Chapter 1

CATASTROPHE THEORY AND THE BUSINESS CYCLE

We use the approach of R. Thom’s “Catastrophe Theory” to construct

a generalization of Kaldor ’s 1940 trade cycle. The model allows for
cyclic behavior which exhibits either rapid recoveries (recessions) or
slow recoveries (depressions).

In this paper we examine a variation on Kaldor’s (1940) model of the

business cycle using some of the methods of catastrophe theory. (Thom
(1975), Zeeman (1977)). The development proceeds in several stages. Sec-
tion I provides a brief outline of catastrophe theory, while Section II applies
some of these techniques to a simple macroeconomic model. This model
yields, as a special case, Kaldor’s business cycles. In an appendix to this
paper, we provide a more rigorous treatment of the cyclic behavior of the
model following the work of Chang and Smyth (1971).
In Section III, we describe a generalization of Kaldor’s model that allows
not only for cyclical recessions, but also allows for long term depressions.
Section IV presents a brief review and summary.
This paper is frankly speculative. It presents, in my opinion, some in-
teresting models concerning important macroeconomic phenomena. How-
ever, the hypotheses of the models are neither derived from microeconomic
models of maximizing behavior, nor are they subjected to serious empirical
testing. The hypotheses are not without economic plausibility, but they
are far from being established truths. Hence, this paper can only be said
to present some interesting stories of macroeconomic instability. Whether
these stories have any empirical basis is an important, and much more
difficult, question.

1.1 Catastrophe theory

Catastrophe theory is a branch of applied mathematics that was originated

by Rene Thom in order to describe certain kinds of biological processes.

University of Michigan. I wish to thank Winston Chang, Paul Evans, Richard

Schmalensee, Robert Solow, and Ramu Ramanathan for some helpful comments on an
earlier version of this paper. This work was supported in part by a grant from the
National Science Foundation.
2 CATASTROPHE THEORY AND THE BUSINESS CYCLE

(Thom (1975)). It has subsequently been applied to a wide range of biolog-

ical, physical and social phenomena, most notably by Christopher Zeeman.
(A complete collection of Zeeman’s work is available in Zeeman (1977).
One of the nicest applications is Zeeman’s study of the heartbeat (Zee-
man (1972)) which is a good starting point for any systematic study of
catastrophe theory.)
Applied catastrophe theory is not without its detractors (Sussman and
Zahler (1978)). Some of the applied work in catastrophe theory has been
criticized for being ad hoc, unscientific, and oversimplified. As with any
new approach to established subjects, catastrophe theory has been to some
extent oversold. In some cases, applications of the techniques may have
been overly hasty. Nevertheless, the basic approach of the subject seems,
to this author at least, potentially fruitful. Catastrophe theory may provide
some descriptive models and some hypotheses which, when coupled with
serious empirical work, may help to explain real phenomena.
Catastrophe theoretic models generally start with a parameterized dy-
namical system, described explicitly or implicitly by a system of differential
equations:

ẋ = f (x, a).

Here x is an n-vector of state variables, ẋ is the n-vector of their time

derivatives, and a is a k-vector of parameters. These parameters may also
change over time, but at a much slower rate than the state variables change.
Thus, it makes sense to model the system as if the state variables adjust
immediately to some “short run” equilibrium, and then the parameters
adjust in some “long run” manner. In the parlance of catastrophe theory,
the state variables are referred to as “fast” variables, and the “parameters”
are referred to as “slow” variables. This distinction is, of course, common
in economic modeling. For example, when we model short run macroeco-
nomic processes we take certain variables, such as the capital stock, as
fixed at some predetermined level. Then when we wish to examine long
run macroeconomic growth processes, we imagine that economy instanta-
neously adjusts to a short run equilibrium, and focus exclusively on the
long run adjustment process.
Catastrophe theory is concerned with the interactions between the short
run equilibria and the long term dynamic process. To be more explicit,
catastrophe theory studies the movements of short run equilibria as the
long run variables evolve.
A particularly interesting kind of movement is when a short run equi-
librium jumps from one region of the state space to another. Such jumps
are known as catastrophes. Under certain assumptions catastrophes can be
classified into a small number of distinct qualitative types. An example of
the simplest type of catastrophe, the fold catastrophe, is given in Figure 1.
 CATASTROPHE THEORY 3

.
x=0

a* a

Figure 1. The fold catastrophe.

