papers on agent-based economics
nr 3
Behavioral foundation and agent-based
simulation of regional innovation dynamics
Frank Beckenbach
Ramón Briegel
Maria Daskalakis
section environmental and innovation economics
university of kassel
papers on agent-based economics
nr 3
Abstract:
Starting from the observation that there exists a broad variety for the level as well as for the
connectedness of innovation activities in a regional context, we try to figure out a behavioral approach
for explaining such a variety. This approach is composed of (i)conceptual and theoretical reflections
about the micro-foundation of innovation activities, (ii)an agent-based simulation model for these
activities and (iii)empirical regional survey studies as a measuring rod for such an approach. Such a
composition can fill the explanatory gap between external conditions for regional innovation activities
on one side and the observable innovation outcome by specifying plausible internal conditions for
novelty creating activities on the other side. Furthermore the emergence of regional innovation
networks can be explained.
Keywords: Regional Innovation Networks,
Economics, Evolutionary Economics.
Multi-Agent-System,
Behavioral
.
2
papers on agent-based economics
nr 3
I. Introduction
For explaining the different levels and the different dimensions as well as the degrees of
connectedness of regional innovation activities (presumably making some regions prosper and
others not) it is not sufficient to refer simply to different institutional, social or cultural
boundary conditions given in these regions. Because the innovation activities and their
connections are generated by (groups of) individuals it is individual action which transmits
these boundary conditions into innovation activity (or not). Furthermore there are cognitive
processes internal to the individuals as an additional explanatory factor to take into
consideration for the problem at stake. Explaining the regional innovation activities in such an
agent-based manner then necessitates answering the following questions:
•
What are the trigger mechanisms inducing a firm agent to figure out an innovation project
and to leave thereby a routinized mode of executing a well experienced sequence of
actions?
•
If an innovation shall be implemented: What are the determining factors and processes for
pursuing an imitation, an individual innovation, or a collaborative innovation?
For answering these questions we refer to different conceptual building blocks in the second
part of the paper. (i)We pick up the insights of modern cognitive psychology as regards to
different modes of activity (i.e. peculiar the combination of perception, evaluation and
selection of possibilities to act). Corresponding to that we refer to the observable tendency of
the human brain to economize on cognitive effort and as a consequence to a situationdependent switching between different modes of action. Among all action modes that meet
the demands of current a situation, the one that requires the lowest cognitive effort is
activated. Hence, a generic explanation for the temporary nature of novelty creation on the
individual level is possible. (ii)The switching between different modes of action is seen as
being heavily influenced by the relationship between a fast moving actual goal attainment of
firm agents and a slowly moving aspiration level. Hence in this respect we refer to the
concept of satisficing and the related notion of aspiration level already well adopted in
economic contexts. (iii)Finally we integrate modern insights generally about the innovative
personality and especially about the entrepreneur by taking into account personal traits
(curiosity, risk attitude etc.) for explaining the type as well as the frequency of innovative
activities on the individual level.
3
papers on agent-based economics
nr 3
These different conceptual building blocks are synthesized in a frame work inspired by
Ajzen's conceptualization of 'planned behavior'. According to this frame work attitudes,
norms and endowments are the main factors influencing the behavior of agents. These states
of the agents in turn influence the selection of modes of action and hence the probability and
type of novel creating activities.
Based on these conceptual reflections a simulation model is set up in the third part of the
paper. Agents can operate in different modes of action: routine, imitation, individual
innovation, and collaborative innovation. Depending on their economic success and on their
internal traits the agents select one of these modes of action for some predefined time steps.
The knowledge endowment of the agents is defined in terms of generalized (explicit and
implicit knowledge), sharable (explicit) and non-sharable (implicit) knowledge. In case of
being in an (individual or collective) innovation mode the demand dynamics an agent will
face is influenced by the amount of this knowledge. According to the composition of their
knowledge, the agents are able and willing to cooperate with other agents. The core activity of
the collaborative innovations is therefore a knowledge transfer being controlled by a process
of building up or dissolving trust.
The fourth and fifth part of the paper will comprise a presentation and analysis of
simulation runs with the model developed in the third part. Here we pick up the data from a
quantitative and qualitative survey we did in the region of Northern Hesse in Germany. The
data will be used to calibrate the model (parameters, starting variables). Furthermore the data
will be used for figuring out a typology of agents. What we are looking for here are typical
combinations of personal attitudes, the willingness to create a novelty and the way this
willingness is implemented in the firm and finally the (probability) of success.
4
papers on agent-based economics
nr 3
II. Behavioral foundations for regional innovation activities
II.1 Micro-foundation of innovative activities?
The search for a micro-foundation of innovative activities usually is confronted with a
methodological caveat according to which the novelty creating process is totally conjectural
without anything to generalize. Due to the idiosyncratic nature of the processes as well as of
the persons involved in innovations there is seen only a limited possibility for some after-thefact analysis on an aggregated level.
Contrary to that pessimistic view we will specify the triggering conditions and the novelty
generating procedures on the level of single agents in a stylized manner. As regards to these
triggering conditions there is a lack of behavioral synthesis in evolutionary economics
originating in the unrelated contributions of Scitovsky and Simon (cf. e.g. Witt 2006, pp15-6).
Furthermore there is no integration in a broader behavioral concept including other forms of
behavior (choice and routines) being the background for a switch to innovative activities.
There exist a lot of conceptual ideas about a behavioral foundation of economic activities in
the literature. 1 Most of them are not related to novelty creation or not even oriented towards
including different modes of activities. Hence in this literature, empirical evidence, if given at
all, is only related to parts of a behavioral frame work needed here. Therefore it is necessary
to include psychological evidence as a criterion for selecting conceptual ideas. Due to this
situation we try to combine the approaches of Ajzen and the Carnegie School (Simon, March)
for elaborating a behavioral synthesis.
II.2 The Ajzen approach
The behavioral approach of Ajzen ("theory of planned behavior") has been specified and
empirically tested for different domains (consumption, leisure etc.)(cf. Ajzen 1991). This
includes the domain of economics and especially innovative behavior of firms in a market
context (cf. Corral 2002; Petts et al. 1998). Due to this domain-dependent flexibility and its
empirical usefulness we choose this approach as a frame work for our behavioral synthesis.
The focus of the approach of Ajzen is to explain intentional activities, i.e. activities resulting
from a conscious plan to do something. According to this approach, this plan (intention) is
1 Most prominent in this respect are revisions of the expected utility theory (e.g. prospect theory; cf.
Kahneman/Tversky 1979) and enhancements of game theory (cf. Gintis 2003)
5
papers on agent-based economics
nr 3
influenced by three different cognitive factors (cf. Ajzen 1991, pp 181): (i) the attitude of the
agent towards the attributes of the planned activity itself, (ii) the appropriateness of this
activity for the social norms the agent is pursuing and finally (iii) the agent’s ability to control
the intended activity. This implies that the agent can anticipate the salient attributes of the
activity under consideration, that he is well aware of norms he wants to follow (as well as the
persons or groups representing these norms) and that he has an idea about his ability to
control the intended activity. All these circumstances are valued by the agent 2 and – in the
case of multiple arguments – aggregated for each factor. Hence, for each of the factors (i)-(iii)
there exists a subjective weight, influencing the overall intensity to pursue an intended
activity.
From an economic perspective this approach can be assessed in a twofold manner: On one
side it contains the essential features of an economic approach in that it combines goals
(attitudes) and constraints/endowments (abilities to control) for explaining activities. On the
other side this frame work is broader than the economic standard approach to decisions in that
it includes a subjective perception of social embeddedness (in terms of reference groups,
norms etc.) as a core element for intentional activities. This approach is also different from a
game theoretic treatment of interaction patterns in that it is not bound to any type of common
knowledge assumptions and well defined strategies.
In the context of explaining the regional innovation dynamics, the first shortcoming of this
approach is the exclusion of dynamics. Even if Ajzen principally concedes that the experience
of past behavior influences the initial conditions (in terms of subjective attitudes, subjective
norms and subjective control beliefs) for explaining present behavior (cf. Ajzen 1991, 203) it
is not clear at all how this influence takes place. How is the present perception of salient
attributes of reference persons and of control attributes of planned action influenced by past
experience? How is the evaluation system necessary for assessing all these factors accessible
for change? Given such an influence of past on present behavior, it can have different
implications: In the case of successful activities it can trigger an increasing importance of
automatic cognitive processing (instead of intentional deliberation) and a reluctance to change
the pattern of activity due to path dependency and corresponding lock-in effects. Contrary to
that, some learning processes can be expected in the case of unsuccessful activities. The
2 Obviously it is supposed that the agent is endowed with such an evaluation capability for all the
cognitive factors. In the work of Ajzen there is no specification of the properties of this individual
evaluation capability (Is it complete? Is it stable? Is it consistent?).
