Internationalization, Technological Change and the Theory of the Firm [1 ed.] 9780203846414, 9780415460712

Citation preview

Internationalization, Technological Change and the Theory of the Firm

This book focuses on three main areas, each of which is central to economic theorising: firms’ organisation and behaviour, technological change and the process of globalisation. Each subject can be analysed by using different methods, which range from purely theoretical abstractions to case studies and from econometrics to simulations. What this collection provides is a broad view of the three topics by concentrating on different aspects of each of them, and utilising different methods of investigation. Internationalization, Technological Change and the Theory of the Firm looks in detail at various questions surrounding firms’ organisation, including why we can observe ordered paths of production, whether proximity between firms matters, and whether patenting is always worthwhile. In addition, several essays explore technology and innovation, including the persistence-­cum-development of old technologies. Furthermore, this book focuses on those processes which concern small- and medium-­sized firms, considering the usefulness of stage theory, the possibilities of production offshoring and the skill composition of manufacturing firms. Overall, the book is characterised by original ideas, renewed applications of mathematical and statistical methods and the use of new databases. This valuable collection will be of interest to postgraduates and researchers focusing on innovation, theories of the firm and globalisation; and should also be useful to a professional readership as it presents up-­to-date research with the aim of improving our understanding of the phenomena of technological change, firms’ strategies, and globalisation. Nicola De Liso is Professor of Economics at the Faculty of Law of the University of Salento, Italy. Riccardo Leoncini is Professor of Economics at the Faculty of Law of the University of Bologna, Italy.

Routledge studies in global competition Edited by John Cantwell University of Reading, UK

and David Mowery

University of California, Berkeley, USA.

  1 Japanese Firms in Europe Edited by Frédérique Sachwald   2 Technological Innovation, Multinational Corporations and New International Competitiveness The case of intermediate countries Edited by José Molero   3 Global Competition and the Labour Market Nigel Driffield   4 The Source of Capital Goods Innovation The role of user firms in Japan and Korea Kong-­Rae Lee

  8 Strategy in Emerging Markets Telecommunications establishments in Europe Anders Pehrsson   9 Going Multinational The Korean experience of direct investment Edited by Frédérique Sachwald 10 Multinational Firms and Impacts on Employment, Trade and Technology New perspectives for a new century Edited by Robert E. Lipsey and Jean-­Louis Mucchielli 11 Multinational Firms The global–local dilemma Edited by John H. Dunning and Jean-­Louis Mucchielli

  5 Climates of Global Competition Maria Bengtsson

12 MIT and the Rise of Entrepreneurial Science Henry Etzkowitz

  6 Multinational Enterprises and Technological Spillovers Tommaso Perez

13 Technological Resources and the Logic of Corporate Diversification Brian Silverman

  7 Governance of International Strategic Alliances Technology and transaction costs Joanne E. Oxley

14 The Economics of Innovation, New Technologies and Structural Change Cristiano Antonelli

15 European Union Direct Investment in China Characteristics, challenges and perspectives Daniel Van Den Bulcke, Haiyan Zhang and Maria do Céu Esteves 16 Biotechnology in Comparative Perspective Edited by Gerhard Fuchs 17 Technological Change and Economic Performance Albert L. Link and Donald S. Siegel 18 Multinational Corporations and European Regional Systems of Innovation John Cantwell and Simona Iammarino 19 Knowledge and Innovation in Regional Industry An entrepreneurial coalition Roel Rutten 20 Local Industrial Clusters Existence, emergence and evolution Thomas Brenner 21 The Emerging Industrial Structure of the Wider Europe Edited by Francis McGowen, Slavo Radosevic and Nick Von Tunzelmann 22 Entrepreneurship A new perspective Thomas Grebel 23 Evaluating Public Research Institutions The U.S. Advanced Technology Program’s Intramural Research Initiative Albert N. Link and John T. Scott

24 Location and Competition Edited by Steven Brakman and Harry Garretsen 25 Entrepreneurship and Dynamics in the Knowledge Economy Edited by Charlie Karlsson, Börje Johansson and Roger R. Stough 26 Evolution and Design of Institutions Edited by Christian Schubert and Georg von Wangenheim 27 The Changing Economic Geography of Globalization Reinventing space Edited by Giovanna Vertova 28 Economics of the Firm Analysis, evolution and history Edited by Michael Dietrich 29 Innovation, Technology and Hypercompetition Hans Gottinger 30 Mergers and Acquisitions in Asia A global perspective Roger Y.W. Tang and Ali M. Metwalli 31 Competitiveness of New Industries Institutional framework and learning in information technology in Japan, the U.S and Germany Edited Cornelia Storz and Andreas Moerke 32 Entry and Post-­Entry Performance of Newborn Firms Marco Vivarelli

33 Changes in Regional Firm Founding Activities A theoretical explanation and empirical evidence Dirk Fornahl

42 Evolutionary Economic Geography Location of production and the European Union Miroslav Jovanovic

34 Risk Appraisal and Venture Capital in High Technology New Ventures Gavin C. Reid and Julia A. Smith

43 Broadband Economics Lessons from Japan Takanori Ida

35 Competing for Knowledge Creating, connecting and growing Robert Huggins and Hiro Izushi

44 Targeting Regional Economic Development Edited by Stephan J. Goetz, Steven C. Deller and Thomas R. Harris

36 Corporate Governance, Finance and the Technological Advantage of Nations Andrew Tylecote and Francesca Visintin

45 Innovation, Knowledge and Power in Organizations Theodora Asimakou

37 Dynamic Capabilities Between Firm Organisation and Local Systems of Production Edited by Riccardo Leoncini and Sandro Montresor 38 Localised Technological Change Towards the economics of complexity Cristiano Antonelli 39 Knowledge Economies Innovation, organization and location Wilfred Dolfsma 40 Governance and Innovation Maria Brouwer 41 Public Policy for Regional Development Edited by Jorge Martinez-­Vazquez and François Vaillancourt

46 Creativity, Innovation and the Cultural Economy Edited by Andy C. Pratt and Paul Jeffcutt 47 Co-­opetition Strategy Giovanni Battista Dagnino and Elena Rocco 48 Knowledge Intensive Entrepreneurship and Innovation Systems Evidence from Europe Edited by Franco Malerba 49 Innovation in Complex Social Systems Edited by Petra Ahrweiler 50 Internationalization, Technological Change and the Theory of the Firm Edited by Nicola De Liso and Riccardo Leoncini

Internationalization, Technological Change and the Theory of the Firm Edited by Nicola De Liso and Riccardo Leoncini

First published 2011 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN Simultaneously published in the USA and Canada by Routledge 270 Madison Avenue, New York, NY 10016 Routledge is an imprint of the Taylor & Francis Group, an informa business This edition published in the Taylor & Francis e-Library, 2010. To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk. © 2011 selection and editorial matter; Nicola De Liso and Riccardo Leoncini, individual chapters; the contributors All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data Internationalization, technological change, and the theory of the firm/ edited by Nicola De Liso and Riccardo Leoncini. p. cm. Includes bibliographical references and index. 1. Industrial organization (Economic theory) 2. Business enterprises. 3. Technological innovations–Economic aspects. 4. Globalization– Economic aspects. I. De Liso, Nicola. II. Leoncini, Riccardo, 1959– HD2326.I58 2010 338.5–dc22 ISBN 0-203-84459-9 Master e-book ISBN

ISBN: 978-0-415-46071-2 (hbk) ISBN: 978-0-203-84641-4 (ebk)

2010004776

Contents



List of illustrations List of contributors

  1 Introduction: firms, technology and globalisation

ix xii 1

N icola D e L iso and R iccardo L eoncini

Part I

Technological change, firms’ organisation and incentives

29

  2 The production process as a complex, dynamic and ordered world

31

M auro L ombardi

  3 Incumbents’ strategies for platform competition: shaping the boundaries of creative destruction

66

S tefano B rusoni and R oberto F ontana

  4 Linking technological change to organisational dynamics: some insights from a pseudo-­NK model

89

T ommaso C iarli , R iccardo L eoncini , S andro M ontresor and M arco V alente

  5 Technological persistence through R&D on an old technology: the ‘sailing ship effect’

119

N icola D e L iso and G iovanni F ilatrella

  6 Software patents and firms’ strategic behaviour in the EU: some empirical insights F rancesco R entocchini and G iuditta D e P rato

141

viii   Contents Part II

Fragmentation and internationalisation of firms and of local systems of production

161

  7 Does spatial proximity matter? Micro-­evidence from Italy

163

G iulio C ainelli and C laudio L upi

  8 Internationalization in Italian medium-­sized firms: does stage theory explain the observed patterns?

187

D onato I acobucci and F rancesca S pigarelli

  9 Production offshoring and the skill composition of Italian manufacturing firms: a counterfactual analysis

210

R oberto A ntonietti and D avide A ntonioli

10 A global network and its local ties: restructuring of the Benetton Group

239

P aolo C restanello and G iuseppe T attara



Index

259

Illustrations

Figures 2.1 2.2 2.3 2.4 2.5 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11 4.12 4.13 4.14 5.1a 5.1b 5.2a 5.2b 5.3a

Suh’s Axiomatic Design Theory developed A binary DSM unpartitioned The binary DSM, partitioned to represent a sequence Mappings based on uncorrelated signals GW as a structured landscape Example landscape for N = 2 and K = 1, using normal functions Example landscape for N = 2 and K = 1, using normal functions. Time series of the V values generated by 100 random strategies that modify one xi per time Example landscape for N = 2 and K = 1, using Gaussian functions Distribution of mjs with different B values Evolution of yis Maximisation of yi and convergence of ξj Random walk of yi and non-­convergence of ξj Lock-in of yi and non-­convergence of ξj Model flowchart: one time period Market shares: firms addressing the global fitness Value of the product characteristics: firms addressing the global fitness Search on the landscape and product innovation: unsuccessful Search on the landscape and product innovation: successful Object structure of the model Performance of technologies A and B as a function of time – ordinary differential model Performance of technologies A and B as a function of time – map model Market share based on performance evolution – ordinary differential model Market share based on performance evolution – map model R&D expenditure based on performance evolution – ordinary differential

38 43 44 56 57 97 98 98 101 103 104 104 105 108 110 110 112 113 115 133 133 134 134 136

x   Illustrations 5.3b 6.1

R&D expenditure based on performance evolution – map model Patents accorded by the European Patent Office to EU15 countries (1997–2002) – high-­tech applications 6.2 Patents accorded by the European Patent Office to EU15 countries (1977–2002) – total 6.3 Patents accorded by the European Patent Office to EU15 countries – propensity to patent 6.4 Innovation equation 6.5 Software patents per designated country (1995–2004) 6.6 Patents accorded by the European Patent Office to EU15 countries – software patents 7.1 Estimated densities of the main variables 7.2 Estimated densities of the posterior distributions of the degrees of the freedom parameter 7.3 Estimated intercepts by sector and firm size 7.4 Estimated b2 parameters by sector and firm size 7.5 Estimated b3 parameters by sector and firm size 7.6 Estimated ‘localisation’ and ‘variety’ parameters 9.1 Employment trends for treated and untreated firms 9.2 Propensity score histogram by treatment status 10.1 New collections structure 10.2 Percentage distribution of garments and accessory items produced by Benetton 10.3 The global value chain of the Benetton Group

136 143 143 144 147 152 156 169 175 178 179 180 181 221 228 245 248 249

Tables 3.1 3.2 4.1 4.2 4.3 5.1 5.2 6.1 6.2 6.3 6.4 6.5 7.1 7.2

Organizations in the context of increasing specialization in technology and complexity in artifacts A typology of platforms and industries Research spaces and solvers’ strategies Parameters for Section 4.1.2 Parameters for Section 5.2 Critical values for the Ordinary Differential Equation approach Critical values for the map approach International Patent Classification and International Standard Industrial Classification of All Economic Activities concordance for GAUSS patents Sample statistics (number of filed patents equal to zero) Sample statistics (number of filed patents greater than zero) Software patenting propensity estimates: software and computer services sector (FTSE 97) 2000–2003 Software patenting propensity estimates: information technology hardware (FTSE 93) 2000–2003 Distribution of firms by distance and industry Distribution of firms and employees by sector

68 81 95 102 111 135 136 151 153 153 154 155 167 170

Illustrations   xi 7.3 7.4 7.5

Distribution of firms and employees by geographical area Definition of the variables used in the empirical analysis Estimates and convergence diagnostics of the main parameters of interest 7.6 Posterior predictive distributions and two-­sided p-­values 7.7 Classification of manufacturing activities 8.1 Companies by sector (Pavitt) and size, 2001 8.2 Export intensity (exports on sales) by sector and size 8.3 Value of foreign subsidiaries in total fixed assets 8.4 Average value of foreign subsidiaries by sector and size 8.5 Typologies of internationalization strategy and patterns 8.6 Distribution of companies by type and size 8.7 Transition matrix of types between 2001 and 2005 8.8 Number of foreing subsidiaries and amount of capital invested, by area 8.9 Companies by geographical span of subsidiaries and size, 2001 8.10 Transition matrix in the geographical span of internationalization between 2001 and 2005 9.1 Production options for a firm 9.2a Sample structure by industry and employment classes 9.2b Sample structure by employment classes and geographical area 9.3 Distribution of firms offshoring production by industry, employment size and geographical area 9.4 Testing the difference in the mean propensity score for treated and controls 9.5 The effect of production offshoring on skilled labour: white collar/blue collar 9.6 The effect of production offshoring on skilled labour: white collar 9.7 The effect of production offshoring on unskilled labour: blue collar 9.8 The effect of production offshoring on unskilled labour: blue collar and clerks 9.9 Variables definition 9.10a Summary statistics – merged sample 9.10b Summary statistics – treated sample (offshoring firms) 9.10c Summary statistics – untreated sample (non-­offshoring firms) 10.1 Benetton Group turnover 10.2 Intra-­trade between Benetton’s company Benind and its foreign affiliated companies in 2007 10.3 Number of garments and accessory items produced by Benetton 10.4 Benetton’s Italian subcontractors: number of firms, employees and production items 10.5 Benetton’s Italian subcontracting firms and their employees 10.6 Benetton’s logistical platforms and subcontracting firms in East Europe and Tunisia, 2007

170 172 174 176 182 195 196 197 197 198 199 199 200 201 201 213 218 218 219 223 225 225 226 226 229 230 231 232 246 247 248 249 250 251

Contributors

Nicola De Liso is Professor of Economics at the Faculty of Law of the University of Salento, Lecce and Research Associate at Ceris-­Cnr, Milan. Riccardo Leoncini is Professor of Economics at the Department of Economics of the University of Bologna and Research Associate at Ceris-­Cnr, Milan. Roberto Antonietti is Researcher at the Department of Economics “Marco Fanno” of the University of Padua. Davide Antonioli is Junior Research Fellow at the Department of Economics, Institutions and Territory of the University of Ferrara. Stefano Brusoni is Associate Professor of Economics at Bocconi University, Milan. Giulio Cainelli is Professor of Economics at the Department of Economics ‘Marco Fanno’ of the University of Padova and Research Associate at Ceris-­Cnr, Milan. Tommaso Ciarli is Researcher at the Evolutionary Economics Group, Max Planck Institute of Economics, Jena. Paolo Crestanello is Contract Professor of Economics at the Faculty of Economics of the University of Venice ‘Cà Foscari’. Giuditta De Prato is Grant Holder at the Joint Research Centre of the European Commission, Institute for Prospective Technological Studies, Information Society Unit, Seville. Giovanni Filatrella is Researcher of Experimental Physics at the Department of Biological and Environmental Science of the University of Sannio in Benevento. Roberto Fontana is Researcher of Applied Economics at the Engineering Faculty of the University of Pavia. Donato Iacobucci is Associate Professor of Economics at the Polytechnic University of Marche.

Contributors   xiii Mauro Lombardi is Associate Professor of Economics at the Department of Economics of the University of Florence. Claudio Lupi is Associate Professor of Economic Statistics at the Faculty of Economics of the University of Molise. Sandro Montresor is Associate Professor of Economics at the Department of Economics of the University of Bologna. Francesco Rentocchini is Junior Research Fellow at the Department of Economics of the University of Trento. Francesca Spigarelli is Researcher of Economics at the Department of Law and Economics of the University of Macerata. Giuseppe Tattara is Professor of Economics at the Faculty of Economics of the University of Venice ‘Cà Foscari’. Marco Valente is Researcher of Economics at the Department of Systems and Institutions for the Economy of the University of L’Aquila.

1 Introduction Firms, technology and globalisation Nicola De Liso and Riccardo Leoncini

1  Introduction Firms, technology and globalisation are three central aspects of both economics and the economy. The different way in which the three dimensions are examined by economists on the one hand and other observers, such as entrepreneurs or engineers, on the other hand, is often astonishing. However, whatever strand (Classical, Neoclassical, Institutionalist, Evolutionary, . . .) and definition (à la Mill, à la Marshall, à la Robbins, . . .) of economics we prefer, this science should always have some connection with economic reality, as, after all, economics can be seen as a theory of the economy. We are of course aware of the fact that a trade-­off may exist between generalisations and the details which may be subsumed; however, what is a theory worth based on ideas such as the average firm, which conceals a basic feature of capitalism, that is firms seeking profits which, because of this search for profit, try to differentiate themselves from their rivals? In this regard we have to be careful with the use of words, since the use of seeking is not casual, and is consciously used instead of maximising. Many questions can be raised on the three topics of which the title of this introduction is made up: mainstream economic analysis of technological change has been unsatisfactory for a long time, and endogenous growth theory has mitigated the problems rather than solved them; a proper theory of the firm does not exist; theories which try to explain globalisation have yet to come as too many questions have to be tackled, from finance to international trade, from international mobility of production factors to global production sharing, and more. Despite the criticisms which have just been considered, economics has gone a long way on the road of science: theories and analyses have been developed and quite a few important principles have been identified. The weight of economics can hardly be overestimated given the emphasis that is placed on it, and despite the criticisms which have arisen as a result of the recent crisis not being forecast by “experts”, all governments and their leaders rely on economic advisors whose tools affect overall societies, while relationships between countries are largely affected by economic considerations. The kind of economic ideas that characterise policies implemented by policy makers have an impact at the micro, meso and macro level.

2   N. De Liso and R. Leoncini In this introduction we briefly outline the way in which firms, technology and globalisation have been considered by economists. Thus, in Section 2 we reconsider some aspects of the debate on the way in which economists have considered firms; in Section 3 we review the way in which economists have tackled technology and its change, while in the fourth section we look at globalisation. The fifth section contains the conclusions and illustrates the structure of the volume.

2  The firm 2.1  Theories of the firm In a private enterprise industrial economy the business firm is the basic unit for the organization of production. The greater part of economic activity is channelled through firms. The patterns of economic life, including the patterns of consumption as well as of production, are largely shaped by the multitude of individual decisions made by the businessmen who guide the actions of the business units we call firms. (Penrose 1959: 9) The importance of firms can hardly be questioned: they are central to both the economy and economic theory. In this section we concentrate on the way in which economics has considered firms. When one looks at the firm through the lens of economic theory the impression one gets is that what we are actually using is not a lens, but a kaleidoscope, so that the image is made up of continuously changing patterns. Things are difficult because of the many aspects which characterise firms (ownership, internal organisation, amount of output to be produced, cost structure, employment relationships, obsolescence and investment decisions, where the boundary of the firm lies, knowledge and others) and because any theory has its needs in terms of consistency. Ronald Coase was aware of the fact that the use of the word “firm” in economics may be different from the use of the term by the “plain man” . . . [so that] it is all the more necessary not only that a clear definition of the word “firm” should be given but that its difference from a firm in the “real world”, if it exists, should be made clear. (Coase 1937: 386) When one tries to provide taxonomies, be it by schools of thought, by themes or any other criterion, one ends up with a picture that is never complete and rarely satisfactory. It is not difficult to find attempts aimed at systematising the contributions to the theory of the firm: examples are Foss (2000a), Gibbons (2004) and Ricketts (2008); older reviews, of course, can easily be found (e.g. Boulding 1942). The problem is what one means by a firm, and the consequent dimensions which have to be considered.

Introduction   3 For instance, Foss refers to the theory of the firm as that body of economics that addresses the existence, the boundaries and the internal organisation of the firm (Foss 2000a); Holmström and Tirole besides considering these three aspects explicitly tackle the firm’s capital structure and the role of management (Holmström and Tirole 1989); Cyert and Hedrick examined the existing works from the standpoint of the generation of new knowledge arising from theoretical models, considering four variants of the neoclassical model and non-­maximising models (Cyert and Hedrick 1972). None of the authors ever claim exhaustiveness while controversies – in terms of classification, theoretical content and relevance – always arise. Given the many dimensions of which a firm is made up, the role of the observer is fundamental as he or she can concentrate on one or more different ones, e.g. make-­orbuy, division of labour, learning within production and so on. Results of the analysis can also be affected by sectoral considerations: too often theories have as their background firms producing manufacturing goods. However, the service sector is heavily predominant – tertiary employment in advanced economies takes more than 70 per cent of total employment – and firms belonging to it may sometimes need different tools of analysis. Let us just hint at two specific problems: first, marginal cost is often too close to zero to be a principle according to which pricing can be put in place; second, given that in capitalist economies innovation is a driving force, innovation in services may require service-­specific tools of analysis (Miles 2005). Juridical aspects combine with economic ones: different analyses and results may be obtained when we consider owner-­run small-­sized firms on the one hand, and joint-­stock companies on the other hand – within the latter the divide between ownership and control creates problems in terms of incentives.1 In this section we do not want to review once more the characteristics of the various approaches – one can refer to the works indicated up to now – but simply to point out two critical points that, we believe, deserve more attention than has been previously given to them. The first consists of the need to reconsider the contribution of classical economists to the theory of the firm; the second point has to do with the non-­market relationships that often develop between firms. 2.2  Theory and historical perspective: a reassessment of the classics Many authors emphasise the fact that the theory of the firm has developed without considering real firms. Cyert and Hedrick (1972) wrote that none of the problems of real firms could find a home within the dominant neoclassical model, and that the controversy over the theory of the firm had arisen over a non-­existent entity – the maximising firm which gets information from the market. Despite the fact that this criticism has been partly superseded by the most recent theoretical developments – bounded rationality, principal-­agent, incomplete contracting, etc. – there remains some truth in it. A fundamental problem, in fact, lies in the willingness to have a nineteenth-­ century physics-­like theory made up of principles which can be applied

4   N. De Liso and R. Leoncini everywhere at any time – and the principle of maximisation or the representation of production in terms of a production function lead in this direction. Firms, however, do not lend themselves to be studied in this way. Furthermore, we have to note that it has become a commonplace to say that “history matters”, but one finds little evidence of the fact that this importance has been actually taken into account in economic studies. Neither criticism, though, applies to classical economists. Before considering some of their contributions let us point out that when one reviews the contributions to the theory of the firm it is quite difficult to find analyses concerned with classical economics, and it is not unusual to find comments such as “classical economists did not elaborate such a theory”.2 The first principle relevant for the theory of the firm that we have to consider consists of Adam Smith’s division of labour. He refers to at least two types of division of labour, which have been defined as vertical (or manufacturing division of labour, or intraoccupational differentiation) on the one hand, and horizontal (or social division of labour or occupational differentiation3) on the other hand. As an example of vertical division of labour we can refer to Smith’s original example of pin manufacture. He noted that even such a “trifling manufacture” could be divided into eighteen distinct operations, and that by organising the work process in such a way that one worker performed only one operation overall productivity and production would increase dramatically. He noted that: This great increase of the quantity of work, which, in consequence of the division of labour, the same number of people are capable of performing, is owing to three different circumstances; first to the increase of dexterity in every particular workman; secondly, to the saving of time which is commonly lost in passing from one species of work to another; and lastly, to the invention of a great number of machines which facilitate and abridge labour, and enable one man to do the work of many. (Smith 1776: 17) This is undoubtedly relevant for a theory of the firm and represents a general economic principle in that “so far as it can be introduced, occasions, in every art, a proportionable increase of the productive power of labour” (Smith 1776: 15). Two aspects that are intrinsically connected with the division of labour concern the necessary co-­operation between workers and the role of the entrepreneur (the capitalist). About the first point we can note that: When numerous workers work together side by side in accordance with a plan, whether in the same process, or in different but connected processes, this form of labour is called co-­operation. . . . [T]he effect of the combined labour could either not be produced at all by isolated individual labour, or it could be produced only by great expenditure of time, on a very dwarf-­like

Introduction   5 scale. Not only do we have here an increase in the productive power of the individual, by means of co-­operation, but the creation of a new productive power, which is intrinsically a collective one. (Marx 1867: 443–4, emphasis added) This helps explain – despite Coase’s criticism4 – why firms exist: the division of labour creates a more efficient way of organising production, and in order to arrange the overall process an entrepreneur is needed: The work of directing, superintending and adjusting becomes one of the functions of capital, from the moment that the labour under capital’s control becomes co-­operative. As a specific function of capital, the directing function acquires its own special characteristics (op. cit. p. 449) One more contributor to the debate who must be referred to is Charles Babbage: his book On the Economy of Machinery and Manufactures5 is full of analyses, suggestions and comments on how a profit-­oriented firm has to be set up. He refers to the division of labour, to mechanisation, to the principles on which large factories are established, and more. With regard to the division of labour he pointed out that it could also be applied to mental labour – and to illustrate the principle he refers to the calculation of logarithms (Babbage 1835: 193–5). The main point is that in capitalist industrial manufactures: “the master manufacturer, by dividing the work to be executed into different processes, each requiring different degrees of skill or of force, can purchase exactly that precise quantity of both which is necessary for each process” (op. cit. p. 175, original emphasis). Classical economists also considered side effects such as the deskilling which could take place as a consequence of the division of labour and mechanisation or the emergence of technological unemployment.6 Smith wrote that “The man whose whole life is spent in performing a few simple operations . . . has no occasion to exert his understanding . . . and generally becomes as stupid and ignorant as is possible for a human creature to become” (Smith 1776: 782). To conclude, we can state that although all of the classical authors referred to real firms and to the history of production and technology, this does not mean that they did not reach theoretical generalisations. 2.3  Firms’ boundary, size and non-­market relationships One of the concerns of the theory of the firm consists of the boundary of the firm itself. Once more, when one addresses this issue, answers are not easy to provide. What do we mean by boundary, or boundaries, when we deal with firms? Is it where the work process starts and ends? Or is it that physical or metaphorical area where the entrepreneur lays down his or her rules? Are they legal? Are they technological? What about the interaction between producers (or service providers) and consumers? And what about the relationships between producers?

6   N. De Liso and R. Leoncini In a key article Richardson (1972) points out that while at first sight one is tempted to refer to firms as islands of planned co-­ordination in a sea of market relations, a deeper analysis points to the fact that a dichotomy between firms and the market does not exist. When we consider economic reality we usually observe that a product is made up of different parts which are often produced by different independent firms which, in turn, have to co-­operate in some way. Think, for instance, of the car industry: the car producer does not produce all of the components, and his or her suppliers have to adapt qualitatively and quantitatively to the specifications set by the former. Many examples can be recalled in which two or more firms co-­ordinate their activities both qualitatively and quantitatively: one of the most immediate forms of co-­operation is that of subcontracting, so that one of the parties agrees to conform its output to the needs of the other; other forms of co-­ operation can be found, e.g. when a manufacturer uses an independent retail chain to sell his or her product(s), or when technology transfer takes place.7 The interaction between firms often gives rise to feedbacks which create a virtuous circle in which process and product improvements occur. As Richardson points out, some form of co-­ordination must exist, and this can occur in three ways: by direction (the activities are subject to a single control and fitted into one coherent plan), through market transactions (spontaneously as an indirect consequence of interactions taking place in response to profit opportunity) or by co-­operation: Co-­ordination is achieved through co-­operation when two or more independent organisations agree to match their related plans in advance. The institutional counterparts to this form of co-­ordination are the complex patterns of co-­operation and affiliation which theoretical formulations too often tend to ignore. (Richardson 1972: 890) Co-­operation between independent firms is a phenomenon that characterises capitalist production, and does not represent an exception. As further proof of this, one can think of industrial districts that can be found throughout the world from Italy (e.g. Sassuolo) to Japan (Ota Ward), from the United States (Seattle) to Brazil (Campinas) and South Korea (Pohang). Industrial districts can be considered as a form of division of labour between firms which have a form of co-­ordination through co-­operation. Among the features which characterise industrial districts one usually finds long-­term relationships between district firms and high importance of intra-­district trade. It looks as if there may be problems in identifying both the boundary and the size of the firm. This is true in a “static” environment, in which the number and type of firms is given, and yet we can have situations like this: “If an apple orchard owner contracts with a beekeeper to pollinate his fruits, is the result one firm or two firms? This question has no clear answer” (Cheung 1983, repr. in Foss 2000b: Vol. I, p. 337).

Introduction   7 It is even truer when we have to face a dynamic economic environment where existing firms may grow, stay as they are or decline, and new ones can emerge.8 The decision on whether to make or buy becomes more and more complex, as growth does not mean to produce more of, or more efficiently, the same product or service: growth often implies diversification, a special form of which is backward or forward vertical integration (Penrose 1959:  145).9 In other cases, particularly when we deal with giant firms already existing, it may mean an opposite process of disintegration-­cum-specialisation. The fact remains that over time both the boundaries and the size of firms change. These changes are due to a set of forces (variables) which are sometimes endogenous and other times exogenous to the firm. A tentative, non-­exhaustive, list of these forces includes entrepreneurial ability, managerial ability, knowledge base of the firm, expectations, proportion of borrowed capital and rate of interest, changes in relative prices of products and services, and changes in factors’ rewards.

3  Technology, innovation and knowledge In the last 25 years a form of competition has emerged among social scientists, international institutions and policymakers whereby they emphasise the importance of technology and its knowledge content for the economy and the society as a whole. Economists and economic institutions have been particularly active in this process. But how have economists dealt with technology, innovation and knowledge? As usual, we have to face many difficulties, which begin with the very definition of each of the dimensions which we want to investigate; furthermore, the three dimensions are inextricably interconnected: there is no technology without some form of knowledge, and knowledge contains the seeds of change. In the next three subsections we will keep the three items artificially distinct. 3.1  Technology Technology is a too broad concept to be encompassed into a single acceptable definition; furthermore, as economists, we are not interested in technology per se, but in its characteristics as related to the economy, economic growth and development. When we deal with technology we have to refer to its inner characteristics and to the economic determinants which contribute to shaping it. Furthermore we have to consider the interaction between technology and science, which have become more and more intertwined at least since the early nineteenth century, when the Industrial Revolution was in full swing in Britain and was spreading throughout Europe and America. Despite the fact that technology is interconnected with science, it preserves some form of autonomy from it. Technology focuses on doing and making and it is usually conceived of with a clear objective needing to be achieved, be it to kill a bird with a bow and an arrow or the construction of a digital computer to

8   N. De Liso and R. Leoncini process numbers. As Layton insisted, technology and science are characterised by – at least partly – separate systems of thought, not least because in technology we observe the existence of non-­verbal modes of thought (Layton 1974: 36). Put another way, even in advanced sectors of the economy, in which one would be inclined to think that the basic inputs come from science, we have technology building on technology – an explicit example being the early development of transistors (Gibbons and Johnson 1970). Science is concerned with knowing and with a kind of knowledge which can be true or false. Technology can be looked at as “applicable knowledge . . . [and it] is determined primarily in terms of successful performances to which such knowledge is relevant” (Polanyi 1962: 175). Let us now turn our attention to the way in which economists have looked at technology. Here we selectively refer to the classical, neoclassical-­mainstream and evolutionary schools of thought. Classical economists have definitely been concerned with technology since the very beginning of their intellectual tradition. Smith explicitly connected the division of labour with the invention of machines and acknowledged the role of thinkers – he call them “philosophers or men of speculation” (Smith 1776: 21) – in contriving (new) machinery. Ricardo, besides having added the chapter on machinery in the third edition of his Principles, bases his theorem of comparative advantage on different labour productivities which, in turn, depend on the technologies being used. Charles Babbage and Karl Marx dedicated a lot of attention to technology and its change. As we have already mentioned in Section 2.2, the title of Babbage’s book is quite telling considering the fact that it is On the Economy of Machinery and Manufactures. The book is too rich to be summarised in a few lines, so we will limit ourselves to providing some highlights on its contents. In the opening chapter Babbage points out that one of the distinguishing features of nineteenth century England consists of the extent and perfection to which the contrivance of machines has been pushed. Regarding the economic principles he states that: The advantages which are derived from machinery and manufactures seem to arise principally from three sources: The addition which they make to human power. – The economy they produce of human time. – The conversion of substances apparently common and worthless into valuable products. (Babbage 1835: 6, original emphasis) Babbage outlines explicitly the distinction between making and manufacturing, where the latter refers to the “economical principles which regulate the application of machinery and which govern the interior of all our great factories” (op. cit. p.  119); put another way, the manufacturer must be aware of the relevant technology which must be arranged in such a way that it minimises costs.10 He reconsiders the Smithian principle of the division of labour, pointing out that it can be also applied to “mental labour”,11 and that the division of labour in terms of conceiving of, drawing and executing must be applied to machine building (op. cit. p. 266).

Introduction   9 Marx’s contribution to the analysis of technology in capitalist economies is widely acknowledged, at least in non-­mainstream economics.12 He devoted a lot of attention to technology particularly in Chapters 14 “Division of Labour and Manufacture” and 15 “Machinery and Large-­scale Industry” of Volume I of Capital. In Chapter 14 Marx reconstructs the dual origin of manufacture, which is that form of co-­operation based on the division of labour which arises: 1. By the assembling together in one workshop, under the control of a single capitalist, of workers belonging to various independent handicrafts [. . .] 2. [or when one] capitalist simultaneously employs in one workshop a number of craftsmen who all do the same work, or the same kind of work . . . (Marx 1867: 455–6) Manufactures, in turn, have two fundamental forms, namely heterogeneous and organic. The latter is the perfected form in which a product goes through step-­ by-step connected operations and phases and at the end we find the final shape of the product itself (op. cit. pp. 461–3). In the fifteenth chapter of Capital, Marx provides an economic analysis of the technology which emerged with large-­scale industry. The investigation is centred on the process of mechanisation which was taking place in more and more industries. Systems of machines had become the rule: they were characterised by the presence of a prime mover, i.e. the steam engine, which gave impulse to the transmitting mechanisms which, in turn, gave movement to the specialised machines which actually performed some operation(s). In this way, says Marx (p. 500) we reach technical unity. Furthermore The transformation of the mode of production in one sphere of industry necessitates a similar transformation in other spheres. . . . Thus machine spinning made machine weaving necessary, and both together made a mechanical and chemical revolution compulsory in bleaching, printing and dyeing. (op. cit. p. 505) Another key passage in the transition to large-­scale industry was the development of a machine-­tool sector, i.e. that branch of mechanical industry which builds the machines for producing machines.13 Of course these economists centred their analyses on mechanical industry which was the key sector to develop industrial production. However their studies are as useful today as they were at the time. The principle of the division of labour is as important today as it was two centuries ago; indeed the systemic view which emerges from the Marxian theory has not lost importance, and it would be easy to provide a long list of relevant topics. The fact that we have referred to the founders of classical theory should not lead us to forget the whole body of literature which has developed along this line, and which has produced a whole theory in which structural change and economic

10   N. De Liso and R. Leoncini dynamics have always been central (e.g. Pasinetti 1981; Steedman 1984; Kurz and Salvadori 1995; Quadrio-­Curzio and Pellizzari 1999). The reswitching of technique problem, which is one of the important results of this theory, deserves an explicit comment: in fact, it contradicted the ultimate truth according to which a lower interest rate necessarily implied a process of capital deepening. The neoclassical-­mainstream school started to investigate technology systematically rather late, with its initial basic tool being the Cobb-­Douglas production function14 (Cobb and Douglas 1928) and the connected isoquants, originally referred to as contour lines.15 For a long time the neoclassical school has treated technology as a black box16 referring to it as the unknown way in which inputs (or factors of production) to the production process are transformed into output. Given the factors’ reward we can identify that technology which minimises costs for a given output. The last comment makes it clear that different techniques, or combinations of inputs, exist according to which we can produce a certain commodity, while the shape of (well-­behaved) isoquants implies the substitutability between factors – the ease of which is measured by the elasticity of substitution. A whole apparatus has been developed around this basic idea, which has been used at both micro and macroeconomic level; such an apparatus, however, has not properly dealt with technology for a long time – and even today many criticisms can be raised (see Sections 3.2 and 3.3). With technology being one of the main drivers of change, the fact that the evolutionary school has always attached great importance to its study will not come as a surprise. Given the self-­imposed label, it should also be clear that this school draws ideas from biology – one of the first names which rightly comes to mind is that of Charles Darwin and his book on The Origin of Species, which was published in 1859.17 Borrowing ideas from other disciplines always implies some risks, and in the case economics-­biology these were largely identified by Penrose18 (1952). The development of the theory has overcome these difficulties so that if at the end of the nineteenth century Veblen asked “Why is economics not an evolutionary science?” (Veblen 1898), today we have a well developed evolutionary economic theory.19 Among the forerunners of this school the most important single contributor is Joseph Schumpeter20 (1912, 1950) who was interested in the process of endogenous economic development: “By ‘development’, therefore, we shall understand only such changes in economic life as are not forced upon it from without, but arise by its own initiative, from within” (Schumpeter 1912, Engl. tr. 1934, p. 63). Technology and innovation, both stimulated by the search of profit, are central to this process of capitalistic economic development. The study of technology has always been at the centre of evolutionary theorising, and it is difficult to take one single starting point.21 Let us consider Dosi’s path-­breaking article on technological paradigms: he defines technology as a set of pieces of both practical and theoretical knowledge, know-­how, methods, procedures, experience of success and failures, and physical devices and equipment which define a state-­of-the-­art and a basis characterised by a limited set of technological alternatives and notional future developments (Dosi

Introduction   11 1982:  151–2). He thus defines a technological paradigm “as ‘model’ and a ‘pattern’ of solution of selected technological problems, based on selected principles derived from natural sciences and on selected material technologies” (Dosi 1982: 152, original emphasis). Technological paradigms have both a positive and negative side; in fact, while on the one hand they are characterised by certain materials, principles, know-­how, beliefs and other features which allow us to achieve a desired result, on the other hand they have a powerful exclusion effect in that they inhibit the exploration of other technological possibilities (op. cit. pp. 152–3). When technology develops in a capitalistic environment the profit incentive constitutes one of the fundamental stimuli to technology creation; in addition, capitalism has also been able to exploit state-­financed science and technology directed at satisfying non-­economic needs, and one of the most important is defence. The best-­known example is the Internet, but others, sometimes not as widely known as in the case of the first numerically-­controlled machine tools,22 could easily be added. 3.2  Innovation What is meant by innovation is usually product and process innovation, that is the creation of new products and services, the improvement of existing ones, and changes in the technology with which products and services are produced and provided. Schumpeter considered as innovation also the opening of new markets, the conquest of new sources of supply of raw materials and/or half-­manufactured goods, and the creation of the new organisation of an industry – e.g. the creation of a monopoly position (Schumpeter 1912: 66). Economists and policymakers’ attention to innovation scarcely deserves any justification: “innovation is of importance not only for increasing the wealth of nations in the narrow sense of increased prosperity, but also in the more fundamental sense of enabling people to do things which have never been done before” (Freeman and Soete 1997: 2). Since the advent of the Industrial Revolution innovation has become a structural feature of capitalist economies. The fact that such a revolution occurred in a capitalistic context does not mean that capitalism is the only form of economic organisation which allows for innovation, nor does it mean that before the Industrial Revolution no improvements occurred. It simply means that capitalistic institutions and incentives, together with science and technology systematically applied to production of goods and services, gave rise to an environment in which technology creation and innovation were a constituent part. This idea has been expressed in various ways by different authors. For instance, Schumpeter wrote that: The essential point to grasp is that in dealing with capitalism we are dealing with an evolutionary process.  . . . Capitalism, then, is by nature, a form or

12   N. De Liso and R. Leoncini method of economic change and not only never is but never can be stationary.  . . . The fundamental impulse that sets and keeps the capitalist engine in motion comes from the new consumers’ goods, the new methods of production or transportation, the new markets, the new forms of industrial organization that capitalist enterprise creates. (Schumpeter 1950: 82–3) The last quotation ties in closely with Schumpeter’s previous work, where he explicitly connected development with innovation (1912: 66). As for the nexus between capitalistic production and innovation it is worth noting that Marx wrote that: “Modern industry never views or treats the existing form of a production process as the definitive one. Its technical basis is therefore revolutionary, whereas all earlier modes of production were essentially conservative” (Marx 1867:  617). Capturing the idea of capitalism as an evolving system, as well as the Schumpeterian description of innovation as an endogenous phenomenon of creative destruction, is the phrase restless capitalism: The label “Restless Capitalism” has two dimensions: searching for improvement and a sense of discomfort that arises as unanticipated structural changes falsify expectations and devalue or revalue existing investments. The first encompasses the process of innovation. . .; the second introduces the evolutionary welfare economics of general system “progress” in the presence of localised “pain”. [. . .] It is fundamental to an understanding of restlessness that modern capitalism is a creative system in which new forms of economic behaviour are continually stimulated. (Metcalfe 2008a: 173–4) Innovation was already central to the founders of classical economics, and has become a central aspect of economic theory, in whatever school of thought we refer to. The concept of innovation, once more, is difficult to encompass with one definition. A first difficulty lies in the distinction which is traditionally drawn between invention on the one hand and innovation on the other; of fundamental importance also are the processes which lead to technology adoption – i.e. the firm’s decision – and thus its diffusion. Even to equate innovation with technological change deserved explicit analysis (Ruttan 1959). For a long while the so-­called Schumpeterian trilogy23 of invention-­ innovation-diffusion has provided the basis on which a linear model of technological change could be proposed: The origin of the linear model can be traced to Schumpeter’s sequence of invention, innovation and diffusion in strict temporal order, with, along the way, the insertion of value judgements about the relative contribution of science and technology to innovation. (Metcalfe 1995: 461–2)

Introduction   13 Furthermore, Schumpeter (1939) explicitly wrote that innovation is possible without anything we should identify as invention, and invention does not necessarily imply innovation. In some cases the distinction between invention and innovation can be meaningful,24 particularly when a time lag occurs between the first occurrence of an original idea and its economic application. However, the linear model has proved fragmented, hiding the interdependence and feedback between the stages as well as other elements which cast light on the continuity of advance; put another way, the boundaries between the three stages are actually blurred (Metcalfe 2008b: 212, 1995: 462). Furthermore, we must not forget the step of technology adoption, which concerns the decision process of single firms which – when they realise that some other firm has accomplished an innovation – will have to decide if and when to adopt the innovation itself. What is implied in the last sentence is the fact that firms somehow learn that an innovation has occurred somewhere in their sector and this process of learning can take different routes, such as participation in sector networks, consultancy, participation in exhibitions, systematic analysis of technical literature, voluntary dissemination of the original innovator, industrial espionage, decreasing market share, diminished profitability and others. One of the characteristics of modern capitalism is that a good deal of innovativeness – which may include an original invention – has become a routine process, particularly in oligopolistic markets (Baumol 2002; Gilfillan 1952). Small firms are also innovation-­conscious and, when they are not the first developers of innovation, they are ready to adopt, or imitate, it. We can thus observe a whole range of situations in which we have firms developing new products, services or processes in-­house, firms which sell/ license their technology to competitors, firms which specialise in the development of a technology which they sell/license to any buyer, technology joint ventures, firms which try to imitate innovation for instance through reverse engineering. This (non-­exhaustive) list makes it clear why it is so difficult to provide general models capable of synthesising innovations and their diffusion. Difficulties may be amplified by the fact that once the first innovation occurs, a wave of further improvements often follows, further boosted through the diffusion process: in fact, in many cases technology cannot be simply adopted, but it must be adapted.25 Let us consider computer technology. Hardware and software have become a necessary component for the bulk of manufacturing and services; just think of what the technology of the car industry has become or of the banking industry. Over time, computers have become faster, cheaper, more reliable, have been interconnected through network technologies, and have changed their nature from being machines which store, update and process data into machines used to exchange communication through active interaction; pervasiveness of computer technology is so great that we can speak of a digital division of labour (De Liso 2008). It should thus be clear that, as a rule, innovation is not a one-­off

14   N. De Liso and R. Leoncini phenomenon but is a process in which feedbacks and further intentional (and sometimes unintentional) developments are always present. This allows us to understand why economists have found it difficult not only to define, but also to measure technological change. In fact, if innovation is characterised by many dimensions, it will be difficult to conceive of measures for all of these same dimensions. Furthermore, measurement may take a different meaning according to whether we think of industrial or service sectors. Finally, the methods and indicators will also be affected by the more or less aggregate level of analysis. For a long time studies based on the Solow (1957) macroeconomic production function approach have been at centre stage. The macroeconomic approach, however, is not very helpful, as it subsumes indistinctly all forms of change, synthesised in a shift of the production function itself. Sectoral studies are more helpful, as in this way it becomes possible to identify key drivers of development and growth – it is enough to think of the importance of the mechanical industry or of the computer industry. Many attempts have been made in the direction of measuring innovation. Patel and Pavitt (1995) and Keith Smith (2005) provide a review of what indicators and methods have been and are being used. The most important ones are R&D expenditure, patenting activity, patent citations, bibliometric data, surveys of technical experts and large scale surveys such as the Community Innovation Survey. These indicators and methods are sometimes mixed in with country, sectoral and large firm studies – just think of the wave of studies concerned with national systems of innovation (Lundvall 1992; Nelson 1993). 3.3  Knowledge in technology and science Whenever we deal with technology and innovation there always exists a knowledge dimension which comes out. We cannot review the epistemological debate on what is meant by knowledge. We refer to it in the first instance in the way in which it is referred to in any dictionary, that is in terms of theoretical and/or practical understanding of a subject, the stock of what is known, the expertise and skills acquired by a person through experience and education. We will later distinguish between technological and scientific knowledge, which overlap but do not necessarily coincide. Let us emphasise the fact that any technology, however elementary, has a knowledge content: just think of the artisans of medieval guilds. Two structural features of modern capitalism lie in the systematic search of ways to apply existing knowledge to economic ends and in the intentional activity – typically R&D – aimed at producing new knowledge directed at economic purposes. As we have already mentioned the phrases knowledge-­based economy and knowledge-­based society have become very popular in the last 25 years, but such an emphasis is not justified. First of all, we have to note that “every economy, always and everywhere is a knowledge economy, for social systems – and

Introduction   15 economies are social systems – could not be arranged otherwise” (Metcalfe and Ramlogan 2005: 656). Should we confine ourselves to modern capitalist economies, the link between knowledge and production was explicitly identified a long time ago; once more we can use Charles Babbage as an example, where he wrote in the final pages of his most famous book that: [I]t is impossible not to perceive that the arts and manufactures of the country are intimately connected with the progress of the severer sciences; and that, as we advance in the career of improvement, every step requires, for its success, that this connexion should be rendered more intimate. [. . .] The experience of the past, has stamped with the indelible character of truth, the maxim, that “Knowledge is power”. (Babbage 1835: 379, 388, original emphasis)26 When we refer to knowledge in technology and in science we may be facing different questions; technologists ask questions such as: “Does it work?”, “Is it reliable?” or “Can it be made compatible with existing devices?”; scientists may ask different questions such as “Why does it work?”, “Can this phenomenon be reproduced?”, “What is the predictive power of my knowledge?”. Different questions may create different forms of knowledge, such as know-­why, know-­ that and know-­how; the first two are typical of science, the third is typical of technology, but no clear-­cut boundaries exist, in the sense that there is always some know-­how in science, while there is some know-­that and know-­why in technology.27 One main difference between scientific and technological knowledge consists of the consequences of the emergence of new knowledge: in technology very often we find explicit efforts aimed at building so-­called bridging technologies, so that old investments can partly be saved, while in science new knowledge usually means a new theory which supersedes the existing one. Examples for technology are the devices conceived to make use of both direct and alternating current or the co-­existence of fibre optics and copper wires for data transmission, while in physics Einstein’s theory “simply” superseded Newton’s. Further differences between scientific and technological knowledge are due to the different roles played by trial and error as well as to the fact that knowledge may emerge as the result of different data, information and beliefs. An example is the chemical industry as it developed throughout the second half of the nineteenth industry; in this period we witness the development of this sector in two ways: on the one hand there were firms which based their production on experience, rule-­ofthumb and trial and error, while on the other hand there started to emerge research laboratories which yielded theoretical work (Landes 2003: 274). The latter comment gives us the opportunity to emphasise the fact that basic scientific knowledge may arise from private research laboratories aimed at seeking knowledge for economic purposes: the usual examples are the Nobel prizes awarded to scholars – technologists and scientists – working on practical problems such as optical communication, digital imaging, and others.28

16   N. De Liso and R. Leoncini Of course we are aware of the fact that technological and “pure” scientific knowledge are characterised by different incentives, which are usually economic for the former and non-­economic for the latter: scientists are usually more concerned with publications and peer recognition than with problem-­solving. And yet many technologists want to be recognised as the first to have invented something, while sometimes scientists turn to entrepreneurs or promote spin-­offs. The first claim is supported by the many patent litigations in which money is not the only point at stake: just think of the case of the telephone patent issued by the US patent office in 1876 in favour of Alexander Graham Bell which generated a huge number of lawsuits, promoted by other inventors who wanted to see their role acknowledged.29 The second claim is buttressed by the increasing numbers of university spin-­offs, typically in hi-­tech sectors, and by the role identity modification of university scientists involved in commercialisation and technology transfer activities (Zucker and Darby 1996; Shane 2004; Jain et al. 2009). The final comments we want to make hint at the science and technology policy issue.30 The importance of knowledge for the economy has been recalled many times, and the fact that governments and non-­profit institutions try to participate and intervene in the process of knowledge creation is hardly surprising. Often government intervention is deemed useful and actively sought because the production of knowledge is subject to market failures – knowledge as a public good, uncertainty and externalities are the first words which come to mind. The debate has concentrated for some time specifically on R&D, the starting point being that “were the field of basic research left exclusively to private firms [. . .] profit incentives would not draw so large a quantity of resources to basic research as is socially desirable” (Nelson 1959:  304).31 And this is why a lot of R&D is carried out in public research laboratories (or in private ones, but with taxpayers’ money). However, R&D is only part of the issue. State intervention, in fact, considers many other important features among which we recall the establishment of compulsory primary education, the regulation of the overall education system, the resources devoted to knowledge concerned with defence or health, the existence of subsidies and incentives, antitrust laws and more. Of course governments are also subject to failure. As one can see, even a simple hint at these problems clarifies that it is not possible to think in terms of optimal allocation of resources to knowledge generation and why different perspectives, such as national systems of innovation, have gained ground.

4  Globalisation32 Globalisation is another word that has become very popular in the recent past. However, once more, a better look at economic reality will show that many characteristics of today’s globalisation have been there for a long time.33 What is meant by economic globalisation is the process of increasing integration between national economies which takes places through international trade of goods and services, the internationalisation of capital markets and the international mobility of labour.

Introduction   17 What we want to do here is to concentrate on the relationships between firms, technology and globalisation without going into the details of all the above dimensions, nor do we want to analyse those institutional changes which have come together with globalisation itself, symbolised by the institutionalisation of the World Trade Organization (WTO), the fall of many barriers and the predominance of financial markets. When we refer to firms we mean not only giant multinational firms – which are globalised by definition – but also small- and medium-­sized ones. The latter, in fact, have become more global in the last 20 years in at least three ways: there has been increasing attention towards exports, on finding new sources of semi-­ manufactured articles and a process of re-­location of (at least) part of their process of production has occurred. A phenomenon which, in general, is experiencing increasing importance is the so-­called global, or international, production sharing (Yeats 1998). What is meant by global production sharing is that form of organisation of production whereby the different phases of the process of production of a good take place in different countries; in this way absolute and comparative advantages, even related to a single phase of the overall process of production, can be exploited. Early forms of global production sharing typically involve the production of a primary commodity (e.g. iron ore) in a developing country, the shipment of this commodity to an industrial nation for processing and the re-­exportation, at least in part, of the processed commodity back to the first country (op. cit. p. 3). The existence of such a phenomenon was explicitly identified by Babbage: The produce of our factories has preceded even our most enterprising travellers. The cotton of India is conveyed by British ships round half our planet, to be woven by British skill in the factories of Lancashire: it is again set in motion by British capital; and, transported to the very plains whereon it grew, is repurchased by the lords of the soil which gave it birth, at a cheaper price than that at which their coarser machinery enables them to manufacture it themselves. (Babbage 1835: 4) There clearly emerges the importance of firms, their technology and innovativeness, and the globalisation of markets. In particular, the exploitation of global production sharing depends heavily on transportation technologies. The modern evolution of transportation technologies and networks, from giant ships such as oil tankers to high-­speed trains and low cost airlines is known to us all. A key historical passage was the application of steam-­engines to ships: it increased carrying capacity and made delivery times reliable. Steam technology was applicable to terrestrial transportation, too. Indeed, the world’s first railway was inaugurated in 1830 and linked Liverpool – then probably the world’s most important sea port – with Manchester, i.e. the heart of the Industrial Revolution. The railway was meant for the transportation of goods, but it was soon used to transport people as well.

18   N. De Liso and R. Leoncini As the reader may notice we are back to the concept of general purpose technology which was introduced in Section 3.2: the steam engine, which was originally built to suck water out of the depths of the coal mines, rapidly became the prime mover of Manchester’s factories, and was applied to transportation over water and over land. Over the course of this process, the performance, size, and reliability of these engines were constantly improved, thanks to improved materials and better knowledge of the principles behind steam technology. Put in another way, the general purpose technology steam-­engine contributed heavily to the first wave of globalisation. Let us clarify the fact that international trade existed well before the Industrial Revolution. Indeed, commercial capitalism precedes industrial capitalism. The words commercial capitalism are used to refer to the period between 1100 and 1690 (Landes 1966); during this period we had trade which was aimed at profit and accumulation. Pirenne notes that as early as the twelfth century it was normal for traders to travel between various European countries, and it is not possible to argue that these traders travelled so far simply for subsistence: their goal was the constant accumulation of wealth, and their tool was large-­scale commerce (Pirenne 1913: 504). The “problem” of commercial capitalism was that while commerce was rapidly expanding, particularly after Europeans reached the Americas, production techniques improved slowly. Although trade was already globalised by the end of the sixteenth century,34 a key impulse was missing: that of production organised on rational and scientific bases. The fusion between economic capitalistic impulse and technological and scientific rationalism applied to production took place with the era of industrial capitalism, that is with the First Industrial Revolution: this is the dividing line. It should thus be clear that globalisation was not driven by commerce, but by the power of the mechanised productive processes that were developed with the application of steam and machinery to the English textile industry. The turning point was the systematic production, organised along capitalistic principles, which exceeded local or domestic demand: production increased so that trade could increase, but the key word was production rather than trade. Worthy of comment is the fact that the Industrial Revolution strengthened the dynamics related to the division of labour, and a further dimension, namely that of the international division of labour, began to emerge. We can also remember here the link between economics and the economy: Adam Smith had already elaborated the theorem according to which the division of labour is limited by the size (extent) of the market, that is the larger the market, the more developed the division of labour. It is also true that the size of the market is determined by the division of labour: in other words, the more production becomes efficient, the greater our ability to penetrate new markets, since our (low) costs and prices lead to greater demand, and therefore to a further expansion of the market. Technology and technological change are at the same time a cause and an effect of globalisation. Technologies become global in that they tend to spread from both a technological and a geographic point of view. One only needs to

Introduction   19 think – besides the steam-­engine – about the spread of electricity or modern information technology. These technologies have had an exceptional impact from a technological point of view, and have spread right across the globe. In other words: the most flexible technologies from a technical point of view become globalised technologies. This is due to the fact that there are global markets in which not only production, but also consumption patterns, tend to become increasingly homogenous. It is important to stress that in some cases the same technology has affected both the production of goods and the capacity to trade them: this was the case of steam-­based technologies. In the most advanced economies, where services are predominant but where an industrial basis persists, information technology has a fourfold aim, as it is used to produce goods, provide services, organise brokering activities which make trading goods and providing services possible, and consume both goods and services. With regard to the latter, one can think of the importance of digital consumption – ranging from mobile phones, which are increasingly used for activities other than speaking, to home banking.

5  Conclusion and plan of the volume It would be too pompous to draw final conclusions on three topics such as the firm, technology and globalisation at the end of a short introduction of a book. What we have done, instead, has been to make an attempt at putting forward some critical ideas, pointing out first of all some of the problems which emerge when we look at the economy through the lens of economics. The issue is crystal clear when we refer to the firm, but it also emerges when we deal with technology and globalisation. A second point which we have made consists of highlighting the structural links which exist between the three items. This, in turn, helps explain the vision of capitalism as an evolving system, with the dimensions of knowledge and innovation playing a key role. The third point which should emerge concerns the wealth of ideas which can be found in economics – or, put another way, economics is not represented by a single school or body of thought. We have reappraised the role of classical economists and we have underlined a few key passages where the importance of the evolutionary school of thought was highlighted. Furthermore, some key features pertaining to the neoclassical-­mainstream theory were reiterated. Nevertheless, we are of course aware of the fact that these schools do not exhaust the list of schools of thought. A fourth point which we hope will emerge – along with the three points which have just been mentioned – from reading the chapters of this volume concerns the richness of methods which can be used in economics to study the economy. Indeed, in what follows, besides finding various background theories, the reader will find various approaches, ranging from simulations to case studies. The book is divided into two parts. In the first one the next five chapters are grouped, and these highlight the complex nature of the firm, and point to the

20   N. De Liso and R. Leoncini necessity of dealing with technology and organisation in order to understand how firms react to turbulent environments. In particular, Chapter 2 by Mauro Lombardi tackles the important and delicate issue of how firms take their production decisions, and in particular how ordered paths of production emerge from the incredibly vast set of alternatives (to which the environment adds another dimension of complexity). By adopting a morphological approach it is possible, starting from the representation of an economy populated by interacting agents, to avoid the so-­called combinatorial explosion of possible results. By advocating compositionality and recursiveness, it is possible to explain how and why order arises from combinatorial spaces, generating a finite number of possible solutions. Organisational and technological choices are at the core of Chapter 3 by Stefano Brusoni and Roberto Fontana, where the above-­mentioned choices determine the strength of the connections among elements and the governing rules. These rules constitute what is termed a platform, and the chapter provides a taxonomy of four different platform types by crossing platform architecture (modular versus non-­modular) and the type of interface (open versus closed). In this way, different cases are analysed in order to understand how and when different platform types are beneficial to incumbents or not, on the basis of rules of competition, opportunities of innovation and relative advantages. Chapter 4 by Tommaso Ciarli, Riccardo Leoncini, Sandro Montresor and Marco Valente, once again focuses on the relationships between organisation, technology and industrial structure. In this case, the organisation of a firm comes to depend on the trade-­off between the decisions to either produce production modules internally or to buy them externally. Again, the organisational structure and technological decisions are intrinsically related, as firms have the choice of producing a module internally by exploring a technological landscape that is corrugate and complex, or of outsourcing it to a more efficient supplier. In so doing, the dimension of the firm (and the industrial dynamics) is endogenously determined by both the internal and the external relationships. Building on their previous work Nicola De Liso and Giovanni Filatrella provide, in Chapter 5, a formal model which describes the so-­called “sailing ship” effect. The latter is the process whereby an incumbent technology is improved in response to the emergence of a new and technologically superior one. The competition between the old and the new technology is based on the performance of the two technologies – performance which can be improved only through explicit efforts which take the form of R&D. The choice thus concerns the amount of resources devoted to R&D given the technological characteristics of both technologies. This chapter has two important features: first, it shows a comparison between rule-­of-thumb and maximisation behaviour in order to make the choice on R&D funding; secondly it considers technological variety and the co-­existence between old and new technologies. Chapter 6, by Francesco Rentocchini and Giuditta De Prato, tackles an important problem which has to do with strategic behaviour of firms with respect to a peculiar and fundamental aspect: that of patents in software. Indeed, the

Introduction   21 issue of innovation and knowledge is implied in this aspect, especially in terms of knowledge appropriability. In particular, seeing as ICTs are assuming an overwhelming importance, software patents are the key to understanding how firms are dealing with knowledge management today. The empirical analysis, performed on patents granted by the European Patent Office, shows how patents are not deemed to be the main instrument in maintaining knowledge appropriability even in software and, as a consequence, the previous economic analysis about non-­patent means of knowledge appropriability is reinforced. The second part of the book focuses on the role of innovation and globalisation in local economies. Chapter 7, by Giulio Cainelli and Claudio Lupi, contributes to the debate on the role and impact of spatial variables on firms’ economic performance. By using a panel of Italian manufacturing firms for the period 1998–2001, the impact of variables of spatial agglomeration is tested across a spectrum of different distances among firms. In this way, it is possible to understand the exact geographical decay of spillovers as distance increases: localisation effects have a positive impact within two kilometres, while it decreases with distance. This analysis calls for a more careful use of standard statistical geographical units (such as, for instance, local labour systems or standard metropolitan area) that squeeze out the distance factor. The three final chapters deal with the phenomenon of globalisation of smalland medium-­sized Italian firms (SMEs) and the case of the Benetton Group. In so doing, they shed light on different aspects brought about by the deepening process of internationalisation that has involved Italian firms and districts. In particular, in Chapter 8, Donato Iacobucci and Francesca Spigarelli investigate the peculiar patterns with which SMEs in the north-­west of Italy responded to the wave of internationalisation in 2000. The export patterns (as measured by several indicators) seem to point to geographically concentrated patterns of foreign direct investment (FDI), which appear to complement rather than substitute exports. Quite diversified patterns emerge if the stage theory is used, with some results pointing to the idiosyncratic nature of the Italian industrial structure. Chapter 9, by Roberto Antonietti and Davide Antonioli, looks at the outsourcing strategies of Italian firms across the period 1995–2003, in order to understand if and how the outsourcing strategies affected the labour composition. Using a balanced panel, a counterfactual experiment was set up. The results point to a skill bias effect that production offshoring has on firms. In particular, an upward shift in the skill ratio is observed, which seems to be significantly driven by the fall in blue collar workers and not by an increase in white collar workers. One specific case is then analysed in Chapter 10 by Paolo Crestanello and Giuseppe Tattara. They provide a detailed analysis of the strategies of the Benetton Group since the mid-­1990s, highlighting the structural changes which they have undergone: from a vertically integrated firm to a strong outsourcing strategy of its suppliers. The supply chain is now articulated along two paths: one for continuous collections linked to geographically close suppliers (located in Eastern Europe and North Africa); another one, for more standard products decentralised in quite distant countries (China and India).

22   N. De Liso and R. Leoncini We cannot close this introduction without acknowledging the fact that the research which has led to the publication of this book would have not been possible without the funding coming from two research projects. The first consists of the National Research Programme (PRIN) “Fragmentation and local development: interpretative models and policy implications”, co-­ordinated by the University of Bologna and sponsored by the Italian Ministry for Education, University and Research (MIUR); the second was funded by the Autonomous Province of Trento, via the OPENLOC research project under the call for proposals “Major Projects 2006” – the project is co-­ordinated by the University of Trento and both the Manchester Institute of Innovation Research and the University of Bologna are partners of the project itself. Last but not least, we wish to thank Ian J. Gavin for his linguistic appraisal of this introduction.

Notes   1 Seminal contributions by Berle and Means (1932), Baumol (1959), Williamson (1963) and Marris (1963) laid down the basis for this kind of analysis.   2 A few exceptions exist, e.g. Foss (1997).   3 See, respectively, Leijonhufvud (1986), Groenewegen (1987) and Landes (1986).   4 Coase writes that “It is sometimes said that the reason for the existence of a firm is to be found in the division of labour” but classifies as “inadmissible” some ideas proposed by Usher and Dobb which refer to this as a principle capable of explaining the existence of firms (Coase 1937: 398). Some critical points have also been raised by Buenstorf (2005), Leoncini et al. (2009), and Lombardi (Chapter 2 in this book).   5 Charles Babbage (1791–1871) was a polymath, and is usually known among computer scientists for his attempt at building two different mechanical computers; his contribution to economics belongs to the classical tradition. The first edition of the Economy was published in 1832; in this work we refer to the third edition which was published in 1835.   6 As for technological unemployment a fierce debate started in the 1820s and saw the participation of many economists such as Malthus, McCulloch, Sismondi, and Say. Worthy of a specific comment is the fact that David Ricardo added the chapter “Of Machinery” to the third edition of his Principles to investigate whether the adoption of machinery could be detrimental to the labour force.   7 This has prompted recent and developing literature on make-­and-buy behaviour; see, for instance, Gulati and Puranam (2006); Parmigiani (2007); Parmigiani and Mitchell (2009); Antonietti et al. (2009).   8 This is a point which has been tackled by the literature on dynamic capabilities, starting from the pioneering work of David Teece (Teece et al. 1997).   9 Chapter VII, The Economics of Diversification, of Penrose’s book (1959) is quite important in order to understand the meaning of growth as attached to the firm. 10 Concerning cost minimisation he says that the manufacturer “must carefully arrange the whole system of his factory in such a manner, that the article he sells to the public may be produced at as small a cost as possible” (Babbage 1835: 121). 11 Op. cit. pp. 191–202. 12 E.g. Rosenberg (1976) or Nelson and Winter (1982). 13 More generally, on the role and importance of the capital-­goods sector in Marx see Rosenberg (1976: 65–74). 14 We are aware of the fact that some authors give priority to Wicksteed or even to Von

Introduction   23 Thunen as the first economists to propose the idea of the production function; we refer to the Cobb-­Douglas as this is the one which has been systematically referred to by economists. 15 In the first note of his article Cobb reminds us that the idea of using contour lines in economics is ascribed by Pareto to Edgeworth (Cobb 1929: 225). 16 For a thorough analysis see Rosenberg (1982). 17 Hayek (1973, p. 152, note 33) points out that Darwin and the other discoverers of the theory of evolution actually owed the suggestion to social theory. 18 Penrose referred specifically to the biological analogies in the theory of the firm; however her criticisms had a broader implication for economic theorising. 19 See, among the many, Nelson and Winter (1982), Dopfer (2005), and Alchian (1950) for an early account. 20 As for the other precursors and allies of evolutionary theory the reader can refer to Nelson and Winter (1982:  33–45); to their list we should add the name of Kenneth Boulding (1950, 1981). 21 To have an idea of the importance of technology and innovation for the evolutionary theory one can refer to Dosi et al. (1988). 22 Numerical control was introduced in the machine tool industry by a defence-­related project aimed at building military aircraft wings (Reintjes 1991). 23 Stoneman (1995: 2). 24 “An important distinction is normally made between invention and innovation. Invention is the first occurrence of an idea for a new product or process, while innovation is the first attempt to carry it out into practice. Sometimes, invention and innovation are closely linked, to the extent that it is hard to distinguish one from another (biotechnology for instance)” (Fagerberg 2005: 4–5). 25 One can think of many examples, the clearest being the ones referred to as general purpose technologies. Technologies can be defined as general purpose if they are (i) pervasive, (ii) subject to further development once introduced, and (iii) capable of leading to innovative complementarities (Bresnahan and Trajtenberg 1995; Helpman 1998). A similar concept can be found in David (1991). 26 That of Babbage is far from being an exception: see, e.g., Rae (1834: 15 and Chapter X entitled “Of the Causes of the Progress of Invention and of the Effects Arising From It”). 27 Despite the fact that explicit analyses of knowledge are not particularly popular among economists, there exists a number of key studies, among which we should point out the ones by Mokyr (2002), Foray (2004), Loasby (1999), Jewkes et al. (1969), Machlup (1962), Hayek (1945) and the bibliographies and references therein. 28 The careful reader may have realised that we are referring here to the 2009 Nobel Prize winners W.S. Boyle, C.K. Kao and G.E. Smith. They do not represent an exception: we could have referred to many other inventions (e.g. the transistor) and discoveries (e.g. cosmic microwave background radiation) which emerged from industrial or applied research. 29 The echoes of the story were going on at least until 2002 when the US House of Representatives issued a resolution acknowledging the role of Antonio Meucci. 30 For thorough analyses see Metcalfe (1995), Lundvall and Borràs (2005) and David (2008). 31 Analogously, Arrow (1962) writes of “underallocation of resources to invention by private enterprise”. 32 This section draws on De Liso and Casilli (2005). 33 For an analysis of globalisation in historical perspective see Bordo et al. (2003). 34 As an example of early global commerce we can recall the fact that Manila was founded in 1571 and that it served as a trade hub between Europe, Asia, Africa and the Americas (Findlay and O’Rourke 2003).

24   N. De Liso and R. Leoncini

Bibliography Alchian A.A. (1950), Uncertainty, Evolution, and Economic Analysis, Journal of Political Economy, 58(3) June: 211–21. Antonietti R., Ferrante M. and Leoncini R. (2009), Local Spillovers, Production Technology and the Choice to Make-­or-Buy. Empirical Evidence from Emilia Romagna Mechanical Industry, Working Paper Openloc 2/2009, University of Trento, online, available at: www. openloc.eu/cms/storage/openloc/working papers/ 2009/AntoniettiFerranteLeoncini2009. pdf. Arndt S.W. (2004), Global Production Networks and Regional Integration, in M.G. Plummer (ed.), Empirical Methods in International Trade, Chapter 8, pp.  129–43, Cheltenham: Elgar. Arrow K.J. (1962), Economic Welfare and the Allocation of Resources for Invention, in NBER, The Rate and Direction of Inventive Activity, Princeton: Princeton University Press, reprinted in P. Mirowski and E.-M. Sent (eds) (2002), Science Bought and Sold. Essays in the Economics of Science, Chicago: University of Chicago Press, pp. 165–80. Babbage C. (1835), On the Economy of Machinery and Manufactures, 4th edition, London: Charles Knight. Reprint 1963, New York: Augustus M. Kelley. Baumol W.J. (1959), Business Behavior, Value and Growth, New York: Macmillan. —— (2002), The Free Market Innovation Machine, Princeton: Princeton University Press. Berle A.A. and Means G.C. (1932), The Modern Corporation and Private Property, New York: Harcourt, Brace & World. Bordo D., Taylor A.M. and Williamson J.G. (eds) (2003), Globalization in Historical Perspective, NBER, Chicago: University of Chicago Press. Boulding K. (1942), The Theory of the Firm in the Last Ten Years, American Economic Review, 32(4) December: 791–802. —— (1950), A Reconstruction of Economics, New York: Wiley & Sons. —— (1981), Evolutionary Economics, London: Sage Publications. Bresnahan T.F. and Trajtenberg M. (1995), General Purpose Technologies: Engines of Growth?, Journal of Econometrics, 65: 83–108. Buenstorf G. (2005), Sequential Production, Modularity and Technological Change, Structural Change and Economic Dynamics, 16(2) June: 221–41. Cheung S.N.S. (1983), The Contractual Nature of the Firm, Journal of Law and Economics, reprinted in Foss (2000b), 1: 325–44. Coase, R. (1937), The Nature of the Firm, Economica, 4(16): 386–405. Cobb C.W. (1929), The Use of Contour Lines in Economics, Journal of Political Economy, 37(2) April: 225–9. Cobb C.W. and Douglas P.H. (1928), A Theory of Production, American Economic Review, 18(1) Supplement, Papers and Proceedings, March: 139–65. Cyert R.M. and Hedrick C.L. (1972), Theory of the Firm: Past, Present, and Future; An Interpretation, Journal of Economic Literature, 10(2): 398–412. David P.A. (1991), Computer and dynamo. The modern productivity paradox in a not-­ too-distant mirror, Technology and Productivity. The Challenge for Economic Policy, Paris: OECD, pp. 315–47. —— (2008), The Historical Origins of “Open Science”: An Essay on Patronage, Reputation and Common Agency Contracting in the Scientific Revolution, Capitalism and Society, 3(2): Article 5. De Liso N. (2008), ICTs and the Digital Division of Labour, in R. Leoncini and S. Montresor (eds), Dynamic Capabilities Between Firm Organization and Local Systems of Production, London: Routledge, pp. 346–74.