Here we have depicted a system with one fast variable, x, and one slow
variable a. For any fixed value of a, the fast variable adjusts to a locally
stable equilibrium, This means that the system moves rapidly to either the
upper or the lower leaf of the ẋ = 0 locus, depending on initial conditions.
For definiteness, we imagine that initially a is rather small, so that x adjusts
to a short run equilibrium position on the upper leaf of the ẋ = 0 locus.
We now suppose that a slowly increases. Then the short run equilibrium
will move continuously to the right along the x = 0 locus, until it reaches
the fold point a∗ . Any further increase in a necessitates a jump from the
upper leaf to the lower leaf of the ẋ = 0 locus. This movement is an
example of the fold catastrophe. Note that any sort of return to the upper
leaf of the ẋ = 0 would involve a similar jump when a declined past the
other fold point.
It can be shown that in a system with one state variable and one param-
eter, this is essentially the only kind of catastrophe that can occur—any
sudden jumps in such two-dimensional systems are generally locally topo-
logically equivalent to Figure 1.1
In higher dimensions things get a bit more complicated. Suppose we
have a three dimensional system with one fast variable, x, and two slow
variables, a and b:

ẋ = f (x, a, b).
Then the ẋ = 0 locus will generally be some two dimensional surface
in R3 . This surface locally sits over the plane R2 in one of three possible
ways, which are depicted in Figure 2.

1 More precisely, one should qualify this statement by requiring that the catastrophe
be “stable” in the sense that its geometry is qualitatively unaffected by any small
perturbations. The word “generally” really should be “generically.”
4 CATASTROPHE THEORY AND THE BUSINESS CYCLE

Slow
return

b c
a
Fast
return

Figure 2. Classification of two-dimensional singularities.

Figure 2a is of course the no catastrophe case: any small movements

of the slow variables a and b result in a continuous adjustment of the
short run equilibrium. Figure 2b depicts the two-dimensional analog of the
fold catastrophe: it is essentially equivalent to the one dimensional case
described earlier. Figure 2c presents the only new and interesting case,
namely, the cusp catastrophe.
The cusp catastrophe is essentially a fold that disappears at some point
(the cusp point). This feature allows for a new kind of movement. Let us
imagine that we have established a short run equilibrium on the upper leaf
of the cusp. Now we let a and b vary so that we fall over the fold, down
to the lower leaf of the ẋ = 0 locus. Let us consider how the system might
return to the upper leaf.
It is apparent that there are two possible paths, both illustrated in Figure
2c. One path is to move back under the fold and then have a jump return
to the top leaf. The other route to the top would be to move around
the cusp, and eventually return to the original starting point. The first
type of return would generally be much faster than the second since the
second return relies exclusively on the movements of the slow variables.
Which type of return occurs depends of course on the nature of the slow
dynamics, the extent of the perturbations and so on. We will return to
this point later in the context of a specific example. For now, we simply
note that any system that allows for both fast and slow returns of the sort
described above must be locally equivalent to the cusp catastrophe.2 (Of
course, globally the shape of the ẋ = 0 locus may be very complex, and
the resulting dynamics may be quite complicated.)
In higher dimensions things get even more complex. In order to establish
a reasonable classification system it is necessary to limit oneself to gradient

2 Again, a rigorous statement of this result would involve the qualification in footnote 1.
 A MODEL OF RECESSIONS 5

dynamical systems. In this case Thom has shown that when k ≤ 4, there
are generically only 7 kinds of local catastrophes. For a development of
some of the mathematical ideas involved in this classification theorem the
reader should consult Golubitsky (1978), Wasserman (1974), Lu (1976),
Thom (1975), or Zeeman (1977).
In the economic model that follows we will only utilize the two simplest
catastrophes, the fold and the cusp. In these low dimensional cases, there
are no restrictions on the nature of dynamical systems involved. In some
ways, the material that follows is more closely related to the mathematical
theory of the singularities of maps, rather than catastrophe theory per se.
(For an elementary survey of singularity theory see Callahan (1974); for a
more advanced exposition see Golubitsky and Guillemin (1973)).

