Ubisoft is one of the flagships of the "creative industries", in which the clear imperative is to sustain creativity on a permanent basis. These reasons explain the choice we made to test our approach of organizational creativity and routines in this firm.

* Patrick Cohendet and Laurent Simon wish to express their gratitude to the members of the executive team of Ubisoft Entertainment Studio in Montreal for their openness and continuing and lasting support for research, and to Ubisoft's employees in Montreal. They also thank the FQRSC and SSHRC for their support, and the funding partners of Mosaic research plateform on managing creativity for innovation at HEC Montréal (http://mosaic.hec.ca/). Laurent Simon is grateful to The Gutenberg Fondation / Cercle Gutenberg in Strasbourg for its support in 2009-2010 that allowed to start up the discussions on this paper. Patrick Llerena would like to thank the partners of the Chair in Management of creativity of the University of Strasbourg: SALM, SOCOMEC and Voirin Conseil en management.

The Routinization of Creativity • 121

Introduction In everyday language, the "routinization of creativity" may sound as an o x y m o r o n . M o s t clichés about creativity refer to some form of free, u n b o u n d e d exploration and expression. Routines, on the contrary, w o u l d rest on stability, regularity, systematization, and standardization of actions and behaviors. For the last 30 years or so, research in economics, strategy a n d management addressed these t w o concepts in depth and suggested a completely different and stimulating view on routines and creativity. O n e important hurdle to consider, though, is that the interest generated by these t w o concepts produced a wide range of different, often divergent, and sometimes conflicting definitions (Becker 2 0 0 4 , 2 0 0 5 ; Felin/Foss 2004). Furthermore, as these definitions are mostly inspired by the background discipline of the researchers, they tend to be onedimensional, to address one level of unit of analysis, and to delude concrete and operational concerns. T o refer to t w o generic metaphors, the scientific status of these t w o concepts evolved f r o m "black b o x " to "mirror ball". Inspired by classic information theory, the black box m e t a p h o r refers to a system the inner working (and origins) of which remains u n k n o w n , but that can be studied and analyzed through its inputs and outputs. 1 W i t h regards to routines, the black box refers to the fact that if routines were k n o w n to be a n essential element of the functioning of organizations, they were mostly taken for granted or briefly defined as " p r o g r a m s for action" (Simon 1981) or "genes of the organization" (Nelson/Winter 1982). The recent literature on routines which literally flourished in the past decade, as emphasized by Becker (2004), suggested multiple, sometimes conflicting definitions, u p to the point of a certain academic confusion a b o u t the very nature of the p h e n o m e n o n (see for instance: Cohen et al. 1996 ). Even if we settle on a definition of routines as "patterns of interaction", as suggested by Becker, the discussion remains open a b o u t the way micro-routines originate and the roles they play as "the basic elements" (Lippman/Rumelt 2003) that drive differences in learning and capability development between organizations. As Felin and Foss (2004: 23) wrote "while references abound to notions of organizational routines and capabilities, at present in evolutionary economics and strategy we have no theory of their origin...." T o reflect this situation, we introduce the complementary m e t a p h o r of the " m i r r o r ball", where the concept being addressed and analyzed by academics f r o m different fields mostly eludes an integrated understanding yet offers interesting reflections about those fields, in terms of theoretical issues, epistemological discussions and empirical methods. As emphasized by some researchers, empirical studies could allow for a more realistic understanding of the nature and evolution of routines (Becker/Lazaric 2009). W i t h this "research prog r a m " on the way, empirical studies provided mixed results, mainly due to the inability to settle on an operational definition of routines, as a p h e n o m e n o n to be observed a n d assessed in action. With regard to "organizational creativity", the concept parallels the situation of routines as a concept evolving f r o m the status of "black b o x " to "mirror ball". Following a few defining papers in the 90's ( W o o d m a n et al. 1993; Drazin et al. 1999), it seems that the concept of organizational creativity is ubiquitous in the literature, covering different realities considering the standpoints of contributors f r o m the fields of psychology, orga1

The origin of the metaphor is usually attributed to von Neumann (1951) and goes back to a lecture delivered in 1948. "The general and logical theory of automata" (Jeffress 1951).

122 · Patrick Cohendet, Patrick Llerena, and Laurent Simon

nizational behavior, management, economics or strategy. The concept was also recently enriched by new perspectives and fields of study such as knowledge management (Nonaka 1994; Fleming et al. 2007), the design field (Hargadon/Sutton 1997; Hargadon 2002; Martin 2009; Brown 2009), economic geography and urban economics (Grabher 2001; Florida 2002; Florida et al. 2008; Scott 2005; Asheim/Gertler 2005; Pratt 2010), engineering (Altshuller 1984; Hatchuel/Weil 2009) as well as computer science (Sosa/Gero 2003; Wierzbicki/Nakamori 2006), for instance. As emphasized by Styhre: "Although organization creativity literature has made some fruitful contributions to organization theory, the literature is still, in comparison to research on for instance knowledge management or organization learning, too disjointed and dispersed to make a broader impact on the field" (2006: 146). In the end, in both cases, "routines" and "organizational creativity", knowledge appears either too generic or broadly idiosyncratic. Furthermore, operational or implementation issues are left to the experimentation and more or less enlightened improvisation of practitioners. However, an emerging literature in management has started considering some virtuous interactions between routines and organizational creativity. As an example, the literature in business has applied recurrently the example of jazz, and more specifically the art of improvisation with which it is associated, to put forward the organizational mechanisms of creativity occurring within and beyond a highly constrained structure of routines (Hatch 1999; Zack 2000; Kamoche/Cunha 2001). The locus of emergence of routines supporting creative processes becomes particularly interesting. On the one hand, the literature on routines is relatively silent on the emergence of the process of routines, particularly concerning routines related to dynamic processes such as innovative ones. On the other hand, organizational creativity offers an interesting hypothesis: As for today, the definition of organizational creativity introduced by W o o d m a n et al. has not been fully questioned. If it remains largely elusive, it still offers an interesting starting point, since it is one of the first to view organizational creativity as a collective endeavour (and not resulting from a purely individual act): "Organizational creativity is the creation of a valuable, useful new product, service, idea, procedure, or process by individuals working together in a complex social system" (Woodman et al. 1993: 293 emphasis ours). This quote balances the active role of "individuals working together", with the "complex social system", the intricate social and technical, formal and informal fabric of the organization. On the one hand, "individual working together" are acting and performing as "knowing communities" sharing knowledge, and building on each other's expertise to conceive "new and useful combinations": "In addition to identifying appropriate and useful knowledge of group members to apply to the group problem solving, groups provide an arena in which members can use others as resources to augment their own knowledge. In this manner, the member does not just add to his own knowledge but uses others' knowledge to stimulate the usefulness of his or her own skills" (Woodman et al. 1993, p. 303). On the other hand, these social groups are interacting with other formal and informal - social groups in the organization (the complex social system) to perform and generate new outputs, framed by formal and informal processes. Drazin et al. (1999:291,293-294) also pointed this key role of communities, integrating the individual contributions, and negotiating new practices with other communities in the organization. In line with Cohendet and Llerena (2003), describing the firm as the interplay of knowing communities, our main hypothesis is that the interactions and confrontation between knowing communities are progressively shaping new "patterns of interaction" (Simon 1981) that are crystallized as new "programs for action" (Becker 2004), i.e. routines.

The Routinization of Creativity · 123

To sum up our argument, we suggest that the locus of emergence of routines is to be found at the intermediary level between individuals and organizational processes, namely in the interplay knowing communities. Moving further beyond this perspective, our aim in this article is to contribute to the organizational creativity literature by considering that routines and creativity in organizations are intrinsically complementary. Our view is that there is no contradiction between creativity and routinization, and that managing creativity in organizations triggers continuous interactions between different types of routines operating at different levels of the organization. From this perspective, we suggest to identify different regimes of routines, based on their origins, and to pay a specific attention to the roles and interactions between the active units of the routines. The paper opens by introducing and combining two bodies of literature (routines and creativity) through an in-depth exploration of the micro-context of the origin of routines, as well as of the conditions of their emergence, and of their intimate link with organizational creativity. Our view is that organizational creativity orchestrates continuous interactions between different types of routines, operating at different levels of the organization. More precisely we propose to distinguish three types of routines: - First, the routines issued from formal structures or hierarchical working groups in the firm (functional groups, project teams, task force, etc.), for which the context of work and coordination of specialized tasks is defined ex ante by the hierarchy of the firm; - Second, the routines emerging from informal structures, the "knowing communities" which is a "generic term that defines different types of autonomous learning groups of individuals (communities of practice, epistemic communities, and other more or less informal learning groups) united by common beliefs and interests who voluntarily share their resources on a long term basis in order to create and diffuse knowledge" (Cohendet et al. 2010); - Third, the routines that are inherently related to the organizational creativity of the firm, which are essentially corporate routines as expression of patterns of thinking, feeling and acting in the corporate culture. In essence they are the genes of collective identity, and take the shape of project management staging and gating principles and practices, framing collective divergent exploration and convergent production toward a creative goal. To a large extent our analysis of the emergence and formation of novelty in organization echoes Ulrich Witt's vision (2009), when he argued that the creation of new cognitive concepts (ideas, imaginings) involves three operations. "One is a generative operation that produces new (re-)combinations elements. The other is an interpretative operation by which the new (re-)combination is integrated into a newly emerging or a more general already existing concept. Yet another operation can often be observed to accompany the interpretative operation, namely an evaluative one. However, where the interpretative operation answers the question of what it is that emerges, the evaluative operation is concerned with what the utility, advantage, benefit of this is" (Witt 2009: 113). We suggest that the generative and interpretative operation results from constant interactions between the routines issued from the formal and informal structures, while the evaluative one is orchestrated by the corporate routines. Our contribution is based on an in-depth analysis of the organizational creativity in the world-leading videogame company, Ubisoft, with a special focus on the studio located in Montréal. As of fall 2013, Ubisoft Montreal studio is the largest videogame development

124 • Patrick Cohendet, Patrick Llerena, and Laurent Simon

studio in the world. This French-owned video game developer and editor established a studio in Montréal, in 1997, to benefit from substantial grants and tax credits offered by the provincial government, but also from the growing experience of the local creative workforce, well-trained in computer-science, cinema, fine arts, literature, theatre, management and marketing (Cohendet/Simon 2007). The Montréal Studio hires employees mainly from Montréal (around 80 %), most of whom have been trained in the Montréal arts and computer schools, and in various university business programs. Ubisoft Montréal now employs over 1800 people, all scattered in the 250.000 square feet of open space offered by the red-brick building situated in the heart of the Mile-End neighborhood considered as one of the hippest and most creative urban areas in the city. Throughout the years, the home environment has offered a fertile ground for individuals to build informal contacts, as well as provided the formal institutional settings supporting the development of cultural life, therefore creating a link between the firm and the local milieu. All these features contribute to creating a strong and positive corporate image. Any new employee at Ubisoft exposed to friendly open office spaces, to a real creative atmosphere, and to permanent story-telling about the recent successes of the company, will quickly adopt the organizational culture of the company, through experiencing what can be called the "corporate routines", that provide common patterns of thinking, feeling and acting, and contribute to shaping the strategies, visions, and norms of all the employees. To some extent, Ubisoft is one of the flagships of the "creative industries", in which the clear imperative is to sustain creativity on a permanent basis. These reasons explain the choice we made to test our approach of organizational creativity in this firm. To a large extent, the empirical case which is presented in this article provides some evidence to Drazin et al. (1999) representation of creativity at the organizational level. However, we suggest moving beyond Drazin's et al. perspective (1999) which principally describes the role of two groups of actors: developers (technical) and the hierarchy (administrative), and considers that these actors generate organizational creativity by crisis or by friction. Our view is that organizational creativity results from the interplay between three types of routines: corporate routines, technical routines, and those innovative routines emerging from knowing communities. With respect to the specific cognitive activities of knowing communities, we consider that the creative contribution of these informal groups is not necessarily by crisis or friction but essentially by exploring new knowledge from the outside world. Thus, the innovative routines generated by the knowing communities can be viewed as ways of generating organizational creativity in addition to the tension and shift view of Drazin et al. (1999). The paper proceeds as follows. Section 1 addresses issues of methodological consideration, which is based on organizational ethnography and research-action. Section 2 introduces the analysis of the ways creative processes are managed at Ubisoft, and particularly analyzes the balance between creative formal and informal forces in the firm. Section 3 proposes an extended discussion on the relationships between routines and organizational creativity. Section 4 offers the conclusions.

1

Methodological considerations

This study started as the reinterpretation of empirical data originating from the organizational ethnography of a creative powerhouse: Ubisoft development studio in Montréal. In the case of Ubisoft, Simon (Simon 2002) was literally embedded in the firm for fourteen

The Routinization of Creativity · 125

months, in line with traditional approaches of organizational ethnography (Van M a a n e n 1979; Schwartzman 1993). This fieldwork led to more than 2 0 0 pages of ethnography, which are at the basis of this particular study (Simon 2002). The researcher followed u p on this w o r k and conducted regular research-action projects with the firm on a yearly basis f r o m 2 0 0 3 to 2 0 0 9 , mostly focusing on production a n d creativity issues (analyzes of productive processes and team dynamics), and subsequently developed an ongoing training p r o g r a m on the management of processes and creativity for the firm's middle and t o p managers. O n average, the researcher has been spending 2 to 3 days per m o n t h with managers and employees of the firm for the last 10 years. Consistent with the definition of organizational ethnography proposed by Rosen (1991), this piece of w o r k revealed the importance of connections between the different projects of the firms, between different modules within a specific project, and between the firm and external social groups a n d organizations. This study helped justifying the choice of observing three types of routines (from projects, f r o m communities and f r o m the organization) in our analysis. Such a diversity of heterogeneous sources of collective patterns is a key in understanding organizational creativity. O u r view is that considering only one category of routines (example the routines related to the management of projects) leads to a very restricted approach of creativity. Drawing on the "emergence" of this topic (Glaser/Strauss 1967; Glaser 1 9 7 8 , 1 9 9 2 ) , specific attention was given to the stories that came out of the interactions with the informants. These narratives regularly stressed: 1 ) T h e roles of individuals' behaviors and knowledge in the creative process; 2) The importance of multiple sources for learning-in-action inside the firm (mainly specialized modules f r o m different projects); 3) The role of social groups, in and outside the firm, and the role of organizations external to the firm, feeding the internal creative processes; 4) The role of organizational design, structure, and processes to orient, channel, and harness individual and collective creativity For this case study, the researcher clearly remained an outside observer (Watson 1999). A constant concern therefore was to faithfully render the observations and to allow the informants to fully express their understandings of their own activities, w i t h o u t introducing the researcher's personal opinions or biases (Geertz 1973, 1983). A second research allowed the researchers to engage in a continuous (and ongoing) set of research activities and research-action projects with firm f r o m 2 0 0 3 to this day. These activities mainly focused on the in-depth analysis of the integration of new content ideas through the implementation of formalized stage-gate process a n d Agile activities. The fine-grained sets of primary data were completed with further direct observations, sets of secondary data on existing and interacting communities in the organizations, as mentioned by the informants (public and corporate sources), as well as interviews with employees, managers and external partners of the firm. T h e data a n d information were finally gathered and compiled into a synthetic case study (as suggested by Eisenhardt 1989; Yin 1994; and discussed by Eisenhardt/Graebner 2007). The sets of data were analyzed through an induction/abduction process aiming at reconstructing knowledge flows and transformations inside the firm, and between firms a n d intermediary communities based on the narratives of the informants and secondary sources, on the one h a n d , and classifying those groups and organizations based on the

126 · Patrick Cohendet, Patrick Llerena, and Laurent Simon

literature on knowledge groups and situated creativity, on the other hand. In turn, some expected entities like diverse forms of knowing communities, including associative professional groups, and also less known elements like creative collectives were identified.

2

The formal management of the creative projects at Ubisoft

On organizational grounds, as for other cultural industries (DeFillippi/Arthur 1998; DeFillippi et al. 2007; Lampel et al. 2000), the management of complex video games projects is the result of a delicate balance. On the one side, a videogame project can be viewed as a complex multimedia artifact, relying on flexible and decentralized expertise, with significant artistic dimensions. It integrates contents from different origins quite parallel to the occupational roles described by Crosby (2000): game-design, programming, 2D and 3D arts, characters' animation, voices, with sound effects, soundtracks, noninteractive parts, and integration/localization. On the other side, the videogame project requires strict managerial attitude looking for the advantages of tight integration of these activities within time, cost and market constraints. The need to fine-tune the level of integration in such an industry is high: too strong an integration could lead to a permanent reduction in diversity and creativity; too loose an integration could lead to divergence, chaos, and inefficiency. To cope with these constraints, Ubisoft, as most of the videogame firms, has adopted a modular form of management of projects. The project is decomposed into relatively small autonomous organizational units (modules) to reduce complexity. Modularization leads to a structure, in which the modules integrate strongly interdependent tasks, while the interdependencies between the modules are weak (Sanchez/Mahoney 1996). The management of a given game is the result of an actualization process - from the idea/concept to the product - accomplished through the progressive divergent/convergent integration of pieces developed in the specialized modules. 2 If modularization within a given project leads to some disaggregation of the traditional form of hierarchical management, it remains on a theoretical basis under the control of the hierarchy of the firm. The modules are managed by specific departments or functions (coined as "métier" at Ubisoft, the French equivalent of specialized craft). In one project, modules are coordinated by the action of a project manager, hereafter "the producer", in charge of this mitigation and integration process. In strategic terms, the video-games industry operates in a very dynamic and globalized market, with an accelerated evolution of technologies and under strong competitive pressure. As with the movie industry, predicting a blockbuster and establishing the internal capabilities for success is not an exact science. To meet to those strategic conditions, large video-games producers are manage a significant portfolio of projects, sometimes more than 20 projects in parallel, where a project can mobilize up to 100 creative employees plus up to 200 quality testers. The dominant organizational form ranges from a balanced matrix structure to a pure project-based organization, with a dynamic equilibrium gravitating toward project-led design (for definitions, see Hobday 2000). 2

As the product is not exactly the same, the development of an animated movie follows the same inner logic, where creativity would be expressed at the level of the story (the theme and game-play in case of a game) and through daily problem-solving and technical incremental innovations. If Pixar's success is driven by unique stories, "a movie" insists Ted Catmull, Pixar's CEO, "contains literally tens of thousands of ideas" of artistic, technical, and managerial origins (2008).

The Routinization of Creativity • 1 2 7

The historical account of the evolution of organizational design at Ubisoft's Montréal studio gives an interesting overview of the challenges to be overcome to define and harness creative processes through organizational design, and especially of the difficulties in diffusing and replicating the best creative practices. Ubisoft's Montréal studio went through three structural designs in its 10 years of existence. From its origins in 1997 to 1999, the studio followed a formal balanced matrix structure, slightly dominated by specialized functions, with game-design and programming in tension as the two dominant "core" functions. During this period, the modules were under the responsibilities of the functions listed earlier. At this point in time, the studio managed a portfolio of less than 10 projects, including 4 major projects. Three "executive producers", with significant previous experience in the video-game, movie or media industry, were responsible for more or less 3 projects. Each project was operationally supervised by a "producer". New to the industry, most producers were recent graduates from business or technical/arts schools with only a few months or years of work experience. As the first round of projects got completed, in 1999, it appeared that the balance of power between producers and executive producers was shifting. Executive producers were busy dealing with the French headquarters or with licensing partners, mostly in the U.S., and spent most of their time outside of the company. Producers were the closest to the action. They would develop personal contacts with most of the employees, and also with their peers. The experience gained through the first round of projects helped them become more efficient and more autonomous. This new state of things received some validation with the arrival of a new CEO. At the end of 1999, the new CEO led a restructuring of the organization and moved towards a purely project-based structure. The rationale was to accelerate the games development process and to instill some competition between the different project teams. In the background was the idea that functions could not play a useful role as repositories of knowledge anymore, since knowledge would be essentially developed on a day to day ad-hoc basis by individuals and teams and, due to market characteristics, would become obsolete almost on the spot. The underlying principle was that, projects being somewhat independent entities with no real incentives for knowledge transfers or complementarities between them, pure competition between the teams would assure speed and efficiency. The existence of a cumulative process in learning was denied or at least significantly underestimated. At the same time, in 1999, the introduction of a Project Office was aimed at standardizing project management practices along the different projects in the organization. It then proved challenging to get the producers and specialists to endorse the recommendations, mostly inspired by a technocratic paradigm, and perceived as „invasive" by most producers. This initiative, viewed as a controlling and constraining stance driven by the hierarchy, generated significant resistance to the point where the hierarchy decided to concede. If the hierarchy recognized the poor performance of the office and dissolved it after a few years of experimentation, it still noticed the importance of formalizing the project management process in order to be able to follow up on the portfolio. One of the internal consultants working with the project office and former consultant in product development for the local aeronautic industry exposed higher management to a more systematic phasing and gating approach, and to the necessity of thinking about it as a learning process rather than a pure controlling and validating one. In fact, the hierarchy recognized the importance of cumulative knowledge and learning, and moreover the importance of the managers in supporting those learning processes.

128 • Patrick Cohendet, Patrick Llerena, and Laurent Simon

In this learning process, the role of the managers as for any project is twofold: First, to make sure that the "routinized corporate procedures" of projects are respected (stage gate procedures, respect of deadlines and cost considerations, as well as, when a project comes to an end, trying to capitalize and codify as much as possible the knowledge gained during the project, etc.). Again, after having tried to impose strict staging and gating procedures, the hierarchy adopted a quite distant stance towards production, and settled on a broad project management "script" left to the interpretation of the producers. The manager makes use of the ambivalence of the rules, exploiting the cognitive dimension of any rule which leaves room to interpretation (Reynaud 1996; Avadikyan et al. 2001). The expected stages were the following: a pre-conception stage to establish a unique "breakthrough" (the element of story and game-play that will make the game stand out from the competition); a conception phase to specify the storyline, the design of the characters, and the "look-and-feel" of the game; a prototyping phase to produce a playable demonstration "map"; and finally, the production phase, mainly re-interpreting the prototype with different specifications to follow up on the storyline. Even if this approach proved efficient in terms of the respect of deadlines, it is criticized for its own virtues: it aims at concentrating "thematic" creativity at the early stages of the process and discourages significant creativity at the later stages. Yet, in these later stages, incremental creativity occurs intensively in "problem-solving" modes in every module. Second, the manager has to undertake specific efforts to delineate, capture, reproduce or replicate the routines that result from the learning by doing processes achieved by the teams involved in a given project. As Winter and Szulaski underlined (2000: 3) "replication of routine is one important process by which organization re-utilize knowledge that is already in use". In the case of hierarchical teams, most of the learning activity results from a learning by doing process. This means that the cognitive construct of the group (the jargon, common grammar and codes, social norms, etc.) is only a by-product of the "main" objectives of the group which are essentially oriented towards coordination mechanisms or incentives determined by the hierarchy (to ensure the task is carried out efficiently, to reach the goal of the project on time, etc.). The cognitive construct that supports these routines is fragile in the sense that it has not been elaborated through the construction of the routine themselves. Most of the time, the hierarchy tries to absorb and replicate the routine of a given team with the global cognitive tools of the organisation (common language and representations) which are necessarily somewhat "distant" from the actual practice of the team (Cohendet et al. 2010). Though considered as essential for sustaining the creativity of the firm in the long run, the actual replication of knowledge accumulated in the routines and practices of the project teams is rather weak and unsatisfactory. This result is not specific to Ubisoft, but is rather generally shared by most firms (Winter/Szulanski 2000). So, only a small part of the knowledge acquired and gained from these formal groups is recuperated by the hierarchy. In fact, the gains of knowledge resulting from the projects can be divided into two main categories. First, most of the knowledge recuperated by the hierarchy is constituted of new ideas about managing projects that could improve the corporate routines. Such a category of knowledge contributes to fuelling the organizational slack of the organization. The organizational slack refers to Penrose (1959), who suggests that organizations always have some stock of unused, or underused resources (e.g., knowledge, relationships, reputation, managerial talent, etc.) that inevitably accumulate in the course of developing, producing and marketing any given product or service. In her view, these unexploited or underexploited productive resources are the primary factors

The Routinization of Creativity · 129

determining both the extent and direction of growth; growth being the d o m i n a n t motivation of firms, limited only by the administrative capacity of the organization. Second, another category of knowledge addresses actual creative issues and insights. As we will see further, this category of knowledge is essentially held by informal communities a n d fuels another type of slack that we call creative slack. Before exploring this new dimension, we turn to the conclusions derived by Ubisoft on the possibility of replication of the routines f r o m the projects. In terms of creative processes, formal "generative operations" (Witt 2009) w o u l d occur at the very first stage of the project. A "core t e a m " , composed of the " p r o d u c e r " , a lead game-designer, an artistic director, a script-writer, and a lead p r o g r a m m e r , w o r k s on defining the concept of the game, based on a minimal "brief" f r o m t o p management and creative direction f r o m the headquarters. They are aiming for an elusive " b r e a k t h r o u g h " idea that would differentiate the product f r o m the competition in terms of contents a n d narratives, and that would also m a k e the best use possible of the available technology. T h e activities w o u l d involve open exploration of the themes, scenario building, inspiration boards, rapid paper prototyping, mock-up development, and sometimes role-playing. The team itself performs regular "interpretative" operations, trying to express the ideas into a more formalized "script" that could be interpreted in the f o r m a t of a typical videogame (game-play, interactivity, progression of the level of difficulties and challenge, m a p s and halls, etc.). Then, every t w o or three weeks, the evolving concept is introduced to a committee composed of t o p management, creative directors and marketing representatives. Active a n d sometimes fierce debates occur during these meetings, focused on the " b r e a k t h r o u g h idea", as they put in tension "interpretative" operations (relevant and interesting ideas) and "evaluative" operations (valuable ideas, in terms of availability of resources, mastery of capacities to actualize the ideas, and market readiness / potential of revenues). In the informal realm, generative operations - expression of new ideas through unplanned intersections of different fields of knowledge - are much dispersed. They could occur through occasional conversations, and also through daily operations and problem-solving activities in and between specialized modules in the project. Informal interpretation and evaluation of ideas then mainly occur inside the specialized modules in projects. As these modules are focused on local often short-term problem-solving activities, the implementation of new ideas in practice, eventually evolving as routines, tend to remain very m u c h localized. For Ubisoft, the possibility to replicate the routines that result f r o m the learning by doing processes achieved by the teams involved in the videogames projects appeared as unsatisfactory. A historical evolution took place with the hiring of a new C E O for the Montreal Studio in 2005. A former producer, involved in one of the most successful blockbusters of the company, the new C E O was very sensitive to the " s u b o p t i m a l " exploitation of knowledge. One m a j o r irritant that he expressed was the fact that a good idea would generally not circulate f r o m a module to the other projects. H e emphasized the existence of a strong tendency to "reinvent the wheel" all a r o u n d the organization. A study conducted at this point would show this position as slightly exaggerated, as specific active units would already intensely foster knowledge circulation and exploration: the "communities of specialists" (Cohendet/Simon 2007). For the hierarchy, the recognition of the role of those communities as active units of knowledge creation a n d diffusion occurred through some research-action projects undertaken f r o m 2 0 0 3 to 2 0 0 8 . A few internal initiatives also revealed the intensity of the cognitive activities of those communities: some virtual forums, experimentally implemented

130 · Patrick Cohendet, Patrick Llerena, and Laurent Simon

in 2006, showed how some experts from different modules in different projects could build common knowledge bases through sharing and discussing. As the hierarchy, in fear of knowledge "leaks" toward the outside of the organization, decided to strictly moderate these forums in order to steer and focus the discussions, with the aim of developing formal, codified knowledge repositories, the attendance at these forums diminished significantly. Then when the hierarchy decided to stop these forums, employees almost started a revolution and asked for their return. This experiment led the hierarchy not to intervene too invasively in those processes and prompted another restructuring. As Ubisoft Montréal's hierarchy was experimenting in learning how to learn about routines, the next step was the implementation of "knowledge-through-people" managers, the "directeurs métier". From late 2008 to today, the hierarchy set up a new position of "directeur métier" (literally: craft director), with the mandate to "facilitate knowledge circulation, exploration and discussion, in order to optimize the implementation of best practices" (field interview). A "directeur métier" position would be created for each specialty, and one for project management, mostly concerned with middle managers in charge of specialized modules (struggling with a dual identity of technical specialist and people manager). Each "directeur métier" had previous experience at Ubisoft Montréal in his/her field of expertise. As the first tendency of the newly hired "directeurs métier" was to try and identify the "best practices", it rapidly appeared disappointing. First, it proved difficult if not impossible to actually reduce an apparent, yet specific success to a formalized and minimal set of rules, due to knowledge tacitness and situatedness. Second, the transmission of a formalized routine from one team to another would generally be costly and counter-productive, as it would detract the team from an already implemented or soon-to-be implemented routine. The option retained, then, appeared to refocus on the possibility of a micro-(re)structuring of routines, betting heavily on inter-individual knowledge transfers, supported by the cognitive work of communities of specialists. Not only did the "directeurs-métier" strongly support the diverse training programs as opportunities for people to "meet and share stories", they also essentially promoted the recombination of knowledge in several ways. First, they would organize formal sharing sessions on a regular basis. Those sessions would provide the opportunity to meet and share knowledge around a well-identified issue or challenge. During those sessions, the "directeur métier" would only play a coordinating role to make sure that everybody would have a chance to voice their concerns and ideas, and would refocus the discussions from time to time if necessary. Such a simple organizational learning device was widely appreciated by the participants, as formal opportunities to share actually were very scarce. At first, most producers were quite reluctant to let their employees attend these sessions. This changed dramatically when the producers came to realize the invaluable role these sessions played in increasing the capability of their respective teams. Employees were therefore strongly encouraged to participate in the sessions. Second, they would play an essential role in combining and recombining specialized teams for new projects, in order to fine-tune the combination of specific individuals in a module to support knowledge creation and creativity. In collaboration with the human resource department (HR) and the producers, the "directeurs métiers" would recommend specific individuals for specific modules of the project at hand. They would focus on previous

The Routinization of Creativity • 131

experience, autonomy, "soft" skills (ability to establish efficient and respectful working relationships with close peers, other modules, and supervisors) and the ability to share knowledge. The focus would aim at a rich mix of previous experiences and "hard skills" to ensure diversity (or require variety) and of so-called "soft skills" to support the integration of knowledge through social convergence toward a shared goal/vision. The guiding philosophy would be to make sure that in every module, at least one person would adopt a questioning, reflective stance towards the established routines, and would challenge work practices, while keeping a strong operational focus to avoid overly disturbing the pace and orientations of the modules. The choice of individuals would be negotiated, and sometimes harshly debated by these three actors: "directeur métier", HR representative and producer. Third, most of them would play a very active role as informal knowledge-brokers. Most "directeurs métier" would consistently "wander around" on the shop floor (according to Peters and Waterman's expression 1982). They would observe more or less formally the work of employees, discuss with them, adopting a listening and coaching stance. They would also informally channel contact between one employee faced with some technical challenge and another employee that may have some parts of the answer. At the time of this study, this organizational reform was still under way and not fully evaluated yet. The different "directeurs métier" have up to this point met with mixed success: as some of them are still trying to gain access to local knowledge through trust building activities, some others are already considered as very efficient "knowledge coaches" (field interview) and respected either as solution providers and/or as connection providers with other, more knowledgeable employees. Some very respected "directeurs métier" (gamedesign, 2D graphic arts, management) have been coined as "fair community players and knowledge-enhancers", stressing the importance of their connections with their respective communities of specialists (field interview). 3

The communities of specialists at Ubisoft as active units of diffusion and creation of routines

In the following part, we introduce communities of specialists and discuss their role in the exploration activities. As mentioned earlier, along the projects, in each module, a small group of specialized employees is in charge of specific elements of the project. Each of these groups should not be seen as an administrative or functional unit, but rather as a part of a community of specialists where members communicate regularly with each other about their practice and trade knowledge through informal cognitive spaces with more or less open boundaries, in a not-so-organized fashion. If indeed a part of their work is determined by the technology they are using (hardware and software), and is also defined by the mandate they received from their hierarchy, a major part is the result of their previous knowledge, experience and shared interpretation of their tasks with the other members of the community. Members of a given community share knowledge on an informal basis. They work in the same building, have lunch and go out together or they just chat online with peers in search of advice or technical solutions. They respect the social norms of their community that drive their behaviours and beliefs. Within a given community, knowledge is continuously exchanged and can circulate through the existence of a local language understandable only by the members. To a large extent, these workspaces are not fully monitored through the formal corporate process. They are not necessarily aligned with corporate goals and strategy. They are also somewhat

132 • Patrick Cohendet, Patrick Llerena, and Laurent Simon

disconnected from the daily pressure of producing an efficient output designed for a specific market purpose. These informal socio-cognitive spaces offer areas where people can meet, wander, confront ideas, build daring assumptions, and validate new creative forms. Members of each of the communities of specialists who are employees of Ubisoft also permanently communicate with the outside world, through global virtual platforms with specialists of the same focus of knowledge, sometimes even with members of competing firms who share the same interest for a given practice. They also directly interact through informal routes with communities of users, and have planted deep local roots in the 'creative city' of Montreal. Through this constant opening to the external world and the permanent search both for the best practices from outside the organization (exploitation activities) and for the new trends and styles in their domains (exploration activities), communities of specialists at Ubisoft are unique devices tapping into the external world to permanently bring useful knowledge and creative ideas to the firm. On a first level, those communities of specialists broadly fit the definition of communities of practice as their members use the same technical "jargon", share practical knowledge, and exchange tricks based on trial-and-error field experiences to increase their competence in a given field of knowledge (thus focusing on exploitation activities) (Lave/Wenger 1990; Brown/Duguid 1991). On a second level, they clearly also have an epistemic dimension, which means that they are focused on the production of new knowledge (exploration activities) (Cowan et al. 2000). As a result, most of the communities of specialists at Ubisoft have a dual dimension in the way they process knowledge, aiming both at exploration and exploitation. As such, by their mixed nature (internal/external; exploration/exploitation, communities of practices/epistemic communities), the communities of specialists are one essential intermediary level, allowing the passage from individually determined creative processes to the macro-dynamics of the firm. The managers at Ubisoft Montréal are fully aware that they cannot directly control or "possess" the creative works of the communities of specialists. Learning by "intrusion" and trying to control the cognitive functioning of the diverse communities would be doomed to failure. What the managers have implemented are integration forces in order to bind the creative units together for achieving effective production, timely delivery and ultimately commercial successes: the staging and gating process. It appears that the nature of the relationships and ties that bind the scattered communities together is generally not a unique platform (such as a given production line or a given modular structure). These communities exchange knowledge through different cognitive platforms (almost in line with the notion of 'ba' in the sense of Nonaka/Konno 1998) which are shaped or enacted by the hierarchy and which have some plasticity and flexibility to take different forms of coordination and may reconfigure through time. From the managers' point of view, this flexibility, partly framed by the generic "script" of the staging and gating process, is the key to the success of the alchemy of combining heterogeneous communities to reach a creative collective video-game product. To go further in this direction of research, we will then develop the idea that the integration forces put forward by the firm are not just for harnessing creative units: they also generate creative slacks for further expansion of creativity. Thus, creativity in this video-game firm seems to unfold through an attenuated, balanced organizational form which combines informal cognitive platforms disseminated in and outside an over-arching hierarchical structure built around an organizational culture and formal processes.

The Routinization of Creativity • 133

As underlined by DeFillippi et al. (2004) organizations attempting to solve the dilemma between creativity and efficiency may physically separate creative work units from more routine work units. "Such de-coupling presumably favours lateral thinking 'outside the box' that is free from the practices and conventions of the routine work of the organization" (Bilton/Leary 2002). However, the implementation of such a solution introduces a major risk of dissonance when creative inputs and creative work practices have to be introduced into the rest of the organization. The "dual" nature of the communities of specialists at Ubisoft (dual in the sense that they have both an exploration and an exploitation mode of activity) contributes to eliminate this risk and by-passes the need of decoupling/re-coupling the organization by providing a specific mode that guarantees the permanent connection between the routine work required in the management of projects, and the creative work done within communities. On the one side, members of a given community of specialists, as any employee of Ubisoft, have fully adopted the cultural global norms of the corporation that shape values and behaviours. They also have accumulated significant competences with regards to the respect of the managerial routinized procedures of achieving projects (or macro routines). To some extent in their current daily practice, these "corporate and projects" routines guide the way they achieve their creative cognitive interactions within their community of specialists. O n the other side, the creative construct made by each community through their constant interactions, inside and outside the firm, greatly contributes to bring novelty on a daily basis into the Ubisoft Montréal projects, in the form of new ideas, new trends, new practices or new codebooks. We argue here that two specific modes of exploration can be identified. A first mode is framed in a top-down way by the hierarchy, and mostly managed through project parameters, and sequential controlling and validation activities (the "gates"). It provides a vision of the output of the project which is only weakly defined and the details of which are left to the interpretation of the project team, actively working on its actualization. This can be defined as macro-routines of exploration. This macro-exploration is completed by multiple micro-exploration activities, essentially managed by members of communities of specialists involved in the modules of different projects. This micro-exploration occurs in a very autonomous, bottom-up and transversal manner, under the radar of top-down control. It plays an essential role in generating new routines and challenging existing ones. Through accretion and breakthrough, micro-creativity can challenge and partly reshape the macro-routines. This permanent connection provides opportunities for feedbacks between the micro creativity that emerged from the daily activities during the project, and the macro-creativity that is the expected output of the interplay of creative communities (and channelled by the hierarchy of the firm). The creativity of a project should not be confined to the macrocreativity designed by the hierarchy once and for all at the beginning of the project. A creative project should be able to incorporate new ideas, innovative suggestions, and all these micro-creative inputs that emerge. According to the managers of Ubisoft, one of the main drawbacks of the Stage-Gate process put forward to strictly control the timing of a project is precisely that this constraint - which excludes any significant feedback in terms of conception - may imply a loss of creativity by killing the micro-creative inputs. The dual identity mitigates this risk, by allowing permanent interactions between micro and macro creativity. In practice, this permanent interaction may lead to two main effects. First, it may happen that if a micro-creative idea that has emerged during a project appears as relevant, it can quickly circulate within the communities through regular exchanges,

134 · Patrick Cohendet, Patrick Llerena, and Laurent Simon

be improved and validated through these exchanges, and be introduced directly into the project. The managers of Ubisoft thus agree that there are cases w h e n the Stage-Gate rule should not fully apply. Second, micro-creative ideas that emerge during a project can be absorbed in the active memory of some communities of specialists, constituting a creative slack that will and/or can be used in further projects. The creative slack resulting f r o m the cognitive w o r k of the communities plays the role of an important reservoir of opportunities of innovative knowledge for the organization, and has guided to a large extent the innovativeness of the organisation. In line with Penrose's vision of a slack, the firm which has accumulated a creative slack is better prepared t h a n any other organisation to derive a benefit f r o m the creative potential of the slack. The creative slack is shaped by the culture of the firm and is essentially understandable through the jargon of the organisation. Because of these idiosyncrasies, it is much cheaper to valorise the slack within the firm which holds it than through any other organisation (including t h r o u g h any isolated communities). Some may argue that the creative slack appears as a cushion of redundancy which is costly to maintain. W e consider that the specific conditions of formation of the creative slack at Ubisoft, which rely on the functioning of quasi a u t o n o m o u s communities, which naturally take in charge at negligible costs the production and conservation of knowledge in their domain of specialisation, is a guarantee of the efficiency of maintaining the creative slack at low costs. T h e remarkable point is that the potential of the slack is diffused in the diverse communities of specialists of the firm that have memorised (thanks t o the knowledge brought by their members) parts of the learning during projects. The slack is not "possessed by the f i r m " . At best, if the firm correctly harnesses the creative w o r k of communities, it can access it at low costs. Although it is well k n o w n that organizations have extreme difficulties in memorizing w h a t was learnt during a project, the interest of communities with regard to this issue is that they rather easily memorise the routines practiced by their members. As Cohendet and Llerena (2003) suggested, "a routine that has naturally emerged within a community of economic agents sharing strong c o m m o n social n o r m s will probably have a much stronger p o w e r of replication than a routine which resulted f r o m the functioning of a t e m p o r a r y team project constituted f r o m heterogeneous agents w h o never met before". T h u s a creative slack has an ambivalent characteristic: it is a specific advantage for the firm, that is the only entity able to derive benefit f r o m it, but at the same time it is held, nurtured and maintained at rather low cost by the diverse communities of the organisation, sometimes even w i t h o u t explicit awareness of the managers. This creative slack m a y also be positively influenced by the existence of multiple projects, where each project acts as a source of knowledge creation and literally feeds the members of every knowing community involved in the project, indirectly increasing the creative potential of all communities and of the firm. The key question that follows logically is whether the organization will benefit in the f u t u r e f r o m this creative slack which is dispersed between the diverse communities and needs an integrative effort to be reassembled and p u t into collective creative practice? O u r view is that the answer depends to a large extent on the culture of the organization. In the case of Ubisoft, the strength of the corporate routines is likely to m a k e the answer a positive one. This rich empirical case offers a complex, yet readable picture of the roots/sources of organizational routines, and of their interplay in the creative process, especially when adopting a multilevel analysis leaving a significant place to meso-level activities (i.e. module teams, communities of practice, communities of specialists). In such a perspective, it appears that the "repertoire" of routines of the firm is a complex one, with three main components.

The Routinization of Creativity • 135

A first c o m p o n e n t is the set of "corporate routines" shared by all the employees of the c o m p a n y which, to a large extent encompasses the organizational cultural traits that shape the c o m m o n vision and general behaviors within the firm. These routines are particularly useful to assure the efficient functioning of the "exploitation activities" of the firm. A second c o m p o n e n t is the set of routines activated through the projects. These routines emerged through activities that can be considered as "guided exploration", with a rather constant control of the hierarchy that has the responsibility to guarantee the respect of project procedures of all sorts, but also to mitigate the risk of losing at the end of a given project, the knowledge and good practices gained and learned during the project. It is at this level that the organizational slack appears and allows for dynamic adjustments and growth in terms of size and n u m b e r of simultaneous projects. A third c o m p o n e n t is the set of routines activated by the communities of specialists. These routines emerged through cognitive interactions that can be considered as "open explor a t i o n " achieved by community members. It is there that the creative slack appears and is elaborated. It allows improving the innovativeness of the firm, its dynamic capabilities and de facto the routinization of creativity to be improved. The key point here is that this domain of the repertoire is not "possessed" by the firm, it can just be harnessed. In order to benefit f r o m these sources of creativity, the integration forces implemented by the managers of the firm to bind the creative units together for achieving commercial successes reveal a hybrid f o r m of project management which combines decentralized platforms with strict constraints on time, and a specific management of space that favours informal interactions. From this perspective, an important related question is to determine w h a t types of competences the firm should keep internally, and w h a t competences it should place in the external environment. This question echoes the idea suggested by Loasby (1991: 9) w h o distinguishes between the firm's internal and external organisation in differentiating the " k n o w l e d g e - h o w " (knowing h o w to d o things for yourself) and the "knowledge-that" (knowing h o w to get things done for you). The firm can thus maintain its direct capabilities internally and place its indirect capabilities in its external environment (Loasby 1998: 9). In the case of Ubisoft, it appears that the firm has delegated its open exploration capabilities to the diverse communities of specialists. Of course the " b e t " of the company is that the conditions of being able to harness the cognitive w o r k of these communities will constantly remain. Those routines which are largely learned outside of the boundaries of the firm will be channeled and exploited by the firm inside its boundaries. As suggested, for instance, by Kogut and Z a n d e r (1992), the role of the firm would then be to provide communities of specialists with "identity" (strategic orientations, corporate culture, and a sense of shared purpose), coordination through a generic script, and opportunities to learn through interactions in multiple platforms. In this regard, in line with Cunha (2005), managers should develop their capacity for "bricolage", ensuring the ongoing interplay of the communities of specialists of the firm.