Introduction   25 De Liso N. and Casilli C. (2005), La Globalizzazione in una Prospettiva Tecnologico-­ Istituzionale, in Economia Politica, XXII(1) April: 43–68. Dopfer K. (2005) (ed.), The Evolutionary Foundations of Economics, Cambridge: Cambridge University Press. Dosi G. (1982), Technological Paradigms and Technological Trajectories, Research Policy, 11(3) June: 147–62. Dosi G., Freeman C., Nelson R.R., Silverberg G. and Soete L. (1988) (eds), Technical Change and Economic Theory, London: Pinter. Fagerberg J. (2005), Innovation. A Guide to the Literature, Chapter 1 of Fagerberg et al. (eds) (2005), pp. 1–26. Fagerberg J., Mowery D.C. and Nelson R.R. (2005) (eds), The Oxford Handbook of Innovation, Oxford: Oxford University Press. Findlay R. and O’Rourke K.H. (2003), Commodity market integration: 1500–2000, in Bordo et al. (2003), pp. 13–62. Foray, D. (2004), The Economics of Knowledge, Cambridge: MIT Press. Foss N.J. (1997), The Classical Theory of Production and the Capabilities View of the Firm, Journal of Economic Studies, 24(5): 307–23. —— (2000a), The Theory of the Firm: An Introduction to Themes and Contributions, introductory chapter to Foss (2000b), 1: xv–lxi. —— (2000b) (ed.), The Theory of the Firm. Critical Perspectives on Business and Management, 4 volume set, London: Routledge. Freeman C. and Soete L. (1997), The Economics of Industrial Innovation, 3rd edition, London: Pinter. Gibbons M. and Johnson C. (1970), Relationship between Science and Technology, Nature, 227(11 July): 125–27. Gibbons R. (2004), Four Formal(izable) Theories of the Firm? MIT Working Paper, September. Gilfillan S.C. (1952), The Prediction of Technical Change, Review of Economics and Statistics, 34(4): 368–85. Grant R.M. (1996), Toward a Knowledge-­Based Theory of the Firm, Strategic Management Journal, 17(Winter, Special Issue): 109–22. Groenewegen P. (1987), Division of Labour, article in The New Palgrave Dictionary Vol 1, edited by J. Eatwell, M. Milgate and P. Newman, London: Macmillan, pp. 901–6. Gulati G. and Puranam P. (2006), Complementarity and Constraints: Why Firms Both Make and Buy the Same Thing, online, available at: www.cbs.dk/content/download/53898/756895/file/Ranjay%20Gulati.%20Phanish%20Puranam,%20. pdf [accessed December 2009]. Hall R.L. and Hitch C.J. (1939), Price Theory and Business Behaviour, Oxford Economic Papers, 2(May): 12–45. Hayek F.A. (1945), The Use of Knowledge in Society, American Economic Review, 35(4) September: 519–30. Hayek F.A. (1973), Law, Legislation and Liberty. Volume I. Rules and Order, London: Routledge & Kegan Paul. Helpman E. (1998) (ed.), General Purpose Technologies and Economic Growth, Cambridge: MIT Press. Holmström B. and Tirole J. (1989), The Theory of the Firm, in Handbook of Industrial Organization, edited by R. Schmalensee and R.D. Willig, Amsterdam: North Holland, Chapter 2, pp. 61–133 reprinted in Foss (2000b), 1: 148–219. Jain S., George G. and Maltarich M. (2009), Academics or Entrepreneurs? Investigating

26   N. De Liso and R. Leoncini Role Identity Modification of University Scientists Involved in Commercialization Activity, Research Policy, 38(6) July: 922–35. Jewkes J., Sawers D. and Stillerman R. (1969), The Sources of Invention, 2nd edition, London: Macmillan. Jones C.I. (2002), Introduction to Economic Growth, 2nd edition, New York: W.W. Norton & Company. Jovanovic B. and Rousseau P.L. (2005), General Purpose Technologies, in P. Aghion and S. Durlauf (eds), Handbook of Economic Growth Vol. 1B, Amsterdam: Elsevier-­North Holland, Chapter 18, pp. 1181–224. Kaldor N. (1934), The Equilibrium of the Firm, Economic Journal, 44(173) March: 60–76. —— (1972), The Irrelevance of Equilibrium Economics, Economic Journal, 82(328) December: 1237–55. Knight F.H. (1921), Risk, Uncertainty, and Profit, reprint 1971 by University of Chicago Press, Chicago with a Foreword by George J. Stigler. Kurz H.D. and Salvadori N. (1995), Theory of Production. A Long-­Period Analysis, Cambridge: Cambridge University Press. Landes D.S. (ed.) (1966), The Rise of Capitalism, New York: Macmillan. —— (1986), What do Bosses Really Do?, Journal of Economic History, 46(September): 585–623. —— (2003), The Unbound Prometheus. Technological Change and Industrial Development in Western Europe from 1750 to the Present, Cambridge: Cambridge University Press. Layton E.T. (1974), Technology as Knowledge, Technology and Culture, 15(1) January: 31–41. Leijonhufvud A. (1986), Capitalism and the Factory System, in R.N. Langlois (ed.), Economics as a Process, Cambridge: Cambridge University Press, pp. 203–23. Leoncini R., Lombardi M. and Montresor S. (2009), From Techno-­Scientific Grammar to Organizational Syntax. New Production Insights on the Nature of the Firm, Working Papers PRIN 006, Department of Economics, University of Bologna, online, available at: http://www2.dse.unibo.it/prin/prin2008/wpaper/WP6_2009. pdf. Loasby B.J. (1998), The Organisation of Capabilities, Journal of Economic Behavior and Organization, 35: 139–60. —— (1999), Knowledge, Institutions and Evolution in Economics, London: Routledge. Lundvall B.-A. (1992) (ed.), National Systems of Innovation, London, Pinter. Lundvall B.-A. and Borràs S. (2005), Science, Technology, and Innovation Policy, Chapter 22 in Fagerberg et al. (2005), pp. 599–631. Machlup F. (1962), The Production and Distribution of Knowledge in the United States, Princeton: Princeton University Press [reprint 1972]. Marris R. (1963), A Model of the “Managerial” Enterprise, The Quarterly Journal of Economics, 77(2): 185–209. Marx K. (1867, English trans. 1976), Capital Vol. 1, Harmondsworth: Penguin Books. Metcalfe J.S. (1995), The Economic Foundations of Technology Policy: Equilibrium and Evolutionary Perspectives, in P. Stoneman (ed.), Chapter 11, pp. 409–512. —— (2008a), Restless Capitalism, in F. Amatori and M. Amendola (eds), Ricerca Avanzata e Alta Divulgazione. Le Momigliano Lectures 1997–2008, Terni: ICSIM, pp. 171–92. —— (2008b), Technical Change, in S.N. Durlauf and L.E. Blume (eds), The New Palgrave Dictionary of Economics Vol.8, 2nd edition, Basingstoke, Palgrave Macmillan, pp. 210–217. Metcalfe J.S. and Ramlogan R. (2005), Limits to the Economy of Knowledge and Knowledge of the Economy, Futures, 37(7) September: 655–74.

Introduction   27 Miles I. (2005), Innovation in Services, in J. Fagerberg et al. (eds), pp. 433–58. Mokyr J. (2002), The Gifts of Athena. Historical Origins of the Knowledge Economy, Princeton: Princeton University Press. Nelson R.R. (1959), The Simple Economics of Basic Scientific Research, Journal of Political Economy, 67(3) June: 297–306. —— (1993) (ed.), National Systems of Innovation. A Comparative Analysis, New York, Oxford University Press. Nelson R.R. and Winter S.G. (1982), An Evolutionary Theory of Economic Change, Cambridge: Belknap Press. Parmigiani A. (2007), Why Do Firms Both Make and Buy? An Investigation of Concurrent Sourcing, Strategic Management Journal, 28: 285–311. Parmigiani A. and Mitchell W. (2009), Complementarity, Capabilities, and the Boundaries of the Firm: The Impact of Within-­Firm and Inter-­Firm Expertise on Concurrent Sourcing of Complementary Components, Strategic Management Journal, 30: 1065–91. Pasinetti L.L. (1959), On Concepts and Measures of Changes in Productivity, Review of Economics and Statistics, 41(3) August: 270–86. —— (1981), Structural Change and Economic Growth, Cambridge: Cambridge University Press. Patel P. and Pavitt K. (1995), Patterns of Technological Activity: Their Measurement and Interpretation, in Stoneman (ed.), pp. 14–51. Penrose E.T. (1952), Biological Analogies in the Theory of the Firm, American Economic Review, 42(5) December: 804–19. —— (1959), The Theory of the Growth of the Firm, Oxford: Basil Blackwell [reprint 1963]. Pirenne H. (1913), The Stages in the Social History of Capitalism, American Historical Review, 19(October): 494–515. Polanyi M. (1962), Personal Knowledge. Towards a Post-­Critical Philosophy, Chicago: University of Chicago Press. Quadrio Curzio A. and Pellizzari F. (1999), Rent, Resources Technology, Berlin: Springer. Rae J. (1834), Statement of Some New Principles on the Subject of Political Economy, New York: A.M. Kelley [reprint 1964]. Rajan R.G. and Zingales L. (2001), The Firm as a Dedicated Hierarchy: A Theory of the Origins and Growth of Firms, Quarterly Journal of Economics, 116(3) August: 805–51. Reintjes J.F. (1991), Numerical Control. Making a New Technology, New York: Oxford University Press. Ricardo D. (1821), On the Principles of Political Economy and Taxation, 3rd edn [1st edn 1817], in P. Sraffa (ed.) The Works and Correspondence of David Ricardo Vol. 1, Cambridge, Cambridge University Press. Richardson G.B. (1972), The Organisation of Industry, Economic Journal, 87(327): 883–96. Ricketts M. (2008) (ed.), The Economics of Modern Business Enterprise, 3 volume set, Cheltenham: Elgar. Robbins L. (1928), The Representative Firm, Economic Journal, 38(151) September: 387–404. Rosenberg N. (1974), Karl Marx on the Economic Role of Science, Journal of Political Economy, 82(4) July/August: 713–28. —— (1976), Marx as a Student of Technology, Monthly Review, 28(3) July/ August: 56–77, reprinted as Chapter 2 of Rosenberg (1982).

28   N. De Liso and R. Leoncini —— (1982), Inside the Black Box. Technology and Economics, Cambridge: Cambridge University Press. —— (1990), Why do Firms do Basic Research (with Their Own Money)?, Research Policy, 19(2) April: 165–74. —— (1994), Exploring the Black Box. Technology, Economics, and History, Cambridge: Cambridge University Press. —— (2009), Some Critical Episodes in the Progress of Medical Innovation: An Anglo-­ American Perspective, Research Policy, 38(2) March: 234–42. Ruttan V.W. (1959), Usher and Schumpeter on Invention, Innovation, and Technological Change, Quarterly Journal of Economics, 73(4) November: 596–606. Schumpeter J.A. (1912 English trans. 1934), The Theory of Economic Development, New York: Oxford University Press [reprint 1961]. —— (1939), Business Cycles, New York: McGraw-­Hill. —— (1950), Capitalism, Socialism, and Democracy, 3rd edition, London: Allen Unwin. Shane S. (2004), Academic Entrepreneurship: University Spinoffs and Wealth Creation, Northampton, USA: E. Elgar. Smith A. (1776), An Inquiry into the Nature and Causes of the Wealth of Nations, edited by R.H. Campbell and A.S. Skinner, textual editor W.B. Todd, Oxford: Oxford University Press [reprint 1976]. Smith K. (2005), Measuring Innovation, in J. Fagerberg et al. (eds), pp. 148–77. Solow R.M. (1957), Technical Change and the Aggregate Production Function, Review of Economics and Statistics, 39(3): 312–20. Steedman I. (1984), Natural Prices, Differential Profit Rates and the Classical Competitive Process, The Manchester School, 52(2): 123–40. Stoneman P. (1995) (ed.), Handbook of the Economics of Innovation and Technological Change, Oxford: Blackwell. Teece D.J., Pisano G. and Shuen A. (1997), Dynamic Capabilities and Strategic Management, Strategic Management Journal, 18(7): 509–33. Veblen T. (1898), Why is Economics not an Evolutionary Science?, Quarterly Journal of Economics, 12(4) July: 373–97. Williamson O.E. (1963), Managerial Discretion and Business Behavior, The American Economic Review, 53: 1032–57. Yeats A.J. (1998), Just How Big is Global Production Sharing?, Working Paper, World Bank, online, available at: http://papers.ssrn.com/sol3/papers.cfm?abstract_id= 597193 [accessed November 2009]. Zucker L.G. and Darby M.R. (1996), Star Scientists and Institutional Transformation: Patterns of Invention and Innovation in the Formation of the Biotechnology Industry, Proceedings of the National Academy of Sciences of the USA, 93(November), Colloquium Paper, pp. 12709–16.

Part I

Technological change, firms’ organisation and incentives

2 The production process as a complex, dynamic and ordered world1 Mauro Lombardi

1  Introduction 1.1  Outline of the chapter This chapter addresses some fundamental questions: how is it possible that an ordered world happens even if it stems from an incredibly vast set of alternatives and it is hit by stochastic perturbations? What mechanisms act within multi-­level processes so that variations and stability can, however, occur? The ordered world which we are referring to is the product conceived as vector of attributes resulting from a huge amount of search developed in multiple spaces. What principles can be formulated in order to explain on the one hand the high-­dimensionality of decision processes and search spaces and on the other their reduction, on the basis of widespread interactions among agents and economic units? In tackling these questions we aim at honing in on a precise research line (Lombardi 2008a) by adopting a cognitive and evolutionary perspective, which allows us to deepen two main topics: 1) the “interaction among competence, transaction costs and scale”, intrinsic to “the multifarious nature of firms” (Morroni 2006: 11); 2) the “co-­relational nature of knowledge” (Saviotti 2007) as interactions between agents and environment unfold. In this chapter we tackle the problem of how stable and variable vectors of attributes (products) are created and compete with each other. The key field of analysis is based on information and knowledge flows which are produced by interacting units (individuals, teams, organizations). Interaction structures unroll through continuous exploiting and exploring activities performed in trying to solve techno-­productive problems within the product development process. The proposed theoretical framework leads us to analyse problem solving activities during the production process in terms of unceasing mappings between multiple search spaces and sub-­spaces, from which vectors of parameters referring to product components have to be sought out. The risk of incurring in a combinatorial explosion of alternatives during multiple mappings is avoided by means of some organizing principles, which are discussed in the chapter: (1) modularity, (2) decentralized research, (3) evolving networks of interdependencies, (4) building blocks for knowledge processes and the creation of rules, which constitute the deep structure organizing the world around us.

32   M. Lombardi Above all during search activities a completely chaotic and random world is circumvented thanks to two general principles, such as recursiveness and compositionality, which are basic properties of human decision processes. The product development process, the design and the production process are viewed as scientific problem-­solving, the dynamics of which show basic properties: stepwise, iterative, evolutionary transformation process. Indeed from an initial idea to its final concrete realization, ready to market, a variable number of requirements have to be satisfied: functionality, reliability, modifiability and performance. Evolutionary features of these processes are fundamentally discussed by recent essays (Braha and Maimon 1998; Bar-­Yam 2005). Different design paradigms have been developed and are used by scholars and experts. The most widespread paradigms are: ASE (Analysis–­Synthesis– Evaluation), Case-­Based, Cognitive, Algorithmic and Artificial Intelligence, Creative Design (SIT, Structured Inventive Thinking).2 Very interesting attempts to define these processes in scientific terms have been also performed such as: “Axiomatic Design Theory” (Suh 1990, 1998), “General Design Theory” (analysed and discussed by Reich 1995), “Formal Design Theory” (Braha and Reich 2003). In this chapter we focus on “Axiomatic Design Theory”, which is an elaborated theoretical frame, based on well-­defined general principles and a precise characterization of given elements: customer needs, functional requirements, design parameters, process variables (see Section 3.1). During the last decades a lot of methodologies have been developed for analysing products and systems developments: Integrated Product Teams (IPT), the Program Evaluation and Review Technique (PERT), the Critical Path Method (CPM), and the Design Structure Matrix (DSM). In this chapter we employ the DSM since it allows us to look into the web of interdependencies which are created during interaction processes among many agents. Our analytical framework is based on some ontological assumptions. 1) The world we analyse is populated by interacting entities (“interactors”). 2) These latter are enmeshed within knowledge processes, which are the main focus of our analysis. In this perspective a basic epistemic assumption is that economic agents act as problem solver within the product development process, which unfolds as a set of multilayered problem-­solving activities, which in turn are developed relating to search activities in multiple spaces: design parameters, engineering characteristics, component and product properties (Sections 2.2, 2.3). The adopted theoretical framework allows us to examine the dynamic mappings between different spaces and sub-­spaces (Section 3.1) in order for changing networks of interdependencies to become fundamental components determining product complexity and organization structures (3.2). The high-­dimensionality of search spaces and the emergence of stable designs and organizational configurations lead us to discuss precise research strategies, elaborated in the literature: decomposability of the product development process and the evolution of modular organizations (Section 4). Then we tackle the basic problem of reducing the high dimensionality through searching for constraints. To this end, by means of a morphogenetic approach we investigate

The production process as a complex world   33 how interdependence and interplay between interactors can develop without bumping into the combinatorial explosion. In fact research strategies for decentralized systems are briefly discussed, in order to reduce the high-­dimensionality of search spaces (Section 5.1) and the means by which they can work, i.e. information packages, building blocks and protocols (Section 5.2). Subsequently we attempt to answer the fundamental question such as “ordering the world” and at the same time producing novelties. We pursue this goal through the concept of rule, which is based on an ideational kernel, that is, the result of dynamic mappings between multiple search spaces. The ideational kernel of this concept is analysed and defined as a result of dynamic mappings between multiple search spaces in Section 6. Section 7 proposes two explanatory principles for understanding determined properties of a world populated with interactors: order, complexity and novelties producing. The principles are compositionality and recursiveness, which are fundamental for producing and absorbing complexity in combinatorial spaces like human language.3 The generative4 nature of technological space is considered as one of the clearest examples of how these principles can help us to understand and explain the economy as a complex, dynamic and ordered world. 1.2  Suggestions from the literature Since Adam Smith’s analysis of division of labour within the pin factory, fundamental elements of firms have been focused on by a growing economic literature: firms’ boundaries and property rights (Langlois 2002), dynamic capabilities and productive knowledge, (Teece et al. 1997; Dosi et al. 2000), coordination and cooperation (Hoopes and Postrel 1999), innovation activity (Gavetti and Levinthal 2000; Fleming and Sorenson 2001), evolution of organizational models (Chandler 1977), complexity in terms of “depth” and “breadth” of firm functions, technology, market, management (Wang and von Tunzelmann 2000). Certainly Nelson and Winter’s book (1982) can be viewed as a kind of watershed, because since this book knowledge and organization are one of the main lines of research (Morroni 2006), as is apparent to all from the relevant strands of studies which have been developed: “resource-­based theory” (Wernerfelt 1995; Barney et al. 2001; Barney 2001), firms as problem-­solving and information processing systems (Radner 1993; DeCanio and Watkins 1998; Dosi and Marengo 2005). Economic theory, even within the evolutionary approach, has devoted less attention to the analysis of production processes, according to Winter (2005), while a huge amount of studies in engineering design theory and system theory has produced a sound body of concepts, analytical tools and simulation models applied to the theory of firms and technological dynamics (Anderson 1999; Dooley and Van de Ven 1999; Morel and Ramanujam 1999; Levin 2002, Antonelli 2007; Ciarli et al. 2008). However if knowledge is the focus of the analysis we stray from the standard economic theory, based on determined assumptions (Lombardi 2001), and we have to enter a “new world”, characterized by seemingly conflicting properties

34   M. Lombardi such as order and instability, high-­dimensionality and complexity of task environments, together with the tendency towards the emergence of stable configurations of products and organizations.

2  The economy as a world populated with interacting entities 2.1  Agents as interactors In order to tackle the question of how these features occur interdependencies between entities populating this world are the focus of our analysis. Thus our starting point is the ontological assumption that the world we are going to examine is populated by interactors, which are defined as “cohesive entities” which interact with their environment (Hull et al. 2001).5 However for the time being we can conceive the interactor as a group of functional related components, that is, parts able to execute different and coordinated activities in information processing. Beyond the ontological, precise epistemic assumptions help us to analyse the dynamics of this world and its properties: interactors are enmeshed within information flows and knowledge processes. This implies that interactors elaborate information and knowledge in many different ways, depending on the combination of “local” and “global” search (March 1991).6 Manifold combinations of different search activities are the results of how the two types of behaviours are adopted by agents within interaction processes. Indeed interactors can alternatively behave as Cartesian or stochast, in this way changing the composition of the population inhabiting this world. The evolving mixes of the two behaviours produce relational topologies, which are at the same time the source and the outcomes of exploratory activities developed in different search spaces. This is particularly important as the development of new products becomes the result of the systemic and dynamic nature of interactions in iterative cycles of problem-­solving tasks (Nightingale 2000). Changing geometry and relational topology then generate a coordination problem first of all in cognitive terms. Indeed as agents are involved in multiple activities it becomes fundamental to actualize what we call the “alignment of models of the world”, which are elaborated by interactors seeking to define demand requirements, commodities, product components, and so on. Interactions between agents and between the agents and the environment, which in turn is generated by their actions, foster a self-­organization process to the extent that structures appear even if agents do not have explicit models of the world. Thus from the perspective adopted here economic processes have to be analysed as collection of goal-­oriented and heterogeneous interactors, who modify individual and collective behaviour in response to changes in their task environment. 2.2  The production process as a global workspace The economy is populated by agents/interactors that we can distinguish in individuals and economic units unceasingly trying to solve techno-­economic

The production process as a complex world   35 problems, such as finding viable technical solutions to produce goods and services in order to satisfy changing needs. We can describe individual agents as vectors [bji, pji, eji], where each i symbol is an index related to choices which can be selected within different spaces of alternatives [Bi, Pi, Ei] by the j agent, that is, the ideal spaces of all possible beliefs, preferences and resource endowments. It is useful to think that every agent draws on the power set of components of these three spaces, in this way many people can share what they consider sound beliefs, what preferences they at least temporarily keep, what resources they hold. Agents are identified by a triplet of values, which synthesize the results of decision processes on the basis of the scanning of a kind of abstract global space for individual agents. This space is composed of sub-­spaces within which each person ideally searches for solutions to be matched to their needs. The same behaviour can be shared by different people as they can be characterized by a totally or partially identical triplet. The abstract global workspace then is the space of all variables belonging to three sub-­spaces (Beliefs, Preferences, Endowments), from which agents have to draw parameters to be coordinated in order to have a feasible representation of needs. This latter in turn has to be satisfied by means of a product. By focusing on the production side of social process economic units are collections of individuals who try out and seek out solutions to technical problems in order to produce artefacts. What is then economic production? Fundamentally it is a set of multi-­layered problem-­solving activities, on the basis of information and knowledge which producers have and make unceasingly evolve. From picking out a market opportunity to obtaining a product available for sale, a multidimensional activity is needed: making assumptions about technology, sharpening productive functions, tuning product parameters with demand requirements. From this point of view “a useful representation of a product” sees any product as “a vector of attributes (e.g. speed, price, reliability, capacity)”,7 viewed either as customer needs or product specifications (Krishnan and Ulrich 2001). The vector of product attributes is the result of vector transformations within a vector space. In this perspective each production cycle can be represented by a vector, whose components are basic coordinates in order to define specific components of the good to be produced. The congruence among multiple vectors, that is, the matrix transformation from n-­dimensional input to final output, can happen through linear sequences of n-­intermediate steps or overlapping and iterated cycles of transformations. The main point is that, in order to have a final product, vector transformations have to produce relatively stable product architectures (Ulrich 1995; Yassine and Wissman 2007) and at the same time they have to relate to unceasing processes of creating new information. The vector of attributes is the result of exploratory activities unfolding in different search spaces: design parameters, engineering characteristics, component and product properties, and so on. So we have sub-­spaces of a global workspace for producers, as they try to elaborate working solutions by means of mappings

36   M. Lombardi between different sub-­spaces. The potential complexity of manifold mapping activities is clear: between the different sub-­spaces so that a working product is obtained on the one hand, and on the other between the two global spaces (abstract global workspace and global workspace) in order to have a feasible product. The process becomes even more complex as the mappings change on the basis of interactions among individuals and between these and the economic units. Before analysing what we have called the Global workspace, that is, the set of all conceivable solutions (either known or unknown) to technical problems, it is worth highlighting some interesting points. 2.3  Issues for a world characterised by very high-­dimensionality Let us look into interactions and interdependencies among people and economic units exploring Abstract global space and Global workspace. First of all, variable choices on the one hand and searching activity on the other, together with reciprocal influences, foster dynamic processes based on micro-­ diversity, micro-­economic turmoil and uneven accumulation of knowledge. Secondly, the mappings between different multiple sub-­spaces imply that all types of agents are characterized by goal-­oriented decision processes8 either in choosing the good to be consumed or in organizing the production sequences so as to satisfy the user needs. In this perspective the great number of variables, that influence all economic agents’ choices, causes the very high-dimensionality of their decision space. It is important to emphasize that changing choices for every decision unit can beget the restructuring of mappings from one space or sub-­ space into another, according to the variations introduced by other agents, who change and adjust their behaviour in reacting to signals and input stemming from operating contexts. In the same way people and organizations, trying to improve their objectives in consuming or producing, are committed to an endless sequence of choices; each of them can be represented as a combination of values, picked up within the Abstract global space or the Global workspace. Scanning and going through the branching process can occur in different ways, depending on the behaviour of agents committed to multilayered problem-­ solving activities, that is, their propensity to search within the neighbourhood of their current practices and of what they know (“existing knowledge space”) or looking at far-­away abstract territory. So problems arise: how does the congruence between sets of parameters drawn from multiple spaces occur and what factors affect it? The starting point of the analysis is the production process as a global workspace which exhibits a general property: a potential very high-dimensionality, because the number of different independent variables affecting the output of the system can be very high, depending on the degrees of freedom existing in each space or sub-­space. In other terms, the dimensionality depends on the constraints imposed on the capacity for modification of the goal-­oriented individual or collective behaviour in response to change in the environment. The constraints reduce the number of dimensions and at the same time the width of search spaces.

The production process as a complex world   37 Thus how is it possible that an ordered world9 happens even if it stems from an incredibly vast set of alternatives and it is hit by unforeseen perturbations? If production processes are potentially infinite sets of tasks to be defined and related to each other, a basic question to be answered is: how do we reduce the number of alternative configurations in a search process? This can be addressed if the focus is on entities subject to an endless series of changes of their organizational micro-­states, emerging from low- and high-­ frequency interactions, as the decomposition of production sequences produces the formation of different forms of hierarchical and nearly decomposable systems (see Section 5).

3  The global workspace for producers 3.1  An unceasing dynamic mapping How do we reduce the dimensionality of the global workspace and at the same time keep an endogenous source of potentially infinite variety? We must start from an adaptive economic system, composed of a population of heterogeneous interactors: they are fundamentally goal-­oriented and interact with each other. This characterization of decision units implies that production processes are fundamentally affected by unceasing elaboration of information, beyond matter and energy flows; what’s more interactions are primarily informational and information flows continuously generate processes and new phenomena like coherent vectors of attributes (as we define products, see Note 7). So let us analyse how they unfold thanks to drawing on the literature on engineering design, which allows us to introduce abstract terms, useful in representing how the mix of exploiting and exploring activities develop together with Cartesian and stochast behaviours. First of all we consider the product development process as unfolding within the Global workspace, which spans from defining customers or market needs to the delivery of a product with precise characteristics. The product development process is composed of a set or sequence of activities conceived and executed in order to attain a vector of attributes. By looking into the product development process we can highlight that it implies the unfolding of an engineering process, which consists in understanding either functional requirements (FRs), which in turn are defined “as the minimum set of independent requirements that completely characterize the functional needs of the product” (Yassine and Falkenburg 1999: 225), and design parameters (DPs), which “are the key variables that characterize the physical entity created by the design process to fulfil the FRs”. In more formal terms the Axiomatic Design Theory (Suh 1990, 1998) anal­yses the mapping process between three domains: 1) Customer needs (CNs), or demand requirements, 2) FRs and constraints, which are effective or acceptable solutions to problems to be solved in order to satisfy CNs, 3) DPs, or design variables (physical parameters, parts, assemblies, modules) to be chosen compatible with constraints and coherent with FRs (see Figure 2.1).

38   M. Lombardi After defining DPs it is important to identify the Process Variables (PVs), that is, resource flows, which can be based on the use of existing processes or the creation of new ones. The situation gets more complicated if we introduce iterative cycles of problem-­solving activities and tasks, with multiple feedback loops (not depicted here. See Figures 2.2 and 2.3). The mapping process between the three domains can be described in a mathematical space by means of vectors representing attributes and properties at different levels: the FR vector expresses design goals, the DP vector represents design variables and the PV vector synthesizes the working flows. The Axiomatic Design Theory starts from the “design equation” {FR} = [A]{DP}

(2.1)

which denotes the relationships between FRs and DPs and [A] is defined as the “design matrix”. The elements of [A] are of the form δFRi _____ Aij = ​     ​  δDPj

(2.2)

indicate how changes in DPs influence FRs. Aij can be constant (for linear design) or functions of DPs (non linear design). For the n-­DP design Eq 2.1 can also be written as {FR} = SAijDPj

(2.3)

Another equation represents the mapping between DPs and PVs in the Global workspace {DP} = [B]{PV}

(2.4)

Mapping

Mapping

Mapping

{CNs}

{FRs}

{DP}

{PVs}

Customer domain

Functional domain

Physical domain

Process domain

Figure 2.1  Suh’s Axiomatic Design Theory developed (source: based on Suh 1999).

The production process as a complex world   39 An important point is: the rows in the design matrix represent the coupling of FRs to DPs in the sense that the Aij indicate how changes in FRs are linked to changes in DPs: “When the design matrix that relates the {FRs} vector to the {DPs} vector is diagonal, the design is defined as uncoupled. When it is triangular, it is a decoupled design. All others are coupled designs” (Suh 1999: 118). Following the engineering design literature we can divide the design process and the product development process into more elemental components insofar as it is necessary to further decompose the PVs, as search activities are developed to the extent that the required vector of final attributes is obtained. Decomposition means to enter multiple pathways, which are scoured during many problem-­solving activities, depending on the nature and the typology of the product. In fact the engineering design can be further divided into specified tasks, defined by means of specification “as nominal value plus/minus a tolerance value” (Yassine and Falkenburg 1999: 228). Design variables are at the same time input and output of the same task, while interface variables are the output of other tasks. Since the product development process is fundamentally fed by the mapping process from CNs vector space through PVs vector space, we must examine how it evolves in order to obtain a “recipe for producing a product” (Browning et al. 2000), keeping in mind that “In design, assumptions about the exact knowledge are almost never true” (Yassine and Braha 2003: 165). So information exchanges across many tasks and information structure analysis are essential insofar as different relationships among tasks (either in design process or production and delivery of a product) can occur depending on the needs of information flows and input. Let us examine how the product development process unfolds on the basis of information exchanges across tasks and people executing them. One important point is the type of good that has to be produced, and how many elements are compounded in order to attain it. In general, different types of goods can be conceived and for the simpler ones these relationships are “usually anecdotes and heuristics”, while for the more complex a lot of elements are required on the basis of the complexity of the product and the related activities which are needed at multiple levels, i.e. “trying, analyzing, evaluating, testing, experimenting, demonstrating, and validating” (Browning et al. 2000). All of them are sources of information influencing the ways tasks are done and how people elaborate knowledge. Following this perspective the product development process is a hierarchy of nested activities, which can develop according to how interactions among elements composing the product evolve. In general terms, they can be parallel (or concurrent), sequential (or dependent), coupled (or interdependent). We must emphasize the implications involved: given the possibility of a variable number of nested hierarchies and the different degrees of interdependency between them at multiple levels, for example subspaces belonging to FRs and DPs vector spaces, the Global workspace is once again a set of search sub-­ spaces, each of them subject to exploratory activity with variable degrees of freedom or constraints.

40   M. Lombardi In order to investigate both it is necessary to discuss three main questions: 1) product complexity, 2) the coupling of product architecture to organizational structure and 3) how components are combined in order to obtain determined final vectors of attributes. 3.2  Changing networks of interdependencies Let us examine each of the listed issues. 3.2.1  Product complexity Regarding the first: Product complexity has three main elements: (1) the number of product components to specify and produce, (2) the extent of interactions to manage between these components (parts coupling), and (3) the degree of product novelty. Variations in product complexity are driven by a number of factors such as choices in performance, technology, and product architecture. (Novak and Eppinger 2001: 189) Product architecture, defined as “the arrangement of functional elements into physical chunks that become the building blocks for a product or family of products” (Browning 2001: 294), is tightly linked with complexity. In order to define the right arrangement the literature on engineering design, especially since the 1990s, suggests a sequence of steps: 1) decompose the system into elements; 2) understand and describe the interactions between the elements (in other terms how they are integrated); 3) analyse potential reintegration of the elements via clustering (integration analysis) (Pimmler and Eppinger 1994; Browning 2001). As for the second, a significant field study, analysed by Gulati and Eppinger (1996), shows the coupling of product architecture to organizational structure, in the sense that there are dynamic relationships between the evolution of organizational capabilities on the one hand and the structures and the type of architectural design on the other. In fact the decomposition choices affect dependencies, interactions, functional differentiation, competences development and managerial decision processes. Technical capabilities are linked to the layout of product architecture, because the arrangement of different pieces influences the pattern of information exchanges, the pathways towards specialization and their changes. At the same time the interdependence between product architecture and organizational design means that “product architecture influences the way firms learn” (Yassine and Wissmann 2007). It must be stressed that in the design process it becomes fundamental to partition design parameters into two categories: hidden information and visible information (Baldwin and Clark 2000). The former refers to cognitive activity developed within organized chunks; the latter refers to the exchanges among chunks or modules and how it flows through the interface specifications. One

The production process as a complex world   41 important implication of this view is the statement that there is a design hierarchy, with “three levels of visibility: global design rules (at the top); locally visible intermediate level design rules (interface rules); and intramodule design rules (at the bottom of the hierarchy)” (Sharman and Yassine 2004: 47). In other terms, hidden and visible information are respectively connected to local and system information within the product development process. However we should bear in mind that the way in which system information is organized matters, because the morphology of the activities network within the product development process is related, on the one hand, to the information exchange with the environment and, on the other, to inner information structure. Indeed, the congruence between parameters belonging to the four domains (see Figure 2.1), the relationships among them and the demand vector are conditional on the web of interdependencies among phases, sub-­phases, activities, teams and organized units. This brings to the fore the importance of product architecture. For example, different typologies of goods can follow different patterns between two extremes. On one end there is a kind of “waterfall model” of information flows10 involving stronger constraints along the process and implying tightly coupled organizations (teams, firms, economic units in general). On the other end there is a very loose coupling of components, where larger degrees of freedom beget horizontal and more open information flows. Thus information structure exercises the role of “embedded coordination” (Sanchez and Mahoney 1996), which is worth analysing. 3.2.2  Organization architecture Clearly we have many forms of coupling and then different forms of learning, which reflect the product architecture, the connected ways of creating and transmitting knowledge within and among organizational units. The literature has analysed “three alternative approaches to creating knowledge and transferring information in product design and component development processes: ‘traditional’ sequential development, overlapping problem solving, and modular product development” (Sanchez and Mahoney 1996: 68). The first is characterized by the precise specifications of interfaces, which allow the sequence of components to unfold, but its information structure bumps into “breakdowns, losses and delays” as information feedbacks are lost, incompletely absorbed, or imperfectly transferred within the sequence. However in the case of an evolving product architecture the incompleteness and indefiniteness are unavoidable properties of the information structure during the product development process, so that important managerial decisions have to be taken and imposed by means of “tightly coupled organization structure” (op. cit. p. 69). The second approach, based on staggered but overlapping stages and problem-­solving activities, has some better properties, like improved and quicker information flows, together with less information loss and an evolving information structure. The last attribute requires an “intensive managerial coordination” as the tasks are incompletely defined. The third is the modular approach and it is characterized by a

42   M. Lombardi complete information structure, based on well-­specified modular architecture and fully specified interfaces. Loose coupled systems of autonomous and concurrent components allow distributed learning through improvements at the module level (Mikkola 2003, 2006). Regarding the question of how components are combined in order to obtain determined final vectors of attributes, let us deepen the analysis by considering elemental components of the product development process. We are suggesting here that a better understanding is possible if we conceive them as a variable number of problem-­solving stages within a sequence based on variable constraints and degrees of freedom in defining CNs-­FRs-DPs-­PVs, depending on information flows and the interactions either among tasks to be executed or people involved. Interdependence between elements become fundamental and the structure of interdependence must be investigated, by bearing in mind that product development frequently implies overlapping tasks, procedural and repeatable activities, iterations and feedback loops. Iterations necessarily occur: 1) when, after transforming preliminary input, new information comes from overlapped tasks; 2) if the reworking of tasks generates new inputs for others already executed; and 3) if the congruence between “information pieces” or parameters is difficult to obtain and then practitioners try again and again. Indeed a lot of empirical and theoretic research shows “the inherent, iterative nature to the design process” (Braha and Bar-­Yam 2004), but a similar statement can be formulated with regard to the entire product development process for every type of product, be it integrated or distributed product development process (Browning et al. 2000; Yassine and Falkenburg 1999; Smith and Eppinger 1997).11 Sequential and parallel iteration, feedback loops as information cycles between multiple operations or tasks, highlight the fact that the product development process and the production process is a sequence “of interdependent individual stages (called operations)” (Buenstorf 2005: 222). In this view the entire process can be conceived and analysed as a variable network of interdependencies, which evolve according to how the product development as a discovery process unfolds: problem-­solving activities (nested, sequential, parallel) and the related search processes within different sub-­spaces are knowledge-­based as information flows are organized and knowledge is elaborated in executing tasks and operations. 3.2.3  Networks of interdependencies It is worth stressing different approaches which have been proposed in order to analyse and model networks of interdependencies within the product development process. One of the most famous is the Dependency (Design) Structure Matrix (DSM), which is a compact representation of the information structure of a design process, but it has been applied to the entire production process and connected with the organizational model of firms. The DSM describes information-­based relations among components, which can be product sub-­ systems, modules, tasks, teams of people involved in designing (Figure 2.2:

The production process as a complex world   43

Figure 2.2  A binary DSM unpartitioned.

Matrix A, adopted from Eppinger et al. 1994. For an overview of DSM models see Sharman and Yassine 2004). The elements of a generic matrix represent activities or components, while the marks indicate parameters (information pieces) required for tasks to be worked. Groups or blocks of marks above the diagonal depict coupling and backward input: they point out that some components need information available only after being executed; thus the arrival of new information (a successive task in the sequence) must be considered. Marks below the diagonal represent the feedforward of information to subsequent tasks. Information-­based interactions and their evolution are then represented, allowing the system analysts to construct models of the activities, problems and teams. In this way some measures can be introduced: dependency strength, volume of information transferred, variability of information exchanged, probability of repetition, impact strength (Yassine 2004). Reordering algorithms12 generally produces sequenced tasks (Figure 2.3: Matrix B, adopted from Eppinger et al. 1994), which can be series of single, parallel tasks, collected groups or coupled blocks depending on the information exchanges, so synthesizing the information structure and the web of interdependencies within the product development process. As the reordering of rows allows us to find interesting components of the DSM (coupling of elements) it is worth going into them and finding how they work and interact, by developing a “molecular analysis” of DSM building blocks. From the literature on engineering design a sort of sequence can be drawn, starting from the parts as “the smallest possible decomposition”, then the chunks or assemblies of elements, modules or modular clusters, where chunks are organized so that “all

44   M. Lombardi

Figure 2.3  The binary DSM, partitioned to represent a sequence.

the parts within the module possess the same relationships with each other and with parts outside of the module” (Sharman and Yassine 2004: 42). The analysis leads us to stress the importance of the decomposition of product development process, viewed as a variable mix of assembling and disassembling, coupling and grouping of elementary parts, depending on information exchanges. The suggested analytical perspective is convergent with the approach centred on “knowledge-­based decomposability” (Buenstorf 2005: 235). The main point of our theoretical frame is the decomposability of the production process tightly linked to knowledge processes, which can be at the same time extremely fluid and condensed. The goal of our research programme is to understand how such conflicting properties are the result of a peculiar dynamic and simultaneously feed it. To this end our aim is to develop the proposed theoretical frame sharpening the intertwined processes of decomposition and knowledge production; in other terms, the way production processes decompose in relation to changes in the state of scientific and technological knowledge.

4  The decomposition of the product development process: beyond the modularity These intertwined processes unfold within the technological and scientific space, viewed as discovering space where problem-­solving activities are pursued and

The production process as a complex world   45 performed. As we have underlined in Section 2, it is exactly within this evolving universe of knowledge that the process we are setting about to analyse is conceptualized as a sequence of searching activities in multiple sub-­spaces: demand requirements, functional requirements, design parameters, process parameters, module clusters, chunks, parts and operations. Frequently there is a vast number of elements to be combined in order to obtain coherent vectors of parameters, sorted out within different search domains. The convergence in solving problems belonging to different search spaces is realized even if the so-­called “design churn”13 is not rare, that is, the multiplication of problems as far as solutions are sought and found. In fact the “churn phenomenon” is caused by a set of factors: incompletely known interdependencies among activities; inadequate forecast of feedbacks, so that instability can arise; delayed iterations which generate hidden information for other components; time phase-­displacement between the effects of local interactions within a team or an operation and its output flowing into other chunks of the global sequence.14 Two general points are worth emphasizing. The first is that as search activities are developed and information exchanges among agents (individuals, teams, organizations) occur, high frequency and low frequency dynamics (Simon 1962) unroll depending on the type of interactions among them. High frequency interactions produce unforeseeable results, which is hidden information viewed from the outside of the dense set of interacting units. However this information must be parameterized in some way in order to be transmitted to other interactors, thus fostering search activities. In this way low frequency interactions are possible which allow exploring other and possibly distant search domains, insofar as long distance units can exchange new information. The second question consists in organizing information flows so that other elements or groups aim at obtaining congruence among parameters at different levels. In this way the decomposition of working flows is directly connected with the organization of activities, while two interesting aspects deserve to be underlined: on the one hand, higher frequency interactions transform incoming information in “local” accumulation of knowledge, conceived as a set of people’s expectations, beliefs and prior learning. On the other hand, information is parameterized knowledge, synthesizing the regularities extracted from data concerning task environment and which is communicated afar. This can be useful to clarify the distinction between data information and knowledge. Data are “discernible differences in physical states of-­the-world – that is, states describable in terms of space, time, and energy” (Boisot and Canals 2004: 46). Information is the result of human activity in extracting regularities from data. Finally, knowledge means significant regularities according to the agent receiving it, in other terms the elaboration of it in relation to the task environment. Data, regularities sorted out in the data (information) and regularities interpreted within a context (knowledge) are constantly elaborated, while agents are involved in exploiting the results of previous learning or in exploring

46   M. Lombardi search spaces. The never-­ending accumulation of knowledge and its parameterization is an iterative process, which is at the same time rich in potential feedback loops and the cause of reworking of tasks and operations, as hidden and visible information are unceasingly produced and transferred. Fundamental issues then are: 1) what mechanisms are able to allow the communication; and 2) how the matching of parameters results from search activities in different spaces. The preceding analysis leads us to directly tackle the problem of the degree of decomposition of the process we want to study. This problem has been expanded by a huge amount of literature from different points of view. For example the engineering perspective is adopted by Ulrich (1995), who investigates the modular architecture starting from “a one-­to-one mapping from functional elements in the function structure to the physical components of the product”. The concept of the module has been sharpened by Yassine and Wissman (2007: 120) “as a grouping, either physical or conceptual, of architectural elements that often result from the interface definition process”. Baldwin and Clark (2000) have further developed this research line by defining precise interface parameters and protocols, called design rules. Modularity begets advantages, such as the reduction of costs, thanks to reusing elements and flexibility to environmental changes. It also has its drawbacks insofar as it is based on the assumption of a complete knowledge of the system and of the interrelationships between the elements belonging to it. Indeed systemic innovations and global effects of environmental turbulence are hardly captured by local accumulation of knowledge within modules. Notwithstanding the described aspects, we think the focus on the degree of decomposition should lead us not to relinquish it, but to opt for reinterpreting it, with the intention of keeping interesting analytic properties and adding other ones, which help us to understand essential elements of the dynamics between multiple search spaces. One promising research line in this direction is relaxing the strong requirements modules have to satisfy from an engineering point of view, that is, completely specified interfaces, and at the same time permitting the acquisition of some important properties like flexibility, adaptability and robustness of components. Another significant characteristic to be pursued is not only variable degree of decomposability, but also the evolution of forms of decomposability. To this end the “knowledge-­based decomposability” (Buenstorf 2005) can be drawn on with some changes, mainly focused on how information exchanges occur at different levels of organizations and within the different sub-­spaces belonging to the Global workspace. We must enlarge the perspective by expanding the concept of modularity proposed by Schilling (2000: 312), in that it refers to a “continuum describing the degree to which the system’s components can be separated and recombined” according to different degrees of couplings and matching between components.15 Aoki and Takizawa (2002) and Aoki (2004) define the module as “a unit of a system within which elements are strongly interrelated to one another, but across

The production process as a complex world   47 modules they are relatively independent”. In general, the decomposability (of natural objects, organisms and its components, products, generic systems) implies that the whole can be partitioned in sub-­units, which in turn can be further decomposed. Modularity depicts a situation where there are partitions which are semi-­independent from each other, while the internal ties are more intense and information exchanges are more frequent than the ties and exchanges among sub-­units (see also Simon 1962). Understanding the formation and functioning of modules within near-­ decomposable wholes requires tackling important issues. First of all, we must analyse how the wholeness lasts in the light of different types and levels of interactions. Secondly, how convergent processes among modules and sub-­modules emerge. Thirdly, how some properties of components and of the whole evolve in relation to changes of interactions. Lastly, we should answer the question of in what way degrees of modularity change over time. The situation gets more complicated and very difficult to understand if we pay attention to the different environments with which each module interacts. So system analysts have to delve into multi-­level processes with two essential properties: variation and stability at global and local level. How can all of this occur? In the view here suggested, one starting point necessarily is the viewing of modules as adaptive entities, with some basic characteristics: 1) internal cohesiveness, 2) variable composition, as the effect of dynamic coupling of multi-­ level problem-­solving activities, 3) changing connective geometry with other adaptive entities, 4) strong interactions with different types of environment, 5) robustness, viewed as “the preservation of particular characteristics despite uncertainty in components or the environment” (Csete and Doyle 2002). Let us tackle the issue by looking into the dynamic matching between multiple spaces. It is quite natural to think that the architecture and the knowledge for simpler products does not raise particular problems, because the entire search space and each sub-­space have low dimensionality, measured through the number of different independent variables affecting the output of the system, that is, the degrees of freedom existing at each level. In these conditions the description of the whole is fully tolerable by the human mind. The dimensionality of the search spaces grows as internal differentiations are introduced and developed within the product development process, when new elements are discovered (technological innovations) or designed (organizational solutions), original ideas are invented and tested, and so on. Thus in the Global workspace, products vary between two extremes: zero or low modularity on one side (integral product architecture), full modularity on the other. The former occurs when many-­to-many mappings evolve according to reciprocal influence among elements belonging to search spaces (coherent parameters to be chosen in order to obtain a good), so that each interacts with any other at different levels. The latter depicts one-­to-one mappings from different domains (see Figure 2.1). The choice process of parameters in each domain has been previously depicted as a multidimensional space;

48   M. Lombardi indeed it is the Cartesian product of the space of values which problem solving activities have to sort out CNs × FRs × DPs × PVs. As the dimensionality of spaces grows, near-­decomposability becomes necessary from a cognitive point of view in order to have a tolerable description of reality, while new phenomena appear such as the partitioning of spaces and the emergence of aggregations of elements. These phenomena led scholars in biology and engineering fields to look at connectivity relations in order to explain modular properties (Rasskin-­Gutman 2003) and to understand their formation and evolution. Let us examine the first. It is worth focusing on one basic peculiarity: the modular structure viewed from a bottom-­up perspective is but a combination of values representing elemental components. So an important point to be investigated is the generative nature of spaces of values during the problem-­solving activities and the linked question of how elements are combined in order to beget viable solutions. Generative nature means that the spaces are created by combining and recombining basic components. Thus the tightly linked questions to be answered are: 1) what are these elemental ingredients? 2) How are they combined in order to beget viable solutions? In order to study the dynamic matching between elements belonging to multiple spaces, it seems particularly promising to take up some suggestions from the morphogenetic approach, defined by Archer (1995) as a theoretical perspective where interdependencies and interplay between interacting elements become central units of analysis. The adopted approach allows us to describe and explain emergent properties as “relational, arising out of combinations” (op. cit. p.  9). We want to sharpen and enrich this approach understanding how it is possible that a final vector of attributes results from a potentially infinite number of problem-­solving activities.

5  Morphogenetic approach: in searching for constraints on search spaces without blocking problem-­solving activities 5.1  Research strategies in decentralized systems The dynamic matching between parameters within the Global workspace can be a formidable challenge for whatever interactor, insofar as the increasing degrees of freedom guide the exploration of different domains (CNs, FRs, DPs, and PVs) by considering interactions among all entities at different levels: the number of combinations could soon generate a combinatorial explosion of alternatives. In this case the very high-­dimensionality of the search space seems conflicting with the goal of attaining a viable solution in an acceptable lapse of time.

The production process as a complex world   49 In order for the dynamic matching to end with an acceptable vector and not an open-­ended achievement, some constraints are needed, which limit the alternatives and at the same time allow the development of research trajectories. The emergence of constraints can be understood if we focus on near-­decomposability as a fundamental direction of research and on its three tightly connected elements: webs of interdependencies, information flowing within them and network thinking, in the sense that the aforementioned dynamic matching is realized through the adaptive behaviour of interactors. The suggested perspective allows us to frame the raised issues from the point of view of information processing within decentralized systems. Indeed modular architecture and variable degrees of modularity are expressions of fundamental properties of the economy as a distributed autonomous system, where components are conceived as interactors which process information in adaptive ways. Information processing in decentralized systems has long been a main topic across disciplines (engineering, biology, computer science), as they strive to understand how complex systems behave.16 On the basis of a huge theoretical and empirical literature, four general principles of information processing in decentralized systems have been proposed (Mitchell 2006: 1208–10): (1) The dynamics of the relational topology begets and transfers global information. (2) “Randomness and probabilities are essential”, in the sense that they allow multiple sources of information and knowledge. (3) “The system carries out a fine-­ grained, parallel search of possibilities”, as within the system many exploratory pathways are simultaneously scoured at different levels and frequently exchanged.17 (4) “The system exhibits a continual interplay of bottom-­up and top-­down processes”. Regarding the three domains of the Global workspace, that means that the elaboration of a design is subject to a never-­ending and variable mix of prior beliefs and expectations on the one side, information gathering and exchanging on the other. This general perspective just outlined is receiving growing attention also in other scientific fields. For example, for more complex engineering projects evolutionary engineering seems particularly attractive to a growing number of scholars (Bar-­Yam 2002, 2004; Magee and de Weck 2003). Indeed, instead of traditional incremental design, based on iterative incremental changes in a precise sequence of steps (such as concept, design, specification, implementation, testing and manufacture), it would be better to develop parallel and quasi-­ autonomous “explorations of product improvements by different companies in a market economy” (Bar-­Yam 2002: 6). This can occur because of the diffusion of networks of interdependencies and the necessity of adopting a multi-­level perspective. So a “multiple iterative parallel evolutionary strategy” has been proposed (op. cit. p. 17), based on determined principles: 1 2 3

It is important to divide tasks into interdependent parts. There are tasks of different dimensions. Adaptability and coordination should be carefully balanced with integration, as both have costs.

50   M. Lombardi 4 5

Central control and detailed planning are effective only for not too complex systems (products). “Parallelism/redundancy provides functional security and enables learning”.

The last point allows us to highlight a significant issue from the morphogenetic approach. The product development process and production processes can show a very large multiplicity of configurational forms: n-­entities like “autonomous” micro-­worlds; networks of cycles of interactions through self-­reinforcing feedbacks (hyperstructures spread worldwide). Indeed cycles involve a sequence of inner phases, each of which can take place either in a dense and compact form (unitary form), or can be decomposed into relatively independent sub-­processes which, in turn, may be further subdivided. In this scenario product development is the complex result of decomposed sequences distributed over dispersed units, while coordination problems emerge and have to be resolved so that essential properties are acquired: 1) redundancy, 2) degeneracy,18 3) robustness or adaptability (Carlson and Doyle 2003; Kitano 2002). For example, fine-­grained systems, which are characterized by “parallel terraced scan”, exhibit adaptability and quicker reactivity to sudden changes of the environment as they enable the information propagation among components. So we meet here a peculiar combination of invariance and stability, which is preserved notwithstanding shocks and “waves” of novelties. But two fundamental questions arise: how information propagates and what is maintained in spite of changes, unless the changes are so radical that everything is deeply affected and completely transformed. In order to tackle them we need to “decompose” information and knowledge processes. 5.2  Information and knowledge processes for networks of interdependencies: protocols and building blocks In answering the two questions some key points have to be emphasized. In the first, the flowing of information through layered networks of interdependencies, feedback loops and multiple problem-­solving activities are possible because shared signalling devices exist, which communicate regular and significant alterations of the physical states-­of-the-­world. These devices are the protocols, which precisely enable information exchanges either in engineering or biological systems (Csete and Doyle 2002), so providing means for absorbing shocks and evolving. Protocols let modular configurations work and at the same time parameterize smaller and bigger changes occurring within decomposable entities. In the same way information flows can go far wherever they are accepted and used in order to describe events or ongoing processes. Decomposition and aggregation of units are then possible depending on the adaptation to protocols. The second question, that is, what is preserved notwithstanding changes, must be carefully treated. It can be formulated in the following terms: how are stable

The production process as a complex world   51 models or configurations of product attained in the product development process and production processes. For simpler products or goods with lower degrees of complexity it is quite easy to hypothesize that convergence towards a well-­ defined final and simplified vector of attributes will be highly probable and fairly stable, provided that stochastic shocks starting from within the task environment are not very frequent. Similarly it can be said regarding the waterfall model of information flows, connected with reduced degrees of freedom along the product development process. Top-­down definition of parameters belonging to different search spaces implies the prevalence of a stable design configuration, while determined constraints restrict the space of all possible solutions at different levels. In both these archetypes it is clear what is maintained and how this happens: there is an evident and shared design/model, which constitutes at the same time the starting point and the goal of the whole process. The situation is completely different when the growing complexity of products begets structural biases towards high-­dimensionality of the Global workspace, in this way leading to decentralized and fine-­grained systems. Protocols are particularly useful but they are not the essential cornerstone. On the contrary, their function needs to be founded on something else, enabling communication among so many entities to happen, even if they meet difficulties and operate within environments characterized by uncertainty. We have to answer a fundamental issue: how is it possible to comply with an imperative such as “ordering” the world and at the same time producing novelties, which could in some cases radically change it? The research line based on the study of combinations of different elements (parts, pieces, chunks, modules) seems particularly promising. In tackling these question our starting point is the generative nature of spaces of values. Indeed the dynamic matching between multiple spaces stems from the exploration either of unbounded sets of all conceivable ideas or unlimited domains of every conceivable solution, subject to never-­ending tests on the basis of well-­accepted criteria. From the elaboration of abstract ideas to the actualization of them in a good there is vast sea of activities and objects (ideal and real) which have to be combined in the process of sequencing and scheduling activities. The perspective adopted here leads us to focus on the couple of components of all processes: knowledge and combination of building blocks, which in turn have to be picked out. Therefore it seems particularly appropriate to view knowledge as “a complex structure of ideational kernels and the connections among them” (Carley 2002: 7258) especially when we want to analyse processes like the product development process, to which we aim at extending the representation, proposed for the production processes as a sequence of operations,19 which must be executed in order to obtain a product able to meet the demand requirements. As it is developed through attaining solutions to problems, the basic building blocks of knowledge are well-­connected information pieces, which help to analyse and efficaciously treat recurrent situations and problems. These building blocks are information packages with an intrinsic property: they

52   M. Lombardi are useful in different contexts and are transformed on the basis of the arrival of never-­ending information flows. In other terms they are rules. So our analysis has reached an important conclusion in treating the two issues (how to reconcile settling down in stable configurations and novelties producing): adaptive information processing by interactors is crucial, based on conceiving new combinations of basic building blocks or information packages, which are unceasingly generated through dynamic matching between elements belonging to multiple spaces of values. The concept of rule is largely employed in economic theory (Dopfer and Potts 2008) and rules play an important function within the research line centred on agents with purposeful behaviour based on programmes. These latter constitute pre-­arranged information, or conjectural knowledge about the world (Vanberg 2002: 15),20 which is unceasingly changed depending on encoding information acquired by interacting with the environment. Likewise the crucial function of rules is analysed in Geels (2004) and Geels and Schot (2007), where a “rule-­ based model of action and a multi-­level perspective are proposed”. The theory of design also assigns an important role to the technological rule (van Aken 2005a: 389) and in the new product development process rules are applied and changed as field-­tested and grounded solution concepts21 are found, for distinct patterns of technological evolution to emerge. If we want to examine the dynamic matching between different search spaces a crucial point deserves to be honed. We are referring here to rules as the building blocks of competencies, either in the individual or the collective view (firms, organization), as we conceive them as packages of information processing devices or cognitive modules (Vromen 2004a, 2004b) which are formed, grouped and decomposed according to the evolution of information and knowledge at different levels. The developed analysis induces us to consider the technology more deeply. Very useful insights to this end can be drawn from some recent papers (Arthur and Polak 2006; Arthur 2007), where technology is viewed as “combination of executables”, which are means fit for fulfilling a purpose or executing a function. Technologies are sets of “components, or assemblies, or subroutines, or stages”, each of them executes a function or pursues a purpose. Thus “a technology is organized in a loose hierarchy of groupings or combinations of parts and subparts that themselves are technologies.” (Arthur 2007: 277). From this point of view the invention assumes an interesting property: the recursiveness, that means nested building blocks which “repeat down to the fundamental level of individual components” and “it repeats until each challenge or problem (and sub-­ problem and sub-­sub-problem) resolves itself into one that can be dealt with using existing elements” (Arthur 2007: 282–3). The discussed concepts of rules and technology as a nested hierarchy with many levels, depending on the combining of parts and pervasive dynamics of knowledge, seem particularly useful in tackling the main issue of this chapter, i.e. how an ordered world is possible such as a product, viewed in terms of a vector of attributes.

The production process as a complex world   53 Let us summarize some partial results so far. At the end of Section 4 we have emphasized the importance of the generative nature of combinatorial spaces to be explored in order to discover coherent vectors of parameters. In Section 5 a precise research line has allowed us to highlight some basic ingredients of our theoretical frame, that is, building blocks and protocols as devices enabling information exchanges. Rules as elemental components of technology with its interesting properties (recursiveness and nested hierarchy) are the points from which it is worth starting for the last part of this chapter, which tries to answer three questions: (1) how are they formed and evolve? (2) How are information and knowledge created and spread? (3) What are the ideational kernels of rules?

6  Rules as outcomes of dynamic mappings between multiple search spaces Searching for the ideational kernel involves starting from what can be considered the “elementary particles”, which are precisely the ideas. Indeed the starting point logically is the nature of technology which is a combinatorial idea space (Olsson 2000, 2005; Olsson and Frey 2003), as we realize that it is the “universal set of all possible technological ideas in the past, in the present and in the future” (Olsson 2005: 40). Let us call the idea space Ω-space, within which all conceivable ideational kernels about the world are allowed. Given the evolutionary bias of human beings as human beings are “patterns seeking” and patterns thinking (Margolis 1987; Tooby and Cosmides 1992), the need for ordering the world leads them to create two types of knowledge: the epistemic base or propositional knowledge and useful knowledge (Mokyr 2005). The first type of knowledge is composed of natural regularities, which act as cognitive ordering devices, enabling humans to direct the searching process of solutions to problems. Indeed the need for constructing an ordered world stimulates humans to discover principles. These can be defined as constraints able to reduce the high dimensionality of the search spaces, in this way helping them to develop knowledge by improving research domains and by exploring other fields. We call this set of principles K-­space, which is a sub-­space of the Ω-space. It must be underlined that all such constraints are not fixed, to the extent that they are synthetic representations of an incompletely known world. Thus the epistemic base is subject to changes depending on the information accumulated, produced by enrichment processes and changes of inputs related to physical states of the world and to the information exchanges among interactors. Let us discuss the second type of knowledge. Never-­ending problem-­solving activities produce cognitive and operational outcomes, which constitute the useful knowledge or technology as it has been called by Mokyr (2005). Technology is the term which precisely sums up this second sub-­space belonging to the Ω-space. We call it λ-space, which in turn can be decomposed into recurrent phenomena within the world around us (let us call it ρ-subspace) and into

54   M. Lombardi prescriptive knowledge or techniques (τ-subspace), that is, “a set of instructions on how to produce goods and services”. Recurrent phenomena and set of instructions are “useful knowledge”, because they allow us to find different ways of satisfying human needs and they are the result of unceasing mappings from either Ω-space and K-­space to λ-space.22 The map correlating varying components of these spaces could be at the same time one-­to-one or many-­to-one in both directions, depending on historical contingencies and then countless ever-­changing factors. It is particularly important to underline that the τ-subspace is the set of feasible techniques relating to the epistemic base: this means that further constraints are introduced, which affect the activity – productive and explorative at the same time – just as propositional knowledge and prescriptive knowledge must be understood, learned and interpreted in the light of the task environment. For example, it is fundamental to know chemical laws and those relating to the effects of combination of substances, but this knowledge must be continuously enriched by information packages generated by acting at the shop level within a given factory. The proposed theoretical framework stresses the unceasing mappings between sub-­spaces belonging to multiple search spaces. These mappings beget information and knowledge flows, which are synthesized by means of cognitive units, in other terms packaged information and knowledge as result of combinatorial activity. The rules are precisely the outcomes of these mappings, which are embedded in individuals and organizations. Given their nature, rules have some significant properties: 1) they evolve, depending on the mix of exploitation and exploration strategies performed by agents (changes in connections between components of Ω-space and elements of λ-space; recombination of pre-­existing entities). 2) They compete, as they produce solutions with different degrees of success in manipulating inputs so to have goods and services. 3) They are selected, as the micro-­diversity is continually threatened by the process of leaving aside less preferred solutions. Macro-­dynamics of general patterns tend to emerge, until new drivers of variety appear (ideas, mappings, recombination and so on).23 Three main points stem from the analysis developed in this section: 1) technology is one of the most important results of search processes which unfold as a combinatorial activity in multiple spaces. 2) Rules are the outcomes of dynamics mappings between search spaces. 3) Combinatorial activity is directly linked to generative relationships, as the intensity and configuration of interactions among interactors foster a continuous recombining of pre-­existing elements and sometimes generate new knowledge. Thus we have identified the process and the mechanisms fostering the emergence of ordering principles in the Global workspace: the propositional and useful knowledge generates a set of ordering principles which limit the search space for problem-­solving activities, thanks to the mappings between different spaces explored by populations of interactors. However we must stress another fundamental element: since technological processes are inherently recursive and they permeate the entire product develop-

The production process as a complex world   55 ment process, we can see this latter as a complex space with a deep structure, precisely the layered sets of rules or grammar (Lombardi 2008a) which on the one hand reduce the high-­dimensionality of search space, and on the other define ranges of possibilities and then allow us to answer the basic question: how is it possible that structures of interactions, stochastic perturbations and generative relationships do not always produce a chaotic or completely random world, but the opposite happens? Indeed they become key processes in generating order and changes within it.

7  The economic processes as a complex, dynamic and ordered world based on two general principles: recursiveness and compositionality The product development process has been viewed as dynamic mappings from sets of consumers’ needs to sets of functional parameters to sets of design parameters, to demand requirements. During multi-­level problem-­solving activities interactors explore many domains, in order to produce fit goods or services. In general terms human beings, who act in a world where decision variables exceed their cognitive capabilities, have evolved as patterns seekers in searching for ordering principles. The inherent human propensity to search for “information structure”, that is, statistical dependency between signals and sets of signals, can be considered one of the main expressions of an evolutionary bias. According to an interesting theoretical framework (Tononi et al. 1996, 1998; Lungarella and Sporns 2005), human beings are evolutionarily constructed in order to “discover” statistical regularities, correlated to statistical dependence viewed as information exchanges and dynamic associations. Dynamic mappings between multiple spaces are nourished thanks to feedback loops and recursive structures, which unfold and spread within the organizational form space, that is, depending on the intensity and the configurations of interactions among teams, groups and economic units. In this perspective combination and the recombination of information packages structure the search activities of scientists and practitioners. It is precisely the technology that constitutes one peculiar expression of this process and it is a kind of “deep structure”. Recursiveness is a fundamental property for the product development process and it requires looking into some sub-­sets within each set shown in Figure 2.1. The unfolding of interactions between individuals and groups working on nested sub-­subsets can very often be unavoidable, especially within the so-­called “non linear model of innovation”. But if this happens how can a very complex world become ordered? In other terms, how do stable designs and organizational forms emerge? Let us analyse the problem from the point of view of mappings between multiple spaces. When topographic mappings are completely casual (Figure 2.4) interactors acting in each of the components are engaged in random walks and are unceasingly searching for solutions.

56   M. Lombardi

Figure 2.4 Mappings based on uncorrelated signals; interactors are engaged in random walks from one space to another. GW as a completely random landscape, with feedback loops among different spaces at multiple levels. The figure focuses on only three of the spaces depicted in Figure 2.1.

In these “fluid” situations agents are hindered by what computational linguistics defines “transmission bottleneck”: given the generative nature of each space and the infinite generative capacity of interactors, they cannot observe the entire sets of information and knowledge on which the others are working. So, “because each agent has infinite generative capacity, and the transmission of this capacity is transmitted on the basis of a finite set, the transmission bottleneck is unavoidable” (Brighton and Kirby 2006: 234). But evolutionary endowments lead human brains to search for statistical dependency between signals and sets of signals and then to construct patterns or sub-­maps. In doing so interactors are induced to create cognitive building blocks, in order to organize completely unforeseeable spaces, where moreover the successive mappings become impossible to manage. Following this line of reasoning we can hypothesize that sub-­sets are constructed belonging to different spaces and correlations between the subsets are elaborated insofar as solutions to problems are found (Figure 2.5). Thus exploration activity is not developed in every direction but it is sub-­ set oriented in the sense that it becomes purposeful, directed within a certain number of sorted out neighbourhoods. In this perspective technology is precisely a means for overcoming the “learning bottleneck” of information flowing through the economic system, by helping to point out trajectories of research. For neighbourhoods we mean that theoretical and applied contexts are defined for exploitation activity, which in turn unfolds by burrowing into them and searching for sub-­sets and sub-­subsets, until the solutions to problems are found. Through this process the Global workspace evolves from a completely random landscape towards a structured one, articulated in nested components and domains.

The production process as a complex world   57

Figure 2.5 GW as a structured landscape. Statistical regularities are discovered and neighbourhoods are pointed out. Layered mappings are formed, even if uncorrelated signals are present.

However this dynamic is only the beginning as the evolutionary founded propensity to sharpen statistical dependency leads people and economic units to elaborate knowledge: mappings and sub-­mappings from one space or sub-­ space to one another (mappings between structured domains) become meaningful as they are concatenated in a particular order. These organized mappings are precisely the “compositionality property” of human language, in other words the peculiar way of ordering the world around us through concatenated statements in order to communicate it.24 The compositionality principle involves the construction of determined systematic associations, able to direct exploration and exploitation activities; in other terms, it is fundamental to find structure preserving mappings either in ordering the world or in smoothing the information flows among individuals and economic units. Thus rules perform exactly this function and changes of a given order involve the loss or the transformation of rules. In this way two general principles, which act in different research domains and probably are deeply rooted in our minds, allow us to explain how the exploration of combinatorial knowledge space does not become combinatorial explosion of alternatives. Recursiveness and compositionality then are fundamental in structuring knowledge and organizational processes since they favour the emergence of interaction structures among entities.

8  Concluding remarks and further research lines Let us summarize the main line of arguments developed in this chapter. In a world populated by agents conceived as interactors, fundamental issues arise. Since product development processes are multilayered problem-­solving activities, how do stable configurations of goods and organizations emerge? What types of organizational patterns emerge?

58   M. Lombardi These problems are tackled on the basis of the morphogenetic approach which allows us to conceive problem-­solving activities as a variable mix of exploitation and exploration pathways in different search spaces. Following this theoretical perspective, dynamic mappings between multiple search spaces are studied, by conceiving the product development process as a combinatorial space, created by interactors thanks to a generative potential stemming from combining and recombining single elements, chunks and aggregates of elements. Basic units of analysis are information and knowledge packages elaborated through these dynamic mappings and the structures of interactions which foster a potentially infinite set of solutions. Human beings and organizations can generate and reduce/absorb the complexity of these processes thanks to two general principles, recursiveness and compositionality. Both already allow them to produce a potentially infinite number of solutions to problems of communication (ordinary language), while the first is applied in explaining technology. The conclusion of our analysis is that compositionality and recursiveness are properties of complex systems (language, science and technology) and at the same time principles for producing, absorbing and managing complexity. Indeed by means of a finite number of elements we can create a potentially infinite universe of entities. One final question has been investigated: are they the necessary results of interactions among adaptive units? In answering, our starting point has been that combinatorial spaces and generative potentials are fundamentally variable mixes of constraints and degrees of freedom: evolution from a random landscape to a structured one; innovations as changes of rules at different levels of hierarchy and so on. We have argued that selection dynamics among rules beget self-­organized biases towards a complex, dynamic and ordered world. Two fundamental reasons ground this statement: 1) completely random signals are forgotten and left aside, because they do not produce learning or adaptation process. 2) Structure creating and preserving mappings are learned and transmitted, as they stir the adaptation process. The next step of our research line is to develop some concepts, analytical tools and simulation models, particularly by representing search activities in a combinatorial space through suggestions and inputs from topology. Indeed it seems interesting to think about search spaces, as those analysed in design theory and in theory of organizations, in terms of topological spaces with suggestive formal and structural properties. A suggestive research line seems to be that of viewing the product development process and the Global workspace as a dynamic cognitive system, which is only seemingly disordered. In this context ideas and suggestions are developed within what we can define as a mental space: multiple mappings (one-­to-one, one-­to-many, many-­to-many) generate ideas, associations of ideas, rules and so on. So the problem could be formalized in terms of convergent processes towards precise vectors belonging to multiple spaces and at the same time as a set of endogenous sources of possible changes and discontinuities. Some properties of such a dynamic cognitive system could be useful starting points: 1) Hierarchical structure, since techno-­scientific grammars and rules act as

The production process as a complex world   59 sequences of constraints, which progressively reduce the degree of freedom in searching for satisfying solutions. In other words, we can conceive a tree of potential paths in searching for required values of vectors from within different domains. 2) Associative dynamics, which allow us to construct ideas and to associate them by evaluating their degree of similarity. 3) Discontinuity, as ideas are not infinitely divisible and it is impossible to postulate their homogeneity. 4) Stability, because the thinking process finds solutions within evolutionary dynamics, where variants and radical novelties are unceasingly conceived and tested. In this perspective we have to model an evolving tree structure which overlaps multiple search spaces, while its manifold branching architecture is the expression of potentially infinite research paths. Searching for and discovering constraints, on the basis of unceasing mappings and rules, imply sequences of trials and often iterative cycles along different paths. Branching and iterations, therefore, have to be considered open-­ended activities until a more acceptable solution is obtained.25

Notes   1 A first version of this chapter has been presented at the Research-­Workshop on “Evolutionary Economics” Schloss Wartensee, 22 and 23 June 2007. I would like to thank the participants to the Conference for valuable comments.   2 For a review see Braha and Maimon (1998).   3 In fact, in human language a finite number of elements can be combined so as to attain a potentially infinite set of entities.   4 By “generative” we mean: through the use of basic principles, guiding rules and heuristics we frequently produce combinations and recombinations of pre-­existent knowledge so that different domains of research can be improved, intermingled and sometimes radically and unpredictably changed.   5 The concept of interactor has been extended to the firm (Hodgson and Knudsen 2004), whose cohesiveness is based on well-­connected routines.   6 In general, be interactors individual or aggregates of them, they can develop information and knowledge in many different ways, from selecting choices and actions on the basis of accumulated experience (“local search”, Gavetti and Levinthal 2000) to penetrating into distant regions of the solution space (“global search”), to combining or alternating them. The real processes are characterized by people with two types of behaviour (Allen and Lesser 1991): Cartesian behaviour, which is the ability to “exploit the available information”, and stochast behaviour, which is the ability to go “beyond the present knowledge”.   7 We attempt to enlarge a particular research line introduced by the Lancaster model based on the vector of product characteristics (1966) and further developed by Saviotti and Metcalfe (1984).   8 By goal-­oriented agents we mean interactors committed to problem-­solving activities at multiple levels so that a working product can be obtained.   9 The ordered world, as it has been defined in introduction, is the product conceived as a vector of attributes resulting from a huge amount of search developed in multiple spaces. 10 “A solid understanding of how to cascade requirements from the system level down to the components ultimately helps to guide system designers through the multitude of trade-­off decisions that arise during product development” (Yassine and Wissmann 2007: 127).

60   M. Lombardi 11 A huge amount of real products have been analysed: brake systems, the driving shaft and particularly the cylindrical part, power screwdriver, gas turbine, audio system, automotive industry, climate control systems and elevator systems. 12 In order to obtain a convergent sequence of parameters, in other words to avoid information cycles and circuits fostering the combinatorial explosion or simply to render the process feasible, a number of algorithms has been elaborated, so that feedbacks are reduced or eliminated (see Sharman and Yassine 2004). When partitioning algorithms are unable to generate a lower triangular form, the system analysts try to minimize the number of the remaining blocks on the diagonal. 13 The “design churn” is “a scenario where the total number of problems being solved (or progress being made) does not reduce (increase) monotonically as the project evolves over time” (Yassine et al. 2003: 145). 14 In order to tackle these problems different strategies for management have been studied in order to realize convergence by reducing the impact of the three main sources of churn (interdependency, concurrency and feedback delays) (Yassine and Braha 2003; Yassine et al. 2003). 15 The concept of the module is widely applied also in biology, where the theory of morphological evolution (Eble 2003) conceives modules as cohesive units, which are able to keep stability and at the same time to change. 16 For heuristic purposes we adopt here a quite informal definition: “a complex system is a large network of relatively simple components with no central control, in which emergent complex behaviour is exhibited.” (Mitchell 2006: 1195). 17 Hofstadter (1979) defines this exploration strategy as “parallel terraced scan”. 18 Degeneracy indicates that “structurally different elements may yield the same or different functions depending on the context in which it is expressed” (Edelman and Gally 2001: 13763). 19 An operation is here defined as “any separable stage of the production process in which the property vector of the work piece is changed in one or several of its dimensions” (Buenstorf 2004: 62). 20 It is in the form of “If . . . then rules which may reach considerable degrees of complexity” (Vanberg 2002: 16). 21 If a rule is “field-­tested” this means it is tested in its intended field of application. If it is “grounded” this means it is known why the intervention or artefact gives the desired performance (van Aken 2005b: 23). 22 We would emphasize that Ω-space comprises all conceivable ideational kernels and K-­space contains general principles ordering the world around us. These principles are fixed in directing search activities and at the same time variables to the extent that the outcomes of search processes trigger the need for exploring new fields within the Ω-space. 23 For an analysis of technology as the combination and recombination of bits of knowledge, while invention is a process of recombinant search, see Fleming and Sorenson (2001) and Fleming (2001). 24 “For an agent with a language like English, a signal for a meaning is constructed by generating strings for subparts of that meaning and concatenating them in a particular order.” (Kirby 1999). 25 For a first attempt in this direction see Lombardi (2008b).