1.2 A model of recessions

Let us begin with a simple dynamic model of national income adjustment.

ẏ/s = C(y) + I(y, k) − y

k̇ = I(y, k) − I0 .
Here y is (gross) national income, k is the capital stock, C(y) is the con-
sumption function, I(y, k) is the gross investment function and I(y, k) − I0
is the net investment function. The autonomous component of the in-
vestment function, I0 , represents investment for the purpose of replacing
depreciated equipment.3 The parameter s represents the speed of adjust-
ment of the first process, which we generally believe is quite rapid at least
when compared to movements of the capital stock.
The dynamic relations postulated are, I believe, relatively standard.
However, they are not without problems. First, note that the specifica-
tion of the k equation implies that any long run equilibrium must involve
zero net investment, and any cyclical behavior by the model must involve
a declining capital stock over part of the cycle. Since such declines are
rarely observed, this feature has cast doubt on the relevance of nonlinear
models of the business cycle such as espoused here (c.f. Rose (1967), p.
153). However, this point is easily dealt with. The k variable is presum-
ably measured relative to trend; thus the capital stock may well continually
increase—but sometimes it may grow at a more rapid pace than at other
times. The cyclic aspect of the model would then be concerned with these
fluctuations about the trend level of growth.

3 Some sort of distinction between gross and net investment is necessary if we wish
savings to be positive even when investment is negative The usual formulation is
to make replacement investment proportional to k. But this tends to obscure some
of the calculations in the appendix. The autonomous investment formulation is a
compromise for the purposes of exposition.
6 CATASTROPHE THEORY AND THE BUSINESS CYCLE

Secondly, there is a problem with the k̇ equation when the y equation is

out of equilibrium; namely, that when desired savings is less than desired
investment, actual investment—and thus actual capital accumulation—will
presumably be constrained. One might argue that this disequilibrium con-
straint shows up as an involuntary accumulation or decumulation of inven-
tories, but this raises even more vexing problems. (See Blinder (1977)).
Another way to deal with this problem would be to specify an adjustment
equation of the form k̇ = min[I(y, k), S(y)]. However, the nondifferentia-
bility in this formulation raises technical difficulties.
Instead, we will take a less satisfying, but perhaps more expeditious
approach, and simply ignore this problem. Since we will be postulating
that the y equation adjusts exceedingly fast, I cannot believe that this
approximation will cause any serious difficulties.

S(y)
I(y,k)
I(y,k1)
S(y)
I(y,k2)

I(y,k3)

Figure 3. Savings and investment behavior which generate the fold catas-
trophe.

We will now proceed to characterize the dynamic behavior of the system

by describing qualitative features of the functional forms. First, we will
suppose that the consumption function is linear so that C(y) = cy + D.
This seems uncontroversial.
For the investment function we will postulate a somewhat more unusual
shape, namely, the sigmoid shape shown in Figure 3. Kaldor (1940) has
argued for this shape as follows:

(The marginal propensity to invest) . . . “will be small for low levels

of activity because when there is a great deal of surplus capacity, an
increase in activity will not induce entrepreneurs to undertake addi-
tional construction: the rise in profits will not stimulate investment.
(At the same time the level of investment will not be zero, for there is
always some investment undertaken for long-period development pur-
poses which is independent of current activity.) But it will also be
 A MODEL OF RECESSIONS 7

small for unusually high levels of activity because rising costs of con-
struction, increasing costs and increasing difficulty of borrowing will
dissuade entrepreneurs from expanding still faster—at a time when
they already have large commitments.” Kaldor (l940), p. 81.

