35

Alison Butler

Al/eon Butler is an economist at the Federal Reserve Bank of

St. Louis. Lora Ho/man provided research assistance. The
author would like to thank William Barnett and Jennifer E/lis
for helpful comments and suggestions.

A MethodologicalApproach to
Chaos: Are Economists Missing
the Point?

“A very slight cause which escapes our notice determines a considerable

effect which we cannot fail to see, and then we say that this effect is
due to chance.”
—Poincar~

HERE IS INCREASING interest among econ- other is a chaotic variable, whose value can be
omists in a new field of study that may offer predicted with certainty. Even the most prac-
an alternative explanation for the seemingly ticed observer, however, would have difficulty
random behavior of many economic variables. determining which of these series, if any, is not
This research, which originated in the physical random. As a result, most economists would
and biological sciences, concerns a phenomenon model or estimate both time series as random
called deterministic chaos.’ processes. The chaotic series is described by a
very simple deterministic equation and identified
Contrary to the common usage of the word, later in this paper.
chaos in this context describes the behavior of a
variable over time which appears to follow no Often, behavior that cannot be explained by
apparent pattern but in fact is completely deter- standard theories and modeling techniques is at-
ministic, that is, each value of the variable over tributed to random forces, even when there is
time can be predicted exactly. In fact, one no theoretical reason to do so. This paper argues
“chaologist” describes chaos as lawless
“... that economists are perhaps not using the ap-
behavior governed entirely by law.” propriate types of models and empirical techni-
ques to explain the behavior of some economic
To demonstrate the difficulty in determining
variables and that the choice of methodology
whether a variable is random or chaotic, figures
needs to be more closely examined.
la and lb show two time series of a variable;
one series is a random variable, whose actual The study of chaos is a recent phenomenon in
value cannot be known with certainty, and the the biological and physical sciences and is just

1 2
The terms “deterministic chaos” and “chaos” are used in- Stewart (1989), page 17.
terchangeably here, although deterministic chaos is the
more precise description.
 35

now beginning to be applied to economics. Un- change over time and are related in a propor-
fortunately, many of the empirical tests for tionate manner. Consider, for example, the
chaos are imprecise and, because of mathemati- following simple production function, which has
cal constraints, the theoretical models used to only labor as an input:
generate chaos are generally limited to systems
with only one or two explanatory variables. (l)Y=AN A>0,
Both of these factors restrict the usefulness of where Y is output, which is completely consumed
applying chaos to economic systems. Neverthe- by workers (there is no saving or investment), N
less, the theory of deterministic chaos has at- is labor employed and A is the productivity
tracted a great deal of attention, both in the parameter. This equation states that output is
popular press and in academic circles. The dis- positively related to the amount of labor em-
cussion that follows attempts to clarify some of ployed. Given the value of A and the labor sup-
the issues and suggests some ways to incorporate ply, the exact value of output can be determined.
chaos into economics.
This type of model is highly restrictive; any
This article first reviews how economic vari-
change in labor changes output by a constant
ables typically are modeled by describing and
percentage. Hence, the production function ex-
evaluating several techniques of economic model-
hibits constant returns to scale.
ing using a simple model of output and popula-
tion growth.3 Next, chaos is defined and its pro- Allowing the model to be nonlinear (that is,
perties demonstrated. The advantages and pit- not necessarily proportionate) provides a more
falls of applying the theories of chaos to eco- general model in which equation 1 is a special
nomics are then discussed and illustrated, case. An example of a nonlinear production
ECONOMIC MODELING function is given by:

There are many different ways to build (2) Y = AN° A > 0, a> 0.
economic models. Four such possibilities are ex- If a = 1, this model is identical to the one
amined here for the case in which all variables shown in equation I By not restricting a to
-

are completely deterministic.~The types of mod- equal one, however, this model can be used to
els examined here are static linear, static non- examine the case in which output can vary
linear, dynamic linear and dynamic nonlinear. A disproportionately with respect to changes in
further distinction, which proves to be signifi- labor. This is illustrated in figure 2, which
cant, is also drawn between discrete and contin- shows the relationship between output and labor
uous time dynamic models. A simple model of for different values of a (for simplicity, A = 1).
output, where labor is the only input, is used to Notice that if a is between zero and one, the
illustrate each approach to modeling as well as production function exhibits decreasing returns
the restrictiveness of many common modeling to scale (that is, output increases less than pro-
techniques. In addition, focusing on economic
portionately with respect to a change in labor);
modeling allows us to show that chaotic dynam-
if a is greater than one, production is character-
ics can only arise in certain types of models
ized by increasing returns to scale (output in-
that have often been excluded, a priori, by
economists. creases more than proportionately with respect
to an increase in labor). Empirical tests of actual
Static MAdeIs production relationships can be performed to
The simplest type of economic model is a determine if a is actually different from or
static linear model, in which variables do not equal to one.