References Allen P.M. and Lesser M. (1991), Evolutionary Human Systems: Learning, Ignorance and Subjectivity, in Metcalfe J.S. and Saviotti P. (eds), Evolutionary Theories of Economic and Technological Change, Harwood Academic Press, Chur. Anderson P. (1997), Some Thoughts About Distribution in Economics, in Arthur W.B.,

The production process as a complex world   61 Durlauf S.N. and Lane D. (eds), The Economy as an Evolving Complex System II, Addison-­Wesley, Reading. —— (1999), Complexity Theory and Organization Science, Organization Science, 10: 216–32. Antonelli C. (2007), The System Dynamics of Collective Knowledge: From Gradualism and Saltationism to Punctuated Change, Journal of Economic Behaviour & Organization, 62: 215–36. Aoki M. (2004), An Organizational Architecture of T-­form: Silicon Valley Clustering and Its Institutional Coherence, Industrial and Corporate Change, 6: 967–81. Aoki M. and Takizawa H. (2002), Information, Incentives and Option Value: the Silicon Valley Model, Journal of Comparative Economics, 30: 759–86. Archer M.S. (1995), Realist Social Theory: The Morphogenetic Approach, Cambridge University Press, Cambridge. Arthur W.B. (2007), The Structure of Invention, Research Policy, 36(2): 274–87. Arthur W.B. and Polak W. (2006), The Evolution of Technology Within a Simple Computer Model, Complexity, 11(5): 23–31. Arthur W.B., Durlauf S.N. and Lane D. (1997), Introduction, in Arthur W.B., Durlauf S.N. and Lane D. (eds), The Economy as an Evolving Complex System II. Addison-­ Wesley, Reading. Baldwin C.Y. and Clark K. (2000), Design Rules: The Power of Modularity, MIT Press, Cambridge, MA. Barney J.B. (2001), Resource-­Based Theories of Competitive Advantage: A Ten Year Retrospective on the Resource-­Based View, Journal of Management, 27(6): 643–50. Barney J.B., Wright M. and Ketchen D.J. (2001), The Resource-­Based View of the Firm: Ten Years After 1991, Journal of Management, 27(6): 625–41. Bar-­Yam Y. (2002), Large Scale Engineering and Evolutionary Change: Useful Concepts for Implementation of FORCEnet, NECSI (New England Complex Systems Institute) Report. —— (2004), A Mathematical Theory of Strong Emergence Using Multiscale Variety, Complexity, 9(6): 15–24. —— (2005), About Engineering Complex Systems: Multiscale Analysis and Evolutionary Engineering, in Brueckner S., Di Marzo Segurendo G., Karageorgos A. and Nagpal R. (eds), Engineering Self-­Organising Systems, Methodologies and Applications [revised versions of papers presented at the Engineering Selforganising Applications (ESOA 2004) Lecture Notes in Computer Science 3464], Springer, Berlin/Heidelberg. Boisot M. and Canals A. (2004), Data, Information and Knowledge: Have We Got It Right?, Journal of Evolutionary Economics, 14(1): 43–67. Braha D. and Bar-­Yam Y. (2004), Information Flow Structure in Large-­Scale Product Development Organizational Networks, Journal of Information Technology, 19(4): 244–53. Braha D. and Maimon O. (1998), Design as Scientific Problem-­Solving, in Braha D. and Maimon O. (eds), A mathematical Theory of Design: Foundations, Algorithms, and Applications, Kluwer-­Springer, Dordrecht, pp. 19–84. Braha D. and Reich Y. (2003), Topological Structures for Modelling Engineering Design Processes, Research in Engineering Design, 14(4): 185–99. Brighton H. and Kirby S. (2006), Understanding Linguistic Evolution by Visualizing the Emergence of Topographic Mappings, Artificial Life, 12(2): 229–42. Browning R.R. (2001), Applying the Design Structure Matrix to System Decomposition and Integration Problems: A Review and New Directions, IEEE Transactions on Engineering Management, 3, August: 292–306.

62   M. Lombardi Browning R.R., Deyst J.J., Eppinger S.D. and Whitney D.E. (2000), Complex System Product Development: Adding Value by Creating Information and Reducing Risk, Proceedings of the Tenth Annual International Symposium of INOCSE, Minneapolis, MN, pp. 16–20. Buenstorf G. (2004), The Economics of Energy and the Production Process, Edward Elgar, Cheltenham —— (2005), Sequential Production, Modularity and Technological Change, Structural Change and Economic Dynamics, 16(2): 221–41. Carley K.M. (2002), Computational Organization Science: A New Frontier, Proceedings of National Academy of Science (PNAS), 14: 7257–62. Carlson J.M. and Doyle J. (2002), Complexity and Robustness, Proceedings of National Academy of Science (PNAS), 1: 2538–45. Chandler A.D. (1977), The Visible Hand: The Managerial Revolution in American Business, The Belknap Press, Cambridge, MA. Ciarli T., Leoncini R., Montresor A., and Valente M. (2008), Technological Change and the Vertical Organization of Industries, Journal of Evolutionary Economics, 18(3–4): 367–87. Csete M.E. and Doyle J.C. (2002), Reverse Engineering of Biological Complexity, Science, 295: 1664–70. DeCanio S. and Watkins W.E. (1998), Information Processing and Organizational Structure, Journal of Economic Behaviour & Organization, 36(3): 275–94. Dooley K.J. and Van de Ven A.H. (1999), Explaining Complex Organizational Dynamics, Organization Science, 10(3): 358–72. Dopfer K. and Potts J. (2008), The General Theory of Economic Evolution, Routledge, Abingdon. Dosi G. and Marengo L. (2005), Division of Labour, Organizational Coordination and Market Mechanisms in Collective Problem-­Solving, Journal of Economic Behaviour & Organization, 58(2): 303–26. Dosi G., Nelson R.R. and Winter S.G. (2000) (eds), The Nature and Dynamics of Organizational Capabilities, Oxford University Press, Oxford. Eble G.J. (2003), Morphological Modularity and Macroevolution: Conceptual and Empirical aspects, in Callebaut W. and Rasskin-­Gutman D. (eds), Modularity: Understanding the Development and Evolution of Complex Natural Systems, MIT Press, Cambridge, MA. Edelman G.M. and Gally J.A. (2001), Degeneracy and Complexity in Biological Systems, Proceedings of National. Academy of Science (PNAS), 98: 13763–8. Eppinger S.D., Whitney D.E., Smith R.P. and Gebala D.A. (1994), A Model-­Based Method for Organising Tasks in Product Development, Research in Engineering Design, 6: 1–13. Fleming L. (2001), Recombinant Uncertainty in Technological Search, Management Science, 47(1): 117–32. Fleming L. and Sorenson O. (2001), Technology as a Complex Adaptive System: Evidence from Patent Data, Research Policy, 30: 1019–39. Garnsey E. and McGlade J. (2006) (eds), Complexity and Co-­Evolution, Edward Elgar, Cheltenham. Gavetti G. and Levinthal D. (2000), Looking Forward and Looking Backward: Cognitive and Experiential Search, Administrative Science Quarterly, 45(1): 113–37. Geels F.W. (2004), From Sectoral Systems of Innovation to Socio-­Technical Systems Insights About Dynamics and Change from Sociology and Institutional Theory, Research Policy, 33(6–7): 897–920.

The production process as a complex world   63 Geels F.W. and Schot J. (2007), Typology of Socio-­Technical Transition Pathways, Research Policy, 36(3): 399–417. Gulati R.K. and Eppinger S. (1996), The Coupling of Product Architecture and Organizational Structure Decisions, Massachusetts Institute of Technology Sloan School of Management Working Paper, No. 3906. Hodgson G.M. and Knudsen T. (2004), The Firm as an Interactor: Firms as Vehicles for Habits and Routines, Journal of Evolutionary Economics, 14(3): 281–307. Hofstadter D.R. (1979), Gödel, Escher, Bach: An Eternal Golden Braid, Basic Books, New York. Hoopes D.L. and Postre1 S. (1999), Shared Knowledge, “Glitches,” and Product Development Performance, Strategic Management Journal, 20(9): 837–65. Hull D.L., Langman R.E. and Glenn S.S. (2001), A General Account of Selection: Biology, Immunology, and Behaviour, Behavioural and Brain Sciences, 24: 511–73. Kirby S. (1999), Learning, Bottlenecks and Infinity: A Working Model of the Evolution of Syntactic Communication, in Dautenhahn K. and Nehaniv C. (eds), Proceedings of the AISB’99 Symposium on Imitation in Animals and Artifacts [published in 2002 by MIT Press, Harvard]. Kitano H. (2002), Systems Biology: A Brief Overview, Science, March, pp. 1662–4. Krishnan V. and Ulrich K.T. (2001), Product Development Decisions: A Review of the Literature, Management Science, 47(1): 1–21. Lancaster K.J. (1966), A New Approach to Consumer Theory, Journal of Political Economy, 74(2): 132–57. Langlois R.N. (2002), Modularity in Technology and Organization, Journal of Economic Behavior & Organization, 1: 19–37. Levin S.A. (2002), Complex Adaptive Systems: Exploring the Known, the Unknown and the Unknowable, Bulletin (New Series) of the American Mathematical Society, 1: 3–19. Lombardi M. (2001), Rethinking Economics with Hodgson’s Evolution and Institutions, History of Economic Ideas, 3: 245–66. —— (2008a), A Morphogenetic Approach to the Evolution of Technological Capabilities, in Leoncini R. and Montresor A. (eds), Dynamic Capabilities Between Firm Organisation and Local Systems of Production, Routledge, Abingdon. —— (2008b), The Product Development Process as a Complex, Dynamic and Ordered Global Workspace, Paper presented at the Fifth European Social Simulation Association Conference, University of Brescia, 1–5 September. Lungarella M. and Sporns O. (2005), Information Self-­Structuring: Key Principle for Learning and Development, Proceedings of 2005 4th IEEE International Conference on Development and Learning available at the IEEE Explore Digital Library, Digital Object Identifier: 10.1109/DEVLRN.2005.1490938. Magee C.L. and de Weck O.L. (2003), An Attempt at Complex Systems Classification, Massachusetts Institute of Technology, Engineering Systems Division, Working Paper series, ESD-­WP-2003–01.02. March J.G. (1991), Exploration and Exploitation in Organizational Learning, Organization Science 2: 71–87 Margolis H. (1987), Patterns, Thinking, and Cognition, The Chicago University Press, Chicago. Mikkola J.H. (2003), Modularity, Component Outsourcing, and Inter-­Firm Learning, R&D Management, 33(4): 439–54. —— (2006), Capturing the Degree of Modularity Embedded in Product Architectures, Journal of Product Innovation Management, 23(2): 128–46.

64   M. Lombardi Mitchell M. (2006), Complex Systems: Network Thinking, Artificial Intelligence, 170(18): 1194–1212. Mokyr J. (2005), Long-­term Economic Growth and the History of Technology, in Aghion P. and Durlauf S. (eds), Handbook of Economic Growth, Elsevier, North-­ Holland. Morel B. and Ramanujam R. (1999), Through the Looking Glass of Complexity: The Dynamics of Organizations as Adaptive and Evolving Systems, Organization Science, 10(3): 278–93. Morroni M. (2006), Knowledge, Scale and Transactions in the Theory of the Firm, Cambridge University Press, Cambridge. Nelson R. and Winter S. (1982), An Evolutionary Theory of Economic Change, Harvard University Press, Cambridge, MA. Nightingale P. (2000), The Product–Process–Organisation Relationship in Complex Development Projects, Research Policy, 29(7–8): 913–30. Novak S. and Eppinger S.D. (2001), Sourcing by Design: Product Complexity and the Supply Chain, Management Science, 47(1): 189–204. Olsson O. (2000), Knowledge as a Set in Idea Space: An Epistemological View on Growth, Journal of Economic Growth, 5(3): 253–75. —— (2005), Technology Opportunity and Growth, Journal of Economic Growth, 10: 35–57. Olsson O. and Frey B.S. (2002), Entrepreneurship as Recombinant Growth, Small Business Economics, 19(2): 69–80. Pimmler T. U. and Eppinger S.D. (1994), Integration Analysis of Product Decompositions, Proceedings of ASME 6th International Conference on Design Theory and Methodology, Minneapolis. Radner R. (1993), The Organization of Decentralized Information Processing, Econometrica, 61(5): 1109–46. Rasskin-­Gutman D. (2003), Modularity: Jumping Forms within Morphospace, in Callebaut W. and Rasskin-­Gutman D. (eds), Modularity: Understanding the Development and Evolution of Complex Natural Systems, MIT Press, Cambridge, MA. Reich Y. (1995), A Critical Review of General Design Theory, Research in Engineering Design, 71(1): 1–18. Sanchez R. and Mahoney J.T. (1996), Modularity, Flexibility, and Knowledge Management in Product and Organization Design, Strategic Management Journal, 17(1): 63–76. Saviotti P. (2007), On the Dynamics of Generation and Utilisation of Knowledge: The Local Character of Knowledge, Structural Change and Economic Dynamics, 18(4): 387–408. Saviotti P. and Metcalfe J.S. (1984), A Theoretical Approach to the Construction of Technological Output Indicators, Research Policy, 13(3): 141–51. Schilling M.A. (2000), Toward a General Modular Systems Theory and Its Application to Interfirm Product Modularity, Academy of Management Review, 25(2): 312–34. Sharman D.M. and Yassine A.A. (2004), Characterizing Complex Product Architectures, Systems Engineering, 7(1): 35–60. Simon H.A. (1962), The Architecture of Complexity, Proceedings of the American Philosophical Society, 106(6): 467–82. Smith R.P. and Eppinger S.D. (1997), Identifying and Controlling Features of Engineering Design Iteration, Management Science, 43(3): 276–93. Suh N.P. (1990), The Principles of Design, Oxford University Press, New York.

The production process as a complex world   65 —— (1998), Axiomatic Design Theory for Systems, Research in Engineering Design, 10: 189–209. —— (1999), A Theory of Complexity, Periodicity and the Design Axioms, Research in Engineering Design, 11: 116–31. Teece D., Pisano G. and Shuen A. (1997), Dynamic Capabilities and Strategic Management, Strategic Management Journal, 18(7): 509–33. Tononi G., Sporns O. and Edelman G.M. (1996), A Complexity Measure for Selective Matching of Signals by the Brain, Proceedings of National. Academy of Science (PNAS), 93: 3422–7. Tononi G., Sporns O. and Edelman G.M. (1998), Complexity and Coherency: Integrating Information in the Brain, Trends in Cognitive Sciences, 12: 474–84. Tooby J. and Cosmides L. (1992), The Psychological Foundations of Culture, in Barkow J.H., Cosmides L. and Tooby J. (eds), The Adapted Mind, Oxford University Press, New York, pp. 19–136 Ulrich K. (1995), The Role of Product Architecture in the Manufacturing Firm, Research Policy, 24(3): 419–41. van Aken J.E. (2005a), Valid Knowledge for the Professional Design of Large and Complex Design Processes, Design Studies, 26(4): 379–404. —— (2005b), Management Research as a Design Science: Articulating the Research Products of Mode 2 Knowledge Production in Management, British Journal of Management, 16(1): 19–36. Vanberg V.J. (2002), Rational Choice vs Program-­Based Behaviour, Rationality and Society, 14(1): 7–54. Vromen J. (2004a), Conjectural Revisionary Economic Ontology: Outline of an Ambitious Research Agenda for Evolutionary Economics, Journal of Economic Methodology, 11(2): 213–47. —— (2004b), Routines, Genes and Program-­Based Behaviour, Max Plank Institute Working Papers, 0420. Wang Q. and von Tunzelmann N. (2000), Complexity and the Functions of the Firm: Breadth and Depth, Research Policy, 29 (7–8): 805–18. Wernerfelt B. (1995), The Resource-­Based View of the Firm: Ten Years After, Strategic Management Journal, 16(3): 1–174. Winter S.G. (2005), Towards an Evolutionary Theory of Production, in Dopfer, K, (ed.), The Evolutionary Foundations of Economics, Cambridge University Press, Cambridge. Yassine A.A. (2004), An Introduction to Modeling and Analyzing Complex Product Development Processes Using the Design Structure Matrix (DSM) Method, Quaderni di Management, 9: 72–88. Yassine A.A. and Braha D. (2003), Complex Concurrent Engineering and the Design Structure Matrix Method, Concurrent Engineering: Research and Applications, 11(3): 164–76. Yassine A.A. and Falkenburg D.R. (1999), A Framework for Design Process Specification vs Management, Journal of Engineering Design, 10(3): 223–34. Yassine A.A. and Wissmann L. (2007), The Implications of Product Architecture on the Firm, Systems Engineering, 2: 118–35. Yassine A.A., Joglekar N., Braha D., Eppinger S. and Whitney D. (2003), Information Hiding in Product Development: The Design Churn Effect, Research in Engineering Design, 14: 145–61.

3 Incumbents’ strategies for platform competition Shaping the boundaries of creative destruction1 Stefano Brusoni and Roberto Fontana 1  Introduction Keith Pavitt has argued that understanding processes of knowledge special­ ization is vital to understand how modern organizations innovate, through the embodiment of a widening range of functions in products; and evolve, adapting their internal processes to the requirements of heterogeneous, yet interrelated, bodies of technical and scientific knowledge. Similarly, Miller et al. (1995) have coined the term Complex Products and Systems (CoPS) to stress the enormous importance acquired in modern societies by complex artifacts, which perform technically sophisticated functions and are made up of wide arrays of complex sub-­systems. In turn, different sub-­systems evolve in different ways, posing dif­ ferent problems, following different life cycles. Levinthal (1997) and co-­workers have stressed that organizations themselves can and should be represented as complex systems made up of many interconnected components, whose dynamics in response to environmental changes also present remarkable degrees of com­ plexity and emergence. Arora et al. (2001) emphasize the increasing importance of sophisticated ‘knowledge markets’ in enabling organizations to access and sell knowledge generated in an increasingly disintegrated and specialized way. Intellectual Property Rights (IPRs) regimes acquire central stage in a context in which one single organization cannot master all relevant bodies of knowledge, nor can it accurately predict where internally funded knowledge generating pro­ cesses are actually leading. The ability of buying in externally generated know­ ledge is matched in importance by the possibility of selling out what knowledge is discovered in non-­core areas. These streams of research lay the foundations to our understanding of platforms, i.e. an evolving system made of interdependent pieces that can each be innovated upon’ (Gawer and Cusumano 2002). Platforms embody both physical technologies and service technologies. The latter aims at improving mainte­ nance, adaptability and users’ satisfaction. Sophisticated users do not require products, but ‘solutions’ to complex sets of problems and needs. Hence, we witness the emergence of providers of ‘Integrated Service Solutions’ (ISS) in a variety of sectors (Davies 2003, 2004). Platforms embody a variety of technical solutions generated relying on different, specialized and very heterogeneous

Strategies for platform competition   67 knowledge bases. Platforms build upon specific institutional and organizational solutions which allow a variety of organizations to interact coherently in their design, production and maintenance. Software interfaces and different types of standard play a central role in understanding how these complex arrays are held together across time and space. The ability of allocating property rights on embodied and disembodied knowledge, the establishment of open standards, the development of modular interfaces: all these factors play a role in our under­ standing of how platforms have come into being, and how they are becoming the central locus of competition in modern economies. In this chapter, we aim at providing a few examples of the emergence of platforms, in different sectors. Our objective is sketching a first, preliminary and merely tentative story of how the evolution of knowledge bases, physical artifacts, complex organizations and IPRs has led us to a world characterized by what Gawer and Cusumano (2002) aptly called ‘platform competition’. Keith Pavitt laid the foundation of our discussion some time ago, developing the table we reproduced here as Table 3.1. The table highlights the implications for the organization of both manufacturing and knowledge-­intensive activities of a number of technological waves. It shows that certain major advances in tech­ nology have been key factors of changes in organizational specialization, some­ times leading to disintegration and sometimes to integration; the degree of disintegration in the production of artifacts has always been greater than in the production of scientific and technological knowledge. This table captures both the increasing complexity of products, and the growing involvement of users as actors of change. Increasing knowledge special­ ization, complexity and users’ involvement challenge the traditional ways of coordinating firms’ activities. Paying attention to the distinction between know­ ledge and product dynamics (see again Table 3.1) is necessary to make sense of recent changes in the international division of labour. The locus of competition through innovation in leading companies is shifting from discrete physical product and process innovations, to innovations in the design, development, inte­ gration and marketing of increasingly complex systems and platforms. Against this background, this chapter analyses the strategies through which established organizations try to influence the competitive landscape in which they operate. It builds upon the idea that products and organizations are complex systems made up of many interconnected elements. How these elements are con­ nected is a matter of technological and organizational design choices. Such choices define and organize some aspects of the inner working of complex systems, define the strengths of the connection among elements, the entry for new components and the exit for obsolete components. The main point that this literature stresses is that firms decreasingly compete on specific products (e.g. washing machines) and increasingly on product families which share some common technological or func­ tional traits (e.g. VHS technology or Microsoft Office). This shift in the locus of competition has led management scholars to coin the notion of ‘platforms’. The chief aim of this chapter is twofold. First, it aims at defining a simple typol­ ogy of platforms on the basis of extant research. We adopt a very simple strategy

Electrical and electronic products

Physics

Increasing product complexity (more Modular product designs components, sub-systems, and bodies Mass customization of knowledge) + ICTs Platform competition

Source: Brusoni and Pavitt (2003)

   Early 2000s

Growth of in-house R&D as dominant source of innovation

Integration of product design, manufacture and marketing

Synthetic products

Organic chemistry

Vertical disintegration in design and production of product components and sub-systems

In- house knowledge of design, and operation of subsystems and components

In-house knowledge of design, Various (e.g. metal cutting, chemistry, Technological convergence Partial vertical computing, ITC) in segments of production disintegration in production and operation of producers’ goods segments (e.g. machine tools, continuous processes and instrumentation, CAD, robots, applications software)

In-house knowledge of design, and operation of capital goods

Vertical disintegration in capital goods

Materials (iron and steel)

Mass production of standard commodities

Energy (coal)

Vertical disintegration

Vertical disintegration

In-house development of specialized skills

Interchangeable parts

Improvements in metal cutting and shaping

Early 1800s

Knowledge

Production

Organizational integration/disintegration of:

Specialization and integration of purchasing, production, marketing

Implications for firms’ management

Changes in technology

Time line

Table 3.1 Organizations in the context of increasing specialization in technology and complexity in artifacts

Strategies for platform competition   69 to characterize different platforms. We focus on two central features of the interdependencies among the technological and organizational components of a plat­ form: their modularity, and their openness. Second, it aims at analysing the dynamics of innovation and competition within each of the identified types of plat­ form. Our focus falls on the problems and opportunities faced by established organizations in coping with a changing environment. The idea is that, by defining platforms of specific characteristics incumbents try to govern the markets in which they are active. This chapter is structured as follows. Section 2 lays the founda­ tions of the analysis. Section 3 identifies the key dimensions that allow us to define a typology of platforms. Section 4 populates this typology discussing the competit­ ive implications. Section 5 concludes.

2  Background: technologies and organizations This section highlights two key dimensions along which to classify different platforms. 2.1  Products as complex systems The idea that artefacts can be conceived as networks of interconnected elements is not new in the literature. Here we build upon recent research on so-­called Complex Product Systems (CoPS) as defined by, among others, Miller et al. (1995) and Hobday (1998). CoPS have been defined as ‘high cost, engineering-­ intensive products, sub-­systems, or constructs’ (Hobday 1998). They differ from simpler, mass-­produced products in terms of the dynamics of the innovation process, competitive strategies, managerial constraints and industrial co-­ ordination. According to Hobday (1998), four characteristics set CoPS apart from mass-­produced goods: (a) they are high-­cost systems composed of many interacting and often customized elements; (b) they exhibit emerging properties; (c) their design, development, and production usually involve several firms; (d) the degree of user involvement is usually very high. First, CoPS are normally high value systems composed of many interacting and sometimes customised elements. Components are organized hierarchically. In addition, different components rely on knowledge bases that may exhibit dif­ ferent rates of change and technological opportunities (e.g. grinding operations versus catalysis). In particular, as different interacting component technologies are considered, the role played by uneven rates of change can be explored as a factor shaping the organization of innovative labour. Second, CoPS may exhibit emerging properties. For example, the scale-­up process of a chemical route is not linear. This means that the basic chemistry of the process changes while moving from the laboratory bench up to full-­scale plant. Reactor behaviour can hardly be theoretically predicted, unless very stand­ ard processes are involved. Existing computer simulation tools can predict the behaviour of a whole process at steady state, but they can hardly help engineers figure out how the process will behave in dynamic situations, e.g. during the

70   S. Brusoni and R. Fontana start-­up phase or during reactivation after a periodic shutdown. Therefore, CoPS exhibit a high degree of ‘systemic uncertainty’ (Bonaccorsi and Pammolli 1996). Systemic uncertainty is present when the interactions between different levels of the system are subjected to uncertainty. Third, in CoPS industries the degree of user involvement is usually very high. This is surely true of chemicals, where the operators are usually in charge of the conceptual design stage. Therefore, CoPS industries lend themselves to the ana­ lysis of the implications of changing patterns of division of labour between heterogeneous firms. Fourth, CoPS design, development and production usually involve several firms. Two points are worth stressing here. First, these networks of firms reflect the hierarchical nature of the product they have to deliver. Different firms will play different roles according to the breadth and depth of their tech­ nical and organizational capabilities. Second, as different agents are involved, ‘epistemic’ uncertainty is also present. Epistemic uncertainty derives from the need to co-­ordinate the solution to a specific problem into a wide network of interdependent technological and managerial processes. For instance, Ancori et al. (2000) stress the problems related to developing ‘common languages, common classification and categorization’ systems when heterogeneous agents interact. While originally put forward to analyse the dynamics of capital goods indus­ tries (e.g. aero-­engines, chemical plants, flight simulators etc.) the idea that any product can be analysed in terms of its constituents has received considerable attention in a variety of consumer industries. This is so for two related reasons. First, products in a variety of industries are embodying a widening range of functionalities. Second, such new functionalities build upon specialized bodies of knowledge. Within this stream of literature, increasing attention has been devoted to the analysis of a specific type of design strategy: modularity. The chief aim of modular design strategies is decomposing complex artefacts in simpler sub-­systems which can evolve independently of each other. In a way, modularity is put forward as the solution to the increasing connectedness of technological and organizational systems. 2.2  Advantages and disadvantages of modular product design strategies Modularity is a design strategy aimed at decomposing complex products into simpler, self-­contained modules. It is based upon a two-­pronged approach. First, standardized interfaces across components are set. Second, all components are being designed in such a way that each performs only one function. Garud and Kumaraswamy (1995) argued that in such contexts the key advantage delivered by modular design strategies is the possibility to reap the benefits of ‘economies of substitution’. That is to say, innovation in complex technical systems, made up of heterogeneous components that change at different rates, can be achieved by replacing certain modules, while retaining others. In their words:

Strategies for platform competition   71 The potential for such economies [of substitution] increases if technological systems are modularly upgradeable. By designing modularly upgradeable systems, firms can reduce product development time, leverage their past investments, and provide customers with continuity. (Garud and Kumaraswamy 1995: 93) By adopting modular design strategies, firms can decouple the design and devel­ opment of separate modules, through the standardization of their interfaces. Thus, they can develop new products more quickly, as integration of the final product is a matter of mix-­and-match of ‘black boxes’ (Baldwin and Clark 2000; Sanchez and Mahoney 1996). Advanced technological knowledge about com­ ponent interactions is used to fully specify and standardize component interfaces and, therefore, to decouple the design of the product architecture (i.e. arrange­ ment of functional elements) from the design of each module. Another major advantage of modularity in fast changing contexts is the reduction in develop­ ment costs made possible by shifting the level of experimentation from an entire system to a single module. Modularity renders complexity manageable by ena­ bling experiments to be run at the level of modules, rather than for an entire arti­ fact, and to be run in parallel (Baldwin and Clark 2000). As stressed by Garud and Kumaraswamy (1995), in fast changing environ­ ments modular upgradeability becomes a key source of competitive advantage for two further reasons. First, on the supply side manufacturers are able to intro­ duce new developments into existing products that produce gains in perform­ ance, functionality and complexity. Second, on the demand side by simply buying a new module, customers can enjoy the benefit of successive develop­ ments to previously adopted products without having to completely change their installed base. Avoiding cannibalization, and allowing customers to continuously upgrade their installed base while acquiring new customers, has been a major driver of competition in Information and Communication Technology (ICT) industries. For example, Christensen (1997) and Christensen and Rosenbloom (1995) have analysed the dangers and dilemmas faced by incumbents when technological change leads to the emergence of new product niches of sophistic­ ated users. The fear of cannibalizing their existing customer base, coupled with the uncertainty associated with the development of a more sophisticated product, may lead incumbents to lose their leadership to the advantage of new entrants. Modularity might help solve such dilemmas. First, it reduces the costs and risks of being locked into early, and inferior, technologies. Hence, it provides manu­ facturers with the incentive for continual development of their products. Second, it provides users with the incentive to become early adopters, since they can acquire new modular upgrades. However, like all design strategies, modularity entails costs and trade-­offs. First, developing modular products is more difficult than developing integral ones (Ulrich 1995). Achieving modularity requires a very precise understanding of product functionalities, how they are allocated to components, and how the components interact as well as a determined choice in terms of the specific

72   S. Brusoni and R. Fontana problem-­solving strategies adopted. Thus, the choice of product architecture should be related to a company’s product strategy. Ulrich (1995) argued that if a company wants to stress product performance, then the most appropriate choice would be an integral architecture, since this type of architecture optimizes global performance characteristics. On the other hand, companies wishing to emphasize product change and variety, flexibility and upgradeability, may as well choose a modular architecture. Second, there are other costs involved in developing a modular architecture. For example, Ethiraj and Levinthal (2004) caution against the dangers of over-­ modularization when the ‘real’ decomposition scheme of the problem at hand is uncertain. In particular, the costs of the opportunities that might have been exploited by adopting different problem-­solving strategies must be considered. This is particularly cogent in an innovative context. The change in the unit of selection implied by modularity in relation to an integral system is crucial: having ‘fine’ rather than ‘coarse’ units of selection makes the search process faster (selection operates on the finer scale of modules and therefore the selec­ tion environment is in a sense ‘richer’), but essentially ‘local’ and quickly tends to lock-­into a local optimum. In an integral system the search is global, which implies that there is no lock-­in, but it is much slower and in complex space there is a lot of wasteful search as nonsensical options can be generated (Marengo and Dosi 2005). In other words, modular design strategies seem capable of providing incremental solutions along given trajectories which embody specific problem-­ solving strategies. However, this could lead to myopic learning processes that make firms reluctant to pursue opportunities beyond the existing trajectory. Looking at how Fujitsu managed to maintain its competitive leadership through a major technological revolution, Chesbrough and Kusunoki (2001) provide an insightful analysis of the kind of ‘tunnel vision’ that can be induced by exces­ sively modular strategies. To conclude, this section has defined a technological platform as ‘an evolving system made of interdependent pieces that can each be innovated upon’ (Gawer and Cusumano 2002). Being complex systems, platforms display many character­ istics (i.e. the presence of positive feedbacks, emerging properties etc.). This section has proposed modularity as one of the key characteristics of technological platforms. A platform can be designed in a modular fashion or not. This choice has major competitive implications in terms of search and entry strategies. The follow­ ing section proposes a second dimension useful to evaluate the dynamics of com­ petition at the platform level: the openness, or closure, of the interfaces. 2.3  Open versus closed interfaces Modularity in the technological interfaces is not the only dimension that matters to our analysis of platforms, though. Organizations too have been analysed as entities made up of interconnected elements. The evolution of organizations is linked to the evolution of the connections among those elements that constitute them (Siggelkow 2002: 126). Generally speaking, the evolution of such connec­

Strategies for platform competition   73 tions appears fundamental to explain radical changes in business strategies (Siggelkow 2001), organization structures (Romanelli and Tushman 1994), insti­ tutional settings (Padgett and McLean 2006) and organizational configurations (Miller 1987; Grandori 1997). Extant research in organization design focuses on the nature of connections which renders organizations more or less likely to succeed in adapting to environmental changes. Some authors argue that tightly coupled organizations have major advantages when dealing with fundamental uncertainty, as tight coupling among elements makes them more sensitive and responsive to the environment (e.g. Weick 1976). Others argue that tight cou­ pling prevents organizations from adapting rapidly: since each change entails many interrelated changes, inertia is the most likely outcome (e.g. Levinthal 1997). Such issues are core to recent research on organizational modularity (Langlois and Robertson 1992; Ulrich 1995; Baldwin and Clark 2000). As also said above, fundamental to this literature is the definition of principles that allo­ cate functions to components, identify the operating principle of each compon­ ent, and determine the interfaces among modules (Baldwin and Clark 2000). Within the boundaries set by these principles, modularity renders complexity manageable by making it possible to run parallel experiments that pursue altern­ ative explorative paths at the level of modules (Baldwin and Clark 2000). The case of the microelectronics industry is often pinpointed as the archetypi­ cal example of the progressive simplification of an interconnected structure through the emergence of simplified, decentralized and globally dispersed net­ works of suppliers of modular components. For example, Sturgeon (2002) has analysed the rise of contract manufacturing in electronics: namely, firms that take over electronic product design from other firms, and do the detailed engin­ eering and manufacture. The technological convergence is based on increasing automation of routine operations (e.g. component insertion), and on the increas­ ing use of standard software tools. The case of contract manufacturing in elec­ tronics is said to exemplify a ‘new’ way of organizing business on a global scale, leading to a neatly specialized system for the production of innovations, with product and systems designers, their components and subsystems sub-­ contractors, and their manufacturers, working together through arm’s-length market relationships. Schilling and Steensma (2001), on the basis of a compre­ hensive survey of US firms, provided evidence of the diffusion of modularity as an organizational strategy in a number of industries. Processes of knowledge codification seem to be pivotal in explaining the changing organization of innovative labor. For example, as long as it is the case that, as Sturgeon (2002) put it, ‘linkages are achieved by the transfer of codified information’ very little room is left for the traditional advantages of communica­ tion enabled by joint ownership (Arrow 1974). Ownership is not the bond that makes a collection of functional units into an organized entity (i.e. a firm). Such a bond is not needed, some argue, because increasingly modularized and codi­ fied knowledge (embodied in modular components) and the transfer of codified information would be enough to achieve coordination (Argyres 1999). In other words, as pointed out by Sanchez and Mahoney (1996), since components’

74   S. Brusoni and R. Fontana interfaces are not permitted to change within a certain period of time, a modular architecture would create an ‘information structure’ that smoothly co-­ordinates decentralized design teams. Thus, the ‘information structure’ would also act as a ‘compensation mechanism’ that holds the systems together without the need to exert explicit managerial authority. The point of departure we take from the above literature is that it is very often assumed that modular products rely on standardized and open interfaces. While we do not question the first feature, we do believe that openness is not neces­ sarily a core feature of any modular products. In other words, modular products do rely on standardized interfaces, but the latter may be either open or closed. In principle a standardized interface can be defined as ‘open’ when it does not embody proprietary specifications. In the Local Area Networking (LAN) indus­ try for instance, hardware interfaces are standardized and open in the sense that they respect technical specifications that are non-­proprietary because they have been defined within standardisation committees (i.e. the IEEE or the ANSI). It has to be noted that reliance on non-­proprietary standards, however, does not automatically lead to platforms that are completely open. As it will become clear in the next section, in many hi-­tech industries, manufacturers build platforms around non-­proprietary standards but find many ways of keeping them closed enough to maintain the control of platform evolution. The openness (or lack thereof ) of the interfaces has major implications in terms of platform competition. Closed modular interfaces (like in the LAN industry) potentially allow incumbents to introduce upgrades without cannibaliz­ ing their existing installed base. Moreover, customers have stronger incentives to become early adopters, as they can anyway upgrade their products by installing a modular upgrade. Of course, open standards also provide advantages by facili­ tating component interconnection and compatibility across manufacturers that may help new firms to enter new markets. The interplay between a characteristic of the platform (i.e. being modular, non-­modular) and the type of interface (open, closed) also has implications for platform competition. It is often claimed that modularity is associated with an increase in market competition because it lowers demand side switching costs and entry barriers (Schilling 2000). Indeed this may happen only when interfaces are fully open. Drawing on the interplay of modularity versus non-­modularity and openness versus closure, the next section presents a typology of platforms in different industries. Section 4 will discuss the competitive implications of the typology.

3  A typology of platforms In this section we review a series of extant empirical studies, to provide some exam­ ples of platforms and industries that can be studied on the basis of the framework introduced in the previous sections. This type of analysis will allow us to gain a better understanding of the different opportunities of innovation offered by each platform type and of the reasons why, at the industry level, relative advantages and disadvantages of incumbents versus potential entrants may dramatically differ.

Strategies for platform competition   75 3.1  The workstation industry as an example of modular-­open platform Modular-­open platforms combine low extent of proprietariness of interfaces with a modular architecture. The workstations market during the mid-­1980s offers an example of how competition from a modular-­open platform can challenge the position of an incumbent. In particular, the competition between Apollo and Sun Microsystems aptly illustrates the advantages of relying on non-­proprietary inter­ faces. Both Apollo and Sun offered apparently similar workstations. Both were modular systems revolving around the Motorola processor. No company was eager to integrate upward in the production of components and both depended on third parties for the supplying of application software (Baldwin and Clark 1997). Despite these similarities, the two competitors displayed different approaches to product design. Apollo (the incumbent) offered a system whose design was modular but intrins­ ically closed. Proprietary interfaces were defined at both the software and the hard­ ware level. At the software level in particular, both the network management system and the O/S were proprietary and closely tied to hardware components (i.e. peripherals, storage etc.). This ‘modular-­closed’ strategy of product design derived from Apollo’s background in the minicomputer industry where firms were used to developing fully integrated systems. When Sun’s workstation was commercial­ ized, Apollo had a 60 per cent share in the workstation market. Sun Microsystems (the entrant) built its system around a modular but open design. In particular, as discussed in Baldwin and Clark (1997: 134–6), the architecture was redefined around several ‘open’ standard components such as a Motorola processor, an Eth­ ernet and a UNIX operating system. Two proprietary hardware components (a memory management unit and a 32 bit internal memory bus) coupled the CPU with the internal memory. This design gave Sun’s workstations several competit­ ive advantages. First, choosing UNIX as the operating system turned out to be particularly appealing for many users who, by exploiting its modular structure, could add new features. Second UNIX was ‘portable’ across several hardware platforms and could be run on a variety of different machines which made Sun’s workstations compatible with the installed base of DEC minicomputers that were used in existing networks. Third, remodularization increased efficiency in the design and turned out to bring about considerable savings in terms of capital requirements. It is not difficult to see how these advantages were the consequence of the choice of a modular-­open platform. As argued by Baldwin and Clark (1997: 138), while non-­proprietariness was the main source of the first two advantages, the third one derived specifically from the redefinition of the architecture and ‘the pursuit of the modularity paradigm in production as well as in engineering’. 3.2  The Local Area Networking (LAN) industry and the software industry as examples of a modular-­closed platform Modular-­closed platforms combine a medium-­high degree of proprietariness of the interfaces with a modular product design. Being LANs, an example of

76   S. Brusoni and R. Fontana components systems technologies, they are characterized by modularity in both design and production and rely heavily on the presence of open standards. At both the hardware and the software level, open standards should be considered as ‘enabling technologies’. They have value in the market only if embodied in products and can be used as ‘channels’ to facilitate complementary innovations (Gawer and Cusumano 2002: 55). When standards are released they provide important information on the technical specifications. However, specifications may be open (i.e. non-­proprietary) but the way they can be actually implemented in products may differ across manufacturers. In particular, manufacturers have a strong interest in keeping control of the plat­ form (for instance, they may profit from the development and commercialization of a range of complementary compatible products). One way to leverage the supply of complementary products while designing products based on open standards is to develop a company specific software platform. When this plat­ form is implemented, pieces of equipment produced by the same manufacturer ‘work better’ together than they would do on a rival software platform. All the major players in the networking industry (Cisco, 3Com, Cabletron Bay Net­ works) have developed specific software platforms. The case of the Internetwork Operating Systems (IOS) developed by Cisco Systems is particularly interesting for the leadership that Cisco has been acquiring in the networking industry throughout the 1990s also thanks to this strategy. The IOS is a type of software developed by Cisco to make its products work and perform well together in the network. IOS consists of many lines of proprie­ tary code that enable Cisco products (mainly routers and switches) to interoper­ ate to be deployed together in the same network. Although Cisco tends to release the code to its customers, an important portion of IOS remains proprietary. Inter­ faces are in principle open but licensees cannot fully modify the code. The open­ ness of the platform is therefore restricted to that portion of the code that licensees need to know to make their products interoperable with Cisco or with those of other manufacturers supporting the IOS. Gawer and Cusumano (2002: 176) define this type of platform ‘open-­but-not-­open’. It has to be noted that the case of the IOS is not the only example of how a software platform can be used to close a system that would otherwise be open. A more general practice in the networking industry consists in developing software platforms, particularly for network management (i.e. the CiscoFusion in the case of Cisco Systems and the Spectrum platform in the case of Cabletron Systems), through which manufacturers can increase the costs of switching to another sup­ plier.2 This strategy limits modularity-­in-use by restricting the extent to which customers can mix-­and-match products from a variety of manufacturers. As a consequence, manufacturers can keep the control of their installed base of cus­ tomers since the presence of an installed base of equipment from a particular manufacturer increases the likelihood of repurchasing from the same vendor if the customer decided to buy again (Chen and Forman 2006). The LAN industry provides also an interesting example of the benefits arising from the presence of a modular-­closed hardware platform. In PC-­based

Strategies for platform competition   77 networking, modularity in hardware started playing an important role at the beginning of the 1990s, a period in which several open and incompatible com­ munication standards (i.e. FDDI, Fast Ethernet, ATM) were ‘battling’ in the market to become the successor to Ethernet, the most largely diffused standard at the time. To compete in this environment characterized by rapid technical change and great uncertainty, manufacturers, particularly of hub equipment, one of the most diffused pieces of equipment at the time, decided to develop modular design. The hardware component of hub equipment revolves around a backplane processing packets and regulating data traffic. Until 1990 the design of the hub was very simple and consisted mainly in a single bus along which data packets travelled in and out the equipment. A single bus design coupled with a transmis­ sion speed of 10 Mbps (Ethernet) was enough to handle all the traffic generated by a few interconnected computers (mainly minis and early microcomputers). This design started to become inadequate when data traffic rapidly increased fol­ lowing the diffusion of PCs, Internet applications and the subsequent advent of client server computing. Hub equipment started representing bottlenecks for the network and some changes in their design were needed. When faster standards were developed (Fast Ethernet potentially runs at 100 Mbps), the design of hub equipment had to change to be able to fully reap the benefits promised by the higher speed. In particular, the design of the internal bus evolved to accommodate the new specifications. Both ‘multi-­bus’ and ‘reconfigurable bus’ designs were proposed. One peculiarity of these design changes was that they were incremental and some of them initially based on pro­ prietary technologies that subsequently were licensed to other manufacturers. Changes in product design and support for the high speed standards went hand in hand thanks to the decision to follow a modular strategy. Relying on modular­ ity, manufacturers could reap the benefits in both production and design by adapting existing interfaces to the new specifications, offering high speed in add-­on modules and bringing new products to market very fast. This strategy allowed them to keep pace with technical change, hedge against uncertainty at the time when it was not sure which standard would have prevailed in the market place and avoid cannibalization of their product line since new standards could be incorporated into existing equipment. It is interesting to note that while a modular-­closed hardware platform enabled hub manufacturers to rapidly change their products’ design, the same strategy did not turn out to be effective in penet­ rating into the new market when it opened up. Indeed, as stressed by Brusoni and Fontana (2005) modular incumbents in the hub market were laggards in entering the switch market where competition required the development of radical product designs. Another well-­known example of a modular-­closed platform coming from the software industry is Microsoft Windows. Windows programming interfaces are ‘open’ in the sense that specifications are distributed to developers. However interfaces they are also ‘not open’ in the sense that the source code is not freely made available. As argued by Schilling (2000: 330): ‘by strategically excluding

78   S. Brusoni and R. Fontana some vendors from access to the full details of the interface, Microsoft retains control over what products can be made compatible’. Although they can both be considered examples of modular-­closed platforms, the cases of Cisco and Microsoft Windows differ in one important aspect which is related to the way the two companies manage the development of their plat­ forms. While Cisco has always tried to manage its platform without explicitly overlapping with the markets in which the manufacturers of complementary products operate, Microsoft has explicitly leveraged upon the ‘open-­not-open’ nature of its platform to expand Windows by incorporating more and more func­ tions into its operating system. As a result, the product has becomes less and less modular in the sense that the one-­to-one mapping between the functional ele­ ments and the physical component of the O/S (i.e. the portions of code) tended to blur as the operating system has increasingly taken over the roles of many other application programs (i.e. browsers, multimedia players, instant messaging systems, etc., etc.). Finally, it is interesting to notice how competition in these platforms resem­ bles to a certain extent the type of competition that prevailed in the mainframe market. In particular, the IBM System/360 system that epitomizes the ‘main­ frame paradigm’ was modular but not open. Underlying visible design para­ meters that contributed to make the system modular there were proprietary and hidden (i.e. secret) interfaces that made it difficult to develop and attach new modules to the system. Moreover, in the case of mainframes, only the hardware was modular. Software was proprietary and non-­modular. Indeed, all these cases seem to point to the role of software as the key component which market leaders should leverage upon in order to reduce the extent of openness of a platform and be able to keep the control of its evolution. As argued by Schilling (2000: 329): ‘by encapsulating proprietary technology within a component that conforms to an open standard-­based architecture firms can reap the advantages of compatibil­ ity while still retaining the rent-­generating potential of their proprietary compon­ ent’. Software may be the ‘Trojan horse’ that makes the encapsulation possible thus contributing to closing the boundaries of an open platform. 3.3  The Linux kernel software and the automotive industry as examples of open-­non-modular platforms The presence of non-modular-open-platforms seem to characterize two appar­ ently different products such as the Linux open source software and cars. Con­ sider the case of the Linux kernel for example. Developed during the 1990s by Linus Torvald, Linux has its origins in the UNIX operating system. UNIX was fully modular and open until 1979 when AT&T, the original developer of the software, decided to stop licensing the source code. Following this decision, several projects to develop UNIX-­like operating systems were started. In 1984 Richard Stallman started redeveloping UNIX on a free software basis giving birth to the so called ‘GNU project’ but it had to redesign the kernel.3 In 1991 Torvald filled the vacant space by developing a version of the GNU/Linux kernel

Strategies for platform competition   79 for his own use and then decided to reveal the code on the Internet to attract con­ tributions from other developers and transform the kernel into a fully working system. Since then Linux has attracted the attention of an increasing number of developers, the number of users has rapidly grown and reached in 1999 a 24 per cent market share for server operating systems (Kogut and Metiu 2001). The Linux platform has two important characteristics. First, its source code is not proprietary (i.e. it is open). This entails that each developer may propose their variants to the existing program that will be evaluated and eventually selected. Second, the community of developers is distributed but subject to hier­ archical control. This structure of governance makes the development of the Linux kernel centralized and less modular. Indeed, reliance on open source code stimulates developers to write specifications which promote customization. However, too much customization can be detrimental for the integrity of the operating system and many specifications are not incorporated into the different versions of the systems that are distributed to the community. Torvald collects the patches from the developers, selects them and decides whether or not to incorporate them into the new version. He is the coordinator of the project and ultimately responsible for its advance. The hierarchical nature of the develop­ ment also reduces the extent to which modularity plays a crucial role for the advance of the core of the platform. Indeed this is reflected in the few numbers of people who actually contribute to the development of the kernel. Most of the members of the community instead participate in the process of debugging rather than code writing. As argued by Kogut and Metiu (2001: 260) ‘[. . .] it is not the modularity that gives the open source a distinctive source of advantage because it too relies on hierarchical development. Rather the source of its advantage lies in the concurrence of development and debugging’. For slightly different reasons the automotive industry provides a similar example of an open-­non-modular platform. The automotive industry is rather known for its transition from the proliferation of modular-­closed platforms that characterized the industry from its inception to modular-­open that started emerg­ ing from the 1920s. As stressed by Langlois and Robertson (1992) this transition was led by small manufacturers who proceed to set detailed standards for many components creating interchangeability across firms. Larger manufacturers instead continued to rely on proprietary standards. A movement from modular toward less modular platforms is instead more recent and is still taking place driven by two factors. The first factor is the ‘electronification of the car’. Indeed, although from the technological viewpoint, the architecture of a car has been rather stable since the 1920s, the switch to the electro-­mechanical architecture has impacted in new ways on the interdependencies between the components of the system. These interdependencies are difficult to understand and to a certain extent unpredicta­ ble. This is particularly true for the connections between the body of the car, the chassis, the engine and the drive-­train that need to be understood to achieve a balance between noise, vibration and harshness and produce a workable car (Sako 2003: 233). Since these interdependencies are difficult to understand and

80   S. Brusoni and R. Fontana their effect on the overall product unpredictable, a centralized coordinator is needed to retain the architectural knowledge required to adjust the interfaces and eventually redesign them. The presence of this ‘system integrator’ reduces the extent of modularity of the platform. A second, and related, factor is the hierarchy of product components that characterizes the design of a car. As argued by Ulrich (1995), ‘pure’ modular­ ity would entail both standardized open interfaces and one-­to-one mapping between the physical component of the system and their functional elements. As cars become more similar to computers it is likely that they are going to experience more frequent upgrades driven by customer requirements. However, this is likely to occur only at low levels of the design hierarchy when the boundaries of components are more finely defined. At an upper level (i.e. the engine) large ‘chunks’ of components exists that would tend to contain multiple functions (Sako 2003: 232). Thus modularity in the sense defined by Ulrich is limited. 3.4  The minicomputer and the early mobile phone industry as examples of non-­modular-closed platforms In non-­modular-closed platforms a high degree of proprietariness of interfaces is accompanied by an absence of modularity in both product and design. These fea­ tures were typically found in minicomputers. Minicomputers typically employed a proprietary processor, a proprietary system bus and were running a proprietary operating system all manufactured by the same company. The competitive advantage of minicomputers with respect to mainframe was that they were cus­ tomized to the specific needs of customers and allowed them to eventually develop new software to suit their customers’ changing needs. A more recent example of a non-­modular-closed platform is constituted by mobile phones. Mobile phones are part of complex telecommunication systems. Within these systems interdependencies across physical components (i.e. switches, transmis­ sion equipment, base station controllers) are particularly strong and difficult to manage especially when components are experiencing very fast and uneven rate of technical change (Davies 1999). Within these systems, making the com­ ponents ‘working well together’ is also crucial for the performance of the overall system, as we also saw above in the case of Local Area Networking. This is easier when interfaces are open and technical compatibility exists across com­ ponents because compatibility allows interconnection. However, synchroniza­ tion between the functioning of all the components of the system and adaptation of the components to the condition that may affect the transfer of voice and/or data (Steinmueller 2003: 137) are also needed. Synchronization and adaptation may be difficult to achieve when components are produced by many different suppliers. As a consequence all the major players (Ericsson and Nokia in par­ ticular), especially in the early years of the industry, were both manufacturers of the handset and also able to install the overall equipment necessary for the func­

Strategies for platform competition   81 tioning of the network. This gave rise to alternative non-­modular and closed platforms. As argued by Davies (1999) being integrated suppliers of the system allowed firms to verify, check and eventually adjust interconnections and adapt more easily to the needs of the network operator.

4  Implications for platform competition and incumbents’ strategy Two types of implications can be drawn from our analysis in the previous sec­ tions. The first implication concerns the characteristics that are likely to be dis­ played by industries in which each type of platform may prevail. The second implication concerns the type of competition that is likely to occur in these industries. In particular, by using our typology (depicted in Table 3.2) we are able to predict how incumbents may respond to new challenges and may or not maintain their leadership. 4.1  Modular -open platforms (top left quadrant of Table 3.2) Modular-­open platforms are characterized by the presence of non-­proprietary interfaces and modularity. As discussed in the previous section modularity in design and production has advantages mainly in terms of design efficiency which translate into capital savings. The presence of non-­proprietary interfaces offer advantages in terms of both availability of off-­the-shelf components and enhanced compatibility across systems which is beneficial to both manufacturers and users. The relative advantages of this type of platform emerged clearly in the competition between Apollo and Sun Microsystems where the presence of an open platform rather than modularity in itself that seems to have made the real difference. The design of Apollo’s workstations was modular and in principle able to reap benefits of economies of substitutions in terms of performance and Table 3.2  A typology of platforms and industries Platform architecture

Type of interfaces

Open

Modular

Non-modular

Workstation and PC Industry Short term success in terms of entry into new segments Loss of control in the long run

Linux kernel and Automotive Industry Important role of systems integration. Strategic choice of key components and capabilities to keep the control of ‘supply chain’

Closed Networking and Software industry Incumbents maintain competitive position if innovative processes are fast and incremental in nature

Early Mobile Phone industry and Minicomputer industry Niche strategy, sophisticated users, in-house development

82   S. Brusoni and R. Fontana speed of upgradeability. However, it was the specific choice of Sun to redefine the system around open standards, particularly the choice of UNIX as the soft­ ware, which gave Sun a competitive advantage against the incumbent. Sun’s design was modular, but most of all it was open. As argued by Baldwin and Clark (1997: 128): ‘Sun’s full blown modularity beat Apollo’s limited modularity’. For these reasons, competition from a modular-­open platform may be difficult to sustain for an incumbent, especially if it is pursuing a closed strategy, and may represent a source of considerable gains in the short term. It is important to stress however, that there are also important risks associated with competition on the basis of modular-­open platforms. In particular, they may undergo the risk of remodularization. Remodularization occurs when the ‘the hidden interior of the former interface becomes a component, and hence prey to competition from alternative versions that conforms to the now-­separate interface definitions’ (op. cit. p. 152). When it started the PC industry, IBM opted for a modular-­open plat­ form. Most of the components of IBM PCs were open and their production out­ sourced to third parties who could also sell them to others (this is what Microsoft did with its O/S). The core of the PC was the basic input-­output system (BIOS) that was stored in memory. The BIOS ‘tailored’ the operating system to the specifications of a particular machine and its code was proprietary (Ceruzzi 1998: 277). Compaq first and then Phoenix Technologies reverse-­engineered the BIOS chip and offered it for sale thus breaking though IBM’s control of the PC architecture, a development that gave birth to the so-­called ‘market for clones’ and signalled the end of IBM dominance in the PC industry. In the presence of modular-­open platforms the same factors that help new entrants to challenge incumbents in the short run may also represent sources of risk in the medium-­ long run. To maintain their leadership within the context of a modular-­open plat­ form, incumbents may either continuously engage in redefining the system or improve its performance over competitors. 4.2  Modular-closed platforms (bottom left quadrant of Table 3.2) Both the case of the internet working industry and that of the software industry shed some light on the dynamics we should expect to prevail in industries where modular-­closed platforms prevail. This type of platform allows a rapid move­ ment across the design space if the firm is an incumbent. It seems ideal for com­ peting in contexts characterized by rapid although incremental technical change. It allows firms to increase the speed of upgrades while catering for different types of customers without cannibalizing the existing installed base. The late tra­ jectory of hub equipment in the LAN industry clearly illustrates this case. More generally, incumbents relying on a modular-­closed platform can respond to a perceived threat to their leadership by extending their platform to incorporate existing features that were previously excluded by adding new features. This is the strategy was followed by Microsoft with the Windows O/S that progressively incorporated a series of applications (i.e. an internet browser, a media player,

Strategies for platform competition   83 etc., etc.) that were previously developed and commercialized by other manufac­ turers. As new functionalities are added and interfaces are being kept almost pro­ prietary, the product becomes less and less modular. This strategy is particularly effective when innovation is mainly incremental. When innovation is radical, incumbents competing on the basis of modular-­closed platforms may be late to respond to radical innovations. Indeed this is what happened to modular hub manufacturers when the switch was introduced. A delay in reacting may lead to a loss of control of the platform itself. To avoid losing control of the platform, incumbents can try to grow through mergers and acquisitions of new innovators both to reduce competition and to acquire the competencies they need to master innovation especially when technical change occur very fast. This is clearly the strategy followed by Cisco Systems, the major incumbent in the networking industry, which is widely known for basing most of its growth on mergers and acquisitions. 4.3  Non-modular-open platforms (top right quadrant Table 3.2) Non-­modular-open platforms combine the presence of non-­proprietary interfaces with a not fully modular system, usually a consequence of the presence of a hier­ archy either in the structure of governance or in the platform technology which leads to the absence of a one-­to-one mapping between the physical components of the platform and their functions. When platforms are non-­modular and open the need for some form of central coordination arises. The coordinator should act as a ‘system integrator’ and perform different tasks. It should take strategic choices in terms of both the identification of the key components of the platform that need to be modularized and the types of capabilities that need to be developed to maintain the control of the platform. It should check, fine-­tune and enforce the principles that define interconnections among the interfaces (Sako 2003). What is required to make these choices is an extent of architectural know­ ledge in excess of what the integrator may actually make within the system (Brusoni et al. 2001). This role is likely to be played by incumbents such as big car manufacturers placed at the centre of a network of component suppliers in the automotive industry or the original developers of the platform such as Linus Torvald in the case of the Linux kernel. 4.4  Non-­modular-closed platforms (bottom right quadrant Table 3.2) Non-­modular-closed platforms are characterized by the presence of proprietary interfaces and absence of modularity. Competition within this type of platform is based on proprietary interfaces, in-­house development of components and is, especially in the early phases of the industry, ‘niche oriented’ (i.e. it is aimed at fulfilling the needs of a specific category of rather sophisticated customers). Minicomputers and the early years of the mobile phone industry are examples of this type of platform competition. Both industries also provide some hints on how the strategies of incumbents may evolve as the industry grows. As argued

84   S. Brusoni and R. Fontana by Schilling (2000) in minicomputers a movement toward more modular design occurred during the 1980s. The VAX family developed by DEC during the 1980s still retained some of the characteristics of previous systems (i.e. a single architecture, a single O/S – the VMS) that were fully proprietary by Digital. However they also started embodying an open standard for networking (Ether­ net) that DEC obtained in an agreement with Xerox and Intel. Data General, another incumbent, increasingly relied on open software such as UNIX. This move toward a more open platform enabled incumbents to survive the shakeout in the minicomputer industry induced by the arrival of the microprocessor and to sustain increasing competition from workstations. Also in the mobile phone industry, we are witnessing a movement toward platforms that are becoming more open and modular. The decision of mobile phone operators to focus on the supply of value added services to the final customer has put even more pressure on the mobile phone suppliers in terms of responsibility for network design and build. As a consequence, incumbents are repositioning themselves toward activ­ ities that enable them to keep up with this requirement and are also more profita­ ble in term of value added. This is accompanied by a progressive modularization of production. As discussed by Davis (2003: 351), ‘a growing portion of Erics­ son’s products – including exchange equipment, 3G audio base stations and mobile handsets – are now outsourced and manufactured under contract’.

5  Overview and conclusions This chapter has developed the idea that products and organizations are complex systems made up of many interconnected elements. How these elements are con­ nected is a matter of technological and organizational design choices. Such choices define and organize some aspects of the inner working of complex systems, define the strengths of the connection among elements, the entry for new components and the exit for obsolete components. Design choices may also provide the foundations of what has been recently called a ‘platform’, i.e. ‘an evolving system made of interdependent pieces that can each be innovated upon’ (Gawer and Cusumano 2002). In this chapter we have provided a simple frame­ work to identify different types of platform. Relying on the interplay between platform architecture (modular versus non-­modular) and type of interface (open versus closed) a typology of four platforms has been proposed. By reviewing a series of extant empirical studies, some examples of platforms and industries have been studied on the basis of the proposed typology. It has been argued that within each platform, the rules of competition, the opportunities of innovation, the relative advantages and disadvantages of incumbents versus potential entrants dramatically differ. Such differences have been explored and implica­ tions for platform competition and incumbents’ strategy drawn. Our examples have shown that the choice of a modular architecture is gener­ ally beneficial for incumbent firms who can maintain their competitive position especially when innovative processes are fast and incremental in nature. However, the choice of the type of interface is crucial. While opting for a closed

Strategies for platform competition   85 interface may entail loss of control in the long run, choosing an open interface may open the door to the ‘gales of creative destruction’ in the form of competi­ tion from new entrants. These results have important implications for platform managers who have to weigh the short-­term benefits of choosing a specific type of interface against its long-­term consequences. Incumbents may also benefit from the choice of a non-­modular architecture. However, also within this context, the outcome of platform competition depends highly on the type of interface. Opting for a specific type of interface has differ­ ent implications for management. Open interfaces are more likely to be chosen by firms that possess the capabilities to act as ‘systems integrators’ in order to choose strategically key components and capabilities to keep in control of the supply chain. Incumbents that do not possess systems integrators’ capabilities should instead opt for closed interfaces. In this case, in-­house development will be coupled with catering to the sophisticated users and niche strategy.

Notes 1 Financial support from the European Commission (FP6) Project: KEINS – Knowledge­Based Entrepreneurship: Innovation, Networks and Systems, Contract n.: CT2-CT­2004–506022 is gratefully acknowledged. 2 There are extensive economies of learning in the management of big communication networks. Management usually entails the connecting and setting-­up of new LAN sta­ tions as well as troubleshooting. Learning how to use the software to perform these activities takes time and requires extensive training. When the software is manufacturer specific, the costs of switching to a different supplier would rise. 3 The kernel is the set of O/S components containing the most basic functions such as system access management, file storage, memory use, device drivers and processor scheduling.

Bibliography Ancori, A., Bureth, A. and Cohendet, P. (2000), The Economics of Knowledge: The Debate about Codification and Tacit Knowledge, Industrial and Corporate Change, 9: 255–87. Argyres, N.S. (1999), The Impact of Information Technology on Coordination: Evidence from the B-­2 ‘Stealth’ bomber, Organization Science, 10: 162–80. Arora, A. and Gambardella, A. (1994), The Changing Technology of Technical Change: General and Abstract Knowledge and the Division of Innovative Labour, Research Policy, 23: 523–32. Arora, A., Fosfuri, A. and Gambardella, A. (2001), Markets for Technology and Their Implications for Corporate Strategy, Industrial and Corporate Change, 10: 417–49. Arora, A., Gambardella, A. and Rullani, E. (1997), Division of Labour and the Locus of Inventive Activity, Journal of Management and Governance, 1: 123–40. Arrow, K.J. (1974), The Limits of Organization, New York, Norton. Baldwin, C.Y. and Clark, K.B. (1997), Sun Wars. Competition Within a Modular Cluster, 1985–1990, in D.B. Yoffie (ed.), Competing in the Age of Digital Convergence, Boston MA, Harvard Business School Press, pp. 123–57. —— and —— (2000), Design Rules: The Power of Modularity, Cambridge MA, MIT Press.

86   S. Brusoni and R. Fontana Bonaccorsi, A. and Pammolli, F. (1996), Knowledge as a Product, Knowledge as a Process – Scientific-­Technological Research and Organizational Firms, in J. Butler and A. Piccaluga (eds), Knowledge, Technology and Innovative Organizations, Milan, Guerini. Brusoni, S. and Fontana, R. (2005), Modularity as an Entry Strategy: The Invasion of New Niches in the LAN Equipment Industry, Working Paper 171, CESPRI – Bocconi University, Milan. Brusoni, S. and Pavitt, K. (2003), Problem Solving and the Coordination of Economic Activities, SPRU WP Series, no. 93, University of Sussex at Brighton. Brusoni, S. and Prencipe, A. (2001), Unpacking the Black Box of Modularity: Technolo­ gies, Products, Organisations, Industrial and Corporate Change, 10: 179–205. Brusoni, S., Prencipe, A. and Pavitt, K. (2001), Knowledge Specialisation, Organizational Coupling and the Boundaries of the Firm: Why Firms Know More Than They Make?, Administrative Science Quarterly, 46: 597–621. Ceruzzi, P.E. (1998), A Modern History of Computing: 1945–1995, Cambridge MA, MIT Press. Chen, P.Y. and Forman, C. (2006), Can Vendors Influence Switching Costs and Compati­ bility in an Environment with Open Standards, MIS Quarterly, 30: 1–22. Chesbrough, H. and Kusunoki, K. (2001), The Modularity Trap: Innovation, Technology Phase Shifts, and the Resulting Limits of Virtual Organizations, in I. Nonaka and D. Teece (eds), Managing Industrial Knowledge: Creation, Transfer and Utilization, Thousand Oaks CA, Sage. Christensen, C.M. (1997), The Innovator’s Dilemma: When New Technologies Cause Great Firms to Fail. Boston MA, Harvard Business School Press. Christensen, C.M. and Rosenbloom, R.S. (1995), Explaining the Attacker’s Advantage: Technological Paradigms, Organisational Dynamics, and the Value Network, Research Policy, 24: 233–257. Davies, A. (1999), Innovation and Competitiveness in Complex Product Systems: The Case of Mobile Phone Systems, in S. Mitter and M. Bastos (eds), Europe and Developing Countries in the Globalised Information Economy, London, ONU Press. —— (2003), Integrated Solutions. The Changing Business of Systems Integration, in A. Prencipe, A. Davies and M. Hobday (eds), The Business of Systems Integration, Oxford, Oxford University Press. —— (2004), Moving Base into High-­Value Integrated Solutions: A Value Stream Approach, Industrial & Corporate Change, 13: 727–56. Ethiraj, S.K. and Levinthal, D. (2004), Modularity and Innovation in Complex Systems, Management Science, 50: 159–73. Forman, C. and Chen, P. (2004), Switching Costs and Network Effects in the Market for Routers and Switches, mimeo, Graduate School of Industrial Administration, Carnegie Mellon University. Gambardella, A. and Torrisi, S. (1998), Does Convergence in Technology Imply Conver­ gence in Markets? Evidence from the Electronics Industry, Research Policy, 27: 445–63. Garud, R. and Kumaraswamy, A. (1995), Technological and Organizational Designs for Realizing Economies of Substitution, Strategic Management Journal, 16: 93–110. Gawer, A. and Cusumano, M.A. (2002), Platform Leadership: How Intel, Microsoft, and Cisco Drive Industry Innovation, Cambridge MA, Harvard Business School Press. Grandori, A. (1997), An Organizational Assessment of Inter-­Firm Coordination Modes, Organization Studies, 6: 897–925.

Strategies for platform competition   87 Granstrand, O., Patel, P. and Pavitt, K. (1997), Multitechnology Corporations: Why They have ‘Distributed’ rather than ‘Distinctive Core’ Capabilities, California Management Review, 39: 8–25. Hobday, M. (1998), Product Complexity, Innovation and Industrial Organisation, Research Policy, 29: 871–93. Kogut, B. and Metiu, A. (2001), Open-­Source Software Development and Distributed Innovation, Oxford Review of Economic Policy, 17: 248–64. Langlois, R.N. and Robertson, P.L. (1992), Networks and Innovation in a Modular System: Lessons from the Microcomputer and Stereo Component Industries, Research Policy, 21: 297–313. Levinthal, D.A. (1997), Adaptation on Rugged Landscapes, Management Science, 43: 934–950. Marengo, L. and Dosi, G. (2005), Division of Labor, Organizational Coordination and Market Mechanisms in Collective Problem-­Solving, Journal of Economic Behavior and Organization, 58: 303–26. Miller, D. (1987), The Genesis of Configuration, The Academy of Management Review, 12(4): 686–701. —— (1996), Configurations Revisited, Strategic Management Journal, 17: 505–12. Miller, R., Hobday, M., Leroux-­Demers, T. and Olleros, X. (1995), Innovation in Complex System Industries: The Case of Flight Simulators, Industrial and Corporate Change, 4: 363–400. Monteverde, K. (1995), Technical Dialog as an Incentive for Vertical Integration in the Semiconductor Industry, Management Science, 41: 1624–38. Mowery, D. (1983), The Relationship Between Intra-­Firm and Contractual Firms of Industrial Research in American Manufacturing, 1900–1940, Explorations in Economic History, 20: 351–74. Nelson, R.R. and Winter, S.G. (1982), An Evolutionary Theory of Economic Change, Cambridge MA, Harvard University Press. Padgett, J. and McLean, P. (2006), Organizational Invention and Elite Transformation: The Birth of Partnership Systems in Renaissance Florence, American Journal of Sociology, 111: 1463–568. Pisano, G.P. (1990), The R&D Boundaries of the Firm: An Empirical Analysis, Administrative Science Quarterly, 35: 153–76. Prencipe, A. (1997), Technological Capabilities and Product Evolutionary Dynamics: A Case Study from the Aero Engine Industry, Research Policy, 25: 1261–76. Romanelli, E. and Tushman, M.L. (1994), Organizational Transformation as Punctuated Equilibrium: An Empirical Test, The Academy of Management Journal, 37: 1141–66. Sako, M. (2003), Modularity and Outsourcing. The Nature of Co-­Evolution of Product Architecture and Organisation Architecture in the Global Automotive Industry, in A. Prencipe, A. Davies and M. Hobday (eds), The Business of Systems Integration, Oxford, Oxford University Press. Sanchez, R. and Mahoney, J.T. (1996), Modularity, Flexibility, and Knowledge Manage­ ment in Product and Organisation Design, Strategic Management Journal, 17: 63–76. Schilling, M.A. (2000), Toward a General Modular Systems Theory and Its Application to Interfirm Product Modularity, Academy of Management Review, 25: 312–34. Schilling, M.A. and Steensma, H.K. (2001), The Use of Modular Organizational Forms: An Industry-­Level Analysis, The Academy of Management Journal, 44: 1149–68. Siggelkow, N. (2001), Change in the Presence of Fit: The Rise, the Fall, and the Renais­ sance of Liz Claiborne, Academy of Management, 44: 838–57.

88   S. Brusoni and R. Fontana —— (2002), Evolution Toward Fit, Administrative Science Quarterly, 47: 125–59. Steinmueller, W.E. (2003), The Role of Technical Standards in Coordinating the Division of Labour in Complex System Industries, in A. Prencipe, A. Davies and M. Hobday (eds), The Business of Systems Integration, Oxford, Oxford University Press. Sturgeon, T.J. (2002), Modular Production Networks: A New American Model of Indus­ trial Organization, Industrial and Corporate Change, 11: 451–96. Ulrich, K.T. (1995), The role of product architecture in the manufacturing firm, Research Policy, vol. 24, pp. 419–40. von Tunzelman, G.N. (1998), Localised Technological Search and Multi-­Technology Companies, Economics of Innovation and New Technology, 6: 231–55. Wang, Q and von Tunzelman, G.N. (2000), Complexity and the Function of the Firm: Breadth and Depth, Research Policy, 29: 805–18. Weick, K.E. (1976), Educational Organizations as Loosely Coupled Systems, Administrative Science Quarterly, 21: 1–19. Williamson, O. (1975), Markets and Hierarchies. Analysis and Antitrust Implications, New York, The Free Press. —— (1985), The Economic Institutions of Capitalism, New York, The Free Press.

4 Linking technological change to organisational dynamics Some insights from a pseudo-­NK model Tommaso Ciarli, Riccardo Leoncini, Sandro Montresor and Marco Valente 1  Introduction This chapter defines the main elements of a research project aimed at investigating firm and industrial organisation by focusing on firms’ ability to understand and manage the knowledge about their production technology. In particular, the project addresses firms’ capacity to cope with the trade-­off between the exploration of technological possibilities and the exploitation of market-­mediated opportunities (Ciarli et al. 2007, 2008). We will argue that three main issues are at stake here, explore them and suggest a formalisation that integrates them. The first is the interrelationship between technological complexity and organisation. In particular, the question is how innovative efforts impact on firms’ performance and the way in which the various aspects of innovative activity is resolved within or outside a firm’s boundaries. The second issue is the management of decisions related to technology and organisation. In particular, the question is how firms, under specific behavioural assumptions, learn and produce new knowledge to implement efficient production processes. Finally, there is the economic systems’ dynamics and networking. Indeed, firms co-­ordinate their activities in many different ways, e.g. upon explicit contracts, with long-­term ‘market’ relations, or with explicit forms of co-­operation, and this crucially affects their techno-­economic behaviour. The remainder of the chapter is structured as follows. Section 2 presents our understanding of the literature on the above issues. In order to formalise these mechanisms, we first propose a model of research in complex landscapes: the pseudo-­NK model (Section 3). We then plug the pseudo-­NK into a simple model of technological competition within a market (Section 4), to study its properties and its dynamics under different configurations by means of numerical simulations (Section 5). Finally, Section 6 offers some concluding remarks on the main results, together with some hints on the main implications for further developments of the research project.