6
papers on agent-based economics
nr 3
second shortcoming of the Ajzen approach in the given context is the neglect of other modes
of action. In focusing the deliberative procedures leading to an intention, this approach deals
exclusively with (a component of) decision processes. Neither circumventing intentions by
automatic processes 3 nor the search for new options of activity are part of this frame work.
Nevertheless attitudes, norms and control still seem to play a role when these other modes of
actions are selected. So it appears to be worthwhile to integrate these modes of actions in the
given frame work.
Obviously there is only a narrow scope for integrating modes of action based on automaticity
into the Ajzen approach. Ajzen himself (1991, 203) suggested that in the case of a very low
level of subjective control (few elements of control with a low influence on behavior)
intentions do not determine behavior. Rather in this case there is a direct influence of the
strong constraints on the resulting behavior. This kind of automaticity urged by the situationspecific circumstances is different from a ‘learned’ automaticity. The latter takes place if the
situation remains rather similar for a longer time and a specific way to act proves to be rather
successful in this situation (cf. Verplanken/Aarts 1999, pp 104). 4 Then the perception of
situational cues seems to be sufficient for performing the well known activity. Implementing
an intention by consciously selecting attributes of action, reference persons and assessing
control abilities is dispensable in such a case. Rather, what takes place here is an unconscious
activation of schemata or scripts stored in the long term memory. This kind of learned
automaticity (different from urged automaticity mentioned above) is at the core of what
usually is called ‘habit’ or ‘routine’. 5
Even if the cognitive factors of the Ajzen model do not play their usual role of determining
the building and the intensity of intention they are still relevant for explaining the ongoing
activation of habits or routines: the latter are related to attitudes in that they are goal-related,
they take place within the realm of norm-conformity and last but not least they only require
minor control capacity (thus leaving room for using the surplus of these control capacities for
3 There is only a short remark about the possibilities of surpassing intention in the case of limited control
capacities (cf. Ajzen 1991, 182).
4 To act in a manner which conforms to the requirement of the situation can be a result of previously
planned behavior; but it can also be derived from other types of activity as e.g. teaching.
5 ‘Habit’ is the notion used in sociology and is meant to include a wide range of social and cultural
explanantia; ‘routine’ is the notion used in economics and is more focused on the activity under
consideration.
7
papers on agent-based economics
nr 3
other purposes). Hence, it seems appropriate to broaden the concept of Ajzen to include the
mode of habit/routine. 6
II.3 The Carnegie approach
The approach of the Carnegie-school is strongly rooted in an empirical analysis of
organizations and firms showing evidence for their main assumptions (cf. Simon 1997, pp 61;
Lant 1992; Mezias et al. 2002). Firstly, this approach sheds light on the usually neglected
modes of decision making: routines and search. Furthermore the decision units (especially
firms) are not conceptualized as a consistent unit but rather as entities with internal conflicts
being moderated in different ways (cf. Cyert/March 1992, pp 214, pp 229). A common
denominator of these features of the decision unit is the restriction of rationality i.e. of the
availability of perceiving information as well as of the ability to transform this information
into activity. Therefore a specification of the bounded rational way to settle and use goals as
well as capacities by constituting an aspiration level, by following satisficing behavior and by
varying organizational slack is a second reason for referring to this approach.
Originally the aspiration level has been conceptualized in psychological field theory backed
by observations about the context-dependence of the expected result of an activity and about
the role of these expectations for future activities. In its economic, adaption the core idea
behind the aspiration level is to internally fix a level of goal attainment which is related to
past experience and/or to the observable experience of other agents being in a similar situation
(cf. Cyert/March 1992, 162, 172). 7 Generally, the divergence between the aspiration level and
the actual performance level is seen as a source for modifications in behavior. This is due to
an evaluation according to which a negative discrepancy (performance level is lower than
aspiration level) leads to the internal state of dissatisfaction whereas a positive discrepancy
(performance level is higher than aspiration level) leads to an increase of ambition. 8
6 Verplanken/Aarts (1999, 125) characterize habitual modes of action as the “enduring ‘default’ mode of
the mind”.
7 In terms of the concept of Ajzen the aspiration level has to be classified as a norm because it indicates a
social interaction leading to an ‘appropriate’ level of goal attainment.
8 A similar asymmetry in evaluation depending on goal attainment above or below a “reference point” is a
feature of the prospect theory of Kahneman and Tversky.
8
papers on agent-based economics
nr 3
Orienting behavior towards such an aspiration level then defines satisficing instead of
optimizing in goal attainment. Hence, the aspiration level gives a cue for dealing with
bounded rationality: “Actually satisficing is less a decision rule than a search rule. It specifies
the conditions under which search is triggered or stopped, and it directs search to areas of
failure.”(March 1994, 27) A negative discrepancy triggers a different mode of action in terms
of information gathering and risk taking (cf. Cyert/March 1992, 228). 9
The aspiration level as well as the satisficing goal attainment do not take into account the
endowment of agents (in terms of knowledge, finance and time) and their corresponding
capacities to act. Given the (fast) dynamics of the endeavors to meet the aspiration level and
the statics (or slow dynamics) of these endowments there is a varying surplus in the capacity
to act. This surplus is called ‘slack’. If this slack is large, it is seen as an additional source for
novelty generating procedures because then constraints are loose giving room for playful
experimentation. Contrary to the “failure-induced” search, this kind of search is “successinduced” (March 1994, 31; cf. Cyert/March 1992, pp 188; March/Simon 1993, pp 203). 10
This success-induced search can be specified on the level of the individual as well as on the
level of organizations (firms). The corresponding individual trait is curiosity as a search for
new information, new knowledge and new experience for its own sake. Recent research in
this field reveals that curiosity is not simply a genetically programmed drive (activated in a
crude stimulus response context) 11 but rather has cognitive sources either in searching for
congruity and sense making or in practicing idle competences (cf. Loewenstein 1994, pp 80).
Hence, curiosity is in the neighborhood of creativity and intimately related to the above
mentioned phenomenon of slack.
On the level of firms, the slack is a necessary implication of the organizational coordination
failure. Given an organization consisting of a multitude of agents and resources, bounded
rationality as well as opportunism will play a role in the coordination of these organizational
9 Basically the notions of aspiration level as well as satisficing are related to the the individual. In
organizations like firms the level of goal attainment may be group specific and therefore be a source for a
conflict. Managing this conflict is then a boundary condition for the pursuit of the original goal (cf.
March/Simon 1993, pp 132; Simon 1997, pp159).
10 Similar to the approach of Ajzen the endowment of the agent (and the correlating control capacity) is
taken into account here. The difference to this approach is given by the assumption that the amount of
control capacity does not simply determine an intention to act but rather a specific way to act: e.g. the
switch to searching behavior.
11 This notion of curiosity is related to a part of the work of Berlyne and was ‘imported’ into economics
by Scitovski (1976).
9
papers on agent-based economics
nr 3
elements. “When the presence of slack relaxes coordination and control pressures, decision
makers are free to pursue idiosyncratic local preferences.”(cf. March 1994, 31; cf.
Cyert/March 1992, pp 41) This will release a search for novelties on the different levels of the
firm because it opens up possibilities for the different organizational departments to
strengthen their relative position. The harmonizing of these different (potentially) innovative
activities will be a central task for the strategic management.
To resume, integrating in the Ajzen frame work the bounded rational way to realize goal
orientation in terms of aspirations and satisficing on one side and curiosity and creativity
effects of idle capacities in term of slack on the other side, opens up a systematic access to the
search modes of human action. Hence the realm of behavioral explanation of the Ajzen frame
work is enhanced by integrating the elements of the Carnegie school. Furthermore these
elements are dynamic due to the necessity of bounded rational agents to make experience and
to learn from that experience. This establishes a feedback from the performance output of
activities to attitudes, norms and endowments.
II.4 A behavioral synthesis
As it can already be deduced from the discussion in section II.3 it is suggested here to use the
concept of Ajzen as a frame work of the behavioral synthesis at stake. But firstly the factors
used by Ajzen are specified for the domain of a firm in a competitive market economy and for
the given context of explaining the dynamics of regional innovation. Curiosity, risk
orientation and expectation as well as goals are taken into account as personal (or
organizational) attitudes. Aspirations in terms of profit and market share are specified as the
relevant ‘norms’. The control component consists of the usual elements of economic
endowment (knowledge, finance and time). Secondly: these factors are not used for explaining
a specific activity (or intention to do that activity); rather, according to the necessary
behavioral enhancement, they are used for explaining the selection of the mode of action. 12
Integrating aspirations and slack in such a frame work allow for dealing with the behavioral
dichotomy in terms of switching between search and routines. According to the context under
investigation, here the search mode is differentiated in a twofold manner: On one side it is
distinguished between a search for novelties already practiced by others (imitation) and
novelties which are created by the searching agent (innovation). It is assumed here that slack
12 Cf. Svenson 1990; Louis/Sutton 1991 and Jager 2000 for such a muli-mode concept of human action.
10
papers on agent-based economics
nr 3
(curiosity) is only relevant in the case of innovations and that the expected costs of imitations
are lower than the expected costs of innovation. On the other side a distinction is made
between an innovative activity of one agent in isolation and a cooperative manner of
innovation.
