The cobweb model

Its instability and the onset of chaos

Carl Chiarella

We introduce a fairly general non-linear supply function into the traditional cobweb model under adaptive expectations. Wefind that the dynamics of the model are driven by a single hump map of the type that occurs in the chaos literature. By applying some recent results of Feigenbaum we are able to show that in its locally unstable region the cobweb model exhibits a regime of period doubling followed by a chaotic regime.
Keywords: Cobweb; Instability; Chaos

One of the most important scientific events in the last two decades has been the discovery of chaotic behaviour in simple, non-linear, deterministic dynamic systems. The essential characteristic of this behaviour is that even though all time paths are bounded, trajectories which start close together separate exponentially. The scientific significance of chaotic behaviour is that precise prediction in deterministic models is impossible. Deterministic models displaying chaotic behaviour generate a fundamental randomness which cannot be removed by the gathering of more information. Chaotic behaviour was first observed by Lorenz [ 141 in computer simulations of a model of atmospheric dynamics. Many of the subsequent developments occurred within the context of models of the dynamics of biological populations and in this regard the works of May [ 173 and Guckenheimer, Oster and Ipaktchi [9] are significant. The books of Schuster [21] and Kaneko [ 1l] are good introductions to the subject of

The author is with the Department of Finance, The University of New South Wales, Sydney, Australia.
This work was performed while the author was on study leave in the School of Economics, University of New South Wales. The author wishes to thank Professor Richard Manning for a conversation which stimulated this paper and for subsequent comments. Some discussions with Dr Daniel Leonard also proved invaluable. Comments from Professor Alistair Mees on some technical aspects of the use of Feigenbaums results are gratefully acknowledged. The author also thanks his colleague, MS Ernestine Gross, for providing an English translation of the Leontief article and some comments thereon. The usual caveat applies. Final manuscript received 18 December 1987.

chaos while the article in Scientific American by Crutchfield, Farmer, Packard and Shaw [4] provides an excellent non-mathematical discussion. Economic theorists were quick to appreciate the significance of these new developments in the theory of dynamic systems. The possibility of chaotic behaviour was found in models of conflict over distributive shares by Pohjola [20], models of economic growth by Day [ 51 and overlapping generation models by Benhabib and Day [2] and Grandmont [8]. Ofequal importance is the development of empirical tests by Barnett and Chen [1] which indicate the existence of chaotic behaviour in certain monetary aggregates. The new developments in the theory of dynamic systems show a clear need for economic theorists to take a fresh look at many of the basic models of economic dynamics. Traditional analysis of such models concentrated on regions of local stability while regions of local instability were ignored or dismissed as improbable because exploding time paths are not observed in real economic data. However, it is local instability which can indicate the emergence of chaotic behaviour, so these regions need to be carefully studied by the introduction of appropriate non-linearities. In this paper we carry out such an analysis for one of the basic models of economic dynamics, namely the cobweb model. The cobweb model of price dynamics in a single (typically agricultural) market has spawned a vast literature and has recently again become topical in the rational expectations literature (see, for example, Sheffrin [22]). Of particular interest are the stability conditions of the model.

02&I-9993/88/040377-08

SO3.00 0 1988 Butterworth & Co (Publishers)

Ltd

377

The cobweb model: C. Chiarella

One of the earliest references to the stability conditions of the cobweb model is Ezekiel [6], though the modern discussion seems to start with Nerlove [ 191. Further refinements were added by Carlson [ 33 and Manning [ 15,163. An important condition for stability of the cobweb model is that the supply elasticity be less than the demand elasticity. However, as Sheffrin points out, it is more likely that this condition will be violated in empirical studies. Our aim in this paper is to investigate the fate of cobweb cycles when the equilibrium is unstable. Scant attention has been paid in the literature to the behaviour of the model as prices move away from the equilibrium. There are, however, two important exceptions. In a largely ignored paper Leontief [ 121 demonstrated the possibility of two cycles when the equilibrium of the cobweb is locally unstable. As we shall see later, the emergence of two cycles is the first step on the path to chaos. Recently Jensen and Urban [lo] showed the existence of chaotic behaviour in a cobweb model having a particular structure. We shall analyse a fairly general cobweb model having a broad class of non-linear supply functions. We shall assume, in conventional manner, that suppliers form prices adaptively but we shall allow for the expectations time lag and production time lag to differ. We find that the dynamics of the model are driven by a quadratic map of the type which arises in much of the chaos literature. In the region of local stability the dynamic behaviour is as predicted by the standard linear theory. In the region of instability price time paths move out to a cycle (starting with the two cycle discovered by Leontief) whose period doubles as a certain parameter is increased. Beyond a certain critical parameter value the cyclical motion becomes chaotic.