90   T. Ciarli et al.

2  The technology–organisation link The idea that the innovative process involves both technological and non-­ technological aspects, mainly of organisational nature, is at least as old as the history of economics. Smith (1776), for first, connected the introduction of a more efficient production process to a typical organisational trait of the firm that is ‘labour-­division’. Innovation found a more explicit organisational equivalent in the pioneering work of Joseph Schumpeter, who more than 70 years ago related it also to ‘the carrying out of the new organisation of any industry, like the creation of a monopoly position (for example through trustification) or the breaking up of a monopoly position’ (Schumpeter 1934). More recently, evolutionary economists have built up a neo-­Schumpeterian theory of economic change in which innovation works through the search of new organisational routines and in which the introduction of new products and/or production processes is contingent to the adoption of new organisational traits (Nelson and Winter 1982). Broadly speaking, we can classify the research on the relationships between technology and organisation in two streams of literature. The first deals with this relationship as a ‘simple’ link, that is as a causality relationship with a certain direction: either by analysing how technological change affects the firm’s organisational structure or how, vice versa, the latter affects the innovativeness of the firm. The second stream, instead, considers the same linkage as a ‘complex’ one, in which technology and organisation mutually affect each other and their inherent dynamics in a co-­evolutionary process. 2.1  Technology on organisation The simplest way to look at the relationship between technology and organisation is based on the analysis of the impact technology has on the governance structure of the firm: i.e. vertical integration – hierarchy-­based mechanisms – versus vertical-­disintegration – market-­based mechanism or outsourcing. Vertical integration is apparently more convenient than outsourcing to deal with the uncertainty technological and non-­technological changes determine in the relationships between the firm and its competitors. Indeed, following transaction cost economics, vertical integration would increase self-­enforcement as the agents will have less incentive to take advantage of the contract incompleteness generated by market disturbances (Williamson 1975:  23–5, 1985:  43–63). It is thus argued that market uncertainty presumably increases vertical integration and decreases outsourcing. Moreover, as Gonzalez-­Diaz et al. (2000) point out, if uncertainty interacts with specificity, the costs of re-­ contracting in front of higher uncertainty could be prohibitive, thus favouring vertical integration only if the transaction requires specific investments. However, if re-­contracting is easy (and considering, for example, the managers’ aversion to profit variability) outsourcing could be more convenient than vertical integration.

Technological change and organisational dynamics   91 Of course, the idea of exogenous uncertainty is not entirely satisfactory, especially in dealing with technological change. Accordingly, a number of concepts have been introduced which are more helpful in dealing with the issue. The concept of the ‘technology development path’, for example, as a set of technological characteristics of the environment in which the firm is embedded and to which the firm actively contributes, is one of them. From an evolutionary-­ innovation perspective, through outsourcing some valuable knowledge of the outsourcing firm may leak and lead the outsourcing firm to increase its innovativeness,1 depending on how the ‘technological regime’ (Dosi 1988) shapes the specific combination of technological opportunity and appropriability conditions, cumulativeness of learning and nature of the knowledge base (Mahnke 2001). 2.2  Organisation on technology Another way of linking technology and organisation is that of the literature addressing how the governance structure affects the firm’s innovation capabilities. Proponents of large vertically integrated firms, on the one hand, and those of networks of small specialised producers – in which disintegration and outsourcing play an important part – on the other hand, have been debating increasingly more in recent years. Vertically integrated structures are recognised, on the one hand, by the ability to access economies of scale in R&D, of co-­ordinating it with other firm activities and of solving relevant problems of power distribution and appropriability in innovating (Teece 1986). Market-­based structures, such as those implemented through outsourcing, have, on the other hand, the capacity of guaranteeing higher specialisation, greater flexibility and a superior adaptability to engender innovations (Sabel 1989). The results however are very much dependent on the kind of technological change and on the sector chosen (Robertson and Langlois 1995; Teece and Pisano 1994), although some clear-­cut arguments can be put forward, especially by referring to either a contractual perspective of the firm or a resource-­based one. Following the contractual perspective, we put forward a positive effect of organisation on technology, by advocating ‘governance inseparability’. An innovative firm, by carrying out a certain amount of research and development, could well extend its activities, together with the relative contractual agreements, to different future prospects. Accordingly, it would be more cautious than a non-­ innovative, or less innovative one, about entering into early commitments through vertical integration. In fact, it could be forced to a relatively low-­value added activity and model its governance structure around it, while finding costly, later on, to break governance inseparability in order to internalise other high-­ value added activities. A similar conclusion can be reached also by looking at the firm in terms of resources and capabilities. Indeed, firms have to find a proper balance between the ‘exploration’ of new capabilities and the ‘exploitation’ of their present ones (e.g. March 1991a; Leonard-­Barton 1992), considering that positive feedbacks from past experience might render exploiting current capabilities easier than

92   T. Ciarli et al. exploring new risky ones (Mahnke 2001). This bias often makes firms to fall into ‘competence traps’ (Levinthal and March 1993), hampering their capabilities-­upgrading. This adverse mechanism could however be overcome, and the innovation capabilities of the firm thus restored (Tushman and Anderson 1986), if the firm is able to experiment with an increase in its external knowledge interfaces and in its sources of learning-­by-interacting. As outsourcing is, in principle, able to foster these opportunities, a positive relationship between it and innovativeness can be envisaged.2 The impact on innovation of organisational issues can be better specified by distinguishing radical from incremental innovations, and product from process ones. In front of radical innovations, vertical integration could be necessary as external specialist suppliers might be unable to understand the viability of radical innovation when it is far from their current capabilities set (Silver 1984).3 A correlation between outsourcing and radical innovations can be envisaged also by thinking of radicality as measured by the firms’ distance from the relevant technological frontier and setting this idea at work in the framework of the property-­rights theory of the firm (Grossman and Hart 1986). In Acemoglu et al. (2005), for example, a non-­overlapping generations model is built up for an economy in which the equilibrium in firm organisation changes with the economy approaching the world technology frontier. Indeed, the distance from such a frontier affects the trade-­off between the flexibility of outsourcing and the control of technological change under vertical integration: when close to the frontier, incremental innovations are required and the advantages of flexible production can be exploited, when far from the frontier and radical innovations are needed, imitation activities are more important and vertical integration is preferred. 2.3  Technology and organisation The relationship between technology and organisation becomes inherently complex when dynamic capabilities are called for and considered from a complex system perspective. This implies going beyond the standard approach to capabilities, retaining some of their insights.4 In strategic management, for example, dynamic capabilities are conceived as a particular ‘ability’ with which the firm manages, more competitively than its rivals, to adapt its ‘internal and external competencies’ in front of a ‘changing environment’ (Teece et al. 1997: 516). However, how both integration and co-­ ordination of the firm’s capabilities favour learning is not taken into consideration. Furthermore, integration and co-­ordination capabilities are mainly conceived as internal features, and the role of external integration is only sketched. In the economics of innovation, instead, dynamic capabilities apply to certain technological elements, such as: the ‘knowledge’ to master a new product dominant-­design (Henderson and Clark 1990); the ‘skills and know-­how’ to deal with new products and processes (Tushman and Anderson 1986); and the

Technological change and organisational dynamics   93 recursive chain of ‘activities’ necessary to engage in technological problem-­ solving activities (Iansiti and Clark 1994). The firm is described in ‘technical’ terms, trying to tackle its technological complexity through its ‘technology integration capacity’ (Iansiti and Clark 1994: 565). However, the firm’s environment is in general portrayed as an ‘industry’ populated by ‘established organisations’ (incumbents) and ‘new entrants’, interacting through competitive relationships (Tushman and Anderson 1986:  445–6). Non-­competitive relationships among firms, and relationships with other institutions and organisations of a different nature, are instead given a marginal role in shaping organisational learning and the capabilities dynamics. In organisation studies, dynamic capabilities are seen as a means to contrast the organisational inertia and the path dependency associated with the learning process. In these accounts (March 1991b; Levitt and March 1988) the interactions between firm/organisation members, as well as organisation-­wide conditions and management models, are more stressed than in the previous approaches (Nonaka 1994). However, these relational considerations are only partially encapsulated in the dynamic capabilities view. First, although both formal and informal external communication channels are considered (op. cit.), the latter are not given any special attention, as they are treated as a mere extension of the standard intra-­organisational case. Finally, in evolutionary economics, dynamic capabilities are related to the minimum ontological element of the evolutionary firm, i.e. to its organisational routines. On the one hand, dynamic capabilities directly apply to ‘operational routines’ – rather than to generic competencies or capabilities – allowing the firm to generate and modify them whenever it is necessary (Zollo and Winter 2002; Zott 2002). On the other hand, dynamic capabilities are distinguished from the routines they apply, because they are intentional and deliberated. However, it must be stressed that the evolutionary framework is confined entirely within the boundaries of the firm, denying a direct dynamic role to the environment. Environmental factors are simply ‘viewed [. . .] as inputs to the dynamic capability building process, rather than part of the process itself ’ (Zollo and Winter 1999: 11). 2.4  Dynamic capabilities and complexity When seen from the complexity perspective, the nature of dynamic capabilities is quite different from the approaches we have referred to above. Dynamic capabilities apply to the firm as a system of capabilities, rather than to one or another of its constitutive elements (be they routines, competencies, or technology). Their core element is the firm’s capacity to undertake a complex adaptive process, rather than to its ability to implement an organisational kind of learning. In fact, such a capacity is neither purely organisational, nor merely consisting of its evolution in response to environmental changes. In particular, firms are typically treated as evolutionary entities (Nelson and Winter 1982), with at least three crucial implications. First, their learning

94   T. Ciarli et al. patterns are informed by a variety generation mechanism, centred on their organisational routines, thus calling for an organisational analysis of their dynamic capabilities. But their evolution is also determined by the selection process operated by the market forces, thus calling for an environmental analysis of their dynamic capabilities. A second implication is that, following a system perspective, the dynamics of the firm cannot be considered in isolation from the set of actors and relationships of the relative technological system. Not only because of the need for an institutional set-­up and an economic structure to carry out their business activities, but also, and above all, because firms learn (and innovate) by interacting with other firms and actors of the technological system (Lundvall 1992). An environmental kind of analysis is thus necessary in this last respect. A third implication concerns the nature of the relationships within a system (‘immaterial’ and ‘material’). While the immaterial relations are conducive to codified knowledge, whose diffusion does not require spatial proximity, closeness is instead essential as far as tacit knowledge is concerned. Its ‘embodied’ diffusion benefits, in fact, from geographical co-­location. Accordingly, while material relations spur a kind of network which can be strategically planned – and thus organisationally investigated – tacit knowledge provides ‘local’ systems of innovation (and production) with a special advantage in ‘socialising’ and ‘externalising’ organisational knowledge (Nonaka 1994) – thus calling for environmental/spatial analysis. In synthesis, the reference to the idea of complexity allows us to make dynamic capabilities more functional to the analysis of the technology-­ organisation relation we focus on. Moreover, it provides some analytical instruments through which the same relationship can be modelled and investigated. We next turn to our proposition of modelling technology as a complex space, which requires firms to undergo a complex search in a non-­modular environment. We first formalise a simple and flexible representation of learning in a complex environment. We then use this representation to model firms’ technological search process. We argue that such a system is a good basis to analyse firms’ technological capabilities and organisational strategies.

3  Modelling complexity NK models provide a formal representation of complex research spaces and the research strategies applied over these spaces (Frenken et al. 1999). The NK representation of a problem consists in: (i) a finite (albeit huge) list of potential solutions; (ii) a distance function upon the solutions; and (iii) a measure of fitness associated to each solution. The solvers’ strategies are defined in terms of algorithms (generally involving some stochasticity), determining, given the current solution, a finite set of different solutions that can be reached from there. Therefore, a research strategy generates a pattern of potential solutions that can be assessed in respect of its capacity to reach the highest fitness solution. Usually, a complex problem is defined when the solver’s strategy cannot guaran-

Technological change and organisational dynamics   95 tee the reaching of the best solution in a reasonable time, and, instead, it is highly likely to terminate in a local maximum after a finite number of steps.5 NK models have been very useful in studying the emergence of complexity and how it affects the search of optimal solutions for a complex problem. In particular, NK models can be used to represent a modularised search space, where the degree of complexity depends on the density of the interactions between modules. If there is no interdependence between modules, then the problem is ‘simple’ in the sense that the solver’s strategy can be applied in parallel to all the modules – each module state is not conditional on any other module state. We can therefore guarantee to obtain the optimal fitness solution to the problem in a relatively small number of steps. Otherwise, when the interdependence between the modules is strong, the best research strategy is bound to take a combinatorial number of steps to reach the optimal solution, or reach a relatively poor fitness (a local optimum). When we formalise a complex space in the NK metaphor, we represent the degree of interdependence among N modules by the K modules with which each module is connected. K can range from 0 (total independence, or full modularity) to N-­1 (maximum interdependence, or no modularity).6 When K = N-­1 the change of one module modifies the fitness contribution of all other modules, requiring the solver to analyse an overwhelming number of combinations at each step. The solver then can opt for a modular strategy, confining the search to subsets of the NK space. NK models have been used to determine formally the results of research strategies applied on different research spaces. Simplifying, we may summarise these results as presented in Table 4.1. A modular research strategy is fast to reach a result that allows for no further improvements, but, in general, does not guarantee to reach the optimal solution for complex problems. Conversely, an integral research strategy always reaches the best solution — as all possible combinations are evaluated at each step, but at the cost of being very slow. The NK model provides formal support to these results, and allows the analysis of any intermediate cases. For example, we can determine the best strategy in terms of the average results produced in a given time span, besides the limit result. NK models have been widely used because the modeller can freely set K and test properties and predictions of a complex space. For example, innovation is usually referred to as a complex technological space explored by firms with their R&D efforts. NK offers the possibility to formally represent the technological Table 4.1  Research spaces and solvers’ strategies Agent’s strategy

Problem Simple

Complex

Modular

Global maximum; fast

Local maximum; fast

Integral

Global maximum; slow

Global maximum; slow

96   T. Ciarli et al. environment and the firms’ research approaches. It can be shown that sub-­ optimal strategies, i.e. modular research in a complex environment, can be economically superior to theoretically optimal strategies. In fact, competition rewards not only technological quality, but also the time of its introduction: optimal solutions can be brought too late to markets, when faster firms, introducing frequent small improvements, already dominate the market (Valente 1999). Given its properties, it would be attractive to plug an NK model representing the technological problem of production and innovation into an economic model where firms adopt different strategies to explore for better inputs and technologies. Unfortunately, though theoretically feasible, the direct application of an NK model into a wider model, for example of a market, is extremely cumbersome. In fact, in order for the properties of NK models to be appreciated it is necessary to take the average behaviour observed in many runs of explorations of a landscape, while in an economic model we usually need to obtain results at each draw (or every few). However, the NK properties can be represented by other means, besides letting them emerge from the structure of the NK format of problems. We call pseudo-­NK models a class of mathematical systems replicating the properties we observe from NK models, but with a more flexible representation that can be easily used as part of larger models. Indeed, pseudo-­NK models are used as search algorithms for optimal solutions, which allow for a finite but large number of local optima. 3.1  The pseudo-­NK model Consider a real-­valued function of N independent real variables upper limited: _​› Max V:ℜN → ℜ such that V(  x ​ . For example, we may consider any bell-­  ​   ) < V  shaped function, like a Gaussian. We consider the partial derivatives of V with respect to K independent variables: ∂V ∂xi

= fi ( xi , xi1 , xi 2 ,..., xiK )

where the xij are the K variables interdependent with xi (for which the derivative is non-­zero). Given that the V function has one peak, the derivative will change sign in respect of changes of its terms. In a pseudo-­NK model, as in the NK case, K indicates which variable affects the validity of a change of another variable. That is, for a change of xi from xi1 to xi2 we can observe both a partial increment and a partial decrement of V for different values of xj, when xj is one of the K interdependent variables of xi. Under _​› these conditions, for K > 0, an exploration starting from a random vector x ​   ​   and considering only changes along one variable per time can reach a local optimum, unless the starting vector is part of the basin of attraction of the global optimum. For example, consider the case where N = 2 (the only one we can plot) with K = 1. Figure 4.1 shows the landscape generated by a V function of x1 and of x2,

Technological change and organisational dynamics   97 interdependent variables, on the basis of a Gaussian function centred on 100. The central ‘dome’ indicates the basin of attraction of the global maximum, while the two diagonal ‘ridges’ are local optima. Unless the search process starts in the neighbourhood of (100, 100), moving along one dimension only leads to a ridge, which is a local optimum. If, for example, the point of departure is (100.5, 98), respectively for x11 and x12, moving along the x1 dimension results in positive partial derivatives until the ridge is reached (in (99.5, 98) or (102, 98)). Any further change in x1 would not lead to any improvement in V. Let us now consider an example of such a ‘complex’ landscape explored by a ‘simple’ strategy, consisting in testing small changes of one of the variables x1 or x2, and accepting the change only if V increases above a minimal threshold.7 Given the landscape (compare with Figure 4.1), we have three local optima, besides the global maximum: two for the two ridges, and one for the floor in between the diagonals. Figure 4.2 shows different possible random patterns of V starting from a random point of the landscape, and converging to one of the maxima. Only some of the V manage to reach the global maximum. And it depends on the random starting point, as well as on the path followed by the x. Figure 4.3 shows the path followed by one typical pair of x for random changes of one of them, randomly selected. In this case the random sequence of change values allows V to reach the global maximum.

4  The model In this section we describe the basic configuration of the pseudo-­NK model, expanding on Section 3.1. We then show how it can be adapted to represent technological research on a number of different components that mutually determine the final quality of a good produced in a sketched economy. The model we present here is agent based on the production side, but has an aggregate demand (or a representative agent on the demand side).

Figure 4.1  Example landscape for N = 2 and K = 1, using normal functions.

98   T. Ciarli et al.

Figure 4.2 Example landscape for N = 2 and K = 1, using normal functions. Time series of the V values generated by 100 random strategies that modify one xi per time.

We build an artificial economy that represents one industrial sector, in which a good for the consumers’ market is produced. The sector is populated by f = 1, . . ., N competing firms that directly sell their good to the market. Firms’ competition is only indirect, and does not allow for direct interaction: the model thus represents an economy with high competition and no strategic behaviour. _ The consumption good y​ ​ f  produced is heterogeneous across firms, and defined over a set of I characteristics. For all firms, the good is assigned an homogeneous number of characteristics yi = 1, . . ., I, which represent the implicit use of

Figure 4.3 Example landscape for N = 2 and K=1, using Gaussian functions. Example _​› path of the x ​  ​   vector before reaching the global optimum at (100,100).

Technological change and organisational dynamics   99 the final consumers in a Lancasterian way (Lancaster 1966). In order to produce the final good, the firm uses j = 1, . . ., M components of quality mj, which we call modules. Each module is thought as an input of the consumption good, and is necessary for its production. Hence, the N firms may also be thought of as assembling firms, and the overall production system represents a production chain for which the technological specialisation has been initially (randomly) defined, as well as the input–output structure.8 The M are then intermediate component producers. They may be vertically integrated in an assembling firm, or they may compete to produce for a number of clients.9 The main contribution of our model is that it allows us to study the relation between (i) technological interdependence of input components; (ii) the innovation strategy; and (iii) industrial organisation (albeit at a first approximation in the present version). With respect to the first relation – technological interdependency and innovation, drawing from Valente (1999) – we relax some of the assumptions on technological complementarities between input components, and the relation between input and output characteristics. The model introduces the search for solutions of complex technological problems directly into firms’ routines. Following a qualitative simulation modelling approach (Valente 2005), in this chapter we investigate one initial aspect, namely the technological interdependency between modules. We first describe the object structure of the model, where each object represents an economic entity. Following on, we focus on the crucial aspect of the model, the interpretation of technological interdependencies in goods’ components. Finally, we complete the model description by plugging technological complementarities and innovation into the economy. 4.1  Representing technological interdependency: a pseudo-­NK model application The research algorithm adopted to allow for economic agents (firms) to explore the complex technological space of input combination has a crucial role in determining the emergent behaviour at the economy level. In this section we discuss the features of the search algorithm, and show its result. Such exploration will ease the understanding of the entire model, and its results.10 The fitness measure we refer to here is the quality level of a product characteristic yi; the quality of the service provided to consumers, in Lancasterian terms. This quality is the result of the combination of the modules mj used as inputs. Each module contributes linearly to the final value of each yi (time index t is suppressed unless required for clarity): yi =

M

∑α

i, j

mi , j

(4.1)

j =1

where mi,j is the value of the j module for the characteristic i, and αi,j = (0, 1) is its contribution to the value of the characteristic.11

100   T. Ciarli et al. While yi is a linear combination of the j modules contributions, the contribution of each module (its fitness, or quality level if we think of a good input) is highly non-­linear, and depends on the architecture of the whole good, the degree and strength of interdependence between modules. Taking into account the interdependency between modules, each mj has a modular, intrinsic, component (xj) and a non-­modular component determined by the position of the interdependent modules (μj). The modular component is the ‘position’ of the module on the multidimensional landscape – a solution in NK terms – where each module j is a dimension. The initial position of xj is exoge__ nous and is drawn from a uniform distribution centred on μ​ ​  . The non-­modular component is endogenous, being defined by its relation with the xk positions of the other mk ≠ j modules: µj =

∑a

k ≠ j∈ℵ

j ,k

ξk + c j

(4.2)

where ℵ is the set of the k = 1, . . ., M-­1 modules interdependent with mj; aj,k = [–1, 1] is the effect of a change in the kth module on the value of mj – via μj; and cj is constructed in order to allow the final value of mj to be independent of the mk __when all a =__0, and to reach its maximum value at xj = xk when all ​  – Sk ≠ j∈ℵaj,k μ​ ​  . a = 1:cj = μ​ The fitness of one single module (its contribution to the quality level of the characteristics) is maximised when its intrinsic position xj (the modular component) is matched with μj (the interdependent, non-­modular component). This is represented as a Gaussian function with mean 0: m j = exp( B (ξ j − µ j ) 2 )

(4.3)

where B is a constant that allows the change of the base of the function with values of mj = [0, 1] affecting the speed at which xj and μj can converge. As shown in Figure 4.4, varying B also affects the slope of mj. As a consequence, for a large B, the change of a module that is close to its maximum has a larger contribution to the quality of the characteristic yi than a change in a module for which the difference between xj and μj is very large. In other words, yi may experience an increase when only for one module |xj – μj| reduces and for all other modules |xk ≠ j – μk ≠ j| increases, if j is close to the average of the Gaussian, and all k are located in its tails. 4.1.1  Model dynamics As shown in Section 3.1, after defining the structure of the landscape we need to have a search strategy. The aim of the research across the landscape is to increase the value of yi. Given the equations above, this requires a change in the position of the components xj. Such a change has a direct effect on mj, but also an indirect effect on all other mk ≠ j (which is a basic feature of NK models). We

Technological change and organisational dynamics   101

Figure 4.4  Distribution of mjs with different B values.

implement an adaptive, modular research strategy: every t time periods each yi randomly draws one of the j modules – __ with equal probability – and changes its position in the landscape xj by d ~ U[​D​,  D]. With reference to an innovation process, this can be seen as the exploratory phase. We define the new position in the landscape as xjd. The new value mjd that results from the change in xj is compared with the initial mj, and xj changed to xjd only if mjd > mj. With reference to an innovation process this should be interpreted as the implementation phase. When the change is accepted, the movement of the single xj is made in accordance with the interdependencies with the other modules, but this does not imply that also all the other mk ≠ j have increased. The final result on yi depends on the number of modules interdependent (complexity of the landscape), on the strength of the relation, and on the initial position. While for some yis it may be comparatively easy to find a path to their highest value through time (still falling into pitfalls, but being able to exit from lock-­ins), other are bounded to converge to a lower local maximum. This is the case when a yi ‘accepts’ changing the position of one module, and this has a negative impact on at least another module, or when the initial positions of the modules, xj, are all very far from the basin of attraction of the global maximum. In fact, given the shape of mj, it is relatively likely that in order to improve the overall quality yi, the initial steps increase the value of the module that is closer to its maximum contribution level (the decrease in the contribution of the other module is lower, given that they are more distant from the top of the Gaussian function). And once the maximum for a module j is reached, it is unlikely that a change in another module k provides a contribution improvement that compensate for the loss in j. This is why this strategy is defined as modular. Any change involves only one module and is evaluated on the performance of that module. An opposite strategy would be to change at all times all the modules, and evaluate the impact on the overall fitness. As mentioned in Table 4.1 the latter strategy, named integral, is far more likely to reach the global maximum, but with a much higher information cost and very slowly (the probability that a change in one module improves its own contribution is much higher than the probability of a simultaneous change of all modules improving the overall fitness). The reason why the integral

102   T. Ciarli et al. strategy is able to achieve the global maximum is that it does not risk climbing onto local maxima that lock-­in the process from future changes. In what follows we briefly analyse the pseudo-­NK model of technological interdependencies. We show three examples depicting different fitness outcomes, generated randomly as the result of the search for fitness improvements, which here represent an increase in the quality level of the good’s characteristic yi. 4.1.2  Paths on a complex technological ‘landscape’ T he model setup

We show results of the dynamic of 100 different yi (I = 100) along 10,000 time steps. Each yi has six modules mi, j (M = 6), and each module is related to the remaining five modules (ℵ = {1, . . ., 6}). The high interdependence describes a system in which modularity is low. The initial position of the modules xI, j is drawn randomly (from a uniform distribution of values between 90 and 110) and changes through time as described in the previous section. The remaining parameters values are summarised in Table 4.2. R esults of the technological search

Figure 4.5 shows the change in the quality level (fitness) for the different yi: the main result is that, given the initial small differences in the position of the Table 4.2  Parameters for Section 4.1.2 Parameter1

Description

I

Number of characteristics yi s

100

J

Number of modules per characteristic mi, js

6

ℵ

Set of modules mi, k ≠ j interdependent with mi, js

{1, . . ., 6}

B

Defines the width of the Normal function for the computation of mi, js

–7

Minimum value of the distribution from which a new xi, j is drawn

xi,j,t – 0.05

D​ ​  

Maximum value of the distribution from which a new xi, j is drawn

xi,j,t + 0.05

xi, j(1)

Initial value of xi, j

RND (90,110)

µ​ ​ 

Average xi, j(1) (centre of the distribution from which it is drawn)

100

ai, j

Contribution of mi, js to yi

RND (0,1)

ai, j, k ≠ j

Relation between two interdependent modules j and k in affecting mi, j

RND (–1,1)

__

D​ ​   __

__

Note 1  Into parenthesis the number of lags of the initial value for the lagged variables.

Value

Technological change and organisational dynamics   103

Figure 4.5  Evolution of yi s.

modules on the multidimensional landscape, the different series follow very dissimilar paths. Each path is the result of the modular search for an improvement in the value of the modules’ contribution to the quality level of the characteristic. It is useful to highlight four main features of the search process that stem from Figure 4.5: (i) all series experience non-­monotonic dynamics, which depend on both technological search (moving the position of one single module at a time) and interdependencies between characteristics. The reason why the series are not monotonic is that the modular solution allows accepting a change when it is positive for the module, and not necessarily for the overall fitness of the quality characteristic; (ii) only some series reach their maximum value, which depends on the opportunity to exit from local maxima, rather than on the speed of technological change (the change in x is equal across series, on average);12 (iii) many series reach a local maximum and stick to it; and (iv) a number of series seem to follow a random walk, raising and decreasing their value. We consider the three cases (ii-­iv) separately. In Figure 4.6 we show the values of yi, xj and μj for a characteristic that reaches its global maximum (i = 85). First, the initial distribution of the xj is quite concentrated. Second, some of the xj quickly converge to the value that maximises their relative contribution with respect to the other xj, μj (e.g. x2, x3 and x4). Most importantly, these modules manage to stick to their relative maximum value while the other modules smoothly converge, because their position on the landscape is quite close to the overall maximum (100).13 Third, y85 increases each time one of the xjs manage to converge to its μj, as it should be by construction. Indeed the series does not follow a monotonic process. For example, when x1 manages to converge to μ1, a relatively better long-­term strategy is to leave the local maximum, reduce m1, while increasing the remaining mj (x5 and x6), although this causes a temporary loss of the overall value of y85.

104   T. Ciarli et al.

Figure 4.6  Maximisation of yi and convergence of ξj.

An opposite dynamic, more similar to a random walk, unfolds when representing the time series of the characteristic y2 in Figure 4.7. Only x2 and x6 manage to converge to their relative maximum value, after a large number of periods. On a few occasions some of the xj approach their maximum, but the negative effect that these changes have on the contribution of the other interdependent modules drop them back. The system, thus, never manages to stabilise on a long-­term value of y. This is due to the fact that the initial random position of the modules is quite spread, and the changes in the module that managed to converge since the beginning are detrimental to the other model, as they push the μj far from the global maximum (100). In other words yi is lost, and there is no modular strategy that can serve the purpose of innovating in the right direction. In order to converge to its maximum, the system should deploy an integral strategy (moving all the modules in a co-­ordinated way) allowing for large jumps in the xj rather than small changes.

Figure 4.7  Random walk of yi and non-convergence of ξj.

Technological change and organisational dynamics   105 A third example shows an even different behaviour (Figure 4.8): y95 converges to a low value – it locks in a local maximum – where it remains for at least 8,000 periods. Only two xjs (1 and 6) reach their relative maximum and manage to stick to it throughout the periods. Conversely, the local strategy adopted by yis to maximise their value (randomly changing only one module every period), keeps outstripping the xj from the positions that would take into account the correlation with the other modules. In fact, the contribution of m1 and m6 to the value of y, is much larger than the contribution that a small change in the xj of any of the remaining mjs would bring. We now turn to the model of the economy, in which firms become the actors of innovation, producing one consumable good for the final market, by means of combining the different modules. The relation between the quality of the good and the interdependent modules is modelled with the pseudo-­NK algorithm. 4.2  The economy Having described the structure and behaviour of the technological components of the model, we now proceed with the description of the general economy structure and the dynamics within which the previous model is embedded. We will therefore describe the way in which the evolution of characteristics affects firms’ behaviour, and how firms manage to deal with the complexity of technological research. Appendix 4.1 describes the structure of the economy, its different levels and the way in which technological search is included as one of the firms’ features. 4.2.1  The market The model represents an aggregate demand of homogeneous consumers that have preferences on both the price and the quality of the characteristics of the  goods produced in the market. We assume that the demand increases

Figure 4.8  Lock-in of yi and non-convergence of ξj.

106   T. Ciarli et al. monotonically with respect to both types of preferences (the price may be easily seen as a further characteristic of the good). As in Section 4.1, each characteristic’s yf,i, ∀f, quality value depends on the contribution of the product components, mi,j. The aggregate demand is computed on the average values of the goods available in the market: D* = H

1 p

εp

I

∏y

εc i



(4.4)

i =1

_

_

where H is a constant, p​ ​   and ​y​ i are respectively the average price and product’s quality characteristics, ep and ec the aggregate preferences for respectively price and quality characteristics. We also acknowledge that it needs time for consumers to adjust their demand, and for firms to interpret the signal. The actual aggregate demand then smoothly adjusts to its target, in an adaptive way: D = σ dDt −1 + (1 − σ d )Dt*

(4.5)

where σ d determines the speed of adjustment. 4.2.2  Firms’ competitiveness The demand preferences also determine the performance of the single firms. We use a competitiveness index cf that has the same form of the target demand D*, although the firm is clearly evaluated on its own features, with respect to both price and quality of the characteristics χf =

1 ε

pfp

I

∏y

εc f ,i

i =1

The cf are then used to determine firms’ market share, mimicking a selection process. As for the demand (and for the same reasons), we first compute a target market share χf msf* = χf

∑

f

to which the actual one adapts through time: msf, t = σ mmsf, t – 1 + (1 – σ m)ms*f, t , where σ m determines the speed of adjustment. 4.2.3  Production and price Firms infinitely respond to their individual demand, using a Leontief technology where the productivity A indicates only labour productivity, and affects only the output price. The individual demand (and production) is a fraction of total demand, determined by the firm market share: δf = msf D.

Technological change and organisational dynamics   107 The price is then determined as a mark-­up on variable costs: wages and input price. Given a wage wf, which we assume rigid, the unitary labour cost for the single firm is given by wf./Af Conversely, the cost of inputs is determined by components suppliers price pmf, j, which we assume to be the same for external supplier and internal modules, and exogenous. The output price is then computed as  wf p f = (1 + n f )  +  A f

M

∑p j =1

m f,j

  

(4.6)

where nf is the mark-­up applied by the firm. Finally, firms supply the market with the good defined over its characteristics yi. The way in which the quality of characteristics is determined is thoroughly described in Section 4.1. In the following section we briefly describe the dynamics of the model, and how firms determine product innovation. 4.3  The dynamics In each time period t, first the market values are computed (mainly the aggregate demand), and then firms’ accordingly produce and undergo technological exploration. Nevertheless, in order to define the demand level, a number of pieces of information at the firm level is needed. Therefore, at the beginning of the period the quality values of the characteristics of the good offered are computed, together with their price (see Figure 4.9). These are used to determine both the competitiveness of each firm – hence their market share – and, following, the weighted average of price and of each characteristic in the market. Note that we are describing an industry in which all firms are able to manage all the modules that enter into their good as inputs, vertically integrated firms. The second part of the dynamics is devoted to technological exploration. The basic operation of technological research on the modules, and according product innovation, was described in Section 4.1. In the context of the economy we divide the innovation process into three main operations: (i) exploration; (ii) evaluation; and (iii) adoption decision. The exploration of the technological space simulates the actual technological experimentation to find a better definition of one product component (one can imagine an R&D lab). For each assembling firm, one module mj is randomly selected, and the technological exploration is undergone, moving it to a new position on the technological landscape – drawing a new value for xj. Following, the innovation outcome is evaluated. We can conceive three alternative routines:14 (a) evaluation of the overall performance of the good: the new component is plugged into the good, and its performance is evaluated,15 with respect to the ‘old’ product. When this routines applies, we represent an assembling firm that is able to tune one module each time, and has complete information on the technological landscape, or has a large influence on its suppliers that autonomously innovate; or (b) evaluation of the performance of one

Time = t t +1

t +1

Each firm provides information on the value of the good’s characteristics and price

Determine the competitiveness of each firm, and consequently market share

Determine the average value of characteristics and price

Compute the demand level in the market

Derive firms’ demand level and produce

TECHNOLOGICAL CHANGE

EXPLORATION For each firm, one module is randomly selected and its component is moved to a new position

NO

INTEGRATED FIRM

YES

FOCUSED INTEGRATED FIRM

NO

MODULAR FIRM

YES

Evaluate the performance of overall characteristics (the good on the market)

YES

Evaluate the performance of one characteristic randomly selected

ACCEPT NEW MODULE

Confirm the new value (position) of the module

Figure 4.9  Model flowchart: one time period.

YES

Evaluate the performance only of the module explored

NO

Return to the initial value, you will be luckier next time

Technological change and organisational dynamics   109 characteristic of the new good, randomly selected; even in this case we represent an integrated firm (a firm that may govern several modules per innovation), but which addresses a particular demand (for which we assume the selected characteristic is important), or which is believed to have a particular comparative advantage; or (c) evaluation of the performance of the single component under exploration: in this case we represent an assembling firm with limited information on the relation between the optimal values of modules, or which has a very limited influence on suppliers that innovate autonomously. Eventually, for each of the three evaluation routines, the innovation is either accepted or rejected. The change in the module is accepted if it produces a higher fitness (either for the good, one characteristic, or the component itself ). Whatever the decision, the final value of the component is used in the following time period t 1 when the process starts again from the beginning.

5  Results We now show and comment the simulation results. First, the general model setup is provided, while the list of parameters and initial variables values is available in Table 4.3. 5.1  Model setup We define an economy populated by 50 firms randomly distributed according to the evaluation strategies. Each firm produces a final good that is evaluated on two characteristics. The good is produced assembling three components (modules), and we assume full interdependence. We initialise the model with homogeneous values across firms, except for the technological values, the xi,js position, which are similar to the ones used in Section 4.1.1, and productivity which slightly differ. Goods have the same architecture for all firms, meaning that: (i) the modules contribute to product characteristics in the same way for all firms, although differently across modules; (ii) we replicate the same structure of interdependence (the ai,j) between modules across all firms, although they differ across modules. We thus undergo this preliminary exercise on a simple system (there are quite few entities – characteristics, modules, firms), characterised by the highest complexity (full interdependence between modules). 5.2  Simulation outcomes Starting from homogeneous firms in a very simple market, we show that the complexity of the technological research, paired with no modularity, produces heterogeneity in the results. In Figure 4.10 we show the evolution over time of the market shares of the firms that attempt to improve the global fitness of the consumption good (with all its characteristics). During the initial periods, the advantage of the first

110   T. Ciarli et al.

Figure 4.10  Market shares: firms addressing the global fitness.

movers (the luckier in this case) is quite apparent. Those firm that are able to increase the value of at least one characteristic at the beginning, suddenly gain market share. Nonetheless, the system settles down into a wide number of stable paths. And only some of the firms that gain share at the beginning, end up on the higher paths. Some of them are not able to modify their good any more. On the contrary, a number of latecomers find their way through the technological landscape and manage to reach the highest quality values (all the rest being equal). This dynamic is also shown in Figure 4.11, where the evolution of the yf,i for the same firms, is reported.

Figure 4.11  Value of the product characteristics: firms addressing the global fitness.

Technological change and organisational dynamics   111 We do not reproduce the results in this chapter, for reason of space and because they are far from unexpected. However, the highest market share, in the long run, is gained by those firms which are able to understand the relation between modules’ values over the total fitness. For those firms there is obviously only one stable path. More interesting is the reason why firms with limited information on the technological landscape may obtain quite different results. Table 4.3  Parameters for Section 5.2 Parameter1

Description

Value

f

Number of firms xf

50

i

Number of characteristics yis

2

j

Number of modules per characteristic mi, js

3

Set of modules mi, k ≠ j interdependent with mi, js

{1, . . ., 3}

ℵ p

Price consumer preferences

1

∈ic

Characteristics consumer preferences

1

σ

Demand smoothing parameter

0.9

σ

Market share smoothing parameter

0.9

H

Constant demand

100

D(1)

Initial demand

150

ms(1)f

Initial market share

0.02

wf

Unitary labour cost per firm

5

Af

Productivity of each firm

RND[1,5]

vf

Mark-up on unitary input costs

0.1

Bi

Defines the width of the Normal function for the computation of mi, js

–7

pmf , j

Unitary price of the inputs

5

D​ ​  

Minimum value of the distribution from which a new xi,j is drawn

xi,j – 0.05

Maximum value of the distribution from which a new xi,j is drawn

xi,j + 0.05

∈

 d  m

__ __

D​ ​   xi, j(1)

Initial value of xi,j

RND(98,102)

µ​ ​ 

Average xi,j (1) (centre of the distribution from which it is drawn)

100

ai, j

Contribution of mi, js to yi

RND[0.1, 0.9] ∀f

ai, j, k ≠ j

Relation between two interdependent modules j and k in affecting mi , j

RND(–1, 1) ∀f

__

Note 1  Into parenthesis the number of lags of the initial value for the lagged variables.

112   T. Ciarli et al. In Figure 4.12 we show the dynamic of technological search for a firm that reaches a low market share, and compare it with a firm that in the long run succeeds to exit all technological lock-­ins and innovate the product successfully (Figure 4.13). The first firm (Figure 4.12) is able to successfully increase the value of one characteristic (as x3 approaches μ3), in quite a short time. Yet, when x3 reaches its maximum, the evaluation of any other innovation is rejected. In fact, any small change in the other xs does not show an increase in the global fitness sufficient to counterbalance the loss of a reduction in x3. Actually, this would have to retreat from its optimal position in order to allow the technological advance in the remaining modules. Conversely, the second firm (Figure 4.13) manages to reach one local maximum (with x1) after a larger number of periods. And even when this is reached, the proximity of x1 with the other xs allows the firm to innovate also with other modules, without sticking on the local maximum: in fact, x1 lowers its value again for a long period, while x2 and x1 approach their maximum, and then x1 gets back to its optimal position.

6  Concluding remarks This chapter has furnished a set of simulation results for a model depicting firms’ search behaviour in front of technological opportunities. Several elements emerged, linked to both the single firm’s capacity to cope with a turbulent environment, and the resulting market structure. The model is built up starting from a pseudo-­NK model that allows us to describe the technological search on a number of different components that jointly determine the set of characteristics of a final good. This algorithm is then plugged into an artificial economy within which firms are able to explore

Figure 4.12  Search on the landscape and product innovation: unsuccessful.

Technological change and organisational dynamics   113

Figure 4.13  Search on the landscape and product innovation: successful.

and/or exploit their technological landscape in order to gain better performances. We are thus able to analyse, by means of computer simulations, the behaviour of firms in dealing with the trade-­off between focusing the innovation on a few components of their product, or conducting a slower, systemic, search. The results confirm that the model is well suited to describe such a kind of dynamic. Our preliminary results highlight some interesting elements, such as the first mover advantage which, however, is not guaranteed, as slow movers are also able to find their way to top performances. Knowledge is important, as results for firms are dependent on their capacity to manage the knowledge about the modules’ relationships. Finally, the resulting industrial organisation is highly dependent on the way in which firms manage to solve the technological trade-­ off. Such a model can be used to analyse firms’ decisions on internalising all the production processes, and their innovation, or rely on the market to gain economies of specialisation. An important element that will be the focus of further research is related to the capacity of firms to act ‘strategically’. Indeed, in front of a change of the relative cost of a module, firms can have two different behaviours, being either proactive or adaptive. In the former case, a proactive firm tries to learn how to cope with technological changes by suitably modifying its technology and its organisational structure. In the latter case, an adaptive firm reacts to a shock by simply modifying its structure in terms of integrating/ disintegrating the module whose cost is changed. Proactive firms are thus characterised by ‘dynamic capabilities’ that are put in place once there is the need to rethink how the technological base of the production process is organised. This happens through a process of exploration of the

114   T. Ciarli et al. technological landscape aimed at finding the most appropriate technology to produce a certain characteristic of the final good, given the new cost structure of the production system. The exploration process ends once a more efficient technology is found for that module. Given the structure of correlation existing between the whole vector of modules, the search process will involve the whole production technology. Adaptive firms instead are not forward looking, and thus they happen to intervene just after the technological change, on the overall module composition of their production structure. In this case, adaptive firms are (myopically) considering that the change in the relative price composition of their array of technological modules has consequences on the set of modules they will be working with. If a change in the cost structure makes a module no longer viable, then that module will be simply outsourced, thus benefiting from the gain due to the reduction of its relative price (net of transaction costs).

Appendix 4.1: Object structure of the model The economic entities included in the model are represented as objects types, hierarchically ordered on different layers. In particular, each object type – up to the last one – may be defined by its micro components, represented by downstream objects. And vice-­versa, each micro object is considered as one characteristic of its parent object. Different object types may be defined at the same layer, and represent a similar level of disaggregation. Finally, each object type usually contains a number of instances that make up the economy. The above should become more clear in the following explanation of the actual model structure, by way of its object configuration (Figure 4.14). The aggregate values of the economy represented in the model are stored at the higher level (Object World). At this level we also define the sectoral (aggregate) demand, aggregate statistics, and the ‘governance’ of the model dynamics in each time period t. The aggregate demand is also defined as a function of the representative consumer’s preferences for the different consumption good characteristics. Both aggregate preferences and the market mean values of the characteristics of the produced good are stored in Object AvChari, which is indexed by i = 1, . . ., I instances. The other objects (left-­hand side in Figure 4.14) define the firm, its production and technology. In Object Firmf (indexed by f = 1, . . ., N) we specify firms’ features in relation to both their behavioural rules (technology and price strategy, wage, productivity) and their indirect outcomes (technological competitiveness, market share), while the characteristics that define the production of a single firm are computed in Object Chari,f. Object Modj,i,f plays a crucial role in mimicking the NK dynamics. It provides the technological value of modules, their relative contribution to product characteristics, and the evaluation of technological change with respect to the interdependencies with the other M-­1 modules. Finally, Object CModk ≠ j, i, f exogenously defines the technological interdependencies between modules.

Technological change and organisational dynamics   115

World Defines sectorial features and its demand

Firm

AvChar

Defines all firm level features, not included in the objects below. Both direct (price, wage, A, R&D strategy) and indirect (technology, market share) consequence of firm’s behaviour

Defines sectorial values of product characteristics and consumers’ preferences

Char

Tree structure

Computes the level of the different product characteristics World 1

...

Mod

1

N

...

Firm

Defines the level of each input module, its relation with the other inputs, the contribution to characteristics and its change (innovation)

1

...

1

AvChar 1

...

1

...

1

Char 1

...

M

...

...

1

...

...

M

Mod

CMod

1

Defines the complementarity between modules

...

M-1

...

...

...

...

1

...

M-1

CMod

Figure 4.14  Object structure of the model.

Notes   1 For example, by combining this leaked knowledge with complementary knowledge already available to the outsourcing firm.   2 On the other hand, tapping into the superior experience of the outsourcing firm does not always entail dynamic efficiency gains (Mahnke 2001). This requires that the outsourced activities are learned and re-­integrated at affordable costs, with significant economies of scale and scope in combining them with the old ones, with no substantial integration investments and no worsening in industrial relations (Hamel 1991; Lyles and Stalk 1996). The literature on post-­merger integration shows that such conditions are hardly satisfied (Haspeslagh and Jemison 1991; Jemison and Sitkin 1986) and that outsourcing might make it more dependent on suppliers in accessing external knowledge (Benson and Ieronimo 1996), if not even inter-­locked with competitors, via learning-­by-interacting (e.g. Dyer and Nobeoka 2000), and with external workers,

116   T. Ciarli et al. via ‘market-­mediated’ work arrangements (Matusik and Hill 1988). Last but not least, outsourcing might compromise the firm’s innovativeness by affecting its ‘absorptive capacity’ (Cohen and Levinthal 1989). Indeed, by focusing on the activities retained in-­house, the firm might then suffer from higher ‘search costs’ in looking for new external knowledge sources and higher ‘cognitive costs’ – both direct and indirect – in articulating and codifying them internally (Foray 2004).   3 However, by assuming that an innovation is radical if it is ‘systemic’ (i.e. it involves interrelated changes in many components (Teece 1986)), vertical integration would not be favoured as it hampers co-­ordination among a set of interdependent elements while outsourcing would be preferred (Mahnke 2001). Contrasting results can be obtained also following a ‘strategic’ perspective focusing on the structure of the ‘technological dialogue’ (Monteverde 1995) between supplier and customer of a certain innovation, in specifying the attributes of the new products and services to be introduced (e.g Christensen et al. 2002).   4 As is well known, the concept of dynamic capabilities has been developed by strategic management to explain, more satisfactorily than the standard neoclassical theory, why firms show different dynamic performances in front of a turbulent environment. Reducing the issue to a pure question of strategic incentives to invest in innovation is in fact not entirely adequate. Indeed, predictions based on these premises are not empirically robust (Henderson 1993).   5 Note that formally we can speak of complexity only considering both the problem’s and the solver’s characteristics.   6 Note, however, that K is actually not the best indicator of complexity.   7 This requirement allows the consideration of the tails of the Gaussian function as flat.   8 In particular, in this first version of the model we assume that all intermediate goods are inputs to the consumption good, and none is also input to another intermediate good. Such an assumption is far from unrealistic for a wide number of sectors, especially when considering the firm level.   9 In this respect, M producers should be seen as independent units of production, or shops. Nonetheless, the version of the model we describe in this chapter does not allow for autonomous behaviour of those intermediate producers. 10 In Appendix 4.1 we provide the object structure of the economy, representing the different levels from micro to macro. 11 From now on we drop the characteristic index i, unless necessary to distinguish between characteristics. 12 Note that the maximum is not identical for all yi, as it also depends on the value of the contribution of each module, α. 13 A quick reference to Figure 4.1 should help in visualising why this is the case in a two-­dimensional landscape. 14 The basic research model in Section 4.1 conceives the process ruled by each characteristic, thus replicating only one evaluation procedure. 15 By evaluating the sum over all the characteristics.

References Acemoglu, D., Aghion, P., Griffith, R. and Zilibotti, F. (2005), Vertical integration and technology: theory and evidence, CEPR Discussion Paper 5258, Centre for Economic Policy Research, London, online, available at www.cepr.org/pubs/dps/ DP5258.asp. Benson, J. and Ieronimo, N. (1996), Outsourcing decisions: evidence from Australia-­ based enterprises, International Labour Review, 135: 59–73. Christensen, C.M., Verlinden, M. and Westerman, G. (2002), Disruption, disintegration and the dissipation of differentiability, Industrial and Corporate Change, 11(5): 955–93.

Technological change and organisational dynamics   117 Ciarli, T., Leoncini, R., Montresor, S. and Valente, M. (2007), Innovation and competition in complex environments, Innovation: Management, Policy & Practice, 9(3–4): 292–310. ——, ——, —— and —— (2008), Organisation of industry and innovation dynamics, Journal of Evolutionary Economics, 18: 367–87. Cohen, W. and Levinthal, D. (1989), Innovation and learning: the two faces of R&D, Economic Journal, 99: 569–96. Dosi, G. (1988), Sources, procedures, and microeconomic effects of innovation, Journal of Economic Literature, 26(September): 1120–71. Dyer, G. and Nobeoka, K. (2000), Creating knowledge and managing a high performance knowledge-­sharing network: the Toyota case, Strategic Management Journal, 21: 345–467. Foray, D. (2004): The Economics of Knowledge. MIT Press, Cambridge, MA. Frenken, K., Marengo, L. and Valente, M. (1999), Interdependencies, nearly-­decomposability and adaptation, in T. Brenner (ed.), Computational Techniques for Modelling Learning in Economics, Kluwer, Boston/Dordrecht/London. Gonzalez-­Diaz, M., Arruñada, B. and Fernandez, A. (2000), Causes of subcontracting: evidence from panel data on construction firms, Journal of Economic Behavior & Organization, 42: 167–87. Grossman, S.J. and Hart, O. (1986), The costs and benefits of ownership: a theory of vertical and lateral integration, Journal of Political Economy, 94: 691–719. Hamel, G. (1991): ‘Competition for determinant and interpartner learning within international strategic alliances’, Strategic Management Journal, 12: 83–103. Haspeslagh, P.C. and Jemison, D.B. (1991), Managing Acquisitions: Creating Value Through Corporate Renewal, Free Press, New York. Henderson, R. (1993), Underinvestment and incompetence as responses to radical innovation: evidence from the photolithographic alignment equipment industry, RAND Journal of Economics, 24: 248–70. Henderson, R.M. and Clark, K.B. (1990), Architectural innovation: the reconfiguration of existing product technologies and the failure of established firms, Administrative Science Quarterly, 35: 9–30. Iansiti, M. and Clark, K.B. (1994), Integration and dynamic capability: evidence from product development in automobiles and mainframe computers’, Industrial and Corporate Change, 3: 557–605. Jemison, D. and Sitkin, S. (1986), Corporate acquisitions: a process perspective, Academy of Management Review, 11: 145–63. Lancaster, K.J. (1966), A new approach to consumer theory, Journal of Political Economy, 74(2): 132–57. Leonard-­Barton, D. (1992), Core capabilities and core rigidities: a paradox in managing new product development, Strategic Management Journal, 12: 111–25. Levinthal, D. and March, J. (1993), The myopia of learning, Strategic Management Journal, 14: 95–112. Levitt, B. and March, J. (1988), Organizational learning, Annual Review of Sociology, 14: 319–40. Lundvall, B. (1992), National Systems of Innovation. Towards a Theory of Innovation and Interactive Learning. Pinter, London. Lyles, M. and Stalk, J. (1996), Knowledge acquisition from foreign parents in international joint ventures, Journal of International Business Studies, 27: 877–904.

118   T. Ciarli et al. Mahnke, V. (2001), The process of vertical dis-­integration: an evolutionary perspective on outsourcing, Journal of Management and Governance, 5(3–4): 353–79. March, J. (1991a), Exploration and exploitation in organizational learning, Organization Science, 2: 71–87. —— (1991b), Continuity and change in theories of organizational action, Administative Science Quarterly, 41: 278–87. Matusik, S.F. and Hill, C.W.L. (1988), The utilization of contingent work, knowledge creation, and competitive advantage, The Academy of Management Review, 23: 680–97. Monteverde, K. (1995), Technical dialog as an incentive for vertical integration in the semiconductor industry, Management Science, 41: 1624–8. Nelson, R.R. and Winter, S.G. (1982), An Evolutionary Theory of Economic Change, The Belknap Press of Harvard University Press, Cambridge, MA. Nonaka, I. (1994), A dynamic theory of organizational knowledge creation, Organization Science, 5: 14–37. Robertson, P. and Langlois, R. (1995), Innovation, networks, and vertical integration, Research Policy, 24: 543–62. Sabel, C.F. (1989), Flexible specialisation and the re-­emergence of regional economies, in P. Hirst and Zeitlin, J. (eds.) Reversing Industrial Decline?, Berg, Oxford/St Martins, New York. Schumpeter, J.A. (1934), The Theory of Economic Development, Harvard University Press, Cambridge, MA. Silver, M. (1984), Enterprise and the Scope of the Firm, Martin Robertson, London. Smith, A. (1776), An Inquiry into the Nature and Causes of the Wealth of Nation, Campbell, R.H., Skinner, A.S. and Todd, W.B. (eds.) [1976], Oxford University Press, Oxford. Teece, D.J. (1986), Profiting from technological innovation: implications for integration, collaboration, licensing, and public policy, Research Policy, 15: 285–305. Teece, D.J. and Pisano, G. (1994), Dynamic capabilities: an introduction, Industrial and Corporate Change, 3(3): 537–56. Teece, D.J., Pisano, G. and Shuen, A. (1997), Dynamic capabilities and strategic management, Strategic Management Journal, 18: 509–33. Tushman, M. and Anderson, P. (1986), Technological discontinuities and organizational environments’, Administrative Science Quarterly, 31: 439–65. Valente, M. (1999), Evolutionary Economics and Computer Simulations. A Model for the Evolution of Markets. Vol I: Consumer Behaviour and Technological Complexity in the Evolution of Markets, Ph.D. thesis, University of Aalborg, Aalborg. —— (2005): Qualitative simulation modelling’, in 4th European Meeting on Applied Evolutionary Economics (EMAEE), Universiteit Utrecht. Williamson, O.E. (1975), Markets and Hierarchies: Analysis and Antitrust Implications, Free Press, New York. —— (1985), The Economic Institutions of Capitalism: Firms, Markets, Relational Contracting, Free Press, New York. Zollo, M. and Winter, S. (1999), From organizational routines to dynamic capabilities, INSEAD R&D Working Papers: 99/48/SM. —— and —— (2002), Deliberate learning and the evolution of dynamic capabilities, Organization Science, 13(3): 339–51. Zott, C. (2002), When adaptation fails: an agent-­based explanation of inefficient bargaining under private information, Journal of Conflict Resolution, 46: 727–53.

5 Technological persistence through R&D on an old technology The ‘sailing ship effect’ Nicola De Liso and Giovanni Filatrella

1  Introduction1 Technological change is undoubtedly one of the driving forces of our economies, and two striking features of our production systems consist of technological variety on the one hand, and technological persistence on the other. The main aim of the present chapter is to analyse the process of competition between two technologies, one which we will call ‘old’ and the other which will be defined as ‘new’, while reviewing some of the determinants of both variety and mechanisms which favour persistence. The interest in such a topic arises from the observation that there exist technologies which experience a life which is much longer than expected, as they are – at least technologically – overwhelmingly outperformed by new and emerging ones. Economic theory has for a long time already hinted at some mechanisms which contribute to prolonging the life of ‘old’ technologies, and in particular learning-­by-doing (Smith 1776; Arrow 1962) and learning-­by-using (Rosenberg 1982), to which we ought to add cost analysis as at first developed by Frankel (1955) and, more thoroughly, by Salter (1969). What we are mainly interested in here is the sort of mechanism which implies intentional action aimed at improving the performance of an ‘old’ technology; this action is stimulated by the emergence of a new technology which is devoted to supplying the same sort of operations or services. This mechanism is often referred to as the ‘sailing ship effect’, after Gilfillan’s study on innovation in ships (Gilfillan 1935). The chapter is divided in the following way: Section 2 hints at technological variety and theoretical justifications of the coexistence of old and new technologies; Section 3 explicitly recalls examples of technological competition between old and new technologies; Section 4 contains the model proposed and the simulations which describe the process of competition between two technologies via improved performance; Section 5 contains the conclusions and makes explicit the pros and cons of the model.

120   N. De Liso and G. Filatrella

2  Technological variety and the coexistence of old and new technologies Ever since the first Industrial Revolution occurred in the late eighteenth century, economic and technological forces have shaped our socio-­economic systems. Economic incentives have strong effects on the direction undertaken by technological creation and development, but the way in which technology can change is not free from constraints. On the other hand, technology opens economic opportunities which would otherwise be unthinkable. The whole matter is made more complex by the existence of rules, standards and limits of various kinds, which in turn, contribute to bias the evolution of technology itself. If, on the one hand, technology can only be developed along specific paths – the notions of technological systems, paradigms and trajectories are analytical tools pointing at these limits – on the other hand we observe that often the same results can be obtained by means of completely different technologies. Let us consider an example. We can produce electricity in many different ways, by burning different kinds of fuel, exploiting nuclear reactions, falling water or by means of photovoltaic cells. Moreover, often within each technology different paths are open, so, for instance, photovoltaic cells can make use of different semiconductors; once a material such as silicon has been chosen, it can be manufactured in three different ways, namely, monocrystalline, polycrystalline and amorphous, the result being always the production of electricity. Technological competition can also be looked at from a different perspective, making reference to broader principles, such as the case of digital and analogue ways of getting a particular result. In fact, one of the most important forms of competition of the recent past is the one between analogue and digital systems. Digital and analogue systems can often provide similar services, but with a different efficiency and quality of the result. A typical path, however, is that of digital technologies supplanting – sometimes gradually, sometimes suddenly – the existing analogue ones.2 The main point we are concerned with is that there exist economic and technological reasons which justify the coexistence of old and new technologies. Two basic types of forces can be identified: (i) forces which affect factors’ productivity and (ii) forces which depend on sunk costs. The two examples often referred to with respect to the first type of force are learning-­by-doing and learning-­by-using. Learning-­by-doing occurs at different levels of production, and concerns both individuals and work processes or organizations as a whole. Typical examples have become the ones referred to by Arrow (1962), i.e. the Horndal iron works in Sweden – where productivity rose by 2 per cent per annum for 15 years, despite the fact that no investment had occurred – and the production of airframes – to produce the nth airframe of a given type the amount of labour required is proportional to n–1/3. In these cases factors’ productivity increases with cumulative output, and the existing techniques/technologies can survive for a long time, even when a new one emerges. We can expect learning-­ by-doing to occur at any level of technological complexity.

The sailing ship effect   121 Learning-­by-using has been referred to by Rosenberg (1982) as that process of learning, and form of disembodied technological change, which concerns the use of machinery, that is the process whereby we learn how to use identical machines in more efficient ways. Typical examples are the ones related to the use of machine tools, whose productivity – given identical machines – increases with their use. The two categories of learning just mentioned can contribute to defining the innovative content of ‘simple’ technology reproduction. In fact, what we learn through production can affect the way in which the technology which is being used can change. Technology reproduction can be looked at from at least two perspectives: symbolic and material. By symbolic reproduction we mean the codified form of transmission of technological knowledge such as the one which we find in technical textbooks, in manuals of use of machines or also in internal reports. By material reproduction we mean the actual process of reproduction of means of production as well as products (including non-­material ones and services, such as, for instance, programming or surgery): along this path we are likely to discover new ways of doing things. The material reproduction of technology contains a high degree of tacit knowledge, part of which can be codified at some future time. Both processes lead to improvements of the existing technologies and products, and thus contribute to prolong their life. Another reason which contributes to keeping an old technology alive lies in the pervasiveness of the technology itself. The more a technology is pervasive and diffused, the more difficult it will be to switch to a new one. It is not only a problem of the direct costs associated with substitution, but of all the connected necessary changes, organizational, institutional and so on. Any new technology, in order to be adopted, must offer a series of incentives in terms of both economic and non-­economic conditions. But let us turn our attention to two explicit analyses of the effects of sunk costs and different cost structures on keeping alive an old technology. First of all, Frankel clarified a long time ago why an old technology can survive together with a new, more efficient, one. In particular his analysis focuses on process innovation, and he considers two cost categories: (i) past or sunk outlays, which ‘denote expenditures already made which are allocable to some future production period and which must be recovered from future revenues’, and (ii) future outlays, which ‘denote all costs that will be incurred over some future production period’. For an old method already in use, past outlays always are some positive amount, while for a new method not yet adopted, past outlays are zero, and total costs are simply given by future outlays. Given these initial definitions, Frankel clarifies when and how an old technology should be continued in use or be immediately replaced3 (Frankel 1955). A few years later, Salter (1969) published his book which clarifies theoretically the adjustment process which takes place within an industry in which there occurs technological change. In its simplest version, the model considers the coexistence of different techniques characterized by different cost structures.

122   N. De Liso and G. Filatrella First of all, this makes it clear that it is inappropriate to discuss productivity movements of an industry in terms of a representative firm, as the dispersion around the average is too important to be excluded from the analysis. Then, by defining the price of the commodity produced as the sum of operating costs (labour and materials rewarded according to their prices) and capital costs (which include normal profits), one obtains the price which defines the oldest plants that can remain in operation. The marginal firm will be that whose operating costs equal the price – at this technological stage. As soon as a new best-­ practice technique becomes available, this marginal firm will have to leave the market as the price would go down, thus rendering its operating costs higher than the new price (op. cit. pp. 58–9). In this model there is room for the coexistence between different techniques characterized by a broad range of costs. The reason why they can coexist lies in that old and new techniques compete upon unequal terms: in fact, while new plants will be constructed when gross revenue is expected to cover the supply price of all factors of production, thus including capital, for plants already in existence the capital at work has no longer a supply price. Thus, the ‘unequal terms of competition give existing capital equipment a lease of life even when outmoded’ (op. cit. p. 62). By way of concluding this section we can say that: Nothing in our understanding of the economics of investments decisions suggests that new technology is necessarily economically superior to existing technology or that today is always the best time to invest in new technology. Delay may well be rational. (Metcalfe 1997: 128, emphasis added)

3  The competition between old and new technologies In this section we recall three examples of competition between old and new technologies – characterized by different principles and different effectiveness – in providing similar services. The general process is constituted by the fact that the development of the old technology is engendered by its eventual supplanter. The first example considered is the one which actually originated the expression ‘sailing ship effect’, the second refers to the so-­called ADSL technology in data transfer, and the third to superconductors and computers’ hard disks. The sailing ship example is recounted by Gilfillan who showed how the ‘old’ sailing ship was improved as steamships emerged during the nineteenth century. Improvements concerned nearly all of the components and materials of the sailing ship which was thus transformed from a basically wooden structure to a basically metallic one, whose carrying capability was massively improved. As Gilfillan pointed out: Large size was 5- to 800 tons in 1800, and for three centuries previously; by mid-­century it was 2,000, carrying hundreds, even 1,000 emigrants; and

The sailing ship effect   123 today 3- to 5,000 tons register. Iron was used first for bolts, then in diagonal straps for strengthening the hull . . . and for the whole hull, 1838, followed . . . by iron (1840) and steel (1863) in masts, spars and standing . . . (Gilfillan 1935: 156–7) The process which saw sailing ships disappearing was thus one characterized by technological progress: in fact the disappearance of the sailing ship was delayed by quite a long period. Eventually, however, the process reached its final point, as the new technology was overwhelmingly superior to the old one, as the steam­engine applied to ships could also be massively improved. The basic point is that: ‘It is paradoxical, but on examination logical, that this noble flowering of the sailing ship, this apotheosis during her decline and just before extermination, was partly vouchsafed by her supplanter, the steamer’ (op. cit. p. 156). Coming to the 1990s, the second example we consider concerns the case of the ‘appropriate’ medium for data transmission. Two basic sets of technologies were available: (i) the traditional ones, based on ordinary copper-­wire telephone lines and TV cables, and (ii) the new ones based on fibre optics. The main benefits of fibre optics consist of high bandwidth, low cost and small diameter of the cables – i.e. fibre optics outperform both copper and TV cables in each relevant characteristic. Given this technological situation one would expect a quick overtaking process of fibre optics over wires. However, things are not as simple. For instance, in 1995 the biggest Italian telephone company launched a campaign worth nearly eight billion dollars in order to fix fibre optics in ten million Italian homes, so that huge quantities of data could travel along the new fibres. However, in its early stages the plan was scrapped: in 1996 broadband modems appeared capable of overcoming the bandwidth limits of the copper wire. The development of such a device was actually stimulated by the need of transferring more and more data (Internet, digital TV and other) from providers to homes by means of the existing old telephone and traditional cable TV networks. Both networks, in fact, are ubiquitous, as they were started long ago, and virtually all families (of the rich countries) do possess a telephone line and a TV set. Two broadband contenders appeared, one for TV cables and another one for telephone lines – the latter technology is called asymmetric digital subscriber line or ADSL4 (Halfhill 1996; Brownstein 1997). These innovations were stimulated by the laggard component of the transmission system, at a time in which fibre optics were competitive under many respects. The third example refers, first of all, to the use of superconductors as better materials for building computer circuits, and serves to illustrate the relevance of the market size on the technological improvements. The promising technology of superconductors has not yet reached the market in spite of its enormous potential, which has been known since the early 1980s. Among the many proposed applications there are logical elements for digital circuits. The so-­called Josephson junctions, essentially two superconductors separated by a tiny non-­ superconductor barrier (Barone and Paternò 1982), are recognized as devices capable of operating at a speed about two orders of magnitude higher than

124   N. De Liso and G. Filatrella conventional silicium-­based semiconducting circuits and about one order of magnitude higher than the gallium-­arsenide compounds. Superconducting devices are capable to operate at very low levels of power dissipation per gate, perhaps three orders of magnitude lower than semiconducting devices (Haya­ kawa 1990), so easing the miniaturization problems and, in principle, making the speed limit even higher. Finally, the discovery of a new generation of superconductors, the so-­called High Tc Superconductors,5 has further moved this physical speed limits upward. There thus arises the question: why the superconducting revolution has not come yet? A first answer is that improvements in semiconductors have been so fast that it is difficult to estimate if and when their betterment will slow down. Not only in everyday life we witness constantly improved processors sold at constant or decreasing prices, but also the data on top-­performing computers (the ones for which the refrigeration necessary for superconductors are presumably not an issue) clearly demonstrate that the clock speed has increased exponentially through time. Thus, despite Josephson’s technology potentialities, semiconductors still constitute the centre of attention for computer makers and until the semiconductor technology will be able to make improvements, it will be more profitable for companies to improve the old technology rather than funding projects to convert a prototype into a marketable product. A second answer lies in that there exists a big difference between the two markets: semiconductors are massively employed, so even a tiny share of profits can boost enormous R&D research. Superconductors, instead, are niche products, and the companies that are developing such a technology are relatively small. Thus the investment in the two technologies is very different, and is actually much larger in the technology with a lower potential. As for hard disks, they give us the opportunity of illustrating the case of a technology approaching its upper limit. In fact, data storage on hard disks is close to experiencing the so-­called superparamagnetic effect, or SPE: Simply described, SPE is a physical phenomenon that occurs in data storage when the energy that holds the magnetic spin in the atoms making up a bit (either a 0 or 1) becomes comparable to the ambient thermal energy. When that happens, bits become subject to random ‘flipping’ between 0’s and 1’s, corrupting the information they represent. (Toigo 2000: 41) A few points seem to emerge from these examples. First of all there does not exist any automatism in the occurrence of the sailing ship effect; improvements in performance and/or cost reduction of the old technology occur as the result of an intentional process; firms may have a double incentive to stick to the old technology: on the one hand, they do not need to scrap the existing capital stock, while, on the other hand, they may find it easier to develop the technology they are already familiar with, rather than developing a new one;6 interrelatedness and the availability of extant networks making use of the old technology make more

The sailing ship effect   125 difficult the development of alternatives; technologies are characterized by an asymptotic upper limit, and usually newer ones have a much higher performance which leads to the displacement of the old.

4  The model 4.1  Performance dynamics of competing technologies In this section we propose our model which sheds some light on the dynamics which develop within the competition process between two technologies. As usual some simplifying assumptions have to be made in order to keep the model itself as simple as possible. We are aware of some of the limits which characterize such modelling and yet we believe that the simulations obtained are capable of depicting technological competition which develops in actual processes. Let us point out that a similar model has been proposed by us (De Liso and Filatrella 2008) but taking as a starting point profit maximization. One of the purposes of the present work is to show that similar results can be obtained without such an explicit operation, but with the help of a rule-­of-thumb procedure. By comparing the two methods we focus on the essential ingredients on which improvements of old technologies are based. In particular we will show that the perceived ‘dangerousness’ of the rival technology, defined as its market share, is a key factor that pushes competition between the two technologies. The initial situation is given by the existence of only one firm which supplies, by using a certain technology, the whole market of a given product. There exist monopolistic profits which attract another firm which will try to enter this market by making use of a newer and potentially better technology.7 Thus, when the second firm enters the market, two technologies exist which provide the same service; we will call A the ‘old’, and B the ‘new’ technology; technology and firm will be used as synonymous, thus we will refer to A (and B) as either technology or firm. Both technologies are characterized by an upper limit in terms of performance, which will be referred to as PAM and PBM; both technologies can be improved by means of an R&D activity, however, A, being the older technology, is characterized by both a lower upper limit, and a closer proximity to the upper limit itself with respect to B (when B appears). Also, in order to avoid needless complications, we assume there do not occur unintentional improvements – such as the ones due to learning-­by-doing – of both technologies’ performance. As already hinted, the initial situation is given by the existence of firm/technology A only, which allows the production of a certain quantity of output sold at a certain price. We do not investigate the growth of output; thus, we assume that demand always matches supply, and we will look at the market share of the two technologies; as far as price is concerned, it is performance-­related. Firm A earns positive profits πA(t), which depend on its market share SA(t) and on its technological performance, which is hereafter synthesized by our term PA(t); performance can be improved by R&D, and RA(t) indicates the expenditure

126   N. De Liso and G. Filatrella on research activities by firm A. In the same way, we will define SB(t), πB(t), PB(t) and RB(t) as the share, profits, technological performance and R&D expenditure of B, respectively. Total profits are standardized to 1.8 One more comment is needed to introduce what we call gains, which we define as GA(t) = πA(t) + RA(t), that is gains and profits coincide when RA(t) = 0. A similar expression will also be used for firm/technology B. We define SA (SB) as the fraction of the market captured by the technology A (B), and we assume that this fraction depends on the performances of the two technologies in the following simple way: S A (t ) =

S B (t ) =

PA (t ) PA (t ) + PB (t ) PB (t ) PA (t ) + PB (t )



(5.1a)



(5.1b)

It is important to notice that the market share depends only on the performances of both technologies at the same time, and therefore in this model no effect is included due to the tendency of the market to keep or ‘prefer’ an old technology. Equations 5.1a and 5.1b have been chosen as the simplest functional form that satisfies some symmetry conditions, to mimic consumers’ choices independent of anything but the quality (measured by the one parameter performance PA). For instance for PA(t) = PB(t), one gets SA(t) = SB(t) = 1/2, and neither the old nor the new one is preferred. Moreover, equations 5.1a and 5.1b satisfy the proper limit conditions: • •

SA(t) = 1 if technology A is much superior to technology B (PA(t) ≫ PB(t)); SA(t) = 0 if technology B is much superior to technology A (PB(t) ≫ PA(t)).

We emphasize that this particular choice does not contain any arbitrary parameter to fit the data, and assumes the simplest linear relationship SA(t)/SB(t) = PA(t)/PB(t). Assuming that profits are simply proportional to the market share, equation 5.1a implies that: πA (t ) = Q ( p − c ) S A (t ) − RA (t ) = Q ( p − c )

PA (t ) PA (t ) + PB (t )

− RA (t ) = GA (t ) − RA (t ) (5.2a)

where π represents the profits of the market in an interval. Here p is the price at which products are sold, and c the cost of production of the unitary product. Notice that Q is the total market at the time t. RA(t) is the flow of resources devoted to R&D, and thus sacrificed from the gains.9 As we have said in the preliminary comments to the model: (i) the old technology, A, will invest more in R&D, up to a certain level, when it sees that the new technology B gains market share; (ii) both technologies can be improved and are characterized by an upper limit, where the latter can be mathematically synthesised by the terms

The sailing ship effect   127 e

−

PAM

PAM − PA ( t )

for technology A and e

−

PBM

PBM − PB ( t )

for the new technology B (for instance, PAM and PBM at the numerator of the exponential function can be thought of as the maximum data carrying capacity of copper wires and fibre optics, respectively); as we shall further clarify, these upper limits will affect R&D spending levels RA and RB. Furthermore, let us point out that the higher the interest rate i the less resources will be devoted to R&D. Anticipating the analysis which will lead to equation 5.7 – see there for explanation – we can write: RA = G A S B e

−

PAM

PAM − PA ( t )

where x A (t ) = S B e

−

PAM  − 1  PAM − PA ( t )  = GA S B e = GA (t ) x A (t )  r r  

1

PAM

PAM − PA ( t )

1 r

(5.2b)

and r = (1 + i )

Thus we can write:10 πA (t ) = GA (t ) − RA (t ) = GA (t ) − GA (t ) x A (t ) = GA (t ) [1 − x A ]

(5.3)

that is technology A, and similarly technology B, faces the problem of how to determine the fraction xA of gains to be invested in R&D. Standard-­analysis efficient resource allocation requires maximization of (future) profits in spite of the temporary ‘loss’ of resources invested in research – and this is the method employed in De Liso and Filatrella (2008). However, what we want to show here is another, more direct, approach. Let us illustrate the dynamics of the competition once the potentially more advanced technology B emerges. At first B’s performance, share and profits equal 0, and SA = 1, i.e. the old technology takes all. For a finite value of the performances PB, we can write for B: πB (t ) = Q ( p − c )

PB (t ) PA (t ) + PB (t )

− RB (t )

(5.4)

that is, B’s profits depend on its own performance as well as on A’s performance. For this technology we can also write an equation for the sacrificed fraction of the gains: πB (t ) = GB (t ) − RB (t ) = GB (t ) [1 − xB ]

(5.5)

128   N. De Liso and G. Filatrella To derive the dynamics of the performance of either technology (at first we will refer to A) we reason as follows: the incremental variation of the performance of a technology is proportional to the resources invested in R&D, RA, times a factor which converts resources into (higher) performance, EA; the latter is the product of the initial efficiency γA, times a factor e

−

PAM

PAM − PA ( t )

– that is −

EA = γAe

PAM

PAM − PA ( t )

– that mimics the increasing difficulty of improving a technology when it approaches its upper physical limit. The dynamics of the performance of technology A can thus be written as: dPA (t ) = RA E A dt = RA γAe

−

PAM

PAM − PA ( t )

dt



(5.6)

In equation 5.6 technological performance can be improved only by investing in intentional R&D activity – RA – whose funding is detrimental to the profits: in fact, as we have pointed out πA(t) = GA(t) – RA(t). In order to determine RA we note that: • •

•

the amount of resources invested in research increases when the ‘dangerousness’, i.e. the share, of the rival technology increases; in the initial phase, the ‘conversion’ efficiency has the highest value, i.e. EA = γAe–1, while the asymptotic limit PA(t) @ PAM implies EA @ 0; research funding is affected by these conditions: the higher the efficiency, the higher the fraction of resources diverted towards R&D; and RA will be held back by alternative investment options, so the higher the interest rate i (or r = 1 + i), the lower the resources allocated to R&D.