‘There is a growing body of theoretical literature incor- ness used in econometric models into the theoretical litera-
porating chaos into many different types of economic ture. Recent papers also look at the properties of chaos in
models. These models include Benhabib and Day (1981), the presence of a random component see, for example,
Deneckere and Pelikan (1986), Grandmont (1985), De Kelsey (1988)]. For simplicity, this paper focuses only on
Grauwe and Vansanten (1990), Kelsey (1988), Day and purely deterministic systems.
Shafer (1985) and Stutzer (1980). For surveys of the
theoretical literature, see Kelsey (1988) and Baumol and
Benhabib (1989).
‘There is also a burgeoning field in stochastic (random)
modeling, which incorporates the assumption of random-
 39

where C and U are constants, and a dot over a

variable means the change in the variable with
Figure 2 respect to a very smali change in time. This
Linear vs. Nonlinear Production type of equation is called a differential equation.
Functions Equation 3 states that the percentage rate of
y=ANa change of the labor force IN(t)/N(t)I, where time
is continuously changing, equals the difference
V
3 between the rate of birth, C, and the rate of
death, given by DN(t)IY(t), where N(t)IY(t) is the
number of individuals who have to subsist on
each good at time t.
Using the linear production function given in
2 equation I and substituting it into equation 3
provides a linear specification of the percentage
change in the population:

(4) N(t)IN(t) = C—D/A.

Notice that when the production function is
linear the rate of death, DfA, is constant.
Solving equation 4 yields the following solu-
tion for the population:

0 (5) N(t) = Ke(C-DIAh,

0 I 2 3
N where K is an arbitrary constant.6
This solution has the property that, unless the
rate of birth (C) is ezcactly equal to the rate of
tinear Dv.nanne MAdeis death (D/A)—in which case the population will
equal K—the population will either rise exponen-
One disadvantage of these static models is that
they can be used to describe the relationship tially or fali to zero. This result is highly restric-
tive, however, because the likelihood of either
between output and employment only if the
labor force or population remains constant over the two rates being identical or the population
time.’ Suppose instead that we want to examine increasing infinitely is, in reality, very small. In
other words if C U/A, the system is unstable.’
the behavior of output over time as it is related
to a continuously changing labor force. A stan- Unfortunately, in models of other types of eco-
nomic variables, results that greatly restrict the
dard equation borrowed from Haavelmo (1954),
used to describe the growth of the labor force possible values of the parameters of the models
where that growth is dependent on the level of are not uncommon. In addition, because of the
output, is given by: complexity of many economic models, the im-
plications of restricting the value of the parame-
(3) N(t)IN(t) = C —DN(t)IY(t) C > 0, D > 0, ters to determine the solution or to ensure a