These different modes compete for being activated by the agent due to endogenously
generated forces. Activating imitation necessitates overriding the ‘default mode’ of routine.
Practicing innovation requires higher “innovation pressure” 13 to overcome the lower pressure
towards mere imitation. Finally, implementing innovation in a cooperative manner calls for a
correspondence in terms of innovation drive and knowledge endowment.
These behavioral elements for explaining the switching between different modes of action are
combined in a dynamic manner. Every mode of action is linked with an amount of the given
endowment (in terms of money, time and knowledge) and has a specific performance output
(in terms of goal attainment). Hence, there is a feedback at least to the ‘norms’ and
capabilities determining the pursuing of specific modes of action. 14 Fig. 1 shows the
behavioral synthesis in graphic terms.
II.5 Role of the regional context for innovation
The behavioral conceptions discussed so far are not explicitly related to a 'regional' context of
innovation activities. Such a context has at least a twofold meaning for the economic
activities of agents: Firstly, a region is tantamount to a set of common boundary conditions in
terms of mentality, culture and institutions corresponding to a commonality in the experience
of agents belonging to a region. Secondly, these commonalities define a potential for
proximity or – to phrase it the other way round – the ease to overcome spatial, mental and
cultural distances. Taking into account that the generation of novelty is an interaction
implying structural uncertainty, these regional commonalities and proximities can be used not
only for reducing uncertainty but also for accessing new economies of scale and scope. Such
an endeavor is the more successful, the more it not only includes firms but also organizations
as well as public and semi-public institutions (cf. Cooke 1997). In distinguishing the relations
and ties between firms, universities, organizations and institutions generated by such a
13 This ignores the differences between the aspiration-induced innovations and the slack-induced
innovations (cf. March 1994, 32, 34).
14 The feedback to the attitudes and cost expectation is not yet implemented in the following model.
11
papers on agent-based economics
nr 3
promotion of regional innovative activities from the normal state of affairs, one can speak of a
"regional innovation system" (RIS). 15
Attitudes
Personal
traits
Norms
Goals
Aspirations
Profit
Risk
Endowment
Curiosity
Knowledge Finance Time
Marketshare
Slack
Selecting mode of action
Executing mode of action
Search
Routine
Expected resource use
Intention
Space of activity
Used resources
Performance
Fig. 1: Causal chain diagram of the behavioral foundation
The literature about RIS is primarily focused on the boundary conditions (e.g. shared mental
and normative models, infrastructure support), on the patterns (e.g. cluster of firms,
interaction between different groups of agents, sectoral structures, types of RIS, and
properties of the system a whole) and on the outcome (e.g. in terms of growth and
employment)(cf Cooke et al. 2004; Asheim/Coenen 2005). How RIS emerge, develop and
possibly disappear is much less investigated. Corresponding to that there is a neglect of the
behavioral background for the agents constituting a RIS. Hence, there is a gap between
15 Cf Iammarino 2005 for a discussion of the conceptual as well as empirical ambiguities of this notion.
12
papers on agent-based economics
nr 3
analyzing the boundary conditions as well as the patterns of RIS on one side and its outcome
in terms of growth and employment on the other side. 16
At the core of a RIS there is an emerging collaboration and cooperation of agents for the sake
of innovation. The driving force for these cooperations is the comparative advantage expected
by all cooperating agents (in terms of their respective goals). This process is fed by
transferring and transforming knowledge coming from universities as well as by mutual trust
and culture. This driving force for a RIS is constrained by bounded rationality in terms of
ignorance about potential cooperation partners, transaction costs and opportunism which are
themselves strongly determined by the agent's endowment and transfer in terms of knowledge
and institutions. These constrains might be resolved by assistance and governance coming
from regional public and semi-public institutions. Correspondingly, the focus of the following
model as regards to regional aspects of innovation is twofold: Firstly, regions are seen as the
level for cooperative innovations. Because it is observable that the existence as well as the
result of these cooperations are easier to communicate on a regional level it is reasonable to
assume that there is a positive regional frequency dependent feedback of these cooperative
innovations. Secondly, agents located in regional proximity can mobilize similar or
complementary knowledge in a comparatively easier way and by exchanging this knowledge
they can built and rely on trust relations making the matching of appropriate innovation
partners more easy. This increases the probability of success for cooparative innovations. 17
III. Architecture of the simulation model
The following simulation model aims at explaining
•
under what conditions agents in a region pursue a novelty generating activity
•
in which way this novelty creation takes place and
•
what will be the outcome of these activities for a regional market.
For these explananda attitudes, norms and endowments of the agents are used as explanantia.
16 Cf. Beckenbach/Briegel/Daskalakis 2006 for an attempt to fill this gap.
17 The knowledge transfers from universities as well as the governance processes involving organizations
and institutions are not yet dealt with in the model.
13
papers on agent-based economics
nr 3
III.1 Agent population and economic activities
The population in our multi-agent model of a regional innovation system (RIS) consists of
firm agents whereby each of them belongs to one of three types: experimental, cautious and
conservative. These types are characterized by different values for behavioral parameters
which reflect personality traits of the entrepreneur resp. behavioral propensities of decision
makers in the firm. These behavioral parameters and the role they play in entrepreneurial
information processing and decision making are focused in this section.
All firm agents in the model produce goods for the consumer market; they sell conventional
(non-innovative) and innovative products. Firm agents are able to develop innovations
individually or within a cooperation with other firms. As an alternative to innovation, firm
agents are able to imitate innovative products that have already been put on the market by
other firms.
The emergence and dynamics of final demand are not modelled in an agent based manner,
i.e., there are no consumer agents and no explicit market transactions in the model. However,
the final demand is endogenous in that it increases (following a simple diffusion submodel 18 )
each time an innovation is put on the market.
III.2 Agent architecture
III.2.1 Endowment of agents
The firm agents are endowed with the following main features:
•
They have cognitive abilities. These comprise the declarative knowledge base of the firm
in terms of technological possibilities and market information, e.g. on consumers’
wants, 19 as well as the procedural knowledge base in terms of the ability to select and
activate an action and cooperation mode which is adapted to the requirements of the
current situation as well as to the amount of currently unused financial and cognitive
resources (see below).
•
They have two (monetary) goals, namely to gain profit and to achieve a high market
share. The degree of achievement of each goal is measured by a corresponding aspiration
level that is dynamically adjusted according to the actually reached current profit resp.
market share (moving target).
18 For details of this diffusion sub-model cf. Beckenbach/Briegel/Daskalakis 2007, pp18. It has to be
mentioned here that this sub-model can be specified according to regional particularities.
19 The structure of the declarative knowledge base is explained in more detail in section III.3
14
papers on agent-based economics
•
nr 3
As mentioned above, they have some personality traits and behavioral propensities. 20
These parameters are the same within each subpopulation (defined by an agent type), but
differ from agent type to agent type, which makes the whole agent population
heterogeneous.
III.2.2 Action modes
As described in more detail in section IV.1.1, we can distinguish several action modes of
firms which differ with respect to the degree of novelty (the amount and depth of newly
acquired knowledge and skills) and the amount of related cognitive effort. If focusing product
innovations, these action modes are radical innovation (i.e. creation of a product which was
not on the market before), incremental innovation (enhancement of an existing product),
imitation, and all remaining action modes without novelty creation, which we integrate into
one single action mode called routine (“business as usual”). However, in order to avoid overcomplexity of the model, we restrict ourselves to one innovation mode which should rather be
thought of as radical innovation.
Consequently, we have the following three action modes in the model:
•
Routine
•
Imitation
•
Innovation, which splits up into the two sub-modes (cooperation modes) of individual and
cooperative innovation.
Both novelty action creating modes (innovation and imitation) require not only cognitive
effort, but also temporal and financial resources. The time to develop an innovation (resp.
imitation) is set probabilistically (drawn from a uniform random distribution) for each
innovation (resp. imitation) project; the minimal and maximal value of the corresponding
distributions are model parameters.
