JOURNAL OF COMPUTING, VOLUME 2, ISSUE 1, JANUARY 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 79

Chaos Models in Economics

Sorin Vlad, Paul Pascu and Nicolae Morariu

Abstract—The paper discusses the main ideas of the chaos theory and presents mainly the importance of the nonlinearities in
the mathematical models. Chaos and order are apparently two opposite terms. The fact that in chaos can be found a certain
precise symmetry (Feigenbaum numbers) is even more surprising. As an illustration of the ubiquity of chaos, three models
among many other existing models that have chaotic features are presented here: the nonlinear feedback profit model, one
model for the simulation of the exchange rate and one application of the chaos theory in the capital markets.

Index Terms—chaos, nonlinear systems, complex behavior, bifurcation diagram.

————————— ‹ ——————————

1 INTRODUCTION

O NE of the axioms of the modern science asserts that if an

accurate description of a physical system can be identi-
fied then the possibility of a deeper understanding of the
ministic. The deterministic chaos represents the chaotic mo-
tion of the nonlinear systems whose dynamic laws deter-
mines uniquely the evolution of the system’s state based on
system and the prediction of the system evolution is possible. the previous evolution.
These assertions are not always correct. For instance, if one The deterministic chaos represents only one particular case
applies the laws of motion stated by Newton, then there is of what is called nondeterministic chaos that exhibits a su-
possible to predict exactly the orbit of the Moon around the perexponential divergence of the trajectories. In this case the
Earth if the influence of other planets is not considered. equations describing the evolution of the system are not
These predictions were verified and proved to be accurate. If known. The both ways of chaos manifestations are short-
the third planet is included, the mathematical model of the term predictable but long term unpredictable.
interaction of the two bodies becomes “the three bodies The chaos and the concepts related to the dynamics of the
problem”, solved by Newton but for a limited set of cases systems and the their modeling using differential equations
and unsolved for the general case. Today by means of a is named the chaos theory and is tightly related with the
computer, “the tree bodies problem” can be solved, but one notion of nonlinearity [4]. The nonlinearity implies the loss
can observe that the prediction of the orbit of the third planet of the causality correlation between the perturbation and
is often impossible. effect propagated in time. The study of the nonlinearity is
A large number of real systems have a nonlinear behavior named nonlinear dynamics – a captivating domain using a
despite the idealized linear behavior used in modeling. The mathematical apparatus still under development.
development of a new way of dealing with nonlinear sys- Despite the fact that the ideas leading to the emergence of
tems is obvious. This “new way of dealing” exists already the chaos theory existed before longtime, Lorenz (1963) cre-
despite the fact that the study of the nonlinearity is still at the ated a mathematical model of the convection currents circu-
beginning. lation in atmosphere and observed that when the systems
Some changes in nonlinear systems can lead to a complex begins with initial conditions slightly changed from the pre-
and erratic behavior called chaos. The nonlinearity is one of vious ones, the results are completely different. This phe-
the conditions needed by a system in order to develop chaos. nomenon will lie at the basis of a very popular paradigm of
The term chaos is used to describe the behavior of a system chaos named “the butterfly effect”, that states that if the
that is aperiodic and apparently random. flapping of a butterfly slightly modifies the atmospherically
S. H. Strogatz defines chaos as an aperiodic long time behav- conditions in the Amazonian jungle, this fact can have an
ior developed by a deterministic system highly sensitive on impact, at the end of a complex cause – effect chain in setting
initial condition. [1] Behind this apparently random behavior off a tornado in Texas.
lies the deterministic character determined by the equations The butterfly effect paradigm contains the essence of the
describing the system. Most of the systems that are used as phenomenon characterizing the chaos: first, the sensitive
examples to explain the concepts of chaos theory are deter- dependence on initial conditions and second – the fact that
ministic. to predict the future state of a chaotic system, the current
There are two types of chaos: deterministic and nondeter- state need to be known with infinite prediction. The manifes-
tation of chaos can be found everywhere in the real world,
———————————————— for instance: the propagation of the avalanches, epidemics
• S. Vlad is with University of Suceava, 13 University Street, Suceava, spreading, climate evolution, heart beats, lasers, electronic
Romania, 720229. circuits, etc.
• P. Pascu is with University of Suceava, 13 University Street, Suceava,
Romania, 720229.
• N. Morariu is with University of Suceava, 13 University Street, Suceava,
Romania, 720229.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 1, JANUARY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 80