Taking into account the three components which have just been mentioned, we can write equation 5.7 in which the fraction of resources dedicated to R&D is proportional to the share of the rival technology times the efficiency,11 divided by r. As we have mentioned, optimal R&D expenditure could be obtained by profit maximization. However we can get similar results by using two simple rule-­of-thumb guidelines; in fact R&D expenditure: (a) increases proportionally to the market share of the rival technology; the rationale behind this behaviour lies in the fact that the old technology will try to limit the expansion of the new technology by improving its own performance, which can only be done through R&D; expenditure will increase up to a certain level, but when the battle is lost

The sailing ship effect   129 the old technology will no longer invest in R&D; (b) is inversely related to the interest rate. These are the main driving forces of technological competition. In formulae, we get: 12

PAM PAM PAM −− PAM( t ) P B A − PA ( t ) A B

PM

PA (tP) (t ) PB (tP) (t ) − P M −−APA (tP)AM1 1 1 1 A M A RA =R G=A SGeS e = Q=( pQ− pc )− c ( PA)(t ) + PB (t ) PA (t ) + PBB (t ) e e PA − PAr(t=) = A r r PA (t ) + PB (t ) PA (t ) + PB (t ) r (5.7) PAM M PA (tP) P(Bt()tP) (t ) − P M −− PA (tP)A 1 1 A M = Q=( pQ− pc )− c ( [ PA) (t ) +APB (t )B]2 e 2 e PA − PAr(t ) r [ PA (t ) + PB (t )] −

Equation 5.7 could be also looked at from the perspective of equations 5.2b and 5.3 in which we have identified xA as the fraction of gains to be invested in R&D. A similar expression would be obtained for technology B. Thus, by reasoning for instance in terms of firm A, the function above states that the higher the share of B, the higher the R&D activity carried out by A. We assume that no part of gains is invested at all in R&D until there emerges the competition of B. The latter assumption, that at first sight might seem very ad hoc and radical, actually does not substantially affect the working of the model.13 Inserting expression 5.7 in equation 5.6 we get: dPA (t ) = GA

PB (t )

1

PA (t ) + PB (t ) r

= Q ( p − c)

−

γAe

2 PAM

PAM − PA ( t )

PA (t ) PB (t )

( PA (t ) + PB (t ) )

2

1 r

dt = −

γAe



2 PAM

PAM − PA ( t )

(5.8)

dt

Summing up, the increments of the performances are proportional to: 1 2 3 4 5

the gains earned by technology A, GA; the weight of technology B’s performance which affects the amount of R&D performed by A (that is the higher PB the more A will invest in R&D); the inverse of the interest rate 1/r (where r = 1 + i); a sort of ‘efficiency converter’ γA of the resources dedicated to R&D (that is the higher γA the more effective the expenditure in R&D); the efficiency weight e

−

PAM

PAM − PA ( t )

which limits A’s performance to its maximum value, PAM (this means that once technology A has reached its best performance PAM, its rate of performance improvement approaches 0);

130   N. De Liso and G. Filatrella 6

an additional factor e

−

PAM

PAM − PA ( t )

,

identical to the efficiency weight,14 that decreases the amount of resources devoted to R&D when the performances approach the limit value PAM; the rationale is that when resources are poorly translated into improved performances the firm is discouraged from funding further R&D activities. Of course, we can write a similar expression for technology B: dPB (t ) = GB

PA (t )

1

PA (t ) + PB (t ) r

= Q ( p − c)

−

γB e

PA (t ) PB (t )

2 PBM

PBM − PB ( t )

( PA (t ) + PB (t ) )

2

1 r

dt = −

γB e



2 PBM

PBM

− PB ( t )

(5.9)

dt

Thus, dividing both sides by dt, we obtain: dPA (t ) dt dPB (t ) dt

= Q ( p − c)

= Q ( p − c)

PA (t ) PB (t )

1

[ PA (t ) + PB (t )]2 r PA (t ) PB (t )

1

[ PA (t ) + PB (t )]2 r

γAe

γB e

−

−

2 PAM

PAM

− PA ( t )

2 PBM

PBM − PB ( t )



(5.10a)



(5.10b)

The factors γA,B discriminate between the case in which technology A is more efficient in using resources to improve its performance (γA > γB) and the case in which technology A is less efficient (γA < γB). In the following we will assume γA = γB, that is both technologies have the same capability of transforming resources into improved performance via R&D activity. 4.2  The evolution of performances, market share and R&D expenditure of the old and new technology In this section we show the simulations of the magnitudes we are interested in, that is performances, market share and R&D expenditure. Furthermore, we will also provide a comparison with results concerning the same magnitudes which have been obtained via profit maximization.15 Worthy of comment is the fact that equations 5.10a and 5.10b represent the Ordinary Differential Equations (ODE) governing the dynamics in continuous time, while the results obtained in De Liso and Filatrella (2008) come from a discrete-­time map approach. The two approaches might be directly compared if one could translate continuous equations into a discrete form, which could be done by using the following equations:

The sailing ship effect   131 t +1

PA (t + 1) =

∫ t

t +1

PB (t + 1) =

∫ t

dPA [ PA (t ′), PB (t ′) ] dt ′ dt ′

(5.11a)

dPB [ PA (t ′), PB (t ′) ] dt ′ dt ′

(5.11b)

However, system (5.10) cannot be analytically integrated, so that integrals 5.11a and 5.11b cannot be solved. Therefore we will draw a comparison between equations 5.10a and 5.10b and subsequent equations 5.12a and 5.12b by means of numerical simulations. For convenience we repeat here the map obtained in De Liso and Filatrella (2008) by explicit profit maximization (albeit with a truncation to the second order in performances’ increment to get an explicit solution16):  2 PA (t ) + PB (t )  r ( PA (t ) + PB (t ) ) 1 − PA (t ) = PA (t − 1) + PAM 2  − PAM − PA ( t )  PB (t ) ( p − c ) γ AQe

    

(5.12a)

 2 PA (t ) + PB (t )  r ( PA (t ) + PB (t ) ) 1 − PB (t ) = PB (t − 1) + PBM 2  − PBM − PB ( t )  PA (t ) ( p − c )γB Qe

    

(5.12b)

Should the term in the square brackets be negative, no resources would be allocated to R&D as performances would deteriorate, so that, at worst, PA(t) = PA(t – 1) and, similarly, PB(t) = PB(t – 1). In the present ODE analysis R&D expenditure results from an heuristic approach while in the map case R&D expenditure is the result of a profit maximization process; in fact, the results for the R&D expenditure in the map approach, analogous to equation 5.7 reads:  2 PA (t ) + PB (t )  r ( PA (t ) + PB (t ) ) 1 − RA (t ) = PAM PAM  − − M PA − PA ( t ) PAM − PA ( t )  PB (t ) ( p − c ) γAQe 2e

    

(5.13)

Let us point out two technical details. First, the two approaches employ the same set of normalizations, so the comparisons shown in Figures 5.1–5.3 share the same set of parameters. Secondly, the time step of the map (from t to t + 1) corresponds to one unit in the ODE time.

132   N. De Liso and G. Filatrella Before drawing a direct comparison let us also derive some analytical considerations from the system of differential equations 5.10a and 5.10b. The structure of these equations reveals that: (i) each term in the system has no inertia, i.e. it does not show any tendency to behave according to its previous ‘motion’; (ii) when the performance of both technologies is far away from the asymptotic values (PA ≪ PAM, PB ≪ PBM ) the increments that one gets in the performance itself are proportional to the initial values. While the first consideration is an immediate consequence of the fact that the differential equations are of the first order, the second statement is not so straightforward. To prove it we note that if the arguments of the exponential terms are both approximately –1, the right-­hand-sides of equations 5.10a and 5.10b simplify, and one can show that the following relation holds: dPA (t ) γB dPB (t ) = γA dt dt

(5.14)

Equation 5.14 can be readily integrated: PA (t ) − PA (0) = ( γB / γA ) [ PB (t ) − PB ( 0)]

(5.15)

The other side of the technological competition consists of the change of the market share of technology A relative to technology B: in fact, A’s share decreases dramatically, until it reaches the asymptotic value which we obtain as: GA (t → ∞) GB (t → ∞)

=

PAM PBM

⇒

S A (t → ∞) S B (t → ∞)

=

PAM PBM



(5.16)

that is, the long-­run profits and shares depend on the maximum values of the performance of both technologies. We are now able to run the simulations which depict the performance (Figures 5.1a and 5.1b) and share (Figure 5.2a and 5.2b) of both technologies through time, comparing what happens according to the ODE approach – panel (a) – with the results of the map approach – panel (b). In the simulations we assume that once B emerges, both A and B invest resources in R&D activities. However, as we will see in the figures and in the tables, despite the fact that A, starting from a dominant position invests at the same rate as B (see Figure 5.3), A’s performance cannot be improved as fast as would be necessary to keep up with B’s. Put another way, A and B invest the same fraction of their gains, but A’s pie is a decreasing one, while B’s is increasing. The behaviour shown in Figure 5.1 is general, but the actual values of t1 and t2 depend upon the initial conditions. Regarding the parameters, let us underline that we have chosen the initial conditions in such a way (PA = 0.1, PB = 0.001) that the emerging technology is confined to a very small market niche, just 1 per cent of total production; the upper limit of the new and better technology B is

The sailing ship effect   133

(a)

(b)

Figure 5.1 Performances of technologies A (solid line) and B (dashed line) as a function of time. The long-dashed horizontal line represents the supposed constant behaviour of technology A. In the figure the overtaking times of the two technologies are also indicated, assuming that technology A does not react (t1) or that it does (t2). Parameters of the simulations are: γa = γb = 5, PAM = 1, PBM = 5, r = 1.05. Initial conditions: PA(0) = 0.1, PB(0) = 0.001. (a) Ordinary differential equation model – ODE (equations 5.10a and 5.10b). (b) Map model –equations 5.12. Circles and crosses refer to technology A and B respectively.

‘only’ five times A’s, so that the maximum technology performances are in a ratio of 1 : 5 (PAM = 1, PBM = 5); finally, we have chosen two identical factors that convert resources into higher performance (γA = γB = 5), to emphasize the fact that this conversion factor is not crucial for our conclusions. Let us underline that if the factors are identical, they do not affect the model’s dynamics because they can be included in the time normalization – see equations 5.10a and 5.10b. The behaviour of performances described in Figure 5.1 is quite interesting. Technology A (solid line) starts reacting when B (dashed line) comes into existence; had we not had reaction of technology A to technology B, A’s performance would be represented by a horizontal line parallel to the time axis (the long-­ dashed line is a continuation of such performance) and the expected time of technological overtaking would be t1. A’s performance however increases, so that the overtaking of B is delayed in time, and the overtaking occurs at a relatively higher level of performance and at a later time, t2. This is the phenomenon which we wanted to create a model for: the dynamics engendered by the emergence of the new technology which stimulates the performance of the old one. Let us remember that in this simulation γA is assumed to be identical to γB. Should we have γA ≠ γB, and in particular should we have γA > γB, A would be

134   N. De Liso and G. Filatrella more efficient than B in converting R&D into improved performance; in terms of our Figure 5.1 the overtaking of technology B over technology A would occur later in time and at a higher level of performance. The converse would be true if we had γA < γB, i.e. B would overtake A earlier and at a lower level of performance. Moreover, we want to stress that the long run position is not affected by the values of γ – see equation 5.16. In Figure 5.2 the solid line depicts the market share of technology A, while the dashed line describes B’s share; the two lines also reflect the fraction of gains devoted to R&D by the other technology, i.e. B’s line is proportional to the fraction of gains devoted to R&D by A, and conversely. Tables 5.1 and 5.2 supply the values which the relevant variables take at some critical points of time for the ODE and the map model, respectively. Table 5.1 shows that at the beginning of the simulation, that is when t = t0 = 0, A gets nearly all of the gains with its performance being PA = 0.10 and PB @ 0 (0.001). After 1.2 units of time (at t1) B would have overtaken A, had A not undertaken its R&D activity: in fact the value of PB at t1 is the same as the value of PA at t0. However, given A’s reaction in terms of R&D, A’s performance is still higher than B’s (PA = 0.17 as compared with PB = 0.10). In this point the old technology is close to investing the maximum amount of resources in R&D (see also Figure 5.3). The two technologies have the same performance and share at t2 = 2.47, with PA = PB = 0.28 and both have a 50 per cent share; it is worthwhile noting that at this stage A is investing in R&D 12 per cent of its gains which correspond to 6 per cent of the overall market value.17 At t2 B is investing 16 per cent of its gains in R&D, which correspond to 8 per cent of the overall market. At t3 we show what happens for a finite but very long time: A’s performances keep increasing and reach the value 0.69 – which is slightly more than twice as

(a)

(b)

Figure 5.2 Market shares – equations 5.1a and 5.1b – based on the evolution of performances of figure 1. When technology B (dashed line) develops market shares reverse in both (a) the ODE and (b) the map approach.

The sailing ship effect   135 Table 5.1  Critical values for the Ordinary Differential Equation approach

t0 t1 t2 t3 t∞

Time

PA

PB

SA %

SB %

R&D Tech. A R&D Tech. B % of market % of market

 0   1.20   2.47 50 ∞

0.10 0.17 0.28 0.69 1

0.001 0.10 0.28 2.48 5

99 63 50 22 17

 1 37 50 78 83

0.4 6 6 0.6 0

1 8 8 2 0

Note We denote with: t0 the initial time at which technology B appears; t1 the time at which technology B should have overtaken technology A if technology A had not reacted; t2 the actual time after which the performance of technology B overtakes technology A; t3 a very long time (not shown in Figure 5.1). Finally, the value of the variables in the asymptotic regime are shown in correspondence of t∞.

much as it was at t2, its share is 22 per cent, and A’s R&D falls to 2.7% (= 0.6 per cent ÷ 0.22); B’s performance increases much faster and reaches the value 2.48 – which represents an increase of a factor 8.9 with respect to t2, its share is 78 per cent, and B’s R&D falls to 2.56 per cent (= 2 per cent ÷ 0.78). Finally, as t → ∞, both technologies will have reached their upper limit in terms of performance (PA = 1, PB = 5), while the share of A falls to 17 per cent and B’s has grown to its maximum of 83 per cent; with A and B both having reached their maximum performances, the R&D activity of both would be 0. It is worth stressing that A’s share does not fall to 0, and thus it continues to survive in a long-­run niche: the actual 17 per cent value is given by rearranging equation 5.16 SA SB

=

SA (1 − S A )

=

PAM PBM

−1

PAM  PAM  ⇒ S A = M 1 + M  PB  PB 

(5.17)

Table 5.2, which refers to the map approach, initially shows the same conditions as in the previous table, but we have to note that there exists a difference in the R&D expenditure, due to the different principle according to which the two firms are behaving – in this case they are maximizing profits. Thus, the old technology finds it profitable not to invest any resource in R&D because missing a mere 1 per cent of the overall market does not provide a strong enough incentive to invest in R&D. Conversely, B finds it profitable to carry out R&D activities up to 10 per cent of its initial gains (= 0.1 ÷ 0.01). After 1.8 units of time (at t1) B would have overtaken A, had A not undertaken its R&D activity: in fact the value of PB at t1 is the same as the value of PA at t0, that is 0.10. Also in this case, A’s reaction results in a performance which is still higher than B’s (PA = 0.14 while PB = 0.10). With regards to R&D expenditure both A and B are heavily expanding this activity. At t2 = 3.4 the two technologies are characterized by the same performance (PA = PB = 0.25) and share (50 per cent each), and are respectively investing 4.8 per cent and 5.7 per cent of the market value (or 9.6 per cent and

136   N. De Liso and G. Filatrella Table 5.2  Critical values for the map approach

t0 t1 t2 t3

Time

PA

PB

SA %

SB %

R&D Tech. A R&D Tech. B % of market % of market

0 1.8 3.4 8

0.10 0.14 0.25 0.30

0.001 0.10 0.25 0.39

99 53 50 43

 1 47 50 57

0 2.6 4.8 0

0.1 3.4 5.7 0

Note We denote with: t0 the initial time at which technology B appears; t1 the time at which technology B should have overtaken technology A if technology A had not reacted; t2 the actual time after which the performance of technology B overtakes technology A; t3 is the time at which the dynamics is stable and shows no further changes

11.4 per cent of their gains). Time t3 = 8 shows the point at which both technologies no longer find it convenient to invest in R&D, and their shares stabilize at 43 per cent for the old technology and 57 per cent for the new. Let us add here some comments on the difference between the maximization used to derive the map (that therefore should be closer to a ‘rational’ approach) and the heuristic derivation of the differential equation based on the rule-­ofthumb described by equation 5.7. 1

From Figure 5.1 it is clear that we do not have a quick and extended deployment of the technologies; furthermore, the asymptotic regime is never reached in the map approach [panel (b)], i.e. it is convenient to stop invest-

(a)

(b)

Figure 5.3 R&D expenditure based on the performance evolution of Figure 5.1; expenditure decreases when performances approach saturation. As in the previous figures technology A is represented by the solid line, while technology B is represented by the dashed line in both (a) the ODE and (b) the map approach.

The sailing ship effect   137

2 3

ments in R&D before full exploitation of the potential maximum performances of technologies is reached. Figure 5.2 makes it clear that a rational allocation of resources leads to the survival of the old technology, which will also be characterized by a high market share (43 per cent). In Figure 5.3 it is evident that the total amount of resources devoted to R&D – represented by the areas under the curves – is much smaller under the maximization hypothesis.

5  Conclusions The features of the model illustrated above constitute at the same time both its strengths and weaknesses. In fact, in order to keep the model as simple as possible we have made use of first-­order deterministic differential equations while we have limited the number of free parameters to two, namely the maximum M performance that can be reached by each technology (P A,B ) and the capability of transforming R&D into improved performance (EA,B). Different functions might have been considered at the price of a much more complicated model, whose heuristic capabilities, however, would have not been improved, that is, the qualitative results would not change. The most basic criticism that can be levelled is that R&D activities actually induce stochastic processes of improvements. We are aware of this, and the answer we have lies in the fact that the two firms develop an existing technology, and this renders the process of improvement less uncertain. Furthermore, if we look at the history of technology, from steel production to the performance of computers or jet engine power, some ‘deterministic laws’ of performance growth have been highlighted – deterministic in the sense that one observes a mechanism that can be relied upon. A further point that could be raised concerns rationality: in fact, one could argue that it is not rational to continue to try improving a technology when a new superior one appears. This argument can be answered with economic, technological and historical examples. In fact, in this model despite the old technology being inferior, it realizes positive profits; in addition, we have seen that old and new technologies compete on unequal terms; thirdly, to master technology A does not imply that one could master the new – often radically – different technology B. To stick to Gilfillan’s example, if one deals perfectly with sails he/she cannot necessarily deal with the steam-­engine with the same degree of understanding and success. The same is true for semiconductors and superconductors: if one can master the first technology, it does not necessarily mean that one can master the latter – not least because of the cryogenic techniques which are necessary to cool down the superconductors. Furthermore, we have to take into account that the user-­owner of the old technology may possess privileged information on the technology that he/she has been using for some time. The use of this information, or latent knowledge, may thus be stimulated only by the emergence of a competitor. On the other hand,

138   N. De Liso and G. Filatrella should the user-­owner of the old technology try to switch to the new one, he or she would face two difficulties: on the one hand, all the knowledge that had been accumulated over time and through experience would be lost; on the other hand, one would find him/herself being a follower, late in adopting and developing the new technology. Furthermore, at the beginning of any new technology a breaking-­in period usually exists which may lead to the belief that the existing technology can successfully fight off the emerging one, keeping it at bay. From all of this there may follow that the superiority of the new technology will emerge only ex post or after a certain amount of time, and thus the rational answer to the appearance of the steam-­engine may well be a further concentration on the sail technology. To conclude, let us note that we started this chapter by pointing at a phenomenon which is observed in economic reality, i.e. the persistence of ‘old’ technologies despite the emergence of new ones capable of offering the same services in a more efficient way. The kind of mechanism we have modelled consists of the new lease of life which comes from the intentional effort activated by the ‘owners’ of the old technology in order to give their technology one more chance – and examples range from the sailing ship during the nineteenth century to broadband modems nowadays. As we have seen, eventually new technology overtakes old; however, this overtaking occurs at a later time than would have been the case if the old technology had not been improved, and then overtaking occurs at a higher level of the performance for both technologies – a higher performance that would have never been reached had the new technology never appeared. We need to stress that this delay in overtaking in performance (t2 rather than t1 in terms of Figure 5.1) is a qualitative result which does not depend upon the values of the parameters. The model is also capable of depicting the evolution of the market share of the two technologies which is reflected in their different performances over time, with the old technology experiencing a dramatic fall as the new one emerges and is improved. The share of old technology, though, does not fall to zero. It is worthwhile stressing that lock-­in does not occur as there does not exist inertia in this system; put another way, the final equilibrium does not depend on history. Last but not least, the final equilibrium, i.e. the share of the two technologies, depends on the asymptotic values of the best performance that each technology can reach.

Notes   1 The authors wish to thank Ian Gavin for the linguistic appraisal of the text.   2 Some attention ought to be devoted to ‘bridging technologies’ which are systematically developed through time between techniques or technologies held to be, at first, incompatible.   3 Briefly, this is Frankel’s argument: by defining P = past outlays of the old method, F = future outlays of the old method, F+P = total costs of the old method, T = total costs of the new method, R1= total revenue from the old method, R2 = total revenue from the new method, the two following criteria emerge: if [R1 – (F + P)] < (R2 – T)
(R1 – F ) the old method should be immediately replaced (Frankel 1955).   4 Those who are interested in the diffusion of ADSL through time can refer to EITO (2007).   5 For a concise summary of the most recent developments in superconductivity one can refer to Collins (2009).   6 In our model we will not consider phenomena such as R&D spillovers (see Breton et al. 2004) or whether a research joint venture would be preferable (see Dakhlia et al. 2006).   7 The model could be generalized to n firms/technologies, giving rise, in what follows, to n equations.   8 Should we assume that profits depend on performances, as the latter have an upper limit, eventually in the asymptotic regime profits themselves would become constant and could be normalized to 1.   9 GA(t) = π A(t) + R A(t) which can be rewritten as π A(t) = GA(t) – R A(t). 10 Occasionally we will omit t in order to have handier equations. 11 In equation 5.7 γA is not considered as the dynamics of the efficiency depends on

e

−

PAM

PAM − PA ( t )

only, γA being constant.

12 See the comments after equation 5.8. 13 Hinting at the result, if we abandon this assumption, and consider that A has an R&D activity going on before B comes into life, the old technology being overtaken by the new technology would occur later in time and at a higher level of performance – see Figure 5.1 and the illustration of the model within this section; however the essential heuristic capabilities of the model do hold. 14 This explains why in equation 5.8 the numerator of the exponential function contains 2PM and not simply PM. 15 See De Liso and Filatrella (2008). 16 Equations 5.12a and 5.12b are a simplified version of eq.s A1 and A2 which can be found in De Liso and Filatrella (2008). Should the reader be interested, with respect to the original equations we set the exponent n to 1 and we omit the J function; the role of the latter was to set the term in the square brackets to 0, so that it cannot become negative. 17 The fraction of gains invested in R&D is obtained by dividing R&D expenditure by the share: the figure for A is thus obtained by dividing 6 by 0.50 (where the latter is the 50 per cent share indicated in column SA).

Bibliography Arrow K. (1962), The Economic Implications of Learning-­by-Doing, Review of Economic Studies 29: 155–73. Barone A. and Paternò G. (1982), Physics and Applications of the Josephson Effect, New York, Wiley. Breton M., Turki A. and Zaccour G. (2004), Dynamic Model of R&D, Spillovers, and Efficiency of Bertrand and Cournot Equilibria, Journal of Optimization Theory and Application, 123(1): 1–25. Brownstein M. (1997), Batter up for Broadband, Byte, 22(10): 71–4. Collins G.P. (2009), An Iron Key to High-­Temperature Superconductivity?, Scientific American, 301(2) August: 56–63. Dakhlia S., Menezes F.M. and Temimi A. (2006), The Role of R&D Technology in Asymmetric Research Joint Ventures, The Manchester School, 74(1) January: 52–63. De Liso N. and Filatrella G. (2008), On Technology Competition: A Formal Analysis of

140   N. De Liso and G. Filatrella the Sailing Ship Effect, Economics of Innovation and New Technology, 17(5–6) July–­ September: 593–610. EITO (2007), European Information Technology Observatory. Annual Report on ICTs, Frankfurt, Eito. Frankel M. (1955), Obsolescence and Technical Change in a Maturing Economy, American Economic Review 45(3): 296–319. Gilfillan S.C. (1935), Inventing the Ship, Chicago, Follett Publishing Co. Halfhill T.R. (1996), Break the Bandwidth Barrier, Byte, 21(9): 68–80. Hayakawa H. (1990), Computing, in S.T. Ruggiero and D.A. Rudman (eds) Superconducting Devices, New York, Academic Press. Kroger H. (1986), Josephson Devices and Technology, in Japanese Assessment, Park Ridge, NJ, Noyes Data Corporation, pp. 250–306. Malone M.S. (1995), The Microprocessor: A Biography, TELOS (The Electronic Library of Science), Santa Clara, Springer. Metcalfe J.S. (1997), On Diffusion and the Process of Technological Change, in G. Antonelli and N. De Liso (eds), Economics of Structural and Technological Change, London, Routledge, pp. 123–44. Rosenberg N. (1982), Inside the Black Box: Technology and Economics, Cambridge, Cambridge University Press. Salter W.E.G. (1969), Productivity and Technical Change, Cambridge, Cambridge University Press. Smith A. (1776), An Inquiry into the Nature and Causes of the Wealth of Nations, 1976 reprint edited by R.H. Campbell, A.S. Skinner and W.B. Todd, Oxford, Clarendon Press. Toigo J.W. (2000), Avoiding a Data Crunch, Scientific American, 282(May): 40–52.

6 Software patents and firms’ strategic behaviour in the EU Some empirical insights1 Francesco Rentocchini and Giuditta De Prato

1  Introduction During the last ten years the number of filed and granted patents2 at the main patent offices (the United States Patents and Trademark Office (USPTO), the European Patent Office (EPO), the World Intellectual Property Organization (WIPO) and the Japan Patent Office (JPO)) has increased spectacularly. This increase has been driven mainly by patent filings in high-­tech classes (Hall 2004; Kim and Marschke 2004). Among these, software patents attract particular interest mainly because of the nature of the technology and because software patentability has been, quite recently, at the centre of a debate at the European level. For a long time, the economic literature has recognised the importance of the patent system in shaping and directing the rate of appropriation of the innovative effort of the firm (Arrow 1962; Nordhaus 1969). Besides ‘classical’ contributions, the literature that has been developed to explain the recent trends in worldwide patenting has relied on Schumpeter’s contributions to economic thought (Schumpeter 1942). More recently, evolutionary economics (Nelson and Winter 1982) has focused on the role of patents in enhancing or hindering innovation depending on sectors where firms compete. Therefore, a number of authors underline that, depending on appropriability conditions of sectors in which they are used, patents might, or not, be a useful institutional mechanism in order to promote the variety of technological solutions and the selection by market forces via competition. Moreover, empirical contributions have shown that firms do not always rate patents as effective appropriability mechanisms (Cohen et al. 2000). Hence, on one side, empirical literature has shown how patents are not suitable appropriability mechanisms in a high number of sectors, but, on the other side, we have witnessed an explosion in the number of patents filed in recent years. Why is there such a trade-­off? Which factors contribute to explain it? One of the main reasons refers to strategic patenting, which is a strategic behaviour of firms aimed at hindering competition, obtain licensing revenues and to have stronger power in negotiations. Our work wants to give an account of the patenting behaviour of a complex technology such as software. While for the US some works have already been presented (Bessen and Hunt 2003; Graham and Mowery 2004), the European

142   F. Rentocchini and G. De Prato Union has been disregarded mainly because of Article 52 of the European Patent Convention (EPC) which regulates patenting activities inside the Union and expressively prohibits the patentability of software and business methods. This exception is not applied in practice indeed we show that more than 40,000 patents have been accorded by the European Patent Office (EPO) in the period 1981–2004. It must be stressed that industries where the innovation process relies mainly on improvements made by others, namely cumulative system technologies (Mazzoleni and Nelson 1998), are more likely to be characterised by strategic patenting behaviours such as cross-­licensing, blocking rivals or gaining licensing revenues. It is for this reason that our work is going to analyse the software patenting by European firms. In order to achieve this, we first discuss the extent of patenting taking place in the EU (Section 2). After that, Section 3 shows that software patenting is a phenomenon common to the European Union as well and not only to the US patent system. Second, a theoretical model explaining factors affecting software patenting at the firm level is put forward in Section 4, together with the most suitable estimation strategy for the problem at stake. In particular, econometric analysis via different types of count data models is put forward in order to find out the most relevant factors affecting software patenting decisions for firms belonging to the software and hardware sectors. As a third step, in Section 5, an original dataset for the period 2000–2003 is constructed, which links the number of software patents filed at the EPO with the R&D spending and other relevant variables of applicants. Finally, in Sections 6 and 7 we present the main results and comment upon them briefly.

2  Patenting in the European Union It is a well-­recognised fact that the EPO faced increasing requests for inventions to be patented. Even if the rate of increase is not comparable with the one of the USPTO, it is surely important and prominent in new technologies. The average annual growth of EPO applications for the period 1995–2001 is more than 8 per cent, with a peak of 12 per cent in both biotechnology and ICT (see Figures 6.1 and 6.2); in addition, the propensity to patent3 rose spectacularly inside the EU with an overall increase of nearly 50 per cent in the period 1995–2000 (see Figure 6.3) (OECD 2004). This upward surge in patenting is due to the joint contribution of member and foreign countries. Moreover, other important changes have taken place into the European patent system. A first relevant aspect is represented by the increase in the length of the granting procedure, now some three and a half years4 (Malerba and Montobbio 2002). Second, it is worth noticing that the number of designated countries, whether considered in absolute terms or compared to the number of designable countries,5 has increased in recent years, pointing out the rise in the number of countries where patent protection is asked for. This means that inventors are keen to internationalise their competences and that they are probably trying to reach new markets using patents as competitive assets. Obviously, the increase in economic integration among member countries and the number of designable countries are important factors to this respect.6

8,000

7,000

Number of patents

6,000

5,000

4,000

3,000

2,000

1,000

0 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

Year

Total high-tech

Computer and automated business equipment Aviation Semiconductors

Micro-organism and genetic engineering Communication technology

Figure 6.1 Patents accorded by the EPO to EU15 countries (1997–2002) – high-tech applications.

70,000

60,000

Number of patents

50,000

40,000

30,000

20,000

10,000

0 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

Year European Union (15 countries)

Figure 6.2  Patents accorded by the EPO to EU15 countries (1977–2002) total.

144   F. Rentocchini and G. De Prato 600

Number of patent applications

500

400

300

200

100

0

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

Year Total number of patent applications by billion EUR of Business enterprise sector’s R&D expenditure (BERD) Total number pf patent applications by billion EUR of total R&D expenditure (GERD)

Figure 6.3  Patents accorded by the EPO to EU15 countries – propensity to patent.

The strong increase in the number of patents filed at the EPO has an unusual character, though. It seems that the growth in the number of patents filed has not been accompanied by a comparable increase in research inputs. The explanation of this last fact is far from being totally agreed upon by the literature. Some authors underline the importance of the increase in the productivity of R&D inputs which yields more inventions and, as a consequence, more patents (Kortum and Lerner 1999). On the other side, Hall and Ziedonis (2001) pointed out the importance of changes in the worldwide patent system which could have brought to an increase in patent propensity without a parallel rise of inventive activity. This last explanation is in line with the increase of strategic patenting as a mean to appropriate monopolistic revenues. In fact, a growing number of large firms started relying on patenting strategies that allow them to collect a huge amount of patents with the only aim of hindering competition or preventing being held-­up by rivals. In this respect, an interesting study has been put forward by Calderini and Scellato (2004) who analysed patenting behaviour in the European telecommunications industry.7 Analysing oppositions to patents filed at the EPO during 1980–2002 in the telecommunications industry, they show that large incumbents have mutual opposition rates lower than those involving one small and one large patentee. This result may indicate a threat effect of large patentees toward small ones, which are usually new and more innovative firms. Thus in a sector of the ICTs, such as telecommunications, incumbents have been found to use patents for reasons other than appropriation of the results of the research and development process. In this sense strategic patenting is devoted to hindering more innovative firms thus eroding their market share. This poses relevant problems for the variety and selection processes inside the sector: it seems that competition is not fair in this case.

Software patents and firms’ strategies   145

3  European software patent Article 52 of the European Patent Convention (EPC) defines what inventions are and when they are patentable under the EPC. Article 52(1) states: European patents shall be granted for any inventions which are susceptible of industrial application, which are new and which involve an inventive step. Definition of what an invention is, or is not, is not provided by the EPC. However, Article 52(2) provides a list of things that shall not be regarded as inventions: 1 2 3 4

Discoveries, scientific theories and mathematical methods; Aesthetic creations; Schemes, rules and methods for performing mental acts, playing games or doing business, and programs for computers; Presentations of information.

The scope of this list is reduced by Article 52(3) which states that the provisions of paragraph 2 shall exclude patentability of the subject-­matter or activities referred to therein only to the extent that an application or a patent relates to such subject-­matter or activities as such. This last statement has been recently used to put forward the idea that inventions having a technical character, and hence that are or may be implemented by computer programs, may well be patentable. The Directive on the patentability of computer-­implemented inventions (2002/0047/COD) was such an attempt. Its aim was two-­fold: first to harmonise national patents law of member countries about software programs, second to stop a situation where software, even if not eligible for patentability, were patented as well.8 The directive, first initiated by the European Commission on 20 February 2002, had a long and difficult journey, ending on 6 July 2005 with the definite withdrawal of the proposal via parliamentary rejection on the second reading. This means that for the moment, even if patenting of software and business methods patents are not permitted under the EPC, the EPO is going to provide a plurality of actors with patents on software. Obviously, this is possible given the presence of a different interpretation of the definition of software and, as a consequence, of that of software patents as well.9

4  The model As it has been previously pointed out, our main goal is to describe the patenting behaviour of firms in the European Union with a particular interest in software patents. In this section we will present the main features of the model from both a theoretical and empirical point of view. In particular, we will first discuss the

146   F. Rentocchini and G. De Prato theoretical framework surrounding our work, namely the knowledge production function approach (see Section 4.1). From the empirical standpoint, this approach relies heavily on the use of patent data as proxy for the dependent variable. Thus, in Section 4.2, we present the empirical model and address some econometric issues regarding the estimation of the parameters of interest. 4.1  Theoretical background The study of the effect of Research and Development spending, and other factors, on the number of patents filed has relied mainly on the Knowledge Production Function (henceforth KPF ) approach. The main idea is that the R&D expenditure at the micro level could be interpreted as a correct proxy for the production of knowledge. Then, if we are able to calculate the stock of knowledge for a certain firm at a fixed point in time, this value might have a high validity in explaining the output of the KPF. The problem remains that what it is produced through the R&D effort of the firm is a rather unobservable quantity, namely technological knowledge. Hence, a good index of the output of this process is needed. In this regard, the economic literature has relied on the number of patents filed by a single firm in a fixed point in time.10 Even if this index has relevant drawbacks, among which the fact that not all new innovations are patented and that patents differ in their economic impact, it has been widely adopted during last 20 years. This happened because patent statistics are easily accessible, which is even more true now after that worldwide patent offices (USPTO, JPO, EPO and WIPO) have computerised their data and have granted the public web access. The amount of R&D expenditure at the firm level is not the only factor affecting the output of the KPF. Other factors, such as the size of the firm and technological effects are likely to be important to this respect. Hence, in order to take into consideration these effects, we rely on a theoretical model that takes into account these factors too (see Figure 6.4). In this figure the R&D expenditure contributes directly to the formation of the knowledge capital of the firm. The relationship between R&D and knowledge capital is straightforward: the amount of R&D expenditure contributes directly to firm’s stock of knowledge.11 Hence, knowledge capital is the factor that influences directly the output of the KPF. But other components have an important role to play. For instance, firm size should affect innovation output as well. But this can happen both directly, through the greater capital that the firm could have at its disposal, and indirectly, thanks to the positive feedback effect that a larger firm has on its R&D spending. The same applies for the other factor indicated in Figure 6.4, namely technological determinants, constituted principally by technological opportunities that we proxy through the industrial sector of activity of the firm. Indeed, the effect of formal R&D spending on the innovation output, mediated by the rate of formation of the stock of knowledge capital, depends on the sector of activity of the firm (Mansfield 1986). Technological opportunity is of particular interest here. In fact, we want to investigate the different behaviour taken by firms belonging

Software patents and firms’ strategies   147

Figure 6.4  Innovation equation.

to two separated sectors, namely hardware and software producers. It has been showed that, during the last ten years, main patentees at the USPTO are likely to be part of electrical, computing and instrument industries (Hall 2004). Moreover, if only software patents are taken into account firms belonging to electrical, machinery and instruments account for more than 60 per cent of software patents accorded at the USPTO, while software publishers and firms from other software industries contribute only 7 per cent to the overall share of software patents (Bessen and Hunt 2003). Hence, if firms not belonging to the software sectors are more likely to patent software inventions, then it seems reasonable to suppose that they are doing it for reasons intrinsically different from spurring innovation spending. In fact, while traditional ‘incentive theory’ advocates that the monopoly power, accorded to the patent holder, acts as an incentive to R&D expenditure,12 recent contributions assert that the high number of patents filed by companies, in particular larger ones, are instead a strategy aimed at hindering competition and increasing their monopolistic position (Merges and Nelson 1990; Hall and Ziedonis 2001). This is particularly true in what have been called ‘cumulative system technologies’, that is technologies where the innovation process is highly cumulative. Therefore the software sector, with the essential cumulativeness of its embedded technology, is also prone to be threatened by strategic patenting activities; and this is what we think it is happening in the EU too. 4.2  Method of estimation On the grounds of the previous section, the main focus of our analysis will be the number of patents a firm applies for. Before continuing and going into the details of how the database was built up, we would like to spend some time on the peculiarity of the different estimation methods that have been implemented.

148   F. Rentocchini and G. De Prato The main object of the analysis is to explain which factors influence the number of software patents a firm applies for at the EPO. Hence, our dependent variable is of a count data type, that is it can assume only positive integer values. Given this particular feature, together with the fact that we are facing micro-­level data repeating through time, we rely on count-­panel-data models. In particular, we use Hausman et al. (1984) and Wooldridge’s (2005) specifications of Count-­Panel-Data models. While the former is usually advocated as the seminal contribution to these kinds of models, the latter is a straightforward procedure which allows taking into account dynamics without using Generalised Method of Moments (GMM) estimation of the parameters of interest.13 The Poisson panel data model states that the dependent variable, being of a count data type, is distributed as a Poisson, formally the probability for firm i of obtaining y patents at time t is equal to: P( yit ) =

exp(−λit ) λ ityit yit !



(6.1)

where lit can be depicted as the average number of patents firm i gets at time t. In more formal terms, it constitutes the expected value of the Poisson distribution and can be defined as:

λit = E[ yit | X it , δi ] = exp(δi + βj X it )



(6.2)

where yit represents the number of patents filed by firm i at time t, di is the natural logarithm of the unobserved heterogeneity term that takes into account effects changing among firms but constant through time and Xit are the relevant explanatory variables characterising firm i at time t. On the grounds of the empirical specification presented above (equations 6.1 and 6.2), we are now able to test empirically the innovation equation put forward in the theoretical section (see Section 4.1). In particular, we specify the variables of interest in equation 6.2, thus obtaining:14 E[ Filed it ] = exp[β1 ln( Emplit ) + β2 ln( RDxEmplit ) + β3Otherpatit + β4 Filedi ,t −1 + δi ] where • •

•

(6.3)

Filedit is the number of software patents filed at the EPO by firm i in year t. Emplit is the number of employees of firm i in year t. This variable proxies for firm size and it influences the number of software patents filed. In fact, larger firms are likely to have more resources at their disposal in order to apply for more patents.15 RDxEmplit is the R&D intensity of firm i in year t. This has been computed dividing the amount spent on R&D spending over the number of employees.

Software patents and firms’ strategies   149 •

•

Otherpatit has been computed as the number of software patents granted to firms, other than firm i, in year t. This variable is an attempt to understand the influence of strategic factors on the software patenting of firms in the sample. Filedi, t – 1 is the number of software patents filed by firm i in year t –1. This variable wants to give an account of the effect on software patenting decision by the number of software patents filed in the former year.

Despite its large diffusion, the Poisson panel data model presents some important drawbacks that are worth noticing: (i) conditional mean and conditional variance cannot vary independently; and (ii) dynamics are not taken into consideration. The first problem is usually solved by means of the negative binomial model developed by Hausman et al. (1984). Indeed, here the conditional variance is allowed to vary against conditional mean.16 The second complication is addressed by adopting the specification of the model presented by Wooldridge (2005). This specification allows us to implement at most one lag of the dependent variable into the Poisson specification of the econometric model thus taking into consideration, at least partially, dynamic effects in the model. We ran regressions using a negative binomial specification, and this gave us results identical to Poisson specification. Hence, given the possibility of incorporating dynamics into the model only through the Poisson specification we decided to rely on the latter.

5  Data In order to test the general theoretical framework put forward in Section 4, we rely on an original dataset which has been obtained by linking the number of patents applied for by a relevant subset of EU and foreign firms at the European Patent Office. The data collection part is certainly a challenging task to be carried out, given a set of important problems to be solved. The optimal solution to the problem at stake would be to have a representative sample of firms containing information on the number of software patents applied and granted, together with other relevant variables (above all R&D spending). To our knowledge, no dataset of this type is currently available for the EU. For this reason, we relied on a particular procedure for the construction of the dataset which allowed us to arrive as closely as possible to the above mentioned optimal solution. Our first step has been to identify, among all patents filed at the EPO, those protecting a software technology. After revising several methods,17 we decided to rely on the GAUSS database (see Section 5.1) which provided us with all the relevant information18 on software patents filed and granted at the EPO. In order to obtain additional firm level data, we propose a methodology to link the GAUSS database with other datasets – namely the R&D scoreboard, Amadeus and Osiris – by applicant’s name (see Section 5.2).

150   F. Rentocchini and G. De Prato In this way, we were able to build a panel dataset containing information on a sample of 844 firms for the years 2000–2003. Section 5.1 proposes a general description of the GAUSS database, with some relevant statistics concerning European software patents. Then, in Section 5.2, we describe the procedure used to build the sample concerning firms’ patenting strategies. The sample subset of data is presented, underlining certain characteristics that call for the use of specific econometric techniques and providing descriptive statistics for the sample itself. 5.1  GAUSS database and descriptive statistics As mentioned, the present analysis of recent trends in software patenting inside the European Union relies on the Gauss.ffii database, which has been accessed through a Postgres Client allowing the performance of SQL queries. The GAUSS database has been created by a group of developers via multiple sources: it includes the FFII’s (Free Foundation for Information Infrastructure) database of software patents, as well as the Stefan Wagners database of 1,900 business method patents. In addition to those sources and in order to keep it up to date, GAUSS performs continuously searches of patent documents by applicants likely to produce software or business method patents. They also make searches for about 150 words occurring in software patents and, furthermore, searches in European Patent Classification (ECLA) classes with a high probability of containing software patents. The database is constructed as a wiki, meaning that users are not only allowed to add content, but also permitted to edit the content; this revealed to be a very effective way to take advantage of improvements through collaborative efforts. The subset of data relevant to the present work had been built by extracting, from the available dataset of software patents filed between 1982 and the end of 2004, all records of information regarding patents filed between 1 January 1995 and 31 December 2004. A subset had been obtained of 77,540 patent records. 13,207 out of the total number already switched to the granted status by the extraction time (June 2006). Therefore, the subset allowed to track 44 per cent of the overall number of software patents granted by EPO between 1995 and 2004 (the remainder having been filed before the end of 1994). Besides the filing date, the date of publication and date of granting had also been collected where present, depending on the state of each patent. Other relevant pieces of information available are the list of designated countries to which each patent refers, the list of the International Patents Classification (IPC) codes relevant to each patent and the applicant name (or the list of applicants where necessary). Some interesting statistics can be drawn, descriptive of the patenting process by EPO in general, and that of software related products or methods in particular, by analysing the number of designated countries where the patent must be enforced, the number of IPC subclasses, which can be thought as a proxy of patent scope, and the average length of the granting procedure for software patents. The subset of software patents in the period between 1995 and 2004

Software patents and firms’ strategies   151 shows an increasing number of filed patents which were not granted or not yet granted at the time of data collection: while about 60 per cent of patents filed in 1995 switched to the granted state before the end of 2004, 83 per cent of patents filed in 2000 were not yet granted in 2006. This justifies the low share of granted patents included in the mentioned subset, and it is connected to an increase in the time required to complete the granting process, whose average length is of three and a half years: while granted patents in 1997 had been filed about one year earlier, those granted in 2004 took as an average more than five years to complete the granting procedure. It must be mentioned that, getting closer to 2004, the database updating procedure has a relevance in justifying a lower share of granted patents. It is possible to point out, anyway, that in the period 1995–2004 a lower number of patents had been granted against a fast increasing number of filed requests, and in general the granting procedure slowed remarkably. By processing the collected data, it is possible to single out 4,992 different IPC codes in the 13,203 patents granted in the period 1995–2004, to which software related patents referred. About 45 per cent of patents declared multiple IPC codes, with an average of 1.65 IPC codes per patent which could be used as an indicator of patent broadness. While the total number of different IPC codes used strongly increases over time, this is mainly due to the higher volume of patents, as the average number of IPC codes declared per patent per year remains fixed. When referred to the International Standard Industrial Classification of All Economic Activities (ISIC) classification, most referred IPCs belongs to the ISIC Electronics and Computers & Office Machines classes, as reported in the table which lists the ten most common IPC (accounting for about 14 per cent of patents) and corresponding ISIC description (see Table 6.1). Software related filed patents in EPO in the period 1995–2004 refer to 28 different designated countries, to which the validity of patents applies; among them, Germany, the United Kingdom and France each takes more than 10 per cent of patents (see Figure 6.5). Table 6.1 International Patent Classification (IPC) and International Standard Industrial Classification of All Economic Activities (ISIC) concordance for GAUSS patents

 1  2  3  4  5  6  7  8  9 10

Number of patents

IPC

ISIC description

608 432 314 295 294 255 253 219 194 186

H04L29/06 G06F17/30 H04L12/56 G06F1/00 G06F17/60 H04Q7/38 G06F9/46 H04Q11/04 G06F9/44 H04N7/173

Electronics Computers and Office Machines Electronics Computers and Office Machines Computers and Office Machines Electronics Computers and Office Machines Electronics Computers and Office Machines Electronics

152   F. Rentocchini and G. De Prato 18000

16000

14000

Number of patents

12000

10000

8000

6000

4000

2000

0 DE

GB

FR

IT

NL

SE

ES

CH

LI

BE

AT

DK

IE

FI

PT

GR

LU

MC

CY

TR

CZ

EE

BG

SK

SI

HU

RO

PL

Designated country

Figure 6.5  Software patents per designated country (1995–2004).

Overall, European, American and Japanese firms are the relevant subset of firms applying, and obtaining, EPO patents. In particular, firms belonging to the latter two countries own nearly an absolute majority of software patents accorded at the EPO: this means that laws and regulations, such as the one that has been rejected by the European Parliament regarding the ‘Patenting of computer implemented inventions’ (see Section 3), must take into account this fact. Indeed, a legal decision that, all of a sudden, allows patentability of software could threaten the future of the European sector. The fact that foreign firms already own large software patent portfolios could hinder competition. This first-­move advantage is something that must be stressed and it must be taken into consideration by policymakers who wish to extend patentability to the software domain. 5.2  Sample construction and description To investigate the determinants of software patenting at the firm level, a link has been established between the collected dataset and the ‘2004 EU Industrial Research Investment Scoreboard’,19 which lists the research investment20 of the top 500 EU and top 500 non-­EU corporate R&D investors in the period 2000–2003, together with other relevant information. In order to establish proper linking relations, a semi-­automatic data process to match companies to applicants has been performed. A specific small software application has been developed performing automatic matching between firms’ values, requiring explicit operator’s confirmation only in cases in which applicants were not univocally identified. A resulting dataset obtained by linking the information available in the two mentioned sources is composed by 1,000 firms whose data concerning R&D spending, sectoral and geographical classification, number of

Software patents and firms’ strategies   153 software patents published are available for the period 2000–2003. With regard to 2003, about 490 companies are revealed to have filed software patents at EPO. As a last step, missing data on net sales and number of employees had been retrieved by the Amadeus and Osiris databases as well, for the period 2000–2003. Not surprisingly, Aerospace and Defence, Automobiles and Parts, Electronic and Electrical, Pharmaceuticals and Biotechnologies and IT Hardware are sectors where the highest level of R&D takes place. Electronic and Electrical, IT Hardware, Media and Entertainment, Telecommunications Services and Software and Computer Services sectors are found to have the highest average number of software patents filed. In line with results from studies carried out for the US (Bessen and Hunt 2003; Hall 2004), we find that also in the EU software patents are filed by firms belonging to industries other than software. Concerning the IT Hardware sector, it is likely that firms belonging to this industry are patenting software for two main orders of reason: (i) to protect their in-­house developed software which is, most of the time, embedded in the sold hardware; (ii) to strategically patent freely available software developed by the software industry. Sample statistics for the final subset of the dataset are presented. Tables 6.2 and 6.3 show an interesting characteristic of the sample, namely the high number of firms not applying for any patent. More than 50 per cent of firms in the sample do not apply for any patent. This structure of the dataset calls for the implementation of a sound econometric model able to take into account the data’s pattern. In this respect, the choice made of adopting count data models is supported from both the nature of the dependent variable and the structure of the dataset.21 Table 6.2  Sample statistics (number of filed patents equal to zero) Statistics

Research and Development (n =2,462) million euro

Sales (n = 2,478) million euro

Employees (n = 1,944)

Mean Max Min StDev Median

190,1353 6.337 0,24 537,2331 61,485

5.090,197 184.383 0,636 9427 15.404,84 1.188

18.502,72 449.594 14 41.228 5.693

Table 6.3  Sample statistics (number of filed patents greater than zero) Statistics

Research and Development (n = 903) million euro

Sales (n = 909) million euro

Employees (n = 766)

Mean Max Min StDev Median

668,5164 6.782 4,82 1.041,226 228,85

12.124,74 142.254 2 19.610,55 4.724

48.606,71 477.100 1 68.693,61 23.345

154   F. Rentocchini and G. De Prato

6  Results Tables 6.4 and 6.5 show results for the panel data Poisson regression model with random effects. Interesting results are obtained for the software and computer service sector; here R&D spending is not significantly related to the number of software patents firm applies for. This result is likely to confirm the fact that, in this sector, a patent is not considered a useful appropriability instrument of the results of R&D process. In fact, the absence of a significant relationship between the two variables supports our belief that R&D contributes to the creation of knowledge capital and innovation but that software patents do not proxy well the innovation output, meaning that they are not deemed as suitable appropriability measures for firms operating in this sector. On the other side, the number of employees seems to play an important role for software patenting at the firm level, pointing out the importance of the presence of a legal department handling intellectual property rights (IPRs) (Lerner 1995). Indeed, larger firms are more likely to benefit from decreasing average costs, thus exploiting better the resulting economies of scale, mainly thanks to the rich endowment of financial resources devoted to IPRs managing departments. The last interesting result coming from our analysis refers to the significance of the variable proxying, at least partially, the role of strategic factors. The number of software patents granted to firms other than the Table 6.4 Software patenting propensity estimates: software and computer services sector (FTSE 97) 2000–2003 Variable

Coefficient (Std. Err.)

Filed_1

0.014 (0.011) 0.913 (0.573) 0.955** (0.272) 0.260*** (0.146) –4.676*** (2.530) 240 –302,262 13,286 13,286

R&D per Employee Employees OtherPat Intercept N Log-likelihood χ2

Note Panel robust standard errors by bootstrap are shown in parentheses. The method of estimation is maximum likelihood for the Poisson model (conditional maximum likelihood). The χ2 is a Wald test for the joint significance of all the explanatory variables taken together. *, ** and *** denote statistical significance at the 1, 5 and the 10% test level.

Software patents and firms’ strategies   155 Table 6.5: Software patenting propensity estimates: information technology hardware (FTSE 93) 2000–2003 Variable

Coefficient (Std. Err.)

Filed_1

0.007** (0.002) 0.641* (0.320) 0.617* (0.266) 0.000 (0.223) –1.961 (2.545) 389 –939,341 27,133 27,133

R&D per Employee Employees OtherPat Intercept N Log-likelihood χ2

Note Panel robust standard errors by bootstrap are shown in parentheses. The method of estimation is maximum likelihood for the Poisson model (conditional maximum likelihood). The χ2 is a Wald test for the joint significance of all the explanatory variables taken together. *, ** and *** denote statistical significance at the 1, 5 and the 10% test level.

firm applying for software patents is significant and positive. This means that the amount of software patents granted in the same year to other firms contributes to the explanation of the number of software patents firm i applies for. We interpret this result as a sign of strong strategic factors inside the software sector. Firms in this sector are not likely to patent software to appropriate results of the R&D process; but, at the same time, they are eager to patent if they feel the threat of other firms. This threat effect is intrinsically due to the nature of the software technology, that is of a cumulative type. An increase in the amount of software patents accorded to neighbour firms can hinder future development of software by the present company spurring it to apply for patents as a strategy of defence. Results for the IT hardware sector differ considerably and are reported in Table 6.5. No relationship can be identified between Otherpatit variable and the number of software patents meaning that no threat effect by other hardware firms is present. Nevertheless, firm size is still significant even if the coefficient is lower than the one found in software sector. On the contrary, R&D spending (RDxEmplit)22 and number of patents filed by the firm in the former year (Filedit) have become significant factors affecting the number of patents a firm applies for in the current year, thus pointing out a trend, characterising firms in the hardware industry, to patent software in sequence.

156   F. Rentocchini and G. De Prato

7  Discussion and conclusion The main goal of our work was to give a preliminary account of a phenomenon that has been disregarded by the scientific literature so far, namely software patenting in the European Union. As we saw, the fact that the European Patent Convention expressly prohibits software patenting has not been a major problem for firms and inventors who have patented software as well. General statistics show the increased overall number of patents accorded by the European Patent Office (EPO) starting from the second half of the 1990s. The overall upsurge has been mainly driven by patents accorded in ICTs and biotechnology fields. Software patents are surely an important part of the former: a rough indication of this could be deduced by the number of patents falling in particular IPC classes (see Table 6.1 and Figure 6.6). A second main goal of our work was to find relevant factors explaining software patenting by firms at the EPO. First of all, a reliable database of software patents is presented. These patents are accorded by the European Patent Office and, to our knowledge, more than 40,000 software patents have been issued up to 2004 to European and non-­European firms. In this respect, a large part of them has been accorded to American and Japanese firms. The fact that nearly the majority of granted patents belong to foreign companies must be due to the greater experience that these firms have acquired dealing with their own patent system. For example, in the US, software has been subject to patentability for a long time; this means that firms have more expertise both in dealing with the application procedures and in identifying the more valuable inventions to be patented. Then, the knowledge production function approach is adopted to understand factors affecting the output of the innovation process at the firm level. The model has been extended to incorporate factors deemed important to explain recent patenting strategies (strategic factors, firm size and technological opportunities) and to deal with our 2500

Number of patents

2000

1500

1000

500

0 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

Year g06-Computing; calculating; counting

h04-Electric communication technique

Figure 6.6  Patents accorded by the EPO to EU15 countries – software patents.

Software patents and firms’ strategies   157 specific interests (namely software and hardware sectors). The way in which the dataset has been built and the method of estimation are then presented, putting forward also some interesting findings (average length of software patents’ granting procedure, sectors applying for the majority of software patents, ownership). Finally, results of the chosen econometric model are presented. The fact that patents are not deemed as useful appropriability instruments by firms belonging to the software sector and the presence of a growing threat effect by other firms are the main outcomes of the analysis. We are aware of limitations to the present work. First of all, the reliability of the GAUSS database must still be tested in a systematic way. Second, the EU Scoreboard, on which we relied to link firms’ characteristics with the number of software patents filed, allows us to take into consideration the behaviour of large firms only. Small and Medium Enterprises (SMEs) together with their software patenting strategies are totally disregarded. Third, it has not been possibile to separate R&D expenditure used in software production from that used for other purposes. Fourth, our data sources do not reveal the number of engineers and programmers. Nevertheless, to our knowledge, our work is one of the first attempts to give an account of the existence, determinants and direction of a diffused phenomenon such as software patenting. The above mentioned deficiencies will be corrected in future works in order to deepen the understanding of the topic.

Notes   1 The authors gratefully acknowledge comments and suggestions from the editors of the book and the participants to the 11th conference of the International Joseph A. Schumpeter Society which took place in Nice/Sophia Antipolis, France during 21–24 June 2006. The usual caveat applies. Francesco Rentocchini gratefully acknowledges the financial support from the Autonomous Province of Trento (Grant 2006: TRACKs).   2 It seems useful to recall that the distinction between filed and granted patents is connected to the different condition of patents’ requests at different steps in the patenting process. In order to obtain a patent, an application must be filed with the authorised body (the relevant patent office) with all the necessary documents and fees; the patent office will then conduct an examination to decide whether to grant or reject the application. In order to award patents, EPO follows the “first to file” principle, while USPTO acts on a “first to invent” basis. If granted the patent provides the applicant with a temporary right to prevent unauthorised use of the technology outlined in the patent itself, and only granted patents are effective and enforceable against infringement (OECD 2006).   3 Propensity to patent is proxied by the number of patents applied for over the R&D expenditure (Scherer 1983).   4 This is mainly due to the decision by WIPO of assigning the granting procedure of international patents to the EPO, contributing to the increase of the average examination period. In fact, substantive examination of international patents is put before European ones. What happens is that the procedure for European applications is postponed, thus increasing the length of the average granting procedure period.   5 Article 79 of the EPC defines designation as the indication of the contracting state(s) in which the protection for the invention is desired. Therefore, designated countries are those countries in which patent applicants wish to protect their invention.

158   F. Rentocchini and G. De Prato Applicants to the EPO have indeed to designate specific countries, even if since January 2004 all international applications filed designate by default all Patent Cooperation Treaty (PCT) contracting countries bound by the PCT treaty (the designable countries) as of the filing date. However, patents have to be validated in the designation country for it to be effective, as designation of a country does not automatically provide patent rights in that country (OECD 2006).   6 Another factor related to this issue is the role played by countries where production has been increasingly outsourced.   7 This sector has recently witnessed to the implementation of two standard setting procedures (GSM and UMTS). The implementation of a standard setting can be quite problematic where firms participating in the standard hide relevant intellectual property rights assets and decide to reveal these once other firms taking part in the procedure have already implemented complementary investments.   8 While presenting the Directive, European commissioner Charlie McCreevy stated that it would have been important to approve it because of the necessity, for the EU, of higher certainty surrounding the topic. Moreover, he stated that more than 7,000 software patents had already been approved.   9 One of the reasons that has prevented the directive being implemented is certainly the discussion on the presence of a ‘technical’ content for the invention to be patented. 10 The seminal contribution to the KPF approach is from Pakes and Griliches (1984). They present a simplified path analysis of the overall KPF model. There, R&D expenditure contributes to the formation of an unobservable variable k, the technological knowledge of the firm, which could be proxied by the number of patents filed. 11 This relationship is even more important nowadays. In fact, the R&D process often takes place inside formally implemented R&D labs which is a sign of the direct link between the amount of resources invested by the firm and its knowledge competence. 12 An inventor, deprived of the exclusive right to exploit his invention for a definite period of time, would not have even started the inventive act if he was aware of it. This is obviously related to the public nature of knowledge (Arrow 1962). 13 Models using GMM estimation to analyse patents filed in a Count-­Panel-Data setting are quite recent and still a work in progress. For a survey see Cameron and Trivedi (1998). 14 Once the variables have been specified, the following step has been to substitute equation 6.3 in equation 6.1, in this way obtaining the probability for firm i of getting y patents at time t. The final step has been to derive the formula for the conditional log-­ likelihood and maximise it, thus obtaining the estimates of the coefficients of interest (for details on the procedure see Cameron and Trivedi (1998)). 15 This is even more likely to happen in the EU where the average cost of a patent is higher than in other patent systems (Malerba and Montobbio 2002). 16 Assuming that the heterogeneity term is distributed as a gamma function, conditional variance is then allowed to vary:

Var ( yit) = λ it (1 + θλ it ) 17 For an in-­depth discussion on the different methods available for the identification of software patents, and the relative advantages and disadvantages, see Rentocchini (2008). 18 It has been possible to extract, for software patents filed and granted by the EPO, data referring to the patent’s number, the International Patent Classification code, the applicant’s name, the inventor’s name and other relevant information. 19 Produced as a part of the ‘Investing in research: an Action Plan for Europe COM (2003) 226 – EC DG Joint Research Centre’. 20 The figure is based on annual audited reports calculated at the consolidated group level.

Software patents and firms’ strategies   159 21 Poisson and negative binomial specifications allow for excess zeros. But, when the zero value originates from a separate decision process other than the one analysed, a zero-­inflated model should be preferred (Lambert 1992). The problem remains that zero-­inflated models have not been developed for panel data yet. 22 Results are likely to be biased upward due to the difficulty of separating the share of R&D expenditures devoted to software development and the share used for other purposes.

References Arrow K. (1962), Economic welfare and the allocation of resources for invention, in Nelson R. (eds) The Rate and Direction of Inventive Activity: Economic and Social Factors, NBER, Princeton NJ, Princeton University Press. Bessen J. and Hunt R. (2003), The software patent experiment, in Patents, Innovation and Economic Performance, OECD Conference Proceedings, Paris, OECD. Calderini M. and Scellato G. (2004), Intellectual property rights as strategic assets: the case of European patent opposition in the telecommunication industry, Discussion Paper 158, CESPRI, Centre for Research on Innovation and Internationalisation, Università Bocconi, Milano, Italy, mimeo. Cameron C. and Trivedi P. (1998), Regression Analysis of Count Data, Cambridge, Cambridge University Press. Cohen W., Nelson R. and Walsh J. (2000), Protecting their Intellectual Assets: Appropriability Conditions and Why U.S. Manufacturing Firms Patent (Or Not), NBER Working Paper 7552, online, available at: http://ideas.repec.org/p/nbr/ nberwo/7552.html. Graham S. and Mowery D. (2004), Submarines in software? Continuations in US software patenting in the 1980s and 1990s, Economics of Innovation and New Technology, 13: 443–56. Hall B. (2004), Exploring the patent explosion, Journal of Technology Transfer, 30: 35–48. Hall B. and Ziedonis R.H. (2001), The patent paradox revisited: an empirical study of patenting in the U.S. semiconductor industry, 1979–1995, RAND Journal of Economics, 32: 101–28. Hausman J., Hall B. and Griliches Z. (1984), Econometric models for count data with an application to patents-­R&D relationship, Econometrica, 52: 909–38. Kim J. and Marschke G. (2004), Accounting for the recent surge in U.S. patenting: changes in R&D expenditures, patent yields, and the high tech sector, Economics of Innovation and New Technology, 13: 543–58. Kortum, S. and Lerner J. (1999), What is behind the recent surge in patenting?, Research Policy, 28: 1–22. Lambert D. (1992), Zero-­inflated Poisson regression, with an application to defects in manufacturing, Technometrics, 34: 1–14. Lerner J. (1995), Patenting in the shadow of competitors, Journal of Law & Economics, 38: 463–95. Malerba F. and Montobbio F. (2002), L’Europa dell’alta tecnologia: lo scenario internazionale in alcuni settori chiave, in Tendenze dell’Industria Italiana, Confindustria, Centro Studi, pp. 63–78, online, available at: www.confindustria.it/ indcong.nsf/e5e343e6b316 e614412565c5004180c2/0a3e04f32e8e7cf9c1256c3a0038589d/$FILE/settori%20 produttivi%202.2.pdf. Mansfield E. (1986), Patents and innovation: an empirical study, Management Science, 32: 173–81.

160   F. Rentocchini and G. De Prato Mazzoleni R. and Nelson R. (1998), The benefits and costs of strong patent protection: a contribution to the current debate, Research Policy, 27: 273–84. Merges R. and Nelson R. (1990), On the complex economics of patent scope, Columbia Law Review, 90: 839–916. Nelson R. and Winter S. (1982), An Evolutionary Theory of Economic Change, Cambridge MA, The Belknap Press of Harvard University Press. Nordhaus W. (1969), Invention, Growth and Welfare, Cambridge MA, MIT Press. OECD (2004), Patents and Innovation: Trends and Policy Challenges, Discussion paper, Paris, OECD. OECD (2006), Glossary of Patent terminology, Economic Analysis and Statistics Division, Directorate for Science, Technology and Industry, Paris, OECD. Pakes A. and Griliches Z. (1984), Patents and R&D at the firm level: a first look, in Griliches Z. (ed.), R&D, Patents and Productivity, Chicago, University of Chicago Press. Rentocchini F. (2008), Sources and characteristics of software patents in the European Union: some empirical considerations, SSRN eLibrary, online, available at: http://ssrn. com/abstract=1141714. Scherer F. (1983), The propensity to patent, International Journal of Industrial Organization, 1: 107–28. Schumpeter J. (1942), Capitalism, Socialism and Democracy, New York, Harper & Row. Wooldridge J. (2005), Simple solutions to the initial conditions problem in dynamic, nonlinear panel data models with unobserved heterogeneity, Journal of Applied Econometrics, 20: 39–54.

Part II

Fragmentation and internationalisation of firms and of local systems of production

7 Does spatial proximity matter? Micro-­evidence from Italy1 Giulio Cainelli and Claudio Lupi

1  Introduction Endogenous growth theories (Romer 1986, 1990; Lucas 1988; Grossman and Helpman 1991) emphasise the role of knowledge spillovers for enhancing technological change and long-­term economic growth. A more recent strand of literature suggests that technological spillovers not only generate externalities, thereby fostering economic growth, but also tend to be spatially bounded (Jaffe et al. 1993; Audretsch and Feldman 1996). Indeed, spatial proximity, stimulating face-­to-face interactions between economic agents (firms and individuals), can facilitate the spread of ideas, information (on new products, production pro­ cesses and markets), and different types of knowledge (codified and tacit, public and private) within a locality (Storper and Venables 2004), thus causing differences not only in firm-­level employment and productivity, but also in growth rates at the local level. As Glaeser et al. (1992:  1127) correctly point out “the cramming of individuals, occupations, and industries into close quarters provides an environment in which ideas flow quickly from person to person”. These “new” insights, largely inspired by the endogenous growth literature, in the early 1990s stimulated some authors such as Glaeser et al. (1992) and Henderson et al. (1995) to empirically investigate the role and the impact of different forms of spatial externalities on measures of subsequent local economic growth. The idea behind these works was that the more a geographic area is specialised in a given local industry (localisation economies), and the more diversified is the economic and productive environment around it (urbanisation economies), the higher will be the subsequent economic growth of that industry in that geographic area: i.e. the higher will be local economic development. These two seminal papers, based on US city data, originated a new stream of empirical work aimed at assessing the impact of geographic agglomeration of productive activities on local employment and firm productivity growth based on information from several countries. The results of this work, which has generally used exogenously defined geographic units, such as “functional” (Standard Metropolitan Areas (SMAs), Local Labour Systems (LLSs) and so on), or “administrative” (regions or provinces) areas, are puzzling and inconclusive. In fact, despite the large number of papers published on this topic, which use data from

164   G. Cainelli and C. Lupi different countries and regions, this stream of literature has not provided definite and universally accepted conclusions about the role and the impact of agglomeration forces on local economic growth and firms’ economic performance. This chapter is an attempt to contribute to this literature, adopting a partially “new” research direction. Using a panel data set of 23,374 Italian manufacturing firms for the period 1998–2001, drawn from AIDA, a commercial database collected by Bureau Van Dijck, and estimating a flexible Bayesian model, this chapter studies the effect of spatial agglomeration economies (i.e. localisation and urbanisation economies) on employment growth over discrete distances (in contrast to the previous literature, which considers pre-­defined geographic units). This work makes three contributions to the empirical literature on agglomeration economies. First, by calculating precise distances between firms for each unit in our large data set, we can compute agglomeration economies at different distances. This is accomplished by using available Geographic Information System (GIS) location coordinates for each firm in our sample. It is worth noting that this approach to calculating agglomeration economies at firm level is novel within this stream of literature (a similar methodological approach was developed by Wallsten (2001) and by Rosenthal and Strange (2008)). Second, using these agglomeration variables we can study the impact of these measures on firms’ employment growth. Thus, by using measures of agglomeration economies over actual distances between firms, we can empirically identify the rate at which knowledge spillovers attenuate over space. Finally, we estimate a flexible Bayesian model, which allows us, on one hand, to build an extremely flexible random coefficient model in a rather natural way and, on the other hand, to derive estimates for the full exact finite-­sample posterior distributions of the parameters. This model also allows us to address potential non-­normality problems. The remainder of the chapter is organised as follows. In Section 2 we briefly survey the related literature. In Section 3 we describe the data set and the empirical framework used in our analysis. Section 4 presents the statistical model and discusses the empirical results. Section 5 concludes.

2  Related literature In the early 1990s, the relationships between spatial agglomeration, knowledge spillovers and economic growth at the urban level were extensively investigated in several seminal papers (Glaeser et al. 1992; Henderson et al. 1995). Glaeser et al. (1992), using a cross-­section of US cities, analysed the impact of three different forms of local knowledge spillovers – Marshall–Arrow–Romer (MAR), Porter and Jacobs externalities – on subsequent urban employment growth. They show in their paper that localisation economies (also known as MAR economies), arising from the spatial concentration of firms belonging to the same industry, and captured by specialisation indicators, have a negative impact on urban economic growth, while urbanisation (or Jacobs) economies, spurred by the variety and diversity of geographically proximate industries, positively affect the subsequent growth of a metropolitan area.