We can express this a bit more systematically as follows. Let us suppose

that prices and wages are fixed, or at least sticky. Then an increase in
national income should initially stimulate investment through the usual
sort of expectations effect. However, an increase in national income will
also push up interest rates through the liquidity balance effect, and this
increase in interest rates might well choke off ensuing investment. The net
results of such interactions could easily be the shape depicted in Figure 3.
Of course, investment also depends on the level of the capital stock. We
will make the natural assumption here: as k increases, other things being
equal, the investment function will shift down. Again, this hardly seems a
controversial assumption.
Now let us describe the shape of the ẏ = 0 locus. This is the set of all
y and k such that the demand for output equals the supply of output or,
equivalently where investment equals savings. In Figure 3 we have super-
imposed the linear savings function and the sigmoid investment function,
for three different values of k. When k is very low, we will have a high level
of investment and thus a high short run equilibrium level of income. As
k increases, we get three possible equilibrium values for y, two stable and
one unstable. Finally when k is rather large (relative to trend), investment
will be small, and thus equilibrium income will be small. These simple
observations allow to depict the ẏ = 0 locus in Figure 4. Note that it has
the shape of the fold catastrophe.

.
y k=0

.
y=0
k

Figure 4. The (y, k) phase space.

We now proceed to the k̇ = 0 locus. It is clear that this must be an

upward sloping line in the (y, k) phase space. It turns out that as long
8 CATASTROPHE THEORY AND THE BUSINESS CYCLE

as the marginal propensity to consume is less than one, the k̇ = 0 locus

intersects the ẏ = 0 locus in exactly one point. This fact is proved in the
appendix.
There are thus three possibilities: we could have a long run equilibrium
on the upper, middle, or lower leaf of the ẏ = 0 locus. The upper and lower
cases are symmetric, so we will omit a discussion of the lower intersection.
The middle intersection is the case considered by Kaldor (1940) and, more
rigorously, by Chang and Smyth (1972). It has been shown by Chang and
Smyth that when the speed of adjustment parameter is large enough, and
certain technical conditions are met, there must exist a limit cycle in the
phase space. In the appendix I prove a slightly simplified and modified
version of this result.
This ”business cycle” proposition is clearly the result intuited by Kaldor
thirty years ago. However, the existence of a regular, periodic business
cycle causes certain theoretical and empirical difficulties. Recent theoretical
work involving rational expectations (Lucas (1975)) and empirical work on
business cycles (McCullough (1975), (1977), Savin (1977)) have argued that
(1) regular cycles seem to be incompatible with rational economic behavior,
and (2) there is little statistically significant evidence for a business cycle
anyway.
However, there does seem to be some evidence for a kind of “cyclic
behavior” in the economy. It is commonplace to hear descriptions of how
exogenous shocks may send the economy spiraling into a recession, from
which it sooner or later recovers. Leijonhufvud (1973) has suggested that
economies operate as if there is a kind of “corridor of stability”: that is,
there is a local stability of equilibrium, but a global instability. Small
shocks are dampened out, but large shocks may be amplified.
The catastrophe model just described offers a way of rationalizing these
features. Let us suppose that the long run equilibrium (or the long run
equilibrium growth path) occurs on the upper leaf of the ẏ = 0 locus, as
is shown in Figure 4. Then the resulting equilibrium is easily seen to be
locally stable. However, it is globally unstable.
The nature of this instability is rather interesting. Suppose for some
reason firms acquire too much capital, perhaps by overinvesting in inven-
tories. Then the investment function might shift down so far that the only
short run equilibrium level of income would be very low. As inventories
are gradually decumulated, k falls and investment begins to recover. At
a critical value of k, income jumps rapidly back to the upper leaf and k
eventually returns to its long run equilibrium level. Such a movement is
depicted in Figure 4.
Such a story seems to me to be a reasonable description of the function-
ing of the commonly described “inventory recession.” It may also describe
other longer term fluctuations in the pattern of capital accumulation. In
either case the cyclic behavior of the model rests on the assumed sigmoid
nonlinearity in the investment functions. Although the postulated nonlin-
 A MODEL OF DEPRESSIONS 9