4

Conclusion: beyond the black box and the mirror ball, a multilevel perspective on routines and organizational creativity

In this contribution we aimed at exploring the micro-context of the origin of routines a n d of their intimate link with organizational creativity. The empirical observations m a d e at Ubisoft confirmed that organizational creativity orchestrates, and is orchestrated by, continuous interactions between different types of routines, operating at different levels

136 · Patrick Cohendet, Patrick Llerena, and Laurent Simon

of the organization. These interactions (which could take different forms such as frictions, reshufflings, clashes or co-evolution) drive differences in learning and capability development within the organization. There are at the origin of the emergence and formation of new routines and improvements of existing routines to support the creative potential of the organization. The diversity of the routines which interact continuously in this creative perspective particularly matters: this diversity is based on the following levels (cf Table 1): 1) The routines issued from the projects carried out by the firm for which the context of work and coordination of specialized tasks is defined ex ante by the hierarchy of the firm. As we have seen, the power of replication of these routines is limited. As Winter and Szulaski (2000: 23) noted, "leveraging knowledge by replication of routines necessarily involves an investment in communication infrastructure, at least in the form of training in the organization's specialized language. Adequate command of language requires, however, substantial knowledge of organizational context: the link of information to action typically depends on the knowledge-based interpretive powers of individual human beings. Hence, the organizational use of symbolic information depends on the stocks of knowledge held by the participants: much of this is tacit and/or context dependent and it reflects the accumulation of local expertise. Under these circumstances, the creation of the requisite knowledge stocks at a new outlet can be accomplished only through a variety of costly processes that are substantially less straightforward than a standard notion of transmission of information would suggest". Though constrained and limited in scope the replication of knowledge from the projects contributes to fuel the organizational slack through new managerial ideas for managing projects recuperated by the hierarchy 2) The routines emerging from informal structures, the communities of specialists. They are at the origin of a wide range (ecology) of local, "situated" routines, partly determined by technology, partly constrained by the hierarchical script, partly socially constructed and interpreted, and tightly or loosely coupled to the very specific local milieu. The "open exploration" is achieved by the members inside the firm but also and sometime mainly outside, in a broader environment. The quality and the richness of the firm's environment become then crucial dimensions for the innovativeness of the firm. 3) The routines that are inherently related to the organizational specificities of the firm, which are essentially corporate routines as expression of patterns of thinking, feeling and acting in the corporate culture. They contribute at the organizational level to the broader organizational slack, and to the building of larger "script" of project management. The "script" is partly imposed by the inner logic of the industry, which defines a specific project-based structure with a phasing-and-gating process. The process is however regularly interpreted and reconfigured by the hierarchy. In essence they are the genes of collective identity, and take the shape of project management staging and gating principles and practices, framing collective divergent exploration and convergent production toward a creative goal. This set of routines permanently organizes a constant friction, abrasion, interactions between the routines issued from the formal and informal structures. It is from this platform that the "improvisational sparks necessary for igniting organizational innovation" (Brown/Duguid 1991: 54) come. From the exploration of this case, in this contribution, our aim was to try and address some main issues about the origin of routines in the organisation, including innovationrelated routines. The theoretical works on routines do insist on understanding "what is a routine", but devote little attention to the nature of the group of agents "who are

The Routinization of Creativity · 137

Table 1 Ecology of routines at Ubisoft Montréal Corporate routines

Routines from projects' structure and design

Routines generated by communities of specialists

Genes of the corporate culture "What we are" Collective identity and purpose Patterns of thinking; feeling and acting

New (mostly managerial) routines, that enrich the "gene pool" of the organization

Routines allowing for some radical exploration (in one specific field, technical or managerial)

Guided, focused exploration

Open-ended, autonomous exploration

Resulting from historical accretion Those routines are transmitted to newcomers in priority

Acquired through learning by doing processes, by teams designed by the hierarchy

Acquired through a deliberate cognitive process of the communities' members, to capitalize on specialized knowledge or to create new specialized knowledge

Shape strategies, visions, norms, focus and convergence

Drive new managerial practices

Genuine sources of novelty

"Owned" by the organization

Supposedly "owned" by the organization, yet difficult to effectively codify, master, and replicate at the end of a project

The organization does not "own" them, but can get access to them, facilitate their expression and transmission, and enact them

Allowing or not the creation and/or exploitation of the slacks

Sources of and location of organizational slacks

Sources of and location of creative slacks

involved in the routine". 3 In other words, the members of the organization involved in a routine are generally considered as anonymous. For instance, the well-known definition of routine given by Cohen et al. (1996: 683) - "A routine is an executable capability for repeated performance in some context that has been learned by an organization in response to selection pressures" - does not specify the type of groups of agents related to the routine. In fact, the evolutionary theory explicitly refers in many examples that it uses, to functional departments or project team as the organisational unit that support the routine, without making any differences between them. The project team is very often referred to, since one of the main issues with routine is its replication w h e n the project is over. W e consider that this view, which concerns the very core of the theory, raises t w o main problems. First, it is only partially relevant. Routines experienced in a functional group, in a project team, in a network of partners, in a community of a different kind, m a y be all different in terms of p o w e r of replication, of degree of inertia, of potential of search. The conditions of emergence of the routines drive to a large extent the modes of evolution of routines and the conditions of their replication for the organisation. These considerations should stand at the heart of the functioning of the knowledge-based firm. Second, the 3

Among the few exceptions, there are Feldman and Rafaeli (2002) and Feldman and Pentland (2005).

138 · Patrick Cohendet, Patrick Llerena, and Laurent Simon

classical evolutionary vision, by focusing on the sole organisational arrangements that are shaped by the hierarchy and that are driven by a pre-existing division of work, tends to leave aside the contribution of informal groups of the firm to the innovative process. Our contribution is to locate at the level of communities of specialists the emergence of routines and practices allowing for a creative slack. As a theoretical approach, the evolutionary perspective allows for a fair and faithful account for the resources creation of firms. However, it lets knowledge creation slip because it proceeds as if the firm possessed (hence the concept of a "repertoire") the knowledge incorporated into routines, and suggests that competence results from the selection of the best routines stored within the repertoire. However, some literature (see, for example, Cook/Brown 1999) shows that most of this knowledge is not accessible through a "given" repertoire, but is instead rooted in the practices of small active groups or "communities" which form the firm. The very nature of a routine (its capacity for replication, degree of inertia and potential for evolution) depends heavily on the group which implements it. Although evolutionary analysis offers a rich context of interpretation of the relations between the individual and collective efforts in the creation of resources through the concept of routine, it still lacks an analysis of the 'intermediate levels' which are the genuine catalysts of creativity in the organization. Once creative ideas have emerged, they require validation and testing: this is precisely why the cognitive role of informal groups at intermediate level is essential to be taken into account in any creative process. Such results open new agenda for research. Moving from considering routines and creativity in organizations as intrinsically complementary, the recent conceptual stream of literature on routines issued form the pioneering article from advances made by Feldman and Pentland (2003) viewing routines as generative systems with internal structures and dynamics, offers a theoretical framework to understand the dynamics of the interplay of routines and creativity. They suggest that the routine dynamics are based on the interaction between the ostensive (the abstract internal representations of routines in the minds of actors) and performative (the ongoing physical practices associated with carrying out the routine) dimensions of routines (Feldman/Pentland 2003: 101), and that the mechanisms for introducing changes to the routine performance as well as maintaining stability are already present in the build-up of a routine. In the perspective of the present study, more research and more patient observation and analyses of case studies could be carried out to precise the dynamics of a new formed routine resulting from an association of an ostensive part of a given routine and a performative part of another routine.

References Asheim, B.T., M. Gertler (2005), The Geography of Innovation: Regional Innovation Systems. Pp. 291-317 in: J. Fagerberg, D. Mowery, R. Nelson (eds.), The Oxford Handbook of Innovation. Oxford University Press, Oxford. Altshuller G.S. (1984), Creativity as an exact science: the theory of the solution of inventive problems. Gorgon and Breach Science Publishers. Avadikyan, Α., P. Llerena, M. Matt, A. Rozan, S. Wolff (2001), Organisational rules, codification and knowledge creation in inter-organisation cooperative agreements. Research Policy 30(9): 1443-1458. Becker, M. (2004), Organizational routines: A review of the literature. Industrial and Corporate Change 13(4): 643-678. Becker, M. (2005), The concept of routines - Some clarifications. Cambridge Journal of Economics 29(2): 249-262.

The Routinization of Creativity · 139

Becker, M . , N . Lazaric (eds) (2009), Organizational routines: Advancing empirical Research. Edward Elgar, Cheltenham UK. Bilton, Ch., R. Leary (2002), W h a t Can Managers D o For Creativity? Brokering Creativity in the Creative Industries. International Journal of Cultural Policy 8(1): 4 9 - 6 4 . Brown, T. (2009), Change by Design. H o w Design Thinking Transforms Organizations and Inspires Innovation. H a r p e r Business. Brown J.S., P. Duguid (1991), Organizational Learning and Communities-of-Practice: T o w a r d a Unified View of Working, Learning, and Innovation. Organization Science 2(1): 4 0 - 5 7 . Catmull, E. (2008), H o w Pixar fosters collective creativity. H a r v a r d Business Review 86(9): 64-72. Cohen, M.D., R. Burkhart, G. Dosi, M . Egidi, L. Marengo, M . Warglien, S.G. Winter (1996), Routines and other recurring action patterns of organizations: contemporary research issues. Industrial and Corporate Change 5(3): 6 5 3 - 6 9 8 . Cohendet, P., P. Llerena (2003), Routines and communities in the theory of the firm. Industrial and Corporate Change 12(3): 2 7 1 - 2 9 7 . Cohendet, P., P. Llerena (2005), A dual theory of the firm between transactions and competences: conceptual analysis and empirical considerations. Revue d'Économie Industrielle 111: 175-199. Cohendet, P., P. Llerena L. Simon (2010), T h e innovative firm: N e x u s of communities and creativity. Revue d'économie industrielle 1 2 9 - 1 3 0 : 1 3 9 - 1 7 0 . Cohendet, P., L. Simon (2007), Playing Across the Playground: Paradoxes of Knowledge Creation in the Video G a m e Industry. Journal of Organizational Behavior 28: 5 8 7 - 6 0 5 . C o o k , S.D.N., J.S. Brown (1999), Bridging epistemologies: the generative dance between organizational knowledge and organizational knowing. Organization Science 10(4): 3 8 1 - 4 0 0 . C o w a n , R., P.A. David, D. Foray (2000), The Explicit Economics of Knowledge Codification and Tacitness. Industrial and C o r p o r a t e Change 9: 2 1 1 - 2 5 3 . Crosby, O. (2000), Working So Others Can Play: Jobs in Video Game Development. Occupational Outlook Quarterly 44: 3 - 1 3 . C u n h a , M.P.E. (2005), Bricolage in Organizations. FEUNL W o r k i n g Paper N o . 4 7 4 . DeFillippi, R., M . Arthur (1998), Paradox in Project-Based Enterprise: T h e Case of Film Making. California M a n a g e m e n t Review 40(2): 1 2 5 - 1 3 9 . DeFillippi, R., G. Grabher, C. Jones (2007), Introduction to paradoxes of creativity: managerial and organizational challenges in the cultural economy. Journal of Organizational Behavior 28(5): 5 1 1 - 6 4 2 . Drazin, R., M.A. Glynn, R.K. Kazanjian (1999), Multilevel Theorizing about Creativity in Organizations: A Sensemaking Perspective. Academy of M a n a g e m e n t Review 24: 2 8 6 - 3 0 7 . Eisenhardt, K.M. (1989), Building theories f r o m case study research. Academy of M a n a g e m e n t Review 14: 5 3 2 - 5 0 . Eisenhardt, K.M., M.E. Graebner (2007), Theory building from cases: opportunities and challenges. Academy of M a n a g e m e n t Journal 50: 2 5 - 3 2 . Feldman, M.S., B.T. Pentland (2003), Reconceptualizing organizational routines as a source of flexibility and change. Administrative Science Quaterly 48: 9 4 - 1 1 3 . Feldman, M.S., A. Rafaeli (2002), Organizational routines as sources of connections and understanding. Journal of M a n a g e m e n t Studies 39(3): 3 0 9 - 3 3 1 . Félin, T., Ν . Foss (2004), Organizational routines: a skeptical look. DRUID Working paper 04-13. Fleming, L., S. Mingo, D. Cheng (2007), Collaborative Brokerage, Generative Creativity and Creative Success. Administrative Science Quarterly 52(3): 443—475. Florida, R. (2002), The Rise of the Creative Class. N e w York, Basic Books. Florida, R., C. Mellander, K. Stolarick (2008), Inside the Black Box of Regional Development - H u m a n Capital, the Creative Class and Tolerance. Journal of Economic Geography 8: 6 1 5 - 6 4 9 . Geertz, C. (1973), The Interpretation of Cultures. N e w York, Basic Books.

140 • Patrick Cohendet, Patrick Llerena, and Laurent Simon

Geertz, C. (1983), Local Knowledge: Further Essays in Interpretive Anthropology. 3rd ed. New York, Basic Books. Glaser, B. (1978), Theoretical Sensitivity: Advances in the methodology of Grounded Theory. Sociology Press. Glaser, Β. (1992), Basics of Grounded Theory Analysis. Mill Valley, Sociology Press. Glaser, Β., A. Strauss (1967), The Discovery of Grounded Theory: Strategies for Qualitative Research. New York, Aldine de Gruyter. Grabher, G. (2001), Ecologies of Creativity: the Village, the Group, and the Heterarchic Organisation of the British Advertising Industry. Environment Sc Planning A 33: 3 5 1 - 3 7 4 . Hargadon, A.B. (2002), Brokering Knowledge: Linking Learning and Innovation. Research in Organizational Behavior 2 4 : 4 1 - 8 5 . Hargadon, A.B., R.I. Sutton (1997), Technology Brokering and Innovation in a Product Development Firm. Administrative Science Quarterly 4 2 : 7 1 6 - 7 4 9 . Hatch, M.J. (1999), Exploring the Empty Spaces of Organizing: How Improvisational Jazz helps Redescribe Organizational Structure. Organization Studies 2 0 : 7 5 - 1 0 0 . Hatchuel, Α., Β. Weil (2009), Les nouveaux régimes de la conception: Langages, théories, métiers. Vuibert. Hobday, M . (2000), The Project-Based Organisation: An ideal form for managing complex products and systems? Research Policy 2 9 ( 7 - 8 ) : 8 7 1 - 8 9 3 . Jeffress, L.A. (ed.) (1951), Cerebral mechanisms in behavior: The Hixon Symposium. New York: John Wiley 8c Sons Kamoche, K.N., M.P. Cunha (2001), Minimal Structures: From Jazz Improvisation to Product Innovation. Organization Studies 22: 7 3 3 - 7 6 4 . Kogut, B., U. Zander (1992), Knowledge of the firm, combinative Capabilities, and the Replication of Technology. Organization Science 3: 3 8 3 - 3 9 7 . Lave, J., E. Wenger (1990), Situated Learning: Legitimate Periperal Participation. Cambridge, UK: Cambridge University Press. Lampel, J., T. Lant, L. Shamsie (2000), Balancing Act: Learning from Organizing Practices in Cultural Industries. Organization Science 11(3): 2 6 3 - 2 6 9 . Lippman, S.A., R.P. Rumelt (2003), A bargaining perspective on resource advantage. Strategic Management Journal 2 4 : 1 0 6 9 - 1 0 8 6 . Loasby, B.J. (1991), Equilibrium and Evolution. Manchester: Manchester University Press. Loasby, B.J. (1998), The organisation of capabilities. Journal of Economic Behaviour and Organisation 35(2): 1 3 9 - 1 6 0 . Martin, R. (2009), The Design of Business: Why Design Thinking is the Next Competitive Advantage. Harvard Business Press. Nelson, R.R., S. Winter (1982), An Evolutionary Theory of Economic Change. Belknap Press of Harvard University Press, Cambridge/MA. Nonaka, I. (1994), A Dynamic Theory of Organizational Knowledge Creation. Organization Science 5: 1 4 - 3 7 . Nonaka, I., N. Konno (1998), The Concept of 'Ba': Building a Foundation for Knowledge Creation. California Management Review 4 0 : 4 0 - 5 4 . Penrose, E. ( 1 9 5 9 - 1 9 9 5 ) , The Theory of the Growth of the Firm. 3rd ed. Oxford: Oxford University Press. Peters, T., J . Waterman (1982), In Search of Excellence. Warner Books, New York Pratt, A.C. (2010), Creative cities: Tensions within and between social, cultural and economic development a critical reading of the UK experience. City, Culture and Society 1: 1 3 - 2 0 . Reynaud, B. (1996), Types of rules, interpretation and collective dynamics: reflections on the introduction of a salary rule in a maintenance workshop. Industrial and Corporate Change 5(3): 6 9 9 - 7 2 1 . Rosen, M . (1991), Coming to Terms with the Field: Understanding and Doing Organizational Ethnography. Journal of Management Studies 28(1) 1 - 2 4 . Sanchez, R., J . T . Mahoney (1996), Modularity, flexibility, and knowledge management in product and organization design. Strategic Management Journal 17: 1 5 8 - 1 7 1 . Schwartzman, H.B. (1993), Ethnography in Organizations. London, Sage.

The Routinization of Creativity • 141

Scott, A.J. (2005), Creative Cities: Conceptual Issues and Policy Questions. OECD International Conference on City Competitiveness. Santa-Cruz de Tenerife, Spain: 3—4 March 2005. Simon, H.A. (1981 [1969]), The Sciences of the Artificial. 2nd. ed. MIT Press, Cambridge, Mass. Simon, L. (2002), Le management en univers ludique jouer et travailler chez Ubi Soft, une entreprise du multimédia à Montréal. Thèse HEC Montréal. Sosa, R., J.S. Gero (2003), Design and change: A model of situated creativity. Pp. 2 5 - 3 4 in: C. Bento, A. Cardosa, J.S. Gero (eds.), Approaches to Creativity in Artificial Intelligence and Cognitive Science. IJCAI03, Acapulco. Styhre, A. (2006), Organization Creativity and the Empiricist Image of Novelty. Creativity and Innovation Management 15(2): 143-149. Van Maanen, J. (eds.) (1979), Special Issue on Qualitative Methodology. Administrative Science Quarterly 24(4): 519-711. von Neumann, J. (1951), The general and logical theory of automata. Pp. 1-31 in: L.A. Jeffress (ed.), Cerebral mechanisms in Behavior: The Hixon Symposium. New York: John Wiley &C Sons. Watson, C.W. (eds.) (1999), Being there: Fieldwork in anthropology. London, Pluto Press. Woodman, R.W., J.E. Sawyer, R.W. Griffin (1993), Toward a Theory of Organizational Creativity. The Academy of Management Review 18(2): 293-321. Wierzbicki, A.P., Y. Nakamori (2006), Creative Space: Models of Creative Processes for Knowledge Civilization Age. Springer. Winter, S, G. Szulanski (2000), Replication of Organizational Routines: Conceptualizing the Exploitation of Knowledge Assets. Mimeo Wharton School, University of Pennsylvania. Witt, U. (2009), Propositions about novelty. Journal of Economic Behavior and Organization 70: 311-320. Yin, R.K. (1994), Case study research: Design and methods. Thousand Oaks, CA, Sage. Zack, M.H. (2000), Jazz improvisation and organizing: Once more from the top. Organization Science 11(2): 227-234. Corresponding author: Laurent Simon, Professor, Service de l'enseignement du management / Department of Management, Co-director, Mosaic, HEC Montréal, 3000 Chemin-de-la-CôteSainte-Catherine, Montréal (Québec), Canada H3T 2A7. [email protected] Patrick Cohendet, Professor, Service de l'enseignement des affaires internationales / Department of International Business, Co-director, Mosaic, HEC Montréal, 3000 Chemin-de-la-CôteSainte-Catherine, Montréal (Québec), Canada H3T 2A7. [email protected] Patrick Llerena, Professeur en Sciences Économiques / Professor in Economics, BETA - Université de Strasbourg, 61, Avenue de la Forêt Noire, 67085 Strasbourg Cedex, France. [email protected]

Jahrbücherf. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

The Evolution of Knowledge and Knowledge of Evolution Brian J. Loasby University of Stirling, Scotland JEL B41; B52; D8; L2; 0 3 Pattern-making; quasi-decomposability; stability and change; methodology; construction systems; uncertainty.

Summary Human knowledge is a human creation: we seek to make sense by creating patterns, which are tested in various ways and with differing degrees and kinds of rigour. For each individual cognition is a scarce resource, but different people can apply it in diverse ways and to diverse subjects: each application has its own range of convenience and its own dangers. Thus the growth of knowledge is an evolutionary process of trial and error, the rate and content of which depends on its organization, both conscious and unconscious. In seeking to develop knowledge methodological choices are unavoidable, but often unconscious. As Simon pointed out, all evolution, of life, economic and social systems, and ideas, depends on quasi-decomposability, the limits of which can never be fully anticipated. Thus uncertainty is inescapable - but it is a condition of innovation.

Introduction I begin with some questions. H o w can each of us make sense of our situation? In particular, why do some things change while others do not, and why do they sometimes switch between these categories? H o w do changes come about? H o w can a single person comprehend systems which are necessarily far more complex than any single brain, and especially in human societies, such as economic systems, which function through the interaction of many brains? Indeed, is 'comprehend' an appropriate term? H o w can we understand, and do the ways in which we understand themselves change? H o w can we draw on ways of understanding which have been developed by other people? Since 'making sense' is an active process of imposing understanding, how do the ways in which we try to understand affect the content and quality of our understanding? Methodology is not an optional extra. The twin foundational premises of this paper are, first, that our environment (and indeed the universe) exhibits multiple combinations of relative stability and change and, second, that there can be no procedure by which we can establish an understanding which can be proven to be permanently correct; the best we can hope for is an understanding based on conjectures which have been thoroughly tested, as Karl Popper argued. This seems to be generally agreed among scientists (see Ziman 1978, 2000b). We are therefore considering an overall process which is composed of many localized processes, and in which

The Evolution of Knowledge and Knowledge of Evolution · 143

at any time there will be stable understandings, each of which we may choose to call an equilibrium - which must be a partial equilibrium. An obvious label for both overall and localized processes is evolution, broadly defined as the emergence of novelty, much of which quickly disappears, but some of which survives for varying lengths of time. W e shall take a closer look at this definition later. We may first observe that there are t w o immediately obvious applications of some notions of evolution to economics. O n e is w h a t is n o w almost officially labelled evolutionary economics, and includes the study of change in both the content of goods and services and production processes and also in organizational forms and the processes of decision making, including concepts of business strategy. T h e other is the history of the subject itself, which also exhibits changes in subject-matter, technology and the modes of decision-making within the discipline. Understanding this history as an evolutionary process may be of value to contemporary evolutionary economists in t w o ways: it may be helpful to c o m p a r e analytical techniques, and they may also gain some insight into the management of relationships with economists w h o d o not think of themselves as evolutionary economists. Unlike most m o d e r n politicians (and many economists) Winston Churchill believed in the value of a historical perspective; he claimed that 'the further back you look, the further ahead you can see'. In considering the prospects for evolutionary economics it should not seem perverse to look back at some features of the history of economics to observe w h a t has helped a n d w h a t has hindered the development of w h a t can n o w be identified (if not precisely defined) as a significant field of both theoretical and empirical research. This will lead us to look further back to early attempts to m a k e sense of the h u m a n environment, and eventually to the most distant past to see h o w a process which permitted the continuous emergence and selective consolidation of novelty, together with selective elimination of established p h e n o m e n a and knowledge, could be possible. 1

The concept of knowledge

M y point of entry is the observation that evolutionary economists are increasingly emphasizing the significance of the growth of h u m a n knowledge in the development of economic systems. This p r o m p t e d the idea that improvements - or, more cautiously, changes - in our understanding of economic development m a y themselves serve as evidence a b o u t ways in which knowledge grows. I begin by noting that knowledge is a tricky concept for anyone seeking to produce a formal analysis. In his contribution to a volume which I recommend to evolutionary economists (Ziman 2000a), which was inspired by the simple analogy between technological innovation, with its 'many starters and very few finishers', and the ruthless selection between genetic mutations which is exhibited in biological evolution, James Fleck (2000: 248) seeks to explore 'the co-evolution of artefacts, knowledge and organization in technological innovation', but soon decides that a 'focus on knowledge makes the evolutionary problem very tough. It is very difficult to put boundaries a r o u n d an idea' (Fleck 2000: 255); he therefore confines his analysis to the relationship between artefacts and activities. The obvious parallel in standard economics is the general exclusion of the concept of knowledge in favour of information, the significance of which is never in d o u b t (as in information theory) because it is defined against a closed set of possibilities. This exclusion of anything currently unthought of is essential to the strict notion of rational choice and the standard concept of efficiency, both of which d e m a n d logical deduction f r o m precisely defined premises. However, as m a n y scientists have observed, the creation of conceptions hitherto unthought of is crucial to the evolution of knowl-

144 · Brian J. Loasby

edge. Here is a nice example of the influence of method on concepts, which indicates why methodology may be of practical importance. The inherent ambiguity of ideas certainly creates problems, which Fleck manages to avoid while still producing some useful knowledge, but it is crucial both to the growth of knowledge, within each individual and within groups, and to its application in economic systems, precisely because it provides the potential for novel connections. Indeed novel applications of knowledge are themselves major contributors to the growth of knowledge, and these are possible only because the boundaries of knowledge are not well defined. This continuous sequence of extending boundaries by making novel connections is the core of Penrose's (1959) account of how firms grow by finding new applications for their newlydeveloped capabilities and new opportunities for these new applications: exploiting these opportunities then creates new capabilities with their own ill-defined boundaries, thus facilitating the perception of new productive services which may generate ideas for new profit opportunities. In order to develop this account Penrose took care (following Schumpeter's example) to locate it in a theoretical space that she clearly distinguished from the theory of the firm in microeconomics, in which all production possibilities and market opportunities are public knowledge. (It is not only in economics that the imposition of a strict boundary is a condition of successful theoretical innovation.) Microeconomic theory, as Coase (1972) pointed out, does not recognize firms as organizations-precisely because, as Coase (1937) argued, firms emerge as pools of resources with a potential range of applications in circumstances which cannot yet be specified, and are therefore unfit for price theory. They have accordingly been selected out, in a classic evolutionary process. Changing theoretical locations is a rather common phenomenon in the growth of both academic and business knowledge; and large-scale relocations are sometimes labelled paradigm shifts. Such relocations may cause trouble not only for theorists but also for many practitioners; and firms may have great - and sometimes fatal - difficulties when faced with structural changes of relevant knowledge in technology or markets. This has become a focus of empirical research in evolutionary economics - an open frontier. Before we proceed any further, it is important to make clear that I am not suggesting that economics is a peculiar subject, or that economists are peculiar people. The point is that human knowledge is necessarily a human product and is therefore shaped by human characteristics. Let me cite a Nobel Prize Winner in the queen of the sciences, Werner Heisenberg. From the very start we are involved in the argument between nature and man in which science plays only a part, so that the common division of the world into subject and object, inner world and outer world, body and soul, is no longer adequate and leads us into difficulties. Thus even in science, the object of research is no longer nature itself, but man's investigation of nature. (Heisenberg 1958: 58) Even in physics, our knowledge is conditioned by the ways in which we attempt to develop it; and this is not simply a matter of the techniques that we use (though these are important) but of the fundamental, but not always explicit, assumptions that we make about the characteristics of the subject-matter, the questions to be posed and the kinds of answers

The Evolution of Knowledge and Knowledge of Evolution · 145

that are considered to be acceptable. C o n t e m p o r a r y physics would have been impossible w i t h o u t radical changes in these assumptions, and some further changes may be in prospect to help resolve current problems. There are three f u n d a m e n t a l obstacles to establishing an unquestionable basis for h u m a n knowledge. T w o are strictly logical; but we m a y begin with a practical consideration which has clear logical implications. The h u m a n brain, for all its remarkable capabilities, is very small in relation to the contents of the universe - a n d even to the sum of the contents of all h u m a n brains; therefore it c a n n o t simply a b s o r b correct information, but must select, simplify and compress in order to construct knowledge. The substantial energy consumption of the brain presents a classic economic problem for any omnipotent designer of humans, a problem which is ignored in models of rational choice; Simon's insistence on the crucial theoretical importance of this ultimate scarce resource seems to be commonly rejected (if it is not ignored) as an advocacy of unrigorous theory. Scientists and philosophers of science, however, take Simon's view. In his Presidential address to the Royal Society of Edinburgh, Sir Michael Atiyah (2008) pointed out that even sight, which has conspicuous priority over our other senses, does not record or reproduce w h a t is within the field of vision, but selects and excludes, thereby creating patterns; and some of these patterns are false. (For example, our very useful ability to judge distance by the apparent size of objects leaves us susceptible to illusions; but the exploitation of this susceptibility by painters using perspective has given us great art.) Atiyah emphasises that although some patterns - notably in mathematics - may have logical f o r m they are created not by logic but by imagination; and he identifies the creation of new patterns which provide fresh insight - not by deduction but by imagination - as the mathematician's greatest delight. It has very recently been suggested that the even greater priority given to sight in Neanderthal brains fatally impeded the capacity for pattern-making which creates intelligence - which if true is a notable example of opportunity costs. Even at the sensory level, our knowledge consists of created patterns; a n d though w e have some ability to modify our mental operations we cannot simply override or replace them. As soon as one becomes sensitized to references to patterns one finds frequent citations of pattern-creation in descriptions of h o w we (or our brains) function in scientific enquiry, business and everyday life; and in all these spheres we then exploit the uncertain boundaries of application for these patterns which Fleck f o u n d so inconvenient. Moreover rules for the proper use of this ability to create patterns cannot be derived f r o m unchallengeable first principles, although a search for foundations may be, and often has been, useful in providing guidance. It is time to look back. The t w o logical difficulties in creating new knowledge were both identified by David H u m e . First, although logic is invaluable in allowing us to derive conclusions f r o m premises - some of them far f r o m obvious w i t h o u t such systematic enquiry we c a n n o t deduce a novel theory f r o m evidence because reason can never produce a new idea (Hume 1878: 164). (Atiyah is familiar with H u m e ' s argument.) W e may note that rational choice, as defined in economics, is a purely logical operation; indeed economists regularly deduce the actions of their theoretical subjects f r o m the specification of their situation, although they believe that their o w n behaviour is not so strictly determined. (The psychologist George Kelly (1963) noted a similar contrast between psychologists' explanations of their own behaviour and their explanations of the behaviour of their h u m a n subjects.) Logic cannot produce innovation; patent law is quite clear a b o u t this. It follows, as Popper and Z i m a n have insisted, that the content of any particular innovation, in theory or technology, c a n n o t be predicted, although we may have plausible,

146 · Brian J. Loasby

though fallible, reasons for expecting innovations of a particular kind within a particular field and at a particular time, unlike the context-independent mutations of genetics. Schumpeter and Penrose follow Smith in recognising the importance of context in the generation as well as the selection of innovations. When a new theory has been created, we encounter Hume's second difficulty. No evidence, however abundant and consistent, can establish the truth of any general proposition beyond doubt, simply because we can never prove that there are no undiscovered counterinstances (Hume 1875: 33). Adam Smith, who we should remember had a close intellectual relationship with Hume, respected both propositions. In his 'History of Astronomy', which is essentially a history of the powerful motives and processes of pattern-creation and acceptance (and takes us much further back in human history), he observed that the evidence in favour of Newton's theory was so persuasive that he had himself been 'insensibly drawn in' to present its principles 'as if they were the real chains which Nature makes use of to bind together her several operations' (Smith [1795] 1980: 105). Nevertheless, he clearly states that, like all preceding theories, these principles were in fact the product of human imagination, and accepted so readily because of their appeal to the imagination of others. As he explicitly observes, they did not appeal to the imagination of those who objected that the postulated gravitational effects relied on action at a distance, for which Newton declined to offer any explanation, and which was disturbingly reminiscent of astrological influences. Smith therefore avoided Kant's great problem, which was simply this: how can we reconcile Hume's obviously correct proposition that the truth of Newton's theory cannot be proved with the inescapable conviction that the theory is true and irrefutable (Popper 1963: 190191)? That Kant's attempt to resolve this problem resulted in a major contribution to philosophy is a reminder that false problems can stimulate the growth of knowledge.

2

Frameworks for knowledge

Human knowledge is a human creation, and is necessarily provisional. How it is created, the effects of the means of creation on its content, and the ways in which it is confirmed, modified or rejected are therefore worth attention, especially for those who are explicitly in the business of knowledge creation, testing and transmission, and even more especially for those who take the development and application of knowledge as a subject of study. We cannot avoid making methodological choices; and these choices have consequences. To indicate both the generality of this problem and its potential importance I shall consider a spectacular example from another subject. Psychologists have often been worried about their scientific status; and between the wars they sought to emulate natural scientists by relying strictly on observation and experimentation. Psychology was to be the study of observable behaviour; and since mental processes were not observable they were to be excluded. What was left was the discovery of correlations between a class of actions and a class of preceding circumstances, without any investigation of the mechanisms by which these circumstances produced these actions - at a time when the search for the mechanisms which produced natural phenomena was the central preoccupation in theoretical physics. Psychologists nevertheless found it difficult to exclude causal language, notably in using such terms as stimulus and response which clearly imply causality. An indication of the opportunity costs for psychology of this aberration can be found in the introduction by Heinrich Klüver, an eminent psychologist, to Friedrich Hayek's theory of the mind which, although published as The Sensory Order in 1952, was based

The Evolution of Knowledge and Knowledge of Evolution · 147

on ideas formulated thirty years earlier. Hayek's first basic idea, in Kliiver's words, is that 'sensory perception must be regarded as an act of classification. W h a t w e perceive are never unique properties of individual objects, but always only properties which the objects have in c o m m o n with other objects. Perception is thus always an interpretation, the placing of something into one or several classes of objects' (Klüver, Introduction, in Hayek 1952: xviii). (This is a crucial principle for evolutionary economists.) H o w it is done is the second basic idea. 'The transmission of impulses f r o m neuron to neuron within the nervous system ... is ... conceived as constituting the a p p a r a t u s of classification' (Hayek 1952: 52); consequently '[t]he qualities which we attribute to the experienced objects are strictly speaking not properties of that object at all, but a set of relations by which our nervous system classifies them' (Hayek 1952: 143). Knowledge is constituted by selective connections. T h u s even as psychologists were resolutely ignoring any explanation of the mental processes which connected the actions of their experimental subjects to some feature of their circumstances they were also failing to recognise that their o w n observations were powerfully influenced by their classification systems. Meanwhile Hayek was laying a f o u n d a t i o n for neuroscience. W h y Hayek's theory matters for our study of the evolution of knowledge, and in particular of the significance of organization in both its development and its application, is the possibility of developing multiple classification systems in many fields, including academic communities and formal and informal organizations such as firms, as well as within individual brains. This perspective provides a basis for exploring both the sources of the ambiguities of which Fleck complained and also the potential which they offer for the growth of knowledge by the modification or replacement of familiar patterns - a domainspecific tendency to variation which produces evolution. Indeed Hayek's enquiry, which arose f r o m his early studies in psychology, including the dissection of brains, was motivated precisely by his realisation that scientific progress had not deepened our knowledge of the sensory order but had created new forms of order by rearranging its elements on different principles. This extraordinary demonstration of the unintended consequences of intelligent enquiry may have had a more p o w e r f u l influence on Hayek's views on the organization of h u m a n society than is c o m m o n l y recognized. As has been pointed out by others, the injunction to 'observe' is almost meaningless; the scientific tradition of requiring doctoral students to w o r k with their professor is intended to habituate them to particular modes of observation that rely on particular classification systems which are believed t o be appropriate to current problems in that particular field. This necessarily implies that many things will not be noticed; and the possibility that a few of them might be potentially significant creates occasional opportunities for those w h o have escaped or rejected the standard conditioning. Scientists create patterns which they find useful for their purposes - which are typically not the purposes which are reflected in the sensory patterns formed by the evolving brain. Explaining this disjunction was Hayek's motivation. The subject-matter of science is indeed m a n ' s investigation of nature, as Heisenberg wrote. This investigation is presumably enabled by the outcome of h u m a n evolution, but since it requires the creation of novel ideas it c a n n o t be genetically determined, any more than the innovations of Penrosian firms. The f u n d a m e n t a l significance of multiple classification systems for h u m a n knowledge and for h u m a n action has been recognised by others. T w o examples seem to be w o r t h attention in the context of this paper. Perhaps the most striking, because it was developed at very nearly the same time as Hayek's psychological studies, is Frank Knight's (1921) analysis. Knight contrasted the simplicity of decision-making in a world of certainty, or

148 • Brian J. Loasby

even in a world of well-defined probability distributions, where no more was required than logical deduction from a closed data set, with the problem of making sense of a world which cannot be adequately represented in this way. He points out that only in an environment of uncertainty is there any role for profit, the firm, or entrepreneurship. We may observe that all three concepts are absent from general equilibrium systems - and necessarily so if these are to be internally consistent. Knight comments that if nothing more than deduction is required then automata will suffice; indeed anything more is a waste of scarce resources. His assessment may be directly applied to the subsequentlycompleted Walrasian model of an economy which is precisely governed by a complete array of contingent contracts, which (as Frank H a h n pointed out) must be in place before the economy opens. Uncertainty - the impossibility of closing the system - is a precondition for the emergence of intelligence, which Knight implicitly regards as superior to rationality because it provides the context for it. To act intelligently we sort phenomena into categories according to some criterion of similarity which seems appropriate for our particular purpose. As Knight points out, there may be many different criteria which are appropriate for different purposes, and each should not be employed beyond its scope - the limits of which may not be obvious. This combination of multiple intersecting classification systems, each with an uncertain range of applications, explains the difficulty of putting boundaries round an idea which caused Fleck to make such a drastic revision to his own initial classification system. It also explains the potential for entrepreneurship, which transcends existing boundaries in order to create new combinations - a possibility which is excluded by a comprehensive and correct specification of everything. Knight's principle of intelligent behaviour justifies the use of mutually incompatible frameworks for different problems, even within one field, and for ways of organizing both economic systems and fields of study (including the sciences) in ways that permit this though it does not justify insouciance about such incompatibilities. Indeed it warns us that we may be led astray by using an inappropriate classification system, especially if it seems to have worked well hitherto: this has happened to many business enterprises, and also to many researchers in all fields of study. This is a major and, in my view, a necessary characteristic of evolutionary processes; I therefore believe that the analysis of such failures deserves more attention by evolutionary economists. It is easy to see that this conception of classification systems, in which perceptions are linked to actions within particular domains, fits naturally into an analysis which is framed in terms of partial equilibria. It also supports a theory of development which rests on specialisation between domains and variation within each. Such was Alfred Marshall's theoretical system, and though the specific evidence of a conscious connection is quite modest it is easy to see - once Raffaelli (2003) had pointed it out - a correspondence between this theory and his early and elaborate formulation of a model in which 'machines' developed effective routines by forming domain-specific linkages between 'ideas of impressions' and 'ideas of action' by trial and error. In view of Marshall's attraction to Darwin's idea of evolution we may find especial interest in his conclusion that machines of identical design, placed in different environments, would develop different patterns of behaviour (Marshall 1994) - an idea to which we shall return. Marshall's model, which is remarkably elaborate, should not be conflated with Hayek's; but (in Knight's terminology) they are similar in certain respects which are relevant in the present context. What does deserve attention is that it produces an outcome which is crucial to Adam Smith's theory of development ([1776] 1976: 29): differences in human abilities are often the consequence rather

The Evolution of Knowledge and Knowledge of Evolution · 149

t h a n the cause of the division of labour. For Smith and Marshall, increasing return is not a property of a production function but a process which modifies production functions. The classic exposition is by Ally η Young (1928). The second advocate of multiple f r a m e w o r k s I have chosen is the American psychologist George Kelly, w h o called this 'constructive alternativism'(Kelly 1963: 8). H e was at least in part reacting against the conceptual basis and inadequate results of the behavioural psychology which we noted earlier. H e observed that this implicitly treated psychologists and their experimental subjects as different species: the scientist, w h o possessed the skills and knowledge necessary to predict and control a class of p h e n o m e n a , a n d other h u m a n organisms whose behaviour was governed by various impulses. W h y not, he asked, give them equal status as people w h o are seeking to make sense by applying interpretative f r a m e w o r k s to their situations as they perceived them? (Kelly 1963: 5). H e even suggested that experimental psychologists were implicitly challenging their subjects to discover the interpretative f r a m e w o r k that was being used by the experimenter (Kelly 1963: 77). Rational choice theorists m a k e a similar implicit claim t o superiority over economic agents in their own ability to go beyond rationality and predictability by creating new models, which agents c a n n o t be allowed to d o if their behaviour is to be predictable. Kelly (1963: 6-7) assumes that the universe exists by happening: w h a t we have to study are processes by which we attempt to develop and apply our understanding of the particular small part of the universe in which each of us is located. These activities drive the evolution of knowledge, technology and organizational structures, all of which are constituted by classification systems with uncertain boundaries. For this to be possible, an essential condition is that, although the universe is a single system, so that every element is ultimately linked to every other element, these connections differ enormously in their strength and the time for the connections to take effect. For clinical psychologists, this created the possibility of changing the patients' behaviour in relation to many - though not all - features of their environment. Kelly's conception has obvious similarities to Simon's ([1962] 1969) architecture of complexity (to which w e shall return in the next-but-one section), not least in its implications for substantial but fallible intelligence. Both give extensive scope for the isolation of subsystems, with a w a r n i n g that the serviceable patterns that we learn within a particular subsystem are always liable to be overridden because the isolation on which they depend may erode over time or be disrupted w i t h o u t warning. A direct implication for economists is that their analysable subsystems cannot be proper subsets of a general equilibrium system; our brains are incapable of formulating the correct full specification of such a system, and any specification that we impose may lead us astray. Partial, not general equilibrium, is the better basis for explaining h o w a complex system can w o r k , and h o w it can break d o w n (Raffaelli 2003). Kelly's (1963: 9-11) f u n d a m e n t a l proposition is that people (including scientists) invent patterns which they use to 'construe the replication of events'; they m a y differ in their criteria for a successful construction, and in their ability first to achieve and then to maintain it. Each theory has a limited range of convenience (including, as he is careful to point out, the theory which he is proposing), although these limits c a n n o t be k n o w n ; moreover it m a y be particularly effective in certain parts of this range. H e argues that there are always, in principle, alternative construction systems which are conceivable - if not yet conceived; so if one pattern is n o longer satisfactory, we can look for another. T h e gradual supersession of the sensory order by the physical order, for scientific purposes but not for most ordinary living, is perhaps the most pervasive illustration of this process.

150 · Brian J. Loasby

Evolutionary economists may be particularly interested both in the process by which a new emphasis by well-established companies on innovation led some of them to m a k e radical changes to their styles of management, and in the interpretation of their behaviour based on a conceptual distinction between 'mechanistic' and 'organic' systems which was developed by a sociologist in order to analyse their experiences (Burns/Stalker 1961). The concept of 'constructive alternativism' was particularly relevant to Kelly's w o r k as a clinical psychologist: the most effective f o r m of treatment for patients might be to help them to look at things in a different way. This may also be the most effective way for a scientist or entrepreneur to resolve an intractable problem, as Smith demonstrated in his 'History of Astronomy'. As Kelly noted, however, patients may find it very difficult to amend a crucial part of a construction system if this is closely linked to another part which seems indispensable. This is not just a problem for patients. Smith noted it as a general h u m a n p h e n o m e n o n , and it may be observed in m a j o r changes in scientific fields. Organizations which rely o n a closely-connected network of construction systems may likewise find it very hard to convert to another network even when it is obvious that the established pattern is n o longer working. The conversion of D u Pont f r o m a functional to a product-based structure was actually impeded by the directors' faith in contemporary organization theory. W e should not be surprised that most organizations eventually disappear, or that reform of the financial system is so problematic. We might consider whether the dissolution of organizations should be made easier, as Drucker (1969: 293) seems t o imply. It certainly seems w o r t h considering whether any scheme for closer integration between complex systems to overcome frictions or failures risks generating problems far worse that those it is intended to resolve. These issues deserve more attention in industrial economics; evolutionary economists m a y provide it.