‘For expositional ease, the terms “population” and “labor have three possible cases: stable converging dynamics
force” are used interchangeably. (such as when C = D/A in the model above), unstable
dynamics (when C * D/A) and cyclical dynamics, which is
‘Equation 4 is solved by the variable separable method of
solving differential equations found in most calculus books. the least common of the three. In nonlinear models, how-
K is the constant of integration, which can be determined ever, cyclical dynamics are far more common, and ex-
by choosing an initial condition. ploding dynamics may not occur. Thus, it is also important
to consider the desirable and realistic stability properties
‘In fact, stability is an important issue which is frequently when choosing a model. Obviously the nonlinear case is
ignored or abstracted from in economics. Stability is impor- more general and the most realistic for variables that ex-
tant because, for example, an unstable equilibrium is not a hibit cyclical variation. For the purpose of this paper,
sustainable equilibrium. Stability is also important in the however, the issue of stability is ignored.
choice between linear and nonlinear models. Linear models
stable solution are not always as obvious as in dynamic models are called difference equations;
the population growth model. Because linear dif- they measure time in distinct intervals rather
ferential equations are far simpler to solve than than the d4’ferential equations used above, which
nonlinear differential equations, and because measure time continuously. Equation 6 can be
their solutions are more often stable and easier transformed into a difference equation by let-
to interpret, however, they are used in econom- ting the rate of change of N (previously given
ic models more often than may be appropriate. by N) equal the difference between the value of
N at time t and t + I. Thus, equation 6 becomes
\imIfr,p~r~ IJvnam.ie Models (7) (N~÷,
—N~)IN = C — D(N,IY,),
1
Combining the nonlinear production function where Y, = AN:.
given by equation 2 with the description of Combining these equations and simplifying the
population growth given in equation 3 provides result yields:
a less restrictive model of population growth:
(8) N,~, = N, [(1+C) — DN,-~/A]
(6) N(t)IN(t) = C — DN(t)’~fA. which, following Stutzer (t980), can be rewritten
Unlike equation 4, equation 6 allows the labor using a change of variables as
force to vary more or less as the current labor (9) X~.÷,
= k x,(1—x~’°),
force changes. Unfortunately, the price of the
generality provided by such nonlinear differen- where k = 1 + C.9
tial equations is that most either cannot be solved The models shown in equations 8 and 9 de-
or have solutions so complex the results cannot scribe the most general specification of popula-
be interpreted. Not surprisingly, economists tion and output growth given the assumptions
often avoid these types of models. made above. Behavior is not restricted to being
linear, nor is population or output restricted to
The model used here, however, was chosen for remaining constant over time. On the other
its tractability and can be solved for the value of hand, as noted earlier, these more general
labor at any time t.’ All that is necessary for a models often cannot be solved or have solutions
stable solution is that the production function ex- without any economic interpretation. Neverthe-
hibits decreasing returns to scale (0 < a -C 1). less, unless there are theoretical reasons for
Regardless of the value of the other parameters, assuming relationships are static or linear,
if a is between zero and one, the population dynamic nonlinear models, which provide the
will reach a stable equilibrium level. Hence, in most general specification of behavior, should at
contrast to the dynamic linear model discussed least be considered in economic analysis. Al-
previously, the results of this model are more though generality for its own sake is not a de-
realistic and provide a more general description sirable goal, using a more general model would
of population and output growth. be appropriate when simpler models have solu-
tions that are highly unrealistic or when param-
.th.serete .Modets eters have to be restricted beyond reason (as in
the model presented here). In addition, if eco-
One problem with using continuous time nomists want to test their models and results
models in economics is that data are available empirically, then these variables should be
only in distinct intervals (daily, weekly, monthly, modeled in the form in which they are esti-
etc.). One approach typically taken by econom- mated—discrete form.b0 As it turns out, these
ists, therefore, is to convert these continuous types of nonlinear dynamic models can exhibit
time models into discrete time models. Discrete chaos.

aFor the solution and discussion of this model, see since in discrete time this model can generate chaotic
Haavelmo (1954), pp. 24-29. dynamics for certain parameter values. Differential equa-
9 tions can also exhibit chaos, although only in more com-
Allowing N, [A(1 +C)ID]”-” X, only changes the scale
of the population and has no effect on the general charac- plicated models. This is discussed in greater detail later.
teristics of the solution. For further discussion of this pro-
cedure and the solution, see Stutzer (1980).
‘°Althoughthe model given by equation 6 is stable in con-
tinuous time, it is not necessarily stable in discrete time
 IN~TRO]flLJf71Tit) IT) THE
Jfl;))fl)( OF (I~.H.r4.OS Figure 3
The Logistic Growth Curve For
The possibility that chaos exists in economic
variables has strong implications for the way in
Various Values of k
which economics is modeled. For example, some = kX~(1 — Xt)

variables that appear to be random processes,

like one of the variables shown in figure 1,
might in fact not be random at all; instead it
might be completely explained using the ap-
propriate deterministic model. This section
demonstrates the properties of chaos, using a
simple model.
In the most general sense, the term chaos is
used to describe the behavior of a variable over
time that appears random but, in fact, is deter-
ministic; more precisely, given the initial value
of the variable, all future values of the variable
can be calculated with exact precision.” In con-
trast, the value of a random variable can never
be predicted with certainty.
More formally, a function is chaotic if, for
certain parameter values, the following two con-
ditions hold: First, the function never reaches
0.0 0.5 1.0
the same point twice under any defined interval
of time. In this case, the function is said to ex-
I xt

