You are on page 1of 22

THE CENTRALISED SOLUTION OF THE UZAWA-LUCAS MODEL WITH

EXTERNALITIES

Alessio Moro


University of Cagliari


The paper solves the centralised version of the Uzawa -Lucas model when an externality is present. By means
of a transformation that uses the average output per unit of physical capital and the consumption to physical capital
ratio, the paper gets information on the characteristics of the transitional dynamics of the model, both locally and
(exceptionally for this kind of very complex models) globally. It is shown that the equation of motion of the average
output per unit of physical capital can be easily solved in closed form to generate a logistic function converging to the
steady state. Furthermore, also the equation of motion for the consumption over physical capital ratio can be explicitly
solved. Then, all the equations of motion of the original variables in the model can be recursively solved.


1. Introduction

After Romer [10],
1
Lucas [7] has greatly contributed to the debate on the ultimate
determinants of the growth process. This author, following the literature on the role of human
capital, particularly Uzawa [15], has developed a two sector model of economic growth with
externalities that explains the evolution of economic systems without any intervention of exogenous
factors.
2

In fact, the Uzawa-Lucas model has become very popular in the literature. One of its
characteristics is that it is a two sector model, with two production functions devoted respectively to
produce physical and human capital. Agents are homogeneous and have the same level of work
qualification and expertise (human capital). Moreover, they devolve a fraction of non leisure time to
produce the final good, and dedicate the remaining to training and studying. The first sector
produces a consumption good by mean of a constant returns to scale Cobb-Douglas technology,
whereas the second one produces human capital. The main peculiarity of the model stays in the
functional form of the technology used in the second sector. In fact, in this sector, a linear
technology is specified in which physical capital does not appear.
Another important aspect worth stressing concerns the presence of externalities that
introduces a wedge between the return on capital as it is perceived by the representative agent, and
the return on capital from a social point of view. The relevance of this difference is twofold. First of
all, one of the usual results of a two sectors growth model is no more obvious, as long as it is no
more guaranteed that the optimal trajectories monotonically converge towards the steady state
equilibrium along the saddle path in the phase diagram. Indeed, as shown in the recent literature
(Benhabib-Perli [2]), when some form of market imperfection is introduced in this kind of models,
a continuum of different equilibrium trajectories, all of them attracted by the same steady state, may
easily emerge. This fact implies that two economies, with the same initial conditions in terms of
state variables, will asymptotically have the same growth rate; however, the transitional dynamics,
and hence the steady state level of the variables, will depend upon the initial values selected for the
control variables.
Secondly, the presence of an externality lead us to distinguish between “decentralised” or
market solution, and “centralised” or social planner solution. As a matter of fact, when a distortion


Alessio Moro <alessiomoro@hotmail.com>, viale Merello 29, 09123 Cagliari (tel. 070-274793; 349-5272420) . With
the usual disclaimers, the paper has greatly benefited from the comments of four anonymous referees. Special thanks
are due to Prof. Paolo Mattana, whose guide and suggestions, together with the consultation of some of his unpublished
papers, were crucial in writing this paper.
1
See also Romer D. [11], p. 116 and f., and Barro R. J.-Sala i Martin X. [1], p. 146 and f.
2
Besides the pioneering contribution of Solow R. [12], the most known models of exogenous growth are Cass D. [3]
and Koopmans T. C. [6].
2
due to such an externality occurs, the market equilibrium solution differs from the optimal solution
that would be chosen by an hypothetical social planner. According with the most recent results on
endogenous growth theory, this work acknowledges the crucial importance, usually neglected in the
literature, of studying the centralised solution of the model when the externality is not zero.
The paper is organised as follows. In section 2, the standard Uzawa-Lucas model and its
decentralised solution is summarised. In section 3, the original system of four differential equations
is reduced into a three equations system, by means of an algebraic transformation of the variables
involved in the model. The new variables we introduce are two ratios: the physical to human capital
ratio and the consumption to physical capital ratio. It can be shown, in fact, that these ratios stay
constant along the balanced growth path (BGP henceforth). The BGP properties can be studied by
using the Jacobian matrix of the model. Indeed, the eigenvalues signs of this matrix reveal that a
stable branch exists, that brings the system to the long-run equilibrium.
In section 4, the system is further reduced into two differential equations in two unknowns,
the first one being the average output per unit of physical capital, while the second is, once again,
the consumption to physical capital ratio. It will be demonstrated that this system has the same
properties as the previous one in R
3
. Anyway, while the system in R
2
is surely less informative than
those in R
3
and R
4
(for instance, in R
2
the difference between state and control variables is no more
distinguishable), at the same time its dynamics can be represented in a phase diagram, which shows
the trajectories of the involved variables.
In section 5, it is suggested how the explicit solution of the reduced two-dimensional model
with externalities can be obtained. This solution, never presented until now in the literature on the
Uzawa- Lucas model, represents the most original and innovative contribution of this work. Some
policy implications of this solution are also illustrated at the end of the section. Finally, section 6
contains some brief conclusions.


2. The centralised solution of the model

2.1. The standard Uzawa-Lucas model
3


In the Uzawa-Lucas model, two economic sectors are considered: in the first one, output is
produced by means of physical capital and human capital, whereas in the second sector human
capital is produced by means of human capital. This means that the production of human capital
involves no physical capital. The model maintains the assumption that education is relatively
intensive in human capital.
In this model, to get endogenous growth it is not necessary that any externality be present.
As a matter of fact, endogenous growth is guaranteed by the fact that human capital has a constant
marginal productivity.
To each sector applies a distinct production function, which both exhibit constant returns to
scale. Thus, in steady state, the rates of return remain constant, and the economy can grow at a
constant rate. We define the production function in the physical sector as follows:

α α −
·
1
) (uhL AK Y (1)

where Y is the level of output, A is the constant technological level of the economy, K is physical
capital, L is labour, which is kept constant, u (with 0<u<1) is the fraction of labour time devolved to

3
A complete version of the Uzawa-Lucas model is presented in Barro R. J.-Sala i Martin X. [1]. Useful comments are
also found in Solow R. [13, 14].

3
produce output, so (1-u) is the fraction of labour time devolved to produce human capital, h is an
index of the average quality of labour, that is the schooling level of workers, therefore hL is the
quantity of labour measured in efficiency units, and finally uhL is the quantity of labour expressed
in efficiency units employed to produce output.
Even if the model could exhibit endogenous growth without the help of any externality, in
fact Lucas introduces this in the form of the average human capital of the working force. In other
words, such an externality can be defined as the average schooling level of workers. Therefore, the
production function in the first sector is completed as follows:

β α α
a
h uhL AK Y

·
1
) ( (2)

where
β
a
h is a measure of the externality.
Dividing by L, equation (2) takes the following form:

β α α
a
h uh Ak y

·
1
) ( (3)

The physical capital accumulation function defines net investment, which is given, ignoring
depreciation, by the difference between production and consumption, that is:

C h uhL AK K
a
− ·
− β α α 1
.
) ( (4)

or, considering that L is constant, we can write (4) in per capita terms:

c h uh Ak k
a
− ·
− β α α 1
.
) ( (5)

where the coherence of the model requires that h
a
= h.
In the second sector, the human capital accumulation function is:

) 1 (
.
u h h − ·φ (6)

where φ is a constant that indicates schooling productivity. So, marginal productivity of human
capital is constant and it does not depend on the level of human capital already accumulated. In fact:

) 1 (
.
u
h
h
− · φ (7)

that is to say, human capital growth rate is given by its own productivity, which is constant in
steady state. As a consequence, also human capital growth rate is constant.


2.2. The maximisation problem

We can, therefore, define the constrained maximisation problem to be solved. The consumer
maximises an isoelastic utility function (CIES) taking into account the two constraints given by
.
k
and
.
h , with given initial level for k and h. We should recall that, with an isoelastic utility function,
the elasticity of marginal utility with respect to consumption (call it σ ) is constant, therefore it
4
does not depend on the consumption level. Such a constant is given by the inverse of the
instantaneous elasticity of substitution, which in turn is given by
( )
( ) c cU
c U
E
' '
'
− · =
σ
1
.
The maximisation problem is then specified as follows:

Max
0
U =
t t
e
c
ρ
σ
σ

∞ −



0
1
1
1
(8)
subject to:
c h u Ak k − ·
+ − − β α α α 1 1
.