Does spatial proximity matter?   165 Using a similar empirical framework, Henderson et al. (1995) find that localisation plays a positive role in mature capital-­goods sectors, while differentiation of the productive structure (variety), which should generate cross-­fertilisation of ideas between different industries, has a positive impact only in the case of high tech industries. Using French data, Combes (2000) also finds a rather negative impact of specialisation on employment growth in both the industry and service sectors. Finally, Forni and Paba (2002), using information on a cross-­section of 995 Italian LLSs for the period 1971–1991 find that in most cases specialisation and variety positively affect growth, but the variety is different for each industry. Moreover, they note that, consistent with Marshall (1920), in order to capture the spillover-­generating process a size effect needs to be added to the specialisation effect. Glaeser et al.’s (1992) approach has been replicated in the contexts of different countries in order to provide further evidence on these issues. Nonetheless, the various results obtained from empirical research in this field are controversial such that currently there is not a unique model explaining the link between employment growth and the structure of the local economy. In particular, some studies referring to the Italian case find that specialisation has a negative impact on local growth, while diversity plays a positive role (see, among others, Cainelli and Leoncini 1999; Cainelli et al. 2001; Cunat and Peri 2001; Usai and Paci 2003; Paci and Usai 2006; Mameli et al. 2007). This empirical literature was extended by several studies that analyse the impact of measures of agglomeration economies both on employment growth (as in the original body of literature referred to), and on productivity or firms’ total factor productivity (TFP) growth (de Lucio et al. 2002; Henderson 2003; Cingano and Schivardi 2004; Martin et al. 2008). The findings within this new strand of empirical research are also rather puzzling. For example, de Lucio et al. (2002) investigating the relationship between labour productivity and spatial agglomeration at the level of the 50 Spanish provinces for the period 1978–1992, find that variety plays a role in labour productivity growth, and find a U-­shaped effect for specialisation. According to their results, low levels of specialisation reduce productivity growth, and high levels foster it. In contrast, Cingano and Schivardi (2004), using firm-­level based TFP indicators, show that specialisation, calculated at the level of the 784 Italian LLSs, has a positive impact on firm productivity growth, but that variety has no significant effect. Taking local employment growth as the dependent variable, Cingano and Schivardi (2004) show that the specialisation effect is reversed and becomes negative, while variety has a significant and positive impact on employment growth, thus confirming Glaeser et al.’s results. On the other hand, Henderson (2003), using the Longitudinal Research Database (LRD) of the US Census Bureau, finds that localisation economies have strong positive effects on productivity at plant level in high tech industries, but not in machinery industries, and finds little evidence of urbanisation economies. Finally, Martin et al. (2008), using French individual firm data from 1996 to 2004, find no significant effect of spatial agglomeration on firm productivity. More precisely, they find that French firms benefit from

166   G. Cainelli and C. Lupi localisation, but not from urbanisation economies. However, the benefits from industrial clustering – even if highly significant from a statistical point of view – are quite modest in terms of magnitude. The use of TFP measures is an obvious and notable improvement to these studies which, however, must also acknowledge some of the drawbacks related to other measurement and empirical issues: for example, the use of sample of plant data (Henderson 2003; Martin et al. 2008) and the problems in the case of Cingano and Schivardi’s paper of sample selection. Another major shortcoming of all these studies is that they refer to exogenously defined geographic units such as SMAs, LLSs, or administrative regions or provinces. A consequence of this choice is that all these studies find it difficult to deal with a rather relevant aspect of these phenomena: namely, the attenuation of agglomeration economies over space. In the recent literature two different approaches were proposed to deal with this issue. First, Desmet and Fafchamps (2005), using US county data for 1972 and 2000, try to overcome it by assuming that a county’s employment growth is not only affected by the county under consideration, but also by all “near” counties. Van Oort (2007) also tries to tackle this issue by considering spatial dependence, though his study has some problems with the robustness of his results. A second approach was developed by Wallsten (2001). Calculating by means of a GIS program the distance between each firm-­pair, he investigates spatial spillovers at the firm level over discrete distances. Finally, Rosenthal and Strange (2008), using 2000 census data to estimate the relationship of agglomeration and proximity to human capital to wages and taking a geographic approach, find that the benefit of spatial concentration tends to attenuate sharply with distance.

3  Data The data source used in this chapter is AIDA, a commercial database collected by Bureau Van Dijck. This large data set of Italian joint stock companies reports balance sheet data such as sales, number of employees, labour costs, etc., as well as the specific sector of activity.2 In addition, it reports firms’ street addresses, information that is very useful for this study. Using these data, we built a balanced panel data set initially composed of 24,089 Italian manufacturing firms for the period 1998–2001. The firms are located across 18 Italian regions. Firms in Sicily and Sardinia were excluded from our investigation because of the insularity of these regions.3 One novelty of this work is that we are able to measure agglomeration economies over space, exploiting information on the actual distance between each pair of firms in the sample. The data on street addresses allow us to recover each firm’s exact longitude and latitude coordinates. Then, using these coordinates – available for all the firms in our sample – we calculate, by means of a Geographical Information System (GIS) program, the actual distance (in kilometres) between each pair of firms (Wallsten 2001). Finally, for each company, we compute the number of other firms located within different actual distances: that

Does spatial proximity matter?   167 is, from 0 to 2 kilometres, from 2 to 10 kilometres, from 10 to 30 kilometres. This allows us to perform a first qualitative analysis, the results of which are presented in Table 7.1. The choice of these distances was not arbitrary but the result of in-­depth empirical analysis, which considered different actual distances ranging from 200 metres to 75 kilometres. Among these distances we chose only three spatial ranges in order to reduce the number of coefficients of the agglomerative variables to be estimated. Table 7.1 shows the distribution of sample firms by distance and industry. From this table it emerges that there is tendency for Italian manufacturing firms to locate close to one another.4 In fact, about 16 per cent of firms belonging to the same industry are located within 10 kilometres, and about 30 per cent are located at distances of 30 kilometres or less (see the appendix, Table 7.7, for key sections). It is also interesting that this tendency towards spatial agglomeration seems to vary according to industry. We find that textiles and leather are – as expected – particularly spatially agglomerated at least if distances from 0 to 2 kilometres are considered. These traditional industries are dominated by the presence of industrial clusters and local systems of small- and medium-­sized firms. The strong agglomeration of industries such as coke and refined petroleum and chemicals, on the other hand, may seem surprising. However, the spatial location of firms belonging to these industries is strongly conditioned by the availability within a given geographic area of specific natural resources, infrastructures or technological competences. It is useful to remember that according to Istat 1991 industrial districts definition, four clusters are active in these sectors which employ more than 60,000 workers. We use these data to calculate two different types of measures of agglomeration economies: (i) an indicator of localisation economies at different distances, and (ii) an indicator of urbanisation (or variety) at the same distances. Table 7.1  Distribution of firms by distance and industry Sectors

0–2 km

2–10 km

10–30 km

DA DB DC DD DE DF DG DH DI DJ DK DL DM DN All sectors

3.92 7.87 7.89 5.80 9.07 13.00 9.68 5.36 4.96 4.90 5.53 6.98 6.63 5.61 6.32

9.23 17.58 15.37 12.48 23.63 18.00 24.25 14.80 11.70 14.25 15.63 20.74 14.83 13.09 16.15

20.10 30.40 27.58 26.71 35.93 28.00 39.14 32.25 22.18 29.97 31.17 35.88 26.70 26.59 30.16

168   G. Cainelli and C. Lupi The variable used to measure localisation (or MAR) economies is calculated as: L(i ,ds) = nsd

(7.1)

where n(d) s is the number of other firms belonging to the same industry s, located within distance d. This is an indicator of the abundance of firms belonging to the same industry in a given area. As already noted, this indicator is calculated over different actual distances d measured in kilometres: that is, from 0 to 2, from 2 to 10 kilometres and so on. In order to measure urbanisation economies, the Herfindahl index is often used in the literature (Glaeser et al. 1992; Cainelli and Leoncini 1999). However, the Herfindahl (or Herfindahl-­Hirschman) index is more properly an index of concentration. Here, the variable Vi,(d)s which we use to measure urbanisation (or Jacobs) economies – that is, variety of the local productive structure – is calculated for a generic firm i belonging to industry s using Shannon’s entropy index (Shannon 1948). Since we want to measure the “richness” of sectors other than the i-­th firm’s sector, in the computation we exclude the i-­th firm’s industry s. The number of firms in the same sector is already accounted for by Li,(d)s . Shannon’s entropy index, also known as the Shannon-­Wiener index is an index of diversity. For this reason, it has been widely used to measure biodiversity (i.e. the richness of species) in ecological studies. One important reason why ecologists became interested in the Shannon-­Wiener index is that it takes account of both the number and the evenness of species that are present in a given environment (see e.g. Ricotta and Szeidl 2006). This is precisely what we want to measure with reference to the “richness” of the economic environment in terms of the number and diversity of firms present in a given area. In other words, we want to measure the variety of “species” of firms (i.e. the presence of firms belonging to different sectors) and the relative abundance of firms in the different sectors in selected districts. Let S denote the number of sectors and N the  total number of firms in the area. nS is the number of firms in each sector s ∈ [1, . . ., S]. Of course in each area the number of firms, N, is N =

∑

S s =1

ns .

We define also with pS the proportion of firms in sector s to the total number of firms in the area, pS = nS/N. Then the Shannon-­Wiener index is defined as H′ = −

S

∑ p ln( p ). s

s

(7.2)

s =1

It can be proved that H9 is maximised when each sector is represented by an equal number of firms. It can also be shown that in this case Hmax = ln(S)0

(7.3)

Therefore it is also possible to compute a relative index, sometimes labeled “equitability”, by considering H9/Hmax. This is the version of the index that we use in our analysis.

Does spatial proximity matter?   169 In order to avoid some evident outlying observations, we did a very mild trimming (0.5 per cent on both tails of the distributions) sequentially on employment growth (our dependent variable) and on production growth and labour cost per employee growth, respectively. This left us with 23,374 valid observations. In order to offer a visual summary of the main variables involved in the analysis, their estimated densities are plotted in Figure 7.1. In this chapter we want explicitly to study the properties of the sample at hand. We do not intend to draw inferences that are valid for the whole population of Italian firms. Indeed, a potential problem with these kinds of samples is that firms are not randomly chosen (Cingano and Schivardi 2004). However, comparisons with the whole population in terms of frequency distribution by sector (Table 7.2), and geographical areas (Table 7.3) show that the structure of our sample generally accords well with the census data. The only (potential) selection problems are that the average size of firms in our sample is generally bigger than in the reference population, and that southern firms tend to be slightly under-­represented. The original 14 2-digit sectors were reduced to 11 by aggregating sectors DB and DC (manufacture of textiles and textile products and manufacture of leather and leather products), DD and DN (manufacture of wood and wood products and manufacturing not elsewhere classified, including furnishings), and DF and DG (manufacture of coke, refined petroleum products and nuclear fuel and manufacture of chemicals, chemical products and man-­made fibres). Definitions of the classifications and other variables may be found in the appendix to this chapter (Table 7.7).

Figure 7.1  Estimated densities of the main variables.

170   G. Cainelli and C. Lupi Table 7.2  Distribution of firms and employees by sector – data refer to joint stock firms Sectors

DA DB DC DD DE DF DG DH DI DJ DK DL DM DN

Sample

Census 2001

Firms %

Employees %

7.67 13.47 3.95 2.37 6.83 0.42 5.05 5.55 5.17 18.06 14.48 9.50 2.34 5.13

6.00 9.71 2.14 1.23 7.04 0.50 12.43 5.05 5.08 16.05 14.95 11.58 5.56 2.68

█

Firms %

Employees %

6.72 12.33 4.34 2.78 8.56 0.27 3.26 5.18 5.17 17.80 13.94 9.78 2.26 7.61

6.59 10.87 3.53 1.68 5.47 0.71 6.08 5.36 5.21 15.32 15.37 10.61 7.83 5.36

4  The statistical model and the empirical results We tested the effects of agglomeration economies on firms’ employment growth. In the analysis we adopted a fairly standard “long differences” specification, that is, time differences computed over more than one year (usually at least three to five years). Long differences are commonly used in the literature to eliminate region-­specific effects (similar to specifications in differences which eliminate fixed effects) and to capture medium- to long-­run relationships between the variables of interest. Long differences partially protect from business cycle and other short-­term fluctuations. Examples of models that use long differences in various fields of application are offered in Holtz-­Eakin and Schwartz (1995), Boarnet (1998), Picci (1999) and Brynjolfsson and Hitt (2003), among others. In our investigation we use the longest difference our data allow. Table 7.3 Distribution of firms and employees by geographical area – data refer to joint stock firms Area

North-West North-East Centre South

Sample

Census 2001

Firms %

Employees %

46.40 32.51 14.80   6.28

52.05 29.37 14.37   4.21

█

Firms %

Employees %

38.56 26.11 19.24 16.10

46.63 27.78 15.31 10.28

Does spatial proximity matter?   171 More specifically, our initial model is parameterized as: ∆ 3 ln(ei ) = β1∆ 3 ln( yi ) + β 2 ∆ 3 ln( wi ) +

D

∑δ d =1

1, d

L(i d ) +

D

∑δ d =1

V ( d ) + ξi

2,d i

(7.4)

where D3 is such that D3zt = zt – zt–3, ei is employment, yi is real output and wi denotes real wage per employee. In addition, b1 and b2 are the elasticities with respect to output and wages, while L(d) and V (d) denote the localisation and i i urbanisation variables, respectively. In a preliminary stage of our investigation we estimated model (7.4) relating employment growth to changes in wages and production, agglomeration variables, as well as size, geographical and industry dummies. The model gave interesting results but was open to three major objections. First, it is reasonable to think that industry effects cannot be fully controlled using only intercept dummies. Second, the model’s residuals, rather than being normal, were approximately distributed according to a Student-­t random variable with about three degrees of freedom. Third, we observed that residuals variance was not constant across firm-­size classes. We decided therefore to build a more sophisticated model that could explicitly address all these points. Following attempts to work out a model that could cope with these problems, we decided to work within the Bayesian framework. This choice was made for three reasons. First, Bayesian methods allow us to build an extremely flexible random coefficients model in a rather natural way; second, they allow us to derive estimates of the full, exact finite-­sample posterior distributions of the parameters,5 rather than simple point estimates; third, they enable us to address potential non-­normality in a straightforward way. Of course, this extra-­flexibility comes at the cost of a significant increase in the computational burden. The “classical” Gaussian linear model assumes that, conditionally upon the explanatory variables X, the dependent variable y is normally distributed with mean Xb. This means that one can write yi/X ≈ N(x9i  b, s2), with x9i the i-­th row of the data matrix X. However, in many empirical applications observations are not necessarily conditionally normally distributed. Indeed, as already highlighted, our preliminary results seem to suggest that, as far as our data are concerned, employment growth might be conditionally t-­distributed. Therefore, a reasonable model for our data should assume yi/X ≈ N(x9i  b, v), where the parameter n indicates the degrees of freedom of the t distribution. Apart from our preliminary evidence, modelling based on Student-­t distribution has two practical advantages. First, using a t distribution allows us to obtain a model that is more robust to outlying observations and helps to cope with heteroschedasticity (see e.g. Gelman et al. 2004; Geweke 1993). Second, it should be emphasised that if the degrees of freedom of parameter n of the t distribution is estimated within the model (as in our case), then the model encompasses a Gaussian distribution. In fact, the model is sufficiently flexible to accommodate a Gaussian distribution, should this be the “true” conditional distribution of the data. In this case, the estimated value of v, vˆ, will be large.

172   G. Cainelli and C. Lupi Turning to our empirical model, let the (n × k) matrix X denote the explanatory variables so that the i-­th row of X, relative to the i-­th firm, is xi' = ( D 3 log ( wi ) ) , D 3 log ( yi ) , di'

(7.5)

where d9i is the row-­vector containing the firm-­specific agglomeration variables and geographical dummies. Specifically, for the i-­th firm belonging to sector s and size class j we assume: ∆ 3 ln(ei ) | X ~ t ( µi , τj , νj)

µi = b1,s , j + b2,s , j ∆3 ln( wi ) + b3,s , j ∆3 ln( yi ) +

+c1 L1i + c2 L 2i + c3 L3i + d1V 1i + d 2V 2i + d3V 3i +

(7.6)

(7.7)

+e1G1i + e2G 2i + e3G3i. Here, D3ln(ei), D3ln(wi), and D3ln(yi) indicate the 1998–2001 growth in employment, labour cost per employee and production, respectively. L1i, . . ., L3i and V1i, . . ., V3i are the localisation and variety variables for various distances.6 G1i, . . ., G3i indicate geographical dummies for the north-­western, north-­eastern and southern regions, respectively.7 Note that we do not assume normality a priori as we would have for the Gaussian linear model. Furthermore, note that some parameters are allowed to vary with the sector and/or the size of the firm. In particular, the precision parameter,8 tj, and the degrees of freedom parameter, nj, of the t distribution are assumed to vary with the size of the firm and we found evidence of this in our initial linear model. Note also that the degrees of freedom parameter is left unspecified and is estimated from the data. However, the b parameters, which are related to technology and institutional factors, are assumed to vary with both sector and firm size. Table 7.4  Definitions of the variables used in the empirical analysis Variables Definition D3ln(ei) D3ln(yi) D3ln(wi) L1i L2i L3i V1i V2i V3i G1i G2i G3i

1998–2001 growth of employment (dependent variable) 1998–2001 growth of production 1998–2001 growth of labour cost per employee Centered log of number of firms of the same sector located within 2 kilometres Centered log of number of firms of the same sector located between 2 and 10 kilometres Centered log of number of firms of the same sector located between 10 and 30 kilometres Centered Shannon-Wiener index computed between 0 and 2 kilometres Centered Shannon-Wiener index computed between 2 and 10 kilometres Centered Shannon-Wiener index computed between 10 and 30 kilometres Firm located in the North-West Firm located in the North-East Firm located in the South

Does spatial proximity matter?   173 Given that our model is specified and estimated in a Bayesian context, model (7.6)–(7.7) is not complete unless some prior distributions are specified.9 On the basis of our previous experience, we expect a low value for nj, so we assign to it a prior such that 1/nj ∼ U(0, 0.5). This is a mildly informative prior, given that, consistent with our expectations, it assigns a value of nj in the interval [2,4] with probability 1/2, i.e. Pr(nj ∈ [2, 4]) = 0.5. The other priors are fairly standard and essentially convey the idea that we know very little (if anything) about the parameters:

τj ≈ Gamma ( 0.0001, 0.001)

(7.8)

bk ,s , j ~ N (0, 0.0001) k = 1,K , 3; s = 1,K ,11; j = 1,K , 4

(7.9)

ck ≈ N ( 0, 0.0001) k = 1,..., 3

(7.10)

d k ≈ N ( 0, 0.0001) k = 1,..., 3

(7.11)

ek ≈ N ( 0, 0.0001) k = 1,..., 3

(7.12)

where N(m,t) denotes a normal distribution with mean m and precision (the inverse of the variance) t. For example, prior (7.10) indicates that we believe that the distribution of the coefficient ck is centred around zero (implying no influence of the associated variable), but with a high dispersion, so that the parameter can actually assume any value within a very large range, in this way reflecting our uncertainty, or ignorance, about ck. The model is estimated by Monte Carlo Markov Chains (MCMC) using R and WinBUGS (R Development Core Team 2006, Spiegelhalter et al. 2004).10 Three independent chains of length 1,000 (excluding the burn-­in replications) are used to derive the posterior distributions of the parameters. The starting values of the chains were randomly selected from uniform distributions. Convergence11 is reached for all the model parameters. Table 7.5 lists the descriptive statistics of the parameters of main interest in the model, computed using the estimated posteriors.12 The table also shows the potential scale reduction factor R� (Gelman and Rubin 1992): if convergence is reached, R� should be close to unity. Prior to commenting in detail on the economic significance of our results, we want to check the plausibility of our model using posterior predictive analysis, along the lines exemplified, for example, in Geweke and McCausland (2001) and Gelman et al. (2004). This approach consists of simulating data using the maintained model. The simulated data are then compared with the actual data, usually on the basis of some chosen descriptive statistics. Models that provide accurate descriptions of the data should replicate the observed ones rather accurately. Specifically, we compare the ability of our model to replicate some interesting features of the data, as opposed to a simple Gaussian model. To derive the

c1 c2 c3 d1 d2 d3 e1 e2 e3 τ1 τ2 τ3 τ4 ν1 ν2 ν3 ν4

0.0040 0.0018 0.0013 –0.0078 –0.0079 0.0031 –0.0031 0.0178 0.0389 27.8026 50.0031 65.5372 60.0378 2.5251 2.9253 2.7469 3.0594

Mean

0.0018 0.0015 0.0014 0.0023 0.0041 0.0062 0.0042 0.0042 0.0068 0.8110 1.4835 2.4488 5.7141 0.0817 0.1083 0.1254 0.3820

SD

0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0001 0.0001 0.0001 0.0148 0.0271 0.0447 0.1043 0.0015 0.0020 0.0023 0.0070

Naive SE 0.0000 0.0000 0.0000 0.0001 0.0001 0.0002 0.0001 0.0001 0.0002 0.0472 0.0900 0.1448 0.3578 0.0054 0.0076 0.0088 0.0243

MC error 0.0000 0.0000 0.0000 0.0001 0.0001 0.0002 0.0001 0.0001 0.0002 0.0475 0.0920 0.1167 0.3005 0.0055 0.0083 0.0065 0.0194

Batch SE 0.0870 0.0151 0.1296 0.2876 –0.2852 0.0887 0.1000 0.1245 0.3481 0.1479 –0.0100 0.1613 0.0263 0.1880 –0.0871 0.1076 0.0566

Batch ACF

Table 7.5  Estimates and convergence diagnostics of the main parameters of interest

0.0011 –0.0007 –0.0010 –0.0116 –0.0147 –0.0070 –0.0101 0.0109 0.0273 26.5495 47.5600 61.5300 51.2190 2.3920 2.7540 2.5580 2.5050

Q(0.05) 0.0040 0.0018 0.0013 –0.0078 –0.0079 0.0030 –0.0032 0.0179 0.0390 27.7800 50.0000 65.5600 59.6500 2.5240 2.9220 2.7370 3.0250

Q(0.50)

0.0069 0.0043 0.0036 –0.0039 –0.0013 0.0132 0.0037 0.0247 0.0498 29.1805 52.4500 69.5505 69.9810 2.6651 3.1110 2.9670 3.7371

Q(0.95)

1.0011 1.0005 1.0005 0.9994 1.0013 1.0007 0.9991 0.9990 1.0003 1.0282 1.0064 1.0174 1.0146 1.0296 1.0131 1.0385 1.0366

Rˆ

Does spatial proximity matter?   175 implications from a Gaussian model, we simply simulate y† ~ N(Xbˆ, sˆ 2I), with bˆ and sˆ 2 being OLS estimates. At each simulation step we compute the statistics of interest on y† and compare them with those computed using actual data. By repeating this procedure several times, it is possible to derive the quantiles of the distributions of the statistics of interest from the simulated data, as well as their p-­values. Then, we can use posterior predictive simulation to compare the implications of our model with respect to the observed same characteristics of the data. In the present study we focus on excess kurtosis, the skewness coefficient, and the quantile ratio defined as (max(y) – min(y))/(y0.75 – y0.25) with y0.75 and y0.25 the third and first quartile of y, respectively. The results, reported in detail in Table 7.6, show that the Gaussian model is not able to reproduce the observed characteristics of the data.13 In no instance are the observed values included within the 5–95 per cent quantile intervals. The p-­values are always 0, up to the third decimal. Our hierarchical model, on the other hand, “fits” the data much better, in the sense that it is able more accurately to reproduce the features that are present in the data. In fact, the 5–95 per cent quantile intervals always cover the observed values and the p-­values are always well above any conventional significance level. Of course, there might still be margins of improvement, but the results are strongly suggestive of the marked superiority of our model over the standard Gaussian alternative.

Figure 7.2 Estimated densities of the posterior distributions of the degrees of freedom parameter, νj. The title of the graph indicates the number of employees. The solid lines indicate the medians of the distributions. The dashed lines denote 90 per cent highest posterior density (HPD) intervals.

Excess kurtosis Skewness Quantile ratio

  3.168 –0.264   9.652

Data

  0.401 –0.093   7.104

Median   0.328 –0.122   6.564

Q(0.05)

Gaussian model

  0.477 –0.066   7.739

Q(0.95)

Table 7.6  Posterior predictive distributions and two-sided p-values

0.000 0.000 0.000

p-value   1.695 –0.210 10.840

  1.534 –0.406   9.020

Hierarchical model █ Median Q(0.05)

11.041 –0.081 24.252

Q(0.95)

0.244 0.202 0.616

p-value

Does spatial proximity matter?   177 Overall, these results confirm our intuition, based on the preliminary results, that the data seem to be conditionally t-­distributed with about three degrees of freedom, rather than being normally distributed (see the values of vˆj in Table 7.5). Indeed, given the importance of this parameter, it could be instructive to plot its posterior density. Figure 7.2 shows that the posterior median of nj is very close to 3 for any size class. Furthermore, all the upper bounds of the 90 per cent highest posterior density (HPD) intervals are well below 4.14 We consider this result as a clear confirmation of non-­normality. Also, substantial heteroschedasticity across size classes is confirmed, as previously suggested (see the values of t in Table 7.5). 4.1  Economic implications An interesting side-­result is that our estimates show that, controlling for other factors, in the period 1998–2001 employment growth was higher in the southern and the north-­eastern regions (parameters e1 and e2 in Table 7.5) compared to the remaining geographic areas of the country. Employment growth in the central and north-­western regions was comparable in size. Of course, the possibility that these growth differentials would be extended over long time horizons, which would reduce the existing unemployment differentials is beyond the scope of the present work.15 A summary of the estimated intercepts and of coefficients b2 and b3 is reported graphically in Figures 7.3–7.5. Estimates show that the parameters have the expected signs and that there is substantial variation across sectors and firm sizes. In terms of the main goal of the analysis, according to our estimates, localisation effects are positive, but decreasing with distance (see Figure 7.6). More precisely, the posterior distributions show that the model predicts that there is about a 99 per cent probability that the localisation effect is positive within two kilometres. The probability decreases to about 89 per cent and 82 per cent for distances between 2–10 kilometres and 10–30 kilometres, respectively. The median effect decreases similarly. This evidence confirms the importance of taking account of attenuation phenomena when considering these types of agglomeration forces. In addition, this evidence suggests that the use of geographic units such as standard metropolitan units, LLSs, administrative regions or provinces (exogenously defined) can be misleading, since the impact of localisation economies on employment growth tends to change with distance. It should be noted that this result, even if not counter-­intuitive with respect to the Italian experience – see, for example, the literature on Italian industrial districts (Signorini 1994) which suggests a positive role for these kinds of agglomeration forces in explaining the success of these local production systems – is not consistent with some of the previous contributions on this issue. In fact, some studies, using specialisation indicators, find a negative role for localisation economies (Glaeser et al. 1992; Henderson et al. 1995; Cainelli

178   G. Cainelli and C. Lupi

Figure 7.3 Estimated intercepts by sector and firm size. The title of the graph indicates the number of employees, the sectors are indicated from 1 to 11. The order of the sectors follows the official NACE Rev. 1.1 classification, with the aggregation described in the main text (DA, DB + DC, DD + DN, DE, DF + DG, DH, DI, DJ, DK, DL, DM). The points are the medians of the posterior distributions. Solid vertical lines represent 90 per cent highest posterior density (HPD) intervals.

and Leoncini 1999). At the same time, other works such as, for example, Forni and Paba (2002) find that the impact of productive specialisation is positive when it is considered jointly with size (in terms of employment) of the local industry under consideration. In other words, the main idea behind these papers is that specialisation and size must be jointly considered in order to capture the knowledge spillovers generating process. In our work, however, unlike some previous studies (see, for example, Glaeser et al. 1992), we do not use specialisation indicators to measure localisation economies, but rather we use direct measures of spatial agglomeration: that is, the number of firms actually located at different distances. In our view, these measures should partially capture the size, in this case in terms of number of firms, of the local industry being considered. This might explain the finding that localisation economies matter for employment growth. A second explanation might be that, as already noted, we do not use pre-­defined geographic units such as LLSs. It is well known that the size of these geographic units tends to change considerably in terms of areas,

Does spatial proximity matter?   179

Figure 7.4 Estimated b2 parameters by sector and firm size. The title of the graph indicates the number of employees, the sectors are indicated from 1 to 11. The order of the sectors follows the official NACE Rev. 1.1 classification, with the aggregation described in the main text (DA, DB + DC, DD + DN, DE, DF + DG, DH, DI, DJ, DK, DL, DM). The points are the medians of the posterior distributions. Solid vertical lines represent 90 per cent highest posterior density (HPD) intervals. .

population and so on, thus imposing a pre-­defined, non-­empirically tested spatial boundary within which agglomeration forces should act. In addition, this non-­empirically tested hypothesis changes according to the LLS being considered. At the same time, the variety effects are negative for distances within two kilometres and between 2–10 kilometres, while these forms of agglomeration forces become positive for distances between 10–30 kilometres. Indeed, our model suggests that variety effects are negative within a distance of 10 kilometres with a probability greater than 97 per cent, while the probability that the effect is positive for longer distances is about 70 per cent. In this case when spatial agglomeration is being analysed, the importance of taking account of attenuation phenomena is again confirmed. Moreover, our evidence suggests that the variety of the production structure has a positive impact on firms’ employment growth only for distances between 10–30 kilometres, which supports previous studies that identified a positive role of variety on employment growth

180   G. Cainelli and C. Lupi

Figure 7.5 Estimated b3 parameters by sector and firm size. The title of the graph indicates the number of employees, the sectors are indicated from 1 to 11. The order of the sectors follows the official NACE Rev. 1.1 classification, with the aggregation described in the main text (DA, DB + DC, DD + DN, DE, DF + DG, DH, DI, DJ, DK, DL, DM). The points are the medians of the posterior distributions. Solid vertical lines represent 90 per cent highest posterior density (HPD) intervals.

(Glaeser et al. 1992), but also suggests that this positive role of variety needs distance to become effective. In this sense, our results are only partially consistent with the previous literature on Italy (see, e.g. Cainelli and Leoncini 1999; Cainelli et al. 2001).

Does spatial proximity matter?   181

Figure 7.6 Estimated “localisation” (ck, k = 1, . . ., 3) and “variety” (dk, k = 1, . . ., 3) parameters. The value of k is reported on the x-axis: on the y-axis is reported the value of the parameter. k = 1 indicates a distance d ≤ 2 km; k = 2 denotes 2 km < d ≤ 10 km; k = 3 stands for 10 km < d ≤ 30 km. Large points are the medians of the posterior distributions. Solid vertical lines represent 90% highest posterior density (HPD) intervals; 80 per cent and 66 per cent HPD intervals are represented by “–” and “x” respectively.

5  Conclusions In this study we used a panel data set of 23,374 Italian manufacturing firms for the period 1998–2001 to estimate a flexible Bayesian model, to examine the impact of spatial agglomeration economies – that is, localisation and urbanisation economies – on employment growth over discrete distances. Our main results can be summarised as follows. We find that localisation effects are positive, but decreasing with distance. More precisely, the posterior distributions show that the model predicts that there is an approximately 99 per cent probability that the localisation effect is positive within two kilometres. This probability decreases to some 89 per cent and 82 per cent for distances between 2–10 kilometres and 10–30 kilometres, respectively. Variety effects, on the other hand, are negative for distances within two kilometres and between 2–10 kilometres, while this form of agglomeration force becomes positive for distances between 10–30 kilometres.

182   G. Cainelli and C. Lupi Our results contrast with the previous literature on agglomeration economies which generally suggests a negative role for localisation economies and it is only partially consistent with those works underlying a role for variety on local growth. In this sense, we believe that the main finding of this work is that the use of geographic units such as SMAs, LLSs, administrative regions or provinces (exogenously defined) can be misleading, since the impact of spatial agglomeration economies on employment growth tends to change with distance, and local knowledge spillovers seem to attenuate over space. This means that some of the results obtained by this literature could be vitiated by using exogemously defined geographic units rather than by directly measuring indicators of spatial agglomeration. It should be noted that, as suggested by Martin et al. (2008), analysis of agglomeration economies is relevant not only for understanding the “mechanics” behind regional and local development, but also for assessing the potential implications for industrial cluster policies. It is not a case that since the end of the 1980s agglomeration economies have been invoked by national and regional/ local institutions to justify public intervention in favour of industrial clusters. In fact, in most countries underlying these policies was the idea that interventions aimed at stimulating productive agglomeration could have a positive effects on firms’ employment and productivity growth. A significant example of these kinds of interventions is represented by public policies in favour of industrial districts in Italy. The most important of these resulted by national law no. 317 of 1991; the most recent examples are the norms introduced in the 2005 Budget Law (Cainelli 2008). The main idea behind these interventions is that agglomeration is always positive. Our findings are partially consistent with this. According to our results, agglomeration forces seem to be an important factor for explaining firms’ competitiveness, but their intensity and sign tend to change with spatial distance. Specialisation seems to apply within a very limited spatial area, while variety works only in the case of longer distance. These aspects, which are not usually considered in the design of industrial cluster policies, should be at the centre of these interventions. Our study could be extended in a number of interesting ways. First, our data-­ set could be enlarged to include non-­joint stock companies. This could be achieved by using microeconomic information drawn from census data, for example. Finally, reliable data on firms’ capital stocks could allow us to investigate the impact of spatial agglomeration economies on firms’ productivity growth or TFP.

Does spatial proximity matter?   183

Appendix Table 7.7  Classification of manufacturing activities Code

Numerical code

Description

DA DB DC DD DE DF DG DH DI DJ DK DL DM DN

15, 16 17, 18 19 20 21, 22 23 24 25 26 27, 28 29, 29 31, 32, 33 34, 35 36

Food products, beverages and tobacco Textiles and clothing Leather and leather products Wood and wood products Pulp, paper and paper products Coke, refined petroleum products and nuclear fuel Chemicals, chemical products and man-made fibres Rubber and plastic products Non-metallic mineral products Basic metals and fabricated metal products Machinery and equipment Electrical and optical equipment Transport equipment Other manufacturing

Notes   1 We thank Nicola De Liso and Riccardo Leoncini for giving us the opportunity to contribute to this book, and for comments and suggestions on a previous version of this chapter, which improved the work.   2 Each firm in the database is assigned to a sector according to the Statistical Classification of Economic Activities in the European Community, NACE Rev. 1.1 (2002). The correspondence between industry codes and their descriptions is provided in Table 7.7 in the Appendix.   3 A similar approach is used by Martin et al. (2008) who drop all firms located in Corsica and in overseas departments.   4 A referee asked us to compare these data with analogous information at European level. Unfortunately, we cannot make this comparison because similar data for other European countries are not currently available. Indirect evidence similar to ours can be found in Duranton and Overman (2005, 2008).   5 The posterior distribution is the central object of interest in a Bayesian inference. It amounts to the distribution of the (possibly vector) random variable (parameter) q conditioned on having observed the data y, p(q|y). Once we have an estimate of the posterior distribution, deriving its mean, median, mode or other quantities (moments) of interest is easy. Common Bayesian point estimates of the parameter are the median or the mean of the posterior, E(q|y).   6 The precise definition of the variables is provided in Table 7.4.   7 With the exception of Sicily and Sardinia which are excluded from the analysis, macro-­areas are defined according to the official classification used by the Italian National Institute of Statistics (Istat). Of course there is no geographical dummy for the centre, taken here as the “reference” area.   8 In Bayesian analysis it is customary to use precision, the inverse of variance, to summarise a distribution’s dispersion.   9 This is because the posterior of parameter q given data y, which is the object of primary interest, can be written as p(q/y) µ f (y/q)π(q) where p(q/y) is the likelihood, and π(q) is the prior. In essence, the prior summarises the investigator’s knowledge about q prior to observing the data y. 10 MCMC methods are simulation-­based estimation procedures, which can be traced back to the seminal contributions of Metropolis et al. (1953) and Hastings (1970).

184   G. Cainelli and C. Lupi Introductory treatments can be found in Geweke (1997, 1999), Brooks (1998), Gelfand (2000) and Jackman (2000). A distinctive feature of MCMC methods is that they focus on sampling, rather than on optimising. Loosely speaking, the idea is to find a simple process (the Markov Chain) from which to sample, in which distribution converges, after an adequate number of burn-­in iterations, to the distribution of interest (the posterior). Then, sampling from this process becomes essentially equivalent to sampling from our unknown distribution of interest. 11 The term “convergence” is used here with reference to the relevant distributions, not with reference to the pointwise convergence of parameters values, as is usual, for example, in non-­linear optimisation methods. 12 The descriptive statistics reported in Table 7.5 are based on 3,000 iterations (three independent chains of 1,000 iterations each). The first two columns of Table 7.5 report the sample mean and the sample standard deviation of the posteriors. Three estimates of the standard error follow: “Naive SE” is a naive estimate (the sample standard deviation divided by the square root of the sample size) which assumes that sampled values are independent, “MC Error” is a time–series estimate (the square root of the spectral density variance estimate divided by sample size) which gives the asymptotic standard error (Geweke 1992), “Batch SE” is a batch estimate calculated as the sample standard deviation of the means from consecutive batches of size 50  divided by the square root of the number of batches. “Batch ACF ” is the autocorrelation between batch means. Q(0.05), Q(0.50), and Q(0.95) are the 5 per cent quantile, the median, and the 95 per cent quantile of the posterior distribution. R� is the potential scale reduction (Gelman and Rubin 1992). R� ≈ 1 denotes good convergence. 13 Column 1 in Table 7.6 reports the statistics of interest computed on actual data. “Median”, Q(0.05) and Q(0.95) are the median, the 5 per cent and the 95 per cent quantiles of the distributions of the statistics for the simulated data, respectively. The results are based on 1,000 simulations. 14 The Highest Posterior Density Intervals are the Bayesian confidence intervals (or credibility intervals) such that for a parameter of interest q, qL and qH determine the shortest interval for which Pr(qL ≤ q ≤ qH) = a(0 < a < 1). In our computations we consider a = 0.9. 15 For an assessment of geographical differences in unemployment in Italy see, among others, Brunello et al. (2001).

References Audretsch, D. and Feldman, M.P. (1996), R&D spillovers and the geography of innovation and production, American Economic Review, 86(3): 630–40. Boarnet, M.G. (1998), Spillovers and the locational effects of public infrastructures, Journal of Regional Science, 38(3): 381–400. Brooks, S.P. (1998), Markov chain Monte Carlo method and its application, The Statistician, 47(1): 69–100. Brunello, G., Lupi, C. and Ordine, P. (2001), Widening differences in Italian regional unemployment, Labour Economics, 8(1): 103–29. Brynjolfsson, E. and Hitt, L.M. (2003), Computing productivity: Firm-­level evidence, Review of Economics and Statistics, 85(4): 793–808. Cainelli, G. and Leoncini, R. (1999), Externalities and long-­term local industrial development: Some empirical evidence for Italy, Revue d’Economie Industrielle, 90(4): 25–39. Cainelli, G., Leoncini, R. and Montini, A. (2001), Struttura produttiva e sviluppo regionale di lungo periodo in Italia, Moneta e Credito, 54(216): 461–85. Cainelli, G. (2008). Industrial districts: theoretical and empirical insights, in Karlsson, C. (ed.), Handbook of Research on Cluster Theory, Edward Elgar Publishing, Cheltenham.

Does spatial proximity matter?   185 Cingano, F. and Schivardi, F. (2004), Identifying the sources of local productivity growth, Journal of the European Economic Association, 2(4): 720–42. Combes, P.-­P. (2000), Economic Structure and Local Growth: France, 1984–1993, Journal of Urban Economics, 47(3): 329–55. Cunat, A. and Peri, G. (2001), Job creation: geography, technology and infrastructures, Giornale degli Economisti e Annali di Economia, 61(1): 23–54. de Lucio, J.J., Herce, J.A. and Goilcolea, A. (2002), The effects of externalities on ­productivity growth in Spanish industry, Regional Science and Urban Economics, 32(2): 241–58. Desmet, K., Fafchamps, M. (2005), Changes in the spatial concentration of employment across U.S. counties: A sectoral analysis 1972–2000, Journal of Economic Geography, 5(3): 261–84. Duranton G. and Overman H. (2005), Testing for localisation using micro-­geographic data, Review of Economic Studies, 72(4): 1077–106. Duranton G. and Overman H. (2008), Exploring the detailed location patterns of UK manufacturing industries using microgeographic data, Journal of Regional Science, 48(1): 313–43. Forni, M. and Paba, S. (2002), Spillover and the growth of local industries, Journal of Industrial Economics, 50(2): 151–71. Gelfand, A.E. (2000), Gibbs sampling, Journal of the American Statistical Association, 95(452): 1300–4. Gelman, A., Carlin, J.B., Stern, H.S. and Rubin, D.B. (2004), Bayesian Data Analysis, Second Edition. Texts in Statistical Science, Chapman & Hall/CRC, Boca Raton/ London/New York/Washington D.C. Gelman, A., Rubin, D. B., (1992), Inference from iterative simulation using multiple sequences, Statistical Science, 7(4): 457–72. Geweke, J. (1992), Evaluating the accuracy of sampling-­based approaches to calculating posterior moments, in Bernardo, J.M., Berger, J.O., Dawid, A.P. and Smith, A.F.M. (eds), Bayesian Statistics, Vol. 4, Oxford University Press, Oxford. Geweke, J. (1993), Bayesian treatment of the independent Student-­t linear model, Journal of Applied Econometrics, 8(supplement): S19–S40. Geweke, J. (1997), Posterior simulators in econometrics, in Kreps, D. and Wallis, K. (eds), Advances in Economics and Econometrics: Theory and Applications, Vol. 3, Cambridge University Press, Cambridge. Geweke, J. (1999), Using simulation methods for Bayesian econometric models: inference, development and communication, Econometric Reviews, 18(1): 1–73. Geweke, J. and McCausland, W. (2001), Bayesian specification analysis in econometrics, American Journal of Agricultural Economics, 83(5): 1181–6. Glaeser, E., Kallal, H.D., Scheinkman, J.A. and Scheifer, A. (1992), Growth in cities, Journal of Political Economy, 100(6): 1126–52. Grossman, G. and Helpman, E. (1991), Innovation and Growth in the Global Economy, MIT Press, Cambridge, MA. Hastings, W.K. (1970), Monte Carlo sampling methods using Markov chains and their applications, Biometrika, 57(1): 97–109. Henderson, V.J. (2003), Marshall’s scale economies, Journal of Urban Economics, 53(1): 1–28. Henderson, V.J., Kunkoro, A. and Turner, M. (1995), Industrial development in cities, Journal of Political Economy, 103(5): 1067–90. Holtz-­Eakin, D. and Schwartz, A. E. (1995), Spatial productivity spillovers from public

186   G. Cainelli and C. Lupi infrastructure: evidence from state highways, International Tax and Public Finance, 2(3): 459–68. Jackman, S. (2000), Estimation and inference via Bayesian simulation: An introduction to Markov chain Monte Carlo, American Journal of Political Science, 44(2): 375–404. Jaffe, A.B., Trajtenberg, M. and Henderson, R. (1993), Geographic localization of knowledge spillovers as evidenced by patent citations, Quarterly Journal of Economics, 63(3): 483–99. Lucas, R.J. (1988), On the mechanics of economic development, Journal of Monetary Economics, 22(1): 3–42. Mameli, F., Faggian, P. and McCann, P. (2007), Agglomeration economies and employment growth in Italian local labour systems, in Joint Congress of the European Regional Science Association and Association de Science Régionale de Langue Française, August 2007, Paris. Marshall, A. (1920), Principles of Economics, 8th edition, Macmillan, London. Martin, P., Mayer, T. and Mayneris, F. (2008), Spatial concentration and firm-­level productivity in France, Discussion Paper series 6858, CEPR, London. Metropolis, N., Rosenbluth, A.W., Rosenbluth, R.N., Teller, A.H. and Teller, E. (1953), Equations of state calculations by fast computing machines, Journal of Chemical Physics, 21(6): 1087–91. Paci, R. and Usai, S. (2006), Agglomeration economies and growth: the case of Italian local labour systems, 1991–2001, Working Paper 2006/12, CRENOS, Cagliari. Picci, L. (1999), Productivity and infrastructure in Italian regions, Giornale degli Economisti e Annali di Economia, 58(3–4): 329–53. R Development Core Team (2006), R 2.4.0 – a language and environment for statistical computing, technical report, R Foundation for Statistical Computing, Vienna. Ricotta, C. and Szeidl, L. (2006), Towards a unifying approach to diversity measures: Bridging the gap between the Shannon entropy and Rao’s quadratic index, Theoretical Population Biology, 70(3): 237–43. Romer, P. (1986), Increasing returns and long-­run growth, Journal of Political Economy, 94(5): 1002–37. Romer, P. (1990), Endogenous technological change, Journal of Political Economy, 98(S5): S71–S102. Rosenthal, S. and Strange, W. (2008), The attenuation of human capital spillovers, Journal of Urban Economics, 64(2): 373–89. Shannon, C.E. (1948), A mathematical theory of communication, The Bell System Technical Journal, 27(July, October): 379–423, 623–656. Signorini, L.F. (1994), The price of Prato, or measuring the industrial district effects, Papers in Regional Science, 73(4): 369–92. Spiegelhalter, D.J., Thomas, A., Best, N. and Lunn, D. (2004), WinBUGS user manual, technical report, MRC Biostatistics Unit, Institute of Public Health, Cambridge. Storper, M. and Venables, A.J. (2004), Buzz: face-­to-face contact and the urban economy, Journal of Economic Geography, 4(4): 351–70. Usai, S. and Paci, R. (2003), Externalities and local economic growth in manufacturing industries, in Fingleton, B. (ed.), European Regional Growth, Springer-­Verlag, Berlin. van Oort, F.G. (2007), Spatial and sectoral composition effects of agglomeration economies in the Netherlands, Papers in Regional Science, 86(1): 5–30. Wallsten, S.J. (2001), An empirical test of geographic knowledge spillovers using geographic information systems and firm-­level data, Regional Science and Urban Economico, 31(5): 571–99.

8 Internationalization in Italian medium-­sized firms Does stage theory explain the observed patterns? Donato Iacobucci and Francesca Spigarelli 1  Introduction This chapter examines the extent to which internationalization strategy theories explain the patterns of internationalization observed in Italian medium-­sized firms in recent years. We focus on stage theory because it seems the most appropriate to explain the patterns of internationalization in small- and medium-­sized enterprises (SMEs). Stage theory assumes that the process of internationalization follows a prescribed path from the lighter mode of entry (based on indirect export) to more intensive investment (in the form of foreign direct investments – FDI). Although this assumption has been challenged on both theoretical and empirical levels, some authors (Gankema et al. 2000) have found that stage theory can be applied to explain SMEs’ behavior. More information about the validity of this explanation would be of interest to both firms and policy makers. In view of the lack of empirical evidence on the internationalization of Italian SMEs and theoretical models explaining their international activities, this chapter aims to analyze the characteristics and recent evolution of patterns of internationalization in Italian medium-­sized firms and to assess the extent to which the observed patterns are compatible with the predictions of stage theory. The empirical analysis refers to 242 manufacturing companies and groups located in the North-­East-Center of Italy (the so-­called NEC regions or “third Italy”). By medium-­sized firms we mean firms with 250–2,500 employees. The focus on these companies is justified by their increasing role in the Italian manufacturing system (Brioschi et al. 2002; Balloni and Iacobucci 2004; Coltorti 2004). The NEC regions were chosen because of their peculiarities in terms of the organization of manufacturing activities, which is based on industrial districts: i.e. local systems of SMEs specialized in the same sector (Cainelli and Zoboli 2004). The SMEs located in these regions have demonstrated remarkable ability to penetrate international markets through export (Menghinello 2004), but found it difficult to develop stable forms of internationalization through productive or commercial units abroad. One of the aims of this chapter is to assess whether medium-­sized firms can change the internationalization patterns observed so far in Italian SMEs, from export to FDI.

188   D. Iacobucci and F. Spigarelli We find that size does not affect average export intensity, but is important for FDI-­based strategies. Size affects the “magnitude” and intensity of investment in foreign markets, but not the willingness to develop internationalization activities abroad: indeed, smaller companies have experienced faster growth in terms of the number of foreign subsidiaries in the period considered. In general, investing abroad, within a gradual approach to internationalization, tends to be complementary to and not a substitute for export. In terms of the geographic span of operations, over time, FDI activities have moved from closer to more distant locations, confirming a “process approach” to internationalization driven by knowledge acquisition. Overall, the evidence accords with the propositions based on stage theory; however, we found some conflicts. When we looked at companies that in 2001 had not embarked on the process of internationalization, we found that the majority jumped directly to an FDI approach, rather than starting with a less intensive mode of internationalization, such as export. We did not expect the internationalization strategies of these firms, which we refer to as “Pioneers”, to be successful. However, this was only partially confirmed by the data as the majority of these Pioneers had survived, in the same stage, at the end of the period. The chapter is organized as follows. Section 2 reviews the current literature on the internationalization strategies of firms, with a specific focus on the stage theory hypothesis. We highlight the predictions of different approaches and discuss the problems emerging from empirical work. Finally, we list and discuss the hypotheses relating to firm behavior that can be deduced from stage theory. Section 3 describes our empirical methodology, providing information on the data, the sample and the variables considered in the quantitative analysis. Section 4 discusses the results of the empirical analysis, and Section 5 provides some conclusions.

2  Background: theory and hypotheses 2.1 Theoretical approaches to SMEs’ internationalization Since 1990 a growing body of literature has focused on the specific role of SMEs in international competition. The internationalization of SMEs is of particular scientific interest because such firms have specific features that affect their attitude to global expansion, compared to that of multinational enterprises (MNEs), the traditional subject for study. Their managerial styles, the role of the entrepreneur, relational social capital, and scale and scope of activities are completely different from those of big firms. Moreover, SMEs usually have to cope with constraints on (or more difficult access to) key resources: financing, management capabilities, skilled labor, information (Erramilli and D’Souza 1993; Lu and Beamish 2001). It has been suggested that new paradigms and multi-­theoretical frameworks are required to understand SMEs’ behavior (Malhotra et al. 2003). The literature focusing on the internationalization of SMEs takes different theoretical positions, which have been mapped and analyzed by numerous authors.1 These contributions can be categorized mainly within the theoretical

Internationalization in Italian firms   189 perspectives stage theory (Cavusgil 1980), network theory (Coviello and McAuley 1999) and FDI theory. We provide a brief review of the literature related to the three main internationalization theories (stage, network, and FDI), with a specific focus on stage theory. Stage theory The argument in stage theory is that internationalization is a gradual strategy, behaviorally oriented, and developed in successive phases (Melin 1992). The main models with this approach are the Uppsala Internationalization model (U-­M) and the Innovation-­related internationalization model (I-­M). The U-­M (Johanson and Vahlne 1977) focuses on managerial competencies: international expansion is driven by the gradual acquisition of competences and experience in foreign markets. Nearby markets are explored through indirect exporting in the early stage of the process, and more physically distant markets are approached though more complex modes of entry, in an incremental way. Four stages have been proposed: no regular export activity; export via independent representatives; establishing an overseas sales subsidiary; establishing overseas production/ manufacturing units (Andersen 1993). In the I-­M, internationalization is seen as a firm innovation strategy (Cavusgil 1980) in which the different stages are linked to different exporting trends and dynamics. Five stages of international commitment are proposed (Cavusgil 1984): preinvolvement, reactive/opportunistic, experimental, active, and committed involvement. Though somewhat different in their theoretical bases, from an empirical point of view the two approaches reach similar conclusions: i.e. in their internationalization strategies firms are expected to follow a sequence of stages characterized by an increasing degree of resource investment and commitment in foreign operations. Some recent studies (Gankema et al. 2000) demonstrate that, within certain limits, Cavusgil’s stage theory holds: i.e. internationalization occurs in stages in European manufacturing SMEs. However, stage theory has encountered growing criticism from both the theoretical and empirical perspectives (Hurmerinta-­ Peltomaki 2003). First, the model is time dependent and assumes a predetermined path of development. It is therefore unsuitable for firms with extensive international experience, and firms in high-­technology, knowledge intensive or service sectors (Bell 1995; Ibeh et al. 2004). Firms in these sectors usually “jump” directly to complex stages of internationalization. Leapfrogging the predetermined stages could also be because SMEs are focused on global market niches or because of decreases in transportation and communication costs (McDougall et al. 1994; Oviatt and McDougall 1994). Internationally experienced management could also help SMEs to jump the first stages of internationalization (Fischer and Reuber 1997; Belso-­Martines 2006). Finally, some SMEs seem born to be global and follow an internationalization strategy from the time they start up (Bell et al. 2001; Moen and Servais 2002; Andersson and Wictor 2003; Rialp et al. 2005). Some authors have underlined that the sequential approach is not suited to explaining the internationalization strategies of emerging global firms which are

190   D. Iacobucci and F. Spigarelli seeking to acquire high value resources (technology, know how, brands) through rapid expansion; in most cases, such firms are supported by government policy (Buckley et al. 2007). In these contexts, a different view of time should be adopted, that is “cyclical time with no fixed directions” (Hurmerinta-­Peltomaki 2003). Another criticism of stage theory is related to the fact that the model emphasizes organizational learning, but fails to explain how the knowledge developed over time affects organizational behavior. In this learning process, the role of key individuals in firms is not taken into account (Anderson 2000). Moreover, if the leaders of the internationalization strategy change over time, the acquisition of experiential knowledge could be interrupted, with unknown consequences for the stage patterns (Bjorkman and Forsgren 2000). Network model The idea underlying the network model is the increasing role of network relationships in firms’ strategic activities. Network alliances are supposed to determine the success or failure of internationalization and the pattern adopted (Coviello and Munro 1997; Coviello and McAuley 1999). This approach focuses on the importance of organizational and social links, based on formal and informal relationships. Network members’ relations and interactions can influence both the decision to export and the mode of entry into different markets. Case study analyses demonstrate the importance of relationships with foreign markets in explaining the internationalization strategies of firms (Johanson and Vahlne 1992). These relations might be business or personal in origin: social and cognitive ties in the business context are important for explaining firm behavior and contrast with the strategic or strictly economic perspective. The main criticism directed to the network model is that it supports less precise conclusions than stage theory on both the empirical effects of a “going abroad” strategy and on the pattern of internationalization. Moreover, as network theory is focused on the presence of a web of multiple relationships (among different firms and social actors) in the business context, it is difficult to use it for predictive purposes (Bjorkman and Forsgren 2000). It also does not explain internationalization by firms that do not belong to a network (Malhotra et al. 2003). FDI approach The main body of FDI theory is underpinned by various theoretical approaches. Based on the seminal works of Coase and Williamson, internationalization is seen as a decision that is affected by transaction costs, in a context of monopolistic advantage and market imperfections. Internationalization strategy and mode of entry are defined in order to resolve the trade-­off between control costs and integration costs (Erramilli and Rao 1993). In the eclectic paradigm (Dunning 1988) attention is drawn to the specific advantages that firms want to acquire by investing abroad: ownership, location and internalization (OLI). An updated version of OLI theory was proposed by

Internationalization in Italian firms   191 Dunning (2000) which takes account of the economic and political changes that occurred in the 1990s by considering the new costs and benefits arising from relationships in the business context, knowledge intensive assets, global alliances, trading blocks, innovation, technological standards, etc. The main weakness to this view is that it is considered too static (Malhotra et al. 2003). Also, the transaction costs view takes account of the fact that costs cannot be measured very accurately. The FDI approach finds some support in the empirical literature, although it mainly focuses on large firms; the internationalization strategies of SMEs seem to be heavily influenced by individual bias, which is unexplained (Apfelthaler 2000). Combined approaches Many studies have tried to combine several different approaches, but the results are often mixed and inconsistent. Studies that try to combine stage theory and the network model (Bell 1995) find little support for either explanation of internationalization in small software firms. Other studies try to integrate all three approaches as none, in isolation, explain SMEs’ expansion abroad: thus, a comparative approach is suggested (Coviello and Martin 1999). Other studies reject the idea of a systematic approach to the internationalization strategies of SMEs (Yip et al. 2000), and Chetty and Campbell-­Hunt (2003) suggest that internationalization is not generally a process that is pre-­planned in absolute detail. Other findings only partially confirm stage theory, suggesting that entrepreneurial behavior seems to be a key factor in explaining a firm’s internationalization strategy, especially in the early stages (Anderson et al. 2000). Even studies that provide direct support for the stage model do not claim that it is the only mode of internationalization followed by SMEs (Jones 1991). Firms often seem to follow quite different and individual paths to entry to overseas markets, where constraints related to tangible and intangible resources are less important than stage and network theory approaches would suggest (Coviello and Martin 1999; Autio et al. 2000). The lack of consistency in the results of these empirical analyses is underlined (Coviello and Martin 1999; McDougall and Oviatt 2000). Poor sample selection, incorrect methodology and, above all, the mixed effects of different strategic actions/ growth strategies developed simultaneously by firms, can cloud the interpretation of SMEs’ internationalization activity. Despite these criticisms, we believe that stage theory is a useful approach, from which testable propositions related to the internationalization process of SMEs can be derived; moreover, it encompasses a gradual approach to foreign markets, which seems to apply to many Italian SMEs. 2.2  Hypotheses The main aim of this chapter is to assess whether the predictions of stage theory are useful for describing the general pattern of evolution of the internationalization

192   D. Iacobucci and F. Spigarelli process in Italian medium-­sized firms. Our empirical analysis uses secondary data from annual reports, thus, the stage theory model we adopt is the U-­M because the different steps in the firm’s internationalization strategy (regular export activity; export via independent representatives; establishing an overseas sales subsidiary; establishing overseas production/manufacturing units) are measurable using such data (see the discussion in Section 3). In what follows, we discuss some research hypotheses related to the stage theory approach. Exporting Stage theory considers exporting to be the first step in the internationalization process: a sort of platform from which to evaluate further international expansion and to enable international experience and practice (Zahra et al. 1997). This mode of entry is regarded as particularly suitable for SMEs, which often lack financial and managerial resources (Dalli 1995; Zahra et al. 1997). Several studies have focused on the relationships between firm size and attitude to export, but results are somewhat contradictory (Wickramasekera and Oczkowski 2004), due to the different measures used to assess size (number of employees, sales, firm’s age) and the different features of the samples analyzed. Some studies find that firm size, measured by number of employees and sales is positively related to export activity, others find no association, while some find a weak link between these features (Bonaccorsi 1992). Calof (1994) found that firm size does not necessarily affect the ability of the firm to enter into exporting activity. Because the smallest firms are excluded from our sample, we can propose the following hypothesis. Hypothesis 1 Firm size does not affect export activity or its intensity FDI Stage theory considers FDI to be a complex mode of entry to foreign markets, to be seen as a more mature step in an internationalization strategy, especially for SMEs.2 Investment abroad requires managerial skills, intensive capital expenditure, and a good knowledge of foreign markets. On the other hand, the more strategic and appropriable the firm’s assets (know how, brand equity, trademarks, patents) become, the more FDIs are essential to avoid distributors’ opportunistic behavior and asset appropriation (Lu and Beamish 2001). As firms acquire knowledge and expertise from international markets, they tend to expand and diversify their FDIs to different locations to take advantage of the different resources and opportunities available (Shan and Song 1997; Deeds and Hill 1998). One aspect specific to FDIs is choice of geographic location. The geographical spread of the markets in which foreign subsidiaries are located is an

Internationalization in Italian firms   193 important variable in decisions about foreign investments. In general, for a variety of reasons, SMEs tend to make their FDIs in nearby, developed countries or in wide-­selling markets. First, SMEs are inclined to follow a market-­seeking approach (Shatz and Venables 2000) in which cost competitiveness is generally less important than it would be for big firms. Their competitive advantage is focused on specialization, differentiation, adaptation to customer needs, and close relationships with customers. Therefore, investment in other countries is usually based on market-­seeking reasons, rather than cost or asset motives (Svetlicic et al. 2007). Moreover, physical distance has much more influence on SMEs’ choices of foreign markets (Dunning 1993). Although these reasons apply in general, there can be considerable differences among sectors. SMEs in highly competitive and volatile products/ markets (Ibeh et al. 2004), such us those operating in short life-­cycle or high technology sectors, tend to use more complex modes of entry (FDIs), for market reasons and for knowledge seeking motives (Burgel and Murray 2000; Ibeh et al. 2004). From the above discussion we can derive the following hypotheses. Hypothesis 2 As companies become more confident about operating in foreign markets, they tend to develop more complex internationalization strategies: from indirect to direct exports and from a single to a multiple presence abroad through increasing numbers of foreign subsidiaries. Hypothesis 3 SMEs’ FDIs tend to follow a market-­seeking approach. Thus, in the early stages of internationalization they choose to locate foreign subsidiaries in nearby or mature markets. Over time, firms developing an FDI-­based strategy will tend to diversify and to increase the number of countries in which foreign units are located, and to choose more distant locations. Hypothesis 4 SMEs belonging to high-­tech sectors tend to establish foreign subsidiaries in countries where high value assets are available, with physical distance being of less importance in their selection of foreign markets. Interaction between export and FDI We focus our analysis on the two main modes of entry within stage theory – export and FDI – and their interaction. Exporting and FDI can be seen as substitutes or as complements, depending on factors such as the stage of development of the host country/market, industry structure, firm strategy, type of FDI (vertical or horizontal), and firm productivity (Helpman et al. 2004). Empirical results vary widely, depending on the level of data aggregation. Some studies

194   D. Iacobucci and F. Spigarelli using industry trade data find strong evidence of complementarity (Clausing 2000; Graham 2000), while studies using product-­level data for specific industries, find evidence of a substitution effect (Blonigen 2001). Although the literature is not conclusive, within stage-­theory FDIs are not seen as a substitute for foreign sales, while export activity is seen as a pre-­ condition and support for an FDI strategy. Hence we can hypothesize that: Hypothesis 5 FDI strategy follows export expansion; moreover, when embarking on an FDI strategy export intensity is maintained, i.e. the two modes of internationalization support one another.

3  Methodology: sample, data sources and variables The empirical analysis considers 242 manufacturing companies and groups located in the NEC or “third Italy” regions. This geographic area was chosen for its peculiarities in the organization of manufacturing activities, based on local systems of SMEs specialized in the same sector (“industrial districts”). Our sample consists of medium-­sized firms defined as companies or groups, with 250–2,500 employees and a turnover of €50–1,000 million. These data refer to 2001, the start year of our observations. Values refer to single companies if they are not members of a group, or to whole groups.3 We excluded companies owned by foreign or domestic groups that exceeded the above employee and turnover limits. We decided on a minimum size of 250 employees because of the increasing role of medium-­sized companies in the Italian industrial system, and especially in the internationalization of industrial districts (Mariotti and Mutinelli 2004). In addition, internationalization activities based on FDIs were significant only for companies above this minimum size (Bugamelli et al. 2000). The list of companies was taken from Bureau Van Dijk’s AIDA database, which provides data from the annual reports of about 700,000 Italian joint-­stock companies. The coverage of this dataset allowed us to consider the population of companies within the size limits mentioned above. In our analysis we consider data referring to 2001 and 2005. A five year period was chosen for two reasons: to verify the predictions of the stage theory hypothesis we need to observe the evolution of internationalization strategies over time, and some of the phenomena under observation, such as number of foreign subsidiaries, are fairly stable over shorter periods of time. Thus, for variations to emerge we need to consider a sufficiently long period of time. The period 2001–2005 is particularly interesting because the Italian economy was encountering increasing difficulties in global markets. In fact, some authors found that the internationalization of Italian companies slowed during this period compared with the 1990s (Mariotti and Mutinelli 2005). The main sources of the data and information are companies’ annual reports, which provide information on company size, performance, asset values, export intensity, and foreign subsidiaries.

Internationalization in Italian firms   195 The population is divided into two sub classes based on number of employees: 250–499 employees and 500–2,500 employees. Companies are also categorized using the Pavitt (1984) classifications to identify differences related to sectors/ typologies of production.4 Table 8.1 shows the distribution of companies by sector and size. Almost half of the companies in our sample belong to the traditional industries (supplier dominated in Pavitt’s classification) and only 5 percent of them to the science-­ based sectors. As proxies for internationalization we chose two variables based on the information available from annual reports. The first is export intensity, i.e. the ratio of foreign to total sales, which is frequently used in the literature as a measure of export intensity.5 The second is related to foreign subsidiaries and branches, i.e. investments in greenfield or non-­greenfield operations in which the company has a total or partial stake. We consider the number of foreign subsidiaries (Delios and Beamish 1999; Lu and Beamish 2004), the value of capital invested in them, and their geographical location.6 In contrast to other studies based on stage theory, in our analysis we do not consider firm age. This is justified by our sample being constituted of medium-­ sized firms which have already passed through the initial stages of growth and transformation. For this reason the information in the dataset relating to age is not completely reliable; in many cases, set-­up date refers to the transformation from a partnership to a joint-­stock company. Also, in our study we consider firms which by definition have already reached a sufficient degree of organizational structuring; for this reason, age difference is assumed to be less relevant than for small and newborn firms. In the next section we discuss the results of our empirical analysis. The anal­ysis is based mainly on descriptive statistics and comparison of internationalization patterns observed in the first and last years of the period considered. The use of descriptive statistics instead of appropriate econometric techniques reduces the possibility of testing the theoretical hypotheses proposed in Section 2. Nevertheless, the methodology used is of value as a first characterization of the evolution of the internationalization patterns of medium-­sized firms and its accordance with stage-­theory predictions. Note also that we consider the Table 8.1  Companies by sector (Pavitt) and size, 2001 Class of employees

    Total

250–499

500–2,500

Scale intensive Science based Specialized supplier Supplier dominated

  51   5   41   86

10  6 16 27

               

Total

183

59

    242

  61   11   57 113

196   D. Iacobucci and F. Spigarelli population of firms within the defined size limits and for this reason descriptive statistics give a reliable picture of the actual size of the internationalization strategies carried out by those firms.

4  Results 4.1  Export and FDI intensity The following tables present some descriptive statistics on export intensity (Table 8.2) and the importance of foreign subsidiaries, in terms of weight in total fixed assets (Table 8.3). Both tables use the Pavitt (Pavitt 1984) categories and size classes. The sample is characterized by a high and growing export intensity. In both 2001 and 2005, the highest level of export activity is among smaller companies in the specialized supplier sector, and bigger companies in the scale intensive sector. Table 8.2 shows that 250 employees is a sufficient size to eliminate any disadvantages in export performance; in fact, when the whole sample is considered, there are no significant differences in the export capacities of these two sub-­classes of companies. However, Table 8.2 also shows that there are significant differences across sectors, in terms of export intensity and in the relationship between size and export performance. In the scale intensive and supplier dominated sectors we observe a positive relationship between size and export intensity, while in the science based and specialized supplier sectors this relationship is negative. Overall, it can be seen that there is no simple direct relation between firm size and export intensity, confirming Hypothesis 1. However, firm size becomes important when we consider the capacity of firms to invest abroad and there are significant differences in both the importance of the value of foreign investments in total fixed assets (Table 8.3) and the average value of investments per unit abroad (Table 8.4) between the two size classes. The exception is firms in the scale intensive sector which show the lowest level of FDI in total assets and no significant difference in this indicator by size Table 8.2  Export intensity (exports on sales) by sector and size (percentage values) 2001

2005

Class of employees 250–499

500–2,500

Scale intensive Science based Specialized supplier Supplier dominated

33.5 42.5 45.9 29.2

44.7 38.0 28.4 34.8

Total

34.5

35.0

Total █ █  firms

█

Class of employees

Total █ firms

250–499

500–2,500

35.3 40.0 41.0 30.5

32.7 43.5 45.2 33.5

45.2 34.7 33.3 37.3

34.8 38.7 41.9 34.4

34.6

36.2

37.3

36.4

Internationalization in Italian firms   197 classes. Comparing with the data in Table 8.2 on export intensity, it seems that firms in the scale intensive sector base their expansion on penetration of foreign markets through exporting rather than direct investment. This strategy is coherent with the exploitation of scale economies by concentrating production activities in one location and selling products in foreign markets. A further indication of this strategy is that firms in the scale intensive sectors have the lowest average value for unitary foreign investment (see Table 8.4); this means that foreign subsidiaries are devoted mainly to commercial rather than production activity. Specialized suppliers and science based firms in both size classes show the highest tendency to invest abroad, but with significantly lower values of foreign assets in total investments for smaller firms. However, over the period the differences between the two groups reduce slightly, based on higher and faster growth in the number and value of foreign investments by the smaller companies. The high values of unitary investments by firms in the science based and specialized supplier sectors (see Table 8.4) confirms the first part of Hypothesis 4, that hightech firms tend to follow a resource seeking approach by establishing foreign subsidiaries in countries where high value assets are available. For total number of foreign subsidiaries, there is a noticeable increase in the period in terms of both units (from 529 units in 2001 to 707 units in 2005) and capital invested (from €1,488 million in 2001 to €2,231 million in 2005). This Table 8.3  Value of foreign subsidiaries in total fixed assets (percentage values) 2001

2005

Class of employees 250–499

500–2,500

Scale intensive Science based Specialized supplier Supplier dominated

5.8 7.4 7.9 6.3

  4.9 17.6 19.3 10.3

Total

6.5

12.7

Total    █ firms

█

Class of employees

Total █  firms

250–499

500–2,500

  5.6 13.0 11.1   7.3

  7.4 15.9   9.4   9.4

  5.1 20.2 21.9 11.3

  7.0 18.3 12.9   9.8

  8.1

  9.0

14.0

10.2

Table 8.4  Average value of foreign subsidiaries by sector and size (million euros) 2001

2005

Class of employees 250–499

500–2,500

Scale intensive Science based Specialized supplier Supplier dominated

1.5 3.0 2.1 1.7

1.7 4.2 5.3 2.9

Total

1.8

3.5

Total  firms █

█

Class of employees

Total █  firms

250–499

500–2,500

1.7 3.6 3.0 2.0

1.9 4.0 3.2 2.5

2.0 4.7 6.3 3.5

2.0 4.4 4.1 2.7

2.2

2.5

4.2

2.9

198   D. Iacobucci and F. Spigarelli confirms the second research hypothesis that, over time, SMEs become more confident in foreign markets and develop more complex internationalization strategies based on a multiple presence abroad. If we look at size classes, we note an interesting phenomenon. While smaller firms show lower values of assets invested abroad, their increased presence in foreign markets is remarkable: +42 percent (from 325 to 462 units) compared to +20 percent for bigger firms (from 204 to 245 units). In general, this demonstrates the attitude and high propensity of medium-­sized companies to invest actively abroad, and their increasing role in globalization. Size affects the “magnitude” and intensity of investments in foreign markets, but not the willingness to develop internationalization activities. 4.2  Internationalization patterns To examine the patterns of internationalization among the firms in our sample, we classified them according to the different possible phases in their internationalization process by combining the two main variables used in the analysis: export intensity and the presence of foreign investment. The typologies of internationalization strategies followed by firms are illustrated in Table 8.5. Domestic firms are those that focus only or mainly on the national market to sell their products: they have at maximum an export intensity of 30 percent of their turnover. Export intensive firms on the contrary are extremely focused on an exclusively export based strategy, selling more than 30 percent of their turnover abroad. These companies do not own overseas units. Pioneer firms are so called because their internationalization strategy focuses on production or on commercial units abroad: they have spent abroad through FDIs, but have a reduced level of exports (less than 30 percent of turnover). In their case, FDIs have replaced exports. Internationalized firms are those that have maintained high export intensity alongside involvement in production and/or commercial units abroad. They adopt both modes of entry: FDIs support selling activities abroad. Table 8.6 presents the distribution of companies according to the above classification for 2001 and 2005. At the beginning of the period a quarter of companies can be considered to be domestic oriented, with low export intensity and no investments abroad, while almost a third can be classified as internationalized according to our typologies. Table 8.5 Typologies of internationalization strategy and patterns Export intensity

Foreign investment No

Yes

Low (≤ 30%)

Domestic

Pioneer

High (> 30%)

Export intensive

Internationalized

Internationalization in Italian firms   199 Table 8.6  Distribution of companies by type and size (percentage values) 2001

2005

Class of employees 250–499 Domestic Export intensive Pioneer Internationalized Total

500–2,500

Total █

█

Class of employees 250–499

500–2,500

Total █

26.2 22.4 20.2 31.2

20.3 15.3 27.1 37.3

24.8 20.7 21.9 32.6

24.0 16.4 21.3 38.3

18.6 11.9 27.1 42.4

22.7 15.3 22.7 39.3

100.0

100.0

100.0

100.0

100.0

100.0

This is depicted in the transition matrix in Table 8.7. The transition matrix is constructed by relating the typology of internationalization patterns observed in 2005 to those obtaining in 2001. The diagonal of the matrix identifies those firms whose pattern has not changed during the period; cells outside the diagonal identify firms whose internationalization patterns changed during the period. The most dynamic typology is export intensive, where 36 percent of companies moved to the internationalized category. In the case of the other typologies, the dynamics during the period of observation are less relevant, with some 80 percent of firms remaining in the same class.7 This suggests that investing abroad tends to complement, not substitute for, export, which is in line with our fifth hypothesis based on stage theory. A significant percentage of firms moved from the domestic to the pioneer category, which contradicts the stage theory hypothesis that this class of firms will adopt an export intensive strategy before embarking on foreign investment. We should highlight the transition pattern for pioneer firms between 2001 and 2005. The large majority of these companies (83 percent) remained in the same typology while those changing status were split between domestic and fully internationalized. This seems to suggest that the “pioneer” strategy does not identify a transient status (as stage theory would suggest) but a specific mode of entry for a specific class of firms. These results are also confirmed when we consider only the smaller firms (250–499 employees). Also, it is interesting to note that the dynamics of internationalized firms are determined exclusively by this size class, while larger Table 8.7  Transition matrix of types between 2001 and 2005 (percentage values) 2005 → Domestic

Export intensive Pioneer Internationalized

Total

5.0 56.0 0.0 7.6

100 100 100 100

2001 Domestic Export intensive Pioneer Internationalized

78.3 8.0 7.6 0.0

11.7 0.0 83.0 5.1

5.0 36.0 9.4 87.3

200   D. Iacobucci and F. Spigarelli companies (500–2,500 employees) remain fairly stable when they become internationalized. This means that size is relevant to explaining not only the propensity to invest abroad, but also the probability of this strategy succeeding. This result is in line with the fundamental hypothesis in stage theory (Hypothesis 2) that firms are not expected to develop complex forms of internationalization (such as FDI) before they have adequate experience of foreign markets, acquired through exporting. 4.3  Location of foreign subsidiaries Table 8.8 shows the number of foreign subsidiaries and the amount of capital invested in them, in 2001 and 2005. Two points emerge from this table: a) the great importance of the EU as an FDI location, in terms of both number of subsidiaries and capital invested; and b) the growth of foreign investment generally (in terms of both number and amounts) in all areas, and especially the countries of Eastern Europe, East and South Asia (China in particular) and North America (specifically the USA). The increase in the number and amount of FDI (from 529 subsidiaries and €1.5 billion in 2001 to 707 subsidiaries and €2.2. billion in 2005) is significantly higher than the growth in export activity. This is coherent with Hypothesis 2 that, as time passes, firms tend to develop FDI based strategies. The concentration of investment in EU countries is in line with Hypothesis 3, which states that SMEs tend to invest in nearby mature markets, following a market-­seeking strategy. It also reveals a “globalization gap” in Italian medium-­ sized companies (and Italian companies in general) (Mariotti and Mutinelli 2005). Considering the geographical and cultural proximity of the EU countries and, even more importantly, the absence of any kind of barriers to trade and capital movements, the EU should be categorized as a domestic rather than a foreign market. Table 8.8 Number of foreign subsidiaries and amount of capital invested (million euros), by area 2001 No. %

Capital % invested

European Union Eastern Europe Other European Countries Middle East Africa North America Centre and South America East and South Asia Oceania

337   13   21   1   11   55   37   47   7

63.7 1,215 2.4 10 4.0 7 0.2 0 2.1 53 10.4 106 7.0 35 8.9 60 1.3 2

Total

529 100.0 1,488

2005 █ No. %

81.6 397 0.7 38 0.5 26 0.0 6 3.6 16 7.1 73 2.4 45 4.0 96 0.1 10

Capital % invested

56.2 1,649 5.4 18 3.7 12 0.8 1 2.3 80 10.3 199 6.4 128 13.5 136 1.4 8

100.0 707 100.0

73.9 0.8 0.5 0.0 3.6 8.9 5.8 6.1 0.4

2,231 100.0

Internationalization in Italian firms   201 According to stage theory (Hypothesis 3), we would expect the size of firms to be related to their geographical span of operations. We would also expect that firms investing in far off markets to have some experience of foreign investment in less distant destinations (Hypothesis 2). To test these hypotheses we divided the foreign country destinations into two areas: near, including the EU, Eastern Europe, Africa and Middle East; and far, including North and South America, Australia, East and South Asia.8 Table 8.9 shows the distribution of companies in 2001 according to the presence of foreign subsidiaries in the above defined two areas. The low percentage of firms (in both size classes) with subsidiaries only in the more distant areas compared with those with investments in both areas, demonstrates that it is unusual for firms to start their internationalization process by investing in far off markets. For small firms the problems involved in investing abroad are related to distance; for larger firms the disadvantages of distance are significant only in the case of far off countries.9 Table 8.10 shows the transition matrix for the span of internationalization between 2001 and 2005. Overall, the movements recorded in the table confirm the stage theory hypothesis that, over time, foreign investment will move from near to more distant locations as companies acquire the knowledge and capabilities required to manage internationalization. The biggest movements are from domestic to near areas, and from near areas to other near and to far off areas. In fact, 25.5 percent of non-­internationalized companies in 2001 appear to be in Table 8.9 Companies by geographical span of subsidiaries and size, 2001 (percentage values) Class of employees 250–499 Non internationalized Near areas Far off areas Both near and far off areas Total

Total

500–2,500

49.2 25.1 4.4 21.3

33.9 27.1 3.4 35.6

45.5 25.6 4.1 24.8

100.0

100.0

100.0

Table 8.10 Transition matrix in the geographical span of internationalization between 2001 and 2005 (percentage values) 2001

2005 → NonNear Faraway Both near and Total Absolute internation areas areas faraway areas value

Non-internationalized Near areas Faraway areas Both near and far off areas

74.6 12.9 20.0 –

13.6 58.1 –   6.7

  3.6   3.2 60.0   5.0

  8.2 25.8 20.0 88.3

100 100 100 100

110   62   10   60

202   D. Iacobucci and F. Spigarelli foreign markets in 2005, moving to near or/and far off areas. In the meantime, 25.8 percent of companies focused on near areas in 2001 expanded their activities in other geographical areas in 2005. In the case of East Asian countries, and China in particular, we should take into account that during the period considered, there was a strong pull effect, due to the high rates of growth experienced by those countries and the policy incentives designed by their governments to attract foreign investments.