hibit aperiodic behavior. Second, the time path

is sensitive to changes in the initial condition, so simplest model that exhibits chaos and provides
that a small change in the value of the initial reasonably interpretable graphical results. This
condition will greatly alter the time path of the equation is given by:
function.”
(10) X,~, = k X~(I—X,),0 < X, < 1,0 -C k < 4.
Chaos only arises in certain types of nonlinear
dynamic systems, although not all nonlinear dy- Equation 10 describes the time path of a vari-
namic equations are chaotic. Moreover, equa- able, X (which for expositional purposes is called
tions that can be characterized as chaotic need a population), that is a function of its previous
not exhibit chaos for all parameter values. Rath- value and a parameter k. To demonstrate chaotic
er, functions that can exhibit chaos will do so behavior in this simple framework, the value of
only for certain parameter values. This is ex- X can only take on values between zero and one.
plained by example below. The value of k, the only parameter in the equa-
And j3iow for Soniethint; tion, is called the ‘tuning” parameter; it deter-
E..7Ainphi.etv tEll rent mines the steepness of the function. Figure 3
shows the function given in equation 10 for
The properties of chaos can be demonstrated various values of k. Increases in the popula-
using a simple mathematical equation, called the tion below X increase future values of X more
logistic growth equation. While this model has than proportionately. Past this point, the popula-
no particular economic interpretation, it is the tion begins to decrease.” For larger values of k,

“For simplicity. only single-variable equations are discussed. tions, see Li and Yorke (1975), Brock and Dechert (1988)
Although chaos exists in multivariate economic systems, and Melese and Transue (1986). For a good mathematical
tests for chaos in these systems are just beginning to be description of chaos and the mathematical tools used in
developed, and the mathematics of such systems are ex- the theory of chaos, see Devaney (1989).
tremely complex. “This behavior is similar to that of a total product curve
12
Thei-e are many different characterizations of deterministic where, once the marginal product becomes negative, fur-
chaos, but they all include the one used here. For more ther increases in an input decreases output.
rigorous definitions and discussion of the different defini-
 42

Figure 4
A Stable Time Path for a
Logistic Growth Curve
X~÷=1 3Xt(1 Xt) —

X0 = .20 t=1 to 500

xt+l
1.0

S
0:5

0.0
0.0 Xo X0.5 X~X~ 1.0
xt

the absolute value of the rate of change of X is which is required for a steady-state equilibrium.
larger. In this example the initial value (when t = 0) is
.20. To determine the value of X1, draw a line
For certain values of the tuning parameter
between the initial value (X0) and the parabola
(k 3), the system is stable; this means the
(line segment X0B). To find the value of X,, set
population will reach some sustainable steady-
= X by drawing a line from point B to the
state value which differs from X.
45-degree line (point C) Then draw a straight
-

Figure 4 illustrates how the time path for X,÷, line from point C to the parabola. This is the
is solved graphically. The parabola represents value of X, (point D). This process, called itera-
equation 10 when k is equal to three; all values tion, can be used to determine as many subse-
of X, and X,~ must lie on this curve. The 45- quent values of X as is desired, once the initial
degree line depicts the points where X,~1 = X,, value is determined.14 As we can see in figure 4,

14
Notice that X,~,,which must always lie on the parabola,
can be either above the 45-degree line (as in point D) or
below it (as in point F). For precision, the equation is
solved numerically and then graphed.
 43

the population appears to be converging to a

steady-state equilibrium value at 2/3 (X).
Figure 5
If the value of k increases past three, how- A Logistic Growth Curve
ever, the equilibrium point becomes unstable
and the time path exhibits a two-period cycle, Exhibiting Chaos
where the variable alternates between two =
X~÷1 &82840 Xt( 1 Xt) —

values. Further increases in k produce a four- X0 =

.0101 t=1 to 500
period cycle (that is, the time path repeats the
same sequence of numbers every fifth iteration),
then an eight-period cycle, and so on, with the
1.oj / all
periodicity increasing by z~(n = 1,2,3, .). If k
.

increases past a certain point called the “point r ‘“““St

I 1St
of accumulation” (for this function, it occurs at
k = 3.5700), the time path enters into a region
in which the function can exhibit chaos,” In the
chaotic region (3.57 < k < 4 for this function),
there can be both an infinite number of periodic 05
cycles and an infinite number of initial condi-
tions that produce an aperiodic time path.16 Us-
ing this simple example, we can demonstrate
some of the properties of chaos in graphical
form.