3

Construction systems in economics

Economists have often tried to emulate w h a t they think are the principles and methods of science. In the 1930s and 1940s one of the manifestations of this desire was the attempt to maintain a clear distinction between positive a n d normative reasoning, and in particular to produce policy advice which avoided any value judgements. Greater attention to logic would have been helpful here. W h a t were proclaimed at the time as m a j o r products of this endeavour were the twin foundational propositions of welfare economics: every perfectly competitive equilibrium is a Pareto o p t i m u m , and every Pareto o p t i m u m can be realised by a perfectly competitive equilibrium. Of course, these propositions relate simply to allocative efficiency, and explicitly exclude any consideration of wealth or income distribution, which requires value judgements. They also exclude any consideration of the effects of different forms of economic organization o n the prospects of enlarging the possibility set - and necessarily so, for the process of innovation is incompatible with perfect competition. The u n f o r t u n a t e consequences of closed-system reasoning may be clearly displayed in a n observation by Paul Samuelson, perhaps the cleverest economist of the twentieth century, a n d w h o w a s certainly capable of good sense on some occasions. 'Increasing returns is the enemy of perfect competition. And therefore it is the enemy of the optimality conditions that perfect competition can ensure' (Samuelson 1967: 39). Samuelson's logic is correct: but the perception of increasing returns as a threat to a n obviously desirable state of affairs is simply a consequence of the desire of economists, like psychologists in the interwar

The Evolution of Knowledge and Knowledge of Evolution · 151

years, to attain a supposedly scientific precision - in contrast to the often-lamented imprecision of Alfred Marshall. (It is an ironic c o m m e n t on the notion of scientific precision that the liberation of economics f r o m psychology by developing a pure logic of choice coincided quite closely with the psychologists' search for scientific precision by excluding the concept of choice f r o m their theories of behaviour.) The powerful effects of highly focussed ways of developing a theoretical system which is often presented as a realisation of the insights of A d a m Smith included the total disregard of Smith's exposition of the welfare gains that could be delivered by organizational arrangements which p r o m o t e d the growth of knowledge. As George Richardson (1975: 353) observed, this exaltation of perfect competition 'might reasonably be regarded as a denial of Smith's central principle erected into a system of political economy'. H o w did this happen? Whereas most economists and most psychologists through much of the last century sought to avoid the contamination of their systems by mental a n d emotional processes (with Schumpeter's theory of economic development as a notable exception), Smith had incorporated these processes both into his theory of the growth of knowledge (Smith [1795] 1980), which has obvious resemblances to both Knight's a n d Kelly's ideas a b o u t the construction of domain-appropriate systems, and also into his theory of h u m a n interaction. T h e latter included the readiness to a d o p t apparently successful practices w i t h o u t understanding why they were successful (Smith [1759] 1976a); this powerful cognitive economy permitted far more rapid diffusion t h a n is possible by the inheritance of superior genetic material. Smith's ([1776] 1976b) theoretical system explained h o w the division of labour fostered the development of new knowledge, which p r o m o t e d economic growth and the development of new markets; these then created opportunities for further division of labour. This is economic dynamics, as subsequently developed by Penrose, a n d Smith gave considerable thought to its organization and stability. Marshall adopted Smith's conception, and supplied a definition of 'increasing returns' which explicitly included changes in organization (Marshall 1920: 318). Allyn Young (1928) subsequently provided the classic exposition of increasing returns, in Marshall's sense, as the key to economic progress. Different ways of organizing activities change the boundaries of existing ideas and so may generate different ideas, some of which can be realized only by further organizational change, which may produce additional ideas - as in Penrose's sequence. This combination has an obvious counterpart in evolutionary biology, but differs in combining chance discoveries with conscious thought and direction. An essential feature of this theoretical system is that the activities of economic agents change the data of the economy, often in ways which c a n n o t in principle be predicted because they rely on novel classifications and novel connections between them. T h a t scientific discovery is likewise unpredictable, for the same reason, has been argued by Popper, by John Z i m a n (1978, 2000b) and with a b u n d a n t illustrations in a lecture by Sir J o h n Meurig T h o m a s (2007). This process is indeed, as Samuelson declared, the enemy of both perfect competition and the supposedly scientific welfare ideal; but it has a natural h o m e in an economics whose 'central idea ...even when its Foundations alone are under discussion, must be that of 'living force and movement' (Marshall 1920: xv) - which is Smith's perspective. T o preserve the particular concept of equilibrium which had been developed in the process of providing a theoretical solution to the intellectual problem of co-ordination in a fully-specified economy, increasing returns can be n o more than a property of an unchanging production function.

152 • Brian J. Loasby

Whereas Smith and Marshall wished to encompass co-ordination and development within a single theoretical system, subsequent economists practised a division of labour between the two, concentrating first on the relatively straightforward co-ordination problem but without realising how difficult their method of dealing with that problem would make it to incorporate economic development within their theoretical structure. The problem is illustrated by what is probably the greatest intellectual achievement of this research programme, the completion of the Walrasian model by defining goods not only by their intrinsic properties but also by their location, date, and the state of the world at that date, in order to allow every agent to optimise with respect to all possible futures. The equilibrium of such a system already incorporates not only everything that might happen but also whenever and wherever it might happen. This is essential to the internal consistency of the model; if all agents are to make rational choices there can be no ambiguity and no surprises. Agents are equipped with well-specified preference functions and probability distributions over all contingencies for every date; but they cannot modify the range and certainly must never conceive a new idea. (As is rarely made explicit, the model must also incorporate all agents who will be active at all future dates.) Within this theoretical system what Schumpeter called development from within the economic system is therefore strictly unthinkable. Schumpeter's own response to the inherent limitations of this intellectual programme (which he admired as an intellectual achievement) was to perceive them as a productive opportunity for himself. By ignoring the orthodox concern with efficiency and optimality which he (like Smith and Marshall) thought were ultimately of less importance than growth through innovation, and the methods of marginal analysis which seemed appropriate for that concern, he was free to explore the importance of system disruption, which was inherently unpredictable by economic agents or economic analysts because it resulted from novel conceptions. Almost as a side-issue, he commented that although rational choice was always a fiction it was a good predictor precisely when people were not consciously choosing at all, but following routines which had proved suitable for familiar contingencies (Schumpeter 1934: 80). What is interesting from our present perspective is that this interpretation of orderly behaviour is a close match to the psychologists' orthodoxy that we noted earlier. Rational choice as defined in orthodox economics is a purely logical operation, and therefore, in Niels Bohr's judgement, excludes thinking (Frisch 1979: 95), as did the psychologists' model. But as Schumpeter realised, the predominance of routine behaviour within an economic system was necessary to provide the reliable data for entrepreneurial planning. That for each individual, the predominance of established connections within the brain was necessary to release cognitive capacity for search was a crucial feature of Marshall's (1994) early model of the brain. What if Schumpeter had known of this model?

4

The precondition of evolution

The central paradox in the development of human knowledge was stated by Hayek ( 1952: 185): 'any apparatus of classification must possess a structure of a higher degree of complexity than is possessed by the objects which it classifies'. We have noted earlier that the human brain cannot match the complexity of its visual environment despite the high priority which this receives, but has to impose its own fallible order within boundaries which are inherently unknowable. We then observed that human knowledge has grown by the imposition of a great variety of created patterns on many phenomena - some of

The Evolution of Knowledge and Knowledge of Evolution · 153

which may even be constituted as phenomena by the patterns that we impose. We should not be surprised that this highly disaggregated structure of knowledge sometimes fails us; the problem is w h y it should ever succeed? Indeed in respect of economic systems the authorised wisdom is that only a single comprehensive model can be relied on. Unfortunately for this view, n o single comprehensive model can be contained within a h u m a n brain. The answer to this puzzle was provided by Herbert Simon ([1962] 1969): w h a t we can hope to understand, at least to a useful degree, are quasi-decomposable systems. In coping with a multi-level system we can, most of the time, m a k e progress by focussing on the interactions at one level, provided that this level interacts with the level above predominantly as a unit and not through its c o m p o n e n t elements, and that each of these c o m p o n e n t elements also functions within this system predominantly as a unity, even though this unity is the product of interactions between its o w n constituent elements. Simon not only argued that the assumption of quasi-decomposability was a precondition of h u m a n knowledge, but that the very existence of a complex universe was overwhelming evidence for its quasi-decomposability - simply because if every element were directly connected to every other, then failure anywhere within the system could precipitate collapse. H e illustrated his argument with the fable of t w o watchmakers, whose work was frequently interrupted: the one with the modular design prospered while the one with an intricate network of connections did not. Simon's argument is neatly counterpointed by that of the highly-distinguished French zoologist and palaeontologist Georges Cuvier ( 1 7 6 9 - 1 8 3 2 ) , a pioneer in the study of fossils. H e argued that since every species was a fully-integrated system, there was n o possibility of any modification; therefore any shock which disrupted its relationship with its environment led to extinction, as was evidenced by the extensive fossil record of failed species. The evolution of species was simply impossible. In seeking to develop a coherent model of general equilibrium economists have replicated Cuvier's argument, which inevitably leads to the same conclusion: because all future possibilities must be incorporated in a single general equilibrium, no f u t u r e adjustment can be permitted. I believe that we should assert clearly that the first principle of any evolutionary theory is that evolution requires decomposability. T h a t this is true both for the organization of scientific enquiry and for the possibility of developing viable theories a b o u t scientific p h e n o m e n a is emphasised by Z i m a n (2000b: 326): 'The assembly of primary entities into more or less distinct c o m p o u n d entities that can interact as wholes ... makes scientific research possible'. It also makes new knowledge unpredictable. I next wish to emphasise that, by Simon's argument, the evolutionary process which we attempt to study started with the origins of the universe. Fortunately his argument also assures us that we d o not need to be professional physicists, chemists a n d biologists before we are fit to study evolutionary economics. Biological evolution is a relatively late example, and it requires not only substantial - though not total - independence of the particular activities of biological organisms f r o m the details of their chemical structure, but also substantial independence of chemical c o m p o u n d s f r o m the atomic structure of their elements. It is quasi-decomposability which makes pattern recognition a n d pattern creation possible, and these are deeply embedded in scientific practice (Ziman 2 0 0 0 b : 120). Within economic systems the principle of quasi-decomposability allows the development of locally-

154 · Brian J. Loasby

appropriate construction systems at three levels: within individual brains, among people engaged in similar activities as members of different organizations (exemplified in Marshall's comments on the value of industrial districts in securing and diffusing the benefits of the tendency to variation among firms in the same trade), and between different trades, including the different occupations to be found in any large organization. The particular arrangements which may be necessary to cope with the limits to decomposability imposed by the need to co-ordinate activities which, though dissimilar, are nevertheless closely complementary, have been examined by Richardson (1972). The development of knowledge within Penrosian firms is both driven by and drives the schemes of decomposition and the (usually sparse) connections between them. Though that principle entails a rejection of general equilibrium thinking, it does not require the rejection of equilibrium in its literal sense of balance: this may be observed in any pattern and procedure which is in regular use. However, what distinguishes this class of equilibria from the conventional concept of a relationship between agents from which none of them has an incentive to depart, is that each equilibrium is maintained by a continuing process. As Kelly argued, the world exists by happening, and so does each of the classification systems that Hayek, Knight and Kelly have written about. When a previously satisfactory process is perceived to have become less successful there is a natural stimulus for insiders or outsiders to imagine ways to modify or replace it. However, because cognition is such a scarce resource, as Hayek and Simon have highlighted in very different (but perhaps complementary) ways, modification or replacement, even within a substantially decomposed system, can receive adequate attention only if most systems are working effectively. Increasing claims on cognitive resources and reduced decomposability can interact with disastrous effect. Because of the factors which permit the development of knowledge, and of economic systems which both produce and use knowledge, systemic breakdown is always a possibility, for individuals and for groups. The evidence is in history and in the present. General equilibrium theorists are right to claim that without such an equilibrium we cannot guarantee that there will be no disorder. Evolutionary economists should not disagree. However they can claim that within such an equilibrium there can be no evolution. What they should perhaps add is that the potential for large-scale breakdowns can probably be reduced by preserving both decomposability and variety within the components of the economic system and the systems of knowledge.

5

Frontiers

I do not think it would be helpful to propose a comprehensive programme for evolutionary economics in the next twenty or thirty years, for reasons which may be deduced from this paper; but I can suggest some topics and ways of thinking which I believe deserve some emphasis. One topic which I believe has now been sufficiently discussed is 'Universal Darwinism'. However there has been a fairly recent development in understanding the role of genes which is worth some attention: an evidence-based critique of genetic determinism which proposed 'one-to-one relationships between specific genes (or specific sets of genes) and complex higher level behaviours like altruism, aggression, intelligence, spatial cognition, or language' (Karmiloff-Smith 2000: 526). This principle, which had been repeatedly endorsed by Steven Pinker, was judged sufficiently misleading to become the theme of

The Evolution of Knowledge and Knowledge of Evolution • 155

the Special Lecture to mark the centenary of the British Psychological Society, delivered by Annette Karmiloff-Smith, who drew on her own and others' clinical experience. A key exhibit in this critique was the Williams syndrome, a genetic disorder which involves the deletion of 17 genes and is strongly associated with a particular set of effects. However, among these effects are 'atypical interactions between brain regions. The WS brain is not a normal brain with parts intact and parts impaired' (Karmiloff-Smith 2002: 528). Karmiloff-Smith focuses on the contrast between the severe impairment of spatial skills and the apparently preserved skills in facial recognition, from which Pinker deduced that the Williams syndrome has no effect on facial recognition. However, detailed study of patients clearly demonstrated that their facial recognition was based not on the normal method of recognising configurations but on scanning individual features, a process which has been definitively associated with other elements of the brain. We may pause to note, first that configuration (pattern-recognition) is a lower-cost method, but with two opportunity costs: substantial difficulties in providing a detailed description even of a familiar face (we know it when we see it), and in recognising inverted faces, for which identification by feature is the standard procedure. Karmiloff-Smith's general argument, developed in this lecture with a range of evidence, is that at least a significant proportion of the genes which guide the development of the human brain are carriers of what students of Penrose would call capabilities, the particular application of which is influenced by the environment. Therefore we need to study 'how genes are expressed through development' (Karmiloff-Smith (2001: 540) like capabilities. Of particular significance for our purpose is her contrast between the initial alternative possibilities in an infant brain and the often-formidable difficulties of reconfiguring a well-established cognitive procedure. We may note a similar contrast in leaf-cutter ants, in which there are sharp distinctions in both size and behaviour between soldiers, large workers who are foragers, and small workers who feed the infants. Since these are all sisters, the differences cannot be innate; they are produced by differences in the length of the feeding period. Here too the genetic material provides a range of possibilities, but the outcomes of development are not reversible. This principle of specialised applications of genetic potential seems to explain Adam Smith's observation that 'the very different genius which appears to distinguish men of different professions, when grown up to maturity, is not upon many occasions so much the cause, as the effect of the division of labour' (Smith ([1776] 1976: 28). It is crucially important for both economic and scientific development that these differences between people, unlike the differences produced by the differentiated rearing of leafcutter ants, are not embodied in narrowly-defined routines; nevertheless (as Smith recognised) it is typically much more difficult, even for economists, to make a substantial change to skill sets, or ways of thinking, than to acquire the first skill. This view of the role of genes reinforces the argument that the familiar Darwinian principle which denies any influence of the selection environment on the characteristics of genetic mutation, in contrast to its influence on the survival of mutations, is not especially relevant to evolutionary economics. Of much greater importance is a cluster of propositions about environmental influences. First, the perception of problems or opportunities motivates search; second, this perception deserves an explanation in each specific case; third, search will be undertaken only if someone can both conceive and implement a research strategy; and finally, the outcome of this search is necessarily unpredictable. Almost all ideas fail to work; outright failure is common (especially at the individual level), and quite often a new discovery is irrelevant to the search objective. The prime

156 · Brian J. Loasby

result of a research programme precisely targeted on the discovery of pesticides which were both highly specific and persistent was the discovery of one which is universal but inactivated by contact with the soil. This changed the role of research targets within the company. Since the overwhelming impression of our universe is diversity, I believe we should not seek to be very prescriptive about the future of evolutionary economics. However I suggest that the concept of organization should be central, because not only is knowledge itself an organizational phenomenon, its development is substantially influenced by its organizational setting - in both the internal structure and external relationships of firms and research institutes, and by institutions both in the sense of formal entities and as sets of standard practices, all of which economise human cognition and influence its application. This proposition is summarised in what may be thought the most important pair of sentences in Marshall's Principles, on the crucial roles of knowledge and the multiple forms of organization which shape its development (Marshall 1920: 138). Organization is ubiquitous (Loasby 2007). My final proposition is that evolutionary economists should recognise the fundamental twin contributions to thinking about their subject which have been provided by Herbert Simon. Cognition must be recognised as the most crucial of all scarce resources, both for the allocation of their own capabilities and as a research topic; and the exploitation of quasi-decomposability, both in their field of enquiry and in their discipline, is the key both to the organisation of their studies and to the understanding of their subject. These are the keys to man's investigation of nature, and to the investigation of economic systems. References Atiyah, M. (2008), Mind, matter and mathematics. Presidential Address to the Royal Society of Edinburgh. Recording available at royalsoced.org.uk/events. Burns, T., G.M. Stalker (1961), The Management of Innovation. London: Tavistock Publications. Coase, R.H. (1937), The nature of the firm. Economica Ν. S. 4: 3 8 6 ^ 0 5 . Reprinted in R.H. Coase (1988), The Firm, The Market and the Law. Chicago: University of Chicago Press, pp. 33-55. Coase, R.H. (1972), Industrial Organization: A Proposal for Research. Pp. 59-73 in: V.R. Fuchs (ed.), Policy Issues and Research Opportunities in Industrial Organization. NBER General Series no. 96. Cambridge MA: National Bureau of Economic Research. Reprinted in R.H. Coase (1988), The Firm, The Market and the Law. Chicago:University of Chicago Press, pp. 57-74. Drucker, P.F. (1969), The Age of Discontinuity. London: Heinemann. Fleck, J. (2000), Artefact •«* activity: the coevolution of artefacts, knowledge and organisation in technological innovation. Pp. 248-266 in: J. Ziman (ed.), Technological Innovation as an Evolutionary Process. Cambridge, Cambridge University Press. Frisch, O. (1979), What Little I Remember. Cambridge: Cambridge University Press. Hayek, F.A. (1952), The Sensory Order. Chicago: University of Chicago Press. Heisenberg, W. (1958), The Physicist's Conception of Nature. New York: Harcourt Brace. Hume, D. (1875), Essays Moral, Political and Literary. Ed. T.H. Green and T.H. Grose. London: Longmans, Green and Co. Hume, D. (1878), A Treatise on Human Nature. Ed. L.A. Selby-Bigge. 2 n d ed. Revised by H. Nidditch. Oxford: Clarendon Press. Karmiloff-Smith, A. (2002), Elementary, my dear Watson, the clue is in the genes ... or is it? Proceedings of the British Academy 117: 525-543. Oxford: Oxford University Press for the British Academy.

The Evolution of Knowledge and Knowledge of Evolution · 157

Kelly, G.A. (1963), A Theory of Personality. New York: W.W. Norton. Knight, F.H. (1921), Risk, Uncertainty and Profit. Boston: Houghton Mifflin. Loasby, B.J. (2007), The ubiquity of organization. Organization Studies 28(11): 1729-1759. Marshall, A. (1920), Principles of Economics. 8th ed. London: Macmillan. Marshall, A. (1994), Ye machine. Research in the History of Economic Thought and Methodology, Archival Supplement 4. Greenwich, CT: JAI Press, pp. 116-132. Penrose, E.T. (1959), The Theory of the Growth of the Firm. Oxford: Basil Blackwell. Popper, K.R. (1963), Conjectures and Refutations. London: Routledge and Kegan Paul. Raffaelli, T. (2003), Marshall's Evolutionary Economics. London: Routledge. Richardson, G.B. (1972), The organisation of industry. Economic Journal 82: 883-896. Richardson, G.B. (1975), Adam Smith on competition and increasing returns. Pp. 350-360 in: AS. Skinner, T. Wilson (eds.), Essays on Adam Smith. Oxford: Clarendon Press. Samuelson, P.A. (1967), The monopolistic competition revolution. In: R.E. Kuenne (ed.), Monopolistic Competition: Studies in Impact. New York: Wiley. Reprinted pp. 18-51 in: R.K. Merton (ed.), The Collected Scientific Papers of Paul A. Samuelson, Vol. 2. Cambridge MA and London: MIT Press. (Reference is to this reprint.) Schumpeter, J.A. (1934), The Theory of Economic Development. Cambridge MA: Harvard University Press. Simon, H.A. (1969), The architecture of complexity. Pp. 84—118 in: The Sciences of the Artificial. Cambridge MA and London: MIT Press. Smith, A. ([1759] 1976a), The Theory of Moral Sentiments. Eds. D.D. Raphael, A.L. Macfie. Oxford: Oxford University Press. Smith, A. ([1776] 1976b), An Inquiry into the Nature and Causes of the Wealth of Nations. Eds. R.H. Campbell, A.S. Skinner, W.B. Todd. Oxford: Oxford University Press. Smith, A. ([1795] 1980), The principles which lead and direct philosophical enquiries: illustrated by the history of astronomy. Pp. 33-105 in: W.P.D. Wightman, J.C. Bryce (eds.), Essays on Philosophical Subjects. Oxford: Oxford University Press. Thomas, J.M. (2007), The unpredictability of science and its consequences. Lecture at the Royal Society of Edinburgh. Recording available at roaylsoced.org.uk/events. Young, A. (1928), Increasing returns and economic progress. Economic Journal 38: 527-542. Ziman, J. (1978), Reliable Knowledge. Cambridge: Cambridge University Press. Ziman, J. (ed.) (2000a), Technological Innovation as an Evolutionary Process. Cambridge: Cambridge University Press, pp. 248-266. Ziman, J. (2000b), Real Science. Cambridge: Cambridge University Press. Brian J. Loasby, Division of Economics, University of Stirling, Stirling FK9 4LA, Scotland. [email protected]

Jahrbücherf. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

Like Doktorvater, like Son? Tracing Role Model Learning in the Evolution of German Laser Research Guido Buenstorf and Matthias Geissler* Institute of Economics and International Center for Higher Education Research (INCHER-Kassel), University of Kassel JEL D83; I23; 033 Open science; role models; observational learning; doctoral dissertations; laser research.

Summary We trace individual-level learning and knowledge transfer in public research by matching about 5 , 0 0 0 doctoral dissertations and their advisors over the full history of German laser research. We study the number of laser-related dissertations per advisor, publication and patent outputs of advisors and doctoral students, as well as the likelihood that former students started laser firms or attained professorships. Our results suggest a substantial relevance of non-codified knowledge and role model learning in public research. There is little evidence of pronounced barriers to entry into laser research.

1

Introduction

Open disclosure and dissemination o f results in publications is the cornerstone o f " o p e n science" ( M e r t o n 1 9 7 3 ) , the institutional setup governing public research. T h e reputation effect o f publications and citations provides researchers with powerful incentives to arrive at novel and relevant findings. At the same time, dissemination o f knowledge via publications ensures that researchers have unrestricted access to other researchers' earlier findings, allowing them to build on the existing stock o f knowledge and to further advance science (Dasgupta/David 1 9 9 4 ; Stephan 1 9 9 6 ) .

* W e thank a large number of university libraries and departments throughout Germany that kindly provided information about dissertation advisors. A special note of thanks is due to the staff of the Deutsche Nationalbibliothek Leipzig for their hospitality, and to Katja Mehlis, Juliane Lucas, Wolfhard Kaus, Claudia Ludwig, Mandy Kramer und Judith Usbeck for research assistance. Sebastian Schmidt and Wolfgang Ziegler from the University of Jena provided assistance in retrieving publication and patent data. W e benefitted from insightful comments on earlier versions by T o m Astebro, M a r k u s Becker, Steven Klepper and Viktor Slavtchev. Financial support from the Volkswagen Foundation, Project "Emergence and Evolution of a Spatial-Sectoral System of Innovation: Laser Technology in Germany, 1 9 6 0 to Present" (LASSSIE) is gratefully acknowledged.

Like Doktorvater,

like Son? • 159

Given this incentive structure, public research should provide a context in which the traditional economic presumption that knowledge is a public good is approximated as closely as anywhere. Accordingly, public research also provides a challenging testing ground for the competing view held by evolutionary economics, which has never subscribed to the public-good perspective on knowledge. Evolutionary economists have long argued that knowledge cannot be presumed to flow freely between agents and organizations. In the evolutionary perspective, "cognitive" issues of who knows what take center stage. Learning and the transfer of non-codified or "tacit" knowledge are seen as crucial processes in economic change and development (Nelson/Winter 1982; Cowan et al. 2000). If empirical evidence supported this view even in the realm of public research, this would strengthen the general perspective on knowledge underlying evolutionary economics. In this paper we explore the transfer of non-codified knowledge through role model learning (Bandura 1986; Witt 1987, 1998, 2000) in public research, emphasizing knowledge flows from more senior to more junior researchers. To this purpose we employ a unique dataset encompassing the entire population of non-medical laser-related doctoral dissertations completed at German universities over a period of about 45 years. Based on extensive archival research, we were able to match the authors of these dissertations to their dissertation advisors. We also retrieved a variety of other types of information about doctoral students and their advisors. Germany is a well-suited empirical context for our analysis because German doctoral students have traditionally worked with a single individual advisor (the "Doktorvater"1) who exerted substantial influence on their research. Metaphorically, if role model learning gives rise to advisor effects on students' research orientation and performance, knowledge is "inherited" from the advisor. Such "vertical" knowledge transfer may also limit the factual openness of "open" science toward entry by outsiders. We therefore first develop a genealogy of German laser research spanning several academic "generations" to assess the degree of openness in this field of research and how it changed over time. Linking the dissertation dataset to patent and publication data, we then turn to the individual level and trace similarities in the research orientation and performance of doctoral students and their advisors, which are interpreted as observable effects of role model learning. We also study whether advisor characteristics are systematically related to the likelihood that (former) doctoral students started a laser firm or attained a professorship. We find that researchers who had written their own dissertation in the field of laser research on average advised three times as many laser-related dissertations as other advisors in the dataset. A substantial fraction of recent laser-related doctoral dissertations have a direct lineage to the first generation of German laser researchers. At the same time, there are few lasting "dynasties", which suggests there were no pronounced barriers to entry into laser research. We moreover find parallels between advisors and students in terms of publication and patent outputs. Overall our results are consistent with a substantial relevance of non-codified knowledge and role model learning in public research. Similar learning processes can account for empirical patterns observed in other contexts, for example labor mobility and employee entrepreneurship (Klepper 2001, 2009). This suggests that better understanding processes and implications of individual-level learning based on direct interaction may be relevant for evolutionary economics more generally.

1

Literally, the "doctor's father." The linguistic gender bias is reflected in the composition of our data, in which the large majority of all doctoral students, and even more so of advisors, are male. Note that, in contrast to recent work on Italy by Durante et al. ( 2 0 1 1 ) , our reference to "inheritance" is purely metaphorical.

160 · Guido Buenstorf and Matthias Geissler

The remainder of the article proceeds as follows. Section 2 provides the theoretical background of the discussion by highlighting the role of non-codified knowledge and learning in evolutionary economics. Section 3 discusses knowledge transfer in the institutional setup of open science. In section 4 we propose that studying doctoral dissertations is a useful empirical approach to trace the relevance of "vertical" transfer of scientific knowledge. Section 5 introduces the data. A genealogy of German laser research is presented in section 6, and results from econometric analyses in section 7. Section 8 concludes by relating our findings to the broader context of evolutionary economics.

2

Non-codified knowledge and learning in evolutionary economics

Economists mostly characterize knowledge as a public good, i.e. non-rival in use and impossible or at least difficult to appropriate (Nelson 1959; Arrow 1962). 2 In line with the conventional microeconomic view of commonly known technologies, this characterization of knowledge implies that new knowledge is (almost) costlessly and instantaneously transferred between individuals and organizations, and also across regions and countries. It underlies, for instance, models of patent races assuming symmetric firms trying to reach some pre-specified innovation target (e.g., Reinganum 1982) and also endogenous growth models (Romer 1990). Given that knowledge transfer and learning are not seen as problematic in this perspective, conventional economics mostly discusses innovation as an issue of getting the incentives right. In particular, the limited appropriability of knowledge provides a rationale for public subsidization of knowledge-generating activities as well as for intellectual property rights. From its beginnings, modern evolutionary economics has proposed a very different view of knowledge. Richard Nelson and Sidney Winter's An Evolutionary Theory of Economic Change (Nelson/Winter 1982) is noteworthy not only for its well-known analogies to evolutionary biology. Nelson and Winter also develop behavioral foundations of organizational learning. In so doing, they draw extensively on the notion of tacit knowledge (Polanyi 1962). Tacit knowledge is associated with skillful individual behavior achieved even though the individual is not able to articulate the rules on which her behavior is based.3 It corresponds to the psychological concept of procedural memory (Cohen/Bacdayan 1994) and is closely related to the distinction between automatic and controlled cognitive processes (Camerer et al. 2005). As discussed in detail by Cowan et al. (2000), tacit knowledge is a multi-faceted concept. Knowledge may be unarticulated (and as a consequence un-codified, i.e. not expressed in a symbolic representation; cf. also Cowan/Foray 1997) for different reasons. For some kinds of knowledge, verbal expression may not be possible in principle. In other instances,

2

3

Nelson (1992: 61) notes that economists have referred to technological knowledge as a "latent" public good because it has "the capacity to benefit many parties" while being "inexpensive (if not literally costless) to teach and learn compared with the cost of invention or discovery in the first place". Because public-goods characteristics are not completely satisfied by all kinds of (codified) knowledge, some economists also refer to knowledge as a "quasi-public good" (cf., e.g., the discussion by Callón 1994). That Nelson and Winter highlight the importance of tacit knowledge is striking because it was one of these authors, Richard Nelson (1959), w h o had first conceptualized knowledge as a public good.

Like Doktorvater,

like Son? · 161

knowledge may remain unarticulated because it is not (yet) understood sufficiently well. This does not imply that articulation, and codification, are inherently impossible. Incentives for codification may simply be lacking, for example because only a small number of agents need to be in command of the respective knowledge and these agents can be trained through direct interaction. It is also possible that knowledge remains unarticulated in a specific context even though it has been codified before in other contexts. Personal expertise may be used as an example. It has long been known that expert problem solving is often based on complex, problem-specific heuristics that are acquired through repeated exposure to the same type of problem and which cannot be articulated by the expert herself (Anderson 2000). Elsewhere (e.g., in academia), the knowledge underlying a practitioner's expertise may have been codified, but it is still non-codified in the context of this practitioner. This restricts the potential of her knowledge to be transferred. In effect, as long as it remains non-codified, it can only be learned by others through direct interaction and observation. 4 Nelson and Winter (1982) were primarily interested in non-codified organizational knowledge embodied in a firm's organizational routines. While they suggested analogies between organizational routines and individual skills, individual-level learning was not in the focus of their analysis. A few years later, Ulrich Witt (1987) adopted a more individualistic perspective highlighting the interplay of inherited human nature, selective cognitive attention, and individual as well as social learning in economic processes. In particular, Witt proposed role model or "observational" learning (Bandura 1986) as a key process underlying the interpersonal transfer of non-codified knowledge. In what follows, we trace the importance of this type of learning in the context of German laser research. Emphasizing the importance of non-codified knowledge has had profound implications for applied evolutionary economics over the past decades. If knowledge is not freely available as a public good, then heterogeneity of agents and variations in their performance due to differential competences, access to knowledge and learning capacities come into focus. This can be seen in the evolutionary work on innovation, which generally focuses on innovative capacities and the direction of change more than on economic incentives (cf. Dosi 1982, 1988; Hanusch/Pyka 2007). Along similar lines, "evolutionary economic geography" (Boschma/Frenken 2006) emphasizes "specific" channels of knowledge transfer such as labor mobility and social networks over the "general" concept of agglomeration economies. Similar to Nelson and Winter's original contribution, individual-level learning and its relevance for organizational changes have not been focused upon in the thrust of the applied evolutionary work. However, adopting a more individualistic approach seems required to better understand dynamics at the organizational, or even regional and national, level. The recent evolutionary work on spin-offs, i. e., startups formed by employees of existing firms (Klepper/Sleeper 2005; Klepper 2009), provides a good illustration of this point.

4

In what follows, we will largely abstract from the issue of whether or not some kind of knowledge could in principle be codified. For the purposes of the present paper, the important insight from the above discussion is that some knowledge held by some specific agents remains non-codified in the context of these agents. As a consequence, this knowledge cannot be accessed by outsiders who do not directly interact with the respective agents. Therefore, consistent with the discussion of Cowan et al. (2000), non-codification or tacitness of knowledge is treated as a context-dependent property.

162 • Guido Buenstorf and Matthias Geissler

Spin-offs tend to outperform other startups, particularly if started by founders leaving successful incumbents (cf., e.g., Klepper 2002,2007; Buenstorf/Klepper 2009). Early work has attributed the direct relationship between parent firm and spin-off performance to inter-organizational "inheritance" of routines (Klepper 2001). It is more challenging to identify the individual-level processes of learning and knowledge transfer underlying this inheritance, but spin-off entrepreneurship seems to entail substantial on-the-job learning of prospective founders, consistent with the notion of "career imprints" (Higgins 2005). 5 The findings on spin-off performance moreover indicate that the knowledge brought in by spin-off founders is not fully accessible to other entrepreneurs, suggesting that at least part of it is non-codified and has been acquired through observation and direct interaction with role models. In turn, once the new firm has been started, spin-off entrepreneurs are in a strategic position to serve as role models and provide shared cognitive frames to coordinate and motivate the workforce (Witt 1998, 2000). In this way, they may shape the organizational routines of the new firm. Based on observational learning and transfer of non-codified knowledge, a micro-level account of the "inheritance" of routines can thus be provided. However, there is only limited empirical evidence to systematically test the relevance of the proposed processes for spin-off entrepreneurship and other forms of private-sector labor mobility. To shed more light on individual-level role model learning, we therefore turn to the empirical context of public research, where we expect similar dynamics to be at work. Using unique data for German laser research, we study whether the long-term evolution of this field of research suggests the importance of differential access to non-codified knowledge and role model learning by doctoral students. While obviously this analysis can only provide evidence for a rather specific empirical context, we understand it as a contribution to a more general question of evolutionary economics: How important are role model learning and the transfer of embodied non-codified knowledge based on labor mobility?

3

Role model learning in open science?

Public research is a competitive system governed by a specific set of institutions and incentive mechanisms (Merton 1973; Dasgupta/David 1994; Stephan 1996). Scientists compete for reputation in the peer group of their research community, which not only provides status but also informs decisions on hiring, tenure, and promotion. Reputation is moreover an important precondition for a researcher's ability to attract research funding and to generate extra monetary income, e.g., from public talks and consulting. It is allocated according to successful research activities that lead to novel findings. Publishing articles in highly regarded academic journals is the single most important activity allowing researchers to establish their priority in attaining a result. Citations by others establish the relevance of their findings. According to the established economics of science, publication-based reputation solves a dilemma caused by the presumed public-good nature of knowledge (Dasgupta/David 1994; cf. Foray 2004). On the one hand, incentives are needed to entice researchers to work hard and reward those who are successful. This is particularly important because

5

Inter alia, this is suggested by the positive association of founder experience and firm performance (e.g, Dahl/Sorenson 2012).

Like Doktorvater,

like Son? · 163

agency problems are endemic in the production of scientific knowledge. Both efforts and inherent quality of researchers (as agents) are difficult to observe and evaluate for society (as the principal) and also for the heads of universities and research organizations. On the other hand, even though scientific knowledge is costly to produce, the marginal costs of using (codified) scientific knowledge are close to zero. Accordingly, it is efficient to make the results of research activities freely available to the public. Alternatively addressing the principal-agent problem by restricting the access to new scientific findings and pricing them according to their usefulness - which would provide researchers with monetary incentives to produce relevant findings - is not in society's best interest. The reputation-based system of open science not only reconciles the "pricing" of research findings based on their usefulness (as perceived by the peer group of other researchers) with open access to these results. It should also provide individual researchers with strong incentives to fully disclose their findings. In principle, open science thus aligns individual incentives toward knowledge dissemination with societal objectives. However, there are reasons to believe that knowledge dissemination through publications is less than perfect, and that scientific knowledge (much like technological knowledge) is not as "public" a good as economic theory would traditionally have it. Qualitative evidence suggests that not all relevant knowledge a researcher possesses can be articulated and codified by the researcher her-/himself. For example, in Collins' (1974) case study of the TEA laser (a classic in the sociology of science), replicating laser designs was complicated by the fact that the importance of some design elements (such as transformer inductance) was not fully understood by the original inventors of the design, while other aspects (e.g., precision machining of electrodes) proved to be less important than the inventors originally thought they were. In-depth familiarity with the research field based on their personal experience helps researchers tackle such problems of imperfect knowledge, where the background knowledge acquired from own experience and from directly observing and working with other researchers is often not codified, but rather shows in the ability to find solutions to problems intuitively. In the case of the TEA laser, no research group was able to construct a workable laser without directly interacting with the original developers (ibid.). 6 The importance of non-codified knowledge is also stressed by the researchers themselves. Zellner (2003) surveyed former doctoral students and post-doctoral researchers of the German M a x Planck Society w h o had left public research for private-sector jobs. The survey asked individuals to assess what kinds of knowledge and skills acquired at M a x Planck had been relevant to their subsequent careers. It found that former researchers generally judged relatively unspecific and frequently non-codified knowledge more important than the highly specific knowledge that is disclosed in publications. The most important type of knowledge, according to the respondents, were non-specific analytical skills allowing researchers to recognize, structure and solve problems in their field. The kinds of knowledge emphasized by the researchers in Zellner (2003) cannot be acquired through access to the scientific literature alone. This leaves room for knowl-

6

Dissemination through publications may be further limited because the disclosure of results is a double-edged sword for the individual researcher (Dasgupta/Maskin 1987). While necessary for publishing one's work and thus build up reputation, full disclosure levels the playing field, eliminating whatever first-mover advantages a researcher may have accumulated in the past.

164 · Guido Buenstorf and Matthias Geissler

edge transfer based on the direct interaction, for instance between co-workers at the researcher's lab (cf. Tartari et al. 2 0 1 0 ; Krabel 2 0 1 2 ) . For the related issue of universityindustry knowledge transfer, quantitative work by Agrawal (2006) indeed shows that direct interaction between university inventors and private-sector licensees increases the likelihood of successful commercialization. For doctoral students the dissertation advisor can be expected to be the most important source of non-codified knowledge. Face-to-face contact to the advisor not only allows for verbally communicating knowledge. It also enables doctoral students to benefit from role model or observational learning (Bandura 1986), giving them access to a broad set of knowledge and skills that their advisors possess. Relevant non-codified kinds of knowledge that can thus be acquired include professional etiquette, the advisor's framing of the research field, the ability to recognize opportunities and evaluate alternative challenges and approaches, intuitive problem-solving skills and expertise (Anderson 2 0 0 0 ) , and also skills related to the design and setup of experiments, equipment and materials handling, etc. 7 Again, this conjecture is supported by empirical evidence. Bercovitz and Feldman (2008) present evidence of organizational "imprinting" during graduate education affecting researchers' participation in technology transfer activities. A survey of 4 0 0 U.S. scientists and engineers by Roach and Sauermann (2010) finds that researchers differ in the extent to which they acquired a "taste" for science during their doctoral studies. These authors also show that individuals self-select into academic vs. private-sector careers based on their preferences for different job characteristics. Self-selection into different careers may not only reflect different "career imprints" received by doctoral students. It may also help account for the observable tendency that junior researchers entering into a research field are often students of insiders already working in this field. For instance, Murray (2010), in her in-depth account of the mouse genetics research community, characterizes this community as a "close network" in which "many of today's leaders can trace their scientific lineage to a single individual". This observation by Murray (2010) leads us to the first question addressed by the subsequent empirical analysis of German laser research: T o what extent are junior researchers entering into a field of research students of researchers already active within the same field? And second, to what extent can we find individual-level parallels between activities and performance of doctoral students and their advisors, as would be consistent with relevant role model learning? Our approach to answer these questions is outlined in the next section.

7

In larger labs, some of the most important interactions will not be between student and (nominal) advisor, i.e., the director of the lab, but intermediates such as post-doctoral researchers and technical staff may play important roles. T o the extent that these have themselves acquired knowledge through observational learning from the lab director, a more indirect, mediated process may nonetheless be at work. In addition, successful scientists running large labs can be expected to provide powerful role models (or "cognitive leadership", cf. W i t t 1 9 9 8 ) shaping the knowledge, skills and attitudes of their doctoral students.

Like Doktorvater, like Son? • 165

4

Laser-related dissertations as indicators of role model learning

Almost by definition, transfer of non-codified knowledge can hardly be observed directly. Our empirical analysis is therefore based on a simple conjecture. If dissemination of codified knowledge through publications led to a perfectly leveled "playing field" of science, then the "vertical" transfer of knowledge from advisors to their doctoral students should not have systematic effects on activities and career outcomes. "Outsiders" and students of weaker advisors should be able to compensate for their disadvantage as long as they have access to the relevant literature (which would seem to depend on university-level factors more than on individual advisors). In contrast, if access to non-codified knowledge based on role model learning and face-to-face communication between advisors and doctoral students is crucial, then there should be similarities between advisors' and students' research activities and productivity, and rates of outsider entry into a field of research should be low. Potential advisor effects are identified with different kinds of empirical probes for laser research in Germany from its inception in 1960 to 2007. We first develop a genealogy of laser research and trace academic "dynasties" based on the links between doctoral students and their advisors. Next, we study the shares of doctoral dissertations submitted at various points in time that were advised by researchers who had authored a laser-related dissertation themselves, or by other early entrants into the field of laser research. We then turn to publication and patent data to see whether there are observable similarities in the research activities and productivity of students and advisors. Finally, we investigate factors that shape the likelihood of laser entrepreneurship by former doctoral students, and also the likelihood that a former doctoral student assumed a professorship at a German university. Our empirical approach exploits institutional particularities of the German academic system. As noted above, individual advisors have a particularly strong impact on the dissertation work of their students, while the role of the dissertation committee is mostly limited to evaluating students' finalized dissertations as well as their performance in oral examinations. 8 This characteristic - which is reflected by the metaphorical notion of the Doktorvater - facilitates our task of tracing advisor effects. It is also expected to enhance the importance of such effects. In addition, the share of doctoral students w h o remain in public research after completing their degree is lower in Germany than elsewhere, notably in the U.S.' This is mostly because German firms have traditionally hired large numbers of individuals holding doctoral degrees in science and engineering for R & D and technical marketing positions. Our analysis is based on the full population of doctoral dissertations in non-medical laser research from 1960 to 2007. Accordingly, we study a research field composed of researchers from different disciplines (mostly physics and engineering) rather than a single academic discipline. We exclude laser-related medical research because we expect that in this research lasers are often used only as laboratory or clinical equipment with no inherent interest of the researcher in laser science and technology. 8

9

Dissertation proposals are not normally discussed or defended before a committee at German universities. Even though the numbers are not fully comparable and censoring issues are relevant, it is illustrative to compare the share of doctoral students in our sample w h o did not attain professorships (more than 95 % ) to the U.S. Survey of Earned Doctorates where only 1 5 % identified a private-sector employer (35 % did not report any specific career plans in that survey; cf. Stephan et al. 2005).