earity has some inherent plausibility, there is no explicit evidence that in

reality investment does behave as suggested.
However, let us, for the sake of argument, accept such a story as providing
a possible explanation of the “cyclic” behavior of an economy. Then there
is yet another puzzle. Each recession in this model will behave rather
similarly: first some kind of shock, then a rapid fall, followed by a slow
change in some stock variables with, eventually, a rapid recovery. Although
this story seems to be descriptive of some recessions, it does not describe
all types of fluctuations of income. Sometimes the economy experiences
depressions. That is, sometimes the return from a crash is very gradual
and drawn out.
Now the fold catastrophe could be shaped so as to exhibit a long drawn
out recovery—simply bend the lower leaf up towards the upper leaf so that
the slow capital adjustment is a bigger fraction of the total movement.
However, the fold catastrophe does not allow for both kinds of movement:
the rapid recovery characterizing a recession and the slow recovery charac-
terizing a depression. To get such behavior we will have to utilize the cusp
catastrophe.

1.3 A model of depressions

In order to describe a model with both fast and slow recoveries, we need
to utilize the cusp catastrophe. The cusp catastrophe requires two slow
variables. One of these must be the capital stock, but what will the other
be? It seems natural to choose the other stock variable of short run macro-
economic theory, namely private wealth. By wealth here, I mean wealth
very broadly construed—including human capital, expectations of future
employment, all types of assets, and in fact any type of stock variable that
affects consumption decisions up to, and including, “consumer sentiment.”
Well, how should changes in wealth affect consumption decisions? It is
generally agreed that an increase in wealth will, other things being equal,
increase consumption. Conversely, a decrease in wealth results in a decrease
in consumption, presumably because consumers wish to save more in order
to rebuild their stocks of wealth.
We are going to assume such a positive relationship between consumption
and wealth and a bit more; namely that there is a positive relationship
between wealth and the marginal propensity to consume. That is, not only
does an increase in wealth increase the level of the consumption function,
it also increases the slope of the consumption function.
In the linear consumption function case that we have been using, this
assumption means that we can write C(y, w) = c(w)y + D(w) with c0 (w) >
0 and D0 (w) > 0. Again this assumption seems quite plausible, but I know
of no theoretical or empirical evidence to support it.
10 CATASTROPHE THEORY AND THE BUSINESS CYCLE

Nevertheless, this simple assumption about wealth effects is sufficient to

generate the cusp catastrophe. The argument is illustrated in Figure 5. For
“intermediate” values of k and w we have the same story as with the fold
catastrophe: shifts in k generate the fold part of the cusp catastrophe, with
the characteristic three sheets of short run equilibria, two stable sheets and
one unstable. If, on the other hand wealth becomes very small, the savings
function shifts upward and becomes more steep. If the savings function
becomes steep enough, then no amount of shifting k will result in multiple
equilibria: that is, if the marginal propensity to save exceeds the marginal
propensity to invest for all levels of income, then there can be only one
intersection of the savings and investment functions.

S(y,w)
I(y,k)
I(y,k1)
S(y,w)
I(y,k2)

I(y,k3)

Figure 5. Savings and investment behavior which generate the cusp catas-
trophe.

Putting these facts together, we see that the ẏ = 0 locus will be three
sheeted for some values of k and w, but will be one sheeted for sufficiently
low values of w. This characterizes the geometry of the cusp catastrophe. 4
It remains to describe the movements of the slow variables k and w. We
will suppose that there is some long run equilibrium (y ∗ .k ∗ , w∗ ) located
on the top sheet of the ẏ = 0 surface. We will also suppose that this
equilibrium is locally stable: when k is greater than k ∗ , k is negative; when
w is less than w∗ , w is positive and so on. An interesting configuration
consistent with these assumptions is illustrated in Figure 6. In this figure
the flow lines illustrating the ẏ movement and the flow lines hidden beneath
the fold have been omitted for clarity.