Properties of Chaos
IC
o.o~~2t—
An example of aperiodic behavior is seen in
figure 5. The first 500 iterations are shown in 0.5 I .0
xt
this figure (that is, t = 1, 2, - . - 500), and no
single point is ever reached twice.17 In fact, no
matter how many times this equation is iterated, initial condition can affect the time path is
never has the same value twice.18 If the data shown in figure 6. In this figure, the values of
are plotted as a time series, it would look similar
X are plotted against time, as in figure 1. This
to figure la, the chaotic series in figure 1, and diagram demonstrates how changing the initial
one might conclude that the data are generated value, X,,, at the fourth decimal place (from
by a random process, such as figure lb. because
.0101 to .0100) causes the time paths generated
they follow no obvious pattern. This is not the by equation 10 to deviate substantially from
case here; the data in figure la and figure 5 each other.” Although not all sections of the
were generated from models without a random time path differ as dramatically as the one
component and therefore are completely shown here, figure 5 graphically demonstrates
deterministic. that the choice of an initial condition or, for
The other characteristic of a chaotic function forecasting purposes, the choice of a time inter-
is that its time path is sensitive to the choice of val (that is, determining where to start the sam-
initial values. An example of how changing the ple), can greatly alter the results. In fact, despite

“This process of increasingly complex periodicity is called icity is required for chaos, tests for chaos in actual data
bifurcation and is discussed in most papers on chaos. For take a different approach, thus avoiding the problem.
a nontechnical discussion of bifurcation, see Gleick (1987) “The time path can avoid having repeat values because the
and Stewart (1989). For a more analytical treatment of number of possible points between zero and one is infinite.
bifurcation, see May (1976) and Baumol and Benhabib
(1989). A more rigorous discussion of the relationship be- “Although it sometimes looks like the function is periodic,
tween bifurcation and chaos is given in Li and Yorke this appearance is a result of the lack of precision of the
(1975). printer and the scale of the graph. In fact, there are no
“Notice that not every initial condition gives rise to an periodic points in this function.
aperiodic time path.
I7This property is unlikely to be found in actual data, how-
ever, because of rounding. Although theoretically, aperiod-
 44

Figure 6
A Segment of the Time Trend Showing Sensitivity
to Initial Conditions
X~+i=3.82840Xt(1 — Xt)

X0=.01o1 X0=AtiOO
xt
1.0

0.8

0.6

0.4

0.2

0.0
600 610 620 630 640 650
I:

seemingly trivial differences in the initial condi- Sensitivity to changes in the parameter values
tions, the time path produced by one initial value is illustrated in figure 7. Here, a fifth-order
will not necessarily be similar to the time path change in the value of the tuning parameter
generated by a marginally different initial value. (from 3.82840 to 3.82844) produces not only a
In general, the two time paths that arise from the substantially different time path from the one in
different initial values wifi have periods during figure 5, but also one that exhibits periodic
which they are arbitrarily close together and rather than chaotic behavior.20
periods during which they deviate substantially.
Are Attractors Strange?
Chaotic functions also exhibit sensitivity to
very small changes in the parameter values. A Another feature often found in chaos,
third. or fourth-order change in the value of a although neither necessary or sufficient for
parameter can alter the time path from stable chaos, is a strange attractor.2’ The properties of
to chaotic or vice versa. attractors and strange attractors are best illus-

20 21
flecall that when a function is in a chaotic region (that is, The only examples of chaos without the presence of a
when the parameters are such that the function can ex- strange attractor are found in certain types of dissipative
hibit chaos), there can be both periodic and aperiodic time systems.
paths.
 43

Figure 7 Figure 8
A Logistic Growth Curve With The Strange Attractor for a
Periodic Points Chaotic Function
Xt.j=3.82844Xt(1 — Xt)
.0101 t=1 to
= 500 = 3.82840X~(1 — Xt)
xt+1
X0 = .0101 t=1 to 1500
1.0 xt.I
1.0
/
/
/ N
/
/ .

