Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Abstract
In recent years, a number of studies have considered the application of chaos theory to economics; the primary focus, however, has been
the implications for stock prices, the foreign exchange market, and the macroeconomy. This paper describes a non-linear model of duopolistic
competition, which focuses on a firm's expenditure and the resulting quality or technological endowment of its product. Results from
computer iteration of the model are presented which indicate that chaotic outcomes are possible for a range of competing managerial policies;
the associated unpredictability is due solely to the dynamics of the interaction. The study also provides the results of some initial work on
how management adaptation may act to forestall chaotic outcomes. D 2000 Elsevier Science Inc. All rights reserved.
0148-2963/01/$ ± see front matter D 2000 Elsevier Science Inc. All rights reserved.
PII: S 0 1 4 8 - 2 9 6 3 ( 9 9 ) 0 0 0 5 0 - 8
180 S. Whitby et al. / Journal of Business Research 51 (2001) 179±191
outside (exogenous) factors. It is solely due to the mathe- The most fundamental requirement for chaos is non-
matics of the relationship between variables endogenous to linearity. In the context of Eqs. (1) and (2), this essentially
the system. In addition, such systems can have periods of means that at least one state variable will appear in the
regular behavior, such as equilibrium and periodicity, that functions fi or gi (where i denotes the index of a general
arise abruptly and disappear just as suddenly. variable) with a non-linear relationship to the rate or the
The state of an evolving system can be represented by a next period value, respectively. Over and above this, precise
number of time-dependent variables; the minimum number specification of the prerequisites for chaos of a general
of variables (n) needed to characterize a system is called its system is difficult; there are presently no general criteria to
dimension. As a system evolves, its state can be considered establish the necessary and sufficient conditions for a
to exist in a space of dimension n called phase space, the system of differential or difference equations to be chaotic.
components of which correspond to the system's n state Requirements have been established, however, regarding
variables. A geometric space with axes given by the phase the minimum dimensionality of the system. Different con-
space components (i.e. the system variables) provides a ditions exist for flows and maps (Ott, 1993). For the case of
setting for the graphic illustration of a system's evolution. In first-order, autonomous ordinary differential equations, i.e.
particular, for a dissipative system (i.e. one possessing a of the form of Eq. (1), the lowest dimension for which
mechanism to dissipate perturbations from steady motion), a chaos is possible (nmin) is 3. For a mapping, the minimum
steady-state behavior can take place in phase space on an required dimensionality differs depending on whether or not
attractor. Crutchfield et al. (1986) define a system's attractors the mapping is invertible. Using notation as in Eq. (2), an
as geometric forms that characterize the system's long-term invertible map is one whose equations, as well as giving a
behavior in phase space. Chaotic evolutions of dissipative unique value for each xi,t + 1 from the set of xi,t can also be
systems in phase space occur on strange attractors, which are solved to give a unique value for each xi,t from the xi,t + 1.
complicated shapes of fractal structure, first observed by For an invertible map, there cannot be chaos unless nmin
Lorenz (1963). Many studies of non-linear dissipative dyna- 2. For a non-invertible map, chaos is possible even for a
mical systems use strange attractors to characterize chaos. dimensionality of one. The logistic equation introduced by
In order to predict future states of a time-dependent May (1976) in a study of population, is a one-dimensional
system, a set of modeling equations can be developed. Such map often used to illustrate chaotic behavior. It is a finite
a set of equations is called a dynamical system, which may difference equation with a quadratic non-linearity or hump
evolve continuously or discretely over time depending on and takes the general form xt + 1 = l x t (1ÿxt). In population
the real system modeled. A continuously evolving dynami- studies, xt + 1 represents the population level and depends
cal system is called a flow, while the discrete case is called a on its immediate predecessor, xt. The parameter l is the
mapping. Flows are represented by differential equations, growth rate from one cycle to another and x is constrained
while difference equations describe the discrete case. The by a resource constraint modeled as 1ÿxt. Through feed-
dependencies governing the evolution over time, t, of a back leading to period doubling (see later), the system's
deterministic flow can be expressed as a set of first-order evolution can become chaotic as the parameter value, l, is
differential equations giving the rates of change of the n increased, with chaos occurring when l is between around
system variables, xi, in terms of the same variables, i.e.: 3.6 and 4.
