Regional knowledge creation in a global industry An empirical agent-based model of the Austrian semiconductor industry

Transcription

1 Regional knowledge creation in a global industry An empirical agent-based model of the Austrian semiconductor industry M. Paier, M. Dünser, A. Unger AIT Austrian Institute of Technology GmbH 6 th European Conference on Corporate R&D and Innovation (CONCORDi 2017) September 2017, Seville, Spain

2 Motivation and background Complexity of the innovation system, w.r.t. technological, economic, social, political aspects etc. challenges understanding and forward-looking analysis Agent-based models (ABMs) are a way to describe path-dependent evolution of systems taking into account underlying micro-dynamics. We observe an increased need for legitimization of policy interventions. Ex-ante impact assessment (IA) of Research, Technology & Innovation (RTI) policy initiatives is enforced by law, including technological aspects. Hence, policy makers call for advanced methods and quantitative tools to support ex-ante IA. We propose a use case of ABM for application in IA Silicon Austria a new policy initiative is being prepared to support the Austrian microelectronics/semiconductor (MESC) industry. ( 2

3 Overview Objectives and approach Model overview Agents, relations, processes Empirical background Analysis of policy scenarios Summary and concluding remarks 3

4 Objectives and approach Development of an industry-specific, empirical ABM of knowledge creation Simulation of potential policy impacts on the industrial knowledge base Scenario analysis referring to specific RTI policy instruments under discussion for the Austrian microelectronic/semiconductor (MESC) industry (Geyer 2015, Caldera et al. 2015) Strengthening internationalization and international networking of Austrian business R&D activities Strengthening public R&D infrastructure in Austria (universities and RTOs) Improving framework conditions for the regional/local industrial ecosystem 4

5 Challenges Strong theoretical and empirical foundation lends credibility to ABMs in support of policymaking Innovation literature, expert validation, empirical calibration, initialization Balancing level of detail of knowledge creation model (Paier et al. 2017) Endogenized R&D strategies Output quantities: patents, knowledge stock, innovations, networks Explicit account of MESC-specific characteristics (Paier et al. 2016) Rapid technological change, high entry barriers, global value chains, strong international competition Agent location: AT, INT Agent type: IND-RTO-EDU Technology space (semiconductor devices +) Influence of the value chain on R&D collaboration 5

6 Model architecture Agents Industry firms Other entities Universities Knowledge endowment Research strategy Knowledge endowment Number of researchers Value chain position Collaboration memory Organisational figures Success score Knowledge endowment RTOs Number of researchers Relationships Internal research Cooperative research Knowledge spillover Environment International industry firms Internat. research organsations Policy institutions 6

7 Output Research processes Input Model processes at the agent level Knowledge endowment Research strategy Organizational figures Value chain position Success score Collaboration memory p sp 1- p sp Spillover k ji found No Cooperative research a co sr co k ji found a int sr int Internal research Yes Yes No Fitness function Knowledge gain p inno No research result p proc Process Patent Innovation p prod Product 7

8 Output Research processes Input Model processes at the agent level Knowledge endowment Research strategy Organizational figures Value chain position Success score Collaboration memory p sp 1- p sp Spillover k ji found No Cooperative research a co sr co k ji found a int sr int Internal research Yes Yes No Fitness function Knowledge gain p inno No research result p proc Process Patent Innovation p prod Product 8

9 Output Research processes Input Model processes at the agent level Knowledge endowment Research strategy Organizational figures Value chain position Success score Collaboration memory p sp 1- p sp Spillover k ji found No Cooperative research a co sr co k ji found a int sr int Internal research Yes Yes No Fitness function Knowledge gain p inno No research result p proc Process Patent Innovation p prod Product 9