II //

0.5
0-s

0.0
0.0 0.5 1 .0 0.0
xt 0.0 0.5 1.0

XI

trated by example. In a stable system, the time

path converges to an equilibrium point (for ex-
therefore never reaches an equilibrium in the
ample, X’ in figure 4). The equilibrium point is
standard sense, it also never leaves the strange
also called the attractor, because the time path
attractor and therefore is not unstable (for ex-
is “attracted” to the equilibrium point. Another ample, never goes to positive or negative infini-
possibility is that the time path has two attract-
ty). An example of this is shown in figure 8,
ors, and the system oscillates between them, which takes the same numerical example as in
never remaining at one equilibrium point. This
figure 5, but iterates it 1500 rather than 500
is found in predator/prey population models,
times. In this picture, the values of X are still
where the population grows until it is so large contained in the same area as in figure 5, but
it begins to die off and then shrinks to a level the distribution of points is becoming denser.
so small it begins to grow again.
The bounded region (shown by the dotted line
A “strange attractor” is the name given to the in figure 8) is the strange attractor for this
case where there is a region, rather than a function, If the function is iterated further, the
finite set of points, that attracts the time path area within the bounded region would appear
of the variable. That is, after some number of to be a solid block, although the function would
iterations, which varies depending on the func- never have the same value twice. In fact, the
tion, the time path of the variable is completely existence of a strange attractor is an important
contained in this region (the strange attractor). way to distinguish between a random and
Thus, even though the path is aperiodic and chaotic time path.22

22
Another definition of a strange attractor is an attractor with they are chaotic. For the purpose of this paper, however,
fractal dimension. In fact, if a strange attractor exists, the the issue of fractals and fractal dimension will be ignored.
variable has fractal dimension. Random variables have in- For a discussion of these topics, see Mandelbrot (1983)
finite dimension, however. As a result, tests for dimension and Gleick (1987).
are one of the main ways data are tested to determine if
LESSONS •EROM P11505 scripts and an error term. Although these sim-
pler models may be more likely to have solutions
Although economists are beginning to incor- with explicit results that can be tested empirical-
porate chaos into their economic and econo- ly, the dynamics that arise may not capture the
metric models, there has been little discussion behavior of the variable of interest. The richness
of the ways in which chaotic dynamics are use- of a model may be found in explaining the be-
ful or realistic for economic models. Clearly, havior of a variable over time as much as in the
chaos holds considerable appeal for economists direct, time-independent (or time-constant) rela-
who are looking for a deterministic explanation tionship between the variables.
of the apparent randomness in economic vari-
ables. Economists frequently assume randomness In addition, the study of deterministic chaos il-
when they are unable to explain the behavior of lustrates some of the pitfalls of first differen-
an economic variable empirically. The presence cing a dynamic model to convert it to discrete
of an alternative explanation, chaos, will require time, as was done in the model of population
them to consider more carefully the rationale growth presented above. This practice is com-
behind their assumptions. mon in economics because data are only
available in discrete intervals.
One problem with incorporating chaos into
economics is that, while economists can either As is shown in Stutzer (1980) and demon-
postulate an equation and test it for the presence strated here, there are first-order differential
of chaos or, alternatively, see if the data them- equation models (such as equation 6) which con-
selves are chaotic, it is especially difficult to verge to a steady-state equilibrium that are cha-
identify the correct functional form that charac- otic when expressed in discrete time (equation
terizes the data. The choice of a functional form 9). Thus, the dynamic properties of the discrete
is always a problem in economics, but, as pre- analog of a differential equation cannot be as-
viously discussed, it is particularly difficult to sumed to be the same. In fact, it has been
model nonlinear dynamics. This problem is ex- shown that, although chaos can arise in first-
acerbated because, as a result of the mathemat- order difference equations, it can only arise in
ics required, the study of nonlinear dynamics third-order or higher differential equations.’~As
has, until recently, been relatively limited in a result, an economist must be careful about
general and largely ignored in economics.23 either converting a continuous time dynamic
Even when it is possible to estimate nonlinear model into a discrete model (such as converting
dynamic equations, the models themselves often equation 6 into equation 7), or taking a static
cannot be solved analytically. Without explicit model and simply adding a time subscript,
solutions to these models, their usefulness is ex- rather than postulating a model that is dynamic
tremely limited. Obviously, the difficulty of de- (in either discrete or continuous time) and esti-
termining the “true” underlying model from a mating or solving it in that form. The choice of
data series is a problem whether or not chaos the appropriate type of model should depend on
exists. The “discovery” of chaos, however, has the economic variables being described rather
focused much more attention on this problem, than analytical convenience. This issue is partic-
especially if the data are nonlinear. ularly important if a continuous-time dynamic
model is estimated in discrete time using the
Econonn.c Modeling an.d Chaos steady-state equilibrium properties of the
continuous-time solution. The discrete-time equa-
The study of chaos emphasizes the impor- tion that is being estimated may not reach a
tance of rigorously modeling the dynamics of a steady state at all, or the solution could differ
system rather than merely taking a static model qualitatively from that found in the continuous-
(like equations I and 2) and adding time sub- time version of the model.