) 1 (
.
u h h − · φ
0
) 0 ( k k · given,
0
) 0 ( h h · given, k, h, c>0 ∀t

where ρ is the instantaneous discount rate.
In this subsection, we derive the centralised solution of the model, when the central planner
takes into account the externality. As for such an externality to be coherent with respect to h, the
production function must be written as follows:

β α α α + − −
·
1 1
h u Ak y (9)

which is the form used in the physical capital accumulation function above.
To solve this problem, we need to define the Hamiltonian function:

)] 1 ( [ ] [
1
1
2
1 1
1
1
u h c h u Ak
c
H − + − +
]
]
]



·
+ − −

φ λ λ
σ
β α α α
σ
(10)

where λ
1
and λ
2
are the two costate variables or auxiliary variables. From (10) it is possible to get
the following first order conditions for a maximum:

0 ·
c
H ⇒
1
λ
σ
·

c (11)

0 ·
u
H ⇒ 0 ] ) 1 ( [
2
1
1
· − −
− + −
φ λ α λ
α β α α
h h u Ak (12)

1 1
ρλ λ + − ·
&
k
H ⇒ [ ]
1
1
.
1 1 1
1
ρλ λ α λ
β α α α
+ − ·
+ − − −
h u Ak (13)

2 2
ρλ λ + − ·
&
h
H ⇒ [ ] [ ]
2 2 2
1
1
) 1 ( ) 1 ( ρλ λ φ λ β α λ
α β α α
+ − · − + + −
− −
&
u h u Ak (14)

.
1
k H ·
λ
⇒ c h u Ak k − ·
+ − − β α α α 1 1
.
(15)

.
2
h H ·
λ
⇒ ) 1 (
.
u h h − · φ (16)

with the transversality condition:

5
[ ] 0 lim
2 1
· +

∞ →
h k e
t
t
λ λ
ρ
(17)


2.3. The Balanced Growth Path (BGP) analysis

From the system given by equations (11)-(17), it is possible to obtain the growth rates of the
variables along the balanced growth path, that is the growth rates for c, k, h and u:

[ ] ρ α
σ
β α α α
− ·
+ − − − 1 1 1
.
1
h u Ak
c
c
(18)
k
c
h u Ak
k
k
− ·
+ − − − β α α α 1 1 1
.
(19)
) 1 (
.
u
h
h
− ·φ (20)
( ) ( )
u
k
c
u
u
α
β α φ
α
β α φ

+ −
+ −
+ −
·
1
1 1
.
(21)

where the growth rate of u is obtained applying logs and deriving with respect to time equation (12).
Now, we look for the existence of optimal growth paths for the variables involved in the
model. To do this, first of all we need to fix the limits of the parametric space such that ) 1 , 0 (
*
∈ u .
The value of
*
u along its balanced growth path is given by:

( )( )
( ) β α σ φ
β φ ρ φ α
+ −
+ − −
− ·
1
1
1
*
u (22)

In order to guarantee that this expression for
*
u be bounded between zero and one, we need
to study the following two inequalities jointly:

( ) ( )
( )
1
1
1
<
+ −
+ − −
β α σ φ
β φ ρ φ α
(23)

( ) ( )
( )
0
1
1
>
+ −
+ − −
β α φσ
β φ ρ φ α
(24)

Thus, we obtain the parametric space ρ and σ , given by:

( ) φψ ρ , 0 ∈ (25)

and

φψ
ρ
σ − >1 (26)

6
where
α
β α
ψ

+ −
·
1
1
.
Once the parametric space has been defined, it is possible to find the steady state value of
the common growth rate for consumption, output, and physical capital. Such a growth rate is given
by:

σ α
β φ
σ
ρ φ
γ
) 1 ( −
+

·
y
(27)

In addition, in steady state we have:

h h k y
ψγ γ β α αγ γ · + − + · ) 1 ( (28)

from which, since in steady state both consumption and physical capital grow at the same rate as
output, it is also possible to obtain the following growth rate for human capital:

) 1 (
) )( 1 (
β α σ
βφ ρ φ α
γ
+ −
+ − −
·
h
(29)

which is different from (27), that is from the common growth rate of consumption, output and
physical capital.
The difference between the two growth rates is due to the externality. If this is zero (β=0),
the growth rate of human capital is exactly the same as the common growth rate of the other
variables, and this common value collapses to:

σ
ρ φ
γ
) ( −
· (30)

3. The reduction of the dynamic system in three dimensions

3.1. The transitional dynamics in R
3


In order to reduce the original system of four differential equations in four unknowns into a
three equations system in three unknowns, we need to define two new variables, obtained by the
following ratios between the original variables:

) 1 (
) 1 (

+ −
·
α
β α
kh p (31)


k
c
q · (32)

In steady state, c and k grow at the same rate, hence their ratio stays constant. As for h, on
the contrary, we know that its growth rate, when the externality is taken into account, is lower than
k’s: in fact
h
γ is equal to
k
γ multiplied by
) 1 (
) 1 (
β α
α
+ −

.
7
It follows that, in order to keep the ratio between these two variables constant, h must be
raised to the power
) 1 (
) 1 (

+ −
α
β α
. It is obvious that, if 0 · β , the last expression reduces to –1,
therefore the ratio between the two variables simply becomes
h
k
. This ratio stays constant because,
in this case, both variables grow at the same common rate.
Using these two new variables, we can reduce by one the dimension of the original system,
which was formed by four differential equations. The new system can be specified as follows:

) 1 (
1
1
1 1
u q u Ap
p


+ −
− − ·
− −
α
β α
φ γ
α α
(33)

( ) ( )
u q
u
α
β α φ
α
β α φ
γ

+ −
+ −
+ −
·
1
1 1
(34)

σ
ρ
σ
σ α
γ
α α


+ ·
− − 1 1
u p A q
q
(35)

Solving this system, we obtain the steady state values of p and q, plus the value of u* just
determined in section 2.3. by equation (22). The growth rates of p and q must be zero in steady
state, being them defined as ratios between variables that are growing at the same rate.
Putting the three equations equal to zero, we then obtain the following steady state values
for p and q:

*
1
1
*
u
A
p

]
]
]

·
α
α
φψ
(36)

( )
* *
1
u q φψ
α
β α φ
+
+ −
· (37)

Substituting for u
*
its steady state value given by (22), we find the following value for q*:

α
φψϕ
σ
ρ
− · * q (38)

where
σ
σ α
ϕ

· .



3.2. Local stability analysis of the steady state

In order to analyse the local stability properties of the fixed point generated by the system
(33)–(35),
4
it is necessary to evaluate the Jacobian matrix J at the stationary point. The computation

4
It is worth noticing that in R
3
the BGP becomes a point in the phase space.
8
of the eigenvalues of this matrix, obtained applying standard mathematical theorems, yields very
interesting information on the movement of the variables in a neighbourhood of the steady state.
In the dynamical system generated by the centralised solution of the Uzawa-Lucas model,
we find that:
( )
BGP
t
t
t
t
t
t
t
t
t
t
t
t
t
t
t
t
t
t
* * *
q
q
u
q
p
q
q
u
u
u
p
u
q
p
u
p
p
p
q , u , p J
]
]
]
]
]
]
]
]
]



















·
& & &
& & &
& & &
(39)

The elements of the matrix are given as follows:

( ) ( )
a
t
a
t
t
t
t t t
a
t
t
t
u Ap a
p
p
u
a
? a
f q u p a A
p
p
J
− − − −
− + · −

+ −
− − ·


·
1 1 a 1 1
11
1 1
1
1 & &

( )
t
a
t
a
t
t
t
p
? a
f u p a A
u
p
J
a 1
1
1
12

+ −
+ − ·


·

&

t
t
t
p
q
p
J − ·


·
&
13

0
21
·


·
t
t
p
u
J
&

( ) ( ) ( )
t
t
t
t t
t
t
u
a
? a
f
u
u
u
a
? a
f q
a
? a f
u
u
J

+ −
+ ·

+ −
+ −
+ −
·


·
1
1
1
1
2
1
22
& &

t
t
t
u
q
u
J − ·


·
&
23

( )
t
a
t
a
t
t
t
q u Ap a
s
s a
p
q
J
− −


·


·
1 1
31
1
&

( )
t
a
t
a
t
t
t
q u Ap a -
s
s a
u
q
J


·


· 1
32
&

t
t
t a
t
a
t t
t
t
q
q
q
s
?
u Ap
s
s a
q
q
q
J + · −

+ ·


·
− −
& &
1 1
33
2


It follows that the Jacobian, evaluated along the BGP, can be written as:

( )
]
]
]
]
]
]
]
]



− +
·
*
*
* *
*
* *
* *
* * *
*
*
*
*
q
u
q f J
p
q f J
u ?u
p ?u J -
u
p
J
J
11 11
11 11
0 φ
φ
(40)

where ( ) ( ) ( )
a
*
a
* *
u p A a J
− −
− − ·
1 1
11
1 .
9
In order to study the stability properties of the steady state, we need to calculate the
eigenvalues of this matrix. The computation is simplified because it is possible to factorise the
characteristic equation of the matrix. The resulting equation is, in fact, the following:

[ ][ ][ ] 0 ? ? ?
3
*
2
*
1
*
11
· − − − q u J φψ (41)

To get the eigenvalues we only need to equalise to zero the three factors of the equation.
Hence, we obtain:

*
J ?
11 1
· (42)

*
2
u φψ κ · (43)

*
q ? ·
3
(44)

Now, substituting the expressions for p
*
and u
*
into the first eingevalue, we obtain its value
as a function of the following parameters:

( )
α
β α φ
κ
+ − −
·
1
1
(45)

We can therefore prove the following proposition:
Proposition 3.1. The characteristic equation associated with the Jacobian matrix J, evaluated at
the steady state, has a negative real eingevalue and two positive real eigenvalues: this implies that
the steady state is locally unique and shows the saddle path stabilities properties.
Proof. The analysis of the signs of (42), (43) and (44) allows us to mark an unambiguous sign to the
eigenvalues. Standard mathematical reasoning leads to the uniqueness result and the stability
properties of the steady state.
Proposition 3.1 implies that, for given initial values of physical and human capital, there
exists only one value for c and u that allows the system to settle on the stable branch. To this
regards, given the explicit form for the eigenvalues, we have the following
Corollary 3.1. The speed of convergence of the variables near the steady state is given by
·
1
κ [ ] α β α φ / ) 1 ( + − − . It increases (in absolute value) as φ increases and decreases as α
increases.
Proof. Standard mathematical reasoning proves that it is the negative eingevalue that regulates the
speed of convergence near the steady state. The sign of partial derivatives with respect to
parameters proves the second part of the corollary.
Moreover, it can be shown that the speed of convergence is a technical relation which is not
influenced by preference parameters.