chaotic dynamics. There is a dispute over the manifestation of

chaotic dynamics in exchange rates. There are many studies
that are positive to the chaotic dynamics (Federici 2001,
Westerhoff, Darvas 1998, Hommes 2005, Vandrocicz 2006)
and also a number of studies that are rejecting the chaos in
exchange rate (Brooks, Serletis).
The chaos theory demonstrates that even the simplest dy-
namical systems can exhibit at some point a very complex
behavior. If the exchange rates variation is caused due to the
chaotic nature of the system, this should lead to the fact that
the smallest influences should have the effect of a nonlinearity
over the exchange rates – exactly what happens in reality.
The first model presented demonstrates the fact that even
Fig. 1. Lorentz attractor – the symbol of chaos theory.
the simplest models can exhibit chaotic behavior. [3]
The demand of foreign currency is determined as percent-
A legitimate question is that the chaos is the rule or the ex- age of the deviation of current exchange rate towards the
ception from the rule. Taking into account that most of the expected one.[2]
systems of the real world are nonlinear (the basic condition
⎛ ee ⎞
for the emergence of chaos), seems that chaos could be one S t = α⎜⎜ − 1⎟⎟, α ≥ 0 (2)
of the not so obvious features of the nature. ⎝ et ⎠
The importance of studying chaos is that chaos offers an
where
alternate method that explains the apparently random be-
et is the domestic price of the foreign currency
havior of the complex systems. The chaos plus the specific
ee is the future estimated exchange rate
mathematical tools is a framework of studying different
α is the sensitivity parameter
models from different fields, models that can be reduced to
elementary models with known chaotic behavior for some
The trade balance (Ti) is a linear function depending on the
values of the parameters.
current exchange rates and the corresponding exchange rate
The way to chaos begins with the phenomenon of period
for the last period, written as deviation from the expected val-
doubling. The period doubling evolves in 2, 4, 8, 16 and so
ues and is given by the equation:
on periods and the system evolution can abruptly fall into
The expected exchange rate represents the stable state at
chaotic regime.
which the speculators on the market do not wish to sell nor
In the case of unimodal function there is an interesting
buy.
symmetry in the parameter values for what the period dou-
bling occurs. ( ) (
Tt = β e t − e e + γ e t −1 − e e β, γ > 0 ) (3)
If A1 is the value of the control parameter for what the The clearing of the exchange markets writes as:
first period doubling occurs and An is the value for what
∆S t = Tt (4)
the nth period doubling occurs, then:
After replacing equations (2) and (1) in (4), equation (4) be-
An − An −1
δ = lim = 4.66920 (1) comes:
n →∞ An +1 − An [ ]
β e t −1 e 2t − (β + γ )e * e t −1 − γe 2t −1 − ae * e t − αe * e t −1 = 0 (5)
where δ is the Feigenbaum number valable for all unimodal The equation 5 has two roots, the positive one being consid-
functions.[5] ered for obvious reasons. The resulting nonlinear equation is:

2 NONLINEAR MODELS et =
[(β + γ )e e
*
− γe 2t −1 − ae *
t −1
+
]
2βe t −1
2.1 Chaos in exchange rates (6)
For the simulation of the volatile behavior of the exchange
rates were created models that treat the exchange rates as be-
[(β + γ )e e
*
t −1 − γe 2
t −1 − ae ]
* 2
+ 4 * βe t −1 * α * e t −1
ing prices of the financial assessments traded on efficient mar- 2βe t −1
kets. The current exchange rate contains the currently avail- for α=β=4 and γ=26.
able information and the changes observed reflect the effect of The graphical representation of the solution et show that the
the new events that are unpredictable by definition. graph presents a peak value of 2.76 and a minimum value of
The theory states that an accurate a priori prediction of the 0.091. Any other value from outside the interval represented
exchange rate evolution is impossible to be made but the sub- by these two values is attracted. The evolution of the system
sequent explanation of the changes is possible. In order to with the specified parameters is chaotic because satisfies the
eliminate these difficulties, the chaos theory and the nonlinear Ly-Yorke condition [3].
models are extensively used. The first researches have been The Figure 2 illustrates the evolution of the system for
carried out starting from 1980. two initial slightly different values: 0.2 and 0.2005 (the
In the majority of situations these models are highly nonlinear dotted line). The values of the two time series are identi-
and result in a wide range of dynamic behavior, including cal for a short period of time (the first 10 iterations) and
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 1, JANUARY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 81

