Chaotic behavior in an economic

model
Clara Grcio, Cristina Janurio and J. Sousa Ramos
Department of Mathematics, Universidade de vora, Rua Romo
Ramalho, 59, 7000-585 vora, Portugal. E-mail:
mgracio@uevora.pt
Department of Chemistry, Mathematics Unit, Instituto Superior de
Engenharia de Lisboa, Rua Conselheiro Emdio Navarro, 1949-014
Lisboa, Portugal. E-mail: cjanuario@deq.isel.ipl.pt
Department of Mathematics, Instituto Superior Tcnico, Av.
Rovisco Pais 1, 1049-001 Lisboa, Portugal. E-mail:
sramos@math.ist.utl.pt
Abstract
The purpose of this work is to study a discrete-time nonlinear
business cycle model of the Kaldor-type. The model is an
extended Kaldor model and it is described by a two-dimensional
dynamical system with income and capital as variables. We
check the orbits of the system, their changes related to changes
of the system parameters and their basins of attraction in order
to understand the dynamic features of the model.
1. Introduction
The complexity of most important and interesting economic
phenomena can only be explained by the use of nonlinear
models. In favor of linearity and the main reason for its
2 Clara Grcio, Cristina Janurio and J. Sousa Ramos
adoption is its simplicity. But simplicity is sometimes not
enough, particularly in the case of macrodynamics and
economic fluctuations. As a result, many economists have
insisted that non-linear structures should be employed
instead. It is impossible to ignore that non-linear dynamical
structures are clearly the more general and common case
and restricting attention to linear structures limits the type
of dynamics that are possible.
In what follows a discrete-time economic model is
considered. It is a particular case of the Kaldor-type
business cycle model and it is described by a two-
dimensional dynamical system. In this work, it is suggested
an extension of the model proposed by Bischi, Dieci,
Rodano, Saltari (2001) and Dieci, Bischi and Gardini (2004).
We can find the original Kaldors model in Kaldor (1940) as
an elementary example of the limit cycle in macroeconomic
models and as one of the early non-linear models.
Consider the following discrete-time version of the Kaldor
model:

'

+
+

+
t
I
t
)K (1
1 t
K
)
t
S
t
(I
t
Y
1 t
Y
,
(1)
where the variables Y, K, I and S represent, respectively,
the income, the capital stock, the investment and savings.
We have that the change of income, between two
consecutive instants, is proportional to the difference
between in investment and savings and (>0) is the
parameter of proportionality. The capital stock in the next
period is equal to new investment plus what remains after
depreciation. The parameter, , represents the capital
stock depreciation rate (0<<1).
The investment function is assumed to take the form of
an increasing arc tangent type function of income, like
proposed in Bischi, Dieci, Rodano, Saltari (2001) and Dieci,
Bischi and Gardini (2004). Since Kaldor concluded, it might
be sensible to assume that the S and I curves are non-
linear, in general, he assumed I = I(Y, K) and S = S(Y, K),
where investment and savings are non-linear functions of
Chaotic behavior in an economic model 3
income and capital. The introduction of nonlinearity to the
capital stock variable is done, considering it a decreasing
trigonometric function of the sin type (Fig. 1.), instead of
being just a linear decreasing function. Therefore,
( ) ( )
t
0.2K sin
t
Y arctan
t
K

1

t
I + +

,
_

,
(2)
where the parameters , , ,
1
, , and are such that 0<, <1
and ,
1
, , >0.
K
I
Fig. 1. Effect of the nonlinearity in the capital stock, K.
Concerning savings, they are assumed to be nonlinear.
So, it depends also on the capital stock in a decreasing
way:
( )
t
0.2K sin
t
K

t
Y
t
S +

,
_

,
(3)
with
2
a positive parameter (
2
>0).
Considering the particular case, when
1
=
2
= , for
convenience, and replacing expressions (2) and (3) in (1),
we get the following two-dimensional map in income and
capital stock variables:
4 Clara Grcio, Cristina Janurio and J. Sousa Ramos
( )
( ) ( )

'

,
_

+ + +
+
+ +
+

t
0.2K sin
t
Y arctan
t
K

t
)K (1
1 t
K
t
Y
t
Y arctan
t
Y
1 t
Y
(4)
2. The model and some considerations about
two-dimensional maps
Let us change the notation of the model (4) to y:=Y and
k:=K.
Consider the family of two-dimensional nonlinear maps,
2
R
2
R : F
, given by
( )
( ) ( )