4 Zeeman has referred to the a and b parameters as representing “normal” and “split-
ting” factors involved in some underlying dynamical system. The method of generat-
ing the cusp catastrophe described in the text seems perfectly general and suggests
that a more descriptive terminology for a and b would be to call them the the “shift”
and the “tilt” factors.
 REVIEW AND SUMMARY 11

Figure 6. Depressions and recessions with the cusp catastrophe.

Here is the interesting feature of the model. Suppose as before, that there
is some kind of perturbation in one of the stock variables. For definiteness
let us suppose that some kind of shock (a stock market crash?) affects
w. If the shock is relatively small, we have much the same story as with
the inventory recession: a gradual decline in k, then a jump return to
the upper sheet, and an eventual return to the long run equilibrium. If
on the other hand the shock is relatively large, wealth may decrease so
much as to significantly affect the propensity to save. In this case, when k
declines, national income will remain at a relatively low level rather than
experience a jump return. Eventually the gradual increase in wealth due
to the increased savings will move the system slowly back towards the long
run equilibrium. (See Figure 6.)
According to this story the major difference between a recession and a
depression is in the effect on consumption. If a shock affects wealth so much
as to change savings propensities, recovery may take a very long time, and
differ quite substantially from the recovery pattern of a recession. This
explanation does not seem to be in contradiction with observed behavior,
but as I have mentioned earlier, it rests on unproven (but not implausible)
assumptions about savings and investment behavior.

1.4 Review and summary

We have shown how nonlinearities in investment behavior can give rise to
cyclic or cycle like behavior in a simple dynamic macroeconomic model.
12 CATASTROPHE THEORY AND THE BUSINESS CYCLE

This behavior shares some features with empirically observed behavior. If

savings behavior also exhibits nonlinearities of a plausible sort, the model
can allow for both rapid recoveries which characterize recessions, as well as
extended recoveries typical of a depression.

1.5 Appendix: the existence of business cycle

In this appendix we examine the behavior of the following dynamical sys-

tem:

ẏ = s[C(y) + I(y, k) − y]
k̇ = I(y, k) − I0 .
We shall make the following maintained assumptions:

ASSUMPTION. The consumption and investment functions are con-

tinuously differentiable functions of y and k. Furthermore, the domain of
2
the dynamical system can be chosen to be some compact region S of R + ,
diffeomorphic to a disk, and (ẏ, k̇) points inward on the boundary of S.

The last part of this assumption simply states that the system is even-
tually self-correcting: for large enough k, k̇ is negative, for large y, ẏ is
negative and so on,
We will use two results from the mathematical theory of dynamical sys-
tems, (For a survey of dynamical systems which describes these and other
useful theorems, see Varian (1978)).
The first theorem gives a criterion for the existence and uniqueness of
equilibria.

THEOREM (Poincare-Hopf ). Let ẋ = f (x) be a dynamical system

on the k-disk that points inward on the boundary of the disk. Furthermore,
suppose that all equilibria x∗i are regular in the sense that each Jacobian
matrix Df (x∗i ) is nonsingular. Then one can define the index of each
equilibrium x∗i , i = 1, . . . , n by


+1 if det(−Df (x∗i )) > 0
index(x∗i ) =
−1 if det(−Df (x∗i )) < 0
and these indices have the property that
n
X
index(x∗i ) = +1.
i=1
 APPENDIX: THE EXISTENCE OF BUSINESS CYCLE 13

Proof. Proof: See Milnor (1965) or Guillemin and Pollack (1974).