“For recent work in nonlinear dynamics, see Grandmont

(1987).
24
The “order” of an equation refers, for a differential equa-
tion, to the highest power attained by the derivative and,
for a difference equation, the highest degree of differen-
cing. For a more complete discussion, see Chiang (1984).
 Empirical Applications of Chaos in Economics
1 here are generally t~to approaches used in For- these tests to be accurate, the data need
the empirical literature to test for the pres- to be especially exact. This degree ot preci-
ence of deternunistic chaos in economic and sion presents a particular problem for cc-a-
financial data ‘l’lie first approach tests for nomics, whei-e controlled experiments are
the presence of nonlinearities in the data essentially impossible, especially on the macro
Since chaos only arises in nonlinear systems, level. Data collection is far from perfect, and
finding nonlinearities in the (laid suggests the quality of the data declines as the degree
that lest ing cli ccci ly for’ the presence of chaos of aggregation increases introducing measure-
is appropriate. In addition. the presence of ment error in the data In addition, because
nonlinear ities in the data pi-ovicles inforina- of rounding, the data are not as precise as
Hon to theorists niodeling these types of they should be. For this reason, tests for chaos
economic systems. Because testing for non- are not simply tests for aperiodicity.
linearities in the dat a is in u cli simpler (arid
less controversial) than testing for chaos, ‘lhe (]uarititv of high-quality data is also cx-
these tests are often performed firsl - tremnely important. Even if the results sho½
aperiodic its’ for a sample of 100 ohsc-,rvations,
Many macroeconomic time series have been the system need not be aperiodic The existing
found to beha’~e in a nonlinear manner. Brock empirical tests br the presence of chaos re-
and Sayers (1988) find such n idence in data quire an ext rernel~large number of highly
for quartet ly eniplovment (1930-83), quarterly accurate data. Barely are both of these avail-
unemployment (1949-82). monthly post-war able to econometricians. As a result, any cvi-
industrial production arid pig-i on pr-oduc’tion dence from tests for chaos should he % iewed
(1877-] 937). Nonlinearities have also been with caution
found in the Dit isia Ml monetary aggregatesA
Other studit’s ha~c’ found nonlI riea cit ic’s ~ Ci ten these cat eats, sortie statistical tests,
financial data as well. I or example, Hinich originating in the physical sciences, do look
and Patterson (1985a. 1985W find strong cvi- for the presence of chaos in economic data
clencc’ of nonlinearity in daily stock returns. ‘I hese tests are run on variables that have
- long time series available, art not aggregate
The second approach is to test directly for variables and are thus more likek to provide
he pr eseru e of chaos.’ There are ma iw pro- a ccu rat c’ restilt s. ‘lest s I tave km rid] evidenc-e
blenis with testing directly for chaos using consistent with chaos in exchange rates (Ellis,
economic data. however. ‘1 he most obvious, it 990), daily gold and silver prices on the 1 on-
arid perhaps t lie most iniport ant, is the sen- clan market (Frank and S terigos, 1 988) at id in
sitivity of chaotic systems to sniall changes in the Di~isia monetary aggregates (Bai nelt and
the par amnetcm- ~‘altm
~s and iiii I ial c-onditions. (hen, 1988) .~

- For a discussion of the tests used, see Brock and Sayers drawbacks see Barnett and Hinich (forthcoming), Brock
(1986). as well as the other papers cited above (1986) and Ramsey (1989)
4
‘For a definition and discussion of the Divrsia monetary This is by no means a comprehensive survey of the
aggregates. see Barnett and Spindt (1982). empirical literature applying chaos to economic and
‘A descr’ptton of the actual ernprricat techniques used to financial data. For a more comprehensivc discussion of
test for chaos is beyond the scope of this paper. For a the empirical work on chaos, see Barnett and Hinrch
description of the tests available for chaos and their (forthcoming) and Ramsey (1989).