4. The further reduction of the dynamic system into two dimensions

4.1. The transitional dynamics in R
2


10
In the centralised solution, the differential equations system of the Uzawa-Lucas model can
be further reduced to only two differential equations.
5
For this purpose the following proposition
holds:
Proposition 4.1. The transitional dynamics of Uzawa-Lucas model with externalities is fully
described by the following differential equation system:

( )
,
`

.
|
− − ·
1
1
.
1
1 z
z
z
α
φψ
α (46)

σ
ρ
ϕ − + ·
1 2
2
.
2
z z
z
z
(47)

Proof. Let us consider the two new variables
k
y
z ·
1
and
k
c
q z · ·
2
. Applying logs and time
derivatives we obtain the two differential equation given in the proposition.
The first variable
1
z , normalising the technological level A to one, is the average output per
unit of physical capital.
6
The second variable
2
z , which has already been defined in section 3.1., is
again the ratio between consumption and physical capital.
This new portrayal of the system does not allow to read directly the original values of the
variables. Also the distinction between state and control variables disappears. However, the
reduction of the model to only two equations permits to study the phase diagram of the system.
Moreover, this system in R
2
is saddle path stable.
Corollary 4.1. Given the initial conditions, the growth rate of
1
z is independent from
2
z .
Proof. Simply reading the equation of motion of
1
z leads to the conclusion in the corollary.
We can find the steady state values of
1
z and
2
z equalising to zero their growth rates. We
therefore obtain:

α
φψ
·
*
1
z (48)

α
φψϕ
σ
ρ
− ·
*
2
z (49)

Obviously,
*
2
z has the same value we already found in equation (38) for
*
q .

4.2. Local stability analysis of the BGP

At this point, we can analyse the stability of the BGP generated by the centralised solution
of the model in the (z
1
, z
2
) space. To do this, we can follow the same procedure as in section 3.2.
Therefore, after determining the Jacobian matrix (evaluated at the BGP), we can determine the signs
of the eigenvalues of the system.

5
The techniques used here are taken from Mattana P. [8, 9].
6
Substitute for y the value given by the production function and put A=1, we obtain
( )
( )
α
α
β α


+ −

,
`

.
|
·
1
1
1
1
k
uh
z
.

11
To write the Jacobian, we need to evaluate the partial derivatives of the two differential
equations with respect to each of the independent variables. We therefore have:

( )
1
1
.
1
1
.
1
1 z
z
z
z
z
α − − ·


(50)

0
2
.
1
·


z
z
(51)
2
1
.
2
z
z
z
ϕ ·


(52)

2
2
.
2
2
.
2
z
z
z
z
z
+ ·


(53)

from which, recalling that in the steady state the growth rates of
1
z and
2
z are zero, we can write
the Jacobian evaluated at the steady state as follows:

( )
]
]
]

− −
·
*
2
*
2
*
1 *
0 1
z z
z
J
ϕ
α
(54)

Now, we need the eigenvalues of the characteristic equation associated with this Jacobian
matrix, which takes the form:

( ) [ ][ ] 0 1
2
*
2 1
*
1
· − − − − κ κ α z z (55)

Setting both factors of this equation equal to zero, we get the following eigenvalues:

( )
*
1 1
1 z α κ − − · (56)

*
2 2
z · κ (57)

The first eingevalue is negative, being the two terms ( ) α − 1 and
*
1
z both positive, while the
second eingevalue is given by
*
2
z . The previously found value for
*
2
z does not allow us to know its
sign. Nevertheless, we can trace it back to the value we found for
*
q in section 3.1., given that
2
z
and q depict the same variable. From (37) we find that the value of
*
q is undoubtedly positive,
hence the second eingevalue is positive as well.
Proposition 4.2. The characteristic equation associated with the matrix J, evaluated at the steady
state, has a negative real eigenvalue and a positive real eigenvalue, given respectively by
( ) 0 1
*
1 1
< − − · z α κ and 0
*
2 2
> · z κ . Therefore, the steady state is saddle path stable and locally
unique.
Proof. Standard mathematical reasoning leads to the results given in the proposition.
We have therefore proved that reducing the dynamic system to only two equations yields the
same results already proved for the three equations system presented in section 3.

12
4.3. The phase diagram

As we know, a system in R
2
can be portrayed and studied in the corresponding phase
diagram. This particular kind of representation shows the dynamic movements of the variables of
the model (in this case
1
z and
2
z ).
In order to draw the phase diagram, it is necessary to set 0
1
· z& and 0
2
· z& . Putting the
growth rate of
1
z equal to zero, after simple algebra, we obtain that 0
1
· z& when:

α
φψ
·
1
z (58)

Following the same procedure for the growth rate of
2
z , we find that 0
2
· z& whenever:

1 2
z z
σ
σ α
σ
ρ −
− · (59)

We can therefore conclude that the locus of stationary points of
1
z , i.e. the locus of the
points where 0
1
· z& , is a perpendicular straight line to the
1
z axis at the
α
φψ
point. Such an
expression is nothing else than the stationary value of
1
z .
As regards to the stationary points of
2
z , we need to distinguish three cases. In fact, the
function (59) is a straight line, with the slope depending on the values of α and σ . Therefore, the
three cases are:
case 1: σ α · ⇒
σ
ρ
·
2
z . In this case, as shown in Figure 1, the function is perpendicular to the
2
z axis at the point
σ
ρ
.

Figure 1. Stationary points for σ =α
0
.
1
· z
z
2



0
.
2
· z


z
1




case 2: σ α < ⇒ the angular coefficient of the straight line
1 2
z z
σ
σ α
σ
ρ −
− · is positive. This
case is depicted in Figure 2.



σ
ρ

α
φψ

13






Figure 2. Stationary points for σ >α

z
2
0
.
1
· z






z
1




case 3: σ α > ⇒ the angular coefficient of the straight line
1 2
z z
σ
σ α
σ
ρ −
− · is negative. In such
a case, we have the situation depicted in Figure 3.

Figure 3. Stationary points for σ <α











Hence, the slope of the straight line 0
2
· z& depends on the ratio between the physical capital
share on output and the elasticity of marginal utility with respect to consumption.
Once we know the stationary points of
1
z and
2
z , we can study the motion of the two
variables in the other points of the diagram. To do that, we need to know the signs of the two
equations
1
z& and
2
z& . We know that:

( )
1 1 1
1 z z z
,
`

.
|
− − ·
α
φψ
α & (60)

and

2 1 2 2
z z z z
,
`

.
|
− + ·
σ
ρ
ϕ & (61)

α
φψ

σ
ρ

0
.
2
· z
α
φψ

1
z
2
z
σ
ρ

0
.
2
· z
0
.
1
· z
14
The first equation is positive until
1
z reaches its steady state value, and then it becomes
negative. The opposite is true for the second equation, which is negative until
2
z reaches its steady
state value.


Figure 4. The phase diagram for σ >α


Figure 5. The phase diagram for σ =α

Figure 6. The phase diagram for σ <α


Therefore, we can formulate the following
15
Proposition 4.3. As
1
z increases, it ranges from a region in which
1
z& >0 to a region in which
1
z& <0.
The boundary between these two regions is the straight line
1
z =
α
φψ
, which also is the steady state
value for
1
z . As
2
z increases, it passes from a region in which
2
z& < 0 to a region in which
2
z& > 0.
The boundary between these two regions is the straight line
1 2
z z
σ
σ α
σ
ρ −
− · .
Proof. The study of the signs of the functions
1
z& and
2
z& yields the results in proposition.
The three cases depicted in Figures 4, 5 and 6, which are computer generated for numerical
values of the parameters, confirm proposition 4.3.


5. The explicit solution of the model

5.1. The optimal path for the average output per unit of physical capital

The solution of the following reduced system of two different ial equations in the two
variables z
1
and z
2
:

( )
,
`

.
|
− − ·
1
1
.
1
1 z
z
z
α
φψ
α (62)

σ
ρ
ϕ − + ·
1 2
2
.
2
z z
z
z
(63)

gives the optimal paths that these variables follow along the transition towards the steady state.
The general solution of the model will contain two arbitrary constants. These constants will
depend on the initial conditions, i.e. on the values given to the variables at time zero. In fact, on the
basis of these initial values, the two variables will evolve differently in time. This is the reason why
it is necessary that the initial values for z
1
and z
2
must be chosen in such a way that the steady state
can be reached. This is guaranteed if the initial values for z
1
and z
2
are both chosen on the stable
branch of the phase diagram. This choice gives the definite solution of the model. If, on the
contrary, the initial values are not chosen on the stable branch, then the time evolution of the two
variables leads to points far away from the steady state.
Then, for the first order differential equations system (62)-(63), the following proposition
holds:
Proposition 5.1. The trajectory of z
1
is given by

( )
) exp( 1
*
1
*
1
1
t z b
z
t z
θ − +
· . (64)

Proof. The equation of motion for z
1
can be rewritten as ( ) θ
1 1
*
1
1
.
z z z z − · , where ) 1 ( α θ − · and
) 0 (
) 0 (
1
1
*
1
z
z z
b

· are control parameters and
α
φψ
·
*
1
z is the steady state value for z
1
.
It is worth noticing that the time path of z
1
depends on its steady state value, the initial
condition ) 0 (
1
z , the physical capital share on output α and, finally, the externality β , but it does
16
not depend on time preference parameters. Moreover, it should be noticed that the time trajectory of
z
1
, as well as the corresponding differential equation, does not depend on the position of z
2
.
Other characteristics of the time path of z
1
can be described as follows. When the value of
1
z
is close to zero, its growth rate, in absolute value, will be close to
α
β α φ ) 1 ( + −
. On the contrary,
when the value of
1
z grows, the growth rate asymptotically tends to zero and reaches this value in
the steady state.
It is possible to draw Figure 7 that shows the time path of z
1
in three different cases,
depending on whether the initial value of
1
z be, respectively, greater than, equal to or less than
α
φψ
·
*
1
z .
It is very interesting to note that the time path of z
1
follows a logistic law in the interesting
case when the level of human capital is lower than the optimal one (lower curve in Figure 7).
Anyway, the standard case applies whenever it is the physical capital that is lower than the optimal
level, which makes z
1
decrease along with decreasing growth rates (upper curve in Figure 7).
Figure 7. Different cases of convergence of z
1
(t) towards the steady state
z
1
(t)



5.2. The optimal path of the consumption/physical capital ratio


After the solution for z
1
is found, we can also find the solution for z
2
. This depends on the
initial choice of the value of z
1
.

