INTERNATIONAL ECONOMIC REVIEW
Vol. 22, No. 3, October, 1981
SOME NEW RESULTS ON THE DYNAMICS OF THE
GENERALIZED TOBIN MODEL*
BY JESS BENHABIB AND TAKAHIRO MIYAO'
The seminal article of Tobin [1965] on money and growth raised many
questions and led to a large and growing literature. In dealing with the unstable
character of the steady state and the explosive nature of almost all trajectories
that are generated by the Tobin model, Sidrauski [1967], Nagatani [1969], Stein
[1969], Hadjimichalakis [1971a, 1971b], Fischer [1972], and others "gener-
alized" the Tobin model by relaxing the short-run perfect foresight assumption
of the model and replacing it with the adaptive expectations hypothesis of Cagan
[1956]. Not surprisingly, it was found that the generalized Tobin model would
be stabilized provided that price expectations adjusted sufficiently slowly. If
expectations adjusted quickly, however, the instability problem remained.
More recently, Burmeister and Turnovsky [1978] constructed a model where
the speed of adjustment of short-run price expectations is characterized as a
parameter reflecting the degree of rationality, at one end beginning from an
irrational range, going through a range of adaptive expectations and to a final
limit of short-run perfect foresight. In our paper, we take a similar approach to
the dynamics of the generalized Tobin model and study exactly how the behavior
of the economic variables will be affected as the short-run price expectations of
the agents adjust more and more quickly. In particular, we demonstrate that
allowing expectations to adjust more quickly, as the agents start to predict faster,
does not necessarily lead to either the explosive, errant paths or the paths which
asymptotically converge to the steady state. Instead, these paths may very well
form trajectories of bounded and persistent oscillations.
First, we review the generalized Tobin model as formulated by Hadjimichalakis
[1971a]:
(1) k = sf(k)-(1 -s)(O-q)m-nk,
(2) ,i = m(O-p-n),
(3) y - - q),
(4) i=e[mn-L(k, q)] + q,
where the endogenous variables, k, mn, q and i are the capital-labor ratio, the
money stock per capita, the expected and the actual rates of inflation, respectively,
and the parameters, s, 0, n, y and E are the saving ratio, the rate of monetary
* Manuscript received May 12, 1980; revised January 12, 1981.
1 The authors are grateful to the editor and anonymous referees for comments and suggestions.
589
This content downloaded from 216.165.95.69 on Mon, 29 Feb 2016 21:25:23 UTC All use subject to JSTOR Terms and Conditions
590 J. BENHABIB AND T. MIYAO
expansion, the natural rate of growth, the speeds of adjustment of expectations
and of the price level, respectively. Note that the price adjustment equation (4)
includes the expected rate of inflation q, as suggested by Fischer [1972], and
Stein [1971], and others.
Substituting (4) into (2) and (3), we obtain three differential equations involv-
ing three variables k, m and q, whose steady state equilibrium values, k*, m* and
q*, can be found by setting k = == 0. Linearizing the dynamical system at
the equilibrium gives
~ k - ~ sf'-n -(1-s)n (1-s)m k-
(5) m =K smL1 -gm m(sL2-1) mm*j,
where all the variables are evaluated at the equilibrium. Let us define
a, - (sj' - n - em -yL
( sf '-n -(1-s)n | -gm m(sL2-1) sf '-n (1-s)m
(6) a2 ~~~+ +
smLI -gm I YE - ysL2 -ysL1 -ysL2
sf '-n -(1-s)n (1-s)m
a3 - gmLl -gm m(sL2-1)
-ysL1 Y8 - sL2
Then it is well known that a necessary and sufficient condition for local stability
of the linear system (5) is:
(7) a, > 0, a2> O, a1a2-a3 > O.
THEOREM 1. The steady state equilibrium is locally asymptotically stable
if the following conditions are satisfied.
(8) Y 0 - + 1m* + (1sk* <
and
(9) ark 1
where
(10) =s - _-q*L2/L, Sk*k*L1/L.