166 • Guido Buenstorf and Matthias Geissler

Laser research is particularly well-suited to our purposes because its beginnings can be precisely dated: the first workable laser was built by U.S. researchers in 1960, and before this date there was little interest in lasers among German researchers (Albrecht 1997). From the beginning, the laser was recognized as a revolutionary technology with potentially profound implications for both research and practical uses. Because of the high degree of visibility attained by the laser, already the early contributions to laser research are categorized as such. This is critical for our ability to retrieve them in the bibliographical data.

5

Data

Starting point of the empirical analysis is the online catalog of the Deutsche Nationalbibliothek (DNB 2007) where, as a mandatory part of all doctoral degrees granted by German universities, a copy of the doctoral dissertation has to be deposited. A simple query for "laser" in any field and "diss" for dissertations in Schrift (denoting the type of registered work) yielded 7,580 entries (documents). After several corrections, the raw dataset includes 6,389 doctoral dissertations published from 1960 to 2007. 10 Figure 1 plots the distribution over time of the dissertations in our full dataset. It shows a constantly increasing number of dissertations (starting from very modest beginnings) over the first four decades of laser research, with a roughly logistic pattern of growth. In the late 1990s, the number of dissertations leveled off at about 320-350 p.a. and then started to decline. Except for the last year when the decline reflects right censoring of the data, we do not have a straightforward explanation for this decline, which is entirely due to non-medical dissertations. For all dissertations, title, author name, year of completion, and university name are provided in the D N B online catalog, and also some information about the field of research. 11 However, our main information of interest (the Doktorvater) is not available from the online catalog but had to be searched manually. In doing so, we proceeded in four steps. Recent dissertations are frequently available online as full texts. These were analyzed first. In step two, we contacted university departments and libraries and asked them to provide the necessary information. Next, for about half of all dissertations, in particular those submitted in the earlier decades of laser research (in total, approximately 3,000 items), advisor information had to be obtained directly via archival research at the DNB facilities in Leipzig. Finally, in step four, dissertations that could not be obtained in any other way, or for which the information was inconclusive, were ordered via inter-library loan. 10

Although the term "laser" is relatively unambiguous, the data obtained had to be corrected with respect to two issues: First, the search terms were entered in a truncated version (i. e. ""laser*"), which resulted in hits where the title or the author's name included this string but were unrelated to lasers (i.e. individuals by the name of "Glaser"). Second, the identification via the D N B catalog yielded entries not items. Therefore, some dissertations entered the dataset twice, because they are available in electronic form as well as in hardcover. Correcting for both issues, 6,305 items remained in the dataset. An additional query for dissertations with "maser" and "quantum optics" in any field was made to enrich the dataset, particularly in the early phase of laser research. After correction for duplicates and naming issues, 139 items were added to the dataset. Finally, Habilitationen were referred to as "Diss. B" in some East German universities prior to 1990. These have been excluded.

11

For about 4 , 0 0 0 dissertations we were able to supplement the catalog data with information about the department and for 3,800 the exact degree awarded to the candidate.

Like Doktorvater, like Son? · 167

Non-medica! Dissertation Si Medicai Dissertation

Figure 1 Distribution over time of laser-related doctoral dissertations at German universities

In identifying dissertation advisors, a challenge arose out of the fact that dissertations are assessed by at least two examiners. Who then is the Doktorvater? Fortunately, in the majority of cases a distinction is made between the "first" and the "second" examiner, which we used to identify the actual dissertation advisor. In the remaining cases we took the first listed examiner to be the advisor. This would lead to invalid inferences if examiners were named in alphabetical order. However, manual inspection did not suggest that examiners are generally listed in alphabetical order. It is moreover conceivable that the ordering of examiners reflects university or departmental politics, with higher-ranking individuals mentioned first. Evidence against this possibility is provided by the fact that we frequently observe that the first examiner is not a Professor, but a lower-ranked Privatdozent or Doktor.12 We were able to establish the identity of the advisor for 6,325 dissertations. These were advised by 2,378 unique examiners. Missing information on advisor degrees and titles was obtained using various editions of Kürschners Gelehrtenkalender, a commercial directory of German university professors issued regularly and covering the entire time period under investigation. The same sources, as well as online queries, were used to deal with homonyms and to identify future professors in the sample of doctoral students. Future laser entrepreneurs among the set of doctoral students were searched using an updated

12

T o validate our convention of identifying the first examiner as dissertation advisor we sampled volumes randomly and searched for the advisor's name in the acknowledgements. There were only a small number of cases where first examiners and advisors did not seem to match. In some of these cases, the person named in the acknowledgements was not at all listed as an examiner, possibly because (s)he was not legally entitled to act as such.

168 · Guido Buenstorf and Matthias Geissler

list of founder names that one of the authors first assembled in prior research (Buenstorf 2007). The empirical investigations presented below are restricted to the subset of non-medical dissertations (identified on the basis of the awarded degrees and field information included in the DNB catalog) .This more limited dataset includes 4,845 dissertations advised by 1,585 individual advisors; i.e. on average each Doktorvater had 3.05 laser-related dissertations. "Entrants" into laser research, i.e. doctoral students whose advisors had not written a laser-related dissertation themselves, dominate the dissertation output in all years. However, 79 individuals were identified both as advisors and as authors of laserrelated dissertations. As advisors these laser-trained researchers account for a total of 692 dissertations (average: 8.8 vs. 2.8 for advisors coming from other fields). By necessity, in the initial years of the data no individuals with laser-related dissertations are found among the advisors. Thus, in these early years our general approach is of limited value for identifying "insiders" to laser research. We therefore employed information about laser-related Habilitationen to supplement the early dissertation data, using the same identification strategy as for dissertations (DNB catalog: "laser" in any field; "habil" or "hab" in the Schrift criterion).13 Until 2007, 197 laser-related Habilitationen were submitted to German universities. Below, the authors of the first-year dissertations and Habilitationen will be used to construct a set of "early entrants" into German laser research. In Figure 2, the non-medical dissertations are further distinguished by type of advisor.14 The share of students whose advisors had written their own dissertation in laser research (cf. Figure 2) rises over time. In the 1976-90 period, they account for about 9 % of all dissertations. After 1990, this share increased to 16 %. There is also a persistent flow of doctoral students advised by laser researchers who entered the field early on but whose own advisors were not active in laser research ("early entrants" defined as researchers who wrote a laser-related dissertation in 1970 or earlier and/or a laser-related Habilitation in 1975 or earlier).15 To study potential advisor effects in more detail, the dissertation dataset was subsequently linked to other individual-level information about German laser research. Patent data were obtained by matching the names of all authors and advisors of laser-related dissertations with the inventor names listed on all applications at the German patent office in the most relevant IPC classes for laser sources, spectroscopy, laser material processing, and medical laser applications. 16 The restriction to laser-relevant patent classes was motivated both by substantive considerations (we wanted to identify activities closely related to laser technology) and to minimize the likelihood of homonyms. Laser-related publications were obtained using the Inspec database for science and engineering publications 13

14 15

16

In the German system, the Habilitation - basically a second dissertation - has traditionally been the entry ticket into an academic career. Only a minority of doctoral students subsequently obtains a Habilitation. As opposed to a doctoral dissertation, which also opens up opportunities for privatesector careers, a Habilitation strongly signals intentions to pursue an academic career. East German dissertations submitted until 1 9 9 0 are excluded. In Figure 2, the dissertations advised by "early entrants" without a laser dissertation are depicted in light color between entrants and students. N o t shown in the figure is the substantial overlap of the "early entrant" group with the group of advisors with a laser-related dissertation. The selection of patents was complex and follows different criteria for the individual fields of application (for example, laser sources: IPC H01S; material processing: IPC B23K 26/00; sensors: IPC G 0 1 N plus "laser" in fulltext).

Like Doktorvater, like Son? · 169

300

250

200

« No laser background of advisor 150

m Early entrant advisor w/o laser dissertation si Advisor with laser dissertation

100

50

σι

w m m (ν σι

Figure 2 Non-medical laser-related doctoral dissertations by advisor background (note: East German dissertations submitted before 1990 are excluded) and results were again matched with the list of doctoral student and advisor names. This yielded a larger set of 4 1 , 7 1 8 (non-unique due to co-authorships) articles published from 1 9 6 0 to 2 0 0 7 . Descriptive statistics and correlations between variables are given in Tables 1 and 2 a - c .

6

A genealogy of German laser research

W e begin our empirical analysis by constructing a genealogy of German laser research based on dissertation advisors and their students. This helps us get a first sense of how open this field of research was toward entry from outside, and how important " s c h o o l s " initiated by early entrants into laser research were in the subsequent development of the field. In line with our primary interest in openness and the relevance of students of laser researchers becoming advisors of laser-related dissertations themselves, we focus on "generations" of advisers and largely suppress the temporal dimension of the evolution of German laser research. Starting point for the genealogy are all 1 , 5 0 6 "first-generation" advisors of doctoral laser-related dissertations whose advisors have not written a laser-related dissertation themselves. These are shown in the first row of Figure 3; in total they advised 4 , 1 5 3 laser-related dissertations. (Numbers of dissertations advised by (groups of) advisors are given in parentheses in Figure 3.) First-generation advisors can be further distinguished into three groups according to the number of student generations that we can trace in the dissertation dataset. The large majority of 1 , 4 4 7 advisors had only one generation of "offspring", i.e. none of their own doctoral students became advisors of laser-related dissertations themselves. These researchers nonetheless seem to have been more than just

170 · Cuido Buenstorf and Matthias Geissler

Table 1 Descriptives Mean (Std. Dev.)

Min

Max

4,841 (diss.) 4,841 (diss.) 4,841 (diss.) 4,841 (diss.) 4,841 (diss.) 4,841 (diss.)

.144 (.351) .089 (.285) .033 (.178) .194 (.396) .773 (.419) 75.616 (73.373)

0

1

0

1

0

1

0

1

0

1

0

366

Integer, sum of all diss, by respective adv. in the population Binary, indicating that this adv. is in the population of dissertations himself Binary, indicating that the adv. of the adv. is in the population as well Binary, indicating that the adv. has an early diss. (bef. 1970) or habilitation (bef. 1975) Integer, computed by subtracting year of first advised diss, from 2008 (final year) Integer, computed by multiplying above variable with itself

1,585 (adv.) 1,585 (adv.) 1,585 (adv.) 1,585 (adv.) 1,585 (adv.) 1,585 (adv.)

3.057 (5.973) .051 (.222) .008 (.087) .016 (.127) 14.765 (9.927) 316.476 (389.747)

1

82

0

1

0

1

0

1

1

49

1

2401

Integer, number of student publications after the dissertation Integer, number of student patents after the dissertation Binary, indicating whether the student founded a laser firm Binary, indicating whether the student subsequently became professor Integer, number of advisor publications one year before dissertation year Integer, number of advisor patents one year before dissertation year Binary, indicating that the adv. had an early diss. (bef. 1970) or habilitation (bef. 1975) Integer, number of student pubi, one year before diss, year (restricted to max 5) Integer, number of student patents one year before diss, year (restricted to max 3) Integer, computed by subtracting (dissertation year) from 2008 Integer, cumulated number of diss, advised at a university up to year before target diss. Binary, indicating diss, in engineering departments

3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.) 3,880 (diss.)

2.980 (11.593) .461 (2.259) .005 (.073) .042 (.200) 22.718 (36.093) 1.655 (3.802) .094 (.291) .579 (1.102) .069 (.327) 13.816 (8.029) 67.209 (64.916) .323 (.468)

0

279

0

88

0

1

0

1

0

237

0

36

0

1

0

5

0

3

4

49

0

314

0

1

Variable

Description

Sample Size

Advisor with own laser diss. Advisor is laser pioneer

Cum. diss, at university (up to t-1)

Binary, indicating that the adv. is in the population as well Binary , indicating that the adv. has an early diss. (bef. 1970) or habilitation (bef. 1975) Binary, indicator for dissertation advised in earliest cohort Binary, indicator for dissertation advised in second cohort Binary, indicator for dissertation advised in latest cohort Integer, cumulated number of diss, advised at a university up to year before target diss.

Number of advised diss. Advisor with own laser diss. Third generation advisor Advisor is laser pioneer Time of entry (inverse) Square (Time of entry (inverse)) Student pubi, (after)

Diss. 1960-1975 Diss. 1976-1990 Diss. 1991-

Student pat. (after) Founded laser related firm Became professor Advisor pubi, (before) Advisor pat. (before) Advisor is laser pioneer Student pubi. (before) Student pat. (before) Time (inverse) Cum. diss, at university (up to t-1) Engineering

Note: Dependent variables in bold script. Numbers of observations differ between variables related to dissertations (N = 4,841 ), dissertation advisors (N = 1,585) and the reduced sample of dissertations accounting for homonyms and timing (N = 3,880). See text for details.

Like Doktorvater, like Son? · 171 Table 2a Correlations between explanatory variables (Models 1 and 2) 4,841 obs.

(1)

Diss. 1960-1975(1)

1

Diss. 1976-1990(2)

-0.091

Cum. diss, at university -0.179 (up to t-1) (3)

(2)

(3)

1 -0.367

1

Values in italics denote significance at the 0.05 level. Individual VIF does not exceed 1.22 (mean: 1.16). VIF computed using colliri-package for STATA.

Table 2b Correlations between explanatory variables (Models 3-6) 1,585 obs.

(1)

Advisor with own laser dissertation (1) Third generation advisor (2) Advisor is laser pioneer (3) Time of entry (inverse) (4) Square (Time of entry (inverse)) (5)

1 0.374

(2)

(3)

(4)

(5)

1

0.261 -0.011

1

-0.064

-0.078

0.151

1

-0.058

-0.061

0.143

0.958

1

Values in italics denote significance at the 0.05 level. Individual VIF does not exceed 1.28 (mean: 1.15). VIF computed using collin-package for STATA.

Table 2c Correlations between explanatory variables (Models 7-17b) 3,880 obs.

(1)

Advisor pubi. (before)(1) Advisor pat. (before)(2) Advisor is laser pioneer (3) Student pubi. (before) (4) Student pat. (before) (5) Time (inverse) (6) Cum. diss, at univ. (up to t-1) (7) Engineering (8)

1

(2)

(3)

(4)

0.424

1

0.326

0.228

1

0.059

0.085

0.095

1

0.079

0.219

0.078

0.147

-0.252 -0.203

0.052 -0.015

0.281

0.333

0.065

-0.114

0.188 -0.124

0.065 -0.007

(5)

(6)

(7)

(8)

1 -0.046

1

0.124 -0.523

1

0.143 -0.084

0.178

Values in italics denote significance at the 0.05 level. Individual VIF does not exceed 1.54 (mean: 1.29). VIF computed using collin-package for STATA.

1

172 · Guido Buenstorf and Matthias Geissler

tangentially involved with laser research; on average, each one of them advised 2.5 laserrelated dissertations. For another 52 advisors we can identify two generations of students. In total, 57 of their 447 doctoral students subsequently advised laser-related dissertations themselves. Finally, there are seven first-generation advisors with three generations of students in laser research. These are singled out as advisors A-G in the first row of Figure 3. Interestingly, with one exception (advisor F), the number of laser-related dissertations these seven individuals advised tends to be small. It is also noteworthy that none of them had more than three students who became advisors of laser-related dissertations themselves. The most important first-generation advisor in our data is Hans Boersch (denoted as advisor A in Figure 3). Even though Boersch, a pioneer of electron microscopy and early entrant into laser research (Albrecht 2001), advised only four laser-related dissertations, his direct and indirect students account for a total of 231 laser-related dissertations or almost 5 % of our dataset. This is because three out of his four doctoral students in the dataset became highly productive advisors themselves, and one of them had a doctoral student who became the most prolific of all listed advisors.

Figure 3 A genealogy of non-medical laser research in Germany, 1960-2007

Looking at individual advisors in more detail, several characteristics of the data are noteworthy. First, the number of dissertations by advisor is highly skewed with a maximum of 82, a mean of 3.1 and a median of 1. Second, even though eight out of the top ten advisors are "first generation" laser researchers, differences in the number of laser-related dissertations by advisor are not fully explained by age and seniority. For instance, the largest number of dissertations was advised by a third-generation advisor who first shows up as a dissertation advisor in our data in 1997. Third, four out of the six leading dissertation advisors had high-profile positions at institutes dedicated to industrial laser applications. Finally, except for the Boersch school, large numbers of "indirect" students of first-generation advisors are due to a single doctoral student who subsequently became a highly productive advisor of laser-related dissertations. Thus, the skewness of the distribution found for the full dataset is also present among the students of individual advisors.

Like Doktorvater, like Son? · 173

7

Econometric analysis

7.1

The openness of German laser research

The discussion in the previous section indicated a considerable openness of German laser research to entrants coming from other fields of research. To see whether this openness underwent systematic changes as the field of laser research matured, we estimate two simple logit models using the full sample of 4,845 non-medical dissertations. 17 In these models, the dependent variable relates to the likelihood that a given dissertation was advised by a researcher who had written a laser-related dissertation her-/himself (Model 1 in Table 3) or by an early entrant as defined above (Model 2). We thus look at both "insider" advisors and early entrants. To analyze changes over time, two cohort dummies denote dissertations submitted 1960-75 and dissertations submitted 1976-1990, with the most recent dissertations making up the control group. We also enter a variable counting the stock of laser-related dissertations at the respective university prior to the year the target dissertation was submitted. This variable controls for university-level differences in the experience with laser research. Table 3 Likelihood of dissertations by advisor background (logistic regressions) for non-medical dissertations

Dissertations 1960-1975 Dissertations 1976-1990 Cum. dissertations at university (up to t-1) Constant Number of observations Log-likelihood (p > chi 2 ) Pseudo R 2

Model 1 (advisors with own laser dissertation)

Model 2 (advisor is laser pioneer)

-1.443*** (0.512) -0.234* (0.134) 0.006*** (0.001) -2.241*** (0.076) 4841

0.166 (0.327) 0.647*** (0.134) 0.004*** (0.001) -2.783*** (0.098) 4841

-1903.761 (0.000)

-1437.553 (0.000)

0.045

0.013

Standard errors in parentheses; *** pj * * r ï /k. /T,s;i V K i t V > V' 1, **

/

- .- .

Starbucks

- - · -

U.S. Steel

Figure 4 Employee growth rates

is still true for GM in recent years because the large drop in sales from 2008 to 2009 was not as large in percentage terms as in 1931-1932.) However, the span of average sales growth rates was very wide, between 6% and 88%. The four oldest companies have substantially lower average sales growth rates than the other three, and nowhere near the total average sales growth rate of 32.62% across all firms. 5.2

Employee growth

Five companies in our sample - Cisco, General Motors, IBM, Starbucks, and Microsoft displayed high maximum employee growth rates of 57% and higher. Three of these companies represented the youngest companies in the sample. Cisco led the pack in terms of fastest employee growth rate. All companies experienced their slowest employee growth rate after attaining their maximum employee growth rate, usually within a decade of one another (IBM and Sears took 30 and 40 years, respectively). Similar to the sales growth rates, the younger firms Cisco, Starbucks, and Microsoft maintained the highest average employee growth rates at 65.57%, 49.21% and 29%, respectively. On the other hand, General Motors, Sears, and US Steel consistently displayed lower average sales and employee growth rates, and experienced a wider range of fluctuating growth rates. 5.3

Some emerging patterns

The younger generation of companies (Microsoft, Starbucks, and Cisco) experienced their fastest sales growth rates within a few years of their IPOs, possibly because they raised resources through the IPOs. In contrast, the four older companies (US Steel, General Motors, Sears, and IBM) experienced their maximum sales growth rates about midway

224 · Johann Peter Murmann, Jenny Korn, and Hagen Worch

through their existence, which points to a cresting effect. In addition, though the four older-generation companies in our sample were operating during the Great Depression in the 1930s, none of them experienced their minimum employee growth rates during that period. In contrast, the older companies Sears and IBM both experienced their minimum sales and employee growth rates in the 1990s, the same decade that contained the slowest sales growth rates of the younger companies, Microsoft, Starbucks, and Cisco. It is worth noting that even Microsoft and Cisco, which provide software and infrastructure for the Internet era and belong to the younger group of firms in the sample, could not maintain their previous high level of growth rates. About half the companies in our sample attained maximum sales and employee growth rates through partnerships with other companies. Surprisingly, Cisco's fastest sales growth rate of 165.23% in 1991 was not attributed to mergers and acquisitions, which became its key growth strategy after 1993. Similarly, General Motors after its initial formation experienced no boost from partnering with other companies through mergers, acquisitions, or joint ventures until the early 1980s, despite reaching its fastest sales growth rate of 94.40% in 1947. Likewise, Sears' first acquisition was in 1981, but its maximum sales growth rate was nearly 35 years earlier in 1946 at 54.28%. About fifteen years after it was founded in 1901 through a merger of many companies, US Steel accomplished its fastest sales and employee growth rates of 72.14% and 32.23%, respectively. In contrast to the above companies, which gained their highest sales growth rates without partnerships, IBM benefited from its acquisition of subsidiary Science Research Associates, Inc., a Chicago-based publisher of educational, testing, and guidance materials in 1964, propelling the company to its maximum sales and employee growth rates of 57.28% and 71.88%. Starbucks' fastest sales growth rate of 75.63% in 1993 stemmed from a combination of completion of its IPO and its relationship with Barnes &C Noble stores, through which Starbucks gained a monopolistic entryway to reach a complementary target market. Microsoft reached its fastest growth rate of 75.12% in 1987 as a result of the company's first acquisition of Forethought, Inc. One striking pattern that emerges from the comparison of the seven companies in our sample is that in all cases the average sales growth exceeds the average employee growth. This means all firms in our sample realized a continued increase of employment efficiency. We can infer from this finding that they not only managed to increase permanently their production and sales capabilities, which is a crucial basis for their sustained growth, but they were also able to coordinate the expansions of their workforces. The seven firms appear to have continuously implemented the required organizational adaptations to suit a growing number of employees. In other words, they tend to have superior capabilities in the organization of their workforces to better leverage available human resources. Inflation could be another explanation for average sales growth rates exceeding average employee growth rates. s Even if it only partly explains the higher average sales growth rates, it counteracts the above argument. Thus, an inflation-based effect weakens the conclusion that the seven firms were permanently increasing their employment efficiency and therefore having substantially superior capabilities to manage employee growth. In fact, constantly growing firms may have such superior capabilities, but to a lesser extent than expected without considering inflation. This paper contributes to one important research stream in Ulrich Witt's broad agenda on an evolutionary approach to economics (Witt 2003), in which the theory of the firm plays 5

We thank an anonymous reviewer for this very valuable comment.

How Fast Can Firms Grow? • 225

a prominent role (Witt 1998, 2000). Witt highlighted the central role that entrepreneurs and their ability to coordinate firm activities play in an economic theory of the firm (1999). H e pioneered the idea that a cognitive dimension is crucial to understanding organizational development (1998). This was a departure f r o m the conventional view of organizational economics, which has focused on transaction costs and optimal sizes rather than the developmental process underlying organizational growth (Rathe/Witt 2001). Entrepreneurs must not only develop a vision of h o w to r u n a profitable firm, they also have to convey that conception to the members of their firm (Witt 2007). Conveying the vision of the business is a coordinative task that becomes increasingly difficult as a firm grows. An entrepreneur may address this challenge in various ways, thus presenting a variety of developmental paths (Witt 2000), such as restructuring an entrepreneurial firm and implementing managerial layers to enable firm growth. Other options include selling the firm or muddling t h r o u g h , either of which results in becoming a niche player or exiting the market. This paper provides some indirect evidence supporting Witt's evolutionary a p p r o a c h to the theory of the firm. O u r results s h o w that coordinating constant organizational growth is required to solve the coordination problem of managing a workforce in such a way that the o u t p u t per staff member is increasing on an ongoing basis, while the number of employees itself is permanently growing. The seven firms tended to mitigate emerging frictions by carrying out adequate structural adjustments. It would be an important topic for further research to look more closely into the relation between firm growth a n d organizational restructuring, and w h a t role the inspiration of the entrepreneurs and the executive management teams plays.

6

Conclusions and future research

Cisco, General M o t o r s , IBM, Microsoft, Sears, Starbucks, and US Steel represented a wide selection of US-based companies f r o m which we extracted sales and employee data to d r a w general trends about firm growth. Based on Cisco's figures, we d r a w the preliminary conclusion for companies that have existed for a few years that 1 6 2 % is the m a x i m u m sales growth rate in any one year that a company can grow w i t h o u t mergers and acquisitions, while a rate of employee g r o w t h a r o u n d 1 1 5 % appears to be the m a x i m u m annual employee growth rate even including some mergers and acquisitions. O n average the highest sales growth achieved by our sample of seven firms was about 9 6 . 5 7 % , and the average highest employee growth was 6 7 . 4 4 % . T o be more certain that these numbers are accurate estimates of the m a x i m u m firm growth rates, it would be valuable to study other fast-growing firms. Since we chose well-known, large firms, we can already say with confidence that the m a x i m u m g r o w t h rates of firms are clearly limited. W e observed one company achieving 1 6 2 % once. The important conclusion to be d r a w n f r o m these findings is that any established firm trying to grow by more than 2 0 0 % per year is unlikely to achieve its goal. M a n a g e r s need to k n o w that there are clear limits to growth of firms. O n the other end of the spectrum, General M o t o r s showed that a c o m p a n y experiencing a negative sales g r o w t h rate as high as —46.55% (—34.90% in terms of employees) can still survive and remain an important player in the economy. However, this tends only to be the case if such a decline is a rare event and not sustained for multiple years. In fact, the age and size of a firm may play an important role in helping a firm to survive a substantial decline.

226 · Johann Peter Murmann, Jenny Korn, and Hagen Worch

This study is not w i t h o u t limitations. O n e is that the sales and employee growth rates of the firms in our sample are observed f r o m the date of their IPO (in t w o cases even substantially later; in one case a few years earlier). We see this as a m a j o r limitation of the study because this means that our findings are applicable primarily to large firms that have already reached a certain size. O u r insights m a y be less relevant to small teams and entrepreneurial new ventures. Furthermore, high growth rates may be the result of acquisitions but also of delayed growth after some years of low growth rates, e.g., due to a recession. O n the other hand, one can argue that even if high g r o w t h rates are due to acquisitions a n d delayed expansions, continuously growing firms are distinct f r o m less successful firms as they are capable of managing these phases of accelerated growth. Future research is needed to analyze the times w h e n these limits were reached in the context of the broader history of the company and general economy - for example, fastest g r o w t h occurred within ten years of founding and slowest growth occurred within the last recession - and then determine whether a general trend for attaining these limits exists. In the future we plan to examine in more detail the conditions under which the fastest growth rates in the history of individual companies were achieved. A study of the history of individual companies f r o m a Penrosian point of view would shed light on the importance of g r o w t h of h u m a n capital, i.e., the management's ability to assess growth opportunities a n d overcome bottlenecks through capability building, networking a n d mergers. It w o u l d also enable researchers to study in more detail the limiting effect of a firm's h u m a n resources on growth, as Penrose (1959) had argued. Furthermore, a longitudinal data set for a large number of firms f r o m a variety of industries, sizes, classes, maturity levels, and national origins needs to be put together to obtain more accurate overall g r o w t h and employee growth rates. This w o u l d provide more accurate estimates as to h o w fast a firm can g r o w even under the best circumstances, and provide managers as well as policymakers with clear yardsticks of w h a t is feasible. It will also provide an empirical demarcation of the speeds of firm growth, which could someday be explained by a new theory of firm growth.

References Acar, A.C. (1993), The Impact of Key Internal Factors on Firm Performance: An Empirical Study of Small Turkish Firms. Journal of Small Business Management 31(4): 86-92. Acs, Z.J., D.B. Audretsch (1990), The Determinants of Small-Firm Growth in U. S. Manufacturing. Applied Economics 22(2): 143-153. Acs, Z.J., P. Mueller (2008), Employment Effects of Business Dynamics: Mice, Gazelles and Elephants. Small Business Economics 30: 85-100. Ahmed, H., N. Hamid (2011), Financing Constraints: Determinants and Implications for Firm Growth in Pakistan. The Lahore Journal of Economics 16: 317-346. Antoncic, J.Α., Β. Antoncic (2011), Employee Loyalty and Its Impact on Firm Growth. International Journal of Management and Information Systems 15: 81-87. Arellano, C., Y. Bai, J. Zhang (2012), Firm Dynamics and Financial Development. Journal of Monetary Economics 59(6): 533. Arrighetti, A. (1994), Entry, Growth and Survival of Manufacturing Firms. Small Business Economics 6(2): 127-137. Ayyagari, M., A. Demirgiiç-Kunt, V. Maksimovic (2010), Formal versus Informal Finance: Evidence from China. The Review of Financial Studies, 23(8): 3048.

How Fast Can Firms Grow? · 227

Baaij, M., A. de Jong, J. van Dalen (2011), The Dynamics of Superior Performance among the Largest Firms in the Global Oil Industry, 1954-2008. Industrial and Corporate Change 20(3): 789-824. Baer, D.H. (1987), H o w to Ensure Your Firm's Future. Public Relations Journal 43(8): 2 9 - 3 0 . Baliamoune-Lutz, M., Z. Brixiova, L. Ndikumana (2011), Credit Constraints and Productive Entrepreneurship in Africa. IZA Discussion Paper N o . 6193. Available at SSRN: http://ssrn.com/abstract= 1976524 (last retrieved 18.09. 2013). Balsmeier, B., D. Czarnitzki (2011), Ownership Concentration, Institutional Development and Firm Performance in Central and Eastern Europe. ZEW - Centre for European Economic Research Discussion Paper No. 10-096. Available at SSRN: http://ssrn.com/ abstract=1739752 (last retrieved 18.09. 2013). Bamiatzi, V.C., R.R. Sinkovics (2010), The Role of H u m a n and Social Capital Traits in SMEs Over-Performance During Industrial Downturns: Theoretical Development. Available at SSRN: http://ssrn.com/abstract= 1620390 (last retrieved 18.09. 2013). Barringer, B.R., D.W. Greening (1998), Small Business Growth through Geographic Expansion: A Comparative Case Study. Journal of Business Venturing 13(6): 467-492. Bertoni, F., M.G. Colombo, L. Grilli (2011), Venture Capital Financing and the Growth of High-tech Start-ups: Disentangling Treatment from Selection Effects. Research Policy 40(7): 1028. Birley, S., P. Westhead (1994), A Taxonomy of Business Start-up Reasons and Their Impact on Firm Growth and Size. Journal of Business Venturing 9: 7 - 3 1 . Bonaccorsi, Α., S. Giannangeli (2010), One or More Growth Processes? Evidence from New Italian Firms. Small Business Economics 35(2): 137-152. Bratkovic, T., Β. Antoncic, A.F. DeNoble (2012), Relationships between Networking, Entrepreneurial Self-Efficacy and Firm Growth: The Case of Slovenian Companies. Ekonomska Istrazivanja 25: 7 3 - 8 7 . Brinckmann, J., S. Salomo, H.G. Gemuenden (2011), Financial Management Competence of Founding Teams and Growth of N e w Technology-Based Firms. Entrepreneurship Theory and Practice 35(2): 217-243. Bronars, S.G., D.R. Deere (1993), Union Organizing Activity, Firm Growth, and the Business Cycle. American Economic Review 83: 203-220. Bryson, J.R., D. Keeble, P. Wood (1997), The Creation and Growth of Small Business Service Firms in Post-industrial Britain. Small Business Economics 9(4): 345-360. Bunnell, D., A. Brate (2000), Making the Cisco Connection: The Story Behind the Real Internet Superpower, New York: John Wiley &c Sons. Bushman, R.M., A.J. Smith, F. Zhang (2011), Investment Cash Flow Sensitivities Really Reflect Related Investment Decisions. Cabrai, L. (1995), Sunk Costs, Firm Size and Firm Growth. Journal of Industrial Economics 43(2): 161-172. Cabrai, L.M.B., J. Mata (2003), On the Evolution of the Firm Size Distribution: Facts and Theory. American Economic Review 93(4): 1075-1090. Capasso, M., E. Cefis (2012), Firm Size and Growth Rate Variance: The Effects of Data Truncation. Review of Industrial Organization 41(3): 193-205. Caves, R.E. (1998), Industrial Organization and N e w Findings on the Turnover and Mobility of Firms. Journal of Economic Literature 36(4): 1947-1982. Chakrabarti, A.K. (1990), Scientific Output of Small and Medium Size Firms in High Tech Industries. IEEE Transactions on Engineering Management 37: 4 8 - 5 2 . Chan, K., Y.-H. Lin, Y. Wang (2011), Bad News or Good News? The Impact of R & D Reductions. http://ssrn.com/abstract=1786951 (last retrieved 18.09. 2013). Chan, R.C. (2011), Financial Constraints, Corruption and Growth: Firm Level Evidence. Available at SSRN: http://ssrn.com/abstract=1718722 (last retrieved 18.09. 2013). Chittenden, F., G. Hall, P. Hutchinson (1996), Small Firm Growth, Access to Capital Markets and Financial Structure: Review of Issues and an Empirical Investigation. Small Business Economics 8: 5 9 - 6 7 .

228 · Johann Peter Murmann, Jenny Korn, and Hagen Worch

Choi, B.P. (2010), The U.S. Property and Liability Insurance Industry: Firm Growth, Size, and Age. Risk Management and Insurance Review 13(2): 207-224. Cisco corporate website, http://www.cisco.com Cisco corporate website's acquisition summary, http://www.cisco.com/web/about/doing_ business/corporate_development/acquisitions/ac_year/about_cisco_acquisition_years_list.html (last retrieved 18.09. 2013). Coad, A. (2009), The Growth of Firms: A Survey of Theories and Empirical Evidence. Cheltenham, UK. Coad, Α., J.P. Tamvada (2012), Firm Growth and Barriers to Growth among Small Firms in India. Small Business Economics 39(2): 383—400. Coad, Α., A. Segarra-Blasco, M . Teruel (2011), Like Milk or Wine: Does Firm Performance Improve with Age? XREAP Working Paper N o . 2010-10. Available at SSRN: http://ssrn.eom/abstract= 1815028 (last retrieved 18.09. 2013). Cordes, C., P.J. Richerson, G. Schwesinger (2010), H o w Corporate Cultures Coevolve with the Business Environment: The Case of Firm Growth Crises and Industry Evolution. Journal of Economic Behavior & Organization 76(3): 465-480. Cordes, C., P.J. Richerson, R. McElreath, P. Strimling (2008), A Naturalistic Approach to the Theory of the Firm: The Role of Cooperation and Cultural Evolution. Journal of Economic Behavior &c Organization 68: 125-139. Cornelius, T. (1979), Organizing a Firm for Growth. Accountancy 90(1027): 81-82. Corsino, M., R. Gabriele (2011), Product Innovation and Firm Growth: Evidence from the Integrated Circuit Industry. Industrial and Corporate Change 20: 2 9 - 5 6 . Covin, J.G., D.P. Slevin (1998), Adherence to Plans, Risk Taking, and Environment as Predictors of Firm Growth. Journal of High Technology Management Research 9(2): 207-237. Cubbin, J., G. Hall (1983), Directors' Remuneration in the Theory of the Firm: Specification and Testing of the Null Hypothesis. European Economic Review 20(1-3): 333-348. Daily, C.M., D.R. Dalton (1992), The Relationship between Governance Structure and Corporate Performance in Entrepreneurial Firms. Journal of Business Venturing 7(5): 375-386. Danis, W.M., D.S. Chiaburu, M.A. Lyles (2010), The Impact of Managerial Networking Intensity and Market-based Strategies on Firm Growth During Institutional Upheaval: A Study of Small and Medium-sized Enterprises in a Transition Economy. Journal of International Business Studies 41(2): 287-307. Das, S. (1995), Size, Age and Firm Growth in an Infant Industry: The Computer Hardware Industry in India. International Journal of Industrial Organization 13: 111-126. Daskalopoulou, I., A. Petrou (2010), Entrepreneurial Growth Expectations and Information Flows in Networks. Journal of Small Business and Enterprise Development 17(3): 334—349. Datta, D.K., J.P. Guthrie (1994), Executive Succession: Organizational Antecedents of CEO Characteristics. Strategic Management Journal 15(7): 569-577. Davidsson, P. (1991), Continued Entrepreneurship: Ability, Need, and Opportunity as Determinants of Small Firm Growth. Journal of Business Venturing 6(6): 405—429. Davies, S.W., P.A. Geroski (1997), Changes in Concentration, Turbulence, and the Dynamics of Market Shares. Review of Economics Sc Statistics 79(3): 383-391. Degryse, H., L. Lu, S.R.G. Ongena (2012), Informal or Formal Financing? Or Both? First Evidence on the Co-Funding of Chinese Firms. Available at SSRN: http://ssrn.com/abstract=2023751 (last retrieved 18.09. 2013). Demirguc-Kunt, Α., V. Maksimovic (1998), Law, Finance, and Firm Growth. Journal of Finance 53(6): 2107-2137. Dunne, P., A. Hughes (1994), Age, Size, Growth and Survival: UK Companies in the 1980s. Journal of Industrial Economics 42(2): 115-140. Eidson, T. (1993), Changing Times Require Rededication to Service. Public Relations Journal 49(8): 39-40. Eisenhardt, K.M., C.B. Schoonhoven (1990), Organizational Growth: Linking Founding Team, Strategy, Environment, and Growth Among U.S. Semiconductor Ventures, 1978-1988. Administrative Science Quarterly 35(3): 504-529.

How Fast Can Firms Grow? • 229

Eriksson, G. (1984), Growth, Entry and Exit of Firms. Scandinavian Journal of Economics 86: 52-67. Evans, D.S. (1987a), The Relationship between Firm Growth, Size, and Age: Estimates for 100 Manufacturing Industries. Journal of Industrial Economics 35(4): 567-581. Evans, D.S. (1987b), Tests of Alternative Theories of Firm Growth, Journal of Political Economy, 95(4): 657-674. Fergus, J. (1995), The New World Order of Lateral Partner Hiring. International Financial Law Review 14(7): 4 2 - 4 4 . Foss, N.J. (1998), Edith Penrose and the Penrosians - Or, Why There is Still So Much to Learn from the Theory of the Growth of the Firm. Special issue of Cahiers de l'ISMEA - Série Oeconomica. Fotopoulos, G., I. Giotopoulos (2010), Gibrat's Law and Persistence of Growth in Greek Manufacturing. Small Business Economics 35(2): 191-202. Fulton, M.E., J.R. Fulton, J.S. Clark, C. Parliament (1995), Cooperative Growth: Is It Constrained? Agribusiness 11(3): 245-261. Gander, J.P. (1991), Managerial Intensity, Firm Size and Growth. Managerial &C Decision Economics 12(3): 261-266. García-Manjón, J.V., M.E. Romero-Merino (2012), Research, Development, and Firm Growth: Empirical Evidence from European Top R & D Spending Firms. Research Policy 41(6): 1084-1092. General Motors corporate website, http://www.gm.com Geroski, P., I. Small, C. Walters (2000), Agglomeration Economies, Technology Spillovers and Company Productivity Growth. Pp. 2 3 6 - 2 6 7 in: R. Barrell, G. Mason, M. O ' M a h o n y (eds.), Productivity, Innovation and Economic Performance. New York: Cambridge University Press. Gibrat, R. (1931), Les inégalités économiques; applications: aux inégalités des richesses, à la concentration des entreprises, aux population des villes, aux statistiques des familles, etc., d'une loi nouvelle, la loi de l'effet proportionnel. Paris. Gifford, S. (1992), Allocation of Entrepreneurial Attention. Journal of Economic Behavior & Organization 19(3): 265-284. Giotopoulos, I., G. Fotopoulos (2010), Infra-Industry Growth Dynamics in the Greek Services Sector: Firm-Level Estimates for ICT-Producing, ICT-Using, and Non-ICT Industries. Review of Industrial Organization 36: 59-74. Glushkov, D., K.S. Bardos (2012), Importance of Catering Incentives for Growth Dynamics. The Journal of Behavioral Finance 13(4): 259-280. Goronzy, F. et al. (1974), Factors in Corporate Growth. Management International Review 14(4/5): 75-85. Greening, D.W., B.R. Barringer, G. Macy (1996), A Qualitative Study of Managerial Challenges Facing Small Business Geographic Expansion. Journal of Business Venturing 11 (4): 233-256. Greenwald, B.C., M . Kohn, J.E. Stiglitz (1990), Financial Market Imperfections and Productivity Growth. Journal of Economic Behavior &c Organization 13(3): 321-345. Hay, M., K. Kamshad (1994), Small Firm Growth: Intentions, Implementation and Impediments. Business Strategy Review 5(3): 4 9 - 6 8 . Hayashi, F., T. Inoue (1991), The Relation Between Firm Growth and Q with Multiple Capital Goods: Theory and Evidence from Panel Data on Japanese Firms. Econometrica 59(3): 731-753. Henrekson, M., D. Johansson (2010), Gazelles as Job Creators: A Survey and Interpretation of the Evidence. Small Business Economics 35(2): 227-244. Hitt, M.A., R.D. Ireland, H.-U. Lee (2000), Technological Learning, Knowledge Management, Firm Growth and Performance: An Introductory Essay. Journal of Engineering & Technology Management 17(3—4): 231-246. Hussain, J.G., J.M. Scott, R.T. Harrison, C. Millman (2010), Enter the Dragoness: Firm Growth, Finance, Guanxi, and Gender in China. Gender in Management 25(2): 137-156. Huynh, K.P., R.J. Petrunia (2010), Age Effects, Leverage and Firm Growth. Journal of Economic Dynamics & Control 34(5): 1003-1013.

230 • Johann Peter Murmann, Jenny Korn, and Hagen Worch

Huynh, K.P., R.J. Petrunia (2011), Post-Entry Struggle for Life and Pre-Exit Shadow of Death from a Financial Perspective. Midwest Finance Association 2012 Annual Meetings Paper. Available at SSRN: http://ssrn.com/abstract=1890649 (last retrieved 18.09. 2013). IBM corporate website, http://www.ibm.com Ikeda, K., N. Doi (1983), The Performance of Merging Firms in Japanese Manufacturing Industry: 1964-75. Journal of Industrial Economics 31(3): 257-266. Iwai, K. (1984), Schumpeterian Dynamics, Part II: Technological Progress, Firm Growth and 'Economic Selection'. Journal of Economic Behavior & Organization 5(3-4): 321-351. JasnaAuer, Α., Β. Antoncic (2011), Employee Satisfaction, Intrapreneurship and Firm Growth: A Model. Industrial Management + Data Systems 111(4): 589-607. Jovanovic, B. (1982), Selection and the Evolution of Industry. Econometrica 50(3): 649-670. Kangasharju, A. (2000), Growth of the Smallest: Determinants of Small Firm Growth during Strong Macroeconomic Fluctuations. International Small Business Journal 19: 28—43. Kania, J.J., J.R. McKean (1981), Profit, Growth, and the Marris Theory of the Firm: A Short Run Empirical Test. American Economist 25: 63-66. Khan, K.S., M.W. Siddiqi (2011 ), Determinants of Firm Growth: An Empirical Examination of SMEs in Gujranwala, Gujrat and Sialkot Districts. Interdisciplinary Journal of Contemporary Research in Business 3: 1389-1409. Kiani, K.M., E.H. Chen, Z. Madjd-Sadjadi (2012), Financial Factors and Firm Growth: Evidence from Financial Data on Taiwanese Firms. Quantitative Finance 12(8): 1299. Klaas, B.S., J. McClendon, T.W. Gainey (2000), Managing HR in the Small and Medium Enterprise: The Impact of Professional Employer Organizations. Entrepreneurship Theory 8c Practice 25: 107-124. Klapper, L.F., I. Love (2011), The Impact of Business Environment Reforms on New Firm Registration. Available at SSRN: http://ssrn.com/abstract=1786802 (last retrieved 18.09. 2013). Klepper, S., E. Graddy (1990), The Evolution of New Industries and the Determinants of Market Structure. Rand Journal of Economics 21: 27-44. Kosová, R., F. Lafontaine (2010), Survival and Growth in Retail and Service Industries: Evidence from Franchised Chains. The Journal of Industrial Economics 58(3): 542. Kotha, S., A. Nair (1995), Strategy and Environment as Determinants of Performance: Evidence from the Japanese Machine Tool Industry. Strategic Management Journal 16(7): 497-518. Kumar, M.S. (1985), Growth, Acquisition Activity and Firm Size: Evidence from the United Kingdom. Journal of Industrial Economics 33(3): 327-338. Lane, S.J. (1995), Business Starts during the Transition in Hungary. Journal of Business Venturing 10(3): 181-194. Lang, L., E. Ofek, R.M. Stulz (1996), Leverage, Investment, and Firm Growth. Journal of Financial Economics 40: 3-29. Langlois, R.N., P.L. Robertson (1995), Firms, Markets, and Economic Change: A Dynamic Theory of Business Institutions. London/New York. Le Bas, C., Ν. Haned, Α. Colombelli (2011), On Firm Growth and Innovation - Some New Empirical Perspectives Using French CIS, 1992-2004. ICER Working Paper No. 7/2011. Available at SSRN: http://ssrn.com/abstract=1857426 (last retrieved 18.09. 2013). Lee, C.-Y. (2010), A Theory of Firm Growth: Learning Capability, Knowledge Threshold, and Patterns of Growth. Research Policy 39(2): 278-289. Levy, D., R.P. Wilder, I.B. Tucker (1984), Testing Stigler's Interpretation of "The Division of Labor Is Limited by the Extent of the Market"/Trends in Vertical Integration: Reply. Journal of Industrial Economics 32(3): 377-392. Lin, E., T.M.Y. Lin, L. Bou-Wen (2010), New High-tech Venturing as Process of Resource Accumulation. Management Decision 48(8): 1230-1246. Lorenzoni, G., O.A. Ornati (1988), Constellations of Firms and New Ventures. Journal of Business Venturing 3: 41-57. Luttmer, E.G.J. (2011), On the Mechanics of Firm Growth. The Review of Economic Studies 78(3): 1042.