The temporal and financial cost of an imitation project is smaller than the one of an
innovation project. However, as regards to the possible returns, there is a disadvantage for an
imitator compared to an innovator as an imitator can only participate in the fraction of the
total demand potential of the imitated innovative product which is not yet exhausted (in the
course of diffusion of the product) at the time of accomplishment of the imitation. Generally,
the expectation value for the demand potential assigned to a novelty creating mode is set
20 It is assumed here that these personal traits are given. To analyse their (long run) dynamics is a topic
for future research.
15
papers on agent-based economics
nr 3
proportionally to the amount of declarative knowledge of the firm (more precisely: to the
number of knowledge domains where the firm has got knowledge). 21
In the following section we describe in more detail the mechanism of the selection between
these three action modes.
III.2.3 Selecting the action mode
III.2.3.1 Basic selection principle
According to the modern insights in cognitive psychology as well as in neuroscience, the
individual is not an entity free of contradictions but rather incorporating counteracting forces
the balance of which determines the way the individual is acting. According to the insights of
modern organization theory, the same seems to be true for organizations as multi person
agencies. Here the counteracting forces are a consequence of different groups interacting in an
organizational context.
The basic counteracting forces in the given context are the force (or pressure) for novelty
creation on one side and the ‘natural’ force to preserve the given processes on the other side.
Taking into consideration the essential difference between innovation and imitation as
different types of novelty, the intention to create novelty is assumed to emerge if the
corresponding force is stronger than the mentioned preservation forces, and the type of
novelty (imitation or innovation) is determined by the strongest corresponding force.
These intentions, however, can only be carried out in so far as the necessary financial means
(see above) are available; therefore, first of all the fulfilment of this budget restriction is
checked in order to determine the set of possible action modes.
In the case of imitation, there is another practical restriction which is checked before the very
action mode selection, namely the (non-)existence of an imitable product. A product is
imitable if and only if it has been put on the market by some other firm at an earlier point of
time and if the temporal distance between this point of time and the present is less than a
model parameter.
Formally, if we encode the three action modes by 0 for routine mode (business as usual), 1 for
imitation and 2 for innovation, we can denote the corresponding forces by Fi for i=0, 1, 2, and
the current action mode am is determined such that Fam is maximal among the Fi for which
21 The background for this assumption is the positive relation between the broadness of knowledge and
the firm's flexibility as regards to the demand side.
16
papers on agent-based economics
nr 3
action mode i is possible in view of the budget restriction and the existence of an imitable
product. 22
III.2.3.2 Specification of the selection forces
Picking up the conceptual reflections in section II, the innovation force (F2) itself consists of
different components such as curiosity, the degree of satisfaction as regards to profits and the
degree of satisfaction as regards to market shares. These degrees of satisfaction are indicated
by the relationship of the aspiration level for profits (asp) and for market shares (asm)
respectively to the corresponding actual performance level (p and m):
asp asm
,
⋅
p m
The activation
level of these different components is influenced by personal attitudes such as the exploration
drive (w0) as well as the weight and elasticity of the profit resp. market share aspiration (w1
and ε1 resp. w2 and ε2). Finally the innovation force is modulated by the expected cost for the
innovation endeavour (cin) and the risk acceptance (α) which maps the willingness to accept
the higher risk of innovation compared to imitation.
The aspiration levels are updated at the end of each time step according to the equation
asp (t + 1) = (1 − φ ) asp (t ) + φ p (t )
[1]
and analogously for asm, where φ is the flexibility of adaptation, which is another personal
trait ( 0 ≤ φ ≤ 1 ).
According to the discussion in section II, curiosity is strongly related to the phenomenon of
‘slack’, i.e. the reserve capacities in terms of knowledge and finance. In any given time step
this slack is tantamount to balancing the given state of knowledge and finance on one side and
the amount needed of these resources for a given type of activity on the other side. Again, the
intensity of curiosity triggered by this slack is depending on a personal trait, the exploration
drive (w0).
These considerations and definitions can be formalized as follows: We define three
component forces fi for curiosity (i=0), profit aspiration (i=1) and market share aspiration
(i=2) by
f 0 = w0 (kr + fr )
⎛ asp ⎞
⎟⎟
f1 = w1 ⎜⎜
⎝ p ⎠
[2]
ε1
[3]
22 If there is more than one am satisfying this condition, the smallest is selected.
17
papers on agent-based economics
nr 3
ε2
⎛ asm ⎞
f 2 = w2 ⎜
⎟ .
⎝ m ⎠
[4]
Here kr and fr denote the knowledge resp. financial reserves. The knowledge reserves are
operationalized as the share of the number of sharable knowledge domains (see below) where
the agent possesses knowledge with respect to the total number of sharable knowledge
domains; the financial resources are operationalized as the share of the current profit with
respect to the current turnover. 23 Finally, the innovation force F2 is defined by
F2 = α
f 0 + f1 + f 2
.
cin
[5]
The imitation force (F1) is different from the innovation force in three respects: Firstly,
curiosity (or the personal exploration drive) plays no role in it (f0 is omitted here). Secondly,
there is a comparative difference in expected cost: cim < cin. Thirdly, the risk acceptance α,
which was introduced to map the willingness to accept the higher risk of innovation compared
to imitation (see above), is omitted here. Hence, the imitation force can be formalized as:
F1 =
f1 + f 2
.
cim
[6]
Finally, we set the preservation (or routine) force
F0 = 1
[7]
as a reference value. 24
23 Further explanations for equations [3] and [4] will be given in section IV.2.1.
24 Setting the preservation force to a constant is no restriction of generality since the absolute values of
the forces Fi don’t matter; it is only the ratio between them which determines the action mode. - There are
three special or exceptional cases in which the selection mechanism mentioned above is not applied (or
even not applicable):
(a)
A new firm (start-up), which at the moment of its entry has zero turnover and zero profit, is
assigned the action mode innovation and the cooperation type (if any) curiosity.
(b)
When a new firm just has finished the development of its first innovation, it is assigned the
routine action mode for a short period (which is determined by the mean development duration of an
innovation).
(c)
A firm with negative or zero profit which doesn’t belong to case (b) is assigned the action mode
innovation and – if it is willing to cooperate – the cooperation reason (see the next subsection for a
formal definition of this notion) is assumed to be profit dissatisfaction.
18
papers on agent-based economics
nr 3
A firm agent which has selected the action mode of innovation still has two options (see
above): He can try to develop an innovation on his own (individual innovation) or he can seek
for cooperation in order to enable or facilitate the development of an innovation.
The selection between these two cooperation modes is modelled probabilistically. The
cooperation probability (cp) depends on the propensity to cooperate (χ) and the share of
agents which participated in an innovation development cooperation in the region under
consideration (i.e. Nc/N, where Nc is the number of such agents and N is the total number of
agents) according to the following equation:
cp = (1 − ifb) χ + ifb
Nc
N
[8]
Here ifb is a parameter which determines the intensity of feedback of cooperations.
III.3 Cooperative innovation
As mentioned above, there are two types of innovation cooperations which may also be seen
as two subsequent phases of a cooperation’s lifespan:
•
An innovation development cooperation aims at the development of a marketable
innovative product. This process involves an exchange of (sharable) knowledge (see
below) between the member firms of the cooperation. When and if the development of
such an innovation is successfully finished, the innovation development cooperation is
transformed into
•
a marketing cooperation. This means that the cooperation partners put the product on the
market together and share the production cost and the returns of the sale of this product.
The structure of (declarative) knowledge and the role it plays in selecting the cooperation
partner as well as in the cooperation process itself is described in the following subsections.
According to the empirical literature (cf. Fritsch/Franke 2003; Antonelli 2000; Sternberg
2000) the endowment with knowledge and the transfer of knowledge between firms is the
main feature of cooperations. Hence, we distinguish three kinds of (declarative) knowledge:
•
generalized (explicit and implicit) knowledge, being equally significant for all agents and
branches and constituting a common base for communication between cooperation
partners;
•
sharable (explicit) knowledge, which can be transferred between cooperation partners and
exploited by each of them and finally
19
papers on agent-based economics
•
nr 3
non-sharable (implicit) knowledge, which cannot be transferred between cooperation
partners.
The whole declarative knowledge base of a firm agent – comprising these three kinds of
knowledge – is mapped in the model as vector components corresponding to knowledge
domains in each of which the agent may or may not have knowledge. This means that there is
no cardinal measure for the knowledge in a certain domain; each component of the agent’s
knowledge vector can take only one of the two values “available” (1) and “not available” (0).