5  Discussion and conclusions We have examined the process of internationalization in a sample of medium-­sized firms in the NEC “third Italy” regions. In the empirical analysis we compared the evolution of the internationalization patterns observed in medium-­sized firms with hypotheses derived from stage theory. Our main findings are as follows. 5.1  Export intensity Although there are significant differences among sectors, we found no simple direct relation between firm size and export intensity. Within the size limits considered in our sample (250–2,500 employees) size does not affect average export capacity (which is particularly high in our sample): we found high values for sales abroad for both smaller (250–499 employees) and bigger (500–2,500 employees) companies. 5.2  FDIs The size of firms becomes important when considering the capacity to invest abroad. In fact, values of FDI on total fixed assets, and average value of investments per unit abroad are considerably higher for bigger firms. Specialized suppliers and science based firms show the highest average values of foreign investments. As demonstrated by other empirical studies, high-­tech companies are more prone to invest worldwide, seeking strategic resources. In terms of the total number of foreign subsidiaries, we found a noticeable increase in the period for all classes and sectors, in both units and capital invested, which is in line with stage theory, which states that as time passes, companies become more confident in foreign markets and tend to develop a more complex internationalization strategy based on a multiple presence abroad. Smaller companies show faster growth in the number of foreign subsidiaries. Although size affects the magnitude and intensity of investments in foreign markets, small firms seem to play an increasing role in internationalization. 5.3  Patterns of export and FDI We found that, in general, investing abroad tends to complement, not substitute for, export. The strongest dynamics between 2001 and 2005 are observed in

Internationalization in Italian firms   203 firms that were export intensive in 2001 and that became fully internationalized during the period through significantly increased FDI. This confirms that FDI is the next step in an international strategy based initially on export (first stage in  the internationalization process) and then on FDI as well. This is coherent with the hypothesis in stage theory, according to which firms are not expected to develop complex forms of internationalization (such as FDI) before acquiring good experience of foreign markets through exporting. Some of our evidence does not accord with the stage theory hypotheses. Firms that from the start embarked on more complex forms of internationalization (here called “pioneers”) are not expected to have a successful internationalization strategy according to stage theory. This is only partially confirmed by our data, which show that the majority of pioneers remain in this stage. Also, we found that among companies that were not internationalized in 2001, the majority had changed their status and jumped directly to FDI in 2005, apparently contradicting stage theory. 5.4  Geographical localization of FDIs Most FDIs are in the EU countries, i.e. in nearby, mature markets. As observed in other studies of Italian companies, our sample confirms the presence of a sort of “globalization gap”, as the EU should more accurately be considered a unique “domestic” market. Analyzing the geographical span of operations, we found that FDI moved through time from closer to more distant locations as companies acquired the knowledge and capabilities to manage these more complex internationalization strategies, confirming a “process approach” to internationalization. Overall, our results provide only partial support for the stage theory of internationalization. This could be due to our methodology, which does not allow us to control for structural variables influencing the behavior of firms. On the other hand, it could be that the acceleration in technology and market changes, together with a more unstable global environment, require a more complex and eclectic model to predict the behavior of firms in terms of their internationalization strategies. Compared to other empirical studies on the Italian situation, our empirical analysis has some distinctive features: we use different measures of internationalization that take account of both export activity and FDI. In examining FDI, we analyze the number of foreign subsidiaries and also their value and geographical location. We consider a five year period to evaluate the evolution of internationalization patterns. However, this study also has some limitations, which we intend to address in future work. Our sample was defined based on specific size measures (number of employees and turnover). As there is no consensus on what constitutes a medium-­sized company, our findings are influenced by the range we chose. We based our measures of internationalization on data available from companies’ annual reports. We therefore do not include measures of internationalization that

204   D. Iacobucci and F. Spigarelli are not “accounting” sensitive, such as non-­equity joint ventures or other types of strategic alliances. This is a major limitation, as these modes of entry are becoming increasingly common for small- and medium-­sized companies. Direct interviews with firms would be the only way to overcome this problem. The empirical analysis is based on descriptive statistics; we need to increase the number of our observations as well as the number of internal and structural variables in order to test the same hypotheses using multivariate analysis.

Notes 1 See the literature reviews in Malhotra et al. (2003) and Cumberland (2006). 2 “Outward FDI by SMEs generally occurs after successful experience gained in exporting and/or forming alliances. The ENSR 2003 survey showed that only 3% of SMEs in Europe have subsidiaries, branches or joint ventures in other countries” (Wilson 2007: 52). 3 A group is a set of companies legally independent, but belonging to the same owner. 4 Pavitt (1984) classifies industries into four sectors according to the innovation regimes characterizing them: supplier dominated, scale intensive, specialized supplier, and science based. 5 Use of the foreign to total sales ratio is widespread (Czinkota and Johnston 1983; Grant 1987; Grant et al. 1988; Geringer et al. 1989; Calof 1993; Tallman and Li 1996; Wolff and Pett 2000; Yu-­Ching et al. 2006). Some authors (Tallman and Li 1996) point out that it does not capture the influence of internal (intra-­corporate) transfers. Bartlett and Ghoshal (1989) suggest using it specifically to map the initial stages of internationalization of firms located in developed countries. Other studies adopt different measures for internationalization strategy, such as: ratio of foreign assets to total assets (Daniels and Bracker 1989); number of foreign countries in which a firm has an operating subsidiary (Tallman and Li 1996) (Lu and Beamish 2004); number of overseas employees to total number of employees (Kim et al. 1989); number of foreign investments and number of countries in which FDI are located (Delios and Beamish 1999); entropy index weighted by foreign sales (Kim et al. 1993) (Hitt et al. 1997). Some authors use more than one measure (Gomes and Ramaswamy 1999). Sullivan (1994) built a single multidimensional indicator but was criticized (Ramaswamy et al. 1996) for lack of validity. 6 In addition to number of foreign subsidiaries, Lu and Beamish (2004) used the number of countries in which firms had overseas subsidiaries. They combined these two measures as suggested by Sanders and Carpenter (1998), to build a complex measure of internationalization. They divided each of the two count measures by either the maximum number of FDIs or the maximum number of FDI countries in the sample, to transform them from counts to ratios. They then computed the average of the two ratios to give a final measure of internationalization in the range 0 to 1, with 1 being the highest level of internationalization. 7 We also classified companies by the same typologies, using a different indicator for foreign investment, i.e. the weight of foreign investment in total fixed assets, considering a cut off of 5 percent. We obtained substantially the same results in terms of company distribution and evolution over time. 8 The division is mainly based on geographic distance. This does not always equate to cultural distance or other types of barriers to trade. Thus, it is a rough approximation of the complex concept of “distance” in stage theory. We also made a distinction between EU and non-­EU countries and obtained the same results. 9 We should also remember that the majority of the companies in nearby countries are in EU countries, which, as already mentioned, for Italian firms should realistically be considered domestic markets.

Internationalization in Italian firms   205

References Andersen, O. (1993), On the internationalization process of firms: a critical analysis, Journal of International Business Studies, 24: 209–231. Anderson, R.C., T.W. Bates, J.M. Bizjak, and Lemmon, M.L. (2000), Corporate governance and firm diversification, Financial Management, 29: 5–22. Anderson, S. (2000), The internationalisation of the firm from an entrepreneurial perspective, International Studies of Management and Organization, 30: 63–92. Andersson, S. and Wictor, I. (2003), Innovative internationalisation in new firms: born globals – the Swedish case, Journal of International Entrepreneurship, 1(3): 249–75. Apfelthaler, G. (2000), Why small enterprises invest abroad: the case of four Austrian firms with U.S. operations, Journal of Small Business Management, 38: 92–8. Autio, E., Sapienza, H.J. and Almeida, J. (2000), Effects of age at entry knowledge intensity and imitability on international growth, Academy of Management Journal, 43: 909–24. Balloni, V. and Iacobucci, D. (2004), The role of medium-­sized and large firms in the evolution of industrial districts. The case of Marche, in G. Cainelli and R. Zoboli (eds.), The Evolution of Industrial Districts, Heidelberg, Physica-­Verlag. Bartlett, C.A. and Ghoshal, S. (1989), Managing across Borders: The Transnational Solution, Boston, Harvard Business School Press. Bell, J. (1995), The internationalization of small computer software firms: a further challenge to stage theories, European Journal of Marketing, 29(8): 60–75. Bell, J., McNaughton, R. and Young, S. (2001), Born again global firms. An extension to the born global phenomenon, Journal of International Management, 7: 173–89. Belso-­Martines, J. (2006), Why are some Spanish manufacturing firms internationalizing rapidly? The role of business and institutional international networks, Entrepreneurship & Regional Development, 18: 207–26. Bjorkman, I. and Forsgren, M. (2000), Nordic international business research, International Studies of Management & Organization, vol. 30(1): 6–25. Blonigen, B.A. (2001), In search of substitution between foreign production and exports, Journal of International Economics, vol. 53(1): 81–104. Bonaccorsi, A. (1992), On the relationship between firm size and export intensity, Journal of International Business Studies, 4: 605–35. Brioschi, F., Brioschi, M.S. and Cainelli, G. (2002), From the industrial district to the district group: an insight to the evolution of local capitalism in Italy, Regional Studies, 36(9): 1037–52. Buckley, P.J., Clegg, L.J., Cross, A.R., Xin, L., Voss, H. and Ping, Z. (2007), The determinants of Chinese outward foreign direct investment, Journal of International Business Studies, 38(4): 499–518. Bugamelli, M.P., Cipollone, P. and Infante, L. (2000), L’internazionalizzazione delle imprese italiane negli anni ’90, Rivista italiana degli economisti 5(3), 349–386. Burgel, O. and Murray, G.C. (2000), The international market entry choices of start-­up companies in high-­technology industries, Journal of International Marketing, 8(2): 30–63. Cainelli, G. and Zoboli, R. (2004) (eds.), The Evolution of Industrial Districts. Changing Governance, Innovation and Internationalization of Local Capitalism in Italy, Heidelberg, Physica-­Verlag. Calof, J.L. (1993), The impact of size on internationalization, Journal of Small Business Management, 31: 60–9.

206   D. Iacobucci and F. Spigarelli Calof, J.L. (1994), The relationship between firm size and export behavior revisited, Journal of International Business Studies, 25: 367–87. Cavusgil, S.T. (1980), On the internationalisation process of firms, European Research, 8(6): 273–81. Cavusgil, S.T. (1984), Changing Nature of Firms International Market Research Activities During the Internationalization Process, New York, Praeger. Chetty, S.K. and Campbell-­Hunt, C. (2003), Paths to internationalisation among small-­to medium-­sized firms. A global versus regional approach, European Journal of Marketing, 37: 796–820. Clausing, K.A. (2000), Does multinational activity displace trade? Economic Inquiry, 38(2): 190–205. Coltorti, F. (2004), Le Mi industriali italiane: nuovi aspetti economico-­finanziari, Economia e Politica Industriale, 31: 5–25. Coviello, N.E. and Martin, K. (1999), Internationalization of service SMEs: an integrated perspective from the engineering consulting sector, Journal of International Marketing, 7(4): 42–66. Coviello, N.E. and Munro, H.J. (1997), Network relationships and the internationalization process of small software firms, International Business Review, 6(4): 361–86. Coviello, N.E. and McAuley, A. (1999), Internationalisation and the smaller firm: a review of contemporary empirical research, Management International Review, 39(3), pp. 223–56. Cumberland, F. (2006), Theory development within international market entry mode. An assessment, Marketing Review, 6: 349–73. Czinkota, M.R. and Johnston, W.J. (1983), Exporting: does sales volume make a difference? Journal of International Business Studies, 14: 147–53. Dalli, D. (1995), The organization of exporting activities: relationships between internal and external arrangements, Journal of Business Research, 34: 107–15. Daniels, J.D. and Bracker, J. (1989), Profit performance: do foreign operations make a difference?, Management International Review, 29(1): 46–56. Deeds, D.L. and Hill, C.W.L. (1998), An examination of opportunistic action within research alliances: evidence from the biotechnology industry, Journal of Business Venturing, 14: 141–63. Delios, A. and Beamish, P.W. (1999), Ownership strategy of Japanese firms: transactional, institutional, and experience influences, Strategic Management Journal, 20: 915–33. Dunning, J.H. (1988), Multinationals, Technology, and Competitiveness, London/ Boston, Unwin Hyman. Dunning, J.H. (1993), Multinational Enterprise and the Global Economy, Reading, Addison Wesley. Dunning, J.H. (2000), The eclectic paradigm as an envelope for economic and business theories of mne activity, International Business Review, 9: 163–90. Erramilli, M.K. and D’Souza, D.E. (1993), Venturing into foreign markets: the case of the small service firm, Entrepreneurship: Theory & Practice, 17: 29–41. Erramilli, M.K. and Rao, C.P. (1993), Service firms’ international entry-­mode choice: a modified transaction-­cost analysis approach, Journal of Marketing, 57: 19–38. Fischer, E. and Reuber, A. (1997), The influence of the management team’s international experience on the internationalization behaviors of SMEs, Journal of International Business Studies, 28: 807–25. Gankema, H.G.J., Snuif, H.R. and Zwart, P.S. (2000), The internationalisation process of small and medium sized enterprises: an evaluation of stage theory, Journal of Small Business Management, 38: 15–27.

Internationalization in Italian firms   207 Geringer, J.M., Beamish, P.W. and daCosta, R.C. (1989), Diversification strategy and internationalization: implications for MNE performance, Strategic Management Journal, 10: 109–19. Gomes, L. and Ramaswamy, K. (1999), An empirical examination of the form of the relationship between internationalization and performance, Journal of International Business Studies, 30: 173–88. Graham, E.M. (2000), The relationships between trade and foreign direct investment in the manufacturing sector: empirical results for the United States and Japan, in D. Encarnation (ed.), Does Ownership Matter? Japanese Multinationals in East Asia London, Oxford University Press. Grant, R.M. (1987), Multinationality and performance among British manufacturing companies, Journal of International Business Studies, 18: 79–89. Grant, R.M., Jammine, A.P. and Thomas, H. (1988), Diversity, diversification, and profitability among British manufacturing companies 1972–84, Academy of Management Journal, 31: 771–801. Helpman, E., Melitz, M.J. and Yeaple, S.R. (2004), Export versus FDI with heterogeneous firms, American Economic Review, 94: 300–16. Hitt, M.A., Hoskisson, R.E. and Kim, H. (1997), International diversification: effects on innovation and firm performance in product-­diversified, Academy of Management Journal, vol. 40(4), 767–98. Hurmerinta-­Peltomaki, L. (2003), Time and internationalisation: theoretical challenges set by rapid internationalisation, Journal of International Entrepreneurship, 1: 217–36. Ibeh, K., Johnson, J.E. Dimitratos, P. and Slow, J. (2004), Micromultinationals: some preliminary evidence on an emergent star of the international entrepreneurship field, Journal of International Entrepreneurship, 2: 289–303. Johanson, J. and Vahlne, J.E. (1977), The internationalization process of the firm – Four Swedish case studies, Journal of Management Studies, 12: 305–22. Johanson, J. and Vahlne, J.E. (1992), Management of foreign market entry, Scandinavian International Business Review, 1: 9–27. Jones, M. (1991), The internationalisation of small high-­technology firms, Journal of International Marketing, 7: 15–41. Kim, W.C., Hwang, P. and Burgers, W.P. (1989), Global diversification strategy and corporate profit performance, Strategic Management Journal, 10: 45–57. Kim, W.C., Hwang, P. and Burgers, W.P. (1993), Multinationals’ diversification and the risk-­return trade-­off, Strategic Management Journal, 14: 275–86. Lu, J.W. and Beamish, P.W. (2001), The internationalization and performance of Smes, Strategic Management Journal, 22: 565–86. Lu, J.W. and Beamish, P.W. (2004), International diversification and firm performance: the S-­curve hypothesis, Academy of Management Journal, 47: 598–609. Malhotra, N.K., Ulgado, F.M. and Agarwal, J. (2003), Internationalization and entry modes: a multitheoretical framework and research propositions, Journal of International Marketing, 11: 1–31. Mariotti, S. and Mutinelli, M. (2004), Small multinational groups in the Italian industrial districts: interpretation and empirical evidence from the mechanical engineering industry, in G. Cainelli and R. Zoboli (eds.), The Evolution of Industrial Districts. Changing Governance, Innovation and Internationalization of Local Capitalism in Italy Heidelberg, Physica-­Verlag. Mariotti, S. and Mutinelli, M. (2005). Italia Multinazionale 2005. Le Partecipazioni Italiane All’estero Ed Estere in Italia. Roma, Istituto Nazionale per il Commercio Estero.

208   D. Iacobucci and F. Spigarelli McDougall, P.P. and Oviatt, B.M. (2000), International entrepreneurship: the intersection of two paths, Academy of Management Journal, 43: 902–6. McDougall, P.P., S. Shane, and Oviatt, B.M. (1994), Explaining the formation of international new ventures: the limits of international business research, Journal of Business Venturing, 9: 469–87. Melin, L. (1992), Internationalization as a strategic process, Strategic Management Journal, 13: 99–118. Menghinello, S. (2004), Local engines of global trade: the case of Italian industrial districts, in G. Cainelli and R. Zoboli (eds.), The Evolution of Industrial Districts. Changing Governance, Innovation and Internationalization of Local Capitalism in Italy Heidelberg, Physica Verlag. Moen, A. and Servais, P. (2002), Born global or gradual global? Examining the export behavior of small and medium-­sized enterprises, Journal of International Marketing, 10: 49–72. Oviatt, B.M., and McDougall, P.P. (1994), Toward a theory of international new ventures, Journal of International Business Studies, 25: 45–64. Pavitt, K. (1984), Sectoral patterns of technical change: towards a taxonomy and a theory, Research Policy, 13: 343–73. Ramaswamy, K., Kroeck, K.G. and Renforth, W. (1996), Measuring the degree of internationalization of a firm: a comment, Journal of International Business Studies, 27: 166–76. Rialp, A., Rialp, J. Urbano, D. and Vaillant, Y. (2005), The born-­global phenomenon: a comparative case study research, Journal of International Entrepreneurship, 3: 133–71. Sanders, W.M.G. and Carpenter, M.A. (1998), Internationalization and firm governance: the roles of CEO compensation, top team composition, and board structure, Academy of Management Journal, 41: 158–78. Shan, W. and Song, J. (1997), Foreign direct investment and the sourcing of technological advantage: evidence from the biotechnology industry, Journal of International Business Studies, 28: 267–84. Shatz, J.H. and Venables, A.J. (2000), The geography of international investment, in Clark, G.L., Fedlman, M.P. and Gertler, M.S. (eds.) The Oxford Handbook of Economic Geography, Oxford, Oxford University Press, pp. 125–45. Sullivan, D. (1994), Measuring the degree of internationalization of a firm, Journal of International Business Studies, 2: 325–42. Svetlicic, M., Jacklic, A. and Burger, A. (2007), Internationalization of small and medium-­size enterprises from selected central European economies, Eastern European Economics, 45: 36–65. Tallman, S. and Li, J. (1996), Effects of international diversity and product diversity on the performance of multinational firms, Academy of Management Journal, 39: 179–96. Wickramasekera, R. and Oczkowski, E. (2004), Key determinants of the stage of internationalisation of Australian wineries, Asia Pacific Journal of Management, 21: 425–44. Wilson, K. (2007), Encouraging the internationalisation of SMEs, in J. Potter and A. Proto (eds.), Promoting Entrepreneurship in South East Europe. Policies and Tools (Vol. 6) Paris, OECD. Wolff, J.A. and Pett, T.L. (2000), Internationalization of small firms: an examination of export competitive patterns, firm size, and export performance, Journal of Small Business Management, 38: 34–47.

Internationalization in Italian firms   209 Yip, G., J. Biscarri, and Monti, J. (2000), The role of internationalisation process in the performance of newly internationalised firms, Journal of International Marketing, 8: 10–35. Yu-­Ching, C., Y. Kuo-­Pin, and Yu, C.-M.J. (2006), Performance, internationalization, and firm-­specific advantages of SMEs in a newly-­industrialized economy, Small Business Economics, 26: 475–92. Zahra, S.A., D.O. Neubaum, and Huse, M. (1997), The effect of the environment on the firm’s export intensity, Entrepreneurship: Theory & Practice, 22: 25–46.

9 Production offshoring and the skill composition of Italian manufacturing firms A counterfactual analysis1 Roberto Antonietti and Davide Antonioli 1  Introduction Since the early 1980s the way goods are manufactured has dramatically changed. A new international division of labour is emerging that is characterized by firms fragmenting the entire value chain – from product design to assembling and distribution – into modules that are moved to different locations on a global scale in order to exploit localization advantages and factor costs differentials. The fall in trade barriers and transportation costs, the rapid diffusion of information and communication technologies, the recent economic transformations occurred in Eastern Europe countries, and the emergence of new ‘powers’ like Brazil, Russia, India and China, have been responsible for the recent increase in the global fragmentation of production by Western economies. In this respect, trade statistics show a steady increase of intra-­industry trade flows between the European Union and the rest of the world in a relatively short period of time. In particular, from 1990 to 1997 and from 2002 onward, an increasing share of trade flows seems to characterize the most advanced economies, mostly in terms of intermediate and unfinished goods being transferred from one country to another in order to be processed (Mariotti and Mutinelli 2005; UNCTAD 2006; Baldone et al. 2007). One of the main consequences of this phenomenon is that, next to an extensive use of IT capital, imported materials and intermediate services, an increasing replacement of low-­skill employment is occurring due to the fact that many firms sub-­contract less knowledge-­intensive activities to cheap labour countries. Trade flows, import competition and foreign direct investment (FDI), thus, can result in a reorganization of production through which home firms can specialize on the high-­value-added phases of the value chain while economizing on production costs. The increasing fear of job losses, particularly referred to low-­skill intensive tasks and occupations (Amiti and Wei 2005), is making international fragmentation of production a ‘hot topic’ both for the media and for academic research. Traditionally, two main explanations have been given to account for the shift in demand away from low-­skilled workers in industrialized countries. The first refers to non-­neutral technological change that, by fostering the demand for

Production offshoring and skills – Italy   211 more qualified workers within technologically advanced industries, tends either to increase wage inequality or to increase the relative unemployment of less qualified workers. Specifically, highly skilled workers seem to benefit from higher wages in relatively flexible labour markets, like the ones characterizing the US and the UK economies, whereas in relatively more rigid ones, like Germany, France, Denmark and Italy, technological change seems to determine relatively higher rates of unemployment for low-­skilled workers.2 The second claims increased international trade and globalization of production, according to which labour is relocated in a way that determines a shift of redundant and routinized activities toward less-­developed countries, while keeping non-­routinized, high skill-­intensive activities at home, thus increasing the domestic firms’ comparative advantage in the production of high-­value added goods. Recent international evidence (Brainard and Litan 2004; Amiti and Wei 2005, 2009), however, shows also that the increasing digitization of production enables firms not only to offshore pure manufacturing processes, but also computing and business-­related services like software programming, medical diagnosis, lab research, product development and analytical activities, hence creating the conditions for the transfer of knowledge-­intensive jobs. However, one should note that, even if service offshoring is growing annually very fast, it is still at very low levels if compared to material offshoring.3 With this piece of work we aim at assessing if, and to what extent, the firm’s decision to offshore production to low-­wage countries alters the skill intensity of domestic employment. For this purpose, we conduct our empirical exercise in a framework close to a laboratory experiment, in which we employ a difference-­ in-differences propensity score matching (DID-­PSM) estimator in order to control for sample selection and unobserved heterogeneity among observations and across time, while avoiding any ad hoc restriction on the relations of interest. The chapter is structured as follows. Section 2 briefly reviews the literature inspecting the skill-­bias effects of international fragmentation of production. Section 3 describes the methodology adopted for the empirical analysis. Section 4 presents the dataset utilized. Section 5 offers some descriptive statistics on the outcome variables and, after testing for the validity of our method of estimation, provides the results of our counterfactual exercise. Finally, Section 6 concludes.

2  Background literature Even if it is considered a ‘hot topic’ for economists, the impact of globalization on the international division of labour and the employment dynamics of workers is still ambiguous. Next to traditional trade and technological change explanations, international fragmentation and outsourcing have been considered as factors potentially responsible for the rising income and employment differentials between skilled and unskilled workers (Feenstra and Hanson 1996; Egger and Falkinger 2003).

212   R. Antonietti and D. Antonioli However, the question whether the international relocation of production determines a change in the skill intensity of jobs is still unanswered, both in theory and in the evidence. While traditional trade models based on the Hecksher–Olin– Samuelson framework argue that the move of low-­skill intensive stages of production abroad decreases the demand for the relatively less abundant factor at home, other studies predict a more ambiguous impact of international outsourcing on low-­skilled workers in the source country, so that multiple outcomes may emerge (Arndt 1997; Glass and Saggi 2001(1); Jones and Kierzkowski 2001). For this purpose, the theoretical literature suggests that such an effect depends on which type of production or service activity is offshored, on the factor intensity of both the processes that remain in the home country and the ones that are relocated internationally (Kohler 2003; Egger and Egger 2003; Egger and Falkinger 2003), and, finally, on the sector in which offshoring occurs (Arndt 1997). Whether, and to what extent, production offshoring impacts the skill composition of a country’s workforce is a matter of empirical research. The empirical literature on the skill bias effects of international outsourcing can be divided into two main lines of research. A first set of studies takes multinational firms and FDI as the units of analysis, distinguishing between vertical and horizontal FDI (Markusen et al. 1996; Lipsey 2002). While the former is mainly driven by the will to exploit the differences in factors endowments and prices, and leads to a net decrease in domestic employment (Agarwal 1997; Braconier and Ekholm 2000; Mariotti et al. 2003), the latter is primarily driven by the will to replicate abroad the whole production process of the home country, with the aim of finding new markets and global opportunities and with the effect of increasing both the employment and the skill intensity of domestic jobs (Markusen et al. 1996; Blömstrom et al. 1997; Mariotti et al. 2003). However, if the literature generally agrees on the total employment effects of FDI, less explored is the issue of the effect of FDI on the skill composition of the workforce. The research question thus becomes: does investing in cheap-­ labour countries lead to a skill upgrading at home? Head and Ries (2002) try to answer by looking at Japanese multinationals in the period 1965–1990: their results point to a positive relationship between offshoring and the demand for skilled labour only if production relocation is directed to developing countries and only when the unit of analysis is the firm instead of the industry. Similarly, Hansson (2005) finds that production delocalization toward less developed countries contributes to the general increase in the average level of qualifications within Swedish multinationals. In contrast to these results, Slaughter (2000), looking at 32 US manufacturing industries in the 1980s, does not show clear results in favour of the positive relationship between FDI and the employment of skilled workers at home. For Italy, Barba Navaretti and Castellani (2004) and Castellani et al. (2006) find a skill upgrading effect of foreign investments by multinationals primarily due to the international relocation of low value-­added segments of the production process that leads to a lower demand for low-­skill labour at home. Using data on a sample of manufacturing firms over the period 1989–1997, Piva and

Production offshoring and skills – Italy   213 Vivarelli (2004), instead, do not find any significant effect of FDI on the skill composition of Italian employment, even if the nature of the data and of results does not exclude a priori any possible influence. A second group of studies, instead, takes an international production approach and considers offshoring as a form of international trade involving intermediate or unfinished goods and processes. According to Jones and Kierzkowski (2001), international fragmentation can be thought as a process of splitting up and spread of previously integrated stages of production over an international network of production sites. More specifically, ‘outsourcing’ refers to the relocation of jobs and processes to external providers regardless of their location, while ‘offshoring’ refers to the relocation of jobs and processes to any foreign country, without distinguishing whether the provider is external or affiliated with the firm.4 According to OECD (2007), the term offshoring designates two distinct situations: (i) production of goods or services partially or totally transferred abroad within the same group of firms, through foreign affiliates (offshore in-­house sourcing); or (ii) the partial or total transfer of the production of goods or ser­ vices abroad to a non-­affiliated unit (offshore outsourcing).5 Table 9.1 summarizes all the outsourcing and offshoring options previously described. The evidence available from the international trade literature provides general support for the skill-­biased nature of production relocation.6 Wood (1994), for instance, calculates that import competition determines a reduction in the demand for unskilled labour by 30 per cent in 1990. On the same line, Sachs and Shatz (1994) conclude that production internationalization exerts a double effect on overall labour composition: it is not only the cause of a general decrease in manufacturing but, together with technological change, is a determinant of the decline in the relative demand for low-­skilled workers. Moreover, Feenstra and Hanson (1996) provide some evidence that, for the period 1972–1990, international outsourcing is responsible for a 30–50 per cent rise in the demand for skilled workers and, thus, for a rise in income inequality. For the UK, Anderton and Brenton (1999) estimate that, between 1970 and 1986, imports from low-­wage countries determine a negative impact of about 40 Table 9.1  Production options for a firm Location

Internal production (in-house)

External production (outsourcing)

Within the country Production within the firm and Production outside the firm but (domestic) the country (domestic in-house) within the country (domestic outsourcing) Abroad (offshoring Production within the group to or cross-border) which the firm belongs but abroad (by its own affiliates) (offshore in-house sourcing)             ↑ Offshoring in the strict sense Source: adapted from OECD (2007, p. 16).

Production outside the firm and outside the country by non-affiliated firms (offshore outsourcing or subcontracting abroad).             ↑ Offshoring in the broad sense

214   R. Antonietti and D. Antonioli per cent on the wage-­bill share and relative employment of low-­skilled workers. This result is further reinforced by Hijzen et al. (2004), who show that, between 1982 and 1996, international outsourcing has a strong negative impact on the demand for semi-­skilled and unskilled labour. For France, Strauss-­Khan (2003) finds that the highly increasing vertical specialization, i.e. the share of imported inputs in production, is the main determinant of the sharp decline in the share of unskilled workers between 1977 and 1993, passed from –15 per cent in the period 1977–1985 to –25 per cent between 1985 and 1993. For Austria, instead, a positive and significant effect on skilled labour comes out only when using proxies of international trade like export openness and outsourcing, while a negative effect arises when considering import penetration (Dell’mour et al. 2000). For the Italian case, finally, the scanty evidence seems to support the positive relationship between offshoring and the relative demand for skilled labour. Helg and Tajoli (2005), for instance, compare the effect of international fragmentation of production on the skilled-­unskilled ratio in Italy and in Germany and show that a positive and significant impact emerges only for the former, while for the latter the effect seems to be not significant.7

3  Methodology In the following analysis we estimate the impact that the choice to segment and internationally relocate production activities exerts on the skill level of employment within Italian manufacturing firms. Differently from previous empirical literature,8 we do this by matching the outcome of a set of firms offshoring production at a certain moment in time to the outcome of a control group of firms (the so-­called counterfactual sample) that, although showing the same average characteristics, did not choose to move production abroad. For this purpose, we set up a quasi-­experiment in which we identify two groups of firms: one group is assigned into the treatment – i.e. the offshoring of previously integrated production activities – on the basis of a random process, while the other is constituted by a subset of the remaining firms that can be considered as similar to the treated except for the fact of not being assigned into treatment. The value added of such a methodology is twofold. First, it constitutes a relatively precise comparison exercise between treated and counterfactual firms: in this case, we do not compare our treated sample to all the (untreated) firms that do not offshore production, but we restrict the attention to only those firms that are supposed to be as similar as possible to treated firms and then can be thought of as potentially selectable for the assignment into treatment. Second, in so doing, we do not adopt any specification of the relations of interest, but, rather, we employ a non-­parametric technique that allow us to avoid any ad hoc restriction on parameters values. Since our data do not come from randomized trials but from observational surveys in which the assignment of firms to the treatment and control group is

Production offshoring and skills – Italy   215 not random, a problem of sample selection may occur due to the presence of confounding factors that can bias the estimation of the treatment effects. More precisely, sample selection concerns the presence of some characteristics of the treated (or control) group that are both associated with receipt of the treatment and with the outcome. This fact, in turn, may lead to biased treatment effects estimations, that is to a false attribution of causality regarding treatment and outcomes (Heckman 1990). In order to reduce the bias in the estimation of treatment effects we use a propensity score matching (PSM) estimator, thanks to which we control for the existence of such confounding factors by comparing the outcome of treated and control subjects that are as similar as possible on the basis of a set of pre-­ treatment characteristics X. Since it is rather unfeasible to match treated and control subjects on large sets of relevant variables, Rosenbaum and Rubin (1983) suggest using the so-­called balancing score b(X), i.e. a function of the relevant observed covariates X such that the conditional distribution of X given b(X) is independent of assignment into treatment. In the analysis we employ a method based on PSM that allows us to summarize pre-­treatment characteristics of each unit into a single-­index variable (the propensity score) which makes the matching feasible.9 Operationally, the propensity score can be defined as the probability of receiving a treatment conditional on a set of pre-­treatment characteristics: p ( X ) ≡ Pr( D = 1 | X ) = E ( D | X )

(9.1)

where D = (0, 1) represents the exposure to treatment – in this case the decision to offshore production – and X is a vector of pre-­treatment characteristics. Rosenbaum and Rubin (1983) demonstrate that, if the assignment to treatment is random within the cells defined by X, then it is also random within the cells defined by p(X). After estimating such a probability, the counterfactual sample is identified on the basis of the outcomes of those untreated units that show the same propensity score as the one of the treated. For the purpose of this chapter we are interested in estimating the average treatment effect on the treated units (ATT), i.e. the average difference in the outcome of treated firms with respect to their counterfactual. Given the propensity score p(X), the ATT can be estimated as follows: ATT = E(Y1 – Y0|D = 1) = E[E(Y1 – Y0|D = 1, p(X)] = E{E[Y1|D = 1, p(X)] – E[Y0|D = 0, p(X)]|D = 1}

(9.2)

where the outer expectation is calculated over the distribution p(X)|D = 1, and Y and Y0 represent the potential outcomes in the two possible situations of treated and untreated. Equation 9.2 simply states that the PSM estimator can be considered as the mean difference in outcome between treated (D = 1) and the untreated units (D = 0), appropriately weighted by the propensity score distribution of participants (p(X)|D = 1).

216   R. Antonietti and D. Antonioli In order to derive (9.2) given (9.1), and then assess the ATT, two basic hypotheses need to be satisfied. The first requires the independence between the participation of a subject into treatment and the outcome that would have occurred if the same subject were not assigned to the treatment. In other words, this balancing property requires that, after conditioning for all the characteristics X, the assignment into treatment or control groups is governed by a random process. Formally, this can be expressed as follows: D ^ X | p(X)

(9.3)

where ^ denotes independence, i.e. given a certain value of the propensity score, treatment status is independent on the characteristics of individuals. When this condition holds, it means that the balancing of pre-­treatment variables is satisfied given the propensity score, a fact occurring when observations with the same propensity score have the same distribution of observable and unobservable characteristics independently of treatment status. Put another way, for a given propensity score, exposure to treatment can be considered as random so that treated and control units should be on average identical, i.e. staying on the so-­called common support (Rosenbaum and Rubin 1983; Dehejia and Wahba 2002). The second assumption requires the treatment to be exogenous with respect to both the sets of characteristics and the outcomes. Different versions of this assumption have been proposed: unconfoundedness, selection on observables or conditional independence assumption. However defined, this assumption implies that systematic differences in outcomes between treated and counterfactual subjects are attributable to treatment. Formally, we can write it as follows: Y0, Y1 ^ D | p(X)

(9.4)

in which ^ means that, given the propensity score, potential outcomes are independent of treatment assignment. This implies that the researcher must be able to simultaneously observe all variables that affect treatment assignment and potential outcomes. Since it is plausible that unobservable factors may have influenced treatment assignment, a different strategy for the identification of the counterfactual is needed. For this purpose, we employ a difference-­in-differences (DID) specification of the PSM, with which we control for unobserved heterogeneity by removing all the time-­invariant differences in outcome between treated and control units that can persist even once conditioning on observable characteristics (Heckman et al. 1997, 1998; Blundell and Costa Dias 2002). Using this algorithm, we thus estimate the ATT by comparing the difference in the outcome between treated and control units before and after the assignment into treatment. However, the simple estimation of the propensity score is not enough to estimate the ATT as in equation 9.2, because, due to the continuous nature of p(X), the probability of observing two units with the same propensity to be

Production offshoring and skills – Italy   217 assigned to treatment is closed to zero. For this purpose, different methodologies have been developed in order to overcome this issue. In our analysis, we choose to implement the nearest neighbour matching (NNM) algorithm10 (Smith and Todd 2005; Caliendo et al. 2008; Caliendo and Kopeinig 2008), where each treated unit is matched with the counterfactual unit that shows the closest propensity score. We also apply this method with replacement, in the sense that we allow for counterfactual units to match more than one treated unit. Once each treated subject is matched with its counterfactual, the difference between the outcome of the treated units and the outcome of the matched control units is computed as follows using a DID specification: ATTDIDPSM =

1 N1

N1





N0

1

N1

1

N0

∑  ∆Y − ∑ w(i, j)∆Y  = N ∑ ∆Y − N ∑ w( j)∆Y i =1



1

0

j =1



1 i =1

1

1 j =1

0

(9.5)

where N1 is the number of treated firms, N0 is the number of counterfactual units, ΔY0 is the difference in the outcome before and after the treatment for counterfactuals, ΔY1 is the difference in outcome before and after the treatment for treated units, w(i, j) is the weight assigned to each comparison unit in the construction of the counterfactual outcome, and w(  j) =

∑ w(i, j) . i

In the NNM case, w(i, j) is equal to the inverse of the number of controls matched with a treated observation if observation j belongs to the set of control units and 0 otherwise.11

4  Data Our dataset consists in a sample of Italian manufacturing firms drawn from the VII, VIII and IX waves of the Survey on Manufacturing Firms (Indagine sulle Imprese Manifatturiere) provided by Capitalia (formerly Mediocredito Centrale) and covering the period 1995–2003. Interviews have been conducted respectively in 1998, 2001 and 2004 for the three surveys of all firms with 500 employees or more and of a representative sample of firms with more than 10 and less than 500 employees, stratified by geographical area, sector of economic activity and size. The three waves, 1995–1997, 1998–2000 and 2001–2003 gather information on 4,497, 4,680 and 4,289 firms respectively. In order to work on a balanced panel, we first merge the three surveys, thus identifying a sample of 414 firms always present across nine years. Firms offshoring production in 1998–2000 number 16, which represents 3.8 per cent of the sample. In order to avoid bad matches in the construction of the counterfactual sample, we further dropped observations belonging to industries in which no firm has moved production abroad (the remaining industries are listed in Table 9.2a). For the same reason, we also excluded other groups of firms potentially

218   R. Antonietti and D. Antonioli conducive to misleading results: specifically, we first dropped firms with missing values in balance sheet data; then we dropped firms undergoing takeovers or break-­ups and, finally, firms subcontracting production abroad before and after the treatment period 1998–2000. This last passage is particularly important for the correct specification of the treatment variable, since we select those firms that only offshored production in 1998–2000, so that we avoid any possible spurious effect on the outcome coming from previous or subsequent offshoring decisions. The final sample is made by a panel of 184 firms suitable for the analysis, and actively operating over the following three periods of time: 1995–1997, 1998–2000 and 2001–2003. Tables 9.2a and 9.2b show the structure of the sample by industry, employment size and geographical location of firms. As expected, the major part of the firms is of small and medium size, and this holds for each industry coded by Table 9.2a  Sample structure by industry and employment classes Industry (2-digit ATECO 1991)

Small (10–49)

Medium (50–249)

Large (250+)

Total

DB: textile and clothing DC: leather and footwear DG: chemicals and allied products DH: rubber and plastic products DJ: metal products DK: industrial machinery

54.06 87.50 100.00 80.00 70.00 58.33

32.43 12.50 0.00 20.00 23.33 33.33

13.51 0.00 0.00 0.00 6.67 8.34

100.0 100.0 100.0 100.0 100.0 100.0

Table 9.2b  Sample structure by employment classes and geographical area 1995–1997 Employment classes 11–20 21–50 11–50 51–250 251+ Total Geographical area North West North East North Centre South Total

21.6 41.0 62.6 26.5 10.9 100.0 40.43 29.57 70.0 17.28 12.72 4,497

1998–2000

39.94 37.14 77.08 16.15 6.77 100.0 37.54 27.44 64.98 20.62 14.40 4,680

1995–2003 Before cleaning 14.01 37.44 51.45 33.82 14.73 100.0 45.17 29.47 74.64 16.18 9.18 414

1995–2003 After cleaning 21.74 42.39 64.13 29.89 5.98 100.0 47.83 28.26 76.09 16.30 7.61 184

Production offshoring and skills – Italy   219 the two-­digit ATECO 1991 standard (textile and clothing, leather and footwear, chemicals, rubber and plastics, metal products, industrial machinery). Moreover, firms are primarily located in the North of Italy, and in particular in the North-­ West. The last two columns of Table 9.2b finally show the structure of the sample before and after the data cleaning process. Table 9.3, instead, shows the distribution of treated (offshoring) and untreated (non-­offshoring) firms by industry, employment size and geographical area. After cleaning, we identify seven firms offshoring production between 1998 and 2000, which represent 3.8% of the final sample, in front of 177 firms which, in the same period, did not decide to transfer their activities. Treated firms are almost equally spread over the six industries, are primarily of small size (57 per cent) and are located in the north of Italy (72 per cent). From a strict econometric point of view, the limited number of treated units does not represent a crucial issue for the application of the DID-­PSM methodology.12 What really matters is the relative dimension of the untreated sample, which needs to be large enough in order to draw an appropriate counterfactual sample of firms. The basic idea of PSM is to find, in a large group of untreated units, those firms that are similar to the treated ones in all relevant pre-­treatment characteristics. In our case, the ratio between treated and untreated units is 1 to 25, so that the pool of possible control units is relatively large and the counterfactual analysis reliable. However, we cannot consider our sample as to be fully representative of the whole Italian manufacturing industry: rather, we consider it as a sort of ‘case study’ referred to six industries and to firms predominantly located in the North of Italy. Table 9.3 Distribution of firms offshoring production by industry, employment size and geographical area (absolute values) Industry (ATECO 1991)

Offshoring

Non-offshoring

Total

DB: textile and clothing DC: leather and footwear DG: chemicals and allied products DH: rubber and plastic products DJ: metal products DK: industrial machinery Total

1 1 1 1 2 1 7

  36   7   1   14   28   35 177

  37   8   2   15   30   36 184

Small (11–49) Medium (50–249) Large (+250) Total

4 1 2 7

115   52   10 177

119   53   12 184

North West North East Centre South

2 3 1 1

  86   49   29   13

  88   52   30   14

Total

7

177

184

220   R. Antonietti and D. Antonioli

5  ATT estimation 5.1  Outcome variables The outcome variable is calculated, in line with the literature (Berman et al. 1994; Autor et al. 2003; Piva and Vivarelli 2004; Helg and Tajoli 2005), as the ratio between skilled and unskilled workers, represented as white collar and blue collar respectively. In the following, we define the ratio between white collar and blue collar (WC/BC) as the skill ratio. According to the International Standard Classification of Occupations (ISCO-­88), white collars are composed of managers, executives and office workers, and blue collars by plant operators. In addition, we define two other outcome variables by splitting the skill ratio into the share of white collar (i.e. the numerator, WC) and the share of blue collar (i.e. the denominator, BC) respectively. By so doing, we can speculate on the direction of the treatment effect on the outcome: if, for example, the firms’ choice to relocate production abroad alters the skill composition of employment by increasing the skill ratio, we are able to identify if such an increase is due to an increase of the numerator or to a decrease of the denominator, or to both. Finally, we define an alternative indicator for unskilled employment by summing up the relative share of blue collars and the relative share of office workers. In this way we build a finer proxy that distinguishes the semi-­skilled and the unskilled components of the labour force (OECD 1998; Hijzen et al. 2004).13 Figure 9.1 shows the trends of our outcome variables for the sample of seven treated firms and the sample of 177 untreated firms respectively. In the top left box is first measured total employment over time: as we can see, firms offshoring production between 1998 and 2000, face a 24 per cent decline in employment from an average of 250 employees in the pre-­treatment period (1995–1997) to an average of 202 employees in the post-­treatment period (2001–2003), while firms not engaged in offshoring production tend to maintain stable their workforce over an average of 78 employees. Interestingly, the top right box of Figure 9.1 shows the dynamics of the skill ratio. In front of a decline in total employment, treated firms seem to upgrade the skills of their workforce by 11 per cent, increasing the WC/BC ratio from a value of 0.302 before the assignment to treatment to a value of 0.411 after the assignment. On the contrary, untreated firms increase their skill ratio by only 1 per cent. The remaining boxes of Figure 9.1 show the dynamics of each element of the skill ratio. The left middle box, for instance, focuses on the number of managers over total employment and shows that, while untreated firms employ on average the same amount of managers (around 1.4 per cent of total employment), treated firms pass from an average of 2 per cent in 1995–1997 to an average of 1.2 per cent in 2001–2003. Differently, as the right middle box illustrates, the share of executives seems to increase for treated firms from an average of 0.2 per cent to an average of 1.6 per cent in the face of a relatively stable trend for untreated firms.

Production offshoring and skills – Italy   221 WC/BC

Total employment 300

0.6

250

0.5

200

0.4

150 100

0.3 0.2

50

0.1 0

0 1995

1996

1997

1998–2000

Treated

2001

2002

2003

1995

1996

1997

1998–2000

Treated

Unreated

2001

2002

2003

Unreated

Managers

Executives

0.025

0.025

0.02

0.02

0.015

0.015

0.01

0.01

0.005

0.005

0

0 1995

1996

1997

1998–2000

Treated

2001

2002

2003

1995

Unreated

1996

1997

Treated

1998–2000

2001

2002

2003

2002

2003

Unreated

Blue collars

Clerks 0.8 0.78 0.76 0.74 0.72 0.7 0.68 0.66

0.25 0.2 0.15 0.1 0.05 0 1995

1996

1997

Treated

1998–2000

2001

Unreated

2002

2003

1995

1996

1997

Treated

1998–2000

2001

Untreated

Figure 9.1 Employment trends for treated (offshoring) and untreated (non-offshoring) firms (period of assignment into treatment: 1998–2000).

Finally, the two bottom boxes of Figure 9.1 concern clerks and office workers on the one side and plant operators (BC) on the other. With respect to the former, the left bottom box shows that, from 1995 to 2003, both treated and untreated firms tended to increase the relative employment of clerks by the same amount (around 2 per cent). Looking at the latter, instead, we note that while untreated firms maintain a stable share of blue collars over time, firms farming out production activities are characterized by a lower average share of plant operators (–4.5 per cent). Generally speaking, the similar dynamics of each variable in the pre-­treatment period between treated and untreated observations seems to support the validity of the unconfoundedness hypothesis at the basis of the DID-­PSM estimation, as expressed in equation 9.4: in the absence of the treatment, the outcome of the treated and of the untreated units seems to follow parallel trends over time. Moreover, the relative stability of each variable for untreated firms also suggests – although simply from a graphical perspective – the absence of any macroeconomic, unobserved, shock that can potentially affect the whole workforce skill composition after the treatment period.

222   R. Antonietti and D. Antonioli In the following empirical analysis we first estimate whether firms’ commitment to offshoring production is responsible for the increase in the WC/BC ratio; secondly, we estimate if such an effect is driven by an increase in white collar employment or by a decrease in the employment of blue collars, or, rather, by both effects together. In this respect, we improve the earlier literature by using a non-­parametric DID PSM approach in order to be able to identify the causal effects on domestic employment composition by skill level. Matching estimators, especially if combined with a DID algorithm, are more appropriate than instrumental variable estimators since no strong exclusion restrictions are required (Blundell and Costa Dias 2002). In the following, we assess the credibility of our matching procedure by using a specific balancing test. 5.2  Assessing the quality of PSM: a test for the balancing property The next step in our treatment effect analysis is to assess the quality of the matching procedure. Since we determine the common support by conditioning on the propensity score, we have to check if the matching procedure can balance the distribution of the relevant variables in both the treated and untreated groups of firms. That is to say, we are looking for a propensity score that guarantees the same distribution of observable (and possibly unobservable) characteristics both for treated and untreated firms. According to Caliendo and Kopeinig (2008), the basic idea is that of comparing the distribution of observable before and after matching and then check if there are differences in such distributions; if this is the case, the matching based on the propensity score is not completely successful and the researcher is called to revise the propensity score determination, because it means that there is a possible misspecification of the model or a failure of the unconfoundedness condition. Many different methodologies can be implemented for testing the balancing property, which satisfies the independence of observable characteristics by the treatment status. For our purpose, we adopt the stratification test developed by Deheja and Wahba (1999, 2002), and implemented by Becker and Ichino (2002), according to which the observations are divided into k equal intervals of the estimated propensity score, in order to verify if some statistically significant differences remain in each interval between the average propensity score for treated and controls. The comparison within each interval between the averages of the propensity score for treated and controls is made conducting simple t-­test for mean equality. The algorithm developed by Becker and Ichino (2002) for STATA software proceeds by splitting the interval for which the propensity score averages between treated and counterfactuals are dissimilar, until the mean equality condition is satisfied in all intervals. In addition, the algorithm tests that within each interval the means of each characteristic do not differ between treated and counterfactual units. If the last condition is not satisfied than it means that the balancing property does not hold. In this case the researcher is called to implement a less parsimonious specification of the regression used to determine the propensity score.

Production offshoring and skills – Italy   223 As can be seen from Table 9.4, the algorithm identifies two blocks (intervals) within which the balancing property is satisfied. This means that, in each block, the mean propensity score for treated firms is not different from the one of the relative counterfactuals. Since the ATT is only defined in the region of common support it is also important to check the overlapping of the propensity score between treatment and counterfactual units. However, since for the ATT estimation it is sufficient to have potential matches in the set of the counterfactuals, following Lechner (2000), we inspect the existence of the common support by relying on a graphical check of the density distribution of the propensity score between treated and untreated units. Table 9.4  Testing the difference in the mean propensity score for treated and controls Test in block 1 Two-sample t-test with equal variances Group

Obs

Mean

Std. error

Std. dev.

Controls (0)

295

  0.054

0.003

0.043

Treated (1)

  18

  0.077

0.012

0.051

Combined

313

  0.055

0.025

0.044

–0.023

0.011

Difference: mean(0)–mean(1) t = –2.2278

Degrees of freedom = 311 H0: diff = 0

H1: diff < 0 Pr(T < t) = 0.0133

H1: diff ≥ 0 Pr(|T | > |t|) = 0.0266

H1: diff > 0 Pr(T > t) = 0.9867

The mean propensity score is not different for treated and controls in block 1 Test in block 2 Two-sample t-test with equal variances Group

Obs

Mean

Std. error

Std. dev.

Controls (0)

4

  0.278

  0.016

0.032

Treated (1)

3

  0.319

  0.005

0.009

Combined

7

  0.295

  0.012

0.032

–0.041

–0.019

Difference: mean(0)-mean(1) t = –2.1420

Degrees of freedom = 5 H0: diff = 0

H1: diff < 0 Pr(T < t) = 0.0425

H1: diff ≥ 0 Pr(|T | > |t|) = 0.0851

H1: diff > 0 Pr(T > t) = 0.9575

The mean propensity score is not different for treated and controls in block 2 Note The number of observations in each block refers to the N × T panel structure of our dataset, where T is equal to 3.

224   R. Antonietti and D. Antonioli The propensity score histogram by treatment status as reported in Figure 9.2 in the Appendix represents a useful instrument to fulfil our objective. As we can see, the propensity score distribution is similar for both treated and untreated, being skewed to the left for both groups. Given the density distribution, we are confident of finding good matches within the set of controls, with the exception of a treated firm which is likely outside the common support: we cannot match it with ‘identical’ firm(s) belonging to the counterfactuals. Thus, imposing the common support condition in the ATT estimation allows us to find a consistent estimator for the observations belonging to the common support, even if we leave outside the estimation procedure one treated unit. 5.3  Estimation results As stated before, the DID-­PSM consists in a two-­stage method of estimation of the ATT. In the first stage, we estimate the propensity score after identifying a set of characteristics which are supposed to be fixed over time or which are measured before the participation into the treatment. This exercise can be done by using any standard probability model, as, for example, a probit model of the type: Pr( Di = 1 | X i ) = Φ[h( X i )]

(9.6)

where Φ stands for the Normal cumulative density function and h(Xi) is a starting specification which includes all the relevant covariates as linear terms without interactions or higher order terms. Variables are selected by relying on economic theory, our knowledge of previous research and on the available information on the structural features of the Italian industrial system. Since the PSM strategy builds on the unconfoundedness hypothesis, only variables that are supposed to simultaneously affect both the offshoring decision and the outcome should be included in the model. In addition, only variables that are unaffected by participation should be taken into account: for this reason, we only select those covariates that are fixed over time or measured before participation, i.e. in the 1995–1997 period. In order to avoid any problem of over-­parameterization, we finally focus on a restricted set of characteristics. In fact, including too many variables in the first stage may create two problems: first, it can complicate the identification of the common support, because of the high number of parameters on the basis of which firms have to be compared; second, as argued by Augurzy and Schmidt (2001) and Bryson et al. (2006), it can raise the variance of the propensity score estimates. Following the theoretical and empirical literature on the determinants of international outsourcing and on the skill-­biased nature of international trade,14 we identify the following vector of covariates: (i) labour costs per employee (ULC); (ii) firm’s export activity (EXPORT); (iii) investments in ICT and network-­ related technologies (ICT); (iv) capital intensity (K/Y); (v) average productivity,

Production offshoring and skills – Italy   225 i.e. net sales per employee (Y/L); (vi) pre-­treatment levels in the outcome vari­ able, as defined by the ratio between white collars and blue collars (WC/BC). We also include geographical and year dummies. The first-­stage dependent variable, i.e. the treatment, consists in a dummy equal to 1 if the firm decides to offshore production for the first time between 1998 and 2000, and 0 if the firm never moved production abroad. Table 9.9 in the Appendix offers a description of all these variables, while Tables 9.10a, 9.10b and 9.10c provide some summary statistics with respect to the whole, the treated and the untreated sample respectively.15 In the second stage, we use the propensity score in order to estimate the ATT, as in (9.5). We identify the counterfactual sample by employing a NNM algorithm that compares the outcome of a treated firm with the outcome of an untreated firm that stands on the common support and shows the closest propensity score. Finally, we use a DID specification in order to control for unobserved heterogeneity across time. Tables 9.5 to 9.8 show the outcome of the estimations. In Table 9.5, we estimate the effect of offshoring production on four specifications of the skill ratio (WC/BC): the first row is the difference between the average skill ratio in 2001–2003 (post-­treatment period) and the average skill ratio in 1995–1997 (pre-­treatment period); the second, third, and fourth rows concern the difference between the skill ratio in each year after the treatment (2001, 2002, and 2003 respectively) and the average skill ratio in the period before offshoring occurred (1995–1997). Table 9.5  The effect of production offshoring on skilled labour: white collar/blue collar Skill ratio

ATT

Bootstrap s.e.

WC/BC2001/2003–WC/BC1995/1997 WC/BC2001–WC/BC1995/1997 WC/BC2002–WC/BC1995/1997 WC/BC2003–WC/BC1995/1997

0.265 0.265 0.274 0.257

0.225 0.248 0.254 0.289

Note Coefficients with (*) are statistically significant at 10%; coefficients with (**) are statistically significant at 5%. Standard errors are bootstrapped with 100 repetitions.

Table 9.6  The effect of production offshoring on skilled labour: white collar White collar employment share

ATT

Bootstrap s.e.

WC2001/2003–WC1995/1997 WC2001–WC1995/1997 WC2002–WC1995/1997 WC2003–WC1995/1997

0.145 0.169 0.175 0.156

0.237 0.583 0.568 0.561

Note Coefficients with (*) are statistically significant at 10%; coefficients with (**) are statistically significant at 5%. Standard errors are bootstrapped with 100 repetitions.

226   R. Antonietti and D. Antonioli The estimates of the ATT point to a ‘potential’ skill-­bias effect of production offshoring: the estimated coefficients are all positive, but their standard errors are not statistically different from zero after 100 repetitions. Although the sign of the coefficients seem to confirm our expectations about the impact of the treatment, we cannot consider the decision to move production abroad as really responsible for a recombination of the employment in favour of more skilled occupations. Even if, in line with recent literature (Piva and Vivarelli 2004; Falzoni and Tajoli 2008), we document a slight general up-­skilling of Italian manufacturing (see Figure 9.1), we cannot clearly assign to offshoring a crucial role. In this respect, Table 9.6 shows the effect of the treatment on the share of white collars (WC), i.e. on the share of skilled workers only. As before, even if the estimated coefficients are all positive, their standard errors are not statistically significant. This means that, although the share of skilled workers is generally increasing over time, production relocation is not responsible for such a phenomenon. Put another way, firms that offshored production in 1998–2000 do not seem to employ more white collars than similar firms that did not move their activities abroad. The only effect that offshoring seems to have on the skill composition within manufacturing firms concerns the employment of unskilled workers, i.e. blue collars (BC). Table 9.7 shows that firms offshoring production tend to employ, in the post-­treatment period, an average 9 per cent less plant operators than their counterfactuals. Moreover, the impact seems to be slightly increasing over time, passing from an average of –8.6 per cent in 2001 to –9.6 per cent in 2003. Table 9.7  The effect of production offshoring on unskilled labour: blue collar Blue collar employment share

ATT

Bootstrap s.e.

BC2001/2003–BC1995/1997 BC2001–BC1995/1997 BC2002–BC1995/1997 BC2003–BC1995/1997

–0.092* –0.086* –0.094** –0.096*

0.048 0.049 0.042 0.053

Note Coefficients with (*) are statistically significant at 10%; coefficients with (**) are statistically significant at 5%. Standard errors are bootstrapped with 100 repetitions.

Table 9.8  The effect of production offshoring on unskilled labour: blue collar and clerks BC + Clerks employment share

ATT

Bootstrap s.e.

BCCL2001/2003–BCCL1995/1997 BCCL2001–BCCL1995/1997 BCCL2002–BCCL1995/1997 BCCL2003–BCCL1995/1997

–0.057* –0.057* –0.053* –0.060*

0.034 0.034 0.032 0.034

Note Coefficients with (*) are statistically significant at 10%; coefficients with (**) are statistically significant at 5%. Standard errors are bootstrapped with 100 repetitions.

Production offshoring and skills – Italy   227 This result is in line with the picture emerging from Figure 9.1, although now we are strictly focusing on firms lying on the common support, and not on the whole sample of untreated firms. This result is also robust to an alternative specification of unskilled workers. In Table 9.8, in fact, we add the share of office workers to the share of plant operators in order to include also the clerical staff (BCCL). The negative effect of offshoring on the employment of unskilled workers still holds but reduces from an average of –5.7 per cent in 2001 to an average of –6 per cent in 2003. This result is also in line with Figure 9.1: while the employment of blue collars has decreased from 1995 to 2003, the employment of clerks and office workers has more or less remained stable over time. Therefore, our estimations seem to confirm the unskilled labour-­saving nature of production offshoring, while the complementarity with skilled labour is not statistically confirmed.16

6  Concluding remarks In this chapter we investigate the effect of offshoring production on the skill composition of manufacturing firms in Italy over the period 1995–2003. In particular, we compare if firms that decided to outsource their production activities on an international scale between 1998 and 2000 tend to employ a higher share of skilled workers with respect to the counterfactual set of firms that, although being similar, did not move their production abroad. We extend and improve the existing literature in three ways. First, differently from previous studies conducted at industry level, our unit of analysis is the single firm and its activities over a time span of nine years. Second, we use a non-­parametric propensity score matching technique in order to be able to identify and estimate the causal effect on domestic employment composition by skill levels while controlling for sample selection and without imposing any strong exclusion restriction to the parameters of the model. Third, we also control for unobserved heterogeneity by combining matching with the difference-­in-differences technique. Finally, we are also careful in assessing the appropriateness of our matching procedure by using a balancing test. Our estimates of the average treatment effect on the treated point only to a ‘potential’ skill-­bias effect of production offshoring. In particular, we only document that firms offshoring production between 1998 and 2000 tend to dismiss an average 10 per cent of blue collars with respect to their counterfactuals, while no significant effect seems to emerge with respect to the employment of white collars. In fact, although the signs of the estimated coefficients for these latter are always positive, they are not statistically different from zero. This result, however, may be due to the limited number of treated units we obtain after cleaning the original dataset. Our evidence seem to be in line with previous works (Sachs and Shatz 1994; Strauss-­Khan 2003; Piva and Vivarelli 2004; Hijzen et al. 2004) that stress how the skill-­bias effect of international outsourcing is mainly determined by the negative dynamics of the employment of unskilled labour. The will to exploit

228   R. Antonietti and D. Antonioli factor cost differentials and the relocation of low-­skill intensive phases of the production process to cheap labour countries has contributed, at least in the short-­medium run, to substitute away for domestic employment of manual workers, while keeping the relative share of non-­manuals relatively stable. We conclude that other factors are probably more responsible for the general increase in the employment of skilled personnel within Italian manufacturing, as, for instance, labour market reforms, the presence and the international performance of industrial districts and technological change. However, due to the evolving nature and complexity of offshoring decisions, in order to fully assess the labour market impact of such an internationalization strategy, more research on the field is required.

Appendix

0

0.1

0.2 Propensity score Untreated

0.3 Treated

Figure 9.2  Propensity score histogram by treatment status.

Dummy

Dummy

Dummy Dummy Continuous Continuous Continuous Continuous Continuous Continuous

North West: Liguria, Lombardia, Piemonte, Valle d’Aosta North East: Emilia-Romagna, Friuli Venezia-Giulia, Trentino Alto Adige, Veneto Center: Lazio, Marche, Toscana, Umbria South: Abruzzo, Basilicata, Calabria, Campania, Molise, Puglia, Sardegna, Sicilia

Treatment 1 if the firm offshored at least one phase of the production process to cheap-labour countries (Eastern Europe and former-Yugoslavia) only between 1998 and 2000, 0 otherwise

Technology 1 if the firm has invested in informatics and ICT in the period 1995–1997, 0 otherwise

Export activity 1 if the firm exported goods in 1995–1997, 0 otherwise

Capital intensity Net capital stock over sales

Labour productivity Net sales per employee

Unit labour cost Labour cost per employee

White Collars Managers/Employment + Executives/Employment + Clerks/Employment

Blue collars Plant operators/Employment

Skill ratio White collars over blue collars

Area

D

ICT

EXPORT

K/Y

Y/L

ULC

WC

BC

WC/BC

Type

Definition

Variable

Table 9.9  Variables definition

1995–1997

1995–1997

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

ICT

EXPORT

Total Employment

Managers/Employment

Executives/Employment

Clerks/Employment

K/Y

Log (K/Y)

Y/L

Log (Y/L)

ULC

Log (ULC)

WC

Log (WC)

BC

Log (BC)

WCBC

Log (WCBC)

1,104

1,104

1,104

1,104

1,104

1,104

  552

  552

  552

  552

  552

  552

1,104

1,104

1,104

1,104

  552

  552

Obs

–1.044

0.475

–0.357

0.717

–1.656

0.224

3.516

50.699

5.292

318.940

3.531

65.647

0.202

0.009

0.014

83.954

0.712

0.701

Mean

0.778

0.456

0.235

0.142

0.620

0.130

0.753

70.915

0.885

426.992

1.147

93.072

0.122

0.025

0.255

166.760

0.453

0.458

Std. Dev.

Note Summary statistics refer to the overall N × T panel, where N is equal to 184 (firms) and T to 3 or 6 (years) respectively.

Years

Variable

Table 9.10a Summary statistics: merged sample

–6.265

0.002

–1.707

0

–4.883

0

1.811

6.116

3.224

25.118

–0.772

0.462

0

0

0

10

0

0

Min

1

1

1.507

4.514

–0.015

0.985

0

1

6.827

922.069

8.625

5,567.856

7.062

1,166.634

1

0.214

0.167

1,801

Max

1995–1997

1995–1997

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

ICT

EXPORT

Total Employment

Managers/Employment

Executives/Employment

Clerks/Employment

K/Y

Ln (K/Y)

Y/L

Ln (Y/L)

ULC

Ln(ULC)

WC

Ln(WC)

BC

Ln (BC)

WCBC

Ln (WCBC)

42

42

42

42

42

42

42

42

42

42

42

42

42

42

42

42

21

21

Obs

–1.356

0.357

–0.284

0.767

–1.850

0.196

3.947

76.766

5.550

445.308

3.898

92.936

0.171

0.009

0.016

225.667

0.857

0.857

Mean*

0.854

0.315

0.200

0.134

0.734

0.122

0.859

75.144

1.073

463.571

1.170

99.453

0.098

0.025

0.023

347.336

0.354

0.354

Std. Dev.

Note Summary statistics refer to the overall N × T panel, where N is equal to 7 (firms) and T to 3 or 6 (years) respectively.

Years

Variable

Table 9.10b Summary statistics: treated sample (offshoring firms)

–3.401

0.033

–0.821

0.44

–3.434

0.032

2.968

19.460

4.467

87.030

2.560

12.925

0

0

0

16

0

0

Min

0.241

1.273

–0.033

0.968

–0.734

0.48

5.440

230.517

7.238

1,391.964

5.676

291.659

0.4

0.12

0.08

1,163

1

1

Max

1,062   531   531   531   531   531   531 1,062 1,062 1,062 1,062 1,062

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

1995–1997 + 2001–2003

Executives/Employment

Clerks/Employment

K/Y

Ln(K/Y)

Y/L

Ln (Y/L)

ULC

Ln (ULC)

WC

Ln (WC)

BC

Ln (BC)

WCBC

Ln (WCBC)

1,062

–1.032

0.479

–0.359

0.715

–1.648

0.225

3.499

49.668

5.28189

313.942

3.517

64.567

0.203

0.009

0.014

78.350

0.706

0.695

Mean*

0.773

0.460

0.236

0.142

0.614

0.130

0.744

70.619

0.877

425.181

1.145

92.746

0.122

0.025

0.026

153.041

0.456

0.461

Std. Dev.*

Note Summary statistics refer to the overall N × T panel, where N is equal to 177 (firms) and T to 3 or 6 (years) respectively.

1,062

1,062

1,062

1995–1997 + 2001–2003

1995–1997 + 2001–2003

Total Employment

  531   531

1995–1997

1995–1997

ICT

EXPORT

Managers/Employment

Obs

Years

Variable

Table 9.10c Summary statistics: untreated sample (non-offshoring firms)

–6.265

0.002

–1.707

0

–4.883

0

1.811

6.116

3.224

25.118

–0.772

0.462

0

0

0

10

0

0

Min*

1.507

4.514

–0.015

0.985

0

1

6.827

922.069

8.625

5,567.856

7.062

1,166.634

1

0.214

0.167

1,801

1

1

Max*

Production offshoring and skills – Italy   233

Notes   1 The authors thank the participants at the XVI AISSEC National Conference, Parma, 21–23 June 2007, at the X EUNIP International Conference, Prato 12–14 September 2007, and at the XXII AIEL National Conference, Naples, 13–14 September 2007. We are particularly grateful to Lucia Tajoli and Guido Pellegrini for useful comments and suggestions. Roberto Antonietti acknowledges with thanks Capitalia for the provision of the datasets. All the opinions expressed and the errors pertain only the authors.   2 See Piva (2004) and Antonietti (2007) for a review on skill-­biased technological change.   3 Amiti and Wei (2009) report that, while the US average annual growth rate of service offshoring between 1992 and 2000 has been 6.26 per cent, compared to the 4.39 per cent growth rate of material offshoring, in 2000 its relative share over total import flows is still around 0.3 per cent with respect to 17 per cent of material offshoring.   4 Alternatively, the Oxford English Dictionary defines offshoring as the action or practice of moving or basing a business operation abroad, usually to take advantage of lower costs (http://dictionary.oed.com/).   5 Offshore outsourcing is also defined as subcontracting abroad or offshoring in the broad sense (OECD 2007).   6 For a comprehensive survey on north-­south trade models explaining the widening wage inequality between skilled and unskilled workers see Chusseau et al. (2008).   7 Similar results for the German case emerge also in Fitzenberger (1999) and Falk and Koebel (2000), who find no evidence that international outsourcing of production and services positively affects the skill composition of the manufacturing workforce. Rather, Fitzenberger gives technology the dominant role in shifting away the employment of unskilled workers.   8 Empirical studies testing the skill-­biased international trade hypothesis are generally based on the estimation of short-­run cost functions, reflecting the cost-­minimizing behaviour of firms. The most utilized flexible cost function is generally the dual of the transcendental logarithmic production function, as proposed by Christensen et al. (1973). However useful, this approach suffers two major limitations: on the one hand, it relies on a ‘simple’ cost function framework, which is subject to a set of ad hoc regularity conditions – i.e. symmetry and homogeneity of parameters and constant returns to scale – that are necessary for its analytical tractability; on the other, it is linked to a linear functional form that constrains the parameters to assume specific values (see Barnett et al. (1985) and Dumont (2006) for a treatment of the drawbacks linked to the use of a translog cost function).   9 The use of propensity score matching as a tool for the evaluation of public programmes has been first proposed by Rosenbaum and Rubin (1983) with a particular reference to medical trials and diagnosis. However, it is after the contribution of Dehejia and Wahba (2002) that PSM has been considered as a primary methodology within the evaluation literature in a context of observational data (see also Angrist et al. 1996; Heckman 1990, 1997; Heckman et al. 1997, 1998; Heckman et al. 1999; Sianesi 2004; Smith and Todd 2005; Wooldridge 2001). 10 The other widely used methods are the radius matching, the calliper matching and the kernel matching (Becker and Ichino 2002; Caliendo and Hujer 2006; Caliendo and Kopeinig 2008). 11 This means that in case a treated unit matches with a single counterfactual unit w(i, j) is equal to 1; if counterfactuals per treated units are two, then w is equal to ½ and so on. 12 If the treated units in the sample were a representative set of the offshoring firms in the population it would be possible to generalize the results; if they are not representative – as in the present case – it is still possible to consistently verify the impact of the treatment on the treated without generalizing the results on a national level.