H o w Fast Can Firms Grow? • 231

Majumdar, S. (2010), Fiber in the Backbone! The Impact of Broadband Adoption on Firm Growth in Network Markets. Economics of Innovation and New Technology 19(3): 283-293. Malevergne, Y., A.I. Saichev, D. Sornette (2010), Zipf's Law for Firms: Relevance of Birth and Death Processes. Available at SSRN: http://ssrn.com/abstract=1083962 (last retrieved 18.09. 2013). Mata, J. (1994), Firm Growth during Infancy. Small Business Economics 6: 27-39. Mateev, M., Y. Anastasov (2010), Determinants of Small and Medium Sized Fast Growing Enterprises in Central and Eastern Europe: A Panel Data Analysis. Financial Theory and Practice 34(3): 269-295. McEvily, B., J. Jaffee, M. Tortoriello (2012), Not All Bridging Ties Are Equal: Network Imprinting and Firm Growth in the Nashville Legal Industry, 1933-1978. Organization Science 23(2): 547-563. McPherson, M.A. (1996), Growth of Micro and Small Enterprises in Southern Africa. Journal of Development Economics 48(2): 253-277. Merz, G.R., P.B. Weber, V.B. Laetz (1994), Linking Small Business Management with Entrepreneurial Growth. Journal of Small Business Management 32(4): 48-60. Microsoft corporate website, http://www.microsoft.com Miner, J.B. (2000), Testing a Psychological Typology of Entrepreneurship Using Business Founders. Journal of Applied Behavioral Science 36: 43-69. Miner, J.B., N.R. Smith, J.S. Bracker (1989), Role of Entrepreneurial Task Motivation in the Growth of Technologically Innovative Firms. Journal of Applied Psychology 74(4): 554—560. Misa, T.J. (1998), A Nation of Steel: The Making of Modern America, 1865-1925. Johns Hopkins University Press. Moreno, J.d.J., L.L. Castillo (2011), Corporate Growth, Age and Ownership Structure: Empirical Evidence in Spanish Firms. Journal of Business Economics and Management 12:164-196. N'Cho-Oguie, C., D.L.P. Blakley, L.W.P. Murray, M.B.P. Smith (2011), Econometric Analysis of Functional Relationship between Inflation and Growth of Firms in South Africa: Empirical Research Findings. Journal of Financial Management & Analysis 24(2): 1-19. Nkurunziza, J.D. (2010), The Effect of Credit on Growth and Convergence of Firm Size in Kenyan Manufacturing. The Journal of International Trade & Economic Development 19(3): 465-494. Orser, B.J., S. Hogarth-Scott, A.L. Riding (2000), Performance, Firm Size, and Management Problem Solving. Journal of Small Business Management 38(4): 42-58. Park, Y., J. Shin, T. Kim (2010), Firm Size, Age, Industrial Networking, and Growth: A Case of the Korean Manufacturing Industry. Small Business Economics 35(2): 153-168. Parker, S.C., D.J. Storey, A. van Witteloostuijn (2010), What Happens to Gazelles? The Importance of Dynamic Management Strategy. Small Business Economics 35(2): 203-226. Pelham, A.M., D.T. Wilson (1996), A Longitudinal Study of the Impact of Market Structure, Firm Structure, Strategy, and Market Orientation Culture on Dimensions of Small-firm Performance. Journal of the Academy of Marketing Science 24: 27-43. Peneder, M.R. (2010), The Impact of Venture Capital on Innovation Behaviour and Firm Growth. An International Journal of Entrepreneurial Finance 12(2), pp. 83-107. Peng, M.W. (1997), Firm Growth in Transitional Economies: Three Longitudinal Cases from China, 1989-96. Organization Studies 18(3): 385-413. Penrose, E.T. (1959), The Theory of the Growth of the Firm. New York: John Wiley. Pervan, M., T.P. Kramaric (2012), Factors That Influence Firm Performance: A Dynamic Panel Data Analysis for Croatian Insurance Industry. The Business Review, Cambridge 19(2): 195-202. Piergiovanni, R. (2010), Gibrat's Law in the "Third Italy": Firm Growth in the Veneto Region. Growth and Change 41: 28-58. Pitelis, C., A. Panagopoulos (2010), Open Innovation and the Management of Intra-firm Conflict. Available at SSRN: http://ssrn.com/abstract=1440279 (last retrieved 18.09. 2013). Ponikvar, N., K.Z. Kejzar (2011 ), Technology Intensity, Foreign Competition and Firm Growth. Challenges of Europe: International Conference Proceedings, Split, Croatia.

232 • Johann Peter Murmann, Jenny Korn, and Hagen Worch

Prescott, E.C., M. Visscher (1980), Organization Capital. Journal of Political Economy 88(3): 446-461. Pugh, E. (1984), Memories That Shaped an Industry: Decisions Leading to IBM System/360. Cambridge, MA: MIT Press. Pugh, E. (1995), Building IBM: Shaping an Industry and Its Technology. MIT Press. Rahaman, M.M. (2011 ), Access to Financing and Firm Growth. Journal of Banking & Finance 35(3): 709-723. Rathe, K., U. Witt (2001 ), The Nature of the Firm - Static versus Developmental Interpretations. Journal of Management and Governance 5(3—4): 331-351. Reichstein, T., M.S. Dahl, Β. Ebersberger, M.B. Jensen (2010),The Devil Dwells in the Tails. Journal of Evolutionary Economics 20(2): 219-231. Rose, P.S. (1992), Interstate Banking: Performance, Market Share, and Market Concentration Issues. Antitrust Bulletin 37(3): 601-630. Santarelli, E., H.T. Tran (2011), Growth of Incumbent Firms and Entrepreneurship in Vietnam. Quaderni DSE Working Paper No. 785. Available at SSRN: http://ssrn.com/ abstract=1927942 (last retrieved 18.09. 2013). Schwarzkopf, Y., J.D. Farmer (2010), What Drives Mutual Fund Asset Concentration? Available at SSRN: http://ssrn.com/abstract=l 173046 (last retrieved 18.09. 2013). Sears corporate website, http://www.sears.com Shamis, G.S., N. Lewandowski (1996), A Piece of the Action. Journal of Accountancy 182(2): 52-54. Shane, S. (1996a), Hybrid Organizational Arrangements and Their Implications for Firm Growth and Survival: A Study of New Franchisors. Academy of Management Journal 39: 216-234. Shane, S. (1996b), Why Franchise Companies Expand Overseas. Journal of Business Venturing, 11(2), 73-88. Sickles, M.W. (1999), Managing the Workforce to Assure Shareholder Value. HR Focus 76(8): 1-14. Siddharthan, N.S., S. Lall (1982), The Recent Growth of the Largest U.S. Multinationals. Oxford Bulletin of Economics & Statistics 44: 1-13. Smith, R.B. (1984), Building on 75 Years of Excellence: The General Motors Story. Newcomen Society in North. Stam, E. (2010), Growth beyond Gibrat: Firm Growth Processes and Strategies. Small Business Economics 35(2): 129-135. Starbucks corporate website, http://www.starbucks.com Starr, E. (1988), The Growth of Small Manufacturers: 1976-84. Business Economics 23(2): 41-45. Stenholm, P. (2011), Innovative Behavior as a Moderator of Growth Intentions. Journal of Small Business Management 49(2): 233-251. Stevenson, H.H., J.C. Jarrillo-Mossi (1986), Preserving Entrepreneurship as Companies Grow. Journal of Business Strategy 7: 10-23. Storey, D.J. (1994a), New Firm Growth and Bank Financing. Small Business Economics 6(2): 139-150. Storey, D.J. (1994b), The Role of Legal Status in Influencing Bank Financing and New Firm Growth. Applied Economics 26(2): 129-136. Sutton, J. (1997), Gibrat's Legacy. Journal of Economic Literature 35: 40-59. Thompson, R.S. (1994), The Franchise Life Cycle and the Penrose Effect. Journal of Economic Behavior and Organization 24(2): 207-218. Tsai, K.-H., M.-H. Hsieh, E.J. Hultink (2011), External Technology Acquisition and Product Innovativeness: The Moderating Roles of RßcD Investment and Configurational Context. Journal of Engineering and Technology Management 28(3): 184-200. Tushman, M.L., E. Romanelli (1985), Organizational Evolution: A Metamorphosis Model of Convergence and Reorientation. Pp. 171-222 in: L. Cummings and B. Staw (eds.), Research in Organizational Behavior. Greenwich, CT. US Steel corporate website, http://www.ussteel.com/

How Fast Can Firms Crow? · 233

Wagner, J. (1992), Firm Size, Firm Growth, and Persistence of Chance: Testing Gibrat's Law with Establishment Data f r o m Lower Saxony, 1 9 7 8 - 1 9 8 9 . Small Business Economics 4(2): 125-131. Wagner, J. (1994), The Post-entry Performance of N e w Small Firms in German M a n u f a c t u r i n g Industries. Journal of Industrial Economics 42(2): 1 4 1 - 1 5 4 . Wallace, J. (1992), H a r d Drive: Bill Gates and the M a k i n g of the Microsoft Empire. N e w York. Wang, C.L., L. Altinay (2012), Social Embeddedness, Entrepreneurial Orientation and Firm G r o w t h in Ethnic Minority Small Businesses in the UK. International Small Business Journal 30: 3 - 2 3 . Weil, G. (1977), Sears, Roebuck, USA: The Great American Catalog Store and H o w It Grew. Briarcliff M a n o r , N Y . W h a r t o n Research Data Services (WRDS) database, http://wrds.wharton.upenn.edu/. Wiley, S. (2011), Building Morale: The Key to Firm Growth. CPA Practice M a n a g e m e n t Forum 7(2): 1 2 - 1 9 . Witt, U. (1998), Imagination and Leadership - The Neglected Dimension of an Evolutionary Theory of the Firm. Journal of Economic Behavior & Organization, 35(2): 1 6 1 - 1 7 7 . Witt, U. (1999), D o Entrepreneurs Need Firms? A Contribution to a Missing Chapter in Austrian Economics. Review of Austrian Economics 11(1-2): 9 9 - 1 0 9 . Witt, U. (2000), Changing Cognitive Frames - Changing Organizational Forms: An Entrepreneurial Theory of Organizational Development. Industrial and Corporate Change 9(4): 7 3 3 - 7 5 5 . Witt, U. (2003), The Evolving Economy: Essays on the Evolutionary Approach to Economics. Cheltenham, N o r t h h a m p t o n , M A . Witt, U. (2007), Firms as Realizations of Entrepreneurial Visions. Journal of M a n a g e m e n t Studies 44(7): 1 1 2 5 - 1 1 4 0 . W o o d w a r d , W., D.T. Dingle (1988), In Pursuit of Profits. Black Enterprise 19(4): 5 7 - 6 2 . Yazdanfar, D. (2012), T h e Impact of Financing Pattern on Firm G r o w t h : Evidence f r o m Swedish Micro Firms. International Business Research 5(9): 1 6 - 2 5 . Yazdanfar, D., A.K. Salman (2012), The Life Cycle of G r o w t h Path a m o n g Micro Firms: Swedish Data. International Business Research 5(10): 1 0 7 - 1 1 4 . Z h a o , L., J.D. A r a m (1995), N e t w o r k i n g and G r o w t h of Young Technology-intensive Ventures in China. Journal of Business Venturing 10(5): 3 4 9 - 3 7 0 . Prof. J o h a n n Peter M u r m a n n , Ph.D., Australian School of Business, University of N e w South Wales, Sydney, N S W 2 0 5 2 , Australia, j [email protected] Jenny Korn, Department of Communication, University o f I l l i n o i s a t C h i c a g o , 1 0 0 7 W Harrison Street, Chicago IL 60607, USA. [email protected] Dr. H a g e n Worch, Institute for M a n a g e m e n t & Innovation, Swiss Distance University of Applied Sciences / Fernfachhochschule Schweiz, Althardstrasse 60, 8105 Regensdorf, Switzerland. [email protected]

Jahrbücherf. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

Firm Growth and the Spatial Impact of Geolocated External Factors Matthias Duschl* Philipps University of Marburg Antje Schimke Karlsruhe Institute of Technology and Fraunhofer-Gesellschaft München Thomas Brenner Philipps University of Marburg Dennis Luxen Karlsruhe Institute of Technology JEL C31; D92; L25; R11 Firm growth; external factors; distances; quantile regression; relatedness; universities; public research; graduates.

Summary This paper studies the relationship between firm growth and external factors. Externalities from related economic, public research and higher educational activities are traced back to specific locations in space. The spatial characteristics of their impact are examined within a distancebased, micro-founded approach. Applying quantile regression techniques on a large sample of German firms, we empirically disentangle the complex interplay between internal factors (firm size), external factors and their spatial extent. In particular, we find that the larger firms are, the more diverse are the activities they benefit from and that the geographical meaning of "nearby" depends on the kind of activity.

1

Introduction

Firms are a central unit of analysis in economics. Understanding their activities and development is fundamental to understand economies and economic growth. Nevertheless, the dynamics of firms have been neglected for a long time within economics. Neoclassical theory has focussed on static analyses of competition scenarios and aggregated production functions, ignoring the individual developments of firms. Empirical studies have argued that firm growth is a purely random process.

* The authors would like to thank Alex Goad, Guido Bünstorf as well as the participants of the Geography of Innovation Conference 2012 in St. Etienne and of the workshop on Firm Growth and Innovation 2012 in Tarragona for helpful comments and suggestions on an earlier draft of this paper. The paper also benefited from detailed comments of two anonymous referees. The usual disclaimers apply.

Firm Growth and the Spatial Impact of Geolocated External Factors • 235

Nelson and Winter (1974) made firms the central actors of a dynamic perspective on economies, explaining their developments on the basis of individual decision rules and selection mechanisms on the market. Since their path-breaking work evolutionary economics as well as the study of firm growth has advanced. Firm growth has been studied empirically in many works, mainly focussing on the characteristics and the internal determinants of growth. Evolutionary economics has extended its scope tremendously with topics reaching from market competition and industry evolution to game theory and consumption behaviour (see, e.g., Witt 2008). These topics also comprise the role of knowledge for economic activities (see, e.g., Witt et al. 2012) and the study of innovation processes as well as their geographic context. Regional innovation systems and clusters are concepts in this field that have received much attention in recent years (see, e.g., Asheim/Gertler 2006 and Asheim et al. 2006). The literature agrees that knowledge and local interaction play an important role for economic growth. This implies that they should be also important determinants for firm growth. Yet another theme of evolutionary economics is the heterogeneity of the economies' constituents. Hence, the question arises which firms are affected by which kinds of activities in their local environment? Unfortunately, empirical studies that examine the influence of local knowledge sources and industrial agglomeration on the growth of firms are rare. Detailed knowledge about these relationships would provide the link between the literature on regional innovation systems and clusters and the original idea by Nelson and Winter to explain economic growth on the level of firms. Given this missing link, the aim of this paper is to examine the impact of spatial externalities, which can be traced back to specific locations in space, on the growth of firms. By doing so, new light is shed on the complex interplay between factors internal and external to the firms. This is achieved by calculating firm-specific location variables, which measure the firms' access (in travel time) to related economic, public research and higher educational activities. Using a quantile regression framework, the impact of these activities is compared for employment growth of small, medium and large firms in Germany. Owing to the lack of a priori knowledge on the impacts' spatial extent, this paper endogenizes distances. The findings suggest that firm size, an indicator for the firms' necessity and capacity to absorb and implement external knowledge, critically shapes how external activities affect firm growth. Firms benefit from public research and higher educational activities at a geographical scale much smaller than usually assumed as "regional", whereas related economic activities tend to transcend predefined regional boundaries. Only for large firms, the negative effect of technologically narrowly related and geographically nearby activities dominate. The paper is structured as follows. In the second section, expectations on the spatial impact of external factors on firm growth are derived from the existing literature. The empirical method is outlined in the third section, while data issues are discussed in the fourth section. Section five discusses the results and section six concludes.

2

Literature and hypothesis

According to the resource-based view of firm growth (Penrose 1959), the ability of generating, combining, and ultimately translating new knowledge into economic opportunities depends on factors internal and external to the firm. Internal factors, like size, age, industry affiliation or R & D , have been repeatedly studied in the economic literature (for an

236 · Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

overview see Coad 2 0 0 9 ) . External factors can be addressed within the conceptual framework of knowledge externalities or spillovers (Raspe/vanOort 2 0 0 8 ) . In this paper, we focus on external knowledge as a driver of firm growth, and distinguish between different kinds of knowledge sources. These are naturally more abundant in agglomerations, yet some technological relatedness is necessary for knowledge spillovers to effectively take place (Frenken et al. 2 0 0 7 ) . Here we ask, what is the right degree of relatedness between the firms' activities and the surrounding economic activities? Furthermore, do public research and higher educational activities, with the explicit aim to transfer new knowledge into the economy, unequivocally foster firm growth? By taking a less deterministic perspective, we expect that it depends on both the firms' necessity and capacity to absorb and apply external knowledge, which in turn are related to the firms' internal resources and growth dynamics. Finally, the spatial dimension of the external factors has to be addressed. Although the literature claims that knowledge externalities are spatially bounded, no consensus exists on their spatial extent (Döring/Schnellenbach 2 0 0 6 ) . By summarising the literature, we derive the hypotheses that aim to disentangle the complex interrelation between firm growth, external factors and internal characteristics. Regarding the spatial dimension of the external factors, we will conclude that this question has to be determined empirically.

2.1

Firm growth and its growth related external factors

The impact of related economic

activities

Already Marshall (1890) stated that firms might perform better when located within agglomerations. In respect to the economic geography literature, traditionally two kinds of externalities are distinguished. Whilst positive effects of localization economies occur through specialization of similar industries, the positive effects of urbanization economies arise from the variety of different industries. After many decades of intensive research, the literature on regional agglomeration remains rather indecisive about the real effect of specialization versus diversification at the regional level (Beaudry/Schiffauerova 2 0 0 9 ) . Reasons for this inconclusiveness are manifold, here we highlight what seems to us most relevant. Studies following this stream of literature mostly focus on regional economies. However, the aggregate regional outcome is composed of the foundation of new firms and the growth and survival of existing firms, which might be affected differently by agglomeration economies. Whilst a positive effect on start-ups (e.g., Sorensen/Audia 200) and survival (e.g., Renski 2 0 1 1 ) is well documented, "the main gap in our empirical understanding concerns the effect of localization economies on firm performance, which some may even consider the key question in economic geography at large" (Frenken et al. 2 0 1 1 ) . Mostly analysing the agglomeration-productivity (e.g., Rigby/Essletzbichler 2 0 0 2 ; Baldwin et al. 2 0 0 8 ; Drucker/Feser 2 0 1 2 ) or agglomeration-innovation relationship (e.g., Feldman 2 0 0 0 ; Beugelsdijk 2 0 0 7 ) , only few exceptions exist that link agglomeration to growth at the micro-level of firms (e.g., Audretsch/Dohse 2 0 0 7 ) . By referring to the resource-based view of firm performance, in which Kogut and Zander (1992) emphasize the critical role of knowledge (Rigby/Brown 2 0 1 3 ) , several mechanisms of agglomeration economies can be expected to have a positive impact on firm growth. Knowledge is exchanged and diffuses within an agglomeration among competing and cooperating firms, either without any direct interaction through constant mutual monitoring (Malmberg/Maskell 2 0 0 2 ) or as a result of direct interactions and learning processes in formal and particularly informal

Firm Growth and the Spatial Impact of Geolocated External Factors · 237

social networks (Singh 2005). Furthermore, the mobility of individuals (Breschi/Lissoni 2009) and the exchange of intermediate goods (Döring/Schnellenbach 2006) cause specialized knowledge embodied in h u m a n and physical capital to circulate. The more economic activities agglomerate a r o u n d a firm, the more readily available growth-relevant knowledge should be. Hence, we simply expect: Hypothesis 1 Agglomerations (of other economic activities) increase the firms' growth prospects. H o w e v e r , recent studies show that not all firms benefit in the same way f r o m co-location (Rigby/Brown 2 0 1 3 ; McCann/Folta 2 0 1 1 ; Knoben et al. 2010). Besides the kind of activities a firm carries out (see Beadury/Swann 2 0 0 9 or Duschl et al. 2 0 1 2 for the industryspecific impact of agglomerations), also its internal capabilities to capture different forms of externalities matter. From a resource-based view, it is often argued that firm size plays a critical role. Furthermore, the organisational structure of firms changes tremendously during their life-time and growth (Witt 2000). V a n O o r t et al. (2012) report that the relationship between firm performance and agglomeration is strongest for medium-sized firms. Neither small firms nor large firms seem to be fully able to internalize externally available resources or knowledge. T h e former tend to lack absorptive capacities (Cohen/Levinthal 1990), whereas the latter reduce their openness t o w a r d s the local environment, as they become organizationally complex and hence inflexible (VanOort et al. 2012). These results are supported by Rigby and Brown (2013), w h o find in their analysis of manufacturing plant performance in C a n a d a that small plants d o not benefit f r o m the local density of upstream suppliers. T h e issue of the moderating effect of firm size becomes more complex, as it additionally depends on the kind and composition of activities surrounding a firm. O n the one h a n d , N o o t e b o o m (2000) presumes for knowledge spillovers to take place that external knowledge should be neither t o o similar nor t o o different to the o w n knowledge base. If it is t o o similar, n o new contribution will be made, and if it is t o o different, the cognitive distance, which constitutes a barrier for effective communication, will cause problems in absorption and implementation (Eriksson 2011). The literature on related variety (e.g., Frenken et al. 2007) concludes that externalities like knowledge spillovers arise f r o m technologically related industries. O n the other h a n d , it is a well-known empirical regularity that firms diversify into new fields of activity with increased size (Bottazzi/Secchi 2006). Hence, the right degree of relatedness depends on firm size, as firms can realize potential benefits f r o m locating in agglomerations "only to the extent that they are capable of using knowledge f r o m other, co-located firms in combination with their own knowledge assets to create value" (Knoben et al. 2 0 1 1 : 8). Smaller firms, which are more specialized in their activities, are expected to benefit most f r o m a more narrowly defined relatedness. Activities that are t o o different d o not match with the internal knowledge base (Nooteb o o m 2000). As the size of the firms increases, their scope of activities broadens, and the meaning of relatedness expands. Knoben et al. (2011 ) show empirically that medium sized firms in general benefit f r o m being located in proximity to rather dissimilar, yet related firms. Other activities that for small firms are complementary to their o w n knowledge base tend to become increasingly internalized by larger firms. Ultimately, such activities, which cannot provide new pieces of knowledge, are reduced to mere sources of rivalry, leading to adverse effects of agglomeration.

238 · Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

Briefly stated, we investigate the theoretical interaction between firm size and the degree of relatedness of agglomerated economic activities regarding their impact on firm growth. Conditioning on firm size, we hypothesize: Hypothesis la With increasing firm size, firms benefit most from more and more broadly related external activities. Hypothesis lb For larger firms, the presence of closely related external activities may even hamper their growth prospects, as these activities become a source of rivalry instead of providing complementary knowledge. The impact of public research activities The explanation of why firms might benefit from industrial clusters has more recently shifted to phenomena like innovation, learning and knowledge spillovers (Malmberg et al. 2000). Growth-relevant knowledge is generated not only by economic activities of competing and cooperating firms, but also by explicit research activities in universities or public research institutes. In this case, the literature that focus on "regions rich in knowledge resources" (Audretsch/Dohse 2007) or "knowledge-intensive environments" (Raspe/vanOort 2011) is less ambiguous. Many studies show that public research in universities and research institutes support the innovativeness of nearby firms (e.g., Jaffe 1989; Anselin et al. 1997) and ultimately their growth performance (e.g., Audretsch/Lehmann 2005; Cassia et al. 2009 or Raspe/vanOort 2011). Scientific knowledge is transferred by various mechanisms, for example directly by public-private research collaborations, or indirectly by "scientific publications, seminars, workshops and informal relationships" (Fritsch/Slavtchev 2007). Usually, these studies do not condition on firm size. Nevertheless, two opposing arguments can be found in the literature: One based on the absorptive capacities of firms, the other based on the dependence of firms on external knowledge imputs. On the one hand, Brenner and Schlump (2013) argue that larger firms, which are also more likely to conduct internal R & D , receive more often and benefit more from knowledge generated by public research activities. Larger firms benefit more because of their higher absorptive capacities (Cohen/Levinthal 1990), or in the words of Laursen and Salter (2004): "The argument contained in previous research is that larger firms are more likely to have the capability to exploit external knowledge sources and to manage interactions with universities." On the other hand, small firms depend more on research conducted outside the firm: "We find substantial evidence that corporate R & D is a relatively more important source for generating innovations in large firms, while spillovers from university research laboratories are more important in producing innovative activity in small firms" (Acs et al. 1994). While large firms are able to conduct substantial research themselves and are able to interact with public research over large distances, small firms depend more on nearby public research (Torre 2008). It is unclear which of these two mechanisms dominates, so that we formulate two contradicting hypotheses: Hypothesis 2a Larger firms benefit from public research activities, whereas no growth effects exist for small ones. Hypothesis 2b Small firms benefit from nearby public research activities, whereas the proximity of public research activities has no effect on larger firms.

Firm Growth and the Spatial Impact of Geolocated External Factors · 239

The impact of higher educational

activities

Studies addressing the impact of universities usually consider the presence (Cassia et al. 2009; Raspe/van Oort 2011) or proximity to the next university (Audretsch/Lehmann 2005). Not much is known about the impact of the different mechanisms of universities on firm growth. Besides academic research, which in the previous section is subsumed under public research activities, universities' function is to perform education (Fritsch/Slavtchev 2007). Both functions differ substantially in their underlying mechanisms. For instance, qualified graduates bring up-to-date knowledge into firms (Brenner/ Schlump 2013). Besides, the mere availability of graduates is a prerequisite for firms that want to expand in their number employees, as graduates contribute to the pool of available and highly qualified workers. The faster firms grow the more workers they need to hire. This holds particularly true for larger firms, as the same growth rate implies a different absolute meaning. Hence, we expect: Hypothesis 3 The more the firms grow in quantitative terms (i.e., larger firms and high growth firms), the more relevant higher educational activities become.

2.2

The spatial dimension of external growth factors

The various sources of externalities discussed above can all be traced back to specific locations in space. Related economic activities take place within co-located firms, new scientific knowledge is published by scientists who work in public research institutes or universities, and from the latter students become graduated who might decide to work in one of the firms. These activities are unequally distributed across space, hence the locations of firms relational to these sources matter (Andersson/Karlsson 2007). Unfortunately, it is virtually impossible to measure the real individual impact of each single source of externality on the growth of each firm. For example, the assumed knowledge flows are intangible (Koo 2005), and data on the entire collaboration network or on the migration patterns of graduates is lacking. Yet the potential strength of the externalities can still be analysed by assessing their expected impact on the growth prospects of firms, to which they are accessible. Karlsson and Manduchi (2001) argue that the accessibility approach, based on early ideas of Weibull (1980), makes the general concept of geographical proximity operational in the first place. A high accessibility means a high potential for interaction, increasing the probability of knowledge spillovers, labour flows or other mechanisms alike. Therefore, we calculate the potential strength of externalities, which is set equal to the accessibility of the firms' locations, as their revealed impact on firm growth. The discussion (implicitly) assumes that agglomeration economies are bounded in space (Frenken et al. 2011). This holds especially true for knowledge spillovers. The inherent properties of the nature of knowledge, like the degree of tacitness or complexity (Sorensen et al. 2006), increase the cost of transmitting knowledge over longer distances. Transferring complex, that means often unstructured, but economically valuable knowledge, demands personal contacts. Because this kind of knowledge is mostly embedded in people, knowledge spillovers are a function of people's mobility and interactions (Andersson/Karlsson 2007). In lieu of recent improvements in ICT (Sonn/Storper 2008), there are still strong empirical findings that social interactions decrease with geographical distance (e.g., Hoekman et al. 2010 on the research collaboration between firms).

240 · Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

Despite the general acceptance of the distance-sensitivity of agglomeration economies, empirical studies lack a consensus on their spatial extent (Döring/Schnellenbach 2006). Distances as diverse as 10 km (Baldwin et al. 2008), 120 km (Anselin et al. 1997) or 300 km (Bottazzi/Peri 2003) are reported. Frenken et al. (2011) concludes that the findings depend on the type of mechanisms and external activities investigated as well as on the composition of the sample firms. Cainelli and Lupi (2008), for instance, show that the effect of similar activities is greatest at very small distances, whereas the effect of variety becomes positive only for larger distances between 10 and 30 kilometres. Considering institutional and geographical proximity simultaneously, Eriksson (2011) argues that greater distances between plants require that their technological knowledge base is more similar, so that differences in the firms' local institutional contexts, another cause of communication problems, can be overcome. Regarding public research activities, the spatial extent of externalities depends amongst others on the spatial distribution and effectiveness of formal and informal science-industry collaborations. As this kind of knowledge transfer often requires social interactions and face-to-face contacts, it might be a narrow local phenomenon. Regarding graduates, we are aware that they are among the most mobile group in society (Möhr 2002). Nonetheless, a study for Germany shows that 70% of them stay in their region of education ten years after graduation (Brenner/ Schlump 2013). The likelihood of outmigration decreases if growing firms are able to provide new job opportunities (Busch 2007). Because the literature does not provide a conclusive picture on the spatial extent of the various externalities, this issue has to be addressed empirically.

3

Empirical method

Distance-based methods, which take into account the bilateral distances between the locations of all activities, are suited to operationalize the effect of externalities as a matter of their accessibility (Andersson/Grasjö 2009). 1 A firm-specific location variable X for each time period t can be calculated by summing up all corresponding external activities χ (here, related economic, public research or higher educational activities), which are discounted by a distance decay function f(d¡m), where dtm indicates the distance between the location of firm i and the location of activity m: (1)

m

This approach requires that two choices are made: the specification of f(d¡m) and the way how distances are measured. Regarding the former, many approaches exist. Mostly, simple linear (e.g., Audretsch/Lehman 2005) or exponential decay functions (e.g., Drucker/Feser 2012) are used. The literature on commuting behaviour (e.g., Andersson/Karlsson 2007) shows that the negative distance sensitivity is not linear in space, but varies between different geographical scales: within a narrow local context, interactions are primarily governed by randomness, because they can take place at short 1

Distance-based methods are an alternative to methods which rely on regional boundaries. The latter try to explain firm performance by means of characteristics of the region the firm is located in. However, results are affected by the arbitrariness of regional boundaries and moreover by the chosen level of aggregation, also known as the Modifiable Areal Unit Problem (MAUP) (Openshaw 1984). By varying the spatial scale of analysis, Buerger et al. (2010) show empirically that the M A U P is highly relevant for agglomeration economies.

Firm Growth and the Spatial Impact of Geolocated External Factors · 241

notice (Thorsen et al. 1999). Thus, within agglomerations interactions are only marginally affected by distance. At some threshold distance, however, the minimal cost principle predominates and consequently, the frequency and contribution of growth-relevant economic interactions become highly distance-sensitive and may decrease rapidly. DeVries et al. (2009) model these spatial interactions by using a sigmoidal log-logistic decay function. Duschl et al. (2012) find that this rather flexible S-shaped curve often converges to a step-wise decay function, which is similar to distance bands, i.e. summing up the activities within specific radii around the firms (e.g., Rosenthal/Strange 2003). Hence, this paper employs such a binary distance function f(dim), which becomes one if the value is below the threshold distance d¡ m . As the literature does not reveal much about an adequate threshold distance (see section 2.2), d¡m is endogenized. In doing so, not only information regarding the magnitude of the impact of external factors is obtained, but also regarding the spatial extent. Furthermore, results may be sensitive to the way how distances are measured. Agglomeration economies are assumed to arise from low transportation costs or the convenience of face-to-face contacts. Thus, the firms' access to external factors not only depends on the location pattern of the corresponding activities, but also on the physical infrastructure (Andersson/Karlsson 2007). Whereas physical distance is still the frame, in which relevant interactions occur (Rodriguez-Pose 2011), it is driving distance or travel time that is directly related to the frequency of interactions (Andersson/Grasjö 2009). The vast majority of distance-based investigations uses orthodromic distances (e.g., km or miles). One of the few exceptions is the work of Audretsch and Lehmann (2005), where the growth of firms is analysed with respect to the firms' driving distance to the their closest university. Owing to high computational costs of route planning, studies in which driving distances are computed to thousands of locations are still very rare. The firm-specific location variables X ( t , which result from equation (1), can then be included in a simple linear model (for spatial econometric issues regarding this kind of spatially discounted variables we refer to Andersson/Grasjö 2009). 2 Because of the flexible specification of f(d¡m), a normalization procedure is first applied to make the corresponding regression coefficients comparable: X«,f = μΧ/,ί

(2)

μ =

(3)

where ΣΙΕΞΞΛ Σ/ Em f(dim)xm,t/N

with Ν denoting the number of firms. This normalization allows for an interpretation of the coefficients as the impact of an additional activity at a distance with an average impact on firm growth. The regression equation reads g u = « + E P i X u . i + E ßk u i , t , k + Ε Ρ ι τ ' · υ + ε ' · > / k ι

with α and β representing the coefficients to be estimated, U the various control variables, and Τ the dummies for the years t of observation. The error term is denoted by e¡ t . 2

Performing an extensive Monte Carlo analysis, these authors confirm that this approach captures substantive spatial dependence in the dependent variable and accounts for both local and global spillovers.

242 • Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

Furthermore, we assume that the variables which are related to public research and higher educational activities have only a positive impact: firms either benefit from these activities or they don't. Considering this, a negative coefficient simply would capture some structural effects, which are rather related to general characteristics of areas which host universities and public research institutes and less to the mechanisms of growth-relevant knowledge transfers. Therefore, the corresponding coefficients are submitted to non-negative constraints. 3 Furthermore, we assume that a driving distances of two hours represent a reasonable upper range even for larger metropolitan areas, so that d¡m is constrained at maximum 120 minutes. The model in (4) is estimated by using quantile regression techniques. Quantile regression seems to be the adequate technique on the grounds of the following reasons. Firstly, the stochastic analysis of section 4 . 1 reveals that firm growth rates are not normally distributed, but show fat tails. In contrast to OLS, quantile regression is free from any distributional assumption in the error term (Buchinsky 1998). Secondly, it is not sensitive to outliers on the dependent variable. This is especially relevant here due to the high frequency of extreme growth events, which also might stem from data problems. Thirdly, the specific conditional quantiles of strongly expanding (00.75) and declining (00.25) firms can be analysed in addition to the median growing firm ($0.5)· These firms significantly contribute to the overall economic development and hence are of interest in their own right (Coad/Rao 2 0 0 8 ) . Our intuition is that high growth events, a dominant feature of firm growth, rely differently on internal as well as external factors. Focusing exclusively on the average firm obscures these relationships (Coad/Rao 2008). Technical details are described amongst others in Koenker (2005). 4 Here we only point out that, likewise to OLS, the coefficient estimates can be interpreted as partial derivatives, meaning the impact of a one-unit change of an independent variable on the firms' growth rate at the 6th quantile holding all other variables fixed. T o recall, the binary distance functions are endogenized and the threshold distances dltn are identified by minimizing the log-likelihood value of the model. One should note that this optimization is done for each conditional quantile θ separately.

4

Data

The BvD-Amadeus database encompasses firms from both manufacturing and service industries. Regarding the firms' location, the addresses of the headquarters are disclosed. Operational and strategic decisions are often made within this organizational unit. Although R & D intensive foreign direct investments have recently become more important (UN C TAD 2 0 0 5 ) , many of the firms' R & D activities still remain located close to the headquarters (Guimón 2 0 0 9 ) . Even for multinational enterprises a home bias for innovation activities is evident (Cohen et al. 2 0 0 9 ; Dunning/Lundan 2 0 0 9 ) . Therefore, we follow Beaudry and Swann (2009) in assuming that the regional environment of the firms' headquarters is most decisive in affecting their growth prospects. This rationale, of course, breaks down for very large firms, which tend to be less focused on their headquarters, but disperse activities in many increasingly independent establish-

3

4

Indeed, the introduction of constraints was necessary, as for certain distances the coefficients turned out to be negative. Standard errors are estimated using bootstrapping techniques. In line with Koenker and Hallock ( 2 0 0 1 ) , we only detected negligible small discrepancies between various available methods.

Firm Growth and the Spatial Impact of Geolocated External Factors • 243 ments across the country and even beyond. Hence, the analysis is restricted to firms with no more than an annual average of 1 0 0 0 employees. 5 Also very small firms with less than ten employees, which growth processes are known to be rather erratic, are omitted (Coad 2 0 0 9 ) . Because in section 2.1 it is argued that the size of the firm determines its growth logic and the impact of externalities, the remaining firms are split into the three different size classes small [ 1 0 - 5 0 ) , medium [ 5 0 - 2 5 0 ) and large [ 2 5 0 - 1 0 0 0 ) according to the European Commission (2003). The composition of the subsamples is presented in Table 1.

Table 1 Sample composition and distributional properties of the growth rates Sample composition Small [10-50) Medium [50-250) Large [250-1000)

4.1

Ν firms

Ν

33062 25199 7423

78653 77147 26011

gi,t

AEP estimates ¿"left

bright

0.864 0.470 0.424

0.381 0.608 0.473

Dependent variable

Firm growth, which is a multi-dimensional process, can be addressed by a variety of measures like employees, turnover, sales or productivity. No universally best indicator exists and the pros and cons of the different measures are discussed in the literature at length (see Coad 2 0 0 9 for an overview). Raspe and van Oort (2008) argue that the employment measure is most adequate from the resource-based view of the firm, because employees represent a firm's most important asset. Besides, employment growth primarily should concern regional policy makers and thus can be regarded as a suitable measure to assess the impact of spatial externalities. Growth rates are calculated by taking the difference of the natural logarithms of the employees Empi of firm i between two successive years t: gij

=

log(Emplit+1)

-

\og(Empll

t

)

(5)

Confronted with an unbalanced panel from 2 0 0 4 to 2 0 1 0 , yearly growth rates are pooled together. Meanwhile, it counts as a stylized fact that the distribution of the growth rates shows an exponential tent-like shape, similar to the one of the Laplace distribution, implying that the tails are much fatter than the normal distribution would suggest. More recently, an asymmetric shape is found with extreme negative growth events being particularly predominant (e.g., Bottazzi et al. 2 0 1 1 ) . Fitting the flexible five-parameter Asymmetric Exponential Power (AEP) distribution (Bottazzi/Secchi 2 0 1 1 ) to our data, we find that the shape parameter b takes values clearly smaller than one, which would be the case of a Laplacian shape (see Table 1). For medium and large firms, especially negative fat tails are much more pronounced, as indicated by a smaller b¡eft. Only for small firms, the asymmetry is reversed. Firms of this subsample are more likely to realize large positive growth jumps, whereas strong decline events are less often observed,

5

W e follow C o a d ( 2 0 0 7 ) and use for all size thresholds the annual average t o avoid a regression fallacy originating f r o m the difficulty o f sorting growing entities into size classes ( F r i e d m a n 1 9 9 2 ) .

244 · Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

maybe due to the fact that exits are not considered in the analysis. Hidden by an adverse shock, smaller firms tend to be prone to disappear from the market altogether (Dunne et al. 1988). Exit as well as entry events are not comprehended as growth events. We are not able to include them in the analysis for two reasons: the growth rate as defined in Equation (5) is not defined for such events and our data only contains information on firms that exist at the end of the observation period. This implies that inference can only be made on the growth of surviving firms. This causes a bias in our data. The exit of a firm represents an extreme decline event in terms of firm growth. Hence, these extreme decline events are underrepresented in our sample. Exits of large firms are rare, but in the case of small firms we have to be aware of the fact that we analyse a biased sample. The most extreme decline events are missing. This impacts mainly the results for the quantile 00.25 and slightly the results for the quantile 6*0.5 in the case of small firms. The results for the quantile $o.75 should not be influenced. Hence, we interpret especially quantile $0.25 regressions for small firms with care. If these results do not fit the overall picture, we do not trust them. A stochastic analysis shows that standard OLS is expected to perform poorly, because errors will not be distributed normally (Maasoumi et al. 2007). Also the discussion on high-growth firms becomes apparent: most firms do not grow (or only slightly), whilst a small, but non-negligible part of firms experiences very rapid growth or decline. As argued in section 3, quantile regression techniques are more adequate for our purpose. 4.2

Independent variables

Firms' potential to benefit from externalities is specific to characteristics of the firms as well as of the corresponding regions (Beugelsdijk 2007). Therefore, the independent variables consist of three different kinds (see Table 2): First, we control for relevant demographic properties of the firms. Second, we include measures of the general environment of the region the firm is located in. Third, the focus of this paper lies on firm-specific location variables reflecting related economic, educational and scientific activities. Control variables: demographic and regional variables Building upon the literature on firm growth, which mostly extends a Gibrat-like growth regression (Coad 2009), we control for four demographic variables: the logarithm of employees (SIZE), years passed since founding date (AGE), a dummy indicating whether or not it is a subsidiary firm (d_SUBS), and a dummy assigning one to firms from knowledge intensive industries ( d _ K N O W ) . ' In the literature of industrial dynamics, it counts as a stylized fact that firm growth is negatively related to both size and age (Evans 1987). Firms that are formally identified as a subsidiary are by definition more or less dependent on their mother institution, with an unknown growth impact. The knowledge intensity dummy should proxy for internal research activities as well as the absorptive capacity, which are expected to increase with the knowledge intensity of the activities a firm is engaged with (Koo 2005). In addition to the firm-specific demographic variables, two variables are chosen to control for the general regional environment. Population density (POP) measures urbanization economies per se, which are rather independent from the surrounding industrial structure (Buerger et al. 2012). Its impact might be either positive or negative: on the one hand, 6

Gehrke et al. (2010) provide a classification for knowledge intensive industries at the 3-digit level.