d x1 Whereas little can be said, a priori, concerning a system's
f1
x1 ; x2 ; . . . ; xn ; t capability of producing chaos, for systems shown to be
dt
chaotic some typical ``routes'' or ``scenarios'' have been
d x2 identified by which the onset of chaos may be approached.
f2
x1 ; x2 ; . . . ; xn ; t
dt
1 In each case, the transition is effected by continually
# changing (usually increasing) one or more ``control'' para-
meters, such as l in the logistics equation above. Normally,
d xn only one parameter is varied, in which case the scenario is
fn
x1 ; x2 ; . . . ; xn ; t
dt known as a codimension Ð one route to chaos. Generally, at
For a discrete-time-mapping, the evolutionary laws can be distinct values of the changing parameter (l) bifurcations
similarly expressed, but as difference equations giving the occur. Bifurcations involve qualitative changes in the beha-
values of the system variables in period t + 1(xi,t + 1) in vior of the system, for example, from equilibrium to
terms of their values in the previous period, t(xi,t), i.e.: periodic or from periodic to chaotic (for a fuller explanation
see, e.g. Berge et al. 1984; Eckmann, 1981; Medio, 1992;
x1;t1 g1
x1;t ; x2;t ; . . . ; xn;t ; t Ott, 1993). One scenario, known as intermittency, involves
x2;t1 g2
x1;t ; x2;t ; . . . ; xn;t ; t interruption of otherwise regular periodic evolution with
2 bursts of irregular, chaos-like behavior as l is varied above/
# below a critical value. The bursts are usually seen to become
more frequent as l is increased/decreased until the original,
xn;t1 gn
x1;t ; x2;t ; . . . ; xn;t ; t regular behavior is lost and becomes chaotic.
S. Whitby et al. / Journal of Business Research 51 (2001) 179±191 181
Another route to chaos for an initial period system, is The interplay between two departments functioning se-
period doubling. As l is increased, bifurcations occur at parately and in parallel, as opposed to in sequence as in the
particular values as additional periodicities complicate the stock management problem, was considered in a study by
system's cycle. The frequency of bifurcation increases as l Rasmussen and Mosekilde (1988). This study proposed a
is progressively increased until at a finite value of l the model of a production company where resources are shared
number accumulates to infinity; increasing l beyond this between production and marketing in accordance with a
point results in the evolution becoming chaotic. Period decision rule reflecting variations in inventory or the back-
doubling will be described in further detail later, when log of customer orders.
considering the results of our model. Other studies have suggested that the incidence of chaos
In the light of what has been described above, we can in business may be much wider. For example, it has been
say that any possible candidate to represent a chaotic suggested that there may be a tendency towards strongly
system must satisfy the non-linearity and minimum dimen- non-linear local interactions where there are nonconvexities
sionality requirements. In addition, the model must be in production technology, due to technological indivisibil-
oscillatory, or at least have the potential to be so, in order ities for instance (Gordon, 1992). Gordon also used a
not to preclude the possibility of its developing from variation of the logistic equation to produce a plausible
periodic to chaotic evolution. model of the relationship between a company's advertising
budget and its resulting sales. Further economic relation-
2.2. Chaos and the firm; inventory control ships subject to feedbacks and timing, which have been
investigated for chaotic outcomes, are investment behavior
If microeconomic institutions such as firms and markets (Mosekilde et al., 1992), the planning of teacher ±student
are governed by linear, or simple non-linear relationships, it ratios in education (Feichtinger and Novak, 1992), and
is possible to study them separately (adopt a reductionist advertising outlays (Baumol and Benhabib, 1989).