0.2
0.2005
alfa=4
3 alfa=4.004
2.5
3

2 2.5

value
1.5
1.5

1 1

0.5
0.5
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 time

Fig. 2. The influence of the initial conditions. Fig. 2. The influence of the initial conditions.

then the trajectories of the systems are diverging. 2.2 The model of the nonlinear feedback mecha-
The scatterplots for the two time series are provided to nism of the profit
demonstrate the independence of the two time series after The current spending of a firm can influence the value of the
10 iterations. The scatterplots presented in Figure 3 and profit obtained at the end of the reference period. The profit
Figure 4 one of the fingerprints of chaos: the distance be- will influence the spending over the next period. The de-
tween two trajectories starting from nearby points in the pendence between the previous value of the profit and the
state space diverge over time. current value is nonlinear because an increase of the spend-
ing does not reflect in an increase of the profit. The law of the
decrease of the efficaciousness asserts that a certain mean
Cycles (Iterations) 1-10 value reaches minimum or maximum value when its magni-
tude equals the marginal value. One can invest in a certain
3
production capability but this doesn’t guarantee an unlim-
2.5
ited increase of the production but the increase up to a cer-
x(0)=0.2005

2
tain point. Beyond that point the increase of the investment
1.5
does not generates a corresponding increase of the produc-
1
tion.
0.5
The dependence between the current profit and the pre-
0
0 0.5 1 1.5 2 2.5 3
vious profit can be modeled by using the equation:
x(0)=0.2 Π t +1 = AΠ t − BΠ 2t (6)
The maximum profit Π max is supposed that it can be de-
Cycles (Iterations) 11-52 termined.
Dividing the equation (6) with Π max the following result
3
is obtained:
Π t +1 Πt ⎛ Π t ⎞ max
2.5 2
x(0)=0.2005

2 = A max − B⎜ max ⎟ Π (7)

1.5
Π max Π ⎝Π ⎠
1
Πt
Let π t = and the equation (7) becomes:
Π max
0.5

0
0 0.5 1 1.5 2 2.5 3

x(0)=0.2 π t +1 = Aπ t − Bπ t Π max (8)

A
If we take Π =
max
Fig. 1. The scatterplot for the first 10 iterations the equation above becomes the
and the scatterplot for the last 41 iterations. B
logistic equation:
When the sensitivity parameter is varied, the same effects can
be observed. Figure 4 presents the trajectories of the system π t +1 = Aπ t − Aπ 2t = A(1 − π t )π t (8)
for two very near values of α.

The apparently irrelevant changes can affect the longtime be- The logistic map exhibits the same dependence on the initial
havior of the exchange rate modeled using the Ellis model and condition: the slightest change of the initial condition causes
some of these small shocks can determine the system to fall a completely different evolution.
into the chaotic regime.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 1, JANUARY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 82

information and the current one are incorporated immedi-

phase space delayed logistic function a=2.2 ately into the price of the assets, then only a new piece of
1
information, or however else we may call it - the “novelty”,
0.9 will be able to determine a change in prices.
0.8
0.7
As novelties are, by definition, unpredictable, then the
0.6 changes in prices (or in turnover) will be unpredictable, too:
no piece of information at time t or previous to this time will
xt-1

0.5
0.4
0.3 be able to help improve the forecast concerning the prices
0.2
0.1
and turnovers (nor contribute to the decrease in the forecast
0 errors made by the agents in this respect).
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
xt
This forecast error independence towards the previous in-
formation is called the feature of orthogonality and is widely
used in testing the efficient market hypothesis.
phase space delayed logistic function a=1.8 M. Larrain elaborated a model combining a classical descrip-
tion of a Keynesian economy with a non-linear model based
0.6
on the evolution of the interest rates. Because he named the
0.5
behavioural model “Z application” and the non-linear
0.4
model “K application”, this led to the so-called K-Z model.
xt-1

