Annals of Operations Research 89(1999)89–100 89

The Kaldor – Kalecki business cycle model★

Adam Krawiec a and Marek Szydl⁄ owski b
a
Department of Economics, Jagiellonian University,
Wislna 2, 31-007 Krakow, Poland
E-mail: uukrawie@cyf-kr.edu.pl
b
Astronomical Observatory, Jagiellonian University,
Orla 171, 30-244 Krakow, Poland
E-mail: uoszydlo@cyf-kr.edu.pl

The question of the determination of investment decisions and their links with economic
activity leads us to formulate a new business cycle model. It is based on the dynamic multi-
plier approach and the distinction between investment and implementation. The study of the
nonlinear behaviour of the Kaldor–Kalecki model represented by the second-order delay
differential equations is presented. It is shown that the dynamics depends crucially on the
time-delay parameter T – the gestation time period of investment. We apply the Poincaré–
Andronov–Hopf bifurcation theorem generalized for functional differential equations. It
allows us to predict the occurrence of a limit cycle bifurcation for the time-delay parameter
T = Tbif . The dependence of T = Tbif on the parameters of our model is discussed. As T is
increased, the system bifurcates to limit cycle behaviour, then to multiply periodic and
aperiodic cycles, and eventually tends towards chaotic behaviour. Our analysis of the dynamics
of the Kaldor–Kalecki model gives us that the limit cycle behaviour is independent of the
assumption of nonlinearity of the investment function. The limit cycle is created only due to
the time-delay parameter via the Hopf bifurcation mechanism. We also show that for a small
time-delay parameter, the Kaldor–Kalecki model assumes the form of the Liénard equation.

1. Introduction

Investment is a subject of great importance in economic theory. The question of

the determination of investment decisions and their links with the level and the rate of
change of economic activity leads us to formulate a new business cycle model which
follows the tradition established by Kalecki and Kaldor.
Both the Kaldor business cycle model and the Kalecki one use an investment
function which is based on the profit principle rather than the acceleration principle.
In the Kaldor model [8], the gross investment depends on the level of output and

★
This work was supported by KBN Grant No. 1 H02B 009 15.

© J.C. Baltzer AG, Science Publishers

90 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model

capital stock. For a given quantity of real capital, investment depends on the level of
profit, which in turn depends on the level of activity. Kaldor presented the assumptions
on nonlinear investment and savings functions and their shift over time which give
rise to a cycle. Thereafter, this model was often discussed, e.g. Ichimura [7] explored
the possibility that the economic system possesses a unique limit cycle. The most
important was the paper by Chang and Smyth [2], where the model was reexamined
and the necessary and sufficient conditions of the existence of a limit cycle were stated.
The co-existence of a limit cycle and an equilibrium was considered in Grasman and
Wentzel’s paper [4].
The Kalecki business cycle model [9] was a few years earlier than the Kaldor
one. Kalecki assumed that the saved part of profit is invested and the capital growth
is due to past investment decisions. There is a gestation period or a time lag, after
which capital equipment is available for production.
We formulate the Kaldor–Kalecki business cycle model based on the multiplier
dynamics which is the core of both the Kaldor (after Keynes) and Kalecki approach.
However, we follow Kalecki’s approach to investment and of a time lag between
investment decisions and implementation.
A study of nonlinear behaviour of the Kaldor–Kalecki business cycle model is
presented. The equations of the Kaldor–Kalecki model of the business cycle are
second-order delay differential equations. For a small time-delay parameter, these
equations assume the form of the Liénard equation similarly to the Kaldor model. For
a large time-delay parameter, they are equivalent to the infinite dimensional dynamical
system. It is shown that the dynamics depends crucially on the time-delay parameter
T conjugated with the investment decisions in the Kalecki theory.
Delay differential equations were extensively studied in different problems of
technics and biology [5,6,11]. The theory of these equations was often applied in
economics [1].
The problem of existence and a number of possible limit cycles for polynomial
vector fields on the plane is known as the XVI Hilbert problem. The extent of the
current mathematical theory of time-delay equations in several variables is quite
minimal, but a few general results and techniques of functional analysis and bifurcation
theory are relevant to the study of our dynamical problems. One of the most funda-
mental tools we have is the Poincaré–Andronov–Hopf bifurcation theorem, which
has been generalized for functional differential equations including delay equations
[3,10]. The power of the Hopf theorem is that, when its conditions are satisfied, it
guarantees both the existence and uniqueness of periodic trajectories. These can
be expressed only in terms of the time-delay parameter, speed of adjustment, initial
disturbances, the positions of savings without the fundamental assumption of an s-
shape of the investment function.
The Hopf theorem allows us to predict the occurrence of a limit cycle bifurcation.
Once this bifurcation occurs, we would like to have an analytic or semianalytic expres-
sion for the limit cycle, and to have the means of predicting secondary bifurcations.
 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model 91