For a fixed cooperation-seeking agent, A, the set of its potential cooperation partners (i.e. the
set in which his search for a cooperation partner has to take place) is defined by the logical
conjunction of six conditions for a potential cooperation partner B. The first four of them are:
1. B is willing to cooperate, i.e., has activated the action mode of cooperative innovation (in
each time step, when all agents have selected their action and cooperation mode, all agents
willing to cooperate are signalling this willingness).
2. B belongs to the same agent type as A or is a supplier or a customer firm of A (at the
beginning, each firm was assigned randomly a set of supplier firms, which are fixed for the
whole simulation).
3. If A and B already had successfully cooperated in the past, then A has sufficient trust in B
and vice versa.
4. B and A have sufficient common generalized knowledge; this means that the number of
generalized knowledge domains where A and B have available knowledge equals at least
some model parameter, reqgen.
To explain the fifth condition, we first need to define the reason for which a firm agent is
seeking a cooperation: The cooperation reason is determined by the dominant component
force of the innovation force, i.e., by the index cr for which the corresponding component
force fcr is maximal. Now the fifth condition for the selection of a cooperation partner
requires:
5. B has the same cooperation reason as A.
As the fourth condition, the sixth one is a knowledge condition, this time on sharable
knowledge. However, the content of this condition depends on the cooperation reason of A:
6. If the cooperation reason of A (and B) is profit aspiration (cr=1), a sufficient amount of
common sharable knowledge is required. If the cooperation reason of A (and B) is
curiosity or market share aspiration (cr=0 or 2), a sufficient amount of complementary
sharable knowledge of each agent for the other is required; this means that the number of
20
papers on agent-based economics
nr 3
sharable knowledge domains where B has available knowledge while A has not equals at
least some other model parameter, reqs, and vice versa with A and B interchanged.
Once the initiator of a cooperation has selected his partner(s), a process of knowledge transfer
is started. This process lasts a number of time steps which is set probabilistically at the
beginning time of the cooperation (see above). In each time step, knowledge in one domain is
transferred from the cooperation initiator to his partner(s) and vice versa with a certain
probability 25 . This probability (pb) depends on one hand positively on the trust (tr) of the
knowledge-giving agent in the knowledge-receiving agent and on the other hand positively on
the absorptive capacity (ac) of the receiving agent 26 . Formally, the probability is given by 27
pb = se (tr − 1) + ac ,
[9]
where se denotes the sensitivity of the transferring probability with respect to trust, another
model parameter.
Each time this knowledge transfer doesn’t happen, the trust of the “receiving” agent in the
“giving” one is diminished by a certain decrement; on the contrary, trust is raised by a certain
increment each time the transfer actually happens. If the trust of a member of the
coopeTration falls beyond a certain threshold, this agent is leaving the cooperation. 28 The
decrement and increment of trust as well as the threshold are model parameters. 29
25 The corresponding random number is drawn for each partner and for each transfer direction, so that
these knowledge transfers are stochastically independent
26 The absorptive capacity is conceptualized here simply as a given probability weight for the happening of
the cooperation.
27 If the right hand side of the formula is negative, the probability is set to 0. We assume 0 ≤ tr ≤ 1 and
0 ≤ ac ≤ 1 .
28 This event can have different consequences for the cooperation depending on the size of the cooperation
and the role the leaving agent plays in it: If the leaving agent is not the initiator of the cooperation (i.e., he
has been selected as a cooperation partner by the initiator) and the number of remaining members is greater
than one, there is no further consequence, i.e., the remaining members continue the cooperation process. If,
on the contrary, the leaving agent is the initiator of the cooperation or if there is only one member left, the
cooperation is completely broken off and the joint innovation development project is cancelled.
29 Analogously to this stochastic process for maintaining or leaving a cooperative innovation, there is a
lottery in the course of the process of individual innovation: during the development of an individual
innovation, in each time step, the innovation project is abandoned with a certain probability, which is a
model parameter.
21
papers on agent-based economics
nr 3
IV. Calibration of the simulation model
IV.1 Empirical calibration of model parameters
The following analysis aims at giving - partially - an empirical foundation of the behavioral
parameters developed in section III. Two datasets dealing with (regional) innovation
behaviour of firms are used for calibrating the model. Both data sets were gained in Northern
Hesse, a region located in the middle of Germany. The first dataset (D1) is based on a written
questionnaire which was sent to 1783 firms. This sample consisted of the whole population of
firms with more than three employees belonging to the manufacturing sector and related firms
of the service sector in the respective region. Altogether, 527 firms responded to the survey
(response rate of 29,6%). To gain the second dataset (D2), a random sample of 400 firms was
drawn from the 527 firms which form the dataset D1. 30 Those firms were surveyed a second
time in order to gain a better understanding of the behavioral foundations of the innovation
process (cf. Daskalakis/Krömker 2007). 207 firms responded to the second survey (51,75 %).
IV.1.1 Calibration of initial firm population
In the first questionnaire (D1), the firms were asked about their innovation activities between
February 2003 and February 2006. Altogether, about 80% of the firms were innovating, most
of them conducted product (and service) innovations (88%), the shares for process
innovations and organisational innovations were 63% respectively 46%. In order to represent
the three agent types of the model (left column of Table 1), we referred to the product
innovators and classified three types of firms (right column of Table 1):
Firm class
Agent type
Radical innovators (F_IR) 31
Experimental
Imitators (F_IIM) 32
Cautious
Routinizing firms (F_ROUT)
Conservative
Table 1: Mapping from empirical firm classes to agent types in the model.
30 Cf. Daskalakis/Krömker 2007.
31 Those firms might also have realized incremental innovation as well as imitation.
32 Those firms solely accomplished imitation.
22
papers on agent-based economics
nr 3
The distribution of the type of firms within Dataset D1 is as follows: The share of Type F_IR
in the product innovators was 30% and the share of the firms type F_IIM 15%. About 20% of
the firms conducted no innovation at all (firm type F_Rout). 33
Having built in a first step the firm population consisting of certain shares for different types
of firm agents, behavioral attributes contained in the survey are assigned to these types of
agents in a second step. In the following this calibration of behavioral parameters for the
different types of agents is explained.
IV.1.2 Calibration of behavioral parameters
The behavioral parameters investigated in the following are (i) risk acceptance, (ii)
exploration drive, (iii) market share aspiration, (iv) profit aspiration, (v) propensity to
cooperate and (vi) trust toward the (regional) cooperation partner. The variables which allow
for representing the behavioral parameters are mostly based on five point or 6 point Likert
scales. Correlation analyses (Kendall-Tau-b) as well as non-parametric tests (Mann-WhitneyU-Test) were applied in order to get insight about the behaviour of the different types of
agents (cf. for the respective evaluations appendix 1). 34
IV.1.2.1 Findings about risk acceptance
Two indicators of our datasets can be used to investigate the risk acceptance of the firms. In
the questionnaire which was used to create dataset D1, the firms were asked about the goals
of their innovation activities. One item dealt with the balance between the acceptance of risk
and the level of profits. The question was: "Does your firm rather aim at a high profit and
accepts therefore a high level of risk or does your firm only approve a moderate risk accepting
a lower profit therewith?" In dataset D2, all of the firms (including the not innovating ones)
were directly asked whether risk acceptance is an attribute of their firm. Both datasets give
evidence that firms of the types F_IR have a higher level of risk acceptance than the firms
which are imitating (F_IIM). However, the results are only significant for dataset D1. Fig. 2
shows the allocation of the answers given in dataset D2 for each type of firm.
33 Product innovators who conducted incremental innovations and are not part of F_IR (55% of the
product innovators) are excluded from our analysis. We thus analyze a very specific, yet quite relevant,
section of possible innovation activities. This allows for implementing the empirical findings in the
simulation model.
34 Because of the relatively low number of firm type F_IIM in D2 (n=20) the usual statistical tests can be
applied only partially when dealing with D2. The respective answers of the firms concerning the
behavioural foundation are thus only used to calibrate the model, if either the statistical evidences
concerning those variables are sufficient and/or if statistical evidences from dataset D1 support those
findings.
23
papers on agent-based economics
nr 3
IV.1.2.2 Findings about exploration drive
For this parameter, several indicators are applicable. An item battery in D2, concerned with
behavioral attributes of the firm, consists of two items capturing the exploration drive: the
firms were asked if they considered themselves as willing to experiment as well as whether
they assessed themselves as creative. Both attributes are highly correlated among each
other 35 as well as with the firms type F_IR. There are no significant differences between the
types F_IR and F_IIM but between F_IR and F_ROUT. Figure 3 shows the frequency
distribution as regards to creativity for all types of firms.
Creativity
50%
40%
30%
20%
10%
0%
1=not at all
2
3
4
5
6=very
much
Level of creativity
F_IIM
F_IR
F_ROUT
Fig. 2: Creativity. Number of F_IR = 44; number of F_IIM=20;
number of F_ROUT=26.