Firm Growth and the Spatial Impact of Geolocated External Factors · 2 4 5

POP is known to foster innovation activities and ultimately growth (Feldman 2000), but on the other hand, negative agglomeration economies, mostly due to higher wages, could prevent firms of hiring employees. Additionally, Fritsch and Slavtchev (2011) argue that this variable catches up other types of unobserved region-specific influences. Unemployment rate (UR) reflects the vitality of the regions' socio-economic conditions. In the special case of Germany it also accounts for structural differences along the east-west and north-south divide. Data for both variables is obtained from the German Federal Statistical Office (destatis). POP is most meaningfully measured at district level, whereas UR at the level of functionally defined regional labour markets (Eckey et al. 2006). Moreover, also the absolute location within Germany might influence the potential magnitude of externalities. Cross-border effects cannot be considered, discriminating firms located close to the border. Owing to historical reasons, two dummies are constructed: one for the location in border regions with the New Member States of the EU ( d _ E U n e w ) and one for all other border regions (d_other). Finally, general macroeconomic conditions, foremost the global recession 2008-10, might systematically affect the firms' growth prospects. Therefore, year dummies are included. Firm-specific

location variables: related employment,

graduates and

publications

In contrast to the regional control variables, which account for a rather diffuse socioeconomic environment (or "social filter", as denominated by Rodriguez-Pose/Crescenzi 2008), other activities as sources of externalities can be traced back to concrete localizations in space: firms compete, cooperate, and learn from each other, new scientific knowledge is published at universities and research institutes, and students get hired after graduation at often nearby universities. The firms' accessibility to these activities and their impact on the growth prospects might be mediated by the general regional environment. For example, the regional environment influences the effectiveness of university-industry knowledge-related linkages (Varga/Parag 2009), and graduates tend to prefer a diverse and open urban atmosphere (Florida 2005). Therefore, the firm-specific location variables, which are distance-based and micro-founded measures of the sources of externalities, are complemented by regional level control variables, as presented above. Related economic activities can be approximated by the accessibility of other employees in the firms' industries. The issue of relatedness is tackled by a hierarchical approach: all employees which belong to the same 2-, 3- or 4-digit industry and are located within a certain distance are taken into account. In case of 3-digit industries, the number of employees of the same 4-digit level is excluded to avoid double counting. Analogously, employees of the 2-digit industry are adjusted by subtracting the same 3-digit employees. The Federal Employment Agency (BA) provides data of the location of industry-specific employment. Three firm-specific agglomeration variables result ( A G G L _ 2 , AGGL_3 and AGGL_4, respectively). Public research activities are reflected in the number of scientific publications. Although firms engage in publishing, the vast majority originates from universities and research institutes. Hence, publications tend to measure public research activities. Data on publications was collected from the ISI Web of Science and can be geolocated on basis of the authors' addresses. We anticipate that the fields of study most relevant are science, technology, engineering, and mathematics, commonly known as the STEM fields. By counting only those publishing activities in STEM that are accessible to the firms, a firm-specific publication variable can be calculated ( P U B L ) .

246 · Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

Higher educational activities can be measured by the number of graduates. Applying the same argument, graduates are restricted to the STEM fields. Data on graduates was taken from destatis and encompasses universities in a narrower sense as well as universities of applied science, that is, all graduates with a technical, diploma, bachelor and master degree. Again, by counting those higher educational activities that are accessible to the firms, a firm-specific graduate variable can be calculated {GRAD). The next section discusses how distances between the firms and related employees, graduates and publications are measured. Table 2 Overview of independent variables and data sources Variable type Variable name Description Demographic variables

SIZE ACE d SUBS d_KNOW

Regional variables

POP UR d_EUnew d_other

Firm-specific location variables

EMPL_2 EMPL_3 EMPL_4 PUBL GRAD

4.3

Data source

Logarithm of employees Years since founding date Subsidiary status (dummy) Sectoral affiliation to a knowledge intensive industry (dummy)

BvD BvD BvD BvD

Amadeus Amadeus Amadeus Amadeus

Population density of the firms' district Unemployment rate of the firms' labour market region (Eckey et al. 2006) Firms' district shares border with one of the new EU member states (dummy) Firms' district shares border with one of the other countries (dummy)

destatis destatis

Accessible employees of the same 2-digit industry Accessible employees of the same 3-digit industry Accessible employees of the same 4-digit industry Accessible publications in STEM Accessible graduates in STEM

BA BA BA Web of Science destatis

Driving distances

The calculation of driving distances is done by exploiting results from graph theory: the road network is modelled as a directed graph with travel time metric as edge weights. 7 Knopp et al. (2007) introduced an algorithm to compute large-scale distance matrices without naively computing a quadratic number of distances. The small search spaces of a speedup technique to Dijkstras seminal algorithm are precomputed and intersected to produce the matrix. This method only needs a linear number of shortest computations and therefore is several orders of magnitude faster than the naive algorithm. In addition, an algorithm of Geisberger et al. (2008) is used that exploits the natural hierarchy of road networks, called Contraction Hierarchies (CH). The method preprocesses a road network and produces a linear sized amount of auxiliary data that is used to speed up 7

Data on the German road network was taken from the OpenStreetMap project as of July 22nd, 2011 and consists of 8,226,112 nodes and 15,501,574 edges.

Firm Growth and the Spatial Impact of Geolocated External Factors · 247

any subsequent queries. CH have the benefit of small search space, i.e. a query has to look at only a few hundred nodes in the graph. Combined with the previous algorithm we can compute distance matrices of 10,000 by 10,000 nodes within a matter of several seconds. To investigate externalities on firm growth from a realistic spatial perspective, the aggregation level of the data should be as low as possible. For Germany, this is given by municipalities, currently with a number of 11249 and an average size of 31.6 km 2 , which Duranton and Overman (2005) would already classify as micro data. Thus, the locations of the firms and all other activities are approximated by using the geocentroids of the corresponding municipalities. In doing so, a new issue arises. If one firm is located in the same municipality as, for instance, a university, it would be inappropriate to set the distance to zero. As a substitute, the existence of a general intra-municipality friction can be assumed. To obtain its value, a random sample of 1000 pairs of firms' address locations is drawn, each belonging to the same municipality, and all bilateral distances are measured. The mean of all intra-municipal travel times is 5.01 minutes.

5 5.1

Results Control variables

The estimated coefficients of the control variables are largely in line with the current literature on firm growth and industrial dynamics (see Table 3). Yearly growth rates generally tend to correlate negatively with SIZE and AGE, even viewed within narrower size classes. The revealed relationships become even more pronounced for highly growing firms at the conditional quantile Ö0.75, implying that growth jumps are less likely the larger or older a firm becomes. SIZE is positively related to growth only at 6>o.25 for small firms, the case that might be influenced by the exclusion of exits. Due to the possible bias in this case we do not trust the result and ignore it. Regarding AGE, we observe an clear overall picture. However, younger firms have higher exit rates, so that the bias leads to more negative estimates than without such a bias. Hence, the negative relationship of AGE with firm growth at $0.25 for small firms might be a consequence of the data bias. This confirms the above statement that especially extreme growth events are negatively related to the size of firms, while we are not able to make a concluding statement for the extreme decline events. Two further demographic control variables, d_SUBS and d_KNOW, are included. Interestingly, the status as subsidiary is beneficial for small firms, but hampers the growth prospects of medium and large firms. This result suggests that the formal reliance on a mother institution might facilitate access to growth-relevant resources and at the same time might provide a shelter against adverse events, because also negative growth events at 00.25 become less likely. With increasing firm size, this dependence turns into a constraint, whereby especially high growth events at Ö0.75 a r e affected. The results for the affiliation to a knowledge intensive industry are quite straightforward. These firms possess greater growth prospects, in particular at higher conditional quantiles. The second group of control variables relate to the general socio-economic environment of the firms' region. POP tends to come along with lower average growth rates. This finding confirms other studies of Germany (e.g., Fornahl/Otto 2008) and suggests that cost aspects due to congestion in densely populated places dominate when agglomeration effects of related employees and of proximate publications and graduates are directly

248 · Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

Table 3 Regression results for demographic and regional control variables small [5, 50) »0.25 SIZE

θ

0.50

medium [50, 250) »0.75

0

»0.25

O.5O

large [250, 1000)

e

0.75

00.25

00.50

%75

. 0 2 7 0 " -.0760*** -.1112*" -.0181 *** - . 0 2 0 9 " * -.0454*** -.0286 * " - . 0 1 8 5 " " - . 0 4 7 5 * "

AGE

-.0001 ** - . 0 0 0 1 * " -.0007*** -.0001 * " - . 0 0 0 2 " * - . 0 0 0 5 " * .0097"*

.0023"*

.0183*** -.0003

d_KNOW

.0017'

.0015*"

.0181"*

Pop

-.0000"

-.0000

UR

.0007

-.0024'

--.0448***

-.0002

-.0018

d_SUBS

d_EUnew -.0016 d_other

-.0014

.0002

.0000

.0048**

.0000

- . 0 0 0 1 " * -.0003***

-.0031*** - . 0 0 7 0 * " -.0005

-.0083"' -.0115"*

.0044 "*

.0051 ***

.0089*"

-.0000 '** - . 0 0 0 0 " * -.0000 .0020

.0025*

.0037*"

-.0000 ** -.0000*

.0048** .0000

-.0090*

-.0275"

-.0098

-.0065

-.0313*

.0010

.0027

-.0012

.0028'

.0049'

.0015"

.0020

-.0004

.0012

.0008

-.0004 .0001

p-values: ' < 0.1, * < 0.05, ** < 0.01, *** < 0.001

taken into account. Put differently, higher wages prevent firms of hiring employees. During phases of high growth, for which no significant relationships are found, the price competition seems to become less relevant. As expected, UR shows a slightly negative correlation with firm growth. Foremost high growth events are hampered in structurally less-favoured regional economies. Finally, the two border-region dummies, which control for a potential underestimation of agglomeration economies across national boundaries, are merely significant, but if so, they show up to be positively correlated with firm growth. Not surprisingly, growth rates were significantly reduced during the years of macroeconomic recession, particularly 2009 (estimates for the yearly dummies are not reported). 5.2

Impact of firm-specific location variables

Having controlled for several demographic and region-specific variables, the impact of the firm-specific location variables can be discussed (see Table 4). Related employees In general, the agglomeration of related employees increases the firms' growth prospects. Estimates tend to be positive and in the majority of cases significant. This supports hypothesis 1. Furthermore, the estimates are larger and more often significant at öo.2S and #o.75> confirming previous literature insofar as agglomeration effects are more relevant for fast growing firms (e.g., Fornahl/Otto 2008). Nonetheless, differences among the size groups and the degree of relatedness exist. Small firms benefit most from being located in proximity to employees of the same 4-digit industries, medium sized firms to employees of the same 3-digit industries, and finally large firms to employees of the same 2-digit industries. This clearly underlines the argument

Firm Growth and the Spatial Impact of Geolocated External Factors • 2 4 9

that the relevant degree of relatedness is conditional on firm size; hypothesis la can be confirmed - the larger the firms, the more diverse the related activities they benefit from. However, in contrast to the expectation that small firms lack the necessary absorptive capacity to benefit from activities that are rather broadly related, still a significantly positive sign is observed for the 2-digit industries. As the effect becomes more pronounced for $0.25 > o n e could conclude that the main role of these rather diverse activities is to reduce the risk of adverse negative growth shocks (by the portfolio effect of diversity); high growth events are not fostered neither for small nor medium sized firms. As predicted in hypothesis 1 b, large firms are hampered by activities of the same 4-digit industries. An agglomeration of very similar activities cannot provide complementary knowledge, but rather tends to become a source of rivalry. This does not hold at #o.25> for which a significant positive impact is found. An explanation can be provided by taking into account the spatial dimension of the externalities. T o recall, distances are endogenized in this paper. Table 5 contains the optimal distances of the corresponding estimates, printed in bold if these are significant. In 13 out 21 cases, in which the impact of related employees is significantly positive, the optimal distance threshold surpasses 30 minutes, and in eleven cases even 90 minutes. This finding reveals that the impact of externalities from related employees tend to have a larger extent, which sometimes even transcend traditional regional boundaries, and which could not been captured by methods purely relying on regions. Interestingly, employees of the same 4digit industry are most beneficial at larger distances. Referring to Erikson (2011), the negative effects of technologically too narrowly related activities can be traded off by variety in the local institutional knowledge, manifested in sharing the same heuristics and routines, which increases by geographical distance. In the two cases of large firms, in which a negative effect at the 4-digit level is found, the corresponding distances are 13 and 14 minutes, confirming the same argument: if both the cognitive and geographic distance is too close, the negative effect dominates. In contrast hereto, large firms, at %25> are able to benefit from activities of the same 4-digit industry which are located within 51 minutes. These shrinking firms, for which the opportunity reducing effect of rivalry seems to become less relevant compared to expanding firms, are able to successfully trade off the technological overlap by other forms of variety that increase with geographical distance. Publications

and

graduates

Regarding the variables PUBL and GRAD, the issue of multicollinearity has to be addressed first. Graduates originate from universities as does a major share of publications; the larger the universities, the higher the output in both measures. This makes the variables strongly correlated. Although the optimal distances d l m for both mechanisms have been determined endogenously, it is not possible to guarantee that their growth effect can be separated entirely. Therefore, besides the simultaneous model with both variables included we estimate additionally the same model but with either PUBL or GRAD excluded and report the significance level in brackets (Table 4). The growth rates of small firms are not significantly correlated with PUBL or GRAD. Although it is not possible to conclude that such relationships are not given, this result tends to confirm the expectation that small firms, because of the lack of absorptive capacities, are not or only marginally able to benefit from public research and higher educational activities. The impact of PUBL becomes significant for medium sized firms at 6Q 5 and

250 • Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

for large firms at 0q.25 and 0Q 5 in the joint model (together with GRAD), and significant for all conditional quantiles in the individual model. N o patterns for differences along the quantiles become apparent. Put differently, after a certain size threshold public research activities are beneficial for all kinds of growth levels. Hence, Hypothesis 2a is confirmed, while Hypothesis 2b is rejected. The medium-sized and large firms are those that benefit from nearby public research activities. Similar findings can be reported for GRAD, with the exception that the estimates clearly increase at higher quantiles (it becomes even slightly significant for small firms at #0.75 )> and that for declining firms at θο.25 they are not significantly different from zero. This confirms hypothesis 3, which states that especially larger firms and high growth firms benefit from higher educational activities. Firms require an adequate pool of available and qualified workers to expand, which in absolute terms holds even more so for large, fastgrowing firms. In contrast to PUBL, which helps to decrease the likelihood of negative growth events, no such effect is observed for GRAD, which supports the idea that this variable measures the contribution of universities to the local labour market. Comparing the effects of GRAD and PUBL in the joint model, it turns out that PUBL can keep its significance in three cases, and GRAD only in one case. Higher educational activities, which seem to directly increase the firms' growth opportunities by their contribution to the local pool of qualified graduates, nonetheless cannot compete with public research activities in explaining firm growth. New scientific knowledge from public research activities, which first has to be transformed by the firms into economic opportunities, seems to be at least equally relevant. Finally, the estimated distances provide some information on the spatial dimension of the growth impact of public research and higher educational activities. In contrast to related employees, considerable smaller distances of a few minutes result for both GRAD and PUBL. Here, we expect that small firms cooperate only with local universities or public research institutes due to travel costs, and that larger firms either cooperate with local ones or with the best global alternatives. If public research activities are performed nearby, firms benefit most as short distances facilitate social interactions and face-toface contacts. Also for graduates, small distances predominate. To conclude, externalities that originate from public research and higher educational activities occur to a large part at a geographical scale much smaller than usually assumed as "regional". Hence, distance-based and micro-founded methods help to avoid underestimating their effect. Agglomeration effects from related economic activities do not take halt at predefined regional boundaries, which in many studies coincide with administrative territories.

6

Conclusions

Frenken et al. (2011) have suggested for future research that one of the main challenges "lies in settling contradictory empirical findings. In particular [...] the main gap in our empirical understanding concerns the effect of localization economies on firm performance, which some may even consider the key question in economic geography at large". In line with these authors, this paper argues that many contradictory empirical findings are closely related to the heterogeneity of firms, the heterogeneity of the sources of externalities, and the heterogeneity of spatial economic landscapes, which are a product of the location of both the firms and the sources.

Firm Growth and the Spatial Impact of Geolocated External Factors • 251

Table 4 Regression results for firm-specific location variables small [5, 5 0 ) 00.25

medium [ 5 0 , 2 5 0 ) θ

«0.75

e0.J0

EMPL_2

.0004* •

.0001 *

EMPL_3

.0001'

.0001*

EMPL_4

. 0 0 0 5 * •*

PUBL GRAD

-.0002

.0004*·'

.0007··

.0004*"

.0002*

.0001'

.0010··

.0003***

.0001

.0004

.0001

.0003

.00020

,0006*(**· )

.0003

.0000

.0012Ό

.0001

.0003(**)

< 0.01, · * *

«0.25

0.75

0.50

.0002"

p-values: ' < 0 . 1 , * < 0 . 0 5 , · ·

large [ 2 5 0 , θ

θ

0.25

-.0003

1000) e

β0.50

0.7S

.0013* ·*

.0009***

.0016"

.0009*"

.0004* *

.0002*

.0003*

.0002*

. 0 0 0 5 * **

.00050 .00060

-.0001*

-.0006*·'

.0014* *

,0010*(*** )

.00170)

.0000

.0008·(·" )

.00120)

< 0.001

Table 5 Regression results for firm- and region-specific control variables small [5, 5 0 ) »0.25

EMPL_2 EMPLJ EMPL_4 PUBL G RAD

θ

0.50

medium [ 5 0 , 2 5 0 )

large [ 2 5 0 ,

1000)

«0.75

»0.25

«0.50

'0.75

«0.25

«0.50

%7J

105

94

18

8

119

24

119

120

120

12

26

10

99

16

25

16

15

14

120

120

120

116

32

32

51

13

14

9

8

11

9

17

12

5

16

16

119

6

24

21

6

6

49

9

15

N u m b e r s in bold if p-value o f corresponding variable is significant

This paper takes the call for a finer resolution seriously. The effect of different sources of externalities on firm growth is compared for small, medium and large firms. Quantile regression techniques additionally shed light on the relationships for highly growing and declining firms. This paper finds that the impact of related economic activities, measured by related employees, depends on both the size of the firm and the degree of relatedness: the larger the firm, the more diverse nearby activities should be. T o put differently, specialised agglomerations rather hamper the growth of large firms, but might stimulate the growth of small firms. For public research and higher educational activities we observe a clear size threshold: only medium and large firms are able to benefit from nearby publications and graduates in the S T E M fields. T h e geographical meaning of " n e a r b y " differ for the analysed sources. By endogenizing distances, it turns out that some externalities, mostly related economic activities, transcend traditional regional boundaries, whereas externalities from public research and higher educational activities are very local phenomena of few driving minutes. This implies that the locations of universities and research institutions are crucial for firms and, hence,

252 · Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

regional economic development. It is important that their location matches the quite local economic activity in order to realise to potential economic benefit. The above findings highlight the advantages of using a distance-based, data-driven approach, which does not rely on predefined regional delimitations. The spatial reach of various relationships between variables differs. More studies in this direction are necessary to obtain a better understanding of the spatial aspect in economic dependencies. Finally, it becomes clear that not only firm internal factors can drive the growth of firms, nor is it exclusively stimulated by firm external factors. Instead, it is the complex interplay between internal factors and external factors, an interplay that depends on the kind (i.e., related employment, graduates and publications) and spatial extent of externalities. Like VanOort et al. (2012) or Rigby and Brown (2013), this paper focuses on differences along firm size. However, the heterogeneity of firms - one of the few invariables of industrial dynamics (Dosi et al. 2010) - also concerns other aspects than size. For instance, firms' age (e.g., Neffke et al. 2012) or industry affiliation (e.g., Duschl et al. 2011, Beaudry/Swann 2009) mediate the growth impact of externalities. Hence, we would motivate for a stronger integration of the various dimensions of heterogeneity in a comprehensive empirical framework. Furthermore, one could explore more systematically the reasons behind the firm-specific differences in the effects of externalities, for example by disentangling the mechanisms in comparative case studies.

References Acs, Z.J., D.B. Audretsch, M.P. Feldman (1994), R&D spillovers and recipient firm size. The Review of Economics and Statistics: 336-340. Andersson, M., U. Grasjö (2009), Spatial dependence and the representation of space in empirical models. The Annals of Regional Science 43: 159-180. Andersson, M., C. Karlsson (2007), Knowledge in Regional Economic Growth - The Role of Knowledge Accessibility. Industry and Innovation 14: 129-149. Anselin, L., A. Varga, Ζ. Acs (1997), Local Geographic Spillovers between University Research and High Technology Innovations. Journal of Urban Economics 42: 422^448. Asheim, B., P. Cooke, R. Martin (2006), The rise of the cluster concept in regional analysis and policy: a critical assessment. Pp. 1-29 in: B. Asheim, P. Cooke, R. Martin (eds.). Clusters and regional development: critical reflections and explorations. London and New York: Routledge. Asheim, B., M.S. Gertler (2006), The Geography of Innovation: Regional Innovation Systems. In J. Fagerberg, D. C. Mowery, R. R. Nelson (eds.): The Oxford Handbook of Innovation, Oxford: Oxford University Press, 291-317. Audretsch, D., D. Dohse (2007), Location: A Neglected Determinant of Firm Growth. Review of World Economics 143: 79-107. Audretsch, D., E. Lehmann (2005), Mansfield's Missing Link: The Impact of Knowledge Spillovers on Firm Growth. Journal of Technology Transfer 30: 207-210. Baldwin, J., D. Beckstead, W.M. Brown, D. Rigby (2008), Agglomeration and the geography of localization economies in Canada. Regional Studies 42: 117-132. Beaudry, C., P. Swann (2009), Firm growth in industrial clusters of the United Kingdom. Small Business Economics 32: 409—424. Beaudry, C., A. Schiffauerova (2009), Who's right, Marshall or Jacobs? The localization versus urbanization debate. Research Policy 38: 318-337. Beugelsdijk, S. (2007), The Regional Environment and a Firm's Innovative Performance: A Plea for a Multilevel Interactionist Approach. Economic Geography 83: 181-199. Bottazzi, G., A. Coad, N. Jacoby, A. Secchi (2011), Corporate growth and industrial dynamics: Evidence from French manufacturing. Applied Economics 43: 103-116.

Firm Growth and the Spatial Impact of Geolocated External Factors · 253

Bottazzi, G., A. Secchi (2006), Gibrat's law and diversification. Industrial and C o r p o r a t e Change 15: 8 4 7 - 8 7 5 . Bottazzi, G., A. Secchi (2011), A new class of asymmetric exponential power densities with application to economics and finance. Industrial and Corporate Change 20: 9 9 1 - 1 0 3 0 . Bottazzi, L., G. Peri (2003), Innovation and spillovers in regions: Evidence f r o m European patent data. European Economic Review 47: 6 8 7 - 7 1 0 . Brenner, T., C. Schlump (2013), Universities, Public Research and Regional Innovation O u t p u t : An Empirical Study of 19 Technologies in Germany. M a r b u r g W o r k i n g Paper on Innovation and Space 03.13. Breschi, S., F. Lissoni (2009), Mobility of skilled workers and co-invention networks: an a n a t o m y of localized knowledge flows. Journal of Economic Geography 9: 4 3 9 - 4 6 8 . Buchinsky, M . (1998), Recent Advances in Quantile Regression Models. T h e Journal of H u m a n Resources 33: 8 8 - 1 2 6 . Buerger, M . , T. Broekel, A. C o a d (2012), Regional Dynamics of Innovation: Investigating the Co-Evolution of Patents, Research and Development ( R & D ) , and Employment. Regional Studies 46: 5 6 5 - 5 8 2 . Buerger, M . , F. v a n O o r t , B. van der K n a a p (2010), A Treatise on the Geographical Scale of Agglomeration Externalities and the MAUP. Scienze Regionali 9: 1 9 - 4 0 . Busch, O . (2007), W h e n have all the graduates gone? Internal cross-state migration of graduates in Germany 1984—2004. SOEPpapers on Multidisciplinary Panel Data Research 26: 1 - 1 7 . Cassia, L., A. Colombelli, S. Paleari (2009), Firms' growth: Does innovation system matter? Structural Change and Economic Dynamics 20: 2 1 1 - 2 2 0 . Cainelli, G., C. Lupi (2008), Does Spatial Proximity Matter? Micro-evidence f r o m Italy. Economics & Statistics Disucssion Paper 42/08. Coad, A. (2007), A Closer Look at Serial G r o w t h Rate Correlation. Review of Industrial Organization 31: 6 9 - 8 2 . Coad, A. (2009), The G r o w t h of Firms. Cheltenham. Coad, Α., R. R a o (2008), Innovation and firm growth in high-tech sectors: A quantile regression approach. Research Policy 37: 6 3 3 - 6 4 8 . Cohen, W., D. Levinthal (1990), Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly 35: 1 2 8 - 1 5 2 . Cohen, S., A. Di Minin, Y. M o t o y a m a , C. Palmberg (2009), The Persistence of H o m e Bias for I m p o r t a n t R & D in Wireless Telecom and Automobiles. Review of Policy Research 26: 55-79. Döring, T., J. Schnellenbach (2006), W h a t D o W e K n o w a b o u t Geographical Knowledge Spillovers and Regional Growth? A Survey of the Literature. Regional Studies 40: 3 7 5 - 3 9 5 . Dosi, G., S. Lechevalier, A. Seechi (2010), Introduction: Interfirm heterogeneity - nature, sources and consequences for industrial dynamics. Industrial and C o r p o r a t e Change 19: 1 8 6 7 - 1 8 9 0 . Drucker J., E. Feser (2012), Regional industrial structure and agglomeration economies: An analysis of productivity in three manufacturing industries. Regional Science and U r b a n Economics 42: 1 - 1 4 . Dunne, T., M . Roberts, L. Samuelson (1988), Patterns of Firm Entry and Exit in U.S. M a n u facturing Industries. The R A N D Journal of Economics 19: 4 9 5 - 5 1 5 . Dunning, J., S. Lundan (2009), The Internationalization of Corporate R & D : A Review of the Evidence and Some Policy Implications for H o m e Countries. Review of Policy Research 26: 13-33. D u r a n t o n , G., H . O v e r m a n (2005), Testing for Localization Using Micro-Geographic Data. Review of Economic Studies 72: 1 0 7 7 - 1 1 0 6 . Duschl, M . , T. Scholl, T. Brenner, D. Luxen, F. Raschke (2012), Industry-specific firm growth and agglomeration. M a r b u r g Working Paper on Innovation and Space 06.12. Eckey, H., R. Kösfeld, M . Tiirck (2006), Abgrenzung deutscher Arbeitsmarktregionen. R a u m forschung und R a u m o r d n u n g 64: 2 9 9 - 3 0 9 . Eriksson R. (2011), Localized Spillovers and Knowledge Flows: H o w Does Proximity Influence the Performance of Plants? Economic Geography 87: 1 2 7 - 1 5 2 .

254 • Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

European Commission (2003), Commission Recommendation of 6 May 2003 concerning the definition of micro, small and medium-sized enterprises. L 125/36. Evans, D. (1987), The relationship between firm growth, size, and age: estimates for 100 manufacturing industries. The Journal of Industrial Economics 35: 567-581. Feldman M . (2000), Location and innovation: the new economic geography of innovation, spillovers, and agglomeration. Pp. 3 7 3 - 3 9 4 in: G. Clark, M. Feldman, M . Gertler (eds.), Oxford Handbook of Economic Geography. Oxford. Florida, R. (2005), Cities and the Creative Class. New York. Fornahl, D., A. Otto (2008), Between snails and gazelles? Analysing the effects of regional specialisation on firm growth: a quantile regression approach. Druid Conference Paper 2008. Frenken, K., F. vanOort, T. Verbürg (2007), Related Variety, Unrelated Variety and Regional Economic Growth. Regional Studies 41: 685-697. Frenken, K., E. Cefis, E. Stam (2011), Industrial dynamics and economic geography: a survey. Working Paper No. 11.07, Eindhoven Centre for Innovation Studies (ECIS). Friedman, M . (1992), Do Old Fallacies Ever Die? Journal of Economic Literature 30: 2129-2132. Fritsch, M., V. Slavtchev (2007), Universities and Innovation in Space. Industry and Innovation 14: 2 0 1 - 2 1 8 . Fritsch, M., V. Slavtchev (2011 ), Determinants of the Efficiency of Regional Innovation Systems. Regional Studies 45: 905-918. Gehrke, B., C. Rammer, R. Frietsch, P. Neuhäusler (2010), Listen wissens- und technologieintensiver Güter und Wirtschaftszweige. Studien zum deutschen Innovationssystem 19-2010. Geisberger, R., P. Sanders, D. Schultes, D. Delling (2008), Contraction Hierarchies: Faster and Simpler Hierarchical Routing in Road Networks. At: 7th Workshop on Experimental Algorithms, May/June 2008. Guimón, J. (2009), Government strategies to attract R&D-intensive FDI. Journal of Technological Transfer 34: 364-379. Hoekman, J., K. Frenken, R. Tijssen (2010), Research collaboration at a distance: Changing spatial patterns of scientific collaboration within Europe. Research Policy 39: 662-673. Jaffe, A. (1989), Real Effects of Academic Research. The American Economic Review 79: 957-970. Karlsson, C., A. Manduchi (2001), Knowledge spillovers in a spatial context - a critical review and assessment. Pp. 101-123 in: M . Fischer, J. Fröhlich (Hrsg.), Knowledge, Complexity and Innovation Systems. Berlin. Knoben, J., O. Raspe, A. Arikan, F. Oort (2011), Agglomerations and Firm Performance: one firm's medicine is another firm's poison. DRUID Winter Conference 2011, Copenhagen. Knopp, S., P. Sanders, D. Schultes, F. Schulz, D. Wagner (2007), Computing Many-to-Many Shortest Paths Using Highway Hierarchies. At: Workshop on Algorithm Engineering and Experiments, January 2007. Koenker, R. (2005), Quantile Regression. Cambridge University Press: N e w York. Koenker, R., K. Hallock (2001), Quantile Regression. Journal of Economic Perspectives 15: 143-156. Kogut, B., U. Zander (1992), Knowledge of the firm, combinative capabilities, and the replication of technology. Organization Science 3: 383-397. Koo, J. (2005), Agglomeration and spillovers in a simultaneous framework. The Annals of Regional Science 39: 3 5 - 4 7 . Laursen, K., A. Salter (2004), Searching high and low: what types of firms use universities as a source of innovation? Research Policy 33: 1201-1205. Maasoumi, E., J. Racine, T. Stengos (2007), Growth and convergence: A profile of distribution dynamics and mobility. Journal of Econometrics 136: 483-508. Malmberg, Α., Β. Malmberg, P. Lundquist (2000), Agglomeration and firm performance: economies of scale, localisation, and urbanisation among Swedish export firms. Environment and Planning A 32: 305-321. McCann, Β., T. Folta (2011), Performance differentials within geographic clusters. Journal of Business Venturing 26: 104-123.

Firm Growth and the Spatial Impact of Geolocated External Factors • 255

Möhr, H . (2002), Räumliche Mobilität von Hochschulabsolventen. Pp. 2 4 9 - 2 8 1 in: L. Bellmann, J. Veiling (eds.), Arbeitsmärkte für Hochqualifizierte. Nürnberg. Neffke, F., M. Henning, R. Boschma (2012), The impact of aging and technological relatedness on agglomeration externalities: a survival analysis. Journal of Economic Geography 12: 485-517. Nelson, R., S. Winter (1974), Neoclassical vs. evolutionary theories of economic growth: Critique and prospectus. Economic Journal 84: 886-905. Nooteboom, B. (2000), Learning by Interaction: Absorptive Capacity, Cognitive Distance and Governance. Journal of Management and Governance 4: 69-92. VanOort, F., M. Burger, J. Knoben, O. Raspe (2012), Multilevel approaches and the firmagglomeration ambiguity in economic growth studies. Journal of Economic Surveys 26: 468-491. Openshaw, S. (1983), The modifiable areal unit problem. Norwich. Penrose, E. (1959), The Theory of the Growth of the Firm. Oxford. Ponds, R., F. van Oort, K. Frenken (2010), Innovation, spillovers, and university-industry collaboration: An extended knowledge production function approach. Journal of Economic Geography 10: 231-255. Raspe, O., F. van Oort (2008), Firm Growth and Localized Knowledge Externalities. The Journal of Regional Analysis & Policy 38: 100-116. Raspe, O., F. van Oort (2011), Growth of new firms and spatially bounded knowledge externalities. The Annals of Regional Science 3: 4 9 5 - 5 1 8 . Renski, H. (2011), External economies of localization, urbanization and industrial diversity and new firm survival. Papers in Regional Science 90: 473-502. Rigby, D., M. Brown (2013), W h o Benefits from Agglomeration? Regional Studies DOLIO. 1080/00343404.2012.753141: 1 - 1 6 . Rigby, D., J. Essletzbichler (2002), Agglomeration economies and productivity differences in US cities. Journal of Economic Geography 2: 407-432. Rodriguez-Pose, A. (2011), Economists as geographers and geographers as something else: on the changing conception of distance in geography and economics. Journal of Economic Geography 11: 347-356. Rodriguez-Pose, Α., R. Crescenzi (2008), Research and Development, Spillovers, Innovation Systems, and the Genesis of Regional Growth in Europe. Regional Studies 42: 5 1 - 6 7 . Rosenthal, S., W. Strange (2003), Geography, industrial organization, and agglomeration. The Review of Economics and Statistics 85: 377-393. Sherman, H., W. Lamb, K. Aspregen (2009), The care and feeding of high-growth businesses in rural areas: the role of universities. Pp. 350-366 in: A. Varga (ed.), Universities, Knowledge Transfer and Regional Development. Cheltenham. Singh, J. (2005), Collaborative networks as determinants of knowledge diffusion patterns. Management Science 51: 756-770. Sonn, J., M. Storper (2008), The increasing importance of geographical proximity in technological innovation: an analysis of US patent citations, 1975-1997. Environment and Planning A 40: 1020-1039. Sorensen, O., P. Audia (2000), The Social Structure of Entrepreneurial Activity: Geographic Concentration of Footwear Production in the United States, 1940-1989. The American Journal of Sociology 106: 424-262. Sorenson, O., J. Rivkin, L. Fleming (2006), Complexity, networks and knowledge flow. Research Policy 25: 994-1017. Thorsen, I., J. Uboe, G. Navdal (1999), A Network Approach to Commuting. Journal of Regional Science 39: 73-101. Tobler, W. (1970), A computer movie simulating urban growth in the Detroit region. Economic Geography 46: 234-240. Torre, A. (2008), On the role played by temporary geographical proximity in knowledge transmission. Regional Studies 42: 869-889. UNCTAD (2005), World investment report 2005. Geneva: United Nations Commission for Trade and Development.

256 • Matthias Duschl, Antje Schimke, Thomas Brenner, and Dennis Luxen

Varga, Α., A. Parag (2009), Academic knowledge transfers and the structure of internal research networks. Working paper IAREG 1.3d. DeVries, J., P. Nijkamp, P. Rietveld (2009), Exponential or Power Distance - decay for Commuting? An Alternative Specification, Environment and Planning A 41: 461-480. Weibull, J. (1980), On the numerical measurement of accessibility. Environment and Planning A 12: 53-67. Witt, U. (2000), Changing cognitive frames - changing organizational forms: an entrepreneurial theory of organizational development. Industrial and Corporate Change 9: 733-755. Witt, U. (2008), Recent developments in evolutionary economics. Cheltenham: Edward Elgar. Witt, U., T. Broekel, T. Brenner (2012), Knowledge and its economic characteristics: a conceptual clarification. Pp. 3 6 9 - 3 8 2 in: R. Arena, A. Festré, Ν . Lazaric (eds.), Handbook of Economics and Knowledge. Cheltenham: Edward Elgar Publishing. Matthias Duschl, Philipps University of Marburg, Department of Geography, Deutschhausstr. 10, 35032 Marburg, Germany. [email protected] Dr. Antje Schimke, Kaiserallee 15, 76185 Karlsruhe. [email protected] Prof. Dr. Dr. Thomas Brenner, Philipps University of Marburg, Department of Geography, Deutschhausstr. 10, 35032 Marburg, Germany. [email protected] Dr. Dennis Luxen, Karlsruhe Institute of Technology, Department of Informatics, Institute of Theoretical Informatics, Am Fasanengarten 5, 76131 Karlsruhe, Germany. [email protected]

Jahrbücherf. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

Competition and Increasing Returns Mario Amendola University of Roma La Sapienza Jean-Luc Gaffard OFCE, SKEMA Business School, Sophia Antipolis Cedex JEL L11; 031 Competition; increasing returns; time to build; viability.

Summary The paper demonstrates the compatibility between competition as a rivalry among firms and increasing returns resulting from innovative choice. The analysis offers the prospect of a general theory of economic evolution. It is carried out by means of a model, which makes it possible to exhibit the time structure of production processes and to sketch out the sequential interaction of decisions in a process of restructuring of productive capacities for the whole economy. It shows that several firms can coexist in the market, despite the existence of increasing returns, yet remain differentiated not so much because they supply differentiated goods, but because they are each one at a different stage of the life cycle of the production process.

1

Introduction

The standard way of looking at the problem of increasing returns from the viewpoint of theory is to ask the question: is a given market structure (namely, perfect competition) compatible with increasing returns? The answer is generally no. According to the standard theory increasing returns, whatever their empirical relevance, are not analytically compatible with full competition. Monopoly, with a falling demand checking decreasing costs, is the only market structure allowing for increasing returns (Sraffa 1926). External economies seem to be the only way to reconcile increasing returns and competitive conditions. Internal economies of scale should induce concentration. But this will be ineluctable only in the absence of innovation, in a context marked by unchanged technologies and preferences, or more exactly when innovation comes to an end. In fact, competition is the very force that leads suppliers and merchants to search and exploit profit opportunities by introducing innovations and changing their environment. We have thus to look at the problem by asking the different, in our view more relevant, question: is innovation (which generates returns) compatible with competition? This, by the way, is in line with Adam Smith (1776) who stressed increasing returns as the main engine of growth in a market (competitive) economy, both at the firm and economy level. The focus must therefore be shifted to the relation between innovation and competition: namely to how, in this light, prices, costs, demand and investment are actually determined in a market economy.

258 · Mario Amendola and Jean-Luc Gaffard

We believe (with Baumol 2002) that competition takes place through innovation rather than through prices. But prices are still relevant as one of the elements in the co-ordination of innovation seen as a process. In this perspective, competition as a co-ordination process is really successful when prices and quantity adjustments are carried out, thus making it possible for the firms to obtain normal profits. That is when, by means of out-ofequilibrium adaptive processes, the industry converges towards a sort of dynamic equilibrium that is supposed to be 'consistent with the coexistence of a number of competing firms, all of them supplying, in conditions of increasing returns, products for the same general market' (Richardson 1998: 172). This is, by the way, in line with Alfred Marshall (1920), who intended to develop an evolutionary analysis of firm and industry, and "saw the 'ordinary business of life' as the interaction in real time of innumerable adaptive processes" (Leijonhufvud 2006). In order to investigate the role of prices in the innovation process (prices that determine shifts in market shares in that they reflect changes in costs brought about by the undergoing innovation process) we need to consider a world of heterogeneous firms (where the heterogeneity is implied by the different position in time over the production process) which exhibit different price trends that sequentially determine changing market situations. As in standard theory, price changes reflect cost changes. But while in standard theory costs are the expressions of given production functions, our costs are characterised 'in time', in that they are the counterparts of the way in which the innovation process takes place (which depends in turn on how co-ordination is realised). On the other hand, only if we assume a steady state are the demand and cost conditions actually independent and exogenously determined by preferences and technology, respectively. Out of equilibrium, as we are during an innovation process, the average cost will depend, at each moment of time, on the successive investment spending (on the temporal structure of productive capacity) and on the sequence of final demand, which are narrowly interrelated. We say (with Hicks 1989) that when each investment by an incumbent belongs to a bundle of complementary investments over time, it would not be rational for the firm to drop it because technological or market information is incomplete. A sort of path dependence phenomenon prevails, which determines the investment behaviour of the firm. Whatever the risk of having to face opportunistic behaviour from other firms (suppliers or customers), incumbents will choose to invest in specific assets. Only the appearance of resource constraints can involve a change in the trajectory followed. New entrants do not have the same temporal constraints so that changes in the environment (i.e. the financial environment or the regulatory environment) create sufficient incentives. In this context the market structure is still relevant, but in relation to the problem of the viability of the innovation process. The prevailing market structure emerges, in fact, as the result of the process of innovation, its specific features depending on the actual evolution of this process, that is, on productivity and demand as they are determined by how this process takes place. Thus, while standard models of oligopoly or monopolistic competition usually deal with the degree of competition and the characteristics of industrial structures as determined by given information and cost conditions, we put the focus on a dynamic process of rivalry that is determined by changing costs and information conditions. This process can result in a waste of productive resources and no real advantage for the customers

Competition and Increasing Returns • 259

or, alternatively, may allow firms and/or customers to benefit f r o m increasing returns. It can likewise result in a strongly unstable market structure (failure of the innovation process) or the opposite, in a fairly stable structure (viability of this process). Within this f r a m e w o r k , the nature of the shocks that affect firms (industry, the economy) does not really matter. These shocks always come d o w n t o a d e m a n d for new productive resources that will result in a productive structure functioning in such a way that the benefits of the change can be enjoyed only if the co-ordination problems raised by the shocks themselves are dealt with properly. The analysis, far f r o m referring to a general equilibrium system characterised by full information, offers the prospect of a general theory of economic evolution. It will be carried out by means of a model that draws on the one built by Amendola and Gaffard (1998, 2006), which makes it possible to exhibit the time structure of production processes and to sketch out the sequential interaction of decisions in a process of restructuring of productive capacities for the whole economy. In this paper, though, the focus will be placed on the behaviour of the single firms and the evolution of market shares resulting f r o m their interaction, leading or less to a concentration process. In particular, the difference between innovating and imitating behaviours will be stressed and the relevance of the conditions of access to external financial resources put into light. The results obtained, on which we shall come back in the conclusions, will allow to give a proper answer to the specific research question of the paper: whether, h o w and when, competition is consistent with the increasing returns that are at the heart of innovation processes.

2

The model

State and control

variables

T h e system is described by state and control variables. The state variables, for a representative firm or for each firm i, are: x'(t) the vector of production processes; m'(t) the money proceeds f r o m sales; h'(t) the monetary idle balances; o'(t) the stock of final output; ω'(ί) the wage f u n d ; ψ'{ί) the available h u m a n resources; d'{t) the volume of final demand; s'(t) the volume of supply; S'(t) the market share. The control variables are: χ j (i) the rate of starts of production processes; u' (t) the rate of scrapping of production processes; τ ' ( ί ) the rate of utilisation of productive capacity; p'(t) the price of final output; w'(t) the wage rate; f'(t) the external financial resources which depend on banking policy; η'it) the fraction of total real stocks actually put back on the market. These are either determined exogenously in the simulations (the open-loop control variables) or according to feedback mechanisms (the close-loop control variables).