However, a full analytic solution is not possible. In itself, the eigenvalue expression
derived from the linearized equations may be useful for determining when a second
complex conjugate eigenvalue pair crosses into the σ > 0 half-plane, but it cannot
predict the secondary bifurcation. However, the power series bifurcation analysis
provides a systematic method for approximating the limit cycle solutions and second-
ary bifurcations.
We apply the Hopf theorem and demonstrate that its conditions are fulfilled for
a generic class of the parameters in the Kaldor–Kalecki model. It allows us to predict
the occurrence of the limit cycle bifurcation for the time-delay parameter T = Tbif .
The dependence of T = Tbif on the parameters of our model is discussed. As T is
increased, the system bifurcates to the limit cycle behaviour, then to multiple periodic
and aperiodic cycles, and eventually tends towards chaotic behaviour.
Our analysis of the dynamics of the modified Kaldor model gives us that the
limit cycle behaviour is independent of the assumption that the investment function is
s-shaped. In our study, we assume the linear function I(Y), and only the time-delay
parameter plays a crucial role in creating the limit cycle.
We also show that for a small time-delay parameter, in the linear approximation,
the Kaldor–Kalecki model assumes the form of the Liénard equation.
A linear stability analysis including the time delay is shown to be an accurate
predictor of the critical T for the first bifurcation. As T is increased, the system bifur-
cates to the limit cycle behaviour, then to multiple periodic and aperiodic cycles or
eventually tends towards the chaotic behaviour via the period doubling cascade route
to the turbulence.
It is noteworthy that all these types of behaviour (without limit cycles) are
accompanied by Kalecki’s time-delay parameter.
We conclude that the business cycle model reconstructed on the basis of the
Kaldor model and the Kalecki time-delay parameter connected with investment deci-
sions generates limit cycles in the phase space. The crucial role in the creation of the
limit cycle is Kalecki’s time-delay parameter, rather than the assumption of the s-
shaped investment function I(Y).
The paper is organized into four sections. In section 2, we reconstruct the new
model of business cycles based on the Kalecki theory in conjuction with the Kaldor
two-dimensional model on the phase plane (Y, K). In section 3, the bifurcation theory
is adopted to the investigation of the limit cycle behaviour in our model. Finally,
section 4 contains conclusions.

2. The Kaldor–Kalecki model of trade cycle with delayed investment

Consider the Kaldor macroeconomic business cycle model as a two-dimensional
autonomous dynamical system in the form
dY
= α[I (Y, K ) − S(Y, K)], (1)
dt
92 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model