Proposition 5.2. The trajectory of
k
c
q z · ·
2
is given by

( ) [ ]
( ) [ ]

+
,
`

.
| −

+
,
`

.
| −
·
dt t z b
t
R
t z b
t
t z
ζ
ζ
θ
σ
ρ
θ
σ
ρ
*
1
*
1
2
exp exp
exp exp
) ( (65)

17
where R is an arbitrary constant and
) 1 ( α σ
σ α
ζ


· .

Proof. The time path of z
2
, after substituting for the value of z
1
, is ruled by a Bernoulli equation that
can be solved.

Note that the complexity of the differential equation does not allow to obtain an explicit
solution for z
2
. This implies that, depending on the sign and the value of ζ, the integral at the
denominator of (65) takes a different form. In order to find the solution of the integral in the
denominator of equation (65), we need therefore to clearly separate two cases: the first when
α σ < , hence ζ is positive; the second when α σ > , hence ζ is negative.
Depending on the relation between these two parameters, the time path of z
2
will be
different. This also depends on b, i.e. the ratio of the difference between the steady state value of z
1

and its initial value over the same initial value.
In the next subsection, therefore, we consider two numerical cases, respectively with ζ >0
and ζ <0, and solve the integral in ) (
2
t z to obtain the form of the time path of this variable in the
two cases. It should also be noticed that when σ =α, then z
2
=z
2
(0) =
s
?
. In this case, there is no
transitional dynamics of z
2
.


5.3. Some numerical simulations of explicit trajectories for z
2


In order to investigate the shapes of alternative trajectories for z
2
, we assign explicit values
to the parameter ζ. In particular, we evaluate the symmetric cases of ζ = t2. In fact, the numerical
analysis shows that when ζ=2 the solution is purely algebraic, whereas the case in which ζ is
negative implies a form of the solution that is partly algebraic and partly logarithmic.
In addition, as it comes out from (65), the time path of z
2
(t) also depends on b, that is on the
position of z
1
(0) with respect to its steady state level, and hence on the sign of the same b. As a
consequence, in what follows, the values of ζ = t2 will be associated to two opposite values for b,
the first one associated to an excess of physical capital [z
1
(0)<
α
φψ
·
*
1
z ⇒ b<0], and the second one
associated to an excess of human capital [z
1
(0)>
α
φψ
·
*
1
z ⇒b>0].

a) First case: ζ=2, for σ <α

First of all, let us consider the case when ζ=2. In this case,
7
we have:
Proposition 5.3. The transitional dynamics of z
2
is decreasing if we start with an excess of physical
capital (b<0), and increasing if we start with an excess of human capital (b>0).
Proof. The numerical simulations invariably yield the results in the proposition.
Two particular cases are depicted in Figures 8 and 9. Once more, it should be noticed the
logistic form of time trajectories of z
2
, in the case in which the transition is due to an economic
shock that determines an excessive physical capital. On this regard, the numerical simulations show
that the logistic form of the trajectory increases with the value of b.




7
The results are confirmed also if we choose positive values for p different from 2.
18








Figure 8. The evolution of z
2
(t) associated to an excess of physical capital when σ<α

z
2
(t)


The computer simulations show more complex dynamics (hardly compatible with the real
evolution of an economy) for values of b smaller than –1. In this case, in fact, it happens that z
2
first
increases, then decreases and finally tends to a monotonic convergence towards the stationary state.
Figure 8, therefore, is obtained for b greater than –1 and smaller than zero.


b) Second case: ζ = -2, for σ >α

The case when σ >α, associated with a value of ζ = -2, is more interesting from the empi-
rical point of view. The numerical simulations performed allow to assert the following
Proposition 5.4. The transitional dynamics of z
2
is increasing if we start with an excess of physical
capital (b<0), and decreasing if we start with an excess of human capital (b>0).
Proof. The numerical simulations invariably yield the result in the proposition.
These cases are depicted in Figures 10 and 11. It should be stressed that, in the case when
we start with an excess of physical capital with respect to human capital, the time path of z
2
still
follows a logistic law: in a first period z
2
grows more than proportionally, then in a second period it
grows less than proportionally, getting asymptotically close to its steady state value.
Differently from the previous case, the arbitrary constant associated to the integration
process is not zero, but has to be chosen exactly in such a away to assure the convergence towards
the stationary state.
Finally, the case in which b<0 is, once more, more complicated to deal with. In fact, when
ζ= -2, the presence of an asymptote restricts the choice of b.

Figure 9. The evolution of z
2
(t) associated to an excess of human capital when σ <α

19
z
2
(t)



Figure 10. The evolution of z
2
(t) associated to an excess of physical capital when σ >α

z
2
(t)


Figure 11. The evolution of z
2
(t) associated to an excess of human capital when σ >α

z
2
(t)



5.3. Some further implications for the optimal time paths of the original variables of the model

The depicted evolution of the two variables z
1
and z
2
are rich of implications in terms of the
solution of the model in the original variables. To see that, let us recall the system of differential
equations obtained by the maximisation problem performed by the social planner:

[ ] ρ α
σ
β α α α
− ·
+ − − − 1 1 1
.
1
h u Ak
c
c
(66)

20
k
c
h u Ak
k
k
− ·
+ − − − β α α α 1 1 1
.
(67)
) 1 (
.
u
h
h
− ·φ (68)

( ) ( )
u
k
c
u
u
α
β α φ
α
β α φ

+ −
+ −
+ −
·
1
1 1
.
(69)

As should be noticed, these dynamic equations are all expressed in terms of z
1
and z
2
.
Although a deeper analysis of this point is beyond the scope of this paper, some numerical
simulations allow us to prove the following interesting:
Proposition 5.5. Given the explicit trajectories for z
1
and z
2
, all the dynamic laws originated by the
centralised solution of the Uzawa-Lucas model can be recursively solved.
Proof. The proposition derives from the numerical simulations depicted above.
Proposition 5.5. has many important effects from the normative economics point of view. If,
as proved by numerical simulations, it is possible to solve the model, then the social planner has
important instruments to manage the economy in order to maximise social welfare.


5.4. Interpretation of α and σ and policy implications

As we know, α is the share of physical capital and σ is the inverse of the elasticity of
substitution, which is also interpreted as a coefficient of risk aversion. If the agent is risk averse
(high value of σ ), he prefers to consume more income today instead of moving consumption (by
savings) to tomorrow. Thus, the dynamics in the Uzawa-Lucas model depends on the fact that the
share of physical capital is larger or smaller than the coefficient of risk aversion.
Proposition 5.5 implies that just one couple of optimal control variables, given by an initial
optimal consumption c*(0) and an initial optimal time devoted to work u*(0), corresponds to each
initial couple of physical capital k(0) and human capital h(0). Furthermore, this choice depends on
the value of the externality β .
Thus, given the externality β , as the ratio k(0)/h(0) varies, c and u take their optimal initial
values according to whether α is bigger or smaller than σ . It is easy to show that, given β and
h(0), when k(0) grows, c*(0) and u*(0) must increase if σ α < and decrease if σ α > . Then, the
policy maker can make initial c and u assume their optimal values. A way to do this is to adopt
fiscal policies that reduce or raise both initial consumption (by a consumption tax) and/or the time
devoted to work (by a labour tax). The measure of these policies depends on the effect of the
externality β on the human capital accumulation
8
.


6. Conclusions

The recent literature has stressed the importance of endogenous growth models, which
explain the evolution of economic systems without turning to the role of exogenous factors. Starting
from Lucas [7], who resumes some of the initial intuitions of Uzawa [15], a large body of this
literature has built on models with two capital goods (or two sectors), with or without the presence
of distorting factors such as externalities.

8
Studies on this kind of policies can be found in Judd K. [5] and Garcia -Castillo P., Sanso M. [4].
21
The peculiarities, in this literature, of the Uzawa-Lucas model, are many. The first one is the
explicit consideration of human capital, in the form of learning. Secondly, the Uzawa-Lucas model
takes into account an externality springing from human capital. This has remarkable effects with
regard both to the evolution of the economy (private or social/centralised) and, more generally, to
the properties of the transitional dynamics and the steady state. A great importance, in the light of
the most recent results on endogenous growth literature, is here attributed to the difference between
centralised and decentralised (or competitive) solution of the maximization problem. In this
framework, the main aim of this work was to obtain and study the centralised solution of the
Uzawa- Lucas model with externalities. As far as we know, the results here derived have not yet
been fully analysed in the literature.
These results are very interesting. Starting from the description of the characteristics of the
original model, we defined the balanced growth paths for the four variables that enter into the
model, i.e. consumption, physical capital, human capital and the fraction of labour time devolved to
produce output. The first result, fairly standardised in the literature of two sector models, is that in
the steady state the growth rates of physical capital, consumption and output are the same, while
human capital, if an externality is present, grows at a lower rate. Only if there is no externality does
the human capital grow at the same rate as the other variables of the model.
A further step exploits the possibility to reduce the dimension of the model from 4 to 3
equations, by mean of a transformation of variables. In this way, it becomes possible to study the
stability properties of the balanced growth path (BGP). However, the most interesting results are
obtained when the model is further reduced to only two equations. By mean of another
transformation that uses the average output per unit of physical capital and the consumption to
physical capital ratio, we get more information on the characteristics of the transitional dynamics,
both locally (through the phase diagram) and (exceptionally for this kind of very complex models)
globally.
As a matter of fact, it is this last point that represents the innovative conclusion of this work.
It is shown that the equation of motion of the average output per unit of physical capital can be
easily solved in closed form (to generate a logistic function converging to the steady state); once
this result has been obtained, also the equation of motion for the consumption over physical capital
ratio can be explicitly solved. Then, all the equations of motion of the original variables in the
model (consumption, output, physical capital, and human capital) can be recursively solved.
The paper ends up with some policy hints about the role of fiscal policy in implementing the
optimal trajectories of the involved variables. Once again, the measure of these policies depends on
the effect of the externality on the human capital accumulation.