iom’ecasting i rio-c’ to impossible -,incc a small

I he stuc]~-ot deterministic chaos also olters error iii the talLie ol the initial condition can
set i’m ,d les on~for c’ronomctr’ iaris. It torerast— leach to hiMhlv utac curate precut tion. (see, lot’
jog i.. a goal cit cc orioiriit’ niocleling. inappi opt iatc’ e~aiiipli’ figurc’ 6) simil,ri’lv, an error ri ant
modeling techniques in Itt1’ presence oF chaos par’iirmietem abc’ can also pm ocluce inc or rc’ct lore—
lnq-oiiu’ tnorc’ costlt - IF Itu’ data arc’ chaotic’ c,i. ts (see tigtim’c’s3aticl It ‘thus. it is itnpor tant
to realize the limitations of economic forecasts ing explicitly how restrictive the assumption of
in the presence of chaotic variables. linearity can be, the study of chaos emphasizes
the importance of allowing for the possibility of
Chaos does not have to be present in the data
nonlinear behavior. The use of chaos in econom-
to find the sort of fluctuating behavior (although
ics also has offered new explanations for behav-
without any clearly defined periodicity) that is
ior that, until recently, has been able to be ex-
often found in economic data. Nonlinear non-
plained only by random forces.
chaotic models often can generate time paths
that appear random, and testing for nonlineari- The techniques that have arisen from the
ties is the likely next step for future research in study of chaos in the physical and biological
this area, In fact, empirical economists are be- sciences are in their infancy. As these techni-
ginning to test for both nonlinearities and chaos ques become more refined, and economists be-
in economic data (see insert on page 47). As a come better trained in working with these types
result, more work needs to be done in under- of models, their ability to explain the behavior
standing nonlinear estimation so that economic of variables such as exchange rates, business
models can describe a greater variety of be- cycles and stock prices is likely to improve.
havior and be more accurate as well. Tn addi- That possibility alone is sufficient reason for
tion, the existence of chaos suggests that econo- economists to take a closer look at deterministic
mists might want to try nonlinear specifications chaos in particular and nonlinear dynamics in
of a variable before resorting to modeling it as a general.
random variable. This in turn will help to im-
prove the quality of economic forecasts in the
presence of nonlinear variables,
D..E.FE:R.]EJ~ItiES
In addition, the use of a random component Barnett, William A., and Ping Chen. “The Aggregation-
in estimation does not necessarily imply that the Theoretic Monetary Aggregates are Chaotic and have
variable itself is random, but rather that other Strange Attractors: An Econometric Application of
Mathematical Chaos?’ in William A. Barnett, Ernst R.
relevant variables might be excluded from the Berndt, and Halbert White, eds., Dynamic Econometric
regression. Although each of these other vari- Modeling, Proceedings of the Third International Sym-
ables could have a small influence on the system posium in Economic Theory and Econometrics (Cambridge
University Press, 1988), pp. 199-245~
by itself, the total effect of these excluded vari-
Barnett, William A., and Melvin J. Hinich. “Has Chaos Been
ables could be substantial. Given both the diffi- Discovered with Economic Data?” in Ping Chen and
culty in detecting what these missing variables Richard Day, eds., Evolutionary Dynamics and Nonlinear
might be and data limitations, such a complex Economics (Oxford University Press, forthcoming).
system might best be approximated by a random Barnett, William A., and Paul A. Spindt. “Divisia Monetary
variable, even if there is no true randomness in Aggregates: Compilation, Data, and Historical Behavior?’
Board of Governors of the Federal Reserve System, Staff
the variable being estimated. In fact, some argue Studies 116 (May 1982).
(see, for example, Kelsey, 1988) that, since eco- Baumol, William J., and Jess Benhabib. “Chaos: Significance,
nomic models do not include such (chaotic) phe- Mechanism, and Economic Applications?’ Journal of
nomena as weather and other biological factors Economic Perspectives (Winter 1989), pp.77-lOS.
which can influence economic variables, it Benhabib, Jess, and Richard H. Day. “Rational Choice and
“seems inevitable that we will have random Erratic Behaviour?’ Review of Economic Studies, (July 1981),
pp. 459-71.
terms in our equations.’25
Brock, W. A. “Distinguishing Random and Deterministic
Systems: Abridged Version?’ Journal of Economic Theory,
(October 1986), pp. 186-95.
(X)NCLUSI.ON
Brock, W. A., and W. D. Dechert. “Theorems on Distinguish-
The study of deterministic chaos and its subse- ing Deterministic from Random Systems?’ in William A.
Barnett, Ernst A. Berndt, and Halbert White, eds., Dynamic
quent application to economics has opened a Econometric Modeling, Proceedings of the Third Interna-
new realm of possibilities for economists trying tional Symposium in Economic Theory and Econometrics
to explain cyclical or erratic behavior in eco- (Cambridge University Press, 1988),
pp. 247-65.
nomic variables. As discussed above, chaos has
Brock, W. A., and Chera L. Sayers. “Is the Business Cycle
implications for both theoretical modeling and Characterized by Deterministic Chaos?” Journal of
empirical applications in economics. By illustrat- Monetary Economics, (July 1988), pp. 71-90.

tm
‘ Kelsey (1988), p. 12.