Figure 1. Non-linear

supply function.

The model
We shall use p, and rc, to denote price and expected price respectively, both at time t. We assume a traditional linear demand function for the good given by
q: = a + bp, a>O, b<O (1)

The conditions that we impose on the non-linear function f are not very stringent. The main condition necessary for the model to generate chaotic behaviour is that there be a unique point of inflection at IL*, say, such that f is convex to the left of this point and concave to the right of it. We have drawn fin Figure 1 satisfying f(0) = 0; however this condition is not necessary to obtain our results. For instance, a supply function displaying zero supply below a threshold expected price is just as easily accommodated into our analysis, provided the point of inflection is maintained. Some readers might like to add the condition f(n) +O as 7c+ co, to reflect supply constraints. Again this condition is not necessary but can be incorporated into the analysis, as we shall indicate at the appropriate point, We assume that price expectations are formed adaptively according to
II 1+1

=(I

-wk,+wp,

(3)

However, we shall assume a non-linear supply function, dependent upon the expected price, which we write as

4: = f(%)

(2)

and suppose that the function f has the general shape shown in Figure 1.

where w ( > 0) is the speed of adjustment parameter and u = l/w may be interpreted as the mean time lag in the formation of expectations. Note that we are implicitly choosing time units so that a unit of time corresponds to the period between transactions. Since it would make little sense for economic agents to form expectations with a time lag less than that between transactions, we assume that u>liew<l.

378

ECONOMIC

MODELLING

October 1988

The cobweb model: C. Chiarella

Equality of supply and demand in the market for the good leads to a + bp, = f (71,) (4)

Equations (3) and (4) simultaneously determine pr and IL,. Using (4) to eliminate p, in (3) we find that n, satisfies the non-linear difference equation
Af+

1=

lid%)

(5)

where g(F) = - aw/b + (1 - w)n + wf(lr)/b The equilibrium of this system is given ii = g(it)

a
(6) by
(7)

A study of the dynamics of Equation (5) proceeds from a study of the local stability properties of the equilibrium ii. In fact il is locally stable for parameter values such that
1

Iiit(~)l< 1
which reduces to

(8)

sl

n2

TI

b
Figure 2.

w< 1
The function g.

1_1<f(ii)<1
W b

(9)
L

In the conventional cobweb model w = 1 is assumed and in this case (9) reduces to the usual stability condition that - 1 < -f(il)/b < 1. We also note that Equation (9) summarizes the familiar result that increasing the expectations lag (ie w decreasing) increases the region of local stability. However, it is the behaviour of the model in the region of instability which is our main concern here. This aspect is best approached by first considering the evolution of the difference equation, (5), from a geometric point of view. To do this we need a sketch of the general features of the function g. Such a sketch is easily obtained by considering the derivative function g and is given in Figure 2, where we see that two cases need to be considered: (i) w = 1, and (ii) w < 1. The turning points n 1, z2 in the case w < 1 are the solutions of
f(n)= -b ;-

-b

$1)

cn nl

n2

1 >

Figure 3.

The function f.

which is illustrated in Figure 3. The dynamics of the difference equation, (5), is seen

geometrically by placing a 45 line on the graphs of Figure 2 and constructing the usual cobweb motions. However, before taking this step it is instructive to

ECONOMIC

MODELLING

October 1988

379

The cobweb model: C. Chiarella

analyse the model in the case w = 1, which is the traditional case. In this case
x,+1

is the case discussed by Leontief [ 121 and Manning [ 151. As we shall see, a more diverse dynamic behaviour emerges when w < 1 is considered.