0.3
Larrain introduces the two components separately.
0.2
Thus, he observes that the future interest rates in the capi-
0.1
tal markets depend both on the previous interest rates (the
0
0 0.1 0.2 0.3 0.4 0.5 0.6 technical analysis conception):
xt
rt +1 = f (rt −n ), = 0,1,2,...n (10)
where f is a non-linear function, and on a series of funda-
Fig. 5. Phase space portrait of the logistic delayed function mental economic variables (fundamentalist conception):
xt=axt-1(1-xt-2). For small values of the parameter a the non-
chaotic attractor is a point. For a larger value of the parameter rt +1 = g ( Z ) (11)
the nonchaotic attractor is a limit cycle.
Where
Z = ( y, M , P,...) , y - being the real GNP, M -
The complex behavior of the apparently simple functions the money offer, P - the consumer price index etc.
can be observed using the bifurcation diagram. The bifurca- The component (1a) shows that the future interest rate de-
tion diagram (Figure 6) is an excellent tool allowing analyz- pends on its previous rates up to a certain lag n. This de-
ing the behavior of a function by varying a control parame- pendence of the future sizes on the previous ones is specific
ter (in the case of logistic function, the control parameter is to the conception of the technical analysis of approaching the
A). capital markets.
The logistic function is known to have a chaotic behavior f (r )
The exact form of t −m is unknown, it may differ
with small isles of periodicity for a value of the parameter A
greater that 3.57. For A∈ [3.57, 4] there are small areas of from one analyst to another.
periodicity, the white stripes that can be observed in the fig- Larrain chooses for this function the expression:
ure. For A>4 the behavior is completely chaotic. rt +1 = a + brt n − crt n+1
2.3 The K-Z model of Larrain where we can notice that if c=b becomes the logistic
The theory and the models regarding the functioning of the equation, which is known to have a chaotic behaviour for
capital markets initially developed on the hypothesis accord- certain values of the control parameter b (or c):
ing to which these markets are efficient. The efficient-market rt +1 = a − brt n (1 − rt ) (12)
hypotheses comprise a series of conditions which basically
This represents the K component (application) of the model.
say that the prices of the assets and their turnover can be
In what follows we shall present the way in which is built
determined based on the supply and demand in the com-
petitive market where there are rational agents. These ra- the Z component:
tional agents quickly assimilate any piece of information that rt +1 = dyt + ePt − f ⋅ M t − g ∑ (Yt − ct ) (13)
is relevant to determining the prices of the assets and their
where d, e, f, g are constants and yt represents the real GNP,
turnover, adjusting the price in accordance with this infor-
Mt – the money offer (expressed through the aggregate Mt), Pt
mation. In other words, the agents do not have different
the - the consumer price index, Yt the real personal and ct – the
comparative advantages in acquiring information.
real personal consumption.
That is to say that such a market does not provide opportu-
This component reflects the fundamentalist conception ac-
nities to obtain a profit on an asset whose turnover is supe-
cording to which the interest rates in the capital markets de-
rior to the risk undertaken by the agent. Thus, the normal
pend on the evolution of fundamental sizes.
profits will be nil, taking into consideration the fact that the
Larrain combines the two components, K and Z, in one
agents procure this piece of information and immediately
single expression as follows:
incorporate it into the price of the assets. If the last piece of
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 1, JANUARY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 83