The differences between the firms get more obvious by analysing the corresponding questions
from dataset D1. Here again, two questions out of the battery regarding the innovation goals
are available to support the results gained from D2. One question asked whether the firms
want to create new needs, the other was about the willingness to develop new markets. Both
items proved to be significant with respect to the differences between the firms type F_IR and
F_IIM. As regard to the requirements of the simulation model both questions were combined
and the means calculated accordingly.
IV.1.2.3 Findings about market share aspiration and profit aspiration
In D2, the firms were asked about the goals underlying their economic behavior. One item
thereby dealt with the relevance of the goal “increasing the market share” (see Fig. 4). The
35 Kendal´s tau b: 0,436; p < 0,01%.
24
papers on agent-based economics
nr 3
statistics show that there is a significant correlation between this item and the firm type F_IR
as well as clear cut differences between the types F_IR and F_IIM as well as between F_IR
and F_ROUT.
Market share
50%
40%
30%
20%
10%
0%
1=not at all
2
3
4
5
6=very
much
Level of aspiration
F_IIM
F_IR
F_ROUT
Fig.3: Market share. Number of F_IR = 44; number of F_IIM=20;
number of F_ROUT=26.
With regard to the profit aspiration (without figure), however, no significant correlations and
differences can be estimated. As profit aspiration is at the core of every economic activity,
these findings are not surprising.
IV.1.2.4 Findings about propensity to cooperate
To investigate the propensity to cooperate, we only refer to dataset D1. As shown above, the
share of innovating firms was about 88%. Nearly 45% of these firms stated that they were
involved in cooperative innovations. With a cooperation rate of about 70%, the firms of type
F_IR were the most cooperative firm type (cooperation rate type F_IIM: 34,5%). In more
detail: about 40% of the firms type F_IR had accomplished a cooperation during the
investigated time frame, about 47% still had been involved in innovation cooperations (type
F_IIM: 9,1% and 29,1%). 36 Accordingly, the correlation between the items capturing the
cooperation activities and both types of firm have a different (significant) level of significance
and the statistics show a remarkable difference between both types of firms.
36 Most of the cooperations (respective over 90%) were aligned to product innovations.
25
papers on agent-based economics
nr 3
IV.1.2.5 Findings about regional trust
On the average, 37% of the cooperative innovations were conducted with regional partners.
The empirical testing shows that, independent from the type of firms, most of the firms had a
high level of trust towards their regional partners. Accordingly, there are no significant
differences between type of firms: trust seems to be a relevant pre-condition for cooperative
innovations (see Fig. 5). However, there can be assed significant differences in the levels of
trust towards regional and towards national partners: the regional cooperation partners were
trusted more profoundly. 37 Daskalakis/Kauffeld-Monz (2007) have shown that the level of
trust and the amount of knowledge exchange within cooperative innovations are positively
correlated. This is approved in the dataset D2 by a high correlation between the item „we have
given our cooperation partners a lot of assistance“ and the cooperative firms 38 . Thus, it can be
assumed, that, on the average, a higher level of trust towards regional partners allows for a
higher level of knowledge transfer between partners in the same region than between partners
of different regions.
Trust tow ards cooperation partners
70%
60%
50%
40%
30%
20%
10%
0%
1
2
3
4
Northern Hesse
5
1
2
3
4
5
Germany
1
2
3
4
5
Other
Level of trust
F_IIM
F_IR
Fig. 4: Regional trust. Number of F_IR = 56/69/; number of F_IIM=11/69/36.
1=not at all; 6=very much.
IV.1.2.6 Applying the empirical findings to the simulation model
For the calibration of the model, the means of the empirical indicators as tested above are
used. Therefore, firstly, the respective means were evaluated. Secondly, these means were
37 Level of significance (Mann-Whitney Test) for the two highest values: 0,001 (2-tailed).
38 Kendall´s tau: 0,202; p<0,01%.
26
papers on agent-based economics
nr 3
linearly transformed to fit the model scales. Table 3 gives an overview of the results with
respect to the different types of firms.
variable
risk
acceptance
exploration
drive
profit
aspiration
Market share
aspiration
cooperation
propensity
regional trust
parameter
empirically derived means
dataset
transformed means
scale
F_IR
F_IIM
F_ROUT
D1
2,3
2,0
-
1..5
D2
4,1
4,2
3,5
1..6
D1 39
3,3
2,6
-
1..5
D2 40
4,6
4,3
3,7
1..6
w1
D2
5,1
4,7
4,7
w2
D2
4,9
3,9
χ
D1
(70%)
tr0 41
D1
4,0
Α
w0
F_IR
F_II
M
F_ROUT
0,6
0,5
0,4
0,6
0,5
0,4
1..6
3
2,5
2
4,3
1..6
0,15
0,1
0,125
(34,5%)
-
-
0,85
0,75
0,65
3,9
-
1..6
0,75
0,75
0,75
Table 3: Statistical calibration of behavioral model parameters
IV.2 Calibration of other model parameters
The other model parameters are set to theoretically plausible values and/or have been adjusted
to produce results which are compatible with empirical observations. In this section we
explain some of the underlying theoretical considerations.
IV.2.1 Elasticity of profit and market share aspiration
Compared to the component force f0 associated to curiosity (cf. formula [2]), the main
variables constituting the other two component forces exhibit specific characteristics that lead
to the introduction and setting of the corresponding elasticity parameters (ε1 and ε2): Profit
and market share are variables which are crucial for the survival (and even the raison d’être)
of the firm; contrary to curiosity, which is related to abundance of cognitive and financial
resources (slack), fluctuations in these variables constitute a potential threat to the existence
of firms. Therefore it seems appropriate to increase the influence of such fluctuations on the
39 This variable comprises the items „developing new markets“ and „creating new needs“.
40 This variable comprises the items „experimental drive“ and “creativity”.
41 This empirical value is only used to calibrate the initial value of the parameter.
27
papers on agent-based economics
nr 3
component forces f1 and f2 and hence on the selection of the action mode. 42 Hence, it is the
elasticity of the component force with respect to profit which must be increased to strengthen
the influence of fluctuations of profit (and analogously for the market share). This is achieved
by introducing the parameters ε1 and ε2 and setting them to values much greater than 1.
Additionally, market shares normally fluctuate less than profits; to compensate this
difference, we set the elasticity associated to market share (ε2) even to a higher value than the
one associated to profit (ε1). In the standard (reference) parameter constellation, we set ε1=8
and ε2=16.
IV.2.2 Parameters related to trust
As the empirical results show that firms have significantly higher trust in regional
cooperation partners than in partners from outside the region, we set the initial value for trust
in the model to a substantially higher value (.75) than the threshold value for breaking off a
cooperation (.5). Together with the setting of .2 for the trust decrement (in the case of failure
of knowledge transfer in the cooperation process), this makes it possible that a cooperation
survives one failure of knowledge transfer while two such failures – if they are consecutive or
separated by only one successful step of knowledge transfer – lead to breaking off the
cooperation.
The trust increment is set to a smaller value (.1) than the trust decrement (.2) in order to
reflect an asymmetry of valuation in trust dynamics: The process of trust building is slower
than the loss of trust; the latter is rather a catastrophe-like development than a gradual
process.
V. Simulation results
V.1 Results for the reference parameter constellation
In this section we will describe a singular simulation run for the parameter constellation
which we use as a reference case. In this parameter constellation, we pick up the values
derived according to the methods and considerations described in section IV.
The dynamics of two types of output variables are depicted:
42 This cannot be done by increasing the weights w1 and w2 as this would lead to a dominance of the
corresponding component forces even if the profit resp. market share are close to or even greater than the
corresponding aspiration level.
28
papers on agent-based economics
•
nr 3
Behavioral and cognitive output variables: the number of current innovation and imitation
projects, the number of current cooperations and the mean knowledge level in the course
of time;
•
Monetary output variables: the total profits and sales over time, aggregated over all firm
agents in the region.
As can be seen from Figure 6, the number of current innovation and imitation projects both
show moderate fluctuations around a relatively stable (only slowly decreasing) level. These
fluctuations are opposite to each other, which can be explained by the fact that innovation and
imitation are mutually exclusive action modes (for each firm at each time step) in the model.
Superposed by short-term stochastic fluctuations, mirror-image cycles of innovation and
imitation activities with periods seem to emerge. This may be explained by the fact that an
imitation project needs an innovative product to be imitated which has to be developed and
put on the market previously; therefore a cumulation of innovation projects can produce a
subsequent cumulation of imitations which follow with a certain time lag. This happens also
at the beginning of the simulation run when there exist not yet any innovative products to be
imitated. The mean knowledge level, i.e. the distribution of sharable knowledge components
between agents is increasing over time due to the knowledge transfer within cooperative
innovations. This indicates the spreading of knowledge, being an essential feature of RIS.