dK
= I(Y, K) − δ K, (2)
dt
where I is the investment and S is the savings function, Y is gross product, K is capital
stock, α is the adjustment coefficient in the goods market, and δ is the depreciation
rate of the capital stock. Kaldor assumed that the investment function I is nonlinear
(s-shaped) on Y.
The system (1)–(2) has been extensively studied from the point of view of the
existence and persistence of cycles in this model [2,4].
In the Kalecki [9] theory, the fundamental role is the time delay parameter T
connected with the investment decisions. He distinguished the three investment stages:
investment orders I, production of capital goods A and deliveries of finished capital
goods D. The change in the capital stock is due to the past investment orders,
d K(t)
= D(t) − U = I(t − T) − U,
dt
where U denotes the capital depreciation.
In order to formulate a new model of the business cycle, we assume Keynes’
proposition that the economy reaches the level of activity where savings and invest-
ment are equal. Then we apply Kalecki’s idea of a time lag in the capital accumulation
equation. We couple the dynamic multiplier approach of the Kaldor model with the
Kaleckian time delay in investment which plays a main role in the capital accumulation
process. Of course, the length of the lag varies between different types of equipment.
Kalecki assumed an average lag between decision and implementation. We will also
treat the lag as a constant.
We formulate the Kaldor–Kalecki model as the time-delay differential equation
system
dY
= α[I(Y(t), K(t)) − S(Y(t), K (t))], (3)
dt
dK
= I(Y(t − T ), K(t)) − δ K (t), (4)
dt
where the time delay T = const.
Investment depends on income at the time investment decisions are taken and on
capital stock at the time investment is finished. The latter is a consequence of the fact
that at time t – T, there are some investments which will be finished between t – T and
T. We assume that capital stock produced in this period is taken into consideration
when new investments are planned.
The savings function S depends only on Y and is linear such that SY = γ ∈(0, 1).
Additionally, we assume that the investment function I(Y, K ) separates with respect to
its two arguments and I Y > 0, I(K) is linear such that IK = β < 0; then

I(Y, K) = I (Y) + β K.
 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model 93

After substituting the above assumption into the Kaldor–Kalecki dynamical

system, we have
dY
Y˙ ≡ = α I(Y(t)) + α βK(t) − αγ Y(t),
dt
dK
K˙ ≡ = I(Y(t − T)) + (β − δ )K(t),
dt
or equivalently
Y˙˙(t) + f (Y) Y˙ (t) + g(Y ) = 0, (7)
where
∂I
f (Y(t)) = − α + α γ − β + δ,
∂Y
g(Y(t), Y(t − T)) = − αγ (β − δ )Y(t) + α(β − δ )I(Y(t)) − α β I(Y(t − T)).

2.1. The Liénard form of the Kaldor–Kalecki model

For a small value of the time-to-build parameter T, i.e. when T ¿ 1, our system
takes the form of the Liénard equation. Let us consider the nonlinear terms in (7), i.e.
I(Y(t)) − I(Y(t − T))
g(Y(t)) = − α δI (Y(t)) − α γ (β − δ)Y(t) + α βT
T
= − α δI (Y(t)) − α γ (β − δ)Y(t) + ∆h(Y(t)),
where
∆I
∆h(Y(t)) = α βT .
∆t
If T is small, ∆h(Y ) can be expressed in terms of a differential quotient and we
obtain
dI
∆h(Y) → d h(Y) = α βT , T ¿ 1. (8)
dt
If we take into account that dI dt = Y˙ ∂I ∂Y for small T, in the linear approximation
we have
Y˙˙ + α(γ + βT IY (Y ) − IY (Y)) Y˙ − α δI (Y ) − αγ (β − δ )Y = 0. (9)

Equation (9) can be reduced to a two-dimensional autonomous dynamical system of

the Liénard type. This equation will be called the Kaldor–Kalecki equation in the
linear approximation for a small time-delay parameter.
The Kaldor–Kalecki equation (9) can be represented in the Liénard form
x˙ = y − F ( x), y˙ = − g( x),

where F (x) = ∫0x f (u)du, with modified function f (Y), and consequently F (x)

f (Y) → f ( y) + α βT IY ,
F ( x) → F (x) + α βTI (x).
94 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model

In the general case of any finite time-delay parameter T, the dynamics is described
by second-order time-delay differential equations which are equivalent to an infinite
set of ordinary differential equations of first order.
If we write ∆I(Y) as a series
`
(− 1) n n d n I (Y(t))
∆ I(Y) = I(Y(t)) − I (Y(t − T)) = − ∑ T
dt n
,
n =1 n!
then the corresponding Kaldor–Kalecki equations take the form of the infinite dimen-
sional dynamical system. To see this fact, it is sufficient to define the subsequent
derivatives of Y as phase variables.
Let us note that in the case of the Liénard system, if the investment function I(Y )
is nonlinear, the existence of a limit cycle can be deduced on the basis of the standard
theorems.