'

,
_

,
_

+ + +
+ +

0.2k sin y arctan k

)k (1
y y arctan y
k
y
F
(5)
where, , , , , and are real parameters such that ,
, , >0 and 0<, <1.
The map is continuously differentiable and has the
structure of a triangular map (skew map) since it has the
form F(y, k) = (f(y), g(y, k)), that is, the variable y does not
depend on k. From the economic point of view, that means
that the dynamic of income is only affected by income
Chaotic behavior in an economic model 5
itself. The first component of the map is called the basis
map and the second is the fiber map. In Fig. 2. it is shown
the graphical representation of the map F for some values
of the parameters.
98.5 99 99.5 100 100.5 101 101.5
y
237.5
238
238.5
239
239.5
240
k
a
96 98 100 102 104
y
115
120
125
130
135
k
b

Fig. 2. Graphical representation of the map F for some values of
the parameters. In plot a) =8.9, =100, =0.3, =0.3, =9.5,
=1.0 and in plot b) =10.0, =100, =0.5, =0.4, =9.5,
=0.8.
As a consequence of the triangular structure, the dynamics
of the map F is deeply influenced by the dynamics of the
one-dimensional basis map, f(y). In particular, many of its
bifurcations are associated to those of the one-dimensional.
The map f is a bimodal map when >1.
90 100 110 120 130
y
90
100
110
120
130
fy
Fig. 3. Graphical representation of the map f when >1.
In this case the turning points are given by
1
- 1

m
c and 1
- 1

M
c + .
(7)
C
m
represents the relative minimum and c
M
the relative
maximum.
6 Clara Grcio, Cristina Janurio and J. Sousa Ramos
To begin with, we present the bifurcation diagrams of the
map f, when the parameters and change:
10 12 14 16 18

90
100
110
y
0.2 0.4 0.6 0.8 1

95
100
105
y
Fig. 3. Bifurcation diagram of the map f as a function of (plot
a)) and as a function of (in plot b)), for =100, =0.2 with the
initial conditions c
m
and c
M
.
The bifurcation diagrams above shows the dynamic
properties of the map f. The long term behavior of the
income variable, y, is deeply influenced by both the
parameters.
To study the dynamic behavior of the map F, first it is
important to make some considerations about two-
dimensional maps.
Let us first define critical curve of rank -1, LC
-1
. When a
map is continuous and continuously differentiable, then LC
-1
is included in the set of points at which de jacobian
determinant vanishes, that is,
{ } 0 ) , ( : ) , ( 1 k y J k y LC
,
(6)
where J denotes the jacobian matrix and the symbol, |.|,
the determinant.
The curve LC
-1
is the two-dimensional analogue of the
points of local extrema of a differentiable one-dimensional
map. Therefore, the iteration of the critical lines can give us
important information about the dynamic features of the
model. The curve LC satisfies F
-1
(LC) LC
-1
and T(LC
-1
)=LC.
The rank k image of LC is a critical curve of rank k+1,
LC
k
=F
k
(LC). For more detailed information see Mira,
Gardini, Barugola, Cathala (1996).
Chaotic behavior in an economic model 7
In this case, the jacobian matrix has the form
1
1
1
1
]
1

+
+

) k cos(0.2 0.2 1
2
) (y 1
1
0
2
) (y 1

1
k) J(y,
(6)
and the roots of its determinant are given by:
1
- 1

1,2
c t and
0.2

0.2
1
Arccos
3
c
+

,
_

.
(7)
Obviously, the two first zeros correspond to the solutions
of the equation f(y)=0 (local extrema of the basis map).
From the geometric point of view, c
1
and c
2
correspond to
vertical straight lines and c
3
symbolizes the infinite
horizontal straight lines (see Fig. 3.).
60 80 100 120 140
y 60
70
80
90
100
k
8 Clara Grcio, Cristina Janurio and J. Sousa Ramos
Fig. 3. Some critical curves for the values of the parameters:
=10.0, =100, =0.2, =0.4, =9.5, =0.8.
The first iteration of LC-1 gives origin to the following
pictures:

95 100 105 110
y
30
40
50
60
70
k
a
80 90 100 110 120 130
y
-38
-36
-34
-32
-30
k
b
Fig. 4. The result of the first iteration of the critical lines. In plot
a) the vertical straight lines and in plot b) one of the horizontal
straight lines. The values of the parameters are: =10.0, =100,
=0.2, =0.4, =9.5, =0.8.
Numerical simulations show that the successive images
of the critical lines corresponding to c
1
and c
2
are always.
Concerning the critical lines corresponding to c3, they start
to appear very complex (see Fig. 5.).
97.5 102.5 107.5
y
47.5
50
52.5
55
57.5
60
k
Fig. 5. The result of five iterations of one horizontal critical
lines. The values of the parameters are: =10.0, =100, =0.2,
=0.4, =9.5, =0.8.
If we iterate all the critical lines a sufficient great number
of times, we get the following picture:
Chaotic behavior in an economic model 9
95 100 105 110
y
45
50
55
60
65
70
k
Fig. 6. Asymptotic behavior of the critical line for the values of
the parameters: =10.0, =100, =0.2, =0.4, =9.5, =0.8.
The figure shows that, asymptotically, the critical lines
end at 6 points.
Such sets of points are called -limit sets. That is, the -
limit set of a point x corresponds to the points q such that
the points F
nk
(x) (k=1, 2,) tend to q (clearly, q belongs to
the limit set of the trajectory). The set (x) is invariant and
can give us information about the long run behavior of the
trajectory from x.
The points in the figure correspond precisely to the points
of the orbit of period 2 or period 4 of the attractor,
depending on the considered initial condition.
The points with the same x-coordinate (that is, in the same
fiber) are due to the fact that, for these parameter values,
the basis map converges to one of the two existing fix
points, depending on the starting point, c
m
or c
M
. Then, in
each fiber, the asymptotic behavior of the fiber map
converges to a period-2 or period-4 orbit (Fig. 7. and Fig.
8.).
10 20 30 40
t
92.4
92.6
92.8
93.2
y

10 20 30 40
t
52
53
54
55
56
57
k
Fig. 7. Representation of the (a) income and (b) capital stock as
a function of time, with initial condition (c
2
, 30.0). The values of
the parameters: =10.0, =100, =0.2, =0.4, =9.5, =0.8.
10 Clara Grcio, Cristina Janurio and J. Sousa Ramos
10 20 30 40
t
106.8
107.2
107.4
107.6
y

10 20 30 40
t
52
54
56
58
60
k
Fig. 8. Representation of the (a) income and (b) capital stock as
a function of time, with initial condition (c
2
, 30.0). The values of
the parameters: =10.0, =100, =0.2, =0.4, =9.5, =0.8.
40 60 80 100
k
30
40
50
60
70
80
90
100
g92.83...,k
40 60 80 100
k
30
40
50
60
70
80
90
100
g107.16...,y
Fig. 9. The period-2 (in plot a)) and period-4 (in plot b)) orbits in
each fiber. The values of the parameters are =10.0, =100,
=0.2, =0.4, =9.5, =0.8.
Therefore, it seams pertinent to illustrate in a color
scheme the initial conditions that, by iteration of F, define a
trajectory that leads to each of the two periodic orbits. The
regions colored with the same color correspond to the
basins of attraction of one orbit. The dark grey color
corresponds to the period-2 orbit and the light grey color to
the period-4 orbit.
Fig. 9. Basins of attraction. The values of the parameters are:
=10.0, =100, =0.2, =0.4, =9.5, =0.8.
Chaotic behavior in an economic model 11
The figures show light and dark grey vertical stripes
revealing the two types of regions mentioned above.
3. Conclusions
In this work it was given a contribution to understand
economic fluctuations with business cycle models.
5. References
Bischi GI.; Dieci R.; Rodano R.; Saltari S. 2001, Multiple attractors and
global bifurcations in a Kaldor-type business cycle model, J. Evol Econ,
no. 11, pp. 527-554.
Dieci R.; Bischi GI.; Gardini, L. 2004, From bi-stability to chaotic
oscillations in a macroeconomic model, Chaos Solitons Fractals,
21, no. 2, pp. 403-412.
Kaldor N. 1940, The Model of the Trade Cycle. Economic Journal
no. 50, pp. 78-92.
Mira C; Gardini L.; Barugola A; Cathala J-C 1996 Chaotic dynamics
in two-dimensional noninvertible maps, World Scientific Publishing
Co. Pte. Ltd,