However, later on there is a saturation effect which is due to devaluation and forgetting of
knowledge.
Figure 5: Novelty creating action modes and mean knowledge stock on the regional level over time
29
papers on agent-based economics
nr 3
Figure 7 shows the growth of sales of innovative products stemming from the gradual
diffusion of these products. The total sales and profits, however, grow only with a lower rate
because of the partial substitution of demand for conventional products by the demand for
innovative products. Towards the end of the simulation run, the curves for the sales of
innovative products and for total sales converge: Conventional products (the products that
were on the market at the beginning of the simulation) are almost completely replaced by
innovative products. In the same way, new innovative products gradually replace older ones
(innovation vintage model).
Figure 6: Monetary variables over time, aggregated on the regional level
In Figure 8 on the next page it is shown how the mode of action changes over time for the
different behavioral types of agents (cf. section IV.1.1) in the standard parameter
configuration. It can be observed that for the 'conservative' agents the level of imitation is
higher (the level of innovation is lower) than for the other types of agents and that these
agents are aversive against cooperations. The difference between the 'cautious' and
'experimental' agents is given by a slightly higher level of innovations (especially of
cooperative innovations) as well as by a higher amplitude of fluctuations between the
innovation and the imitation mode for the latter type of agents. 43
43 Corresponding to that is the survival for the firms of the different types during the simulation: whereas
the number of 'conservative' firms is increasing, the 'cautious' firms remain about the same and the
'experimental' firms are decreasing.
30
papers on agent-based economics
nr 3
Figure 7: Modes of action for 'conservative' (above), 'cautious' (middle) and 'experimental' agents (below)
31
papers on agent-based economics
nr 3
V.2 Sensitivity analysis
The model which was elaborated in the previous sections contains a couple of behavioral
parameters. According to the behavioral foundation these parameters can be distinguished as
parameters belonging to the attitude component of the explanantia on one side and to the
norm component of the explanantia on the other side. 44 Furthermore these parameters are
different as regards to the procedure by which they are calibrated: either by reflections of the
scientific common sense or by rearranging empirical data. For broadening the explanatory
range of the model, we take the standard parameter configuration as a starting point for
varying these parameters in a sensitivity analysis (simulation runs over 120 time steps). This
gives some clues about the importance of the behavioral traits for the performance of regional
innovation activities. In the following we therefore present the performance results of a two
dimensional parameter analysis. We use the cumulated frequency of innovations and
cooperations respectively as a performance indicator and illustrate how this performance
changes if one parameter related to attitudes and one parameter related to norms is changed
simultaneously. 45
If the flexibility (φ cf. formula [1] above) and risk acceptance (α; cf. formula [5]) are varied
(fig. 9, first row), the number of innovations increases the more the higher these parameters
are. More precisely, the innovation increasing effect of increasing risk acceptance requires a
high level of flexibility, i.e. a high sensitivity of the firm agents as regards to its market
performance. Nevertheless for high levels of risk acceptance an increase of flexibility may
lead to a decline of innovation frequencies. Either a medium range of risk acceptance
combined with high level of flexibility or a high level of risk acceptance combined with a
medium level of flexibility seem to be the parameter areas for the most innovative outcomes
in the region.
Considering the influence of the flexibility (i.e. an increase of φ; cf. formula [1] above) as
well as of the cooperation propensity (i.e. an increase of χ; cf formula [8]) on the cumulated
number of inventions and cooperations as depicted in the second row of figure 9 an increase
in flexibility is accompanied by an increase of innovations and cooperations respectively.
44 This difference is not only of academic importance: presumably the norm-related parameters change
faster than the attitude-related parameters.
45 The results for the types of agents being differently parametrized are aggregated over all types of
agents.
32
papers on agent-based economics
nr 3
According to formula [5] and [6] an increase in flexibility means that both the innovation
force and the imitation force are growing but the imitation force is growing comparatively
faster. Hence, beyond a critical value for the flexibility parameter the amount of innovations
is decreasing again. For a level of flexibility which is not too low (avoiding the dominance of
routine forces) and not too high (avoiding the dominance of imitation forces) and a low level
for the propensity to cooperate (avoiding the binding effect of cooperations 46 ) there is a
maximum of innovations.
In row 3 of figure 9 the influence of the exploration drive (w0 of formula [3]) and of the
weight for the profit aspiration (w1 of formula [3]) is shown. The completed innovations as
well as the completed cooperations are increasing (for the whole range of the aspiration
weight as regards to profits) if the exploration drive is augmented. The reason for that is the
growing innovation force compared with the imitation force (cf. formula [5] and [6]).
Contrary to that, an increase in the weight of profit aspiration is comparatively strengthening
the propensity to innovate. Therefore the maximum amount of innovations and cooperations
respectively is realized if high values for the exploration drive parameter and low values for
the aspiration weight parameter are given.
These selected sensitivity analyses reveal a 'behavioral landscape' behind the observable
novelty creating processes (taking place on a regional level). At the same token, the
singularity of a situation in a given time step and a path starting therefrom is impressively
illustrated. Broadening the perspective to include such a behavioral landscape sheds light on
the possibility of contradicting behavioral requirements for a good performance. Taking for
example the flexibility parameter, one can conclude from the analysis above that in the case
of risk acceptance and cooperation propensity a medium range of that flexibility performs
well in terms of innovations but not so well in terms of cooperations. Another example is the
required low cooperation propensity (if combined with flexibility) and the required high or
low cooperation propensity (if combined with the weight of market share aspiration). This
multi-parameter requirements can be used as a starting point for analysing the implications of
a possible behavioral change (be that intended or not).
46 A high share of cooperating agents limits the number of innovation projects that can be developed
simultaneously: All agents in one cooperation can only develop one innovation together at a time, while
each of these agents could develop its own innovation if they did not cooperate.
33
papers on agent-based economics
nr 3
Completed innovations HcumulatedL
3
2.8
2.6
2.4
2.2
2
0.1
0.2
0.3
0.4
0.5
54.
62.
70.
78.
87.
95.
104
112
120
129
137
145
154
162
170
179
187
195
203
212
220
3.4
3.2
3
Risk acceptance
3.2
Risk acceptance
Completed cooperations HcumulatedL
456
483
510
538
565
592
619
646
674
701
728
755
782
810
837
864
891
918
946
973
100
3.4
2.8
2.6
2.4
2.2
2
0.6
0.1
0.2
0.3
Flexibility
Completed innovations HcumulatedL
0.7
0.65
0.6
0.55
0.4
0.5
0.6
Completed innovations HcumulatedL
289
317
346
374
402
430
459
487
515
543
572
600
628
656
685
713
741
769
798
826
854
0.7
0.6
Exploration drive
0.5
0.4
0.3
0.2
0.1
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.75
0.7
0.65
0.6
0.55
0.1
.
Flexibility
34.
43.
53.
62.
72.
81.
91.
101
110
120
129
139
148
158
167
177
186
196
205
215
224
0.8
Cooperation propensity
0.75
0.3
0.2
0.6
0.5
0.4
0.3
0.2
0.1
0.08
0.7
0.65
0.6
0.55
0.18
0.2
0.12
0.14
0.16
0.18
0.2
0.22
Completed cooperations HcumulatedL
62.
68.
74.
81.
87.
94.
101
107
114
120
127
133
139
146
152
159
165
172
178
185
191
0.85
0.8
Cooperatioin propensity
Cooperation propensitiy
0.75
0.16
0.1
Weight of profit aspiration
0.8
0.14
0.6
27.
33.
40.
47.
54.
60.
67.
74.
81.
87.
94.
101
108
115
122
128
135
142
149
155
162
0.06
494
509
524
539
555
570
585
600
615
630
646
661
676
691
706
721
736
752
767
782
797
0.12
0.5
Completed cooperations HcumulatedL
Completed innovations HcumulatedL
Weight of market share asp.
0.4
0.7
0.2
0.85
0.1
0.3
Flexibility
Weight of profit asp.
0.08
0.6
0.85
Exploration drive
Cooperation propensity
0.8
0.2
0.5
Completed cooperations HcumulatedL
395
423
452
480
509
537
566
594
623
651
680
708
736
765
793
822
850
879
907
936
964
0.85
0.1
0.4
Flexibility
0.75
0.7
0.65
0.6
0.55
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0.22
Weight of market share asp.