rt +1 = a − brt n (1 − rt ) + dyt + ePt − f ⋅ M t − g ∑ (Yt − ct ) (14) [3] R.C.  Hilborn,  “Chaos  and  Nonlinear  Dynamics“,  Oxford  Uni‐
t versity Press, 1999. 
[4] S.  Kuchta,  “Nonlinearity  and  Chaos  in  Macroeconomics  and 
This expression shows that the future interest rates are a
Financial Markets“ 
combined function of technical and fundamental factors.
[5] S.  H.  Strogatz,  “Nonlinear  dynamics  and  chaos  with  applica‐
Meanwhile, the former or latter of the two components can
tions  to  Physics,  Chemistry  and  Engineering“,  Perseus  Books, 
dominate the other one. Thus, during the stability periods, 1994. 
the capital markets are efficient and the interest rates will [6] V.  Mărăcine,  E.  Scarlat,  Aplicații  ale  teoriei  haosului  în  economie, 
depend on the Z component, to a larger extent. In the unsta- “Informatică Economică“, no.1(21), 2002. 
ble periods of the markets in question, the investors lose
their trust in fundamental variables, making decisions by Sorin Vlad graduated from "Ştefan cel Mare" University of Suceava,
extrapolating tendencies. Thus, it is the K component that Electrical Engineering Faculty, section Computer and System's Sci-
ence, 1998. He is Phd Candidate in Computer Science, field of re-
becomes dominant. search - Chaotic systems behavior modeling, University "Ştefan cel
In this situation, under certain circumstances, the c control Mare" of Suceava, since 2004. He is with the Informatics Depart-
parameter can take values in intervals for which the logistic ment, University "Ştefan cel Mare" of Suceava, Economic Sciences
and Public Administration. His research interests include: neural
equation has a chaotic behaviour, thus inducing crises and networks, expert systems, logic programming, chaos theory, chaotic
chaos episodes in the markets in question. time series analysis and prediction.
The tests made with Larrain’s model led to a series of inter-
esting conclusions. Paul Pascu graduated from "Gheorghe Asachi" University of Iasi,
Faculty of Computer Science, 1999. He is Phd Candidate in econo-
Thus, for the stable capital markets, such as the bond market mie, field of research – Cybernetics models in economy, Economics
or the security market, the obtained forecasts covered quite Bucharest Academy of Economic, since 2006. He is with the Infor-
well the evolution of the interest rates observed in reality, matics Department, University "Ştefan cel Mare" of Suceava, Eco-
which for such a model represents a success. Still, for the nomic Sciences and Public Administration. His research interests
include: cybernetic economy, economic modeling, databases, logic
estimation of the equation parameters (14), the model used programming.
techniques of linear regression, which annulled the premise
that one or another of the two components can be dominant Nicolae Morariu graduated from "Alexandru Ioan Cuza" University
in one period or another. of Iaşi, Mathematics - Mechanics Faculty, section Computing Ma-
In order to introduce such an alternation of the dominance chines, 1972. He obtained the PhD degree from "Ştefan cel Mare"
University of Suceava in 2004, with the thesis entitled: Contribution
of the K-Z application components it is necessary that the to the development of data and knowledge bases. His postgraduate
parameters of the function in question should be variable in activity includes: The design and implementation of the applications
time, which the model in its initial form cannot allow. and informatics systems within the Regional Electronic Computing
Center of Suceava and Informatics Services Society of Suceava
Improving such a model could reconcile the two big tenden- (1972-1993), research projects within the national research pro-
cies in the analysis of the capital markets, technical and fun- grams (1993-2002), SSI Suceava design-research manager depart-
damentalist, offering a powerful instrument of forecasting ment (1998-2002), associate professor "Stefan cel Mare" University
these markets. of Suceava, Electrical Engineering Faculty (1991-1998). Presently
he is associate professor at the Economic Sciences and Public Ad-
ministration Faculty, "Stefan cel Mare" University of Suceava. His
research interests include databases: FoxPro, Access, Oracle ad-
3 CONCLUSION ministration and SQL programming, Deductive databases. Artificial
Chaos is can be found almost everywhere in the nature. intelligence: expert systems, pattern recognition, neural networks,
vegetal infogenetics.
Chaos theory and fractals are currently applied in the
study of the natural phenomenon.
An essential condition needed in order that chaos to
emerge is to have nonlinear systems. In fact very few of all
models are purely linear, the vast majority of the systems
are nonlinear.
The paper emphasizes two of the features of the chaotic
systems: dependence to initial conditions and the diver-
gence of nearby trajectories.
The chaos theory has a significant impact on economy and
especially on capital markets. If the behavior of one eco-
nomic system is proved to be chaotic this guarantees that,
using appropriate methods, a short-term prediction can be
made.

REFERENCES
[1] E. Peters, “Chaos and Order in the Capital Markets“, New York: 
John Wiley & Sons, 1996. 
[2] J. Ellis, Non‐linearities and chaos in exchange rates, in “Chaos and 
Non‐Linear  Models  in  Economics:  Theory  and  Applications“, 
pp.187‐195, Edward Elgar Publishing, 1994.