methodology) without losing crucial characteristics of their The effect of research and development (R&D) activity
behavior. Interactions with other firms or markets can be on a firm's time path has been modeled by Baumol and
relatively easily assimilated into the models as generally Wolff (1983), Feichtinger and Kopel (1993), and Kopel
well-behaved positive and negative feedback effects. Equi- (1996). Baumol and Wolff focused on the relationship
librium outcomes result and uncertainty as to the model's between R&D and productivity growth, showing chaos to
outcome is minimized. If economic systems are significantly be possible for certain conditions. The Feichtinger and
non-linear, however, then their behavior may be highly Kopel and Kopel models, on the other hand, described a
unpredictable, and this limits the application of both tradi- non-linear increase of sales for higher R&D allocations,
tional neoclassical market models and more recent games employing a treatment closer to that presented here. Nota-
theoretic modeling involving Bayesian techniques. bly, Feichtinger and Kopel placed particular emphasis on the
Many of the non-linear models, which have identified the influence of the decision-maker (i.e. the manager) whose
potential for chaos in microeconomics, involve discrete-time input amounted to periodic adjustments to R&D funding in
queuing or consideration of inventory. A typical example is the light of current sales. This was represented by a non-
the beer distribution model, an exercise in beer-stock linear decision rule that is well-founded in behavioral
management involving non-linearity and feedback (Mose- decision theory, and which also features in the model used
kilde and Larsen, 1988; Sterman, 1988; Van Ackere et al., in this study, namely the anchoring and adjustment heur-
1993). Mosekilde and Larsen presented an overview of this istic, described in Tversky and Kahneman (1974). A variety
model's behavior, producing a time series through iterations of temporal behavior was observed, including chaos.
of the model and identifying chaos for certain parameter In the above studies, the model parameters representing
values. Sterman's approach was notably different in his use human behavior were kept at a constant value throughout a
of the model to provide a simulated environment in which given evolution. This failure to allow parameter variability
subjects could take part in a beer distribution game. To within evolutions effectively ignores the fact that decision-
describe the subjects' behaviors, a general heuristic was makers may adjust their responses over time; their actions
constructed, which, on further iteration, suggested chaos for may evolve through some form of iterative process, for
a significant minority of values. example. There is a comment on this point in Sterman
The common feature in these inventory control models (1988, p. 174).
is a delay between the ordering of new stock and its
delivery. Furthermore, disruption in one part of the supply 2.3. Chaos and oligopoly
chain leads to a sequence of changes in other parts;
expansion of output, for example, leads to excess capacity Oligopolistic industries demonstrate phases of co-opera-
developing when orders fall. Overall, this suggests a tive behavior and phases of intense price and non-price
tendency for oscillation in supply behavior, which, for competition. Such phases result from oligopolists taking
certain conditions, can become chaotic (Erramilli and into consideration the reactions of their competitors, as well
Forys, 1991; Levy, 1994). as consumers. Competitors may retaliate to price cuts, for
182 S. Whitby et al. / Journal of Business Research 51 (2001) 179±191
example, by also cutting prices so as not to lose market standard of achievement in terms of the resulting product,
share. Developments in oligopoly theory have shown that relative to that of the competition. In addition, this stan-
various other outcomes are possible depending, for exam- dard is, for both firms, a direct result of the level of inputs,
ple, upon the financial strength and production conditions i.e. resources applied. In constructing the model, particular
for each of the firms. In recent years, it has been recognized emphasis was placed on the resources that each allocates
that competition among a few firms can be usefully modeled to R&D, and the resultant (measurable or subjective)
in a game theoretic framework that contains feedback, time, product standard/quality, such as the number of features
and uncertainty. In game theoretic models of oligopoly, or durability. The latter are means by which discerning
rivalry is studied from an equilibrium (usually a Nash consumers may differentiate in a market where product
equilibrium) perspective and uncertainty is dealt with by standard is an important order winning criterion. Expendi-
an expected profit or payoff maximization. The outcome ture on R&D could also result in a change in the process
then depends upon the assumptions made about actions and of production rather than a change in the product, having
reactions at the various stages of the game (e.g. see implications for costs of supply; this impact on process is
Hargreaves Heap and Varoufakis, 1995; Phlips, 1995). not modeled here. In the model, an increase in a compa-
Game theory has introduced dynamism into the study of ny's resource allocation to R&D is considered to lead to an
the competitive process, opening up the possibility of improved product standard (quality or technological en-
sufficiently non-linear reaction functions to give chaotic dowment of the product), while having a product standard
outcomes. Rand (1978) considered the possible conditions superior to that of the competition is deemed to constitute
in which chaotic behavior might result, and more recently, a competitive advantage leading ultimately to greater sales
Byers and Peel (1994) flagged the possible contribution the and profits. Current sales, therefore, are considered to be a
analysis of dynamic non-linear behavior might make to the direct result of the current, relative product standards of the
study of industrial economics. To date, however, there has competitors and so do not need to feature explicitly in the
been limited analysis of the potential for chaos in oligopo- model. Each firm is assumed to be able to monitor readily
listic markets. Noteworthy studies include that by Dana and the product standard of its competitor.