REFERENCES

[1] Barro R.J., Sala-i-Martin X., Economic Growth, McGraw Hill, New York, 1995.
[2] Benhabib J., Perli R., “Uniqueness and Indeterminacy: on the Dynamics of Endogenous
Growth”, Journal of Economic Theory, vol. 63, 1994, pp. 113-142.
[3] Cass D., “Optimum Growth in an Aggregative Model of Capital Accumulation”, Review of
Economic Studies, vol. 32, 1965, pp. 233-240.
[4] Garcia-Castrillo P., Sanso M., “Human Capital and Optimal Policy in Lucas-Type Model”,
Review of Economics Dynamics, vol. 3, 2000, pp. 757-770.
[5] Judd K., “Optimal Taxation and Spending in General Competitive Growth Models”, Journal of
Public Economics, vol. 71, 1999, pp. 1-26.
[6] Koopmans T. C., On the Concept of Optimal Economic Growth, in The Econometric Approach
to Development Planning, North Holland, Amsterdam, 1965.
[7] Lucas R.E., “On the Mechanics of Economic Development”, Journal of Monetary Economics,
vol. 22, 1988, pp. 3-42.
22
[8] Mattana P., A Global Characterisation of the Optimal Solution of the Uzawa-Lucas Model,
University of Cagliari, Department of Economics Working Paper n. 23, 2001.
[9] Mattana P., The Uzawa-Lucas Model: a Complete Treatment, Ashgate, Aldershot, 2002
(forthcoming).
[10] Romer P. M., “Increasing Returns and Long Run Growth”, Journal of Political Economy, vol.
94, 1986, pp. 1002-1037.
[11] Romer D., Advanced Macroeconomics, McGraw Hill, New York, 1996.
[12] Solow R. M., “A Contribution to the Theory of Economic Growth”, Quarterly Journal of
Economics, vol. 70, 1956, pp. 65-94.
[13] Solow R. M., Lezioni sulla teoria della crescita endogena, La Nuova Italia Scientifica, Roma,
1994.
[14] Solow R. M., Growth Theory: an Exposition, Oxford University Press, New York-Oxford,
2000.
[15] Uzawa H., “Optimum Technical Change in an Aggregative Model of Economic Growth”,
International Economic Review, vol. 6, 1965, pp. 18-31.

2 due to such an externality occurs, the market equilibrium solution differs from the optimal solution that would be chosen by an hypothetical social planner. According with the most recent results on endogenous growth theory, this work acknowledges the crucial importance, usually neglected in the literature, of studying the centralised solution of the model when the externality is not zero. The paper is organised as follows. In section 2, the standard Uzawa-Lucas model and its decentralised solution is summarised. In section 3, the original system of four differential equations is reduced into a three equations system, by means of an algebraic transformation of the variables involved in the model. The new variables we introduce are two ratios: the physical to human capital ratio and the consumption to physical capital ratio. It can be shown, in fact, that these ratios stay constant along the balanc ed growth path (BGP henceforth). The BGP properties can be studied by using the Jacobian matrix of the model. Indeed, the eigenvalues signs of this matrix reveal that a stable branch exists, that brings the system to the long-run equilibrium. In section 4, the system is further reduced into two differential equations in two unknowns, the first one being the average output per unit of physical capital, while the second is, once again, the consumption to physical capital ratio. It will be demonstrated that this system has the same properties as the previous one in R3 . Anyway, while the system in R2 is surely less informative than those in R3 and R4 (for instance, in R2 the difference between state and control variables is no more distinguishable), at the same time its dynamics can be represented in a phase diagram, which shows the trajectories of the involved variables. In section 5, it is suggested how the explicit solution of the reduced two-dimensional model with externalities can be obtained. This solution, never presented until now in the literature on the Uzawa- Lucas model, represents the most original and innovative contribution of this work. Some policy implications of this solution are also illustrated at the end of the section. Finally, section 6 contains some brief conclusions.

2. The centralised solution of the model 2.1. The standard Uzawa-Lucas model3 In the Uzawa-Lucas model, two economic sectors are considered: in the first one, output is produced by means of physical capital and human capital, whereas in the second sector human capital is produced by means of human capital. This means that the production of human capital involves no physical capital. The model maintains the assumption that education is relatively intensive in human capital. In this model, to get endogenous growth it is not necessary that any externality be present. As a matter of fact, endogenous growth is guaranteed by the fact that human capital has a constant marginal productivity. To each sector applies a distinct production function, which both exhibit constant returns to scale. Thus, in steady state, the rates of return remain constant, and the economy can grow at a constant rate. We define the production function in the physical sector as follows: Y = AK α (uhL ) 1−α (1)

where Y is the level of output, A is the constant technological level of the economy, K is physical capital, L is labour, which is kept constant, u (with 0<u<1) is the fraction of labour time devolved to
3

A complete version of the Uzawa-Lucas model is presented in Barro R. J.-Sala i Martin X. [1]. Useful comments are also found in Solow R. [13, 14].

so (1-u) is the fraction of labour time devolved to produce human capital. we can write (4) in per capita terms: k = Ak α ( uh) 1−α h aβ − c . the production function in the first sector is completed as follows: β Y = AK α (uhL ) 1−α h a (2) where h a is a measure of the externality. considering that L is constant. the elasticity of marginal utility with respect to consumption (call it σ ) is constant. In fact: . The maximisation problem We can. h is an index of the average quality of labour.3 produce output. The consumer . maximises an isoelastic utility function (CIES) taking into account the two constraints given by k . Therefore. (5) where the coherence of the model requires that ha = h. which is given. that is: K = AK α (uhL) 1−α h aβ − C . (4) or. the human capital accumulation function is: h = φh (1 − u ) . 2. with an isoelastic utility function. that is the schooling level of workers. ignoring depreciation. In other words. and h . Dividing by L. with given initial level for k and h. equation (2) takes the following form: β y = Ak α (uh ) 1−α h a β (3) The physical capital accumulation function defines net investment. by the difference between production and consumption. also human capital growth rate is constant. which is constant in steady state. (6) where φ is a constant that indicates schooling productivity. In the second sector. therefore hL is the quantity of labour measured in efficiency units. define the constrained maximisation problem to be solved. So. marginal productivity of human capital is constant and it does not depend on the level of human capital already accumulated. As a consequence. Even if the model could exhibit endogenous growth without the help of any externality.2. such an externality can be defined as the average schooling level of workers. We should recall that. h = φ (1 − u ) h (7) that is to say. in fact Lucas introduces this in the form of the average human capital of the working force. and finally uhL is the quantity of labour expressed in efficiency units employed to produce output. therefore it . therefore. human capital growth rate is given by its own productivity.

we need to define the Hamiltonian function: (9) . To solve this problem.  c 1−σ − 1 α 1−α 1−α + β H = − c] + λ 2 [ hφ (1 − u )]  + λ1 [ Ak u h  1−σ  (10) where λ1 and λ2 are the two costate variables or auxiliary variables. the production function must be written as follows: y = Ak α u 1−α h1−α + β which is the form used in the physical capital accumulation function above. . In this subsection. Such a constant is given by the inverse of the U ' (c ) 1 instantaneous elasticity of substitution. which in turn is given by E = − = . ⇒ ⇒ k = Ak α u 1−α h1−α + β − c h = φh(1 − u) . k. h. . h (0) = h0 given. & λ1 (1 − α + β ) Ak α u1−α h β −α + λ2 [φ (1 − u )] = −λ2 + ρλ2 [ ] H λ1 = k H λ2 = h . ⇒ . we derive the centralised solution of the model. From (10) it is possible to get the following first order conditions for a maximum: Hc = 0 Hu = 0 ⇒ ⇒ c −σ = λ1 (11) (12) (13) (14) (15) (16) λ1[ Ak α (1 − α )u − α h 1+ β −α ] − λ2 hφ = 0 λ1 αAkα −1u1−α h1−α + β = − λ1 + ρλ1 & H k = −λ1 + ρλ1 ⇒ & H h = −λ2 + ρλ2 [ ] .4 does not depend on the consumption level. As for such an externality to be coherent with respect to h. with the transversality condition: . cU '' (c ) σ The maximisation problem is then specified as follows: ct1−σ −1 − ρt Max U 0 = ∫ e 1 −σ 0 subject to: ∞ (8) k = Ak α u 1−α h1−α + β − c h = φh(1 − u ) k ( 0) = k 0 given. when the central planner takes into account the externality. c>0 ∀ t where ρ is the instantaneous discount rate.

we need to study the following two inequalities jointly: (1 − α )(φ − ρ ) + φ β φ σ (1 − α + β ) (1 − α )(φ − ρ ) + φ β φσ (1 − α + β ) <1 (23) >0 (24) Thus. h = φ (1 − u) h u φ (1 − α + β ) c φ (1 − α + β ) = − + u u α k 1−α . where the growth rate of u is obtained applying logs and deriving with respect to time equation (12).3. The value of u * along its balanced growth path is given by: u * = 1− (1 − α )(φ − ρ ) + φ β φ σ (1 − α + β ) (22) In order to guarantee that this expression for u * be bounded between zero and one. Now. we obtain the parametric space ρ and σ . that is the growth rates for c. φψ ) and (25) σ >1− ρ φψ (26) . The Balanced Growth Path (BGP) analysis From the system given by equations (11)-(17). To do this.5 lim e − ρt [λ1k + λ 2 h] = 0 t→ ∞ (17) 2.1) . it is possible to obtain the growth rates of the variables along the balanced growth path. k. c 1 = αAk 1−α u 1−α h 1−α + β − ρ c σ . first of all we need to fix the limits of the parametric space such that u *∈(0. h and u: . given by: ρ ∈ (0. we look for the existence of optimal growth paths for the variables involved in the model. [ ] (18) (19) (20) (21) k c = Ak α −1u 1−α h1−α + β − k k .