We can apply this theorem directly to the problem at hand. The Jacobian
matrix referred to in the Poincare-Hopf theorem takes the form:
" h i
s 1 − ∂C(y) ∂I(y,k)
−s ∂I(y,k)
#
∂y − ∂y ∂k
−Df (x) =
− ∂I(y,k)
∂y − ∂I(y,k)
∂k

The determinant of this matrix is easily seen to be

 
∂C(y) ∂I(y, k)
det(−Df (x)) = −s 1 − .
∂y ∂k
Thus as long as the marginal propensity to consume is less than one, all
equilibria must have index +1. The Poincare-Hopf theorem then implies
that there is exactly one equilibrium.
The second mathematical result used requires a bit of notation. Again
we let ẋ = f (x) be a dynamical system on the k-disk and we let φt (z)
be the state of the system at time t, if the system was in state x at time
zero. The function φt (x) is called the flow of the dynamical system. A
point x is a limit point of x̂ if there is some sequence tn → ∞ such that
limn→∞ φtn (x) = x̂. The set of all limit points of x is called the limit set
of x. For a description of the properties of limit sets as well as some other
aspects of the qualitative theory of differential equations, see Hirsch and
Smale (1974).
Intuitively speaking, the limit sets of a system are where the flow ends up
as time goes to infinity. One example of a limit set is an equilibrium such
as x∗ . If we start at x∗ , then we stay there forever more so it is certainly
a limit set. Another example of a limit set is a closed orbit. If we start
at any point on a closed orbit, we can find a sequence (tn ) such that the
flow converges to any other point. The Poincare-Bendixson theorem shows
that these are about the only examples of limit sets for dynamical systems
in the plane.

THEOREM (Poincare-Bendixson). A nonempty compact limit set of

a continuously differentiable dynamical system in the plane, which contains
no equilibrium point, is a closed orbit.

Proof. See Hirsch and Smale (1974), Chapter 11.

Let ẋ = f (x) be a dynamical system with equilibrium x∗ . We will say

that x∗ is totally unstable if x∗ is an asymptotically stable equilibrium of
the system ẋ = −f (x). The following lemma is an immediate consequence
of the Poincare-Bendixson theorem:
14 CATASTROPHE THEORY AND THE BUSINESS CYCLE

LEMMA. Let (y ∗ , k ∗ ) be the unique equilibrium of the macrodynamic

system (1). Suppose that (y ∗ , k ∗ ) is totally unstable. Then if (y, k) 6=
(y ∗ , k ∗ ) is any other point of S, the limit set of (y, k) is a closed orbit.

Proof. Proof: Let (y, k) 6= (y ∗ , k ∗ ) be a point of S. The limit set of (y, k) is

compact and nonempty since S is compact. It cannot contain (y ∗ , k ∗ ) since
(y ∗ , k ∗ ) is totally unstable, and it cannot contain any other equilibrium
point since (y ∗ , k ∗ ) is the unique equilibrium. By the Poincare-Bendixson
theorem, the limit set of (y, k) must be a closed orbit.

We now apply this lemma to the macrodynamic system (1). We already

know that (y ∗ , k ∗ ) is the unique equilibrium of the system; hence we only
need consider configurations where (y ∗ , k ∗ ) is totally unstable. We reex-
amine the Jacobian matrix of the system as given in expression (2). Since
the determinant of this matrix is positive, both eigenvalues have the same
sign. The trace of the Jacobian matrix is:
 
∂C(y ∗ ) ∂I(y ∗ , k ∗ ) ∂I(y ∗ , k ∗ )
Tr(Df (x∗ )) = s + −1 + .
∂y ∂y ∂k
(Remember that (2) is the negative of the Jacobian matrix.)
The second term of this expression is negative by assumption—an in-
crease in the capital stock will tend to decrease investment. If the long run
equilibrium is located on the upper leaf of the ẏ = 0 locus, then the first
expression will also be negative and the long run equilibrium will be locally
stable.
If, however, the long run equilibrium is located on the middle leaf of
the ẏ = 0 locus, the first term will be positive. If the speed of adjust-
ment parameter s is large enough this term will dominate, and the trace
will be positive. This, along with the positive determinant, implies that
both eigenvalues of the Jacobian matrix have positive real part, and hence
(y ∗ , k ∗ ) will be totally unstable. Applying the earlier lemma completes the
argument.

