A Survey of Neo-Schumpeterian Simulation Models:

1

• A Survey of Neo-Schumpeterian Simulation Models
• Review and Prospects
• Paul Windrum
• presented at DIME-ETIC
• The Economy As A Complex Evolving System,
• UNU-MERIT, Maastricht, 15 - 19 October 2007

2

• References
• Empirical Validation of Agent-Based Models,
  Special Issue of Computational Economics,
  Computational Economics, 2007 Vol. 30 (3), Chris
  Birchenhall, Giorgio Fagiolo and Paul Windrum
  (editors).
• Windrum, P., 2007, Neo-Schumpeterian simulation
  models, in The Edward Elgar Companion to
  Neo-Schumpeterian Economics, H. Hanusch and A.
  Pyka (eds.), Cheltenham Edward Elgar.
• Windrum, P., Fagiolo, G., and Moneta, A., 2007,
  Empirical validation of agent-based models
  alternatives and prospects, Journal of
  Artificial Societies and Social Simulation, 10(2)
  8, lthttp//jasss.soc.surrey.ac.uk/10/2/8.htmlgt.
• Windrum, P., 1999, Simulation models of
  technological innovation a review, American
  Behavioral Scientist, 42 (10), pp.1531-1550.
  ISSN 0002-7642.

3

• Neo-Classical (Type 1) and Schumpeterian (Type 2)
  Models
• Different views about the world in which real
  economic agents operate.
• Type 1 world can, in principle, be known and
  understood.
• In the Type 2 world the set is unknown, and
  agents must engage in an open-ended search for
  new objects.
• Primary interest of models differ
• Type 1 models learning that leads to
  improvements in allocative efficiency.
• Type 2 models open-ended search of dynamically
  changing environments. Due to (i) ongoing
  introduction of novelty generation of new
  patterns of behaviour (Knightian uncertainty),
  and (ii) complexity of interactions between
  heterogeneous agents.

4

• Equilibrium versus non-equilibrium
• Type 1 models view the underlying structure of
  the economic system as an equilibrium structure.
  
• In Type 2 models, aggregate regularities are not
  equilibrium properties but emergent properties
  that arise from an evolutionary process process
  in which variety generation and selection interact

5

• A bottom-up perspective
• Heterogeneity
• The evolving complex system (ECS) approach
• Non-reductionism
• Non-linearity
• Direct (endogenous) interactions
• Bounded rationality
• The nature of learning
• True dynamics (irreversibility)
• Endogenous and persistent novelty
• Selection-based market mechanisms

6

• Early Neo-Schumpeterian (Type 2) Models
• Stylised Man of early models (Nelson Winter,
  Dosi et al., Silverberg-Verspagen).
• Focus develop models containing key evolutionary
  mechanisms that can generate stylised facts
  observed at industry/macro level.
• Key algorithms variety generation (search) and
  selection algorithms
• Key elements heterogeneity of agents, stochastic
  processes (notably of innovation), interaction
  between agents, feedbacks between decision making
  and emergent properties (path dependency),
  absence of perfect foresight and learning as a
  process of open-ended search (constrained
  rationality/ myopia).

7

• Success of Early Models
• Agenda setting a viable, neo-Schumpeterian
  alternative.
• Method Nelson Winter 2-step method to approach
  to empirical validation of simulation models
• Demonstration demonstration of the feasibility
  of the new approach using simulation.
• Models generated outputs accorded with
  empirically observed phenomena and evolutionary
  neo-Schumpeterian explanations for these
  phenomena.
• Innovation open-ended search as basis for
  learning in worlds with Knightian uncertainty.

8

• Limitations of Early Models
• Generality of the studies
• A very limited range of agents were considered
  and the agent representations were highly
  stylised
• Reports conducted on very few simulation runs
  (illustrations from handful of individual runs)
• High dimensionality of the models (random walk?)
• Lack of sensitivity analysis on key variables and
  parameters (whats really driving the model)
• Lack of rigorous testing procedures for model
  outputs.
• No comparison of alternative theories or models.

9

• More Recent Neo-Schumpeterian Models
• Motivations
• To address the limitation of the early models
• Exploit new algorithms, procedures developed in
  computer sciences, statistics etc., and make use
  of improved software / hardware.
• Will consider 2 examples
• Malerba-Nelson-Orsenigo-Winter, and
• Windrum-Birchenhall models.

10

• History friendly modelling Methodology
• suggests tying down simulation models to
  carefully specified, empirical histories of
  individual industries.
• Detailed empirical data to inform the simulation
  work
• Act as a guide when specifying the
  representations of agents (their behaviour,
  decision rules, and interactions), and the
  environment in which they operate.
• Assist in identification of particular parameters
  on key variables likely to generate the observed
  history.
• Enable more demanding tests on model outputs
  to be specified - evaluate a model by comparing
  its output (simulated trace history) with the
  actual history of an industry.

11

• Issues Regarding the History friendly Method
• Category 1 implementation issues.
• Modelling a special case Malerba et al. (1999,
  2001) is not informed by a history of the
  computer industry as whole, but of one particular
  company IBM. The research questions addressed
  are not relevant to others in the industry.
• IBM account is itself highly stylised and
  subjective
• Lack of data on industry as a whole empirical
  data regarding RD spend, market shares, and
  profitability of computer firms. No empirical
  data on key variables relative sizes of network
  externalities and branding in the mainframe and
  PC markets.
• Question is the empirical data required by the
  method readily obtained in practice?

12

• Other factors influencing the modelling choices
• modelling is informed by theory as well as
  available empirical data
• Empirical data is itself informed by theory
• Do not present a rigorous sensitivity analysis of
  the initial seedings or the random parameter
  values used in the 50 simulation runs reported.