Montrucchio (1986), which showed how chaotic dynamical The model is structured to represent two companies, A
paths might arise in duopoly where there are small rates of and B, for which the resource allocations to R&D for period
discount, and that by Puu (1994), who considered an t are denoted by RAt and RBt, resulting in product quality
adjustment process for a pair of Cournot duopolists. Puu standards denoted for A and B during period t by SAt and
demonstrated mathematically that with an iso-elastic de- SBt, respectively. Modeling R&D to impact on the product
mand function and constant marginal costs, the duopoly standard in the same time period is admittedly a simplifica-
system could exhibit periodic, and even chaotic behavior. tion Ð R&D might be expected to affect product standard in
In Section 3, a new, non-linear model of duopolistic t + n where n varies depending on the product and industry;
interaction is described in which the effect of managerial the simplification is introduced for ease of modeling. A
policy plays a central role. The formulation is based on the notable consequence of the simplification is that for condi-
relationships specified in the R&D model of Feichtinger and tions under which the system oscillates over time, the
Kopel (1993). However, while their study included manage- fluctuations of the separate companies will be either in
rial judgements relative to an internally determined target phase (peaks coincide) or in anti-phase (one's peaks coin-
performance, the following extends the analysis by applying cide with the other's troughs).
the relationships to the interaction between two separate, To facilitate description, X, RXt and SXt are used to
competing companies, the policy judgements of each being represent either company and its associated system vari-
made in light of its own performance relative to the other. In ables for period t, in the general case. Arising from
addition, while Feichtinger and Kopel's treatment, along managerial decisions made at the end of each time period,
with the vast majority of model simulation studies, makes changes to R&D allocation are modeled as dependent on
no allowance for policy adjustment over time, this study the current relative quality or technological standing of the
makes some attempt to include the management's capacity firm's products.
to adapt its strategy. The interactions within the system are shown in Fig. 1,
where the arrows between variables represent deterministic
cause and effect Ð the ``tail'' variables affecting the ``heads,''
3. A basic model of duopolistic competition with e.g. RAt affects SAt. The dotted arrows describe similar
non-linear dynamic interactions effects but between tail variables and the head variables of
the following period, e.g. SAtÿSBt affects RAt + 1.
The proposed model describes two manufacturers pro- For both companies, the demand for the output in the
ducing a similar technological product and competing market is assumed to be a function of the product standard.
through product innovation. In both cases, managerial Other variables that may impact on demand, such as price,
decisions regarding the extent of an input or resource are assumed to be unimportant or constants. While it is
allocation are made in the light of the organization's recent recognized that the amount of resources dedicated to R&D
S. Whitby et al. / Journal of Business Research 51 (2001) 179±191 183
4.2. Arrangement group (+,ÿ)/(ÿ,+) the smaller allocation (RA0 > RB0). The system again
showed monotonic behavior rather than oscillation or chaos.