PROOF. (7) is satisfied under conditions (8) and (9) (See Appendix 1).
Since the stationary values k*, m* and q* are not affected by changes in y or a,
we can state from (8) that the smaller the value of y (the slower the adjustment of
expectations) or the greater the value of e (the faster the adjustment of the price
This content downloaded from 216.165.95.69 on Mon, 29 Feb 2016 21:25:23 UTC All use subject to JSTOR Terms and Conditions
DYNAMICS OF THE GENERALIZED TOBIN MODEL 591
level), the more likely is condition (8) satisfied. Also, the smaller the elasticity
of the money demand function with respect to q, or the greater the elasticity with
respect to k, the more likely is stability.2
In order to make the main point of this paper, let us go beyond the conventional
local stability analysis and examine what can happen if the steady state equilibri-
um becomes unstable due to certain changes in parameter values. The stability
conditions (8) and (9) indicate that increasing the speed of adjustment of expec-
tations beyond a certain value destabilizes the model. Thus it would seem that
as we move from adaptive expectations towards perfect foresight, we should obtain
"saddle-point" instability results analogous to those of the Tobin model. In
effect, the non-local situation is quite a bit more complex. As expectations
adjust faster, the emergence of instability at the steady state equilibrium seems to
be associated with persistent but bounded oscillations in the variables of the system.
For the sake of illustration, consider an increase in the stock of money at the
equilibrium. The immediate impact of this is to increase the price level and the
real money stock tends to fall back to its original level, but the initial increase in
money also tends to increase price expectations and reduce the capital stock.
The latter two effects reinforce the fall of the money supply, and depending on
their strength, they may cause the money stock to overshoot its long-run equilibri-
um level. As the money supply keeps falling beyond its equilibrium level, how-
ever, the effects on the other two variables are reversed: the capital stock rises
and expectations fall. Combined with the direct effect of the money stock on
the accumulation of money, the fall of the money stock will now be reversed.
This explanation, describing a possibility of divergent or convergent oscil-
lations, is based on local analysis in the neighborhood of the steady state. When
combined with the non-linearities of the system, oscillations which change from
being locally convergent to locally divergent as we vary the parameters, in general,
will also become non-trivial and persistent, forming closed orbits. We will now
prove the existence of closed orbits for the present system. We assume the
following:
(A.1) The functions L(k, q) and f(k) are Cs, where s?2 (continuous and at
least twice differentiable).
(A.2) There exists a set of parameter values such that the stationary point of
the dynamical system (I)-(4) is locally asymptotically stable.
LEMMA. Under (A.1) and (A.2), there exists a value of y=y*, where the
Jacobian of (5) has a pair of pure imaginary roots.
PROOF. By (A.2) there exist parameter values such that the necessary conditions
for the Jacobian of (5) to have roots with negative real parts, a,> 0, a2>0 and
2 Our stability conditions (8) and (9) are somewhat different from those obtained by
Hadjimichalakis [1971a, 1971b]. It is interesting to note that condition (9) guarantees the
uniqueness of the steady state equilibrium, as shown by Hadjimichalakis [1971b].
This content downloaded from 216.165.95.69 on Mon, 29 Feb 2016 21:25:23 UTC All use subject to JSTOR Terms and Conditions
592 J. BENHABIB AND T. MIYAO
a1a2-a3O0 (given by (7)), are satisfied. Since a1 = -(sf'-n)+cm+y8L2 and
the stationary values k*, m* and q* are independent of y, there exists a finite y=-
such that j > - (sf'- n - sm)/(cL2) so that a,<0. This implies that not all the
roots of the Jacobian can have negative real parts. Since the roots of the
Jacobian vary continuously with y, there must exist a value of y, say y*, where
either a real root or the real part of a pair of complex roots is equal to zero. If
a real root becomes zero, the determinant of the Jacobian must vanish, that is,
for y =y* we must have a3=0. But as can be seen from the proof of Theorem 1,
the sign of a3 is independent of y, as long as y does not change sign. Thus,
variations in y that do not alter its sign cannot make the determinant vanish, and
the Jacobian of (5) has a pair of pure imaginary roots at y =y*. Q. E. D.