10 Patent output filter: Fitness function The agent s patenting probability is determined subject to its empirical patenting propensity that is evaluated through a fitness function p i = p 0 f (E β 1, L β 2, I β 3, A β 4 ) The coefficients β 1, β 2, β 3 and β 4 are estimated by means of a zeroinflated negative binomial model for a sample of patenting and nonpatenting Austrian MESC firms, where p 0 system paramter R i R&D expenditures of agent a i β 1 = L i number of employees of agent a i β 2 = I i assets of agent a i β 3 = A i age of agent a i β 4 =

11 Empirical Basis Patenting activity, industry structures and innovation processes in the Austrian MESC industry (Paier et al. 2016) Agents (decision-making agents in the model) 138 Austrian firms (with EPO patents (in semiconductor devices, H01L) + relevant firms from the database of the industry association) Other entities (static, empirically initialized agents in the model) 50 international firms, 39 Austrian and foreign universities and RTOs Data sources PATSTAT ( ), Orbis, FirmenABC, Statistik Austria, CIS, Eurostat Technology space the 73 most frequent technology classes (3-digit-IPC) in the set of identified patents Distance measure: Jaccard index of co-occurrences (Rip and Courtial 1984) 11

12 Overview of policy scenarios Increased networking of business R&D with international firms international universities and RTOs national firms national universities and RTOs Upgrade of national public R&D Increased collaboration with national universities and RTOs ~ plus increased supply of specific technological know-how ~ plus increased demand for specific technological know-how ~ plus increase of both supply and demand Improved framework conditions for the national industrial ecosystem Increase R&D performance of small agents Support for lead firms 12

13 Scenarios 1a,b: International networking Scenario 1a Increased collaboration with international firms Scenario 1b Increased collaboration with international public research - Collaboration with + - Collaboration with

14 Scenarios 1c,d: National networking Scenario 1c Increased collaboration with national firms Scenario 1d Increased collaboration with national public research + Collaboration with - + Collaboration with

15 Patent output in scenarios 1-a,b,c,d 15

16 Scenarios 2: Upgrade of national public R&D Networking Scenario 2a + Collaboration with

17 Scenarios 2: Upgrade of national public R&D Networking Scenario 2b + Supply Technology profile AT EDU/RTO + Collaboration with

18 Scenarios 2: Upgrade of national public R&D Networking Scenario 2c Collaboration with Demand Research target AT IND H01. Subfield Expertise 19

19 Scenarios 2: Upgrade of national public R&D Networking Scenario 2d + Supply Technology profile AT EDU/RTO Collaboration with Demand Research target AT IND H01. Subfield Expertise 20

20 No Scenario 2a Increase of demand Yes Scenario 2c Technological profiles in supplied technologies (2-a,b,c,d) Number of patents total: 354 H01: 53 F H L M Q R No Scenario 2a Increase of supply Yes Scenario 2b 21

21 No Scenario 2a Increase of demand Yes Scenario 2c Technological profiles in supplied technologies (2-a,b,c,d) Number of patents total: 354 H01: 53 Number of patents total: 351 H01: 55 F H L M Q R F H L M Q R No Scenario 2a Increase of supply Yes Scenario 2b 22

22 No Scenario 2a Increase of demand Yes Scenario 2c Technological profiles in supplied technologies (2-a,b,c,d) Number of patents total: 340 H01: 64 F H L M Q R Number of patents total: 354 H01: 53 Number of patents total: 351 H01: 55 F H L M Q R F H L M Q R No Scenario 2a Increase of supply Yes Scenario 2b 23

23 No Scenario 2a Increase of demand Yes Scenario 2c Technological profiles in supplied technologies (2-a,b,c,d) Number of patents total: 340 H01: 64 Number of patents total: 365 H01: 77 F H L M Q R F H L M Q R Number of patents total: 354 H01: 53 Number of patents total: 351 H01: 55 F H L M Q R F H L M Q R No Scenario 2a Increase of supply Yes Scenario 2b 24

24 No Scenario 2a Increase of demand Yes Scenario 2c Technological profiles in supplied technologies (2-a,b,c,d) Number of patents total: 340 H01: 64 Number of patents total: 365 H01: 77 F H L M Q R +21% F H L M Q R +45% Number of patents total: 354 H01: 53 +4% Number of patents total: 351 H01: 55 F H L M Q R F H L M Q R No Scenario 2a Increase of supply Yes Scenario 2b 25