The groups of opposing managerial policies (+,ÿ) and This behavior can be explained by the fact that here, B, with
(ÿ,+) are identical due to the structural symmetry of the its sensitive reinforcing policy, increases its initial disad-
model and while the following refers to (+,ÿ), it applies vantage and so directs SBt towards zero. Meanwhile, the
equally to (ÿ,+). For these arrangements, on occasions smaller counteracting policy of A means this company acts
when the firm with the reinforcing policy is ahead in terms so as to relax on its advantage. The overall result of these
of product standard, this firm invests more in R&D. In tendencies is that both SBt and SAt fall to zero, with the
such cases, the trailing competitor (with a counteracting former having the more sensitive policy, thus reducing the
policy) does the same in the hope of making good its more rapidly.
present loss of competitive advantage. Equally for this
arrangement group, circumstances occur for which the 4.2.3. Scenario 3
counteracting firm has the advantage and, true to its policy, Investigations were also carried out on a third type of
relaxes and invests less in R&D, perhaps expecting to (+,ÿ) arrangement, somewhat different from the two
maintain its competitive advantage without additional R&D considered above. In this case, the one company (B) had
effort. In this case, the trailing, reinforcing competitor also a sensitive counteracting policy (d = a large negative
invests less, representing a firm which, when trailing, sees number), while the other (A) had a less sensitive reinfor-
no point in trying to compete through R&D to improve its cing policy. Further, the initial conditions were such that
market share. the counteracting company started with a greater allocation
Investigation of the model showed the potential for (RB0 > RA0), which translated to SB0 > SA0. For these
oscillation to depend not only on the modulus and sign conditions, for certain parameter values, oscillatory beha-
of the parameters, but also on the initial relative standing vior was observed, shown for variables RAt and RBt in
of the two companies. This led to the consideration of four Fig. 3(a) and (b), respectively. It is notable that while the
general scenarios. two variables took values over different ranges, their
oscillations were in phase.
4.2.1. Scenario 1 The potential for chaos in non-linear systems is often
Arrangements were investigated for which one company investigated by varying one system parameter while the
(say B) had a sensitive reinforcing policy (d = a large others are kept constant. The occurrence of increasingly
positive number), while the other (A) had a less sensitive complicated periodic behavior leading to chaos as a para-
counteracting policy (c = a smaller negative number), with meter is varied and is well-recognized in chaotic systems.
initial conditions such that the reinforcing company started Where one parameter is varied, the scenario is known as a
with the greater allocation (RB0 > RA0). For these condi-
tions, as for (+,+) arrangements, the model showed no
potential for oscillation or chaos. In this case, the RAt and
RBt (and so SAt and SBt) variables both increased repeat-
edly over time. As before, this tendency for monotonic
variation rather than oscillation can be understood from the
system's feedback structure, as determined by the policies.
For these arrangements and initial conditions, the larger
initial funding of B translates to a larger initial product
standard SB0. As the model is iterated and the system
evolves, B, with its sensitive reinforcing policy, will
further increase funding and so amplify its advantage. At
the same time, A, with its counteracting policy, attempts to
remedy its disadvantageous initial position by increasing
its funding. Due to the lower sensitivity of its policy,
however, A repeatedly redresses B's advantage less than
the extent to which B further increases it. Overall, there-
fore, SBt and SAt will diverge to the upper bound of the
model, but with SBt > SAt at all times before the upper
bound is reached.
4.2.2. Scenario 2
Further (+,ÿ) arrangements were investigated for poli- Fig. 3. (a) Period 1 oscillations of RAt for system arrangement (c, d) =
cies with relative sensitivities as above, but with initial (+0.1, ÿ3.8), initial conditions (RA0, RB0) = (10.0, 15.0). (b) Period 1
conditions such that the reinforcing company started with oscillations of RBt, with system arrangement and initial conditions as in (a).