THEOREM 2. Let (A.1) and (A.2) hold. Then there exists a continuous func-
tion y(b) with y(O)=y*, and for all sufficiently small values of 600, there exists
a continuous family of non-constant, positive periodic solutions [k(t, 6), in(t, 6),
q(t, 6)] for the dynamical system (1)-(4), which collapse to the stationary point
(k*, m*, q*) as 6-0.
PROOF. The proof follows from the Hopf Bifurcation Theorem (see Hopf
[1976] and Marsden and McCracken [1976, pp. 197-198]).3 We only have to
verify that the assumptions of the Hopf Theorem are satisfied. By (A.1) our
dynamical system is continuous and differentiable, and by (A.2) a stable stationary
point (k*, mn*, q*) exists and is independent of y. From the above lemma, the
Jacobian of the dynamical system has a pair of pure imaginary roots for y=Y*.
It is shown in Appendix 2 that the real part of the pure imaginary roots is not
stationary with respect to y. This covers the assumptions of the Hopf Theorem.
The existence of positive periodic solutions follows because the stationary point
is positive and the Hopf Theorem states that the periodic solutions collapse con-
tinuously onto the stationary point as b6-O. In other words, the amplitude of
the orbits approaches zero as b6-O. Q. E. D.
It is interesting to compare the above theorem with the early results on adaptive
expectations by Cagan [1956] and their application to the Tobin model by
Sidrauski [1967], Hadjimichalakis [1971a, 1971b], and others. Our results
show that the loss of stability that occurs as expectations adjust faster is associated
with the emergence of bounded, persistent oscillations in price, output and expec-
tations.4 This result holds no matter how quickly prices adjust, that is, no matter
how large the value of s is, since there always exists a value of y at which the
stability of the stationary point is lost. It should be noted that as y grows larger
8 Hopf states and proves the theorem for analytic functions. Howard and Koppel in thier
"editorial comments" in Hopf [1976] revise Hopf's proof and provide the C8 version used in
our proof.
I Similar techniques have been used by Benhabib and Nishimura [1979) to show the existence
of closed orbits in prices, outputs and stocks in a real rational expectations growth model with
long-run perfect foresight.
This content downloaded from 216.165.95.69 on Mon, 29 Feb 2016 21:25:23 UTC All use subject to JSTOR Terms and Conditions
DYNAMICS OF THE GENERALIZED TOBIN MODEL 593
we approach perfect foresight and instantaneous market clearing, and yet,
bounded, robust,5 and persistent orbits may exist globally beyond the local
neighborhood of the bifurcation value of y.6
The stability properties of the bifurcating orbits also require complex analysis,
but it is clear from the Hopf Theorem that the orbits will in general exist either in
a left (sub-critical) or a right (super-critical) neighborhood of y*. Given that the
third real root is negative, the bifurcating orbit will be locally attracting (locally
orbitally stable) if it is super-critical, and will be locally repelling (locally orbitally
unstable) if it is sub-critical. Whether the super-critical or sub-critical case holds
depends on the higher order non-linear terms in the Taylor expansion of the
dynamical system at the stationary point. For example, if the stationary point
is locally asymptotically stable at y* (at which point the Jacobian with pure
imaginary roots does not determine local stability), the bifurcating orbits will be
super-critical.7
The sub-critical case is also of special interest, and it corresponds to the "cor-
ridor stability" concept introduced by Leijonhufvud [1973]. For a left neighbor-
hood of y*, the economy will be locally stable for small perturbations around the
stationary point. The region of local stability in the neighborhood of the station-
ary point, however, is bounded by the closed orbit around it. A larger shock
may throw the economy out or beyond the orbit, in which case it does not have
a natural tendency to return to the stationary point. From an economic point
of view, sub-critical and super-critical orbits both seem plausible. Economic
considerations alone cannot restrict the..sign of the second, third and higher
derivatives of the functions appearing in the dynamical system (5) that determine
whether the orbits are sub-critical or super-critical.