260 · Mario Amendola and Jean-Luc Gaffard

The structure of productive capacity In each firm i production is carried out by means of processes of a Neo-Austrian type. An elementary process of production is defined by the input vector: ή =

k = \,...,nc+

nu

whose elements represent the quantities of labour required in the successive periods of the phase of construction c (from 1 to nc) and following it, of the phase of utilisation u (from nc + 1 to nc + n") of the productive capacity of commodity j, so that: a)

=

[ 1, that is when increasing returns to adoption prevail, the only change is that the number of firms that characterises the dynamic equilibrium will be smaller (Figure 4). In contrast, a weak resource constraint (k = 1), by favouring investment on the part of the incumbents themselves, makes it difficult for new firms to enter and simultaneously results in a relatively strong instability in the market shares, which is associated with an increase in the concentration index. Costs and unit margins strongly fluctuate. There are actually no productivity gains from innovations. In the case of an initial atomistic structure, a selection process takes place (a so-called shake-out of firms), which has a cumulative character. As a matter of fact, any exit results in a reduction of the average market price (the exiting firms being of course the less competitive ones, that is, those

Competition and Increasing Returns • 269

DISPERSION OF MARKET SHARES

LABOUR PRODUCTIVITY

PERIOD

PERIOD

Figure 3 Innovation and competition with strong financial resources constraint (large initial number of firms) (multiple runs)

25

NUMBER OF FIRMS

LABOUR PRODUCTIVITY

PERIOD

PERIOD

Figure 4 Innovation and competition with increasing returns to adoption

270 · Mario Amendola and Jean-Luc Gaffard NUMBER OF FIRMS

PRICES 2.61

40

1

[

[

[

PERIOD

PERIOD

MARKET SHARES

LABOUR PRODUCTIVITY

PERIOD

PERIOD

UNIT MARGIN

UNIT COST

0.5

20

,

60

80

100

PERIOD

120

140

PERIOD

Figure 5 Innovation and competition with weak resources constraint (large initial number of firms)

Competition and Increasing Returns · 271

NUMBER OF FIRMS

20

LABOUR PRODUCTIVITY

40

100

PERIOD

1 o

PERIOD

Figure 6 Innovation and competition with strong resources constraint and price volatility NUMBER OF FIRMS

LABOUR PRODUCTIVITY

100

120

140

160

180

200

PERIOD

Figure 7 Innovation and competition with strong resources constraint, price volatility and horizontal differentiations of product

charging higher prices) which makes marginal firms more fragile and can even push them out of the market (Figure 5). In other words, the instability of the market structure appears as an obstacle to the viability of the innovation process. W h e n prices evolve together with the current costs and hence are volatile, despite the existence of resource constraints, strong turbulence occurs that prevents the economy f r o m being viable. There is a shaken out process that does not necessarily result in the stabilised market structure that would be associated with the capability of each firm to really get the productivity gains (Figure 6). Nevertheless, with the latter price regime, in the case of monopolistic competition {β < 1), w h e n the global market is segmented between customers that d o not react to changes in prices charged by different firms, the market structure is stabilised and productivity gains obtained. This is because market shares are much less sensitive to price gaps. Price fluctuations do not p e r t u r b the d e m a n d profile and hence the temporal structure of productive capacity (Figure 7). A f o r m of market p o w e r is fruitful. T h u s technological advances per se d o not determine the dynamics of

272 • Mario Amendola and Jean-Luc Gaffard

the number of firms. On the contrary, this should actually be identified only once the stabilisation of the market structure signals that the viability conditions have been fulfilled. Viability conditions can be associated with different institutional frameworks. Different market structures can emerge from the same kind of innovation process, depending on the effective working of the co-ordination mechanism that reflects the prevailing rules and institutions. 4

Conclusion

In a stable market structure, competition among the surviving firms still prevails, in the sense that these firms continue to innovate, obtain the gains of innovation and distribute them, though they do not push others out of the market (see Richardson 1998). In this sense competition is compatible with increasing returns. There is no intrinsic contradiction between thinking that internal economies are a main factor of 'progress towards equilibrium', and imagining that increasing returns can be consistent with a given state of the industry (Robertson 1924). A good scenario will be the result of the way competition operates. In this context, the reality of increasing returns is twofold. On the one hand: The cost savings made available by an increase in the scale of a particular economic activity do not manifest themselves uniformly in all its stage and, as a result, an increase in demand may lead, not to an increase in the size of the enterprises undertaking the activity, not to an increase in the number of such enterprises, but to a change in industrial structure, those stages exhibiting the greatest scale economies becoming the business of specialist suppliers (Richardson 1998: 168). As a consequence, 'increasing returns would lead to specialisation and interdependence rather than to straightforward concentration' (ibid.: 163). On the other hand, increasing returns, which can be obtained by each firm in a context of a sequential competition consisting in the introduction of new products belonging to the same general market at different moments of time, raise only a transitory competitive advantage. Several firms can coexist in the market, despite the existence of increasing returns, yet remain differentiated not so much because they supply differentiated goods, but because they are each one at a different stage of the life cycle of the production process. The reason is that production takes time and new products gain market share only after a lapse of time "as their merits become apparent and as the capacity to produce them is built up" (Richardson 1998: 172). The latter situation can be really defined as a dynamic equilibrium. This is a situation in which competition causes 'the rate of investment in product development to rise or fall towards the level at which this investment yields only a normal return' (Richardson 1998:172). This is a situation in which prices charged by firms reflect decreasing average costs so that the benefits from innovation can also be distributed to consumers. Finally, this is a situation in which the stability of market shares obtains: there are neither new entries nor exits from the market, or more precisely, entries and exits do not change the market structure and look like a purely random phenomenon. All these considerations not only qualify a dynamic equilibrium, but also what competitive conditions consistent with increasing returns should be. The results of the analysis carried out offer, finally, interesting policy suggestions. In particular, that monetary and banking policies must be such as to allow investments

Competition and Increasing Returns · 273

to be neither insufficient nor excessive with respect to those required to keep both the internal and the external equilibrium structure of the productive capacity of firms. And that in reaction to current market disequilibria successful innovation processes require not only sufficiently strong financial constraints but also price (and wage) rigidities, that is, the opposite of the flexibility usually stressed as the advocated feature of markets. The market structure that emerges from successful innovation processes, thus, involves price stickiness's and resource constraints, that is, the so-called market imperfections looked at as the main obstacle to innovation and growth.

References Amendola, M., J.-L. Gaffard (1998), Out of Equilibrium. Oxford: Clarendon Press. Amendola, M., J.-L. Gaffard (2006), The Market Way to Riches: Beyond the Myth. Cheltenham: E. Elgar. Baumol, W.J. (2002), The Free-Market Innovation Machine. Princeton : Princeton University Press. Hicks, J.R. (1973), Capital and Time. Oxford: Clarendon Press. Hicks, J.R. (1989), A Market Theory of Money. Oxford: Clarendon Press. Leijonhufvud, A. (2006), Market Adjustment Processes. Pp. 226-236 in: T. Raffaelli, G. Becattini, M. Dardi (eds.), The Elgar Companion to Alfred Marshall. Cheltenham: E. Elgar. Marshall A. (1920), Principles of Economics. 8th edition. London: Macmillan. Richardson, G.B. (1998), The Economics of Imperfect Knowledge. Cheltenham: E. Elgar. Robertson, D.H. (1924), Those Empty Boxes. Economic Journal 34: 16-30. Reprinted in G.J. Stigler, K.E. Boulding (eds.) (1953), Readings in Price Theory. London: Allen and Unwin. Smith, A. (1976), An Inquiry into the Nature and Causes of The Wealth of Nations. Edited by Edwin Cannan. Chicago: The University of Chicago Press. Sraffa, P. (1926), The Laws of Returns under Competitive Conditions. Economic Journal 36: 535-550. Reprinted in G.J. Stigler, K.E. Boulding (eds.) (1952), Readings in Price Theory. Homewood, 111.: Irwin. Mario Amendola, University of Roma La Sapienza, Dipartimento di Economia, Piazzale Aldo Moro, 5, 00185 Roma, Box 83 Roma 62, Italy. [email protected] Jean-Luc Gaffard, OFCE, SKEMA Business School, 60 rue Dostoïevski, 06902 Sophia Antipolis Cedex, French. [email protected]

Jahrbücherf. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

Innovation and Market Structure in Pharmaceuticals: An Econometric Analysis on Simulated Data Christian Garavagliaabr Franco Malerbab,cr Luigi Orsenigob'd, Michele Pezzonib,e,f* a DEMS,University of Milano-Bicocca b CRIOS, Bocconi University c Bocconi University, Department of Economics d IUSS (Institute for Advanced Studies), Pavia e OST, Observatoire des Sciences et des Techniques, Paris f CEMI, École Polytechnique Fédérale De Lausanne JEL C63; C15; L10; L65; 0 3 0 Innovation; market structure; history-friendly; pharmaceuticals; Monte Carlo.

Summary The contribution of this paper is twofold. First, it presents the results of a "history-friendly" simulation model of evolution of the pharmaceutical industry. Second, it aims at contributing to a more general methodological discussion about agent-based models by proposing an econometric analysis of the results of the simulations. The case of the pharmaceutical industry has been studied extensively by scholars because, despite the high level of R & D intensity, the industry has been characterized by a relatively low levels of concentration. The model is able to reproduce the main stylized facts of the industry in an evolutionary perspective. In this paper we extend the analysis conducted in two previous works (Garavaglia et al. 2012, 2013) by further qualifying the findings with an extensive econometric investigation of the model outputs. The paper focuses the attention on the determinants of market structure, the innovative performance of the industry, the diversification in multiple submarkets and the level of prices. We find that the properties of the technological and demand regimes are key determinants of the patterns of industry evolution and that the main mechanisms driving the model are the random processes of search, the discovery of new submarkets as well as the interactions between patent protection, imitation and price competition. In addition, this paper emphasizes how the emerging leaders in the industry are those innovative early entrants which entered in large submarkets, showing the importance of the first mover advantage and of the size of the "prize" accruing to innovators when they discover a new rich submarket.

1

Introduction

This paper presents an econometric analysis of the results generated by a "historyfriendly" model of the evolution of the pharmaceutical industry. This systematic statistical investigation allows us to strengthen and further control for sensitivity and robustness * The authors thank two anonymous referees and the Editor for their useful suggestions. Christian Garavaglia would like to thank the participants of the 13th Conference of the International Schumpeter Society. The usual disclaimers apply.

Innovation and Market Structure in Pharmaceuticals · 275

of our previous results and of their interpretation. In so doing, this paper hopefully contributes also to the broader methodological discussion about the validation of simulated, agent-based, models (Windrum et al. 2007). Our history-friendly model examines the so-called "golden age" or "random screening" era of the pharmaceutical industry (Garavaglia et al. 2012, 2013). In particular, we focus our attention on a suggested interpretation of a set of stylised facts characterising the evolution of this industry which the model is able to replicate. Specifically, despite being traditionally a high R & D and marketing intensive sector, pharmaceuticals has been consistently showing low level of concentration over its whole history. Yet, adding to the puzzle, the industry has been dominated by a core of innovative firms which has remained quite small and stable for a very long period of time, suggesting the existence of first mover advantages. In Garavaglia et al. (2013), we showed that our model did a good job in replicating simultaneously these facts. In Garavaglia et al. (2012), we showed - by inspecting qualitatively the output of simulated data and by performing "counterfactual" runs - that these observed patterns could be explained by the specific features of the technological and demand regimes which characterize the pharmaceutical industry. Technological regimes are defined by the opportunity and appropriability conditions denoting innovative activities and by the degree of cumulativeness of technological advances. Demand regimes are defined by the size and the degree of fragmentation of the market in several independent submarkets (Malerba/Orsenigo 1995, 1996; Breschi et al. 2000; Garavaglia et al. 2012). In this paper, we implement an econometric investigation of our simulated data to probe further for the robustness of these results. Specifically, we investigate - implementing different econometric exercises - how the observed patterns of the evolutionary dynamics depend on factors related to the nature of the relevant technological regime and/or to the structure of demand. In particular, building on Garavaglia et al. (2012), we argue that the combination of the following main factors goes a long way to explain the patterns of industry evolution in this industry: a) The nature of the processes of drug discovery, which has been characterized for a very long time by low degrees of cumulativeness and by "quasi-random" procedures of search (random screening). Thus, innovation in one market (a TC, therapeutic category) does not entail much higher probabilities of success in another one. b) Patents are a fundamental appropriability device in this industry, which confers sizable economic advantages to innovators. Yet, competition remains strong through processes of "inventing a r o u n d " , the development of so-called "me-too drugs" and after patent expiry - imitation and the entry of generics. c) Pharmaceuticals are composed by a large number of independent submarkets (TCs): for example, cardiovascular products do not compete with antidepressants. Given the "quasi-random" nature of the innovative processes, innovation in one TC bears little direct consequences on the ability to innovate in another submarket, thus limiting the emergence of monopolistic positions. This interpretation is partly in line with other, more conventional, explanations of market structure in pharmaceuticals in the economic literature. In particular, Sutton (1998) provides a game theoretic framework for analyzing this issue and he identifies very similar properties of technology and demand as the ultimate determinants of market structure. However, the mechanisms, processes and basic assumptions driving Sutton's and

2 7 6 · C. Garavaglia, F. M a l e r b a , L. O r s e n i g o , a n d M . Pezzoni

our approaches are quite different, in that our analysis is dynamic and evolutionary in nature. Thus, our model is able to account also for additional aspects of the history of this industry, like the existence of significant first mover advantages. The paper is organized as follows: section 2 presents a stylized history about the evolution of market structure and the nature of competition in pharmaceuticals. Section 3 presents the model and the baseline simulation results, and in section 4 we implement an econometric analysis on the determinants of market concentration. Section 5 concludes the paper. 2

Relevant industry b a c k g r o u n d : the m a i n facts to be explained

2.1

The evolution of market structure in pharmaceuticals

The history of the pharmaceutical industry has been described and discussed by a large literature. 1 In Garavaglia et al. ( 2 0 1 2 , 2 0 1 3 ) we provide a succinct account of the main traits of the evolution of this industry which our model attempts at explaining. In this paper, our analysis focuses on one puzzling feature of the evolution of this industry, namely that throughout all its history, pharmaceuticals has been characterized by relatively low levels of concentration, especially when compared to other R & D and marketing intensive industries. Ever since its start, market structure has been characterized by a core of leading firms and a large fringe of smaller ones, with no clearly dominant positions emerging in the USA and in the other large European economies. 2 Over the last decade concentration has been increasing, despite the entry of the new biotechnology firms and the expansion of the generic segment of the industry, mainly as a consequence of mergers and acquisitions. Yet, in 2 0 0 4 , the largest pharmaceutical firm held a world market share close to 10 % and the C R 5 concentration ratio was around 1/3 in the US and in the EU. Within the core of the first 1 0 - 2 0 largest firms around the world, competition is intense; changes in the hierarchy of the leaders occur but, despite this mobility within the core, the club of the major firms has been remarkably stable (Pammolli 1 9 9 6 ) . Similarly, entry has not been a significant phenomenon at least until the biotechnology revolution. This picture is different, however, at the level of the individual submarkets or therapeutic categories, ( T C s ) , like e.g. cardiovascular, diuretics, tranquilizers, etc. The largest firms

1

1

See, among others, Pisano ( 1 9 9 6 ) , Henderson et al. ( 1 9 9 9 ) , Malerba and Orsenigo ( 2 0 0 2 ) , Sutton ( 1 9 9 8 ) , Pammolli ( 1 9 9 6 ) , Grabowski and Vernon ( 1 9 7 7 , 1 9 9 4 ) , Chandler ( 1 9 9 0 , 1 9 9 8 , 2 0 0 5 ) , Galambos and Sewell ( 1 9 9 6 ) , Galambos and Sturchio ( 1 9 9 6 ) , Gambardella ( 1 9 9 5 ) . As compared to these previous versions, the model has been modified in many respects (see Garavaglia et al. 2 0 1 3 ) . The main change concerns the possibility of running parallel projects. The empirical evidence shows that firms engage several simultaneous drug developments in order to differentiate the high the risk of failure: therefore now firms are able to develop several R & D projects at the same time. W e also improved the pricing rule moving from a fixed (exogenous) mark-up to a dynamic (endogenous) pricing routine. In this version of the model, the price of a product at time t depends on the time series of its previous prices, the elasticity of demand, and the market power of the firm within the TC. W e updated also the routine which assigns value to each T C in each simulation period, which is now a function of the number of patients and the degree of competition within the TC. Also, the firms' propensity to innovate and imitate has been updated: in the new version of the model we introduced heterogeneity of consumers in terms of preferences for effective drugs. Finally, three new exit rules has been introduced in order to take into account research inefficiencies, financial difficulties, unattractive marginal position both of the product and of the firm.

Innovation and Market Structure in Pharmaceuticals · 277 are present in a large number of submarkets where they hold indeed dominant positions. In some submarkets, the CR4 index was above 80 percent in 1995, and in many others only two or three drugs account for more than 50 percent of the market sales (Chong et al. 2003). These firms also represent the most active firms in terms of innovative output (measured by the introduction of NCE (new chemical entity) in the market). Yet, even in submarkets, dominant positions are not infrequently temporary and contestable. A further significant observation is that even in this highly competitive industry some forms of first mover advantages seem to play an important role. The "oligopolistic core" of the industry has been composed of the early innovative Swiss and German firms, joined after World War II by American and British companies, all of which maintained over time an innovation-oriented strategy. Many of the leading firms during this period companies such as Roche, Ciba, Hoechst, Merck, Pfizer, and Lilly - had their origins in the "pre-R&D" era of the industry. 2.2

Factors affecting the evolution of market structure

Which factors and processes can explain at the same time low degrees of concentration in the aggregate, coexisting with persistent stability of a relatively large core of innovative, diversified (in multiple individual submarkets) industry leaders, but also with higher dynamism and market power in individual submarkets? The historical and empirical literature suggests three main factors which may explain these patterns: a) The nature of the processes of drug discovery, i.e. the properties of the space of technological opportunities and of the search procedures through which firms explore it. Innovative processes have been characterized for a very long time by "quasi-random" procedures of search (random screening), extreme uncertainty and very high rates of attrition. In this respect, as suggested by Sutton (1998) the process of discovery and development of a drug closely resembles a lottery. Moreover, it is very difficult to leverage the results of past innovative efforts into new products: in other words, economies of scope and cumulativeness of technological advances are limited. b) The type of competition and the role of patents and imitation. In pharmaceuticals patents are a fundamental appropriability device. They confer temporary monopoly power to the innovator but competition remains strong through processes of "inventing around" and - after patent expiry - imitation. c) The fragmented nature of the relevant markets. Pharmaceuticals are composed by a large number of independent submarkets (therapeutic categories, TCs): for example, cardiovascular products do not compete with antidepressants. And, given the "quasirandom" nature of the innovative processes, innovation in one therapeutic category does not provide further advantages on the ability to innovate in another market. Hence, concentration tends to be lower as compared to a homogenous market. The point is seemingly very simple. If a market of size 100 is homogeneous, a firm gaining a competitive advantage can maintain and reinforce it through learning processes and success-breeds-success processes. By contrast, if the same market is divided into 10 markets of size 10, absent scope economies and cumulative learning, there is a higher probability that different firms dominate different niches. Successful firms still have an advantage since they have more resources to invest and therefore higher chances to discover new drugs in new submarkets. Yet, as long as the differential profits accruing to an innovator are not too large, there is always a positive probability that a competitor will discover a new drug in a different submarket.

278 • C. Garavaglia, F. Malerba, L. Orsenigo, and ΛΛ. Pezzoni

As mentioned earlier, a few formal, theoretical models have been developed to explain these stylized facts. In particular, Sutton (1998) has applied his model of the relationship between market structure and innovation to the pharmaceutical industry. In this approach, a main determinant of industry concentration is the incentive for firms to outspend their competitors: if the profit deriving from spending a little bit more in R & D than competing firms is large, then an escalation mechanism is set in motion which leads to high concentration (in Sutton's terms, this is called the "escalation parameter" alpha). In the case of pharmaceuticals, the random nature of the discovery process, the absence of economies of scope in R & D and particularly the fragmentation of the drug market in several independent submarkets imply that the value of alpha and hence the degree of concentration are low. Our analysis overlaps with Sutton's analysis in identifying very similar properties of technology and demand as the ultimate determinants of market structure. However, the mechanisms, processes and basic assumptions driving Sutton's and our approaches are quite different. First, in an evolutionary perspective, we do not assume full rationality on the part of the agents nor do we pre-impose equilibrium conditions. Second, our setting is explicitly dynamic, where competition unfolds over time and market structure emerges as the outcome of (complex) interactions among agents and structural variables. Third, we do not exogenously fix the number of submarkets: although there is a fixed number of "potential" submarkets, only some of them are actually discovered through R & D efforts, so that the degree of market fragmentation is partially endogenous. Finally, as a consequence, our model is also able to account for further intrinsically dynamic phenomena, primarily the existence of first mover advantages. In a different context, Klepper (1997) suggests that product differentiation and demand fragmentation into many niches may prevent shakeouts and the emergence of concentration. Generalizing this intuition, Klepper and Thompson (2006) develop a model in which the process of (exogenous) creation and destruction of submarkets drives industry evolution. Firms expand by exploiting new opportunities that arrive in the form of new submarkets, while they shrink when the submarkets in which they operate are destroyed. The model predicts that a shakeout occurs and concentration increases if the rate of creation of new submarkets slows down and/or a new very large submarket appears. The exploitation of economies of scale and especially economies of scope across different product varieties reinforces this tendency (see also Buenstorf/Klepper 2010). Yet, these models are based on an extremely simplified conceptualization of the innovative process and of the exogenous arrival/destruction of new submarkets. 3 3.1

The model History-friendly models

The aims and the logic of history-friendly models (HFMs) have been presented and discussed at length elsewhere (see for example, Malerba et al. 1 9 9 9 , 2 0 0 7 , 2008; Garavaglia 2010). Suffice it to briefly remind here that H F M s aim at the construction of models of industry evolution based on the formalization of appreciative theories about mechanisms and factors affecting the evolution of specific industries suggested by empirical research. H F M s attempt at bridging the gap which is often observed between detailed-rich, empirical and historical accounts of specific phenomena and "general theories", almost always formalised in mathematical models. Formal models are thus considered as attempts at

Innovation and M a r k e t Structure in Pharmaceuticals · 2 7 9

checking the logical consistency and the robustness of the verbal arguments that constitute the appreciative theory. T h e construction of formal models of specific industries might be useful in forcing the theorist trying to apply a general model to a specific case and calling both for more realism t h a n it is sometimes the case and for stronger awareness of the distance that might exist between any "general" theory and the issue under investigation. History-friendly models, and in general evolutionary and agent-based models, aim at and allow the researcher to deal with a much higher degree of complexity than in conventional o r t h o d o x models. Detailed characterization of agents results in a bottom-up perspective of the model, starting f r o m an agent's definition at micro-level and ending u p with emerging macro-dynamics that usually is w h a t interests to analysts. The micro-level approach, together with the b o u n d e d rationality assumption and the explicit description of the interactions between agents, generates the high level of complexity that characterizes these models (Windrum et al. 2007). Simulation is then a f u n d a m e n t a l tool for capturing such complexity in the agents' characteristics and in their inter-relationships (Garavaglia 2010). 3.2

The essential features of the model 3

This model is an improved version of Malerba and Orsenigo (2002) and Garavaglia et al. (2006). 4 It is not the purpose of this section to formally discuss the essence of the model. The complete formal structure of the model is presented in Garavaglia et al. (2013). Here we limit ourselves to summarise its basic logic and functioning. In the model, the market is composed by a given number of therapeutic categories or submarkets, TCs, characterized by a different n u m b e r of potential patients expressing the d e m a n d for drugs. Patients in each T C are grouped according to their willingness to buy drugs characterized by different qualities. A number of firms (a parameter of the model) compete to discover, develop and market new drugs for a large variety of diseases a n d , other things being equal, TCs having a larger economic value tend to be more attractive for firms. 5 Firms are heterogeneous, in that they are characterised by different propensities - i.e. firm-specific, time invariant fractions of the available budget invested - t o w a r d s innovation, imitation and marketing. Each T C is characterized by a given spectrum of opportunities, represented by the n u m b e r of molecules having a therapeutic and (therefore potential) commercial value (quality Q). Firms aim to discover new valuable molecules. At the beginning, firms face a large space of unexplored opportunities. The search for new promising molecules is essentially r a n d o m , because the knowledge of w h y a certain molecule can " c u r e " a particular disease is limited. Thus, innovating firms explore randomly the "space of molecules" until they find one which might become a useful drug a n d patent it. T o d o this, firms invest in a search process, spending a fixed cost t o " d r a w "

3 4

5

This Section, as well as Section 3.3 and 3.4, draw heavily on Garavaglia et al. (2010). Up until the mid-Nineties, no firm had a worldwide market share larger than 4.5 %: the market shares of the top twenty firms ranged in 1996 from 1.3 % to 4 . 4 % . The CR4 index was equal to 2 8 % in 1947, 2 4 % in 1967 and 2 2 % in 1987 (Sutton 1998). As just mentioned in footnote 2, the economic value of each TC is endogenously determined by summing the revenues of each drug /' sold at a given time-variable price (Pricejj). Therefore, even if the number of patients is exogenously given, the economic value of the TC changes during the simulation according to the monopolistic power stemming from patents and the degree of competition among firms.

280 • C. Garavaglia, F. Malerba, L. Orsenigo, and M. Pezzoni a molecule. Thus, the number of molecules drawn by a firm in each period is determined by the ratio between the fraction of the available budget, and the cost of a draw. Firms do not know the quality Q of the molecule that they have drawn: they only know whether Q is greater than zero or not. If the molecule has Q > 0 and it has not been patented by others, then a patent for that molecule is obtained. The patented molecules become part of an individual 'portfolio' that each firm maintains for potential drug development. The patent provides protection from imitation for a certain amount of time and over a given range of "similar" molecules. Imitators select molecules among those whose patent has expired: for imitators the cost of search equals zero. Both innovator and imitator develop products from molecules by engaging in drug development activities. If the molecule is potentially interesting (i.e. it has a quality Q greater than zero), the firm starts a costly development project. The time and the cost necessary to complete a development project are assumed - for sake of simplicity - to be fixed and equal for all molecules and firms, the only difference being that innovative are more expensive and time consuming than imitation. If the cost of development of a potential drug is larger than the available budget, the molecule remains stored in the firm's portfolio, and the firm until enough financial resources have been accumulated to start one or more new projects. When the project ends the quality of the molecule (the new drug) is revealed. Products must have a minimum quality to be allowed to be sold in the marketplace. In other words, products are subject to a "quality check" by an external agency (e.g. the FDA). Below this value, the drug cannot be commercialized and the project fails. When a product originates from a molecule which has never been used before, it is labelled as an innovative product, otherwise it is considered an imitative product. If the drug is authorised by the FDA, marketing expenditures allow firms to increase the number of patients they can access. Decision to buy a drug by consumers6 depends on several factors, which together yield a specific "merit" to each drug at time t. The "merit" of a drug to the consumers eyes is a function of its quality, of the level of marketing "image", and of its cheapness (i.e. the inverse of price). The relevance of each of these factors is different (and randomly determined) across TCs. Prices are defined by firms according to a mark-up rule over the costs of manufacturing (equal for all firms and drugs). The mark-up varies as a function of the competitive pressure in the TC and of the price elasticity of demand, ε, in the submarket. At the beginning, the new drug is the only product available on that particular TC. Hence, the innovator enjoys high prices and profits. Profits increase the budget available for further investment in drug discovery, development and marketing. But other firms can discover competing drugs or imitate in the same TC. The innovator will therefore experience a burst of growth following the introduction of the new drug, but later on its revenues and market shares will be eroded away by competitors and imitators. The discovery of a drug in a particular TC does not entail any advantage in the discovery of another drug in a different TC - except for the volume of profits they can earn. Moreover, the various TCs that define the aggregate pharmaceutical industry are independent from one another also on the demand side: for example, an anti-ulcer drug is useless for a patient suffering Alzheimer. As a consequence, the discovery of new promising molecule and 6

For reasons of simplicity, we do not distinguish between patients who use the drug and physicians who prescribe it.

Innovation and Market Structure in Pharmaceuticals • 281

hence diversification into different TCs is also purely random. Firms will start searching again randomly for a new product everywhere in the space of molecules. Firms' growth will then depend on the number of drugs they have discovered, on the number of markets they are active in (i.e. diversification into different TCs), on the number of competitors, on the relative quality and price of their drug vis-à-vis competitors. In few cases, a firm may discover a blockbuster, i.e. a high quality drug that has a large market. Higher profits set in motion a success-breeds-success process sustained by the reinvestment of profits, the random discovery of a "blockbuster" and to diversification. Thus, a few firms will grow and become larger. However, given the random nature of the search process, the large number of TCs and the absence of any form of cumulativeness in the innovative processes, firms have little hope to be able to gain a large aggregate market share, but if anything - only in specific TCs for a limited period of time. As a result, the degree of concentration in the aggregate market for pharmaceuticals will tend to remain low. Conversely, firms will withdraw a product from the (sub-) market when its (sub-) market share falls below a given threshold. Also, firms may exit the aggregate market altogether for a number of different reasons, reflecting research inefficiency (innovative firms exit if they persistently fail to discover new potentially valuable molecules), financial difficulties (failure occurs when a firm does not have the minimum budget needed to complete one project and it is not selling or making other products), and unattractive position of the firm in the market (firms exit when their aggregate market share is lower than a minimum threshold). Under these conditions (absence of significant economies of scale and scope, independence across submarkets), a skewed distribution of firms' size is likely to emerge. 3.3

Parameters setting

The results of the simulations are averages over 100 runs. The "Standard Set" parameterization reflects both some fundamental theoretical hypotheses and, in a qualitative way, some empirical evidence. The Standard Set is broadly considered as "history-friendly" and it serves the purpose to produce a benchmark for subsequent analyses. 7 The calibration of the model is the result of a process of repeated changes in the parameters and methods of the model in order to obtain a satisfactory specification. Some parameters are selected on the basis of the knowledge we have about their meanings and values as shown by the empirical literature and the evidence provided by industry's specialists. The value of other parameters has been selected with the view to preserve coherence. In many cases, the parameterization of some key variables of the model is largely ad hoc·. for example, we do not know the "real" distributions of the opportunities of discovery and we have only some rather generic knowledge about the economic value of the developed drugs. Because most parameters fall into groups with a particular mechanism in the model, there is typically some common-sense guidance available for choosing plausible orders of magnitude. Many value choices for parameters involve implicit unit choices for variables, which means that the quantitative values are in the end arbitrary (or matters of convenience), but also means that relations among parameters affecting the behaviour of the same variables have to be made with a view to consistency. For example, it does not matter what range of numerical values represents the aggregate value of sales in our model, but the relationship of production costs or R & D spending to that total sales does 7

In the Standard Set there are no economies of scale, no economies of scope and no processes of mergers and acquisitions. Even more important, there are no exogenous advances in knowledge that allow firms to focus their search activities and to increase the productivity of their research.

282 · C. Garavaglia, F. Malerba, L. Orsenigo, and M. Pezzoni

matter. An additional constraint disciplining the choice of parameters values is provided by the time structure of the model. History-friendly models p u r p o r t t o generate sequences of events that take place in (approximations to) real time. And the definition of w h a t "one p e r i o d " means in real time (one year in this model) is crucial for establishing which actions take place at any one period, which follow, etc. Hence, the time structure of the model imposes restrictions in order to respect consistency (Windrum et al. 2 0 0 7 , discuss alternative methods of model validation, different f r o m history-friendly approach). In our model, the landscape explored by firms is sufficiently rich in terms of o p p o r t u nities of discovery to allow for the survival of the industry and the introduction of a large n u m b e r of new drugs. However, search remains a very risky a n d most of the time unsuccessful activity: the parameter describing the probability of finding a "zero quality" molecule is set equal to 0.97: this means that only 3 % of the available molecules are potentially valuable, thus reflecting a very uncertain innovative process. Moreover, the quality value of the molecules is highly skewed. Search, development and marketing activities are expensive and take time. In line with the empirical evidence, we set the development time of a drug equal to 8 and 4 periods respectively for innovative and imitative products. Patent duration is set equal to 2 0 periods. The relative costs of search, development and marketing broadly reflect the costs currently observed in the industry (Di Masi et al. 2003). The number of TCs is also very high (200). Marketing expenditures have an i m p o r t a n t role in accessing a large number of customers and the sensitivity of d e m a n d to price is rather low. Emerging results are encouraging.

3.4

Baseline simulation results

Here, we report briefly the main results of the model under the " s t a n d a r d " parametrization. In each TC, concentration (measured by the Herfindahl index) tends to decrease quickly after an initial upsurge (Figure 1), reaching value of 0.3 at the end of the simulation runs. This pattern is due to the monopoly p o w e r of early entrants in each TC: gradually, after the introduction of new competitive innovative and imitative products in the same TC, the degree of competition rises and concentration decreases. N o t surprisingly, given market fragmentation in several TCs, aggregate market concentration is always much lower than in individual TCs (Figure 2). It increases after period 50 as bigger firms, exploiting their larger financial resources, are able to enter new TCs by discovering and developing new products and imitating existing ones. However, this tendency is countered by the intrinsic randomness of innovation a n d by imitation. Thus, aggregate concentration remains relatively low, reaching a value of 0.23 at the end of the simulation. Selection is particularly intense in the first half of the simulation. Only 2 0 firms out of 50 potential entrants succeed in actually entering the market, because either they fail to discover promising molecules or they are unable to complete the development process. Other firms d o succeed in entering the market but - for the same reasons as above and as a result of the pressure of competition - they subsequently exit. As time goes by, the industry becomes more stable: both entry and exit decrease and most of the surviving firms tend to remain alive until the end of the simulation. The leadership of the industry changes rather frequently at the beginning of the simulations, given the

Innovation and Market Structure in Pharmaceuticals · 283

Figure 1 Average Herfindahl index in each TC

Figure 2 Herfindahl index in the aggregate market non-cumulative nature of innovation. After this period of instability, however, a stable core of few large firms emerges. As a consequence of the processes of innovation, imitation and diversification, firms grow quickly and enjoy high levels of profitability. A skewed firm size distribution emerges, with some larger firms (mainly innovators) who are present in a high number of TCs, and many smaller companies (mainly imitators), 8 in line with the empirical evidence (Cabral/Mata 2003). There is a significant heterogeneity across TCs: in a few TCs there are no firms, while in the richer TCs several firms are present. Firms compete simultaneously through pro-

8

See Garavaglia et al. ( 2 0 1 3 ) for further details. A thorough analysis of the properties of firm's growth in the model as compared to real data is the object of further research.

284 · C. Garavaglia, F. Malerba, L. Orsenigo, and M. Pezzoni

Time

Figure 3 Number of innovative and imitative products

Time

Figure 4 Number of discovered TCs

cesses of innovation, imitation and diversification into new TCs. Innovation constitutes a fundamental ingredient of firms' competitiveness. In the early periods of the simulation, most products are - almost by definition - innovative. Afterwards, as time goes by, imitation starts to take place: after the patents of the first set of innovative products expire imitation occurs frequently. Both the number of innovative and even more so of imitative products increase (Figure 3). Also, firms increasingly diversify into new TCs. The rate of discovery of new TCs is quite high in the first part of the runs, but then it slows down (Figure 4). The decreasing rate of discovery is also correlated to the higher competitive pressure that imitative products have on innovative ones: firms tend to select imitative drugs in large submarkets rather than developing new products in smaller TCs. As time goes by, the prices of drugs decrease (Figure 5). The decline of prices of innovative products under patent protection is due to the higher degree of competition (the entry of

Innovation and Market Structure in Pharmaceuticals • 285

Time

Figure 5 Average price of products competing innovative products in each TC) along the simulation time span. The decline in prices of products whose patent has expired is even more pronounced because of the fiercer competition they suffer from imitation. Prices of imitative products, after an initial peak due to cases of duopoly between the first innovative incumbent and the second entrant, remain stable over time. The patterns of average earnings for these typologies of products follow a similar dynamics.

4 4.1

Econometric analysis of the model results Why an econometric analysis?

Simulation results are encouraging in reproducing the main stylised facts in the evolution of pharmaceuticals: a) aggregate market concentration remains quite low while concentration in each TC is higher; b) a stable core of leading firms emerge and the size distribution of firms is highly skewed; c) firms compete through innovation and imitation strategies, but market leaders are typically innovators; d) firms' diversification increases over time. In Garavaglia et al. (2012, 2013), we probed the robustness of the model and the consistency of our results by running counterfactual simulations (e.g., how would concentration change if patent protection had been longer or shorter?) and by conducting systematic sensitivity analysis. Those exercised confirmed that the dynamic of the model is fundamentally driven by the parameters capturing the properties of the technological regime (opportunity and appropriability conditions, cumulativeness of technological advances) and of the demand regime (market size and especially the degree of fragmentation of the aggregate market in multiple independent submarkets). More specifically, in Garavaglia et al. (2012) we observed that all the variables describing the technological regime tend to increase concentration in individual submarkets, but

286 • C. Garavaglia, F. Malerba, L. Orsenigo, and M. Pezzoni to decrease aggregate concentration. Conversely, other variables which are customarily referred to as important determinants of market structure, primarily economies of scale in production, have only marginal effects on aggregate concentration. The explanation of this result was identified in the high degree of fragmentation of the market. Higher concentration in individual submarkets implies higher profits, more resources available for the discovery of new molecules, and therefore the opening of new market niches. Indeed, in simulations with a low number of submarkets, both aggregate and individual concentration in submarkets tend to increase unambiguously. More specifically, the analysis suggested that the cause of these results is not the degree of market fragmentation per se, but the size of the "prize", i.e. profit, that an innovator gains when a new molecule is discovered. In a fragmented market, the "prize" obtained by an innovator is by definition smaller than what could have been earned in the presence of fewer submarkets (in the limit, just one market). Thus, the differential probability to further discover new molecules - through higher profits and more "innovative draws" - is comparatively lower. In other words, market structure is determined by the relative premium - the jackpot - that innovators can take advantage of with respect to the subsequent innovating and imitating firms: the higher this premium, the stronger are success-breeds-success effects, the higher market concentration will be.' This causal mechanism was found to account also for the existence of first mover advantages in the model: early innovators discovering a rich submarket gain a significant advantage vis-à-vis competitors and by reinvesting profits are able to maintain it over time. In this section, we broaden our analysis by running an extensive econometric investigation on the model outputs combined with a Monte Carlo sampling of the inputs. This technique is defined by some authors as an uncertainty analysis followed by a regression exercise, others simply define it sensitivity analysis (Kleijnen 1997). As it is generally intended, this technique aims at assessing the effect of input variations on outputs, in presence of complex stochastic (or deterministic) simulated systems. In this paper our aim is first to verify the robustness of our theoretical hypotheses and previous results and second to check systematically how the model output reacts to changes of the key parameters. Indeed, it is often observed that the interpretation of the causal relation between inputs and outputs of the simulation model may not be immediately transparent, given a) the complexity of the agents' interactions governing the micro-dynamics of the model and b) the presence of a stochastic component. An econometric analysis of the results of the simulations run over a different combination of the parameters can therefore help in disentangling and in identifying more precisely the role played by the key explanatory variables. ' Indeed, in Garavaglia et al. (2012) we show that holding constant the number of TCs and the value of the overall market, changes in the variance and in the skewness of the values of individual submarkets have a substantial impact on concentration. The explanation is that, if the overall market is composed by very few extremely rich TCs and many poor ones, concentration increases drastically: the firm discovering the large submarket gains also a large fraction of the overall market; the "size of the prize" matters. This result is in tune also with the theoretical expectations of other models. In particular, Sutton's model predicts that market fragmentation leads to lower concentration because the "escalation parameter", alpha, is lower. When markets are fragmented, the additional profits obtainable by a firm outspending rivals are limited: concentration remains low. Our model confirms - in a radically different theoretical context - this intuition: when the market is fragmented, the prize accruing to an innovator is limited. In dynamic terms, this observation implies also that the discovery of a rich submarket will raise abruptly concentration, as in Buenstorf and Klepper (2010).

Innovation and Market Structure in Pharmaceuticals · 287

Further, we extend our analysis by considering additional dependent and explanatory variables that has not been systematically examined in our previous papers. Among the dependent variables, we focus our attention not only on concentration but also on the innovative performance of the industry and on prices. We also consider the role played by other variables and parameters which may be potentially important in influencing results. Specifically, we examine first the initial number of potential entrants: how does the model behave if the initial population of firms is set larger or smaller? Second, we look at the intensity of price competition: what are its effects on concentration, innovativeness and prices? Third, we investigate the relevance of firms' strategies: how do the key dependent variables behave with different mixes of innovative and imitative strategic orientations? For some of these additional explanatory variables we have strong a priori expectations on their effects on dependent variables. For instance we expect that the number of potential entrants should impact negatively on industry concentration as well as price competition should decrease average drug prices. We consider these expected results as a validity test for our simulation model. 4.2

Methodology: econometrics as a tool for analysing simulation results

In this section we describe the methodology we implement to analyse simulated data. We identify a model output of interest, Yt, and a set of time invariant parameters Ζ that represent some basic characteristics of the industry. Parameters Ζ are exogenously defined and are expected to influence the selected output. We are interested in assessing the relation between Yt and Z, however in order to conduct our analysis we have to consider two aspects characterizing history-friendly models. Firstly, Windrum et al. (2007) point out that the characteristics of agent-based simulation models cause a strong path-dependency effect, i.e. the values of Yt depend on the past states of the simulated system. In order to control for path-dependency, we include as determinants of Yt a set of time variant factors Xt~i that characterize the state of the system in t — 1 X t _ j are all the variables considered to be important for the dynamics of the system (e.g. number of innovative and imitative firms on the market at time t — 1). Secondly, history-friendly models are characterized by the presence of a stochastic component (ξί), which is what accounts for variability of output in two (or more) distinct simulations even when all the parameters Ζ are kept constant. We, then, represent the relation of interest between Yt and Ζ as in Equation 1 where we include also factors X f _ i , and the stochastic component that contribute in determining the model output.

Yt = giz,xt.1,b)

(i)

The function g is in principle known but in practice too complex to be handled. It represents all the routines governing the micro-dynamics of the system. Then, the effect of the parameters Ζ on Yt is not easily measurable because of the complexity of g. In order to investigate the causal effects between parameters and the variable of interest, we follow a two-step analysis. We start by running a standard Monte Carlo sampling on the parameters Z. The data generation process lies on the assumption that each parameter Ζ is a random variable with a given uniform distribution. The values of the inputs in each simulation run are the result of independent random draws from these distributions.

288 · C. Garavaglia, F. Malerba, L. Orsenigo, and M. Pezzoni In the second step, we specify a meta-model as in Equation 2, where coefficients ¿o, bz and bx measure the impact on Y¿ of the parameters Ζ and of the variables X¿_ ), while the stochastic component is mainly included in the error term e. Coefficients of equation 2 can be easily estimated by mean of OLS. Yt = b0 + bz * Zs + bx *Xt-i

(2)

Although the use of the meta-model in Equation 2 makes some simplifying assumptions such as the linearity of g, it provides at least three major advantages. First, it isolates the effect of each parameter Z, ceteris-paribus in terms of other variables. Second, simulation models, and in particular history-friendly models, are sometimes too rich in terms of results and it becomes difficult to identify interesting effects in such a mass of possible inputs and outputs. Using a meta-model allows us to make inference and then to identify which of the parameters are actually relevant in influencing the output Yt. Finally, a metamodel allows to rank the variables according to the extent of their effects on Yt (i.e. it allows to compare sizes of coefficients b in case they are standardized). In section 4.4, we apply the methodology described to five distinct model outputs Yt namely, aggregate concentration (H), concentration in TCs (HTC), number of innovative products (Inno), number of explored TCs (% of TC viewed), price of innovative products (Price inno). In section 4.5 we conclude our analysis by investigating also some the micro-dynamics of the model. We consider a regression analysis at the level of firm to identify which characteristics impact on the firms' chances of survival, their size and the extent of diversification across submarkets. This analysis is crucial for testing the relevance of the mechanisms just discussed in Section 4.1, i.e. the presence of a substantial advantage for early entrants, the role of the "size of the jackpot" effect and the linkage between innovative strategies and market leadership. On the existence of these mechanisms we base large part of our interpretation of simulation results.