186 S. Whitby et al. / Journal of Business Research 51 (2001) 179±191
codimension-one route to chaos. Generally, at distinct For some non-linear systems, if the control parameter is
values of the changing parameter, bifurcations occur. Bifur- progressed past a critical value, the motion becomes aper-
cations involve qualitative changes in the behavior of the iodic, producing the phenomenon known as deterministic
system, for example, from equilibrium to periodic (for a chaos. Fig. 4(c) shows the model exhibiting such chaotic
fuller explanation see e.g. Berge et al., 1984; Eckmann, behavior; there is no longer a discernible repeating pattern to
1981; Medio, 1992; Ott, 1993). Such a process was ob- the series, which has become aperiodic. A comparison of
served for these system conditions, as shown in Fig. 4(a), Fig. 4(c) and (d) illustrates sensitive dependence on initial
(b), and (c) for which the system's behavior was studied for conditions (sdic), known to be a characteristic of chaotic
increasingly negative settings of d. These changes corre- evolutions. Both series emanate from identical system
spond to a company (B in this case) becoming increasingly arrangements, i.e. the company policies are the same in
sensitive in its counteracting policy, while sensitivity at the each case, but the initial values of the evolutions, as
reinforcing competitor remains unchanged. specified by (RA0, RB0), differ slightly: initial conditions
The plots in Fig. 4 indicate classic period doubling. In were (10.0, 15.0) and (10.1, 15.0) for Fig. 4(c) and (d),
Fig. 4(a), for which policy parameter d = ÿ4, the evolution respectively. The marked difference in the two (chaotic)
of RBt was found to undergo two oscillations, with two evolutions, as a result of sdic, suggests serious, fundamental
different maxima and minima, before a full cycle back to the limitations to forecasting, control and planning where such
initial value was completed. Behavior of this type is known system arrangements exist.
as a period 2 evolution, while the most basic oscillations of A useful illustrative tool in the analysis of systems,
Fig. 3 are termed period 1, the system bifurcating from the which pursue the period doubling route to chaos, is a
latter to the former as the control parameter (d) is varied. In bifurcation diagram. A bifurcation diagram indicates how
Fig. 4(b), d has become more negative, d = ÿ4.18, and the post-transient, limiting behavior of a system alters as one
another bifurcation has occurred so that the periodic beha- parameter is varied. One of its great strengths is that it
vior has become even more complicated; each complete provides a more general, global view of the system's
period now comprises four sub-oscillations: a period 4 behavior for a variety of parameter settings; in this case,
evolution. In theory, this period doubling as a system company policies. More specifically, for this discretely
parameter is increased/decreased will continue for even evolving model, the bifurcation diagram shown in Fig. 5
higher periods. For increasingly higher period numbers, identifies the values of a particular system variable (say
however, the bifurcations occur for progressively smaller RAt) present in the post-transient evolution, plots these
changes in the increasing/decreasing parameter, and there- values against the corresponding value of the parameter,
fore, become progressively more difficult to observe. and repeats this for a range of parameter values. The
Fig. 4. (a) Period 2 oscillations of RBt for system arrangement (c, d) = (+0.1, ÿ4.0), initial conditions (RA0, RB0) = (10.0, 15.0). (b) Period 4 oscillations
of RBt for system arrangement (c, d) = (+0.1, ÿ4.18), initial conditions (RA0, RB0) = (10.0, 15.0). (c) Chaotic behavior of RBt for system arrangement
(c, d) = (+0.1, ÿ4.5), initial conditions (RA0, RB0) = (10.0, 15.0). (d) Chaotic behavior of RBt for system arrangement as in (c), initial conditions (RA0,
RB0) = (10.1, 15.0).
S. Whitby et al. / Journal of Business Research 51 (2001) 179±191 187
4.2.4. Scenario 4
Oscillations, period doubling, and chaotic behavior simi-
lar to that described above were again obtained for certain
parameter values for a fourth type of (+,ÿ) arrangement.
The relative sensitivities of the policies were similar to those
for the conditions producing the oscillations and chaos
described above, but here, the initial conditions were such
that the counteracting company started with the smaller
allocation (RA0 > RB0). The tendency for oscillation could
be explained by the fact that for these conditions, the
sensitive counteracting B acts to remedy its initial disad-
vantage, while the less sensitive reinforcing A tends to
increase its advantage. As before, B's effect is the stronger,
so that the system evolves to a state where RBt > RAt, and
the policies now engender opposite effects. This change in
direction of the effects due to policy occurs repeatedly, and
it follows that for these conditions, there is a tendency for
Fig. 5. Bifurcation diagram for (+,ÿ) system arrangements; policy
the system again to oscillate.
parameter d varying.