(1 − α + β ) . If this is zero (β=0). 1− α Once the parametric space has been defined. it is also possible to obtain the following growth rate for human capital: γh = (1 − α )(φ − ρ ) + βφ σ (1 − α + β ) (29) which is different from (27). The difference between the two growth rates is due to the externality. we know that its growth rate. we need to define two new variables. and physical capital. when the externality is taken into account. is lower than (1 − α ) k’s: in fact γ h is equal to γ k multiplied by . that is from the common growth rate of consumption. output. The transitional dynamics in R3 In order to reduce the original system of four differential equations in four unknowns into a three equations system in three unknowns. output and physical capital. it is possible to find the steady state value of the common growth rate for consumption. As for h. on the contrary. The reduction of the dynamic system in three dimensions 3.6 1−α + β . since in steady state both consumption and physical capital grow at the same rate as output. the growth rate of human capital is exactly the same as the common growth rate of the other variables.1. Such a growth rate is given by: where ψ = γy = φ −ρ φβ + σ (1 − α )σ In addition. hence their ratio stays constant. in steady state we have: (27) γ y = αγ k + (1 − α + β )γ h = ψγ h (28) from which. obtained by the following ratios between the original variables: p = kh (1 −α + β ) (α −1 ) (31) q= c k (32) In steady state. and this common value collapses to: γ = (φ − ρ ) σ (30) 3. c and k grow at the same rate.

h must be (1 − α + β ) raised to the power . (α − 1) k therefore the ratio between the two variables simply becomes . if β = 0 . being them defined as ratios between variables that are growing at the same rate. in order to keep the ratio between these two variables constant. The computation 4 It is worth noticing that in R3 the BGP becomes a point in the phase space. This ratio stays constant because. we find the following value for q*: q* = ρ φψϕ − σ α α −σ . . the last expression reduces to –1. both variables grow at the same common rate. It is obvious that.2. we obtain the steady state values of p and q. 4 it is necessary to evaluate the Jacobian matrix J at the stationary point. The new system can be specified as follows: γ p = Ap α −1 u 1−α − q − φ 1−α + β (1 − u ) 1−α (33) γu = φ (1 − α + β ) φ (1 − α + β ) −q + u α 1 −α (34) γ q = q+ A α − σ α −1 1−α ρ p u − σ σ (35) Solving this system. we then obtain the following steady state values for p and q: φψ  α −1 * p =  u  αA  * 1 (36) q* = φ (1 − α + β ) + φψ u * α (37) Substituting for u* its steady state value given by (22). Putting the three equations equal to zero.7 It follows that. σ (38) where ϕ = 3. The growth rates of p and q must be zero in steady state. which was formed by four differential equations. plus the value of u* just determined in section 2. by equation (22). Using these two new variables. Local stability analysis of the steady state In order to analyse the local stability properties of the fixed point generated by the system (33)–(35). we can reduce by one the dimension of the original system. h in this case.3.

q = (39)  ∂pt ∂u t ∂qt   ∂q ∂q & & t ∂qt  &  t   ∂pt ∂u t ∂qt  BGP   ( ) The elements of the matrix are given as follows: ∂p t & p & 1 −a + ? = Aa p a −1u t1−a − q t −f (1 − u t ) = t + (a − 1) Ap a −1u t1−a t t ∂p t 1−a pt & ∂p t 1 −a + ? a −a J 12 = = A(1 −a ) p t u t + f pt ∂u t 1−a ∂p & J 13 = t = − pt ∂q t & ∂u J 21 = t = 0 ∂pt & & ∂u f (1 −a +? ) (1−a +? ) ut (1−a +? ) J22 = t = − qt + 2f ut = +f ut ∂ut a 1−a ut 1 −a & ∂u J 23 = t = −u t ∂qt ∂q & a −s J 31 = t = (a − 1) Apa −1u 1−a q t t t ∂p t s & ∂q t a − s (1 .J − * 11 + φ? u * ) φ ? u* * f J 11 q* u* a −1 * 1−a  − p*   − u*   *  q   .a ) Ap a u t−a q t J 32 = = t ∂u t s ∂q & & a −s ? q J 33 = t = 2q t + Ap a −1u a −1 − = t + q t t t ∂qt s s qt J 11 = It follows that the Jacobian.u . obtained applying standard mathematical theorems. we find that: & & &  ∂pt ∂p t ∂pt   ∂p ∂u ∂qt  t  t  & & & * * *  ∂u t ∂u t ∂u t  J p . In the dynamical system generated by the centralised solution of the Uzawa-Lucas model. evaluated along the BGP. yields very interesting information on the movement of the variables in a neighbourhood of the steady state. can be written as:  *  J 11  J* =  0  * *  f J 11q  p*  p* u* (.8 of the eigenvalues of this matrix. (40) * where J 11 = −(1 − a ) A p * ( ) (u ) .

substituting the expressions for p* and u* into the first eingevalue. Proof.1. To this regards. The analysis of the signs of (42). The sign of partial derivatives with respect to parameters proves the second part of the corollary. for given initial va lues of physical and human capital. Moreover. Standard mathematical reasoning proves that it is the negative eingevalue that regulates the speed of convergence near the steady state. The transitional dynamics in R2 . has a negative real eingevalue and two positive real eigenvalues: this implies that the steady state is locally unique and shows the saddle path stabilities properties. Proposition 3. Standard mathematical reasoning leads to the uniqueness result and the stability properties of the steady state. Proof. The resulting equation is. given the explicit form for the eigenvalues. we need to calculate the eigenvalues of this matrix. Hence.1. 4. The computation is simplified because it is possible to factorise the characteristic equation of the matrix.1.1 implies that. evaluated at the steady state. The further reduction of the dynamic system into two dimensions 4. it can be shown that the speed of convergence is a technical relation which is not influenced by preference parameters. the following: [J * 11 − ? 1 φψ u * − ? 2 q * − ? 3 = 0 ][ ][ ] (41) To get the eigenvalues we only need to equalise to zero the three factors of the equation. there exists only one value for c and u that allows the system to settle on the stable branch. we obtain its value as a function of the following parameters: κ1 = − φ (1 − α + β ) α (45) We can therefore prove the following proposition: Proposition 3. in fact. The characteristic equation associated with the Jacobian matrix J. we have the following Corollary 3. The speed of convergence o the variables near the steady state is given by f κ 1 = [− φ (1 − α + β ) /α ] . we obtain: * ? 1 = J 11 (42) (43) (44) κ 2 = φψ u * ? 3 = q* Now.9 In order to study the stability properties of the steady state. (43) and (44) allows us to mark an unambiguous sign to the eigenvalues. It increases (in absolute value) as φ increases and decreases as α increases.

which has already been defined in section 3. 2 4. The transitional dynamics of Uzawa-Lucas model with externalities is fully described by the following differential equation system: z1  φψ  = (1 − α )  − z1  z1  α  z2 ρ = z 2 + ϕ z1 − z2 σ Proof. Given the initial conditions. Applying logs and time k k derivatives we obtain the two differential equation given in the proposition. the reduction of the model to only two equations permits to study the phase diagram of the system. We can find the steady state values of z1 and z 2 equalising to zero their growth rates. the differential equations system of the Uzawa-Lucas model can be further reduced to only two differential equations. 6 The second variable z 2 .10 In the centralised solution. . Substitute for y the value given by the production function and put A=1. Proof. The first variable z 1 .2.1. (46) (47) y c and z 2 = q = .. is again the ratio between consumption and physical capital. Moreover. 9]. the growth rate of z 1 is independent from z 2 . 5 6 The techniques used here are taken from Mattana P. Let us consider the two new variables z1 = . . This new portrayal of the system does not allow to read directly the original values of the variables. z * has the same value we already found in equation (38) for q * . [8. this system in R2 is saddle path stable. To do this. We therefore obtain: z1* = φψ α (48) z* = 2 ρ φψϕ − σ α (49) Obviously. However. Local stability analysis of the BGP At this point. is the average output per unit of physical capital. Simply reading the equation of motion of z 1 leads to the conclusion in the corollary. Therefore.1. normalising the technological level A to one. Corollary 4. we can determine the signs of the eigenvalues of the system. we obtain +  (1(−α α )β )  uh 1 − z1 =  k        1−α . z2 ) space. Also the distinction between state and control variables disappears. after determining the Jacobian matrix (evaluated at the BGP). 5 For this purpose the following proposition holds: Proposition 4. we can follow the same procedure as in section 3.1.2. we can analyse the stability of the BGP generated by the centralised solution of the model in the (z1 .