234   R. Antonietti and D. Antonioli 13 We also build an indicator for the high-­skilled and low-­skilled components of white collars, summing up the employment shares of managers and of executives. However, due to the high number of firms with zero managers and zero executives, when we operate the logarithmic transformation of their ratio, we miss a lot of data, so that our estimates, although significant, are not easily comparable with those of blue collars. Tables and outputs on white collars are not reported here, but are available on request from the authors. 14 In particular, unit labour costs are considered, for instance, in Abraham and Taylor (1996) and in Computer Weekly and Morgan Chambers (2001); export behaviour in Girma et al. (2004), Helpman et al. (2004) and Bartel et al. (2005); technology and ICT in Basile et al. (2003), Abramovsky and Griffith (2006), Tomiura (2005) and Bartel et al. (2005); capital intensity in Bartel et al. (2005); productivity in Tomiura (2005), Castellani and Zanfei (2007) and Benfratello and Razzolini (2007). Following Helg and Tajoli (2005) we also add the lagged skill ratio variable. 15 We do not report here the first-­stage estimation results, since they are only functional to the identification of the propensity score. Output and comments are available in Antonietti and Antonioli (2010). 16 These results are robust to alternative specifications of the first-­stage model and, then, of the propensity score. Tables and outcomes are not reported here but are available in Antonietti and Antonioli (2010).

References Abraham K.G. and Taylor S.K. (1996), Firm’s use of outside contractors: theory and evidence, Journal of Labor Economics, 14: 394–424. Abramovsky L. and Griffith R. (2006), Outsourcing and offshoring of business services: how important is ICT?, Journal of the European Economic Association, 4(2–3): 594–601. Agarwal J.P. (1997), Effect of foreign direct investment on employment in home countries, Transnational Corporation, 2: 1–28. Amiti M. and Wei S.J. (2005), Fear of outsourcing: is it justified?, Economic Policy, 20: 307–47. Amiti M. and Wei S.J. (2009), Service offshoring and productivity: evidence from the United States, The World Economy, 32(2): 203–20. Anderton B. and Brenton P. (1999), Outsourcing and low-­skilled workers in the UK, Bulletin of Economic Research, 51: 267–86. Angrist J.D., Imbens G.W. and Rubin D.B. (1996), Identification of causal effect using instrumental variables, Journal of Econometrics, 71: 145–60. Antonietti R. (2007), Opening the ‘skill-­biased technological change’ black box: a look at the microfoundations of the technology-­skill relationship, Economia Politica, 24: 451–76. Antonietti R. and Antonioli D. (2010), Production offshoring and the skill composition of manufacturing firms: a counterfactual analysis, International Review of Applied Economics, forthcoming. Arndt S.W. (1997), Globalization and the open economy, North American Journal of Economics and Finance, 8: 71–9. Augurzy B. and Schmidt C. (2001), The propensity score: a means to an end, IZA Working Paper no. 271, Bonn, IZA. Autor D.H., Levy F. and Murnane R.J. (2003), The skill content of recent technological change: an empirical investigation, Quarterly Journal of Economics, 118: 1279–333.

Production offshoring and skills – Italy   235 Baldone S., Sdogati F. and Tajoli L. (2007), On some effects of international fragmentation of production on comparative advantages, trade flows, and the income of countries, World Economy, 30: 1726–69. Barba Navaretti G. and Castellani D. (2004), Does investing abroad affect performance at home? Comparing Italian multinationals, CEPR Discussion Paper no. 4284, London, CEPR, online, available at: www.cepr.org/pubs/dps/DP4284.asp. Barnett W.A., Lee Y.W. and Wolfe M.D. (1985), The three-­dimensional global properties of the Minflex Laurent, Generalized Leontief, and Translog Flexible functional forms, Journal of Econometrics, 30(1–2): 3–31. Bartel A.P., Lach S. and Sicherman N. (2005), Outsourcing and Technological Change, Working Paper no. 11158, Cambridge, MA, NBER, online, available at: www.nber. org/papers/w11158. Basile R., Giunta A. and Nugent J.B. (2003), Foreign expansion of Italian manufacturing firms in the nineties: an ordered probit analysis, Review of Industrial Organization, 23: 1–24. Becker S. and Ichino A. (2002), Estimation of average treatment effects based on propensity scores, The Stata Journal, 2: 358–77. Benfratello L. and Razzolini T. (2007), Firms’ productivity and internationalisation choices: evidence for a large sample of Italian firms, in Piscitello L. and Santangelo G. (eds.), Multinationals and Local Competitiveness, Milan, Franco Angeli. Berman E., Bound J. and Griliches Z. (1994), Changes in the demand for skilled labour within US manufacturing: evidence from the annual survey of manufacturers, The Quarterly Journal of Economics, 109: 367–97. Blömstrom M., Fors G. and Lipsey R.E. (1997), Foreign direct investment and employment: home country experience and in the United States and Sweden, The Economic Journal, 107: 1787–97. Blundell R. and Costa Dias M. (2002), Alternative approaches to evaluation in empirical microeconomics, Portuguese Journal of Economics, 91: 91–115. Braconier H. and Ekholm K. (2000), Swedish multinationals and competition from high to low-­wage location, Review of International Economics, 8: 448–61. Brainard L. and Litan R. (2004), Offshoring service jobs: bane or boom or what to do?, The Brookings Institutions, Policy Brief no. 132. Bryson A., Dorsett R. and Purdon S. (2006), The use of propensity score matching in the evaluation of labour market policies, in Caliendo, M. (ed.) Microeconomic Evaluation of Labour Market Policies, Berlin/Heidelberg, Springer, pp. 71–96. Caliendo M. and Hujer R. (2006), The microeconometric estimation of treatment effects: an overview, Allgemeines Statisches Archiv, 90(1): 199–215. Caliendo M. and Kopeinig S. (2008), Some practical guidance for the implementation of propensity score matching, Journal of Economic Surveys, 22: 31–72. Caliendo M., Hujer R. and Thomsen S. (2008), The Employment effects of job creation schemes in Germany: a microeconometric evaluation, in Fomby, T., Carter Hill, R., Millimet, D.L. Smith, J.A. and Vytlacil, E.J. (eds.), Modelling and Evaluating Treatment Effects in Econometrics (Advances in Econometrics Series Vol 14), Bingley, Emerald, pp. 381–428. Castellani D., Mariotti I. and Piscitello L. (2006), Attività estere delle imprese multinazionali italiane e skill upgrading, in Mariotti M. and Piscitello L. (eds.), Multinazionali, innovazione e strategie per la competitività, Bologna, Il Mulino. Castellani D. and Zanfei A. (2007), Internationalisation, innovation and productivity: how do firms differ in Italy?, The World Economy, 30: 156–76.

236   R. Antonietti and D. Antonioli Christensen L.R., Jorgenson D.W. and Lau L.J. (1973), Transcendental logarithmic production frontier, The Review of Economics and Statistics, 55: 28–45. Chusseau N., Dumont M. and Hellier J. (2008), Explaining rising inequality: skill-­biased technical change and North-­South trade, Journal of Economic Surveys, 22: 409–57. Dehejia R.H. and Wahba S. (1999), Causal effects in nonexperimental studies: reevaluating the evaluation of training programs, Journal of the American Association, 94: 1053–62. Dehejia R.H. and Wahba S. (2002), Propensity score matching methods for nonexperimental causal studies, The Review of Economics and Statistics, 84: 151–61. Dell’mour R., Egger P., Gugler K. and Pfaffermayr M. (2000), Outsourcing of Austrian manufacturing to Eastern Europe countries: effects on the productivity and the labour market, in Arndt S.W., Handler H. and Salvatore D. (eds.), Fragmentation of the Value Added Chain, Vienna Conference. Dumont M. (2006), Foreign outsourcing, labour demand and the choice of functional form, Journal of Applied Economics, 9: 255–73. Egger H. and Egger P. (2003), Outsourcing and skill-­specific employment in a small open economy: Austria after the fall of the iron curtain, Oxford Economic Papers, 55: 625–43. Egger H. and Falkinger J. (2003), The distributional effects of international outsourcing in a 2 × 2 production model, North American Journal of Economics and Finance, 14: 189–206. Falk M. and Koebel B. (2000), Outsourcing of Services, Imported Materials and the Demand for Heterogeneous Labour: An Application of a Generalized Box-­Cox Function, Discussion Paper no. 00–51, Berlin, ZEW, online, available at: www.zew.de/en/ publikationen/publikation.php3?action=detail&art=12&nr=6740. Falzoni A.M. and Tajoli L. (2008), Offshoring and the Skill Composition of Employment in the Italian Manufacturing Industries, Working Paper no. 219, Bocconi University, Milan, CESPRI, online, available at: http://ideas.repec.org/p/ cri/ cespri/wp219.html. Feenstra R. and Hanson G. (1996), Globalization, outsourcing, and wage inequality, American Economic Review Paper and Proceedings, 86: 240–48. Fitzenberger B. (1999), Wages and employment across skill groups: an analysis for West Germany, ZEW Economic Studies Vol. 6, Heidelberg, Physica. Glass A.J. and Saggi K. (2001), Innovation and wage effects of international outsourcing, European Economic Review, 45: 67–86. Girma S., Görg H. and Strobl E. (2004), Exports, international investments, and plant performance: evidence from a non-­parametric test, Economic Letters, 83: 317–24. Hansson P. (2005), Skill upgrading and production transfer within swedish multinationals in the 1990s, Scandinavian Journal of Economics, 197(4): 673–92. Head K. and Ries J. (2002), Offshore production and skill upgrading by Japanese manufacturing firms, Journal of International Economics, 58: 81–105. Heckman J.J., Ichimura H. and Todd P.E. (1997), Matching as an econometric evaluation estimation: evidence from evaluating a job training programme, Review of Economic Studies, 64: 605–54. Heckman J.J., Ichimura H. and Todd P.E. (1998), Matching as an econometric evaluation estimation, Review of Economics Studies, 65: 261–94. Heckman J.J., Lalonde R. and Smith J. (1999), The economics and econometrics of active labor market programs, in Ashenfelter O. and Card D. (eds.), Handbook of Labour Economics Vol. 3A, North Holland, Elsevier. Heckman J.J. (1990), Varieties of selection bias, American Economic Review, 80: 313–18.

Production offshoring and skills – Italy   237 Heckman J.J. (1997), Instrumental variables: a study of implicit behavioural assumptions used in making program evaluations, Journal of Human Resources, 32(Summer): 441–62. Helg R. and Tajoli L. (2005), Patterns of international fragmentation of production and the relative demand for labour, North American Journal of Economics and Finance, 16: 233–54. Helpman E., Melitz M. and Yeaple S. (2004), Export versus FDI with heterogeneous firms, American Economic Review, 94: 300–16. Hijzen A., Görg H. and Hine R.C. (2004), International outsourcing and the skill structure of labour demand in the United Kingdom, Economic Journal, 115: 860–78. Jones R.W. and Kierzkowski H. (2001), A framework for fragmentation, in Arndt S. and Kierzkowski H. (eds.), Fragmentation and International Trade, Oxford, Oxford University Press. Kohler W. (2003), The distributional effects of international fragmentation, German Economic Review, 4: 89–120. Lechner M. (2000), A note on the common support problem in applied evaluation studies, Discussion Paper 09, University of St. Gallen, SIAW, online, available at: www.siaw. unisg.ch/org/siaw/webold.nsf/SysWebRessources/ML_2000_L_09/$FILE/00l09.pdf. Lipsey R.E. (2002), Home and host country effects of FDI, Working Paper no. 9293, Cambridge, MA, NBER, online, available at: www.nber.org/papers/w9293. Mariotti S. and Mutinelli M. (2005), Italia multinazionale 2005. Le partecipazioni italiane all’estero ed estere in Italia, Politechnic University of Milan-­ICE. Mariottti S., Mutinelli M. and Piscitello L. (2003), Home country employment and foreign direct investment: evidence from the Italian case, Cambridge Journal of Economics, 27(3): 419–31. Markusen J.R., Konan D., Venables A. and Zhang K. (1996), A Unified Treatment of Horizontal Direct Investment Vertical Direct Investment and the Pattern of Trade in Goods and Services, Working Paper no. 5696, Cambridge, Ma, NBER, online, available at: www.nber.org/papers/w5696.pdf. Computer Weekly and Morgan Chambers (2001), Outsourcing in the FTSE 100 Firms: The Definitive Study, formerly online at: www.morgan-­chambers.com/ studyftse.htm. OECD (1998), OECD Data on Skills: Employment by Industry and Occupation, DSTI Working Paper no. 1998/4, Paris, OECD. OECD (2007), Offshoring and Employment. Trends and Impacts, Paris, OECD. Piva M. (2004), Evoluzione della domanda di lavoro qualificato: teorie e verifiche empiriche, Economia Politica, 21(2): 323–59. Piva M. and Vivarelli M. (2004), The determinants of the skill bias in Italy: R&D, organization or globalizaton?, Economics of Innovation and New Technology, 13: 329–47. Rosenbaum P.R. and Rubin D.B. (1983), The central role of propensity score in observational studies for causal effects, Biometrika, 70: 41–4. Sachs J.D. and Shatz H.J. (1994), Trade and jobs in U.S. manufacturing, Broking Papers on Economic Activity, 1994(1): 1–84, online, available at: www.jstor.org/ stable/ i343493. Sianesi B. (2004), An evaluation of the Swedish system of active labour market programmes in the 1990s, Review of Economics and Statistics, 86: 133–55. Slaughter M.J. (2000), Production transfer within multinational enterprises and American wages, Journal of International Economics, 50: 449–72. Smith J.A. and Todd P. (2005), Does matching overcome LaLonde’s critique of nonexperimental estimators?, Journal of Econometrics, 125: 305–53.

238   R. Antonietti and D. Antonioli Strauss-­Khan V. (2003), The Role of Globalization in the Within-­Industry Shift Away From Unskilled Workers in France, Working Paper no. 9716, Cambridge, MA, NBER, online, available at: www.nber.org/papers/w9716. Tomiura E. (2005), Foreign outsourcing and firm-­level characteristics: evidence from Japanese manufacturers, Journal of the Japanese and International Economies, 19: 255–71. UNCTAD (2006), World Investment Report 2006, Geneva, United Nations. Wood A. (1994), North–South Trade, Employment, and Inequality: Changing Fortunes in a Skill-­Driven World, Oxford, Clarendon Press. Wooldridge J.M. (2001), Asymptotic properties of weighted M-­estimators for standard stratified samples, Econometric Theory, 17: 451–70.

10 A global network and its local ties Restructuring of the Benetton Group1 Paolo Crestanello and Giuseppe Tattara

1  Introduction Since the 1990s, the European clothing market has experienced a massive process of restructuring. There are several causes: from the modest growth of consumption caused by the stagnating income of Western countries in most recent years, to the appreciation of the euro over the dollar that held back European exports and to the strong competition coming from low wage countries, particularly China after the abolition of clothing quotas (Adam 1971; Finger 1976, 1977; Baden 2002; Graziani 2001). The consequences for the richer European countries have been a growing import of garments and the loss of hundreds of thousands of jobs in the domestic textile and clothing industry. Italy, as the only exception in the EU15 countries, maintains a clothing trade surplus that is however diminishing through time. In fact, between 1995 and 2007, the normalized Italian balance of clothing trade fell from 50 percent to 17 percent and during the same period the imports from China, as a share of total Italian imports, rose from 11 percent to 25 percent. In these last few years, the clothing market overall has appeared stagnant, but the companies who have adopted strategies based on a quick response have grown faster than others (Ciappei and Sani 2006: 55). These companies have increased the variety of products that can be brought to the market in a short time. The fashion market is volatile and risky and consumers are incessantly stimulated to renew their purchases through a rich offer of new products. The successful marketer on the one hand needs to meet the variable fashion demand of the customer but on the other hand requires to build a network of fast and reliable suppliers (Gereffi et al. 2005). Fast fashion and production flexibility contrast with the necessity, especially for firms that are in the low- or medium-­end market, to produce at competitive prices. It is possible to reduce costs of production by sourcing to countries with lower wages, but importing from distant countries increases the time it takes to have the garments on the market and makes production less flexible. Consequently, the governance of the supply chain is now much more complex as it requires more efforts on design (to create a greater number of new products) and on controlling and managing distant suppliers (Abecassis-­Moedas 2006).

240   P. Crestanello and G. Tattara Benetton is one of the largest European garment producers and its core business consists of designing, producing and selling garments for men, women and children in wool and cotton. Its clothing production is marketed under several brands: “United Colors of Benetton”, the flagship brand mainly for men and children, and Silsey, the fashion-­oriented brand for women, represent respectively 77 percent and 18 percent of the total sales. Other minor brands, “The Hip Side” for trendy, “Play Life and Killer Loop” for the leisurewear, cover the remaining 5 percent. Benetton, with Zara, is considered an exception in a context of large international brands mainly owned by pure retailers, such as Gap, H&M, and Mango, that do not own factories, but keep market relations with large independent suppliers (Tokatly 2008: 23). Benetton is still renowned in the international literature as a firm that manufactures the majority of its products in Italy (Berger 2006; Dicken 2007; Tokatly 2008). If this was true in the past, today it is no more. In the last five years, Benetton has quickly increased the process of production relocation abroad and at present only 20 per cent of its products are manufactured in Italy, and it is foreseen that this percentage will halve in the next few years. The majority of the production is made in North Africa and in East Europe, but the recourse to the Asian producers, that did not appear in Benetton’s suppliers list until 2003, is greatly increasing. Parallel to this, production planning has deeply changed. Up to 2003, production was entirely based on two seasonal collections and 80 percent of the orders from the retailers were received before the beginning of the selling season. The remaining 20 percent were mainly reorders and, only in a small amount, orders of new products designed during the selling season. Today the number of collections has increased and the amount of orders received before the selling season has greatly decreased. In the contest of broadening production variety, the increase of foreign supplying, in particular from Asia, has deepened Benetton’s logistical problems. Furthermore, the newly adopted strategies have had tangible consequences on the garment district of Treviso that in the last few years has greatly downsized.

2  The roots of success: the period 1960–1970 The firm Benetton was founded in 1965 at Ponzano Veneto, a small town near Treviso, by four brothers’ initiative. In the beginning, Benetton was only a small company that was producing sweaters for local independent retailers. The keys to the success consisted in some innovations related to the product and its distribution and to an efficient production organization based on the work of a large network of small local subcontractors specialized in knitting, cutting and sewing garments. Right from the beginning, Benetton offered a new product characterized by bright colours and targeted at young people. The full fashion knitwear was made on cotton looms and it was strictly in the plain colour.2 In this way it is possible to knit plain wool into sweaters and postpone dyeing the entire stock until just before going to the market, according to the latest fashion trends.

A global network and its local ties   241 Retailers could order plain sweaters in advance and specify the colour during the selling season.3 Together with the advantage of a rapid response to the fashion market, the dyeing postponement process allowed a drastic reduction of costs due to less expensive inventories and to a smaller unsold stock. This process was made possible thanks to an advanced dyeing process set up by Benetton, able to offer a wide number of colours and the guarantee that garments did not lose their colours when washed. Benetton internalized the dyeing process to take advantage fully of its dyeing know-­how (Dapiran 1992). Shortly after the production of knitwear, followed the production of shirts and jeans. In the beginning Benetton sold them under different brands (Tomato, Jeans West, etc.) because the quality of these new products was not yet comparable to the one obtained for the sweaters and there was a fear that it might damage the reputation that the firm had achieved as a knitwear producer. It is estimated that in the second part of the 1970s around 60–70 percent of the overall Benetton production was made by a hundred of subcontractors located mainly in Treviso and in the surrounding provinces of Veneto (Nardin 1987; Benetton and Lee 1990). The activities such as design, quality control and the manufacturing stages which required greater investments (such as knitting, cutting and dyeing), were instead undertaken in the two factories of Villorba and Monzambano which employed about 1,000 workers. From the very beginning, tight control was imposed on subcontractors, to whom raw materials and precise technical details to make the garment were sent. The price paid by Benetton to its subcontractors was generally lower than the one paid by other firms, and it was updated yearly according to the rate of inflation. Lower prices, however, were compensated by the certainty and punctuality of payments, by long production runs (which could surpass 10,000 items per model which was large for the market of the time) and by the guarantee of continuous orders that permitted the subcontractors to work at full production capacity (Brusco and Crestanello 1995; Crestanello 1999). Parallel to the productive developments, Benetton carried out another revolution: it was the first Italian firm to apply a quasi-­franchising system to retailing. This system permitted a fast growth of sales thanks to the fact that there was no need to have great financial resources to open new stores. That was good for Benetton, that at the beginning of its success lacked the necessary capital. The first Benetton’s shop opened in Belluno in 1966 and in just few years Benetton’s stores covered all Italian’s provinces. At the beginning of the 1970s, there were about 500 stores under different Benetton’s brands (as well as Benetton, also Tomato, My Market, and Merceria). The relationships with the retailers were similar but not equal to those of the franchising contract. In fact, there was not a written contract and royalties were not requested. On the other hand Benetton did not guarantee the retailers an exclusivity of territory, did not repurchase the unsold products and imposed the retail prices (Favero 2005: 79). The shops, generally of small size, constituted an innovation in the Italian market because they offered, at a good price, good quality and highly fashionable sweaters which were displayed in a way so that customers were

242   P. Crestanello and G. Tattara able to pick them up from the shelves, touch and try them. Even if the retailers worked with a limited mark-­up, the selling activity was profitable as it guaranteed a high turnover per square meter and per worked hour.4 The growth of Benetton depended more and more on the capacity to increase the number of stores under its own brands involving in the business some of its agents who became owners of many stores. In those years, Benetton contributed to the creation of the casual style, targeted at the beginning for young people, but which shortly after spread to other age groups. At the beginning almost all Benetton production was sold on the domestic market and exports became significant toward the end of the 1970s with stores opened in France, Germany, United Kingdom, Holland and Belgium (Benetton Group 1990: 112). Between 1973 and 1979, the Benetton’s sales increased from 31 to 287 million euros (Benetton and Lee 1990: 104). In the 1970s, thanks to Benetton and to other firms that followed the trail of its success, Italy became the major producer of knitwear in Europe. Another important producer of Treviso, Stefanel, in those years experienced a market success following the same Benetton’s business model (coloured sweaters sold in franchising) and becoming very soon one of its main competitors.

3  The period 1980–1990: growth by vertical and horizontal integration The 1980s saw the passage from a family-­owned company to a managerial one with the recruitment of a manager from outside the Ponzano factory.5 The firm was quoted in Milan’s stock exchange in 1985 and later in the New York stock exchange (from whose quotation it was withdrawn in 2007). In those years, an important role was played by the promotional campaigns carried out in collaboration with Oliviero Toscani who strengthened Benetton image, building an identity brand that was missing until then. The name “United Colors of Benetton” became the flagship brand of the Benetton Group. Benetton grew through a strategy of vertical and horizontal integration. At the end of the 1970s Benetton’s organization could be defined as “quasi-­verticalintegration” (Blois 1972) as the company controlled the whole value chain, even though various activities were not organized through an exclusive hierarchical control. In fact Benetton represented the main, if not the only, client of its subcontractors and could decide the price paid and the general terms of supply (Belussi and Festa 1990). As in the case of the franchisees, there was no written contract and the orders were tacitly placed again every season. Benetton established with its subcontractors long-­term relationships based on cooperation and trust. Although there was an evident asymmetry in the negotiation power (subcontractors employed an average of 15–20 workers), Benetton, thanks to the constant growth of sales, was able to renew and increase the orders at every season, favouring the subcontractors who updated their equipment. Benetton used to advise its subcontractors about which new machines were most profitable and provided to some of them financial assistance through its leasing and factoring company.

A global network and its local ties   243 It was at the end of the 1980s that Benetton started the process of entering directly into the upstream stages of the clothing value chain.6 It acquired important textile and knitting factories through the affiliated company Olimpias that today owns, in several Italian provinces, ten plants supplying the majority of the raw materials necessary to the Group’s clothing division. The control of the entire value chain was then completed: from retailing to clothing and textile manufacturing, to which wool production was later added. In 1991 in fact, the Benetton family acquired the company Tierras Del Sur Argentino, becoming the owner of 900,000 hectares of breeding area for sheeps, for a total production of over six million kilos of wool.7 The process of horizontal integration was also achieved. The strategy of providing a total look was completed with the introduction of products such as shoes, spectacles, perfumes, watches and, most recently, jewellery. This strategy was carried out both through acquisitions, as in the case of “Calzaturificio di Varese” in 1988, and through production licences as in the case of perfumes, spectacles and watches. In 1989 it was decided to enter into the sporting goods sector with the acquisition (near Treviso) of Nordica, an important producer of boots, skis, skates, skateboards and tennis rackets. The new business was not successful and it was sold in 2003. Until the 1980s all the Benetton products were made in Italy. The beginning of Benetton’s foreign production can be traced back to 1982 with factories established in France, Scotland and the United States. The US factory was linked to a failed attempt to enter in the North American market (Nardin 1987: 46). In the beginning of the 1990s, the factories in Scotland and in the United States closed and two new plants opened in Spain and Portugal. The share of the foreign production rose to 20 percent, a limited percentage if compared to that of other large Italian clothing brands. During the 1990s, in consequence of the growth of sales the number of Benetton’s Italian subcontractors increased and reached its maximum of 866 units in 2000. Therefore, in that period, production remained mainly in Italy and the strategy of producing abroad, rather than for reducing costs, was driven by the desire to move production closer to the consumption market avoiding the currency exchange risk. Starting from the second half of the 1990s, Benetton faced an increase in competition, in particular from Zara, H&M and Mango, whose products had generally lower prices.8 The yarn and fabrics produced by Olimpias became expensive compared to those sold in the market. Furthermore, the export advantages due to the weak Italian currency, the lira, ended in 1995 with the decision of Italy to peg the lira rate of exchange in order to enter the European Monetary Union two years later. The fixed exchange rate made it impossible for Italian companies to transfer the increase of production costs onto higher export prices and the progressive opening of the Eastern European markets to foreign investments induced Benetton to follow the strategy, already adopted by other Italian firms, to delocalize production first in Hungary, then in Romania and Croatia.9 Also Tunisia was interested by this process, mostly for cotton sweaters and jeans, in which this country is specialized. The manufacturing factories in France, Spain and Portugal lost their importance little by little and stopped their activity.

244   P. Crestanello and G. Tattara

4  The last decade: change of the collections timetable and productive internationalization At the beginning of 2000, Benetton speeded up the process of changing the production organization, in consequence of the strong competition mainly coming from Zara, H&M, and Mango, which are the main foreign brands to have their own stores in Italy. The process of restructuring was extremely fast: in 2003, 48 percent of the volume of production was still manufactured abroad and 52 percent in Italy. Production abroad increased in just one year, between 2004 and 2005, by 13 million items and the employment in Benetton’s Italian subcontracting firms shrank, from 2003 to 2005, by 3,100 workers.10 This great shift was due to the decision taken in 2004 to move production to China. The recourse to Asian suppliers with a large autonomy in managing a broader range of manufacturing functions, including the sourcing of inputs and sometimes logistics, is described as “full package production.” Benetton provides the design, often a simple sketch, and buys the final product that is delivered to its warehouse and then distributed to the stores. In 2007, Benetton’s full package production represented, in terms of volume, 37.6 percent of the total11 and the increasing importance of this form of sourcing has made Benetton much more similar to the large clothing international retailers (e.g. H&M, The Gap, Marks & Spencer) than to a clothing manufacturer. In 2005 Benetton’s organization shifted from a system based on productive units referring to the different product categories (such as wool, cotton, etc.), to a structure based on the different activities (such as design, quality control, marketing, etc), a move that underlines the change in the governance of the value chain (Annual Report 2006: 24; Camuffo et al. 2001). Also the structure and the number of collections changed radically. Until 2003, the production was based on two seasonal collections (spring/summer and autumn/winter) that were designed much in advance of the selling season and 80 percent of the production was decided on the basis of orders collected before the season by Benetton’s agents. The remaining 20 percent came from reorders. The products designed during the selling seasons “flashes” were a very small part of the production and were made just to “refresh” the shop windows. This organization did not permit taking advantage of the market opportunities, and was not encouraging consumers to pay more visits to the shops in search of the last fashion trends.12 Following the success of Zara, able to offer constantly updated products in its stores, Benetton changed its collections timetable. The traditional seasonal collection was split taking the names of Contemporary1 and Contemporary2. Each one of these collections has a time-­to-market that varies between four and eight months and is articulated in four launches: spring, summer, autumn and winter (Figure 10.1). Additionally, during the selling season, Benetton introduced three collections: “Trend” a collection more sensitive to the fashion tendencies with time-­to-market between one and four months and the collections “Just in time” and “Continuative items” that use standardised raw materials (“Continuative items” is manufactured from stock)

A global network and its local ties   245 and are brought to the market in a very short time (7 days if the products are made in Italy and 15 days if imported from abroad). While “Just in time” aims to satisfy fashion sensitive consumers, “Continuative items” guarantees that a collection’s core products are restocked in a very short time.13 The passage from a production planned well in advance to a flexible one, with a reduction of the time-­to-market and an increase in the number of collections, required a new selling organization. The independent retailers, in fact, have to bear the risk of the end of season markdown and they place their orders only after having seen the products. There is in fact the need for the agents to visit the retailers more than one time in a season to show the collections and this implies high transactional costs and difficulties in planning production. A direct control of the shops, instead, guarantees a better coordination of the entire value chain reducing the time needed for the independent retailers to decide on their purchases. For this reason, Benetton, in the last few years, has increased the number of its own stores that now sell about a quarter of the value of total sales. Furthermore, in the last two years Benetton invested a great deal of resources in retailing activities, opening new stores in new markets, giving economic incentives to the franchisees and linking the production, logistic and retailing units through a new information system, in order to receive information about the sell-­out and the retailers can have immediate confirmation and guaranteed delivery times for their orders.14 This shift of focus from production to retail activities confirms the transformation of Benetton from a manufacturing to a buying company.15

5  The present production organization In 2007, Benetton produced 145.2 million items (garments and clothing accessories, such as footwear, bags, belts, etc.), for a turnover of 1,856 million euros

Figure 10.1  New collections structure (source: Benetton website).

246   P. Crestanello and G. Tattara (Table 10.1). Other sales, concerning textile production sold to the independent clothing firms, royalties and other minor incomes, led the total Benetton’s turnover to a value of 2,085 million euros.16 About half of the Benetton’s production was sold in the Italian market, 36 percent in other European countries, 11 percent in Asia and a minor part in the Americas. Benetton sold its products in 124 countries17 through 5,800 mono-­brand stores, 95 percent of which are in franchising. Benetton owns only 300 stores that sell a proportionately larger amount of overall sale value, about 21 percent. In order to keep relations with its many independent retailers Benetton maintains a large network of agents (some of them own several shops) coordinated by a team of area managers. In 2007, the Group employed 8,896 workers of which 7,737 in various clothing activities such as textile (17 percent) and retailing (42 percent). Three thousand eight hundred workers, 43 percent of the total, were employed in Treviso headquarters and in ten textile and knitting factories located in various parts of Italy. The production activities are carried out by the Benetton’s affiliate Benind, which governs, through its own subsidiaries, factories and logistic platforms in Hungary, Croatia, Tunisia, and India (Table 10.2), and purchases full package products from the Benetton’s Asian suppliers through the Group’s trade company Asia Pacific. The garments manufactured abroad are imported from Benind and then sold to Bencom (the commercial division of the Group), which distributes them to the stores. In 2007 Benind sold raw materials to its foreign affiliates (Benetton India and Asia Pacific excluded) for 188 million euros and imported from all foreign companies finished garments and clothing accessories for 692 million euros. In the same year, Benind purchased garments from Italian subcontractors and paid dye-­work services for a total of 103 million euros. In 2007, 32.4 percent of Benetton’s production, in terms of volume, was produced in Asia, 20 percent in Tunisia and 28.5 percent in East European countries. Italy produced only 10.3 percent (Table 10.3). The choice of the outsourcing countries depends on several factors: labour cost, economic and fiscal incentives, availability of skilled labour, flexibility and time-­to-market. Shifting production from Europe to Asia has cut the unit cost of production because of the lower cost of labour, the use of cheaper local raw materials, and because several Asian countries have their currencies linked to the Table 10.1  Benetton Group turnover (in millions of euros) 2007

2006

2005

Clothing, accessories and footwear Yarns and fabrics Royalties Others

1,956.0 88.0 41.0 –

1,715.1 85.6 11.9 98.4

1,579.4 88.6 15.5 81.6

Total

2,085.0

1,911.0

1,765.1

Source: Annual reports: various years.

  36.3   62.9 139.9 231.0   2.3

Osijek

Labin

Sahline

Hong Kong

Gurgaon

Benetton Istria1

Benetton Tunisia1

Benetton Asia Pacific2

Benetton India1

Notes 1  outward processing production. 2  full package production.

Source: Benetton Group 2007 pp. 43–4.

Total

1

692.2

  36.8

Sibiu

Benrom Romania1

Benetton Croatia

183.1

Nagykálló

Benetton Hungary1

187.7

  0.2

–

  48.6

  30.9

  14.8

  18.7

  74.6

–

–

  91.3

  32.1

  21.5

  18.1

108.6

Cotton garments

Cotton garments, knitwear and accessories

Cotton garments, knitwear, dyeing

Knitwear, wool garments, samples (part of)

Wool garments, cotton knitwear, dyeing

Cotton garments

Garments and footwear

Purchases of finished Sales of raw material Manufacturing value Main activity products (1) and accessories (2) (3) = (1) – (2)

Table 10.2 Intra-trade between Benetton’s company Benind and its foreign affiliated companies in 2007 (in millions of Euros)

248   P. Crestanello and G. Tattara 1993

Spain, France

2004

Italy

Tunisia

Hungary

Italy

Others

2007

Croatia

China

Tunisia

Hungary

Croatia

Italy

Others

Figure 10.2 Percentage distribution of garments and accessory items produced by Benetton (number of items).

weak dollar.18 The increasing full package imports from Asia have reduced, as well, transactional costs simplifying the control of the production value chain. From 2003, in spite of the presence of a yearly sale growth of 8–9 percent, and even if a large part of production moved to Asia was to the detriment of East European countries, there has been a further reduction in the level of activity performed in Italy. This is because the recourse to full package production reduces the activities carried out by Italian workers (employed both by Benetton and by its subcontractors) such as quality control and logistics that were connected to the sourcing of part of the manufacturing production in Tunisia and in Eastern Europe. In addition, the amount of raw materials produced by the Benetton textile-­knitting division and sent to the European and Tunisian factories, has declined and this has further reduced the value of production made in Italy. At present, Benetton’s production is organized according to two supply chains (Figure 10.3). The first one uses Italian, European and Tunisian suppliers to produce fast fashion and more complex products, while the second one employs Asian suppliers for more standardized production, made on long runs and planned in advance. The majority of Asian production is imported to Treviso where, thanks to a fully robotized warehouse, it is efficiently stocked and sent to the worldwide shops. The warehouse in Shanghai serves mainly the Asian markets. Table 10.3  Number of garments and accessory items produced by Benetton 2006 Million of items Italy Eastern Europe Tunisia Asia Other countries Total

2007 %

█

Million of items

%

19.7 43.0 26.1 34.9 8.8

14.9 32.5 19.7 26.3 6.6

15.0 41.4 29.0 47.1 12.7

10.3 28.5 20.0 32.4 8.8

132.5

100.0

145.2

100.0

Source: interviews with Benetton staff.

A global network and its local ties   249 Productive and logistic units in Italy, Hungary, Croatia, Tunisia Productions: Contemporary1, Contemporary2, Trend, Just in time)

Logistic units in India, Thailand, China Productions: Contemporary1, Contemporary2, Continuative items, clothing accessories

Warehouse (Castrette in Treviso)

Warehouse (Shanghai)

Markets: Europe, Americas, Asia

Markets: Asia

Figure 10.3 The global value chain of the Benetton Group (source: interviews with Benetton staff).

5.1  The Italian production Between 2003 and 2007 the Italian share of Benetton’s production, in terms of volume, shifted from 38 percent (41.3 million units) to 10.5 percent (15 million units) with a reduction of 26.3 millions items19 (Table 10.4). Between 2000 and 2007 the number of Benetton’s Italian subcontractors has shrunk from 580 to 295, while the number of employees, between 2003 (first year for which data are available) and 2005, has declined from 8,249 to 5,136. In the last three years, the number of subcontracting firms in Treviso diminished from 208 to 116, while their employment in 2005 (last available data) was 2,085 (Table 10.5). At present 80 percent of Benetton’s Italian subcontractors are located in Veneto. Table 10.4 Benetton’s Italian subcontractors: number of firms, employees and production items Production (millions) Italian subcontractors Production Employees █ █ █ Total Italy Firms Employees Italy/total Per firm Per firm 2003 2004 2005 2006 2007

108.7 109.4 113.0 134.0 145.2

41.3 30.6 20.3 18.0 15.0

525 458 327 351 295

Source: data supplied by the Trade Unions.

8,249 5,884 5,136 n.a. n.a.

38.0% 28.0% 18.0% 13.4% 10.5%

78,667 66,812 62,080 51,282 50,847

15.7 12.8 15.7 n..a. n.a.

250   P. Crestanello and G. Tattara Between 2003 and 2007 the average dimension of the Benetton’s orders per subcontracting firm fell from 79,000 to 51,000 items, a volume of production that a subcontractor of average size (15 employees) can deal with in three to five months. The majority of the firms, that in the past used to work exclusively for Benetton, has now diversified their client portfolio. Furthermore, the only orders which remain in Italy are products made on small runs and with short delivery time. Nowadays, Italian subcontractors have the role to guarantee flexibility to the value chain, solving problems that arise in dealing with distant sourcing: transport delays, errors in production plans, product faults, etc. In 2006, 37 percent of Benetton’s Italian suppliers were located in Treviso representing a share of 30 percent of the number of the clothing subcontractors with an employment share of 40 percent in 2004 (last available year). We can estimate that in 2006 the number of garments produced in Treviso was about two million items (Crestanello 2008). In its headquarters and in the robotized warehouse located in the surroundings of Treviso, Benetton employs 1,800 workers involved in design, marketing, and logistic activities. Table 10.5  Benetton’s Italian subcontracting firms and their employees Italy

2000 2003 2004 2005 2006 2007

Veneto region

Firms

Employees

768 525 458 327 351 295

– 8,249 5,884 5,136 n.a. n.a.

█

Firms 580 395 354 272 276 240

Province of Treviso █

Firms

Employees

283 208 181 137 132 116

– 3,147 2,464 2,085 n.a. n.a.

Source: data supplied by the Trade Unions.

5.2  Production in Eastern European countries and in Tunisia Benetton owns in Europe four logistic-­productive platforms – two in Hungary, two in Croatia – plus one in Tunisia. Raw materials are distributed from Italy to the local subcontractors mainly specialized in sewing, ironing, and wrapping the final product that is sent back to Italy ready to be distributed to the shops. In Tunisia and in European countries all the production occurs according to an outward processing model that uses raw materials sent from the buyer and requires continuous technical assistance to keep up the quality and fulfill the delivery times. This is possible thanks to the presence of skilled employees placed at the end of the productive lines (Pickles et al. 2006; Crestanello and Tattara 2006). In 2007, the five platforms of Benetton employed 1,047 workers and the relative value chains involved 312 foreign suppliers with 19,500 workers (Table 10.6). The average size of the foreign firms is three times bigger than that

A global network and its local ties   251 Table 10.6 Benetton’s logistic platforms and subcontracting firms in East Europe and Tunisia, 2007 Benetton employees

Subcontracting Employees in the Number of firms subcontracting employees firms per firm

Hungary and Romania   281 Croatia   375 Tunisia   391

126   43 143

  9,200   2,800   7,500

73.2 65.1 52.4

Total

312

19,500

62.5

1,047

Source: interviews.

of the firms that work for Benetton in Italy and the reduction in the number of subcontractors is undoubtedly an element of organizational simplification. In these countries there are several Italian entrepreneurs who have been convinced by the Benetton’s management to transfer their activities abroad, so as to be able to continue to work for the Group; on the contrary their orders would have been interrupted (Crestanello and Tattara 2006). 5.2.1  East Europe 41.4 million garments are produced in the Eastern European countries, equal to 28.5 percent of the whole volume made from Benetton. Production is organized by four logistic-­productive platforms. Those of Nagikallo in Hungary and of Sibiu in Romania employ 281 workers and manage a network of 126 suppliers localized in several Eastern European countries (Hungary, Romania, Poland, Moldavia, Slovakia, and Ukraine) for a total of 9,200 employees. In addition, the Hungarian plant carries out cutting and printing operations. The third and fourth platforms are located in Croatia (Osijek and Labin) and carry out dyeing and knitting, employing 375 employees. The garments output is made by 43 local suppliers with 2,800 employees.20 In the last few years, production in Hungary has started to decline,21 while that in Croatia has increased, specializing in fast response production. For this reason Croatia today has become a formidable competitor for Italian subcontractors (Crestanello 2008). 5.2.2  Tunisia In Tunisia, the local network of subcontractors, mainly specialized in sewing and upstream operations, is managed by the Benetton’s platform of Sahline. Here Benetton owns two factories with 391 workers which make cotton fabric, and carry out knitting, dyeing and special finishing (like printing, stone wash, stone bleach, etc.). The subcontractors’ network is made of 143 firms employing 7,500 employees that produce yearly 29 million garments, mainly knitwear.

252   P. Crestanello and G. Tattara The production in Tunisia has recently increased and 20 million euros are to be invested in a new factory that will produce knitted cotton fabrics. The new production, estimated at 3.6 million kilos will serve local subcontractors, with a decline in unit cost and increase in efficiency (the network forecasted production will increase by 21 million garment units). The present preference for the Tunisian suppliers rather than the Eastern European is due to the fiscal benefits allowed to the new factories,22 to a lower and stable cost of labour and to the high competences available in Tunisia. 5.3  Production in Asia In Asia production is commissioned from more autonomous local subcontractors which purchase directly the necessary raw materials, often from selected producers, whose production process is monitored by Benetton. In East Europe and in Tunisia the subcontractors receive the fabric (made in Italy or in other countries, such as Turkey) directly from Benetton, and all accessories, as well with a detailed manufacturing schedule. In Asia are produced, as well as garments, all Benetton’s accessories such as shoes, bags belts, umbrellas, toys, perfumes, etc. The imports from Asia have largely increased in the last few years and today they represent in terms of volume 47.1 million items equal to 32.4 percent of Benetton’s total production.23 Benetton’s network of Asian suppliers is managed by Asia Pacific, an affiliated trade company established in Hong Kong, which controls the logistic platforms of Shenzhen and Shanghai. Only a small part of the products manufactured in Asia is sold on the domestic market, the majority is imported in Italy (231 million euros out of a total of 248). A growing amount of Asian production is manufactured in Vietnam, Cambodia and Bangladesh and is governed by the recently established logistics platform of Bangkok. At present, the number of garments produced in these three countries is around two million items, forecasted to reach 18 million items by 2011. The increase of production in these countries is explained by a labour cost which is lower than that of China. Full package production makes the outsourcing organization simpler and, in the case of standardized products made from stock, such as a great part of the accessories, permits a short delivery time too.24 In India production occurs according to a network model that combines vertical integration and subcontracting. In Gurgaon (close to New Delhi) the Benetton factory employs 300 workers and organizes the work of many local subcontractors that employ 5,000 workers. The subcontractors receive the raw materials and a precise technical schedule and they produce about 50 percent of the entire Benetton Indian production. Differently from China, almost all the production made in India (about six million units) is targeted to the domestic market.25

A global network and its local ties   253

6  New markets and production partnerships Benetton’s strategy not only aims to reduce production costs, but also to expand the number of mono-­brand stores, mainly in new markets which offer better business opportunities. In the last few years, in fact, the Russian, Turkish, and Asian markets, to which the Mexican market can be added a few years later, have sustained Benetton’s sales. The rate of growth of Benetton revenues in these markets has been 27 percent in 2008 (nine months) and 40 percent in 2007, while revenues in Europe have grown respectively at 4 percent and 12 percent (Benetton Group 2008: 7, 2007: 17). In China, at the end of 2006, Benetton had opened more than 100 shops and two mega-­stores in Shanghai and Beijing. In India where Benetton has been present since 1990, 106 shops have been opened in 43 towns, while in Russia the 150th shop was inaugurated in 2007 (Benetton Group 2007: 17). In the last two years economic resources have been invested in opening new stores managed directly by Benetton. The majority of these new shops has been set up in the new markets such as India, Korea and Russia, where the franchising system is difficult to introduce. In some countries Benetton uses productive and commercial partnerships to expand its presence faster. In China, an agreement has been signed with Hemply International Company which will open 150 Sisley shops in the next five years. In India, Benetton’s garments are distributed thanks to a deal with the Trent company, which is part of the Tata Group, while in Mexico a commercial agreement has been made with the distribution company Sears. As regards the licensed products,26 Benetton has two agreements in Turkey: one is concerning house products under the Sisley brand with the Zorlu Group, leader in this sector, the other with the Boyner Group for the production and distribution of garments also in Turkey’s neighbouring countries.

7  Conclusions In the 1970s Benetton’s competitive advantage was grounded on product innovation based on the use of bright colours, on delaying the dyeing process as long as possible and on the ability to manage a network of small subcontractors located in important industrial districts of Italy. These firms were linked to Benetton by an exclusive pact and by informal and long-­term relationships based on trust. Production costs were relatively low – at that time Italian wages were noticeably much lower than French and German ones – and Benetton’s subcontractors benefited from long production runs and working at full production capacity. Benetton in fact granted a constant flow of orders to its suppliers that could confidently make new investments and update their machinery. The Benetton success has been inextricably linked with the development of the Treviso district where the largest share of Benetton production was traditionally localized. The innovations that were at the roots of its success have developed from immersion in the crafts of production, in close contact with

254   P. Crestanello and G. Tattara people who had daily practice with design, machinery, raw materials, and accessories.27 At the same time the familiarity that derived from the direct selling experience of Luciano Benetton has contributed to build a retail structure that, without direct capital investment, mushroomed in just a few years and now is one of the largest retail structures in the European clothing market.28 During the last decade and under a strong competitive pressure, Benetton has focused more on price, moving a large share of its production abroad, and on the need to change the timetable of its collections, introducing many new products during the selling season. There is an evident trade-­off between these two targets. While the recourse to cheap foreign suppliers risks increasing the delivery time and creates rigidities in production organization, a shorter product life cycle requires more flexibility. This more complex organization requires a tight and efficient integration between the production, logistic, and retailing activities. A difficult task if, as in the case of Benetton, the large majority of mono-­brand shops is run by independent retailers. To face this challenge Benetton has increased the number of its own shops and at the same time has recently granted greater mark-­up margins to the independent retailers (bearing a heavy cost) in order to strengthen the retail network and receive from the retailers better feedback (information on sell­out, support for the many seasonal proposals, etc. – see Benetton Group 2007: 42). The retail network has thus experienced a process of increasing hierarchical control; whereas the production organization is moving towards arm’s-length market relations, a phenomenon confirmed by the large amount of full package production outsourced abroad, mainly from Asia. Such a radical shift of focus from production to retail activities has required new competencies and has changed the relationships with the economic actors in the different stages of the supply chain. In particular, this change has negatively impacted on the clothing district of Treviso where the largest share of Benetton production was traditionally localized. Not only the number of Italian Benetton’s subcontractors has reduced, substituted with foreign, low-­wage suppliers, but also the relationships based on trust, that had in the past linked the Group to its subcontractors, are now much weaker. Many of these firms have diversified their client portfolio and now Benetton is no longer their exclusive buyer. The clothing district of Treviso is today specialized in activities such as design, quality control, logistics, marketing, and in producing short runs with quick delivery times, having lost the large volumes of production that Benetton and other leading brands now get from abroad.29

Notes   1 We wish to thank Pietro Arnaboldi, Mara Di Giorgio, Giuliano Franco, Diego Favaro, and Lorenzo Zago of the Benetton Group staff and Tiziana Crosato and Sergio Spiller, of the Trade Unions, who provided us with the necessary information. We also thank participants to the Prin “Dynamic Capabilities between firm organization and local systems of production”, to the Sase Meeting Copenhagen, June 29, 2007, and to other seminars held at the University of Venice, CDS Trivandrum, IDIS Delhi, and Bank of Italy where a preliminary version of this chapter has been discussed.

A global network and its local ties   255   2 The Carpi district, which had the Italian knitwear’s leadership at that time, was instead specialized in a production of cut and sew knitwear with a very wide offer of models. So Benetton, differently from Carpi, offered a limited number of models, using the colours as strategy to differentiate its products (Brusco and Crestanello 1995).   3 Gaeta in the introduction of Nardin’s book (1987: 8) defined the Benetton’s organization as Fordist and the Ford’s expression: “any customer can have a car painted any colour that he wants so long as it is black” could have been adapted in Benetton’s case to: “any sweater he wants so long as it is basic and fully coloured.”   4 The mark-­up of Benetton’s stores was 70 percent against an average of 100 percent applied by the other stores.   5 In 1981, Aldo Palmeri, a Bank of Italy officer, became CEO of Benetton. Two years later, Giovanni Cantagalli, another manager coming from an American multinational company, was recruited in charge of personnel and shortly a team of managers was created to reorganize the Benettons’ family-­owned company.   6 In 1987 Benetton tried unsuccessfully to acquire Lanerossi, a large Italian textile company, which was its main supplier of carded yarn. The bidding winner was Marzotto which, after this acquisition, became the largest textile company in Italy.   7 benetton.linefeed.org/archives/000058.html.   8 Zara is a vertical integrated firm both upstream, with factories located in Galicia, and downstream, with a large network of mono-­brand stores. All the outlets transmit to the headquarters a continuous flow of information on their sell-­out, allowing in this way a quick production response to the market’s requests.   9 According to Benetton in that period “the increase of Ebitda margins came from a significant reduction of industrial costs due to the production delocalization to East Europe” (Benetton Group 2003). 10 The loss of jobs was the result both of the closing of firms and the reduction of employees per firm. Furthermore, several subcontractors stopped working exclusively for Benetton and had to diversify their customer portfolio to face a situation of uncertainty. 11 At the end of the first decade of the twenty-­first century, 90 percent of the Benetton’s full package production comes from Asia. 12 Zara customers pay an average of 17 visits a year to the brand’s shops, against an average of four visits for competitors (Gallaugher 2008). Furthermore, it is a strong consumption incentive to know the product you see on the shelves might be no longer available the day after. 13 The two collections Contemporary1 and Contemporary2 represent 75 percent of the annual production sold under the brand United Colors of Benetton. The collections Trend, Just in Time and “Continuative items” have instead a greater relevance for the fashion-­oriented brand Sisley and they represent 50 percent of the total production under this brand. 14 Starting from 2005 Benetton has renewed its information system. Furthermore, Benetton monitors constantly what is selling and what is not in 500 mono-­brand shops, to have a better production planning and to keep its inventory low. 15 This change was anticipated in 2005 by the CEO Silvano Cassano who explained how Benetton was shifting investments from the manufacturing area to the retailing one (Benetton Group 2005). This strategy was also confirmed recently by a financial analyst: “Benetton is implementing new organisational procedures, with a better coordination among all the stages of the value chain, without focalising exclusively on the production efficiency. The result will lead to improve the variety of the offer and to reduce the lead time. Benetton is changing its own organisation with the focus on retailing and not on production . . .” (Benetton Group 2007). 16 Olimpias, the Benetton’s textile company, sells two-­thirds of its yarn and fabric production to the Group’s clothing division and one-­third to independent firms.

256   P. Crestanello and G. Tattara 17 More than 2,000 shops are located in Italy, 2,000 in other European countries, 500 in Asia and 250 in the Americas. 18 According to an interview carried out with a manager of Benetton, the production delocalised in India has permitted a reduction in production costs of 20 percent (see also Na Thalang 2007). 19 The average value per garment is higher in Italy than in the foreign countries, because of the more complex products made by Italian subcontractors. 20 In 2000 the Osijek factory received 24 large looms from the Benetton’s factory of Troyes (France) which was closing. 21 The decades of fiscal benefits of the Hungarian factory are ending and this, together with the abandonment of the sporting goods production (made in this factory), explain the progressive loss of importance of this productive pole. 22 In Tunisia, Benetton and other clothing firms like The Gap, Lee Cooper and Yves Saint Laurent benefit from duty exemptions on imports, low VAT on the imported capital goods, low taxes on profits (and total exemption on the reinvested profits) and other incentives (Tunisie 2008). 23 It has to be considered that in 2004 Benetton imports from Asia were only 14 million euros against 231 million in 2007. 24 As Benetton says: “A particular attention is given to the full package production that regards specific markets, such as China. The expansion of this production has produced a competitive advantage in terms of lower costs and shorter lead time” (Annual Report 2004: 9). 25 It mainly produces clothing for children and apparel accessories. 26 Benetton production licenses regard different products: spectacles, perfumes, house products. 27 The second generation of Benetton’s managers (and the sons of the four Benetton founders with them) do not have the same level of familiarity with the tools and materials that their parents had when they started the business. 28 See Lane (2004), for an approach to the district as an evolving structure under pressure from globalization. 29 Nowadays local producers deal with high fashion and more sophisticated products, such as seamless garments, and relatively capital intensive manufacturing operations such as CAD-­CAM, dyeing and printing (Gomirato 2004).

Bibliography Abecassis-­Moedas C. (2006), Integrating design and retail in the clothing value chain. An empirical study of the organisation of design, International Journal of Operations and Production Management, 26(4): 412–28. Adam G. (1971), New trends in international business: worldwide sourcing and dedomiciling, Acta Oeconomica, 7(3–4): 349–67. Baden S. (2002), Trade Policy, Retail Markets and Value Chain Restructuring in the EU Clothing Sector, Prus Working Paper, no. 9. Belussi F. and M. Festa (1990), L’impresa rete del modello veneto: dal postfordismo al toyotismo? Alcune note illustrative sulle strutture organizzative dell’indotto Benetton, Oltre il Ponte, 31: 3–81. Benetton L. and A. Lee (1990), Io e i miei fratelli, Sperling & Kupfer, Milano Benetton Group (anni vari), Bilancio consolidato. Benetton Group (1990), Benetton Group Annual Report, Benetton, Ponzano Veneto. Benetton Group (2003), Benetton presenta la linee guida di sviluppo 2003–2007, Press release 9 December, Benetton, Ponzano Veneto Benetton Group (2004), Benetton Group Annual Report, Benetton, Ponzano Veneto.

A global network and its local ties   257 Benetton Group (2005), Approvati dal CdA Benetton i risultati dei primi nove mesi 2005, Press release 11 November, Benetton, Ponzano Veneto. Benetton Group (2006), Benetton Group Annual Report, Benetton, Ponzano Veneto. Benetton Group (2007a), Benetton Group Annual Report, Benetton, Ponzano Veneto. Benetton Group (2007b), Analisti finanziari, 19 September, Benetton, Ponzano Veneto. Benetton Group (2008), Benetton Group Nine Months Report, Benetton, Ponzano Veneto. Berger S. (2006), How we Compete: What Companies Around the World are Doing to Make it in Today’s Global Economy, Currency/Doubleday, NewYork. Blois K.J. (1972), Vertical Quasi Integration, Journal of Industrial Economics, 20. Brusco S. and P. Crestanello (1995), Tre diversi sistemi produttivi locali: un confronto fra Carpi, Thiene e Benetton, Oltre il Ponte, 52: 5–21. Camuffo A., P. Romano, and A. Vinelli (2001), Back to the Future. Benetton Transforms its Global Network, MIT Sloan Management Review, 43(1): 45–52. CEPS (2005), An Enlarged Community and the Outlook in the Candidate States, Final Report. Part 1. The Textiles and Clothing Industries, Centre for European Policy Studies, Brussels. Ciappei C. and A. Sani (2006) Strategie di internazionalizzazione e grande distribuzione nel settore dell’abbigliamento. Focus sulla realtà fiorentina, University Press, Firenze. Crestanello P. (1999), L’industria veneta dell’abbigliamento: internazionalizzazione produttiva e imprese di subfornitura, Franco Angeli, Milan. Crestanello P. and G. Tattara (2006), Connessioni e competenze nei processi di delocalizzazione delle industrie venete dell’abbigliamento e delle calzature in Romania, in G. Tattara, G. Corò, and M. Volpe (eds.), Andarsene per continuare a crescere, Carocci, Rome. Crestanello P. and P.E. Dalla Libera (2003), La delocalizzazione produttiva all’estero nell’industria della moda: il caso di Vicenza, Economia e società regionale, 82(2): 5–46. Crestanello P. (2008), I processi di trasformazione dell’industria dell’abbigliamento, Economia e società regionale, 101(1): 33–66. Dapiran P. (1992), Benetton. Global Logistics in Action, International Journal of Physical Distribution & Logistics Management, 22(6): 7–11. Dicken P. (2007), Global Shift. Mapping the changing contours of the world economy, 5th edn, Sage Publications, London. Dunford M. (2006), Industrial Districts, Magic Circles, and the Restructuring of the Italian Textiles and Clothing Chain, Economic Geography, 82(1): 27–59. Favero G. (2005), Benetton. I colori del successo, Egea, Milano. Finger J.M. (1976), Trade and domestic effects of offshore assembly provision in the U.S. tariff, American Economic Review, 66(4): 598–611. Finger J.M. (1977), Offshore Assembly Provision in the West German and Netherlands Tariffs: Trade and Domestic Effects, Weltwirtschaftliches Archiv, 113(2): 237–49. Gallaugher J. (2008), Zara Case: Fast Fashion from Savvy Systems, online, available at: www.gallaugher.com/Zara%20Case.pdf. Gereffi G., J. Humphrey, and T. Sturgeon (2005), The Governance of Global Value Chain, Review of International Political Economy, 12(1): 78–104. Gomirato E. (2004), La delocalizzazione dell’abbigliamento in Romania: il caso Stefanel, Economia e società regionale, 86: 36–91. Graziani G. (2001), International subcontracting in the textile and clothing industry, in S.W. Arndt and H. Kierzkowsky (eds.), Fragmentation. New Production Patterns in the World Economy, Oxford University Press, Oxford.

258   P. Crestanello and G. Tattara The Hindu Business Line (2007), Trent, Benetton join hands for strategic partnership, September 19, online, available at: www.blonnet.com/2007/09/19/stories/ 2007091951220500.htm. Lane D.A. (2004), Innovation led policies for clusters and business networks: Globalization, ontological uncertainty and degeneracy, paper presented at the conference “Knowledge-­based regional development and innovation”, Florence, 25–26 November. Nardin G. (1987), La Benetton. Strategia e struttura di un’impresa di successo, Edizioni lavoro, Rome. Na Thalang J. (2007), Outsourcing to India enables Benetton to sell 20% cheaper, The Nation, online, available at: www.nationmultimedia.com/2007/05/14/business/ business_30034122.php. Pickles J., A. Smith, M. Buck, P. Roukova, and R. Begg, (2006), Upgrading, changing competitive pressures, and diverse practices in the East and Central European apparel industry, Environment and Planning, 38(12): 2305–24. Sasi A. (2007), Textile firms leverage Italian tie-­ups to move up value chain, July 18, online, available at: www.blonnet.com/2007/07/18/stories/2007071850101100. htm. Tokatli N. (2008), Global sourcing: insights from the global clothing industry. The case of Zara, a fast fashion retailer, Journal of Economic Geography, 8(1): 21–38. Tunisie (2008), How to invest in Tunisia?, November 12, online, available at: www.animaweb.org/en/pays_tunisie_pourquoiinvestir_en.php.

Index

Page numbers in italic refer to tables. Page numbers in bold refer to figures. abstract global space 35, 36 adaptive firms 113, 114 ADSL technology 123 agents as interactors 34–5, 57 agglomeration economies 164, 165, 166, 167, 170, 178, 181, 182 AIDA database 164, 166, 194 Amadeus database 149, 153 Apollo 75, 81, 82 appropriability mechanisms 21, 141, 154, 157 ATT estimation, offshoring 220–7 attenuation phenomena 166, 177, 179 automotive industry 79–80, 83 Axiomatic Design Theory 32, 37–8 Babbage, Charles 5, 8, 15, 17 Bayesian model 164, 171, 180 beliefs 35, 45, 49 Benetton Group 239–54; Asian production 252; Eastern European production 250–1; history 240–5; Italian production 249–50; new markets 253; production organization 245–52, 254; production partnerships 253; productive internationalization 244–5; retail operation 240–3, 245, 246, 254; Tunisian production 250, 251–2; value chain 242, 243, 244, 245, 249, 250 broadband 123, 138 building blocks of knowledge 51–2 Bureau Van Dijck 164, 166, 194 business methods patents 142, 145, 150 cannibalization of customer base 71, 74, 82 capital investment in foreign subsidiaries 197, 200, 202

capitalism 11–12, 13, 14, 18, 19 car manufacturers 79–80, 83 Cartesian behaviours 34, 37 Cisco Systems 76, 78, 83 classical economics 4–5, 8, 12 closed interfaces 74, 84–5 clothing market 239 Cobb-Douglas production function 10 cohesive entities 34 combinatorial explosion 31, 33, 48, 54, 57 combinatorial spaces 33, 53, 57, 58 commercial capitalism 18 commercialization 16 Complex Product Systems (CoPS) 69–70 complexity: and dynamic capabilities 93–4; modelling of 94–112; and modularity 71, 73; product 39, 40–1, 67; technological 89, 93, 120 components, product 31, 32, 35, 41–2, 80 compositionality 32, 33, 57, 58 co-operation between firms 6 co-operation of labour 4–5 counterfactual firms 214–17, 219, 227 Count-Panel-Data models 148 cumulative system technologies 142, 147, 155 currency exchange risk 243 customer needs (CNs) 37–9, 42 decentralized systems 48–50, 51 decomposability 44, 46–7 decomposition of product development process 37, 39, 40, 44–8 delocalization 212, 243 design churn 45 design parameters (DPs) 37–9, 42 Design Structure Matrix (DSM) 32, 42–4

260  Index DID-PSM methodology 211, 216, 219, 221, 222, 227 difference-in-differences see DID-PSM methodology differentiation 165, 193 diffusion and innovation 12, 13 digital division of labour 13 division of labour 4, 5, 6, 8, 18 domestic firms 198, 199 dynamic capabilities 92–4, 113 dynamic mapping 32, 33, 37–40, 53–5 economic production 35 economic theory and firms 2–5 economy structure model 105–7 employment dynamics of workers 211 employment growth of firms 164, 165, 170, 175, 179, 181 endogenous economic development 10, 163 entrepreneurs 4, 5, 16, 188, 191 epistemic base 53, 54 epistemic uncertainty 70 EU Industrial Research Investment Scoreboard, 2004 149, 152, 157 European Patent Classification (ECLA) 150 European Patent Convention (EPC) 142, 145, 156 European Patent Office (EPO) 141, 142–4, 149, 150–2, 156 European software patent 145 European Union, patenting 142–4 evolutionary economics 10, 90, 93, 141 exogenous uncertainty 91 export intensity 194, 195, 196–8, 202 export intensive firms 198, 199 exporting 192, 193–4, 202–3 fashion market 239 FDI: approach and theory 188, 189, 190–1; and export intensity 193–4, 196–8, 202–3; and firm size 196–7, 201, 202; and foreign markets 192–3, 202; and foreign subsidiaries 200; and geographic location 192–3; geographical localization of 203; and multinationals 212; and offshoring 210, 212–13; and stage theory 192–3 feedback loops 42, 46, 50, 55 FFII (Free Foundation for Information Infrastructure) 150 fibre optics 123 firm size 6–7; and exporting 188, 192, 196, 202; and FDI 196–7, 201, 202

firms: boundaries 5–7; co-operation between 6; co-ordination of activities 6; dynamic capabilities 92–4; and economic theory 2–5; employment growth of 164, 165, 170, 175, 179, 181; governance structure of 90, 91; and innovation 13, 91–2; and the market 5; non-market relationships 5–7; size 6–7; technological search process 94, 103, 105, 112–13, 114; theories of 2–5 foreign direct investment see FDI foreign markets 188, 189, 192–3, 197, 202, 203 foreign subsidiaries 192, 193, 195, 196–8, 200–2 functional requirements (FRs) 37–9, 42 GAUSS database 149, 150–2, 157 Gaussian linear model 171, 172, 173, 175 general purpose technology 18 generative nature of spaces 33, 48, 51, 53, 56 geographic agglomeration 163 Geographic Information System (GIS) 164, 166 geographic localization of FDIs 203 geographic location and FDI 192–3 geographical span of operations 188, 201, 203 global networks 239–54 global production sharing 17 global workspace 34–7, 47, 49; for producers 35, 37–44; and product development 46, 47, 49, 51, 58; production process 34–7 globalization 16–19, 198, 211 globalization gap 200, 203 governance inseparability 91 governance structure of firms 90, 91 government intervention 16, 182 hard disks 124 hardware patents 153, 155 hardware platforms 76–7 Herfindahl index 168 hidden information 40–1, 45, 46 hierarchy-based mechanisms 90 high-dimensionality of search spaces 36–7, 48, 51, 53, 55 horizontal integration 242, 243 IBM 78, 82 idea space 53 ideational kernel 33, 51, 53

Index  261 imitation activities 92 immaterial relations 94 import competition 210 incremental innovations 83, 84, 92 incumbents’ strategy 81–4 indirect exporting 187, 189 industrial capitalism 18 industrial clustering 166, 167, 182 industrial districts 6, 177, 182, 187, 194, 228 industrial organization 89, 99, 113 Industrial Revolution 7, 11, 18, 120 information exchanges 39, 45, 50 information flows 34, 45 information processes 50–3 information processing in decentralized systems 49 information structure analysis 39 innovation: capabilities of firms 91–2; and capitalistic economic development 10; definition 11–14; measuring 14; modelling 105–9; and patents 14, 141, 142, 146, 154; and platform competition 67, 69, 74, 83, 84; process 107–9; strategy for internationalization 189 innovation equation 147, 148 Innovation-related internationalization model (I-M) 189 integral research strategies 95, 101–2 intellectual property rights (IPRs) 66, 154 interacting entities 34–7 international: competition 188; division of labour 18, 210, 211; fragmentation of production 210, 211, 213; Patents Classification (IPC) 150; Standard Industrial Classification of All Economic Activities (ISIC) 151; trade 16, 18, 211, 213, 224 internationalization: and Benetton Group 244–5; of capital markets 16; geographical span of operations 188, 201, 203; and inventors 142; in Italian medium-sized firms 187–204; of production 213; strategies 198–200, 228 internationalized firms 198, 199 Internet 11, 79, 82 Internetwork Operating Systems (IOS) 76 intervention 16, 182 invention-innovation-diffusion 12, 13 inventions 142, 144, 145, 147, 152, 156 IPC classes 156 IPC codes 150, 151 isoquants 10 Italian manufacturing firms 164, 166–7, 177, 180, 210–28

Italian medium-sized firms 187–204 iterations, in design process 42 Japan Patent Office (JPO) 141 knowledge: capital 146, 154; codification 73; in technology and science 14–16; processes 34, 50–3; Knowledge Production Function (KPF) approach 146; specialisation 66, 67; spillovers 163, 164, 178, 181; knowledge-based decomposability 44, 46 labour market and offshoring 210–14, 227–8 labour productivity growth 165 learning bottleneck 56 learning-by-doing 120, 125 learning-by-interacting 92 learning-by-using 120, 121 Linux kernel software 78–9, 83 Local Area Networking (LAN) industry 74, 75–8, 82 local economic growth 163 Local Labour Systems (LLSs) 163, 164, 165, 166, 177, 178, 179 localisation economies 163, 164, 165, 166, 168, 177, 178, 180, 181 localized production 253, 254 long differences 170 macroeconomic production function approach 14 managerial competencies 189 MAR economies 164, 168 market share 125, 126, 130–8 market-based mechanisms 90, 91 market-seeking approach 193, 200 Marx, Karl 8–9, 12 material offshoring 211 material relations 94 material reproduction 121 mechanisation 5, 9 mental labour 5, 8 Microsoft Windows 77–8, 82 minicomputers 80, 83–4 mobile phone industry 80–1, 83, 84 modular: architecture 42, 46, 49, 72, 74, 84–5; product design 70–2; research strategies 95, 101; modular-closed platforms 75–8, 82–3; modular-open platforms 75, 81–2 modularity 46–8, 49, 70–2, 73–4 monopolistic position and patenting 144, 147

262  Index Monte Carlo Markov Chains (MCMC) 173 morphogenetic approach 48–53, 58 multinationals and FDI 212 multiple iterative parallel evolutionary strategy 49–50 multiple search spaces see search spaces national systems of innovation 14, 16 near-decomposability 48, 49 nearest neighbour matching (NNM) algorithm 217 negative binomial model 149 neoclassical model 3 neoclassical-mainstream school 10, 19 network model of internationalization 190, 191 networks of interdependencies 42–4, 50–3 NK models 94–112 non-maximising models 3 non-modular architecture 85 non-modular-closed platforms 80–1, 83–4 non-modular-open platforms 78–80, 83 non-proprietary interfaces 81, 83 non-proprietary standards 74, 75 novelties 33, 50, 51, 59 object structure, economy model 114–15 offshoring 210–28 old technologies 120–5, 137–8 oligopolistic markets 13 open interfaces 74, 85 open source code 79 open standards 74, 75, 76, 82 Ordinary Differential Equations (ODE) 130–4 organization architecture 41–2 organisational: interfaces 72–4; learning 93, 190; modularity 73; specialization 67, 68; structure 40 Osiris database 149, 153 outsourcing 90, 91, 92, 211, 212, 213, 246 ownership, location and internalization (OLI) 190–1 patents 14, 16, 141–57 PC industry 82 performance dynamics of competing technologies 125–38 pioneer firms 198, 199, 203 platform competition 66–85 platform typology 74–81 Poisson panel data model 148–9 posterior predictive analysis 173, 175, 176, 177

preferences 35 proactive firms 113 problem-solving activities 31–58 process variables (PVs) 38–9, 42 product: architecture 40, 41, 72; complexity 39, 40–1, 67; components 31, 32, 35, 41–2, 80; development 31–59; product as complex systems 69–70 production: systematic 18; offshoring 210–28; process 10, 31–59 productivity forces 120 propensity score matching estimator see PSM propositional knowledge 53, 54 proprietary interfaces 83 protocols 50, 51 pseudo-NK model 96–7, 99–109, 112 PSM 215–17, 219, 221, 222, 227 radical innovations 83, 92 random coefficient model 164, 171 R&D: economies of scale in 91; expenditure 146–7, 152, 154–5, 157; government intervention 16; scoreboard 149, 152, 157 recursiveness 32, 33, 52, 55, 57, 58 relational topologies 34 remodularization 75, 82 research spaces 94, 95 research strategies 48–50, 94, 95 resource endowments 35 restless capitalism 12 Ricardo, David 8 rules 52, 53–5 sailing ship effect 119, 122–3, 124 Schumpeter, Joseph 10, 11–12, 13, 90, 141 search activities 32, 34, 45, 55, 58 search spaces 31–59, 95 service offshoring 211 Shannon-Wiener index 168 shocks 50, 51, 113 skill composition 210–28 skill intensity of jobs 211, 212 small- and medium-sized enterprises (SMEs) 187–204 Smith, Adam 4, 8 software industry 76, 77–8, 82 software patents 141–57 software platforms 76 spatial agglomeration economies see agglomeration economies spatial externalities 163

Index  263 spatial proximity 163–82 specialisation 40, 91, 113, 165, 178, 182 spin-offs 16 stage theory 187–204 state-financed science and technology 11 Stefan Wagners database 150 stochast behaviours 34, 37, 51, 55 strategic patenting 144, 147, 155 subcontracting 240–54 sub-spaces 35–6 substitution 10, 70–1, 81, 121 Sun Microsystems 75, 81–2 sunk costs 120, 121–2 superconductors 123–4 superparamagnetic effect (SPE) 124 symbolic reproduction 121 systems integrators 83, 85 technological: change 71, 91, 92, 210–11; competition 120, 125–30; complexity 89, 93, 120; frontier 92; interdependencies 99–109; knowledge 15, 121, 146; landscape 102–5, 107, 110, 111; paradigms 10–11; persistence 119–38; platforms 66–85; search process 94, 103, 105, 112–13, 114; spillovers 163; variety 119, 120–2 technology 7–11; adoption 13; and change 8, 10, 12, 18–19; coexistence of old and new 120–2; competition between old and new 122–5; creation and innovation 11; development path 91; and organization 90–4; reproduction 121; and rules 52; and science 7–8; and search processes 53, 54; transfer 16 Torvald, Linus 78–9, 83

total factor productivity (TFP) 165, 166 trade flows 210 transition matrix for internationalization types 199 transition matrix for span of internationalization 201 transportation technologies 17 typology of platforms 74–81 unceasing dynamic mapping 37–40 United States Patents and Trademark Office (USPTO) 141, 142, 147 UNIX 75, 78, 82 unskilled labour and offshoring 226–7 Uppsala Internationalization model (U-M) 189, 192 urbanisation economies 163, 164, 165, 166, 168, 180 useful knowledge 53–4 value chain 210, 242, 243, 244, 245, 249, 250 variety effects 165, 179, 180 vector of attributes 35, 37, 42 vector transformations 35 vertical disintegration 90 vertical integration 90, 91, 92, 242, 252 visible information 40–1, 46 workstation industry 75 World Intellectual Property Organization (WIPO) 141 World Trade Organization (WTO) 17 Zara 240, 243, 244