Fig. 6. (a) Period 1 oscillations of RAt for system arrangement (c, d) = (ÿ0.1, ÿ3.0), initial conditions (RA0, RB0) = (10, 5.0). (b) Period 1 oscillations of RBt
for system arrangement and initial conditions as for (a). (c) Period 2 oscillations of RBt for system arrangement (c, d) = (ÿ0.1, ÿ3.8), initial conditions (RA0,
RB0) = (10, 5.0). (d) Period 4 oscillations of RBt for system arrangement (c, d) = (ÿ0.1, ÿ3.99), initial conditions (RA0, RB0) = (10, 5.0). (e) Chaotic behavior
of RAt for system arrangement (c, d) = (ÿ0.1, ÿ4.15), initial conditions (RA0, RB0) = (10, 5.0). (f) Chaotic behavior of RBt, system arrangement and initial
conditions as for (e).
suggests that for these conditions, the system is prone to tion process by permitting variation over time of the policy
oscillation where the scale of variation of the more parameters c and d.
sensitive company is substantially greater than that of
the competition. 5.1. Adaptation mechanisms
Fig. 8. (a) Evolution of RBt with proportional adaptation mechanism. (b) Fig. 9. (a) Evolution of RAt with absolute adaptation mechanism. (b)
Further evolution of RBt with proportional adaptation mechanism. Further evolution of RAt with absolute adaptation mechanism.
190 S. Whitby et al. / Journal of Business Research 51 (2001) 179±191
6. Conclusions
tions of actions when paired with those of the competition Cable J, editor. Current Issues in Industrial Economics. London: Mac-
millan, 1994. pp. 227 ± 59.
(i.e. a more global view of system behavior, such as that
Crutchfield JP, Farmer JD, Packard NH, Shaw RS. Chaos. Sci Am
given by bifurcation diagrams), appropriate management 1986;255:38 ± 49.
actions could be taken to forestall chaotic outcomes. Dana RA, Montrucchio L. Dynamic complexity in duopoly games. J Econ
Conditions of chaos will obviously be very troublesome Theory 1986;40:40 ± 56.
to managers intent on forecasting markets and planning Day RH. The evolving economy. Eur J Oper Res 1987;30:251 ± 57.
and controlling their organization: both the lack of order Eckmann JP. Roads to turbulence in dissipative dynamical systems. Rev
Mod Phys 1981;53(4):643 ± 54.
in behavior over time and the property of sdic reduce the Erramilli A, Forys LJ. Oscillations and chaos in a flow model of a switching
possibility of effective management judgement. It follows, system. IEEE J Sel Areas Commun 1991;9:171 ± 8.
therefore, that for two duopolistic competitors intent on Feichtinger G, Kopel M. Chaos in nonlinear dynamical systems exempli-
ordered trajectories, there are great benefits to be had if fied by an R&D model. Eur J Oper Res 1993;68:145 ± 59.
both have a knowledge of their combined behavior and Feichtinger G, Novak AJ. How Stock Dependent Flow Rates may Imply
Chaos in Educational Planning, mimeo Institute of Econometrics and
the pairings of their individual policies over the relevant Department of Statistics, University of Vienna, 1992.
range of parameter space. If this were the case, then for Gordon TJ. Chaos in social systems. Technological Forecasting and Social
both organizations, chaos could be more easily avoided Change 1992;42:1 ± 15.
and more regular time paths achieved. This implies that an Hargreaves Heap SP, Varoufakis Y. Game Theory: a Critical Introduction.
aspect of co-operation (or at least honesty with a compe- Routledge, London, 1985.
Kelsey D. The economics of chaos or the chaos of economics. Oxford Econ
titor regarding policy) would be desirable if unexpected Pap March 1988;40(1):1 ± 31.
outcomes are to be prevented. This conclusion, in turn, Kopel M. Periodic and chaotic behaviour of a simple R&D model. Ric
has implications for competition policy: collaboration Econ 1996;50:235 ± 65.
between duopolists may help to dampen what otherwise Levy D. Chaos theory and strategy: theory, application, and managerial
would be highly erratic fluctuations in the market place implications. Strategic Manage J (special issue, summer) 1994;15:
167 ± 78.