1.. which takes the form: [− (1 − α )z * 1 − κ1 z * − κ 2 = 0 2 ][ ] (55) Setting both factors of this equation equal to zero. we need the eigenvalues of the characteristic equation associated with this Jacobian matrix. . Therefore. recalling that in the steady state the growth rates of z1 and z 2 are zero. we need to evaluate the partial derivatives of the two differential equations with respect to each of the independent variables. Proof. the steady state is saddle path stable and locally unique. . given respectively by * * κ 1 = −(1 − α )z 1 < 0 and κ 2 = z 2 > 0 .11 To write the Jacobian. has a negative real eigenvalue and a positive real eigenvalue. . being the two terms (1 − α ) and z 1 both positive.2. we can trace it back to the va lue we found for q * in section 3. we can write the Jacobian evaluated at the steady state as follows: − (1 − α )z1* J = *  ϕ z2 * 0  z*  2 (54) Now. The characteristic equation associated with the matrix J. Proposition 4. The previously found value for z * does not allow us to know its 2 2 sign. . From (37) we find that the value of q * is undoubtedly positive. hence the second eingevalue is positive as well. Standard mathematical reasoning leads to the results given in the proposition. Nevertheless. We have therefore proved that reducing the dynamic system to only two equations yields the same results already proved for the three equations system presented in section 3. . We therefore have: ∂ z1 z1 = − (1 − α ) z 1 ∂ z1 z1 ∂ z1 =0 ∂ z2 ∂ z2 = ϕ z2 ∂ z1 ∂ z2 z2 = + z2 ∂ z2 z2 . given that z 2 and q depict the same variable. (50) (51) (52) (53) from which. we get the following eigenvalues: * κ 1 = −(1 − α )z1 (56) (57) κ 2 = z* 2 * The first eingevalue is negative. evaluated at the steady state. . while the second eingevalue is given by z * .

Putting the & & growth rate of z 1 equal to zero. after simple algebra.3. i. the locus of the φψ points where z 1 = 0 . In this case. ρ α −σ − z1 is positive. is a perpendicular straight line to the z 1 axis at the & point. This particular kind of representation shows the dynamic movements of the variables of the model (in this case z 1 and z 2 ). with the slope depending on the values of α and σ .12 4. ρ σ z2 = 0 . z1 φψ α case 2: α < σ ⇒ the angular coefficient of the straight line z 2 = case is depicted in Figure 2. In order to draw the phase diagram. the function (59) is a straight line. we need to distinguish three cases. σ Figure 1. we find that z 2 = 0 whenever: & z2 = ρ α −σ − z1 σ σ (59) We can therefore conclude that the locus of stationary points of z 1 . This σ σ . a system in R2 can be portrayed and studied in the corresponding phase diagram. Such an α expression is nothing else than the stationary value of z 1 . we obtain that z 1 = 0 when: & z1 = φψ α (58) Following the same procedure for the growth rate of z 2 . it is necessary to set z 1 = 0 and z 2 = 0 . as shown in Figure 1.e. the three cases are: ρ case 1: α = σ ⇒ z 2 = . As regards to the stationary points of z 2 . Stationary points for σ =α z1 = 0 z2 . Therefore. The phase diagram As we know. the function is perpendicular to the σ ρ z 2 axis at the point . In fact.

In such σ σ z2 ρ σ φψ α z1 = 0 . we can study the motion of the two variables in the other points of the diagram. Figure 3. To do that. we need to know the signs of the two equations z1 and z 2 .13 Figure 2. we have the situation depicted in Figure 3. ρ σ φψ α z1 case 3: α > σ ⇒ the angular coefficient of the straight line z 2 = a case. Stationary points for σ <α ρ α −σ − z1 is negative. z2 = 0 . Once we know the stationary points of z 1 and z 2 . We know that: & &  φψ  & z1 = (1 − α )  − z1  z 1  α  (60) and ρ  & z 2 =  z 2 + ϕ z1 −  z 2 σ  (61) . z2 = 0 . z1 Hence. the slope of the straight line z 2 = 0 depends on the ratio between the physical capital & share on output and the elasticity of marginal utility with respect to consumption. Stationary points for σ >α z2 z1 = 0 .

14 The first equation is positive until z 1 reaches its steady state value. Figure 4. The phase diagram for σ =α Figure 6. and then it becomes negative. we can formulate the following . which is negative until z 2 reaches its steady state value. The phase diagram for σ >α Figure 5. The opposite is true for the second equation. The phase diagram for σ <α Therefore.

This is the reason why it is necessary that the initial values for z1 and z2 must be chosen in such a way that the steady state can be reached. where θ = (1 − α ) and ( ) z − z 1 (0) φψ are control parameters and z1* = is the steady state value for z1 . then the time evolution of the two variables leads to points far away from the steady state. the physical capital share on output α and. As z 1 increases. it passes from a region in which z 2 < 0 to a region in which z 2 > 0. on the values given to the variables at time zero. σ σ Proof. (64) * Proof. for the first order differential equations system (62)-(63). * 1 + b exp( −θz 1 t ) . on the basis of these initial values. it ranges from a region in which z1 >0 to a region in which z1 <0. & & The three cases depicted in Figures 4. These constants will depend on the initial conditions. the externality β . the following proposition holds: Proposition 5. The general solution of the model will contain two arbitrary constants.1. 5 and 6. the initial values are not chosen on the stable branch.3. Then. . confirm proposition 4. The equation of motion for z1 can be rewritten as z 1 = z 1 − z 1 z1θ . The trajectory of z1 is given by z1 (t ) = z1* . which are computer generated for numerical values of the parameters. which also is the steady state α value for z 1 . This choice gives the definite solution of the model. 5. If. The optimal path for the average output per unit of physical capital The solution of the following reduced system of two different ial equations in the two variables z1 and z2 : z1  φψ  = (1 − α )  − z1  z1  α  z2 ρ = z 2 + ϕz1 − z2 σ . z 1 (0) α It is worth noticing that the time path of z1 depends on its steady state value. As z 2 increases. the two variables will evolve differently in time. on the contrary. The explicit solution of the model 5. The study of the signs of the functions z1 and z 2 yields the results in proposition.e. & & ρ α −σ The boundary between these two regions is the straight line z 2 = − z1 . In fact. This is guaranteed if the initial values for z1 and z2 are both chosen on the stable branch of the phase diagram. (62) (63) gives the optimal paths that these variables follow along the transition towards the steady state.1. i. & & φψ The boundary between these two regions is the straight line z 1 = .15 Proposition 4. the initial condition z 1 ( 0) . finally.3. but it does b= * 1 .

which makes z1 decrease along with decreasing growth rates (upper curve in Figure 7). it should be noticed that the time trajectory of z1 . Other characteristics of the time path of z1 can be described as follows. The trajectory of z 2 = q = ζ  −ρ t  exp   b + exp θ z1*t  σ  z 2 (t) = ζ  −ρ t  * R − ∫ exp   b + exp θ z1 t dt  σ  c is given by k [ ( )] [ ( )] (65) .2. This depends on the initial choice of the value of z1 . respectively. Proposition 5. greater than. will be close to . as well as the corresponding differential equation. depending on whether the initial value of z 1 be. equal to or less than φψ z1* = . its growth rate. we can also find the solution for z2 . When the value of z1 φ (1 − α + β ) is close to zero. α when the value of z1 grows. Figure 7. does not depend on the position of z2 . the growth rate asymptotically tends to zero and reaches this value in the steady state. On the contrary. The optimal path of the consumption/physical capital ratio After the solution for z1 is found. Different cases of convergence of z1 (t) towards the steady state z1 (t) 5. the standard case applies whenever it is the physical capital that is lower than the optimal level. α It is very interesting to note that the time path of z1 follows a logistic law in the interesting case when the level of human capital is lower than the optimal one (lower curve in Figure 7). Moreover.2. in absolute value. It is possible to draw Figure 7 that shows the time path of z1 in three different cases. Anyway.16 not depend on time preference parameters.

respectively with ζ >0 and ζ <0. after substituting for the value of z1 . whereas the case in which ζ is negative implies a form of the solution that is partly algebraic and partly logarithmic. i. and solve the integral in z 2 (t ) to obtain the form of the time path of this variable in the ? two cases. In this case. the second whe n σ > α . and increasing if we start with an excess of human capital (b>0). In the next subsection. Some numerical simulations of explicit trajectories for z2 In order to investigate the shapes of alternative trajectories for z2 .17 where R is an arbitrary constant and ζ = α −σ . . The time path of z2 . In addition. Note that the complexity of the differential equation does not allow to obtain an explicit solution for z2 . This also depends on b. and hence on the sign of the same b. This implies that. It should also be noticed that when σ =α. The numerical simulations invariably yield the results in the proposition. On this regard.3. in the case in which the transition is due to an economic shock that determines an excessive physical capital. In order to find the solution of the integral in the denominator of equation (65). let us consider the case when ζ=2. the numerical analysis shows that when ζ=2 the solution is purely algebraic. the ratio of the difference between the steady state value of z1 and its initial value over the same initial value.3. As a consequence. the integral at the denominator of (65) takes a different form. 7 we have: Proposition 5. in what follows. In fact. φψ the first one associated to an excess of physical capital [z1 (0)< z1* = ⇒ b<0]. Two particular cases are depicted in Figures 8 and 9. the numerical simulations show that the logistic form of the trajectory increases with the value of b. hence ζ is negative. the values of ζ = ±2 will be associated to two opposite values for b. therefore. In particular. Depending on the relation between these two parameters. the time path of z2 will be different. 5. for σ <α First of all. as it comes out from (65). depending on the sign and the value of ζ. it should be noticed the logistic form of time trajectories of z2 . we evaluate the symmetric cases of ζ = ±2. 7 The results are confirmed also if we choose positive values for p different from 2. is ruled by a Bernoulli equation that can be solved. α a) First case: ζ=2. Once more. the time path of z2 (t) also depends on b. then z2 =z2 (0) = . that is on the position of z1 (0) with respect to its steady state level. hence ζ is positive. there is no s transitional dynamics of z2 . σ (1 − α ) Proof. and the second one α φψ associated to an excess of human capital [z1 (0)> z1* = ⇒b>0].e. we assign explicit values to the parameter ζ. In this case. The transitional dynamics of z2 is decreasing if we start with an excess of physical capital (b<0). we consider two numerical cases. Proof. we need therefore to clearly separate two cases: the first when σ < α .