Fig. 8: Selected results of sensitivity analysis
34
papers on agent-based economics
nr 3
VI. Discussion and conclusions
We proposed an agent-based simulation model for explaining the occurrence as well as the
outcome of novelty creating processes on the regional level. The features of this simulation
model are legitimized by referring to behavioral concepts as well as to the findings of
empirical surveys. The outcome of this model is an explanation for the modes and frequencies
of novelty generating activities by referring to different groups of explanantia (attitudes,
norms and endowments). This implies that the behavioral component of the novelty
generating procedures can be specified and by this are accessible for reconfiguration.
Moreover such a model can demonstrate the special importance of cooperative innovations
for the emergence of a regional network of knowledge and trust relations.
As regards to the literature about agent-based explanations of innovation activities (cf. Dawid
2006 for a recent survey) the specificity our approach is to consider the novelty creation not
simply as driven by a rule but rather as an activity triggered in particular situations by the
behavioral traits of the agents. This includes differentiating between triggering conditions for
the various modes of novelty creation (imitation, individual innovation and cooperative
innovation). Furthermore the role of the endowment as well as of the transfer of different
types of knowledge for the success of firm's endeavours to create novelties is demonstrated
(especially as regards to the cooperative innovation). What is not yet specified in the model is
a search landscape as well as firm-specific innovations strategies related to such a landscape
and the evaluation of corresponding search results (cf. Beckenbach 2005; Fleming/Sorensen
2003).
As regards to the literature about RIS (cf. Cooke et al. 2004 for a recent survey) the
specificity of our approach is its bottom up perspective by focussing the behavioral
foundation for the emergence (as well as the disappearance) of regional knowledge networks.
Such a behavioral foundation is also necessary for assessing the top down implementation or
RIS as it responds to the question how the behavior of regional agents can be affected by
public incentives and regulations. For answering this purpose the simulation model has to be
enhanced in several respects: Firstly, the role of 'soft' and 'hard' regional institutions (such as
e.g. reputation or subsidies) has to be taken into account. Secondly, the inclusion of
knowledge generating public or semi-public institutions (like universities and research
facilities) is necessary for a full range explanation of regional knowledge diffusion. Thirdly,
35
papers on agent-based economics
nr 3
the empirical classification of agents has to be enriched by picking up more findings of the
surveys. Possibly this can be done by feeding a cluster analysis with available empirical data.
Nevertheless at least partially some 'stylized facts' of regional innovation processes can be
reproduced by the simulation model: the ongoing heterogeneity of firm agents and their
actions, the changing core of an innovation network between firms and the corresponding
spread of knowledge within the region.
36
papers on agent-based economics
nr 3
References
Ajzen, I. (1991). The Theory of Planned Behavior. Organizational Behavior and Human
Decision Process 50: 179-211.
Antonelli, C. (2000), Collective knowledge communication and innovation: the evidence of
technological districts, Regional Studies 34: 535-547.
Asheim, B. T. and L. Coenen (2005). Industrial Dynamic, Innovation and Development.
Research Policy 34: 1173.
Beckenbach, F. (2005). Knowledge Representation and Seach Processes: A Contribution to
the Microeconomics of Invention and Innovation. Volkswirtschaftliche
Diskussionsbeiträge. Kassel.
Beckenbach, F., R. Briegel and M. Daskalakis (2007). The Influence of Regional Innovation
Systems on Regional Economic Growth: Linking Regional Input-Output Analysis
and Agent Based Modeling. papers on agent-based economics. Kassel.
Braczyk, H.M., P. Cooke, and M. Heidenreich (eds) (2004). Regional Innovation Systems:
The Role of Governance in a Globalized World. London, Routledge
Corral, C. M. (2002). Environmental Policy and Technological Innovation. Cheltenham,
Edward Elgar.
Cyert, R. M. and J. G. March (1992). A Behavioral Theory of the Firm. Oxford, Blackwell.
Dawid, H. (2006). Agent-Based Models of Innovation and Technological Change. Handbook
of Computational Economics, Vol 2. L. Tesfatsion and K. L. Judd. Amsterdam,
North-Holland: 1235-72.
Daskalakis,M. and Kauffeld, M. (2006): The microeconomics of innovation networks: On the
dynamics of knowledge generation and trust building (mimeo).
Daskalakis, M. and D. Krömker (2007): Investigating innovation with behavioral-sciences:
empirical evidences for the utility of an agent based approach. (mimeo)
Fleming, L. and O. Sorenson (2003). Navigating the Technology Landscape of Innovation.
MIT Sloan Management Review(Winter): 15-23.
Fritsch, Michael and Franke,Grit (2004), Innovation, regional knowledge spillovers and R&D
cooperation, Research Policy 33, 245–255.
Gintis, H. (2003). Towards a Unity of the Human Behavioral Sciences. Santa Fe
Institute/working paper. Santa Fe.
Iammarino, S. (2005). An Evolutionary Integrated View of Regional Systems of Innovation:
Concepts, Measures and Historical Perspectives. European Planning Studies 13(4):
497-519.
Jager, W. (2000). Modelling Consumer Behavior. Groningen: Universal Press.
37
papers on agent-based economics
nr 3
Kahneman, D. and A. Tversky (1979). Prospect Theory: An Analysis of Decision under Risk.
Econometrica 47: 265-91.
Lant, T. K. (1992). Aspiration Level Adaptation: an Empirical Exploration. Management
Science 38(5): 623-44.
Loewenstein, G. (1994). The Psychology of Curiosity: A Review and Reinterpretation.
Psychological Bulletin 116(1): 75-98.
Louis, M.R. and R.I. Sutton. (1991). Switching Cognitive Gears: From Habits to Active
Thinking. Human Relations,44(1): 55-76.
March, J. G. (1994). A Primer on Decision Making. New York, The Free Press.
March, J. G. and H. Simon (1993). Organizations. Cambridge/Mass., Blackwell.
Mezias, S. J., Y.R. Chen and P.R. Murphy (2002). Aspiration-level Adaptation in an
American Financial Services Organization: A Field Study. Management Science
48(10): 1285-1300.
Petts, J. (1998). Environmental Responsiveness, Individuals and Organizational Learning:
SME Experience. Journal of Environmental Planning and Management 41(6): 71130.
Scitovsky, T. (1976). The Joyless Economy. New York, Oxford University Press.
Simon, H. a. (1997). Towards an Empirically Based Microeconomics. Cambridge, Cambridge
University Press.
Svenson, O. (1990). Some Propositions for the Classification of Decision Situations, in:
Borcherding, I.O. (ed.) Contemporary Issues in Decision Making, Amsterdam: NorthHolland.
Sternberg, Rolf (2000), Innovation Networks and Regional Development—Evidence from the
European Regional Innovation Survey (ERIS): Theoretical Concepts, Methodological
Approach, Empirical Basis and Introduction to the Theme Issue., European Planning
Studies 8.
Verplanken, B. and H. Aarts (1999). "Habit, Attitude, and Planned Behaviour: Is Habit an
Empty Construct or an Interesting Case of Goal-Directed Automaticity?" European
Review of Social Psychology 10: 102-134.
Witt, U. (2006). Evolutionary Economics and Psychology. Papers on Economics & Evolution.
Jena.
38
papers on agent-based economics
nr 3
Appendix
Mann Whitney Test significances
(2-tailed)
variable
Risk
acceptance
Exploration
drive
Profit
aspiration
Market share
aspiration
Cooperation
propensity
Regional
trust
para- datameter set
F_IIM
and
F_IR
F_ROUT
and
F_IIM
Kendall’s Tau b
correlation coefficient
F_ROUT
and
F_IR
F_IR
F_IIM
F_ROUT
α
D1
D2
0.051
0.724
0.075
0.092
-0.54
0.082
-0.091
-0.057
-0.098
0.000
-
-
0.236**
-0.205**
-
w0
D1 47
D2 48
0.231
0.143
0.001
0.199**
-0.089
-0.139*
w1
D2
0.483
0.686
0.128
0.019
0.103
-0.125
w2
D2
0.021
0.413
0.030
0.168*
-0.025
-0.067
Χ
D1
0.000
-
-
0.276**
- 0.120*
-
tr0
D1
0.788
-
-
0.082
0.004
-
**indicates significance at 1% level (2-tailed);
* indicates significance at 5% level (2-tailed).
Impressum:
papers on agent-based economics
Herausgeber:
Universität Kassel
Fachbereich Wirtschaftswissenschaften (Prof. Dr. Frank Beckenbach)
Fachgebiet Umwelt- und Innovationsökonomik
Nora-Platiel- Str. 4
34127 Kassel
www.ivwl.uni-kassel.de/beckenbach/
ISSN: 1864-5585
47 This variable comprises the items „developing new markets“„creating new need“.
48 This variable comprises the items „experimental drive“ and “creativity”.
39