=P,

(11) Period doubling to chaos

The dynamics of the difference equation (5) in the case of local instability of the equilibrium point are illustrated in Figure 5. Consider first of all Figure 5a, which is the story of Figure 4 viewed from the viewpoint of expected prices x, rather than actual prices pr, with r-c,and rcL corresponding to pu and pL respectively.

so we are able to obtain a difference equation for pt directly without having to resort to the function g. By equating supply and demand we find that pt satisfies the difference equation

a+bh=f(Pt-1)

(12)

In Figure 4 we construct the cobweb in the case of instability. There we see that as prices diverge from the equilibrium p, they may ultimately settle down into the two cycle {pL, pu, pL, p,, . . . .}, with quantities settling down into the corresponding two cycle {qL, qu, qL, 4.9 . . .}. Such a cycle may arise because of the existence of a lower supply constraint qL, from which p,, qu and pL are determined via
Sd(P.) =
4L

4 = qs(P.)
cld(PLL) = 4.

(13)

This is the situation depicted in Figure 4 and could be justified by the argument that producers are unwilling to allow output to drop below some minimum level so as to maintain productive capacity. Alternatively we may dispense with the lower supply constraint and postulate the existence of a price pair pu and pL satisfying the condition q"(p,) = qd( pL) (this

9:

PL

P"

w<

Figure 4. The two cycle when w = 1.

Figure 5. The two cycle.

380

ECONOMIC

MODELLING

October 1988

The cobweb model: C. Chiarella

Notice that if we define the iterated mapping

gt2(.rr) = g(iz(n))

(14)

then rc, and rcL are both stable fixed points of g(n) ie are both solutions of n = gZ(72) (15)

since rr, = g(nJ and IC,_ = g(a,). Now consider Figure 5b when w < 1. For w still close to 1 we might expect that as 7c, diverges away from Tr it might settle down into a two cycle as in the case w = 1. This will be so if rr, and A~ are stable fixed points of the iterated mapping g)(rt) (now with w < 1, whereas in (14) we had w = 1). It indeed turns out that such rr, and 7~~ exist; however rather than attempting to derive the sketch of gt2(n), which would be a rather messy procedure, we are able to appeal to some powerful results given by Feigenbaum [7]. This author discusses at length the difference equation x ,+ 1 = AX,( 1 - X,) = G(X,) (16)

The function G on the righthand side is a concave parabola cutting the horizontal axis at X =0 and X = 1, as shown in Figure 6. While the form of the function G may seem very particular, Feigenbaum shows that any difference equation whose righthand side has the basic concave, folding over nature of G will rapidly settle down to behave qualitatively in the same manner as Equation (16). Hence we shall first discuss the qualitative behaviour of (16) and then relate the behaviour of the difference equation (5) to it. For a complete discussion of the difference equation (16) we refer the reader to May [ 173. Here we reproduce the elements of the period doubling process so as to make the discussion self contained for the reader not familiar with chaos literature. Figure 6 displays the situation when the equilibrium X is locally stable. In the lower panel we sketch the function Gc2)(X) = G(G(X)), of which X is also a fixed point, stable in the case depicted. In fact any fixed point of G is a fixed point of G, but Gt2 may have fixed points which are not fixed points of G. It is this latter fact which is the basic reason for the emergence of bounded oscillatory behaviour when X becomes unstable. The equilibrium point X becomes unstable when G(X) = - 1, which occurs at I = A, = 3.0. At this point Gc2 is tangential to the 45 line at X =X. As G(X) falls below - 1 and X becomes an unstable fixed point of G, it also becomes an unstable fixed point of G (2). What is more important, however, is

Figure 6. IG(x)l < 1 - asymptotic stability.

that two additional, stable, fixed points, X, and X2, of Gc2 now emerge, as displayed in the lower panel of Figure 7. The motion of X, now spirals towards the stable two cycle {X,, X2, X,, X2, . . . >, as shown in the upper panel of Figure 7. It turns out that Gc2(X2) = G2(X,) and the two cycle remains stable for as long as IG(X,)( < 1 ie as long as X, and X2 are stable fixed points of Gf2). So the next critical value of Iz is that for which Gt2)(X2) = - 1 and the value is 1= A, = 3.449. As ;i increases through A2, the iterated function G*(X) = G2)(G(2)(X)) passes from having three fixed points (X,, X, X2) to having seven fixed points (X;, Xi, XT, X, X;, X2,X:), of which (X;,X:,X;,X:) are stable and (X,, X, X2) are unstable. Now X, tends towards the stable four cycle {X;, XT, X;, X:, X;, XT, . ..} (see Figure 8). As with the stable fixed points of Gt2) it turns out that Gc4) has the same slope at each of its stable fixed points. Obviously the next critical value occurs at