13
Attractor state
Cheapness
PC
Mainframe
Quality
14

• Compare with
• Windrum, P. and Birchenhall, C., 2005,
  Structural change in the presence of network
  externalities a co-evolutionary model of
  technological successions, Journal of
  Evolutionary Economics, 15(2), pp.123-148.
• This paper opens up
• quality is unpacked a complex
  multi-dimensional variable its own right
• relationship between heterogeneous consumer
  demand and different sets of characteristics
  offered by old new technology products
• some performance characteristics of old
  technology are better than those of new
  technology
• some performance characteristics of new
  technology are better than those of old technology

15

• some performance characteristics of offered by
  the new technology are NOT offered by the old
  technology
• some performance characteristics of offered by
  the old technology are NOT offered by the new
  technology
• Now can address the question of co-evolution new
  consumer groups with different preferences
  demanding new technology products with new
  characteristics (a succession).
• Can distinguish between successions and
  substitutions
• Rigorous sensitivity analysis is conducted on
  1000 runs using different parameter values of
  variables and different seedings.
• Then a statistical model is used to test which
  variables affect the probability of a succession
  occurring (here to predict the probability of a
  succession occurring).

16

• Results successions occur if
• direct utility of new tech product gt old tech
  product
• (product characteristics)
• indirect utility of new tech product gt old tech
  product
• (price production economies efficiency of
  production techniques)
• Important initial new design(s) needs to be
  highly competitive have a supporting new
  customer group(s)
• rate of innovative improvements of new tech firms
  gt rate of innovative improvements of new tech
  firms (sail ship effect is possible).
• Complex interplay between quality, price and
  cost is NOT open to investigation in Malerba et
  al model.

17

• Results
• Time
• Time old technology firms have to improve
  performance, price and cost of their designs
  prior to new technology firms arriving in the
  market (entry)
• Time the new firms have to improve to undertake
  innovation and improve their product process
  performance (degree of competition in the market
  strength of replicator OR degree to which new
  consumer preference are distinguished from old
  consumer preferences).

18

• Issues comparing the models
• Both Malerba et al and Windrum-Birchenhall
  consider sequential competitions, and the
  conditions under which old technology and
  established firms may be replaced by new
  technologies and new firms.
• But the models are very different in terms of
• what they are trying to explain Malerba et al
  want to understand the conditions under which
  established firms can survive by switching
  production from old to new technology
• the elements used in each model
• the empirical data they draw upon when building
  their models (input)
• the empirical data that is selected as the
  stylised facts that each model is expected to
  reproduce

19

• Category 2 Methodological Issues.
• Can history to be the final arbiter in
  theoretical and modelling debates? (as suggested
  by Malerba et al, and Brenner)
• E.H Carr (1961) History itself is neither simple
  nor uncontested
• Records that exist are fortuitously bequeathed.
• In-built biases Yin (1994) verbal reports are
  subject to problems of bias, poor recall and
  inaccurate articulation
• Missing data
• Contestability of events, current and past.
• Process of writing academic history is
  open-ended process in which many pieces of
  data, bequeathed from the past, are filtered by
  the historian. Some data accorded status of
  facts by the community of historians. But this
  status is open to review.

20

• Upshot
• Need to develop high quality accounts, open to
  critical scrutiny.
• On the basis of these accounts, guidance is taken
  on particular modelling choices, on parameter
  testing, and output evaluation.
• In recognising the limitations of any historical
  account, we simultaneously recognise the
  limitations of decisions based on that account.
• Applies to all historical / empirical
  approaches to modelling

21

• Further Issues.
• goal of modelling what is the advantage of
  having a very accurate description of one case?
  (Silverbergs example perfect description of the
  fall of an individual leaf from a tree, or
  Brownian motion equations)
• unconditional objects and alternative model
  testing (Brock, 1999)
• alternative methods of sensitivity analysis
• counterfactuals (Cowan Foray, 2002)
• ergodicity (what if the system is non-ergodic?)

22

• Further Issues cont.
• structural change - the relationship of
  statistical data to evolutionary models, timing
  effects and lag structures in simulation models
• calibration alternative ways to calibrate
  initial conditions and parameters
• Indirect calibration
• First validate, then indirectly calibrate the
  model by focusing on the parameters that are
  consistent with output validation.

23

• Werker-Brenner approach
• Step 1 use existing empirical knowledge to
  calibrate initial conditions and the ranges of
  model parameters Step 2 empirically validate the
  outputs for each of the model specifications
  derived from Step 1. This reduces the plausible
  set of models still further.
• Step 3 further round of calibration on surviving
  set of models and, where helpful, recourse to
  expert testimony from historians (so-called
  methodological abduction).

24

• Conclusions and Forward Look
• Rapid development consistent reappraisal of the
  boundaries of research, both with respect to the
  range of phenomena studied and model content.
• Development of distinctive features which set
  neo-Schumpeterian models apart from other models,
  and which gives them a collective coherence.
• A distinctive view about the type of world in
  which real economic agents operate.
• An identifiable set of algorithms that make up a
  neo-Schumpeterian simulation model a search
  algorithm, a selection algorithm, and a
  population of objects in which variation is
  expressed and on which selection operates.

25

• Limitations of the early models are starting to
  be addressed in various ways.
• Malerba et al. have put forward a new methodology
  and a new model structure.
• BUT key issues relating to use of data -
  applicable to all historical / empirically based
  modelling
• unconditional objects and alternative model
  testing
• alternative methods of sensitivity analysis
• counterfactuals
• ergodicity and structural change - the
  relationship of statistical data to evolutionary
  models, timing effects and lag structures in
  simulation models, and calibration.