(here reflected in applications of R&D resources), suggest- Lorenz E. Deterministic nonperiodic flow. J Atmos Sci 1963;20(2):130 ± 41.
ing that collaboration could, in this sense, possibly be May R. Simple mathematical models with very complicated dynamics.
welfare-enhancing. Nature 1976;262:459 ± 67.
Turning to the possibilities for future research, an ob- Medio A. Chaotic Dynamics: Theory and Applications to Economics. Cam-
bridge: Cambridge Univ. Press, 1992 (in collaboration with G. Gallo).
vious refinement to the study would be the extension of the
Mosekilde E, Larsen ER. Deterministic chaos in the beer-production dis-
decision rules to allow for more flexible management tribution model. Syst Dyn Rev 1988;4(1 ± 2):131 ± 47.
policies and the inclusion of time lags between resource Mosekilde E, Larsen ER, Sterman JD, Thomsen JS. Nonlinear mode-inter-
allocation and the resulting changes, e.g. in product stan- action in the macroeconomy. Ann Oper Res 1992;37:185 ± 215.
dard. Initial work incorporating such a time lag suggests that Ott E. Chaos in Dynamical Systems. Cambridge: Cambridge Univ.
the system is dampened, but for reasons that are not entirely Press, 1993.
Parker D, Stacey R. Chaos, Management and Economics: the Implications
clear. In addition, the model could be amplified to include of Non-linear Thinking, Hobart Paper 125. London: Institute of Eco-
additional competition, i.e. an extension from duopoly to nomic Affairs, 1995.
oligopoly. At a time when industry is moving towards ``lean Phlips L. Competition Policy: A Game-Theoretic Perspective. Cambridge:
supply'' manufacturing, which involves reduced time lags Cambridge Univ. Press, 1995.
between consumer orders and input supplies, the possible Puu T. The Chaotic Duopolist, Working Paper no. 64. Munich: Centre for
Economic Studies, University of Munich, 1994.
critical effect of time lags on the behavior of economic Radzicki MJ. Institutional dynamicals, deterministic chaos, and self-orga-
systems has important policy implications. nizing systems. J Econ Issues 1990;24(1):57 ± 102.
Rand D. Exotic phenomena in games and duopoly models. J Math Econ
1978;5:173 ± 84.
References Rasmussen DR, Mosekilde E. Bifurcations and chaos in a generic manage-
ment model. Eur J Oper Res 1988;35:80 ± 8.
Allen PM. Coherence, chaos, and evolution in the social context. Futures. Sterman JD. Deterministic chaos in models of human behavior: meth-
1994;26(6):583 ± 97. odological issues and experimental results. Syst Dyn Rev 1988;4
Anderson PW, Arrow KJ, Pines D, editors. The Economy as an Evolving (1 ± 2):148 ± 78.
Complex System. Proceedings of the Evolutionary Paths of the Global Sterman JD. Modeling managerial behaviour: misperceptions of feed-
Economy Workshop, September 1987, Santa Fe, New Mexico. Red- back in a dynamic decision making experiment. Manage Sci
wood City, CA: Addison-Wesley, 1988. 1989;35:321 ± 39.
Baumol WJ, Benhabib J. Chaos: significance, mechanism, and economic Stewart I, Cohen J. Why are there simple rules in a complicated universe.
applications. J Econ Perspect 1989;3(1):77 ± 105. Futures 1994;26(6):648 ± 64.
Baumol WJ, Wolff EN. Feedback from productivity growth to R&D. Scand Tversky A, Kahneman D. Judgement under uncertainty: Heuristics and
J Econ 1983;85(2):147 ± 57. biases. Science 1974;185:1124 ± 31.
Berge P, Pomeau Y, Vidal C. Order Within Chaos. New York: Wiley, 1984. Van Ackere A, Larsen ER, Morecroft JDW. Systems thinking and business
Byers JD, Peel DA. Linear and non-linear models of economic time process redesign: an application to the beer game. Eur Manage J
series: an introduction with applications to industrial economics. In: 1993;11(4):412 ± 23.