3. Time delay parameter dependence in the Kaldor–Kalecki model

Because the oscillatory and chaotic tendencies of the Kaldor–Kalecki model
should have a number of observationally interesting consequences, it is worthwhile to
take a more detailed look at the parameter dependences.
In the Kaldor model, the investment function was nonlinear with respect to gross
product. We abandon this assumption and the investment function becomes
I (Y, K) = ηY + β K,
where η > 0.
If we assume a solution of (7) in the form Y(t) = e λt, we derive the eigenvalue
equation
λ2 + Aλ + B + De − λT = 0, (10)
where A, B and D are constant:
A = − α η + α γ − ( β − δ ),
B = α(β − δ)η − α γ( β − δ ),
D = − α β η.
As usual for time-delay equations, we have the exponential term in (10), making
the equation transcendental. The first consequence is that usually we cannot solve
the general eigenvalue equation analytically, and the second one is that there is an
indefinite number of eigenvalues. The latter is a reflection of the fact that, like partial
differential equations, the differential time-delay equations are equivalent to an in-
definite set of ordinary differential equations.
In essence, we have two main tools besides direct numerical integration; firstly,
the linear stability analysis, especially in the limit T ¿ 1, and secondly, the formal
results of the bifurcation theory that have been extended to time-delay equations,
notably including the Hopf bifurcation theorem. In the subsequent subsections, we
discuss both approaches.
 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model 95

3.1. Linear stability analysis

Let us consider the most important limit case when T ¿ 1. Then, exp(– λT) ≈
1 – λT and the eigenvalue equation becomes
λ2 + ( A − DT)λ + B + D = 0. (11)
Now this is a simple quadratic equation in λ. Solving this expression shows that
the real parts of eigenvalues are negative at very small values of T (if A > 0), but
above a critical value of T bif , at least one real part becomes positive. Nonetheless, the
usefulness of the small T expression is very limited because the linearization theorem
is not true for the time-delay differential equations.
It is useful to write any eigenvalue in terms of its real and imaginary parts,
λ = σ + i ω , and write the real and imaginary parts of the general eigenvalue equation.
For small T ¿ 1, we have
σ 2 − ω 2 + σ ( A − DT) + B + D = 0, (12)
2σω + ω( A − DT) = 0, (13)
the solution of which is
DT − A
σ = ,
2
T = Tbif ⇔ σ = 0,
A − αη + α γ − (β − δ )
Tbif = = .
D −α β η
Therefore, if A > 0, then there is a bifurcation parameter T = Tbif for which the real
part of the eigenvalue changes sign. Let us note that σ (T ) is an increasing function of
its argument T such that ∂λ ∂T | T =Tbif > 0. The imaginary part of the eigenvalues takes
the form
ω = B + D, (14)
where the condition B + D ≥ 0 is equivalent to
γ (δ − β)
≥ η, δ ≠ 0.
δ
We see that the value of ω for T = Tbif is positive. This fact guarantees the bifurcation
to a limit cycle for T = T bif = A D. The function σ (T) is increasing and linear with
respect to its argument.

3.2. The Hopf bifurcation to a limit cycle

It is useful to write any eigenvalue in terms of its real and imaginary parts,
λ = σ + i ω, and write the real and imaginary parts of the general eigenvalue equation.
96 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model

We obtain the following system:

σ 2 − ω 2 + σ A + B + De −σ T cos ωT = 0, (15)
2σω + ω A − De −σ T sin ω T = 0. (16)
The general eigenvalue expression (15)–(16) can be used for the prediction of
limit cycles in the Kaldor–Kalecki model. This can be solved easily only numerically.
The limit cycle bifurcation occurs when the real part of a complex conjugate pair of
eingenvalues changes its sign from negative to positive. To find the first bifurcation
point, we set σ = 0. We might naïvely suspect that subsequent bifurcations are the
result of more eigenvalue pairs acquiring a positive real part. From the eigenvalue
equations we see that this occurs when the limit cycle period and the time delay satisfy

ω T = ω bif Tbif + 2π n, n = 1, 2,… .