New York University, U.S.A.
University of Southern California, U.S.A.
The orbits are "robust" in the sense that under any small perturbation to the parameters
of the system (to the underlying vector field) the orbits will not disappear, provided a certain
"characteristic multiplier" is not identically equal to unity. The case with the characteristic
multiplier identically equal to unity essentially corresponds to the linear case, and any small
nonlinearity will restore the robustness of the orbits. For a rigorous discussion, see Hirsch
and Smale [1976, pp. 309].
6 While a full global discussion of how the orbits change as theparameter A is varied beyond
the local neighborhood of its bifurcation value may be too technical to include here, some recent
results of Alexander and Yorke [1978] show that either (1) the bifurcation parameter of the
orbits or the period of the orbits will be pushed out of its defined domain (or go to infinity), or
(2) the orbit will close onto a stationary point distinct from the stationary point of the original
bifurcation.
See Marsden and McCracken [1976], and also Bruslinskaya [1961].
This content downloaded from 216.165.95.69 on Mon, 29 Feb 2016 21:25:23 UTC All use subject to JSTOR Terms and Conditions
594 J. BENHABIB AND T. MIYAO
APPENDIX 1
To prove Theorem 1, we find from (6) and (7)
(1-1) a, = -(sf'-n-ein-yeL2)
=- (sf '-s f + (1 -s)nm -sem-yeL2)
S f -f' + em + 'yL2-(1-s)nm >0?
= \k ,/k>0
because, from (8)
(1-2) 1+ y L2 (1 -s)n > 0.
(1-3) a2 =-(sf -n)em + (1 - s)nemL1 - emyeL2 - ysm(eL2 - 1)
-(sf -n)yeL2 + (1 - s)myeL,
=-(sf -n)(em + yeL2) + (1- s)emL,(n + y) + yeni
-(Sf -s + (1 - s)nM )(em + yeL2)
+ (1 - s)emL,(n + y) + ysm
=S({ -f')(em +ysL2) + (1 -s)ne k2 (n + ' Lk _I
-(1- s)n k yeL2 + ySm > 0,
from (8) (or (1-2)) and (9).
(1-4) a3 =-[(sf'- n)emyeL2 + (1 - s)nm(eL2 - 1)yeL1 + (1 - s)memL1ys
-(1- s)menyeL -(sf - n)m(eL2 - 1)yS - (1-s)neinLyeL2]
=-(1 -s)nm(eL2- 1)yeL + (1 - s)nemL1yeL2
-Sf'- fk + (I1-s)n MT mys
=S({ -f') mye+ (1 -s)nys m2 (L1 k _1)I
which leads to
This content downloaded from 216.165.95.69 on Mon, 29 Feb 2016 21:25:23 UTC All use subject to JSTOR Terms and Conditions
DYNAMICS OF THE GENERALIZED TOBIN MODEL 595
(1-5) ala2 - a3 = sQ -f') + gm + ysL2 - (1 - s)nmk !]Len
+ ( s)en (L k- I) + S -f ')(em + yeL2)
+ (1 -s)ycnL1 - (1- s)n!- yeL2j s(-{ -i)
yrnz+y(l - s)en (L1 I > O.
because, from (8)
(1-6) em + ysL2 - (1- s)nrnlk ? y.