4.3

Data

The statistics relative to the variables Yt, Z, and Xt are reported in Table 1. The upper and lower bounds of the distributions of parameters are set with coherence: negative lower bound of the patent duration distribution does not make any sense, as well a negative level of opportunities, size of the market or the number of possible entrants. Similarly, we set reasonable upper bounds. In addition to the parameters defining the technological regime (Opportunity, Patent Duration (PD) and Cumulativeness), we also assess the effect of variation of the number of potential entrants (Firms), the size of the market in terms of potential patients (Market size), market fragmentation (nTC), and a measure of the intensity of price competition (i.e. price elasticity of demand, ε, in the mark-up, section 3.2). We also control for a set of time dependent variables defined in order to describe the state of the system in t — 1 (X f _i ) in each period of simulation: the number of innovative firms (i.e. firms whose profits are mainly originated by innovative products) (N. of innovative firmst), the number of imitative firms (i.e. firms whose profits are mainly originated by imitative products) (N. of imitative firmst), the average R & D propensity of active firms, the count of changes of market leadership at time t since the begin of the simulation

I n n o v a t i o n a n d M a r k e t S t r u c t u r e in P h a r m a c e u t i c a l s

l/l 4) ΐ % Ë i

S S

I

I

I

o o

o o

I

I

l

O ΓΝ O ΓΜ

s α

(N ID m ΓΜ (Y! r i N m q m Ò O O fN ^f ^ τ-

c rt 01 s

T - i o o t m τ- in in

ι

ι

S fe

o ο ο

m ο γμ o ο ò

τo ο

m γμ πη σ\ 5 γμ in ο ο ó γμ ò ιη γμ τοο

in r^ σ\ οο οο νο οο ai c > > .Η ρ ' ε ·>

where p¡ = Χ , / Σ ^ Χ ^ is the relative demand for the i'th good. A successful tax system would lead to a post-tax mean impact equal to the target level i.e. ε = η. We offer the following remarks on a successful tax policy. 1. A necessary condition for success is that the target η lie between the lowest impact min¡ ε, and the highest impact max, ε,. That is to say the available impacts must span the target. In particular there must be goods with an impact less than the target. 2. Assuming existing impacts span the target there must be sufficient substitutability to allow relative price changes to engender the requisite shift in relative demands. 3. The net tax revenue G from the tax regime is G = Σ,Χ,ί; = Σ,Χ,τ(ε, - η) = τ [Σ,Χ,ε, - ηΈ,Χ,] = τ(ε -

η)Σ,Χ,.

For a successful regime this revenue would be zero. There would be no surplus to fund public goods other than the perceived benefits of the reduced impact of consumption. 22

See Bansal and Gangopadhyay (2003) for a more sophisticated and complex model.

380 · Chris Birchenhall and Paul Windrum

4. While the budget is balanced (G = 0) the implicit redistribution of resources is likely to be met by political resistance f r o m those consumers and businesses w h o suffer a net loss. As noted by Sandler any attempt to reduce the carbon emissions will have a wide spread impact on consumers and businesses implying the political pressures will be high compared with the issue of CFCs. The t a x w o u l d hit hardest those w h o find it most difficult to substitute t o w a r d low impact goods. 5. Insofar as the proposed t a x policy is seen as part of an international agreement on carbon emissions voters (consumers) perceived "public g o o d " benefit, in the f o r m of the slowing of global warming, will in n o small part reflect their assessment of that global agreement, both in terms of its likely effectiveness and its "fairness". 6. T o delve a little deeper into the politics of this t a x regime we can look at its impact on individual consumer/voter. Using χ^ to be the post t a x d e m a n d of individual k for good i the net t a x payed g^ by individual k will be gk = Σ,-ί,·*? = S f T(e,· -

= T(e* -

η ) ^

where Σ,-ε,·*^ Σ,·*? is the mean impact for individual k. As G = = 0 there will be some w h o have a net subsidy, with g* < 0, and some w h o pay a net tax, with g^ > 0. N o t e we can rewrite G as follows G =

= Σ έ τ ( ε έ - η)Σ,·*^ = τ [vkwk{sk

- η)] Σ,·Χ,·

where W

k

Σ χ >ϊ = —-f Σ/Χ,

is the relative d e m a n d or " w e i g h t " of individual k. G = 0 implies — η) = 0. W e view individuals with a high weight as being " r i c h e r " a n d those will low weight as been " p o o r e r " . The evidence is that the richer tend to have higher impacts then the poorer, see for example Weber and M a t t h e w s (2008). This implies this t a x regime would be progressive. Whether that makes the t a x system more likely to succeed politically will depend on the balance of power. 7. Insofar as consumers are also voters we can ask if each w o u l d benefit f r o m the t a x structure. A positive g* would not necessarily imply the individual being against the tax. The relevant measure is the change in the individual's total utility if the t a x were to be introduced, where total utility includes both the individual's direct utility f r o m consumption and the utility they gain f r o m an anticipated improvement in the environment. This difference in direct utility will not only reflect the net t a x paid but also the individual's ability and willingness to substitute away f r o m high impact goods to low impact goods. W e can anticipate that the willingness to change will be positively

Global Warming: Technology, Preferences and Policy • 381

correlated with the individual's utility from an improved environment. As to an individual's ability to change we have to suggest this will be positively correlated with wealth. All in all an investigation of the degree of support for such a tax requires an in depth empirical study of the population distributions over all of these relevant characteristics: mean impacts, relative demand, willingness and ability to adjust and the willingness to pay for an improved environment. We are unaware of a thorough empirical study along these lines. Without that evidence we can not make statements about the "interests" associated with the tax policy. At the time of writing the world is more preoccupied with the continued financial fallout from the 2008 crisis; it is not a time to be optimistic that a global agreement on GHGs will be forthcoming as the political pressures point to reducing sovereign debt and voters are preoccupied in defending themselves against the implications of various austerity packages. There is little ground for optimism with regard to a successful global agreement on GHGs in the near future. 8. Just as there will be winners and losers in the population of consumers/voters so there will be winners and losers among producers. In the case of GHGs some of these producers have large stakes in current technology and are powerful lobbyists. All in all, as Sandler has noted an IEA on GHGs is not going to be easy and the fundamental institutional failure remains. Just how these considerations impinge on a nation's policy choice will reflect the polity of that nation. Even in a democracy it is unlikely that a decision would be based on a referendum so medium voter analysis may not be apt. 23 Models of decision making in non-democratic states is underdeveloped. Where do we see a technological shift impinging on this simple story? One of the major impediments to a successful tax policy, and thus a constraint on the win set of implementable targets η, is the degree of substitutability toward low impact goods and from high impact goods. We can view a paradigm substitution working through the introduction of new products and an associated shifting preferences that enhance substitutability. This opens up a longer term strategy with upfront subsidies and planned stepped rise in the tax rate τ allowing consumers and producers time to adjust, in particular to give producers time to innovate. This in turn will expand win sets and enhance the prospect of a global agreement.

9

Evolutionary games and institutions

The previous discussion assumed a set of fixed institutions - national polities and international organisations. At this juncture we can add some notes on the use of evolutionary game theory to investigate institutions. One feature that is of interest in the context of technological innovation and paradigm substitution is that evolutionary games do not require the same degree of common knowledge normally assumed in standard game theory. Our primary reference here is Young (1998). Consider fictitious play in recurrent games. In a recurrent game players are drawn at random from a population; unlike repeated games, where the same players repeatedly play against the same set of players, in recurrent games

23

See Roemer Chapter 58 in Weingast and Wittman (2006: Ch58) for a modern introduction to modelling votes and the fragility of the "median voter theorem".

382 · Chris Birchenhall and Paul Windrum

players are matched randomly and punishment strategies are not relevant. In fictitious play each player can estimate the expected payoff from each possible action; this estimate is based on historical data taken from a finite number of previous plays of the game. It does not require players to know the payoffs for other roles; what is needed is an estimate of relative number of times an action has been chosen. In this sense there is no need for common knowledge of payoffs. Given the expected payoffs each player chooses a best response, an action that maximises the expected payoff. Young's adaptive play combines fictitious play with a error process - there is a small positive probability, called the error rate, that a player does not play a best response but chooses an action at random. With an error rate of zero adaptive play can become "locked in" to a strict Nash equilibrium. For example consider the Stag Hunt game. If the historical data shows everyone has played Hunt Hare then the best response is Hunt Hare; in the next play the data continues to show everyone playing Hunt Hare and the process has been absorbed by the Hunt Hare equilibrium. Such histories of play Young calls a convention. It is to be stressed that a convention is a history of play not a Nash equilibrium. Young (1998: Ch4) provides analytical results on small games such as that in Figure 1. One result Young (1998: Theorem 4.1) tells us that if the game is a coordination game (a > d, a > c and d > b), sampling of history is "low" and the error rate is zero then with probability one the process converges to and locks into a convention. In the Stag Hunt game that may imply everyone hunts Stag or everyone hunts Hare. The same theorem offers a result for coordination games with a risk dominant strategy such as Hunt Hare in the Stag Hunt game i.e. c + d > a + b. With choices being based on large samples and a positive error rate the stochastically stable states are the risk dominant conventions. This would imply in the Stag Hunt game the process with almost always be in a Hunt Hare convention. While it is possible for the process to jump between conventions the probability of jumping into a risk dominant convention is higher than the probability of jumping out of such conventions and thus over the long run it would almost always be in risk dominant conventions. As we have noted before Sugden (2004, Afterwood) took a similar position by suggesting the risk dominant strategy was evolutionary stable. Skyrms (2004) looks at variations in the rules of the game that allow the efficient equilibrium (Hunt Stag) to be maintained. How does this relate to our discussion of paradigm substitutions? Let us interpret the game in Figure 1 as follows. A player who opts for BAU sticks with the old life style and technology, a player who opts for Abate is choosing a new low impact life style and technology. The population's choices are evolving through adaptive play. If the game is a coordination game and BAU is risk dominant then Young's results suggests the population will almost always get "locked in" to a BAU convention. The Pareto dominance of Abate will not prevail. While a preferred paradigm substitution seems to be available in the Abate equilibrium continued shocks (errors) will make it difficult to maintain. This suggests that a paradigm substitution needs to be shock-proof. Here we point to one possible source of such shock-proof paradigm substitutions. As in standard game theory, Young assumes payoffs are fixed. Yet we know that the real costs of production of goods tends to fall as experience with the good increases. Imagine the population seemingly locked into BAU but experiences "errors" that are biased toward Abate. We can imagine "green warriors" choosing Abate even if this does not seem to be a best response. Such biased actions could be the basis for a niche allowing producers to experiment with low impact goods, that experimentation may lead to lower costs for

Global Warming: Technology, Preferences and Policy • 383

such goods and thus increase the payoffs for Abate. This raises the possibility that BAU looses its risk dominance or even its status as an equilibrium. In either case the Abate convention may be stable and we can have a paradigm substitution. It is to be stressed to develop this prospect we will need to significantly modify Young's model; a task beyond this initial paper. It is to be stressed that this brief discussion of Young's results abstracts from the details of international agreements outlined above and puts focus on the underlying issue of shifting lifestyles toward an Abate form.

10

Paradigms shifts and international polity

T h e Copenhagen Accord did not involve agreed emission targets. Rather developed nations would commit to reductions and provide financial support to Nationally Appropriate Mitigation Actions (NAMAs) submitted by developing nations. N A M A s moved the emphasis " f r o m marginal emissions to low-carbon development" Neuhoff ( 2 0 1 1 : 1). The developed country emission targets and N A M A s were to be submitted by individual nations. It is clear the bargain set at Copenhagen was ill-defined and severely limited. While Neuhoff seems to think this Accord has solved the free rider problem the final conclusion is that "international co-operation will be essential" Neuhoff ( 2 0 1 1 : 2 4 7 ) . Victor ( 2 0 1 1 ) starts by questioning the fundamental approach underlying the search for "universal, legally binding agreements that national governments implement back at h o m e " . Victor is also sceptical about a simplistic "engineer's m y t h " that establishing new paradigms is quick and cheap. Credible policies that address the fundamental risks of investing in new technologies are not readily identifiable. Much of Victor's discussion is consistent with our framework. W h a t is of particular interest here is the suggestion that the focus should be on Olson-like small groups or Buchanan-like " c l u b s " of nations working on flexible nonbinding agreements and the suggestion that G A T T and W T O offer workable models. A reading of Victor's final chapter is readily translated into bargain set terminology. Removing states from negotiations will reduce the number of win sets and can lead to an expansion of the bargain set. Expanding the set of instruments used in formulating agreements, for example going beyond simple emission targets, can also expand the bargain set. Selective grouping could enhance such expansion of bargain sets. Victor is suggesting a paradigm shift in the nature of international negotiations over G H G s . Any global mitigation policy will need to emerge from the ground up with small groups more likely to find common ground. At this juncture it is apt to mention the work of Elinor Ostrom, see for example Ostrom ( 1 9 9 0 , 2 0 0 5 ) . This work is relevant as Ostrom focuses on institutional solutions to common-pool resources (CPRs) where market and state solutions have failed. Ostrom explicitly counters those arguments, based on Hardin and Olson, which look to the use o f coercion or market orientated property rights to successfully manage CPRs, see Ostrom ( 1 9 9 0 : C h i ) . 2 4 W e are unaware of a clear statement how Ostrom's framework sheds

24

Ostrom is aware of Olson's discussion of small or intermediary groups. Her critique is aimed at those who use Olson's work to argue coercion is necessary in the management of CPRs.

384 · Chris Birchenhall and Paul Windrum

light on how to design international institutions for G H G mitigation. Where Ostrom is most relevant is in the discussion of local institutions aimed at sustaining "green" lifestyles.25 11

Conclusion

In this paper we have outlined a framework for the coevolutionary analysis of production, lifestyles and international political policies, with a focus on the mitigation of G H G emissions. At the core of this framework is what we have called the bargain set in international agreements. This term encapsulates the central analytical tool of Putman's twostage game. Essentially negotiations are constrained to agreements in the bargain set. The Copenhagen Accord's focus on nationally defined actions suggests the bargain set for agreed GHG mitigation targets is currently empty. The issue then is how to expand the bargain set. Our recurrent theme has been the potential role of technological paradigm shifts in this expansion. Each nation entering into international negotiations will have its own win set·, namely, the set of potential agreements that its domestic polity will ratify. An agreement is in the bargain set if it is in all nations' win sets. Consider negotiations regarding an emission profile that defines an emission target for each nation. Here the bargain set will be the set of emission profiles which will be ratified by all nations. Just which profiles a nation will ratify largely depends on what domestic targets are feasible within that nation's polity. It will also depend on that nation's assessment of the whole profile. Each nation's set of feasible targets will reflect national production, lifestyles and political structure. Here is where a coevolutionary analysis of production, lifestyles and policies comes into play. As firms, households and governments interact so technologies, lifestyles and policies will coevolve. Paradigm shifts involve policy shifts as well as changes in technology and lifestyles. In turn, win and bargain sets will evolve. We have presented a simple tax-subsidy analysis to highlight the issues of technological spanning, substitutability and political feasibility. For a nation to successfully meet an emission target there must be a mix of available products which would give rise to that target, i.e. the available emission profile of goods must span the target. If this technological condition is satisfied, a tax-subsidy policy, which biases relative prices in favour of low-carbon products, can only succeed if there is sufficient substitutability between high carbon and low carbon alternatives. Finally, given a sufficient substitutability exists, a successful tax-subsidy policy must be politically feasible. Substitutability and political feasibility will be enhanced if there are sufficient households who are "green warriors" who commit to "green" lifestyles. These "green warriors" provide a niche market for firms to pursue low-carbon innovation. To further develop the analysis of win sets will require a number of case studies that focus on low carbon innovation within the context of national production, lifestyles and polity. 25

See her interview with Der Spiegel at the Copenhagen meeting, Ostrom (2009). "One treaty will not solve the problem entirely. This is why I propose a so-called polycentric approach to tackling climate change. We need all levels of human society to work on this to be effective in the long run. Cities, villages, communities and networks of people have been neglected as players. ... we need to take action on smaller levels. If the politicians do not agree in Copenhagen, I would like to embarrass the hell out of them by getting some agreements going where people are doing something - essentially saying: 'We are tired of waiting for you.' The city of Freiburg is a very good place to see what that actually means."

Global Warming: Technology, Preferences and Policy · 385

Each of the key nations have very different profiles across the three dimensions and the case studies should not impose any simple model but be designed to inform systematic analysis of the core issues. 26 These national case studies could include, or be enhanced by, investigations of local institutions that help sustain low-carbon lifestyles. Beyond national win sets further work is required to consider alternative forms of international agreements targeting global warming. We have illustrated how bargain set analysis can help organise thoughts by looking at the suggestions of Victor (2011 ) to move beyond emission targets and focus on small groups of nations. The observation that governments can only deliver policies, and not emission targets, seems a good starting point. Moving toward agreements on national policy mixes and away from narrow emission targets can only expand bargain sets.

References Abernathy, W.J., J.M. Utterback (1975), A dynamic model of product and process innovation. Omega 3: 639-656. Acemoglu, D., J.A. Robinson (2012), Why Nations Fail: The Origins of Power, Prosperity and Poverty. Profile Books. Atkinson, A.B., J.E. Stiglitz (1969), A new view of technological change. Economic Journal 79: 573-578. Axelrod, R. (1990 [1984]), The Evolution of Co-operation. Penguin Books (with Foreword by Richard Dawkins). Bansal, S., S. Gangopadhyay (2003), Tax/subsidy policies in the presence of environmentally friendly aware consumers. Journal of Environmental Economics and Management 45: 333-355. Barrett, S. (2003), Environment 8c Statecraft: The Strategy of Environmental Treaty-Making. Oxford University Press. Beckenbach, F., R. Briegel, R. (2010), Multi-agent modeling of economic innovation dynamics and its implications for analyzing emission impacts. International Economics and Economic Policy 7: 317-341. Binmore, K. (2005), Natural Justice. Oxford UP. Brown, J., C. Hendry (2009), Public demonstration projects and field trials: Accelerating commercialisation of sustainable technology in solar photovoltaics. Energy Policy 37:2560-2573. Cooper, R.W. (1999), Coordination Games: Complementarities and Macroeconomics. Cambridge UP. Crick, B. (2002), Democracy: A Very Short Introduction. Oxford UP. Dasgupta, P. (2008), Discounting Climate Change. Journal of Risk and Uncertainty 37(2-3): 141-169. David, P.A. (1975), Technical Choice, Innovation and Economic Growth. Essays on American and British Experience in the Nineteenth Century. Cambridge: Cambridge University Press. Desmarchelier, B., F. Djellal, F. Gallouj (2013), Environmental policies and ecoinnovations by service firms: An agent-based model. Technological Forecasting and Social Change. In press. DOI: TFS-17684. Dijk, M., M. Yarime (2010), The emergence of hybrid-electric cars: Innovation path creation through coevolution of supply and demand. Technological Forecasting and Social Change 77: 1371-1390. Dixit, Α., S. Skeath (1999), Games of Strategy. Norton. Dixit, A. (2004), Lawlessness and Economics: Alternative Models of Governance, Princeton UP.

26

Following on from footnote 8 any further study should consider systematically applying the idea of interest and learn lessons from the literature on Open Economy Politics.

386 • Chris Birchenhall and Paul Windrum

Dosi, G. (1982), Technological paradigms and technological trajectories. Research Policy 11: 147-162. Dutta, P.K., R. Radner (2004), Self-enforcing climate-change treaties. Proceedings of the National Academy of Sciences of the United States of America, April 6. Evans, P.B., H.K. Jacobson, R.D. Putnam (1993), Double-Edged Diplomacy: International Bargaining and Domestic Politics. University of California Press. Fenby, J. (2008), The Penguin History of China. Penguin. Finus, M. (2001), Game Theory and International Environmental Cooperation. Edward Elgar. Foxon, T.J (2003), Inducing Innovation for a low-carbon future: drivers, barriers and policies. A Report for the Carbon Trust. Available under publications at: www.carbontrust.co.uk. Freeman, C. (1982), The Economics of Industrial Innovation. 2nd ed., London: Pinter. Geerlings, H., K.M. Gwilliam (1994), N e w technologies and their potential to reduce the environmental impact of transportation. Transportation Research 28(4): 307-319. Greif, A. (2006), Institutions and the Path to the Modern Economy: Lessons from Medieval Trade. Cambridge UP. Held, D. (2006), Models of Democracy. Third edition, Polity Press. Heckbert, S., T. Baynes, A. Reeson (2010), Agent-based modeling in ecological economics. Annals of The New York Academy of Sciences 1185(1): 39-53. Kuhn, Th.S. (1962), The Structure of Scientific Revolutions. University of Chicago Press. Mayer, C. (2013), Firm Commitment: Why the corporation is failing us and how to restore trust. Oxford UP. McGregor, R. (2010), The Party: The Secret World of China's Communist Rulers. Penguin. Mitter, R. (2004), A Bitter Revolution: China's Struggle with the Modern World. Oxford UP. Nelson, R.R., S.G. Winter (1977), In search of useful theory of innovation. Research Policy 6: 36-76. Neuhoff, Κ. (2011), Climate Policy after Copenhagen: The Role of Carbon Pricing. Cambridge UP. Nordhaus, W.D. (2007), A Review of the Stern Review on the Economics of Global Warming. Journal of Economic Literature 45: 686-702. Nordhaus, W.D. (2010), Economic Aspects of global warming in a post-Copenhagen environment. Proceedings of the National Academy of Sciences of the United States of America, June 14. Olson, M. (1971 [1965]), Logic of Collective Action: Public Goods and the Theory of Groups. Harvard UP. Oltra, V., M. Saint Jean (2005), The dynamics of environmental innovations: three stylised trajectories of clean technology. Economics of Innovation and N e w Technology 14(3): 189-212. Osborne, M.J., A. Rubinstein (1994), A Course in Game Theory. MIT Press. Ostrom, E. (1990), Governing the Commons. Cambridge UP. Ostrom, E. (2005), Understanding Institutional Diversity. Princeton UP. Ostrom, E. (2009), Interview with Der Spiegel. Search: www.spiegel.de/international/ (last seen March 10 2013). Putnam, R.D. (1988), Diplomacy and domestic politics: the logic of two-level games. International Organization 42: 427-460. Rosenberg, N . (1969), The direction of technical change: inducement mechanisms and focussing devices. Economic Development and Cultural Change 1 8 : 1 - 2 4 (reprinted 1976 in: N . Rosenberg, Perspectives on Technology, Cambridge: Cambridge University Press). Sandler, T. (2004), Global Collective Action. Cambridge UP. Safarzynska, K., J.C.J.M. van den Bergh (2010), Demand-supply coevolution with multiple increasing returns: Policy analysis for unlocking and system transitions. Technological Forecasting and Social Change 77: 297-317. Schelling, Th.C. (1960), The Strategy of Conict. Harvard UP. Skyrms, B. (2004), The Stag Hunt and the Evolution of Social Structure. Cambridge UP. Stern, N . (2007), The Economics of Climate Change. Cambridge UP. Also available online at the UK's Treasury website.

Global Warming: Technology, Preferences and Policy · 387

Sugden, R. (2004), The Economics of Rights, Cooperation and Welfare. Second Edition, Palgrave. Vernon, R. (1966), International investment and international trade in the product life cycle. Quarterly Journal of Economics 80: 1 9 0 - 2 0 7 . Victor, D. (2011), Global Warming Gridlock: Creating More effective Strategies for Protecting the Planet. Cambridge UP. Vogel, E. (2011), Deng Xiaoping and the Transformation of China. The Belknap Press of Harvard UP. Watts, J. (2010), When a Billion Chinese Jump: Voice from the Front Line of Climate Change. Faber and Faber. Weber, Ch.L., H.S. Matthews (2008), Quantifying the global and distributional of American household carbon footprint. Ecological Economics 66: 3 7 9 - 3 9 1 . Weingast, B.R., D.A. Wittman (eds.) (2006), The Oxford Handbook of Political Economy. Oxford UP. Windrum, P., C. Birchenhall (1998), Is life cycle theory a special case?: dominant designs and the emergence of market niches through co-evolutionary learning. Structural Change and Economic Dynamics 9: 1 0 9 - 1 3 4 . Windrum, P., C. Birchenhall (2005), Structural change in the presence of network externalities: a co-evolutionary model of technological successions. Journal of Evolutionary Economics 15(2): 1 2 3 - 1 4 8 . Windrum, P., T. Ciarli, C. Birchenhall (2009), Consumer heterogeneity and the development of environmentally friendly technologies. Technological Forecasting and Social Change 76: 533-551. Young, H.P. (1998), Individual Strategy and Social Structure: An Evolutionary Theory of Institutions. Princeton UP. Corresponding author: Chris Birchenhall, Honorary Fellow, Economics, Arthur Lewis Building2.016, School of Social Science, University of Manchester, Manchester M 1 3 9PL, UK. [email protected] Dr. Paul Windrum, Associate Professor of Strategy, Nottingham University Business School, Jubilee Campus, Nottingham NG8 IBB, UK. [email protected]

Jahrbücherf. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2014) Bd. (Vol.) 234/2+3

Naturalizing Institutions: Evolutionary Principles and Application on the Case of Money Carsten Herrmann-Pillath* Frankfurt School of Finance and Management JEL B52; D02; D87; E40; Z1

Generalized Darwinism; institutions; replicator/interactor; Searle; Aoki; naturalism; memes; emotions; money.

Summary In recent extensions of Darwinism into economics, the replicator-interactor duality looms large. I propose a naturalistic approach to this duality in the context of the theory of institutions, which means that its use is necessarily dependent on identifying a physical realization. I introduce a general framework, which synthesizes Searle's and Aoki's theories, especially with regard to the role of public representations (signs) in the coordination of actions, and the function of cognitive processes that underlie rule-following as a behavioural disposition. Institutions are causal circuits that connect the population-level dynamics of interactions with cognitive phenomena on the individual level which ultimately root in neuronal structures. I propose a new conceptualization of the replicator in the context of institutions: the replicator is a causal conjunction between (physical) signs and neuronal structures which undergirds the dispositions that generate rule-following actions. Signs, in turn, are outcomes of population-level interactions. I apply this framework on the case of money, analysing the emotions that go along with the use of money, and presenting a stylized account of the emergence of money in terms of the naturalized Searle-Aoki model. In this view, money is a neuronally anchored sign for emotions relating with social exchange and reciprocity. The money replicator is physically realized in a causal conjunction of money artefacts and money emotions.

1

The naturalistic turn in the evolutionary approach to institutions

O n e o f the m a j o r challenges in generalizing the t h e o r y o f evolution is t o include h u m a n culture a n d institutions i n t o the picture ( M e s o u d i et al. 2 0 0 6 ; M e s o u d i 2 0 1 1 ) . In eco-

* I dedicate this paper to Ulrich Witt. Professor Witt has been my mentor in evolutionary economics since my times as a PhD student, though not under his supervision. He always gave support to my academic advancement in critical situations. His idea of 'ontological continuity' continually served as a beacon for my own research, which does not mean that we agree on all details, however. What I appreciate most is his openness to fresh and sometimes apparently outlandish ideas, which I am especially fond of, sometimes overstretching my points. The ideas in this paper matured over half a decade, and I hope that the final result meets with his high standards of research on evolution and the economy.

Naturalizing Institutions • 389

nomics, this research agenda was launched by Thorstein Vehlen (1899) for the first time, but was lost out of sight for most of the 2 0 t h century. Outside economics, the co-evolution of human biology and culture has received considerable attention in anthropology and biology after sociobiology had attacked the foundations of the social sciences and humanities as independent research traditions. Today, diverse approaches to gene-culture evolution are at hand which avoid fully-fledged reductionism but also extend the evolutionary concepts into the realm of culture (e.g. Richerson/Boyd 2005; Jablonka/Lamb 2006); the concept of 'inclusive inheritance' is emerging as a unifying framework (Danchin et al. 2011). In economics, unified approaches have only been brought back on the research agenda with the recent program of 'Generalized Darwinism' (Hodgson 2002; Aldrich et al. 2008; Hodgson/Knudsen 2010). One conceptual problem in all these extensions is the question of ontology, in the specific sense of social ontology. Social ontology defines the major ontological difference between 'old' and 'new' institutionalisms in economics: In approaches of the former, seminally launched by Veblen, amongst others, institutions are treated as constituent units of social reality, whereas new institutionalisms mostly follow the standard assumption of methodological individualism in economics, which would only treat 'individuals' as 'real' units of larger social systems (for an overview, see Hodgson 1999). For evolutionary approaches and Darwinism in particular, this disjunction applies as well, in the context of the tensions between claims of genetic reductionism and the possible role of alternative approaches which would highlight the role of higher-level units in evolution. I reduce these complexities to one question: Can we construct an extension of evolutionary theory that treats institutions as units of evolution, alongside with genes as units of biological evolution? H o w can we elaborate on the general hypothesis on the 'ontological continuity' between different levels of evolution (Witt 2003)? In this paper, I present an argument in favour of treating institutions as 'real' and as units of evolution on an ontological level which is independent from the genetic level. As such, the paper picks up a distinction which is currently seen as being obsolete by the vast majority of researchers, namely the distinction between genes and 'memes', which was posited in one early universalization of Darwinism by Dawkins (1989) (for a collection of viewpoints, see Aunger 2000). The problems in fixing the concept of the 'meme' are just special examples of the troubles with generalizations of another Dawkins concept that underlies the notion of meme, i.e. the 'replicator' (Hull/Wilkins 2005), and which has been put at the centre of the efforts of Knudsen and Hodgson in generalizing the Darwinian paradigm to economics (Knudsen 2002; Hodgson/Knudsen 2006). In particular, what is the physical realization of replicators in the institutional domain, beyond chemical mechanisms such as genetics? I will put together different theoretical resources from different disciplines to offer my solution to this quandary. The first starts out from Aunger's (2002) theory of (neuro)memes. Aunger, too, posits that one of the intricate questions of the generalization of Darwinism is the distinction between replicators and interactors, i.e. the generic conceptual counterpart to the genotype / phenotype dualism. Replicators would define the information accumulation, transmission and retention function; interactors would define the functions of this information relative to selective environments. Although it is possible to make sense of this distinction in purely information-theoretic terms (as in Hodgson/Knudsen 2010, following related views such as Dennett 1995), this approach is methodologically problematic because it implicitly gives up the naturalistic ontology underlying Darwinism. In fact, the purely information-theoretic approach is a disguised

390 · Carsten Herrmann-Pillath

Cartesian substance dualism of mind vs. matter which builds the universalization of the theory on the distinction between a material domain, where biology reigns, and an abstract domain of 'information', in which the generalization holds (for a pertinent discussion on the notion of information in biology, see M a y n a r d Smith 2 0 0 0 vs. Griffiths 2 0 0 1 ; also compare O y a m a 2001). Instead of this, I present an entirely naturalistic account of institutions (for a related view, see Sperber 2011). I define naturalism as the ontological hypothesis that the world is physically closed in causal terms, thus eschewing any sort of substance dualism, a n d that therefore 'existence' is defined in terms of physical causal powers (Papineau 2 0 0 7 , 2009). Naturalism does not preclude the possibility of emergence, i.e. ontological novelty (Bunge 1977/1979); that means, I propose a nonreductionist evolutionary account of institutions (in the sense of H o d g s o n 1999). M y naturalistic approach focuses on the causal circuitry between institutional artefacts and neuronal structures as the physical realization of replicator functions. Then, one central question is h o w we can understand the causality between artefacts and neuronal states. M y solution to this problem is to synthesize the categories of 'meaning' and 'function' in an evolutionary account of institutions. This synthesis starts out f r o m the recent 'cognitive t u r n ' in institutional economics (e.g. N o r t h 2005), which sees institutions as combinations of incentive mechanisms and cognitive schemes (mental models etc.). I present a detailed proposal on h o w to conceptualize replicators in the context of cognitive theories of institutions, taking Aoki's (2007, 2 0 1 0 , 2011) theory as a workhorse (building on my extension in Herrmann-Pillath 2012a). Aoki has shown that for the emergence and sustainability of institutions, a specific kind of causal circuitry between external artefacts (his 'public representations') and states of individuals (his 'beliefs') is essential. This causal circuitry mediates between individual-level and population-level processes. I show that this view can be translated into purely naturalistic terms (compare Sperber 2011). As a side effect, my argument points t o w a r d a lacuna in recent debates in evolutionary economics, namely the integration of the brain sciences and recent progress in neuroeconomics. So, I propose an extension of the cognitive a p p r o a c h to institutions to include a neurocognitive f o u n d a t i o n , following recent theories a b o u t 'grounded cognition' a n d related approaches which emphasize the essential role of externalized actions in enabling cognitive formations (e.g. Barsalou 1999, Pecher/Zwaan 2005). This argument basically follows the example of H a y e k , w h o had put the analysis of the brain at the centre of his entire a p p r o a c h to institutions, starting out f r o m his seminal 'Sensory O r d e r ' (Hayek 1952). As a result, I argue that the evolution of h u m a n institutions takes place at the interface of t w o levels of evolutionary processes, namely the evolution of states of the brain (Neural Darwinism, as launched by Edelman 1987) and the evolution of signs (carriers of information) in the most general sense. These t w o processes connect with the process of genetic evolution via epigenetic mechanisms and the phylogenetic heritage of value functions that guide h u m a n learning in an institutional context. Another central idea of the Hayekian a p p r o a c h to institutions that can also be detailed analytically within Aoki's f r a m e w o r k is that institutions are media of distributed knowledge. Aoki analyses institutions in terms of a specific causal circuitry that relates interactions under institutions with external sets of public representations (which I call 'signs'), which have the essential function of 'information compression'. This idea matches exactly with the information-theoretic interpretation of the replicator / interactor duality. In Aoki's conceptualization, the signs have the function of information compression, and they generate certain dispositions that result into actions which reproduce certain behavioural regularities as well as the public representations. Thus, we get an empirical

Naturalizing Institutions • 3 9 1

interpretation of the general replicator function in the context of a fully-fledged evolutionary approach to institutions. The replicator is a conjunction of signs and neuronal states, and the interactor is the resulting behaviour, however in terms of its aggregate population level patterns, i.e. the 'states of play' in Aoki's sense. This analysis catches the important fact that both the interactor and the replicator must be population-level phenomena, such as in the classical distinction between the genotype and the phenotype (Lewontin 2007). Even though the neuronal states are strictly individual, the signs are population-level phenomena, and their functional relation depends on the sustainability of collective behavioural patterns in the population of agents. I call these patterns 'institutionally guided behavioural patterns' IGBP; thus, I highlight that behaviour is always individual behaviour in ontological terms, yet manifests emergent collective properties because of the existence of institutions. Thus, in summary, I present a Darwinian account of institutions that interprets institutions in terms of the replicator-interactor duality. The interactor is the pattern of sustainable behavioural regularities on the population level which manifest functions relative to a selective environment. The replicator is a stable causal conjunction of signs, which are generated on the population level, and neuronal states on the individual level. The replicator connects signs and individual behavioural dispositions, such that the meaning of the sign is the function that it has in sustaining the population-level patterns. I summarize this basic structure in Figure 1.

REPLICATOR Distribution of neuronal states in a population

Distribution of signs / artefacts in a population

f

Ν INTERACTOR

V

Institutionally guided behavioural patterns

>

ENVIRONMENT Figure 1 Replicator and interactor in the evolutionary approach t o institutions

The paper proceeds as follows. In section two, I present a detailed account of the theory sketched so far. In section three, I apply this theory to the institution of money, putting together three different sources of insights: first, the empirical record of the role of emotions in the societal use of money; second, a specific proposal by Lea and Webley (2006) about the core emotion that undergirds the use of money, which I interpret in Darwinian

392 · Carsten Herrmann-Pillath terms, namely the human instinct of social exchange; and third, a conceptual model of the historical emergence of coins presented by Hutter (1994), which I analyse in terms of my generic neurocognitive model. Section four summarizes the argument in terms of stating a general replicator / interactor structure for institutional analysis and glimpses at the larger research agenda of a naturalistic theory of institutions.

2

Institutions, distributed cognition and neuromemes: Outline of a naturalistic approach to institutions

In this section I develop the theoretical framework in more detail. I will relate different theoretical resources, and I present my own interpretations of these contributions. This is especially true for the pivotal theory, Searle's theory of institutions. I will introduce many of Searle's concepts, but impose a strictly naturalistic interpretation. Searle himself is arguing in naturalistic terms (e.g. Searle 2004), but also maintains what I call a 'mentalistic' approach, or, in other words, internalism with regard to mind. Especially, I focus on one notion that has been retreating in Searle's own work recently (e.g. 2010), which is the 'background'. The background is a set of enabling capacities of agents, which makes rule following possible. In his earlier (1995) work Searle argues that the background generates behavioural dispositions (which I distinguish sharply from actions or intentions). This idea I will relate with another theory of institutions in economics, which I find congenial to Searle's approach, namely Aoki's, especially in its most recent versions. For Aoki's approach, too, I develop a naturalistic account, and I will show that the linkage between the two theories rests on the notion of a functional causal circuitry, mediated via language (more generally, sign systems) and dispositions, such that institutions emerge as population-level regularities in individual behavioural patterns. 2.1

Searle: Institutional facts and functions

To begin with, Searle (1995: 129ff.; compare Searle 2004) argues that institutionalized behaviour builds on behavioural dispositions, which are neurophysiologically anchored. 'Following a rule' does not require knowing the rule as such (in the 'knowing that' sense), so there is no need for a fully-fledged mental representation (just 'knowing how' is enough). It suffices to be able to process environmental cues which trigger neurophysiological reactions that produce the required behaviour. Thus, in this view institutions are not fully reflected in cognitive models, but in complex conjunctions of partial cognitive representations and neurophysiological dispositions. This viewpoint seems to be complementary to Aoki's (2001) notion of the stabilization of institutions by summary representations of the underlying game structures. Summary representations are partial cognitive models which do not need to be shared in a population (contra Denzau/North 1994), but form part and parcel of the reproduction of the institution by means of coordinated behaviour, as long as pay-offs stabilize both the different summary representations and the behaviour. In elaborating on this model, Aoki (2007, 2010, 2011) introduces the notion of 'substantive institutions'. This compares with the mentalism of many theories about institutions, especially in game theory. Mentalism approaches institutions as coordinated states of mind between individuals, especially in the sense of mutually confirming expectations, based on common knowledge. To the contrary, substantive institutions are external determinants of mental states, i.e. beliefs. I argue that this approach can be directly connected to Searle's theory of institutions as facts, which is in turn based on

Naturalizing Institutions · 393

a theory of functions. This theory of functions is central to understand the replicatorinteractor relationship in the complete model that I propose: A replicator is a special kind of function relative to the interactor, and the interactor realizes functions relative to the environment. Searle puts his theory in the broader context of a general theory of facts (Searle 1995: 120ff.). He distinguishes between observer independent and observer relative facts (in 2010, he changes the terminology from 'observer' to 'mind', which I do not follow here; for a full treatment of these issues, see Herrmann-Pillath 2013). A metal coin is a piece of metal, which is observer independent. But the function as money is entirely dependent on the observer, hence observer relative. Observer relativity ultimately roots in collective intentionality. This is a crucial step, if we further consider that Searle distinguishes between two kinds of the subjective / objective distinction, i.e. the ontological one, referring to facts about entities, and the epistemological one, referring to judgments about facts (see Table 1). A fact can be ontologically subjective and epistemologically objective, such as in case of a technological artefact, which is observer relative, but the functioning of which follows physical laws. On the other hand, there can be ontologically objective facts which are epistemologically subjective, such as the so-called qualia, i.e. inner perceptions of feelings, which are physical states of the brain, but nevertheless cannot be directly accessed by outside observers. From these distinctions, it becomes clear that institutions are ontologically subjective but epistemologically objective. So, money is a part of an ontology which is observer relative, but its functionings can be analysed by objective epistemic tools, such as the quantity theory of money. Table 1 Types of facts and examples entity

judgement

Ontologically subjective Ontologically objective

Epistemically subjective

Epistemically objective

Subjective fact psychoneural fact (qualia)

Institutional fact Biological fact

There are further important distinctions, especially between agentive and non-agentive functions, and, on the level of institutions, regulatory and constitutive ones. Agentive functions involve intentional agents not only in the ascription of the function, but also in its workings. That is, the function of my heart is non-agentive because its works independently from my intention. If I use money, this function depends on me and all other agents who agree to use money. However, in institutional analysis many functions are also non-agentive, if there are collectively unintended consequences which might be only perceivable to the external observer. Regulatory institutions refer to a pre-existing field of activity, such as institutions governing the exploitation of fish resources; constitutive institutions create the very activity that is governed by the institution, as in the case of a financial market. Money can be regarded as a constitutive institution in the case of modern money, whereas the transition from pure commodity money to coins starts out from regulatory institutions of barter. Further, agentive functions can result into states in which the process of collective intentionality actually retreats into the 'background', such as when coins as money are taken for granted. This implies a shift from agentive to non-agentive functions. Actually, we can state that in the evolution of institutions, the transformation from agentive to non-agentive functions is the essential process in the general phenomenon of institutional scaffolding of individual behaviour (North 2005).

394 • Carsten Herrmann-Pillath

2.2

Aoki: Signs and dispositions in the emergence and reproduction of institutions

This analysis is essential to develop on a radicalization of Aoki's theory (for more detail, see Herrmann-Pillath 2012a). This is because in Aoki's original argument, deployed in 2007 and 2011, there are still some traces of mentalism in two senses. One is that the public representations are seen in terms of a semantics of representation (survey in Lycan 1999), and the other is that they still generate 'beliefs' qua mental states. I propose to substitute this with a functional semantics which follows recent developments in teleosemantics (overview in MacDonald/Papineau 2006 and Neander 2009; compare Millikan 1998, 2005). This means to analyse the public representations exclusively in terms of their causal effects in the causal circuitry of institutions. To indicate this change of perspective, I use the term 'sign' instead of 'public representations', also following recent developments in game theory (e.g. Skyrms 2004, 2010). A sign does not 'represent' a state of play, but has a function, which consists of triggering certain responses by the agents that in turn support those states of play dynamically, which includes the production of those signs. This causal circuitry is the ontological feature that justifies treating institutions as facts in the naturalistic sense. Then, the functions of institutions are partly independent from mental states of the agents, because they are partly non-agentive on the population level. Treating institutions as combinations of agentive with non-agentive functions grasps the Aoki concept of information compression, i.e. there can be no full assignment of functions by any purely internal mental states of individual observers because they are lacking the necessary knowledge of doing that. In other words, following an institution is normally based on a certain understanding of an institution, but also includes many unintended effects which are essential for the functioning of the institution, and which enable the generic function of information compression. This causal circuitry of the revised Aoki model describes an externalist approach to institutions, which allows arguing that via the institutions cognitive functions are externalized on population-level processes. I summarize the revised Aoki model in Figure 2. This keeps the original distinction between the individual (micro) and the aggregate population (macro) level, but changes the original orthogonal distinction between the behavioural and the cognitive level into the two notions of distributed cognition and performativity, which actually connect the individual level with the macro-level (indicated by the diagonal) (for more detail, see Herrmann-Pillath 2012a). I also maintain the idea that there are strategic interactions in populations which can be analysed by different tools familiar from game theory and other approaches in economics and complex systems sciences (top of diagram). These interactions result in states of play which include the generation of signs or sign systems. Signs are artefacts which may be partly produced intentionally, but their functioning in the causal circuitry does not rely on this property (i.e. the function is non-agentive in essence). This is essential to understand the role of signs in processes of distributed cognition. Next, the signs produce causal effects that are mediated via neuronal structures of the agents. Thus, I merge the two notions of function and meaning in the sense that the meaning of the sign is its function relative to the neuronal states of the agents. The neuronal states create dispositions to act, again, without the essential requirement of consciously reflected choices. Dispositions cause actions within a certain range of random variation, which renders the entire model evolutionary in the Darwinian sense. The actions of different agents play together on the population level, producing certain aggregate results, including the reproduction of the sign systems.

Naturalizing Institutions · 395

Population level interaction dynamics PERFORMATIVE FUNCTION strategic actions

generate

recursive states of play S (Β

Stochastic generation of actions Neuronally embodied dispositions