ECONOMIC

MODELLING

October 1988

381

The cobweb model: C. Chiarella

Figure 7. IG(x)I > 1 - the

emergence of a stable two cycle.

where Gc4(X;) = - 1 and it is by now clear that by considering the fixed points of Gf6)( X) = Gt4(Gt4)( X)) a stable eight cycle will appear. This process whereby the stable cycle doubles its period as 2 passes through the critical values A,, AZ, A 3, . . . is known as period doubling. The mathematical details of determining the sequence {A.} are discussed by Feigenbaum and it is worth noting that it is possible by a sequence of resealing transformations to reduce the problem to continually considering the non-zero fixed point of X = G(X), with G appropriately redefined. The important observation for our discussion is that A,, +A, = 3.570. Beyond this value of rZ the stable cycle has infinite period which means the cycle never repeats itself. The time path of X, is bounded oscillatory, does not converge to any fixed value of X and is not periodic; the term chaos has been coined to describe such dynamic behaviour. Feigenbaum shows an even more important property

I = A3 = 3.544

Figure 8.

The emergence of a four cycle.

of the period doubling phenomenon, namely that such behaviour is not dependent on the particular functional form of G. Any difference equation X,, 1 = G(X,), where G has a quadratic extremum will, after a few iterations, settle down to behave like the system (16). The sequence {A} is different for each particular functional form but of the same order of magnitude. Furthermore A,+A, at the rate 6- where the number 6 is independent of the particular form. In fact 6 x 4.67.

The onset of chaos in the cobweb model

The upshot of the discussion in the previous section is that the difference equation (5) will quickly settle down to exhibit the same qualitative behaviour as the

382

ECONOMIC

MODELLING

October 1988

The cobweb model: C. Chiarella

difference equation
7ct+

equals the ratio elasticity.

(17)

of supply

elasticity

to demand

1=

G(T)

where G is the quadratic approximation to g which passes through the origin, the point (8, g(ii)) and satisfies G(ii) = g( 5). The function G satisfying these requirements is given by G(n) = n(A - jln) where A= 2g( 3) > 0 and /I=(1 -g(ii))/?i>O (19) If we define the resealed variable X, = @n,/A,then (17) reduces to the difference equation (16), for which we know that chaotic behaviour sets in for A z 3.57 ie for g(?i) < - 1.57. We are now able to give a complete characterization of the dynamics of the cobweb model: 1 - 2/w < f(it)/b < 1 (asymptotically
1 - 2.57/w < f(%)/b < 1 - 2/w

Discussion
By introducing a fairly general non-linear supply function and distinguishing between production and expectations time lags we have shown that in its locally unstable state the cobweb model will enter either a period doubling regime or a chaotic regime. We see from Figure 9 that the chaotic regime forms the larger part of the locally unstable region. Meadows [18] pointed to the need to introduce non-linearities and time lags into the cobweb model so as to obtain more realistic, oscillatory, price time paths and he did so by use of computer simulation. Here we have obtained such behaviour with a very simple non-linear model. Sheffrin [22] makes the point that the cobweb model with adaptive expectations is more likely to be in the locally unstable region. He then argues that such local instability favours the use of a rational expectations framework for such models. Our analysis shows that the adaptive expectations framework can yield realistic behaviour in the locally unstable region. It may be that on other grounds the rational expectations framework is preferable to the adaptive expectations one, but the local instability of adaptive expectations alone does not provide such grounds. It is often argued that under adaptive expectations economic agents will be making systematic prediction errors and that therefore the only sensible expectations mechanism for rational economic agents is rational expectations. It is difficult to sustain this line of reasoning in relation to models exhibiting chaotic behaviour. Now economic agents do not see systematic errors and indeed even though they live in a deterministic environment they are not able to make long-term predictions with certainty. This latter fact follows from the well known property of chaotic motion that time paths which start close together separate exponentially. We refer the reader to Lichtenberg and Lieberman [ 133 for more detail on this point. We have considered a fairly general non-linear S-shaped supply function. Other cases which would warrant further investigation are backward bending supply curves and non-linear demand curves.