APPENDIX 2
In this appendix we prove that the zero real parts of the pure imaginary roots
of the Jacobian of (5) are not stationary with respect to y. Let
(2-1) A3 + aA2 + a2A + a3 = 0
be the characteristic equation of the Jacobian. Then, by Orlando's formula (Sec
Gantmacher [1954, p. 197]),
(2-2) - 9l + A2)(Q'1 + X3)(2 + )3) = ala2- a3
must always hold. As stated in Gantmacher [1954], A and -A are both roots
of (2-1), if and only if ala2-a3=0. Note that a1, a2 and a3 are given by (6)
and also in the proof of Theorem 1. Now assume that x + yi and A are the roots
of (2-1) with x=O, which means that a~a2-a3=0. Differentiating both sides of
(2-2) with respect to y, we obtain
(2-3) -2 dx (A2 +y2) = d(ala2- a3)
dy dy
Then, from (1-1), (1-3) and (1-4) in the proof of Theorem 1, together with
a1a2-a3=O, it follows that
(2-4) /d(ala2 - a3) = da2 + a da2 - da3
dy -da +a dy dy
= EL2a ( + a, P)+ _9-- = a,(sL2 - P) < 0,
where
(2-5) p ( tg th fer+ (l - s)an d t /d 1) 0>
(2-3) anld (2-4) togetller lead to dxlxdy fO.
This content downloaded from 216.165.95.69 on Mon, 29 Feb 2016 21:25:23 UTC All use subject to JSTOR Terms and Conditions
596 J. BENHABIB AND T. MIYAO
REFERENCES
ALEXANDER, J. C. AND J. A. YORKE, "Global Bifurcation of Periodic Orbits," American Journal
of Mathematics, 100 (1978), 263-292.
BENHABIB, J. AND K. NISHIMURA, "The Hopf Bifurcation and the Existence and Stability of
Closed Orbits in Multisector Models of Optimal Economic Growth," Journal of Economic
Theory, 21 (December, 1979), 421-444.
BRUSLINSKAYA, N. N., "Qualitative Integration of a System of Differential Equations in a
Region Containing a Singular Point and a Limit Cycle," Soviet Mathematics Doklady,
2 (1961), 9-12.
BURMEISTER, E. AND S. J. TURNOVSKY, "Price Expectations, Disequilibrium Adjustments and
Macroeconomic Price Stability," Journal of Economic Theory, 17 (April, 1978), 287-311.
CAGAN, P., "The Monetary Dynamics of Hyperinflation," in, M. Friedman, ed., Studies in the
Quantity Theory of Money (Chicago: University of Chicago, 1956).
FISCHER, S., "Keynes-Wicksell and Neoclassical Models of Money and Growth," American
Economic Review, 62 (December, 1972), 880-890.
GANTMACHER, F. R., Theory of Matrices (New York: Interscience Publishers, 1954).
HADJIMICHALAKIS, M., "Money, Expectations, and Dynamics: an Alternative View," Inter-
national Economic Review, 12 (October, 1971a), 381-402.
, "Equilibrium and Disequilibrium Growth with Money: The Tobin Models," Review
of Economic Studies, 38 (October, 1971b), 457-479.
HIRSCH, M. W. AND S. SMALE, Differential Equations, Dynamic Systmnes and Linear Algebra
(New York: Academic Press, 1976).
HOPF, E., "Bifurcation of a Periodic Solution from a Stationary Solution of a System of Differ-
ential Equations," in, J. E. Marsden and M. McCracken, eds., The Hopf Bifurcation and
Its Applications (Berlin/New York: Springer-Verlag, 1976).
LEIJONHUFVUD, A., "Effective Demand Failure," Swedish Journal of Economics. 75 (March,
1973), 27-48.
MARSDEN, J. E. AND M. MCCRACKEN, The Hopf Bifurcation and Its Applications (Berlin/New
York: Springer-Verlag, 1976).
NAGATANI, K., "A Monetary Growth Model with Variable Employment," Journal of Money,
Credit and Banking, 1 (May, 1969), 188-206.
SIDRAUSKI, M., "Inflation and Economic Growth," Journal of Political Economy, 75 (December,
1967), 796-810.
STEIN, J. L., "Neoclassical and Keynes-Wicksell Monetary Growth Models," Journal of Money,
Credit and Banking, 1 (May, 1969), 153-171.
TOBIN, J., "Money and Economic Growth," Econometrica, 33 (October, 1965), 671-684.
This content downloaded from 216.165.95.69 on Mon, 29 Feb 2016 21:25:23 UTC All use subject to JSTOR Terms and Conditions