Actually, it is more convenient to use the new parametric variable y = ωT, since this is
the argument of the trigonometric functions in (15)–(16). The imaginary part, equation
(16), can be solved for ω:
D sin y
ω = . (17)
A
This result is in turn substituted into the real part, equation (15). Then, with the help
of the trigonometric identity sin2 y = 1 – cos2 y, we obtain

D2 cos2 y + DA 2 cos y + E = 0, (18)

2 2
where E = BA – D .
This is readily solved for y, (17) is then solved for ω, and finally
y A y A 1
Tbif = = = . (19)
ω D sin y D (sin y) y
From the last formula, we can see that as the time-delay parameter is small T ¿ 1 (or
sin y y ≈ 1), T = Tbif = A D as in the previous linearized case.
The dependence of Tbif on the parameters of our model is determined from the
following relation:
A y A arccos z A arccos z
Tbif = = = ,
D sin y D sin(arccos z) D π 2−z
where
z1,2 =
1
2D
(
− A2 m A4 − 4(A 2 B − D2 ) . )
In figure 1, the dependence of the bifurcation parameter on z is shown. The
amplitude of this relation is equal to A D and corresponds to the value of z = 0. In
figure 2, it is checked that the derivative of the real part of the eigenvalue with respect
to T is positive for small y.
 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model 97

Figure 1. Diagram to illustrate the dependence of bifurcation parameter Tbif

on z = cos y. When z → 0, Tbif → A D, as in the linear approximation case.

Figure 2. Dependence of ∂σ ∂T |T dif on y.

The Hopf theorem predicts a bifurcation to a limit cycle when (a) the real part σ
of a pair of complex conjugate (with nonzero imaginary part) eigenvalues changes its
sign from negative to positive as parameter T is varied, and (b) the derivative of the
real part with respect to parameter T is positive as σ passes through zero. We saw that
condition (a) occurs in the Kaldor–Kalecki model as the time-delay parameter is
increased. We can verify condition (b) by differentiating the eigenvalue equation (10),
which yields
∂λ λ De− λT
= ,
∂T 2λ + A − DTe − λT
and it follows that

∂σ ∂λ De − σ T [ A(σ cos ω T + ω sin ωT ) + 2 c o sω T (σ 2 + ω 2 ) − TDσ e −σ T ]

= Re =
∂T ∂T ( 2ω + TDe −σ T sinω T ) 2 + ( 2σ + A − T D e−σ T cos ω T ) 2
98 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model

or
∂σ D( Aω sin ω T + 2ω 2 cos ω T)
= .
∂T T = Tbif (2ω + T Dsin ω T) 2 + (A − T Dcos ω T)2

If we assume that A D ≥ 0 (D is always positive), then ∂σ ∂T | T = Tbif is positive

for the argument y = ωT ∈[– π 2, π 2].
In sum, the linear stability analysis shows that the creation of a limit cycle is due
to the Hopf bifurcation (i.e. the real part of a complex conjugate eigenvalue pair
changes its sign from negative to positive) and, as expected in the Hopf bifurcation,
the limit cycle is approximately an ellipse near the bifurcation. In the general case,
one can check that for T < Tcrit = (Tbif y)arctan –1 (– ATbif (2y)), we have ∂σ ∂T > 0,
which means that σ (T) is an increasing function of its argument.
Moreover, we can determine the period of an orbit by
2π 2π Tbif
P= =
|λ(Tbif )| | y|
and the radius of an orbit by R ∝ T.

Figure 3. Real σ (triangles) and imaginary ω (circles)

part of the eigenvalue as a function of the time delay
T in the region of the limit cycle bifurcation Tbif .