associated with a value of ζ = -2. Figure 8. b) Second case: ζ = -2. then decreases and finally tends to a monotonic convergence towards the stationary state. Finally. is more interesting from the empirical point of view. once more. Differently from the previous case. then in a second period it grows less than proportionally. Figure 9. more complicated to deal with. In fact. in the case when we start with an excess of physical capital with respect to human capital.4. is obtained for b greater than –1 and smaller than zero. getting asymptotically close to its steady state value. therefore. These cases are depicted in Figures 10 and 11. and decreasing if we start with an excess of human capital (b>0). the case in which b<0 is. In this case. The numerical simulations performed allow to assert the following Proposition 5. for σ >α The case when σ >α. it happens that z2 first increases. the time path of z2 still follows a logistic law: in a first period z2 grows more than proportionally. It should be stressed that. in fact. The evolution of z2 (t) associated to an excess of human capital when σ <α . when ζ= -2.18 Figure 8. but has to be chosen exactly in such a away to assure the convergence towards the stationary state. the presence of an asymptote restricts the choice of b. Proof. The transitional dynamics of z2 is increasing if we start with an excess of physical capital (b<0). the arbitrary constant associated to the integration process is not zero. The evolution of z2 (t) associated to an excess of physical capital when σ<α z2 (t) The computer simulations show more complex dynamics (hardly compatible with the real evolution of an economy) for values of b smaller than –1. The numerical simulations invariably yield the result in the proposition.

The evolution of z2 (t) associated to an excess of human capital when σ >α z2 (t) 5. let us recall the system of differential equations obtained by the maximisation problem performed by the social planner: . Some further implications for the optimal time paths of the original variables of the model The depicted evolution of the two variables z1 and z2 are rich of implications in terms of the solution of the model in the original variables. The evolution of z2 (t) associated to an excess of physical capital when σ >α z2 (t) Figure 11. To see that.3.19 z2 (t) Figure 10. c 1 = αAk 1−α u 1−α h 1−α + β − ρ c σ [ ] (66) .

5. some numerical simulations allow us to prove the following interesting: Proposition 5. a large body of this literature has built on models with two capital goods (or two sectors). Proposition 5. the policy maker can make initial c and u assume their optimal values. Furthermore. It is easy to show that. (67) (68) h = φ (1 − u) h u φ (1 − α + β ) c φ (1 − α + β ) = − + u u α k 1−α . he prefers to consume more income today instead of moving consumption (by savings) to tomorrow. 8 Studies on this kind of policies can be found in Judd K.5. has many important effects from the normative economics point of view. A way to do this is to adopt fiscal policies that reduce or raise both initial consumption (by a consumption tax) and/or the time devoted to work (by a labour tax). 6. [5] and Garcia -Castillo P. [4]. the dynamics in the Uzawa-Lucas model depends on the fact that the share of physical capital is larger or smaller than the coefficient of risk aversion.20 . The measure of these policies depends on the effect of the externality β on the human capital accumulation8 . Then. . (69) As should be noticed. Sanso M. then the social planner has important instruments to manage the economy in order to maximise social welfare. Thus. If the agent is risk averse (high value of σ ). when k(0) grows. α is the share of physical capital and σ is the inverse of the elasticity of substitution. with or without the presence of distorting factors such as externalities.. as proved by numerical simulations. which is also interpreted as a coefficient of risk aversion. which explain the evolution of economic systems without turning to the role of exogenous factors. 5. who resumes some of the initial intuitions of Uzawa [15]. given β and h(0). Although a deeper analysis of this point is beyond the scope of this paper. Proof. c*(0) and u*(0) must increase if α < σ and decrease if α > σ . given the externality β . all the dynamic laws originated by the centralised solution of the Uzawa-Lucas model can be recursively solved. If. Given the explicit trajectories for z1 and z2 . Starting from Lucas [7]. Interpretation of α and σ and policy implications As we know. k c = Ak α −1u 1−α h1−α + β − k k . as the ratio k(0)/h(0) varies. these dynamic equations are all expressed in terms of z1 and z2 . given by an initial optimal consumption c*(0) and an initial optimal time devoted to work u*(0). The proposition derives from the numerical simulations depicted above. Conclusions The recent literature has stressed the importance of endogenous growth models. corresponds to each initial couple of physical capital k(0) and human capital h(0). Proposition 5. it is possible to solve the model. Thus. c and u take their optimal initial values according to whether α is bigger or smaller than σ .5 implies that just one couple of optimal control variables. this choice depends on the value of the externality β .4.

.. is here attributed to the difference between centralised and decentralised (or competitive) solution of the maximization problem.e. are many. if an externality is present. grows at a lower rate. the measure of these policies depends on the effect of the externality on the human capital accumulation.. Perli R. of the Uzawa-Lucas model. and human capital) can be recursively solved.Lucas model with externalities. once this result has been obtained. The paper ends up with some policy hints about the role of fiscal policy in implementing the optimal trajectories of the involved variables.. 1-26. vol. more generally. 22. the main aim of this work was to obtain and study the centralised solution of the Uzawa. 233-240. It is shown that the equation of motion of the average output per unit of physical capital can be easily solved in closed form (to generate a logistic function converging to the steady state). the most interesting results are obtained when the model is further reduced to only two equations. also the equation of motion for the cons umption over physical capital ratio can be explicitly solved. we defined the balanced growth paths for the four variables that enter into the model. 113-142. all the equations of motion of the original variables in the model (consumption. Review of Economics Dynamics. [5] Judd K. 2000. Journal of Economic Theory. Once again. The first one is the explicit consideration of human capital. 1988. In this way. we get more information on the characteristics of the transitional dynamics. McGraw Hill. The first result. 3-42.. pp. both locally (through the phase diagram) and (exceptionally for this kind of very complex models) globally. Only if there is no externality does the human capital grow at the same rate as the other variables of the model. in The Econometric Approach to Development Planning. A great importance. Sala-i-Martin X. 32. pp. New York. Review of Economic Studies. C. is that in the steady state the growth rates of physical capital. in the form of learning. vol. pp.J. physical capital. it becomes possible to study the stability properties of the balanced growth path (BGP). On the Concept of Optimal Economic Growth. REFERENCES [1] Barro R. while human capital. Economic Growth. by mean of a transformation of variables. “Uniqueness and Indeterminacy: on the Dynamics of Endogenous Growth”. . fairly standardised in the literature of two sector models. Then. “Human Capital and Optimal Policy in Lucas-Type Model”. In this framework. “Optimum Growth in an Aggregative Model of Capital Accumulation”. Secondly. 71. in the light of the most recent results on endogenous growth literature. vol.21 The peculiarities. 63. 757-770. [6] Koopmans T.. consumption. [2] Benhabib J. A further step exploits the possibility to reduce the dimension of the model from 4 to 3 equations. Amsterdam. 1965. [4] Garcia-Castrillo P.. By mean of another transformation that uses the average output per unit of physical capital and the consumption to physical capital ratio. [7] Lucas R. “Optimal Taxation and Spending in General Competitive Growth Models”.. These results are very interesting. human capital and the fraction of labour time devolved to produce output. pp. the Uzawa-Lucas model takes into account an externality springing from human capital. 1965. “On the Mechanics of Economic Development”. North Holland. This has remarkable effects with regard both to the evolution of the economy (private or social/centralised) and. [3] Cass D. vol. output. 1999.E. pp. in this literature.. consumption and output are the same. As far as we know. to the properties of the transitional dynamics and the steady state. it is this last point that represents the innovative conclusion of this work. Journal of Public Economics. 3. Journal of Monetary Economics. Starting from the description of the characteristics of the original model. i. physical capital. 1994. As a matter of fact. 1995. However. Sanso M. the results here derived have not yet been fully analysed in the literature. vol..

pp. 1994. M. [15] Uzawa H.. La Nuova Italia Scientifica.. M. 6. Oxford University Press. International Economic Review. M. “A Contribution to the Theory of Economic Growth”. 1996. Department of Economics Working Paper n. 23. “Increasing Returns and Long Run Growth”. . Lezioni sulla teoria della crescita endogena. Quarterly Journal of Economics. 1965.22 [8] Mattana P... 94.. A Global Characterisation of the Optimal Solution of the Uzawa-Lucas Model. Journal of Political Economy. [9] Mattana P. [12] Solow R. Advanced Macroeconomics. Growth Theory: an Exposition. University of Cagliari. [14] Solow R. [13] Solow R. New York.. vol. 1956. 65-94. The Uzawa-Lucas Model: a Complete Treatment. pp. 70. vol. 2002 (forthcoming). 1002-1037. 18-31. Roma. M.. New York-Oxford. 2001. [11] Romer D. 1986. 2000.. vol. Ashgate. Aldershot. [10] Romer P. “Optimum Technical Change in an Aggregative Model of Economic Growth”. McGraw Hill. pp.