25 Scenarios 3: Improved framework conditions Scenario 3a Improved performance of small agents Scenario 3b Support for lead firms (22%) 26

26 Scenario 3a: Improved performance of small firms # Patents # Innovations Scenario All firms Small firms All firms Small firms Reference 351 (100%) 4 (1%) 1392 (100%) 262 (19%) Scenario 3a 355 (100%) 11 (3%) 1752 (100%) 622 (36%) Δ +1% +175% +26% +137% 27

27 Scenario 3b: Support for lead firms 28

28 Scenario 3b: Support for lead firms Scenario # Patents All firms Lead firms Other firms Reference 358 (100%) 218 (61%) 140 (39%) Scenario 3b 383 (100%) 235 (61%) 148 (39%) Δ +7% +8% +6% 29

29 Summary and concluding remarks Scenario analyses refer to Austrian microelectronics/semiconductor industry, current RTI policy making in Austria Support the high impact potential of international R&D collaboration, particularly with public research organizations National upgrade of technological competence required Networking with existing R&D organizations alone not sufficient Multiplier effects with lead firm but not with small firm support Further advancement of the approach Improved representation of knowledge and performance (publications, collaboration networks etc.) and of budgetary aspects Geography of spillovers (spatial extension of the model) Integration of the modelling approach in a ex-ante impact assessment 30

30 References Caldera, P., C. Hartmann, et al. (2015). Electronic Based Systems in Österreich Zahlen, Daten und Fakten. Endbericht. Studie im Auftrag der FFG. Graz/Wien, Austria, Joanneum Research GmbH. Fagiolo, G., P. Windrum, et al. (2007). "A Critical Guide to Empirical Validation of Agent-Based Economics Models: Methodologies, Procedures, and Open Problems." Computational Economics 30: Geyer, A. (2015). Optionsanalyse Silicon Austria Cluster. Studie im Auftrag des Bundesministeriums für Verkehr, Innovation und Technologie, Abteilung III/I5 Informations-und industrielle Technologien, Raumfahrt. Wien, Technopolis. Gilbert, N. (2008). Agent-Based Models. Los Angeles, London, New Delhi, Singapore, SAGE Publications. Hassan, S., L. Antunes, et al. (2008). Stepping on Earth: A Roadmap for Data-driven Agent-Based Modelling. Surrey Scholarship Online. Guildford, Surrey, The Centre for Research in Social Simulation University of Surrey. Macal, C. M. and M. J. North (2010). "Tutorial on agent-based modelling and simulation." Journal of Simulation 4: Paier, M., M. Dünser, et al. (2017). Knowledge Creation and Research Policy in Science-Based Industries: An Empirical Agent-Based Model. Innovation Networks for Regional Development. Concepts, Case studies, and Agent- Based Models. B. Vermeulen and M. Paier, Springer International Publishing Switzerland: Paier, M., M. Dünser, et al. (2016). Innovationsprozesse in der österreichischen Mikroelektronikindustrie. Quantitative und qualitative Fundierung eines akteursbasierten Simulationsmodells. AIT-IS Report Vol Wien, Projekt gefördert durch das Bundesministerium für Verkehr, Innovation und Technologie. AIT Austrian Institute of Technology GmbH. Schneider, H. W., R. Dorfmayr, et al. (2016). Internationale Leitbetriebe in Österreich 2015/2016. Studie im Auftrag der Industriellenvereinigung (IV). Wien, Industriewissenschaftliches Institut (IWI): 38. Smajgl, A. and O. Barreteau (2014). Designing Empirical Agent-Based Models: An Issue of Matching Data, Technical Requirements and Stakeholders Expectations. Empirical Agent-Based Modelling - Challenges and Solutions. Volume 1: The Characterisation and Parameterisation of Empirical Agent-Based Models. A. Smajgl and O. Barreteau. New York Heidelberg Dordrecht London, Springer. 1:

31 Thank you for your attention! MANFRED PAIER Scientist Center for Innovation Systems & Policy AIT Austrian Institute of Technology GmbH Donau-City-Straße Vienna Austria T +43(0) F +43(0) manfred.paier@ait.ac.at

32 Model details

33 Agent characteristics Name Type Updating Initialization Nationality AT, INT fixed empirical user input Value chain position Organizational figures supplier, microsystems, semiconductor, customer fixed empirical user input Assets R + fixed empirical user input No. of employees N fixed empirical user input R&D expenditures R + fixed empirical user input Age N periodicity 12ω empirical user input Knowledge endowment K i = T 1 (τ 1 ) S 1 E 1, T 2 (τ 2 ) S 2 E 2,, T J (τ J ) S J E J periodicity ω, 3ω, 6ω empirical user input, random Research strategy inactive, conservative, incremental, radical periodicity ω user input; endogenized Success score Z periodicity ω 0 Collaboration memory M i = a 1,, a 10 periodicity ω 0 Knowledge acquisition strategy spillover, internal research, cooperative research periodicity ω random according to user input probabilities 34

34 Strategy options and rules Inactive No research, no new target chosen Incrementally innovative Diversification to related technology Conservative Research focus on current core technology Radically innovative Diversification to new technology class General guideline Retain strategy in case of successful research Change strategy in case of repeated failure Exception: Radically innovative - change strategy to Conservative if successful 35

35 R&D: search in technology space G06K H01L H05B 36

36 Position in the value chain Suppliers Materials Manufacturing Tools EDA Tools Semiconductors IDM Fabless Design House Foundry Microsystems Components Software System integration Customers Communication Lighting, Instruments Transportation, Automotive Consumer Electronics Production 37

37 Collaboration probabilities the value chain AT IND Su Se Mi Cu Collaboration with IND-IND collaboration triggered by transition matrix Empirically based probabilities (26.1 NACE) AT INT Accounts for vertically integrated R&D collaboration, horizontal competition IND EDU RTO IND EDU RTO Su Se Mi Cu Su Se Mi Cu 38

38 Modelling steps Input data Experimental design Simulation model Model output Simulation output analysis Uncertainty analysis Sensitivity analysis Data analysis Calibration Validation Empirical calibration and validation 39

39 Number of employees Agents by size Representing a total 138 Austrian and 50 foreign MESC firms > Foreign firms Austrian firms Number of firms 40

40 Agents by value chain position Representing138 firms with a total of 64,262 employees Materials Sup1 Manufacturing Tools Sup2 Integrated Device Manufacturers Sem1 Microsystem Components Mic1 System integration, Packaging Mic3 Communication Cus1 Transportation, Automotive Cus2 Production Cus3 Lighting, Instruments Cus4 Consumer Electronics Cus5 Number of employees Number of firms 41

41 Technology portfolio of the Austrian MESC industry IPC-3-digit classes (73 of 366) of the selected semiconductor patents (2,106) 22 F21V Lighting devices 39 G06F Digital data processing 40 G06K Recognition of data; record carriers 50 H01L Semiconductor devices 59 H02M Conversion between AC and DC 66 H04L Transmission of digital information 72 H05B Electric heating; lighting 42

42 Empirical Calibration Five Parameters are empirically calibrated Share of cooperative agents, probability for spillover, success rate of own research, success rate of collaborative research, search horizon in technology space Search for the parameter combination, which starting from the empirical initialization of the model population ( ) resembles best the empirical patent output ( ) Calibration using a cost function approach Total number of patents Pearson-Correlation between simulated and empirical patent profile 43

43 Model implementation and application Model implementation in Java Model application to policy scenarios using R Simulation and analysis Empirical calibration and Validation ( Reference scenario) Simulation runs over 120 steps (10 years) Output quantities: averages over 100 runs 44