1.6 Relationship to previous work

The applicability of the Poincare-Bendixson Theorem to the Kaldor model
was first noticed by Chang and Smyth (1971). Their proof utilized a dif-
ferent version of the theorem and differs in some details, but is essentially
similar to that given above. Rose (1967) has applied Poincare-Bendixson
to a rather different sort of business cycle model, while Ichimura (1954) has
examined a parameterized version of the model and has shown it to give
rise to a form of Lienard’s equation.
 RELATIONSHIP TO PREVIOUS WORK 15

References

Blinder, A. (1977) “A Difficulty with Keynesian Models of Aggregate De-

mand,” in Natural Resource Uncertainty, and General Equilibrium Sys-
tems, ed. A. Blinder and P. Friedman. New York: Academic Press.
Callahan, J. (1974) “Singularities and Plane Mapsm,” American Mathe-
matical Monthly, March 1974.
Chang, W. and Smyth D. (1971) “The Existence and Persistence of Cycles
in a Nonlinear Model: Kaldor’s 1940 Model Reexamined,” Review of
Economic Studies, 38, 37–44.
Golubitsky, M. (1978) “An Introduction to Catastrophe Theory and its
Applications,” SIAM Review, 20, 352–387.
Golubitsky, M. and Guillemin, V. (1974) Stable Mappings and Their Sin-
gularities. New York: Springer-Verlag.
Guillemin, V. and Pollack, A. (1974) Differential Topology. Englewood
Cliffs, New Jersey: Prentice Hall.
Hirsch, M. and Smale, S. (1974) Differential Equations, Dynamical Sys-
tems, and Linear Algebra. New York: Academic Press.
Ichimura, S. (1954) “Towards a General Nonlinear Macrodynamic Theory,”
in Post Keynesian Economics, ed. K. Kurihara. New Brunswick, New
Jersey: Rutgers University Press.
Kaldor, N. (1940) “A model of the Trade Cycle,” Economic Journal, March
1940.
Kaldor, N. (1971) “A Comment,” Review of Economic Studies, 38, 45–46.
Leijonhufvud, A. (1973) “Effective Demand Failures,” Swedish Journal of
Economics, , 27–48.
Lu, Y. (1976) Singularity Theory and an Introduction to Catastrophe The-
ory. New York: Springer-Verlag.
Lucas, R. (1975) “An Equilibrium Model of the Business Cycle,” Journal
of Political Economy, 83, 1113–1144.
McCullough, J. (1975) “The Monte Carlo Cycle in Business Activity,” Eco-
nomic Inquiry, 13, 303-321.
McCullough, J. (1977) “The Monte Carlo Hypothesis: Reply,” Economic
Inquiry, 15, 618.
Milnor, J. (1965) Topology from the Differentiable Viewpoint. Charlottesville:
University of Virginia Press.
16 CATASTROPHE THEORY AND THE BUSINESS CYCLE

Rose, H. (1967) “On the Nonlinear Theory of the Employment Cycle,”

Review of Economic Studies, 34, 153–173.
Savin, N. (1977) “A Test of the Monte Carlo Hypothesis: Comment,”
Economic Inquiry, 15, 613–617.
Sussman, H. and Zahler, R. (1978) “Catastrophe Theory as Applied to the
Social and Biological Sciences: A Critique,” Synthese, 37, 1–216.
Thom, R. (1975) Structural Stability and Morphogenesis. New York: Ben-
jamin.
Varian, H. (1978) “Dynamical Systems with Applications to Economics,” in
Handbook of Mathematical Economics, ed. K. Arrow and M. Intriligator.
Amsterdam: North-Holland Press.
Wasserman, C. (1974) Stability of Unfoldings. New York: Springer-Verlag.
Zeeman, C. (1972) “Differential Equations for the Heartbeat and Nerve
Impulses” in Towards a Theoretical Biology 4: Essays, ed. Waddington,
Edinburgh University Press.
Zeeman, C. (1977) Catastrophe Theory. New York: Addison-Wesley.