(18)

stable to 5)

(period doubling around it) f(il)/b < 1 - 2.57/w (chaotic motion around ii) (20) These various regions are succinctly represented in Figure 9. These conditions may also be expressed in terms of elasticities if we bear in mind that f(?t)/b

(ii)/6

u-l

Iw

References
1 W. Bamett and P. Chen, The Aggregation-Theoretic Figure 9. Dynamic regimes of the cobweb model.
Monetary Aggregates are Chaotic and Have Strange Attractors: An Econometric Application of Mathematical

ECONOMIC

MODELLING

October 1988

383

The cobweb model: C. Chiarella

3 4

8 9

10

11

Chaos, IC2 Institute, The University of Texas at Austin, April 1986. J. Benhabib and R.H. Day, A characterization of erratic dynamics in the overlapping generations model, Journal of Economic Dynamics and Control, Vol4,1982, pp 37-55. J.A. Carlson, An invariably stable cobweb model, Review of Economic Studies, Vol 35, 1968. J.P. Crutchfield, J. Doyne Farmer, N.H. Packard and R.S. Shaw, Chaos, Scientijc American, Vol 255, December 1986, pp 38-49. R.H. Day, Irregular growth cycles, American Economic Review, Vol 72, 1982, pp 406-414. M. Ezekiel, The cobweb theorem, Quarterly Journal of Economics, Vol 52, 1938. M.J. Feigenbaum, Universal behaviour in nonlinear systems, in G.I. Ioos, ed, Nonlinear Dynamics and Turbulence, 1983, Pitman, London. J.-M. Grandmont, On endogenous competitive business cycles, Econometrica, Vol 53, 1985, pp 995-1045. J. Guckenheimer, G. Oster and A. Ipaktchi, The dynamics of density dependent population models, Journal of Mathematical Biology, Vol4,1977, pp lOl- 147. R.V. Jensen and R. Urban, Chaotic behaviour in a non-linear cobweb model, Economics Letters, Vol 15, 1984, pp 235-240. K. Kaneko, Collapse of Tori and Genesis of Chaos in Dissipatioe Systems, 1986, World Scientific Publishing

Co., Singapore. und 12 W.W. Leontief, Verzogerte Angebotsanpassing partielles Gleichgewicht, Zeitschri#fur Nationalokonomie, Vol 5, No 5, 1934, pp 670-676. 13 A.J. Lichtenberg and M.A. Lieberman, Regular and Stochastic Motion, 1983, Spring-Verlag, New York. 14 E.N. Lorenz, Deterministic non-periodic flows, Journal of Atmospheric Science, Vol 20, 1963, pp 130-141. 15 R. Manning, Stability of cobwebs, Economics Record,. Vol 46, 1970, pp 588-589. 16 R. Manning, A generalization of a cobweb theorem, Review of Economic Studies, Vol 38, 1971, pp 123-125. 17 R. May, Simple mathematical models with very complicated dynamics, Nature, Vol 261, 1976, pp 459-467. 18 D.L. Meadows, Dynamics of commodity production cycles, 1970, Wright-Allen Press, Cambridge, MA. 19 M. Nerlove, Adaptive expectations and cobweb phenomena, Quarterly Journal of Economics, Vol 72, 1958, pp 227-240. 20 M.T. Pohjola, Stable cyclic and chaotic growth: the dynamics of a discrete-time version of Goodwins growth cycle model, Zeitschrift fur Nationalokonomie, Vol 41, 1981, pp 27-38. 21 H.G. Schuster, Deterministic Chaos, Physik-Verlag, Weinheim, FR Germany. 22 S.M. Sheffrin, Rational Expectations, 1983, Cambridge. University Press, London.

384

ECONOMIC

MODELLING

October

1988