Figure 3 shows a sequence σ, ω(mod 2π T ) values as T increases, for a different

typical parameter set. It was obtained by numerical experiments. The change of the
sign of σ from negative to positive is confirmed by the linear stability analysis.

4. Conclusion
Investment is often seen only as a component of aggregate demand. However,
investment is a more complex problem. When we take into account the distinction
between investment decision and expenditure, we come to the problem of gestation
 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model 99

lags in investment. This led us to formulate the simple Kaldor–Kalecki business cycle
model which, due to the time delay in investment, is represented as a time-delay
differential equation system.
A crucial problem in any business cycle theory is the persistence of cycles.
Kalecki adopted the criticized idea (but later dropped it) to assert that the values of
parameters are such that there is a cycle with a constant amplitude. Therefore, we take
another route and formulate a simple model with gestation lags which is capable of
generating stable cycles. By the methods of nonlinear dynamics and bifurcation theory,
we establish the existence of a stable limit cycle in the model by the Hopf bifurcation
mechanism.
We use the Poincaré–Andronov–Hopf theorem to predict the occurrence of a
bifurcation to a limit cycle for some values of the time-delay parameter. We obtain the
limit cycle behaviour in an alternative way to the Poincaré–Bendixson limit cycle
theorem used in the Kaldor model. Our model admits the limit cycle behaviour even
for a linear function of investment. We prove that Kalecki’s time-delay parameter can
create the limit cycle analogously to the Kaldor proposition of the s-shaped invest-
ment function.
In the approximation of small time-delay parameter T, our model takes the form
of the two-dimensional autonomous dynamical system of the Liénard type. It can be
shown that the persistence of cycles, in general, in the linear approximation T ¿ 1
depends crucially on the delay parameter values T and, additionally, on the speed of
adjustment and initial disturbances. We stress that T = A D is a good approximation
of the first bifurcation. The problem of the existence of further bifurcations is interest-
ing if it can lead to a cascade of bifurcations and eventually to chaotic behaviour.
There is no doubt that the time lags in investment and their capacity for generat-
ing cycles should be taken into account when we speak about business cycles. The
other question is the interdependence of the different aspects of investment which are
responsible for generating the cyclic behaviour, i.e. the influence of nonlinearities of
the s-shaped investment function and or time lags in investment on the amplitude and
period of oscillations.

Acknowledgements

The authors are grateful to Dr. Janusz Tobol⁄ a for drawing the figures. We thank
the anonymous referees for helpful comments.

References

[1] R.G.D. Allen, Mathematical Economics, 2nd ed., Macmillan, 1959.

[2] W.W. Chang and D.J. Smyth, The existence and persistence of cycles in a nonlinear model: Kaldor’s
1940 model re-examined, Review of Economic Studies 38(1971)37–44.
[3] S.N. Chow and J.K. Hale, Methods of Bifurcation Theory, Springer, 1982.
100 A. Krawiec, M. Szydl⁄owski y The Kaldor–Kalecki model

[4] J. Grasman and J.J. Wentzel, Co-existence of a limit cycle and an equilibrium in Kaldor’s business
cycle model and its consequences, Journal of Economic Behavior and Organization 24(1994)369–
377.
[5] J.K. Hale, Functional equations: Some models and perturbations, in: Differential Equations and
Dynamical Systems, Academic Press, 1967.
[6] J.K. Hale, Theory of Functional Equations, Springer, 1977.
[7] S. Ichimura, Toward a general nonlinear macrodynamics theory of economic fluctuations, in: Post-
Keynesian Economics, ed. K. Kurihara, Rutgers University Press, 1954, pp. 192–226.
[8] N. Kaldor, A model of the trade cycle, Economic Journal 50(1940)78–92.
[9] M. Kalecki, A macrodynamic theory of business cycles, Econometrica 3(1935)327–344.
[10] J.E. Marsden and M. McCraken, The Hopf Bifurcation and its Applications, Springer, 1976.
[11] R.M. May, Stability and Complexity in Ecosystem Models, Princeton University Press, 1975.